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);
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);
289 m->mode = SVC_POOL_DEFAULT;
297 mutex_unlock(&svc_pool_map_mutex);
302 * Set the given thread's cpus_allowed mask so that it
303 * will only run on cpus in the given pool.
306 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
308 struct svc_pool_map *m = &svc_pool_map;
309 unsigned int node = m->pool_to[pidx];
312 * The caller checks for sv_nrpools > 1, which
313 * implies that we've been initialized.
315 BUG_ON(m->count == 0);
318 case SVC_POOL_PERCPU:
320 set_cpus_allowed_ptr(task, cpumask_of(node));
323 case SVC_POOL_PERNODE:
325 set_cpus_allowed_ptr(task, cpumask_of_node(node));
332 * Use the mapping mode to choose a pool for a given CPU.
333 * Used when enqueueing an incoming RPC. Always returns
334 * a non-NULL pool pointer.
337 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
339 struct svc_pool_map *m = &svc_pool_map;
340 unsigned int pidx = 0;
343 * An uninitialised map happens in a pure client when
344 * lockd is brought up, so silently treat it the
345 * same as SVC_POOL_GLOBAL.
347 if (svc_serv_is_pooled(serv)) {
349 case SVC_POOL_PERCPU:
350 pidx = m->to_pool[cpu];
352 case SVC_POOL_PERNODE:
353 pidx = m->to_pool[cpu_to_node(cpu)];
357 return &serv->sv_pools[pidx % serv->sv_nrpools];
362 * Create an RPC service
364 static struct svc_serv *
365 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
366 void (*shutdown)(struct svc_serv *serv))
368 struct svc_serv *serv;
370 unsigned int xdrsize;
373 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
375 serv->sv_name = prog->pg_name;
376 serv->sv_program = prog;
377 serv->sv_nrthreads = 1;
378 serv->sv_stats = prog->pg_stats;
379 if (bufsize > RPCSVC_MAXPAYLOAD)
380 bufsize = RPCSVC_MAXPAYLOAD;
381 serv->sv_max_payload = bufsize? bufsize : 4096;
382 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
383 serv->sv_shutdown = shutdown;
386 prog->pg_lovers = prog->pg_nvers-1;
387 for (vers=0; vers<prog->pg_nvers ; vers++)
388 if (prog->pg_vers[vers]) {
389 prog->pg_hivers = vers;
390 if (prog->pg_lovers > vers)
391 prog->pg_lovers = vers;
392 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
393 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
395 prog = prog->pg_next;
397 serv->sv_xdrsize = xdrsize;
398 INIT_LIST_HEAD(&serv->sv_tempsocks);
399 INIT_LIST_HEAD(&serv->sv_permsocks);
400 init_timer(&serv->sv_temptimer);
401 spin_lock_init(&serv->sv_lock);
403 serv->sv_nrpools = npools;
405 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
407 if (!serv->sv_pools) {
412 for (i = 0; i < serv->sv_nrpools; i++) {
413 struct svc_pool *pool = &serv->sv_pools[i];
415 dprintk("svc: initialising pool %u for %s\n",
419 INIT_LIST_HEAD(&pool->sp_threads);
420 INIT_LIST_HEAD(&pool->sp_sockets);
421 INIT_LIST_HEAD(&pool->sp_all_threads);
422 spin_lock_init(&pool->sp_lock);
425 /* Remove any stale portmap registrations */
426 svc_unregister(serv);
432 svc_create(struct svc_program *prog, unsigned int bufsize,
433 void (*shutdown)(struct svc_serv *serv))
435 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
437 EXPORT_SYMBOL_GPL(svc_create);
440 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
441 void (*shutdown)(struct svc_serv *serv),
442 svc_thread_fn func, struct module *mod)
444 struct svc_serv *serv;
445 unsigned int npools = svc_pool_map_get();
447 serv = __svc_create(prog, bufsize, npools, shutdown);
450 serv->sv_function = func;
451 serv->sv_module = mod;
456 EXPORT_SYMBOL_GPL(svc_create_pooled);
459 * Destroy an RPC service. Should be called with appropriate locking to
460 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
463 svc_destroy(struct svc_serv *serv)
465 dprintk("svc: svc_destroy(%s, %d)\n",
466 serv->sv_program->pg_name,
469 if (serv->sv_nrthreads) {
470 if (--(serv->sv_nrthreads) != 0) {
471 svc_sock_update_bufs(serv);
475 printk("svc_destroy: no threads for serv=%p!\n", serv);
477 del_timer_sync(&serv->sv_temptimer);
479 * The set of xprts (contained in the sv_tempsocks and
480 * sv_permsocks lists) is now constant, since it is modified
481 * only by accepting new sockets (done by service threads in
482 * svc_recv) or aging old ones (done by sv_temptimer), or
483 * configuration changes (excluded by whatever locking the
484 * caller is using--nfsd_mutex in the case of nfsd). So it's
485 * safe to traverse those lists and shut everything down:
489 if (serv->sv_shutdown)
490 serv->sv_shutdown(serv);
492 cache_clean_deferred(serv);
494 if (svc_serv_is_pooled(serv))
497 svc_unregister(serv);
498 kfree(serv->sv_pools);
501 EXPORT_SYMBOL_GPL(svc_destroy);
504 * Allocate an RPC server's buffer space.
505 * We allocate pages and place them in rq_argpages.
508 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
510 unsigned int pages, arghi;
512 /* bc_xprt uses fore channel allocated buffers */
513 if (svc_is_backchannel(rqstp))
516 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
517 * We assume one is at most one page
520 BUG_ON(pages > RPCSVC_MAXPAGES);
522 struct page *p = alloc_page(GFP_KERNEL);
525 rqstp->rq_pages[arghi++] = p;
532 * Release an RPC server buffer
535 svc_release_buffer(struct svc_rqst *rqstp)
539 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
540 if (rqstp->rq_pages[i])
541 put_page(rqstp->rq_pages[i]);
545 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
547 struct svc_rqst *rqstp;
549 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
553 init_waitqueue_head(&rqstp->rq_wait);
555 serv->sv_nrthreads++;
556 spin_lock_bh(&pool->sp_lock);
557 pool->sp_nrthreads++;
558 list_add(&rqstp->rq_all, &pool->sp_all_threads);
559 spin_unlock_bh(&pool->sp_lock);
560 rqstp->rq_server = serv;
561 rqstp->rq_pool = pool;
563 rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
567 rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
571 if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
576 svc_exit_thread(rqstp);
578 return ERR_PTR(-ENOMEM);
580 EXPORT_SYMBOL_GPL(svc_prepare_thread);
583 * Choose a pool in which to create a new thread, for svc_set_num_threads
585 static inline struct svc_pool *
586 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
591 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
595 * Choose a thread to kill, for svc_set_num_threads
597 static inline struct task_struct *
598 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
601 struct task_struct *task = NULL;
604 spin_lock_bh(&pool->sp_lock);
606 /* choose a pool in round-robin fashion */
607 for (i = 0; i < serv->sv_nrpools; i++) {
608 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
609 spin_lock_bh(&pool->sp_lock);
610 if (!list_empty(&pool->sp_all_threads))
612 spin_unlock_bh(&pool->sp_lock);
618 if (!list_empty(&pool->sp_all_threads)) {
619 struct svc_rqst *rqstp;
622 * Remove from the pool->sp_all_threads list
623 * so we don't try to kill it again.
625 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
626 list_del_init(&rqstp->rq_all);
627 task = rqstp->rq_task;
629 spin_unlock_bh(&pool->sp_lock);
635 * Create or destroy enough new threads to make the number
636 * of threads the given number. If `pool' is non-NULL, applies
637 * only to threads in that pool, otherwise round-robins between
638 * all pools. Must be called with a svc_get() reference and
639 * the BKL or another lock to protect access to svc_serv fields.
641 * Destroying threads relies on the service threads filling in
642 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
643 * has been created using svc_create_pooled().
645 * Based on code that used to be in nfsd_svc() but tweaked
649 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
651 struct svc_rqst *rqstp;
652 struct task_struct *task;
653 struct svc_pool *chosen_pool;
655 unsigned int state = serv->sv_nrthreads-1;
658 /* The -1 assumes caller has done a svc_get() */
659 nrservs -= (serv->sv_nrthreads-1);
661 spin_lock_bh(&pool->sp_lock);
662 nrservs -= pool->sp_nrthreads;
663 spin_unlock_bh(&pool->sp_lock);
666 /* create new threads */
667 while (nrservs > 0) {
669 chosen_pool = choose_pool(serv, pool, &state);
671 rqstp = svc_prepare_thread(serv, chosen_pool);
673 error = PTR_ERR(rqstp);
677 __module_get(serv->sv_module);
678 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
680 error = PTR_ERR(task);
681 module_put(serv->sv_module);
682 svc_exit_thread(rqstp);
686 rqstp->rq_task = task;
687 if (serv->sv_nrpools > 1)
688 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
690 svc_sock_update_bufs(serv);
691 wake_up_process(task);
693 /* destroy old threads */
694 while (nrservs < 0 &&
695 (task = choose_victim(serv, pool, &state)) != NULL) {
696 send_sig(SIGINT, task, 1);
702 EXPORT_SYMBOL_GPL(svc_set_num_threads);
705 * Called from a server thread as it's exiting. Caller must hold the BKL or
706 * the "service mutex", whichever is appropriate for the service.
709 svc_exit_thread(struct svc_rqst *rqstp)
711 struct svc_serv *serv = rqstp->rq_server;
712 struct svc_pool *pool = rqstp->rq_pool;
714 svc_release_buffer(rqstp);
715 kfree(rqstp->rq_resp);
716 kfree(rqstp->rq_argp);
717 kfree(rqstp->rq_auth_data);
719 spin_lock_bh(&pool->sp_lock);
720 pool->sp_nrthreads--;
721 list_del(&rqstp->rq_all);
722 spin_unlock_bh(&pool->sp_lock);
726 /* Release the server */
730 EXPORT_SYMBOL_GPL(svc_exit_thread);
733 * Register an "inet" protocol family netid with the local
734 * rpcbind daemon via an rpcbind v4 SET request.
736 * No netconfig infrastructure is available in the kernel, so
737 * we map IP_ protocol numbers to netids by hand.
739 * Returns zero on success; a negative errno value is returned
740 * if any error occurs.
742 static int __svc_rpcb_register4(const u32 program, const u32 version,
743 const unsigned short protocol,
744 const unsigned short port)
746 const struct sockaddr_in sin = {
747 .sin_family = AF_INET,
748 .sin_addr.s_addr = htonl(INADDR_ANY),
749 .sin_port = htons(port),
756 netid = RPCBIND_NETID_UDP;
759 netid = RPCBIND_NETID_TCP;
765 error = rpcb_v4_register(program, version,
766 (const struct sockaddr *)&sin, netid);
769 * User space didn't support rpcbind v4, so retry this
770 * registration request with the legacy rpcbind v2 protocol.
772 if (error == -EPROTONOSUPPORT)
773 error = rpcb_register(program, version, protocol, port);
778 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
780 * Register an "inet6" protocol family netid with the local
781 * rpcbind daemon via an rpcbind v4 SET request.
783 * No netconfig infrastructure is available in the kernel, so
784 * we map IP_ protocol numbers to netids by hand.
786 * Returns zero on success; a negative errno value is returned
787 * if any error occurs.
789 static int __svc_rpcb_register6(const u32 program, const u32 version,
790 const unsigned short protocol,
791 const unsigned short port)
793 const struct sockaddr_in6 sin6 = {
794 .sin6_family = AF_INET6,
795 .sin6_addr = IN6ADDR_ANY_INIT,
796 .sin6_port = htons(port),
803 netid = RPCBIND_NETID_UDP6;
806 netid = RPCBIND_NETID_TCP6;
812 error = rpcb_v4_register(program, version,
813 (const struct sockaddr *)&sin6, netid);
816 * User space didn't support rpcbind version 4, so we won't
817 * use a PF_INET6 listener.
819 if (error == -EPROTONOSUPPORT)
820 error = -EAFNOSUPPORT;
824 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
827 * Register a kernel RPC service via rpcbind version 4.
829 * Returns zero on success; a negative errno value is returned
830 * if any error occurs.
832 static int __svc_register(const char *progname,
833 const u32 program, const u32 version,
835 const unsigned short protocol,
836 const unsigned short port)
838 int error = -EAFNOSUPPORT;
842 error = __svc_rpcb_register4(program, version,
845 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
847 error = __svc_rpcb_register6(program, version,
849 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
853 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
854 "service (errno %d).\n", progname, version, -error);
859 * svc_register - register an RPC service with the local portmapper
860 * @serv: svc_serv struct for the service to register
861 * @family: protocol family of service's listener socket
862 * @proto: transport protocol number to advertise
863 * @port: port to advertise
865 * Service is registered for any address in the passed-in protocol family
867 int svc_register(const struct svc_serv *serv, const int family,
868 const unsigned short proto, const unsigned short port)
870 struct svc_program *progp;
874 BUG_ON(proto == 0 && port == 0);
876 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
877 for (i = 0; i < progp->pg_nvers; i++) {
878 if (progp->pg_vers[i] == NULL)
881 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
884 proto == IPPROTO_UDP? "udp" : "tcp",
887 progp->pg_vers[i]->vs_hidden?
888 " (but not telling portmap)" : "");
890 if (progp->pg_vers[i]->vs_hidden)
893 error = __svc_register(progp->pg_name, progp->pg_prog,
894 i, family, proto, port);
904 * If user space is running rpcbind, it should take the v4 UNSET
905 * and clear everything for this [program, version]. If user space
906 * is running portmap, it will reject the v4 UNSET, but won't have
907 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
908 * in this case to clear all existing entries for [program, version].
910 static void __svc_unregister(const u32 program, const u32 version,
911 const char *progname)
915 error = rpcb_v4_register(program, version, NULL, "");
918 * User space didn't support rpcbind v4, so retry this
919 * request with the legacy rpcbind v2 protocol.
921 if (error == -EPROTONOSUPPORT)
922 error = rpcb_register(program, version, 0, 0);
924 dprintk("svc: %s(%sv%u), error %d\n",
925 __func__, progname, version, error);
929 * All netids, bind addresses and ports registered for [program, version]
930 * are removed from the local rpcbind database (if the service is not
931 * hidden) to make way for a new instance of the service.
933 * The result of unregistration is reported via dprintk for those who want
934 * verification of the result, but is otherwise not important.
936 static void svc_unregister(const struct svc_serv *serv)
938 struct svc_program *progp;
942 clear_thread_flag(TIF_SIGPENDING);
944 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
945 for (i = 0; i < progp->pg_nvers; i++) {
946 if (progp->pg_vers[i] == NULL)
948 if (progp->pg_vers[i]->vs_hidden)
951 dprintk("svc: attempting to unregister %sv%u\n",
953 __svc_unregister(progp->pg_prog, i, progp->pg_name);
957 spin_lock_irqsave(¤t->sighand->siglock, flags);
959 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
963 * Printk the given error with the address of the client that caused it.
966 __attribute__ ((format (printf, 2, 3)))
967 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
971 char buf[RPC_MAX_ADDRBUFLEN];
973 if (!net_ratelimit())
976 printk(KERN_WARNING "svc: %s: ",
977 svc_print_addr(rqstp, buf, sizeof(buf)));
980 r = vprintk(fmt, args);
987 * Common routine for processing the RPC request.
990 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
992 struct svc_program *progp;
993 struct svc_version *versp = NULL; /* compiler food */
994 struct svc_procedure *procp = NULL;
995 struct svc_serv *serv = rqstp->rq_server;
998 u32 prog, vers, proc;
999 __be32 auth_stat, rpc_stat;
1001 __be32 *reply_statp;
1003 rpc_stat = rpc_success;
1005 if (argv->iov_len < 6*4)
1008 /* Will be turned off only in gss privacy case: */
1009 rqstp->rq_splice_ok = 1;
1010 /* Will be turned off only when NFSv4 Sessions are used */
1011 rqstp->rq_usedeferral = 1;
1012 rqstp->rq_dropme = false;
1014 /* Setup reply header */
1015 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1017 svc_putu32(resv, rqstp->rq_xid);
1019 vers = svc_getnl(argv);
1021 /* First words of reply: */
1022 svc_putnl(resv, 1); /* REPLY */
1024 if (vers != 2) /* RPC version number */
1027 /* Save position in case we later decide to reject: */
1028 reply_statp = resv->iov_base + resv->iov_len;
1030 svc_putnl(resv, 0); /* ACCEPT */
1032 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1033 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1034 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1036 progp = serv->sv_program;
1038 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1039 if (prog == progp->pg_prog)
1043 * Decode auth data, and add verifier to reply buffer.
1044 * We do this before anything else in order to get a decent
1047 auth_res = svc_authenticate(rqstp, &auth_stat);
1048 /* Also give the program a chance to reject this call: */
1049 if (auth_res == SVC_OK && progp) {
1050 auth_stat = rpc_autherr_badcred;
1051 auth_res = progp->pg_authenticate(rqstp);
1059 rpc_stat = rpc_system_err;
1064 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1065 svc_close_xprt(rqstp->rq_xprt);
1075 if (vers >= progp->pg_nvers ||
1076 !(versp = progp->pg_vers[vers]))
1079 procp = versp->vs_proc + proc;
1080 if (proc >= versp->vs_nproc || !procp->pc_func)
1082 rqstp->rq_procinfo = procp;
1084 /* Syntactic check complete */
1085 serv->sv_stats->rpccnt++;
1087 /* Build the reply header. */
1088 statp = resv->iov_base +resv->iov_len;
1089 svc_putnl(resv, RPC_SUCCESS);
1091 /* Bump per-procedure stats counter */
1094 /* Initialize storage for argp and resp */
1095 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1096 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1098 /* un-reserve some of the out-queue now that we have a
1099 * better idea of reply size
1101 if (procp->pc_xdrressize)
1102 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1104 /* Call the function that processes the request. */
1105 if (!versp->vs_dispatch) {
1106 /* Decode arguments */
1107 xdr = procp->pc_decode;
1108 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1111 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1114 if (rqstp->rq_dropme) {
1115 if (procp->pc_release)
1116 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1119 if (*statp == rpc_success &&
1120 (xdr = procp->pc_encode) &&
1121 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1122 dprintk("svc: failed to encode reply\n");
1123 /* serv->sv_stats->rpcsystemerr++; */
1124 *statp = rpc_system_err;
1127 dprintk("svc: calling dispatcher\n");
1128 if (!versp->vs_dispatch(rqstp, statp)) {
1129 /* Release reply info */
1130 if (procp->pc_release)
1131 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1136 /* Check RPC status result */
1137 if (*statp != rpc_success)
1138 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1140 /* Release reply info */
1141 if (procp->pc_release)
1142 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1144 if (procp->pc_encode == NULL)
1148 if (svc_authorise(rqstp))
1150 return 1; /* Caller can now send it */
1153 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1154 dprintk("svc: svc_process dropit\n");
1158 svc_printk(rqstp, "short len %Zd, dropping request\n",
1161 goto dropit; /* drop request */
1164 serv->sv_stats->rpcbadfmt++;
1165 svc_putnl(resv, 1); /* REJECT */
1166 svc_putnl(resv, 0); /* RPC_MISMATCH */
1167 svc_putnl(resv, 2); /* Only RPCv2 supported */
1172 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1173 serv->sv_stats->rpcbadauth++;
1174 /* Restore write pointer to location of accept status: */
1175 xdr_ressize_check(rqstp, reply_statp);
1176 svc_putnl(resv, 1); /* REJECT */
1177 svc_putnl(resv, 1); /* AUTH_ERROR */
1178 svc_putnl(resv, ntohl(auth_stat)); /* status */
1182 dprintk("svc: unknown program %d\n", prog);
1183 serv->sv_stats->rpcbadfmt++;
1184 svc_putnl(resv, RPC_PROG_UNAVAIL);
1188 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1189 vers, prog, progp->pg_name);
1191 serv->sv_stats->rpcbadfmt++;
1192 svc_putnl(resv, RPC_PROG_MISMATCH);
1193 svc_putnl(resv, progp->pg_lovers);
1194 svc_putnl(resv, progp->pg_hivers);
1198 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1200 serv->sv_stats->rpcbadfmt++;
1201 svc_putnl(resv, RPC_PROC_UNAVAIL);
1205 svc_printk(rqstp, "failed to decode args\n");
1207 rpc_stat = rpc_garbage_args;
1209 serv->sv_stats->rpcbadfmt++;
1210 svc_putnl(resv, ntohl(rpc_stat));
1213 EXPORT_SYMBOL_GPL(svc_process);
1216 * Process the RPC request.
1219 svc_process(struct svc_rqst *rqstp)
1221 struct kvec *argv = &rqstp->rq_arg.head[0];
1222 struct kvec *resv = &rqstp->rq_res.head[0];
1223 struct svc_serv *serv = rqstp->rq_server;
1227 * Setup response xdr_buf.
1228 * Initially it has just one page
1230 rqstp->rq_resused = 1;
1231 resv->iov_base = page_address(rqstp->rq_respages[0]);
1233 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1234 rqstp->rq_res.len = 0;
1235 rqstp->rq_res.page_base = 0;
1236 rqstp->rq_res.page_len = 0;
1237 rqstp->rq_res.buflen = PAGE_SIZE;
1238 rqstp->rq_res.tail[0].iov_base = NULL;
1239 rqstp->rq_res.tail[0].iov_len = 0;
1241 rqstp->rq_xid = svc_getu32(argv);
1243 dir = svc_getnl(argv);
1245 /* direction != CALL */
1246 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1247 serv->sv_stats->rpcbadfmt++;
1252 /* Returns 1 for send, 0 for drop */
1253 if (svc_process_common(rqstp, argv, resv))
1254 return svc_send(rqstp);
1261 #if defined(CONFIG_NFS_V4_1)
1263 * Process a backchannel RPC request that arrived over an existing
1264 * outbound connection
1267 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1268 struct svc_rqst *rqstp)
1270 struct kvec *argv = &rqstp->rq_arg.head[0];
1271 struct kvec *resv = &rqstp->rq_res.head[0];
1273 /* Build the svc_rqst used by the common processing routine */
1274 rqstp->rq_xprt = serv->sv_bc_xprt;
1275 rqstp->rq_xid = req->rq_xid;
1276 rqstp->rq_prot = req->rq_xprt->prot;
1277 rqstp->rq_server = serv;
1279 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1280 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1281 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1282 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1284 /* reset result send buffer "put" position */
1287 if (rqstp->rq_prot != IPPROTO_TCP) {
1288 printk(KERN_ERR "No support for Non-TCP transports!\n");
1293 * Skip the next two words because they've already been
1294 * processed in the trasport
1296 svc_getu32(argv); /* XID */
1297 svc_getnl(argv); /* CALLDIR */
1299 /* Returns 1 for send, 0 for drop */
1300 if (svc_process_common(rqstp, argv, resv)) {
1301 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1302 sizeof(req->rq_snd_buf));
1303 return bc_send(req);
1306 xprt_free_bc_request(req);
1310 EXPORT_SYMBOL(bc_svc_process);
1311 #endif /* CONFIG_NFS_V4_1 */
1314 * Return (transport-specific) limit on the rpc payload.
1316 u32 svc_max_payload(const struct svc_rqst *rqstp)
1318 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1320 if (rqstp->rq_server->sv_max_payload < max)
1321 max = rqstp->rq_server->sv_max_payload;
1324 EXPORT_SYMBOL_GPL(svc_max_payload);