Fix netfs to erase the contents of a hole created after the EOF by an
ordinary write if dirty data has been previously left there by writes
through an mmapped region. Neither the buffered nor the unbuffered/DIO
write path clears that stale pagecache.
Zero the tail of the folio straddling the EOF before an extending
write. That is the only folio that can hold data written past the EOF
through an mmap, as pages wholly beyond the EOF can't be faulted in.
The folio is zeroed rather than dropped so a concurrent extending write
can't lose data.
Both write paths downgrade the i_rwsem to shared, so extending writes
can run concurrently and the i_size read by the caller may be stale by
the time the folio is locked. Re-read i_size under the folio lock and
clamp the zeroed range up to it, so a racing write that already put
data into the folio isn't clobbered.
Wait for any writeback on the folio to finish before zeroing it so that
the pagecache isn't modified while it may still be read by the transport
during transmission. Honour IOCB_NOWAIT by returning -EAGAIN rather
than blocking on the folio lock, on writeback, or in folio_mkclean()'s
rmap walk when the folio is mapped.
truncate_pagecache() can't be used here: it must be called with the
i_rwsem held exclusively, but these write paths only hold it shared,
and it would block unconditionally, breaking IOCB_NOWAIT.
Callers that hold i_rwsem exclusively for the whole resize (truncate,
setattr, fallocate, clone) exclude any genuine concurrent buffered
writer, so staleness can instead be decided from the folio's dirty
state, as pagecache_isize_extended() already does for filesystems that
serialise writes against truncate/setattr via a single i_rwsem.
Export netfs_clear_stale_post_isize() helper to handle such case.
The helper is required by the CIFS client to fix generic/363.
Closes: https://sashiko.dev/#/patchset/20260921230755.1133425-1-pc%40manguebit.org
Fixes: 938e13a73b ("netfs: Implement buffered write API")
Fixes: 153a9961b5 ("netfs: Implement unbuffered/DIO write support")
Reviewed-by: David Howells <[email protected]>
Reviewed-by: Namjae Jeon <[email protected]>
Signed-off-by: Paulo Alcantara <[email protected]>
Cc: Christian Brauner <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Ronnie Sahlberg <[email protected]>
Cc: Shyam Prasad N <[email protected]>
Cc: Tom Talpey <[email protected]>
Cc: Bharath SM <[email protected]>
Cc: [email protected]
For really big read RPC ops that span multiple folios, netfslib allows the
filesystem to give progress notifications to wake up the collector thread
to do a collection of folios that have now been fetched, even if the RPC is
still ongoing, thereby allowing the application to make progress.
This works by taking the current rreq->cleaned_to value (which indicates
which folios have been unlocked) and adding the stashed size of the next
folio to it. cleaned_to, however, is subject to 64-bit tearing on a 32-bit
arch.
Fix this by stashing the next progress notification point as a size_t
(which won't tear) to be added to rreq->start (which won't change), with
the collector thread calculating that from cleaned_to plus the next folio
size.
Further, however, if the folios are small, the collector thread gets
constantly woken up - which has a negative performance impact on the
system.
Fix that too by setting a minimum trigger of 256KiB or the size of the
folio at the front of the queue, whichever is larger. Note that this has
an issue that different subreqs have different need-to-be-cached
properties; this is solved by a preceding patch that marks the property on
the folios whilst issuing subreqs rather than when collecting them.
Also, make sure rreq->cleaned_to is initialised up front, along with
rreq->collected_to and stream->collected_to.
Fixes: e2d46f2ec3 ("netfs: Change the read result collector to only use one work item")
Link: https://sashiko.dev/#/patchset/20260804100224.2748935-1-dhowells%40redhat.com
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
Acked-by: Paulo Alcantara <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner (Amutable) <[email protected]>
Mark folios with NETFS_FOLIO_COPY_TO_CACHE whilst issuing subreqs rather than
when collecting them. This means that the collector thread doesn't have to
try and keep track of which subreqs contribute to which folios - and thus
which folios will need to be copied to the cache because at least one byte
wasn't in the cache. Instead, this is marked on the folios up front and the
collector need only consider the folios.
For PG_private_2-using filesystems, PG_private_2 is set instead of
NETFS_FOLIO_COPY_TO_CACHE, but otherwise it works the same.
The NETFS_RREQ_COPY_TO_CACHE is replaced with NETFS_RREQ_CANCEL_CACHING, which
is now set if caching fails somewhere, thereby causing the collection thread
to cancel the copy-to-cache marks on the remaining folios.
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
Acked-by: Paulo Alcantara <[email protected]>
cc: Paulo Alcantara (Red Hat) <[email protected]>
cc: Matthew Wilcox <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner (Amutable) <[email protected]>
There are some synchronisation issues that derive from the app thread
adding more folios to the rolling buffer whilst the collector thread is
looking at them or trying to clear them, such as determining the setting of
front_folio_order when the next folio hasn't been added yet,
The reason for the rolling buffer approach is that loading the buffer
upfront and then dropping all the refs just acquired is quite a slow
operation, and loading progressively allows some of the cost to be deferred
until after at least some of the I/O is started.
Instead, a better way is to load all the folios into the rolling buffer
upfront - and then drop the refs later, once the I/O is in progress. (Even
better would be for the refs not to be there at all.)
Fix this by changing the rolling buffer loader to load all the folios
selected by the VM for readahead upfront into the folio queue. The folio
queue is allocated a batch worth at a time as we don't know how many folios
are involved (the readahead_control struct, alas, has a page count, not a
folio count).
The folio refs acquired from readahead are then dropped in bulk once the
first subrequest is dispatched as it's quite a slow operation. The
collector waits for NETFS_RREQ_NEED_PUT_RA_REFS to be cleared so that it
doesn't unlock folios before the xarray has been scanned for them.
This simplifies the buffer handling later and isn't noticeably slower as
the xarray doesn't need to be modified and the folios are all already
pre-locked.
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Link: https://sashiko.dev/#/patchset/20260824120224.504575-1-dhowells%40redhat.com
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
Acked-by: Paulo Alcantara <[email protected]>
cc: Paulo Alcantara (Red Hat) <[email protected]>
cc: Matthew Wilcox <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner (Amutable) <[email protected]>
Fix the handling of folio_queue allocation failure in writeback by adding a
mempool and passing in gfp_t flags to the rolling buffer functions that
allocate memory, using the mempool if gfp != GFP_KERNEL.
This is then extended upwards and the gfp to be used for a request is stored
in the netfs_io_request struct and is then used for both requests and
subrequests, eliminating the sleeping loops there.
The failure caused:
folio != NULL
WARNING: fs/netfs/write_issue.c:603 at netfs_writepages+0x883/0xa10 fs/netfs/write_issue.c:603, CPU#3: syz.0.17/5919
Fixes: cd0277ed0c ("netfs: Use new folio_queue data type and iterator instead of xarray iter")
Reported-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=0da43efa72f88bd3a8af
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
Tested-by: [email protected]
cc: Paulo Alcantara <[email protected]>
cc: Yun Zhou <[email protected]>
cc: Matthew Wilcox <[email protected]>
cc: Christoph Hellwig <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner (Amutable) <[email protected]>
The netfs_inode::wb_lock mutex is used to prevent multiple simultaneous
writebacks from fighting each other (a writeback thread will write multiple
discontiguous regions within the same request). The mutex, however, only
serialises the issuing of subrequests; it doesn't serialise the collection
of results, and, in particular, the updating of file size information and
fscache populatedness data.
Unfortunately, the mutex cannot be held around the entire process as it has
to be unlocked in the same thread in which it is locked - and we don't want
to hold up the allocator whilst we complete the writeback.
Fix this by replacing the mutex with a bit flag and a list of lock waiters
so that the lock can be dropped in the collector thread after collection is
complete.
Link: https://sashiko.dev/#/patchset/20260608145432.681865-1-dhowells%40redhat.com
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
cc: Paulo Alcantara <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner (Amutable) <[email protected]>
netfs_unlock_abandoned_read_pages(rreq) accesses the index of the folios it
is wanting to unlock and compares that to rreq->no_unlock_folio so that it
doesn't unlock a folio being read for netfs_perform_write() or
netfs_write_begin().
However, given that netfs_unlock_abandoned_read_pages() is called _after_
NETFS_RREQ_IN_PROGRESS is cleared, the one folio that it's not allowed to
dereference is the one specified by ->no_unlock_folio as ownership
immediately reverts to the caller.
Fix this by storing the folio pointer instead and using that rather than
the index. Also fix netfs_unlock_read_folio() where the same applies.
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Closes: https://sashiko.dev/#/patchset/20260414082004.3756080-1-dhowells%40redhat.com
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
cc: Paulo Alcantara <[email protected]>
cc: Viacheslav Dubeyko <[email protected]>
cc: Matthew Wilcox <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
The netfs_io_stream::front member is meant to point to the subrequest
currently being collected on a stream, but it isn't actually used this way
by direct write (which mostly ignores it). However, there's a tracepoint
which looks at it. Further, stream->front is actually redundant with
stream->subrequests.next.
Fix the potential problem in the direct code by just removing the member
and using stream->subrequests.next instead, thereby also simplifying the
code.
Fixes: a0b4c7a491 ("netfs: Fix unbuffered/DIO writes to dispatch subrequests in strict sequence")
Reported-by: Paulo Alcantara <[email protected]>
Signed-off-by: David Howells <[email protected]>
Link: https://patch.msgid.link/[email protected]
Reviewed-by: Paulo Alcantara (Red Hat) <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
If all the subrequests in an unbuffered write stream fail, the subrequest
collector doesn't update the stream->transferred value and it retains its
initial LONG_MAX value. Unfortunately, if all active streams fail, then we
take the smallest value of { LONG_MAX, LONG_MAX, ... } as the value to set
in wreq->transferred - which is then returned from ->write_iter().
LONG_MAX was chosen as the initial value so that all the streams can be
quickly assessed by taking the smallest value of all stream->transferred -
but this only works if we've set any of them.
Fix this by adding a flag to indicate whether the value in
stream->transferred is valid and checking that when we integrate the
values. stream->transferred can then be initialised to zero.
This was found by running the generic/750 xfstest against cifs with
cache=none. It splices data to the target file. Once (if) it has used up
all the available scratch space, the writes start failing with ENOSPC.
This causes ->write_iter() to fail. However, it was returning
wreq->transferred, i.e. LONG_MAX, rather than an error (because it thought
the amount transferred was non-zero) and iter_file_splice_write() would
then try to clean up that amount of pipe bufferage - leading to an oops
when it overran. The kernel log showed:
CIFS: VFS: Send error in write = -28
followed by:
BUG: kernel NULL pointer dereference, address: 0000000000000008
with:
RIP: 0010:iter_file_splice_write+0x3a4/0x520
do_splice+0x197/0x4e0
or:
RIP: 0010:pipe_buf_release (include/linux/pipe_fs_i.h:282)
iter_file_splice_write (fs/splice.c:755)
Also put a warning check into splice to announce if ->write_iter() returned
that it had written more than it was asked to.
Fixes: 288ace2f57 ("netfs: New writeback implementation")
Reported-by: Xiaoli Feng <[email protected]>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220445
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/[email protected]
cc: Paulo Alcantara <[email protected]>
cc: Steve French <[email protected]>
cc: Shyam Prasad N <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
Pull misc VFS updates from Christian Brauner:
"This contains the usual selections of misc updates for this cycle.
Features:
- Add ext4 IOCB_DONTCACHE support
This refactors the address_space_operations write_begin() and
write_end() callbacks to take const struct kiocb * as their first
argument, allowing IOCB flags such as IOCB_DONTCACHE to propagate
to the filesystem's buffered I/O path.
Ext4 is updated to implement handling of the IOCB_DONTCACHE flag
and advertises support via the FOP_DONTCACHE file operation flag.
Additionally, the i915 driver's shmem write paths are updated to
bypass the legacy write_begin/write_end interface in favor of
directly calling write_iter() with a constructed synchronous kiocb.
Another i915 change replaces a manual write loop with
kernel_write() during GEM shmem object creation.
Cleanups:
- don't duplicate vfs_open() in kernel_file_open()
- proc_fd_getattr(): don't bother with S_ISDIR() check
- fs/ecryptfs: replace snprintf with sysfs_emit in show function
- vfs: Remove unnecessary list_for_each_entry_safe() from
evict_inodes()
- filelock: add new locks_wake_up_waiter() helper
- fs: Remove three arguments from block_write_end()
- VFS: change old_dir and new_dir in struct renamedata to dentrys
- netfs: Remove unused declaration netfs_queue_write_request()
Fixes:
- eventpoll: Fix semi-unbounded recursion
- eventpoll: fix sphinx documentation build warning
- fs/read_write: Fix spelling typo
- fs: annotate data race between poll_schedule_timeout() and
pollwake()
- fs/pipe: set FMODE_NOWAIT in create_pipe_files()
- docs/vfs: update references to i_mutex to i_rwsem
- fs/buffer: remove comment about hard sectorsize
- fs/buffer: remove the min and max limit checks in __getblk_slow()
- fs/libfs: don't assume blocksize <= PAGE_SIZE in
generic_check_addressable
- fs_context: fix parameter name in infofc() macro
- fs: Prevent file descriptor table allocations exceeding INT_MAX"
* tag 'vfs-6.17-rc1.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (24 commits)
netfs: Remove unused declaration netfs_queue_write_request()
eventpoll: fix sphinx documentation build warning
ext4: support uncached buffered I/O
mm/pagemap: add write_begin_get_folio() helper function
fs: change write_begin/write_end interface to take struct kiocb *
drm/i915: Refactor shmem_pwrite() to use kiocb and write_iter
drm/i915: Use kernel_write() in shmem object create
eventpoll: Fix semi-unbounded recursion
vfs: Remove unnecessary list_for_each_entry_safe() from evict_inodes()
fs/libfs: don't assume blocksize <= PAGE_SIZE in generic_check_addressable
fs/buffer: remove the min and max limit checks in __getblk_slow()
fs: Prevent file descriptor table allocations exceeding INT_MAX
fs: Remove three arguments from block_write_end()
fs/ecryptfs: replace snprintf with sysfs_emit in show function
fs: annotate suspected data race between poll_schedule_timeout() and pollwake()
docs/vfs: update references to i_mutex to i_rwsem
fs/buffer: remove comment about hard sectorsize
fs_context: fix parameter name in infofc() macro
VFS: change old_dir and new_dir in struct renamedata to dentrys
proc_fd_getattr(): don't bother with S_ISDIR() check
...
Renumber the NETFS_RREQ_* flags to put the most useful status bits in the
bottom nibble - and therefore the last hex digit in the trace output -
making it easier to grasp the state at a glance.
In particular, put the IN_PROGRESS flag in bit 0 and ALL_QUEUED at bit 1.
Also make the flags field in /proc/fs/netfs/requests larger to accommodate
all the flags.
Also make the flags field in the netfs_sreq tracepoint larger to
accommodate all the NETFS_SREQ_* flags.
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/[email protected]
Reviewed-by: Paulo Alcantara <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
Netfslib has two functions for updating the i_size after a write: one for
buffered writes into the pagecache and one for direct/unbuffered writes.
However, what needs to be done is much the same in both cases, so merge
them together.
This does raise one question, though: should updating the i_size after a
direct write do the same estimated update of i_blocks as is done for
buffered writes.
Also get rid of the cleanup function pointer from netfs_io_request as it's
only used for direct write to update i_size; instead do the i_size setting
directly from write collection.
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/[email protected]
cc: Steve French <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
David Howells <[email protected]> says:
Here are some miscellaneous fixes and changes for netfslib, if you could
pull them:
(1) Fix an oops in write-retry due to mis-resetting the I/O iterator.
(2) Fix the recording of transferred bytes for short DIO reads.
(3) Fix a request's work item to not require a reference, thereby avoiding
the need to get rid of it in BH/IRQ context.
(4) Fix waiting and waking to be consistent about the waitqueue used.
* patches from https://lore.kernel.org/[email protected]:
netfs: Fix wait/wake to be consistent about the waitqueue used
netfs: Fix the request's work item to not require a ref
netfs: Fix setting of transferred bytes with short DIO reads
netfs: Fix oops in write-retry from mis-resetting the subreq iterator
Link: https://lore.kernel.org/[email protected]
Signed-off-by: Christian Brauner <[email protected]>
When the netfs_io_request struct's work item is queued, it must be supplied
with a ref to the work item struct to prevent it being deallocated whilst
on the queue or whilst it is being processed. This is tricky to manage as
we have to get a ref before we try and queue it and then we may find it's
already queued and is thus already holding a ref - in which case we have to
try and get rid of the ref again.
The problem comes if we're in BH or IRQ context and need to drop the ref:
if netfs_put_request() reduces the count to 0, we have to do the cleanup -
but the cleanup may need to wait.
Fix this by adding a new work item to the request, ->cleanup_work, and
dispatching that when the refcount hits zero. That can then synchronously
cancel any outstanding work on the main work item before doing the cleanup.
Adding a new work item also deals with another problem upstream where it's
sometimes changing the work func in the put function and requeuing it -
which has occasionally in the past caused the cleanup to happen
incorrectly.
As a bonus, this allows us to get rid of the 'was_async' parameter from a
bunch of functions. This indicated whether the put function might not be
permitted to sleep.
Fixes: 3d3c950467 ("netfs: Provide readahead and readpage netfs helpers")
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/[email protected]
cc: Paulo Alcantara <[email protected]>
cc: Marc Dionne <[email protected]>
cc: Steve French <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
A netfslib request comprises an ordered stream of subrequests that, when
doing an unbuffered/DIO read, are contiguous. The subrequests may be
performed in parallel, but may not be fully completed.
For instance, if we try and make a 256KiB DIO read from a 3-byte file with
a 64KiB rsize and 256KiB bsize, netfslib will attempt to make a read of
256KiB, broken up into four 64KiB subreads, with the expectation that the
first will be short and the subsequent three be completely devoid - but we
do all four on the basis that the file may have been changed by a third
party.
The read-collection code, however, walks through all the subreqs and
advances the notion of how much data has been read in the stream to the
start of each subreq plus its amount transferred (which are 3, 0, 0, 0 for
the example above) - which gives an amount apparently read of 3*64KiB -
which is incorrect.
Fix the collection code to cut short the calculation of the transferred
amount with the first short subrequest in an unbuffered read; everything
beyond that must be ignored as there's a hole that cannot be filled. This
applies both to shortness due to hitting the EOF and shortness due to an
error.
This is achieved by setting a flag on the request when we collect the first
short subrequest (collection is done in ascending order).
This can be tested by mounting a cifs volume with rsize=65536,bsize=262144
and doing a 256k DIO read of a very small file (e.g. 3 bytes). read()
should return 3, not >3.
This problem came in when netfs_read_collection() set rreq->transferred to
stream->transferred, even for DIO. Prior to that, netfs_rreq_assess_dio()
just went over the list and added up the subreqs till it met a short one -
but now the subreqs are discarded earlier.
Fixes: e2d46f2ec3 ("netfs: Change the read result collector to only use one work item")
Reported-by: Nicolas Baranger <[email protected]>
Closes: https://lore.kernel.org/all/[email protected]/
Signed-off-by: "Paulo Alcantara (Red Hat)" <[email protected]>
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/[email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
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 <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Tested-by: Marc Dionne <[email protected]>
Tested-by: Steve French <[email protected]>
cc: Ihor Solodrai <[email protected]>
cc: Eric Van Hensbergen <[email protected]>
cc: Latchesar Ionkov <[email protected]>
cc: Dominique Martinet <[email protected]>
cc: Christian Schoenebeck <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
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 <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
Add two netfslib functions to build up or clean up a buffer in a
folio_queue. The first, netfs_alloc_folioq_buffer() will add folios to a
buffer, extending up at least to the given size. If it can, it will add
multipage folios. The folios are optionally have the mapping set and will
have the index set according to the distance from the front of the folio
queue.
The second function will free up a folio queue and put any folios in the
queue that have the first mark set.
The netfs_folio tracepoint is also altered to cope with folios that have a
NULL mapping, and the folios being added/put will have trace lines emitted
and will be accounted in the stats.
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
cc: Jeff Layton <[email protected]>
cc: Marc Dionne <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
A rolling buffer is a series of folios held in a list of folio_queues. New
folios and folio_queue structs may be inserted at the head simultaneously
with spent ones being removed from the tail without the need for locking.
The rolling buffer includes an iov_iter and it has to be careful managing
this as the list of folio_queues is extended such that an oops doesn't
incurred because the iterator was pointing to the end of a folio_queue
segment that got appended to and then removed.
We need to use the mechanism twice, once for read and once for write, and,
in future patches, we will use a second rolling buffer to handle bounce
buffering for content encryption.
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
netfs: Fix is-caching check in read-retry
The read-retry code checks the NETFS_RREQ_COPY_TO_CACHE flag to determine
if there might be failed reads from the cache that need turning into reads
from the server, with the intention of skipping the complicated part if it
can. The code that set the flag, however, got lost during the read-side
rewrite.
Fix the check to see if the cache_resources are valid instead. The flag
can then be removed.
Fixes: ee4cdf7ba8 ("netfs: Speed up buffered reading")
Signed-off-by: David Howells <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
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 <[email protected]>
Link: https://lore.kernel.org/r/[email protected]/ [1]
Link: https://lore.kernel.org/r/[email protected]
Tested-by: [email protected]
Suggested-by: Lizhi Xu <[email protected]>
cc: Dominique Martinet <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Reported-by: [email protected]
Signed-off-by: Christian Brauner <[email protected]>
Because it uses DIO writes, cachefiles is unable to make a write to the
backing file if that write is not aligned to and sized according to the
backing file's DIO block alignment. This makes it tricky to handle a write
to the cache where the EOF on the network file is not correctly aligned.
To get around this, netfslib attempts to tell the driver it is calling how
much more data there is available beyond the EOF that it can use to pad the
write (netfslib preclears the part of the folio above the EOF). However,
it tries to tell the cache what the maximum length is, but doesn't
calculate this correctly; and, in any case, cachefiles actually ignores the
value and just skips the block.
Fix this by:
(1) Change the value passed to indicate the amount of extra data that can
be added to the operation (now ->submit_extendable_to). This is much
simpler to calculate as it's just the end of the folio minus the top
of the data within the folio - rather than having to account for data
spread over multiple folios.
(2) Make cachefiles add some of this data if the subrequest it is given
ends at the network file's i_size if the extra data is sufficient to
pad out to a whole block.
Signed-off-by: David Howells <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]/ # v2
Signed-off-by: Christian Brauner <[email protected]>
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 <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]/ # v2
Signed-off-by: Christian Brauner <[email protected]>
Use the new folio_queue structures to simplify the writeback code. The
problem with referring to the i_pages xarray directly is that we may have
gaps in the sequence of folios we're writing from that we need to skip when
we're removing the writeback mark from the folios we're writing back from.
At the moment the code tries to deal with this by carefully tracking the
gaps in each writeback stream (eg. write to server and write to cache) and
divining when there's a gap that spans folios (something that's not helped
by folios not being a consistent size).
Instead, the folio_queue buffer contains pointers only the folios we're
dealing with, has them in ascending order and indicates a gap by placing
non-consequitive folios next to each other. This makes it possible to
track where we need to clean up to by just keeping track of where we've
processed to on each stream and taking the minimum.
Note that the I/O iterator is always rounded up to the end of the folio,
even if that is beyond the EOF position, so that the cache can do DIO from
the page. The excess space is cleared, though mmapped writes clobber it.
Signed-off-by: David Howells <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]/ # v2
Signed-off-by: Christian Brauner <[email protected]>
Make the netfs write-side routines use the new folio_queue struct to hold a
rolling buffer of folios, with the issuer adding folios at the tail and the
collector removing them from the head as they're processed instead of using
an xarray.
This will allow a subsequent patch to simplify the write collector.
The primary mark (as tested by folioq_is_marked()) is used to note if the
corresponding folio needs putting.
Signed-off-by: David Howells <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]/ # v2
Signed-off-by: Christian Brauner <[email protected]>
Short DIO reads, particularly in relation to cifs, are not being handled
correctly by cifs and netfslib. This can be tested by doing a DIO read of
a file where the size of read is larger than the size of the file. When it
crosses the EOF, it gets a short read and this gets retried, and in the
case of cifs, the retry read fails, with the failure being translated to
ENODATA.
Fix this by the following means:
(1) Add a flag, NETFS_SREQ_HIT_EOF, for the filesystem to set when it
detects that the read did hit the EOF.
(2) Make the netfslib read assessment stop processing subrequests when it
encounters one with that flag set.
(3) Return rreq->transferred, the accumulated contiguous amount read to
that point, to userspace for a DIO read.
(4) Make cifs set the flag and clear the error if the read RPC returned
ENODATA.
(5) Make cifs set the flag and clear the error if a short read occurred
without error and the read-to file position is now at the remote inode
size.
Fixes: 69c3c023af ("cifs: Implement netfslib hooks")
Signed-off-by: David Howells <[email protected]>
cc: Steve French <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
cc: [email protected]
Signed-off-by: Steve French <[email protected]>
Paulo Alcantara