Linux-2.6.12-rc2
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / clnt.c
blob02bc029d46fe780caf634c6f080b5e8f84a54b79
1 /*
2 * linux/net/sunrpc/rpcclnt.c
4 * This file contains the high-level RPC interface.
5 * It is modeled as a finite state machine to support both synchronous
6 * and asynchronous requests.
8 * - RPC header generation and argument serialization.
9 * - Credential refresh.
10 * - TCP connect handling.
11 * - Retry of operation when it is suspected the operation failed because
12 * of uid squashing on the server, or when the credentials were stale
13 * and need to be refreshed, or when a packet was damaged in transit.
14 * This may be have to be moved to the VFS layer.
16 * NB: BSD uses a more intelligent approach to guessing when a request
17 * or reply has been lost by keeping the RTO estimate for each procedure.
18 * We currently make do with a constant timeout value.
20 * Copyright (C) 1992,1993 Rick Sladkey <jrs@world.std.com>
21 * Copyright (C) 1995,1996 Olaf Kirch <okir@monad.swb.de>
24 #include <asm/system.h>
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/mm.h>
29 #include <linux/slab.h>
30 #include <linux/in.h>
31 #include <linux/utsname.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/workqueue.h>
35 #include <linux/sunrpc/rpc_pipe_fs.h>
37 #include <linux/nfs.h>
40 #define RPC_SLACK_SPACE (1024) /* total overkill */
42 #ifdef RPC_DEBUG
43 # define RPCDBG_FACILITY RPCDBG_CALL
44 #endif
46 static DECLARE_WAIT_QUEUE_HEAD(destroy_wait);
49 static void call_start(struct rpc_task *task);
50 static void call_reserve(struct rpc_task *task);
51 static void call_reserveresult(struct rpc_task *task);
52 static void call_allocate(struct rpc_task *task);
53 static void call_encode(struct rpc_task *task);
54 static void call_decode(struct rpc_task *task);
55 static void call_bind(struct rpc_task *task);
56 static void call_transmit(struct rpc_task *task);
57 static void call_status(struct rpc_task *task);
58 static void call_refresh(struct rpc_task *task);
59 static void call_refreshresult(struct rpc_task *task);
60 static void call_timeout(struct rpc_task *task);
61 static void call_connect(struct rpc_task *task);
62 static void call_connect_status(struct rpc_task *task);
63 static u32 * call_header(struct rpc_task *task);
64 static u32 * call_verify(struct rpc_task *task);
67 static int
68 rpc_setup_pipedir(struct rpc_clnt *clnt, char *dir_name)
70 static uint32_t clntid;
71 int error;
73 if (dir_name == NULL)
74 return 0;
75 for (;;) {
76 snprintf(clnt->cl_pathname, sizeof(clnt->cl_pathname),
77 "%s/clnt%x", dir_name,
78 (unsigned int)clntid++);
79 clnt->cl_pathname[sizeof(clnt->cl_pathname) - 1] = '\0';
80 clnt->cl_dentry = rpc_mkdir(clnt->cl_pathname, clnt);
81 if (!IS_ERR(clnt->cl_dentry))
82 return 0;
83 error = PTR_ERR(clnt->cl_dentry);
84 if (error != -EEXIST) {
85 printk(KERN_INFO "RPC: Couldn't create pipefs entry %s, error %d\n",
86 clnt->cl_pathname, error);
87 return error;
93 * Create an RPC client
94 * FIXME: This should also take a flags argument (as in task->tk_flags).
95 * It's called (among others) from pmap_create_client, which may in
96 * turn be called by an async task. In this case, rpciod should not be
97 * made to sleep too long.
99 struct rpc_clnt *
100 rpc_create_client(struct rpc_xprt *xprt, char *servname,
101 struct rpc_program *program, u32 vers,
102 rpc_authflavor_t flavor)
104 struct rpc_version *version;
105 struct rpc_clnt *clnt = NULL;
106 int err;
107 int len;
109 dprintk("RPC: creating %s client for %s (xprt %p)\n",
110 program->name, servname, xprt);
112 err = -EINVAL;
113 if (!xprt)
114 goto out_err;
115 if (vers >= program->nrvers || !(version = program->version[vers]))
116 goto out_err;
118 err = -ENOMEM;
119 clnt = (struct rpc_clnt *) kmalloc(sizeof(*clnt), GFP_KERNEL);
120 if (!clnt)
121 goto out_err;
122 memset(clnt, 0, sizeof(*clnt));
123 atomic_set(&clnt->cl_users, 0);
124 atomic_set(&clnt->cl_count, 1);
125 clnt->cl_parent = clnt;
127 clnt->cl_server = clnt->cl_inline_name;
128 len = strlen(servname) + 1;
129 if (len > sizeof(clnt->cl_inline_name)) {
130 char *buf = kmalloc(len, GFP_KERNEL);
131 if (buf != 0)
132 clnt->cl_server = buf;
133 else
134 len = sizeof(clnt->cl_inline_name);
136 strlcpy(clnt->cl_server, servname, len);
138 clnt->cl_xprt = xprt;
139 clnt->cl_procinfo = version->procs;
140 clnt->cl_maxproc = version->nrprocs;
141 clnt->cl_protname = program->name;
142 clnt->cl_pmap = &clnt->cl_pmap_default;
143 clnt->cl_port = xprt->addr.sin_port;
144 clnt->cl_prog = program->number;
145 clnt->cl_vers = version->number;
146 clnt->cl_prot = xprt->prot;
147 clnt->cl_stats = program->stats;
148 rpc_init_wait_queue(&clnt->cl_pmap_default.pm_bindwait, "bindwait");
150 if (!clnt->cl_port)
151 clnt->cl_autobind = 1;
153 clnt->cl_rtt = &clnt->cl_rtt_default;
154 rpc_init_rtt(&clnt->cl_rtt_default, xprt->timeout.to_initval);
156 err = rpc_setup_pipedir(clnt, program->pipe_dir_name);
157 if (err < 0)
158 goto out_no_path;
160 err = -ENOMEM;
161 if (!rpcauth_create(flavor, clnt)) {
162 printk(KERN_INFO "RPC: Couldn't create auth handle (flavor %u)\n",
163 flavor);
164 goto out_no_auth;
167 /* save the nodename */
168 clnt->cl_nodelen = strlen(system_utsname.nodename);
169 if (clnt->cl_nodelen > UNX_MAXNODENAME)
170 clnt->cl_nodelen = UNX_MAXNODENAME;
171 memcpy(clnt->cl_nodename, system_utsname.nodename, clnt->cl_nodelen);
172 return clnt;
174 out_no_auth:
175 rpc_rmdir(clnt->cl_pathname);
176 out_no_path:
177 if (clnt->cl_server != clnt->cl_inline_name)
178 kfree(clnt->cl_server);
179 kfree(clnt);
180 out_err:
181 return ERR_PTR(err);
185 * This function clones the RPC client structure. It allows us to share the
186 * same transport while varying parameters such as the authentication
187 * flavour.
189 struct rpc_clnt *
190 rpc_clone_client(struct rpc_clnt *clnt)
192 struct rpc_clnt *new;
194 new = (struct rpc_clnt *)kmalloc(sizeof(*new), GFP_KERNEL);
195 if (!new)
196 goto out_no_clnt;
197 memcpy(new, clnt, sizeof(*new));
198 atomic_set(&new->cl_count, 1);
199 atomic_set(&new->cl_users, 0);
200 new->cl_parent = clnt;
201 atomic_inc(&clnt->cl_count);
202 /* Duplicate portmapper */
203 rpc_init_wait_queue(&new->cl_pmap_default.pm_bindwait, "bindwait");
204 /* Turn off autobind on clones */
205 new->cl_autobind = 0;
206 new->cl_oneshot = 0;
207 new->cl_dead = 0;
208 rpc_init_rtt(&new->cl_rtt_default, clnt->cl_xprt->timeout.to_initval);
209 if (new->cl_auth)
210 atomic_inc(&new->cl_auth->au_count);
211 return new;
212 out_no_clnt:
213 printk(KERN_INFO "RPC: out of memory in %s\n", __FUNCTION__);
214 return ERR_PTR(-ENOMEM);
218 * Properly shut down an RPC client, terminating all outstanding
219 * requests. Note that we must be certain that cl_oneshot and
220 * cl_dead are cleared, or else the client would be destroyed
221 * when the last task releases it.
224 rpc_shutdown_client(struct rpc_clnt *clnt)
226 dprintk("RPC: shutting down %s client for %s, tasks=%d\n",
227 clnt->cl_protname, clnt->cl_server,
228 atomic_read(&clnt->cl_users));
230 while (atomic_read(&clnt->cl_users) > 0) {
231 /* Don't let rpc_release_client destroy us */
232 clnt->cl_oneshot = 0;
233 clnt->cl_dead = 0;
234 rpc_killall_tasks(clnt);
235 sleep_on_timeout(&destroy_wait, 1*HZ);
238 if (atomic_read(&clnt->cl_users) < 0) {
239 printk(KERN_ERR "RPC: rpc_shutdown_client clnt %p tasks=%d\n",
240 clnt, atomic_read(&clnt->cl_users));
241 #ifdef RPC_DEBUG
242 rpc_show_tasks();
243 #endif
244 BUG();
247 return rpc_destroy_client(clnt);
251 * Delete an RPC client
254 rpc_destroy_client(struct rpc_clnt *clnt)
256 if (!atomic_dec_and_test(&clnt->cl_count))
257 return 1;
258 BUG_ON(atomic_read(&clnt->cl_users) != 0);
260 dprintk("RPC: destroying %s client for %s\n",
261 clnt->cl_protname, clnt->cl_server);
262 if (clnt->cl_auth) {
263 rpcauth_destroy(clnt->cl_auth);
264 clnt->cl_auth = NULL;
266 if (clnt->cl_parent != clnt) {
267 rpc_destroy_client(clnt->cl_parent);
268 goto out_free;
270 if (clnt->cl_pathname[0])
271 rpc_rmdir(clnt->cl_pathname);
272 if (clnt->cl_xprt) {
273 xprt_destroy(clnt->cl_xprt);
274 clnt->cl_xprt = NULL;
276 if (clnt->cl_server != clnt->cl_inline_name)
277 kfree(clnt->cl_server);
278 out_free:
279 kfree(clnt);
280 return 0;
284 * Release an RPC client
286 void
287 rpc_release_client(struct rpc_clnt *clnt)
289 dprintk("RPC: rpc_release_client(%p, %d)\n",
290 clnt, atomic_read(&clnt->cl_users));
292 if (!atomic_dec_and_test(&clnt->cl_users))
293 return;
294 wake_up(&destroy_wait);
295 if (clnt->cl_oneshot || clnt->cl_dead)
296 rpc_destroy_client(clnt);
300 * Default callback for async RPC calls
302 static void
303 rpc_default_callback(struct rpc_task *task)
308 * Export the signal mask handling for aysnchronous code that
309 * sleeps on RPC calls
312 void rpc_clnt_sigmask(struct rpc_clnt *clnt, sigset_t *oldset)
314 unsigned long sigallow = sigmask(SIGKILL);
315 unsigned long irqflags;
317 /* Turn off various signals */
318 if (clnt->cl_intr) {
319 struct k_sigaction *action = current->sighand->action;
320 if (action[SIGINT-1].sa.sa_handler == SIG_DFL)
321 sigallow |= sigmask(SIGINT);
322 if (action[SIGQUIT-1].sa.sa_handler == SIG_DFL)
323 sigallow |= sigmask(SIGQUIT);
325 spin_lock_irqsave(&current->sighand->siglock, irqflags);
326 *oldset = current->blocked;
327 siginitsetinv(&current->blocked, sigallow & ~oldset->sig[0]);
328 recalc_sigpending();
329 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
332 void rpc_clnt_sigunmask(struct rpc_clnt *clnt, sigset_t *oldset)
334 unsigned long irqflags;
336 spin_lock_irqsave(&current->sighand->siglock, irqflags);
337 current->blocked = *oldset;
338 recalc_sigpending();
339 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
343 * New rpc_call implementation
345 int rpc_call_sync(struct rpc_clnt *clnt, struct rpc_message *msg, int flags)
347 struct rpc_task *task;
348 sigset_t oldset;
349 int status;
351 /* If this client is slain all further I/O fails */
352 if (clnt->cl_dead)
353 return -EIO;
355 BUG_ON(flags & RPC_TASK_ASYNC);
357 rpc_clnt_sigmask(clnt, &oldset);
359 status = -ENOMEM;
360 task = rpc_new_task(clnt, NULL, flags);
361 if (task == NULL)
362 goto out;
364 rpc_call_setup(task, msg, 0);
366 /* Set up the call info struct and execute the task */
367 if (task->tk_status == 0)
368 status = rpc_execute(task);
369 else {
370 status = task->tk_status;
371 rpc_release_task(task);
374 out:
375 rpc_clnt_sigunmask(clnt, &oldset);
377 return status;
381 * New rpc_call implementation
384 rpc_call_async(struct rpc_clnt *clnt, struct rpc_message *msg, int flags,
385 rpc_action callback, void *data)
387 struct rpc_task *task;
388 sigset_t oldset;
389 int status;
391 /* If this client is slain all further I/O fails */
392 if (clnt->cl_dead)
393 return -EIO;
395 flags |= RPC_TASK_ASYNC;
397 rpc_clnt_sigmask(clnt, &oldset);
399 /* Create/initialize a new RPC task */
400 if (!callback)
401 callback = rpc_default_callback;
402 status = -ENOMEM;
403 if (!(task = rpc_new_task(clnt, callback, flags)))
404 goto out;
405 task->tk_calldata = data;
407 rpc_call_setup(task, msg, 0);
409 /* Set up the call info struct and execute the task */
410 status = task->tk_status;
411 if (status == 0)
412 rpc_execute(task);
413 else
414 rpc_release_task(task);
416 out:
417 rpc_clnt_sigunmask(clnt, &oldset);
419 return status;
423 void
424 rpc_call_setup(struct rpc_task *task, struct rpc_message *msg, int flags)
426 task->tk_msg = *msg;
427 task->tk_flags |= flags;
428 /* Bind the user cred */
429 if (task->tk_msg.rpc_cred != NULL)
430 rpcauth_holdcred(task);
431 else
432 rpcauth_bindcred(task);
434 if (task->tk_status == 0)
435 task->tk_action = call_start;
436 else
437 task->tk_action = NULL;
440 void
441 rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize)
443 struct rpc_xprt *xprt = clnt->cl_xprt;
445 xprt->sndsize = 0;
446 if (sndsize)
447 xprt->sndsize = sndsize + RPC_SLACK_SPACE;
448 xprt->rcvsize = 0;
449 if (rcvsize)
450 xprt->rcvsize = rcvsize + RPC_SLACK_SPACE;
451 if (xprt_connected(xprt))
452 xprt_sock_setbufsize(xprt);
456 * Return size of largest payload RPC client can support, in bytes
458 * For stream transports, this is one RPC record fragment (see RFC
459 * 1831), as we don't support multi-record requests yet. For datagram
460 * transports, this is the size of an IP packet minus the IP, UDP, and
461 * RPC header sizes.
463 size_t rpc_max_payload(struct rpc_clnt *clnt)
465 return clnt->cl_xprt->max_payload;
467 EXPORT_SYMBOL(rpc_max_payload);
470 * Restart an (async) RPC call. Usually called from within the
471 * exit handler.
473 void
474 rpc_restart_call(struct rpc_task *task)
476 if (RPC_ASSASSINATED(task))
477 return;
479 task->tk_action = call_start;
483 * 0. Initial state
485 * Other FSM states can be visited zero or more times, but
486 * this state is visited exactly once for each RPC.
488 static void
489 call_start(struct rpc_task *task)
491 struct rpc_clnt *clnt = task->tk_client;
493 dprintk("RPC: %4d call_start %s%d proc %d (%s)\n", task->tk_pid,
494 clnt->cl_protname, clnt->cl_vers, task->tk_msg.rpc_proc->p_proc,
495 (RPC_IS_ASYNC(task) ? "async" : "sync"));
497 /* Increment call count */
498 task->tk_msg.rpc_proc->p_count++;
499 clnt->cl_stats->rpccnt++;
500 task->tk_action = call_reserve;
504 * 1. Reserve an RPC call slot
506 static void
507 call_reserve(struct rpc_task *task)
509 dprintk("RPC: %4d call_reserve\n", task->tk_pid);
511 if (!rpcauth_uptodatecred(task)) {
512 task->tk_action = call_refresh;
513 return;
516 task->tk_status = 0;
517 task->tk_action = call_reserveresult;
518 xprt_reserve(task);
522 * 1b. Grok the result of xprt_reserve()
524 static void
525 call_reserveresult(struct rpc_task *task)
527 int status = task->tk_status;
529 dprintk("RPC: %4d call_reserveresult (status %d)\n",
530 task->tk_pid, task->tk_status);
533 * After a call to xprt_reserve(), we must have either
534 * a request slot or else an error status.
536 task->tk_status = 0;
537 if (status >= 0) {
538 if (task->tk_rqstp) {
539 task->tk_action = call_allocate;
540 return;
543 printk(KERN_ERR "%s: status=%d, but no request slot, exiting\n",
544 __FUNCTION__, status);
545 rpc_exit(task, -EIO);
546 return;
550 * Even though there was an error, we may have acquired
551 * a request slot somehow. Make sure not to leak it.
553 if (task->tk_rqstp) {
554 printk(KERN_ERR "%s: status=%d, request allocated anyway\n",
555 __FUNCTION__, status);
556 xprt_release(task);
559 switch (status) {
560 case -EAGAIN: /* woken up; retry */
561 task->tk_action = call_reserve;
562 return;
563 case -EIO: /* probably a shutdown */
564 break;
565 default:
566 printk(KERN_ERR "%s: unrecognized error %d, exiting\n",
567 __FUNCTION__, status);
568 break;
570 rpc_exit(task, status);
574 * 2. Allocate the buffer. For details, see sched.c:rpc_malloc.
575 * (Note: buffer memory is freed in rpc_task_release).
577 static void
578 call_allocate(struct rpc_task *task)
580 unsigned int bufsiz;
582 dprintk("RPC: %4d call_allocate (status %d)\n",
583 task->tk_pid, task->tk_status);
584 task->tk_action = call_bind;
585 if (task->tk_buffer)
586 return;
588 /* FIXME: compute buffer requirements more exactly using
589 * auth->au_wslack */
590 bufsiz = task->tk_msg.rpc_proc->p_bufsiz + RPC_SLACK_SPACE;
592 if (rpc_malloc(task, bufsiz << 1) != NULL)
593 return;
594 printk(KERN_INFO "RPC: buffer allocation failed for task %p\n", task);
596 if (RPC_IS_ASYNC(task) || !(task->tk_client->cl_intr && signalled())) {
597 xprt_release(task);
598 task->tk_action = call_reserve;
599 rpc_delay(task, HZ>>4);
600 return;
603 rpc_exit(task, -ERESTARTSYS);
607 * 3. Encode arguments of an RPC call
609 static void
610 call_encode(struct rpc_task *task)
612 struct rpc_clnt *clnt = task->tk_client;
613 struct rpc_rqst *req = task->tk_rqstp;
614 struct xdr_buf *sndbuf = &req->rq_snd_buf;
615 struct xdr_buf *rcvbuf = &req->rq_rcv_buf;
616 unsigned int bufsiz;
617 kxdrproc_t encode;
618 int status;
619 u32 *p;
621 dprintk("RPC: %4d call_encode (status %d)\n",
622 task->tk_pid, task->tk_status);
624 /* Default buffer setup */
625 bufsiz = task->tk_bufsize >> 1;
626 sndbuf->head[0].iov_base = (void *)task->tk_buffer;
627 sndbuf->head[0].iov_len = bufsiz;
628 sndbuf->tail[0].iov_len = 0;
629 sndbuf->page_len = 0;
630 sndbuf->len = 0;
631 sndbuf->buflen = bufsiz;
632 rcvbuf->head[0].iov_base = (void *)((char *)task->tk_buffer + bufsiz);
633 rcvbuf->head[0].iov_len = bufsiz;
634 rcvbuf->tail[0].iov_len = 0;
635 rcvbuf->page_len = 0;
636 rcvbuf->len = 0;
637 rcvbuf->buflen = bufsiz;
639 /* Encode header and provided arguments */
640 encode = task->tk_msg.rpc_proc->p_encode;
641 if (!(p = call_header(task))) {
642 printk(KERN_INFO "RPC: call_header failed, exit EIO\n");
643 rpc_exit(task, -EIO);
644 return;
646 if (encode && (status = rpcauth_wrap_req(task, encode, req, p,
647 task->tk_msg.rpc_argp)) < 0) {
648 printk(KERN_WARNING "%s: can't encode arguments: %d\n",
649 clnt->cl_protname, -status);
650 rpc_exit(task, status);
655 * 4. Get the server port number if not yet set
657 static void
658 call_bind(struct rpc_task *task)
660 struct rpc_clnt *clnt = task->tk_client;
661 struct rpc_xprt *xprt = clnt->cl_xprt;
663 dprintk("RPC: %4d call_bind xprt %p %s connected\n", task->tk_pid,
664 xprt, (xprt_connected(xprt) ? "is" : "is not"));
666 task->tk_action = (xprt_connected(xprt)) ? call_transmit : call_connect;
668 if (!clnt->cl_port) {
669 task->tk_action = call_connect;
670 task->tk_timeout = RPC_CONNECT_TIMEOUT;
671 rpc_getport(task, clnt);
676 * 4a. Connect to the RPC server (TCP case)
678 static void
679 call_connect(struct rpc_task *task)
681 struct rpc_clnt *clnt = task->tk_client;
683 dprintk("RPC: %4d call_connect status %d\n",
684 task->tk_pid, task->tk_status);
686 if (xprt_connected(clnt->cl_xprt)) {
687 task->tk_action = call_transmit;
688 return;
690 task->tk_action = call_connect_status;
691 if (task->tk_status < 0)
692 return;
693 xprt_connect(task);
697 * 4b. Sort out connect result
699 static void
700 call_connect_status(struct rpc_task *task)
702 struct rpc_clnt *clnt = task->tk_client;
703 int status = task->tk_status;
705 task->tk_status = 0;
706 if (status >= 0) {
707 clnt->cl_stats->netreconn++;
708 task->tk_action = call_transmit;
709 return;
712 /* Something failed: we may have to rebind */
713 if (clnt->cl_autobind)
714 clnt->cl_port = 0;
715 switch (status) {
716 case -ENOTCONN:
717 case -ETIMEDOUT:
718 case -EAGAIN:
719 task->tk_action = (clnt->cl_port == 0) ? call_bind : call_connect;
720 break;
721 default:
722 rpc_exit(task, -EIO);
727 * 5. Transmit the RPC request, and wait for reply
729 static void
730 call_transmit(struct rpc_task *task)
732 dprintk("RPC: %4d call_transmit (status %d)\n",
733 task->tk_pid, task->tk_status);
735 task->tk_action = call_status;
736 if (task->tk_status < 0)
737 return;
738 task->tk_status = xprt_prepare_transmit(task);
739 if (task->tk_status != 0)
740 return;
741 /* Encode here so that rpcsec_gss can use correct sequence number. */
742 if (!task->tk_rqstp->rq_bytes_sent)
743 call_encode(task);
744 if (task->tk_status < 0)
745 return;
746 xprt_transmit(task);
747 if (task->tk_status < 0)
748 return;
749 if (!task->tk_msg.rpc_proc->p_decode) {
750 task->tk_action = NULL;
751 rpc_wake_up_task(task);
756 * 6. Sort out the RPC call status
758 static void
759 call_status(struct rpc_task *task)
761 struct rpc_clnt *clnt = task->tk_client;
762 struct rpc_rqst *req = task->tk_rqstp;
763 int status;
765 if (req->rq_received > 0 && !req->rq_bytes_sent)
766 task->tk_status = req->rq_received;
768 dprintk("RPC: %4d call_status (status %d)\n",
769 task->tk_pid, task->tk_status);
771 status = task->tk_status;
772 if (status >= 0) {
773 task->tk_action = call_decode;
774 return;
777 task->tk_status = 0;
778 switch(status) {
779 case -ETIMEDOUT:
780 task->tk_action = call_timeout;
781 break;
782 case -ECONNREFUSED:
783 case -ENOTCONN:
784 req->rq_bytes_sent = 0;
785 if (clnt->cl_autobind)
786 clnt->cl_port = 0;
787 task->tk_action = call_bind;
788 break;
789 case -EAGAIN:
790 task->tk_action = call_transmit;
791 break;
792 case -EIO:
793 /* shutdown or soft timeout */
794 rpc_exit(task, status);
795 break;
796 default:
797 if (clnt->cl_chatty)
798 printk("%s: RPC call returned error %d\n",
799 clnt->cl_protname, -status);
800 rpc_exit(task, status);
801 break;
806 * 6a. Handle RPC timeout
807 * We do not release the request slot, so we keep using the
808 * same XID for all retransmits.
810 static void
811 call_timeout(struct rpc_task *task)
813 struct rpc_clnt *clnt = task->tk_client;
815 if (xprt_adjust_timeout(task->tk_rqstp) == 0) {
816 dprintk("RPC: %4d call_timeout (minor)\n", task->tk_pid);
817 goto retry;
820 dprintk("RPC: %4d call_timeout (major)\n", task->tk_pid);
821 if (RPC_IS_SOFT(task)) {
822 if (clnt->cl_chatty)
823 printk(KERN_NOTICE "%s: server %s not responding, timed out\n",
824 clnt->cl_protname, clnt->cl_server);
825 rpc_exit(task, -EIO);
826 return;
829 if (clnt->cl_chatty && !(task->tk_flags & RPC_CALL_MAJORSEEN)) {
830 task->tk_flags |= RPC_CALL_MAJORSEEN;
831 printk(KERN_NOTICE "%s: server %s not responding, still trying\n",
832 clnt->cl_protname, clnt->cl_server);
834 if (clnt->cl_autobind)
835 clnt->cl_port = 0;
837 retry:
838 clnt->cl_stats->rpcretrans++;
839 task->tk_action = call_bind;
840 task->tk_status = 0;
844 * 7. Decode the RPC reply
846 static void
847 call_decode(struct rpc_task *task)
849 struct rpc_clnt *clnt = task->tk_client;
850 struct rpc_rqst *req = task->tk_rqstp;
851 kxdrproc_t decode = task->tk_msg.rpc_proc->p_decode;
852 u32 *p;
854 dprintk("RPC: %4d call_decode (status %d)\n",
855 task->tk_pid, task->tk_status);
857 if (clnt->cl_chatty && (task->tk_flags & RPC_CALL_MAJORSEEN)) {
858 printk(KERN_NOTICE "%s: server %s OK\n",
859 clnt->cl_protname, clnt->cl_server);
860 task->tk_flags &= ~RPC_CALL_MAJORSEEN;
863 if (task->tk_status < 12) {
864 if (!RPC_IS_SOFT(task)) {
865 task->tk_action = call_bind;
866 clnt->cl_stats->rpcretrans++;
867 goto out_retry;
869 printk(KERN_WARNING "%s: too small RPC reply size (%d bytes)\n",
870 clnt->cl_protname, task->tk_status);
871 rpc_exit(task, -EIO);
872 return;
875 req->rq_rcv_buf.len = req->rq_private_buf.len;
877 /* Check that the softirq receive buffer is valid */
878 WARN_ON(memcmp(&req->rq_rcv_buf, &req->rq_private_buf,
879 sizeof(req->rq_rcv_buf)) != 0);
881 /* Verify the RPC header */
882 if (!(p = call_verify(task))) {
883 if (task->tk_action == NULL)
884 return;
885 goto out_retry;
888 task->tk_action = NULL;
890 if (decode)
891 task->tk_status = rpcauth_unwrap_resp(task, decode, req, p,
892 task->tk_msg.rpc_resp);
893 dprintk("RPC: %4d call_decode result %d\n", task->tk_pid,
894 task->tk_status);
895 return;
896 out_retry:
897 req->rq_received = req->rq_private_buf.len = 0;
898 task->tk_status = 0;
902 * 8. Refresh the credentials if rejected by the server
904 static void
905 call_refresh(struct rpc_task *task)
907 dprintk("RPC: %4d call_refresh\n", task->tk_pid);
909 xprt_release(task); /* Must do to obtain new XID */
910 task->tk_action = call_refreshresult;
911 task->tk_status = 0;
912 task->tk_client->cl_stats->rpcauthrefresh++;
913 rpcauth_refreshcred(task);
917 * 8a. Process the results of a credential refresh
919 static void
920 call_refreshresult(struct rpc_task *task)
922 int status = task->tk_status;
923 dprintk("RPC: %4d call_refreshresult (status %d)\n",
924 task->tk_pid, task->tk_status);
926 task->tk_status = 0;
927 task->tk_action = call_reserve;
928 if (status >= 0 && rpcauth_uptodatecred(task))
929 return;
930 if (status == -EACCES) {
931 rpc_exit(task, -EACCES);
932 return;
934 task->tk_action = call_refresh;
935 if (status != -ETIMEDOUT)
936 rpc_delay(task, 3*HZ);
937 return;
941 * Call header serialization
943 static u32 *
944 call_header(struct rpc_task *task)
946 struct rpc_clnt *clnt = task->tk_client;
947 struct rpc_xprt *xprt = clnt->cl_xprt;
948 struct rpc_rqst *req = task->tk_rqstp;
949 u32 *p = req->rq_svec[0].iov_base;
951 /* FIXME: check buffer size? */
952 if (xprt->stream)
953 *p++ = 0; /* fill in later */
954 *p++ = req->rq_xid; /* XID */
955 *p++ = htonl(RPC_CALL); /* CALL */
956 *p++ = htonl(RPC_VERSION); /* RPC version */
957 *p++ = htonl(clnt->cl_prog); /* program number */
958 *p++ = htonl(clnt->cl_vers); /* program version */
959 *p++ = htonl(task->tk_msg.rpc_proc->p_proc); /* procedure */
960 return rpcauth_marshcred(task, p);
964 * Reply header verification
966 static u32 *
967 call_verify(struct rpc_task *task)
969 struct kvec *iov = &task->tk_rqstp->rq_rcv_buf.head[0];
970 int len = task->tk_rqstp->rq_rcv_buf.len >> 2;
971 u32 *p = iov->iov_base, n;
972 int error = -EACCES;
974 if ((len -= 3) < 0)
975 goto out_overflow;
976 p += 1; /* skip XID */
978 if ((n = ntohl(*p++)) != RPC_REPLY) {
979 printk(KERN_WARNING "call_verify: not an RPC reply: %x\n", n);
980 goto out_retry;
982 if ((n = ntohl(*p++)) != RPC_MSG_ACCEPTED) {
983 if (--len < 0)
984 goto out_overflow;
985 switch ((n = ntohl(*p++))) {
986 case RPC_AUTH_ERROR:
987 break;
988 case RPC_MISMATCH:
989 printk(KERN_WARNING "%s: RPC call version mismatch!\n", __FUNCTION__);
990 goto out_eio;
991 default:
992 printk(KERN_WARNING "%s: RPC call rejected, unknown error: %x\n", __FUNCTION__, n);
993 goto out_eio;
995 if (--len < 0)
996 goto out_overflow;
997 switch ((n = ntohl(*p++))) {
998 case RPC_AUTH_REJECTEDCRED:
999 case RPC_AUTH_REJECTEDVERF:
1000 case RPCSEC_GSS_CREDPROBLEM:
1001 case RPCSEC_GSS_CTXPROBLEM:
1002 if (!task->tk_cred_retry)
1003 break;
1004 task->tk_cred_retry--;
1005 dprintk("RPC: %4d call_verify: retry stale creds\n",
1006 task->tk_pid);
1007 rpcauth_invalcred(task);
1008 task->tk_action = call_refresh;
1009 return NULL;
1010 case RPC_AUTH_BADCRED:
1011 case RPC_AUTH_BADVERF:
1012 /* possibly garbled cred/verf? */
1013 if (!task->tk_garb_retry)
1014 break;
1015 task->tk_garb_retry--;
1016 dprintk("RPC: %4d call_verify: retry garbled creds\n",
1017 task->tk_pid);
1018 task->tk_action = call_bind;
1019 return NULL;
1020 case RPC_AUTH_TOOWEAK:
1021 printk(KERN_NOTICE "call_verify: server requires stronger "
1022 "authentication.\n");
1023 break;
1024 default:
1025 printk(KERN_WARNING "call_verify: unknown auth error: %x\n", n);
1026 error = -EIO;
1028 dprintk("RPC: %4d call_verify: call rejected %d\n",
1029 task->tk_pid, n);
1030 goto out_err;
1032 if (!(p = rpcauth_checkverf(task, p))) {
1033 printk(KERN_WARNING "call_verify: auth check failed\n");
1034 goto out_retry; /* bad verifier, retry */
1036 len = p - (u32 *)iov->iov_base - 1;
1037 if (len < 0)
1038 goto out_overflow;
1039 switch ((n = ntohl(*p++))) {
1040 case RPC_SUCCESS:
1041 return p;
1042 case RPC_PROG_UNAVAIL:
1043 printk(KERN_WARNING "RPC: call_verify: program %u is unsupported by server %s\n",
1044 (unsigned int)task->tk_client->cl_prog,
1045 task->tk_client->cl_server);
1046 goto out_eio;
1047 case RPC_PROG_MISMATCH:
1048 printk(KERN_WARNING "RPC: call_verify: program %u, version %u unsupported by server %s\n",
1049 (unsigned int)task->tk_client->cl_prog,
1050 (unsigned int)task->tk_client->cl_vers,
1051 task->tk_client->cl_server);
1052 goto out_eio;
1053 case RPC_PROC_UNAVAIL:
1054 printk(KERN_WARNING "RPC: call_verify: proc %p unsupported by program %u, version %u on server %s\n",
1055 task->tk_msg.rpc_proc,
1056 task->tk_client->cl_prog,
1057 task->tk_client->cl_vers,
1058 task->tk_client->cl_server);
1059 goto out_eio;
1060 case RPC_GARBAGE_ARGS:
1061 dprintk("RPC: %4d %s: server saw garbage\n", task->tk_pid, __FUNCTION__);
1062 break; /* retry */
1063 default:
1064 printk(KERN_WARNING "call_verify: server accept status: %x\n", n);
1065 /* Also retry */
1068 out_retry:
1069 task->tk_client->cl_stats->rpcgarbage++;
1070 if (task->tk_garb_retry) {
1071 task->tk_garb_retry--;
1072 dprintk(KERN_WARNING "RPC %s: retrying %4d\n", __FUNCTION__, task->tk_pid);
1073 task->tk_action = call_bind;
1074 return NULL;
1076 printk(KERN_WARNING "RPC %s: retry failed, exit EIO\n", __FUNCTION__);
1077 out_eio:
1078 error = -EIO;
1079 out_err:
1080 rpc_exit(task, error);
1081 return NULL;
1082 out_overflow:
1083 printk(KERN_WARNING "RPC %s: server reply was truncated.\n", __FUNCTION__);
1084 goto out_retry;