* [Cluster-devel] [PATCH] gfs2: Clean up gfs2_file_write_iter and fix O_SYNC write handling
@ 2020-02-06 15:37 Andreas Gruenbacher
2020-02-06 16:34 ` Christoph Hellwig
0 siblings, 1 reply; 3+ messages in thread
From: Andreas Gruenbacher @ 2020-02-06 15:37 UTC (permalink / raw)
To: cluster-devel.redhat.com
Clean up gfs2_file_write_iter. When falling back to a buffered write in the
O_DIRECT case, use generic_write_sync + IOCB_DSYNC to sync the buffered write
to disk instead of hand-rolling the sync. Currently, the error checking in the
buffered write fallback case is incomplete.
Based on a proposed fix by Christoph Hellwig <hch@lst.de>.
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
fs/gfs2/file.c | 51 +++++++++++++++++++++-----------------------------
1 file changed, 21 insertions(+), 30 deletions(-)
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 21d032c4b077..d68530b12706 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -847,7 +847,7 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
struct file *file = iocb->ki_filp;
struct inode *inode = file_inode(file);
struct gfs2_inode *ip = GFS2_I(inode);
- ssize_t written = 0, ret;
+ ssize_t ret;
ret = gfs2_rsqa_alloc(ip);
if (ret)
@@ -879,55 +879,46 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
if (iocb->ki_flags & IOCB_DIRECT) {
struct address_space *mapping = file->f_mapping;
- loff_t pos, endbyte;
- ssize_t buffered;
+ ssize_t buffered, ret2;
- written = gfs2_file_direct_write(iocb, from);
- if (written < 0 || !iov_iter_count(from))
+ ret = gfs2_file_direct_write(iocb, from);
+ if (ret < 0 || !iov_iter_count(from))
goto out_unlock;
current->backing_dev_info = inode_to_bdi(inode);
- ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+ buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
current->backing_dev_info = NULL;
- if (unlikely(ret < 0))
+ if (unlikely(buffered <= 0))
goto out_unlock;
- buffered = ret;
/*
* We need to ensure that the page cache pages are written to
* disk and invalidated to preserve the expected O_DIRECT
- * semantics.
+ * semantics. If the writeback or invalidate fails, only report
+ * the direct I/O range as we don't know if the buffered pages
+ * made it to disk.
*/
- pos = iocb->ki_pos;
- endbyte = pos + buffered - 1;
- ret = filemap_write_and_wait_range(mapping, pos, endbyte);
- if (!ret) {
- iocb->ki_pos += buffered;
- written += buffered;
- invalidate_mapping_pages(mapping,
- pos >> PAGE_SHIFT,
- endbyte >> PAGE_SHIFT);
- } else {
- /*
- * We don't know how much we wrote, so just return
- * the number of bytes which were direct-written
- */
- }
+ iocb->ki_pos += buffered;
+ iocb->ki_flags |= IOCB_DSYNC;
+ ret2 = generic_write_sync(iocb, buffered);
+ invalidate_mapping_pages(mapping,
+ (iocb->ki_pos - buffered) >> PAGE_SHIFT,
+ (iocb->ki_pos - 1) >> PAGE_SHIFT);
+ if (!ret || ret2 > 0)
+ ret += ret2;
} else {
current->backing_dev_info = inode_to_bdi(inode);
ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
current->backing_dev_info = NULL;
- if (likely(ret > 0))
+ if (likely(ret > 0)) {
iocb->ki_pos += ret;
+ ret = generic_write_sync(iocb, ret);
+ }
}
out_unlock:
inode_unlock(inode);
- if (likely(ret > 0)) {
- /* Handle various SYNC-type writes */
- ret = generic_write_sync(iocb, ret);
- }
- return written ? written : ret;
+ return ret;
}
static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
--
2.24.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
* [Cluster-devel] [PATCH] gfs2: Clean up gfs2_file_write_iter and fix O_SYNC write handling
2020-02-06 15:37 [Cluster-devel] [PATCH] gfs2: Clean up gfs2_file_write_iter and fix O_SYNC write handling Andreas Gruenbacher
@ 2020-02-06 16:34 ` Christoph Hellwig
2020-02-06 17:02 ` Andreas Gruenbacher
0 siblings, 1 reply; 3+ messages in thread
From: Christoph Hellwig @ 2020-02-06 16:34 UTC (permalink / raw)
To: cluster-devel.redhat.com
> if (iocb->ki_flags & IOCB_DIRECT) {
> struct address_space *mapping = file->f_mapping;
> + ssize_t buffered, ret2;
>
> + ret = gfs2_file_direct_write(iocb, from);
> + if (ret < 0 || !iov_iter_count(from))
> goto out_unlock;
>
> current->backing_dev_info = inode_to_bdi(inode);
> + buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
> current->backing_dev_info = NULL;
> - if (unlikely(ret < 0))
> + if (unlikely(buffered <= 0))
> goto out_unlock;
>
> /*
> * We need to ensure that the page cache pages are written to
> * disk and invalidated to preserve the expected O_DIRECT
> + * semantics. If the writeback or invalidate fails, only report
> + * the direct I/O range as we don't know if the buffered pages
> + * made it to disk.
> */
> + iocb->ki_pos += buffered;
> + iocb->ki_flags |= IOCB_DSYNC;
I think I'd rather add IOCB_DSYNC before calling
iomap_file_buffered_write, just in case we ever do optimizations for
synchronous I/O there.
> + ret2 = generic_write_sync(iocb, buffered);
> + invalidate_mapping_pages(mapping,
> + (iocb->ki_pos - buffered) >> PAGE_SHIFT,
This adds a line > 80 chars.
Otherwise this looks fine to me, although I'd just put the fix in the
subject line and just mention the cleanup at the end of the actual
commit log.
^ permalink raw reply [flat|nested] 3+ messages in thread
* [Cluster-devel] [PATCH] gfs2: Clean up gfs2_file_write_iter and fix O_SYNC write handling
2020-02-06 16:34 ` Christoph Hellwig
@ 2020-02-06 17:02 ` Andreas Gruenbacher
0 siblings, 0 replies; 3+ messages in thread
From: Andreas Gruenbacher @ 2020-02-06 17:02 UTC (permalink / raw)
To: cluster-devel.redhat.com
On Thu, Feb 6, 2020 at 5:34 PM Christoph Hellwig <hch@lst.de> wrote:
> > if (iocb->ki_flags & IOCB_DIRECT) {
> > struct address_space *mapping = file->f_mapping;
> > + ssize_t buffered, ret2;
> >
> > + ret = gfs2_file_direct_write(iocb, from);
> > + if (ret < 0 || !iov_iter_count(from))
> > goto out_unlock;
> >
> > current->backing_dev_info = inode_to_bdi(inode);
> > + buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
> > current->backing_dev_info = NULL;
> > - if (unlikely(ret < 0))
> > + if (unlikely(buffered <= 0))
> > goto out_unlock;
> >
> > /*
> > * We need to ensure that the page cache pages are written to
> > * disk and invalidated to preserve the expected O_DIRECT
> > + * semantics. If the writeback or invalidate fails, only report
> > + * the direct I/O range as we don't know if the buffered pages
> > + * made it to disk.
> > */
> > + iocb->ki_pos += buffered;
> > + iocb->ki_flags |= IOCB_DSYNC;
>
> I think I'd rather add IOCB_DSYNC before calling
> iomap_file_buffered_write, just in case we ever do optimizations for
> synchronous I/O there.
Okay, that shouldn't hurt.
> > + ret2 = generic_write_sync(iocb, buffered);
> > + invalidate_mapping_pages(mapping,
> > + (iocb->ki_pos - buffered) >> PAGE_SHIFT,
>
> This adds a line > 80 chars.
>
> Otherwise this looks fine to me, although I'd just put the fix in the
> subject line and just mention the cleanup at the end of the actual
> commit log.
Alright.
Thanks,
Andreas
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2020-02-06 17:02 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-02-06 15:37 [Cluster-devel] [PATCH] gfs2: Clean up gfs2_file_write_iter and fix O_SYNC write handling Andreas Gruenbacher
2020-02-06 16:34 ` Christoph Hellwig
2020-02-06 17:02 ` Andreas Gruenbacher
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.