blob: c2c8bb20d07f7171aad6a74e9c4cf323fd4f03ca [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Greg Banksbfd24162006-10-02 02:18:01 -07007 *
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
10 * by Greg Banks <gnb@melbourne.sgi.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
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>
Greg Banksa7455442006-10-02 02:17:59 -070019#include <linux/interrupt.h>
20#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <linux/sunrpc/types.h>
23#include <linux/sunrpc/xdr.h>
24#include <linux/sunrpc/stats.h>
25#include <linux/sunrpc/svcsock.h>
26#include <linux/sunrpc/clnt.h>
27
28#define RPCDBG_FACILITY RPCDBG_SVCDSP
29#define RPC_PARANOIA 1
30
31/*
Greg Banksbfd24162006-10-02 02:18:01 -070032 * Mode for mapping cpus to pools.
33 */
34enum {
35 SVC_POOL_NONE = -1, /* uninitialised, choose one of the others */
36 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
37 * (legacy & UP mode) */
38 SVC_POOL_PERCPU, /* one pool per cpu */
39 SVC_POOL_PERNODE /* one pool per numa node */
40};
41
42/*
43 * Structure for mapping cpus to pools and vice versa.
44 * Setup once during sunrpc initialisation.
45 */
46static struct svc_pool_map {
47 int mode; /* Note: int not enum to avoid
48 * warnings about "enumeration value
49 * not handled in switch" */
50 unsigned int npools;
51 unsigned int *pool_to; /* maps pool id to cpu or node */
52 unsigned int *to_pool; /* maps cpu or node to pool id */
53} svc_pool_map = {
54 .mode = SVC_POOL_NONE
55};
56
57
58/*
59 * Detect best pool mapping mode heuristically,
60 * according to the machine's topology.
61 */
62static int
63svc_pool_map_choose_mode(void)
64{
65 unsigned int node;
66
67 if (num_online_nodes() > 1) {
68 /*
69 * Actually have multiple NUMA nodes,
70 * so split pools on NUMA node boundaries
71 */
72 return SVC_POOL_PERNODE;
73 }
74
75 node = any_online_node(node_online_map);
76 if (nr_cpus_node(node) > 2) {
77 /*
78 * Non-trivial SMP, or CONFIG_NUMA on
79 * non-NUMA hardware, e.g. with a generic
80 * x86_64 kernel on Xeons. In this case we
81 * want to divide the pools on cpu boundaries.
82 */
83 return SVC_POOL_PERCPU;
84 }
85
86 /* default: one global pool */
87 return SVC_POOL_GLOBAL;
88}
89
90/*
91 * Allocate the to_pool[] and pool_to[] arrays.
92 * Returns 0 on success or an errno.
93 */
94static int
95svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
96{
97 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
98 if (!m->to_pool)
99 goto fail;
100 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
101 if (!m->pool_to)
102 goto fail_free;
103
104 return 0;
105
106fail_free:
107 kfree(m->to_pool);
108fail:
109 return -ENOMEM;
110}
111
112/*
113 * Initialise the pool map for SVC_POOL_PERCPU mode.
114 * Returns number of pools or <0 on error.
115 */
116static int
117svc_pool_map_init_percpu(struct svc_pool_map *m)
118{
119 unsigned int maxpools = highest_possible_processor_id()+1;
120 unsigned int pidx = 0;
121 unsigned int cpu;
122 int err;
123
124 err = svc_pool_map_alloc_arrays(m, maxpools);
125 if (err)
126 return err;
127
128 for_each_online_cpu(cpu) {
129 BUG_ON(pidx > maxpools);
130 m->to_pool[cpu] = pidx;
131 m->pool_to[pidx] = cpu;
132 pidx++;
133 }
134 /* cpus brought online later all get mapped to pool0, sorry */
135
136 return pidx;
137};
138
139
140/*
141 * Initialise the pool map for SVC_POOL_PERNODE mode.
142 * Returns number of pools or <0 on error.
143 */
144static int
145svc_pool_map_init_pernode(struct svc_pool_map *m)
146{
147 unsigned int maxpools = highest_possible_node_id()+1;
148 unsigned int pidx = 0;
149 unsigned int node;
150 int err;
151
152 err = svc_pool_map_alloc_arrays(m, maxpools);
153 if (err)
154 return err;
155
156 for_each_node_with_cpus(node) {
157 /* some architectures (e.g. SN2) have cpuless nodes */
158 BUG_ON(pidx > maxpools);
159 m->to_pool[node] = pidx;
160 m->pool_to[pidx] = node;
161 pidx++;
162 }
163 /* nodes brought online later all get mapped to pool0, sorry */
164
165 return pidx;
166}
167
168
169/*
170 * Build the global map of cpus to pools and vice versa.
171 */
172static unsigned int
173svc_pool_map_init(void)
174{
175 struct svc_pool_map *m = &svc_pool_map;
176 int npools = -1;
177
178 if (m->mode != SVC_POOL_NONE)
179 return m->npools;
180
181 m->mode = svc_pool_map_choose_mode();
182
183 switch (m->mode) {
184 case SVC_POOL_PERCPU:
185 npools = svc_pool_map_init_percpu(m);
186 break;
187 case SVC_POOL_PERNODE:
188 npools = svc_pool_map_init_pernode(m);
189 break;
190 }
191
192 if (npools < 0) {
193 /* default, or memory allocation failure */
194 npools = 1;
195 m->mode = SVC_POOL_GLOBAL;
196 }
197 m->npools = npools;
198
199 return m->npools;
200}
201
202/*
203 * Set the current thread's cpus_allowed mask so that it
204 * will only run on cpus in the given pool.
205 *
206 * Returns 1 and fills in oldmask iff a cpumask was applied.
207 */
208static inline int
209svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask)
210{
211 struct svc_pool_map *m = &svc_pool_map;
212 unsigned int node; /* or cpu */
213
214 /*
215 * The caller checks for sv_nrpools > 1, which
216 * implies that we've been initialized and the
217 * map mode is not NONE.
218 */
219 BUG_ON(m->mode == SVC_POOL_NONE);
220
221 switch (m->mode)
222 {
223 default:
224 return 0;
225 case SVC_POOL_PERCPU:
226 node = m->pool_to[pidx];
227 *oldmask = current->cpus_allowed;
228 set_cpus_allowed(current, cpumask_of_cpu(node));
229 return 1;
230 case SVC_POOL_PERNODE:
231 node = m->pool_to[pidx];
232 *oldmask = current->cpus_allowed;
233 set_cpus_allowed(current, node_to_cpumask(node));
234 return 1;
235 }
236}
237
238/*
239 * Use the mapping mode to choose a pool for a given CPU.
240 * Used when enqueueing an incoming RPC. Always returns
241 * a non-NULL pool pointer.
242 */
243struct svc_pool *
244svc_pool_for_cpu(struct svc_serv *serv, int cpu)
245{
246 struct svc_pool_map *m = &svc_pool_map;
247 unsigned int pidx = 0;
248
249 /*
250 * SVC_POOL_NONE happens in a pure client when
251 * lockd is brought up, so silently treat it the
252 * same as SVC_POOL_GLOBAL.
253 */
254
255 switch (m->mode) {
256 case SVC_POOL_PERCPU:
257 pidx = m->to_pool[cpu];
258 break;
259 case SVC_POOL_PERNODE:
260 pidx = m->to_pool[cpu_to_node(cpu)];
261 break;
262 }
263 return &serv->sv_pools[pidx % serv->sv_nrpools];
264}
265
266
267/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 * Create an RPC service
269 */
Greg Banksa7455442006-10-02 02:17:59 -0700270static struct svc_serv *
271__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
NeilBrownbc591cc2006-10-02 02:17:44 -0700272 void (*shutdown)(struct svc_serv *serv))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 struct svc_serv *serv;
275 int vers;
276 unsigned int xdrsize;
Greg Banks3262c812006-10-02 02:17:58 -0700277 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700279 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return NULL;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000281 serv->sv_name = prog->pg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 serv->sv_program = prog;
283 serv->sv_nrthreads = 1;
284 serv->sv_stats = prog->pg_stats;
285 serv->sv_bufsz = bufsize? bufsize : 4096;
NeilBrownbc591cc2006-10-02 02:17:44 -0700286 serv->sv_shutdown = shutdown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 xdrsize = 0;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000288 while (prog) {
289 prog->pg_lovers = prog->pg_nvers-1;
290 for (vers=0; vers<prog->pg_nvers ; vers++)
291 if (prog->pg_vers[vers]) {
292 prog->pg_hivers = vers;
293 if (prog->pg_lovers > vers)
294 prog->pg_lovers = vers;
295 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
296 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
297 }
298 prog = prog->pg_next;
299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 serv->sv_xdrsize = xdrsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 INIT_LIST_HEAD(&serv->sv_tempsocks);
302 INIT_LIST_HEAD(&serv->sv_permsocks);
Greg Banks36bdfc82006-10-02 02:17:54 -0700303 init_timer(&serv->sv_temptimer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 spin_lock_init(&serv->sv_lock);
305
Greg Banksa7455442006-10-02 02:17:59 -0700306 serv->sv_nrpools = npools;
Greg Banks3262c812006-10-02 02:17:58 -0700307 serv->sv_pools =
308 kcalloc(sizeof(struct svc_pool), serv->sv_nrpools,
309 GFP_KERNEL);
310 if (!serv->sv_pools) {
311 kfree(serv);
312 return NULL;
313 }
314
315 for (i = 0; i < serv->sv_nrpools; i++) {
316 struct svc_pool *pool = &serv->sv_pools[i];
317
318 dprintk("initialising pool %u for %s\n",
319 i, serv->sv_name);
320
321 pool->sp_id = i;
322 INIT_LIST_HEAD(&pool->sp_threads);
323 INIT_LIST_HEAD(&pool->sp_sockets);
Greg Banksa7455442006-10-02 02:17:59 -0700324 INIT_LIST_HEAD(&pool->sp_all_threads);
Greg Banks3262c812006-10-02 02:17:58 -0700325 spin_lock_init(&pool->sp_lock);
326 }
327
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 /* Remove any stale portmap registrations */
330 svc_register(serv, 0, 0);
331
332 return serv;
333}
334
Greg Banksa7455442006-10-02 02:17:59 -0700335struct svc_serv *
336svc_create(struct svc_program *prog, unsigned int bufsize,
337 void (*shutdown)(struct svc_serv *serv))
338{
339 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
340}
341
342struct svc_serv *
343svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
344 void (*shutdown)(struct svc_serv *serv),
345 svc_thread_fn func, int sig, struct module *mod)
346{
347 struct svc_serv *serv;
Greg Banksbfd24162006-10-02 02:18:01 -0700348 unsigned int npools = svc_pool_map_init();
Greg Banksa7455442006-10-02 02:17:59 -0700349
Greg Banksbfd24162006-10-02 02:18:01 -0700350 serv = __svc_create(prog, bufsize, npools, shutdown);
Greg Banksa7455442006-10-02 02:17:59 -0700351
352 if (serv != NULL) {
353 serv->sv_function = func;
354 serv->sv_kill_signal = sig;
355 serv->sv_module = mod;
356 }
357
358 return serv;
359}
360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361/*
Greg Banks3262c812006-10-02 02:17:58 -0700362 * Destroy an RPC service. Should be called with the BKL held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 */
364void
365svc_destroy(struct svc_serv *serv)
366{
367 struct svc_sock *svsk;
368
369 dprintk("RPC: svc_destroy(%s, %d)\n",
370 serv->sv_program->pg_name,
371 serv->sv_nrthreads);
372
373 if (serv->sv_nrthreads) {
374 if (--(serv->sv_nrthreads) != 0) {
375 svc_sock_update_bufs(serv);
376 return;
377 }
378 } else
379 printk("svc_destroy: no threads for serv=%p!\n", serv);
380
Greg Banks36bdfc82006-10-02 02:17:54 -0700381 del_timer_sync(&serv->sv_temptimer);
382
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 while (!list_empty(&serv->sv_tempsocks)) {
384 svsk = list_entry(serv->sv_tempsocks.next,
385 struct svc_sock,
386 sk_list);
387 svc_delete_socket(svsk);
388 }
NeilBrownbc591cc2006-10-02 02:17:44 -0700389 if (serv->sv_shutdown)
390 serv->sv_shutdown(serv);
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 while (!list_empty(&serv->sv_permsocks)) {
393 svsk = list_entry(serv->sv_permsocks.next,
394 struct svc_sock,
395 sk_list);
396 svc_delete_socket(svsk);
397 }
398
399 cache_clean_deferred(serv);
400
401 /* Unregister service with the portmapper */
402 svc_register(serv, 0, 0);
Greg Banks3262c812006-10-02 02:17:58 -0700403 kfree(serv->sv_pools);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 kfree(serv);
405}
406
407/*
408 * Allocate an RPC server's buffer space.
409 * We allocate pages and place them in rq_argpages.
410 */
411static int
412svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
413{
414 int pages;
415 int arghi;
416
417 if (size > RPCSVC_MAXPAYLOAD)
418 size = RPCSVC_MAXPAYLOAD;
419 pages = 2 + (size+ PAGE_SIZE -1) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 arghi = 0;
Kris Katterjohn09a62662006-01-08 22:24:28 -0800421 BUG_ON(pages > RPCSVC_MAXPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 while (pages) {
423 struct page *p = alloc_page(GFP_KERNEL);
424 if (!p)
425 break;
NeilBrown44524352006-10-04 02:15:46 -0700426 rqstp->rq_pages[arghi++] = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 pages--;
428 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 return ! pages;
430}
431
432/*
433 * Release an RPC server buffer
434 */
435static void
436svc_release_buffer(struct svc_rqst *rqstp)
437{
NeilBrown44524352006-10-04 02:15:46 -0700438 int i;
439 for (i=0; i<ARRAY_SIZE(rqstp->rq_pages); i++)
440 if (rqstp->rq_pages[i])
441 put_page(rqstp->rq_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
444/*
Greg Banks3262c812006-10-02 02:17:58 -0700445 * Create a thread in the given pool. Caller must hold BKL.
Greg Banksbfd24162006-10-02 02:18:01 -0700446 * On a NUMA or SMP machine, with a multi-pool serv, the thread
447 * will be restricted to run on the cpus belonging to the pool.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 */
Greg Banks3262c812006-10-02 02:17:58 -0700449static int
450__svc_create_thread(svc_thread_fn func, struct svc_serv *serv,
451 struct svc_pool *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
453 struct svc_rqst *rqstp;
454 int error = -ENOMEM;
Greg Banksbfd24162006-10-02 02:18:01 -0700455 int have_oldmask = 0;
456 cpumask_t oldmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700458 rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (!rqstp)
460 goto out;
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 init_waitqueue_head(&rqstp->rq_wait);
463
Jesper Juhl12fe2c52006-01-10 13:08:21 -0800464 if (!(rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL))
465 || !(rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 || !svc_init_buffer(rqstp, serv->sv_bufsz))
467 goto out_thread;
468
469 serv->sv_nrthreads++;
Greg Banks3262c812006-10-02 02:17:58 -0700470 spin_lock_bh(&pool->sp_lock);
471 pool->sp_nrthreads++;
Greg Banksa7455442006-10-02 02:17:59 -0700472 list_add(&rqstp->rq_all, &pool->sp_all_threads);
Greg Banks3262c812006-10-02 02:17:58 -0700473 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 rqstp->rq_server = serv;
Greg Banks3262c812006-10-02 02:17:58 -0700475 rqstp->rq_pool = pool;
Greg Banksbfd24162006-10-02 02:18:01 -0700476
477 if (serv->sv_nrpools > 1)
478 have_oldmask = svc_pool_map_set_cpumask(pool->sp_id, &oldmask);
479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 error = kernel_thread((int (*)(void *)) func, rqstp, 0);
Greg Banksbfd24162006-10-02 02:18:01 -0700481
482 if (have_oldmask)
483 set_cpus_allowed(current, oldmask);
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 if (error < 0)
486 goto out_thread;
487 svc_sock_update_bufs(serv);
488 error = 0;
489out:
490 return error;
491
492out_thread:
493 svc_exit_thread(rqstp);
494 goto out;
495}
496
497/*
Greg Banks3262c812006-10-02 02:17:58 -0700498 * Create a thread in the default pool. Caller must hold BKL.
499 */
500int
501svc_create_thread(svc_thread_fn func, struct svc_serv *serv)
502{
503 return __svc_create_thread(func, serv, &serv->sv_pools[0]);
504}
505
506/*
Greg Banksa7455442006-10-02 02:17:59 -0700507 * Choose a pool in which to create a new thread, for svc_set_num_threads
508 */
509static inline struct svc_pool *
510choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
511{
512 if (pool != NULL)
513 return pool;
514
515 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
516}
517
518/*
519 * Choose a thread to kill, for svc_set_num_threads
520 */
521static inline struct task_struct *
522choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
523{
524 unsigned int i;
525 struct task_struct *task = NULL;
526
527 if (pool != NULL) {
528 spin_lock_bh(&pool->sp_lock);
529 } else {
530 /* choose a pool in round-robin fashion */
531 for (i = 0; i < serv->sv_nrpools; i++) {
532 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
533 spin_lock_bh(&pool->sp_lock);
534 if (!list_empty(&pool->sp_all_threads))
535 goto found_pool;
536 spin_unlock_bh(&pool->sp_lock);
537 }
538 return NULL;
539 }
540
541found_pool:
542 if (!list_empty(&pool->sp_all_threads)) {
543 struct svc_rqst *rqstp;
544
545 /*
546 * Remove from the pool->sp_all_threads list
547 * so we don't try to kill it again.
548 */
549 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
550 list_del_init(&rqstp->rq_all);
551 task = rqstp->rq_task;
552 }
553 spin_unlock_bh(&pool->sp_lock);
554
555 return task;
556}
557
558/*
559 * Create or destroy enough new threads to make the number
560 * of threads the given number. If `pool' is non-NULL, applies
561 * only to threads in that pool, otherwise round-robins between
562 * all pools. Must be called with a svc_get() reference and
563 * the BKL held.
564 *
565 * Destroying threads relies on the service threads filling in
566 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
567 * has been created using svc_create_pooled().
568 *
569 * Based on code that used to be in nfsd_svc() but tweaked
570 * to be pool-aware.
571 */
572int
573svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
574{
575 struct task_struct *victim;
576 int error = 0;
577 unsigned int state = serv->sv_nrthreads-1;
578
579 if (pool == NULL) {
580 /* The -1 assumes caller has done a svc_get() */
581 nrservs -= (serv->sv_nrthreads-1);
582 } else {
583 spin_lock_bh(&pool->sp_lock);
584 nrservs -= pool->sp_nrthreads;
585 spin_unlock_bh(&pool->sp_lock);
586 }
587
588 /* create new threads */
589 while (nrservs > 0) {
590 nrservs--;
591 __module_get(serv->sv_module);
592 error = __svc_create_thread(serv->sv_function, serv,
593 choose_pool(serv, pool, &state));
594 if (error < 0) {
595 module_put(serv->sv_module);
596 break;
597 }
598 }
599 /* destroy old threads */
600 while (nrservs < 0 &&
601 (victim = choose_victim(serv, pool, &state)) != NULL) {
602 send_sig(serv->sv_kill_signal, victim, 1);
603 nrservs++;
604 }
605
606 return error;
607}
608
609/*
Greg Banks3262c812006-10-02 02:17:58 -0700610 * Called from a server thread as it's exiting. Caller must hold BKL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
612void
613svc_exit_thread(struct svc_rqst *rqstp)
614{
615 struct svc_serv *serv = rqstp->rq_server;
Greg Banks3262c812006-10-02 02:17:58 -0700616 struct svc_pool *pool = rqstp->rq_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 svc_release_buffer(rqstp);
Jesper Juhla51482b2005-11-08 09:41:34 -0800619 kfree(rqstp->rq_resp);
620 kfree(rqstp->rq_argp);
621 kfree(rqstp->rq_auth_data);
Greg Banks3262c812006-10-02 02:17:58 -0700622
623 spin_lock_bh(&pool->sp_lock);
624 pool->sp_nrthreads--;
Greg Banksa7455442006-10-02 02:17:59 -0700625 list_del(&rqstp->rq_all);
Greg Banks3262c812006-10-02 02:17:58 -0700626 spin_unlock_bh(&pool->sp_lock);
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 kfree(rqstp);
629
630 /* Release the server */
631 if (serv)
632 svc_destroy(serv);
633}
634
635/*
636 * Register an RPC service with the local portmapper.
637 * To unregister a service, call this routine with
638 * proto and port == 0.
639 */
640int
641svc_register(struct svc_serv *serv, int proto, unsigned short port)
642{
643 struct svc_program *progp;
644 unsigned long flags;
645 int i, error = 0, dummy;
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (!port)
648 clear_thread_flag(TIF_SIGPENDING);
649
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700650 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
651 for (i = 0; i < progp->pg_nvers; i++) {
652 if (progp->pg_vers[i] == NULL)
653 continue;
654
655 dprintk("RPC: svc_register(%s, %s, %d, %d)%s\n",
656 progp->pg_name,
657 proto == IPPROTO_UDP? "udp" : "tcp",
658 port,
659 i,
660 progp->pg_vers[i]->vs_hidden?
661 " (but not telling portmap)" : "");
662
663 if (progp->pg_vers[i]->vs_hidden)
664 continue;
665
666 error = rpc_register(progp->pg_prog, i, proto, port, &dummy);
667 if (error < 0)
668 break;
669 if (port && !dummy) {
670 error = -EACCES;
671 break;
672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 }
674 }
675
676 if (!port) {
677 spin_lock_irqsave(&current->sighand->siglock, flags);
678 recalc_sigpending();
679 spin_unlock_irqrestore(&current->sighand->siglock, flags);
680 }
681
682 return error;
683}
684
685/*
686 * Process the RPC request.
687 */
688int
NeilBrown6fb2b472006-10-02 02:17:50 -0700689svc_process(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 struct svc_program *progp;
692 struct svc_version *versp = NULL; /* compiler food */
693 struct svc_procedure *procp = NULL;
694 struct kvec * argv = &rqstp->rq_arg.head[0];
695 struct kvec * resv = &rqstp->rq_res.head[0];
NeilBrown6fb2b472006-10-02 02:17:50 -0700696 struct svc_serv *serv = rqstp->rq_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 kxdrproc_t xdr;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700698 __be32 *statp;
699 u32 dir, prog, vers, proc;
700 __be32 auth_stat, rpc_stat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 int auth_res;
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -0700702 __be32 *reply_statp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
704 rpc_stat = rpc_success;
705
706 if (argv->iov_len < 6*4)
707 goto err_short_len;
708
709 /* setup response xdr_buf.
710 * Initially it has just one page
711 */
NeilBrown44524352006-10-04 02:15:46 -0700712 rqstp->rq_resused = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 resv->iov_base = page_address(rqstp->rq_respages[0]);
714 resv->iov_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700715 rqstp->rq_res.pages = rqstp->rq_respages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 rqstp->rq_res.len = 0;
717 rqstp->rq_res.page_base = 0;
718 rqstp->rq_res.page_len = 0;
Trond Myklebust334ccfd2005-06-22 17:16:19 +0000719 rqstp->rq_res.buflen = PAGE_SIZE;
J. Bruce Fields7c9fdcf2006-06-30 01:56:19 -0700720 rqstp->rq_res.tail[0].iov_base = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 rqstp->rq_res.tail[0].iov_len = 0;
J. Bruce Fields5c04c462006-06-30 01:56:19 -0700722 /* Will be turned off only in gss privacy case: */
723 rqstp->rq_sendfile_ok = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 /* tcp needs a space for the record length... */
725 if (rqstp->rq_prot == IPPROTO_TCP)
Alexey Dobriyan76994312006-09-26 22:28:46 -0700726 svc_putnl(resv, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 rqstp->rq_xid = svc_getu32(argv);
729 svc_putu32(resv, rqstp->rq_xid);
730
Alexey Dobriyan76994312006-09-26 22:28:46 -0700731 dir = svc_getnl(argv);
732 vers = svc_getnl(argv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 /* First words of reply: */
Alexey Dobriyan76994312006-09-26 22:28:46 -0700735 svc_putnl(resv, 1); /* REPLY */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737 if (dir != 0) /* direction != CALL */
738 goto err_bad_dir;
739 if (vers != 2) /* RPC version number */
740 goto err_bad_rpc;
741
742 /* Save position in case we later decide to reject: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -0700743 reply_statp = resv->iov_base + resv->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Alexey Dobriyan76994312006-09-26 22:28:46 -0700745 svc_putnl(resv, 0); /* ACCEPT */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Alexey Dobriyan76994312006-09-26 22:28:46 -0700747 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
748 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
749 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751 progp = serv->sv_program;
NeilBrown80d188a2005-11-07 01:00:27 -0800752
753 for (progp = serv->sv_program; progp; progp = progp->pg_next)
754 if (prog == progp->pg_prog)
755 break;
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 /*
758 * Decode auth data, and add verifier to reply buffer.
759 * We do this before anything else in order to get a decent
760 * auth verifier.
761 */
762 auth_res = svc_authenticate(rqstp, &auth_stat);
763 /* Also give the program a chance to reject this call: */
NeilBrown80d188a2005-11-07 01:00:27 -0800764 if (auth_res == SVC_OK && progp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 auth_stat = rpc_autherr_badcred;
766 auth_res = progp->pg_authenticate(rqstp);
767 }
768 switch (auth_res) {
769 case SVC_OK:
770 break;
771 case SVC_GARBAGE:
772 rpc_stat = rpc_garbage_args;
773 goto err_bad;
774 case SVC_SYSERR:
775 rpc_stat = rpc_system_err;
776 goto err_bad;
777 case SVC_DENIED:
778 goto err_bad_auth;
779 case SVC_DROP:
780 goto dropit;
781 case SVC_COMPLETE:
782 goto sendit;
783 }
NeilBrown80d188a2005-11-07 01:00:27 -0800784
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000785 if (progp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 goto err_bad_prog;
787
788 if (vers >= progp->pg_nvers ||
789 !(versp = progp->pg_vers[vers]))
790 goto err_bad_vers;
791
792 procp = versp->vs_proc + proc;
793 if (proc >= versp->vs_nproc || !procp->pc_func)
794 goto err_bad_proc;
795 rqstp->rq_server = serv;
796 rqstp->rq_procinfo = procp;
797
798 /* Syntactic check complete */
799 serv->sv_stats->rpccnt++;
800
801 /* Build the reply header. */
802 statp = resv->iov_base +resv->iov_len;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700803 svc_putnl(resv, RPC_SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805 /* Bump per-procedure stats counter */
806 procp->pc_count++;
807
808 /* Initialize storage for argp and resp */
809 memset(rqstp->rq_argp, 0, procp->pc_argsize);
810 memset(rqstp->rq_resp, 0, procp->pc_ressize);
811
812 /* un-reserve some of the out-queue now that we have a
813 * better idea of reply size
814 */
815 if (procp->pc_xdrressize)
816 svc_reserve(rqstp, procp->pc_xdrressize<<2);
817
818 /* Call the function that processes the request. */
819 if (!versp->vs_dispatch) {
820 /* Decode arguments */
821 xdr = procp->pc_decode;
822 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
823 goto err_garbage;
824
825 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
826
827 /* Encode reply */
828 if (*statp == rpc_success && (xdr = procp->pc_encode)
829 && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
830 dprintk("svc: failed to encode reply\n");
831 /* serv->sv_stats->rpcsystemerr++; */
832 *statp = rpc_system_err;
833 }
834 } else {
835 dprintk("svc: calling dispatcher\n");
836 if (!versp->vs_dispatch(rqstp, statp)) {
837 /* Release reply info */
838 if (procp->pc_release)
839 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
840 goto dropit;
841 }
842 }
843
844 /* Check RPC status result */
845 if (*statp != rpc_success)
846 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
847
848 /* Release reply info */
849 if (procp->pc_release)
850 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
851
852 if (procp->pc_encode == NULL)
853 goto dropit;
854
855 sendit:
856 if (svc_authorise(rqstp))
857 goto dropit;
858 return svc_send(rqstp);
859
860 dropit:
861 svc_authorise(rqstp); /* doesn't hurt to call this twice */
862 dprintk("svc: svc_process dropit\n");
863 svc_drop(rqstp);
864 return 0;
865
866err_short_len:
867#ifdef RPC_PARANOIA
868 printk("svc: short len %Zd, dropping request\n", argv->iov_len);
869#endif
870 goto dropit; /* drop request */
871
872err_bad_dir:
873#ifdef RPC_PARANOIA
874 printk("svc: bad direction %d, dropping request\n", dir);
875#endif
876 serv->sv_stats->rpcbadfmt++;
877 goto dropit; /* drop request */
878
879err_bad_rpc:
880 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700881 svc_putnl(resv, 1); /* REJECT */
882 svc_putnl(resv, 0); /* RPC_MISMATCH */
883 svc_putnl(resv, 2); /* Only RPCv2 supported */
884 svc_putnl(resv, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 goto sendit;
886
887err_bad_auth:
888 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
889 serv->sv_stats->rpcbadauth++;
890 /* Restore write pointer to location of accept status: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -0700891 xdr_ressize_check(rqstp, reply_statp);
Alexey Dobriyan76994312006-09-26 22:28:46 -0700892 svc_putnl(resv, 1); /* REJECT */
893 svc_putnl(resv, 1); /* AUTH_ERROR */
894 svc_putnl(resv, ntohl(auth_stat)); /* status */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 goto sendit;
896
897err_bad_prog:
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000898 dprintk("svc: unknown program %d\n", prog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700900 svc_putnl(resv, RPC_PROG_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 goto sendit;
902
903err_bad_vers:
904#ifdef RPC_PARANOIA
905 printk("svc: unknown version (%d)\n", vers);
906#endif
907 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700908 svc_putnl(resv, RPC_PROG_MISMATCH);
909 svc_putnl(resv, progp->pg_lovers);
910 svc_putnl(resv, progp->pg_hivers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 goto sendit;
912
913err_bad_proc:
914#ifdef RPC_PARANOIA
915 printk("svc: unknown procedure (%d)\n", proc);
916#endif
917 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700918 svc_putnl(resv, RPC_PROC_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 goto sendit;
920
921err_garbage:
922#ifdef RPC_PARANOIA
923 printk("svc: failed to decode args\n");
924#endif
925 rpc_stat = rpc_garbage_args;
926err_bad:
927 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700928 svc_putnl(resv, ntohl(rpc_stat));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 goto sendit;
930}
Greg Banks7adae482006-10-04 02:15:47 -0700931
932/*
933 * Return (transport-specific) limit on the rpc payload.
934 */
935u32 svc_max_payload(const struct svc_rqst *rqstp)
936{
937 int max = RPCSVC_MAXPAYLOAD_TCP;
938
939 if (rqstp->rq_sock->sk_sock->type == SOCK_DGRAM)
940 max = RPCSVC_MAXPAYLOAD_UDP;
941 if (rqstp->rq_server->sv_bufsz < max)
942 max = rqstp->rq_server->sv_bufsz;
943 return max;
944}
945EXPORT_SYMBOL_GPL(svc_max_payload);