2 * linux/net/sunrpc/svc.c
4 * High-level RPC service routines
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.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.
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" */
59 unsigned int *pool_to; /* maps pool id to cpu or node */
60 unsigned int *to_pool; /* maps cpu or node to pool id */
63 .mode = SVC_POOL_DEFAULT
65 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
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;
74 mutex_lock(&svc_pool_map_mutex);
81 if (!strncmp(val, "auto", 4))
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;
93 mutex_unlock(&svc_pool_map_mutex);
98 param_get_pool_mode(char *buf, struct kernel_param *kp)
100 int *ip = (int *)kp->arg;
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);
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.
125 svc_pool_map_choose_mode(void)
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.
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);
162 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
175 * Initialise the pool map for SVC_POOL_PERCPU mode.
176 * Returns number of pools or <0 on error.
179 svc_pool_map_init_percpu(struct svc_pool_map *m)
181 unsigned int maxpools = nr_cpu_ids;
182 unsigned int pidx = 0;
186 err = svc_pool_map_alloc_arrays(m, maxpools);
190 for_each_online_cpu(cpu) {
191 BUG_ON(pidx > maxpools);
192 m->to_pool[cpu] = pidx;
193 m->pool_to[pidx] = cpu;
196 /* cpus brought online later all get mapped to pool0, sorry */
203 * Initialise the pool map for SVC_POOL_PERNODE mode.
204 * Returns number of pools or <0 on error.
207 svc_pool_map_init_pernode(struct svc_pool_map *m)
209 unsigned int maxpools = nr_node_ids;
210 unsigned int pidx = 0;
214 err = svc_pool_map_alloc_arrays(m, maxpools);
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;
225 /* nodes brought online later all get mapped to pool0, sorry */
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.
237 svc_pool_map_get(void)
239 struct svc_pool_map *m = &svc_pool_map;
242 mutex_lock(&svc_pool_map_mutex);
245 mutex_unlock(&svc_pool_map_mutex);
249 if (m->mode == SVC_POOL_AUTO)
250 m->mode = svc_pool_map_choose_mode();
253 case SVC_POOL_PERCPU:
254 npools = svc_pool_map_init_percpu(m);
256 case SVC_POOL_PERNODE:
257 npools = svc_pool_map_init_pernode(m);
262 /* default, or memory allocation failure */
264 m->mode = SVC_POOL_GLOBAL;
268 mutex_unlock(&svc_pool_map_mutex);
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.
281 svc_pool_map_put(void)
283 struct svc_pool_map *m = &svc_pool_map;
285 mutex_lock(&svc_pool_map_mutex);
288 m->mode = SVC_POOL_DEFAULT;
294 mutex_unlock(&svc_pool_map_mutex);
298 static int svc_pool_map_get_node(unsigned int pidx)
300 const struct svc_pool_map *m = &svc_pool_map;
303 if (m->mode == SVC_POOL_PERCPU)
304 return cpu_to_node(m->pool_to[pidx]);
305 if (m->mode == SVC_POOL_PERNODE)
306 return m->pool_to[pidx];
311 * Set the given thread's cpus_allowed mask so that it
312 * will only run on cpus in the given pool.
315 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
317 struct svc_pool_map *m = &svc_pool_map;
318 unsigned int node = m->pool_to[pidx];
321 * The caller checks for sv_nrpools > 1, which
322 * implies that we've been initialized.
324 BUG_ON(m->count == 0);
327 case SVC_POOL_PERCPU:
329 set_cpus_allowed_ptr(task, cpumask_of(node));
332 case SVC_POOL_PERNODE:
334 set_cpus_allowed_ptr(task, cpumask_of_node(node));
341 * Use the mapping mode to choose a pool for a given CPU.
342 * Used when enqueueing an incoming RPC. Always returns
343 * a non-NULL pool pointer.
346 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
348 struct svc_pool_map *m = &svc_pool_map;
349 unsigned int pidx = 0;
352 * An uninitialised map happens in a pure client when
353 * lockd is brought up, so silently treat it the
354 * same as SVC_POOL_GLOBAL.
356 if (svc_serv_is_pooled(serv)) {
358 case SVC_POOL_PERCPU:
359 pidx = m->to_pool[cpu];
361 case SVC_POOL_PERNODE:
362 pidx = m->to_pool[cpu_to_node(cpu)];
366 return &serv->sv_pools[pidx % serv->sv_nrpools];
369 static int svc_rpcb_setup(struct svc_serv *serv)
373 err = rpcb_create_local();
377 /* Remove any stale portmap registrations */
378 svc_unregister(serv);
382 void svc_rpcb_cleanup(struct svc_serv *serv)
384 svc_unregister(serv);
387 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
389 static int svc_uses_rpcbind(struct svc_serv *serv)
391 struct svc_program *progp;
394 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
395 for (i = 0; i < progp->pg_nvers; i++) {
396 if (progp->pg_vers[i] == NULL)
398 if (progp->pg_vers[i]->vs_hidden == 0)
407 * Create an RPC service
409 static struct svc_serv *
410 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
411 void (*shutdown)(struct svc_serv *serv))
413 struct svc_serv *serv;
415 unsigned int xdrsize;
418 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
420 serv->sv_name = prog->pg_name;
421 serv->sv_program = prog;
422 serv->sv_nrthreads = 1;
423 serv->sv_stats = prog->pg_stats;
424 if (bufsize > RPCSVC_MAXPAYLOAD)
425 bufsize = RPCSVC_MAXPAYLOAD;
426 serv->sv_max_payload = bufsize? bufsize : 4096;
427 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
428 serv->sv_shutdown = shutdown;
431 prog->pg_lovers = prog->pg_nvers-1;
432 for (vers=0; vers<prog->pg_nvers ; vers++)
433 if (prog->pg_vers[vers]) {
434 prog->pg_hivers = vers;
435 if (prog->pg_lovers > vers)
436 prog->pg_lovers = vers;
437 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
438 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
440 prog = prog->pg_next;
442 serv->sv_xdrsize = xdrsize;
443 INIT_LIST_HEAD(&serv->sv_tempsocks);
444 INIT_LIST_HEAD(&serv->sv_permsocks);
445 init_timer(&serv->sv_temptimer);
446 spin_lock_init(&serv->sv_lock);
448 serv->sv_nrpools = npools;
450 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
452 if (!serv->sv_pools) {
457 for (i = 0; i < serv->sv_nrpools; i++) {
458 struct svc_pool *pool = &serv->sv_pools[i];
460 dprintk("svc: initialising pool %u for %s\n",
464 INIT_LIST_HEAD(&pool->sp_threads);
465 INIT_LIST_HEAD(&pool->sp_sockets);
466 INIT_LIST_HEAD(&pool->sp_all_threads);
467 spin_lock_init(&pool->sp_lock);
470 if (svc_uses_rpcbind(serv)) {
471 if (svc_rpcb_setup(serv) < 0) {
472 kfree(serv->sv_pools);
476 if (!serv->sv_shutdown)
477 serv->sv_shutdown = svc_rpcb_cleanup;
484 svc_create(struct svc_program *prog, unsigned int bufsize,
485 void (*shutdown)(struct svc_serv *serv))
487 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
489 EXPORT_SYMBOL_GPL(svc_create);
492 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
493 void (*shutdown)(struct svc_serv *serv),
494 svc_thread_fn func, struct module *mod)
496 struct svc_serv *serv;
497 unsigned int npools = svc_pool_map_get();
499 serv = __svc_create(prog, bufsize, npools, shutdown);
502 serv->sv_function = func;
503 serv->sv_module = mod;
508 EXPORT_SYMBOL_GPL(svc_create_pooled);
511 * Destroy an RPC service. Should be called with appropriate locking to
512 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
515 svc_destroy(struct svc_serv *serv)
517 dprintk("svc: svc_destroy(%s, %d)\n",
518 serv->sv_program->pg_name,
521 if (serv->sv_nrthreads) {
522 if (--(serv->sv_nrthreads) != 0) {
523 svc_sock_update_bufs(serv);
527 printk("svc_destroy: no threads for serv=%p!\n", serv);
529 del_timer_sync(&serv->sv_temptimer);
531 * The set of xprts (contained in the sv_tempsocks and
532 * sv_permsocks lists) is now constant, since it is modified
533 * only by accepting new sockets (done by service threads in
534 * svc_recv) or aging old ones (done by sv_temptimer), or
535 * configuration changes (excluded by whatever locking the
536 * caller is using--nfsd_mutex in the case of nfsd). So it's
537 * safe to traverse those lists and shut everything down:
541 if (serv->sv_shutdown)
542 serv->sv_shutdown(serv);
544 cache_clean_deferred(serv);
546 if (svc_serv_is_pooled(serv))
549 kfree(serv->sv_pools);
552 EXPORT_SYMBOL_GPL(svc_destroy);
555 * Allocate an RPC server's buffer space.
556 * We allocate pages and place them in rq_argpages.
559 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
561 unsigned int pages, arghi;
563 /* bc_xprt uses fore channel allocated buffers */
564 if (svc_is_backchannel(rqstp))
567 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
568 * We assume one is at most one page
571 BUG_ON(pages > RPCSVC_MAXPAGES);
573 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
576 rqstp->rq_pages[arghi++] = p;
583 * Release an RPC server buffer
586 svc_release_buffer(struct svc_rqst *rqstp)
590 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
591 if (rqstp->rq_pages[i])
592 put_page(rqstp->rq_pages[i]);
596 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
598 struct svc_rqst *rqstp;
600 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
604 init_waitqueue_head(&rqstp->rq_wait);
606 serv->sv_nrthreads++;
607 spin_lock_bh(&pool->sp_lock);
608 pool->sp_nrthreads++;
609 list_add(&rqstp->rq_all, &pool->sp_all_threads);
610 spin_unlock_bh(&pool->sp_lock);
611 rqstp->rq_server = serv;
612 rqstp->rq_pool = pool;
614 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
618 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
622 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
627 svc_exit_thread(rqstp);
629 return ERR_PTR(-ENOMEM);
631 EXPORT_SYMBOL_GPL(svc_prepare_thread);
634 * Choose a pool in which to create a new thread, for svc_set_num_threads
636 static inline struct svc_pool *
637 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
642 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
646 * Choose a thread to kill, for svc_set_num_threads
648 static inline struct task_struct *
649 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
652 struct task_struct *task = NULL;
655 spin_lock_bh(&pool->sp_lock);
657 /* choose a pool in round-robin fashion */
658 for (i = 0; i < serv->sv_nrpools; i++) {
659 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
660 spin_lock_bh(&pool->sp_lock);
661 if (!list_empty(&pool->sp_all_threads))
663 spin_unlock_bh(&pool->sp_lock);
669 if (!list_empty(&pool->sp_all_threads)) {
670 struct svc_rqst *rqstp;
673 * Remove from the pool->sp_all_threads list
674 * so we don't try to kill it again.
676 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
677 list_del_init(&rqstp->rq_all);
678 task = rqstp->rq_task;
680 spin_unlock_bh(&pool->sp_lock);
686 * Create or destroy enough new threads to make the number
687 * of threads the given number. If `pool' is non-NULL, applies
688 * only to threads in that pool, otherwise round-robins between
689 * all pools. Must be called with a svc_get() reference and
690 * the BKL or another lock to protect access to svc_serv fields.
692 * Destroying threads relies on the service threads filling in
693 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
694 * has been created using svc_create_pooled().
696 * Based on code that used to be in nfsd_svc() but tweaked
700 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
702 struct svc_rqst *rqstp;
703 struct task_struct *task;
704 struct svc_pool *chosen_pool;
706 unsigned int state = serv->sv_nrthreads-1;
710 /* The -1 assumes caller has done a svc_get() */
711 nrservs -= (serv->sv_nrthreads-1);
713 spin_lock_bh(&pool->sp_lock);
714 nrservs -= pool->sp_nrthreads;
715 spin_unlock_bh(&pool->sp_lock);
718 /* create new threads */
719 while (nrservs > 0) {
721 chosen_pool = choose_pool(serv, pool, &state);
723 node = svc_pool_map_get_node(chosen_pool->sp_id);
724 rqstp = svc_prepare_thread(serv, chosen_pool, node);
726 error = PTR_ERR(rqstp);
730 __module_get(serv->sv_module);
731 task = kthread_create_on_node(serv->sv_function, rqstp,
732 node, serv->sv_name);
734 error = PTR_ERR(task);
735 module_put(serv->sv_module);
736 svc_exit_thread(rqstp);
740 rqstp->rq_task = task;
741 if (serv->sv_nrpools > 1)
742 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
744 svc_sock_update_bufs(serv);
745 wake_up_process(task);
747 /* destroy old threads */
748 while (nrservs < 0 &&
749 (task = choose_victim(serv, pool, &state)) != NULL) {
750 send_sig(SIGINT, task, 1);
756 EXPORT_SYMBOL_GPL(svc_set_num_threads);
759 * Called from a server thread as it's exiting. Caller must hold the BKL or
760 * the "service mutex", whichever is appropriate for the service.
763 svc_exit_thread(struct svc_rqst *rqstp)
765 struct svc_serv *serv = rqstp->rq_server;
766 struct svc_pool *pool = rqstp->rq_pool;
768 svc_release_buffer(rqstp);
769 kfree(rqstp->rq_resp);
770 kfree(rqstp->rq_argp);
771 kfree(rqstp->rq_auth_data);
773 spin_lock_bh(&pool->sp_lock);
774 pool->sp_nrthreads--;
775 list_del(&rqstp->rq_all);
776 spin_unlock_bh(&pool->sp_lock);
780 /* Release the server */
784 EXPORT_SYMBOL_GPL(svc_exit_thread);
787 * Register an "inet" protocol family netid with the local
788 * rpcbind daemon via an rpcbind v4 SET request.
790 * No netconfig infrastructure is available in the kernel, so
791 * we map IP_ protocol numbers to netids by hand.
793 * Returns zero on success; a negative errno value is returned
794 * if any error occurs.
796 static int __svc_rpcb_register4(const u32 program, const u32 version,
797 const unsigned short protocol,
798 const unsigned short port)
800 const struct sockaddr_in sin = {
801 .sin_family = AF_INET,
802 .sin_addr.s_addr = htonl(INADDR_ANY),
803 .sin_port = htons(port),
810 netid = RPCBIND_NETID_UDP;
813 netid = RPCBIND_NETID_TCP;
819 error = rpcb_v4_register(program, version,
820 (const struct sockaddr *)&sin, netid);
823 * User space didn't support rpcbind v4, so retry this
824 * registration request with the legacy rpcbind v2 protocol.
826 if (error == -EPROTONOSUPPORT)
827 error = rpcb_register(program, version, protocol, port);
832 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
834 * Register an "inet6" protocol family netid with the local
835 * rpcbind daemon via an rpcbind v4 SET request.
837 * No netconfig infrastructure is available in the kernel, so
838 * we map IP_ protocol numbers to netids by hand.
840 * Returns zero on success; a negative errno value is returned
841 * if any error occurs.
843 static int __svc_rpcb_register6(const u32 program, const u32 version,
844 const unsigned short protocol,
845 const unsigned short port)
847 const struct sockaddr_in6 sin6 = {
848 .sin6_family = AF_INET6,
849 .sin6_addr = IN6ADDR_ANY_INIT,
850 .sin6_port = htons(port),
857 netid = RPCBIND_NETID_UDP6;
860 netid = RPCBIND_NETID_TCP6;
866 error = rpcb_v4_register(program, version,
867 (const struct sockaddr *)&sin6, netid);
870 * User space didn't support rpcbind version 4, so we won't
871 * use a PF_INET6 listener.
873 if (error == -EPROTONOSUPPORT)
874 error = -EAFNOSUPPORT;
878 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
881 * Register a kernel RPC service via rpcbind version 4.
883 * Returns zero on success; a negative errno value is returned
884 * if any error occurs.
886 static int __svc_register(const char *progname,
887 const u32 program, const u32 version,
889 const unsigned short protocol,
890 const unsigned short port)
892 int error = -EAFNOSUPPORT;
896 error = __svc_rpcb_register4(program, version,
899 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
901 error = __svc_rpcb_register6(program, version,
903 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
907 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
908 "service (errno %d).\n", progname, version, -error);
913 * svc_register - register an RPC service with the local portmapper
914 * @serv: svc_serv struct for the service to register
915 * @family: protocol family of service's listener socket
916 * @proto: transport protocol number to advertise
917 * @port: port to advertise
919 * Service is registered for any address in the passed-in protocol family
921 int svc_register(const struct svc_serv *serv, const int family,
922 const unsigned short proto, const unsigned short port)
924 struct svc_program *progp;
928 BUG_ON(proto == 0 && port == 0);
930 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
931 for (i = 0; i < progp->pg_nvers; i++) {
932 if (progp->pg_vers[i] == NULL)
935 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
938 proto == IPPROTO_UDP? "udp" : "tcp",
941 progp->pg_vers[i]->vs_hidden?
942 " (but not telling portmap)" : "");
944 if (progp->pg_vers[i]->vs_hidden)
947 error = __svc_register(progp->pg_name, progp->pg_prog,
948 i, family, proto, port);
958 * If user space is running rpcbind, it should take the v4 UNSET
959 * and clear everything for this [program, version]. If user space
960 * is running portmap, it will reject the v4 UNSET, but won't have
961 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
962 * in this case to clear all existing entries for [program, version].
964 static void __svc_unregister(const u32 program, const u32 version,
965 const char *progname)
969 error = rpcb_v4_register(program, version, NULL, "");
972 * User space didn't support rpcbind v4, so retry this
973 * request with the legacy rpcbind v2 protocol.
975 if (error == -EPROTONOSUPPORT)
976 error = rpcb_register(program, version, 0, 0);
978 dprintk("svc: %s(%sv%u), error %d\n",
979 __func__, progname, version, error);
983 * All netids, bind addresses and ports registered for [program, version]
984 * are removed from the local rpcbind database (if the service is not
985 * hidden) to make way for a new instance of the service.
987 * The result of unregistration is reported via dprintk for those who want
988 * verification of the result, but is otherwise not important.
990 static void svc_unregister(const struct svc_serv *serv)
992 struct svc_program *progp;
996 clear_thread_flag(TIF_SIGPENDING);
998 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
999 for (i = 0; i < progp->pg_nvers; i++) {
1000 if (progp->pg_vers[i] == NULL)
1002 if (progp->pg_vers[i]->vs_hidden)
1005 dprintk("svc: attempting to unregister %sv%u\n",
1007 __svc_unregister(progp->pg_prog, i, progp->pg_name);
1011 spin_lock_irqsave(¤t->sighand->siglock, flags);
1012 recalc_sigpending();
1013 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1017 * Printk the given error with the address of the client that caused it.
1019 static __printf(2, 3)
1020 int svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1024 char buf[RPC_MAX_ADDRBUFLEN];
1026 if (!net_ratelimit())
1029 printk(KERN_WARNING "svc: %s: ",
1030 svc_print_addr(rqstp, buf, sizeof(buf)));
1032 va_start(args, fmt);
1033 r = vprintk(fmt, args);
1040 * Common routine for processing the RPC request.
1043 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1045 struct svc_program *progp;
1046 struct svc_version *versp = NULL; /* compiler food */
1047 struct svc_procedure *procp = NULL;
1048 struct svc_serv *serv = rqstp->rq_server;
1051 u32 prog, vers, proc;
1052 __be32 auth_stat, rpc_stat;
1054 __be32 *reply_statp;
1056 rpc_stat = rpc_success;
1058 if (argv->iov_len < 6*4)
1061 /* Will be turned off only in gss privacy case: */
1062 rqstp->rq_splice_ok = 1;
1063 /* Will be turned off only when NFSv4 Sessions are used */
1064 rqstp->rq_usedeferral = 1;
1065 rqstp->rq_dropme = false;
1067 /* Setup reply header */
1068 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1070 svc_putu32(resv, rqstp->rq_xid);
1072 vers = svc_getnl(argv);
1074 /* First words of reply: */
1075 svc_putnl(resv, 1); /* REPLY */
1077 if (vers != 2) /* RPC version number */
1080 /* Save position in case we later decide to reject: */
1081 reply_statp = resv->iov_base + resv->iov_len;
1083 svc_putnl(resv, 0); /* ACCEPT */
1085 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1086 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1087 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1089 progp = serv->sv_program;
1091 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1092 if (prog == progp->pg_prog)
1096 * Decode auth data, and add verifier to reply buffer.
1097 * We do this before anything else in order to get a decent
1100 auth_res = svc_authenticate(rqstp, &auth_stat);
1101 /* Also give the program a chance to reject this call: */
1102 if (auth_res == SVC_OK && progp) {
1103 auth_stat = rpc_autherr_badcred;
1104 auth_res = progp->pg_authenticate(rqstp);
1112 rpc_stat = rpc_system_err;
1117 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1118 svc_close_xprt(rqstp->rq_xprt);
1128 if (vers >= progp->pg_nvers ||
1129 !(versp = progp->pg_vers[vers]))
1132 procp = versp->vs_proc + proc;
1133 if (proc >= versp->vs_nproc || !procp->pc_func)
1135 rqstp->rq_procinfo = procp;
1137 /* Syntactic check complete */
1138 serv->sv_stats->rpccnt++;
1140 /* Build the reply header. */
1141 statp = resv->iov_base +resv->iov_len;
1142 svc_putnl(resv, RPC_SUCCESS);
1144 /* Bump per-procedure stats counter */
1147 /* Initialize storage for argp and resp */
1148 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1149 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1151 /* un-reserve some of the out-queue now that we have a
1152 * better idea of reply size
1154 if (procp->pc_xdrressize)
1155 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1157 /* Call the function that processes the request. */
1158 if (!versp->vs_dispatch) {
1159 /* Decode arguments */
1160 xdr = procp->pc_decode;
1161 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1164 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1167 if (rqstp->rq_dropme) {
1168 if (procp->pc_release)
1169 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1172 if (*statp == rpc_success &&
1173 (xdr = procp->pc_encode) &&
1174 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1175 dprintk("svc: failed to encode reply\n");
1176 /* serv->sv_stats->rpcsystemerr++; */
1177 *statp = rpc_system_err;
1180 dprintk("svc: calling dispatcher\n");
1181 if (!versp->vs_dispatch(rqstp, statp)) {
1182 /* Release reply info */
1183 if (procp->pc_release)
1184 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1189 /* Check RPC status result */
1190 if (*statp != rpc_success)
1191 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1193 /* Release reply info */
1194 if (procp->pc_release)
1195 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1197 if (procp->pc_encode == NULL)
1201 if (svc_authorise(rqstp))
1203 return 1; /* Caller can now send it */
1206 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1207 dprintk("svc: svc_process dropit\n");
1211 svc_printk(rqstp, "short len %Zd, dropping request\n",
1214 goto dropit; /* drop request */
1217 serv->sv_stats->rpcbadfmt++;
1218 svc_putnl(resv, 1); /* REJECT */
1219 svc_putnl(resv, 0); /* RPC_MISMATCH */
1220 svc_putnl(resv, 2); /* Only RPCv2 supported */
1225 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1226 serv->sv_stats->rpcbadauth++;
1227 /* Restore write pointer to location of accept status: */
1228 xdr_ressize_check(rqstp, reply_statp);
1229 svc_putnl(resv, 1); /* REJECT */
1230 svc_putnl(resv, 1); /* AUTH_ERROR */
1231 svc_putnl(resv, ntohl(auth_stat)); /* status */
1235 dprintk("svc: unknown program %d\n", prog);
1236 serv->sv_stats->rpcbadfmt++;
1237 svc_putnl(resv, RPC_PROG_UNAVAIL);
1241 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1242 vers, prog, progp->pg_name);
1244 serv->sv_stats->rpcbadfmt++;
1245 svc_putnl(resv, RPC_PROG_MISMATCH);
1246 svc_putnl(resv, progp->pg_lovers);
1247 svc_putnl(resv, progp->pg_hivers);
1251 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1253 serv->sv_stats->rpcbadfmt++;
1254 svc_putnl(resv, RPC_PROC_UNAVAIL);
1258 svc_printk(rqstp, "failed to decode args\n");
1260 rpc_stat = rpc_garbage_args;
1262 serv->sv_stats->rpcbadfmt++;
1263 svc_putnl(resv, ntohl(rpc_stat));
1266 EXPORT_SYMBOL_GPL(svc_process);
1269 * Process the RPC request.
1272 svc_process(struct svc_rqst *rqstp)
1274 struct kvec *argv = &rqstp->rq_arg.head[0];
1275 struct kvec *resv = &rqstp->rq_res.head[0];
1276 struct svc_serv *serv = rqstp->rq_server;
1280 * Setup response xdr_buf.
1281 * Initially it has just one page
1283 rqstp->rq_resused = 1;
1284 resv->iov_base = page_address(rqstp->rq_respages[0]);
1286 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1287 rqstp->rq_res.len = 0;
1288 rqstp->rq_res.page_base = 0;
1289 rqstp->rq_res.page_len = 0;
1290 rqstp->rq_res.buflen = PAGE_SIZE;
1291 rqstp->rq_res.tail[0].iov_base = NULL;
1292 rqstp->rq_res.tail[0].iov_len = 0;
1294 rqstp->rq_xid = svc_getu32(argv);
1296 dir = svc_getnl(argv);
1298 /* direction != CALL */
1299 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1300 serv->sv_stats->rpcbadfmt++;
1305 /* Returns 1 for send, 0 for drop */
1306 if (svc_process_common(rqstp, argv, resv))
1307 return svc_send(rqstp);
1314 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1316 * Process a backchannel RPC request that arrived over an existing
1317 * outbound connection
1320 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1321 struct svc_rqst *rqstp)
1323 struct kvec *argv = &rqstp->rq_arg.head[0];
1324 struct kvec *resv = &rqstp->rq_res.head[0];
1326 /* Build the svc_rqst used by the common processing routine */
1327 rqstp->rq_xprt = serv->sv_bc_xprt;
1328 rqstp->rq_xid = req->rq_xid;
1329 rqstp->rq_prot = req->rq_xprt->prot;
1330 rqstp->rq_server = serv;
1332 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1333 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1334 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1335 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1337 /* reset result send buffer "put" position */
1340 if (rqstp->rq_prot != IPPROTO_TCP) {
1341 printk(KERN_ERR "No support for Non-TCP transports!\n");
1346 * Skip the next two words because they've already been
1347 * processed in the trasport
1349 svc_getu32(argv); /* XID */
1350 svc_getnl(argv); /* CALLDIR */
1352 /* Returns 1 for send, 0 for drop */
1353 if (svc_process_common(rqstp, argv, resv)) {
1354 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1355 sizeof(req->rq_snd_buf));
1356 return bc_send(req);
1358 /* Nothing to do to drop request */
1362 EXPORT_SYMBOL_GPL(bc_svc_process);
1363 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1366 * Return (transport-specific) limit on the rpc payload.
1368 u32 svc_max_payload(const struct svc_rqst *rqstp)
1370 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1372 if (rqstp->rq_server->sv_max_payload < max)
1373 max = rqstp->rq_server->sv_max_payload;
1376 EXPORT_SYMBOL_GPL(svc_max_payload);