blob: 30d70abb4e2c7e7577758b4857401a04b1992b0f [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090022#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <linux/sunrpc/types.h>
25#include <linux/sunrpc/xdr.h>
26#include <linux/sunrpc/stats.h>
27#include <linux/sunrpc/svcsock.h>
28#include <linux/sunrpc/clnt.h>
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -040029#include <linux/sunrpc/bc_xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#define RPCDBG_FACILITY RPCDBG_SVCDSP
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Chuck Lever7252d572008-08-18 19:34:08 -040033static void svc_unregister(const struct svc_serv *serv);
34
Greg Banks42a7fc42007-03-06 01:42:23 -080035#define svc_serv_is_pooled(serv) ((serv)->sv_function)
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
Greg Banksbfd24162006-10-02 02:18:01 -070038 * Mode for mapping cpus to pools.
39 */
40enum {
Greg Banks42a7fc42007-03-06 01:42:23 -080041 SVC_POOL_AUTO = -1, /* choose one of the others */
Greg Banksbfd24162006-10-02 02:18:01 -070042 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
43 * (legacy & UP mode) */
44 SVC_POOL_PERCPU, /* one pool per cpu */
45 SVC_POOL_PERNODE /* one pool per numa node */
46};
Greg Banks42a7fc42007-03-06 01:42:23 -080047#define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
Greg Banksbfd24162006-10-02 02:18:01 -070048
49/*
50 * Structure for mapping cpus to pools and vice versa.
51 * Setup once during sunrpc initialisation.
52 */
53static struct svc_pool_map {
Greg Banks42a7fc42007-03-06 01:42:23 -080054 int count; /* How many svc_servs use us */
Greg Banksbfd24162006-10-02 02:18:01 -070055 int mode; /* Note: int not enum to avoid
56 * warnings about "enumeration value
57 * not handled in switch" */
58 unsigned int npools;
59 unsigned int *pool_to; /* maps pool id to cpu or node */
60 unsigned int *to_pool; /* maps cpu or node to pool id */
61} svc_pool_map = {
Greg Banks42a7fc42007-03-06 01:42:23 -080062 .count = 0,
63 .mode = SVC_POOL_DEFAULT
Greg Banksbfd24162006-10-02 02:18:01 -070064};
Greg Banks42a7fc42007-03-06 01:42:23 -080065static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
Greg Banksbfd24162006-10-02 02:18:01 -070066
Greg Banks42a7fc42007-03-06 01:42:23 -080067static int
68param_set_pool_mode(const char *val, struct kernel_param *kp)
69{
70 int *ip = (int *)kp->arg;
71 struct svc_pool_map *m = &svc_pool_map;
72 int err;
73
74 mutex_lock(&svc_pool_map_mutex);
75
76 err = -EBUSY;
77 if (m->count)
78 goto out;
79
80 err = 0;
81 if (!strncmp(val, "auto", 4))
82 *ip = SVC_POOL_AUTO;
83 else if (!strncmp(val, "global", 6))
84 *ip = SVC_POOL_GLOBAL;
85 else if (!strncmp(val, "percpu", 6))
86 *ip = SVC_POOL_PERCPU;
87 else if (!strncmp(val, "pernode", 7))
88 *ip = SVC_POOL_PERNODE;
89 else
90 err = -EINVAL;
91
92out:
93 mutex_unlock(&svc_pool_map_mutex);
94 return err;
95}
96
97static int
98param_get_pool_mode(char *buf, struct kernel_param *kp)
99{
100 int *ip = (int *)kp->arg;
101
102 switch (*ip)
103 {
104 case SVC_POOL_AUTO:
105 return strlcpy(buf, "auto", 20);
106 case SVC_POOL_GLOBAL:
107 return strlcpy(buf, "global", 20);
108 case SVC_POOL_PERCPU:
109 return strlcpy(buf, "percpu", 20);
110 case SVC_POOL_PERNODE:
111 return strlcpy(buf, "pernode", 20);
112 default:
113 return sprintf(buf, "%d", *ip);
114 }
115}
116
117module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
118 &svc_pool_map.mode, 0644);
Greg Banksbfd24162006-10-02 02:18:01 -0700119
120/*
121 * Detect best pool mapping mode heuristically,
122 * according to the machine's topology.
123 */
124static int
125svc_pool_map_choose_mode(void)
126{
127 unsigned int node;
128
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700129 if (nr_online_nodes > 1) {
Greg Banksbfd24162006-10-02 02:18:01 -0700130 /*
131 * Actually have multiple NUMA nodes,
132 * so split pools on NUMA node boundaries
133 */
134 return SVC_POOL_PERNODE;
135 }
136
H Hartley Sweeten72c33682010-03-05 13:42:43 -0800137 node = first_online_node;
Greg Banksbfd24162006-10-02 02:18:01 -0700138 if (nr_cpus_node(node) > 2) {
139 /*
140 * Non-trivial SMP, or CONFIG_NUMA on
141 * non-NUMA hardware, e.g. with a generic
142 * x86_64 kernel on Xeons. In this case we
143 * want to divide the pools on cpu boundaries.
144 */
145 return SVC_POOL_PERCPU;
146 }
147
148 /* default: one global pool */
149 return SVC_POOL_GLOBAL;
150}
151
152/*
153 * Allocate the to_pool[] and pool_to[] arrays.
154 * Returns 0 on success or an errno.
155 */
156static int
157svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
158{
159 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
160 if (!m->to_pool)
161 goto fail;
162 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
163 if (!m->pool_to)
164 goto fail_free;
165
166 return 0;
167
168fail_free:
169 kfree(m->to_pool);
170fail:
171 return -ENOMEM;
172}
173
174/*
175 * Initialise the pool map for SVC_POOL_PERCPU mode.
176 * Returns number of pools or <0 on error.
177 */
178static int
179svc_pool_map_init_percpu(struct svc_pool_map *m)
180{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800181 unsigned int maxpools = nr_cpu_ids;
Greg Banksbfd24162006-10-02 02:18:01 -0700182 unsigned int pidx = 0;
183 unsigned int cpu;
184 int err;
185
186 err = svc_pool_map_alloc_arrays(m, maxpools);
187 if (err)
188 return err;
189
190 for_each_online_cpu(cpu) {
191 BUG_ON(pidx > maxpools);
192 m->to_pool[cpu] = pidx;
193 m->pool_to[pidx] = cpu;
194 pidx++;
195 }
196 /* cpus brought online later all get mapped to pool0, sorry */
197
198 return pidx;
199};
200
201
202/*
203 * Initialise the pool map for SVC_POOL_PERNODE mode.
204 * Returns number of pools or <0 on error.
205 */
206static int
207svc_pool_map_init_pernode(struct svc_pool_map *m)
208{
Christoph Lameter74c7aa82007-02-20 13:57:51 -0800209 unsigned int maxpools = nr_node_ids;
Greg Banksbfd24162006-10-02 02:18:01 -0700210 unsigned int pidx = 0;
211 unsigned int node;
212 int err;
213
214 err = svc_pool_map_alloc_arrays(m, maxpools);
215 if (err)
216 return err;
217
218 for_each_node_with_cpus(node) {
219 /* some architectures (e.g. SN2) have cpuless nodes */
220 BUG_ON(pidx > maxpools);
221 m->to_pool[node] = pidx;
222 m->pool_to[pidx] = node;
223 pidx++;
224 }
225 /* nodes brought online later all get mapped to pool0, sorry */
226
227 return pidx;
228}
229
230
231/*
Greg Banks42a7fc42007-03-06 01:42:23 -0800232 * Add a reference to the global map of cpus to pools (and
233 * vice versa). Initialise the map if we're the first user.
234 * Returns the number of pools.
Greg Banksbfd24162006-10-02 02:18:01 -0700235 */
236static unsigned int
Greg Banks42a7fc42007-03-06 01:42:23 -0800237svc_pool_map_get(void)
Greg Banksbfd24162006-10-02 02:18:01 -0700238{
239 struct svc_pool_map *m = &svc_pool_map;
240 int npools = -1;
241
Greg Banks42a7fc42007-03-06 01:42:23 -0800242 mutex_lock(&svc_pool_map_mutex);
Greg Banksbfd24162006-10-02 02:18:01 -0700243
Greg Banks42a7fc42007-03-06 01:42:23 -0800244 if (m->count++) {
245 mutex_unlock(&svc_pool_map_mutex);
246 return m->npools;
247 }
248
249 if (m->mode == SVC_POOL_AUTO)
250 m->mode = svc_pool_map_choose_mode();
Greg Banksbfd24162006-10-02 02:18:01 -0700251
252 switch (m->mode) {
253 case SVC_POOL_PERCPU:
254 npools = svc_pool_map_init_percpu(m);
255 break;
256 case SVC_POOL_PERNODE:
257 npools = svc_pool_map_init_pernode(m);
258 break;
259 }
260
261 if (npools < 0) {
262 /* default, or memory allocation failure */
263 npools = 1;
264 m->mode = SVC_POOL_GLOBAL;
265 }
266 m->npools = npools;
267
Greg Banks42a7fc42007-03-06 01:42:23 -0800268 mutex_unlock(&svc_pool_map_mutex);
Greg Banksbfd24162006-10-02 02:18:01 -0700269 return m->npools;
270}
271
Greg Banks42a7fc42007-03-06 01:42:23 -0800272
273/*
274 * Drop a reference to the global map of cpus to pools.
275 * When the last reference is dropped, the map data is
276 * freed; this allows the sysadmin to change the pool
277 * mode using the pool_mode module option without
278 * rebooting or re-loading sunrpc.ko.
279 */
280static void
281svc_pool_map_put(void)
282{
283 struct svc_pool_map *m = &svc_pool_map;
284
285 mutex_lock(&svc_pool_map_mutex);
286
287 if (!--m->count) {
288 m->mode = SVC_POOL_DEFAULT;
289 kfree(m->to_pool);
290 kfree(m->pool_to);
291 m->npools = 0;
292 }
293
294 mutex_unlock(&svc_pool_map_mutex);
295}
296
297
Eric Dumazet11fd1652011-07-28 20:04:09 +0200298static int svc_pool_map_get_node(unsigned int pidx)
299{
300 const struct svc_pool_map *m = &svc_pool_map;
301
302 if (m->count) {
303 if (m->mode == SVC_POOL_PERCPU)
304 return cpu_to_node(m->pool_to[pidx]);
305 if (m->mode == SVC_POOL_PERNODE)
306 return m->pool_to[pidx];
307 }
308 return NUMA_NO_NODE;
309}
Greg Banksbfd24162006-10-02 02:18:01 -0700310/*
Jeff Layton9867d762008-06-10 08:40:38 -0400311 * Set the given thread's cpus_allowed mask so that it
Greg Banksbfd24162006-10-02 02:18:01 -0700312 * will only run on cpus in the given pool.
Greg Banksbfd24162006-10-02 02:18:01 -0700313 */
Jeff Layton9867d762008-06-10 08:40:38 -0400314static inline void
315svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
Greg Banksbfd24162006-10-02 02:18:01 -0700316{
317 struct svc_pool_map *m = &svc_pool_map;
Jeff Layton9867d762008-06-10 08:40:38 -0400318 unsigned int node = m->pool_to[pidx];
Greg Banksbfd24162006-10-02 02:18:01 -0700319
320 /*
321 * The caller checks for sv_nrpools > 1, which
Greg Banks42a7fc42007-03-06 01:42:23 -0800322 * implies that we've been initialized.
Greg Banksbfd24162006-10-02 02:18:01 -0700323 */
Greg Banks42a7fc42007-03-06 01:42:23 -0800324 BUG_ON(m->count == 0);
Greg Banksbfd24162006-10-02 02:18:01 -0700325
Jeff Layton9867d762008-06-10 08:40:38 -0400326 switch (m->mode) {
Greg Banksbfd24162006-10-02 02:18:01 -0700327 case SVC_POOL_PERCPU:
Mike Travisc5f59f02008-04-04 18:11:10 -0700328 {
Rusty Russellaa85ea52009-03-30 22:05:15 -0600329 set_cpus_allowed_ptr(task, cpumask_of(node));
Jeff Layton9867d762008-06-10 08:40:38 -0400330 break;
Mike Travisc5f59f02008-04-04 18:11:10 -0700331 }
Greg Banksbfd24162006-10-02 02:18:01 -0700332 case SVC_POOL_PERNODE:
Mike Travisc5f59f02008-04-04 18:11:10 -0700333 {
Rusty Russella70f7302009-03-13 14:49:46 +1030334 set_cpus_allowed_ptr(task, cpumask_of_node(node));
Jeff Layton9867d762008-06-10 08:40:38 -0400335 break;
Greg Banksbfd24162006-10-02 02:18:01 -0700336 }
Mike Travisc5f59f02008-04-04 18:11:10 -0700337 }
Greg Banksbfd24162006-10-02 02:18:01 -0700338}
339
340/*
341 * Use the mapping mode to choose a pool for a given CPU.
342 * Used when enqueueing an incoming RPC. Always returns
343 * a non-NULL pool pointer.
344 */
345struct svc_pool *
346svc_pool_for_cpu(struct svc_serv *serv, int cpu)
347{
348 struct svc_pool_map *m = &svc_pool_map;
349 unsigned int pidx = 0;
350
351 /*
Greg Banks42a7fc42007-03-06 01:42:23 -0800352 * An uninitialised map happens in a pure client when
Greg Banksbfd24162006-10-02 02:18:01 -0700353 * lockd is brought up, so silently treat it the
354 * same as SVC_POOL_GLOBAL.
355 */
Greg Banks42a7fc42007-03-06 01:42:23 -0800356 if (svc_serv_is_pooled(serv)) {
357 switch (m->mode) {
358 case SVC_POOL_PERCPU:
359 pidx = m->to_pool[cpu];
360 break;
361 case SVC_POOL_PERNODE:
362 pidx = m->to_pool[cpu_to_node(cpu)];
363 break;
364 }
Greg Banksbfd24162006-10-02 02:18:01 -0700365 }
366 return &serv->sv_pools[pidx % serv->sv_nrpools];
367}
368
369
370/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 * Create an RPC service
372 */
Greg Banksa7455442006-10-02 02:17:59 -0700373static struct svc_serv *
374__svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
Chuck Lever49a90722009-03-18 20:46:29 -0400375 void (*shutdown)(struct svc_serv *serv))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
377 struct svc_serv *serv;
Chuck Leverea339d42007-10-26 13:32:56 -0400378 unsigned int vers;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 unsigned int xdrsize;
Greg Banks3262c812006-10-02 02:17:58 -0700380 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700382 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return NULL;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000384 serv->sv_name = prog->pg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 serv->sv_program = prog;
386 serv->sv_nrthreads = 1;
387 serv->sv_stats = prog->pg_stats;
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700388 if (bufsize > RPCSVC_MAXPAYLOAD)
389 bufsize = RPCSVC_MAXPAYLOAD;
390 serv->sv_max_payload = bufsize? bufsize : 4096;
391 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
NeilBrownbc591cc2006-10-02 02:17:44 -0700392 serv->sv_shutdown = shutdown;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 xdrsize = 0;
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000394 while (prog) {
395 prog->pg_lovers = prog->pg_nvers-1;
396 for (vers=0; vers<prog->pg_nvers ; vers++)
397 if (prog->pg_vers[vers]) {
398 prog->pg_hivers = vers;
399 if (prog->pg_lovers > vers)
400 prog->pg_lovers = vers;
401 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
402 xdrsize = prog->pg_vers[vers]->vs_xdrsize;
403 }
404 prog = prog->pg_next;
405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 serv->sv_xdrsize = xdrsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 INIT_LIST_HEAD(&serv->sv_tempsocks);
408 INIT_LIST_HEAD(&serv->sv_permsocks);
Greg Banks36bdfc82006-10-02 02:17:54 -0700409 init_timer(&serv->sv_temptimer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 spin_lock_init(&serv->sv_lock);
411
Greg Banksa7455442006-10-02 02:17:59 -0700412 serv->sv_nrpools = npools;
Greg Banks3262c812006-10-02 02:17:58 -0700413 serv->sv_pools =
Robert P. J. Daycd861282006-12-13 00:34:52 -0800414 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
Greg Banks3262c812006-10-02 02:17:58 -0700415 GFP_KERNEL);
416 if (!serv->sv_pools) {
417 kfree(serv);
418 return NULL;
419 }
420
421 for (i = 0; i < serv->sv_nrpools; i++) {
422 struct svc_pool *pool = &serv->sv_pools[i];
423
Chuck Lever46121cf2007-01-31 12:14:08 -0500424 dprintk("svc: initialising pool %u for %s\n",
Greg Banks3262c812006-10-02 02:17:58 -0700425 i, serv->sv_name);
426
427 pool->sp_id = i;
428 INIT_LIST_HEAD(&pool->sp_threads);
429 INIT_LIST_HEAD(&pool->sp_sockets);
Greg Banksa7455442006-10-02 02:17:59 -0700430 INIT_LIST_HEAD(&pool->sp_all_threads);
Greg Banks3262c812006-10-02 02:17:58 -0700431 spin_lock_init(&pool->sp_lock);
432 }
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 /* Remove any stale portmap registrations */
Chuck Lever7252d572008-08-18 19:34:08 -0400435 svc_unregister(serv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437 return serv;
438}
439
Greg Banksa7455442006-10-02 02:17:59 -0700440struct svc_serv *
441svc_create(struct svc_program *prog, unsigned int bufsize,
Chuck Lever49a90722009-03-18 20:46:29 -0400442 void (*shutdown)(struct svc_serv *serv))
Greg Banksa7455442006-10-02 02:17:59 -0700443{
Chuck Lever49a90722009-03-18 20:46:29 -0400444 return __svc_create(prog, bufsize, /*npools*/1, shutdown);
Greg Banksa7455442006-10-02 02:17:59 -0700445}
Trond Myklebust24c37672008-12-23 16:30:12 -0500446EXPORT_SYMBOL_GPL(svc_create);
Greg Banksa7455442006-10-02 02:17:59 -0700447
448struct svc_serv *
449svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
Chuck Lever49a90722009-03-18 20:46:29 -0400450 void (*shutdown)(struct svc_serv *serv),
Jeff Laytona75c5d02008-06-10 08:40:39 -0400451 svc_thread_fn func, struct module *mod)
Greg Banksa7455442006-10-02 02:17:59 -0700452{
453 struct svc_serv *serv;
Greg Banks42a7fc42007-03-06 01:42:23 -0800454 unsigned int npools = svc_pool_map_get();
Greg Banksa7455442006-10-02 02:17:59 -0700455
Chuck Lever49a90722009-03-18 20:46:29 -0400456 serv = __svc_create(prog, bufsize, npools, shutdown);
Greg Banksa7455442006-10-02 02:17:59 -0700457
458 if (serv != NULL) {
459 serv->sv_function = func;
Greg Banksa7455442006-10-02 02:17:59 -0700460 serv->sv_module = mod;
461 }
462
463 return serv;
464}
Trond Myklebust24c37672008-12-23 16:30:12 -0500465EXPORT_SYMBOL_GPL(svc_create_pooled);
Greg Banksa7455442006-10-02 02:17:59 -0700466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467/*
Neil Brownbedbdd82008-06-10 08:40:35 -0400468 * Destroy an RPC service. Should be called with appropriate locking to
469 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
471void
472svc_destroy(struct svc_serv *serv)
473{
Chuck Lever46121cf2007-01-31 12:14:08 -0500474 dprintk("svc: svc_destroy(%s, %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 serv->sv_program->pg_name,
476 serv->sv_nrthreads);
477
478 if (serv->sv_nrthreads) {
479 if (--(serv->sv_nrthreads) != 0) {
480 svc_sock_update_bufs(serv);
481 return;
482 }
483 } else
484 printk("svc_destroy: no threads for serv=%p!\n", serv);
485
Greg Banks36bdfc82006-10-02 02:17:54 -0700486 del_timer_sync(&serv->sv_temptimer);
487
Tom Tucker7a182082007-12-30 21:07:53 -0600488 svc_close_all(&serv->sv_tempsocks);
NeilBrowncda1fd42007-03-06 01:42:22 -0800489
NeilBrownbc591cc2006-10-02 02:17:44 -0700490 if (serv->sv_shutdown)
491 serv->sv_shutdown(serv);
492
Tom Tucker7a182082007-12-30 21:07:53 -0600493 svc_close_all(&serv->sv_permsocks);
NeilBrowncda1fd42007-03-06 01:42:22 -0800494
495 BUG_ON(!list_empty(&serv->sv_permsocks));
496 BUG_ON(!list_empty(&serv->sv_tempsocks));
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800497
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 cache_clean_deferred(serv);
499
Greg Banks42a7fc42007-03-06 01:42:23 -0800500 if (svc_serv_is_pooled(serv))
501 svc_pool_map_put();
502
Chuck Lever7252d572008-08-18 19:34:08 -0400503 svc_unregister(serv);
Greg Banks3262c812006-10-02 02:17:58 -0700504 kfree(serv->sv_pools);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 kfree(serv);
506}
Trond Myklebust24c37672008-12-23 16:30:12 -0500507EXPORT_SYMBOL_GPL(svc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509/*
510 * Allocate an RPC server's buffer space.
511 * We allocate pages and place them in rq_argpages.
512 */
513static int
Eric Dumazet11fd1652011-07-28 20:04:09 +0200514svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
Chuck Lever0dc220f2008-04-14 12:27:52 -0400516 unsigned int pages, arghi;
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800517
Andy Adamsonba176862010-01-26 21:24:04 -0500518 /* bc_xprt uses fore channel allocated buffers */
519 if (svc_is_backchannel(rqstp))
520 return 1;
521
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700522 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
523 * We assume one is at most one page
524 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 arghi = 0;
Kris Katterjohn09a62662006-01-08 22:24:28 -0800526 BUG_ON(pages > RPCSVC_MAXPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 while (pages) {
Eric Dumazet11fd1652011-07-28 20:04:09 +0200528 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (!p)
530 break;
NeilBrown44524352006-10-04 02:15:46 -0700531 rqstp->rq_pages[arghi++] = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 pages--;
533 }
Chuck Lever0dc220f2008-04-14 12:27:52 -0400534 return pages == 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
537/*
538 * Release an RPC server buffer
539 */
540static void
541svc_release_buffer(struct svc_rqst *rqstp)
542{
Chuck Lever50c8bb12008-04-14 12:27:45 -0400543 unsigned int i;
544
545 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
NeilBrown44524352006-10-04 02:15:46 -0700546 if (rqstp->rq_pages[i])
547 put_page(rqstp->rq_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
549
Jeff Layton0113ab32008-01-29 10:30:54 -0500550struct svc_rqst *
Eric Dumazet11fd1652011-07-28 20:04:09 +0200551svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
Jeff Layton0113ab32008-01-29 10:30:54 -0500552{
553 struct svc_rqst *rqstp;
554
Eric Dumazet11fd1652011-07-28 20:04:09 +0200555 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node);
Jeff Layton0113ab32008-01-29 10:30:54 -0500556 if (!rqstp)
557 goto out_enomem;
558
559 init_waitqueue_head(&rqstp->rq_wait);
560
561 serv->sv_nrthreads++;
562 spin_lock_bh(&pool->sp_lock);
563 pool->sp_nrthreads++;
564 list_add(&rqstp->rq_all, &pool->sp_all_threads);
565 spin_unlock_bh(&pool->sp_lock);
566 rqstp->rq_server = serv;
567 rqstp->rq_pool = pool;
568
Eric Dumazet11fd1652011-07-28 20:04:09 +0200569 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
Jeff Layton0113ab32008-01-29 10:30:54 -0500570 if (!rqstp->rq_argp)
571 goto out_thread;
572
Eric Dumazet11fd1652011-07-28 20:04:09 +0200573 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node);
Jeff Layton0113ab32008-01-29 10:30:54 -0500574 if (!rqstp->rq_resp)
575 goto out_thread;
576
Eric Dumazet11fd1652011-07-28 20:04:09 +0200577 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
Jeff Layton0113ab32008-01-29 10:30:54 -0500578 goto out_thread;
579
580 return rqstp;
581out_thread:
582 svc_exit_thread(rqstp);
583out_enomem:
584 return ERR_PTR(-ENOMEM);
585}
Trond Myklebust24c37672008-12-23 16:30:12 -0500586EXPORT_SYMBOL_GPL(svc_prepare_thread);
Jeff Layton0113ab32008-01-29 10:30:54 -0500587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588/*
Greg Banksa7455442006-10-02 02:17:59 -0700589 * Choose a pool in which to create a new thread, for svc_set_num_threads
590 */
591static inline struct svc_pool *
592choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
593{
594 if (pool != NULL)
595 return pool;
596
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800597 return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
Greg Banksa7455442006-10-02 02:17:59 -0700598}
599
600/*
601 * Choose a thread to kill, for svc_set_num_threads
602 */
603static inline struct task_struct *
604choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
605{
606 unsigned int i;
607 struct task_struct *task = NULL;
608
609 if (pool != NULL) {
610 spin_lock_bh(&pool->sp_lock);
611 } else {
612 /* choose a pool in round-robin fashion */
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800613 for (i = 0; i < serv->sv_nrpools; i++) {
614 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
Greg Banksa7455442006-10-02 02:17:59 -0700615 spin_lock_bh(&pool->sp_lock);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800616 if (!list_empty(&pool->sp_all_threads))
617 goto found_pool;
Greg Banksa7455442006-10-02 02:17:59 -0700618 spin_unlock_bh(&pool->sp_lock);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800619 }
Greg Banksa7455442006-10-02 02:17:59 -0700620 return NULL;
621 }
622
623found_pool:
624 if (!list_empty(&pool->sp_all_threads)) {
625 struct svc_rqst *rqstp;
626
627 /*
628 * Remove from the pool->sp_all_threads list
629 * so we don't try to kill it again.
630 */
631 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
632 list_del_init(&rqstp->rq_all);
633 task = rqstp->rq_task;
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800634 }
Greg Banksa7455442006-10-02 02:17:59 -0700635 spin_unlock_bh(&pool->sp_lock);
636
637 return task;
638}
639
640/*
641 * Create or destroy enough new threads to make the number
642 * of threads the given number. If `pool' is non-NULL, applies
643 * only to threads in that pool, otherwise round-robins between
644 * all pools. Must be called with a svc_get() reference and
Neil Brownbedbdd82008-06-10 08:40:35 -0400645 * the BKL or another lock to protect access to svc_serv fields.
Greg Banksa7455442006-10-02 02:17:59 -0700646 *
647 * Destroying threads relies on the service threads filling in
648 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
649 * has been created using svc_create_pooled().
650 *
651 * Based on code that used to be in nfsd_svc() but tweaked
652 * to be pool-aware.
653 */
654int
655svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
656{
Jeff Layton9867d762008-06-10 08:40:38 -0400657 struct svc_rqst *rqstp;
658 struct task_struct *task;
659 struct svc_pool *chosen_pool;
Greg Banksa7455442006-10-02 02:17:59 -0700660 int error = 0;
661 unsigned int state = serv->sv_nrthreads-1;
Eric Dumazet11fd1652011-07-28 20:04:09 +0200662 int node;
Greg Banksa7455442006-10-02 02:17:59 -0700663
664 if (pool == NULL) {
665 /* The -1 assumes caller has done a svc_get() */
666 nrservs -= (serv->sv_nrthreads-1);
667 } else {
668 spin_lock_bh(&pool->sp_lock);
669 nrservs -= pool->sp_nrthreads;
670 spin_unlock_bh(&pool->sp_lock);
671 }
672
673 /* create new threads */
674 while (nrservs > 0) {
675 nrservs--;
Jeff Layton9867d762008-06-10 08:40:38 -0400676 chosen_pool = choose_pool(serv, pool, &state);
677
Eric Dumazet11fd1652011-07-28 20:04:09 +0200678 node = svc_pool_map_get_node(chosen_pool->sp_id);
679 rqstp = svc_prepare_thread(serv, chosen_pool, node);
Jeff Layton9867d762008-06-10 08:40:38 -0400680 if (IS_ERR(rqstp)) {
681 error = PTR_ERR(rqstp);
Greg Banksa7455442006-10-02 02:17:59 -0700682 break;
683 }
Jeff Layton9867d762008-06-10 08:40:38 -0400684
685 __module_get(serv->sv_module);
Eric Dumazet11fd1652011-07-28 20:04:09 +0200686 task = kthread_create_on_node(serv->sv_function, rqstp,
687 node, serv->sv_name);
Jeff Layton9867d762008-06-10 08:40:38 -0400688 if (IS_ERR(task)) {
689 error = PTR_ERR(task);
690 module_put(serv->sv_module);
691 svc_exit_thread(rqstp);
692 break;
693 }
694
695 rqstp->rq_task = task;
696 if (serv->sv_nrpools > 1)
697 svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
698
699 svc_sock_update_bufs(serv);
700 wake_up_process(task);
Greg Banksa7455442006-10-02 02:17:59 -0700701 }
702 /* destroy old threads */
703 while (nrservs < 0 &&
Jeff Layton9867d762008-06-10 08:40:38 -0400704 (task = choose_victim(serv, pool, &state)) != NULL) {
Jeff Laytona75c5d02008-06-10 08:40:39 -0400705 send_sig(SIGINT, task, 1);
Greg Banksa7455442006-10-02 02:17:59 -0700706 nrservs++;
707 }
708
709 return error;
710}
Trond Myklebust24c37672008-12-23 16:30:12 -0500711EXPORT_SYMBOL_GPL(svc_set_num_threads);
Greg Banksa7455442006-10-02 02:17:59 -0700712
713/*
Neil Brownbedbdd82008-06-10 08:40:35 -0400714 * Called from a server thread as it's exiting. Caller must hold the BKL or
715 * the "service mutex", whichever is appropriate for the service.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
717void
718svc_exit_thread(struct svc_rqst *rqstp)
719{
720 struct svc_serv *serv = rqstp->rq_server;
Greg Banks3262c812006-10-02 02:17:58 -0700721 struct svc_pool *pool = rqstp->rq_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 svc_release_buffer(rqstp);
Jesper Juhla51482b2005-11-08 09:41:34 -0800724 kfree(rqstp->rq_resp);
725 kfree(rqstp->rq_argp);
726 kfree(rqstp->rq_auth_data);
Greg Banks3262c812006-10-02 02:17:58 -0700727
728 spin_lock_bh(&pool->sp_lock);
729 pool->sp_nrthreads--;
Greg Banksa7455442006-10-02 02:17:59 -0700730 list_del(&rqstp->rq_all);
Greg Banks3262c812006-10-02 02:17:58 -0700731 spin_unlock_bh(&pool->sp_lock);
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 kfree(rqstp);
734
735 /* Release the server */
736 if (serv)
737 svc_destroy(serv);
738}
Trond Myklebust24c37672008-12-23 16:30:12 -0500739EXPORT_SYMBOL_GPL(svc_exit_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741/*
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500742 * Register an "inet" protocol family netid with the local
743 * rpcbind daemon via an rpcbind v4 SET request.
Chuck Levera26cfad2008-08-18 19:34:16 -0400744 *
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500745 * No netconfig infrastructure is available in the kernel, so
746 * we map IP_ protocol numbers to netids by hand.
Chuck Levera26cfad2008-08-18 19:34:16 -0400747 *
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500748 * Returns zero on success; a negative errno value is returned
749 * if any error occurs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 */
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500751static int __svc_rpcb_register4(const u32 program, const u32 version,
752 const unsigned short protocol,
753 const unsigned short port)
Chuck Levera26cfad2008-08-18 19:34:16 -0400754{
Chuck Levercadc0fa2009-03-18 20:47:36 -0400755 const struct sockaddr_in sin = {
Chuck Levera26cfad2008-08-18 19:34:16 -0400756 .sin_family = AF_INET,
757 .sin_addr.s_addr = htonl(INADDR_ANY),
758 .sin_port = htons(port),
759 };
Chuck Levercadc0fa2009-03-18 20:47:36 -0400760 const char *netid;
761 int error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500762
763 switch (protocol) {
764 case IPPROTO_UDP:
765 netid = RPCBIND_NETID_UDP;
766 break;
767 case IPPROTO_TCP:
768 netid = RPCBIND_NETID_TCP;
769 break;
770 default:
Chuck Leverba5c35e2009-03-18 20:46:59 -0400771 return -ENOPROTOOPT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500772 }
773
Chuck Levercadc0fa2009-03-18 20:47:36 -0400774 error = rpcb_v4_register(program, version,
775 (const struct sockaddr *)&sin, netid);
776
777 /*
778 * User space didn't support rpcbind v4, so retry this
779 * registration request with the legacy rpcbind v2 protocol.
780 */
781 if (error == -EPROTONOSUPPORT)
782 error = rpcb_register(program, version, protocol, port);
783
784 return error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500785}
786
Chuck Levercadc0fa2009-03-18 20:47:36 -0400787#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500788/*
789 * Register an "inet6" protocol family netid with the local
790 * rpcbind daemon via an rpcbind v4 SET request.
791 *
792 * No netconfig infrastructure is available in the kernel, so
793 * we map IP_ protocol numbers to netids by hand.
794 *
795 * Returns zero on success; a negative errno value is returned
796 * if any error occurs.
797 */
798static int __svc_rpcb_register6(const u32 program, const u32 version,
799 const unsigned short protocol,
800 const unsigned short port)
801{
Chuck Levercadc0fa2009-03-18 20:47:36 -0400802 const struct sockaddr_in6 sin6 = {
Chuck Levera26cfad2008-08-18 19:34:16 -0400803 .sin6_family = AF_INET6,
804 .sin6_addr = IN6ADDR_ANY_INIT,
805 .sin6_port = htons(port),
806 };
Chuck Levercadc0fa2009-03-18 20:47:36 -0400807 const char *netid;
808 int error;
Chuck Levera26cfad2008-08-18 19:34:16 -0400809
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500810 switch (protocol) {
811 case IPPROTO_UDP:
812 netid = RPCBIND_NETID_UDP6;
813 break;
814 case IPPROTO_TCP:
815 netid = RPCBIND_NETID_TCP6;
816 break;
817 default:
Chuck Leverba5c35e2009-03-18 20:46:59 -0400818 return -ENOPROTOOPT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500819 }
820
Chuck Levercadc0fa2009-03-18 20:47:36 -0400821 error = rpcb_v4_register(program, version,
822 (const struct sockaddr *)&sin6, netid);
823
824 /*
825 * User space didn't support rpcbind version 4, so we won't
826 * use a PF_INET6 listener.
827 */
828 if (error == -EPROTONOSUPPORT)
829 error = -EAFNOSUPPORT;
830
831 return error;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500832}
Chuck Levercadc0fa2009-03-18 20:47:36 -0400833#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500834
835/*
836 * Register a kernel RPC service via rpcbind version 4.
837 *
838 * Returns zero on success; a negative errno value is returned
839 * if any error occurs.
840 */
Chuck Lever363f7242009-03-18 20:47:44 -0400841static int __svc_register(const char *progname,
842 const u32 program, const u32 version,
Chuck Lever4b62e582009-03-18 20:46:06 -0400843 const int family,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500844 const unsigned short protocol,
845 const unsigned short port)
846{
Chuck Lever363f7242009-03-18 20:47:44 -0400847 int error = -EAFNOSUPPORT;
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500848
Chuck Levera26cfad2008-08-18 19:34:16 -0400849 switch (family) {
Chuck Lever4b62e582009-03-18 20:46:06 -0400850 case PF_INET:
Chuck Lever363f7242009-03-18 20:47:44 -0400851 error = __svc_rpcb_register4(program, version,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500852 protocol, port);
Chuck Levercadc0fa2009-03-18 20:47:36 -0400853 break;
854#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever4b62e582009-03-18 20:46:06 -0400855 case PF_INET6:
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500856 error = __svc_rpcb_register6(program, version,
857 protocol, port);
Chuck Levercadc0fa2009-03-18 20:47:36 -0400858#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Levera26cfad2008-08-18 19:34:16 -0400859 }
860
Chuck Lever363f7242009-03-18 20:47:44 -0400861 if (error < 0)
862 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
863 "service (errno %d).\n", progname, version, -error);
864 return error;
Chuck Levera26cfad2008-08-18 19:34:16 -0400865}
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500866
Chuck Levera26cfad2008-08-18 19:34:16 -0400867/**
868 * svc_register - register an RPC service with the local portmapper
869 * @serv: svc_serv struct for the service to register
Chuck Lever4b62e582009-03-18 20:46:06 -0400870 * @family: protocol family of service's listener socket
Chuck Levera26cfad2008-08-18 19:34:16 -0400871 * @proto: transport protocol number to advertise
872 * @port: port to advertise
873 *
Chuck Lever4b62e582009-03-18 20:46:06 -0400874 * Service is registered for any address in the passed-in protocol family
Chuck Levera26cfad2008-08-18 19:34:16 -0400875 */
Chuck Lever4b62e582009-03-18 20:46:06 -0400876int svc_register(const struct svc_serv *serv, const int family,
877 const unsigned short proto, const unsigned short port)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
879 struct svc_program *progp;
Chuck Leverea339d42007-10-26 13:32:56 -0400880 unsigned int i;
Chuck Lever14aeb212008-08-18 19:34:00 -0400881 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Chuck Lever7252d572008-08-18 19:34:08 -0400883 BUG_ON(proto == 0 && port == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700885 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
886 for (i = 0; i < progp->pg_nvers; i++) {
887 if (progp->pg_vers[i] == NULL)
888 continue;
889
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500890 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700891 progp->pg_name,
Chuck Lever2c7eb0b2008-09-15 16:27:23 -0500892 i,
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700893 proto == IPPROTO_UDP? "udp" : "tcp",
894 port,
Chuck Lever4b62e582009-03-18 20:46:06 -0400895 family,
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700896 progp->pg_vers[i]->vs_hidden?
897 " (but not telling portmap)" : "");
898
899 if (progp->pg_vers[i]->vs_hidden)
900 continue;
901
Chuck Lever363f7242009-03-18 20:47:44 -0400902 error = __svc_register(progp->pg_name, progp->pg_prog,
903 i, family, proto, port);
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700904 if (error < 0)
905 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
907 }
908
Chuck Lever7252d572008-08-18 19:34:08 -0400909 return error;
910}
911
Chuck Leverd5a86202009-03-18 20:47:29 -0400912/*
913 * If user space is running rpcbind, it should take the v4 UNSET
914 * and clear everything for this [program, version]. If user space
915 * is running portmap, it will reject the v4 UNSET, but won't have
916 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
917 * in this case to clear all existing entries for [program, version].
918 */
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400919static void __svc_unregister(const u32 program, const u32 version,
920 const char *progname)
921{
922 int error;
923
Chuck Leverd5a86202009-03-18 20:47:29 -0400924 error = rpcb_v4_register(program, version, NULL, "");
925
926 /*
927 * User space didn't support rpcbind v4, so retry this
928 * request with the legacy rpcbind v2 protocol.
929 */
930 if (error == -EPROTONOSUPPORT)
931 error = rpcb_register(program, version, 0, 0);
932
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400933 dprintk("svc: %s(%sv%u), error %d\n",
934 __func__, progname, version, error);
935}
936
Chuck Lever7252d572008-08-18 19:34:08 -0400937/*
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400938 * All netids, bind addresses and ports registered for [program, version]
939 * are removed from the local rpcbind database (if the service is not
940 * hidden) to make way for a new instance of the service.
Chuck Lever7252d572008-08-18 19:34:08 -0400941 *
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400942 * The result of unregistration is reported via dprintk for those who want
943 * verification of the result, but is otherwise not important.
Chuck Lever7252d572008-08-18 19:34:08 -0400944 */
945static void svc_unregister(const struct svc_serv *serv)
946{
947 struct svc_program *progp;
948 unsigned long flags;
949 unsigned int i;
Chuck Lever7252d572008-08-18 19:34:08 -0400950
951 clear_thread_flag(TIF_SIGPENDING);
952
953 for (progp = serv->sv_program; progp; progp = progp->pg_next) {
954 for (i = 0; i < progp->pg_nvers; i++) {
955 if (progp->pg_vers[i] == NULL)
956 continue;
957 if (progp->pg_vers[i]->vs_hidden)
958 continue;
959
Chuck Lever7402ab192011-05-09 15:22:55 -0400960 dprintk("svc: attempting to unregister %sv%u\n",
961 progp->pg_name, i);
Chuck Leverf6fb3f62008-09-25 11:56:57 -0400962 __svc_unregister(progp->pg_prog, i, progp->pg_name);
Chuck Lever7252d572008-08-18 19:34:08 -0400963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
965
Chuck Lever7252d572008-08-18 19:34:08 -0400966 spin_lock_irqsave(&current->sighand->siglock, flags);
967 recalc_sigpending();
968 spin_unlock_irqrestore(&current->sighand->siglock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
970
971/*
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +0100972 * Printk the given error with the address of the client that caused it.
973 */
974static int
975__attribute__ ((format (printf, 2, 3)))
976svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
977{
978 va_list args;
979 int r;
980 char buf[RPC_MAX_ADDRBUFLEN];
981
982 if (!net_ratelimit())
983 return 0;
984
985 printk(KERN_WARNING "svc: %s: ",
986 svc_print_addr(rqstp, buf, sizeof(buf)));
987
988 va_start(args, fmt);
989 r = vprintk(fmt, args);
990 va_end(args);
991
992 return r;
993}
994
995/*
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -0400996 * Common routine for processing the RPC request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -0400998static int
999svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
1001 struct svc_program *progp;
1002 struct svc_version *versp = NULL; /* compiler food */
1003 struct svc_procedure *procp = NULL;
NeilBrown6fb2b472006-10-02 02:17:50 -07001004 struct svc_serv *serv = rqstp->rq_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 kxdrproc_t xdr;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001006 __be32 *statp;
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001007 u32 prog, vers, proc;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001008 __be32 auth_stat, rpc_stat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 int auth_res;
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -07001010 __be32 *reply_statp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 rpc_stat = rpc_success;
1013
1014 if (argv->iov_len < 6*4)
1015 goto err_short_len;
1016
J. Bruce Fields5c04c462006-06-30 01:56:19 -07001017 /* Will be turned off only in gss privacy case: */
Jens Axboecf8208d2007-06-12 21:22:14 +02001018 rqstp->rq_splice_ok = 1;
Andy Adamson2f425872009-04-03 08:27:32 +03001019 /* Will be turned off only when NFSv4 Sessions are used */
1020 rqstp->rq_usedeferral = 1;
J. Bruce Fields9e701c62011-01-02 21:56:36 -05001021 rqstp->rq_dropme = false;
Tom Tuckere831fe62007-12-30 21:07:29 -06001022
1023 /* Setup reply header */
1024 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 svc_putu32(resv, rqstp->rq_xid);
1027
Alexey Dobriyan76994312006-09-26 22:28:46 -07001028 vers = svc_getnl(argv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 /* First words of reply: */
Alexey Dobriyan76994312006-09-26 22:28:46 -07001031 svc_putnl(resv, 1); /* REPLY */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (vers != 2) /* RPC version number */
1034 goto err_bad_rpc;
1035
1036 /* Save position in case we later decide to reject: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -07001037 reply_statp = resv->iov_base + resv->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Alexey Dobriyan76994312006-09-26 22:28:46 -07001039 svc_putnl(resv, 0); /* ACCEPT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Alexey Dobriyan76994312006-09-26 22:28:46 -07001041 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */
1042 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */
1043 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 progp = serv->sv_program;
NeilBrown80d188a2005-11-07 01:00:27 -08001046
1047 for (progp = serv->sv_program; progp; progp = progp->pg_next)
1048 if (prog == progp->pg_prog)
1049 break;
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /*
1052 * Decode auth data, and add verifier to reply buffer.
1053 * We do this before anything else in order to get a decent
1054 * auth verifier.
1055 */
1056 auth_res = svc_authenticate(rqstp, &auth_stat);
1057 /* Also give the program a chance to reject this call: */
NeilBrown80d188a2005-11-07 01:00:27 -08001058 if (auth_res == SVC_OK && progp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 auth_stat = rpc_autherr_badcred;
1060 auth_res = progp->pg_authenticate(rqstp);
1061 }
1062 switch (auth_res) {
1063 case SVC_OK:
1064 break;
1065 case SVC_GARBAGE:
Harshula Jayasuriyadd352102008-02-20 10:56:56 +11001066 goto err_garbage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 case SVC_SYSERR:
1068 rpc_stat = rpc_system_err;
1069 goto err_bad;
1070 case SVC_DENIED:
1071 goto err_bad_auth;
NeilBrown1ebede82010-08-12 17:04:07 +10001072 case SVC_CLOSE:
1073 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags))
1074 svc_close_xprt(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 case SVC_DROP:
1076 goto dropit;
1077 case SVC_COMPLETE:
1078 goto sendit;
1079 }
NeilBrown80d188a2005-11-07 01:00:27 -08001080
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +00001081 if (progp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto err_bad_prog;
1083
1084 if (vers >= progp->pg_nvers ||
1085 !(versp = progp->pg_vers[vers]))
1086 goto err_bad_vers;
1087
1088 procp = versp->vs_proc + proc;
1089 if (proc >= versp->vs_nproc || !procp->pc_func)
1090 goto err_bad_proc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 rqstp->rq_procinfo = procp;
1092
1093 /* Syntactic check complete */
1094 serv->sv_stats->rpccnt++;
1095
1096 /* Build the reply header. */
1097 statp = resv->iov_base +resv->iov_len;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001098 svc_putnl(resv, RPC_SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 /* Bump per-procedure stats counter */
1101 procp->pc_count++;
1102
1103 /* Initialize storage for argp and resp */
1104 memset(rqstp->rq_argp, 0, procp->pc_argsize);
1105 memset(rqstp->rq_resp, 0, procp->pc_ressize);
1106
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001107 /* un-reserve some of the out-queue now that we have a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 * better idea of reply size
1109 */
1110 if (procp->pc_xdrressize)
Jeff Laytoncd123012007-05-09 02:34:50 -07001111 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 /* Call the function that processes the request. */
1114 if (!versp->vs_dispatch) {
1115 /* Decode arguments */
1116 xdr = procp->pc_decode;
1117 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1118 goto err_garbage;
1119
1120 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1121
1122 /* Encode reply */
J. Bruce Fields9e701c62011-01-02 21:56:36 -05001123 if (rqstp->rq_dropme) {
NeilBrownd343fce2006-10-17 00:10:18 -07001124 if (procp->pc_release)
1125 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1126 goto dropit;
1127 }
Joe Perchesf64f9e72009-11-29 16:55:45 -08001128 if (*statp == rpc_success &&
1129 (xdr = procp->pc_encode) &&
1130 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 dprintk("svc: failed to encode reply\n");
1132 /* serv->sv_stats->rpcsystemerr++; */
1133 *statp = rpc_system_err;
1134 }
1135 } else {
1136 dprintk("svc: calling dispatcher\n");
1137 if (!versp->vs_dispatch(rqstp, statp)) {
1138 /* Release reply info */
1139 if (procp->pc_release)
1140 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1141 goto dropit;
1142 }
1143 }
1144
1145 /* Check RPC status result */
1146 if (*statp != rpc_success)
1147 resv->iov_len = ((void*)statp) - resv->iov_base + 4;
1148
1149 /* Release reply info */
1150 if (procp->pc_release)
1151 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1152
1153 if (procp->pc_encode == NULL)
1154 goto dropit;
1155
1156 sendit:
1157 if (svc_authorise(rqstp))
1158 goto dropit;
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001159 return 1; /* Caller can now send it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 dropit:
1162 svc_authorise(rqstp); /* doesn't hurt to call this twice */
1163 dprintk("svc: svc_process dropit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return 0;
1165
1166err_short_len:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001167 svc_printk(rqstp, "short len %Zd, dropping request\n",
1168 argv->iov_len);
NeilBrown34e9a632007-01-29 13:19:52 -08001169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 goto dropit; /* drop request */
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172err_bad_rpc:
1173 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001174 svc_putnl(resv, 1); /* REJECT */
1175 svc_putnl(resv, 0); /* RPC_MISMATCH */
1176 svc_putnl(resv, 2); /* Only RPCv2 supported */
1177 svc_putnl(resv, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 goto sendit;
1179
1180err_bad_auth:
1181 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1182 serv->sv_stats->rpcbadauth++;
1183 /* Restore write pointer to location of accept status: */
J.Bruce Fields8f8e05c2006-10-04 02:16:08 -07001184 xdr_ressize_check(rqstp, reply_statp);
Alexey Dobriyan76994312006-09-26 22:28:46 -07001185 svc_putnl(resv, 1); /* REJECT */
1186 svc_putnl(resv, 1); /* AUTH_ERROR */
1187 svc_putnl(resv, ntohl(auth_stat)); /* status */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 goto sendit;
1189
1190err_bad_prog:
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +00001191 dprintk("svc: unknown program %d\n", prog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001193 svc_putnl(resv, RPC_PROG_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 goto sendit;
1195
1196err_bad_vers:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001197 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
NeilBrown34e9a632007-01-29 13:19:52 -08001198 vers, prog, progp->pg_name);
1199
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001201 svc_putnl(resv, RPC_PROG_MISMATCH);
1202 svc_putnl(resv, progp->pg_lovers);
1203 svc_putnl(resv, progp->pg_hivers);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 goto sendit;
1205
1206err_bad_proc:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001207 svc_printk(rqstp, "unknown procedure (%d)\n", proc);
NeilBrown34e9a632007-01-29 13:19:52 -08001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001210 svc_putnl(resv, RPC_PROC_UNAVAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 goto sendit;
1212
1213err_garbage:
Dr. David Alan Gilbert354ecbb2007-08-25 16:09:27 +01001214 svc_printk(rqstp, "failed to decode args\n");
NeilBrown34e9a632007-01-29 13:19:52 -08001215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 rpc_stat = rpc_garbage_args;
1217err_bad:
1218 serv->sv_stats->rpcbadfmt++;
Alexey Dobriyan76994312006-09-26 22:28:46 -07001219 svc_putnl(resv, ntohl(rpc_stat));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 goto sendit;
1221}
Trond Myklebust24c37672008-12-23 16:30:12 -05001222EXPORT_SYMBOL_GPL(svc_process);
Greg Banks7adae482006-10-04 02:15:47 -07001223
1224/*
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001225 * Process the RPC request.
1226 */
1227int
1228svc_process(struct svc_rqst *rqstp)
1229{
1230 struct kvec *argv = &rqstp->rq_arg.head[0];
1231 struct kvec *resv = &rqstp->rq_res.head[0];
1232 struct svc_serv *serv = rqstp->rq_server;
1233 u32 dir;
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001234
1235 /*
1236 * Setup response xdr_buf.
1237 * Initially it has just one page
1238 */
1239 rqstp->rq_resused = 1;
1240 resv->iov_base = page_address(rqstp->rq_respages[0]);
1241 resv->iov_len = 0;
1242 rqstp->rq_res.pages = rqstp->rq_respages + 1;
1243 rqstp->rq_res.len = 0;
1244 rqstp->rq_res.page_base = 0;
1245 rqstp->rq_res.page_len = 0;
1246 rqstp->rq_res.buflen = PAGE_SIZE;
1247 rqstp->rq_res.tail[0].iov_base = NULL;
1248 rqstp->rq_res.tail[0].iov_len = 0;
1249
1250 rqstp->rq_xid = svc_getu32(argv);
1251
1252 dir = svc_getnl(argv);
1253 if (dir != 0) {
1254 /* direction != CALL */
1255 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1256 serv->sv_stats->rpcbadfmt++;
1257 svc_drop(rqstp);
1258 return 0;
1259 }
1260
Andy Adamson4b5b3ba2011-01-06 02:04:24 +00001261 /* Returns 1 for send, 0 for drop */
1262 if (svc_process_common(rqstp, argv, resv))
1263 return svc_send(rqstp);
1264 else {
1265 svc_drop(rqstp);
1266 return 0;
1267 }
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001268}
1269
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001270#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001271/*
1272 * Process a backchannel RPC request that arrived over an existing
1273 * outbound connection
1274 */
1275int
1276bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1277 struct svc_rqst *rqstp)
1278{
1279 struct kvec *argv = &rqstp->rq_arg.head[0];
1280 struct kvec *resv = &rqstp->rq_res.head[0];
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001281
1282 /* Build the svc_rqst used by the common processing routine */
Andy Adamson4a19de02011-01-06 02:04:35 +00001283 rqstp->rq_xprt = serv->sv_bc_xprt;
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001284 rqstp->rq_xid = req->rq_xid;
1285 rqstp->rq_prot = req->rq_xprt->prot;
1286 rqstp->rq_server = serv;
1287
1288 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1289 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1290 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1291 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1292
1293 /* reset result send buffer "put" position */
1294 resv->iov_len = 0;
1295
1296 if (rqstp->rq_prot != IPPROTO_TCP) {
1297 printk(KERN_ERR "No support for Non-TCP transports!\n");
1298 BUG();
1299 }
1300
1301 /*
1302 * Skip the next two words because they've already been
1303 * processed in the trasport
1304 */
1305 svc_getu32(argv); /* XID */
1306 svc_getnl(argv); /* CALLDIR */
1307
Andy Adamson4b5b3ba2011-01-06 02:04:24 +00001308 /* Returns 1 for send, 0 for drop */
1309 if (svc_process_common(rqstp, argv, resv)) {
1310 memcpy(&req->rq_snd_buf, &rqstp->rq_res,
1311 sizeof(req->rq_snd_buf));
1312 return bc_send(req);
1313 } else {
1314 /* Nothing to do to drop request */
1315 return 0;
1316 }
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001317}
Trond Myklebust0d961aa2011-07-13 19:24:15 -04001318EXPORT_SYMBOL_GPL(bc_svc_process);
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001319#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Ricardo Labiaga4d6bbb62009-04-01 09:23:07 -04001320
Ricardo Labiaga1cad7ea2009-04-01 09:23:06 -04001321/*
Greg Banks7adae482006-10-04 02:15:47 -07001322 * Return (transport-specific) limit on the rpc payload.
1323 */
1324u32 svc_max_payload(const struct svc_rqst *rqstp)
1325{
Tom Tucker49023152007-12-30 21:07:21 -06001326 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
Greg Banks7adae482006-10-04 02:15:47 -07001327
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001328 if (rqstp->rq_server->sv_max_payload < max)
1329 max = rqstp->rq_server->sv_max_payload;
Greg Banks7adae482006-10-04 02:15:47 -07001330 return max;
1331}
1332EXPORT_SYMBOL_GPL(svc_max_payload);