xref: /linux/net/sunrpc/svc.c (revision d53b8e36925256097a08d7cb749198d85cbf9b2b)
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 	atomic_inc(&pool->sp_nrthreads);
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 retry:
772 	pool = target_pool;
773 
774 	if (pool != NULL) {
775 		if (atomic_inc_not_zero(&pool->sp_nrthreads))
776 			goto found_pool;
777 		return NULL;
778 	} else {
779 		for (i = 0; i < serv->sv_nrpools; i++) {
780 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
781 			if (atomic_inc_not_zero(&pool->sp_nrthreads))
782 				goto found_pool;
783 		}
784 		return NULL;
785 	}
786 
787 found_pool:
788 	set_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
789 	set_bit(SP_NEED_VICTIM, &pool->sp_flags);
790 	if (!atomic_dec_and_test(&pool->sp_nrthreads))
791 		return pool;
792 	/* Nothing left in this pool any more */
793 	clear_bit(SP_NEED_VICTIM, &pool->sp_flags);
794 	clear_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
795 	goto retry;
796 }
797 
798 static int
799 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
800 {
801 	struct svc_rqst	*rqstp;
802 	struct task_struct *task;
803 	struct svc_pool *chosen_pool;
804 	unsigned int state = serv->sv_nrthreads-1;
805 	int node;
806 
807 	do {
808 		nrservs--;
809 		chosen_pool = svc_pool_next(serv, pool, &state);
810 		node = svc_pool_map_get_node(chosen_pool->sp_id);
811 
812 		rqstp = svc_prepare_thread(serv, chosen_pool, node);
813 		if (IS_ERR(rqstp))
814 			return PTR_ERR(rqstp);
815 		task = kthread_create_on_node(serv->sv_threadfn, rqstp,
816 					      node, "%s", serv->sv_name);
817 		if (IS_ERR(task)) {
818 			svc_exit_thread(rqstp);
819 			return PTR_ERR(task);
820 		}
821 
822 		rqstp->rq_task = task;
823 		if (serv->sv_nrpools > 1)
824 			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
825 
826 		svc_sock_update_bufs(serv);
827 		wake_up_process(task);
828 	} while (nrservs > 0);
829 
830 	return 0;
831 }
832 
833 static int
834 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
835 {
836 	unsigned int state = serv->sv_nrthreads-1;
837 	struct svc_pool *victim;
838 
839 	do {
840 		victim = svc_pool_victim(serv, pool, &state);
841 		if (!victim)
842 			break;
843 		svc_pool_wake_idle_thread(victim);
844 		wait_on_bit(&victim->sp_flags, SP_VICTIM_REMAINS,
845 			    TASK_IDLE);
846 		nrservs++;
847 	} while (nrservs < 0);
848 	return 0;
849 }
850 
851 /**
852  * svc_set_num_threads - adjust number of threads per RPC service
853  * @serv: RPC service to adjust
854  * @pool: Specific pool from which to choose threads, or NULL
855  * @nrservs: New number of threads for @serv (0 or less means kill all threads)
856  *
857  * Create or destroy threads to make the number of threads for @serv the
858  * given number. If @pool is non-NULL, change only threads in that pool;
859  * otherwise, round-robin between all pools for @serv. @serv's
860  * sv_nrthreads is adjusted for each thread created or destroyed.
861  *
862  * Caller must ensure mutual exclusion between this and server startup or
863  * shutdown.
864  *
865  * Returns zero on success or a negative errno if an error occurred while
866  * starting a thread.
867  */
868 int
869 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
870 {
871 	if (!pool)
872 		nrservs -= serv->sv_nrthreads;
873 	else
874 		nrservs -= atomic_read(&pool->sp_nrthreads);
875 
876 	if (nrservs > 0)
877 		return svc_start_kthreads(serv, pool, nrservs);
878 	if (nrservs < 0)
879 		return svc_stop_kthreads(serv, pool, nrservs);
880 	return 0;
881 }
882 EXPORT_SYMBOL_GPL(svc_set_num_threads);
883 
884 /**
885  * svc_rqst_replace_page - Replace one page in rq_pages[]
886  * @rqstp: svc_rqst with pages to replace
887  * @page: replacement page
888  *
889  * When replacing a page in rq_pages, batch the release of the
890  * replaced pages to avoid hammering the page allocator.
891  *
892  * Return values:
893  *   %true: page replaced
894  *   %false: array bounds checking failed
895  */
896 bool svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
897 {
898 	struct page **begin = rqstp->rq_pages;
899 	struct page **end = &rqstp->rq_pages[RPCSVC_MAXPAGES];
900 
901 	if (unlikely(rqstp->rq_next_page < begin || rqstp->rq_next_page > end)) {
902 		trace_svc_replace_page_err(rqstp);
903 		return false;
904 	}
905 
906 	if (*rqstp->rq_next_page) {
907 		if (!folio_batch_add(&rqstp->rq_fbatch,
908 				page_folio(*rqstp->rq_next_page)))
909 			__folio_batch_release(&rqstp->rq_fbatch);
910 	}
911 
912 	get_page(page);
913 	*(rqstp->rq_next_page++) = page;
914 	return true;
915 }
916 EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
917 
918 /**
919  * svc_rqst_release_pages - Release Reply buffer pages
920  * @rqstp: RPC transaction context
921  *
922  * Release response pages that might still be in flight after
923  * svc_send, and any spliced filesystem-owned pages.
924  */
925 void svc_rqst_release_pages(struct svc_rqst *rqstp)
926 {
927 	int i, count = rqstp->rq_next_page - rqstp->rq_respages;
928 
929 	if (count) {
930 		release_pages(rqstp->rq_respages, count);
931 		for (i = 0; i < count; i++)
932 			rqstp->rq_respages[i] = NULL;
933 	}
934 }
935 
936 /*
937  * Called from a server thread as it's exiting. Caller must hold the "service
938  * mutex" for the service.
939  */
940 void
941 svc_rqst_free(struct svc_rqst *rqstp)
942 {
943 	folio_batch_release(&rqstp->rq_fbatch);
944 	svc_release_buffer(rqstp);
945 	if (rqstp->rq_scratch_page)
946 		put_page(rqstp->rq_scratch_page);
947 	kfree(rqstp->rq_resp);
948 	kfree(rqstp->rq_argp);
949 	kfree(rqstp->rq_auth_data);
950 	kfree_rcu(rqstp, rq_rcu_head);
951 }
952 EXPORT_SYMBOL_GPL(svc_rqst_free);
953 
954 void
955 svc_exit_thread(struct svc_rqst *rqstp)
956 {
957 	struct svc_serv	*serv = rqstp->rq_server;
958 	struct svc_pool	*pool = rqstp->rq_pool;
959 
960 	list_del_rcu(&rqstp->rq_all);
961 
962 	atomic_dec(&pool->sp_nrthreads);
963 
964 	spin_lock_bh(&serv->sv_lock);
965 	serv->sv_nrthreads -= 1;
966 	spin_unlock_bh(&serv->sv_lock);
967 	svc_sock_update_bufs(serv);
968 
969 	svc_rqst_free(rqstp);
970 
971 	clear_and_wake_up_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
972 }
973 EXPORT_SYMBOL_GPL(svc_exit_thread);
974 
975 /*
976  * Register an "inet" protocol family netid with the local
977  * rpcbind daemon via an rpcbind v4 SET request.
978  *
979  * No netconfig infrastructure is available in the kernel, so
980  * we map IP_ protocol numbers to netids by hand.
981  *
982  * Returns zero on success; a negative errno value is returned
983  * if any error occurs.
984  */
985 static int __svc_rpcb_register4(struct net *net, const u32 program,
986 				const u32 version,
987 				const unsigned short protocol,
988 				const unsigned short port)
989 {
990 	const struct sockaddr_in sin = {
991 		.sin_family		= AF_INET,
992 		.sin_addr.s_addr	= htonl(INADDR_ANY),
993 		.sin_port		= htons(port),
994 	};
995 	const char *netid;
996 	int error;
997 
998 	switch (protocol) {
999 	case IPPROTO_UDP:
1000 		netid = RPCBIND_NETID_UDP;
1001 		break;
1002 	case IPPROTO_TCP:
1003 		netid = RPCBIND_NETID_TCP;
1004 		break;
1005 	default:
1006 		return -ENOPROTOOPT;
1007 	}
1008 
1009 	error = rpcb_v4_register(net, program, version,
1010 					(const struct sockaddr *)&sin, netid);
1011 
1012 	/*
1013 	 * User space didn't support rpcbind v4, so retry this
1014 	 * registration request with the legacy rpcbind v2 protocol.
1015 	 */
1016 	if (error == -EPROTONOSUPPORT)
1017 		error = rpcb_register(net, program, version, protocol, port);
1018 
1019 	return error;
1020 }
1021 
1022 #if IS_ENABLED(CONFIG_IPV6)
1023 /*
1024  * Register an "inet6" protocol family netid with the local
1025  * rpcbind daemon via an rpcbind v4 SET request.
1026  *
1027  * No netconfig infrastructure is available in the kernel, so
1028  * we map IP_ protocol numbers to netids by hand.
1029  *
1030  * Returns zero on success; a negative errno value is returned
1031  * if any error occurs.
1032  */
1033 static int __svc_rpcb_register6(struct net *net, const u32 program,
1034 				const u32 version,
1035 				const unsigned short protocol,
1036 				const unsigned short port)
1037 {
1038 	const struct sockaddr_in6 sin6 = {
1039 		.sin6_family		= AF_INET6,
1040 		.sin6_addr		= IN6ADDR_ANY_INIT,
1041 		.sin6_port		= htons(port),
1042 	};
1043 	const char *netid;
1044 	int error;
1045 
1046 	switch (protocol) {
1047 	case IPPROTO_UDP:
1048 		netid = RPCBIND_NETID_UDP6;
1049 		break;
1050 	case IPPROTO_TCP:
1051 		netid = RPCBIND_NETID_TCP6;
1052 		break;
1053 	default:
1054 		return -ENOPROTOOPT;
1055 	}
1056 
1057 	error = rpcb_v4_register(net, program, version,
1058 					(const struct sockaddr *)&sin6, netid);
1059 
1060 	/*
1061 	 * User space didn't support rpcbind version 4, so we won't
1062 	 * use a PF_INET6 listener.
1063 	 */
1064 	if (error == -EPROTONOSUPPORT)
1065 		error = -EAFNOSUPPORT;
1066 
1067 	return error;
1068 }
1069 #endif	/* IS_ENABLED(CONFIG_IPV6) */
1070 
1071 /*
1072  * Register a kernel RPC service via rpcbind version 4.
1073  *
1074  * Returns zero on success; a negative errno value is returned
1075  * if any error occurs.
1076  */
1077 static int __svc_register(struct net *net, const char *progname,
1078 			  const u32 program, const u32 version,
1079 			  const int family,
1080 			  const unsigned short protocol,
1081 			  const unsigned short port)
1082 {
1083 	int error = -EAFNOSUPPORT;
1084 
1085 	switch (family) {
1086 	case PF_INET:
1087 		error = __svc_rpcb_register4(net, program, version,
1088 						protocol, port);
1089 		break;
1090 #if IS_ENABLED(CONFIG_IPV6)
1091 	case PF_INET6:
1092 		error = __svc_rpcb_register6(net, program, version,
1093 						protocol, port);
1094 #endif
1095 	}
1096 
1097 	trace_svc_register(progname, version, family, protocol, port, error);
1098 	return error;
1099 }
1100 
1101 int svc_rpcbind_set_version(struct net *net,
1102 			    const struct svc_program *progp,
1103 			    u32 version, int family,
1104 			    unsigned short proto,
1105 			    unsigned short port)
1106 {
1107 	return __svc_register(net, progp->pg_name, progp->pg_prog,
1108 				version, family, proto, port);
1109 
1110 }
1111 EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1112 
1113 int svc_generic_rpcbind_set(struct net *net,
1114 			    const struct svc_program *progp,
1115 			    u32 version, int family,
1116 			    unsigned short proto,
1117 			    unsigned short port)
1118 {
1119 	const struct svc_version *vers = progp->pg_vers[version];
1120 	int error;
1121 
1122 	if (vers == NULL)
1123 		return 0;
1124 
1125 	if (vers->vs_hidden) {
1126 		trace_svc_noregister(progp->pg_name, version, proto,
1127 				     port, family, 0);
1128 		return 0;
1129 	}
1130 
1131 	/*
1132 	 * Don't register a UDP port if we need congestion
1133 	 * control.
1134 	 */
1135 	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1136 		return 0;
1137 
1138 	error = svc_rpcbind_set_version(net, progp, version,
1139 					family, proto, port);
1140 
1141 	return (vers->vs_rpcb_optnl) ? 0 : error;
1142 }
1143 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1144 
1145 /**
1146  * svc_register - register an RPC service with the local portmapper
1147  * @serv: svc_serv struct for the service to register
1148  * @net: net namespace for the service to register
1149  * @family: protocol family of service's listener socket
1150  * @proto: transport protocol number to advertise
1151  * @port: port to advertise
1152  *
1153  * Service is registered for any address in the passed-in protocol family
1154  */
1155 int svc_register(const struct svc_serv *serv, struct net *net,
1156 		 const int family, const unsigned short proto,
1157 		 const unsigned short port)
1158 {
1159 	struct svc_program	*progp;
1160 	unsigned int		i;
1161 	int			error = 0;
1162 
1163 	WARN_ON_ONCE(proto == 0 && port == 0);
1164 	if (proto == 0 && port == 0)
1165 		return -EINVAL;
1166 
1167 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1168 		for (i = 0; i < progp->pg_nvers; i++) {
1169 
1170 			error = progp->pg_rpcbind_set(net, progp, i,
1171 					family, proto, port);
1172 			if (error < 0) {
1173 				printk(KERN_WARNING "svc: failed to register "
1174 					"%sv%u RPC service (errno %d).\n",
1175 					progp->pg_name, i, -error);
1176 				break;
1177 			}
1178 		}
1179 	}
1180 
1181 	return error;
1182 }
1183 
1184 /*
1185  * If user space is running rpcbind, it should take the v4 UNSET
1186  * and clear everything for this [program, version].  If user space
1187  * is running portmap, it will reject the v4 UNSET, but won't have
1188  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1189  * in this case to clear all existing entries for [program, version].
1190  */
1191 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1192 			     const char *progname)
1193 {
1194 	int error;
1195 
1196 	error = rpcb_v4_register(net, program, version, NULL, "");
1197 
1198 	/*
1199 	 * User space didn't support rpcbind v4, so retry this
1200 	 * request with the legacy rpcbind v2 protocol.
1201 	 */
1202 	if (error == -EPROTONOSUPPORT)
1203 		error = rpcb_register(net, program, version, 0, 0);
1204 
1205 	trace_svc_unregister(progname, version, error);
1206 }
1207 
1208 /*
1209  * All netids, bind addresses and ports registered for [program, version]
1210  * are removed from the local rpcbind database (if the service is not
1211  * hidden) to make way for a new instance of the service.
1212  *
1213  * The result of unregistration is reported via dprintk for those who want
1214  * verification of the result, but is otherwise not important.
1215  */
1216 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1217 {
1218 	struct sighand_struct *sighand;
1219 	struct svc_program *progp;
1220 	unsigned long flags;
1221 	unsigned int i;
1222 
1223 	clear_thread_flag(TIF_SIGPENDING);
1224 
1225 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1226 		for (i = 0; i < progp->pg_nvers; i++) {
1227 			if (progp->pg_vers[i] == NULL)
1228 				continue;
1229 			if (progp->pg_vers[i]->vs_hidden)
1230 				continue;
1231 			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1232 		}
1233 	}
1234 
1235 	rcu_read_lock();
1236 	sighand = rcu_dereference(current->sighand);
1237 	spin_lock_irqsave(&sighand->siglock, flags);
1238 	recalc_sigpending();
1239 	spin_unlock_irqrestore(&sighand->siglock, flags);
1240 	rcu_read_unlock();
1241 }
1242 
1243 /*
1244  * dprintk the given error with the address of the client that caused it.
1245  */
1246 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1247 static __printf(2, 3)
1248 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1249 {
1250 	struct va_format vaf;
1251 	va_list args;
1252 	char 	buf[RPC_MAX_ADDRBUFLEN];
1253 
1254 	va_start(args, fmt);
1255 
1256 	vaf.fmt = fmt;
1257 	vaf.va = &args;
1258 
1259 	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1260 
1261 	va_end(args);
1262 }
1263 #else
1264 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1265 #endif
1266 
1267 __be32
1268 svc_generic_init_request(struct svc_rqst *rqstp,
1269 		const struct svc_program *progp,
1270 		struct svc_process_info *ret)
1271 {
1272 	const struct svc_version *versp = NULL;	/* compiler food */
1273 	const struct svc_procedure *procp = NULL;
1274 
1275 	if (rqstp->rq_vers >= progp->pg_nvers )
1276 		goto err_bad_vers;
1277 	versp = progp->pg_vers[rqstp->rq_vers];
1278 	if (!versp)
1279 		goto err_bad_vers;
1280 
1281 	/*
1282 	 * Some protocol versions (namely NFSv4) require some form of
1283 	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1284 	 * In other words, UDP is not allowed. We mark those when setting
1285 	 * up the svc_xprt, and verify that here.
1286 	 *
1287 	 * The spec is not very clear about what error should be returned
1288 	 * when someone tries to access a server that is listening on UDP
1289 	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1290 	 * fit.
1291 	 */
1292 	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1293 	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1294 		goto err_bad_vers;
1295 
1296 	if (rqstp->rq_proc >= versp->vs_nproc)
1297 		goto err_bad_proc;
1298 	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1299 
1300 	/* Initialize storage for argp and resp */
1301 	memset(rqstp->rq_argp, 0, procp->pc_argzero);
1302 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1303 
1304 	/* Bump per-procedure stats counter */
1305 	this_cpu_inc(versp->vs_count[rqstp->rq_proc]);
1306 
1307 	ret->dispatch = versp->vs_dispatch;
1308 	return rpc_success;
1309 err_bad_vers:
1310 	ret->mismatch.lovers = progp->pg_lovers;
1311 	ret->mismatch.hivers = progp->pg_hivers;
1312 	return rpc_prog_mismatch;
1313 err_bad_proc:
1314 	return rpc_proc_unavail;
1315 }
1316 EXPORT_SYMBOL_GPL(svc_generic_init_request);
1317 
1318 /*
1319  * Common routine for processing the RPC request.
1320  */
1321 static int
1322 svc_process_common(struct svc_rqst *rqstp)
1323 {
1324 	struct xdr_stream	*xdr = &rqstp->rq_res_stream;
1325 	struct svc_program	*progp;
1326 	const struct svc_procedure *procp = NULL;
1327 	struct svc_serv		*serv = rqstp->rq_server;
1328 	struct svc_process_info process;
1329 	enum svc_auth_status	auth_res;
1330 	unsigned int		aoffset;
1331 	int			rc;
1332 	__be32			*p;
1333 
1334 	/* Will be turned off only when NFSv4 Sessions are used */
1335 	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1336 	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1337 
1338 	/* Construct the first words of the reply: */
1339 	svcxdr_init_encode(rqstp);
1340 	xdr_stream_encode_be32(xdr, rqstp->rq_xid);
1341 	xdr_stream_encode_be32(xdr, rpc_reply);
1342 
1343 	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 4);
1344 	if (unlikely(!p))
1345 		goto err_short_len;
1346 	if (*p++ != cpu_to_be32(RPC_VERSION))
1347 		goto err_bad_rpc;
1348 
1349 	xdr_stream_encode_be32(xdr, rpc_msg_accepted);
1350 
1351 	rqstp->rq_prog = be32_to_cpup(p++);
1352 	rqstp->rq_vers = be32_to_cpup(p++);
1353 	rqstp->rq_proc = be32_to_cpup(p);
1354 
1355 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1356 		if (rqstp->rq_prog == progp->pg_prog)
1357 			break;
1358 
1359 	/*
1360 	 * Decode auth data, and add verifier to reply buffer.
1361 	 * We do this before anything else in order to get a decent
1362 	 * auth verifier.
1363 	 */
1364 	auth_res = svc_authenticate(rqstp);
1365 	/* Also give the program a chance to reject this call: */
1366 	if (auth_res == SVC_OK && progp)
1367 		auth_res = progp->pg_authenticate(rqstp);
1368 	trace_svc_authenticate(rqstp, auth_res);
1369 	switch (auth_res) {
1370 	case SVC_OK:
1371 		break;
1372 	case SVC_GARBAGE:
1373 		goto err_garbage_args;
1374 	case SVC_SYSERR:
1375 		goto err_system_err;
1376 	case SVC_DENIED:
1377 		goto err_bad_auth;
1378 	case SVC_CLOSE:
1379 		goto close;
1380 	case SVC_DROP:
1381 		goto dropit;
1382 	case SVC_COMPLETE:
1383 		goto sendit;
1384 	default:
1385 		pr_warn_once("Unexpected svc_auth_status (%d)\n", auth_res);
1386 		goto err_system_err;
1387 	}
1388 
1389 	if (progp == NULL)
1390 		goto err_bad_prog;
1391 
1392 	switch (progp->pg_init_request(rqstp, progp, &process)) {
1393 	case rpc_success:
1394 		break;
1395 	case rpc_prog_unavail:
1396 		goto err_bad_prog;
1397 	case rpc_prog_mismatch:
1398 		goto err_bad_vers;
1399 	case rpc_proc_unavail:
1400 		goto err_bad_proc;
1401 	}
1402 
1403 	procp = rqstp->rq_procinfo;
1404 	/* Should this check go into the dispatcher? */
1405 	if (!procp || !procp->pc_func)
1406 		goto err_bad_proc;
1407 
1408 	/* Syntactic check complete */
1409 	if (serv->sv_stats)
1410 		serv->sv_stats->rpccnt++;
1411 	trace_svc_process(rqstp, progp->pg_name);
1412 
1413 	aoffset = xdr_stream_pos(xdr);
1414 
1415 	/* un-reserve some of the out-queue now that we have a
1416 	 * better idea of reply size
1417 	 */
1418 	if (procp->pc_xdrressize)
1419 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1420 
1421 	/* Call the function that processes the request. */
1422 	rc = process.dispatch(rqstp);
1423 	if (procp->pc_release)
1424 		procp->pc_release(rqstp);
1425 	xdr_finish_decode(xdr);
1426 
1427 	if (!rc)
1428 		goto dropit;
1429 	if (rqstp->rq_auth_stat != rpc_auth_ok)
1430 		goto err_bad_auth;
1431 
1432 	if (*rqstp->rq_accept_statp != rpc_success)
1433 		xdr_truncate_encode(xdr, aoffset);
1434 
1435 	if (procp->pc_encode == NULL)
1436 		goto dropit;
1437 
1438  sendit:
1439 	if (svc_authorise(rqstp))
1440 		goto close_xprt;
1441 	return 1;		/* Caller can now send it */
1442 
1443  dropit:
1444 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1445 	dprintk("svc: svc_process dropit\n");
1446 	return 0;
1447 
1448  close:
1449 	svc_authorise(rqstp);
1450 close_xprt:
1451 	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1452 		svc_xprt_close(rqstp->rq_xprt);
1453 	dprintk("svc: svc_process close\n");
1454 	return 0;
1455 
1456 err_short_len:
1457 	svc_printk(rqstp, "short len %u, dropping request\n",
1458 		   rqstp->rq_arg.len);
1459 	goto close_xprt;
1460 
1461 err_bad_rpc:
1462 	if (serv->sv_stats)
1463 		serv->sv_stats->rpcbadfmt++;
1464 	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1465 	xdr_stream_encode_u32(xdr, RPC_MISMATCH);
1466 	/* Only RPCv2 supported */
1467 	xdr_stream_encode_u32(xdr, RPC_VERSION);
1468 	xdr_stream_encode_u32(xdr, RPC_VERSION);
1469 	return 1;	/* don't wrap */
1470 
1471 err_bad_auth:
1472 	dprintk("svc: authentication failed (%d)\n",
1473 		be32_to_cpu(rqstp->rq_auth_stat));
1474 	if (serv->sv_stats)
1475 		serv->sv_stats->rpcbadauth++;
1476 	/* Restore write pointer to location of reply status: */
1477 	xdr_truncate_encode(xdr, XDR_UNIT * 2);
1478 	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1479 	xdr_stream_encode_u32(xdr, RPC_AUTH_ERROR);
1480 	xdr_stream_encode_be32(xdr, rqstp->rq_auth_stat);
1481 	goto sendit;
1482 
1483 err_bad_prog:
1484 	dprintk("svc: unknown program %d\n", rqstp->rq_prog);
1485 	if (serv->sv_stats)
1486 		serv->sv_stats->rpcbadfmt++;
1487 	*rqstp->rq_accept_statp = rpc_prog_unavail;
1488 	goto sendit;
1489 
1490 err_bad_vers:
1491 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1492 		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1493 
1494 	if (serv->sv_stats)
1495 		serv->sv_stats->rpcbadfmt++;
1496 	*rqstp->rq_accept_statp = rpc_prog_mismatch;
1497 
1498 	/*
1499 	 * svc_authenticate() has already added the verifier and
1500 	 * advanced the stream just past rq_accept_statp.
1501 	 */
1502 	xdr_stream_encode_u32(xdr, process.mismatch.lovers);
1503 	xdr_stream_encode_u32(xdr, process.mismatch.hivers);
1504 	goto sendit;
1505 
1506 err_bad_proc:
1507 	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1508 
1509 	if (serv->sv_stats)
1510 		serv->sv_stats->rpcbadfmt++;
1511 	*rqstp->rq_accept_statp = rpc_proc_unavail;
1512 	goto sendit;
1513 
1514 err_garbage_args:
1515 	svc_printk(rqstp, "failed to decode RPC header\n");
1516 
1517 	if (serv->sv_stats)
1518 		serv->sv_stats->rpcbadfmt++;
1519 	*rqstp->rq_accept_statp = rpc_garbage_args;
1520 	goto sendit;
1521 
1522 err_system_err:
1523 	if (serv->sv_stats)
1524 		serv->sv_stats->rpcbadfmt++;
1525 	*rqstp->rq_accept_statp = rpc_system_err;
1526 	goto sendit;
1527 }
1528 
1529 /**
1530  * svc_process - Execute one RPC transaction
1531  * @rqstp: RPC transaction context
1532  *
1533  */
1534 void svc_process(struct svc_rqst *rqstp)
1535 {
1536 	struct kvec		*resv = &rqstp->rq_res.head[0];
1537 	__be32 *p;
1538 
1539 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1540 	if (!fail_sunrpc.ignore_server_disconnect &&
1541 	    should_fail(&fail_sunrpc.attr, 1))
1542 		svc_xprt_deferred_close(rqstp->rq_xprt);
1543 #endif
1544 
1545 	/*
1546 	 * Setup response xdr_buf.
1547 	 * Initially it has just one page
1548 	 */
1549 	rqstp->rq_next_page = &rqstp->rq_respages[1];
1550 	resv->iov_base = page_address(rqstp->rq_respages[0]);
1551 	resv->iov_len = 0;
1552 	rqstp->rq_res.pages = rqstp->rq_next_page;
1553 	rqstp->rq_res.len = 0;
1554 	rqstp->rq_res.page_base = 0;
1555 	rqstp->rq_res.page_len = 0;
1556 	rqstp->rq_res.buflen = PAGE_SIZE;
1557 	rqstp->rq_res.tail[0].iov_base = NULL;
1558 	rqstp->rq_res.tail[0].iov_len = 0;
1559 
1560 	svcxdr_init_decode(rqstp);
1561 	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2);
1562 	if (unlikely(!p))
1563 		goto out_drop;
1564 	rqstp->rq_xid = *p++;
1565 	if (unlikely(*p != rpc_call))
1566 		goto out_baddir;
1567 
1568 	if (!svc_process_common(rqstp))
1569 		goto out_drop;
1570 	svc_send(rqstp);
1571 	return;
1572 
1573 out_baddir:
1574 	svc_printk(rqstp, "bad direction 0x%08x, dropping request\n",
1575 		   be32_to_cpu(*p));
1576 	if (rqstp->rq_server->sv_stats)
1577 		rqstp->rq_server->sv_stats->rpcbadfmt++;
1578 out_drop:
1579 	svc_drop(rqstp);
1580 }
1581 
1582 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1583 /**
1584  * svc_process_bc - process a reverse-direction RPC request
1585  * @req: RPC request to be used for client-side processing
1586  * @rqstp: server-side execution context
1587  *
1588  */
1589 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp)
1590 {
1591 	struct rpc_timeout timeout = {
1592 		.to_increment		= 0,
1593 	};
1594 	struct rpc_task *task;
1595 	int proc_error;
1596 
1597 	/* Build the svc_rqst used by the common processing routine */
1598 	rqstp->rq_xid = req->rq_xid;
1599 	rqstp->rq_prot = req->rq_xprt->prot;
1600 	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1601 
1602 	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1603 	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1604 	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1605 	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1606 
1607 	/* Adjust the argument buffer length */
1608 	rqstp->rq_arg.len = req->rq_private_buf.len;
1609 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1610 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1611 		rqstp->rq_arg.page_len = 0;
1612 	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1613 			rqstp->rq_arg.page_len)
1614 		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1615 			rqstp->rq_arg.head[0].iov_len;
1616 	else
1617 		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1618 			rqstp->rq_arg.page_len;
1619 
1620 	/* Reset the response buffer */
1621 	rqstp->rq_res.head[0].iov_len = 0;
1622 
1623 	/*
1624 	 * Skip the XID and calldir fields because they've already
1625 	 * been processed by the caller.
1626 	 */
1627 	svcxdr_init_decode(rqstp);
1628 	if (!xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2))
1629 		return;
1630 
1631 	/* Parse and execute the bc call */
1632 	proc_error = svc_process_common(rqstp);
1633 
1634 	atomic_dec(&req->rq_xprt->bc_slot_count);
1635 	if (!proc_error) {
1636 		/* Processing error: drop the request */
1637 		xprt_free_bc_request(req);
1638 		return;
1639 	}
1640 	/* Finally, send the reply synchronously */
1641 	if (rqstp->bc_to_initval > 0) {
1642 		timeout.to_initval = rqstp->bc_to_initval;
1643 		timeout.to_retries = rqstp->bc_to_retries;
1644 	} else {
1645 		timeout.to_initval = req->rq_xprt->timeout->to_initval;
1646 		timeout.to_retries = req->rq_xprt->timeout->to_retries;
1647 	}
1648 	timeout.to_maxval = timeout.to_initval;
1649 	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1650 	task = rpc_run_bc_task(req, &timeout);
1651 
1652 	if (IS_ERR(task))
1653 		return;
1654 
1655 	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1656 	rpc_put_task(task);
1657 }
1658 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1659 
1660 /**
1661  * svc_max_payload - Return transport-specific limit on the RPC payload
1662  * @rqstp: RPC transaction context
1663  *
1664  * Returns the maximum number of payload bytes the current transport
1665  * allows.
1666  */
1667 u32 svc_max_payload(const struct svc_rqst *rqstp)
1668 {
1669 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1670 
1671 	if (rqstp->rq_server->sv_max_payload < max)
1672 		max = rqstp->rq_server->sv_max_payload;
1673 	return max;
1674 }
1675 EXPORT_SYMBOL_GPL(svc_max_payload);
1676 
1677 /**
1678  * svc_proc_name - Return RPC procedure name in string form
1679  * @rqstp: svc_rqst to operate on
1680  *
1681  * Return value:
1682  *   Pointer to a NUL-terminated string
1683  */
1684 const char *svc_proc_name(const struct svc_rqst *rqstp)
1685 {
1686 	if (rqstp && rqstp->rq_procinfo)
1687 		return rqstp->rq_procinfo->pc_name;
1688 	return "unknown";
1689 }
1690 
1691 
1692 /**
1693  * svc_encode_result_payload - mark a range of bytes as a result payload
1694  * @rqstp: svc_rqst to operate on
1695  * @offset: payload's byte offset in rqstp->rq_res
1696  * @length: size of payload, in bytes
1697  *
1698  * Returns zero on success, or a negative errno if a permanent
1699  * error occurred.
1700  */
1701 int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1702 			      unsigned int length)
1703 {
1704 	return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1705 							   length);
1706 }
1707 EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1708 
1709 /**
1710  * svc_fill_write_vector - Construct data argument for VFS write call
1711  * @rqstp: svc_rqst to operate on
1712  * @payload: xdr_buf containing only the write data payload
1713  *
1714  * Fills in rqstp::rq_vec, and returns the number of elements.
1715  */
1716 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1717 				   struct xdr_buf *payload)
1718 {
1719 	struct page **pages = payload->pages;
1720 	struct kvec *first = payload->head;
1721 	struct kvec *vec = rqstp->rq_vec;
1722 	size_t total = payload->len;
1723 	unsigned int i;
1724 
1725 	/* Some types of transport can present the write payload
1726 	 * entirely in rq_arg.pages. In this case, @first is empty.
1727 	 */
1728 	i = 0;
1729 	if (first->iov_len) {
1730 		vec[i].iov_base = first->iov_base;
1731 		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1732 		total -= vec[i].iov_len;
1733 		++i;
1734 	}
1735 
1736 	while (total) {
1737 		vec[i].iov_base = page_address(*pages);
1738 		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1739 		total -= vec[i].iov_len;
1740 		++i;
1741 		++pages;
1742 	}
1743 
1744 	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1745 	return i;
1746 }
1747 EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1748 
1749 /**
1750  * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1751  * @rqstp: svc_rqst to operate on
1752  * @first: buffer containing first section of pathname
1753  * @p: buffer containing remaining section of pathname
1754  * @total: total length of the pathname argument
1755  *
1756  * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1757  * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1758  * the returned string.
1759  */
1760 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1761 				void *p, size_t total)
1762 {
1763 	size_t len, remaining;
1764 	char *result, *dst;
1765 
1766 	result = kmalloc(total + 1, GFP_KERNEL);
1767 	if (!result)
1768 		return ERR_PTR(-ESERVERFAULT);
1769 
1770 	dst = result;
1771 	remaining = total;
1772 
1773 	len = min_t(size_t, total, first->iov_len);
1774 	if (len) {
1775 		memcpy(dst, first->iov_base, len);
1776 		dst += len;
1777 		remaining -= len;
1778 	}
1779 
1780 	if (remaining) {
1781 		len = min_t(size_t, remaining, PAGE_SIZE);
1782 		memcpy(dst, p, len);
1783 		dst += len;
1784 	}
1785 
1786 	*dst = '\0';
1787 
1788 	/* Sanity check: Linux doesn't allow the pathname argument to
1789 	 * contain a NUL byte.
1790 	 */
1791 	if (strlen(result) != total) {
1792 		kfree(result);
1793 		return ERR_PTR(-EINVAL);
1794 	}
1795 	return result;
1796 }
1797 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1798