ASoC: Intel: Convert all sst_codecs data definition to c99 style
[linux-2.6/btrfs-unstable.git] / net / sunrpc / svc.c
blobbc0f5a0ecbdce2a5203bf707ebfba8c99821d0f7
1 /*
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/signal.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
17 #include <linux/in.h>
18 #include <linux/mm.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 */
52 static int
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;
57 int err;
59 mutex_lock(&svc_pool_map_mutex);
61 err = -EBUSY;
62 if (m->count)
63 goto out;
65 err = 0;
66 if (!strncmp(val, "auto", 4))
67 *ip = SVC_POOL_AUTO;
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;
74 else
75 err = -EINVAL;
77 out:
78 mutex_unlock(&svc_pool_map_mutex);
79 return err;
82 static int
83 param_get_pool_mode(char *buf, struct kernel_param *kp)
85 int *ip = (int *)kp->arg;
87 switch (*ip)
89 case SVC_POOL_AUTO:
90 return strlcpy(buf, "auto", 20);
91 case SVC_POOL_GLOBAL:
92 return strlcpy(buf, "global", 20);
93 case SVC_POOL_PERCPU:
94 return strlcpy(buf, "percpu", 20);
95 case SVC_POOL_PERNODE:
96 return strlcpy(buf, "pernode", 20);
97 default:
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.
109 static int
110 svc_pool_map_choose_mode(void)
112 unsigned int node;
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.
141 static int
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);
145 if (!m->to_pool)
146 goto fail;
147 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
148 if (!m->pool_to)
149 goto fail_free;
151 return 0;
153 fail_free:
154 kfree(m->to_pool);
155 m->to_pool = NULL;
156 fail:
157 return -ENOMEM;
161 * Initialise the pool map for SVC_POOL_PERCPU mode.
162 * Returns number of pools or <0 on error.
164 static int
165 svc_pool_map_init_percpu(struct svc_pool_map *m)
167 unsigned int maxpools = nr_cpu_ids;
168 unsigned int pidx = 0;
169 unsigned int cpu;
170 int err;
172 err = svc_pool_map_alloc_arrays(m, maxpools);
173 if (err)
174 return err;
176 for_each_online_cpu(cpu) {
177 BUG_ON(pidx >= maxpools);
178 m->to_pool[cpu] = pidx;
179 m->pool_to[pidx] = cpu;
180 pidx++;
182 /* cpus brought online later all get mapped to pool0, sorry */
184 return pidx;
189 * Initialise the pool map for SVC_POOL_PERNODE mode.
190 * Returns number of pools or <0 on error.
192 static int
193 svc_pool_map_init_pernode(struct svc_pool_map *m)
195 unsigned int maxpools = nr_node_ids;
196 unsigned int pidx = 0;
197 unsigned int node;
198 int err;
200 err = svc_pool_map_alloc_arrays(m, maxpools);
201 if (err)
202 return err;
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;
209 pidx++;
211 /* nodes brought online later all get mapped to pool0, sorry */
213 return pidx;
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.
222 unsigned int
223 svc_pool_map_get(void)
225 struct svc_pool_map *m = &svc_pool_map;
226 int npools = -1;
228 mutex_lock(&svc_pool_map_mutex);
230 if (m->count++) {
231 mutex_unlock(&svc_pool_map_mutex);
232 return m->npools;
235 if (m->mode == SVC_POOL_AUTO)
236 m->mode = svc_pool_map_choose_mode();
238 switch (m->mode) {
239 case SVC_POOL_PERCPU:
240 npools = svc_pool_map_init_percpu(m);
241 break;
242 case SVC_POOL_PERNODE:
243 npools = svc_pool_map_init_pernode(m);
244 break;
247 if (npools < 0) {
248 /* default, or memory allocation failure */
249 npools = 1;
250 m->mode = SVC_POOL_GLOBAL;
252 m->npools = npools;
254 mutex_unlock(&svc_pool_map_mutex);
255 return m->npools;
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.
266 void
267 svc_pool_map_put(void)
269 struct svc_pool_map *m = &svc_pool_map;
271 mutex_lock(&svc_pool_map_mutex);
273 if (!--m->count) {
274 kfree(m->to_pool);
275 m->to_pool = NULL;
276 kfree(m->pool_to);
277 m->pool_to = NULL;
278 m->npools = 0;
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;
289 if (m->count) {
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];
295 return NUMA_NO_NODE;
298 * Set the given thread's cpus_allowed mask so that it
299 * will only run on cpus in the given pool.
301 static inline void
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);
312 if (m->count == 0)
313 return;
315 switch (m->mode) {
316 case SVC_POOL_PERCPU:
318 set_cpus_allowed_ptr(task, cpumask_of(node));
319 break;
321 case SVC_POOL_PERNODE:
323 set_cpus_allowed_ptr(task, cpumask_of_node(node));
324 break;
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.
334 struct svc_pool *
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)) {
346 switch (m->mode) {
347 case SVC_POOL_PERCPU:
348 pidx = m->to_pool[cpu];
349 break;
350 case SVC_POOL_PERNODE:
351 pidx = m->to_pool[cpu_to_node(cpu)];
352 break;
355 return &serv->sv_pools[pidx % serv->sv_nrpools];
358 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
360 int err;
362 err = rpcb_create_local(net);
363 if (err)
364 return err;
366 /* Remove any stale portmap registrations */
367 svc_unregister(serv, net);
368 return 0;
370 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
372 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
374 svc_unregister(serv, net);
375 rpcb_put_local(net);
377 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
379 static int svc_uses_rpcbind(struct svc_serv *serv)
381 struct svc_program *progp;
382 unsigned int i;
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)
387 continue;
388 if (!progp->pg_vers[i]->vs_hidden)
389 return 1;
393 return 0;
396 int svc_bind(struct svc_serv *serv, struct net *net)
398 if (!svc_uses_rpcbind(serv))
399 return 0;
400 return svc_rpcb_setup(serv, net);
402 EXPORT_SYMBOL_GPL(svc_bind);
404 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
405 static void
406 __svc_init_bc(struct svc_serv *serv)
408 INIT_LIST_HEAD(&serv->sv_cb_list);
409 spin_lock_init(&serv->sv_cb_lock);
410 init_waitqueue_head(&serv->sv_cb_waitq);
412 #else
413 static void
414 __svc_init_bc(struct svc_serv *serv)
417 #endif
420 * Create an RPC service
422 static struct svc_serv *
423 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
424 struct svc_serv_ops *ops)
426 struct svc_serv *serv;
427 unsigned int vers;
428 unsigned int xdrsize;
429 unsigned int i;
431 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
432 return NULL;
433 serv->sv_name = prog->pg_name;
434 serv->sv_program = prog;
435 serv->sv_nrthreads = 1;
436 serv->sv_stats = prog->pg_stats;
437 if (bufsize > RPCSVC_MAXPAYLOAD)
438 bufsize = RPCSVC_MAXPAYLOAD;
439 serv->sv_max_payload = bufsize? bufsize : 4096;
440 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
441 serv->sv_ops = ops;
442 xdrsize = 0;
443 while (prog) {
444 prog->pg_lovers = prog->pg_nvers-1;
445 for (vers=0; vers<prog->pg_nvers ; vers++)
446 if (prog->pg_vers[vers]) {
447 prog->pg_hivers = vers;
448 if (prog->pg_lovers > vers)
449 prog->pg_lovers = vers;
450 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
451 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
453 prog = prog->pg_next;
455 serv->sv_xdrsize = xdrsize;
456 INIT_LIST_HEAD(&serv->sv_tempsocks);
457 INIT_LIST_HEAD(&serv->sv_permsocks);
458 init_timer(&serv->sv_temptimer);
459 spin_lock_init(&serv->sv_lock);
461 __svc_init_bc(serv);
463 serv->sv_nrpools = npools;
464 serv->sv_pools =
465 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
466 GFP_KERNEL);
467 if (!serv->sv_pools) {
468 kfree(serv);
469 return NULL;
472 for (i = 0; i < serv->sv_nrpools; i++) {
473 struct svc_pool *pool = &serv->sv_pools[i];
475 dprintk("svc: initialising pool %u for %s\n",
476 i, serv->sv_name);
478 pool->sp_id = i;
479 INIT_LIST_HEAD(&pool->sp_sockets);
480 INIT_LIST_HEAD(&pool->sp_all_threads);
481 spin_lock_init(&pool->sp_lock);
484 return serv;
487 struct svc_serv *
488 svc_create(struct svc_program *prog, unsigned int bufsize,
489 struct svc_serv_ops *ops)
491 return __svc_create(prog, bufsize, /*npools*/1, ops);
493 EXPORT_SYMBOL_GPL(svc_create);
495 struct svc_serv *
496 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
497 struct svc_serv_ops *ops)
499 struct svc_serv *serv;
500 unsigned int npools = svc_pool_map_get();
502 serv = __svc_create(prog, bufsize, npools, ops);
503 if (!serv)
504 goto out_err;
505 return serv;
506 out_err:
507 svc_pool_map_put();
508 return NULL;
510 EXPORT_SYMBOL_GPL(svc_create_pooled);
512 void svc_shutdown_net(struct svc_serv *serv, struct net *net)
514 svc_close_net(serv, net);
516 if (serv->sv_ops->svo_shutdown)
517 serv->sv_ops->svo_shutdown(serv, net);
519 EXPORT_SYMBOL_GPL(svc_shutdown_net);
522 * Destroy an RPC service. Should be called with appropriate locking to
523 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
525 void
526 svc_destroy(struct svc_serv *serv)
528 dprintk("svc: svc_destroy(%s, %d)\n",
529 serv->sv_program->pg_name,
530 serv->sv_nrthreads);
532 if (serv->sv_nrthreads) {
533 if (--(serv->sv_nrthreads) != 0) {
534 svc_sock_update_bufs(serv);
535 return;
537 } else
538 printk("svc_destroy: no threads for serv=%p!\n", serv);
540 del_timer_sync(&serv->sv_temptimer);
543 * The last user is gone and thus all sockets have to be destroyed to
544 * the point. Check this.
546 BUG_ON(!list_empty(&serv->sv_permsocks));
547 BUG_ON(!list_empty(&serv->sv_tempsocks));
549 cache_clean_deferred(serv);
551 if (svc_serv_is_pooled(serv))
552 svc_pool_map_put();
554 kfree(serv->sv_pools);
555 kfree(serv);
557 EXPORT_SYMBOL_GPL(svc_destroy);
560 * Allocate an RPC server's buffer space.
561 * We allocate pages and place them in rq_argpages.
563 static int
564 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
566 unsigned int pages, arghi;
568 /* bc_xprt uses fore channel allocated buffers */
569 if (svc_is_backchannel(rqstp))
570 return 1;
572 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
573 * We assume one is at most one page
575 arghi = 0;
576 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
577 if (pages > RPCSVC_MAXPAGES)
578 pages = RPCSVC_MAXPAGES;
579 while (pages) {
580 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
581 if (!p)
582 break;
583 rqstp->rq_pages[arghi++] = p;
584 pages--;
586 return pages == 0;
590 * Release an RPC server buffer
592 static void
593 svc_release_buffer(struct svc_rqst *rqstp)
595 unsigned int i;
597 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
598 if (rqstp->rq_pages[i])
599 put_page(rqstp->rq_pages[i]);
602 struct svc_rqst *
603 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
605 struct svc_rqst *rqstp;
607 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
608 if (!rqstp)
609 return rqstp;
611 __set_bit(RQ_BUSY, &rqstp->rq_flags);
612 spin_lock_init(&rqstp->rq_lock);
613 rqstp->rq_server = serv;
614 rqstp->rq_pool = pool;
616 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
617 if (!rqstp->rq_argp)
618 goto out_enomem;
620 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
621 if (!rqstp->rq_resp)
622 goto out_enomem;
624 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
625 goto out_enomem;
627 return rqstp;
628 out_enomem:
629 svc_rqst_free(rqstp);
630 return NULL;
632 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
634 struct svc_rqst *
635 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
637 struct svc_rqst *rqstp;
639 rqstp = svc_rqst_alloc(serv, pool, node);
640 if (!rqstp)
641 return ERR_PTR(-ENOMEM);
643 serv->sv_nrthreads++;
644 spin_lock_bh(&pool->sp_lock);
645 pool->sp_nrthreads++;
646 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
647 spin_unlock_bh(&pool->sp_lock);
648 return rqstp;
650 EXPORT_SYMBOL_GPL(svc_prepare_thread);
653 * Choose a pool in which to create a new thread, for svc_set_num_threads
655 static inline struct svc_pool *
656 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
658 if (pool != NULL)
659 return pool;
661 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
665 * Choose a thread to kill, for svc_set_num_threads
667 static inline struct task_struct *
668 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
670 unsigned int i;
671 struct task_struct *task = NULL;
673 if (pool != NULL) {
674 spin_lock_bh(&pool->sp_lock);
675 } else {
676 /* choose a pool in round-robin fashion */
677 for (i = 0; i < serv->sv_nrpools; i++) {
678 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
679 spin_lock_bh(&pool->sp_lock);
680 if (!list_empty(&pool->sp_all_threads))
681 goto found_pool;
682 spin_unlock_bh(&pool->sp_lock);
684 return NULL;
687 found_pool:
688 if (!list_empty(&pool->sp_all_threads)) {
689 struct svc_rqst *rqstp;
692 * Remove from the pool->sp_all_threads list
693 * so we don't try to kill it again.
695 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
696 set_bit(RQ_VICTIM, &rqstp->rq_flags);
697 list_del_rcu(&rqstp->rq_all);
698 task = rqstp->rq_task;
700 spin_unlock_bh(&pool->sp_lock);
702 return task;
705 /* create new threads */
706 static int
707 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
709 struct svc_rqst *rqstp;
710 struct task_struct *task;
711 struct svc_pool *chosen_pool;
712 unsigned int state = serv->sv_nrthreads-1;
713 int node;
715 do {
716 nrservs--;
717 chosen_pool = choose_pool(serv, pool, &state);
719 node = svc_pool_map_get_node(chosen_pool->sp_id);
720 rqstp = svc_prepare_thread(serv, chosen_pool, node);
721 if (IS_ERR(rqstp))
722 return PTR_ERR(rqstp);
724 __module_get(serv->sv_ops->svo_module);
725 task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
726 node, "%s", serv->sv_name);
727 if (IS_ERR(task)) {
728 module_put(serv->sv_ops->svo_module);
729 svc_exit_thread(rqstp);
730 return PTR_ERR(task);
733 rqstp->rq_task = task;
734 if (serv->sv_nrpools > 1)
735 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
737 svc_sock_update_bufs(serv);
738 wake_up_process(task);
739 } while (nrservs > 0);
741 return 0;
745 /* destroy old threads */
746 static int
747 svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
749 struct task_struct *task;
750 unsigned int state = serv->sv_nrthreads-1;
752 /* destroy old threads */
753 do {
754 task = choose_victim(serv, pool, &state);
755 if (task == NULL)
756 break;
757 send_sig(SIGINT, task, 1);
758 nrservs++;
759 } while (nrservs < 0);
761 return 0;
765 * Create or destroy enough new threads to make the number
766 * of threads the given number. If `pool' is non-NULL, applies
767 * only to threads in that pool, otherwise round-robins between
768 * all pools. Caller must ensure that mutual exclusion between this and
769 * server startup or shutdown.
771 * Destroying threads relies on the service threads filling in
772 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
773 * has been created using svc_create_pooled().
775 * Based on code that used to be in nfsd_svc() but tweaked
776 * to be pool-aware.
779 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
781 if (pool == NULL) {
782 /* The -1 assumes caller has done a svc_get() */
783 nrservs -= (serv->sv_nrthreads-1);
784 } else {
785 spin_lock_bh(&pool->sp_lock);
786 nrservs -= pool->sp_nrthreads;
787 spin_unlock_bh(&pool->sp_lock);
790 if (nrservs > 0)
791 return svc_start_kthreads(serv, pool, nrservs);
792 if (nrservs < 0)
793 return svc_signal_kthreads(serv, pool, nrservs);
794 return 0;
796 EXPORT_SYMBOL_GPL(svc_set_num_threads);
798 /* destroy old threads */
799 static int
800 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
802 struct task_struct *task;
803 unsigned int state = serv->sv_nrthreads-1;
805 /* destroy old threads */
806 do {
807 task = choose_victim(serv, pool, &state);
808 if (task == NULL)
809 break;
810 kthread_stop(task);
811 nrservs++;
812 } while (nrservs < 0);
813 return 0;
817 svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
819 if (pool == NULL) {
820 /* The -1 assumes caller has done a svc_get() */
821 nrservs -= (serv->sv_nrthreads-1);
822 } else {
823 spin_lock_bh(&pool->sp_lock);
824 nrservs -= pool->sp_nrthreads;
825 spin_unlock_bh(&pool->sp_lock);
828 if (nrservs > 0)
829 return svc_start_kthreads(serv, pool, nrservs);
830 if (nrservs < 0)
831 return svc_stop_kthreads(serv, pool, nrservs);
832 return 0;
834 EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
837 * Called from a server thread as it's exiting. Caller must hold the "service
838 * mutex" for the service.
840 void
841 svc_rqst_free(struct svc_rqst *rqstp)
843 svc_release_buffer(rqstp);
844 kfree(rqstp->rq_resp);
845 kfree(rqstp->rq_argp);
846 kfree(rqstp->rq_auth_data);
847 kfree_rcu(rqstp, rq_rcu_head);
849 EXPORT_SYMBOL_GPL(svc_rqst_free);
851 void
852 svc_exit_thread(struct svc_rqst *rqstp)
854 struct svc_serv *serv = rqstp->rq_server;
855 struct svc_pool *pool = rqstp->rq_pool;
857 spin_lock_bh(&pool->sp_lock);
858 pool->sp_nrthreads--;
859 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
860 list_del_rcu(&rqstp->rq_all);
861 spin_unlock_bh(&pool->sp_lock);
863 svc_rqst_free(rqstp);
865 /* Release the server */
866 if (serv)
867 svc_destroy(serv);
869 EXPORT_SYMBOL_GPL(svc_exit_thread);
872 * Register an "inet" protocol family netid with the local
873 * rpcbind daemon via an rpcbind v4 SET request.
875 * No netconfig infrastructure is available in the kernel, so
876 * we map IP_ protocol numbers to netids by hand.
878 * Returns zero on success; a negative errno value is returned
879 * if any error occurs.
881 static int __svc_rpcb_register4(struct net *net, const u32 program,
882 const u32 version,
883 const unsigned short protocol,
884 const unsigned short port)
886 const struct sockaddr_in sin = {
887 .sin_family = AF_INET,
888 .sin_addr.s_addr = htonl(INADDR_ANY),
889 .sin_port = htons(port),
891 const char *netid;
892 int error;
894 switch (protocol) {
895 case IPPROTO_UDP:
896 netid = RPCBIND_NETID_UDP;
897 break;
898 case IPPROTO_TCP:
899 netid = RPCBIND_NETID_TCP;
900 break;
901 default:
902 return -ENOPROTOOPT;
905 error = rpcb_v4_register(net, program, version,
906 (const struct sockaddr *)&sin, netid);
909 * User space didn't support rpcbind v4, so retry this
910 * registration request with the legacy rpcbind v2 protocol.
912 if (error == -EPROTONOSUPPORT)
913 error = rpcb_register(net, program, version, protocol, port);
915 return error;
918 #if IS_ENABLED(CONFIG_IPV6)
920 * Register an "inet6" protocol family netid with the local
921 * rpcbind daemon via an rpcbind v4 SET request.
923 * No netconfig infrastructure is available in the kernel, so
924 * we map IP_ protocol numbers to netids by hand.
926 * Returns zero on success; a negative errno value is returned
927 * if any error occurs.
929 static int __svc_rpcb_register6(struct net *net, const u32 program,
930 const u32 version,
931 const unsigned short protocol,
932 const unsigned short port)
934 const struct sockaddr_in6 sin6 = {
935 .sin6_family = AF_INET6,
936 .sin6_addr = IN6ADDR_ANY_INIT,
937 .sin6_port = htons(port),
939 const char *netid;
940 int error;
942 switch (protocol) {
943 case IPPROTO_UDP:
944 netid = RPCBIND_NETID_UDP6;
945 break;
946 case IPPROTO_TCP:
947 netid = RPCBIND_NETID_TCP6;
948 break;
949 default:
950 return -ENOPROTOOPT;
953 error = rpcb_v4_register(net, program, version,
954 (const struct sockaddr *)&sin6, netid);
957 * User space didn't support rpcbind version 4, so we won't
958 * use a PF_INET6 listener.
960 if (error == -EPROTONOSUPPORT)
961 error = -EAFNOSUPPORT;
963 return error;
965 #endif /* IS_ENABLED(CONFIG_IPV6) */
968 * Register a kernel RPC service via rpcbind version 4.
970 * Returns zero on success; a negative errno value is returned
971 * if any error occurs.
973 static int __svc_register(struct net *net, const char *progname,
974 const u32 program, const u32 version,
975 const int family,
976 const unsigned short protocol,
977 const unsigned short port)
979 int error = -EAFNOSUPPORT;
981 switch (family) {
982 case PF_INET:
983 error = __svc_rpcb_register4(net, program, version,
984 protocol, port);
985 break;
986 #if IS_ENABLED(CONFIG_IPV6)
987 case PF_INET6:
988 error = __svc_rpcb_register6(net, program, version,
989 protocol, port);
990 #endif
993 return error;
997 * svc_register - register an RPC service with the local portmapper
998 * @serv: svc_serv struct for the service to register
999 * @net: net namespace for the service to register
1000 * @family: protocol family of service's listener socket
1001 * @proto: transport protocol number to advertise
1002 * @port: port to advertise
1004 * Service is registered for any address in the passed-in protocol family
1006 int svc_register(const struct svc_serv *serv, struct net *net,
1007 const int family, const unsigned short proto,
1008 const unsigned short port)
1010 struct svc_program *progp;
1011 struct svc_version *vers;
1012 unsigned int i;
1013 int error = 0;
1015 WARN_ON_ONCE(proto == 0 && port == 0);
1016 if (proto == 0 && port == 0)
1017 return -EINVAL;
1019 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1020 for (i = 0; i < progp->pg_nvers; i++) {
1021 vers = progp->pg_vers[i];
1022 if (vers == NULL)
1023 continue;
1025 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
1026 progp->pg_name,
1028 proto == IPPROTO_UDP? "udp" : "tcp",
1029 port,
1030 family,
1031 vers->vs_hidden ?
1032 " (but not telling portmap)" : "");
1034 if (vers->vs_hidden)
1035 continue;
1038 * Don't register a UDP port if we need congestion
1039 * control.
1041 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1042 continue;
1044 error = __svc_register(net, progp->pg_name, progp->pg_prog,
1045 i, family, proto, port);
1047 if (vers->vs_rpcb_optnl) {
1048 error = 0;
1049 continue;
1052 if (error < 0) {
1053 printk(KERN_WARNING "svc: failed to register "
1054 "%sv%u RPC service (errno %d).\n",
1055 progp->pg_name, i, -error);
1056 break;
1061 return error;
1065 * If user space is running rpcbind, it should take the v4 UNSET
1066 * and clear everything for this [program, version]. If user space
1067 * is running portmap, it will reject the v4 UNSET, but won't have
1068 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1069 * in this case to clear all existing entries for [program, version].
1071 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1072 const char *progname)
1074 int error;
1076 error = rpcb_v4_register(net, program, version, NULL, "");
1079 * User space didn't support rpcbind v4, so retry this
1080 * request with the legacy rpcbind v2 protocol.
1082 if (error == -EPROTONOSUPPORT)
1083 error = rpcb_register(net, program, version, 0, 0);
1085 dprintk("svc: %s(%sv%u), error %d\n",
1086 __func__, progname, version, error);
1090 * All netids, bind addresses and ports registered for [program, version]
1091 * are removed from the local rpcbind database (if the service is not
1092 * hidden) to make way for a new instance of the service.
1094 * The result of unregistration is reported via dprintk for those who want
1095 * verification of the result, but is otherwise not important.
1097 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1099 struct svc_program *progp;
1100 unsigned long flags;
1101 unsigned int i;
1103 clear_thread_flag(TIF_SIGPENDING);
1105 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1106 for (i = 0; i < progp->pg_nvers; i++) {
1107 if (progp->pg_vers[i] == NULL)
1108 continue;
1109 if (progp->pg_vers[i]->vs_hidden)
1110 continue;
1112 dprintk("svc: attempting to unregister %sv%u\n",
1113 progp->pg_name, i);
1114 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1118 spin_lock_irqsave(&current->sighand->siglock, flags);
1119 recalc_sigpending();
1120 spin_unlock_irqrestore(&current->sighand->siglock, flags);
1124 * dprintk the given error with the address of the client that caused it.
1126 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1127 static __printf(2, 3)
1128 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1130 struct va_format vaf;
1131 va_list args;
1132 char buf[RPC_MAX_ADDRBUFLEN];
1134 va_start(args, fmt);
1136 vaf.fmt = fmt;
1137 vaf.va = &args;
1139 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1141 va_end(args);
1143 #else
1144 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1145 #endif
1148 * Common routine for processing the RPC request.
1150 static int
1151 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1153 struct svc_program *progp;
1154 struct svc_version *versp = NULL; /* compiler food */
1155 struct svc_procedure *procp = NULL;
1156 struct svc_serv *serv = rqstp->rq_server;
1157 kxdrproc_t xdr;
1158 __be32 *statp;
1159 u32 prog, vers, proc;
1160 __be32 auth_stat, rpc_stat;
1161 int auth_res;
1162 __be32 *reply_statp;
1164 rpc_stat = rpc_success;
1166 if (argv->iov_len < 6*4)
1167 goto err_short_len;
1169 /* Will be turned off only in gss privacy case: */
1170 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1171 /* Will be turned off only when NFSv4 Sessions are used */
1172 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1173 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1175 /* Setup reply header */
1176 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1178 svc_putu32(resv, rqstp->rq_xid);
1180 vers = svc_getnl(argv);
1182 /* First words of reply: */
1183 svc_putnl(resv, 1); /* REPLY */
1185 if (vers != 2) /* RPC version number */
1186 goto err_bad_rpc;
1188 /* Save position in case we later decide to reject: */
1189 reply_statp = resv->iov_base + resv->iov_len;
1191 svc_putnl(resv, 0); /* ACCEPT */
1193 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1194 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1195 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1197 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1198 if (prog == progp->pg_prog)
1199 break;
1202 * Decode auth data, and add verifier to reply buffer.
1203 * We do this before anything else in order to get a decent
1204 * auth verifier.
1206 auth_res = svc_authenticate(rqstp, &auth_stat);
1207 /* Also give the program a chance to reject this call: */
1208 if (auth_res == SVC_OK && progp) {
1209 auth_stat = rpc_autherr_badcred;
1210 auth_res = progp->pg_authenticate(rqstp);
1212 switch (auth_res) {
1213 case SVC_OK:
1214 break;
1215 case SVC_GARBAGE:
1216 goto err_garbage;
1217 case SVC_SYSERR:
1218 rpc_stat = rpc_system_err;
1219 goto err_bad;
1220 case SVC_DENIED:
1221 goto err_bad_auth;
1222 case SVC_CLOSE:
1223 goto close;
1224 case SVC_DROP:
1225 goto dropit;
1226 case SVC_COMPLETE:
1227 goto sendit;
1230 if (progp == NULL)
1231 goto err_bad_prog;
1233 if (vers >= progp->pg_nvers ||
1234 !(versp = progp->pg_vers[vers]))
1235 goto err_bad_vers;
1238 * Some protocol versions (namely NFSv4) require some form of
1239 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1240 * In other words, UDP is not allowed. We mark those when setting
1241 * up the svc_xprt, and verify that here.
1243 * The spec is not very clear about what error should be returned
1244 * when someone tries to access a server that is listening on UDP
1245 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1246 * fit.
1248 if (versp->vs_need_cong_ctrl &&
1249 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1250 goto err_bad_vers;
1252 procp = versp->vs_proc + proc;
1253 if (proc >= versp->vs_nproc || !procp->pc_func)
1254 goto err_bad_proc;
1255 rqstp->rq_procinfo = procp;
1257 /* Syntactic check complete */
1258 serv->sv_stats->rpccnt++;
1260 /* Build the reply header. */
1261 statp = resv->iov_base +resv->iov_len;
1262 svc_putnl(resv, RPC_SUCCESS);
1264 /* Bump per-procedure stats counter */
1265 procp->pc_count++;
1267 /* Initialize storage for argp and resp */
1268 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1269 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1271 /* un-reserve some of the out-queue now that we have a
1272 * better idea of reply size
1274 if (procp->pc_xdrressize)
1275 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1277 /* Call the function that processes the request. */
1278 if (!versp->vs_dispatch) {
1279 /* Decode arguments */
1280 xdr = procp->pc_decode;
1281 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1282 goto err_garbage;
1284 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1286 /* Encode reply */
1287 if (*statp == rpc_drop_reply ||
1288 test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1289 if (procp->pc_release)
1290 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1291 goto dropit;
1293 if (*statp == rpc_autherr_badcred) {
1294 if (procp->pc_release)
1295 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1296 goto err_bad_auth;
1298 if (*statp == rpc_success &&
1299 (xdr = procp->pc_encode) &&
1300 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1301 dprintk("svc: failed to encode reply\n");
1302 /* serv->sv_stats->rpcsystemerr++; */
1303 *statp = rpc_system_err;
1305 } else {
1306 dprintk("svc: calling dispatcher\n");
1307 if (!versp->vs_dispatch(rqstp, statp)) {
1308 /* Release reply info */
1309 if (procp->pc_release)
1310 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1311 goto dropit;
1315 /* Check RPC status result */
1316 if (*statp != rpc_success)
1317 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1319 /* Release reply info */
1320 if (procp->pc_release)
1321 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1323 if (procp->pc_encode == NULL)
1324 goto dropit;
1326 sendit:
1327 if (svc_authorise(rqstp))
1328 goto close;
1329 return 1; /* Caller can now send it */
1331 dropit:
1332 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1333 dprintk("svc: svc_process dropit\n");
1334 return 0;
1336 close:
1337 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1338 svc_close_xprt(rqstp->rq_xprt);
1339 dprintk("svc: svc_process close\n");
1340 return 0;
1342 err_short_len:
1343 svc_printk(rqstp, "short len %zd, dropping request\n",
1344 argv->iov_len);
1345 goto close;
1347 err_bad_rpc:
1348 serv->sv_stats->rpcbadfmt++;
1349 svc_putnl(resv, 1); /* REJECT */
1350 svc_putnl(resv, 0); /* RPC_MISMATCH */
1351 svc_putnl(resv, 2); /* Only RPCv2 supported */
1352 svc_putnl(resv, 2);
1353 goto sendit;
1355 err_bad_auth:
1356 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1357 serv->sv_stats->rpcbadauth++;
1358 /* Restore write pointer to location of accept status: */
1359 xdr_ressize_check(rqstp, reply_statp);
1360 svc_putnl(resv, 1); /* REJECT */
1361 svc_putnl(resv, 1); /* AUTH_ERROR */
1362 svc_putnl(resv, ntohl(auth_stat)); /* status */
1363 goto sendit;
1365 err_bad_prog:
1366 dprintk("svc: unknown program %d\n", prog);
1367 serv->sv_stats->rpcbadfmt++;
1368 svc_putnl(resv, RPC_PROG_UNAVAIL);
1369 goto sendit;
1371 err_bad_vers:
1372 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1373 vers, prog, progp->pg_name);
1375 serv->sv_stats->rpcbadfmt++;
1376 svc_putnl(resv, RPC_PROG_MISMATCH);
1377 svc_putnl(resv, progp->pg_lovers);
1378 svc_putnl(resv, progp->pg_hivers);
1379 goto sendit;
1381 err_bad_proc:
1382 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1384 serv->sv_stats->rpcbadfmt++;
1385 svc_putnl(resv, RPC_PROC_UNAVAIL);
1386 goto sendit;
1388 err_garbage:
1389 svc_printk(rqstp, "failed to decode args\n");
1391 rpc_stat = rpc_garbage_args;
1392 err_bad:
1393 serv->sv_stats->rpcbadfmt++;
1394 svc_putnl(resv, ntohl(rpc_stat));
1395 goto sendit;
1399 * Process the RPC request.
1402 svc_process(struct svc_rqst *rqstp)
1404 struct kvec *argv = &rqstp->rq_arg.head[0];
1405 struct kvec *resv = &rqstp->rq_res.head[0];
1406 struct svc_serv *serv = rqstp->rq_server;
1407 u32 dir;
1410 * Setup response xdr_buf.
1411 * Initially it has just one page
1413 rqstp->rq_next_page = &rqstp->rq_respages[1];
1414 resv->iov_base = page_address(rqstp->rq_respages[0]);
1415 resv->iov_len = 0;
1416 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1417 rqstp->rq_res.len = 0;
1418 rqstp->rq_res.page_base = 0;
1419 rqstp->rq_res.page_len = 0;
1420 rqstp->rq_res.buflen = PAGE_SIZE;
1421 rqstp->rq_res.tail[0].iov_base = NULL;
1422 rqstp->rq_res.tail[0].iov_len = 0;
1424 dir = svc_getnl(argv);
1425 if (dir != 0) {
1426 /* direction != CALL */
1427 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1428 serv->sv_stats->rpcbadfmt++;
1429 goto out_drop;
1432 /* Returns 1 for send, 0 for drop */
1433 if (likely(svc_process_common(rqstp, argv, resv))) {
1434 int ret = svc_send(rqstp);
1436 trace_svc_process(rqstp, ret);
1437 return ret;
1439 out_drop:
1440 trace_svc_process(rqstp, 0);
1441 svc_drop(rqstp);
1442 return 0;
1444 EXPORT_SYMBOL_GPL(svc_process);
1446 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1448 * Process a backchannel RPC request that arrived over an existing
1449 * outbound connection
1452 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1453 struct svc_rqst *rqstp)
1455 struct kvec *argv = &rqstp->rq_arg.head[0];
1456 struct kvec *resv = &rqstp->rq_res.head[0];
1457 struct rpc_task *task;
1458 int proc_error;
1459 int error;
1461 dprintk("svc: %s(%p)\n", __func__, req);
1463 /* Build the svc_rqst used by the common processing routine */
1464 rqstp->rq_xprt = serv->sv_bc_xprt;
1465 rqstp->rq_xid = req->rq_xid;
1466 rqstp->rq_prot = req->rq_xprt->prot;
1467 rqstp->rq_server = serv;
1469 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1470 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1471 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1472 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1474 /* Adjust the argument buffer length */
1475 rqstp->rq_arg.len = req->rq_private_buf.len;
1476 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1477 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1478 rqstp->rq_arg.page_len = 0;
1479 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1480 rqstp->rq_arg.page_len)
1481 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1482 rqstp->rq_arg.head[0].iov_len;
1483 else
1484 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1485 rqstp->rq_arg.page_len;
1487 /* reset result send buffer "put" position */
1488 resv->iov_len = 0;
1491 * Skip the next two words because they've already been
1492 * processed in the transport
1494 svc_getu32(argv); /* XID */
1495 svc_getnl(argv); /* CALLDIR */
1497 /* Parse and execute the bc call */
1498 proc_error = svc_process_common(rqstp, argv, resv);
1500 atomic_inc(&req->rq_xprt->bc_free_slots);
1501 if (!proc_error) {
1502 /* Processing error: drop the request */
1503 xprt_free_bc_request(req);
1504 return 0;
1507 /* Finally, send the reply synchronously */
1508 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1509 task = rpc_run_bc_task(req);
1510 if (IS_ERR(task)) {
1511 error = PTR_ERR(task);
1512 goto out;
1515 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1516 error = task->tk_status;
1517 rpc_put_task(task);
1519 out:
1520 dprintk("svc: %s(), error=%d\n", __func__, error);
1521 return error;
1523 EXPORT_SYMBOL_GPL(bc_svc_process);
1524 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1527 * Return (transport-specific) limit on the rpc payload.
1529 u32 svc_max_payload(const struct svc_rqst *rqstp)
1531 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1533 if (rqstp->rq_server->sv_max_payload < max)
1534 max = rqstp->rq_server->sv_max_payload;
1535 return max;
1537 EXPORT_SYMBOL_GPL(svc_max_payload);