[PATCH] NFS: pin the superblock across migration recovery

[PATCH] NFS: pin the superblock across migration recovery

[Cheng Yongkang]

The NFSv4 state manager walks clp->cl_superblocks under RCU in
nfs4_handle_migration() and nfs4_handle_lease_moved(), then drops the
lock and calls nfs4_try_migration() without holding a reference on the
nfs_server or its superblock.  A concurrent umount can therefore run
nfs_kill_super() -> nfs_free_server() while migration recovery is still
inside nfs4_update_server(): the server is removed from cl_superblocks a
second time (DEBUG_LIST BUG), and once freed by call_rcu() the migration
path still writes server->mig_status and touches server->super->s_root.

Pin the superblock with nfs_sb_active() before dropping RCU, as
nfs_client_for_each_server() already does, and drop it again with
nfs_sb_deactive() on every path: umount then detaches the superblock but
cannot reach nfs_kill_super()/nfs_free_server() until migration recovery
has finished and restored the server's list membership.
nfs4_handle_migration() releases the pin after nfs4_try_migration();
nfs4_handle_lease_moved() releases it on both restart paths and before
its final return.

Umount may now wait for the migration attempt to finish, bounded by the
mount's RPC timeouts, the same trade-off as the other nfs_sb_active()
users.

Assisted-by: LLM
Signed-off-by: Cheng Yongkang <teel4res@gmail.com>
---
 fs/nfs/nfs4state.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index a5dec04..ff866f6 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2169,8 +2169,12 @@ restart:
 						&server->mig_status))
 			continue;

+		if (!server->super || !nfs_sb_active(server->super))
+			continue;
+
 		rcu_read_unlock();
 		status = nfs4_try_migration(server, cred);
+		nfs_sb_deactive(server->super);
 		if (status < 0) {
 			put_cred(cred);
 			return status;
@@ -2212,13 +2216,20 @@ restart:
 			continue;
 		server->mig_gen = clp->cl_mig_gen;

+		if (!server->super || !nfs_sb_active(server->super))
+			continue;
+
 		rcu_read_unlock();

 		inode = d_inode(server->super->s_root);
 		status = nfs4_proc_fsid_present(inode, cred);
-		if (status != -NFS4ERR_MOVED)
+		if (status != -NFS4ERR_MOVED) {
+			nfs_sb_deactive(server->super);
 			goto restart;	/* wasn't this one */
-		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
+		}
+		status = nfs4_try_migration(server, cred);
+		nfs_sb_deactive(server->super);
+		if (status == -NFS4ERR_LEASE_MOVED)
 			goto restart;	/* there are more */
 		goto out;
 	}