2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:15 -07:00
|
|
|
|
2019-06-03 20:19:52 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:06 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
2026-07-24 00:26:16 +00:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:05 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:06 -07:00
|
|
|
|
2019-05-24 14:43:05 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2024-03-06 13:49:20 +01:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:07 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
2026-07-24 00:26:16 +00:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:05 -07:00
|
|
|
|
2019-05-24 14:43:06 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-03 16:18:13 -07:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:17 +01:00
|
|
|
|
net: nexthop: Increase weight to u16
In CLOS networks, as link failures occur at various points in the network,
ECMP weights of the involved nodes are adjusted to compensate. With high
fan-out of the involved nodes, and overall high number of nodes,
a (non-)ECMP weight ratio that we would like to configure does not fit into
8 bits. Instead of, say, 255:254, we might like to configure something like
1000:999. For these deployments, the 8-bit weight may not be enough.
To that end, in this patch increase the next hop weight from u8 to u16.
Increasing the width of an integral type can be tricky, because while the
code still compiles, the types may not check out anymore, and numerical
errors come up. To prevent this, the conversion was done in two steps.
First the type was changed from u8 to a single-member structure, which
invalidated all uses of the field. This allowed going through them one by
one and audit for type correctness. Then the structure was replaced with a
vanilla u16 again. This should ensure that no place was missed.
The UAPI for configuring nexthop group members is that an attribute
NHA_GROUP carries an array of struct nexthop_grp entries:
struct nexthop_grp {
__u32 id; /* nexthop id - must exist */
__u8 weight; /* weight of this nexthop */
__u8 resvd1;
__u16 resvd2;
};
The field resvd1 is currently validated and required to be zero. We can
lift this requirement and carry high-order bits of the weight in the
reserved field:
struct nexthop_grp {
__u32 id; /* nexthop id - must exist */
__u8 weight; /* weight of this nexthop */
__u8 weight_high;
__u16 resvd2;
};
Keeping the fields split this way was chosen in case an existing userspace
makes assumptions about the width of the weight field, and to sidestep any
endianness issues.
The weight field is currently encoded as the weight value minus one,
because weight of 0 is invalid. This same trick is impossible for the new
weight_high field, because zero must mean actual zero. With this in place:
- Old userspace is guaranteed to carry weight_high of 0, therefore
configuring 8-bit weights as appropriate. When dumping nexthops with
16-bit weight, it would only show the lower 8 bits. But configuring such
nexthops implies existence of userspace aware of the extension in the
first place.
- New userspace talking to an old kernel will work as long as it only
attempts to configure 8-bit weights, where the high-order bits are zero.
Old kernel will bounce attempts at configuring >8-bit weights.
Renaming reserved fields as they are allocated for some purpose is commonly
done in Linux. Whoever touches a reserved field is doing so at their own
risk. nexthop_grp::resvd1 in particular is currently used by at least
strace, however they carry an own copy of UAPI headers, and the conversion
should be trivial. A helper is provided for decoding the weight out of the
two fields. Forcing a conversion seems preferable to bending backwards and
introducing anonymous unions or whatever.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://patch.msgid.link/483e2fcf4beb0d9135d62e7d27b46fa2685479d4.1723036486.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-08-07 16:13:47 +02:00
|
|
|
|
2021-01-28 13:49:15 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-26 14:20:22 +01:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-28 13:49:15 +01:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:15 -06:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2021-03-11 19:03:14 +01:00
|
|
|
|
2021-03-26 14:20:22 +01:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
2020-06-08 20:54:43 -06:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2024-03-06 13:49:19 +01:00
|
|
|
|
nexthop: Add implementation of resilient next-hop groups
At this moment, there is only one type of next-hop group: an mpath group,
which implements the hash-threshold algorithm.
To select a next hop, hash-threshold algorithm first assigns a range of
hashes to each next hop in the group, and then selects the next hop by
comparing the SKB hash with the individual ranges. When a next hop is
removed from the group, the ranges are recomputed, which leads to
reassignment of parts of hash space from one next hop to another. While
there will usually be some overlap between the previous and the new
distribution, some traffic flows change the next hop that they resolve to.
That causes problems e.g. as established TCP connections are reset, because
the traffic is forwarded to a server that is not familiar with the
connection.
Resilient hashing is a technique to address the above problem. Resilient
next-hop group has another layer of indirection between the group itself
and its constituent next hops: a hash table. The selection algorithm uses a
straightforward modulo operation to choose a hash bucket, and then reads
the next hop that this bucket contains, and forwards traffic there.
This indirection brings an important feature. In the hash-threshold
algorithm, the range of hashes associated with a next hop must be
continuous. With a hash table, mapping between the hash table buckets and
the individual next hops is arbitrary. Therefore when a next hop is deleted
the buckets that held it are simply reassigned to other next hops. When
weights of next hops in a group are altered, it may be possible to choose a
subset of buckets that are currently not used for forwarding traffic, and
use those to satisfy the new next-hop distribution demands, keeping the
"busy" buckets intact. This way, established flows are ideally kept being
forwarded to the same endpoints through the same paths as before the
next-hop group change.
In a nutshell, the algorithm works as follows. Each next hop has a number
of buckets that it wants to have, according to its weight and the number of
buckets in the hash table. In case of an event that might cause bucket
allocation change, the numbers for individual next hops are updated,
similarly to how ranges are updated for mpath group next hops. Following
that, a new "upkeep" algorithm runs, and for idle buckets that belong to a
next hop that is currently occupying more buckets than it wants (it is
"overweight"), it migrates the buckets to one of the next hops that has
fewer buckets than it wants (it is "underweight"). If, after this, there
are still underweight next hops, another upkeep run is scheduled to a
future time.
Chances are there are not enough "idle" buckets to satisfy the new demands.
The algorithm has knobs to select both what it means for a bucket to be
idle, and for whether and when to forcefully migrate buckets if there keeps
being an insufficient number of idle buckets.
There are three users of the resilient data structures.
- The forwarding code accesses them under RCU, and does not modify them
except for updating the time a selected bucket was last used.
- Netlink code, running under RTNL, which may modify the data.
- The delayed upkeep code, which may modify the data. This runs unlocked,
and mutual exclusion between the RTNL code and the delayed upkeep is
maintained by canceling the delayed work synchronously before the RTNL
code touches anything. Later it restarts the delayed work if necessary.
The RTNL code has to implement next-hop group replacement, next hop
removal, etc. For removal, the mpath code uses a neat trick of having a
backup next hop group structure, doing the necessary changes offline, and
then RCU-swapping them in. However, the hash tables for resilient hashing
are about an order of magnitude larger than the groups themselves (the size
might be e.g. 4K entries), and it was felt that keeping two of them is an
overkill. Both the primary next-hop group and the spare therefore use the
same resilient table, and writers are careful to keep all references valid
for the forwarding code. The hash table references next-hop group entries
from the next-hop group that is currently in the primary role (i.e. not
spare). During the transition from primary to spare, the table references a
mix of both the primary group and the spare. When a next hop is deleted,
the corresponding buckets are not set to NULL, but instead marked as empty,
so that the pointer is valid and can be used by the forwarding code. The
buckets are then migrated to a new next-hop group entry during upkeep. The
only times that the hash table is invalid is the very beginning and very
end of its lifetime. Between those points, it is always kept valid.
This patch introduces the core support code itself. It does not handle
notifications towards drivers, which are kept as if the group were an mpath
one. It does not handle netlink either. The only bit currently exposed to
user space is the new next-hop group type, and that is currently bounced.
There is therefore no way to actually access this code.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-03-11 19:03:16 +01:00
|
|
|
|
|
|
|
|
|
2023-10-03 18:44:49 -07:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
2019-06-03 20:19:52 -07:00
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2025-04-17 17:03:55 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:15 -07:00
|
|
|
|
2020-11-04 15:30:29 +02:00
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:17 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
2020-05-21 22:26:15 -07:00
|
|
|
|
|
|
|
|
|
2021-01-28 13:49:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:17 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
2021-01-28 13:49:17 +01:00
|
|
|
|
|
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-08 13:59:46 +01:00
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: nexthop: Increase weight to u16
In CLOS networks, as link failures occur at various points in the network,
ECMP weights of the involved nodes are adjusted to compensate. With high
fan-out of the involved nodes, and overall high number of nodes,
a (non-)ECMP weight ratio that we would like to configure does not fit into
8 bits. Instead of, say, 255:254, we might like to configure something like
1000:999. For these deployments, the 8-bit weight may not be enough.
To that end, in this patch increase the next hop weight from u8 to u16.
Increasing the width of an integral type can be tricky, because while the
code still compiles, the types may not check out anymore, and numerical
errors come up. To prevent this, the conversion was done in two steps.
First the type was changed from u8 to a single-member structure, which
invalidated all uses of the field. This allowed going through them one by
one and audit for type correctness. Then the structure was replaced with a
vanilla u16 again. This should ensure that no place was missed.
The UAPI for configuring nexthop group members is that an attribute
NHA_GROUP carries an array of struct nexthop_grp entries:
struct nexthop_grp {
__u32 id; /* nexthop id - must exist */
__u8 weight; /* weight of this nexthop */
__u8 resvd1;
__u16 resvd2;
};
The field resvd1 is currently validated and required to be zero. We can
lift this requirement and carry high-order bits of the weight in the
reserved field:
struct nexthop_grp {
__u32 id; /* nexthop id - must exist */
__u8 weight; /* weight of this nexthop */
__u8 weight_high;
__u16 resvd2;
};
Keeping the fields split this way was chosen in case an existing userspace
makes assumptions about the width of the weight field, and to sidestep any
endianness issues.
The weight field is currently encoded as the weight value minus one,
because weight of 0 is invalid. This same trick is impossible for the new
weight_high field, because zero must mean actual zero. With this in place:
- Old userspace is guaranteed to carry weight_high of 0, therefore
configuring 8-bit weights as appropriate. When dumping nexthops with
16-bit weight, it would only show the lower 8 bits. But configuring such
nexthops implies existence of userspace aware of the extension in the
first place.
- New userspace talking to an old kernel will work as long as it only
attempts to configure 8-bit weights, where the high-order bits are zero.
Old kernel will bounce attempts at configuring >8-bit weights.
Renaming reserved fields as they are allocated for some purpose is commonly
done in Linux. Whoever touches a reserved field is doing so at their own
risk. nexthop_grp::resvd1 in particular is currently used by at least
strace, however they carry an own copy of UAPI headers, and the conversion
should be trivial. A helper is provided for decoding the weight out of the
two fields. Forcing a conversion seems preferable to bending backwards and
introducing anonymous unions or whatever.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Link: https://patch.msgid.link/483e2fcf4beb0d9135d62e7d27b46fa2685479d4.1723036486.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-08-07 16:13:47 +02:00
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:19 +01:00
|
|
|
|
2023-10-03 16:21:47 -07:00
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:19 +01:00
|
|
|
|
2023-10-03 16:18:18 -07:00
|
|
|
|
2021-03-11 19:03:17 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-28 13:49:17 +01:00
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:17 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
2020-11-04 15:30:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-11-04 15:30:34 +02:00
|
|
|
|
|
|
|
|
|
2024-02-06 14:43:04 +00:00
|
|
|
|
2020-05-21 22:26:15 -07:00
|
|
|
|
2020-11-04 15:30:28 +02:00
|
|
|
|
2021-03-11 19:03:19 +01:00
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:20 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 13:49:21 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:15 -07:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-23 20:54:08 +00:00
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-06-08 20:54:43 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-06-09 17:27:28 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:14 +01:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:14 +01:00
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-07 18:31:07 +03:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-09-04 10:11:58 -04:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2024-02-22 10:50:18 +00:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-09-23 23:03:19 +08:00
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-11 19:03:14 +01:00
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:17 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:18 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:51 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:49 -07:00
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-09 18:20:35 -08:00
|
|
|
|
2019-06-03 20:19:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-26 12:56:16 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 20:19:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-08 14:53:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-24 00:26:17 +00:00
|
|
|
|
|
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-24 00:26:17 +00:00
|
|
|
|
2020-05-21 22:26:13 -07:00
|
|
|
|
|
|
|
|
|
2019-05-24 14:43:04 -07:00
|
|
|
|
