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> |
bee42f68 | 23 | #include <linux/nsproxy.h> |
1da177e4 LT |
24 | |
25 | #include <linux/sunrpc/types.h> | |
26 | #include <linux/sunrpc/xdr.h> | |
27 | #include <linux/sunrpc/stats.h> | |
28 | #include <linux/sunrpc/svcsock.h> | |
29 | #include <linux/sunrpc/clnt.h> | |
4d6bbb62 | 30 | #include <linux/sunrpc/bc_xprt.h> |
1da177e4 LT |
31 | |
32 | #define RPCDBG_FACILITY RPCDBG_SVCDSP | |
1da177e4 | 33 | |
5247fab5 | 34 | static void svc_unregister(const struct svc_serv *serv, struct net *net); |
7252d575 | 35 | |
42a7fc4a GB |
36 | #define svc_serv_is_pooled(serv) ((serv)->sv_function) |
37 | ||
bfd24160 GB |
38 | /* |
39 | * Mode for mapping cpus to pools. | |
40 | */ | |
41 | enum { | |
42a7fc4a | 42 | SVC_POOL_AUTO = -1, /* choose one of the others */ |
bfd24160 GB |
43 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool |
44 | * (legacy & UP mode) */ | |
45 | SVC_POOL_PERCPU, /* one pool per cpu */ | |
46 | SVC_POOL_PERNODE /* one pool per numa node */ | |
47 | }; | |
42a7fc4a | 48 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
bfd24160 GB |
49 | |
50 | /* | |
51 | * Structure for mapping cpus to pools and vice versa. | |
52 | * Setup once during sunrpc initialisation. | |
53 | */ | |
54 | static struct svc_pool_map { | |
42a7fc4a | 55 | int count; /* How many svc_servs use us */ |
bfd24160 GB |
56 | int mode; /* Note: int not enum to avoid |
57 | * warnings about "enumeration value | |
58 | * not handled in switch" */ | |
59 | unsigned int npools; | |
60 | unsigned int *pool_to; /* maps pool id to cpu or node */ | |
61 | unsigned int *to_pool; /* maps cpu or node to pool id */ | |
62 | } svc_pool_map = { | |
42a7fc4a GB |
63 | .count = 0, |
64 | .mode = SVC_POOL_DEFAULT | |
bfd24160 | 65 | }; |
42a7fc4a GB |
66 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
67 | ||
68 | static int | |
69 | param_set_pool_mode(const char *val, struct kernel_param *kp) | |
70 | { | |
71 | int *ip = (int *)kp->arg; | |
72 | struct svc_pool_map *m = &svc_pool_map; | |
73 | int err; | |
74 | ||
75 | mutex_lock(&svc_pool_map_mutex); | |
76 | ||
77 | err = -EBUSY; | |
78 | if (m->count) | |
79 | goto out; | |
80 | ||
81 | err = 0; | |
82 | if (!strncmp(val, "auto", 4)) | |
83 | *ip = SVC_POOL_AUTO; | |
84 | else if (!strncmp(val, "global", 6)) | |
85 | *ip = SVC_POOL_GLOBAL; | |
86 | else if (!strncmp(val, "percpu", 6)) | |
87 | *ip = SVC_POOL_PERCPU; | |
88 | else if (!strncmp(val, "pernode", 7)) | |
89 | *ip = SVC_POOL_PERNODE; | |
90 | else | |
91 | err = -EINVAL; | |
92 | ||
93 | out: | |
94 | mutex_unlock(&svc_pool_map_mutex); | |
95 | return err; | |
96 | } | |
97 | ||
98 | static int | |
99 | param_get_pool_mode(char *buf, struct kernel_param *kp) | |
100 | { | |
101 | int *ip = (int *)kp->arg; | |
102 | ||
103 | switch (*ip) | |
104 | { | |
105 | case SVC_POOL_AUTO: | |
106 | return strlcpy(buf, "auto", 20); | |
107 | case SVC_POOL_GLOBAL: | |
108 | return strlcpy(buf, "global", 20); | |
109 | case SVC_POOL_PERCPU: | |
110 | return strlcpy(buf, "percpu", 20); | |
111 | case SVC_POOL_PERNODE: | |
112 | return strlcpy(buf, "pernode", 20); | |
113 | default: | |
114 | return sprintf(buf, "%d", *ip); | |
115 | } | |
116 | } | |
bfd24160 | 117 | |
42a7fc4a GB |
118 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
119 | &svc_pool_map.mode, 0644); | |
bfd24160 GB |
120 | |
121 | /* | |
122 | * Detect best pool mapping mode heuristically, | |
123 | * according to the machine's topology. | |
124 | */ | |
125 | static int | |
126 | svc_pool_map_choose_mode(void) | |
127 | { | |
128 | unsigned int node; | |
129 | ||
62bc62a8 | 130 | if (nr_online_nodes > 1) { |
bfd24160 GB |
131 | /* |
132 | * Actually have multiple NUMA nodes, | |
133 | * so split pools on NUMA node boundaries | |
134 | */ | |
135 | return SVC_POOL_PERNODE; | |
136 | } | |
137 | ||
72c33688 | 138 | node = first_online_node; |
bfd24160 GB |
139 | if (nr_cpus_node(node) > 2) { |
140 | /* | |
141 | * Non-trivial SMP, or CONFIG_NUMA on | |
142 | * non-NUMA hardware, e.g. with a generic | |
143 | * x86_64 kernel on Xeons. In this case we | |
144 | * want to divide the pools on cpu boundaries. | |
145 | */ | |
146 | return SVC_POOL_PERCPU; | |
147 | } | |
148 | ||
149 | /* default: one global pool */ | |
150 | return SVC_POOL_GLOBAL; | |
151 | } | |
152 | ||
153 | /* | |
154 | * Allocate the to_pool[] and pool_to[] arrays. | |
155 | * Returns 0 on success or an errno. | |
156 | */ | |
157 | static int | |
158 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
159 | { | |
160 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
161 | if (!m->to_pool) | |
162 | goto fail; | |
163 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
164 | if (!m->pool_to) | |
165 | goto fail_free; | |
166 | ||
167 | return 0; | |
168 | ||
169 | fail_free: | |
170 | kfree(m->to_pool); | |
61c8504c | 171 | m->to_pool = NULL; |
bfd24160 GB |
172 | fail: |
173 | return -ENOMEM; | |
174 | } | |
175 | ||
176 | /* | |
177 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
178 | * Returns number of pools or <0 on error. | |
179 | */ | |
180 | static int | |
181 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
182 | { | |
53b8a315 | 183 | unsigned int maxpools = nr_cpu_ids; |
bfd24160 GB |
184 | unsigned int pidx = 0; |
185 | unsigned int cpu; | |
186 | int err; | |
187 | ||
188 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
189 | if (err) | |
190 | return err; | |
191 | ||
192 | for_each_online_cpu(cpu) { | |
193 | BUG_ON(pidx > maxpools); | |
194 | m->to_pool[cpu] = pidx; | |
195 | m->pool_to[pidx] = cpu; | |
196 | pidx++; | |
197 | } | |
198 | /* cpus brought online later all get mapped to pool0, sorry */ | |
199 | ||
200 | return pidx; | |
201 | }; | |
202 | ||
203 | ||
204 | /* | |
205 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
206 | * Returns number of pools or <0 on error. | |
207 | */ | |
208 | static int | |
209 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
210 | { | |
74c7aa8b | 211 | unsigned int maxpools = nr_node_ids; |
bfd24160 GB |
212 | unsigned int pidx = 0; |
213 | unsigned int node; | |
214 | int err; | |
215 | ||
216 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
217 | if (err) | |
218 | return err; | |
219 | ||
220 | for_each_node_with_cpus(node) { | |
221 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
222 | BUG_ON(pidx > maxpools); | |
223 | m->to_pool[node] = pidx; | |
224 | m->pool_to[pidx] = node; | |
225 | pidx++; | |
226 | } | |
227 | /* nodes brought online later all get mapped to pool0, sorry */ | |
228 | ||
229 | return pidx; | |
230 | } | |
231 | ||
232 | ||
233 | /* | |
42a7fc4a GB |
234 | * Add a reference to the global map of cpus to pools (and |
235 | * vice versa). Initialise the map if we're the first user. | |
236 | * Returns the number of pools. | |
bfd24160 GB |
237 | */ |
238 | static unsigned int | |
42a7fc4a | 239 | svc_pool_map_get(void) |
bfd24160 GB |
240 | { |
241 | struct svc_pool_map *m = &svc_pool_map; | |
242 | int npools = -1; | |
243 | ||
42a7fc4a GB |
244 | mutex_lock(&svc_pool_map_mutex); |
245 | ||
246 | if (m->count++) { | |
247 | mutex_unlock(&svc_pool_map_mutex); | |
bfd24160 | 248 | return m->npools; |
42a7fc4a | 249 | } |
bfd24160 | 250 | |
42a7fc4a GB |
251 | if (m->mode == SVC_POOL_AUTO) |
252 | m->mode = svc_pool_map_choose_mode(); | |
bfd24160 GB |
253 | |
254 | switch (m->mode) { | |
255 | case SVC_POOL_PERCPU: | |
256 | npools = svc_pool_map_init_percpu(m); | |
257 | break; | |
258 | case SVC_POOL_PERNODE: | |
259 | npools = svc_pool_map_init_pernode(m); | |
260 | break; | |
261 | } | |
262 | ||
263 | if (npools < 0) { | |
264 | /* default, or memory allocation failure */ | |
265 | npools = 1; | |
266 | m->mode = SVC_POOL_GLOBAL; | |
267 | } | |
268 | m->npools = npools; | |
269 | ||
42a7fc4a | 270 | mutex_unlock(&svc_pool_map_mutex); |
bfd24160 GB |
271 | return m->npools; |
272 | } | |
273 | ||
42a7fc4a GB |
274 | |
275 | /* | |
276 | * Drop a reference to the global map of cpus to pools. | |
277 | * When the last reference is dropped, the map data is | |
278 | * freed; this allows the sysadmin to change the pool | |
279 | * mode using the pool_mode module option without | |
280 | * rebooting or re-loading sunrpc.ko. | |
281 | */ | |
282 | static void | |
283 | svc_pool_map_put(void) | |
284 | { | |
285 | struct svc_pool_map *m = &svc_pool_map; | |
286 | ||
287 | mutex_lock(&svc_pool_map_mutex); | |
288 | ||
289 | if (!--m->count) { | |
42a7fc4a | 290 | kfree(m->to_pool); |
61c8504c | 291 | m->to_pool = NULL; |
42a7fc4a | 292 | kfree(m->pool_to); |
61c8504c | 293 | m->pool_to = NULL; |
42a7fc4a GB |
294 | m->npools = 0; |
295 | } | |
296 | ||
297 | mutex_unlock(&svc_pool_map_mutex); | |
298 | } | |
299 | ||
300 | ||
11fd165c ED |
301 | static int svc_pool_map_get_node(unsigned int pidx) |
302 | { | |
303 | const struct svc_pool_map *m = &svc_pool_map; | |
304 | ||
305 | if (m->count) { | |
306 | if (m->mode == SVC_POOL_PERCPU) | |
307 | return cpu_to_node(m->pool_to[pidx]); | |
308 | if (m->mode == SVC_POOL_PERNODE) | |
309 | return m->pool_to[pidx]; | |
310 | } | |
311 | return NUMA_NO_NODE; | |
312 | } | |
bfd24160 | 313 | /* |
9867d76c | 314 | * Set the given thread's cpus_allowed mask so that it |
bfd24160 | 315 | * will only run on cpus in the given pool. |
bfd24160 | 316 | */ |
9867d76c JL |
317 | static inline void |
318 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) | |
bfd24160 GB |
319 | { |
320 | struct svc_pool_map *m = &svc_pool_map; | |
9867d76c | 321 | unsigned int node = m->pool_to[pidx]; |
bfd24160 GB |
322 | |
323 | /* | |
324 | * The caller checks for sv_nrpools > 1, which | |
42a7fc4a | 325 | * implies that we've been initialized. |
bfd24160 | 326 | */ |
1bd58aaf WAA |
327 | WARN_ON_ONCE(m->count == 0); |
328 | if (m->count == 0) | |
329 | return; | |
bfd24160 | 330 | |
9867d76c | 331 | switch (m->mode) { |
bfd24160 | 332 | case SVC_POOL_PERCPU: |
c5f59f08 | 333 | { |
aa85ea5b | 334 | set_cpus_allowed_ptr(task, cpumask_of(node)); |
9867d76c | 335 | break; |
c5f59f08 | 336 | } |
bfd24160 | 337 | case SVC_POOL_PERNODE: |
c5f59f08 | 338 | { |
a70f7302 | 339 | set_cpus_allowed_ptr(task, cpumask_of_node(node)); |
9867d76c | 340 | break; |
bfd24160 | 341 | } |
c5f59f08 | 342 | } |
bfd24160 GB |
343 | } |
344 | ||
345 | /* | |
346 | * Use the mapping mode to choose a pool for a given CPU. | |
347 | * Used when enqueueing an incoming RPC. Always returns | |
348 | * a non-NULL pool pointer. | |
349 | */ | |
350 | struct svc_pool * | |
351 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | |
352 | { | |
353 | struct svc_pool_map *m = &svc_pool_map; | |
354 | unsigned int pidx = 0; | |
355 | ||
356 | /* | |
42a7fc4a | 357 | * An uninitialised map happens in a pure client when |
bfd24160 GB |
358 | * lockd is brought up, so silently treat it the |
359 | * same as SVC_POOL_GLOBAL. | |
360 | */ | |
42a7fc4a GB |
361 | if (svc_serv_is_pooled(serv)) { |
362 | switch (m->mode) { | |
363 | case SVC_POOL_PERCPU: | |
364 | pidx = m->to_pool[cpu]; | |
365 | break; | |
366 | case SVC_POOL_PERNODE: | |
367 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
368 | break; | |
369 | } | |
bfd24160 GB |
370 | } |
371 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | |
372 | } | |
373 | ||
bb2224df | 374 | int svc_rpcb_setup(struct svc_serv *serv, struct net *net) |
d9908560 SK |
375 | { |
376 | int err; | |
377 | ||
bee42f68 | 378 | err = rpcb_create_local(net); |
d9908560 SK |
379 | if (err) |
380 | return err; | |
381 | ||
382 | /* Remove any stale portmap registrations */ | |
bee42f68 | 383 | svc_unregister(serv, net); |
d9908560 SK |
384 | return 0; |
385 | } | |
bb2224df | 386 | EXPORT_SYMBOL_GPL(svc_rpcb_setup); |
d9908560 | 387 | |
5ecebb7c | 388 | void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) |
d9908560 | 389 | { |
5ecebb7c SK |
390 | svc_unregister(serv, net); |
391 | rpcb_put_local(net); | |
d9908560 | 392 | } |
16d05870 | 393 | EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); |
d9908560 SK |
394 | |
395 | static int svc_uses_rpcbind(struct svc_serv *serv) | |
396 | { | |
397 | struct svc_program *progp; | |
398 | unsigned int i; | |
399 | ||
400 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
401 | for (i = 0; i < progp->pg_nvers; i++) { | |
402 | if (progp->pg_vers[i] == NULL) | |
403 | continue; | |
404 | if (progp->pg_vers[i]->vs_hidden == 0) | |
405 | return 1; | |
406 | } | |
407 | } | |
408 | ||
409 | return 0; | |
410 | } | |
bfd24160 | 411 | |
9793f7c8 SK |
412 | int svc_bind(struct svc_serv *serv, struct net *net) |
413 | { | |
414 | if (!svc_uses_rpcbind(serv)) | |
415 | return 0; | |
416 | return svc_rpcb_setup(serv, net); | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(svc_bind); | |
419 | ||
1da177e4 LT |
420 | /* |
421 | * Create an RPC service | |
422 | */ | |
a7455442 GB |
423 | static struct svc_serv * |
424 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
5ecebb7c | 425 | void (*shutdown)(struct svc_serv *serv, struct net *net)) |
1da177e4 LT |
426 | { |
427 | struct svc_serv *serv; | |
ea339d46 | 428 | unsigned int vers; |
1da177e4 | 429 | unsigned int xdrsize; |
3262c816 | 430 | unsigned int i; |
1da177e4 | 431 | |
0da974f4 | 432 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 433 | return NULL; |
9ba02638 | 434 | serv->sv_name = prog->pg_name; |
1da177e4 LT |
435 | serv->sv_program = prog; |
436 | serv->sv_nrthreads = 1; | |
437 | serv->sv_stats = prog->pg_stats; | |
c6b0a9f8 N |
438 | if (bufsize > RPCSVC_MAXPAYLOAD) |
439 | bufsize = RPCSVC_MAXPAYLOAD; | |
440 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
441 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
bc591ccf | 442 | serv->sv_shutdown = shutdown; |
1da177e4 | 443 | xdrsize = 0; |
9ba02638 AG |
444 | while (prog) { |
445 | prog->pg_lovers = prog->pg_nvers-1; | |
446 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
447 | if (prog->pg_vers[vers]) { | |
448 | prog->pg_hivers = vers; | |
449 | if (prog->pg_lovers > vers) | |
450 | prog->pg_lovers = vers; | |
451 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
452 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
453 | } | |
454 | prog = prog->pg_next; | |
455 | } | |
1da177e4 | 456 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
457 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
458 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
36bdfc8b | 459 | init_timer(&serv->sv_temptimer); |
1da177e4 LT |
460 | spin_lock_init(&serv->sv_lock); |
461 | ||
a7455442 | 462 | serv->sv_nrpools = npools; |
3262c816 | 463 | serv->sv_pools = |
cd861280 | 464 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
465 | GFP_KERNEL); |
466 | if (!serv->sv_pools) { | |
467 | kfree(serv); | |
468 | return NULL; | |
469 | } | |
470 | ||
471 | for (i = 0; i < serv->sv_nrpools; i++) { | |
472 | struct svc_pool *pool = &serv->sv_pools[i]; | |
473 | ||
46121cf7 | 474 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
475 | i, serv->sv_name); |
476 | ||
477 | pool->sp_id = i; | |
478 | INIT_LIST_HEAD(&pool->sp_threads); | |
479 | INIT_LIST_HEAD(&pool->sp_sockets); | |
a7455442 | 480 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
481 | spin_lock_init(&pool->sp_lock); |
482 | } | |
483 | ||
9793f7c8 SK |
484 | if (svc_uses_rpcbind(serv) && (!serv->sv_shutdown)) |
485 | serv->sv_shutdown = svc_rpcb_cleanup; | |
1da177e4 LT |
486 | |
487 | return serv; | |
488 | } | |
489 | ||
a7455442 GB |
490 | struct svc_serv * |
491 | svc_create(struct svc_program *prog, unsigned int bufsize, | |
5ecebb7c | 492 | void (*shutdown)(struct svc_serv *serv, struct net *net)) |
a7455442 | 493 | { |
49a9072f | 494 | return __svc_create(prog, bufsize, /*npools*/1, shutdown); |
a7455442 | 495 | } |
24c3767e | 496 | EXPORT_SYMBOL_GPL(svc_create); |
a7455442 GB |
497 | |
498 | struct svc_serv * | |
499 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | |
5ecebb7c | 500 | void (*shutdown)(struct svc_serv *serv, struct net *net), |
a75c5d01 | 501 | svc_thread_fn func, struct module *mod) |
a7455442 GB |
502 | { |
503 | struct svc_serv *serv; | |
42a7fc4a | 504 | unsigned int npools = svc_pool_map_get(); |
a7455442 | 505 | |
49a9072f | 506 | serv = __svc_create(prog, bufsize, npools, shutdown); |
a7455442 GB |
507 | |
508 | if (serv != NULL) { | |
509 | serv->sv_function = func; | |
a7455442 GB |
510 | serv->sv_module = mod; |
511 | } | |
512 | ||
513 | return serv; | |
514 | } | |
24c3767e | 515 | EXPORT_SYMBOL_GPL(svc_create_pooled); |
a7455442 | 516 | |
074d0f67 SK |
517 | void svc_shutdown_net(struct svc_serv *serv, struct net *net) |
518 | { | |
519 | /* | |
520 | * The set of xprts (contained in the sv_tempsocks and | |
521 | * sv_permsocks lists) is now constant, since it is modified | |
522 | * only by accepting new sockets (done by service threads in | |
523 | * svc_recv) or aging old ones (done by sv_temptimer), or | |
524 | * configuration changes (excluded by whatever locking the | |
525 | * caller is using--nfsd_mutex in the case of nfsd). So it's | |
526 | * safe to traverse those lists and shut everything down: | |
527 | */ | |
528 | svc_close_net(serv, net); | |
529 | ||
530 | if (serv->sv_shutdown) | |
531 | serv->sv_shutdown(serv, net); | |
532 | } | |
533 | EXPORT_SYMBOL_GPL(svc_shutdown_net); | |
534 | ||
1da177e4 | 535 | /* |
bedbdd8b NB |
536 | * Destroy an RPC service. Should be called with appropriate locking to |
537 | * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. | |
1da177e4 LT |
538 | */ |
539 | void | |
540 | svc_destroy(struct svc_serv *serv) | |
541 | { | |
46121cf7 | 542 | dprintk("svc: svc_destroy(%s, %d)\n", |
1da177e4 LT |
543 | serv->sv_program->pg_name, |
544 | serv->sv_nrthreads); | |
545 | ||
546 | if (serv->sv_nrthreads) { | |
547 | if (--(serv->sv_nrthreads) != 0) { | |
548 | svc_sock_update_bufs(serv); | |
549 | return; | |
550 | } | |
551 | } else | |
552 | printk("svc_destroy: no threads for serv=%p!\n", serv); | |
553 | ||
36bdfc8b | 554 | del_timer_sync(&serv->sv_temptimer); |
074d0f67 | 555 | |
7b147f1f SK |
556 | /* |
557 | * The last user is gone and thus all sockets have to be destroyed to | |
558 | * the point. Check this. | |
559 | */ | |
560 | BUG_ON(!list_empty(&serv->sv_permsocks)); | |
561 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | |
cda1fd4a | 562 | |
1da177e4 LT |
563 | cache_clean_deferred(serv); |
564 | ||
42a7fc4a GB |
565 | if (svc_serv_is_pooled(serv)) |
566 | svc_pool_map_put(); | |
567 | ||
3262c816 | 568 | kfree(serv->sv_pools); |
1da177e4 LT |
569 | kfree(serv); |
570 | } | |
24c3767e | 571 | EXPORT_SYMBOL_GPL(svc_destroy); |
1da177e4 LT |
572 | |
573 | /* | |
574 | * Allocate an RPC server's buffer space. | |
575 | * We allocate pages and place them in rq_argpages. | |
576 | */ | |
577 | static int | |
11fd165c | 578 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) |
1da177e4 | 579 | { |
0dc220f0 | 580 | unsigned int pages, arghi; |
cca5172a | 581 | |
ba17686f AA |
582 | /* bc_xprt uses fore channel allocated buffers */ |
583 | if (svc_is_backchannel(rqstp)) | |
584 | return 1; | |
585 | ||
c6b0a9f8 N |
586 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
587 | * We assume one is at most one page | |
588 | */ | |
1da177e4 | 589 | arghi = 0; |
b25cd058 WAA |
590 | WARN_ON_ONCE(pages > RPCSVC_MAXPAGES); |
591 | if (pages > RPCSVC_MAXPAGES) | |
592 | pages = RPCSVC_MAXPAGES; | |
1da177e4 | 593 | while (pages) { |
11fd165c | 594 | struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); |
1da177e4 LT |
595 | if (!p) |
596 | break; | |
44524359 | 597 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
598 | pages--; |
599 | } | |
0dc220f0 | 600 | return pages == 0; |
1da177e4 LT |
601 | } |
602 | ||
603 | /* | |
604 | * Release an RPC server buffer | |
605 | */ | |
606 | static void | |
607 | svc_release_buffer(struct svc_rqst *rqstp) | |
608 | { | |
50c8bb13 CL |
609 | unsigned int i; |
610 | ||
611 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) | |
44524359 N |
612 | if (rqstp->rq_pages[i]) |
613 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
614 | } |
615 | ||
0113ab34 | 616 | struct svc_rqst * |
11fd165c | 617 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) |
1da177e4 LT |
618 | { |
619 | struct svc_rqst *rqstp; | |
1da177e4 | 620 | |
11fd165c | 621 | rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); |
1da177e4 | 622 | if (!rqstp) |
0113ab34 | 623 | goto out_enomem; |
1da177e4 | 624 | |
1da177e4 LT |
625 | init_waitqueue_head(&rqstp->rq_wait); |
626 | ||
1da177e4 | 627 | serv->sv_nrthreads++; |
3262c816 GB |
628 | spin_lock_bh(&pool->sp_lock); |
629 | pool->sp_nrthreads++; | |
a7455442 | 630 | list_add(&rqstp->rq_all, &pool->sp_all_threads); |
3262c816 | 631 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 632 | rqstp->rq_server = serv; |
3262c816 | 633 | rqstp->rq_pool = pool; |
bfd24160 | 634 | |
11fd165c | 635 | rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 JL |
636 | if (!rqstp->rq_argp) |
637 | goto out_thread; | |
638 | ||
11fd165c | 639 | rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 JL |
640 | if (!rqstp->rq_resp) |
641 | goto out_thread; | |
642 | ||
11fd165c | 643 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) |
0113ab34 JL |
644 | goto out_thread; |
645 | ||
646 | return rqstp; | |
647 | out_thread: | |
648 | svc_exit_thread(rqstp); | |
649 | out_enomem: | |
650 | return ERR_PTR(-ENOMEM); | |
651 | } | |
24c3767e | 652 | EXPORT_SYMBOL_GPL(svc_prepare_thread); |
0113ab34 | 653 | |
a7455442 GB |
654 | /* |
655 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
656 | */ | |
657 | static inline struct svc_pool * | |
658 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
659 | { | |
660 | if (pool != NULL) | |
661 | return pool; | |
662 | ||
cca5172a | 663 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
664 | } |
665 | ||
666 | /* | |
667 | * Choose a thread to kill, for svc_set_num_threads | |
668 | */ | |
669 | static inline struct task_struct * | |
670 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
671 | { | |
672 | unsigned int i; | |
673 | struct task_struct *task = NULL; | |
674 | ||
675 | if (pool != NULL) { | |
676 | spin_lock_bh(&pool->sp_lock); | |
677 | } else { | |
678 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
679 | for (i = 0; i < serv->sv_nrpools; i++) { |
680 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 681 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
682 | if (!list_empty(&pool->sp_all_threads)) |
683 | goto found_pool; | |
a7455442 | 684 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 685 | } |
a7455442 GB |
686 | return NULL; |
687 | } | |
688 | ||
689 | found_pool: | |
690 | if (!list_empty(&pool->sp_all_threads)) { | |
691 | struct svc_rqst *rqstp; | |
692 | ||
693 | /* | |
694 | * Remove from the pool->sp_all_threads list | |
695 | * so we don't try to kill it again. | |
696 | */ | |
697 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
698 | list_del_init(&rqstp->rq_all); | |
699 | task = rqstp->rq_task; | |
cca5172a | 700 | } |
a7455442 GB |
701 | spin_unlock_bh(&pool->sp_lock); |
702 | ||
703 | return task; | |
704 | } | |
705 | ||
706 | /* | |
707 | * Create or destroy enough new threads to make the number | |
708 | * of threads the given number. If `pool' is non-NULL, applies | |
709 | * only to threads in that pool, otherwise round-robins between | |
94cf3179 BF |
710 | * all pools. Caller must ensure that mutual exclusion between this and |
711 | * server startup or shutdown. | |
a7455442 GB |
712 | * |
713 | * Destroying threads relies on the service threads filling in | |
714 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | |
715 | * has been created using svc_create_pooled(). | |
716 | * | |
717 | * Based on code that used to be in nfsd_svc() but tweaked | |
718 | * to be pool-aware. | |
719 | */ | |
720 | int | |
721 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
722 | { | |
9867d76c JL |
723 | struct svc_rqst *rqstp; |
724 | struct task_struct *task; | |
725 | struct svc_pool *chosen_pool; | |
a7455442 GB |
726 | int error = 0; |
727 | unsigned int state = serv->sv_nrthreads-1; | |
11fd165c | 728 | int node; |
a7455442 GB |
729 | |
730 | if (pool == NULL) { | |
731 | /* The -1 assumes caller has done a svc_get() */ | |
732 | nrservs -= (serv->sv_nrthreads-1); | |
733 | } else { | |
734 | spin_lock_bh(&pool->sp_lock); | |
735 | nrservs -= pool->sp_nrthreads; | |
736 | spin_unlock_bh(&pool->sp_lock); | |
737 | } | |
738 | ||
739 | /* create new threads */ | |
740 | while (nrservs > 0) { | |
741 | nrservs--; | |
9867d76c JL |
742 | chosen_pool = choose_pool(serv, pool, &state); |
743 | ||
11fd165c ED |
744 | node = svc_pool_map_get_node(chosen_pool->sp_id); |
745 | rqstp = svc_prepare_thread(serv, chosen_pool, node); | |
9867d76c JL |
746 | if (IS_ERR(rqstp)) { |
747 | error = PTR_ERR(rqstp); | |
748 | break; | |
749 | } | |
750 | ||
a7455442 | 751 | __module_get(serv->sv_module); |
11fd165c ED |
752 | task = kthread_create_on_node(serv->sv_function, rqstp, |
753 | node, serv->sv_name); | |
9867d76c JL |
754 | if (IS_ERR(task)) { |
755 | error = PTR_ERR(task); | |
a7455442 | 756 | module_put(serv->sv_module); |
9867d76c | 757 | svc_exit_thread(rqstp); |
a7455442 GB |
758 | break; |
759 | } | |
9867d76c JL |
760 | |
761 | rqstp->rq_task = task; | |
762 | if (serv->sv_nrpools > 1) | |
763 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); | |
764 | ||
765 | svc_sock_update_bufs(serv); | |
766 | wake_up_process(task); | |
a7455442 GB |
767 | } |
768 | /* destroy old threads */ | |
769 | while (nrservs < 0 && | |
9867d76c | 770 | (task = choose_victim(serv, pool, &state)) != NULL) { |
a75c5d01 | 771 | send_sig(SIGINT, task, 1); |
a7455442 GB |
772 | nrservs++; |
773 | } | |
774 | ||
775 | return error; | |
776 | } | |
24c3767e | 777 | EXPORT_SYMBOL_GPL(svc_set_num_threads); |
a7455442 | 778 | |
3262c816 | 779 | /* |
bedbdd8b NB |
780 | * Called from a server thread as it's exiting. Caller must hold the BKL or |
781 | * the "service mutex", whichever is appropriate for the service. | |
1da177e4 LT |
782 | */ |
783 | void | |
784 | svc_exit_thread(struct svc_rqst *rqstp) | |
785 | { | |
786 | struct svc_serv *serv = rqstp->rq_server; | |
3262c816 | 787 | struct svc_pool *pool = rqstp->rq_pool; |
1da177e4 LT |
788 | |
789 | svc_release_buffer(rqstp); | |
a51482bd JJ |
790 | kfree(rqstp->rq_resp); |
791 | kfree(rqstp->rq_argp); | |
792 | kfree(rqstp->rq_auth_data); | |
3262c816 GB |
793 | |
794 | spin_lock_bh(&pool->sp_lock); | |
795 | pool->sp_nrthreads--; | |
a7455442 | 796 | list_del(&rqstp->rq_all); |
3262c816 GB |
797 | spin_unlock_bh(&pool->sp_lock); |
798 | ||
1da177e4 LT |
799 | kfree(rqstp); |
800 | ||
801 | /* Release the server */ | |
802 | if (serv) | |
803 | svc_destroy(serv); | |
804 | } | |
24c3767e | 805 | EXPORT_SYMBOL_GPL(svc_exit_thread); |
1da177e4 LT |
806 | |
807 | /* | |
2c7eb0b2 CL |
808 | * Register an "inet" protocol family netid with the local |
809 | * rpcbind daemon via an rpcbind v4 SET request. | |
a26cfad6 | 810 | * |
2c7eb0b2 CL |
811 | * No netconfig infrastructure is available in the kernel, so |
812 | * we map IP_ protocol numbers to netids by hand. | |
a26cfad6 | 813 | * |
2c7eb0b2 CL |
814 | * Returns zero on success; a negative errno value is returned |
815 | * if any error occurs. | |
1da177e4 | 816 | */ |
5247fab5 SK |
817 | static int __svc_rpcb_register4(struct net *net, const u32 program, |
818 | const u32 version, | |
2c7eb0b2 CL |
819 | const unsigned short protocol, |
820 | const unsigned short port) | |
a26cfad6 | 821 | { |
cadc0fa5 | 822 | const struct sockaddr_in sin = { |
a26cfad6 CL |
823 | .sin_family = AF_INET, |
824 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
825 | .sin_port = htons(port), | |
826 | }; | |
cadc0fa5 CL |
827 | const char *netid; |
828 | int error; | |
2c7eb0b2 CL |
829 | |
830 | switch (protocol) { | |
831 | case IPPROTO_UDP: | |
832 | netid = RPCBIND_NETID_UDP; | |
833 | break; | |
834 | case IPPROTO_TCP: | |
835 | netid = RPCBIND_NETID_TCP; | |
836 | break; | |
837 | default: | |
ba5c35e0 | 838 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
839 | } |
840 | ||
5247fab5 | 841 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
842 | (const struct sockaddr *)&sin, netid); |
843 | ||
844 | /* | |
845 | * User space didn't support rpcbind v4, so retry this | |
846 | * registration request with the legacy rpcbind v2 protocol. | |
847 | */ | |
848 | if (error == -EPROTONOSUPPORT) | |
5247fab5 | 849 | error = rpcb_register(net, program, version, protocol, port); |
cadc0fa5 CL |
850 | |
851 | return error; | |
2c7eb0b2 CL |
852 | } |
853 | ||
dfd56b8b | 854 | #if IS_ENABLED(CONFIG_IPV6) |
2c7eb0b2 CL |
855 | /* |
856 | * Register an "inet6" protocol family netid with the local | |
857 | * rpcbind daemon via an rpcbind v4 SET request. | |
858 | * | |
859 | * No netconfig infrastructure is available in the kernel, so | |
860 | * we map IP_ protocol numbers to netids by hand. | |
861 | * | |
862 | * Returns zero on success; a negative errno value is returned | |
863 | * if any error occurs. | |
864 | */ | |
5247fab5 SK |
865 | static int __svc_rpcb_register6(struct net *net, const u32 program, |
866 | const u32 version, | |
2c7eb0b2 CL |
867 | const unsigned short protocol, |
868 | const unsigned short port) | |
869 | { | |
cadc0fa5 | 870 | const struct sockaddr_in6 sin6 = { |
a26cfad6 CL |
871 | .sin6_family = AF_INET6, |
872 | .sin6_addr = IN6ADDR_ANY_INIT, | |
873 | .sin6_port = htons(port), | |
874 | }; | |
cadc0fa5 CL |
875 | const char *netid; |
876 | int error; | |
a26cfad6 | 877 | |
2c7eb0b2 CL |
878 | switch (protocol) { |
879 | case IPPROTO_UDP: | |
880 | netid = RPCBIND_NETID_UDP6; | |
a26cfad6 | 881 | break; |
2c7eb0b2 | 882 | case IPPROTO_TCP: |
a26cfad6 | 883 | netid = RPCBIND_NETID_TCP6; |
a26cfad6 CL |
884 | break; |
885 | default: | |
ba5c35e0 | 886 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
887 | } |
888 | ||
5247fab5 | 889 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
890 | (const struct sockaddr *)&sin6, netid); |
891 | ||
892 | /* | |
893 | * User space didn't support rpcbind version 4, so we won't | |
894 | * use a PF_INET6 listener. | |
895 | */ | |
896 | if (error == -EPROTONOSUPPORT) | |
897 | error = -EAFNOSUPPORT; | |
898 | ||
899 | return error; | |
2c7eb0b2 | 900 | } |
dfd56b8b | 901 | #endif /* IS_ENABLED(CONFIG_IPV6) */ |
2c7eb0b2 CL |
902 | |
903 | /* | |
904 | * Register a kernel RPC service via rpcbind version 4. | |
905 | * | |
906 | * Returns zero on success; a negative errno value is returned | |
907 | * if any error occurs. | |
908 | */ | |
5247fab5 | 909 | static int __svc_register(struct net *net, const char *progname, |
363f724c | 910 | const u32 program, const u32 version, |
4b62e58c | 911 | const int family, |
2c7eb0b2 CL |
912 | const unsigned short protocol, |
913 | const unsigned short port) | |
914 | { | |
363f724c | 915 | int error = -EAFNOSUPPORT; |
2c7eb0b2 CL |
916 | |
917 | switch (family) { | |
4b62e58c | 918 | case PF_INET: |
5247fab5 | 919 | error = __svc_rpcb_register4(net, program, version, |
2c7eb0b2 | 920 | protocol, port); |
cadc0fa5 | 921 | break; |
dfd56b8b | 922 | #if IS_ENABLED(CONFIG_IPV6) |
4b62e58c | 923 | case PF_INET6: |
5247fab5 | 924 | error = __svc_rpcb_register6(net, program, version, |
2c7eb0b2 | 925 | protocol, port); |
dfd56b8b | 926 | #endif |
a26cfad6 CL |
927 | } |
928 | ||
363f724c CL |
929 | if (error < 0) |
930 | printk(KERN_WARNING "svc: failed to register %sv%u RPC " | |
931 | "service (errno %d).\n", progname, version, -error); | |
932 | return error; | |
a26cfad6 | 933 | } |
2c7eb0b2 | 934 | |
a26cfad6 CL |
935 | /** |
936 | * svc_register - register an RPC service with the local portmapper | |
937 | * @serv: svc_serv struct for the service to register | |
5247fab5 | 938 | * @net: net namespace for the service to register |
4b62e58c | 939 | * @family: protocol family of service's listener socket |
a26cfad6 CL |
940 | * @proto: transport protocol number to advertise |
941 | * @port: port to advertise | |
942 | * | |
4b62e58c | 943 | * Service is registered for any address in the passed-in protocol family |
a26cfad6 | 944 | */ |
5247fab5 SK |
945 | int svc_register(const struct svc_serv *serv, struct net *net, |
946 | const int family, const unsigned short proto, | |
947 | const unsigned short port) | |
1da177e4 LT |
948 | { |
949 | struct svc_program *progp; | |
ea339d46 | 950 | unsigned int i; |
14aeb211 | 951 | int error = 0; |
1da177e4 | 952 | |
7252d575 | 953 | BUG_ON(proto == 0 && port == 0); |
1da177e4 | 954 | |
bc5fea42 OK |
955 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
956 | for (i = 0; i < progp->pg_nvers; i++) { | |
957 | if (progp->pg_vers[i] == NULL) | |
958 | continue; | |
959 | ||
2c7eb0b2 | 960 | dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", |
bc5fea42 | 961 | progp->pg_name, |
2c7eb0b2 | 962 | i, |
bc5fea42 OK |
963 | proto == IPPROTO_UDP? "udp" : "tcp", |
964 | port, | |
4b62e58c | 965 | family, |
bc5fea42 OK |
966 | progp->pg_vers[i]->vs_hidden? |
967 | " (but not telling portmap)" : ""); | |
968 | ||
969 | if (progp->pg_vers[i]->vs_hidden) | |
970 | continue; | |
971 | ||
5247fab5 | 972 | error = __svc_register(net, progp->pg_name, progp->pg_prog, |
363f724c | 973 | i, family, proto, port); |
bc5fea42 OK |
974 | if (error < 0) |
975 | break; | |
1da177e4 LT |
976 | } |
977 | } | |
978 | ||
7252d575 CL |
979 | return error; |
980 | } | |
981 | ||
d5a8620f CL |
982 | /* |
983 | * If user space is running rpcbind, it should take the v4 UNSET | |
984 | * and clear everything for this [program, version]. If user space | |
985 | * is running portmap, it will reject the v4 UNSET, but won't have | |
986 | * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient | |
987 | * in this case to clear all existing entries for [program, version]. | |
988 | */ | |
5247fab5 | 989 | static void __svc_unregister(struct net *net, const u32 program, const u32 version, |
f6fb3f6f CL |
990 | const char *progname) |
991 | { | |
f6fb3f6f CL |
992 | int error; |
993 | ||
5247fab5 | 994 | error = rpcb_v4_register(net, program, version, NULL, ""); |
f6fb3f6f | 995 | |
d5a8620f CL |
996 | /* |
997 | * User space didn't support rpcbind v4, so retry this | |
998 | * request with the legacy rpcbind v2 protocol. | |
999 | */ | |
1000 | if (error == -EPROTONOSUPPORT) | |
5247fab5 | 1001 | error = rpcb_register(net, program, version, 0, 0); |
f6fb3f6f | 1002 | |
f6fb3f6f CL |
1003 | dprintk("svc: %s(%sv%u), error %d\n", |
1004 | __func__, progname, version, error); | |
1005 | } | |
1006 | ||
7252d575 | 1007 | /* |
f6fb3f6f CL |
1008 | * All netids, bind addresses and ports registered for [program, version] |
1009 | * are removed from the local rpcbind database (if the service is not | |
1010 | * hidden) to make way for a new instance of the service. | |
7252d575 | 1011 | * |
f6fb3f6f CL |
1012 | * The result of unregistration is reported via dprintk for those who want |
1013 | * verification of the result, but is otherwise not important. | |
7252d575 | 1014 | */ |
5247fab5 | 1015 | static void svc_unregister(const struct svc_serv *serv, struct net *net) |
7252d575 CL |
1016 | { |
1017 | struct svc_program *progp; | |
1018 | unsigned long flags; | |
1019 | unsigned int i; | |
7252d575 CL |
1020 | |
1021 | clear_thread_flag(TIF_SIGPENDING); | |
1022 | ||
1023 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
1024 | for (i = 0; i < progp->pg_nvers; i++) { | |
1025 | if (progp->pg_vers[i] == NULL) | |
1026 | continue; | |
1027 | if (progp->pg_vers[i]->vs_hidden) | |
1028 | continue; | |
1029 | ||
7402ab19 CL |
1030 | dprintk("svc: attempting to unregister %sv%u\n", |
1031 | progp->pg_name, i); | |
5247fab5 | 1032 | __svc_unregister(net, progp->pg_prog, i, progp->pg_name); |
7252d575 | 1033 | } |
1da177e4 LT |
1034 | } |
1035 | ||
7252d575 CL |
1036 | spin_lock_irqsave(¤t->sighand->siglock, flags); |
1037 | recalc_sigpending(); | |
1038 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
1da177e4 LT |
1039 | } |
1040 | ||
354ecbb9 DDAG |
1041 | /* |
1042 | * Printk the given error with the address of the client that caused it. | |
1043 | */ | |
b9075fa9 | 1044 | static __printf(2, 3) |
e87cc472 | 1045 | void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) |
354ecbb9 | 1046 | { |
e87cc472 | 1047 | struct va_format vaf; |
354ecbb9 | 1048 | va_list args; |
354ecbb9 DDAG |
1049 | char buf[RPC_MAX_ADDRBUFLEN]; |
1050 | ||
e87cc472 | 1051 | va_start(args, fmt); |
354ecbb9 | 1052 | |
e87cc472 JP |
1053 | vaf.fmt = fmt; |
1054 | vaf.va = &args; | |
354ecbb9 | 1055 | |
e87cc472 JP |
1056 | net_warn_ratelimited("svc: %s: %pV", |
1057 | svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); | |
354ecbb9 | 1058 | |
e87cc472 | 1059 | va_end(args); |
354ecbb9 DDAG |
1060 | } |
1061 | ||
1da177e4 | 1062 | /* |
1cad7ea6 | 1063 | * Common routine for processing the RPC request. |
1da177e4 | 1064 | */ |
1cad7ea6 RL |
1065 | static int |
1066 | svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) | |
1da177e4 LT |
1067 | { |
1068 | struct svc_program *progp; | |
1069 | struct svc_version *versp = NULL; /* compiler food */ | |
1070 | struct svc_procedure *procp = NULL; | |
6fb2b47f | 1071 | struct svc_serv *serv = rqstp->rq_server; |
1da177e4 | 1072 | kxdrproc_t xdr; |
d8ed029d | 1073 | __be32 *statp; |
1cad7ea6 | 1074 | u32 prog, vers, proc; |
d8ed029d | 1075 | __be32 auth_stat, rpc_stat; |
1da177e4 | 1076 | int auth_res; |
8f8e05c5 | 1077 | __be32 *reply_statp; |
1da177e4 LT |
1078 | |
1079 | rpc_stat = rpc_success; | |
1080 | ||
1081 | if (argv->iov_len < 6*4) | |
1082 | goto err_short_len; | |
1083 | ||
5c04c46a | 1084 | /* Will be turned off only in gss privacy case: */ |
cf8208d0 | 1085 | rqstp->rq_splice_ok = 1; |
2f425878 AA |
1086 | /* Will be turned off only when NFSv4 Sessions are used */ |
1087 | rqstp->rq_usedeferral = 1; | |
9e701c61 | 1088 | rqstp->rq_dropme = false; |
e831fe65 TT |
1089 | |
1090 | /* Setup reply header */ | |
1091 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); | |
1da177e4 | 1092 | |
1da177e4 LT |
1093 | svc_putu32(resv, rqstp->rq_xid); |
1094 | ||
76994313 | 1095 | vers = svc_getnl(argv); |
1da177e4 LT |
1096 | |
1097 | /* First words of reply: */ | |
76994313 | 1098 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 | 1099 | |
1da177e4 LT |
1100 | if (vers != 2) /* RPC version number */ |
1101 | goto err_bad_rpc; | |
1102 | ||
1103 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 1104 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 1105 | |
76994313 | 1106 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 1107 | |
76994313 AD |
1108 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
1109 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | |
1110 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 LT |
1111 | |
1112 | progp = serv->sv_program; | |
80d188a6 N |
1113 | |
1114 | for (progp = serv->sv_program; progp; progp = progp->pg_next) | |
1115 | if (prog == progp->pg_prog) | |
1116 | break; | |
1117 | ||
1da177e4 LT |
1118 | /* |
1119 | * Decode auth data, and add verifier to reply buffer. | |
1120 | * We do this before anything else in order to get a decent | |
1121 | * auth verifier. | |
1122 | */ | |
1123 | auth_res = svc_authenticate(rqstp, &auth_stat); | |
1124 | /* Also give the program a chance to reject this call: */ | |
80d188a6 | 1125 | if (auth_res == SVC_OK && progp) { |
1da177e4 LT |
1126 | auth_stat = rpc_autherr_badcred; |
1127 | auth_res = progp->pg_authenticate(rqstp); | |
1128 | } | |
1129 | switch (auth_res) { | |
1130 | case SVC_OK: | |
1131 | break; | |
1132 | case SVC_GARBAGE: | |
dd35210e | 1133 | goto err_garbage; |
1da177e4 LT |
1134 | case SVC_SYSERR: |
1135 | rpc_stat = rpc_system_err; | |
1136 | goto err_bad; | |
1137 | case SVC_DENIED: | |
1138 | goto err_bad_auth; | |
1ebede86 N |
1139 | case SVC_CLOSE: |
1140 | if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) | |
1141 | svc_close_xprt(rqstp->rq_xprt); | |
1da177e4 LT |
1142 | case SVC_DROP: |
1143 | goto dropit; | |
1144 | case SVC_COMPLETE: | |
1145 | goto sendit; | |
1146 | } | |
80d188a6 | 1147 | |
9ba02638 | 1148 | if (progp == NULL) |
1da177e4 LT |
1149 | goto err_bad_prog; |
1150 | ||
1151 | if (vers >= progp->pg_nvers || | |
1152 | !(versp = progp->pg_vers[vers])) | |
1153 | goto err_bad_vers; | |
1154 | ||
1155 | procp = versp->vs_proc + proc; | |
1156 | if (proc >= versp->vs_nproc || !procp->pc_func) | |
1157 | goto err_bad_proc; | |
1da177e4 LT |
1158 | rqstp->rq_procinfo = procp; |
1159 | ||
1160 | /* Syntactic check complete */ | |
1161 | serv->sv_stats->rpccnt++; | |
1162 | ||
1163 | /* Build the reply header. */ | |
1164 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 1165 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 LT |
1166 | |
1167 | /* Bump per-procedure stats counter */ | |
1168 | procp->pc_count++; | |
1169 | ||
1170 | /* Initialize storage for argp and resp */ | |
1171 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | |
1172 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | |
1173 | ||
cca5172a | 1174 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
1175 | * better idea of reply size |
1176 | */ | |
1177 | if (procp->pc_xdrressize) | |
cd123012 | 1178 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1da177e4 LT |
1179 | |
1180 | /* Call the function that processes the request. */ | |
1181 | if (!versp->vs_dispatch) { | |
1182 | /* Decode arguments */ | |
1183 | xdr = procp->pc_decode; | |
1184 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | |
1185 | goto err_garbage; | |
1186 | ||
1187 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
1188 | ||
1189 | /* Encode reply */ | |
9e701c61 | 1190 | if (rqstp->rq_dropme) { |
d343fce1 N |
1191 | if (procp->pc_release) |
1192 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1193 | goto dropit; | |
1194 | } | |
f64f9e71 JP |
1195 | if (*statp == rpc_success && |
1196 | (xdr = procp->pc_encode) && | |
1197 | !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | |
1da177e4 LT |
1198 | dprintk("svc: failed to encode reply\n"); |
1199 | /* serv->sv_stats->rpcsystemerr++; */ | |
1200 | *statp = rpc_system_err; | |
1201 | } | |
1202 | } else { | |
1203 | dprintk("svc: calling dispatcher\n"); | |
1204 | if (!versp->vs_dispatch(rqstp, statp)) { | |
1205 | /* Release reply info */ | |
1206 | if (procp->pc_release) | |
1207 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1208 | goto dropit; | |
1209 | } | |
1210 | } | |
1211 | ||
1212 | /* Check RPC status result */ | |
1213 | if (*statp != rpc_success) | |
1214 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
1215 | ||
1216 | /* Release reply info */ | |
1217 | if (procp->pc_release) | |
1218 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1219 | ||
1220 | if (procp->pc_encode == NULL) | |
1221 | goto dropit; | |
1222 | ||
1223 | sendit: | |
1224 | if (svc_authorise(rqstp)) | |
1225 | goto dropit; | |
1cad7ea6 | 1226 | return 1; /* Caller can now send it */ |
1da177e4 LT |
1227 | |
1228 | dropit: | |
1229 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
1230 | dprintk("svc: svc_process dropit\n"); | |
1da177e4 LT |
1231 | return 0; |
1232 | ||
1233 | err_short_len: | |
354ecbb9 DDAG |
1234 | svc_printk(rqstp, "short len %Zd, dropping request\n", |
1235 | argv->iov_len); | |
34e9a63b | 1236 | |
1da177e4 LT |
1237 | goto dropit; /* drop request */ |
1238 | ||
1da177e4 LT |
1239 | err_bad_rpc: |
1240 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
1241 | svc_putnl(resv, 1); /* REJECT */ |
1242 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
1243 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
1244 | svc_putnl(resv, 2); | |
1da177e4 LT |
1245 | goto sendit; |
1246 | ||
1247 | err_bad_auth: | |
1248 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | |
1249 | serv->sv_stats->rpcbadauth++; | |
1250 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 1251 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
1252 | svc_putnl(resv, 1); /* REJECT */ |
1253 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
1254 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | |
1da177e4 LT |
1255 | goto sendit; |
1256 | ||
1257 | err_bad_prog: | |
9ba02638 | 1258 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 1259 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1260 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
1261 | goto sendit; |
1262 | ||
1263 | err_bad_vers: | |
354ecbb9 | 1264 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
34e9a63b N |
1265 | vers, prog, progp->pg_name); |
1266 | ||
1da177e4 | 1267 | serv->sv_stats->rpcbadfmt++; |
76994313 AD |
1268 | svc_putnl(resv, RPC_PROG_MISMATCH); |
1269 | svc_putnl(resv, progp->pg_lovers); | |
1270 | svc_putnl(resv, progp->pg_hivers); | |
1da177e4 LT |
1271 | goto sendit; |
1272 | ||
1273 | err_bad_proc: | |
354ecbb9 | 1274 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); |
34e9a63b | 1275 | |
1da177e4 | 1276 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1277 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
1278 | goto sendit; |
1279 | ||
1280 | err_garbage: | |
354ecbb9 | 1281 | svc_printk(rqstp, "failed to decode args\n"); |
34e9a63b | 1282 | |
1da177e4 LT |
1283 | rpc_stat = rpc_garbage_args; |
1284 | err_bad: | |
1285 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 1286 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
1287 | goto sendit; |
1288 | } | |
24c3767e | 1289 | EXPORT_SYMBOL_GPL(svc_process); |
7adae489 | 1290 | |
1cad7ea6 RL |
1291 | /* |
1292 | * Process the RPC request. | |
1293 | */ | |
1294 | int | |
1295 | svc_process(struct svc_rqst *rqstp) | |
1296 | { | |
1297 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1298 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1299 | struct svc_serv *serv = rqstp->rq_server; | |
1300 | u32 dir; | |
1cad7ea6 RL |
1301 | |
1302 | /* | |
1303 | * Setup response xdr_buf. | |
1304 | * Initially it has just one page | |
1305 | */ | |
1306 | rqstp->rq_resused = 1; | |
1307 | resv->iov_base = page_address(rqstp->rq_respages[0]); | |
1308 | resv->iov_len = 0; | |
1309 | rqstp->rq_res.pages = rqstp->rq_respages + 1; | |
1310 | rqstp->rq_res.len = 0; | |
1311 | rqstp->rq_res.page_base = 0; | |
1312 | rqstp->rq_res.page_len = 0; | |
1313 | rqstp->rq_res.buflen = PAGE_SIZE; | |
1314 | rqstp->rq_res.tail[0].iov_base = NULL; | |
1315 | rqstp->rq_res.tail[0].iov_len = 0; | |
1316 | ||
1317 | rqstp->rq_xid = svc_getu32(argv); | |
1318 | ||
1319 | dir = svc_getnl(argv); | |
1320 | if (dir != 0) { | |
1321 | /* direction != CALL */ | |
1322 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); | |
1323 | serv->sv_stats->rpcbadfmt++; | |
1324 | svc_drop(rqstp); | |
1325 | return 0; | |
1326 | } | |
1327 | ||
4b5b3ba1 AA |
1328 | /* Returns 1 for send, 0 for drop */ |
1329 | if (svc_process_common(rqstp, argv, resv)) | |
1330 | return svc_send(rqstp); | |
1331 | else { | |
1332 | svc_drop(rqstp); | |
1333 | return 0; | |
1334 | } | |
1cad7ea6 RL |
1335 | } |
1336 | ||
9e00abc3 | 1337 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
4d6bbb62 RL |
1338 | /* |
1339 | * Process a backchannel RPC request that arrived over an existing | |
1340 | * outbound connection | |
1341 | */ | |
1342 | int | |
1343 | bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, | |
1344 | struct svc_rqst *rqstp) | |
1345 | { | |
1346 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1347 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
4d6bbb62 RL |
1348 | |
1349 | /* Build the svc_rqst used by the common processing routine */ | |
4a19de0f | 1350 | rqstp->rq_xprt = serv->sv_bc_xprt; |
4d6bbb62 RL |
1351 | rqstp->rq_xid = req->rq_xid; |
1352 | rqstp->rq_prot = req->rq_xprt->prot; | |
1353 | rqstp->rq_server = serv; | |
1354 | ||
1355 | rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); | |
1356 | memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); | |
1357 | memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); | |
1358 | memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); | |
1359 | ||
1360 | /* reset result send buffer "put" position */ | |
1361 | resv->iov_len = 0; | |
1362 | ||
1363 | if (rqstp->rq_prot != IPPROTO_TCP) { | |
1364 | printk(KERN_ERR "No support for Non-TCP transports!\n"); | |
1365 | BUG(); | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Skip the next two words because they've already been | |
1370 | * processed in the trasport | |
1371 | */ | |
1372 | svc_getu32(argv); /* XID */ | |
1373 | svc_getnl(argv); /* CALLDIR */ | |
1374 | ||
4b5b3ba1 AA |
1375 | /* Returns 1 for send, 0 for drop */ |
1376 | if (svc_process_common(rqstp, argv, resv)) { | |
1377 | memcpy(&req->rq_snd_buf, &rqstp->rq_res, | |
1378 | sizeof(req->rq_snd_buf)); | |
1379 | return bc_send(req); | |
1380 | } else { | |
b3b02ae5 TM |
1381 | /* drop request */ |
1382 | xprt_free_bc_request(req); | |
4b5b3ba1 AA |
1383 | return 0; |
1384 | } | |
4d6bbb62 | 1385 | } |
0d961aa9 | 1386 | EXPORT_SYMBOL_GPL(bc_svc_process); |
9e00abc3 | 1387 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
4d6bbb62 | 1388 | |
7adae489 GB |
1389 | /* |
1390 | * Return (transport-specific) limit on the rpc payload. | |
1391 | */ | |
1392 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
1393 | { | |
49023155 | 1394 | u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; |
7adae489 | 1395 | |
c6b0a9f8 N |
1396 | if (rqstp->rq_server->sv_max_payload < max) |
1397 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
1398 | return max; |
1399 | } | |
1400 | EXPORT_SYMBOL_GPL(svc_max_payload); |