Fix a number of hangs in the netfslib read-retry code, including:
(1) netfs_reissue_read() doubles up the getting of references on
subrequests, thereby leaking the subrequest and causing inode eviction
to wait indefinitely. This can lead to the kernel reporting a hang in
the filesystem's evict_inode().
Fix this by removing the get from netfs_reissue_read() and adding one
to netfs_retry_read_subrequests() to deal with the one place that
didn't double up.
(2) The loop in netfs_retry_read_subrequests() that retries a sequence of
failed subrequests doesn't record whether or not it retried the one
that the "subreq" pointer points to when it leaves the loop. It may
not if renegotiation/repreparation of the subrequests means that fewer
subrequests are needed to span the cumulative range of the sequence.
Because it doesn't record this, the piece of code that discards
now-superfluous subrequests doesn't know whether it should discard the
one "subreq" points to - and so it doesn't.
Fix this by noting whether the last subreq it examines is superfluous
and if it is, then getting rid of it and all subsequent subrequests.
If that one one wasn't superfluous, then we would have tried to go
round the previous loop again and so there can be no further unretried
subrequests in the sequence.
(3) netfs_retry_read_subrequests() gets yet an extra ref on any additional
subrequests it has to get because it ran out of ones it could reuse to
to renegotiation/repreparation shrinking the subrequests.
Fix this by removing that extra ref.
(4) In netfs_retry_reads(), it was using wait_on_bit() to wait for
NETFS_SREQ_IN_PROGRESS to be cleared on all subrequests in the
sequence - but netfs_read_subreq_terminated() is now using a wait
queue on the request instead and so this wait will never finish.
Fix this by waiting on the wait queue instead. To make this work, a
new flag, NETFS_RREQ_RETRYING, is now set around the wait loop to tell
the wake-up code to wake up the wait queue rather than requeuing the
request's work item.
Note that this flag replaces the NETFS_RREQ_NEED_RETRY flag which is
no longer used.
(5) Whilst not strictly anything to do with the hang,
netfs_retry_read_subrequests() was also doubly incrementing the
subreq_counter and re-setting the debug index, leaving a gap in the
trace. This is also fixed.
One of these hangs was observed with 9p and with cifs. Others were forced
by manual code injection into fs/afs/file.c. Firstly, afs_prepare_read()
was created to provide an changing pattern of maximum subrequest sizes:
static int afs_prepare_read(struct netfs_io_subrequest *subreq)
{
struct netfs_io_request *rreq = subreq->rreq;
if (!S_ISREG(subreq->rreq->inode->i_mode))
return 0;
if (subreq->retry_count < 20)
rreq->io_streams[0].sreq_max_len =
umax(200, 2222 - subreq->retry_count * 40);
else
rreq->io_streams[0].sreq_max_len = 3333;
return 0;
}
and pointed to by afs_req_ops. Then the following:
struct netfs_io_subrequest *subreq = op->fetch.subreq;
if (subreq->error == 0 &&
S_ISREG(subreq->rreq->inode->i_mode) &&
subreq->retry_count < 20) {
subreq->transferred = subreq->already_done;
__clear_bit(NETFS_SREQ_HIT_EOF, &subreq->flags);
__set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
afs_fetch_data_notify(op);
return;
}
was inserted into afs_fetch_data_success() at the beginning and struct
netfs_io_subrequest given an extra field, "already_done" that was set to
the value in "subreq->transferred" by netfs_reissue_read().
When reading a 4K file, the subrequests would get gradually smaller, a new
subrequest would be allocated around the 3rd retry and then eventually be
rendered superfluous when the 20th retry was hit and the limit on the first
subrequest was eased.
Fixes: e2d46f2ec3 ("netfs: Change the read result collector to only use one work item")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20250212222402.3618494-2-dhowells@redhat.com
Tested-by: Marc Dionne <marc.dionne@auristor.com>
Tested-by: Steve French <stfrench@microsoft.com>
cc: Ihor Solodrai <ihor.solodrai@pm.me>
cc: Eric Van Hensbergen <ericvh@kernel.org>
cc: Latchesar Ionkov <lucho@ionkov.net>
cc: Dominique Martinet <asmadeus@codewreck.org>
cc: Christian Schoenebeck <linux_oss@crudebyte.com>
cc: Paulo Alcantara <pc@manguebit.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: v9fs@lists.linux.dev
cc: linux-cifs@vger.kernel.org
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Change the way netfslib collects read results to do all the collection for
a particular read request using a single work item that walks along the
subrequest queue as subrequests make progress or complete, unlocking folios
progressively rather than doing the unlock in parallel as parallel requests
come in.
The code is remodelled to be more like the write-side code, though only
using a single stream. This makes it more directly comparable and thus
easier to duplicate fixes between the two sides.
This has a number of advantages:
(1) It's simpler. There doesn't need to be a complex donation mechanism
to handle mismatches between the size and alignment of subrequests and
folios. The collector unlocks folios as the subrequests covering each
complete.
(2) It should cause less scheduler overhead as there's a single work item
in play unlocking pages in parallel when a read gets split up into a
lot of subrequests instead of one per subrequest.
Whilst the parallellism is nice in theory, in practice, the vast
majority of loads are sequential reads of the whole file, so
committing a bunch of threads to unlocking folios out of order doesn't
help in those cases.
(3) It should make it easier to implement content decryption. A folio
cannot be decrypted until all the requests that contribute to it have
completed - and, again, most loads are sequential and so, most of the
time, we want to begin decryption sequentially (though it's great if
the decryption can happen in parallel).
There is a disadvantage in that we're losing the ability to decrypt and
unlock things on an as-things-arrive basis which may affect some
applications.
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241216204124.3752367-28-dhowells@redhat.com
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Drop the was_async argument from netfs_read_subreq_terminated(). Almost
every caller is either in process context and passes false. Some
filesystems delegate the call to a workqueue to avoid doing the work in
their network message queue parsing thread.
The only exception is netfs_cache_read_terminated() which handles
completion in the cache - which is usually a callback from the backing
filesystem in softirq context, though it can be from process context if an
error occurred. In this case, delegate to a workqueue.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/r/CAHk-=wiVC5Cgyz6QKXFu6fTaA6h4CjexDR-OV9kL6Vo5x9v8=A@mail.gmail.com/
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241216204124.3752367-10-dhowells@redhat.com
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Add a tracepoint to log the lifespan of folio_queue structs. For tracing
illustrative purposes, folio_queues are tagged with the debug ID of
whatever they're related to (typically a netfs_io_request) and a debug ID
of their own.
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241216204124.3752367-5-dhowells@redhat.com
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
At the end of netfs_unlock_read_folio() in which folios are marked
appropriately for copying to the cache (either with by being marked dirty
and having their private data set or by having PG_private_2 set) and then
unlocked, the folio_queue struct has the entry pointing to the folio
cleared. This presents a problem for netfs_pgpriv2_write_to_the_cache(),
which is used to write folios marked with PG_private_2 to the cache as it
expects to be able to trawl the folio_queue list thereafter to find the
relevant folios, leading to a hang.
Fix this by not clearing the folio_queue entry if we're going to do the
deprecated copy-to-cache. The clearance will be done instead as the folios
are written to the cache.
This can be reproduced by starting cachefiles, mounting a ceph filesystem
with "-o fsc" and writing to it.
Fixes: 796a404964 ("netfs: In readahead, put the folio refs as soon extracted")
Reported-by: Max Kellermann <max.kellermann@ionos.com>
Closes: https://lore.kernel.org/r/CAKPOu+_4m80thNy5_fvROoxBm689YtA0dZ-=gcmkzwYSY4syqw@mail.gmail.com/
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241213135013.2964079-10-dhowells@redhat.com
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
cc: Jeff Layton <jlayton@kernel.org>
cc: Ilya Dryomov <idryomov@gmail.com>
cc: Xiubo Li <xiubli@redhat.com>
cc: netfs@lists.linux.dev
cc: ceph-devel@vger.kernel.org
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
syzkaller reported recursion with a loop of three calls (netfs_rreq_assess,
netfs_retry_reads and netfs_rreq_terminated) hitting the limit of the stack
during an unbuffered or direct I/O read.
There are a number of issues:
(1) There is no limit on the number of retries.
(2) A subrequest is supposed to be abandoned if it does not transfer
anything (NETFS_SREQ_NO_PROGRESS), but that isn't checked under all
circumstances.
(3) The actual root cause, which is this:
if (atomic_dec_and_test(&rreq->nr_outstanding))
netfs_rreq_terminated(rreq, ...);
When we do a retry, we bump the rreq->nr_outstanding counter to
prevent the final cleanup phase running before we've finished
dispatching the retries. The problem is if we hit 0, we have to do
the cleanup phase - but we're in the cleanup phase and end up
repeating the retry cycle, hence the recursion.
Work around the problem by limiting the number of retries. This is based
on Lizhi Xu's patch[1], and makes the following changes:
(1) Replace NETFS_SREQ_NO_PROGRESS with NETFS_SREQ_MADE_PROGRESS and make
the filesystem set it if it managed to read or write at least one byte
of data. Clear this bit before issuing a subrequest.
(2) Add a ->retry_count member to the subrequest and increment it any time
we do a retry.
(3) Remove the NETFS_SREQ_RETRYING flag as it is superfluous with
->retry_count. If the latter is non-zero, we're doing a retry.
(4) Abandon a subrequest if retry_count is non-zero and we made no
progress.
(5) Use ->retry_count in both the write-side and the read-size.
[?] Question: Should I set a hard limit on retry_count in both read and
write? Say it hits 50, we always abandon it. The problem is that
these changes only mitigate the issue. As long as it made at least one
byte of progress, the recursion is still an issue. This patch
mitigates the problem, but does not fix the underlying cause. I have
patches that will do that, but it's an intrusive fix that's currently
pending for the next merge window.
The oops generated by KASAN looks something like:
BUG: TASK stack guard page was hit at ffffc9000482ff48 (stack is ffffc90004830000..ffffc90004838000)
Oops: stack guard page: 0000 [#1] PREEMPT SMP KASAN NOPTI
...
RIP: 0010:mark_lock+0x25/0xc60 kernel/locking/lockdep.c:4686
...
mark_usage kernel/locking/lockdep.c:4646 [inline]
__lock_acquire+0x906/0x3ce0 kernel/locking/lockdep.c:5156
lock_acquire.part.0+0x11b/0x380 kernel/locking/lockdep.c:5825
local_lock_acquire include/linux/local_lock_internal.h:29 [inline]
___slab_alloc+0x123/0x1880 mm/slub.c:3695
__slab_alloc.constprop.0+0x56/0xb0 mm/slub.c:3908
__slab_alloc_node mm/slub.c:3961 [inline]
slab_alloc_node mm/slub.c:4122 [inline]
kmem_cache_alloc_noprof+0x2a7/0x2f0 mm/slub.c:4141
radix_tree_node_alloc.constprop.0+0x1e8/0x350 lib/radix-tree.c:253
idr_get_free+0x528/0xa40 lib/radix-tree.c:1506
idr_alloc_u32+0x191/0x2f0 lib/idr.c:46
idr_alloc+0xc1/0x130 lib/idr.c:87
p9_tag_alloc+0x394/0x870 net/9p/client.c:321
p9_client_prepare_req+0x19f/0x4d0 net/9p/client.c:644
p9_client_zc_rpc.constprop.0+0x105/0x880 net/9p/client.c:793
p9_client_read_once+0x443/0x820 net/9p/client.c:1570
p9_client_read+0x13f/0x1b0 net/9p/client.c:1534
v9fs_issue_read+0x115/0x310 fs/9p/vfs_addr.c:74
netfs_retry_read_subrequests fs/netfs/read_retry.c:60 [inline]
netfs_retry_reads+0x153a/0x1d00 fs/netfs/read_retry.c:232
netfs_rreq_assess+0x5d3/0x870 fs/netfs/read_collect.c:371
netfs_rreq_terminated+0xe5/0x110 fs/netfs/read_collect.c:407
netfs_retry_reads+0x155e/0x1d00 fs/netfs/read_retry.c:235
netfs_rreq_assess+0x5d3/0x870 fs/netfs/read_collect.c:371
netfs_rreq_terminated+0xe5/0x110 fs/netfs/read_collect.c:407
netfs_retry_reads+0x155e/0x1d00 fs/netfs/read_retry.c:235
netfs_rreq_assess+0x5d3/0x870 fs/netfs/read_collect.c:371
...
netfs_rreq_terminated+0xe5/0x110 fs/netfs/read_collect.c:407
netfs_retry_reads+0x155e/0x1d00 fs/netfs/read_retry.c:235
netfs_rreq_assess+0x5d3/0x870 fs/netfs/read_collect.c:371
netfs_rreq_terminated+0xe5/0x110 fs/netfs/read_collect.c:407
netfs_retry_reads+0x155e/0x1d00 fs/netfs/read_retry.c:235
netfs_rreq_assess+0x5d3/0x870 fs/netfs/read_collect.c:371
netfs_rreq_terminated+0xe5/0x110 fs/netfs/read_collect.c:407
netfs_dispatch_unbuffered_reads fs/netfs/direct_read.c:103 [inline]
netfs_unbuffered_read fs/netfs/direct_read.c:127 [inline]
netfs_unbuffered_read_iter_locked+0x12f6/0x19b0 fs/netfs/direct_read.c:221
netfs_unbuffered_read_iter+0xc5/0x100 fs/netfs/direct_read.c:256
v9fs_file_read_iter+0xbf/0x100 fs/9p/vfs_file.c:361
do_iter_readv_writev+0x614/0x7f0 fs/read_write.c:832
vfs_readv+0x4cf/0x890 fs/read_write.c:1025
do_preadv fs/read_write.c:1142 [inline]
__do_sys_preadv fs/read_write.c:1192 [inline]
__se_sys_preadv fs/read_write.c:1187 [inline]
__x64_sys_preadv+0x22d/0x310 fs/read_write.c:1187
do_syscall_x64 arch/x86/entry/common.c:52 [inline]
do_syscall_64+0xcd/0x250 arch/x86/entry/common.c:83
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Closes: https://syzkaller.appspot.com/bug?extid=1fc6f64c40a9d143cfb6
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241108034020.3695718-1-lizhi.xu@windriver.com/ [1]
Link: https://lore.kernel.org/r/20241213135013.2964079-9-dhowells@redhat.com
Tested-by: syzbot+885c03ad650731743489@syzkaller.appspotmail.com
Suggested-by: Lizhi Xu <lizhi.xu@windriver.com>
cc: Dominique Martinet <asmadeus@codewreck.org>
cc: Jeff Layton <jlayton@kernel.org>
cc: v9fs@lists.linux.dev
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Reported-by: syzbot+885c03ad650731743489@syzkaller.appspotmail.com
Signed-off-by: Christian Brauner <brauner@kernel.org>
Use clear_and_wake_up_bit() rather than something like:
clear_bit_unlock(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
wake_up_bit(&rreq->flags, NETFS_RREQ_IN_PROGRESS);
as there needs to be a barrier inserted between which is present in
clear_and_wake_up_bit().
Fixes: 288ace2f57 ("netfs: New writeback implementation")
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20241213135013.2964079-8-dhowells@redhat.com
Reviewed-by: Akira Yokosawa <akiyks@gmail.com>
cc: Zilin Guan <zilin@seu.edu.cn>
cc: Akira Yokosawa <akiyks@gmail.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
When a read subrequest finishes, if it doesn't have sufficient coverage to
complete the folio(s) covering either side of it, it will donate the excess
coverage to the adjacent subrequests on either side, offloading
responsibility for unlocking the folio(s) covered to them.
Now, preference is given to donating down to a lower file offset over
donating up because that check is done first - but there's no check that
the lower subreq is actually contiguous, and so we can end up donating
incorrectly.
The scenario seen[1] is that an 8MiB readahead request spanning four 2MiB
folios is split into eight 1MiB subreqs (numbered 1 through 8). These
terminate in the order 1,6,2,5,3,7,4,8. What happens is:
- 1 donates to 2
- 6 donates to 5
- 2 completes, unlocking the first folio (with 1).
- 5 completes, unlocking the third folio (with 6).
- 3 donates to 4
- 7 donates to 4 incorrectly
- 4 completes, unlocking the second folio (with 3), but can't use
the excess from 7.
- 8 donates to 4, also incorrectly.
Fix this by preventing downward donation if the subreqs are not contiguous
(in the example above, 7 donates to 4 across the gap left by 5 and 6).
Reported-by: Shyam Prasad N <nspmangalore@gmail.com>
Closes: https://lore.kernel.org/r/CANT5p=qBwjBm-D8soFVVtswGEfmMtQXVW83=TNfUtvyHeFQZBA@mail.gmail.com/
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/526707.1733224486@warthog.procyon.org.uk/ [1]
Link: https://lore.kernel.org/r/20241213135013.2964079-3-dhowells@redhat.com
cc: Steve French <sfrench@samba.org>
cc: Paulo Alcantara <pc@manguebit.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: linux-cifs@vger.kernel.org
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
netfslib currently defers dropping the ref on the folios it obtains during
readahead to after it has started I/O on the basis that we can do it whilst
we wait for the I/O to complete, but this runs the risk of the I/O
collection racing with this in future.
Furthermore, Matthew Wilcox strongly suggests that the refs should be
dropped immediately, as readahead_folio() does (netfslib is using
__readahead_batch() which doesn't drop the refs).
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Suggested-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/3771538.1728052438@warthog.procyon.org.uk
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Improve the efficiency of buffered reads in a number of ways:
(1) Overhaul the algorithm in general so that it's a lot more compact and
split the read submission code between buffered and unbuffered
versions. The unbuffered version can be vastly simplified.
(2) Read-result collection is handed off to a work queue rather than being
done in the I/O thread. Multiple subrequests can be processes
simultaneously.
(3) When a subrequest is collected, any folios it fully spans are
collected and "spare" data on either side is donated to either the
previous or the next subrequest in the sequence.
Notes:
(*) Readahead expansion is massively slows down fio, presumably because it
causes a load of extra allocations, both folio and xarray, up front
before RPC requests can be transmitted.
(*) RDMA with cifs does appear to work, both with SIW and RXE.
(*) PG_private_2-based reading and copy-to-cache is split out into its own
file and altered to use folio_queue. Note that the copy to the cache
now creates a new write transaction against the cache and adds the
folios to be copied into it. This allows it to use part of the
writeback I/O code.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Link: https://lore.kernel.org/r/20240814203850.2240469-20-dhowells@redhat.com/ # v2
Signed-off-by: Christian Brauner <brauner@kernel.org>
