diff mbox series

[6/9] btrfs: stop submitting I/O after an error in extent_write_locked_range

Message ID 20230724132701.816771-7-hch@lst.de (mailing list archive)
State New, archived
Headers show
Series [1/9] btrfs: don't stop integrity writeback too early | expand

Commit Message

Christoph Hellwig July 24, 2023, 1:26 p.m. UTC 
As soon a __extent_writepage_io returns an error we know that the
extent_map is corrupted or we're out of memory, and there is no point
in continuing to submit I/O.  Follow the behavior in extent_write_cache
pages and stop submitting I/O, and instead just unlock all pages.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/extent_io.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)

Comments

Boris Burkov Aug. 3, 2023, 12:42 a.m. UTC | #1 
On Mon, Jul 24, 2023 at 06:26:58AM -0700, Christoph Hellwig wrote:
> As soon a __extent_writepage_io returns an error we know that the
> extent_map is corrupted or we're out of memory, and there is no point
> in continuing to submit I/O.  Follow the behavior in extent_write_cache
> pages and stop submitting I/O, and instead just unlock all pages.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Boris Burkov <boris@bur.io>
diff mbox series

Patch

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 85d7a219040d07..fada7a1931b130 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2151,7 +2151,6 @@  void extent_write_locked_range(struct inode *inode, struct page *locked_page,
 			       u64 start, u64 end, struct writeback_control *wbc,
 			       bool pages_dirty)
 {
-	bool found_error = false;
 	int ret = 0;
 	struct address_space *mapping = inode->i_mapping;
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
@@ -2180,16 +2179,29 @@  void extent_write_locked_range(struct inode *inode, struct page *locked_page,
 			clear_page_dirty_for_io(page);
 		}
 
-		ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
-					    i_size);
+		/*
+		 * The entire page range that we were called on is locked and
+		 * covered by an ordered_extent.  Make sure we continue the
+		 * loop after an initial writeback error to unwind these as well
+		 * as clearing the dirty bit on the page by starting and ending
+		 * writeback.
+		 */
+		if (ret) {
+			btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
+						       cur, cur_len, false);
+			btrfs_page_clear_uptodate(fs_info, page, cur, cur_len);
+			set_page_writeback(page);
+			end_page_writeback(page);
+		} else {
+			ret = __extent_writepage_io(BTRFS_I(inode), page,
+						    &bio_ctrl, i_size);
+		}
 		btrfs_page_unlock_writer(fs_info, page, cur, cur_len);
-		if (ret < 0)
-			found_error = true;
 		put_page(page);
 		cur = cur_end + 1;
 	}
 
-	submit_write_bio(&bio_ctrl, found_error ? ret : 0);
+	submit_write_bio(&bio_ctrl, ret);
 }
 
 int extent_writepages(struct address_space *mapping,