NFSv4: Add functions to order RPC calls
[pandora-kernel.git] / fs / nfs / nfs4state.c
index afe587d..f535c21 100644 (file)
@@ -264,13 +264,16 @@ nfs4_alloc_state_owner(void)
 {
        struct nfs4_state_owner *sp;
 
-       sp = kmalloc(sizeof(*sp),GFP_KERNEL);
+       sp = kzalloc(sizeof(*sp),GFP_KERNEL);
        if (!sp)
                return NULL;
        init_MUTEX(&sp->so_sema);
-       sp->so_seqid = 0;                 /* arbitrary */
        INIT_LIST_HEAD(&sp->so_states);
        INIT_LIST_HEAD(&sp->so_delegations);
+       rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
+       sp->so_seqid.sequence = &sp->so_sequence;
+       spin_lock_init(&sp->so_sequence.lock);
+       INIT_LIST_HEAD(&sp->so_sequence.list);
        atomic_set(&sp->so_count, 1);
        return sp;
 }
@@ -553,12 +556,10 @@ static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, f
        struct nfs4_lock_state *lsp;
        struct nfs4_client *clp = state->owner->so_client;
 
-       lsp = kmalloc(sizeof(*lsp), GFP_KERNEL);
+       lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
        if (lsp == NULL)
                return NULL;
-       lsp->ls_flags = 0;
-       lsp->ls_seqid = 0;      /* arbitrary */
-       memset(lsp->ls_stateid.data, 0, sizeof(lsp->ls_stateid.data));
+       lsp->ls_seqid.sequence = &state->owner->so_sequence;
        atomic_set(&lsp->ls_count, 1);
        lsp->ls_owner = fl_owner;
        spin_lock(&clp->cl_lock);
@@ -673,29 +674,102 @@ void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t f
        nfs4_put_lock_state(lsp);
 }
 
-/*
-* Called with state->lock_sema and clp->cl_sem held.
-*/
-void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *lsp)
+struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
+{
+       struct rpc_sequence *sequence = counter->sequence;
+       struct nfs_seqid *new;
+
+       new = kmalloc(sizeof(*new), GFP_KERNEL);
+       if (new != NULL) {
+               new->sequence = counter;
+               new->task = NULL;
+               spin_lock(&sequence->lock);
+               list_add_tail(&new->list, &sequence->list);
+               spin_unlock(&sequence->lock);
+       }
+       return new;
+}
+
+void nfs_free_seqid(struct nfs_seqid *seqid)
 {
-       if (status == NFS_OK || seqid_mutating_err(-status))
-               lsp->ls_seqid++;
+       struct rpc_sequence *sequence = seqid->sequence->sequence;
+       struct rpc_task *next = NULL;
+
+       spin_lock(&sequence->lock);
+       list_del(&seqid->list);
+       if (!list_empty(&sequence->list)) {
+               next = list_entry(sequence->list.next, struct nfs_seqid, list)->task;
+               if (next)
+                       rpc_wake_up_task(next);
+       }
+       spin_unlock(&sequence->lock);
+       kfree(seqid);
 }
 
 /*
-* Called with sp->so_sema and clp->cl_sem held.
-*
-* Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
-* failed with a seqid incrementing error -
-* see comments nfs_fs.h:seqid_mutating_error()
-*/
-void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp)
-{
-       if (status == NFS_OK || seqid_mutating_err(-status))
-               sp->so_seqid++;
-       /* If the server returns BAD_SEQID, unhash state_owner here */
-       if (status == -NFS4ERR_BAD_SEQID)
+ * Called with sp->so_sema and clp->cl_sem held.
+ *
+ * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
+ * failed with a seqid incrementing error -
+ * see comments nfs_fs.h:seqid_mutating_error()
+ */
+static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
+{
+       switch (status) {
+               case 0:
+                       break;
+               case -NFS4ERR_BAD_SEQID:
+               case -NFS4ERR_STALE_CLIENTID:
+               case -NFS4ERR_STALE_STATEID:
+               case -NFS4ERR_BAD_STATEID:
+               case -NFS4ERR_BADXDR:
+               case -NFS4ERR_RESOURCE:
+               case -NFS4ERR_NOFILEHANDLE:
+                       /* Non-seqid mutating errors */
+                       return;
+       };
+       /*
+        * Note: no locking needed as we are guaranteed to be first
+        * on the sequence list
+        */
+       seqid->sequence->counter++;
+}
+
+void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
+{
+       if (status == -NFS4ERR_BAD_SEQID) {
+               struct nfs4_state_owner *sp = container_of(seqid->sequence,
+                               struct nfs4_state_owner, so_seqid);
                nfs4_drop_state_owner(sp);
+       }
+       return nfs_increment_seqid(status, seqid);
+}
+
+/*
+ * Called with ls->lock_sema and clp->cl_sem held.
+ *
+ * Increment the seqid if the LOCK/LOCKU succeeded, or
+ * failed with a seqid incrementing error -
+ * see comments nfs_fs.h:seqid_mutating_error()
+ */
+void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
+{
+       return nfs_increment_seqid(status, seqid);
+}
+
+int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
+{
+       struct rpc_sequence *sequence = seqid->sequence->sequence;
+       int status = 0;
+
+       spin_lock(&sequence->lock);
+       if (sequence->list.next != &seqid->list) {
+               seqid->task = task;
+               rpc_sleep_on(&sequence->wait, task, NULL, NULL);
+               status = -EAGAIN;
+       }
+       spin_unlock(&sequence->lock);
+       return status;
 }
 
 static int reclaimer(void *);
@@ -791,8 +865,6 @@ static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct n
                if (state->state == 0)
                        continue;
                status = ops->recover_open(sp, state);
-               list_for_each_entry(lock, &state->lock_states, ls_locks)
-                       lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
                if (status >= 0) {
                        status = nfs4_reclaim_locks(ops, state);
                        if (status < 0)
@@ -831,6 +903,26 @@ out_err:
        return status;
 }
 
+static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
+{
+       struct nfs4_state_owner *sp;
+       struct nfs4_state *state;
+       struct nfs4_lock_state *lock;
+
+       /* Reset all sequence ids to zero */
+       list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
+               sp->so_seqid.counter = 0;
+               sp->so_seqid.flags = 0;
+               list_for_each_entry(state, &sp->so_states, open_states) {
+                       list_for_each_entry(lock, &state->lock_states, ls_locks) {
+                               lock->ls_seqid.counter = 0;
+                               lock->ls_seqid.flags = 0;
+                               lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
+                       }
+               }
+       }
+}
+
 static int reclaimer(void *ptr)
 {
        struct reclaimer_args *args = (struct reclaimer_args *)ptr;
@@ -864,6 +956,7 @@ restart_loop:
                default:
                        ops = &nfs4_network_partition_recovery_ops;
        };
+       nfs4_state_mark_reclaim(clp);
        status = __nfs4_init_client(clp);
        if (status)
                goto out_error;