From: Jchao Sun <sunjunchao2870@xxxxxxxxx>
commit 10e14073107dd0b6d97d9516a02845a8e501c2c9 upstream.
Commit b35250c0816c ("writeback: Protect inode->i_io_list with
inode->i_lock") made inode->i_io_list not only protected by
wb->list_lock but also inode->i_lock, but inode_io_list_move_locked()
was missed. Add lock there and also update comment describing
things protected by inode->i_lock. This also fixes a race where
__mark_inode_dirty() could move inode under flush worker's hands
and thus sync(2) could miss writing some inodes.
Fixes: b35250c0816c ("writeback: Protect inode->i_io_list with inode->i_lock")
Link: https://lore.kernel.org/r/20220524150540.12552-1-sunjunchao2870@xxxxxxxxx
CC: stable@xxxxxxxxxxxxxxx
Signed-off-by: Jchao Sun <sunjunchao2870@xxxxxxxxx>
Signed-off-by: Jan Kara <jack@xxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
fs/fs-writeback.c | 37 ++++++++++++++++++++++++++++---------
fs/inode.c | 2 +-
2 files changed, 29 insertions(+), 10 deletions(-)
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -120,6 +120,7 @@ static bool inode_io_list_move_locked(st
struct list_head *head)
{
assert_spin_locked(&wb->list_lock);
+ assert_spin_locked(&inode->i_lock);
list_move(&inode->i_io_list, head);
@@ -1400,9 +1401,9 @@ static int move_expired_inodes(struct li
inode = wb_inode(delaying_queue->prev);
if (inode_dirtied_after(inode, dirtied_before))
break;
+ spin_lock(&inode->i_lock);
list_move(&inode->i_io_list, &tmp);
moved++;
- spin_lock(&inode->i_lock);
inode->i_state |= I_SYNC_QUEUED;
spin_unlock(&inode->i_lock);
if (sb_is_blkdev_sb(inode->i_sb))
@@ -1418,7 +1419,12 @@ static int move_expired_inodes(struct li
goto out;
}
- /* Move inodes from one superblock together */
+ /*
+ * Although inode's i_io_list is moved from 'tmp' to 'dispatch_queue',
+ * we don't take inode->i_lock here because it is just a pointless overhead.
+ * Inode is already marked as I_SYNC_QUEUED so writeback list handling is
+ * fully under our control.
+ */
while (!list_empty(&tmp)) {
sb = wb_inode(tmp.prev)->i_sb;
list_for_each_prev_safe(pos, node, &tmp) {
@@ -1853,8 +1859,8 @@ static long writeback_sb_inodes(struct s
* We'll have another go at writing back this inode
* when we completed a full scan of b_io.
*/
- spin_unlock(&inode->i_lock);
requeue_io(inode, wb);
+ spin_unlock(&inode->i_lock);
trace_writeback_sb_inodes_requeue(inode);
continue;
}
@@ -2389,6 +2395,7 @@ void __mark_inode_dirty(struct inode *in
{
struct super_block *sb = inode->i_sb;
int dirtytime = 0;
+ struct bdi_writeback *wb = NULL;
trace_writeback_mark_inode_dirty(inode, flags);
@@ -2441,13 +2448,24 @@ void __mark_inode_dirty(struct inode *in
inode->i_state |= flags;
/*
+ * Grab inode's wb early because it requires dropping i_lock and we
+ * need to make sure following checks happen atomically with dirty
+ * list handling so that we don't move inodes under flush worker's
+ * hands.
+ */
+ if (!was_dirty) {
+ wb = locked_inode_to_wb_and_lock_list(inode);
+ spin_lock(&inode->i_lock);
+ }
+
+ /*
* If the inode is queued for writeback by flush worker, just
* update its dirty state. Once the flush worker is done with
* the inode it will place it on the appropriate superblock
* list, based upon its state.
*/
if (inode->i_state & I_SYNC_QUEUED)
- goto out_unlock_inode;
+ goto out_unlock;
/*
* Only add valid (hashed) inodes to the superblock's
@@ -2455,22 +2473,19 @@ void __mark_inode_dirty(struct inode *in
*/
if (!S_ISBLK(inode->i_mode)) {
if (inode_unhashed(inode))
- goto out_unlock_inode;
+ goto out_unlock;
}
if (inode->i_state & I_FREEING)
- goto out_unlock_inode;
+ goto out_unlock;
/*
* If the inode was already on b_dirty/b_io/b_more_io, don't
* reposition it (that would break b_dirty time-ordering).
*/
if (!was_dirty) {
- struct bdi_writeback *wb;
struct list_head *dirty_list;
bool wakeup_bdi = false;
- wb = locked_inode_to_wb_and_lock_list(inode);
-
inode->dirtied_when = jiffies;
if (dirtytime)
inode->dirtied_time_when = jiffies;
@@ -2484,6 +2499,7 @@ void __mark_inode_dirty(struct inode *in
dirty_list);
spin_unlock(&wb->list_lock);
+ spin_unlock(&inode->i_lock);
trace_writeback_dirty_inode_enqueue(inode);
/*
@@ -2498,6 +2514,9 @@ void __mark_inode_dirty(struct inode *in
return;
}
}
+out_unlock:
+ if (wb)
+ spin_unlock(&wb->list_lock);
out_unlock_inode:
spin_unlock(&inode->i_lock);
}
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -27,7 +27,7 @@
* Inode locking rules:
*
* inode->i_lock protects:
- * inode->i_state, inode->i_hash, __iget()
+ * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
* Inode LRU list locks protect:
* inode->i_sb->s_inode_lru, inode->i_lru
* inode->i_sb->s_inode_list_lock protects:
