Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svc.c | |
3 | * | |
4 | * High-level RPC service routines | |
5 | * | |
6 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
bfd24160 GB |
7 | * |
8 | * Multiple threads pools and NUMAisation | |
9 | * Copyright (c) 2006 Silicon Graphics, Inc. | |
10 | * by Greg Banks <gnb@melbourne.sgi.com> | |
1da177e4 LT |
11 | */ |
12 | ||
13 | #include <linux/linkage.h> | |
14 | #include <linux/sched.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/net.h> | |
17 | #include <linux/in.h> | |
18 | #include <linux/mm.h> | |
a7455442 GB |
19 | #include <linux/interrupt.h> |
20 | #include <linux/module.h> | |
df3825c5 | 21 | #include <linux/sched.h> |
1da177e4 LT |
22 | |
23 | #include <linux/sunrpc/types.h> | |
24 | #include <linux/sunrpc/xdr.h> | |
25 | #include <linux/sunrpc/stats.h> | |
26 | #include <linux/sunrpc/svcsock.h> | |
27 | #include <linux/sunrpc/clnt.h> | |
28 | ||
29 | #define RPCDBG_FACILITY RPCDBG_SVCDSP | |
1da177e4 | 30 | |
42a7fc4a GB |
31 | #define svc_serv_is_pooled(serv) ((serv)->sv_function) |
32 | ||
bfd24160 GB |
33 | /* |
34 | * Mode for mapping cpus to pools. | |
35 | */ | |
36 | enum { | |
42a7fc4a | 37 | SVC_POOL_AUTO = -1, /* choose one of the others */ |
bfd24160 GB |
38 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool |
39 | * (legacy & UP mode) */ | |
40 | SVC_POOL_PERCPU, /* one pool per cpu */ | |
41 | SVC_POOL_PERNODE /* one pool per numa node */ | |
42 | }; | |
42a7fc4a | 43 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
bfd24160 GB |
44 | |
45 | /* | |
46 | * Structure for mapping cpus to pools and vice versa. | |
47 | * Setup once during sunrpc initialisation. | |
48 | */ | |
49 | static struct svc_pool_map { | |
42a7fc4a | 50 | int count; /* How many svc_servs use us */ |
bfd24160 GB |
51 | int mode; /* Note: int not enum to avoid |
52 | * warnings about "enumeration value | |
53 | * not handled in switch" */ | |
54 | unsigned int npools; | |
55 | unsigned int *pool_to; /* maps pool id to cpu or node */ | |
56 | unsigned int *to_pool; /* maps cpu or node to pool id */ | |
57 | } svc_pool_map = { | |
42a7fc4a GB |
58 | .count = 0, |
59 | .mode = SVC_POOL_DEFAULT | |
bfd24160 | 60 | }; |
42a7fc4a GB |
61 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
62 | ||
63 | static int | |
64 | param_set_pool_mode(const char *val, struct kernel_param *kp) | |
65 | { | |
66 | int *ip = (int *)kp->arg; | |
67 | struct svc_pool_map *m = &svc_pool_map; | |
68 | int err; | |
69 | ||
70 | mutex_lock(&svc_pool_map_mutex); | |
71 | ||
72 | err = -EBUSY; | |
73 | if (m->count) | |
74 | goto out; | |
75 | ||
76 | err = 0; | |
77 | if (!strncmp(val, "auto", 4)) | |
78 | *ip = SVC_POOL_AUTO; | |
79 | else if (!strncmp(val, "global", 6)) | |
80 | *ip = SVC_POOL_GLOBAL; | |
81 | else if (!strncmp(val, "percpu", 6)) | |
82 | *ip = SVC_POOL_PERCPU; | |
83 | else if (!strncmp(val, "pernode", 7)) | |
84 | *ip = SVC_POOL_PERNODE; | |
85 | else | |
86 | err = -EINVAL; | |
87 | ||
88 | out: | |
89 | mutex_unlock(&svc_pool_map_mutex); | |
90 | return err; | |
91 | } | |
92 | ||
93 | static int | |
94 | param_get_pool_mode(char *buf, struct kernel_param *kp) | |
95 | { | |
96 | int *ip = (int *)kp->arg; | |
97 | ||
98 | switch (*ip) | |
99 | { | |
100 | case SVC_POOL_AUTO: | |
101 | return strlcpy(buf, "auto", 20); | |
102 | case SVC_POOL_GLOBAL: | |
103 | return strlcpy(buf, "global", 20); | |
104 | case SVC_POOL_PERCPU: | |
105 | return strlcpy(buf, "percpu", 20); | |
106 | case SVC_POOL_PERNODE: | |
107 | return strlcpy(buf, "pernode", 20); | |
108 | default: | |
109 | return sprintf(buf, "%d", *ip); | |
110 | } | |
111 | } | |
bfd24160 | 112 | |
42a7fc4a GB |
113 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
114 | &svc_pool_map.mode, 0644); | |
bfd24160 GB |
115 | |
116 | /* | |
117 | * Detect best pool mapping mode heuristically, | |
118 | * according to the machine's topology. | |
119 | */ | |
120 | static int | |
121 | svc_pool_map_choose_mode(void) | |
122 | { | |
123 | unsigned int node; | |
124 | ||
125 | if (num_online_nodes() > 1) { | |
126 | /* | |
127 | * Actually have multiple NUMA nodes, | |
128 | * so split pools on NUMA node boundaries | |
129 | */ | |
130 | return SVC_POOL_PERNODE; | |
131 | } | |
132 | ||
133 | node = any_online_node(node_online_map); | |
134 | if (nr_cpus_node(node) > 2) { | |
135 | /* | |
136 | * Non-trivial SMP, or CONFIG_NUMA on | |
137 | * non-NUMA hardware, e.g. with a generic | |
138 | * x86_64 kernel on Xeons. In this case we | |
139 | * want to divide the pools on cpu boundaries. | |
140 | */ | |
141 | return SVC_POOL_PERCPU; | |
142 | } | |
143 | ||
144 | /* default: one global pool */ | |
145 | return SVC_POOL_GLOBAL; | |
146 | } | |
147 | ||
148 | /* | |
149 | * Allocate the to_pool[] and pool_to[] arrays. | |
150 | * Returns 0 on success or an errno. | |
151 | */ | |
152 | static int | |
153 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
154 | { | |
155 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
156 | if (!m->to_pool) | |
157 | goto fail; | |
158 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
159 | if (!m->pool_to) | |
160 | goto fail_free; | |
161 | ||
162 | return 0; | |
163 | ||
164 | fail_free: | |
165 | kfree(m->to_pool); | |
166 | fail: | |
167 | return -ENOMEM; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
172 | * Returns number of pools or <0 on error. | |
173 | */ | |
174 | static int | |
175 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
176 | { | |
53b8a315 | 177 | unsigned int maxpools = nr_cpu_ids; |
bfd24160 GB |
178 | unsigned int pidx = 0; |
179 | unsigned int cpu; | |
180 | int err; | |
181 | ||
182 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
183 | if (err) | |
184 | return err; | |
185 | ||
186 | for_each_online_cpu(cpu) { | |
187 | BUG_ON(pidx > maxpools); | |
188 | m->to_pool[cpu] = pidx; | |
189 | m->pool_to[pidx] = cpu; | |
190 | pidx++; | |
191 | } | |
192 | /* cpus brought online later all get mapped to pool0, sorry */ | |
193 | ||
194 | return pidx; | |
195 | }; | |
196 | ||
197 | ||
198 | /* | |
199 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
200 | * Returns number of pools or <0 on error. | |
201 | */ | |
202 | static int | |
203 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
204 | { | |
74c7aa8b | 205 | unsigned int maxpools = nr_node_ids; |
bfd24160 GB |
206 | unsigned int pidx = 0; |
207 | unsigned int node; | |
208 | int err; | |
209 | ||
210 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
211 | if (err) | |
212 | return err; | |
213 | ||
214 | for_each_node_with_cpus(node) { | |
215 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
216 | BUG_ON(pidx > maxpools); | |
217 | m->to_pool[node] = pidx; | |
218 | m->pool_to[pidx] = node; | |
219 | pidx++; | |
220 | } | |
221 | /* nodes brought online later all get mapped to pool0, sorry */ | |
222 | ||
223 | return pidx; | |
224 | } | |
225 | ||
226 | ||
227 | /* | |
42a7fc4a GB |
228 | * Add a reference to the global map of cpus to pools (and |
229 | * vice versa). Initialise the map if we're the first user. | |
230 | * Returns the number of pools. | |
bfd24160 GB |
231 | */ |
232 | static unsigned int | |
42a7fc4a | 233 | svc_pool_map_get(void) |
bfd24160 GB |
234 | { |
235 | struct svc_pool_map *m = &svc_pool_map; | |
236 | int npools = -1; | |
237 | ||
42a7fc4a GB |
238 | mutex_lock(&svc_pool_map_mutex); |
239 | ||
240 | if (m->count++) { | |
241 | mutex_unlock(&svc_pool_map_mutex); | |
bfd24160 | 242 | return m->npools; |
42a7fc4a | 243 | } |
bfd24160 | 244 | |
42a7fc4a GB |
245 | if (m->mode == SVC_POOL_AUTO) |
246 | m->mode = svc_pool_map_choose_mode(); | |
bfd24160 GB |
247 | |
248 | switch (m->mode) { | |
249 | case SVC_POOL_PERCPU: | |
250 | npools = svc_pool_map_init_percpu(m); | |
251 | break; | |
252 | case SVC_POOL_PERNODE: | |
253 | npools = svc_pool_map_init_pernode(m); | |
254 | break; | |
255 | } | |
256 | ||
257 | if (npools < 0) { | |
258 | /* default, or memory allocation failure */ | |
259 | npools = 1; | |
260 | m->mode = SVC_POOL_GLOBAL; | |
261 | } | |
262 | m->npools = npools; | |
263 | ||
42a7fc4a | 264 | mutex_unlock(&svc_pool_map_mutex); |
bfd24160 GB |
265 | return m->npools; |
266 | } | |
267 | ||
42a7fc4a GB |
268 | |
269 | /* | |
270 | * Drop a reference to the global map of cpus to pools. | |
271 | * When the last reference is dropped, the map data is | |
272 | * freed; this allows the sysadmin to change the pool | |
273 | * mode using the pool_mode module option without | |
274 | * rebooting or re-loading sunrpc.ko. | |
275 | */ | |
276 | static void | |
277 | svc_pool_map_put(void) | |
278 | { | |
279 | struct svc_pool_map *m = &svc_pool_map; | |
280 | ||
281 | mutex_lock(&svc_pool_map_mutex); | |
282 | ||
283 | if (!--m->count) { | |
284 | m->mode = SVC_POOL_DEFAULT; | |
285 | kfree(m->to_pool); | |
286 | kfree(m->pool_to); | |
287 | m->npools = 0; | |
288 | } | |
289 | ||
290 | mutex_unlock(&svc_pool_map_mutex); | |
291 | } | |
292 | ||
293 | ||
bfd24160 GB |
294 | /* |
295 | * Set the current thread's cpus_allowed mask so that it | |
296 | * will only run on cpus in the given pool. | |
297 | * | |
298 | * Returns 1 and fills in oldmask iff a cpumask was applied. | |
299 | */ | |
300 | static inline int | |
301 | svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask) | |
302 | { | |
303 | struct svc_pool_map *m = &svc_pool_map; | |
304 | unsigned int node; /* or cpu */ | |
305 | ||
306 | /* | |
307 | * The caller checks for sv_nrpools > 1, which | |
42a7fc4a | 308 | * implies that we've been initialized. |
bfd24160 | 309 | */ |
42a7fc4a | 310 | BUG_ON(m->count == 0); |
bfd24160 GB |
311 | |
312 | switch (m->mode) | |
313 | { | |
314 | default: | |
315 | return 0; | |
316 | case SVC_POOL_PERCPU: | |
317 | node = m->pool_to[pidx]; | |
318 | *oldmask = current->cpus_allowed; | |
319 | set_cpus_allowed(current, cpumask_of_cpu(node)); | |
320 | return 1; | |
321 | case SVC_POOL_PERNODE: | |
322 | node = m->pool_to[pidx]; | |
323 | *oldmask = current->cpus_allowed; | |
324 | set_cpus_allowed(current, node_to_cpumask(node)); | |
325 | return 1; | |
326 | } | |
327 | } | |
328 | ||
329 | /* | |
330 | * Use the mapping mode to choose a pool for a given CPU. | |
331 | * Used when enqueueing an incoming RPC. Always returns | |
332 | * a non-NULL pool pointer. | |
333 | */ | |
334 | struct svc_pool * | |
335 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | |
336 | { | |
337 | struct svc_pool_map *m = &svc_pool_map; | |
338 | unsigned int pidx = 0; | |
339 | ||
340 | /* | |
42a7fc4a | 341 | * An uninitialised map happens in a pure client when |
bfd24160 GB |
342 | * lockd is brought up, so silently treat it the |
343 | * same as SVC_POOL_GLOBAL. | |
344 | */ | |
42a7fc4a GB |
345 | if (svc_serv_is_pooled(serv)) { |
346 | switch (m->mode) { | |
347 | case SVC_POOL_PERCPU: | |
348 | pidx = m->to_pool[cpu]; | |
349 | break; | |
350 | case SVC_POOL_PERNODE: | |
351 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
352 | break; | |
353 | } | |
bfd24160 GB |
354 | } |
355 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | |
356 | } | |
357 | ||
358 | ||
1da177e4 LT |
359 | /* |
360 | * Create an RPC service | |
361 | */ | |
a7455442 GB |
362 | static struct svc_serv * |
363 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
bc591ccf | 364 | void (*shutdown)(struct svc_serv *serv)) |
1da177e4 LT |
365 | { |
366 | struct svc_serv *serv; | |
367 | int vers; | |
368 | unsigned int xdrsize; | |
3262c816 | 369 | unsigned int i; |
1da177e4 | 370 | |
0da974f4 | 371 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 372 | return NULL; |
9ba02638 | 373 | serv->sv_name = prog->pg_name; |
1da177e4 LT |
374 | serv->sv_program = prog; |
375 | serv->sv_nrthreads = 1; | |
376 | serv->sv_stats = prog->pg_stats; | |
c6b0a9f8 N |
377 | if (bufsize > RPCSVC_MAXPAYLOAD) |
378 | bufsize = RPCSVC_MAXPAYLOAD; | |
379 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
380 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
bc591ccf | 381 | serv->sv_shutdown = shutdown; |
1da177e4 | 382 | xdrsize = 0; |
9ba02638 AG |
383 | while (prog) { |
384 | prog->pg_lovers = prog->pg_nvers-1; | |
385 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
386 | if (prog->pg_vers[vers]) { | |
387 | prog->pg_hivers = vers; | |
388 | if (prog->pg_lovers > vers) | |
389 | prog->pg_lovers = vers; | |
390 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
391 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
392 | } | |
393 | prog = prog->pg_next; | |
394 | } | |
1da177e4 | 395 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
396 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
397 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
36bdfc8b | 398 | init_timer(&serv->sv_temptimer); |
1da177e4 LT |
399 | spin_lock_init(&serv->sv_lock); |
400 | ||
a7455442 | 401 | serv->sv_nrpools = npools; |
3262c816 | 402 | serv->sv_pools = |
cd861280 | 403 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
404 | GFP_KERNEL); |
405 | if (!serv->sv_pools) { | |
406 | kfree(serv); | |
407 | return NULL; | |
408 | } | |
409 | ||
410 | for (i = 0; i < serv->sv_nrpools; i++) { | |
411 | struct svc_pool *pool = &serv->sv_pools[i]; | |
412 | ||
46121cf7 | 413 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
414 | i, serv->sv_name); |
415 | ||
416 | pool->sp_id = i; | |
417 | INIT_LIST_HEAD(&pool->sp_threads); | |
418 | INIT_LIST_HEAD(&pool->sp_sockets); | |
a7455442 | 419 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
420 | spin_lock_init(&pool->sp_lock); |
421 | } | |
422 | ||
423 | ||
1da177e4 LT |
424 | /* Remove any stale portmap registrations */ |
425 | svc_register(serv, 0, 0); | |
426 | ||
427 | return serv; | |
428 | } | |
429 | ||
a7455442 GB |
430 | struct svc_serv * |
431 | svc_create(struct svc_program *prog, unsigned int bufsize, | |
432 | void (*shutdown)(struct svc_serv *serv)) | |
433 | { | |
434 | return __svc_create(prog, bufsize, /*npools*/1, shutdown); | |
435 | } | |
436 | ||
437 | struct svc_serv * | |
438 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | |
439 | void (*shutdown)(struct svc_serv *serv), | |
440 | svc_thread_fn func, int sig, struct module *mod) | |
441 | { | |
442 | struct svc_serv *serv; | |
42a7fc4a | 443 | unsigned int npools = svc_pool_map_get(); |
a7455442 | 444 | |
bfd24160 | 445 | serv = __svc_create(prog, bufsize, npools, shutdown); |
a7455442 GB |
446 | |
447 | if (serv != NULL) { | |
448 | serv->sv_function = func; | |
449 | serv->sv_kill_signal = sig; | |
450 | serv->sv_module = mod; | |
451 | } | |
452 | ||
453 | return serv; | |
454 | } | |
455 | ||
1da177e4 | 456 | /* |
3262c816 | 457 | * Destroy an RPC service. Should be called with the BKL held |
1da177e4 LT |
458 | */ |
459 | void | |
460 | svc_destroy(struct svc_serv *serv) | |
461 | { | |
462 | struct svc_sock *svsk; | |
cda1fd4a | 463 | struct svc_sock *tmp; |
1da177e4 | 464 | |
46121cf7 | 465 | dprintk("svc: svc_destroy(%s, %d)\n", |
1da177e4 LT |
466 | serv->sv_program->pg_name, |
467 | serv->sv_nrthreads); | |
468 | ||
469 | if (serv->sv_nrthreads) { | |
470 | if (--(serv->sv_nrthreads) != 0) { | |
471 | svc_sock_update_bufs(serv); | |
472 | return; | |
473 | } | |
474 | } else | |
475 | printk("svc_destroy: no threads for serv=%p!\n", serv); | |
476 | ||
36bdfc8b GB |
477 | del_timer_sync(&serv->sv_temptimer); |
478 | ||
cda1fd4a N |
479 | list_for_each_entry_safe(svsk, tmp, &serv->sv_tempsocks, sk_list) |
480 | svc_force_close_socket(svsk); | |
481 | ||
bc591ccf N |
482 | if (serv->sv_shutdown) |
483 | serv->sv_shutdown(serv); | |
484 | ||
cda1fd4a N |
485 | list_for_each_entry_safe(svsk, tmp, &serv->sv_permsocks, sk_list) |
486 | svc_force_close_socket(svsk); | |
487 | ||
488 | BUG_ON(!list_empty(&serv->sv_permsocks)); | |
489 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | |
cca5172a | 490 | |
1da177e4 LT |
491 | cache_clean_deferred(serv); |
492 | ||
42a7fc4a GB |
493 | if (svc_serv_is_pooled(serv)) |
494 | svc_pool_map_put(); | |
495 | ||
1da177e4 LT |
496 | /* Unregister service with the portmapper */ |
497 | svc_register(serv, 0, 0); | |
3262c816 | 498 | kfree(serv->sv_pools); |
1da177e4 LT |
499 | kfree(serv); |
500 | } | |
501 | ||
502 | /* | |
503 | * Allocate an RPC server's buffer space. | |
504 | * We allocate pages and place them in rq_argpages. | |
505 | */ | |
506 | static int | |
507 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) | |
508 | { | |
509 | int pages; | |
510 | int arghi; | |
cca5172a | 511 | |
c6b0a9f8 N |
512 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
513 | * We assume one is at most one page | |
514 | */ | |
1da177e4 | 515 | arghi = 0; |
09a62660 | 516 | BUG_ON(pages > RPCSVC_MAXPAGES); |
1da177e4 LT |
517 | while (pages) { |
518 | struct page *p = alloc_page(GFP_KERNEL); | |
519 | if (!p) | |
520 | break; | |
44524359 | 521 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
522 | pages--; |
523 | } | |
1da177e4 LT |
524 | return ! pages; |
525 | } | |
526 | ||
527 | /* | |
528 | * Release an RPC server buffer | |
529 | */ | |
530 | static void | |
531 | svc_release_buffer(struct svc_rqst *rqstp) | |
532 | { | |
44524359 N |
533 | int i; |
534 | for (i=0; i<ARRAY_SIZE(rqstp->rq_pages); i++) | |
535 | if (rqstp->rq_pages[i]) | |
536 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
537 | } |
538 | ||
539 | /* | |
3262c816 | 540 | * Create a thread in the given pool. Caller must hold BKL. |
bfd24160 GB |
541 | * On a NUMA or SMP machine, with a multi-pool serv, the thread |
542 | * will be restricted to run on the cpus belonging to the pool. | |
1da177e4 | 543 | */ |
3262c816 GB |
544 | static int |
545 | __svc_create_thread(svc_thread_fn func, struct svc_serv *serv, | |
546 | struct svc_pool *pool) | |
1da177e4 LT |
547 | { |
548 | struct svc_rqst *rqstp; | |
549 | int error = -ENOMEM; | |
bfd24160 GB |
550 | int have_oldmask = 0; |
551 | cpumask_t oldmask; | |
1da177e4 | 552 | |
0da974f4 | 553 | rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL); |
1da177e4 LT |
554 | if (!rqstp) |
555 | goto out; | |
556 | ||
1da177e4 LT |
557 | init_waitqueue_head(&rqstp->rq_wait); |
558 | ||
12fe2c58 JJ |
559 | if (!(rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL)) |
560 | || !(rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL)) | |
c6b0a9f8 | 561 | || !svc_init_buffer(rqstp, serv->sv_max_mesg)) |
1da177e4 LT |
562 | goto out_thread; |
563 | ||
564 | serv->sv_nrthreads++; | |
3262c816 GB |
565 | spin_lock_bh(&pool->sp_lock); |
566 | pool->sp_nrthreads++; | |
a7455442 | 567 | list_add(&rqstp->rq_all, &pool->sp_all_threads); |
3262c816 | 568 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 569 | rqstp->rq_server = serv; |
3262c816 | 570 | rqstp->rq_pool = pool; |
bfd24160 GB |
571 | |
572 | if (serv->sv_nrpools > 1) | |
573 | have_oldmask = svc_pool_map_set_cpumask(pool->sp_id, &oldmask); | |
574 | ||
1da177e4 | 575 | error = kernel_thread((int (*)(void *)) func, rqstp, 0); |
bfd24160 GB |
576 | |
577 | if (have_oldmask) | |
578 | set_cpus_allowed(current, oldmask); | |
579 | ||
1da177e4 LT |
580 | if (error < 0) |
581 | goto out_thread; | |
582 | svc_sock_update_bufs(serv); | |
583 | error = 0; | |
584 | out: | |
585 | return error; | |
586 | ||
587 | out_thread: | |
588 | svc_exit_thread(rqstp); | |
589 | goto out; | |
590 | } | |
591 | ||
592 | /* | |
3262c816 GB |
593 | * Create a thread in the default pool. Caller must hold BKL. |
594 | */ | |
595 | int | |
596 | svc_create_thread(svc_thread_fn func, struct svc_serv *serv) | |
597 | { | |
598 | return __svc_create_thread(func, serv, &serv->sv_pools[0]); | |
599 | } | |
600 | ||
a7455442 GB |
601 | /* |
602 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
603 | */ | |
604 | static inline struct svc_pool * | |
605 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
606 | { | |
607 | if (pool != NULL) | |
608 | return pool; | |
609 | ||
cca5172a | 610 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
611 | } |
612 | ||
613 | /* | |
614 | * Choose a thread to kill, for svc_set_num_threads | |
615 | */ | |
616 | static inline struct task_struct * | |
617 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
618 | { | |
619 | unsigned int i; | |
620 | struct task_struct *task = NULL; | |
621 | ||
622 | if (pool != NULL) { | |
623 | spin_lock_bh(&pool->sp_lock); | |
624 | } else { | |
625 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
626 | for (i = 0; i < serv->sv_nrpools; i++) { |
627 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 628 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
629 | if (!list_empty(&pool->sp_all_threads)) |
630 | goto found_pool; | |
a7455442 | 631 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 632 | } |
a7455442 GB |
633 | return NULL; |
634 | } | |
635 | ||
636 | found_pool: | |
637 | if (!list_empty(&pool->sp_all_threads)) { | |
638 | struct svc_rqst *rqstp; | |
639 | ||
640 | /* | |
641 | * Remove from the pool->sp_all_threads list | |
642 | * so we don't try to kill it again. | |
643 | */ | |
644 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
645 | list_del_init(&rqstp->rq_all); | |
646 | task = rqstp->rq_task; | |
cca5172a | 647 | } |
a7455442 GB |
648 | spin_unlock_bh(&pool->sp_lock); |
649 | ||
650 | return task; | |
651 | } | |
652 | ||
653 | /* | |
654 | * Create or destroy enough new threads to make the number | |
655 | * of threads the given number. If `pool' is non-NULL, applies | |
656 | * only to threads in that pool, otherwise round-robins between | |
657 | * all pools. Must be called with a svc_get() reference and | |
658 | * the BKL held. | |
659 | * | |
660 | * Destroying threads relies on the service threads filling in | |
661 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | |
662 | * has been created using svc_create_pooled(). | |
663 | * | |
664 | * Based on code that used to be in nfsd_svc() but tweaked | |
665 | * to be pool-aware. | |
666 | */ | |
667 | int | |
668 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
669 | { | |
670 | struct task_struct *victim; | |
671 | int error = 0; | |
672 | unsigned int state = serv->sv_nrthreads-1; | |
673 | ||
674 | if (pool == NULL) { | |
675 | /* The -1 assumes caller has done a svc_get() */ | |
676 | nrservs -= (serv->sv_nrthreads-1); | |
677 | } else { | |
678 | spin_lock_bh(&pool->sp_lock); | |
679 | nrservs -= pool->sp_nrthreads; | |
680 | spin_unlock_bh(&pool->sp_lock); | |
681 | } | |
682 | ||
683 | /* create new threads */ | |
684 | while (nrservs > 0) { | |
685 | nrservs--; | |
686 | __module_get(serv->sv_module); | |
687 | error = __svc_create_thread(serv->sv_function, serv, | |
688 | choose_pool(serv, pool, &state)); | |
689 | if (error < 0) { | |
690 | module_put(serv->sv_module); | |
691 | break; | |
692 | } | |
693 | } | |
694 | /* destroy old threads */ | |
695 | while (nrservs < 0 && | |
696 | (victim = choose_victim(serv, pool, &state)) != NULL) { | |
697 | send_sig(serv->sv_kill_signal, victim, 1); | |
698 | nrservs++; | |
699 | } | |
700 | ||
701 | return error; | |
702 | } | |
703 | ||
3262c816 GB |
704 | /* |
705 | * Called from a server thread as it's exiting. Caller must hold BKL. | |
1da177e4 LT |
706 | */ |
707 | void | |
708 | svc_exit_thread(struct svc_rqst *rqstp) | |
709 | { | |
710 | struct svc_serv *serv = rqstp->rq_server; | |
3262c816 | 711 | struct svc_pool *pool = rqstp->rq_pool; |
1da177e4 LT |
712 | |
713 | svc_release_buffer(rqstp); | |
a51482bd JJ |
714 | kfree(rqstp->rq_resp); |
715 | kfree(rqstp->rq_argp); | |
716 | kfree(rqstp->rq_auth_data); | |
3262c816 GB |
717 | |
718 | spin_lock_bh(&pool->sp_lock); | |
719 | pool->sp_nrthreads--; | |
a7455442 | 720 | list_del(&rqstp->rq_all); |
3262c816 GB |
721 | spin_unlock_bh(&pool->sp_lock); |
722 | ||
1da177e4 LT |
723 | kfree(rqstp); |
724 | ||
725 | /* Release the server */ | |
726 | if (serv) | |
727 | svc_destroy(serv); | |
728 | } | |
729 | ||
730 | /* | |
731 | * Register an RPC service with the local portmapper. | |
cca5172a | 732 | * To unregister a service, call this routine with |
1da177e4 LT |
733 | * proto and port == 0. |
734 | */ | |
735 | int | |
736 | svc_register(struct svc_serv *serv, int proto, unsigned short port) | |
737 | { | |
738 | struct svc_program *progp; | |
739 | unsigned long flags; | |
740 | int i, error = 0, dummy; | |
741 | ||
1da177e4 LT |
742 | if (!port) |
743 | clear_thread_flag(TIF_SIGPENDING); | |
744 | ||
bc5fea42 OK |
745 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
746 | for (i = 0; i < progp->pg_nvers; i++) { | |
747 | if (progp->pg_vers[i] == NULL) | |
748 | continue; | |
749 | ||
46121cf7 | 750 | dprintk("svc: svc_register(%s, %s, %d, %d)%s\n", |
bc5fea42 OK |
751 | progp->pg_name, |
752 | proto == IPPROTO_UDP? "udp" : "tcp", | |
753 | port, | |
754 | i, | |
755 | progp->pg_vers[i]->vs_hidden? | |
756 | " (but not telling portmap)" : ""); | |
757 | ||
758 | if (progp->pg_vers[i]->vs_hidden) | |
759 | continue; | |
760 | ||
26080014 | 761 | error = rpcb_register(progp->pg_prog, i, proto, port, &dummy); |
bc5fea42 OK |
762 | if (error < 0) |
763 | break; | |
764 | if (port && !dummy) { | |
765 | error = -EACCES; | |
766 | break; | |
767 | } | |
1da177e4 LT |
768 | } |
769 | } | |
770 | ||
771 | if (!port) { | |
772 | spin_lock_irqsave(¤t->sighand->siglock, flags); | |
773 | recalc_sigpending(); | |
774 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
775 | } | |
776 | ||
777 | return error; | |
778 | } | |
779 | ||
354ecbb9 DDAG |
780 | /* |
781 | * Printk the given error with the address of the client that caused it. | |
782 | */ | |
783 | static int | |
784 | __attribute__ ((format (printf, 2, 3))) | |
785 | svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) | |
786 | { | |
787 | va_list args; | |
788 | int r; | |
789 | char buf[RPC_MAX_ADDRBUFLEN]; | |
790 | ||
791 | if (!net_ratelimit()) | |
792 | return 0; | |
793 | ||
794 | printk(KERN_WARNING "svc: %s: ", | |
795 | svc_print_addr(rqstp, buf, sizeof(buf))); | |
796 | ||
797 | va_start(args, fmt); | |
798 | r = vprintk(fmt, args); | |
799 | va_end(args); | |
800 | ||
801 | return r; | |
802 | } | |
803 | ||
1da177e4 LT |
804 | /* |
805 | * Process the RPC request. | |
806 | */ | |
807 | int | |
6fb2b47f | 808 | svc_process(struct svc_rqst *rqstp) |
1da177e4 LT |
809 | { |
810 | struct svc_program *progp; | |
811 | struct svc_version *versp = NULL; /* compiler food */ | |
812 | struct svc_procedure *procp = NULL; | |
813 | struct kvec * argv = &rqstp->rq_arg.head[0]; | |
814 | struct kvec * resv = &rqstp->rq_res.head[0]; | |
6fb2b47f | 815 | struct svc_serv *serv = rqstp->rq_server; |
1da177e4 | 816 | kxdrproc_t xdr; |
d8ed029d AD |
817 | __be32 *statp; |
818 | u32 dir, prog, vers, proc; | |
819 | __be32 auth_stat, rpc_stat; | |
1da177e4 | 820 | int auth_res; |
8f8e05c5 | 821 | __be32 *reply_statp; |
1da177e4 LT |
822 | |
823 | rpc_stat = rpc_success; | |
824 | ||
825 | if (argv->iov_len < 6*4) | |
826 | goto err_short_len; | |
827 | ||
828 | /* setup response xdr_buf. | |
cca5172a | 829 | * Initially it has just one page |
1da177e4 | 830 | */ |
44524359 | 831 | rqstp->rq_resused = 1; |
1da177e4 LT |
832 | resv->iov_base = page_address(rqstp->rq_respages[0]); |
833 | resv->iov_len = 0; | |
44524359 | 834 | rqstp->rq_res.pages = rqstp->rq_respages + 1; |
1da177e4 LT |
835 | rqstp->rq_res.len = 0; |
836 | rqstp->rq_res.page_base = 0; | |
837 | rqstp->rq_res.page_len = 0; | |
334ccfd5 | 838 | rqstp->rq_res.buflen = PAGE_SIZE; |
7c9fdcfb | 839 | rqstp->rq_res.tail[0].iov_base = NULL; |
1da177e4 | 840 | rqstp->rq_res.tail[0].iov_len = 0; |
5c04c46a | 841 | /* Will be turned off only in gss privacy case: */ |
cf8208d0 | 842 | rqstp->rq_splice_ok = 1; |
e831fe65 TT |
843 | |
844 | /* Setup reply header */ | |
845 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); | |
1da177e4 LT |
846 | |
847 | rqstp->rq_xid = svc_getu32(argv); | |
848 | svc_putu32(resv, rqstp->rq_xid); | |
849 | ||
76994313 AD |
850 | dir = svc_getnl(argv); |
851 | vers = svc_getnl(argv); | |
1da177e4 LT |
852 | |
853 | /* First words of reply: */ | |
76994313 | 854 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 LT |
855 | |
856 | if (dir != 0) /* direction != CALL */ | |
857 | goto err_bad_dir; | |
858 | if (vers != 2) /* RPC version number */ | |
859 | goto err_bad_rpc; | |
860 | ||
861 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 862 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 863 | |
76994313 | 864 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 865 | |
76994313 AD |
866 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
867 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | |
868 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 LT |
869 | |
870 | progp = serv->sv_program; | |
80d188a6 N |
871 | |
872 | for (progp = serv->sv_program; progp; progp = progp->pg_next) | |
873 | if (prog == progp->pg_prog) | |
874 | break; | |
875 | ||
1da177e4 LT |
876 | /* |
877 | * Decode auth data, and add verifier to reply buffer. | |
878 | * We do this before anything else in order to get a decent | |
879 | * auth verifier. | |
880 | */ | |
881 | auth_res = svc_authenticate(rqstp, &auth_stat); | |
882 | /* Also give the program a chance to reject this call: */ | |
80d188a6 | 883 | if (auth_res == SVC_OK && progp) { |
1da177e4 LT |
884 | auth_stat = rpc_autherr_badcred; |
885 | auth_res = progp->pg_authenticate(rqstp); | |
886 | } | |
887 | switch (auth_res) { | |
888 | case SVC_OK: | |
889 | break; | |
890 | case SVC_GARBAGE: | |
891 | rpc_stat = rpc_garbage_args; | |
892 | goto err_bad; | |
893 | case SVC_SYSERR: | |
894 | rpc_stat = rpc_system_err; | |
895 | goto err_bad; | |
896 | case SVC_DENIED: | |
897 | goto err_bad_auth; | |
898 | case SVC_DROP: | |
899 | goto dropit; | |
900 | case SVC_COMPLETE: | |
901 | goto sendit; | |
902 | } | |
80d188a6 | 903 | |
9ba02638 | 904 | if (progp == NULL) |
1da177e4 LT |
905 | goto err_bad_prog; |
906 | ||
907 | if (vers >= progp->pg_nvers || | |
908 | !(versp = progp->pg_vers[vers])) | |
909 | goto err_bad_vers; | |
910 | ||
911 | procp = versp->vs_proc + proc; | |
912 | if (proc >= versp->vs_nproc || !procp->pc_func) | |
913 | goto err_bad_proc; | |
914 | rqstp->rq_server = serv; | |
915 | rqstp->rq_procinfo = procp; | |
916 | ||
917 | /* Syntactic check complete */ | |
918 | serv->sv_stats->rpccnt++; | |
919 | ||
920 | /* Build the reply header. */ | |
921 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 922 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 LT |
923 | |
924 | /* Bump per-procedure stats counter */ | |
925 | procp->pc_count++; | |
926 | ||
927 | /* Initialize storage for argp and resp */ | |
928 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | |
929 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | |
930 | ||
cca5172a | 931 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
932 | * better idea of reply size |
933 | */ | |
934 | if (procp->pc_xdrressize) | |
cd123012 | 935 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1da177e4 LT |
936 | |
937 | /* Call the function that processes the request. */ | |
938 | if (!versp->vs_dispatch) { | |
939 | /* Decode arguments */ | |
940 | xdr = procp->pc_decode; | |
941 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | |
942 | goto err_garbage; | |
943 | ||
944 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
945 | ||
946 | /* Encode reply */ | |
d343fce1 N |
947 | if (*statp == rpc_drop_reply) { |
948 | if (procp->pc_release) | |
949 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
950 | goto dropit; | |
951 | } | |
1da177e4 LT |
952 | if (*statp == rpc_success && (xdr = procp->pc_encode) |
953 | && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | |
954 | dprintk("svc: failed to encode reply\n"); | |
955 | /* serv->sv_stats->rpcsystemerr++; */ | |
956 | *statp = rpc_system_err; | |
957 | } | |
958 | } else { | |
959 | dprintk("svc: calling dispatcher\n"); | |
960 | if (!versp->vs_dispatch(rqstp, statp)) { | |
961 | /* Release reply info */ | |
962 | if (procp->pc_release) | |
963 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
964 | goto dropit; | |
965 | } | |
966 | } | |
967 | ||
968 | /* Check RPC status result */ | |
969 | if (*statp != rpc_success) | |
970 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
971 | ||
972 | /* Release reply info */ | |
973 | if (procp->pc_release) | |
974 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
975 | ||
976 | if (procp->pc_encode == NULL) | |
977 | goto dropit; | |
978 | ||
979 | sendit: | |
980 | if (svc_authorise(rqstp)) | |
981 | goto dropit; | |
982 | return svc_send(rqstp); | |
983 | ||
984 | dropit: | |
985 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
986 | dprintk("svc: svc_process dropit\n"); | |
987 | svc_drop(rqstp); | |
988 | return 0; | |
989 | ||
990 | err_short_len: | |
354ecbb9 DDAG |
991 | svc_printk(rqstp, "short len %Zd, dropping request\n", |
992 | argv->iov_len); | |
34e9a63b | 993 | |
1da177e4 LT |
994 | goto dropit; /* drop request */ |
995 | ||
996 | err_bad_dir: | |
354ecbb9 | 997 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); |
34e9a63b | 998 | |
1da177e4 LT |
999 | serv->sv_stats->rpcbadfmt++; |
1000 | goto dropit; /* drop request */ | |
1001 | ||
1002 | err_bad_rpc: | |
1003 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
1004 | svc_putnl(resv, 1); /* REJECT */ |
1005 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
1006 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
1007 | svc_putnl(resv, 2); | |
1da177e4 LT |
1008 | goto sendit; |
1009 | ||
1010 | err_bad_auth: | |
1011 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | |
1012 | serv->sv_stats->rpcbadauth++; | |
1013 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 1014 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
1015 | svc_putnl(resv, 1); /* REJECT */ |
1016 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
1017 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | |
1da177e4 LT |
1018 | goto sendit; |
1019 | ||
1020 | err_bad_prog: | |
9ba02638 | 1021 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 1022 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1023 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
1024 | goto sendit; |
1025 | ||
1026 | err_bad_vers: | |
354ecbb9 | 1027 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
34e9a63b N |
1028 | vers, prog, progp->pg_name); |
1029 | ||
1da177e4 | 1030 | serv->sv_stats->rpcbadfmt++; |
76994313 AD |
1031 | svc_putnl(resv, RPC_PROG_MISMATCH); |
1032 | svc_putnl(resv, progp->pg_lovers); | |
1033 | svc_putnl(resv, progp->pg_hivers); | |
1da177e4 LT |
1034 | goto sendit; |
1035 | ||
1036 | err_bad_proc: | |
354ecbb9 | 1037 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); |
34e9a63b | 1038 | |
1da177e4 | 1039 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1040 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
1041 | goto sendit; |
1042 | ||
1043 | err_garbage: | |
354ecbb9 | 1044 | svc_printk(rqstp, "failed to decode args\n"); |
34e9a63b | 1045 | |
1da177e4 LT |
1046 | rpc_stat = rpc_garbage_args; |
1047 | err_bad: | |
1048 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 1049 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
1050 | goto sendit; |
1051 | } | |
7adae489 GB |
1052 | |
1053 | /* | |
1054 | * Return (transport-specific) limit on the rpc payload. | |
1055 | */ | |
1056 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
1057 | { | |
49023155 | 1058 | u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; |
7adae489 | 1059 | |
c6b0a9f8 N |
1060 | if (rqstp->rq_server->sv_max_payload < max) |
1061 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
1062 | return max; |
1063 | } | |
1064 | EXPORT_SYMBOL_GPL(svc_max_payload); |