xref: /openbmc/linux/net/sunrpc/svc.c (revision 643d1f7f)
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/sched.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 #define svc_serv_is_pooled(serv)    ((serv)->sv_function)
32 
33 /*
34  * Mode for mapping cpus to pools.
35  */
36 enum {
37 	SVC_POOL_AUTO = -1,	/* choose one of the others */
38 	SVC_POOL_GLOBAL,	/* no mapping, just a single global pool
39 				 * (legacy & UP mode) */
40 	SVC_POOL_PERCPU,	/* one pool per cpu */
41 	SVC_POOL_PERNODE	/* one pool per numa node */
42 };
43 #define SVC_POOL_DEFAULT	SVC_POOL_GLOBAL
44 
45 /*
46  * Structure for mapping cpus to pools and vice versa.
47  * Setup once during sunrpc initialisation.
48  */
49 static struct svc_pool_map {
50 	int count;			/* How many svc_servs use us */
51 	int mode;			/* Note: int not enum to avoid
52 					 * warnings about "enumeration value
53 					 * not handled in switch" */
54 	unsigned int npools;
55 	unsigned int *pool_to;		/* maps pool id to cpu or node */
56 	unsigned int *to_pool;		/* maps cpu or node to pool id */
57 } svc_pool_map = {
58 	.count = 0,
59 	.mode = SVC_POOL_DEFAULT
60 };
61 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
62 
63 static int
64 param_set_pool_mode(const char *val, struct kernel_param *kp)
65 {
66 	int *ip = (int *)kp->arg;
67 	struct svc_pool_map *m = &svc_pool_map;
68 	int err;
69 
70 	mutex_lock(&svc_pool_map_mutex);
71 
72 	err = -EBUSY;
73 	if (m->count)
74 		goto out;
75 
76 	err = 0;
77 	if (!strncmp(val, "auto", 4))
78 		*ip = SVC_POOL_AUTO;
79 	else if (!strncmp(val, "global", 6))
80 		*ip = SVC_POOL_GLOBAL;
81 	else if (!strncmp(val, "percpu", 6))
82 		*ip = SVC_POOL_PERCPU;
83 	else if (!strncmp(val, "pernode", 7))
84 		*ip = SVC_POOL_PERNODE;
85 	else
86 		err = -EINVAL;
87 
88 out:
89 	mutex_unlock(&svc_pool_map_mutex);
90 	return err;
91 }
92 
93 static int
94 param_get_pool_mode(char *buf, struct kernel_param *kp)
95 {
96 	int *ip = (int *)kp->arg;
97 
98 	switch (*ip)
99 	{
100 	case SVC_POOL_AUTO:
101 		return strlcpy(buf, "auto", 20);
102 	case SVC_POOL_GLOBAL:
103 		return strlcpy(buf, "global", 20);
104 	case SVC_POOL_PERCPU:
105 		return strlcpy(buf, "percpu", 20);
106 	case SVC_POOL_PERNODE:
107 		return strlcpy(buf, "pernode", 20);
108 	default:
109 		return sprintf(buf, "%d", *ip);
110 	}
111 }
112 
113 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
114 		 &svc_pool_map.mode, 0644);
115 
116 /*
117  * Detect best pool mapping mode heuristically,
118  * according to the machine's topology.
119  */
120 static int
121 svc_pool_map_choose_mode(void)
122 {
123 	unsigned int node;
124 
125 	if (num_online_nodes() > 1) {
126 		/*
127 		 * Actually have multiple NUMA nodes,
128 		 * so split pools on NUMA node boundaries
129 		 */
130 		return SVC_POOL_PERNODE;
131 	}
132 
133 	node = any_online_node(node_online_map);
134 	if (nr_cpus_node(node) > 2) {
135 		/*
136 		 * Non-trivial SMP, or CONFIG_NUMA on
137 		 * non-NUMA hardware, e.g. with a generic
138 		 * x86_64 kernel on Xeons.  In this case we
139 		 * want to divide the pools on cpu boundaries.
140 		 */
141 		return SVC_POOL_PERCPU;
142 	}
143 
144 	/* default: one global pool */
145 	return SVC_POOL_GLOBAL;
146 }
147 
148 /*
149  * Allocate the to_pool[] and pool_to[] arrays.
150  * Returns 0 on success or an errno.
151  */
152 static int
153 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
154 {
155 	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
156 	if (!m->to_pool)
157 		goto fail;
158 	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
159 	if (!m->pool_to)
160 		goto fail_free;
161 
162 	return 0;
163 
164 fail_free:
165 	kfree(m->to_pool);
166 fail:
167 	return -ENOMEM;
168 }
169 
170 /*
171  * Initialise the pool map for SVC_POOL_PERCPU mode.
172  * Returns number of pools or <0 on error.
173  */
174 static int
175 svc_pool_map_init_percpu(struct svc_pool_map *m)
176 {
177 	unsigned int maxpools = nr_cpu_ids;
178 	unsigned int pidx = 0;
179 	unsigned int cpu;
180 	int err;
181 
182 	err = svc_pool_map_alloc_arrays(m, maxpools);
183 	if (err)
184 		return err;
185 
186 	for_each_online_cpu(cpu) {
187 		BUG_ON(pidx > maxpools);
188 		m->to_pool[cpu] = pidx;
189 		m->pool_to[pidx] = cpu;
190 		pidx++;
191 	}
192 	/* cpus brought online later all get mapped to pool0, sorry */
193 
194 	return pidx;
195 };
196 
197 
198 /*
199  * Initialise the pool map for SVC_POOL_PERNODE mode.
200  * Returns number of pools or <0 on error.
201  */
202 static int
203 svc_pool_map_init_pernode(struct svc_pool_map *m)
204 {
205 	unsigned int maxpools = nr_node_ids;
206 	unsigned int pidx = 0;
207 	unsigned int node;
208 	int err;
209 
210 	err = svc_pool_map_alloc_arrays(m, maxpools);
211 	if (err)
212 		return err;
213 
214 	for_each_node_with_cpus(node) {
215 		/* some architectures (e.g. SN2) have cpuless nodes */
216 		BUG_ON(pidx > maxpools);
217 		m->to_pool[node] = pidx;
218 		m->pool_to[pidx] = node;
219 		pidx++;
220 	}
221 	/* nodes brought online later all get mapped to pool0, sorry */
222 
223 	return pidx;
224 }
225 
226 
227 /*
228  * Add a reference to the global map of cpus to pools (and
229  * vice versa).  Initialise the map if we're the first user.
230  * Returns the number of pools.
231  */
232 static unsigned int
233 svc_pool_map_get(void)
234 {
235 	struct svc_pool_map *m = &svc_pool_map;
236 	int npools = -1;
237 
238 	mutex_lock(&svc_pool_map_mutex);
239 
240 	if (m->count++) {
241 		mutex_unlock(&svc_pool_map_mutex);
242 		return m->npools;
243 	}
244 
245 	if (m->mode == SVC_POOL_AUTO)
246 		m->mode = svc_pool_map_choose_mode();
247 
248 	switch (m->mode) {
249 	case SVC_POOL_PERCPU:
250 		npools = svc_pool_map_init_percpu(m);
251 		break;
252 	case SVC_POOL_PERNODE:
253 		npools = svc_pool_map_init_pernode(m);
254 		break;
255 	}
256 
257 	if (npools < 0) {
258 		/* default, or memory allocation failure */
259 		npools = 1;
260 		m->mode = SVC_POOL_GLOBAL;
261 	}
262 	m->npools = npools;
263 
264 	mutex_unlock(&svc_pool_map_mutex);
265 	return m->npools;
266 }
267 
268 
269 /*
270  * Drop a reference to the global map of cpus to pools.
271  * When the last reference is dropped, the map data is
272  * freed; this allows the sysadmin to change the pool
273  * mode using the pool_mode module option without
274  * rebooting or re-loading sunrpc.ko.
275  */
276 static void
277 svc_pool_map_put(void)
278 {
279 	struct svc_pool_map *m = &svc_pool_map;
280 
281 	mutex_lock(&svc_pool_map_mutex);
282 
283 	if (!--m->count) {
284 		m->mode = SVC_POOL_DEFAULT;
285 		kfree(m->to_pool);
286 		kfree(m->pool_to);
287 		m->npools = 0;
288 	}
289 
290 	mutex_unlock(&svc_pool_map_mutex);
291 }
292 
293 
294 /*
295  * Set the current thread's cpus_allowed mask so that it
296  * will only run on cpus in the given pool.
297  *
298  * Returns 1 and fills in oldmask iff a cpumask was applied.
299  */
300 static inline int
301 svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask)
302 {
303 	struct svc_pool_map *m = &svc_pool_map;
304 	unsigned int node; /* or cpu */
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 	{
314 	default:
315 		return 0;
316 	case SVC_POOL_PERCPU:
317 		node = m->pool_to[pidx];
318 		*oldmask = current->cpus_allowed;
319 		set_cpus_allowed(current, cpumask_of_cpu(node));
320 		return 1;
321 	case SVC_POOL_PERNODE:
322 		node = m->pool_to[pidx];
323 		*oldmask = current->cpus_allowed;
324 		set_cpus_allowed(current, node_to_cpumask(node));
325 		return 1;
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 
359 /*
360  * Create an RPC service
361  */
362 static struct svc_serv *
363 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
364 	   void (*shutdown)(struct svc_serv *serv))
365 {
366 	struct svc_serv	*serv;
367 	unsigned int vers;
368 	unsigned int xdrsize;
369 	unsigned int i;
370 
371 	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
372 		return NULL;
373 	serv->sv_name      = prog->pg_name;
374 	serv->sv_program   = prog;
375 	serv->sv_nrthreads = 1;
376 	serv->sv_stats     = prog->pg_stats;
377 	if (bufsize > RPCSVC_MAXPAYLOAD)
378 		bufsize = RPCSVC_MAXPAYLOAD;
379 	serv->sv_max_payload = bufsize? bufsize : 4096;
380 	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
381 	serv->sv_shutdown  = shutdown;
382 	xdrsize = 0;
383 	while (prog) {
384 		prog->pg_lovers = prog->pg_nvers-1;
385 		for (vers=0; vers<prog->pg_nvers ; vers++)
386 			if (prog->pg_vers[vers]) {
387 				prog->pg_hivers = vers;
388 				if (prog->pg_lovers > vers)
389 					prog->pg_lovers = vers;
390 				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
391 					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
392 			}
393 		prog = prog->pg_next;
394 	}
395 	serv->sv_xdrsize   = xdrsize;
396 	INIT_LIST_HEAD(&serv->sv_tempsocks);
397 	INIT_LIST_HEAD(&serv->sv_permsocks);
398 	init_timer(&serv->sv_temptimer);
399 	spin_lock_init(&serv->sv_lock);
400 
401 	serv->sv_nrpools = npools;
402 	serv->sv_pools =
403 		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
404 			GFP_KERNEL);
405 	if (!serv->sv_pools) {
406 		kfree(serv);
407 		return NULL;
408 	}
409 
410 	for (i = 0; i < serv->sv_nrpools; i++) {
411 		struct svc_pool *pool = &serv->sv_pools[i];
412 
413 		dprintk("svc: initialising pool %u for %s\n",
414 				i, serv->sv_name);
415 
416 		pool->sp_id = i;
417 		INIT_LIST_HEAD(&pool->sp_threads);
418 		INIT_LIST_HEAD(&pool->sp_sockets);
419 		INIT_LIST_HEAD(&pool->sp_all_threads);
420 		spin_lock_init(&pool->sp_lock);
421 	}
422 
423 
424 	/* Remove any stale portmap registrations */
425 	svc_register(serv, 0, 0);
426 
427 	return serv;
428 }
429 
430 struct svc_serv *
431 svc_create(struct svc_program *prog, unsigned int bufsize,
432 		void (*shutdown)(struct svc_serv *serv))
433 {
434 	return __svc_create(prog, bufsize, /*npools*/1, shutdown);
435 }
436 EXPORT_SYMBOL(svc_create);
437 
438 struct svc_serv *
439 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
440 		void (*shutdown)(struct svc_serv *serv),
441 		  svc_thread_fn func, int sig, struct module *mod)
442 {
443 	struct svc_serv *serv;
444 	unsigned int npools = svc_pool_map_get();
445 
446 	serv = __svc_create(prog, bufsize, npools, shutdown);
447 
448 	if (serv != NULL) {
449 		serv->sv_function = func;
450 		serv->sv_kill_signal = sig;
451 		serv->sv_module = mod;
452 	}
453 
454 	return serv;
455 }
456 EXPORT_SYMBOL(svc_create_pooled);
457 
458 /*
459  * Destroy an RPC service.  Should be called with the BKL held
460  */
461 void
462 svc_destroy(struct svc_serv *serv)
463 {
464 	dprintk("svc: svc_destroy(%s, %d)\n",
465 				serv->sv_program->pg_name,
466 				serv->sv_nrthreads);
467 
468 	if (serv->sv_nrthreads) {
469 		if (--(serv->sv_nrthreads) != 0) {
470 			svc_sock_update_bufs(serv);
471 			return;
472 		}
473 	} else
474 		printk("svc_destroy: no threads for serv=%p!\n", serv);
475 
476 	del_timer_sync(&serv->sv_temptimer);
477 
478 	svc_close_all(&serv->sv_tempsocks);
479 
480 	if (serv->sv_shutdown)
481 		serv->sv_shutdown(serv);
482 
483 	svc_close_all(&serv->sv_permsocks);
484 
485 	BUG_ON(!list_empty(&serv->sv_permsocks));
486 	BUG_ON(!list_empty(&serv->sv_tempsocks));
487 
488 	cache_clean_deferred(serv);
489 
490 	if (svc_serv_is_pooled(serv))
491 		svc_pool_map_put();
492 
493 	/* Unregister service with the portmapper */
494 	svc_register(serv, 0, 0);
495 	kfree(serv->sv_pools);
496 	kfree(serv);
497 }
498 EXPORT_SYMBOL(svc_destroy);
499 
500 /*
501  * Allocate an RPC server's buffer space.
502  * We allocate pages and place them in rq_argpages.
503  */
504 static int
505 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
506 {
507 	int pages;
508 	int arghi;
509 
510 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
511 				       * We assume one is at most one page
512 				       */
513 	arghi = 0;
514 	BUG_ON(pages > RPCSVC_MAXPAGES);
515 	while (pages) {
516 		struct page *p = alloc_page(GFP_KERNEL);
517 		if (!p)
518 			break;
519 		rqstp->rq_pages[arghi++] = p;
520 		pages--;
521 	}
522 	return ! pages;
523 }
524 
525 /*
526  * Release an RPC server buffer
527  */
528 static void
529 svc_release_buffer(struct svc_rqst *rqstp)
530 {
531 	int i;
532 	for (i=0; i<ARRAY_SIZE(rqstp->rq_pages); i++)
533 		if (rqstp->rq_pages[i])
534 			put_page(rqstp->rq_pages[i]);
535 }
536 
537 struct svc_rqst *
538 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
539 {
540 	struct svc_rqst	*rqstp;
541 
542 	rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
543 	if (!rqstp)
544 		goto out_enomem;
545 
546 	init_waitqueue_head(&rqstp->rq_wait);
547 
548 	serv->sv_nrthreads++;
549 	spin_lock_bh(&pool->sp_lock);
550 	pool->sp_nrthreads++;
551 	list_add(&rqstp->rq_all, &pool->sp_all_threads);
552 	spin_unlock_bh(&pool->sp_lock);
553 	rqstp->rq_server = serv;
554 	rqstp->rq_pool = pool;
555 
556 	rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
557 	if (!rqstp->rq_argp)
558 		goto out_thread;
559 
560 	rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
561 	if (!rqstp->rq_resp)
562 		goto out_thread;
563 
564 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
565 		goto out_thread;
566 
567 	return rqstp;
568 out_thread:
569 	svc_exit_thread(rqstp);
570 out_enomem:
571 	return ERR_PTR(-ENOMEM);
572 }
573 EXPORT_SYMBOL(svc_prepare_thread);
574 
575 /*
576  * Create a thread in the given pool.  Caller must hold BKL.
577  * On a NUMA or SMP machine, with a multi-pool serv, the thread
578  * will be restricted to run on the cpus belonging to the pool.
579  */
580 static int
581 __svc_create_thread(svc_thread_fn func, struct svc_serv *serv,
582 		    struct svc_pool *pool)
583 {
584 	struct svc_rqst	*rqstp;
585 	int		error = -ENOMEM;
586 	int		have_oldmask = 0;
587 	cpumask_t	oldmask;
588 
589 	rqstp = svc_prepare_thread(serv, pool);
590 	if (IS_ERR(rqstp)) {
591 		error = PTR_ERR(rqstp);
592 		goto out;
593 	}
594 
595 	if (serv->sv_nrpools > 1)
596 		have_oldmask = svc_pool_map_set_cpumask(pool->sp_id, &oldmask);
597 
598 	error = kernel_thread((int (*)(void *)) func, rqstp, 0);
599 
600 	if (have_oldmask)
601 		set_cpus_allowed(current, oldmask);
602 
603 	if (error < 0)
604 		goto out_thread;
605 	svc_sock_update_bufs(serv);
606 	error = 0;
607 out:
608 	return error;
609 
610 out_thread:
611 	svc_exit_thread(rqstp);
612 	goto out;
613 }
614 
615 /*
616  * Create a thread in the default pool.  Caller must hold BKL.
617  */
618 int
619 svc_create_thread(svc_thread_fn func, struct svc_serv *serv)
620 {
621 	return __svc_create_thread(func, serv, &serv->sv_pools[0]);
622 }
623 EXPORT_SYMBOL(svc_create_thread);
624 
625 /*
626  * Choose a pool in which to create a new thread, for svc_set_num_threads
627  */
628 static inline struct svc_pool *
629 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
630 {
631 	if (pool != NULL)
632 		return pool;
633 
634 	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
635 }
636 
637 /*
638  * Choose a thread to kill, for svc_set_num_threads
639  */
640 static inline struct task_struct *
641 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
642 {
643 	unsigned int i;
644 	struct task_struct *task = NULL;
645 
646 	if (pool != NULL) {
647 		spin_lock_bh(&pool->sp_lock);
648 	} else {
649 		/* choose a pool in round-robin fashion */
650 		for (i = 0; i < serv->sv_nrpools; i++) {
651 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
652 			spin_lock_bh(&pool->sp_lock);
653 			if (!list_empty(&pool->sp_all_threads))
654 				goto found_pool;
655 			spin_unlock_bh(&pool->sp_lock);
656 		}
657 		return NULL;
658 	}
659 
660 found_pool:
661 	if (!list_empty(&pool->sp_all_threads)) {
662 		struct svc_rqst *rqstp;
663 
664 		/*
665 		 * Remove from the pool->sp_all_threads list
666 		 * so we don't try to kill it again.
667 		 */
668 		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
669 		list_del_init(&rqstp->rq_all);
670 		task = rqstp->rq_task;
671 	}
672 	spin_unlock_bh(&pool->sp_lock);
673 
674 	return task;
675 }
676 
677 /*
678  * Create or destroy enough new threads to make the number
679  * of threads the given number.  If `pool' is non-NULL, applies
680  * only to threads in that pool, otherwise round-robins between
681  * all pools.  Must be called with a svc_get() reference and
682  * the BKL held.
683  *
684  * Destroying threads relies on the service threads filling in
685  * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
686  * has been created using svc_create_pooled().
687  *
688  * Based on code that used to be in nfsd_svc() but tweaked
689  * to be pool-aware.
690  */
691 int
692 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
693 {
694 	struct task_struct *victim;
695 	int error = 0;
696 	unsigned int state = serv->sv_nrthreads-1;
697 
698 	if (pool == NULL) {
699 		/* The -1 assumes caller has done a svc_get() */
700 		nrservs -= (serv->sv_nrthreads-1);
701 	} else {
702 		spin_lock_bh(&pool->sp_lock);
703 		nrservs -= pool->sp_nrthreads;
704 		spin_unlock_bh(&pool->sp_lock);
705 	}
706 
707 	/* create new threads */
708 	while (nrservs > 0) {
709 		nrservs--;
710 		__module_get(serv->sv_module);
711 		error = __svc_create_thread(serv->sv_function, serv,
712 					    choose_pool(serv, pool, &state));
713 		if (error < 0) {
714 			module_put(serv->sv_module);
715 			break;
716 		}
717 	}
718 	/* destroy old threads */
719 	while (nrservs < 0 &&
720 	       (victim = choose_victim(serv, pool, &state)) != NULL) {
721 		send_sig(serv->sv_kill_signal, victim, 1);
722 		nrservs++;
723 	}
724 
725 	return error;
726 }
727 EXPORT_SYMBOL(svc_set_num_threads);
728 
729 /*
730  * Called from a server thread as it's exiting.  Caller must hold BKL.
731  */
732 void
733 svc_exit_thread(struct svc_rqst *rqstp)
734 {
735 	struct svc_serv	*serv = rqstp->rq_server;
736 	struct svc_pool	*pool = rqstp->rq_pool;
737 
738 	svc_release_buffer(rqstp);
739 	kfree(rqstp->rq_resp);
740 	kfree(rqstp->rq_argp);
741 	kfree(rqstp->rq_auth_data);
742 
743 	spin_lock_bh(&pool->sp_lock);
744 	pool->sp_nrthreads--;
745 	list_del(&rqstp->rq_all);
746 	spin_unlock_bh(&pool->sp_lock);
747 
748 	kfree(rqstp);
749 
750 	/* Release the server */
751 	if (serv)
752 		svc_destroy(serv);
753 }
754 EXPORT_SYMBOL(svc_exit_thread);
755 
756 /*
757  * Register an RPC service with the local portmapper.
758  * To unregister a service, call this routine with
759  * proto and port == 0.
760  */
761 int
762 svc_register(struct svc_serv *serv, int proto, unsigned short port)
763 {
764 	struct svc_program	*progp;
765 	unsigned long		flags;
766 	unsigned int		i;
767 	int			error = 0, dummy;
768 
769 	if (!port)
770 		clear_thread_flag(TIF_SIGPENDING);
771 
772 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
773 		for (i = 0; i < progp->pg_nvers; i++) {
774 			if (progp->pg_vers[i] == NULL)
775 				continue;
776 
777 			dprintk("svc: svc_register(%s, %s, %d, %d)%s\n",
778 					progp->pg_name,
779 					proto == IPPROTO_UDP?  "udp" : "tcp",
780 					port,
781 					i,
782 					progp->pg_vers[i]->vs_hidden?
783 						" (but not telling portmap)" : "");
784 
785 			if (progp->pg_vers[i]->vs_hidden)
786 				continue;
787 
788 			error = rpcb_register(progp->pg_prog, i, proto, port, &dummy);
789 			if (error < 0)
790 				break;
791 			if (port && !dummy) {
792 				error = -EACCES;
793 				break;
794 			}
795 		}
796 	}
797 
798 	if (!port) {
799 		spin_lock_irqsave(&current->sighand->siglock, flags);
800 		recalc_sigpending();
801 		spin_unlock_irqrestore(&current->sighand->siglock, flags);
802 	}
803 
804 	return error;
805 }
806 
807 /*
808  * Printk the given error with the address of the client that caused it.
809  */
810 static int
811 __attribute__ ((format (printf, 2, 3)))
812 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
813 {
814 	va_list args;
815 	int 	r;
816 	char 	buf[RPC_MAX_ADDRBUFLEN];
817 
818 	if (!net_ratelimit())
819 		return 0;
820 
821 	printk(KERN_WARNING "svc: %s: ",
822 		svc_print_addr(rqstp, buf, sizeof(buf)));
823 
824 	va_start(args, fmt);
825 	r = vprintk(fmt, args);
826 	va_end(args);
827 
828 	return r;
829 }
830 
831 /*
832  * Process the RPC request.
833  */
834 int
835 svc_process(struct svc_rqst *rqstp)
836 {
837 	struct svc_program	*progp;
838 	struct svc_version	*versp = NULL;	/* compiler food */
839 	struct svc_procedure	*procp = NULL;
840 	struct kvec *		argv = &rqstp->rq_arg.head[0];
841 	struct kvec *		resv = &rqstp->rq_res.head[0];
842 	struct svc_serv		*serv = rqstp->rq_server;
843 	kxdrproc_t		xdr;
844 	__be32			*statp;
845 	u32			dir, prog, vers, proc;
846 	__be32			auth_stat, rpc_stat;
847 	int			auth_res;
848 	__be32			*reply_statp;
849 
850 	rpc_stat = rpc_success;
851 
852 	if (argv->iov_len < 6*4)
853 		goto err_short_len;
854 
855 	/* setup response xdr_buf.
856 	 * Initially it has just one page
857 	 */
858 	rqstp->rq_resused = 1;
859 	resv->iov_base = page_address(rqstp->rq_respages[0]);
860 	resv->iov_len = 0;
861 	rqstp->rq_res.pages = rqstp->rq_respages + 1;
862 	rqstp->rq_res.len = 0;
863 	rqstp->rq_res.page_base = 0;
864 	rqstp->rq_res.page_len = 0;
865 	rqstp->rq_res.buflen = PAGE_SIZE;
866 	rqstp->rq_res.tail[0].iov_base = NULL;
867 	rqstp->rq_res.tail[0].iov_len = 0;
868 	/* Will be turned off only in gss privacy case: */
869 	rqstp->rq_splice_ok = 1;
870 
871 	/* Setup reply header */
872 	rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
873 
874 	rqstp->rq_xid = svc_getu32(argv);
875 	svc_putu32(resv, rqstp->rq_xid);
876 
877 	dir  = svc_getnl(argv);
878 	vers = svc_getnl(argv);
879 
880 	/* First words of reply: */
881 	svc_putnl(resv, 1);		/* REPLY */
882 
883 	if (dir != 0)		/* direction != CALL */
884 		goto err_bad_dir;
885 	if (vers != 2)		/* RPC version number */
886 		goto err_bad_rpc;
887 
888 	/* Save position in case we later decide to reject: */
889 	reply_statp = resv->iov_base + resv->iov_len;
890 
891 	svc_putnl(resv, 0);		/* ACCEPT */
892 
893 	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
894 	rqstp->rq_vers = vers = svc_getnl(argv);	/* version number */
895 	rqstp->rq_proc = proc = svc_getnl(argv);	/* procedure number */
896 
897 	progp = serv->sv_program;
898 
899 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
900 		if (prog == progp->pg_prog)
901 			break;
902 
903 	/*
904 	 * Decode auth data, and add verifier to reply buffer.
905 	 * We do this before anything else in order to get a decent
906 	 * auth verifier.
907 	 */
908 	auth_res = svc_authenticate(rqstp, &auth_stat);
909 	/* Also give the program a chance to reject this call: */
910 	if (auth_res == SVC_OK && progp) {
911 		auth_stat = rpc_autherr_badcred;
912 		auth_res = progp->pg_authenticate(rqstp);
913 	}
914 	switch (auth_res) {
915 	case SVC_OK:
916 		break;
917 	case SVC_GARBAGE:
918 		rpc_stat = rpc_garbage_args;
919 		goto err_bad;
920 	case SVC_SYSERR:
921 		rpc_stat = rpc_system_err;
922 		goto err_bad;
923 	case SVC_DENIED:
924 		goto err_bad_auth;
925 	case SVC_DROP:
926 		goto dropit;
927 	case SVC_COMPLETE:
928 		goto sendit;
929 	}
930 
931 	if (progp == NULL)
932 		goto err_bad_prog;
933 
934 	if (vers >= progp->pg_nvers ||
935 	  !(versp = progp->pg_vers[vers]))
936 		goto err_bad_vers;
937 
938 	procp = versp->vs_proc + proc;
939 	if (proc >= versp->vs_nproc || !procp->pc_func)
940 		goto err_bad_proc;
941 	rqstp->rq_server   = serv;
942 	rqstp->rq_procinfo = procp;
943 
944 	/* Syntactic check complete */
945 	serv->sv_stats->rpccnt++;
946 
947 	/* Build the reply header. */
948 	statp = resv->iov_base +resv->iov_len;
949 	svc_putnl(resv, RPC_SUCCESS);
950 
951 	/* Bump per-procedure stats counter */
952 	procp->pc_count++;
953 
954 	/* Initialize storage for argp and resp */
955 	memset(rqstp->rq_argp, 0, procp->pc_argsize);
956 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
957 
958 	/* un-reserve some of the out-queue now that we have a
959 	 * better idea of reply size
960 	 */
961 	if (procp->pc_xdrressize)
962 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
963 
964 	/* Call the function that processes the request. */
965 	if (!versp->vs_dispatch) {
966 		/* Decode arguments */
967 		xdr = procp->pc_decode;
968 		if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
969 			goto err_garbage;
970 
971 		*statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
972 
973 		/* Encode reply */
974 		if (*statp == rpc_drop_reply) {
975 			if (procp->pc_release)
976 				procp->pc_release(rqstp, NULL, rqstp->rq_resp);
977 			goto dropit;
978 		}
979 		if (*statp == rpc_success && (xdr = procp->pc_encode)
980 		 && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
981 			dprintk("svc: failed to encode reply\n");
982 			/* serv->sv_stats->rpcsystemerr++; */
983 			*statp = rpc_system_err;
984 		}
985 	} else {
986 		dprintk("svc: calling dispatcher\n");
987 		if (!versp->vs_dispatch(rqstp, statp)) {
988 			/* Release reply info */
989 			if (procp->pc_release)
990 				procp->pc_release(rqstp, NULL, rqstp->rq_resp);
991 			goto dropit;
992 		}
993 	}
994 
995 	/* Check RPC status result */
996 	if (*statp != rpc_success)
997 		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
998 
999 	/* Release reply info */
1000 	if (procp->pc_release)
1001 		procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1002 
1003 	if (procp->pc_encode == NULL)
1004 		goto dropit;
1005 
1006  sendit:
1007 	if (svc_authorise(rqstp))
1008 		goto dropit;
1009 	return svc_send(rqstp);
1010 
1011  dropit:
1012 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1013 	dprintk("svc: svc_process dropit\n");
1014 	svc_drop(rqstp);
1015 	return 0;
1016 
1017 err_short_len:
1018 	svc_printk(rqstp, "short len %Zd, dropping request\n",
1019 			argv->iov_len);
1020 
1021 	goto dropit;			/* drop request */
1022 
1023 err_bad_dir:
1024 	svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1025 
1026 	serv->sv_stats->rpcbadfmt++;
1027 	goto dropit;			/* drop request */
1028 
1029 err_bad_rpc:
1030 	serv->sv_stats->rpcbadfmt++;
1031 	svc_putnl(resv, 1);	/* REJECT */
1032 	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1033 	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1034 	svc_putnl(resv, 2);
1035 	goto sendit;
1036 
1037 err_bad_auth:
1038 	dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1039 	serv->sv_stats->rpcbadauth++;
1040 	/* Restore write pointer to location of accept status: */
1041 	xdr_ressize_check(rqstp, reply_statp);
1042 	svc_putnl(resv, 1);	/* REJECT */
1043 	svc_putnl(resv, 1);	/* AUTH_ERROR */
1044 	svc_putnl(resv, ntohl(auth_stat));	/* status */
1045 	goto sendit;
1046 
1047 err_bad_prog:
1048 	dprintk("svc: unknown program %d\n", prog);
1049 	serv->sv_stats->rpcbadfmt++;
1050 	svc_putnl(resv, RPC_PROG_UNAVAIL);
1051 	goto sendit;
1052 
1053 err_bad_vers:
1054 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1055 		       vers, prog, progp->pg_name);
1056 
1057 	serv->sv_stats->rpcbadfmt++;
1058 	svc_putnl(resv, RPC_PROG_MISMATCH);
1059 	svc_putnl(resv, progp->pg_lovers);
1060 	svc_putnl(resv, progp->pg_hivers);
1061 	goto sendit;
1062 
1063 err_bad_proc:
1064 	svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1065 
1066 	serv->sv_stats->rpcbadfmt++;
1067 	svc_putnl(resv, RPC_PROC_UNAVAIL);
1068 	goto sendit;
1069 
1070 err_garbage:
1071 	svc_printk(rqstp, "failed to decode args\n");
1072 
1073 	rpc_stat = rpc_garbage_args;
1074 err_bad:
1075 	serv->sv_stats->rpcbadfmt++;
1076 	svc_putnl(resv, ntohl(rpc_stat));
1077 	goto sendit;
1078 }
1079 EXPORT_SYMBOL(svc_process);
1080 
1081 /*
1082  * Return (transport-specific) limit on the rpc payload.
1083  */
1084 u32 svc_max_payload(const struct svc_rqst *rqstp)
1085 {
1086 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1087 
1088 	if (rqstp->rq_server->sv_max_payload < max)
1089 		max = rqstp->rq_server->sv_max_payload;
1090 	return max;
1091 }
1092 EXPORT_SYMBOL_GPL(svc_max_payload);
1093