diff mbox series

[1/3] iomap: dio Return zero in case of unsuccessful pagecache invalidation

Message ID 20200605204838.10765-2-rgoldwyn@suse.de (mailing list archive)
State New, archived
Headers show
Series Transient errors in Direct I/O | expand

Commit Message

Goldwyn Rodrigues June 5, 2020, 8:48 p.m. UTC 
From: Goldwyn Rodrigues <rgoldwyn@suse.com>

Filesystems such as btrfs are unable to guarantee page invalidation
because extents could be locked because of an ongoing I/O. This happens
even though a filemap_write_and_wait() has been called because
btrfs locks extents in a separate cleanup thread until all ordered
extents in range have performed the tree changes.

Return zero in case a page cache invalidation is unsuccessful so
filesystems can fallback to buffered I/O.

This takes care of the following invalidation warning during btrfs
mixed buffered and direct I/O using iomap_dio_rw():

Page cache invalidation failure on direct I/O.  Possible data
corruption due to collision with buffered I/O!

This is similar to the behavior of generic_file_direct_write().

Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
---
 fs/iomap/direct-io.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

Comments

Matthew Wilcox (Oracle) June 6, 2020, 3:13 a.m. UTC | #1 
On Fri, Jun 05, 2020 at 03:48:36PM -0500, Goldwyn Rodrigues wrote:
> Return zero in case a page cache invalidation is unsuccessful so
> filesystems can fallback to buffered I/O.

I don't think it's acceptable to change common code to fix a btrfs bug
at this point in the release cycle.  This change needs to be agreed to
before the -rc5 stage, not during the pre-rc1 window.

If you can't fix this in btrfs alone, then back out the btrfs changes
for now.
diff mbox series

Patch

diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index e4addfc58107..215315be6233 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -483,9 +483,15 @@  iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 	 */
 	ret = invalidate_inode_pages2_range(mapping,
 			pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
-	if (ret)
-		dio_warn_stale_pagecache(iocb->ki_filp);
-	ret = 0;
+	/*
+	 * If a page can not be invalidated, return 0 to fall back
+	 * to buffered write.
+	 */
+	if (ret) {
+		if (ret == -EBUSY)
+			ret = 0;
+		goto out_free_dio;
+	}
 
 	if (iov_iter_rw(iter) == WRITE && !wait_for_completion &&
 	    !inode->i_sb->s_dio_done_wq) {