nfsd: always hold the fi_lock when bumping fi_access refcounts
authorJeff Layton <jlayton@primarydata.com>
Thu, 10 Jul 2014 18:07:33 +0000 (14:07 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Fri, 11 Jul 2014 15:06:17 +0000 (11:06 -0400)
Once we remove the client_mutex, there's an unlikely but possible race
that could occur. It will be possible for nfs4_file_put_access to race
with nfs4_file_get_access. The refcount will go to zero (briefly) and
then bumped back to one. If that happens we set ourselves up for a
use-after-free and the potential for a lock to race onto the i_flock
list as a filp is being torn down.

Ensure that we can safely bump the refcount on the file by holding the
fi_lock whenever that's done. The only place it currently isn't is in
get_lock_access.

In order to ensure atomicity with finding the file, use the
find_*_file_locked variants and then call get_lock_access to get new
access references on the nfs4_file under the same lock.

Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4state.c

index 5f7294712ad4761db5990120388782c4f8ab12d4..8f320f2f8b846623eb709e3a6cf088907c96eb3d 100644 (file)
@@ -377,6 +377,8 @@ static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 static void
 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 {
+       lockdep_assert_held(&fp->fi_lock);
+
        if (access & NFS4_SHARE_ACCESS_WRITE)
                atomic_inc(&fp->fi_access[O_WRONLY]);
        if (access & NFS4_SHARE_ACCESS_READ)
@@ -386,6 +388,8 @@ __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 static __be32
 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 {
+       lockdep_assert_held(&fp->fi_lock);
+
        /* Does this access mode make sense? */
        if (access & ~NFS4_SHARE_ACCESS_BOTH)
                return nfserr_inval;
@@ -4572,6 +4576,8 @@ static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
 {
        struct nfs4_file *fp = lock_stp->st_file;
 
+       lockdep_assert_held(&fp->fi_lock);
+
        if (test_access(access, lock_stp))
                return;
        __nfs4_file_get_access(fp, access);
@@ -4623,6 +4629,7 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        struct nfs4_openowner *open_sop = NULL;
        struct nfs4_lockowner *lock_sop = NULL;
        struct nfs4_ol_stateid *lock_stp;
+       struct nfs4_file *fp;
        struct file *filp = NULL;
        struct file_lock *file_lock = NULL;
        struct file_lock *conflock = NULL;
@@ -4703,20 +4710,25 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                goto out;
        }
 
+       fp = lock_stp->st_file;
        locks_init_lock(file_lock);
        switch (lock->lk_type) {
                case NFS4_READ_LT:
                case NFS4_READW_LT:
-                       filp = find_readable_file(lock_stp->st_file);
+                       spin_lock(&fp->fi_lock);
+                       filp = find_readable_file_locked(fp);
                        if (filp)
                                get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
+                       spin_unlock(&fp->fi_lock);
                        file_lock->fl_type = F_RDLCK;
                        break;
                case NFS4_WRITE_LT:
                case NFS4_WRITEW_LT:
-                       filp = find_writeable_file(lock_stp->st_file);
+                       spin_lock(&fp->fi_lock);
+                       filp = find_writeable_file_locked(fp);
                        if (filp)
                                get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
+                       spin_unlock(&fp->fi_lock);
                        file_lock->fl_type = F_WRLCK;
                        break;
                default: