The compat version of settimeofday() uses '>' instead of '>=' when
validating tv_usec against USEC_PER_SEC, allowing the value 1000000 to pass
the check. After the subsequent conversion to nanoseconds (tv_nsec *=
NSEC_PER_USEC), this results in tv_nsec == NSEC_PER_SEC, which violates the
timespec invariant that tv_nsec must be strictly less than NSEC_PER_SEC.
The native settimeofday() was already fixed in commit ce4abda5e1 ("time:
Fix off-by-one in settimeofday() usec validation"), but the compat
counterpart was missed.
Fix it by using '>=' to reject tv_usec values outside the valid range [0,
USEC_PER_SEC - 1].
Fixes: 5e0fb1b57b ("y2038: time: avoid timespec usage in settimeofday()")
Signed-off-by: Wang Yan <wangyan01@kylinos.cn>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Cc: stable@vger.kernel.org
Link: https://patch.msgid.link/20260622103348.120255-1-wangyan01@kylinos.cn
Protected guest faults charge long term pins to the VM's mm. Teardown
can run later from file release, where current->mm may be unrelated.
Drop the charge from kvm->mm instead.
Fixes: 4e6e03f9ea ("KVM: arm64: Hook up reclaim hypercall to pkvm_pgtable_stage2_destroy()")
Signed-off-by: Bradley Morgan <include@grrlz.net>
Reviewed-by: Fuad Tabba <fuad.tabba@linux.dev>
Tested-by: Fuad Tabba <fuad.tabba@linux.dev>
Link: https://patch.msgid.link/20260621213155.6019-1-include@grrlz.net
Signed-off-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
__pkvm_memshare_page_req() constructs a fake DABT ESR_EL2 to exit to
the host without setting IL. The ESR has ISV=0, so IL must be 1 per the
architecture. The host does not read IL on this path, but the
constructed syndrome should still be architecturally valid.
Set ESR_ELx_IL.
Fixes: 03313efed5 ("KVM: arm64: Implement the MEM_SHARE hypercall for protected VMs")
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://patch.msgid.link/20260618121643.4105064-8-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
The FPAC syndrome constructed during nested ERET emulation does not set
IL. For FPAC (EC=0x1C), IL reflects the instruction length. ERET and
its authenticated variants are always A64 32-bit instructions, so IL
must be 1.
Fixes: 213b3d1ea1 ("KVM: arm64: nv: Handle ERETA[AB] instructions")
Suggested-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://patch.msgid.link/20260618121643.4105064-5-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
inject_abt64() derives IL from the triggering trap's instruction length
(kvm_vcpu_trap_il_is32bit()), but the IL of the injected abort is fixed
by its EC, not by the triggering instruction. The architecture mandates
IL=1 for Instruction Aborts unconditionally and for Data Aborts with
ISV=0, and this function never sets ISV (the FSC is always EXTABT or
SEA_TTW). For a 16-bit T32 trap (a 32-bit EL0 task under an AArch64 EL1
guest) the trap has IL=0, so the abort is injected with the wrong IL.
Set ESR_ELx_IL unconditionally.
Fixes: aa8eff9bfb ("arm64: KVM: fault injection into a guest")
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://patch.msgid.link/20260618121643.4105064-4-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
inject_undef64() derives IL from the triggering trap's instruction
length (kvm_vcpu_trap_il_is32bit()), but the IL of the injected
exception is fixed by its EC, not by the triggering instruction. The
architecture mandates IL=1 for EC=0 (Unknown) unconditionally, so the
conditional is wrong. The undef-injection paths are not reached from
16-bit instructions, so there is no functional change today, but the
logic should not rely on that.
Set ESR_ELx_IL unconditionally.
Fixes: aa8eff9bfb ("arm64: KVM: fault injection into a guest")
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://patch.msgid.link/20260618121643.4105064-3-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
inject_undef64() constructs an ESR with EC=0 (Unknown) but does not set
IL. The architecture mandates IL=1 for EC=0 unconditionally (ARM DDI
0487, ESR_ELx.IL description), so the injected syndrome is one that
conforming hardware cannot produce.
Set ESR_ELx_IL in the constructed syndrome.
Fixes: e5d40a5a97 ("KVM: arm64: pkvm: Add a generic synchronous exception injection primitive")
Reported-by: sashiko <sashiko@sashiko.dev>
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://patch.msgid.link/20260618121643.4105064-2-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
KVM is unlikely to resolve an unexpected VNCR abort, meaning that
returning to the guest will likely leave the vCPU stuck in an abort
loop. Bug the VM and exit to userspace instead.
Signed-off-by: Oliver Upton <oupton@kernel.org>
Link: https://patch.msgid.link/20260618234207.1063941-6-oupton@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
When constructing an L1 VNCR mapping, KVM unconditionally uses cacheable
memory attributes, even if the underlying PFN isn't memory. This gets
particularly hairy if the endpoint doesn't support cacheable memory
attributes, potentially throwing an SError on writeback...
While KVM does permit cacheable memory attributes on certain PFNMAP
VMAs, kvm_translate_vncr() isn't currently grabbing the VMA. So do the
simpler thing for now and just reject everything that isn't memory.
Cc: stable@vger.kernel.org
Fixes: 2a359e0725 ("KVM: arm64: nv: Handle mapping of VNCR_EL2 at EL2")
Signed-off-by: Oliver Upton <oupton@kernel.org>
Link: https://patch.msgid.link/20260618234207.1063941-5-oupton@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
KVM faults in the VNCR page with FOLL_WRITE whenever the guest aborts
for a write, similar to how a regular stage-2 mapping is handled. It is
entirely possible that the guest reads from the VNCR before writing to
it, in which case the PFN could only be read-only.
Invalidate the VNCR TLB and re-fetch the translation upon taking a VNCR
abort, allowing the host mapping to be faulted in for write the second
time around. Interestingly enough, this also satisfies the ordering
requirements of FEAT_ETS2/3 between descriptor updates and MMU faults.
Cc: stable@vger.kernel.org
Fixes: 2a359e0725 ("KVM: arm64: nv: Handle mapping of VNCR_EL2 at EL2")
Reported-by: Sashiko <sashiko-bot@kernel.org>
Signed-off-by: Oliver Upton <oupton@kernel.org>
Link: https://patch.msgid.link/20260618234207.1063941-4-oupton@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
kvm_handle_vncr_abort() assumes that s1_walk_result conveys an abort
when kvm_translate_vncr() returns -EFAULT. This is not always the case
as it's possible to encounter 'late' failures on the output of S1
translation, e.g. a GFN outside of the memslots.
Fix it by preparing an external abort before returning from
kvm_translate_vncr(). Get rid of the BUG_ON() in the fault injection
path while at it.
Cc: stable@vger.kernel.org
Fixes: 2a359e0725 ("KVM: arm64: nv: Handle mapping of VNCR_EL2 at EL2")
Signed-off-by: Oliver Upton <oupton@kernel.org>
Link: https://patch.msgid.link/20260618234207.1063941-3-oupton@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
KVM currently maps the L1 VNCR into the host stage-1 by relying entirely
on the permissions of the guest stage-1. At the same time, it is
entirely possible that the backing PFN is read-only (e.g. RO memslot),
meaning that the L1 VNCR should use at most a read-only mapping.
Cache the writability of the PFN in the VNCR TLB and use it to constrain
the resulting fixmap permissions. Promote VNCR permission faults to an
SEA in the case where the guest attempts to write to a read-only
endpoint. Conveniently, this also plugs a page leak found by Sashiko [*]
resulting from the early return for a read-only PFN.
Cc: stable@vger.kernel.org
Fixes: 2a359e0725 ("KVM: arm64: nv: Handle mapping of VNCR_EL2 at EL2")
Link: https://lore.kernel.org/kvm/20260608082603.16AEC1F00893@smtp.kernel.org/
Signed-off-by: Oliver Upton <oupton@kernel.org>
Link: https://patch.msgid.link/20260618234207.1063941-2-oupton@kernel.org
Signed-off-by: Marc Zyngier <maz@kernel.org>
Passing a maximum as 64-bit type to drm_sysfb_get_validated_int0()
can truncate the value to 32 bits. Use drm_sysfb_get_validated_size0(),
which uses 64-bit arithmetics. Then test the returned stride against
the limits of int to avoid truncations in the returned value. A valid
stride is in the range of [1, INT_MAX] inclusive.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/dri-devel/20260617114016.5A5991F000E9@smtp.kernel.org/
Fixes: 32ae90c66f ("drm/sysfb: Add efidrm for EFI displays")
Fixes: a84eb6abe2 ("drm/sysfb: Add vesadrm for VESA displays")
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: dri-devel@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v6.16+
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patch.msgid.link/20260618084327.46567-5-tzimmermann@suse.de
Change the return type of drm_sysfb_get_visible_size() to s64 so
that it returns a possible errno code from _get_validated_size0().
Fix callers to handle the errno code.
The currently returned unsigned type converts an errno code to a
very large size value, which drivers interpret as visible size of
the system framebuffer. Later efforts to reserve the framebuffer
resource fail.
The bug has been present since efidrm and vesadrm got merged. It
was then part of each driver.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Fixes: 32ae90c66f ("drm/sysfb: Add efidrm for EFI displays")
Fixes: a84eb6abe2 ("drm/sysfb: Add vesadrm for VESA displays")
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: dri-devel@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v6.16+
Link: https://patch.msgid.link/20260618084327.46567-4-tzimmermann@suse.de
Calculating the visible size of the system framebuffer can result in
truncation of the result. The calculation uses 32-bit arithmetics,
which can overflow if the values for height and stride are large. Fix
the issue by multiplying with mul_u32_u32().
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Fixes: 32ae90c66f ("drm/sysfb: Add efidrm for EFI displays")
Fixes: a84eb6abe2 ("drm/sysfb: Add vesadrm for VESA displays")
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/dri-devel/20260617114027.1F2A71F000E9@smtp.kernel.org/
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: dri-devel@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v6.16+
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patch.msgid.link/20260618084327.46567-3-tzimmermann@suse.de
Only return the actually visible size of the system framebuffer in
drm_sysfb_get_visible_size_si(). Drivers use this size value for
reserving access to framebuffer memory. Increasing the value can
make later attempts to do so fail.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Fixes: 32ae90c66f ("drm/sysfb: Add efidrm for EFI displays")
Fixes: a84eb6abe2 ("drm/sysfb: Add vesadrm for VESA displays")
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: dri-devel@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v6.16+
Link: https://patch.msgid.link/20260618084327.46567-2-tzimmermann@suse.de
The Yoga 7 16IAP7 (board LNVNB161216, codec SSID 17aa:386a) has pin
complex 0x17 (bass speakers) wrongly reported as unconnected, causing
only one of four speaker pins (0x14) to be configured and resulting in
mono/tinny audio.
SOF corrupts the PCI subsystem ID to 17aa:0000, preventing SND_PCI_QUIRK
from matching. HDA_CODEC_QUIRK is used instead, which matches against
codec->core.subsystem_id read directly from the HDA codec register and
unaffected by the SOF bug.
Applies ALC287_FIXUP_YOGA9_14IAP7_BASS_SPK_PIN, the same fixup used for
the Yoga 7 14IAL7, which corrects pin 0x17's default configuration and
enables both speaker pairs.
Signed-off-by: Chris Aherin <chrisaherin@gmail.com>
Link: https://patch.msgid.link/20260622001210.20553-1-chrisaherin@gmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
drm_parse_tiled_block() casts the DisplayID block to a
struct displayid_tiled_block and reads the full fixed layout up to
tile->topology_id[7] without checking block->num_bytes. The DisplayID
iterator only validates the declared payload length, so a crafted EDID
can advertise a tiled-display block (tag DATA_BLOCK_TILED_DISPLAY, or
DATA_BLOCK_2_TILED_DISPLAY_TOPOLOGY for v2.0) with a small num_bytes at
the end of a DisplayID extension. The read then runs past the end of the
exact-sized kmemdup()'d EDID allocation, a heap out-of-bounds read.
Reject blocks shorter than the spec's 22-byte tiled payload before
reading the fixed struct, as drm_parse_vesa_mso_data() already does.
BUG: KASAN: slab-out-of-bounds in drm_edid_connector_update
Read of size 2 at addr ffff888010077700 by task exploit/147
dump_stack_lvl (lib/dump_stack.c:94 ...)
print_report (mm/kasan/report.c:378 ...)
kasan_report (mm/kasan/report.c:595)
drm_edid_connector_update (drivers/gpu/drm/drm_edid.c:7581)
bochs_connector_helper_get_modes (drivers/gpu/drm/tiny/bochs.c:574)
drm_helper_probe_single_connector_modes (drivers/gpu/drm/drm_probe_helper.c:426)
status_store (drivers/gpu/drm/drm_sysfs.c:219)
...
vfs_write (fs/read_write.c:595 fs/read_write.c:688)
ksys_write (fs/read_write.c:740)
Fixes: 40d9b043a8 ("drm/connector: store tile information from displayid (v3)")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patch.msgid.link/20260615184737.899892-1-xmei5@asu.edu
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Fix build warning by adding the missing structure name and
description to the kernel-doc comment block.
Signed-off-by: Igor Putko <igorpetindev@gmail.com>
Link: https://patch.msgid.link/20260618155626.18751-2-igorpetindev@gmail.com
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
gpiochip_set_multiple() falls back to setting lines one by one when the
chip does not provide set_multiple(). If the fallback path receives an
empty mask, the loop is skipped and ret is returned without being
initialized.
Initialize ret to 0 so an empty mask is treated as a successful no-op.
Fixes: 9b40731275 ("gpiolib: rework the wrapper around gpio_chip::set_multiple()")
Signed-off-by: Ruoyu Wang <ruoyuw560@gmail.com>
Acked-by: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Link: https://patch.msgid.link/20260620155319.79994-1-ruoyuw560@gmail.com
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
The pwrseq core takes three locks in consistent order everywhere:
pwrseq_sem -> pwrseq->rw_lock -> pwrseq->state_lock
pwrseq_get() -> pwrseq_match_device() takes pwrseq_sem for reading, then
rw_lock for reading. pwrseq_power_on()/pwrseq_power_off() take rw_lock
for reading and then state_lock.
pwrseq_device_unregister() is the only exception, it takes: state_lock,
then rw_lock for writing and finally pwrseq_sem for writing. This created
two potential ABBA deadlock situations that sashiko pointed out.
- pwrseq_power_on/off() take rw_lock for reading then state_lock, while
pwrseq_unregister() takes state_lock then rw_lock for writing
- pwrseq_get() takes pwrseq_sem for reading then rw_lock for reading,
while pwrseq_unregister() takes rw_lock for writing then pwrseq_sem
for writing
Reorder the unregister path to taking pwrseq_sem for writing -> rw_lock
for writing and drop the state_lock entirely. This is safe as
enable_count is only ever written under rw_lock held for read (via
pwrseq_unit_enable()/disable(), reached only from pwrseq_power_on/off()),
so holding rw_lock for writing already excludes every other writer and
reader and the active-users WARN() stays race-free without state_lock.
Fixes: 249ebf3f65 ("power: sequencing: implement the pwrseq core")
Closes: https://sashiko.dev/#/patchset/20260616151049.1705503-1-vulab%40iscas.ac.cn
Link: https://patch.msgid.link/20260618-pwrseq-abba-deadlock-v1-1-943a3fd81c06@oss.qualcomm.com
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
Sort the entries in pwrseq_m2_pci_ids[] by device ID in ascending order:
0x1103 (WCN6855) before 0x1107 (WCN7850).
Fixes: 2abcfdd91e ("power: sequencing: pcie-m2: Add PCI ID 0x1103 for WCN6855 Bluetooth")
Reviewed-by: Manivannan Sadhasivam <mani@kernel.org>
Signed-off-by: Wei Deng <wei.deng@oss.qualcomm.com>
Link: https://patch.msgid.link/20260617143055.820096-1-wei.deng@oss.qualcomm.com
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
pwrseq_debugfs_seq_next() declares 'next' with __free(put_device),
which causes put_device() to be called on the returned pointer when
the variable goes out of scope. This results in a use-after-free
since the seq_file framework receives a pointer whose reference has
already been dropped.
Simply removing __free(put_device) would fix the UAF but would leak
the reference acquired by bus_find_next_device(), as stop() only
calls up_read(&pwrseq_sem) and never releases the device reference.
Fix this by making the reference counting consistent across all
seq_file callbacks, matching the standard pattern used by PCI and
SCSI:
- start(): use get_device() so it returns a referenced pointer.
- next(): explicitly put_device(curr) to release the previous
device's reference (no NULL check needed - the seq_file framework
only calls next() while the previous return was non-NULL).
- stop(): put_device(data) to release the last iterated device's
reference, with a NULL guard since stop() may be called with NULL
when start() returned NULL or next() reached end-of-sequence.
Cc: stable@vger.kernel.org
Fixes: 249ebf3f65 ("power: sequencing: implement the pwrseq core")
Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
Link: https://patch.msgid.link/20260616151049.1705503-1-vulab@iscas.ac.cn
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
My Grace host started to show this warning:
WARNING: drivers/soc/tegra/fuse/tegra-apbmisc.c:120 at tegra_read_straps
tegra30_fuse_add_randomness
tegra30_fuse_init
tegra_fuse_probe
tegra_read_straps() warns when the static "chipid" cache is still zero,
using it as a proxy for "APBMISC has been initialised". However chipid
is only ever populated lazily by tegra_read_chipid() when it reads the
APBMISC register.
Guard on apbmisc_base instead, which is set unconditionally in
tegra_init_apbmisc_resources() for all platforms and is already the
sentinel used by tegra_read_chipid().
Fixes: 8b8ee2e56f ("soc/tegra: Use ARM SMCCC to get chip ID, revision, and platform info")
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Thierry Reding <treding@nvidia.com>
When build testing on ARM without the PMC driver, the other drivers
fail to link:
ld.lld: error: undefined symbol: tegra_pmc_core_domain_state_synced
>>> referenced by regulators-tegra30.c
>>> drivers/soc/tegra/regulators-tegra30.o:(tegra30_regulator_balance_voltage) in archive vmlinux.a
>>> referenced by regulators-tegra20.c
>>> drivers/soc/tegra/regulators-tegra20.o:(tegra20_core_rtc_update) in archive vmlinux.a
Adapt the checks in the header to cover both cases on other architectures
and without PMC.
Fixes: 8318af5dd2 ("soc/tegra: pmc: Move legacy code behind CONFIG_ARM guard")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Thierry Reding <treding@nvidia.com>
The resource freed at the 'put_aux' label is "sor->aux->dev".
However, this resource is taken after devm_tegra_pmc_get(), so there is no
point to release it in this error handling path.
This is harmless because put_device() will be called with a NULL pointer,
but this is confusing.
So, fix the logic and return directly.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Thierry Reding <treding@nvidia.com>
EROFS over fscache was introduced to provide image lazy pulling
functionality. After the feature landed, the fscache subsystem made
netfs a new hard dependency, which is unexpected for a local filesystem
and has an kernel-defined caching hierarchy which could be inflexible
compared to the fanotify pre-content hooks. Therefore, this feature has
been deprecated for almost two years.
As EROFS file-backed mounts and fanotify pre-content hooks both upstream
for a while and already providing equivalent functionality (erofs-utils
has supported fanotify pre-content hooks), let's remove the fscache
backend now.
The main application of this feature is Nydus [1], and they plan to move
to use fanotify pre-content hooks in the near future too.
I hope this patch can be merged into Linux 7.2, which is also motivated
by newly found implementation issues [2][3] that are not worth
investigating given the deprecation and limited development resources.
The associated fscache/cachefiles cleanup patch will follow separately
through the vfs tree (netfs) later: it seems fine since the codebase is
isolated by CONFIG_CACHEFILES_ONDEMAND.
[1] https://github.com/dragonflyoss/nydus/blob/v2.1.0/docs/nydus-fscache.md
[2] https://github.com/dragonflyoss/nydus/pull/1824
[3] https://lore.kernel.org/r/20260619135800.1594811-1-michael.bommarito@gmail.com
Acked-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Amery Hung says:
====================
Fix effective prog array indexing with BPF_F_PREORDER
This patchset fixes a cgroup effective-array indexing bug where
replace_effective_prog() and purge_effective_progs() used a linear hlist
position that doesn't match the array layout when BPF_F_PREORDER
programs are present, corrupting the array on link update and risking a
use-after-free in the detach fallback. It computes the slot via a shared
effective_prog_pos() helper and adds a cgroup_preorder selftest.
Changelog
v1 -> v2:
- Also fix purge_effective_progs(), in addition to
replace_effective_prog() (Sashiko).
- selftest: Set err on bpf_link_create() failure so the failure is
reported to the caller (Sashiko).
====================
Link: https://patch.msgid.link/20260619063520.2690547-1-ameryhung@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add a cgroup_preorder case that attaches a normal and a BPF_F_PREORDER
program to a cgroup (effective order [2, 1]), then replaces the normal
link's program via bpf_link_update() and checks the effective order
becomes [2, 3] — i.e. only the non-preorder slot changes. Without the
replace_effective_prog() fix the array is corrupted and the order is
wrong.
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20260619063520.2690547-3-ameryhung@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
replace_effective_prog() and purge_effective_progs() located the slot in
the effective array by walking the program hlist and counting entries
linearly. That count does not match the array layout: compute_effective_
progs() places BPF_F_PREORDER programs at the front (ancestor cgroup
first, attach order within a cgroup) and the rest after them (descendant
cgroup first). So when a preorder program is present, the linear hlist
position no longer equals the program's index in the effective array.
For replace_effective_prog() (bpf_link_update()) this overwrote the
wrong slot, corrupting the effective order. For purge_effective_progs(),
it could dummy out a slot belonging to a different program and leave the
detached program in the array while bpf_prog_put() drops its reference,
i.e. a use-after-free.
Fix both by replaying compute_effective_progs()'s placement (including
the per-cgroup preorder reversal) in a shared effective_prog_pos()
helper. Identify the entry by its struct bpf_prog_list pointer rather
than by (prog, link) value, so the lookup resolves to exactly the
attachment the syscall selected even when the same bpf_prog is attached
to several cgroups in the hierarchy.
Fixes: 4b82b181a2 ("bpf: Allow pre-ordering for bpf cgroup progs")
Signed-off-by: Amery Hung <ameryhung@gmail.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20260619063520.2690547-2-ameryhung@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When struct prog_assoc_struct_ops was added,
BPF_PROG_ASSOC_STRUCT_OPS_LAST_FIELD referenced prog_fd instead of the
actual last field, flags.
Fixes: b5709f6d26 ("bpf: Support associating BPF program with struct_ops")
Signed-off-by: Thiébaud Weksteen <tweek@google.com>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20260618040934.4113938-1-tweek@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
bpf_ipv4_fib_lookup() and bpf_ipv6_fib_lookup() build the flow key on
the stack with a bare "struct flowi4 fl4;" / "struct flowi6 fl6;" and
fill it field by field, but never set flowi4_l3mdev / flowi6_l3mdev.
On the non-DIRECT path the lookup goes through the fib rules whenever the
netns has custom rules, which a VRF installs:
bpf_ipv4_fib_lookup() -> fib_lookup() -> __fib_lookup()
-> l3mdev_update_flow() reads !fl->flowi_l3mdev
-> fib_rules_lookup() -> fib_rule_match()
-> l3mdev_fib_rule_match() uses fl->flowi_l3mdev
l3mdev_update_flow() resolves the l3mdev master from the ingress device
only while the field is still zero. Left at a nonzero stack value the
resolution is skipped, and l3mdev_fib_rule_match() then tests that value
as an ifindex, so the VRF master is not resolved and the rule fails to
match: an ingress enslaved to a VRF can fail to select its table. FIB
rules matching on an L3 master device (l3mdev_fib_rule_iif_match()/
_oif_match()) read the same value, so an "ip rule iif/oif <vrf>"
mismatches the same way.
Zero-initialize the whole flow struct rather than adding one more
field assignment, so any flowi field added later is covered too.
ip_route_input_slow() likewise zeroes the field before its input lookup.
CONFIG_INIT_STACK_ALL_ZERO masks this by default, but it depends on
compiler support (CC_HAS_AUTO_VAR_INIT_ZERO), so INIT_STACK_NONE builds,
including older toolchains that fall back to it, are exposed. Built with
INIT_STACK_ALL_PATTERN, a plain bpf_fib_lookup (no VLAN, no DIRECT) over a
VRF slave whose destination is routed only in the VRF table returns
BPF_FIB_LKUP_RET_NOT_FWDED, and resolves with this patch. On the default
config the lookup succeeds either way, so ordinary testing does not catch
the bug.
Fixes: 40867d74c3 ("net: Add l3mdev index to flow struct and avoid oif reset for port devices")
Signed-off-by: Avinash Duduskar <avinash.duduskar@gmail.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/r/20260617224719.1428599-1-avinash.duduskar@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
bpftool cgroup show and tree call libbpf_find_kernel_btf() to
resolve attach_btf names, but never release the returned BTF object.
For cgroup tree, do_show_tree_fn() is called once for each cgroup
visited by nftw(). When more than one cgroup has attached programs,
each callback overwrites btf_vmlinux with a new object and loses the
previous allocation.
Load vmlinux BTF only once during a tree walk and release it when
cgroup show or tree completes. Reset btf_vmlinux_id at the same time
so batch mode starts with clean state.
Fixes: 596f5fb2ea ("bpftool: implement cgroup tree for BPF_LSM_CGROUP")
Signed-off-by: Yichong Chen <chenyichong@uniontech.com>
Reviewed-by: Quentin Monnet <qmo@kernel.org>
Link: https://lore.kernel.org/r/24357C69B4405079+20260617090117.280222-1-chenyichong@uniontech.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
llvm commit [1] allowed attaching type tag records to modifier BTF
records. This is useful for using typedefs that encompass a base type
and a type tag, e.g.:
typedef struct rbtree __arena rbtree_t;
Modify btf_check_type_tags() so that it allows this sequence of records.
The function now only checks for record loops in BTF modifier record
chains. Rename to btf_check_modifier_chain_length to reflect this.
Also expand the BTF modifier traversal code to take into account that
type record can be interleaved with other modifier records. In effect
this means traversing all modifiers to collect the type tags.
Also modify existing selftests to now accept modifier records (const,
typedef) that point to type tag records.
[1] https://github.com/llvm/llvm-project/pull/203089
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260616061454.7869-1-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When BPF_WRITE goes through a PTR_TO_BTF_ID register, check_ptr_to_btf_access()
delegates to env->ops->btf_struct_access(). Most implementations
(bpf_scx_btf_struct_access, tc_cls_act_btf_struct_access, etc.) return
-EACCES for disallowed fields without logging anything, so the verifier
rejects the program with an empty message. For example a scx program doing
1: R1=trusted_ptr_task_struct()
...
4: (7b) *(u64 *)(r1 +0) = r2
verification time 83 usec
the program is rejected
leaves the user guessing which field is off-limits.
Emit verbose message.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20260615232146.5491-1-alexei.starovoitov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch is a follow up to recent implementation of
stack_map_get_build_id_offset_sleepable() [1].
stack_map_get_build_id_offset() and its sleepable variant each cached
only the last successfully resolved VMA, with separate bookkeeping in
each function. A run of IPs in a VMA with no usable build ID will
repeat the lookup for every frame: find_vma() in the non-sleepable
path, a VMA lock and a blocking build_id_parse_file() in the sleepable.
Factor the per-call cache into a shared struct stack_map_build_id_cache
with two independent slots [2][3], used by both functions:
* resolved - last VMA that produced a build ID (file, build_id and
range), reused to skip the lookup and the parse;
* unresolved - last VMA with no usable build ID (range only), reused to
emit a raw IP without another lookup or parse.
Keeping the slots independent means a build-ID-less VMA no longer evicts
the last resolved build ID, so a trace alternating between a binary and a
region without one stops re-resolving the binary on every return.
The shared lookup tests [vm_start, vm_end), matching the sleepable path;
the non-sleepable path previously reused a build ID for ip == vm_end
(range_in_vma() is inclusive) and now re-resolves it correctly.
[1] https://lore.kernel.org/bpf/20260525223948.1920986-1-ihor.solodrai@linux.dev/
[2] https://lore.kernel.org/bpf/CAEf4Bza2fRDGhLQoPE-EzM7F34xaEJfi5Exmxb-iWVUN3F06=g@mail.gmail.com/
[3] https://lore.kernel.org/bpf/CAEf4BzZXJFr=1iiVx937ht=4PYQkQHg=eFk810zhMDzXQG3ihw@mail.gmail.com/
Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Ihor Solodrai <ihor.solodrai@linux.dev>
Link: https://lore.kernel.org/r/20260615195536.1065107-1-ihor.solodrai@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
For non-linear test_run output, bpf_test_finish() derives the linear
data copy length from copy_size - frag_size. This only matches the
linear data length when copy_size is the full packet size.
When userspace provides a short data_out buffer, copy_size is clamped to
that buffer size. If copy_size is smaller than frag_size, the computed
length becomes negative and bpf_test_finish() returns -ENOSPC before
copying the packet prefix or updating data_size_out.
Compute the linear data length from the packet layout instead, and clamp
the linear copy length to copy_size. This preserves the expected
partial-copy semantics: return -ENOSPC, copy the packet prefix that fits
in data_out, and report the full packet length through data_size_out.
Fixes: 7855e0db15 ("bpf: test_run: add xdp_shared_info pointer in bpf_test_finish signature")
Signed-off-by: Sun Jian <sun.jian.kdev@gmail.com>
Acked-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/20260617093557.63880-2-sun.jian.kdev@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Nuoqi Gui says:
====================
bpf: Fix stack slot index for Spectre v4 nospec checks
check_stack_write_fixed_off() uses one byte-indexing scheme when checking
whether a fixed-offset stack write needs Spectre v4 sanitization, and another
scheme when recording the write into slot_type[].
For sub-8-byte writes this can make the sanitization check look at bytes that
are not overwritten by the write. A zeroed lower half-slot followed by a write
to the upper half-slot can therefore miss the nospec barrier for the second
write.
Use the same stack-byte index for the sanitization check and the slot update,
and add a focused verifier selftest that expects both half-slot writes to emit
nospec through the unprivileged loader lane.
Bounded impact: this fixes verifier/JIT Spectre v4 mitigation emission for a
fixed-offset stack-write corner case. No architectural verifier memory-safety
bypass, exploit chain, CVE, embargo, or security escalation is claimed.
Fixes: 2039f26f3a ("bpf: Fix leakage due to insufficient speculative store bypass mitigation")
Signed-off-by: Nuoqi Gui <gnq25@mails.tsinghua.edu.cn>
---
Changes in v3:
- selftests/bpf: drop the stray space in the __xlated_unpriv stack-store
expectations ("(r10 - 4)"/"(r10 - 8)" -> "(r10 -4)"/"(r10 -8)")
- Link to v2: https://lore.kernel.org/bpf/20260618-f01-11-stack-nospec-slot-index-v2-0-ede9495359b6@mails.tsinghua.edu.cn/
Changes in v2:
- drop __caps_unpriv(CAP_BPF) from the selftest
- fix selftest style
- use Fixes: 2039f26f3a per review
- Link to v1: https://lore.kernel.org/bpf/20260617-f01-11-stack-nospec-slot-index-v1-0-e3a080b0cd7e@mails.tsinghua.edu.cn/
====================
Link: https://patch.msgid.link/20260618-f01-11-stack-nospec-slot-index-v3-0-780297041721@mails.tsinghua.edu.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add a verifier test for the fixed-offset stack write case where two 4-byte
stores initialize opposite halves of the same stack slot.
The test runs through the unprivileged loader lane and expects both
half-slot writes to emit nospec in the translated program.
Acked-by: Luis Gerhorst <luis.gerhorst@fau.de>
Signed-off-by: Nuoqi Gui <gnq25@mails.tsinghua.edu.cn>
Link: https://lore.kernel.org/r/20260618-f01-11-stack-nospec-slot-index-v3-2-780297041721@mails.tsinghua.edu.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
check_stack_write_fixed_off() computes the byte slot for a fixed-offset
stack write as -off - 1, and records each written byte in slot_type[] with
(slot - i) % BPF_REG_SIZE.
The Spectre v4 sanitization pre-check uses slot_type[i] instead. For a
4-byte write at fp-8 after the lower half of fp-8 has been zeroed, the
pre-check scans bytes 0..3 and sees STACK_ZERO while the actual write updates
bytes 7..4. That can leave the second half-slot write without nospec_result
even though the bytes being overwritten still require sanitization.
Use the same slot index in the sanitization pre-check that the write path uses
when updating slot_type[].
Fixes: 2039f26f3a ("bpf: Fix leakage due to insufficient speculative store bypass mitigation")
Acked-by: Luis Gerhorst <luis.gerhorst@fau.de>
Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Nuoqi Gui <gnq25@mails.tsinghua.edu.cn>
Link: https://lore.kernel.org/r/20260618-f01-11-stack-nospec-slot-index-v3-1-780297041721@mails.tsinghua.edu.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Named initializers are better readable and more robust to changes of the
struct definition. This robustness is relevant for a planned change to
struct platform_device_id replacing .driver_data by an anonymous union.
While touching these arrays unify spacing and usage of commas.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Reviewed-by: Matti Vaittinen <mazziesaccount@gmail.com>
Reviewed-by: Linus Walleij <linusw@kernel.org>
Acked-by: Karel Balej <balejk@matfyz.cz> # for Marvell 88PM886
Link: https://patch.msgid.link/d14b9076b2c7703708bcc5cc35f339cd97fc10cd.1779950275.git.u.kleine-koenig@baylibre.com
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Instead of hiding the rtc ops for the only supported device behind an
abstraction for multi-device support, hardcode the used ops which gets rid
of the need to call platform_get_device_id and two casts.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Reviewed-by: Linus Walleij <linusw@kernel.org>
Link: https://patch.msgid.link/a909d3c59d00756130ac16051ceedbec0ce9cec7.1779950275.git.u.kleine-koenig@baylibre.com
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
The two drivers explicitly set the .driver_data member of struct
platform_device_id to zero without relying on that value. Drop this
unused assignments.
While touching these array unify spacing, usage of commas and use named
initializers for .name.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Reviewed-by: Tzung-Bi Shih <tzungbi@kernel.org>
Link: https://patch.msgid.link/9ec7a174605a17dd19c011ee2253de28d09b02bd.1779950275.git.u.kleine-koenig@baylibre.com
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
