loop: Fix lost writes caused by missing flag
Subsystems (3)
PATCH 1/1 loop: Fix lost writes caused by missing flag
The following commit:
commit aa4d86163e4e ("block: loop: switch to VFS ITER_BVEC")
replaced __do_lo_send_write(), which used ITER_KVEC iterators, with lo_write_bvec() which uses ITER_BVEC iterators. In this change, though, the WRITE flag was lost:
- iov_iter_kvec(&from, ITER_KVEC | WRITE, &kvec, 1, len);
+ iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
This flag is necessary for the DAX case because we make decisions based on whether or not the iterator is a READ or a WRITE in dax_iomap_actor() and in dax_iomap_rw().
We end up going through this path in configurations where we combine a PMEM device with 4k sectors, a loopback device and DAX. The consequence of this missed flag is that what we intend as a write actually turns into a read in the DAX code, so no data is ever written.
The very simplest test case is to create a loopback device and try and write a small string to it, then hexdump a few bytes of the device to see if the write took. Without this patch you read back all zeros, with this you read back the string you wrote.
For XFS this causes us to fail or panic during the following xfstests:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250
For ext4 we have a similar issue where writes never happen, but we don't currently have any xfstests that use loopback and show this issue.
Fix this by restoring the WRITE flag argument to iov_iter_bvec(). This causes the xfstests to all pass.
--- drivers/block/loop.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/block/loop.c +1 -1
@@ -266,7 +266,7 @@ static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)struct iov_iter i;ssize_t bw; -iov_iter_bvec (&i,ITER_BVEC , bvec,1 , bvec->bv_len); +iov_iter_bvec (&i,ITER_BVEC |WRITE , bvec,1 , bvec->bv_len);file_start_write (file); bw =vfs_iter_write (file, &i, ppos,0 );
Signature · 6 lines
2.14.3 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Looks good:
Can you wire up your test cases for blktests?
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Quoted 5 lines · Christoph Hellwig
Looks good:
Reviewed-by Christoph Hellwig <[email protected]>Can you wire up your test cases for blktests?
Is blktests really the right place for this test? This failure is highly dependent on the configuration of the filesystem that is holding the file that we are using for the loopback device. It doesn't seem like blktests has support for mount options (dax), etc?
Because of the interaction with the underlying filesystem this seems like a better fit with xfstests to me, but I don't know if we need to add tests there because we already have pretty good coverage of loopback device failures. That's how we found this - this bug causes all these tests to fail:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Quoted 12 lines · Ross Zwisler
Quoted 5 lines · Christoph Hellwig
Looks good:
Reviewed-by Christoph Hellwig <[email protected]>Can you wire up your test cases for blktests?
Is blktests really the right place for this test? This failure is highly dependent on the configuration of the filesystem that is holding the file that we are using for the loopback device. It doesn't seem like blktests has support for mount options (dax), etc?
Quoted 5 lines
Because of the interaction with the underlying filesystem this seems like a better fit with xfstests to me, but I don't know if we need to add tests there because we already have pretty good coverage of loopback device failures. That's how we found this - this bug causes all these tests to fail:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250
The problem is that those tests don't configure the device in 4K sector mode, so we're still missing a regression test. That seems to be where blktests can come into play.
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Quoted 58 lines · Ross Zwisler
The following commit:
commit
aa4d86163e4e("block: loop: switch to VFSITER_BVEC")replaced
__do_lo_send_write(), which usedITER_KVECiterators, withlo_write_bvec()which usesITER_BVECiterators. In this change, though, the WRITE flag was lost:(quoted diff) +1
-1- iov_iter_kvec(&from, ITER_KVEC | WRITE, &kvec, 1, len); + iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);This flag is necessary for the DAX case because we make decisions based on whether or not the iterator is a READ or a WRITE in
dax_iomap_actor()and indax_iomap_rw().We end up going through this path in configurations where we combine a PMEM device with 4k sectors, a loopback device and DAX. The consequence of this missed flag is that what we intend as a write actually turns into a read in the DAX code, so no data is ever written.
The very simplest test case is to create a loopback device and try and write a small string to it, then hexdump a few bytes of the device to see if the write took. Without this patch you read back all zeros, with this you read back the string you wrote.
For XFS this causes us to fail or panic during the following xfstests:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250For ext4 we have a similar issue where writes never happen, but we don't currently have any xfstests that use loopback and show this issue.
Fix this by restoring the WRITE flag argument to
iov_iter_bvec(). This causes the xfstests to all pass.Signed-off-by Ross Zwisler <[email protected]>Fixes commit aa4d86163e4e ("block: loop: switch to VFS ITER_BVEC")Cc Christoph Hellwig <[email protected]>Cc Al Viro <[email protected]>--- drivers/block/loop.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/block/loop.c +1
-2@@ -266,7 +266,7 @@ static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)struct iov_iter i;ssize_t bw; -iov_iter_bvec (&i,ITER_BVEC , bvec,1 , bvec->bv_len); +iov_iter_bvec (&i,ITER_BVEC |WRITE , bvec,1 , bvec->bv_len);file_start_write (file); bw =vfs_iter_write (file, &i, ppos,0 ); --
Quoted 2 lines
2.14.3
Signature · 5 lines
Ming Lei _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
This has gotten Reviewed-by tags from Christoph and Ming Lei.
Al, are you the right person to merge this? Or is the correct person Jens, whom I accidentally didn't include when I sent this out?
Just wanted to make sure this got merged, and to see whether it was targeting v4.16 or v4.17.
Thanks,
- Ross
Quoted 60 lines · Ross Zwisler
The following commit:
commit
aa4d86163e4e("block: loop: switch to VFSITER_BVEC")replaced
__do_lo_send_write(), which usedITER_KVECiterators, withlo_write_bvec()which usesITER_BVECiterators. In this change, though, the WRITE flag was lost:(quoted diff) +1
-1- iov_iter_kvec(&from, ITER_KVEC | WRITE, &kvec, 1, len); + iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);This flag is necessary for the DAX case because we make decisions based on whether or not the iterator is a READ or a WRITE in
dax_iomap_actor()and indax_iomap_rw().We end up going through this path in configurations where we combine a PMEM device with 4k sectors, a loopback device and DAX. The consequence of this missed flag is that what we intend as a write actually turns into a read in the DAX code, so no data is ever written.
The very simplest test case is to create a loopback device and try and write a small string to it, then hexdump a few bytes of the device to see if the write took. Without this patch you read back all zeros, with this you read back the string you wrote.
For XFS this causes us to fail or panic during the following xfstests:
xfs/074 xfs/078 xfs/216 xfs/217 xfs/250For ext4 we have a similar issue where writes never happen, but we don't currently have any xfstests that use loopback and show this issue.
Fix this by restoring the WRITE flag argument to
iov_iter_bvec(). This causes the xfstests to all pass.Signed-off-by Ross Zwisler <[email protected]>Fixes commit aa4d86163e4e ("block: loop: switch to VFS ITER_BVEC")Cc Christoph Hellwig <[email protected]>Cc Al Viro <[email protected]>--- drivers/block/loop.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/block/loop.c +1
-1@@ -266,7 +266,7 @@ static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)struct iov_iter i;ssize_t bw; -iov_iter_bvec (&i,ITER_BVEC , bvec,1 , bvec->bv_len); +iov_iter_bvec (&i,ITER_BVEC |WRITE , bvec,1 , bvec->bv_len);file_start_write (file); bw =vfs_iter_write (file, &i, ppos,0 );Signature · 1 line
2.14.3
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Jens Axboe
maintainerQuoted 7 lines · Ross Zwisler
This has gotten Reviewed-by tags from Christoph and Ming Lei.
Al, are you the right person to merge this? Or is the correct person Jens, whom I accidentally didn't include when I sent this out?
Just wanted to make sure this got merged, and to see whether it was targeting v4.16 or v4.17.
I'm the right guy, but I don't see patches if I'm not cc'ed on them... I have queued this up for 4.16, thanks.
Signature · 6 lines
Jens Axboe _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Jens Axboe
maintainerQuoted 11 lines · Jens Axboe
Quoted 7 lines · Ross Zwisler
This has gotten Reviewed-by tags from Christoph and Ming Lei.
Al, are you the right person to merge this? Or is the correct person Jens, whom I accidentally didn't include when I sent this out?
Just wanted to make sure this got merged, and to see whether it was targeting v4.16 or v4.17.
I'm the right guy, but I don't see patches if I'm not cc'ed on them... I have queued this up for 4.16, thanks.
I do see patches sent to linux-block as well, but you didn't CC that one either.
Signature · 6 lines
Jens Axboe _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
Quoted 7 lines · Jens Axboe
Quoted 6 lines · Jens Axboe
Quoted 5 lines · Ross Zwisler
This has gotten Reviewed-by tags from Christoph and Ming Lei.
Al, are you the right person to merge this? Or is the correct person Jens, whom I accidentally didn't include when I sent this out?
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
Just wanted to make sure this got merged, and to see whether it was targeting v4.16 or v4.17.
I'm the right guy, but I don't see patches if I'm not cc'ed on them... I have queued this up for 4.16, thanks.
I do see patches sent to linux-block as well, but you didn't CC that one either.
I figure out who to CC on my patches by using scripts/get_maintainer.pl, and it didn't know about you or linux-block because drivers/block wasn't covered in MAINTAINERS. I'll send a patch to fix this.
As it was I just CC'd the folks involved in the original patch, and that one went up through Al.
Thanks for picking this up.
_______________________________________________
Linux-nvdimm mailing list
[email protected]
https://lists.01.org/mailman/listinfo/linux-nvdimm
Jens Axboe
maintainerQuoted 14 lines · Ross Zwisler
Quoted 13 lines · Jens Axboe
Quoted 11 lines · Jens Axboe
Quoted 7 lines · Ross Zwisler
This has gotten Reviewed-by tags from Christoph and Ming Lei.
Al, are you the right person to merge this? Or is the correct person Jens, whom I accidentally didn't include when I sent this out?
Just wanted to make sure this got merged, and to see whether it was targeting v4.16 or v4.17.
I'm the right guy, but I don't see patches if I'm not cc'ed on them... I have queued this up for 4.16, thanks.
Quoted 6 lines
Quoted 2 lines
I do see patches sent to linux-block aswell, but you didn't CC that one either.
I figure out who to CC on my patches by using
scripts/get_maintainer.pl, and it didn't know about you or linux-block becausedrivers/blockwasn't covered inMAINTAINERS. I'll send a patch to fix this.
I'm the first person for a check on drivers/block/ or drivers/block/loop.c, though...
Quoted 4 lines
As it was I just CC'd the folks involved in the original patch, and that one went up through Al.
Thanks for picking this up.
No problem, glad it worked out.
Signature · 6 lines
Jens Axboe _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm
PATCH 1/1 MAINTAINERS: add coverage for drivers/block
In reply to Jens AxboeTo help folks like me that use scripts/get_maintainer.pl.
--- MAINTAINERS | 1 + 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
@@ -2685,6 +2685,7 @@ L: [email protected] T: git git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git S: Maintained F: block/ +F: drivers/block/ F: kernel/trace/blktrace.c F: lib/sbitmap.c
Signature · 6 lines
2.14.3 _______________________________________________ Linux-nvdimm mailing list [email protected] https://lists.01.org/mailman/listinfo/linux-nvdimm