xref: /linux/net/sunrpc/svc.c (revision f4738f56d1dc62aaba69b33702a5ab098f1b8c63)
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, const struct kernel_param *kp)
76 {
77 	int *ip = (int *)kp->arg;
78 	struct svc_pool_map *m = &svc_pool_map;
79 	int err;
80 
81 	mutex_lock(&svc_pool_map_mutex);
82 
83 	err = -EBUSY;
84 	if (m->count)
85 		goto out;
86 
87 	err = 0;
88 	if (!strncmp(val, "auto", 4))
89 		*ip = SVC_POOL_AUTO;
90 	else if (!strncmp(val, "global", 6))
91 		*ip = SVC_POOL_GLOBAL;
92 	else if (!strncmp(val, "percpu", 6))
93 		*ip = SVC_POOL_PERCPU;
94 	else if (!strncmp(val, "pernode", 7))
95 		*ip = SVC_POOL_PERNODE;
96 	else
97 		err = -EINVAL;
98 
99 out:
100 	mutex_unlock(&svc_pool_map_mutex);
101 	return err;
102 }
103 
104 static int
105 param_get_pool_mode(char *buf, const struct kernel_param *kp)
106 {
107 	int *ip = (int *)kp->arg;
108 
109 	switch (*ip)
110 	{
111 	case SVC_POOL_AUTO:
112 		return sysfs_emit(buf, "auto\n");
113 	case SVC_POOL_GLOBAL:
114 		return sysfs_emit(buf, "global\n");
115 	case SVC_POOL_PERCPU:
116 		return sysfs_emit(buf, "percpu\n");
117 	case SVC_POOL_PERNODE:
118 		return sysfs_emit(buf, "pernode\n");
119 	default:
120 		return sysfs_emit(buf, "%d\n", *ip);
121 	}
122 }
123 
124 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
125 		 &svc_pool_map.mode, 0644);
126 
127 /*
128  * Detect best pool mapping mode heuristically,
129  * according to the machine's topology.
130  */
131 static int
132 svc_pool_map_choose_mode(void)
133 {
134 	unsigned int node;
135 
136 	if (nr_online_nodes > 1) {
137 		/*
138 		 * Actually have multiple NUMA nodes,
139 		 * so split pools on NUMA node boundaries
140 		 */
141 		return SVC_POOL_PERNODE;
142 	}
143 
144 	node = first_online_node;
145 	if (nr_cpus_node(node) > 2) {
146 		/*
147 		 * Non-trivial SMP, or CONFIG_NUMA on
148 		 * non-NUMA hardware, e.g. with a generic
149 		 * x86_64 kernel on Xeons.  In this case we
150 		 * want to divide the pools on cpu boundaries.
151 		 */
152 		return SVC_POOL_PERCPU;
153 	}
154 
155 	/* default: one global pool */
156 	return SVC_POOL_GLOBAL;
157 }
158 
159 /*
160  * Allocate the to_pool[] and pool_to[] arrays.
161  * Returns 0 on success or an errno.
162  */
163 static int
164 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
165 {
166 	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
167 	if (!m->to_pool)
168 		goto fail;
169 	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
170 	if (!m->pool_to)
171 		goto fail_free;
172 
173 	return 0;
174 
175 fail_free:
176 	kfree(m->to_pool);
177 	m->to_pool = NULL;
178 fail:
179 	return -ENOMEM;
180 }
181 
182 /*
183  * Initialise the pool map for SVC_POOL_PERCPU mode.
184  * Returns number of pools or <0 on error.
185  */
186 static int
187 svc_pool_map_init_percpu(struct svc_pool_map *m)
188 {
189 	unsigned int maxpools = nr_cpu_ids;
190 	unsigned int pidx = 0;
191 	unsigned int cpu;
192 	int err;
193 
194 	err = svc_pool_map_alloc_arrays(m, maxpools);
195 	if (err)
196 		return err;
197 
198 	for_each_online_cpu(cpu) {
199 		BUG_ON(pidx >= maxpools);
200 		m->to_pool[cpu] = pidx;
201 		m->pool_to[pidx] = cpu;
202 		pidx++;
203 	}
204 	/* cpus brought online later all get mapped to pool0, sorry */
205 
206 	return pidx;
207 };
208 
209 
210 /*
211  * Initialise the pool map for SVC_POOL_PERNODE mode.
212  * Returns number of pools or <0 on error.
213  */
214 static int
215 svc_pool_map_init_pernode(struct svc_pool_map *m)
216 {
217 	unsigned int maxpools = nr_node_ids;
218 	unsigned int pidx = 0;
219 	unsigned int node;
220 	int err;
221 
222 	err = svc_pool_map_alloc_arrays(m, maxpools);
223 	if (err)
224 		return err;
225 
226 	for_each_node_with_cpus(node) {
227 		/* some architectures (e.g. SN2) have cpuless nodes */
228 		BUG_ON(pidx > maxpools);
229 		m->to_pool[node] = pidx;
230 		m->pool_to[pidx] = node;
231 		pidx++;
232 	}
233 	/* nodes brought online later all get mapped to pool0, sorry */
234 
235 	return pidx;
236 }
237 
238 
239 /*
240  * Add a reference to the global map of cpus to pools (and
241  * vice versa) if pools are in use.
242  * Initialise the map if we're the first user.
243  * Returns the number of pools. If this is '1', no reference
244  * was taken.
245  */
246 static unsigned int
247 svc_pool_map_get(void)
248 {
249 	struct svc_pool_map *m = &svc_pool_map;
250 	int npools = -1;
251 
252 	mutex_lock(&svc_pool_map_mutex);
253 
254 	if (m->count++) {
255 		mutex_unlock(&svc_pool_map_mutex);
256 		WARN_ON_ONCE(m->npools <= 1);
257 		return m->npools;
258 	}
259 
260 	if (m->mode == SVC_POOL_AUTO)
261 		m->mode = svc_pool_map_choose_mode();
262 
263 	switch (m->mode) {
264 	case SVC_POOL_PERCPU:
265 		npools = svc_pool_map_init_percpu(m);
266 		break;
267 	case SVC_POOL_PERNODE:
268 		npools = svc_pool_map_init_pernode(m);
269 		break;
270 	}
271 
272 	if (npools <= 0) {
273 		/* default, or memory allocation failure */
274 		npools = 1;
275 		m->mode = SVC_POOL_GLOBAL;
276 	}
277 	m->npools = npools;
278 
279 	if (npools == 1)
280 		/* service is unpooled, so doesn't hold a reference */
281 		m->count--;
282 
283 	mutex_unlock(&svc_pool_map_mutex);
284 	return npools;
285 }
286 
287 /*
288  * Drop a reference to the global map of cpus to pools, if
289  * pools were in use, i.e. if npools > 1.
290  * When the last reference is dropped, the map data is
291  * freed; this allows the sysadmin to change the pool
292  * mode using the pool_mode module option without
293  * rebooting or re-loading sunrpc.ko.
294  */
295 static void
296 svc_pool_map_put(int npools)
297 {
298 	struct svc_pool_map *m = &svc_pool_map;
299 
300 	if (npools <= 1)
301 		return;
302 	mutex_lock(&svc_pool_map_mutex);
303 
304 	if (!--m->count) {
305 		kfree(m->to_pool);
306 		m->to_pool = NULL;
307 		kfree(m->pool_to);
308 		m->pool_to = NULL;
309 		m->npools = 0;
310 	}
311 
312 	mutex_unlock(&svc_pool_map_mutex);
313 }
314 
315 static int svc_pool_map_get_node(unsigned int pidx)
316 {
317 	const struct svc_pool_map *m = &svc_pool_map;
318 
319 	if (m->count) {
320 		if (m->mode == SVC_POOL_PERCPU)
321 			return cpu_to_node(m->pool_to[pidx]);
322 		if (m->mode == SVC_POOL_PERNODE)
323 			return m->pool_to[pidx];
324 	}
325 	return NUMA_NO_NODE;
326 }
327 /*
328  * Set the given thread's cpus_allowed mask so that it
329  * will only run on cpus in the given pool.
330  */
331 static inline void
332 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
333 {
334 	struct svc_pool_map *m = &svc_pool_map;
335 	unsigned int node = m->pool_to[pidx];
336 
337 	/*
338 	 * The caller checks for sv_nrpools > 1, which
339 	 * implies that we've been initialized.
340 	 */
341 	WARN_ON_ONCE(m->count == 0);
342 	if (m->count == 0)
343 		return;
344 
345 	switch (m->mode) {
346 	case SVC_POOL_PERCPU:
347 	{
348 		set_cpus_allowed_ptr(task, cpumask_of(node));
349 		break;
350 	}
351 	case SVC_POOL_PERNODE:
352 	{
353 		set_cpus_allowed_ptr(task, cpumask_of_node(node));
354 		break;
355 	}
356 	}
357 }
358 
359 /**
360  * svc_pool_for_cpu - Select pool to run a thread on this cpu
361  * @serv: An RPC service
362  *
363  * Use the active CPU and the svc_pool_map's mode setting to
364  * select the svc thread pool to use. Once initialized, the
365  * svc_pool_map does not change.
366  *
367  * Return value:
368  *   A pointer to an svc_pool
369  */
370 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv)
371 {
372 	struct svc_pool_map *m = &svc_pool_map;
373 	int cpu = raw_smp_processor_id();
374 	unsigned int pidx = 0;
375 
376 	if (serv->sv_nrpools <= 1)
377 		return serv->sv_pools;
378 
379 	switch (m->mode) {
380 	case SVC_POOL_PERCPU:
381 		pidx = m->to_pool[cpu];
382 		break;
383 	case SVC_POOL_PERNODE:
384 		pidx = m->to_pool[cpu_to_node(cpu)];
385 		break;
386 	}
387 
388 	return &serv->sv_pools[pidx % serv->sv_nrpools];
389 }
390 
391 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
392 {
393 	int err;
394 
395 	err = rpcb_create_local(net);
396 	if (err)
397 		return err;
398 
399 	/* Remove any stale portmap registrations */
400 	svc_unregister(serv, net);
401 	return 0;
402 }
403 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
404 
405 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
406 {
407 	svc_unregister(serv, net);
408 	rpcb_put_local(net);
409 }
410 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
411 
412 static int svc_uses_rpcbind(struct svc_serv *serv)
413 {
414 	struct svc_program	*progp;
415 	unsigned int		i;
416 
417 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
418 		for (i = 0; i < progp->pg_nvers; i++) {
419 			if (progp->pg_vers[i] == NULL)
420 				continue;
421 			if (!progp->pg_vers[i]->vs_hidden)
422 				return 1;
423 		}
424 	}
425 
426 	return 0;
427 }
428 
429 int svc_bind(struct svc_serv *serv, struct net *net)
430 {
431 	if (!svc_uses_rpcbind(serv))
432 		return 0;
433 	return svc_rpcb_setup(serv, net);
434 }
435 EXPORT_SYMBOL_GPL(svc_bind);
436 
437 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
438 static void
439 __svc_init_bc(struct svc_serv *serv)
440 {
441 	INIT_LIST_HEAD(&serv->sv_cb_list);
442 	spin_lock_init(&serv->sv_cb_lock);
443 	init_waitqueue_head(&serv->sv_cb_waitq);
444 }
445 #else
446 static void
447 __svc_init_bc(struct svc_serv *serv)
448 {
449 }
450 #endif
451 
452 /*
453  * Create an RPC service
454  */
455 static struct svc_serv *
456 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
457 	     int (*threadfn)(void *data))
458 {
459 	struct svc_serv	*serv;
460 	unsigned int vers;
461 	unsigned int xdrsize;
462 	unsigned int i;
463 
464 	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
465 		return NULL;
466 	serv->sv_name      = prog->pg_name;
467 	serv->sv_program   = prog;
468 	kref_init(&serv->sv_refcnt);
469 	serv->sv_stats     = prog->pg_stats;
470 	if (bufsize > RPCSVC_MAXPAYLOAD)
471 		bufsize = RPCSVC_MAXPAYLOAD;
472 	serv->sv_max_payload = bufsize? bufsize : 4096;
473 	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
474 	serv->sv_threadfn = threadfn;
475 	xdrsize = 0;
476 	while (prog) {
477 		prog->pg_lovers = prog->pg_nvers-1;
478 		for (vers=0; vers<prog->pg_nvers ; vers++)
479 			if (prog->pg_vers[vers]) {
480 				prog->pg_hivers = vers;
481 				if (prog->pg_lovers > vers)
482 					prog->pg_lovers = vers;
483 				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
484 					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
485 			}
486 		prog = prog->pg_next;
487 	}
488 	serv->sv_xdrsize   = xdrsize;
489 	INIT_LIST_HEAD(&serv->sv_tempsocks);
490 	INIT_LIST_HEAD(&serv->sv_permsocks);
491 	timer_setup(&serv->sv_temptimer, NULL, 0);
492 	spin_lock_init(&serv->sv_lock);
493 
494 	__svc_init_bc(serv);
495 
496 	serv->sv_nrpools = npools;
497 	serv->sv_pools =
498 		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
499 			GFP_KERNEL);
500 	if (!serv->sv_pools) {
501 		kfree(serv);
502 		return NULL;
503 	}
504 
505 	for (i = 0; i < serv->sv_nrpools; i++) {
506 		struct svc_pool *pool = &serv->sv_pools[i];
507 
508 		dprintk("svc: initialising pool %u for %s\n",
509 				i, serv->sv_name);
510 
511 		pool->sp_id = i;
512 		INIT_LIST_HEAD(&pool->sp_sockets);
513 		INIT_LIST_HEAD(&pool->sp_all_threads);
514 		spin_lock_init(&pool->sp_lock);
515 
516 		percpu_counter_init(&pool->sp_messages_arrived, 0, GFP_KERNEL);
517 		percpu_counter_init(&pool->sp_sockets_queued, 0, GFP_KERNEL);
518 		percpu_counter_init(&pool->sp_threads_woken, 0, GFP_KERNEL);
519 	}
520 
521 	return serv;
522 }
523 
524 /**
525  * svc_create - Create an RPC service
526  * @prog: the RPC program the new service will handle
527  * @bufsize: maximum message size for @prog
528  * @threadfn: a function to service RPC requests for @prog
529  *
530  * Returns an instantiated struct svc_serv object or NULL.
531  */
532 struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize,
533 			    int (*threadfn)(void *data))
534 {
535 	return __svc_create(prog, bufsize, 1, threadfn);
536 }
537 EXPORT_SYMBOL_GPL(svc_create);
538 
539 /**
540  * svc_create_pooled - Create an RPC service with pooled threads
541  * @prog: the RPC program the new service will handle
542  * @bufsize: maximum message size for @prog
543  * @threadfn: a function to service RPC requests for @prog
544  *
545  * Returns an instantiated struct svc_serv object or NULL.
546  */
547 struct svc_serv *svc_create_pooled(struct svc_program *prog,
548 				   unsigned int bufsize,
549 				   int (*threadfn)(void *data))
550 {
551 	struct svc_serv *serv;
552 	unsigned int npools = svc_pool_map_get();
553 
554 	serv = __svc_create(prog, bufsize, npools, threadfn);
555 	if (!serv)
556 		goto out_err;
557 	return serv;
558 out_err:
559 	svc_pool_map_put(npools);
560 	return NULL;
561 }
562 EXPORT_SYMBOL_GPL(svc_create_pooled);
563 
564 /*
565  * Destroy an RPC service. Should be called with appropriate locking to
566  * protect sv_permsocks and sv_tempsocks.
567  */
568 void
569 svc_destroy(struct kref *ref)
570 {
571 	struct svc_serv *serv = container_of(ref, struct svc_serv, sv_refcnt);
572 	unsigned int i;
573 
574 	dprintk("svc: svc_destroy(%s)\n", serv->sv_program->pg_name);
575 	timer_shutdown_sync(&serv->sv_temptimer);
576 
577 	/*
578 	 * The last user is gone and thus all sockets have to be destroyed to
579 	 * the point. Check this.
580 	 */
581 	BUG_ON(!list_empty(&serv->sv_permsocks));
582 	BUG_ON(!list_empty(&serv->sv_tempsocks));
583 
584 	cache_clean_deferred(serv);
585 
586 	svc_pool_map_put(serv->sv_nrpools);
587 
588 	for (i = 0; i < serv->sv_nrpools; i++) {
589 		struct svc_pool *pool = &serv->sv_pools[i];
590 
591 		percpu_counter_destroy(&pool->sp_messages_arrived);
592 		percpu_counter_destroy(&pool->sp_sockets_queued);
593 		percpu_counter_destroy(&pool->sp_threads_woken);
594 	}
595 	kfree(serv->sv_pools);
596 	kfree(serv);
597 }
598 EXPORT_SYMBOL_GPL(svc_destroy);
599 
600 static bool
601 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
602 {
603 	unsigned long pages, ret;
604 
605 	/* bc_xprt uses fore channel allocated buffers */
606 	if (svc_is_backchannel(rqstp))
607 		return true;
608 
609 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
610 				       * We assume one is at most one page
611 				       */
612 	WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
613 	if (pages > RPCSVC_MAXPAGES)
614 		pages = RPCSVC_MAXPAGES;
615 
616 	ret = alloc_pages_bulk_array_node(GFP_KERNEL, node, pages,
617 					  rqstp->rq_pages);
618 	return ret == pages;
619 }
620 
621 /*
622  * Release an RPC server buffer
623  */
624 static void
625 svc_release_buffer(struct svc_rqst *rqstp)
626 {
627 	unsigned int i;
628 
629 	for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
630 		if (rqstp->rq_pages[i])
631 			put_page(rqstp->rq_pages[i]);
632 }
633 
634 struct svc_rqst *
635 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
636 {
637 	struct svc_rqst	*rqstp;
638 
639 	rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
640 	if (!rqstp)
641 		return rqstp;
642 
643 	folio_batch_init(&rqstp->rq_fbatch);
644 
645 	__set_bit(RQ_BUSY, &rqstp->rq_flags);
646 	rqstp->rq_server = serv;
647 	rqstp->rq_pool = pool;
648 
649 	rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
650 	if (!rqstp->rq_scratch_page)
651 		goto out_enomem;
652 
653 	rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
654 	if (!rqstp->rq_argp)
655 		goto out_enomem;
656 
657 	rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
658 	if (!rqstp->rq_resp)
659 		goto out_enomem;
660 
661 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
662 		goto out_enomem;
663 
664 	return rqstp;
665 out_enomem:
666 	svc_rqst_free(rqstp);
667 	return NULL;
668 }
669 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
670 
671 static struct svc_rqst *
672 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
673 {
674 	struct svc_rqst	*rqstp;
675 
676 	rqstp = svc_rqst_alloc(serv, pool, node);
677 	if (!rqstp)
678 		return ERR_PTR(-ENOMEM);
679 
680 	svc_get(serv);
681 	spin_lock_bh(&serv->sv_lock);
682 	serv->sv_nrthreads += 1;
683 	spin_unlock_bh(&serv->sv_lock);
684 
685 	spin_lock_bh(&pool->sp_lock);
686 	pool->sp_nrthreads++;
687 	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
688 	spin_unlock_bh(&pool->sp_lock);
689 	return rqstp;
690 }
691 
692 /**
693  * svc_pool_wake_idle_thread - Awaken an idle thread in @pool
694  * @pool: service thread pool
695  *
696  * Can be called from soft IRQ or process context. Finding an idle
697  * service thread and marking it BUSY is atomic with respect to
698  * other calls to svc_pool_wake_idle_thread().
699  *
700  */
701 void svc_pool_wake_idle_thread(struct svc_pool *pool)
702 {
703 	struct svc_rqst	*rqstp;
704 
705 	rcu_read_lock();
706 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
707 		if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
708 			continue;
709 
710 		WRITE_ONCE(rqstp->rq_qtime, ktime_get());
711 		wake_up_process(rqstp->rq_task);
712 		rcu_read_unlock();
713 		percpu_counter_inc(&pool->sp_threads_woken);
714 		trace_svc_wake_up(rqstp->rq_task->pid);
715 		return;
716 	}
717 	rcu_read_unlock();
718 
719 	set_bit(SP_CONGESTED, &pool->sp_flags);
720 }
721 
722 static struct svc_pool *
723 svc_pool_next(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
724 {
725 	return pool ? pool : &serv->sv_pools[(*state)++ % serv->sv_nrpools];
726 }
727 
728 static struct task_struct *
729 svc_pool_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
730 {
731 	unsigned int i;
732 	struct task_struct *task = NULL;
733 
734 	if (pool != NULL) {
735 		spin_lock_bh(&pool->sp_lock);
736 	} else {
737 		for (i = 0; i < serv->sv_nrpools; i++) {
738 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
739 			spin_lock_bh(&pool->sp_lock);
740 			if (!list_empty(&pool->sp_all_threads))
741 				goto found_pool;
742 			spin_unlock_bh(&pool->sp_lock);
743 		}
744 		return NULL;
745 	}
746 
747 found_pool:
748 	if (!list_empty(&pool->sp_all_threads)) {
749 		struct svc_rqst *rqstp;
750 
751 		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
752 		set_bit(RQ_VICTIM, &rqstp->rq_flags);
753 		list_del_rcu(&rqstp->rq_all);
754 		task = rqstp->rq_task;
755 	}
756 	spin_unlock_bh(&pool->sp_lock);
757 	return task;
758 }
759 
760 static int
761 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
762 {
763 	struct svc_rqst	*rqstp;
764 	struct task_struct *task;
765 	struct svc_pool *chosen_pool;
766 	unsigned int state = serv->sv_nrthreads-1;
767 	int node;
768 
769 	do {
770 		nrservs--;
771 		chosen_pool = svc_pool_next(serv, pool, &state);
772 		node = svc_pool_map_get_node(chosen_pool->sp_id);
773 
774 		rqstp = svc_prepare_thread(serv, chosen_pool, node);
775 		if (IS_ERR(rqstp))
776 			return PTR_ERR(rqstp);
777 		task = kthread_create_on_node(serv->sv_threadfn, rqstp,
778 					      node, "%s", serv->sv_name);
779 		if (IS_ERR(task)) {
780 			svc_exit_thread(rqstp);
781 			return PTR_ERR(task);
782 		}
783 
784 		rqstp->rq_task = task;
785 		if (serv->sv_nrpools > 1)
786 			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
787 
788 		svc_sock_update_bufs(serv);
789 		wake_up_process(task);
790 	} while (nrservs > 0);
791 
792 	return 0;
793 }
794 
795 static int
796 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
797 {
798 	struct svc_rqst	*rqstp;
799 	struct task_struct *task;
800 	unsigned int state = serv->sv_nrthreads-1;
801 
802 	do {
803 		task = svc_pool_victim(serv, pool, &state);
804 		if (task == NULL)
805 			break;
806 		rqstp = kthread_data(task);
807 		/* Did we lose a race to svo_function threadfn? */
808 		if (kthread_stop(task) == -EINTR)
809 			svc_exit_thread(rqstp);
810 		nrservs++;
811 	} while (nrservs < 0);
812 	return 0;
813 }
814 
815 /**
816  * svc_set_num_threads - adjust number of threads per RPC service
817  * @serv: RPC service to adjust
818  * @pool: Specific pool from which to choose threads, or NULL
819  * @nrservs: New number of threads for @serv (0 or less means kill all threads)
820  *
821  * Create or destroy threads to make the number of threads for @serv the
822  * given number. If @pool is non-NULL, change only threads in that pool;
823  * otherwise, round-robin between all pools for @serv. @serv's
824  * sv_nrthreads is adjusted for each thread created or destroyed.
825  *
826  * Caller must ensure mutual exclusion between this and server startup or
827  * shutdown.
828  *
829  * Returns zero on success or a negative errno if an error occurred while
830  * starting a thread.
831  */
832 int
833 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
834 {
835 	if (pool == NULL) {
836 		nrservs -= serv->sv_nrthreads;
837 	} else {
838 		spin_lock_bh(&pool->sp_lock);
839 		nrservs -= pool->sp_nrthreads;
840 		spin_unlock_bh(&pool->sp_lock);
841 	}
842 
843 	if (nrservs > 0)
844 		return svc_start_kthreads(serv, pool, nrservs);
845 	if (nrservs < 0)
846 		return svc_stop_kthreads(serv, pool, nrservs);
847 	return 0;
848 }
849 EXPORT_SYMBOL_GPL(svc_set_num_threads);
850 
851 /**
852  * svc_rqst_replace_page - Replace one page in rq_pages[]
853  * @rqstp: svc_rqst with pages to replace
854  * @page: replacement page
855  *
856  * When replacing a page in rq_pages, batch the release of the
857  * replaced pages to avoid hammering the page allocator.
858  *
859  * Return values:
860  *   %true: page replaced
861  *   %false: array bounds checking failed
862  */
863 bool svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
864 {
865 	struct page **begin = rqstp->rq_pages;
866 	struct page **end = &rqstp->rq_pages[RPCSVC_MAXPAGES];
867 
868 	if (unlikely(rqstp->rq_next_page < begin || rqstp->rq_next_page > end)) {
869 		trace_svc_replace_page_err(rqstp);
870 		return false;
871 	}
872 
873 	if (*rqstp->rq_next_page) {
874 		if (!folio_batch_add(&rqstp->rq_fbatch,
875 				page_folio(*rqstp->rq_next_page)))
876 			__folio_batch_release(&rqstp->rq_fbatch);
877 	}
878 
879 	get_page(page);
880 	*(rqstp->rq_next_page++) = page;
881 	return true;
882 }
883 EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
884 
885 /**
886  * svc_rqst_release_pages - Release Reply buffer pages
887  * @rqstp: RPC transaction context
888  *
889  * Release response pages that might still be in flight after
890  * svc_send, and any spliced filesystem-owned pages.
891  */
892 void svc_rqst_release_pages(struct svc_rqst *rqstp)
893 {
894 	int i, count = rqstp->rq_next_page - rqstp->rq_respages;
895 
896 	if (count) {
897 		release_pages(rqstp->rq_respages, count);
898 		for (i = 0; i < count; i++)
899 			rqstp->rq_respages[i] = NULL;
900 	}
901 }
902 
903 /*
904  * Called from a server thread as it's exiting. Caller must hold the "service
905  * mutex" for the service.
906  */
907 void
908 svc_rqst_free(struct svc_rqst *rqstp)
909 {
910 	folio_batch_release(&rqstp->rq_fbatch);
911 	svc_release_buffer(rqstp);
912 	if (rqstp->rq_scratch_page)
913 		put_page(rqstp->rq_scratch_page);
914 	kfree(rqstp->rq_resp);
915 	kfree(rqstp->rq_argp);
916 	kfree(rqstp->rq_auth_data);
917 	kfree_rcu(rqstp, rq_rcu_head);
918 }
919 EXPORT_SYMBOL_GPL(svc_rqst_free);
920 
921 void
922 svc_exit_thread(struct svc_rqst *rqstp)
923 {
924 	struct svc_serv	*serv = rqstp->rq_server;
925 	struct svc_pool	*pool = rqstp->rq_pool;
926 
927 	spin_lock_bh(&pool->sp_lock);
928 	pool->sp_nrthreads--;
929 	if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
930 		list_del_rcu(&rqstp->rq_all);
931 	spin_unlock_bh(&pool->sp_lock);
932 
933 	spin_lock_bh(&serv->sv_lock);
934 	serv->sv_nrthreads -= 1;
935 	spin_unlock_bh(&serv->sv_lock);
936 	svc_sock_update_bufs(serv);
937 
938 	svc_rqst_free(rqstp);
939 
940 	svc_put(serv);
941 }
942 EXPORT_SYMBOL_GPL(svc_exit_thread);
943 
944 /*
945  * Register an "inet" protocol family netid with the local
946  * rpcbind daemon via an rpcbind v4 SET request.
947  *
948  * No netconfig infrastructure is available in the kernel, so
949  * we map IP_ protocol numbers to netids by hand.
950  *
951  * Returns zero on success; a negative errno value is returned
952  * if any error occurs.
953  */
954 static int __svc_rpcb_register4(struct net *net, const u32 program,
955 				const u32 version,
956 				const unsigned short protocol,
957 				const unsigned short port)
958 {
959 	const struct sockaddr_in sin = {
960 		.sin_family		= AF_INET,
961 		.sin_addr.s_addr	= htonl(INADDR_ANY),
962 		.sin_port		= htons(port),
963 	};
964 	const char *netid;
965 	int error;
966 
967 	switch (protocol) {
968 	case IPPROTO_UDP:
969 		netid = RPCBIND_NETID_UDP;
970 		break;
971 	case IPPROTO_TCP:
972 		netid = RPCBIND_NETID_TCP;
973 		break;
974 	default:
975 		return -ENOPROTOOPT;
976 	}
977 
978 	error = rpcb_v4_register(net, program, version,
979 					(const struct sockaddr *)&sin, netid);
980 
981 	/*
982 	 * User space didn't support rpcbind v4, so retry this
983 	 * registration request with the legacy rpcbind v2 protocol.
984 	 */
985 	if (error == -EPROTONOSUPPORT)
986 		error = rpcb_register(net, program, version, protocol, port);
987 
988 	return error;
989 }
990 
991 #if IS_ENABLED(CONFIG_IPV6)
992 /*
993  * Register an "inet6" protocol family netid with the local
994  * rpcbind daemon via an rpcbind v4 SET request.
995  *
996  * No netconfig infrastructure is available in the kernel, so
997  * we map IP_ protocol numbers to netids by hand.
998  *
999  * Returns zero on success; a negative errno value is returned
1000  * if any error occurs.
1001  */
1002 static int __svc_rpcb_register6(struct net *net, const u32 program,
1003 				const u32 version,
1004 				const unsigned short protocol,
1005 				const unsigned short port)
1006 {
1007 	const struct sockaddr_in6 sin6 = {
1008 		.sin6_family		= AF_INET6,
1009 		.sin6_addr		= IN6ADDR_ANY_INIT,
1010 		.sin6_port		= htons(port),
1011 	};
1012 	const char *netid;
1013 	int error;
1014 
1015 	switch (protocol) {
1016 	case IPPROTO_UDP:
1017 		netid = RPCBIND_NETID_UDP6;
1018 		break;
1019 	case IPPROTO_TCP:
1020 		netid = RPCBIND_NETID_TCP6;
1021 		break;
1022 	default:
1023 		return -ENOPROTOOPT;
1024 	}
1025 
1026 	error = rpcb_v4_register(net, program, version,
1027 					(const struct sockaddr *)&sin6, netid);
1028 
1029 	/*
1030 	 * User space didn't support rpcbind version 4, so we won't
1031 	 * use a PF_INET6 listener.
1032 	 */
1033 	if (error == -EPROTONOSUPPORT)
1034 		error = -EAFNOSUPPORT;
1035 
1036 	return error;
1037 }
1038 #endif	/* IS_ENABLED(CONFIG_IPV6) */
1039 
1040 /*
1041  * Register a kernel RPC service via rpcbind version 4.
1042  *
1043  * Returns zero on success; a negative errno value is returned
1044  * if any error occurs.
1045  */
1046 static int __svc_register(struct net *net, const char *progname,
1047 			  const u32 program, const u32 version,
1048 			  const int family,
1049 			  const unsigned short protocol,
1050 			  const unsigned short port)
1051 {
1052 	int error = -EAFNOSUPPORT;
1053 
1054 	switch (family) {
1055 	case PF_INET:
1056 		error = __svc_rpcb_register4(net, program, version,
1057 						protocol, port);
1058 		break;
1059 #if IS_ENABLED(CONFIG_IPV6)
1060 	case PF_INET6:
1061 		error = __svc_rpcb_register6(net, program, version,
1062 						protocol, port);
1063 #endif
1064 	}
1065 
1066 	trace_svc_register(progname, version, family, protocol, port, error);
1067 	return error;
1068 }
1069 
1070 int svc_rpcbind_set_version(struct net *net,
1071 			    const struct svc_program *progp,
1072 			    u32 version, int family,
1073 			    unsigned short proto,
1074 			    unsigned short port)
1075 {
1076 	return __svc_register(net, progp->pg_name, progp->pg_prog,
1077 				version, family, proto, port);
1078 
1079 }
1080 EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1081 
1082 int svc_generic_rpcbind_set(struct net *net,
1083 			    const struct svc_program *progp,
1084 			    u32 version, int family,
1085 			    unsigned short proto,
1086 			    unsigned short port)
1087 {
1088 	const struct svc_version *vers = progp->pg_vers[version];
1089 	int error;
1090 
1091 	if (vers == NULL)
1092 		return 0;
1093 
1094 	if (vers->vs_hidden) {
1095 		trace_svc_noregister(progp->pg_name, version, proto,
1096 				     port, family, 0);
1097 		return 0;
1098 	}
1099 
1100 	/*
1101 	 * Don't register a UDP port if we need congestion
1102 	 * control.
1103 	 */
1104 	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1105 		return 0;
1106 
1107 	error = svc_rpcbind_set_version(net, progp, version,
1108 					family, proto, port);
1109 
1110 	return (vers->vs_rpcb_optnl) ? 0 : error;
1111 }
1112 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1113 
1114 /**
1115  * svc_register - register an RPC service with the local portmapper
1116  * @serv: svc_serv struct for the service to register
1117  * @net: net namespace for the service to register
1118  * @family: protocol family of service's listener socket
1119  * @proto: transport protocol number to advertise
1120  * @port: port to advertise
1121  *
1122  * Service is registered for any address in the passed-in protocol family
1123  */
1124 int svc_register(const struct svc_serv *serv, struct net *net,
1125 		 const int family, const unsigned short proto,
1126 		 const unsigned short port)
1127 {
1128 	struct svc_program	*progp;
1129 	unsigned int		i;
1130 	int			error = 0;
1131 
1132 	WARN_ON_ONCE(proto == 0 && port == 0);
1133 	if (proto == 0 && port == 0)
1134 		return -EINVAL;
1135 
1136 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1137 		for (i = 0; i < progp->pg_nvers; i++) {
1138 
1139 			error = progp->pg_rpcbind_set(net, progp, i,
1140 					family, proto, port);
1141 			if (error < 0) {
1142 				printk(KERN_WARNING "svc: failed to register "
1143 					"%sv%u RPC service (errno %d).\n",
1144 					progp->pg_name, i, -error);
1145 				break;
1146 			}
1147 		}
1148 	}
1149 
1150 	return error;
1151 }
1152 
1153 /*
1154  * If user space is running rpcbind, it should take the v4 UNSET
1155  * and clear everything for this [program, version].  If user space
1156  * is running portmap, it will reject the v4 UNSET, but won't have
1157  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1158  * in this case to clear all existing entries for [program, version].
1159  */
1160 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1161 			     const char *progname)
1162 {
1163 	int error;
1164 
1165 	error = rpcb_v4_register(net, program, version, NULL, "");
1166 
1167 	/*
1168 	 * User space didn't support rpcbind v4, so retry this
1169 	 * request with the legacy rpcbind v2 protocol.
1170 	 */
1171 	if (error == -EPROTONOSUPPORT)
1172 		error = rpcb_register(net, program, version, 0, 0);
1173 
1174 	trace_svc_unregister(progname, version, error);
1175 }
1176 
1177 /*
1178  * All netids, bind addresses and ports registered for [program, version]
1179  * are removed from the local rpcbind database (if the service is not
1180  * hidden) to make way for a new instance of the service.
1181  *
1182  * The result of unregistration is reported via dprintk for those who want
1183  * verification of the result, but is otherwise not important.
1184  */
1185 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1186 {
1187 	struct sighand_struct *sighand;
1188 	struct svc_program *progp;
1189 	unsigned long flags;
1190 	unsigned int i;
1191 
1192 	clear_thread_flag(TIF_SIGPENDING);
1193 
1194 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1195 		for (i = 0; i < progp->pg_nvers; i++) {
1196 			if (progp->pg_vers[i] == NULL)
1197 				continue;
1198 			if (progp->pg_vers[i]->vs_hidden)
1199 				continue;
1200 			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1201 		}
1202 	}
1203 
1204 	rcu_read_lock();
1205 	sighand = rcu_dereference(current->sighand);
1206 	spin_lock_irqsave(&sighand->siglock, flags);
1207 	recalc_sigpending();
1208 	spin_unlock_irqrestore(&sighand->siglock, flags);
1209 	rcu_read_unlock();
1210 }
1211 
1212 /*
1213  * dprintk the given error with the address of the client that caused it.
1214  */
1215 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1216 static __printf(2, 3)
1217 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1218 {
1219 	struct va_format vaf;
1220 	va_list args;
1221 	char 	buf[RPC_MAX_ADDRBUFLEN];
1222 
1223 	va_start(args, fmt);
1224 
1225 	vaf.fmt = fmt;
1226 	vaf.va = &args;
1227 
1228 	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1229 
1230 	va_end(args);
1231 }
1232 #else
1233 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1234 #endif
1235 
1236 __be32
1237 svc_generic_init_request(struct svc_rqst *rqstp,
1238 		const struct svc_program *progp,
1239 		struct svc_process_info *ret)
1240 {
1241 	const struct svc_version *versp = NULL;	/* compiler food */
1242 	const struct svc_procedure *procp = NULL;
1243 
1244 	if (rqstp->rq_vers >= progp->pg_nvers )
1245 		goto err_bad_vers;
1246 	versp = progp->pg_vers[rqstp->rq_vers];
1247 	if (!versp)
1248 		goto err_bad_vers;
1249 
1250 	/*
1251 	 * Some protocol versions (namely NFSv4) require some form of
1252 	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1253 	 * In other words, UDP is not allowed. We mark those when setting
1254 	 * up the svc_xprt, and verify that here.
1255 	 *
1256 	 * The spec is not very clear about what error should be returned
1257 	 * when someone tries to access a server that is listening on UDP
1258 	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1259 	 * fit.
1260 	 */
1261 	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1262 	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1263 		goto err_bad_vers;
1264 
1265 	if (rqstp->rq_proc >= versp->vs_nproc)
1266 		goto err_bad_proc;
1267 	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1268 	if (!procp)
1269 		goto err_bad_proc;
1270 
1271 	/* Initialize storage for argp and resp */
1272 	memset(rqstp->rq_argp, 0, procp->pc_argzero);
1273 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1274 
1275 	/* Bump per-procedure stats counter */
1276 	this_cpu_inc(versp->vs_count[rqstp->rq_proc]);
1277 
1278 	ret->dispatch = versp->vs_dispatch;
1279 	return rpc_success;
1280 err_bad_vers:
1281 	ret->mismatch.lovers = progp->pg_lovers;
1282 	ret->mismatch.hivers = progp->pg_hivers;
1283 	return rpc_prog_mismatch;
1284 err_bad_proc:
1285 	return rpc_proc_unavail;
1286 }
1287 EXPORT_SYMBOL_GPL(svc_generic_init_request);
1288 
1289 /*
1290  * Common routine for processing the RPC request.
1291  */
1292 static int
1293 svc_process_common(struct svc_rqst *rqstp)
1294 {
1295 	struct xdr_stream	*xdr = &rqstp->rq_res_stream;
1296 	struct svc_program	*progp;
1297 	const struct svc_procedure *procp = NULL;
1298 	struct svc_serv		*serv = rqstp->rq_server;
1299 	struct svc_process_info process;
1300 	enum svc_auth_status	auth_res;
1301 	unsigned int		aoffset;
1302 	int			rc;
1303 	__be32			*p;
1304 
1305 	/* Will be turned off by GSS integrity and privacy services */
1306 	set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1307 	/* Will be turned off only when NFSv4 Sessions are used */
1308 	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1309 	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1310 
1311 	/* Construct the first words of the reply: */
1312 	svcxdr_init_encode(rqstp);
1313 	xdr_stream_encode_be32(xdr, rqstp->rq_xid);
1314 	xdr_stream_encode_be32(xdr, rpc_reply);
1315 
1316 	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 4);
1317 	if (unlikely(!p))
1318 		goto err_short_len;
1319 	if (*p++ != cpu_to_be32(RPC_VERSION))
1320 		goto err_bad_rpc;
1321 
1322 	xdr_stream_encode_be32(xdr, rpc_msg_accepted);
1323 
1324 	rqstp->rq_prog = be32_to_cpup(p++);
1325 	rqstp->rq_vers = be32_to_cpup(p++);
1326 	rqstp->rq_proc = be32_to_cpup(p);
1327 
1328 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1329 		if (rqstp->rq_prog == progp->pg_prog)
1330 			break;
1331 
1332 	/*
1333 	 * Decode auth data, and add verifier to reply buffer.
1334 	 * We do this before anything else in order to get a decent
1335 	 * auth verifier.
1336 	 */
1337 	auth_res = svc_authenticate(rqstp);
1338 	/* Also give the program a chance to reject this call: */
1339 	if (auth_res == SVC_OK && progp)
1340 		auth_res = progp->pg_authenticate(rqstp);
1341 	trace_svc_authenticate(rqstp, auth_res);
1342 	switch (auth_res) {
1343 	case SVC_OK:
1344 		break;
1345 	case SVC_GARBAGE:
1346 		goto err_garbage_args;
1347 	case SVC_SYSERR:
1348 		goto err_system_err;
1349 	case SVC_DENIED:
1350 		goto err_bad_auth;
1351 	case SVC_CLOSE:
1352 		goto close;
1353 	case SVC_DROP:
1354 		goto dropit;
1355 	case SVC_COMPLETE:
1356 		goto sendit;
1357 	default:
1358 		pr_warn_once("Unexpected svc_auth_status (%d)\n", auth_res);
1359 		goto err_system_err;
1360 	}
1361 
1362 	if (progp == NULL)
1363 		goto err_bad_prog;
1364 
1365 	switch (progp->pg_init_request(rqstp, progp, &process)) {
1366 	case rpc_success:
1367 		break;
1368 	case rpc_prog_unavail:
1369 		goto err_bad_prog;
1370 	case rpc_prog_mismatch:
1371 		goto err_bad_vers;
1372 	case rpc_proc_unavail:
1373 		goto err_bad_proc;
1374 	}
1375 
1376 	procp = rqstp->rq_procinfo;
1377 	/* Should this check go into the dispatcher? */
1378 	if (!procp || !procp->pc_func)
1379 		goto err_bad_proc;
1380 
1381 	/* Syntactic check complete */
1382 	serv->sv_stats->rpccnt++;
1383 	trace_svc_process(rqstp, progp->pg_name);
1384 
1385 	aoffset = xdr_stream_pos(xdr);
1386 
1387 	/* un-reserve some of the out-queue now that we have a
1388 	 * better idea of reply size
1389 	 */
1390 	if (procp->pc_xdrressize)
1391 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1392 
1393 	/* Call the function that processes the request. */
1394 	rc = process.dispatch(rqstp);
1395 	if (procp->pc_release)
1396 		procp->pc_release(rqstp);
1397 	xdr_finish_decode(xdr);
1398 
1399 	if (!rc)
1400 		goto dropit;
1401 	if (rqstp->rq_auth_stat != rpc_auth_ok)
1402 		goto err_bad_auth;
1403 
1404 	if (*rqstp->rq_accept_statp != rpc_success)
1405 		xdr_truncate_encode(xdr, aoffset);
1406 
1407 	if (procp->pc_encode == NULL)
1408 		goto dropit;
1409 
1410  sendit:
1411 	if (svc_authorise(rqstp))
1412 		goto close_xprt;
1413 	return 1;		/* Caller can now send it */
1414 
1415  dropit:
1416 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1417 	dprintk("svc: svc_process dropit\n");
1418 	return 0;
1419 
1420  close:
1421 	svc_authorise(rqstp);
1422 close_xprt:
1423 	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1424 		svc_xprt_close(rqstp->rq_xprt);
1425 	dprintk("svc: svc_process close\n");
1426 	return 0;
1427 
1428 err_short_len:
1429 	svc_printk(rqstp, "short len %u, dropping request\n",
1430 		   rqstp->rq_arg.len);
1431 	goto close_xprt;
1432 
1433 err_bad_rpc:
1434 	serv->sv_stats->rpcbadfmt++;
1435 	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1436 	xdr_stream_encode_u32(xdr, RPC_MISMATCH);
1437 	/* Only RPCv2 supported */
1438 	xdr_stream_encode_u32(xdr, RPC_VERSION);
1439 	xdr_stream_encode_u32(xdr, RPC_VERSION);
1440 	return 1;	/* don't wrap */
1441 
1442 err_bad_auth:
1443 	dprintk("svc: authentication failed (%d)\n",
1444 		be32_to_cpu(rqstp->rq_auth_stat));
1445 	serv->sv_stats->rpcbadauth++;
1446 	/* Restore write pointer to location of reply status: */
1447 	xdr_truncate_encode(xdr, XDR_UNIT * 2);
1448 	xdr_stream_encode_u32(xdr, RPC_MSG_DENIED);
1449 	xdr_stream_encode_u32(xdr, RPC_AUTH_ERROR);
1450 	xdr_stream_encode_be32(xdr, rqstp->rq_auth_stat);
1451 	goto sendit;
1452 
1453 err_bad_prog:
1454 	dprintk("svc: unknown program %d\n", rqstp->rq_prog);
1455 	serv->sv_stats->rpcbadfmt++;
1456 	*rqstp->rq_accept_statp = rpc_prog_unavail;
1457 	goto sendit;
1458 
1459 err_bad_vers:
1460 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1461 		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1462 
1463 	serv->sv_stats->rpcbadfmt++;
1464 	*rqstp->rq_accept_statp = rpc_prog_mismatch;
1465 
1466 	/*
1467 	 * svc_authenticate() has already added the verifier and
1468 	 * advanced the stream just past rq_accept_statp.
1469 	 */
1470 	xdr_stream_encode_u32(xdr, process.mismatch.lovers);
1471 	xdr_stream_encode_u32(xdr, process.mismatch.hivers);
1472 	goto sendit;
1473 
1474 err_bad_proc:
1475 	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1476 
1477 	serv->sv_stats->rpcbadfmt++;
1478 	*rqstp->rq_accept_statp = rpc_proc_unavail;
1479 	goto sendit;
1480 
1481 err_garbage_args:
1482 	svc_printk(rqstp, "failed to decode RPC header\n");
1483 
1484 	serv->sv_stats->rpcbadfmt++;
1485 	*rqstp->rq_accept_statp = rpc_garbage_args;
1486 	goto sendit;
1487 
1488 err_system_err:
1489 	serv->sv_stats->rpcbadfmt++;
1490 	*rqstp->rq_accept_statp = rpc_system_err;
1491 	goto sendit;
1492 }
1493 
1494 /**
1495  * svc_process - Execute one RPC transaction
1496  * @rqstp: RPC transaction context
1497  *
1498  */
1499 void svc_process(struct svc_rqst *rqstp)
1500 {
1501 	struct kvec		*resv = &rqstp->rq_res.head[0];
1502 	__be32 *p;
1503 
1504 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1505 	if (!fail_sunrpc.ignore_server_disconnect &&
1506 	    should_fail(&fail_sunrpc.attr, 1))
1507 		svc_xprt_deferred_close(rqstp->rq_xprt);
1508 #endif
1509 
1510 	/*
1511 	 * Setup response xdr_buf.
1512 	 * Initially it has just one page
1513 	 */
1514 	rqstp->rq_next_page = &rqstp->rq_respages[1];
1515 	resv->iov_base = page_address(rqstp->rq_respages[0]);
1516 	resv->iov_len = 0;
1517 	rqstp->rq_res.pages = rqstp->rq_next_page;
1518 	rqstp->rq_res.len = 0;
1519 	rqstp->rq_res.page_base = 0;
1520 	rqstp->rq_res.page_len = 0;
1521 	rqstp->rq_res.buflen = PAGE_SIZE;
1522 	rqstp->rq_res.tail[0].iov_base = NULL;
1523 	rqstp->rq_res.tail[0].iov_len = 0;
1524 
1525 	svcxdr_init_decode(rqstp);
1526 	p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2);
1527 	if (unlikely(!p))
1528 		goto out_drop;
1529 	rqstp->rq_xid = *p++;
1530 	if (unlikely(*p != rpc_call))
1531 		goto out_baddir;
1532 
1533 	if (!svc_process_common(rqstp))
1534 		goto out_drop;
1535 	svc_send(rqstp);
1536 	return;
1537 
1538 out_baddir:
1539 	svc_printk(rqstp, "bad direction 0x%08x, dropping request\n",
1540 		   be32_to_cpu(*p));
1541 	rqstp->rq_server->sv_stats->rpcbadfmt++;
1542 out_drop:
1543 	svc_drop(rqstp);
1544 }
1545 
1546 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1547 /*
1548  * Process a backchannel RPC request that arrived over an existing
1549  * outbound connection
1550  */
1551 int
1552 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1553 	       struct svc_rqst *rqstp)
1554 {
1555 	struct rpc_task *task;
1556 	int proc_error;
1557 	int error;
1558 
1559 	dprintk("svc: %s(%p)\n", __func__, req);
1560 
1561 	/* Build the svc_rqst used by the common processing routine */
1562 	rqstp->rq_xid = req->rq_xid;
1563 	rqstp->rq_prot = req->rq_xprt->prot;
1564 	rqstp->rq_server = serv;
1565 	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1566 
1567 	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1568 	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1569 	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1570 	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1571 
1572 	/* Adjust the argument buffer length */
1573 	rqstp->rq_arg.len = req->rq_private_buf.len;
1574 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1575 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1576 		rqstp->rq_arg.page_len = 0;
1577 	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1578 			rqstp->rq_arg.page_len)
1579 		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1580 			rqstp->rq_arg.head[0].iov_len;
1581 	else
1582 		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1583 			rqstp->rq_arg.page_len;
1584 
1585 	/* Reset the response buffer */
1586 	rqstp->rq_res.head[0].iov_len = 0;
1587 
1588 	/*
1589 	 * Skip the XID and calldir fields because they've already
1590 	 * been processed by the caller.
1591 	 */
1592 	svcxdr_init_decode(rqstp);
1593 	if (!xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2)) {
1594 		error = -EINVAL;
1595 		goto out;
1596 	}
1597 
1598 	/* Parse and execute the bc call */
1599 	proc_error = svc_process_common(rqstp);
1600 
1601 	atomic_dec(&req->rq_xprt->bc_slot_count);
1602 	if (!proc_error) {
1603 		/* Processing error: drop the request */
1604 		xprt_free_bc_request(req);
1605 		error = -EINVAL;
1606 		goto out;
1607 	}
1608 	/* Finally, send the reply synchronously */
1609 	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1610 	task = rpc_run_bc_task(req);
1611 	if (IS_ERR(task)) {
1612 		error = PTR_ERR(task);
1613 		goto out;
1614 	}
1615 
1616 	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1617 	error = task->tk_status;
1618 	rpc_put_task(task);
1619 
1620 out:
1621 	dprintk("svc: %s(), error=%d\n", __func__, error);
1622 	return error;
1623 }
1624 EXPORT_SYMBOL_GPL(bc_svc_process);
1625 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1626 
1627 /**
1628  * svc_max_payload - Return transport-specific limit on the RPC payload
1629  * @rqstp: RPC transaction context
1630  *
1631  * Returns the maximum number of payload bytes the current transport
1632  * allows.
1633  */
1634 u32 svc_max_payload(const struct svc_rqst *rqstp)
1635 {
1636 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1637 
1638 	if (rqstp->rq_server->sv_max_payload < max)
1639 		max = rqstp->rq_server->sv_max_payload;
1640 	return max;
1641 }
1642 EXPORT_SYMBOL_GPL(svc_max_payload);
1643 
1644 /**
1645  * svc_proc_name - Return RPC procedure name in string form
1646  * @rqstp: svc_rqst to operate on
1647  *
1648  * Return value:
1649  *   Pointer to a NUL-terminated string
1650  */
1651 const char *svc_proc_name(const struct svc_rqst *rqstp)
1652 {
1653 	if (rqstp && rqstp->rq_procinfo)
1654 		return rqstp->rq_procinfo->pc_name;
1655 	return "unknown";
1656 }
1657 
1658 
1659 /**
1660  * svc_encode_result_payload - mark a range of bytes as a result payload
1661  * @rqstp: svc_rqst to operate on
1662  * @offset: payload's byte offset in rqstp->rq_res
1663  * @length: size of payload, in bytes
1664  *
1665  * Returns zero on success, or a negative errno if a permanent
1666  * error occurred.
1667  */
1668 int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1669 			      unsigned int length)
1670 {
1671 	return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1672 							   length);
1673 }
1674 EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1675 
1676 /**
1677  * svc_fill_write_vector - Construct data argument for VFS write call
1678  * @rqstp: svc_rqst to operate on
1679  * @payload: xdr_buf containing only the write data payload
1680  *
1681  * Fills in rqstp::rq_vec, and returns the number of elements.
1682  */
1683 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1684 				   struct xdr_buf *payload)
1685 {
1686 	struct page **pages = payload->pages;
1687 	struct kvec *first = payload->head;
1688 	struct kvec *vec = rqstp->rq_vec;
1689 	size_t total = payload->len;
1690 	unsigned int i;
1691 
1692 	/* Some types of transport can present the write payload
1693 	 * entirely in rq_arg.pages. In this case, @first is empty.
1694 	 */
1695 	i = 0;
1696 	if (first->iov_len) {
1697 		vec[i].iov_base = first->iov_base;
1698 		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1699 		total -= vec[i].iov_len;
1700 		++i;
1701 	}
1702 
1703 	while (total) {
1704 		vec[i].iov_base = page_address(*pages);
1705 		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1706 		total -= vec[i].iov_len;
1707 		++i;
1708 		++pages;
1709 	}
1710 
1711 	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1712 	return i;
1713 }
1714 EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1715 
1716 /**
1717  * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1718  * @rqstp: svc_rqst to operate on
1719  * @first: buffer containing first section of pathname
1720  * @p: buffer containing remaining section of pathname
1721  * @total: total length of the pathname argument
1722  *
1723  * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1724  * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1725  * the returned string.
1726  */
1727 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1728 				void *p, size_t total)
1729 {
1730 	size_t len, remaining;
1731 	char *result, *dst;
1732 
1733 	result = kmalloc(total + 1, GFP_KERNEL);
1734 	if (!result)
1735 		return ERR_PTR(-ESERVERFAULT);
1736 
1737 	dst = result;
1738 	remaining = total;
1739 
1740 	len = min_t(size_t, total, first->iov_len);
1741 	if (len) {
1742 		memcpy(dst, first->iov_base, len);
1743 		dst += len;
1744 		remaining -= len;
1745 	}
1746 
1747 	if (remaining) {
1748 		len = min_t(size_t, remaining, PAGE_SIZE);
1749 		memcpy(dst, p, len);
1750 		dst += len;
1751 	}
1752 
1753 	*dst = '\0';
1754 
1755 	/* Sanity check: Linux doesn't allow the pathname argument to
1756 	 * contain a NUL byte.
1757 	 */
1758 	if (strlen(result) != total) {
1759 		kfree(result);
1760 		return ERR_PTR(-EINVAL);
1761 	}
1762 	return result;
1763 }
1764 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1765