linux/fs/btrfs/extent_map.c

1396 lines
39 KiB

License cleanup: add SPDX GPL-2.0 license identifier to files with no license Many source files in the tree are missing licensing information, which makes it harder for compliance tools to determine the correct license. By default all files without license information are under the default license of the kernel, which is GPL version 2. Update the files which contain no license information with the 'GPL-2.0' SPDX license identifier. The SPDX identifier is a legally binding shorthand, which can be used instead of the full boiler plate text. This patch is based on work done by Thomas Gleixner and Kate Stewart and Philippe Ombredanne. How this work was done: Patches were generated and checked against linux-4.14-rc6 for a subset of the use cases: - file had no licensing information it it. - file was a */uapi/* one with no licensing information in it, - file was a */uapi/* one with existing licensing information, Further patches will be generated in subsequent months to fix up cases where non-standard license headers were used, and references to license had to be inferred by heuristics based on keywords. The analysis to determine which SPDX License Identifier to be applied to a file was done in a spreadsheet of side by side results from of the output of two independent scanners (ScanCode & Windriver) producing SPDX tag:value files created by Philippe Ombredanne. Philippe prepared the base worksheet, and did an initial spot review of a few 1000 files. The 4.13 kernel was the starting point of the analysis with 60,537 files assessed. Kate Stewart did a file by file comparison of the scanner results in the spreadsheet to determine which SPDX license identifier(s) to be applied to the file. She confirmed any determination that was not immediately clear with lawyers working with the Linux Foundation. Criteria used to select files for SPDX license identifier tagging was: - Files considered eligible had to be source code files. - Make and config files were included as candidates if they contained >5 lines of source - File already had some variant of a license header in it (even if <5 lines). All documentation files were explicitly excluded. The following heuristics were used to determine which SPDX license identifiers to apply. - when both scanners couldn't find any license traces, file was considered to have no license information in it, and the top level COPYING file license applied. For non */uapi/* files that summary was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 11139 and resulted in the first patch in this series. If that file was a */uapi/* path one, it was "GPL-2.0 WITH Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 WITH Linux-syscall-note 930 and resulted in the second patch in this series. - if a file had some form of licensing information in it, and was one of the */uapi/* ones, it was denoted with the Linux-syscall-note if any GPL family license was found in the file or had no licensing in it (per prior point). Results summary: SPDX license identifier # files ---------------------------------------------------|------ GPL-2.0 WITH Linux-syscall-note 270 GPL-2.0+ WITH Linux-syscall-note 169 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17 LGPL-2.1+ WITH Linux-syscall-note 15 GPL-1.0+ WITH Linux-syscall-note 14 ((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5 LGPL-2.0+ WITH Linux-syscall-note 4 LGPL-2.1 WITH Linux-syscall-note 3 ((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3 ((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1 and that resulted in the third patch in this series. - when the two scanners agreed on the detected license(s), that became the concluded license(s). - when there was disagreement between the two scanners (one detected a license but the other didn't, or they both detected different licenses) a manual inspection of the file occurred. - In most cases a manual inspection of the information in the file resulted in a clear resolution of the license that should apply (and which scanner probably needed to revisit its heuristics). - When it was not immediately clear, the license identifier was confirmed with lawyers working with the Linux Foundation. - If there was any question as to the appropriate license identifier, the file was flagged for further research and to be revisited later in time. In total, over 70 hours of logged manual review was done on the spreadsheet to determine the SPDX license identifiers to apply to the source files by Kate, Philippe, Thomas and, in some cases, confirmation by lawyers working with the Linux Foundation. Kate also obtained a third independent scan of the 4.13 code base from FOSSology, and compared selected files where the other two scanners disagreed against that SPDX file, to see if there was new insights. The Windriver scanner is based on an older version of FOSSology in part, so they are related. Thomas did random spot checks in about 500 files from the spreadsheets for the uapi headers and agreed with SPDX license identifier in the files he inspected. For the non-uapi files Thomas did random spot checks in about 15000 files. In initial set of patches against 4.14-rc6, 3 files were found to have copy/paste license identifier errors, and have been fixed to reflect the correct identifier. Additionally Philippe spent 10 hours this week doing a detailed manual inspection and review of the 12,461 patched files from the initial patch version early this week with: - a full scancode scan run, collecting the matched texts, detected license ids and scores - reviewing anything where there was a license detected (about 500+ files) to ensure that the applied SPDX license was correct - reviewing anything where there was no detection but the patch license was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied SPDX license was correct This produced a worksheet with 20 files needing minor correction. This worksheet was then exported into 3 different .csv files for the different types of files to be modified. These .csv files were then reviewed by Greg. Thomas wrote a script to parse the csv files and add the proper SPDX tag to the file, in the format that the file expected. This script was further refined by Greg based on the output to detect more types of files automatically and to distinguish between header and source .c files (which need different comment types.) Finally Greg ran the script using the .csv files to generate the patches. Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org> Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-11-01 15:07:57 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 14:49:59 -04:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
2013-04-05 16:51:15 -04:00
2013-04-05 16:51:15 -04:00
btrfs: fix extent map merging not happening for adjacent extents If we have 3 or more adjacent extents in a file, that is, consecutive file extent items pointing to adjacent extents, within a contiguous file range and compatible flags, we end up not merging all the extents into a single extent map. For example: $ mkfs.btrfs -f /dev/sdc $ mount /dev/sdc /mnt/sdc $ xfs_io -f -d -c "pwrite -b 64K 0 64K" \ -c "pwrite -b 64K 64K 64K" \ -c "pwrite -b 64K 128K 64K" \ -c "pwrite -b 64K 192K 64K" \ /mnt/sdc/foo After all the ordered extents complete we unpin the extent maps and try to merge them, but instead of getting a single extent map we get two because: 1) When the first ordered extent completes (file range [0, 64K)) we unpin its extent map and attempt to merge it with the extent map for the range [64K, 128K), but we can't because that extent map is still pinned; 2) When the second ordered extent completes (file range [64K, 128K)), we unpin its extent map and merge it with the previous extent map, for file range [0, 64K), but we can't merge with the next extent map, for the file range [128K, 192K), because this one is still pinned. The merged extent map for the file range [0, 128K) gets the flag EXTENT_MAP_MERGED set; 3) When the third ordered extent completes (file range [128K, 192K)), we unpin its extent map and attempt to merge it with the previous extent map, for file range [0, 128K), but we can't because that extent map has the flag EXTENT_MAP_MERGED set (mergeable_maps() returns false due to different flags) while the extent map for the range [128K, 192K) doesn't have that flag set. We also can't merge it with the next extent map, for file range [192K, 256K), because that one is still pinned. At this moment we have 3 extent maps: One for file range [0, 128K), with the flag EXTENT_MAP_MERGED set. One for file range [128K, 192K). One for file range [192K, 256K) which is still pinned; 4) When the fourth and final extent completes (file range [192K, 256K)), we unpin its extent map and attempt to merge it with the previous extent map, for file range [128K, 192K), which succeeds since none of these extent maps have the EXTENT_MAP_MERGED flag set. So we end up with 2 extent maps: One for file range [0, 128K), with the flag EXTENT_MAP_MERGED set. One for file range [128K, 256K), with the flag EXTENT_MAP_MERGED set. Since after merging extent maps we don't attempt to merge again, that is, merge the resulting extent map with the one that is now preceding it (and the one following it), we end up with those two extent maps, when we could have had a single extent map to represent the whole file. Fix this by making mergeable_maps() ignore the EXTENT_MAP_MERGED flag. While this doesn't present any functional issue, it prevents the merging of extent maps which allows to save memory, and can make defrag not merging extents too (that will be addressed in the next patch). Fixes: 199257a78bb0 ("btrfs: defrag: don't use merged extent map for their generation check") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-28 16:23:00 +00:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: fix read corruption due to race with extent map merging In debugging some corrupt squashfs files, we observed symptoms of corrupt page cache pages but correct on-disk contents. Further investigation revealed that the exact symptom was a correct page followed by an incorrect, duplicate, page. This got us thinking about extent maps. commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them") enforces a reference count on the primary `em` extent_map being merged, as that one gets modified. However, since, commit 3d2ac9922465 ("btrfs: introduce new members for extent_map") both 'em' and 'merge' get modified, which started modifying 'merge' and thus introduced the same race. We were able to reproduce this by looping the affected squashfs workload in parallel on a bunch of separate btrfs-es while also dropping caches. We are still working on a simple enough reproducer to make into an fstest. The simplest fix is to stop modifying 'merge', which is not essential, as it is dropped immediately after the merge. This behavior is simply a consequence of the order of the two extent maps being important in computing the new values. Modify merge_ondisk_extents to take prev and next by const* and also take a third merged parameter that it puts the results in. Note that this introduces the rather odd behavior of passing 'em' to merge_ondisk_extents as a const * and as a regular ptr. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") CC: stable@vger.kernel.org # 6.11+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-18 15:44:34 -07:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: fix read corruption due to race with extent map merging In debugging some corrupt squashfs files, we observed symptoms of corrupt page cache pages but correct on-disk contents. Further investigation revealed that the exact symptom was a correct page followed by an incorrect, duplicate, page. This got us thinking about extent maps. commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them") enforces a reference count on the primary `em` extent_map being merged, as that one gets modified. However, since, commit 3d2ac9922465 ("btrfs: introduce new members for extent_map") both 'em' and 'merge' get modified, which started modifying 'merge' and thus introduced the same race. We were able to reproduce this by looping the affected squashfs workload in parallel on a bunch of separate btrfs-es while also dropping caches. We are still working on a simple enough reproducer to make into an fstest. The simplest fix is to stop modifying 'merge', which is not essential, as it is dropped immediately after the merge. This behavior is simply a consequence of the order of the two extent maps being important in computing the new values. Modify merge_ondisk_extents to take prev and next by const* and also take a third merged parameter that it puts the results in. Note that this introduces the rather odd behavior of passing 'em' to merge_ondisk_extents as a const * and as a regular ptr. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") CC: stable@vger.kernel.org # 6.11+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-18 15:44:34 -07:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: fix read corruption due to race with extent map merging In debugging some corrupt squashfs files, we observed symptoms of corrupt page cache pages but correct on-disk contents. Further investigation revealed that the exact symptom was a correct page followed by an incorrect, duplicate, page. This got us thinking about extent maps. commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them") enforces a reference count on the primary `em` extent_map being merged, as that one gets modified. However, since, commit 3d2ac9922465 ("btrfs: introduce new members for extent_map") both 'em' and 'merge' get modified, which started modifying 'merge' and thus introduced the same race. We were able to reproduce this by looping the affected squashfs workload in parallel on a bunch of separate btrfs-es while also dropping caches. We are still working on a simple enough reproducer to make into an fstest. The simplest fix is to stop modifying 'merge', which is not essential, as it is dropped immediately after the merge. This behavior is simply a consequence of the order of the two extent maps being important in computing the new values. Modify merge_ondisk_extents to take prev and next by const* and also take a third merged parameter that it puts the results in. Note that this introduces the rather odd behavior of passing 'em' to merge_ondisk_extents as a const * and as a regular ptr. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") CC: stable@vger.kernel.org # 6.11+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-18 15:44:34 -07:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
Btrfs: fix race between using extent maps and merging them We have a few cases where we allow an extent map that is in an extent map tree to be merged with other extents in the tree. Such cases include the unpinning of an extent after the respective ordered extent completed or after logging an extent during a fast fsync. This can lead to subtle and dangerous problems because when doing the merge some other task might be using the same extent map and as consequence see an inconsistent state of the extent map - for example sees the new length but has seen the old start offset. With luck this triggers a BUG_ON(), and not some silent bug, such as the following one in __do_readpage(): $ cat -n fs/btrfs/extent_io.c 3061 static int __do_readpage(struct extent_io_tree *tree, 3062 struct page *page, (...) 3127 em = __get_extent_map(inode, page, pg_offset, cur, 3128 end - cur + 1, get_extent, em_cached); 3129 if (IS_ERR_OR_NULL(em)) { 3130 SetPageError(page); 3131 unlock_extent(tree, cur, end); 3132 break; 3133 } 3134 extent_offset = cur - em->start; 3135 BUG_ON(extent_map_end(em) <= cur); (...) Consider the following example scenario, where we end up hitting the BUG_ON() in __do_readpage(). We have an inode with a size of 8KiB and 2 extent maps: extent A: file offset 0, length 4KiB, disk_bytenr = X, persisted on disk by a previous transaction extent B: file offset 4KiB, length 4KiB, disk_bytenr = X + 4KiB, not yet persisted but writeback started for it already. The extent map is pinned since there's writeback and an ordered extent in progress, so it can not be merged with extent map A yet The following sequence of steps leads to the BUG_ON(): 1) The ordered extent for extent B completes, the respective page gets its writeback bit cleared and the extent map is unpinned, at that point it is not yet merged with extent map A because it's in the list of modified extents; 2) Due to memory pressure, or some other reason, the MM subsystem releases the page corresponding to extent B - btrfs_releasepage() is called and returns 1, meaning the page can be released as it's not dirty, not under writeback anymore and the extent range is not locked in the inode's iotree. However the extent map is not released, either because we are not in a context that allows memory allocations to block or because the inode's size is smaller than 16MiB - in this case our inode has a size of 8KiB; 3) Task B needs to read extent B and ends up __do_readpage() through the btrfs_readpage() callback. At __do_readpage() it gets a reference to extent map B; 4) Task A, doing a fast fsync, calls clear_em_loggin() against extent map B while holding the write lock on the inode's extent map tree - this results in try_merge_map() being called and since it's possible to merge extent map B with extent map A now (the extent map B was removed from the list of modified extents), the merging begins - it sets extent map B's start offset to 0 (was 4KiB), but before it increments the map's length to 8KiB (4kb + 4KiB), task A is at: BUG_ON(extent_map_end(em) <= cur); The call to extent_map_end() sees the extent map has a start of 0 and a length still at 4KiB, so it returns 4KiB and 'cur' is 4KiB, so the BUG_ON() is triggered. So it's dangerous to modify an extent map that is in the tree, because some other task might have got a reference to it before and still using it, and needs to see a consistent map while using it. Generally this is very rare since most paths that lookup and use extent maps also have the file range locked in the inode's iotree. The fsync path is pretty much the only exception where we don't do it to avoid serialization with concurrent reads. Fix this by not allowing an extent map do be merged if if it's being used by tasks other then the one attempting to merge the extent map (when the reference count of the extent map is greater than 2). Reported-by: ryusuke1925 <st13s20@gm.ibaraki-ct.ac.jp> Reported-by: Koki Mitani <koki.mitani.xg@hco.ntt.co.jp> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=206211 CC: stable@vger.kernel.org # 4.4+ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-01-31 14:06:07 +00:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: fix read corruption due to race with extent map merging In debugging some corrupt squashfs files, we observed symptoms of corrupt page cache pages but correct on-disk contents. Further investigation revealed that the exact symptom was a correct page followed by an incorrect, duplicate, page. This got us thinking about extent maps. commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them") enforces a reference count on the primary `em` extent_map being merged, as that one gets modified. However, since, commit 3d2ac9922465 ("btrfs: introduce new members for extent_map") both 'em' and 'merge' get modified, which started modifying 'merge' and thus introduced the same race. We were able to reproduce this by looping the affected squashfs workload in parallel on a bunch of separate btrfs-es while also dropping caches. We are still working on a simple enough reproducer to make into an fstest. The simplest fix is to stop modifying 'merge', which is not essential, as it is dropped immediately after the merge. This behavior is simply a consequence of the order of the two extent maps being important in computing the new values. Modify merge_ondisk_extents to take prev and next by const* and also take a third merged parameter that it puts the results in. Note that this introduces the rather odd behavior of passing 'em' to merge_ondisk_extents as a const * and as a regular ptr. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") CC: stable@vger.kernel.org # 6.11+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-18 15:44:34 -07:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: fix read corruption due to race with extent map merging In debugging some corrupt squashfs files, we observed symptoms of corrupt page cache pages but correct on-disk contents. Further investigation revealed that the exact symptom was a correct page followed by an incorrect, duplicate, page. This got us thinking about extent maps. commit ac05ca913e9f ("Btrfs: fix race between using extent maps and merging them") enforces a reference count on the primary `em` extent_map being merged, as that one gets modified. However, since, commit 3d2ac9922465 ("btrfs: introduce new members for extent_map") both 'em' and 'merge' get modified, which started modifying 'merge' and thus introduced the same race. We were able to reproduce this by looping the affected squashfs workload in parallel on a bunch of separate btrfs-es while also dropping caches. We are still working on a simple enough reproducer to make into an fstest. The simplest fix is to stop modifying 'merge', which is not essential, as it is dropped immediately after the merge. This behavior is simply a consequence of the order of the two extent maps being important in computing the new values. Modify merge_ondisk_extents to take prev and next by const* and also take a third merged parameter that it puts the results in. Note that this introduces the rather odd behavior of passing 'em' to merge_ondisk_extents as a const * and as a regular ptr. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") CC: stable@vger.kernel.org # 6.11+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2024-10-18 15:44:34 -07:00
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
Btrfs: turbo charge fsync At least for the vm workload. Currently on fsync we will 1) Truncate all items in the log tree for the given inode if they exist and 2) Copy all items for a given inode into the log The problem with this is that for things like VMs you can have lots of extents from the fragmented writing behavior, and worst yet you may have only modified a few extents, not the entire thing. This patch fixes this problem by tracking which transid modified our extent, and then when we do the tree logging we find all of the extents we've modified in our current transaction, sort them and commit them. We also only truncate up to the xattrs of the inode and copy that stuff in normally, and then just drop any extents in the range we have that exist in the log already. Here are some numbers of a 50 meg fio job that does random writes and fsync()s after every write Original Patched SATA drive 82KB/s 140KB/s Fusion drive 431KB/s 2532KB/s So around 2-6 times faster depending on your hardware. There are a few corner cases, for example if you truncate at all we have to do it the old way since there is no way to be sure what is in the log is ok. This probably could be done smarter, but if you write-fsync-truncate-write-fsync you deserve what you get. All this work is in RAM of course so if your inode gets evicted from cache and you read it in and fsync it we'll do it the slow way if we are still in the same transaction that we last modified the inode in. The biggest cool part of this is that it requires no changes to the recovery code, so if you fsync with this patch and crash and load an old kernel, it will run the recovery and be a-ok. I have tested this pretty thoroughly with an fsync tester and everything comes back fine, as well as xfstests. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2012-08-17 13:14:17 -04:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: introduce extra sanity checks for extent maps Since extent_map structure has the all the needed members to represent a file extent directly, we can apply all the file extent sanity checks to an extent map. The new sanity checks will cross check both the old members (block_start/block_len/orig_start) and the new members (disk_bytenr/disk_num_bytes/offset). There is a special case for offset/orig_start/start cross check, we only do such sanity check for compressed extent, as only compressed read/encoded write really utilize orig_start. This can be proved by the cleanup patch of orig_start. The checks happens at the following times: - add_extent_mapping() This is for newly added extent map - replace_extent_mapping() This is for btrfs_drop_extent_map_range() and split_extent_map() - try_merge_map() For a lot of call sites we have to properly populate all the members to pass the sanity check, meanwhile the following code needs extra modification: - setup_file_extents() from inode-tests The file extents layout of setup_file_extents() is already too invalid that tree-checker would reject most of them in real world. However there is just a special unaligned regular extent which has mismatched disk_num_bytes (4096) and ram_bytes (4096 - 1). So instead of dropping the whole test case, here we just unify disk_num_bytes and ram_bytes to 4096 - 1. - test_case_7() from extent-map-tests An extent is inserted with 16K length, but on-disk extent size is only 4K. This means it must be a compressed extent, so set the compressed flag for it. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:03 +09:30
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
Btrfs: more efficient btrfs_drop_extent_cache While droping extent map structures from the extent cache that cover our target range, we would remove each extent map structure from the red black tree and then add either 1 or 2 new extent map structures if the former extent map covered sections outside our target range. This change simply attempts to replace the existing extent map structure with a new one that covers the subsection we're not interested in, instead of doing a red black remove operation followed by an insertion operation. The number of elements in an inode's extent map tree can get very high for large files under random writes. For example, while running the following test: sysbench --test=fileio --file-num=1 --file-total-size=10G \ --file-test-mode=rndrw --num-threads=32 --file-block-size=32768 \ --max-requests=500000 --file-rw-ratio=2 [prepare|run] I captured the following histogram capturing the number of extent_map items in the red black tree while that test was running: Count: 122462 Range: 1.000 - 172231.000; Mean: 96415.831; Median: 101855.000; Stddev: 49700.981 Percentiles: 90th: 160120.000; 95th: 166335.000; 99th: 171070.000 1.000 - 5.231: 452 | 5.231 - 187.392: 87 | 187.392 - 585.911: 206 | 585.911 - 1827.438: 623 | 1827.438 - 5695.245: 1962 # 5695.245 - 17744.861: 6204 #### 17744.861 - 55283.764: 21115 ############ 55283.764 - 172231.000: 91813 ##################################################### Benchmark: sysbench --test=fileio --file-num=1 --file-total-size=10G --file-test-mode=rndwr \ --num-threads=64 --file-block-size=32768 --max-requests=0 --max-time=60 \ --file-io-mode=sync --file-fsync-freq=0 [prepare|run] Before this change: 122.1Mb/sec After this change: 125.07Mb/sec (averages of 5 test runs) Test machine: quad core intel i5-3570K, 32Gb of ram, SSD Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com> Signed-off-by: Josef Bacik <jbacik@fb.com>
2014-02-25 14:15:13 +00:00
btrfs: fix corrupt read due to bad offset of a compressed extent map If we attempt to insert a compressed extent map that has a range that overlaps another extent map we have in the inode's extent map tree, we can end up with an incorrect offset after adjusting the new extent map at merge_extent_mapping() because we don't update the extent map's offset. For example consider the following scenario: 1) We have a file extent item for a compressed extent covering the file range [108K, 144K) and currently there's no corresponding extent map in the inode's extent map tree; 2) The inode's size is 141K; 3) We have an encoded write (compressed) into the file range [120K, 128K), which overlaps the existing file extent item. The encoded write creates a matching extent map, adds it to the inode's extent map tree and creates an ordered extent for it. Note that the corresponding file extent item is added to the subvolume tree only when the ordered extent completes (when executing btrfs_finish_one_ordered()); 4) We have a write into the file range [160K, 164K). This writes increases the i_size of the file, and there's a hole between the current i_size (141K) and the start offset of this write, and since the old i_size is in the middle of the block [140K, 144K), we have to write zeroes to the range [141K, 144K) (3072 bytes) and therefore dirty that page. We then call btrfs_set_extent_delalloc() with a start offset of 140K. We then end up at btrfs_find_new_delalloc_bytes() which will call btrfs_get_extent() for the range [140K, 144K); 5) The btrfs_get_extent() doesn't find any extent map in the inode's extent map tree covering the range [140K, 144K), so it searches the subvolume tree for any file extent items covering that range. There it finds the file extent item for the range [108K, 144K), creates a compressed extent map for that range and then calls btrfs_add_extent_mapping() with that extent map and passes the range [140K, 144K) via the "start" and "len" parameters; 6) The call to add_extent_mapping() done by btrfs_add_extent_mapping() fails with -EEXIST because there's an extent map, created at step 2 for the [120K, 128K) range, that covers that overlaps with the range of the given extent map ([108K, 144K)). Then it does a lookup for extent map from step 2 add calls merge_extent_mapping() to adjust the input extent map ([108K, 144K)). That adjust the extent map to a start offset of 128K and a length of 16K (starting just after the extent map from step 2), but it does not update the offset field of the extent map, leaving it with a value of zero instead of updating to a value of 20K (128K - 108K = 20K). As a result any read for the range [128K, 144K) can return incorrect data since we read from a wrong section of the extent (unless both the correct and incorrect ranges happen to have the same data). So fix this by changing merge_extent_mapping() to update the extent map's offset even if it's compressed. Also add a test case to the self tests. This didn't happen before the patchset that does big changes in the extent map structure (which includes the commit in the Fixes tag below) because we kept track of the original start offset in the extent map (member "orig_start") so we could always calculate the correct offset by subtracting that offset from the start offset. A test case for fstests that triggered this problem using send/receive with compressed writes will be added soon. Fixes: 3d2ac9922465 ("btrfs: introduce new members for extent_map") Reviewed-by: Qu Wenruo <wqu@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-07-11 13:33:23 +01:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: fix incorrect splitting in btrfs_drop_extent_map_range In production we were seeing a variety of WARN_ON()'s in the extent_map code, specifically in btrfs_drop_extent_map_range() when we have to call add_extent_mapping() for our second split. Consider the following extent map layout PINNED [0 16K) [32K, 48K) and then we call btrfs_drop_extent_map_range for [0, 36K), with skip_pinned == true. The initial loop will have start = 0 end = 36K len = 36K we will find the [0, 16k) extent, but since we are pinned we will skip it, which has this code start = em_end; if (end != (u64)-1) len = start + len - em_end; em_end here is 16K, so now the values are start = 16K len = 16K + 36K - 16K = 36K len should instead be 20K. This is a problem when we find the next extent at [32K, 48K), we need to split this extent to leave [36K, 48k), however the code for the split looks like this split->start = start + len; split->len = em_end - (start + len); In this case we have em_end = 48K split->start = 16K + 36K // this should be 16K + 20K split->len = 48K - (16K + 36K) // this overflows as 16K + 36K is 52K and now we have an invalid extent_map in the tree that potentially overlaps other entries in the extent map. Even in the non-overlapping case we will have split->start set improperly, which will cause problems with any block related calculations. We don't actually need len in this loop, we can simply use end as our end point, and only adjust start up when we find a pinned extent we need to skip. Adjust the logic to do this, which keeps us from inserting an invalid extent map. We only skip_pinned in the relocation case, so this is relatively rare, except in the case where you are running relocation a lot, which can happen with auto relocation on. Fixes: 55ef68990029 ("Btrfs: Fix btrfs_drop_extent_cache for skip pinned case") CC: stable@vger.kernel.org # 4.14+ Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-08-17 16:57:30 -04:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: drop extent map range more efficiently Currently when dropping extent maps for a file range, through btrfs_drop_extent_map_range(), we do the following non-optimal things: 1) We lookup for extent maps one by one, always starting the search from the root of the extent map tree. This is not efficient if we have multiple extent maps in the range; 2) We check on every iteration if we have the 'split' and 'split2' spare extent maps in case we need to split an extent map that intersects our range but also crosses its boundaries (to the left, to the right or both cases). If our target range is for example: [2M, 8M) And we have 3 extents maps in the range: [1M, 3M) [3M, 6M) [6M, 10M[ The on the first iteration we allocate two extent maps for 'split' and 'split2', and use the 'split' to split the first extent map, so after the split we set 'split' to 'split2' and then set 'split2' to NULL. On the second iteration, we don't need to split the second extent map, but because 'split2' is now NULL, we allocate a new extent map for 'split2'. On the third iteration we need to split the third extent map, so we use the extent map pointed by 'split'. So we ended up allocating 3 extent maps for splitting, but all we needed was 2 extent maps. We never need to allocate more than 2, because extent maps that need to be split are always the first one and the last one in the target range. Improve on this by: 1) Using rb_next() to move on to the next extent map. This results in iterating over less nodes of the tree and it does not require comparing the ranges of nodes to our start/end offset; 2) Allocate the 2 extent maps for splitting before entering the loop and never allocate more than 2. In practice it's very rare to have the combination of both extent map allocations fail, since we have a dedicated slab for extent maps, and also have the need to split two extent maps. This patch is part of a patchset comprised of the following patches: btrfs: fix missed extent on fsync after dropping extent maps btrfs: move btrfs_drop_extent_cache() to extent_map.c btrfs: use extent_map_end() at btrfs_drop_extent_map_range() btrfs: use cond_resched_rwlock_write() during inode eviction btrfs: move open coded extent map tree deletion out of inode eviction btrfs: add helper to replace extent map range with a new extent map btrfs: remove the refcount warning/check at free_extent_map() btrfs: remove unnecessary extent map initializations btrfs: assert tree is locked when clearing extent map from logging btrfs: remove unnecessary NULL pointer checks when searching extent maps btrfs: remove unnecessary next extent map search btrfs: avoid pointless extent map tree search when flushing delalloc btrfs: drop extent map range more efficiently And the following fio test was done before and after applying the whole patchset, on a non-debug kernel (Debian's default kernel config) on a 12 cores Intel box with 64G of ram: $ cat test.sh #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/nvme0n1 MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="-R free-space-tree -O no-holes" cat <<EOF > /tmp/fio-job.ini [writers] rw=randwrite fsync=8 fallocate=none group_reporting=1 direct=0 bssplit=4k/20:8k/20:16k/20:32k/10:64k/10:128k/5:256k/5:512k/5:1m/5 ioengine=psync filesize=2G runtime=300 time_based directory=$MNT numjobs=8 thread EOF echo performance | \ tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor echo echo "Using config:" echo cat /tmp/fio-job.ini echo umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Result before applying the patchset: WRITE: bw=197MiB/s (206MB/s), 197MiB/s-197MiB/s (206MB/s-206MB/s), io=57.7GiB (61.9GB), run=300188-300188msec Result after applying the patchset: WRITE: bw=203MiB/s (213MB/s), 203MiB/s-203MiB/s (213MB/s-213MB/s), io=59.5GiB (63.9GB), run=300019-300019msec Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2022-09-19 15:06:40 +01:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: use the flags of an extent map to identify the compression type Currently, in struct extent_map, we use an unsigned int (32 bits) to identify the compression type of an extent and an unsigned long (64 bits on a 64 bits platform, 32 bits otherwise) for flags. We are only using 6 different flags, so an unsigned long is excessive and we can use flags to identify the compression type instead of using a dedicated 32 bits field. We can easily have tens or hundreds of thousands (or more) of extent maps on busy and large filesystems, specially with compression enabled or many or large files with tons of small extents. So it's convenient to have the extent_map structure as small as possible in order to use less memory. So remove the compression type field from struct extent_map, use flags to identify the compression type and shorten the flags field from an unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and reduces the size of the structure from 136 bytes down to 128 bytes, using now only two cache lines, and increases the number of extent maps we can have per 4K page from 30 to 32. By using a u32 for the flags instead of an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(), but that level of atomicity is not needed as most flags are never cleared once set (before adding an extent map to the tree), and the ones that can be cleared or set after an extent map is added to the tree, are always performed while holding the write lock on the extent map tree, while the reader holds a lock on the tree or tests for a flag that never changes once the extent map is in the tree (such as compression flags). Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-04 16:20:33 +00:00
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: introduce new members for extent_map Introduce two new members for extent_map: - disk_bytenr - offset Both are matching the members with the same name inside btrfs_file_extent_items. For now this patch only touches those members when: - Reading btrfs_file_extent_items from disk - Inserting new holes - Merging two extent maps With the new disk_bytenr and disk_num_bytes, doing merging would be a little more complex, as we have 3 different cases: * Both extent maps are referring to the same data extents |<----- data extent A ----->| |<- em 1 ->|<- em 2 ->| * Both extent maps are referring to different data extents |<-- data extent A -->|<-- data extent B -->| |<- em 1 ->|<- em 2 ->| * One of the extent maps is referring to a merged and larger data extent that covers both extent maps This is not really valid case other than some selftests. So this test case would be removed. A new helper merge_ondisk_extents() is introduced to handle the above valid cases. To properly assign values for those new members, a new btrfs_file_extent parameter is introduced to all the involved call sites. - For NOCOW writes the btrfs_file_extent would be exposed from can_nocow_file_extent(). - For other writes, the members can be easily calculated As most of them have 0 offset and utilizing the whole on-disk data extent. The exception is encoded write, but thankfully that interface provided offset directly and all other needed info. For now, both the old members (block_start/block_len/orig_start) are co-existing with the new members (disk_bytenr/offset), meanwhile all the critical code is still using the old members only. The cleanup will happen later after all the old and new members are properly validated. There would be some re-ordering for the assignment of the extent_map members, now we follow the new ordering: - start and len Or file_pos and num_bytes for other structures. - disk_bytenr and disk_num_bytes - offset and ram_bytes - compression So expect some seemingly unrelated line movement. Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-30 07:53:02 +09:30
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: use a regular rb_root instead of cached rb_root for extent_map_tree We are currently using a cached rb_root (struct rb_root_cached) for the rb root of struct extent_map_tree. This doesn't offer much of an advantage here because: 1) It's only advantage over the regular rb_root is that it caches a pointer to the left most node (first node), so a call to rb_first_cached() doesn't have to chase pointers until it reaches the left most node; 2) We only have two scenarios that access left most node with rb_first_cached(): When dropping all extent maps from an inode, during inode eviction; When iterating over extent maps during the extent map shrinker; 3) In both cases we keep removing extent maps, which causes deletion of the left most node so rb_erase_cached() has to call rb_next() to find out what's the next left most node and assign it to struct rb_root_cached::rb_leftmost; 4) We can do that ourselves in those two uses cases and stop using a rb_root_cached rb tree and use instead a regular rb_root rb tree. This reduces the size of struct extent_map_tree by 8 bytes and, since this structure is embedded in struct btrfs_inode, it also reduces the size of that structure by 8 bytes. So on a 64 bits platform the size of btrfs_inode is reduced from 1032 bytes down to 1024 bytes. This means we will be able to have 4 inodes per 4K page instead of 3. Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-05-10 17:41:04 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: stop extent map shrinker if reschedule is needed The extent map shrinker can be called in a variety of contexts where we are under memory pressure, and of them is when a task is trying to allocate memory. For this reason the shrinker is typically called with a value of struct shrink_control::nr_to_scan that is much smaller than what we return in the nr_cached_objects callback of struct super_operations (fs/btrfs/super.c:btrfs_nr_cached_objects()), so that the shrinker does not take a long time and cause high latencies. However we can still take a lot of time in the shrinker even for a limited amount of nr_to_scan: 1) When traversing the red black tree that tracks open inodes in a root, as for example with millions of open inodes we get a deep tree which takes time searching for an inode; 2) Iterating over the extent map tree, which is a red black tree, of an inode when doing the rb_next() calls and when removing an extent map from the tree, since often that requires rebalancing the red black tree; 3) When trying to write lock an inode's extent map tree we may wait for a significant amount of time, because there's either another task about to do IO and searching for an extent map in the tree or inserting an extent map in the tree, and we can have thousands or even millions of extent maps for an inode. Furthermore, there can be concurrent calls to the shrinker so the lock might be busy simply because there is already another task shrinking extent maps for the same inode; 4) We often reschedule if we need to, which further increases latency. So improve on this by stopping the extent map shrinking code whenever we need to reschedule and make it skip an inode if we can't immediately lock its extent map tree. Reported-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com> Reported-by: Andrea Gelmini <andrea.gelmini@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CABXGCsMmmb36ym8hVNGTiU8yfUS_cGvoUmGCcBrGWq9OxTrs+A@mail.gmail.com/ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-07-08 15:42:44 +01:00
btrfs: only run the extent map shrinker from kswapd tasks Currently the extent map shrinker can be run by any task when attempting to allocate memory and there's enough memory pressure to trigger it. To avoid too much latency we stop iterating over extent maps and removing them once the task needs to reschedule. This logic was introduced in commit b3ebb9b7e92a ("btrfs: stop extent map shrinker if reschedule is needed"). While that solved high latency problems for some use cases, it's still not enough because with a too high number of tasks entering the extent map shrinker code, either due to memory allocations or because they are a kswapd task, we end up having a very high level of contention on some spin locks, namely: 1) The fs_info->fs_roots_radix_lock spin lock, which we need to find roots to iterate over their inodes; 2) The spin lock of the xarray used to track open inodes for a root (struct btrfs_root::inodes) - on 6.10 kernels and below, it used to be a red black tree and the spin lock was root->inode_lock; 3) The fs_info->delayed_iput_lock spin lock since the shrinker adds delayed iputs (calls btrfs_add_delayed_iput()). Instead of allowing the extent map shrinker to be run by any task, make it run only by kswapd tasks. This still solves the problem of running into OOM situations due to an unbounded extent map creation, which is simple to trigger by direct IO writes, as described in the changelog of commit 956a17d9d050 ("btrfs: add a shrinker for extent maps"), and by a similar case when doing buffered IO on files with a very large number of holes (keeping the file open and creating many holes, whose extent maps are only released when the file is closed). Reported-by: kzd <kzd@56709.net> Link: https://bugzilla.kernel.org/show_bug.cgi?id=219121 Reported-by: Octavia Togami <octavia.togami@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CAHPNGSSt-a4ZZWrtJdVyYnJFscFjP9S7rMcvEMaNSpR556DdLA@mail.gmail.com/ Fixes: 956a17d9d050 ("btrfs: add a shrinker for extent maps") CC: stable@vger.kernel.org # 6.10+ Tested-by: kzd <kzd@56709.net> Tested-by: Octavia Togami <octavia.togami@gmail.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-11 11:53:42 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: stop extent map shrinker if reschedule is needed The extent map shrinker can be called in a variety of contexts where we are under memory pressure, and of them is when a task is trying to allocate memory. For this reason the shrinker is typically called with a value of struct shrink_control::nr_to_scan that is much smaller than what we return in the nr_cached_objects callback of struct super_operations (fs/btrfs/super.c:btrfs_nr_cached_objects()), so that the shrinker does not take a long time and cause high latencies. However we can still take a lot of time in the shrinker even for a limited amount of nr_to_scan: 1) When traversing the red black tree that tracks open inodes in a root, as for example with millions of open inodes we get a deep tree which takes time searching for an inode; 2) Iterating over the extent map tree, which is a red black tree, of an inode when doing the rb_next() calls and when removing an extent map from the tree, since often that requires rebalancing the red black tree; 3) When trying to write lock an inode's extent map tree we may wait for a significant amount of time, because there's either another task about to do IO and searching for an extent map in the tree or inserting an extent map in the tree, and we can have thousands or even millions of extent maps for an inode. Furthermore, there can be concurrent calls to the shrinker so the lock might be busy simply because there is already another task shrinking extent maps for the same inode; 4) We often reschedule if we need to, which further increases latency. So improve on this by stopping the extent map shrinking code whenever we need to reschedule and make it skip an inode if we can't immediately lock its extent map tree. Reported-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com> Reported-by: Andrea Gelmini <andrea.gelmini@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CABXGCsMmmb36ym8hVNGTiU8yfUS_cGvoUmGCcBrGWq9OxTrs+A@mail.gmail.com/ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-07-08 15:42:44 +01:00
for-6.11-tag -----BEGIN PGP SIGNATURE----- iQIzBAABCgAdFiEE8rQSAMVO+zA4DBdWxWXV+ddtWDsFAmaVN3MACgkQxWXV+ddt WDtpIRAAl+1NjsEj8e5V/UYn8Jr06ujTOnrkR3PCTICxDHbUaMLkQEw21H0K/ogQ 3fOiEVpSlZOfKdYXtXaMQbC0jd/Af2eA10Uht96nAEjAtxu1uJ4cFZGu2meNdXZP xUioivJ/CElMPH2aluG6FaQvUTqmhrEr8tSoYbxzQmUd434q9kqqyjtw1tfzYDG1 VDn2f7ykhpB/8P0aoqgWSshWTmaCzG0GkuI28o1o0iZUIF/P9TKdzxlLRW6BVHE7 T2oGLEQjN1GQbCH75L4IeNJDkCBVfcDcbZkUDJ/ae4Pt/jJQTFY53YIP9wXFZQnd mdfHmK7Atpsk75ATftYSq+ENkbQ5fsuut5CD63u54gAqA4M1FncDXTAWS1Y30F76 P8juSCmsSy0o3gTflDIo/IMdntoh/JmncwwStF6oKzmyUZZzzarsqM8mc1P03ZNt 3ttlnbY7lC1TDAlD5J2wXE0INCT2pN+4C9IToWdRypeuLu6qrI7cQ0oylyp9OVQM t9umTXm0B6s1cyqEDjJf0xJZS/JTHYwu7S4EmAJwicgiLpOjABVTmO8021rVmDJy TAUu6yEhSsrTT6Dxm7/2Et1EEOKFF5hhsG1SiGD9oUIZK6B5+0waT+rbkEWl7osR 4/TAv2zX6tuCc7HIW0fQloM/6/Gyd5wcDVaQNDUzFA075uKstwY= =k5d3 -----END PGP SIGNATURE----- Merge tag 'for-6.11-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux Pull btrfs updates from David Sterba: "The highlights are new logic behind background block group reclaim, automatic removal of qgroup after removing a subvolume and new 'rescue=' mount options. The rest is optimizations, cleanups and refactoring. User visible features: - dynamic block group reclaim: - tunable framework to avoid situations where eager data allocations prevent creating new metadata chunks due to lack of unallocated space - reuse sysfs knob bg_reclaim_threshold (otherwise used only in zoned mode) for a fixed value threshold - new on/off sysfs knob "dynamic_reclaim" calculating the value based on heuristics, aiming to keep spare working space for relocating chunks but not to needlessly relocate partially utilized block groups or reclaim newly allocated ones - stats are exported in sysfs per block group type, files "reclaim_*" - this may increase IO load at unexpected times but the corner case of no allocatable block groups is known to be worse - automatically remove qgroup of deleted subvolumes: - adjust qgroup removal conditions, make sure all related subvolume data are already removed, or return EBUSY, also take into account setting of sysfs drop_subtree_threshold - also works in squota mode - mount option updates: new modes of 'rescue=' that allow to mount images (read-only) that could have been partially converted by user space tools - ignoremetacsums - invalid metadata checksums are ignored - ignoresuperflags - super block flags that track conversion in progress (like UUID or checksums) Core: - size of struct btrfs_inode is now below 1024 (on a release config), improved memory packing and other secondary effects - switch tracking of open inodes from rb-tree to xarray, minor performance improvement - reduce number of empty transaction commits when there are no dirty data/metadata - memory allocation optimizations (reduced numbers, reordering out of critical sections) - extent map structure optimizations and refactoring, more sanity checks - more subpage in zoned mode preparations or fixes - general snapshot code cleanups, improvements and documentation - tree-checker updates: more file extent ram_bytes fixes, continued - raid-stripe-tree update (not backward compatible): - remove extent encoding field from the structure, can be inferred from other information - requires btrfs-progs 6.9.1 or newer - cleanups and refactoring - error message updates - error handling improvements - return type and parameter cleanups and improvements" * tag 'for-6.11-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (152 commits) btrfs: fix extent map use-after-free when adding pages to compressed bio btrfs: fix bitmap leak when loading free space cache on duplicate entry btrfs: remove the BUG_ON() inside extent_range_clear_dirty_for_io() btrfs: move extent_range_clear_dirty_for_io() into inode.c btrfs: enhance compression error messages btrfs: fix data race when accessing the last_trans field of a root btrfs: rename the extra_gfp parameter of btrfs_alloc_page_array() btrfs: remove the extra_gfp parameter from btrfs_alloc_folio_array() btrfs: introduce new "rescue=ignoresuperflags" mount option btrfs: introduce new "rescue=ignoremetacsums" mount option btrfs: output the unrecognized super block flags as hex btrfs: remove unused Opt enums btrfs: tree-checker: add extra ram_bytes and disk_num_bytes check btrfs: fix the ram_bytes assignment for truncated ordered extents btrfs: make validate_extent_map() catch ram_bytes mismatch btrfs: ignore incorrect btrfs_file_extent_item::ram_bytes btrfs: cleanup the bytenr usage inside btrfs_extent_item_to_extent_map() btrfs: fix typo in error message in btrfs_validate_super() btrfs: move the direct IO code into its own file btrfs: pass a btrfs_inode to btrfs_set_prop() ...
2024-07-17 12:38:04 -07:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: always drop root->inodes lock before cond_resched() find_first_inode() and find_first_inode_to_shrink() lock root->inodes, then loop over them, occasionally skipping some inodes. When they skip an inode, they attempt to share the cpu/lock with cond_resched_lock(). However, that has a subtle problem associated with it. cond_resched_lock() only drops the lock if it needs to actually call schedule(). With CONFIG_PREEMPT_NONE, this means the full timeslice as detected at ticks. With 8+ cpus and default tunables, this is 2.8ms. So regardless of HZ, we will run for at least 2.8ms in this loop without dropping the lock, assuming it finds no suitable inodes. If HZ is small enough, it might be even worse as the tick granularity becomes bigger than the timeslice. The knock-on effect of this is that callers to btrfs_del_inode_from_root() like kswapd trying to shrink the inode slab or userspace threads calling evict() will spin on xa_lock(&root->inodes) for 2.8ms, so the extent map shrinker dominates the lock even though ostensibly it is intending to share it. This produces memory pressure as there is only one kswapd and it runs sequentially so it can get stuck in the inode slab shrinking. To fix it, simply replace cond_resched_lock() with an open coded variant which unconditionally does unlock/lock around cond_resched. Sharing the lock is decoupled from sharing the CPU, and all the users of the lock now share it fairly. I was able to reproduce this on test systems by producing a lot of empty files (to make a big root->inodes xarray), then producing memory pressure by reading large files larger than ram, triggering kswapd and the extent_map shrinker. The lock contention is visible with perf or lockstat. This patch also relieved a user-apparent bottleneck on a production system from the original report. Tested-by: Rik van Riel <riel@surriel.com> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2026-05-08 13:11:26 -07:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: only run the extent map shrinker from kswapd tasks Currently the extent map shrinker can be run by any task when attempting to allocate memory and there's enough memory pressure to trigger it. To avoid too much latency we stop iterating over extent maps and removing them once the task needs to reschedule. This logic was introduced in commit b3ebb9b7e92a ("btrfs: stop extent map shrinker if reschedule is needed"). While that solved high latency problems for some use cases, it's still not enough because with a too high number of tasks entering the extent map shrinker code, either due to memory allocations or because they are a kswapd task, we end up having a very high level of contention on some spin locks, namely: 1) The fs_info->fs_roots_radix_lock spin lock, which we need to find roots to iterate over their inodes; 2) The spin lock of the xarray used to track open inodes for a root (struct btrfs_root::inodes) - on 6.10 kernels and below, it used to be a red black tree and the spin lock was root->inode_lock; 3) The fs_info->delayed_iput_lock spin lock since the shrinker adds delayed iputs (calls btrfs_add_delayed_iput()). Instead of allowing the extent map shrinker to be run by any task, make it run only by kswapd tasks. This still solves the problem of running into OOM situations due to an unbounded extent map creation, which is simple to trigger by direct IO writes, as described in the changelog of commit 956a17d9d050 ("btrfs: add a shrinker for extent maps"), and by a similar case when doing buffered IO on files with a very large number of holes (keeping the file open and creating many holes, whose extent maps are only released when the file is closed). Reported-by: kzd <kzd@56709.net> Link: https://bugzilla.kernel.org/show_bug.cgi?id=219121 Reported-by: Octavia Togami <octavia.togami@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CAHPNGSSt-a4ZZWrtJdVyYnJFscFjP9S7rMcvEMaNSpR556DdLA@mail.gmail.com/ Fixes: 956a17d9d050 ("btrfs: add a shrinker for extent maps") CC: stable@vger.kernel.org # 6.10+ Tested-by: kzd <kzd@56709.net> Tested-by: Octavia Togami <octavia.togami@gmail.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-11 11:53:42 +01:00
btrfs: stop extent map shrinker if reschedule is needed The extent map shrinker can be called in a variety of contexts where we are under memory pressure, and of them is when a task is trying to allocate memory. For this reason the shrinker is typically called with a value of struct shrink_control::nr_to_scan that is much smaller than what we return in the nr_cached_objects callback of struct super_operations (fs/btrfs/super.c:btrfs_nr_cached_objects()), so that the shrinker does not take a long time and cause high latencies. However we can still take a lot of time in the shrinker even for a limited amount of nr_to_scan: 1) When traversing the red black tree that tracks open inodes in a root, as for example with millions of open inodes we get a deep tree which takes time searching for an inode; 2) Iterating over the extent map tree, which is a red black tree, of an inode when doing the rb_next() calls and when removing an extent map from the tree, since often that requires rebalancing the red black tree; 3) When trying to write lock an inode's extent map tree we may wait for a significant amount of time, because there's either another task about to do IO and searching for an extent map in the tree or inserting an extent map in the tree, and we can have thousands or even millions of extent maps for an inode. Furthermore, there can be concurrent calls to the shrinker so the lock might be busy simply because there is already another task shrinking extent maps for the same inode; 4) We often reschedule if we need to, which further increases latency. So improve on this by stopping the extent map shrinking code whenever we need to reschedule and make it skip an inode if we can't immediately lock its extent map tree. Reported-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com> Reported-by: Andrea Gelmini <andrea.gelmini@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CABXGCsMmmb36ym8hVNGTiU8yfUS_cGvoUmGCcBrGWq9OxTrs+A@mail.gmail.com/ Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-07-08 15:42:44 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: only run the extent map shrinker from kswapd tasks Currently the extent map shrinker can be run by any task when attempting to allocate memory and there's enough memory pressure to trigger it. To avoid too much latency we stop iterating over extent maps and removing them once the task needs to reschedule. This logic was introduced in commit b3ebb9b7e92a ("btrfs: stop extent map shrinker if reschedule is needed"). While that solved high latency problems for some use cases, it's still not enough because with a too high number of tasks entering the extent map shrinker code, either due to memory allocations or because they are a kswapd task, we end up having a very high level of contention on some spin locks, namely: 1) The fs_info->fs_roots_radix_lock spin lock, which we need to find roots to iterate over their inodes; 2) The spin lock of the xarray used to track open inodes for a root (struct btrfs_root::inodes) - on 6.10 kernels and below, it used to be a red black tree and the spin lock was root->inode_lock; 3) The fs_info->delayed_iput_lock spin lock since the shrinker adds delayed iputs (calls btrfs_add_delayed_iput()). Instead of allowing the extent map shrinker to be run by any task, make it run only by kswapd tasks. This still solves the problem of running into OOM situations due to an unbounded extent map creation, which is simple to trigger by direct IO writes, as described in the changelog of commit 956a17d9d050 ("btrfs: add a shrinker for extent maps"), and by a similar case when doing buffered IO on files with a very large number of holes (keeping the file open and creating many holes, whose extent maps are only released when the file is closed). Reported-by: kzd <kzd@56709.net> Link: https://bugzilla.kernel.org/show_bug.cgi?id=219121 Reported-by: Octavia Togami <octavia.togami@gmail.com> Link: https://lore.kernel.org/linux-btrfs/CAHPNGSSt-a4ZZWrtJdVyYnJFscFjP9S7rMcvEMaNSpR556DdLA@mail.gmail.com/ Fixes: 956a17d9d050 ("btrfs: add a shrinker for extent maps") CC: stable@vger.kernel.org # 6.10+ Tested-by: kzd <kzd@56709.net> Tested-by: Octavia Togami <octavia.togami@gmail.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-11 11:53:42 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: make the extent map shrinker run asynchronously as a work queue job Currently the extent map shrinker is run synchronously for kswapd tasks that end up calling the fs shrinker (fs/super.c:super_cache_scan()). This has some disadvantages and for some heavy workloads with memory pressure it can cause some delays and stalls that make a machine unresponsive for some periods. This happens because: 1) We can have several kswapd tasks on machines with multiple NUMA zones, and running the extent map shrinker concurrently can cause high contention on some spin locks, namely the spin locks that protect the radix tree that tracks roots, the per root xarray that tracks open inodes and the list of delayed iputs. This not only delays the shrinker but also causes high CPU consumption and makes the task running the shrinker monopolize a core, resulting in the symptoms of an unresponsive system. This was noted in previous commits such as commit ae1e766f623f ("btrfs: only run the extent map shrinker from kswapd tasks"); 2) The extent map shrinker's iteration over inodes can often be slow, even after changing the data structure that tracks open inodes for a root from a red black tree (up to kernel 6.10) to an xarray (kernel 6.10+). The transition to the xarray while it made things a bit faster, it's still somewhat slow - for example in a test scenario with 10000 inodes that have no extent maps loaded, the extent map shrinker took between 5ms to 8ms, using a release, non-debug kernel. Iterating over the extent maps of an inode can also be slow if have an inode with many thousands of extent maps, since we use a red black tree to track and search extent maps. So having the extent map shrinker run synchronously adds extra delay for other things a kswapd task does. So make the extent map shrinker run asynchronously as a job for the system unbounded workqueue, just like what we do for data and metadata space reclaim jobs. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-08-29 15:23:32 +01:00
btrfs: add a shrinker for extent maps Extent maps are used either to represent existing file extent items, or to represent new extents that are going to be written and the respective file extent items are created when the ordered extent completes. We currently don't have any limit for how many extent maps we can have, neither per inode nor globally. Most of the time this not too noticeable because extent maps are removed in the following situations: 1) When evicting an inode; 2) When releasing folios (pages) through the btrfs_release_folio() address space operation callback. However we won't release extent maps in the folio range if the folio is either dirty or under writeback or if the inode's i_size is less than or equals to 16M (see try_release_extent_mapping(). This 16M i_size constraint was added back in 2008 with commit 70dec8079d78 ("Btrfs: extent_io and extent_state optimizations"), but there's no explanation about why we have it or why the 16M value. This means that for buffered IO we can reach an OOM situation due to too many extent maps if either of the following happens: 1) There's a set of tasks constantly doing IO on many files with a size not larger than 16M, specially if they keep the files open for very long periods, therefore preventing inode eviction. This requires a really high number of such files, and having many non mergeable extent maps (due to random 4K writes for example) and a machine with very little memory; 2) There's a set tasks constantly doing random write IO (therefore creating many non mergeable extent maps) on files and keeping them open for long periods of time, so inode eviction doesn't happen and there's always a lot of dirty pages or pages under writeback, preventing btrfs_release_folio() from releasing the respective extent maps. This second case was actually reported in the thread pointed by the Link tag below, and it requires a very large file under heavy IO and a machine with very little amount of RAM, which is probably hard to happen in practice in a real world use case. However when using direct IO this is not so hard to happen, because the page cache is not used, and therefore btrfs_release_folio() is never called. Which means extent maps are dropped only when evicting the inode, and that means that if we have tasks that keep a file descriptor open and keep doing IO on a very large file (or files), we can exhaust memory due to an unbounded amount of extent maps. This is especially easy to happen if we have a huge file with millions of small extents and their extent maps are not mergeable (non contiguous offsets and disk locations). This was reported in that thread with the following fio test: $ cat test.sh #!/bin/bash DEV=/dev/sdj MNT=/mnt/sdj MOUNT_OPTIONS="-o ssd" MKFS_OPTIONS="" cat <<EOF > /tmp/fio-job.ini [global] name=fio-rand-write filename=$MNT/fio-rand-write rw=randwrite bs=4K direct=1 numjobs=16 fallocate=none time_based runtime=90000 [file1] size=300G ioengine=libaio iodepth=16 EOF umount $MNT &> /dev/null mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT fio /tmp/fio-job.ini umount $MNT Monitoring the btrfs_extent_map slab while running the test with: $ watch -d -n 1 'cat /sys/kernel/slab/btrfs_extent_map/objects \ /sys/kernel/slab/btrfs_extent_map/total_objects' Shows the number of active and total extent maps skyrocketing to tens of millions, and on systems with a short amount of memory it's easy and quick to get into an OOM situation, as reported in that thread. So to avoid this issue add a shrinker that will remove extents maps, as long as they are not pinned, and takes proper care with any concurrent fsync to avoid missing extents (setting the full sync flag while in the middle of a fast fsync). This shrinker is triggered through the callbacks nr_cached_objects and free_cached_objects of struct super_operations. The shrinker will iterate over all roots and over all inodes of each root, and keeps track of the last scanned root and inode, so that the next time it runs, it starts from that root and from the next inode. This is similar to what xfs does for its inode reclaim (implements those callbacks, and cycles through inodes by starting from where it ended last time). Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2024-04-15 17:09:26 +01:00