Merge branch 'for-2.6.38' of git://linux-nfs.org/~bfields/linux
[pandora-kernel.git] / fs / nfsd / nfs4callback.c
index c3c6a90..3be975e 100644 (file)
@@ -499,34 +499,28 @@ out_default:
 /*
  * NB: Without this zero space reservation, callbacks over krb5p fail
  */
-static int nfs4_xdr_enc_cb_null(struct rpc_rqst *req, __be32 *p, void *__unused)
+static void nfs4_xdr_enc_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
+                                void *__unused)
 {
-       struct xdr_stream xdrs, *xdr = &xdrs;
-
-       xdr_init_encode(&xdrs, &req->rq_snd_buf, p);
        xdr_reserve_space(xdr, 0);
-       return 0;
 }
 
 /*
  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
  */
-static int nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, __be32 *p,
-                                 const struct nfsd4_callback *cb)
+static void nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, struct xdr_stream *xdr,
+                                  const struct nfsd4_callback *cb)
 {
-       struct xdr_stream xdr;
        const struct nfs4_delegation *args = cb->cb_op;
        struct nfs4_cb_compound_hdr hdr = {
                .ident = cb->cb_clp->cl_cb_ident,
                .minorversion = cb->cb_minorversion,
        };
 
-       xdr_init_encode(&xdr, &req->rq_snd_buf, p);
-       encode_cb_compound4args(&xdr, &hdr);
-       encode_cb_sequence4args(&xdr, cb, &hdr);
-       encode_cb_recall4args(&xdr, args, &hdr);
+       encode_cb_compound4args(xdr, &hdr);
+       encode_cb_sequence4args(xdr, cb, &hdr);
+       encode_cb_recall4args(xdr, args, &hdr);
        encode_cb_nops(&hdr);
-       return 0;
 }
 
 
@@ -539,7 +533,8 @@ static int nfs4_xdr_enc_cb_recall(struct rpc_rqst *req, __be32 *p,
  * Protocol".
  */
 
-static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, __be32 *p, void *__unused)
+static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, struct xdr_stream *xdr,
+                               void *__unused)
 {
        return 0;
 }
@@ -547,26 +542,25 @@ static int nfs4_xdr_dec_cb_null(struct rpc_rqst *req, __be32 *p, void *__unused)
 /*
  * 20.2. Operation 4: CB_RECALL - Recall a Delegation
  */
-static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp, __be32 *p,
+static int nfs4_xdr_dec_cb_recall(struct rpc_rqst *rqstp,
+                                 struct xdr_stream *xdr,
                                  struct nfsd4_callback *cb)
 {
-       struct xdr_stream xdr;
        struct nfs4_cb_compound_hdr hdr;
        enum nfsstat4 nfserr;
        int status;
 
-       xdr_init_decode(&xdr, &rqstp->rq_rcv_buf, p);
-       status = decode_cb_compound4res(&xdr, &hdr);
+       status = decode_cb_compound4res(xdr, &hdr);
        if (unlikely(status))
                goto out;
 
        if (cb != NULL) {
-               status = decode_cb_sequence4res(&xdr, cb);
+               status = decode_cb_sequence4res(xdr, cb);
                if (unlikely(status))
                        goto out;
        }
 
-       status = decode_cb_op_status(&xdr, OP_CB_RECALL, &nfserr);
+       status = decode_cb_op_status(xdr, OP_CB_RECALL, &nfserr);
        if (unlikely(status))
                goto out;
        if (unlikely(nfserr != NFS4_OK))
@@ -580,23 +574,23 @@ out_default:
 /*
  * RPC procedure tables
  */
-#define PROC(proc, call, argtype, restype)                              \
-[NFSPROC4_CLNT_##proc] = {                                             \
-        .p_proc   = NFSPROC4_CB_##call,                                        \
-        .p_encode = (kxdrproc_t) nfs4_xdr_##argtype,                    \
-        .p_decode = (kxdrproc_t) nfs4_xdr_##restype,                    \
-        .p_arglen = NFS4_##argtype##_sz,                                \
-        .p_replen = NFS4_##restype##_sz,                                \
-        .p_statidx = NFSPROC4_CB_##call,                               \
-       .p_name   = #proc,                                              \
-}
-
-static struct rpc_procinfo     nfs4_cb_procedures[] = {
-    PROC(CB_NULL,      NULL,     enc_cb_null,     dec_cb_null),
-    PROC(CB_RECALL,    COMPOUND,   enc_cb_recall,      dec_cb_recall),
+#define PROC(proc, call, argtype, restype)                             \
+[NFSPROC4_CLNT_##proc] = {                                             \
+       .p_proc    = NFSPROC4_CB_##call,                                \
+       .p_encode  = (kxdreproc_t)nfs4_xdr_enc_##argtype,               \
+       .p_decode  = (kxdrdproc_t)nfs4_xdr_dec_##restype,               \
+       .p_arglen  = NFS4_enc_##argtype##_sz,                           \
+       .p_replen  = NFS4_dec_##restype##_sz,                           \
+       .p_statidx = NFSPROC4_CB_##call,                                \
+       .p_name    = #proc,                                             \
+}
+
+static struct rpc_procinfo nfs4_cb_procedures[] = {
+       PROC(CB_NULL,   NULL,           cb_null,        cb_null),
+       PROC(CB_RECALL, COMPOUND,       cb_recall,      cb_recall),
 };
 
-static struct rpc_version       nfs_cb_version4 = {
+static struct rpc_version nfs_cb_version4 = {
 /*
  * Note on the callback rpc program version number: despite language in rfc
  * 5661 section 18.36.3 requiring servers to use 4 in this field, the
@@ -604,29 +598,29 @@ static struct rpc_version       nfs_cb_version4 = {
  * in practice that appears to be what implementations use.  The section
  * 18.36.3 language is expected to be fixed in an erratum.
  */
-        .number                 = 1,
-        .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
-        .procs                  = nfs4_cb_procedures
+       .number                 = 1,
+       .nrprocs                = ARRAY_SIZE(nfs4_cb_procedures),
+       .procs                  = nfs4_cb_procedures
 };
 
-static struct rpc_version *    nfs_cb_version[] = {
+static struct rpc_version *nfs_cb_version[] = {
        &nfs_cb_version4,
 };
 
 static struct rpc_program cb_program;
 
 static struct rpc_stat cb_stats = {
-               .program        = &cb_program
+       .program                = &cb_program
 };
 
 #define NFS4_CALLBACK 0x40000000
 static struct rpc_program cb_program = {
-               .name           = "nfs4_cb",
-               .number         = NFS4_CALLBACK,
-               .nrvers         = ARRAY_SIZE(nfs_cb_version),
-               .version        = nfs_cb_version,
-               .stats          = &cb_stats,
-               .pipe_dir_name  = "/nfsd4_cb",
+       .name                   = "nfs4_cb",
+       .number                 = NFS4_CALLBACK,
+       .nrvers                 = ARRAY_SIZE(nfs_cb_version),
+       .version                = nfs_cb_version,
+       .stats                  = &cb_stats,
+       .pipe_dir_name          = "/nfsd4_cb",
 };
 
 static int max_cb_time(void)
@@ -634,10 +628,8 @@ static int max_cb_time(void)
        return max(nfsd4_lease/10, (time_t)1) * HZ;
 }
 
-/* Reference counting, callback cleanup, etc., all look racy as heck.
- * And why is cl_cb_set an atomic? */
 
-int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
+static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn, struct nfsd4_session *ses)
 {
        struct rpc_timeout      timeparms = {
                .to_initval     = max_cb_time(),
@@ -647,6 +639,7 @@ int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
                .net            = &init_net,
                .address        = (struct sockaddr *) &conn->cb_addr,
                .addrsize       = conn->cb_addrlen,
+               .saddress       = (struct sockaddr *) &conn->cb_saddr,
                .timeout        = &timeparms,
                .program        = &cb_program,
                .version        = 0,
@@ -663,6 +656,10 @@ int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
                args.protocol = XPRT_TRANSPORT_TCP;
                clp->cl_cb_ident = conn->cb_ident;
        } else {
+               if (!conn->cb_xprt)
+                       return -EINVAL;
+               clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
+               clp->cl_cb_session = ses;
                args.bc_xprt = conn->cb_xprt;
                args.prognumber = clp->cl_cb_session->se_cb_prog;
                args.protocol = XPRT_TRANSPORT_BC_TCP;
@@ -685,14 +682,20 @@ static void warn_no_callback_path(struct nfs4_client *clp, int reason)
                (int)clp->cl_name.len, clp->cl_name.data, reason);
 }
 
+static void nfsd4_mark_cb_down(struct nfs4_client *clp, int reason)
+{
+       clp->cl_cb_state = NFSD4_CB_DOWN;
+       warn_no_callback_path(clp, reason);
+}
+
 static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
 {
        struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
 
        if (task->tk_status)
-               warn_no_callback_path(clp, task->tk_status);
+               nfsd4_mark_cb_down(clp, task->tk_status);
        else
-               atomic_set(&clp->cl_cb_set, 1);
+               clp->cl_cb_state = NFSD4_CB_UP;
 }
 
 static const struct rpc_call_ops nfsd4_cb_probe_ops = {
@@ -715,6 +718,11 @@ int set_callback_cred(void)
 
 static struct workqueue_struct *callback_wq;
 
+static void run_nfsd4_cb(struct nfsd4_callback *cb)
+{
+       queue_work(callback_wq, &cb->cb_work);
+}
+
 static void do_probe_callback(struct nfs4_client *clp)
 {
        struct nfsd4_callback *cb = &clp->cl_cb_null;
@@ -729,7 +737,7 @@ static void do_probe_callback(struct nfs4_client *clp)
 
        cb->cb_ops = &nfsd4_cb_probe_ops;
 
-       queue_work(callback_wq, &cb->cb_work);
+       run_nfsd4_cb(cb);
 }
 
 /*
@@ -738,14 +746,21 @@ static void do_probe_callback(struct nfs4_client *clp)
  */
 void nfsd4_probe_callback(struct nfs4_client *clp)
 {
+       /* XXX: atomicity?  Also, should we be using cl_cb_flags? */
+       clp->cl_cb_state = NFSD4_CB_UNKNOWN;
        set_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_cb_flags);
        do_probe_callback(clp);
 }
 
-void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
+void nfsd4_probe_callback_sync(struct nfs4_client *clp)
 {
-       BUG_ON(atomic_read(&clp->cl_cb_set));
+       nfsd4_probe_callback(clp);
+       flush_workqueue(callback_wq);
+}
 
+void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
+{
+       clp->cl_cb_state = NFSD4_CB_UNKNOWN;
        spin_lock(&clp->cl_lock);
        memcpy(&clp->cl_cb_conn, conn, sizeof(struct nfs4_cb_conn));
        spin_unlock(&clp->cl_lock);
@@ -756,24 +771,14 @@ void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
  * If the slot is available, then mark it busy.  Otherwise, set the
  * thread for sleeping on the callback RPC wait queue.
  */
-static int nfsd41_cb_setup_sequence(struct nfs4_client *clp,
-               struct rpc_task *task)
+static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
 {
-       u32 *ptr = (u32 *)clp->cl_cb_session->se_sessionid.data;
-       int status = 0;
-
-       dprintk("%s: %u:%u:%u:%u\n", __func__,
-               ptr[0], ptr[1], ptr[2], ptr[3]);
-
        if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
                rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
                dprintk("%s slot is busy\n", __func__);
-               status = -EAGAIN;
-               goto out;
+               return false;
        }
-out:
-       dprintk("%s status=%d\n", __func__, status);
-       return status;
+       return true;
 }
 
 /*
@@ -786,20 +791,19 @@ static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
        struct nfs4_delegation *dp = container_of(cb, struct nfs4_delegation, dl_recall);
        struct nfs4_client *clp = dp->dl_client;
        u32 minorversion = clp->cl_minorversion;
-       int status = 0;
 
        cb->cb_minorversion = minorversion;
        if (minorversion) {
-               status = nfsd41_cb_setup_sequence(clp, task);
-               if (status) {
-                       if (status != -EAGAIN) {
-                               /* terminate rpc task */
-                               task->tk_status = status;
-                               task->tk_action = NULL;
-                       }
+               if (!nfsd41_cb_get_slot(clp, task))
                        return;
-               }
        }
+       spin_lock(&clp->cl_lock);
+       if (list_empty(&cb->cb_per_client)) {
+               /* This is the first call, not a restart */
+               cb->cb_done = false;
+               list_add(&cb->cb_per_client, &clp->cl_callbacks);
+       }
+       spin_unlock(&clp->cl_lock);
        rpc_call_start(task);
 }
 
@@ -835,15 +839,18 @@ static void nfsd4_cb_recall_done(struct rpc_task *task, void *calldata)
 
        nfsd4_cb_done(task, calldata);
 
-       if (current_rpc_client == NULL) {
-               /* We're shutting down; give up. */
-               /* XXX: err, or is it ok just to fall through
-                * and rpc_restart_call? */
+       if (current_rpc_client != task->tk_client) {
+               /* We're shutting down or changing cl_cb_client; leave
+                * it to nfsd4_process_cb_update to restart the call if
+                * necessary. */
                return;
        }
 
+       if (cb->cb_done)
+               return;
        switch (task->tk_status) {
        case 0:
+               cb->cb_done = true;
                return;
        case -EBADHANDLE:
        case -NFS4ERR_BAD_STATEID:
@@ -852,32 +859,30 @@ static void nfsd4_cb_recall_done(struct rpc_task *task, void *calldata)
                break;
        default:
                /* Network partition? */
-               atomic_set(&clp->cl_cb_set, 0);
-               warn_no_callback_path(clp, task->tk_status);
-               if (current_rpc_client != task->tk_client) {
-                       /* queue a callback on the new connection: */
-                       atomic_inc(&dp->dl_count);
-                       nfsd4_cb_recall(dp);
-                       return;
-               }
+               nfsd4_mark_cb_down(clp, task->tk_status);
        }
        if (dp->dl_retries--) {
                rpc_delay(task, 2*HZ);
                task->tk_status = 0;
                rpc_restart_call_prepare(task);
                return;
-       } else {
-               atomic_set(&clp->cl_cb_set, 0);
-               warn_no_callback_path(clp, task->tk_status);
        }
+       nfsd4_mark_cb_down(clp, task->tk_status);
+       cb->cb_done = true;
 }
 
 static void nfsd4_cb_recall_release(void *calldata)
 {
        struct nfsd4_callback *cb = calldata;
+       struct nfs4_client *clp = cb->cb_clp;
        struct nfs4_delegation *dp = container_of(cb, struct nfs4_delegation, dl_recall);
 
-       nfs4_put_delegation(dp);
+       if (cb->cb_done) {
+               spin_lock(&clp->cl_lock);
+               list_del(&cb->cb_per_client);
+               spin_unlock(&clp->cl_lock);
+               nfs4_put_delegation(dp);
+       }
 }
 
 static const struct rpc_call_ops nfsd4_cb_recall_ops = {
@@ -912,16 +917,33 @@ void nfsd4_shutdown_callback(struct nfs4_client *clp)
        flush_workqueue(callback_wq);
 }
 
-void nfsd4_release_cb(struct nfsd4_callback *cb)
+static void nfsd4_release_cb(struct nfsd4_callback *cb)
 {
        if (cb->cb_ops->rpc_release)
                cb->cb_ops->rpc_release(cb);
 }
 
-void nfsd4_process_cb_update(struct nfsd4_callback *cb)
+/* requires cl_lock: */
+static struct nfsd4_conn * __nfsd4_find_backchannel(struct nfs4_client *clp)
+{
+       struct nfsd4_session *s;
+       struct nfsd4_conn *c;
+
+       list_for_each_entry(s, &clp->cl_sessions, se_perclnt) {
+               list_for_each_entry(c, &s->se_conns, cn_persession) {
+                       if (c->cn_flags & NFS4_CDFC4_BACK)
+                               return c;
+               }
+       }
+       return NULL;
+}
+
+static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 {
        struct nfs4_cb_conn conn;
        struct nfs4_client *clp = cb->cb_clp;
+       struct nfsd4_session *ses = NULL;
+       struct nfsd4_conn *c;
        int err;
 
        /*
@@ -932,6 +954,10 @@ void nfsd4_process_cb_update(struct nfsd4_callback *cb)
                rpc_shutdown_client(clp->cl_cb_client);
                clp->cl_cb_client = NULL;
        }
+       if (clp->cl_cb_conn.cb_xprt) {
+               svc_xprt_put(clp->cl_cb_conn.cb_xprt);
+               clp->cl_cb_conn.cb_xprt = NULL;
+       }
        if (test_bit(NFSD4_CLIENT_KILL, &clp->cl_cb_flags))
                return;
        spin_lock(&clp->cl_lock);
@@ -942,11 +968,22 @@ void nfsd4_process_cb_update(struct nfsd4_callback *cb)
        BUG_ON(!clp->cl_cb_flags);
        clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_cb_flags);
        memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
+       c = __nfsd4_find_backchannel(clp);
+       if (c) {
+               svc_xprt_get(c->cn_xprt);
+               conn.cb_xprt = c->cn_xprt;
+               ses = c->cn_session;
+       }
        spin_unlock(&clp->cl_lock);
 
-       err = setup_callback_client(clp, &conn);
-       if (err)
+       err = setup_callback_client(clp, &conn, ses);
+       if (err) {
                warn_no_callback_path(clp, err);
+               return;
+       }
+       /* Yay, the callback channel's back! Restart any callbacks: */
+       list_for_each_entry(cb, &clp->cl_callbacks, cb_per_client)
+               run_nfsd4_cb(cb);
 }
 
 void nfsd4_do_callback_rpc(struct work_struct *w)
@@ -971,10 +1008,11 @@ void nfsd4_do_callback_rpc(struct work_struct *w)
 void nfsd4_cb_recall(struct nfs4_delegation *dp)
 {
        struct nfsd4_callback *cb = &dp->dl_recall;
+       struct nfs4_client *clp = dp->dl_client;
 
        dp->dl_retries = 1;
        cb->cb_op = dp;
-       cb->cb_clp = dp->dl_client;
+       cb->cb_clp = clp;
        cb->cb_msg.rpc_proc = &nfs4_cb_procedures[NFSPROC4_CLNT_CB_RECALL];
        cb->cb_msg.rpc_argp = cb;
        cb->cb_msg.rpc_resp = cb;
@@ -983,5 +1021,8 @@ void nfsd4_cb_recall(struct nfs4_delegation *dp)
        cb->cb_ops = &nfsd4_cb_recall_ops;
        dp->dl_retries = 1;
 
-       queue_work(callback_wq, &dp->dl_recall.cb_work);
+       INIT_LIST_HEAD(&cb->cb_per_client);
+       cb->cb_done = true;
+
+       run_nfsd4_cb(&dp->dl_recall);
 }