fs: writeback: fix use-after-free in __mark_inode_dirty()
authorJiufei Xue <jiufei.xue@samsung.com>
Mon, 28 Jul 2025 10:07:15 +0000 (18:07 +0800)
committerChristian Brauner <brauner@kernel.org>
Mon, 11 Aug 2025 12:51:45 +0000 (14:51 +0200)
An use-after-free issue occurred when __mark_inode_dirty() get the
bdi_writeback that was in the progress of switching.

CPU: 1 PID: 562 Comm: systemd-random- Not tainted 6.6.56-gb4403bd46a8e #1
......
pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : __mark_inode_dirty+0x124/0x418
lr : __mark_inode_dirty+0x118/0x418
sp : ffffffc08c9dbbc0
........
Call trace:
 __mark_inode_dirty+0x124/0x418
 generic_update_time+0x4c/0x60
 file_modified+0xcc/0xd0
 ext4_buffered_write_iter+0x58/0x124
 ext4_file_write_iter+0x54/0x704
 vfs_write+0x1c0/0x308
 ksys_write+0x74/0x10c
 __arm64_sys_write+0x1c/0x28
 invoke_syscall+0x48/0x114
 el0_svc_common.constprop.0+0xc0/0xe0
 do_el0_svc+0x1c/0x28
 el0_svc+0x40/0xe4
 el0t_64_sync_handler+0x120/0x12c
 el0t_64_sync+0x194/0x198

Root cause is:

systemd-random-seed                         kworker
----------------------------------------------------------------------
___mark_inode_dirty                     inode_switch_wbs_work_fn

  spin_lock(&inode->i_lock);
  inode_attach_wb
  locked_inode_to_wb_and_lock_list
     get inode->i_wb
     spin_unlock(&inode->i_lock);
     spin_lock(&wb->list_lock)
  spin_lock(&inode->i_lock)
  inode_io_list_move_locked
  spin_unlock(&wb->list_lock)
  spin_unlock(&inode->i_lock)
                                    spin_lock(&old_wb->list_lock)
                                      inode_do_switch_wbs
                                        spin_lock(&inode->i_lock)
                                        inode->i_wb = new_wb
                                        spin_unlock(&inode->i_lock)
                                    spin_unlock(&old_wb->list_lock)
                                    wb_put_many(old_wb, nr_switched)
                                      cgwb_release
                                      old wb released
  wb_wakeup_delayed() accesses wb,
  then trigger the use-after-free
  issue

Fix this race condition by holding inode spinlock until
wb_wakeup_delayed() finished.

Signed-off-by: Jiufei Xue <jiufei.xue@samsung.com>
Link: https://lore.kernel.org/20250728100715.3863241-1-jiufei.xue@samsung.com
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
fs/fs-writeback.c

index cc57367fb641d7bb448d65f132ef5734c3795422..a07b8cf73ae271fdb77c7da962d5069e284886eb 100644 (file)
@@ -2608,10 +2608,6 @@ void __mark_inode_dirty(struct inode *inode, int flags)
                        wakeup_bdi = inode_io_list_move_locked(inode, wb,
                                                               dirty_list);
 
-                       spin_unlock(&wb->list_lock);
-                       spin_unlock(&inode->i_lock);
-                       trace_writeback_dirty_inode_enqueue(inode);
-
                        /*
                         * If this is the first dirty inode for this bdi,
                         * we have to wake-up the corresponding bdi thread
@@ -2621,6 +2617,11 @@ void __mark_inode_dirty(struct inode *inode, int flags)
                        if (wakeup_bdi &&
                            (wb->bdi->capabilities & BDI_CAP_WRITEBACK))
                                wb_wakeup_delayed(wb);
+
+                       spin_unlock(&wb->list_lock);
+                       spin_unlock(&inode->i_lock);
+                       trace_writeback_dirty_inode_enqueue(inode);
+
                        return;
                }
        }