From: Filipe Manana <fdman...@suse.com> While removing a qgroup's sysfs entry we end up taking the kernfs_mutex, through kobject_del(), while holding the fs_info->qgroup_lock spinlock, producing the following trace:
[ 821.843637] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:281 [ 821.843641] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 28214, name: podman [ 821.843644] CPU: 3 PID: 28214 Comm: podman Tainted: G W 5.11.6 #15 [ 821.843646] Hardware name: Dell Inc. PowerEdge R330/084XW4, BIOS 2.11.0 12/08/2020 [ 821.843647] Call Trace: [ 821.843650] dump_stack+0xa1/0xfb [ 821.843656] ___might_sleep+0x144/0x160 [ 821.843659] mutex_lock+0x17/0x40 [ 821.843662] kernfs_remove_by_name_ns+0x1f/0x80 [ 821.843666] sysfs_remove_group+0x7d/0xe0 [ 821.843668] sysfs_remove_groups+0x28/0x40 [ 821.843670] kobject_del+0x2a/0x80 [ 821.843672] btrfs_sysfs_del_one_qgroup+0x2b/0x40 [btrfs] [ 821.843685] __del_qgroup_rb+0x12/0x150 [btrfs] [ 821.843696] btrfs_remove_qgroup+0x288/0x2a0 [btrfs] [ 821.843707] btrfs_ioctl+0x3129/0x36a0 [btrfs] [ 821.843717] ? __mod_lruvec_page_state+0x5e/0xb0 [ 821.843719] ? page_add_new_anon_rmap+0xbc/0x150 [ 821.843723] ? kfree+0x1b4/0x300 [ 821.843725] ? mntput_no_expire+0x55/0x330 [ 821.843728] __x64_sys_ioctl+0x5a/0xa0 [ 821.843731] do_syscall_64+0x33/0x70 [ 821.843733] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 821.843736] RIP: 0033:0x4cd3fb [ 821.843739] Code: fa ff eb bd e8 86 8b fa ff e9 61 ff ff ff cc e8 fb 55 fa ff 48 8b 7c 24 10 48 8b 74 24 18 48 8b 54 24 20 48 8b 44 24 08 0f 05 <48> 3d 01 f0 ff ff 76 20 48 c7 44 24 28 ff ff ff ff 48 c7 44 24 30 [ 821.843741] RSP: 002b:000000c000906b20 EFLAGS: 00000206 ORIG_RAX: 0000000000000010 [ 821.843744] RAX: ffffffffffffffda RBX: 000000c000050000 RCX: 00000000004cd3fb [ 821.843745] RDX: 000000c000906b98 RSI: 000000004010942a RDI: 000000000000000f [ 821.843747] RBP: 000000c000907cd0 R08: 000000c000622901 R09: 0000000000000000 [ 821.843748] R10: 000000c000d992c0 R11: 0000000000000206 R12: 000000000000012d [ 821.843749] R13: 000000000000012c R14: 0000000000000200 R15: 0000000000000049 Fix this by removing the qgroup sysfs entry while not holding the spinlock, since the spinlock is only meant for protection of the qgroup rbtree. Reported-by: Stuart Shelton <srcshel...@gmail.com> Link: https://lore.kernel.org/linux-btrfs/7a5485bb-0628-419d-a4d3-27b1af47e...@gmail.com/ Signed-off-by: Filipe Manana <fdman...@suse.com> --- fs/btrfs/qgroup.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c index 9d8d1ac86962..2319c923c9e6 100644 --- a/fs/btrfs/qgroup.c +++ b/fs/btrfs/qgroup.c @@ -227,7 +227,6 @@ static void __del_qgroup_rb(struct btrfs_fs_info *fs_info, { struct btrfs_qgroup_list *list; - btrfs_sysfs_del_one_qgroup(fs_info, qgroup); list_del(&qgroup->dirty); while (!list_empty(&qgroup->groups)) { list = list_first_entry(&qgroup->groups, @@ -244,7 +243,6 @@ static void __del_qgroup_rb(struct btrfs_fs_info *fs_info, list_del(&list->next_member); kfree(list); } - kfree(qgroup); } /* must be called with qgroup_lock held */ @@ -570,6 +568,8 @@ void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info) qgroup = rb_entry(n, struct btrfs_qgroup, node); rb_erase(n, &fs_info->qgroup_tree); __del_qgroup_rb(fs_info, qgroup); + btrfs_sysfs_del_one_qgroup(fs_info, qgroup); + kfree(qgroup); } /* * We call btrfs_free_qgroup_config() when unmounting @@ -1579,6 +1579,14 @@ int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid) spin_lock(&fs_info->qgroup_lock); del_qgroup_rb(fs_info, qgroupid); spin_unlock(&fs_info->qgroup_lock); + + /* + * Remove the qgroup from sysfs now without holding the qgroup_lock + * spinlock, since the sysfs_remove_group() function needs to take + * the mutex kernfs_mutex through kernfs_remove_by_name_ns(). + */ + btrfs_sysfs_del_one_qgroup(fs_info, qgroup); + kfree(qgroup); out: mutex_unlock(&fs_info->qgroup_ioctl_lock); return ret; -- 2.28.0
