On Sat, Nov 16, 2024 at 01:58:17AM +0900, Jeongjun Park wrote:
Chuck Lever chuck.lever@oracle.com wrote:
On Sat, Nov 16, 2024 at 01:33:19AM +0900, Jeongjun Park wrote:
2024년 11월 16일 (토) 오전 1:25, Chuck Lever chuck.lever@oracle.com님이 작성:
On Fri, Nov 15, 2024 at 11:04:56AM -0500, Chuck Lever wrote:
I've found that NFS access to an exported tmpfs file system hangs indefinitely when the client first performs a GETATTR. The hanging nfsd thread is waiting for the inode lock in shmem_getattr():
task:nfsd state:D stack:0 pid:1775 tgid:1775 ppid:2 flags:0x00004000 Call Trace:
<TASK> __schedule+0x770/0x7b0 schedule+0x33/0x50 schedule_preempt_disabled+0x19/0x30 rwsem_down_read_slowpath+0x206/0x230 down_read+0x3f/0x60 shmem_getattr+0x84/0xf0 vfs_getattr_nosec+0x9e/0xc0 vfs_getattr+0x49/0x50 fh_getattr+0x43/0x50 [nfsd] fh_fill_pre_attrs+0x4e/0xd0 [nfsd] nfsd4_open+0x51f/0x910 [nfsd] nfsd4_proc_compound+0x492/0x5d0 [nfsd] nfsd_dispatch+0x117/0x1f0 [nfsd] svc_process_common+0x3b2/0x5e0 [sunrpc] ? __pfx_nfsd_dispatch+0x10/0x10 [nfsd] svc_process+0xcf/0x130 [sunrpc] svc_recv+0x64e/0x750 [sunrpc] ? __wake_up_bit+0x4b/0x60 ? __pfx_nfsd+0x10/0x10 [nfsd] nfsd+0xc6/0xf0 [nfsd] kthread+0xed/0x100 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x2e/0x50 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 </TASK>
I bisected the problem to:
d949d1d14fa281ace388b1de978e8f2cd52875cf is the first bad commit commit d949d1d14fa281ace388b1de978e8f2cd52875cf Author: Jeongjun Park aha310510@gmail.com AuthorDate: Mon Sep 9 21:35:58 2024 +0900 Commit: Andrew Morton akpm@linux-foundation.org CommitDate: Mon Oct 28 21:40:39 2024 -0700
mm: shmem: fix data-race in shmem_getattr()...
Link: https://lkml.kernel.org/r/20240909123558.70229-1-aha310510@gmail.com Fixes: 44a30220bc0a ("shmem: recalculate file inode when fstat") Signed-off-by: Jeongjun Park <aha310510@gmail.com> Reported-by: syzbot <syzkaller@googlegroup.com> Cc: Hugh Dickins <hughd@google.com> Cc: Yu Zhao <yuzhao@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>which first appeared in v6.12-rc6, and adds the line that is waiting on the inode lock when my NFS server hangs.
I haven't yet found the process that is holding the inode lock for this inode.
It is likely that the caller (nfsd4_open()-> fh_fill_pre_attrs()) is already holding the inode semaphore in this case.
Thanks for letting me know!
It seems that the previous patch I wrote was wrong in how to prevent data-race. It seems that the problem occurs in nfsd because nfsd4_create_file() already holds the inode_lock.
After further analysis, I found that this data-race mainly occurs when vfs_statx_path does not acquire the inode_lock, and in other filesystems, it is confirmed that inode_lock is acquired in many cases, so I will send a new patch that fixes this problem right away.
Thanks for your quick response!
My brief sample of file system ->getattr methods shows that these functions do not grab the inode semaphore at all when calling generic_fillattr(). Likely they expect the method's caller to take it.
I strongly prefer to see this commit reverted for v6.12-rc first, and then the new fix should be merged via a normal merge window to permit a lengthy period of testing.
Hmm... Of course, revert this patch is not a bad idea, but I think that patching it like below can effectively prevent data-race without causing recursive locking:
mm/shmem.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/mm/shmem.c b/mm/shmem.c index e87f5d6799a7..d061f8b34d49 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -1153,6 +1153,12 @@ static int shmem_getattr(struct mnt_idmap *idmap, { struct inode *inode = path->dentry->d_inode; struct shmem_inode_info *info = SHMEM_I(inode);
bool inode_locked = NULL;
if (!inode_is_locked(inode)) {
inode_lock_shared(inode);inode_locked = true;}
if (info->alloced - info->swapped != inode->i_mapping->nrpages) shmem_recalc_inode(inode, 0, 0);
@@ -1166,9 +1172,7 @@ static int shmem_getattr(struct mnt_idmap *idmap, stat->attributes_mask |= (STATX_ATTR_APPEND | STATX_ATTR_IMMUTABLE | STATX_ATTR_NODUMP);
inode_lock_shared(inode); generic_fillattr(idmap, request_mask, inode, stat);
inode_unlock_shared(inode);
if (shmem_huge_global_enabled(inode, 0, 0, false, NULL, 0)) stat->blksize = HPAGE_PMD_SIZE;
@@ -1179,6 +1183,9 @@ static int shmem_getattr(struct mnt_idmap *idmap, stat->btime.tv_nsec = info->i_crtime.tv_nsec; }
- if (inode_locked)
inode_unlock_shared(inode);- return 0;
}
--
What do you think?
As I said before, I've failed to find any file system getattr method that explicitly takes the inode's semaphore around a generic_fillattr() call. My understanding is that these methods assume that their caller handles appropriate serialization. Therefore, taking the inode semaphore at all in shmem_getattr() seems to me to be the wrong approach entirely.
The point of reverting immediately is that any fix can't possibly get the review and testing it deserves three days before v6.12 becomes final. Also, it's not clear what the rush to fix the KCSAN splat is; according to the Fixes: tag, this issue has been present for years without causing overt problems. It doesn't feel like this change belongs in an -rc in the first place.
Please revert d949d1d14fa2, then let's discuss a proper fix in a separate thread. Thanks!
Regards,
Jeongjun Park
Because this commit addresses only a KCSAN splat that has been present since v4.3, and does not address a reported behavioral issue, I respectfully request that this commit be reverted immediately so that it does not appear in v6.12 final. Troubleshooting and testing should continue until a fix to the KCSAN issue can be found that does not deadlock NFS exports of tmpfs.
-- Chuck Lever
-- Chuck Lever
-- Chuck Lever