Skip to contentSkip to conversation

btrfs: sysfs: fix NULL pointer dereference at btrfs_sysfs_del_qgroups()

PATCH In review Qu Wenruo 4 participants 5 messages Aug 3, 2020 – Aug 11, 2020 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH 1/1 btrfs: sysfs: fix NULL pointer dereference at btrfs_sysfs_del_qgroups()

[BUG]
With next-20200731 tag (079ad2fb4bf9eba8a0aaab014b49705cd7f07c66), unmounting a btrfs with quota disabled will cause the following NULL pointer dereference:

BTRFS info (device dm-5): has skinny extents

  BUG: kernel NULL pointer dereference, address: 0000000000000018
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  CPU: 7 PID: 637 Comm: umount Not tainted 5.8.0-rc7-next-20200731-custom #76
  RIP: 0010:kobject_del+0x6/0x20
  Call Trace:
   btrfs_sysfs_del_qgroups+0xac/0xf0 [btrfs]
   btrfs_free_qgroup_config+0x63/0x70 [btrfs]
   close_ctree+0x1f5/0x323 [btrfs]
   btrfs_put_super+0x15/0x17 [btrfs]
   generic_shutdown_super+0x72/0x110
   kill_anon_super+0x18/0x30
   btrfs_kill_super+0x17/0x30 [btrfs]
   deactivate_locked_super+0x3b/0xa0
   deactivate_super+0x40/0x50
   cleanup_mnt+0x135/0x190
   __cleanup_mnt+0x12/0x20
   task_work_run+0x64/0xb0
   exit_to_user_mode_prepare+0x18a/0x190
   syscall_exit_to_user_mode+0x4f/0x270
   do_syscall_64+0x45/0x50
   entry_SYSCALL_64_after_hwframe+0x44/0xa9
  ---[ end trace 37b7adca5c1d5c5d ]---

[CAUSE]
Commit 079ad2fb4bf9 ("kobject: Avoid premature parent object freeing in kobject_cleanup()") changed kobject_del() that it no longer accepts NULL pointer.

Before that commit, kobject_del() and kobject_put() all accept NULL pointers and just ignore such NULL pointers.

But that mentioned commit needs to access the parent node, killing the old NULL pointer behavior.

Unfortunately btrfs is relying on that hidden feature thus we will trigger such NULL pointer dereference.

[FIX]
Instead of just saving several lines, do proper fs_info->qgroups_kobj check before calling kobject_del() and kobject_put().

---
 fs/btrfs/sysfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/btrfs/sysfs.c +5 -3
index 104c80caaa74..c8df2edafd85 100644
@@ -1565,9 +1565,11 @@ void btrfs_sysfs_del_qgroups(struct btrfs_fs_info *fs_info)
 	rbtree_postorder_for_each_entry_safe(qgroup, next,
 					     &fs_info->qgroup_tree, node)
 		btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
-	kobject_del(fs_info->qgroups_kobj);
-	kobject_put(fs_info->qgroups_kobj);
-	fs_info->qgroups_kobj = NULL;
+	if (fs_info->qgroups_kobj) {
+		kobject_del(fs_info->qgroups_kobj);
+		kobject_put(fs_info->qgroups_kobj);
+		fs_info->qgroups_kobj = NULL;
+	}
 }
 
 /* Called when qgroups get initialized, thus there is no need for locking */
Signature · 1 line
2.28.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 41 lines · Qu Wenruo

[BUG]
With next-20200731 tag (079ad2fb4bf9eba8a0aaab014b49705cd7f07c66), unmounting a btrfs with quota disabled will cause the following NULL pointer dereference:

BTRFS info (device dm-5): has skinny extents

  BUG: kernel NULL pointer dereference, address: 0000000000000018
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  CPU: 7 PID: 637 Comm: umount Not tainted 5.8.0-rc7-next-20200731-custom #76
  RIP: 0010:kobject_del+0x6/0x20
  Call Trace:
   btrfs_sysfs_del_qgroups+0xac/0xf0 [btrfs]
   btrfs_free_qgroup_config+0x63/0x70 [btrfs]
   close_ctree+0x1f5/0x323 [btrfs]
   btrfs_put_super+0x15/0x17 [btrfs]
   generic_shutdown_super+0x72/0x110
   kill_anon_super+0x18/0x30
   btrfs_kill_super+0x17/0x30 [btrfs]
   deactivate_locked_super+0x3b/0xa0
   deactivate_super+0x40/0x50
   cleanup_mnt+0x135/0x190
   __cleanup_mnt+0x12/0x20
   task_work_run+0x64/0xb0
   exit_to_user_mode_prepare+0x18a/0x190
   syscall_exit_to_user_mode+0x4f/0x270
   do_syscall_64+0x45/0x50
   entry_SYSCALL_64_after_hwframe+0x44/0xa9
  ---[ end trace 37b7adca5c1d5c5d ]---

[CAUSE]
Commit 079ad2fb4bf9 ("kobject: Avoid premature parent object freeing in kobject_cleanup()") changed kobject_del() that it no longer accepts NULL pointer.

Before that commit, kobject_del() and kobject_put() all accept NULL pointers and just ignore such NULL pointers.

But that mentioned commit needs to access the parent node, killing the old NULL pointer behavior.

Quoted 8 lines

Unfortunately btrfs is relying on that hidden feature thus we will trigger such NULL pointer dereference.

[FIX]
Instead of just saving several lines, do proper fs_info->qgroups_kobj check before calling kobject_del() and kobject_put().

Quoted 24 lines
---
 fs/btrfs/sysfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/btrfs/sysfs.c +5 -3
index 104c80caaa74..c8df2edafd85 100644
@@ -1565,9 +1565,11 @@ void btrfs_sysfs_del_qgroups(struct btrfs_fs_info *fs_info)
 	rbtree_postorder_for_each_entry_safe(qgroup, next,
 					     &fs_info->qgroup_tree, node)
 		btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
-	kobject_del(fs_info->qgroups_kobj);
-	kobject_put(fs_info->qgroups_kobj);
-	fs_info->qgroups_kobj = NULL;
+	if (fs_info->qgroups_kobj) {
+		kobject_del(fs_info->qgroups_kobj);
+		kobject_put(fs_info->qgroups_kobj);
+		fs_info->qgroups_kobj = NULL;
+	}
 }
 
 /* Called when qgroups get initialized, thus there is no need for locking */
#2 lore raw
In reply to Qu Wenruo
Quoted 2 lines · Qu Wenruo

[BUG]
With next-20200731 tag (079ad2fb4bf9eba8a0aaab014b49705cd7f07c66),

I don't think linux-next commit ids are useful, even the tags get removed after a month so the exact commit causing the crash is what we want.

Quoted 28 lines

unmounting a btrfs with quota disabled will cause the following NULL pointer dereference:

BTRFS info (device dm-5): has skinny extents

  BUG: kernel NULL pointer dereference, address: 0000000000000018
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  CPU: 7 PID: 637 Comm: umount Not tainted 5.8.0-rc7-next-20200731-custom #76
  RIP: 0010:kobject_del+0x6/0x20
  Call Trace:
   btrfs_sysfs_del_qgroups+0xac/0xf0 [btrfs]
   btrfs_free_qgroup_config+0x63/0x70 [btrfs]
   close_ctree+0x1f5/0x323 [btrfs]
   btrfs_put_super+0x15/0x17 [btrfs]
   generic_shutdown_super+0x72/0x110
   kill_anon_super+0x18/0x30
   btrfs_kill_super+0x17/0x30 [btrfs]
   deactivate_locked_super+0x3b/0xa0
   deactivate_super+0x40/0x50
   cleanup_mnt+0x135/0x190
   __cleanup_mnt+0x12/0x20
   task_work_run+0x64/0xb0
   exit_to_user_mode_prepare+0x18a/0x190
   syscall_exit_to_user_mode+0x4f/0x270
   do_syscall_64+0x45/0x50
   entry_SYSCALL_64_after_hwframe+0x44/0xa9
  ---[ end trace 37b7adca5c1d5c5d ]---
Quoted 4 lines

[CAUSE]
Commit 079ad2fb4bf9 ("kobject: Avoid premature parent object freeing in kobject_cleanup()") changed kobject_del() that it no longer accepts NULL pointer.

That commit reference should be sufficient.

Quoted 6 lines

Before that commit, kobject_del() and kobject_put() all accept NULL pointers and just ignore such NULL pointers.

But that mentioned commit needs to access the parent node, killing the old NULL pointer behavior.

Quoted 8 lines

Unfortunately btrfs is relying on that hidden feature thus we will trigger such NULL pointer dereference.

[FIX]
Instead of just saving several lines, do proper fs_info->qgroups_kobj check before calling kobject_del() and kobject_put().

Added to misc-next, thanks.

#3 lore raw
Quoted 52 lines · David Sterba
Quoted 2 lines · Qu Wenruo

[BUG]
With next-20200731 tag (079ad2fb4bf9eba8a0aaab014b49705cd7f07c66),

I don't think linux-next commit ids are useful, even the tags get removed after a month so the exact commit causing the crash is what we want.

Quoted 28 lines

unmounting a btrfs with quota disabled will cause the following NULL pointer dereference:

BTRFS info (device dm-5): has skinny extents

  BUG: kernel NULL pointer dereference, address: 0000000000000018
  #PF: supervisor read access in kernel mode
  #PF: error_code(0x0000) - not-present page
  CPU: 7 PID: 637 Comm: umount Not tainted 5.8.0-rc7-next-20200731-custom #76
  RIP: 0010:kobject_del+0x6/0x20
  Call Trace:
   btrfs_sysfs_del_qgroups+0xac/0xf0 [btrfs]
   btrfs_free_qgroup_config+0x63/0x70 [btrfs]
   close_ctree+0x1f5/0x323 [btrfs]
   btrfs_put_super+0x15/0x17 [btrfs]
   generic_shutdown_super+0x72/0x110
   kill_anon_super+0x18/0x30
   btrfs_kill_super+0x17/0x30 [btrfs]
   deactivate_locked_super+0x3b/0xa0
   deactivate_super+0x40/0x50
   cleanup_mnt+0x135/0x190
   __cleanup_mnt+0x12/0x20
   task_work_run+0x64/0xb0
   exit_to_user_mode_prepare+0x18a/0x190
   syscall_exit_to_user_mode+0x4f/0x270
   do_syscall_64+0x45/0x50
   entry_SYSCALL_64_after_hwframe+0x44/0xa9
  ---[ end trace 37b7adca5c1d5c5d ]---
Quoted 4 lines

[CAUSE]
Commit 079ad2fb4bf9 ("kobject: Avoid premature parent object freeing in kobject_cleanup()") changed kobject_del() that it no longer accepts NULL pointer.

That commit reference should be sufficient.

Quoted 9 lines

Before that commit, kobject_del() and kobject_put() all accept NULL pointers and just ignore such NULL pointers.

But that mentioned commit needs to access the parent node, killing the old NULL pointer behavior.

Unfortunately btrfs is relying on that hidden feature thus we will trigger such NULL pointer dereference.

Quoted 8 lines
Quoted 5 lines

[FIX]
Instead of just saving several lines, do proper fs_info->qgroups_kobj check before calling kobject_del() and kobject_put().

Added to misc-next, thanks.
Just to add a note, kobject guys are going to restore the NULL-awareness

behavior, so the patch here is just to be extra safe.

Thanks,
Qu

#4 lore raw
Quoted 4 lines · Qu Wenruo
Quoted 1 line

Added to misc-next, thanks.

Just to add a note, kobject guys are going to restore the NULL-awareness behavior, so the patch here is just to be extra safe.

We have the additional pointer checks in other locations too, so adding it here is fine.