[PATCH v6 02/15] ubifs: Fix deadlock in concurrent rename whiteout and inode writeback

[Zhihao Cheng]

Following hung tasks:
[   77.028764] task:kworker/u8:4    state:D stack:    0 pid:  132
[   77.028820] Call Trace:
[   77.029027]  schedule+0x8c/0x1b0
[   77.029067]  mutex_lock+0x50/0x60
[   77.029074]  ubifs_write_inode+0x68/0x1f0 [ubifs]
[   77.029117]  __writeback_single_inode+0x43c/0x570
[   77.029128]  writeback_sb_inodes+0x259/0x740
[   77.029148]  wb_writeback+0x107/0x4d0
[   77.029163]  wb_workfn+0x162/0x7b0

[   92.390442] task:aa              state:D stack:    0 pid: 1506
[   92.390448] Call Trace:
[   92.390458]  schedule+0x8c/0x1b0
[   92.390461]  wb_wait_for_completion+0x82/0xd0
[   92.390469]  __writeback_inodes_sb_nr+0xb2/0x110
[   92.390472]  writeback_inodes_sb_nr+0x14/0x20
[   92.390476]  ubifs_budget_space+0x705/0xdd0 [ubifs]
[   92.390503]  do_rename.cold+0x7f/0x187 [ubifs]
[   92.390549]  ubifs_rename+0x8b/0x180 [ubifs]
[   92.390571]  vfs_rename+0xdb2/0x1170
[   92.390580]  do_renameat2+0x554/0x770

, are caused by concurrent rename whiteout and inode writeback processes:
	rename_whiteout(Thread 1)	        wb_workfn(Thread2)
ubifs_rename
  do_rename
    lock_4_inodes (Hold ui_mutex)
    ubifs_budget_space
      make_free_space
        shrink_liability
	  __writeback_inodes_sb_nr
	    bdi_split_work_to_wbs (Queue new wb work)
					      wb_do_writeback(wb work)
						__writeback_single_inode
					          ubifs_write_inode
					            LOCK(ui_mutex)
							   ↑
	      wb_wait_for_completion (Wait wb work) <-- deadlock!

Reproducer (Detail program in [Link]):
  1. SYS_renameat2("/mp/dir/file", "/mp/dir/whiteout", RENAME_WHITEOUT)
  2. Consume out of space before kernel(mdelay) doing budget for whiteout

Fix it by doing whiteout space budget before locking ubifs inodes.
BTW, it also fixes wrong goto tag 'out_release' in whiteout budget
error handling path(It should at least recover dir i_size and unlock
4 ubifs inodes).

Fixes: 9e0a1fff8db56ea ("ubifs: Implement RENAME_WHITEOUT")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=214733
Signed-off-by: Zhihao Cheng <chengzhihao1@xxxxxxxxxx>
---
 fs/ubifs/dir.c | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)

diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
index cfa8881d8cca..2735ad1affed 100644
--- a/fs/ubifs/dir.c
+++ b/fs/ubifs/dir.c
@@ -1324,6 +1324,7 @@ static int do_rename(struct inode *old_dir, struct dentry *old_dentry,
 
 	if (flags & RENAME_WHITEOUT) {
 		union ubifs_dev_desc *dev = NULL;
+		struct ubifs_budget_req wht_req;
 
 		dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
 		if (!dev) {
@@ -1345,6 +1346,20 @@ static int do_rename(struct inode *old_dir, struct dentry *old_dentry,
 		whiteout_ui->data = dev;
 		whiteout_ui->data_len = ubifs_encode_dev(dev, MKDEV(0, 0));
 		ubifs_assert(c, !whiteout_ui->dirty);
+
+		memset(&wht_req, 0, sizeof(struct ubifs_budget_req));
+		wht_req.dirtied_ino = 1;
+		wht_req.dirtied_ino_d = ALIGN(whiteout_ui->data_len, 8);
+		/*
+		 * To avoid deadlock between space budget (holds ui_mutex and
+		 * waits wb work) and writeback work(waits ui_mutex), do space
+		 * budget before ubifs inodes locked.
+		 */
+		err = ubifs_budget_space(c, &wht_req);
+		if (err) {
+			iput(whiteout);
+			goto out_release;
+		}
 	}
 
 	lock_4_inodes(old_dir, new_dir, new_inode, whiteout);
@@ -1419,16 +1434,6 @@ static int do_rename(struct inode *old_dir, struct dentry *old_dentry,
 	}
 
 	if (whiteout) {
-		struct ubifs_budget_req wht_req = { .dirtied_ino = 1,
-				.dirtied_ino_d = \
-				ALIGN(ubifs_inode(whiteout)->data_len, 8) };
-
-		err = ubifs_budget_space(c, &wht_req);
-		if (err) {
-			iput(whiteout);
-			goto out_release;
-		}
-
 		inc_nlink(whiteout);
 		mark_inode_dirty(whiteout);
 
-- 
2.31.1