xref: /openbmc/linux/net/sunrpc/svc.c (revision 151f4e2b)
1 /*
2  * linux/net/sunrpc/svc.c
3  *
4  * High-level RPC service routines
5  *
6  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7  *
8  * Multiple threads pools and NUMAisation
9  * Copyright (c) 2006 Silicon Graphics, Inc.
10  * by Greg Banks <gnb@melbourne.sgi.com>
11  */
12 
13 #include <linux/linkage.h>
14 #include <linux/sched/signal.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
17 #include <linux/in.h>
18 #include <linux/mm.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
23 
24 #include <linux/sunrpc/types.h>
25 #include <linux/sunrpc/xdr.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svcsock.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/bc_xprt.h>
30 
31 #include <trace/events/sunrpc.h>
32 
33 #define RPCDBG_FACILITY	RPCDBG_SVCDSP
34 
35 static void svc_unregister(const struct svc_serv *serv, struct net *net);
36 
37 #define svc_serv_is_pooled(serv)    ((serv)->sv_ops->svo_function)
38 
39 #define SVC_POOL_DEFAULT	SVC_POOL_GLOBAL
40 
41 /*
42  * Structure for mapping cpus to pools and vice versa.
43  * Setup once during sunrpc initialisation.
44  */
45 struct svc_pool_map svc_pool_map = {
46 	.mode = SVC_POOL_DEFAULT
47 };
48 EXPORT_SYMBOL_GPL(svc_pool_map);
49 
50 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
51 
52 static int
53 param_set_pool_mode(const char *val, const struct kernel_param *kp)
54 {
55 	int *ip = (int *)kp->arg;
56 	struct svc_pool_map *m = &svc_pool_map;
57 	int err;
58 
59 	mutex_lock(&svc_pool_map_mutex);
60 
61 	err = -EBUSY;
62 	if (m->count)
63 		goto out;
64 
65 	err = 0;
66 	if (!strncmp(val, "auto", 4))
67 		*ip = SVC_POOL_AUTO;
68 	else if (!strncmp(val, "global", 6))
69 		*ip = SVC_POOL_GLOBAL;
70 	else if (!strncmp(val, "percpu", 6))
71 		*ip = SVC_POOL_PERCPU;
72 	else if (!strncmp(val, "pernode", 7))
73 		*ip = SVC_POOL_PERNODE;
74 	else
75 		err = -EINVAL;
76 
77 out:
78 	mutex_unlock(&svc_pool_map_mutex);
79 	return err;
80 }
81 
82 static int
83 param_get_pool_mode(char *buf, const struct kernel_param *kp)
84 {
85 	int *ip = (int *)kp->arg;
86 
87 	switch (*ip)
88 	{
89 	case SVC_POOL_AUTO:
90 		return strlcpy(buf, "auto", 20);
91 	case SVC_POOL_GLOBAL:
92 		return strlcpy(buf, "global", 20);
93 	case SVC_POOL_PERCPU:
94 		return strlcpy(buf, "percpu", 20);
95 	case SVC_POOL_PERNODE:
96 		return strlcpy(buf, "pernode", 20);
97 	default:
98 		return sprintf(buf, "%d", *ip);
99 	}
100 }
101 
102 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
103 		 &svc_pool_map.mode, 0644);
104 
105 /*
106  * Detect best pool mapping mode heuristically,
107  * according to the machine's topology.
108  */
109 static int
110 svc_pool_map_choose_mode(void)
111 {
112 	unsigned int node;
113 
114 	if (nr_online_nodes > 1) {
115 		/*
116 		 * Actually have multiple NUMA nodes,
117 		 * so split pools on NUMA node boundaries
118 		 */
119 		return SVC_POOL_PERNODE;
120 	}
121 
122 	node = first_online_node;
123 	if (nr_cpus_node(node) > 2) {
124 		/*
125 		 * Non-trivial SMP, or CONFIG_NUMA on
126 		 * non-NUMA hardware, e.g. with a generic
127 		 * x86_64 kernel on Xeons.  In this case we
128 		 * want to divide the pools on cpu boundaries.
129 		 */
130 		return SVC_POOL_PERCPU;
131 	}
132 
133 	/* default: one global pool */
134 	return SVC_POOL_GLOBAL;
135 }
136 
137 /*
138  * Allocate the to_pool[] and pool_to[] arrays.
139  * Returns 0 on success or an errno.
140  */
141 static int
142 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
143 {
144 	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
145 	if (!m->to_pool)
146 		goto fail;
147 	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
148 	if (!m->pool_to)
149 		goto fail_free;
150 
151 	return 0;
152 
153 fail_free:
154 	kfree(m->to_pool);
155 	m->to_pool = NULL;
156 fail:
157 	return -ENOMEM;
158 }
159 
160 /*
161  * Initialise the pool map for SVC_POOL_PERCPU mode.
162  * Returns number of pools or <0 on error.
163  */
164 static int
165 svc_pool_map_init_percpu(struct svc_pool_map *m)
166 {
167 	unsigned int maxpools = nr_cpu_ids;
168 	unsigned int pidx = 0;
169 	unsigned int cpu;
170 	int err;
171 
172 	err = svc_pool_map_alloc_arrays(m, maxpools);
173 	if (err)
174 		return err;
175 
176 	for_each_online_cpu(cpu) {
177 		BUG_ON(pidx >= maxpools);
178 		m->to_pool[cpu] = pidx;
179 		m->pool_to[pidx] = cpu;
180 		pidx++;
181 	}
182 	/* cpus brought online later all get mapped to pool0, sorry */
183 
184 	return pidx;
185 };
186 
187 
188 /*
189  * Initialise the pool map for SVC_POOL_PERNODE mode.
190  * Returns number of pools or <0 on error.
191  */
192 static int
193 svc_pool_map_init_pernode(struct svc_pool_map *m)
194 {
195 	unsigned int maxpools = nr_node_ids;
196 	unsigned int pidx = 0;
197 	unsigned int node;
198 	int err;
199 
200 	err = svc_pool_map_alloc_arrays(m, maxpools);
201 	if (err)
202 		return err;
203 
204 	for_each_node_with_cpus(node) {
205 		/* some architectures (e.g. SN2) have cpuless nodes */
206 		BUG_ON(pidx > maxpools);
207 		m->to_pool[node] = pidx;
208 		m->pool_to[pidx] = node;
209 		pidx++;
210 	}
211 	/* nodes brought online later all get mapped to pool0, sorry */
212 
213 	return pidx;
214 }
215 
216 
217 /*
218  * Add a reference to the global map of cpus to pools (and
219  * vice versa).  Initialise the map if we're the first user.
220  * Returns the number of pools.
221  */
222 unsigned int
223 svc_pool_map_get(void)
224 {
225 	struct svc_pool_map *m = &svc_pool_map;
226 	int npools = -1;
227 
228 	mutex_lock(&svc_pool_map_mutex);
229 
230 	if (m->count++) {
231 		mutex_unlock(&svc_pool_map_mutex);
232 		return m->npools;
233 	}
234 
235 	if (m->mode == SVC_POOL_AUTO)
236 		m->mode = svc_pool_map_choose_mode();
237 
238 	switch (m->mode) {
239 	case SVC_POOL_PERCPU:
240 		npools = svc_pool_map_init_percpu(m);
241 		break;
242 	case SVC_POOL_PERNODE:
243 		npools = svc_pool_map_init_pernode(m);
244 		break;
245 	}
246 
247 	if (npools < 0) {
248 		/* default, or memory allocation failure */
249 		npools = 1;
250 		m->mode = SVC_POOL_GLOBAL;
251 	}
252 	m->npools = npools;
253 
254 	mutex_unlock(&svc_pool_map_mutex);
255 	return m->npools;
256 }
257 EXPORT_SYMBOL_GPL(svc_pool_map_get);
258 
259 /*
260  * Drop a reference to the global map of cpus to pools.
261  * When the last reference is dropped, the map data is
262  * freed; this allows the sysadmin to change the pool
263  * mode using the pool_mode module option without
264  * rebooting or re-loading sunrpc.ko.
265  */
266 void
267 svc_pool_map_put(void)
268 {
269 	struct svc_pool_map *m = &svc_pool_map;
270 
271 	mutex_lock(&svc_pool_map_mutex);
272 
273 	if (!--m->count) {
274 		kfree(m->to_pool);
275 		m->to_pool = NULL;
276 		kfree(m->pool_to);
277 		m->pool_to = NULL;
278 		m->npools = 0;
279 	}
280 
281 	mutex_unlock(&svc_pool_map_mutex);
282 }
283 EXPORT_SYMBOL_GPL(svc_pool_map_put);
284 
285 static int svc_pool_map_get_node(unsigned int pidx)
286 {
287 	const struct svc_pool_map *m = &svc_pool_map;
288 
289 	if (m->count) {
290 		if (m->mode == SVC_POOL_PERCPU)
291 			return cpu_to_node(m->pool_to[pidx]);
292 		if (m->mode == SVC_POOL_PERNODE)
293 			return m->pool_to[pidx];
294 	}
295 	return NUMA_NO_NODE;
296 }
297 /*
298  * Set the given thread's cpus_allowed mask so that it
299  * will only run on cpus in the given pool.
300  */
301 static inline void
302 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
303 {
304 	struct svc_pool_map *m = &svc_pool_map;
305 	unsigned int node = m->pool_to[pidx];
306 
307 	/*
308 	 * The caller checks for sv_nrpools > 1, which
309 	 * implies that we've been initialized.
310 	 */
311 	WARN_ON_ONCE(m->count == 0);
312 	if (m->count == 0)
313 		return;
314 
315 	switch (m->mode) {
316 	case SVC_POOL_PERCPU:
317 	{
318 		set_cpus_allowed_ptr(task, cpumask_of(node));
319 		break;
320 	}
321 	case SVC_POOL_PERNODE:
322 	{
323 		set_cpus_allowed_ptr(task, cpumask_of_node(node));
324 		break;
325 	}
326 	}
327 }
328 
329 /*
330  * Use the mapping mode to choose a pool for a given CPU.
331  * Used when enqueueing an incoming RPC.  Always returns
332  * a non-NULL pool pointer.
333  */
334 struct svc_pool *
335 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
336 {
337 	struct svc_pool_map *m = &svc_pool_map;
338 	unsigned int pidx = 0;
339 
340 	/*
341 	 * An uninitialised map happens in a pure client when
342 	 * lockd is brought up, so silently treat it the
343 	 * same as SVC_POOL_GLOBAL.
344 	 */
345 	if (svc_serv_is_pooled(serv)) {
346 		switch (m->mode) {
347 		case SVC_POOL_PERCPU:
348 			pidx = m->to_pool[cpu];
349 			break;
350 		case SVC_POOL_PERNODE:
351 			pidx = m->to_pool[cpu_to_node(cpu)];
352 			break;
353 		}
354 	}
355 	return &serv->sv_pools[pidx % serv->sv_nrpools];
356 }
357 
358 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
359 {
360 	int err;
361 
362 	err = rpcb_create_local(net);
363 	if (err)
364 		return err;
365 
366 	/* Remove any stale portmap registrations */
367 	svc_unregister(serv, net);
368 	return 0;
369 }
370 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
371 
372 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
373 {
374 	svc_unregister(serv, net);
375 	rpcb_put_local(net);
376 }
377 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
378 
379 static int svc_uses_rpcbind(struct svc_serv *serv)
380 {
381 	struct svc_program	*progp;
382 	unsigned int		i;
383 
384 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
385 		for (i = 0; i < progp->pg_nvers; i++) {
386 			if (progp->pg_vers[i] == NULL)
387 				continue;
388 			if (!progp->pg_vers[i]->vs_hidden)
389 				return 1;
390 		}
391 	}
392 
393 	return 0;
394 }
395 
396 int svc_bind(struct svc_serv *serv, struct net *net)
397 {
398 	if (!svc_uses_rpcbind(serv))
399 		return 0;
400 	return svc_rpcb_setup(serv, net);
401 }
402 EXPORT_SYMBOL_GPL(svc_bind);
403 
404 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
405 static void
406 __svc_init_bc(struct svc_serv *serv)
407 {
408 	INIT_LIST_HEAD(&serv->sv_cb_list);
409 	spin_lock_init(&serv->sv_cb_lock);
410 	init_waitqueue_head(&serv->sv_cb_waitq);
411 }
412 #else
413 static void
414 __svc_init_bc(struct svc_serv *serv)
415 {
416 }
417 #endif
418 
419 /*
420  * Create an RPC service
421  */
422 static struct svc_serv *
423 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
424 	     const struct svc_serv_ops *ops)
425 {
426 	struct svc_serv	*serv;
427 	unsigned int vers;
428 	unsigned int xdrsize;
429 	unsigned int i;
430 
431 	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
432 		return NULL;
433 	serv->sv_name      = prog->pg_name;
434 	serv->sv_program   = prog;
435 	serv->sv_nrthreads = 1;
436 	serv->sv_stats     = prog->pg_stats;
437 	if (bufsize > RPCSVC_MAXPAYLOAD)
438 		bufsize = RPCSVC_MAXPAYLOAD;
439 	serv->sv_max_payload = bufsize? bufsize : 4096;
440 	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
441 	serv->sv_ops = ops;
442 	xdrsize = 0;
443 	while (prog) {
444 		prog->pg_lovers = prog->pg_nvers-1;
445 		for (vers=0; vers<prog->pg_nvers ; vers++)
446 			if (prog->pg_vers[vers]) {
447 				prog->pg_hivers = vers;
448 				if (prog->pg_lovers > vers)
449 					prog->pg_lovers = vers;
450 				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
451 					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
452 			}
453 		prog = prog->pg_next;
454 	}
455 	serv->sv_xdrsize   = xdrsize;
456 	INIT_LIST_HEAD(&serv->sv_tempsocks);
457 	INIT_LIST_HEAD(&serv->sv_permsocks);
458 	timer_setup(&serv->sv_temptimer, NULL, 0);
459 	spin_lock_init(&serv->sv_lock);
460 
461 	__svc_init_bc(serv);
462 
463 	serv->sv_nrpools = npools;
464 	serv->sv_pools =
465 		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
466 			GFP_KERNEL);
467 	if (!serv->sv_pools) {
468 		kfree(serv);
469 		return NULL;
470 	}
471 
472 	for (i = 0; i < serv->sv_nrpools; i++) {
473 		struct svc_pool *pool = &serv->sv_pools[i];
474 
475 		dprintk("svc: initialising pool %u for %s\n",
476 				i, serv->sv_name);
477 
478 		pool->sp_id = i;
479 		INIT_LIST_HEAD(&pool->sp_sockets);
480 		INIT_LIST_HEAD(&pool->sp_all_threads);
481 		spin_lock_init(&pool->sp_lock);
482 	}
483 
484 	return serv;
485 }
486 
487 struct svc_serv *
488 svc_create(struct svc_program *prog, unsigned int bufsize,
489 	   const struct svc_serv_ops *ops)
490 {
491 	return __svc_create(prog, bufsize, /*npools*/1, ops);
492 }
493 EXPORT_SYMBOL_GPL(svc_create);
494 
495 struct svc_serv *
496 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
497 		  const struct svc_serv_ops *ops)
498 {
499 	struct svc_serv *serv;
500 	unsigned int npools = svc_pool_map_get();
501 
502 	serv = __svc_create(prog, bufsize, npools, ops);
503 	if (!serv)
504 		goto out_err;
505 	return serv;
506 out_err:
507 	svc_pool_map_put();
508 	return NULL;
509 }
510 EXPORT_SYMBOL_GPL(svc_create_pooled);
511 
512 void svc_shutdown_net(struct svc_serv *serv, struct net *net)
513 {
514 	svc_close_net(serv, net);
515 
516 	if (serv->sv_ops->svo_shutdown)
517 		serv->sv_ops->svo_shutdown(serv, net);
518 }
519 EXPORT_SYMBOL_GPL(svc_shutdown_net);
520 
521 /*
522  * Destroy an RPC service. Should be called with appropriate locking to
523  * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
524  */
525 void
526 svc_destroy(struct svc_serv *serv)
527 {
528 	dprintk("svc: svc_destroy(%s, %d)\n",
529 				serv->sv_program->pg_name,
530 				serv->sv_nrthreads);
531 
532 	if (serv->sv_nrthreads) {
533 		if (--(serv->sv_nrthreads) != 0) {
534 			svc_sock_update_bufs(serv);
535 			return;
536 		}
537 	} else
538 		printk("svc_destroy: no threads for serv=%p!\n", serv);
539 
540 	del_timer_sync(&serv->sv_temptimer);
541 
542 	/*
543 	 * The last user is gone and thus all sockets have to be destroyed to
544 	 * the point. Check this.
545 	 */
546 	BUG_ON(!list_empty(&serv->sv_permsocks));
547 	BUG_ON(!list_empty(&serv->sv_tempsocks));
548 
549 	cache_clean_deferred(serv);
550 
551 	if (svc_serv_is_pooled(serv))
552 		svc_pool_map_put();
553 
554 	kfree(serv->sv_pools);
555 	kfree(serv);
556 }
557 EXPORT_SYMBOL_GPL(svc_destroy);
558 
559 /*
560  * Allocate an RPC server's buffer space.
561  * We allocate pages and place them in rq_argpages.
562  */
563 static int
564 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
565 {
566 	unsigned int pages, arghi;
567 
568 	/* bc_xprt uses fore channel allocated buffers */
569 	if (svc_is_backchannel(rqstp))
570 		return 1;
571 
572 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
573 				       * We assume one is at most one page
574 				       */
575 	arghi = 0;
576 	WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
577 	if (pages > RPCSVC_MAXPAGES)
578 		pages = RPCSVC_MAXPAGES;
579 	while (pages) {
580 		struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
581 		if (!p)
582 			break;
583 		rqstp->rq_pages[arghi++] = p;
584 		pages--;
585 	}
586 	return pages == 0;
587 }
588 
589 /*
590  * Release an RPC server buffer
591  */
592 static void
593 svc_release_buffer(struct svc_rqst *rqstp)
594 {
595 	unsigned int i;
596 
597 	for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
598 		if (rqstp->rq_pages[i])
599 			put_page(rqstp->rq_pages[i]);
600 }
601 
602 struct svc_rqst *
603 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
604 {
605 	struct svc_rqst	*rqstp;
606 
607 	rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
608 	if (!rqstp)
609 		return rqstp;
610 
611 	__set_bit(RQ_BUSY, &rqstp->rq_flags);
612 	spin_lock_init(&rqstp->rq_lock);
613 	rqstp->rq_server = serv;
614 	rqstp->rq_pool = pool;
615 
616 	rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
617 	if (!rqstp->rq_argp)
618 		goto out_enomem;
619 
620 	rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
621 	if (!rqstp->rq_resp)
622 		goto out_enomem;
623 
624 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
625 		goto out_enomem;
626 
627 	return rqstp;
628 out_enomem:
629 	svc_rqst_free(rqstp);
630 	return NULL;
631 }
632 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
633 
634 struct svc_rqst *
635 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
636 {
637 	struct svc_rqst	*rqstp;
638 
639 	rqstp = svc_rqst_alloc(serv, pool, node);
640 	if (!rqstp)
641 		return ERR_PTR(-ENOMEM);
642 
643 	serv->sv_nrthreads++;
644 	spin_lock_bh(&pool->sp_lock);
645 	pool->sp_nrthreads++;
646 	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
647 	spin_unlock_bh(&pool->sp_lock);
648 	return rqstp;
649 }
650 EXPORT_SYMBOL_GPL(svc_prepare_thread);
651 
652 /*
653  * Choose a pool in which to create a new thread, for svc_set_num_threads
654  */
655 static inline struct svc_pool *
656 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
657 {
658 	if (pool != NULL)
659 		return pool;
660 
661 	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
662 }
663 
664 /*
665  * Choose a thread to kill, for svc_set_num_threads
666  */
667 static inline struct task_struct *
668 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
669 {
670 	unsigned int i;
671 	struct task_struct *task = NULL;
672 
673 	if (pool != NULL) {
674 		spin_lock_bh(&pool->sp_lock);
675 	} else {
676 		/* choose a pool in round-robin fashion */
677 		for (i = 0; i < serv->sv_nrpools; i++) {
678 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
679 			spin_lock_bh(&pool->sp_lock);
680 			if (!list_empty(&pool->sp_all_threads))
681 				goto found_pool;
682 			spin_unlock_bh(&pool->sp_lock);
683 		}
684 		return NULL;
685 	}
686 
687 found_pool:
688 	if (!list_empty(&pool->sp_all_threads)) {
689 		struct svc_rqst *rqstp;
690 
691 		/*
692 		 * Remove from the pool->sp_all_threads list
693 		 * so we don't try to kill it again.
694 		 */
695 		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
696 		set_bit(RQ_VICTIM, &rqstp->rq_flags);
697 		list_del_rcu(&rqstp->rq_all);
698 		task = rqstp->rq_task;
699 	}
700 	spin_unlock_bh(&pool->sp_lock);
701 
702 	return task;
703 }
704 
705 /* create new threads */
706 static int
707 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
708 {
709 	struct svc_rqst	*rqstp;
710 	struct task_struct *task;
711 	struct svc_pool *chosen_pool;
712 	unsigned int state = serv->sv_nrthreads-1;
713 	int node;
714 
715 	do {
716 		nrservs--;
717 		chosen_pool = choose_pool(serv, pool, &state);
718 
719 		node = svc_pool_map_get_node(chosen_pool->sp_id);
720 		rqstp = svc_prepare_thread(serv, chosen_pool, node);
721 		if (IS_ERR(rqstp))
722 			return PTR_ERR(rqstp);
723 
724 		__module_get(serv->sv_ops->svo_module);
725 		task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
726 					      node, "%s", serv->sv_name);
727 		if (IS_ERR(task)) {
728 			module_put(serv->sv_ops->svo_module);
729 			svc_exit_thread(rqstp);
730 			return PTR_ERR(task);
731 		}
732 
733 		rqstp->rq_task = task;
734 		if (serv->sv_nrpools > 1)
735 			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
736 
737 		svc_sock_update_bufs(serv);
738 		wake_up_process(task);
739 	} while (nrservs > 0);
740 
741 	return 0;
742 }
743 
744 
745 /* destroy old threads */
746 static int
747 svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
748 {
749 	struct task_struct *task;
750 	unsigned int state = serv->sv_nrthreads-1;
751 
752 	/* destroy old threads */
753 	do {
754 		task = choose_victim(serv, pool, &state);
755 		if (task == NULL)
756 			break;
757 		send_sig(SIGINT, task, 1);
758 		nrservs++;
759 	} while (nrservs < 0);
760 
761 	return 0;
762 }
763 
764 /*
765  * Create or destroy enough new threads to make the number
766  * of threads the given number.  If `pool' is non-NULL, applies
767  * only to threads in that pool, otherwise round-robins between
768  * all pools.  Caller must ensure that mutual exclusion between this and
769  * server startup or shutdown.
770  *
771  * Destroying threads relies on the service threads filling in
772  * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
773  * has been created using svc_create_pooled().
774  *
775  * Based on code that used to be in nfsd_svc() but tweaked
776  * to be pool-aware.
777  */
778 int
779 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
780 {
781 	if (pool == NULL) {
782 		/* The -1 assumes caller has done a svc_get() */
783 		nrservs -= (serv->sv_nrthreads-1);
784 	} else {
785 		spin_lock_bh(&pool->sp_lock);
786 		nrservs -= pool->sp_nrthreads;
787 		spin_unlock_bh(&pool->sp_lock);
788 	}
789 
790 	if (nrservs > 0)
791 		return svc_start_kthreads(serv, pool, nrservs);
792 	if (nrservs < 0)
793 		return svc_signal_kthreads(serv, pool, nrservs);
794 	return 0;
795 }
796 EXPORT_SYMBOL_GPL(svc_set_num_threads);
797 
798 /* destroy old threads */
799 static int
800 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
801 {
802 	struct task_struct *task;
803 	unsigned int state = serv->sv_nrthreads-1;
804 
805 	/* destroy old threads */
806 	do {
807 		task = choose_victim(serv, pool, &state);
808 		if (task == NULL)
809 			break;
810 		kthread_stop(task);
811 		nrservs++;
812 	} while (nrservs < 0);
813 	return 0;
814 }
815 
816 int
817 svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
818 {
819 	if (pool == NULL) {
820 		/* The -1 assumes caller has done a svc_get() */
821 		nrservs -= (serv->sv_nrthreads-1);
822 	} else {
823 		spin_lock_bh(&pool->sp_lock);
824 		nrservs -= pool->sp_nrthreads;
825 		spin_unlock_bh(&pool->sp_lock);
826 	}
827 
828 	if (nrservs > 0)
829 		return svc_start_kthreads(serv, pool, nrservs);
830 	if (nrservs < 0)
831 		return svc_stop_kthreads(serv, pool, nrservs);
832 	return 0;
833 }
834 EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
835 
836 /*
837  * Called from a server thread as it's exiting. Caller must hold the "service
838  * mutex" for the service.
839  */
840 void
841 svc_rqst_free(struct svc_rqst *rqstp)
842 {
843 	svc_release_buffer(rqstp);
844 	kfree(rqstp->rq_resp);
845 	kfree(rqstp->rq_argp);
846 	kfree(rqstp->rq_auth_data);
847 	kfree_rcu(rqstp, rq_rcu_head);
848 }
849 EXPORT_SYMBOL_GPL(svc_rqst_free);
850 
851 void
852 svc_exit_thread(struct svc_rqst *rqstp)
853 {
854 	struct svc_serv	*serv = rqstp->rq_server;
855 	struct svc_pool	*pool = rqstp->rq_pool;
856 
857 	spin_lock_bh(&pool->sp_lock);
858 	pool->sp_nrthreads--;
859 	if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
860 		list_del_rcu(&rqstp->rq_all);
861 	spin_unlock_bh(&pool->sp_lock);
862 
863 	svc_rqst_free(rqstp);
864 
865 	/* Release the server */
866 	if (serv)
867 		svc_destroy(serv);
868 }
869 EXPORT_SYMBOL_GPL(svc_exit_thread);
870 
871 /*
872  * Register an "inet" protocol family netid with the local
873  * rpcbind daemon via an rpcbind v4 SET request.
874  *
875  * No netconfig infrastructure is available in the kernel, so
876  * we map IP_ protocol numbers to netids by hand.
877  *
878  * Returns zero on success; a negative errno value is returned
879  * if any error occurs.
880  */
881 static int __svc_rpcb_register4(struct net *net, const u32 program,
882 				const u32 version,
883 				const unsigned short protocol,
884 				const unsigned short port)
885 {
886 	const struct sockaddr_in sin = {
887 		.sin_family		= AF_INET,
888 		.sin_addr.s_addr	= htonl(INADDR_ANY),
889 		.sin_port		= htons(port),
890 	};
891 	const char *netid;
892 	int error;
893 
894 	switch (protocol) {
895 	case IPPROTO_UDP:
896 		netid = RPCBIND_NETID_UDP;
897 		break;
898 	case IPPROTO_TCP:
899 		netid = RPCBIND_NETID_TCP;
900 		break;
901 	default:
902 		return -ENOPROTOOPT;
903 	}
904 
905 	error = rpcb_v4_register(net, program, version,
906 					(const struct sockaddr *)&sin, netid);
907 
908 	/*
909 	 * User space didn't support rpcbind v4, so retry this
910 	 * registration request with the legacy rpcbind v2 protocol.
911 	 */
912 	if (error == -EPROTONOSUPPORT)
913 		error = rpcb_register(net, program, version, protocol, port);
914 
915 	return error;
916 }
917 
918 #if IS_ENABLED(CONFIG_IPV6)
919 /*
920  * Register an "inet6" protocol family netid with the local
921  * rpcbind daemon via an rpcbind v4 SET request.
922  *
923  * No netconfig infrastructure is available in the kernel, so
924  * we map IP_ protocol numbers to netids by hand.
925  *
926  * Returns zero on success; a negative errno value is returned
927  * if any error occurs.
928  */
929 static int __svc_rpcb_register6(struct net *net, const u32 program,
930 				const u32 version,
931 				const unsigned short protocol,
932 				const unsigned short port)
933 {
934 	const struct sockaddr_in6 sin6 = {
935 		.sin6_family		= AF_INET6,
936 		.sin6_addr		= IN6ADDR_ANY_INIT,
937 		.sin6_port		= htons(port),
938 	};
939 	const char *netid;
940 	int error;
941 
942 	switch (protocol) {
943 	case IPPROTO_UDP:
944 		netid = RPCBIND_NETID_UDP6;
945 		break;
946 	case IPPROTO_TCP:
947 		netid = RPCBIND_NETID_TCP6;
948 		break;
949 	default:
950 		return -ENOPROTOOPT;
951 	}
952 
953 	error = rpcb_v4_register(net, program, version,
954 					(const struct sockaddr *)&sin6, netid);
955 
956 	/*
957 	 * User space didn't support rpcbind version 4, so we won't
958 	 * use a PF_INET6 listener.
959 	 */
960 	if (error == -EPROTONOSUPPORT)
961 		error = -EAFNOSUPPORT;
962 
963 	return error;
964 }
965 #endif	/* IS_ENABLED(CONFIG_IPV6) */
966 
967 /*
968  * Register a kernel RPC service via rpcbind version 4.
969  *
970  * Returns zero on success; a negative errno value is returned
971  * if any error occurs.
972  */
973 static int __svc_register(struct net *net, const char *progname,
974 			  const u32 program, const u32 version,
975 			  const int family,
976 			  const unsigned short protocol,
977 			  const unsigned short port)
978 {
979 	int error = -EAFNOSUPPORT;
980 
981 	switch (family) {
982 	case PF_INET:
983 		error = __svc_rpcb_register4(net, program, version,
984 						protocol, port);
985 		break;
986 #if IS_ENABLED(CONFIG_IPV6)
987 	case PF_INET6:
988 		error = __svc_rpcb_register6(net, program, version,
989 						protocol, port);
990 #endif
991 	}
992 
993 	return error;
994 }
995 
996 int svc_rpcbind_set_version(struct net *net,
997 			    const struct svc_program *progp,
998 			    u32 version, int family,
999 			    unsigned short proto,
1000 			    unsigned short port)
1001 {
1002 	dprintk("svc: svc_register(%sv%d, %s, %u, %u)\n",
1003 		progp->pg_name, version,
1004 		proto == IPPROTO_UDP?  "udp" : "tcp",
1005 		port, family);
1006 
1007 	return __svc_register(net, progp->pg_name, progp->pg_prog,
1008 				version, family, proto, port);
1009 
1010 }
1011 EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1012 
1013 int svc_generic_rpcbind_set(struct net *net,
1014 			    const struct svc_program *progp,
1015 			    u32 version, int family,
1016 			    unsigned short proto,
1017 			    unsigned short port)
1018 {
1019 	const struct svc_version *vers = progp->pg_vers[version];
1020 	int error;
1021 
1022 	if (vers == NULL)
1023 		return 0;
1024 
1025 	if (vers->vs_hidden) {
1026 		dprintk("svc: svc_register(%sv%d, %s, %u, %u)"
1027 			" (but not telling portmap)\n",
1028 			progp->pg_name, version,
1029 			proto == IPPROTO_UDP?  "udp" : "tcp",
1030 			port, family);
1031 		return 0;
1032 	}
1033 
1034 	/*
1035 	 * Don't register a UDP port if we need congestion
1036 	 * control.
1037 	 */
1038 	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1039 		return 0;
1040 
1041 	error = svc_rpcbind_set_version(net, progp, version,
1042 					family, proto, port);
1043 
1044 	return (vers->vs_rpcb_optnl) ? 0 : error;
1045 }
1046 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1047 
1048 /**
1049  * svc_register - register an RPC service with the local portmapper
1050  * @serv: svc_serv struct for the service to register
1051  * @net: net namespace for the service to register
1052  * @family: protocol family of service's listener socket
1053  * @proto: transport protocol number to advertise
1054  * @port: port to advertise
1055  *
1056  * Service is registered for any address in the passed-in protocol family
1057  */
1058 int svc_register(const struct svc_serv *serv, struct net *net,
1059 		 const int family, const unsigned short proto,
1060 		 const unsigned short port)
1061 {
1062 	struct svc_program	*progp;
1063 	unsigned int		i;
1064 	int			error = 0;
1065 
1066 	WARN_ON_ONCE(proto == 0 && port == 0);
1067 	if (proto == 0 && port == 0)
1068 		return -EINVAL;
1069 
1070 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1071 		for (i = 0; i < progp->pg_nvers; i++) {
1072 
1073 			error = progp->pg_rpcbind_set(net, progp, i,
1074 					family, proto, port);
1075 			if (error < 0) {
1076 				printk(KERN_WARNING "svc: failed to register "
1077 					"%sv%u RPC service (errno %d).\n",
1078 					progp->pg_name, i, -error);
1079 				break;
1080 			}
1081 		}
1082 	}
1083 
1084 	return error;
1085 }
1086 
1087 /*
1088  * If user space is running rpcbind, it should take the v4 UNSET
1089  * and clear everything for this [program, version].  If user space
1090  * is running portmap, it will reject the v4 UNSET, but won't have
1091  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1092  * in this case to clear all existing entries for [program, version].
1093  */
1094 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1095 			     const char *progname)
1096 {
1097 	int error;
1098 
1099 	error = rpcb_v4_register(net, program, version, NULL, "");
1100 
1101 	/*
1102 	 * User space didn't support rpcbind v4, so retry this
1103 	 * request with the legacy rpcbind v2 protocol.
1104 	 */
1105 	if (error == -EPROTONOSUPPORT)
1106 		error = rpcb_register(net, program, version, 0, 0);
1107 
1108 	dprintk("svc: %s(%sv%u), error %d\n",
1109 			__func__, progname, version, error);
1110 }
1111 
1112 /*
1113  * All netids, bind addresses and ports registered for [program, version]
1114  * are removed from the local rpcbind database (if the service is not
1115  * hidden) to make way for a new instance of the service.
1116  *
1117  * The result of unregistration is reported via dprintk for those who want
1118  * verification of the result, but is otherwise not important.
1119  */
1120 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1121 {
1122 	struct svc_program *progp;
1123 	unsigned long flags;
1124 	unsigned int i;
1125 
1126 	clear_thread_flag(TIF_SIGPENDING);
1127 
1128 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1129 		for (i = 0; i < progp->pg_nvers; i++) {
1130 			if (progp->pg_vers[i] == NULL)
1131 				continue;
1132 			if (progp->pg_vers[i]->vs_hidden)
1133 				continue;
1134 
1135 			dprintk("svc: attempting to unregister %sv%u\n",
1136 				progp->pg_name, i);
1137 			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1138 		}
1139 	}
1140 
1141 	spin_lock_irqsave(&current->sighand->siglock, flags);
1142 	recalc_sigpending();
1143 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
1144 }
1145 
1146 /*
1147  * dprintk the given error with the address of the client that caused it.
1148  */
1149 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1150 static __printf(2, 3)
1151 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1152 {
1153 	struct va_format vaf;
1154 	va_list args;
1155 	char 	buf[RPC_MAX_ADDRBUFLEN];
1156 
1157 	va_start(args, fmt);
1158 
1159 	vaf.fmt = fmt;
1160 	vaf.va = &args;
1161 
1162 	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1163 
1164 	va_end(args);
1165 }
1166 #else
1167 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1168 #endif
1169 
1170 __be32
1171 svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err)
1172 {
1173 	set_bit(RQ_AUTHERR, &rqstp->rq_flags);
1174 	return auth_err;
1175 }
1176 EXPORT_SYMBOL_GPL(svc_return_autherr);
1177 
1178 static __be32
1179 svc_get_autherr(struct svc_rqst *rqstp, __be32 *statp)
1180 {
1181 	if (test_and_clear_bit(RQ_AUTHERR, &rqstp->rq_flags))
1182 		return *statp;
1183 	return rpc_auth_ok;
1184 }
1185 
1186 static int
1187 svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1188 {
1189 	struct kvec *argv = &rqstp->rq_arg.head[0];
1190 	struct kvec *resv = &rqstp->rq_res.head[0];
1191 	const struct svc_procedure *procp = rqstp->rq_procinfo;
1192 
1193 	/*
1194 	 * Decode arguments
1195 	 * XXX: why do we ignore the return value?
1196 	 */
1197 	if (procp->pc_decode &&
1198 	    !procp->pc_decode(rqstp, argv->iov_base)) {
1199 		*statp = rpc_garbage_args;
1200 		return 1;
1201 	}
1202 
1203 	*statp = procp->pc_func(rqstp);
1204 
1205 	if (*statp == rpc_drop_reply ||
1206 	    test_bit(RQ_DROPME, &rqstp->rq_flags))
1207 		return 0;
1208 
1209 	if (test_bit(RQ_AUTHERR, &rqstp->rq_flags))
1210 		return 1;
1211 
1212 	if (*statp != rpc_success)
1213 		return 1;
1214 
1215 	/* Encode reply */
1216 	if (procp->pc_encode &&
1217 	    !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1218 		dprintk("svc: failed to encode reply\n");
1219 		/* serv->sv_stats->rpcsystemerr++; */
1220 		*statp = rpc_system_err;
1221 	}
1222 	return 1;
1223 }
1224 
1225 __be32
1226 svc_generic_init_request(struct svc_rqst *rqstp,
1227 		const struct svc_program *progp,
1228 		struct svc_process_info *ret)
1229 {
1230 	const struct svc_version *versp = NULL;	/* compiler food */
1231 	const struct svc_procedure *procp = NULL;
1232 
1233 	if (rqstp->rq_vers >= progp->pg_nvers )
1234 		goto err_bad_vers;
1235 	  versp = progp->pg_vers[rqstp->rq_vers];
1236 	  if (!versp)
1237 		goto err_bad_vers;
1238 
1239 	/*
1240 	 * Some protocol versions (namely NFSv4) require some form of
1241 	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1242 	 * In other words, UDP is not allowed. We mark those when setting
1243 	 * up the svc_xprt, and verify that here.
1244 	 *
1245 	 * The spec is not very clear about what error should be returned
1246 	 * when someone tries to access a server that is listening on UDP
1247 	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1248 	 * fit.
1249 	 */
1250 	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1251 	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1252 		goto err_bad_vers;
1253 
1254 	if (rqstp->rq_proc >= versp->vs_nproc)
1255 		goto err_bad_proc;
1256 	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1257 	if (!procp)
1258 		goto err_bad_proc;
1259 
1260 	/* Initialize storage for argp and resp */
1261 	memset(rqstp->rq_argp, 0, procp->pc_argsize);
1262 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1263 
1264 	/* Bump per-procedure stats counter */
1265 	versp->vs_count[rqstp->rq_proc]++;
1266 
1267 	ret->dispatch = versp->vs_dispatch;
1268 	return rpc_success;
1269 err_bad_vers:
1270 	ret->mismatch.lovers = progp->pg_lovers;
1271 	ret->mismatch.hivers = progp->pg_hivers;
1272 	return rpc_prog_mismatch;
1273 err_bad_proc:
1274 	return rpc_proc_unavail;
1275 }
1276 EXPORT_SYMBOL_GPL(svc_generic_init_request);
1277 
1278 /*
1279  * Common routine for processing the RPC request.
1280  */
1281 static int
1282 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1283 {
1284 	struct svc_program	*progp;
1285 	const struct svc_procedure *procp = NULL;
1286 	struct svc_serv		*serv = rqstp->rq_server;
1287 	struct svc_process_info process;
1288 	__be32			*statp;
1289 	u32			prog, vers;
1290 	__be32			auth_stat, rpc_stat;
1291 	int			auth_res;
1292 	__be32			*reply_statp;
1293 
1294 	rpc_stat = rpc_success;
1295 
1296 	if (argv->iov_len < 6*4)
1297 		goto err_short_len;
1298 
1299 	/* Will be turned off by GSS integrity and privacy services */
1300 	set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1301 	/* Will be turned off only when NFSv4 Sessions are used */
1302 	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1303 	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1304 
1305 	svc_putu32(resv, rqstp->rq_xid);
1306 
1307 	vers = svc_getnl(argv);
1308 
1309 	/* First words of reply: */
1310 	svc_putnl(resv, 1);		/* REPLY */
1311 
1312 	if (vers != 2)		/* RPC version number */
1313 		goto err_bad_rpc;
1314 
1315 	/* Save position in case we later decide to reject: */
1316 	reply_statp = resv->iov_base + resv->iov_len;
1317 
1318 	svc_putnl(resv, 0);		/* ACCEPT */
1319 
1320 	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
1321 	rqstp->rq_vers = svc_getnl(argv);	/* version number */
1322 	rqstp->rq_proc = svc_getnl(argv);	/* procedure number */
1323 
1324 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1325 		if (prog == progp->pg_prog)
1326 			break;
1327 
1328 	/*
1329 	 * Decode auth data, and add verifier to reply buffer.
1330 	 * We do this before anything else in order to get a decent
1331 	 * auth verifier.
1332 	 */
1333 	auth_res = svc_authenticate(rqstp, &auth_stat);
1334 	/* Also give the program a chance to reject this call: */
1335 	if (auth_res == SVC_OK && progp) {
1336 		auth_stat = rpc_autherr_badcred;
1337 		auth_res = progp->pg_authenticate(rqstp);
1338 	}
1339 	switch (auth_res) {
1340 	case SVC_OK:
1341 		break;
1342 	case SVC_GARBAGE:
1343 		goto err_garbage;
1344 	case SVC_SYSERR:
1345 		rpc_stat = rpc_system_err;
1346 		goto err_bad;
1347 	case SVC_DENIED:
1348 		goto err_bad_auth;
1349 	case SVC_CLOSE:
1350 		goto close;
1351 	case SVC_DROP:
1352 		goto dropit;
1353 	case SVC_COMPLETE:
1354 		goto sendit;
1355 	}
1356 
1357 	if (progp == NULL)
1358 		goto err_bad_prog;
1359 
1360 	rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1361 	switch (rpc_stat) {
1362 	case rpc_success:
1363 		break;
1364 	case rpc_prog_unavail:
1365 		goto err_bad_prog;
1366 	case rpc_prog_mismatch:
1367 		goto err_bad_vers;
1368 	case rpc_proc_unavail:
1369 		goto err_bad_proc;
1370 	}
1371 
1372 	procp = rqstp->rq_procinfo;
1373 	/* Should this check go into the dispatcher? */
1374 	if (!procp || !procp->pc_func)
1375 		goto err_bad_proc;
1376 
1377 	/* Syntactic check complete */
1378 	serv->sv_stats->rpccnt++;
1379 	trace_svc_process(rqstp, progp->pg_name);
1380 
1381 	/* Build the reply header. */
1382 	statp = resv->iov_base +resv->iov_len;
1383 	svc_putnl(resv, RPC_SUCCESS);
1384 
1385 	/* un-reserve some of the out-queue now that we have a
1386 	 * better idea of reply size
1387 	 */
1388 	if (procp->pc_xdrressize)
1389 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1390 
1391 	/* Call the function that processes the request. */
1392 	if (!process.dispatch) {
1393 		if (!svc_generic_dispatch(rqstp, statp))
1394 			goto release_dropit;
1395 		if (*statp == rpc_garbage_args)
1396 			goto err_garbage;
1397 		auth_stat = svc_get_autherr(rqstp, statp);
1398 		if (auth_stat != rpc_auth_ok)
1399 			goto err_release_bad_auth;
1400 	} else {
1401 		dprintk("svc: calling dispatcher\n");
1402 		if (!process.dispatch(rqstp, statp))
1403 			goto release_dropit; /* Release reply info */
1404 	}
1405 
1406 	/* Check RPC status result */
1407 	if (*statp != rpc_success)
1408 		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1409 
1410 	/* Release reply info */
1411 	if (procp->pc_release)
1412 		procp->pc_release(rqstp);
1413 
1414 	if (procp->pc_encode == NULL)
1415 		goto dropit;
1416 
1417  sendit:
1418 	if (svc_authorise(rqstp))
1419 		goto close;
1420 	return 1;		/* Caller can now send it */
1421 
1422 release_dropit:
1423 	if (procp->pc_release)
1424 		procp->pc_release(rqstp);
1425  dropit:
1426 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1427 	dprintk("svc: svc_process dropit\n");
1428 	return 0;
1429 
1430  close:
1431 	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1432 		svc_close_xprt(rqstp->rq_xprt);
1433 	dprintk("svc: svc_process close\n");
1434 	return 0;
1435 
1436 err_short_len:
1437 	svc_printk(rqstp, "short len %zd, dropping request\n",
1438 			argv->iov_len);
1439 	goto close;
1440 
1441 err_bad_rpc:
1442 	serv->sv_stats->rpcbadfmt++;
1443 	svc_putnl(resv, 1);	/* REJECT */
1444 	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1445 	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1446 	svc_putnl(resv, 2);
1447 	goto sendit;
1448 
1449 err_release_bad_auth:
1450 	if (procp->pc_release)
1451 		procp->pc_release(rqstp);
1452 err_bad_auth:
1453 	dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1454 	serv->sv_stats->rpcbadauth++;
1455 	/* Restore write pointer to location of accept status: */
1456 	xdr_ressize_check(rqstp, reply_statp);
1457 	svc_putnl(resv, 1);	/* REJECT */
1458 	svc_putnl(resv, 1);	/* AUTH_ERROR */
1459 	svc_putnl(resv, ntohl(auth_stat));	/* status */
1460 	goto sendit;
1461 
1462 err_bad_prog:
1463 	dprintk("svc: unknown program %d\n", prog);
1464 	serv->sv_stats->rpcbadfmt++;
1465 	svc_putnl(resv, RPC_PROG_UNAVAIL);
1466 	goto sendit;
1467 
1468 err_bad_vers:
1469 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1470 		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1471 
1472 	serv->sv_stats->rpcbadfmt++;
1473 	svc_putnl(resv, RPC_PROG_MISMATCH);
1474 	svc_putnl(resv, process.mismatch.lovers);
1475 	svc_putnl(resv, process.mismatch.hivers);
1476 	goto sendit;
1477 
1478 err_bad_proc:
1479 	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1480 
1481 	serv->sv_stats->rpcbadfmt++;
1482 	svc_putnl(resv, RPC_PROC_UNAVAIL);
1483 	goto sendit;
1484 
1485 err_garbage:
1486 	svc_printk(rqstp, "failed to decode args\n");
1487 
1488 	rpc_stat = rpc_garbage_args;
1489 err_bad:
1490 	serv->sv_stats->rpcbadfmt++;
1491 	svc_putnl(resv, ntohl(rpc_stat));
1492 	goto sendit;
1493 }
1494 
1495 /*
1496  * Process the RPC request.
1497  */
1498 int
1499 svc_process(struct svc_rqst *rqstp)
1500 {
1501 	struct kvec		*argv = &rqstp->rq_arg.head[0];
1502 	struct kvec		*resv = &rqstp->rq_res.head[0];
1503 	struct svc_serv		*serv = rqstp->rq_server;
1504 	u32			dir;
1505 
1506 	/*
1507 	 * Setup response xdr_buf.
1508 	 * Initially it has just one page
1509 	 */
1510 	rqstp->rq_next_page = &rqstp->rq_respages[1];
1511 	resv->iov_base = page_address(rqstp->rq_respages[0]);
1512 	resv->iov_len = 0;
1513 	rqstp->rq_res.pages = rqstp->rq_respages + 1;
1514 	rqstp->rq_res.len = 0;
1515 	rqstp->rq_res.page_base = 0;
1516 	rqstp->rq_res.page_len = 0;
1517 	rqstp->rq_res.buflen = PAGE_SIZE;
1518 	rqstp->rq_res.tail[0].iov_base = NULL;
1519 	rqstp->rq_res.tail[0].iov_len = 0;
1520 
1521 	dir  = svc_getnl(argv);
1522 	if (dir != 0) {
1523 		/* direction != CALL */
1524 		svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1525 		serv->sv_stats->rpcbadfmt++;
1526 		goto out_drop;
1527 	}
1528 
1529 	/* Reserve space for the record marker */
1530 	if (rqstp->rq_prot == IPPROTO_TCP)
1531 		svc_putnl(resv, 0);
1532 
1533 	/* Returns 1 for send, 0 for drop */
1534 	if (likely(svc_process_common(rqstp, argv, resv)))
1535 		return svc_send(rqstp);
1536 
1537 out_drop:
1538 	svc_drop(rqstp);
1539 	return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(svc_process);
1542 
1543 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1544 /*
1545  * Process a backchannel RPC request that arrived over an existing
1546  * outbound connection
1547  */
1548 int
1549 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1550 	       struct svc_rqst *rqstp)
1551 {
1552 	struct kvec	*argv = &rqstp->rq_arg.head[0];
1553 	struct kvec	*resv = &rqstp->rq_res.head[0];
1554 	struct rpc_task *task;
1555 	int proc_error;
1556 	int error;
1557 
1558 	dprintk("svc: %s(%p)\n", __func__, req);
1559 
1560 	/* Build the svc_rqst used by the common processing routine */
1561 	rqstp->rq_xid = req->rq_xid;
1562 	rqstp->rq_prot = req->rq_xprt->prot;
1563 	rqstp->rq_server = serv;
1564 	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1565 
1566 	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1567 	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1568 	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1569 	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1570 
1571 	/* Adjust the argument buffer length */
1572 	rqstp->rq_arg.len = req->rq_private_buf.len;
1573 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1574 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1575 		rqstp->rq_arg.page_len = 0;
1576 	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1577 			rqstp->rq_arg.page_len)
1578 		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1579 			rqstp->rq_arg.head[0].iov_len;
1580 	else
1581 		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1582 			rqstp->rq_arg.page_len;
1583 
1584 	/* reset result send buffer "put" position */
1585 	resv->iov_len = 0;
1586 
1587 	/*
1588 	 * Skip the next two words because they've already been
1589 	 * processed in the transport
1590 	 */
1591 	svc_getu32(argv);	/* XID */
1592 	svc_getnl(argv);	/* CALLDIR */
1593 
1594 	/* Parse and execute the bc call */
1595 	proc_error = svc_process_common(rqstp, argv, resv);
1596 
1597 	atomic_inc(&req->rq_xprt->bc_free_slots);
1598 	if (!proc_error) {
1599 		/* Processing error: drop the request */
1600 		xprt_free_bc_request(req);
1601 		error = -EINVAL;
1602 		goto out;
1603 	}
1604 	/* Finally, send the reply synchronously */
1605 	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1606 	task = rpc_run_bc_task(req);
1607 	if (IS_ERR(task)) {
1608 		error = PTR_ERR(task);
1609 		goto out;
1610 	}
1611 
1612 	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1613 	error = task->tk_status;
1614 	rpc_put_task(task);
1615 
1616 out:
1617 	dprintk("svc: %s(), error=%d\n", __func__, error);
1618 	return error;
1619 }
1620 EXPORT_SYMBOL_GPL(bc_svc_process);
1621 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1622 
1623 /*
1624  * Return (transport-specific) limit on the rpc payload.
1625  */
1626 u32 svc_max_payload(const struct svc_rqst *rqstp)
1627 {
1628 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1629 
1630 	if (rqstp->rq_server->sv_max_payload < max)
1631 		max = rqstp->rq_server->sv_max_payload;
1632 	return max;
1633 }
1634 EXPORT_SYMBOL_GPL(svc_max_payload);
1635 
1636 /**
1637  * svc_fill_write_vector - Construct data argument for VFS write call
1638  * @rqstp: svc_rqst to operate on
1639  * @pages: list of pages containing data payload
1640  * @first: buffer containing first section of write payload
1641  * @total: total number of bytes of write payload
1642  *
1643  * Fills in rqstp::rq_vec, and returns the number of elements.
1644  */
1645 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, struct page **pages,
1646 				   struct kvec *first, size_t total)
1647 {
1648 	struct kvec *vec = rqstp->rq_vec;
1649 	unsigned int i;
1650 
1651 	/* Some types of transport can present the write payload
1652 	 * entirely in rq_arg.pages. In this case, @first is empty.
1653 	 */
1654 	i = 0;
1655 	if (first->iov_len) {
1656 		vec[i].iov_base = first->iov_base;
1657 		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1658 		total -= vec[i].iov_len;
1659 		++i;
1660 	}
1661 
1662 	while (total) {
1663 		vec[i].iov_base = page_address(*pages);
1664 		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1665 		total -= vec[i].iov_len;
1666 		++i;
1667 		++pages;
1668 	}
1669 
1670 	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1671 	return i;
1672 }
1673 EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1674 
1675 /**
1676  * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1677  * @rqstp: svc_rqst to operate on
1678  * @first: buffer containing first section of pathname
1679  * @p: buffer containing remaining section of pathname
1680  * @total: total length of the pathname argument
1681  *
1682  * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1683  * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1684  * the returned string.
1685  */
1686 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1687 				void *p, size_t total)
1688 {
1689 	size_t len, remaining;
1690 	char *result, *dst;
1691 
1692 	result = kmalloc(total + 1, GFP_KERNEL);
1693 	if (!result)
1694 		return ERR_PTR(-ESERVERFAULT);
1695 
1696 	dst = result;
1697 	remaining = total;
1698 
1699 	len = min_t(size_t, total, first->iov_len);
1700 	if (len) {
1701 		memcpy(dst, first->iov_base, len);
1702 		dst += len;
1703 		remaining -= len;
1704 	}
1705 
1706 	if (remaining) {
1707 		len = min_t(size_t, remaining, PAGE_SIZE);
1708 		memcpy(dst, p, len);
1709 		dst += len;
1710 	}
1711 
1712 	*dst = '\0';
1713 
1714 	/* Sanity check: Linux doesn't allow the pathname argument to
1715 	 * contain a NUL byte.
1716 	 */
1717 	if (strlen(result) != total) {
1718 		kfree(result);
1719 		return ERR_PTR(-EINVAL);
1720 	}
1721 	return result;
1722 }
1723 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1724