On Tue, 2022-04-12 at 14:24 +0800, chenxiaosong (A) wrote:
> 在 2022/4/12 5:33, trondmy@xxxxxxxxxx 写道:
> > From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> >
> > Any errors reported by the write() system call need to be cleared
> > from
> > the file descriptor's error tracking. The current call to
> > nfs_wb_all()
> > causes the error to be reported, but since it doesn't call
> > file_check_and_advance_wb_err(), we can end up reporting the same
> > error
> > a second time when the application calls fsync().
> >
> > Note that since Linux 4.13, the rule is that EIO may be reported
> > for
> > write(), but it must be reported by a subsequent fsync(), so let's
> > just
> > drop reporting it in write.
> >
> > The check for nfs_ctx_key_to_expire() is just a duplicate to the
> > one
> > already in nfs_write_end(), so let's drop that too.
> >
> > Reported-by: ChenXiaoSong <chenxiaosong2@xxxxxxxxxx>
> > Fixes: ce368536dd61 ("nfs: nfs_file_write() should check for
> > writeback errors")
> > Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> > ---
> > fs/nfs/file.c | 33 +++++++++++++--------------------
> > 1 file changed, 13 insertions(+), 20 deletions(-)
>
>
> # 1. wb error mechanism of other filesystem
>
> Other filesystem only clear the wb error when calling fsync(), async
> write will not clear the wb error.
>
>
> # 2. still report unexpected error ... again
>
> After merging this patchset(5 patches), second `dd` of the following
> reproducer will still report unexpected error: No space left on
> device.
>
> Reproducer:
> nfs server | nfs client
> ------------------------------|--------------------------------------
> -------
> # No space left on server |
> fallocate -l 100G /svr/nospc |
> | mount -t nfs $nfs_server_ip:/ /mnt
> |
> | # Expected error: No space left on
> device
> | dd if=/dev/zero of=/mnt/file count=1
> ibs=10K
> |
> | # Release space on mountpoint
> | rm /mnt/nospc
> |
> | # Unexpected error: No space left on
> device
> | dd if=/dev/zero of=/mnt/file count=1
> ibs=10K
>
>
> # 3. my patchset
>
> https://patchwork.kernel.org/project/linux-nfs/list/?series=628066
>
> My patchset can fix bug of above reproducer.
>
> filemap_sample_wb_err() always return 0 if old writeback error
> have not been consumed. filemap_check_wb_err() will return the old
> error
> even if there is no new writeback error between
> filemap_sample_wb_err() and
> filemap_check_wb_err().
>
> ```c
> since = filemap_sample_wb_err() = 0
> errseq_sample
> if (!(old & ERRSEQ_SEEN)) // nobody see the error
> return 0;
> nfs_wb_all // no new error
> error = filemap_check_wb_err(..., since) != 0 // unexpected error
> ```
>
> So we still need record writeback error in address_space flags. The
> writeback
> error in address_space flags is not used to be reported to userspace,
> it
> is just
> used to detect if there is new error while writeback.
I understand all that. The point you appear to be missing is that this
is in fact in agreement with the documented behaviour in the write(2)
and fsync(2) manpages. These errors are supposed to be reported once,
even if they were caused by a write to a different file descriptor.
In fact, even with your change, if you make the second 'dd' call fsync
(by adding a conv=fsync), I would expect it to report the exact same
ENOSPC error, and that would be correct behaviour!
So my patches are more concerned with the fact that we appear to be
reporting the same error more than once, rather than the fact that
we're reporting them in the second attempt at I/O. As far as I'm
concerned, that is the main change that is needed to meet the behaviour
that is documented in the manpages.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx
