writeback: avoid use-after-free after removing device

jira VULN-4108
cve CVE-2024-0562
commit-author Khazhismel Kumykov <[email protected]>
commit f87904c

When a disk is removed, bdi_unregister gets called to stop further
writeback and wait for associated delayed work to complete.  However,
wb_inode_writeback_end() may schedule bandwidth estimation dwork after
this has completed, which can result in the timer attempting to access the
just freed bdi_writeback.

Fix this by checking if the bdi_writeback is alive, similar to when
scheduling writeback work.

Since this requires wb->work_lock, and wb_inode_writeback_end() may get
called from interrupt, switch wb->work_lock to an irqsafe lock.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 45a2966 ("writeback: fix bandwidth estimate for spiky workload")
	Signed-off-by: Khazhismel Kumykov <[email protected]>
	Reviewed-by: Jan Kara <[email protected]>
	Cc: Michael Stapelberg <[email protected]>
	Cc: Wu Fengguang <[email protected]>
	Cc: Alexander Viro <[email protected]>
	Cc: <[email protected]>
	Signed-off-by: Andrew Morton <[email protected]>
(cherry picked from commit f87904c)
	Signed-off-by: Pratham Patel <[email protected]>

Author: thefossguy-ciq

fs/fs-writeback.c

@@ -151,10 +151,10 @@ static void inode_io_list_del_locked(struct inode *inode,
 
 static void wb_wakeup(struct bdi_writeback *wb)
 {
-	spin_lock_bh(&wb->work_lock);
+	spin_lock_irq(&wb->work_lock);
 	if (test_bit(WB_registered, &wb->state))
 		mod_delayed_work(bdi_wq, &wb->dwork, 0);
-	spin_unlock_bh(&wb->work_lock);
+	spin_unlock_irq(&wb->work_lock);
 }
 
 static void finish_writeback_work(struct bdi_writeback *wb,
@@ -181,15 +181,15 @@ static void wb_queue_work(struct bdi_writeback *wb,
 	if (work->done)
 		atomic_inc(&work->done->cnt);
 
-	spin_lock_bh(&wb->work_lock);
+	spin_lock_irq(&wb->work_lock);
 
 	if (test_bit(WB_registered, &wb->state)) {
 		list_add_tail(&work->list, &wb->work_list);
 		mod_delayed_work(bdi_wq, &wb->dwork, 0);
 	} else
 		finish_writeback_work(wb, work);
 
-	spin_unlock_bh(&wb->work_lock);
+	spin_unlock_irq(&wb->work_lock);
 }
 
 /**
@@ -1938,13 +1938,13 @@ static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
 {
 	struct wb_writeback_work *work = NULL;
 
-	spin_lock_bh(&wb->work_lock);
+	spin_lock_irq(&wb->work_lock);
 	if (!list_empty(&wb->work_list)) {
 		work = list_entry(wb->work_list.next,
 				  struct wb_writeback_work, list);
 		list_del_init(&work->list);
 	}
-	spin_unlock_bh(&wb->work_lock);
+	spin_unlock_irq(&wb->work_lock);
 	return work;
 }
 

mm/backing-dev.c

@@ -280,10 +280,10 @@ void wb_wakeup_delayed(struct bdi_writeback *wb)
 	unsigned long timeout;
 
 	timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
-	spin_lock_bh(&wb->work_lock);
+	spin_lock_irq(&wb->work_lock);
 	if (test_bit(WB_registered, &wb->state))
 		queue_delayed_work(bdi_wq, &wb->dwork, timeout);
-	spin_unlock_bh(&wb->work_lock);
+	spin_unlock_irq(&wb->work_lock);
 }
 
 static void wb_update_bandwidth_workfn(struct work_struct *work)
@@ -376,12 +376,12 @@ static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
 static void wb_shutdown(struct bdi_writeback *wb)
 {
 	/* Make sure nobody queues further work */
-	spin_lock_bh(&wb->work_lock);
+	spin_lock_irq(&wb->work_lock);
 	if (!test_and_clear_bit(WB_registered, &wb->state)) {
-		spin_unlock_bh(&wb->work_lock);
+		spin_unlock_irq(&wb->work_lock);
 		return;
 	}
-	spin_unlock_bh(&wb->work_lock);
+	spin_unlock_irq(&wb->work_lock);
 
 	cgwb_remove_from_bdi_list(wb);
 	/*

mm/page-writeback.c

@@ -2754,6 +2754,7 @@ static void wb_inode_writeback_start(struct bdi_writeback *wb)
 
 static void wb_inode_writeback_end(struct bdi_writeback *wb)
 {
+	unsigned long flags;
 	atomic_dec(&wb->writeback_inodes);
 	/*
 	 * Make sure estimate of writeback throughput gets updated after
@@ -2762,7 +2763,10 @@ static void wb_inode_writeback_end(struct bdi_writeback *wb)
 	 * that if multiple inodes end writeback at a similar time, they get
 	 * batched into one bandwidth update.
 	 */
-	queue_delayed_work(bdi_wq, wb->bw_dwork, BANDWIDTH_INTERVAL);
+	spin_lock_irqsave(&wb->work_lock, flags);
+	if (test_bit(WB_registered, &wb->state))
+		queue_delayed_work(bdi_wq, wb->bw_dwork, BANDWIDTH_INTERVAL);
+	spin_unlock_irqrestore(&wb->work_lock, flags);
 }
 
 int test_clear_page_writeback(struct page *page)