[PATCH] f2fs: Fix deadlock under storage almost full/dirty condition

[Sahitya Tummala]

There could be a potential deadlock when the storage capacity
is almost full and theren't enough free segments available, due
to which FG_GC is needed in the atomic commit ioctl as shown in
the below callstack -

schedule_timeout
io_schedule_timeout
congestion_wait
f2fs_drop_inmem_pages_all
f2fs_gc
f2fs_balance_fs
__write_node_page
f2fs_fsync_node_pages
f2fs_do_sync_file
f2fs_ioctl

If this inode doesn't have i_gc_failures[GC_FAILURE_ATOMIC] set,
then it waits forever in f2fs_drop_inmem_pages_all(), for this
atomic inode to be dropped. And the rest of the system is stuck
waiting for sbi->gc_mutex lock, which is acquired by f2fs_balance_fs()
in the stack above.

Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
---
 fs/f2fs/segment.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index da830fc..335ec09 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -300,7 +300,8 @@ void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
 
 	if (inode) {
 		if (gc_failure) {
-			if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
+			if (fi->i_gc_failures[GC_FAILURE_ATOMIC] ||
+				F2FS_I(inode)->inmem_task == current)
 				goto drop;
 			goto skip;
 		}
-- 
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.