Loading...
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#include <trace/events/sunrpc.h>
32
33#define RPCDBG_FACILITY RPCDBG_SVCDSP
34
35static void svc_unregister(const struct svc_serv *serv, struct net *net);
36
37#define svc_serv_is_pooled(serv) ((serv)->sv_ops->svo_function)
38
39#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
40
41/*
42 * Structure for mapping cpus to pools and vice versa.
43 * Setup once during sunrpc initialisation.
44 */
45struct svc_pool_map svc_pool_map = {
46 .mode = SVC_POOL_DEFAULT
47};
48EXPORT_SYMBOL_GPL(svc_pool_map);
49
50static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
51
52static int
53param_set_pool_mode(const char *val, struct kernel_param *kp)
54{
55 int *ip = (int *)kp->arg;
56 struct svc_pool_map *m = &svc_pool_map;
57 int err;
58
59 mutex_lock(&svc_pool_map_mutex);
60
61 err = -EBUSY;
62 if (m->count)
63 goto out;
64
65 err = 0;
66 if (!strncmp(val, "auto", 4))
67 *ip = SVC_POOL_AUTO;
68 else if (!strncmp(val, "global", 6))
69 *ip = SVC_POOL_GLOBAL;
70 else if (!strncmp(val, "percpu", 6))
71 *ip = SVC_POOL_PERCPU;
72 else if (!strncmp(val, "pernode", 7))
73 *ip = SVC_POOL_PERNODE;
74 else
75 err = -EINVAL;
76
77out:
78 mutex_unlock(&svc_pool_map_mutex);
79 return err;
80}
81
82static int
83param_get_pool_mode(char *buf, struct kernel_param *kp)
84{
85 int *ip = (int *)kp->arg;
86
87 switch (*ip)
88 {
89 case SVC_POOL_AUTO:
90 return strlcpy(buf, "auto", 20);
91 case SVC_POOL_GLOBAL:
92 return strlcpy(buf, "global", 20);
93 case SVC_POOL_PERCPU:
94 return strlcpy(buf, "percpu", 20);
95 case SVC_POOL_PERNODE:
96 return strlcpy(buf, "pernode", 20);
97 default:
98 return sprintf(buf, "%d", *ip);
99 }
100}
101
102module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
103 &svc_pool_map.mode, 0644);
104
105/*
106 * Detect best pool mapping mode heuristically,
107 * according to the machine's topology.
108 */
109static int
110svc_pool_map_choose_mode(void)
111{
112 unsigned int node;
113
114 if (nr_online_nodes > 1) {
115 /*
116 * Actually have multiple NUMA nodes,
117 * so split pools on NUMA node boundaries
118 */
119 return SVC_POOL_PERNODE;
120 }
121
122 node = first_online_node;
123 if (nr_cpus_node(node) > 2) {
124 /*
125 * Non-trivial SMP, or CONFIG_NUMA on
126 * non-NUMA hardware, e.g. with a generic
127 * x86_64 kernel on Xeons. In this case we
128 * want to divide the pools on cpu boundaries.
129 */
130 return SVC_POOL_PERCPU;
131 }
132
133 /* default: one global pool */
134 return SVC_POOL_GLOBAL;
135}
136
137/*
138 * Allocate the to_pool[] and pool_to[] arrays.
139 * Returns 0 on success or an errno.
140 */
141static int
142svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
143{
144 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
145 if (!m->to_pool)
146 goto fail;
147 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
148 if (!m->pool_to)
149 goto fail_free;
150
151 return 0;
152
153fail_free:
154 kfree(m->to_pool);
155 m->to_pool = NULL;
156fail:
157 return -ENOMEM;
158}
159
160/*
161 * Initialise the pool map for SVC_POOL_PERCPU mode.
162 * Returns number of pools or <0 on error.
163 */
164static int
165svc_pool_map_init_percpu(struct svc_pool_map *m)
166{
167 unsigned int maxpools = nr_cpu_ids;
168 unsigned int pidx = 0;
169 unsigned int cpu;
170 int err;
171
172 err = svc_pool_map_alloc_arrays(m, maxpools);
173 if (err)
174 return err;
175
176 for_each_online_cpu(cpu) {
177 BUG_ON(pidx >= maxpools);
178 m->to_pool[cpu] = pidx;
179 m->pool_to[pidx] = cpu;
180 pidx++;
181 }
182 /* cpus brought online later all get mapped to pool0, sorry */
183
184 return pidx;
185};
186
187
188/*
189 * Initialise the pool map for SVC_POOL_PERNODE mode.
190 * Returns number of pools or <0 on error.
191 */
192static int
193svc_pool_map_init_pernode(struct svc_pool_map *m)
194{
195 unsigned int maxpools = nr_node_ids;
196 unsigned int pidx = 0;
197 unsigned int node;
198 int err;
199
200 err = svc_pool_map_alloc_arrays(m, maxpools);
201 if (err)
202 return err;
203
204 for_each_node_with_cpus(node) {
205 /* some architectures (e.g. SN2) have cpuless nodes */
206 BUG_ON(pidx > maxpools);
207 m->to_pool[node] = pidx;
208 m->pool_to[pidx] = node;
209 pidx++;
210 }
211 /* nodes brought online later all get mapped to pool0, sorry */
212
213 return pidx;
214}
215
216
217/*
218 * Add a reference to the global map of cpus to pools (and
219 * vice versa). Initialise the map if we're the first user.
220 * Returns the number of pools.
221 */
222unsigned int
223svc_pool_map_get(void)
224{
225 struct svc_pool_map *m = &svc_pool_map;
226 int npools = -1;
227
228 mutex_lock(&svc_pool_map_mutex);
229
230 if (m->count++) {
231 mutex_unlock(&svc_pool_map_mutex);
232 return m->npools;
233 }
234
235 if (m->mode == SVC_POOL_AUTO)
236 m->mode = svc_pool_map_choose_mode();
237
238 switch (m->mode) {
239 case SVC_POOL_PERCPU:
240 npools = svc_pool_map_init_percpu(m);
241 break;
242 case SVC_POOL_PERNODE:
243 npools = svc_pool_map_init_pernode(m);
244 break;
245 }
246
247 if (npools < 0) {
248 /* default, or memory allocation failure */
249 npools = 1;
250 m->mode = SVC_POOL_GLOBAL;
251 }
252 m->npools = npools;
253
254 mutex_unlock(&svc_pool_map_mutex);
255 return m->npools;
256}
257EXPORT_SYMBOL_GPL(svc_pool_map_get);
258
259/*
260 * Drop a reference to the global map of cpus to pools.
261 * When the last reference is dropped, the map data is
262 * freed; this allows the sysadmin to change the pool
263 * mode using the pool_mode module option without
264 * rebooting or re-loading sunrpc.ko.
265 */
266void
267svc_pool_map_put(void)
268{
269 struct svc_pool_map *m = &svc_pool_map;
270
271 mutex_lock(&svc_pool_map_mutex);
272
273 if (!--m->count) {
274 kfree(m->to_pool);
275 m->to_pool = NULL;
276 kfree(m->pool_to);
277 m->pool_to = NULL;
278 m->npools = 0;
279 }
280
281 mutex_unlock(&svc_pool_map_mutex);
282}
283EXPORT_SYMBOL_GPL(svc_pool_map_put);
284
285static int svc_pool_map_get_node(unsigned int pidx)
286{
287 const struct svc_pool_map *m = &svc_pool_map;
288
289 if (m->count) {
290 if (m->mode == SVC_POOL_PERCPU)
291 return cpu_to_node(m->pool_to[pidx]);
292 if (m->mode == SVC_POOL_PERNODE)
293 return m->pool_to[pidx];
294 }
295 return NUMA_NO_NODE;
296}
297/*
298 * Set the given thread's cpus_allowed mask so that it
299 * will only run on cpus in the given pool.
300 */
301static inline void
302svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
303{
304 struct svc_pool_map *m = &svc_pool_map;
305 unsigned int node = m->pool_to[pidx];
306
307 /*
308 * The caller checks for sv_nrpools > 1, which
309 * implies that we've been initialized.
310 */
311 WARN_ON_ONCE(m->count == 0);
312 if (m->count == 0)
313 return;
314
315 switch (m->mode) {
316 case SVC_POOL_PERCPU:
317 {
318 set_cpus_allowed_ptr(task, cpumask_of(node));
319 break;
320 }
321 case SVC_POOL_PERNODE:
322 {
323 set_cpus_allowed_ptr(task, cpumask_of_node(node));
324 break;
325 }
326 }
327}
328
329/*
330 * Use the mapping mode to choose a pool for a given CPU.
331 * Used when enqueueing an incoming RPC. Always returns
332 * a non-NULL pool pointer.
333 */
334struct svc_pool *
335svc_pool_for_cpu(struct svc_serv *serv, int cpu)
336{
337 struct svc_pool_map *m = &svc_pool_map;
338 unsigned int pidx = 0;
339
340 /*
341 * An uninitialised map happens in a pure client when
342 * lockd is brought up, so silently treat it the
343 * same as SVC_POOL_GLOBAL.
344 */
345 if (svc_serv_is_pooled(serv)) {
346 switch (m->mode) {
347 case SVC_POOL_PERCPU:
348 pidx = m->to_pool[cpu];
349 break;
350 case SVC_POOL_PERNODE:
351 pidx = m->to_pool[cpu_to_node(cpu)];
352 break;
353 }
354 }
355 return &serv->sv_pools[pidx % serv->sv_nrpools];
356}
357
358int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
359{
360 int err;
361
362 err = rpcb_create_local(net);
363 if (err)
364 return err;
365
366 /* Remove any stale portmap registrations */
367 svc_unregister(serv, net);
368 return 0;
369}
370EXPORT_SYMBOL_GPL(svc_rpcb_setup);
371
372void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
373{
374 svc_unregister(serv, net);
375 rpcb_put_local(net);
376}
377EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
378
379static int svc_uses_rpcbind(struct svc_serv *serv)
380{
381 struct svc_program *progp;
382 unsigned int i;
383
384 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
385 for (i = 0; i < progp->pg_nvers; i++) {
386 if (progp->pg_vers[i] == NULL)
387 continue;
388 if (progp->pg_vers[i]->vs_hidden == 0)
389 return 1;
390 }
391 }
392
393 return 0;
394}
395
396int svc_bind(struct svc_serv *serv, struct net *net)
397{
398 if (!svc_uses_rpcbind(serv))
399 return 0;
400 return svc_rpcb_setup(serv, net);
401}
402EXPORT_SYMBOL_GPL(svc_bind);
403
404/*
405 * Create an RPC service
406 */
407static struct svc_serv *
408__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
409 struct svc_serv_ops *ops)
410{
411 struct svc_serv *serv;
412 unsigned int vers;
413 unsigned int xdrsize;
414 unsigned int i;
415
416 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
417 return NULL;
418 serv->sv_name = prog->pg_name;
419 serv->sv_program = prog;
420 serv->sv_nrthreads = 1;
421 serv->sv_stats = prog->pg_stats;
422 if (bufsize > RPCSVC_MAXPAYLOAD)
423 bufsize = RPCSVC_MAXPAYLOAD;
424 serv->sv_max_payload = bufsize? bufsize : 4096;
425 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
426 serv->sv_ops = ops;
427 xdrsize = 0;
428 while (prog) {
429 prog->pg_lovers = prog->pg_nvers-1;
430 for (vers=0; vers<prog->pg_nvers ; vers++)
431 if (prog->pg_vers[vers]) {
432 prog->pg_hivers = vers;
433 if (prog->pg_lovers > vers)
434 prog->pg_lovers = vers;
435 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
436 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
437 }
438 prog = prog->pg_next;
439 }
440 serv->sv_xdrsize = xdrsize;
441 INIT_LIST_HEAD(&serv->sv_tempsocks);
442 INIT_LIST_HEAD(&serv->sv_permsocks);
443 init_timer(&serv->sv_temptimer);
444 spin_lock_init(&serv->sv_lock);
445
446 serv->sv_nrpools = npools;
447 serv->sv_pools =
448 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
449 GFP_KERNEL);
450 if (!serv->sv_pools) {
451 kfree(serv);
452 return NULL;
453 }
454
455 for (i = 0; i < serv->sv_nrpools; i++) {
456 struct svc_pool *pool = &serv->sv_pools[i];
457
458 dprintk("svc: initialising pool %u for %s\n",
459 i, serv->sv_name);
460
461 pool->sp_id = i;
462 INIT_LIST_HEAD(&pool->sp_sockets);
463 INIT_LIST_HEAD(&pool->sp_all_threads);
464 spin_lock_init(&pool->sp_lock);
465 }
466
467 return serv;
468}
469
470struct svc_serv *
471svc_create(struct svc_program *prog, unsigned int bufsize,
472 struct svc_serv_ops *ops)
473{
474 return __svc_create(prog, bufsize, /*npools*/1, ops);
475}
476EXPORT_SYMBOL_GPL(svc_create);
477
478struct svc_serv *
479svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
480 struct svc_serv_ops *ops)
481{
482 struct svc_serv *serv;
483 unsigned int npools = svc_pool_map_get();
484
485 serv = __svc_create(prog, bufsize, npools, ops);
486 if (!serv)
487 goto out_err;
488 return serv;
489out_err:
490 svc_pool_map_put();
491 return NULL;
492}
493EXPORT_SYMBOL_GPL(svc_create_pooled);
494
495void svc_shutdown_net(struct svc_serv *serv, struct net *net)
496{
497 svc_close_net(serv, net);
498
499 if (serv->sv_ops->svo_shutdown)
500 serv->sv_ops->svo_shutdown(serv, net);
501}
502EXPORT_SYMBOL_GPL(svc_shutdown_net);
503
504/*
505 * Destroy an RPC service. Should be called with appropriate locking to
506 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
507 */
508void
509svc_destroy(struct svc_serv *serv)
510{
511 dprintk("svc: svc_destroy(%s, %d)\n",
512 serv->sv_program->pg_name,
513 serv->sv_nrthreads);
514
515 if (serv->sv_nrthreads) {
516 if (--(serv->sv_nrthreads) != 0) {
517 svc_sock_update_bufs(serv);
518 return;
519 }
520 } else
521 printk("svc_destroy: no threads for serv=%p!\n", serv);
522
523 del_timer_sync(&serv->sv_temptimer);
524
525 /*
526 * The last user is gone and thus all sockets have to be destroyed to
527 * the point. Check this.
528 */
529 BUG_ON(!list_empty(&serv->sv_permsocks));
530 BUG_ON(!list_empty(&serv->sv_tempsocks));
531
532 cache_clean_deferred(serv);
533
534 if (svc_serv_is_pooled(serv))
535 svc_pool_map_put();
536
537 kfree(serv->sv_pools);
538 kfree(serv);
539}
540EXPORT_SYMBOL_GPL(svc_destroy);
541
542/*
543 * Allocate an RPC server's buffer space.
544 * We allocate pages and place them in rq_argpages.
545 */
546static int
547svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
548{
549 unsigned int pages, arghi;
550
551 /* bc_xprt uses fore channel allocated buffers */
552 if (svc_is_backchannel(rqstp))
553 return 1;
554
555 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
556 * We assume one is at most one page
557 */
558 arghi = 0;
559 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
560 if (pages > RPCSVC_MAXPAGES)
561 pages = RPCSVC_MAXPAGES;
562 while (pages) {
563 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
564 if (!p)
565 break;
566 rqstp->rq_pages[arghi++] = p;
567 pages--;
568 }
569 return pages == 0;
570}
571
572/*
573 * Release an RPC server buffer
574 */
575static void
576svc_release_buffer(struct svc_rqst *rqstp)
577{
578 unsigned int i;
579
580 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
581 if (rqstp->rq_pages[i])
582 put_page(rqstp->rq_pages[i]);
583}
584
585struct svc_rqst *
586svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
587{
588 struct svc_rqst *rqstp;
589
590 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
591 if (!rqstp)
592 return rqstp;
593
594 __set_bit(RQ_BUSY, &rqstp->rq_flags);
595 spin_lock_init(&rqstp->rq_lock);
596 rqstp->rq_server = serv;
597 rqstp->rq_pool = pool;
598
599 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
600 if (!rqstp->rq_argp)
601 goto out_enomem;
602
603 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
604 if (!rqstp->rq_resp)
605 goto out_enomem;
606
607 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
608 goto out_enomem;
609
610 return rqstp;
611out_enomem:
612 svc_rqst_free(rqstp);
613 return NULL;
614}
615EXPORT_SYMBOL_GPL(svc_rqst_alloc);
616
617struct svc_rqst *
618svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
619{
620 struct svc_rqst *rqstp;
621
622 rqstp = svc_rqst_alloc(serv, pool, node);
623 if (!rqstp)
624 return ERR_PTR(-ENOMEM);
625
626 serv->sv_nrthreads++;
627 spin_lock_bh(&pool->sp_lock);
628 pool->sp_nrthreads++;
629 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
630 spin_unlock_bh(&pool->sp_lock);
631 return rqstp;
632}
633EXPORT_SYMBOL_GPL(svc_prepare_thread);
634
635/*
636 * Choose a pool in which to create a new thread, for svc_set_num_threads
637 */
638static inline struct svc_pool *
639choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
640{
641 if (pool != NULL)
642 return pool;
643
644 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
645}
646
647/*
648 * Choose a thread to kill, for svc_set_num_threads
649 */
650static inline struct task_struct *
651choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
652{
653 unsigned int i;
654 struct task_struct *task = NULL;
655
656 if (pool != NULL) {
657 spin_lock_bh(&pool->sp_lock);
658 } else {
659 /* choose a pool in round-robin fashion */
660 for (i = 0; i < serv->sv_nrpools; i++) {
661 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
662 spin_lock_bh(&pool->sp_lock);
663 if (!list_empty(&pool->sp_all_threads))
664 goto found_pool;
665 spin_unlock_bh(&pool->sp_lock);
666 }
667 return NULL;
668 }
669
670found_pool:
671 if (!list_empty(&pool->sp_all_threads)) {
672 struct svc_rqst *rqstp;
673
674 /*
675 * Remove from the pool->sp_all_threads list
676 * so we don't try to kill it again.
677 */
678 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
679 set_bit(RQ_VICTIM, &rqstp->rq_flags);
680 list_del_rcu(&rqstp->rq_all);
681 task = rqstp->rq_task;
682 }
683 spin_unlock_bh(&pool->sp_lock);
684
685 return task;
686}
687
688/*
689 * Create or destroy enough new threads to make the number
690 * of threads the given number. If `pool' is non-NULL, applies
691 * only to threads in that pool, otherwise round-robins between
692 * all pools. Caller must ensure that mutual exclusion between this and
693 * server startup or shutdown.
694 *
695 * Destroying threads relies on the service threads filling in
696 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
697 * has been created using svc_create_pooled().
698 *
699 * Based on code that used to be in nfsd_svc() but tweaked
700 * to be pool-aware.
701 */
702int
703svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
704{
705 struct svc_rqst *rqstp;
706 struct task_struct *task;
707 struct svc_pool *chosen_pool;
708 int error = 0;
709 unsigned int state = serv->sv_nrthreads-1;
710 int node;
711
712 if (pool == NULL) {
713 /* The -1 assumes caller has done a svc_get() */
714 nrservs -= (serv->sv_nrthreads-1);
715 } else {
716 spin_lock_bh(&pool->sp_lock);
717 nrservs -= pool->sp_nrthreads;
718 spin_unlock_bh(&pool->sp_lock);
719 }
720
721 /* create new threads */
722 while (nrservs > 0) {
723 nrservs--;
724 chosen_pool = choose_pool(serv, pool, &state);
725
726 node = svc_pool_map_get_node(chosen_pool->sp_id);
727 rqstp = svc_prepare_thread(serv, chosen_pool, node);
728 if (IS_ERR(rqstp)) {
729 error = PTR_ERR(rqstp);
730 break;
731 }
732
733 __module_get(serv->sv_ops->svo_module);
734 task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
735 node, "%s", serv->sv_name);
736 if (IS_ERR(task)) {
737 error = PTR_ERR(task);
738 module_put(serv->sv_ops->svo_module);
739 svc_exit_thread(rqstp);
740 break;
741 }
742
743 rqstp->rq_task = task;
744 if (serv->sv_nrpools > 1)
745 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
746
747 svc_sock_update_bufs(serv);
748 wake_up_process(task);
749 }
750 /* destroy old threads */
751 while (nrservs < 0 &&
752 (task = choose_victim(serv, pool, &state)) != NULL) {
753 send_sig(SIGINT, task, 1);
754 nrservs++;
755 }
756
757 return error;
758}
759EXPORT_SYMBOL_GPL(svc_set_num_threads);
760
761/*
762 * Called from a server thread as it's exiting. Caller must hold the "service
763 * mutex" for the service.
764 */
765void
766svc_rqst_free(struct svc_rqst *rqstp)
767{
768 svc_release_buffer(rqstp);
769 kfree(rqstp->rq_resp);
770 kfree(rqstp->rq_argp);
771 kfree(rqstp->rq_auth_data);
772 kfree_rcu(rqstp, rq_rcu_head);
773}
774EXPORT_SYMBOL_GPL(svc_rqst_free);
775
776void
777svc_exit_thread(struct svc_rqst *rqstp)
778{
779 struct svc_serv *serv = rqstp->rq_server;
780 struct svc_pool *pool = rqstp->rq_pool;
781
782 spin_lock_bh(&pool->sp_lock);
783 pool->sp_nrthreads--;
784 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
785 list_del_rcu(&rqstp->rq_all);
786 spin_unlock_bh(&pool->sp_lock);
787
788 svc_rqst_free(rqstp);
789
790 /* Release the server */
791 if (serv)
792 svc_destroy(serv);
793}
794EXPORT_SYMBOL_GPL(svc_exit_thread);
795
796/*
797 * Register an "inet" protocol family netid with the local
798 * rpcbind daemon via an rpcbind v4 SET request.
799 *
800 * No netconfig infrastructure is available in the kernel, so
801 * we map IP_ protocol numbers to netids by hand.
802 *
803 * Returns zero on success; a negative errno value is returned
804 * if any error occurs.
805 */
806static int __svc_rpcb_register4(struct net *net, const u32 program,
807 const u32 version,
808 const unsigned short protocol,
809 const unsigned short port)
810{
811 const struct sockaddr_in sin = {
812 .sin_family = AF_INET,
813 .sin_addr.s_addr = htonl(INADDR_ANY),
814 .sin_port = htons(port),
815 };
816 const char *netid;
817 int error;
818
819 switch (protocol) {
820 case IPPROTO_UDP:
821 netid = RPCBIND_NETID_UDP;
822 break;
823 case IPPROTO_TCP:
824 netid = RPCBIND_NETID_TCP;
825 break;
826 default:
827 return -ENOPROTOOPT;
828 }
829
830 error = rpcb_v4_register(net, program, version,
831 (const struct sockaddr *)&sin, netid);
832
833 /*
834 * User space didn't support rpcbind v4, so retry this
835 * registration request with the legacy rpcbind v2 protocol.
836 */
837 if (error == -EPROTONOSUPPORT)
838 error = rpcb_register(net, program, version, protocol, port);
839
840 return error;
841}
842
843#if IS_ENABLED(CONFIG_IPV6)
844/*
845 * Register an "inet6" protocol family netid with the local
846 * rpcbind daemon via an rpcbind v4 SET request.
847 *
848 * No netconfig infrastructure is available in the kernel, so
849 * we map IP_ protocol numbers to netids by hand.
850 *
851 * Returns zero on success; a negative errno value is returned
852 * if any error occurs.
853 */
854static int __svc_rpcb_register6(struct net *net, const u32 program,
855 const u32 version,
856 const unsigned short protocol,
857 const unsigned short port)
858{
859 const struct sockaddr_in6 sin6 = {
860 .sin6_family = AF_INET6,
861 .sin6_addr = IN6ADDR_ANY_INIT,
862 .sin6_port = htons(port),
863 };
864 const char *netid;
865 int error;
866
867 switch (protocol) {
868 case IPPROTO_UDP:
869 netid = RPCBIND_NETID_UDP6;
870 break;
871 case IPPROTO_TCP:
872 netid = RPCBIND_NETID_TCP6;
873 break;
874 default:
875 return -ENOPROTOOPT;
876 }
877
878 error = rpcb_v4_register(net, program, version,
879 (const struct sockaddr *)&sin6, netid);
880
881 /*
882 * User space didn't support rpcbind version 4, so we won't
883 * use a PF_INET6 listener.
884 */
885 if (error == -EPROTONOSUPPORT)
886 error = -EAFNOSUPPORT;
887
888 return error;
889}
890#endif /* IS_ENABLED(CONFIG_IPV6) */
891
892/*
893 * Register a kernel RPC service via rpcbind version 4.
894 *
895 * Returns zero on success; a negative errno value is returned
896 * if any error occurs.
897 */
898static int __svc_register(struct net *net, const char *progname,
899 const u32 program, const u32 version,
900 const int family,
901 const unsigned short protocol,
902 const unsigned short port)
903{
904 int error = -EAFNOSUPPORT;
905
906 switch (family) {
907 case PF_INET:
908 error = __svc_rpcb_register4(net, program, version,
909 protocol, port);
910 break;
911#if IS_ENABLED(CONFIG_IPV6)
912 case PF_INET6:
913 error = __svc_rpcb_register6(net, program, version,
914 protocol, port);
915#endif
916 }
917
918 return error;
919}
920
921/**
922 * svc_register - register an RPC service with the local portmapper
923 * @serv: svc_serv struct for the service to register
924 * @net: net namespace for the service to register
925 * @family: protocol family of service's listener socket
926 * @proto: transport protocol number to advertise
927 * @port: port to advertise
928 *
929 * Service is registered for any address in the passed-in protocol family
930 */
931int svc_register(const struct svc_serv *serv, struct net *net,
932 const int family, const unsigned short proto,
933 const unsigned short port)
934{
935 struct svc_program *progp;
936 struct svc_version *vers;
937 unsigned int i;
938 int error = 0;
939
940 WARN_ON_ONCE(proto == 0 && port == 0);
941 if (proto == 0 && port == 0)
942 return -EINVAL;
943
944 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
945 for (i = 0; i < progp->pg_nvers; i++) {
946 vers = progp->pg_vers[i];
947 if (vers == NULL)
948 continue;
949
950 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
951 progp->pg_name,
952 i,
953 proto == IPPROTO_UDP? "udp" : "tcp",
954 port,
955 family,
956 vers->vs_hidden ?
957 " (but not telling portmap)" : "");
958
959 if (vers->vs_hidden)
960 continue;
961
962 error = __svc_register(net, progp->pg_name, progp->pg_prog,
963 i, family, proto, port);
964
965 if (vers->vs_rpcb_optnl) {
966 error = 0;
967 continue;
968 }
969
970 if (error < 0) {
971 printk(KERN_WARNING "svc: failed to register "
972 "%sv%u RPC service (errno %d).\n",
973 progp->pg_name, i, -error);
974 break;
975 }
976 }
977 }
978
979 return error;
980}
981
982/*
983 * If user space is running rpcbind, it should take the v4 UNSET
984 * and clear everything for this [program, version]. If user space
985 * is running portmap, it will reject the v4 UNSET, but won't have
986 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
987 * in this case to clear all existing entries for [program, version].
988 */
989static void __svc_unregister(struct net *net, const u32 program, const u32 version,
990 const char *progname)
991{
992 int error;
993
994 error = rpcb_v4_register(net, program, version, NULL, "");
995
996 /*
997 * User space didn't support rpcbind v4, so retry this
998 * request with the legacy rpcbind v2 protocol.
999 */
1000 if (error == -EPROTONOSUPPORT)
1001 error = rpcb_register(net, program, version, 0, 0);
1002
1003 dprintk("svc: %s(%sv%u), error %d\n",
1004 __func__, progname, version, error);
1005}
1006
1007/*
1008 * All netids, bind addresses and ports registered for [program, version]
1009 * are removed from the local rpcbind database (if the service is not
1010 * hidden) to make way for a new instance of the service.
1011 *
1012 * The result of unregistration is reported via dprintk for those who want
1013 * verification of the result, but is otherwise not important.
1014 */
1015static void svc_unregister(const struct svc_serv *serv, struct net *net)
1016{
1017 struct svc_program *progp;
1018 unsigned long flags;
1019 unsigned int i;
1020
1021 clear_thread_flag(TIF_SIGPENDING);
1022
1023 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1024 for (i = 0; i < progp->pg_nvers; i++) {
1025 if (progp->pg_vers[i] == NULL)
1026 continue;
1027 if (progp->pg_vers[i]->vs_hidden)
1028 continue;
1029
1030 dprintk("svc: attempting to unregister %sv%u\n",
1031 progp->pg_name, i);
1032 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1033 }
1034 }
1035
1036 spin_lock_irqsave(¤t->sighand->siglock, flags);
1037 recalc_sigpending();
1038 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1039}
1040
1041/*
1042 * dprintk the given error with the address of the client that caused it.
1043 */
1044#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1045static __printf(2, 3)
1046void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1047{
1048 struct va_format vaf;
1049 va_list args;
1050 char buf[RPC_MAX_ADDRBUFLEN];
1051
1052 va_start(args, fmt);
1053
1054 vaf.fmt = fmt;
1055 vaf.va = &args;
1056
1057 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1058
1059 va_end(args);
1060}
1061#else
1062static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1063#endif
1064
1065/*
1066 * Common routine for processing the RPC request.
1067 */
1068static int
1069svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1070{
1071 struct svc_program *progp;
1072 struct svc_version *versp = NULL; /* compiler food */
1073 struct svc_procedure *procp = NULL;
1074 struct svc_serv *serv = rqstp->rq_server;
1075 kxdrproc_t xdr;
1076 __be32 *statp;
1077 u32 prog, vers, proc;
1078 __be32 auth_stat, rpc_stat;
1079 int auth_res;
1080 __be32 *reply_statp;
1081
1082 rpc_stat = rpc_success;
1083
1084 if (argv->iov_len < 6*4)
1085 goto err_short_len;
1086
1087 /* Will be turned off only in gss privacy case: */
1088 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1089 /* Will be turned off only when NFSv4 Sessions are used */
1090 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1091 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1092
1093 /* Setup reply header */
1094 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1095
1096 svc_putu32(resv, rqstp->rq_xid);
1097
1098 vers = svc_getnl(argv);
1099
1100 /* First words of reply: */
1101 svc_putnl(resv, 1); /* REPLY */
1102
1103 if (vers != 2) /* RPC version number */
1104 goto err_bad_rpc;
1105
1106 /* Save position in case we later decide to reject: */
1107 reply_statp = resv->iov_base + resv->iov_len;
1108
1109 svc_putnl(resv, 0); /* ACCEPT */
1110
1111 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1112 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1113 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
1114
1115 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1116 if (prog == progp->pg_prog)
1117 break;
1118
1119 /*
1120 * Decode auth data, and add verifier to reply buffer.
1121 * We do this before anything else in order to get a decent
1122 * auth verifier.
1123 */
1124 auth_res = svc_authenticate(rqstp, &auth_stat);
1125 /* Also give the program a chance to reject this call: */
1126 if (auth_res == SVC_OK && progp) {
1127 auth_stat = rpc_autherr_badcred;
1128 auth_res = progp->pg_authenticate(rqstp);
1129 }
1130 switch (auth_res) {
1131 case SVC_OK:
1132 break;
1133 case SVC_GARBAGE:
1134 goto err_garbage;
1135 case SVC_SYSERR:
1136 rpc_stat = rpc_system_err;
1137 goto err_bad;
1138 case SVC_DENIED:
1139 goto err_bad_auth;
1140 case SVC_CLOSE:
1141 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1142 svc_close_xprt(rqstp->rq_xprt);
1143 case SVC_DROP:
1144 goto dropit;
1145 case SVC_COMPLETE:
1146 goto sendit;
1147 }
1148
1149 if (progp == NULL)
1150 goto err_bad_prog;
1151
1152 if (vers >= progp->pg_nvers ||
1153 !(versp = progp->pg_vers[vers]))
1154 goto err_bad_vers;
1155
1156 procp = versp->vs_proc + proc;
1157 if (proc >= versp->vs_nproc || !procp->pc_func)
1158 goto err_bad_proc;
1159 rqstp->rq_procinfo = procp;
1160
1161 /* Syntactic check complete */
1162 serv->sv_stats->rpccnt++;
1163
1164 /* Build the reply header. */
1165 statp = resv->iov_base +resv->iov_len;
1166 svc_putnl(resv, RPC_SUCCESS);
1167
1168 /* Bump per-procedure stats counter */
1169 procp->pc_count++;
1170
1171 /* Initialize storage for argp and resp */
1172 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1173 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1174
1175 /* un-reserve some of the out-queue now that we have a
1176 * better idea of reply size
1177 */
1178 if (procp->pc_xdrressize)
1179 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1180
1181 /* Call the function that processes the request. */
1182 if (!versp->vs_dispatch) {
1183 /* Decode arguments */
1184 xdr = procp->pc_decode;
1185 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1186 goto err_garbage;
1187
1188 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1189
1190 /* Encode reply */
1191 if (test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1192 if (procp->pc_release)
1193 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1194 goto dropit;
1195 }
1196 if (*statp == rpc_success &&
1197 (xdr = procp->pc_encode) &&
1198 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1199 dprintk("svc: failed to encode reply\n");
1200 /* serv->sv_stats->rpcsystemerr++; */
1201 *statp = rpc_system_err;
1202 }
1203 } else {
1204 dprintk("svc: calling dispatcher\n");
1205 if (!versp->vs_dispatch(rqstp, statp)) {
1206 /* Release reply info */
1207 if (procp->pc_release)
1208 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1209 goto dropit;
1210 }
1211 }
1212
1213 /* Check RPC status result */
1214 if (*statp != rpc_success)
1215 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1216
1217 /* Release reply info */
1218 if (procp->pc_release)
1219 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1220
1221 if (procp->pc_encode == NULL)
1222 goto dropit;
1223
1224 sendit:
1225 if (svc_authorise(rqstp))
1226 goto dropit;
1227 return 1; /* Caller can now send it */
1228
1229 dropit:
1230 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1231 dprintk("svc: svc_process dropit\n");
1232 return 0;
1233
1234err_short_len:
1235 svc_printk(rqstp, "short len %Zd, dropping request\n",
1236 argv->iov_len);
1237
1238 goto dropit; /* drop request */
1239
1240err_bad_rpc:
1241 serv->sv_stats->rpcbadfmt++;
1242 svc_putnl(resv, 1); /* REJECT */
1243 svc_putnl(resv, 0); /* RPC_MISMATCH */
1244 svc_putnl(resv, 2); /* Only RPCv2 supported */
1245 svc_putnl(resv, 2);
1246 goto sendit;
1247
1248err_bad_auth:
1249 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1250 serv->sv_stats->rpcbadauth++;
1251 /* Restore write pointer to location of accept status: */
1252 xdr_ressize_check(rqstp, reply_statp);
1253 svc_putnl(resv, 1); /* REJECT */
1254 svc_putnl(resv, 1); /* AUTH_ERROR */
1255 svc_putnl(resv, ntohl(auth_stat)); /* status */
1256 goto sendit;
1257
1258err_bad_prog:
1259 dprintk("svc: unknown program %d\n", prog);
1260 serv->sv_stats->rpcbadfmt++;
1261 svc_putnl(resv, RPC_PROG_UNAVAIL);
1262 goto sendit;
1263
1264err_bad_vers:
1265 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1266 vers, prog, progp->pg_name);
1267
1268 serv->sv_stats->rpcbadfmt++;
1269 svc_putnl(resv, RPC_PROG_MISMATCH);
1270 svc_putnl(resv, progp->pg_lovers);
1271 svc_putnl(resv, progp->pg_hivers);
1272 goto sendit;
1273
1274err_bad_proc:
1275 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1276
1277 serv->sv_stats->rpcbadfmt++;
1278 svc_putnl(resv, RPC_PROC_UNAVAIL);
1279 goto sendit;
1280
1281err_garbage:
1282 svc_printk(rqstp, "failed to decode args\n");
1283
1284 rpc_stat = rpc_garbage_args;
1285err_bad:
1286 serv->sv_stats->rpcbadfmt++;
1287 svc_putnl(resv, ntohl(rpc_stat));
1288 goto sendit;
1289}
1290
1291/*
1292 * Process the RPC request.
1293 */
1294int
1295svc_process(struct svc_rqst *rqstp)
1296{
1297 struct kvec *argv = &rqstp->rq_arg.head[0];
1298 struct kvec *resv = &rqstp->rq_res.head[0];
1299 struct svc_serv *serv = rqstp->rq_server;
1300 u32 dir;
1301
1302 /*
1303 * Setup response xdr_buf.
1304 * Initially it has just one page
1305 */
1306 rqstp->rq_next_page = &rqstp->rq_respages[1];
1307 resv->iov_base = page_address(rqstp->rq_respages[0]);
1308 resv->iov_len = 0;
1309 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1310 rqstp->rq_res.len = 0;
1311 rqstp->rq_res.page_base = 0;
1312 rqstp->rq_res.page_len = 0;
1313 rqstp->rq_res.buflen = PAGE_SIZE;
1314 rqstp->rq_res.tail[0].iov_base = NULL;
1315 rqstp->rq_res.tail[0].iov_len = 0;
1316
1317 dir = svc_getnl(argv);
1318 if (dir != 0) {
1319 /* direction != CALL */
1320 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1321 serv->sv_stats->rpcbadfmt++;
1322 goto out_drop;
1323 }
1324
1325 /* Returns 1 for send, 0 for drop */
1326 if (likely(svc_process_common(rqstp, argv, resv))) {
1327 int ret = svc_send(rqstp);
1328
1329 trace_svc_process(rqstp, ret);
1330 return ret;
1331 }
1332out_drop:
1333 trace_svc_process(rqstp, 0);
1334 svc_drop(rqstp);
1335 return 0;
1336}
1337EXPORT_SYMBOL_GPL(svc_process);
1338
1339#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1340/*
1341 * Process a backchannel RPC request that arrived over an existing
1342 * outbound connection
1343 */
1344int
1345bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1346 struct svc_rqst *rqstp)
1347{
1348 struct kvec *argv = &rqstp->rq_arg.head[0];
1349 struct kvec *resv = &rqstp->rq_res.head[0];
1350 struct rpc_task *task;
1351 int proc_error;
1352 int error;
1353
1354 dprintk("svc: %s(%p)\n", __func__, req);
1355
1356 /* Build the svc_rqst used by the common processing routine */
1357 rqstp->rq_xprt = serv->sv_bc_xprt;
1358 rqstp->rq_xid = req->rq_xid;
1359 rqstp->rq_prot = req->rq_xprt->prot;
1360 rqstp->rq_server = serv;
1361
1362 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1363 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1364 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1365 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1366
1367 /* Adjust the argument buffer length */
1368 rqstp->rq_arg.len = req->rq_private_buf.len;
1369 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1370 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1371 rqstp->rq_arg.page_len = 0;
1372 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1373 rqstp->rq_arg.page_len)
1374 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1375 rqstp->rq_arg.head[0].iov_len;
1376 else
1377 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1378 rqstp->rq_arg.page_len;
1379
1380 /* reset result send buffer "put" position */
1381 resv->iov_len = 0;
1382
1383 /*
1384 * Skip the next two words because they've already been
1385 * processed in the transport
1386 */
1387 svc_getu32(argv); /* XID */
1388 svc_getnl(argv); /* CALLDIR */
1389
1390 /* Parse and execute the bc call */
1391 proc_error = svc_process_common(rqstp, argv, resv);
1392
1393 atomic_inc(&req->rq_xprt->bc_free_slots);
1394 if (!proc_error) {
1395 /* Processing error: drop the request */
1396 xprt_free_bc_request(req);
1397 return 0;
1398 }
1399
1400 /* Finally, send the reply synchronously */
1401 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1402 task = rpc_run_bc_task(req);
1403 if (IS_ERR(task)) {
1404 error = PTR_ERR(task);
1405 goto out;
1406 }
1407
1408 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1409 error = task->tk_status;
1410 rpc_put_task(task);
1411
1412out:
1413 dprintk("svc: %s(), error=%d\n", __func__, error);
1414 return error;
1415}
1416EXPORT_SYMBOL_GPL(bc_svc_process);
1417#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1418
1419/*
1420 * Return (transport-specific) limit on the rpc payload.
1421 */
1422u32 svc_max_payload(const struct svc_rqst *rqstp)
1423{
1424 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1425
1426 if (rqstp->rq_server->sv_max_payload < max)
1427 max = rqstp->rq_server->sv_max_payload;
1428 return max;
1429}
1430EXPORT_SYMBOL_GPL(svc_max_payload);
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\n", 20);
92 case SVC_POOL_GLOBAL:
93 return strlcpy(buf, "global\n", 20);
94 case SVC_POOL_PERCPU:
95 return strlcpy(buf, "percpu\n", 20);
96 case SVC_POOL_PERNODE:
97 return strlcpy(buf, "pernode\n", 20);
98 default:
99 return sprintf(buf, "%d\n", *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_pages.
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_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
618 if (!rqstp->rq_scratch_page)
619 goto out_enomem;
620
621 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
622 if (!rqstp->rq_argp)
623 goto out_enomem;
624
625 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
626 if (!rqstp->rq_resp)
627 goto out_enomem;
628
629 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
630 goto out_enomem;
631
632 return rqstp;
633out_enomem:
634 svc_rqst_free(rqstp);
635 return NULL;
636}
637EXPORT_SYMBOL_GPL(svc_rqst_alloc);
638
639struct svc_rqst *
640svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
641{
642 struct svc_rqst *rqstp;
643
644 rqstp = svc_rqst_alloc(serv, pool, node);
645 if (!rqstp)
646 return ERR_PTR(-ENOMEM);
647
648 serv->sv_nrthreads++;
649 spin_lock_bh(&pool->sp_lock);
650 pool->sp_nrthreads++;
651 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
652 spin_unlock_bh(&pool->sp_lock);
653 return rqstp;
654}
655EXPORT_SYMBOL_GPL(svc_prepare_thread);
656
657/*
658 * Choose a pool in which to create a new thread, for svc_set_num_threads
659 */
660static inline struct svc_pool *
661choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
662{
663 if (pool != NULL)
664 return pool;
665
666 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
667}
668
669/*
670 * Choose a thread to kill, for svc_set_num_threads
671 */
672static inline struct task_struct *
673choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
674{
675 unsigned int i;
676 struct task_struct *task = NULL;
677
678 if (pool != NULL) {
679 spin_lock_bh(&pool->sp_lock);
680 } else {
681 /* choose a pool in round-robin fashion */
682 for (i = 0; i < serv->sv_nrpools; i++) {
683 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
684 spin_lock_bh(&pool->sp_lock);
685 if (!list_empty(&pool->sp_all_threads))
686 goto found_pool;
687 spin_unlock_bh(&pool->sp_lock);
688 }
689 return NULL;
690 }
691
692found_pool:
693 if (!list_empty(&pool->sp_all_threads)) {
694 struct svc_rqst *rqstp;
695
696 /*
697 * Remove from the pool->sp_all_threads list
698 * so we don't try to kill it again.
699 */
700 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
701 set_bit(RQ_VICTIM, &rqstp->rq_flags);
702 list_del_rcu(&rqstp->rq_all);
703 task = rqstp->rq_task;
704 }
705 spin_unlock_bh(&pool->sp_lock);
706
707 return task;
708}
709
710/* create new threads */
711static int
712svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
713{
714 struct svc_rqst *rqstp;
715 struct task_struct *task;
716 struct svc_pool *chosen_pool;
717 unsigned int state = serv->sv_nrthreads-1;
718 int node;
719
720 do {
721 nrservs--;
722 chosen_pool = choose_pool(serv, pool, &state);
723
724 node = svc_pool_map_get_node(chosen_pool->sp_id);
725 rqstp = svc_prepare_thread(serv, chosen_pool, node);
726 if (IS_ERR(rqstp))
727 return PTR_ERR(rqstp);
728
729 __module_get(serv->sv_ops->svo_module);
730 task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
731 node, "%s", serv->sv_name);
732 if (IS_ERR(task)) {
733 module_put(serv->sv_ops->svo_module);
734 svc_exit_thread(rqstp);
735 return PTR_ERR(task);
736 }
737
738 rqstp->rq_task = task;
739 if (serv->sv_nrpools > 1)
740 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
741
742 svc_sock_update_bufs(serv);
743 wake_up_process(task);
744 } while (nrservs > 0);
745
746 return 0;
747}
748
749
750/* destroy old threads */
751static int
752svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
753{
754 struct task_struct *task;
755 unsigned int state = serv->sv_nrthreads-1;
756
757 /* destroy old threads */
758 do {
759 task = choose_victim(serv, pool, &state);
760 if (task == NULL)
761 break;
762 send_sig(SIGINT, task, 1);
763 nrservs++;
764 } while (nrservs < 0);
765
766 return 0;
767}
768
769/*
770 * Create or destroy enough new threads to make the number
771 * of threads the given number. If `pool' is non-NULL, applies
772 * only to threads in that pool, otherwise round-robins between
773 * all pools. Caller must ensure that mutual exclusion between this and
774 * server startup or shutdown.
775 *
776 * Destroying threads relies on the service threads filling in
777 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
778 * has been created using svc_create_pooled().
779 *
780 * Based on code that used to be in nfsd_svc() but tweaked
781 * to be pool-aware.
782 */
783int
784svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
785{
786 if (pool == NULL) {
787 /* The -1 assumes caller has done a svc_get() */
788 nrservs -= (serv->sv_nrthreads-1);
789 } else {
790 spin_lock_bh(&pool->sp_lock);
791 nrservs -= pool->sp_nrthreads;
792 spin_unlock_bh(&pool->sp_lock);
793 }
794
795 if (nrservs > 0)
796 return svc_start_kthreads(serv, pool, nrservs);
797 if (nrservs < 0)
798 return svc_signal_kthreads(serv, pool, nrservs);
799 return 0;
800}
801EXPORT_SYMBOL_GPL(svc_set_num_threads);
802
803/* destroy old threads */
804static int
805svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
806{
807 struct task_struct *task;
808 unsigned int state = serv->sv_nrthreads-1;
809
810 /* destroy old threads */
811 do {
812 task = choose_victim(serv, pool, &state);
813 if (task == NULL)
814 break;
815 kthread_stop(task);
816 nrservs++;
817 } while (nrservs < 0);
818 return 0;
819}
820
821int
822svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
823{
824 if (pool == NULL) {
825 /* The -1 assumes caller has done a svc_get() */
826 nrservs -= (serv->sv_nrthreads-1);
827 } else {
828 spin_lock_bh(&pool->sp_lock);
829 nrservs -= pool->sp_nrthreads;
830 spin_unlock_bh(&pool->sp_lock);
831 }
832
833 if (nrservs > 0)
834 return svc_start_kthreads(serv, pool, nrservs);
835 if (nrservs < 0)
836 return svc_stop_kthreads(serv, pool, nrservs);
837 return 0;
838}
839EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
840
841/*
842 * Called from a server thread as it's exiting. Caller must hold the "service
843 * mutex" for the service.
844 */
845void
846svc_rqst_free(struct svc_rqst *rqstp)
847{
848 svc_release_buffer(rqstp);
849 if (rqstp->rq_scratch_page)
850 put_page(rqstp->rq_scratch_page);
851 kfree(rqstp->rq_resp);
852 kfree(rqstp->rq_argp);
853 kfree(rqstp->rq_auth_data);
854 kfree_rcu(rqstp, rq_rcu_head);
855}
856EXPORT_SYMBOL_GPL(svc_rqst_free);
857
858void
859svc_exit_thread(struct svc_rqst *rqstp)
860{
861 struct svc_serv *serv = rqstp->rq_server;
862 struct svc_pool *pool = rqstp->rq_pool;
863
864 spin_lock_bh(&pool->sp_lock);
865 pool->sp_nrthreads--;
866 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
867 list_del_rcu(&rqstp->rq_all);
868 spin_unlock_bh(&pool->sp_lock);
869
870 svc_rqst_free(rqstp);
871
872 /* Release the server */
873 if (serv)
874 svc_destroy(serv);
875}
876EXPORT_SYMBOL_GPL(svc_exit_thread);
877
878/*
879 * Register an "inet" protocol family netid with the local
880 * rpcbind daemon via an rpcbind v4 SET request.
881 *
882 * No netconfig infrastructure is available in the kernel, so
883 * we map IP_ protocol numbers to netids by hand.
884 *
885 * Returns zero on success; a negative errno value is returned
886 * if any error occurs.
887 */
888static int __svc_rpcb_register4(struct net *net, const u32 program,
889 const u32 version,
890 const unsigned short protocol,
891 const unsigned short port)
892{
893 const struct sockaddr_in sin = {
894 .sin_family = AF_INET,
895 .sin_addr.s_addr = htonl(INADDR_ANY),
896 .sin_port = htons(port),
897 };
898 const char *netid;
899 int error;
900
901 switch (protocol) {
902 case IPPROTO_UDP:
903 netid = RPCBIND_NETID_UDP;
904 break;
905 case IPPROTO_TCP:
906 netid = RPCBIND_NETID_TCP;
907 break;
908 default:
909 return -ENOPROTOOPT;
910 }
911
912 error = rpcb_v4_register(net, program, version,
913 (const struct sockaddr *)&sin, netid);
914
915 /*
916 * User space didn't support rpcbind v4, so retry this
917 * registration request with the legacy rpcbind v2 protocol.
918 */
919 if (error == -EPROTONOSUPPORT)
920 error = rpcb_register(net, program, version, protocol, port);
921
922 return error;
923}
924
925#if IS_ENABLED(CONFIG_IPV6)
926/*
927 * Register an "inet6" protocol family netid with the local
928 * rpcbind daemon via an rpcbind v4 SET request.
929 *
930 * No netconfig infrastructure is available in the kernel, so
931 * we map IP_ protocol numbers to netids by hand.
932 *
933 * Returns zero on success; a negative errno value is returned
934 * if any error occurs.
935 */
936static int __svc_rpcb_register6(struct net *net, const u32 program,
937 const u32 version,
938 const unsigned short protocol,
939 const unsigned short port)
940{
941 const struct sockaddr_in6 sin6 = {
942 .sin6_family = AF_INET6,
943 .sin6_addr = IN6ADDR_ANY_INIT,
944 .sin6_port = htons(port),
945 };
946 const char *netid;
947 int error;
948
949 switch (protocol) {
950 case IPPROTO_UDP:
951 netid = RPCBIND_NETID_UDP6;
952 break;
953 case IPPROTO_TCP:
954 netid = RPCBIND_NETID_TCP6;
955 break;
956 default:
957 return -ENOPROTOOPT;
958 }
959
960 error = rpcb_v4_register(net, program, version,
961 (const struct sockaddr *)&sin6, netid);
962
963 /*
964 * User space didn't support rpcbind version 4, so we won't
965 * use a PF_INET6 listener.
966 */
967 if (error == -EPROTONOSUPPORT)
968 error = -EAFNOSUPPORT;
969
970 return error;
971}
972#endif /* IS_ENABLED(CONFIG_IPV6) */
973
974/*
975 * Register a kernel RPC service via rpcbind version 4.
976 *
977 * Returns zero on success; a negative errno value is returned
978 * if any error occurs.
979 */
980static int __svc_register(struct net *net, const char *progname,
981 const u32 program, const u32 version,
982 const int family,
983 const unsigned short protocol,
984 const unsigned short port)
985{
986 int error = -EAFNOSUPPORT;
987
988 switch (family) {
989 case PF_INET:
990 error = __svc_rpcb_register4(net, program, version,
991 protocol, port);
992 break;
993#if IS_ENABLED(CONFIG_IPV6)
994 case PF_INET6:
995 error = __svc_rpcb_register6(net, program, version,
996 protocol, port);
997#endif
998 }
999
1000 trace_svc_register(progname, version, protocol, port, family, error);
1001 return error;
1002}
1003
1004int svc_rpcbind_set_version(struct net *net,
1005 const struct svc_program *progp,
1006 u32 version, int family,
1007 unsigned short proto,
1008 unsigned short port)
1009{
1010 return __svc_register(net, progp->pg_name, progp->pg_prog,
1011 version, family, proto, port);
1012
1013}
1014EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1015
1016int svc_generic_rpcbind_set(struct net *net,
1017 const struct svc_program *progp,
1018 u32 version, int family,
1019 unsigned short proto,
1020 unsigned short port)
1021{
1022 const struct svc_version *vers = progp->pg_vers[version];
1023 int error;
1024
1025 if (vers == NULL)
1026 return 0;
1027
1028 if (vers->vs_hidden) {
1029 trace_svc_noregister(progp->pg_name, version, proto,
1030 port, family, 0);
1031 return 0;
1032 }
1033
1034 /*
1035 * Don't register a UDP port if we need congestion
1036 * control.
1037 */
1038 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1039 return 0;
1040
1041 error = svc_rpcbind_set_version(net, progp, version,
1042 family, proto, port);
1043
1044 return (vers->vs_rpcb_optnl) ? 0 : error;
1045}
1046EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1047
1048/**
1049 * svc_register - register an RPC service with the local portmapper
1050 * @serv: svc_serv struct for the service to register
1051 * @net: net namespace for the service to register
1052 * @family: protocol family of service's listener socket
1053 * @proto: transport protocol number to advertise
1054 * @port: port to advertise
1055 *
1056 * Service is registered for any address in the passed-in protocol family
1057 */
1058int svc_register(const struct svc_serv *serv, struct net *net,
1059 const int family, const unsigned short proto,
1060 const unsigned short port)
1061{
1062 struct svc_program *progp;
1063 unsigned int i;
1064 int error = 0;
1065
1066 WARN_ON_ONCE(proto == 0 && port == 0);
1067 if (proto == 0 && port == 0)
1068 return -EINVAL;
1069
1070 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1071 for (i = 0; i < progp->pg_nvers; i++) {
1072
1073 error = progp->pg_rpcbind_set(net, progp, i,
1074 family, proto, port);
1075 if (error < 0) {
1076 printk(KERN_WARNING "svc: failed to register "
1077 "%sv%u RPC service (errno %d).\n",
1078 progp->pg_name, i, -error);
1079 break;
1080 }
1081 }
1082 }
1083
1084 return error;
1085}
1086
1087/*
1088 * If user space is running rpcbind, it should take the v4 UNSET
1089 * and clear everything for this [program, version]. If user space
1090 * is running portmap, it will reject the v4 UNSET, but won't have
1091 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1092 * in this case to clear all existing entries for [program, version].
1093 */
1094static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1095 const char *progname)
1096{
1097 int error;
1098
1099 error = rpcb_v4_register(net, program, version, NULL, "");
1100
1101 /*
1102 * User space didn't support rpcbind v4, so retry this
1103 * request with the legacy rpcbind v2 protocol.
1104 */
1105 if (error == -EPROTONOSUPPORT)
1106 error = rpcb_register(net, program, version, 0, 0);
1107
1108 trace_svc_unregister(progname, version, error);
1109}
1110
1111/*
1112 * All netids, bind addresses and ports registered for [program, version]
1113 * are removed from the local rpcbind database (if the service is not
1114 * hidden) to make way for a new instance of the service.
1115 *
1116 * The result of unregistration is reported via dprintk for those who want
1117 * verification of the result, but is otherwise not important.
1118 */
1119static void svc_unregister(const struct svc_serv *serv, struct net *net)
1120{
1121 struct svc_program *progp;
1122 unsigned long flags;
1123 unsigned int i;
1124
1125 clear_thread_flag(TIF_SIGPENDING);
1126
1127 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1128 for (i = 0; i < progp->pg_nvers; i++) {
1129 if (progp->pg_vers[i] == NULL)
1130 continue;
1131 if (progp->pg_vers[i]->vs_hidden)
1132 continue;
1133 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1134 }
1135 }
1136
1137 spin_lock_irqsave(¤t->sighand->siglock, flags);
1138 recalc_sigpending();
1139 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
1140}
1141
1142/*
1143 * dprintk the given error with the address of the client that caused it.
1144 */
1145#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1146static __printf(2, 3)
1147void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1148{
1149 struct va_format vaf;
1150 va_list args;
1151 char buf[RPC_MAX_ADDRBUFLEN];
1152
1153 va_start(args, fmt);
1154
1155 vaf.fmt = fmt;
1156 vaf.va = &args;
1157
1158 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1159
1160 va_end(args);
1161}
1162#else
1163static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1164#endif
1165
1166__be32
1167svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1168{
1169 set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1170 return auth_err;
1171}
1172EXPORT_SYMBOL_GPL(svc_return_autherr);
1173
1174static __be32
1175svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1176{
1177 if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1178 return *statp;
1179 return rpc_auth_ok;
1180}
1181
1182static int
1183svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1184{
1185 struct kvec *argv = &rqstp->rq_arg.head[0];
1186 struct kvec *resv = &rqstp->rq_res.head[0];
1187 const struct svc_procedure *procp = rqstp->rq_procinfo;
1188
1189 /*
1190 * Decode arguments
1191 * XXX: why do we ignore the return value?
1192 */
1193 if (procp->pc_decode &&
1194 !procp->pc_decode(rqstp, argv->iov_base)) {
1195 *statp = rpc_garbage_args;
1196 return 1;
1197 }
1198
1199 *statp = procp->pc_func(rqstp);
1200
1201 if (*statp == rpc_drop_reply ||
1202 test_bit(RQ_DROPME, &rqstp->rq_flags))
1203 return 0;
1204
1205 if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1206 return 1;
1207
1208 if (*statp != rpc_success)
1209 return 1;
1210
1211 /* Encode reply */
1212 if (procp->pc_encode &&
1213 !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1214 dprintk("svc: failed to encode reply\n");
1215 /* serv->sv_stats->rpcsystemerr++; */
1216 *statp = rpc_system_err;
1217 }
1218 return 1;
1219}
1220
1221__be32
1222svc_generic_init_request(struct svc_rqst *rqstp,
1223 const struct svc_program *progp,
1224 struct svc_process_info *ret)
1225{
1226 const struct svc_version *versp = NULL; /* compiler food */
1227 const struct svc_procedure *procp = NULL;
1228
1229 if (rqstp->rq_vers >= progp->pg_nvers )
1230 goto err_bad_vers;
1231 versp = progp->pg_vers[rqstp->rq_vers];
1232 if (!versp)
1233 goto err_bad_vers;
1234
1235 /*
1236 * Some protocol versions (namely NFSv4) require some form of
1237 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1238 * In other words, UDP is not allowed. We mark those when setting
1239 * up the svc_xprt, and verify that here.
1240 *
1241 * The spec is not very clear about what error should be returned
1242 * when someone tries to access a server that is listening on UDP
1243 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1244 * fit.
1245 */
1246 if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1247 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1248 goto err_bad_vers;
1249
1250 if (rqstp->rq_proc >= versp->vs_nproc)
1251 goto err_bad_proc;
1252 rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1253 if (!procp)
1254 goto err_bad_proc;
1255
1256 /* Initialize storage for argp and resp */
1257 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1258 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1259
1260 /* Bump per-procedure stats counter */
1261 versp->vs_count[rqstp->rq_proc]++;
1262
1263 ret->dispatch = versp->vs_dispatch;
1264 return rpc_success;
1265err_bad_vers:
1266 ret->mismatch.lovers = progp->pg_lovers;
1267 ret->mismatch.hivers = progp->pg_hivers;
1268 return rpc_prog_mismatch;
1269err_bad_proc:
1270 return rpc_proc_unavail;
1271}
1272EXPORT_SYMBOL_GPL(svc_generic_init_request);
1273
1274/*
1275 * Common routine for processing the RPC request.
1276 */
1277static int
1278svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1279{
1280 struct svc_program *progp;
1281 const struct svc_procedure *procp = NULL;
1282 struct svc_serv *serv = rqstp->rq_server;
1283 struct svc_process_info process;
1284 __be32 *statp;
1285 u32 prog, vers;
1286 __be32 auth_stat, rpc_stat;
1287 int auth_res;
1288 __be32 *reply_statp;
1289
1290 rpc_stat = rpc_success;
1291
1292 if (argv->iov_len < 6*4)
1293 goto err_short_len;
1294
1295 /* Will be turned off by GSS integrity and privacy services */
1296 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1297 /* Will be turned off only when NFSv4 Sessions are used */
1298 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1299 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1300
1301 svc_putu32(resv, rqstp->rq_xid);
1302
1303 vers = svc_getnl(argv);
1304
1305 /* First words of reply: */
1306 svc_putnl(resv, 1); /* REPLY */
1307
1308 if (vers != 2) /* RPC version number */
1309 goto err_bad_rpc;
1310
1311 /* Save position in case we later decide to reject: */
1312 reply_statp = resv->iov_base + resv->iov_len;
1313
1314 svc_putnl(resv, 0); /* ACCEPT */
1315
1316 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1317 rqstp->rq_vers = svc_getnl(argv); /* version number */
1318 rqstp->rq_proc = svc_getnl(argv); /* procedure number */
1319
1320 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1321 if (prog == progp->pg_prog)
1322 break;
1323
1324 /*
1325 * Decode auth data, and add verifier to reply buffer.
1326 * We do this before anything else in order to get a decent
1327 * auth verifier.
1328 */
1329 auth_res = svc_authenticate(rqstp, &auth_stat);
1330 /* Also give the program a chance to reject this call: */
1331 if (auth_res == SVC_OK && progp) {
1332 auth_stat = rpc_autherr_badcred;
1333 auth_res = progp->pg_authenticate(rqstp);
1334 }
1335 if (auth_res != SVC_OK)
1336 trace_svc_authenticate(rqstp, auth_res, auth_stat);
1337 switch (auth_res) {
1338 case SVC_OK:
1339 break;
1340 case SVC_GARBAGE:
1341 goto err_garbage;
1342 case SVC_SYSERR:
1343 rpc_stat = rpc_system_err;
1344 goto err_bad;
1345 case SVC_DENIED:
1346 goto err_bad_auth;
1347 case SVC_CLOSE:
1348 goto close;
1349 case SVC_DROP:
1350 goto dropit;
1351 case SVC_COMPLETE:
1352 goto sendit;
1353 }
1354
1355 if (progp == NULL)
1356 goto err_bad_prog;
1357
1358 rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1359 switch (rpc_stat) {
1360 case rpc_success:
1361 break;
1362 case rpc_prog_unavail:
1363 goto err_bad_prog;
1364 case rpc_prog_mismatch:
1365 goto err_bad_vers;
1366 case rpc_proc_unavail:
1367 goto err_bad_proc;
1368 }
1369
1370 procp = rqstp->rq_procinfo;
1371 /* Should this check go into the dispatcher? */
1372 if (!procp || !procp->pc_func)
1373 goto err_bad_proc;
1374
1375 /* Syntactic check complete */
1376 serv->sv_stats->rpccnt++;
1377 trace_svc_process(rqstp, progp->pg_name);
1378
1379 /* Build the reply header. */
1380 statp = resv->iov_base +resv->iov_len;
1381 svc_putnl(resv, RPC_SUCCESS);
1382
1383 /* un-reserve some of the out-queue now that we have a
1384 * better idea of reply size
1385 */
1386 if (procp->pc_xdrressize)
1387 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1388
1389 /* Call the function that processes the request. */
1390 if (!process.dispatch) {
1391 if (!svc_generic_dispatch(rqstp, statp))
1392 goto release_dropit;
1393 if (*statp == rpc_garbage_args)
1394 goto err_garbage;
1395 auth_stat = svc_get_autherr(rqstp, statp);
1396 if (auth_stat != rpc_auth_ok)
1397 goto err_release_bad_auth;
1398 } else {
1399 dprintk("svc: calling dispatcher\n");
1400 if (!process.dispatch(rqstp, statp))
1401 goto release_dropit; /* Release reply info */
1402 }
1403
1404 /* Check RPC status result */
1405 if (*statp != rpc_success)
1406 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1407
1408 /* Release reply info */
1409 if (procp->pc_release)
1410 procp->pc_release(rqstp);
1411
1412 if (procp->pc_encode == NULL)
1413 goto dropit;
1414
1415 sendit:
1416 if (svc_authorise(rqstp))
1417 goto close_xprt;
1418 return 1; /* Caller can now send it */
1419
1420release_dropit:
1421 if (procp->pc_release)
1422 procp->pc_release(rqstp);
1423 dropit:
1424 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1425 dprintk("svc: svc_process dropit\n");
1426 return 0;
1427
1428 close:
1429 svc_authorise(rqstp);
1430close_xprt:
1431 if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1432 svc_close_xprt(rqstp->rq_xprt);
1433 dprintk("svc: svc_process close\n");
1434 return 0;
1435
1436err_short_len:
1437 svc_printk(rqstp, "short len %zd, dropping request\n",
1438 argv->iov_len);
1439 goto close_xprt;
1440
1441err_bad_rpc:
1442 serv->sv_stats->rpcbadfmt++;
1443 svc_putnl(resv, 1); /* REJECT */
1444 svc_putnl(resv, 0); /* RPC_MISMATCH */
1445 svc_putnl(resv, 2); /* Only RPCv2 supported */
1446 svc_putnl(resv, 2);
1447 goto sendit;
1448
1449err_release_bad_auth:
1450 if (procp->pc_release)
1451 procp->pc_release(rqstp);
1452err_bad_auth:
1453 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1454 serv->sv_stats->rpcbadauth++;
1455 /* Restore write pointer to location of accept status: */
1456 xdr_ressize_check(rqstp, reply_statp);
1457 svc_putnl(resv, 1); /* REJECT */
1458 svc_putnl(resv, 1); /* AUTH_ERROR */
1459 svc_putnl(resv, ntohl(auth_stat)); /* status */
1460 goto sendit;
1461
1462err_bad_prog:
1463 dprintk("svc: unknown program %d\n", prog);
1464 serv->sv_stats->rpcbadfmt++;
1465 svc_putnl(resv, RPC_PROG_UNAVAIL);
1466 goto sendit;
1467
1468err_bad_vers:
1469 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1470 rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1471
1472 serv->sv_stats->rpcbadfmt++;
1473 svc_putnl(resv, RPC_PROG_MISMATCH);
1474 svc_putnl(resv, process.mismatch.lovers);
1475 svc_putnl(resv, process.mismatch.hivers);
1476 goto sendit;
1477
1478err_bad_proc:
1479 svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1480
1481 serv->sv_stats->rpcbadfmt++;
1482 svc_putnl(resv, RPC_PROC_UNAVAIL);
1483 goto sendit;
1484
1485err_garbage:
1486 svc_printk(rqstp, "failed to decode args\n");
1487
1488 rpc_stat = rpc_garbage_args;
1489err_bad:
1490 serv->sv_stats->rpcbadfmt++;
1491 svc_putnl(resv, ntohl(rpc_stat));
1492 goto sendit;
1493}
1494
1495/*
1496 * Process the RPC request.
1497 */
1498int
1499svc_process(struct svc_rqst *rqstp)
1500{
1501 struct kvec *argv = &rqstp->rq_arg.head[0];
1502 struct kvec *resv = &rqstp->rq_res.head[0];
1503 struct svc_serv *serv = rqstp->rq_server;
1504 u32 dir;
1505
1506 /*
1507 * Setup response xdr_buf.
1508 * Initially it has just one page
1509 */
1510 rqstp->rq_next_page = &rqstp->rq_respages[1];
1511 resv->iov_base = page_address(rqstp->rq_respages[0]);
1512 resv->iov_len = 0;
1513 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1514 rqstp->rq_res.len = 0;
1515 rqstp->rq_res.page_base = 0;
1516 rqstp->rq_res.page_len = 0;
1517 rqstp->rq_res.buflen = PAGE_SIZE;
1518 rqstp->rq_res.tail[0].iov_base = NULL;
1519 rqstp->rq_res.tail[0].iov_len = 0;
1520
1521 dir = svc_getnl(argv);
1522 if (dir != 0) {
1523 /* direction != CALL */
1524 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1525 serv->sv_stats->rpcbadfmt++;
1526 goto out_drop;
1527 }
1528
1529 /* Returns 1 for send, 0 for drop */
1530 if (likely(svc_process_common(rqstp, argv, resv)))
1531 return svc_send(rqstp);
1532
1533out_drop:
1534 svc_drop(rqstp);
1535 return 0;
1536}
1537EXPORT_SYMBOL_GPL(svc_process);
1538
1539#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1540/*
1541 * Process a backchannel RPC request that arrived over an existing
1542 * outbound connection
1543 */
1544int
1545bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1546 struct svc_rqst *rqstp)
1547{
1548 struct kvec *argv = &rqstp->rq_arg.head[0];
1549 struct kvec *resv = &rqstp->rq_res.head[0];
1550 struct rpc_task *task;
1551 int proc_error;
1552 int error;
1553
1554 dprintk("svc: %s(%p)\n", __func__, req);
1555
1556 /* Build the svc_rqst used by the common processing routine */
1557 rqstp->rq_xid = req->rq_xid;
1558 rqstp->rq_prot = req->rq_xprt->prot;
1559 rqstp->rq_server = serv;
1560 rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1561
1562 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1563 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1564 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1565 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1566
1567 /* Adjust the argument buffer length */
1568 rqstp->rq_arg.len = req->rq_private_buf.len;
1569 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1570 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1571 rqstp->rq_arg.page_len = 0;
1572 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1573 rqstp->rq_arg.page_len)
1574 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1575 rqstp->rq_arg.head[0].iov_len;
1576 else
1577 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1578 rqstp->rq_arg.page_len;
1579
1580 /* reset result send buffer "put" position */
1581 resv->iov_len = 0;
1582
1583 /*
1584 * Skip the next two words because they've already been
1585 * processed in the transport
1586 */
1587 svc_getu32(argv); /* XID */
1588 svc_getnl(argv); /* CALLDIR */
1589
1590 /* Parse and execute the bc call */
1591 proc_error = svc_process_common(rqstp, argv, resv);
1592
1593 atomic_dec(&req->rq_xprt->bc_slot_count);
1594 if (!proc_error) {
1595 /* Processing error: drop the request */
1596 xprt_free_bc_request(req);
1597 error = -EINVAL;
1598 goto out;
1599 }
1600 /* Finally, send the reply synchronously */
1601 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1602 task = rpc_run_bc_task(req);
1603 if (IS_ERR(task)) {
1604 error = PTR_ERR(task);
1605 goto out;
1606 }
1607
1608 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1609 error = task->tk_status;
1610 rpc_put_task(task);
1611
1612out:
1613 dprintk("svc: %s(), error=%d\n", __func__, error);
1614 return error;
1615}
1616EXPORT_SYMBOL_GPL(bc_svc_process);
1617#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1618
1619/*
1620 * Return (transport-specific) limit on the rpc payload.
1621 */
1622u32 svc_max_payload(const struct svc_rqst *rqstp)
1623{
1624 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1625
1626 if (rqstp->rq_server->sv_max_payload < max)
1627 max = rqstp->rq_server->sv_max_payload;
1628 return max;
1629}
1630EXPORT_SYMBOL_GPL(svc_max_payload);
1631
1632/**
1633 * svc_proc_name - Return RPC procedure name in string form
1634 * @rqstp: svc_rqst to operate on
1635 *
1636 * Return value:
1637 * Pointer to a NUL-terminated string
1638 */
1639const char *svc_proc_name(const struct svc_rqst *rqstp)
1640{
1641 if (rqstp && rqstp->rq_procinfo)
1642 return rqstp->rq_procinfo->pc_name;
1643 return "unknown";
1644}
1645
1646
1647/**
1648 * svc_encode_result_payload - mark a range of bytes as a result payload
1649 * @rqstp: svc_rqst to operate on
1650 * @offset: payload's byte offset in rqstp->rq_res
1651 * @length: size of payload, in bytes
1652 *
1653 * Returns zero on success, or a negative errno if a permanent
1654 * error occurred.
1655 */
1656int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1657 unsigned int length)
1658{
1659 return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1660 length);
1661}
1662EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1663
1664/**
1665 * svc_fill_write_vector - Construct data argument for VFS write call
1666 * @rqstp: svc_rqst to operate on
1667 * @pages: list of pages containing data payload
1668 * @first: buffer containing first section of write payload
1669 * @total: total number of bytes of write payload
1670 *
1671 * Fills in rqstp::rq_vec, and returns the number of elements.
1672 */
1673unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1674 struct kvec *first, size_t total)
1675{
1676 struct kvec *vec = rqstp->rq_vec;
1677 unsigned int i;
1678
1679 /* Some types of transport can present the write payload
1680 * entirely in rq_arg.pages. In this case, @first is empty.
1681 */
1682 i = 0;
1683 if (first->iov_len) {
1684 vec[i].iov_base = first->iov_base;
1685 vec[i].iov_len = min_t(size_t, total, first->iov_len);
1686 total -= vec[i].iov_len;
1687 ++i;
1688 }
1689
1690 while (total) {
1691 vec[i].iov_base = page_address(*pages);
1692 vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1693 total -= vec[i].iov_len;
1694 ++i;
1695 ++pages;
1696 }
1697
1698 WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1699 return i;
1700}
1701EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1702
1703/**
1704 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1705 * @rqstp: svc_rqst to operate on
1706 * @first: buffer containing first section of pathname
1707 * @p: buffer containing remaining section of pathname
1708 * @total: total length of the pathname argument
1709 *
1710 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1711 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1712 * the returned string.
1713 */
1714char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1715 void *p, size_t total)
1716{
1717 size_t len, remaining;
1718 char *result, *dst;
1719
1720 result = kmalloc(total + 1, GFP_KERNEL);
1721 if (!result)
1722 return ERR_PTR(-ESERVERFAULT);
1723
1724 dst = result;
1725 remaining = total;
1726
1727 len = min_t(size_t, total, first->iov_len);
1728 if (len) {
1729 memcpy(dst, first->iov_base, len);
1730 dst += len;
1731 remaining -= len;
1732 }
1733
1734 if (remaining) {
1735 len = min_t(size_t, remaining, PAGE_SIZE);
1736 memcpy(dst, p, len);
1737 dst += len;
1738 }
1739
1740 *dst = '\0';
1741
1742 /* Sanity check: Linux doesn't allow the pathname argument to
1743 * contain a NUL byte.
1744 */
1745 if (strlen(result) != total) {
1746 kfree(result);
1747 return ERR_PTR(-EINVAL);
1748 }
1749 return result;
1750}
1751EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);