Merge branch 'media_fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
[cris-mirror.git] / net / sunrpc / svc.c
blob08e05a8ce0255b338cc9c812006302a79e7ad4a9
1 /*
2 * linux/net/sunrpc/svc.c
4 * High-level RPC service routines
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
10 * by Greg Banks <gnb@melbourne.sgi.com>
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
17 #include <linux/in.h>
18 #include <linux/mm.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
24 #include <linux/sunrpc/types.h>
25 #include <linux/sunrpc/xdr.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svcsock.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/bc_xprt.h>
31 #define RPCDBG_FACILITY RPCDBG_SVCDSP
33 static void svc_unregister(const struct svc_serv *serv);
35 #define svc_serv_is_pooled(serv) ((serv)->sv_function)
38 * Mode for mapping cpus to pools.
40 enum {
41 SVC_POOL_AUTO = -1, /* choose one of the others */
42 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
43 * (legacy & UP mode) */
44 SVC_POOL_PERCPU, /* one pool per cpu */
45 SVC_POOL_PERNODE /* one pool per numa node */
47 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
50 * Structure for mapping cpus to pools and vice versa.
51 * Setup once during sunrpc initialisation.
53 static struct svc_pool_map {
54 int count; /* How many svc_servs use us */
55 int mode; /* Note: int not enum to avoid
56 * warnings about "enumeration value
57 * not handled in switch" */
58 unsigned int npools;
59 unsigned int *pool_to; /* maps pool id to cpu or node */
60 unsigned int *to_pool; /* maps cpu or node to pool id */
61 } svc_pool_map = {
62 .count = 0,
63 .mode = SVC_POOL_DEFAULT
65 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
67 static int
68 param_set_pool_mode(const char *val, struct kernel_param *kp)
70 int *ip = (int *)kp->arg;
71 struct svc_pool_map *m = &svc_pool_map;
72 int err;
74 mutex_lock(&svc_pool_map_mutex);
76 err = -EBUSY;
77 if (m->count)
78 goto out;
80 err = 0;
81 if (!strncmp(val, "auto", 4))
82 *ip = SVC_POOL_AUTO;
83 else if (!strncmp(val, "global", 6))
84 *ip = SVC_POOL_GLOBAL;
85 else if (!strncmp(val, "percpu", 6))
86 *ip = SVC_POOL_PERCPU;
87 else if (!strncmp(val, "pernode", 7))
88 *ip = SVC_POOL_PERNODE;
89 else
90 err = -EINVAL;
92 out:
93 mutex_unlock(&svc_pool_map_mutex);
94 return err;
97 static int
98 param_get_pool_mode(char *buf, struct kernel_param *kp)
100 int *ip = (int *)kp->arg;
102 switch (*ip)
104 case SVC_POOL_AUTO:
105 return strlcpy(buf, "auto", 20);
106 case SVC_POOL_GLOBAL:
107 return strlcpy(buf, "global", 20);
108 case SVC_POOL_PERCPU:
109 return strlcpy(buf, "percpu", 20);
110 case SVC_POOL_PERNODE:
111 return strlcpy(buf, "pernode", 20);
112 default:
113 return sprintf(buf, "%d", *ip);
117 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
118 &svc_pool_map.mode, 0644);
121 * Detect best pool mapping mode heuristically,
122 * according to the machine's topology.
124 static int
125 svc_pool_map_choose_mode(void)
127 unsigned int node;
129 if (nr_online_nodes > 1) {
131 * Actually have multiple NUMA nodes,
132 * so split pools on NUMA node boundaries
134 return SVC_POOL_PERNODE;
137 node = first_online_node;
138 if (nr_cpus_node(node) > 2) {
140 * Non-trivial SMP, or CONFIG_NUMA on
141 * non-NUMA hardware, e.g. with a generic
142 * x86_64 kernel on Xeons. In this case we
143 * want to divide the pools on cpu boundaries.
145 return SVC_POOL_PERCPU;
148 /* default: one global pool */
149 return SVC_POOL_GLOBAL;
153 * Allocate the to_pool[] and pool_to[] arrays.
154 * Returns 0 on success or an errno.
156 static int
157 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
159 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
160 if (!m->to_pool)
161 goto fail;
162 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
163 if (!m->pool_to)
164 goto fail_free;
166 return 0;
168 fail_free:
169 kfree(m->to_pool);
170 fail:
171 return -ENOMEM;
175 * Initialise the pool map for SVC_POOL_PERCPU mode.
176 * Returns number of pools or <0 on error.
178 static int
179 svc_pool_map_init_percpu(struct svc_pool_map *m)
181 unsigned int maxpools = nr_cpu_ids;
182 unsigned int pidx = 0;
183 unsigned int cpu;
184 int err;
186 err = svc_pool_map_alloc_arrays(m, maxpools);
187 if (err)
188 return err;
190 for_each_online_cpu(cpu) {
191 BUG_ON(pidx > maxpools);
192 m->to_pool[cpu] = pidx;
193 m->pool_to[pidx] = cpu;
194 pidx++;
196 /* cpus brought online later all get mapped to pool0, sorry */
198 return pidx;
203 * Initialise the pool map for SVC_POOL_PERNODE mode.
204 * Returns number of pools or <0 on error.
206 static int
207 svc_pool_map_init_pernode(struct svc_pool_map *m)
209 unsigned int maxpools = nr_node_ids;
210 unsigned int pidx = 0;
211 unsigned int node;
212 int err;
214 err = svc_pool_map_alloc_arrays(m, maxpools);
215 if (err)
216 return err;
218 for_each_node_with_cpus(node) {
219 /* some architectures (e.g. SN2) have cpuless nodes */
220 BUG_ON(pidx > maxpools);
221 m->to_pool[node] = pidx;
222 m->pool_to[pidx] = node;
223 pidx++;
225 /* nodes brought online later all get mapped to pool0, sorry */
227 return pidx;
232 * Add a reference to the global map of cpus to pools (and
233 * vice versa). Initialise the map if we're the first user.
234 * Returns the number of pools.
236 static unsigned int
237 svc_pool_map_get(void)
239 struct svc_pool_map *m = &svc_pool_map;
240 int npools = -1;
242 mutex_lock(&svc_pool_map_mutex);
244 if (m->count++) {
245 mutex_unlock(&svc_pool_map_mutex);
246 return m->npools;
249 if (m->mode == SVC_POOL_AUTO)
250 m->mode = svc_pool_map_choose_mode();
252 switch (m->mode) {
253 case SVC_POOL_PERCPU:
254 npools = svc_pool_map_init_percpu(m);
255 break;
256 case SVC_POOL_PERNODE:
257 npools = svc_pool_map_init_pernode(m);
258 break;
261 if (npools < 0) {
262 /* default, or memory allocation failure */
263 npools = 1;
264 m->mode = SVC_POOL_GLOBAL;
266 m->npools = npools;
268 mutex_unlock(&svc_pool_map_mutex);
269 return m->npools;
274 * Drop a reference to the global map of cpus to pools.
275 * When the last reference is dropped, the map data is
276 * freed; this allows the sysadmin to change the pool
277 * mode using the pool_mode module option without
278 * rebooting or re-loading sunrpc.ko.
280 static void
281 svc_pool_map_put(void)
283 struct svc_pool_map *m = &svc_pool_map;
285 mutex_lock(&svc_pool_map_mutex);
287 if (!--m->count) {
288 m->mode = SVC_POOL_DEFAULT;
289 kfree(m->to_pool);
290 kfree(m->pool_to);
291 m->npools = 0;
294 mutex_unlock(&svc_pool_map_mutex);
299 * Set the given thread's cpus_allowed mask so that it
300 * will only run on cpus in the given pool.
302 static inline void
303 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
305 struct svc_pool_map *m = &svc_pool_map;
306 unsigned int node = m->pool_to[pidx];
309 * The caller checks for sv_nrpools > 1, which
310 * implies that we've been initialized.
312 BUG_ON(m->count == 0);
314 switch (m->mode) {
315 case SVC_POOL_PERCPU:
317 set_cpus_allowed_ptr(task, cpumask_of(node));
318 break;
320 case SVC_POOL_PERNODE:
322 set_cpus_allowed_ptr(task, cpumask_of_node(node));
323 break;
329 * Use the mapping mode to choose a pool for a given CPU.
330 * Used when enqueueing an incoming RPC. Always returns
331 * a non-NULL pool pointer.
333 struct svc_pool *
334 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
336 struct svc_pool_map *m = &svc_pool_map;
337 unsigned int pidx = 0;
340 * An uninitialised map happens in a pure client when
341 * lockd is brought up, so silently treat it the
342 * same as SVC_POOL_GLOBAL.
344 if (svc_serv_is_pooled(serv)) {
345 switch (m->mode) {
346 case SVC_POOL_PERCPU:
347 pidx = m->to_pool[cpu];
348 break;
349 case SVC_POOL_PERNODE:
350 pidx = m->to_pool[cpu_to_node(cpu)];
351 break;
354 return &serv->sv_pools[pidx % serv->sv_nrpools];
359 * Create an RPC service
361 static struct svc_serv *
362 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
363 void (*shutdown)(struct svc_serv *serv))
365 struct svc_serv *serv;
366 unsigned int vers;
367 unsigned int xdrsize;
368 unsigned int i;
370 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
371 return NULL;
372 serv->sv_name = prog->pg_name;
373 serv->sv_program = prog;
374 serv->sv_nrthreads = 1;
375 serv->sv_stats = prog->pg_stats;
376 if (bufsize > RPCSVC_MAXPAYLOAD)
377 bufsize = RPCSVC_MAXPAYLOAD;
378 serv->sv_max_payload = bufsize? bufsize : 4096;
379 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
380 serv->sv_shutdown = shutdown;
381 xdrsize = 0;
382 while (prog) {
383 prog->pg_lovers = prog->pg_nvers-1;
384 for (vers=0; vers<prog->pg_nvers ; vers++)
385 if (prog->pg_vers[vers]) {
386 prog->pg_hivers = vers;
387 if (prog->pg_lovers > vers)
388 prog->pg_lovers = vers;
389 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
390 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
392 prog = prog->pg_next;
394 serv->sv_xdrsize = xdrsize;
395 INIT_LIST_HEAD(&serv->sv_tempsocks);
396 INIT_LIST_HEAD(&serv->sv_permsocks);
397 init_timer(&serv->sv_temptimer);
398 spin_lock_init(&serv->sv_lock);
400 serv->sv_nrpools = npools;
401 serv->sv_pools =
402 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
403 GFP_KERNEL);
404 if (!serv->sv_pools) {
405 kfree(serv);
406 return NULL;
409 for (i = 0; i < serv->sv_nrpools; i++) {
410 struct svc_pool *pool = &serv->sv_pools[i];
412 dprintk("svc: initialising pool %u for %s\n",
413 i, serv->sv_name);
415 pool->sp_id = i;
416 INIT_LIST_HEAD(&pool->sp_threads);
417 INIT_LIST_HEAD(&pool->sp_sockets);
418 INIT_LIST_HEAD(&pool->sp_all_threads);
419 spin_lock_init(&pool->sp_lock);
422 /* Remove any stale portmap registrations */
423 svc_unregister(serv);
425 return serv;
428 struct svc_serv *
429 svc_create(struct svc_program *prog, unsigned int bufsize,
430 void (*shutdown)(struct svc_serv *serv))
432 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
434 EXPORT_SYMBOL_GPL(svc_create);
436 struct svc_serv *
437 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
438 void (*shutdown)(struct svc_serv *serv),
439 svc_thread_fn func, struct module *mod)
441 struct svc_serv *serv;
442 unsigned int npools = svc_pool_map_get();
444 serv = __svc_create(prog, bufsize, npools, shutdown);
446 if (serv != NULL) {
447 serv->sv_function = func;
448 serv->sv_module = mod;
451 return serv;
453 EXPORT_SYMBOL_GPL(svc_create_pooled);
456 * Destroy an RPC service. Should be called with appropriate locking to
457 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
459 void
460 svc_destroy(struct svc_serv *serv)
462 dprintk("svc: svc_destroy(%s, %d)\n",
463 serv->sv_program->pg_name,
464 serv->sv_nrthreads);
466 if (serv->sv_nrthreads) {
467 if (--(serv->sv_nrthreads) != 0) {
468 svc_sock_update_bufs(serv);
469 return;
471 } else
472 printk("svc_destroy: no threads for serv=%p!\n", serv);
474 del_timer_sync(&serv->sv_temptimer);
476 svc_close_all(&serv->sv_tempsocks);
478 if (serv->sv_shutdown)
479 serv->sv_shutdown(serv);
481 svc_close_all(&serv->sv_permsocks);
483 BUG_ON(!list_empty(&serv->sv_permsocks));
484 BUG_ON(!list_empty(&serv->sv_tempsocks));
486 cache_clean_deferred(serv);
488 if (svc_serv_is_pooled(serv))
489 svc_pool_map_put();
491 svc_unregister(serv);
492 kfree(serv->sv_pools);
493 kfree(serv);
495 EXPORT_SYMBOL_GPL(svc_destroy);
498 * Allocate an RPC server's buffer space.
499 * We allocate pages and place them in rq_argpages.
501 static int
502 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
504 unsigned int pages, arghi;
506 /* bc_xprt uses fore channel allocated buffers */
507 if (svc_is_backchannel(rqstp))
508 return 1;
510 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
511 * We assume one is at most one page
513 arghi = 0;
514 BUG_ON(pages > RPCSVC_MAXPAGES);
515 while (pages) {
516 struct page *p = alloc_page(GFP_KERNEL);
517 if (!p)
518 break;
519 rqstp->rq_pages[arghi++] = p;
520 pages--;
522 return pages == 0;
526 * Release an RPC server buffer
528 static void
529 svc_release_buffer(struct svc_rqst *rqstp)
531 unsigned int i;
533 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
534 if (rqstp->rq_pages[i])
535 put_page(rqstp->rq_pages[i]);
538 struct svc_rqst *
539 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
541 struct svc_rqst *rqstp;
543 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
544 if (!rqstp)
545 goto out_enomem;
547 init_waitqueue_head(&rqstp->rq_wait);
549 serv->sv_nrthreads++;
550 spin_lock_bh(&pool->sp_lock);
551 pool->sp_nrthreads++;
552 list_add(&rqstp->rq_all, &pool->sp_all_threads);
553 spin_unlock_bh(&pool->sp_lock);
554 rqstp->rq_server = serv;
555 rqstp->rq_pool = pool;
557 rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
558 if (!rqstp->rq_argp)
559 goto out_thread;
561 rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
562 if (!rqstp->rq_resp)
563 goto out_thread;
565 if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
566 goto out_thread;
568 return rqstp;
569 out_thread:
570 svc_exit_thread(rqstp);
571 out_enomem:
572 return ERR_PTR(-ENOMEM);
574 EXPORT_SYMBOL_GPL(svc_prepare_thread);
577 * Choose a pool in which to create a new thread, for svc_set_num_threads
579 static inline struct svc_pool *
580 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
582 if (pool != NULL)
583 return pool;
585 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
589 * Choose a thread to kill, for svc_set_num_threads
591 static inline struct task_struct *
592 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
594 unsigned int i;
595 struct task_struct *task = NULL;
597 if (pool != NULL) {
598 spin_lock_bh(&pool->sp_lock);
599 } else {
600 /* choose a pool in round-robin fashion */
601 for (i = 0; i < serv->sv_nrpools; i++) {
602 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
603 spin_lock_bh(&pool->sp_lock);
604 if (!list_empty(&pool->sp_all_threads))
605 goto found_pool;
606 spin_unlock_bh(&pool->sp_lock);
608 return NULL;
611 found_pool:
612 if (!list_empty(&pool->sp_all_threads)) {
613 struct svc_rqst *rqstp;
616 * Remove from the pool->sp_all_threads list
617 * so we don't try to kill it again.
619 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
620 list_del_init(&rqstp->rq_all);
621 task = rqstp->rq_task;
623 spin_unlock_bh(&pool->sp_lock);
625 return task;
629 * Create or destroy enough new threads to make the number
630 * of threads the given number. If `pool' is non-NULL, applies
631 * only to threads in that pool, otherwise round-robins between
632 * all pools. Must be called with a svc_get() reference and
633 * the BKL or another lock to protect access to svc_serv fields.
635 * Destroying threads relies on the service threads filling in
636 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
637 * has been created using svc_create_pooled().
639 * Based on code that used to be in nfsd_svc() but tweaked
640 * to be pool-aware.
643 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
645 struct svc_rqst *rqstp;
646 struct task_struct *task;
647 struct svc_pool *chosen_pool;
648 int error = 0;
649 unsigned int state = serv->sv_nrthreads-1;
651 if (pool == NULL) {
652 /* The -1 assumes caller has done a svc_get() */
653 nrservs -= (serv->sv_nrthreads-1);
654 } else {
655 spin_lock_bh(&pool->sp_lock);
656 nrservs -= pool->sp_nrthreads;
657 spin_unlock_bh(&pool->sp_lock);
660 /* create new threads */
661 while (nrservs > 0) {
662 nrservs--;
663 chosen_pool = choose_pool(serv, pool, &state);
665 rqstp = svc_prepare_thread(serv, chosen_pool);
666 if (IS_ERR(rqstp)) {
667 error = PTR_ERR(rqstp);
668 break;
671 __module_get(serv->sv_module);
672 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
673 if (IS_ERR(task)) {
674 error = PTR_ERR(task);
675 module_put(serv->sv_module);
676 svc_exit_thread(rqstp);
677 break;
680 rqstp->rq_task = task;
681 if (serv->sv_nrpools > 1)
682 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
684 svc_sock_update_bufs(serv);
685 wake_up_process(task);
687 /* destroy old threads */
688 while (nrservs < 0 &&
689 (task = choose_victim(serv, pool, &state)) != NULL) {
690 send_sig(SIGINT, task, 1);
691 nrservs++;
694 return error;
696 EXPORT_SYMBOL_GPL(svc_set_num_threads);
699 * Called from a server thread as it's exiting. Caller must hold the BKL or
700 * the "service mutex", whichever is appropriate for the service.
702 void
703 svc_exit_thread(struct svc_rqst *rqstp)
705 struct svc_serv *serv = rqstp->rq_server;
706 struct svc_pool *pool = rqstp->rq_pool;
708 svc_release_buffer(rqstp);
709 kfree(rqstp->rq_resp);
710 kfree(rqstp->rq_argp);
711 kfree(rqstp->rq_auth_data);
713 spin_lock_bh(&pool->sp_lock);
714 pool->sp_nrthreads--;
715 list_del(&rqstp->rq_all);
716 spin_unlock_bh(&pool->sp_lock);
718 kfree(rqstp);
720 /* Release the server */
721 if (serv)
722 svc_destroy(serv);
724 EXPORT_SYMBOL_GPL(svc_exit_thread);
727 * Register an "inet" protocol family netid with the local
728 * rpcbind daemon via an rpcbind v4 SET request.
730 * No netconfig infrastructure is available in the kernel, so
731 * we map IP_ protocol numbers to netids by hand.
733 * Returns zero on success; a negative errno value is returned
734 * if any error occurs.
736 static int __svc_rpcb_register4(const u32 program, const u32 version,
737 const unsigned short protocol,
738 const unsigned short port)
740 const struct sockaddr_in sin = {
741 .sin_family = AF_INET,
742 .sin_addr.s_addr = htonl(INADDR_ANY),
743 .sin_port = htons(port),
745 const char *netid;
746 int error;
748 switch (protocol) {
749 case IPPROTO_UDP:
750 netid = RPCBIND_NETID_UDP;
751 break;
752 case IPPROTO_TCP:
753 netid = RPCBIND_NETID_TCP;
754 break;
755 default:
756 return -ENOPROTOOPT;
759 error = rpcb_v4_register(program, version,
760 (const struct sockaddr *)&sin, netid);
763 * User space didn't support rpcbind v4, so retry this
764 * registration request with the legacy rpcbind v2 protocol.
766 if (error == -EPROTONOSUPPORT)
767 error = rpcb_register(program, version, protocol, port);
769 return error;
772 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
774 * Register an "inet6" protocol family netid with the local
775 * rpcbind daemon via an rpcbind v4 SET request.
777 * No netconfig infrastructure is available in the kernel, so
778 * we map IP_ protocol numbers to netids by hand.
780 * Returns zero on success; a negative errno value is returned
781 * if any error occurs.
783 static int __svc_rpcb_register6(const u32 program, const u32 version,
784 const unsigned short protocol,
785 const unsigned short port)
787 const struct sockaddr_in6 sin6 = {
788 .sin6_family = AF_INET6,
789 .sin6_addr = IN6ADDR_ANY_INIT,
790 .sin6_port = htons(port),
792 const char *netid;
793 int error;
795 switch (protocol) {
796 case IPPROTO_UDP:
797 netid = RPCBIND_NETID_UDP6;
798 break;
799 case IPPROTO_TCP:
800 netid = RPCBIND_NETID_TCP6;
801 break;
802 default:
803 return -ENOPROTOOPT;
806 error = rpcb_v4_register(program, version,
807 (const struct sockaddr *)&sin6, netid);
810 * User space didn't support rpcbind version 4, so we won't
811 * use a PF_INET6 listener.
813 if (error == -EPROTONOSUPPORT)
814 error = -EAFNOSUPPORT;
816 return error;
818 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
821 * Register a kernel RPC service via rpcbind version 4.
823 * Returns zero on success; a negative errno value is returned
824 * if any error occurs.
826 static int __svc_register(const char *progname,
827 const u32 program, const u32 version,
828 const int family,
829 const unsigned short protocol,
830 const unsigned short port)
832 int error = -EAFNOSUPPORT;
834 switch (family) {
835 case PF_INET:
836 error = __svc_rpcb_register4(program, version,
837 protocol, port);
838 break;
839 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
840 case PF_INET6:
841 error = __svc_rpcb_register6(program, version,
842 protocol, port);
843 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
846 if (error < 0)
847 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
848 "service (errno %d).\n", progname, version, -error);
849 return error;
853 * svc_register - register an RPC service with the local portmapper
854 * @serv: svc_serv struct for the service to register
855 * @family: protocol family of service's listener socket
856 * @proto: transport protocol number to advertise
857 * @port: port to advertise
859 * Service is registered for any address in the passed-in protocol family
861 int svc_register(const struct svc_serv *serv, const int family,
862 const unsigned short proto, const unsigned short port)
864 struct svc_program *progp;
865 unsigned int i;
866 int error = 0;
868 BUG_ON(proto == 0 && port == 0);
870 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
871 for (i = 0; i < progp->pg_nvers; i++) {
872 if (progp->pg_vers[i] == NULL)
873 continue;
875 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
876 progp->pg_name,
878 proto == IPPROTO_UDP? "udp" : "tcp",
879 port,
880 family,
881 progp->pg_vers[i]->vs_hidden?
882 " (but not telling portmap)" : "");
884 if (progp->pg_vers[i]->vs_hidden)
885 continue;
887 error = __svc_register(progp->pg_name, progp->pg_prog,
888 i, family, proto, port);
889 if (error < 0)
890 break;
894 return error;
898 * If user space is running rpcbind, it should take the v4 UNSET
899 * and clear everything for this [program, version]. If user space
900 * is running portmap, it will reject the v4 UNSET, but won't have
901 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
902 * in this case to clear all existing entries for [program, version].
904 static void __svc_unregister(const u32 program, const u32 version,
905 const char *progname)
907 int error;
909 error = rpcb_v4_register(program, version, NULL, "");
912 * User space didn't support rpcbind v4, so retry this
913 * request with the legacy rpcbind v2 protocol.
915 if (error == -EPROTONOSUPPORT)
916 error = rpcb_register(program, version, 0, 0);
918 dprintk("svc: %s(%sv%u), error %d\n",
919 __func__, progname, version, error);
923 * All netids, bind addresses and ports registered for [program, version]
924 * are removed from the local rpcbind database (if the service is not
925 * hidden) to make way for a new instance of the service.
927 * The result of unregistration is reported via dprintk for those who want
928 * verification of the result, but is otherwise not important.
930 static void svc_unregister(const struct svc_serv *serv)
932 struct svc_program *progp;
933 unsigned long flags;
934 unsigned int i;
936 clear_thread_flag(TIF_SIGPENDING);
938 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
939 for (i = 0; i < progp->pg_nvers; i++) {
940 if (progp->pg_vers[i] == NULL)
941 continue;
942 if (progp->pg_vers[i]->vs_hidden)
943 continue;
945 __svc_unregister(progp->pg_prog, i, progp->pg_name);
949 spin_lock_irqsave(&current->sighand->siglock, flags);
950 recalc_sigpending();
951 spin_unlock_irqrestore(&current->sighand->siglock, flags);
955 * Printk the given error with the address of the client that caused it.
957 static int
958 __attribute__ ((format (printf, 2, 3)))
959 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
961 va_list args;
962 int r;
963 char buf[RPC_MAX_ADDRBUFLEN];
965 if (!net_ratelimit())
966 return 0;
968 printk(KERN_WARNING "svc: %s: ",
969 svc_print_addr(rqstp, buf, sizeof(buf)));
971 va_start(args, fmt);
972 r = vprintk(fmt, args);
973 va_end(args);
975 return r;
979 * Common routine for processing the RPC request.
981 static int
982 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
984 struct svc_program *progp;
985 struct svc_version *versp = NULL; /* compiler food */
986 struct svc_procedure *procp = NULL;
987 struct svc_serv *serv = rqstp->rq_server;
988 kxdrproc_t xdr;
989 __be32 *statp;
990 u32 prog, vers, proc;
991 __be32 auth_stat, rpc_stat;
992 int auth_res;
993 __be32 *reply_statp;
995 rpc_stat = rpc_success;
997 if (argv->iov_len < 6*4)
998 goto err_short_len;
1000 /* Will be turned off only in gss privacy case: */
1001 rqstp->rq_splice_ok = 1;
1002 /* Will be turned off only when NFSv4 Sessions are used */
1003 rqstp->rq_usedeferral = 1;
1004 rqstp->rq_dropme = false;
1006 /* Setup reply header */
1007 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1009 svc_putu32(resv, rqstp->rq_xid);
1011 vers = svc_getnl(argv);
1013 /* First words of reply: */
1014 svc_putnl(resv, 1); /* REPLY */
1016 if (vers != 2) /* RPC version number */
1017 goto err_bad_rpc;
1019 /* Save position in case we later decide to reject: */
1020 reply_statp = resv->iov_base + resv->iov_len;
1022 svc_putnl(resv, 0); /* ACCEPT */
1024 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1025 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1026 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1028 progp = serv->sv_program;
1030 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1031 if (prog == progp->pg_prog)
1032 break;
1035 * Decode auth data, and add verifier to reply buffer.
1036 * We do this before anything else in order to get a decent
1037 * auth verifier.
1039 auth_res = svc_authenticate(rqstp, &auth_stat);
1040 /* Also give the program a chance to reject this call: */
1041 if (auth_res == SVC_OK && progp) {
1042 auth_stat = rpc_autherr_badcred;
1043 auth_res = progp->pg_authenticate(rqstp);
1045 switch (auth_res) {
1046 case SVC_OK:
1047 break;
1048 case SVC_GARBAGE:
1049 goto err_garbage;
1050 case SVC_SYSERR:
1051 rpc_stat = rpc_system_err;
1052 goto err_bad;
1053 case SVC_DENIED:
1054 goto err_bad_auth;
1055 case SVC_CLOSE:
1056 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1057 svc_close_xprt(rqstp->rq_xprt);
1058 case SVC_DROP:
1059 goto dropit;
1060 case SVC_COMPLETE:
1061 goto sendit;
1064 if (progp == NULL)
1065 goto err_bad_prog;
1067 if (vers >= progp->pg_nvers ||
1068 !(versp = progp->pg_vers[vers]))
1069 goto err_bad_vers;
1071 procp = versp->vs_proc + proc;
1072 if (proc >= versp->vs_nproc || !procp->pc_func)
1073 goto err_bad_proc;
1074 rqstp->rq_procinfo = procp;
1076 /* Syntactic check complete */
1077 serv->sv_stats->rpccnt++;
1079 /* Build the reply header. */
1080 statp = resv->iov_base +resv->iov_len;
1081 svc_putnl(resv, RPC_SUCCESS);
1083 /* Bump per-procedure stats counter */
1084 procp->pc_count++;
1086 /* Initialize storage for argp and resp */
1087 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1088 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1090 /* un-reserve some of the out-queue now that we have a
1091 * better idea of reply size
1093 if (procp->pc_xdrressize)
1094 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1096 /* Call the function that processes the request. */
1097 if (!versp->vs_dispatch) {
1098 /* Decode arguments */
1099 xdr = procp->pc_decode;
1100 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1101 goto err_garbage;
1103 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1105 /* Encode reply */
1106 if (rqstp->rq_dropme) {
1107 if (procp->pc_release)
1108 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1109 goto dropit;
1111 if (*statp == rpc_success &&
1112 (xdr = procp->pc_encode) &&
1113 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1114 dprintk("svc: failed to encode reply\n");
1115 /* serv->sv_stats->rpcsystemerr++; */
1116 *statp = rpc_system_err;
1118 } else {
1119 dprintk("svc: calling dispatcher\n");
1120 if (!versp->vs_dispatch(rqstp, statp)) {
1121 /* Release reply info */
1122 if (procp->pc_release)
1123 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1124 goto dropit;
1128 /* Check RPC status result */
1129 if (*statp != rpc_success)
1130 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1132 /* Release reply info */
1133 if (procp->pc_release)
1134 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1136 if (procp->pc_encode == NULL)
1137 goto dropit;
1139 sendit:
1140 if (svc_authorise(rqstp))
1141 goto dropit;
1142 return 1; /* Caller can now send it */
1144 dropit:
1145 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1146 dprintk("svc: svc_process dropit\n");
1147 return 0;
1149 err_short_len:
1150 svc_printk(rqstp, "short len %Zd, dropping request\n",
1151 argv->iov_len);
1153 goto dropit; /* drop request */
1155 err_bad_rpc:
1156 serv->sv_stats->rpcbadfmt++;
1157 svc_putnl(resv, 1); /* REJECT */
1158 svc_putnl(resv, 0); /* RPC_MISMATCH */
1159 svc_putnl(resv, 2); /* Only RPCv2 supported */
1160 svc_putnl(resv, 2);
1161 goto sendit;
1163 err_bad_auth:
1164 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1165 serv->sv_stats->rpcbadauth++;
1166 /* Restore write pointer to location of accept status: */
1167 xdr_ressize_check(rqstp, reply_statp);
1168 svc_putnl(resv, 1); /* REJECT */
1169 svc_putnl(resv, 1); /* AUTH_ERROR */
1170 svc_putnl(resv, ntohl(auth_stat)); /* status */
1171 goto sendit;
1173 err_bad_prog:
1174 dprintk("svc: unknown program %d\n", prog);
1175 serv->sv_stats->rpcbadfmt++;
1176 svc_putnl(resv, RPC_PROG_UNAVAIL);
1177 goto sendit;
1179 err_bad_vers:
1180 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1181 vers, prog, progp->pg_name);
1183 serv->sv_stats->rpcbadfmt++;
1184 svc_putnl(resv, RPC_PROG_MISMATCH);
1185 svc_putnl(resv, progp->pg_lovers);
1186 svc_putnl(resv, progp->pg_hivers);
1187 goto sendit;
1189 err_bad_proc:
1190 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1192 serv->sv_stats->rpcbadfmt++;
1193 svc_putnl(resv, RPC_PROC_UNAVAIL);
1194 goto sendit;
1196 err_garbage:
1197 svc_printk(rqstp, "failed to decode args\n");
1199 rpc_stat = rpc_garbage_args;
1200 err_bad:
1201 serv->sv_stats->rpcbadfmt++;
1202 svc_putnl(resv, ntohl(rpc_stat));
1203 goto sendit;
1205 EXPORT_SYMBOL_GPL(svc_process);
1208 * Process the RPC request.
1211 svc_process(struct svc_rqst *rqstp)
1213 struct kvec *argv = &rqstp->rq_arg.head[0];
1214 struct kvec *resv = &rqstp->rq_res.head[0];
1215 struct svc_serv *serv = rqstp->rq_server;
1216 u32 dir;
1219 * Setup response xdr_buf.
1220 * Initially it has just one page
1222 rqstp->rq_resused = 1;
1223 resv->iov_base = page_address(rqstp->rq_respages[0]);
1224 resv->iov_len = 0;
1225 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1226 rqstp->rq_res.len = 0;
1227 rqstp->rq_res.page_base = 0;
1228 rqstp->rq_res.page_len = 0;
1229 rqstp->rq_res.buflen = PAGE_SIZE;
1230 rqstp->rq_res.tail[0].iov_base = NULL;
1231 rqstp->rq_res.tail[0].iov_len = 0;
1233 rqstp->rq_xid = svc_getu32(argv);
1235 dir = svc_getnl(argv);
1236 if (dir != 0) {
1237 /* direction != CALL */
1238 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1239 serv->sv_stats->rpcbadfmt++;
1240 svc_drop(rqstp);
1241 return 0;
1244 /* Returns 1 for send, 0 for drop */
1245 if (svc_process_common(rqstp, argv, resv))
1246 return svc_send(rqstp);
1247 else {
1248 svc_drop(rqstp);
1249 return 0;
1253 #if defined(CONFIG_NFS_V4_1)
1255 * Process a backchannel RPC request that arrived over an existing
1256 * outbound connection
1259 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1260 struct svc_rqst *rqstp)
1262 struct kvec *argv = &rqstp->rq_arg.head[0];
1263 struct kvec *resv = &rqstp->rq_res.head[0];
1265 /* Build the svc_rqst used by the common processing routine */
1266 rqstp->rq_xprt = serv->sv_bc_xprt;
1267 rqstp->rq_xid = req->rq_xid;
1268 rqstp->rq_prot = req->rq_xprt->prot;
1269 rqstp->rq_server = serv;
1271 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1272 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1273 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1274 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1276 /* reset result send buffer "put" position */
1277 resv->iov_len = 0;
1279 if (rqstp->rq_prot != IPPROTO_TCP) {
1280 printk(KERN_ERR "No support for Non-TCP transports!\n");
1281 BUG();
1285 * Skip the next two words because they've already been
1286 * processed in the trasport
1288 svc_getu32(argv); /* XID */
1289 svc_getnl(argv); /* CALLDIR */
1291 /* Returns 1 for send, 0 for drop */
1292 if (svc_process_common(rqstp, argv, resv)) {
1293 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1294 sizeof(req->rq_snd_buf));
1295 return bc_send(req);
1296 } else {
1297 /* Nothing to do to drop request */
1298 return 0;
1301 EXPORT_SYMBOL(bc_svc_process);
1302 #endif /* CONFIG_NFS_V4_1 */
1305 * Return (transport-specific) limit on the rpc payload.
1307 u32 svc_max_payload(const struct svc_rqst *rqstp)
1309 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1311 if (rqstp->rq_server->sv_max_payload < max)
1312 max = rqstp->rq_server->sv_max_payload;
1313 return max;
1315 EXPORT_SYMBOL_GPL(svc_max_payload);