Eric has noticed that after pagecache read rework, generic/418 is
occasionally failing for ext4 when blocksize < pagesize. In fact, the
pagecache rework just made hard to hit race in ext4 more likely. The
problem is that since ext4 conversion of direct IO writes to iomap
framework (commit 378f32bab371), we update inode size after direct IO
write only after invalidating page cache. Thus if buffered read sneaks
at unfortunate moment like:
CPU1 - write at offset 1k CPU2 - read from offset 0
iomap_dio_rw(..., IOMAP_DIO_FORCE_WAIT);
ext4_readpage();
ext4_handle_inode_extension()
the read will zero out tail of the page as it still sees smaller inode
size and thus page cache becomes inconsistent with on-disk contents with
all the consequences.
Fix the problem by moving inode size update into end_io handler which
gets called before the page cache is invalidated.
Reported-by: Eric Whitney <[email protected]>
Fixes: 378f32bab371 ("ext4: introduce direct I/O write using iomap infrastructure")
CC: [email protected]
Signed-off-by: Jan Kara <[email protected]>
---
fs/ext4/file.c | 20 ++++++++++++++++----
1 file changed, 16 insertions(+), 4 deletions(-)
Eric, can you please try whether this patch fixes the failures you are
occasionally seeing?
Changes since v1:
* Rewritten the fix to avoid the need for separate transaction handle for
orphan list update
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 194f5d00fa32..be1e80af61be 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -371,15 +371,27 @@ static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
int error, unsigned int flags)
{
- loff_t offset = iocb->ki_pos;
+ loff_t pos = iocb->ki_pos;
struct inode *inode = file_inode(iocb->ki_filp);
if (error)
return error;
- if (size && flags & IOMAP_DIO_UNWRITTEN)
- return ext4_convert_unwritten_extents(NULL, inode,
- offset, size);
+ if (size && flags & IOMAP_DIO_UNWRITTEN) {
+ error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
+ if (error < 0)
+ return error;
+ }
+ /*
+ * If we are extending the file, we have to update i_size here before
+ * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
+ * buffered reads could zero out too much from page cache pages. Update
+ * of on-disk size will happen later in ext4_dio_write_iter() where
+ * we have enough information to also perform orphan list handling etc.
+ */
+ pos += size;
+ if (pos > i_size_read(inode))
+ i_size_write(inode, pos);
return 0;
}
--
2.26.2
* Jan Kara <[email protected]>:
> Eric has noticed that after pagecache read rework, generic/418 is
> occasionally failing for ext4 when blocksize < pagesize. In fact, the
> pagecache rework just made hard to hit race in ext4 more likely. The
> problem is that since ext4 conversion of direct IO writes to iomap
> framework (commit 378f32bab371), we update inode size after direct IO
> write only after invalidating page cache. Thus if buffered read sneaks
> at unfortunate moment like:
>
> CPU1 - write at offset 1k CPU2 - read from offset 0
> iomap_dio_rw(..., IOMAP_DIO_FORCE_WAIT);
> ext4_readpage();
> ext4_handle_inode_extension()
>
> the read will zero out tail of the page as it still sees smaller inode
> size and thus page cache becomes inconsistent with on-disk contents with
> all the consequences.
>
> Fix the problem by moving inode size update into end_io handler which
> gets called before the page cache is invalidated.
>
> Reported-by: Eric Whitney <[email protected]>
> Fixes: 378f32bab371 ("ext4: introduce direct I/O write using iomap infrastructure")
> CC: [email protected]
> Signed-off-by: Jan Kara <[email protected]>
> ---
> fs/ext4/file.c | 20 ++++++++++++++++----
> 1 file changed, 16 insertions(+), 4 deletions(-)
>
> Eric, can you please try whether this patch fixes the failures you are
> occasionally seeing?
Sure - the tests are running. They'll need to run overnight to get us a big
enough sample to verify the fix with a high degree of confidence. I'll report
when done.
Thanks for the patch!
Eric
>
> Changes since v1:
> * Rewritten the fix to avoid the need for separate transaction handle for
> orphan list update
>
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index 194f5d00fa32..be1e80af61be 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -371,15 +371,27 @@ static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
> static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
> int error, unsigned int flags)
> {
> - loff_t offset = iocb->ki_pos;
> + loff_t pos = iocb->ki_pos;
> struct inode *inode = file_inode(iocb->ki_filp);
>
> if (error)
> return error;
>
> - if (size && flags & IOMAP_DIO_UNWRITTEN)
> - return ext4_convert_unwritten_extents(NULL, inode,
> - offset, size);
> + if (size && flags & IOMAP_DIO_UNWRITTEN) {
> + error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
> + if (error < 0)
> + return error;
> + }
> + /*
> + * If we are extending the file, we have to update i_size here before
> + * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
> + * buffered reads could zero out too much from page cache pages. Update
> + * of on-disk size will happen later in ext4_dio_write_iter() where
> + * we have enough information to also perform orphan list handling etc.
> + */
> + pos += size;
> + if (pos > i_size_read(inode))
> + i_size_write(inode, pos);
>
> return 0;
> }
> --
> 2.26.2
>
On Wed, Apr 14, 2021 at 03:14:53PM +0200, Jan Kara wrote:
> Eric has noticed that after pagecache read rework, generic/418 is
> occasionally failing for ext4 when blocksize < pagesize. In fact, the
> pagecache rework just made hard to hit race in ext4 more likely. The
> problem is that since ext4 conversion of direct IO writes to iomap
> framework (commit 378f32bab371), we update inode size after direct IO
> write only after invalidating page cache. Thus if buffered read sneaks
> at unfortunate moment like:
>
> CPU1 - write at offset 1k CPU2 - read from offset 0
> iomap_dio_rw(..., IOMAP_DIO_FORCE_WAIT);
> ext4_readpage();
> ext4_handle_inode_extension()
>
> the read will zero out tail of the page as it still sees smaller inode
> size and thus page cache becomes inconsistent with on-disk contents with
> all the consequences.
>
> Fix the problem by moving inode size update into end_io handler which
> gets called before the page cache is invalidated.
>
> Reported-by: Eric Whitney <[email protected]>
> Fixes: 378f32bab371 ("ext4: introduce direct I/O write using iomap infrastructure")
> CC: [email protected]
> Signed-off-by: Jan Kara <[email protected]>
> ---
> fs/ext4/file.c | 20 ++++++++++++++++----
> 1 file changed, 16 insertions(+), 4 deletions(-)
>
> Eric, can you please try whether this patch fixes the failures you are
> occasionally seeing?
>
> Changes since v1:
> * Rewritten the fix to avoid the need for separate transaction handle for
> orphan list update
>
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index 194f5d00fa32..be1e80af61be 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -371,15 +371,27 @@ static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
> static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
> int error, unsigned int flags)
> {
> - loff_t offset = iocb->ki_pos;
> + loff_t pos = iocb->ki_pos;
> struct inode *inode = file_inode(iocb->ki_filp);
>
> if (error)
> return error;
>
> - if (size && flags & IOMAP_DIO_UNWRITTEN)
> - return ext4_convert_unwritten_extents(NULL, inode,
> - offset, size);
> + if (size && flags & IOMAP_DIO_UNWRITTEN) {
> + error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
> + if (error < 0)
> + return error;
> + }
> + /*
> + * If we are extending the file, we have to update i_size here before
> + * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
> + * buffered reads could zero out too much from page cache pages. Update
> + * of on-disk size will happen later in ext4_dio_write_iter() where
> + * we have enough information to also perform orphan list handling etc.
> + */
> + pos += size;
> + if (pos > i_size_read(inode))
> + i_size_write(inode, pos);
Might be worth explaining why this doesn't require locking to
prevent racing completions from updating the inode size and
potentially losing an EOF update. I know why but it might not be so
obvious to others (DIO extending writes are serialised
at submission in ext4) but it's probably worth having a comment
similar to the one in xfs_dio_write_end_io() that explains why XFS
needs locking.
Cheers,
Dave.
--
Dave Chinner
[email protected]
On Thu 15-04-21 12:47:24, Dave Chinner wrote:
> On Wed, Apr 14, 2021 at 03:14:53PM +0200, Jan Kara wrote:
> > Eric has noticed that after pagecache read rework, generic/418 is
> > occasionally failing for ext4 when blocksize < pagesize. In fact, the
> > pagecache rework just made hard to hit race in ext4 more likely. The
> > problem is that since ext4 conversion of direct IO writes to iomap
> > framework (commit 378f32bab371), we update inode size after direct IO
> > write only after invalidating page cache. Thus if buffered read sneaks
> > at unfortunate moment like:
> >
> > CPU1 - write at offset 1k CPU2 - read from offset 0
> > iomap_dio_rw(..., IOMAP_DIO_FORCE_WAIT);
> > ext4_readpage();
> > ext4_handle_inode_extension()
> >
> > the read will zero out tail of the page as it still sees smaller inode
> > size and thus page cache becomes inconsistent with on-disk contents with
> > all the consequences.
> >
> > Fix the problem by moving inode size update into end_io handler which
> > gets called before the page cache is invalidated.
> >
> > Reported-by: Eric Whitney <[email protected]>
> > Fixes: 378f32bab371 ("ext4: introduce direct I/O write using iomap infrastructure")
> > CC: [email protected]
> > Signed-off-by: Jan Kara <[email protected]>
> > ---
> > fs/ext4/file.c | 20 ++++++++++++++++----
> > 1 file changed, 16 insertions(+), 4 deletions(-)
> >
> > Eric, can you please try whether this patch fixes the failures you are
> > occasionally seeing?
> >
> > Changes since v1:
> > * Rewritten the fix to avoid the need for separate transaction handle for
> > orphan list update
> >
> > diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> > index 194f5d00fa32..be1e80af61be 100644
> > --- a/fs/ext4/file.c
> > +++ b/fs/ext4/file.c
> > @@ -371,15 +371,27 @@ static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
> > static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
> > int error, unsigned int flags)
> > {
> > - loff_t offset = iocb->ki_pos;
> > + loff_t pos = iocb->ki_pos;
> > struct inode *inode = file_inode(iocb->ki_filp);
> >
> > if (error)
> > return error;
> >
> > - if (size && flags & IOMAP_DIO_UNWRITTEN)
> > - return ext4_convert_unwritten_extents(NULL, inode,
> > - offset, size);
> > + if (size && flags & IOMAP_DIO_UNWRITTEN) {
> > + error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
> > + if (error < 0)
> > + return error;
> > + }
> > + /*
> > + * If we are extending the file, we have to update i_size here before
> > + * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
> > + * buffered reads could zero out too much from page cache pages. Update
> > + * of on-disk size will happen later in ext4_dio_write_iter() where
> > + * we have enough information to also perform orphan list handling etc.
> > + */
> > + pos += size;
> > + if (pos > i_size_read(inode))
> > + i_size_write(inode, pos);
>
> Might be worth explaining why this doesn't require locking to
> prevent racing completions from updating the inode size and
> potentially losing an EOF update. I know why but it might not be so
> obvious to others (DIO extending writes are serialised
> at submission in ext4) but it's probably worth having a comment
> similar to the one in xfs_dio_write_end_io() that explains why XFS
> needs locking.
Good idea. There's also the somewhat subtle reason why this completion for
non-extending write cannot race with truncate and falsely seeing pos >
i_size. I've added comments explaining that.
Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR
* Jan Kara <[email protected]>:
> Eric has noticed that after pagecache read rework, generic/418 is
> occasionally failing for ext4 when blocksize < pagesize. In fact, the
> pagecache rework just made hard to hit race in ext4 more likely. The
> problem is that since ext4 conversion of direct IO writes to iomap
> framework (commit 378f32bab371), we update inode size after direct IO
> write only after invalidating page cache. Thus if buffered read sneaks
> at unfortunate moment like:
>
> CPU1 - write at offset 1k CPU2 - read from offset 0
> iomap_dio_rw(..., IOMAP_DIO_FORCE_WAIT);
> ext4_readpage();
> ext4_handle_inode_extension()
>
> the read will zero out tail of the page as it still sees smaller inode
> size and thus page cache becomes inconsistent with on-disk contents with
> all the consequences.
>
> Fix the problem by moving inode size update into end_io handler which
> gets called before the page cache is invalidated.
>
> Reported-by: Eric Whitney <[email protected]>
> Fixes: 378f32bab371 ("ext4: introduce direct I/O write using iomap infrastructure")
> CC: [email protected]
> Signed-off-by: Jan Kara <[email protected]>
> ---
> fs/ext4/file.c | 20 ++++++++++++++++----
> 1 file changed, 16 insertions(+), 4 deletions(-)
>
> Eric, can you please try whether this patch fixes the failures you are
> occasionally seeing?
>
I applied this patch to 5.12-rc7 and ran tests on two different sets of
hardware using kvm-xfstests.
5000 runs of generic/418 passed on the 1k test configuration without
incident. On an unmodified -rc7 kernel, I'm typically getting 100-200 failures
for that many runs. So, that looks good.
After running through all the appliance's test configurations with the
patched kernel, I did see one unexpected failure of generic/068 on
data=journal due to a burst of kernel warnings. Unexpected means I haven't
seen this failure in recent memory on upstream -rc* regression runs. A
10 run attempt to reproduce the failure failed to do so. An instance of the
warning follows.
Thanks,
Eric
[35456.178585] WARNING: CPU: 1 PID: 22225 at fs/ext4/ext4_jbd2.c:75 ext4_journal_check_start+0x4c/0x90
[35456.180802] Modules linked in:
[35456.181554] CPU: 1 PID: 22225 Comm: kworker/u4:0 Not tainted 5.12.0-rc7+ #2
[35456.183169] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
[35456.185030] Workqueue: writeback wb_workfn (flush-254:32)
[35456.186290] RIP: 0010:ext4_journal_check_start+0x4c/0x90
[35456.187540] Code: e1 02 75 24 f6 43 50 01 75 54 83 bb 20 03 00 00 04 74 17 48 8b 92 70 03 00 00 31 c0 48 85 d2 74 07 8b 02 83 e0 02 75 06 5b c3 <0f> 0b eb e5 44 8b 42 08 68 11 23 0f 82 48 89 df 45 31 c9 b9 01 00
[35456.190939] RSP: 0018:ffffc90004167850 EFLAGS: 00010246
[35456.191845] RAX: 00000000fffffffb RBX: ffff888009f76000 RCX: 0000000000000000
[35456.193668] RDX: ffff888009f73000 RSI: ffffffff8235dd60 RDI: ffff888009f76000
[35456.195461] RBP: 0000000000000009 R08: 0000000000000000 R09: ffffffff813c9fd8
[35456.197365] R10: 0000000000000000 R11: 000000000002f6b8 R12: 0000000000000000
[35456.199276] R13: 0000000000000008 R14: 0000000000000002 R15: 0000000000000771
[35456.201080] FS: 0000000000000000(0000) GS:ffff88807dd00000(0000) knlGS:0000000000000000
[35456.202513] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[35456.203521] CR2: 00007f30cf413000 CR3: 0000000001c04006 CR4: 0000000000370ee0
[35456.205554] Call Trace:
[35456.206275] __ext4_journal_start_sb+0x118/0x180
[35456.207568] ext4_writepage+0x368/0x740
[35456.209077] ? _raw_spin_unlock_irqrestore+0x30/0x40
[35456.210638] ? page_mkclean+0x6e/0xc0
[35456.211899] ? lock_is_held_type+0x99/0x100
[35456.213341] __writepage+0x19/0x70
[35456.214614] write_cache_pages+0x1a6/0x470
[35456.216026] ? __wb_calc_thresh+0xd0/0xd0
[35456.217617] generic_writepages+0x59/0x90
[35456.219119] ? lock_is_held_type+0x99/0x100
[35456.220673] ? do_writepages+0x41/0xd0
[35456.222038] ? do_writepages+0x41/0xd0
[35456.223341] ext4_writepages+0xc40/0x1080
[35456.224750] ? __lock_acquire+0x3d5/0x2380
[35456.226175] ? __lock_acquire+0x3d5/0x2380
[35456.227586] ? do_writepages+0x41/0xd0
[35456.228563] ? __ext4_mark_inode_dirty+0x280/0x280
[35456.230308] do_writepages+0x41/0xd0
[35456.231540] ? lock_is_held_type+0x99/0x100
[35456.233032] __writeback_single_inode+0x58/0x530
[35456.234542] writeback_sb_inodes+0x204/0x4d0
[35456.235995] wb_writeback+0x107/0x430
[35456.237489] wb_workfn+0xb5/0x5f0
[35456.238645] ? lock_acquire+0xb5/0x380
[35456.240026] ? process_one_work+0x223/0x5d0
[35456.241522] ? lock_is_held_type+0x99/0x100
[35456.243139] process_one_work+0x2a5/0x5d0
[35456.244615] ? process_one_work+0x5d0/0x5d0
[35456.246108] worker_thread+0x2d/0x3c0
[35456.247429] ? process_one_work+0x5d0/0x5d0
[35456.248926] kthread+0x127/0x140
[35456.250090] ? kthread_park+0x80/0x80
[35456.251379] ret_from_fork+0x22/0x30
[35456.252559] irq event stamp: 4858465
[35456.253785] hardirqs last enabled at (4858473): [<ffffffff8115ddee>] console_unlock+0x36e/0x550
[35456.256235] hardirqs last disabled at (4858482): [<ffffffff8115de77>] console_unlock+0x3f7/0x550
[35456.259117] softirqs last enabled at (4858496): [<ffffffff81c00303>] __do_softirq+0x303/0x41a
[35456.261485] softirqs last disabled at (4858491): [<ffffffff810ecade>] irq_exit_rcu+0xae/0xb0
[35456.263742] ---[ end trace ea82423dcfaf8730 ]---
> Changes since v1:
> * Rewritten the fix to avoid the need for separate transaction handle for
> orphan list update
>
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index 194f5d00fa32..be1e80af61be 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -371,15 +371,27 @@ static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
> static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
> int error, unsigned int flags)
> {
> - loff_t offset = iocb->ki_pos;
> + loff_t pos = iocb->ki_pos;
> struct inode *inode = file_inode(iocb->ki_filp);
>
> if (error)
> return error;
>
> - if (size && flags & IOMAP_DIO_UNWRITTEN)
> - return ext4_convert_unwritten_extents(NULL, inode,
> - offset, size);
> + if (size && flags & IOMAP_DIO_UNWRITTEN) {
> + error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
> + if (error < 0)
> + return error;
> + }
> + /*
> + * If we are extending the file, we have to update i_size here before
> + * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
> + * buffered reads could zero out too much from page cache pages. Update
> + * of on-disk size will happen later in ext4_dio_write_iter() where
> + * we have enough information to also perform orphan list handling etc.
> + */
> + pos += size;
> + if (pos > i_size_read(inode))
> + i_size_write(inode, pos);
>
> return 0;
> }
> --
> 2.26.2
>