Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svc.c
4 *
5 * High-level RPC service routines
6 *
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 *
9 * Multiple threads pools and NUMAisation
10 * Copyright (c) 2006 Silicon Graphics, Inc.
11 * by Greg Banks <gnb@melbourne.sgi.com>
12 */
13
14#include <linux/linkage.h>
15#include <linux/sched/signal.h>
16#include <linux/errno.h>
17#include <linux/net.h>
18#include <linux/in.h>
19#include <linux/mm.h>
20#include <linux/interrupt.h>
21#include <linux/module.h>
22#include <linux/kthread.h>
23#include <linux/slab.h>
24
25#include <linux/sunrpc/types.h>
26#include <linux/sunrpc/xdr.h>
27#include <linux/sunrpc/stats.h>
28#include <linux/sunrpc/svcsock.h>
29#include <linux/sunrpc/clnt.h>
30#include <linux/sunrpc/bc_xprt.h>
31
32#include <trace/events/sunrpc.h>
33
34#include "fail.h"
35
36#define RPCDBG_FACILITY RPCDBG_SVCDSP
37
38static void svc_unregister(const struct svc_serv *serv, struct net *net);
39
40#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
41
42/*
43 * Mode for mapping cpus to pools.
44 */
45enum {
46 SVC_POOL_AUTO = -1, /* choose one of the others */
47 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
48 * (legacy & UP mode) */
49 SVC_POOL_PERCPU, /* one pool per cpu */
50 SVC_POOL_PERNODE /* one pool per numa node */
51};
52
53/*
54 * Structure for mapping cpus to pools and vice versa.
55 * Setup once during sunrpc initialisation.
56 */
57
58struct svc_pool_map {
59 int count; /* How many svc_servs use us */
60 int mode; /* Note: int not enum to avoid
61 * warnings about "enumeration value
62 * not handled in switch" */
63 unsigned int npools;
64 unsigned int *pool_to; /* maps pool id to cpu or node */
65 unsigned int *to_pool; /* maps cpu or node to pool id */
66};
67
68static struct svc_pool_map svc_pool_map = {
69 .mode = SVC_POOL_DEFAULT
70};
71
72static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
73
74static int
75param_set_pool_mode(const char *val, const struct kernel_param *kp)
76{
77 int *ip = (int *)kp->arg;
78 struct svc_pool_map *m = &svc_pool_map;
79 int err;
80
81 mutex_lock(&svc_pool_map_mutex);
82
83 err = -EBUSY;
84 if (m->count)
85 goto out;
86
87 err = 0;
88 if (!strncmp(val, "auto", 4))
89 *ip = SVC_POOL_AUTO;
90 else if (!strncmp(val, "global", 6))
91 *ip = SVC_POOL_GLOBAL;
92 else if (!strncmp(val, "percpu", 6))
93 *ip = SVC_POOL_PERCPU;
94 else if (!strncmp(val, "pernode", 7))
95 *ip = SVC_POOL_PERNODE;
96 else
97 err = -EINVAL;
98
99out:
100 mutex_unlock(&svc_pool_map_mutex);
101 return err;
102}
103
104static int
105param_get_pool_mode(char *buf, const struct kernel_param *kp)
106{
107 int *ip = (int *)kp->arg;
108
109 switch (*ip)
110 {
111 case SVC_POOL_AUTO:
112 return strlcpy(buf, "auto\n", 20);
113 case SVC_POOL_GLOBAL:
114 return strlcpy(buf, "global\n", 20);
115 case SVC_POOL_PERCPU:
116 return strlcpy(buf, "percpu\n", 20);
117 case SVC_POOL_PERNODE:
118 return strlcpy(buf, "pernode\n", 20);
119 default:
120 return sprintf(buf, "%d\n", *ip);
121 }
122}
123
124module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
125 &svc_pool_map.mode, 0644);
126
127/*
128 * Detect best pool mapping mode heuristically,
129 * according to the machine's topology.
130 */
131static int
132svc_pool_map_choose_mode(void)
133{
134 unsigned int node;
135
136 if (nr_online_nodes > 1) {
137 /*
138 * Actually have multiple NUMA nodes,
139 * so split pools on NUMA node boundaries
140 */
141 return SVC_POOL_PERNODE;
142 }
143
144 node = first_online_node;
145 if (nr_cpus_node(node) > 2) {
146 /*
147 * Non-trivial SMP, or CONFIG_NUMA on
148 * non-NUMA hardware, e.g. with a generic
149 * x86_64 kernel on Xeons. In this case we
150 * want to divide the pools on cpu boundaries.
151 */
152 return SVC_POOL_PERCPU;
153 }
154
155 /* default: one global pool */
156 return SVC_POOL_GLOBAL;
157}
158
159/*
160 * Allocate the to_pool[] and pool_to[] arrays.
161 * Returns 0 on success or an errno.
162 */
163static int
164svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
165{
166 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
167 if (!m->to_pool)
168 goto fail;
169 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
170 if (!m->pool_to)
171 goto fail_free;
172
173 return 0;
174
175fail_free:
176 kfree(m->to_pool);
177 m->to_pool = NULL;
178fail:
179 return -ENOMEM;
180}
181
182/*
183 * Initialise the pool map for SVC_POOL_PERCPU mode.
184 * Returns number of pools or <0 on error.
185 */
186static int
187svc_pool_map_init_percpu(struct svc_pool_map *m)
188{
189 unsigned int maxpools = nr_cpu_ids;
190 unsigned int pidx = 0;
191 unsigned int cpu;
192 int err;
193
194 err = svc_pool_map_alloc_arrays(m, maxpools);
195 if (err)
196 return err;
197
198 for_each_online_cpu(cpu) {
199 BUG_ON(pidx >= maxpools);
200 m->to_pool[cpu] = pidx;
201 m->pool_to[pidx] = cpu;
202 pidx++;
203 }
204 /* cpus brought online later all get mapped to pool0, sorry */
205
206 return pidx;
207};
208
209
210/*
211 * Initialise the pool map for SVC_POOL_PERNODE mode.
212 * Returns number of pools or <0 on error.
213 */
214static int
215svc_pool_map_init_pernode(struct svc_pool_map *m)
216{
217 unsigned int maxpools = nr_node_ids;
218 unsigned int pidx = 0;
219 unsigned int node;
220 int err;
221
222 err = svc_pool_map_alloc_arrays(m, maxpools);
223 if (err)
224 return err;
225
226 for_each_node_with_cpus(node) {
227 /* some architectures (e.g. SN2) have cpuless nodes */
228 BUG_ON(pidx > maxpools);
229 m->to_pool[node] = pidx;
230 m->pool_to[pidx] = node;
231 pidx++;
232 }
233 /* nodes brought online later all get mapped to pool0, sorry */
234
235 return pidx;
236}
237
238
239/*
240 * Add a reference to the global map of cpus to pools (and
241 * vice versa) if pools are in use.
242 * Initialise the map if we're the first user.
243 * Returns the number of pools. If this is '1', no reference
244 * was taken.
245 */
246static unsigned int
247svc_pool_map_get(void)
248{
249 struct svc_pool_map *m = &svc_pool_map;
250 int npools = -1;
251
252 mutex_lock(&svc_pool_map_mutex);
253
254 if (m->count++) {
255 mutex_unlock(&svc_pool_map_mutex);
256 WARN_ON_ONCE(m->npools <= 1);
257 return m->npools;
258 }
259
260 if (m->mode == SVC_POOL_AUTO)
261 m->mode = svc_pool_map_choose_mode();
262
263 switch (m->mode) {
264 case SVC_POOL_PERCPU:
265 npools = svc_pool_map_init_percpu(m);
266 break;
267 case SVC_POOL_PERNODE:
268 npools = svc_pool_map_init_pernode(m);
269 break;
270 }
271
272 if (npools <= 0) {
273 /* default, or memory allocation failure */
274 npools = 1;
275 m->mode = SVC_POOL_GLOBAL;
276 }
277 m->npools = npools;
278
279 if (npools == 1)
280 /* service is unpooled, so doesn't hold a reference */
281 m->count--;
282
283 mutex_unlock(&svc_pool_map_mutex);
284 return npools;
285}
286
287/*
288 * Drop a reference to the global map of cpus to pools, if
289 * pools were in use, i.e. if npools > 1.
290 * When the last reference is dropped, the map data is
291 * freed; this allows the sysadmin to change the pool
292 * mode using the pool_mode module option without
293 * rebooting or re-loading sunrpc.ko.
294 */
295static void
296svc_pool_map_put(int npools)
297{
298 struct svc_pool_map *m = &svc_pool_map;
299
300 if (npools <= 1)
301 return;
302 mutex_lock(&svc_pool_map_mutex);
303
304 if (!--m->count) {
305 kfree(m->to_pool);
306 m->to_pool = NULL;
307 kfree(m->pool_to);
308 m->pool_to = NULL;
309 m->npools = 0;
310 }
311
312 mutex_unlock(&svc_pool_map_mutex);
313}
314
315static int svc_pool_map_get_node(unsigned int pidx)
316{
317 const struct svc_pool_map *m = &svc_pool_map;
318
319 if (m->count) {
320 if (m->mode == SVC_POOL_PERCPU)
321 return cpu_to_node(m->pool_to[pidx]);
322 if (m->mode == SVC_POOL_PERNODE)
323 return m->pool_to[pidx];
324 }
325 return NUMA_NO_NODE;
326}
327/*
328 * Set the given thread's cpus_allowed mask so that it
329 * will only run on cpus in the given pool.
330 */
331static inline void
332svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
333{
334 struct svc_pool_map *m = &svc_pool_map;
335 unsigned int node = m->pool_to[pidx];
336
337 /*
338 * The caller checks for sv_nrpools > 1, which
339 * implies that we've been initialized.
340 */
341 WARN_ON_ONCE(m->count == 0);
342 if (m->count == 0)
343 return;
344
345 switch (m->mode) {
346 case SVC_POOL_PERCPU:
347 {
348 set_cpus_allowed_ptr(task, cpumask_of(node));
349 break;
350 }
351 case SVC_POOL_PERNODE:
352 {
353 set_cpus_allowed_ptr(task, cpumask_of_node(node));
354 break;
355 }
356 }
357}
358
359/**
360 * svc_pool_for_cpu - Select pool to run a thread on this cpu
361 * @serv: An RPC service
362 *
363 * Use the active CPU and the svc_pool_map's mode setting to
364 * select the svc thread pool to use. Once initialized, the
365 * svc_pool_map does not change.
366 *
367 * Return value:
368 * A pointer to an svc_pool
369 */
370struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv)
371{
372 struct svc_pool_map *m = &svc_pool_map;
373 int cpu = raw_smp_processor_id();
374 unsigned int pidx = 0;
375
376 if (serv->sv_nrpools <= 1)
377 return serv->sv_pools;
378
379 switch (m->mode) {
380 case SVC_POOL_PERCPU:
381 pidx = m->to_pool[cpu];
382 break;
383 case SVC_POOL_PERNODE:
384 pidx = m->to_pool[cpu_to_node(cpu)];
385 break;
386 }
387
388 return &serv->sv_pools[pidx % serv->sv_nrpools];
389}
390
391int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
392{
393 int err;
394
395 err = rpcb_create_local(net);
396 if (err)
397 return err;
398
399 /* Remove any stale portmap registrations */
400 svc_unregister(serv, net);
401 return 0;
402}
403EXPORT_SYMBOL_GPL(svc_rpcb_setup);
404
405void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
406{
407 svc_unregister(serv, net);
408 rpcb_put_local(net);
409}
410EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
411
412static int svc_uses_rpcbind(struct svc_serv *serv)
413{
414 struct svc_program *progp;
415 unsigned int i;
416
417 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
418 for (i = 0; i < progp->pg_nvers; i++) {
419 if (progp->pg_vers[i] == NULL)
420 continue;
421 if (!progp->pg_vers[i]->vs_hidden)
422 return 1;
423 }
424 }
425
426 return 0;
427}
428
429int svc_bind(struct svc_serv *serv, struct net *net)
430{
431 if (!svc_uses_rpcbind(serv))
432 return 0;
433 return svc_rpcb_setup(serv, net);
434}
435EXPORT_SYMBOL_GPL(svc_bind);
436
437#if defined(CONFIG_SUNRPC_BACKCHANNEL)
438static void
439__svc_init_bc(struct svc_serv *serv)
440{
441 INIT_LIST_HEAD(&serv->sv_cb_list);
442 spin_lock_init(&serv->sv_cb_lock);
443 init_waitqueue_head(&serv->sv_cb_waitq);
444}
445#else
446static void
447__svc_init_bc(struct svc_serv *serv)
448{
449}
450#endif
451
452/*
453 * Create an RPC service
454 */
455static struct svc_serv *
456__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
457 int (*threadfn)(void *data))
458{
459 struct svc_serv *serv;
460 unsigned int vers;
461 unsigned int xdrsize;
462 unsigned int i;
463
464 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
465 return NULL;
466 serv->sv_name = prog->pg_name;
467 serv->sv_program = prog;
468 kref_init(&serv->sv_refcnt);
469 serv->sv_stats = prog->pg_stats;
470 if (bufsize > RPCSVC_MAXPAYLOAD)
471 bufsize = RPCSVC_MAXPAYLOAD;
472 serv->sv_max_payload = bufsize? bufsize : 4096;
473 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
474 serv->sv_threadfn = threadfn;
475 xdrsize = 0;
476 while (prog) {
477 prog->pg_lovers = prog->pg_nvers-1;
478 for (vers=0; vers<prog->pg_nvers ; vers++)
479 if (prog->pg_vers[vers]) {
480 prog->pg_hivers = vers;
481 if (prog->pg_lovers > vers)
482 prog->pg_lovers = vers;
483 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
484 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
485 }
486 prog = prog->pg_next;
487 }
488 serv->sv_xdrsize = xdrsize;
489 INIT_LIST_HEAD(&serv->sv_tempsocks);
490 INIT_LIST_HEAD(&serv->sv_permsocks);
491 timer_setup(&serv->sv_temptimer, NULL, 0);
492 spin_lock_init(&serv->sv_lock);
493
494 __svc_init_bc(serv);
495
496 serv->sv_nrpools = npools;
497 serv->sv_pools =
498 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
499 GFP_KERNEL);
500 if (!serv->sv_pools) {
501 kfree(serv);
502 return NULL;
503 }
504
505 for (i = 0; i < serv->sv_nrpools; i++) {
506 struct svc_pool *pool = &serv->sv_pools[i];
507
508 dprintk("svc: initialising pool %u for %s\n",
509 i, serv->sv_name);
510
511 pool->sp_id = i;
512 INIT_LIST_HEAD(&pool->sp_sockets);
513 INIT_LIST_HEAD(&pool->sp_all_threads);
514 spin_lock_init(&pool->sp_lock);
515 }
516
517 return serv;
518}
519
520/**
521 * svc_create - Create an RPC service
522 * @prog: the RPC program the new service will handle
523 * @bufsize: maximum message size for @prog
524 * @threadfn: a function to service RPC requests for @prog
525 *
526 * Returns an instantiated struct svc_serv object or NULL.
527 */
528struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize,
529 int (*threadfn)(void *data))
530{
531 return __svc_create(prog, bufsize, 1, threadfn);
532}
533EXPORT_SYMBOL_GPL(svc_create);
534
535/**
536 * svc_create_pooled - Create an RPC service with pooled threads
537 * @prog: the RPC program the new service will handle
538 * @bufsize: maximum message size for @prog
539 * @threadfn: a function to service RPC requests for @prog
540 *
541 * Returns an instantiated struct svc_serv object or NULL.
542 */
543struct svc_serv *svc_create_pooled(struct svc_program *prog,
544 unsigned int bufsize,
545 int (*threadfn)(void *data))
546{
547 struct svc_serv *serv;
548 unsigned int npools = svc_pool_map_get();
549
550 serv = __svc_create(prog, bufsize, npools, threadfn);
551 if (!serv)
552 goto out_err;
553 return serv;
554out_err:
555 svc_pool_map_put(npools);
556 return NULL;
557}
558EXPORT_SYMBOL_GPL(svc_create_pooled);
559
560/*
561 * Destroy an RPC service. Should be called with appropriate locking to
562 * protect sv_permsocks and sv_tempsocks.
563 */
564void
565svc_destroy(struct kref *ref)
566{
567 struct svc_serv *serv = container_of(ref, struct svc_serv, sv_refcnt);
568
569 dprintk("svc: svc_destroy(%s)\n", serv->sv_program->pg_name);
570 timer_shutdown_sync(&serv->sv_temptimer);
571
572 /*
573 * The last user is gone and thus all sockets have to be destroyed to
574 * the point. Check this.
575 */
576 BUG_ON(!list_empty(&serv->sv_permsocks));
577 BUG_ON(!list_empty(&serv->sv_tempsocks));
578
579 cache_clean_deferred(serv);
580
581 svc_pool_map_put(serv->sv_nrpools);
582
583 kfree(serv->sv_pools);
584 kfree(serv);
585}
586EXPORT_SYMBOL_GPL(svc_destroy);
587
588/*
589 * Allocate an RPC server's buffer space.
590 * We allocate pages and place them in rq_pages.
591 */
592static int
593svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
594{
595 unsigned int pages, arghi;
596
597 /* bc_xprt uses fore channel allocated buffers */
598 if (svc_is_backchannel(rqstp))
599 return 1;
600
601 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
602 * We assume one is at most one page
603 */
604 arghi = 0;
605 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
606 if (pages > RPCSVC_MAXPAGES)
607 pages = RPCSVC_MAXPAGES;
608 while (pages) {
609 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
610 if (!p)
611 break;
612 rqstp->rq_pages[arghi++] = p;
613 pages--;
614 }
615 return pages == 0;
616}
617
618/*
619 * Release an RPC server buffer
620 */
621static void
622svc_release_buffer(struct svc_rqst *rqstp)
623{
624 unsigned int i;
625
626 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
627 if (rqstp->rq_pages[i])
628 put_page(rqstp->rq_pages[i]);
629}
630
631struct svc_rqst *
632svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
633{
634 struct svc_rqst *rqstp;
635
636 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
637 if (!rqstp)
638 return rqstp;
639
640 __set_bit(RQ_BUSY, &rqstp->rq_flags);
641 rqstp->rq_server = serv;
642 rqstp->rq_pool = pool;
643
644 rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
645 if (!rqstp->rq_scratch_page)
646 goto out_enomem;
647
648 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
649 if (!rqstp->rq_argp)
650 goto out_enomem;
651
652 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
653 if (!rqstp->rq_resp)
654 goto out_enomem;
655
656 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
657 goto out_enomem;
658
659 return rqstp;
660out_enomem:
661 svc_rqst_free(rqstp);
662 return NULL;
663}
664EXPORT_SYMBOL_GPL(svc_rqst_alloc);
665
666static struct svc_rqst *
667svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
668{
669 struct svc_rqst *rqstp;
670
671 rqstp = svc_rqst_alloc(serv, pool, node);
672 if (!rqstp)
673 return ERR_PTR(-ENOMEM);
674
675 svc_get(serv);
676 spin_lock_bh(&serv->sv_lock);
677 serv->sv_nrthreads += 1;
678 spin_unlock_bh(&serv->sv_lock);
679
680 spin_lock_bh(&pool->sp_lock);
681 pool->sp_nrthreads++;
682 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
683 spin_unlock_bh(&pool->sp_lock);
684 return rqstp;
685}
686
687/*
688 * Choose a pool in which to create a new thread, for svc_set_num_threads
689 */
690static inline struct svc_pool *
691choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
692{
693 if (pool != NULL)
694 return pool;
695
696 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
697}
698
699/*
700 * Choose a thread to kill, for svc_set_num_threads
701 */
702static inline struct task_struct *
703choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
704{
705 unsigned int i;
706 struct task_struct *task = NULL;
707
708 if (pool != NULL) {
709 spin_lock_bh(&pool->sp_lock);
710 } else {
711 /* choose a pool in round-robin fashion */
712 for (i = 0; i < serv->sv_nrpools; i++) {
713 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
714 spin_lock_bh(&pool->sp_lock);
715 if (!list_empty(&pool->sp_all_threads))
716 goto found_pool;
717 spin_unlock_bh(&pool->sp_lock);
718 }
719 return NULL;
720 }
721
722found_pool:
723 if (!list_empty(&pool->sp_all_threads)) {
724 struct svc_rqst *rqstp;
725
726 /*
727 * Remove from the pool->sp_all_threads list
728 * so we don't try to kill it again.
729 */
730 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
731 set_bit(RQ_VICTIM, &rqstp->rq_flags);
732 list_del_rcu(&rqstp->rq_all);
733 task = rqstp->rq_task;
734 }
735 spin_unlock_bh(&pool->sp_lock);
736
737 return task;
738}
739
740/* create new threads */
741static int
742svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
743{
744 struct svc_rqst *rqstp;
745 struct task_struct *task;
746 struct svc_pool *chosen_pool;
747 unsigned int state = serv->sv_nrthreads-1;
748 int node;
749
750 do {
751 nrservs--;
752 chosen_pool = choose_pool(serv, pool, &state);
753
754 node = svc_pool_map_get_node(chosen_pool->sp_id);
755 rqstp = svc_prepare_thread(serv, chosen_pool, node);
756 if (IS_ERR(rqstp))
757 return PTR_ERR(rqstp);
758
759 task = kthread_create_on_node(serv->sv_threadfn, rqstp,
760 node, "%s", serv->sv_name);
761 if (IS_ERR(task)) {
762 svc_exit_thread(rqstp);
763 return PTR_ERR(task);
764 }
765
766 rqstp->rq_task = task;
767 if (serv->sv_nrpools > 1)
768 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
769
770 svc_sock_update_bufs(serv);
771 wake_up_process(task);
772 } while (nrservs > 0);
773
774 return 0;
775}
776
777/*
778 * Create or destroy enough new threads to make the number
779 * of threads the given number. If `pool' is non-NULL, applies
780 * only to threads in that pool, otherwise round-robins between
781 * all pools. Caller must ensure that mutual exclusion between this and
782 * server startup or shutdown.
783 */
784
785/* destroy old threads */
786static int
787svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
788{
789 struct task_struct *task;
790 unsigned int state = serv->sv_nrthreads-1;
791
792 /* destroy old threads */
793 do {
794 task = choose_victim(serv, pool, &state);
795 if (task == NULL)
796 break;
797 kthread_stop(task);
798 nrservs++;
799 } while (nrservs < 0);
800 return 0;
801}
802
803int
804svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
805{
806 if (pool == NULL) {
807 nrservs -= serv->sv_nrthreads;
808 } else {
809 spin_lock_bh(&pool->sp_lock);
810 nrservs -= pool->sp_nrthreads;
811 spin_unlock_bh(&pool->sp_lock);
812 }
813
814 if (nrservs > 0)
815 return svc_start_kthreads(serv, pool, nrservs);
816 if (nrservs < 0)
817 return svc_stop_kthreads(serv, pool, nrservs);
818 return 0;
819}
820EXPORT_SYMBOL_GPL(svc_set_num_threads);
821
822/**
823 * svc_rqst_replace_page - Replace one page in rq_pages[]
824 * @rqstp: svc_rqst with pages to replace
825 * @page: replacement page
826 *
827 * When replacing a page in rq_pages, batch the release of the
828 * replaced pages to avoid hammering the page allocator.
829 */
830void svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
831{
832 if (*rqstp->rq_next_page) {
833 if (!pagevec_space(&rqstp->rq_pvec))
834 __pagevec_release(&rqstp->rq_pvec);
835 pagevec_add(&rqstp->rq_pvec, *rqstp->rq_next_page);
836 }
837
838 get_page(page);
839 *(rqstp->rq_next_page++) = page;
840}
841EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
842
843/*
844 * Called from a server thread as it's exiting. Caller must hold the "service
845 * mutex" for the service.
846 */
847void
848svc_rqst_free(struct svc_rqst *rqstp)
849{
850 svc_release_buffer(rqstp);
851 if (rqstp->rq_scratch_page)
852 put_page(rqstp->rq_scratch_page);
853 kfree(rqstp->rq_resp);
854 kfree(rqstp->rq_argp);
855 kfree(rqstp->rq_auth_data);
856 kfree_rcu(rqstp, rq_rcu_head);
857}
858EXPORT_SYMBOL_GPL(svc_rqst_free);
859
860void
861svc_exit_thread(struct svc_rqst *rqstp)
862{
863 struct svc_serv *serv = rqstp->rq_server;
864 struct svc_pool *pool = rqstp->rq_pool;
865
866 spin_lock_bh(&pool->sp_lock);
867 pool->sp_nrthreads--;
868 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
869 list_del_rcu(&rqstp->rq_all);
870 spin_unlock_bh(&pool->sp_lock);
871
872 spin_lock_bh(&serv->sv_lock);
873 serv->sv_nrthreads -= 1;
874 spin_unlock_bh(&serv->sv_lock);
875 svc_sock_update_bufs(serv);
876
877 svc_rqst_free(rqstp);
878
879 svc_put(serv);
880}
881EXPORT_SYMBOL_GPL(svc_exit_thread);
882
883/*
884 * Register an "inet" protocol family netid with the local
885 * rpcbind daemon via an rpcbind v4 SET request.
886 *
887 * No netconfig infrastructure is available in the kernel, so
888 * we map IP_ protocol numbers to netids by hand.
889 *
890 * Returns zero on success; a negative errno value is returned
891 * if any error occurs.
892 */
893static int __svc_rpcb_register4(struct net *net, const u32 program,
894 const u32 version,
895 const unsigned short protocol,
896 const unsigned short port)
897{
898 const struct sockaddr_in sin = {
899 .sin_family = AF_INET,
900 .sin_addr.s_addr = htonl(INADDR_ANY),
901 .sin_port = htons(port),
902 };
903 const char *netid;
904 int error;
905
906 switch (protocol) {
907 case IPPROTO_UDP:
908 netid = RPCBIND_NETID_UDP;
909 break;
910 case IPPROTO_TCP:
911 netid = RPCBIND_NETID_TCP;
912 break;
913 default:
914 return -ENOPROTOOPT;
915 }
916
917 error = rpcb_v4_register(net, program, version,
918 (const struct sockaddr *)&sin, netid);
919
920 /*
921 * User space didn't support rpcbind v4, so retry this
922 * registration request with the legacy rpcbind v2 protocol.
923 */
924 if (error == -EPROTONOSUPPORT)
925 error = rpcb_register(net, program, version, protocol, port);
926
927 return error;
928}
929
930#if IS_ENABLED(CONFIG_IPV6)
931/*
932 * Register an "inet6" protocol family netid with the local
933 * rpcbind daemon via an rpcbind v4 SET request.
934 *
935 * No netconfig infrastructure is available in the kernel, so
936 * we map IP_ protocol numbers to netids by hand.
937 *
938 * Returns zero on success; a negative errno value is returned
939 * if any error occurs.
940 */
941static int __svc_rpcb_register6(struct net *net, const u32 program,
942 const u32 version,
943 const unsigned short protocol,
944 const unsigned short port)
945{
946 const struct sockaddr_in6 sin6 = {
947 .sin6_family = AF_INET6,
948 .sin6_addr = IN6ADDR_ANY_INIT,
949 .sin6_port = htons(port),
950 };
951 const char *netid;
952 int error;
953
954 switch (protocol) {
955 case IPPROTO_UDP:
956 netid = RPCBIND_NETID_UDP6;
957 break;
958 case IPPROTO_TCP:
959 netid = RPCBIND_NETID_TCP6;
960 break;
961 default:
962 return -ENOPROTOOPT;
963 }
964
965 error = rpcb_v4_register(net, program, version,
966 (const struct sockaddr *)&sin6, netid);
967
968 /*
969 * User space didn't support rpcbind version 4, so we won't
970 * use a PF_INET6 listener.
971 */
972 if (error == -EPROTONOSUPPORT)
973 error = -EAFNOSUPPORT;
974
975 return error;
976}
977#endif /* IS_ENABLED(CONFIG_IPV6) */
978
979/*
980 * Register a kernel RPC service via rpcbind version 4.
981 *
982 * Returns zero on success; a negative errno value is returned
983 * if any error occurs.
984 */
985static int __svc_register(struct net *net, const char *progname,
986 const u32 program, const u32 version,
987 const int family,
988 const unsigned short protocol,
989 const unsigned short port)
990{
991 int error = -EAFNOSUPPORT;
992
993 switch (family) {
994 case PF_INET:
995 error = __svc_rpcb_register4(net, program, version,
996 protocol, port);
997 break;
998#if IS_ENABLED(CONFIG_IPV6)
999 case PF_INET6:
1000 error = __svc_rpcb_register6(net, program, version,
1001 protocol, port);
1002#endif
1003 }
1004
1005 trace_svc_register(progname, version, protocol, port, family, error);
1006 return error;
1007}
1008
1009int svc_rpcbind_set_version(struct net *net,
1010 const struct svc_program *progp,
1011 u32 version, int family,
1012 unsigned short proto,
1013 unsigned short port)
1014{
1015 return __svc_register(net, progp->pg_name, progp->pg_prog,
1016 version, family, proto, port);
1017
1018}
1019EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1020
1021int svc_generic_rpcbind_set(struct net *net,
1022 const struct svc_program *progp,
1023 u32 version, int family,
1024 unsigned short proto,
1025 unsigned short port)
1026{
1027 const struct svc_version *vers = progp->pg_vers[version];
1028 int error;
1029
1030 if (vers == NULL)
1031 return 0;
1032
1033 if (vers->vs_hidden) {
1034 trace_svc_noregister(progp->pg_name, version, proto,
1035 port, family, 0);
1036 return 0;
1037 }
1038
1039 /*
1040 * Don't register a UDP port if we need congestion
1041 * control.
1042 */
1043 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1044 return 0;
1045
1046 error = svc_rpcbind_set_version(net, progp, version,
1047 family, proto, port);
1048
1049 return (vers->vs_rpcb_optnl) ? 0 : error;
1050}
1051EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1052
1053/**
1054 * svc_register - register an RPC service with the local portmapper
1055 * @serv: svc_serv struct for the service to register
1056 * @net: net namespace for the service to register
1057 * @family: protocol family of service's listener socket
1058 * @proto: transport protocol number to advertise
1059 * @port: port to advertise
1060 *
1061 * Service is registered for any address in the passed-in protocol family
1062 */
1063int svc_register(const struct svc_serv *serv, struct net *net,
1064 const int family, const unsigned short proto,
1065 const unsigned short port)
1066{
1067 struct svc_program *progp;
1068 unsigned int i;
1069 int error = 0;
1070
1071 WARN_ON_ONCE(proto == 0 && port == 0);
1072 if (proto == 0 && port == 0)
1073 return -EINVAL;
1074
1075 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1076 for (i = 0; i < progp->pg_nvers; i++) {
1077
1078 error = progp->pg_rpcbind_set(net, progp, i,
1079 family, proto, port);
1080 if (error < 0) {
1081 printk(KERN_WARNING "svc: failed to register "
1082 "%sv%u RPC service (errno %d).\n",
1083 progp->pg_name, i, -error);
1084 break;
1085 }
1086 }
1087 }
1088
1089 return error;
1090}
1091
1092/*
1093 * If user space is running rpcbind, it should take the v4 UNSET
1094 * and clear everything for this [program, version]. If user space
1095 * is running portmap, it will reject the v4 UNSET, but won't have
1096 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1097 * in this case to clear all existing entries for [program, version].
1098 */
1099static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1100 const char *progname)
1101{
1102 int error;
1103
1104 error = rpcb_v4_register(net, program, version, NULL, "");
1105
1106 /*
1107 * User space didn't support rpcbind v4, so retry this
1108 * request with the legacy rpcbind v2 protocol.
1109 */
1110 if (error == -EPROTONOSUPPORT)
1111 error = rpcb_register(net, program, version, 0, 0);
1112
1113 trace_svc_unregister(progname, version, error);
1114}
1115
1116/*
1117 * All netids, bind addresses and ports registered for [program, version]
1118 * are removed from the local rpcbind database (if the service is not
1119 * hidden) to make way for a new instance of the service.
1120 *
1121 * The result of unregistration is reported via dprintk for those who want
1122 * verification of the result, but is otherwise not important.
1123 */
1124static void svc_unregister(const struct svc_serv *serv, struct net *net)
1125{
1126 struct svc_program *progp;
1127 unsigned long flags;
1128 unsigned int i;
1129
1130 clear_thread_flag(TIF_SIGPENDING);
1131
1132 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1133 for (i = 0; i < progp->pg_nvers; i++) {
1134 if (progp->pg_vers[i] == NULL)
1135 continue;
1136 if (progp->pg_vers[i]->vs_hidden)
1137 continue;
1138 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1139 }
1140 }
1141
1142 spin_lock_irqsave(¤t->sighand->siglock, flags);
1143 recalc_sigpending();
1144 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1145}
1146
1147/*
1148 * dprintk the given error with the address of the client that caused it.
1149 */
1150#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1151static __printf(2, 3)
1152void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1153{
1154 struct va_format vaf;
1155 va_list args;
1156 char buf[RPC_MAX_ADDRBUFLEN];
1157
1158 va_start(args, fmt);
1159
1160 vaf.fmt = fmt;
1161 vaf.va = &args;
1162
1163 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1164
1165 va_end(args);
1166}
1167#else
1168static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1169#endif
1170
1171__be32
1172svc_generic_init_request(struct svc_rqst *rqstp,
1173 const struct svc_program *progp,
1174 struct svc_process_info *ret)
1175{
1176 const struct svc_version *versp = NULL; /* compiler food */
1177 const struct svc_procedure *procp = NULL;
1178
1179 if (rqstp->rq_vers >= progp->pg_nvers )
1180 goto err_bad_vers;
1181 versp = progp->pg_vers[rqstp->rq_vers];
1182 if (!versp)
1183 goto err_bad_vers;
1184
1185 /*
1186 * Some protocol versions (namely NFSv4) require some form of
1187 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1188 * In other words, UDP is not allowed. We mark those when setting
1189 * up the svc_xprt, and verify that here.
1190 *
1191 * The spec is not very clear about what error should be returned
1192 * when someone tries to access a server that is listening on UDP
1193 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1194 * fit.
1195 */
1196 if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1197 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1198 goto err_bad_vers;
1199
1200 if (rqstp->rq_proc >= versp->vs_nproc)
1201 goto err_bad_proc;
1202 rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1203 if (!procp)
1204 goto err_bad_proc;
1205
1206 /* Initialize storage for argp and resp */
1207 memset(rqstp->rq_argp, 0, procp->pc_argzero);
1208 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1209
1210 /* Bump per-procedure stats counter */
1211 versp->vs_count[rqstp->rq_proc]++;
1212
1213 ret->dispatch = versp->vs_dispatch;
1214 return rpc_success;
1215err_bad_vers:
1216 ret->mismatch.lovers = progp->pg_lovers;
1217 ret->mismatch.hivers = progp->pg_hivers;
1218 return rpc_prog_mismatch;
1219err_bad_proc:
1220 return rpc_proc_unavail;
1221}
1222EXPORT_SYMBOL_GPL(svc_generic_init_request);
1223
1224/*
1225 * Common routine for processing the RPC request.
1226 */
1227static int
1228svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1229{
1230 struct svc_program *progp;
1231 const struct svc_procedure *procp = NULL;
1232 struct svc_serv *serv = rqstp->rq_server;
1233 struct svc_process_info process;
1234 __be32 *statp;
1235 u32 prog, vers;
1236 __be32 rpc_stat;
1237 int auth_res, rc;
1238 __be32 *reply_statp;
1239
1240 rpc_stat = rpc_success;
1241
1242 if (argv->iov_len < 6*4)
1243 goto err_short_len;
1244
1245 /* Will be turned off by GSS integrity and privacy services */
1246 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1247 /* Will be turned off only when NFSv4 Sessions are used */
1248 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1249 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1250
1251 svc_putu32(resv, rqstp->rq_xid);
1252
1253 vers = svc_getnl(argv);
1254
1255 /* First words of reply: */
1256 svc_putnl(resv, 1); /* REPLY */
1257
1258 if (vers != 2) /* RPC version number */
1259 goto err_bad_rpc;
1260
1261 /* Save position in case we later decide to reject: */
1262 reply_statp = resv->iov_base + resv->iov_len;
1263
1264 svc_putnl(resv, 0); /* ACCEPT */
1265
1266 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1267 rqstp->rq_vers = svc_getnl(argv); /* version number */
1268 rqstp->rq_proc = svc_getnl(argv); /* procedure number */
1269
1270 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1271 if (prog == progp->pg_prog)
1272 break;
1273
1274 /*
1275 * Decode auth data, and add verifier to reply buffer.
1276 * We do this before anything else in order to get a decent
1277 * auth verifier.
1278 */
1279 auth_res = svc_authenticate(rqstp);
1280 /* Also give the program a chance to reject this call: */
1281 if (auth_res == SVC_OK && progp)
1282 auth_res = progp->pg_authenticate(rqstp);
1283 trace_svc_authenticate(rqstp, auth_res);
1284 switch (auth_res) {
1285 case SVC_OK:
1286 break;
1287 case SVC_GARBAGE:
1288 goto err_garbage;
1289 case SVC_SYSERR:
1290 rpc_stat = rpc_system_err;
1291 goto err_bad;
1292 case SVC_DENIED:
1293 goto err_bad_auth;
1294 case SVC_CLOSE:
1295 goto close;
1296 case SVC_DROP:
1297 goto dropit;
1298 case SVC_COMPLETE:
1299 goto sendit;
1300 }
1301
1302 if (progp == NULL)
1303 goto err_bad_prog;
1304
1305 rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1306 switch (rpc_stat) {
1307 case rpc_success:
1308 break;
1309 case rpc_prog_unavail:
1310 goto err_bad_prog;
1311 case rpc_prog_mismatch:
1312 goto err_bad_vers;
1313 case rpc_proc_unavail:
1314 goto err_bad_proc;
1315 }
1316
1317 procp = rqstp->rq_procinfo;
1318 /* Should this check go into the dispatcher? */
1319 if (!procp || !procp->pc_func)
1320 goto err_bad_proc;
1321
1322 /* Syntactic check complete */
1323 serv->sv_stats->rpccnt++;
1324 trace_svc_process(rqstp, progp->pg_name);
1325
1326 /* Build the reply header. */
1327 statp = resv->iov_base +resv->iov_len;
1328 svc_putnl(resv, RPC_SUCCESS);
1329
1330 /* un-reserve some of the out-queue now that we have a
1331 * better idea of reply size
1332 */
1333 if (procp->pc_xdrressize)
1334 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1335
1336 /* Call the function that processes the request. */
1337 rc = process.dispatch(rqstp, statp);
1338 if (procp->pc_release)
1339 procp->pc_release(rqstp);
1340 if (!rc)
1341 goto dropit;
1342 if (rqstp->rq_auth_stat != rpc_auth_ok)
1343 goto err_bad_auth;
1344
1345 /* Check RPC status result */
1346 if (*statp != rpc_success)
1347 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1348
1349 if (procp->pc_encode == NULL)
1350 goto dropit;
1351
1352 sendit:
1353 if (svc_authorise(rqstp))
1354 goto close_xprt;
1355 return 1; /* Caller can now send it */
1356
1357 dropit:
1358 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1359 dprintk("svc: svc_process dropit\n");
1360 return 0;
1361
1362 close:
1363 svc_authorise(rqstp);
1364close_xprt:
1365 if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1366 svc_xprt_close(rqstp->rq_xprt);
1367 dprintk("svc: svc_process close\n");
1368 return 0;
1369
1370err_short_len:
1371 svc_printk(rqstp, "short len %zd, dropping request\n",
1372 argv->iov_len);
1373 goto close_xprt;
1374
1375err_bad_rpc:
1376 serv->sv_stats->rpcbadfmt++;
1377 svc_putnl(resv, 1); /* REJECT */
1378 svc_putnl(resv, 0); /* RPC_MISMATCH */
1379 svc_putnl(resv, 2); /* Only RPCv2 supported */
1380 svc_putnl(resv, 2);
1381 goto sendit;
1382
1383err_bad_auth:
1384 dprintk("svc: authentication failed (%d)\n",
1385 be32_to_cpu(rqstp->rq_auth_stat));
1386 serv->sv_stats->rpcbadauth++;
1387 /* Restore write pointer to location of accept status: */
1388 xdr_ressize_check(rqstp, reply_statp);
1389 svc_putnl(resv, 1); /* REJECT */
1390 svc_putnl(resv, 1); /* AUTH_ERROR */
1391 svc_putu32(resv, rqstp->rq_auth_stat); /* status */
1392 goto sendit;
1393
1394err_bad_prog:
1395 dprintk("svc: unknown program %d\n", prog);
1396 serv->sv_stats->rpcbadfmt++;
1397 svc_putnl(resv, RPC_PROG_UNAVAIL);
1398 goto sendit;
1399
1400err_bad_vers:
1401 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1402 rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1403
1404 serv->sv_stats->rpcbadfmt++;
1405 svc_putnl(resv, RPC_PROG_MISMATCH);
1406 svc_putnl(resv, process.mismatch.lovers);
1407 svc_putnl(resv, process.mismatch.hivers);
1408 goto sendit;
1409
1410err_bad_proc:
1411 svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1412
1413 serv->sv_stats->rpcbadfmt++;
1414 svc_putnl(resv, RPC_PROC_UNAVAIL);
1415 goto sendit;
1416
1417err_garbage:
1418 svc_printk(rqstp, "failed to decode args\n");
1419
1420 rpc_stat = rpc_garbage_args;
1421err_bad:
1422 serv->sv_stats->rpcbadfmt++;
1423 svc_putnl(resv, ntohl(rpc_stat));
1424 goto sendit;
1425}
1426
1427/*
1428 * Process the RPC request.
1429 */
1430int
1431svc_process(struct svc_rqst *rqstp)
1432{
1433 struct kvec *argv = &rqstp->rq_arg.head[0];
1434 struct kvec *resv = &rqstp->rq_res.head[0];
1435 __be32 dir;
1436
1437#if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1438 if (!fail_sunrpc.ignore_server_disconnect &&
1439 should_fail(&fail_sunrpc.attr, 1))
1440 svc_xprt_deferred_close(rqstp->rq_xprt);
1441#endif
1442
1443 /*
1444 * Setup response xdr_buf.
1445 * Initially it has just one page
1446 */
1447 rqstp->rq_next_page = &rqstp->rq_respages[1];
1448 resv->iov_base = page_address(rqstp->rq_respages[0]);
1449 resv->iov_len = 0;
1450 rqstp->rq_res.pages = rqstp->rq_next_page;
1451 rqstp->rq_res.len = 0;
1452 rqstp->rq_res.page_base = 0;
1453 rqstp->rq_res.page_len = 0;
1454 rqstp->rq_res.buflen = PAGE_SIZE;
1455 rqstp->rq_res.tail[0].iov_base = NULL;
1456 rqstp->rq_res.tail[0].iov_len = 0;
1457
1458 dir = svc_getu32(argv);
1459 if (dir != rpc_call)
1460 goto out_baddir;
1461 if (!svc_process_common(rqstp, argv, resv))
1462 goto out_drop;
1463 return svc_send(rqstp);
1464
1465out_baddir:
1466 svc_printk(rqstp, "bad direction 0x%08x, dropping request\n",
1467 be32_to_cpu(dir));
1468 rqstp->rq_server->sv_stats->rpcbadfmt++;
1469out_drop:
1470 svc_drop(rqstp);
1471 return 0;
1472}
1473EXPORT_SYMBOL_GPL(svc_process);
1474
1475#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1476/*
1477 * Process a backchannel RPC request that arrived over an existing
1478 * outbound connection
1479 */
1480int
1481bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1482 struct svc_rqst *rqstp)
1483{
1484 struct kvec *argv = &rqstp->rq_arg.head[0];
1485 struct kvec *resv = &rqstp->rq_res.head[0];
1486 struct rpc_task *task;
1487 int proc_error;
1488 int error;
1489
1490 dprintk("svc: %s(%p)\n", __func__, req);
1491
1492 /* Build the svc_rqst used by the common processing routine */
1493 rqstp->rq_xid = req->rq_xid;
1494 rqstp->rq_prot = req->rq_xprt->prot;
1495 rqstp->rq_server = serv;
1496 rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1497
1498 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1499 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1500 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1501 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1502
1503 /* Adjust the argument buffer length */
1504 rqstp->rq_arg.len = req->rq_private_buf.len;
1505 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1506 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1507 rqstp->rq_arg.page_len = 0;
1508 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1509 rqstp->rq_arg.page_len)
1510 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1511 rqstp->rq_arg.head[0].iov_len;
1512 else
1513 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1514 rqstp->rq_arg.page_len;
1515
1516 /* reset result send buffer "put" position */
1517 resv->iov_len = 0;
1518
1519 /*
1520 * Skip the next two words because they've already been
1521 * processed in the transport
1522 */
1523 svc_getu32(argv); /* XID */
1524 svc_getnl(argv); /* CALLDIR */
1525
1526 /* Parse and execute the bc call */
1527 proc_error = svc_process_common(rqstp, argv, resv);
1528
1529 atomic_dec(&req->rq_xprt->bc_slot_count);
1530 if (!proc_error) {
1531 /* Processing error: drop the request */
1532 xprt_free_bc_request(req);
1533 error = -EINVAL;
1534 goto out;
1535 }
1536 /* Finally, send the reply synchronously */
1537 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1538 task = rpc_run_bc_task(req);
1539 if (IS_ERR(task)) {
1540 error = PTR_ERR(task);
1541 goto out;
1542 }
1543
1544 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1545 error = task->tk_status;
1546 rpc_put_task(task);
1547
1548out:
1549 dprintk("svc: %s(), error=%d\n", __func__, error);
1550 return error;
1551}
1552EXPORT_SYMBOL_GPL(bc_svc_process);
1553#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1554
1555/**
1556 * svc_max_payload - Return transport-specific limit on the RPC payload
1557 * @rqstp: RPC transaction context
1558 *
1559 * Returns the maximum number of payload bytes the current transport
1560 * allows.
1561 */
1562u32 svc_max_payload(const struct svc_rqst *rqstp)
1563{
1564 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1565
1566 if (rqstp->rq_server->sv_max_payload < max)
1567 max = rqstp->rq_server->sv_max_payload;
1568 return max;
1569}
1570EXPORT_SYMBOL_GPL(svc_max_payload);
1571
1572/**
1573 * svc_proc_name - Return RPC procedure name in string form
1574 * @rqstp: svc_rqst to operate on
1575 *
1576 * Return value:
1577 * Pointer to a NUL-terminated string
1578 */
1579const char *svc_proc_name(const struct svc_rqst *rqstp)
1580{
1581 if (rqstp && rqstp->rq_procinfo)
1582 return rqstp->rq_procinfo->pc_name;
1583 return "unknown";
1584}
1585
1586
1587/**
1588 * svc_encode_result_payload - mark a range of bytes as a result payload
1589 * @rqstp: svc_rqst to operate on
1590 * @offset: payload's byte offset in rqstp->rq_res
1591 * @length: size of payload, in bytes
1592 *
1593 * Returns zero on success, or a negative errno if a permanent
1594 * error occurred.
1595 */
1596int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1597 unsigned int length)
1598{
1599 return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1600 length);
1601}
1602EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1603
1604/**
1605 * svc_fill_write_vector - Construct data argument for VFS write call
1606 * @rqstp: svc_rqst to operate on
1607 * @payload: xdr_buf containing only the write data payload
1608 *
1609 * Fills in rqstp::rq_vec, and returns the number of elements.
1610 */
1611unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1612 struct xdr_buf *payload)
1613{
1614 struct page **pages = payload->pages;
1615 struct kvec *first = payload->head;
1616 struct kvec *vec = rqstp->rq_vec;
1617 size_t total = payload->len;
1618 unsigned int i;
1619
1620 /* Some types of transport can present the write payload
1621 * entirely in rq_arg.pages. In this case, @first is empty.
1622 */
1623 i = 0;
1624 if (first->iov_len) {
1625 vec[i].iov_base = first->iov_base;
1626 vec[i].iov_len = min_t(size_t, total, first->iov_len);
1627 total -= vec[i].iov_len;
1628 ++i;
1629 }
1630
1631 while (total) {
1632 vec[i].iov_base = page_address(*pages);
1633 vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1634 total -= vec[i].iov_len;
1635 ++i;
1636 ++pages;
1637 }
1638
1639 WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1640 return i;
1641}
1642EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1643
1644/**
1645 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1646 * @rqstp: svc_rqst to operate on
1647 * @first: buffer containing first section of pathname
1648 * @p: buffer containing remaining section of pathname
1649 * @total: total length of the pathname argument
1650 *
1651 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1652 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1653 * the returned string.
1654 */
1655char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1656 void *p, size_t total)
1657{
1658 size_t len, remaining;
1659 char *result, *dst;
1660
1661 result = kmalloc(total + 1, GFP_KERNEL);
1662 if (!result)
1663 return ERR_PTR(-ESERVERFAULT);
1664
1665 dst = result;
1666 remaining = total;
1667
1668 len = min_t(size_t, total, first->iov_len);
1669 if (len) {
1670 memcpy(dst, first->iov_base, len);
1671 dst += len;
1672 remaining -= len;
1673 }
1674
1675 if (remaining) {
1676 len = min_t(size_t, remaining, PAGE_SIZE);
1677 memcpy(dst, p, len);
1678 dst += len;
1679 }
1680
1681 *dst = '\0';
1682
1683 /* Sanity check: Linux doesn't allow the pathname argument to
1684 * contain a NUL byte.
1685 */
1686 if (strlen(result) != total) {
1687 kfree(result);
1688 return ERR_PTR(-EINVAL);
1689 }
1690 return result;
1691}
1692EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svc.c
4 *
5 * High-level RPC service routines
6 *
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 *
9 * Multiple threads pools and NUMAisation
10 * Copyright (c) 2006 Silicon Graphics, Inc.
11 * by Greg Banks <gnb@melbourne.sgi.com>
12 */
13
14#include <linux/linkage.h>
15#include <linux/sched/signal.h>
16#include <linux/errno.h>
17#include <linux/net.h>
18#include <linux/in.h>
19#include <linux/mm.h>
20#include <linux/interrupt.h>
21#include <linux/module.h>
22#include <linux/kthread.h>
23#include <linux/slab.h>
24
25#include <linux/sunrpc/types.h>
26#include <linux/sunrpc/xdr.h>
27#include <linux/sunrpc/stats.h>
28#include <linux/sunrpc/svcsock.h>
29#include <linux/sunrpc/clnt.h>
30#include <linux/sunrpc/bc_xprt.h>
31
32#include <trace/events/sunrpc.h>
33
34#define RPCDBG_FACILITY RPCDBG_SVCDSP
35
36static void svc_unregister(const struct svc_serv *serv, struct net *net);
37
38#define svc_serv_is_pooled(serv) ((serv)->sv_ops->svo_function)
39
40#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
41
42/*
43 * Structure for mapping cpus to pools and vice versa.
44 * Setup once during sunrpc initialisation.
45 */
46struct svc_pool_map svc_pool_map = {
47 .mode = SVC_POOL_DEFAULT
48};
49EXPORT_SYMBOL_GPL(svc_pool_map);
50
51static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
52
53static int
54param_set_pool_mode(const char *val, const struct kernel_param *kp)
55{
56 int *ip = (int *)kp->arg;
57 struct svc_pool_map *m = &svc_pool_map;
58 int err;
59
60 mutex_lock(&svc_pool_map_mutex);
61
62 err = -EBUSY;
63 if (m->count)
64 goto out;
65
66 err = 0;
67 if (!strncmp(val, "auto", 4))
68 *ip = SVC_POOL_AUTO;
69 else if (!strncmp(val, "global", 6))
70 *ip = SVC_POOL_GLOBAL;
71 else if (!strncmp(val, "percpu", 6))
72 *ip = SVC_POOL_PERCPU;
73 else if (!strncmp(val, "pernode", 7))
74 *ip = SVC_POOL_PERNODE;
75 else
76 err = -EINVAL;
77
78out:
79 mutex_unlock(&svc_pool_map_mutex);
80 return err;
81}
82
83static int
84param_get_pool_mode(char *buf, const struct kernel_param *kp)
85{
86 int *ip = (int *)kp->arg;
87
88 switch (*ip)
89 {
90 case SVC_POOL_AUTO:
91 return strlcpy(buf, "auto", 20);
92 case SVC_POOL_GLOBAL:
93 return strlcpy(buf, "global", 20);
94 case SVC_POOL_PERCPU:
95 return strlcpy(buf, "percpu", 20);
96 case SVC_POOL_PERNODE:
97 return strlcpy(buf, "pernode", 20);
98 default:
99 return sprintf(buf, "%d", *ip);
100 }
101}
102
103module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
104 &svc_pool_map.mode, 0644);
105
106/*
107 * Detect best pool mapping mode heuristically,
108 * according to the machine's topology.
109 */
110static int
111svc_pool_map_choose_mode(void)
112{
113 unsigned int node;
114
115 if (nr_online_nodes > 1) {
116 /*
117 * Actually have multiple NUMA nodes,
118 * so split pools on NUMA node boundaries
119 */
120 return SVC_POOL_PERNODE;
121 }
122
123 node = first_online_node;
124 if (nr_cpus_node(node) > 2) {
125 /*
126 * Non-trivial SMP, or CONFIG_NUMA on
127 * non-NUMA hardware, e.g. with a generic
128 * x86_64 kernel on Xeons. In this case we
129 * want to divide the pools on cpu boundaries.
130 */
131 return SVC_POOL_PERCPU;
132 }
133
134 /* default: one global pool */
135 return SVC_POOL_GLOBAL;
136}
137
138/*
139 * Allocate the to_pool[] and pool_to[] arrays.
140 * Returns 0 on success or an errno.
141 */
142static int
143svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
144{
145 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
146 if (!m->to_pool)
147 goto fail;
148 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
149 if (!m->pool_to)
150 goto fail_free;
151
152 return 0;
153
154fail_free:
155 kfree(m->to_pool);
156 m->to_pool = NULL;
157fail:
158 return -ENOMEM;
159}
160
161/*
162 * Initialise the pool map for SVC_POOL_PERCPU mode.
163 * Returns number of pools or <0 on error.
164 */
165static int
166svc_pool_map_init_percpu(struct svc_pool_map *m)
167{
168 unsigned int maxpools = nr_cpu_ids;
169 unsigned int pidx = 0;
170 unsigned int cpu;
171 int err;
172
173 err = svc_pool_map_alloc_arrays(m, maxpools);
174 if (err)
175 return err;
176
177 for_each_online_cpu(cpu) {
178 BUG_ON(pidx >= maxpools);
179 m->to_pool[cpu] = pidx;
180 m->pool_to[pidx] = cpu;
181 pidx++;
182 }
183 /* cpus brought online later all get mapped to pool0, sorry */
184
185 return pidx;
186};
187
188
189/*
190 * Initialise the pool map for SVC_POOL_PERNODE mode.
191 * Returns number of pools or <0 on error.
192 */
193static int
194svc_pool_map_init_pernode(struct svc_pool_map *m)
195{
196 unsigned int maxpools = nr_node_ids;
197 unsigned int pidx = 0;
198 unsigned int node;
199 int err;
200
201 err = svc_pool_map_alloc_arrays(m, maxpools);
202 if (err)
203 return err;
204
205 for_each_node_with_cpus(node) {
206 /* some architectures (e.g. SN2) have cpuless nodes */
207 BUG_ON(pidx > maxpools);
208 m->to_pool[node] = pidx;
209 m->pool_to[pidx] = node;
210 pidx++;
211 }
212 /* nodes brought online later all get mapped to pool0, sorry */
213
214 return pidx;
215}
216
217
218/*
219 * Add a reference to the global map of cpus to pools (and
220 * vice versa). Initialise the map if we're the first user.
221 * Returns the number of pools.
222 */
223unsigned int
224svc_pool_map_get(void)
225{
226 struct svc_pool_map *m = &svc_pool_map;
227 int npools = -1;
228
229 mutex_lock(&svc_pool_map_mutex);
230
231 if (m->count++) {
232 mutex_unlock(&svc_pool_map_mutex);
233 return m->npools;
234 }
235
236 if (m->mode == SVC_POOL_AUTO)
237 m->mode = svc_pool_map_choose_mode();
238
239 switch (m->mode) {
240 case SVC_POOL_PERCPU:
241 npools = svc_pool_map_init_percpu(m);
242 break;
243 case SVC_POOL_PERNODE:
244 npools = svc_pool_map_init_pernode(m);
245 break;
246 }
247
248 if (npools < 0) {
249 /* default, or memory allocation failure */
250 npools = 1;
251 m->mode = SVC_POOL_GLOBAL;
252 }
253 m->npools = npools;
254
255 mutex_unlock(&svc_pool_map_mutex);
256 return m->npools;
257}
258EXPORT_SYMBOL_GPL(svc_pool_map_get);
259
260/*
261 * Drop a reference to the global map of cpus to pools.
262 * When the last reference is dropped, the map data is
263 * freed; this allows the sysadmin to change the pool
264 * mode using the pool_mode module option without
265 * rebooting or re-loading sunrpc.ko.
266 */
267void
268svc_pool_map_put(void)
269{
270 struct svc_pool_map *m = &svc_pool_map;
271
272 mutex_lock(&svc_pool_map_mutex);
273
274 if (!--m->count) {
275 kfree(m->to_pool);
276 m->to_pool = NULL;
277 kfree(m->pool_to);
278 m->pool_to = NULL;
279 m->npools = 0;
280 }
281
282 mutex_unlock(&svc_pool_map_mutex);
283}
284EXPORT_SYMBOL_GPL(svc_pool_map_put);
285
286static int svc_pool_map_get_node(unsigned int pidx)
287{
288 const struct svc_pool_map *m = &svc_pool_map;
289
290 if (m->count) {
291 if (m->mode == SVC_POOL_PERCPU)
292 return cpu_to_node(m->pool_to[pidx]);
293 if (m->mode == SVC_POOL_PERNODE)
294 return m->pool_to[pidx];
295 }
296 return NUMA_NO_NODE;
297}
298/*
299 * Set the given thread's cpus_allowed mask so that it
300 * will only run on cpus in the given pool.
301 */
302static inline void
303svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
304{
305 struct svc_pool_map *m = &svc_pool_map;
306 unsigned int node = m->pool_to[pidx];
307
308 /*
309 * The caller checks for sv_nrpools > 1, which
310 * implies that we've been initialized.
311 */
312 WARN_ON_ONCE(m->count == 0);
313 if (m->count == 0)
314 return;
315
316 switch (m->mode) {
317 case SVC_POOL_PERCPU:
318 {
319 set_cpus_allowed_ptr(task, cpumask_of(node));
320 break;
321 }
322 case SVC_POOL_PERNODE:
323 {
324 set_cpus_allowed_ptr(task, cpumask_of_node(node));
325 break;
326 }
327 }
328}
329
330/*
331 * Use the mapping mode to choose a pool for a given CPU.
332 * Used when enqueueing an incoming RPC. Always returns
333 * a non-NULL pool pointer.
334 */
335struct svc_pool *
336svc_pool_for_cpu(struct svc_serv *serv, int cpu)
337{
338 struct svc_pool_map *m = &svc_pool_map;
339 unsigned int pidx = 0;
340
341 /*
342 * An uninitialised map happens in a pure client when
343 * lockd is brought up, so silently treat it the
344 * same as SVC_POOL_GLOBAL.
345 */
346 if (svc_serv_is_pooled(serv)) {
347 switch (m->mode) {
348 case SVC_POOL_PERCPU:
349 pidx = m->to_pool[cpu];
350 break;
351 case SVC_POOL_PERNODE:
352 pidx = m->to_pool[cpu_to_node(cpu)];
353 break;
354 }
355 }
356 return &serv->sv_pools[pidx % serv->sv_nrpools];
357}
358
359int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
360{
361 int err;
362
363 err = rpcb_create_local(net);
364 if (err)
365 return err;
366
367 /* Remove any stale portmap registrations */
368 svc_unregister(serv, net);
369 return 0;
370}
371EXPORT_SYMBOL_GPL(svc_rpcb_setup);
372
373void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
374{
375 svc_unregister(serv, net);
376 rpcb_put_local(net);
377}
378EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
379
380static int svc_uses_rpcbind(struct svc_serv *serv)
381{
382 struct svc_program *progp;
383 unsigned int i;
384
385 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
386 for (i = 0; i < progp->pg_nvers; i++) {
387 if (progp->pg_vers[i] == NULL)
388 continue;
389 if (!progp->pg_vers[i]->vs_hidden)
390 return 1;
391 }
392 }
393
394 return 0;
395}
396
397int svc_bind(struct svc_serv *serv, struct net *net)
398{
399 if (!svc_uses_rpcbind(serv))
400 return 0;
401 return svc_rpcb_setup(serv, net);
402}
403EXPORT_SYMBOL_GPL(svc_bind);
404
405#if defined(CONFIG_SUNRPC_BACKCHANNEL)
406static void
407__svc_init_bc(struct svc_serv *serv)
408{
409 INIT_LIST_HEAD(&serv->sv_cb_list);
410 spin_lock_init(&serv->sv_cb_lock);
411 init_waitqueue_head(&serv->sv_cb_waitq);
412}
413#else
414static void
415__svc_init_bc(struct svc_serv *serv)
416{
417}
418#endif
419
420/*
421 * Create an RPC service
422 */
423static struct svc_serv *
424__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
425 const struct svc_serv_ops *ops)
426{
427 struct svc_serv *serv;
428 unsigned int vers;
429 unsigned int xdrsize;
430 unsigned int i;
431
432 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
433 return NULL;
434 serv->sv_name = prog->pg_name;
435 serv->sv_program = prog;
436 serv->sv_nrthreads = 1;
437 serv->sv_stats = prog->pg_stats;
438 if (bufsize > RPCSVC_MAXPAYLOAD)
439 bufsize = RPCSVC_MAXPAYLOAD;
440 serv->sv_max_payload = bufsize? bufsize : 4096;
441 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
442 serv->sv_ops = ops;
443 xdrsize = 0;
444 while (prog) {
445 prog->pg_lovers = prog->pg_nvers-1;
446 for (vers=0; vers<prog->pg_nvers ; vers++)
447 if (prog->pg_vers[vers]) {
448 prog->pg_hivers = vers;
449 if (prog->pg_lovers > vers)
450 prog->pg_lovers = vers;
451 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
452 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
453 }
454 prog = prog->pg_next;
455 }
456 serv->sv_xdrsize = xdrsize;
457 INIT_LIST_HEAD(&serv->sv_tempsocks);
458 INIT_LIST_HEAD(&serv->sv_permsocks);
459 timer_setup(&serv->sv_temptimer, NULL, 0);
460 spin_lock_init(&serv->sv_lock);
461
462 __svc_init_bc(serv);
463
464 serv->sv_nrpools = npools;
465 serv->sv_pools =
466 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
467 GFP_KERNEL);
468 if (!serv->sv_pools) {
469 kfree(serv);
470 return NULL;
471 }
472
473 for (i = 0; i < serv->sv_nrpools; i++) {
474 struct svc_pool *pool = &serv->sv_pools[i];
475
476 dprintk("svc: initialising pool %u for %s\n",
477 i, serv->sv_name);
478
479 pool->sp_id = i;
480 INIT_LIST_HEAD(&pool->sp_sockets);
481 INIT_LIST_HEAD(&pool->sp_all_threads);
482 spin_lock_init(&pool->sp_lock);
483 }
484
485 return serv;
486}
487
488struct svc_serv *
489svc_create(struct svc_program *prog, unsigned int bufsize,
490 const struct svc_serv_ops *ops)
491{
492 return __svc_create(prog, bufsize, /*npools*/1, ops);
493}
494EXPORT_SYMBOL_GPL(svc_create);
495
496struct svc_serv *
497svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
498 const struct svc_serv_ops *ops)
499{
500 struct svc_serv *serv;
501 unsigned int npools = svc_pool_map_get();
502
503 serv = __svc_create(prog, bufsize, npools, ops);
504 if (!serv)
505 goto out_err;
506 return serv;
507out_err:
508 svc_pool_map_put();
509 return NULL;
510}
511EXPORT_SYMBOL_GPL(svc_create_pooled);
512
513void svc_shutdown_net(struct svc_serv *serv, struct net *net)
514{
515 svc_close_net(serv, net);
516
517 if (serv->sv_ops->svo_shutdown)
518 serv->sv_ops->svo_shutdown(serv, net);
519}
520EXPORT_SYMBOL_GPL(svc_shutdown_net);
521
522/*
523 * Destroy an RPC service. Should be called with appropriate locking to
524 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
525 */
526void
527svc_destroy(struct svc_serv *serv)
528{
529 dprintk("svc: svc_destroy(%s, %d)\n",
530 serv->sv_program->pg_name,
531 serv->sv_nrthreads);
532
533 if (serv->sv_nrthreads) {
534 if (--(serv->sv_nrthreads) != 0) {
535 svc_sock_update_bufs(serv);
536 return;
537 }
538 } else
539 printk("svc_destroy: no threads for serv=%p!\n", serv);
540
541 del_timer_sync(&serv->sv_temptimer);
542
543 /*
544 * The last user is gone and thus all sockets have to be destroyed to
545 * the point. Check this.
546 */
547 BUG_ON(!list_empty(&serv->sv_permsocks));
548 BUG_ON(!list_empty(&serv->sv_tempsocks));
549
550 cache_clean_deferred(serv);
551
552 if (svc_serv_is_pooled(serv))
553 svc_pool_map_put();
554
555 kfree(serv->sv_pools);
556 kfree(serv);
557}
558EXPORT_SYMBOL_GPL(svc_destroy);
559
560/*
561 * Allocate an RPC server's buffer space.
562 * We allocate pages and place them in rq_argpages.
563 */
564static int
565svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
566{
567 unsigned int pages, arghi;
568
569 /* bc_xprt uses fore channel allocated buffers */
570 if (svc_is_backchannel(rqstp))
571 return 1;
572
573 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
574 * We assume one is at most one page
575 */
576 arghi = 0;
577 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
578 if (pages > RPCSVC_MAXPAGES)
579 pages = RPCSVC_MAXPAGES;
580 while (pages) {
581 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
582 if (!p)
583 break;
584 rqstp->rq_pages[arghi++] = p;
585 pages--;
586 }
587 return pages == 0;
588}
589
590/*
591 * Release an RPC server buffer
592 */
593static void
594svc_release_buffer(struct svc_rqst *rqstp)
595{
596 unsigned int i;
597
598 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
599 if (rqstp->rq_pages[i])
600 put_page(rqstp->rq_pages[i]);
601}
602
603struct svc_rqst *
604svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
605{
606 struct svc_rqst *rqstp;
607
608 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
609 if (!rqstp)
610 return rqstp;
611
612 __set_bit(RQ_BUSY, &rqstp->rq_flags);
613 spin_lock_init(&rqstp->rq_lock);
614 rqstp->rq_server = serv;
615 rqstp->rq_pool = pool;
616
617 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
618 if (!rqstp->rq_argp)
619 goto out_enomem;
620
621 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
622 if (!rqstp->rq_resp)
623 goto out_enomem;
624
625 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
626 goto out_enomem;
627
628 return rqstp;
629out_enomem:
630 svc_rqst_free(rqstp);
631 return NULL;
632}
633EXPORT_SYMBOL_GPL(svc_rqst_alloc);
634
635struct svc_rqst *
636svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
637{
638 struct svc_rqst *rqstp;
639
640 rqstp = svc_rqst_alloc(serv, pool, node);
641 if (!rqstp)
642 return ERR_PTR(-ENOMEM);
643
644 serv->sv_nrthreads++;
645 spin_lock_bh(&pool->sp_lock);
646 pool->sp_nrthreads++;
647 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
648 spin_unlock_bh(&pool->sp_lock);
649 return rqstp;
650}
651EXPORT_SYMBOL_GPL(svc_prepare_thread);
652
653/*
654 * Choose a pool in which to create a new thread, for svc_set_num_threads
655 */
656static inline struct svc_pool *
657choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
658{
659 if (pool != NULL)
660 return pool;
661
662 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
663}
664
665/*
666 * Choose a thread to kill, for svc_set_num_threads
667 */
668static inline struct task_struct *
669choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
670{
671 unsigned int i;
672 struct task_struct *task = NULL;
673
674 if (pool != NULL) {
675 spin_lock_bh(&pool->sp_lock);
676 } else {
677 /* choose a pool in round-robin fashion */
678 for (i = 0; i < serv->sv_nrpools; i++) {
679 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
680 spin_lock_bh(&pool->sp_lock);
681 if (!list_empty(&pool->sp_all_threads))
682 goto found_pool;
683 spin_unlock_bh(&pool->sp_lock);
684 }
685 return NULL;
686 }
687
688found_pool:
689 if (!list_empty(&pool->sp_all_threads)) {
690 struct svc_rqst *rqstp;
691
692 /*
693 * Remove from the pool->sp_all_threads list
694 * so we don't try to kill it again.
695 */
696 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
697 set_bit(RQ_VICTIM, &rqstp->rq_flags);
698 list_del_rcu(&rqstp->rq_all);
699 task = rqstp->rq_task;
700 }
701 spin_unlock_bh(&pool->sp_lock);
702
703 return task;
704}
705
706/* create new threads */
707static int
708svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
709{
710 struct svc_rqst *rqstp;
711 struct task_struct *task;
712 struct svc_pool *chosen_pool;
713 unsigned int state = serv->sv_nrthreads-1;
714 int node;
715
716 do {
717 nrservs--;
718 chosen_pool = choose_pool(serv, pool, &state);
719
720 node = svc_pool_map_get_node(chosen_pool->sp_id);
721 rqstp = svc_prepare_thread(serv, chosen_pool, node);
722 if (IS_ERR(rqstp))
723 return PTR_ERR(rqstp);
724
725 __module_get(serv->sv_ops->svo_module);
726 task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
727 node, "%s", serv->sv_name);
728 if (IS_ERR(task)) {
729 module_put(serv->sv_ops->svo_module);
730 svc_exit_thread(rqstp);
731 return PTR_ERR(task);
732 }
733
734 rqstp->rq_task = task;
735 if (serv->sv_nrpools > 1)
736 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
737
738 svc_sock_update_bufs(serv);
739 wake_up_process(task);
740 } while (nrservs > 0);
741
742 return 0;
743}
744
745
746/* destroy old threads */
747static int
748svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
749{
750 struct task_struct *task;
751 unsigned int state = serv->sv_nrthreads-1;
752
753 /* destroy old threads */
754 do {
755 task = choose_victim(serv, pool, &state);
756 if (task == NULL)
757 break;
758 send_sig(SIGINT, task, 1);
759 nrservs++;
760 } while (nrservs < 0);
761
762 return 0;
763}
764
765/*
766 * Create or destroy enough new threads to make the number
767 * of threads the given number. If `pool' is non-NULL, applies
768 * only to threads in that pool, otherwise round-robins between
769 * all pools. Caller must ensure that mutual exclusion between this and
770 * server startup or shutdown.
771 *
772 * Destroying threads relies on the service threads filling in
773 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
774 * has been created using svc_create_pooled().
775 *
776 * Based on code that used to be in nfsd_svc() but tweaked
777 * to be pool-aware.
778 */
779int
780svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
781{
782 if (pool == NULL) {
783 /* The -1 assumes caller has done a svc_get() */
784 nrservs -= (serv->sv_nrthreads-1);
785 } else {
786 spin_lock_bh(&pool->sp_lock);
787 nrservs -= pool->sp_nrthreads;
788 spin_unlock_bh(&pool->sp_lock);
789 }
790
791 if (nrservs > 0)
792 return svc_start_kthreads(serv, pool, nrservs);
793 if (nrservs < 0)
794 return svc_signal_kthreads(serv, pool, nrservs);
795 return 0;
796}
797EXPORT_SYMBOL_GPL(svc_set_num_threads);
798
799/* destroy old threads */
800static int
801svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
802{
803 struct task_struct *task;
804 unsigned int state = serv->sv_nrthreads-1;
805
806 /* destroy old threads */
807 do {
808 task = choose_victim(serv, pool, &state);
809 if (task == NULL)
810 break;
811 kthread_stop(task);
812 nrservs++;
813 } while (nrservs < 0);
814 return 0;
815}
816
817int
818svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
819{
820 if (pool == NULL) {
821 /* The -1 assumes caller has done a svc_get() */
822 nrservs -= (serv->sv_nrthreads-1);
823 } else {
824 spin_lock_bh(&pool->sp_lock);
825 nrservs -= pool->sp_nrthreads;
826 spin_unlock_bh(&pool->sp_lock);
827 }
828
829 if (nrservs > 0)
830 return svc_start_kthreads(serv, pool, nrservs);
831 if (nrservs < 0)
832 return svc_stop_kthreads(serv, pool, nrservs);
833 return 0;
834}
835EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
836
837/*
838 * Called from a server thread as it's exiting. Caller must hold the "service
839 * mutex" for the service.
840 */
841void
842svc_rqst_free(struct svc_rqst *rqstp)
843{
844 svc_release_buffer(rqstp);
845 kfree(rqstp->rq_resp);
846 kfree(rqstp->rq_argp);
847 kfree(rqstp->rq_auth_data);
848 kfree_rcu(rqstp, rq_rcu_head);
849}
850EXPORT_SYMBOL_GPL(svc_rqst_free);
851
852void
853svc_exit_thread(struct svc_rqst *rqstp)
854{
855 struct svc_serv *serv = rqstp->rq_server;
856 struct svc_pool *pool = rqstp->rq_pool;
857
858 spin_lock_bh(&pool->sp_lock);
859 pool->sp_nrthreads--;
860 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
861 list_del_rcu(&rqstp->rq_all);
862 spin_unlock_bh(&pool->sp_lock);
863
864 svc_rqst_free(rqstp);
865
866 /* Release the server */
867 if (serv)
868 svc_destroy(serv);
869}
870EXPORT_SYMBOL_GPL(svc_exit_thread);
871
872/*
873 * Register an "inet" protocol family netid with the local
874 * rpcbind daemon via an rpcbind v4 SET request.
875 *
876 * No netconfig infrastructure is available in the kernel, so
877 * we map IP_ protocol numbers to netids by hand.
878 *
879 * Returns zero on success; a negative errno value is returned
880 * if any error occurs.
881 */
882static int __svc_rpcb_register4(struct net *net, const u32 program,
883 const u32 version,
884 const unsigned short protocol,
885 const unsigned short port)
886{
887 const struct sockaddr_in sin = {
888 .sin_family = AF_INET,
889 .sin_addr.s_addr = htonl(INADDR_ANY),
890 .sin_port = htons(port),
891 };
892 const char *netid;
893 int error;
894
895 switch (protocol) {
896 case IPPROTO_UDP:
897 netid = RPCBIND_NETID_UDP;
898 break;
899 case IPPROTO_TCP:
900 netid = RPCBIND_NETID_TCP;
901 break;
902 default:
903 return -ENOPROTOOPT;
904 }
905
906 error = rpcb_v4_register(net, program, version,
907 (const struct sockaddr *)&sin, netid);
908
909 /*
910 * User space didn't support rpcbind v4, so retry this
911 * registration request with the legacy rpcbind v2 protocol.
912 */
913 if (error == -EPROTONOSUPPORT)
914 error = rpcb_register(net, program, version, protocol, port);
915
916 return error;
917}
918
919#if IS_ENABLED(CONFIG_IPV6)
920/*
921 * Register an "inet6" protocol family netid with the local
922 * rpcbind daemon via an rpcbind v4 SET request.
923 *
924 * No netconfig infrastructure is available in the kernel, so
925 * we map IP_ protocol numbers to netids by hand.
926 *
927 * Returns zero on success; a negative errno value is returned
928 * if any error occurs.
929 */
930static int __svc_rpcb_register6(struct net *net, const u32 program,
931 const u32 version,
932 const unsigned short protocol,
933 const unsigned short port)
934{
935 const struct sockaddr_in6 sin6 = {
936 .sin6_family = AF_INET6,
937 .sin6_addr = IN6ADDR_ANY_INIT,
938 .sin6_port = htons(port),
939 };
940 const char *netid;
941 int error;
942
943 switch (protocol) {
944 case IPPROTO_UDP:
945 netid = RPCBIND_NETID_UDP6;
946 break;
947 case IPPROTO_TCP:
948 netid = RPCBIND_NETID_TCP6;
949 break;
950 default:
951 return -ENOPROTOOPT;
952 }
953
954 error = rpcb_v4_register(net, program, version,
955 (const struct sockaddr *)&sin6, netid);
956
957 /*
958 * User space didn't support rpcbind version 4, so we won't
959 * use a PF_INET6 listener.
960 */
961 if (error == -EPROTONOSUPPORT)
962 error = -EAFNOSUPPORT;
963
964 return error;
965}
966#endif /* IS_ENABLED(CONFIG_IPV6) */
967
968/*
969 * Register a kernel RPC service via rpcbind version 4.
970 *
971 * Returns zero on success; a negative errno value is returned
972 * if any error occurs.
973 */
974static int __svc_register(struct net *net, const char *progname,
975 const u32 program, const u32 version,
976 const int family,
977 const unsigned short protocol,
978 const unsigned short port)
979{
980 int error = -EAFNOSUPPORT;
981
982 switch (family) {
983 case PF_INET:
984 error = __svc_rpcb_register4(net, program, version,
985 protocol, port);
986 break;
987#if IS_ENABLED(CONFIG_IPV6)
988 case PF_INET6:
989 error = __svc_rpcb_register6(net, program, version,
990 protocol, port);
991#endif
992 }
993
994 return error;
995}
996
997int svc_rpcbind_set_version(struct net *net,
998 const struct svc_program *progp,
999 u32 version, int family,
1000 unsigned short proto,
1001 unsigned short port)
1002{
1003 dprintk("svc: svc_register(%sv%d, %s, %u, %u)\n",
1004 progp->pg_name, version,
1005 proto == IPPROTO_UDP? "udp" : "tcp",
1006 port, family);
1007
1008 return __svc_register(net, progp->pg_name, progp->pg_prog,
1009 version, family, proto, port);
1010
1011}
1012EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1013
1014int svc_generic_rpcbind_set(struct net *net,
1015 const struct svc_program *progp,
1016 u32 version, int family,
1017 unsigned short proto,
1018 unsigned short port)
1019{
1020 const struct svc_version *vers = progp->pg_vers[version];
1021 int error;
1022
1023 if (vers == NULL)
1024 return 0;
1025
1026 if (vers->vs_hidden) {
1027 dprintk("svc: svc_register(%sv%d, %s, %u, %u)"
1028 " (but not telling portmap)\n",
1029 progp->pg_name, version,
1030 proto == IPPROTO_UDP? "udp" : "tcp",
1031 port, family);
1032 return 0;
1033 }
1034
1035 /*
1036 * Don't register a UDP port if we need congestion
1037 * control.
1038 */
1039 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1040 return 0;
1041
1042 error = svc_rpcbind_set_version(net, progp, version,
1043 family, proto, port);
1044
1045 return (vers->vs_rpcb_optnl) ? 0 : error;
1046}
1047EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1048
1049/**
1050 * svc_register - register an RPC service with the local portmapper
1051 * @serv: svc_serv struct for the service to register
1052 * @net: net namespace for the service to register
1053 * @family: protocol family of service's listener socket
1054 * @proto: transport protocol number to advertise
1055 * @port: port to advertise
1056 *
1057 * Service is registered for any address in the passed-in protocol family
1058 */
1059int svc_register(const struct svc_serv *serv, struct net *net,
1060 const int family, const unsigned short proto,
1061 const unsigned short port)
1062{
1063 struct svc_program *progp;
1064 unsigned int i;
1065 int error = 0;
1066
1067 WARN_ON_ONCE(proto == 0 && port == 0);
1068 if (proto == 0 && port == 0)
1069 return -EINVAL;
1070
1071 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1072 for (i = 0; i < progp->pg_nvers; i++) {
1073
1074 error = progp->pg_rpcbind_set(net, progp, i,
1075 family, proto, port);
1076 if (error < 0) {
1077 printk(KERN_WARNING "svc: failed to register "
1078 "%sv%u RPC service (errno %d).\n",
1079 progp->pg_name, i, -error);
1080 break;
1081 }
1082 }
1083 }
1084
1085 return error;
1086}
1087
1088/*
1089 * If user space is running rpcbind, it should take the v4 UNSET
1090 * and clear everything for this [program, version]. If user space
1091 * is running portmap, it will reject the v4 UNSET, but won't have
1092 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1093 * in this case to clear all existing entries for [program, version].
1094 */
1095static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1096 const char *progname)
1097{
1098 int error;
1099
1100 error = rpcb_v4_register(net, program, version, NULL, "");
1101
1102 /*
1103 * User space didn't support rpcbind v4, so retry this
1104 * request with the legacy rpcbind v2 protocol.
1105 */
1106 if (error == -EPROTONOSUPPORT)
1107 error = rpcb_register(net, program, version, 0, 0);
1108
1109 dprintk("svc: %s(%sv%u), error %d\n",
1110 __func__, progname, version, error);
1111}
1112
1113/*
1114 * All netids, bind addresses and ports registered for [program, version]
1115 * are removed from the local rpcbind database (if the service is not
1116 * hidden) to make way for a new instance of the service.
1117 *
1118 * The result of unregistration is reported via dprintk for those who want
1119 * verification of the result, but is otherwise not important.
1120 */
1121static void svc_unregister(const struct svc_serv *serv, struct net *net)
1122{
1123 struct svc_program *progp;
1124 unsigned long flags;
1125 unsigned int i;
1126
1127 clear_thread_flag(TIF_SIGPENDING);
1128
1129 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1130 for (i = 0; i < progp->pg_nvers; i++) {
1131 if (progp->pg_vers[i] == NULL)
1132 continue;
1133 if (progp->pg_vers[i]->vs_hidden)
1134 continue;
1135
1136 dprintk("svc: attempting to unregister %sv%u\n",
1137 progp->pg_name, i);
1138 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1139 }
1140 }
1141
1142 spin_lock_irqsave(¤t->sighand->siglock, flags);
1143 recalc_sigpending();
1144 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1145}
1146
1147/*
1148 * dprintk the given error with the address of the client that caused it.
1149 */
1150#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1151static __printf(2, 3)
1152void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1153{
1154 struct va_format vaf;
1155 va_list args;
1156 char buf[RPC_MAX_ADDRBUFLEN];
1157
1158 va_start(args, fmt);
1159
1160 vaf.fmt = fmt;
1161 vaf.va = &args;
1162
1163 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1164
1165 va_end(args);
1166}
1167#else
1168static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1169#endif
1170
1171__be32
1172svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1173{
1174 set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1175 return auth_err;
1176}
1177EXPORT_SYMBOL_GPL(svc_return_autherr);
1178
1179static __be32
1180svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1181{
1182 if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1183 return *statp;
1184 return rpc_auth_ok;
1185}
1186
1187static int
1188svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1189{
1190 struct kvec *argv = &rqstp->rq_arg.head[0];
1191 struct kvec *resv = &rqstp->rq_res.head[0];
1192 const struct svc_procedure *procp = rqstp->rq_procinfo;
1193
1194 /*
1195 * Decode arguments
1196 * XXX: why do we ignore the return value?
1197 */
1198 if (procp->pc_decode &&
1199 !procp->pc_decode(rqstp, argv->iov_base)) {
1200 *statp = rpc_garbage_args;
1201 return 1;
1202 }
1203
1204 *statp = procp->pc_func(rqstp);
1205
1206 if (*statp == rpc_drop_reply ||
1207 test_bit(RQ_DROPME, &rqstp->rq_flags))
1208 return 0;
1209
1210 if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1211 return 1;
1212
1213 if (*statp != rpc_success)
1214 return 1;
1215
1216 /* Encode reply */
1217 if (procp->pc_encode &&
1218 !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1219 dprintk("svc: failed to encode reply\n");
1220 /* serv->sv_stats->rpcsystemerr++; */
1221 *statp = rpc_system_err;
1222 }
1223 return 1;
1224}
1225
1226__be32
1227svc_generic_init_request(struct svc_rqst *rqstp,
1228 const struct svc_program *progp,
1229 struct svc_process_info *ret)
1230{
1231 const struct svc_version *versp = NULL; /* compiler food */
1232 const struct svc_procedure *procp = NULL;
1233
1234 if (rqstp->rq_vers >= progp->pg_nvers )
1235 goto err_bad_vers;
1236 versp = progp->pg_vers[rqstp->rq_vers];
1237 if (!versp)
1238 goto err_bad_vers;
1239
1240 /*
1241 * Some protocol versions (namely NFSv4) require some form of
1242 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1243 * In other words, UDP is not allowed. We mark those when setting
1244 * up the svc_xprt, and verify that here.
1245 *
1246 * The spec is not very clear about what error should be returned
1247 * when someone tries to access a server that is listening on UDP
1248 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1249 * fit.
1250 */
1251 if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1252 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1253 goto err_bad_vers;
1254
1255 if (rqstp->rq_proc >= versp->vs_nproc)
1256 goto err_bad_proc;
1257 rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1258 if (!procp)
1259 goto err_bad_proc;
1260
1261 /* Initialize storage for argp and resp */
1262 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1263 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1264
1265 /* Bump per-procedure stats counter */
1266 versp->vs_count[rqstp->rq_proc]++;
1267
1268 ret->dispatch = versp->vs_dispatch;
1269 return rpc_success;
1270err_bad_vers:
1271 ret->mismatch.lovers = progp->pg_lovers;
1272 ret->mismatch.hivers = progp->pg_hivers;
1273 return rpc_prog_mismatch;
1274err_bad_proc:
1275 return rpc_proc_unavail;
1276}
1277EXPORT_SYMBOL_GPL(svc_generic_init_request);
1278
1279/*
1280 * Common routine for processing the RPC request.
1281 */
1282static int
1283svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1284{
1285 struct svc_program *progp;
1286 const struct svc_procedure *procp = NULL;
1287 struct svc_serv *serv = rqstp->rq_server;
1288 struct svc_process_info process;
1289 __be32 *statp;
1290 u32 prog, vers;
1291 __be32 auth_stat, rpc_stat;
1292 int auth_res;
1293 __be32 *reply_statp;
1294
1295 rpc_stat = rpc_success;
1296
1297 if (argv->iov_len < 6*4)
1298 goto err_short_len;
1299
1300 /* Will be turned off by GSS integrity and privacy services */
1301 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1302 /* Will be turned off only when NFSv4 Sessions are used */
1303 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1304 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1305
1306 svc_putu32(resv, rqstp->rq_xid);
1307
1308 vers = svc_getnl(argv);
1309
1310 /* First words of reply: */
1311 svc_putnl(resv, 1); /* REPLY */
1312
1313 if (vers != 2) /* RPC version number */
1314 goto err_bad_rpc;
1315
1316 /* Save position in case we later decide to reject: */
1317 reply_statp = resv->iov_base + resv->iov_len;
1318
1319 svc_putnl(resv, 0); /* ACCEPT */
1320
1321 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1322 rqstp->rq_vers = svc_getnl(argv); /* version number */
1323 rqstp->rq_proc = svc_getnl(argv); /* procedure number */
1324
1325 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1326 if (prog == progp->pg_prog)
1327 break;
1328
1329 /*
1330 * Decode auth data, and add verifier to reply buffer.
1331 * We do this before anything else in order to get a decent
1332 * auth verifier.
1333 */
1334 auth_res = svc_authenticate(rqstp, &auth_stat);
1335 /* Also give the program a chance to reject this call: */
1336 if (auth_res == SVC_OK && progp) {
1337 auth_stat = rpc_autherr_badcred;
1338 auth_res = progp->pg_authenticate(rqstp);
1339 }
1340 switch (auth_res) {
1341 case SVC_OK:
1342 break;
1343 case SVC_GARBAGE:
1344 goto err_garbage;
1345 case SVC_SYSERR:
1346 rpc_stat = rpc_system_err;
1347 goto err_bad;
1348 case SVC_DENIED:
1349 goto err_bad_auth;
1350 case SVC_CLOSE:
1351 goto close;
1352 case SVC_DROP:
1353 goto dropit;
1354 case SVC_COMPLETE:
1355 goto sendit;
1356 }
1357
1358 if (progp == NULL)
1359 goto err_bad_prog;
1360
1361 rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1362 switch (rpc_stat) {
1363 case rpc_success:
1364 break;
1365 case rpc_prog_unavail:
1366 goto err_bad_prog;
1367 case rpc_prog_mismatch:
1368 goto err_bad_vers;
1369 case rpc_proc_unavail:
1370 goto err_bad_proc;
1371 }
1372
1373 procp = rqstp->rq_procinfo;
1374 /* Should this check go into the dispatcher? */
1375 if (!procp || !procp->pc_func)
1376 goto err_bad_proc;
1377
1378 /* Syntactic check complete */
1379 serv->sv_stats->rpccnt++;
1380 trace_svc_process(rqstp, progp->pg_name);
1381
1382 /* Build the reply header. */
1383 statp = resv->iov_base +resv->iov_len;
1384 svc_putnl(resv, RPC_SUCCESS);
1385
1386 /* un-reserve some of the out-queue now that we have a
1387 * better idea of reply size
1388 */
1389 if (procp->pc_xdrressize)
1390 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1391
1392 /* Call the function that processes the request. */
1393 if (!process.dispatch) {
1394 if (!svc_generic_dispatch(rqstp, statp))
1395 goto release_dropit;
1396 if (*statp == rpc_garbage_args)
1397 goto err_garbage;
1398 auth_stat = svc_get_autherr(rqstp, statp);
1399 if (auth_stat != rpc_auth_ok)
1400 goto err_release_bad_auth;
1401 } else {
1402 dprintk("svc: calling dispatcher\n");
1403 if (!process.dispatch(rqstp, statp))
1404 goto release_dropit; /* Release reply info */
1405 }
1406
1407 /* Check RPC status result */
1408 if (*statp != rpc_success)
1409 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1410
1411 /* Release reply info */
1412 if (procp->pc_release)
1413 procp->pc_release(rqstp);
1414
1415 if (procp->pc_encode == NULL)
1416 goto dropit;
1417
1418 sendit:
1419 if (svc_authorise(rqstp))
1420 goto close;
1421 return 1; /* Caller can now send it */
1422
1423release_dropit:
1424 if (procp->pc_release)
1425 procp->pc_release(rqstp);
1426 dropit:
1427 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1428 dprintk("svc: svc_process dropit\n");
1429 return 0;
1430
1431 close:
1432 if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1433 svc_close_xprt(rqstp->rq_xprt);
1434 dprintk("svc: svc_process close\n");
1435 return 0;
1436
1437err_short_len:
1438 svc_printk(rqstp, "short len %zd, dropping request\n",
1439 argv->iov_len);
1440 goto close;
1441
1442err_bad_rpc:
1443 serv->sv_stats->rpcbadfmt++;
1444 svc_putnl(resv, 1); /* REJECT */
1445 svc_putnl(resv, 0); /* RPC_MISMATCH */
1446 svc_putnl(resv, 2); /* Only RPCv2 supported */
1447 svc_putnl(resv, 2);
1448 goto sendit;
1449
1450err_release_bad_auth:
1451 if (procp->pc_release)
1452 procp->pc_release(rqstp);
1453err_bad_auth:
1454 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1455 serv->sv_stats->rpcbadauth++;
1456 /* Restore write pointer to location of accept status: */
1457 xdr_ressize_check(rqstp, reply_statp);
1458 svc_putnl(resv, 1); /* REJECT */
1459 svc_putnl(resv, 1); /* AUTH_ERROR */
1460 svc_putnl(resv, ntohl(auth_stat)); /* status */
1461 goto sendit;
1462
1463err_bad_prog:
1464 dprintk("svc: unknown program %d\n", prog);
1465 serv->sv_stats->rpcbadfmt++;
1466 svc_putnl(resv, RPC_PROG_UNAVAIL);
1467 goto sendit;
1468
1469err_bad_vers:
1470 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1471 rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1472
1473 serv->sv_stats->rpcbadfmt++;
1474 svc_putnl(resv, RPC_PROG_MISMATCH);
1475 svc_putnl(resv, process.mismatch.lovers);
1476 svc_putnl(resv, process.mismatch.hivers);
1477 goto sendit;
1478
1479err_bad_proc:
1480 svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1481
1482 serv->sv_stats->rpcbadfmt++;
1483 svc_putnl(resv, RPC_PROC_UNAVAIL);
1484 goto sendit;
1485
1486err_garbage:
1487 svc_printk(rqstp, "failed to decode args\n");
1488
1489 rpc_stat = rpc_garbage_args;
1490err_bad:
1491 serv->sv_stats->rpcbadfmt++;
1492 svc_putnl(resv, ntohl(rpc_stat));
1493 goto sendit;
1494}
1495
1496/*
1497 * Process the RPC request.
1498 */
1499int
1500svc_process(struct svc_rqst *rqstp)
1501{
1502 struct kvec *argv = &rqstp->rq_arg.head[0];
1503 struct kvec *resv = &rqstp->rq_res.head[0];
1504 struct svc_serv *serv = rqstp->rq_server;
1505 u32 dir;
1506
1507 /*
1508 * Setup response xdr_buf.
1509 * Initially it has just one page
1510 */
1511 rqstp->rq_next_page = &rqstp->rq_respages[1];
1512 resv->iov_base = page_address(rqstp->rq_respages[0]);
1513 resv->iov_len = 0;
1514 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1515 rqstp->rq_res.len = 0;
1516 rqstp->rq_res.page_base = 0;
1517 rqstp->rq_res.page_len = 0;
1518 rqstp->rq_res.buflen = PAGE_SIZE;
1519 rqstp->rq_res.tail[0].iov_base = NULL;
1520 rqstp->rq_res.tail[0].iov_len = 0;
1521
1522 dir = svc_getnl(argv);
1523 if (dir != 0) {
1524 /* direction != CALL */
1525 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1526 serv->sv_stats->rpcbadfmt++;
1527 goto out_drop;
1528 }
1529
1530 /* Reserve space for the record marker */
1531 if (rqstp->rq_prot == IPPROTO_TCP)
1532 svc_putnl(resv, 0);
1533
1534 /* Returns 1 for send, 0 for drop */
1535 if (likely(svc_process_common(rqstp, argv, resv)))
1536 return svc_send(rqstp);
1537
1538out_drop:
1539 svc_drop(rqstp);
1540 return 0;
1541}
1542EXPORT_SYMBOL_GPL(svc_process);
1543
1544#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1545/*
1546 * Process a backchannel RPC request that arrived over an existing
1547 * outbound connection
1548 */
1549int
1550bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1551 struct svc_rqst *rqstp)
1552{
1553 struct kvec *argv = &rqstp->rq_arg.head[0];
1554 struct kvec *resv = &rqstp->rq_res.head[0];
1555 struct rpc_task *task;
1556 int proc_error;
1557 int error;
1558
1559 dprintk("svc: %s(%p)\n", __func__, req);
1560
1561 /* Build the svc_rqst used by the common processing routine */
1562 rqstp->rq_xid = req->rq_xid;
1563 rqstp->rq_prot = req->rq_xprt->prot;
1564 rqstp->rq_server = serv;
1565 rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1566
1567 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1568 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1569 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1570 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1571
1572 /* Adjust the argument buffer length */
1573 rqstp->rq_arg.len = req->rq_private_buf.len;
1574 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1575 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1576 rqstp->rq_arg.page_len = 0;
1577 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1578 rqstp->rq_arg.page_len)
1579 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1580 rqstp->rq_arg.head[0].iov_len;
1581 else
1582 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1583 rqstp->rq_arg.page_len;
1584
1585 /* reset result send buffer "put" position */
1586 resv->iov_len = 0;
1587
1588 /*
1589 * Skip the next two words because they've already been
1590 * processed in the transport
1591 */
1592 svc_getu32(argv); /* XID */
1593 svc_getnl(argv); /* CALLDIR */
1594
1595 /* Parse and execute the bc call */
1596 proc_error = svc_process_common(rqstp, argv, resv);
1597
1598 atomic_dec(&req->rq_xprt->bc_slot_count);
1599 if (!proc_error) {
1600 /* Processing error: drop the request */
1601 xprt_free_bc_request(req);
1602 error = -EINVAL;
1603 goto out;
1604 }
1605 /* Finally, send the reply synchronously */
1606 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1607 task = rpc_run_bc_task(req);
1608 if (IS_ERR(task)) {
1609 error = PTR_ERR(task);
1610 goto out;
1611 }
1612
1613 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1614 error = task->tk_status;
1615 rpc_put_task(task);
1616
1617out:
1618 dprintk("svc: %s(), error=%d\n", __func__, error);
1619 return error;
1620}
1621EXPORT_SYMBOL_GPL(bc_svc_process);
1622#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1623
1624/*
1625 * Return (transport-specific) limit on the rpc payload.
1626 */
1627u32 svc_max_payload(const struct svc_rqst *rqstp)
1628{
1629 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1630
1631 if (rqstp->rq_server->sv_max_payload < max)
1632 max = rqstp->rq_server->sv_max_payload;
1633 return max;
1634}
1635EXPORT_SYMBOL_GPL(svc_max_payload);
1636
1637/**
1638 * svc_fill_write_vector - Construct data argument for VFS write call
1639 * @rqstp: svc_rqst to operate on
1640 * @pages: list of pages containing data payload
1641 * @first: buffer containing first section of write payload
1642 * @total: total number of bytes of write payload
1643 *
1644 * Fills in rqstp::rq_vec, and returns the number of elements.
1645 */
1646unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1647 struct kvec *first, size_t total)
1648{
1649 struct kvec *vec = rqstp->rq_vec;
1650 unsigned int i;
1651
1652 /* Some types of transport can present the write payload
1653 * entirely in rq_arg.pages. In this case, @first is empty.
1654 */
1655 i = 0;
1656 if (first->iov_len) {
1657 vec[i].iov_base = first->iov_base;
1658 vec[i].iov_len = min_t(size_t, total, first->iov_len);
1659 total -= vec[i].iov_len;
1660 ++i;
1661 }
1662
1663 while (total) {
1664 vec[i].iov_base = page_address(*pages);
1665 vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1666 total -= vec[i].iov_len;
1667 ++i;
1668 ++pages;
1669 }
1670
1671 WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1672 return i;
1673}
1674EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1675
1676/**
1677 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1678 * @rqstp: svc_rqst to operate on
1679 * @first: buffer containing first section of pathname
1680 * @p: buffer containing remaining section of pathname
1681 * @total: total length of the pathname argument
1682 *
1683 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1684 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1685 * the returned string.
1686 */
1687char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1688 void *p, size_t total)
1689{
1690 size_t len, remaining;
1691 char *result, *dst;
1692
1693 result = kmalloc(total + 1, GFP_KERNEL);
1694 if (!result)
1695 return ERR_PTR(-ESERVERFAULT);
1696
1697 dst = result;
1698 remaining = total;
1699
1700 len = min_t(size_t, total, first->iov_len);
1701 if (len) {
1702 memcpy(dst, first->iov_base, len);
1703 dst += len;
1704 remaining -= len;
1705 }
1706
1707 if (remaining) {
1708 len = min_t(size_t, remaining, PAGE_SIZE);
1709 memcpy(dst, p, len);
1710 dst += len;
1711 }
1712
1713 *dst = '\0';
1714
1715 /* Sanity check: Linux doesn't allow the pathname argument to
1716 * contain a NUL byte.
1717 */
1718 if (strlen(result) != total) {
1719 kfree(result);
1720 return ERR_PTR(-EINVAL);
1721 }
1722 return result;
1723}
1724EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);