Skip to contentSkip to conversation

fs: do not update freeing inode i_io_list

PATCH In review Svyatoslav Feldsherov 3 participants 3 messages Nov 15, 2022 – Nov 28, 2022 VFS

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 fs: do not update freeing inode i_io_list

After commit cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE") writeback_single_inode can push inode with I_DIRTY_TIME set to b_dirty_time list. In case of freeing inode with I_DIRTY_TIME set this can happen after deletion of inode from i_io_list at evict. Stack trace is following.

evict
fat_evict_inode
fat_truncate_blocks
fat_flush_inodes
writeback_inode
sync_inode_metadata(inode, sync=0) writeback_single_inode(inode, wbc) <- wbc->sync_mode == WB_SYNC_NONE

This will lead to use after free in flusher thread.

Similar issue can be triggered if writeback_single_inode in the stack trace update inode->i_io_list. Add explicit check to avoid it.

Fixes cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE")
Reported-by [email protected]

---
V2 -> V3:
- fix grammar in commit message and comments

V1 -> V2:
- address review comments
- skip inode_cgwb_move_to_attached for freeing inode

 fs/fs-writeback.c | 30 +++++++++++++++++++-----------
 1 file changed, 19 insertions(+), 11 deletions(-)
fs/fs-writeback.c +19 -11
index 443f83382b9b..9958d4020771 100644
@@ -1712,18 +1712,26 @@ static int writeback_single_inode(struct inode *inode,
 	wb = inode_to_wb_and_lock_list(inode);
 	spin_lock(&inode->i_lock);
 	/*
-	 * If the inode is now fully clean, then it can be safely removed from
-	 * its writeback list (if any).  Otherwise the flusher threads are
-	 * responsible for the writeback lists.
+	 * If the inode is freeing, its i_io_list shoudn't be updated
+	 * as it can be finally deleted at this moment.
 	 */
-	if (!(inode->i_state & I_DIRTY_ALL))
-		inode_cgwb_move_to_attached(inode, wb);
-	else if (!(inode->i_state & I_SYNC_QUEUED)) {
-		if ((inode->i_state & I_DIRTY))
-			redirty_tail_locked(inode, wb);
-		else if (inode->i_state & I_DIRTY_TIME) {
-			inode->dirtied_when = jiffies;
-			inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
+	if (!(inode->i_state & I_FREEING)) {
+		/*
+		 * If the inode is now fully clean, then it can be safely
+		 * removed from its writeback list (if any). Otherwise the
+		 * flusher threads are responsible for the writeback lists.
+		 */
+		if (!(inode->i_state & I_DIRTY_ALL))
+			inode_cgwb_move_to_attached(inode, wb);
+		else if (!(inode->i_state & I_SYNC_QUEUED)) {
+			if ((inode->i_state & I_DIRTY))
+				redirty_tail_locked(inode, wb);
+			else if (inode->i_state & I_DIRTY_TIME) {
+				inode->dirtied_when = jiffies;
+				inode_io_list_move_locked(inode,
+							  wb,
+							  &wb->b_dirty_time);
+			}
 		}
 	}
 
Signature · 1 line
2.38.1.431.g37b22c650d-goog
reviewer
#1 lore raw
Quoted 16 lines · Svyatoslav Feldsherov

After commit cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE") writeback_single_inode can push inode with I_DIRTY_TIME set to b_dirty_time list. In case of freeing inode with I_DIRTY_TIME set this can happen after deletion of inode from i_io_list at evict. Stack trace is following.

evict
fat_evict_inode
fat_truncate_blocks
fat_flush_inodes
writeback_inode
sync_inode_metadata(inode, sync=0) writeback_single_inode(inode, wbc) <- wbc->sync_mode == WB_SYNC_NONE

This will lead to use after free in flusher thread.

Quoted 7 lines

Similar issue can be triggered if writeback_single_inode in the stack trace update inode->i_io_list. Add explicit check to avoid it.

Fixes cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE")
Reported-by [email protected]

Ted, I guess you will merge this patch since you've merged the one from Lukas this patch is fixing?

Honza

Quoted 56 lines

---
V2 -> V3:
- fix grammar in commit message and comments

V1 -> V2:
- address review comments
- skip inode_cgwb_move_to_attached for freeing inode

 fs/fs-writeback.c | 30 +++++++++++++++++++-----------
 1 file changed, 19 insertions(+), 11 deletions(-)
fs/fs-writeback.c +19 -11
index 443f83382b9b..9958d4020771 100644
@@ -1712,18 +1712,26 @@ static int writeback_single_inode(struct inode *inode,
 	wb = inode_to_wb_and_lock_list(inode);
 	spin_lock(&inode->i_lock);
 	/*
-	 * If the inode is now fully clean, then it can be safely removed from
-	 * its writeback list (if any).  Otherwise the flusher threads are
-	 * responsible for the writeback lists.
+	 * If the inode is freeing, its i_io_list shoudn't be updated
+	 * as it can be finally deleted at this moment.
 	 */
-	if (!(inode->i_state & I_DIRTY_ALL))
-		inode_cgwb_move_to_attached(inode, wb);
-	else if (!(inode->i_state & I_SYNC_QUEUED)) {
-		if ((inode->i_state & I_DIRTY))
-			redirty_tail_locked(inode, wb);
-		else if (inode->i_state & I_DIRTY_TIME) {
-			inode->dirtied_when = jiffies;
-			inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
+	if (!(inode->i_state & I_FREEING)) {
+		/*
+		 * If the inode is now fully clean, then it can be safely
+		 * removed from its writeback list (if any). Otherwise the
+		 * flusher threads are responsible for the writeback lists.
+		 */
+		if (!(inode->i_state & I_DIRTY_ALL))
+			inode_cgwb_move_to_attached(inode, wb);
+		else if (!(inode->i_state & I_SYNC_QUEUED)) {
+			if ((inode->i_state & I_DIRTY))
+				redirty_tail_locked(inode, wb);
+			else if (inode->i_state & I_DIRTY_TIME) {
+				inode->dirtied_when = jiffies;
+				inode_io_list_move_locked(inode,
+							  wb,
+							  &wb->b_dirty_time);
+			}
 		}
 	}
 
Signature · 1 line
2.38.1.431.g37b22c650d-goog
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#2 lore raw
Quoted 20 lines · Jan Kara
Quoted 19 lines · Svyatoslav Feldsherov

After commit cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE") writeback_single_inode can push inode with I_DIRTY_TIME set to b_dirty_time list. In case of freeing inode with I_DIRTY_TIME set this can happen after deletion of inode from i_io_list at evict. Stack trace is following.

evict
fat_evict_inode
fat_truncate_blocks
fat_flush_inodes
writeback_inode
sync_inode_metadata(inode, sync=0) writeback_single_inode(inode, wbc) <- wbc->sync_mode == WB_SYNC_NONE

This will lead to use after free in flusher thread.

Similar issue can be triggered if writeback_single_inode in the stack trace update inode->i_io_list. Add explicit check to avoid it.

Quoted 7 lines
Quoted 4 lines
Fixes cbfecb927f42 ("fs: record I_DIRTY_TIME even if inode already has I_DIRTY_INODE")
Reported-by [email protected]

Ted, I guess you will merge this patch since you've merged the one from Lukas this patch is fixing?

Sorry, I forgot to ack this earlier, but this was pushed to Linus and it's in 6.1-rc7.

- Ted