• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/net/sunrpc/svc.c
4  *
5  * High-level RPC service routines
6  *
7  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8  *
9  * Multiple threads pools and NUMAisation
10  * Copyright (c) 2006 Silicon Graphics, Inc.
11  * by Greg Banks <gnb@melbourne.sgi.com>
12  */
13 
14 #include <linux/linkage.h>
15 #include <linux/sched/signal.h>
16 #include <linux/errno.h>
17 #include <linux/net.h>
18 #include <linux/in.h>
19 #include <linux/mm.h>
20 #include <linux/interrupt.h>
21 #include <linux/module.h>
22 #include <linux/kthread.h>
23 #include <linux/slab.h>
24 
25 #include <linux/sunrpc/types.h>
26 #include <linux/sunrpc/xdr.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/svcsock.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/bc_xprt.h>
31 
32 #include <trace/events/sunrpc.h>
33 
34 #include "fail.h"
35 
36 #define RPCDBG_FACILITY	RPCDBG_SVCDSP
37 
38 static void svc_unregister(const struct svc_serv *serv, struct net *net);
39 
40 #define svc_serv_is_pooled(serv)    ((serv)->sv_ops->svo_function)
41 
42 #define SVC_POOL_DEFAULT	SVC_POOL_GLOBAL
43 
44 /*
45  * Structure for mapping cpus to pools and vice versa.
46  * Setup once during sunrpc initialisation.
47  */
48 struct svc_pool_map svc_pool_map = {
49 	.mode = SVC_POOL_DEFAULT
50 };
51 EXPORT_SYMBOL_GPL(svc_pool_map);
52 
53 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
54 
55 static int
param_set_pool_mode(const char * val,const struct kernel_param * kp)56 param_set_pool_mode(const char *val, const struct kernel_param *kp)
57 {
58 	int *ip = (int *)kp->arg;
59 	struct svc_pool_map *m = &svc_pool_map;
60 	int err;
61 
62 	mutex_lock(&svc_pool_map_mutex);
63 
64 	err = -EBUSY;
65 	if (m->count)
66 		goto out;
67 
68 	err = 0;
69 	if (!strncmp(val, "auto", 4))
70 		*ip = SVC_POOL_AUTO;
71 	else if (!strncmp(val, "global", 6))
72 		*ip = SVC_POOL_GLOBAL;
73 	else if (!strncmp(val, "percpu", 6))
74 		*ip = SVC_POOL_PERCPU;
75 	else if (!strncmp(val, "pernode", 7))
76 		*ip = SVC_POOL_PERNODE;
77 	else
78 		err = -EINVAL;
79 
80 out:
81 	mutex_unlock(&svc_pool_map_mutex);
82 	return err;
83 }
84 
85 static int
param_get_pool_mode(char * buf,const struct kernel_param * kp)86 param_get_pool_mode(char *buf, const struct kernel_param *kp)
87 {
88 	int *ip = (int *)kp->arg;
89 
90 	switch (*ip)
91 	{
92 	case SVC_POOL_AUTO:
93 		return strlcpy(buf, "auto\n", 20);
94 	case SVC_POOL_GLOBAL:
95 		return strlcpy(buf, "global\n", 20);
96 	case SVC_POOL_PERCPU:
97 		return strlcpy(buf, "percpu\n", 20);
98 	case SVC_POOL_PERNODE:
99 		return strlcpy(buf, "pernode\n", 20);
100 	default:
101 		return sprintf(buf, "%d\n", *ip);
102 	}
103 }
104 
105 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
106 		 &svc_pool_map.mode, 0644);
107 
108 /*
109  * Detect best pool mapping mode heuristically,
110  * according to the machine's topology.
111  */
112 static int
svc_pool_map_choose_mode(void)113 svc_pool_map_choose_mode(void)
114 {
115 	unsigned int node;
116 
117 	if (nr_online_nodes > 1) {
118 		/*
119 		 * Actually have multiple NUMA nodes,
120 		 * so split pools on NUMA node boundaries
121 		 */
122 		return SVC_POOL_PERNODE;
123 	}
124 
125 	node = first_online_node;
126 	if (nr_cpus_node(node) > 2) {
127 		/*
128 		 * Non-trivial SMP, or CONFIG_NUMA on
129 		 * non-NUMA hardware, e.g. with a generic
130 		 * x86_64 kernel on Xeons.  In this case we
131 		 * want to divide the pools on cpu boundaries.
132 		 */
133 		return SVC_POOL_PERCPU;
134 	}
135 
136 	/* default: one global pool */
137 	return SVC_POOL_GLOBAL;
138 }
139 
140 /*
141  * Allocate the to_pool[] and pool_to[] arrays.
142  * Returns 0 on success or an errno.
143  */
144 static int
svc_pool_map_alloc_arrays(struct svc_pool_map * m,unsigned int maxpools)145 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
146 {
147 	m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
148 	if (!m->to_pool)
149 		goto fail;
150 	m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
151 	if (!m->pool_to)
152 		goto fail_free;
153 
154 	return 0;
155 
156 fail_free:
157 	kfree(m->to_pool);
158 	m->to_pool = NULL;
159 fail:
160 	return -ENOMEM;
161 }
162 
163 /*
164  * Initialise the pool map for SVC_POOL_PERCPU mode.
165  * Returns number of pools or <0 on error.
166  */
167 static int
svc_pool_map_init_percpu(struct svc_pool_map * m)168 svc_pool_map_init_percpu(struct svc_pool_map *m)
169 {
170 	unsigned int maxpools = nr_cpu_ids;
171 	unsigned int pidx = 0;
172 	unsigned int cpu;
173 	int err;
174 
175 	err = svc_pool_map_alloc_arrays(m, maxpools);
176 	if (err)
177 		return err;
178 
179 	for_each_online_cpu(cpu) {
180 		BUG_ON(pidx >= maxpools);
181 		m->to_pool[cpu] = pidx;
182 		m->pool_to[pidx] = cpu;
183 		pidx++;
184 	}
185 	/* cpus brought online later all get mapped to pool0, sorry */
186 
187 	return pidx;
188 };
189 
190 
191 /*
192  * Initialise the pool map for SVC_POOL_PERNODE mode.
193  * Returns number of pools or <0 on error.
194  */
195 static int
svc_pool_map_init_pernode(struct svc_pool_map * m)196 svc_pool_map_init_pernode(struct svc_pool_map *m)
197 {
198 	unsigned int maxpools = nr_node_ids;
199 	unsigned int pidx = 0;
200 	unsigned int node;
201 	int err;
202 
203 	err = svc_pool_map_alloc_arrays(m, maxpools);
204 	if (err)
205 		return err;
206 
207 	for_each_node_with_cpus(node) {
208 		/* some architectures (e.g. SN2) have cpuless nodes */
209 		BUG_ON(pidx > maxpools);
210 		m->to_pool[node] = pidx;
211 		m->pool_to[pidx] = node;
212 		pidx++;
213 	}
214 	/* nodes brought online later all get mapped to pool0, sorry */
215 
216 	return pidx;
217 }
218 
219 
220 /*
221  * Add a reference to the global map of cpus to pools (and
222  * vice versa).  Initialise the map if we're the first user.
223  * Returns the number of pools.
224  */
225 unsigned int
svc_pool_map_get(void)226 svc_pool_map_get(void)
227 {
228 	struct svc_pool_map *m = &svc_pool_map;
229 	int npools = -1;
230 
231 	mutex_lock(&svc_pool_map_mutex);
232 
233 	if (m->count++) {
234 		mutex_unlock(&svc_pool_map_mutex);
235 		return m->npools;
236 	}
237 
238 	if (m->mode == SVC_POOL_AUTO)
239 		m->mode = svc_pool_map_choose_mode();
240 
241 	switch (m->mode) {
242 	case SVC_POOL_PERCPU:
243 		npools = svc_pool_map_init_percpu(m);
244 		break;
245 	case SVC_POOL_PERNODE:
246 		npools = svc_pool_map_init_pernode(m);
247 		break;
248 	}
249 
250 	if (npools < 0) {
251 		/* default, or memory allocation failure */
252 		npools = 1;
253 		m->mode = SVC_POOL_GLOBAL;
254 	}
255 	m->npools = npools;
256 
257 	mutex_unlock(&svc_pool_map_mutex);
258 	return m->npools;
259 }
260 EXPORT_SYMBOL_GPL(svc_pool_map_get);
261 
262 /*
263  * Drop a reference to the global map of cpus to pools.
264  * When the last reference is dropped, the map data is
265  * freed; this allows the sysadmin to change the pool
266  * mode using the pool_mode module option without
267  * rebooting or re-loading sunrpc.ko.
268  */
269 void
svc_pool_map_put(void)270 svc_pool_map_put(void)
271 {
272 	struct svc_pool_map *m = &svc_pool_map;
273 
274 	mutex_lock(&svc_pool_map_mutex);
275 
276 	if (!--m->count) {
277 		kfree(m->to_pool);
278 		m->to_pool = NULL;
279 		kfree(m->pool_to);
280 		m->pool_to = NULL;
281 		m->npools = 0;
282 	}
283 
284 	mutex_unlock(&svc_pool_map_mutex);
285 }
286 EXPORT_SYMBOL_GPL(svc_pool_map_put);
287 
svc_pool_map_get_node(unsigned int pidx)288 static int svc_pool_map_get_node(unsigned int pidx)
289 {
290 	const struct svc_pool_map *m = &svc_pool_map;
291 
292 	if (m->count) {
293 		if (m->mode == SVC_POOL_PERCPU)
294 			return cpu_to_node(m->pool_to[pidx]);
295 		if (m->mode == SVC_POOL_PERNODE)
296 			return m->pool_to[pidx];
297 	}
298 	return NUMA_NO_NODE;
299 }
300 /*
301  * Set the given thread's cpus_allowed mask so that it
302  * will only run on cpus in the given pool.
303  */
304 static inline void
svc_pool_map_set_cpumask(struct task_struct * task,unsigned int pidx)305 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
306 {
307 	struct svc_pool_map *m = &svc_pool_map;
308 	unsigned int node = m->pool_to[pidx];
309 
310 	/*
311 	 * The caller checks for sv_nrpools > 1, which
312 	 * implies that we've been initialized.
313 	 */
314 	WARN_ON_ONCE(m->count == 0);
315 	if (m->count == 0)
316 		return;
317 
318 	switch (m->mode) {
319 	case SVC_POOL_PERCPU:
320 	{
321 		set_cpus_allowed_ptr(task, cpumask_of(node));
322 		break;
323 	}
324 	case SVC_POOL_PERNODE:
325 	{
326 		set_cpus_allowed_ptr(task, cpumask_of_node(node));
327 		break;
328 	}
329 	}
330 }
331 
332 /*
333  * Use the mapping mode to choose a pool for a given CPU.
334  * Used when enqueueing an incoming RPC.  Always returns
335  * a non-NULL pool pointer.
336  */
337 struct svc_pool *
svc_pool_for_cpu(struct svc_serv * serv,int cpu)338 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
339 {
340 	struct svc_pool_map *m = &svc_pool_map;
341 	unsigned int pidx = 0;
342 
343 	/*
344 	 * An uninitialised map happens in a pure client when
345 	 * lockd is brought up, so silently treat it the
346 	 * same as SVC_POOL_GLOBAL.
347 	 */
348 	if (svc_serv_is_pooled(serv)) {
349 		switch (m->mode) {
350 		case SVC_POOL_PERCPU:
351 			pidx = m->to_pool[cpu];
352 			break;
353 		case SVC_POOL_PERNODE:
354 			pidx = m->to_pool[cpu_to_node(cpu)];
355 			break;
356 		}
357 	}
358 	return &serv->sv_pools[pidx % serv->sv_nrpools];
359 }
360 
svc_rpcb_setup(struct svc_serv * serv,struct net * net)361 int svc_rpcb_setup(struct svc_serv *serv, struct net *net)
362 {
363 	int err;
364 
365 	err = rpcb_create_local(net);
366 	if (err)
367 		return err;
368 
369 	/* Remove any stale portmap registrations */
370 	svc_unregister(serv, net);
371 	return 0;
372 }
373 EXPORT_SYMBOL_GPL(svc_rpcb_setup);
374 
svc_rpcb_cleanup(struct svc_serv * serv,struct net * net)375 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net)
376 {
377 	svc_unregister(serv, net);
378 	rpcb_put_local(net);
379 }
380 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup);
381 
svc_uses_rpcbind(struct svc_serv * serv)382 static int svc_uses_rpcbind(struct svc_serv *serv)
383 {
384 	struct svc_program	*progp;
385 	unsigned int		i;
386 
387 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
388 		for (i = 0; i < progp->pg_nvers; i++) {
389 			if (progp->pg_vers[i] == NULL)
390 				continue;
391 			if (!progp->pg_vers[i]->vs_hidden)
392 				return 1;
393 		}
394 	}
395 
396 	return 0;
397 }
398 
svc_bind(struct svc_serv * serv,struct net * net)399 int svc_bind(struct svc_serv *serv, struct net *net)
400 {
401 	if (!svc_uses_rpcbind(serv))
402 		return 0;
403 	return svc_rpcb_setup(serv, net);
404 }
405 EXPORT_SYMBOL_GPL(svc_bind);
406 
407 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
408 static void
__svc_init_bc(struct svc_serv * serv)409 __svc_init_bc(struct svc_serv *serv)
410 {
411 	INIT_LIST_HEAD(&serv->sv_cb_list);
412 	spin_lock_init(&serv->sv_cb_lock);
413 	init_waitqueue_head(&serv->sv_cb_waitq);
414 }
415 #else
416 static void
__svc_init_bc(struct svc_serv * serv)417 __svc_init_bc(struct svc_serv *serv)
418 {
419 }
420 #endif
421 
422 /*
423  * Create an RPC service
424  */
425 static struct svc_serv *
__svc_create(struct svc_program * prog,unsigned int bufsize,int npools,const struct svc_serv_ops * ops)426 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
427 	     const struct svc_serv_ops *ops)
428 {
429 	struct svc_serv	*serv;
430 	unsigned int vers;
431 	unsigned int xdrsize;
432 	unsigned int i;
433 
434 	if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
435 		return NULL;
436 	serv->sv_name      = prog->pg_name;
437 	serv->sv_program   = prog;
438 	serv->sv_nrthreads = 1;
439 	serv->sv_stats     = prog->pg_stats;
440 	if (bufsize > RPCSVC_MAXPAYLOAD)
441 		bufsize = RPCSVC_MAXPAYLOAD;
442 	serv->sv_max_payload = bufsize? bufsize : 4096;
443 	serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
444 	serv->sv_ops = ops;
445 	xdrsize = 0;
446 	while (prog) {
447 		prog->pg_lovers = prog->pg_nvers-1;
448 		for (vers=0; vers<prog->pg_nvers ; vers++)
449 			if (prog->pg_vers[vers]) {
450 				prog->pg_hivers = vers;
451 				if (prog->pg_lovers > vers)
452 					prog->pg_lovers = vers;
453 				if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
454 					xdrsize = prog->pg_vers[vers]->vs_xdrsize;
455 			}
456 		prog = prog->pg_next;
457 	}
458 	serv->sv_xdrsize   = xdrsize;
459 	INIT_LIST_HEAD(&serv->sv_tempsocks);
460 	INIT_LIST_HEAD(&serv->sv_permsocks);
461 	timer_setup(&serv->sv_temptimer, NULL, 0);
462 	spin_lock_init(&serv->sv_lock);
463 
464 	__svc_init_bc(serv);
465 
466 	serv->sv_nrpools = npools;
467 	serv->sv_pools =
468 		kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
469 			GFP_KERNEL);
470 	if (!serv->sv_pools) {
471 		kfree(serv);
472 		return NULL;
473 	}
474 
475 	for (i = 0; i < serv->sv_nrpools; i++) {
476 		struct svc_pool *pool = &serv->sv_pools[i];
477 
478 		dprintk("svc: initialising pool %u for %s\n",
479 				i, serv->sv_name);
480 
481 		pool->sp_id = i;
482 		INIT_LIST_HEAD(&pool->sp_sockets);
483 		INIT_LIST_HEAD(&pool->sp_all_threads);
484 		spin_lock_init(&pool->sp_lock);
485 	}
486 
487 	return serv;
488 }
489 
490 struct svc_serv *
svc_create(struct svc_program * prog,unsigned int bufsize,const struct svc_serv_ops * ops)491 svc_create(struct svc_program *prog, unsigned int bufsize,
492 	   const struct svc_serv_ops *ops)
493 {
494 	return __svc_create(prog, bufsize, /*npools*/1, ops);
495 }
496 EXPORT_SYMBOL_GPL(svc_create);
497 
498 struct svc_serv *
svc_create_pooled(struct svc_program * prog,unsigned int bufsize,const struct svc_serv_ops * ops)499 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
500 		  const struct svc_serv_ops *ops)
501 {
502 	struct svc_serv *serv;
503 	unsigned int npools = svc_pool_map_get();
504 
505 	serv = __svc_create(prog, bufsize, npools, ops);
506 	if (!serv)
507 		goto out_err;
508 	return serv;
509 out_err:
510 	svc_pool_map_put();
511 	return NULL;
512 }
513 EXPORT_SYMBOL_GPL(svc_create_pooled);
514 
svc_shutdown_net(struct svc_serv * serv,struct net * net)515 void svc_shutdown_net(struct svc_serv *serv, struct net *net)
516 {
517 	svc_close_net(serv, net);
518 
519 	if (serv->sv_ops->svo_shutdown)
520 		serv->sv_ops->svo_shutdown(serv, net);
521 }
522 EXPORT_SYMBOL_GPL(svc_shutdown_net);
523 
524 /*
525  * Destroy an RPC service. Should be called with appropriate locking to
526  * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
527  */
528 void
svc_destroy(struct svc_serv * serv)529 svc_destroy(struct svc_serv *serv)
530 {
531 	dprintk("svc: svc_destroy(%s, %d)\n",
532 				serv->sv_program->pg_name,
533 				serv->sv_nrthreads);
534 
535 	if (serv->sv_nrthreads) {
536 		if (--(serv->sv_nrthreads) != 0) {
537 			svc_sock_update_bufs(serv);
538 			return;
539 		}
540 	} else
541 		printk("svc_destroy: no threads for serv=%p!\n", serv);
542 
543 	del_timer_sync(&serv->sv_temptimer);
544 
545 	/*
546 	 * The last user is gone and thus all sockets have to be destroyed to
547 	 * the point. Check this.
548 	 */
549 	BUG_ON(!list_empty(&serv->sv_permsocks));
550 	BUG_ON(!list_empty(&serv->sv_tempsocks));
551 
552 	cache_clean_deferred(serv);
553 
554 	if (svc_serv_is_pooled(serv))
555 		svc_pool_map_put();
556 
557 	kfree(serv->sv_pools);
558 	kfree(serv);
559 }
560 EXPORT_SYMBOL_GPL(svc_destroy);
561 
562 /*
563  * Allocate an RPC server's buffer space.
564  * We allocate pages and place them in rq_pages.
565  */
566 static int
svc_init_buffer(struct svc_rqst * rqstp,unsigned int size,int node)567 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
568 {
569 	unsigned int pages, arghi;
570 
571 	/* bc_xprt uses fore channel allocated buffers */
572 	if (svc_is_backchannel(rqstp))
573 		return 1;
574 
575 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
576 				       * We assume one is at most one page
577 				       */
578 	arghi = 0;
579 	WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
580 	if (pages > RPCSVC_MAXPAGES)
581 		pages = RPCSVC_MAXPAGES;
582 	while (pages) {
583 		struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
584 		if (!p)
585 			break;
586 		rqstp->rq_pages[arghi++] = p;
587 		pages--;
588 	}
589 	return pages == 0;
590 }
591 
592 /*
593  * Release an RPC server buffer
594  */
595 static void
svc_release_buffer(struct svc_rqst * rqstp)596 svc_release_buffer(struct svc_rqst *rqstp)
597 {
598 	unsigned int i;
599 
600 	for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
601 		if (rqstp->rq_pages[i])
602 			put_page(rqstp->rq_pages[i]);
603 }
604 
605 struct svc_rqst *
svc_rqst_alloc(struct svc_serv * serv,struct svc_pool * pool,int node)606 svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
607 {
608 	struct svc_rqst	*rqstp;
609 
610 	rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
611 	if (!rqstp)
612 		return rqstp;
613 
614 	__set_bit(RQ_BUSY, &rqstp->rq_flags);
615 	spin_lock_init(&rqstp->rq_lock);
616 	rqstp->rq_server = serv;
617 	rqstp->rq_pool = pool;
618 
619 	rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0);
620 	if (!rqstp->rq_scratch_page)
621 		goto out_enomem;
622 
623 	rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
624 	if (!rqstp->rq_argp)
625 		goto out_enomem;
626 
627 	rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
628 	if (!rqstp->rq_resp)
629 		goto out_enomem;
630 
631 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
632 		goto out_enomem;
633 
634 	return rqstp;
635 out_enomem:
636 	svc_rqst_free(rqstp);
637 	return NULL;
638 }
639 EXPORT_SYMBOL_GPL(svc_rqst_alloc);
640 
641 struct svc_rqst *
svc_prepare_thread(struct svc_serv * serv,struct svc_pool * pool,int node)642 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
643 {
644 	struct svc_rqst	*rqstp;
645 
646 	rqstp = svc_rqst_alloc(serv, pool, node);
647 	if (!rqstp)
648 		return ERR_PTR(-ENOMEM);
649 
650 	serv->sv_nrthreads++;
651 	spin_lock_bh(&pool->sp_lock);
652 	pool->sp_nrthreads++;
653 	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
654 	spin_unlock_bh(&pool->sp_lock);
655 	return rqstp;
656 }
657 EXPORT_SYMBOL_GPL(svc_prepare_thread);
658 
659 /*
660  * Choose a pool in which to create a new thread, for svc_set_num_threads
661  */
662 static inline struct svc_pool *
choose_pool(struct svc_serv * serv,struct svc_pool * pool,unsigned int * state)663 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
664 {
665 	if (pool != NULL)
666 		return pool;
667 
668 	return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
669 }
670 
671 /*
672  * Choose a thread to kill, for svc_set_num_threads
673  */
674 static inline struct task_struct *
choose_victim(struct svc_serv * serv,struct svc_pool * pool,unsigned int * state)675 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
676 {
677 	unsigned int i;
678 	struct task_struct *task = NULL;
679 
680 	if (pool != NULL) {
681 		spin_lock_bh(&pool->sp_lock);
682 	} else {
683 		/* choose a pool in round-robin fashion */
684 		for (i = 0; i < serv->sv_nrpools; i++) {
685 			pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
686 			spin_lock_bh(&pool->sp_lock);
687 			if (!list_empty(&pool->sp_all_threads))
688 				goto found_pool;
689 			spin_unlock_bh(&pool->sp_lock);
690 		}
691 		return NULL;
692 	}
693 
694 found_pool:
695 	if (!list_empty(&pool->sp_all_threads)) {
696 		struct svc_rqst *rqstp;
697 
698 		/*
699 		 * Remove from the pool->sp_all_threads list
700 		 * so we don't try to kill it again.
701 		 */
702 		rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
703 		set_bit(RQ_VICTIM, &rqstp->rq_flags);
704 		list_del_rcu(&rqstp->rq_all);
705 		task = rqstp->rq_task;
706 	}
707 	spin_unlock_bh(&pool->sp_lock);
708 
709 	return task;
710 }
711 
712 /* create new threads */
713 static int
svc_start_kthreads(struct svc_serv * serv,struct svc_pool * pool,int nrservs)714 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
715 {
716 	struct svc_rqst	*rqstp;
717 	struct task_struct *task;
718 	struct svc_pool *chosen_pool;
719 	unsigned int state = serv->sv_nrthreads-1;
720 	int node;
721 
722 	do {
723 		nrservs--;
724 		chosen_pool = choose_pool(serv, pool, &state);
725 
726 		node = svc_pool_map_get_node(chosen_pool->sp_id);
727 		rqstp = svc_prepare_thread(serv, chosen_pool, node);
728 		if (IS_ERR(rqstp))
729 			return PTR_ERR(rqstp);
730 
731 		__module_get(serv->sv_ops->svo_module);
732 		task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp,
733 					      node, "%s", serv->sv_name);
734 		if (IS_ERR(task)) {
735 			module_put(serv->sv_ops->svo_module);
736 			svc_exit_thread(rqstp);
737 			return PTR_ERR(task);
738 		}
739 
740 		rqstp->rq_task = task;
741 		if (serv->sv_nrpools > 1)
742 			svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
743 
744 		svc_sock_update_bufs(serv);
745 		wake_up_process(task);
746 	} while (nrservs > 0);
747 
748 	return 0;
749 }
750 
751 
752 /* destroy old threads */
753 static int
svc_signal_kthreads(struct svc_serv * serv,struct svc_pool * pool,int nrservs)754 svc_signal_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
755 {
756 	struct task_struct *task;
757 	unsigned int state = serv->sv_nrthreads-1;
758 
759 	/* destroy old threads */
760 	do {
761 		task = choose_victim(serv, pool, &state);
762 		if (task == NULL)
763 			break;
764 		send_sig(SIGINT, task, 1);
765 		nrservs++;
766 	} while (nrservs < 0);
767 
768 	return 0;
769 }
770 
771 /*
772  * Create or destroy enough new threads to make the number
773  * of threads the given number.  If `pool' is non-NULL, applies
774  * only to threads in that pool, otherwise round-robins between
775  * all pools.  Caller must ensure that mutual exclusion between this and
776  * server startup or shutdown.
777  *
778  * Destroying threads relies on the service threads filling in
779  * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
780  * has been created using svc_create_pooled().
781  *
782  * Based on code that used to be in nfsd_svc() but tweaked
783  * to be pool-aware.
784  */
785 int
svc_set_num_threads(struct svc_serv * serv,struct svc_pool * pool,int nrservs)786 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
787 {
788 	if (pool == NULL) {
789 		/* The -1 assumes caller has done a svc_get() */
790 		nrservs -= (serv->sv_nrthreads-1);
791 	} else {
792 		spin_lock_bh(&pool->sp_lock);
793 		nrservs -= pool->sp_nrthreads;
794 		spin_unlock_bh(&pool->sp_lock);
795 	}
796 
797 	if (nrservs > 0)
798 		return svc_start_kthreads(serv, pool, nrservs);
799 	if (nrservs < 0)
800 		return svc_signal_kthreads(serv, pool, nrservs);
801 	return 0;
802 }
803 EXPORT_SYMBOL_GPL(svc_set_num_threads);
804 
805 /* destroy old threads */
806 static int
svc_stop_kthreads(struct svc_serv * serv,struct svc_pool * pool,int nrservs)807 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
808 {
809 	struct svc_rqst	*rqstp;
810 	struct task_struct *task;
811 	unsigned int state = serv->sv_nrthreads-1;
812 
813 	/* destroy old threads */
814 	do {
815 		task = choose_victim(serv, pool, &state);
816 		if (task == NULL)
817 			break;
818 		rqstp = kthread_data(task);
819 		/* Did we lose a race to svo_function threadfn? */
820 		if (kthread_stop(task) == -EINTR)
821 			svc_exit_thread(rqstp);
822 		nrservs++;
823 	} while (nrservs < 0);
824 	return 0;
825 }
826 
827 int
svc_set_num_threads_sync(struct svc_serv * serv,struct svc_pool * pool,int nrservs)828 svc_set_num_threads_sync(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
829 {
830 	if (pool == NULL) {
831 		/* The -1 assumes caller has done a svc_get() */
832 		nrservs -= (serv->sv_nrthreads-1);
833 	} else {
834 		spin_lock_bh(&pool->sp_lock);
835 		nrservs -= pool->sp_nrthreads;
836 		spin_unlock_bh(&pool->sp_lock);
837 	}
838 
839 	if (nrservs > 0)
840 		return svc_start_kthreads(serv, pool, nrservs);
841 	if (nrservs < 0)
842 		return svc_stop_kthreads(serv, pool, nrservs);
843 	return 0;
844 }
845 EXPORT_SYMBOL_GPL(svc_set_num_threads_sync);
846 
847 /**
848  * svc_rqst_replace_page - Replace one page in rq_pages[]
849  * @rqstp: svc_rqst with pages to replace
850  * @page: replacement page
851  *
852  * When replacing a page in rq_pages, batch the release of the
853  * replaced pages to avoid hammering the page allocator.
854  */
svc_rqst_replace_page(struct svc_rqst * rqstp,struct page * page)855 void svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page)
856 {
857 	if (*rqstp->rq_next_page) {
858 		if (!pagevec_space(&rqstp->rq_pvec))
859 			__pagevec_release(&rqstp->rq_pvec);
860 		pagevec_add(&rqstp->rq_pvec, *rqstp->rq_next_page);
861 	}
862 
863 	get_page(page);
864 	*(rqstp->rq_next_page++) = page;
865 }
866 EXPORT_SYMBOL_GPL(svc_rqst_replace_page);
867 
868 /*
869  * Called from a server thread as it's exiting. Caller must hold the "service
870  * mutex" for the service.
871  */
872 void
svc_rqst_free(struct svc_rqst * rqstp)873 svc_rqst_free(struct svc_rqst *rqstp)
874 {
875 	svc_release_buffer(rqstp);
876 	if (rqstp->rq_scratch_page)
877 		put_page(rqstp->rq_scratch_page);
878 	kfree(rqstp->rq_resp);
879 	kfree(rqstp->rq_argp);
880 	kfree(rqstp->rq_auth_data);
881 	kfree_rcu(rqstp, rq_rcu_head);
882 }
883 EXPORT_SYMBOL_GPL(svc_rqst_free);
884 
885 void
svc_exit_thread(struct svc_rqst * rqstp)886 svc_exit_thread(struct svc_rqst *rqstp)
887 {
888 	struct svc_serv	*serv = rqstp->rq_server;
889 	struct svc_pool	*pool = rqstp->rq_pool;
890 
891 	spin_lock_bh(&pool->sp_lock);
892 	pool->sp_nrthreads--;
893 	if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags))
894 		list_del_rcu(&rqstp->rq_all);
895 	spin_unlock_bh(&pool->sp_lock);
896 
897 	svc_rqst_free(rqstp);
898 
899 	/* Release the server */
900 	if (serv)
901 		svc_destroy(serv);
902 }
903 EXPORT_SYMBOL_GPL(svc_exit_thread);
904 
905 /*
906  * Register an "inet" protocol family netid with the local
907  * rpcbind daemon via an rpcbind v4 SET request.
908  *
909  * No netconfig infrastructure is available in the kernel, so
910  * we map IP_ protocol numbers to netids by hand.
911  *
912  * Returns zero on success; a negative errno value is returned
913  * if any error occurs.
914  */
__svc_rpcb_register4(struct net * net,const u32 program,const u32 version,const unsigned short protocol,const unsigned short port)915 static int __svc_rpcb_register4(struct net *net, const u32 program,
916 				const u32 version,
917 				const unsigned short protocol,
918 				const unsigned short port)
919 {
920 	const struct sockaddr_in sin = {
921 		.sin_family		= AF_INET,
922 		.sin_addr.s_addr	= htonl(INADDR_ANY),
923 		.sin_port		= htons(port),
924 	};
925 	const char *netid;
926 	int error;
927 
928 	switch (protocol) {
929 	case IPPROTO_UDP:
930 		netid = RPCBIND_NETID_UDP;
931 		break;
932 	case IPPROTO_TCP:
933 		netid = RPCBIND_NETID_TCP;
934 		break;
935 	default:
936 		return -ENOPROTOOPT;
937 	}
938 
939 	error = rpcb_v4_register(net, program, version,
940 					(const struct sockaddr *)&sin, netid);
941 
942 	/*
943 	 * User space didn't support rpcbind v4, so retry this
944 	 * registration request with the legacy rpcbind v2 protocol.
945 	 */
946 	if (error == -EPROTONOSUPPORT)
947 		error = rpcb_register(net, program, version, protocol, port);
948 
949 	return error;
950 }
951 
952 #if IS_ENABLED(CONFIG_IPV6)
953 /*
954  * Register an "inet6" protocol family netid with the local
955  * rpcbind daemon via an rpcbind v4 SET request.
956  *
957  * No netconfig infrastructure is available in the kernel, so
958  * we map IP_ protocol numbers to netids by hand.
959  *
960  * Returns zero on success; a negative errno value is returned
961  * if any error occurs.
962  */
__svc_rpcb_register6(struct net * net,const u32 program,const u32 version,const unsigned short protocol,const unsigned short port)963 static int __svc_rpcb_register6(struct net *net, const u32 program,
964 				const u32 version,
965 				const unsigned short protocol,
966 				const unsigned short port)
967 {
968 	const struct sockaddr_in6 sin6 = {
969 		.sin6_family		= AF_INET6,
970 		.sin6_addr		= IN6ADDR_ANY_INIT,
971 		.sin6_port		= htons(port),
972 	};
973 	const char *netid;
974 	int error;
975 
976 	switch (protocol) {
977 	case IPPROTO_UDP:
978 		netid = RPCBIND_NETID_UDP6;
979 		break;
980 	case IPPROTO_TCP:
981 		netid = RPCBIND_NETID_TCP6;
982 		break;
983 	default:
984 		return -ENOPROTOOPT;
985 	}
986 
987 	error = rpcb_v4_register(net, program, version,
988 					(const struct sockaddr *)&sin6, netid);
989 
990 	/*
991 	 * User space didn't support rpcbind version 4, so we won't
992 	 * use a PF_INET6 listener.
993 	 */
994 	if (error == -EPROTONOSUPPORT)
995 		error = -EAFNOSUPPORT;
996 
997 	return error;
998 }
999 #endif	/* IS_ENABLED(CONFIG_IPV6) */
1000 
1001 /*
1002  * Register a kernel RPC service via rpcbind version 4.
1003  *
1004  * Returns zero on success; a negative errno value is returned
1005  * if any error occurs.
1006  */
__svc_register(struct net * net,const char * progname,const u32 program,const u32 version,const int family,const unsigned short protocol,const unsigned short port)1007 static int __svc_register(struct net *net, const char *progname,
1008 			  const u32 program, const u32 version,
1009 			  const int family,
1010 			  const unsigned short protocol,
1011 			  const unsigned short port)
1012 {
1013 	int error = -EAFNOSUPPORT;
1014 
1015 	switch (family) {
1016 	case PF_INET:
1017 		error = __svc_rpcb_register4(net, program, version,
1018 						protocol, port);
1019 		break;
1020 #if IS_ENABLED(CONFIG_IPV6)
1021 	case PF_INET6:
1022 		error = __svc_rpcb_register6(net, program, version,
1023 						protocol, port);
1024 #endif
1025 	}
1026 
1027 	trace_svc_register(progname, version, family, protocol, port, error);
1028 	return error;
1029 }
1030 
svc_rpcbind_set_version(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)1031 int svc_rpcbind_set_version(struct net *net,
1032 			    const struct svc_program *progp,
1033 			    u32 version, int family,
1034 			    unsigned short proto,
1035 			    unsigned short port)
1036 {
1037 	return __svc_register(net, progp->pg_name, progp->pg_prog,
1038 				version, family, proto, port);
1039 
1040 }
1041 EXPORT_SYMBOL_GPL(svc_rpcbind_set_version);
1042 
svc_generic_rpcbind_set(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)1043 int svc_generic_rpcbind_set(struct net *net,
1044 			    const struct svc_program *progp,
1045 			    u32 version, int family,
1046 			    unsigned short proto,
1047 			    unsigned short port)
1048 {
1049 	const struct svc_version *vers = progp->pg_vers[version];
1050 	int error;
1051 
1052 	if (vers == NULL)
1053 		return 0;
1054 
1055 	if (vers->vs_hidden) {
1056 		trace_svc_noregister(progp->pg_name, version, proto,
1057 				     port, family, 0);
1058 		return 0;
1059 	}
1060 
1061 	/*
1062 	 * Don't register a UDP port if we need congestion
1063 	 * control.
1064 	 */
1065 	if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP)
1066 		return 0;
1067 
1068 	error = svc_rpcbind_set_version(net, progp, version,
1069 					family, proto, port);
1070 
1071 	return (vers->vs_rpcb_optnl) ? 0 : error;
1072 }
1073 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set);
1074 
1075 /**
1076  * svc_register - register an RPC service with the local portmapper
1077  * @serv: svc_serv struct for the service to register
1078  * @net: net namespace for the service to register
1079  * @family: protocol family of service's listener socket
1080  * @proto: transport protocol number to advertise
1081  * @port: port to advertise
1082  *
1083  * Service is registered for any address in the passed-in protocol family
1084  */
svc_register(const struct svc_serv * serv,struct net * net,const int family,const unsigned short proto,const unsigned short port)1085 int svc_register(const struct svc_serv *serv, struct net *net,
1086 		 const int family, const unsigned short proto,
1087 		 const unsigned short port)
1088 {
1089 	struct svc_program	*progp;
1090 	unsigned int		i;
1091 	int			error = 0;
1092 
1093 	WARN_ON_ONCE(proto == 0 && port == 0);
1094 	if (proto == 0 && port == 0)
1095 		return -EINVAL;
1096 
1097 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1098 		for (i = 0; i < progp->pg_nvers; i++) {
1099 
1100 			error = progp->pg_rpcbind_set(net, progp, i,
1101 					family, proto, port);
1102 			if (error < 0) {
1103 				printk(KERN_WARNING "svc: failed to register "
1104 					"%sv%u RPC service (errno %d).\n",
1105 					progp->pg_name, i, -error);
1106 				break;
1107 			}
1108 		}
1109 	}
1110 
1111 	return error;
1112 }
1113 
1114 /*
1115  * If user space is running rpcbind, it should take the v4 UNSET
1116  * and clear everything for this [program, version].  If user space
1117  * is running portmap, it will reject the v4 UNSET, but won't have
1118  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
1119  * in this case to clear all existing entries for [program, version].
1120  */
__svc_unregister(struct net * net,const u32 program,const u32 version,const char * progname)1121 static void __svc_unregister(struct net *net, const u32 program, const u32 version,
1122 			     const char *progname)
1123 {
1124 	int error;
1125 
1126 	error = rpcb_v4_register(net, program, version, NULL, "");
1127 
1128 	/*
1129 	 * User space didn't support rpcbind v4, so retry this
1130 	 * request with the legacy rpcbind v2 protocol.
1131 	 */
1132 	if (error == -EPROTONOSUPPORT)
1133 		error = rpcb_register(net, program, version, 0, 0);
1134 
1135 	trace_svc_unregister(progname, version, error);
1136 }
1137 
1138 /*
1139  * All netids, bind addresses and ports registered for [program, version]
1140  * are removed from the local rpcbind database (if the service is not
1141  * hidden) to make way for a new instance of the service.
1142  *
1143  * The result of unregistration is reported via dprintk for those who want
1144  * verification of the result, but is otherwise not important.
1145  */
svc_unregister(const struct svc_serv * serv,struct net * net)1146 static void svc_unregister(const struct svc_serv *serv, struct net *net)
1147 {
1148 	struct svc_program *progp;
1149 	unsigned long flags;
1150 	unsigned int i;
1151 
1152 	clear_thread_flag(TIF_SIGPENDING);
1153 
1154 	for (progp = serv->sv_program; progp; progp = progp->pg_next) {
1155 		for (i = 0; i < progp->pg_nvers; i++) {
1156 			if (progp->pg_vers[i] == NULL)
1157 				continue;
1158 			if (progp->pg_vers[i]->vs_hidden)
1159 				continue;
1160 			__svc_unregister(net, progp->pg_prog, i, progp->pg_name);
1161 		}
1162 	}
1163 
1164 	spin_lock_irqsave(&current->sighand->siglock, flags);
1165 	recalc_sigpending();
1166 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
1167 }
1168 
1169 /*
1170  * dprintk the given error with the address of the client that caused it.
1171  */
1172 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1173 static __printf(2, 3)
svc_printk(struct svc_rqst * rqstp,const char * fmt,...)1174 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
1175 {
1176 	struct va_format vaf;
1177 	va_list args;
1178 	char 	buf[RPC_MAX_ADDRBUFLEN];
1179 
1180 	va_start(args, fmt);
1181 
1182 	vaf.fmt = fmt;
1183 	vaf.va = &args;
1184 
1185 	dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf);
1186 
1187 	va_end(args);
1188 }
1189 #else
svc_printk(struct svc_rqst * rqstp,const char * fmt,...)1190 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {}
1191 #endif
1192 
1193 static int
svc_generic_dispatch(struct svc_rqst * rqstp,__be32 * statp)1194 svc_generic_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1195 {
1196 	struct kvec *argv = &rqstp->rq_arg.head[0];
1197 	struct kvec *resv = &rqstp->rq_res.head[0];
1198 	const struct svc_procedure *procp = rqstp->rq_procinfo;
1199 
1200 	/*
1201 	 * Decode arguments
1202 	 * XXX: why do we ignore the return value?
1203 	 */
1204 	if (procp->pc_decode &&
1205 	    !procp->pc_decode(rqstp, argv->iov_base)) {
1206 		*statp = rpc_garbage_args;
1207 		return 1;
1208 	}
1209 
1210 	*statp = procp->pc_func(rqstp);
1211 
1212 	if (*statp == rpc_drop_reply ||
1213 	    test_bit(RQ_DROPME, &rqstp->rq_flags))
1214 		return 0;
1215 
1216 	if (rqstp->rq_auth_stat != rpc_auth_ok)
1217 		return 1;
1218 
1219 	if (*statp != rpc_success)
1220 		return 1;
1221 
1222 	/* Encode reply */
1223 	if (procp->pc_encode &&
1224 	    !procp->pc_encode(rqstp, resv->iov_base + resv->iov_len)) {
1225 		dprintk("svc: failed to encode reply\n");
1226 		/* serv->sv_stats->rpcsystemerr++; */
1227 		*statp = rpc_system_err;
1228 	}
1229 	return 1;
1230 }
1231 
1232 __be32
svc_generic_init_request(struct svc_rqst * rqstp,const struct svc_program * progp,struct svc_process_info * ret)1233 svc_generic_init_request(struct svc_rqst *rqstp,
1234 		const struct svc_program *progp,
1235 		struct svc_process_info *ret)
1236 {
1237 	const struct svc_version *versp = NULL;	/* compiler food */
1238 	const struct svc_procedure *procp = NULL;
1239 
1240 	if (rqstp->rq_vers >= progp->pg_nvers )
1241 		goto err_bad_vers;
1242 	versp = progp->pg_vers[rqstp->rq_vers];
1243 	if (!versp)
1244 		goto err_bad_vers;
1245 
1246 	/*
1247 	 * Some protocol versions (namely NFSv4) require some form of
1248 	 * congestion control.  (See RFC 7530 section 3.1 paragraph 2)
1249 	 * In other words, UDP is not allowed. We mark those when setting
1250 	 * up the svc_xprt, and verify that here.
1251 	 *
1252 	 * The spec is not very clear about what error should be returned
1253 	 * when someone tries to access a server that is listening on UDP
1254 	 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1255 	 * fit.
1256 	 */
1257 	if (versp->vs_need_cong_ctrl && rqstp->rq_xprt &&
1258 	    !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags))
1259 		goto err_bad_vers;
1260 
1261 	if (rqstp->rq_proc >= versp->vs_nproc)
1262 		goto err_bad_proc;
1263 	rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc];
1264 	if (!procp)
1265 		goto err_bad_proc;
1266 
1267 	/* Initialize storage for argp and resp */
1268 	memset(rqstp->rq_argp, 0, procp->pc_argsize);
1269 	memset(rqstp->rq_resp, 0, procp->pc_ressize);
1270 
1271 	/* Bump per-procedure stats counter */
1272 	versp->vs_count[rqstp->rq_proc]++;
1273 
1274 	ret->dispatch = versp->vs_dispatch;
1275 	return rpc_success;
1276 err_bad_vers:
1277 	ret->mismatch.lovers = progp->pg_lovers;
1278 	ret->mismatch.hivers = progp->pg_hivers;
1279 	return rpc_prog_mismatch;
1280 err_bad_proc:
1281 	return rpc_proc_unavail;
1282 }
1283 EXPORT_SYMBOL_GPL(svc_generic_init_request);
1284 
1285 /*
1286  * Common routine for processing the RPC request.
1287  */
1288 static int
svc_process_common(struct svc_rqst * rqstp,struct kvec * argv,struct kvec * resv)1289 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
1290 {
1291 	struct svc_program	*progp;
1292 	const struct svc_procedure *procp = NULL;
1293 	struct svc_serv		*serv = rqstp->rq_server;
1294 	struct svc_process_info process;
1295 	__be32			*statp;
1296 	u32			prog, vers;
1297 	__be32			rpc_stat;
1298 	int			auth_res;
1299 	__be32			*reply_statp;
1300 
1301 	rpc_stat = rpc_success;
1302 
1303 	if (argv->iov_len < 6*4)
1304 		goto err_short_len;
1305 
1306 	/* Will be turned off by GSS integrity and privacy services */
1307 	set_bit(RQ_SPLICE_OK, &rqstp->rq_flags);
1308 	/* Will be turned off only when NFSv4 Sessions are used */
1309 	set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags);
1310 	clear_bit(RQ_DROPME, &rqstp->rq_flags);
1311 
1312 	svc_putu32(resv, rqstp->rq_xid);
1313 
1314 	vers = svc_getnl(argv);
1315 
1316 	/* First words of reply: */
1317 	svc_putnl(resv, 1);		/* REPLY */
1318 
1319 	if (vers != 2)		/* RPC version number */
1320 		goto err_bad_rpc;
1321 
1322 	/* Save position in case we later decide to reject: */
1323 	reply_statp = resv->iov_base + resv->iov_len;
1324 
1325 	svc_putnl(resv, 0);		/* ACCEPT */
1326 
1327 	rqstp->rq_prog = prog = svc_getnl(argv);	/* program number */
1328 	rqstp->rq_vers = svc_getnl(argv);	/* version number */
1329 	rqstp->rq_proc = svc_getnl(argv);	/* procedure number */
1330 
1331 	for (progp = serv->sv_program; progp; progp = progp->pg_next)
1332 		if (prog == progp->pg_prog)
1333 			break;
1334 
1335 	/*
1336 	 * Decode auth data, and add verifier to reply buffer.
1337 	 * We do this before anything else in order to get a decent
1338 	 * auth verifier.
1339 	 */
1340 	auth_res = svc_authenticate(rqstp);
1341 	/* Also give the program a chance to reject this call: */
1342 	if (auth_res == SVC_OK && progp)
1343 		auth_res = progp->pg_authenticate(rqstp);
1344 	if (auth_res != SVC_OK)
1345 		trace_svc_authenticate(rqstp, auth_res);
1346 	switch (auth_res) {
1347 	case SVC_OK:
1348 		break;
1349 	case SVC_GARBAGE:
1350 		goto err_garbage;
1351 	case SVC_SYSERR:
1352 		rpc_stat = rpc_system_err;
1353 		goto err_bad;
1354 	case SVC_DENIED:
1355 		goto err_bad_auth;
1356 	case SVC_CLOSE:
1357 		goto close;
1358 	case SVC_DROP:
1359 		goto dropit;
1360 	case SVC_COMPLETE:
1361 		goto sendit;
1362 	}
1363 
1364 	if (progp == NULL)
1365 		goto err_bad_prog;
1366 
1367 	rpc_stat = progp->pg_init_request(rqstp, progp, &process);
1368 	switch (rpc_stat) {
1369 	case rpc_success:
1370 		break;
1371 	case rpc_prog_unavail:
1372 		goto err_bad_prog;
1373 	case rpc_prog_mismatch:
1374 		goto err_bad_vers;
1375 	case rpc_proc_unavail:
1376 		goto err_bad_proc;
1377 	}
1378 
1379 	procp = rqstp->rq_procinfo;
1380 	/* Should this check go into the dispatcher? */
1381 	if (!procp || !procp->pc_func)
1382 		goto err_bad_proc;
1383 
1384 	/* Syntactic check complete */
1385 	serv->sv_stats->rpccnt++;
1386 	trace_svc_process(rqstp, progp->pg_name);
1387 
1388 	/* Build the reply header. */
1389 	statp = resv->iov_base +resv->iov_len;
1390 	svc_putnl(resv, RPC_SUCCESS);
1391 
1392 	/* un-reserve some of the out-queue now that we have a
1393 	 * better idea of reply size
1394 	 */
1395 	if (procp->pc_xdrressize)
1396 		svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1397 
1398 	/* Call the function that processes the request. */
1399 	if (!process.dispatch) {
1400 		if (!svc_generic_dispatch(rqstp, statp))
1401 			goto release_dropit;
1402 		if (*statp == rpc_garbage_args)
1403 			goto err_garbage;
1404 	} else {
1405 		dprintk("svc: calling dispatcher\n");
1406 		if (!process.dispatch(rqstp, statp))
1407 			goto release_dropit; /* Release reply info */
1408 	}
1409 
1410 	if (rqstp->rq_auth_stat != rpc_auth_ok)
1411 		goto err_release_bad_auth;
1412 
1413 	/* Check RPC status result */
1414 	if (*statp != rpc_success)
1415 		resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1416 
1417 	/* Release reply info */
1418 	if (procp->pc_release)
1419 		procp->pc_release(rqstp);
1420 
1421 	if (procp->pc_encode == NULL)
1422 		goto dropit;
1423 
1424  sendit:
1425 	if (svc_authorise(rqstp))
1426 		goto close_xprt;
1427 	return 1;		/* Caller can now send it */
1428 
1429 release_dropit:
1430 	if (procp->pc_release)
1431 		procp->pc_release(rqstp);
1432  dropit:
1433 	svc_authorise(rqstp);	/* doesn't hurt to call this twice */
1434 	dprintk("svc: svc_process dropit\n");
1435 	return 0;
1436 
1437  close:
1438 	svc_authorise(rqstp);
1439 close_xprt:
1440 	if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1441 		svc_close_xprt(rqstp->rq_xprt);
1442 	dprintk("svc: svc_process close\n");
1443 	return 0;
1444 
1445 err_short_len:
1446 	svc_printk(rqstp, "short len %zd, dropping request\n",
1447 			argv->iov_len);
1448 	goto close_xprt;
1449 
1450 err_bad_rpc:
1451 	serv->sv_stats->rpcbadfmt++;
1452 	svc_putnl(resv, 1);	/* REJECT */
1453 	svc_putnl(resv, 0);	/* RPC_MISMATCH */
1454 	svc_putnl(resv, 2);	/* Only RPCv2 supported */
1455 	svc_putnl(resv, 2);
1456 	goto sendit;
1457 
1458 err_release_bad_auth:
1459 	if (procp->pc_release)
1460 		procp->pc_release(rqstp);
1461 err_bad_auth:
1462 	dprintk("svc: authentication failed (%d)\n",
1463 		be32_to_cpu(rqstp->rq_auth_stat));
1464 	serv->sv_stats->rpcbadauth++;
1465 	/* Restore write pointer to location of accept status: */
1466 	xdr_ressize_check(rqstp, reply_statp);
1467 	svc_putnl(resv, 1);	/* REJECT */
1468 	svc_putnl(resv, 1);	/* AUTH_ERROR */
1469 	svc_putu32(resv, rqstp->rq_auth_stat);	/* status */
1470 	goto sendit;
1471 
1472 err_bad_prog:
1473 	dprintk("svc: unknown program %d\n", prog);
1474 	serv->sv_stats->rpcbadfmt++;
1475 	svc_putnl(resv, RPC_PROG_UNAVAIL);
1476 	goto sendit;
1477 
1478 err_bad_vers:
1479 	svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1480 		       rqstp->rq_vers, rqstp->rq_prog, progp->pg_name);
1481 
1482 	serv->sv_stats->rpcbadfmt++;
1483 	svc_putnl(resv, RPC_PROG_MISMATCH);
1484 	svc_putnl(resv, process.mismatch.lovers);
1485 	svc_putnl(resv, process.mismatch.hivers);
1486 	goto sendit;
1487 
1488 err_bad_proc:
1489 	svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc);
1490 
1491 	serv->sv_stats->rpcbadfmt++;
1492 	svc_putnl(resv, RPC_PROC_UNAVAIL);
1493 	goto sendit;
1494 
1495 err_garbage:
1496 	svc_printk(rqstp, "failed to decode args\n");
1497 
1498 	rpc_stat = rpc_garbage_args;
1499 err_bad:
1500 	serv->sv_stats->rpcbadfmt++;
1501 	svc_putnl(resv, ntohl(rpc_stat));
1502 	goto sendit;
1503 }
1504 
1505 /*
1506  * Process the RPC request.
1507  */
1508 int
svc_process(struct svc_rqst * rqstp)1509 svc_process(struct svc_rqst *rqstp)
1510 {
1511 	struct kvec		*argv = &rqstp->rq_arg.head[0];
1512 	struct kvec		*resv = &rqstp->rq_res.head[0];
1513 	struct svc_serv		*serv = rqstp->rq_server;
1514 	u32			dir;
1515 
1516 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
1517 	if (!fail_sunrpc.ignore_server_disconnect &&
1518 	    should_fail(&fail_sunrpc.attr, 1))
1519 		svc_xprt_deferred_close(rqstp->rq_xprt);
1520 #endif
1521 
1522 	/*
1523 	 * Setup response xdr_buf.
1524 	 * Initially it has just one page
1525 	 */
1526 	rqstp->rq_next_page = &rqstp->rq_respages[1];
1527 	resv->iov_base = page_address(rqstp->rq_respages[0]);
1528 	resv->iov_len = 0;
1529 	rqstp->rq_res.pages = rqstp->rq_respages + 1;
1530 	rqstp->rq_res.len = 0;
1531 	rqstp->rq_res.page_base = 0;
1532 	rqstp->rq_res.page_len = 0;
1533 	rqstp->rq_res.buflen = PAGE_SIZE;
1534 	rqstp->rq_res.tail[0].iov_base = NULL;
1535 	rqstp->rq_res.tail[0].iov_len = 0;
1536 
1537 	dir  = svc_getnl(argv);
1538 	if (dir != 0) {
1539 		/* direction != CALL */
1540 		svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1541 		serv->sv_stats->rpcbadfmt++;
1542 		goto out_drop;
1543 	}
1544 
1545 	/* Returns 1 for send, 0 for drop */
1546 	if (likely(svc_process_common(rqstp, argv, resv)))
1547 		return svc_send(rqstp);
1548 
1549 out_drop:
1550 	svc_drop(rqstp);
1551 	return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(svc_process);
1554 
1555 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1556 /*
1557  * Process a backchannel RPC request that arrived over an existing
1558  * outbound connection
1559  */
1560 int
bc_svc_process(struct svc_serv * serv,struct rpc_rqst * req,struct svc_rqst * rqstp)1561 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1562 	       struct svc_rqst *rqstp)
1563 {
1564 	struct kvec	*argv = &rqstp->rq_arg.head[0];
1565 	struct kvec	*resv = &rqstp->rq_res.head[0];
1566 	struct rpc_task *task;
1567 	int proc_error;
1568 	int error;
1569 
1570 	dprintk("svc: %s(%p)\n", __func__, req);
1571 
1572 	/* Build the svc_rqst used by the common processing routine */
1573 	rqstp->rq_xid = req->rq_xid;
1574 	rqstp->rq_prot = req->rq_xprt->prot;
1575 	rqstp->rq_server = serv;
1576 	rqstp->rq_bc_net = req->rq_xprt->xprt_net;
1577 
1578 	rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1579 	memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1580 	memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1581 	memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1582 
1583 	/* Adjust the argument buffer length */
1584 	rqstp->rq_arg.len = req->rq_private_buf.len;
1585 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1586 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1587 		rqstp->rq_arg.page_len = 0;
1588 	} else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len +
1589 			rqstp->rq_arg.page_len)
1590 		rqstp->rq_arg.page_len = rqstp->rq_arg.len -
1591 			rqstp->rq_arg.head[0].iov_len;
1592 	else
1593 		rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len +
1594 			rqstp->rq_arg.page_len;
1595 
1596 	/* reset result send buffer "put" position */
1597 	resv->iov_len = 0;
1598 
1599 	/*
1600 	 * Skip the next two words because they've already been
1601 	 * processed in the transport
1602 	 */
1603 	svc_getu32(argv);	/* XID */
1604 	svc_getnl(argv);	/* CALLDIR */
1605 
1606 	/* Parse and execute the bc call */
1607 	proc_error = svc_process_common(rqstp, argv, resv);
1608 
1609 	atomic_dec(&req->rq_xprt->bc_slot_count);
1610 	if (!proc_error) {
1611 		/* Processing error: drop the request */
1612 		xprt_free_bc_request(req);
1613 		error = -EINVAL;
1614 		goto out;
1615 	}
1616 	/* Finally, send the reply synchronously */
1617 	memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1618 	task = rpc_run_bc_task(req);
1619 	if (IS_ERR(task)) {
1620 		error = PTR_ERR(task);
1621 		goto out;
1622 	}
1623 
1624 	WARN_ON_ONCE(atomic_read(&task->tk_count) != 1);
1625 	error = task->tk_status;
1626 	rpc_put_task(task);
1627 
1628 out:
1629 	dprintk("svc: %s(), error=%d\n", __func__, error);
1630 	return error;
1631 }
1632 EXPORT_SYMBOL_GPL(bc_svc_process);
1633 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1634 
1635 /*
1636  * Return (transport-specific) limit on the rpc payload.
1637  */
svc_max_payload(const struct svc_rqst * rqstp)1638 u32 svc_max_payload(const struct svc_rqst *rqstp)
1639 {
1640 	u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1641 
1642 	if (rqstp->rq_server->sv_max_payload < max)
1643 		max = rqstp->rq_server->sv_max_payload;
1644 	return max;
1645 }
1646 EXPORT_SYMBOL_GPL(svc_max_payload);
1647 
1648 /**
1649  * svc_proc_name - Return RPC procedure name in string form
1650  * @rqstp: svc_rqst to operate on
1651  *
1652  * Return value:
1653  *   Pointer to a NUL-terminated string
1654  */
svc_proc_name(const struct svc_rqst * rqstp)1655 const char *svc_proc_name(const struct svc_rqst *rqstp)
1656 {
1657 	if (rqstp && rqstp->rq_procinfo)
1658 		return rqstp->rq_procinfo->pc_name;
1659 	return "unknown";
1660 }
1661 
1662 
1663 /**
1664  * svc_encode_result_payload - mark a range of bytes as a result payload
1665  * @rqstp: svc_rqst to operate on
1666  * @offset: payload's byte offset in rqstp->rq_res
1667  * @length: size of payload, in bytes
1668  *
1669  * Returns zero on success, or a negative errno if a permanent
1670  * error occurred.
1671  */
svc_encode_result_payload(struct svc_rqst * rqstp,unsigned int offset,unsigned int length)1672 int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1673 			      unsigned int length)
1674 {
1675 	return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset,
1676 							   length);
1677 }
1678 EXPORT_SYMBOL_GPL(svc_encode_result_payload);
1679 
1680 /**
1681  * svc_fill_write_vector - Construct data argument for VFS write call
1682  * @rqstp: svc_rqst to operate on
1683  * @payload: xdr_buf containing only the write data payload
1684  *
1685  * Fills in rqstp::rq_vec, and returns the number of elements.
1686  */
svc_fill_write_vector(struct svc_rqst * rqstp,struct xdr_buf * payload)1687 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
1688 				   struct xdr_buf *payload)
1689 {
1690 	struct page **pages = payload->pages;
1691 	struct kvec *first = payload->head;
1692 	struct kvec *vec = rqstp->rq_vec;
1693 	size_t total = payload->len;
1694 	unsigned int i;
1695 
1696 	/* Some types of transport can present the write payload
1697 	 * entirely in rq_arg.pages. In this case, @first is empty.
1698 	 */
1699 	i = 0;
1700 	if (first->iov_len) {
1701 		vec[i].iov_base = first->iov_base;
1702 		vec[i].iov_len = min_t(size_t, total, first->iov_len);
1703 		total -= vec[i].iov_len;
1704 		++i;
1705 	}
1706 
1707 	while (total) {
1708 		vec[i].iov_base = page_address(*pages);
1709 		vec[i].iov_len = min_t(size_t, total, PAGE_SIZE);
1710 		total -= vec[i].iov_len;
1711 		++i;
1712 		++pages;
1713 	}
1714 
1715 	WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec));
1716 	return i;
1717 }
1718 EXPORT_SYMBOL_GPL(svc_fill_write_vector);
1719 
1720 /**
1721  * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call
1722  * @rqstp: svc_rqst to operate on
1723  * @first: buffer containing first section of pathname
1724  * @p: buffer containing remaining section of pathname
1725  * @total: total length of the pathname argument
1726  *
1727  * The VFS symlink API demands a NUL-terminated pathname in mapped memory.
1728  * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free
1729  * the returned string.
1730  */
svc_fill_symlink_pathname(struct svc_rqst * rqstp,struct kvec * first,void * p,size_t total)1731 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first,
1732 				void *p, size_t total)
1733 {
1734 	size_t len, remaining;
1735 	char *result, *dst;
1736 
1737 	result = kmalloc(total + 1, GFP_KERNEL);
1738 	if (!result)
1739 		return ERR_PTR(-ESERVERFAULT);
1740 
1741 	dst = result;
1742 	remaining = total;
1743 
1744 	len = min_t(size_t, total, first->iov_len);
1745 	if (len) {
1746 		memcpy(dst, first->iov_base, len);
1747 		dst += len;
1748 		remaining -= len;
1749 	}
1750 
1751 	if (remaining) {
1752 		len = min_t(size_t, remaining, PAGE_SIZE);
1753 		memcpy(dst, p, len);
1754 		dst += len;
1755 	}
1756 
1757 	*dst = '\0';
1758 
1759 	/* Sanity check: Linux doesn't allow the pathname argument to
1760 	 * contain a NUL byte.
1761 	 */
1762 	if (strlen(result) != total) {
1763 		kfree(result);
1764 		return ERR_PTR(-EINVAL);
1765 	}
1766 	return result;
1767 }
1768 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname);
1769