kernel/drivers/staging/lustre/lustre/ptlrpc/sec.c

   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
  43 #include "../../include/linux/libcfs/libcfs.h"
  44 #include <linux/crypto.h>
  45 #include <linux/key.h>
  46
  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"
  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
 113 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
 114 {
 115         static DEFINE_MUTEX(load_mutex);
 116         static atomic_t loaded = ATOMIC_INIT(0);
 117         struct ptlrpc_sec_policy *policy;
 118         __u16 number = SPTLRPC_FLVR_POLICY(flavor);
 119         __u16 flag = 0;
 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 {
 212         strlcpy(buf, sptlrpc_flavor2name_base(sf->sf_rpc), bufsize);
 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);
 223                 strlcat(buf, bspec, bufsize);
 224         }
 225
 226         return buf;
 227 }
 228 EXPORT_SYMBOL(sptlrpc_flavor2name);
 229
 230 static 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
 248 /**************************************************
 249  * client context APIs                      *
 250  **************************************************/
 251
 252 static
 253 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 254 {
 255         struct vfs_cred vcred;
 256         int create = 1, remove_dead = 1;
 257
 258         LASSERT(sec);
 259         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 260
 261         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 262                                      PTLRPC_SEC_FL_ROOTONLY)) {
 263                 vcred.vc_uid = 0;
 264                 vcred.vc_gid = 0;
 265                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 266                         create = 0;
 267                         remove_dead = 0;
 268                 }
 269         } else {
 270                 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
 271                 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
 272         }
 273
 274         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
 275                                                    create, remove_dead);
 276 }
 277
 278 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 279 {
 280         atomic_inc(&ctx->cc_refcount);
 281         return ctx;
 282 }
 283 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 284
 285 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 286 {
 287         struct ptlrpc_sec *sec = ctx->cc_sec;
 288
 289         LASSERT(sec);
 290         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 291
 292         if (!atomic_dec_and_test(&ctx->cc_refcount))
 293                 return;
 294
 295         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 296 }
 297 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 298
 299 static int import_sec_check_expire(struct obd_import *imp)
 300 {
 301         int adapt = 0;
 302
 303         spin_lock(&imp->imp_lock);
 304         if (imp->imp_sec_expire &&
 305             imp->imp_sec_expire < ktime_get_real_seconds()) {
 306                 adapt = 1;
 307                 imp->imp_sec_expire = 0;
 308         }
 309         spin_unlock(&imp->imp_lock);
 310
 311         if (!adapt)
 312                 return 0;
 313
 314         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 315         return sptlrpc_import_sec_adapt(imp, NULL, NULL);
 316 }
 317
 318 static int import_sec_validate_get(struct obd_import *imp,
 319                                    struct ptlrpc_sec **sec)
 320 {
 321         int rc;
 322
 323         if (unlikely(imp->imp_sec_expire)) {
 324                 rc = import_sec_check_expire(imp);
 325                 if (rc)
 326                         return rc;
 327         }
 328
 329         *sec = sptlrpc_import_sec_ref(imp);
 330         if (*sec == NULL) {
 331                 CERROR("import %p (%s) with no sec\n",
 332                        imp, ptlrpc_import_state_name(imp->imp_state));
 333                 return -EACCES;
 334         }
 335
 336         if (unlikely((*sec)->ps_dying)) {
 337                 CERROR("attempt to use dying sec %p\n", sec);
 338                 sptlrpc_sec_put(*sec);
 339                 return -EACCES;
 340         }
 341
 342         return 0;
 343 }
 344
 345 /**
 346  * Given a \a req, find or allocate a appropriate context for it.
 347  * \pre req->rq_cli_ctx == NULL.
 348  *
 349  * \retval 0 succeed, and req->rq_cli_ctx is set.
 350  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 351  */
 352 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 353 {
 354         struct obd_import *imp = req->rq_import;
 355         struct ptlrpc_sec *sec;
 356         int             rc;
 357
 358         LASSERT(!req->rq_cli_ctx);
 359         LASSERT(imp);
 360
 361         rc = import_sec_validate_get(imp, &sec);
 362         if (rc)
 363                 return rc;
 364
 365         req->rq_cli_ctx = get_my_ctx(sec);
 366
 367         sptlrpc_sec_put(sec);
 368
 369         if (!req->rq_cli_ctx) {
 370                 CERROR("req %p: fail to get context\n", req);
 371                 return -ENOMEM;
 372         }
 373
 374         return 0;
 375 }
 376
 377 /**
 378  * Drop the context for \a req.
 379  * \pre req->rq_cli_ctx != NULL.
 380  * \post req->rq_cli_ctx == NULL.
 381  *
 382  * If \a sync == 0, this function should return quickly without sleep;
 383  * otherwise it might trigger and wait for the whole process of sending
 384  * an context-destroying rpc to server.
 385  */
 386 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 387 {
 388         LASSERT(req);
 389         LASSERT(req->rq_cli_ctx);
 390
 391         /* request might be asked to release earlier while still
 392          * in the context waiting list.
 393          */
 394         if (!list_empty(&req->rq_ctx_chain)) {
 395                 spin_lock(&req->rq_cli_ctx->cc_lock);
 396                 list_del_init(&req->rq_ctx_chain);
 397                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 398         }
 399
 400         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 401         req->rq_cli_ctx = NULL;
 402 }
 403
 404 static
 405 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 406                            struct ptlrpc_cli_ctx *oldctx,
 407                            struct ptlrpc_cli_ctx *newctx)
 408 {
 409         struct sptlrpc_flavor old_flvr;
 410         char *reqmsg = NULL; /* to workaround old gcc */
 411         int reqmsg_size;
 412         int rc = 0;
 413
 414         LASSERT(req->rq_reqmsg);
 415         LASSERT(req->rq_reqlen);
 416         LASSERT(req->rq_replen);
 417
 418         CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
 419                req,
 420                oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
 421                newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
 422                oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
 423                newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 424
 425         /* save flavor */
 426         old_flvr = req->rq_flvr;
 427
 428         /* save request message */
 429         reqmsg_size = req->rq_reqlen;
 430         if (reqmsg_size != 0) {
 431                 reqmsg = libcfs_kvzalloc(reqmsg_size, GFP_NOFS);
 432                 if (reqmsg == NULL)
 433                         return -ENOMEM;
 434                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 435         }
 436
 437         /* release old req/rep buf */
 438         req->rq_cli_ctx = oldctx;
 439         sptlrpc_cli_free_reqbuf(req);
 440         sptlrpc_cli_free_repbuf(req);
 441         req->rq_cli_ctx = newctx;
 442
 443         /* recalculate the flavor */
 444         sptlrpc_req_set_flavor(req, 0);
 445
 446         /* alloc new request buffer
 447          * we don't need to alloc reply buffer here, leave it to the
 448          * rest procedure of ptlrpc */
 449         if (reqmsg_size != 0) {
 450                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 451                 if (!rc) {
 452                         LASSERT(req->rq_reqmsg);
 453                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 454                 } else {
 455                         CWARN("failed to alloc reqbuf: %d\n", rc);
 456                         req->rq_flvr = old_flvr;
 457                 }
 458
 459                 kvfree(reqmsg);
 460         }
 461         return rc;
 462 }
 463
 464 /**
 465  * If current context of \a req is dead somehow, e.g. we just switched flavor
 466  * thus marked original contexts dead, we'll find a new context for it. if
 467  * no switch is needed, \a req will end up with the same context.
 468  *
 469  * \note a request must have a context, to keep other parts of code happy.
 470  * In any case of failure during the switching, we must restore the old one.
 471  */
 472 static int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 473 {
 474         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 475         struct ptlrpc_cli_ctx *newctx;
 476         int rc;
 477
 478         LASSERT(oldctx);
 479
 480         sptlrpc_cli_ctx_get(oldctx);
 481         sptlrpc_req_put_ctx(req, 0);
 482
 483         rc = sptlrpc_req_get_ctx(req);
 484         if (unlikely(rc)) {
 485                 LASSERT(!req->rq_cli_ctx);
 486
 487                 /* restore old ctx */
 488                 req->rq_cli_ctx = oldctx;
 489                 return rc;
 490         }
 491
 492         newctx = req->rq_cli_ctx;
 493         LASSERT(newctx);
 494
 495         if (unlikely(newctx == oldctx &&
 496                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 497                 /*
 498                  * still get the old dead ctx, usually means system too busy
 499                  */
 500                 CDEBUG(D_SEC,
 501                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 502                        newctx, newctx->cc_flags);
 503
 504                 set_current_state(TASK_INTERRUPTIBLE);
 505                 schedule_timeout(HZ);
 506         } else {
 507                 /*
 508                  * it's possible newctx == oldctx if we're switching
 509                  * subflavor with the same sec.
 510                  */
 511                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 512                 if (rc) {
 513                         /* restore old ctx */
 514                         sptlrpc_req_put_ctx(req, 0);
 515                         req->rq_cli_ctx = oldctx;
 516                         return rc;
 517                 }
 518
 519                 LASSERT(req->rq_cli_ctx == newctx);
 520         }
 521
 522         sptlrpc_cli_ctx_put(oldctx, 1);
 523         return 0;
 524 }
 525
 526 static
 527 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 528 {
 529         if (cli_ctx_is_refreshed(ctx))
 530                 return 1;
 531         return 0;
 532 }
 533
 534 static
 535 int ctx_refresh_timeout(void *data)
 536 {
 537         struct ptlrpc_request *req = data;
 538         int rc;
 539
 540         /* conn_cnt is needed in expire_one_request */
 541         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 542
 543         rc = ptlrpc_expire_one_request(req, 1);
 544         /* if we started recovery, we should mark this ctx dead; otherwise
 545          * in case of lgssd died nobody would retire this ctx, following
 546          * connecting will still find the same ctx thus cause deadlock.
 547          * there's an assumption that expire time of the request should be
 548          * later than the context refresh expire time.
 549          */
 550         if (rc == 0)
 551                 req->rq_cli_ctx->cc_ops->force_die(req->rq_cli_ctx, 0);
 552         return rc;
 553 }
 554
 555 static
 556 void ctx_refresh_interrupt(void *data)
 557 {
 558         struct ptlrpc_request *req = data;
 559
 560         spin_lock(&req->rq_lock);
 561         req->rq_intr = 1;
 562         spin_unlock(&req->rq_lock);
 563 }
 564
 565 static
 566 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 567 {
 568         spin_lock(&ctx->cc_lock);
 569         if (!list_empty(&req->rq_ctx_chain))
 570                 list_del_init(&req->rq_ctx_chain);
 571         spin_unlock(&ctx->cc_lock);
 572 }
 573
 574 /**
 575  * To refresh the context of \req, if it's not up-to-date.
 576  * \param timeout
 577  * - < 0: don't wait
 578  * - = 0: wait until success or fatal error occur
 579  * - > 0: timeout value (in seconds)
 580  *
 581  * The status of the context could be subject to be changed by other threads
 582  * at any time. We allow this race, but once we return with 0, the caller will
 583  * suppose it's uptodated and keep using it until the owning rpc is done.
 584  *
 585  * \retval 0 only if the context is uptodated.
 586  * \retval -ev error number.
 587  */
 588 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 589 {
 590         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 591         struct ptlrpc_sec *sec;
 592         struct l_wait_info lwi;
 593         int rc;
 594
 595         LASSERT(ctx);
 596
 597         if (req->rq_ctx_init || req->rq_ctx_fini)
 598                 return 0;
 599
 600         /*
 601          * during the process a request's context might change type even
 602          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 603          * everything
 604          */
 605 again:
 606         rc = import_sec_validate_get(req->rq_import, &sec);
 607         if (rc)
 608                 return rc;
 609
 610         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 611                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 612                       req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 613                 req_off_ctx_list(req, ctx);
 614                 sptlrpc_req_replace_dead_ctx(req);
 615                 ctx = req->rq_cli_ctx;
 616         }
 617         sptlrpc_sec_put(sec);
 618
 619         if (cli_ctx_is_eternal(ctx))
 620                 return 0;
 621
 622         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 623                 LASSERT(ctx->cc_ops->refresh);
 624                 ctx->cc_ops->refresh(ctx);
 625         }
 626         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 627
 628         LASSERT(ctx->cc_ops->validate);
 629         if (ctx->cc_ops->validate(ctx) == 0) {
 630                 req_off_ctx_list(req, ctx);
 631                 return 0;
 632         }
 633
 634         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 635                 spin_lock(&req->rq_lock);
 636                 req->rq_err = 1;
 637                 spin_unlock(&req->rq_lock);
 638                 req_off_ctx_list(req, ctx);
 639                 return -EPERM;
 640         }
 641
 642         /*
 643          * There's a subtle issue for resending RPCs, suppose following
 644          * situation:
 645          *  1. the request was sent to server.
 646          *  2. recovery was kicked start, after finished the request was
 647          *     marked as resent.
 648          *  3. resend the request.
 649          *  4. old reply from server received, we accept and verify the reply.
 650          *     this has to be success, otherwise the error will be aware
 651          *     by application.
 652          *  5. new reply from server received, dropped by LNet.
 653          *
 654          * Note the xid of old & new request is the same. We can't simply
 655          * change xid for the resent request because the server replies on
 656          * it for reply reconstruction.
 657          *
 658          * Commonly the original context should be uptodate because we
 659          * have a expiry nice time; server will keep its context because
 660          * we at least hold a ref of old context which prevent context
 661          * destroying RPC being sent. So server still can accept the request
 662          * and finish the RPC. But if that's not the case:
 663          *  1. If server side context has been trimmed, a NO_CONTEXT will
 664          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 665          *     context by force.
 666          *  2. Current context never be refreshed, then we are fine: we
 667          *     never really send request with old context before.
 668          */
 669         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 670             unlikely(req->rq_reqmsg) &&
 671             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 672                 req_off_ctx_list(req, ctx);
 673                 return 0;
 674         }
 675
 676         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 677                 req_off_ctx_list(req, ctx);
 678                 /*
 679                  * don't switch ctx if import was deactivated
 680                  */
 681                 if (req->rq_import->imp_deactive) {
 682                         spin_lock(&req->rq_lock);
 683                         req->rq_err = 1;
 684                         spin_unlock(&req->rq_lock);
 685                         return -EINTR;
 686                 }
 687
 688                 rc = sptlrpc_req_replace_dead_ctx(req);
 689                 if (rc) {
 690                         LASSERT(ctx == req->rq_cli_ctx);
 691                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 692                                req, ctx, rc);
 693                         spin_lock(&req->rq_lock);
 694                         req->rq_err = 1;
 695                         spin_unlock(&req->rq_lock);
 696                         return rc;
 697                 }
 698
 699                 ctx = req->rq_cli_ctx;
 700                 goto again;
 701         }
 702
 703         /*
 704          * Now we're sure this context is during upcall, add myself into
 705          * waiting list
 706          */
 707         spin_lock(&ctx->cc_lock);
 708         if (list_empty(&req->rq_ctx_chain))
 709                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 710         spin_unlock(&ctx->cc_lock);
 711
 712         if (timeout < 0)
 713                 return -EWOULDBLOCK;
 714
 715         /* Clear any flags that may be present from previous sends */
 716         LASSERT(req->rq_receiving_reply == 0);
 717         spin_lock(&req->rq_lock);
 718         req->rq_err = 0;
 719         req->rq_timedout = 0;
 720         req->rq_resend = 0;
 721         req->rq_restart = 0;
 722         spin_unlock(&req->rq_lock);
 723
 724         lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout,
 725                                ctx_refresh_interrupt, req);
 726         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 727
 728         /*
 729          * following cases could lead us here:
 730          * - successfully refreshed;
 731          * - interrupted;
 732          * - timedout, and we don't want recover from the failure;
 733          * - timedout, and waked up upon recovery finished;
 734          * - someone else mark this ctx dead by force;
 735          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 736          *   e.g. ptlrpc_abort_inflight();
 737          */
 738         if (!cli_ctx_is_refreshed(ctx)) {
 739                 /* timed out or interrupted */
 740                 req_off_ctx_list(req, ctx);
 741
 742                 LASSERT(rc != 0);
 743                 return rc;
 744         }
 745
 746         goto again;
 747 }
 748
 749 /**
 750  * Initialize flavor settings for \a req, according to \a opcode.
 751  *
 752  * \note this could be called in two situations:
 753  * - new request from ptlrpc_pre_req(), with proper @opcode
 754  * - old request which changed ctx in the middle, with @opcode == 0
 755  */
 756 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 757 {
 758         struct ptlrpc_sec *sec;
 759
 760         LASSERT(req->rq_import);
 761         LASSERT(req->rq_cli_ctx);
 762         LASSERT(req->rq_cli_ctx->cc_sec);
 763         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 764
 765         /* special security flags according to opcode */
 766         switch (opcode) {
 767         case OST_READ:
 768         case MDS_READPAGE:
 769         case MGS_CONFIG_READ:
 770         case OBD_IDX_READ:
 771                 req->rq_bulk_read = 1;
 772                 break;
 773         case OST_WRITE:
 774         case MDS_WRITEPAGE:
 775                 req->rq_bulk_write = 1;
 776                 break;
 777         case SEC_CTX_INIT:
 778                 req->rq_ctx_init = 1;
 779                 break;
 780         case SEC_CTX_FINI:
 781                 req->rq_ctx_fini = 1;
 782                 break;
 783         case 0:
 784                 /* init/fini rpc won't be resend, so can't be here */
 785                 LASSERT(req->rq_ctx_init == 0);
 786                 LASSERT(req->rq_ctx_fini == 0);
 787
 788                 /* cleanup flags, which should be recalculated */
 789                 req->rq_pack_udesc = 0;
 790                 req->rq_pack_bulk = 0;
 791                 break;
 792         }
 793
 794         sec = req->rq_cli_ctx->cc_sec;
 795
 796         spin_lock(&sec->ps_lock);
 797         req->rq_flvr = sec->ps_flvr;
 798         spin_unlock(&sec->ps_lock);
 799
 800         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
 801          * destruction rpc */
 802         if (unlikely(req->rq_ctx_init))
 803                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 804         else if (unlikely(req->rq_ctx_fini))
 805                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 806
 807         /* user descriptor flag, null security can't do it anyway */
 808         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 809             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 810                 req->rq_pack_udesc = 1;
 811
 812         /* bulk security flag */
 813         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 814             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 815                 req->rq_pack_bulk = 1;
 816 }
 817
 818 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 819 {
 820         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 821                 return;
 822
 823         LASSERT(req->rq_clrbuf);
 824         if (req->rq_pool || !req->rq_reqbuf)
 825                 return;
 826
 827         kfree(req->rq_reqbuf);
 828         req->rq_reqbuf = NULL;
 829         req->rq_reqbuf_len = 0;
 830 }
 831
 832 /**
 833  * Given an import \a imp, check whether current user has a valid context
 834  * or not. We may create a new context and try to refresh it, and try
 835  * repeatedly try in case of non-fatal errors. Return 0 means success.
 836  */
 837 int sptlrpc_import_check_ctx(struct obd_import *imp)
 838 {
 839         struct ptlrpc_sec *sec;
 840         struct ptlrpc_cli_ctx *ctx;
 841         struct ptlrpc_request *req = NULL;
 842         int rc;
 843
 844         might_sleep();
 845
 846         sec = sptlrpc_import_sec_ref(imp);
 847         ctx = get_my_ctx(sec);
 848         sptlrpc_sec_put(sec);
 849
 850         if (!ctx)
 851                 return -ENOMEM;
 852
 853         if (cli_ctx_is_eternal(ctx) ||
 854             ctx->cc_ops->validate(ctx) == 0) {
 855                 sptlrpc_cli_ctx_put(ctx, 1);
 856                 return 0;
 857         }
 858
 859         if (cli_ctx_is_error(ctx)) {
 860                 sptlrpc_cli_ctx_put(ctx, 1);
 861                 return -EACCES;
 862         }
 863
 864         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 865         if (!req)
 866                 return -ENOMEM;
 867
 868         spin_lock_init(&req->rq_lock);
 869         atomic_set(&req->rq_refcount, 10000);
 870         INIT_LIST_HEAD(&req->rq_ctx_chain);
 871         init_waitqueue_head(&req->rq_reply_waitq);
 872         init_waitqueue_head(&req->rq_set_waitq);
 873         req->rq_import = imp;
 874         req->rq_flvr = sec->ps_flvr;
 875         req->rq_cli_ctx = ctx;
 876
 877         rc = sptlrpc_req_refresh_ctx(req, 0);
 878         LASSERT(list_empty(&req->rq_ctx_chain));
 879         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 880         ptlrpc_request_cache_free(req);
 881
 882         return rc;
 883 }
 884
 885 /**
 886  * Used by ptlrpc client, to perform the pre-defined security transformation
 887  * upon the request message of \a req. After this function called,
 888  * req->rq_reqmsg is still accessible as clear text.
 889  */
 890 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 891 {
 892         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 893         int rc = 0;
 894
 895         LASSERT(ctx);
 896         LASSERT(ctx->cc_sec);
 897         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
 898
 899         /* we wrap bulk request here because now we can be sure
 900          * the context is uptodate.
 901          */
 902         if (req->rq_bulk) {
 903                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
 904                 if (rc)
 905                         return rc;
 906         }
 907
 908         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 909         case SPTLRPC_SVC_NULL:
 910         case SPTLRPC_SVC_AUTH:
 911         case SPTLRPC_SVC_INTG:
 912                 LASSERT(ctx->cc_ops->sign);
 913                 rc = ctx->cc_ops->sign(ctx, req);
 914                 break;
 915         case SPTLRPC_SVC_PRIV:
 916                 LASSERT(ctx->cc_ops->seal);
 917                 rc = ctx->cc_ops->seal(ctx, req);
 918                 break;
 919         default:
 920                 LBUG();
 921         }
 922
 923         if (rc == 0) {
 924                 LASSERT(req->rq_reqdata_len);
 925                 LASSERT(req->rq_reqdata_len % 8 == 0);
 926                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
 927         }
 928
 929         return rc;
 930 }
 931
 932 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
 933 {
 934         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 935         int rc;
 936
 937         LASSERT(ctx);
 938         LASSERT(ctx->cc_sec);
 939         LASSERT(req->rq_repbuf);
 940         LASSERT(req->rq_repdata);
 941         LASSERT(req->rq_repmsg == NULL);
 942
 943         req->rq_rep_swab_mask = 0;
 944
 945         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
 946         switch (rc) {
 947         case 1:
 948                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
 949         case 0:
 950                 break;
 951         default:
 952                 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
 953                 return -EPROTO;
 954         }
 955
 956         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
 957                 CERROR("replied data length %d too small\n",
 958                        req->rq_repdata_len);
 959                 return -EPROTO;
 960         }
 961
 962         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
 963             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
 964                 CERROR("reply policy %u doesn't match request policy %u\n",
 965                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
 966                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
 967                 return -EPROTO;
 968         }
 969
 970         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 971         case SPTLRPC_SVC_NULL:
 972         case SPTLRPC_SVC_AUTH:
 973         case SPTLRPC_SVC_INTG:
 974                 LASSERT(ctx->cc_ops->verify);
 975                 rc = ctx->cc_ops->verify(ctx, req);
 976                 break;
 977         case SPTLRPC_SVC_PRIV:
 978                 LASSERT(ctx->cc_ops->unseal);
 979                 rc = ctx->cc_ops->unseal(ctx, req);
 980                 break;
 981         default:
 982                 LBUG();
 983         }
 984         LASSERT(rc || req->rq_repmsg || req->rq_resend);
 985
 986         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
 987             !req->rq_ctx_init)
 988                 req->rq_rep_swab_mask = 0;
 989         return rc;
 990 }
 991
 992 /**
 993  * Used by ptlrpc client, to perform security transformation upon the reply
 994  * message of \a req. After return successfully, req->rq_repmsg points to
 995  * the reply message in clear text.
 996  *
 997  * \pre the reply buffer should have been un-posted from LNet, so nothing is
 998  * going to change.
 999  */
1000 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1001 {
1002         LASSERT(req->rq_repbuf);
1003         LASSERT(req->rq_repdata == NULL);
1004         LASSERT(req->rq_repmsg == NULL);
1005         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1006
1007         if (req->rq_reply_off == 0 &&
1008             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1009                 CERROR("real reply with offset 0\n");
1010                 return -EPROTO;
1011         }
1012
1013         if (req->rq_reply_off % 8 != 0) {
1014                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1015                 return -EPROTO;
1016         }
1017
1018         req->rq_repdata = (struct lustre_msg *)
1019                                 (req->rq_repbuf + req->rq_reply_off);
1020         req->rq_repdata_len = req->rq_nob_received;
1021
1022         return do_cli_unwrap_reply(req);
1023 }
1024
1025 /**
1026  * Used by ptlrpc client, to perform security transformation upon the early
1027  * reply message of \a req. We expect the rq_reply_off is 0, and
1028  * rq_nob_received is the early reply size.
1029  *
1030  * Because the receive buffer might be still posted, the reply data might be
1031  * changed at any time, no matter we're holding rq_lock or not. For this reason
1032  * we allocate a separate ptlrpc_request and reply buffer for early reply
1033  * processing.
1034  *
1035  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1036  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1037  * \a *req_ret to release it.
1038  * \retval -ev error number, and \a req_ret will not be set.
1039  */
1040 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1041                                    struct ptlrpc_request **req_ret)
1042 {
1043         struct ptlrpc_request *early_req;
1044         char *early_buf;
1045         int early_bufsz, early_size;
1046         int rc;
1047
1048         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1049         if (early_req == NULL)
1050                 return -ENOMEM;
1051
1052         early_size = req->rq_nob_received;
1053         early_bufsz = size_roundup_power2(early_size);
1054         early_buf = libcfs_kvzalloc(early_bufsz, GFP_NOFS);
1055         if (early_buf == NULL) {
1056                 rc = -ENOMEM;
1057                 goto err_req;
1058         }
1059
1060         /* sanity checkings and copy data out, do it inside spinlock */
1061         spin_lock(&req->rq_lock);
1062
1063         if (req->rq_replied) {
1064                 spin_unlock(&req->rq_lock);
1065                 rc = -EALREADY;
1066                 goto err_buf;
1067         }
1068
1069         LASSERT(req->rq_repbuf);
1070         LASSERT(req->rq_repdata == NULL);
1071         LASSERT(req->rq_repmsg == NULL);
1072
1073         if (req->rq_reply_off != 0) {
1074                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1075                 spin_unlock(&req->rq_lock);
1076                 rc = -EPROTO;
1077                 goto err_buf;
1078         }
1079
1080         if (req->rq_nob_received != early_size) {
1081                 /* even another early arrived the size should be the same */
1082                 CERROR("data size has changed from %u to %u\n",
1083                        early_size, req->rq_nob_received);
1084                 spin_unlock(&req->rq_lock);
1085                 rc = -EINVAL;
1086                 goto err_buf;
1087         }
1088
1089         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1090                 CERROR("early reply length %d too small\n",
1091                        req->rq_nob_received);
1092                 spin_unlock(&req->rq_lock);
1093                 rc = -EALREADY;
1094                 goto err_buf;
1095         }
1096
1097         memcpy(early_buf, req->rq_repbuf, early_size);
1098         spin_unlock(&req->rq_lock);
1099
1100         spin_lock_init(&early_req->rq_lock);
1101         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1102         early_req->rq_flvr = req->rq_flvr;
1103         early_req->rq_repbuf = early_buf;
1104         early_req->rq_repbuf_len = early_bufsz;
1105         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1106         early_req->rq_repdata_len = early_size;
1107         early_req->rq_early = 1;
1108         early_req->rq_reqmsg = req->rq_reqmsg;
1109
1110         rc = do_cli_unwrap_reply(early_req);
1111         if (rc) {
1112                 DEBUG_REQ(D_ADAPTTO, early_req,
1113                           "error %d unwrap early reply", rc);
1114                 goto err_ctx;
1115         }
1116
1117         LASSERT(early_req->rq_repmsg);
1118         *req_ret = early_req;
1119         return 0;
1120
1121 err_ctx:
1122         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1123 err_buf:
1124         kvfree(early_buf);
1125 err_req:
1126         ptlrpc_request_cache_free(early_req);
1127         return rc;
1128 }
1129
1130 /**
1131  * Used by ptlrpc client, to release a processed early reply \a early_req.
1132  *
1133  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1134  */
1135 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1136 {
1137         LASSERT(early_req->rq_repbuf);
1138         LASSERT(early_req->rq_repdata);
1139         LASSERT(early_req->rq_repmsg);
1140
1141         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1142         kvfree(early_req->rq_repbuf);
1143         ptlrpc_request_cache_free(early_req);
1144 }
1145
1146 /**************************************************
1147  * sec ID                                        *
1148  **************************************************/
1149
1150 /*
1151  * "fixed" sec (e.g. null) use sec_id < 0
1152  */
1153 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1154
1155 int sptlrpc_get_next_secid(void)
1156 {
1157         return atomic_inc_return(&sptlrpc_sec_id);
1158 }
1159 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1160
1161 /**************************************************
1162  * client side high-level security APIs    *
1163  **************************************************/
1164
1165 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1166                                    int grace, int force)
1167 {
1168         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1169
1170         LASSERT(policy->sp_cops);
1171         LASSERT(policy->sp_cops->flush_ctx_cache);
1172
1173         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1174 }
1175
1176 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1177 {
1178         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1179
1180         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1181         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1182         LASSERT(policy->sp_cops->destroy_sec);
1183
1184         CDEBUG(D_SEC, "%s@%p: being destroyed\n", sec->ps_policy->sp_name, sec);
1185
1186         policy->sp_cops->destroy_sec(sec);
1187         sptlrpc_policy_put(policy);
1188 }
1189
1190 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1191 {
1192         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1193
1194         if (sec->ps_policy->sp_cops->kill_sec) {
1195                 sec->ps_policy->sp_cops->kill_sec(sec);
1196
1197                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1198         }
1199 }
1200
1201 static struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1202 {
1203         if (sec)
1204                 atomic_inc(&sec->ps_refcount);
1205
1206         return sec;
1207 }
1208
1209 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1210 {
1211         if (sec) {
1212                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1213
1214                 if (atomic_dec_and_test(&sec->ps_refcount)) {
1215                         sptlrpc_gc_del_sec(sec);
1216                         sec_cop_destroy_sec(sec);
1217                 }
1218         }
1219 }
1220 EXPORT_SYMBOL(sptlrpc_sec_put);
1221
1222 /*
1223  * policy module is responsible for taking reference of import
1224  */
1225 static
1226 struct ptlrpc_sec *sptlrpc_sec_create(struct obd_import *imp,
1227                                       struct ptlrpc_svc_ctx *svc_ctx,
1228                                       struct sptlrpc_flavor *sf,
1229                                       enum lustre_sec_part sp)
1230 {
1231         struct ptlrpc_sec_policy *policy;
1232         struct ptlrpc_sec *sec;
1233         char str[32];
1234
1235         if (svc_ctx) {
1236                 LASSERT(imp->imp_dlm_fake == 1);
1237
1238                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1239                        imp->imp_obd->obd_type->typ_name,
1240                        imp->imp_obd->obd_name,
1241                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1242
1243                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1244                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1245         } else {
1246                 LASSERT(imp->imp_dlm_fake == 0);
1247
1248                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1249                        imp->imp_obd->obd_type->typ_name,
1250                        imp->imp_obd->obd_name,
1251                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1252
1253                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1254                 if (!policy) {
1255                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1256                         return NULL;
1257                 }
1258         }
1259
1260         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1261         if (sec) {
1262                 atomic_inc(&sec->ps_refcount);
1263
1264                 sec->ps_part = sp;
1265
1266                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1267                         sptlrpc_gc_add_sec(sec);
1268         } else {
1269                 sptlrpc_policy_put(policy);
1270         }
1271
1272         return sec;
1273 }
1274
1275 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1276 {
1277         struct ptlrpc_sec *sec;
1278
1279         spin_lock(&imp->imp_lock);
1280         sec = sptlrpc_sec_get(imp->imp_sec);
1281         spin_unlock(&imp->imp_lock);
1282
1283         return sec;
1284 }
1285 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1286
1287 static void sptlrpc_import_sec_install(struct obd_import *imp,
1288                                        struct ptlrpc_sec *sec)
1289 {
1290         struct ptlrpc_sec *old_sec;
1291
1292         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1293
1294         spin_lock(&imp->imp_lock);
1295         old_sec = imp->imp_sec;
1296         imp->imp_sec = sec;
1297         spin_unlock(&imp->imp_lock);
1298
1299         if (old_sec) {
1300                 sptlrpc_sec_kill(old_sec);
1301
1302                 /* balance the ref taken by this import */
1303                 sptlrpc_sec_put(old_sec);
1304         }
1305 }
1306
1307 static inline
1308 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1309 {
1310         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1311 }
1312
1313 static inline
1314 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1315 {
1316         *dst = *src;
1317 }
1318
1319 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1320                                              struct ptlrpc_sec *sec,
1321                                              struct sptlrpc_flavor *sf)
1322 {
1323         char str1[32], str2[32];
1324
1325         if (sec->ps_flvr.sf_flags != sf->sf_flags)
1326                 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1327                        sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1328                                             str1, sizeof(str1)),
1329                        sptlrpc_secflags2str(sf->sf_flags,
1330                                             str2, sizeof(str2)));
1331
1332         spin_lock(&sec->ps_lock);
1333         flavor_copy(&sec->ps_flvr, sf);
1334         spin_unlock(&sec->ps_lock);
1335 }
1336
1337 /**
1338  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1339  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1340  *
1341  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1342  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1343  */
1344 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1345                              struct ptlrpc_svc_ctx *svc_ctx,
1346                              struct sptlrpc_flavor *flvr)
1347 {
1348         struct ptlrpc_connection *conn;
1349         struct sptlrpc_flavor sf;
1350         struct ptlrpc_sec *sec, *newsec;
1351         enum lustre_sec_part sp;
1352         char str[24];
1353         int rc = 0;
1354
1355         might_sleep();
1356
1357         if (imp == NULL)
1358                 return 0;
1359
1360         conn = imp->imp_connection;
1361
1362         if (svc_ctx == NULL) {
1363                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1364                 /*
1365                  * normal import, determine flavor from rule set, except
1366                  * for mgc the flavor is predetermined.
1367                  */
1368                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1369                         sf = cliobd->cl_flvr_mgc;
1370                 else
1371                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1372                                                    cliobd->cl_sp_to,
1373                                                    &cliobd->cl_target_uuid,
1374                                                    conn->c_self, &sf);
1375
1376                 sp = imp->imp_obd->u.cli.cl_sp_me;
1377         } else {
1378                 /* reverse import, determine flavor from incoming request */
1379                 sf = *flvr;
1380
1381                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1382                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1383                                       PTLRPC_SEC_FL_ROOTONLY;
1384
1385                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1386         }
1387
1388         sec = sptlrpc_import_sec_ref(imp);
1389         if (sec) {
1390                 char str2[24];
1391
1392                 if (flavor_equal(&sf, &sec->ps_flvr))
1393                         goto out;
1394
1395                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1396                        imp->imp_obd->obd_name,
1397                        obd_uuid2str(&conn->c_remote_uuid),
1398                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1399                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1400
1401                 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1402                     SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1403                     SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1404                     SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1405                         sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1406                         goto out;
1407                 }
1408         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1409                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1410                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1411                        imp->imp_obd->obd_name,
1412                        obd_uuid2str(&conn->c_remote_uuid),
1413                        LNET_NIDNET(conn->c_self),
1414                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1415         }
1416
1417         mutex_lock(&imp->imp_sec_mutex);
1418
1419         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1420         if (newsec) {
1421                 sptlrpc_import_sec_install(imp, newsec);
1422         } else {
1423                 CERROR("import %s->%s: failed to create new sec\n",
1424                        imp->imp_obd->obd_name,
1425                        obd_uuid2str(&conn->c_remote_uuid));
1426                 rc = -EPERM;
1427         }
1428
1429         mutex_unlock(&imp->imp_sec_mutex);
1430 out:
1431         sptlrpc_sec_put(sec);
1432         return rc;
1433 }
1434
1435 void sptlrpc_import_sec_put(struct obd_import *imp)
1436 {
1437         if (imp->imp_sec) {
1438                 sptlrpc_sec_kill(imp->imp_sec);
1439
1440                 sptlrpc_sec_put(imp->imp_sec);
1441                 imp->imp_sec = NULL;
1442         }
1443 }
1444
1445 static void import_flush_ctx_common(struct obd_import *imp,
1446                                     uid_t uid, int grace, int force)
1447 {
1448         struct ptlrpc_sec *sec;
1449
1450         if (imp == NULL)
1451                 return;
1452
1453         sec = sptlrpc_import_sec_ref(imp);
1454         if (sec == NULL)
1455                 return;
1456
1457         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1458         sptlrpc_sec_put(sec);
1459 }
1460
1461 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1462 {
1463         import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1464                                 1, 1);
1465 }
1466 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1467
1468 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1469 {
1470         import_flush_ctx_common(imp, -1, 1, 1);
1471 }
1472 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1473
1474 /**
1475  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1476  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1477  */
1478 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1479 {
1480         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1481         struct ptlrpc_sec_policy *policy;
1482         int rc;
1483
1484         LASSERT(ctx);
1485         LASSERT(ctx->cc_sec);
1486         LASSERT(ctx->cc_sec->ps_policy);
1487         LASSERT(req->rq_reqmsg == NULL);
1488         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1489
1490         policy = ctx->cc_sec->ps_policy;
1491         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1492         if (!rc) {
1493                 LASSERT(req->rq_reqmsg);
1494                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1495
1496                 /* zeroing preallocated buffer */
1497                 if (req->rq_pool)
1498                         memset(req->rq_reqmsg, 0, msgsize);
1499         }
1500
1501         return rc;
1502 }
1503
1504 /**
1505  * Used by ptlrpc client to free request buffer of \a req. After this
1506  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1507  */
1508 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1509 {
1510         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1511         struct ptlrpc_sec_policy *policy;
1512
1513         LASSERT(ctx);
1514         LASSERT(ctx->cc_sec);
1515         LASSERT(ctx->cc_sec->ps_policy);
1516         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1517
1518         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1519                 return;
1520
1521         policy = ctx->cc_sec->ps_policy;
1522         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1523         req->rq_reqmsg = NULL;
1524 }
1525
1526 /*
1527  * NOTE caller must guarantee the buffer size is enough for the enlargement
1528  */
1529 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1530                                   int segment, int newsize)
1531 {
1532         void *src, *dst;
1533         int oldsize, oldmsg_size, movesize;
1534
1535         LASSERT(segment < msg->lm_bufcount);
1536         LASSERT(msg->lm_buflens[segment] <= newsize);
1537
1538         if (msg->lm_buflens[segment] == newsize)
1539                 return;
1540
1541         /* nothing to do if we are enlarging the last segment */
1542         if (segment == msg->lm_bufcount - 1) {
1543                 msg->lm_buflens[segment] = newsize;
1544                 return;
1545         }
1546
1547         oldsize = msg->lm_buflens[segment];
1548
1549         src = lustre_msg_buf(msg, segment + 1, 0);
1550         msg->lm_buflens[segment] = newsize;
1551         dst = lustre_msg_buf(msg, segment + 1, 0);
1552         msg->lm_buflens[segment] = oldsize;
1553
1554         /* move from segment + 1 to end segment */
1555         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1556         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1557         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1558         LASSERT(movesize >= 0);
1559
1560         if (movesize)
1561                 memmove(dst, src, movesize);
1562
1563         /* note we don't clear the ares where old data live, not secret */
1564
1565         /* finally set new segment size */
1566         msg->lm_buflens[segment] = newsize;
1567 }
1568 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1569
1570 /**
1571  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1572  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1573  * preserved after the enlargement. this must be called after original request
1574  * buffer being allocated.
1575  *
1576  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1577  * so caller should refresh its local pointers if needed.
1578  */
1579 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1580                                int segment, int newsize)
1581 {
1582         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1583         struct ptlrpc_sec_cops *cops;
1584         struct lustre_msg *msg = req->rq_reqmsg;
1585
1586         LASSERT(ctx);
1587         LASSERT(msg);
1588         LASSERT(msg->lm_bufcount > segment);
1589         LASSERT(msg->lm_buflens[segment] <= newsize);
1590
1591         if (msg->lm_buflens[segment] == newsize)
1592                 return 0;
1593
1594         cops = ctx->cc_sec->ps_policy->sp_cops;
1595         LASSERT(cops->enlarge_reqbuf);
1596         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1597 }
1598 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1599
1600 /**
1601  * Used by ptlrpc client to allocate reply buffer of \a req.
1602  *
1603  * \note After this, req->rq_repmsg is still not accessible.
1604  */
1605 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1606 {
1607         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1608         struct ptlrpc_sec_policy *policy;
1609
1610         LASSERT(ctx);
1611         LASSERT(ctx->cc_sec);
1612         LASSERT(ctx->cc_sec->ps_policy);
1613
1614         if (req->rq_repbuf)
1615                 return 0;
1616
1617         policy = ctx->cc_sec->ps_policy;
1618         return policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize);
1619 }
1620
1621 /**
1622  * Used by ptlrpc client to free reply buffer of \a req. After this
1623  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1624  */
1625 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1626 {
1627         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1628         struct ptlrpc_sec_policy *policy;
1629
1630         LASSERT(ctx);
1631         LASSERT(ctx->cc_sec);
1632         LASSERT(ctx->cc_sec->ps_policy);
1633         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1634
1635         if (req->rq_repbuf == NULL)
1636                 return;
1637         LASSERT(req->rq_repbuf_len);
1638
1639         policy = ctx->cc_sec->ps_policy;
1640         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1641         req->rq_repmsg = NULL;
1642 }
1643
1644 static int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1645                                        struct ptlrpc_svc_ctx *ctx)
1646 {
1647         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1648
1649         if (!policy->sp_sops->install_rctx)
1650                 return 0;
1651         return policy->sp_sops->install_rctx(imp, ctx);
1652 }
1653
1654 /****************************************
1655  * server side security          *
1656  ****************************************/
1657
1658 static int flavor_allowed(struct sptlrpc_flavor *exp,
1659                           struct ptlrpc_request *req)
1660 {
1661         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1662
1663         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1664                 return 1;
1665
1666         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1667             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1668             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1669             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1670                 return 1;
1671
1672         return 0;
1673 }
1674
1675 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1676
1677 /**
1678  * Given an export \a exp, check whether the flavor of incoming \a req
1679  * is allowed by the export \a exp. Main logic is about taking care of
1680  * changing configurations. Return 0 means success.
1681  */
1682 int sptlrpc_target_export_check(struct obd_export *exp,
1683                                 struct ptlrpc_request *req)
1684 {
1685         struct sptlrpc_flavor flavor;
1686
1687         if (exp == NULL)
1688                 return 0;
1689
1690         /* client side export has no imp_reverse, skip
1691          * FIXME maybe we should check flavor this as well??? */
1692         if (exp->exp_imp_reverse == NULL)
1693                 return 0;
1694
1695         /* don't care about ctx fini rpc */
1696         if (req->rq_ctx_fini)
1697                 return 0;
1698
1699         spin_lock(&exp->exp_lock);
1700
1701         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1702          * the first req with the new flavor, then treat it as current flavor,
1703          * adapt reverse sec according to it.
1704          * note the first rpc with new flavor might not be with root ctx, in
1705          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1706         if (unlikely(exp->exp_flvr_changed) &&
1707             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1708                 /* make the new flavor as "current", and old ones as
1709                  * about-to-expire */
1710                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1711                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1712                 flavor = exp->exp_flvr_old[1];
1713                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1714                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1715                 exp->exp_flvr_old[0] = exp->exp_flvr;
1716                 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1717                                           EXP_FLVR_UPDATE_EXPIRE;
1718                 exp->exp_flvr = flavor;
1719
1720                 /* flavor change finished */
1721                 exp->exp_flvr_changed = 0;
1722                 LASSERT(exp->exp_flvr_adapt == 1);
1723
1724                 /* if it's gss, we only interested in root ctx init */
1725                 if (req->rq_auth_gss &&
1726                     !(req->rq_ctx_init &&
1727                       (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1728                        req->rq_auth_usr_ost))) {
1729                         spin_unlock(&exp->exp_lock);
1730                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1731                                req->rq_auth_gss, req->rq_ctx_init,
1732                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1733                                req->rq_auth_usr_ost);
1734                         return 0;
1735                 }
1736
1737                 exp->exp_flvr_adapt = 0;
1738                 spin_unlock(&exp->exp_lock);
1739
1740                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1741                                                 req->rq_svc_ctx, &flavor);
1742         }
1743
1744         /* if it equals to the current flavor, we accept it, but need to
1745          * dealing with reverse sec/ctx */
1746         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1747                 /* most cases should return here, we only interested in
1748                  * gss root ctx init */
1749                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1750                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1751                      !req->rq_auth_usr_ost)) {
1752                         spin_unlock(&exp->exp_lock);
1753                         return 0;
1754                 }
1755
1756                 /* if flavor just changed, we should not proceed, just leave
1757                  * it and current flavor will be discovered and replaced
1758                  * shortly, and let _this_ rpc pass through */
1759                 if (exp->exp_flvr_changed) {
1760                         LASSERT(exp->exp_flvr_adapt);
1761                         spin_unlock(&exp->exp_lock);
1762                         return 0;
1763                 }
1764
1765                 if (exp->exp_flvr_adapt) {
1766                         exp->exp_flvr_adapt = 0;
1767                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1768                                exp, exp->exp_flvr.sf_rpc,
1769                                exp->exp_flvr_old[0].sf_rpc,
1770                                exp->exp_flvr_old[1].sf_rpc);
1771                         flavor = exp->exp_flvr;
1772                         spin_unlock(&exp->exp_lock);
1773
1774                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1775                                                         req->rq_svc_ctx,
1776                                                         &flavor);
1777                 } else {
1778                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
1779                                exp, exp->exp_flvr.sf_rpc,
1780                                exp->exp_flvr_old[0].sf_rpc,
1781                                exp->exp_flvr_old[1].sf_rpc);
1782                         spin_unlock(&exp->exp_lock);
1783
1784                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1785                                                            req->rq_svc_ctx);
1786                 }
1787         }
1788
1789         if (exp->exp_flvr_expire[0]) {
1790                 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
1791                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1792                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
1793                                        exp->exp_flvr.sf_rpc,
1794                                        exp->exp_flvr_old[0].sf_rpc,
1795                                        exp->exp_flvr_old[1].sf_rpc,
1796                                        (s64)(exp->exp_flvr_expire[0] -
1797                                        ktime_get_real_seconds()));
1798                                 spin_unlock(&exp->exp_lock);
1799                                 return 0;
1800                         }
1801                 } else {
1802                         CDEBUG(D_SEC, "mark middle expired\n");
1803                         exp->exp_flvr_expire[0] = 0;
1804                 }
1805                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1806                        exp->exp_flvr.sf_rpc,
1807                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1808                        req->rq_flvr.sf_rpc);
1809         }
1810
1811         /* now it doesn't match the current flavor, the only chance we can
1812          * accept it is match the old flavors which is not expired. */
1813         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1814                 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
1815                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1816                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
1817                                        exp,
1818                                        exp->exp_flvr.sf_rpc,
1819                                        exp->exp_flvr_old[0].sf_rpc,
1820                                        exp->exp_flvr_old[1].sf_rpc,
1821                                        (s64)(exp->exp_flvr_expire[1] -
1822                                        ktime_get_real_seconds()));
1823                                 spin_unlock(&exp->exp_lock);
1824                                 return 0;
1825                         }
1826                 } else {
1827                         CDEBUG(D_SEC, "mark oldest expired\n");
1828                         exp->exp_flvr_expire[1] = 0;
1829                 }
1830                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1831                        exp, exp->exp_flvr.sf_rpc,
1832                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1833                        req->rq_flvr.sf_rpc);
1834         } else {
1835                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1836                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1837                        exp->exp_flvr_old[1].sf_rpc);
1838         }
1839
1840         spin_unlock(&exp->exp_lock);
1841
1842         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
1843               exp, exp->exp_obd->obd_name,
1844               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1845               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1846               req->rq_flvr.sf_rpc,
1847               exp->exp_flvr.sf_rpc,
1848               exp->exp_flvr_old[0].sf_rpc,
1849               exp->exp_flvr_expire[0] ?
1850               (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
1851               exp->exp_flvr_old[1].sf_rpc,
1852               exp->exp_flvr_expire[1] ?
1853               (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
1854         return -EACCES;
1855 }
1856 EXPORT_SYMBOL(sptlrpc_target_export_check);
1857
1858 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1859 {
1860         /* peer's claim is unreliable unless gss is being used */
1861         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1862                 return svc_rc;
1863
1864         switch (req->rq_sp_from) {
1865         case LUSTRE_SP_CLI:
1866                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1867                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
1868                         svc_rc = SECSVC_DROP;
1869                 }
1870                 break;
1871         case LUSTRE_SP_MDT:
1872                 if (!req->rq_auth_usr_mdt) {
1873                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
1874                         svc_rc = SECSVC_DROP;
1875                 }
1876                 break;
1877         case LUSTRE_SP_OST:
1878                 if (!req->rq_auth_usr_ost) {
1879                         DEBUG_REQ(D_ERROR, req, "faked source OST");
1880                         svc_rc = SECSVC_DROP;
1881                 }
1882                 break;
1883         case LUSTRE_SP_MGS:
1884         case LUSTRE_SP_MGC:
1885                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1886                     !req->rq_auth_usr_ost) {
1887                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
1888                         svc_rc = SECSVC_DROP;
1889                 }
1890                 break;
1891         case LUSTRE_SP_ANY:
1892         default:
1893                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
1894                 svc_rc = SECSVC_DROP;
1895         }
1896
1897         return svc_rc;
1898 }
1899
1900 /**
1901  * Used by ptlrpc server, to perform transformation upon request message of
1902  * incoming \a req. This must be the first thing to do with a incoming
1903  * request in ptlrpc layer.
1904  *
1905  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
1906  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
1907  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
1908  * reply message has been prepared.
1909  * \retval SECSVC_DROP failed, this request should be dropped.
1910  */
1911 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
1912 {
1913         struct ptlrpc_sec_policy *policy;
1914         struct lustre_msg *msg = req->rq_reqbuf;
1915         int rc;
1916
1917         LASSERT(msg);
1918         LASSERT(req->rq_reqmsg == NULL);
1919         LASSERT(req->rq_repmsg == NULL);
1920         LASSERT(req->rq_svc_ctx == NULL);
1921
1922         req->rq_req_swab_mask = 0;
1923
1924         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
1925         switch (rc) {
1926         case 1:
1927                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1928         case 0:
1929                 break;
1930         default:
1931                 CERROR("error unpacking request from %s x%llu\n",
1932                        libcfs_id2str(req->rq_peer), req->rq_xid);
1933                 return SECSVC_DROP;
1934         }
1935
1936         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
1937         req->rq_sp_from = LUSTRE_SP_ANY;
1938         req->rq_auth_uid = -1;
1939         req->rq_auth_mapped_uid = -1;
1940
1941         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
1942         if (!policy) {
1943                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
1944                 return SECSVC_DROP;
1945         }
1946
1947         LASSERT(policy->sp_sops->accept);
1948         rc = policy->sp_sops->accept(req);
1949         sptlrpc_policy_put(policy);
1950         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
1951         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
1952
1953         /*
1954          * if it's not null flavor (which means embedded packing msg),
1955          * reset the swab mask for the coming inner msg unpacking.
1956          */
1957         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
1958                 req->rq_req_swab_mask = 0;
1959
1960         /* sanity check for the request source */
1961         rc = sptlrpc_svc_check_from(req, rc);
1962         return rc;
1963 }
1964
1965 /**
1966  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
1967  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
1968  * a buffer of \a msglen size.
1969  */
1970 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
1971 {
1972         struct ptlrpc_sec_policy *policy;
1973         struct ptlrpc_reply_state *rs;
1974         int rc;
1975
1976         LASSERT(req->rq_svc_ctx);
1977         LASSERT(req->rq_svc_ctx->sc_policy);
1978
1979         policy = req->rq_svc_ctx->sc_policy;
1980         LASSERT(policy->sp_sops->alloc_rs);
1981
1982         rc = policy->sp_sops->alloc_rs(req, msglen);
1983         if (unlikely(rc == -ENOMEM)) {
1984                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1985
1986                 if (svcpt->scp_service->srv_max_reply_size <
1987                    msglen + sizeof(struct ptlrpc_reply_state)) {
1988                         /* Just return failure if the size is too big */
1989                         CERROR("size of message is too big (%zd), %d allowed",
1990                                 msglen + sizeof(struct ptlrpc_reply_state),
1991                                 svcpt->scp_service->srv_max_reply_size);
1992                         return -ENOMEM;
1993                 }
1994
1995                 /* failed alloc, try emergency pool */
1996                 rs = lustre_get_emerg_rs(svcpt);
1997                 if (rs == NULL)
1998                         return -ENOMEM;
1999
2000                 req->rq_reply_state = rs;
2001                 rc = policy->sp_sops->alloc_rs(req, msglen);
2002                 if (rc) {
2003                         lustre_put_emerg_rs(rs);
2004                         req->rq_reply_state = NULL;
2005                 }
2006         }
2007
2008         LASSERT(rc != 0 ||
2009                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2010
2011         return rc;
2012 }
2013
2014 /**
2015  * Used by ptlrpc server, to perform transformation upon reply message.
2016  *
2017  * \post req->rq_reply_off is set to appropriate server-controlled reply offset.
2018  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2019  */
2020 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2021 {
2022         struct ptlrpc_sec_policy *policy;
2023         int rc;
2024
2025         LASSERT(req->rq_svc_ctx);
2026         LASSERT(req->rq_svc_ctx->sc_policy);
2027
2028         policy = req->rq_svc_ctx->sc_policy;
2029         LASSERT(policy->sp_sops->authorize);
2030
2031         rc = policy->sp_sops->authorize(req);
2032         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2033
2034         return rc;
2035 }
2036
2037 /**
2038  * Used by ptlrpc server, to free reply_state.
2039  */
2040 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2041 {
2042         struct ptlrpc_sec_policy *policy;
2043         unsigned int prealloc;
2044
2045         LASSERT(rs->rs_svc_ctx);
2046         LASSERT(rs->rs_svc_ctx->sc_policy);
2047
2048         policy = rs->rs_svc_ctx->sc_policy;
2049         LASSERT(policy->sp_sops->free_rs);
2050
2051         prealloc = rs->rs_prealloc;
2052         policy->sp_sops->free_rs(rs);
2053
2054         if (prealloc)
2055                 lustre_put_emerg_rs(rs);
2056 }
2057
2058 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2059 {
2060         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2061
2062         if (ctx != NULL)
2063                 atomic_inc(&ctx->sc_refcount);
2064 }
2065
2066 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2067 {
2068         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2069
2070         if (ctx == NULL)
2071                 return;
2072
2073         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2074         if (atomic_dec_and_test(&ctx->sc_refcount)) {
2075                 if (ctx->sc_policy->sp_sops->free_ctx)
2076                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2077         }
2078         req->rq_svc_ctx = NULL;
2079 }
2080
2081 /****************************************
2082  * bulk security                        *
2083  ****************************************/
2084
2085 /**
2086  * Perform transformation upon bulk data pointed by \a desc. This is called
2087  * before transforming the request message.
2088  */
2089 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2090                           struct ptlrpc_bulk_desc *desc)
2091 {
2092         struct ptlrpc_cli_ctx *ctx;
2093
2094         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2095
2096         if (!req->rq_pack_bulk)
2097                 return 0;
2098
2099         ctx = req->rq_cli_ctx;
2100         if (ctx->cc_ops->wrap_bulk)
2101                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2102         return 0;
2103 }
2104 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2105
2106 /**
2107  * This is called after unwrap the reply message.
2108  * return nob of actual plain text size received, or error code.
2109  */
2110 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2111                                  struct ptlrpc_bulk_desc *desc,
2112                                  int nob)
2113 {
2114         struct ptlrpc_cli_ctx *ctx;
2115         int rc;
2116
2117         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2118
2119         if (!req->rq_pack_bulk)
2120                 return desc->bd_nob_transferred;
2121
2122         ctx = req->rq_cli_ctx;
2123         if (ctx->cc_ops->unwrap_bulk) {
2124                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2125                 if (rc < 0)
2126                         return rc;
2127         }
2128         return desc->bd_nob_transferred;
2129 }
2130 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2131
2132 /**
2133  * This is called after unwrap the reply message.
2134  * return 0 for success or error code.
2135  */
2136 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2137                                   struct ptlrpc_bulk_desc *desc)
2138 {
2139         struct ptlrpc_cli_ctx *ctx;
2140         int rc;
2141
2142         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2143
2144         if (!req->rq_pack_bulk)
2145                 return 0;
2146
2147         ctx = req->rq_cli_ctx;
2148         if (ctx->cc_ops->unwrap_bulk) {
2149                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2150                 if (rc < 0)
2151                         return rc;
2152         }
2153
2154         /*
2155          * if everything is going right, nob should equals to nob_transferred.
2156          * in case of privacy mode, nob_transferred needs to be adjusted.
2157          */
2158         if (desc->bd_nob != desc->bd_nob_transferred) {
2159                 CERROR("nob %d doesn't match transferred nob %d",
2160                        desc->bd_nob, desc->bd_nob_transferred);
2161                 return -EPROTO;
2162         }
2163
2164         return 0;
2165 }
2166 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2167
2168 /****************************************
2169  * user descriptor helpers            *
2170  ****************************************/
2171
2172 int sptlrpc_current_user_desc_size(void)
2173 {
2174         int ngroups;
2175
2176         ngroups = current_ngroups;
2177
2178         if (ngroups > LUSTRE_MAX_GROUPS)
2179                 ngroups = LUSTRE_MAX_GROUPS;
2180         return sptlrpc_user_desc_size(ngroups);
2181 }
2182 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2183
2184 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2185 {
2186         struct ptlrpc_user_desc *pud;
2187
2188         pud = lustre_msg_buf(msg, offset, 0);
2189
2190         pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2191         pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2192         pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2193         pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2194         pud->pud_cap = cfs_curproc_cap_pack();
2195         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2196
2197         task_lock(current);
2198         if (pud->pud_ngroups > current_ngroups)
2199                 pud->pud_ngroups = current_ngroups;
2200         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2201                pud->pud_ngroups * sizeof(__u32));
2202         task_unlock(current);
2203
2204         return 0;
2205 }
2206 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2207
2208 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2209 {
2210         struct ptlrpc_user_desc *pud;
2211         int i;
2212
2213         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2214         if (!pud)
2215                 return -EINVAL;
2216
2217         if (swabbed) {
2218                 __swab32s(&pud->pud_uid);
2219                 __swab32s(&pud->pud_gid);
2220                 __swab32s(&pud->pud_fsuid);
2221                 __swab32s(&pud->pud_fsgid);
2222                 __swab32s(&pud->pud_cap);
2223                 __swab32s(&pud->pud_ngroups);
2224         }
2225
2226         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2227                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2228                 return -EINVAL;
2229         }
2230
2231         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2232             msg->lm_buflens[offset]) {
2233                 CERROR("%u groups are claimed but bufsize only %u\n",
2234                        pud->pud_ngroups, msg->lm_buflens[offset]);
2235                 return -EINVAL;
2236         }
2237
2238         if (swabbed) {
2239                 for (i = 0; i < pud->pud_ngroups; i++)
2240                         __swab32s(&pud->pud_groups[i]);
2241         }
2242
2243         return 0;
2244 }
2245 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2246
2247 /****************************************
2248  * misc helpers                  *
2249  ****************************************/
2250
2251 const char *sec2target_str(struct ptlrpc_sec *sec)
2252 {
2253         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2254                 return "*";
2255         if (sec_is_reverse(sec))
2256                 return "c";
2257         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2258 }
2259 EXPORT_SYMBOL(sec2target_str);
2260
2261 /*
2262  * return true if the bulk data is protected
2263  */
2264 bool sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2265 {
2266         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2267         case SPTLRPC_BULK_SVC_INTG:
2268         case SPTLRPC_BULK_SVC_PRIV:
2269                 return true;
2270         default:
2271                 return false;
2272         }
2273 }
2274 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2275
2276 /****************************************
2277  * crypto API helper/alloc blkciper     *
2278  ****************************************/
2279
2280 /****************************************
2281  * initialize/finalize            *
2282  ****************************************/
2283
2284 int sptlrpc_init(void)
2285 {
2286         int rc;
2287
2288         rwlock_init(&policy_lock);
2289
2290         rc = sptlrpc_gc_init();
2291         if (rc)
2292                 goto out;
2293
2294         rc = sptlrpc_conf_init();
2295         if (rc)
2296                 goto out_gc;
2297
2298         rc = sptlrpc_enc_pool_init();
2299         if (rc)
2300                 goto out_conf;
2301
2302         rc = sptlrpc_null_init();
2303         if (rc)
2304                 goto out_pool;
2305
2306         rc = sptlrpc_plain_init();
2307         if (rc)
2308                 goto out_null;
2309
2310         rc = sptlrpc_lproc_init();
2311         if (rc)
2312                 goto out_plain;
2313
2314         return 0;
2315
2316 out_plain:
2317         sptlrpc_plain_fini();
2318 out_null:
2319         sptlrpc_null_fini();
2320 out_pool:
2321         sptlrpc_enc_pool_fini();
2322 out_conf:
2323         sptlrpc_conf_fini();
2324 out_gc:
2325         sptlrpc_gc_fini();
2326 out:
2327         return rc;
2328 }
2329
2330 void sptlrpc_fini(void)
2331 {
2332         sptlrpc_lproc_fini();
2333         sptlrpc_plain_fini();
2334         sptlrpc_null_fini();
2335         sptlrpc_enc_pool_fini();
2336         sptlrpc_conf_fini();
2337         sptlrpc_gc_fini();
2338 }