Re: [PATCH v2] ocfs2: fix deadlock caused by recursive locking in xattr
From: Eric Ren
Date: Thu Jun 22 2017 - 19:35:45 EST
Hi Andrew,
On 06/23/2017 05:24 AM, Andrew Morton wrote:
On Thu, 22 Jun 2017 14:10:38 +0800 Joseph Qi <jiangqi903@xxxxxxxxx> wrote:
Looks good.
Reviewed-by: Joseph Qi <jiangqi903@xxxxxxxxx>
Should this fix be backported into -stable kernels?
No, I think, because the previous patches that this one needs to be on,
- commit 439a36b8ef38 ("ocfs2/dlmglue: prepare tracking logic to avoid recursive cluster lock").
- commit b891fa5024a9 ("ocfs2: fix deadlock issue when taking inode lock at vfs entry points")
is not in --stable too.
I don't know if it's possible to make them all into stable.
Thanks,
Eric
On 17/6/22 09:47, Eric Ren wrote:
Another deadlock path caused by recursive locking is reported.
This kind of issue was introduced since commit 743b5f1434f5 ("ocfs2:
take inode lock in ocfs2_iop_set/get_acl()"). Two deadlock paths
have been fixed by commit b891fa5024a9 ("ocfs2: fix deadlock issue when
taking inode lock at vfs entry points"). Yes, we intend to fix this
kind of case in incremental way, because it's hard to find out all
possible paths at once.
This one can be reproduced like this. On node1, cp a large file from
home directory to ocfs2 mountpoint. While on node2, run setfacl/getfacl.
Both nodes will hang up there. The backtraces:
On node1:
[<ffffffffc06a1f67>] __ocfs2_cluster_lock.isra.39+0x357/0x740 [ocfs2]
[<ffffffffc06a2d3d>] ocfs2_inode_lock_full_nested+0x17d/0x840 [ocfs2]
[<ffffffffc0692043>] ocfs2_write_begin+0x43/0x1a0 [ocfs2]
[<ffffffffa21ac719>] generic_perform_write+0xa9/0x180
[<ffffffffa21aecda>] __generic_file_write_iter+0x1aa/0x1d0
[<ffffffffc06abc24>] ocfs2_file_write_iter+0x4f4/0xb40 [ocfs2]
[<ffffffffa223c3b3>] __vfs_write+0xc3/0x130
[<ffffffffa223cae1>] vfs_write+0xb1/0x1a0
[<ffffffffa223dfe6>] SyS_write+0x46/0xa0
On node2:
[<ffffffffc07b6f67>] __ocfs2_cluster_lock.isra.39+0x357/0x740 [ocfs2]
[<ffffffffc07b7d3d>] ocfs2_inode_lock_full_nested+0x17d/0x840 [ocfs2]
[<ffffffffc0815c1e>] ocfs2_xattr_set+0x12e/0xe80 [ocfs2]
[<ffffffffc081783d>] ocfs2_set_acl+0x22d/0x260 [ocfs2]
[<ffffffffc08178d5>] ocfs2_iop_set_acl+0x65/0xb0 [ocfs2]
[<ffffffffa62a43f5>] set_posix_acl+0x75/0xb0
[<ffffffffa62a4479>] posix_acl_xattr_set+0x49/0xa0
[<ffffffffa6265c69>] __vfs_setxattr+0x69/0x80
[<ffffffffa6266832>] __vfs_setxattr_noperm+0x72/0x1a0
[<ffffffffa6266a07>] vfs_setxattr+0xa7/0xb0
[<ffffffffa6266b3d>] setxattr+0x12d/0x190
[<ffffffffa6266c3f>] path_setxattr+0x9f/0xb0
[<ffffffffa6266d74>] SyS_setxattr+0x14/0x20
Fixes this one by using ocfs2_inode_{lock|unlock}_tracker, which is
exported by commit 439a36b8ef38 ("ocfs2/dlmglue: prepare tracking
logic to avoid recursive cluster lock").
Changes since v1:
- Revised git commit description style in commit log.
Reported-by: Thomas Voegtle <tv@xxxxxxxx>
Tested-by: Thomas Voegtle <tv@xxxxxxxx>
Signed-off-by: Eric Ren <zren@xxxxxxxx>
---
fs/ocfs2/dlmglue.c | 4 ++++
fs/ocfs2/xattr.c | 23 +++++++++++++----------
2 files changed, 17 insertions(+), 10 deletions(-)
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index 3b7c937..4689940 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2591,6 +2591,10 @@ void ocfs2_inode_unlock_tracker(struct inode *inode,
struct ocfs2_lock_res *lockres;
lockres = &OCFS2_I(inode)->ip_inode_lockres;
+ /* had_lock means that the currect process already takes the cluster
+ * lock previously. If had_lock is 1, we have nothing to do here, and
+ * it will get unlocked where we got the lock.
+ */
if (!had_lock) {
ocfs2_remove_holder(lockres, oh);
ocfs2_inode_unlock(inode, ex);
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 3c5384d..f70c377 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -1328,20 +1328,21 @@ static int ocfs2_xattr_get(struct inode *inode,
void *buffer,
size_t buffer_size)
{
- int ret;
+ int ret, had_lock;
struct buffer_head *di_bh = NULL;
+ struct ocfs2_lock_holder oh;
- ret = ocfs2_inode_lock(inode, &di_bh, 0);
- if (ret < 0) {
- mlog_errno(ret);
- return ret;
+ had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
+ if (had_lock < 0) {
+ mlog_errno(had_lock);
+ return had_lock;
}
down_read(&OCFS2_I(inode)->ip_xattr_sem);
ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
name, buffer, buffer_size);
up_read(&OCFS2_I(inode)->ip_xattr_sem);
- ocfs2_inode_unlock(inode, 0);
+ ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
brelse(di_bh);
@@ -3537,11 +3538,12 @@ int ocfs2_xattr_set(struct inode *inode,
{
struct buffer_head *di_bh = NULL;
struct ocfs2_dinode *di;
- int ret, credits, ref_meta = 0, ref_credits = 0;
+ int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct inode *tl_inode = osb->osb_tl_inode;
struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
struct ocfs2_refcount_tree *ref_tree = NULL;
+ struct ocfs2_lock_holder oh;
struct ocfs2_xattr_info xi = {
.xi_name_index = name_index,
@@ -3572,8 +3574,9 @@ int ocfs2_xattr_set(struct inode *inode,
return -ENOMEM;
}
- ret = ocfs2_inode_lock(inode, &di_bh, 1);
- if (ret < 0) {
+ had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
+ if (had_lock < 0) {
+ ret = had_lock;
mlog_errno(ret);
goto cleanup_nolock;
}
@@ -3670,7 +3673,7 @@ int ocfs2_xattr_set(struct inode *inode,
if (ret)
mlog_errno(ret);
}
- ocfs2_inode_unlock(inode, 1);
+ ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
cleanup_nolock:
brelse(di_bh);
brelse(xbs.xattr_bh);