在 2022/4/1 11:44, ChenXiaoSong 写道:
> v1:
> cover letter: (nfs: check writeback errors correctly)
>
> v2:
> - return more nuanced writeback errors in nfs_file_write().
> - return writeback error in close()->flush() without consumed it.
> - fix: nfs_file_write() will always call nfs_wb_all() even if there is no
> new writeback error.
>
>
> ChenXiaoSong (3):
> NFS: return more nuanced writeback errors in nfs_file_write()
> NFS: nfs{,4}_file_flush() return correct writeback errors
> Revert "nfs: nfs_file_write() should check for writeback errors"
>
> fs/nfs/file.c | 23 ++++++++++-------------
> fs/nfs/nfs4file.c | 8 ++++----
> fs/nfs/write.c | 5 +----
> 3 files changed, 15 insertions(+), 21 deletions(-)
>
It is not a good idea to modify error sequence mechanism, as the
`lib/errseq.c` described:
22 * Note that there is a risk of collisions if new errors are being
recorded
23 * frequently, since we have so few bits to use as a counter.
24 *
25 * To mitigate this, one bit is used as a flag to tell whether the
value has
26 * been sampled since a new value was recorded. That allows us to
avoid bumping
27 * the counter if no one has sampled it since the last time an error was
28 * recorded.
So, if we want to report nuanced writeback error, it is better to detect
wb error from filemap_check_errors(), and then return
-(file->f_mapping->wb_err & MAX_ERRNO) to userspace without consume it.
nfs_mapping_set_error
mapping_set_error
__filemap_set_wb_err // record error sequence
errseq_set
set_bit(..., &mapping->flags) // record address_space flag
// it is not used to be reported, just used to detect
error = filemap_check_errors // -ENOSPC or -EIO
test_and_clear_bit(..., &mapping->flags) // error bit cleared
// now we try to return nuanced writeback error
if (error)
return filemap_check_wb_err(file->f_mapping, 0);
return -(file->f_mapping->wb_err & MAX_ERRNO)
ping ...
在 2022/4/1 15:03, chenxiaosong (A) 写道:
> 在 2022/4/1 11:44, ChenXiaoSong 写道:
>> v1:
>> cover letter: (nfs: check writeback errors correctly)
>>
>> v2:
>> - return more nuanced writeback errors in nfs_file_write().
>> - return writeback error in close()->flush() without consumed it.
>> - fix: nfs_file_write() will always call nfs_wb_all() even if there is no
>> new writeback error.
>>
>>
>> ChenXiaoSong (3):
>> NFS: return more nuanced writeback errors in nfs_file_write()
>> NFS: nfs{,4}_file_flush() return correct writeback errors
>> Revert "nfs: nfs_file_write() should check for writeback errors"
>>
>> fs/nfs/file.c | 23 ++++++++++-------------
>> fs/nfs/nfs4file.c | 8 ++++----
>> fs/nfs/write.c | 5 +----
>> 3 files changed, 15 insertions(+), 21 deletions(-)
>>
>
> It is not a good idea to modify error sequence mechanism, as the
> `lib/errseq.c` described:
>
> 22 * Note that there is a risk of collisions if new errors are
> being recorded
> 23 * frequently, since we have so few bits to use as a counter.
> 24 *
> 25 * To mitigate this, one bit is used as a flag to tell whether
> the value has
> 26 * been sampled since a new value was recorded. That allows us
> to avoid bumping
> 27 * the counter if no one has sampled it since the last time an
> error was
> 28 * recorded.
>
>
> So, if we want to report nuanced writeback error, it is better to detect
> wb error from filemap_check_errors(), and then return
> -(file->f_mapping->wb_err & MAX_ERRNO) to userspace without consume it.
>
> nfs_mapping_set_error
> mapping_set_error
> __filemap_set_wb_err // record error sequence
> errseq_set
> set_bit(..., &mapping->flags) // record address_space flag
>
> // it is not used to be reported, just used to detect
> error = filemap_check_errors // -ENOSPC or -EIO
> test_and_clear_bit(..., &mapping->flags) // error bit cleared
>
> // now we try to return nuanced writeback error
> if (error)
> return filemap_check_wb_err(file->f_mapping, 0);
> return -(file->f_mapping->wb_err & MAX_ERRNO)
>