Skip to contentSkip to conversation

ocfs2: xattr problems with 5.4.0-rc7

DISCUSSION Thomas Voegtle 2 participants 3 messages Nov 12, 2019 – Nov 13, 2019

lore
Share & tools
lore
#0 lore raw

Hello,

with 5.4.0-rc7 and 4.9.200 we see the following errors with mkdir or touch on a ocfs2 mountpoint:

root@s2:/shared/ClusterShareDisk# mkdir dir
mkdir: cannot create directory ‘dir’: Invalid argument

which produces this output:

root@s2:/shared/ClusterShareDisk# dmesg
[ 6918.815770] (mkdir,19461,0):ocfs2_xa_set:2242 ERROR: status = -22
[ 6918.815772] (mkdir,19461,0):ocfs2_mknod:408 ERROR: status = -22
[ 6918.816215] (mkdir,19461,0):ocfs2_mknod:486 ERROR: status = -22
[ 6918.816216] (mkdir,19461,0):ocfs2_mkdir:652 ERROR: status = -22

We got some ACLs:

$ getfacl /shared/ClusterShareDisk/
getfacl: Removing leading '/' from absolute path names
# file: shared/ClusterShareDisk/
# owner: root
# group: root
user::rwx
user:admin:rwx
group::rwx
mask::rwx
other::---
default:user::rwx
default:user:admin:rwx
default:group::rwx
default:mask::rwx
default:other::---

And of course it is mounted with user_xattr and acl option.

Reverting
commit 56e94ea132bb5c2c1d0b60a6aeb34dcb7d71a53d

Author: Jia-Ju Bai <[email protected]>
Date:   Sun Oct 6 17:57:50 2019 -0700

fs: ocfs2: fix possible null-pointer dereferences in
ocfs2_xa_prepare_entry()

fixes the problem.

Greetings,

Thomas

#1 lore raw

Hi Thomas,
Thanks for reporting this issue.
I have some doubts on it, please see my comments below.

Quoted 8 lines · Thomas Voegtle

Hello,

with 5.4.0-rc7 and 4.9.200 we see the following errors with mkdir or touch on a ocfs2 mountpoint:

root@s2:/shared/ClusterShareDisk# mkdir dir
mkdir: cannot create directory ‘dir’: Invalid argument
Quoted 8 lines

which produces this output:

root@s2:/shared/ClusterShareDisk# dmesg
[ 6918.815770] (mkdir,19461,0):ocfs2_xa_set:2242 ERROR: status = -22
[ 6918.815772] (mkdir,19461,0):ocfs2_mknod:408 ERROR: status = -22
[ 6918.816215] (mkdir,19461,0):ocfs2_mknod:486 ERROR: status = -22
[ 6918.816216] (mkdir,19461,0):ocfs2_mkdir:652 ERROR: status = -22

ocfs2_xa_set
ocfs2_xa_prepare_entry

Since ocfs2_xa_set() returns -EINVAL, it means loc->xl_entry is NULL.

	if (!loc->xl_entry) {
		rc = -EINVAL;
		goto out;
	}

After reverting
commit 56e94ea132bb "fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()", it will call ocfs2_xa_add_entry(),

	if (loc->xl_entry) {
		...
	} else
		ocfs2_xa_add_entry(loc, name_hash);

Theoretically it will cause NULL pointer dereference when access attributes of loc->xl_entry. So could you please check if the issue is caused by NULL loc->xl_entry?

Thanks,
Joseph

Quoted 33 lines

We got some ACLs:

$ getfacl /shared/ClusterShareDisk/
getfacl: Removing leading '/' from absolute path names
# file: shared/ClusterShareDisk/
# owner: root
# group: root
user::rwx
user:admin:rwx
group::rwx
mask::rwx
other::---
default:user::rwx
default:user:admin:rwx
default:group::rwx
default:mask::rwx
default:other::---

And of course it is mounted with user_xattr and acl option.

Reverting
commit 56e94ea132bb5c2c1d0b60a6aeb34dcb7d71a53d
Author: Jia-Ju Bai <[email protected]>
Date:   Sun Oct 6 17:57:50 2019 -0700

    fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()

fixes the problem.

Greetings,

    Thomas

#2 lore raw
Quoted 35 lines · Joseph Qi

Hi Thomas,
Thanks for reporting this issue.
I have some doubts on it, please see my comments below.

Quoted 8 lines · Thomas Voegtle

Hello,

with 5.4.0-rc7 and 4.9.200 we see the following errors with mkdir or touch on a ocfs2 mountpoint:

root@s2:/shared/ClusterShareDisk# mkdir dir
mkdir: cannot create directory ‘dir’: Invalid argument
Quoted 8 lines

which produces this output:

root@s2:/shared/ClusterShareDisk# dmesg
[ 6918.815770] (mkdir,19461,0):ocfs2_xa_set:2242 ERROR: status = -22
[ 6918.815772] (mkdir,19461,0):ocfs2_mknod:408 ERROR: status = -22
[ 6918.816215] (mkdir,19461,0):ocfs2_mknod:486 ERROR: status = -22
[ 6918.816216] (mkdir,19461,0):ocfs2_mkdir:652 ERROR: status = -22

ocfs2_xa_set
ocfs2_xa_prepare_entry

Since ocfs2_xa_set() returns -EINVAL, it means loc->xl_entry is NULL.

	if (!loc->xl_entry) {
		rc = -EINVAL;
		goto out;
	}

After reverting
commit 56e94ea132bb "fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()", it will call ocfs2_xa_add_entry(),

Quoted 8 lines
	if (loc->xl_entry) {
		...
	} else
		ocfs2_xa_add_entry(loc, name_hash);

Theoretically it will cause NULL pointer dereference when access attributes of loc->xl_entry. So could you please check if the issue is caused by NULL loc->xl_entry?

I think I've got the answer. loc->xl_ops->xlo_add_entry() will handle this case. I'll revert this patch.

Thanks,
Joseph

Quoted 34 lines
Quoted 34 lines

We got some ACLs:

$ getfacl /shared/ClusterShareDisk/
getfacl: Removing leading '/' from absolute path names
# file: shared/ClusterShareDisk/
# owner: root
# group: root
user::rwx
user:admin:rwx
group::rwx
mask::rwx
other::---
default:user::rwx
default:user:admin:rwx
default:group::rwx
default:mask::rwx
default:other::---

And of course it is mounted with user_xattr and acl option.

Reverting
commit 56e94ea132bb5c2c1d0b60a6aeb34dcb7d71a53d
Author: Jia-Ju Bai <[email protected]>
Date:   Sun Oct 6 17:57:50 2019 -0700

    fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()

fixes the problem.

Greetings,

    Thomas