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/*
2 * linux/net/sunrpc/svc.c
3 *
4 * High-level RPC service routines
5 *
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 *
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
10 * by Greg Banks <gnb@melbourne.sgi.com>
11 */
12
13#include <linux/linkage.h>
14#include <linux/sched.h>
15#include <linux/errno.h>
16#include <linux/net.h>
17#include <linux/in.h>
18#include <linux/mm.h>
19#include <linux/interrupt.h>
20#include <linux/module.h>
21#include <linux/kthread.h>
22#include <linux/slab.h>
23
24#include <linux/sunrpc/types.h>
25#include <linux/sunrpc/xdr.h>
26#include <linux/sunrpc/stats.h>
27#include <linux/sunrpc/svcsock.h>
28#include <linux/sunrpc/clnt.h>
29#include <linux/sunrpc/bc_xprt.h>
30
31#define RPCDBG_FACILITY RPCDBG_SVCDSP
32
33static void svc_unregister(const struct svc_serv *serv);
34
35#define svc_serv_is_pooled(serv) ((serv)->sv_function)
36
37/*
38 * Mode for mapping cpus to pools.
39 */
40enum {
41 SVC_POOL_AUTO = -1, /* choose one of the others */
42 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
43 * (legacy & UP mode) */
44 SVC_POOL_PERCPU, /* one pool per cpu */
45 SVC_POOL_PERNODE /* one pool per numa node */
46};
47#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
48
49/*
50 * Structure for mapping cpus to pools and vice versa.
51 * Setup once during sunrpc initialisation.
52 */
53static struct svc_pool_map {
54 int count; /* How many svc_servs use us */
55 int mode; /* Note: int not enum to avoid
56 * warnings about "enumeration value
57 * not handled in switch" */
58 unsigned int npools;
59 unsigned int *pool_to; /* maps pool id to cpu or node */
60 unsigned int *to_pool; /* maps cpu or node to pool id */
61} svc_pool_map = {
62 .count = 0,
63 .mode = SVC_POOL_DEFAULT
64};
65static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
66
67static int
68param_set_pool_mode(const char *val, struct kernel_param *kp)
69{
70 int *ip = (int *)kp->arg;
71 struct svc_pool_map *m = &svc_pool_map;
72 int err;
73
74 mutex_lock(&svc_pool_map_mutex);
75
76 err = -EBUSY;
77 if (m->count)
78 goto out;
79
80 err = 0;
81 if (!strncmp(val, "auto", 4))
82 *ip = SVC_POOL_AUTO;
83 else if (!strncmp(val, "global", 6))
84 *ip = SVC_POOL_GLOBAL;
85 else if (!strncmp(val, "percpu", 6))
86 *ip = SVC_POOL_PERCPU;
87 else if (!strncmp(val, "pernode", 7))
88 *ip = SVC_POOL_PERNODE;
89 else
90 err = -EINVAL;
91
92out:
93 mutex_unlock(&svc_pool_map_mutex);
94 return err;
95}
96
97static int
98param_get_pool_mode(char *buf, struct kernel_param *kp)
99{
100 int *ip = (int *)kp->arg;
101
102 switch (*ip)
103 {
104 case SVC_POOL_AUTO:
105 return strlcpy(buf, "auto", 20);
106 case SVC_POOL_GLOBAL:
107 return strlcpy(buf, "global", 20);
108 case SVC_POOL_PERCPU:
109 return strlcpy(buf, "percpu", 20);
110 case SVC_POOL_PERNODE:
111 return strlcpy(buf, "pernode", 20);
112 default:
113 return sprintf(buf, "%d", *ip);
114 }
115}
116
117module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
118 &svc_pool_map.mode, 0644);
119
120/*
121 * Detect best pool mapping mode heuristically,
122 * according to the machine's topology.
123 */
124static int
125svc_pool_map_choose_mode(void)
126{
127 unsigned int node;
128
129 if (nr_online_nodes > 1) {
130 /*
131 * Actually have multiple NUMA nodes,
132 * so split pools on NUMA node boundaries
133 */
134 return SVC_POOL_PERNODE;
135 }
136
137 node = first_online_node;
138 if (nr_cpus_node(node) > 2) {
139 /*
140 * Non-trivial SMP, or CONFIG_NUMA on
141 * non-NUMA hardware, e.g. with a generic
142 * x86_64 kernel on Xeons. In this case we
143 * want to divide the pools on cpu boundaries.
144 */
145 return SVC_POOL_PERCPU;
146 }
147
148 /* default: one global pool */
149 return SVC_POOL_GLOBAL;
150}
151
152/*
153 * Allocate the to_pool[] and pool_to[] arrays.
154 * Returns 0 on success or an errno.
155 */
156static int
157svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
158{
159 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
160 if (!m->to_pool)
161 goto fail;
162 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
163 if (!m->pool_to)
164 goto fail_free;
165
166 return 0;
167
168fail_free:
169 kfree(m->to_pool);
170fail:
171 return -ENOMEM;
172}
173
174/*
175 * Initialise the pool map for SVC_POOL_PERCPU mode.
176 * Returns number of pools or <0 on error.
177 */
178static int
179svc_pool_map_init_percpu(struct svc_pool_map *m)
180{
181 unsigned int maxpools = nr_cpu_ids;
182 unsigned int pidx = 0;
183 unsigned int cpu;
184 int err;
185
186 err = svc_pool_map_alloc_arrays(m, maxpools);
187 if (err)
188 return err;
189
190 for_each_online_cpu(cpu) {
191 BUG_ON(pidx > maxpools);
192 m->to_pool[cpu] = pidx;
193 m->pool_to[pidx] = cpu;
194 pidx++;
195 }
196 /* cpus brought online later all get mapped to pool0, sorry */
197
198 return pidx;
199};
200
201
202/*
203 * Initialise the pool map for SVC_POOL_PERNODE mode.
204 * Returns number of pools or <0 on error.
205 */
206static int
207svc_pool_map_init_pernode(struct svc_pool_map *m)
208{
209 unsigned int maxpools = nr_node_ids;
210 unsigned int pidx = 0;
211 unsigned int node;
212 int err;
213
214 err = svc_pool_map_alloc_arrays(m, maxpools);
215 if (err)
216 return err;
217
218 for_each_node_with_cpus(node) {
219 /* some architectures (e.g. SN2) have cpuless nodes */
220 BUG_ON(pidx > maxpools);
221 m->to_pool[node] = pidx;
222 m->pool_to[pidx] = node;
223 pidx++;
224 }
225 /* nodes brought online later all get mapped to pool0, sorry */
226
227 return pidx;
228}
229
230
231/*
232 * Add a reference to the global map of cpus to pools (and
233 * vice versa). Initialise the map if we're the first user.
234 * Returns the number of pools.
235 */
236static unsigned int
237svc_pool_map_get(void)
238{
239 struct svc_pool_map *m = &svc_pool_map;
240 int npools = -1;
241
242 mutex_lock(&svc_pool_map_mutex);
243
244 if (m->count++) {
245 mutex_unlock(&svc_pool_map_mutex);
246 return m->npools;
247 }
248
249 if (m->mode == SVC_POOL_AUTO)
250 m->mode = svc_pool_map_choose_mode();
251
252 switch (m->mode) {
253 case SVC_POOL_PERCPU:
254 npools = svc_pool_map_init_percpu(m);
255 break;
256 case SVC_POOL_PERNODE:
257 npools = svc_pool_map_init_pernode(m);
258 break;
259 }
260
261 if (npools < 0) {
262 /* default, or memory allocation failure */
263 npools = 1;
264 m->mode = SVC_POOL_GLOBAL;
265 }
266 m->npools = npools;
267
268 mutex_unlock(&svc_pool_map_mutex);
269 return m->npools;
270}
271
272
273/*
274 * Drop a reference to the global map of cpus to pools.
275 * When the last reference is dropped, the map data is
276 * freed; this allows the sysadmin to change the pool
277 * mode using the pool_mode module option without
278 * rebooting or re-loading sunrpc.ko.
279 */
280static void
281svc_pool_map_put(void)
282{
283 struct svc_pool_map *m = &svc_pool_map;
284
285 mutex_lock(&svc_pool_map_mutex);
286
287 if (!--m->count) {
288 m->mode = SVC_POOL_DEFAULT;
289 kfree(m->to_pool);
290 kfree(m->pool_to);
291 m->npools = 0;
292 }
293
294 mutex_unlock(&svc_pool_map_mutex);
295}
296
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 BUG_ON(m->count == 0);
313
314 switch (m->mode) {
315 case SVC_POOL_PERCPU:
316 {
317 set_cpus_allowed_ptr(task, cpumask_of(node));
318 break;
319 }
320 case SVC_POOL_PERNODE:
321 {
322 set_cpus_allowed_ptr(task, cpumask_of_node(node));
323 break;
324 }
325 }
326}
327
328/*
329 * Use the mapping mode to choose a pool for a given CPU.
330 * Used when enqueueing an incoming RPC. Always returns
331 * a non-NULL pool pointer.
332 */
333struct svc_pool *
334svc_pool_for_cpu(struct svc_serv *serv, int cpu)
335{
336 struct svc_pool_map *m = &svc_pool_map;
337 unsigned int pidx = 0;
338
339 /*
340 * An uninitialised map happens in a pure client when
341 * lockd is brought up, so silently treat it the
342 * same as SVC_POOL_GLOBAL.
343 */
344 if (svc_serv_is_pooled(serv)) {
345 switch (m->mode) {
346 case SVC_POOL_PERCPU:
347 pidx = m->to_pool[cpu];
348 break;
349 case SVC_POOL_PERNODE:
350 pidx = m->to_pool[cpu_to_node(cpu)];
351 break;
352 }
353 }
354 return &serv->sv_pools[pidx % serv->sv_nrpools];
355}
356
357
358/*
359 * Create an RPC service
360 */
361static struct svc_serv *
362__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
363 void (*shutdown)(struct svc_serv *serv))
364{
365 struct svc_serv *serv;
366 unsigned int vers;
367 unsigned int xdrsize;
368 unsigned int i;
369
370 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
371 return NULL;
372 serv->sv_name = prog->pg_name;
373 serv->sv_program = prog;
374 serv->sv_nrthreads = 1;
375 serv->sv_stats = prog->pg_stats;
376 if (bufsize > RPCSVC_MAXPAYLOAD)
377 bufsize = RPCSVC_MAXPAYLOAD;
378 serv->sv_max_payload = bufsize? bufsize : 4096;
379 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
380 serv->sv_shutdown = shutdown;
381 xdrsize = 0;
382 while (prog) {
383 prog->pg_lovers = prog->pg_nvers-1;
384 for (vers=0; vers<prog->pg_nvers ; vers++)
385 if (prog->pg_vers[vers]) {
386 prog->pg_hivers = vers;
387 if (prog->pg_lovers > vers)
388 prog->pg_lovers = vers;
389 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
390 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
391 }
392 prog = prog->pg_next;
393 }
394 serv->sv_xdrsize = xdrsize;
395 INIT_LIST_HEAD(&serv->sv_tempsocks);
396 INIT_LIST_HEAD(&serv->sv_permsocks);
397 init_timer(&serv->sv_temptimer);
398 spin_lock_init(&serv->sv_lock);
399
400 serv->sv_nrpools = npools;
401 serv->sv_pools =
402 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
403 GFP_KERNEL);
404 if (!serv->sv_pools) {
405 kfree(serv);
406 return NULL;
407 }
408
409 for (i = 0; i < serv->sv_nrpools; i++) {
410 struct svc_pool *pool = &serv->sv_pools[i];
411
412 dprintk("svc: initialising pool %u for %s\n",
413 i, serv->sv_name);
414
415 pool->sp_id = i;
416 INIT_LIST_HEAD(&pool->sp_threads);
417 INIT_LIST_HEAD(&pool->sp_sockets);
418 INIT_LIST_HEAD(&pool->sp_all_threads);
419 spin_lock_init(&pool->sp_lock);
420 }
421
422 /* Remove any stale portmap registrations */
423 svc_unregister(serv);
424
425 return serv;
426}
427
428struct svc_serv *
429svc_create(struct svc_program *prog, unsigned int bufsize,
430 void (*shutdown)(struct svc_serv *serv))
431{
432 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
433}
434EXPORT_SYMBOL_GPL(svc_create);
435
436struct svc_serv *
437svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
438 void (*shutdown)(struct svc_serv *serv),
439 svc_thread_fn func, struct module *mod)
440{
441 struct svc_serv *serv;
442 unsigned int npools = svc_pool_map_get();
443
444 serv = __svc_create(prog, bufsize, npools, shutdown);
445
446 if (serv != NULL) {
447 serv->sv_function = func;
448 serv->sv_module = mod;
449 }
450
451 return serv;
452}
453EXPORT_SYMBOL_GPL(svc_create_pooled);
454
455/*
456 * Destroy an RPC service. Should be called with appropriate locking to
457 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
458 */
459void
460svc_destroy(struct svc_serv *serv)
461{
462 dprintk("svc: svc_destroy(%s, %d)\n",
463 serv->sv_program->pg_name,
464 serv->sv_nrthreads);
465
466 if (serv->sv_nrthreads) {
467 if (--(serv->sv_nrthreads) != 0) {
468 svc_sock_update_bufs(serv);
469 return;
470 }
471 } else
472 printk("svc_destroy: no threads for serv=%p!\n", serv);
473
474 del_timer_sync(&serv->sv_temptimer);
475
476 svc_close_all(&serv->sv_tempsocks);
477
478 if (serv->sv_shutdown)
479 serv->sv_shutdown(serv);
480
481 svc_close_all(&serv->sv_permsocks);
482
483 BUG_ON(!list_empty(&serv->sv_permsocks));
484 BUG_ON(!list_empty(&serv->sv_tempsocks));
485
486 cache_clean_deferred(serv);
487
488 if (svc_serv_is_pooled(serv))
489 svc_pool_map_put();
490
491 svc_unregister(serv);
492 kfree(serv->sv_pools);
493 kfree(serv);
494}
495EXPORT_SYMBOL_GPL(svc_destroy);
496
497/*
498 * Allocate an RPC server's buffer space.
499 * We allocate pages and place them in rq_argpages.
500 */
501static int
502svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
503{
504 unsigned int pages, arghi;
505
506 /* bc_xprt uses fore channel allocated buffers */
507 if (svc_is_backchannel(rqstp))
508 return 1;
509
510 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
511 * We assume one is at most one page
512 */
513 arghi = 0;
514 BUG_ON(pages > RPCSVC_MAXPAGES);
515 while (pages) {
516 struct page *p = alloc_page(GFP_KERNEL);
517 if (!p)
518 break;
519 rqstp->rq_pages[arghi++] = p;
520 pages--;
521 }
522 return pages == 0;
523}
524
525/*
526 * Release an RPC server buffer
527 */
528static void
529svc_release_buffer(struct svc_rqst *rqstp)
530{
531 unsigned int i;
532
533 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
534 if (rqstp->rq_pages[i])
535 put_page(rqstp->rq_pages[i]);
536}
537
538struct svc_rqst *
539svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
540{
541 struct svc_rqst *rqstp;
542
543 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
544 if (!rqstp)
545 goto out_enomem;
546
547 init_waitqueue_head(&rqstp->rq_wait);
548
549 serv->sv_nrthreads++;
550 spin_lock_bh(&pool->sp_lock);
551 pool->sp_nrthreads++;
552 list_add(&rqstp->rq_all, &pool->sp_all_threads);
553 spin_unlock_bh(&pool->sp_lock);
554 rqstp->rq_server = serv;
555 rqstp->rq_pool = pool;
556
557 rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
558 if (!rqstp->rq_argp)
559 goto out_thread;
560
561 rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
562 if (!rqstp->rq_resp)
563 goto out_thread;
564
565 if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
566 goto out_thread;
567
568 return rqstp;
569out_thread:
570 svc_exit_thread(rqstp);
571out_enomem:
572 return ERR_PTR(-ENOMEM);
573}
574EXPORT_SYMBOL_GPL(svc_prepare_thread);
575
576/*
577 * Choose a pool in which to create a new thread, for svc_set_num_threads
578 */
579static inline struct svc_pool *
580choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
581{
582 if (pool != NULL)
583 return pool;
584
585 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
586}
587
588/*
589 * Choose a thread to kill, for svc_set_num_threads
590 */
591static inline struct task_struct *
592choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
593{
594 unsigned int i;
595 struct task_struct *task = NULL;
596
597 if (pool != NULL) {
598 spin_lock_bh(&pool->sp_lock);
599 } else {
600 /* choose a pool in round-robin fashion */
601 for (i = 0; i < serv->sv_nrpools; i++) {
602 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
603 spin_lock_bh(&pool->sp_lock);
604 if (!list_empty(&pool->sp_all_threads))
605 goto found_pool;
606 spin_unlock_bh(&pool->sp_lock);
607 }
608 return NULL;
609 }
610
611found_pool:
612 if (!list_empty(&pool->sp_all_threads)) {
613 struct svc_rqst *rqstp;
614
615 /*
616 * Remove from the pool->sp_all_threads list
617 * so we don't try to kill it again.
618 */
619 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
620 list_del_init(&rqstp->rq_all);
621 task = rqstp->rq_task;
622 }
623 spin_unlock_bh(&pool->sp_lock);
624
625 return task;
626}
627
628/*
629 * Create or destroy enough new threads to make the number
630 * of threads the given number. If `pool' is non-NULL, applies
631 * only to threads in that pool, otherwise round-robins between
632 * all pools. Must be called with a svc_get() reference and
633 * the BKL or another lock to protect access to svc_serv fields.
634 *
635 * Destroying threads relies on the service threads filling in
636 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
637 * has been created using svc_create_pooled().
638 *
639 * Based on code that used to be in nfsd_svc() but tweaked
640 * to be pool-aware.
641 */
642int
643svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
644{
645 struct svc_rqst *rqstp;
646 struct task_struct *task;
647 struct svc_pool *chosen_pool;
648 int error = 0;
649 unsigned int state = serv->sv_nrthreads-1;
650
651 if (pool == NULL) {
652 /* The -1 assumes caller has done a svc_get() */
653 nrservs -= (serv->sv_nrthreads-1);
654 } else {
655 spin_lock_bh(&pool->sp_lock);
656 nrservs -= pool->sp_nrthreads;
657 spin_unlock_bh(&pool->sp_lock);
658 }
659
660 /* create new threads */
661 while (nrservs > 0) {
662 nrservs--;
663 chosen_pool = choose_pool(serv, pool, &state);
664
665 rqstp = svc_prepare_thread(serv, chosen_pool);
666 if (IS_ERR(rqstp)) {
667 error = PTR_ERR(rqstp);
668 break;
669 }
670
671 __module_get(serv->sv_module);
672 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
673 if (IS_ERR(task)) {
674 error = PTR_ERR(task);
675 module_put(serv->sv_module);
676 svc_exit_thread(rqstp);
677 break;
678 }
679
680 rqstp->rq_task = task;
681 if (serv->sv_nrpools > 1)
682 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
683
684 svc_sock_update_bufs(serv);
685 wake_up_process(task);
686 }
687 /* destroy old threads */
688 while (nrservs < 0 &&
689 (task = choose_victim(serv, pool, &state)) != NULL) {
690 send_sig(SIGINT, task, 1);
691 nrservs++;
692 }
693
694 return error;
695}
696EXPORT_SYMBOL_GPL(svc_set_num_threads);
697
698/*
699 * Called from a server thread as it's exiting. Caller must hold the BKL or
700 * the "service mutex", whichever is appropriate for the service.
701 */
702void
703svc_exit_thread(struct svc_rqst *rqstp)
704{
705 struct svc_serv *serv = rqstp->rq_server;
706 struct svc_pool *pool = rqstp->rq_pool;
707
708 svc_release_buffer(rqstp);
709 kfree(rqstp->rq_resp);
710 kfree(rqstp->rq_argp);
711 kfree(rqstp->rq_auth_data);
712
713 spin_lock_bh(&pool->sp_lock);
714 pool->sp_nrthreads--;
715 list_del(&rqstp->rq_all);
716 spin_unlock_bh(&pool->sp_lock);
717
718 kfree(rqstp);
719
720 /* Release the server */
721 if (serv)
722 svc_destroy(serv);
723}
724EXPORT_SYMBOL_GPL(svc_exit_thread);
725
726/*
727 * Register an "inet" protocol family netid with the local
728 * rpcbind daemon via an rpcbind v4 SET request.
729 *
730 * No netconfig infrastructure is available in the kernel, so
731 * we map IP_ protocol numbers to netids by hand.
732 *
733 * Returns zero on success; a negative errno value is returned
734 * if any error occurs.
735 */
736static int __svc_rpcb_register4(const u32 program, const u32 version,
737 const unsigned short protocol,
738 const unsigned short port)
739{
740 const struct sockaddr_in sin = {
741 .sin_family = AF_INET,
742 .sin_addr.s_addr = htonl(INADDR_ANY),
743 .sin_port = htons(port),
744 };
745 const char *netid;
746 int error;
747
748 switch (protocol) {
749 case IPPROTO_UDP:
750 netid = RPCBIND_NETID_UDP;
751 break;
752 case IPPROTO_TCP:
753 netid = RPCBIND_NETID_TCP;
754 break;
755 default:
756 return -ENOPROTOOPT;
757 }
758
759 error = rpcb_v4_register(program, version,
760 (const struct sockaddr *)&sin, netid);
761
762 /*
763 * User space didn't support rpcbind v4, so retry this
764 * registration request with the legacy rpcbind v2 protocol.
765 */
766 if (error == -EPROTONOSUPPORT)
767 error = rpcb_register(program, version, protocol, port);
768
769 return error;
770}
771
772#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
773/*
774 * Register an "inet6" protocol family netid with the local
775 * rpcbind daemon via an rpcbind v4 SET request.
776 *
777 * No netconfig infrastructure is available in the kernel, so
778 * we map IP_ protocol numbers to netids by hand.
779 *
780 * Returns zero on success; a negative errno value is returned
781 * if any error occurs.
782 */
783static int __svc_rpcb_register6(const u32 program, const u32 version,
784 const unsigned short protocol,
785 const unsigned short port)
786{
787 const struct sockaddr_in6 sin6 = {
788 .sin6_family = AF_INET6,
789 .sin6_addr = IN6ADDR_ANY_INIT,
790 .sin6_port = htons(port),
791 };
792 const char *netid;
793 int error;
794
795 switch (protocol) {
796 case IPPROTO_UDP:
797 netid = RPCBIND_NETID_UDP6;
798 break;
799 case IPPROTO_TCP:
800 netid = RPCBIND_NETID_TCP6;
801 break;
802 default:
803 return -ENOPROTOOPT;
804 }
805
806 error = rpcb_v4_register(program, version,
807 (const struct sockaddr *)&sin6, netid);
808
809 /*
810 * User space didn't support rpcbind version 4, so we won't
811 * use a PF_INET6 listener.
812 */
813 if (error == -EPROTONOSUPPORT)
814 error = -EAFNOSUPPORT;
815
816 return error;
817}
818#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
819
820/*
821 * Register a kernel RPC service via rpcbind version 4.
822 *
823 * Returns zero on success; a negative errno value is returned
824 * if any error occurs.
825 */
826static int __svc_register(const char *progname,
827 const u32 program, const u32 version,
828 const int family,
829 const unsigned short protocol,
830 const unsigned short port)
831{
832 int error = -EAFNOSUPPORT;
833
834 switch (family) {
835 case PF_INET:
836 error = __svc_rpcb_register4(program, version,
837 protocol, port);
838 break;
839#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
840 case PF_INET6:
841 error = __svc_rpcb_register6(program, version,
842 protocol, port);
843#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
844 }
845
846 if (error < 0)
847 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
848 "service (errno %d).\n", progname, version, -error);
849 return error;
850}
851
852/**
853 * svc_register - register an RPC service with the local portmapper
854 * @serv: svc_serv struct for the service to register
855 * @family: protocol family of service's listener socket
856 * @proto: transport protocol number to advertise
857 * @port: port to advertise
858 *
859 * Service is registered for any address in the passed-in protocol family
860 */
861int svc_register(const struct svc_serv *serv, const int family,
862 const unsigned short proto, const unsigned short port)
863{
864 struct svc_program *progp;
865 unsigned int i;
866 int error = 0;
867
868 BUG_ON(proto == 0 && port == 0);
869
870 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
871 for (i = 0; i < progp->pg_nvers; i++) {
872 if (progp->pg_vers[i] == NULL)
873 continue;
874
875 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
876 progp->pg_name,
877 i,
878 proto == IPPROTO_UDP? "udp" : "tcp",
879 port,
880 family,
881 progp->pg_vers[i]->vs_hidden?
882 " (but not telling portmap)" : "");
883
884 if (progp->pg_vers[i]->vs_hidden)
885 continue;
886
887 error = __svc_register(progp->pg_name, progp->pg_prog,
888 i, family, proto, port);
889 if (error < 0)
890 break;
891 }
892 }
893
894 return error;
895}
896
897/*
898 * If user space is running rpcbind, it should take the v4 UNSET
899 * and clear everything for this [program, version]. If user space
900 * is running portmap, it will reject the v4 UNSET, but won't have
901 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
902 * in this case to clear all existing entries for [program, version].
903 */
904static void __svc_unregister(const u32 program, const u32 version,
905 const char *progname)
906{
907 int error;
908
909 error = rpcb_v4_register(program, version, NULL, "");
910
911 /*
912 * User space didn't support rpcbind v4, so retry this
913 * request with the legacy rpcbind v2 protocol.
914 */
915 if (error == -EPROTONOSUPPORT)
916 error = rpcb_register(program, version, 0, 0);
917
918 dprintk("svc: %s(%sv%u), error %d\n",
919 __func__, progname, version, error);
920}
921
922/*
923 * All netids, bind addresses and ports registered for [program, version]
924 * are removed from the local rpcbind database (if the service is not
925 * hidden) to make way for a new instance of the service.
926 *
927 * The result of unregistration is reported via dprintk for those who want
928 * verification of the result, but is otherwise not important.
929 */
930static void svc_unregister(const struct svc_serv *serv)
931{
932 struct svc_program *progp;
933 unsigned long flags;
934 unsigned int i;
935
936 clear_thread_flag(TIF_SIGPENDING);
937
938 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
939 for (i = 0; i < progp->pg_nvers; i++) {
940 if (progp->pg_vers[i] == NULL)
941 continue;
942 if (progp->pg_vers[i]->vs_hidden)
943 continue;
944
945 dprintk("svc: attempting to unregister %sv%u\n",
946 progp->pg_name, i);
947 __svc_unregister(progp->pg_prog, i, progp->pg_name);
948 }
949 }
950
951 spin_lock_irqsave(¤t->sighand->siglock, flags);
952 recalc_sigpending();
953 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
954}
955
956/*
957 * Printk the given error with the address of the client that caused it.
958 */
959static int
960__attribute__ ((format (printf, 2, 3)))
961svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
962{
963 va_list args;
964 int r;
965 char buf[RPC_MAX_ADDRBUFLEN];
966
967 if (!net_ratelimit())
968 return 0;
969
970 printk(KERN_WARNING "svc: %s: ",
971 svc_print_addr(rqstp, buf, sizeof(buf)));
972
973 va_start(args, fmt);
974 r = vprintk(fmt, args);
975 va_end(args);
976
977 return r;
978}
979
980/*
981 * Common routine for processing the RPC request.
982 */
983static int
984svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
985{
986 struct svc_program *progp;
987 struct svc_version *versp = NULL; /* compiler food */
988 struct svc_procedure *procp = NULL;
989 struct svc_serv *serv = rqstp->rq_server;
990 kxdrproc_t xdr;
991 __be32 *statp;
992 u32 prog, vers, proc;
993 __be32 auth_stat, rpc_stat;
994 int auth_res;
995 __be32 *reply_statp;
996
997 rpc_stat = rpc_success;
998
999 if (argv->iov_len < 6*4)
1000 goto err_short_len;
1001
1002 /* Will be turned off only in gss privacy case: */
1003 rqstp->rq_splice_ok = 1;
1004 /* Will be turned off only when NFSv4 Sessions are used */
1005 rqstp->rq_usedeferral = 1;
1006 rqstp->rq_dropme = false;
1007
1008 /* Setup reply header */
1009 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1010
1011 svc_putu32(resv, rqstp->rq_xid);
1012
1013 vers = svc_getnl(argv);
1014
1015 /* First words of reply: */
1016 svc_putnl(resv, 1); /* REPLY */
1017
1018 if (vers != 2) /* RPC version number */
1019 goto err_bad_rpc;
1020
1021 /* Save position in case we later decide to reject: */
1022 reply_statp = resv->iov_base + resv->iov_len;
1023
1024 svc_putnl(resv, 0); /* ACCEPT */
1025
1026 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1027 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1028 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1029
1030 progp = serv->sv_program;
1031
1032 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1033 if (prog == progp->pg_prog)
1034 break;
1035
1036 /*
1037 * Decode auth data, and add verifier to reply buffer.
1038 * We do this before anything else in order to get a decent
1039 * auth verifier.
1040 */
1041 auth_res = svc_authenticate(rqstp, &auth_stat);
1042 /* Also give the program a chance to reject this call: */
1043 if (auth_res == SVC_OK && progp) {
1044 auth_stat = rpc_autherr_badcred;
1045 auth_res = progp->pg_authenticate(rqstp);
1046 }
1047 switch (auth_res) {
1048 case SVC_OK:
1049 break;
1050 case SVC_GARBAGE:
1051 goto err_garbage;
1052 case SVC_SYSERR:
1053 rpc_stat = rpc_system_err;
1054 goto err_bad;
1055 case SVC_DENIED:
1056 goto err_bad_auth;
1057 case SVC_CLOSE:
1058 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1059 svc_close_xprt(rqstp->rq_xprt);
1060 case SVC_DROP:
1061 goto dropit;
1062 case SVC_COMPLETE:
1063 goto sendit;
1064 }
1065
1066 if (progp == NULL)
1067 goto err_bad_prog;
1068
1069 if (vers >= progp->pg_nvers ||
1070 !(versp = progp->pg_vers[vers]))
1071 goto err_bad_vers;
1072
1073 procp = versp->vs_proc + proc;
1074 if (proc >= versp->vs_nproc || !procp->pc_func)
1075 goto err_bad_proc;
1076 rqstp->rq_procinfo = procp;
1077
1078 /* Syntactic check complete */
1079 serv->sv_stats->rpccnt++;
1080
1081 /* Build the reply header. */
1082 statp = resv->iov_base +resv->iov_len;
1083 svc_putnl(resv, RPC_SUCCESS);
1084
1085 /* Bump per-procedure stats counter */
1086 procp->pc_count++;
1087
1088 /* Initialize storage for argp and resp */
1089 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1090 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1091
1092 /* un-reserve some of the out-queue now that we have a
1093 * better idea of reply size
1094 */
1095 if (procp->pc_xdrressize)
1096 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1097
1098 /* Call the function that processes the request. */
1099 if (!versp->vs_dispatch) {
1100 /* Decode arguments */
1101 xdr = procp->pc_decode;
1102 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1103 goto err_garbage;
1104
1105 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1106
1107 /* Encode reply */
1108 if (rqstp->rq_dropme) {
1109 if (procp->pc_release)
1110 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1111 goto dropit;
1112 }
1113 if (*statp == rpc_success &&
1114 (xdr = procp->pc_encode) &&
1115 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1116 dprintk("svc: failed to encode reply\n");
1117 /* serv->sv_stats->rpcsystemerr++; */
1118 *statp = rpc_system_err;
1119 }
1120 } else {
1121 dprintk("svc: calling dispatcher\n");
1122 if (!versp->vs_dispatch(rqstp, statp)) {
1123 /* Release reply info */
1124 if (procp->pc_release)
1125 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1126 goto dropit;
1127 }
1128 }
1129
1130 /* Check RPC status result */
1131 if (*statp != rpc_success)
1132 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1133
1134 /* Release reply info */
1135 if (procp->pc_release)
1136 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1137
1138 if (procp->pc_encode == NULL)
1139 goto dropit;
1140
1141 sendit:
1142 if (svc_authorise(rqstp))
1143 goto dropit;
1144 return 1; /* Caller can now send it */
1145
1146 dropit:
1147 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1148 dprintk("svc: svc_process dropit\n");
1149 return 0;
1150
1151err_short_len:
1152 svc_printk(rqstp, "short len %Zd, dropping request\n",
1153 argv->iov_len);
1154
1155 goto dropit; /* drop request */
1156
1157err_bad_rpc:
1158 serv->sv_stats->rpcbadfmt++;
1159 svc_putnl(resv, 1); /* REJECT */
1160 svc_putnl(resv, 0); /* RPC_MISMATCH */
1161 svc_putnl(resv, 2); /* Only RPCv2 supported */
1162 svc_putnl(resv, 2);
1163 goto sendit;
1164
1165err_bad_auth:
1166 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1167 serv->sv_stats->rpcbadauth++;
1168 /* Restore write pointer to location of accept status: */
1169 xdr_ressize_check(rqstp, reply_statp);
1170 svc_putnl(resv, 1); /* REJECT */
1171 svc_putnl(resv, 1); /* AUTH_ERROR */
1172 svc_putnl(resv, ntohl(auth_stat)); /* status */
1173 goto sendit;
1174
1175err_bad_prog:
1176 dprintk("svc: unknown program %d\n", prog);
1177 serv->sv_stats->rpcbadfmt++;
1178 svc_putnl(resv, RPC_PROG_UNAVAIL);
1179 goto sendit;
1180
1181err_bad_vers:
1182 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1183 vers, prog, progp->pg_name);
1184
1185 serv->sv_stats->rpcbadfmt++;
1186 svc_putnl(resv, RPC_PROG_MISMATCH);
1187 svc_putnl(resv, progp->pg_lovers);
1188 svc_putnl(resv, progp->pg_hivers);
1189 goto sendit;
1190
1191err_bad_proc:
1192 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1193
1194 serv->sv_stats->rpcbadfmt++;
1195 svc_putnl(resv, RPC_PROC_UNAVAIL);
1196 goto sendit;
1197
1198err_garbage:
1199 svc_printk(rqstp, "failed to decode args\n");
1200
1201 rpc_stat = rpc_garbage_args;
1202err_bad:
1203 serv->sv_stats->rpcbadfmt++;
1204 svc_putnl(resv, ntohl(rpc_stat));
1205 goto sendit;
1206}
1207EXPORT_SYMBOL_GPL(svc_process);
1208
1209/*
1210 * Process the RPC request.
1211 */
1212int
1213svc_process(struct svc_rqst *rqstp)
1214{
1215 struct kvec *argv = &rqstp->rq_arg.head[0];
1216 struct kvec *resv = &rqstp->rq_res.head[0];
1217 struct svc_serv *serv = rqstp->rq_server;
1218 u32 dir;
1219
1220 /*
1221 * Setup response xdr_buf.
1222 * Initially it has just one page
1223 */
1224 rqstp->rq_resused = 1;
1225 resv->iov_base = page_address(rqstp->rq_respages[0]);
1226 resv->iov_len = 0;
1227 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1228 rqstp->rq_res.len = 0;
1229 rqstp->rq_res.page_base = 0;
1230 rqstp->rq_res.page_len = 0;
1231 rqstp->rq_res.buflen = PAGE_SIZE;
1232 rqstp->rq_res.tail[0].iov_base = NULL;
1233 rqstp->rq_res.tail[0].iov_len = 0;
1234
1235 rqstp->rq_xid = svc_getu32(argv);
1236
1237 dir = svc_getnl(argv);
1238 if (dir != 0) {
1239 /* direction != CALL */
1240 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1241 serv->sv_stats->rpcbadfmt++;
1242 svc_drop(rqstp);
1243 return 0;
1244 }
1245
1246 /* Returns 1 for send, 0 for drop */
1247 if (svc_process_common(rqstp, argv, resv))
1248 return svc_send(rqstp);
1249 else {
1250 svc_drop(rqstp);
1251 return 0;
1252 }
1253}
1254
1255#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1256/*
1257 * Process a backchannel RPC request that arrived over an existing
1258 * outbound connection
1259 */
1260int
1261bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1262 struct svc_rqst *rqstp)
1263{
1264 struct kvec *argv = &rqstp->rq_arg.head[0];
1265 struct kvec *resv = &rqstp->rq_res.head[0];
1266
1267 /* Build the svc_rqst used by the common processing routine */
1268 rqstp->rq_xprt = serv->sv_bc_xprt;
1269 rqstp->rq_xid = req->rq_xid;
1270 rqstp->rq_prot = req->rq_xprt->prot;
1271 rqstp->rq_server = serv;
1272
1273 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1274 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1275 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1276 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1277
1278 /* reset result send buffer "put" position */
1279 resv->iov_len = 0;
1280
1281 if (rqstp->rq_prot != IPPROTO_TCP) {
1282 printk(KERN_ERR "No support for Non-TCP transports!\n");
1283 BUG();
1284 }
1285
1286 /*
1287 * Skip the next two words because they've already been
1288 * processed in the trasport
1289 */
1290 svc_getu32(argv); /* XID */
1291 svc_getnl(argv); /* CALLDIR */
1292
1293 /* Returns 1 for send, 0 for drop */
1294 if (svc_process_common(rqstp, argv, resv)) {
1295 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1296 sizeof(req->rq_snd_buf));
1297 return bc_send(req);
1298 } else {
1299 /* Nothing to do to drop request */
1300 return 0;
1301 }
1302}
1303EXPORT_SYMBOL_GPL(bc_svc_process);
1304#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1305
1306/*
1307 * Return (transport-specific) limit on the rpc payload.
1308 */
1309u32 svc_max_payload(const struct svc_rqst *rqstp)
1310{
1311 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1312
1313 if (rqstp->rq_server->sv_max_payload < max)
1314 max = rqstp->rq_server->sv_max_payload;
1315 return max;
1316}
1317EXPORT_SYMBOL_GPL(svc_max_payload);