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>
23 #include <linux/sunrpc/types.h>
24 #include <linux/sunrpc/xdr.h>
25 #include <linux/sunrpc/stats.h>
26 #include <linux/sunrpc/svcsock.h>
27 #include <linux/sunrpc/clnt.h>
29 #define RPCDBG_FACILITY RPCDBG_SVCDSP
31 static void svc_unregister(const struct svc_serv *serv);
33 #define svc_serv_is_pooled(serv) ((serv)->sv_function)
36 * Mode for mapping cpus to pools.
39 SVC_POOL_AUTO = -1, /* choose one of the others */
40 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
41 * (legacy & UP mode) */
42 SVC_POOL_PERCPU, /* one pool per cpu */
43 SVC_POOL_PERNODE /* one pool per numa node */
45 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
48 * Structure for mapping cpus to pools and vice versa.
49 * Setup once during sunrpc initialisation.
51 static struct svc_pool_map {
52 int count; /* How many svc_servs use us */
53 int mode; /* Note: int not enum to avoid
54 * warnings about "enumeration value
55 * not handled in switch" */
57 unsigned int *pool_to; /* maps pool id to cpu or node */
58 unsigned int *to_pool; /* maps cpu or node to pool id */
61 .mode = SVC_POOL_DEFAULT
63 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
66 param_set_pool_mode(const char *val, struct kernel_param *kp)
68 int *ip = (int *)kp->arg;
69 struct svc_pool_map *m = &svc_pool_map;
72 mutex_lock(&svc_pool_map_mutex);
79 if (!strncmp(val, "auto", 4))
81 else if (!strncmp(val, "global", 6))
82 *ip = SVC_POOL_GLOBAL;
83 else if (!strncmp(val, "percpu", 6))
84 *ip = SVC_POOL_PERCPU;
85 else if (!strncmp(val, "pernode", 7))
86 *ip = SVC_POOL_PERNODE;
91 mutex_unlock(&svc_pool_map_mutex);
96 param_get_pool_mode(char *buf, struct kernel_param *kp)
98 int *ip = (int *)kp->arg;
103 return strlcpy(buf, "auto", 20);
104 case SVC_POOL_GLOBAL:
105 return strlcpy(buf, "global", 20);
106 case SVC_POOL_PERCPU:
107 return strlcpy(buf, "percpu", 20);
108 case SVC_POOL_PERNODE:
109 return strlcpy(buf, "pernode", 20);
111 return sprintf(buf, "%d", *ip);
115 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
116 &svc_pool_map.mode, 0644);
119 * Detect best pool mapping mode heuristically,
120 * according to the machine's topology.
123 svc_pool_map_choose_mode(void)
127 if (num_online_nodes() > 1) {
129 * Actually have multiple NUMA nodes,
130 * so split pools on NUMA node boundaries
132 return SVC_POOL_PERNODE;
135 node = any_online_node(node_online_map);
136 if (nr_cpus_node(node) > 2) {
138 * Non-trivial SMP, or CONFIG_NUMA on
139 * non-NUMA hardware, e.g. with a generic
140 * x86_64 kernel on Xeons. In this case we
141 * want to divide the pools on cpu boundaries.
143 return SVC_POOL_PERCPU;
146 /* default: one global pool */
147 return SVC_POOL_GLOBAL;
151 * Allocate the to_pool[] and pool_to[] arrays.
152 * Returns 0 on success or an errno.
155 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
157 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
160 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
173 * Initialise the pool map for SVC_POOL_PERCPU mode.
174 * Returns number of pools or <0 on error.
177 svc_pool_map_init_percpu(struct svc_pool_map *m)
179 unsigned int maxpools = nr_cpu_ids;
180 unsigned int pidx = 0;
184 err = svc_pool_map_alloc_arrays(m, maxpools);
188 for_each_online_cpu(cpu) {
189 BUG_ON(pidx > maxpools);
190 m->to_pool[cpu] = pidx;
191 m->pool_to[pidx] = cpu;
194 /* cpus brought online later all get mapped to pool0, sorry */
201 * Initialise the pool map for SVC_POOL_PERNODE mode.
202 * Returns number of pools or <0 on error.
205 svc_pool_map_init_pernode(struct svc_pool_map *m)
207 unsigned int maxpools = nr_node_ids;
208 unsigned int pidx = 0;
212 err = svc_pool_map_alloc_arrays(m, maxpools);
216 for_each_node_with_cpus(node) {
217 /* some architectures (e.g. SN2) have cpuless nodes */
218 BUG_ON(pidx > maxpools);
219 m->to_pool[node] = pidx;
220 m->pool_to[pidx] = node;
223 /* nodes brought online later all get mapped to pool0, sorry */
230 * Add a reference to the global map of cpus to pools (and
231 * vice versa). Initialise the map if we're the first user.
232 * Returns the number of pools.
235 svc_pool_map_get(void)
237 struct svc_pool_map *m = &svc_pool_map;
240 mutex_lock(&svc_pool_map_mutex);
243 mutex_unlock(&svc_pool_map_mutex);
247 if (m->mode == SVC_POOL_AUTO)
248 m->mode = svc_pool_map_choose_mode();
251 case SVC_POOL_PERCPU:
252 npools = svc_pool_map_init_percpu(m);
254 case SVC_POOL_PERNODE:
255 npools = svc_pool_map_init_pernode(m);
260 /* default, or memory allocation failure */
262 m->mode = SVC_POOL_GLOBAL;
266 mutex_unlock(&svc_pool_map_mutex);
272 * Drop a reference to the global map of cpus to pools.
273 * When the last reference is dropped, the map data is
274 * freed; this allows the sysadmin to change the pool
275 * mode using the pool_mode module option without
276 * rebooting or re-loading sunrpc.ko.
279 svc_pool_map_put(void)
281 struct svc_pool_map *m = &svc_pool_map;
283 mutex_lock(&svc_pool_map_mutex);
286 m->mode = SVC_POOL_DEFAULT;
292 mutex_unlock(&svc_pool_map_mutex);
297 * Set the given thread's cpus_allowed mask so that it
298 * will only run on cpus in the given pool.
301 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
303 struct svc_pool_map *m = &svc_pool_map;
304 unsigned int node = m->pool_to[pidx];
307 * The caller checks for sv_nrpools > 1, which
308 * implies that we've been initialized.
310 BUG_ON(m->count == 0);
313 case SVC_POOL_PERCPU:
315 set_cpus_allowed_ptr(task, cpumask_of(node));
318 case SVC_POOL_PERNODE:
320 set_cpus_allowed_ptr(task, cpumask_of_node(node));
327 * Use the mapping mode to choose a pool for a given CPU.
328 * Used when enqueueing an incoming RPC. Always returns
329 * a non-NULL pool pointer.
332 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
334 struct svc_pool_map *m = &svc_pool_map;
335 unsigned int pidx = 0;
338 * An uninitialised map happens in a pure client when
339 * lockd is brought up, so silently treat it the
340 * same as SVC_POOL_GLOBAL.
342 if (svc_serv_is_pooled(serv)) {
344 case SVC_POOL_PERCPU:
345 pidx = m->to_pool[cpu];
347 case SVC_POOL_PERNODE:
348 pidx = m->to_pool[cpu_to_node(cpu)];
352 return &serv->sv_pools[pidx % serv->sv_nrpools];
357 * Create an RPC service
359 static struct svc_serv *
360 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
361 void (*shutdown)(struct svc_serv *serv))
363 struct svc_serv *serv;
365 unsigned int xdrsize;
368 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
370 serv->sv_name = prog->pg_name;
371 serv->sv_program = prog;
372 serv->sv_nrthreads = 1;
373 serv->sv_stats = prog->pg_stats;
374 if (bufsize > RPCSVC_MAXPAYLOAD)
375 bufsize = RPCSVC_MAXPAYLOAD;
376 serv->sv_max_payload = bufsize? bufsize : 4096;
377 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
378 serv->sv_shutdown = shutdown;
381 prog->pg_lovers = prog->pg_nvers-1;
382 for (vers=0; vers<prog->pg_nvers ; vers++)
383 if (prog->pg_vers[vers]) {
384 prog->pg_hivers = vers;
385 if (prog->pg_lovers > vers)
386 prog->pg_lovers = vers;
387 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
388 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
390 prog = prog->pg_next;
392 serv->sv_xdrsize = xdrsize;
393 INIT_LIST_HEAD(&serv->sv_tempsocks);
394 INIT_LIST_HEAD(&serv->sv_permsocks);
395 init_timer(&serv->sv_temptimer);
396 spin_lock_init(&serv->sv_lock);
398 serv->sv_nrpools = npools;
400 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
402 if (!serv->sv_pools) {
407 for (i = 0; i < serv->sv_nrpools; i++) {
408 struct svc_pool *pool = &serv->sv_pools[i];
410 dprintk("svc: initialising pool %u for %s\n",
414 INIT_LIST_HEAD(&pool->sp_threads);
415 INIT_LIST_HEAD(&pool->sp_sockets);
416 INIT_LIST_HEAD(&pool->sp_all_threads);
417 spin_lock_init(&pool->sp_lock);
420 /* Remove any stale portmap registrations */
421 svc_unregister(serv);
427 svc_create(struct svc_program *prog, unsigned int bufsize,
428 void (*shutdown)(struct svc_serv *serv))
430 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
432 EXPORT_SYMBOL_GPL(svc_create);
435 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
436 void (*shutdown)(struct svc_serv *serv),
437 svc_thread_fn func, struct module *mod)
439 struct svc_serv *serv;
440 unsigned int npools = svc_pool_map_get();
442 serv = __svc_create(prog, bufsize, npools, shutdown);
445 serv->sv_function = func;
446 serv->sv_module = mod;
451 EXPORT_SYMBOL_GPL(svc_create_pooled);
454 * Destroy an RPC service. Should be called with appropriate locking to
455 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
458 svc_destroy(struct svc_serv *serv)
460 dprintk("svc: svc_destroy(%s, %d)\n",
461 serv->sv_program->pg_name,
464 if (serv->sv_nrthreads) {
465 if (--(serv->sv_nrthreads) != 0) {
466 svc_sock_update_bufs(serv);
470 printk("svc_destroy: no threads for serv=%p!\n", serv);
472 del_timer_sync(&serv->sv_temptimer);
474 svc_close_all(&serv->sv_tempsocks);
476 if (serv->sv_shutdown)
477 serv->sv_shutdown(serv);
479 svc_close_all(&serv->sv_permsocks);
481 BUG_ON(!list_empty(&serv->sv_permsocks));
482 BUG_ON(!list_empty(&serv->sv_tempsocks));
484 cache_clean_deferred(serv);
486 if (svc_serv_is_pooled(serv))
489 svc_unregister(serv);
490 kfree(serv->sv_pools);
493 EXPORT_SYMBOL_GPL(svc_destroy);
496 * Allocate an RPC server's buffer space.
497 * We allocate pages and place them in rq_argpages.
500 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
502 unsigned int pages, arghi;
504 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
505 * We assume one is at most one page
508 BUG_ON(pages > RPCSVC_MAXPAGES);
510 struct page *p = alloc_page(GFP_KERNEL);
513 rqstp->rq_pages[arghi++] = p;
520 * Release an RPC server buffer
523 svc_release_buffer(struct svc_rqst *rqstp)
527 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
528 if (rqstp->rq_pages[i])
529 put_page(rqstp->rq_pages[i]);
533 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
535 struct svc_rqst *rqstp;
537 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
541 init_waitqueue_head(&rqstp->rq_wait);
543 serv->sv_nrthreads++;
544 spin_lock_bh(&pool->sp_lock);
545 pool->sp_nrthreads++;
546 list_add(&rqstp->rq_all, &pool->sp_all_threads);
547 spin_unlock_bh(&pool->sp_lock);
548 rqstp->rq_server = serv;
549 rqstp->rq_pool = pool;
551 rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
555 rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
559 if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
564 svc_exit_thread(rqstp);
566 return ERR_PTR(-ENOMEM);
568 EXPORT_SYMBOL_GPL(svc_prepare_thread);
571 * Choose a pool in which to create a new thread, for svc_set_num_threads
573 static inline struct svc_pool *
574 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
579 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
583 * Choose a thread to kill, for svc_set_num_threads
585 static inline struct task_struct *
586 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
589 struct task_struct *task = NULL;
592 spin_lock_bh(&pool->sp_lock);
594 /* choose a pool in round-robin fashion */
595 for (i = 0; i < serv->sv_nrpools; i++) {
596 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
597 spin_lock_bh(&pool->sp_lock);
598 if (!list_empty(&pool->sp_all_threads))
600 spin_unlock_bh(&pool->sp_lock);
606 if (!list_empty(&pool->sp_all_threads)) {
607 struct svc_rqst *rqstp;
610 * Remove from the pool->sp_all_threads list
611 * so we don't try to kill it again.
613 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
614 list_del_init(&rqstp->rq_all);
615 task = rqstp->rq_task;
617 spin_unlock_bh(&pool->sp_lock);
623 * Create or destroy enough new threads to make the number
624 * of threads the given number. If `pool' is non-NULL, applies
625 * only to threads in that pool, otherwise round-robins between
626 * all pools. Must be called with a svc_get() reference and
627 * the BKL or another lock to protect access to svc_serv fields.
629 * Destroying threads relies on the service threads filling in
630 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
631 * has been created using svc_create_pooled().
633 * Based on code that used to be in nfsd_svc() but tweaked
637 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
639 struct svc_rqst *rqstp;
640 struct task_struct *task;
641 struct svc_pool *chosen_pool;
643 unsigned int state = serv->sv_nrthreads-1;
646 /* The -1 assumes caller has done a svc_get() */
647 nrservs -= (serv->sv_nrthreads-1);
649 spin_lock_bh(&pool->sp_lock);
650 nrservs -= pool->sp_nrthreads;
651 spin_unlock_bh(&pool->sp_lock);
654 /* create new threads */
655 while (nrservs > 0) {
657 chosen_pool = choose_pool(serv, pool, &state);
659 rqstp = svc_prepare_thread(serv, chosen_pool);
661 error = PTR_ERR(rqstp);
665 __module_get(serv->sv_module);
666 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
668 error = PTR_ERR(task);
669 module_put(serv->sv_module);
670 svc_exit_thread(rqstp);
674 rqstp->rq_task = task;
675 if (serv->sv_nrpools > 1)
676 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
678 svc_sock_update_bufs(serv);
679 wake_up_process(task);
681 /* destroy old threads */
682 while (nrservs < 0 &&
683 (task = choose_victim(serv, pool, &state)) != NULL) {
684 send_sig(SIGINT, task, 1);
690 EXPORT_SYMBOL_GPL(svc_set_num_threads);
693 * Called from a server thread as it's exiting. Caller must hold the BKL or
694 * the "service mutex", whichever is appropriate for the service.
697 svc_exit_thread(struct svc_rqst *rqstp)
699 struct svc_serv *serv = rqstp->rq_server;
700 struct svc_pool *pool = rqstp->rq_pool;
702 svc_release_buffer(rqstp);
703 kfree(rqstp->rq_resp);
704 kfree(rqstp->rq_argp);
705 kfree(rqstp->rq_auth_data);
707 spin_lock_bh(&pool->sp_lock);
708 pool->sp_nrthreads--;
709 list_del(&rqstp->rq_all);
710 spin_unlock_bh(&pool->sp_lock);
714 /* Release the server */
718 EXPORT_SYMBOL_GPL(svc_exit_thread);
721 * Register an "inet" protocol family netid with the local
722 * rpcbind daemon via an rpcbind v4 SET request.
724 * No netconfig infrastructure is available in the kernel, so
725 * we map IP_ protocol numbers to netids by hand.
727 * Returns zero on success; a negative errno value is returned
728 * if any error occurs.
730 static int __svc_rpcb_register4(const u32 program, const u32 version,
731 const unsigned short protocol,
732 const unsigned short port)
734 const struct sockaddr_in sin = {
735 .sin_family = AF_INET,
736 .sin_addr.s_addr = htonl(INADDR_ANY),
737 .sin_port = htons(port),
744 netid = RPCBIND_NETID_UDP;
747 netid = RPCBIND_NETID_TCP;
753 error = rpcb_v4_register(program, version,
754 (const struct sockaddr *)&sin, netid);
757 * User space didn't support rpcbind v4, so retry this
758 * registration request with the legacy rpcbind v2 protocol.
760 if (error == -EPROTONOSUPPORT)
761 error = rpcb_register(program, version, protocol, port);
766 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
768 * Register an "inet6" protocol family netid with the local
769 * rpcbind daemon via an rpcbind v4 SET request.
771 * No netconfig infrastructure is available in the kernel, so
772 * we map IP_ protocol numbers to netids by hand.
774 * Returns zero on success; a negative errno value is returned
775 * if any error occurs.
777 static int __svc_rpcb_register6(const u32 program, const u32 version,
778 const unsigned short protocol,
779 const unsigned short port)
781 const struct sockaddr_in6 sin6 = {
782 .sin6_family = AF_INET6,
783 .sin6_addr = IN6ADDR_ANY_INIT,
784 .sin6_port = htons(port),
791 netid = RPCBIND_NETID_UDP6;
794 netid = RPCBIND_NETID_TCP6;
800 error = rpcb_v4_register(program, version,
801 (const struct sockaddr *)&sin6, netid);
804 * User space didn't support rpcbind version 4, so we won't
805 * use a PF_INET6 listener.
807 if (error == -EPROTONOSUPPORT)
808 error = -EAFNOSUPPORT;
812 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
815 * Register a kernel RPC service via rpcbind version 4.
817 * Returns zero on success; a negative errno value is returned
818 * if any error occurs.
820 static int __svc_register(const char *progname,
821 const u32 program, const u32 version,
823 const unsigned short protocol,
824 const unsigned short port)
826 int error = -EAFNOSUPPORT;
830 error = __svc_rpcb_register4(program, version,
833 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
835 error = __svc_rpcb_register6(program, version,
837 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
841 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
842 "service (errno %d).\n", progname, version, -error);
847 * svc_register - register an RPC service with the local portmapper
848 * @serv: svc_serv struct for the service to register
849 * @family: protocol family of service's listener socket
850 * @proto: transport protocol number to advertise
851 * @port: port to advertise
853 * Service is registered for any address in the passed-in protocol family
855 int svc_register(const struct svc_serv *serv, const int family,
856 const unsigned short proto, const unsigned short port)
858 struct svc_program *progp;
862 BUG_ON(proto == 0 && port == 0);
864 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
865 for (i = 0; i < progp->pg_nvers; i++) {
866 if (progp->pg_vers[i] == NULL)
869 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
872 proto == IPPROTO_UDP? "udp" : "tcp",
875 progp->pg_vers[i]->vs_hidden?
876 " (but not telling portmap)" : "");
878 if (progp->pg_vers[i]->vs_hidden)
881 error = __svc_register(progp->pg_name, progp->pg_prog,
882 i, family, proto, port);
892 * If user space is running rpcbind, it should take the v4 UNSET
893 * and clear everything for this [program, version]. If user space
894 * is running portmap, it will reject the v4 UNSET, but won't have
895 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
896 * in this case to clear all existing entries for [program, version].
898 static void __svc_unregister(const u32 program, const u32 version,
899 const char *progname)
903 error = rpcb_v4_register(program, version, NULL, "");
906 * User space didn't support rpcbind v4, so retry this
907 * request with the legacy rpcbind v2 protocol.
909 if (error == -EPROTONOSUPPORT)
910 error = rpcb_register(program, version, 0, 0);
912 dprintk("svc: %s(%sv%u), error %d\n",
913 __func__, progname, version, error);
917 * All netids, bind addresses and ports registered for [program, version]
918 * are removed from the local rpcbind database (if the service is not
919 * hidden) to make way for a new instance of the service.
921 * The result of unregistration is reported via dprintk for those who want
922 * verification of the result, but is otherwise not important.
924 static void svc_unregister(const struct svc_serv *serv)
926 struct svc_program *progp;
930 clear_thread_flag(TIF_SIGPENDING);
932 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
933 for (i = 0; i < progp->pg_nvers; i++) {
934 if (progp->pg_vers[i] == NULL)
936 if (progp->pg_vers[i]->vs_hidden)
939 __svc_unregister(progp->pg_prog, i, progp->pg_name);
943 spin_lock_irqsave(¤t->sighand->siglock, flags);
945 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
949 * Printk the given error with the address of the client that caused it.
952 __attribute__ ((format (printf, 2, 3)))
953 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
957 char buf[RPC_MAX_ADDRBUFLEN];
959 if (!net_ratelimit())
962 printk(KERN_WARNING "svc: %s: ",
963 svc_print_addr(rqstp, buf, sizeof(buf)));
966 r = vprintk(fmt, args);
973 * Process the RPC request.
976 svc_process(struct svc_rqst *rqstp)
978 struct svc_program *progp;
979 struct svc_version *versp = NULL; /* compiler food */
980 struct svc_procedure *procp = NULL;
981 struct kvec * argv = &rqstp->rq_arg.head[0];
982 struct kvec * resv = &rqstp->rq_res.head[0];
983 struct svc_serv *serv = rqstp->rq_server;
986 u32 dir, prog, vers, proc;
987 __be32 auth_stat, rpc_stat;
991 rpc_stat = rpc_success;
993 if (argv->iov_len < 6*4)
996 /* setup response xdr_buf.
997 * Initially it has just one page
999 rqstp->rq_resused = 1;
1000 resv->iov_base = page_address(rqstp->rq_respages[0]);
1002 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1003 rqstp->rq_res.len = 0;
1004 rqstp->rq_res.page_base = 0;
1005 rqstp->rq_res.page_len = 0;
1006 rqstp->rq_res.buflen = PAGE_SIZE;
1007 rqstp->rq_res.tail[0].iov_base = NULL;
1008 rqstp->rq_res.tail[0].iov_len = 0;
1009 /* Will be turned off only in gss privacy case: */
1010 rqstp->rq_splice_ok = 1;
1012 /* Setup reply header */
1013 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1015 rqstp->rq_xid = svc_getu32(argv);
1016 svc_putu32(resv, rqstp->rq_xid);
1018 dir = svc_getnl(argv);
1019 vers = svc_getnl(argv);
1021 /* First words of reply: */
1022 svc_putnl(resv, 1); /* REPLY */
1024 if (dir != 0) /* direction != CALL */
1026 if (vers != 2) /* RPC version number */
1029 /* Save position in case we later decide to reject: */
1030 reply_statp = resv->iov_base + resv->iov_len;
1032 svc_putnl(resv, 0); /* ACCEPT */
1034 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1035 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1036 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1038 progp = serv->sv_program;
1040 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1041 if (prog == progp->pg_prog)
1045 * Decode auth data, and add verifier to reply buffer.
1046 * We do this before anything else in order to get a decent
1049 auth_res = svc_authenticate(rqstp, &auth_stat);
1050 /* Also give the program a chance to reject this call: */
1051 if (auth_res == SVC_OK && progp) {
1052 auth_stat = rpc_autherr_badcred;
1053 auth_res = progp->pg_authenticate(rqstp);
1061 rpc_stat = rpc_system_err;
1074 if (vers >= progp->pg_nvers ||
1075 !(versp = progp->pg_vers[vers]))
1078 procp = versp->vs_proc + proc;
1079 if (proc >= versp->vs_nproc || !procp->pc_func)
1081 rqstp->rq_server = serv;
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 (*statp == rpc_drop_reply) {
1115 if (procp->pc_release)
1116 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1119 if (*statp == rpc_success && (xdr = procp->pc_encode)
1120 && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1121 dprintk("svc: failed to encode reply\n");
1122 /* serv->sv_stats->rpcsystemerr++; */
1123 *statp = rpc_system_err;
1126 dprintk("svc: calling dispatcher\n");
1127 if (!versp->vs_dispatch(rqstp, statp)) {
1128 /* Release reply info */
1129 if (procp->pc_release)
1130 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1135 /* Check RPC status result */
1136 if (*statp != rpc_success)
1137 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1139 /* Release reply info */
1140 if (procp->pc_release)
1141 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1143 if (procp->pc_encode == NULL)
1147 if (svc_authorise(rqstp))
1149 return svc_send(rqstp);
1152 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1153 dprintk("svc: svc_process dropit\n");
1158 svc_printk(rqstp, "short len %Zd, dropping request\n",
1161 goto dropit; /* drop request */
1164 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1166 serv->sv_stats->rpcbadfmt++;
1167 goto dropit; /* drop request */
1170 serv->sv_stats->rpcbadfmt++;
1171 svc_putnl(resv, 1); /* REJECT */
1172 svc_putnl(resv, 0); /* RPC_MISMATCH */
1173 svc_putnl(resv, 2); /* Only RPCv2 supported */
1178 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1179 serv->sv_stats->rpcbadauth++;
1180 /* Restore write pointer to location of accept status: */
1181 xdr_ressize_check(rqstp, reply_statp);
1182 svc_putnl(resv, 1); /* REJECT */
1183 svc_putnl(resv, 1); /* AUTH_ERROR */
1184 svc_putnl(resv, ntohl(auth_stat)); /* status */
1188 dprintk("svc: unknown program %d\n", prog);
1189 serv->sv_stats->rpcbadfmt++;
1190 svc_putnl(resv, RPC_PROG_UNAVAIL);
1194 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1195 vers, prog, progp->pg_name);
1197 serv->sv_stats->rpcbadfmt++;
1198 svc_putnl(resv, RPC_PROG_MISMATCH);
1199 svc_putnl(resv, progp->pg_lovers);
1200 svc_putnl(resv, progp->pg_hivers);
1204 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1206 serv->sv_stats->rpcbadfmt++;
1207 svc_putnl(resv, RPC_PROC_UNAVAIL);
1211 svc_printk(rqstp, "failed to decode args\n");
1213 rpc_stat = rpc_garbage_args;
1215 serv->sv_stats->rpcbadfmt++;
1216 svc_putnl(resv, ntohl(rpc_stat));
1219 EXPORT_SYMBOL_GPL(svc_process);
1222 * Return (transport-specific) limit on the rpc payload.
1224 u32 svc_max_payload(const struct svc_rqst *rqstp)
1226 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1228 if (rqstp->rq_server->sv_max_payload < max)
1229 max = rqstp->rq_server->sv_max_payload;
1232 EXPORT_SYMBOL_GPL(svc_max_payload);