]>
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 | 30 | |
860a0d9e JL |
31 | #include <trace/events/sunrpc.h> |
32 | ||
1da177e4 | 33 | #define RPCDBG_FACILITY RPCDBG_SVCDSP |
1da177e4 | 34 | |
5247fab5 | 35 | static void svc_unregister(const struct svc_serv *serv, struct net *net); |
7252d575 | 36 | |
c369014f | 37 | #define svc_serv_is_pooled(serv) ((serv)->sv_ops->svo_function) |
42a7fc4a | 38 | |
42a7fc4a | 39 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
bfd24160 GB |
40 | |
41 | /* | |
42 | * Structure for mapping cpus to pools and vice versa. | |
43 | * Setup once during sunrpc initialisation. | |
44 | */ | |
d70bc0c6 | 45 | struct svc_pool_map svc_pool_map = { |
42a7fc4a | 46 | .mode = SVC_POOL_DEFAULT |
bfd24160 | 47 | }; |
d70bc0c6 JL |
48 | EXPORT_SYMBOL_GPL(svc_pool_map); |
49 | ||
42a7fc4a GB |
50 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
51 | ||
52 | static int | |
53 | param_set_pool_mode(const char *val, struct kernel_param *kp) | |
54 | { | |
55 | int *ip = (int *)kp->arg; | |
56 | struct svc_pool_map *m = &svc_pool_map; | |
57 | int err; | |
58 | ||
59 | mutex_lock(&svc_pool_map_mutex); | |
60 | ||
61 | err = -EBUSY; | |
62 | if (m->count) | |
63 | goto out; | |
64 | ||
65 | err = 0; | |
66 | if (!strncmp(val, "auto", 4)) | |
67 | *ip = SVC_POOL_AUTO; | |
68 | else if (!strncmp(val, "global", 6)) | |
69 | *ip = SVC_POOL_GLOBAL; | |
70 | else if (!strncmp(val, "percpu", 6)) | |
71 | *ip = SVC_POOL_PERCPU; | |
72 | else if (!strncmp(val, "pernode", 7)) | |
73 | *ip = SVC_POOL_PERNODE; | |
74 | else | |
75 | err = -EINVAL; | |
76 | ||
77 | out: | |
78 | mutex_unlock(&svc_pool_map_mutex); | |
79 | return err; | |
80 | } | |
81 | ||
82 | static int | |
83 | param_get_pool_mode(char *buf, struct kernel_param *kp) | |
84 | { | |
85 | int *ip = (int *)kp->arg; | |
86 | ||
87 | switch (*ip) | |
88 | { | |
89 | case SVC_POOL_AUTO: | |
90 | return strlcpy(buf, "auto", 20); | |
91 | case SVC_POOL_GLOBAL: | |
92 | return strlcpy(buf, "global", 20); | |
93 | case SVC_POOL_PERCPU: | |
94 | return strlcpy(buf, "percpu", 20); | |
95 | case SVC_POOL_PERNODE: | |
96 | return strlcpy(buf, "pernode", 20); | |
97 | default: | |
98 | return sprintf(buf, "%d", *ip); | |
99 | } | |
100 | } | |
bfd24160 | 101 | |
42a7fc4a GB |
102 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
103 | &svc_pool_map.mode, 0644); | |
bfd24160 GB |
104 | |
105 | /* | |
106 | * Detect best pool mapping mode heuristically, | |
107 | * according to the machine's topology. | |
108 | */ | |
109 | static int | |
110 | svc_pool_map_choose_mode(void) | |
111 | { | |
112 | unsigned int node; | |
113 | ||
62bc62a8 | 114 | if (nr_online_nodes > 1) { |
bfd24160 GB |
115 | /* |
116 | * Actually have multiple NUMA nodes, | |
117 | * so split pools on NUMA node boundaries | |
118 | */ | |
119 | return SVC_POOL_PERNODE; | |
120 | } | |
121 | ||
72c33688 | 122 | node = first_online_node; |
bfd24160 GB |
123 | if (nr_cpus_node(node) > 2) { |
124 | /* | |
125 | * Non-trivial SMP, or CONFIG_NUMA on | |
126 | * non-NUMA hardware, e.g. with a generic | |
127 | * x86_64 kernel on Xeons. In this case we | |
128 | * want to divide the pools on cpu boundaries. | |
129 | */ | |
130 | return SVC_POOL_PERCPU; | |
131 | } | |
132 | ||
133 | /* default: one global pool */ | |
134 | return SVC_POOL_GLOBAL; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Allocate the to_pool[] and pool_to[] arrays. | |
139 | * Returns 0 on success or an errno. | |
140 | */ | |
141 | static int | |
142 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
143 | { | |
144 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
145 | if (!m->to_pool) | |
146 | goto fail; | |
147 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
148 | if (!m->pool_to) | |
149 | goto fail_free; | |
150 | ||
151 | return 0; | |
152 | ||
153 | fail_free: | |
154 | kfree(m->to_pool); | |
61c8504c | 155 | m->to_pool = NULL; |
bfd24160 GB |
156 | fail: |
157 | return -ENOMEM; | |
158 | } | |
159 | ||
160 | /* | |
161 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
162 | * Returns number of pools or <0 on error. | |
163 | */ | |
164 | static int | |
165 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
166 | { | |
53b8a315 | 167 | unsigned int maxpools = nr_cpu_ids; |
bfd24160 GB |
168 | unsigned int pidx = 0; |
169 | unsigned int cpu; | |
170 | int err; | |
171 | ||
172 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
173 | if (err) | |
174 | return err; | |
175 | ||
176 | for_each_online_cpu(cpu) { | |
eb63192b | 177 | BUG_ON(pidx >= maxpools); |
bfd24160 GB |
178 | m->to_pool[cpu] = pidx; |
179 | m->pool_to[pidx] = cpu; | |
180 | pidx++; | |
181 | } | |
182 | /* cpus brought online later all get mapped to pool0, sorry */ | |
183 | ||
184 | return pidx; | |
185 | }; | |
186 | ||
187 | ||
188 | /* | |
189 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
190 | * Returns number of pools or <0 on error. | |
191 | */ | |
192 | static int | |
193 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
194 | { | |
74c7aa8b | 195 | unsigned int maxpools = nr_node_ids; |
bfd24160 GB |
196 | unsigned int pidx = 0; |
197 | unsigned int node; | |
198 | int err; | |
199 | ||
200 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
201 | if (err) | |
202 | return err; | |
203 | ||
204 | for_each_node_with_cpus(node) { | |
205 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
206 | BUG_ON(pidx > maxpools); | |
207 | m->to_pool[node] = pidx; | |
208 | m->pool_to[pidx] = node; | |
209 | pidx++; | |
210 | } | |
211 | /* nodes brought online later all get mapped to pool0, sorry */ | |
212 | ||
213 | return pidx; | |
214 | } | |
215 | ||
216 | ||
217 | /* | |
42a7fc4a GB |
218 | * Add a reference to the global map of cpus to pools (and |
219 | * vice versa). Initialise the map if we're the first user. | |
220 | * Returns the number of pools. | |
bfd24160 | 221 | */ |
d70bc0c6 | 222 | unsigned int |
42a7fc4a | 223 | svc_pool_map_get(void) |
bfd24160 GB |
224 | { |
225 | struct svc_pool_map *m = &svc_pool_map; | |
226 | int npools = -1; | |
227 | ||
42a7fc4a GB |
228 | mutex_lock(&svc_pool_map_mutex); |
229 | ||
230 | if (m->count++) { | |
231 | mutex_unlock(&svc_pool_map_mutex); | |
bfd24160 | 232 | return m->npools; |
42a7fc4a | 233 | } |
bfd24160 | 234 | |
42a7fc4a GB |
235 | if (m->mode == SVC_POOL_AUTO) |
236 | m->mode = svc_pool_map_choose_mode(); | |
bfd24160 GB |
237 | |
238 | switch (m->mode) { | |
239 | case SVC_POOL_PERCPU: | |
240 | npools = svc_pool_map_init_percpu(m); | |
241 | break; | |
242 | case SVC_POOL_PERNODE: | |
243 | npools = svc_pool_map_init_pernode(m); | |
244 | break; | |
245 | } | |
246 | ||
247 | if (npools < 0) { | |
248 | /* default, or memory allocation failure */ | |
249 | npools = 1; | |
250 | m->mode = SVC_POOL_GLOBAL; | |
251 | } | |
252 | m->npools = npools; | |
253 | ||
42a7fc4a | 254 | mutex_unlock(&svc_pool_map_mutex); |
bfd24160 GB |
255 | return m->npools; |
256 | } | |
d70bc0c6 | 257 | EXPORT_SYMBOL_GPL(svc_pool_map_get); |
42a7fc4a GB |
258 | |
259 | /* | |
260 | * Drop a reference to the global map of cpus to pools. | |
261 | * When the last reference is dropped, the map data is | |
262 | * freed; this allows the sysadmin to change the pool | |
263 | * mode using the pool_mode module option without | |
264 | * rebooting or re-loading sunrpc.ko. | |
265 | */ | |
d70bc0c6 | 266 | void |
42a7fc4a GB |
267 | svc_pool_map_put(void) |
268 | { | |
269 | struct svc_pool_map *m = &svc_pool_map; | |
270 | ||
271 | mutex_lock(&svc_pool_map_mutex); | |
272 | ||
273 | if (!--m->count) { | |
42a7fc4a | 274 | kfree(m->to_pool); |
61c8504c | 275 | m->to_pool = NULL; |
42a7fc4a | 276 | kfree(m->pool_to); |
61c8504c | 277 | m->pool_to = NULL; |
42a7fc4a GB |
278 | m->npools = 0; |
279 | } | |
280 | ||
281 | mutex_unlock(&svc_pool_map_mutex); | |
282 | } | |
d70bc0c6 | 283 | EXPORT_SYMBOL_GPL(svc_pool_map_put); |
42a7fc4a | 284 | |
11fd165c ED |
285 | static int svc_pool_map_get_node(unsigned int pidx) |
286 | { | |
287 | const struct svc_pool_map *m = &svc_pool_map; | |
288 | ||
289 | if (m->count) { | |
290 | if (m->mode == SVC_POOL_PERCPU) | |
291 | return cpu_to_node(m->pool_to[pidx]); | |
292 | if (m->mode == SVC_POOL_PERNODE) | |
293 | return m->pool_to[pidx]; | |
294 | } | |
295 | return NUMA_NO_NODE; | |
296 | } | |
bfd24160 | 297 | /* |
9867d76c | 298 | * Set the given thread's cpus_allowed mask so that it |
bfd24160 | 299 | * will only run on cpus in the given pool. |
bfd24160 | 300 | */ |
9867d76c JL |
301 | static inline void |
302 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) | |
bfd24160 GB |
303 | { |
304 | struct svc_pool_map *m = &svc_pool_map; | |
9867d76c | 305 | unsigned int node = m->pool_to[pidx]; |
bfd24160 GB |
306 | |
307 | /* | |
308 | * The caller checks for sv_nrpools > 1, which | |
42a7fc4a | 309 | * implies that we've been initialized. |
bfd24160 | 310 | */ |
1bd58aaf WAA |
311 | WARN_ON_ONCE(m->count == 0); |
312 | if (m->count == 0) | |
313 | return; | |
bfd24160 | 314 | |
9867d76c | 315 | switch (m->mode) { |
bfd24160 | 316 | case SVC_POOL_PERCPU: |
c5f59f08 | 317 | { |
aa85ea5b | 318 | set_cpus_allowed_ptr(task, cpumask_of(node)); |
9867d76c | 319 | break; |
c5f59f08 | 320 | } |
bfd24160 | 321 | case SVC_POOL_PERNODE: |
c5f59f08 | 322 | { |
a70f7302 | 323 | set_cpus_allowed_ptr(task, cpumask_of_node(node)); |
9867d76c | 324 | break; |
bfd24160 | 325 | } |
c5f59f08 | 326 | } |
bfd24160 GB |
327 | } |
328 | ||
329 | /* | |
330 | * Use the mapping mode to choose a pool for a given CPU. | |
331 | * Used when enqueueing an incoming RPC. Always returns | |
332 | * a non-NULL pool pointer. | |
333 | */ | |
334 | struct svc_pool * | |
335 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | |
336 | { | |
337 | struct svc_pool_map *m = &svc_pool_map; | |
338 | unsigned int pidx = 0; | |
339 | ||
340 | /* | |
42a7fc4a | 341 | * An uninitialised map happens in a pure client when |
bfd24160 GB |
342 | * lockd is brought up, so silently treat it the |
343 | * same as SVC_POOL_GLOBAL. | |
344 | */ | |
42a7fc4a GB |
345 | if (svc_serv_is_pooled(serv)) { |
346 | switch (m->mode) { | |
347 | case SVC_POOL_PERCPU: | |
348 | pidx = m->to_pool[cpu]; | |
349 | break; | |
350 | case SVC_POOL_PERNODE: | |
351 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
352 | break; | |
353 | } | |
bfd24160 GB |
354 | } |
355 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | |
356 | } | |
357 | ||
bb2224df | 358 | int svc_rpcb_setup(struct svc_serv *serv, struct net *net) |
d9908560 SK |
359 | { |
360 | int err; | |
361 | ||
bee42f68 | 362 | err = rpcb_create_local(net); |
d9908560 SK |
363 | if (err) |
364 | return err; | |
365 | ||
366 | /* Remove any stale portmap registrations */ | |
bee42f68 | 367 | svc_unregister(serv, net); |
d9908560 SK |
368 | return 0; |
369 | } | |
bb2224df | 370 | EXPORT_SYMBOL_GPL(svc_rpcb_setup); |
d9908560 | 371 | |
5ecebb7c | 372 | void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) |
d9908560 | 373 | { |
5ecebb7c SK |
374 | svc_unregister(serv, net); |
375 | rpcb_put_local(net); | |
d9908560 | 376 | } |
16d05870 | 377 | EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); |
d9908560 SK |
378 | |
379 | static int svc_uses_rpcbind(struct svc_serv *serv) | |
380 | { | |
381 | struct svc_program *progp; | |
382 | unsigned int i; | |
383 | ||
384 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
385 | for (i = 0; i < progp->pg_nvers; i++) { | |
386 | if (progp->pg_vers[i] == NULL) | |
387 | continue; | |
388 | if (progp->pg_vers[i]->vs_hidden == 0) | |
389 | return 1; | |
390 | } | |
391 | } | |
392 | ||
393 | return 0; | |
394 | } | |
bfd24160 | 395 | |
9793f7c8 SK |
396 | int svc_bind(struct svc_serv *serv, struct net *net) |
397 | { | |
398 | if (!svc_uses_rpcbind(serv)) | |
399 | return 0; | |
400 | return svc_rpcb_setup(serv, net); | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(svc_bind); | |
403 | ||
1da177e4 LT |
404 | /* |
405 | * Create an RPC service | |
406 | */ | |
a7455442 GB |
407 | static struct svc_serv * |
408 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
ea126e74 | 409 | struct svc_serv_ops *ops) |
1da177e4 LT |
410 | { |
411 | struct svc_serv *serv; | |
ea339d46 | 412 | unsigned int vers; |
1da177e4 | 413 | unsigned int xdrsize; |
3262c816 | 414 | unsigned int i; |
1da177e4 | 415 | |
0da974f4 | 416 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 417 | return NULL; |
9ba02638 | 418 | serv->sv_name = prog->pg_name; |
1da177e4 LT |
419 | serv->sv_program = prog; |
420 | serv->sv_nrthreads = 1; | |
421 | serv->sv_stats = prog->pg_stats; | |
c6b0a9f8 N |
422 | if (bufsize > RPCSVC_MAXPAYLOAD) |
423 | bufsize = RPCSVC_MAXPAYLOAD; | |
424 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
425 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
ea126e74 | 426 | serv->sv_ops = ops; |
1da177e4 | 427 | xdrsize = 0; |
9ba02638 AG |
428 | while (prog) { |
429 | prog->pg_lovers = prog->pg_nvers-1; | |
430 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
431 | if (prog->pg_vers[vers]) { | |
432 | prog->pg_hivers = vers; | |
433 | if (prog->pg_lovers > vers) | |
434 | prog->pg_lovers = vers; | |
435 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
436 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
437 | } | |
438 | prog = prog->pg_next; | |
439 | } | |
1da177e4 | 440 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
441 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
442 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
36bdfc8b | 443 | init_timer(&serv->sv_temptimer); |
1da177e4 LT |
444 | spin_lock_init(&serv->sv_lock); |
445 | ||
a7455442 | 446 | serv->sv_nrpools = npools; |
3262c816 | 447 | serv->sv_pools = |
cd861280 | 448 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
449 | GFP_KERNEL); |
450 | if (!serv->sv_pools) { | |
451 | kfree(serv); | |
452 | return NULL; | |
453 | } | |
454 | ||
455 | for (i = 0; i < serv->sv_nrpools; i++) { | |
456 | struct svc_pool *pool = &serv->sv_pools[i]; | |
457 | ||
46121cf7 | 458 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
459 | i, serv->sv_name); |
460 | ||
461 | pool->sp_id = i; | |
3262c816 | 462 | INIT_LIST_HEAD(&pool->sp_sockets); |
a7455442 | 463 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
464 | spin_lock_init(&pool->sp_lock); |
465 | } | |
466 | ||
1da177e4 LT |
467 | return serv; |
468 | } | |
469 | ||
a7455442 GB |
470 | struct svc_serv * |
471 | svc_create(struct svc_program *prog, unsigned int bufsize, | |
ea126e74 | 472 | struct svc_serv_ops *ops) |
a7455442 | 473 | { |
ea126e74 | 474 | return __svc_create(prog, bufsize, /*npools*/1, ops); |
a7455442 | 475 | } |
24c3767e | 476 | EXPORT_SYMBOL_GPL(svc_create); |
a7455442 GB |
477 | |
478 | struct svc_serv * | |
479 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | |
758f62ff | 480 | struct svc_serv_ops *ops) |
a7455442 GB |
481 | { |
482 | struct svc_serv *serv; | |
42a7fc4a | 483 | unsigned int npools = svc_pool_map_get(); |
a7455442 | 484 | |
ea126e74 | 485 | serv = __svc_create(prog, bufsize, npools, ops); |
067f96ef JL |
486 | if (!serv) |
487 | goto out_err; | |
a7455442 | 488 | return serv; |
067f96ef JL |
489 | out_err: |
490 | svc_pool_map_put(); | |
491 | return NULL; | |
a7455442 | 492 | } |
24c3767e | 493 | EXPORT_SYMBOL_GPL(svc_create_pooled); |
a7455442 | 494 | |
074d0f67 SK |
495 | void svc_shutdown_net(struct svc_serv *serv, struct net *net) |
496 | { | |
074d0f67 SK |
497 | svc_close_net(serv, net); |
498 | ||
ea126e74 JL |
499 | if (serv->sv_ops->svo_shutdown) |
500 | serv->sv_ops->svo_shutdown(serv, net); | |
074d0f67 SK |
501 | } |
502 | EXPORT_SYMBOL_GPL(svc_shutdown_net); | |
503 | ||
1da177e4 | 504 | /* |
bedbdd8b NB |
505 | * Destroy an RPC service. Should be called with appropriate locking to |
506 | * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. | |
1da177e4 LT |
507 | */ |
508 | void | |
509 | svc_destroy(struct svc_serv *serv) | |
510 | { | |
46121cf7 | 511 | dprintk("svc: svc_destroy(%s, %d)\n", |
1da177e4 LT |
512 | serv->sv_program->pg_name, |
513 | serv->sv_nrthreads); | |
514 | ||
515 | if (serv->sv_nrthreads) { | |
516 | if (--(serv->sv_nrthreads) != 0) { | |
517 | svc_sock_update_bufs(serv); | |
518 | return; | |
519 | } | |
520 | } else | |
521 | printk("svc_destroy: no threads for serv=%p!\n", serv); | |
522 | ||
36bdfc8b | 523 | del_timer_sync(&serv->sv_temptimer); |
074d0f67 | 524 | |
7b147f1f SK |
525 | /* |
526 | * The last user is gone and thus all sockets have to be destroyed to | |
527 | * the point. Check this. | |
528 | */ | |
529 | BUG_ON(!list_empty(&serv->sv_permsocks)); | |
530 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | |
cda1fd4a | 531 | |
1da177e4 LT |
532 | cache_clean_deferred(serv); |
533 | ||
42a7fc4a GB |
534 | if (svc_serv_is_pooled(serv)) |
535 | svc_pool_map_put(); | |
536 | ||
3262c816 | 537 | kfree(serv->sv_pools); |
1da177e4 LT |
538 | kfree(serv); |
539 | } | |
24c3767e | 540 | EXPORT_SYMBOL_GPL(svc_destroy); |
1da177e4 LT |
541 | |
542 | /* | |
543 | * Allocate an RPC server's buffer space. | |
544 | * We allocate pages and place them in rq_argpages. | |
545 | */ | |
546 | static int | |
11fd165c | 547 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) |
1da177e4 | 548 | { |
0dc220f0 | 549 | unsigned int pages, arghi; |
cca5172a | 550 | |
ba17686f AA |
551 | /* bc_xprt uses fore channel allocated buffers */ |
552 | if (svc_is_backchannel(rqstp)) | |
553 | return 1; | |
554 | ||
c6b0a9f8 N |
555 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
556 | * We assume one is at most one page | |
557 | */ | |
1da177e4 | 558 | arghi = 0; |
b25cd058 WAA |
559 | WARN_ON_ONCE(pages > RPCSVC_MAXPAGES); |
560 | if (pages > RPCSVC_MAXPAGES) | |
561 | pages = RPCSVC_MAXPAGES; | |
1da177e4 | 562 | while (pages) { |
11fd165c | 563 | struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); |
1da177e4 LT |
564 | if (!p) |
565 | break; | |
44524359 | 566 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
567 | pages--; |
568 | } | |
0dc220f0 | 569 | return pages == 0; |
1da177e4 LT |
570 | } |
571 | ||
572 | /* | |
573 | * Release an RPC server buffer | |
574 | */ | |
575 | static void | |
576 | svc_release_buffer(struct svc_rqst *rqstp) | |
577 | { | |
50c8bb13 CL |
578 | unsigned int i; |
579 | ||
580 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) | |
44524359 N |
581 | if (rqstp->rq_pages[i]) |
582 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
583 | } |
584 | ||
0113ab34 | 585 | struct svc_rqst * |
1b6dc1df | 586 | svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node) |
1da177e4 LT |
587 | { |
588 | struct svc_rqst *rqstp; | |
1da177e4 | 589 | |
11fd165c | 590 | rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); |
1da177e4 | 591 | if (!rqstp) |
1b6dc1df | 592 | return rqstp; |
1da177e4 | 593 | |
b1691bc0 JL |
594 | __set_bit(RQ_BUSY, &rqstp->rq_flags); |
595 | spin_lock_init(&rqstp->rq_lock); | |
596 | rqstp->rq_server = serv; | |
597 | rqstp->rq_pool = pool; | |
bfd24160 | 598 | |
11fd165c | 599 | rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 | 600 | if (!rqstp->rq_argp) |
1b6dc1df | 601 | goto out_enomem; |
0113ab34 | 602 | |
11fd165c | 603 | rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 | 604 | if (!rqstp->rq_resp) |
1b6dc1df | 605 | goto out_enomem; |
0113ab34 | 606 | |
11fd165c | 607 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) |
1b6dc1df | 608 | goto out_enomem; |
0113ab34 JL |
609 | |
610 | return rqstp; | |
0113ab34 | 611 | out_enomem: |
1b6dc1df JL |
612 | svc_rqst_free(rqstp); |
613 | return NULL; | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(svc_rqst_alloc); | |
616 | ||
617 | struct svc_rqst * | |
618 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) | |
619 | { | |
620 | struct svc_rqst *rqstp; | |
621 | ||
622 | rqstp = svc_rqst_alloc(serv, pool, node); | |
623 | if (!rqstp) | |
624 | return ERR_PTR(-ENOMEM); | |
625 | ||
626 | serv->sv_nrthreads++; | |
627 | spin_lock_bh(&pool->sp_lock); | |
628 | pool->sp_nrthreads++; | |
629 | list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads); | |
630 | spin_unlock_bh(&pool->sp_lock); | |
631 | return rqstp; | |
0113ab34 | 632 | } |
24c3767e | 633 | EXPORT_SYMBOL_GPL(svc_prepare_thread); |
0113ab34 | 634 | |
a7455442 GB |
635 | /* |
636 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
637 | */ | |
638 | static inline struct svc_pool * | |
639 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
640 | { | |
641 | if (pool != NULL) | |
642 | return pool; | |
643 | ||
cca5172a | 644 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
645 | } |
646 | ||
647 | /* | |
648 | * Choose a thread to kill, for svc_set_num_threads | |
649 | */ | |
650 | static inline struct task_struct * | |
651 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
652 | { | |
653 | unsigned int i; | |
654 | struct task_struct *task = NULL; | |
655 | ||
656 | if (pool != NULL) { | |
657 | spin_lock_bh(&pool->sp_lock); | |
658 | } else { | |
659 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
660 | for (i = 0; i < serv->sv_nrpools; i++) { |
661 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 662 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
663 | if (!list_empty(&pool->sp_all_threads)) |
664 | goto found_pool; | |
a7455442 | 665 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 666 | } |
a7455442 GB |
667 | return NULL; |
668 | } | |
669 | ||
670 | found_pool: | |
671 | if (!list_empty(&pool->sp_all_threads)) { | |
672 | struct svc_rqst *rqstp; | |
673 | ||
674 | /* | |
675 | * Remove from the pool->sp_all_threads list | |
676 | * so we don't try to kill it again. | |
677 | */ | |
678 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
81244386 JL |
679 | set_bit(RQ_VICTIM, &rqstp->rq_flags); |
680 | list_del_rcu(&rqstp->rq_all); | |
a7455442 | 681 | task = rqstp->rq_task; |
cca5172a | 682 | } |
a7455442 GB |
683 | spin_unlock_bh(&pool->sp_lock); |
684 | ||
685 | return task; | |
686 | } | |
687 | ||
688 | /* | |
689 | * Create or destroy enough new threads to make the number | |
690 | * of threads the given number. If `pool' is non-NULL, applies | |
691 | * only to threads in that pool, otherwise round-robins between | |
94cf3179 BF |
692 | * all pools. Caller must ensure that mutual exclusion between this and |
693 | * server startup or shutdown. | |
a7455442 GB |
694 | * |
695 | * Destroying threads relies on the service threads filling in | |
696 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | |
697 | * has been created using svc_create_pooled(). | |
698 | * | |
699 | * Based on code that used to be in nfsd_svc() but tweaked | |
700 | * to be pool-aware. | |
701 | */ | |
702 | int | |
703 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
704 | { | |
9867d76c JL |
705 | struct svc_rqst *rqstp; |
706 | struct task_struct *task; | |
707 | struct svc_pool *chosen_pool; | |
a7455442 GB |
708 | int error = 0; |
709 | unsigned int state = serv->sv_nrthreads-1; | |
11fd165c | 710 | int node; |
a7455442 GB |
711 | |
712 | if (pool == NULL) { | |
713 | /* The -1 assumes caller has done a svc_get() */ | |
714 | nrservs -= (serv->sv_nrthreads-1); | |
715 | } else { | |
716 | spin_lock_bh(&pool->sp_lock); | |
717 | nrservs -= pool->sp_nrthreads; | |
718 | spin_unlock_bh(&pool->sp_lock); | |
719 | } | |
720 | ||
721 | /* create new threads */ | |
722 | while (nrservs > 0) { | |
723 | nrservs--; | |
9867d76c JL |
724 | chosen_pool = choose_pool(serv, pool, &state); |
725 | ||
11fd165c ED |
726 | node = svc_pool_map_get_node(chosen_pool->sp_id); |
727 | rqstp = svc_prepare_thread(serv, chosen_pool, node); | |
9867d76c JL |
728 | if (IS_ERR(rqstp)) { |
729 | error = PTR_ERR(rqstp); | |
730 | break; | |
731 | } | |
732 | ||
758f62ff | 733 | __module_get(serv->sv_ops->svo_module); |
c369014f | 734 | task = kthread_create_on_node(serv->sv_ops->svo_function, rqstp, |
f170168b | 735 | node, "%s", serv->sv_name); |
9867d76c JL |
736 | if (IS_ERR(task)) { |
737 | error = PTR_ERR(task); | |
758f62ff | 738 | module_put(serv->sv_ops->svo_module); |
9867d76c | 739 | svc_exit_thread(rqstp); |
a7455442 GB |
740 | break; |
741 | } | |
9867d76c JL |
742 | |
743 | rqstp->rq_task = task; | |
744 | if (serv->sv_nrpools > 1) | |
745 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); | |
746 | ||
747 | svc_sock_update_bufs(serv); | |
748 | wake_up_process(task); | |
a7455442 GB |
749 | } |
750 | /* destroy old threads */ | |
751 | while (nrservs < 0 && | |
9867d76c | 752 | (task = choose_victim(serv, pool, &state)) != NULL) { |
a75c5d01 | 753 | send_sig(SIGINT, task, 1); |
a7455442 GB |
754 | nrservs++; |
755 | } | |
756 | ||
757 | return error; | |
758 | } | |
24c3767e | 759 | EXPORT_SYMBOL_GPL(svc_set_num_threads); |
a7455442 | 760 | |
3262c816 | 761 | /* |
3c519914 JL |
762 | * Called from a server thread as it's exiting. Caller must hold the "service |
763 | * mutex" for the service. | |
1da177e4 LT |
764 | */ |
765 | void | |
1b6dc1df | 766 | svc_rqst_free(struct svc_rqst *rqstp) |
1da177e4 | 767 | { |
1da177e4 | 768 | svc_release_buffer(rqstp); |
a51482bd JJ |
769 | kfree(rqstp->rq_resp); |
770 | kfree(rqstp->rq_argp); | |
771 | kfree(rqstp->rq_auth_data); | |
1b6dc1df JL |
772 | kfree_rcu(rqstp, rq_rcu_head); |
773 | } | |
774 | EXPORT_SYMBOL_GPL(svc_rqst_free); | |
775 | ||
776 | void | |
777 | svc_exit_thread(struct svc_rqst *rqstp) | |
778 | { | |
779 | struct svc_serv *serv = rqstp->rq_server; | |
780 | struct svc_pool *pool = rqstp->rq_pool; | |
3262c816 GB |
781 | |
782 | spin_lock_bh(&pool->sp_lock); | |
783 | pool->sp_nrthreads--; | |
81244386 JL |
784 | if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags)) |
785 | list_del_rcu(&rqstp->rq_all); | |
3262c816 GB |
786 | spin_unlock_bh(&pool->sp_lock); |
787 | ||
1b6dc1df | 788 | svc_rqst_free(rqstp); |
1da177e4 LT |
789 | |
790 | /* Release the server */ | |
791 | if (serv) | |
792 | svc_destroy(serv); | |
793 | } | |
24c3767e | 794 | EXPORT_SYMBOL_GPL(svc_exit_thread); |
1da177e4 LT |
795 | |
796 | /* | |
2c7eb0b2 CL |
797 | * Register an "inet" protocol family netid with the local |
798 | * rpcbind daemon via an rpcbind v4 SET request. | |
a26cfad6 | 799 | * |
2c7eb0b2 CL |
800 | * No netconfig infrastructure is available in the kernel, so |
801 | * we map IP_ protocol numbers to netids by hand. | |
a26cfad6 | 802 | * |
2c7eb0b2 CL |
803 | * Returns zero on success; a negative errno value is returned |
804 | * if any error occurs. | |
1da177e4 | 805 | */ |
5247fab5 SK |
806 | static int __svc_rpcb_register4(struct net *net, const u32 program, |
807 | const u32 version, | |
2c7eb0b2 CL |
808 | const unsigned short protocol, |
809 | const unsigned short port) | |
a26cfad6 | 810 | { |
cadc0fa5 | 811 | const struct sockaddr_in sin = { |
a26cfad6 CL |
812 | .sin_family = AF_INET, |
813 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
814 | .sin_port = htons(port), | |
815 | }; | |
cadc0fa5 CL |
816 | const char *netid; |
817 | int error; | |
2c7eb0b2 CL |
818 | |
819 | switch (protocol) { | |
820 | case IPPROTO_UDP: | |
821 | netid = RPCBIND_NETID_UDP; | |
822 | break; | |
823 | case IPPROTO_TCP: | |
824 | netid = RPCBIND_NETID_TCP; | |
825 | break; | |
826 | default: | |
ba5c35e0 | 827 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
828 | } |
829 | ||
5247fab5 | 830 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
831 | (const struct sockaddr *)&sin, netid); |
832 | ||
833 | /* | |
834 | * User space didn't support rpcbind v4, so retry this | |
835 | * registration request with the legacy rpcbind v2 protocol. | |
836 | */ | |
837 | if (error == -EPROTONOSUPPORT) | |
5247fab5 | 838 | error = rpcb_register(net, program, version, protocol, port); |
cadc0fa5 CL |
839 | |
840 | return error; | |
2c7eb0b2 CL |
841 | } |
842 | ||
dfd56b8b | 843 | #if IS_ENABLED(CONFIG_IPV6) |
2c7eb0b2 CL |
844 | /* |
845 | * Register an "inet6" protocol family netid with the local | |
846 | * rpcbind daemon via an rpcbind v4 SET request. | |
847 | * | |
848 | * No netconfig infrastructure is available in the kernel, so | |
849 | * we map IP_ protocol numbers to netids by hand. | |
850 | * | |
851 | * Returns zero on success; a negative errno value is returned | |
852 | * if any error occurs. | |
853 | */ | |
5247fab5 SK |
854 | static int __svc_rpcb_register6(struct net *net, const u32 program, |
855 | const u32 version, | |
2c7eb0b2 CL |
856 | const unsigned short protocol, |
857 | const unsigned short port) | |
858 | { | |
cadc0fa5 | 859 | const struct sockaddr_in6 sin6 = { |
a26cfad6 CL |
860 | .sin6_family = AF_INET6, |
861 | .sin6_addr = IN6ADDR_ANY_INIT, | |
862 | .sin6_port = htons(port), | |
863 | }; | |
cadc0fa5 CL |
864 | const char *netid; |
865 | int error; | |
a26cfad6 | 866 | |
2c7eb0b2 CL |
867 | switch (protocol) { |
868 | case IPPROTO_UDP: | |
869 | netid = RPCBIND_NETID_UDP6; | |
a26cfad6 | 870 | break; |
2c7eb0b2 | 871 | case IPPROTO_TCP: |
a26cfad6 | 872 | netid = RPCBIND_NETID_TCP6; |
a26cfad6 CL |
873 | break; |
874 | default: | |
ba5c35e0 | 875 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
876 | } |
877 | ||
5247fab5 | 878 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
879 | (const struct sockaddr *)&sin6, netid); |
880 | ||
881 | /* | |
882 | * User space didn't support rpcbind version 4, so we won't | |
883 | * use a PF_INET6 listener. | |
884 | */ | |
885 | if (error == -EPROTONOSUPPORT) | |
886 | error = -EAFNOSUPPORT; | |
887 | ||
888 | return error; | |
2c7eb0b2 | 889 | } |
dfd56b8b | 890 | #endif /* IS_ENABLED(CONFIG_IPV6) */ |
2c7eb0b2 CL |
891 | |
892 | /* | |
893 | * Register a kernel RPC service via rpcbind version 4. | |
894 | * | |
895 | * Returns zero on success; a negative errno value is returned | |
896 | * if any error occurs. | |
897 | */ | |
5247fab5 | 898 | static int __svc_register(struct net *net, const char *progname, |
363f724c | 899 | const u32 program, const u32 version, |
4b62e58c | 900 | const int family, |
2c7eb0b2 CL |
901 | const unsigned short protocol, |
902 | const unsigned short port) | |
903 | { | |
363f724c | 904 | int error = -EAFNOSUPPORT; |
2c7eb0b2 CL |
905 | |
906 | switch (family) { | |
4b62e58c | 907 | case PF_INET: |
5247fab5 | 908 | error = __svc_rpcb_register4(net, program, version, |
2c7eb0b2 | 909 | protocol, port); |
cadc0fa5 | 910 | break; |
dfd56b8b | 911 | #if IS_ENABLED(CONFIG_IPV6) |
4b62e58c | 912 | case PF_INET6: |
5247fab5 | 913 | error = __svc_rpcb_register6(net, program, version, |
2c7eb0b2 | 914 | protocol, port); |
dfd56b8b | 915 | #endif |
a26cfad6 CL |
916 | } |
917 | ||
363f724c | 918 | return error; |
a26cfad6 | 919 | } |
2c7eb0b2 | 920 | |
a26cfad6 CL |
921 | /** |
922 | * svc_register - register an RPC service with the local portmapper | |
923 | * @serv: svc_serv struct for the service to register | |
5247fab5 | 924 | * @net: net namespace for the service to register |
4b62e58c | 925 | * @family: protocol family of service's listener socket |
a26cfad6 CL |
926 | * @proto: transport protocol number to advertise |
927 | * @port: port to advertise | |
928 | * | |
4b62e58c | 929 | * Service is registered for any address in the passed-in protocol family |
a26cfad6 | 930 | */ |
5247fab5 SK |
931 | int svc_register(const struct svc_serv *serv, struct net *net, |
932 | const int family, const unsigned short proto, | |
933 | const unsigned short port) | |
1da177e4 LT |
934 | { |
935 | struct svc_program *progp; | |
7e55b59b | 936 | struct svc_version *vers; |
ea339d46 | 937 | unsigned int i; |
14aeb211 | 938 | int error = 0; |
1da177e4 | 939 | |
0af39507 WAA |
940 | WARN_ON_ONCE(proto == 0 && port == 0); |
941 | if (proto == 0 && port == 0) | |
942 | return -EINVAL; | |
1da177e4 | 943 | |
bc5fea42 OK |
944 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
945 | for (i = 0; i < progp->pg_nvers; i++) { | |
7e55b59b KM |
946 | vers = progp->pg_vers[i]; |
947 | if (vers == NULL) | |
bc5fea42 OK |
948 | continue; |
949 | ||
2c7eb0b2 | 950 | dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", |
bc5fea42 | 951 | progp->pg_name, |
2c7eb0b2 | 952 | i, |
bc5fea42 OK |
953 | proto == IPPROTO_UDP? "udp" : "tcp", |
954 | port, | |
4b62e58c | 955 | family, |
7e55b59b KM |
956 | vers->vs_hidden ? |
957 | " (but not telling portmap)" : ""); | |
bc5fea42 | 958 | |
7e55b59b | 959 | if (vers->vs_hidden) |
bc5fea42 OK |
960 | continue; |
961 | ||
5247fab5 | 962 | error = __svc_register(net, progp->pg_name, progp->pg_prog, |
363f724c | 963 | i, family, proto, port); |
7e55b59b KM |
964 | |
965 | if (vers->vs_rpcb_optnl) { | |
966 | error = 0; | |
967 | continue; | |
968 | } | |
969 | ||
970 | if (error < 0) { | |
971 | printk(KERN_WARNING "svc: failed to register " | |
972 | "%sv%u RPC service (errno %d).\n", | |
973 | progp->pg_name, i, -error); | |
bc5fea42 | 974 | break; |
7e55b59b | 975 | } |
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 | 1041 | /* |
7032a3dd | 1042 | * dprintk the given error with the address of the client that caused it. |
354ecbb9 | 1043 | */ |
f895b252 | 1044 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
b9075fa9 | 1045 | static __printf(2, 3) |
e87cc472 | 1046 | void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) |
354ecbb9 | 1047 | { |
e87cc472 | 1048 | struct va_format vaf; |
354ecbb9 | 1049 | va_list args; |
354ecbb9 DDAG |
1050 | char buf[RPC_MAX_ADDRBUFLEN]; |
1051 | ||
e87cc472 | 1052 | va_start(args, fmt); |
354ecbb9 | 1053 | |
e87cc472 JP |
1054 | vaf.fmt = fmt; |
1055 | vaf.va = &args; | |
354ecbb9 | 1056 | |
7032a3dd | 1057 | dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); |
354ecbb9 | 1058 | |
e87cc472 | 1059 | va_end(args); |
354ecbb9 | 1060 | } |
624ab464 BF |
1061 | #else |
1062 | static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {} | |
1063 | #endif | |
354ecbb9 | 1064 | |
1da177e4 | 1065 | /* |
1cad7ea6 | 1066 | * Common routine for processing the RPC request. |
1da177e4 | 1067 | */ |
1cad7ea6 RL |
1068 | static int |
1069 | svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) | |
1da177e4 LT |
1070 | { |
1071 | struct svc_program *progp; | |
1072 | struct svc_version *versp = NULL; /* compiler food */ | |
1073 | struct svc_procedure *procp = NULL; | |
6fb2b47f | 1074 | struct svc_serv *serv = rqstp->rq_server; |
1da177e4 | 1075 | kxdrproc_t xdr; |
d8ed029d | 1076 | __be32 *statp; |
1cad7ea6 | 1077 | u32 prog, vers, proc; |
d8ed029d | 1078 | __be32 auth_stat, rpc_stat; |
1da177e4 | 1079 | int auth_res; |
8f8e05c5 | 1080 | __be32 *reply_statp; |
1da177e4 LT |
1081 | |
1082 | rpc_stat = rpc_success; | |
1083 | ||
1084 | if (argv->iov_len < 6*4) | |
1085 | goto err_short_len; | |
1086 | ||
5c04c46a | 1087 | /* Will be turned off only in gss privacy case: */ |
779fb0f3 | 1088 | set_bit(RQ_SPLICE_OK, &rqstp->rq_flags); |
2f425878 | 1089 | /* Will be turned off only when NFSv4 Sessions are used */ |
30660e04 | 1090 | set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags); |
78b65eb3 | 1091 | clear_bit(RQ_DROPME, &rqstp->rq_flags); |
e831fe65 TT |
1092 | |
1093 | /* Setup reply header */ | |
1094 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); | |
1da177e4 | 1095 | |
1da177e4 LT |
1096 | svc_putu32(resv, rqstp->rq_xid); |
1097 | ||
76994313 | 1098 | vers = svc_getnl(argv); |
1da177e4 LT |
1099 | |
1100 | /* First words of reply: */ | |
76994313 | 1101 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 | 1102 | |
1da177e4 LT |
1103 | if (vers != 2) /* RPC version number */ |
1104 | goto err_bad_rpc; | |
1105 | ||
1106 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 1107 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 1108 | |
76994313 | 1109 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 1110 | |
76994313 AD |
1111 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
1112 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | |
1113 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 | 1114 | |
80d188a6 N |
1115 | for (progp = serv->sv_program; progp; progp = progp->pg_next) |
1116 | if (prog == progp->pg_prog) | |
1117 | break; | |
1118 | ||
1da177e4 LT |
1119 | /* |
1120 | * Decode auth data, and add verifier to reply buffer. | |
1121 | * We do this before anything else in order to get a decent | |
1122 | * auth verifier. | |
1123 | */ | |
1124 | auth_res = svc_authenticate(rqstp, &auth_stat); | |
1125 | /* Also give the program a chance to reject this call: */ | |
80d188a6 | 1126 | if (auth_res == SVC_OK && progp) { |
1da177e4 LT |
1127 | auth_stat = rpc_autherr_badcred; |
1128 | auth_res = progp->pg_authenticate(rqstp); | |
1129 | } | |
1130 | switch (auth_res) { | |
1131 | case SVC_OK: | |
1132 | break; | |
1133 | case SVC_GARBAGE: | |
dd35210e | 1134 | goto err_garbage; |
1da177e4 LT |
1135 | case SVC_SYSERR: |
1136 | rpc_stat = rpc_system_err; | |
1137 | goto err_bad; | |
1138 | case SVC_DENIED: | |
1139 | goto err_bad_auth; | |
1ebede86 N |
1140 | case SVC_CLOSE: |
1141 | if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) | |
1142 | svc_close_xprt(rqstp->rq_xprt); | |
1da177e4 LT |
1143 | case SVC_DROP: |
1144 | goto dropit; | |
1145 | case SVC_COMPLETE: | |
1146 | goto sendit; | |
1147 | } | |
80d188a6 | 1148 | |
9ba02638 | 1149 | if (progp == NULL) |
1da177e4 LT |
1150 | goto err_bad_prog; |
1151 | ||
1152 | if (vers >= progp->pg_nvers || | |
1153 | !(versp = progp->pg_vers[vers])) | |
1154 | goto err_bad_vers; | |
1155 | ||
1156 | procp = versp->vs_proc + proc; | |
1157 | if (proc >= versp->vs_nproc || !procp->pc_func) | |
1158 | goto err_bad_proc; | |
1da177e4 LT |
1159 | rqstp->rq_procinfo = procp; |
1160 | ||
1161 | /* Syntactic check complete */ | |
1162 | serv->sv_stats->rpccnt++; | |
1163 | ||
1164 | /* Build the reply header. */ | |
1165 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 1166 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 LT |
1167 | |
1168 | /* Bump per-procedure stats counter */ | |
1169 | procp->pc_count++; | |
1170 | ||
1171 | /* Initialize storage for argp and resp */ | |
1172 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | |
1173 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | |
1174 | ||
cca5172a | 1175 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
1176 | * better idea of reply size |
1177 | */ | |
1178 | if (procp->pc_xdrressize) | |
cd123012 | 1179 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1da177e4 LT |
1180 | |
1181 | /* Call the function that processes the request. */ | |
1182 | if (!versp->vs_dispatch) { | |
1183 | /* Decode arguments */ | |
1184 | xdr = procp->pc_decode; | |
1185 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | |
1186 | goto err_garbage; | |
1187 | ||
1188 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
1189 | ||
1190 | /* Encode reply */ | |
78b65eb3 | 1191 | if (test_bit(RQ_DROPME, &rqstp->rq_flags)) { |
d343fce1 N |
1192 | if (procp->pc_release) |
1193 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1194 | goto dropit; | |
1195 | } | |
f64f9e71 JP |
1196 | if (*statp == rpc_success && |
1197 | (xdr = procp->pc_encode) && | |
1198 | !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | |
1da177e4 LT |
1199 | dprintk("svc: failed to encode reply\n"); |
1200 | /* serv->sv_stats->rpcsystemerr++; */ | |
1201 | *statp = rpc_system_err; | |
1202 | } | |
1203 | } else { | |
1204 | dprintk("svc: calling dispatcher\n"); | |
1205 | if (!versp->vs_dispatch(rqstp, statp)) { | |
1206 | /* Release reply info */ | |
1207 | if (procp->pc_release) | |
1208 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1209 | goto dropit; | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | /* Check RPC status result */ | |
1214 | if (*statp != rpc_success) | |
1215 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
1216 | ||
1217 | /* Release reply info */ | |
1218 | if (procp->pc_release) | |
1219 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | |
1220 | ||
1221 | if (procp->pc_encode == NULL) | |
1222 | goto dropit; | |
1223 | ||
1224 | sendit: | |
1225 | if (svc_authorise(rqstp)) | |
1226 | goto dropit; | |
1cad7ea6 | 1227 | return 1; /* Caller can now send it */ |
1da177e4 LT |
1228 | |
1229 | dropit: | |
1230 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
1231 | dprintk("svc: svc_process dropit\n"); | |
1da177e4 LT |
1232 | return 0; |
1233 | ||
1234 | err_short_len: | |
354ecbb9 DDAG |
1235 | svc_printk(rqstp, "short len %Zd, dropping request\n", |
1236 | argv->iov_len); | |
34e9a63b | 1237 | |
1da177e4 LT |
1238 | goto dropit; /* drop request */ |
1239 | ||
1da177e4 LT |
1240 | err_bad_rpc: |
1241 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
1242 | svc_putnl(resv, 1); /* REJECT */ |
1243 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
1244 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
1245 | svc_putnl(resv, 2); | |
1da177e4 LT |
1246 | goto sendit; |
1247 | ||
1248 | err_bad_auth: | |
1249 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | |
1250 | serv->sv_stats->rpcbadauth++; | |
1251 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 1252 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
1253 | svc_putnl(resv, 1); /* REJECT */ |
1254 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
1255 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | |
1da177e4 LT |
1256 | goto sendit; |
1257 | ||
1258 | err_bad_prog: | |
9ba02638 | 1259 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 1260 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1261 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
1262 | goto sendit; |
1263 | ||
1264 | err_bad_vers: | |
354ecbb9 | 1265 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
34e9a63b N |
1266 | vers, prog, progp->pg_name); |
1267 | ||
1da177e4 | 1268 | serv->sv_stats->rpcbadfmt++; |
76994313 AD |
1269 | svc_putnl(resv, RPC_PROG_MISMATCH); |
1270 | svc_putnl(resv, progp->pg_lovers); | |
1271 | svc_putnl(resv, progp->pg_hivers); | |
1da177e4 LT |
1272 | goto sendit; |
1273 | ||
1274 | err_bad_proc: | |
354ecbb9 | 1275 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); |
34e9a63b | 1276 | |
1da177e4 | 1277 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1278 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
1279 | goto sendit; |
1280 | ||
1281 | err_garbage: | |
354ecbb9 | 1282 | svc_printk(rqstp, "failed to decode args\n"); |
34e9a63b | 1283 | |
1da177e4 LT |
1284 | rpc_stat = rpc_garbage_args; |
1285 | err_bad: | |
1286 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 1287 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
1288 | goto sendit; |
1289 | } | |
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 | */ | |
afc59400 | 1306 | rqstp->rq_next_page = &rqstp->rq_respages[1]; |
1cad7ea6 RL |
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 | ||
1cad7ea6 RL |
1317 | dir = svc_getnl(argv); |
1318 | if (dir != 0) { | |
1319 | /* direction != CALL */ | |
1320 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); | |
1321 | serv->sv_stats->rpcbadfmt++; | |
860a0d9e | 1322 | goto out_drop; |
1cad7ea6 RL |
1323 | } |
1324 | ||
4b5b3ba1 | 1325 | /* Returns 1 for send, 0 for drop */ |
860a0d9e JL |
1326 | if (likely(svc_process_common(rqstp, argv, resv))) { |
1327 | int ret = svc_send(rqstp); | |
1328 | ||
1329 | trace_svc_process(rqstp, ret); | |
1330 | return ret; | |
4b5b3ba1 | 1331 | } |
860a0d9e JL |
1332 | out_drop: |
1333 | trace_svc_process(rqstp, 0); | |
1334 | svc_drop(rqstp); | |
1335 | return 0; | |
1cad7ea6 | 1336 | } |
3f87d5d6 | 1337 | EXPORT_SYMBOL_GPL(svc_process); |
1cad7ea6 | 1338 | |
9e00abc3 | 1339 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
4d6bbb62 RL |
1340 | /* |
1341 | * Process a backchannel RPC request that arrived over an existing | |
1342 | * outbound connection | |
1343 | */ | |
1344 | int | |
1345 | bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, | |
1346 | struct svc_rqst *rqstp) | |
1347 | { | |
1348 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1349 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
632dda83 | 1350 | struct rpc_task *task; |
0d2a970d | 1351 | int proc_error; |
632dda83 CL |
1352 | int error; |
1353 | ||
1354 | dprintk("svc: %s(%p)\n", __func__, req); | |
4d6bbb62 RL |
1355 | |
1356 | /* Build the svc_rqst used by the common processing routine */ | |
4a19de0f | 1357 | rqstp->rq_xprt = serv->sv_bc_xprt; |
4d6bbb62 RL |
1358 | rqstp->rq_xid = req->rq_xid; |
1359 | rqstp->rq_prot = req->rq_xprt->prot; | |
1360 | rqstp->rq_server = serv; | |
1361 | ||
1362 | rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); | |
1363 | memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); | |
1364 | memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); | |
1365 | memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); | |
756b9b37 TM |
1366 | |
1367 | /* Adjust the argument buffer length */ | |
38b7631f | 1368 | rqstp->rq_arg.len = req->rq_private_buf.len; |
756b9b37 TM |
1369 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1370 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1371 | rqstp->rq_arg.page_len = 0; | |
1372 | } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len + | |
1373 | rqstp->rq_arg.page_len) | |
1374 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - | |
1375 | rqstp->rq_arg.head[0].iov_len; | |
1376 | else | |
1377 | rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len + | |
1378 | rqstp->rq_arg.page_len; | |
4d6bbb62 RL |
1379 | |
1380 | /* reset result send buffer "put" position */ | |
1381 | resv->iov_len = 0; | |
1382 | ||
4d6bbb62 RL |
1383 | /* |
1384 | * Skip the next two words because they've already been | |
632dda83 | 1385 | * processed in the transport |
4d6bbb62 RL |
1386 | */ |
1387 | svc_getu32(argv); /* XID */ | |
1388 | svc_getnl(argv); /* CALLDIR */ | |
1389 | ||
632dda83 | 1390 | /* Parse and execute the bc call */ |
0d2a970d TM |
1391 | proc_error = svc_process_common(rqstp, argv, resv); |
1392 | ||
1393 | atomic_inc(&req->rq_xprt->bc_free_slots); | |
1394 | if (!proc_error) { | |
632dda83 | 1395 | /* Processing error: drop the request */ |
b3b02ae5 | 1396 | xprt_free_bc_request(req); |
4b5b3ba1 AA |
1397 | return 0; |
1398 | } | |
632dda83 CL |
1399 | |
1400 | /* Finally, send the reply synchronously */ | |
1401 | memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf)); | |
0f419791 | 1402 | task = rpc_run_bc_task(req); |
632dda83 CL |
1403 | if (IS_ERR(task)) { |
1404 | error = PTR_ERR(task); | |
1405 | goto out; | |
1406 | } | |
1407 | ||
1408 | WARN_ON_ONCE(atomic_read(&task->tk_count) != 1); | |
1409 | error = task->tk_status; | |
1410 | rpc_put_task(task); | |
1411 | ||
1412 | out: | |
1413 | dprintk("svc: %s(), error=%d\n", __func__, error); | |
1414 | return error; | |
4d6bbb62 | 1415 | } |
0d961aa9 | 1416 | EXPORT_SYMBOL_GPL(bc_svc_process); |
9e00abc3 | 1417 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
4d6bbb62 | 1418 | |
7adae489 GB |
1419 | /* |
1420 | * Return (transport-specific) limit on the rpc payload. | |
1421 | */ | |
1422 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
1423 | { | |
49023155 | 1424 | u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; |
7adae489 | 1425 | |
c6b0a9f8 N |
1426 | if (rqstp->rq_server->sv_max_payload < max) |
1427 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
1428 | return max; |
1429 | } | |
1430 | EXPORT_SYMBOL_GPL(svc_max_payload); |