On Apr 6, 2015, at 4:09 PM, Al Viro wrote:
> On Mon, Apr 06, 2015 at 08:04:01PM +0000, Drokin, Oleg wrote:
>> Hello!
>>
>> On Apr 6, 2015, at 12:02 PM, Al Viro wrote:
>>> 2) should we ever update the current position when write() returns 0?
>>> IOW, what effect should zero-length write() on O_APPEND file have upon its
>>> current position? POSIX seems to imply that it should do nothing, and
>>> generally that's what happens, but e.g. ext4 *does* update position to
>>> the EOF, whether we will write anything or not. So does FUSE when server
>>> requests to bypass the page cache. AFAICS, lustre is the same way,
>>> but I might be missing something; everything else definitely does not
>>
>> Lustre is not the same way.
>> drivers/staging/lustre/lustre/llite/file.c::ll_file_io_generic() has this
>> if (io-> ci_nob > 0) {
>> result = io->ci_nob;
>> *ppos = io->u.ci_wr.wr.crw_pos;
>> }
>>
>> and ppos is passed in as &iocb->ki_pos. ci_nob is number of bytes that we managed to read/write,
>> and if it was 0 (either due to 0 bytes io request or due to error from the get go)
>> we won't update it.
>
> Eh? vvp_io_write_start():
> result = generic_file_write_iter(cio->cui_iocb, cio->cui_iter);
Oh, hm, I guess you mean
if (cl_io_is_append(io)) {
pos = io->u.ci_wr.wr.crw_pos = i_size_read(inode);
cio->cui_iocb->ki_pos = pos;
missed that one.
It's a leftover from the times we used to call __generic_file_aio_write there and that took
&iocb->ki_pos argument.
Now that we pass entire iocb in, probably can just drop that ki_pos assignment, I am going to try that now.
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
