btrfs: qgroup: fix sleep from invalid context bug in update_qgroup_limit_item()
PATCH v2 1/1 btrfs: qgroup: fix sleep from invalid context bug in update_qgroup_limit_item()
Syzkaller reported BUG as follows:
BUG: sleeping function called from invalid context at
include/linux/sched/mm.h:274
Call Trace:
<TASK>
dump_stack_lvl+0xcd/0x134
__might_resched.cold+0x222/0x26b
kmem_cache_alloc+0x2e7/0x3c0
update_qgroup_limit_item+0xe1/0x390
btrfs_qgroup_inherit+0x147b/0x1ee0
create_subvol+0x4eb/0x1710
btrfs_mksubvol+0xfe5/0x13f0
__btrfs_ioctl_snap_create+0x2b0/0x430
btrfs_ioctl_snap_create_v2+0x25a/0x520
btrfs_ioctl+0x2a1c/0x5ce0
__x64_sys_ioctl+0x193/0x200
do_syscall_64+0x35/0x80Fix this by delaying the limit item updates until unlock the spin lock.
--- fs/btrfs/qgroup.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)
fs/btrfs/qgroup.c +9 -9
@@ -2860,6 +2860,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,bool need_rescan = false;u32 level_size =0 ;u64 nums; +bool update_limit = false;/* * There are only two callers of this function.
@@ -2950,15 +2951,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, dstgroup->max_excl = inherit->lim.max_excl; dstgroup->rsv_rfer = inherit->lim.rsv_rfer; dstgroup->rsv_excl = inherit->lim.rsv_excl; - - ret =update_qgroup_limit_item (trans, dstgroup); -if (ret) { -qgroup_mark_inconsistent (fs_info); -btrfs_info (fs_info, -"unable to update quota limit for %llu" , - dstgroup->qgroupid); - goto unlock; - } + update_limit = true; }if (srcid) {
@@ -2985,6 +2978,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, dstgroup->max_excl = srcgroup->max_excl; dstgroup->rsv_rfer = srcgroup->rsv_rfer; dstgroup->rsv_excl = srcgroup->rsv_excl; + update_limit = true;qgroup_dirty (fs_info, dstgroup);qgroup_dirty (fs_info, srcgroup);
@@ -3053,6 +3047,12 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, unlock:spin_unlock (&fs_info->qgroup_lock); +if (update_limit &&update_qgroup_limit_item (trans, dstgroup)) { +qgroup_mark_inconsistent (fs_info); +btrfs_info (fs_info, +"unable to update quota limit for %llu" , + dstgroup->qgroupid); + }if (!ret) ret =btrfs_sysfs_add_one_qgroup (fs_info, dstgroup); out:
Signature · 1 line
2.31.1
Quoted 19 lines · ChenXiaoSong
Syzkaller reported BUG as follows:
BUG: sleeping function called from invalid context at include/linux/sched/mm.h:274 Call Trace: <TASK> dump_stack_lvl+0xcd/0x134 __might_resched.cold+0x222/0x26b kmem_cache_alloc+0x2e7/0x3c0 update_qgroup_limit_item+0xe1/0x390 btrfs_qgroup_inherit+0x147b/0x1ee0 create_subvol+0x4eb/0x1710 btrfs_mksubvol+0xfe5/0x13f0 __btrfs_ioctl_snap_create+0x2b0/0x430 btrfs_ioctl_snap_create_v2+0x25a/0x520 btrfs_ioctl+0x2a1c/0x5ce0 __x64_sys_ioctl+0x193/0x200 do_syscall_64+0x35/0x80
Quoted 1 line
Fix this by delaying the limit item updates until unlock the spin lock.
The overall idea is way better now.
But sorry that I didn't immediately find out the best solution at the first glance.
In fact, your v2 path just lets me remember what is the correct way to handle such situation.
Quoted 10 lines
Signed-off-by ChenXiaoSong <[email protected]>--- fs/btrfs/qgroup.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)fs/btrfs/qgroup.c +0
-0
[...]
fs/btrfs/qgroup.c int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, · line 2984
Quoted 8 lines
fs/btrfs/qgroup.c +1
-0@@ -2985,6 +2978,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, dstgroup->max_excl = srcgroup->max_excl; dstgroup->rsv_rfer = srcgroup->rsv_rfer; dstgroup->rsv_excl = srcgroup->rsv_excl; + update_limit = true;qgroup_dirty (fs_info, dstgroup);qgroup_dirty (fs_info, srcgroup);
You caught the "if (srcid)" branch, which also changed the limit/rfer/excl numbers of the destination qgroup, but didn't call update_qgroup_limit_item().
But if you check the function qgroup_dirty() a little deeper, you can find out that, qgroup_dirty() will move the target qgroup into fs_info->dirty_qgroups list.
And later at btrfs_run_qgroups() (which is called during btrfs_commit_transaction(), and also after create_pending_snapshots()), we will update the quota tree to reflect the result.
So this means, all you need is just call qgroup_dirty() on @dstqgroup, and call it a day. Everything else will be properly handled.
Sorry I didn't notice this earlier...
Thanks,
Qu
Quoted 13 lines
fs/btrfs/qgroup.c +6
-0@@ -3053,6 +3047,12 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, unlock:spin_unlock (&fs_info->qgroup_lock); +if (update_limit &&update_qgroup_limit_item (trans, dstgroup)) { +qgroup_mark_inconsistent (fs_info); +btrfs_info (fs_info, +"unable to update quota limit for %llu" , + dstgroup->qgroupid); + }if (!ret) ret =btrfs_sysfs_add_one_qgroup (fs_info, dstgroup); out:
Thanks for your suggestions, I will try to send a new version patch.
Quoted 80 lines · Qu Wenruo
Quoted 19 lines · ChenXiaoSong
Syzkaller reported BUG as follows:
BUG: sleeping function called from invalid context at include/linux/sched/mm.h:274 Call Trace: <TASK> dump_stack_lvl+0xcd/0x134 __might_resched.cold+0x222/0x26b kmem_cache_alloc+0x2e7/0x3c0 update_qgroup_limit_item+0xe1/0x390 btrfs_qgroup_inherit+0x147b/0x1ee0 create_subvol+0x4eb/0x1710 btrfs_mksubvol+0xfe5/0x13f0 __btrfs_ioctl_snap_create+0x2b0/0x430 btrfs_ioctl_snap_create_v2+0x25a/0x520 btrfs_ioctl+0x2a1c/0x5ce0 __x64_sys_ioctl+0x193/0x200 do_syscall_64+0x35/0x80Quoted 1 line
Fix this by delaying the limit item updates until unlock the spin lock.
The overall idea is way better now.
But sorry that I didn't immediately find out the best solution at the first glance.
In fact, your v2 path just lets me remember what is the correct way to handle such situation.
Quoted 10 lines
Signed-off-by ChenXiaoSong <[email protected]>---
fs/btrfs/qgroup.c| 18 +++++++++---------1 file changed, 9 insertions(+), 9 deletions(-)fs/btrfs/qgroup.c +0
-0[...]
fs/btrfs/qgroup.c int btrfs_qgroup_inherit(struct · line 2984
Quoted 8 lines
fs/btrfs/qgroup.c +1
-0@@ -2985,6 +2978,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, dstgroup->max_excl = srcgroup->max_excl; dstgroup->rsv_rfer = srcgroup->rsv_rfer; dstgroup->rsv_excl = srcgroup->rsv_excl; + update_limit = true; qgroup_dirty(fs_info, dstgroup); qgroup_dirty(fs_info, srcgroup);You caught the "if (srcid)" branch, which also changed the limit/rfer/excl numbers of the destination qgroup, but didn't call
update_qgroup_limit_item().But if you check the function
qgroup_dirty()a little deeper, you can find out that,qgroup_dirty()will move the target qgroup intofs_info->dirty_qgroupslist.And later at
btrfs_run_qgroups()(which is called duringbtrfs_commit_transaction(), and also aftercreate_pending_snapshots()), we will update the quota tree to reflect the result.So this means, all you need is just call
qgroup_dirty()on @dstqgroup, and call it a day. Everything else will be properly handled.Sorry I didn't notice this earlier...
Thanks,
QuQuoted 5 lines
fs/btrfs/qgroup.c +0
-0@@ -3053,6 +3047,12 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,unlock:
spin_unlock(&fs_info->qgroup_lock);
+ if (update_limit&&update_qgroup_limit_item(trans, dstgroup)) {Quoted 8 lines
+
qgroup_mark_inconsistent(fs_info);
+ btrfs_info(fs_info,
+ "unable to update quota limit for %llu", +dstgroup->qgroupid);
+ }
if (!ret)
ret =btrfs_sysfs_add_one_qgroup(fs_info, dstgroup); out:.