Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/net/sunrpc/svc.c | |
4 | * | |
5 | * High-level RPC service routines | |
6 | * | |
7 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
bfd24160 GB |
8 | * |
9 | * Multiple threads pools and NUMAisation | |
10 | * Copyright (c) 2006 Silicon Graphics, Inc. | |
11 | * by Greg Banks <gnb@melbourne.sgi.com> | |
1da177e4 LT |
12 | */ |
13 | ||
14 | #include <linux/linkage.h> | |
3f07c014 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/errno.h> |
17 | #include <linux/net.h> | |
18 | #include <linux/in.h> | |
19 | #include <linux/mm.h> | |
a7455442 GB |
20 | #include <linux/interrupt.h> |
21 | #include <linux/module.h> | |
9867d76c | 22 | #include <linux/kthread.h> |
5a0e3ad6 | 23 | #include <linux/slab.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 | 31 | |
860a0d9e JL |
32 | #include <trace/events/sunrpc.h> |
33 | ||
3a126180 CL |
34 | #include "fail.h" |
35 | ||
1da177e4 | 36 | #define RPCDBG_FACILITY RPCDBG_SVCDSP |
1da177e4 | 37 | |
5247fab5 | 38 | static void svc_unregister(const struct svc_serv *serv, struct net *net); |
7252d575 | 39 | |
42a7fc4a | 40 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
bfd24160 | 41 | |
cf0e124e N |
42 | /* |
43 | * Mode for mapping cpus to pools. | |
44 | */ | |
45 | enum { | |
46 | SVC_POOL_AUTO = -1, /* choose one of the others */ | |
47 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool | |
48 | * (legacy & UP mode) */ | |
49 | SVC_POOL_PERCPU, /* one pool per cpu */ | |
50 | SVC_POOL_PERNODE /* one pool per numa node */ | |
51 | }; | |
52 | ||
bfd24160 GB |
53 | /* |
54 | * Structure for mapping cpus to pools and vice versa. | |
55 | * Setup once during sunrpc initialisation. | |
56 | */ | |
cf0e124e N |
57 | |
58 | struct svc_pool_map { | |
59 | int count; /* How many svc_servs use us */ | |
60 | int mode; /* Note: int not enum to avoid | |
61 | * warnings about "enumeration value | |
62 | * not handled in switch" */ | |
63 | unsigned int npools; | |
64 | unsigned int *pool_to; /* maps pool id to cpu or node */ | |
65 | unsigned int *to_pool; /* maps cpu or node to pool id */ | |
66 | }; | |
67 | ||
68 | static struct svc_pool_map svc_pool_map = { | |
42a7fc4a | 69 | .mode = SVC_POOL_DEFAULT |
bfd24160 | 70 | }; |
d70bc0c6 | 71 | |
42a7fc4a GB |
72 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
73 | ||
74 | static int | |
e4dca7b7 | 75 | param_set_pool_mode(const char *val, const struct kernel_param *kp) |
42a7fc4a GB |
76 | { |
77 | int *ip = (int *)kp->arg; | |
78 | struct svc_pool_map *m = &svc_pool_map; | |
79 | int err; | |
80 | ||
81 | mutex_lock(&svc_pool_map_mutex); | |
82 | ||
83 | err = -EBUSY; | |
84 | if (m->count) | |
85 | goto out; | |
86 | ||
87 | err = 0; | |
88 | if (!strncmp(val, "auto", 4)) | |
89 | *ip = SVC_POOL_AUTO; | |
90 | else if (!strncmp(val, "global", 6)) | |
91 | *ip = SVC_POOL_GLOBAL; | |
92 | else if (!strncmp(val, "percpu", 6)) | |
93 | *ip = SVC_POOL_PERCPU; | |
94 | else if (!strncmp(val, "pernode", 7)) | |
95 | *ip = SVC_POOL_PERNODE; | |
96 | else | |
97 | err = -EINVAL; | |
98 | ||
99 | out: | |
100 | mutex_unlock(&svc_pool_map_mutex); | |
101 | return err; | |
102 | } | |
103 | ||
104 | static int | |
e4dca7b7 | 105 | param_get_pool_mode(char *buf, const struct kernel_param *kp) |
42a7fc4a GB |
106 | { |
107 | int *ip = (int *)kp->arg; | |
108 | ||
109 | switch (*ip) | |
110 | { | |
111 | case SVC_POOL_AUTO: | |
746c6237 | 112 | return strlcpy(buf, "auto\n", 20); |
42a7fc4a | 113 | case SVC_POOL_GLOBAL: |
746c6237 | 114 | return strlcpy(buf, "global\n", 20); |
42a7fc4a | 115 | case SVC_POOL_PERCPU: |
746c6237 | 116 | return strlcpy(buf, "percpu\n", 20); |
42a7fc4a | 117 | case SVC_POOL_PERNODE: |
746c6237 | 118 | return strlcpy(buf, "pernode\n", 20); |
42a7fc4a | 119 | default: |
746c6237 | 120 | return sprintf(buf, "%d\n", *ip); |
42a7fc4a GB |
121 | } |
122 | } | |
bfd24160 | 123 | |
42a7fc4a GB |
124 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
125 | &svc_pool_map.mode, 0644); | |
bfd24160 GB |
126 | |
127 | /* | |
128 | * Detect best pool mapping mode heuristically, | |
129 | * according to the machine's topology. | |
130 | */ | |
131 | static int | |
132 | svc_pool_map_choose_mode(void) | |
133 | { | |
134 | unsigned int node; | |
135 | ||
62bc62a8 | 136 | if (nr_online_nodes > 1) { |
bfd24160 GB |
137 | /* |
138 | * Actually have multiple NUMA nodes, | |
139 | * so split pools on NUMA node boundaries | |
140 | */ | |
141 | return SVC_POOL_PERNODE; | |
142 | } | |
143 | ||
72c33688 | 144 | node = first_online_node; |
bfd24160 GB |
145 | if (nr_cpus_node(node) > 2) { |
146 | /* | |
147 | * Non-trivial SMP, or CONFIG_NUMA on | |
148 | * non-NUMA hardware, e.g. with a generic | |
149 | * x86_64 kernel on Xeons. In this case we | |
150 | * want to divide the pools on cpu boundaries. | |
151 | */ | |
152 | return SVC_POOL_PERCPU; | |
153 | } | |
154 | ||
155 | /* default: one global pool */ | |
156 | return SVC_POOL_GLOBAL; | |
157 | } | |
158 | ||
159 | /* | |
160 | * Allocate the to_pool[] and pool_to[] arrays. | |
161 | * Returns 0 on success or an errno. | |
162 | */ | |
163 | static int | |
164 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | |
165 | { | |
166 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
167 | if (!m->to_pool) | |
168 | goto fail; | |
169 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | |
170 | if (!m->pool_to) | |
171 | goto fail_free; | |
172 | ||
173 | return 0; | |
174 | ||
175 | fail_free: | |
176 | kfree(m->to_pool); | |
61c8504c | 177 | m->to_pool = NULL; |
bfd24160 GB |
178 | fail: |
179 | return -ENOMEM; | |
180 | } | |
181 | ||
182 | /* | |
183 | * Initialise the pool map for SVC_POOL_PERCPU mode. | |
184 | * Returns number of pools or <0 on error. | |
185 | */ | |
186 | static int | |
187 | svc_pool_map_init_percpu(struct svc_pool_map *m) | |
188 | { | |
53b8a315 | 189 | unsigned int maxpools = nr_cpu_ids; |
bfd24160 GB |
190 | unsigned int pidx = 0; |
191 | unsigned int cpu; | |
192 | int err; | |
193 | ||
194 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
195 | if (err) | |
196 | return err; | |
197 | ||
198 | for_each_online_cpu(cpu) { | |
eb63192b | 199 | BUG_ON(pidx >= maxpools); |
bfd24160 GB |
200 | m->to_pool[cpu] = pidx; |
201 | m->pool_to[pidx] = cpu; | |
202 | pidx++; | |
203 | } | |
204 | /* cpus brought online later all get mapped to pool0, sorry */ | |
205 | ||
206 | return pidx; | |
207 | }; | |
208 | ||
209 | ||
210 | /* | |
211 | * Initialise the pool map for SVC_POOL_PERNODE mode. | |
212 | * Returns number of pools or <0 on error. | |
213 | */ | |
214 | static int | |
215 | svc_pool_map_init_pernode(struct svc_pool_map *m) | |
216 | { | |
74c7aa8b | 217 | unsigned int maxpools = nr_node_ids; |
bfd24160 GB |
218 | unsigned int pidx = 0; |
219 | unsigned int node; | |
220 | int err; | |
221 | ||
222 | err = svc_pool_map_alloc_arrays(m, maxpools); | |
223 | if (err) | |
224 | return err; | |
225 | ||
226 | for_each_node_with_cpus(node) { | |
227 | /* some architectures (e.g. SN2) have cpuless nodes */ | |
228 | BUG_ON(pidx > maxpools); | |
229 | m->to_pool[node] = pidx; | |
230 | m->pool_to[pidx] = node; | |
231 | pidx++; | |
232 | } | |
233 | /* nodes brought online later all get mapped to pool0, sorry */ | |
234 | ||
235 | return pidx; | |
236 | } | |
237 | ||
238 | ||
239 | /* | |
42a7fc4a | 240 | * Add a reference to the global map of cpus to pools (and |
93aa619e N |
241 | * vice versa) if pools are in use. |
242 | * Initialise the map if we're the first user. | |
243 | * Returns the number of pools. If this is '1', no reference | |
244 | * was taken. | |
bfd24160 | 245 | */ |
cf0e124e | 246 | static unsigned int |
42a7fc4a | 247 | svc_pool_map_get(void) |
bfd24160 GB |
248 | { |
249 | struct svc_pool_map *m = &svc_pool_map; | |
250 | int npools = -1; | |
251 | ||
42a7fc4a GB |
252 | mutex_lock(&svc_pool_map_mutex); |
253 | ||
254 | if (m->count++) { | |
255 | mutex_unlock(&svc_pool_map_mutex); | |
93aa619e | 256 | WARN_ON_ONCE(m->npools <= 1); |
bfd24160 | 257 | return m->npools; |
42a7fc4a | 258 | } |
bfd24160 | 259 | |
42a7fc4a GB |
260 | if (m->mode == SVC_POOL_AUTO) |
261 | m->mode = svc_pool_map_choose_mode(); | |
bfd24160 GB |
262 | |
263 | switch (m->mode) { | |
264 | case SVC_POOL_PERCPU: | |
265 | npools = svc_pool_map_init_percpu(m); | |
266 | break; | |
267 | case SVC_POOL_PERNODE: | |
268 | npools = svc_pool_map_init_pernode(m); | |
269 | break; | |
270 | } | |
271 | ||
93aa619e | 272 | if (npools <= 0) { |
bfd24160 GB |
273 | /* default, or memory allocation failure */ |
274 | npools = 1; | |
275 | m->mode = SVC_POOL_GLOBAL; | |
276 | } | |
277 | m->npools = npools; | |
278 | ||
93aa619e N |
279 | if (npools == 1) |
280 | /* service is unpooled, so doesn't hold a reference */ | |
281 | m->count--; | |
282 | ||
42a7fc4a | 283 | mutex_unlock(&svc_pool_map_mutex); |
93aa619e | 284 | return npools; |
bfd24160 | 285 | } |
42a7fc4a GB |
286 | |
287 | /* | |
93aa619e N |
288 | * Drop a reference to the global map of cpus to pools, if |
289 | * pools were in use, i.e. if npools > 1. | |
42a7fc4a GB |
290 | * When the last reference is dropped, the map data is |
291 | * freed; this allows the sysadmin to change the pool | |
292 | * mode using the pool_mode module option without | |
293 | * rebooting or re-loading sunrpc.ko. | |
294 | */ | |
cf0e124e | 295 | static void |
93aa619e | 296 | svc_pool_map_put(int npools) |
42a7fc4a GB |
297 | { |
298 | struct svc_pool_map *m = &svc_pool_map; | |
299 | ||
93aa619e N |
300 | if (npools <= 1) |
301 | return; | |
42a7fc4a GB |
302 | mutex_lock(&svc_pool_map_mutex); |
303 | ||
304 | if (!--m->count) { | |
42a7fc4a | 305 | kfree(m->to_pool); |
61c8504c | 306 | m->to_pool = NULL; |
42a7fc4a | 307 | kfree(m->pool_to); |
61c8504c | 308 | m->pool_to = NULL; |
42a7fc4a GB |
309 | m->npools = 0; |
310 | } | |
311 | ||
312 | mutex_unlock(&svc_pool_map_mutex); | |
313 | } | |
42a7fc4a | 314 | |
11fd165c ED |
315 | static int svc_pool_map_get_node(unsigned int pidx) |
316 | { | |
317 | const struct svc_pool_map *m = &svc_pool_map; | |
318 | ||
319 | if (m->count) { | |
320 | if (m->mode == SVC_POOL_PERCPU) | |
321 | return cpu_to_node(m->pool_to[pidx]); | |
322 | if (m->mode == SVC_POOL_PERNODE) | |
323 | return m->pool_to[pidx]; | |
324 | } | |
325 | return NUMA_NO_NODE; | |
326 | } | |
bfd24160 | 327 | /* |
9867d76c | 328 | * Set the given thread's cpus_allowed mask so that it |
bfd24160 | 329 | * will only run on cpus in the given pool. |
bfd24160 | 330 | */ |
9867d76c JL |
331 | static inline void |
332 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) | |
bfd24160 GB |
333 | { |
334 | struct svc_pool_map *m = &svc_pool_map; | |
9867d76c | 335 | unsigned int node = m->pool_to[pidx]; |
bfd24160 GB |
336 | |
337 | /* | |
338 | * The caller checks for sv_nrpools > 1, which | |
42a7fc4a | 339 | * implies that we've been initialized. |
bfd24160 | 340 | */ |
1bd58aaf WAA |
341 | WARN_ON_ONCE(m->count == 0); |
342 | if (m->count == 0) | |
343 | return; | |
bfd24160 | 344 | |
9867d76c | 345 | switch (m->mode) { |
bfd24160 | 346 | case SVC_POOL_PERCPU: |
c5f59f08 | 347 | { |
aa85ea5b | 348 | set_cpus_allowed_ptr(task, cpumask_of(node)); |
9867d76c | 349 | break; |
c5f59f08 | 350 | } |
bfd24160 | 351 | case SVC_POOL_PERNODE: |
c5f59f08 | 352 | { |
a70f7302 | 353 | set_cpus_allowed_ptr(task, cpumask_of_node(node)); |
9867d76c | 354 | break; |
bfd24160 | 355 | } |
c5f59f08 | 356 | } |
bfd24160 GB |
357 | } |
358 | ||
2059b698 CL |
359 | /** |
360 | * svc_pool_for_cpu - Select pool to run a thread on this cpu | |
361 | * @serv: An RPC service | |
362 | * | |
363 | * Use the active CPU and the svc_pool_map's mode setting to | |
364 | * select the svc thread pool to use. Once initialized, the | |
365 | * svc_pool_map does not change. | |
366 | * | |
367 | * Return value: | |
368 | * A pointer to an svc_pool | |
bfd24160 | 369 | */ |
2059b698 | 370 | struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv) |
bfd24160 GB |
371 | { |
372 | struct svc_pool_map *m = &svc_pool_map; | |
2059b698 | 373 | int cpu = raw_smp_processor_id(); |
bfd24160 GB |
374 | unsigned int pidx = 0; |
375 | ||
93aa619e N |
376 | if (serv->sv_nrpools <= 1) |
377 | return serv->sv_pools; | |
378 | ||
379 | switch (m->mode) { | |
380 | case SVC_POOL_PERCPU: | |
381 | pidx = m->to_pool[cpu]; | |
382 | break; | |
383 | case SVC_POOL_PERNODE: | |
384 | pidx = m->to_pool[cpu_to_node(cpu)]; | |
385 | break; | |
bfd24160 | 386 | } |
93aa619e | 387 | |
bfd24160 GB |
388 | return &serv->sv_pools[pidx % serv->sv_nrpools]; |
389 | } | |
390 | ||
bb2224df | 391 | int svc_rpcb_setup(struct svc_serv *serv, struct net *net) |
d9908560 SK |
392 | { |
393 | int err; | |
394 | ||
bee42f68 | 395 | err = rpcb_create_local(net); |
d9908560 SK |
396 | if (err) |
397 | return err; | |
398 | ||
399 | /* Remove any stale portmap registrations */ | |
bee42f68 | 400 | svc_unregister(serv, net); |
d9908560 SK |
401 | return 0; |
402 | } | |
bb2224df | 403 | EXPORT_SYMBOL_GPL(svc_rpcb_setup); |
d9908560 | 404 | |
5ecebb7c | 405 | void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) |
d9908560 | 406 | { |
5ecebb7c SK |
407 | svc_unregister(serv, net); |
408 | rpcb_put_local(net); | |
d9908560 | 409 | } |
16d05870 | 410 | EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); |
d9908560 SK |
411 | |
412 | static int svc_uses_rpcbind(struct svc_serv *serv) | |
413 | { | |
414 | struct svc_program *progp; | |
415 | unsigned int i; | |
416 | ||
417 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
418 | for (i = 0; i < progp->pg_nvers; i++) { | |
419 | if (progp->pg_vers[i] == NULL) | |
420 | continue; | |
05a45a2d | 421 | if (!progp->pg_vers[i]->vs_hidden) |
d9908560 SK |
422 | return 1; |
423 | } | |
424 | } | |
425 | ||
426 | return 0; | |
427 | } | |
bfd24160 | 428 | |
9793f7c8 SK |
429 | int svc_bind(struct svc_serv *serv, struct net *net) |
430 | { | |
431 | if (!svc_uses_rpcbind(serv)) | |
432 | return 0; | |
433 | return svc_rpcb_setup(serv, net); | |
434 | } | |
435 | EXPORT_SYMBOL_GPL(svc_bind); | |
436 | ||
d0025268 TM |
437 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
438 | static void | |
439 | __svc_init_bc(struct svc_serv *serv) | |
440 | { | |
441 | INIT_LIST_HEAD(&serv->sv_cb_list); | |
442 | spin_lock_init(&serv->sv_cb_lock); | |
443 | init_waitqueue_head(&serv->sv_cb_waitq); | |
444 | } | |
445 | #else | |
446 | static void | |
447 | __svc_init_bc(struct svc_serv *serv) | |
448 | { | |
449 | } | |
450 | #endif | |
451 | ||
1da177e4 LT |
452 | /* |
453 | * Create an RPC service | |
454 | */ | |
a7455442 GB |
455 | static struct svc_serv * |
456 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | |
37902c63 | 457 | int (*threadfn)(void *data)) |
1da177e4 LT |
458 | { |
459 | struct svc_serv *serv; | |
ea339d46 | 460 | unsigned int vers; |
1da177e4 | 461 | unsigned int xdrsize; |
3262c816 | 462 | unsigned int i; |
1da177e4 | 463 | |
0da974f4 | 464 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
1da177e4 | 465 | return NULL; |
9ba02638 | 466 | serv->sv_name = prog->pg_name; |
1da177e4 | 467 | serv->sv_program = prog; |
ec52361d | 468 | kref_init(&serv->sv_refcnt); |
1da177e4 | 469 | serv->sv_stats = prog->pg_stats; |
c6b0a9f8 N |
470 | if (bufsize > RPCSVC_MAXPAYLOAD) |
471 | bufsize = RPCSVC_MAXPAYLOAD; | |
472 | serv->sv_max_payload = bufsize? bufsize : 4096; | |
473 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | |
37902c63 | 474 | serv->sv_threadfn = threadfn; |
1da177e4 | 475 | xdrsize = 0; |
9ba02638 AG |
476 | while (prog) { |
477 | prog->pg_lovers = prog->pg_nvers-1; | |
478 | for (vers=0; vers<prog->pg_nvers ; vers++) | |
479 | if (prog->pg_vers[vers]) { | |
480 | prog->pg_hivers = vers; | |
481 | if (prog->pg_lovers > vers) | |
482 | prog->pg_lovers = vers; | |
483 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | |
484 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | |
485 | } | |
486 | prog = prog->pg_next; | |
487 | } | |
1da177e4 | 488 | serv->sv_xdrsize = xdrsize; |
1da177e4 LT |
489 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
490 | INIT_LIST_HEAD(&serv->sv_permsocks); | |
ff861c4d | 491 | timer_setup(&serv->sv_temptimer, NULL, 0); |
1da177e4 LT |
492 | spin_lock_init(&serv->sv_lock); |
493 | ||
d0025268 TM |
494 | __svc_init_bc(serv); |
495 | ||
a7455442 | 496 | serv->sv_nrpools = npools; |
3262c816 | 497 | serv->sv_pools = |
cd861280 | 498 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
3262c816 GB |
499 | GFP_KERNEL); |
500 | if (!serv->sv_pools) { | |
501 | kfree(serv); | |
502 | return NULL; | |
503 | } | |
504 | ||
505 | for (i = 0; i < serv->sv_nrpools; i++) { | |
506 | struct svc_pool *pool = &serv->sv_pools[i]; | |
507 | ||
46121cf7 | 508 | dprintk("svc: initialising pool %u for %s\n", |
3262c816 GB |
509 | i, serv->sv_name); |
510 | ||
511 | pool->sp_id = i; | |
3262c816 | 512 | INIT_LIST_HEAD(&pool->sp_sockets); |
a7455442 | 513 | INIT_LIST_HEAD(&pool->sp_all_threads); |
3262c816 GB |
514 | spin_lock_init(&pool->sp_lock); |
515 | } | |
516 | ||
1da177e4 LT |
517 | return serv; |
518 | } | |
519 | ||
37902c63 CL |
520 | /** |
521 | * svc_create - Create an RPC service | |
522 | * @prog: the RPC program the new service will handle | |
523 | * @bufsize: maximum message size for @prog | |
524 | * @threadfn: a function to service RPC requests for @prog | |
525 | * | |
526 | * Returns an instantiated struct svc_serv object or NULL. | |
527 | */ | |
528 | struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize, | |
529 | int (*threadfn)(void *data)) | |
a7455442 | 530 | { |
37902c63 | 531 | return __svc_create(prog, bufsize, 1, threadfn); |
a7455442 | 532 | } |
24c3767e | 533 | EXPORT_SYMBOL_GPL(svc_create); |
a7455442 | 534 | |
37902c63 CL |
535 | /** |
536 | * svc_create_pooled - Create an RPC service with pooled threads | |
537 | * @prog: the RPC program the new service will handle | |
538 | * @bufsize: maximum message size for @prog | |
539 | * @threadfn: a function to service RPC requests for @prog | |
540 | * | |
541 | * Returns an instantiated struct svc_serv object or NULL. | |
542 | */ | |
543 | struct svc_serv *svc_create_pooled(struct svc_program *prog, | |
544 | unsigned int bufsize, | |
545 | int (*threadfn)(void *data)) | |
a7455442 GB |
546 | { |
547 | struct svc_serv *serv; | |
42a7fc4a | 548 | unsigned int npools = svc_pool_map_get(); |
a7455442 | 549 | |
37902c63 | 550 | serv = __svc_create(prog, bufsize, npools, threadfn); |
067f96ef JL |
551 | if (!serv) |
552 | goto out_err; | |
a7455442 | 553 | return serv; |
067f96ef | 554 | out_err: |
93aa619e | 555 | svc_pool_map_put(npools); |
067f96ef | 556 | return NULL; |
a7455442 | 557 | } |
24c3767e | 558 | EXPORT_SYMBOL_GPL(svc_create_pooled); |
a7455442 | 559 | |
1da177e4 | 560 | /* |
bedbdd8b | 561 | * Destroy an RPC service. Should be called with appropriate locking to |
2a36395f | 562 | * protect sv_permsocks and sv_tempsocks. |
1da177e4 LT |
563 | */ |
564 | void | |
ec52361d | 565 | svc_destroy(struct kref *ref) |
1da177e4 | 566 | { |
ec52361d | 567 | struct svc_serv *serv = container_of(ref, struct svc_serv, sv_refcnt); |
1da177e4 | 568 | |
ec52361d | 569 | dprintk("svc: svc_destroy(%s)\n", serv->sv_program->pg_name); |
36bdfc8b | 570 | del_timer_sync(&serv->sv_temptimer); |
074d0f67 | 571 | |
7b147f1f SK |
572 | /* |
573 | * The last user is gone and thus all sockets have to be destroyed to | |
574 | * the point. Check this. | |
575 | */ | |
576 | BUG_ON(!list_empty(&serv->sv_permsocks)); | |
577 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | |
cda1fd4a | 578 | |
1da177e4 LT |
579 | cache_clean_deferred(serv); |
580 | ||
93aa619e | 581 | svc_pool_map_put(serv->sv_nrpools); |
42a7fc4a | 582 | |
3262c816 | 583 | kfree(serv->sv_pools); |
1da177e4 LT |
584 | kfree(serv); |
585 | } | |
24c3767e | 586 | EXPORT_SYMBOL_GPL(svc_destroy); |
1da177e4 LT |
587 | |
588 | /* | |
589 | * Allocate an RPC server's buffer space. | |
4ff923ce | 590 | * We allocate pages and place them in rq_pages. |
1da177e4 LT |
591 | */ |
592 | static int | |
11fd165c | 593 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) |
1da177e4 | 594 | { |
0dc220f0 | 595 | unsigned int pages, arghi; |
cca5172a | 596 | |
ba17686f AA |
597 | /* bc_xprt uses fore channel allocated buffers */ |
598 | if (svc_is_backchannel(rqstp)) | |
599 | return 1; | |
600 | ||
c6b0a9f8 N |
601 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
602 | * We assume one is at most one page | |
603 | */ | |
1da177e4 | 604 | arghi = 0; |
b25cd058 WAA |
605 | WARN_ON_ONCE(pages > RPCSVC_MAXPAGES); |
606 | if (pages > RPCSVC_MAXPAGES) | |
607 | pages = RPCSVC_MAXPAGES; | |
1da177e4 | 608 | while (pages) { |
11fd165c | 609 | struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); |
1da177e4 LT |
610 | if (!p) |
611 | break; | |
44524359 | 612 | rqstp->rq_pages[arghi++] = p; |
1da177e4 LT |
613 | pages--; |
614 | } | |
0dc220f0 | 615 | return pages == 0; |
1da177e4 LT |
616 | } |
617 | ||
618 | /* | |
619 | * Release an RPC server buffer | |
620 | */ | |
621 | static void | |
622 | svc_release_buffer(struct svc_rqst *rqstp) | |
623 | { | |
50c8bb13 CL |
624 | unsigned int i; |
625 | ||
626 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) | |
44524359 N |
627 | if (rqstp->rq_pages[i]) |
628 | put_page(rqstp->rq_pages[i]); | |
1da177e4 LT |
629 | } |
630 | ||
0113ab34 | 631 | struct svc_rqst * |
1b6dc1df | 632 | svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node) |
1da177e4 LT |
633 | { |
634 | struct svc_rqst *rqstp; | |
1da177e4 | 635 | |
11fd165c | 636 | rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); |
1da177e4 | 637 | if (!rqstp) |
1b6dc1df | 638 | return rqstp; |
1da177e4 | 639 | |
b1691bc0 JL |
640 | __set_bit(RQ_BUSY, &rqstp->rq_flags); |
641 | spin_lock_init(&rqstp->rq_lock); | |
642 | rqstp->rq_server = serv; | |
643 | rqstp->rq_pool = pool; | |
bfd24160 | 644 | |
5191955d CL |
645 | rqstp->rq_scratch_page = alloc_pages_node(node, GFP_KERNEL, 0); |
646 | if (!rqstp->rq_scratch_page) | |
647 | goto out_enomem; | |
648 | ||
11fd165c | 649 | rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 | 650 | if (!rqstp->rq_argp) |
1b6dc1df | 651 | goto out_enomem; |
0113ab34 | 652 | |
11fd165c | 653 | rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); |
0113ab34 | 654 | if (!rqstp->rq_resp) |
1b6dc1df | 655 | goto out_enomem; |
0113ab34 | 656 | |
11fd165c | 657 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) |
1b6dc1df | 658 | goto out_enomem; |
0113ab34 JL |
659 | |
660 | return rqstp; | |
0113ab34 | 661 | out_enomem: |
1b6dc1df JL |
662 | svc_rqst_free(rqstp); |
663 | return NULL; | |
664 | } | |
665 | EXPORT_SYMBOL_GPL(svc_rqst_alloc); | |
666 | ||
6b044fba | 667 | static struct svc_rqst * |
1b6dc1df JL |
668 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) |
669 | { | |
670 | struct svc_rqst *rqstp; | |
671 | ||
672 | rqstp = svc_rqst_alloc(serv, pool, node); | |
673 | if (!rqstp) | |
674 | return ERR_PTR(-ENOMEM); | |
675 | ||
ec52361d | 676 | svc_get(serv); |
2a36395f N |
677 | spin_lock_bh(&serv->sv_lock); |
678 | serv->sv_nrthreads += 1; | |
679 | spin_unlock_bh(&serv->sv_lock); | |
680 | ||
1b6dc1df JL |
681 | spin_lock_bh(&pool->sp_lock); |
682 | pool->sp_nrthreads++; | |
683 | list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads); | |
684 | spin_unlock_bh(&pool->sp_lock); | |
685 | return rqstp; | |
0113ab34 | 686 | } |
0113ab34 | 687 | |
a7455442 GB |
688 | /* |
689 | * Choose a pool in which to create a new thread, for svc_set_num_threads | |
690 | */ | |
691 | static inline struct svc_pool * | |
692 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
693 | { | |
694 | if (pool != NULL) | |
695 | return pool; | |
696 | ||
cca5172a | 697 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
a7455442 GB |
698 | } |
699 | ||
700 | /* | |
701 | * Choose a thread to kill, for svc_set_num_threads | |
702 | */ | |
703 | static inline struct task_struct * | |
704 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | |
705 | { | |
706 | unsigned int i; | |
707 | struct task_struct *task = NULL; | |
708 | ||
709 | if (pool != NULL) { | |
710 | spin_lock_bh(&pool->sp_lock); | |
711 | } else { | |
712 | /* choose a pool in round-robin fashion */ | |
cca5172a YH |
713 | for (i = 0; i < serv->sv_nrpools; i++) { |
714 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | |
a7455442 | 715 | spin_lock_bh(&pool->sp_lock); |
cca5172a YH |
716 | if (!list_empty(&pool->sp_all_threads)) |
717 | goto found_pool; | |
a7455442 | 718 | spin_unlock_bh(&pool->sp_lock); |
cca5172a | 719 | } |
a7455442 GB |
720 | return NULL; |
721 | } | |
722 | ||
723 | found_pool: | |
724 | if (!list_empty(&pool->sp_all_threads)) { | |
725 | struct svc_rqst *rqstp; | |
726 | ||
727 | /* | |
728 | * Remove from the pool->sp_all_threads list | |
729 | * so we don't try to kill it again. | |
730 | */ | |
731 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | |
81244386 JL |
732 | set_bit(RQ_VICTIM, &rqstp->rq_flags); |
733 | list_del_rcu(&rqstp->rq_all); | |
a7455442 | 734 | task = rqstp->rq_task; |
cca5172a | 735 | } |
a7455442 GB |
736 | spin_unlock_bh(&pool->sp_lock); |
737 | ||
738 | return task; | |
739 | } | |
740 | ||
9e0d8768 TM |
741 | /* create new threads */ |
742 | static int | |
743 | svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
a7455442 | 744 | { |
9867d76c JL |
745 | struct svc_rqst *rqstp; |
746 | struct task_struct *task; | |
747 | struct svc_pool *chosen_pool; | |
a7455442 | 748 | unsigned int state = serv->sv_nrthreads-1; |
11fd165c | 749 | int node; |
a7455442 | 750 | |
9e0d8768 | 751 | do { |
a7455442 | 752 | nrservs--; |
9867d76c JL |
753 | chosen_pool = choose_pool(serv, pool, &state); |
754 | ||
11fd165c ED |
755 | node = svc_pool_map_get_node(chosen_pool->sp_id); |
756 | rqstp = svc_prepare_thread(serv, chosen_pool, node); | |
9e0d8768 TM |
757 | if (IS_ERR(rqstp)) |
758 | return PTR_ERR(rqstp); | |
9867d76c | 759 | |
37902c63 | 760 | task = kthread_create_on_node(serv->sv_threadfn, rqstp, |
f170168b | 761 | node, "%s", serv->sv_name); |
9867d76c | 762 | if (IS_ERR(task)) { |
9867d76c | 763 | svc_exit_thread(rqstp); |
9e0d8768 | 764 | return PTR_ERR(task); |
a7455442 | 765 | } |
9867d76c JL |
766 | |
767 | rqstp->rq_task = task; | |
768 | if (serv->sv_nrpools > 1) | |
769 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); | |
770 | ||
771 | svc_sock_update_bufs(serv); | |
772 | wake_up_process(task); | |
9e0d8768 TM |
773 | } while (nrservs > 0); |
774 | ||
775 | return 0; | |
776 | } | |
777 | ||
9e0d8768 TM |
778 | /* |
779 | * Create or destroy enough new threads to make the number | |
780 | * of threads the given number. If `pool' is non-NULL, applies | |
781 | * only to threads in that pool, otherwise round-robins between | |
782 | * all pools. Caller must ensure that mutual exclusion between this and | |
783 | * server startup or shutdown. | |
9e0d8768 | 784 | */ |
a7455442 | 785 | |
ed6473dd TM |
786 | /* destroy old threads */ |
787 | static int | |
788 | svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | |
789 | { | |
790 | struct task_struct *task; | |
791 | unsigned int state = serv->sv_nrthreads-1; | |
792 | ||
793 | /* destroy old threads */ | |
794 | do { | |
795 | task = choose_victim(serv, pool, &state); | |
796 | if (task == NULL) | |
797 | break; | |
798 | kthread_stop(task); | |
799 | nrservs++; | |
800 | } while (nrservs < 0); | |
801 | return 0; | |
802 | } | |
803 | ||
804 | int | |
3ebdbe52 | 805 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
ed6473dd TM |
806 | { |
807 | if (pool == NULL) { | |
ec52361d | 808 | nrservs -= serv->sv_nrthreads; |
ed6473dd TM |
809 | } else { |
810 | spin_lock_bh(&pool->sp_lock); | |
811 | nrservs -= pool->sp_nrthreads; | |
812 | spin_unlock_bh(&pool->sp_lock); | |
813 | } | |
814 | ||
815 | if (nrservs > 0) | |
816 | return svc_start_kthreads(serv, pool, nrservs); | |
817 | if (nrservs < 0) | |
818 | return svc_stop_kthreads(serv, pool, nrservs); | |
819 | return 0; | |
820 | } | |
3ebdbe52 | 821 | EXPORT_SYMBOL_GPL(svc_set_num_threads); |
ed6473dd | 822 | |
2f0f88f4 CL |
823 | /** |
824 | * svc_rqst_replace_page - Replace one page in rq_pages[] | |
825 | * @rqstp: svc_rqst with pages to replace | |
826 | * @page: replacement page | |
827 | * | |
828 | * When replacing a page in rq_pages, batch the release of the | |
829 | * replaced pages to avoid hammering the page allocator. | |
830 | */ | |
831 | void svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page) | |
832 | { | |
833 | if (*rqstp->rq_next_page) { | |
834 | if (!pagevec_space(&rqstp->rq_pvec)) | |
835 | __pagevec_release(&rqstp->rq_pvec); | |
836 | pagevec_add(&rqstp->rq_pvec, *rqstp->rq_next_page); | |
837 | } | |
838 | ||
839 | get_page(page); | |
840 | *(rqstp->rq_next_page++) = page; | |
841 | } | |
842 | EXPORT_SYMBOL_GPL(svc_rqst_replace_page); | |
843 | ||
3262c816 | 844 | /* |
3c519914 JL |
845 | * Called from a server thread as it's exiting. Caller must hold the "service |
846 | * mutex" for the service. | |
1da177e4 LT |
847 | */ |
848 | void | |
1b6dc1df | 849 | svc_rqst_free(struct svc_rqst *rqstp) |
1da177e4 | 850 | { |
1da177e4 | 851 | svc_release_buffer(rqstp); |
b9f83ffa YW |
852 | if (rqstp->rq_scratch_page) |
853 | put_page(rqstp->rq_scratch_page); | |
a51482bd JJ |
854 | kfree(rqstp->rq_resp); |
855 | kfree(rqstp->rq_argp); | |
856 | kfree(rqstp->rq_auth_data); | |
1b6dc1df JL |
857 | kfree_rcu(rqstp, rq_rcu_head); |
858 | } | |
859 | EXPORT_SYMBOL_GPL(svc_rqst_free); | |
860 | ||
861 | void | |
862 | svc_exit_thread(struct svc_rqst *rqstp) | |
863 | { | |
864 | struct svc_serv *serv = rqstp->rq_server; | |
865 | struct svc_pool *pool = rqstp->rq_pool; | |
3262c816 GB |
866 | |
867 | spin_lock_bh(&pool->sp_lock); | |
868 | pool->sp_nrthreads--; | |
81244386 JL |
869 | if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags)) |
870 | list_del_rcu(&rqstp->rq_all); | |
3262c816 GB |
871 | spin_unlock_bh(&pool->sp_lock); |
872 | ||
2a36395f | 873 | spin_lock_bh(&serv->sv_lock); |
ec52361d | 874 | serv->sv_nrthreads -= 1; |
2a36395f | 875 | spin_unlock_bh(&serv->sv_lock); |
ec52361d N |
876 | svc_sock_update_bufs(serv); |
877 | ||
1b6dc1df | 878 | svc_rqst_free(rqstp); |
1da177e4 | 879 | |
ec52361d | 880 | svc_put(serv); |
1da177e4 | 881 | } |
24c3767e | 882 | EXPORT_SYMBOL_GPL(svc_exit_thread); |
1da177e4 LT |
883 | |
884 | /* | |
2c7eb0b2 CL |
885 | * Register an "inet" protocol family netid with the local |
886 | * rpcbind daemon via an rpcbind v4 SET request. | |
a26cfad6 | 887 | * |
2c7eb0b2 CL |
888 | * No netconfig infrastructure is available in the kernel, so |
889 | * we map IP_ protocol numbers to netids by hand. | |
a26cfad6 | 890 | * |
2c7eb0b2 CL |
891 | * Returns zero on success; a negative errno value is returned |
892 | * if any error occurs. | |
1da177e4 | 893 | */ |
5247fab5 SK |
894 | static int __svc_rpcb_register4(struct net *net, const u32 program, |
895 | const u32 version, | |
2c7eb0b2 CL |
896 | const unsigned short protocol, |
897 | const unsigned short port) | |
a26cfad6 | 898 | { |
cadc0fa5 | 899 | const struct sockaddr_in sin = { |
a26cfad6 CL |
900 | .sin_family = AF_INET, |
901 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
902 | .sin_port = htons(port), | |
903 | }; | |
cadc0fa5 CL |
904 | const char *netid; |
905 | int error; | |
2c7eb0b2 CL |
906 | |
907 | switch (protocol) { | |
908 | case IPPROTO_UDP: | |
909 | netid = RPCBIND_NETID_UDP; | |
910 | break; | |
911 | case IPPROTO_TCP: | |
912 | netid = RPCBIND_NETID_TCP; | |
913 | break; | |
914 | default: | |
ba5c35e0 | 915 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
916 | } |
917 | ||
5247fab5 | 918 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
919 | (const struct sockaddr *)&sin, netid); |
920 | ||
921 | /* | |
922 | * User space didn't support rpcbind v4, so retry this | |
923 | * registration request with the legacy rpcbind v2 protocol. | |
924 | */ | |
925 | if (error == -EPROTONOSUPPORT) | |
5247fab5 | 926 | error = rpcb_register(net, program, version, protocol, port); |
cadc0fa5 CL |
927 | |
928 | return error; | |
2c7eb0b2 CL |
929 | } |
930 | ||
dfd56b8b | 931 | #if IS_ENABLED(CONFIG_IPV6) |
2c7eb0b2 CL |
932 | /* |
933 | * Register an "inet6" protocol family netid with the local | |
934 | * rpcbind daemon via an rpcbind v4 SET request. | |
935 | * | |
936 | * No netconfig infrastructure is available in the kernel, so | |
937 | * we map IP_ protocol numbers to netids by hand. | |
938 | * | |
939 | * Returns zero on success; a negative errno value is returned | |
940 | * if any error occurs. | |
941 | */ | |
5247fab5 SK |
942 | static int __svc_rpcb_register6(struct net *net, const u32 program, |
943 | const u32 version, | |
2c7eb0b2 CL |
944 | const unsigned short protocol, |
945 | const unsigned short port) | |
946 | { | |
cadc0fa5 | 947 | const struct sockaddr_in6 sin6 = { |
a26cfad6 CL |
948 | .sin6_family = AF_INET6, |
949 | .sin6_addr = IN6ADDR_ANY_INIT, | |
950 | .sin6_port = htons(port), | |
951 | }; | |
cadc0fa5 CL |
952 | const char *netid; |
953 | int error; | |
a26cfad6 | 954 | |
2c7eb0b2 CL |
955 | switch (protocol) { |
956 | case IPPROTO_UDP: | |
957 | netid = RPCBIND_NETID_UDP6; | |
a26cfad6 | 958 | break; |
2c7eb0b2 | 959 | case IPPROTO_TCP: |
a26cfad6 | 960 | netid = RPCBIND_NETID_TCP6; |
a26cfad6 CL |
961 | break; |
962 | default: | |
ba5c35e0 | 963 | return -ENOPROTOOPT; |
2c7eb0b2 CL |
964 | } |
965 | ||
5247fab5 | 966 | error = rpcb_v4_register(net, program, version, |
cadc0fa5 CL |
967 | (const struct sockaddr *)&sin6, netid); |
968 | ||
969 | /* | |
970 | * User space didn't support rpcbind version 4, so we won't | |
971 | * use a PF_INET6 listener. | |
972 | */ | |
973 | if (error == -EPROTONOSUPPORT) | |
974 | error = -EAFNOSUPPORT; | |
975 | ||
976 | return error; | |
2c7eb0b2 | 977 | } |
dfd56b8b | 978 | #endif /* IS_ENABLED(CONFIG_IPV6) */ |
2c7eb0b2 CL |
979 | |
980 | /* | |
981 | * Register a kernel RPC service via rpcbind version 4. | |
982 | * | |
983 | * Returns zero on success; a negative errno value is returned | |
984 | * if any error occurs. | |
985 | */ | |
5247fab5 | 986 | static int __svc_register(struct net *net, const char *progname, |
363f724c | 987 | const u32 program, const u32 version, |
4b62e58c | 988 | const int family, |
2c7eb0b2 CL |
989 | const unsigned short protocol, |
990 | const unsigned short port) | |
991 | { | |
363f724c | 992 | int error = -EAFNOSUPPORT; |
2c7eb0b2 CL |
993 | |
994 | switch (family) { | |
4b62e58c | 995 | case PF_INET: |
5247fab5 | 996 | error = __svc_rpcb_register4(net, program, version, |
2c7eb0b2 | 997 | protocol, port); |
cadc0fa5 | 998 | break; |
dfd56b8b | 999 | #if IS_ENABLED(CONFIG_IPV6) |
4b62e58c | 1000 | case PF_INET6: |
5247fab5 | 1001 | error = __svc_rpcb_register6(net, program, version, |
2c7eb0b2 | 1002 | protocol, port); |
dfd56b8b | 1003 | #endif |
a26cfad6 CL |
1004 | } |
1005 | ||
b4af5932 | 1006 | trace_svc_register(progname, version, protocol, port, family, error); |
363f724c | 1007 | return error; |
a26cfad6 | 1008 | } |
2c7eb0b2 | 1009 | |
642ee6b2 TM |
1010 | int svc_rpcbind_set_version(struct net *net, |
1011 | const struct svc_program *progp, | |
1012 | u32 version, int family, | |
1013 | unsigned short proto, | |
1014 | unsigned short port) | |
1015 | { | |
642ee6b2 TM |
1016 | return __svc_register(net, progp->pg_name, progp->pg_prog, |
1017 | version, family, proto, port); | |
1018 | ||
1019 | } | |
1020 | EXPORT_SYMBOL_GPL(svc_rpcbind_set_version); | |
1021 | ||
1022 | int svc_generic_rpcbind_set(struct net *net, | |
1023 | const struct svc_program *progp, | |
1024 | u32 version, int family, | |
1025 | unsigned short proto, | |
1026 | unsigned short port) | |
1027 | { | |
1028 | const struct svc_version *vers = progp->pg_vers[version]; | |
1029 | int error; | |
1030 | ||
1031 | if (vers == NULL) | |
1032 | return 0; | |
1033 | ||
1034 | if (vers->vs_hidden) { | |
b4af5932 CL |
1035 | trace_svc_noregister(progp->pg_name, version, proto, |
1036 | port, family, 0); | |
642ee6b2 TM |
1037 | return 0; |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Don't register a UDP port if we need congestion | |
1042 | * control. | |
1043 | */ | |
1044 | if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP) | |
1045 | return 0; | |
1046 | ||
1047 | error = svc_rpcbind_set_version(net, progp, version, | |
1048 | family, proto, port); | |
1049 | ||
1050 | return (vers->vs_rpcb_optnl) ? 0 : error; | |
1051 | } | |
1052 | EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set); | |
1053 | ||
a26cfad6 CL |
1054 | /** |
1055 | * svc_register - register an RPC service with the local portmapper | |
1056 | * @serv: svc_serv struct for the service to register | |
5247fab5 | 1057 | * @net: net namespace for the service to register |
4b62e58c | 1058 | * @family: protocol family of service's listener socket |
a26cfad6 CL |
1059 | * @proto: transport protocol number to advertise |
1060 | * @port: port to advertise | |
1061 | * | |
4b62e58c | 1062 | * Service is registered for any address in the passed-in protocol family |
a26cfad6 | 1063 | */ |
5247fab5 SK |
1064 | int svc_register(const struct svc_serv *serv, struct net *net, |
1065 | const int family, const unsigned short proto, | |
1066 | const unsigned short port) | |
1da177e4 LT |
1067 | { |
1068 | struct svc_program *progp; | |
ea339d46 | 1069 | unsigned int i; |
14aeb211 | 1070 | int error = 0; |
1da177e4 | 1071 | |
0af39507 WAA |
1072 | WARN_ON_ONCE(proto == 0 && port == 0); |
1073 | if (proto == 0 && port == 0) | |
1074 | return -EINVAL; | |
1da177e4 | 1075 | |
bc5fea42 OK |
1076 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
1077 | for (i = 0; i < progp->pg_nvers; i++) { | |
7e55b59b | 1078 | |
642ee6b2 TM |
1079 | error = progp->pg_rpcbind_set(net, progp, i, |
1080 | family, proto, port); | |
7e55b59b KM |
1081 | if (error < 0) { |
1082 | printk(KERN_WARNING "svc: failed to register " | |
1083 | "%sv%u RPC service (errno %d).\n", | |
1084 | progp->pg_name, i, -error); | |
bc5fea42 | 1085 | break; |
7e55b59b | 1086 | } |
1da177e4 LT |
1087 | } |
1088 | } | |
1089 | ||
7252d575 CL |
1090 | return error; |
1091 | } | |
1092 | ||
d5a8620f CL |
1093 | /* |
1094 | * If user space is running rpcbind, it should take the v4 UNSET | |
1095 | * and clear everything for this [program, version]. If user space | |
1096 | * is running portmap, it will reject the v4 UNSET, but won't have | |
1097 | * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient | |
1098 | * in this case to clear all existing entries for [program, version]. | |
1099 | */ | |
5247fab5 | 1100 | static void __svc_unregister(struct net *net, const u32 program, const u32 version, |
f6fb3f6f CL |
1101 | const char *progname) |
1102 | { | |
f6fb3f6f CL |
1103 | int error; |
1104 | ||
5247fab5 | 1105 | error = rpcb_v4_register(net, program, version, NULL, ""); |
f6fb3f6f | 1106 | |
d5a8620f CL |
1107 | /* |
1108 | * User space didn't support rpcbind v4, so retry this | |
1109 | * request with the legacy rpcbind v2 protocol. | |
1110 | */ | |
1111 | if (error == -EPROTONOSUPPORT) | |
5247fab5 | 1112 | error = rpcb_register(net, program, version, 0, 0); |
f6fb3f6f | 1113 | |
b4af5932 | 1114 | trace_svc_unregister(progname, version, error); |
f6fb3f6f CL |
1115 | } |
1116 | ||
7252d575 | 1117 | /* |
f6fb3f6f CL |
1118 | * All netids, bind addresses and ports registered for [program, version] |
1119 | * are removed from the local rpcbind database (if the service is not | |
1120 | * hidden) to make way for a new instance of the service. | |
7252d575 | 1121 | * |
f6fb3f6f CL |
1122 | * The result of unregistration is reported via dprintk for those who want |
1123 | * verification of the result, but is otherwise not important. | |
7252d575 | 1124 | */ |
5247fab5 | 1125 | static void svc_unregister(const struct svc_serv *serv, struct net *net) |
7252d575 CL |
1126 | { |
1127 | struct svc_program *progp; | |
1128 | unsigned long flags; | |
1129 | unsigned int i; | |
7252d575 CL |
1130 | |
1131 | clear_thread_flag(TIF_SIGPENDING); | |
1132 | ||
1133 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | |
1134 | for (i = 0; i < progp->pg_nvers; i++) { | |
1135 | if (progp->pg_vers[i] == NULL) | |
1136 | continue; | |
1137 | if (progp->pg_vers[i]->vs_hidden) | |
1138 | continue; | |
5247fab5 | 1139 | __svc_unregister(net, progp->pg_prog, i, progp->pg_name); |
7252d575 | 1140 | } |
1da177e4 LT |
1141 | } |
1142 | ||
7252d575 CL |
1143 | spin_lock_irqsave(¤t->sighand->siglock, flags); |
1144 | recalc_sigpending(); | |
1145 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | |
1da177e4 LT |
1146 | } |
1147 | ||
354ecbb9 | 1148 | /* |
7032a3dd | 1149 | * dprintk the given error with the address of the client that caused it. |
354ecbb9 | 1150 | */ |
f895b252 | 1151 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
b9075fa9 | 1152 | static __printf(2, 3) |
e87cc472 | 1153 | void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) |
354ecbb9 | 1154 | { |
e87cc472 | 1155 | struct va_format vaf; |
354ecbb9 | 1156 | va_list args; |
354ecbb9 DDAG |
1157 | char buf[RPC_MAX_ADDRBUFLEN]; |
1158 | ||
e87cc472 | 1159 | va_start(args, fmt); |
354ecbb9 | 1160 | |
e87cc472 JP |
1161 | vaf.fmt = fmt; |
1162 | vaf.va = &args; | |
354ecbb9 | 1163 | |
7032a3dd | 1164 | dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); |
354ecbb9 | 1165 | |
e87cc472 | 1166 | va_end(args); |
354ecbb9 | 1167 | } |
624ab464 BF |
1168 | #else |
1169 | static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {} | |
1170 | #endif | |
354ecbb9 | 1171 | |
8e5b6773 TM |
1172 | __be32 |
1173 | svc_generic_init_request(struct svc_rqst *rqstp, | |
1174 | const struct svc_program *progp, | |
1175 | struct svc_process_info *ret) | |
1176 | { | |
1177 | const struct svc_version *versp = NULL; /* compiler food */ | |
1178 | const struct svc_procedure *procp = NULL; | |
1179 | ||
1180 | if (rqstp->rq_vers >= progp->pg_nvers ) | |
1181 | goto err_bad_vers; | |
e41f9efb CIK |
1182 | versp = progp->pg_vers[rqstp->rq_vers]; |
1183 | if (!versp) | |
8e5b6773 TM |
1184 | goto err_bad_vers; |
1185 | ||
1186 | /* | |
1187 | * Some protocol versions (namely NFSv4) require some form of | |
1188 | * congestion control. (See RFC 7530 section 3.1 paragraph 2) | |
1189 | * In other words, UDP is not allowed. We mark those when setting | |
1190 | * up the svc_xprt, and verify that here. | |
1191 | * | |
1192 | * The spec is not very clear about what error should be returned | |
1193 | * when someone tries to access a server that is listening on UDP | |
1194 | * for lower versions. RPC_PROG_MISMATCH seems to be the closest | |
1195 | * fit. | |
1196 | */ | |
1197 | if (versp->vs_need_cong_ctrl && rqstp->rq_xprt && | |
1198 | !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags)) | |
1199 | goto err_bad_vers; | |
1200 | ||
1201 | if (rqstp->rq_proc >= versp->vs_nproc) | |
1202 | goto err_bad_proc; | |
1203 | rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc]; | |
1204 | if (!procp) | |
1205 | goto err_bad_proc; | |
1206 | ||
1207 | /* Initialize storage for argp and resp */ | |
103cc1fa | 1208 | memset(rqstp->rq_argp, 0, procp->pc_argzero); |
8e5b6773 TM |
1209 | memset(rqstp->rq_resp, 0, procp->pc_ressize); |
1210 | ||
1211 | /* Bump per-procedure stats counter */ | |
1212 | versp->vs_count[rqstp->rq_proc]++; | |
1213 | ||
1214 | ret->dispatch = versp->vs_dispatch; | |
1215 | return rpc_success; | |
1216 | err_bad_vers: | |
1217 | ret->mismatch.lovers = progp->pg_lovers; | |
1218 | ret->mismatch.hivers = progp->pg_hivers; | |
1219 | return rpc_prog_mismatch; | |
1220 | err_bad_proc: | |
1221 | return rpc_proc_unavail; | |
1222 | } | |
1223 | EXPORT_SYMBOL_GPL(svc_generic_init_request); | |
1224 | ||
1da177e4 | 1225 | /* |
1cad7ea6 | 1226 | * Common routine for processing the RPC request. |
1da177e4 | 1227 | */ |
1cad7ea6 RL |
1228 | static int |
1229 | svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) | |
1da177e4 LT |
1230 | { |
1231 | struct svc_program *progp; | |
860bda29 | 1232 | const struct svc_procedure *procp = NULL; |
6fb2b47f | 1233 | struct svc_serv *serv = rqstp->rq_server; |
8e5b6773 | 1234 | struct svc_process_info process; |
d8ed029d | 1235 | __be32 *statp; |
8e5b6773 | 1236 | u32 prog, vers; |
438623a0 | 1237 | __be32 rpc_stat; |
5b747a59 | 1238 | int auth_res, rc; |
8f8e05c5 | 1239 | __be32 *reply_statp; |
1da177e4 LT |
1240 | |
1241 | rpc_stat = rpc_success; | |
1242 | ||
1243 | if (argv->iov_len < 6*4) | |
1244 | goto err_short_len; | |
1245 | ||
06eb8a56 | 1246 | /* Will be turned off by GSS integrity and privacy services */ |
28df0988 | 1247 | __set_bit(RQ_SPLICE_OK, &rqstp->rq_flags); |
2f425878 | 1248 | /* Will be turned off only when NFSv4 Sessions are used */ |
28df0988 CL |
1249 | __set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags); |
1250 | __clear_bit(RQ_DROPME, &rqstp->rq_flags); | |
e831fe65 | 1251 | |
1da177e4 LT |
1252 | svc_putu32(resv, rqstp->rq_xid); |
1253 | ||
76994313 | 1254 | vers = svc_getnl(argv); |
1da177e4 LT |
1255 | |
1256 | /* First words of reply: */ | |
76994313 | 1257 | svc_putnl(resv, 1); /* REPLY */ |
1da177e4 | 1258 | |
1da177e4 LT |
1259 | if (vers != 2) /* RPC version number */ |
1260 | goto err_bad_rpc; | |
1261 | ||
1262 | /* Save position in case we later decide to reject: */ | |
8f8e05c5 | 1263 | reply_statp = resv->iov_base + resv->iov_len; |
1da177e4 | 1264 | |
76994313 | 1265 | svc_putnl(resv, 0); /* ACCEPT */ |
1da177e4 | 1266 | |
76994313 | 1267 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
8e5b6773 TM |
1268 | rqstp->rq_vers = svc_getnl(argv); /* version number */ |
1269 | rqstp->rq_proc = svc_getnl(argv); /* procedure number */ | |
1da177e4 | 1270 | |
80d188a6 N |
1271 | for (progp = serv->sv_program; progp; progp = progp->pg_next) |
1272 | if (prog == progp->pg_prog) | |
1273 | break; | |
1274 | ||
1da177e4 LT |
1275 | /* |
1276 | * Decode auth data, and add verifier to reply buffer. | |
1277 | * We do this before anything else in order to get a decent | |
1278 | * auth verifier. | |
1279 | */ | |
438623a0 | 1280 | auth_res = svc_authenticate(rqstp); |
1da177e4 | 1281 | /* Also give the program a chance to reject this call: */ |
5c2465df | 1282 | if (auth_res == SVC_OK && progp) |
1da177e4 | 1283 | auth_res = progp->pg_authenticate(rqstp); |
ff27e9f7 | 1284 | if (auth_res != SVC_OK) |
438623a0 | 1285 | trace_svc_authenticate(rqstp, auth_res); |
1da177e4 LT |
1286 | switch (auth_res) { |
1287 | case SVC_OK: | |
1288 | break; | |
1289 | case SVC_GARBAGE: | |
dd35210e | 1290 | goto err_garbage; |
1da177e4 LT |
1291 | case SVC_SYSERR: |
1292 | rpc_stat = rpc_system_err; | |
1293 | goto err_bad; | |
1294 | case SVC_DENIED: | |
1295 | goto err_bad_auth; | |
1ebede86 | 1296 | case SVC_CLOSE: |
4d712ef1 | 1297 | goto close; |
1da177e4 LT |
1298 | case SVC_DROP: |
1299 | goto dropit; | |
1300 | case SVC_COMPLETE: | |
1301 | goto sendit; | |
1302 | } | |
80d188a6 | 1303 | |
9ba02638 | 1304 | if (progp == NULL) |
1da177e4 LT |
1305 | goto err_bad_prog; |
1306 | ||
8e5b6773 TM |
1307 | rpc_stat = progp->pg_init_request(rqstp, progp, &process); |
1308 | switch (rpc_stat) { | |
1309 | case rpc_success: | |
1310 | break; | |
1311 | case rpc_prog_unavail: | |
1312 | goto err_bad_prog; | |
1313 | case rpc_prog_mismatch: | |
5283b03e | 1314 | goto err_bad_vers; |
8e5b6773 TM |
1315 | case rpc_proc_unavail: |
1316 | goto err_bad_proc; | |
1317 | } | |
5283b03e | 1318 | |
8e5b6773 TM |
1319 | procp = rqstp->rq_procinfo; |
1320 | /* Should this check go into the dispatcher? */ | |
1321 | if (!procp || !procp->pc_func) | |
1da177e4 | 1322 | goto err_bad_proc; |
1da177e4 LT |
1323 | |
1324 | /* Syntactic check complete */ | |
1325 | serv->sv_stats->rpccnt++; | |
0b9547bf | 1326 | trace_svc_process(rqstp, progp->pg_name); |
1da177e4 LT |
1327 | |
1328 | /* Build the reply header. */ | |
1329 | statp = resv->iov_base +resv->iov_len; | |
76994313 | 1330 | svc_putnl(resv, RPC_SUCCESS); |
1da177e4 | 1331 | |
cca5172a | 1332 | /* un-reserve some of the out-queue now that we have a |
1da177e4 LT |
1333 | * better idea of reply size |
1334 | */ | |
1335 | if (procp->pc_xdrressize) | |
cd123012 | 1336 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1da177e4 LT |
1337 | |
1338 | /* Call the function that processes the request. */ | |
5b747a59 CL |
1339 | rc = process.dispatch(rqstp, statp); |
1340 | if (procp->pc_release) | |
1341 | procp->pc_release(rqstp); | |
1342 | if (!rc) | |
1343 | goto dropit; | |
9082e1d9 | 1344 | if (rqstp->rq_auth_stat != rpc_auth_ok) |
5b747a59 | 1345 | goto err_bad_auth; |
9082e1d9 | 1346 | |
1da177e4 LT |
1347 | /* Check RPC status result */ |
1348 | if (*statp != rpc_success) | |
1349 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | |
1350 | ||
1da177e4 LT |
1351 | if (procp->pc_encode == NULL) |
1352 | goto dropit; | |
1353 | ||
1354 | sendit: | |
1355 | if (svc_authorise(rqstp)) | |
f1442d63 | 1356 | goto close_xprt; |
1cad7ea6 | 1357 | return 1; /* Caller can now send it */ |
1da177e4 LT |
1358 | |
1359 | dropit: | |
1360 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | |
1361 | dprintk("svc: svc_process dropit\n"); | |
1da177e4 LT |
1362 | return 0; |
1363 | ||
4d712ef1 | 1364 | close: |
f1442d63 DK |
1365 | svc_authorise(rqstp); |
1366 | close_xprt: | |
d4b09acf | 1367 | if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) |
4355d767 | 1368 | svc_xprt_close(rqstp->rq_xprt); |
4d712ef1 CL |
1369 | dprintk("svc: svc_process close\n"); |
1370 | return 0; | |
1371 | ||
1da177e4 | 1372 | err_short_len: |
5b5e0928 | 1373 | svc_printk(rqstp, "short len %zd, dropping request\n", |
354ecbb9 | 1374 | argv->iov_len); |
f1442d63 | 1375 | goto close_xprt; |
1da177e4 | 1376 | |
1da177e4 LT |
1377 | err_bad_rpc: |
1378 | serv->sv_stats->rpcbadfmt++; | |
76994313 AD |
1379 | svc_putnl(resv, 1); /* REJECT */ |
1380 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | |
1381 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | |
1382 | svc_putnl(resv, 2); | |
1da177e4 LT |
1383 | goto sendit; |
1384 | ||
1385 | err_bad_auth: | |
438623a0 CL |
1386 | dprintk("svc: authentication failed (%d)\n", |
1387 | be32_to_cpu(rqstp->rq_auth_stat)); | |
1da177e4 LT |
1388 | serv->sv_stats->rpcbadauth++; |
1389 | /* Restore write pointer to location of accept status: */ | |
8f8e05c5 | 1390 | xdr_ressize_check(rqstp, reply_statp); |
76994313 AD |
1391 | svc_putnl(resv, 1); /* REJECT */ |
1392 | svc_putnl(resv, 1); /* AUTH_ERROR */ | |
438623a0 | 1393 | svc_putu32(resv, rqstp->rq_auth_stat); /* status */ |
1da177e4 LT |
1394 | goto sendit; |
1395 | ||
1396 | err_bad_prog: | |
9ba02638 | 1397 | dprintk("svc: unknown program %d\n", prog); |
1da177e4 | 1398 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1399 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1da177e4 LT |
1400 | goto sendit; |
1401 | ||
1402 | err_bad_vers: | |
354ecbb9 | 1403 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
8e5b6773 | 1404 | rqstp->rq_vers, rqstp->rq_prog, progp->pg_name); |
34e9a63b | 1405 | |
1da177e4 | 1406 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1407 | svc_putnl(resv, RPC_PROG_MISMATCH); |
8e5b6773 TM |
1408 | svc_putnl(resv, process.mismatch.lovers); |
1409 | svc_putnl(resv, process.mismatch.hivers); | |
1da177e4 LT |
1410 | goto sendit; |
1411 | ||
1412 | err_bad_proc: | |
8e5b6773 | 1413 | svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc); |
34e9a63b | 1414 | |
1da177e4 | 1415 | serv->sv_stats->rpcbadfmt++; |
76994313 | 1416 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1da177e4 LT |
1417 | goto sendit; |
1418 | ||
1419 | err_garbage: | |
354ecbb9 | 1420 | svc_printk(rqstp, "failed to decode args\n"); |
34e9a63b | 1421 | |
1da177e4 LT |
1422 | rpc_stat = rpc_garbage_args; |
1423 | err_bad: | |
1424 | serv->sv_stats->rpcbadfmt++; | |
76994313 | 1425 | svc_putnl(resv, ntohl(rpc_stat)); |
1da177e4 LT |
1426 | goto sendit; |
1427 | } | |
7adae489 | 1428 | |
1cad7ea6 RL |
1429 | /* |
1430 | * Process the RPC request. | |
1431 | */ | |
1432 | int | |
1433 | svc_process(struct svc_rqst *rqstp) | |
1434 | { | |
1435 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1436 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
81593c4d | 1437 | __be32 dir; |
1cad7ea6 | 1438 | |
3a126180 CL |
1439 | #if IS_ENABLED(CONFIG_FAIL_SUNRPC) |
1440 | if (!fail_sunrpc.ignore_server_disconnect && | |
1441 | should_fail(&fail_sunrpc.attr, 1)) | |
1442 | svc_xprt_deferred_close(rqstp->rq_xprt); | |
1443 | #endif | |
1444 | ||
1cad7ea6 RL |
1445 | /* |
1446 | * Setup response xdr_buf. | |
1447 | * Initially it has just one page | |
1448 | */ | |
afc59400 | 1449 | rqstp->rq_next_page = &rqstp->rq_respages[1]; |
1cad7ea6 RL |
1450 | resv->iov_base = page_address(rqstp->rq_respages[0]); |
1451 | resv->iov_len = 0; | |
81593c4d | 1452 | rqstp->rq_res.pages = rqstp->rq_next_page; |
1cad7ea6 RL |
1453 | rqstp->rq_res.len = 0; |
1454 | rqstp->rq_res.page_base = 0; | |
1455 | rqstp->rq_res.page_len = 0; | |
1456 | rqstp->rq_res.buflen = PAGE_SIZE; | |
1457 | rqstp->rq_res.tail[0].iov_base = NULL; | |
1458 | rqstp->rq_res.tail[0].iov_len = 0; | |
1459 | ||
81593c4d CL |
1460 | dir = svc_getu32(argv); |
1461 | if (dir != rpc_call) | |
1462 | goto out_baddir; | |
1463 | if (!svc_process_common(rqstp, argv, resv)) | |
860a0d9e | 1464 | goto out_drop; |
81593c4d | 1465 | return svc_send(rqstp); |
860a0d9e | 1466 | |
81593c4d CL |
1467 | out_baddir: |
1468 | svc_printk(rqstp, "bad direction 0x%08x, dropping request\n", | |
1469 | be32_to_cpu(dir)); | |
1470 | rqstp->rq_server->sv_stats->rpcbadfmt++; | |
860a0d9e | 1471 | out_drop: |
860a0d9e JL |
1472 | svc_drop(rqstp); |
1473 | return 0; | |
1cad7ea6 | 1474 | } |
3f87d5d6 | 1475 | EXPORT_SYMBOL_GPL(svc_process); |
1cad7ea6 | 1476 | |
9e00abc3 | 1477 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
4d6bbb62 RL |
1478 | /* |
1479 | * Process a backchannel RPC request that arrived over an existing | |
1480 | * outbound connection | |
1481 | */ | |
1482 | int | |
1483 | bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, | |
1484 | struct svc_rqst *rqstp) | |
1485 | { | |
1486 | struct kvec *argv = &rqstp->rq_arg.head[0]; | |
1487 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
632dda83 | 1488 | struct rpc_task *task; |
0d2a970d | 1489 | int proc_error; |
632dda83 CL |
1490 | int error; |
1491 | ||
1492 | dprintk("svc: %s(%p)\n", __func__, req); | |
4d6bbb62 RL |
1493 | |
1494 | /* Build the svc_rqst used by the common processing routine */ | |
1495 | rqstp->rq_xid = req->rq_xid; | |
1496 | rqstp->rq_prot = req->rq_xprt->prot; | |
1497 | rqstp->rq_server = serv; | |
d4b09acf | 1498 | rqstp->rq_bc_net = req->rq_xprt->xprt_net; |
4d6bbb62 RL |
1499 | |
1500 | rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); | |
1501 | memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); | |
1502 | memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); | |
1503 | memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); | |
756b9b37 TM |
1504 | |
1505 | /* Adjust the argument buffer length */ | |
38b7631f | 1506 | rqstp->rq_arg.len = req->rq_private_buf.len; |
756b9b37 TM |
1507 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1508 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1509 | rqstp->rq_arg.page_len = 0; | |
1510 | } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len + | |
1511 | rqstp->rq_arg.page_len) | |
1512 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - | |
1513 | rqstp->rq_arg.head[0].iov_len; | |
1514 | else | |
1515 | rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len + | |
1516 | rqstp->rq_arg.page_len; | |
4d6bbb62 RL |
1517 | |
1518 | /* reset result send buffer "put" position */ | |
1519 | resv->iov_len = 0; | |
1520 | ||
4d6bbb62 RL |
1521 | /* |
1522 | * Skip the next two words because they've already been | |
632dda83 | 1523 | * processed in the transport |
4d6bbb62 RL |
1524 | */ |
1525 | svc_getu32(argv); /* XID */ | |
1526 | svc_getnl(argv); /* CALLDIR */ | |
1527 | ||
632dda83 | 1528 | /* Parse and execute the bc call */ |
0d2a970d TM |
1529 | proc_error = svc_process_common(rqstp, argv, resv); |
1530 | ||
7402a4fe | 1531 | atomic_dec(&req->rq_xprt->bc_slot_count); |
0d2a970d | 1532 | if (!proc_error) { |
632dda83 | 1533 | /* Processing error: drop the request */ |
b3b02ae5 | 1534 | xprt_free_bc_request(req); |
8f7766c8 VA |
1535 | error = -EINVAL; |
1536 | goto out; | |
4b5b3ba1 | 1537 | } |
632dda83 CL |
1538 | /* Finally, send the reply synchronously */ |
1539 | memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf)); | |
0f419791 | 1540 | task = rpc_run_bc_task(req); |
632dda83 CL |
1541 | if (IS_ERR(task)) { |
1542 | error = PTR_ERR(task); | |
1543 | goto out; | |
1544 | } | |
1545 | ||
1546 | WARN_ON_ONCE(atomic_read(&task->tk_count) != 1); | |
1547 | error = task->tk_status; | |
1548 | rpc_put_task(task); | |
1549 | ||
1550 | out: | |
1551 | dprintk("svc: %s(), error=%d\n", __func__, error); | |
1552 | return error; | |
4d6bbb62 | 1553 | } |
0d961aa9 | 1554 | EXPORT_SYMBOL_GPL(bc_svc_process); |
9e00abc3 | 1555 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
4d6bbb62 | 1556 | |
7adae489 GB |
1557 | /* |
1558 | * Return (transport-specific) limit on the rpc payload. | |
1559 | */ | |
1560 | u32 svc_max_payload(const struct svc_rqst *rqstp) | |
1561 | { | |
49023155 | 1562 | u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; |
7adae489 | 1563 | |
c6b0a9f8 N |
1564 | if (rqstp->rq_server->sv_max_payload < max) |
1565 | max = rqstp->rq_server->sv_max_payload; | |
7adae489 GB |
1566 | return max; |
1567 | } | |
1568 | EXPORT_SYMBOL_GPL(svc_max_payload); | |
8154ef27 | 1569 | |
5c117207 CL |
1570 | /** |
1571 | * svc_proc_name - Return RPC procedure name in string form | |
1572 | * @rqstp: svc_rqst to operate on | |
1573 | * | |
1574 | * Return value: | |
1575 | * Pointer to a NUL-terminated string | |
1576 | */ | |
1577 | const char *svc_proc_name(const struct svc_rqst *rqstp) | |
1578 | { | |
1579 | if (rqstp && rqstp->rq_procinfo) | |
1580 | return rqstp->rq_procinfo->pc_name; | |
1581 | return "unknown"; | |
1582 | } | |
1583 | ||
1584 | ||
41205539 | 1585 | /** |
03493bca | 1586 | * svc_encode_result_payload - mark a range of bytes as a result payload |
41205539 CL |
1587 | * @rqstp: svc_rqst to operate on |
1588 | * @offset: payload's byte offset in rqstp->rq_res | |
1589 | * @length: size of payload, in bytes | |
1590 | * | |
1591 | * Returns zero on success, or a negative errno if a permanent | |
1592 | * error occurred. | |
1593 | */ | |
03493bca CL |
1594 | int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset, |
1595 | unsigned int length) | |
41205539 | 1596 | { |
03493bca CL |
1597 | return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset, |
1598 | length); | |
41205539 | 1599 | } |
03493bca | 1600 | EXPORT_SYMBOL_GPL(svc_encode_result_payload); |
41205539 | 1601 | |
8154ef27 CL |
1602 | /** |
1603 | * svc_fill_write_vector - Construct data argument for VFS write call | |
1604 | * @rqstp: svc_rqst to operate on | |
dae9a6ca | 1605 | * @payload: xdr_buf containing only the write data payload |
8154ef27 | 1606 | * |
3fd9557a | 1607 | * Fills in rqstp::rq_vec, and returns the number of elements. |
8154ef27 | 1608 | */ |
dae9a6ca CL |
1609 | unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, |
1610 | struct xdr_buf *payload) | |
8154ef27 | 1611 | { |
dae9a6ca CL |
1612 | struct page **pages = payload->pages; |
1613 | struct kvec *first = payload->head; | |
8154ef27 | 1614 | struct kvec *vec = rqstp->rq_vec; |
dae9a6ca | 1615 | size_t total = payload->len; |
8154ef27 CL |
1616 | unsigned int i; |
1617 | ||
1618 | /* Some types of transport can present the write payload | |
1619 | * entirely in rq_arg.pages. In this case, @first is empty. | |
1620 | */ | |
1621 | i = 0; | |
1622 | if (first->iov_len) { | |
1623 | vec[i].iov_base = first->iov_base; | |
1624 | vec[i].iov_len = min_t(size_t, total, first->iov_len); | |
1625 | total -= vec[i].iov_len; | |
1626 | ++i; | |
1627 | } | |
1628 | ||
8154ef27 CL |
1629 | while (total) { |
1630 | vec[i].iov_base = page_address(*pages); | |
1631 | vec[i].iov_len = min_t(size_t, total, PAGE_SIZE); | |
1632 | total -= vec[i].iov_len; | |
1633 | ++i; | |
8154ef27 CL |
1634 | ++pages; |
1635 | } | |
1636 | ||
1637 | WARN_ON_ONCE(i > ARRAY_SIZE(rqstp->rq_vec)); | |
1638 | return i; | |
1639 | } | |
1640 | EXPORT_SYMBOL_GPL(svc_fill_write_vector); | |
38a70315 CL |
1641 | |
1642 | /** | |
1643 | * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call | |
1644 | * @rqstp: svc_rqst to operate on | |
1645 | * @first: buffer containing first section of pathname | |
11b4d66e | 1646 | * @p: buffer containing remaining section of pathname |
38a70315 CL |
1647 | * @total: total length of the pathname argument |
1648 | * | |
11b4d66e CL |
1649 | * The VFS symlink API demands a NUL-terminated pathname in mapped memory. |
1650 | * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free | |
1651 | * the returned string. | |
38a70315 CL |
1652 | */ |
1653 | char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first, | |
11b4d66e | 1654 | void *p, size_t total) |
38a70315 | 1655 | { |
11b4d66e CL |
1656 | size_t len, remaining; |
1657 | char *result, *dst; | |
38a70315 | 1658 | |
11b4d66e CL |
1659 | result = kmalloc(total + 1, GFP_KERNEL); |
1660 | if (!result) | |
1661 | return ERR_PTR(-ESERVERFAULT); | |
38a70315 | 1662 | |
11b4d66e CL |
1663 | dst = result; |
1664 | remaining = total; | |
38a70315 | 1665 | |
11b4d66e CL |
1666 | len = min_t(size_t, total, first->iov_len); |
1667 | if (len) { | |
38a70315 CL |
1668 | memcpy(dst, first->iov_base, len); |
1669 | dst += len; | |
1670 | remaining -= len; | |
11b4d66e | 1671 | } |
38a70315 | 1672 | |
11b4d66e CL |
1673 | if (remaining) { |
1674 | len = min_t(size_t, remaining, PAGE_SIZE); | |
1675 | memcpy(dst, p, len); | |
1676 | dst += len; | |
38a70315 CL |
1677 | } |
1678 | ||
11b4d66e CL |
1679 | *dst = '\0'; |
1680 | ||
1681 | /* Sanity check: Linux doesn't allow the pathname argument to | |
38a70315 CL |
1682 | * contain a NUL byte. |
1683 | */ | |
11b4d66e CL |
1684 | if (strlen(result) != total) { |
1685 | kfree(result); | |
38a70315 | 1686 | return ERR_PTR(-EINVAL); |
11b4d66e | 1687 | } |
38a70315 CL |
1688 | return result; |
1689 | } | |
1690 | EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname); |