The aim of this patch is to reduce the fragmentation of block groups
under certain unhappy workloads. It is particularly effective when the
size of extents correlates with their lifetime, which is something we
have observed causing fragmentation in the fleet at Meta.
This patch categorizes extents into size classes:
- x < 128KiB: "small"
- 128KiB < x < 8MiB: "medium"
- x > 8MiB: "large"
and as much as possible reduces allocations of extents into block groups
that don't match the size class. This takes advantage of any (possible)
correlation between size and lifetime and also leaves behind predictable
re-usable gaps when extents are freed; small writes don't gum up bigger
holes.
Size classes are implemented in the following way:
- Mark each new block group with a size class of the first allocation
that goes into it.
- Add two new passes to ffe: "unset size class" and "wrong size class".
First, try only matching block groups, then try unset ones, then allow
allocation of new ones, and finally allow mismatched block groups.
- Filtering is done just by skipping inappropriate ones, there is no
special size class indexing.
Other solutions I considered were:
- A best fit allocator with an rb-tree. This worked well, as small
writes didn't leak big holes from large freed extents, but led to
regressions in ffe and write performance due to lock contention on
the rb-tree with every allocation possibly updating it in parallel.
Perhaps something clever could be done to do the updates in the
background while being "right enough".
- A fixed size "working set". This prevents freeing an extent
drastically changing where writes currently land, and seems like a
good option too. Doesn't take advantage of size in any way.
- The same size class idea, but implemented with xarray marks. This
turned out to be slower than looping the linked list and skipping
wrong block groups, and is also less flexible since we must have only
3 size classes (max #marks). With the current approach we can have as
many as we like.
Performance testing was done via: https://github.com/josefbacik/fsperf
Of particular relevance are the new fragmentation specific tests.
A brief summary of the testing results:
- Neutral results on existing tests. There are some minor regressions
and improvements here and there, but nothing that truly stands out as
notable.
- Improvement on new tests where size class and extent lifetime are
correlated. Fragmentation in these cases is completely eliminated
and write performance is generally a little better. There is also
significant improvement where extent sizes are just a bit larger than
the size class boundaries.
- Regression on one new tests: where the allocations are sized
intentionally a hair under the borders of the size classes. Results
are neutral on the test that intentionally attacks this new scheme by
mixing extent size and lifetime.
The full dump of the performance results can be found here:
https://bur.io/fsperf/size-class-2022-11-15.txt
(there are ANSI escape codes, so best to curl and view in terminal)
Here is a snippet from the full results for a new test which mixes
buffered writes appending to a long lived set of files and large short
lived fallocates:
bufferedappendvsfallocate results
metric baseline current stdev diff
======================================================================================
avg_commit_ms 31.13 29.20 2.67 -6.22%
bg_count 14 15.60 0 11.43%
commits 11.10 12.20 0.32 9.91%
elapsed 27.30 26.40 2.98 -3.30%
end_state_mount_ns 11122551.90 10635118.90 851143.04 -4.38%
end_state_umount_ns 1.36e+09 1.35e+09 12248056.65 -1.07%
find_free_extent_calls 116244.30 114354.30 964.56 -1.63%
find_free_extent_ns_max 599507.20 1047168.20 103337.08 74.67%
find_free_extent_ns_mean 3607.19 3672.11 101.20 1.80%
find_free_extent_ns_min 500 512 6.67 2.40%
find_free_extent_ns_p50 2848 2876 37.65 0.98%
find_free_extent_ns_p95 4916 5000 75.45 1.71%
find_free_extent_ns_p99 20734.49 20920.48 1670.93 0.90%
frag_pct_max 61.67 0 8.05 -100.00%
frag_pct_mean 43.59 0 6.10 -100.00%
frag_pct_min 25.91 0 16.60 -100.00%
frag_pct_p50 42.53 0 7.25 -100.00%
frag_pct_p95 61.67 0 8.05 -100.00%
frag_pct_p99 61.67 0 8.05 -100.00%
fragmented_bg_count 6.10 0 1.45 -100.00%
max_commit_ms 49.80 46 5.37 -7.63%
sys_cpu 2.59 2.62 0.29 1.39%
write_bw_bytes 1.62e+08 1.68e+08 17975843.50 3.23%
write_clat_ns_mean 57426.39 54475.95 2292.72 -5.14%
write_clat_ns_p50 46950.40 42905.60 2101.35 -8.62%
write_clat_ns_p99 148070.40 143769.60 2115.17 -2.90%
write_io_kbytes 4194304 4194304 0 0.00%
write_iops 2476.15 2556.10 274.29 3.23%
write_lat_ns_max 2101667.60 2251129.50 370556.59 7.11%
write_lat_ns_mean 59374.91 55682.00 2523.09 -6.22%
write_lat_ns_min 17353.10 16250 1646.08 -6.36%
There are some mixed improvements/regressions in most metrics along with
an elimination of fragmentation in this workload.
On the balance, the drastic 1->0 improvement in the happy cases seems
worth the mix of regressions and improvements we do observe.
Some considerations for future work:
- Experimenting with more size classes
- More hinting/search ordering work to approximate a best-fit allocator
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
find_free_extent is a complicated function. It consists (at least) of:
- a hint that jumps into the middle of a for loop macro
- a middle loop trying every raid level
- an outer loop ascending through ffe loop levels
- complicated logic for skipping some of those ffe loop levels
- multiple underlying in-bg allocators (zoned, cluster, no cluster)
Which is all to say that more tracing is helpful for debugging its
behavior. Add two new tracepoints: at the entrance to the block_groups
loop (hit for every raid level and every ffe_ctl loop) and at the point
we seriously consider a block_group for allocation. This way we can see
the whole path through the algorithm, including hints, multiple loops,
etc.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The allocator tracepoints currently have a pile of values from ffe_ctl.
In modifying the allocator and adding more tracepoints, I found myself
adding to the already long argument list of the tracepoints. It makes it
a lot simpler to just send in the ffe_ctl itself.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Boris Burkov <boris@bur.io>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Merge the general CXL updates with fixes targeting v6.2-rc for v6.3.
Resolve a conflict with the fix and move of cxl_report_and_clear() from
pci.c to core/pci.c.
Convert to use iostat_lat_type as parameter instead of raw number.
BTW, move NUM_PREALLOC_IOSTAT_CTXS to the header file, adjust
iostat_lat[{0,1,2}] to iostat_lat[{READ_IO,WRITE_SYNC_IO,WRITE_ASYNC_IO}]
in tracepoint function, and rename iotype to page_type to match the definition.
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Yangtao Li <frank.li@vivo.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
there is one dwc3 trace event declare as below,
DECLARE_EVENT_CLASS(dwc3_log_event,
TP_PROTO(u32 event, struct dwc3 *dwc),
TP_ARGS(event, dwc),
TP_STRUCT__entry(
__field(u32, event)
__field(u32, ep0state)
__dynamic_array(char, str, DWC3_MSG_MAX)
),
TP_fast_assign(
__entry->event = event;
__entry->ep0state = dwc->ep0state;
),
TP_printk("event (%08x): %s", __entry->event,
dwc3_decode_event(__get_str(str), DWC3_MSG_MAX,
__entry->event, __entry->ep0state))
);
the problem is when trace function called, it will allocate up to
DWC3_MSG_MAX bytes from trace event buffer, but never fill the buffer
during fast assignment, it only fill the buffer when output function are
called, so this means if output function are not called, the buffer will
never used.
add __get_buf(len) which acquiree buffer from iter->tmp_seq when trace
output function called, it allow user write string to acquired buffer.
the mentioned dwc3 trace event will changed as below,
DECLARE_EVENT_CLASS(dwc3_log_event,
TP_PROTO(u32 event, struct dwc3 *dwc),
TP_ARGS(event, dwc),
TP_STRUCT__entry(
__field(u32, event)
__field(u32, ep0state)
),
TP_fast_assign(
__entry->event = event;
__entry->ep0state = dwc->ep0state;
),
TP_printk("event (%08x): %s", __entry->event,
dwc3_decode_event(__get_buf(DWC3_MSG_MAX), DWC3_MSG_MAX,
__entry->event, __entry->ep0state))
);.
Link: https://lore.kernel.org/linux-trace-kernel/1675065249-23368-1-git-send-email-quic_linyyuan@quicinc.com
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Linyu Yuan <quic_linyyuan@quicinc.com>
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
The following patch will add two more maximum MDB allowances to the global
one, mcast_hash_max, that exists today. In all these cases, attempts to add
MDB entries above the configured maximums through netlink, fail noisily and
obviously. Such visibility is missing when adding entries through the
control plane traffic, by IGMP or MLD packets.
To improve visibility in those cases, add a trace point that reports the
violation, including the relevant netdevice (be it a slave or the bridge
itself), and the MDB entry parameters:
# perf record -e bridge:br_mdb_full &
# [...]
# perf script | cut -d: -f4-
dev v2 af 2 src ::ffff:0.0.0.0 grp ::ffff:239.1.1.112/00:00:00:00:00:00 vid 0
dev v2 af 10 src :: grp ff0e::112/00:00:00:00:00:00 vid 0
dev v2 af 2 src ::ffff:0.0.0.0 grp ::ffff:239.1.1.112/00:00:00:00:00:00 vid 10
dev v2 af 10 src 2001:db8:1::1 grp ff0e::1/00:00:00:00:00:00 vid 10
dev v2 af 2 src ::ffff:192.0.2.1 grp ::ffff:239.1.1.1/00:00:00:00:00:00 vid 10
CC: Steven Rostedt <rostedt@goodmis.org>
CC: linux-trace-kernel@vger.kernel.org
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
__GFP_ATOMIC serves little purpose. Its main effect is to set
ALLOC_HARDER which adds a few little boosts to increase the chance of an
allocation succeeding, one of which is to lower the water-mark at which it
will succeed.
It is *always* paired with __GFP_HIGH which sets ALLOC_HIGH which also
adjusts this watermark. It is probable that other users of __GFP_HIGH
should benefit from the other little bonuses that __GFP_ATOMIC gets.
__GFP_ATOMIC also gives a warning if used with __GFP_DIRECT_RECLAIM.
There is little point to this. We already get a might_sleep() warning if
__GFP_DIRECT_RECLAIM is set.
__GFP_ATOMIC allows the "watermark_boost" to be side-stepped. It is
probable that testing ALLOC_HARDER is a better fit here.
__GFP_ATOMIC is used by tegra-smmu.c to check if the allocation might
sleep. This should test __GFP_DIRECT_RECLAIM instead.
This patch:
- removes __GFP_ATOMIC
- allows __GFP_HIGH allocations to ignore watermark boosting as well
as GFP_ATOMIC requests.
- makes other adjustments as suggested by the above.
The net result is not change to GFP_ATOMIC allocations. Other
allocations that use __GFP_HIGH will benefit from a few different extra
privileges. This affects:
xen, dm, md, ntfs3
the vermillion frame buffer
hibernation
ksm
swap
all of which likely produce more benefit than cost if these selected
allocation are more likely to succeed quickly.
[mgorman: Minor adjustments to rework on top of a series]
Link: https://lkml.kernel.org/r/163712397076.13692.4727608274002939094@noble.neil.brown.name
Link: https://lkml.kernel.org/r/20230113111217.14134-7-mgorman@techsingularity.net
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Thierry Reding <thierry.reding@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Change the rx_packet tracepoint to display the securityIndex from the
packet header instead of displaying the type in numeric form. There's no
need for the latter, as the display of the type in symbolic form will fall
back automatically to displaying the hex value if no symbol is available.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Now that general ACK transmission is done from the same thread as incoming
DATA packet wrangling, there's no possibility that the SACK table will be
being updated by the latter whilst the former is trying to copy it to an
ACK.
This means that we can safely rotate the SACK table whilst updating it
without having to take a lock, rather than keeping all the bits inside it
in fixed place and copying and then rotating it in the transmitter.
Therefore, simplify SACK handing by keeping track of starting point in the
ring and rotate slots down as we consume them.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
call->ackr_window doesn't need to be atomic as ACK generation and ACK
transmission are now done in the same thread, so drop the atomic64 handling
and split it into two separate members.
Similarly, call->ackr_nr_unacked doesn't need to be atomic now either.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
When doing a call that has a single transmitted data packet and a massive
amount of received data packets, we only ping for one RTT sample, which
means we don't get a good reading on it.
Fix this by converting occasional IDLE ACKs into PING ACKs to elicit a
response.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Shrink the tabulation in the rxrpc trace header a bit to allow for fields
with long type names that have been removed.
Signed-off-by: David Howells <dhowells@redhat.com>
Work around checkpatch warnings in the rxrpc trace header by removing
whitespace before ')' on lines defining the trace record struct.
Signed-off-by: David Howells <dhowells@redhat.com>
Due to holidays we started -next with more -fixes in-flight than
usual, and people have been asking where they are. Backmerge to get
things better in sync.
Conflicts:
- Tiny conflict in drm_fbdev_generic.c between variable rename and
missing error handling that got added.
- Conflict in drm_fb_helper.c between the added call to vgaswitcheroo
in drm_fb_helper_single_fb_probe and a refactor patch that extracted
lots of helpers and incidentally removed the dev local variable.
Readd it to make things compile.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Commit 3db1de0e58 ("f2fs: change the current atomic write way")
removed old tracepoints, but it missed to add new one, this patch
fixes to introduce trace_f2fs_replace_atomic_write_block to trace
atomic_write commit flow.
Fixes: 3db1de0e58 ("f2fs: change the current atomic write way")
Signed-off-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
There are cases where it may be useful to dump the whole LBW configs.
Yet, doing so while spamming the kernel log will probably shade other
important messages since the LBW access is done in sheer volume.
To answer this we add trace events for those too.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
As the COMMS protocol is being used more widely in our driver,
an available debug tool for the handshake will be handy.
This commit defines tracepoints to various key points of the COMMS
protocol.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Arm SCMI updates for v6.3
The main addition is a unified userspace interface for SCMI irrespective
of the underlying transport and along with some changed to refactor the
SCMI stack probing sequence.
1. SCMI unified userspace interface
This is to have a unified way of testing an SCMI platform firmware
implementation for compliance, fuzzing etc., from the perspective of
the non-secure OSPM irrespective of the underlying transport supporting
SCMI. It is just for testing/development and not a feature intended fo
use in production.
Currently an SCMI Compliance Suite[1] can only work by injecting SCMI
messages using the mailbox test driver only which makes it transport
specific and can't be used with any other transport like virtio,
smc/hvc, optee, etc. Also the shared memory can be transport specific
and it is better to even abstract/hide those details while providing
the userspace access. So in order to scale with any transport, we need
a unified interface for the same.
In order to achieve that, SCMI "raw mode support" is being added through
debugfs which is more configurable as well. A userspace application
can inject bare SCMI binary messages into the SCMI core stack; such
messages will be routed by the SCMI regular kernel stack to the backend
platform firmware using the configured transport transparently. This
eliminates the to know about the specific underlying transport
internals that will be taken care of by the SCMI core stack itself.
Further no additional changes needed in the device tree like in the
mailbox-test driver.
[1] https://gitlab.arm.com/tests/scmi-tests
2. Refactoring of the SCMI stack probing sequence
On some platforms, SCMI transport can be provide by OPTEE/TEE which
introduces certain dependency in the probe ordering. In order to address
the same, the SCMI bus is split into its own module which continues to
be initialized at subsys_initcall, while the SCMI core stack, including
its various transport backends (like optee, mailbox, virtio, smc), is
now moved into a separate module at module_init level.
This allows the other possibly dependent subsystems to register and/or
access SCMI bus well before the core SCMI stack and its dependent
transport backends.
* tag 'scmi-updates-6.3' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux: (31 commits)
firmware: arm_scmi: Clarify raw per-channel ABI documentation
firmware: arm_scmi: Add per-channel raw injection support
firmware: arm_scmi: Add the raw mode co-existence support
firmware: arm_scmi: Call raw mode hooks from the core stack
firmware: arm_scmi: Reject SCMI drivers when configured in raw mode
firmware: arm_scmi: Add debugfs ABI documentation for raw mode
firmware: arm_scmi: Add core raw transmission support
firmware: arm_scmi: Add debugfs ABI documentation for common entries
firmware: arm_scmi: Populate a common SCMI debugfs root
debugfs: Export debugfs_create_str symbol
include: trace: Add platform and channel instance references
firmware: arm_scmi: Add internal platform/channel identifiers
firmware: arm_scmi: Move errors defs and code to common.h
firmware: arm_scmi: Add xfer helpers to provide raw access
firmware: arm_scmi: Add flags field to xfer
firmware: arm_scmi: Refactor scmi_wait_for_message_response
firmware: arm_scmi: Refactor polling helpers
firmware: arm_scmi: Refactor xfer in-flight registration routines
firmware: arm_scmi: Split bus and driver into distinct modules
firmware: arm_scmi: Introduce a new lifecycle for protocol devices
...
Link: https://lore.kernel.org/r/20230120162152.1438456-1-sudeep.holla@arm.com
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
As suggested by Cong, introduce a tracepoint for all ->sk_data_ready()
callback implementations. For example:
<...>
iperf-609 [002] ..... 70.660425: sk_data_ready: family=2 protocol=6 func=sock_def_readable
iperf-609 [002] ..... 70.660436: sk_data_ready: family=2 protocol=6 func=sock_def_readable
<...>
Suggested-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The result of the allocation attempt is not printed in
trace_cma_alloc_finish, but it's important to do it so we can set filters
to catch specific errors on allocation or to trigger some operations on
specific errors.
We have printed the result in log, but the log is conditional and could
not be filtered by tracing events.
It introduces little overhead to print this result. The result of
allocation is named `errorno' in the trace.
Link: https://lkml.kernel.org/r/20221208142130.1501195-1-haowenchao@huawei.com
Signed-off-by: Wenchao Hao <haowenchao@huawei.com>
Cc: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Cc: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Add 2 tracepoints to monitor the tcp/udp traffic
of per process and per cgroup.
Regarding monitoring the tcp/udp traffic of each process, there are two
existing solutions, the first one is https://www.atoptool.nl/netatop.php.
The second is via kprobe/kretprobe.
Netatop solution is implemented by registering the hook function at the
hook point provided by the netfilter framework.
These hook functions may be in the soft interrupt context and cannot
directly obtain the pid. Some data structures are added to bind packets
and processes. For example, struct taskinfobucket, struct taskinfo ...
Every time the process sends and receives packets it needs multiple
hashmaps,resulting in low performance and it has the problem fo inaccurate
tcp/udp traffic statistics(for example: multiple threads share sockets).
We can obtain the information with kretprobe, but as we know, kprobe gets
the result by trappig in an exception, which loses performance compared
to tracepoint.
We compared the performance of tracepoints with the above two methods, and
the results are as follows:
ab -n 1000000 -c 1000 -r http://127.0.0.1/index.html
without trace:
Time per request: 39.660 [ms] (mean)
Time per request: 0.040 [ms] (mean, across all concurrent requests)
netatop:
Time per request: 50.717 [ms] (mean)
Time per request: 0.051 [ms] (mean, across all concurrent requests)
kr:
Time per request: 43.168 [ms] (mean)
Time per request: 0.043 [ms] (mean, across all concurrent requests)
tracepoint:
Time per request: 41.004 [ms] (mean)
Time per request: 0.041 [ms] (mean, across all concurrent requests
It can be seen that tracepoint has better performance.
Signed-off-by: Yunhui Cui <cuiyunhui@bytedance.com>
Signed-off-by: Xiongchun Duan <duanxiongchun@bytedance.com>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously, we supported to account iostat io_bytes,
in this patch, it adds to account iostat count and avg_bytes:
time: 1671648667
io_bytes count avg_bytes
[WRITE]
app buffered data: 31 2 15
Signed-off-by: Yangtao Li <frank.li@vivo.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The create argument is always identicaly to map->m_may_create, so use
that consistently.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Move the connection setup of client calls to the I/O thread so that a whole
load of locking and barrierage can be eliminated. This necessitates the
app thread waiting for connection to complete before it can begin
encrypting data.
This also completes the fix for a race that exists between call connection
and call disconnection whereby the data transmission code adds the call to
the peer error distribution list after the call has been disconnected (say
by the rxrpc socket getting closed).
The fix is to complete the process of moving call connection, data
transmission and call disconnection into the I/O thread and thus forcibly
serialising them.
Note that the issue may predate the overhaul to an I/O thread model that
were included in the merge window for v6.2, but the timing is very much
changed by the change given below.
Fixes: cf37b59875 ("rxrpc: Move DATA transmission into call processor work item")
Reported-by: syzbot+c22650d2844392afdcfd@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Use the information now stored in struct rxrpc_call to configure the
connection bundle and thence the connection, rather than using the
rxrpc_conn_parameters struct.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Offload the completion of the challenge/response cycle on a service
connection to the I/O thread. After the RESPONSE packet has been
successfully decrypted and verified by the work queue, offloading the
changing of the call states to the I/O thread makes iteration over the
conn's channel list simpler.
Do this by marking the RESPONSE skbuff and putting it onto the receive
queue for the I/O thread to collect. We put it on the front of the queue
as we've already received the packet for it.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Tidy up the abort generation infrastructure in the following ways:
(1) Create an enum and string mapping table to list the reasons an abort
might be generated in tracing.
(2) Replace the 3-char string with the values from (1) in the places that
use that to log the abort source. This gets rid of a memcpy() in the
tracepoint.
(3) Subsume the rxrpc_rx_eproto tracepoint with the rxrpc_abort tracepoint
and use values from (1) to indicate the trace reason.
(4) Always make a call to an abort function at the point of the abort
rather than stashing the values into variables and using goto to get
to a place where it reported. The C optimiser will collapse the calls
together as appropriate. The abort functions return a value that can
be returned directly if appropriate.
Note that this extends into afs also at the points where that generates an
abort. To aid with this, the afs sources need to #define
RXRPC_TRACE_ONLY_DEFINE_ENUMS before including the rxrpc tracing header
because they don't have access to the rxrpc internal structures that some
of the tracepoints make use of.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Clean up connection abort, using the connection state_lock to gate access
to change that state, and use an rxrpc_call_completion value to indicate
the difference between local and remote aborts as these can be pasted
directly into the call state.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Provide a means by which an event notification can be sent to a connection
through such that the I/O thread can pick it up and handle it rather than
doing it in a separate workqueue.
This is then used to move the deferred final ACK of a call into the I/O
thread rather than a separate work queue as part of the drive to do all
transmission from the I/O thread.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Only perform call disconnection in the I/O thread to reduce the locking
requirement.
This is the first part of a fix for a race that exists between call
connection and call disconnection whereby the data transmission code adds
the call to the peer error distribution list after the call has been
disconnected (say by the rxrpc socket getting closed).
The fix is to complete the process of moving call connection, data
transmission and call disconnection into the I/O thread and thus forcibly
serialising them.
Note that the issue may predate the overhaul to an I/O thread model that
were included in the merge window for v6.2, but the timing is very much
changed by the change given below.
Fixes: cf37b59875 ("rxrpc: Move DATA transmission into call processor work item")
Reported-by: syzbot+c22650d2844392afdcfd@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Only set the abort call completion state in the I/O thread and only
transmit ABORT packets from there. rxrpc_abort_call() can then be made to
actually send the packet.
Further, ABORT packets should only be sent if the call has been exposed to
the network (ie. at least one attempted DATA transmission has occurred for
it).
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Make the local endpoint and it's I/O thread hold a reference on a connected
call until that call is disconnected. Without this, we're reliant on
either the AF_RXRPC socket to hold a ref (which is dropped when the call is
released) or a queued work item to hold a ref (the work item is being
replaced with the I/O thread).
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Pull networking fixes from Jakub Kicinski:
"Including fixes from bpf, wifi, and netfilter.
Current release - regressions:
- bpf: fix nullness propagation for reg to reg comparisons, avoid
null-deref
- inet: control sockets should not use current thread task_frag
- bpf: always use maximal size for copy_array()
- eth: bnxt_en: don't link netdev to a devlink port for VFs
Current release - new code bugs:
- rxrpc: fix a couple of potential use-after-frees
- netfilter: conntrack: fix IPv6 exthdr error check
- wifi: iwlwifi: fw: skip PPAG for JF, avoid FW crashes
- eth: dsa: qca8k: various fixes for the in-band register access
- eth: nfp: fix schedule in atomic context when sync mc address
- eth: renesas: rswitch: fix getting mac address from device tree
- mobile: ipa: use proper endpoint mask for suspend
Previous releases - regressions:
- tcp: add TIME_WAIT sockets in bhash2, fix regression caught by
Jiri / python tests
- net: tc: don't intepret cls results when asked to drop, fix
oob-access
- vrf: determine the dst using the original ifindex for multicast
- eth: bnxt_en:
- fix XDP RX path if BPF adjusted packet length
- fix HDS (header placement) and jumbo thresholds for RX packets
- eth: ice: xsk: do not use xdp_return_frame() on tx_buf->raw_buf,
avoid memory corruptions
Previous releases - always broken:
- ulp: prevent ULP without clone op from entering the LISTEN status
- veth: fix race with AF_XDP exposing old or uninitialized
descriptors
- bpf:
- pull before calling skb_postpull_rcsum() (fix checksum support
and avoid a WARN())
- fix panic due to wrong pageattr of im->image (when livepatch and
kretfunc coexist)
- keep a reference to the mm, in case the task is dead
- mptcp: fix deadlock in fastopen error path
- netfilter:
- nf_tables: perform type checking for existing sets
- nf_tables: honor set timeout and garbage collection updates
- ipset: fix hash:net,port,net hang with /0 subnet
- ipset: avoid hung task warning when adding/deleting entries
- selftests: net:
- fix cmsg_so_mark.sh test hang on non-x86 systems
- fix the arp_ndisc_evict_nocarrier test for IPv6
- usb: rndis_host: secure rndis_query check against int overflow
- eth: r8169: fix dmar pte write access during suspend/resume with
WOL
- eth: lan966x: fix configuration of the PCS
- eth: sparx5: fix reading of the MAC address
- eth: qed: allow sleep in qed_mcp_trace_dump()
- eth: hns3:
- fix interrupts re-initialization after VF FLR
- fix handling of promisc when MAC addr table gets full
- refine the handling for VF heartbeat
- eth: mlx5:
- properly handle ingress QinQ-tagged packets on VST
- fix io_eq_size and event_eq_size params validation on big endian
- fix RoCE setting at HCA level if not supported at all
- don't turn CQE compression on by default for IPoIB
- eth: ena:
- fix toeplitz initial hash key value
- account for the number of XDP-processed bytes in interface stats
- fix rx_copybreak value update
Misc:
- ethtool: harden phy stat handling against buggy drivers
- docs: netdev: convert maintainer's doc from FAQ to a normal
document"
* tag 'net-6.2-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (112 commits)
caif: fix memory leak in cfctrl_linkup_request()
inet: control sockets should not use current thread task_frag
net/ulp: prevent ULP without clone op from entering the LISTEN status
qed: allow sleep in qed_mcp_trace_dump()
MAINTAINERS: Update maintainers for ptp_vmw driver
usb: rndis_host: Secure rndis_query check against int overflow
net: dpaa: Fix dtsec check for PCS availability
octeontx2-pf: Fix lmtst ID used in aura free
drivers/net/bonding/bond_3ad: return when there's no aggregator
netfilter: ipset: Rework long task execution when adding/deleting entries
netfilter: ipset: fix hash:net,port,net hang with /0 subnet
net: sparx5: Fix reading of the MAC address
vxlan: Fix memory leaks in error path
net: sched: htb: fix htb_classify() kernel-doc
net: sched: cbq: dont intepret cls results when asked to drop
net: sched: atm: dont intepret cls results when asked to drop
dt-bindings: net: marvell,orion-mdio: Fix examples
dt-bindings: net: sun8i-emac: Add phy-supply property
net: ipa: use proper endpoint mask for suspend
selftests: net: return non-zero for failures reported in arp_ndisc_evict_nocarrier
...
CXL is using tracepoints for reporting RAS capability register payloads
for AER events, and has plans to use tracepoints for the output payload
of Get Poison List and Get Event Records commands. For organization
purposes it would be nice to keep those all under a single + local CXL
trace system. This also organization also potentially helps in the
future when CXL drivers expand beyond generic memory expanders, however
that would also entail a move away from the expander-specific
cxl_dev_state context, save that for later.
Note that the powerpc-specific drivers/misc/cxl/ also defines a 'cxl'
trace system, however, it is unlikely that a single platform will ever
load both drivers simultaneously.
Cc: Steven Rostedt <rostedt@goodmis.org>
Tested-by: Alison Schofield <alison.schofield@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/167051869176.436579.9728373544811641087.stgit@dwillia2-xfh.jf.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Pull btrfs fixes from David Sterba:
"First batch of regression and regular fixes:
- regressions:
- fix error handling after conversion to qstr for paths
- fix raid56/scrub recovery caused by uninitialized variable
after conversion to error bitmaps
- restore qgroup backref lookup behaviour after recent
refactoring
- fix leak of device lists at module exit time
- fix resolving backrefs for inline extent followed by prealloc
- reset defrag ioctl buffer on memory allocation error"
* tag 'for-6.2-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: fix fscrypt name leak after failure to join log transaction
btrfs: scrub: fix uninitialized return value in recover_scrub_rbio
btrfs: fix resolving backrefs for inline extent followed by prealloc
btrfs: fix trace event name typo for FLUSH_DELAYED_REFS
btrfs: restore BTRFS_SEQ_LAST when looking up qgroup backref lookup
btrfs: fix leak of fs devices after removing btrfs module
btrfs: fix an error handling path in btrfs_defrag_leaves()
btrfs: fix an error handling path in btrfs_rename()
At the end of rxrpc_recvmsg(), if a call is found, the call is put and then
a trace line is emitted referencing that call in a couple of places - but
the call may have been deallocated by the time those traces happen.
Fix this by stashing the call debug_id in a variable and passing that to
the tracepoint rather than the call pointer.
Fixes: 849979051c ("rxrpc: Add a tracepoint to follow what recvmsg does")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Adding the NUMA node id to struct rseq is a straightforward thing to do,
and a good way to figure out if anything in the user-space ecosystem
prevents extending struct rseq.
This NUMA node id field allows memory allocators such as tcmalloc to
take advantage of fast access to the current NUMA node id to perform
NUMA-aware memory allocation.
It can also be useful for implementing fast-paths for NUMA-aware
user-space mutexes.
It also allows implementing getcpu(2) purely in user-space.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20221122203932.231377-5-mathieu.desnoyers@efficios.com
