netfs: Fix io_uring based write-through
Subsystems (5)
David Howells
maintainerPATCH v2 1/1 netfs: Fix io_uring based write-through
This can be triggered by mounting a cifs filesystem with a cache=strict mount option and then, using the fsx program from xfstests, doing:
ltp/fsx -A -d -N 1000 -S 11463 -P /tmp /cifs-mount/foo \
--replay-ops=gen112-fsxopsWhere gen112-fsxops holds:
fallocate 0x6be7 0x8fc5 0x377d3
copy_range 0x9c71 0x77e8 0x2edaf 0x377d3
write 0x2776d 0x8f65 0x377d3The problem is that netfs_io_request::len is being used for two purposes and ends up getting set to the amount of data we transferred, not the amount of data the caller asked to be transferred (for various reasons, such as mmap'd writes, we might end up rounding out the data written to the server to include the entire folio at each end).
Fix this by keeping the amount we were asked to write in ->len and using ->submitted to track what we issued ops for. Then, when we come to calling ->ki_complete(), ->len is the right size.
This also required netfs_cleanup_dio_write() to change since we're no longer advancing wreq->len. Use wreq->transferred instead as we might have done a short read and wreq->len must be set when setting up a direct write.
With this, the generic/112 xfstest passes if cifs is forced to put all non-DIO opens into write-through mode.
---
Changes
=======
ver #2)
- Set wreq->len when doing direct writes.
fs/netfs/direct_write.c | 5 +++-- fs/netfs/write_collect.c | 7 ++++--- fs/netfs/write_issue.c | 2 +- 3 files changed, 8 insertions(+), 6 deletions(-)
fs/netfs/direct_write.c +3 -2
@@ -12,7 +12,7 @@static void netfs_cleanup_dio_write (struct netfs_io_request *wreq) {struct inode *inode = wreq->inode; -unsigned long long end = wreq->start + wreq->len; +unsigned long long end = wreq->start + wreq->transferred;if (!wreq->error &&i_size_read (inode) < end) {
@@ -92,8 +92,9 @@ static ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov__set_bit (NETFS_RREQ_UPLOAD_TO_SERVER , &wreq->flags);if (async) wreq->iocb = iocb; + wreq->len =iov_iter_count (&wreq->io_iter); wreq->cleanup = netfs_cleanup_dio_write; - ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb),iov_iter_count (&wreq->io_iter)); + ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb), wreq->len);if (ret <0 ) {_debug ("begin = %zd" , ret); goto out;
fs/netfs/write_collect.c +4 -3
@@ -510,7 +510,7 @@ static void netfs_collect_write_results(struct netfs_io_request *wreq)* stream has a gap that can be jumped. */ if (notes &SOME_EMPTY ) { -unsigned long long jump_to = wreq->start + wreq->len; +unsigned long long jump_to = wreq->start +READ_ONCE (wreq->submitted);for (s =0 ; s <NR_IO_STREAMS ; s++) { stream = &wreq->io_streams[s];
@@ -690,10 +690,11 @@ void netfs_write_collection_worker(struct work_struct *work)wake_up_bit (&wreq->flags,NETFS_RREQ_IN_PROGRESS );if (wreq->iocb) { - wreq->iocb->ki_pos += wreq->transferred; +size_t written =min (wreq->transferred, wreq->len); + wreq->iocb->ki_pos += written;if (wreq->iocb->ki_complete) wreq->iocb->ki_complete ( - wreq->iocb, wreq->error ? wreq->error : wreq->transferred); + wreq->iocb, wreq->error ? wreq->error : written); wreq->iocb =VFS_PTR_POISON ; }
fs/netfs/write_issue.c +1 -1
@@ -254,7 +254,7 @@ static void netfs_issue_write(struct netfs_io_request *wreq, stream->construct =NULL ;if (subreq->start + subreq->len > wreq->start + wreq->submitted) - wreq->len = wreq->submitted = subreq->start + subreq->len - wreq->start; +WRITE_ONCE (wreq->submitted, subreq->start + subreq->len - wreq->start);netfs_do_issue_write (stream, subreq); }
fixed minor checpatch warning (updated patch attached)
Quoted 117 lines · David Howells
This can be triggered by mounting a cifs filesystem with a cache=strict mount option and then, using the fsx program from xfstests, doing:
ltp/fsx -A -d -N 1000 -S 11463 -P /tmp /cifs-mount/foo \ --replay-ops=gen112-fsxopsWhere gen112-fsxops holds:
fallocate 0x6be7 0x8fc5 0x377d3 copy_range 0x9c71 0x77e8 0x2edaf 0x377d3 write 0x2776d 0x8f65 0x377d3The problem is that
netfs_io_request::lenis being used for two purposes and ends up getting set to the amount of data we transferred, not the amount of data the caller asked to be transferred (for various reasons, such as mmap'd writes, we might end up rounding out the data written to the server to include the entire folio at each end).Fix this by keeping the amount we were asked to write in ->len and using ->submitted to track what we issued ops for. Then, when we come to calling ->ki_complete(), ->len is the right size.
This also required
netfs_cleanup_dio_write()to change since we're no longer advancingwreq->len. Usewreq->transferredinstead as we might have done a short read andwreq->lenmust be set when setting up a direct write.With this, the generic/112 xfstest passes if cifs is forced to put all non-DIO opens into write-through mode.
Fixes 288ace2f57c9 ("netfs: New writeback implementation")Signed-off-by David Howells <[email protected]>cc Jeff Layton <[email protected]>cc Steve French <[email protected]>cc Enzo Matsumiya <[email protected]>---
Changes
=======
ver #2)
- Setwreq->lenwhen doing direct writes.fs/netfs/direct_write.c | 5 +++-- fs/netfs/write_collect.c | 7 ++++--- fs/netfs/write_issue.c | 2 +- 3 files changed, 8 insertions(+), 6 deletions(-)fs/netfs/direct_write.c +3
-2@@ -12,7 +12,7 @@static void netfs_cleanup_dio_write (struct netfs_io_request *wreq) {struct inode *inode = wreq->inode; -unsigned long long end = wreq->start + wreq->len; +unsigned long long end = wreq->start + wreq->transferred;if (!wreq->error &&i_size_read (inode) < end) {@@ -92,8 +92,9 @@ static ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov__set_bit (NETFS_RREQ_UPLOAD_TO_SERVER , &wreq->flags);if (async) wreq->iocb = iocb; + wreq->len =iov_iter_count (&wreq->io_iter); wreq->cleanup = netfs_cleanup_dio_write; - ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb),iov_iter_count (&wreq->io_iter)); + ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb), wreq->len);if (ret <0 ) {_debug ("begin = %zd" , ret); goto out;fs/netfs/write_collect.c +4
-3@@ -510,7 +510,7 @@ static void netfs_collect_write_results(struct netfs_io_request *wreq)* stream has a gap that can be jumped. */ if (notes &SOME_EMPTY ) { -unsigned long long jump_to = wreq->start + wreq->len; +unsigned long long jump_to = wreq->start +READ_ONCE (wreq->submitted);for (s =0 ; s <NR_IO_STREAMS ; s++) { stream = &wreq->io_streams[s];@@ -690,10 +690,11 @@ void netfs_write_collection_worker(struct work_struct *work)wake_up_bit (&wreq->flags,NETFS_RREQ_IN_PROGRESS );if (wreq->iocb) { - wreq->iocb->ki_pos += wreq->transferred; +size_t written =min (wreq->transferred, wreq->len); + wreq->iocb->ki_pos += written;if (wreq->iocb->ki_complete) wreq->iocb->ki_complete ( - wreq->iocb, wreq->error ? wreq->error : wreq->transferred); + wreq->iocb, wreq->error ? wreq->error : written); wreq->iocb =VFS_PTR_POISON ; }fs/netfs/write_issue.c +1
-1@@ -254,7 +254,7 @@ static void netfs_issue_write(struct netfs_io_request *wreq, stream->construct =NULL ;if (subreq->start + subreq->len > wreq->start + wreq->submitted) - wreq->len = wreq->submitted = subreq->start + subreq->len - wreq->start; +WRITE_ONCE (wreq->submitted, subreq->start + subreq->len - wreq->start);netfs_do_issue_write (stream, subreq); }
Signature · 3 lines
Thanks, Steve
From 73263d2e6023e4d7ef2dc6d917e01d38fd2a1a28 Mon Sep 17 00:00:00 2001
From: David Howells <[email protected]>
Date: Wed, 22 May 2024 01:02:30 +0100
Subject: [PATCH] netfs: Fix io_uring based write-through
This can be triggered by mounting a cifs filesystem with a cache=strict mount option and then, using the fsx program from xfstests, doing:
ltp/fsx -A -d -N 1000 -S 11463 -P /tmp /cifs-mount/foo \
--replay-ops=gen112-fsxopsWhere gen112-fsxops holds:
fallocate 0x6be7 0x8fc5 0x377d3
copy_range 0x9c71 0x77e8 0x2edaf 0x377d3
write 0x2776d 0x8f65 0x377d3The problem is that netfs_io_request::len is being used for two purposes and ends up getting set to the amount of data we transferred, not the amount of data the caller asked to be transferred (for various reasons, such as mmap'd writes, we might end up rounding out the data written to the server to include the entire folio at each end).
Fix this by keeping the amount we were asked to write in ->len and using ->submitted to track what we issued ops for. Then, when we come to calling ->ki_complete(), ->len is the right size.
This also required netfs_cleanup_dio_write() to change since we're no longer advancing wreq->len. Use wreq->transferred instead as we might have done a short read and wreq->len must be set when setting up a direct write.
With this, the generic/112 xfstest passes if cifs is forced to put all non-DIO opens into write-through mode.
--- fs/netfs/direct_write.c | 5 +++-- fs/netfs/write_collect.c | 8 +++++--- fs/netfs/write_issue.c | 2 +- 3 files changed, 9 insertions(+), 6 deletions(-)
fs/netfs/direct_write.c +3 -2
@@ -12,7 +12,7 @@static void netfs_cleanup_dio_write (struct netfs_io_request *wreq) {struct inode *inode = wreq->inode; -unsigned long long end = wreq->start + wreq->len; +unsigned long long end = wreq->start + wreq->transferred;if (!wreq->error &&i_size_read (inode) < end) {
@@ -92,8 +92,9 @@ ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *__set_bit (NETFS_RREQ_UPLOAD_TO_SERVER , &wreq->flags);if (async) wreq->iocb = iocb; + wreq->len =iov_iter_count (&wreq->io_iter); wreq->cleanup = netfs_cleanup_dio_write; - ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb),iov_iter_count (&wreq->io_iter)); + ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb), wreq->len);if (ret <0 ) {_debug ("begin = %zd" , ret); goto out;
fs/netfs/write_collect.c +5 -3
@@ -510,7 +510,7 @@ static void netfs_collect_write_results(struct netfs_io_request *wreq)* stream has a gap that can be jumped. */ if (notes &SOME_EMPTY ) { -unsigned long long jump_to = wreq->start + wreq->len; +unsigned long long jump_to = wreq->start +READ_ONCE (wreq->submitted);for (s =0 ; s <NR_IO_STREAMS ; s++) { stream = &wreq->io_streams[s];
@@ -690,10 +690,12 @@ void netfs_write_collection_worker(struct work_struct *work)wake_up_bit (&wreq->flags,NETFS_RREQ_IN_PROGRESS );if (wreq->iocb) { - wreq->iocb->ki_pos += wreq->transferred; +size_t written =min (wreq->transferred, wreq->len); + + wreq->iocb->ki_pos += written;if (wreq->iocb->ki_complete) wreq->iocb->ki_complete ( - wreq->iocb, wreq->error ? wreq->error : wreq->transferred); + wreq->iocb, wreq->error ? wreq->error : written); wreq->iocb =VFS_PTR_POISON ; }
fs/netfs/write_issue.c +1 -1
@@ -254,7 +254,7 @@ static void netfs_issue_write(struct netfs_io_request *wreq, stream->construct =NULL ;if (subreq->start + subreq->len > wreq->start + wreq->submitted) - wreq->len = wreq->submitted = subreq->start + subreq->len - wreq->start; +WRITE_ONCE (wreq->submitted, subreq->start + subreq->len - wreq->start);netfs_do_issue_write (stream, subreq); }
Signature · 1 line
2.40.1
Can add my Tested-by if you would like. Hopefully this (and "netfs:
Fix setting of BDP_ASYNC from iocb flags") can be in mainline relatively soon as they both fix some issues we hit with netfs testing
cifs.ko. There are still a few more netfs bugs to work through but
this is progress.
Quoted 127 lines · Steve French
fixed minor checpatch warning (updated patch attached)
Quoted 109 lines · David Howells
This can be triggered by mounting a cifs filesystem with a cache=strict mount option and then, using the fsx program from xfstests, doing:
ltp/fsx -A -d -N 1000 -S 11463 -P /tmp /cifs-mount/foo \ --replay-ops=gen112-fsxopsWhere gen112-fsxops holds:
fallocate 0x6be7 0x8fc5 0x377d3 copy_range 0x9c71 0x77e8 0x2edaf 0x377d3 write 0x2776d 0x8f65 0x377d3The problem is that
netfs_io_request::lenis being used for two purposes and ends up getting set to the amount of data we transferred, not the amount of data the caller asked to be transferred (for various reasons, such as mmap'd writes, we might end up rounding out the data written to the server to include the entire folio at each end).Fix this by keeping the amount we were asked to write in ->len and using ->submitted to track what we issued ops for. Then, when we come to calling ->ki_complete(), ->len is the right size.
This also required
netfs_cleanup_dio_write()to change since we're no longer advancingwreq->len. Usewreq->transferredinstead as we might have done a short read andwreq->lenmust be set when setting up a direct write.With this, the generic/112 xfstest passes if cifs is forced to put all non-DIO opens into write-through mode.
Fixes 288ace2f57c9 ("netfs: New writeback implementation")Signed-off-by David Howells <[email protected]>cc Jeff Layton <[email protected]>cc Steve French <[email protected]>cc Enzo Matsumiya <[email protected]>---
Changes
=======
ver #2)
- Setwreq->lenwhen doing direct writes.fs/netfs/direct_write.c | 5 +++-- fs/netfs/write_collect.c | 7 ++++--- fs/netfs/write_issue.c | 2 +- 3 files changed, 8 insertions(+), 6 deletions(-)fs/netfs/direct_write.c +3
-2@@ -12,7 +12,7 @@static void netfs_cleanup_dio_write (struct netfs_io_request *wreq) {struct inode *inode = wreq->inode; -unsigned long long end = wreq->start + wreq->len; +unsigned long long end = wreq->start + wreq->transferred;if (!wreq->error &&i_size_read (inode) < end) {@@ -92,8 +92,9 @@ static ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov__set_bit (NETFS_RREQ_UPLOAD_TO_SERVER , &wreq->flags);if (async) wreq->iocb = iocb; + wreq->len =iov_iter_count (&wreq->io_iter); wreq->cleanup = netfs_cleanup_dio_write; - ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb),iov_iter_count (&wreq->io_iter)); + ret =netfs_unbuffered_write (wreq,is_sync_kiocb (iocb), wreq->len);if (ret <0 ) {_debug ("begin = %zd" , ret); goto out;fs/netfs/write_collect.c +4
-3@@ -510,7 +510,7 @@ static void netfs_collect_write_results(struct netfs_io_request *wreq)* stream has a gap that can be jumped. */ if (notes &SOME_EMPTY ) { -unsigned long long jump_to = wreq->start + wreq->len; +unsigned long long jump_to = wreq->start +READ_ONCE (wreq->submitted);for (s =0 ; s <NR_IO_STREAMS ; s++) { stream = &wreq->io_streams[s];@@ -690,10 +690,11 @@ void netfs_write_collection_worker(struct work_struct *work)wake_up_bit (&wreq->flags,NETFS_RREQ_IN_PROGRESS );if (wreq->iocb) { - wreq->iocb->ki_pos += wreq->transferred; +size_t written =min (wreq->transferred, wreq->len); + wreq->iocb->ki_pos += written;if (wreq->iocb->ki_complete) wreq->iocb->ki_complete ( - wreq->iocb, wreq->error ? wreq->error : wreq->transferred); + wreq->iocb, wreq->error ? wreq->error : written); wreq->iocb =VFS_PTR_POISON ; }fs/netfs/write_issue.c +0
-0fs/netfs/write_issue.c static void netfs_issue_write(struct netfs_io_request *wreq, · line 260
Quoted 8 lines
fs/netfs/write_issue.c +1
-1@@ -254,7 +254,7 @@ static void netfs_issue_write(struct netfs_io_request *wreq, stream->construct =NULL ;if (subreq->start + subreq->len > wreq->start + wreq->submitted) - wreq->len = wreq->submitted = subreq->start + subreq->len - wreq->start; +WRITE_ONCE (wreq->submitted, subreq->start + subreq->len - wreq->start);netfs_do_issue_write (stream, subreq); }--
Thanks,Steve
Signature · 3 lines
Thanks, Steve