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
38 static 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 */
45 enum {
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
58 struct 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
68 static struct svc_pool_map svc_pool_map = {
69 .mode = SVC_POOL_DEFAULT
70 };
71
72 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
73
74 static int
75 __param_set_pool_mode(const char *val, struct svc_pool_map *m)
76 {
77 int err, mode;
78
79 mutex_lock(&svc_pool_map_mutex);
80
81 err = 0;
82 if (!strncmp(val, "auto", 4))
83 mode = SVC_POOL_AUTO;
84 else if (!strncmp(val, "global", 6))
85 mode = SVC_POOL_GLOBAL;
86 else if (!strncmp(val, "percpu", 6))
87 mode = SVC_POOL_PERCPU;
88 else if (!strncmp(val, "pernode", 7))
89 mode = SVC_POOL_PERNODE;
90 else
91 err = -EINVAL;
92
93 if (err)
94 goto out;
95
96 if (m->count == 0)
97 m->mode = mode;
98 else if (mode != m->mode)
99 err = -EBUSY;
100 out:
101 mutex_unlock(&svc_pool_map_mutex);
102 return err;
103 }
104
105 static int
106 param_set_pool_mode(const char *val, const struct kernel_param *kp)
107 {
108 struct svc_pool_map *m = kp->arg;
109
110 return __param_set_pool_mode(val, m);
111 }
112
113 int sunrpc_set_pool_mode(const char *val)
114 {
115 return __param_set_pool_mode(val, &svc_pool_map);
116 }
117 EXPORT_SYMBOL(sunrpc_set_pool_mode);
118
119 /**
120 * sunrpc_get_pool_mode - get the current pool_mode for the host
121 * @buf: where to write the current pool_mode
122 * @size: size of @buf
123 *
124 * Grab the current pool_mode from the svc_pool_map and write
125 * the resulting string to @buf. Returns the number of characters
126 * written to @buf (a'la snprintf()).
127 */
128 int
129 sunrpc_get_pool_mode(char *buf, size_t size)
130 {
131 struct svc_pool_map *m = &svc_pool_map;
132
133 switch (m->mode)
134 {
135 case SVC_POOL_AUTO:
136 return snprintf(buf, size, "auto");
137 case SVC_POOL_GLOBAL:
138 return snprintf(buf, size, "global");
139 case SVC_POOL_PERCPU:
140 return snprintf(buf, size, "percpu");
141 case SVC_POOL_PERNODE:
142 return snprintf(buf, size, "pernode");
143 default:
144 return snprintf(buf, size, "%d", m->mode);
145 }
146 }
147 EXPORT_SYMBOL(sunrpc_get_pool_mode);
148
149 static int
150 param_get_pool_mode(char *buf, const struct kernel_param *kp)
151 {
152 char str[16];
153 int len;
154
155 len = sunrpc_get_pool_mode(str, ARRAY_SIZE(str));
156
157 /* Ensure we have room for newline and NUL */
158 len = min_t(int, len, ARRAY_SIZE(str) - 2);
159
160 /* tack on the newline */
161 str[len] = '\n';
162 str[len + 1] = '\0';
163
164 return sysfs_emit(buf, "%s", str);
165 }
166
167 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
168 &svc_pool_map, 0644);
169
170 /*
171 * Detect best pool mapping mode heuristically,
172 * according to the machine's topology.
173 */
174 static int
175 svc_pool_map_choose_mode(void)
176 {
177 unsigned int node;
178
179 if (nr_online_nodes > 1) {
180 /*
181 * Actually have multiple NUMA nodes,
182 * so split pools on NUMA node boundaries
183 */
184 return SVC_POOL_PERNODE;
185 }
186
187 node = first_online_node;
188 if (nr_cpus_node(node) > 2) {
189 /*
190 * Non-trivial SMP, or CONFIG_NUMA on
191 * non-NUMA hardware, e.g. with a generic
192 * x86_64 kernel on Xeons. In this case we
193 * want to divide the pools on cpu boundaries.
194 */
195 return SVC_POOL_PERCPU;
196 }
197
198 /* default: one global pool */
199 return SVC_POOL_GLOBAL;
200 }
201
202 /*
203 * Allocate the to_pool[] and pool_to[] arrays.
204 * Returns 0 on success or an errno.
205 */
206 static int
207 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
208 {
209 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
210 if (!m->to_pool)
211 goto fail;
212 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
213 if (!m->pool_to)
214 goto fail_free;
215
216 return 0;
217
218 fail_free:
219 kfree(m->to_pool);
220 m->to_pool = NULL;
221 fail:
222 return -ENOMEM;
223 }
224
225 /*
226 * Initialise the pool map for SVC_POOL_PERCPU mode.
227 * Returns number of pools or <0 on error.
228 */
229 static int
230 svc_pool_map_init_percpu(struct svc_pool_map *m)
231 {
232 unsigned int maxpools = nr_cpu_ids;
233 unsigned int pidx = 0;
234 unsigned int cpu;
235 int err;
236
237 err = svc_pool_map_alloc_arrays(m, maxpools);
238 if (err)
239 return err;
240
241 for_each_online_cpu(cpu) {
242 BUG_ON(pidx >= maxpools);
243 m->to_pool[cpu] = pidx;
244 m->pool_to[pidx] = cpu;
245 pidx++;
246 }
247 /* cpus brought online later all get mapped to pool0, sorry */
248
249 return pidx;
250 };
251
252
253 /*
254 * Initialise the pool map for SVC_POOL_PERNODE mode.
255 * Returns number of pools or <0 on error.
256 */
257 static int
258 svc_pool_map_init_pernode(struct svc_pool_map *m)
259 {
260 unsigned int maxpools = nr_node_ids;
261 unsigned int pidx = 0;
262 unsigned int node;
263 int err;
264
265 err = svc_pool_map_alloc_arrays(m, maxpools);
266 if (err)
267 return err;
268
269 for_each_node_with_cpus(node) {
270 /* some architectures (e.g. SN2) have cpuless nodes */
271 BUG_ON(pidx > maxpools);
272 m->to_pool[node] = pidx;
273 m->pool_to[pidx] = node;
274 pidx++;
275 }
276 /* nodes brought online later all get mapped to pool0, sorry */
277
278 return pidx;
279 }
280
281
282 /*
283 * Add a reference to the global map of cpus to pools (and
284 * vice versa) if pools are in use.
285 * Initialise the map if we're the first user.
286 * Returns the number of pools. If this is '1', no reference
287 * was taken.
288 */
289 static unsigned int
290 svc_pool_map_get(void)
291 {
292 struct svc_pool_map *m = &svc_pool_map;
293 int npools = -1;
294
295 mutex_lock(&svc_pool_map_mutex);
296 if (m->count++) {
297 mutex_unlock(&svc_pool_map_mutex);
298 return m->npools;
299 }
300
301 if (m->mode == SVC_POOL_AUTO)
302 m->mode = svc_pool_map_choose_mode();
303
304 switch (m->mode) {
305 case SVC_POOL_PERCPU:
306 npools = svc_pool_map_init_percpu(m);
307 break;
308 case SVC_POOL_PERNODE:
309 npools = svc_pool_map_init_pernode(m);
310 break;
311 }
312
313 if (npools <= 0) {
314 /* default, or memory allocation failure */
315 npools = 1;
316 m->mode = SVC_POOL_GLOBAL;
317 }
318 m->npools = npools;
319 mutex_unlock(&svc_pool_map_mutex);
320 return npools;
321 }
322
323 /*
324 * Drop a reference to the global map of cpus to pools.
325 * When the last reference is dropped, the map data is
326 * freed; this allows the sysadmin to change the pool.
327 */
328 static void
329 svc_pool_map_put(void)
330 {
331 struct svc_pool_map *m = &svc_pool_map;
332
333 mutex_lock(&svc_pool_map_mutex);
334 if (!--m->count) {
335 kfree(m->to_pool);
336 m->to_pool = NULL;
337 kfree(m->pool_to);
338 m->pool_to = NULL;
339 m->npools = 0;
340 }
341 mutex_unlock(&svc_pool_map_mutex);
342 }
343
344 static int svc_pool_map_get_node(unsigned int pidx)
345 {
346 const struct svc_pool_map *m = &svc_pool_map;
347
348 if (m->count) {
349 if (m->mode == SVC_POOL_PERCPU)
350 return cpu_to_node(m->pool_to[pidx]);
351 if (m->mode == SVC_POOL_PERNODE)
352 return m->pool_to[pidx];
353 }
354 return NUMA_NO_NODE;
355 }
356 /*
357 * Set the given thread's cpus_allowed mask so that it
358 * will only run on cpus in the given pool.
359 */
360 static inline void
361 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
362 {
363 struct svc_pool_map *m = &svc_pool_map;
364 unsigned int node = m->pool_to[pidx];
365
366 /*
367 * The caller checks for sv_nrpools > 1, which
368 * implies that we've been initialized.
369 */
370 WARN_ON_ONCE(m->count == 0);
371 if (m->count == 0)
372 return;
373
374 switch (m->mode) {
375 case SVC_POOL_PERCPU:
376 {
377 set_cpus_allowed_ptr(task, cpumask_of(node));
378 break;
379 }
380 case SVC_POOL_PERNODE:
381 {
382 set_cpus_allowed_ptr(task, cpumask_of_node(node));
383 break;
384 }
385 }
386 }
387
388 /**
389 * svc_pool_for_cpu - Select pool to run a thread on this cpu
390 * @serv: An RPC service
391 *
392 * Use the active CPU and the svc_pool_map's mode setting to
393 * select the svc thread pool to use. Once initialized, the
394 * svc_pool_map does not change.
395 *
396 * Return value:
397 * A pointer to an svc_pool
398 */
399 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv)
400 {
401 struct svc_pool_map *m = &svc_pool_map;
402 int cpu = raw_smp_processor_id();
403 unsigned int pidx = 0;
404
405 if (serv->sv_nrpools <= 1)
406 return serv->sv_pools;
407
408 switch (m->mode) {
409 case SVC_POOL_PERCPU:
410 pidx = m->to_pool[cpu];
411 break;
412 case SVC_POOL_PERNODE:
413 pidx = m->to_pool[cpu_to_node(cpu)];
414 break;
415 }
416
417 return &serv->sv_pools[pidx % serv->sv_nrpools];
418 }
419
420 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
421 {
422 int err;
423
424 err = rpcb_create_local(net);
425 if (err)
426 return err;
427
428 /* Remove any stale portmap registrations */
429 svc_unregister(serv, net);
430 return 0;
431 }
432 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
433
434 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
435 {
436 svc_unregister(serv, net);
437 rpcb_put_local(net);
438 }
439 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
440
441 static int svc_uses_rpcbind(struct svc_serv *serv)
442 {
443 struct svc_program *progp;
444 unsigned int i;
445
446 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
447 for (i = 0; i < progp->pg_nvers; i++) {
448 if (progp->pg_vers[i] == NULL)
449 continue;
450 if (!progp->pg_vers[i]->vs_hidden)
451 return 1;
452 }
453 }
454
455 return 0;
456 }
457
458 int svc_bind(struct svc_serv *serv, struct net *net)
459 {
460 if (!svc_uses_rpcbind(serv))
461 return 0;
462 return svc_rpcb_setup(serv, net);
463 }
464 EXPORT_SYMBOL_GPL(svc_bind);
465
466 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
467 static void
468 __svc_init_bc(struct svc_serv *serv)
469 {
470 lwq_init(&serv->sv_cb_list);
471 }
472 #else
473 static void
474 __svc_init_bc(struct svc_serv *serv)
475 {
476 }
477 #endif
478
479 /*
480 * Create an RPC service
481 */
482 static struct svc_serv *
483 __svc_create(struct svc_program *prog, struct svc_stat *stats,
484 unsigned int bufsize, int npools, int (*threadfn)(void *data))
485 {
486 struct svc_serv *serv;
487 unsigned int vers;
488 unsigned int xdrsize;
489 unsigned int i;
490
491 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
492 return NULL;
493 serv->sv_name = prog->pg_name;
494 serv->sv_program = prog;
495 serv->sv_stats = stats;
496 if (bufsize > RPCSVC_MAXPAYLOAD)
497 bufsize = RPCSVC_MAXPAYLOAD;
498 serv->sv_max_payload = bufsize? bufsize : 4096;
499 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
500 serv->sv_threadfn = threadfn;
501 xdrsize = 0;
502 while (prog) {
503 prog->pg_lovers = prog->pg_nvers-1;
504 for (vers=0; vers<prog->pg_nvers ; vers++)
505 if (prog->pg_vers[vers]) {
506 prog->pg_hivers = vers;
507 if (prog->pg_lovers > vers)
508 prog->pg_lovers = vers;
509 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
510 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
511 }
512 prog = prog->pg_next;
513 }
514 serv->sv_xdrsize = xdrsize;
515 INIT_LIST_HEAD(&serv->sv_tempsocks);
516 INIT_LIST_HEAD(&serv->sv_permsocks);
517 timer_setup(&serv->sv_temptimer, NULL, 0);
518 spin_lock_init(&serv->sv_lock);
519
520 __svc_init_bc(serv);
521
522 serv->sv_nrpools = npools;
523 serv->sv_pools =
524 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
525 GFP_KERNEL);
526 if (!serv->sv_pools) {
527 kfree(serv);
528 return NULL;
529 }
530
531 for (i = 0; i < serv->sv_nrpools; i++) {
532 struct svc_pool *pool = &serv->sv_pools[i];
533
534 dprintk("svc: initialising pool %u for %s\n",
535 i, serv->sv_name);
536
537 pool->sp_id = i;
538 lwq_init(&pool->sp_xprts);
539 INIT_LIST_HEAD(&pool->sp_all_threads);
540 init_llist_head(&pool->sp_idle_threads);
541
542 percpu_counter_init(&pool->sp_messages_arrived, 0, GFP_KERNEL);
543 percpu_counter_init(&pool->sp_sockets_queued, 0, GFP_KERNEL);
544 percpu_counter_init(&pool->sp_threads_woken, 0, GFP_KERNEL);
545 }
546
547 return serv;
548 }
549
550 /**
551 * svc_create - Create an RPC service
552 * @prog: the RPC program the new service will handle
553 * @bufsize: maximum message size for @prog
554 * @threadfn: a function to service RPC requests for @prog
555 *
556 * Returns an instantiated struct svc_serv object or NULL.
557 */
558 struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize,
559 int (*threadfn)(void *data))
560 {
561 return __svc_create(prog, NULL, bufsize, 1, threadfn);
562 }
563 EXPORT_SYMBOL_GPL(svc_create);
564
565 /**
566 * svc_create_pooled - Create an RPC service with pooled threads
567 * @prog: the RPC program the new service will handle
568 * @stats: the stats struct if desired
569 * @bufsize: maximum message size for @prog
570 * @threadfn: a function to service RPC requests for @prog
571 *
572 * Returns an instantiated struct svc_serv object or NULL.
573 */
574 struct svc_serv *svc_create_pooled(struct svc_program *prog,
575 struct svc_stat *stats,
576 unsigned int bufsize,
577 int (*threadfn)(void *data))
578 {
579 struct svc_serv *serv;
580 unsigned int npools = svc_pool_map_get();
581
582 serv = __svc_create(prog, stats, bufsize, npools, threadfn);
583 if (!serv)
584 goto out_err;
585 serv->sv_is_pooled = true;
586 return serv;
587 out_err:
588 svc_pool_map_put();
589 return NULL;
590 }
591 EXPORT_SYMBOL_GPL(svc_create_pooled);
592
593 /*
594 * Destroy an RPC service. Should be called with appropriate locking to
595 * protect sv_permsocks and sv_tempsocks.
596 */
597 void
598 svc_destroy(struct svc_serv **servp)
599 {
600 struct svc_serv *serv = *servp;
601 unsigned int i;
602
603 *servp = NULL;
604
605 dprintk("svc: svc_destroy(%s)\n", serv->sv_program->pg_name);
606 timer_shutdown_sync(&serv->sv_temptimer);
607
608 /*
609 * Remaining transports at this point are not expected.
610 */
611 WARN_ONCE(!list_empty(&serv->sv_permsocks),
612 "SVC: permsocks remain for %s\n", serv->sv_program->pg_name);
613 WARN_ONCE(!list_empty(&serv->sv_tempsocks),
614 "SVC: tempsocks remain for %s\n", serv->sv_program->pg_name);
615
616 cache_clean_deferred(serv);
617
618 if (serv->sv_is_pooled)
619 svc_pool_map_put();
620
621 for (i = 0; i < serv->sv_nrpools; i++) {
622 struct svc_pool *pool = &serv->sv_pools[i];
623
624 percpu_counter_destroy(&pool->sp_messages_arrived);
625 percpu_counter_destroy(&pool->sp_sockets_queued);
626 percpu_counter_destroy(&pool->sp_threads_woken);
627 }
628 kfree(serv->sv_pools);
629 kfree(serv);
630 }
631 EXPORT_SYMBOL_GPL(svc_destroy);
632
633 static bool
634 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
635 {
636 unsigned long pages, ret;
637
638 /* bc_xprt uses fore channel allocated buffers */
639 if (svc_is_backchannel(rqstp))
640 return true;
641
642 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
643 * We assume one is at most one page
644 */
645 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
646 if (pages > RPCSVC_MAXPAGES)
647 pages = RPCSVC_MAXPAGES;
648
649 ret = alloc_pages_bulk_array_node(GFP_KERNEL, node, pages,
650 rqstp->rq_pages);
651 return ret == pages;
652 }
653
654 /*
655 * Release an RPC server buffer
656 */
657 static void
658 svc_release_buffer(struct svc_rqst *rqstp)
659 {
660 unsigned int i;
661
662 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
663 if (rqstp->rq_pages[i])
664 put_page(rqstp->rq_pages[i]);
665 }
666
667 struct svc_rqst *
668 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
669 {
670 struct svc_rqst *rqstp;
671
672 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
673 if (!rqstp)
674 return rqstp;
675
676 folio_batch_init(&rqstp->rq_fbatch);
677
678 rqstp->rq_server = serv;
679 rqstp->rq_pool = pool;
680
681 rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
682 if (!rqstp->rq_scratch_page)
683 goto out_enomem;
684
685 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
686 if (!rqstp->rq_argp)
687 goto out_enomem;
688
689 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
690 if (!rqstp->rq_resp)
691 goto out_enomem;
692
693 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
694 goto out_enomem;
695
696 return rqstp;
697 out_enomem:
698 svc_rqst_free(rqstp);
699 return NULL;
700 }
701 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
702
703 static struct svc_rqst *
704 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
705 {
706 struct svc_rqst *rqstp;
707
708 rqstp = svc_rqst_alloc(serv, pool, node);
709 if (!rqstp)
710 return ERR_PTR(-ENOMEM);
711
712 spin_lock_bh(&serv->sv_lock);
713 serv->sv_nrthreads += 1;
714 spin_unlock_bh(&serv->sv_lock);
715
716 pool->sp_nrthreads += 1;
717
718 /* Protected by whatever lock the service uses when calling
719 * svc_set_num_threads()
720 */
721 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
722
723 return rqstp;
724 }
725
726 /**
727 * svc_pool_wake_idle_thread - Awaken an idle thread in @pool
728 * @pool: service thread pool
729 *
730 * Can be called from soft IRQ or process context. Finding an idle
731 * service thread and marking it BUSY is atomic with respect to
732 * other calls to svc_pool_wake_idle_thread().
733 *
734 */
735 void svc_pool_wake_idle_thread(struct svc_pool *pool)
736 {
737 struct svc_rqst *rqstp;
738 struct llist_node *ln;
739
740 rcu_read_lock();
741 ln = READ_ONCE(pool->sp_idle_threads.first);
742 if (ln) {
743 rqstp = llist_entry(ln, struct svc_rqst, rq_idle);
744 WRITE_ONCE(rqstp->rq_qtime, ktime_get());
745 if (!task_is_running(rqstp->rq_task)) {
746 wake_up_process(rqstp->rq_task);
747 trace_svc_wake_up(rqstp->rq_task->pid);
748 percpu_counter_inc(&pool->sp_threads_woken);
749 }
750 rcu_read_unlock();
751 return;
752 }
753 rcu_read_unlock();
754
755 }
756 EXPORT_SYMBOL_GPL(svc_pool_wake_idle_thread);
757
758 static struct svc_pool *
759 svc_pool_next(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
760 {
761 return pool ? pool : &serv->sv_pools[(*state)++ % serv->sv_nrpools];
762 }
763
764 static struct svc_pool *
765 svc_pool_victim(struct svc_serv *serv, struct svc_pool *target_pool,
766 unsigned int *state)
767 {
768 struct svc_pool *pool;
769 unsigned int i;
770
771 pool = target_pool;
772
773 if (!pool) {
774 for (i = 0; i < serv->sv_nrpools; i++) {
775 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
776 if (pool->sp_nrthreads)
777 break;
778 }
779 }
780
781 if (pool && pool->sp_nrthreads) {
782 set_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
783 set_bit(SP_NEED_VICTIM, &pool->sp_flags);
784 return pool;
785 }
786 return NULL;
787 }
788
789 static int
790 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
791 {
792 struct svc_rqst *rqstp;
793 struct task_struct *task;
794 struct svc_pool *chosen_pool;
795 unsigned int state = serv->sv_nrthreads-1;
796 int node;
797
798 do {
799 nrservs--;
800 chosen_pool = svc_pool_next(serv, pool, &state);
801 node = svc_pool_map_get_node(chosen_pool->sp_id);
802
803 rqstp = svc_prepare_thread(serv, chosen_pool, node);
804 if (IS_ERR(rqstp))
805 return PTR_ERR(rqstp);
806 task = kthread_create_on_node(serv->sv_threadfn, rqstp,
807 node, "%s", serv->sv_name);
808 if (IS_ERR(task)) {
809 svc_exit_thread(rqstp);
810 return PTR_ERR(task);
811 }
812
813 rqstp->rq_task = task;
814 if (serv->sv_nrpools > 1)
815 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
816
817 svc_sock_update_bufs(serv);
818 wake_up_process(task);
819 } while (nrservs > 0);
820
821 return 0;
822 }
823
824 static int
825 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
826 {
827 unsigned int state = serv->sv_nrthreads-1;
828 struct svc_pool *victim;
829
830 do {
831 victim = svc_pool_victim(serv, pool, &state);
832 if (!victim)
833 break;
834 svc_pool_wake_idle_thread(victim);
835 wait_on_bit(&victim->sp_flags, SP_VICTIM_REMAINS,
836 TASK_IDLE);
837 nrservs++;
838 } while (nrservs < 0);
839 return 0;
840 }
841
842 /**
843 * svc_set_num_threads - adjust number of threads per RPC service
844 * @serv: RPC service to adjust
845 * @pool: Specific pool from which to choose threads, or NULL
846 * @nrservs: New number of threads for @serv (0 or less means kill all threads)
847 *
848 * Create or destroy threads to make the number of threads for @serv the
849 * given number. If @pool is non-NULL, change only threads in that pool;
850 * otherwise, round-robin between all pools for @serv. @serv's
851 * sv_nrthreads is adjusted for each thread created or destroyed.
852 *
853 * Caller must ensure mutual exclusion between this and server startup or
854 * shutdown.
855 *
856 * Returns zero on success or a negative errno if an error occurred while
857 * starting a thread.
858 */
859 int
860 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
861 {
862 if (!pool)
863 nrservs -= serv->sv_nrthreads;
864 else
865 nrservs -= pool->sp_nrthreads;
866
867 if (nrservs > 0)
868 return svc_start_kthreads(serv, pool, nrservs);
869 if (nrservs < 0)
870 return svc_stop_kthreads(serv, pool, nrservs);
871 return 0;
872 }
873 EXPORT_SYMBOL_GPL(svc_set_num_threads);
874
875 /**
876 * svc_rqst_replace_page - Replace one page in rq_pages[]
877 * @rqstp: svc_rqst with pages to replace
878 * @page: replacement page
879 *
880 * When replacing a page in rq_pages, batch the release of the
881 * replaced pages to avoid hammering the page allocator.
882 *
883 * Return values:
884 * %true: page replaced
885 * %false: array bounds checking failed
886 */
887 bool svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
888 {
889 struct page **begin = rqstp->rq_pages;
890 struct page **end = &rqstp->rq_pages[RPCSVC_MAXPAGES];
891
892 if (unlikely(rqstp->rq_next_page < begin || rqstp->rq_next_page > end)) {
893 trace_svc_replace_page_err(rqstp);
894 return false;
895 }
896
897 if (*rqstp->rq_next_page) {
898 if (!folio_batch_add(&rqstp->rq_fbatch,
899 page_folio(*rqstp->rq_next_page)))
900 __folio_batch_release(&rqstp->rq_fbatch);
901 }
902
903 get_page(page);
904 *(rqstp->rq_next_page++) = page;
905 return true;
906 }
907 EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
908
909 /**
910 * svc_rqst_release_pages - Release Reply buffer pages
911 * @rqstp: RPC transaction context
912 *
913 * Release response pages that might still be in flight after
914 * svc_send, and any spliced filesystem-owned pages.
915 */
916 void svc_rqst_release_pages(struct svc_rqst *rqstp)
917 {
918 int i, count = rqstp->rq_next_page - rqstp->rq_respages;
919
920 if (count) {
921 release_pages(rqstp->rq_respages, count);
922 for (i = 0; i < count; i++)
923 rqstp->rq_respages[i] = NULL;
924 }
925 }
926
927 /*
928 * Called from a server thread as it's exiting. Caller must hold the "service
929 * mutex" for the service.
930 */
931 void
932 svc_rqst_free(struct svc_rqst *rqstp)
933 {
934 folio_batch_release(&rqstp->rq_fbatch);
935 svc_release_buffer(rqstp);
936 if (rqstp->rq_scratch_page)
937 put_page(rqstp->rq_scratch_page);
938 kfree(rqstp->rq_resp);
939 kfree(rqstp->rq_argp);
940 kfree(rqstp->rq_auth_data);
941 kfree_rcu(rqstp, rq_rcu_head);
942 }
943 EXPORT_SYMBOL_GPL(svc_rqst_free);
944
945 void
946 svc_exit_thread(struct svc_rqst *rqstp)
947 {
948 struct svc_serv *serv = rqstp->rq_server;
949 struct svc_pool *pool = rqstp->rq_pool;
950
951 list_del_rcu(&rqstp->rq_all);
952
953 pool->sp_nrthreads -= 1;
954
955 spin_lock_bh(&serv->sv_lock);
956 serv->sv_nrthreads -= 1;
957 spin_unlock_bh(&serv->sv_lock);
958 svc_sock_update_bufs(serv);
959
960 svc_rqst_free(rqstp);
961
962 clear_and_wake_up_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
963 }
964 EXPORT_SYMBOL_GPL(svc_exit_thread);
965
966 /*
967 * Register an "inet" protocol family netid with the local
968 * rpcbind daemon via an rpcbind v4 SET request.
969 *
970 * No netconfig infrastructure is available in the kernel, so
971 * we map IP_ protocol numbers to netids by hand.
972 *
973 * Returns zero on success; a negative errno value is returned
974 * if any error occurs.
975 */
976 static int __svc_rpcb_register4(struct net *net, const u32 program,
977 const u32 version,
978 const unsigned short protocol,
979 const unsigned short port)
980 {
981 const struct sockaddr_in sin = {
982 .sin_family = AF_INET,
983 .sin_addr.s_addr = htonl(INADDR_ANY),
984 .sin_port = htons(port),
985 };
986 const char *netid;
987 int error;
988
989 switch (protocol) {
990 case IPPROTO_UDP:
991 netid = RPCBIND_NETID_UDP;
992 break;
993 case IPPROTO_TCP:
994 netid = RPCBIND_NETID_TCP;
995 break;
996 default:
997 return -ENOPROTOOPT;
998 }
999
1000 error = rpcb_v4_register(net, program, version,
1001 (const struct sockaddr *)&sin, netid);
1002
1003 /*
1004 * User space didn't support rpcbind v4, so retry this
1005 * registration request with the legacy rpcbind v2 protocol.
1006 */
1007 if (error == -EPROTONOSUPPORT)
1008 error = rpcb_register(net, program, version, protocol, port);
1009
1010 return error;
1011 }
1012
1013 #if IS_ENABLED(CONFIG_IPV6)
1014 /*
1015 * Register an "inet6" protocol family netid with the local
1016 * rpcbind daemon via an rpcbind v4 SET request.
1017 *
1018 * No netconfig infrastructure is available in the kernel, so
1019 * we map IP_ protocol numbers to netids by hand.
1020 *
1021 * Returns zero on success; a negative errno value is returned
1022 * if any error occurs.
1023 */
1024 static int __svc_rpcb_register6(struct net *net, const u32 program,
1025 const u32 version,
1026 const unsigned short protocol,
1027 const unsigned short port)
1028 {
1029 const struct sockaddr_in6 sin6 = {
1030 .sin6_family = AF_INET6,
1031 .sin6_addr = IN6ADDR_ANY_INIT,
1032 .sin6_port = htons(port),
1033 };
1034 const char *netid;
1035 int error;
1036
1037 switch (protocol) {
1038 case IPPROTO_UDP:
1039 netid = RPCBIND_NETID_UDP6;
1040 break;
1041 case IPPROTO_TCP:
1042 netid = RPCBIND_NETID_TCP6;
1043 break;
1044 default:
1045 return -ENOPROTOOPT;
1046 }
1047
1048 error = rpcb_v4_register(net, program, version,
1049 (const struct sockaddr *)&sin6, netid);
1050
1051 /*
1052 * User space didn't support rpcbind version 4, so we won't
1053 * use a PF_INET6 listener.
1054 */
1055 if (error == -EPROTONOSUPPORT)
1056 error = -EAFNOSUPPORT;
1057
1058 return error;
1059 }
1060 #endif /* IS_ENABLED(CONFIG_IPV6) */
1061
1062 /*
1063 * Register a kernel RPC service via rpcbind version 4.
1064 *
1065 * Returns zero on success; a negative errno value is returned
1066 * if any error occurs.
1067 */
1068 static int __svc_register(struct net *net, const char *progname,
1069 const u32 program, const u32 version,
1070 const int family,
1071 const unsigned short protocol,
1072 const unsigned short port)
1073 {
1074 int error = -EAFNOSUPPORT;
1075
1076 switch (family) {
1077 case PF_INET:
1078 error = __svc_rpcb_register4(net, program, version,
1079 protocol, port);
1080 break;
1081 #if IS_ENABLED(CONFIG_IPV6)
1082 case PF_INET6:
1083 error = __svc_rpcb_register6(net, program, version,
1084 protocol, port);
1085 #endif
1086 }
1087
1088 trace_svc_register(progname, version, family, protocol, port, error);
1089 return error;
1090 }
1091
1092 int svc_rpcbind_set_version(struct net *net,
1093 const struct svc_program *progp,
1094 u32 version, int family,
1095 unsigned short proto,
1096 unsigned short port)
1097 {
1098 return __svc_register(net, progp->pg_name, progp->pg_prog,
1099 version, family, proto, port);
1100
1101 }
1102 EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1103
1104 int svc_generic_rpcbind_set(struct net *net,
1105 const struct svc_program *progp,
1106 u32 version, int family,
1107 unsigned short proto,
1108 unsigned short port)
1109 {
1110 const struct svc_version *vers = progp->pg_vers[version];
1111 int error;
1112
1113 if (vers == NULL)
1114 return 0;
1115
1116 if (vers->vs_hidden) {
1117 trace_svc_noregister(progp->pg_name, version, proto,
1118 port, family, 0);
1119 return 0;
1120 }
1121
1122 /*
1123 * Don't register a UDP port if we need congestion
1124 * control.
1125 */
1126 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1127 return 0;
1128
1129 error = svc_rpcbind_set_version(net, progp, version,
1130 family, proto, port);
1131
1132 return (vers->vs_rpcb_optnl) ? 0 : error;
1133 }
1134 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1135
1136 /**
1137 * svc_register - register an RPC service with the local portmapper
1138 * @serv: svc_serv struct for the service to register
1139 * @net: net namespace for the service to register
1140 * @family: protocol family of service's listener socket
1141 * @proto: transport protocol number to advertise
1142 * @port: port to advertise
1143 *
1144 * Service is registered for any address in the passed-in protocol family
1145 */
1146 int svc_register(const struct svc_serv *serv, struct net *net,
1147 const int family, const unsigned short proto,
1148 const unsigned short port)
1149 {
1150 struct svc_program *progp;
1151 unsigned int i;
1152 int error = 0;
1153
1154 WARN_ON_ONCE(proto == 0 && port == 0);
1155 if (proto == 0 && port == 0)
1156 return -EINVAL;
1157
1158 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1159 for (i = 0; i < progp->pg_nvers; i++) {
1160
1161 error = progp->pg_rpcbind_set(net, progp, i,
1162 family, proto, port);
1163 if (error < 0) {
1164 printk(KERN_WARNING "svc: failed to register "
1165 "%sv%u RPC service (errno %d).\n",
1166 progp->pg_name, i, -error);
1167 break;
1168 }
1169 }
1170 }
1171
1172 return error;
1173 }
1174
1175 /*
1176 * If user space is running rpcbind, it should take the v4 UNSET
1177 * and clear everything for this [program, version]. If user space
1178 * is running portmap, it will reject the v4 UNSET, but won't have
1179 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1180 * in this case to clear all existing entries for [program, version].
1181 */
1182 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1183 const char *progname)
1184 {
1185 int error;
1186
1187 error = rpcb_v4_register(net, program, version, NULL, "");
1188
1189 /*
1190 * User space didn't support rpcbind v4, so retry this
1191 * request with the legacy rpcbind v2 protocol.
1192 */
1193 if (error == -EPROTONOSUPPORT)
1194 error = rpcb_register(net, program, version, 0, 0);
1195
1196 trace_svc_unregister(progname, version, error);
1197 }
1198
1199 /*
1200 * All netids, bind addresses and ports registered for [program, version]
1201 * are removed from the local rpcbind database (if the service is not
1202 * hidden) to make way for a new instance of the service.
1203 *
1204 * The result of unregistration is reported via dprintk for those who want
1205 * verification of the result, but is otherwise not important.
1206 */
1207 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1208 {
1209 struct sighand_struct *sighand;
1210 struct svc_program *progp;
1211 unsigned long flags;
1212 unsigned int i;
1213
1214 clear_thread_flag(TIF_SIGPENDING);
1215
1216 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1217 for (i = 0; i < progp->pg_nvers; i++) {
1218 if (progp->pg_vers[i] == NULL)
1219 continue;
1220 if (progp->pg_vers[i]->vs_hidden)
1221 continue;
1222 __svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1223 }
1224 }
1225
1226 rcu_read_lock();
1227 sighand = rcu_dereference(current->sighand);
1228 spin_lock_irqsave(&sighand->siglock, flags);
1229 recalc_sigpending();
1230 spin_unlock_irqrestore(&sighand->siglock, flags);
1231 rcu_read_unlock();
1232 }
1233
1234 /*
1235 * dprintk the given error with the address of the client that caused it.
1236 */
1237 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1238 static __printf(2, 3)
1239 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1240 {
1241 struct va_format vaf;
1242 va_list args;
1243 char buf[RPC_MAX_ADDRBUFLEN];
1244
1245 va_start(args, fmt);
1246
1247 vaf.fmt = fmt;
1248 vaf.va = &args;
1249
1250 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1251
1252 va_end(args);
1253 }
1254 #else
1255 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1256 #endif
1257
1258 __be32
1259 svc_generic_init_request(struct svc_rqst *rqstp,
1260 const struct svc_program *progp,
1261 struct svc_process_info *ret)
1262 {
1263 const struct svc_version *versp = NULL; /* compiler food */
1264 const struct svc_procedure *procp = NULL;
1265
1266 if (rqstp->rq_vers >= progp->pg_nvers )
1267 goto err_bad_vers;
1268 versp = progp->pg_vers[rqstp->rq_vers];
1269 if (!versp)
1270 goto err_bad_vers;
1271
1272 /*
1273 * Some protocol versions (namely NFSv4) require some form of
1274 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1275 * In other words, UDP is not allowed. We mark those when setting
1276 * up the svc_xprt, and verify that here.
1277 *
1278 * The spec is not very clear about what error should be returned
1279 * when someone tries to access a server that is listening on UDP
1280 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1281 * fit.
1282 */
1283 if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1284 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1285 goto err_bad_vers;
1286
1287 if (rqstp->rq_proc >= versp->vs_nproc)
1288 goto err_bad_proc;
1289 rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1290
1291 /* Initialize storage for argp and resp */
1292 memset(rqstp->rq_argp, 0, procp->pc_argzero);
1293 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1294
1295 /* Bump per-procedure stats counter */
1296 this_cpu_inc(versp->vs_count[rqstp->rq_proc]);
1297
1298 ret->dispatch = versp->vs_dispatch;
1299 return rpc_success;
1300 err_bad_vers:
1301 ret->mismatch.lovers = progp->pg_lovers;
1302 ret->mismatch.hivers = progp->pg_hivers;
1303 return rpc_prog_mismatch;
1304 err_bad_proc:
1305 return rpc_proc_unavail;
1306 }
1307 EXPORT_SYMBOL_GPL(svc_generic_init_request);
1308
1309 /*
1310 * Common routine for processing the RPC request.
1311 */
1312 static int
1313 svc_process_common(struct svc_rqst *rqstp)
1314 {
1315 struct xdr_stream *xdr = &rqstp->rq_res_stream;
1316 struct svc_program *progp;
1317 const struct svc_procedure *procp = NULL;
1318 struct svc_serv *serv = rqstp->rq_server;
1319 struct svc_process_info process;
1320 enum svc_auth_status auth_res;
1321 unsigned int aoffset;
1322 int rc;
1323 __be32 *p;
1324
1325 /* Will be turned off only when NFSv4 Sessions are used */
1326 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1327 clear_bit(RQ_DROPME, &rqstp->rq_flags);
1328
1329 /* Construct the first words of the reply: */
1330 svcxdr_init_encode(rqstp);
1331 xdr_stream_encode_be32(xdr, rqstp->rq_xid);
1332 xdr_stream_encode_be32(xdr, rpc_reply);
1333
1334 p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 4);
1335 if (unlikely(!p))
1336 goto err_short_len;
1337 if (*p++ != cpu_to_be32(RPC_VERSION))
1338 goto err_bad_rpc;
1339
1340 xdr_stream_encode_be32(xdr, rpc_msg_accepted);
1341
1342 rqstp->rq_prog = be32_to_cpup(p++);
1343 rqstp->rq_vers = be32_to_cpup(p++);
1344 rqstp->rq_proc = be32_to_cpup(p);
1345
1346 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1347 if (rqstp->rq_prog == progp->pg_prog)
1348 break;
1349
1350 /*
1351 * Decode auth data, and add verifier to reply buffer.
1352 * We do this before anything else in order to get a decent
1353 * auth verifier.
1354 */
1355 auth_res = svc_authenticate(rqstp);
1356 /* Also give the program a chance to reject this call: */
1357 if (auth_res == SVC_OK && progp)
1358 auth_res = progp->pg_authenticate(rqstp);
1359 trace_svc_authenticate(rqstp, auth_res);
1360 switch (auth_res) {
1361 case SVC_OK:
1362 break;
1363 case SVC_GARBAGE:
1364 goto err_garbage_args;
1365 case SVC_SYSERR:
1366 goto err_system_err;
1367 case SVC_DENIED:
1368 goto err_bad_auth;
1369 case SVC_CLOSE:
1370 goto close;
1371 case SVC_DROP:
1372 goto dropit;
1373 case SVC_COMPLETE:
1374 goto sendit;
1375 default:
1376 pr_warn_once("Unexpected svc_auth_status (%d)\n", auth_res);
1377 goto err_system_err;
1378 }
1379
1380 if (progp == NULL)
1381 goto err_bad_prog;
1382
1383 switch (progp->pg_init_request(rqstp, progp, &process)) {
1384 case rpc_success:
1385 break;
1386 case rpc_prog_unavail:
1387 goto err_bad_prog;
1388 case rpc_prog_mismatch:
1389 goto err_bad_vers;
1390 case rpc_proc_unavail:
1391 goto err_bad_proc;
1392 }
1393
1394 procp = rqstp->rq_procinfo;
1395 /* Should this check go into the dispatcher? */
1396 if (!procp || !procp->pc_func)
1397 goto err_bad_proc;
1398
1399 /* Syntactic check complete */
1400 if (serv->sv_stats)
1401 serv->sv_stats->rpccnt++;
1402 trace_svc_process(rqstp, progp->pg_name);
1403
1404 aoffset = xdr_stream_pos(xdr);
1405
1406 /* un-reserve some of the out-queue now that we have a
1407 * better idea of reply size
1408 */
1409 if (procp->pc_xdrressize)
1410 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1411
1412 /* Call the function that processes the request. */
1413 rc = process.dispatch(rqstp);
1414 if (procp->pc_release)
1415 procp->pc_release(rqstp);
1416 xdr_finish_decode(xdr);
1417
1418 if (!rc)
1419 goto dropit;
1420 if (rqstp->rq_auth_stat != rpc_auth_ok)
1421 goto err_bad_auth;
1422
1423 if (*rqstp->rq_accept_statp != rpc_success)
1424 xdr_truncate_encode(xdr, aoffset);
1425
1426 if (procp->pc_encode == NULL)
1427 goto dropit;
1428
1429 sendit:
1430 if (svc_authorise(rqstp))
1431 goto close_xprt;
1432 return 1; /* Caller can now send it */
1433
1434 dropit:
1435 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1436 dprintk("svc: svc_process dropit\n");
1437 return 0;
1438
1439 close:
1440 svc_authorise(rqstp);
1441 close_xprt:
1442 if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1443 svc_xprt_close(rqstp->rq_xprt);
1444 dprintk("svc: svc_process close\n");
1445 return 0;
1446
1447 err_short_len:
1448 svc_printk(rqstp, "short len %u, dropping request\n",
1449 rqstp->rq_arg.len);
1450 goto close_xprt;
1451
1452 err_bad_rpc:
1453 if (serv->sv_stats)
1454 serv->sv_stats->rpcbadfmt++;
1455 xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1456 xdr_stream_encode_u32(xdr, RPC_MISMATCH);
1457 /* Only RPCv2 supported */
1458 xdr_stream_encode_u32(xdr, RPC_VERSION);
1459 xdr_stream_encode_u32(xdr, RPC_VERSION);
1460 return 1; /* don't wrap */
1461
1462 err_bad_auth:
1463 dprintk("svc: authentication failed (%d)\n",
1464 be32_to_cpu(rqstp->rq_auth_stat));
1465 if (serv->sv_stats)
1466 serv->sv_stats->rpcbadauth++;
1467 /* Restore write pointer to location of reply status: */
1468 xdr_truncate_encode(xdr, XDR_UNIT * 2);
1469 xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1470 xdr_stream_encode_u32(xdr, RPC_AUTH_ERROR);
1471 xdr_stream_encode_be32(xdr, rqstp->rq_auth_stat);
1472 goto sendit;
1473
1474 err_bad_prog:
1475 dprintk("svc: unknown program %d\n", rqstp->rq_prog);
1476 if (serv->sv_stats)
1477 serv->sv_stats->rpcbadfmt++;
1478 *rqstp->rq_accept_statp = rpc_prog_unavail;
1479 goto sendit;
1480
1481 err_bad_vers:
1482 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1483 rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1484
1485 if (serv->sv_stats)
1486 serv->sv_stats->rpcbadfmt++;
1487 *rqstp->rq_accept_statp = rpc_prog_mismatch;
1488
1489 /*
1490 * svc_authenticate() has already added the verifier and
1491 * advanced the stream just past rq_accept_statp.
1492 */
1493 xdr_stream_encode_u32(xdr, process.mismatch.lovers);
1494 xdr_stream_encode_u32(xdr, process.mismatch.hivers);
1495 goto sendit;
1496
1497 err_bad_proc:
1498 svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1499
1500 if (serv->sv_stats)
1501 serv->sv_stats->rpcbadfmt++;
1502 *rqstp->rq_accept_statp = rpc_proc_unavail;
1503 goto sendit;
1504
1505 err_garbage_args:
1506 svc_printk(rqstp, "failed to decode RPC header\n");
1507
1508 if (serv->sv_stats)
1509 serv->sv_stats->rpcbadfmt++;
1510 *rqstp->rq_accept_statp = rpc_garbage_args;
1511 goto sendit;
1512
1513 err_system_err:
1514 if (serv->sv_stats)
1515 serv->sv_stats->rpcbadfmt++;
1516 *rqstp->rq_accept_statp = rpc_system_err;
1517 goto sendit;
1518 }
1519
1520 /**
1521 * svc_process - Execute one RPC transaction
1522 * @rqstp: RPC transaction context
1523 *
1524 */
1525 void svc_process(struct svc_rqst *rqstp)
1526 {
1527 struct kvec *resv = &rqstp->rq_res.head[0];
1528 __be32 *p;
1529
1530 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1531 if (!fail_sunrpc.ignore_server_disconnect &&
1532 should_fail(&fail_sunrpc.attr, 1))
1533 svc_xprt_deferred_close(rqstp->rq_xprt);
1534 #endif
1535
1536 /*
1537 * Setup response xdr_buf.
1538 * Initially it has just one page
1539 */
1540 rqstp->rq_next_page = &rqstp->rq_respages[1];
1541 resv->iov_base = page_address(rqstp->rq_respages[0]);
1542 resv->iov_len = 0;
1543 rqstp->rq_res.pages = rqstp->rq_next_page;
1544 rqstp->rq_res.len = 0;
1545 rqstp->rq_res.page_base = 0;
1546 rqstp->rq_res.page_len = 0;
1547 rqstp->rq_res.buflen = PAGE_SIZE;
1548 rqstp->rq_res.tail[0].iov_base = NULL;
1549 rqstp->rq_res.tail[0].iov_len = 0;
1550
1551 svcxdr_init_decode(rqstp);
1552 p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2);
1553 if (unlikely(!p))
1554 goto out_drop;
1555 rqstp->rq_xid = *p++;
1556 if (unlikely(*p != rpc_call))
1557 goto out_baddir;
1558
1559 if (!svc_process_common(rqstp))
1560 goto out_drop;
1561 svc_send(rqstp);
1562 return;
1563
1564 out_baddir:
1565 svc_printk(rqstp, "bad direction 0x%08x, dropping request\n",
1566 be32_to_cpu(*p));
1567 if (rqstp->rq_server->sv_stats)
1568 rqstp->rq_server->sv_stats->rpcbadfmt++;
1569 out_drop:
1570 svc_drop(rqstp);
1571 }
1572
1573 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1574 /**
1575 * svc_process_bc - process a reverse-direction RPC request
1576 * @req: RPC request to be used for client-side processing
1577 * @rqstp: server-side execution context
1578 *
1579 */
1580 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp)
1581 {
1582 struct rpc_timeout timeout = {
1583 .to_increment = 0,
1584 };
1585 struct rpc_task *task;
1586 int proc_error;
1587
1588 /* Build the svc_rqst used by the common processing routine */
1589 rqstp->rq_xid = req->rq_xid;
1590 rqstp->rq_prot = req->rq_xprt->prot;
1591 rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1592
1593 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1594 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1595 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1596 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1597
1598 /* Adjust the argument buffer length */
1599 rqstp->rq_arg.len = req->rq_private_buf.len;
1600 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1601 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1602 rqstp->rq_arg.page_len = 0;
1603 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1604 rqstp->rq_arg.page_len)
1605 rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1606 rqstp->rq_arg.head[0].iov_len;
1607 else
1608 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1609 rqstp->rq_arg.page_len;
1610
1611 /* Reset the response buffer */
1612 rqstp->rq_res.head[0].iov_len = 0;
1613
1614 /*
1615 * Skip the XID and calldir fields because they've already
1616 * been processed by the caller.
1617 */
1618 svcxdr_init_decode(rqstp);
1619 if (!xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2))
1620 return;
1621
1622 /* Parse and execute the bc call */
1623 proc_error = svc_process_common(rqstp);
1624
1625 atomic_dec(&req->rq_xprt->bc_slot_count);
1626 if (!proc_error) {
1627 /* Processing error: drop the request */
1628 xprt_free_bc_request(req);
1629 return;
1630 }
1631 /* Finally, send the reply synchronously */
1632 if (rqstp->bc_to_initval > 0) {
1633 timeout.to_initval = rqstp->bc_to_initval;
1634 timeout.to_retries = rqstp->bc_to_retries;
1635 } else {
1636 timeout.to_initval = req->rq_xprt->timeout->to_initval;
1637 timeout.to_retries = req->rq_xprt->timeout->to_retries;
1638 }
1639 timeout.to_maxval = timeout.to_initval;
1640 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1641 task = rpc_run_bc_task(req, &timeout);
1642
1643 if (IS_ERR(task))
1644 return;
1645
1646 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1647 rpc_put_task(task);
1648 }
1649 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1650
1651 /**
1652 * svc_max_payload - Return transport-specific limit on the RPC payload
1653 * @rqstp: RPC transaction context
1654 *
1655 * Returns the maximum number of payload bytes the current transport
1656 * allows.
1657 */
1658 u32 svc_max_payload(const struct svc_rqst *rqstp)
1659 {
1660 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1661
1662 if (rqstp->rq_server->sv_max_payload < max)
1663 max = rqstp->rq_server->sv_max_payload;
1664 return max;
1665 }
1666 EXPORT_SYMBOL_GPL(svc_max_payload);
1667
1668 /**
1669 * svc_proc_name - Return RPC procedure name in string form
1670 * @rqstp: svc_rqst to operate on
1671 *
1672 * Return value:
1673 * Pointer to a NUL-terminated string
1674 */
1675 const char *svc_proc_name(const struct svc_rqst *rqstp)
1676 {
1677 if (rqstp && rqstp->rq_procinfo)
1678 return rqstp->rq_procinfo->pc_name;
1679 return "unknown";
1680 }
1681
1682
1683 /**
1684 * svc_encode_result_payload - mark a range of bytes as a result payload
1685 * @rqstp: svc_rqst to operate on
1686 * @offset: payload's byte offset in rqstp->rq_res
1687 * @length: size of payload, in bytes
1688 *
1689 * Returns zero on success, or a negative errno if a permanent
1690 * error occurred.
1691 */
1692 int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1693 unsigned int length)
1694 {
1695 return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1696 length);
1697 }
1698 EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1699
1700 /**
1701 * svc_fill_write_vector - Construct data argument for VFS write call
1702 * @rqstp: svc_rqst to operate on
1703 * @payload: xdr_buf containing only the write data payload
1704 *
1705 * Fills in rqstp::rq_vec, and returns the number of elements.
1706 */
1707 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1708 struct xdr_buf *payload)
1709 {
1710 struct page **pages = payload->pages;
1711 struct kvec *first = payload->head;
1712 struct kvec *vec = rqstp->rq_vec;
1713 size_t total = payload->len;
1714 unsigned int i;
1715
1716 /* Some types of transport can present the write payload
1717 * entirely in rq_arg.pages. In this case, @first is empty.
1718 */
1719 i = 0;
1720 if (first->iov_len) {
1721 vec[i].iov_base = first->iov_base;
1722 vec[i].iov_len = min_t(size_t, total, first->iov_len);
1723 total -= vec[i].iov_len;
1724 ++i;
1725 }
1726
1727 while (total) {
1728 vec[i].iov_base = page_address(*pages);
1729 vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1730 total -= vec[i].iov_len;
1731 ++i;
1732 ++pages;
1733 }
1734
1735 WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1736 return i;
1737 }
1738 EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1739
1740 /**
1741 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1742 * @rqstp: svc_rqst to operate on
1743 * @first: buffer containing first section of pathname
1744 * @p: buffer containing remaining section of pathname
1745 * @total: total length of the pathname argument
1746 *
1747 * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1748 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1749 * the returned string.
1750 */
1751 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1752 void *p, size_t total)
1753 {
1754 size_t len, remaining;
1755 char *result, *dst;
1756
1757 result = kmalloc(total + 1, GFP_KERNEL);
1758 if (!result)
1759 return ERR_PTR(-ESERVERFAULT);
1760
1761 dst = result;
1762 remaining = total;
1763
1764 len = min_t(size_t, total, first->iov_len);
1765 if (len) {
1766 memcpy(dst, first->iov_base, len);
1767 dst += len;
1768 remaining -= len;
1769 }
1770
1771 if (remaining) {
1772 len = min_t(size_t, remaining, PAGE_SIZE);
1773 memcpy(dst, p, len);
1774 dst += len;
1775 }
1776
1777 *dst = '\0';
1778
1779 /* Sanity check: Linux doesn't allow the pathname argument to
1780 * contain a NUL byte.
1781 */
1782 if (strlen(result) != total) {
1783 kfree(result);
1784 return ERR_PTR(-EINVAL);
1785 }
1786 return result;
1787 }
1788 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1789
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.