On Tue, 9 Nov 2010 21:16:44 +0300
Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> 2010/11/9 Jeff Layton <jlayton@xxxxxxxxxxxxxxx>:
> > On Fri, 5 Nov 2010 11:29:34 +0300
> > Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> >
> >> On strict cache mode if we don't have Exclusive oplock we write a data to
> >> the server through cifs_user_write. Then if we Level II oplock store it in
> >> the cache, otherwise - invalidate inode pages affected by this writing.
> >>
> >> Signed-off-by: Pavel Shilovsky <piastryyy@xxxxxxxxx>
> >> ---
> >> fs/cifs/cifsfs.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++-----
> >> fs/cifs/file.c | 14 ++++++++++++--
> >> 2 files changed, 58 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/fs/cifs/cifsfs.c b/fs/cifs/cifsfs.c
> >> index bb7f36e..fbcd219 100644
> >> --- a/fs/cifs/cifsfs.c
> >> +++ b/fs/cifs/cifsfs.c
> >> @@ -596,12 +596,53 @@ static ssize_t cifs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
> >> static ssize_t cifs_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
> >> unsigned long nr_segs, loff_t pos)
> >> {
> >> - struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
> >> - ssize_t written;
> >> + struct inode *inode;
> >> + struct cifs_sb_info *cifs_sb;
> >> + ssize_t written, cache_written;
> >> + loff_t saved_pos;
> >> +
> >> + inode = iocb->ki_filp->f_path.dentry->d_inode;
> >> +
> >> + if (CIFS_I(inode)->clientCanCacheAll)
> >> + return generic_file_aio_write(iocb, iov, nr_segs, pos);
> >> +
> >> + cifs_sb = CIFS_SB(iocb->ki_filp->f_path.dentry->d_sb);
> >> +
> >> + if ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) == 0) {
> >> + int rc;
> >> +
> >> + written = generic_file_aio_write(iocb, iov, nr_segs, pos);
> >> +
> >> + rc = filemap_fdatawrite(inode->i_mapping);
> >> + if (rc)
> >> + cFYI(1, "cifs_file_aio_write: %d rc on %p inode",
> >> + rc, inode);
> >> + return written;
> >> + }
> >> +
> >> + saved_pos = pos;
> >> + written = cifs_user_write(iocb->ki_filp, iov->iov_base,
> >> + iov->iov_len, &pos);
> >> +
> >> + if (written > 0 && CIFS_I(inode)->clientCanCacheRead) {
> >> + /* we have data written to the server and at least oplock
> >> + for reading - store the date in the cache */
> >> + cache_written = generic_file_aio_write(iocb, iov,
> >> + nr_segs, saved_pos);
> >> + if (cache_written != written)
> >> + cERROR(1, "Cache written and server written data "
> >> + "lengths are different");
> >> + return written;
> >> + }
> >> +
> >> + if (written > 0)
> >> + invalidate_mapping_pages(inode->i_mapping,
> >> + saved_pos >> PAGE_CACHE_SHIFT,
> >> + (saved_pos+iov->iov_len-1)
> >> + >> PAGE_CACHE_SHIFT);
> >> +
> >> + iocb->ki_pos = pos;
> >>
> >> - written = generic_file_aio_write(iocb, iov, nr_segs, pos);
> >> - if (!CIFS_I(inode)->clientCanCacheAll)
> >> - filemap_fdatawrite(inode->i_mapping);
> >> return written;
> >> }
> >>
> >> diff --git a/fs/cifs/file.c b/fs/cifs/file.c
> >> index b36de2e..85824c0 100644
> >> --- a/fs/cifs/file.c
> >> +++ b/fs/cifs/file.c
> >> @@ -1537,7 +1537,11 @@ static int cifs_write_end(struct file *file, struct address_space *mapping,
> >> struct page *page, void *fsdata)
> >> {
> >> int rc;
> >> - struct inode *inode = mapping->host;
> >> + struct inode *inode;
> >> + struct cifs_sb_info *cifs_sb;
> >> +
> >> + inode = mapping->host;
> >> + cifs_sb = CIFS_SB(inode->i_sb);
> >>
> >> cFYI(1, "write_end for page %p from pos %lld with %d bytes",
> >> page, pos, copied);
> >> @@ -1570,7 +1574,13 @@ static int cifs_write_end(struct file *file, struct address_space *mapping,
> >> } else {
> >> rc = copied;
> >> pos += copied;
> >> - set_page_dirty(page);
> >> + /* if we have strict cache switched on and don't have Exclusive
> >> + oplock for the inode, we don't have to set_page_dirty
> >> + because we flushed the data to the server in
> >> + cifs_file_aio_write before */
> >> + if ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) == 0 ||
> >> + CIFS_I(inode)->clientCanCacheAll)
> >> + set_page_dirty(page);
> >
> > Is there a potential for a race here? Suppose I have clientCanCacheAll
> > set, I write to the cache, but before I get to cifs_write_end, an
> > oplock break is processed and clientCanCacheAll is cleared. Now I have
> > dirty data in cache but the dirty flag doesn't get set. Does anything
> > prevent that?
>
> Yes, you are right. In the case of clientCanCacheAll and STRICT_IO set
> - can we call set_page_dirty for every page affected by our write in
> cifs_file_aio_write? It should prevent races, I think.
>
Hmm...setting the dirty bit before the data has actually been copied to
the page? I'm not sure that's a good idea. If you want to do that you
may want to run it by linux-fsdevel first.
Maybe you could do something instead with the fsdata pointer in
write_begin and write_end?
--
Jeff Layton <jlayton@xxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
