nfsd: have nfsd4_lock use blocking locks for v4.1+ locks
authorJeff Layton <jlayton@redhat.com>
Fri, 16 Sep 2016 20:28:24 +0000 (16:28 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Mon, 26 Sep 2016 19:20:36 +0000 (15:20 -0400)
Create a new per-lockowner+per-inode structure that contains a
file_lock. Have nfsd4_lock add this structure to the lockowner's list
prior to setting the lock. Then call the vfs and request a blocking lock
(by setting FL_SLEEP). If we get anything besides FILE_LOCK_DEFERRED
back, then we dequeue the block structure and free it. When the next
lock request comes in, we'll look for an existing block for the same
filehandle and dequeue and reuse it if there is one.

When the lock comes free (a'la an lm_notify call), we dequeue it
from the lockowner's list and kick off a CB_NOTIFY_LOCK callback to
inform the client that it should retry the lock request.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4state.c
fs/nfsd/state.h

index a204d7e109d4d63a76d01a31198b00f3f1cd09be..ca0db4974e5b194aa9454f19128788c56345212f 100644 (file)
@@ -99,6 +99,7 @@ static struct kmem_cache *odstate_slab;
 static void free_session(struct nfsd4_session *);
 
 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
+static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 
 static bool is_session_dead(struct nfsd4_session *ses)
 {
@@ -210,6 +211,84 @@ static void nfsd4_put_session(struct nfsd4_session *ses)
        spin_unlock(&nn->client_lock);
 }
 
+static struct nfsd4_blocked_lock *
+find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
+                       struct nfsd_net *nn)
+{
+       struct nfsd4_blocked_lock *cur, *found = NULL;
+
+       spin_lock(&nn->client_lock);
+       list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
+               if (fh_match(fh, &cur->nbl_fh)) {
+                       list_del_init(&cur->nbl_list);
+                       found = cur;
+                       break;
+               }
+       }
+       spin_unlock(&nn->client_lock);
+       if (found)
+               posix_unblock_lock(&found->nbl_lock);
+       return found;
+}
+
+static struct nfsd4_blocked_lock *
+find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
+                       struct nfsd_net *nn)
+{
+       struct nfsd4_blocked_lock *nbl;
+
+       nbl = find_blocked_lock(lo, fh, nn);
+       if (!nbl) {
+               nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
+               if (nbl) {
+                       fh_copy_shallow(&nbl->nbl_fh, fh);
+                       locks_init_lock(&nbl->nbl_lock);
+                       nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
+                                       &nfsd4_cb_notify_lock_ops,
+                                       NFSPROC4_CLNT_CB_NOTIFY_LOCK);
+               }
+       }
+       return nbl;
+}
+
+static void
+free_blocked_lock(struct nfsd4_blocked_lock *nbl)
+{
+       locks_release_private(&nbl->nbl_lock);
+       kfree(nbl);
+}
+
+static int
+nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
+{
+       /*
+        * Since this is just an optimization, we don't try very hard if it
+        * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
+        * just quit trying on anything else.
+        */
+       switch (task->tk_status) {
+       case -NFS4ERR_DELAY:
+               rpc_delay(task, 1 * HZ);
+               return 0;
+       default:
+               return 1;
+       }
+}
+
+static void
+nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
+{
+       struct nfsd4_blocked_lock       *nbl = container_of(cb,
+                                               struct nfsd4_blocked_lock, nbl_cb);
+
+       free_blocked_lock(nbl);
+}
+
+static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
+       .done           = nfsd4_cb_notify_lock_done,
+       .release        = nfsd4_cb_notify_lock_release,
+};
+
 static inline struct nfs4_stateowner *
 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 {
@@ -5309,7 +5388,29 @@ nfsd4_fl_put_owner(fl_owner_t owner)
                nfs4_put_stateowner(&lo->lo_owner);
 }
 
+static void
+nfsd4_lm_notify(struct file_lock *fl)
+{
+       struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
+       struct net                      *net = lo->lo_owner.so_client->net;
+       struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
+       struct nfsd4_blocked_lock       *nbl = container_of(fl,
+                                               struct nfsd4_blocked_lock, nbl_lock);
+       bool queue = false;
+
+       spin_lock(&nn->client_lock);
+       if (!list_empty(&nbl->nbl_list)) {
+               list_del_init(&nbl->nbl_list);
+               queue = true;
+       }
+       spin_unlock(&nn->client_lock);
+
+       if (queue)
+               nfsd4_run_cb(&nbl->nbl_cb);
+}
+
 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
+       .lm_notify = nfsd4_lm_notify,
        .lm_get_owner = nfsd4_fl_get_owner,
        .lm_put_owner = nfsd4_fl_put_owner,
 };
@@ -5407,6 +5508,7 @@ alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
        lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
        if (!lo)
                return NULL;
+       INIT_LIST_HEAD(&lo->lo_blocked);
        INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
        lo->lo_owner.so_is_open_owner = 0;
        lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
@@ -5588,12 +5690,15 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        struct nfs4_ol_stateid *open_stp = NULL;
        struct nfs4_file *fp;
        struct file *filp = NULL;
+       struct nfsd4_blocked_lock *nbl = NULL;
        struct file_lock *file_lock = NULL;
        struct file_lock *conflock = NULL;
        __be32 status = 0;
        int lkflg;
        int err;
        bool new = false;
+       unsigned char fl_type;
+       unsigned int fl_flags = FL_POSIX;
        struct net *net = SVC_NET(rqstp);
        struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 
@@ -5658,46 +5763,55 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        if (!locks_in_grace(net) && lock->lk_reclaim)
                goto out;
 
-       file_lock = locks_alloc_lock();
-       if (!file_lock) {
-               dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
-               status = nfserr_jukebox;
-               goto out;
-       }
-
        fp = lock_stp->st_stid.sc_file;
        switch (lock->lk_type) {
-               case NFS4_READ_LT:
                case NFS4_READW_LT:
+                       if (nfsd4_has_session(cstate))
+                               fl_flags |= FL_SLEEP;
+                       /* Fallthrough */
+               case NFS4_READ_LT:
                        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;
+                       fl_type = F_RDLCK;
                        break;
-               case NFS4_WRITE_LT:
                case NFS4_WRITEW_LT:
+                       if (nfsd4_has_session(cstate))
+                               fl_flags |= FL_SLEEP;
+                       /* Fallthrough */
+               case NFS4_WRITE_LT:
                        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;
+                       fl_type = F_WRLCK;
                        break;
                default:
                        status = nfserr_inval;
                goto out;
        }
+
        if (!filp) {
                status = nfserr_openmode;
                goto out;
        }
 
+       nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
+       if (!nbl) {
+               dprintk("NFSD: %s: unable to allocate block!\n", __func__);
+               status = nfserr_jukebox;
+               goto out;
+       }
+
+       file_lock = &nbl->nbl_lock;
+       file_lock->fl_type = fl_type;
        file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
        file_lock->fl_pid = current->tgid;
        file_lock->fl_file = filp;
-       file_lock->fl_flags = FL_POSIX;
+       file_lock->fl_flags = fl_flags;
        file_lock->fl_lmops = &nfsd_posix_mng_ops;
        file_lock->fl_start = lock->lk_offset;
        file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
@@ -5710,18 +5824,27 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                goto out;
        }
 
+       if (fl_flags & FL_SLEEP) {
+               spin_lock(&nn->client_lock);
+               list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
+               spin_unlock(&nn->client_lock);
+       }
+
        err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
-       switch (-err) {
+       switch (err) {
        case 0: /* success! */
                nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
                status = 0;
                break;
-       case (EAGAIN):          /* conflock holds conflicting lock */
+       case FILE_LOCK_DEFERRED:
+               nbl = NULL;
+               /* Fallthrough */
+       case -EAGAIN:           /* conflock holds conflicting lock */
                status = nfserr_denied;
                dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
                nfs4_set_lock_denied(conflock, &lock->lk_denied);
                break;
-       case (EDEADLK):
+       case -EDEADLK:
                status = nfserr_deadlock;
                break;
        default:
@@ -5730,6 +5853,15 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                break;
        }
 out:
+       if (nbl) {
+               /* dequeue it if we queued it before */
+               if (fl_flags & FL_SLEEP) {
+                       spin_lock(&nn->client_lock);
+                       list_del_init(&nbl->nbl_list);
+                       spin_unlock(&nn->client_lock);
+               }
+               free_blocked_lock(nbl);
+       }
        if (filp)
                fput(filp);
        if (lock_stp) {
@@ -5753,8 +5885,6 @@ out:
        if (open_stp)
                nfs4_put_stid(&open_stp->st_stid);
        nfsd4_bump_seqid(cstate, status);
-       if (file_lock)
-               locks_free_lock(file_lock);
        if (conflock)
                locks_free_lock(conflock);
        return status;
index 88d029dd13aaefe03763efd79328525b9e4ff239..e45c183a8bf79a25a4ff37781587b49213af7156 100644 (file)
@@ -440,11 +440,11 @@ struct nfs4_openowner {
 /*
  * Represents a generic "lockowner". Similar to an openowner. References to it
  * are held by the lock stateids that are created on its behalf. This object is
- * a superset of the nfs4_stateowner struct (or would be if it needed any extra
- * fields).
+ * a superset of the nfs4_stateowner struct.
  */
 struct nfs4_lockowner {
-       struct nfs4_stateowner  lo_owner; /* must be first element */
+       struct nfs4_stateowner  lo_owner;       /* must be first element */
+       struct list_head        lo_blocked;     /* blocked file_locks */
 };
 
 static inline struct nfs4_openowner * openowner(struct nfs4_stateowner *so)
@@ -580,7 +580,13 @@ static inline bool nfsd4_stateid_generation_after(stateid_t *a, stateid_t *b)
        return (s32)(a->si_generation - b->si_generation) > 0;
 }
 
+/*
+ * When a client tries to get a lock on a file, we set one of these objects
+ * on the blocking lock. When the lock becomes free, we can then issue a
+ * CB_NOTIFY_LOCK to the server.
+ */
 struct nfsd4_blocked_lock {
+       struct list_head        nbl_list;
        struct file_lock        nbl_lock;
        struct knfsd_fh         nbl_fh;
        struct nfsd4_callback   nbl_cb;