Christian Schoenebeck wrote on Tue, Jun 14, 2022 at 02:10:01PM +0200:
> It definitely goes into the right direction, but I think it's going a bit too
> far by using writeback_fid also in cases where it is not necessary and wasn't
> used before in the past.
Would help if I had an idea of what was used where in the past.. :)
>From a quick look at the code, checking out v5.10,
v9fs_vfs_writepage_locked() used the writeback fid always for all writes
v9fs_vfs_readpages is a bit more complex but only seems to be using the
"direct" private_data fid for reads...
It took me a bit of time but I think the reads you were seeing on
writeback fid come from v9fs_write_begin that does some readpage on the
writeback fid to populate the page before a non-filling write happens.
> What about something like this in v9fs_init_request() (yet untested):
>
> /* writeback_fid is always opened O_RDWR (instead of just O_WRONLY)
> * explicitly for this case: partial write backs that require a read
> * prior to actual write and therefore requires a fid with read
> * capability.
> */
> if (rreq->origin == NETFS_READ_FOR_WRITE)
> fid = v9inode->writeback_fid;
... Which seems to be exactly what this origin is about, so if that
works I'm all for it.
> If desired, this could be further constrained later on like:
>
> if (rreq->origin == NETFS_READ_FOR_WRITE &&
> (fid->mode & O_ACCMODE) == O_WRONLY)
> {
> fid = v9inode->writeback_fid;
> }
That also makes sense, if the fid mode has read permissions we might as
well use these as the writeback fid would needlessly be doing root IOs.
> I will definitely give these options some test spins here, a short feedback
> ahead would be appreciated though.
Please let me know how that works out, I'd be happy to use either of
your versions instead of mine.
If I can be greedy though I'd like to post it together with the other
couple of fixes next week, so having something before the end of the
week would be great -- I think even my first overkill version early and
building on it would make sense at this point.
But I think you've got the right end, so hopefully won't be needing to
delay
Cheers,
--
Dominique
