blob: 538ca433a56cc72e2bae671e2b74174cb8efb3a8 [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>
Jeff Layton9867d762008-06-10 08:40:38 -040021#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
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>
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -040028#include <linux/sunrpc/bc_xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#define RPCDBG_FACILITY RPCDBG_SVCDSP
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Chuck Lever7252d572008-08-18 19:34:08 -040032static void svc_unregister(const struct svc_serv *serv);
33
Greg Banks42a7fc42007-03-06 01:42:23 -080034#define svc_serv_is_pooled(serv) ((serv)->sv_function)
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
Greg Banksbfd24162006-10-02 02:18:01 -070037 * Mode for mapping cpus to pools.
38 */
39enum {
Greg Banks42a7fc42007-03-06 01:42:23 -080040 SVC_POOL_AUTO = -1, /* choose one of the others */
Greg Banksbfd24162006-10-02 02:18:01 -070041 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
42 * (legacy & UP mode) */
43 SVC_POOL_PERCPU, /* one pool per cpu */
44 SVC_POOL_PERNODE /* one pool per numa node */
45};
Greg Banks42a7fc42007-03-06 01:42:23 -080046#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
Greg Banksbfd24162006-10-02 02:18:01 -070047
48/*
49 * Structure for mapping cpus to pools and vice versa.
50 * Setup once during sunrpc initialisation.
51 */
52static struct svc_pool_map {
Greg Banks42a7fc42007-03-06 01:42:23 -080053 int count; /* How many svc_servs use us */
Greg Banksbfd24162006-10-02 02:18:01 -070054 int mode; /* Note: int not enum to avoid
55 * warnings about "enumeration value
56 * not handled in switch" */
57 unsigned int npools;
58 unsigned int *pool_to; /* maps pool id to cpu or node */
59 unsigned int *to_pool; /* maps cpu or node to pool id */
60} svc_pool_map = {
Greg Banks42a7fc42007-03-06 01:42:23 -080061 .count = 0,
62 .mode = SVC_POOL_DEFAULT
Greg Banksbfd24162006-10-02 02:18:01 -070063};
Greg Banks42a7fc42007-03-06 01:42:23 -080064static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
Greg Banksbfd24162006-10-02 02:18:01 -070065
Greg Banks42a7fc42007-03-06 01:42:23 -080066static int
67param_set_pool_mode(const char *val, struct kernel_param *kp)
68{
69 int *ip = (int *)kp->arg;
70 struct svc_pool_map *m = &svc_pool_map;
71 int err;
72
73 mutex_lock(&svc_pool_map_mutex);
74
75 err = -EBUSY;
76 if (m->count)
77 goto out;
78
79 err = 0;
80 if (!strncmp(val, "auto", 4))
81 *ip = SVC_POOL_AUTO;
82 else if (!strncmp(val, "global", 6))
83 *ip = SVC_POOL_GLOBAL;
84 else if (!strncmp(val, "percpu", 6))
85 *ip = SVC_POOL_PERCPU;
86 else if (!strncmp(val, "pernode", 7))
87 *ip = SVC_POOL_PERNODE;
88 else
89 err = -EINVAL;
90
91out:
92 mutex_unlock(&svc_pool_map_mutex);
93 return err;
94}
95
96static int
97param_get_pool_mode(char *buf, struct kernel_param *kp)
98{
99 int *ip = (int *)kp->arg;
100
101 switch (*ip)
102 {
103 case SVC_POOL_AUTO:
104 return strlcpy(buf, "auto", 20);
105 case SVC_POOL_GLOBAL:
106 return strlcpy(buf, "global", 20);
107 case SVC_POOL_PERCPU:
108 return strlcpy(buf, "percpu", 20);
109 case SVC_POOL_PERNODE:
110 return strlcpy(buf, "pernode", 20);
111 default:
112 return sprintf(buf, "%d", *ip);
113 }
114}
115
116module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
117 &svc_pool_map.mode, 0644);
Greg Banksbfd24162006-10-02 02:18:01 -0700118
119/*
120 * Detect best pool mapping mode heuristically,
121 * according to the machine's topology.
122 */
123static int
124svc_pool_map_choose_mode(void)
125{
126 unsigned int node;
127
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700128 if (nr_online_nodes > 1) {
Greg Banksbfd24162006-10-02 02:18:01 -0700129 /*
130 * Actually have multiple NUMA nodes,
131 * so split pools on NUMA node boundaries
132 */
133 return SVC_POOL_PERNODE;
134 }
135
136 node = any_online_node(node_online_map);
137 if (nr_cpus_node(node) > 2) {
138 /*
139 * Non-trivial SMP, or CONFIG_NUMA on
140 * non-NUMA hardware, e.g. with a generic
141 * x86_64 kernel on Xeons. In this case we
142 * want to divide the pools on cpu boundaries.
143 */
144 return SVC_POOL_PERCPU;
145 }
146
147 /* default: one global pool */
148 return SVC_POOL_GLOBAL;
149}
150
151/*
152 * Allocate the to_pool[] and pool_to[] arrays.
153 * Returns 0 on success or an errno.
154 */
155static int
156svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
157{
158 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
159 if (!m->to_pool)
160 goto fail;
161 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
162 if (!m->pool_to)
163 goto fail_free;
164
165 return 0;
166
167fail_free:
168 kfree(m->to_pool);
169fail:
170 return -ENOMEM;
171}
172
173/*
174 * Initialise the pool map for SVC_POOL_PERCPU mode.
175 * Returns number of pools or <0 on error.
176 */
177static int
178svc_pool_map_init_percpu(struct svc_pool_map *m)
179{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800180 unsigned int maxpools = nr_cpu_ids;
Greg Banksbfd24162006-10-02 02:18:01 -0700181 unsigned int pidx = 0;
182 unsigned int cpu;
183 int err;
184
185 err = svc_pool_map_alloc_arrays(m, maxpools);
186 if (err)
187 return err;
188
189 for_each_online_cpu(cpu) {
190 BUG_ON(pidx > maxpools);
191 m->to_pool[cpu] = pidx;
192 m->pool_to[pidx] = cpu;
193 pidx++;
194 }
195 /* cpus brought online later all get mapped to pool0, sorry */
196
197 return pidx;
198};
199
200
201/*
202 * Initialise the pool map for SVC_POOL_PERNODE mode.
203 * Returns number of pools or <0 on error.
204 */
205static int
206svc_pool_map_init_pernode(struct svc_pool_map *m)
207{
Christoph Lameter74c7aa82007-02-20 13:57:51 -0800208 unsigned int maxpools = nr_node_ids;
Greg Banksbfd24162006-10-02 02:18:01 -0700209 unsigned int pidx = 0;
210 unsigned int node;
211 int err;
212
213 err = svc_pool_map_alloc_arrays(m, maxpools);
214 if (err)
215 return err;
216
217 for_each_node_with_cpus(node) {
218 /* some architectures (e.g. SN2) have cpuless nodes */
219 BUG_ON(pidx > maxpools);
220 m->to_pool[node] = pidx;
221 m->pool_to[pidx] = node;
222 pidx++;
223 }
224 /* nodes brought online later all get mapped to pool0, sorry */
225
226 return pidx;
227}
228
229
230/*
Greg Banks42a7fc42007-03-06 01:42:23 -0800231 * Add a reference to the global map of cpus to pools (and
232 * vice versa). Initialise the map if we're the first user.
233 * Returns the number of pools.
Greg Banksbfd24162006-10-02 02:18:01 -0700234 */
235static unsigned int
Greg Banks42a7fc42007-03-06 01:42:23 -0800236svc_pool_map_get(void)
Greg Banksbfd24162006-10-02 02:18:01 -0700237{
238 struct svc_pool_map *m = &svc_pool_map;
239 int npools = -1;
240
Greg Banks42a7fc42007-03-06 01:42:23 -0800241 mutex_lock(&svc_pool_map_mutex);
Greg Banksbfd24162006-10-02 02:18:01 -0700242
Greg Banks42a7fc42007-03-06 01:42:23 -0800243 if (m->count++) {
244 mutex_unlock(&svc_pool_map_mutex);
245 return m->npools;
246 }
247
248 if (m->mode == SVC_POOL_AUTO)
249 m->mode = svc_pool_map_choose_mode();
Greg Banksbfd24162006-10-02 02:18:01 -0700250
251 switch (m->mode) {
252 case SVC_POOL_PERCPU:
253 npools = svc_pool_map_init_percpu(m);
254 break;
255 case SVC_POOL_PERNODE:
256 npools = svc_pool_map_init_pernode(m);
257 break;
258 }
259
260 if (npools < 0) {
261 /* default, or memory allocation failure */
262 npools = 1;
263 m->mode = SVC_POOL_GLOBAL;
264 }
265 m->npools = npools;
266
Greg Banks42a7fc42007-03-06 01:42:23 -0800267 mutex_unlock(&svc_pool_map_mutex);
Greg Banksbfd24162006-10-02 02:18:01 -0700268 return m->npools;
269}
270
Greg Banks42a7fc42007-03-06 01:42:23 -0800271
272/*
273 * Drop a reference to the global map of cpus to pools.
274 * When the last reference is dropped, the map data is
275 * freed; this allows the sysadmin to change the pool
276 * mode using the pool_mode module option without
277 * rebooting or re-loading sunrpc.ko.
278 */
279static void
280svc_pool_map_put(void)
281{
282 struct svc_pool_map *m = &svc_pool_map;
283
284 mutex_lock(&svc_pool_map_mutex);
285
286 if (!--m->count) {
287 m->mode = SVC_POOL_DEFAULT;
288 kfree(m->to_pool);
289 kfree(m->pool_to);
290 m->npools = 0;
291 }
292
293 mutex_unlock(&svc_pool_map_mutex);
294}
295
296
Greg Banksbfd24162006-10-02 02:18:01 -0700297/*
Jeff Layton9867d762008-06-10 08:40:38 -0400298 * Set the given thread's cpus_allowed mask so that it
Greg Banksbfd24162006-10-02 02:18:01 -0700299 * will only run on cpus in the given pool.
Greg Banksbfd24162006-10-02 02:18:01 -0700300 */
Jeff Layton9867d762008-06-10 08:40:38 -0400301static inline void
302svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
Greg Banksbfd24162006-10-02 02:18:01 -0700303{
304 struct svc_pool_map *m = &svc_pool_map;
Jeff Layton9867d762008-06-10 08:40:38 -0400305 unsigned int node = m->pool_to[pidx];
Greg Banksbfd24162006-10-02 02:18:01 -0700306
307 /*
308 * The caller checks for sv_nrpools > 1, which
Greg Banks42a7fc42007-03-06 01:42:23 -0800309 * implies that we've been initialized.
Greg Banksbfd24162006-10-02 02:18:01 -0700310 */
Greg Banks42a7fc42007-03-06 01:42:23 -0800311 BUG_ON(m->count == 0);
Greg Banksbfd24162006-10-02 02:18:01 -0700312
Jeff Layton9867d762008-06-10 08:40:38 -0400313 switch (m->mode) {
Greg Banksbfd24162006-10-02 02:18:01 -0700314 case SVC_POOL_PERCPU:
Mike Travisc5f59f02008-04-04 18:11:10 -0700315 {
Rusty Russellaa85ea52009-03-30 22:05:15 -0600316 set_cpus_allowed_ptr(task, cpumask_of(node));
Jeff Layton9867d762008-06-10 08:40:38 -0400317 break;
Mike Travisc5f59f02008-04-04 18:11:10 -0700318 }
Greg Banksbfd24162006-10-02 02:18:01 -0700319 case SVC_POOL_PERNODE:
Mike Travisc5f59f02008-04-04 18:11:10 -0700320 {
Rusty Russella70f7302009-03-13 14:49:46 +1030321 set_cpus_allowed_ptr(task, cpumask_of_node(node));
Jeff Layton9867d762008-06-10 08:40:38 -0400322 break;
Greg Banksbfd24162006-10-02 02:18:01 -0700323 }
Mike Travisc5f59f02008-04-04 18:11:10 -0700324 }
Greg Banksbfd24162006-10-02 02:18:01 -0700325}
326
327/*
328 * Use the mapping mode to choose a pool for a given CPU.
329 * Used when enqueueing an incoming RPC. Always returns
330 * a non-NULL pool pointer.
331 */
332struct svc_pool *
333svc_pool_for_cpu(struct svc_serv *serv, int cpu)
334{
335 struct svc_pool_map *m = &svc_pool_map;
336 unsigned int pidx = 0;
337
338 /*
Greg Banks42a7fc42007-03-06 01:42:23 -0800339 * An uninitialised map happens in a pure client when
Greg Banksbfd24162006-10-02 02:18:01 -0700340 * lockd is brought up, so silently treat it the
341 * same as SVC_POOL_GLOBAL.
342 */
Greg Banks42a7fc42007-03-06 01:42:23 -0800343 if (svc_serv_is_pooled(serv)) {
344 switch (m->mode) {
345 case SVC_POOL_PERCPU:
346 pidx = m->to_pool[cpu];
347 break;
348 case SVC_POOL_PERNODE:
349 pidx = m->to_pool[cpu_to_node(cpu)];
350 break;
351 }
Greg Banksbfd24162006-10-02 02:18:01 -0700352 }
353 return &serv->sv_pools[pidx % serv->sv_nrpools];
354}
355
356
357/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 * Create an RPC service
359 */
Greg Banksa7455442006-10-02 02:17:59 -0700360static struct svc_serv *
361__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
Chuck Lever49a90722009-03-18 20:46:29 -0400362 void (*shutdown)(struct svc_serv *serv))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
364 struct svc_serv *serv;
Chuck Leverea339d42007-10-26 13:32:56 -0400365 unsigned int vers;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 unsigned int xdrsize;
Greg Banks3262c812006-10-02 02:17:58 -0700367 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700369 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 return NULL;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000371 serv->sv_name = prog->pg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 serv->sv_program = prog;
373 serv->sv_nrthreads = 1;
374 serv->sv_stats = prog->pg_stats;
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700375 if (bufsize > RPCSVC_MAXPAYLOAD)
376 bufsize = RPCSVC_MAXPAYLOAD;
377 serv->sv_max_payload = bufsize? bufsize : 4096;
378 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
NeilBrownbc591cc2006-10-02 02:17:44 -0700379 serv->sv_shutdown = shutdown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 xdrsize = 0;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000381 while (prog) {
382 prog->pg_lovers = prog->pg_nvers-1;
383 for (vers=0; vers<prog->pg_nvers ; vers++)
384 if (prog->pg_vers[vers]) {
385 prog->pg_hivers = vers;
386 if (prog->pg_lovers > vers)
387 prog->pg_lovers = vers;
388 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
389 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
390 }
391 prog = prog->pg_next;
392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 serv->sv_xdrsize = xdrsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 INIT_LIST_HEAD(&serv->sv_tempsocks);
395 INIT_LIST_HEAD(&serv->sv_permsocks);
Greg Banks36bdfc82006-10-02 02:17:54 -0700396 init_timer(&serv->sv_temptimer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 spin_lock_init(&serv->sv_lock);
398
Greg Banksa7455442006-10-02 02:17:59 -0700399 serv->sv_nrpools = npools;
Greg Banks3262c812006-10-02 02:17:58 -0700400 serv->sv_pools =
Robert P. J. Daycd861282006-12-13 00:34:52 -0800401 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
Greg Banks3262c812006-10-02 02:17:58 -0700402 GFP_KERNEL);
403 if (!serv->sv_pools) {
404 kfree(serv);
405 return NULL;
406 }
407
408 for (i = 0; i < serv->sv_nrpools; i++) {
409 struct svc_pool *pool = &serv->sv_pools[i];
410
Chuck Lever46121cf2007-01-31 12:14:08 -0500411 dprintk("svc: initialising pool %u for %s\n",
Greg Banks3262c812006-10-02 02:17:58 -0700412 i, serv->sv_name);
413
414 pool->sp_id = i;
415 INIT_LIST_HEAD(&pool->sp_threads);
416 INIT_LIST_HEAD(&pool->sp_sockets);
Greg Banksa7455442006-10-02 02:17:59 -0700417 INIT_LIST_HEAD(&pool->sp_all_threads);
Greg Banks3262c812006-10-02 02:17:58 -0700418 spin_lock_init(&pool->sp_lock);
419 }
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 /* Remove any stale portmap registrations */
Chuck Lever7252d572008-08-18 19:34:08 -0400422 svc_unregister(serv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424 return serv;
425}
426
Greg Banksa7455442006-10-02 02:17:59 -0700427struct svc_serv *
428svc_create(struct svc_program *prog, unsigned int bufsize,
Chuck Lever49a90722009-03-18 20:46:29 -0400429 void (*shutdown)(struct svc_serv *serv))
Greg Banksa7455442006-10-02 02:17:59 -0700430{
Chuck Lever49a90722009-03-18 20:46:29 -0400431 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
Greg Banksa7455442006-10-02 02:17:59 -0700432}
Trond Myklebust24c37672008-12-23 16:30:12 -0500433EXPORT_SYMBOL_GPL(svc_create);
Greg Banksa7455442006-10-02 02:17:59 -0700434
435struct svc_serv *
436svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
Chuck Lever49a90722009-03-18 20:46:29 -0400437 void (*shutdown)(struct svc_serv *serv),
Jeff Laytona75c5d02008-06-10 08:40:39 -0400438 svc_thread_fn func, struct module *mod)
Greg Banksa7455442006-10-02 02:17:59 -0700439{
440 struct svc_serv *serv;
Greg Banks42a7fc42007-03-06 01:42:23 -0800441 unsigned int npools = svc_pool_map_get();
Greg Banksa7455442006-10-02 02:17:59 -0700442
Chuck Lever49a90722009-03-18 20:46:29 -0400443 serv = __svc_create(prog, bufsize, npools, shutdown);
Greg Banksa7455442006-10-02 02:17:59 -0700444
445 if (serv != NULL) {
446 serv->sv_function = func;
Greg Banksa7455442006-10-02 02:17:59 -0700447 serv->sv_module = mod;
448 }
449
450 return serv;
451}
Trond Myklebust24c37672008-12-23 16:30:12 -0500452EXPORT_SYMBOL_GPL(svc_create_pooled);
Greg Banksa7455442006-10-02 02:17:59 -0700453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454/*
Neil Brownbedbdd82008-06-10 08:40:35 -0400455 * Destroy an RPC service. Should be called with appropriate locking to
456 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 */
458void
459svc_destroy(struct svc_serv *serv)
460{
Chuck Lever46121cf2007-01-31 12:14:08 -0500461 dprintk("svc: svc_destroy(%s, %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 serv->sv_program->pg_name,
463 serv->sv_nrthreads);
464
465 if (serv->sv_nrthreads) {
466 if (--(serv->sv_nrthreads) != 0) {
467 svc_sock_update_bufs(serv);
468 return;
469 }
470 } else
471 printk("svc_destroy: no threads for serv=%p!\n", serv);
472
Greg Banks36bdfc82006-10-02 02:17:54 -0700473 del_timer_sync(&serv->sv_temptimer);
474
Tom Tucker7a182082007-12-30 21:07:53 -0600475 svc_close_all(&serv->sv_tempsocks);
NeilBrowncda1fd42007-03-06 01:42:22 -0800476
NeilBrownbc591cc2006-10-02 02:17:44 -0700477 if (serv->sv_shutdown)
478 serv->sv_shutdown(serv);
479
Tom Tucker7a182082007-12-30 21:07:53 -0600480 svc_close_all(&serv->sv_permsocks);
NeilBrowncda1fd42007-03-06 01:42:22 -0800481
482 BUG_ON(!list_empty(&serv->sv_permsocks));
483 BUG_ON(!list_empty(&serv->sv_tempsocks));
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 cache_clean_deferred(serv);
486
Greg Banks42a7fc42007-03-06 01:42:23 -0800487 if (svc_serv_is_pooled(serv))
488 svc_pool_map_put();
489
Andy Adamson9c9f3f52009-04-01 09:23:10 -0400490#if defined(CONFIG_NFS_V4_1)
491 svc_sock_destroy(serv->bc_xprt);
492#endif /* CONFIG_NFS_V4_1 */
493
Chuck Lever7252d572008-08-18 19:34:08 -0400494 svc_unregister(serv);
Greg Banks3262c812006-10-02 02:17:58 -0700495 kfree(serv->sv_pools);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 kfree(serv);
497}
Trond Myklebust24c37672008-12-23 16:30:12 -0500498EXPORT_SYMBOL_GPL(svc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500/*
501 * Allocate an RPC server's buffer space.
502 * We allocate pages and place them in rq_argpages.
503 */
504static int
505svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
506{
Chuck Lever0dc220f2008-04-14 12:27:52 -0400507 unsigned int pages, arghi;
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800508
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700509 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
510 * We assume one is at most one page
511 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 arghi = 0;
Kris Katterjohn09a62662006-01-08 22:24:28 -0800513 BUG_ON(pages > RPCSVC_MAXPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 while (pages) {
515 struct page *p = alloc_page(GFP_KERNEL);
516 if (!p)
517 break;
NeilBrown44524352006-10-04 02:15:46 -0700518 rqstp->rq_pages[arghi++] = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 pages--;
520 }
Chuck Lever0dc220f2008-04-14 12:27:52 -0400521 return pages == 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524/*
525 * Release an RPC server buffer
526 */
527static void
528svc_release_buffer(struct svc_rqst *rqstp)
529{
Chuck Lever50c8bb12008-04-14 12:27:45 -0400530 unsigned int i;
531
532 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
NeilBrown44524352006-10-04 02:15:46 -0700533 if (rqstp->rq_pages[i])
534 put_page(rqstp->rq_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
Jeff Layton0113ab32008-01-29 10:30:54 -0500537struct svc_rqst *
538svc_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;
568out_thread:
569 svc_exit_thread(rqstp);
570out_enomem:
571 return ERR_PTR(-ENOMEM);
572}
Trond Myklebust24c37672008-12-23 16:30:12 -0500573EXPORT_SYMBOL_GPL(svc_prepare_thread);
Jeff Layton0113ab32008-01-29 10:30:54 -0500574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575/*
Greg Banksa7455442006-10-02 02:17:59 -0700576 * Choose a pool in which to create a new thread, for svc_set_num_threads
577 */
578static inline struct svc_pool *
579choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
580{
581 if (pool != NULL)
582 return pool;
583
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800584 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
Greg Banksa7455442006-10-02 02:17:59 -0700585}
586
587/*
588 * Choose a thread to kill, for svc_set_num_threads
589 */
590static inline struct task_struct *
591choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
592{
593 unsigned int i;
594 struct task_struct *task = NULL;
595
596 if (pool != NULL) {
597 spin_lock_bh(&pool->sp_lock);
598 } else {
599 /* choose a pool in round-robin fashion */
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800600 for (i = 0; i < serv->sv_nrpools; i++) {
601 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
Greg Banksa7455442006-10-02 02:17:59 -0700602 spin_lock_bh(&pool->sp_lock);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800603 if (!list_empty(&pool->sp_all_threads))
604 goto found_pool;
Greg Banksa7455442006-10-02 02:17:59 -0700605 spin_unlock_bh(&pool->sp_lock);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800606 }
Greg Banksa7455442006-10-02 02:17:59 -0700607 return NULL;
608 }
609
610found_pool:
611 if (!list_empty(&pool->sp_all_threads)) {
612 struct svc_rqst *rqstp;
613
614 /*
615 * Remove from the pool->sp_all_threads list
616 * so we don't try to kill it again.
617 */
618 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
619 list_del_init(&rqstp->rq_all);
620 task = rqstp->rq_task;
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800621 }
Greg Banksa7455442006-10-02 02:17:59 -0700622 spin_unlock_bh(&pool->sp_lock);
623
624 return task;
625}
626
627/*
628 * Create or destroy enough new threads to make the number
629 * of threads the given number. If `pool' is non-NULL, applies
630 * only to threads in that pool, otherwise round-robins between
631 * all pools. Must be called with a svc_get() reference and
Neil Brownbedbdd82008-06-10 08:40:35 -0400632 * the BKL or another lock to protect access to svc_serv fields.
Greg Banksa7455442006-10-02 02:17:59 -0700633 *
634 * Destroying threads relies on the service threads filling in
635 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
636 * has been created using svc_create_pooled().
637 *
638 * Based on code that used to be in nfsd_svc() but tweaked
639 * to be pool-aware.
640 */
641int
642svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
643{
Jeff Layton9867d762008-06-10 08:40:38 -0400644 struct svc_rqst *rqstp;
645 struct task_struct *task;
646 struct svc_pool *chosen_pool;
Greg Banksa7455442006-10-02 02:17:59 -0700647 int error = 0;
648 unsigned int state = serv->sv_nrthreads-1;
649
650 if (pool == NULL) {
651 /* The -1 assumes caller has done a svc_get() */
652 nrservs -= (serv->sv_nrthreads-1);
653 } else {
654 spin_lock_bh(&pool->sp_lock);
655 nrservs -= pool->sp_nrthreads;
656 spin_unlock_bh(&pool->sp_lock);
657 }
658
659 /* create new threads */
660 while (nrservs > 0) {
661 nrservs--;
Jeff Layton9867d762008-06-10 08:40:38 -0400662 chosen_pool = choose_pool(serv, pool, &state);
663
664 rqstp = svc_prepare_thread(serv, chosen_pool);
665 if (IS_ERR(rqstp)) {
666 error = PTR_ERR(rqstp);
Greg Banksa7455442006-10-02 02:17:59 -0700667 break;
668 }
Jeff Layton9867d762008-06-10 08:40:38 -0400669
670 __module_get(serv->sv_module);
671 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
672 if (IS_ERR(task)) {
673 error = PTR_ERR(task);
674 module_put(serv->sv_module);
675 svc_exit_thread(rqstp);
676 break;
677 }
678
679 rqstp->rq_task = task;
680 if (serv->sv_nrpools > 1)
681 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
682
683 svc_sock_update_bufs(serv);
684 wake_up_process(task);
Greg Banksa7455442006-10-02 02:17:59 -0700685 }
686 /* destroy old threads */
687 while (nrservs < 0 &&
Jeff Layton9867d762008-06-10 08:40:38 -0400688 (task = choose_victim(serv, pool, &state)) != NULL) {
Jeff Laytona75c5d02008-06-10 08:40:39 -0400689 send_sig(SIGINT, task, 1);
Greg Banksa7455442006-10-02 02:17:59 -0700690 nrservs++;
691 }
692
693 return error;
694}
Trond Myklebust24c37672008-12-23 16:30:12 -0500695EXPORT_SYMBOL_GPL(svc_set_num_threads);
Greg Banksa7455442006-10-02 02:17:59 -0700696
697/*
Neil Brownbedbdd82008-06-10 08:40:35 -0400698 * Called from a server thread as it's exiting. Caller must hold the BKL or
699 * the "service mutex", whichever is appropriate for the service.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 */
701void
702svc_exit_thread(struct svc_rqst *rqstp)
703{
704 struct svc_serv *serv = rqstp->rq_server;
Greg Banks3262c812006-10-02 02:17:58 -0700705 struct svc_pool *pool = rqstp->rq_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 svc_release_buffer(rqstp);
Jesper Juhla51482b2005-11-08 09:41:34 -0800708 kfree(rqstp->rq_resp);
709 kfree(rqstp->rq_argp);
710 kfree(rqstp->rq_auth_data);
Greg Banks3262c812006-10-02 02:17:58 -0700711
712 spin_lock_bh(&pool->sp_lock);
713 pool->sp_nrthreads--;
Greg Banksa7455442006-10-02 02:17:59 -0700714 list_del(&rqstp->rq_all);
Greg Banks3262c812006-10-02 02:17:58 -0700715 spin_unlock_bh(&pool->sp_lock);
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 kfree(rqstp);
718
719 /* Release the server */
720 if (serv)
721 svc_destroy(serv);
722}
Trond Myklebust24c37672008-12-23 16:30:12 -0500723EXPORT_SYMBOL_GPL(svc_exit_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
725/*
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500726 * Register an "inet" protocol family netid with the local
727 * rpcbind daemon via an rpcbind v4 SET request.
Chuck Levera26cfad2008-08-18 19:34:16 -0400728 *
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500729 * No netconfig infrastructure is available in the kernel, so
730 * we map IP_ protocol numbers to netids by hand.
Chuck Levera26cfad2008-08-18 19:34:16 -0400731 *
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500732 * Returns zero on success; a negative errno value is returned
733 * if any error occurs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 */
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500735static int __svc_rpcb_register4(const u32 program, const u32 version,
736 const unsigned short protocol,
737 const unsigned short port)
Chuck Levera26cfad2008-08-18 19:34:16 -0400738{
Chuck Levercadc0fa2009-03-18 20:47:36 -0400739 const struct sockaddr_in sin = {
Chuck Levera26cfad2008-08-18 19:34:16 -0400740 .sin_family = AF_INET,
741 .sin_addr.s_addr = htonl(INADDR_ANY),
742 .sin_port = htons(port),
743 };
Chuck Levercadc0fa2009-03-18 20:47:36 -0400744 const char *netid;
745 int error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500746
747 switch (protocol) {
748 case IPPROTO_UDP:
749 netid = RPCBIND_NETID_UDP;
750 break;
751 case IPPROTO_TCP:
752 netid = RPCBIND_NETID_TCP;
753 break;
754 default:
Chuck Leverba5c35e2009-03-18 20:46:59 -0400755 return -ENOPROTOOPT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500756 }
757
Chuck Levercadc0fa2009-03-18 20:47:36 -0400758 error = rpcb_v4_register(program, version,
759 (const struct sockaddr *)&sin, netid);
760
761 /*
762 * User space didn't support rpcbind v4, so retry this
763 * registration request with the legacy rpcbind v2 protocol.
764 */
765 if (error == -EPROTONOSUPPORT)
766 error = rpcb_register(program, version, protocol, port);
767
768 return error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500769}
770
Chuck Levercadc0fa2009-03-18 20:47:36 -0400771#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500772/*
773 * Register an "inet6" protocol family netid with the local
774 * rpcbind daemon via an rpcbind v4 SET request.
775 *
776 * No netconfig infrastructure is available in the kernel, so
777 * we map IP_ protocol numbers to netids by hand.
778 *
779 * Returns zero on success; a negative errno value is returned
780 * if any error occurs.
781 */
782static int __svc_rpcb_register6(const u32 program, const u32 version,
783 const unsigned short protocol,
784 const unsigned short port)
785{
Chuck Levercadc0fa2009-03-18 20:47:36 -0400786 const struct sockaddr_in6 sin6 = {
Chuck Levera26cfad2008-08-18 19:34:16 -0400787 .sin6_family = AF_INET6,
788 .sin6_addr = IN6ADDR_ANY_INIT,
789 .sin6_port = htons(port),
790 };
Chuck Levercadc0fa2009-03-18 20:47:36 -0400791 const char *netid;
792 int error;
Chuck Levera26cfad2008-08-18 19:34:16 -0400793
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500794 switch (protocol) {
795 case IPPROTO_UDP:
796 netid = RPCBIND_NETID_UDP6;
797 break;
798 case IPPROTO_TCP:
799 netid = RPCBIND_NETID_TCP6;
800 break;
801 default:
Chuck Leverba5c35e2009-03-18 20:46:59 -0400802 return -ENOPROTOOPT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500803 }
804
Chuck Levercadc0fa2009-03-18 20:47:36 -0400805 error = rpcb_v4_register(program, version,
806 (const struct sockaddr *)&sin6, netid);
807
808 /*
809 * User space didn't support rpcbind version 4, so we won't
810 * use a PF_INET6 listener.
811 */
812 if (error == -EPROTONOSUPPORT)
813 error = -EAFNOSUPPORT;
814
815 return error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500816}
Chuck Levercadc0fa2009-03-18 20:47:36 -0400817#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500818
819/*
820 * Register a kernel RPC service via rpcbind version 4.
821 *
822 * Returns zero on success; a negative errno value is returned
823 * if any error occurs.
824 */
Chuck Lever363f7242009-03-18 20:47:44 -0400825static int __svc_register(const char *progname,
826 const u32 program, const u32 version,
Chuck Lever4b62e582009-03-18 20:46:06 -0400827 const int family,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500828 const unsigned short protocol,
829 const unsigned short port)
830{
Chuck Lever363f7242009-03-18 20:47:44 -0400831 int error = -EAFNOSUPPORT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500832
Chuck Levera26cfad2008-08-18 19:34:16 -0400833 switch (family) {
Chuck Lever4b62e582009-03-18 20:46:06 -0400834 case PF_INET:
Chuck Lever363f7242009-03-18 20:47:44 -0400835 error = __svc_rpcb_register4(program, version,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500836 protocol, port);
Chuck Levercadc0fa2009-03-18 20:47:36 -0400837 break;
838#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever4b62e582009-03-18 20:46:06 -0400839 case PF_INET6:
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500840 error = __svc_rpcb_register6(program, version,
841 protocol, port);
Chuck Levercadc0fa2009-03-18 20:47:36 -0400842#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Levera26cfad2008-08-18 19:34:16 -0400843 }
844
Chuck Lever363f7242009-03-18 20:47:44 -0400845 if (error < 0)
846 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
847 "service (errno %d).\n", progname, version, -error);
848 return error;
Chuck Levera26cfad2008-08-18 19:34:16 -0400849}
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500850
Chuck Levera26cfad2008-08-18 19:34:16 -0400851/**
852 * svc_register - register an RPC service with the local portmapper
853 * @serv: svc_serv struct for the service to register
Chuck Lever4b62e582009-03-18 20:46:06 -0400854 * @family: protocol family of service's listener socket
Chuck Levera26cfad2008-08-18 19:34:16 -0400855 * @proto: transport protocol number to advertise
856 * @port: port to advertise
857 *
Chuck Lever4b62e582009-03-18 20:46:06 -0400858 * Service is registered for any address in the passed-in protocol family
Chuck Levera26cfad2008-08-18 19:34:16 -0400859 */
Chuck Lever4b62e582009-03-18 20:46:06 -0400860int svc_register(const struct svc_serv *serv, const int family,
861 const unsigned short proto, const unsigned short port)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
863 struct svc_program *progp;
Chuck Leverea339d42007-10-26 13:32:56 -0400864 unsigned int i;
Chuck Lever14aeb212008-08-18 19:34:00 -0400865 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
Chuck Lever7252d572008-08-18 19:34:08 -0400867 BUG_ON(proto == 0 && port == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700869 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
870 for (i = 0; i < progp->pg_nvers; i++) {
871 if (progp->pg_vers[i] == NULL)
872 continue;
873
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500874 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700875 progp->pg_name,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500876 i,
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700877 proto == IPPROTO_UDP? "udp" : "tcp",
878 port,
Chuck Lever4b62e582009-03-18 20:46:06 -0400879 family,
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700880 progp->pg_vers[i]->vs_hidden?
881 " (but not telling portmap)" : "");
882
883 if (progp->pg_vers[i]->vs_hidden)
884 continue;
885
Chuck Lever363f7242009-03-18 20:47:44 -0400886 error = __svc_register(progp->pg_name, progp->pg_prog,
887 i, family, proto, port);
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700888 if (error < 0)
889 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
891 }
892
Chuck Lever7252d572008-08-18 19:34:08 -0400893 return error;
894}
895
Chuck Leverd5a86202009-03-18 20:47:29 -0400896/*
897 * If user space is running rpcbind, it should take the v4 UNSET
898 * and clear everything for this [program, version]. If user space
899 * is running portmap, it will reject the v4 UNSET, but won't have
900 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
901 * in this case to clear all existing entries for [program, version].
902 */
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400903static void __svc_unregister(const u32 program, const u32 version,
904 const char *progname)
905{
906 int error;
907
Chuck Leverd5a86202009-03-18 20:47:29 -0400908 error = rpcb_v4_register(program, version, NULL, "");
909
910 /*
911 * User space didn't support rpcbind v4, so retry this
912 * request with the legacy rpcbind v2 protocol.
913 */
914 if (error == -EPROTONOSUPPORT)
915 error = rpcb_register(program, version, 0, 0);
916
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400917 dprintk("svc: %s(%sv%u), error %d\n",
918 __func__, progname, version, error);
919}
920
Chuck Lever7252d572008-08-18 19:34:08 -0400921/*
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400922 * All netids, bind addresses and ports registered for [program, version]
923 * are removed from the local rpcbind database (if the service is not
924 * hidden) to make way for a new instance of the service.
Chuck Lever7252d572008-08-18 19:34:08 -0400925 *
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400926 * The result of unregistration is reported via dprintk for those who want
927 * verification of the result, but is otherwise not important.
Chuck Lever7252d572008-08-18 19:34:08 -0400928 */
929static void svc_unregister(const struct svc_serv *serv)
930{
931 struct svc_program *progp;
932 unsigned long flags;
933 unsigned int i;
Chuck Lever7252d572008-08-18 19:34:08 -0400934
935 clear_thread_flag(TIF_SIGPENDING);
936
937 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
938 for (i = 0; i < progp->pg_nvers; i++) {
939 if (progp->pg_vers[i] == NULL)
940 continue;
941 if (progp->pg_vers[i]->vs_hidden)
942 continue;
943
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400944 __svc_unregister(progp->pg_prog, i, progp->pg_name);
Chuck Lever7252d572008-08-18 19:34:08 -0400945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
947
Chuck Lever7252d572008-08-18 19:34:08 -0400948 spin_lock_irqsave(&current->sighand->siglock, flags);
949 recalc_sigpending();
950 spin_unlock_irqrestore(&current->sighand->siglock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
953/*
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +0100954 * Printk the given error with the address of the client that caused it.
955 */
956static int
957__attribute__ ((format (printf, 2, 3)))
958svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
959{
960 va_list args;
961 int r;
962 char buf[RPC_MAX_ADDRBUFLEN];
963
964 if (!net_ratelimit())
965 return 0;
966
967 printk(KERN_WARNING "svc: %s: ",
968 svc_print_addr(rqstp, buf, sizeof(buf)));
969
970 va_start(args, fmt);
971 r = vprintk(fmt, args);
972 va_end(args);
973
974 return r;
975}
976
977/*
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -0400978 * Common routine for processing the RPC request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 */
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -0400980static int
981svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
983 struct svc_program *progp;
984 struct svc_version *versp = NULL; /* compiler food */
985 struct svc_procedure *procp = NULL;
NeilBrown6fb2b472006-10-02 02:17:50 -0700986 struct svc_serv *serv = rqstp->rq_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 kxdrproc_t xdr;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700988 __be32 *statp;
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -0400989 u32 prog, vers, proc;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700990 __be32 auth_stat, rpc_stat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 int auth_res;
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -0700992 __be32 *reply_statp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 rpc_stat = rpc_success;
995
996 if (argv->iov_len < 6*4)
997 goto err_short_len;
998
J. Bruce Fields5c04c462006-06-30 01:56:19 -0700999 /* Will be turned off only in gss privacy case: */
Jens Axboecf8208d2007-06-12 21:22:14 +02001000 rqstp->rq_splice_ok = 1;
Andy Adamson2f425872009-04-03 08:27:32 +03001001 /* Will be turned off only when NFSv4 Sessions are used */
1002 rqstp->rq_usedeferral = 1;
Tom Tuckere831fe62007-12-30 21:07:29 -06001003
1004 /* Setup reply header */
1005 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 svc_putu32(resv, rqstp->rq_xid);
1008
Alexey Dobriyan76994312006-09-26 22:28:46 -07001009 vers = svc_getnl(argv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 /* First words of reply: */
Alexey Dobriyan76994312006-09-26 22:28:46 -07001012 svc_putnl(resv, 1); /* REPLY */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (vers != 2) /* RPC version number */
1015 goto err_bad_rpc;
1016
1017 /* Save position in case we later decide to reject: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -07001018 reply_statp = resv->iov_base + resv->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Alexey Dobriyan76994312006-09-26 22:28:46 -07001020 svc_putnl(resv, 0); /* ACCEPT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Alexey Dobriyan76994312006-09-26 22:28:46 -07001022 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1023 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1024 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 progp = serv->sv_program;
NeilBrown80d188a2005-11-07 01:00:27 -08001027
1028 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1029 if (prog == progp->pg_prog)
1030 break;
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 /*
1033 * Decode auth data, and add verifier to reply buffer.
1034 * We do this before anything else in order to get a decent
1035 * auth verifier.
1036 */
1037 auth_res = svc_authenticate(rqstp, &auth_stat);
1038 /* Also give the program a chance to reject this call: */
NeilBrown80d188a2005-11-07 01:00:27 -08001039 if (auth_res == SVC_OK && progp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 auth_stat = rpc_autherr_badcred;
1041 auth_res = progp->pg_authenticate(rqstp);
1042 }
1043 switch (auth_res) {
1044 case SVC_OK:
1045 break;
1046 case SVC_GARBAGE:
Harshula Jayasuriyadd352102008-02-20 10:56:56 +11001047 goto err_garbage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 case SVC_SYSERR:
1049 rpc_stat = rpc_system_err;
1050 goto err_bad;
1051 case SVC_DENIED:
1052 goto err_bad_auth;
1053 case SVC_DROP:
1054 goto dropit;
1055 case SVC_COMPLETE:
1056 goto sendit;
1057 }
NeilBrown80d188a2005-11-07 01:00:27 -08001058
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +00001059 if (progp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 goto err_bad_prog;
1061
1062 if (vers >= progp->pg_nvers ||
1063 !(versp = progp->pg_vers[vers]))
1064 goto err_bad_vers;
1065
1066 procp = versp->vs_proc + proc;
1067 if (proc >= versp->vs_nproc || !procp->pc_func)
1068 goto err_bad_proc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 rqstp->rq_procinfo = procp;
1070
1071 /* Syntactic check complete */
1072 serv->sv_stats->rpccnt++;
1073
1074 /* Build the reply header. */
1075 statp = resv->iov_base +resv->iov_len;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001076 svc_putnl(resv, RPC_SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 /* Bump per-procedure stats counter */
1079 procp->pc_count++;
1080
1081 /* Initialize storage for argp and resp */
1082 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1083 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1084
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001085 /* un-reserve some of the out-queue now that we have a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 * better idea of reply size
1087 */
1088 if (procp->pc_xdrressize)
Jeff Laytoncd123012007-05-09 02:34:50 -07001089 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
1091 /* Call the function that processes the request. */
1092 if (!versp->vs_dispatch) {
1093 /* Decode arguments */
1094 xdr = procp->pc_decode;
1095 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1096 goto err_garbage;
1097
1098 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1099
1100 /* Encode reply */
NeilBrownd343fce2006-10-17 00:10:18 -07001101 if (*statp == rpc_drop_reply) {
1102 if (procp->pc_release)
1103 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1104 goto dropit;
1105 }
Joe Perchesf64f9e72009-11-29 16:55:45 -08001106 if (*statp == rpc_success &&
1107 (xdr = procp->pc_encode) &&
1108 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 dprintk("svc: failed to encode reply\n");
1110 /* serv->sv_stats->rpcsystemerr++; */
1111 *statp = rpc_system_err;
1112 }
1113 } else {
1114 dprintk("svc: calling dispatcher\n");
1115 if (!versp->vs_dispatch(rqstp, statp)) {
1116 /* Release reply info */
1117 if (procp->pc_release)
1118 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1119 goto dropit;
1120 }
1121 }
1122
1123 /* Check RPC status result */
1124 if (*statp != rpc_success)
1125 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1126
1127 /* Release reply info */
1128 if (procp->pc_release)
1129 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1130
1131 if (procp->pc_encode == NULL)
1132 goto dropit;
1133
1134 sendit:
1135 if (svc_authorise(rqstp))
1136 goto dropit;
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001137 return 1; /* Caller can now send it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 dropit:
1140 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1141 dprintk("svc: svc_process dropit\n");
1142 svc_drop(rqstp);
1143 return 0;
1144
1145err_short_len:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001146 svc_printk(rqstp, "short len %Zd, dropping request\n",
1147 argv->iov_len);
NeilBrown34e9a632007-01-29 13:19:52 -08001148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 goto dropit; /* drop request */
1150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151err_bad_rpc:
1152 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001153 svc_putnl(resv, 1); /* REJECT */
1154 svc_putnl(resv, 0); /* RPC_MISMATCH */
1155 svc_putnl(resv, 2); /* Only RPCv2 supported */
1156 svc_putnl(resv, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 goto sendit;
1158
1159err_bad_auth:
1160 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1161 serv->sv_stats->rpcbadauth++;
1162 /* Restore write pointer to location of accept status: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -07001163 xdr_ressize_check(rqstp, reply_statp);
Alexey Dobriyan76994312006-09-26 22:28:46 -07001164 svc_putnl(resv, 1); /* REJECT */
1165 svc_putnl(resv, 1); /* AUTH_ERROR */
1166 svc_putnl(resv, ntohl(auth_stat)); /* status */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 goto sendit;
1168
1169err_bad_prog:
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +00001170 dprintk("svc: unknown program %d\n", prog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001172 svc_putnl(resv, RPC_PROG_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 goto sendit;
1174
1175err_bad_vers:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001176 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
NeilBrown34e9a632007-01-29 13:19:52 -08001177 vers, prog, progp->pg_name);
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001180 svc_putnl(resv, RPC_PROG_MISMATCH);
1181 svc_putnl(resv, progp->pg_lovers);
1182 svc_putnl(resv, progp->pg_hivers);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 goto sendit;
1184
1185err_bad_proc:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001186 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
NeilBrown34e9a632007-01-29 13:19:52 -08001187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001189 svc_putnl(resv, RPC_PROC_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 goto sendit;
1191
1192err_garbage:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001193 svc_printk(rqstp, "failed to decode args\n");
NeilBrown34e9a632007-01-29 13:19:52 -08001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 rpc_stat = rpc_garbage_args;
1196err_bad:
1197 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001198 svc_putnl(resv, ntohl(rpc_stat));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 goto sendit;
1200}
Trond Myklebust24c37672008-12-23 16:30:12 -05001201EXPORT_SYMBOL_GPL(svc_process);
Greg Banks7adae482006-10-04 02:15:47 -07001202
1203/*
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001204 * Process the RPC request.
1205 */
1206int
1207svc_process(struct svc_rqst *rqstp)
1208{
1209 struct kvec *argv = &rqstp->rq_arg.head[0];
1210 struct kvec *resv = &rqstp->rq_res.head[0];
1211 struct svc_serv *serv = rqstp->rq_server;
1212 u32 dir;
1213 int error;
1214
1215 /*
1216 * Setup response xdr_buf.
1217 * Initially it has just one page
1218 */
1219 rqstp->rq_resused = 1;
1220 resv->iov_base = page_address(rqstp->rq_respages[0]);
1221 resv->iov_len = 0;
1222 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1223 rqstp->rq_res.len = 0;
1224 rqstp->rq_res.page_base = 0;
1225 rqstp->rq_res.page_len = 0;
1226 rqstp->rq_res.buflen = PAGE_SIZE;
1227 rqstp->rq_res.tail[0].iov_base = NULL;
1228 rqstp->rq_res.tail[0].iov_len = 0;
1229
1230 rqstp->rq_xid = svc_getu32(argv);
1231
1232 dir = svc_getnl(argv);
1233 if (dir != 0) {
1234 /* direction != CALL */
1235 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1236 serv->sv_stats->rpcbadfmt++;
1237 svc_drop(rqstp);
1238 return 0;
1239 }
1240
1241 error = svc_process_common(rqstp, argv, resv);
1242 if (error <= 0)
1243 return error;
1244
1245 return svc_send(rqstp);
1246}
1247
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001248#if defined(CONFIG_NFS_V4_1)
1249/*
1250 * Process a backchannel RPC request that arrived over an existing
1251 * outbound connection
1252 */
1253int
1254bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1255 struct svc_rqst *rqstp)
1256{
1257 struct kvec *argv = &rqstp->rq_arg.head[0];
1258 struct kvec *resv = &rqstp->rq_res.head[0];
1259 int error;
1260
1261 /* Build the svc_rqst used by the common processing routine */
Benny Halevy8f975242009-04-01 09:23:11 -04001262 rqstp->rq_xprt = serv->bc_xprt;
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001263 rqstp->rq_xid = req->rq_xid;
1264 rqstp->rq_prot = req->rq_xprt->prot;
1265 rqstp->rq_server = serv;
1266
1267 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1268 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1269 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1270 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1271
1272 /* reset result send buffer "put" position */
1273 resv->iov_len = 0;
1274
1275 if (rqstp->rq_prot != IPPROTO_TCP) {
1276 printk(KERN_ERR "No support for Non-TCP transports!\n");
1277 BUG();
1278 }
1279
1280 /*
1281 * Skip the next two words because they've already been
1282 * processed in the trasport
1283 */
1284 svc_getu32(argv); /* XID */
1285 svc_getnl(argv); /* CALLDIR */
1286
1287 error = svc_process_common(rqstp, argv, resv);
1288 if (error <= 0)
1289 return error;
1290
1291 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1292 return bc_send(req);
1293}
1294EXPORT_SYMBOL(bc_svc_process);
1295#endif /* CONFIG_NFS_V4_1 */
1296
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001297/*
Greg Banks7adae482006-10-04 02:15:47 -07001298 * Return (transport-specific) limit on the rpc payload.
1299 */
1300u32 svc_max_payload(const struct svc_rqst *rqstp)
1301{
Tom Tucker49023152007-12-30 21:07:21 -06001302 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
Greg Banks7adae482006-10-04 02:15:47 -07001303
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001304 if (rqstp->rq_server->sv_max_payload < max)
1305 max = rqstp->rq_server->sv_max_payload;
Greg Banks7adae482006-10-04 02:15:47 -07001306 return max;
1307}
1308EXPORT_SYMBOL_GPL(svc_max_payload);