f2fs_read_data_large_folio() can keep a read bio across multiple
readahead folios. If a later folio hits an error before any of its
blocks are added to the bio, folio_in_bio is false and the current error
path returns immediately after ending that folio.
This can leave the bio accumulated for earlier folios unsubmitted. Those
folios then never receive read completion, and readers can wait
indefinitely on the locked folios.
Route errors through the common out path so any pending bio is submitted
before returning. Stop consuming more readahead folios once an error is
seen, and only wait on and clear the current folio when it was actually
added to the bio.
Fixes: a5d8b9d94e18 ("f2fs: fix to unlock folio in f2fs_read_data_large_folio()")
Signed-off-by: Wenjie Qi <qiwenjie@xiaomi.com>
---
Changes in v2:
- Rebase on current jaegeuk/f2fs.git dev branch.
Tested with a local-only fault hook which injected -EIO when
f2fs_read_data_large_folio() had a pending read bio but the current
readahead folio had not been added to it. Without this fix, a 256MiB
immutable file read timed out after 20 seconds. With this fix, the same
read completed without hanging.
fs/f2fs/data.c | 11 +++++------
1 file changed, 5 insertions(+), 6 deletions(-)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index bdcf396038f8..9c6440a7db0e 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2529,7 +2529,7 @@ static int f2fs_read_data_large_folio(struct inode *inode,
unsigned nrpages;
struct f2fs_folio_state *ffs;
int ret = 0;
- bool folio_in_bio;
+ bool folio_in_bio = false;
if (f2fs_compressed_file(inode)) {
if (folio)
@@ -2645,18 +2645,17 @@ static int f2fs_read_data_large_folio(struct inode *inode,
}
trace_f2fs_read_folio(folio, DATA);
err_out:
- if (!folio_in_bio) {
+ if (!folio_in_bio)
folio_end_read(folio, !ret);
- if (ret)
- return ret;
- }
+ if (ret)
+ goto out;
if (rac) {
folio = readahead_folio(rac);
goto next_folio;
}
out:
f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
- if (ret) {
+ if (ret && folio_in_bio) {
/* Wait bios and clear uptodate. */
folio_lock(folio);
folio_clear_uptodate(folio);
base-commit: 520760b9f9156bf9698de38dc44c614fad68a1f9
--
2.43.0