Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * lustre/ptlrpc/sec.c | |
37 | * | |
38 | * Author: Eric Mei <ericm@clusterfs.com> | |
39 | */ | |
40 | ||
41 | #define DEBUG_SUBSYSTEM S_SEC | |
42 | ||
9fdaf8c0 | 43 | #include "../../include/linux/libcfs/libcfs.h" |
d7e09d03 PT |
44 | #include <linux/crypto.h> |
45 | #include <linux/key.h> | |
46 | ||
e27db149 GKH |
47 | #include "../include/obd.h" |
48 | #include "../include/obd_class.h" | |
49 | #include "../include/obd_support.h" | |
50 | #include "../include/lustre_net.h" | |
51 | #include "../include/lustre_import.h" | |
52 | #include "../include/lustre_dlm.h" | |
53 | #include "../include/lustre_sec.h" | |
d7e09d03 PT |
54 | |
55 | #include "ptlrpc_internal.h" | |
56 | ||
57 | /*********************************************** | |
58 | * policy registers * | |
59 | ***********************************************/ | |
60 | ||
61 | static rwlock_t policy_lock; | |
62 | static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = { | |
63 | NULL, | |
64 | }; | |
65 | ||
66 | int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy) | |
67 | { | |
68 | __u16 number = policy->sp_policy; | |
69 | ||
70 | LASSERT(policy->sp_name); | |
71 | LASSERT(policy->sp_cops); | |
72 | LASSERT(policy->sp_sops); | |
73 | ||
74 | if (number >= SPTLRPC_POLICY_MAX) | |
75 | return -EINVAL; | |
76 | ||
77 | write_lock(&policy_lock); | |
78 | if (unlikely(policies[number])) { | |
79 | write_unlock(&policy_lock); | |
80 | return -EALREADY; | |
81 | } | |
82 | policies[number] = policy; | |
83 | write_unlock(&policy_lock); | |
84 | ||
85 | CDEBUG(D_SEC, "%s: registered\n", policy->sp_name); | |
86 | return 0; | |
87 | } | |
88 | EXPORT_SYMBOL(sptlrpc_register_policy); | |
89 | ||
90 | int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy) | |
91 | { | |
92 | __u16 number = policy->sp_policy; | |
93 | ||
94 | LASSERT(number < SPTLRPC_POLICY_MAX); | |
95 | ||
96 | write_lock(&policy_lock); | |
97 | if (unlikely(policies[number] == NULL)) { | |
98 | write_unlock(&policy_lock); | |
99 | CERROR("%s: already unregistered\n", policy->sp_name); | |
100 | return -EINVAL; | |
101 | } | |
102 | ||
103 | LASSERT(policies[number] == policy); | |
104 | policies[number] = NULL; | |
105 | write_unlock(&policy_lock); | |
106 | ||
107 | CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name); | |
108 | return 0; | |
109 | } | |
110 | EXPORT_SYMBOL(sptlrpc_unregister_policy); | |
111 | ||
112 | static | |
aff9d8e8 | 113 | struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor) |
d7e09d03 PT |
114 | { |
115 | static DEFINE_MUTEX(load_mutex); | |
d0bfef31 | 116 | static atomic_t loaded = ATOMIC_INIT(0); |
d7e09d03 | 117 | struct ptlrpc_sec_policy *policy; |
d0bfef31 CH |
118 | __u16 number = SPTLRPC_FLVR_POLICY(flavor); |
119 | __u16 flag = 0; | |
d7e09d03 PT |
120 | |
121 | if (number >= SPTLRPC_POLICY_MAX) | |
122 | return NULL; | |
123 | ||
124 | while (1) { | |
125 | read_lock(&policy_lock); | |
126 | policy = policies[number]; | |
127 | if (policy && !try_module_get(policy->sp_owner)) | |
128 | policy = NULL; | |
129 | if (policy == NULL) | |
130 | flag = atomic_read(&loaded); | |
131 | read_unlock(&policy_lock); | |
132 | ||
133 | if (policy != NULL || flag != 0 || | |
134 | number != SPTLRPC_POLICY_GSS) | |
135 | break; | |
136 | ||
137 | /* try to load gss module, once */ | |
138 | mutex_lock(&load_mutex); | |
139 | if (atomic_read(&loaded) == 0) { | |
140 | if (request_module("ptlrpc_gss") == 0) | |
141 | CDEBUG(D_SEC, | |
142 | "module ptlrpc_gss loaded on demand\n"); | |
143 | else | |
144 | CERROR("Unable to load module ptlrpc_gss\n"); | |
145 | ||
146 | atomic_set(&loaded, 1); | |
147 | } | |
148 | mutex_unlock(&load_mutex); | |
149 | } | |
150 | ||
151 | return policy; | |
152 | } | |
153 | ||
154 | __u32 sptlrpc_name2flavor_base(const char *name) | |
155 | { | |
156 | if (!strcmp(name, "null")) | |
157 | return SPTLRPC_FLVR_NULL; | |
158 | if (!strcmp(name, "plain")) | |
159 | return SPTLRPC_FLVR_PLAIN; | |
160 | if (!strcmp(name, "krb5n")) | |
161 | return SPTLRPC_FLVR_KRB5N; | |
162 | if (!strcmp(name, "krb5a")) | |
163 | return SPTLRPC_FLVR_KRB5A; | |
164 | if (!strcmp(name, "krb5i")) | |
165 | return SPTLRPC_FLVR_KRB5I; | |
166 | if (!strcmp(name, "krb5p")) | |
167 | return SPTLRPC_FLVR_KRB5P; | |
168 | ||
169 | return SPTLRPC_FLVR_INVALID; | |
170 | } | |
171 | EXPORT_SYMBOL(sptlrpc_name2flavor_base); | |
172 | ||
173 | const char *sptlrpc_flavor2name_base(__u32 flvr) | |
174 | { | |
175 | __u32 base = SPTLRPC_FLVR_BASE(flvr); | |
176 | ||
177 | if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) | |
178 | return "null"; | |
179 | else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN)) | |
180 | return "plain"; | |
181 | else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N)) | |
182 | return "krb5n"; | |
183 | else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A)) | |
184 | return "krb5a"; | |
185 | else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I)) | |
186 | return "krb5i"; | |
187 | else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P)) | |
188 | return "krb5p"; | |
189 | ||
190 | CERROR("invalid wire flavor 0x%x\n", flvr); | |
191 | return "invalid"; | |
192 | } | |
193 | EXPORT_SYMBOL(sptlrpc_flavor2name_base); | |
194 | ||
195 | char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf, | |
196 | char *buf, int bufsize) | |
197 | { | |
198 | if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) | |
199 | snprintf(buf, bufsize, "hash:%s", | |
200 | sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg)); | |
201 | else | |
202 | snprintf(buf, bufsize, "%s", | |
203 | sptlrpc_flavor2name_base(sf->sf_rpc)); | |
204 | ||
205 | buf[bufsize - 1] = '\0'; | |
206 | return buf; | |
207 | } | |
208 | EXPORT_SYMBOL(sptlrpc_flavor2name_bulk); | |
209 | ||
210 | char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize) | |
211 | { | |
22e1dd69 | 212 | strlcpy(buf, sptlrpc_flavor2name_base(sf->sf_rpc), bufsize); |
d7e09d03 PT |
213 | |
214 | /* | |
215 | * currently we don't support customized bulk specification for | |
216 | * flavors other than plain | |
217 | */ | |
218 | if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) { | |
219 | char bspec[16]; | |
220 | ||
221 | bspec[0] = '-'; | |
222 | sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1); | |
22e1dd69 | 223 | strlcat(buf, bspec, bufsize); |
d7e09d03 PT |
224 | } |
225 | ||
d7e09d03 PT |
226 | return buf; |
227 | } | |
228 | EXPORT_SYMBOL(sptlrpc_flavor2name); | |
229 | ||
230 | char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize) | |
231 | { | |
232 | buf[0] = '\0'; | |
233 | ||
234 | if (flags & PTLRPC_SEC_FL_REVERSE) | |
235 | strlcat(buf, "reverse,", bufsize); | |
236 | if (flags & PTLRPC_SEC_FL_ROOTONLY) | |
237 | strlcat(buf, "rootonly,", bufsize); | |
238 | if (flags & PTLRPC_SEC_FL_UDESC) | |
239 | strlcat(buf, "udesc,", bufsize); | |
240 | if (flags & PTLRPC_SEC_FL_BULK) | |
241 | strlcat(buf, "bulk,", bufsize); | |
242 | if (buf[0] == '\0') | |
243 | strlcat(buf, "-,", bufsize); | |
244 | ||
245 | return buf; | |
246 | } | |
247 | EXPORT_SYMBOL(sptlrpc_secflags2str); | |
248 | ||
249 | /************************************************** | |
250 | * client context APIs * | |
251 | **************************************************/ | |
252 | ||
253 | static | |
254 | struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec) | |
255 | { | |
256 | struct vfs_cred vcred; | |
257 | int create = 1, remove_dead = 1; | |
258 | ||
259 | LASSERT(sec); | |
260 | LASSERT(sec->ps_policy->sp_cops->lookup_ctx); | |
261 | ||
262 | if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE | | |
263 | PTLRPC_SEC_FL_ROOTONLY)) { | |
264 | vcred.vc_uid = 0; | |
265 | vcred.vc_gid = 0; | |
266 | if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) { | |
267 | create = 0; | |
268 | remove_dead = 0; | |
269 | } | |
270 | } else { | |
4b1a25f0 PT |
271 | vcred.vc_uid = from_kuid(&init_user_ns, current_uid()); |
272 | vcred.vc_gid = from_kgid(&init_user_ns, current_gid()); | |
d7e09d03 PT |
273 | } |
274 | ||
275 | return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, | |
276 | create, remove_dead); | |
277 | } | |
278 | ||
279 | struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx) | |
280 | { | |
281 | atomic_inc(&ctx->cc_refcount); | |
282 | return ctx; | |
283 | } | |
284 | EXPORT_SYMBOL(sptlrpc_cli_ctx_get); | |
285 | ||
286 | void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync) | |
287 | { | |
288 | struct ptlrpc_sec *sec = ctx->cc_sec; | |
289 | ||
290 | LASSERT(sec); | |
291 | LASSERT_ATOMIC_POS(&ctx->cc_refcount); | |
292 | ||
293 | if (!atomic_dec_and_test(&ctx->cc_refcount)) | |
294 | return; | |
295 | ||
296 | sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync); | |
297 | } | |
298 | EXPORT_SYMBOL(sptlrpc_cli_ctx_put); | |
299 | ||
300 | /** | |
301 | * Expire the client context immediately. | |
302 | * | |
303 | * \pre Caller must hold at least 1 reference on the \a ctx. | |
304 | */ | |
305 | void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx) | |
306 | { | |
d0f17b64 CG |
307 | LASSERT(ctx->cc_ops->force_die); |
308 | ctx->cc_ops->force_die(ctx, 0); | |
d7e09d03 PT |
309 | } |
310 | EXPORT_SYMBOL(sptlrpc_cli_ctx_expire); | |
311 | ||
312 | /** | |
313 | * To wake up the threads who are waiting for this client context. Called | |
314 | * after some status change happened on \a ctx. | |
315 | */ | |
316 | void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx) | |
317 | { | |
318 | struct ptlrpc_request *req, *next; | |
319 | ||
320 | spin_lock(&ctx->cc_lock); | |
321 | list_for_each_entry_safe(req, next, &ctx->cc_req_list, | |
322 | rq_ctx_chain) { | |
323 | list_del_init(&req->rq_ctx_chain); | |
324 | ptlrpc_client_wake_req(req); | |
325 | } | |
326 | spin_unlock(&ctx->cc_lock); | |
327 | } | |
328 | EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup); | |
329 | ||
330 | int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize) | |
331 | { | |
332 | LASSERT(ctx->cc_ops); | |
333 | ||
334 | if (ctx->cc_ops->display == NULL) | |
335 | return 0; | |
336 | ||
337 | return ctx->cc_ops->display(ctx, buf, bufsize); | |
338 | } | |
339 | ||
340 | static int import_sec_check_expire(struct obd_import *imp) | |
341 | { | |
d0bfef31 | 342 | int adapt = 0; |
d7e09d03 PT |
343 | |
344 | spin_lock(&imp->imp_lock); | |
345 | if (imp->imp_sec_expire && | |
7264b8a5 | 346 | imp->imp_sec_expire < get_seconds()) { |
d7e09d03 PT |
347 | adapt = 1; |
348 | imp->imp_sec_expire = 0; | |
349 | } | |
350 | spin_unlock(&imp->imp_lock); | |
351 | ||
352 | if (!adapt) | |
353 | return 0; | |
354 | ||
355 | CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n"); | |
81fb955d | 356 | return sptlrpc_import_sec_adapt(imp, NULL, NULL); |
d7e09d03 PT |
357 | } |
358 | ||
359 | static int import_sec_validate_get(struct obd_import *imp, | |
360 | struct ptlrpc_sec **sec) | |
361 | { | |
d0bfef31 | 362 | int rc; |
d7e09d03 PT |
363 | |
364 | if (unlikely(imp->imp_sec_expire)) { | |
365 | rc = import_sec_check_expire(imp); | |
366 | if (rc) | |
367 | return rc; | |
368 | } | |
369 | ||
370 | *sec = sptlrpc_import_sec_ref(imp); | |
371 | if (*sec == NULL) { | |
372 | CERROR("import %p (%s) with no sec\n", | |
373 | imp, ptlrpc_import_state_name(imp->imp_state)); | |
374 | return -EACCES; | |
375 | } | |
376 | ||
377 | if (unlikely((*sec)->ps_dying)) { | |
378 | CERROR("attempt to use dying sec %p\n", sec); | |
379 | sptlrpc_sec_put(*sec); | |
380 | return -EACCES; | |
381 | } | |
382 | ||
383 | return 0; | |
384 | } | |
385 | ||
386 | /** | |
387 | * Given a \a req, find or allocate a appropriate context for it. | |
388 | * \pre req->rq_cli_ctx == NULL. | |
389 | * | |
390 | * \retval 0 succeed, and req->rq_cli_ctx is set. | |
391 | * \retval -ev error number, and req->rq_cli_ctx == NULL. | |
392 | */ | |
393 | int sptlrpc_req_get_ctx(struct ptlrpc_request *req) | |
394 | { | |
395 | struct obd_import *imp = req->rq_import; | |
396 | struct ptlrpc_sec *sec; | |
397 | int rc; | |
d7e09d03 PT |
398 | |
399 | LASSERT(!req->rq_cli_ctx); | |
400 | LASSERT(imp); | |
401 | ||
402 | rc = import_sec_validate_get(imp, &sec); | |
403 | if (rc) | |
0a3bdb00 | 404 | return rc; |
d7e09d03 PT |
405 | |
406 | req->rq_cli_ctx = get_my_ctx(sec); | |
407 | ||
408 | sptlrpc_sec_put(sec); | |
409 | ||
410 | if (!req->rq_cli_ctx) { | |
411 | CERROR("req %p: fail to get context\n", req); | |
0a3bdb00 | 412 | return -ENOMEM; |
d7e09d03 PT |
413 | } |
414 | ||
0a3bdb00 | 415 | return 0; |
d7e09d03 PT |
416 | } |
417 | ||
418 | /** | |
419 | * Drop the context for \a req. | |
420 | * \pre req->rq_cli_ctx != NULL. | |
421 | * \post req->rq_cli_ctx == NULL. | |
422 | * | |
423 | * If \a sync == 0, this function should return quickly without sleep; | |
424 | * otherwise it might trigger and wait for the whole process of sending | |
425 | * an context-destroying rpc to server. | |
426 | */ | |
427 | void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync) | |
428 | { | |
d7e09d03 PT |
429 | LASSERT(req); |
430 | LASSERT(req->rq_cli_ctx); | |
431 | ||
432 | /* request might be asked to release earlier while still | |
433 | * in the context waiting list. | |
434 | */ | |
435 | if (!list_empty(&req->rq_ctx_chain)) { | |
436 | spin_lock(&req->rq_cli_ctx->cc_lock); | |
437 | list_del_init(&req->rq_ctx_chain); | |
438 | spin_unlock(&req->rq_cli_ctx->cc_lock); | |
439 | } | |
440 | ||
441 | sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync); | |
442 | req->rq_cli_ctx = NULL; | |
d7e09d03 PT |
443 | } |
444 | ||
445 | static | |
446 | int sptlrpc_req_ctx_switch(struct ptlrpc_request *req, | |
447 | struct ptlrpc_cli_ctx *oldctx, | |
448 | struct ptlrpc_cli_ctx *newctx) | |
449 | { | |
d0bfef31 CH |
450 | struct sptlrpc_flavor old_flvr; |
451 | char *reqmsg = NULL; /* to workaround old gcc */ | |
452 | int reqmsg_size; | |
453 | int rc = 0; | |
d7e09d03 PT |
454 | |
455 | LASSERT(req->rq_reqmsg); | |
456 | LASSERT(req->rq_reqlen); | |
457 | LASSERT(req->rq_replen); | |
458 | ||
2d00bd17 JP |
459 | CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n", |
460 | req, | |
d7e09d03 PT |
461 | oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec), |
462 | newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec), | |
463 | oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name, | |
464 | newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name); | |
465 | ||
466 | /* save flavor */ | |
467 | old_flvr = req->rq_flvr; | |
468 | ||
469 | /* save request message */ | |
470 | reqmsg_size = req->rq_reqlen; | |
471 | if (reqmsg_size != 0) { | |
472 | OBD_ALLOC_LARGE(reqmsg, reqmsg_size); | |
473 | if (reqmsg == NULL) | |
474 | return -ENOMEM; | |
475 | memcpy(reqmsg, req->rq_reqmsg, reqmsg_size); | |
476 | } | |
477 | ||
478 | /* release old req/rep buf */ | |
479 | req->rq_cli_ctx = oldctx; | |
480 | sptlrpc_cli_free_reqbuf(req); | |
481 | sptlrpc_cli_free_repbuf(req); | |
482 | req->rq_cli_ctx = newctx; | |
483 | ||
484 | /* recalculate the flavor */ | |
485 | sptlrpc_req_set_flavor(req, 0); | |
486 | ||
487 | /* alloc new request buffer | |
488 | * we don't need to alloc reply buffer here, leave it to the | |
489 | * rest procedure of ptlrpc */ | |
490 | if (reqmsg_size != 0) { | |
491 | rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size); | |
492 | if (!rc) { | |
493 | LASSERT(req->rq_reqmsg); | |
494 | memcpy(req->rq_reqmsg, reqmsg, reqmsg_size); | |
495 | } else { | |
496 | CWARN("failed to alloc reqbuf: %d\n", rc); | |
497 | req->rq_flvr = old_flvr; | |
498 | } | |
499 | ||
500 | OBD_FREE_LARGE(reqmsg, reqmsg_size); | |
501 | } | |
502 | return rc; | |
503 | } | |
504 | ||
505 | /** | |
506 | * If current context of \a req is dead somehow, e.g. we just switched flavor | |
507 | * thus marked original contexts dead, we'll find a new context for it. if | |
508 | * no switch is needed, \a req will end up with the same context. | |
509 | * | |
510 | * \note a request must have a context, to keep other parts of code happy. | |
511 | * In any case of failure during the switching, we must restore the old one. | |
512 | */ | |
513 | int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req) | |
514 | { | |
515 | struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx; | |
516 | struct ptlrpc_cli_ctx *newctx; | |
d0bfef31 | 517 | int rc; |
d7e09d03 PT |
518 | |
519 | LASSERT(oldctx); | |
520 | ||
521 | sptlrpc_cli_ctx_get(oldctx); | |
522 | sptlrpc_req_put_ctx(req, 0); | |
523 | ||
524 | rc = sptlrpc_req_get_ctx(req); | |
525 | if (unlikely(rc)) { | |
526 | LASSERT(!req->rq_cli_ctx); | |
527 | ||
528 | /* restore old ctx */ | |
529 | req->rq_cli_ctx = oldctx; | |
0a3bdb00 | 530 | return rc; |
d7e09d03 PT |
531 | } |
532 | ||
533 | newctx = req->rq_cli_ctx; | |
534 | LASSERT(newctx); | |
535 | ||
536 | if (unlikely(newctx == oldctx && | |
537 | test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) { | |
538 | /* | |
539 | * still get the old dead ctx, usually means system too busy | |
540 | */ | |
541 | CDEBUG(D_SEC, | |
542 | "ctx (%p, fl %lx) doesn't switch, relax a little bit\n", | |
543 | newctx, newctx->cc_flags); | |
544 | ||
18fd5baa PT |
545 | set_current_state(TASK_INTERRUPTIBLE); |
546 | schedule_timeout(HZ); | |
d7e09d03 PT |
547 | } else { |
548 | /* | |
549 | * it's possible newctx == oldctx if we're switching | |
550 | * subflavor with the same sec. | |
551 | */ | |
552 | rc = sptlrpc_req_ctx_switch(req, oldctx, newctx); | |
553 | if (rc) { | |
554 | /* restore old ctx */ | |
555 | sptlrpc_req_put_ctx(req, 0); | |
556 | req->rq_cli_ctx = oldctx; | |
0a3bdb00 | 557 | return rc; |
d7e09d03 PT |
558 | } |
559 | ||
560 | LASSERT(req->rq_cli_ctx == newctx); | |
561 | } | |
562 | ||
563 | sptlrpc_cli_ctx_put(oldctx, 1); | |
0a3bdb00 | 564 | return 0; |
d7e09d03 PT |
565 | } |
566 | EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx); | |
567 | ||
568 | static | |
569 | int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx) | |
570 | { | |
571 | if (cli_ctx_is_refreshed(ctx)) | |
572 | return 1; | |
573 | return 0; | |
574 | } | |
575 | ||
576 | static | |
577 | int ctx_refresh_timeout(void *data) | |
578 | { | |
579 | struct ptlrpc_request *req = data; | |
580 | int rc; | |
581 | ||
582 | /* conn_cnt is needed in expire_one_request */ | |
583 | lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt); | |
584 | ||
585 | rc = ptlrpc_expire_one_request(req, 1); | |
586 | /* if we started recovery, we should mark this ctx dead; otherwise | |
587 | * in case of lgssd died nobody would retire this ctx, following | |
588 | * connecting will still find the same ctx thus cause deadlock. | |
589 | * there's an assumption that expire time of the request should be | |
590 | * later than the context refresh expire time. | |
591 | */ | |
592 | if (rc == 0) | |
d0f17b64 | 593 | req->rq_cli_ctx->cc_ops->force_die(req->rq_cli_ctx, 0); |
d7e09d03 PT |
594 | return rc; |
595 | } | |
596 | ||
597 | static | |
598 | void ctx_refresh_interrupt(void *data) | |
599 | { | |
600 | struct ptlrpc_request *req = data; | |
601 | ||
602 | spin_lock(&req->rq_lock); | |
603 | req->rq_intr = 1; | |
604 | spin_unlock(&req->rq_lock); | |
605 | } | |
606 | ||
607 | static | |
608 | void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx) | |
609 | { | |
610 | spin_lock(&ctx->cc_lock); | |
611 | if (!list_empty(&req->rq_ctx_chain)) | |
612 | list_del_init(&req->rq_ctx_chain); | |
613 | spin_unlock(&ctx->cc_lock); | |
614 | } | |
615 | ||
616 | /** | |
617 | * To refresh the context of \req, if it's not up-to-date. | |
618 | * \param timeout | |
619 | * - < 0: don't wait | |
620 | * - = 0: wait until success or fatal error occur | |
621 | * - > 0: timeout value (in seconds) | |
622 | * | |
623 | * The status of the context could be subject to be changed by other threads | |
624 | * at any time. We allow this race, but once we return with 0, the caller will | |
625 | * suppose it's uptodated and keep using it until the owning rpc is done. | |
626 | * | |
627 | * \retval 0 only if the context is uptodated. | |
628 | * \retval -ev error number. | |
629 | */ | |
630 | int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout) | |
631 | { | |
d0bfef31 CH |
632 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; |
633 | struct ptlrpc_sec *sec; | |
634 | struct l_wait_info lwi; | |
635 | int rc; | |
d7e09d03 PT |
636 | |
637 | LASSERT(ctx); | |
638 | ||
639 | if (req->rq_ctx_init || req->rq_ctx_fini) | |
0a3bdb00 | 640 | return 0; |
d7e09d03 PT |
641 | |
642 | /* | |
643 | * during the process a request's context might change type even | |
644 | * (e.g. from gss ctx to null ctx), so each loop we need to re-check | |
645 | * everything | |
646 | */ | |
647 | again: | |
648 | rc = import_sec_validate_get(req->rq_import, &sec); | |
649 | if (rc) | |
0a3bdb00 | 650 | return rc; |
d7e09d03 PT |
651 | |
652 | if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) { | |
653 | CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n", | |
654 | req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc); | |
655 | req_off_ctx_list(req, ctx); | |
656 | sptlrpc_req_replace_dead_ctx(req); | |
657 | ctx = req->rq_cli_ctx; | |
658 | } | |
659 | sptlrpc_sec_put(sec); | |
660 | ||
661 | if (cli_ctx_is_eternal(ctx)) | |
0a3bdb00 | 662 | return 0; |
d7e09d03 PT |
663 | |
664 | if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) { | |
665 | LASSERT(ctx->cc_ops->refresh); | |
666 | ctx->cc_ops->refresh(ctx); | |
667 | } | |
668 | LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0); | |
669 | ||
670 | LASSERT(ctx->cc_ops->validate); | |
671 | if (ctx->cc_ops->validate(ctx) == 0) { | |
672 | req_off_ctx_list(req, ctx); | |
0a3bdb00 | 673 | return 0; |
d7e09d03 PT |
674 | } |
675 | ||
676 | if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) { | |
677 | spin_lock(&req->rq_lock); | |
678 | req->rq_err = 1; | |
679 | spin_unlock(&req->rq_lock); | |
680 | req_off_ctx_list(req, ctx); | |
0a3bdb00 | 681 | return -EPERM; |
d7e09d03 PT |
682 | } |
683 | ||
684 | /* | |
685 | * There's a subtle issue for resending RPCs, suppose following | |
686 | * situation: | |
687 | * 1. the request was sent to server. | |
688 | * 2. recovery was kicked start, after finished the request was | |
689 | * marked as resent. | |
690 | * 3. resend the request. | |
691 | * 4. old reply from server received, we accept and verify the reply. | |
692 | * this has to be success, otherwise the error will be aware | |
693 | * by application. | |
694 | * 5. new reply from server received, dropped by LNet. | |
695 | * | |
696 | * Note the xid of old & new request is the same. We can't simply | |
697 | * change xid for the resent request because the server replies on | |
698 | * it for reply reconstruction. | |
699 | * | |
700 | * Commonly the original context should be uptodate because we | |
701 | * have a expiry nice time; server will keep its context because | |
702 | * we at least hold a ref of old context which prevent context | |
703 | * destroying RPC being sent. So server still can accept the request | |
704 | * and finish the RPC. But if that's not the case: | |
705 | * 1. If server side context has been trimmed, a NO_CONTEXT will | |
706 | * be returned, gss_cli_ctx_verify/unseal will switch to new | |
707 | * context by force. | |
708 | * 2. Current context never be refreshed, then we are fine: we | |
709 | * never really send request with old context before. | |
710 | */ | |
711 | if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) && | |
712 | unlikely(req->rq_reqmsg) && | |
713 | lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { | |
714 | req_off_ctx_list(req, ctx); | |
0a3bdb00 | 715 | return 0; |
d7e09d03 PT |
716 | } |
717 | ||
718 | if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) { | |
719 | req_off_ctx_list(req, ctx); | |
720 | /* | |
721 | * don't switch ctx if import was deactivated | |
722 | */ | |
723 | if (req->rq_import->imp_deactive) { | |
724 | spin_lock(&req->rq_lock); | |
725 | req->rq_err = 1; | |
726 | spin_unlock(&req->rq_lock); | |
0a3bdb00 | 727 | return -EINTR; |
d7e09d03 PT |
728 | } |
729 | ||
730 | rc = sptlrpc_req_replace_dead_ctx(req); | |
731 | if (rc) { | |
732 | LASSERT(ctx == req->rq_cli_ctx); | |
733 | CERROR("req %p: failed to replace dead ctx %p: %d\n", | |
734 | req, ctx, rc); | |
735 | spin_lock(&req->rq_lock); | |
736 | req->rq_err = 1; | |
737 | spin_unlock(&req->rq_lock); | |
0a3bdb00 | 738 | return rc; |
d7e09d03 PT |
739 | } |
740 | ||
741 | ctx = req->rq_cli_ctx; | |
742 | goto again; | |
743 | } | |
744 | ||
745 | /* | |
746 | * Now we're sure this context is during upcall, add myself into | |
747 | * waiting list | |
748 | */ | |
749 | spin_lock(&ctx->cc_lock); | |
750 | if (list_empty(&req->rq_ctx_chain)) | |
751 | list_add(&req->rq_ctx_chain, &ctx->cc_req_list); | |
752 | spin_unlock(&ctx->cc_lock); | |
753 | ||
754 | if (timeout < 0) | |
0a3bdb00 | 755 | return -EWOULDBLOCK; |
d7e09d03 PT |
756 | |
757 | /* Clear any flags that may be present from previous sends */ | |
758 | LASSERT(req->rq_receiving_reply == 0); | |
759 | spin_lock(&req->rq_lock); | |
760 | req->rq_err = 0; | |
761 | req->rq_timedout = 0; | |
762 | req->rq_resend = 0; | |
763 | req->rq_restart = 0; | |
764 | spin_unlock(&req->rq_lock); | |
765 | ||
766 | lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout, | |
767 | ctx_refresh_interrupt, req); | |
768 | rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi); | |
769 | ||
770 | /* | |
771 | * following cases could lead us here: | |
772 | * - successfully refreshed; | |
773 | * - interrupted; | |
774 | * - timedout, and we don't want recover from the failure; | |
775 | * - timedout, and waked up upon recovery finished; | |
776 | * - someone else mark this ctx dead by force; | |
777 | * - someone invalidate the req and call ptlrpc_client_wake_req(), | |
778 | * e.g. ptlrpc_abort_inflight(); | |
779 | */ | |
780 | if (!cli_ctx_is_refreshed(ctx)) { | |
0c2bc758 | 781 | /* timed out or interrupted */ |
d7e09d03 PT |
782 | req_off_ctx_list(req, ctx); |
783 | ||
784 | LASSERT(rc != 0); | |
0a3bdb00 | 785 | return rc; |
d7e09d03 PT |
786 | } |
787 | ||
788 | goto again; | |
789 | } | |
790 | ||
791 | /** | |
792 | * Initialize flavor settings for \a req, according to \a opcode. | |
793 | * | |
794 | * \note this could be called in two situations: | |
795 | * - new request from ptlrpc_pre_req(), with proper @opcode | |
796 | * - old request which changed ctx in the middle, with @opcode == 0 | |
797 | */ | |
798 | void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode) | |
799 | { | |
800 | struct ptlrpc_sec *sec; | |
801 | ||
802 | LASSERT(req->rq_import); | |
803 | LASSERT(req->rq_cli_ctx); | |
804 | LASSERT(req->rq_cli_ctx->cc_sec); | |
805 | LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0); | |
806 | ||
0c2bc758 | 807 | /* special security flags according to opcode */ |
d7e09d03 PT |
808 | switch (opcode) { |
809 | case OST_READ: | |
810 | case MDS_READPAGE: | |
811 | case MGS_CONFIG_READ: | |
812 | case OBD_IDX_READ: | |
813 | req->rq_bulk_read = 1; | |
814 | break; | |
815 | case OST_WRITE: | |
816 | case MDS_WRITEPAGE: | |
817 | req->rq_bulk_write = 1; | |
818 | break; | |
819 | case SEC_CTX_INIT: | |
820 | req->rq_ctx_init = 1; | |
821 | break; | |
822 | case SEC_CTX_FINI: | |
823 | req->rq_ctx_fini = 1; | |
824 | break; | |
825 | case 0: | |
826 | /* init/fini rpc won't be resend, so can't be here */ | |
827 | LASSERT(req->rq_ctx_init == 0); | |
828 | LASSERT(req->rq_ctx_fini == 0); | |
829 | ||
830 | /* cleanup flags, which should be recalculated */ | |
831 | req->rq_pack_udesc = 0; | |
832 | req->rq_pack_bulk = 0; | |
833 | break; | |
834 | } | |
835 | ||
836 | sec = req->rq_cli_ctx->cc_sec; | |
837 | ||
838 | spin_lock(&sec->ps_lock); | |
839 | req->rq_flvr = sec->ps_flvr; | |
840 | spin_unlock(&sec->ps_lock); | |
841 | ||
842 | /* force SVC_NULL for context initiation rpc, SVC_INTG for context | |
843 | * destruction rpc */ | |
844 | if (unlikely(req->rq_ctx_init)) | |
845 | flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL); | |
846 | else if (unlikely(req->rq_ctx_fini)) | |
847 | flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG); | |
848 | ||
849 | /* user descriptor flag, null security can't do it anyway */ | |
850 | if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) && | |
851 | (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL)) | |
852 | req->rq_pack_udesc = 1; | |
853 | ||
854 | /* bulk security flag */ | |
855 | if ((req->rq_bulk_read || req->rq_bulk_write) && | |
856 | sptlrpc_flavor_has_bulk(&req->rq_flvr)) | |
857 | req->rq_pack_bulk = 1; | |
858 | } | |
859 | ||
860 | void sptlrpc_request_out_callback(struct ptlrpc_request *req) | |
861 | { | |
862 | if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV) | |
863 | return; | |
864 | ||
865 | LASSERT(req->rq_clrbuf); | |
866 | if (req->rq_pool || !req->rq_reqbuf) | |
867 | return; | |
868 | ||
9ae10597 | 869 | kfree(req->rq_reqbuf); |
d7e09d03 PT |
870 | req->rq_reqbuf = NULL; |
871 | req->rq_reqbuf_len = 0; | |
872 | } | |
873 | ||
874 | /** | |
875 | * Given an import \a imp, check whether current user has a valid context | |
876 | * or not. We may create a new context and try to refresh it, and try | |
877 | * repeatedly try in case of non-fatal errors. Return 0 means success. | |
878 | */ | |
879 | int sptlrpc_import_check_ctx(struct obd_import *imp) | |
880 | { | |
d0bfef31 | 881 | struct ptlrpc_sec *sec; |
d7e09d03 PT |
882 | struct ptlrpc_cli_ctx *ctx; |
883 | struct ptlrpc_request *req = NULL; | |
884 | int rc; | |
d7e09d03 PT |
885 | |
886 | might_sleep(); | |
887 | ||
888 | sec = sptlrpc_import_sec_ref(imp); | |
889 | ctx = get_my_ctx(sec); | |
890 | sptlrpc_sec_put(sec); | |
891 | ||
892 | if (!ctx) | |
0a3bdb00 | 893 | return -ENOMEM; |
d7e09d03 PT |
894 | |
895 | if (cli_ctx_is_eternal(ctx) || | |
896 | ctx->cc_ops->validate(ctx) == 0) { | |
897 | sptlrpc_cli_ctx_put(ctx, 1); | |
0a3bdb00 | 898 | return 0; |
d7e09d03 PT |
899 | } |
900 | ||
901 | if (cli_ctx_is_error(ctx)) { | |
902 | sptlrpc_cli_ctx_put(ctx, 1); | |
0a3bdb00 | 903 | return -EACCES; |
d7e09d03 PT |
904 | } |
905 | ||
0be19afa | 906 | req = ptlrpc_request_cache_alloc(GFP_NOFS); |
d7e09d03 | 907 | if (!req) |
0a3bdb00 | 908 | return -ENOMEM; |
d7e09d03 PT |
909 | |
910 | spin_lock_init(&req->rq_lock); | |
911 | atomic_set(&req->rq_refcount, 10000); | |
912 | INIT_LIST_HEAD(&req->rq_ctx_chain); | |
913 | init_waitqueue_head(&req->rq_reply_waitq); | |
914 | init_waitqueue_head(&req->rq_set_waitq); | |
915 | req->rq_import = imp; | |
916 | req->rq_flvr = sec->ps_flvr; | |
917 | req->rq_cli_ctx = ctx; | |
918 | ||
919 | rc = sptlrpc_req_refresh_ctx(req, 0); | |
920 | LASSERT(list_empty(&req->rq_ctx_chain)); | |
921 | sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1); | |
35b2e1b7 | 922 | ptlrpc_request_cache_free(req); |
d7e09d03 | 923 | |
0a3bdb00 | 924 | return rc; |
d7e09d03 PT |
925 | } |
926 | ||
927 | /** | |
928 | * Used by ptlrpc client, to perform the pre-defined security transformation | |
929 | * upon the request message of \a req. After this function called, | |
930 | * req->rq_reqmsg is still accessible as clear text. | |
931 | */ | |
932 | int sptlrpc_cli_wrap_request(struct ptlrpc_request *req) | |
933 | { | |
934 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
935 | int rc = 0; | |
d7e09d03 PT |
936 | |
937 | LASSERT(ctx); | |
938 | LASSERT(ctx->cc_sec); | |
939 | LASSERT(req->rq_reqbuf || req->rq_clrbuf); | |
940 | ||
941 | /* we wrap bulk request here because now we can be sure | |
942 | * the context is uptodate. | |
943 | */ | |
944 | if (req->rq_bulk) { | |
945 | rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk); | |
946 | if (rc) | |
0a3bdb00 | 947 | return rc; |
d7e09d03 PT |
948 | } |
949 | ||
950 | switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) { | |
951 | case SPTLRPC_SVC_NULL: | |
952 | case SPTLRPC_SVC_AUTH: | |
953 | case SPTLRPC_SVC_INTG: | |
954 | LASSERT(ctx->cc_ops->sign); | |
955 | rc = ctx->cc_ops->sign(ctx, req); | |
956 | break; | |
957 | case SPTLRPC_SVC_PRIV: | |
958 | LASSERT(ctx->cc_ops->seal); | |
959 | rc = ctx->cc_ops->seal(ctx, req); | |
960 | break; | |
961 | default: | |
962 | LBUG(); | |
963 | } | |
964 | ||
965 | if (rc == 0) { | |
966 | LASSERT(req->rq_reqdata_len); | |
967 | LASSERT(req->rq_reqdata_len % 8 == 0); | |
968 | LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len); | |
969 | } | |
970 | ||
0a3bdb00 | 971 | return rc; |
d7e09d03 PT |
972 | } |
973 | ||
974 | static int do_cli_unwrap_reply(struct ptlrpc_request *req) | |
975 | { | |
976 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
d0bfef31 | 977 | int rc; |
d7e09d03 PT |
978 | |
979 | LASSERT(ctx); | |
980 | LASSERT(ctx->cc_sec); | |
981 | LASSERT(req->rq_repbuf); | |
982 | LASSERT(req->rq_repdata); | |
983 | LASSERT(req->rq_repmsg == NULL); | |
984 | ||
985 | req->rq_rep_swab_mask = 0; | |
986 | ||
987 | rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len); | |
988 | switch (rc) { | |
989 | case 1: | |
990 | lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF); | |
991 | case 0: | |
992 | break; | |
993 | default: | |
b0f5aad5 | 994 | CERROR("failed unpack reply: x%llu\n", req->rq_xid); |
0a3bdb00 | 995 | return -EPROTO; |
d7e09d03 PT |
996 | } |
997 | ||
998 | if (req->rq_repdata_len < sizeof(struct lustre_msg)) { | |
999 | CERROR("replied data length %d too small\n", | |
1000 | req->rq_repdata_len); | |
0a3bdb00 | 1001 | return -EPROTO; |
d7e09d03 PT |
1002 | } |
1003 | ||
1004 | if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) != | |
1005 | SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) { | |
1006 | CERROR("reply policy %u doesn't match request policy %u\n", | |
1007 | SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr), | |
1008 | SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)); | |
0a3bdb00 | 1009 | return -EPROTO; |
d7e09d03 PT |
1010 | } |
1011 | ||
1012 | switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) { | |
1013 | case SPTLRPC_SVC_NULL: | |
1014 | case SPTLRPC_SVC_AUTH: | |
1015 | case SPTLRPC_SVC_INTG: | |
1016 | LASSERT(ctx->cc_ops->verify); | |
1017 | rc = ctx->cc_ops->verify(ctx, req); | |
1018 | break; | |
1019 | case SPTLRPC_SVC_PRIV: | |
1020 | LASSERT(ctx->cc_ops->unseal); | |
1021 | rc = ctx->cc_ops->unseal(ctx, req); | |
1022 | break; | |
1023 | default: | |
1024 | LBUG(); | |
1025 | } | |
1026 | LASSERT(rc || req->rq_repmsg || req->rq_resend); | |
1027 | ||
1028 | if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL && | |
1029 | !req->rq_ctx_init) | |
1030 | req->rq_rep_swab_mask = 0; | |
0a3bdb00 | 1031 | return rc; |
d7e09d03 PT |
1032 | } |
1033 | ||
1034 | /** | |
1035 | * Used by ptlrpc client, to perform security transformation upon the reply | |
1036 | * message of \a req. After return successfully, req->rq_repmsg points to | |
1037 | * the reply message in clear text. | |
1038 | * | |
1039 | * \pre the reply buffer should have been un-posted from LNet, so nothing is | |
1040 | * going to change. | |
1041 | */ | |
1042 | int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req) | |
1043 | { | |
1044 | LASSERT(req->rq_repbuf); | |
1045 | LASSERT(req->rq_repdata == NULL); | |
1046 | LASSERT(req->rq_repmsg == NULL); | |
1047 | LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len); | |
1048 | ||
1049 | if (req->rq_reply_off == 0 && | |
1050 | (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) { | |
1051 | CERROR("real reply with offset 0\n"); | |
1052 | return -EPROTO; | |
1053 | } | |
1054 | ||
1055 | if (req->rq_reply_off % 8 != 0) { | |
1056 | CERROR("reply at odd offset %u\n", req->rq_reply_off); | |
1057 | return -EPROTO; | |
1058 | } | |
1059 | ||
1060 | req->rq_repdata = (struct lustre_msg *) | |
1061 | (req->rq_repbuf + req->rq_reply_off); | |
1062 | req->rq_repdata_len = req->rq_nob_received; | |
1063 | ||
1064 | return do_cli_unwrap_reply(req); | |
1065 | } | |
1066 | ||
1067 | /** | |
1068 | * Used by ptlrpc client, to perform security transformation upon the early | |
1069 | * reply message of \a req. We expect the rq_reply_off is 0, and | |
1070 | * rq_nob_received is the early reply size. | |
1071 | * | |
1072 | * Because the receive buffer might be still posted, the reply data might be | |
1073 | * changed at any time, no matter we're holding rq_lock or not. For this reason | |
1074 | * we allocate a separate ptlrpc_request and reply buffer for early reply | |
1075 | * processing. | |
1076 | * | |
1077 | * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request. | |
1078 | * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned | |
1079 | * \a *req_ret to release it. | |
1080 | * \retval -ev error number, and \a req_ret will not be set. | |
1081 | */ | |
1082 | int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req, | |
1083 | struct ptlrpc_request **req_ret) | |
1084 | { | |
d0bfef31 CH |
1085 | struct ptlrpc_request *early_req; |
1086 | char *early_buf; | |
1087 | int early_bufsz, early_size; | |
1088 | int rc; | |
d7e09d03 | 1089 | |
0be19afa | 1090 | early_req = ptlrpc_request_cache_alloc(GFP_NOFS); |
d7e09d03 | 1091 | if (early_req == NULL) |
0a3bdb00 | 1092 | return -ENOMEM; |
d7e09d03 PT |
1093 | |
1094 | early_size = req->rq_nob_received; | |
1095 | early_bufsz = size_roundup_power2(early_size); | |
1096 | OBD_ALLOC_LARGE(early_buf, early_bufsz); | |
a9b3e8f3 JL |
1097 | if (early_buf == NULL) { |
1098 | rc = -ENOMEM; | |
1099 | goto err_req; | |
1100 | } | |
d7e09d03 PT |
1101 | |
1102 | /* sanity checkings and copy data out, do it inside spinlock */ | |
1103 | spin_lock(&req->rq_lock); | |
1104 | ||
1105 | if (req->rq_replied) { | |
1106 | spin_unlock(&req->rq_lock); | |
a9b3e8f3 JL |
1107 | rc = -EALREADY; |
1108 | goto err_buf; | |
d7e09d03 PT |
1109 | } |
1110 | ||
1111 | LASSERT(req->rq_repbuf); | |
1112 | LASSERT(req->rq_repdata == NULL); | |
1113 | LASSERT(req->rq_repmsg == NULL); | |
1114 | ||
1115 | if (req->rq_reply_off != 0) { | |
1116 | CERROR("early reply with offset %u\n", req->rq_reply_off); | |
1117 | spin_unlock(&req->rq_lock); | |
a9b3e8f3 JL |
1118 | rc = -EPROTO; |
1119 | goto err_buf; | |
d7e09d03 PT |
1120 | } |
1121 | ||
1122 | if (req->rq_nob_received != early_size) { | |
1123 | /* even another early arrived the size should be the same */ | |
1124 | CERROR("data size has changed from %u to %u\n", | |
1125 | early_size, req->rq_nob_received); | |
1126 | spin_unlock(&req->rq_lock); | |
a9b3e8f3 JL |
1127 | rc = -EINVAL; |
1128 | goto err_buf; | |
d7e09d03 PT |
1129 | } |
1130 | ||
1131 | if (req->rq_nob_received < sizeof(struct lustre_msg)) { | |
1132 | CERROR("early reply length %d too small\n", | |
1133 | req->rq_nob_received); | |
1134 | spin_unlock(&req->rq_lock); | |
a9b3e8f3 JL |
1135 | rc = -EALREADY; |
1136 | goto err_buf; | |
d7e09d03 PT |
1137 | } |
1138 | ||
1139 | memcpy(early_buf, req->rq_repbuf, early_size); | |
1140 | spin_unlock(&req->rq_lock); | |
1141 | ||
1142 | spin_lock_init(&early_req->rq_lock); | |
1143 | early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx); | |
1144 | early_req->rq_flvr = req->rq_flvr; | |
1145 | early_req->rq_repbuf = early_buf; | |
1146 | early_req->rq_repbuf_len = early_bufsz; | |
1147 | early_req->rq_repdata = (struct lustre_msg *) early_buf; | |
1148 | early_req->rq_repdata_len = early_size; | |
1149 | early_req->rq_early = 1; | |
1150 | early_req->rq_reqmsg = req->rq_reqmsg; | |
1151 | ||
1152 | rc = do_cli_unwrap_reply(early_req); | |
1153 | if (rc) { | |
1154 | DEBUG_REQ(D_ADAPTTO, early_req, | |
1155 | "error %d unwrap early reply", rc); | |
a9b3e8f3 | 1156 | goto err_ctx; |
d7e09d03 PT |
1157 | } |
1158 | ||
1159 | LASSERT(early_req->rq_repmsg); | |
1160 | *req_ret = early_req; | |
0a3bdb00 | 1161 | return 0; |
d7e09d03 PT |
1162 | |
1163 | err_ctx: | |
1164 | sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1); | |
1165 | err_buf: | |
1166 | OBD_FREE_LARGE(early_buf, early_bufsz); | |
1167 | err_req: | |
35b2e1b7 | 1168 | ptlrpc_request_cache_free(early_req); |
0a3bdb00 | 1169 | return rc; |
d7e09d03 PT |
1170 | } |
1171 | ||
1172 | /** | |
1173 | * Used by ptlrpc client, to release a processed early reply \a early_req. | |
1174 | * | |
1175 | * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply(). | |
1176 | */ | |
1177 | void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req) | |
1178 | { | |
1179 | LASSERT(early_req->rq_repbuf); | |
1180 | LASSERT(early_req->rq_repdata); | |
1181 | LASSERT(early_req->rq_repmsg); | |
1182 | ||
1183 | sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1); | |
1184 | OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len); | |
35b2e1b7 | 1185 | ptlrpc_request_cache_free(early_req); |
d7e09d03 PT |
1186 | } |
1187 | ||
1188 | /************************************************** | |
1189 | * sec ID * | |
1190 | **************************************************/ | |
1191 | ||
1192 | /* | |
1193 | * "fixed" sec (e.g. null) use sec_id < 0 | |
1194 | */ | |
1195 | static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1); | |
1196 | ||
1197 | int sptlrpc_get_next_secid(void) | |
1198 | { | |
1199 | return atomic_inc_return(&sptlrpc_sec_id); | |
1200 | } | |
1201 | EXPORT_SYMBOL(sptlrpc_get_next_secid); | |
1202 | ||
1203 | /************************************************** | |
1204 | * client side high-level security APIs * | |
1205 | **************************************************/ | |
1206 | ||
1207 | static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid, | |
1208 | int grace, int force) | |
1209 | { | |
1210 | struct ptlrpc_sec_policy *policy = sec->ps_policy; | |
1211 | ||
1212 | LASSERT(policy->sp_cops); | |
1213 | LASSERT(policy->sp_cops->flush_ctx_cache); | |
1214 | ||
1215 | return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force); | |
1216 | } | |
1217 | ||
1218 | static void sec_cop_destroy_sec(struct ptlrpc_sec *sec) | |
1219 | { | |
1220 | struct ptlrpc_sec_policy *policy = sec->ps_policy; | |
1221 | ||
1222 | LASSERT_ATOMIC_ZERO(&sec->ps_refcount); | |
1223 | LASSERT_ATOMIC_ZERO(&sec->ps_nctx); | |
1224 | LASSERT(policy->sp_cops->destroy_sec); | |
1225 | ||
0c2bc758 | 1226 | CDEBUG(D_SEC, "%s@%p: being destroyed\n", sec->ps_policy->sp_name, sec); |
d7e09d03 PT |
1227 | |
1228 | policy->sp_cops->destroy_sec(sec); | |
1229 | sptlrpc_policy_put(policy); | |
1230 | } | |
1231 | ||
1232 | void sptlrpc_sec_destroy(struct ptlrpc_sec *sec) | |
1233 | { | |
1234 | sec_cop_destroy_sec(sec); | |
1235 | } | |
1236 | EXPORT_SYMBOL(sptlrpc_sec_destroy); | |
1237 | ||
1238 | static void sptlrpc_sec_kill(struct ptlrpc_sec *sec) | |
1239 | { | |
1240 | LASSERT_ATOMIC_POS(&sec->ps_refcount); | |
1241 | ||
1242 | if (sec->ps_policy->sp_cops->kill_sec) { | |
1243 | sec->ps_policy->sp_cops->kill_sec(sec); | |
1244 | ||
1245 | sec_cop_flush_ctx_cache(sec, -1, 1, 1); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec) | |
1250 | { | |
1251 | if (sec) | |
1252 | atomic_inc(&sec->ps_refcount); | |
1253 | ||
1254 | return sec; | |
1255 | } | |
1256 | EXPORT_SYMBOL(sptlrpc_sec_get); | |
1257 | ||
1258 | void sptlrpc_sec_put(struct ptlrpc_sec *sec) | |
1259 | { | |
1260 | if (sec) { | |
1261 | LASSERT_ATOMIC_POS(&sec->ps_refcount); | |
1262 | ||
1263 | if (atomic_dec_and_test(&sec->ps_refcount)) { | |
1264 | sptlrpc_gc_del_sec(sec); | |
1265 | sec_cop_destroy_sec(sec); | |
1266 | } | |
1267 | } | |
1268 | } | |
1269 | EXPORT_SYMBOL(sptlrpc_sec_put); | |
1270 | ||
1271 | /* | |
0c2bc758 | 1272 | * policy module is responsible for taking reference of import |
d7e09d03 PT |
1273 | */ |
1274 | static | |
aff9d8e8 | 1275 | struct ptlrpc_sec *sptlrpc_sec_create(struct obd_import *imp, |
d0bfef31 CH |
1276 | struct ptlrpc_svc_ctx *svc_ctx, |
1277 | struct sptlrpc_flavor *sf, | |
1278 | enum lustre_sec_part sp) | |
d7e09d03 PT |
1279 | { |
1280 | struct ptlrpc_sec_policy *policy; | |
d0bfef31 CH |
1281 | struct ptlrpc_sec *sec; |
1282 | char str[32]; | |
d7e09d03 PT |
1283 | |
1284 | if (svc_ctx) { | |
1285 | LASSERT(imp->imp_dlm_fake == 1); | |
1286 | ||
1287 | CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n", | |
1288 | imp->imp_obd->obd_type->typ_name, | |
1289 | imp->imp_obd->obd_name, | |
1290 | sptlrpc_flavor2name(sf, str, sizeof(str))); | |
1291 | ||
1292 | policy = sptlrpc_policy_get(svc_ctx->sc_policy); | |
1293 | sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY; | |
1294 | } else { | |
1295 | LASSERT(imp->imp_dlm_fake == 0); | |
1296 | ||
1297 | CDEBUG(D_SEC, "%s %s: select security flavor %s\n", | |
1298 | imp->imp_obd->obd_type->typ_name, | |
1299 | imp->imp_obd->obd_name, | |
1300 | sptlrpc_flavor2name(sf, str, sizeof(str))); | |
1301 | ||
1302 | policy = sptlrpc_wireflavor2policy(sf->sf_rpc); | |
1303 | if (!policy) { | |
1304 | CERROR("invalid flavor 0x%x\n", sf->sf_rpc); | |
0a3bdb00 | 1305 | return NULL; |
d7e09d03 PT |
1306 | } |
1307 | } | |
1308 | ||
1309 | sec = policy->sp_cops->create_sec(imp, svc_ctx, sf); | |
1310 | if (sec) { | |
1311 | atomic_inc(&sec->ps_refcount); | |
1312 | ||
1313 | sec->ps_part = sp; | |
1314 | ||
1315 | if (sec->ps_gc_interval && policy->sp_cops->gc_ctx) | |
1316 | sptlrpc_gc_add_sec(sec); | |
1317 | } else { | |
1318 | sptlrpc_policy_put(policy); | |
1319 | } | |
1320 | ||
0a3bdb00 | 1321 | return sec; |
d7e09d03 PT |
1322 | } |
1323 | ||
1324 | struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp) | |
1325 | { | |
1326 | struct ptlrpc_sec *sec; | |
1327 | ||
1328 | spin_lock(&imp->imp_lock); | |
1329 | sec = sptlrpc_sec_get(imp->imp_sec); | |
1330 | spin_unlock(&imp->imp_lock); | |
1331 | ||
1332 | return sec; | |
1333 | } | |
1334 | EXPORT_SYMBOL(sptlrpc_import_sec_ref); | |
1335 | ||
1336 | static void sptlrpc_import_sec_install(struct obd_import *imp, | |
1337 | struct ptlrpc_sec *sec) | |
1338 | { | |
1339 | struct ptlrpc_sec *old_sec; | |
1340 | ||
1341 | LASSERT_ATOMIC_POS(&sec->ps_refcount); | |
1342 | ||
1343 | spin_lock(&imp->imp_lock); | |
1344 | old_sec = imp->imp_sec; | |
1345 | imp->imp_sec = sec; | |
1346 | spin_unlock(&imp->imp_lock); | |
1347 | ||
1348 | if (old_sec) { | |
1349 | sptlrpc_sec_kill(old_sec); | |
1350 | ||
1351 | /* balance the ref taken by this import */ | |
1352 | sptlrpc_sec_put(old_sec); | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | static inline | |
1357 | int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2) | |
1358 | { | |
1359 | return (memcmp(sf1, sf2, sizeof(*sf1)) == 0); | |
1360 | } | |
1361 | ||
1362 | static inline | |
1363 | void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src) | |
1364 | { | |
1365 | *dst = *src; | |
1366 | } | |
1367 | ||
1368 | static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp, | |
1369 | struct ptlrpc_sec *sec, | |
1370 | struct sptlrpc_flavor *sf) | |
1371 | { | |
d0bfef31 | 1372 | char str1[32], str2[32]; |
d7e09d03 PT |
1373 | |
1374 | if (sec->ps_flvr.sf_flags != sf->sf_flags) | |
1375 | CDEBUG(D_SEC, "changing sec flags: %s -> %s\n", | |
1376 | sptlrpc_secflags2str(sec->ps_flvr.sf_flags, | |
1377 | str1, sizeof(str1)), | |
1378 | sptlrpc_secflags2str(sf->sf_flags, | |
1379 | str2, sizeof(str2))); | |
1380 | ||
1381 | spin_lock(&sec->ps_lock); | |
1382 | flavor_copy(&sec->ps_flvr, sf); | |
1383 | spin_unlock(&sec->ps_lock); | |
1384 | } | |
1385 | ||
1386 | /** | |
1387 | * To get an appropriate ptlrpc_sec for the \a imp, according to the current | |
1388 | * configuration. Upon called, imp->imp_sec may or may not be NULL. | |
1389 | * | |
1390 | * - regular import: \a svc_ctx should be NULL and \a flvr is ignored; | |
1391 | * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request. | |
1392 | */ | |
1393 | int sptlrpc_import_sec_adapt(struct obd_import *imp, | |
1394 | struct ptlrpc_svc_ctx *svc_ctx, | |
1395 | struct sptlrpc_flavor *flvr) | |
1396 | { | |
d0bfef31 CH |
1397 | struct ptlrpc_connection *conn; |
1398 | struct sptlrpc_flavor sf; | |
1399 | struct ptlrpc_sec *sec, *newsec; | |
1400 | enum lustre_sec_part sp; | |
1401 | char str[24]; | |
1402 | int rc = 0; | |
d7e09d03 PT |
1403 | |
1404 | might_sleep(); | |
1405 | ||
1406 | if (imp == NULL) | |
0a3bdb00 | 1407 | return 0; |
d7e09d03 PT |
1408 | |
1409 | conn = imp->imp_connection; | |
1410 | ||
1411 | if (svc_ctx == NULL) { | |
1412 | struct client_obd *cliobd = &imp->imp_obd->u.cli; | |
1413 | /* | |
1414 | * normal import, determine flavor from rule set, except | |
1415 | * for mgc the flavor is predetermined. | |
1416 | */ | |
1417 | if (cliobd->cl_sp_me == LUSTRE_SP_MGC) | |
1418 | sf = cliobd->cl_flvr_mgc; | |
1419 | else | |
1420 | sptlrpc_conf_choose_flavor(cliobd->cl_sp_me, | |
1421 | cliobd->cl_sp_to, | |
1422 | &cliobd->cl_target_uuid, | |
1423 | conn->c_self, &sf); | |
1424 | ||
1425 | sp = imp->imp_obd->u.cli.cl_sp_me; | |
1426 | } else { | |
0c2bc758 | 1427 | /* reverse import, determine flavor from incoming request */ |
d7e09d03 PT |
1428 | sf = *flvr; |
1429 | ||
1430 | if (sf.sf_rpc != SPTLRPC_FLVR_NULL) | |
1431 | sf.sf_flags = PTLRPC_SEC_FL_REVERSE | | |
1432 | PTLRPC_SEC_FL_ROOTONLY; | |
1433 | ||
1434 | sp = sptlrpc_target_sec_part(imp->imp_obd); | |
1435 | } | |
1436 | ||
1437 | sec = sptlrpc_import_sec_ref(imp); | |
1438 | if (sec) { | |
d0bfef31 | 1439 | char str2[24]; |
d7e09d03 PT |
1440 | |
1441 | if (flavor_equal(&sf, &sec->ps_flvr)) | |
a9b3e8f3 | 1442 | goto out; |
d7e09d03 PT |
1443 | |
1444 | CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n", | |
1445 | imp->imp_obd->obd_name, | |
1446 | obd_uuid2str(&conn->c_remote_uuid), | |
1447 | sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)), | |
1448 | sptlrpc_flavor2name(&sf, str2, sizeof(str2))); | |
1449 | ||
1450 | if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) == | |
1451 | SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) && | |
1452 | SPTLRPC_FLVR_MECH(sf.sf_rpc) == | |
1453 | SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) { | |
1454 | sptlrpc_import_sec_adapt_inplace(imp, sec, &sf); | |
a9b3e8f3 | 1455 | goto out; |
d7e09d03 PT |
1456 | } |
1457 | } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) != | |
1458 | SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) { | |
1459 | CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n", | |
1460 | imp->imp_obd->obd_name, | |
1461 | obd_uuid2str(&conn->c_remote_uuid), | |
1462 | LNET_NIDNET(conn->c_self), | |
1463 | sptlrpc_flavor2name(&sf, str, sizeof(str))); | |
1464 | } | |
1465 | ||
1466 | mutex_lock(&imp->imp_sec_mutex); | |
1467 | ||
1468 | newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp); | |
1469 | if (newsec) { | |
1470 | sptlrpc_import_sec_install(imp, newsec); | |
1471 | } else { | |
1472 | CERROR("import %s->%s: failed to create new sec\n", | |
1473 | imp->imp_obd->obd_name, | |
1474 | obd_uuid2str(&conn->c_remote_uuid)); | |
1475 | rc = -EPERM; | |
1476 | } | |
1477 | ||
1478 | mutex_unlock(&imp->imp_sec_mutex); | |
1479 | out: | |
1480 | sptlrpc_sec_put(sec); | |
0a3bdb00 | 1481 | return rc; |
d7e09d03 PT |
1482 | } |
1483 | ||
1484 | void sptlrpc_import_sec_put(struct obd_import *imp) | |
1485 | { | |
1486 | if (imp->imp_sec) { | |
1487 | sptlrpc_sec_kill(imp->imp_sec); | |
1488 | ||
1489 | sptlrpc_sec_put(imp->imp_sec); | |
1490 | imp->imp_sec = NULL; | |
1491 | } | |
1492 | } | |
1493 | ||
1494 | static void import_flush_ctx_common(struct obd_import *imp, | |
1495 | uid_t uid, int grace, int force) | |
1496 | { | |
1497 | struct ptlrpc_sec *sec; | |
1498 | ||
1499 | if (imp == NULL) | |
1500 | return; | |
1501 | ||
1502 | sec = sptlrpc_import_sec_ref(imp); | |
1503 | if (sec == NULL) | |
1504 | return; | |
1505 | ||
1506 | sec_cop_flush_ctx_cache(sec, uid, grace, force); | |
1507 | sptlrpc_sec_put(sec); | |
1508 | } | |
1509 | ||
1510 | void sptlrpc_import_flush_root_ctx(struct obd_import *imp) | |
1511 | { | |
1512 | /* it's important to use grace mode, see explain in | |
1513 | * sptlrpc_req_refresh_ctx() */ | |
1514 | import_flush_ctx_common(imp, 0, 1, 1); | |
1515 | } | |
1516 | ||
1517 | void sptlrpc_import_flush_my_ctx(struct obd_import *imp) | |
1518 | { | |
4b1a25f0 PT |
1519 | import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()), |
1520 | 1, 1); | |
d7e09d03 PT |
1521 | } |
1522 | EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx); | |
1523 | ||
1524 | void sptlrpc_import_flush_all_ctx(struct obd_import *imp) | |
1525 | { | |
1526 | import_flush_ctx_common(imp, -1, 1, 1); | |
1527 | } | |
1528 | EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx); | |
1529 | ||
1530 | /** | |
1531 | * Used by ptlrpc client to allocate request buffer of \a req. Upon return | |
1532 | * successfully, req->rq_reqmsg points to a buffer with size \a msgsize. | |
1533 | */ | |
1534 | int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize) | |
1535 | { | |
1536 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
1537 | struct ptlrpc_sec_policy *policy; | |
1538 | int rc; | |
1539 | ||
1540 | LASSERT(ctx); | |
1541 | LASSERT(ctx->cc_sec); | |
1542 | LASSERT(ctx->cc_sec->ps_policy); | |
1543 | LASSERT(req->rq_reqmsg == NULL); | |
1544 | LASSERT_ATOMIC_POS(&ctx->cc_refcount); | |
1545 | ||
1546 | policy = ctx->cc_sec->ps_policy; | |
1547 | rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize); | |
1548 | if (!rc) { | |
1549 | LASSERT(req->rq_reqmsg); | |
1550 | LASSERT(req->rq_reqbuf || req->rq_clrbuf); | |
1551 | ||
1552 | /* zeroing preallocated buffer */ | |
1553 | if (req->rq_pool) | |
1554 | memset(req->rq_reqmsg, 0, msgsize); | |
1555 | } | |
1556 | ||
1557 | return rc; | |
1558 | } | |
1559 | ||
1560 | /** | |
1561 | * Used by ptlrpc client to free request buffer of \a req. After this | |
1562 | * req->rq_reqmsg is set to NULL and should not be accessed anymore. | |
1563 | */ | |
1564 | void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req) | |
1565 | { | |
1566 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
1567 | struct ptlrpc_sec_policy *policy; | |
1568 | ||
1569 | LASSERT(ctx); | |
1570 | LASSERT(ctx->cc_sec); | |
1571 | LASSERT(ctx->cc_sec->ps_policy); | |
1572 | LASSERT_ATOMIC_POS(&ctx->cc_refcount); | |
1573 | ||
1574 | if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL) | |
1575 | return; | |
1576 | ||
1577 | policy = ctx->cc_sec->ps_policy; | |
1578 | policy->sp_cops->free_reqbuf(ctx->cc_sec, req); | |
1579 | req->rq_reqmsg = NULL; | |
1580 | } | |
1581 | ||
1582 | /* | |
1583 | * NOTE caller must guarantee the buffer size is enough for the enlargement | |
1584 | */ | |
1585 | void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg, | |
1586 | int segment, int newsize) | |
1587 | { | |
d0bfef31 CH |
1588 | void *src, *dst; |
1589 | int oldsize, oldmsg_size, movesize; | |
d7e09d03 PT |
1590 | |
1591 | LASSERT(segment < msg->lm_bufcount); | |
1592 | LASSERT(msg->lm_buflens[segment] <= newsize); | |
1593 | ||
1594 | if (msg->lm_buflens[segment] == newsize) | |
1595 | return; | |
1596 | ||
1597 | /* nothing to do if we are enlarging the last segment */ | |
1598 | if (segment == msg->lm_bufcount - 1) { | |
1599 | msg->lm_buflens[segment] = newsize; | |
1600 | return; | |
1601 | } | |
1602 | ||
1603 | oldsize = msg->lm_buflens[segment]; | |
1604 | ||
1605 | src = lustre_msg_buf(msg, segment + 1, 0); | |
1606 | msg->lm_buflens[segment] = newsize; | |
1607 | dst = lustre_msg_buf(msg, segment + 1, 0); | |
1608 | msg->lm_buflens[segment] = oldsize; | |
1609 | ||
1610 | /* move from segment + 1 to end segment */ | |
1611 | LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2); | |
1612 | oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens); | |
1613 | movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg); | |
1614 | LASSERT(movesize >= 0); | |
1615 | ||
1616 | if (movesize) | |
1617 | memmove(dst, src, movesize); | |
1618 | ||
1619 | /* note we don't clear the ares where old data live, not secret */ | |
1620 | ||
1621 | /* finally set new segment size */ | |
1622 | msg->lm_buflens[segment] = newsize; | |
1623 | } | |
1624 | EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace); | |
1625 | ||
1626 | /** | |
1627 | * Used by ptlrpc client to enlarge the \a segment of request message pointed | |
1628 | * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be | |
1629 | * preserved after the enlargement. this must be called after original request | |
1630 | * buffer being allocated. | |
1631 | * | |
1632 | * \note after this be called, rq_reqmsg and rq_reqlen might have been changed, | |
1633 | * so caller should refresh its local pointers if needed. | |
1634 | */ | |
1635 | int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req, | |
1636 | int segment, int newsize) | |
1637 | { | |
d0bfef31 CH |
1638 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; |
1639 | struct ptlrpc_sec_cops *cops; | |
1640 | struct lustre_msg *msg = req->rq_reqmsg; | |
d7e09d03 PT |
1641 | |
1642 | LASSERT(ctx); | |
1643 | LASSERT(msg); | |
1644 | LASSERT(msg->lm_bufcount > segment); | |
1645 | LASSERT(msg->lm_buflens[segment] <= newsize); | |
1646 | ||
1647 | if (msg->lm_buflens[segment] == newsize) | |
1648 | return 0; | |
1649 | ||
1650 | cops = ctx->cc_sec->ps_policy->sp_cops; | |
1651 | LASSERT(cops->enlarge_reqbuf); | |
1652 | return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize); | |
1653 | } | |
1654 | EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf); | |
1655 | ||
1656 | /** | |
1657 | * Used by ptlrpc client to allocate reply buffer of \a req. | |
1658 | * | |
1659 | * \note After this, req->rq_repmsg is still not accessible. | |
1660 | */ | |
1661 | int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize) | |
1662 | { | |
1663 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
1664 | struct ptlrpc_sec_policy *policy; | |
d7e09d03 PT |
1665 | |
1666 | LASSERT(ctx); | |
1667 | LASSERT(ctx->cc_sec); | |
1668 | LASSERT(ctx->cc_sec->ps_policy); | |
1669 | ||
1670 | if (req->rq_repbuf) | |
0a3bdb00 | 1671 | return 0; |
d7e09d03 PT |
1672 | |
1673 | policy = ctx->cc_sec->ps_policy; | |
0a3bdb00 | 1674 | return policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize); |
d7e09d03 PT |
1675 | } |
1676 | ||
1677 | /** | |
1678 | * Used by ptlrpc client to free reply buffer of \a req. After this | |
1679 | * req->rq_repmsg is set to NULL and should not be accessed anymore. | |
1680 | */ | |
1681 | void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req) | |
1682 | { | |
1683 | struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; | |
1684 | struct ptlrpc_sec_policy *policy; | |
d7e09d03 PT |
1685 | |
1686 | LASSERT(ctx); | |
1687 | LASSERT(ctx->cc_sec); | |
1688 | LASSERT(ctx->cc_sec->ps_policy); | |
1689 | LASSERT_ATOMIC_POS(&ctx->cc_refcount); | |
1690 | ||
1691 | if (req->rq_repbuf == NULL) | |
1692 | return; | |
1693 | LASSERT(req->rq_repbuf_len); | |
1694 | ||
1695 | policy = ctx->cc_sec->ps_policy; | |
1696 | policy->sp_cops->free_repbuf(ctx->cc_sec, req); | |
1697 | req->rq_repmsg = NULL; | |
d7e09d03 PT |
1698 | } |
1699 | ||
1700 | int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp, | |
1701 | struct ptlrpc_cli_ctx *ctx) | |
1702 | { | |
1703 | struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy; | |
1704 | ||
1705 | if (!policy->sp_cops->install_rctx) | |
1706 | return 0; | |
1707 | return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx); | |
1708 | } | |
1709 | ||
1710 | int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp, | |
1711 | struct ptlrpc_svc_ctx *ctx) | |
1712 | { | |
1713 | struct ptlrpc_sec_policy *policy = ctx->sc_policy; | |
1714 | ||
1715 | if (!policy->sp_sops->install_rctx) | |
1716 | return 0; | |
1717 | return policy->sp_sops->install_rctx(imp, ctx); | |
1718 | } | |
1719 | ||
1720 | /**************************************** | |
1721 | * server side security * | |
1722 | ****************************************/ | |
1723 | ||
1724 | static int flavor_allowed(struct sptlrpc_flavor *exp, | |
1725 | struct ptlrpc_request *req) | |
1726 | { | |
1727 | struct sptlrpc_flavor *flvr = &req->rq_flvr; | |
1728 | ||
1729 | if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc) | |
1730 | return 1; | |
1731 | ||
1732 | if ((req->rq_ctx_init || req->rq_ctx_fini) && | |
1733 | SPTLRPC_FLVR_POLICY(exp->sf_rpc) == | |
1734 | SPTLRPC_FLVR_POLICY(flvr->sf_rpc) && | |
1735 | SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc)) | |
1736 | return 1; | |
1737 | ||
1738 | return 0; | |
1739 | } | |
1740 | ||
1741 | #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10) | |
1742 | ||
1743 | /** | |
1744 | * Given an export \a exp, check whether the flavor of incoming \a req | |
1745 | * is allowed by the export \a exp. Main logic is about taking care of | |
1746 | * changing configurations. Return 0 means success. | |
1747 | */ | |
1748 | int sptlrpc_target_export_check(struct obd_export *exp, | |
1749 | struct ptlrpc_request *req) | |
1750 | { | |
d0bfef31 | 1751 | struct sptlrpc_flavor flavor; |
d7e09d03 PT |
1752 | |
1753 | if (exp == NULL) | |
1754 | return 0; | |
1755 | ||
1756 | /* client side export has no imp_reverse, skip | |
1757 | * FIXME maybe we should check flavor this as well??? */ | |
1758 | if (exp->exp_imp_reverse == NULL) | |
1759 | return 0; | |
1760 | ||
1761 | /* don't care about ctx fini rpc */ | |
1762 | if (req->rq_ctx_fini) | |
1763 | return 0; | |
1764 | ||
1765 | spin_lock(&exp->exp_lock); | |
1766 | ||
1767 | /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for | |
1768 | * the first req with the new flavor, then treat it as current flavor, | |
1769 | * adapt reverse sec according to it. | |
1770 | * note the first rpc with new flavor might not be with root ctx, in | |
1771 | * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */ | |
1772 | if (unlikely(exp->exp_flvr_changed) && | |
1773 | flavor_allowed(&exp->exp_flvr_old[1], req)) { | |
1774 | /* make the new flavor as "current", and old ones as | |
1775 | * about-to-expire */ | |
1776 | CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp, | |
1777 | exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc); | |
1778 | flavor = exp->exp_flvr_old[1]; | |
1779 | exp->exp_flvr_old[1] = exp->exp_flvr_old[0]; | |
1780 | exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0]; | |
1781 | exp->exp_flvr_old[0] = exp->exp_flvr; | |
7264b8a5 | 1782 | exp->exp_flvr_expire[0] = get_seconds() + |
d7e09d03 PT |
1783 | EXP_FLVR_UPDATE_EXPIRE; |
1784 | exp->exp_flvr = flavor; | |
1785 | ||
1786 | /* flavor change finished */ | |
1787 | exp->exp_flvr_changed = 0; | |
1788 | LASSERT(exp->exp_flvr_adapt == 1); | |
1789 | ||
1790 | /* if it's gss, we only interested in root ctx init */ | |
1791 | if (req->rq_auth_gss && | |
1792 | !(req->rq_ctx_init && | |
1793 | (req->rq_auth_usr_root || req->rq_auth_usr_mdt || | |
1794 | req->rq_auth_usr_ost))) { | |
1795 | spin_unlock(&exp->exp_lock); | |
1796 | CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n", | |
1797 | req->rq_auth_gss, req->rq_ctx_init, | |
1798 | req->rq_auth_usr_root, req->rq_auth_usr_mdt, | |
1799 | req->rq_auth_usr_ost); | |
1800 | return 0; | |
1801 | } | |
1802 | ||
1803 | exp->exp_flvr_adapt = 0; | |
1804 | spin_unlock(&exp->exp_lock); | |
1805 | ||
1806 | return sptlrpc_import_sec_adapt(exp->exp_imp_reverse, | |
1807 | req->rq_svc_ctx, &flavor); | |
1808 | } | |
1809 | ||
1810 | /* if it equals to the current flavor, we accept it, but need to | |
1811 | * dealing with reverse sec/ctx */ | |
1812 | if (likely(flavor_allowed(&exp->exp_flvr, req))) { | |
1813 | /* most cases should return here, we only interested in | |
1814 | * gss root ctx init */ | |
1815 | if (!req->rq_auth_gss || !req->rq_ctx_init || | |
1816 | (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt && | |
1817 | !req->rq_auth_usr_ost)) { | |
1818 | spin_unlock(&exp->exp_lock); | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | /* if flavor just changed, we should not proceed, just leave | |
1823 | * it and current flavor will be discovered and replaced | |
1824 | * shortly, and let _this_ rpc pass through */ | |
1825 | if (exp->exp_flvr_changed) { | |
1826 | LASSERT(exp->exp_flvr_adapt); | |
1827 | spin_unlock(&exp->exp_lock); | |
1828 | return 0; | |
1829 | } | |
1830 | ||
1831 | if (exp->exp_flvr_adapt) { | |
1832 | exp->exp_flvr_adapt = 0; | |
1833 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n", | |
1834 | exp, exp->exp_flvr.sf_rpc, | |
1835 | exp->exp_flvr_old[0].sf_rpc, | |
1836 | exp->exp_flvr_old[1].sf_rpc); | |
1837 | flavor = exp->exp_flvr; | |
1838 | spin_unlock(&exp->exp_lock); | |
1839 | ||
1840 | return sptlrpc_import_sec_adapt(exp->exp_imp_reverse, | |
1841 | req->rq_svc_ctx, | |
1842 | &flavor); | |
1843 | } else { | |
2d00bd17 JP |
1844 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n", |
1845 | exp, exp->exp_flvr.sf_rpc, | |
d7e09d03 PT |
1846 | exp->exp_flvr_old[0].sf_rpc, |
1847 | exp->exp_flvr_old[1].sf_rpc); | |
1848 | spin_unlock(&exp->exp_lock); | |
1849 | ||
1850 | return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse, | |
1851 | req->rq_svc_ctx); | |
1852 | } | |
1853 | } | |
1854 | ||
1855 | if (exp->exp_flvr_expire[0]) { | |
7264b8a5 | 1856 | if (exp->exp_flvr_expire[0] >= get_seconds()) { |
d7e09d03 | 1857 | if (flavor_allowed(&exp->exp_flvr_old[0], req)) { |
2d00bd17 | 1858 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (" CFS_DURATION_T ")\n", exp, |
d7e09d03 PT |
1859 | exp->exp_flvr.sf_rpc, |
1860 | exp->exp_flvr_old[0].sf_rpc, | |
1861 | exp->exp_flvr_old[1].sf_rpc, | |
1862 | exp->exp_flvr_expire[0] - | |
2d00bd17 | 1863 | get_seconds()); |
d7e09d03 PT |
1864 | spin_unlock(&exp->exp_lock); |
1865 | return 0; | |
1866 | } | |
1867 | } else { | |
1868 | CDEBUG(D_SEC, "mark middle expired\n"); | |
1869 | exp->exp_flvr_expire[0] = 0; | |
1870 | } | |
1871 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp, | |
1872 | exp->exp_flvr.sf_rpc, | |
1873 | exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc, | |
1874 | req->rq_flvr.sf_rpc); | |
1875 | } | |
1876 | ||
1877 | /* now it doesn't match the current flavor, the only chance we can | |
1878 | * accept it is match the old flavors which is not expired. */ | |
1879 | if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) { | |
7264b8a5 | 1880 | if (exp->exp_flvr_expire[1] >= get_seconds()) { |
d7e09d03 | 1881 | if (flavor_allowed(&exp->exp_flvr_old[1], req)) { |
2d00bd17 JP |
1882 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (" CFS_DURATION_T ")\n", |
1883 | exp, | |
d7e09d03 PT |
1884 | exp->exp_flvr.sf_rpc, |
1885 | exp->exp_flvr_old[0].sf_rpc, | |
1886 | exp->exp_flvr_old[1].sf_rpc, | |
1887 | exp->exp_flvr_expire[1] - | |
2d00bd17 | 1888 | get_seconds()); |
d7e09d03 PT |
1889 | spin_unlock(&exp->exp_lock); |
1890 | return 0; | |
1891 | } | |
1892 | } else { | |
1893 | CDEBUG(D_SEC, "mark oldest expired\n"); | |
1894 | exp->exp_flvr_expire[1] = 0; | |
1895 | } | |
1896 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n", | |
1897 | exp, exp->exp_flvr.sf_rpc, | |
1898 | exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc, | |
1899 | req->rq_flvr.sf_rpc); | |
1900 | } else { | |
1901 | CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n", | |
1902 | exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc, | |
1903 | exp->exp_flvr_old[1].sf_rpc); | |
1904 | } | |
1905 | ||
1906 | spin_unlock(&exp->exp_lock); | |
1907 | ||
2d00bd17 | 1908 | CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n", |
d7e09d03 PT |
1909 | exp, exp->exp_obd->obd_name, |
1910 | req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini, | |
1911 | req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost, | |
1912 | req->rq_flvr.sf_rpc, | |
1913 | exp->exp_flvr.sf_rpc, | |
1914 | exp->exp_flvr_old[0].sf_rpc, | |
1915 | exp->exp_flvr_expire[0] ? | |
1916 | (unsigned long) (exp->exp_flvr_expire[0] - | |
7264b8a5 | 1917 | get_seconds()) : 0, |
d7e09d03 PT |
1918 | exp->exp_flvr_old[1].sf_rpc, |
1919 | exp->exp_flvr_expire[1] ? | |
1920 | (unsigned long) (exp->exp_flvr_expire[1] - | |
7264b8a5 | 1921 | get_seconds()) : 0); |
d7e09d03 PT |
1922 | return -EACCES; |
1923 | } | |
1924 | EXPORT_SYMBOL(sptlrpc_target_export_check); | |
1925 | ||
1926 | void sptlrpc_target_update_exp_flavor(struct obd_device *obd, | |
1927 | struct sptlrpc_rule_set *rset) | |
1928 | { | |
d0bfef31 CH |
1929 | struct obd_export *exp; |
1930 | struct sptlrpc_flavor new_flvr; | |
d7e09d03 PT |
1931 | |
1932 | LASSERT(obd); | |
1933 | ||
1934 | spin_lock(&obd->obd_dev_lock); | |
1935 | ||
1936 | list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) { | |
1937 | if (exp->exp_connection == NULL) | |
1938 | continue; | |
1939 | ||
1940 | /* note if this export had just been updated flavor | |
1941 | * (exp_flvr_changed == 1), this will override the | |
1942 | * previous one. */ | |
1943 | spin_lock(&exp->exp_lock); | |
1944 | sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer, | |
1945 | exp->exp_connection->c_peer.nid, | |
1946 | &new_flvr); | |
1947 | if (exp->exp_flvr_changed || | |
1948 | !flavor_equal(&new_flvr, &exp->exp_flvr)) { | |
1949 | exp->exp_flvr_old[1] = new_flvr; | |
1950 | exp->exp_flvr_expire[1] = 0; | |
1951 | exp->exp_flvr_changed = 1; | |
1952 | exp->exp_flvr_adapt = 1; | |
1953 | ||
1954 | CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n", | |
1955 | exp, sptlrpc_part2name(exp->exp_sp_peer), | |
1956 | exp->exp_flvr.sf_rpc, | |
1957 | exp->exp_flvr_old[1].sf_rpc); | |
1958 | } | |
1959 | spin_unlock(&exp->exp_lock); | |
1960 | } | |
1961 | ||
1962 | spin_unlock(&obd->obd_dev_lock); | |
1963 | } | |
1964 | EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor); | |
1965 | ||
1966 | static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc) | |
1967 | { | |
1968 | /* peer's claim is unreliable unless gss is being used */ | |
1969 | if (!req->rq_auth_gss || svc_rc == SECSVC_DROP) | |
1970 | return svc_rc; | |
1971 | ||
1972 | switch (req->rq_sp_from) { | |
1973 | case LUSTRE_SP_CLI: | |
1974 | if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) { | |
1975 | DEBUG_REQ(D_ERROR, req, "faked source CLI"); | |
1976 | svc_rc = SECSVC_DROP; | |
1977 | } | |
1978 | break; | |
1979 | case LUSTRE_SP_MDT: | |
1980 | if (!req->rq_auth_usr_mdt) { | |
1981 | DEBUG_REQ(D_ERROR, req, "faked source MDT"); | |
1982 | svc_rc = SECSVC_DROP; | |
1983 | } | |
1984 | break; | |
1985 | case LUSTRE_SP_OST: | |
1986 | if (!req->rq_auth_usr_ost) { | |
1987 | DEBUG_REQ(D_ERROR, req, "faked source OST"); | |
1988 | svc_rc = SECSVC_DROP; | |
1989 | } | |
1990 | break; | |
1991 | case LUSTRE_SP_MGS: | |
1992 | case LUSTRE_SP_MGC: | |
1993 | if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt && | |
1994 | !req->rq_auth_usr_ost) { | |
1995 | DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS"); | |
1996 | svc_rc = SECSVC_DROP; | |
1997 | } | |
1998 | break; | |
1999 | case LUSTRE_SP_ANY: | |
2000 | default: | |
2001 | DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from); | |
2002 | svc_rc = SECSVC_DROP; | |
2003 | } | |
2004 | ||
2005 | return svc_rc; | |
2006 | } | |
2007 | ||
2008 | /** | |
2009 | * Used by ptlrpc server, to perform transformation upon request message of | |
2010 | * incoming \a req. This must be the first thing to do with a incoming | |
2011 | * request in ptlrpc layer. | |
2012 | * | |
2013 | * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in | |
2014 | * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set. | |
2015 | * \retval SECSVC_COMPLETE success, the request has been fully processed, and | |
2016 | * reply message has been prepared. | |
2017 | * \retval SECSVC_DROP failed, this request should be dropped. | |
2018 | */ | |
2019 | int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req) | |
2020 | { | |
2021 | struct ptlrpc_sec_policy *policy; | |
d0bfef31 CH |
2022 | struct lustre_msg *msg = req->rq_reqbuf; |
2023 | int rc; | |
d7e09d03 PT |
2024 | |
2025 | LASSERT(msg); | |
2026 | LASSERT(req->rq_reqmsg == NULL); | |
2027 | LASSERT(req->rq_repmsg == NULL); | |
2028 | LASSERT(req->rq_svc_ctx == NULL); | |
2029 | ||
2030 | req->rq_req_swab_mask = 0; | |
2031 | ||
2032 | rc = __lustre_unpack_msg(msg, req->rq_reqdata_len); | |
2033 | switch (rc) { | |
2034 | case 1: | |
2035 | lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF); | |
2036 | case 0: | |
2037 | break; | |
2038 | default: | |
b0f5aad5 | 2039 | CERROR("error unpacking request from %s x%llu\n", |
d7e09d03 | 2040 | libcfs_id2str(req->rq_peer), req->rq_xid); |
0a3bdb00 | 2041 | return SECSVC_DROP; |
d7e09d03 PT |
2042 | } |
2043 | ||
2044 | req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr); | |
2045 | req->rq_sp_from = LUSTRE_SP_ANY; | |
4b1a25f0 PT |
2046 | req->rq_auth_uid = -1; |
2047 | req->rq_auth_mapped_uid = -1; | |
d7e09d03 PT |
2048 | |
2049 | policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc); | |
2050 | if (!policy) { | |
2051 | CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc); | |
0a3bdb00 | 2052 | return SECSVC_DROP; |
d7e09d03 PT |
2053 | } |
2054 | ||
2055 | LASSERT(policy->sp_sops->accept); | |
2056 | rc = policy->sp_sops->accept(req); | |
2057 | sptlrpc_policy_put(policy); | |
2058 | LASSERT(req->rq_reqmsg || rc != SECSVC_OK); | |
2059 | LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP); | |
2060 | ||
2061 | /* | |
2062 | * if it's not null flavor (which means embedded packing msg), | |
0c2bc758 | 2063 | * reset the swab mask for the coming inner msg unpacking. |
d7e09d03 PT |
2064 | */ |
2065 | if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL) | |
2066 | req->rq_req_swab_mask = 0; | |
2067 | ||
2068 | /* sanity check for the request source */ | |
2069 | rc = sptlrpc_svc_check_from(req, rc); | |
0a3bdb00 | 2070 | return rc; |
d7e09d03 PT |
2071 | } |
2072 | ||
2073 | /** | |
2074 | * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed, | |
2075 | * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to | |
2076 | * a buffer of \a msglen size. | |
2077 | */ | |
2078 | int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen) | |
2079 | { | |
2080 | struct ptlrpc_sec_policy *policy; | |
2081 | struct ptlrpc_reply_state *rs; | |
2082 | int rc; | |
d7e09d03 PT |
2083 | |
2084 | LASSERT(req->rq_svc_ctx); | |
2085 | LASSERT(req->rq_svc_ctx->sc_policy); | |
2086 | ||
2087 | policy = req->rq_svc_ctx->sc_policy; | |
2088 | LASSERT(policy->sp_sops->alloc_rs); | |
2089 | ||
2090 | rc = policy->sp_sops->alloc_rs(req, msglen); | |
2091 | if (unlikely(rc == -ENOMEM)) { | |
a34041af PF |
2092 | struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt; |
2093 | if (svcpt->scp_service->srv_max_reply_size < | |
2094 | msglen + sizeof(struct ptlrpc_reply_state)) { | |
2095 | /* Just return failure if the size is too big */ | |
2096 | CERROR("size of message is too big (%zd), %d allowed", | |
2097 | msglen + sizeof(struct ptlrpc_reply_state), | |
2098 | svcpt->scp_service->srv_max_reply_size); | |
2099 | return -ENOMEM; | |
2100 | } | |
2101 | ||
d7e09d03 | 2102 | /* failed alloc, try emergency pool */ |
a34041af | 2103 | rs = lustre_get_emerg_rs(svcpt); |
d7e09d03 | 2104 | if (rs == NULL) |
0a3bdb00 | 2105 | return -ENOMEM; |
d7e09d03 PT |
2106 | |
2107 | req->rq_reply_state = rs; | |
2108 | rc = policy->sp_sops->alloc_rs(req, msglen); | |
2109 | if (rc) { | |
2110 | lustre_put_emerg_rs(rs); | |
2111 | req->rq_reply_state = NULL; | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | LASSERT(rc != 0 || | |
2116 | (req->rq_reply_state && req->rq_reply_state->rs_msg)); | |
2117 | ||
0a3bdb00 | 2118 | return rc; |
d7e09d03 PT |
2119 | } |
2120 | ||
2121 | /** | |
2122 | * Used by ptlrpc server, to perform transformation upon reply message. | |
2123 | * | |
0c2bc758 | 2124 | * \post req->rq_reply_off is set to appropriate server-controlled reply offset. |
d7e09d03 PT |
2125 | * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible. |
2126 | */ | |
2127 | int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req) | |
2128 | { | |
2129 | struct ptlrpc_sec_policy *policy; | |
2130 | int rc; | |
d7e09d03 PT |
2131 | |
2132 | LASSERT(req->rq_svc_ctx); | |
2133 | LASSERT(req->rq_svc_ctx->sc_policy); | |
2134 | ||
2135 | policy = req->rq_svc_ctx->sc_policy; | |
2136 | LASSERT(policy->sp_sops->authorize); | |
2137 | ||
2138 | rc = policy->sp_sops->authorize(req); | |
2139 | LASSERT(rc || req->rq_reply_state->rs_repdata_len); | |
2140 | ||
0a3bdb00 | 2141 | return rc; |
d7e09d03 PT |
2142 | } |
2143 | ||
2144 | /** | |
2145 | * Used by ptlrpc server, to free reply_state. | |
2146 | */ | |
2147 | void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs) | |
2148 | { | |
2149 | struct ptlrpc_sec_policy *policy; | |
2150 | unsigned int prealloc; | |
d7e09d03 PT |
2151 | |
2152 | LASSERT(rs->rs_svc_ctx); | |
2153 | LASSERT(rs->rs_svc_ctx->sc_policy); | |
2154 | ||
2155 | policy = rs->rs_svc_ctx->sc_policy; | |
2156 | LASSERT(policy->sp_sops->free_rs); | |
2157 | ||
2158 | prealloc = rs->rs_prealloc; | |
2159 | policy->sp_sops->free_rs(rs); | |
2160 | ||
2161 | if (prealloc) | |
2162 | lustre_put_emerg_rs(rs); | |
d7e09d03 PT |
2163 | } |
2164 | ||
2165 | void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req) | |
2166 | { | |
2167 | struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx; | |
2168 | ||
2169 | if (ctx != NULL) | |
2170 | atomic_inc(&ctx->sc_refcount); | |
2171 | } | |
2172 | ||
2173 | void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req) | |
2174 | { | |
2175 | struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx; | |
2176 | ||
2177 | if (ctx == NULL) | |
2178 | return; | |
2179 | ||
2180 | LASSERT_ATOMIC_POS(&ctx->sc_refcount); | |
2181 | if (atomic_dec_and_test(&ctx->sc_refcount)) { | |
2182 | if (ctx->sc_policy->sp_sops->free_ctx) | |
2183 | ctx->sc_policy->sp_sops->free_ctx(ctx); | |
2184 | } | |
2185 | req->rq_svc_ctx = NULL; | |
2186 | } | |
2187 | ||
2188 | void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req) | |
2189 | { | |
2190 | struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx; | |
2191 | ||
2192 | if (ctx == NULL) | |
2193 | return; | |
2194 | ||
2195 | LASSERT_ATOMIC_POS(&ctx->sc_refcount); | |
2196 | if (ctx->sc_policy->sp_sops->invalidate_ctx) | |
2197 | ctx->sc_policy->sp_sops->invalidate_ctx(ctx); | |
2198 | } | |
2199 | EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate); | |
2200 | ||
2201 | /**************************************** | |
2202 | * bulk security * | |
2203 | ****************************************/ | |
2204 | ||
2205 | /** | |
2206 | * Perform transformation upon bulk data pointed by \a desc. This is called | |
2207 | * before transforming the request message. | |
2208 | */ | |
2209 | int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req, | |
2210 | struct ptlrpc_bulk_desc *desc) | |
2211 | { | |
2212 | struct ptlrpc_cli_ctx *ctx; | |
2213 | ||
2214 | LASSERT(req->rq_bulk_read || req->rq_bulk_write); | |
2215 | ||
2216 | if (!req->rq_pack_bulk) | |
2217 | return 0; | |
2218 | ||
2219 | ctx = req->rq_cli_ctx; | |
2220 | if (ctx->cc_ops->wrap_bulk) | |
2221 | return ctx->cc_ops->wrap_bulk(ctx, req, desc); | |
2222 | return 0; | |
2223 | } | |
2224 | EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk); | |
2225 | ||
2226 | /** | |
2227 | * This is called after unwrap the reply message. | |
2228 | * return nob of actual plain text size received, or error code. | |
2229 | */ | |
2230 | int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req, | |
2231 | struct ptlrpc_bulk_desc *desc, | |
2232 | int nob) | |
2233 | { | |
d0bfef31 CH |
2234 | struct ptlrpc_cli_ctx *ctx; |
2235 | int rc; | |
d7e09d03 PT |
2236 | |
2237 | LASSERT(req->rq_bulk_read && !req->rq_bulk_write); | |
2238 | ||
2239 | if (!req->rq_pack_bulk) | |
2240 | return desc->bd_nob_transferred; | |
2241 | ||
2242 | ctx = req->rq_cli_ctx; | |
2243 | if (ctx->cc_ops->unwrap_bulk) { | |
2244 | rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc); | |
2245 | if (rc < 0) | |
2246 | return rc; | |
2247 | } | |
2248 | return desc->bd_nob_transferred; | |
2249 | } | |
2250 | EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read); | |
2251 | ||
2252 | /** | |
2253 | * This is called after unwrap the reply message. | |
2254 | * return 0 for success or error code. | |
2255 | */ | |
2256 | int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req, | |
2257 | struct ptlrpc_bulk_desc *desc) | |
2258 | { | |
d0bfef31 CH |
2259 | struct ptlrpc_cli_ctx *ctx; |
2260 | int rc; | |
d7e09d03 PT |
2261 | |
2262 | LASSERT(!req->rq_bulk_read && req->rq_bulk_write); | |
2263 | ||
2264 | if (!req->rq_pack_bulk) | |
2265 | return 0; | |
2266 | ||
2267 | ctx = req->rq_cli_ctx; | |
2268 | if (ctx->cc_ops->unwrap_bulk) { | |
2269 | rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc); | |
2270 | if (rc < 0) | |
2271 | return rc; | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * if everything is going right, nob should equals to nob_transferred. | |
2276 | * in case of privacy mode, nob_transferred needs to be adjusted. | |
2277 | */ | |
2278 | if (desc->bd_nob != desc->bd_nob_transferred) { | |
2279 | CERROR("nob %d doesn't match transferred nob %d", | |
2280 | desc->bd_nob, desc->bd_nob_transferred); | |
2281 | return -EPROTO; | |
2282 | } | |
2283 | ||
2284 | return 0; | |
2285 | } | |
2286 | EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write); | |
2287 | ||
2288 | ||
2289 | /**************************************** | |
2290 | * user descriptor helpers * | |
2291 | ****************************************/ | |
2292 | ||
2293 | int sptlrpc_current_user_desc_size(void) | |
2294 | { | |
2295 | int ngroups; | |
2296 | ||
2297 | ngroups = current_ngroups; | |
2298 | ||
2299 | if (ngroups > LUSTRE_MAX_GROUPS) | |
2300 | ngroups = LUSTRE_MAX_GROUPS; | |
2301 | return sptlrpc_user_desc_size(ngroups); | |
2302 | } | |
2303 | EXPORT_SYMBOL(sptlrpc_current_user_desc_size); | |
2304 | ||
2305 | int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset) | |
2306 | { | |
2307 | struct ptlrpc_user_desc *pud; | |
2308 | ||
2309 | pud = lustre_msg_buf(msg, offset, 0); | |
2310 | ||
4b1a25f0 PT |
2311 | pud->pud_uid = from_kuid(&init_user_ns, current_uid()); |
2312 | pud->pud_gid = from_kgid(&init_user_ns, current_gid()); | |
2313 | pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid()); | |
2314 | pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid()); | |
d7e09d03 PT |
2315 | pud->pud_cap = cfs_curproc_cap_pack(); |
2316 | pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4; | |
2317 | ||
2318 | task_lock(current); | |
2319 | if (pud->pud_ngroups > current_ngroups) | |
2320 | pud->pud_ngroups = current_ngroups; | |
2321 | memcpy(pud->pud_groups, current_cred()->group_info->blocks[0], | |
2322 | pud->pud_ngroups * sizeof(__u32)); | |
2323 | task_unlock(current); | |
2324 | ||
2325 | return 0; | |
2326 | } | |
2327 | EXPORT_SYMBOL(sptlrpc_pack_user_desc); | |
2328 | ||
2329 | int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed) | |
2330 | { | |
2331 | struct ptlrpc_user_desc *pud; | |
d0bfef31 | 2332 | int i; |
d7e09d03 PT |
2333 | |
2334 | pud = lustre_msg_buf(msg, offset, sizeof(*pud)); | |
2335 | if (!pud) | |
2336 | return -EINVAL; | |
2337 | ||
2338 | if (swabbed) { | |
2339 | __swab32s(&pud->pud_uid); | |
2340 | __swab32s(&pud->pud_gid); | |
2341 | __swab32s(&pud->pud_fsuid); | |
2342 | __swab32s(&pud->pud_fsgid); | |
2343 | __swab32s(&pud->pud_cap); | |
2344 | __swab32s(&pud->pud_ngroups); | |
2345 | } | |
2346 | ||
2347 | if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) { | |
2348 | CERROR("%u groups is too large\n", pud->pud_ngroups); | |
2349 | return -EINVAL; | |
2350 | } | |
2351 | ||
2352 | if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) > | |
2353 | msg->lm_buflens[offset]) { | |
2354 | CERROR("%u groups are claimed but bufsize only %u\n", | |
2355 | pud->pud_ngroups, msg->lm_buflens[offset]); | |
2356 | return -EINVAL; | |
2357 | } | |
2358 | ||
2359 | if (swabbed) { | |
2360 | for (i = 0; i < pud->pud_ngroups; i++) | |
2361 | __swab32s(&pud->pud_groups[i]); | |
2362 | } | |
2363 | ||
2364 | return 0; | |
2365 | } | |
2366 | EXPORT_SYMBOL(sptlrpc_unpack_user_desc); | |
2367 | ||
2368 | /**************************************** | |
2369 | * misc helpers * | |
2370 | ****************************************/ | |
2371 | ||
aff9d8e8 | 2372 | const char *sec2target_str(struct ptlrpc_sec *sec) |
d7e09d03 PT |
2373 | { |
2374 | if (!sec || !sec->ps_import || !sec->ps_import->imp_obd) | |
2375 | return "*"; | |
2376 | if (sec_is_reverse(sec)) | |
2377 | return "c"; | |
2378 | return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid); | |
2379 | } | |
2380 | EXPORT_SYMBOL(sec2target_str); | |
2381 | ||
2382 | /* | |
2383 | * return true if the bulk data is protected | |
2384 | */ | |
2385 | int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr) | |
2386 | { | |
2387 | switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) { | |
2388 | case SPTLRPC_BULK_SVC_INTG: | |
2389 | case SPTLRPC_BULK_SVC_PRIV: | |
2390 | return 1; | |
2391 | default: | |
2392 | return 0; | |
2393 | } | |
2394 | } | |
2395 | EXPORT_SYMBOL(sptlrpc_flavor_has_bulk); | |
2396 | ||
2397 | /**************************************** | |
2398 | * crypto API helper/alloc blkciper * | |
2399 | ****************************************/ | |
2400 | ||
2401 | /**************************************** | |
2402 | * initialize/finalize * | |
2403 | ****************************************/ | |
2404 | ||
2405 | int sptlrpc_init(void) | |
2406 | { | |
2407 | int rc; | |
2408 | ||
2409 | rwlock_init(&policy_lock); | |
2410 | ||
2411 | rc = sptlrpc_gc_init(); | |
2412 | if (rc) | |
2413 | goto out; | |
2414 | ||
2415 | rc = sptlrpc_conf_init(); | |
2416 | if (rc) | |
2417 | goto out_gc; | |
2418 | ||
2419 | rc = sptlrpc_enc_pool_init(); | |
2420 | if (rc) | |
2421 | goto out_conf; | |
2422 | ||
2423 | rc = sptlrpc_null_init(); | |
2424 | if (rc) | |
2425 | goto out_pool; | |
2426 | ||
2427 | rc = sptlrpc_plain_init(); | |
2428 | if (rc) | |
2429 | goto out_null; | |
2430 | ||
2431 | rc = sptlrpc_lproc_init(); | |
2432 | if (rc) | |
2433 | goto out_plain; | |
2434 | ||
2435 | return 0; | |
2436 | ||
2437 | out_plain: | |
2438 | sptlrpc_plain_fini(); | |
2439 | out_null: | |
2440 | sptlrpc_null_fini(); | |
2441 | out_pool: | |
2442 | sptlrpc_enc_pool_fini(); | |
2443 | out_conf: | |
2444 | sptlrpc_conf_fini(); | |
2445 | out_gc: | |
2446 | sptlrpc_gc_fini(); | |
2447 | out: | |
2448 | return rc; | |
2449 | } | |
2450 | ||
2451 | void sptlrpc_fini(void) | |
2452 | { | |
2453 | sptlrpc_lproc_fini(); | |
2454 | sptlrpc_plain_fini(); | |
2455 | sptlrpc_null_fini(); | |
2456 | sptlrpc_enc_pool_fini(); | |
2457 | sptlrpc_conf_fini(); | |
2458 | sptlrpc_gc_fini(); | |
2459 | } |