[PATCH AUTOSEL 5.8 08/27] f2fs: should avoid inode eviction in synchronous path

From: Sasha Levin
Date: Wed Aug 19 2020 - 20:12:52 EST


From: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>

[ Upstream commit b0f3b87fb3abc42c81d76c6c5795f26dbdb2f04b ]

https://bugzilla.kernel.org/show_bug.cgi?id=208565

PID: 257 TASK: ecdd0000 CPU: 0 COMMAND: "init"
#0 [<c0b420ec>] (__schedule) from [<c0b423c8>]
#1 [<c0b423c8>] (schedule) from [<c0b459d4>]
#2 [<c0b459d4>] (rwsem_down_read_failed) from [<c0b44fa0>]
#3 [<c0b44fa0>] (down_read) from [<c044233c>]
#4 [<c044233c>] (f2fs_truncate_blocks) from [<c0442890>]
#5 [<c0442890>] (f2fs_truncate) from [<c044d408>]
#6 [<c044d408>] (f2fs_evict_inode) from [<c030be18>]
#7 [<c030be18>] (evict) from [<c030a558>]
#8 [<c030a558>] (iput) from [<c047c600>]
#9 [<c047c600>] (f2fs_sync_node_pages) from [<c0465414>]
#10 [<c0465414>] (f2fs_write_checkpoint) from [<c04575f4>]
#11 [<c04575f4>] (f2fs_sync_fs) from [<c0441918>]
#12 [<c0441918>] (f2fs_do_sync_file) from [<c0441098>]
#13 [<c0441098>] (f2fs_sync_file) from [<c0323fa0>]
#14 [<c0323fa0>] (vfs_fsync_range) from [<c0324294>]
#15 [<c0324294>] (do_fsync) from [<c0324014>]
#16 [<c0324014>] (sys_fsync) from [<c0108bc0>]

This can be caused by flush_dirty_inode() in f2fs_sync_node_pages() where
iput() requires f2fs_lock_op() again resulting in livelock.

Reported-by: Zhiguo Niu <Zhiguo.Niu@xxxxxxxxxx>
Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
fs/f2fs/node.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index 03e24df1c84f5..e61ce7fb0958b 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -1924,8 +1924,12 @@ int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
goto continue_unlock;
}

- /* flush inline_data, if it's async context. */
- if (do_balance && is_inline_node(page)) {
+ /* flush inline_data/inode, if it's async context. */
+ if (!do_balance)
+ goto write_node;
+
+ /* flush inline_data */
+ if (is_inline_node(page)) {
clear_inline_node(page);
unlock_page(page);
flush_inline_data(sbi, ino_of_node(page));
@@ -1938,7 +1942,7 @@ int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
if (flush_dirty_inode(page))
goto lock_node;
}
-
+write_node:
f2fs_wait_on_page_writeback(page, NODE, true, true);

if (!clear_page_dirty_for_io(page))
--
2.25.1