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>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
23 #include <linux/nsproxy.h>
25 #include <linux/sunrpc/types.h>
26 #include <linux/sunrpc/xdr.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/svcsock.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/bc_xprt.h>
32 #define RPCDBG_FACILITY RPCDBG_SVCDSP
34 static void svc_unregister(const struct svc_serv *serv, struct net *net);
36 #define svc_serv_is_pooled(serv) ((serv)->sv_function)
39 * Mode for mapping cpus to pools.
42 SVC_POOL_AUTO = -1, /* choose one of the others */
43 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
44 * (legacy & UP mode) */
45 SVC_POOL_PERCPU, /* one pool per cpu */
46 SVC_POOL_PERNODE /* one pool per numa node */
48 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
51 * Structure for mapping cpus to pools and vice versa.
52 * Setup once during sunrpc initialisation.
54 static struct svc_pool_map {
55 int count; /* How many svc_servs use us */
56 int mode; /* Note: int not enum to avoid
57 * warnings about "enumeration value
58 * not handled in switch" */
60 unsigned int *pool_to; /* maps pool id to cpu or node */
61 unsigned int *to_pool; /* maps cpu or node to pool id */
64 .mode = SVC_POOL_DEFAULT
66 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
69 param_set_pool_mode(const char *val, struct kernel_param *kp)
71 int *ip = (int *)kp->arg;
72 struct svc_pool_map *m = &svc_pool_map;
75 mutex_lock(&svc_pool_map_mutex);
82 if (!strncmp(val, "auto", 4))
84 else if (!strncmp(val, "global", 6))
85 *ip = SVC_POOL_GLOBAL;
86 else if (!strncmp(val, "percpu", 6))
87 *ip = SVC_POOL_PERCPU;
88 else if (!strncmp(val, "pernode", 7))
89 *ip = SVC_POOL_PERNODE;
94 mutex_unlock(&svc_pool_map_mutex);
99 param_get_pool_mode(char *buf, struct kernel_param *kp)
101 int *ip = (int *)kp->arg;
106 return strlcpy(buf, "auto", 20);
107 case SVC_POOL_GLOBAL:
108 return strlcpy(buf, "global", 20);
109 case SVC_POOL_PERCPU:
110 return strlcpy(buf, "percpu", 20);
111 case SVC_POOL_PERNODE:
112 return strlcpy(buf, "pernode", 20);
114 return sprintf(buf, "%d", *ip);
118 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
119 &svc_pool_map.mode, 0644);
122 * Detect best pool mapping mode heuristically,
123 * according to the machine's topology.
126 svc_pool_map_choose_mode(void)
130 if (nr_online_nodes > 1) {
132 * Actually have multiple NUMA nodes,
133 * so split pools on NUMA node boundaries
135 return SVC_POOL_PERNODE;
138 node = first_online_node;
139 if (nr_cpus_node(node) > 2) {
141 * Non-trivial SMP, or CONFIG_NUMA on
142 * non-NUMA hardware, e.g. with a generic
143 * x86_64 kernel on Xeons. In this case we
144 * want to divide the pools on cpu boundaries.
146 return SVC_POOL_PERCPU;
149 /* default: one global pool */
150 return SVC_POOL_GLOBAL;
154 * Allocate the to_pool[] and pool_to[] arrays.
155 * Returns 0 on success or an errno.
158 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
160 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
163 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
177 * Initialise the pool map for SVC_POOL_PERCPU mode.
178 * Returns number of pools or <0 on error.
181 svc_pool_map_init_percpu(struct svc_pool_map *m)
183 unsigned int maxpools = nr_cpu_ids;
184 unsigned int pidx = 0;
188 err = svc_pool_map_alloc_arrays(m, maxpools);
192 for_each_online_cpu(cpu) {
193 BUG_ON(pidx > maxpools);
194 m->to_pool[cpu] = pidx;
195 m->pool_to[pidx] = cpu;
198 /* cpus brought online later all get mapped to pool0, sorry */
205 * Initialise the pool map for SVC_POOL_PERNODE mode.
206 * Returns number of pools or <0 on error.
209 svc_pool_map_init_pernode(struct svc_pool_map *m)
211 unsigned int maxpools = nr_node_ids;
212 unsigned int pidx = 0;
216 err = svc_pool_map_alloc_arrays(m, maxpools);
220 for_each_node_with_cpus(node) {
221 /* some architectures (e.g. SN2) have cpuless nodes */
222 BUG_ON(pidx > maxpools);
223 m->to_pool[node] = pidx;
224 m->pool_to[pidx] = node;
227 /* nodes brought online later all get mapped to pool0, sorry */
234 * Add a reference to the global map of cpus to pools (and
235 * vice versa). Initialise the map if we're the first user.
236 * Returns the number of pools.
239 svc_pool_map_get(void)
241 struct svc_pool_map *m = &svc_pool_map;
244 mutex_lock(&svc_pool_map_mutex);
247 mutex_unlock(&svc_pool_map_mutex);
251 if (m->mode == SVC_POOL_AUTO)
252 m->mode = svc_pool_map_choose_mode();
255 case SVC_POOL_PERCPU:
256 npools = svc_pool_map_init_percpu(m);
258 case SVC_POOL_PERNODE:
259 npools = svc_pool_map_init_pernode(m);
264 /* default, or memory allocation failure */
266 m->mode = SVC_POOL_GLOBAL;
270 mutex_unlock(&svc_pool_map_mutex);
276 * Drop a reference to the global map of cpus to pools.
277 * When the last reference is dropped, the map data is
278 * freed; this allows the sysadmin to change the pool
279 * mode using the pool_mode module option without
280 * rebooting or re-loading sunrpc.ko.
283 svc_pool_map_put(void)
285 struct svc_pool_map *m = &svc_pool_map;
287 mutex_lock(&svc_pool_map_mutex);
297 mutex_unlock(&svc_pool_map_mutex);
301 static int svc_pool_map_get_node(unsigned int pidx)
303 const struct svc_pool_map *m = &svc_pool_map;
306 if (m->mode == SVC_POOL_PERCPU)
307 return cpu_to_node(m->pool_to[pidx]);
308 if (m->mode == SVC_POOL_PERNODE)
309 return m->pool_to[pidx];
314 * Set the given thread's cpus_allowed mask so that it
315 * will only run on cpus in the given pool.
318 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
320 struct svc_pool_map *m = &svc_pool_map;
321 unsigned int node = m->pool_to[pidx];
324 * The caller checks for sv_nrpools > 1, which
325 * implies that we've been initialized.
327 BUG_ON(m->count == 0);
330 case SVC_POOL_PERCPU:
332 set_cpus_allowed_ptr(task, cpumask_of(node));
335 case SVC_POOL_PERNODE:
337 set_cpus_allowed_ptr(task, cpumask_of_node(node));
344 * Use the mapping mode to choose a pool for a given CPU.
345 * Used when enqueueing an incoming RPC. Always returns
346 * a non-NULL pool pointer.
349 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
351 struct svc_pool_map *m = &svc_pool_map;
352 unsigned int pidx = 0;
355 * An uninitialised map happens in a pure client when
356 * lockd is brought up, so silently treat it the
357 * same as SVC_POOL_GLOBAL.
359 if (svc_serv_is_pooled(serv)) {
361 case SVC_POOL_PERCPU:
362 pidx = m->to_pool[cpu];
364 case SVC_POOL_PERNODE:
365 pidx = m->to_pool[cpu_to_node(cpu)];
369 return &serv->sv_pools[pidx % serv->sv_nrpools];
372 static int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
376 err = rpcb_create_local(net);
380 /* Remove any stale portmap registrations */
381 svc_unregister(serv, net);
385 void svc_rpcb_cleanup(struct svc_serv *serv)
387 svc_unregister(serv, &init_net);
388 rpcb_put_local(&init_net);
390 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
392 static int svc_uses_rpcbind(struct svc_serv *serv)
394 struct svc_program *progp;
397 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
398 for (i = 0; i < progp->pg_nvers; i++) {
399 if (progp->pg_vers[i] == NULL)
401 if (progp->pg_vers[i]->vs_hidden == 0)
410 * Create an RPC service
412 static struct svc_serv *
413 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
414 void (*shutdown)(struct svc_serv *serv))
416 struct svc_serv *serv;
418 unsigned int xdrsize;
421 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
423 serv->sv_name = prog->pg_name;
424 serv->sv_program = prog;
425 serv->sv_nrthreads = 1;
426 serv->sv_stats = prog->pg_stats;
427 if (bufsize > RPCSVC_MAXPAYLOAD)
428 bufsize = RPCSVC_MAXPAYLOAD;
429 serv->sv_max_payload = bufsize? bufsize : 4096;
430 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
431 serv->sv_shutdown = shutdown;
434 prog->pg_lovers = prog->pg_nvers-1;
435 for (vers=0; vers<prog->pg_nvers ; vers++)
436 if (prog->pg_vers[vers]) {
437 prog->pg_hivers = vers;
438 if (prog->pg_lovers > vers)
439 prog->pg_lovers = vers;
440 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
441 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
443 prog = prog->pg_next;
445 serv->sv_xdrsize = xdrsize;
446 INIT_LIST_HEAD(&serv->sv_tempsocks);
447 INIT_LIST_HEAD(&serv->sv_permsocks);
448 init_timer(&serv->sv_temptimer);
449 spin_lock_init(&serv->sv_lock);
451 serv->sv_nrpools = npools;
453 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
455 if (!serv->sv_pools) {
460 for (i = 0; i < serv->sv_nrpools; i++) {
461 struct svc_pool *pool = &serv->sv_pools[i];
463 dprintk("svc: initialising pool %u for %s\n",
467 INIT_LIST_HEAD(&pool->sp_threads);
468 INIT_LIST_HEAD(&pool->sp_sockets);
469 INIT_LIST_HEAD(&pool->sp_all_threads);
470 spin_lock_init(&pool->sp_lock);
473 if (svc_uses_rpcbind(serv)) {
474 if (svc_rpcb_setup(serv, current->nsproxy->net_ns) < 0) {
475 kfree(serv->sv_pools);
479 if (!serv->sv_shutdown)
480 serv->sv_shutdown = svc_rpcb_cleanup;
487 svc_create(struct svc_program *prog, unsigned int bufsize,
488 void (*shutdown)(struct svc_serv *serv))
490 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
492 EXPORT_SYMBOL_GPL(svc_create);
495 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
496 void (*shutdown)(struct svc_serv *serv),
497 svc_thread_fn func, struct module *mod)
499 struct svc_serv *serv;
500 unsigned int npools = svc_pool_map_get();
502 serv = __svc_create(prog, bufsize, npools, shutdown);
505 serv->sv_function = func;
506 serv->sv_module = mod;
511 EXPORT_SYMBOL_GPL(svc_create_pooled);
514 * Destroy an RPC service. Should be called with appropriate locking to
515 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
518 svc_destroy(struct svc_serv *serv)
520 dprintk("svc: svc_destroy(%s, %d)\n",
521 serv->sv_program->pg_name,
524 if (serv->sv_nrthreads) {
525 if (--(serv->sv_nrthreads) != 0) {
526 svc_sock_update_bufs(serv);
530 printk("svc_destroy: no threads for serv=%p!\n", serv);
532 del_timer_sync(&serv->sv_temptimer);
534 * The set of xprts (contained in the sv_tempsocks and
535 * sv_permsocks lists) is now constant, since it is modified
536 * only by accepting new sockets (done by service threads in
537 * svc_recv) or aging old ones (done by sv_temptimer), or
538 * configuration changes (excluded by whatever locking the
539 * caller is using--nfsd_mutex in the case of nfsd). So it's
540 * safe to traverse those lists and shut everything down:
544 if (serv->sv_shutdown)
545 serv->sv_shutdown(serv);
547 cache_clean_deferred(serv);
549 if (svc_serv_is_pooled(serv))
552 kfree(serv->sv_pools);
555 EXPORT_SYMBOL_GPL(svc_destroy);
558 * Allocate an RPC server's buffer space.
559 * We allocate pages and place them in rq_argpages.
562 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
564 unsigned int pages, arghi;
566 /* bc_xprt uses fore channel allocated buffers */
567 if (svc_is_backchannel(rqstp))
570 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
571 * We assume one is at most one page
574 BUG_ON(pages > RPCSVC_MAXPAGES);
576 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
579 rqstp->rq_pages[arghi++] = p;
586 * Release an RPC server buffer
589 svc_release_buffer(struct svc_rqst *rqstp)
593 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
594 if (rqstp->rq_pages[i])
595 put_page(rqstp->rq_pages[i]);
599 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
601 struct svc_rqst *rqstp;
603 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
607 init_waitqueue_head(&rqstp->rq_wait);
609 serv->sv_nrthreads++;
610 spin_lock_bh(&pool->sp_lock);
611 pool->sp_nrthreads++;
612 list_add(&rqstp->rq_all, &pool->sp_all_threads);
613 spin_unlock_bh(&pool->sp_lock);
614 rqstp->rq_server = serv;
615 rqstp->rq_pool = pool;
617 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
621 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
625 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
630 svc_exit_thread(rqstp);
632 return ERR_PTR(-ENOMEM);
634 EXPORT_SYMBOL_GPL(svc_prepare_thread);
637 * Choose a pool in which to create a new thread, for svc_set_num_threads
639 static inline struct svc_pool *
640 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
645 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
649 * Choose a thread to kill, for svc_set_num_threads
651 static inline struct task_struct *
652 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
655 struct task_struct *task = NULL;
658 spin_lock_bh(&pool->sp_lock);
660 /* choose a pool in round-robin fashion */
661 for (i = 0; i < serv->sv_nrpools; i++) {
662 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
663 spin_lock_bh(&pool->sp_lock);
664 if (!list_empty(&pool->sp_all_threads))
666 spin_unlock_bh(&pool->sp_lock);
672 if (!list_empty(&pool->sp_all_threads)) {
673 struct svc_rqst *rqstp;
676 * Remove from the pool->sp_all_threads list
677 * so we don't try to kill it again.
679 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
680 list_del_init(&rqstp->rq_all);
681 task = rqstp->rq_task;
683 spin_unlock_bh(&pool->sp_lock);
689 * Create or destroy enough new threads to make the number
690 * of threads the given number. If `pool' is non-NULL, applies
691 * only to threads in that pool, otherwise round-robins between
692 * all pools. Caller must ensure that mutual exclusion between this and
693 * server startup or shutdown.
695 * Destroying threads relies on the service threads filling in
696 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
697 * has been created using svc_create_pooled().
699 * Based on code that used to be in nfsd_svc() but tweaked
703 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
705 struct svc_rqst *rqstp;
706 struct task_struct *task;
707 struct svc_pool *chosen_pool;
709 unsigned int state = serv->sv_nrthreads-1;
713 /* The -1 assumes caller has done a svc_get() */
714 nrservs -= (serv->sv_nrthreads-1);
716 spin_lock_bh(&pool->sp_lock);
717 nrservs -= pool->sp_nrthreads;
718 spin_unlock_bh(&pool->sp_lock);
721 /* create new threads */
722 while (nrservs > 0) {
724 chosen_pool = choose_pool(serv, pool, &state);
726 node = svc_pool_map_get_node(chosen_pool->sp_id);
727 rqstp = svc_prepare_thread(serv, chosen_pool, node);
729 error = PTR_ERR(rqstp);
733 __module_get(serv->sv_module);
734 task = kthread_create_on_node(serv->sv_function, rqstp,
735 node, serv->sv_name);
737 error = PTR_ERR(task);
738 module_put(serv->sv_module);
739 svc_exit_thread(rqstp);
743 rqstp->rq_task = task;
744 if (serv->sv_nrpools > 1)
745 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
747 svc_sock_update_bufs(serv);
748 wake_up_process(task);
750 /* destroy old threads */
751 while (nrservs < 0 &&
752 (task = choose_victim(serv, pool, &state)) != NULL) {
753 send_sig(SIGINT, task, 1);
759 EXPORT_SYMBOL_GPL(svc_set_num_threads);
762 * Called from a server thread as it's exiting. Caller must hold the BKL or
763 * the "service mutex", whichever is appropriate for the service.
766 svc_exit_thread(struct svc_rqst *rqstp)
768 struct svc_serv *serv = rqstp->rq_server;
769 struct svc_pool *pool = rqstp->rq_pool;
771 svc_release_buffer(rqstp);
772 kfree(rqstp->rq_resp);
773 kfree(rqstp->rq_argp);
774 kfree(rqstp->rq_auth_data);
776 spin_lock_bh(&pool->sp_lock);
777 pool->sp_nrthreads--;
778 list_del(&rqstp->rq_all);
779 spin_unlock_bh(&pool->sp_lock);
783 /* Release the server */
787 EXPORT_SYMBOL_GPL(svc_exit_thread);
790 * Register an "inet" protocol family netid with the local
791 * rpcbind daemon via an rpcbind v4 SET request.
793 * No netconfig infrastructure is available in the kernel, so
794 * we map IP_ protocol numbers to netids by hand.
796 * Returns zero on success; a negative errno value is returned
797 * if any error occurs.
799 static int __svc_rpcb_register4(struct net *net, const u32 program,
801 const unsigned short protocol,
802 const unsigned short port)
804 const struct sockaddr_in sin = {
805 .sin_family = AF_INET,
806 .sin_addr.s_addr = htonl(INADDR_ANY),
807 .sin_port = htons(port),
814 netid = RPCBIND_NETID_UDP;
817 netid = RPCBIND_NETID_TCP;
823 error = rpcb_v4_register(net, program, version,
824 (const struct sockaddr *)&sin, netid);
827 * User space didn't support rpcbind v4, so retry this
828 * registration request with the legacy rpcbind v2 protocol.
830 if (error == -EPROTONOSUPPORT)
831 error = rpcb_register(net, program, version, protocol, port);
836 #if IS_ENABLED(CONFIG_IPV6)
838 * Register an "inet6" protocol family netid with the local
839 * rpcbind daemon via an rpcbind v4 SET request.
841 * No netconfig infrastructure is available in the kernel, so
842 * we map IP_ protocol numbers to netids by hand.
844 * Returns zero on success; a negative errno value is returned
845 * if any error occurs.
847 static int __svc_rpcb_register6(struct net *net, const u32 program,
849 const unsigned short protocol,
850 const unsigned short port)
852 const struct sockaddr_in6 sin6 = {
853 .sin6_family = AF_INET6,
854 .sin6_addr = IN6ADDR_ANY_INIT,
855 .sin6_port = htons(port),
862 netid = RPCBIND_NETID_UDP6;
865 netid = RPCBIND_NETID_TCP6;
871 error = rpcb_v4_register(net, program, version,
872 (const struct sockaddr *)&sin6, netid);
875 * User space didn't support rpcbind version 4, so we won't
876 * use a PF_INET6 listener.
878 if (error == -EPROTONOSUPPORT)
879 error = -EAFNOSUPPORT;
883 #endif /* IS_ENABLED(CONFIG_IPV6) */
886 * Register a kernel RPC service via rpcbind version 4.
888 * Returns zero on success; a negative errno value is returned
889 * if any error occurs.
891 static int __svc_register(struct net *net, const char *progname,
892 const u32 program, const u32 version,
894 const unsigned short protocol,
895 const unsigned short port)
897 int error = -EAFNOSUPPORT;
901 error = __svc_rpcb_register4(net, program, version,
904 #if IS_ENABLED(CONFIG_IPV6)
906 error = __svc_rpcb_register6(net, program, version,
912 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
913 "service (errno %d).\n", progname, version, -error);
918 * svc_register - register an RPC service with the local portmapper
919 * @serv: svc_serv struct for the service to register
920 * @net: net namespace for the service to register
921 * @family: protocol family of service's listener socket
922 * @proto: transport protocol number to advertise
923 * @port: port to advertise
925 * Service is registered for any address in the passed-in protocol family
927 int svc_register(const struct svc_serv *serv, struct net *net,
928 const int family, const unsigned short proto,
929 const unsigned short port)
931 struct svc_program *progp;
935 BUG_ON(proto == 0 && port == 0);
937 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
938 for (i = 0; i < progp->pg_nvers; i++) {
939 if (progp->pg_vers[i] == NULL)
942 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
945 proto == IPPROTO_UDP? "udp" : "tcp",
948 progp->pg_vers[i]->vs_hidden?
949 " (but not telling portmap)" : "");
951 if (progp->pg_vers[i]->vs_hidden)
954 error = __svc_register(net, progp->pg_name, progp->pg_prog,
955 i, family, proto, port);
965 * If user space is running rpcbind, it should take the v4 UNSET
966 * and clear everything for this [program, version]. If user space
967 * is running portmap, it will reject the v4 UNSET, but won't have
968 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
969 * in this case to clear all existing entries for [program, version].
971 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
972 const char *progname)
976 error = rpcb_v4_register(net, program, version, NULL, "");
979 * User space didn't support rpcbind v4, so retry this
980 * request with the legacy rpcbind v2 protocol.
982 if (error == -EPROTONOSUPPORT)
983 error = rpcb_register(net, program, version, 0, 0);
985 dprintk("svc: %s(%sv%u), error %d\n",
986 __func__, progname, version, error);
990 * All netids, bind addresses and ports registered for [program, version]
991 * are removed from the local rpcbind database (if the service is not
992 * hidden) to make way for a new instance of the service.
994 * The result of unregistration is reported via dprintk for those who want
995 * verification of the result, but is otherwise not important.
997 static void svc_unregister(const struct svc_serv *serv, struct net *net)
999 struct svc_program *progp;
1000 unsigned long flags;
1003 clear_thread_flag(TIF_SIGPENDING);
1005 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1006 for (i = 0; i < progp->pg_nvers; i++) {
1007 if (progp->pg_vers[i] == NULL)
1009 if (progp->pg_vers[i]->vs_hidden)
1012 dprintk("svc: attempting to unregister %sv%u\n",
1014 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1018 spin_lock_irqsave(¤t->sighand->siglock, flags);
1019 recalc_sigpending();
1020 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1024 * Printk the given error with the address of the client that caused it.
1026 static __printf(2, 3)
1027 int svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1031 char buf[RPC_MAX_ADDRBUFLEN];
1033 if (!net_ratelimit())
1036 printk(KERN_WARNING "svc: %s: ",
1037 svc_print_addr(rqstp, buf, sizeof(buf)));
1039 va_start(args, fmt);
1040 r = vprintk(fmt, args);
1047 * Common routine for processing the RPC request.
1050 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1052 struct svc_program *progp;
1053 struct svc_version *versp = NULL; /* compiler food */
1054 struct svc_procedure *procp = NULL;
1055 struct svc_serv *serv = rqstp->rq_server;
1058 u32 prog, vers, proc;
1059 __be32 auth_stat, rpc_stat;
1061 __be32 *reply_statp;
1063 rpc_stat = rpc_success;
1065 if (argv->iov_len < 6*4)
1068 /* Will be turned off only in gss privacy case: */
1069 rqstp->rq_splice_ok = 1;
1070 /* Will be turned off only when NFSv4 Sessions are used */
1071 rqstp->rq_usedeferral = 1;
1072 rqstp->rq_dropme = false;
1074 /* Setup reply header */
1075 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1077 svc_putu32(resv, rqstp->rq_xid);
1079 vers = svc_getnl(argv);
1081 /* First words of reply: */
1082 svc_putnl(resv, 1); /* REPLY */
1084 if (vers != 2) /* RPC version number */
1087 /* Save position in case we later decide to reject: */
1088 reply_statp = resv->iov_base + resv->iov_len;
1090 svc_putnl(resv, 0); /* ACCEPT */
1092 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1093 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1094 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1096 progp = serv->sv_program;
1098 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1099 if (prog == progp->pg_prog)
1103 * Decode auth data, and add verifier to reply buffer.
1104 * We do this before anything else in order to get a decent
1107 auth_res = svc_authenticate(rqstp, &auth_stat);
1108 /* Also give the program a chance to reject this call: */
1109 if (auth_res == SVC_OK && progp) {
1110 auth_stat = rpc_autherr_badcred;
1111 auth_res = progp->pg_authenticate(rqstp);
1119 rpc_stat = rpc_system_err;
1124 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1125 svc_close_xprt(rqstp->rq_xprt);
1135 if (vers >= progp->pg_nvers ||
1136 !(versp = progp->pg_vers[vers]))
1139 procp = versp->vs_proc + proc;
1140 if (proc >= versp->vs_nproc || !procp->pc_func)
1142 rqstp->rq_procinfo = procp;
1144 /* Syntactic check complete */
1145 serv->sv_stats->rpccnt++;
1147 /* Build the reply header. */
1148 statp = resv->iov_base +resv->iov_len;
1149 svc_putnl(resv, RPC_SUCCESS);
1151 /* Bump per-procedure stats counter */
1154 /* Initialize storage for argp and resp */
1155 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1156 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1158 /* un-reserve some of the out-queue now that we have a
1159 * better idea of reply size
1161 if (procp->pc_xdrressize)
1162 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1164 /* Call the function that processes the request. */
1165 if (!versp->vs_dispatch) {
1166 /* Decode arguments */
1167 xdr = procp->pc_decode;
1168 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1171 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1174 if (rqstp->rq_dropme) {
1175 if (procp->pc_release)
1176 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1179 if (*statp == rpc_success &&
1180 (xdr = procp->pc_encode) &&
1181 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1182 dprintk("svc: failed to encode reply\n");
1183 /* serv->sv_stats->rpcsystemerr++; */
1184 *statp = rpc_system_err;
1187 dprintk("svc: calling dispatcher\n");
1188 if (!versp->vs_dispatch(rqstp, statp)) {
1189 /* Release reply info */
1190 if (procp->pc_release)
1191 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1196 /* Check RPC status result */
1197 if (*statp != rpc_success)
1198 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1200 /* Release reply info */
1201 if (procp->pc_release)
1202 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1204 if (procp->pc_encode == NULL)
1208 if (svc_authorise(rqstp))
1210 return 1; /* Caller can now send it */
1213 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1214 dprintk("svc: svc_process dropit\n");
1218 svc_printk(rqstp, "short len %Zd, dropping request\n",
1221 goto dropit; /* drop request */
1224 serv->sv_stats->rpcbadfmt++;
1225 svc_putnl(resv, 1); /* REJECT */
1226 svc_putnl(resv, 0); /* RPC_MISMATCH */
1227 svc_putnl(resv, 2); /* Only RPCv2 supported */
1232 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1233 serv->sv_stats->rpcbadauth++;
1234 /* Restore write pointer to location of accept status: */
1235 xdr_ressize_check(rqstp, reply_statp);
1236 svc_putnl(resv, 1); /* REJECT */
1237 svc_putnl(resv, 1); /* AUTH_ERROR */
1238 svc_putnl(resv, ntohl(auth_stat)); /* status */
1242 dprintk("svc: unknown program %d\n", prog);
1243 serv->sv_stats->rpcbadfmt++;
1244 svc_putnl(resv, RPC_PROG_UNAVAIL);
1248 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1249 vers, prog, progp->pg_name);
1251 serv->sv_stats->rpcbadfmt++;
1252 svc_putnl(resv, RPC_PROG_MISMATCH);
1253 svc_putnl(resv, progp->pg_lovers);
1254 svc_putnl(resv, progp->pg_hivers);
1258 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1260 serv->sv_stats->rpcbadfmt++;
1261 svc_putnl(resv, RPC_PROC_UNAVAIL);
1265 svc_printk(rqstp, "failed to decode args\n");
1267 rpc_stat = rpc_garbage_args;
1269 serv->sv_stats->rpcbadfmt++;
1270 svc_putnl(resv, ntohl(rpc_stat));
1273 EXPORT_SYMBOL_GPL(svc_process);
1276 * Process the RPC request.
1279 svc_process(struct svc_rqst *rqstp)
1281 struct kvec *argv = &rqstp->rq_arg.head[0];
1282 struct kvec *resv = &rqstp->rq_res.head[0];
1283 struct svc_serv *serv = rqstp->rq_server;
1287 * Setup response xdr_buf.
1288 * Initially it has just one page
1290 rqstp->rq_resused = 1;
1291 resv->iov_base = page_address(rqstp->rq_respages[0]);
1293 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1294 rqstp->rq_res.len = 0;
1295 rqstp->rq_res.page_base = 0;
1296 rqstp->rq_res.page_len = 0;
1297 rqstp->rq_res.buflen = PAGE_SIZE;
1298 rqstp->rq_res.tail[0].iov_base = NULL;
1299 rqstp->rq_res.tail[0].iov_len = 0;
1301 rqstp->rq_xid = svc_getu32(argv);
1303 dir = svc_getnl(argv);
1305 /* direction != CALL */
1306 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1307 serv->sv_stats->rpcbadfmt++;
1312 /* Returns 1 for send, 0 for drop */
1313 if (svc_process_common(rqstp, argv, resv))
1314 return svc_send(rqstp);
1321 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1323 * Process a backchannel RPC request that arrived over an existing
1324 * outbound connection
1327 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1328 struct svc_rqst *rqstp)
1330 struct kvec *argv = &rqstp->rq_arg.head[0];
1331 struct kvec *resv = &rqstp->rq_res.head[0];
1333 /* Build the svc_rqst used by the common processing routine */
1334 rqstp->rq_xprt = serv->sv_bc_xprt;
1335 rqstp->rq_xid = req->rq_xid;
1336 rqstp->rq_prot = req->rq_xprt->prot;
1337 rqstp->rq_server = serv;
1339 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1340 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1341 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1342 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1344 /* reset result send buffer "put" position */
1347 if (rqstp->rq_prot != IPPROTO_TCP) {
1348 printk(KERN_ERR "No support for Non-TCP transports!\n");
1353 * Skip the next two words because they've already been
1354 * processed in the trasport
1356 svc_getu32(argv); /* XID */
1357 svc_getnl(argv); /* CALLDIR */
1359 /* Returns 1 for send, 0 for drop */
1360 if (svc_process_common(rqstp, argv, resv)) {
1361 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1362 sizeof(req->rq_snd_buf));
1363 return bc_send(req);
1365 /* Nothing to do to drop request */
1369 EXPORT_SYMBOL_GPL(bc_svc_process);
1370 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1373 * Return (transport-specific) limit on the rpc payload.
1375 u32 svc_max_payload(const struct svc_rqst *rqstp)
1377 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1379 if (rqstp->rq_server->sv_max_payload < max)
1380 max = rqstp->rq_server->sv_max_payload;
1383 EXPORT_SYMBOL_GPL(svc_max_payload);