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