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, struct net *net);
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);
176 * Initialise the pool map for SVC_POOL_PERCPU mode.
177 * Returns number of pools or <0 on error.
180 svc_pool_map_init_percpu(struct svc_pool_map *m)
182 unsigned int maxpools = nr_cpu_ids;
183 unsigned int pidx = 0;
187 err = svc_pool_map_alloc_arrays(m, maxpools);
191 for_each_online_cpu(cpu) {
192 BUG_ON(pidx > maxpools);
193 m->to_pool[cpu] = pidx;
194 m->pool_to[pidx] = cpu;
197 /* cpus brought online later all get mapped to pool0, sorry */
204 * Initialise the pool map for SVC_POOL_PERNODE mode.
205 * Returns number of pools or <0 on error.
208 svc_pool_map_init_pernode(struct svc_pool_map *m)
210 unsigned int maxpools = nr_node_ids;
211 unsigned int pidx = 0;
215 err = svc_pool_map_alloc_arrays(m, maxpools);
219 for_each_node_with_cpus(node) {
220 /* some architectures (e.g. SN2) have cpuless nodes */
221 BUG_ON(pidx > maxpools);
222 m->to_pool[node] = pidx;
223 m->pool_to[pidx] = node;
226 /* nodes brought online later all get mapped to pool0, sorry */
233 * Add a reference to the global map of cpus to pools (and
234 * vice versa). Initialise the map if we're the first user.
235 * Returns the number of pools.
238 svc_pool_map_get(void)
240 struct svc_pool_map *m = &svc_pool_map;
243 mutex_lock(&svc_pool_map_mutex);
246 mutex_unlock(&svc_pool_map_mutex);
250 if (m->mode == SVC_POOL_AUTO)
251 m->mode = svc_pool_map_choose_mode();
254 case SVC_POOL_PERCPU:
255 npools = svc_pool_map_init_percpu(m);
257 case SVC_POOL_PERNODE:
258 npools = svc_pool_map_init_pernode(m);
263 /* default, or memory allocation failure */
265 m->mode = SVC_POOL_GLOBAL;
269 mutex_unlock(&svc_pool_map_mutex);
275 * Drop a reference to the global map of cpus to pools.
276 * When the last reference is dropped, the map data is
277 * freed; this allows the sysadmin to change the pool
278 * mode using the pool_mode module option without
279 * rebooting or re-loading sunrpc.ko.
282 svc_pool_map_put(void)
284 struct svc_pool_map *m = &svc_pool_map;
286 mutex_lock(&svc_pool_map_mutex);
296 mutex_unlock(&svc_pool_map_mutex);
300 static int svc_pool_map_get_node(unsigned int pidx)
302 const struct svc_pool_map *m = &svc_pool_map;
305 if (m->mode == SVC_POOL_PERCPU)
306 return cpu_to_node(m->pool_to[pidx]);
307 if (m->mode == SVC_POOL_PERNODE)
308 return m->pool_to[pidx];
313 * Set the given thread's cpus_allowed mask so that it
314 * will only run on cpus in the given pool.
317 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
319 struct svc_pool_map *m = &svc_pool_map;
320 unsigned int node = m->pool_to[pidx];
323 * The caller checks for sv_nrpools > 1, which
324 * implies that we've been initialized.
326 WARN_ON_ONCE(m->count == 0);
331 case SVC_POOL_PERCPU:
333 set_cpus_allowed_ptr(task, cpumask_of(node));
336 case SVC_POOL_PERNODE:
338 set_cpus_allowed_ptr(task, cpumask_of_node(node));
345 * Use the mapping mode to choose a pool for a given CPU.
346 * Used when enqueueing an incoming RPC. Always returns
347 * a non-NULL pool pointer.
350 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
352 struct svc_pool_map *m = &svc_pool_map;
353 unsigned int pidx = 0;
356 * An uninitialised map happens in a pure client when
357 * lockd is brought up, so silently treat it the
358 * same as SVC_POOL_GLOBAL.
360 if (svc_serv_is_pooled(serv)) {
362 case SVC_POOL_PERCPU:
363 pidx = m->to_pool[cpu];
365 case SVC_POOL_PERNODE:
366 pidx = m->to_pool[cpu_to_node(cpu)];
370 return &serv->sv_pools[pidx % serv->sv_nrpools];
373 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
377 err = rpcb_create_local(net);
381 /* Remove any stale portmap registrations */
382 svc_unregister(serv, net);
385 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
387 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
389 svc_unregister(serv, net);
392 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
394 static int svc_uses_rpcbind(struct svc_serv *serv)
396 struct svc_program *progp;
399 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
400 for (i = 0; i < progp->pg_nvers; i++) {
401 if (progp->pg_vers[i] == NULL)
403 if (progp->pg_vers[i]->vs_hidden == 0)
411 int svc_bind(struct svc_serv *serv, struct net *net)
413 if (!svc_uses_rpcbind(serv))
415 return svc_rpcb_setup(serv, net);
417 EXPORT_SYMBOL_GPL(svc_bind);
420 * Create an RPC service
422 static struct svc_serv *
423 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
424 void (*shutdown)(struct svc_serv *serv, struct net *net))
426 struct svc_serv *serv;
428 unsigned int xdrsize;
431 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
433 serv->sv_name = prog->pg_name;
434 serv->sv_program = prog;
435 serv->sv_nrthreads = 1;
436 serv->sv_stats = prog->pg_stats;
437 if (bufsize > RPCSVC_MAXPAYLOAD)
438 bufsize = RPCSVC_MAXPAYLOAD;
439 serv->sv_max_payload = bufsize? bufsize : 4096;
440 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
441 serv->sv_shutdown = shutdown;
444 prog->pg_lovers = prog->pg_nvers-1;
445 for (vers=0; vers<prog->pg_nvers ; vers++)
446 if (prog->pg_vers[vers]) {
447 prog->pg_hivers = vers;
448 if (prog->pg_lovers > vers)
449 prog->pg_lovers = vers;
450 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
451 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
453 prog = prog->pg_next;
455 serv->sv_xdrsize = xdrsize;
456 INIT_LIST_HEAD(&serv->sv_tempsocks);
457 INIT_LIST_HEAD(&serv->sv_permsocks);
458 init_timer(&serv->sv_temptimer);
459 spin_lock_init(&serv->sv_lock);
461 serv->sv_nrpools = npools;
463 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
465 if (!serv->sv_pools) {
470 for (i = 0; i < serv->sv_nrpools; i++) {
471 struct svc_pool *pool = &serv->sv_pools[i];
473 dprintk("svc: initialising pool %u for %s\n",
477 INIT_LIST_HEAD(&pool->sp_threads);
478 INIT_LIST_HEAD(&pool->sp_sockets);
479 INIT_LIST_HEAD(&pool->sp_all_threads);
480 spin_lock_init(&pool->sp_lock);
483 if (svc_uses_rpcbind(serv) && (!serv->sv_shutdown))
484 serv->sv_shutdown = svc_rpcb_cleanup;
490 svc_create(struct svc_program *prog, unsigned int bufsize,
491 void (*shutdown)(struct svc_serv *serv, struct net *net))
493 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
495 EXPORT_SYMBOL_GPL(svc_create);
498 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
499 void (*shutdown)(struct svc_serv *serv, struct net *net),
500 svc_thread_fn func, struct module *mod)
502 struct svc_serv *serv;
503 unsigned int npools = svc_pool_map_get();
505 serv = __svc_create(prog, bufsize, npools, shutdown);
508 serv->sv_function = func;
509 serv->sv_module = mod;
514 EXPORT_SYMBOL_GPL(svc_create_pooled);
516 void svc_shutdown_net(struct svc_serv *serv, struct net *net)
518 svc_close_net(serv, net);
520 if (serv->sv_shutdown)
521 serv->sv_shutdown(serv, net);
523 EXPORT_SYMBOL_GPL(svc_shutdown_net);
526 * Destroy an RPC service. Should be called with appropriate locking to
527 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
530 svc_destroy(struct svc_serv *serv)
532 dprintk("svc: svc_destroy(%s, %d)\n",
533 serv->sv_program->pg_name,
536 if (serv->sv_nrthreads) {
537 if (--(serv->sv_nrthreads) != 0) {
538 svc_sock_update_bufs(serv);
542 printk("svc_destroy: no threads for serv=%p!\n", serv);
544 del_timer_sync(&serv->sv_temptimer);
547 * The last user is gone and thus all sockets have to be destroyed to
548 * the point. Check this.
550 BUG_ON(!list_empty(&serv->sv_permsocks));
551 BUG_ON(!list_empty(&serv->sv_tempsocks));
553 cache_clean_deferred(serv);
555 if (svc_serv_is_pooled(serv))
558 kfree(serv->sv_pools);
561 EXPORT_SYMBOL_GPL(svc_destroy);
564 * Allocate an RPC server's buffer space.
565 * We allocate pages and place them in rq_argpages.
568 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
570 unsigned int pages, arghi;
572 /* bc_xprt uses fore channel allocated buffers */
573 if (svc_is_backchannel(rqstp))
576 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
577 * We assume one is at most one page
580 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
581 if (pages > RPCSVC_MAXPAGES)
582 pages = RPCSVC_MAXPAGES;
584 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
587 rqstp->rq_pages[arghi++] = p;
594 * Release an RPC server buffer
597 svc_release_buffer(struct svc_rqst *rqstp)
601 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
602 if (rqstp->rq_pages[i])
603 put_page(rqstp->rq_pages[i]);
607 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
609 struct svc_rqst *rqstp;
611 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
615 init_waitqueue_head(&rqstp->rq_wait);
617 serv->sv_nrthreads++;
618 spin_lock_bh(&pool->sp_lock);
619 pool->sp_nrthreads++;
620 list_add(&rqstp->rq_all, &pool->sp_all_threads);
621 spin_unlock_bh(&pool->sp_lock);
622 rqstp->rq_server = serv;
623 rqstp->rq_pool = pool;
625 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
629 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
633 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
638 svc_exit_thread(rqstp);
640 return ERR_PTR(-ENOMEM);
642 EXPORT_SYMBOL_GPL(svc_prepare_thread);
645 * Choose a pool in which to create a new thread, for svc_set_num_threads
647 static inline struct svc_pool *
648 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
653 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
657 * Choose a thread to kill, for svc_set_num_threads
659 static inline struct task_struct *
660 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
663 struct task_struct *task = NULL;
666 spin_lock_bh(&pool->sp_lock);
668 /* choose a pool in round-robin fashion */
669 for (i = 0; i < serv->sv_nrpools; i++) {
670 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
671 spin_lock_bh(&pool->sp_lock);
672 if (!list_empty(&pool->sp_all_threads))
674 spin_unlock_bh(&pool->sp_lock);
680 if (!list_empty(&pool->sp_all_threads)) {
681 struct svc_rqst *rqstp;
684 * Remove from the pool->sp_all_threads list
685 * so we don't try to kill it again.
687 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
688 list_del_init(&rqstp->rq_all);
689 task = rqstp->rq_task;
691 spin_unlock_bh(&pool->sp_lock);
697 * Create or destroy enough new threads to make the number
698 * of threads the given number. If `pool' is non-NULL, applies
699 * only to threads in that pool, otherwise round-robins between
700 * all pools. Caller must ensure that mutual exclusion between this and
701 * server startup or shutdown.
703 * Destroying threads relies on the service threads filling in
704 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
705 * has been created using svc_create_pooled().
707 * Based on code that used to be in nfsd_svc() but tweaked
711 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
713 struct svc_rqst *rqstp;
714 struct task_struct *task;
715 struct svc_pool *chosen_pool;
717 unsigned int state = serv->sv_nrthreads-1;
721 /* The -1 assumes caller has done a svc_get() */
722 nrservs -= (serv->sv_nrthreads-1);
724 spin_lock_bh(&pool->sp_lock);
725 nrservs -= pool->sp_nrthreads;
726 spin_unlock_bh(&pool->sp_lock);
729 /* create new threads */
730 while (nrservs > 0) {
732 chosen_pool = choose_pool(serv, pool, &state);
734 node = svc_pool_map_get_node(chosen_pool->sp_id);
735 rqstp = svc_prepare_thread(serv, chosen_pool, node);
737 error = PTR_ERR(rqstp);
741 __module_get(serv->sv_module);
742 task = kthread_create_on_node(serv->sv_function, rqstp,
743 node, "%s", serv->sv_name);
745 error = PTR_ERR(task);
746 module_put(serv->sv_module);
747 svc_exit_thread(rqstp);
751 rqstp->rq_task = task;
752 if (serv->sv_nrpools > 1)
753 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
755 svc_sock_update_bufs(serv);
756 wake_up_process(task);
758 /* destroy old threads */
759 while (nrservs < 0 &&
760 (task = choose_victim(serv, pool, &state)) != NULL) {
761 send_sig(SIGINT, task, 1);
767 EXPORT_SYMBOL_GPL(svc_set_num_threads);
770 * Called from a server thread as it's exiting. Caller must hold the BKL or
771 * the "service mutex", whichever is appropriate for the service.
774 svc_exit_thread(struct svc_rqst *rqstp)
776 struct svc_serv *serv = rqstp->rq_server;
777 struct svc_pool *pool = rqstp->rq_pool;
779 svc_release_buffer(rqstp);
780 kfree(rqstp->rq_resp);
781 kfree(rqstp->rq_argp);
782 kfree(rqstp->rq_auth_data);
784 spin_lock_bh(&pool->sp_lock);
785 pool->sp_nrthreads--;
786 list_del(&rqstp->rq_all);
787 spin_unlock_bh(&pool->sp_lock);
791 /* Release the server */
795 EXPORT_SYMBOL_GPL(svc_exit_thread);
798 * Register an "inet" protocol family netid with the local
799 * rpcbind daemon via an rpcbind v4 SET request.
801 * No netconfig infrastructure is available in the kernel, so
802 * we map IP_ protocol numbers to netids by hand.
804 * Returns zero on success; a negative errno value is returned
805 * if any error occurs.
807 static int __svc_rpcb_register4(struct net *net, const u32 program,
809 const unsigned short protocol,
810 const unsigned short port)
812 const struct sockaddr_in sin = {
813 .sin_family = AF_INET,
814 .sin_addr.s_addr = htonl(INADDR_ANY),
815 .sin_port = htons(port),
822 netid = RPCBIND_NETID_UDP;
825 netid = RPCBIND_NETID_TCP;
831 error = rpcb_v4_register(net, program, version,
832 (const struct sockaddr *)&sin, netid);
835 * User space didn't support rpcbind v4, so retry this
836 * registration request with the legacy rpcbind v2 protocol.
838 if (error == -EPROTONOSUPPORT)
839 error = rpcb_register(net, program, version, protocol, port);
844 #if IS_ENABLED(CONFIG_IPV6)
846 * Register an "inet6" protocol family netid with the local
847 * rpcbind daemon via an rpcbind v4 SET request.
849 * No netconfig infrastructure is available in the kernel, so
850 * we map IP_ protocol numbers to netids by hand.
852 * Returns zero on success; a negative errno value is returned
853 * if any error occurs.
855 static int __svc_rpcb_register6(struct net *net, const u32 program,
857 const unsigned short protocol,
858 const unsigned short port)
860 const struct sockaddr_in6 sin6 = {
861 .sin6_family = AF_INET6,
862 .sin6_addr = IN6ADDR_ANY_INIT,
863 .sin6_port = htons(port),
870 netid = RPCBIND_NETID_UDP6;
873 netid = RPCBIND_NETID_TCP6;
879 error = rpcb_v4_register(net, program, version,
880 (const struct sockaddr *)&sin6, netid);
883 * User space didn't support rpcbind version 4, so we won't
884 * use a PF_INET6 listener.
886 if (error == -EPROTONOSUPPORT)
887 error = -EAFNOSUPPORT;
891 #endif /* IS_ENABLED(CONFIG_IPV6) */
894 * Register a kernel RPC service via rpcbind version 4.
896 * Returns zero on success; a negative errno value is returned
897 * if any error occurs.
899 static int __svc_register(struct net *net, const char *progname,
900 const u32 program, const u32 version,
902 const unsigned short protocol,
903 const unsigned short port)
905 int error = -EAFNOSUPPORT;
909 error = __svc_rpcb_register4(net, program, version,
912 #if IS_ENABLED(CONFIG_IPV6)
914 error = __svc_rpcb_register6(net, program, version,
923 * svc_register - register an RPC service with the local portmapper
924 * @serv: svc_serv struct for the service to register
925 * @net: net namespace for the service to register
926 * @family: protocol family of service's listener socket
927 * @proto: transport protocol number to advertise
928 * @port: port to advertise
930 * Service is registered for any address in the passed-in protocol family
932 int svc_register(const struct svc_serv *serv, struct net *net,
933 const int family, const unsigned short proto,
934 const unsigned short port)
936 struct svc_program *progp;
937 struct svc_version *vers;
941 WARN_ON_ONCE(proto == 0 && port == 0);
942 if (proto == 0 && port == 0)
945 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
946 for (i = 0; i < progp->pg_nvers; i++) {
947 vers = progp->pg_vers[i];
951 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
954 proto == IPPROTO_UDP? "udp" : "tcp",
958 " (but not telling portmap)" : "");
963 error = __svc_register(net, progp->pg_name, progp->pg_prog,
964 i, family, proto, port);
966 if (vers->vs_rpcb_optnl) {
972 printk(KERN_WARNING "svc: failed to register "
973 "%sv%u RPC service (errno %d).\n",
974 progp->pg_name, i, -error);
984 * If user space is running rpcbind, it should take the v4 UNSET
985 * and clear everything for this [program, version]. If user space
986 * is running portmap, it will reject the v4 UNSET, but won't have
987 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
988 * in this case to clear all existing entries for [program, version].
990 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
991 const char *progname)
995 error = rpcb_v4_register(net, program, version, NULL, "");
998 * User space didn't support rpcbind v4, so retry this
999 * request with the legacy rpcbind v2 protocol.
1001 if (error == -EPROTONOSUPPORT)
1002 error = rpcb_register(net, program, version, 0, 0);
1004 dprintk("svc: %s(%sv%u), error %d\n",
1005 __func__, progname, version, error);
1009 * All netids, bind addresses and ports registered for [program, version]
1010 * are removed from the local rpcbind database (if the service is not
1011 * hidden) to make way for a new instance of the service.
1013 * The result of unregistration is reported via dprintk for those who want
1014 * verification of the result, but is otherwise not important.
1016 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1018 struct svc_program *progp;
1019 unsigned long flags;
1022 clear_thread_flag(TIF_SIGPENDING);
1024 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1025 for (i = 0; i < progp->pg_nvers; i++) {
1026 if (progp->pg_vers[i] == NULL)
1028 if (progp->pg_vers[i]->vs_hidden)
1031 dprintk("svc: attempting to unregister %sv%u\n",
1033 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1037 spin_lock_irqsave(¤t->sighand->siglock, flags);
1038 recalc_sigpending();
1039 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1043 * dprintk the given error with the address of the client that caused it.
1046 static __printf(2, 3)
1047 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1049 struct va_format vaf;
1051 char buf[RPC_MAX_ADDRBUFLEN];
1053 va_start(args, fmt);
1058 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1063 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1067 * Common routine for processing the RPC request.
1070 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1072 struct svc_program *progp;
1073 struct svc_version *versp = NULL; /* compiler food */
1074 struct svc_procedure *procp = NULL;
1075 struct svc_serv *serv = rqstp->rq_server;
1078 u32 prog, vers, proc;
1079 __be32 auth_stat, rpc_stat;
1081 __be32 *reply_statp;
1083 rpc_stat = rpc_success;
1085 if (argv->iov_len < 6*4)
1088 /* Will be turned off only in gss privacy case: */
1089 rqstp->rq_splice_ok = 1;
1090 /* Will be turned off only when NFSv4 Sessions are used */
1091 rqstp->rq_usedeferral = 1;
1092 rqstp->rq_dropme = false;
1094 /* Setup reply header */
1095 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1097 svc_putu32(resv, rqstp->rq_xid);
1099 vers = svc_getnl(argv);
1101 /* First words of reply: */
1102 svc_putnl(resv, 1); /* REPLY */
1104 if (vers != 2) /* RPC version number */
1107 /* Save position in case we later decide to reject: */
1108 reply_statp = resv->iov_base + resv->iov_len;
1110 svc_putnl(resv, 0); /* ACCEPT */
1112 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1113 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1114 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1116 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1117 if (prog == progp->pg_prog)
1121 * Decode auth data, and add verifier to reply buffer.
1122 * We do this before anything else in order to get a decent
1125 auth_res = svc_authenticate(rqstp, &auth_stat);
1126 /* Also give the program a chance to reject this call: */
1127 if (auth_res == SVC_OK && progp) {
1128 auth_stat = rpc_autherr_badcred;
1129 auth_res = progp->pg_authenticate(rqstp);
1137 rpc_stat = rpc_system_err;
1142 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1143 svc_close_xprt(rqstp->rq_xprt);
1153 if (vers >= progp->pg_nvers ||
1154 !(versp = progp->pg_vers[vers]))
1157 procp = versp->vs_proc + proc;
1158 if (proc >= versp->vs_nproc || !procp->pc_func)
1160 rqstp->rq_procinfo = procp;
1162 /* Syntactic check complete */
1163 serv->sv_stats->rpccnt++;
1165 /* Build the reply header. */
1166 statp = resv->iov_base +resv->iov_len;
1167 svc_putnl(resv, RPC_SUCCESS);
1169 /* Bump per-procedure stats counter */
1172 /* Initialize storage for argp and resp */
1173 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1174 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1176 /* un-reserve some of the out-queue now that we have a
1177 * better idea of reply size
1179 if (procp->pc_xdrressize)
1180 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1182 /* Call the function that processes the request. */
1183 if (!versp->vs_dispatch) {
1184 /* Decode arguments */
1185 xdr = procp->pc_decode;
1186 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1189 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1192 if (rqstp->rq_dropme) {
1193 if (procp->pc_release)
1194 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1197 if (*statp == rpc_success &&
1198 (xdr = procp->pc_encode) &&
1199 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1200 dprintk("svc: failed to encode reply\n");
1201 /* serv->sv_stats->rpcsystemerr++; */
1202 *statp = rpc_system_err;
1205 dprintk("svc: calling dispatcher\n");
1206 if (!versp->vs_dispatch(rqstp, statp)) {
1207 /* Release reply info */
1208 if (procp->pc_release)
1209 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1214 /* Check RPC status result */
1215 if (*statp != rpc_success)
1216 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1218 /* Release reply info */
1219 if (procp->pc_release)
1220 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1222 if (procp->pc_encode == NULL)
1226 if (svc_authorise(rqstp))
1228 return 1; /* Caller can now send it */
1231 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1232 dprintk("svc: svc_process dropit\n");
1236 svc_printk(rqstp, "short len %Zd, dropping request\n",
1239 goto dropit; /* drop request */
1242 serv->sv_stats->rpcbadfmt++;
1243 svc_putnl(resv, 1); /* REJECT */
1244 svc_putnl(resv, 0); /* RPC_MISMATCH */
1245 svc_putnl(resv, 2); /* Only RPCv2 supported */
1250 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1251 serv->sv_stats->rpcbadauth++;
1252 /* Restore write pointer to location of accept status: */
1253 xdr_ressize_check(rqstp, reply_statp);
1254 svc_putnl(resv, 1); /* REJECT */
1255 svc_putnl(resv, 1); /* AUTH_ERROR */
1256 svc_putnl(resv, ntohl(auth_stat)); /* status */
1260 dprintk("svc: unknown program %d\n", prog);
1261 serv->sv_stats->rpcbadfmt++;
1262 svc_putnl(resv, RPC_PROG_UNAVAIL);
1266 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1267 vers, prog, progp->pg_name);
1269 serv->sv_stats->rpcbadfmt++;
1270 svc_putnl(resv, RPC_PROG_MISMATCH);
1271 svc_putnl(resv, progp->pg_lovers);
1272 svc_putnl(resv, progp->pg_hivers);
1276 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1278 serv->sv_stats->rpcbadfmt++;
1279 svc_putnl(resv, RPC_PROC_UNAVAIL);
1283 svc_printk(rqstp, "failed to decode args\n");
1285 rpc_stat = rpc_garbage_args;
1287 serv->sv_stats->rpcbadfmt++;
1288 svc_putnl(resv, ntohl(rpc_stat));
1291 EXPORT_SYMBOL_GPL(svc_process);
1294 * Process the RPC request.
1297 svc_process(struct svc_rqst *rqstp)
1299 struct kvec *argv = &rqstp->rq_arg.head[0];
1300 struct kvec *resv = &rqstp->rq_res.head[0];
1301 struct svc_serv *serv = rqstp->rq_server;
1305 * Setup response xdr_buf.
1306 * Initially it has just one page
1308 rqstp->rq_next_page = &rqstp->rq_respages[1];
1309 resv->iov_base = page_address(rqstp->rq_respages[0]);
1311 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1312 rqstp->rq_res.len = 0;
1313 rqstp->rq_res.page_base = 0;
1314 rqstp->rq_res.page_len = 0;
1315 rqstp->rq_res.buflen = PAGE_SIZE;
1316 rqstp->rq_res.tail[0].iov_base = NULL;
1317 rqstp->rq_res.tail[0].iov_len = 0;
1319 rqstp->rq_xid = svc_getu32(argv);
1321 dir = svc_getnl(argv);
1323 /* direction != CALL */
1324 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1325 serv->sv_stats->rpcbadfmt++;
1330 /* Returns 1 for send, 0 for drop */
1331 if (svc_process_common(rqstp, argv, resv))
1332 return svc_send(rqstp);
1339 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1341 * Process a backchannel RPC request that arrived over an existing
1342 * outbound connection
1345 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1346 struct svc_rqst *rqstp)
1348 struct kvec *argv = &rqstp->rq_arg.head[0];
1349 struct kvec *resv = &rqstp->rq_res.head[0];
1351 /* Build the svc_rqst used by the common processing routine */
1352 rqstp->rq_xprt = serv->sv_bc_xprt;
1353 rqstp->rq_xid = req->rq_xid;
1354 rqstp->rq_prot = req->rq_xprt->prot;
1355 rqstp->rq_server = serv;
1357 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1358 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1359 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1360 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1362 /* reset result send buffer "put" position */
1365 if (rqstp->rq_prot != IPPROTO_TCP) {
1366 printk(KERN_ERR "No support for Non-TCP transports!\n");
1371 * Skip the next two words because they've already been
1372 * processed in the trasport
1374 svc_getu32(argv); /* XID */
1375 svc_getnl(argv); /* CALLDIR */
1377 /* Returns 1 for send, 0 for drop */
1378 if (svc_process_common(rqstp, argv, resv)) {
1379 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1380 sizeof(req->rq_snd_buf));
1381 return bc_send(req);
1384 xprt_free_bc_request(req);
1388 EXPORT_SYMBOL_GPL(bc_svc_process);
1389 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1392 * Return (transport-specific) limit on the rpc payload.
1394 u32 svc_max_payload(const struct svc_rqst *rqstp)
1396 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1398 if (rqstp->rq_server->sv_max_payload < max)
1399 max = rqstp->rq_server->sv_max_payload;
1402 EXPORT_SYMBOL_GPL(svc_max_payload);