xref: /linux/net/sunrpc/svc.c (revision bec36eca6f5d1d83a9c3733fc40ba173ad849df2)
1 /*
2  * linux/net/sunrpc/svc.c
3  *
4  * High-level RPC service routines
5  *
6  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7  *
8  * Multiple threads pools and NUMAisation
9  * Copyright (c) 2006 Silicon Graphics, Inc.
10  * by Greg Banks <gnb@melbourne.sgi.com>
11  */
12 
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
17 #include <linux/in.h>
18 #include <linux/mm.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 
23 #include <linux/sunrpc/types.h>
24 #include <linux/sunrpc/xdr.h>
25 #include <linux/sunrpc/stats.h>
26 #include <linux/sunrpc/svcsock.h>
27 #include <linux/sunrpc/clnt.h>
28 
29 #define RPCDBG_FACILITY	RPCDBG_SVCDSP
30 
31 static void svc_unregister(const struct svc_serv *serv);
32 
33 #define svc_serv_is_pooled(serv)    ((serv)->sv_function)
34 
35 /*
36  * Mode for mapping cpus to pools.
37  */
38 enum {
39 	SVC_POOL_AUTO = -1,	/* choose one of the others */
40 	SVC_POOL_GLOBAL,	/* no mapping, just a single global pool
41 				 * (legacy & UP mode) */
42 	SVC_POOL_PERCPU,	/* one pool per cpu */
43 	SVC_POOL_PERNODE	/* one pool per numa node */
44 };
45 #define SVC_POOL_DEFAULT	SVC_POOL_GLOBAL
46 
47 /*
48  * Structure for mapping cpus to pools and vice versa.
49  * Setup once during sunrpc initialisation.
50  */
51 static struct svc_pool_map {
52 	int count;			/* How many svc_servs use us */
53 	int mode;			/* Note: int not enum to avoid
54 					 * warnings about "enumeration value
55 					 * not handled in switch" */
56 	unsigned int npools;
57 	unsigned int *pool_to;		/* maps pool id to cpu or node */
58 	unsigned int *to_pool;		/* maps cpu or node to pool id */
59 } svc_pool_map = {
60 	.count = 0,
61 	.mode = SVC_POOL_DEFAULT
62 };
63 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
64 
65 static int
66 param_set_pool_mode(const char *val, struct kernel_param *kp)
67 {
68 	int *ip = (int *)kp->arg;
69 	struct svc_pool_map *m = &svc_pool_map;
70 	int err;
71 
72 	mutex_lock(&svc_pool_map_mutex);
73 
74 	err = -EBUSY;
75 	if (m->count)
76 		goto out;
77 
78 	err = 0;
79 	if (!strncmp(val, "auto", 4))
80 		*ip = SVC_POOL_AUTO;
81 	else if (!strncmp(val, "global", 6))
82 		*ip = SVC_POOL_GLOBAL;
83 	else if (!strncmp(val, "percpu", 6))
84 		*ip = SVC_POOL_PERCPU;
85 	else if (!strncmp(val, "pernode", 7))
86 		*ip = SVC_POOL_PERNODE;
87 	else
88 		err = -EINVAL;
89 
90 out:
91 	mutex_unlock(&svc_pool_map_mutex);
92 	return err;
93 }
94 
95 static int
96 param_get_pool_mode(char *buf, struct kernel_param *kp)
97 {
98 	int *ip = (int *)kp->arg;
99 
100 	switch (*ip)
101 	{
102 	case SVC_POOL_AUTO:
103 		return strlcpy(buf, "auto", 20);
104 	case SVC_POOL_GLOBAL:
105 		return strlcpy(buf, "global", 20);
106 	case SVC_POOL_PERCPU:
107 		return strlcpy(buf, "percpu", 20);
108 	case SVC_POOL_PERNODE:
109 		return strlcpy(buf, "pernode", 20);
110 	default:
111 		return sprintf(buf, "%d", *ip);
112 	}
113 }
114 
115 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
116 		 &svc_pool_map.mode, 0644);
117 
118 /*
119  * Detect best pool mapping mode heuristically,
120  * according to the machine's topology.
121  */
122 static int
123 svc_pool_map_choose_mode(void)
124 {
125 	unsigned int node;
126 
127 	if (num_online_nodes() > 1) {
128 		/*
129 		 * Actually have multiple NUMA nodes,
130 		 * so split pools on NUMA node boundaries
131 		 */
132 		return SVC_POOL_PERNODE;
133 	}
134 
135 	node = any_online_node(node_online_map);
136 	if (nr_cpus_node(node) > 2) {
137 		/*
138 		 * Non-trivial SMP, or CONFIG_NUMA on
139 		 * non-NUMA hardware, e.g. with a generic
140 		 * x86_64 kernel on Xeons.  In this case we
141 		 * want to divide the pools on cpu boundaries.
142 		 */
143 		return SVC_POOL_PERCPU;
144 	}
145 
146 	/* default: one global pool */
147 	return SVC_POOL_GLOBAL;
148 }
149 
150 /*
151  * Allocate the to_pool[] and pool_to[] arrays.
152  * Returns 0 on success or an errno.
153  */
154 static int
155 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
156 {
157 	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
158 	if (!m->to_pool)
159 		goto fail;
160 	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
161 	if (!m->pool_to)
162 		goto fail_free;
163 
164 	return 0;
165 
166 fail_free:
167 	kfree(m->to_pool);
168 fail:
169 	return -ENOMEM;
170 }
171 
172 /*
173  * Initialise the pool map for SVC_POOL_PERCPU mode.
174  * Returns number of pools or <0 on error.
175  */
176 static int
177 svc_pool_map_init_percpu(struct svc_pool_map *m)
178 {
179 	unsigned int maxpools = nr_cpu_ids;
180 	unsigned int pidx = 0;
181 	unsigned int cpu;
182 	int err;
183 
184 	err = svc_pool_map_alloc_arrays(m, maxpools);
185 	if (err)
186 		return err;
187 
188 	for_each_online_cpu(cpu) {
189 		BUG_ON(pidx > maxpools);
190 		m->to_pool[cpu] = pidx;
191 		m->pool_to[pidx] = cpu;
192 		pidx++;
193 	}
194 	/* cpus brought online later all get mapped to pool0, sorry */
195 
196 	return pidx;
197 };
198 
199 
200 /*
201  * Initialise the pool map for SVC_POOL_PERNODE mode.
202  * Returns number of pools or <0 on error.
203  */
204 static int
205 svc_pool_map_init_pernode(struct svc_pool_map *m)
206 {
207 	unsigned int maxpools = nr_node_ids;
208 	unsigned int pidx = 0;
209 	unsigned int node;
210 	int err;
211 
212 	err = svc_pool_map_alloc_arrays(m, maxpools);
213 	if (err)
214 		return err;
215 
216 	for_each_node_with_cpus(node) {
217 		/* some architectures (e.g. SN2) have cpuless nodes */
218 		BUG_ON(pidx > maxpools);
219 		m->to_pool[node] = pidx;
220 		m->pool_to[pidx] = node;
221 		pidx++;
222 	}
223 	/* nodes brought online later all get mapped to pool0, sorry */
224 
225 	return pidx;
226 }
227 
228 
229 /*
230  * Add a reference to the global map of cpus to pools (and
231  * vice versa).  Initialise the map if we're the first user.
232  * Returns the number of pools.
233  */
234 static unsigned int
235 svc_pool_map_get(void)
236 {
237 	struct svc_pool_map *m = &svc_pool_map;
238 	int npools = -1;
239 
240 	mutex_lock(&svc_pool_map_mutex);
241 
242 	if (m->count++) {
243 		mutex_unlock(&svc_pool_map_mutex);
244 		return m->npools;
245 	}
246 
247 	if (m->mode == SVC_POOL_AUTO)
248 		m->mode = svc_pool_map_choose_mode();
249 
250 	switch (m->mode) {
251 	case SVC_POOL_PERCPU:
252 		npools = svc_pool_map_init_percpu(m);
253 		break;
254 	case SVC_POOL_PERNODE:
255 		npools = svc_pool_map_init_pernode(m);
256 		break;
257 	}
258 
259 	if (npools < 0) {
260 		/* default, or memory allocation failure */
261 		npools = 1;
262 		m->mode = SVC_POOL_GLOBAL;
263 	}
264 	m->npools = npools;
265 
266 	mutex_unlock(&svc_pool_map_mutex);
267 	return m->npools;
268 }
269 
270 
271 /*
272  * Drop a reference to the global map of cpus to pools.
273  * When the last reference is dropped, the map data is
274  * freed; this allows the sysadmin to change the pool
275  * mode using the pool_mode module option without
276  * rebooting or re-loading sunrpc.ko.
277  */
278 static void
279 svc_pool_map_put(void)
280 {
281 	struct svc_pool_map *m = &svc_pool_map;
282 
283 	mutex_lock(&svc_pool_map_mutex);
284 
285 	if (!--m->count) {
286 		m->mode = SVC_POOL_DEFAULT;
287 		kfree(m->to_pool);
288 		kfree(m->pool_to);
289 		m->npools = 0;
290 	}
291 
292 	mutex_unlock(&svc_pool_map_mutex);
293 }
294 
295 
296 /*
297  * Set the given thread's cpus_allowed mask so that it
298  * will only run on cpus in the given pool.
299  */
300 static inline void
301 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
302 {
303 	struct svc_pool_map *m = &svc_pool_map;
304 	unsigned int node = m->pool_to[pidx];
305 
306 	/*
307 	 * The caller checks for sv_nrpools > 1, which
308 	 * implies that we've been initialized.
309 	 */
310 	BUG_ON(m->count == 0);
311 
312 	switch (m->mode) {
313 	case SVC_POOL_PERCPU:
314 	{
315 		set_cpus_allowed_ptr(task, cpumask_of(node));
316 		break;
317 	}
318 	case SVC_POOL_PERNODE:
319 	{
320 		set_cpus_allowed_ptr(task, cpumask_of_node(node));
321 		break;
322 	}
323 	}
324 }
325 
326 /*
327  * Use the mapping mode to choose a pool for a given CPU.
328  * Used when enqueueing an incoming RPC.  Always returns
329  * a non-NULL pool pointer.
330  */
331 struct svc_pool *
332 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
333 {
334 	struct svc_pool_map *m = &svc_pool_map;
335 	unsigned int pidx = 0;
336 
337 	/*
338 	 * An uninitialised map happens in a pure client when
339 	 * lockd is brought up, so silently treat it the
340 	 * same as SVC_POOL_GLOBAL.
341 	 */
342 	if (svc_serv_is_pooled(serv)) {
343 		switch (m->mode) {
344 		case SVC_POOL_PERCPU:
345 			pidx = m->to_pool[cpu];
346 			break;
347 		case SVC_POOL_PERNODE:
348 			pidx = m->to_pool[cpu_to_node(cpu)];
349 			break;
350 		}
351 	}
352 	return &serv->sv_pools[pidx % serv->sv_nrpools];
353 }
354 
355 
356 /*
357  * Create an RPC service
358  */
359 static struct svc_serv *
360 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
361 	     void (*shutdown)(struct svc_serv *serv))
362 {
363 	struct svc_serv	*serv;
364 	unsigned int vers;
365 	unsigned int xdrsize;
366 	unsigned int i;
367 
368 	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
369 		return NULL;
370 	serv->sv_name      = prog->pg_name;
371 	serv->sv_program   = prog;
372 	serv->sv_nrthreads = 1;
373 	serv->sv_stats     = prog->pg_stats;
374 	if (bufsize > RPCSVC_MAXPAYLOAD)
375 		bufsize = RPCSVC_MAXPAYLOAD;
376 	serv->sv_max_payload = bufsize? bufsize : 4096;
377 	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
378 	serv->sv_shutdown  = shutdown;
379 	xdrsize = 0;
380 	while (prog) {
381 		prog->pg_lovers = prog->pg_nvers-1;
382 		for (vers=0; vers<prog->pg_nvers ; vers++)
383 			if (prog->pg_vers[vers]) {
384 				prog->pg_hivers = vers;
385 				if (prog->pg_lovers > vers)
386 					prog->pg_lovers = vers;
387 				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
388 					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
389 			}
390 		prog = prog->pg_next;
391 	}
392 	serv->sv_xdrsize   = xdrsize;
393 	INIT_LIST_HEAD(&serv->sv_tempsocks);
394 	INIT_LIST_HEAD(&serv->sv_permsocks);
395 	init_timer(&serv->sv_temptimer);
396 	spin_lock_init(&serv->sv_lock);
397 
398 	serv->sv_nrpools = npools;
399 	serv->sv_pools =
400 		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
401 			GFP_KERNEL);
402 	if (!serv->sv_pools) {
403 		kfree(serv);
404 		return NULL;
405 	}
406 
407 	for (i = 0; i < serv->sv_nrpools; i++) {
408 		struct svc_pool *pool = &serv->sv_pools[i];
409 
410 		dprintk("svc: initialising pool %u for %s\n",
411 				i, serv->sv_name);
412 
413 		pool->sp_id = i;
414 		INIT_LIST_HEAD(&pool->sp_threads);
415 		INIT_LIST_HEAD(&pool->sp_sockets);
416 		INIT_LIST_HEAD(&pool->sp_all_threads);
417 		spin_lock_init(&pool->sp_lock);
418 	}
419 
420 	/* Remove any stale portmap registrations */
421 	svc_unregister(serv);
422 
423 	return serv;
424 }
425 
426 struct svc_serv *
427 svc_create(struct svc_program *prog, unsigned int bufsize,
428 	   void (*shutdown)(struct svc_serv *serv))
429 {
430 	return __svc_create(prog, bufsize, /*npools*/1, shutdown);
431 }
432 EXPORT_SYMBOL_GPL(svc_create);
433 
434 struct svc_serv *
435 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
436 		  void (*shutdown)(struct svc_serv *serv),
437 		  svc_thread_fn func, struct module *mod)
438 {
439 	struct svc_serv *serv;
440 	unsigned int npools = svc_pool_map_get();
441 
442 	serv = __svc_create(prog, bufsize, npools, shutdown);
443 
444 	if (serv != NULL) {
445 		serv->sv_function = func;
446 		serv->sv_module = mod;
447 	}
448 
449 	return serv;
450 }
451 EXPORT_SYMBOL_GPL(svc_create_pooled);
452 
453 /*
454  * Destroy an RPC service. Should be called with appropriate locking to
455  * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
456  */
457 void
458 svc_destroy(struct svc_serv *serv)
459 {
460 	dprintk("svc: svc_destroy(%s, %d)\n",
461 				serv->sv_program->pg_name,
462 				serv->sv_nrthreads);
463 
464 	if (serv->sv_nrthreads) {
465 		if (--(serv->sv_nrthreads) != 0) {
466 			svc_sock_update_bufs(serv);
467 			return;
468 		}
469 	} else
470 		printk("svc_destroy: no threads for serv=%p!\n", serv);
471 
472 	del_timer_sync(&serv->sv_temptimer);
473 
474 	svc_close_all(&serv->sv_tempsocks);
475 
476 	if (serv->sv_shutdown)
477 		serv->sv_shutdown(serv);
478 
479 	svc_close_all(&serv->sv_permsocks);
480 
481 	BUG_ON(!list_empty(&serv->sv_permsocks));
482 	BUG_ON(!list_empty(&serv->sv_tempsocks));
483 
484 	cache_clean_deferred(serv);
485 
486 	if (svc_serv_is_pooled(serv))
487 		svc_pool_map_put();
488 
489 	svc_unregister(serv);
490 	kfree(serv->sv_pools);
491 	kfree(serv);
492 }
493 EXPORT_SYMBOL_GPL(svc_destroy);
494 
495 /*
496  * Allocate an RPC server's buffer space.
497  * We allocate pages and place them in rq_argpages.
498  */
499 static int
500 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
501 {
502 	unsigned int pages, arghi;
503 
504 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
505 				       * We assume one is at most one page
506 				       */
507 	arghi = 0;
508 	BUG_ON(pages > RPCSVC_MAXPAGES);
509 	while (pages) {
510 		struct page *p = alloc_page(GFP_KERNEL);
511 		if (!p)
512 			break;
513 		rqstp->rq_pages[arghi++] = p;
514 		pages--;
515 	}
516 	return pages == 0;
517 }
518 
519 /*
520  * Release an RPC server buffer
521  */
522 static void
523 svc_release_buffer(struct svc_rqst *rqstp)
524 {
525 	unsigned int i;
526 
527 	for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
528 		if (rqstp->rq_pages[i])
529 			put_page(rqstp->rq_pages[i]);
530 }
531 
532 struct svc_rqst *
533 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
534 {
535 	struct svc_rqst	*rqstp;
536 
537 	rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
538 	if (!rqstp)
539 		goto out_enomem;
540 
541 	init_waitqueue_head(&rqstp->rq_wait);
542 
543 	serv->sv_nrthreads++;
544 	spin_lock_bh(&pool->sp_lock);
545 	pool->sp_nrthreads++;
546 	list_add(&rqstp->rq_all, &pool->sp_all_threads);
547 	spin_unlock_bh(&pool->sp_lock);
548 	rqstp->rq_server = serv;
549 	rqstp->rq_pool = pool;
550 
551 	rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
552 	if (!rqstp->rq_argp)
553 		goto out_thread;
554 
555 	rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
556 	if (!rqstp->rq_resp)
557 		goto out_thread;
558 
559 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
560 		goto out_thread;
561 
562 	return rqstp;
563 out_thread:
564 	svc_exit_thread(rqstp);
565 out_enomem:
566 	return ERR_PTR(-ENOMEM);
567 }
568 EXPORT_SYMBOL_GPL(svc_prepare_thread);
569 
570 /*
571  * Choose a pool in which to create a new thread, for svc_set_num_threads
572  */
573 static inline struct svc_pool *
574 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
575 {
576 	if (pool != NULL)
577 		return pool;
578 
579 	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
580 }
581 
582 /*
583  * Choose a thread to kill, for svc_set_num_threads
584  */
585 static inline struct task_struct *
586 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
587 {
588 	unsigned int i;
589 	struct task_struct *task = NULL;
590 
591 	if (pool != NULL) {
592 		spin_lock_bh(&pool->sp_lock);
593 	} else {
594 		/* choose a pool in round-robin fashion */
595 		for (i = 0; i < serv->sv_nrpools; i++) {
596 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
597 			spin_lock_bh(&pool->sp_lock);
598 			if (!list_empty(&pool->sp_all_threads))
599 				goto found_pool;
600 			spin_unlock_bh(&pool->sp_lock);
601 		}
602 		return NULL;
603 	}
604 
605 found_pool:
606 	if (!list_empty(&pool->sp_all_threads)) {
607 		struct svc_rqst *rqstp;
608 
609 		/*
610 		 * Remove from the pool->sp_all_threads list
611 		 * so we don't try to kill it again.
612 		 */
613 		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
614 		list_del_init(&rqstp->rq_all);
615 		task = rqstp->rq_task;
616 	}
617 	spin_unlock_bh(&pool->sp_lock);
618 
619 	return task;
620 }
621 
622 /*
623  * Create or destroy enough new threads to make the number
624  * of threads the given number.  If `pool' is non-NULL, applies
625  * only to threads in that pool, otherwise round-robins between
626  * all pools.  Must be called with a svc_get() reference and
627  * the BKL or another lock to protect access to svc_serv fields.
628  *
629  * Destroying threads relies on the service threads filling in
630  * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
631  * has been created using svc_create_pooled().
632  *
633  * Based on code that used to be in nfsd_svc() but tweaked
634  * to be pool-aware.
635  */
636 int
637 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
638 {
639 	struct svc_rqst	*rqstp;
640 	struct task_struct *task;
641 	struct svc_pool *chosen_pool;
642 	int error = 0;
643 	unsigned int state = serv->sv_nrthreads-1;
644 
645 	if (pool == NULL) {
646 		/* The -1 assumes caller has done a svc_get() */
647 		nrservs -= (serv->sv_nrthreads-1);
648 	} else {
649 		spin_lock_bh(&pool->sp_lock);
650 		nrservs -= pool->sp_nrthreads;
651 		spin_unlock_bh(&pool->sp_lock);
652 	}
653 
654 	/* create new threads */
655 	while (nrservs > 0) {
656 		nrservs--;
657 		chosen_pool = choose_pool(serv, pool, &state);
658 
659 		rqstp = svc_prepare_thread(serv, chosen_pool);
660 		if (IS_ERR(rqstp)) {
661 			error = PTR_ERR(rqstp);
662 			break;
663 		}
664 
665 		__module_get(serv->sv_module);
666 		task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
667 		if (IS_ERR(task)) {
668 			error = PTR_ERR(task);
669 			module_put(serv->sv_module);
670 			svc_exit_thread(rqstp);
671 			break;
672 		}
673 
674 		rqstp->rq_task = task;
675 		if (serv->sv_nrpools > 1)
676 			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
677 
678 		svc_sock_update_bufs(serv);
679 		wake_up_process(task);
680 	}
681 	/* destroy old threads */
682 	while (nrservs < 0 &&
683 	       (task = choose_victim(serv, pool, &state)) != NULL) {
684 		send_sig(SIGINT, task, 1);
685 		nrservs++;
686 	}
687 
688 	return error;
689 }
690 EXPORT_SYMBOL_GPL(svc_set_num_threads);
691 
692 /*
693  * Called from a server thread as it's exiting. Caller must hold the BKL or
694  * the "service mutex", whichever is appropriate for the service.
695  */
696 void
697 svc_exit_thread(struct svc_rqst *rqstp)
698 {
699 	struct svc_serv	*serv = rqstp->rq_server;
700 	struct svc_pool	*pool = rqstp->rq_pool;
701 
702 	svc_release_buffer(rqstp);
703 	kfree(rqstp->rq_resp);
704 	kfree(rqstp->rq_argp);
705 	kfree(rqstp->rq_auth_data);
706 
707 	spin_lock_bh(&pool->sp_lock);
708 	pool->sp_nrthreads--;
709 	list_del(&rqstp->rq_all);
710 	spin_unlock_bh(&pool->sp_lock);
711 
712 	kfree(rqstp);
713 
714 	/* Release the server */
715 	if (serv)
716 		svc_destroy(serv);
717 }
718 EXPORT_SYMBOL_GPL(svc_exit_thread);
719 
720 /*
721  * Register an "inet" protocol family netid with the local
722  * rpcbind daemon via an rpcbind v4 SET request.
723  *
724  * No netconfig infrastructure is available in the kernel, so
725  * we map IP_ protocol numbers to netids by hand.
726  *
727  * Returns zero on success; a negative errno value is returned
728  * if any error occurs.
729  */
730 static int __svc_rpcb_register4(const u32 program, const u32 version,
731 				const unsigned short protocol,
732 				const unsigned short port)
733 {
734 	const struct sockaddr_in sin = {
735 		.sin_family		= AF_INET,
736 		.sin_addr.s_addr	= htonl(INADDR_ANY),
737 		.sin_port		= htons(port),
738 	};
739 	const char *netid;
740 	int error;
741 
742 	switch (protocol) {
743 	case IPPROTO_UDP:
744 		netid = RPCBIND_NETID_UDP;
745 		break;
746 	case IPPROTO_TCP:
747 		netid = RPCBIND_NETID_TCP;
748 		break;
749 	default:
750 		return -ENOPROTOOPT;
751 	}
752 
753 	error = rpcb_v4_register(program, version,
754 					(const struct sockaddr *)&sin, netid);
755 
756 	/*
757 	 * User space didn't support rpcbind v4, so retry this
758 	 * registration request with the legacy rpcbind v2 protocol.
759 	 */
760 	if (error == -EPROTONOSUPPORT)
761 		error = rpcb_register(program, version, protocol, port);
762 
763 	return error;
764 }
765 
766 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
767 /*
768  * Register an "inet6" protocol family netid with the local
769  * rpcbind daemon via an rpcbind v4 SET request.
770  *
771  * No netconfig infrastructure is available in the kernel, so
772  * we map IP_ protocol numbers to netids by hand.
773  *
774  * Returns zero on success; a negative errno value is returned
775  * if any error occurs.
776  */
777 static int __svc_rpcb_register6(const u32 program, const u32 version,
778 				const unsigned short protocol,
779 				const unsigned short port)
780 {
781 	const struct sockaddr_in6 sin6 = {
782 		.sin6_family		= AF_INET6,
783 		.sin6_addr		= IN6ADDR_ANY_INIT,
784 		.sin6_port		= htons(port),
785 	};
786 	const char *netid;
787 	int error;
788 
789 	switch (protocol) {
790 	case IPPROTO_UDP:
791 		netid = RPCBIND_NETID_UDP6;
792 		break;
793 	case IPPROTO_TCP:
794 		netid = RPCBIND_NETID_TCP6;
795 		break;
796 	default:
797 		return -ENOPROTOOPT;
798 	}
799 
800 	error = rpcb_v4_register(program, version,
801 					(const struct sockaddr *)&sin6, netid);
802 
803 	/*
804 	 * User space didn't support rpcbind version 4, so we won't
805 	 * use a PF_INET6 listener.
806 	 */
807 	if (error == -EPROTONOSUPPORT)
808 		error = -EAFNOSUPPORT;
809 
810 	return error;
811 }
812 #endif	/* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
813 
814 /*
815  * Register a kernel RPC service via rpcbind version 4.
816  *
817  * Returns zero on success; a negative errno value is returned
818  * if any error occurs.
819  */
820 static int __svc_register(const char *progname,
821 			  const u32 program, const u32 version,
822 			  const int family,
823 			  const unsigned short protocol,
824 			  const unsigned short port)
825 {
826 	int error = -EAFNOSUPPORT;
827 
828 	switch (family) {
829 	case PF_INET:
830 		error = __svc_rpcb_register4(program, version,
831 						protocol, port);
832 		break;
833 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
834 	case PF_INET6:
835 		error = __svc_rpcb_register6(program, version,
836 						protocol, port);
837 #endif	/* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
838 	}
839 
840 	if (error < 0)
841 		printk(KERN_WARNING "svc: failed to register %sv%u RPC "
842 			"service (errno %d).\n", progname, version, -error);
843 	return error;
844 }
845 
846 /**
847  * svc_register - register an RPC service with the local portmapper
848  * @serv: svc_serv struct for the service to register
849  * @family: protocol family of service's listener socket
850  * @proto: transport protocol number to advertise
851  * @port: port to advertise
852  *
853  * Service is registered for any address in the passed-in protocol family
854  */
855 int svc_register(const struct svc_serv *serv, const int family,
856 		 const unsigned short proto, const unsigned short port)
857 {
858 	struct svc_program	*progp;
859 	unsigned int		i;
860 	int			error = 0;
861 
862 	BUG_ON(proto == 0 && port == 0);
863 
864 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
865 		for (i = 0; i < progp->pg_nvers; i++) {
866 			if (progp->pg_vers[i] == NULL)
867 				continue;
868 
869 			dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
870 					progp->pg_name,
871 					i,
872 					proto == IPPROTO_UDP?  "udp" : "tcp",
873 					port,
874 					family,
875 					progp->pg_vers[i]->vs_hidden?
876 						" (but not telling portmap)" : "");
877 
878 			if (progp->pg_vers[i]->vs_hidden)
879 				continue;
880 
881 			error = __svc_register(progp->pg_name, progp->pg_prog,
882 						i, family, proto, port);
883 			if (error < 0)
884 				break;
885 		}
886 	}
887 
888 	return error;
889 }
890 
891 /*
892  * If user space is running rpcbind, it should take the v4 UNSET
893  * and clear everything for this [program, version].  If user space
894  * is running portmap, it will reject the v4 UNSET, but won't have
895  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
896  * in this case to clear all existing entries for [program, version].
897  */
898 static void __svc_unregister(const u32 program, const u32 version,
899 			     const char *progname)
900 {
901 	int error;
902 
903 	error = rpcb_v4_register(program, version, NULL, "");
904 
905 	/*
906 	 * User space didn't support rpcbind v4, so retry this
907 	 * request with the legacy rpcbind v2 protocol.
908 	 */
909 	if (error == -EPROTONOSUPPORT)
910 		error = rpcb_register(program, version, 0, 0);
911 
912 	dprintk("svc: %s(%sv%u), error %d\n",
913 			__func__, progname, version, error);
914 }
915 
916 /*
917  * All netids, bind addresses and ports registered for [program, version]
918  * are removed from the local rpcbind database (if the service is not
919  * hidden) to make way for a new instance of the service.
920  *
921  * The result of unregistration is reported via dprintk for those who want
922  * verification of the result, but is otherwise not important.
923  */
924 static void svc_unregister(const struct svc_serv *serv)
925 {
926 	struct svc_program *progp;
927 	unsigned long flags;
928 	unsigned int i;
929 
930 	clear_thread_flag(TIF_SIGPENDING);
931 
932 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
933 		for (i = 0; i < progp->pg_nvers; i++) {
934 			if (progp->pg_vers[i] == NULL)
935 				continue;
936 			if (progp->pg_vers[i]->vs_hidden)
937 				continue;
938 
939 			__svc_unregister(progp->pg_prog, i, progp->pg_name);
940 		}
941 	}
942 
943 	spin_lock_irqsave(&current->sighand->siglock, flags);
944 	recalc_sigpending();
945 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
946 }
947 
948 /*
949  * Printk the given error with the address of the client that caused it.
950  */
951 static int
952 __attribute__ ((format (printf, 2, 3)))
953 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
954 {
955 	va_list args;
956 	int 	r;
957 	char 	buf[RPC_MAX_ADDRBUFLEN];
958 
959 	if (!net_ratelimit())
960 		return 0;
961 
962 	printk(KERN_WARNING "svc: %s: ",
963 		svc_print_addr(rqstp, buf, sizeof(buf)));
964 
965 	va_start(args, fmt);
966 	r = vprintk(fmt, args);
967 	va_end(args);
968 
969 	return r;
970 }
971 
972 /*
973  * Process the RPC request.
974  */
975 int
976 svc_process(struct svc_rqst *rqstp)
977 {
978 	struct svc_program	*progp;
979 	struct svc_version	*versp = NULL;	/* compiler food */
980 	struct svc_procedure	*procp = NULL;
981 	struct kvec *		argv = &rqstp->rq_arg.head[0];
982 	struct kvec *		resv = &rqstp->rq_res.head[0];
983 	struct svc_serv		*serv = rqstp->rq_server;
984 	kxdrproc_t		xdr;
985 	__be32			*statp;
986 	u32			dir, prog, vers, proc;
987 	__be32			auth_stat, rpc_stat;
988 	int			auth_res;
989 	__be32			*reply_statp;
990 
991 	rpc_stat = rpc_success;
992 
993 	if (argv->iov_len < 6*4)
994 		goto err_short_len;
995 
996 	/* setup response xdr_buf.
997 	 * Initially it has just one page
998 	 */
999 	rqstp->rq_resused = 1;
1000 	resv->iov_base = page_address(rqstp->rq_respages[0]);
1001 	resv->iov_len = 0;
1002 	rqstp->rq_res.pages = rqstp->rq_respages + 1;
1003 	rqstp->rq_res.len = 0;
1004 	rqstp->rq_res.page_base = 0;
1005 	rqstp->rq_res.page_len = 0;
1006 	rqstp->rq_res.buflen = PAGE_SIZE;
1007 	rqstp->rq_res.tail[0].iov_base = NULL;
1008 	rqstp->rq_res.tail[0].iov_len = 0;
1009 	/* Will be turned off only in gss privacy case: */
1010 	rqstp->rq_splice_ok = 1;
1011 	/* Will be turned off only when NFSv4 Sessions are used */
1012 	rqstp->rq_usedeferral = 1;
1013 
1014 	/* Setup reply header */
1015 	rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1016 
1017 	rqstp->rq_xid = svc_getu32(argv);
1018 	svc_putu32(resv, rqstp->rq_xid);
1019 
1020 	dir  = svc_getnl(argv);
1021 	vers = svc_getnl(argv);
1022 
1023 	/* First words of reply: */
1024 	svc_putnl(resv, 1);		/* REPLY */
1025 
1026 	if (dir != 0)		/* direction != CALL */
1027 		goto err_bad_dir;
1028 	if (vers != 2)		/* RPC version number */
1029 		goto err_bad_rpc;
1030 
1031 	/* Save position in case we later decide to reject: */
1032 	reply_statp = resv->iov_base + resv->iov_len;
1033 
1034 	svc_putnl(resv, 0);		/* ACCEPT */
1035 
1036 	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
1037 	rqstp->rq_vers = vers = svc_getnl(argv);	/* version number */
1038 	rqstp->rq_proc = proc = svc_getnl(argv);	/* procedure number */
1039 
1040 	progp = serv->sv_program;
1041 
1042 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1043 		if (prog == progp->pg_prog)
1044 			break;
1045 
1046 	/*
1047 	 * Decode auth data, and add verifier to reply buffer.
1048 	 * We do this before anything else in order to get a decent
1049 	 * auth verifier.
1050 	 */
1051 	auth_res = svc_authenticate(rqstp, &auth_stat);
1052 	/* Also give the program a chance to reject this call: */
1053 	if (auth_res == SVC_OK && progp) {
1054 		auth_stat = rpc_autherr_badcred;
1055 		auth_res = progp->pg_authenticate(rqstp);
1056 	}
1057 	switch (auth_res) {
1058 	case SVC_OK:
1059 		break;
1060 	case SVC_GARBAGE:
1061 		goto err_garbage;
1062 	case SVC_SYSERR:
1063 		rpc_stat = rpc_system_err;
1064 		goto err_bad;
1065 	case SVC_DENIED:
1066 		goto err_bad_auth;
1067 	case SVC_DROP:
1068 		goto dropit;
1069 	case SVC_COMPLETE:
1070 		goto sendit;
1071 	}
1072 
1073 	if (progp == NULL)
1074 		goto err_bad_prog;
1075 
1076 	if (vers >= progp->pg_nvers ||
1077 	  !(versp = progp->pg_vers[vers]))
1078 		goto err_bad_vers;
1079 
1080 	procp = versp->vs_proc + proc;
1081 	if (proc >= versp->vs_nproc || !procp->pc_func)
1082 		goto err_bad_proc;
1083 	rqstp->rq_procinfo = procp;
1084 
1085 	/* Syntactic check complete */
1086 	serv->sv_stats->rpccnt++;
1087 
1088 	/* Build the reply header. */
1089 	statp = resv->iov_base +resv->iov_len;
1090 	svc_putnl(resv, RPC_SUCCESS);
1091 
1092 	/* Bump per-procedure stats counter */
1093 	procp->pc_count++;
1094 
1095 	/* Initialize storage for argp and resp */
1096 	memset(rqstp->rq_argp, 0, procp->pc_argsize);
1097 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1098 
1099 	/* un-reserve some of the out-queue now that we have a
1100 	 * better idea of reply size
1101 	 */
1102 	if (procp->pc_xdrressize)
1103 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1104 
1105 	/* Call the function that processes the request. */
1106 	if (!versp->vs_dispatch) {
1107 		/* Decode arguments */
1108 		xdr = procp->pc_decode;
1109 		if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1110 			goto err_garbage;
1111 
1112 		*statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1113 
1114 		/* Encode reply */
1115 		if (*statp == rpc_drop_reply) {
1116 			if (procp->pc_release)
1117 				procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1118 			goto dropit;
1119 		}
1120 		if (*statp == rpc_success && (xdr = procp->pc_encode)
1121 		 && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1122 			dprintk("svc: failed to encode reply\n");
1123 			/* serv->sv_stats->rpcsystemerr++; */
1124 			*statp = rpc_system_err;
1125 		}
1126 	} else {
1127 		dprintk("svc: calling dispatcher\n");
1128 		if (!versp->vs_dispatch(rqstp, statp)) {
1129 			/* Release reply info */
1130 			if (procp->pc_release)
1131 				procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1132 			goto dropit;
1133 		}
1134 	}
1135 
1136 	/* Check RPC status result */
1137 	if (*statp != rpc_success)
1138 		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1139 
1140 	/* Release reply info */
1141 	if (procp->pc_release)
1142 		procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1143 
1144 	if (procp->pc_encode == NULL)
1145 		goto dropit;
1146 
1147  sendit:
1148 	if (svc_authorise(rqstp))
1149 		goto dropit;
1150 	return svc_send(rqstp);
1151 
1152  dropit:
1153 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1154 	dprintk("svc: svc_process dropit\n");
1155 	svc_drop(rqstp);
1156 	return 0;
1157 
1158 err_short_len:
1159 	svc_printk(rqstp, "short len %Zd, dropping request\n",
1160 			argv->iov_len);
1161 
1162 	goto dropit;			/* drop request */
1163 
1164 err_bad_dir:
1165 	svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1166 
1167 	serv->sv_stats->rpcbadfmt++;
1168 	goto dropit;			/* drop request */
1169 
1170 err_bad_rpc:
1171 	serv->sv_stats->rpcbadfmt++;
1172 	svc_putnl(resv, 1);	/* REJECT */
1173 	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1174 	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1175 	svc_putnl(resv, 2);
1176 	goto sendit;
1177 
1178 err_bad_auth:
1179 	dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1180 	serv->sv_stats->rpcbadauth++;
1181 	/* Restore write pointer to location of accept status: */
1182 	xdr_ressize_check(rqstp, reply_statp);
1183 	svc_putnl(resv, 1);	/* REJECT */
1184 	svc_putnl(resv, 1);	/* AUTH_ERROR */
1185 	svc_putnl(resv, ntohl(auth_stat));	/* status */
1186 	goto sendit;
1187 
1188 err_bad_prog:
1189 	dprintk("svc: unknown program %d\n", prog);
1190 	serv->sv_stats->rpcbadfmt++;
1191 	svc_putnl(resv, RPC_PROG_UNAVAIL);
1192 	goto sendit;
1193 
1194 err_bad_vers:
1195 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1196 		       vers, prog, progp->pg_name);
1197 
1198 	serv->sv_stats->rpcbadfmt++;
1199 	svc_putnl(resv, RPC_PROG_MISMATCH);
1200 	svc_putnl(resv, progp->pg_lovers);
1201 	svc_putnl(resv, progp->pg_hivers);
1202 	goto sendit;
1203 
1204 err_bad_proc:
1205 	svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1206 
1207 	serv->sv_stats->rpcbadfmt++;
1208 	svc_putnl(resv, RPC_PROC_UNAVAIL);
1209 	goto sendit;
1210 
1211 err_garbage:
1212 	svc_printk(rqstp, "failed to decode args\n");
1213 
1214 	rpc_stat = rpc_garbage_args;
1215 err_bad:
1216 	serv->sv_stats->rpcbadfmt++;
1217 	svc_putnl(resv, ntohl(rpc_stat));
1218 	goto sendit;
1219 }
1220 EXPORT_SYMBOL_GPL(svc_process);
1221 
1222 /*
1223  * Return (transport-specific) limit on the rpc payload.
1224  */
1225 u32 svc_max_payload(const struct svc_rqst *rqstp)
1226 {
1227 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1228 
1229 	if (rqstp->rq_server->sv_max_payload < max)
1230 		max = rqstp->rq_server->sv_max_payload;
1231 	return max;
1232 }
1233 EXPORT_SYMBOL_GPL(svc_max_payload);
1234