On Tue, 2022-04-12 at 21:46 +0800, chenxiaosong (A) wrote:
> 在 2022/3/6 23:08, Trond Myklebust 写道:
> >
> > Just to clarify a little.
> >
> > I don't see a need to consume the writeback errors on close(),
> > unless
> > other filesystems do the same. If the intention is that fsync()
> > should
> > see _all_ errors that haven't already been seen, then NFS should
> > follow
> > the same semantics as all the other filesystems.
> >
>
> Other filesystem will _not_ clear writeback error on close().
> And other filesystem will _not_ clear writeback error on async
> write() too.
>
> Other filesystem _only_ clear writeback error on fsync() or sync
> write().
>
Yes. We might even consider not reporting writeback errors at all in
close(), since most developers don't check it. We certainly don't want
to clear those errors there because the manpages don't document that as
being the case.
> Should NFS follow the same semantics as all the other filesystems?
It needs to follow the semantics described in the manpage for write(2)
and fsync(2) as closely as possible, yes. That documentation is
supposed to be normative for application developers.
We won't guarantee to immediately report ENOSPC like other filesystems
do (because that would require us to only support synchronous writes),
however that behaviour is already documented in the manpage.
We may also report some errors that are not documented in the manpage
(e.g. EACCES or EROFS) simply because those errors cannot always be
reported at open() time, as would be the case for a local filesystem.
That's just how the NFS protocol works (particularly for the case of
the stateless NFSv3 protocol).
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
在 2022/4/12 21:56, Trond Myklebust 写道:
> On Tue, 2022-04-12 at 21:46 +0800, chenxiaosong (A) wrote:
>>
>> Other filesystem will _not_ clear writeback error on close().
>> And other filesystem will _not_ clear writeback error on async
>> write() too.
>>
>> Other filesystem _only_ clear writeback error on fsync() or sync
>> write().
>>
>
> Yes. We might even consider not reporting writeback errors at all in
> close(), since most developers don't check it. We certainly don't want
> to clear those errors there because the manpages don't document that as
> being the case.
>
>> Should NFS follow the same semantics as all the other filesystems?
>
> It needs to follow the semantics described in the manpage for write(2)
> and fsync(2) as closely as possible, yes. That documentation is
> supposed to be normative for application developers.
>
> We won't guarantee to immediately report ENOSPC like other filesystems
> do (because that would require us to only support synchronous writes),
> however that behaviour is already documented in the manpage.
>
> We may also report some errors that are not documented in the manpage
> (e.g. EACCES or EROFS) simply because those errors cannot always be
> reported at open() time, as would be the case for a local filesystem.
> That's just how the NFS protocol works (particularly for the case of
> the stateless NFSv3 protocol).
>
After merging your patchset, NFS will clear wb error on async write(),
is this reasonable?
And more importantly, we can not detect new error by using
filemap_sample_wb_err()/filemap_sample_wb_err() while nfs_wb_all(),just
as I described:
```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
```