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>
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);
299 * Set the given thread's cpus_allowed mask so that it
300 * will only run on cpus in the given pool.
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);
315 case SVC_POOL_PERCPU:
317 set_cpus_allowed_ptr(task, cpumask_of(node));
320 case SVC_POOL_PERNODE:
322 set_cpus_allowed_ptr(task, cpumask_of_node(node));
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.
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)) {
346 case SVC_POOL_PERCPU:
347 pidx = m->to_pool[cpu];
349 case SVC_POOL_PERNODE:
350 pidx = m->to_pool[cpu_to_node(cpu)];
354 return &serv->sv_pools[pidx % serv->sv_nrpools];
357 static int svc_rpcb_setup(struct svc_serv *serv)
361 err = rpcb_create_local();
365 /* Remove any stale portmap registrations */
366 svc_unregister(serv);
370 static void svc_rpcb_cleanup(struct svc_serv *serv)
372 svc_unregister(serv);
376 static int svc_uses_rpcbind(struct svc_serv *serv)
378 struct svc_program *progp;
381 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
382 for (i = 0; i < progp->pg_nvers; i++) {
383 if (progp->pg_vers[i] == NULL)
385 if (progp->pg_vers[i]->vs_hidden == 0)
394 * Create an RPC service
396 static struct svc_serv *
397 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
398 void (*shutdown)(struct svc_serv *serv))
400 struct svc_serv *serv;
402 unsigned int xdrsize;
405 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
407 serv->sv_name = prog->pg_name;
408 serv->sv_program = prog;
409 serv->sv_nrthreads = 1;
410 serv->sv_stats = prog->pg_stats;
411 if (bufsize > RPCSVC_MAXPAYLOAD)
412 bufsize = RPCSVC_MAXPAYLOAD;
413 serv->sv_max_payload = bufsize? bufsize : 4096;
414 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
415 serv->sv_shutdown = shutdown;
418 prog->pg_lovers = prog->pg_nvers-1;
419 for (vers=0; vers<prog->pg_nvers ; vers++)
420 if (prog->pg_vers[vers]) {
421 prog->pg_hivers = vers;
422 if (prog->pg_lovers > vers)
423 prog->pg_lovers = vers;
424 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
425 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
427 prog = prog->pg_next;
429 serv->sv_xdrsize = xdrsize;
430 INIT_LIST_HEAD(&serv->sv_tempsocks);
431 INIT_LIST_HEAD(&serv->sv_permsocks);
432 init_timer(&serv->sv_temptimer);
433 spin_lock_init(&serv->sv_lock);
435 serv->sv_nrpools = npools;
437 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
439 if (!serv->sv_pools) {
444 for (i = 0; i < serv->sv_nrpools; i++) {
445 struct svc_pool *pool = &serv->sv_pools[i];
447 dprintk("svc: initialising pool %u for %s\n",
451 INIT_LIST_HEAD(&pool->sp_threads);
452 INIT_LIST_HEAD(&pool->sp_sockets);
453 INIT_LIST_HEAD(&pool->sp_all_threads);
454 spin_lock_init(&pool->sp_lock);
457 if (svc_uses_rpcbind(serv)) {
458 if (svc_rpcb_setup(serv) < 0) {
459 kfree(serv->sv_pools);
463 if (!serv->sv_shutdown)
464 serv->sv_shutdown = svc_rpcb_cleanup;
471 svc_create(struct svc_program *prog, unsigned int bufsize,
472 void (*shutdown)(struct svc_serv *serv))
474 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
476 EXPORT_SYMBOL_GPL(svc_create);
479 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
480 void (*shutdown)(struct svc_serv *serv),
481 svc_thread_fn func, struct module *mod)
483 struct svc_serv *serv;
484 unsigned int npools = svc_pool_map_get();
486 serv = __svc_create(prog, bufsize, npools, shutdown);
489 serv->sv_function = func;
490 serv->sv_module = mod;
495 EXPORT_SYMBOL_GPL(svc_create_pooled);
498 * Destroy an RPC service. Should be called with appropriate locking to
499 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
502 svc_destroy(struct svc_serv *serv)
504 dprintk("svc: svc_destroy(%s, %d)\n",
505 serv->sv_program->pg_name,
508 if (serv->sv_nrthreads) {
509 if (--(serv->sv_nrthreads) != 0) {
510 svc_sock_update_bufs(serv);
514 printk("svc_destroy: no threads for serv=%p!\n", serv);
516 del_timer_sync(&serv->sv_temptimer);
518 svc_close_all(&serv->sv_tempsocks);
520 if (serv->sv_shutdown)
521 serv->sv_shutdown(serv);
523 svc_close_all(&serv->sv_permsocks);
525 BUG_ON(!list_empty(&serv->sv_permsocks));
526 BUG_ON(!list_empty(&serv->sv_tempsocks));
528 cache_clean_deferred(serv);
530 if (svc_serv_is_pooled(serv))
533 svc_unregister(serv);
534 kfree(serv->sv_pools);
537 EXPORT_SYMBOL_GPL(svc_destroy);
540 * Allocate an RPC server's buffer space.
541 * We allocate pages and place them in rq_argpages.
544 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
546 unsigned int pages, arghi;
548 /* bc_xprt uses fore channel allocated buffers */
549 if (svc_is_backchannel(rqstp))
552 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
553 * We assume one is at most one page
556 BUG_ON(pages > RPCSVC_MAXPAGES);
558 struct page *p = alloc_page(GFP_KERNEL);
561 rqstp->rq_pages[arghi++] = p;
568 * Release an RPC server buffer
571 svc_release_buffer(struct svc_rqst *rqstp)
575 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
576 if (rqstp->rq_pages[i])
577 put_page(rqstp->rq_pages[i]);
581 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
583 struct svc_rqst *rqstp;
585 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
589 init_waitqueue_head(&rqstp->rq_wait);
591 serv->sv_nrthreads++;
592 spin_lock_bh(&pool->sp_lock);
593 pool->sp_nrthreads++;
594 list_add(&rqstp->rq_all, &pool->sp_all_threads);
595 spin_unlock_bh(&pool->sp_lock);
596 rqstp->rq_server = serv;
597 rqstp->rq_pool = pool;
599 rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
603 rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
607 if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
612 svc_exit_thread(rqstp);
614 return ERR_PTR(-ENOMEM);
616 EXPORT_SYMBOL_GPL(svc_prepare_thread);
619 * Choose a pool in which to create a new thread, for svc_set_num_threads
621 static inline struct svc_pool *
622 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
627 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
631 * Choose a thread to kill, for svc_set_num_threads
633 static inline struct task_struct *
634 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
637 struct task_struct *task = NULL;
640 spin_lock_bh(&pool->sp_lock);
642 /* choose a pool in round-robin fashion */
643 for (i = 0; i < serv->sv_nrpools; i++) {
644 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
645 spin_lock_bh(&pool->sp_lock);
646 if (!list_empty(&pool->sp_all_threads))
648 spin_unlock_bh(&pool->sp_lock);
654 if (!list_empty(&pool->sp_all_threads)) {
655 struct svc_rqst *rqstp;
658 * Remove from the pool->sp_all_threads list
659 * so we don't try to kill it again.
661 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
662 list_del_init(&rqstp->rq_all);
663 task = rqstp->rq_task;
665 spin_unlock_bh(&pool->sp_lock);
671 * Create or destroy enough new threads to make the number
672 * of threads the given number. If `pool' is non-NULL, applies
673 * only to threads in that pool, otherwise round-robins between
674 * all pools. Must be called with a svc_get() reference and
675 * the BKL or another lock to protect access to svc_serv fields.
677 * Destroying threads relies on the service threads filling in
678 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
679 * has been created using svc_create_pooled().
681 * Based on code that used to be in nfsd_svc() but tweaked
685 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
687 struct svc_rqst *rqstp;
688 struct task_struct *task;
689 struct svc_pool *chosen_pool;
691 unsigned int state = serv->sv_nrthreads-1;
694 /* The -1 assumes caller has done a svc_get() */
695 nrservs -= (serv->sv_nrthreads-1);
697 spin_lock_bh(&pool->sp_lock);
698 nrservs -= pool->sp_nrthreads;
699 spin_unlock_bh(&pool->sp_lock);
702 /* create new threads */
703 while (nrservs > 0) {
705 chosen_pool = choose_pool(serv, pool, &state);
707 rqstp = svc_prepare_thread(serv, chosen_pool);
709 error = PTR_ERR(rqstp);
713 __module_get(serv->sv_module);
714 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
716 error = PTR_ERR(task);
717 module_put(serv->sv_module);
718 svc_exit_thread(rqstp);
722 rqstp->rq_task = task;
723 if (serv->sv_nrpools > 1)
724 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
726 svc_sock_update_bufs(serv);
727 wake_up_process(task);
729 /* destroy old threads */
730 while (nrservs < 0 &&
731 (task = choose_victim(serv, pool, &state)) != NULL) {
732 send_sig(SIGINT, task, 1);
738 EXPORT_SYMBOL_GPL(svc_set_num_threads);
741 * Called from a server thread as it's exiting. Caller must hold the BKL or
742 * the "service mutex", whichever is appropriate for the service.
745 svc_exit_thread(struct svc_rqst *rqstp)
747 struct svc_serv *serv = rqstp->rq_server;
748 struct svc_pool *pool = rqstp->rq_pool;
750 svc_release_buffer(rqstp);
751 kfree(rqstp->rq_resp);
752 kfree(rqstp->rq_argp);
753 kfree(rqstp->rq_auth_data);
755 spin_lock_bh(&pool->sp_lock);
756 pool->sp_nrthreads--;
757 list_del(&rqstp->rq_all);
758 spin_unlock_bh(&pool->sp_lock);
762 /* Release the server */
766 EXPORT_SYMBOL_GPL(svc_exit_thread);
769 * Register an "inet" protocol family netid with the local
770 * rpcbind daemon via an rpcbind v4 SET request.
772 * No netconfig infrastructure is available in the kernel, so
773 * we map IP_ protocol numbers to netids by hand.
775 * Returns zero on success; a negative errno value is returned
776 * if any error occurs.
778 static int __svc_rpcb_register4(const u32 program, const u32 version,
779 const unsigned short protocol,
780 const unsigned short port)
782 const struct sockaddr_in sin = {
783 .sin_family = AF_INET,
784 .sin_addr.s_addr = htonl(INADDR_ANY),
785 .sin_port = htons(port),
792 netid = RPCBIND_NETID_UDP;
795 netid = RPCBIND_NETID_TCP;
801 error = rpcb_v4_register(program, version,
802 (const struct sockaddr *)&sin, netid);
805 * User space didn't support rpcbind v4, so retry this
806 * registration request with the legacy rpcbind v2 protocol.
808 if (error == -EPROTONOSUPPORT)
809 error = rpcb_register(program, version, protocol, port);
814 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
816 * Register an "inet6" protocol family netid with the local
817 * rpcbind daemon via an rpcbind v4 SET request.
819 * No netconfig infrastructure is available in the kernel, so
820 * we map IP_ protocol numbers to netids by hand.
822 * Returns zero on success; a negative errno value is returned
823 * if any error occurs.
825 static int __svc_rpcb_register6(const u32 program, const u32 version,
826 const unsigned short protocol,
827 const unsigned short port)
829 const struct sockaddr_in6 sin6 = {
830 .sin6_family = AF_INET6,
831 .sin6_addr = IN6ADDR_ANY_INIT,
832 .sin6_port = htons(port),
839 netid = RPCBIND_NETID_UDP6;
842 netid = RPCBIND_NETID_TCP6;
848 error = rpcb_v4_register(program, version,
849 (const struct sockaddr *)&sin6, netid);
852 * User space didn't support rpcbind version 4, so we won't
853 * use a PF_INET6 listener.
855 if (error == -EPROTONOSUPPORT)
856 error = -EAFNOSUPPORT;
860 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
863 * Register a kernel RPC service via rpcbind version 4.
865 * Returns zero on success; a negative errno value is returned
866 * if any error occurs.
868 static int __svc_register(const char *progname,
869 const u32 program, const u32 version,
871 const unsigned short protocol,
872 const unsigned short port)
874 int error = -EAFNOSUPPORT;
878 error = __svc_rpcb_register4(program, version,
881 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
883 error = __svc_rpcb_register6(program, version,
885 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
889 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
890 "service (errno %d).\n", progname, version, -error);
895 * svc_register - register an RPC service with the local portmapper
896 * @serv: svc_serv struct for the service to register
897 * @family: protocol family of service's listener socket
898 * @proto: transport protocol number to advertise
899 * @port: port to advertise
901 * Service is registered for any address in the passed-in protocol family
903 int svc_register(const struct svc_serv *serv, const int family,
904 const unsigned short proto, const unsigned short port)
906 struct svc_program *progp;
910 BUG_ON(proto == 0 && port == 0);
912 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
913 for (i = 0; i < progp->pg_nvers; i++) {
914 if (progp->pg_vers[i] == NULL)
917 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
920 proto == IPPROTO_UDP? "udp" : "tcp",
923 progp->pg_vers[i]->vs_hidden?
924 " (but not telling portmap)" : "");
926 if (progp->pg_vers[i]->vs_hidden)
929 error = __svc_register(progp->pg_name, progp->pg_prog,
930 i, family, proto, port);
940 * If user space is running rpcbind, it should take the v4 UNSET
941 * and clear everything for this [program, version]. If user space
942 * is running portmap, it will reject the v4 UNSET, but won't have
943 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
944 * in this case to clear all existing entries for [program, version].
946 static void __svc_unregister(const u32 program, const u32 version,
947 const char *progname)
951 error = rpcb_v4_register(program, version, NULL, "");
954 * User space didn't support rpcbind v4, so retry this
955 * request with the legacy rpcbind v2 protocol.
957 if (error == -EPROTONOSUPPORT)
958 error = rpcb_register(program, version, 0, 0);
960 dprintk("svc: %s(%sv%u), error %d\n",
961 __func__, progname, version, error);
965 * All netids, bind addresses and ports registered for [program, version]
966 * are removed from the local rpcbind database (if the service is not
967 * hidden) to make way for a new instance of the service.
969 * The result of unregistration is reported via dprintk for those who want
970 * verification of the result, but is otherwise not important.
972 static void svc_unregister(const struct svc_serv *serv)
974 struct svc_program *progp;
978 clear_thread_flag(TIF_SIGPENDING);
980 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
981 for (i = 0; i < progp->pg_nvers; i++) {
982 if (progp->pg_vers[i] == NULL)
984 if (progp->pg_vers[i]->vs_hidden)
987 dprintk("svc: attempting to unregister %sv%u\n",
989 __svc_unregister(progp->pg_prog, i, progp->pg_name);
993 spin_lock_irqsave(¤t->sighand->siglock, flags);
995 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
999 * Printk the given error with the address of the client that caused it.
1002 __attribute__ ((format (printf, 2, 3)))
1003 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1007 char buf[RPC_MAX_ADDRBUFLEN];
1009 if (!net_ratelimit())
1012 printk(KERN_WARNING "svc: %s: ",
1013 svc_print_addr(rqstp, buf, sizeof(buf)));
1015 va_start(args, fmt);
1016 r = vprintk(fmt, args);
1023 * Common routine for processing the RPC request.
1026 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1028 struct svc_program *progp;
1029 struct svc_version *versp = NULL; /* compiler food */
1030 struct svc_procedure *procp = NULL;
1031 struct svc_serv *serv = rqstp->rq_server;
1034 u32 prog, vers, proc;
1035 __be32 auth_stat, rpc_stat;
1037 __be32 *reply_statp;
1039 rpc_stat = rpc_success;
1041 if (argv->iov_len < 6*4)
1044 /* Will be turned off only in gss privacy case: */
1045 rqstp->rq_splice_ok = 1;
1046 /* Will be turned off only when NFSv4 Sessions are used */
1047 rqstp->rq_usedeferral = 1;
1048 rqstp->rq_dropme = false;
1050 /* Setup reply header */
1051 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1053 svc_putu32(resv, rqstp->rq_xid);
1055 vers = svc_getnl(argv);
1057 /* First words of reply: */
1058 svc_putnl(resv, 1); /* REPLY */
1060 if (vers != 2) /* RPC version number */
1063 /* Save position in case we later decide to reject: */
1064 reply_statp = resv->iov_base + resv->iov_len;
1066 svc_putnl(resv, 0); /* ACCEPT */
1068 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1069 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1070 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1072 progp = serv->sv_program;
1074 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1075 if (prog == progp->pg_prog)
1079 * Decode auth data, and add verifier to reply buffer.
1080 * We do this before anything else in order to get a decent
1083 auth_res = svc_authenticate(rqstp, &auth_stat);
1084 /* Also give the program a chance to reject this call: */
1085 if (auth_res == SVC_OK && progp) {
1086 auth_stat = rpc_autherr_badcred;
1087 auth_res = progp->pg_authenticate(rqstp);
1095 rpc_stat = rpc_system_err;
1100 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1101 svc_close_xprt(rqstp->rq_xprt);
1111 if (vers >= progp->pg_nvers ||
1112 !(versp = progp->pg_vers[vers]))
1115 procp = versp->vs_proc + proc;
1116 if (proc >= versp->vs_nproc || !procp->pc_func)
1118 rqstp->rq_procinfo = procp;
1120 /* Syntactic check complete */
1121 serv->sv_stats->rpccnt++;
1123 /* Build the reply header. */
1124 statp = resv->iov_base +resv->iov_len;
1125 svc_putnl(resv, RPC_SUCCESS);
1127 /* Bump per-procedure stats counter */
1130 /* Initialize storage for argp and resp */
1131 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1132 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1134 /* un-reserve some of the out-queue now that we have a
1135 * better idea of reply size
1137 if (procp->pc_xdrressize)
1138 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1140 /* Call the function that processes the request. */
1141 if (!versp->vs_dispatch) {
1142 /* Decode arguments */
1143 xdr = procp->pc_decode;
1144 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1147 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1150 if (rqstp->rq_dropme) {
1151 if (procp->pc_release)
1152 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1155 if (*statp == rpc_success &&
1156 (xdr = procp->pc_encode) &&
1157 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1158 dprintk("svc: failed to encode reply\n");
1159 /* serv->sv_stats->rpcsystemerr++; */
1160 *statp = rpc_system_err;
1163 dprintk("svc: calling dispatcher\n");
1164 if (!versp->vs_dispatch(rqstp, statp)) {
1165 /* Release reply info */
1166 if (procp->pc_release)
1167 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1172 /* Check RPC status result */
1173 if (*statp != rpc_success)
1174 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1176 /* Release reply info */
1177 if (procp->pc_release)
1178 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1180 if (procp->pc_encode == NULL)
1184 if (svc_authorise(rqstp))
1186 return 1; /* Caller can now send it */
1189 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1190 dprintk("svc: svc_process dropit\n");
1194 svc_printk(rqstp, "short len %Zd, dropping request\n",
1197 goto dropit; /* drop request */
1200 serv->sv_stats->rpcbadfmt++;
1201 svc_putnl(resv, 1); /* REJECT */
1202 svc_putnl(resv, 0); /* RPC_MISMATCH */
1203 svc_putnl(resv, 2); /* Only RPCv2 supported */
1208 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1209 serv->sv_stats->rpcbadauth++;
1210 /* Restore write pointer to location of accept status: */
1211 xdr_ressize_check(rqstp, reply_statp);
1212 svc_putnl(resv, 1); /* REJECT */
1213 svc_putnl(resv, 1); /* AUTH_ERROR */
1214 svc_putnl(resv, ntohl(auth_stat)); /* status */
1218 dprintk("svc: unknown program %d\n", prog);
1219 serv->sv_stats->rpcbadfmt++;
1220 svc_putnl(resv, RPC_PROG_UNAVAIL);
1224 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1225 vers, prog, progp->pg_name);
1227 serv->sv_stats->rpcbadfmt++;
1228 svc_putnl(resv, RPC_PROG_MISMATCH);
1229 svc_putnl(resv, progp->pg_lovers);
1230 svc_putnl(resv, progp->pg_hivers);
1234 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1236 serv->sv_stats->rpcbadfmt++;
1237 svc_putnl(resv, RPC_PROC_UNAVAIL);
1241 svc_printk(rqstp, "failed to decode args\n");
1243 rpc_stat = rpc_garbage_args;
1245 serv->sv_stats->rpcbadfmt++;
1246 svc_putnl(resv, ntohl(rpc_stat));
1249 EXPORT_SYMBOL_GPL(svc_process);
1252 * Process the RPC request.
1255 svc_process(struct svc_rqst *rqstp)
1257 struct kvec *argv = &rqstp->rq_arg.head[0];
1258 struct kvec *resv = &rqstp->rq_res.head[0];
1259 struct svc_serv *serv = rqstp->rq_server;
1263 * Setup response xdr_buf.
1264 * Initially it has just one page
1266 rqstp->rq_resused = 1;
1267 resv->iov_base = page_address(rqstp->rq_respages[0]);
1269 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1270 rqstp->rq_res.len = 0;
1271 rqstp->rq_res.page_base = 0;
1272 rqstp->rq_res.page_len = 0;
1273 rqstp->rq_res.buflen = PAGE_SIZE;
1274 rqstp->rq_res.tail[0].iov_base = NULL;
1275 rqstp->rq_res.tail[0].iov_len = 0;
1277 rqstp->rq_xid = svc_getu32(argv);
1279 dir = svc_getnl(argv);
1281 /* direction != CALL */
1282 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1283 serv->sv_stats->rpcbadfmt++;
1288 /* Returns 1 for send, 0 for drop */
1289 if (svc_process_common(rqstp, argv, resv))
1290 return svc_send(rqstp);
1297 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1299 * Process a backchannel RPC request that arrived over an existing
1300 * outbound connection
1303 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1304 struct svc_rqst *rqstp)
1306 struct kvec *argv = &rqstp->rq_arg.head[0];
1307 struct kvec *resv = &rqstp->rq_res.head[0];
1309 /* Build the svc_rqst used by the common processing routine */
1310 rqstp->rq_xprt = serv->sv_bc_xprt;
1311 rqstp->rq_xid = req->rq_xid;
1312 rqstp->rq_prot = req->rq_xprt->prot;
1313 rqstp->rq_server = serv;
1315 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1316 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1317 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1318 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1320 /* reset result send buffer "put" position */
1323 if (rqstp->rq_prot != IPPROTO_TCP) {
1324 printk(KERN_ERR "No support for Non-TCP transports!\n");
1329 * Skip the next two words because they've already been
1330 * processed in the trasport
1332 svc_getu32(argv); /* XID */
1333 svc_getnl(argv); /* CALLDIR */
1335 /* Returns 1 for send, 0 for drop */
1336 if (svc_process_common(rqstp, argv, resv)) {
1337 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1338 sizeof(req->rq_snd_buf));
1339 return bc_send(req);
1341 /* Nothing to do to drop request */
1345 EXPORT_SYMBOL_GPL(bc_svc_process);
1346 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1349 * Return (transport-specific) limit on the rpc payload.
1351 u32 svc_max_payload(const struct svc_rqst *rqstp)
1353 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1355 if (rqstp->rq_server->sv_max_payload < max)
1356 max = rqstp->rq_server->sv_max_payload;
1359 EXPORT_SYMBOL_GPL(svc_max_payload);