On Thu, Mar 18, 2021 at 11:25 AM <fdman...@kernel.org> wrote: > > 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>
Missed: Fixes: 49e5fb46211de0 ("btrfs: qgroup: export qgroups in sysfs") > --- > 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 >