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> | |
9867d76c | 21 | #include <linux/kthread.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
1da177e4 LT |
23 | |
24 | #include <linux/sunrpc/types.h> | |
25 | #include <linux/sunrpc/xdr.h> | |
26 | #include <linux/sunrpc/stats.h> | |
27 | #include <linux/sunrpc/svcsock.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
4d6bbb62 | 29 | #include <linux/sunrpc/bc_xprt.h> |
1da177e4 LT |
30 | |
31 | #define RPCDBG_FACILITY RPCDBG_SVCDSP | |
1da177e4 | 32 | |
7252d575 CL |
33 | static void svc_unregister(const struct svc_serv *serv); |
34 | ||
42a7fc4a GB |
35 | #define svc_serv_is_pooled(serv) ((serv)->sv_function) |
36 | ||
bfd24160 GB |
37 | /* |
38 | * Mode for mapping cpus to pools. | |
39 | */ | |
40 | enum { | |
42a7fc4a | 41 | SVC_POOL_AUTO = -1, /* choose one of the others */ |
bfd24160 GB |
42 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool |
43 | * (legacy & UP mode) */ | |
44 | SVC_POOL_PERCPU, /* one pool per cpu */ | |
45 | SVC_POOL_PERNODE /* one pool per numa node */ | |
46 | }; | |
42a7fc4a | 47 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
bfd24160 GB |
48 | |
49 | /* | |
50 | * Structure for mapping cpus to pools and vice versa. | |
51 | * Setup once during sunrpc initialisation. | |
52 | */ | |
53 | static struct svc_pool_map { | |
42a7fc4a | 54 | int count; /* How many svc_servs use us */ |
bfd24160 GB |
55 | int mode; /* Note: int not enum to avoid |
56 | * warnings about "enumeration value | |
57 | * not handled in switch" */ | |
58 | unsigned int npools; | |
59 | unsigned int *pool_to; /* maps pool id to cpu or node */ | |
60 | unsigned int *to_pool; /* maps cpu or node to pool id */ | |
61 | } svc_pool_map = { | |
42a7fc4a GB |
62 | .count = 0, |
63 | .mode = SVC_POOL_DEFAULT | |
bfd24160 | 64 | }; |
42a7fc4a GB |
65 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
66 | ||
67 | static int | |
68 | param_set_pool_mode(const char *val, struct kernel_param *kp) | |
69 | { | |
70 | int *ip = (int *)kp->arg; | |
71 | struct svc_pool_map *m = &svc_pool_map; | |
72 | int err; | |
73 | ||
74 | mutex_lock(&svc_pool_map_mutex); | |
75 | ||
76 | err = -EBUSY; | |
77 | if (m->count) | |
78 | goto out; | |
79 | ||
80 | err = 0; | |
81 | if (!strncmp(val, "auto", 4)) | |
82 | *ip = SVC_POOL_AUTO; | |
83 | else if (!strncmp(val, "global", 6)) | |
84 | *ip = SVC_POOL_GLOBAL; | |
85 | else if (!strncmp(val, "percpu", 6)) | |
86 | *ip = SVC_POOL_PERCPU; | |
87 | else if (!strncmp(val, "pernode", 7)) | |
88 | *ip = SVC_POOL_PERNODE; | |
89 | else | |
90 | err = -EINVAL; | |
91 | ||
92 | out: | |
93 | mutex_unlock(&svc_pool_map_mutex); | |
94 | return err; | |
95 | } | |
96 | ||
97 | static int | |
98 | param_get_pool_mode(char *buf, struct kernel_param *kp) | |
99 | { | |
100 | int *ip = (int *)kp->arg; | |
101 | ||
102 | switch (*ip) | |
103 | { | |
104 | case SVC_POOL_AUTO: | |
105 | return strlcpy(buf, "auto", 20); | |
106 | case SVC_POOL_GLOBAL: | |
107 | return strlcpy(buf, "global", 20); | |
108 | case SVC_POOL_PERCPU: | |
109 | return strlcpy(buf, "percpu", 20); | |
110 | case SVC_POOL_PERNODE: | |
111 | return strlcpy(buf, "pernode", 20); | |
112 | default: | |
113 | return sprintf(buf, "%d", *ip); | |
114 | } | |
115 | } | |
bfd24160 | 116 | |
42a7fc4a GB |
117 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
118 | &svc_pool_map.mode, 0644); | |
bfd24160 GB |
119 | |
120 | /* | |
121 | * Detect best pool mapping mode heuristically, | |
122 | * according to the machine's topology. | |
123 | */ | |
124 | static int | |
125 | svc_pool_map_choose_mode(void) | |
126 | { | |
127 | unsigned int node; | |
128 | ||
62bc62a8 | 129 | if (nr_online_nodes > 1) { |
bfd24160 GB |
130 | /* |
131 | * Actually have multiple NUMA nodes, | |
132 | * so split pools on NUMA node boundaries | |
133 | */ | |
134 | return SVC_POOL_PERNODE; | |
135 | } | |
136 | ||
72c33688 | 137 | node = first_online_node; |
bfd24160 GB |
138 | if (nr_cpus_node(node) > 2) { |
139 | /* | |
140 | * Non-trivial SMP, or CONFIG_NUMA on | |
141 | * non-NUMA hardware, e.g. with a generic | |
142 | * x86_64 kernel on Xeons. In this case we | |
143 | * want to divide the pools on cpu boundaries. | |
144 | */ | |
145 | return SVC_POOL_PERCPU; | |
146 | } | |
147 | ||
148 | /* default: one global pool */ | |
149 | return SVC_POOL_GLOBAL; | |
150 | } | |
151 | ||
152 | /* | |
153 | * Allocate the to_pool[] and pool_to[] arrays. | |
154 | * Returns 0 on success or an errno. | |
155 | */ | |
156 | static int | |
157 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
158 | { | |
159 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
160 | if (!m->to_pool) | |
161 | goto fail; | |
162 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
163 | if (!m->pool_to) | |
164 | goto fail_free; | |
165 | ||
166 | return 0; | |
167 | ||
168 | fail_free: | |
169 | kfree(m->to_pool); | |
170 | fail: | |
171 | return -ENOMEM; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
176 | * Returns number of pools or <0 on error. | |
177 | */ | |
178 | static int | |
179 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
180 | { | |
53b8a315 | 181 | unsigned int maxpools = nr_cpu_ids; |
bfd24160 GB |
182 | unsigned int pidx = 0; |
183 | unsigned int cpu; | |
184 | int err; | |
185 | ||
186 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
187 | if (err) | |
188 | return err; | |
189 | ||
190 | for_each_online_cpu(cpu) { | |
191 | BUG_ON(pidx > maxpools); | |
192 | m->to_pool[cpu] = pidx; | |
193 | m->pool_to[pidx] = cpu; | |
194 | pidx++; | |
195 | } | |
196 | /* cpus brought online later all get mapped to pool0, sorry */ | |
197 | ||
198 | return pidx; | |
199 | }; | |
200 | ||
201 | ||
202 | /* | |
203 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
204 | * Returns number of pools or <0 on error. | |
205 | */ | |
206 | static int | |
207 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
208 | { | |
74c7aa8b | 209 | unsigned int maxpools = nr_node_ids; |
bfd24160 GB |
210 | unsigned int pidx = 0; |
211 | unsigned int node; | |
212 | int err; | |
213 | ||
214 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
215 | if (err) | |
216 | return err; | |
217 | ||
218 | for_each_node_with_cpus(node) { | |
219 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
220 | BUG_ON(pidx > maxpools); | |
221 | m->to_pool[node] = pidx; | |
222 | m->pool_to[pidx] = node; | |
223 | pidx++; | |
224 | } | |
225 | /* nodes brought online later all get mapped to pool0, sorry */ | |
226 | ||
227 | return pidx; | |
228 | } | |
229 | ||
230 | ||
231 | /* | |
42a7fc4a GB |
232 | * Add a reference to the global map of cpus to pools (and |
233 | * vice versa). Initialise the map if we're the first user. | |
234 | * Returns the number of pools. | |
bfd24160 GB |
235 | */ |
236 | static unsigned int | |
42a7fc4a | 237 | svc_pool_map_get(void) |
bfd24160 GB |
238 | { |
239 | struct svc_pool_map *m = &svc_pool_map; | |
240 | int npools = -1; | |
241 | ||
42a7fc4a GB |
242 | mutex_lock(&svc_pool_map_mutex); |
243 | ||
244 | if (m->count++) { | |
245 | mutex_unlock(&svc_pool_map_mutex); | |
bfd24160 | 246 | return m->npools; |
42a7fc4a | 247 | } |
bfd24160 | 248 | |
42a7fc4a GB |
249 | if (m->mode == SVC_POOL_AUTO) |
250 | m->mode = svc_pool_map_choose_mode(); | |
bfd24160 GB |
251 | |
252 | switch (m->mode) { | |
253 | case SVC_POOL_PERCPU: | |
254 | npools = svc_pool_map_init_percpu(m); | |
255 | break; | |
256 | case SVC_POOL_PERNODE: | |
257 | npools = svc_pool_map_init_pernode(m); | |
258 | break; | |
259 | } | |
260 | ||
261 | if (npools < 0) { | |
262 | /* default, or memory allocation failure */ | |
263 | npools = 1; | |
264 | m->mode = SVC_POOL_GLOBAL; | |
265 | } | |
266 | m->npools = npools; | |
267 | ||
42a7fc4a | 268 | mutex_unlock(&svc_pool_map_mutex); |
bfd24160 GB |
269 | return m->npools; |
270 | } | |
271 | ||
42a7fc4a GB |
272 | |
273 | /* | |
274 | * Drop a reference to the global map of cpus to pools. | |
275 | * When the last reference is dropped, the map data is | |
276 | * freed; this allows the sysadmin to change the pool | |
277 | * mode using the pool_mode module option without | |
278 | * rebooting or re-loading sunrpc.ko. | |
279 | */ | |
280 | static void | |
281 | svc_pool_map_put(void) | |
282 | { | |
283 | struct svc_pool_map *m = &svc_pool_map; | |
284 | ||
285 | mutex_lock(&svc_pool_map_mutex); | |
286 | ||
287 | if (!--m->count) { | |
288 | m->mode = SVC_POOL_DEFAULT; | |
289 | kfree(m->to_pool); | |
290 | kfree(m->pool_to); | |
291 | m->npools = 0; | |
292 | } | |
293 | ||
294 | mutex_unlock(&svc_pool_map_mutex); | |
295 | } | |
296 | ||
297 | ||
11fd165c ED |
298 | static int svc_pool_map_get_node(unsigned int pidx) |
299 | { | |
300 | const struct svc_pool_map *m = &svc_pool_map; | |
301 | ||
302 | if (m->count) { | |
303 | if (m->mode == SVC_POOL_PERCPU) | |
304 | return cpu_to_node(m->pool_to[pidx]); | |
305 | if (m->mode == SVC_POOL_PERNODE) | |
306 | return m->pool_to[pidx]; | |
307 | } | |
308 | return NUMA_NO_NODE; | |
309 | } | |
bfd24160 | 310 | /* |
9867d76c | 311 | * Set the given thread's cpus_allowed mask so that it |
bfd24160 | 312 | * will only run on cpus in the given pool. |
bfd24160 | 313 | */ |
9867d76c JL |
314 | static inline void |
315 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) | |
bfd24160 GB |
316 | { |
317 | struct svc_pool_map *m = &svc_pool_map; | |
9867d76c | 318 | unsigned int node = m->pool_to[pidx]; |
bfd24160 GB |
319 | |
320 | /* | |
321 | * The caller checks for sv_nrpools > 1, which | |
42a7fc4a | 322 | * implies that we've been initialized. |
bfd24160 | 323 | */ |
42a7fc4a | 324 | BUG_ON(m->count == 0); |
bfd24160 | 325 | |
9867d76c | 326 | switch (m->mode) { |
bfd24160 | 327 | case SVC_POOL_PERCPU: |
c5f59f08 | 328 | { |
aa85ea5b | 329 | set_cpus_allowed_ptr(task, cpumask_of(node)); |
9867d76c | 330 | break; |
c5f59f08 | 331 | } |
bfd24160 | 332 | case SVC_POOL_PERNODE: |
c5f59f08 | 333 | { |
a70f7302 | 334 | set_cpus_allowed_ptr(task, cpumask_of_node(node)); |
9867d76c | 335 | break; |
bfd24160 | 336 | } |
c5f59f08 | 337 | } |
bfd24160 GB |
338 | } |
339 | ||
340 | /* | |
341 | * Use the mapping mode to choose a pool for a given CPU. | |
342 | * Used when enqueueing an incoming RPC. Always returns | |
343 | * a non-NULL pool pointer. | |
344 | */ | |
345 | struct svc_pool * | |
346 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | |
347 | { | |
348 | struct svc_pool_map *m = &svc_pool_map; | |
349 | unsigned int pidx = 0; | |
350 | ||
351 | /* | |
42a7fc4a | 352 | * An uninitialised map happens in a pure client when |
bfd24160 GB |
353 | * lockd is brought up, so silently treat it the |
354 | * same as SVC_POOL_GLOBAL. | |
355 | */ | |
42a7fc4a GB |
356 | if (svc_serv_is_pooled(serv)) { |
357 | switch (m->mode) { | |
358 | case SVC_POOL_PERCPU: | |
359 | pidx = m->to_pool[cpu]; | |
360 | break; | |
361 | case SVC_POOL_PERNODE: | |
362 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
363 | break; | |
364 | } | |
bfd24160 GB |
365 | } |
366 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | |
367 | } | |
368 | ||
d9908560 SK |
369 | static int svc_rpcb_setup(struct svc_serv *serv) |
370 | { | |
371 | int err; | |
372 | ||
373 | err = rpcb_create_local(); | |
374 | if (err) | |
375 | return err; | |
376 | ||
377 | /* Remove any stale portmap registrations */ | |
378 | svc_unregister(serv); | |
379 | return 0; | |
380 | } | |
381 | ||
16d05870 | 382 | void svc_rpcb_cleanup(struct svc_serv *serv) |
d9908560 SK |
383 | { |
384 | svc_unregister(serv); | |
385 | rpcb_put_local(); | |
386 | } | |
16d05870 | 387 | EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); |
d9908560 SK |
388 | |
389 | static int svc_uses_rpcbind(struct svc_serv *serv) | |
390 | { | |
391 | struct svc_program *progp; | |
392 | unsigned int i; | |
393 | ||
394 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
395 | for (i = 0; i < progp->pg_nvers; i++) { | |
396 | if (progp->pg_vers[i] == NULL) | |
397 | continue; | |
398 | if (progp->pg_vers[i]->vs_hidden == 0) | |
399 | return 1; | |
400 | } | |
401 | } | |
402 | ||
403 | return 0; | |
404 | } | |
bfd24160 | 405 | |
1da177e4 LT |
406 | /* |
407 | * Create an RPC service | |
408 | */ | |
a7455442 GB |
409 | static struct svc_serv * |
410 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
49a9072f | 411 | void (*shutdown)(struct svc_serv *serv)) |
1da177e4 LT |
412 | { |
413 | struct svc_serv *serv; | |
ea339d46 | 414 | unsigned int vers; |
1da177e4 | 415 | unsigned int xdrsize; |
3262c816 | 416 | unsigned int i; |
1da177e4 | 417 | |
0da974f4 | 418 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 419 | return NULL; |
9ba02638 | 420 | serv->sv_name = prog->pg_name; |
1da177e4 LT |
421 | serv->sv_program = prog; |
422 | serv->sv_nrthreads = 1; | |
423 | serv->sv_stats = prog->pg_stats; | |
c6b0a9f8 N |
424 | if (bufsize > RPCSVC_MAXPAYLOAD) |
425 | bufsize = RPCSVC_MAXPAYLOAD; | |
426 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
427 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
bc591ccf | 428 | serv->sv_shutdown = shutdown; |
1da177e4 | 429 | xdrsize = 0; |
9ba02638 AG |
430 | while (prog) { |
431 | prog->pg_lovers = prog->pg_nvers-1; | |
432 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
433 | if (prog->pg_vers[vers]) { | |
434 | prog->pg_hivers = vers; | |
435 | if (prog->pg_lovers > vers) | |
436 | prog->pg_lovers = vers; | |
437 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
438 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
439 | } | |
440 | prog = prog->pg_next; | |
441 | } | |
1da177e4 | 442 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
443 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
444 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
36bdfc8b | 445 | init_timer(&serv->sv_temptimer); |
1da177e4 LT |
446 | spin_lock_init(&serv->sv_lock); |
447 | ||
a7455442 | 448 | serv->sv_nrpools = npools; |
3262c816 | 449 | serv->sv_pools = |
cd861280 | 450 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
451 | GFP_KERNEL); |
452 | if (!serv->sv_pools) { | |
453 | kfree(serv); | |
454 | return NULL; | |
455 | } | |
456 | ||
457 | for (i = 0; i < serv->sv_nrpools; i++) { | |
458 | struct svc_pool *pool = &serv->sv_pools[i]; | |
459 | ||
46121cf7 | 460 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
461 | i, serv->sv_name); |
462 | ||
463 | pool->sp_id = i; | |
464 | INIT_LIST_HEAD(&pool->sp_threads); | |
465 | INIT_LIST_HEAD(&pool->sp_sockets); | |
a7455442 | 466 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
467 | spin_lock_init(&pool->sp_lock); |
468 | } | |
469 | ||
e40f5e29 SK |
470 | if (svc_uses_rpcbind(serv)) { |
471 | if (svc_rpcb_setup(serv) < 0) { | |
472 | kfree(serv->sv_pools); | |
473 | kfree(serv); | |
474 | return NULL; | |
475 | } | |
476 | if (!serv->sv_shutdown) | |
477 | serv->sv_shutdown = svc_rpcb_cleanup; | |
478 | } | |
1da177e4 LT |
479 | |
480 | return serv; | |
481 | } | |
482 | ||
a7455442 GB |
483 | struct svc_serv * |
484 | svc_create(struct svc_program *prog, unsigned int bufsize, | |
49a9072f | 485 | void (*shutdown)(struct svc_serv *serv)) |
a7455442 | 486 | { |
49a9072f | 487 | return __svc_create(prog, bufsize, /*npools*/1, shutdown); |
a7455442 | 488 | } |
24c3767e | 489 | EXPORT_SYMBOL_GPL(svc_create); |
a7455442 GB |
490 | |
491 | struct svc_serv * | |
492 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | |
49a9072f | 493 | void (*shutdown)(struct svc_serv *serv), |
a75c5d01 | 494 | svc_thread_fn func, struct module *mod) |
a7455442 GB |
495 | { |
496 | struct svc_serv *serv; | |
42a7fc4a | 497 | unsigned int npools = svc_pool_map_get(); |
a7455442 | 498 | |
49a9072f | 499 | serv = __svc_create(prog, bufsize, npools, shutdown); |
a7455442 GB |
500 | |
501 | if (serv != NULL) { | |
502 | serv->sv_function = func; | |
a7455442 GB |
503 | serv->sv_module = mod; |
504 | } | |
505 | ||
506 | return serv; | |
507 | } | |
24c3767e | 508 | EXPORT_SYMBOL_GPL(svc_create_pooled); |
a7455442 | 509 | |
1da177e4 | 510 | /* |
bedbdd8b NB |
511 | * Destroy an RPC service. Should be called with appropriate locking to |
512 | * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. | |
1da177e4 LT |
513 | */ |
514 | void | |
515 | svc_destroy(struct svc_serv *serv) | |
516 | { | |
46121cf7 | 517 | dprintk("svc: svc_destroy(%s, %d)\n", |
1da177e4 LT |
518 | serv->sv_program->pg_name, |
519 | serv->sv_nrthreads); | |
520 | ||
521 | if (serv->sv_nrthreads) { | |
522 | if (--(serv->sv_nrthreads) != 0) { | |
523 | svc_sock_update_bufs(serv); | |
524 | return; | |
525 | } | |
526 | } else | |
527 | printk("svc_destroy: no threads for serv=%p!\n", serv); | |
528 | ||
36bdfc8b GB |
529 | del_timer_sync(&serv->sv_temptimer); |
530 | ||
7a182083 | 531 | svc_close_all(&serv->sv_tempsocks); |
cda1fd4a | 532 | |
bc591ccf N |
533 | if (serv->sv_shutdown) |
534 | serv->sv_shutdown(serv); | |
535 | ||
7a182083 | 536 | svc_close_all(&serv->sv_permsocks); |
cda1fd4a N |
537 | |
538 | BUG_ON(!list_empty(&serv->sv_permsocks)); | |
539 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | |
cca5172a | 540 | |
1da177e4 LT |
541 | cache_clean_deferred(serv); |
542 | ||
42a7fc4a GB |
543 | if (svc_serv_is_pooled(serv)) |
544 | svc_pool_map_put(); | |
545 | ||
3262c816 | 546 | kfree(serv->sv_pools); |
1da177e4 LT |
547 | kfree(serv); |
548 | } | |
24c3767e | 549 | EXPORT_SYMBOL_GPL(svc_destroy); |
1da177e4 LT |
550 | |
551 | /* | |
552 | * Allocate an RPC server's buffer space. | |
553 | * We allocate pages and place them in rq_argpages. | |
554 | */ | |
555 | static int | |
11fd165c | 556 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) |
1da177e4 | 557 | { |
0dc220f0 | 558 | unsigned int pages, arghi; |
cca5172a | 559 | |
ba17686f AA |
560 | /* bc_xprt uses fore channel allocated buffers */ |
561 | if (svc_is_backchannel(rqstp)) | |
562 | return 1; | |
563 | ||
c6b0a9f8 N |
564 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
565 | * We assume one is at most one page | |
566 | */ | |
1da177e4 | 567 | arghi = 0; |
09a62660 | 568 | BUG_ON(pages > RPCSVC_MAXPAGES); |
1da177e4 | 569 | while (pages) { |
11fd165c | 570 | struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); |
1da177e4 LT |
571 | if (!p) |
572 | break; | |
44524359 | 573 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
574 | pages--; |
575 | } | |
0dc220f0 | 576 | return pages == 0; |
1da177e4 LT |
577 | } |
578 | ||
579 | /* | |
580 | * Release an RPC server buffer | |
581 | */ | |
582 | static void | |
583 | svc_release_buffer(struct svc_rqst *rqstp) | |
584 | { | |
50c8bb13 CL |
585 | unsigned int i; |
586 | ||
587 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) | |
44524359 N |
588 | if (rqstp->rq_pages[i]) |
589 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
590 | } |
591 | ||
0113ab34 | 592 | struct svc_rqst * |
11fd165c | 593 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) |
1da177e4 LT |
594 | { |
595 | struct svc_rqst *rqstp; | |
1da177e4 | 596 | |
11fd165c | 597 | rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); |
1da177e4 | 598 | if (!rqstp) |
0113ab34 | 599 | goto out_enomem; |
1da177e4 | 600 | |
1da177e4 LT |
601 | init_waitqueue_head(&rqstp->rq_wait); |
602 | ||
1da177e4 | 603 | serv->sv_nrthreads++; |
3262c816 GB |
604 | spin_lock_bh(&pool->sp_lock); |
605 | pool->sp_nrthreads++; | |
a7455442 | 606 | list_add(&rqstp->rq_all, &pool->sp_all_threads); |
3262c816 | 607 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 608 | rqstp->rq_server = serv; |
3262c816 | 609 | rqstp->rq_pool = pool; |
bfd24160 | 610 | |
11fd165c | 611 | rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 JL |
612 | if (!rqstp->rq_argp) |
613 | goto out_thread; | |
614 | ||
11fd165c | 615 | rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 JL |
616 | if (!rqstp->rq_resp) |
617 | goto out_thread; | |
618 | ||
11fd165c | 619 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) |
0113ab34 JL |
620 | goto out_thread; |
621 | ||
622 | return rqstp; | |
623 | out_thread: | |
624 | svc_exit_thread(rqstp); | |
625 | out_enomem: | |
626 | return ERR_PTR(-ENOMEM); | |
627 | } | |
24c3767e | 628 | EXPORT_SYMBOL_GPL(svc_prepare_thread); |
0113ab34 | 629 | |
a7455442 GB |
630 | /* |
631 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
632 | */ | |
633 | static inline struct svc_pool * | |
634 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
635 | { | |
636 | if (pool != NULL) | |
637 | return pool; | |
638 | ||
cca5172a | 639 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
640 | } |
641 | ||
642 | /* | |
643 | * Choose a thread to kill, for svc_set_num_threads | |
644 | */ | |
645 | static inline struct task_struct * | |
646 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
647 | { | |
648 | unsigned int i; | |
649 | struct task_struct *task = NULL; | |
650 | ||
651 | if (pool != NULL) { | |
652 | spin_lock_bh(&pool->sp_lock); | |
653 | } else { | |
654 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
655 | for (i = 0; i < serv->sv_nrpools; i++) { |
656 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 657 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
658 | if (!list_empty(&pool->sp_all_threads)) |
659 | goto found_pool; | |
a7455442 | 660 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 661 | } |
a7455442 GB |
662 | return NULL; |
663 | } | |
664 | ||
665 | found_pool: | |
666 | if (!list_empty(&pool->sp_all_threads)) { | |
667 | struct svc_rqst *rqstp; | |
668 | ||
669 | /* | |
670 | * Remove from the pool->sp_all_threads list | |
671 | * so we don't try to kill it again. | |
672 | */ | |
673 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
674 | list_del_init(&rqstp->rq_all); | |
675 | task = rqstp->rq_task; | |
cca5172a | 676 | } |
a7455442 GB |
677 | spin_unlock_bh(&pool->sp_lock); |
678 | ||
679 | return task; | |
680 | } | |
681 | ||
682 | /* | |
683 | * Create or destroy enough new threads to make the number | |
684 | * of threads the given number. If `pool' is non-NULL, applies | |
685 | * only to threads in that pool, otherwise round-robins between | |
686 | * all pools. Must be called with a svc_get() reference and | |
bedbdd8b | 687 | * the BKL or another lock to protect access to svc_serv fields. |
a7455442 GB |
688 | * |
689 | * Destroying threads relies on the service threads filling in | |
690 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | |
691 | * has been created using svc_create_pooled(). | |
692 | * | |
693 | * Based on code that used to be in nfsd_svc() but tweaked | |
694 | * to be pool-aware. | |
695 | */ | |
696 | int | |
697 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
698 | { | |
9867d76c JL |
699 | struct svc_rqst *rqstp; |
700 | struct task_struct *task; | |
701 | struct svc_pool *chosen_pool; | |
a7455442 GB |
702 | int error = 0; |
703 | unsigned int state = serv->sv_nrthreads-1; | |
11fd165c | 704 | int node; |
a7455442 GB |
705 | |
706 | if (pool == NULL) { | |
707 | /* The -1 assumes caller has done a svc_get() */ | |
708 | nrservs -= (serv->sv_nrthreads-1); | |
709 | } else { | |
710 | spin_lock_bh(&pool->sp_lock); | |
711 | nrservs -= pool->sp_nrthreads; | |
712 | spin_unlock_bh(&pool->sp_lock); | |
713 | } | |
714 | ||
715 | /* create new threads */ | |
716 | while (nrservs > 0) { | |
717 | nrservs--; | |
9867d76c JL |
718 | chosen_pool = choose_pool(serv, pool, &state); |
719 | ||
11fd165c ED |
720 | node = svc_pool_map_get_node(chosen_pool->sp_id); |
721 | rqstp = svc_prepare_thread(serv, chosen_pool, node); | |
9867d76c JL |
722 | if (IS_ERR(rqstp)) { |
723 | error = PTR_ERR(rqstp); | |
724 | break; | |
725 | } | |
726 | ||
a7455442 | 727 | __module_get(serv->sv_module); |
11fd165c ED |
728 | task = kthread_create_on_node(serv->sv_function, rqstp, |
729 | node, serv->sv_name); | |
9867d76c JL |
730 | if (IS_ERR(task)) { |
731 | error = PTR_ERR(task); | |
a7455442 | 732 | module_put(serv->sv_module); |
9867d76c | 733 | svc_exit_thread(rqstp); |
a7455442 GB |
734 | break; |
735 | } | |
9867d76c JL |
736 | |
737 | rqstp->rq_task = task; | |
738 | if (serv->sv_nrpools > 1) | |
739 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); | |
740 | ||
741 | svc_sock_update_bufs(serv); | |
742 | wake_up_process(task); | |
a7455442 GB |
743 | } |
744 | /* destroy old threads */ | |
745 | while (nrservs < 0 && | |
9867d76c | 746 | (task = choose_victim(serv, pool, &state)) != NULL) { |
a75c5d01 | 747 | send_sig(SIGINT, task, 1); |
a7455442 GB |
748 | nrservs++; |
749 | } | |
750 | ||
751 | return error; | |
752 | } | |
24c3767e | 753 | EXPORT_SYMBOL_GPL(svc_set_num_threads); |
a7455442 | 754 | |
3262c816 | 755 | /* |
bedbdd8b NB |
756 | * Called from a server thread as it's exiting. Caller must hold the BKL or |
757 | * the "service mutex", whichever is appropriate for the service. | |
1da177e4 LT |
758 | */ |
759 | void | |
760 | svc_exit_thread(struct svc_rqst *rqstp) | |
761 | { | |
762 | struct svc_serv *serv = rqstp->rq_server; | |
3262c816 | 763 | struct svc_pool *pool = rqstp->rq_pool; |
1da177e4 LT |
764 | |
765 | svc_release_buffer(rqstp); | |
a51482bd JJ |
766 | kfree(rqstp->rq_resp); |
767 | kfree(rqstp->rq_argp); | |
768 | kfree(rqstp->rq_auth_data); | |
3262c816 GB |
769 | |
770 | spin_lock_bh(&pool->sp_lock); | |
771 | pool->sp_nrthreads--; | |
a7455442 | 772 | list_del(&rqstp->rq_all); |
3262c816 GB |
773 | spin_unlock_bh(&pool->sp_lock); |
774 | ||
1da177e4 LT |
775 | kfree(rqstp); |
776 | ||
777 | /* Release the server */ | |
778 | if (serv) | |
779 | svc_destroy(serv); | |
780 | } | |
24c3767e | 781 | EXPORT_SYMBOL_GPL(svc_exit_thread); |
1da177e4 LT |
782 | |
783 | /* | |
2c7eb0b2 CL |
784 | * Register an "inet" protocol family netid with the local |
785 | * rpcbind daemon via an rpcbind v4 SET request. | |
a26cfad6 | 786 | * |
2c7eb0b2 CL |
787 | * No netconfig infrastructure is available in the kernel, so |
788 | * we map IP_ protocol numbers to netids by hand. | |
a26cfad6 | 789 | * |
2c7eb0b2 CL |
790 | * Returns zero on success; a negative errno value is returned |
791 | * if any error occurs. | |
1da177e4 | 792 | */ |
2c7eb0b2 CL |
793 | static int __svc_rpcb_register4(const u32 program, const u32 version, |
794 | const unsigned short protocol, | |
795 | const unsigned short port) | |
a26cfad6 | 796 | { |
cadc0fa5 | 797 | const struct sockaddr_in sin = { |
a26cfad6 CL |
798 | .sin_family = AF_INET, |
799 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
800 | .sin_port = htons(port), | |
801 | }; | |
cadc0fa5 CL |
802 | const char *netid; |
803 | int error; | |
2c7eb0b2 CL |
804 | |
805 | switch (protocol) { | |
806 | case IPPROTO_UDP: | |
807 | netid = RPCBIND_NETID_UDP; | |
808 | break; | |
809 | case IPPROTO_TCP: | |
810 | netid = RPCBIND_NETID_TCP; | |
811 | break; | |
812 | default: | |
ba5c35e0 | 813 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
814 | } |
815 | ||
cadc0fa5 CL |
816 | error = rpcb_v4_register(program, version, |
817 | (const struct sockaddr *)&sin, netid); | |
818 | ||
819 | /* | |
820 | * User space didn't support rpcbind v4, so retry this | |
821 | * registration request with the legacy rpcbind v2 protocol. | |
822 | */ | |
823 | if (error == -EPROTONOSUPPORT) | |
824 | error = rpcb_register(program, version, protocol, port); | |
825 | ||
826 | return error; | |
2c7eb0b2 CL |
827 | } |
828 | ||
dfd56b8b | 829 | #if IS_ENABLED(CONFIG_IPV6) |
2c7eb0b2 CL |
830 | /* |
831 | * Register an "inet6" protocol family netid with the local | |
832 | * rpcbind daemon via an rpcbind v4 SET request. | |
833 | * | |
834 | * No netconfig infrastructure is available in the kernel, so | |
835 | * we map IP_ protocol numbers to netids by hand. | |
836 | * | |
837 | * Returns zero on success; a negative errno value is returned | |
838 | * if any error occurs. | |
839 | */ | |
840 | static int __svc_rpcb_register6(const u32 program, const u32 version, | |
841 | const unsigned short protocol, | |
842 | const unsigned short port) | |
843 | { | |
cadc0fa5 | 844 | const struct sockaddr_in6 sin6 = { |
a26cfad6 CL |
845 | .sin6_family = AF_INET6, |
846 | .sin6_addr = IN6ADDR_ANY_INIT, | |
847 | .sin6_port = htons(port), | |
848 | }; | |
cadc0fa5 CL |
849 | const char *netid; |
850 | int error; | |
a26cfad6 | 851 | |
2c7eb0b2 CL |
852 | switch (protocol) { |
853 | case IPPROTO_UDP: | |
854 | netid = RPCBIND_NETID_UDP6; | |
a26cfad6 | 855 | break; |
2c7eb0b2 | 856 | case IPPROTO_TCP: |
a26cfad6 | 857 | netid = RPCBIND_NETID_TCP6; |
a26cfad6 CL |
858 | break; |
859 | default: | |
ba5c35e0 | 860 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
861 | } |
862 | ||
cadc0fa5 CL |
863 | error = rpcb_v4_register(program, version, |
864 | (const struct sockaddr *)&sin6, netid); | |
865 | ||
866 | /* | |
867 | * User space didn't support rpcbind version 4, so we won't | |
868 | * use a PF_INET6 listener. | |
869 | */ | |
870 | if (error == -EPROTONOSUPPORT) | |
871 | error = -EAFNOSUPPORT; | |
872 | ||
873 | return error; | |
2c7eb0b2 | 874 | } |
dfd56b8b | 875 | #endif /* IS_ENABLED(CONFIG_IPV6) */ |
2c7eb0b2 CL |
876 | |
877 | /* | |
878 | * Register a kernel RPC service via rpcbind version 4. | |
879 | * | |
880 | * Returns zero on success; a negative errno value is returned | |
881 | * if any error occurs. | |
882 | */ | |
363f724c CL |
883 | static int __svc_register(const char *progname, |
884 | const u32 program, const u32 version, | |
4b62e58c | 885 | const int family, |
2c7eb0b2 CL |
886 | const unsigned short protocol, |
887 | const unsigned short port) | |
888 | { | |
363f724c | 889 | int error = -EAFNOSUPPORT; |
2c7eb0b2 CL |
890 | |
891 | switch (family) { | |
4b62e58c | 892 | case PF_INET: |
363f724c | 893 | error = __svc_rpcb_register4(program, version, |
2c7eb0b2 | 894 | protocol, port); |
cadc0fa5 | 895 | break; |
dfd56b8b | 896 | #if IS_ENABLED(CONFIG_IPV6) |
4b62e58c | 897 | case PF_INET6: |
2c7eb0b2 CL |
898 | error = __svc_rpcb_register6(program, version, |
899 | protocol, port); | |
dfd56b8b | 900 | #endif |
a26cfad6 CL |
901 | } |
902 | ||
363f724c CL |
903 | if (error < 0) |
904 | printk(KERN_WARNING "svc: failed to register %sv%u RPC " | |
905 | "service (errno %d).\n", progname, version, -error); | |
906 | return error; | |
a26cfad6 | 907 | } |
2c7eb0b2 | 908 | |
a26cfad6 CL |
909 | /** |
910 | * svc_register - register an RPC service with the local portmapper | |
911 | * @serv: svc_serv struct for the service to register | |
4b62e58c | 912 | * @family: protocol family of service's listener socket |
a26cfad6 CL |
913 | * @proto: transport protocol number to advertise |
914 | * @port: port to advertise | |
915 | * | |
4b62e58c | 916 | * Service is registered for any address in the passed-in protocol family |
a26cfad6 | 917 | */ |
4b62e58c CL |
918 | int svc_register(const struct svc_serv *serv, const int family, |
919 | const unsigned short proto, const unsigned short port) | |
1da177e4 LT |
920 | { |
921 | struct svc_program *progp; | |
ea339d46 | 922 | unsigned int i; |
14aeb211 | 923 | int error = 0; |
1da177e4 | 924 | |
7252d575 | 925 | BUG_ON(proto == 0 && port == 0); |
1da177e4 | 926 | |
bc5fea42 OK |
927 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
928 | for (i = 0; i < progp->pg_nvers; i++) { | |
929 | if (progp->pg_vers[i] == NULL) | |
930 | continue; | |
931 | ||
2c7eb0b2 | 932 | dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", |
bc5fea42 | 933 | progp->pg_name, |
2c7eb0b2 | 934 | i, |
bc5fea42 OK |
935 | proto == IPPROTO_UDP? "udp" : "tcp", |
936 | port, | |
4b62e58c | 937 | family, |
bc5fea42 OK |
938 | progp->pg_vers[i]->vs_hidden? |
939 | " (but not telling portmap)" : ""); | |
940 | ||
941 | if (progp->pg_vers[i]->vs_hidden) | |
942 | continue; | |
943 | ||
363f724c CL |
944 | error = __svc_register(progp->pg_name, progp->pg_prog, |
945 | i, family, proto, port); | |
bc5fea42 OK |
946 | if (error < 0) |
947 | break; | |
1da177e4 LT |
948 | } |
949 | } | |
950 | ||
7252d575 CL |
951 | return error; |
952 | } | |
953 | ||
d5a8620f CL |
954 | /* |
955 | * If user space is running rpcbind, it should take the v4 UNSET | |
956 | * and clear everything for this [program, version]. If user space | |
957 | * is running portmap, it will reject the v4 UNSET, but won't have | |
958 | * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient | |
959 | * in this case to clear all existing entries for [program, version]. | |
960 | */ | |
f6fb3f6f CL |
961 | static void __svc_unregister(const u32 program, const u32 version, |
962 | const char *progname) | |
963 | { | |
f6fb3f6f CL |
964 | int error; |
965 | ||
d5a8620f | 966 | error = rpcb_v4_register(program, version, NULL, ""); |
f6fb3f6f | 967 | |
d5a8620f CL |
968 | /* |
969 | * User space didn't support rpcbind v4, so retry this | |
970 | * request with the legacy rpcbind v2 protocol. | |
971 | */ | |
972 | if (error == -EPROTONOSUPPORT) | |
973 | error = rpcb_register(program, version, 0, 0); | |
f6fb3f6f | 974 | |
f6fb3f6f CL |
975 | dprintk("svc: %s(%sv%u), error %d\n", |
976 | __func__, progname, version, error); | |
977 | } | |
978 | ||
7252d575 | 979 | /* |
f6fb3f6f CL |
980 | * All netids, bind addresses and ports registered for [program, version] |
981 | * are removed from the local rpcbind database (if the service is not | |
982 | * hidden) to make way for a new instance of the service. | |
7252d575 | 983 | * |
f6fb3f6f CL |
984 | * The result of unregistration is reported via dprintk for those who want |
985 | * verification of the result, but is otherwise not important. | |
7252d575 CL |
986 | */ |
987 | static void svc_unregister(const struct svc_serv *serv) | |
988 | { | |
989 | struct svc_program *progp; | |
990 | unsigned long flags; | |
991 | unsigned int i; | |
7252d575 CL |
992 | |
993 | clear_thread_flag(TIF_SIGPENDING); | |
994 | ||
995 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
996 | for (i = 0; i < progp->pg_nvers; i++) { | |
997 | if (progp->pg_vers[i] == NULL) | |
998 | continue; | |
999 | if (progp->pg_vers[i]->vs_hidden) | |
1000 | continue; | |
1001 | ||
7402ab19 CL |
1002 | dprintk("svc: attempting to unregister %sv%u\n", |
1003 | progp->pg_name, i); | |
f6fb3f6f | 1004 | __svc_unregister(progp->pg_prog, i, progp->pg_name); |
7252d575 | 1005 | } |
1da177e4 LT |
1006 | } |
1007 | ||
7252d575 CL |
1008 | spin_lock_irqsave(¤t->sighand->siglock, flags); |
1009 | recalc_sigpending(); | |
1010 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
1da177e4 LT |
1011 | } |
1012 | ||
354ecbb9 DDAG |
1013 | /* |
1014 | * Printk the given error with the address of the client that caused it. | |
1015 | */ | |
b9075fa9 JP |
1016 | static __printf(2, 3) |
1017 | int svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) | |
354ecbb9 DDAG |
1018 | { |
1019 | va_list args; | |
1020 | int r; | |
1021 | char buf[RPC_MAX_ADDRBUFLEN]; | |
1022 | ||
1023 | if (!net_ratelimit()) | |
1024 | return 0; | |
1025 | ||
1026 | printk(KERN_WARNING "svc: %s: ", | |
1027 | svc_print_addr(rqstp, buf, sizeof(buf))); | |
1028 | ||
1029 | va_start(args, fmt); | |
1030 | r = vprintk(fmt, args); | |
1031 | va_end(args); | |
1032 | ||
1033 | return r; | |
1034 | } | |
1035 | ||
1da177e4 | 1036 | /* |
1cad7ea6 | 1037 | * Common routine for processing the RPC request. |
1da177e4 | 1038 | */ |
1cad7ea6 RL |
1039 | static int |
1040 | svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) | |
1da177e4 LT |
1041 | { |
1042 | struct svc_program *progp; | |
1043 | struct svc_version *versp = NULL; /* compiler food */ | |
1044 | struct svc_procedure *procp = NULL; | |
6fb2b47f | 1045 | struct svc_serv *serv = rqstp->rq_server; |
1da177e4 | 1046 | kxdrproc_t xdr; |
d8ed029d | 1047 | __be32 *statp; |
1cad7ea6 | 1048 | u32 prog, vers, proc; |
d8ed029d | 1049 | __be32 auth_stat, rpc_stat; |
1da177e4 | 1050 | int auth_res; |
8f8e05c5 | 1051 | __be32 *reply_statp; |
1da177e4 LT |
1052 | |
1053 | rpc_stat = rpc_success; | |
1054 | ||
1055 | if (argv->iov_len < 6*4) | |
1056 | goto err_short_len; | |
1057 | ||
5c04c46a | 1058 | /* Will be turned off only in gss privacy case: */ |
cf8208d0 | 1059 | rqstp->rq_splice_ok = 1; |
2f425878 AA |
1060 | /* Will be turned off only when NFSv4 Sessions are used */ |
1061 | rqstp->rq_usedeferral = 1; | |
9e701c61 | 1062 | rqstp->rq_dropme = false; |
e831fe65 TT |
1063 | |
1064 | /* Setup reply header */ | |
1065 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); | |
1da177e4 | 1066 | |
1da177e4 LT |
1067 | svc_putu32(resv, rqstp->rq_xid); |
1068 | ||
76994313 | 1069 | vers = svc_getnl(argv); |
1da177e4 LT |
1070 | |
1071 | /* First words of reply: */ | |
76994313 | 1072 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 | 1073 | |
1da177e4 LT |
1074 | if (vers != 2) /* RPC version number */ |
1075 | goto err_bad_rpc; | |
1076 | ||
1077 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 1078 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 1079 | |
76994313 | 1080 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 1081 | |
76994313 AD |
1082 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
1083 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | |
1084 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 LT |
1085 | |
1086 | progp = serv->sv_program; | |
80d188a6 N |
1087 | |
1088 | for (progp = serv->sv_program; progp; progp = progp->pg_next) | |
1089 | if (prog == progp->pg_prog) | |
1090 | break; | |
1091 | ||
1da177e4 LT |
1092 | /* |
1093 | * Decode auth data, and add verifier to reply buffer. | |
1094 | * We do this before anything else in order to get a decent | |
1095 | * auth verifier. | |
1096 | */ | |
1097 | auth_res = svc_authenticate(rqstp, &auth_stat); | |
1098 | /* Also give the program a chance to reject this call: */ | |
80d188a6 | 1099 | if (auth_res == SVC_OK && progp) { |
1da177e4 LT |
1100 | auth_stat = rpc_autherr_badcred; |
1101 | auth_res = progp->pg_authenticate(rqstp); | |
1102 | } | |
1103 | switch (auth_res) { | |
1104 | case SVC_OK: | |
1105 | break; | |
1106 | case SVC_GARBAGE: | |
dd35210e | 1107 | goto err_garbage; |
1da177e4 LT |
1108 | case SVC_SYSERR: |
1109 | rpc_stat = rpc_system_err; | |
1110 | goto err_bad; | |
1111 | case SVC_DENIED: | |
1112 | goto err_bad_auth; | |
1ebede86 N |
1113 | case SVC_CLOSE: |
1114 | if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) | |
1115 | svc_close_xprt(rqstp->rq_xprt); | |
1da177e4 LT |
1116 | case SVC_DROP: |
1117 | goto dropit; | |
1118 | case SVC_COMPLETE: | |
1119 | goto sendit; | |
1120 | } | |
80d188a6 | 1121 | |
9ba02638 | 1122 | if (progp == NULL) |
1da177e4 LT |
1123 | goto err_bad_prog; |
1124 | ||
1125 | if (vers >= progp->pg_nvers || | |
1126 | !(versp = progp->pg_vers[vers])) | |
1127 | goto err_bad_vers; | |
1128 | ||
1129 | procp = versp->vs_proc + proc; | |
1130 | if (proc >= versp->vs_nproc || !procp->pc_func) | |
1131 | goto err_bad_proc; | |
1da177e4 LT |
1132 | rqstp->rq_procinfo = procp; |
1133 | ||
1134 | /* Syntactic check complete */ | |
1135 | serv->sv_stats->rpccnt++; | |
1136 | ||
1137 | /* Build the reply header. */ | |
1138 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 1139 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 LT |
1140 | |
1141 | /* Bump per-procedure stats counter */ | |
1142 | procp->pc_count++; | |
1143 | ||
1144 | /* Initialize storage for argp and resp */ | |
1145 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | |
1146 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | |
1147 | ||
cca5172a | 1148 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
1149 | * better idea of reply size |
1150 | */ | |
1151 | if (procp->pc_xdrressize) | |
cd123012 | 1152 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1da177e4 LT |
1153 | |
1154 | /* Call the function that processes the request. */ | |
1155 | if (!versp->vs_dispatch) { | |
1156 | /* Decode arguments */ | |
1157 | xdr = procp->pc_decode; | |
1158 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | |
1159 | goto err_garbage; | |
1160 | ||
1161 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
1162 | ||
1163 | /* Encode reply */ | |
9e701c61 | 1164 | if (rqstp->rq_dropme) { |
d343fce1 N |
1165 | if (procp->pc_release) |
1166 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1167 | goto dropit; | |
1168 | } | |
f64f9e71 JP |
1169 | if (*statp == rpc_success && |
1170 | (xdr = procp->pc_encode) && | |
1171 | !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | |
1da177e4 LT |
1172 | dprintk("svc: failed to encode reply\n"); |
1173 | /* serv->sv_stats->rpcsystemerr++; */ | |
1174 | *statp = rpc_system_err; | |
1175 | } | |
1176 | } else { | |
1177 | dprintk("svc: calling dispatcher\n"); | |
1178 | if (!versp->vs_dispatch(rqstp, statp)) { | |
1179 | /* Release reply info */ | |
1180 | if (procp->pc_release) | |
1181 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1182 | goto dropit; | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | /* Check RPC status result */ | |
1187 | if (*statp != rpc_success) | |
1188 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
1189 | ||
1190 | /* Release reply info */ | |
1191 | if (procp->pc_release) | |
1192 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1193 | ||
1194 | if (procp->pc_encode == NULL) | |
1195 | goto dropit; | |
1196 | ||
1197 | sendit: | |
1198 | if (svc_authorise(rqstp)) | |
1199 | goto dropit; | |
1cad7ea6 | 1200 | return 1; /* Caller can now send it */ |
1da177e4 LT |
1201 | |
1202 | dropit: | |
1203 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
1204 | dprintk("svc: svc_process dropit\n"); | |
1da177e4 LT |
1205 | return 0; |
1206 | ||
1207 | err_short_len: | |
354ecbb9 DDAG |
1208 | svc_printk(rqstp, "short len %Zd, dropping request\n", |
1209 | argv->iov_len); | |
34e9a63b | 1210 | |
1da177e4 LT |
1211 | goto dropit; /* drop request */ |
1212 | ||
1da177e4 LT |
1213 | err_bad_rpc: |
1214 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
1215 | svc_putnl(resv, 1); /* REJECT */ |
1216 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
1217 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
1218 | svc_putnl(resv, 2); | |
1da177e4 LT |
1219 | goto sendit; |
1220 | ||
1221 | err_bad_auth: | |
1222 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | |
1223 | serv->sv_stats->rpcbadauth++; | |
1224 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 1225 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
1226 | svc_putnl(resv, 1); /* REJECT */ |
1227 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
1228 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | |
1da177e4 LT |
1229 | goto sendit; |
1230 | ||
1231 | err_bad_prog: | |
9ba02638 | 1232 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 1233 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1234 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
1235 | goto sendit; |
1236 | ||
1237 | err_bad_vers: | |
354ecbb9 | 1238 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
34e9a63b N |
1239 | vers, prog, progp->pg_name); |
1240 | ||
1da177e4 | 1241 | serv->sv_stats->rpcbadfmt++; |
76994313 AD |
1242 | svc_putnl(resv, RPC_PROG_MISMATCH); |
1243 | svc_putnl(resv, progp->pg_lovers); | |
1244 | svc_putnl(resv, progp->pg_hivers); | |
1da177e4 LT |
1245 | goto sendit; |
1246 | ||
1247 | err_bad_proc: | |
354ecbb9 | 1248 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); |
34e9a63b | 1249 | |
1da177e4 | 1250 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1251 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
1252 | goto sendit; |
1253 | ||
1254 | err_garbage: | |
354ecbb9 | 1255 | svc_printk(rqstp, "failed to decode args\n"); |
34e9a63b | 1256 | |
1da177e4 LT |
1257 | rpc_stat = rpc_garbage_args; |
1258 | err_bad: | |
1259 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 1260 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
1261 | goto sendit; |
1262 | } | |
24c3767e | 1263 | EXPORT_SYMBOL_GPL(svc_process); |
7adae489 | 1264 | |
1cad7ea6 RL |
1265 | /* |
1266 | * Process the RPC request. | |
1267 | */ | |
1268 | int | |
1269 | svc_process(struct svc_rqst *rqstp) | |
1270 | { | |
1271 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1272 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1273 | struct svc_serv *serv = rqstp->rq_server; | |
1274 | u32 dir; | |
1cad7ea6 RL |
1275 | |
1276 | /* | |
1277 | * Setup response xdr_buf. | |
1278 | * Initially it has just one page | |
1279 | */ | |
1280 | rqstp->rq_resused = 1; | |
1281 | resv->iov_base = page_address(rqstp->rq_respages[0]); | |
1282 | resv->iov_len = 0; | |
1283 | rqstp->rq_res.pages = rqstp->rq_respages + 1; | |
1284 | rqstp->rq_res.len = 0; | |
1285 | rqstp->rq_res.page_base = 0; | |
1286 | rqstp->rq_res.page_len = 0; | |
1287 | rqstp->rq_res.buflen = PAGE_SIZE; | |
1288 | rqstp->rq_res.tail[0].iov_base = NULL; | |
1289 | rqstp->rq_res.tail[0].iov_len = 0; | |
1290 | ||
1291 | rqstp->rq_xid = svc_getu32(argv); | |
1292 | ||
1293 | dir = svc_getnl(argv); | |
1294 | if (dir != 0) { | |
1295 | /* direction != CALL */ | |
1296 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); | |
1297 | serv->sv_stats->rpcbadfmt++; | |
1298 | svc_drop(rqstp); | |
1299 | return 0; | |
1300 | } | |
1301 | ||
4b5b3ba1 AA |
1302 | /* Returns 1 for send, 0 for drop */ |
1303 | if (svc_process_common(rqstp, argv, resv)) | |
1304 | return svc_send(rqstp); | |
1305 | else { | |
1306 | svc_drop(rqstp); | |
1307 | return 0; | |
1308 | } | |
1cad7ea6 RL |
1309 | } |
1310 | ||
9e00abc3 | 1311 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
4d6bbb62 RL |
1312 | /* |
1313 | * Process a backchannel RPC request that arrived over an existing | |
1314 | * outbound connection | |
1315 | */ | |
1316 | int | |
1317 | bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, | |
1318 | struct svc_rqst *rqstp) | |
1319 | { | |
1320 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1321 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
4d6bbb62 RL |
1322 | |
1323 | /* Build the svc_rqst used by the common processing routine */ | |
4a19de0f | 1324 | rqstp->rq_xprt = serv->sv_bc_xprt; |
4d6bbb62 RL |
1325 | rqstp->rq_xid = req->rq_xid; |
1326 | rqstp->rq_prot = req->rq_xprt->prot; | |
1327 | rqstp->rq_server = serv; | |
1328 | ||
1329 | rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); | |
1330 | memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); | |
1331 | memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); | |
1332 | memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); | |
1333 | ||
1334 | /* reset result send buffer "put" position */ | |
1335 | resv->iov_len = 0; | |
1336 | ||
1337 | if (rqstp->rq_prot != IPPROTO_TCP) { | |
1338 | printk(KERN_ERR "No support for Non-TCP transports!\n"); | |
1339 | BUG(); | |
1340 | } | |
1341 | ||
1342 | /* | |
1343 | * Skip the next two words because they've already been | |
1344 | * processed in the trasport | |
1345 | */ | |
1346 | svc_getu32(argv); /* XID */ | |
1347 | svc_getnl(argv); /* CALLDIR */ | |
1348 | ||
4b5b3ba1 AA |
1349 | /* Returns 1 for send, 0 for drop */ |
1350 | if (svc_process_common(rqstp, argv, resv)) { | |
1351 | memcpy(&req->rq_snd_buf, &rqstp->rq_res, | |
1352 | sizeof(req->rq_snd_buf)); | |
1353 | return bc_send(req); | |
1354 | } else { | |
1355 | /* Nothing to do to drop request */ | |
1356 | return 0; | |
1357 | } | |
4d6bbb62 | 1358 | } |
0d961aa9 | 1359 | EXPORT_SYMBOL_GPL(bc_svc_process); |
9e00abc3 | 1360 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
4d6bbb62 | 1361 | |
7adae489 GB |
1362 | /* |
1363 | * Return (transport-specific) limit on the rpc payload. | |
1364 | */ | |
1365 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
1366 | { | |
49023155 | 1367 | u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; |
7adae489 | 1368 | |
c6b0a9f8 N |
1369 | if (rqstp->rq_server->sv_max_payload < max) |
1370 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
1371 | return max; |
1372 | } | |
1373 | EXPORT_SYMBOL_GPL(svc_max_payload); |