2 * linux/net/sunrpc/svc.c
4 * High-level RPC service routines
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
24 #include <linux/sunrpc/types.h>
25 #include <linux/sunrpc/xdr.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svcsock.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/bc_xprt.h>
31 #include <trace/events/sunrpc.h>
33 #define RPCDBG_FACILITY RPCDBG_SVCDSP
35 static void svc_unregister(const struct svc_serv *serv, struct net *net);
37 #define svc_serv_is_pooled(serv) ((serv)->sv_ops->svo_function)
39 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
42 * Structure for mapping cpus to pools and vice versa.
43 * Setup once during sunrpc initialisation.
45 struct svc_pool_map svc_pool_map = {
46 .mode = SVC_POOL_DEFAULT
48 EXPORT_SYMBOL_GPL(svc_pool_map);
50 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
53 param_set_pool_mode(const char *val, struct kernel_param *kp)
55 int *ip = (int *)kp->arg;
56 struct svc_pool_map *m = &svc_pool_map;
59 mutex_lock(&svc_pool_map_mutex);
66 if (!strncmp(val, "auto", 4))
68 else if (!strncmp(val, "global", 6))
69 *ip = SVC_POOL_GLOBAL;
70 else if (!strncmp(val, "percpu", 6))
71 *ip = SVC_POOL_PERCPU;
72 else if (!strncmp(val, "pernode", 7))
73 *ip = SVC_POOL_PERNODE;
78 mutex_unlock(&svc_pool_map_mutex);
83 param_get_pool_mode(char *buf, struct kernel_param *kp)
85 int *ip = (int *)kp->arg;
90 return strlcpy(buf, "auto", 20);
92 return strlcpy(buf, "global", 20);
94 return strlcpy(buf, "percpu", 20);
95 case SVC_POOL_PERNODE:
96 return strlcpy(buf, "pernode", 20);
98 return sprintf(buf, "%d", *ip);
102 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
103 &svc_pool_map.mode, 0644);
106 * Detect best pool mapping mode heuristically,
107 * according to the machine's topology.
110 svc_pool_map_choose_mode(void)
114 if (nr_online_nodes > 1) {
116 * Actually have multiple NUMA nodes,
117 * so split pools on NUMA node boundaries
119 return SVC_POOL_PERNODE;
122 node = first_online_node;
123 if (nr_cpus_node(node) > 2) {
125 * Non-trivial SMP, or CONFIG_NUMA on
126 * non-NUMA hardware, e.g. with a generic
127 * x86_64 kernel on Xeons. In this case we
128 * want to divide the pools on cpu boundaries.
130 return SVC_POOL_PERCPU;
133 /* default: one global pool */
134 return SVC_POOL_GLOBAL;
138 * Allocate the to_pool[] and pool_to[] arrays.
139 * Returns 0 on success or an errno.
142 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
144 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
147 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
161 * Initialise the pool map for SVC_POOL_PERCPU mode.
162 * Returns number of pools or <0 on error.
165 svc_pool_map_init_percpu(struct svc_pool_map *m)
167 unsigned int maxpools = nr_cpu_ids;
168 unsigned int pidx = 0;
172 err = svc_pool_map_alloc_arrays(m, maxpools);
176 for_each_online_cpu(cpu) {
177 BUG_ON(pidx >= maxpools);
178 m->to_pool[cpu] = pidx;
179 m->pool_to[pidx] = cpu;
182 /* cpus brought online later all get mapped to pool0, sorry */
189 * Initialise the pool map for SVC_POOL_PERNODE mode.
190 * Returns number of pools or <0 on error.
193 svc_pool_map_init_pernode(struct svc_pool_map *m)
195 unsigned int maxpools = nr_node_ids;
196 unsigned int pidx = 0;
200 err = svc_pool_map_alloc_arrays(m, maxpools);
204 for_each_node_with_cpus(node) {
205 /* some architectures (e.g. SN2) have cpuless nodes */
206 BUG_ON(pidx > maxpools);
207 m->to_pool[node] = pidx;
208 m->pool_to[pidx] = node;
211 /* nodes brought online later all get mapped to pool0, sorry */
218 * Add a reference to the global map of cpus to pools (and
219 * vice versa). Initialise the map if we're the first user.
220 * Returns the number of pools.
223 svc_pool_map_get(void)
225 struct svc_pool_map *m = &svc_pool_map;
228 mutex_lock(&svc_pool_map_mutex);
231 mutex_unlock(&svc_pool_map_mutex);
235 if (m->mode == SVC_POOL_AUTO)
236 m->mode = svc_pool_map_choose_mode();
239 case SVC_POOL_PERCPU:
240 npools = svc_pool_map_init_percpu(m);
242 case SVC_POOL_PERNODE:
243 npools = svc_pool_map_init_pernode(m);
248 /* default, or memory allocation failure */
250 m->mode = SVC_POOL_GLOBAL;
254 mutex_unlock(&svc_pool_map_mutex);
257 EXPORT_SYMBOL_GPL(svc_pool_map_get);
260 * Drop a reference to the global map of cpus to pools.
261 * When the last reference is dropped, the map data is
262 * freed; this allows the sysadmin to change the pool
263 * mode using the pool_mode module option without
264 * rebooting or re-loading sunrpc.ko.
267 svc_pool_map_put(void)
269 struct svc_pool_map *m = &svc_pool_map;
271 mutex_lock(&svc_pool_map_mutex);
281 mutex_unlock(&svc_pool_map_mutex);
283 EXPORT_SYMBOL_GPL(svc_pool_map_put);
285 static int svc_pool_map_get_node(unsigned int pidx)
287 const struct svc_pool_map *m = &svc_pool_map;
290 if (m->mode == SVC_POOL_PERCPU)
291 return cpu_to_node(m->pool_to[pidx]);
292 if (m->mode == SVC_POOL_PERNODE)
293 return m->pool_to[pidx];
298 * Set the given thread's cpus_allowed mask so that it
299 * will only run on cpus in the given pool.
302 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
304 struct svc_pool_map *m = &svc_pool_map;
305 unsigned int node = m->pool_to[pidx];
308 * The caller checks for sv_nrpools > 1, which
309 * implies that we've been initialized.
311 WARN_ON_ONCE(m->count == 0);
316 case SVC_POOL_PERCPU:
318 set_cpus_allowed_ptr(task, cpumask_of(node));
321 case SVC_POOL_PERNODE:
323 set_cpus_allowed_ptr(task, cpumask_of_node(node));
330 * Use the mapping mode to choose a pool for a given CPU.
331 * Used when enqueueing an incoming RPC. Always returns
332 * a non-NULL pool pointer.
335 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
337 struct svc_pool_map *m = &svc_pool_map;
338 unsigned int pidx = 0;
341 * An uninitialised map happens in a pure client when
342 * lockd is brought up, so silently treat it the
343 * same as SVC_POOL_GLOBAL.
345 if (svc_serv_is_pooled(serv)) {
347 case SVC_POOL_PERCPU:
348 pidx = m->to_pool[cpu];
350 case SVC_POOL_PERNODE:
351 pidx = m->to_pool[cpu_to_node(cpu)];
355 return &serv->sv_pools[pidx % serv->sv_nrpools];
358 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
362 err = rpcb_create_local(net);
366 /* Remove any stale portmap registrations */
367 svc_unregister(serv, net);
370 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
372 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
374 svc_unregister(serv, net);
377 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
379 static int svc_uses_rpcbind(struct svc_serv *serv)
381 struct svc_program *progp;
384 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
385 for (i = 0; i < progp->pg_nvers; i++) {
386 if (progp->pg_vers[i] == NULL)
388 if (progp->pg_vers[i]->vs_hidden == 0)
396 int svc_bind(struct svc_serv *serv, struct net *net)
398 if (!svc_uses_rpcbind(serv))
400 return svc_rpcb_setup(serv, net);
402 EXPORT_SYMBOL_GPL(svc_bind);
405 * Create an RPC service
407 static struct svc_serv *
408 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
409 struct svc_serv_ops *ops)
411 struct svc_serv *serv;
413 unsigned int xdrsize;
416 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
418 serv->sv_name = prog->pg_name;
419 serv->sv_program = prog;
420 serv->sv_nrthreads = 1;
421 serv->sv_stats = prog->pg_stats;
422 if (bufsize > RPCSVC_MAXPAYLOAD)
423 bufsize = RPCSVC_MAXPAYLOAD;
424 serv->sv_max_payload = bufsize? bufsize : 4096;
425 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
429 prog->pg_lovers = prog->pg_nvers-1;
430 for (vers=0; vers<prog->pg_nvers ; vers++)
431 if (prog->pg_vers[vers]) {
432 prog->pg_hivers = vers;
433 if (prog->pg_lovers > vers)
434 prog->pg_lovers = vers;
435 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
436 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
438 prog = prog->pg_next;
440 serv->sv_xdrsize = xdrsize;
441 INIT_LIST_HEAD(&serv->sv_tempsocks);
442 INIT_LIST_HEAD(&serv->sv_permsocks);
443 init_timer(&serv->sv_temptimer);
444 spin_lock_init(&serv->sv_lock);
446 serv->sv_nrpools = npools;
448 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
450 if (!serv->sv_pools) {
455 for (i = 0; i < serv->sv_nrpools; i++) {
456 struct svc_pool *pool = &serv->sv_pools[i];
458 dprintk("svc: initialising pool %u for %s\n",
462 INIT_LIST_HEAD(&pool->sp_sockets);
463 INIT_LIST_HEAD(&pool->sp_all_threads);
464 spin_lock_init(&pool->sp_lock);
471 svc_create(struct svc_program *prog, unsigned int bufsize,
472 struct svc_serv_ops *ops)
474 return __svc_create(prog, bufsize, /*npools*/1, ops);
476 EXPORT_SYMBOL_GPL(svc_create);
479 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
480 struct svc_serv_ops *ops)
482 struct svc_serv *serv;
483 unsigned int npools = svc_pool_map_get();
485 serv = __svc_create(prog, bufsize, npools, ops);
493 EXPORT_SYMBOL_GPL(svc_create_pooled);
495 void svc_shutdown_net(struct svc_serv *serv, struct net *net)
497 svc_close_net(serv, net);
499 if (serv->sv_ops->svo_shutdown)
500 serv->sv_ops->svo_shutdown(serv, net);
502 EXPORT_SYMBOL_GPL(svc_shutdown_net);
505 * Destroy an RPC service. Should be called with appropriate locking to
506 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
509 svc_destroy(struct svc_serv *serv)
511 dprintk("svc: svc_destroy(%s, %d)\n",
512 serv->sv_program->pg_name,
515 if (serv->sv_nrthreads) {
516 if (--(serv->sv_nrthreads) != 0) {
517 svc_sock_update_bufs(serv);
521 printk("svc_destroy: no threads for serv=%p!\n", serv);
523 del_timer_sync(&serv->sv_temptimer);
526 * The last user is gone and thus all sockets have to be destroyed to
527 * the point. Check this.
529 BUG_ON(!list_empty(&serv->sv_permsocks));
530 BUG_ON(!list_empty(&serv->sv_tempsocks));
532 cache_clean_deferred(serv);
534 if (svc_serv_is_pooled(serv))
537 kfree(serv->sv_pools);
540 EXPORT_SYMBOL_GPL(svc_destroy);
543 * Allocate an RPC server's buffer space.
544 * We allocate pages and place them in rq_argpages.
547 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
549 unsigned int pages, arghi;
551 /* bc_xprt uses fore channel allocated buffers */
552 if (svc_is_backchannel(rqstp))
555 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
556 * We assume one is at most one page
559 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
560 if (pages > RPCSVC_MAXPAGES)
561 pages = RPCSVC_MAXPAGES;
563 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
566 rqstp->rq_pages[arghi++] = p;
573 * Release an RPC server buffer
576 svc_release_buffer(struct svc_rqst *rqstp)
580 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
581 if (rqstp->rq_pages[i])
582 put_page(rqstp->rq_pages[i]);
586 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
588 struct svc_rqst *rqstp;
590 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
594 __set_bit(RQ_BUSY, &rqstp->rq_flags);
595 spin_lock_init(&rqstp->rq_lock);
596 rqstp->rq_server = serv;
597 rqstp->rq_pool = pool;
599 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
603 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
607 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
612 svc_rqst_free(rqstp);
615 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
618 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
620 struct svc_rqst *rqstp;
622 rqstp = svc_rqst_alloc(serv, pool, node);
624 return ERR_PTR(-ENOMEM);
626 serv->sv_nrthreads++;
627 spin_lock_bh(&pool->sp_lock);
628 pool->sp_nrthreads++;
629 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
630 spin_unlock_bh(&pool->sp_lock);
633 EXPORT_SYMBOL_GPL(svc_prepare_thread);
636 * Choose a pool in which to create a new thread, for svc_set_num_threads
638 static inline struct svc_pool *
639 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
644 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
648 * Choose a thread to kill, for svc_set_num_threads
650 static inline struct task_struct *
651 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
654 struct task_struct *task = NULL;
657 spin_lock_bh(&pool->sp_lock);
659 /* choose a pool in round-robin fashion */
660 for (i = 0; i < serv->sv_nrpools; i++) {
661 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
662 spin_lock_bh(&pool->sp_lock);
663 if (!list_empty(&pool->sp_all_threads))
665 spin_unlock_bh(&pool->sp_lock);
671 if (!list_empty(&pool->sp_all_threads)) {
672 struct svc_rqst *rqstp;
675 * Remove from the pool->sp_all_threads list
676 * so we don't try to kill it again.
678 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
679 set_bit(RQ_VICTIM, &rqstp->rq_flags);
680 list_del_rcu(&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_ops->svo_module);
734 task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
735 node, "%s", serv->sv_name);
737 error = PTR_ERR(task);
738 module_put(serv->sv_ops->svo_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 "service
763 * mutex" for the service.
766 svc_rqst_free(struct svc_rqst *rqstp)
768 svc_release_buffer(rqstp);
769 kfree(rqstp->rq_resp);
770 kfree(rqstp->rq_argp);
771 kfree(rqstp->rq_auth_data);
772 kfree_rcu(rqstp, rq_rcu_head);
774 EXPORT_SYMBOL_GPL(svc_rqst_free);
777 svc_exit_thread(struct svc_rqst *rqstp)
779 struct svc_serv *serv = rqstp->rq_server;
780 struct svc_pool *pool = rqstp->rq_pool;
782 spin_lock_bh(&pool->sp_lock);
783 pool->sp_nrthreads--;
784 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
785 list_del_rcu(&rqstp->rq_all);
786 spin_unlock_bh(&pool->sp_lock);
788 svc_rqst_free(rqstp);
790 /* Release the server */
794 EXPORT_SYMBOL_GPL(svc_exit_thread);
797 * Register an "inet" protocol family netid with the local
798 * rpcbind daemon via an rpcbind v4 SET request.
800 * No netconfig infrastructure is available in the kernel, so
801 * we map IP_ protocol numbers to netids by hand.
803 * Returns zero on success; a negative errno value is returned
804 * if any error occurs.
806 static int __svc_rpcb_register4(struct net *net, const u32 program,
808 const unsigned short protocol,
809 const unsigned short port)
811 const struct sockaddr_in sin = {
812 .sin_family = AF_INET,
813 .sin_addr.s_addr = htonl(INADDR_ANY),
814 .sin_port = htons(port),
821 netid = RPCBIND_NETID_UDP;
824 netid = RPCBIND_NETID_TCP;
830 error = rpcb_v4_register(net, program, version,
831 (const struct sockaddr *)&sin, netid);
834 * User space didn't support rpcbind v4, so retry this
835 * registration request with the legacy rpcbind v2 protocol.
837 if (error == -EPROTONOSUPPORT)
838 error = rpcb_register(net, program, version, protocol, port);
843 #if IS_ENABLED(CONFIG_IPV6)
845 * Register an "inet6" protocol family netid with the local
846 * rpcbind daemon via an rpcbind v4 SET request.
848 * No netconfig infrastructure is available in the kernel, so
849 * we map IP_ protocol numbers to netids by hand.
851 * Returns zero on success; a negative errno value is returned
852 * if any error occurs.
854 static int __svc_rpcb_register6(struct net *net, const u32 program,
856 const unsigned short protocol,
857 const unsigned short port)
859 const struct sockaddr_in6 sin6 = {
860 .sin6_family = AF_INET6,
861 .sin6_addr = IN6ADDR_ANY_INIT,
862 .sin6_port = htons(port),
869 netid = RPCBIND_NETID_UDP6;
872 netid = RPCBIND_NETID_TCP6;
878 error = rpcb_v4_register(net, program, version,
879 (const struct sockaddr *)&sin6, netid);
882 * User space didn't support rpcbind version 4, so we won't
883 * use a PF_INET6 listener.
885 if (error == -EPROTONOSUPPORT)
886 error = -EAFNOSUPPORT;
890 #endif /* IS_ENABLED(CONFIG_IPV6) */
893 * Register a kernel RPC service via rpcbind version 4.
895 * Returns zero on success; a negative errno value is returned
896 * if any error occurs.
898 static int __svc_register(struct net *net, const char *progname,
899 const u32 program, const u32 version,
901 const unsigned short protocol,
902 const unsigned short port)
904 int error = -EAFNOSUPPORT;
908 error = __svc_rpcb_register4(net, program, version,
911 #if IS_ENABLED(CONFIG_IPV6)
913 error = __svc_rpcb_register6(net, program, version,
922 * svc_register - register an RPC service with the local portmapper
923 * @serv: svc_serv struct for the service to register
924 * @net: net namespace for the service to register
925 * @family: protocol family of service's listener socket
926 * @proto: transport protocol number to advertise
927 * @port: port to advertise
929 * Service is registered for any address in the passed-in protocol family
931 int svc_register(const struct svc_serv *serv, struct net *net,
932 const int family, const unsigned short proto,
933 const unsigned short port)
935 struct svc_program *progp;
936 struct svc_version *vers;
940 WARN_ON_ONCE(proto == 0 && port == 0);
941 if (proto == 0 && port == 0)
944 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
945 for (i = 0; i < progp->pg_nvers; i++) {
946 vers = progp->pg_vers[i];
950 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
953 proto == IPPROTO_UDP? "udp" : "tcp",
957 " (but not telling portmap)" : "");
962 error = __svc_register(net, progp->pg_name, progp->pg_prog,
963 i, family, proto, port);
965 if (vers->vs_rpcb_optnl) {
971 printk(KERN_WARNING "svc: failed to register "
972 "%sv%u RPC service (errno %d).\n",
973 progp->pg_name, i, -error);
983 * If user space is running rpcbind, it should take the v4 UNSET
984 * and clear everything for this [program, version]. If user space
985 * is running portmap, it will reject the v4 UNSET, but won't have
986 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
987 * in this case to clear all existing entries for [program, version].
989 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
990 const char *progname)
994 error = rpcb_v4_register(net, program, version, NULL, "");
997 * User space didn't support rpcbind v4, so retry this
998 * request with the legacy rpcbind v2 protocol.
1000 if (error == -EPROTONOSUPPORT)
1001 error = rpcb_register(net, program, version, 0, 0);
1003 dprintk("svc: %s(%sv%u), error %d\n",
1004 __func__, progname, version, error);
1008 * All netids, bind addresses and ports registered for [program, version]
1009 * are removed from the local rpcbind database (if the service is not
1010 * hidden) to make way for a new instance of the service.
1012 * The result of unregistration is reported via dprintk for those who want
1013 * verification of the result, but is otherwise not important.
1015 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1017 struct svc_program *progp;
1018 unsigned long flags;
1021 clear_thread_flag(TIF_SIGPENDING);
1023 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1024 for (i = 0; i < progp->pg_nvers; i++) {
1025 if (progp->pg_vers[i] == NULL)
1027 if (progp->pg_vers[i]->vs_hidden)
1030 dprintk("svc: attempting to unregister %sv%u\n",
1032 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1036 spin_lock_irqsave(¤t->sighand->siglock, flags);
1037 recalc_sigpending();
1038 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1042 * dprintk the given error with the address of the client that caused it.
1044 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1045 static __printf(2, 3)
1046 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1048 struct va_format vaf;
1050 char buf[RPC_MAX_ADDRBUFLEN];
1052 va_start(args, fmt);
1057 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1062 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1066 * Common routine for processing the RPC request.
1069 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1071 struct svc_program *progp;
1072 struct svc_version *versp = NULL; /* compiler food */
1073 struct svc_procedure *procp = NULL;
1074 struct svc_serv *serv = rqstp->rq_server;
1077 u32 prog, vers, proc;
1078 __be32 auth_stat, rpc_stat;
1080 __be32 *reply_statp;
1082 rpc_stat = rpc_success;
1084 if (argv->iov_len < 6*4)
1087 /* Will be turned off only in gss privacy case: */
1088 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1089 /* Will be turned off only when NFSv4 Sessions are used */
1090 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1091 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1093 /* Setup reply header */
1094 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1096 svc_putu32(resv, rqstp->rq_xid);
1098 vers = svc_getnl(argv);
1100 /* First words of reply: */
1101 svc_putnl(resv, 1); /* REPLY */
1103 if (vers != 2) /* RPC version number */
1106 /* Save position in case we later decide to reject: */
1107 reply_statp = resv->iov_base + resv->iov_len;
1109 svc_putnl(resv, 0); /* ACCEPT */
1111 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1112 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1113 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1115 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1116 if (prog == progp->pg_prog)
1120 * Decode auth data, and add verifier to reply buffer.
1121 * We do this before anything else in order to get a decent
1124 auth_res = svc_authenticate(rqstp, &auth_stat);
1125 /* Also give the program a chance to reject this call: */
1126 if (auth_res == SVC_OK && progp) {
1127 auth_stat = rpc_autherr_badcred;
1128 auth_res = progp->pg_authenticate(rqstp);
1136 rpc_stat = rpc_system_err;
1141 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1142 svc_close_xprt(rqstp->rq_xprt);
1152 if (vers >= progp->pg_nvers ||
1153 !(versp = progp->pg_vers[vers]))
1156 procp = versp->vs_proc + proc;
1157 if (proc >= versp->vs_nproc || !procp->pc_func)
1159 rqstp->rq_procinfo = procp;
1161 /* Syntactic check complete */
1162 serv->sv_stats->rpccnt++;
1164 /* Build the reply header. */
1165 statp = resv->iov_base +resv->iov_len;
1166 svc_putnl(resv, RPC_SUCCESS);
1168 /* Bump per-procedure stats counter */
1171 /* Initialize storage for argp and resp */
1172 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1173 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1175 /* un-reserve some of the out-queue now that we have a
1176 * better idea of reply size
1178 if (procp->pc_xdrressize)
1179 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1181 /* Call the function that processes the request. */
1182 if (!versp->vs_dispatch) {
1183 /* Decode arguments */
1184 xdr = procp->pc_decode;
1185 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1188 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1191 if (test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1192 if (procp->pc_release)
1193 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1196 if (*statp == rpc_success &&
1197 (xdr = procp->pc_encode) &&
1198 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1199 dprintk("svc: failed to encode reply\n");
1200 /* serv->sv_stats->rpcsystemerr++; */
1201 *statp = rpc_system_err;
1204 dprintk("svc: calling dispatcher\n");
1205 if (!versp->vs_dispatch(rqstp, statp)) {
1206 /* Release reply info */
1207 if (procp->pc_release)
1208 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1213 /* Check RPC status result */
1214 if (*statp != rpc_success)
1215 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1217 /* Release reply info */
1218 if (procp->pc_release)
1219 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1221 if (procp->pc_encode == NULL)
1225 if (svc_authorise(rqstp))
1227 return 1; /* Caller can now send it */
1230 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1231 dprintk("svc: svc_process dropit\n");
1235 svc_printk(rqstp, "short len %Zd, dropping request\n",
1238 goto dropit; /* drop request */
1241 serv->sv_stats->rpcbadfmt++;
1242 svc_putnl(resv, 1); /* REJECT */
1243 svc_putnl(resv, 0); /* RPC_MISMATCH */
1244 svc_putnl(resv, 2); /* Only RPCv2 supported */
1249 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1250 serv->sv_stats->rpcbadauth++;
1251 /* Restore write pointer to location of accept status: */
1252 xdr_ressize_check(rqstp, reply_statp);
1253 svc_putnl(resv, 1); /* REJECT */
1254 svc_putnl(resv, 1); /* AUTH_ERROR */
1255 svc_putnl(resv, ntohl(auth_stat)); /* status */
1259 dprintk("svc: unknown program %d\n", prog);
1260 serv->sv_stats->rpcbadfmt++;
1261 svc_putnl(resv, RPC_PROG_UNAVAIL);
1265 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1266 vers, prog, progp->pg_name);
1268 serv->sv_stats->rpcbadfmt++;
1269 svc_putnl(resv, RPC_PROG_MISMATCH);
1270 svc_putnl(resv, progp->pg_lovers);
1271 svc_putnl(resv, progp->pg_hivers);
1275 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1277 serv->sv_stats->rpcbadfmt++;
1278 svc_putnl(resv, RPC_PROC_UNAVAIL);
1282 svc_printk(rqstp, "failed to decode args\n");
1284 rpc_stat = rpc_garbage_args;
1286 serv->sv_stats->rpcbadfmt++;
1287 svc_putnl(resv, ntohl(rpc_stat));
1292 * Process the RPC request.
1295 svc_process(struct svc_rqst *rqstp)
1297 struct kvec *argv = &rqstp->rq_arg.head[0];
1298 struct kvec *resv = &rqstp->rq_res.head[0];
1299 struct svc_serv *serv = rqstp->rq_server;
1303 * Setup response xdr_buf.
1304 * Initially it has just one page
1306 rqstp->rq_next_page = &rqstp->rq_respages[1];
1307 resv->iov_base = page_address(rqstp->rq_respages[0]);
1309 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1310 rqstp->rq_res.len = 0;
1311 rqstp->rq_res.page_base = 0;
1312 rqstp->rq_res.page_len = 0;
1313 rqstp->rq_res.buflen = PAGE_SIZE;
1314 rqstp->rq_res.tail[0].iov_base = NULL;
1315 rqstp->rq_res.tail[0].iov_len = 0;
1317 dir = svc_getnl(argv);
1319 /* direction != CALL */
1320 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1321 serv->sv_stats->rpcbadfmt++;
1325 /* Returns 1 for send, 0 for drop */
1326 if (likely(svc_process_common(rqstp, argv, resv))) {
1327 int ret = svc_send(rqstp);
1329 trace_svc_process(rqstp, ret);
1333 trace_svc_process(rqstp, 0);
1337 EXPORT_SYMBOL_GPL(svc_process);
1339 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1341 * Process a backchannel RPC request that arrived over an existing
1342 * outbound connection
1345 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1346 struct svc_rqst *rqstp)
1348 struct kvec *argv = &rqstp->rq_arg.head[0];
1349 struct kvec *resv = &rqstp->rq_res.head[0];
1350 struct rpc_task *task;
1354 dprintk("svc: %s(%p)\n", __func__, req);
1356 /* Build the svc_rqst used by the common processing routine */
1357 rqstp->rq_xprt = serv->sv_bc_xprt;
1358 rqstp->rq_xid = req->rq_xid;
1359 rqstp->rq_prot = req->rq_xprt->prot;
1360 rqstp->rq_server = serv;
1362 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1363 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1364 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1365 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1367 /* Adjust the argument buffer length */
1368 rqstp->rq_arg.len = req->rq_private_buf.len;
1369 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1370 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1371 rqstp->rq_arg.page_len = 0;
1372 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1373 rqstp->rq_arg.page_len)
1374 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1375 rqstp->rq_arg.head[0].iov_len;
1377 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1378 rqstp->rq_arg.page_len;
1380 /* reset result send buffer "put" position */
1384 * Skip the next two words because they've already been
1385 * processed in the transport
1387 svc_getu32(argv); /* XID */
1388 svc_getnl(argv); /* CALLDIR */
1390 /* Parse and execute the bc call */
1391 proc_error = svc_process_common(rqstp, argv, resv);
1393 atomic_inc(&req->rq_xprt->bc_free_slots);
1395 /* Processing error: drop the request */
1396 xprt_free_bc_request(req);
1400 /* Finally, send the reply synchronously */
1401 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1402 task = rpc_run_bc_task(req);
1404 error = PTR_ERR(task);
1408 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1409 error = task->tk_status;
1413 dprintk("svc: %s(), error=%d\n", __func__, error);
1416 EXPORT_SYMBOL_GPL(bc_svc_process);
1417 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1420 * Return (transport-specific) limit on the rpc payload.
1422 u32 svc_max_payload(const struct svc_rqst *rqstp)
1424 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1426 if (rqstp->rq_server->sv_max_payload < max)
1427 max = rqstp->rq_server->sv_max_payload;
1430 EXPORT_SYMBOL_GPL(svc_max_payload);