]>
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 | |
1da177e4 | 29 | |
bfd24160 GB |
30 | /* |
31 | * Mode for mapping cpus to pools. | |
32 | */ | |
33 | enum { | |
34 | SVC_POOL_NONE = -1, /* uninitialised, choose one of the others */ | |
35 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool | |
36 | * (legacy & UP mode) */ | |
37 | SVC_POOL_PERCPU, /* one pool per cpu */ | |
38 | SVC_POOL_PERNODE /* one pool per numa node */ | |
39 | }; | |
40 | ||
41 | /* | |
42 | * Structure for mapping cpus to pools and vice versa. | |
43 | * Setup once during sunrpc initialisation. | |
44 | */ | |
45 | static struct svc_pool_map { | |
46 | int mode; /* Note: int not enum to avoid | |
47 | * warnings about "enumeration value | |
48 | * not handled in switch" */ | |
49 | unsigned int npools; | |
50 | unsigned int *pool_to; /* maps pool id to cpu or node */ | |
51 | unsigned int *to_pool; /* maps cpu or node to pool id */ | |
52 | } svc_pool_map = { | |
53 | .mode = SVC_POOL_NONE | |
54 | }; | |
55 | ||
56 | ||
57 | /* | |
58 | * Detect best pool mapping mode heuristically, | |
59 | * according to the machine's topology. | |
60 | */ | |
61 | static int | |
62 | svc_pool_map_choose_mode(void) | |
63 | { | |
64 | unsigned int node; | |
65 | ||
66 | if (num_online_nodes() > 1) { | |
67 | /* | |
68 | * Actually have multiple NUMA nodes, | |
69 | * so split pools on NUMA node boundaries | |
70 | */ | |
71 | return SVC_POOL_PERNODE; | |
72 | } | |
73 | ||
74 | node = any_online_node(node_online_map); | |
75 | if (nr_cpus_node(node) > 2) { | |
76 | /* | |
77 | * Non-trivial SMP, or CONFIG_NUMA on | |
78 | * non-NUMA hardware, e.g. with a generic | |
79 | * x86_64 kernel on Xeons. In this case we | |
80 | * want to divide the pools on cpu boundaries. | |
81 | */ | |
82 | return SVC_POOL_PERCPU; | |
83 | } | |
84 | ||
85 | /* default: one global pool */ | |
86 | return SVC_POOL_GLOBAL; | |
87 | } | |
88 | ||
89 | /* | |
90 | * Allocate the to_pool[] and pool_to[] arrays. | |
91 | * Returns 0 on success or an errno. | |
92 | */ | |
93 | static int | |
94 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
95 | { | |
96 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
97 | if (!m->to_pool) | |
98 | goto fail; | |
99 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
100 | if (!m->pool_to) | |
101 | goto fail_free; | |
102 | ||
103 | return 0; | |
104 | ||
105 | fail_free: | |
106 | kfree(m->to_pool); | |
107 | fail: | |
108 | return -ENOMEM; | |
109 | } | |
110 | ||
111 | /* | |
112 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
113 | * Returns number of pools or <0 on error. | |
114 | */ | |
115 | static int | |
116 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
117 | { | |
118 | unsigned int maxpools = highest_possible_processor_id()+1; | |
119 | unsigned int pidx = 0; | |
120 | unsigned int cpu; | |
121 | int err; | |
122 | ||
123 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
124 | if (err) | |
125 | return err; | |
126 | ||
127 | for_each_online_cpu(cpu) { | |
128 | BUG_ON(pidx > maxpools); | |
129 | m->to_pool[cpu] = pidx; | |
130 | m->pool_to[pidx] = cpu; | |
131 | pidx++; | |
132 | } | |
133 | /* cpus brought online later all get mapped to pool0, sorry */ | |
134 | ||
135 | return pidx; | |
136 | }; | |
137 | ||
138 | ||
139 | /* | |
140 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
141 | * Returns number of pools or <0 on error. | |
142 | */ | |
143 | static int | |
144 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
145 | { | |
146 | unsigned int maxpools = highest_possible_node_id()+1; | |
147 | unsigned int pidx = 0; | |
148 | unsigned int node; | |
149 | int err; | |
150 | ||
151 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
152 | if (err) | |
153 | return err; | |
154 | ||
155 | for_each_node_with_cpus(node) { | |
156 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
157 | BUG_ON(pidx > maxpools); | |
158 | m->to_pool[node] = pidx; | |
159 | m->pool_to[pidx] = node; | |
160 | pidx++; | |
161 | } | |
162 | /* nodes brought online later all get mapped to pool0, sorry */ | |
163 | ||
164 | return pidx; | |
165 | } | |
166 | ||
167 | ||
168 | /* | |
169 | * Build the global map of cpus to pools and vice versa. | |
170 | */ | |
171 | static unsigned int | |
172 | svc_pool_map_init(void) | |
173 | { | |
174 | struct svc_pool_map *m = &svc_pool_map; | |
175 | int npools = -1; | |
176 | ||
177 | if (m->mode != SVC_POOL_NONE) | |
178 | return m->npools; | |
179 | ||
180 | m->mode = svc_pool_map_choose_mode(); | |
181 | ||
182 | switch (m->mode) { | |
183 | case SVC_POOL_PERCPU: | |
184 | npools = svc_pool_map_init_percpu(m); | |
185 | break; | |
186 | case SVC_POOL_PERNODE: | |
187 | npools = svc_pool_map_init_pernode(m); | |
188 | break; | |
189 | } | |
190 | ||
191 | if (npools < 0) { | |
192 | /* default, or memory allocation failure */ | |
193 | npools = 1; | |
194 | m->mode = SVC_POOL_GLOBAL; | |
195 | } | |
196 | m->npools = npools; | |
197 | ||
198 | return m->npools; | |
199 | } | |
200 | ||
201 | /* | |
202 | * Set the current thread's cpus_allowed mask so that it | |
203 | * will only run on cpus in the given pool. | |
204 | * | |
205 | * Returns 1 and fills in oldmask iff a cpumask was applied. | |
206 | */ | |
207 | static inline int | |
208 | svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask) | |
209 | { | |
210 | struct svc_pool_map *m = &svc_pool_map; | |
211 | unsigned int node; /* or cpu */ | |
212 | ||
213 | /* | |
214 | * The caller checks for sv_nrpools > 1, which | |
215 | * implies that we've been initialized and the | |
216 | * map mode is not NONE. | |
217 | */ | |
218 | BUG_ON(m->mode == SVC_POOL_NONE); | |
219 | ||
220 | switch (m->mode) | |
221 | { | |
222 | default: | |
223 | return 0; | |
224 | case SVC_POOL_PERCPU: | |
225 | node = m->pool_to[pidx]; | |
226 | *oldmask = current->cpus_allowed; | |
227 | set_cpus_allowed(current, cpumask_of_cpu(node)); | |
228 | return 1; | |
229 | case SVC_POOL_PERNODE: | |
230 | node = m->pool_to[pidx]; | |
231 | *oldmask = current->cpus_allowed; | |
232 | set_cpus_allowed(current, node_to_cpumask(node)); | |
233 | return 1; | |
234 | } | |
235 | } | |
236 | ||
237 | /* | |
238 | * Use the mapping mode to choose a pool for a given CPU. | |
239 | * Used when enqueueing an incoming RPC. Always returns | |
240 | * a non-NULL pool pointer. | |
241 | */ | |
242 | struct svc_pool * | |
243 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | |
244 | { | |
245 | struct svc_pool_map *m = &svc_pool_map; | |
246 | unsigned int pidx = 0; | |
247 | ||
248 | /* | |
249 | * SVC_POOL_NONE happens in a pure client when | |
250 | * lockd is brought up, so silently treat it the | |
251 | * same as SVC_POOL_GLOBAL. | |
252 | */ | |
253 | ||
254 | switch (m->mode) { | |
255 | case SVC_POOL_PERCPU: | |
256 | pidx = m->to_pool[cpu]; | |
257 | break; | |
258 | case SVC_POOL_PERNODE: | |
259 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
260 | break; | |
261 | } | |
262 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | |
263 | } | |
264 | ||
265 | ||
1da177e4 LT |
266 | /* |
267 | * Create an RPC service | |
268 | */ | |
a7455442 GB |
269 | static struct svc_serv * |
270 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
bc591ccf | 271 | void (*shutdown)(struct svc_serv *serv)) |
1da177e4 LT |
272 | { |
273 | struct svc_serv *serv; | |
274 | int vers; | |
275 | unsigned int xdrsize; | |
3262c816 | 276 | unsigned int i; |
1da177e4 | 277 | |
0da974f4 | 278 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 279 | return NULL; |
9ba02638 | 280 | serv->sv_name = prog->pg_name; |
1da177e4 LT |
281 | serv->sv_program = prog; |
282 | serv->sv_nrthreads = 1; | |
283 | serv->sv_stats = prog->pg_stats; | |
c6b0a9f8 N |
284 | if (bufsize > RPCSVC_MAXPAYLOAD) |
285 | bufsize = RPCSVC_MAXPAYLOAD; | |
286 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
287 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
bc591ccf | 288 | serv->sv_shutdown = shutdown; |
1da177e4 | 289 | xdrsize = 0; |
9ba02638 AG |
290 | while (prog) { |
291 | prog->pg_lovers = prog->pg_nvers-1; | |
292 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
293 | if (prog->pg_vers[vers]) { | |
294 | prog->pg_hivers = vers; | |
295 | if (prog->pg_lovers > vers) | |
296 | prog->pg_lovers = vers; | |
297 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
298 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
299 | } | |
300 | prog = prog->pg_next; | |
301 | } | |
1da177e4 | 302 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
303 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
304 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
36bdfc8b | 305 | init_timer(&serv->sv_temptimer); |
1da177e4 LT |
306 | spin_lock_init(&serv->sv_lock); |
307 | ||
a7455442 | 308 | serv->sv_nrpools = npools; |
3262c816 | 309 | serv->sv_pools = |
cd861280 | 310 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
311 | GFP_KERNEL); |
312 | if (!serv->sv_pools) { | |
313 | kfree(serv); | |
314 | return NULL; | |
315 | } | |
316 | ||
317 | for (i = 0; i < serv->sv_nrpools; i++) { | |
318 | struct svc_pool *pool = &serv->sv_pools[i]; | |
319 | ||
46121cf7 | 320 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
321 | i, serv->sv_name); |
322 | ||
323 | pool->sp_id = i; | |
324 | INIT_LIST_HEAD(&pool->sp_threads); | |
325 | INIT_LIST_HEAD(&pool->sp_sockets); | |
a7455442 | 326 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
327 | spin_lock_init(&pool->sp_lock); |
328 | } | |
329 | ||
330 | ||
1da177e4 LT |
331 | /* Remove any stale portmap registrations */ |
332 | svc_register(serv, 0, 0); | |
333 | ||
334 | return serv; | |
335 | } | |
336 | ||
a7455442 GB |
337 | struct svc_serv * |
338 | svc_create(struct svc_program *prog, unsigned int bufsize, | |
339 | void (*shutdown)(struct svc_serv *serv)) | |
340 | { | |
341 | return __svc_create(prog, bufsize, /*npools*/1, shutdown); | |
342 | } | |
343 | ||
344 | struct svc_serv * | |
345 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | |
346 | void (*shutdown)(struct svc_serv *serv), | |
347 | svc_thread_fn func, int sig, struct module *mod) | |
348 | { | |
349 | struct svc_serv *serv; | |
bfd24160 | 350 | unsigned int npools = svc_pool_map_init(); |
a7455442 | 351 | |
bfd24160 | 352 | serv = __svc_create(prog, bufsize, npools, shutdown); |
a7455442 GB |
353 | |
354 | if (serv != NULL) { | |
355 | serv->sv_function = func; | |
356 | serv->sv_kill_signal = sig; | |
357 | serv->sv_module = mod; | |
358 | } | |
359 | ||
360 | return serv; | |
361 | } | |
362 | ||
1da177e4 | 363 | /* |
3262c816 | 364 | * Destroy an RPC service. Should be called with the BKL held |
1da177e4 LT |
365 | */ |
366 | void | |
367 | svc_destroy(struct svc_serv *serv) | |
368 | { | |
369 | struct svc_sock *svsk; | |
370 | ||
46121cf7 | 371 | dprintk("svc: svc_destroy(%s, %d)\n", |
1da177e4 LT |
372 | serv->sv_program->pg_name, |
373 | serv->sv_nrthreads); | |
374 | ||
375 | if (serv->sv_nrthreads) { | |
376 | if (--(serv->sv_nrthreads) != 0) { | |
377 | svc_sock_update_bufs(serv); | |
378 | return; | |
379 | } | |
380 | } else | |
381 | printk("svc_destroy: no threads for serv=%p!\n", serv); | |
382 | ||
36bdfc8b GB |
383 | del_timer_sync(&serv->sv_temptimer); |
384 | ||
1da177e4 LT |
385 | while (!list_empty(&serv->sv_tempsocks)) { |
386 | svsk = list_entry(serv->sv_tempsocks.next, | |
387 | struct svc_sock, | |
388 | sk_list); | |
aaf68cfb | 389 | svc_close_socket(svsk); |
1da177e4 | 390 | } |
bc591ccf N |
391 | if (serv->sv_shutdown) |
392 | serv->sv_shutdown(serv); | |
393 | ||
1da177e4 LT |
394 | while (!list_empty(&serv->sv_permsocks)) { |
395 | svsk = list_entry(serv->sv_permsocks.next, | |
396 | struct svc_sock, | |
397 | sk_list); | |
aaf68cfb | 398 | svc_close_socket(svsk); |
1da177e4 | 399 | } |
cca5172a | 400 | |
1da177e4 LT |
401 | cache_clean_deferred(serv); |
402 | ||
403 | /* Unregister service with the portmapper */ | |
404 | svc_register(serv, 0, 0); | |
3262c816 | 405 | kfree(serv->sv_pools); |
1da177e4 LT |
406 | kfree(serv); |
407 | } | |
408 | ||
409 | /* | |
410 | * Allocate an RPC server's buffer space. | |
411 | * We allocate pages and place them in rq_argpages. | |
412 | */ | |
413 | static int | |
414 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) | |
415 | { | |
416 | int pages; | |
417 | int arghi; | |
cca5172a | 418 | |
c6b0a9f8 N |
419 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
420 | * We assume one is at most one page | |
421 | */ | |
1da177e4 | 422 | arghi = 0; |
09a62660 | 423 | BUG_ON(pages > RPCSVC_MAXPAGES); |
1da177e4 LT |
424 | while (pages) { |
425 | struct page *p = alloc_page(GFP_KERNEL); | |
426 | if (!p) | |
427 | break; | |
44524359 | 428 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
429 | pages--; |
430 | } | |
1da177e4 LT |
431 | return ! pages; |
432 | } | |
433 | ||
434 | /* | |
435 | * Release an RPC server buffer | |
436 | */ | |
437 | static void | |
438 | svc_release_buffer(struct svc_rqst *rqstp) | |
439 | { | |
44524359 N |
440 | int i; |
441 | for (i=0; i<ARRAY_SIZE(rqstp->rq_pages); i++) | |
442 | if (rqstp->rq_pages[i]) | |
443 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
444 | } |
445 | ||
446 | /* | |
3262c816 | 447 | * Create a thread in the given pool. Caller must hold BKL. |
bfd24160 GB |
448 | * On a NUMA or SMP machine, with a multi-pool serv, the thread |
449 | * will be restricted to run on the cpus belonging to the pool. | |
1da177e4 | 450 | */ |
3262c816 GB |
451 | static int |
452 | __svc_create_thread(svc_thread_fn func, struct svc_serv *serv, | |
453 | struct svc_pool *pool) | |
1da177e4 LT |
454 | { |
455 | struct svc_rqst *rqstp; | |
456 | int error = -ENOMEM; | |
bfd24160 GB |
457 | int have_oldmask = 0; |
458 | cpumask_t oldmask; | |
1da177e4 | 459 | |
0da974f4 | 460 | rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL); |
1da177e4 LT |
461 | if (!rqstp) |
462 | goto out; | |
463 | ||
1da177e4 LT |
464 | init_waitqueue_head(&rqstp->rq_wait); |
465 | ||
12fe2c58 JJ |
466 | if (!(rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL)) |
467 | || !(rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL)) | |
c6b0a9f8 | 468 | || !svc_init_buffer(rqstp, serv->sv_max_mesg)) |
1da177e4 LT |
469 | goto out_thread; |
470 | ||
471 | serv->sv_nrthreads++; | |
3262c816 GB |
472 | spin_lock_bh(&pool->sp_lock); |
473 | pool->sp_nrthreads++; | |
a7455442 | 474 | list_add(&rqstp->rq_all, &pool->sp_all_threads); |
3262c816 | 475 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 476 | rqstp->rq_server = serv; |
3262c816 | 477 | rqstp->rq_pool = pool; |
bfd24160 GB |
478 | |
479 | if (serv->sv_nrpools > 1) | |
480 | have_oldmask = svc_pool_map_set_cpumask(pool->sp_id, &oldmask); | |
481 | ||
1da177e4 | 482 | error = kernel_thread((int (*)(void *)) func, rqstp, 0); |
bfd24160 GB |
483 | |
484 | if (have_oldmask) | |
485 | set_cpus_allowed(current, oldmask); | |
486 | ||
1da177e4 LT |
487 | if (error < 0) |
488 | goto out_thread; | |
489 | svc_sock_update_bufs(serv); | |
490 | error = 0; | |
491 | out: | |
492 | return error; | |
493 | ||
494 | out_thread: | |
495 | svc_exit_thread(rqstp); | |
496 | goto out; | |
497 | } | |
498 | ||
499 | /* | |
3262c816 GB |
500 | * Create a thread in the default pool. Caller must hold BKL. |
501 | */ | |
502 | int | |
503 | svc_create_thread(svc_thread_fn func, struct svc_serv *serv) | |
504 | { | |
505 | return __svc_create_thread(func, serv, &serv->sv_pools[0]); | |
506 | } | |
507 | ||
a7455442 GB |
508 | /* |
509 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
510 | */ | |
511 | static inline struct svc_pool * | |
512 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
513 | { | |
514 | if (pool != NULL) | |
515 | return pool; | |
516 | ||
cca5172a | 517 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
518 | } |
519 | ||
520 | /* | |
521 | * Choose a thread to kill, for svc_set_num_threads | |
522 | */ | |
523 | static inline struct task_struct * | |
524 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
525 | { | |
526 | unsigned int i; | |
527 | struct task_struct *task = NULL; | |
528 | ||
529 | if (pool != NULL) { | |
530 | spin_lock_bh(&pool->sp_lock); | |
531 | } else { | |
532 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
533 | for (i = 0; i < serv->sv_nrpools; i++) { |
534 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 535 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
536 | if (!list_empty(&pool->sp_all_threads)) |
537 | goto found_pool; | |
a7455442 | 538 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 539 | } |
a7455442 GB |
540 | return NULL; |
541 | } | |
542 | ||
543 | found_pool: | |
544 | if (!list_empty(&pool->sp_all_threads)) { | |
545 | struct svc_rqst *rqstp; | |
546 | ||
547 | /* | |
548 | * Remove from the pool->sp_all_threads list | |
549 | * so we don't try to kill it again. | |
550 | */ | |
551 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
552 | list_del_init(&rqstp->rq_all); | |
553 | task = rqstp->rq_task; | |
cca5172a | 554 | } |
a7455442 GB |
555 | spin_unlock_bh(&pool->sp_lock); |
556 | ||
557 | return task; | |
558 | } | |
559 | ||
560 | /* | |
561 | * Create or destroy enough new threads to make the number | |
562 | * of threads the given number. If `pool' is non-NULL, applies | |
563 | * only to threads in that pool, otherwise round-robins between | |
564 | * all pools. Must be called with a svc_get() reference and | |
565 | * the BKL held. | |
566 | * | |
567 | * Destroying threads relies on the service threads filling in | |
568 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | |
569 | * has been created using svc_create_pooled(). | |
570 | * | |
571 | * Based on code that used to be in nfsd_svc() but tweaked | |
572 | * to be pool-aware. | |
573 | */ | |
574 | int | |
575 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
576 | { | |
577 | struct task_struct *victim; | |
578 | int error = 0; | |
579 | unsigned int state = serv->sv_nrthreads-1; | |
580 | ||
581 | if (pool == NULL) { | |
582 | /* The -1 assumes caller has done a svc_get() */ | |
583 | nrservs -= (serv->sv_nrthreads-1); | |
584 | } else { | |
585 | spin_lock_bh(&pool->sp_lock); | |
586 | nrservs -= pool->sp_nrthreads; | |
587 | spin_unlock_bh(&pool->sp_lock); | |
588 | } | |
589 | ||
590 | /* create new threads */ | |
591 | while (nrservs > 0) { | |
592 | nrservs--; | |
593 | __module_get(serv->sv_module); | |
594 | error = __svc_create_thread(serv->sv_function, serv, | |
595 | choose_pool(serv, pool, &state)); | |
596 | if (error < 0) { | |
597 | module_put(serv->sv_module); | |
598 | break; | |
599 | } | |
600 | } | |
601 | /* destroy old threads */ | |
602 | while (nrservs < 0 && | |
603 | (victim = choose_victim(serv, pool, &state)) != NULL) { | |
604 | send_sig(serv->sv_kill_signal, victim, 1); | |
605 | nrservs++; | |
606 | } | |
607 | ||
608 | return error; | |
609 | } | |
610 | ||
3262c816 GB |
611 | /* |
612 | * Called from a server thread as it's exiting. Caller must hold BKL. | |
1da177e4 LT |
613 | */ |
614 | void | |
615 | svc_exit_thread(struct svc_rqst *rqstp) | |
616 | { | |
617 | struct svc_serv *serv = rqstp->rq_server; | |
3262c816 | 618 | struct svc_pool *pool = rqstp->rq_pool; |
1da177e4 LT |
619 | |
620 | svc_release_buffer(rqstp); | |
a51482bd JJ |
621 | kfree(rqstp->rq_resp); |
622 | kfree(rqstp->rq_argp); | |
623 | kfree(rqstp->rq_auth_data); | |
3262c816 GB |
624 | |
625 | spin_lock_bh(&pool->sp_lock); | |
626 | pool->sp_nrthreads--; | |
a7455442 | 627 | list_del(&rqstp->rq_all); |
3262c816 GB |
628 | spin_unlock_bh(&pool->sp_lock); |
629 | ||
1da177e4 LT |
630 | kfree(rqstp); |
631 | ||
632 | /* Release the server */ | |
633 | if (serv) | |
634 | svc_destroy(serv); | |
635 | } | |
636 | ||
637 | /* | |
638 | * Register an RPC service with the local portmapper. | |
cca5172a | 639 | * To unregister a service, call this routine with |
1da177e4 LT |
640 | * proto and port == 0. |
641 | */ | |
642 | int | |
643 | svc_register(struct svc_serv *serv, int proto, unsigned short port) | |
644 | { | |
645 | struct svc_program *progp; | |
646 | unsigned long flags; | |
647 | int i, error = 0, dummy; | |
648 | ||
1da177e4 LT |
649 | if (!port) |
650 | clear_thread_flag(TIF_SIGPENDING); | |
651 | ||
bc5fea42 OK |
652 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
653 | for (i = 0; i < progp->pg_nvers; i++) { | |
654 | if (progp->pg_vers[i] == NULL) | |
655 | continue; | |
656 | ||
46121cf7 | 657 | dprintk("svc: svc_register(%s, %s, %d, %d)%s\n", |
bc5fea42 OK |
658 | progp->pg_name, |
659 | proto == IPPROTO_UDP? "udp" : "tcp", | |
660 | port, | |
661 | i, | |
662 | progp->pg_vers[i]->vs_hidden? | |
663 | " (but not telling portmap)" : ""); | |
664 | ||
665 | if (progp->pg_vers[i]->vs_hidden) | |
666 | continue; | |
667 | ||
668 | error = rpc_register(progp->pg_prog, i, proto, port, &dummy); | |
669 | if (error < 0) | |
670 | break; | |
671 | if (port && !dummy) { | |
672 | error = -EACCES; | |
673 | break; | |
674 | } | |
1da177e4 LT |
675 | } |
676 | } | |
677 | ||
678 | if (!port) { | |
679 | spin_lock_irqsave(¤t->sighand->siglock, flags); | |
680 | recalc_sigpending(); | |
681 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
682 | } | |
683 | ||
684 | return error; | |
685 | } | |
686 | ||
687 | /* | |
688 | * Process the RPC request. | |
689 | */ | |
690 | int | |
6fb2b47f | 691 | svc_process(struct svc_rqst *rqstp) |
1da177e4 LT |
692 | { |
693 | struct svc_program *progp; | |
694 | struct svc_version *versp = NULL; /* compiler food */ | |
695 | struct svc_procedure *procp = NULL; | |
696 | struct kvec * argv = &rqstp->rq_arg.head[0]; | |
697 | struct kvec * resv = &rqstp->rq_res.head[0]; | |
6fb2b47f | 698 | struct svc_serv *serv = rqstp->rq_server; |
1da177e4 | 699 | kxdrproc_t xdr; |
d8ed029d AD |
700 | __be32 *statp; |
701 | u32 dir, prog, vers, proc; | |
702 | __be32 auth_stat, rpc_stat; | |
1da177e4 | 703 | int auth_res; |
8f8e05c5 | 704 | __be32 *reply_statp; |
1da177e4 LT |
705 | |
706 | rpc_stat = rpc_success; | |
707 | ||
708 | if (argv->iov_len < 6*4) | |
709 | goto err_short_len; | |
710 | ||
711 | /* setup response xdr_buf. | |
cca5172a | 712 | * Initially it has just one page |
1da177e4 | 713 | */ |
44524359 | 714 | rqstp->rq_resused = 1; |
1da177e4 LT |
715 | resv->iov_base = page_address(rqstp->rq_respages[0]); |
716 | resv->iov_len = 0; | |
44524359 | 717 | rqstp->rq_res.pages = rqstp->rq_respages + 1; |
1da177e4 LT |
718 | rqstp->rq_res.len = 0; |
719 | rqstp->rq_res.page_base = 0; | |
720 | rqstp->rq_res.page_len = 0; | |
334ccfd5 | 721 | rqstp->rq_res.buflen = PAGE_SIZE; |
7c9fdcfb | 722 | rqstp->rq_res.tail[0].iov_base = NULL; |
1da177e4 | 723 | rqstp->rq_res.tail[0].iov_len = 0; |
5c04c46a BF |
724 | /* Will be turned off only in gss privacy case: */ |
725 | rqstp->rq_sendfile_ok = 1; | |
1da177e4 LT |
726 | /* tcp needs a space for the record length... */ |
727 | if (rqstp->rq_prot == IPPROTO_TCP) | |
76994313 | 728 | svc_putnl(resv, 0); |
1da177e4 LT |
729 | |
730 | rqstp->rq_xid = svc_getu32(argv); | |
731 | svc_putu32(resv, rqstp->rq_xid); | |
732 | ||
76994313 AD |
733 | dir = svc_getnl(argv); |
734 | vers = svc_getnl(argv); | |
1da177e4 LT |
735 | |
736 | /* First words of reply: */ | |
76994313 | 737 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 LT |
738 | |
739 | if (dir != 0) /* direction != CALL */ | |
740 | goto err_bad_dir; | |
741 | if (vers != 2) /* RPC version number */ | |
742 | goto err_bad_rpc; | |
743 | ||
744 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 745 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 746 | |
76994313 | 747 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 748 | |
76994313 AD |
749 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
750 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | |
751 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 LT |
752 | |
753 | progp = serv->sv_program; | |
80d188a6 N |
754 | |
755 | for (progp = serv->sv_program; progp; progp = progp->pg_next) | |
756 | if (prog == progp->pg_prog) | |
757 | break; | |
758 | ||
1da177e4 LT |
759 | /* |
760 | * Decode auth data, and add verifier to reply buffer. | |
761 | * We do this before anything else in order to get a decent | |
762 | * auth verifier. | |
763 | */ | |
764 | auth_res = svc_authenticate(rqstp, &auth_stat); | |
765 | /* Also give the program a chance to reject this call: */ | |
80d188a6 | 766 | if (auth_res == SVC_OK && progp) { |
1da177e4 LT |
767 | auth_stat = rpc_autherr_badcred; |
768 | auth_res = progp->pg_authenticate(rqstp); | |
769 | } | |
770 | switch (auth_res) { | |
771 | case SVC_OK: | |
772 | break; | |
773 | case SVC_GARBAGE: | |
774 | rpc_stat = rpc_garbage_args; | |
775 | goto err_bad; | |
776 | case SVC_SYSERR: | |
777 | rpc_stat = rpc_system_err; | |
778 | goto err_bad; | |
779 | case SVC_DENIED: | |
780 | goto err_bad_auth; | |
781 | case SVC_DROP: | |
782 | goto dropit; | |
783 | case SVC_COMPLETE: | |
784 | goto sendit; | |
785 | } | |
80d188a6 | 786 | |
9ba02638 | 787 | if (progp == NULL) |
1da177e4 LT |
788 | goto err_bad_prog; |
789 | ||
790 | if (vers >= progp->pg_nvers || | |
791 | !(versp = progp->pg_vers[vers])) | |
792 | goto err_bad_vers; | |
793 | ||
794 | procp = versp->vs_proc + proc; | |
795 | if (proc >= versp->vs_nproc || !procp->pc_func) | |
796 | goto err_bad_proc; | |
797 | rqstp->rq_server = serv; | |
798 | rqstp->rq_procinfo = procp; | |
799 | ||
800 | /* Syntactic check complete */ | |
801 | serv->sv_stats->rpccnt++; | |
802 | ||
803 | /* Build the reply header. */ | |
804 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 805 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 LT |
806 | |
807 | /* Bump per-procedure stats counter */ | |
808 | procp->pc_count++; | |
809 | ||
810 | /* Initialize storage for argp and resp */ | |
811 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | |
812 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | |
813 | ||
cca5172a | 814 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
815 | * better idea of reply size |
816 | */ | |
817 | if (procp->pc_xdrressize) | |
818 | svc_reserve(rqstp, procp->pc_xdrressize<<2); | |
819 | ||
820 | /* Call the function that processes the request. */ | |
821 | if (!versp->vs_dispatch) { | |
822 | /* Decode arguments */ | |
823 | xdr = procp->pc_decode; | |
824 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | |
825 | goto err_garbage; | |
826 | ||
827 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
828 | ||
829 | /* Encode reply */ | |
d343fce1 N |
830 | if (*statp == rpc_drop_reply) { |
831 | if (procp->pc_release) | |
832 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
833 | goto dropit; | |
834 | } | |
1da177e4 LT |
835 | if (*statp == rpc_success && (xdr = procp->pc_encode) |
836 | && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | |
837 | dprintk("svc: failed to encode reply\n"); | |
838 | /* serv->sv_stats->rpcsystemerr++; */ | |
839 | *statp = rpc_system_err; | |
840 | } | |
841 | } else { | |
842 | dprintk("svc: calling dispatcher\n"); | |
843 | if (!versp->vs_dispatch(rqstp, statp)) { | |
844 | /* Release reply info */ | |
845 | if (procp->pc_release) | |
846 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
847 | goto dropit; | |
848 | } | |
849 | } | |
850 | ||
851 | /* Check RPC status result */ | |
852 | if (*statp != rpc_success) | |
853 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
854 | ||
855 | /* Release reply info */ | |
856 | if (procp->pc_release) | |
857 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
858 | ||
859 | if (procp->pc_encode == NULL) | |
860 | goto dropit; | |
861 | ||
862 | sendit: | |
863 | if (svc_authorise(rqstp)) | |
864 | goto dropit; | |
865 | return svc_send(rqstp); | |
866 | ||
867 | dropit: | |
868 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
869 | dprintk("svc: svc_process dropit\n"); | |
870 | svc_drop(rqstp); | |
871 | return 0; | |
872 | ||
873 | err_short_len: | |
34e9a63b N |
874 | if (net_ratelimit()) |
875 | printk("svc: short len %Zd, dropping request\n", argv->iov_len); | |
876 | ||
1da177e4 LT |
877 | goto dropit; /* drop request */ |
878 | ||
879 | err_bad_dir: | |
34e9a63b N |
880 | if (net_ratelimit()) |
881 | printk("svc: bad direction %d, dropping request\n", dir); | |
882 | ||
1da177e4 LT |
883 | serv->sv_stats->rpcbadfmt++; |
884 | goto dropit; /* drop request */ | |
885 | ||
886 | err_bad_rpc: | |
887 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
888 | svc_putnl(resv, 1); /* REJECT */ |
889 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
890 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
891 | svc_putnl(resv, 2); | |
1da177e4 LT |
892 | goto sendit; |
893 | ||
894 | err_bad_auth: | |
895 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | |
896 | serv->sv_stats->rpcbadauth++; | |
897 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 898 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
899 | svc_putnl(resv, 1); /* REJECT */ |
900 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
901 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | |
1da177e4 LT |
902 | goto sendit; |
903 | ||
904 | err_bad_prog: | |
9ba02638 | 905 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 906 | serv->sv_stats->rpcbadfmt++; |
76994313 | 907 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
908 | goto sendit; |
909 | ||
910 | err_bad_vers: | |
34e9a63b N |
911 | if (net_ratelimit()) |
912 | printk("svc: unknown version (%d for prog %d, %s)\n", | |
913 | vers, prog, progp->pg_name); | |
914 | ||
1da177e4 | 915 | serv->sv_stats->rpcbadfmt++; |
76994313 AD |
916 | svc_putnl(resv, RPC_PROG_MISMATCH); |
917 | svc_putnl(resv, progp->pg_lovers); | |
918 | svc_putnl(resv, progp->pg_hivers); | |
1da177e4 LT |
919 | goto sendit; |
920 | ||
921 | err_bad_proc: | |
34e9a63b N |
922 | if (net_ratelimit()) |
923 | printk("svc: unknown procedure (%d)\n", proc); | |
924 | ||
1da177e4 | 925 | serv->sv_stats->rpcbadfmt++; |
76994313 | 926 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
927 | goto sendit; |
928 | ||
929 | err_garbage: | |
34e9a63b N |
930 | if (net_ratelimit()) |
931 | printk("svc: failed to decode args\n"); | |
932 | ||
1da177e4 LT |
933 | rpc_stat = rpc_garbage_args; |
934 | err_bad: | |
935 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 936 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
937 | goto sendit; |
938 | } | |
7adae489 GB |
939 | |
940 | /* | |
941 | * Return (transport-specific) limit on the rpc payload. | |
942 | */ | |
943 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
944 | { | |
945 | int max = RPCSVC_MAXPAYLOAD_TCP; | |
946 | ||
947 | if (rqstp->rq_sock->sk_sock->type == SOCK_DGRAM) | |
948 | max = RPCSVC_MAXPAYLOAD_UDP; | |
c6b0a9f8 N |
949 | if (rqstp->rq_server->sv_max_payload < max) |
950 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
951 | return max; |
952 | } | |
953 | EXPORT_SYMBOL_GPL(svc_max_payload); |