lustre/ptlrpc/sec.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  42 #define EXPORT_SYMTAB
  43 #endif
  44 #define DEBUG_SUBSYSTEM S_SEC
  45
  46 #include <libcfs/libcfs.h>
  47 #ifndef __KERNEL__
  48 #include <liblustre.h>
  49 #include <libcfs/list.h>
  50 #else
  51 #include <linux/crypto.h>
  52 #include <linux/key.h>
  53 #endif
  54
  55 #include <obd.h>
  56 #include <obd_class.h>
  57 #include <obd_support.h>
  58 #include <lustre_net.h>
  59 #include <lustre_import.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_sec.h>
  62
  63 #include "ptlrpc_internal.h"
  64
  65 /***********************************************
  66  * policy registers                            *
  67  ***********************************************/
  68
  69 static cfs_rwlock_t policy_lock;
  70 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  71         NULL,
  72 };
  73
  74 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  75 {
  76         __u16 number = policy->sp_policy;
  77
  78         LASSERT(policy->sp_name);
  79         LASSERT(policy->sp_cops);
  80         LASSERT(policy->sp_sops);
  81
  82         if (number >= SPTLRPC_POLICY_MAX)
  83                 return -EINVAL;
  84
  85         cfs_write_lock(&policy_lock);
  86         if (unlikely(policies[number])) {
  87                 cfs_write_unlock(&policy_lock);
  88                 return -EALREADY;
  89         }
  90         policies[number] = policy;
  91         cfs_write_unlock(&policy_lock);
  92
  93         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  94         return 0;
  95 }
  96 EXPORT_SYMBOL(sptlrpc_register_policy);
  97
  98 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  99 {
 100         __u16 number = policy->sp_policy;
 101
 102         LASSERT(number < SPTLRPC_POLICY_MAX);
 103
 104         cfs_write_lock(&policy_lock);
 105         if (unlikely(policies[number] == NULL)) {
 106                 cfs_write_unlock(&policy_lock);
 107                 CERROR("%s: already unregistered\n", policy->sp_name);
 108                 return -EINVAL;
 109         }
 110
 111         LASSERT(policies[number] == policy);
 112         policies[number] = NULL;
 113         cfs_write_unlock(&policy_lock);
 114
 115         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 116         return 0;
 117 }
 118 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 119
 120 static
 121 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
 122 {
 123         static CFS_DECLARE_MUTEX(load_mutex);
 124         static cfs_atomic_t       loaded = CFS_ATOMIC_INIT(0);
 125         struct ptlrpc_sec_policy *policy;
 126         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 127         __u16                     flag = 0;
 128
 129         if (number >= SPTLRPC_POLICY_MAX)
 130                 return NULL;
 131
 132         while (1) {
 133                 cfs_read_lock(&policy_lock);
 134                 policy = policies[number];
 135                 if (policy && !cfs_try_module_get(policy->sp_owner))
 136                         policy = NULL;
 137                 if (policy == NULL)
 138                         flag = cfs_atomic_read(&loaded);
 139                 cfs_read_unlock(&policy_lock);
 140
 141                 if (policy != NULL || flag != 0 ||
 142                     number != SPTLRPC_POLICY_GSS)
 143                         break;
 144
 145                 /* try to load gss module, once */
 146                 cfs_mutex_down(&load_mutex);
 147                 if (cfs_atomic_read(&loaded) == 0) {
 148                         if (cfs_request_module("ptlrpc_gss") == 0)
 149                                 CWARN("module ptlrpc_gss loaded on demand\n");
 150                         else
 151                                 CERROR("Unable to load module ptlrpc_gss\n");
 152
 153                         cfs_atomic_set(&loaded, 1);
 154                 }
 155                 cfs_mutex_up(&load_mutex);
 156         }
 157
 158         return policy;
 159 }
 160
 161 __u32 sptlrpc_name2flavor_base(const char *name)
 162 {
 163         if (!strcmp(name, "null"))
 164                 return SPTLRPC_FLVR_NULL;
 165         if (!strcmp(name, "plain"))
 166                 return SPTLRPC_FLVR_PLAIN;
 167         if (!strcmp(name, "krb5n"))
 168                 return SPTLRPC_FLVR_KRB5N;
 169         if (!strcmp(name, "krb5a"))
 170                 return SPTLRPC_FLVR_KRB5A;
 171         if (!strcmp(name, "krb5i"))
 172                 return SPTLRPC_FLVR_KRB5I;
 173         if (!strcmp(name, "krb5p"))
 174                 return SPTLRPC_FLVR_KRB5P;
 175
 176         return SPTLRPC_FLVR_INVALID;
 177 }
 178 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 179
 180 const char *sptlrpc_flavor2name_base(__u32 flvr)
 181 {
 182         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 183
 184         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 185                 return "null";
 186         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 187                 return "plain";
 188         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 189                 return "krb5n";
 190         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 191                 return "krb5a";
 192         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 193                 return "krb5i";
 194         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 195                 return "krb5p";
 196
 197         CERROR("invalid wire flavor 0x%x\n", flvr);
 198         return "invalid";
 199 }
 200 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 201
 202 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 203                                char *buf, int bufsize)
 204 {
 205         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 206                 snprintf(buf, bufsize, "hash:%s",
 207                          sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 208         else
 209                 snprintf(buf, bufsize, "%s",
 210                          sptlrpc_flavor2name_base(sf->sf_rpc));
 211
 212         buf[bufsize - 1] = '\0';
 213         return buf;
 214 }
 215 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 216
 217 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 218 {
 219         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 220
 221         /*
 222          * currently we don't support customized bulk specification for
 223          * flavors other than plain
 224          */
 225         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 226                 char bspec[16];
 227
 228                 bspec[0] = '-';
 229                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 230                 strncat(buf, bspec, bufsize);
 231         }
 232
 233         buf[bufsize - 1] = '\0';
 234         return buf;
 235 }
 236 EXPORT_SYMBOL(sptlrpc_flavor2name);
 237
 238 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 239 {
 240         buf[0] = '\0';
 241
 242         if (flags & PTLRPC_SEC_FL_REVERSE)
 243                 strncat(buf, "reverse,", bufsize);
 244         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 245                 strncat(buf, "rootonly,", bufsize);
 246         if (flags & PTLRPC_SEC_FL_UDESC)
 247                 strncat(buf, "udesc,", bufsize);
 248         if (flags & PTLRPC_SEC_FL_BULK)
 249                 strncat(buf, "bulk,", bufsize);
 250         if (buf[0] == '\0')
 251                 strncat(buf, "-,", bufsize);
 252
 253         buf[bufsize - 1] = '\0';
 254         return buf;
 255 }
 256 EXPORT_SYMBOL(sptlrpc_secflags2str);
 257
 258 /**************************************************
 259  * client context APIs                            *
 260  **************************************************/
 261
 262 static
 263 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 264 {
 265         struct vfs_cred vcred;
 266         int create = 1, remove_dead = 1;
 267
 268         LASSERT(sec);
 269         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 270
 271         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 272                                      PTLRPC_SEC_FL_ROOTONLY)) {
 273                 vcred.vc_uid = 0;
 274                 vcred.vc_gid = 0;
 275                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 276                         create = 0;
 277                         remove_dead = 0;
 278                 }
 279         } else {
 280                 vcred.vc_uid = cfs_curproc_uid();
 281                 vcred.vc_gid = cfs_curproc_gid();
 282         }
 283
 284         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
 285                                                    create, remove_dead);
 286 }
 287
 288 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 289 {
 290         cfs_atomic_inc(&ctx->cc_refcount);
 291         return ctx;
 292 }
 293 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 294
 295 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 296 {
 297         struct ptlrpc_sec *sec = ctx->cc_sec;
 298
 299         LASSERT(sec);
 300         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 301
 302         if (!cfs_atomic_dec_and_test(&ctx->cc_refcount))
 303                 return;
 304
 305         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 306 }
 307 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 308
 309 /**
 310  * Expire the client context immediately.
 311  *
 312  * \pre Caller must hold at least 1 reference on the \a ctx.
 313  */
 314 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 315 {
 316         LASSERT(ctx->cc_ops->die);
 317         ctx->cc_ops->die(ctx, 0);
 318 }
 319 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 320
 321 /**
 322  * To wake up the threads who are waiting for this client context. Called
 323  * after some status change happened on \a ctx.
 324  */
 325 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 326 {
 327         struct ptlrpc_request *req, *next;
 328
 329         cfs_spin_lock(&ctx->cc_lock);
 330         cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 331                                      rq_ctx_chain) {
 332                 cfs_list_del_init(&req->rq_ctx_chain);
 333                 ptlrpc_client_wake_req(req);
 334         }
 335         cfs_spin_unlock(&ctx->cc_lock);
 336 }
 337 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 338
 339 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 340 {
 341         LASSERT(ctx->cc_ops);
 342
 343         if (ctx->cc_ops->display == NULL)
 344                 return 0;
 345
 346         return ctx->cc_ops->display(ctx, buf, bufsize);
 347 }
 348
 349 static int import_sec_check_expire(struct obd_import *imp)
 350 {
 351         int     adapt = 0;
 352
 353         cfs_spin_lock(&imp->imp_lock);
 354         if (imp->imp_sec_expire &&
 355             imp->imp_sec_expire < cfs_time_current_sec()) {
 356                 adapt = 1;
 357                 imp->imp_sec_expire = 0;
 358         }
 359         cfs_spin_unlock(&imp->imp_lock);
 360
 361         if (!adapt)
 362                 return 0;
 363
 364         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 365         return sptlrpc_import_sec_adapt(imp, NULL, 0);
 366 }
 367
 368 static int import_sec_validate_get(struct obd_import *imp,
 369                                    struct ptlrpc_sec **sec)
 370 {
 371         int     rc;
 372
 373         if (unlikely(imp->imp_sec_expire)) {
 374                 rc = import_sec_check_expire(imp);
 375                 if (rc)
 376                         return rc;
 377         }
 378
 379         *sec = sptlrpc_import_sec_ref(imp);
 380         if (*sec == NULL) {
 381                 CERROR("import %p (%s) with no sec\n",
 382                        imp, ptlrpc_import_state_name(imp->imp_state));
 383                 return -EACCES;
 384         }
 385
 386         if (unlikely((*sec)->ps_dying)) {
 387                 CERROR("attempt to use dying sec %p\n", sec);
 388                 sptlrpc_sec_put(*sec);
 389                 return -EACCES;
 390         }
 391
 392         return 0;
 393 }
 394
 395 /**
 396  * Given a \a req, find or allocate a appropriate context for it.
 397  * \pre req->rq_cli_ctx == NULL.
 398  *
 399  * \retval 0 succeed, and req->rq_cli_ctx is set.
 400  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 401  */
 402 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 403 {
 404         struct obd_import *imp = req->rq_import;
 405         struct ptlrpc_sec *sec;
 406         int                rc;
 407         ENTRY;
 408
 409         LASSERT(!req->rq_cli_ctx);
 410         LASSERT(imp);
 411
 412         rc = import_sec_validate_get(imp, &sec);
 413         if (rc)
 414                 RETURN(rc);
 415
 416         req->rq_cli_ctx = get_my_ctx(sec);
 417
 418         sptlrpc_sec_put(sec);
 419
 420         if (!req->rq_cli_ctx) {
 421                 CERROR("req %p: fail to get context\n", req);
 422                 RETURN(-ENOMEM);
 423         }
 424
 425         RETURN(0);
 426 }
 427
 428 /**
 429  * Drop the context for \a req.
 430  * \pre req->rq_cli_ctx != NULL.
 431  * \post req->rq_cli_ctx == NULL.
 432  *
 433  * If \a sync == 0, this function should return quickly without sleep;
 434  * otherwise it might trigger and wait for the whole process of sending
 435  * an context-destroying rpc to server.
 436  */
 437 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 438 {
 439         ENTRY;
 440
 441         LASSERT(req);
 442         LASSERT(req->rq_cli_ctx);
 443
 444         /* request might be asked to release earlier while still
 445          * in the context waiting list.
 446          */
 447         if (!cfs_list_empty(&req->rq_ctx_chain)) {
 448                 cfs_spin_lock(&req->rq_cli_ctx->cc_lock);
 449                 cfs_list_del_init(&req->rq_ctx_chain);
 450                 cfs_spin_unlock(&req->rq_cli_ctx->cc_lock);
 451         }
 452
 453         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 454         req->rq_cli_ctx = NULL;
 455         EXIT;
 456 }
 457
 458 static
 459 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 460                            struct ptlrpc_cli_ctx *oldctx,
 461                            struct ptlrpc_cli_ctx *newctx)
 462 {
 463         struct sptlrpc_flavor   old_flvr;
 464         char                   *reqmsg = NULL; /* to workaround old gcc */
 465         int                     reqmsg_size;
 466         int                     rc = 0;
 467
 468         LASSERT(req->rq_reqmsg);
 469         LASSERT(req->rq_reqlen);
 470         LASSERT(req->rq_replen);
 471
 472         CWARN("req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
 473               "switch sec %p(%s) -> %p(%s)\n", req,
 474               oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
 475               newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
 476               oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
 477               newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 478
 479         /* save flavor */
 480         old_flvr = req->rq_flvr;
 481
 482         /* save request message */
 483         reqmsg_size = req->rq_reqlen;
 484         if (reqmsg_size != 0) {
 485                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 486                 if (reqmsg == NULL)
 487                         return -ENOMEM;
 488                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 489         }
 490
 491         /* release old req/rep buf */
 492         req->rq_cli_ctx = oldctx;
 493         sptlrpc_cli_free_reqbuf(req);
 494         sptlrpc_cli_free_repbuf(req);
 495         req->rq_cli_ctx = newctx;
 496
 497         /* recalculate the flavor */
 498         sptlrpc_req_set_flavor(req, 0);
 499
 500         /* alloc new request buffer
 501          * we don't need to alloc reply buffer here, leave it to the
 502          * rest procedure of ptlrpc */
 503         if (reqmsg_size != 0) {
 504                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 505                 if (!rc) {
 506                         LASSERT(req->rq_reqmsg);
 507                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 508                 } else {
 509                         CWARN("failed to alloc reqbuf: %d\n", rc);
 510                         req->rq_flvr = old_flvr;
 511                 }
 512
 513                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 514         }
 515         return rc;
 516 }
 517
 518 /**
 519  * If current context of \a req is dead somehow, e.g. we just switched flavor
 520  * thus marked original contexts dead, we'll find a new context for it. if
 521  * no switch is needed, \a req will end up with the same context.
 522  *
 523  * \note a request must have a context, to keep other parts of code happy.
 524  * In any case of failure during the switching, we must restore the old one.
 525  */
 526 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 527 {
 528         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 529         struct ptlrpc_cli_ctx *newctx;
 530         int                    rc;
 531         ENTRY;
 532
 533         LASSERT(oldctx);
 534
 535         sptlrpc_cli_ctx_get(oldctx);
 536         sptlrpc_req_put_ctx(req, 0);
 537
 538         rc = sptlrpc_req_get_ctx(req);
 539         if (unlikely(rc)) {
 540                 LASSERT(!req->rq_cli_ctx);
 541
 542                 /* restore old ctx */
 543                 req->rq_cli_ctx = oldctx;
 544                 RETURN(rc);
 545         }
 546
 547         newctx = req->rq_cli_ctx;
 548         LASSERT(newctx);
 549
 550         if (unlikely(newctx == oldctx &&
 551                      cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 552                 /*
 553                  * still get the old dead ctx, usually means system too busy
 554                  */
 555                 CWARN("ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 556                       newctx, newctx->cc_flags);
 557
 558                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
 559                                                    CFS_HZ);
 560         } else {
 561                 /*
 562                  * it's possible newctx == oldctx if we're switching
 563                  * subflavor with the same sec.
 564                  */
 565                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 566                 if (rc) {
 567                         /* restore old ctx */
 568                         sptlrpc_req_put_ctx(req, 0);
 569                         req->rq_cli_ctx = oldctx;
 570                         RETURN(rc);
 571                 }
 572
 573                 LASSERT(req->rq_cli_ctx == newctx);
 574         }
 575
 576         sptlrpc_cli_ctx_put(oldctx, 1);
 577         RETURN(0);
 578 }
 579 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 580
 581 static
 582 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 583 {
 584         if (cli_ctx_is_refreshed(ctx))
 585                 return 1;
 586         return 0;
 587 }
 588
 589 static
 590 int ctx_refresh_timeout(void *data)
 591 {
 592         struct ptlrpc_request *req = data;
 593         int rc;
 594
 595         /* conn_cnt is needed in expire_one_request */
 596         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 597
 598         rc = ptlrpc_expire_one_request(req, 1);
 599         /* if we started recovery, we should mark this ctx dead; otherwise
 600          * in case of lgssd died nobody would retire this ctx, following
 601          * connecting will still find the same ctx thus cause deadlock.
 602          * there's an assumption that expire time of the request should be
 603          * later than the context refresh expire time.
 604          */
 605         if (rc == 0)
 606                 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
 607         return rc;
 608 }
 609
 610 static
 611 void ctx_refresh_interrupt(void *data)
 612 {
 613         struct ptlrpc_request *req = data;
 614
 615         cfs_spin_lock(&req->rq_lock);
 616         req->rq_intr = 1;
 617         cfs_spin_unlock(&req->rq_lock);
 618 }
 619
 620 static
 621 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 622 {
 623         cfs_spin_lock(&ctx->cc_lock);
 624         if (!cfs_list_empty(&req->rq_ctx_chain))
 625                 cfs_list_del_init(&req->rq_ctx_chain);
 626         cfs_spin_unlock(&ctx->cc_lock);
 627 }
 628
 629 /**
 630  * To refresh the context of \req, if it's not up-to-date.
 631  * \param timeout
 632  * - < 0: don't wait
 633  * - = 0: wait until success or fatal error occur
 634  * - > 0: timeout value (in seconds)
 635  *
 636  * The status of the context could be subject to be changed by other threads
 637  * at any time. We allow this race, but once we return with 0, the caller will
 638  * suppose it's uptodated and keep using it until the owning rpc is done.
 639  *
 640  * \retval 0 only if the context is uptodated.
 641  * \retval -ev error number.
 642  */
 643 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 644 {
 645         struct ptlrpc_cli_ctx  *ctx = req->rq_cli_ctx;
 646         struct ptlrpc_sec      *sec;
 647         struct l_wait_info      lwi;
 648         int                     rc;
 649         ENTRY;
 650
 651         LASSERT(ctx);
 652
 653         if (req->rq_ctx_init || req->rq_ctx_fini)
 654                 RETURN(0);
 655
 656         /*
 657          * during the process a request's context might change type even
 658          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 659          * everything
 660          */
 661 again:
 662         rc = import_sec_validate_get(req->rq_import, &sec);
 663         if (rc)
 664                 RETURN(rc);
 665
 666         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 667                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 668                       req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 669                 req_off_ctx_list(req, ctx);
 670                 sptlrpc_req_replace_dead_ctx(req);
 671                 ctx = req->rq_cli_ctx;
 672         }
 673         sptlrpc_sec_put(sec);
 674
 675         if (cli_ctx_is_eternal(ctx))
 676                 RETURN(0);
 677
 678         if (unlikely(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 679                 LASSERT(ctx->cc_ops->refresh);
 680                 ctx->cc_ops->refresh(ctx);
 681         }
 682         LASSERT(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 683
 684         LASSERT(ctx->cc_ops->validate);
 685         if (ctx->cc_ops->validate(ctx) == 0) {
 686                 req_off_ctx_list(req, ctx);
 687                 RETURN(0);
 688         }
 689
 690         if (unlikely(cfs_test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 691                 cfs_spin_lock(&req->rq_lock);
 692                 req->rq_err = 1;
 693                 cfs_spin_unlock(&req->rq_lock);
 694                 req_off_ctx_list(req, ctx);
 695                 RETURN(-EPERM);
 696         }
 697
 698         /*
 699          * There's a subtle issue for resending RPCs, suppose following
 700          * situation:
 701          *  1. the request was sent to server.
 702          *  2. recovery was kicked start, after finished the request was
 703          *     marked as resent.
 704          *  3. resend the request.
 705          *  4. old reply from server received, we accept and verify the reply.
 706          *     this has to be success, otherwise the error will be aware
 707          *     by application.
 708          *  5. new reply from server received, dropped by LNet.
 709          *
 710          * Note the xid of old & new request is the same. We can't simply
 711          * change xid for the resent request because the server replies on
 712          * it for reply reconstruction.
 713          *
 714          * Commonly the original context should be uptodate because we
 715          * have a expiry nice time; server will keep its context because
 716          * we at least hold a ref of old context which prevent context
 717          * destroying RPC being sent. So server still can accept the request
 718          * and finish the RPC. But if that's not the case:
 719          *  1. If server side context has been trimmed, a NO_CONTEXT will
 720          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 721          *     context by force.
 722          *  2. Current context never be refreshed, then we are fine: we
 723          *     never really send request with old context before.
 724          */
 725         if (cfs_test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 726             unlikely(req->rq_reqmsg) &&
 727             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 728                 req_off_ctx_list(req, ctx);
 729                 RETURN(0);
 730         }
 731
 732         if (unlikely(cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 733                 req_off_ctx_list(req, ctx);
 734                 /*
 735                  * don't switch ctx if import was deactivated
 736                  */
 737                 if (req->rq_import->imp_deactive) {
 738                         cfs_spin_lock(&req->rq_lock);
 739                         req->rq_err = 1;
 740                         cfs_spin_unlock(&req->rq_lock);
 741                         RETURN(-EINTR);
 742                 }
 743
 744                 rc = sptlrpc_req_replace_dead_ctx(req);
 745                 if (rc) {
 746                         LASSERT(ctx == req->rq_cli_ctx);
 747                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 748                                 req, ctx, rc);
 749                         cfs_spin_lock(&req->rq_lock);
 750                         req->rq_err = 1;
 751                         cfs_spin_unlock(&req->rq_lock);
 752                         RETURN(rc);
 753                 }
 754
 755                 ctx = req->rq_cli_ctx;
 756                 goto again;
 757         }
 758
 759         /*
 760          * Now we're sure this context is during upcall, add myself into
 761          * waiting list
 762          */
 763         cfs_spin_lock(&ctx->cc_lock);
 764         if (cfs_list_empty(&req->rq_ctx_chain))
 765                 cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 766         cfs_spin_unlock(&ctx->cc_lock);
 767
 768         if (timeout < 0)
 769                 RETURN(-EWOULDBLOCK);
 770
 771         /* Clear any flags that may be present from previous sends */
 772         LASSERT(req->rq_receiving_reply == 0);
 773         cfs_spin_lock(&req->rq_lock);
 774         req->rq_err = 0;
 775         req->rq_timedout = 0;
 776         req->rq_resend = 0;
 777         req->rq_restart = 0;
 778         cfs_spin_unlock(&req->rq_lock);
 779
 780         lwi = LWI_TIMEOUT_INTR(timeout * CFS_HZ, ctx_refresh_timeout,
 781                                ctx_refresh_interrupt, req);
 782         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 783
 784         /*
 785          * following cases could lead us here:
 786          * - successfully refreshed;
 787          * - interrupted;
 788          * - timedout, and we don't want recover from the failure;
 789          * - timedout, and waked up upon recovery finished;
 790          * - someone else mark this ctx dead by force;
 791          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 792          *   e.g. ptlrpc_abort_inflight();
 793          */
 794         if (!cli_ctx_is_refreshed(ctx)) {
 795                 /* timed out or interruptted */
 796                 req_off_ctx_list(req, ctx);
 797
 798                 LASSERT(rc != 0);
 799                 RETURN(rc);
 800         }
 801
 802         goto again;
 803 }
 804
 805 /**
 806  * Initialize flavor settings for \a req, according to \a opcode.
 807  *
 808  * \note this could be called in two situations:
 809  * - new request from ptlrpc_pre_req(), with proper @opcode
 810  * - old request which changed ctx in the middle, with @opcode == 0
 811  */
 812 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 813 {
 814         struct ptlrpc_sec *sec;
 815
 816         LASSERT(req->rq_import);
 817         LASSERT(req->rq_cli_ctx);
 818         LASSERT(req->rq_cli_ctx->cc_sec);
 819         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 820
 821         /* special security flags accoding to opcode */
 822         switch (opcode) {
 823         case OST_READ:
 824         case MDS_READPAGE:
 825         case MGS_CONFIG_READ:
 826                 req->rq_bulk_read = 1;
 827                 break;
 828         case OST_WRITE:
 829         case MDS_WRITEPAGE:
 830                 req->rq_bulk_write = 1;
 831                 break;
 832         case SEC_CTX_INIT:
 833                 req->rq_ctx_init = 1;
 834                 break;
 835         case SEC_CTX_FINI:
 836                 req->rq_ctx_fini = 1;
 837                 break;
 838         case 0:
 839                 /* init/fini rpc won't be resend, so can't be here */
 840                 LASSERT(req->rq_ctx_init == 0);
 841                 LASSERT(req->rq_ctx_fini == 0);
 842
 843                 /* cleanup flags, which should be recalculated */
 844                 req->rq_pack_udesc = 0;
 845                 req->rq_pack_bulk = 0;
 846                 break;
 847         }
 848
 849         sec = req->rq_cli_ctx->cc_sec;
 850
 851         cfs_spin_lock(&sec->ps_lock);
 852         req->rq_flvr = sec->ps_flvr;
 853         cfs_spin_unlock(&sec->ps_lock);
 854
 855         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
 856          * destruction rpc */
 857         if (unlikely(req->rq_ctx_init))
 858                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 859         else if (unlikely(req->rq_ctx_fini))
 860                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 861
 862         /* user descriptor flag, null security can't do it anyway */
 863         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 864             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 865                 req->rq_pack_udesc = 1;
 866
 867         /* bulk security flag */
 868         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 869             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 870                 req->rq_pack_bulk = 1;
 871 }
 872
 873 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 874 {
 875         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 876                 return;
 877
 878         LASSERT(req->rq_clrbuf);
 879         if (req->rq_pool || !req->rq_reqbuf)
 880                 return;
 881
 882         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 883         req->rq_reqbuf = NULL;
 884         req->rq_reqbuf_len = 0;
 885 }
 886
 887 /**
 888  * Given an import \a imp, check whether current user has a valid context
 889  * or not. We may create a new context and try to refresh it, and try
 890  * repeatedly try in case of non-fatal errors. Return 0 means success.
 891  */
 892 int sptlrpc_import_check_ctx(struct obd_import *imp)
 893 {
 894         struct ptlrpc_sec     *sec;
 895         struct ptlrpc_cli_ctx *ctx;
 896         struct ptlrpc_request *req = NULL;
 897         int rc;
 898         ENTRY;
 899
 900         cfs_might_sleep();
 901
 902         sec = sptlrpc_import_sec_ref(imp);
 903         ctx = get_my_ctx(sec);
 904         sptlrpc_sec_put(sec);
 905
 906         if (!ctx)
 907                 RETURN(-ENOMEM);
 908
 909         if (cli_ctx_is_eternal(ctx) ||
 910             ctx->cc_ops->validate(ctx) == 0) {
 911                 sptlrpc_cli_ctx_put(ctx, 1);
 912                 RETURN(0);
 913         }
 914
 915         if (cli_ctx_is_error(ctx)) {
 916                 sptlrpc_cli_ctx_put(ctx, 1);
 917                 RETURN(-EACCES);
 918         }
 919
 920         OBD_ALLOC_PTR(req);
 921         if (!req)
 922                 RETURN(-ENOMEM);
 923
 924         cfs_spin_lock_init(&req->rq_lock);
 925         cfs_atomic_set(&req->rq_refcount, 10000);
 926         CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
 927         cfs_waitq_init(&req->rq_reply_waitq);
 928         cfs_waitq_init(&req->rq_set_waitq);
 929         req->rq_import = imp;
 930         req->rq_flvr = sec->ps_flvr;
 931         req->rq_cli_ctx = ctx;
 932
 933         rc = sptlrpc_req_refresh_ctx(req, 0);
 934         LASSERT(cfs_list_empty(&req->rq_ctx_chain));
 935         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 936         OBD_FREE_PTR(req);
 937
 938         RETURN(rc);
 939 }
 940
 941 /**
 942  * Used by ptlrpc client, to perform the pre-defined security transformation
 943  * upon the request message of \a req. After this function called,
 944  * req->rq_reqmsg is still accessible as clear text.
 945  */
 946 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 947 {
 948         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 949         int rc = 0;
 950         ENTRY;
 951
 952         LASSERT(ctx);
 953         LASSERT(ctx->cc_sec);
 954         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
 955
 956         /* we wrap bulk request here because now we can be sure
 957          * the context is uptodate.
 958          */
 959         if (req->rq_bulk) {
 960                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
 961                 if (rc)
 962                         RETURN(rc);
 963         }
 964
 965         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 966         case SPTLRPC_SVC_NULL:
 967         case SPTLRPC_SVC_AUTH:
 968         case SPTLRPC_SVC_INTG:
 969                 LASSERT(ctx->cc_ops->sign);
 970                 rc = ctx->cc_ops->sign(ctx, req);
 971                 break;
 972         case SPTLRPC_SVC_PRIV:
 973                 LASSERT(ctx->cc_ops->seal);
 974                 rc = ctx->cc_ops->seal(ctx, req);
 975                 break;
 976         default:
 977                 LBUG();
 978         }
 979
 980         if (rc == 0) {
 981                 LASSERT(req->rq_reqdata_len);
 982                 LASSERT(req->rq_reqdata_len % 8 == 0);
 983                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
 984         }
 985
 986         RETURN(rc);
 987 }
 988
 989 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
 990 {
 991         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 992         int                    rc;
 993         ENTRY;
 994
 995         LASSERT(ctx);
 996         LASSERT(ctx->cc_sec);
 997         LASSERT(req->rq_repbuf);
 998         LASSERT(req->rq_repdata);
 999         LASSERT(req->rq_repmsg == NULL);
1000
1001         req->rq_rep_swab_mask = 0;
1002
1003         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1004         switch (rc) {
1005         case 1:
1006                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1007         case 0:
1008                 break;
1009         default:
1010                 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1011                 RETURN(-EPROTO);
1012         }
1013
1014         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1015                 CERROR("replied data length %d too small\n",
1016                        req->rq_repdata_len);
1017                 RETURN(-EPROTO);
1018         }
1019
1020         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1021             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1022                 CERROR("reply policy %u doesn't match request policy %u\n",
1023                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1024                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1025                 RETURN(-EPROTO);
1026         }
1027
1028         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1029         case SPTLRPC_SVC_NULL:
1030         case SPTLRPC_SVC_AUTH:
1031         case SPTLRPC_SVC_INTG:
1032                 LASSERT(ctx->cc_ops->verify);
1033                 rc = ctx->cc_ops->verify(ctx, req);
1034                 break;
1035         case SPTLRPC_SVC_PRIV:
1036                 LASSERT(ctx->cc_ops->unseal);
1037                 rc = ctx->cc_ops->unseal(ctx, req);
1038                 break;
1039         default:
1040                 LBUG();
1041         }
1042         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1043
1044         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1045             !req->rq_ctx_init)
1046                 req->rq_rep_swab_mask = 0;
1047         RETURN(rc);
1048 }
1049
1050 /**
1051  * Used by ptlrpc client, to perform security transformation upon the reply
1052  * message of \a req. After return successfully, req->rq_repmsg points to
1053  * the reply message in clear text.
1054  *
1055  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1056  * going to change.
1057  */
1058 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1059 {
1060         LASSERT(req->rq_repbuf);
1061         LASSERT(req->rq_repdata == NULL);
1062         LASSERT(req->rq_repmsg == NULL);
1063         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1064
1065         if (req->rq_reply_off == 0 &&
1066             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1067                 CERROR("real reply with offset 0\n");
1068                 return -EPROTO;
1069         }
1070
1071         if (req->rq_reply_off % 8 != 0) {
1072                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1073                 return -EPROTO;
1074         }
1075
1076         req->rq_repdata = (struct lustre_msg *)
1077                                 (req->rq_repbuf + req->rq_reply_off);
1078         req->rq_repdata_len = req->rq_nob_received;
1079
1080         return do_cli_unwrap_reply(req);
1081 }
1082
1083 /**
1084  * Used by ptlrpc client, to perform security transformation upon the early
1085  * reply message of \a req. We expect the rq_reply_off is 0, and
1086  * rq_nob_received is the early reply size.
1087  *
1088  * Because the receive buffer might be still posted, the reply data might be
1089  * changed at any time, no matter we're holding rq_lock or not. For this reason
1090  * we allocate a separate ptlrpc_request and reply buffer for early reply
1091  * processing.
1092  *
1093  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1094  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1095  * \a *req_ret to release it.
1096  * \retval -ev error number, and \a req_ret will not be set.
1097  */
1098 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1099                                    struct ptlrpc_request **req_ret)
1100 {
1101         struct ptlrpc_request  *early_req;
1102         char                   *early_buf;
1103         int                     early_bufsz, early_size;
1104         int                     rc;
1105         ENTRY;
1106
1107         OBD_ALLOC_PTR(early_req);
1108         if (early_req == NULL)
1109                 RETURN(-ENOMEM);
1110
1111         early_size = req->rq_nob_received;
1112         early_bufsz = size_roundup_power2(early_size);
1113         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1114         if (early_buf == NULL)
1115                 GOTO(err_req, rc = -ENOMEM);
1116
1117         /* sanity checkings and copy data out, do it inside spinlock */
1118         cfs_spin_lock(&req->rq_lock);
1119
1120         if (req->rq_replied) {
1121                 cfs_spin_unlock(&req->rq_lock);
1122                 GOTO(err_buf, rc = -EALREADY);
1123         }
1124
1125         LASSERT(req->rq_repbuf);
1126         LASSERT(req->rq_repdata == NULL);
1127         LASSERT(req->rq_repmsg == NULL);
1128
1129         if (req->rq_reply_off != 0) {
1130                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1131                 cfs_spin_unlock(&req->rq_lock);
1132                 GOTO(err_buf, rc = -EPROTO);
1133         }
1134
1135         if (req->rq_nob_received != early_size) {
1136                 /* even another early arrived the size should be the same */
1137                 CERROR("data size has changed from %u to %u\n",
1138                        early_size, req->rq_nob_received);
1139                 cfs_spin_unlock(&req->rq_lock);
1140                 GOTO(err_buf, rc = -EINVAL);
1141         }
1142
1143         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1144                 CERROR("early reply length %d too small\n",
1145                        req->rq_nob_received);
1146                 cfs_spin_unlock(&req->rq_lock);
1147                 GOTO(err_buf, rc = -EALREADY);
1148         }
1149
1150         memcpy(early_buf, req->rq_repbuf, early_size);
1151         cfs_spin_unlock(&req->rq_lock);
1152
1153         cfs_spin_lock_init(&early_req->rq_lock);
1154         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1155         early_req->rq_flvr = req->rq_flvr;
1156         early_req->rq_repbuf = early_buf;
1157         early_req->rq_repbuf_len = early_bufsz;
1158         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1159         early_req->rq_repdata_len = early_size;
1160         early_req->rq_early = 1;
1161         early_req->rq_reqmsg = req->rq_reqmsg;
1162
1163         rc = do_cli_unwrap_reply(early_req);
1164         if (rc) {
1165                 DEBUG_REQ(D_ADAPTTO, early_req,
1166                           "error %d unwrap early reply", rc);
1167                 GOTO(err_ctx, rc);
1168         }
1169
1170         LASSERT(early_req->rq_repmsg);
1171         *req_ret = early_req;
1172         RETURN(0);
1173
1174 err_ctx:
1175         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1176 err_buf:
1177         OBD_FREE_LARGE(early_buf, early_bufsz);
1178 err_req:
1179         OBD_FREE_PTR(early_req);
1180         RETURN(rc);
1181 }
1182
1183 /**
1184  * Used by ptlrpc client, to release a processed early reply \a early_req.
1185  *
1186  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1187  */
1188 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1189 {
1190         LASSERT(early_req->rq_repbuf);
1191         LASSERT(early_req->rq_repdata);
1192         LASSERT(early_req->rq_repmsg);
1193
1194         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1195         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1196         OBD_FREE_PTR(early_req);
1197 }
1198
1199 /**************************************************
1200  * sec ID                                         *
1201  **************************************************/
1202
1203 /*
1204  * "fixed" sec (e.g. null) use sec_id < 0
1205  */
1206 static cfs_atomic_t sptlrpc_sec_id = CFS_ATOMIC_INIT(1);
1207
1208 int sptlrpc_get_next_secid(void)
1209 {
1210         return cfs_atomic_inc_return(&sptlrpc_sec_id);
1211 }
1212 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1213
1214 /**************************************************
1215  * client side high-level security APIs           *
1216  **************************************************/
1217
1218 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1219                                    int grace, int force)
1220 {
1221         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1222
1223         LASSERT(policy->sp_cops);
1224         LASSERT(policy->sp_cops->flush_ctx_cache);
1225
1226         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1227 }
1228
1229 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1230 {
1231         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1232
1233         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1234         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1235         LASSERT(policy->sp_cops->destroy_sec);
1236
1237         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1238
1239         policy->sp_cops->destroy_sec(sec);
1240         sptlrpc_policy_put(policy);
1241 }
1242
1243 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1244 {
1245         sec_cop_destroy_sec(sec);
1246 }
1247 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1248
1249 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1250 {
1251         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1252
1253         if (sec->ps_policy->sp_cops->kill_sec) {
1254                 sec->ps_policy->sp_cops->kill_sec(sec);
1255
1256                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1257         }
1258 }
1259
1260 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1261 {
1262         if (sec)
1263                 cfs_atomic_inc(&sec->ps_refcount);
1264
1265         return sec;
1266 }
1267 EXPORT_SYMBOL(sptlrpc_sec_get);
1268
1269 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1270 {
1271         if (sec) {
1272                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1273
1274                 if (cfs_atomic_dec_and_test(&sec->ps_refcount)) {
1275                         sptlrpc_gc_del_sec(sec);
1276                         sec_cop_destroy_sec(sec);
1277                 }
1278         }
1279 }
1280 EXPORT_SYMBOL(sptlrpc_sec_put);
1281
1282 /*
1283  * policy module is responsible for taking refrence of import
1284  */
1285 static
1286 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1287                                        struct ptlrpc_svc_ctx *svc_ctx,
1288                                        struct sptlrpc_flavor *sf,
1289                                        enum lustre_sec_part sp)
1290 {
1291         struct ptlrpc_sec_policy *policy;
1292         struct ptlrpc_sec        *sec;
1293         char                      str[32];
1294         ENTRY;
1295
1296         if (svc_ctx) {
1297                 LASSERT(imp->imp_dlm_fake == 1);
1298
1299                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1300                        imp->imp_obd->obd_type->typ_name,
1301                        imp->imp_obd->obd_name,
1302                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1303
1304                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1305                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1306         } else {
1307                 LASSERT(imp->imp_dlm_fake == 0);
1308
1309                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1310                        imp->imp_obd->obd_type->typ_name,
1311                        imp->imp_obd->obd_name,
1312                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1313
1314                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1315                 if (!policy) {
1316                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1317                         RETURN(NULL);
1318                 }
1319         }
1320
1321         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1322         if (sec) {
1323                 cfs_atomic_inc(&sec->ps_refcount);
1324
1325                 sec->ps_part = sp;
1326
1327                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1328                         sptlrpc_gc_add_sec(sec);
1329         } else {
1330                 sptlrpc_policy_put(policy);
1331         }
1332
1333         RETURN(sec);
1334 }
1335
1336 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1337 {
1338         struct ptlrpc_sec *sec;
1339
1340         cfs_spin_lock(&imp->imp_lock);
1341         sec = sptlrpc_sec_get(imp->imp_sec);
1342         cfs_spin_unlock(&imp->imp_lock);
1343
1344         return sec;
1345 }
1346 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1347
1348 static void sptlrpc_import_sec_install(struct obd_import *imp,
1349                                        struct ptlrpc_sec *sec)
1350 {
1351         struct ptlrpc_sec *old_sec;
1352
1353         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1354
1355         cfs_spin_lock(&imp->imp_lock);
1356         old_sec = imp->imp_sec;
1357         imp->imp_sec = sec;
1358         cfs_spin_unlock(&imp->imp_lock);
1359
1360         if (old_sec) {
1361                 sptlrpc_sec_kill(old_sec);
1362
1363                 /* balance the ref taken by this import */
1364                 sptlrpc_sec_put(old_sec);
1365         }
1366 }
1367
1368 static inline
1369 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1370 {
1371         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1372 }
1373
1374 static inline
1375 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1376 {
1377         *dst = *src;
1378 }
1379
1380 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1381                                              struct ptlrpc_sec *sec,
1382                                              struct sptlrpc_flavor *sf)
1383 {
1384         char    str1[32], str2[32];
1385
1386         if (sec->ps_flvr.sf_flags != sf->sf_flags)
1387                 CWARN("changing sec flags: %s -> %s\n",
1388                       sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1389                                            str1, sizeof(str1)),
1390                       sptlrpc_secflags2str(sf->sf_flags,
1391                                            str2, sizeof(str2)));
1392
1393         cfs_spin_lock(&sec->ps_lock);
1394         flavor_copy(&sec->ps_flvr, sf);
1395         cfs_spin_unlock(&sec->ps_lock);
1396 }
1397
1398 /**
1399  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1400  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1401  *
1402  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1403  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1404  */
1405 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1406                              struct ptlrpc_svc_ctx *svc_ctx,
1407                              struct sptlrpc_flavor *flvr)
1408 {
1409         struct ptlrpc_connection   *conn;
1410         struct sptlrpc_flavor       sf;
1411         struct ptlrpc_sec          *sec, *newsec;
1412         enum lustre_sec_part        sp;
1413         char                        str[24];
1414         int                         rc = 0;
1415         ENTRY;
1416
1417         cfs_might_sleep();
1418
1419         if (imp == NULL)
1420                 RETURN(0);
1421
1422         conn = imp->imp_connection;
1423
1424         if (svc_ctx == NULL) {
1425                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1426                 /*
1427                  * normal import, determine flavor from rule set, except
1428                  * for mgc the flavor is predetermined.
1429                  */
1430                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1431                         sf = cliobd->cl_flvr_mgc;
1432                 else
1433                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1434                                                    cliobd->cl_sp_to,
1435                                                    &cliobd->cl_target_uuid,
1436                                                    conn->c_self, &sf);
1437
1438                 sp = imp->imp_obd->u.cli.cl_sp_me;
1439         } else {
1440                 /* reverse import, determine flavor from incoming reqeust */
1441                 sf = *flvr;
1442
1443                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1444                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1445                                       PTLRPC_SEC_FL_ROOTONLY;
1446
1447                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1448         }
1449
1450         sec = sptlrpc_import_sec_ref(imp);
1451         if (sec) {
1452                 char    str2[24];
1453
1454                 if (flavor_equal(&sf, &sec->ps_flvr))
1455                         GOTO(out, rc);
1456
1457                 CWARN("import %s->%s: changing flavor %s -> %s\n",
1458                       imp->imp_obd->obd_name,
1459                       obd_uuid2str(&conn->c_remote_uuid),
1460                       sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1461                       sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1462
1463                 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1464                     SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1465                     SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1466                     SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1467                         sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1468                         GOTO(out, rc);
1469                 }
1470         } else if (sptlrpc_flavor2name_base(sf.sf_rpc) != SPTLRPC_FLVR_NULL) {
1471                 LCONSOLE_INFO("import %s->%s netid %x: select flavor %s\n",
1472                               imp->imp_obd->obd_name,
1473                               obd_uuid2str(&conn->c_remote_uuid),
1474                               LNET_NIDNET(conn->c_self),
1475                               sptlrpc_flavor2name(&sf, str, sizeof(str)));
1476         }
1477
1478         cfs_mutex_down(&imp->imp_sec_mutex);
1479
1480         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1481         if (newsec) {
1482                 sptlrpc_import_sec_install(imp, newsec);
1483         } else {
1484                 CERROR("import %s->%s: failed to create new sec\n",
1485                        imp->imp_obd->obd_name,
1486                        obd_uuid2str(&conn->c_remote_uuid));
1487                 rc = -EPERM;
1488         }
1489
1490         cfs_mutex_up(&imp->imp_sec_mutex);
1491 out:
1492         sptlrpc_sec_put(sec);
1493         RETURN(rc);
1494 }
1495
1496 void sptlrpc_import_sec_put(struct obd_import *imp)
1497 {
1498         if (imp->imp_sec) {
1499                 sptlrpc_sec_kill(imp->imp_sec);
1500
1501                 sptlrpc_sec_put(imp->imp_sec);
1502                 imp->imp_sec = NULL;
1503         }
1504 }
1505
1506 static void import_flush_ctx_common(struct obd_import *imp,
1507                                     uid_t uid, int grace, int force)
1508 {
1509         struct ptlrpc_sec *sec;
1510
1511         if (imp == NULL)
1512                 return;
1513
1514         sec = sptlrpc_import_sec_ref(imp);
1515         if (sec == NULL)
1516                 return;
1517
1518         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1519         sptlrpc_sec_put(sec);
1520 }
1521
1522 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1523 {
1524         /* it's important to use grace mode, see explain in
1525          * sptlrpc_req_refresh_ctx() */
1526         import_flush_ctx_common(imp, 0, 1, 1);
1527 }
1528
1529 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1530 {
1531         import_flush_ctx_common(imp, cfs_curproc_uid(), 1, 1);
1532 }
1533 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1534
1535 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1536 {
1537         import_flush_ctx_common(imp, -1, 1, 1);
1538 }
1539 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1540
1541 /**
1542  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1543  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1544  */
1545 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1546 {
1547         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1548         struct ptlrpc_sec_policy *policy;
1549         int rc;
1550
1551         LASSERT(ctx);
1552         LASSERT(ctx->cc_sec);
1553         LASSERT(ctx->cc_sec->ps_policy);
1554         LASSERT(req->rq_reqmsg == NULL);
1555         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1556
1557         policy = ctx->cc_sec->ps_policy;
1558         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1559         if (!rc) {
1560                 LASSERT(req->rq_reqmsg);
1561                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1562
1563                 /* zeroing preallocated buffer */
1564                 if (req->rq_pool)
1565                         memset(req->rq_reqmsg, 0, msgsize);
1566         }
1567
1568         return rc;
1569 }
1570
1571 /**
1572  * Used by ptlrpc client to free request buffer of \a req. After this
1573  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1574  */
1575 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1576 {
1577         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1578         struct ptlrpc_sec_policy *policy;
1579
1580         LASSERT(ctx);
1581         LASSERT(ctx->cc_sec);
1582         LASSERT(ctx->cc_sec->ps_policy);
1583         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1584
1585         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1586                 return;
1587
1588         policy = ctx->cc_sec->ps_policy;
1589         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1590         req->rq_reqmsg = NULL;
1591 }
1592
1593 /*
1594  * NOTE caller must guarantee the buffer size is enough for the enlargement
1595  */
1596 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1597                                   int segment, int newsize)
1598 {
1599         void   *src, *dst;
1600         int     oldsize, oldmsg_size, movesize;
1601
1602         LASSERT(segment < msg->lm_bufcount);
1603         LASSERT(msg->lm_buflens[segment] <= newsize);
1604
1605         if (msg->lm_buflens[segment] == newsize)
1606                 return;
1607
1608         /* nothing to do if we are enlarging the last segment */
1609         if (segment == msg->lm_bufcount - 1) {
1610                 msg->lm_buflens[segment] = newsize;
1611                 return;
1612         }
1613
1614         oldsize = msg->lm_buflens[segment];
1615
1616         src = lustre_msg_buf(msg, segment + 1, 0);
1617         msg->lm_buflens[segment] = newsize;
1618         dst = lustre_msg_buf(msg, segment + 1, 0);
1619         msg->lm_buflens[segment] = oldsize;
1620
1621         /* move from segment + 1 to end segment */
1622         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1623         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1624         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1625         LASSERT(movesize >= 0);
1626
1627         if (movesize)
1628                 memmove(dst, src, movesize);
1629
1630         /* note we don't clear the ares where old data live, not secret */
1631
1632         /* finally set new segment size */
1633         msg->lm_buflens[segment] = newsize;
1634 }
1635 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1636
1637 /**
1638  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1639  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1640  * preserved after the enlargement. this must be called after original request
1641  * buffer being allocated.
1642  *
1643  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1644  * so caller should refresh its local pointers if needed.
1645  */
1646 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1647                                int segment, int newsize)
1648 {
1649         struct ptlrpc_cli_ctx    *ctx = req->rq_cli_ctx;
1650         struct ptlrpc_sec_cops   *cops;
1651         struct lustre_msg        *msg = req->rq_reqmsg;
1652
1653         LASSERT(ctx);
1654         LASSERT(msg);
1655         LASSERT(msg->lm_bufcount > segment);
1656         LASSERT(msg->lm_buflens[segment] <= newsize);
1657
1658         if (msg->lm_buflens[segment] == newsize)
1659                 return 0;
1660
1661         cops = ctx->cc_sec->ps_policy->sp_cops;
1662         LASSERT(cops->enlarge_reqbuf);
1663         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1664 }
1665 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1666
1667 /**
1668  * Used by ptlrpc client to allocate reply buffer of \a req.
1669  *
1670  * \note After this, req->rq_repmsg is still not accessible.
1671  */
1672 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1673 {
1674         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1675         struct ptlrpc_sec_policy *policy;
1676         ENTRY;
1677
1678         LASSERT(ctx);
1679         LASSERT(ctx->cc_sec);
1680         LASSERT(ctx->cc_sec->ps_policy);
1681
1682         if (req->rq_repbuf)
1683                 RETURN(0);
1684
1685         policy = ctx->cc_sec->ps_policy;
1686         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1687 }
1688
1689 /**
1690  * Used by ptlrpc client to free reply buffer of \a req. After this
1691  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1692  */
1693 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1694 {
1695         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1696         struct ptlrpc_sec_policy *policy;
1697         ENTRY;
1698
1699         LASSERT(ctx);
1700         LASSERT(ctx->cc_sec);
1701         LASSERT(ctx->cc_sec->ps_policy);
1702         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1703
1704         if (req->rq_repbuf == NULL)
1705                 return;
1706         LASSERT(req->rq_repbuf_len);
1707
1708         policy = ctx->cc_sec->ps_policy;
1709         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1710         req->rq_repmsg = NULL;
1711         EXIT;
1712 }
1713
1714 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1715                                 struct ptlrpc_cli_ctx *ctx)
1716 {
1717         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1718
1719         if (!policy->sp_cops->install_rctx)
1720                 return 0;
1721         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1722 }
1723
1724 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1725                                 struct ptlrpc_svc_ctx *ctx)
1726 {
1727         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1728
1729         if (!policy->sp_sops->install_rctx)
1730                 return 0;
1731         return policy->sp_sops->install_rctx(imp, ctx);
1732 }
1733
1734 /****************************************
1735  * server side security                 *
1736  ****************************************/
1737
1738 static int flavor_allowed(struct sptlrpc_flavor *exp,
1739                           struct ptlrpc_request *req)
1740 {
1741         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1742
1743         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1744                 return 1;
1745
1746         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1747             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1748             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1749             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1750                 return 1;
1751
1752         return 0;
1753 }
1754
1755 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1756
1757 /**
1758  * Given an export \a exp, check whether the flavor of incoming \a req
1759  * is allowed by the export \a exp. Main logic is about taking care of
1760  * changing configurations. Return 0 means success.
1761  */
1762 int sptlrpc_target_export_check(struct obd_export *exp,
1763                                 struct ptlrpc_request *req)
1764 {
1765         struct sptlrpc_flavor   flavor;
1766
1767         if (exp == NULL)
1768                 return 0;
1769
1770         /* client side export has no imp_reverse, skip
1771          * FIXME maybe we should check flavor this as well??? */
1772         if (exp->exp_imp_reverse == NULL)
1773                 return 0;
1774
1775         /* don't care about ctx fini rpc */
1776         if (req->rq_ctx_fini)
1777                 return 0;
1778
1779         cfs_spin_lock(&exp->exp_lock);
1780
1781         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1782          * the first req with the new flavor, then treat it as current flavor,
1783          * adapt reverse sec according to it.
1784          * note the first rpc with new flavor might not be with root ctx, in
1785          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1786         if (unlikely(exp->exp_flvr_changed) &&
1787             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1788                 /* make the new flavor as "current", and old ones as
1789                  * about-to-expire */
1790                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1791                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1792                 flavor = exp->exp_flvr_old[1];
1793                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1794                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1795                 exp->exp_flvr_old[0] = exp->exp_flvr;
1796                 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1797                                           EXP_FLVR_UPDATE_EXPIRE;
1798                 exp->exp_flvr = flavor;
1799
1800                 /* flavor change finished */
1801                 exp->exp_flvr_changed = 0;
1802                 LASSERT(exp->exp_flvr_adapt == 1);
1803
1804                 /* if it's gss, we only interested in root ctx init */
1805                 if (req->rq_auth_gss &&
1806                     !(req->rq_ctx_init &&
1807                       (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1808                        req->rq_auth_usr_ost))) {
1809                         cfs_spin_unlock(&exp->exp_lock);
1810                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1811                                req->rq_auth_gss, req->rq_ctx_init,
1812                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1813                                req->rq_auth_usr_ost);
1814                         return 0;
1815                 }
1816
1817                 exp->exp_flvr_adapt = 0;
1818                 cfs_spin_unlock(&exp->exp_lock);
1819
1820                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1821                                                 req->rq_svc_ctx, &flavor);
1822         }
1823
1824         /* if it equals to the current flavor, we accept it, but need to
1825          * dealing with reverse sec/ctx */
1826         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1827                 /* most cases should return here, we only interested in
1828                  * gss root ctx init */
1829                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1830                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1831                      !req->rq_auth_usr_ost)) {
1832                         cfs_spin_unlock(&exp->exp_lock);
1833                         return 0;
1834                 }
1835
1836                 /* if flavor just changed, we should not proceed, just leave
1837                  * it and current flavor will be discovered and replaced
1838                  * shortly, and let _this_ rpc pass through */
1839                 if (exp->exp_flvr_changed) {
1840                         LASSERT(exp->exp_flvr_adapt);
1841                         cfs_spin_unlock(&exp->exp_lock);
1842                         return 0;
1843                 }
1844
1845                 if (exp->exp_flvr_adapt) {
1846                         exp->exp_flvr_adapt = 0;
1847                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1848                                exp, exp->exp_flvr.sf_rpc,
1849                                exp->exp_flvr_old[0].sf_rpc,
1850                                exp->exp_flvr_old[1].sf_rpc);
1851                         flavor = exp->exp_flvr;
1852                         cfs_spin_unlock(&exp->exp_lock);
1853
1854                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1855                                                         req->rq_svc_ctx,
1856                                                         &flavor);
1857                 } else {
1858                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1859                                "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1860                                exp->exp_flvr_old[0].sf_rpc,
1861                                exp->exp_flvr_old[1].sf_rpc);
1862                         cfs_spin_unlock(&exp->exp_lock);
1863
1864                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1865                                                            req->rq_svc_ctx);
1866                 }
1867         }
1868
1869         if (exp->exp_flvr_expire[0]) {
1870                 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1871                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1872                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1873                                        "middle one ("CFS_DURATION_T")\n", exp,
1874                                        exp->exp_flvr.sf_rpc,
1875                                        exp->exp_flvr_old[0].sf_rpc,
1876                                        exp->exp_flvr_old[1].sf_rpc,
1877                                        exp->exp_flvr_expire[0] -
1878                                                 cfs_time_current_sec());
1879                                 cfs_spin_unlock(&exp->exp_lock);
1880                                 return 0;
1881                         }
1882                 } else {
1883                         CDEBUG(D_SEC, "mark middle expired\n");
1884                         exp->exp_flvr_expire[0] = 0;
1885                 }
1886                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1887                        exp->exp_flvr.sf_rpc,
1888                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1889                        req->rq_flvr.sf_rpc);
1890         }
1891
1892         /* now it doesn't match the current flavor, the only chance we can
1893          * accept it is match the old flavors which is not expired. */
1894         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1895                 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1896                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1897                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1898                                        "oldest one ("CFS_DURATION_T")\n", exp,
1899                                        exp->exp_flvr.sf_rpc,
1900                                        exp->exp_flvr_old[0].sf_rpc,
1901                                        exp->exp_flvr_old[1].sf_rpc,
1902                                        exp->exp_flvr_expire[1] -
1903                                                 cfs_time_current_sec());
1904                                 cfs_spin_unlock(&exp->exp_lock);
1905                                 return 0;
1906                         }
1907                 } else {
1908                         CDEBUG(D_SEC, "mark oldest expired\n");
1909                         exp->exp_flvr_expire[1] = 0;
1910                 }
1911                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1912                        exp, exp->exp_flvr.sf_rpc,
1913                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1914                        req->rq_flvr.sf_rpc);
1915         } else {
1916                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1917                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1918                        exp->exp_flvr_old[1].sf_rpc);
1919         }
1920
1921         cfs_spin_unlock(&exp->exp_lock);
1922
1923         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1924               "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1925               exp, exp->exp_obd->obd_name,
1926               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1927               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1928               req->rq_flvr.sf_rpc,
1929               exp->exp_flvr.sf_rpc,
1930               exp->exp_flvr_old[0].sf_rpc,
1931               exp->exp_flvr_expire[0] ?
1932               (unsigned long) (exp->exp_flvr_expire[0] -
1933                                cfs_time_current_sec()) : 0,
1934               exp->exp_flvr_old[1].sf_rpc,
1935               exp->exp_flvr_expire[1] ?
1936               (unsigned long) (exp->exp_flvr_expire[1] -
1937                                cfs_time_current_sec()) : 0);
1938         return -EACCES;
1939 }
1940 EXPORT_SYMBOL(sptlrpc_target_export_check);
1941
1942 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1943                                       struct sptlrpc_rule_set *rset)
1944 {
1945         struct obd_export       *exp;
1946         struct sptlrpc_flavor    new_flvr;
1947
1948         LASSERT(obd);
1949
1950         cfs_spin_lock(&obd->obd_dev_lock);
1951
1952         cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1953                 if (exp->exp_connection == NULL)
1954                         continue;
1955
1956                 /* note if this export had just been updated flavor
1957                  * (exp_flvr_changed == 1), this will override the
1958                  * previous one. */
1959                 cfs_spin_lock(&exp->exp_lock);
1960                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1961                                              exp->exp_connection->c_peer.nid,
1962                                              &new_flvr);
1963                 if (exp->exp_flvr_changed ||
1964                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1965                         exp->exp_flvr_old[1] = new_flvr;
1966                         exp->exp_flvr_expire[1] = 0;
1967                         exp->exp_flvr_changed = 1;
1968                         exp->exp_flvr_adapt = 1;
1969
1970                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1971                                exp, sptlrpc_part2name(exp->exp_sp_peer),
1972                                exp->exp_flvr.sf_rpc,
1973                                exp->exp_flvr_old[1].sf_rpc);
1974                 }
1975                 cfs_spin_unlock(&exp->exp_lock);
1976         }
1977
1978         cfs_spin_unlock(&obd->obd_dev_lock);
1979 }
1980 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1981
1982 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1983 {
1984         /* peer's claim is unreliable unless gss is being used */
1985         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1986                 return svc_rc;
1987
1988         switch (req->rq_sp_from) {
1989         case LUSTRE_SP_CLI:
1990                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1991                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
1992                         svc_rc = SECSVC_DROP;
1993                 }
1994                 break;
1995         case LUSTRE_SP_MDT:
1996                 if (!req->rq_auth_usr_mdt) {
1997                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
1998                         svc_rc = SECSVC_DROP;
1999                 }
2000                 break;
2001         case LUSTRE_SP_OST:
2002                 if (!req->rq_auth_usr_ost) {
2003                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2004                         svc_rc = SECSVC_DROP;
2005                 }
2006                 break;
2007         case LUSTRE_SP_MGS:
2008         case LUSTRE_SP_MGC:
2009                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2010                     !req->rq_auth_usr_ost) {
2011                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2012                         svc_rc = SECSVC_DROP;
2013                 }
2014                 break;
2015         case LUSTRE_SP_ANY:
2016         default:
2017                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2018                 svc_rc = SECSVC_DROP;
2019         }
2020
2021         return svc_rc;
2022 }
2023
2024 /**
2025  * Used by ptlrpc server, to perform transformation upon request message of
2026  * incoming \a req. This must be the first thing to do with a incoming
2027  * request in ptlrpc layer.
2028  *
2029  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2030  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2031  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2032  * reply message has been prepared.
2033  * \retval SECSVC_DROP failed, this request should be dropped.
2034  */
2035 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2036 {
2037         struct ptlrpc_sec_policy *policy;
2038         struct lustre_msg        *msg = req->rq_reqbuf;
2039         int                       rc;
2040         ENTRY;
2041
2042         LASSERT(msg);
2043         LASSERT(req->rq_reqmsg == NULL);
2044         LASSERT(req->rq_repmsg == NULL);
2045         LASSERT(req->rq_svc_ctx == NULL);
2046
2047         req->rq_req_swab_mask = 0;
2048
2049         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2050         switch (rc) {
2051         case 1:
2052                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2053         case 0:
2054                 break;
2055         default:
2056                 CERROR("error unpacking request from %s x"LPU64"\n",
2057                        libcfs_id2str(req->rq_peer), req->rq_xid);
2058                 RETURN(SECSVC_DROP);
2059         }
2060
2061         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2062         req->rq_sp_from = LUSTRE_SP_ANY;
2063         req->rq_auth_uid = INVALID_UID;
2064         req->rq_auth_mapped_uid = INVALID_UID;
2065
2066         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2067         if (!policy) {
2068                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2069                 RETURN(SECSVC_DROP);
2070         }
2071
2072         LASSERT(policy->sp_sops->accept);
2073         rc = policy->sp_sops->accept(req);
2074         sptlrpc_policy_put(policy);
2075         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2076         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2077
2078         /*
2079          * if it's not null flavor (which means embedded packing msg),
2080          * reset the swab mask for the comming inner msg unpacking.
2081          */
2082         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2083                 req->rq_req_swab_mask = 0;
2084
2085         /* sanity check for the request source */
2086         rc = sptlrpc_svc_check_from(req, rc);
2087         RETURN(rc);
2088 }
2089
2090 /**
2091  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2092  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2093  * a buffer of \a msglen size.
2094  */
2095 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2096 {
2097         struct ptlrpc_sec_policy *policy;
2098         struct ptlrpc_reply_state *rs;
2099         int rc;
2100         ENTRY;
2101
2102         LASSERT(req->rq_svc_ctx);
2103         LASSERT(req->rq_svc_ctx->sc_policy);
2104
2105         policy = req->rq_svc_ctx->sc_policy;
2106         LASSERT(policy->sp_sops->alloc_rs);
2107
2108         rc = policy->sp_sops->alloc_rs(req, msglen);
2109         if (unlikely(rc == -ENOMEM)) {
2110                 /* failed alloc, try emergency pool */
2111                 rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service);
2112                 if (rs == NULL)
2113                         RETURN(-ENOMEM);
2114
2115                 req->rq_reply_state = rs;
2116                 rc = policy->sp_sops->alloc_rs(req, msglen);
2117                 if (rc) {
2118                         lustre_put_emerg_rs(rs);
2119                         req->rq_reply_state = NULL;
2120                 }
2121         }
2122
2123         LASSERT(rc != 0 ||
2124                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2125
2126         RETURN(rc);
2127 }
2128
2129 /**
2130  * Used by ptlrpc server, to perform transformation upon reply message.
2131  *
2132  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2133  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2134  */
2135 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2136 {
2137         struct ptlrpc_sec_policy *policy;
2138         int rc;
2139         ENTRY;
2140
2141         LASSERT(req->rq_svc_ctx);
2142         LASSERT(req->rq_svc_ctx->sc_policy);
2143
2144         policy = req->rq_svc_ctx->sc_policy;
2145         LASSERT(policy->sp_sops->authorize);
2146
2147         rc = policy->sp_sops->authorize(req);
2148         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2149
2150         RETURN(rc);
2151 }
2152
2153 /**
2154  * Used by ptlrpc server, to free reply_state.
2155  */
2156 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2157 {
2158         struct ptlrpc_sec_policy *policy;
2159         unsigned int prealloc;
2160         ENTRY;
2161
2162         LASSERT(rs->rs_svc_ctx);
2163         LASSERT(rs->rs_svc_ctx->sc_policy);
2164
2165         policy = rs->rs_svc_ctx->sc_policy;
2166         LASSERT(policy->sp_sops->free_rs);
2167
2168         prealloc = rs->rs_prealloc;
2169         policy->sp_sops->free_rs(rs);
2170
2171         if (prealloc)
2172                 lustre_put_emerg_rs(rs);
2173         EXIT;
2174 }
2175
2176 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2177 {
2178         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2179
2180         if (ctx != NULL)
2181                 cfs_atomic_inc(&ctx->sc_refcount);
2182 }
2183
2184 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2185 {
2186         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2187
2188         if (ctx == NULL)
2189                 return;
2190
2191         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2192         if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) {
2193                 if (ctx->sc_policy->sp_sops->free_ctx)
2194                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2195         }
2196         req->rq_svc_ctx = NULL;
2197 }
2198
2199 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2200 {
2201         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2202
2203         if (ctx == NULL)
2204                 return;
2205
2206         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2207         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2208                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2209 }
2210 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2211
2212 /****************************************
2213  * bulk security                        *
2214  ****************************************/
2215
2216 /**
2217  * Perform transformation upon bulk data pointed by \a desc. This is called
2218  * before transforming the request message.
2219  */
2220 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2221                           struct ptlrpc_bulk_desc *desc)
2222 {
2223         struct ptlrpc_cli_ctx *ctx;
2224
2225         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2226
2227         if (!req->rq_pack_bulk)
2228                 return 0;
2229
2230         ctx = req->rq_cli_ctx;
2231         if (ctx->cc_ops->wrap_bulk)
2232                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2233         return 0;
2234 }
2235 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2236
2237 /**
2238  * This is called after unwrap the reply message.
2239  * return nob of actual plain text size received, or error code.
2240  */
2241 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2242                                  struct ptlrpc_bulk_desc *desc,
2243                                  int nob)
2244 {
2245         struct ptlrpc_cli_ctx  *ctx;
2246         int                     rc;
2247
2248         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2249
2250         if (!req->rq_pack_bulk)
2251                 return desc->bd_nob_transferred;
2252
2253         ctx = req->rq_cli_ctx;
2254         if (ctx->cc_ops->unwrap_bulk) {
2255                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2256                 if (rc < 0)
2257                         return rc;
2258         }
2259         return desc->bd_nob_transferred;
2260 }
2261 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2262
2263 /**
2264  * This is called after unwrap the reply message.
2265  * return 0 for success or error code.
2266  */
2267 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2268                                   struct ptlrpc_bulk_desc *desc)
2269 {
2270         struct ptlrpc_cli_ctx  *ctx;
2271         int                     rc;
2272
2273         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2274
2275         if (!req->rq_pack_bulk)
2276                 return 0;
2277
2278         ctx = req->rq_cli_ctx;
2279         if (ctx->cc_ops->unwrap_bulk) {
2280                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2281                 if (rc < 0)
2282                         return rc;
2283         }
2284
2285         /*
2286          * if everything is going right, nob should equals to nob_transferred.
2287          * in case of privacy mode, nob_transferred needs to be adjusted.
2288          */
2289         if (desc->bd_nob != desc->bd_nob_transferred) {
2290                 CERROR("nob %d doesn't match transferred nob %d",
2291                        desc->bd_nob, desc->bd_nob_transferred);
2292                 return -EPROTO;
2293         }
2294
2295         return 0;
2296 }
2297 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2298
2299 /**
2300  * Performe transformation upon outgoing bulk read.
2301  */
2302 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2303                           struct ptlrpc_bulk_desc *desc)
2304 {
2305         struct ptlrpc_svc_ctx *ctx;
2306
2307         LASSERT(req->rq_bulk_read);
2308
2309         if (!req->rq_pack_bulk)
2310                 return 0;
2311
2312         ctx = req->rq_svc_ctx;
2313         if (ctx->sc_policy->sp_sops->wrap_bulk)
2314                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2315
2316         return 0;
2317 }
2318 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2319
2320 /**
2321  * Performe transformation upon incoming bulk write.
2322  */
2323 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2324                             struct ptlrpc_bulk_desc *desc)
2325 {
2326         struct ptlrpc_svc_ctx *ctx;
2327         int                    rc;
2328
2329         LASSERT(req->rq_bulk_write);
2330
2331         /*
2332          * if it's in privacy mode, transferred should >= expected; otherwise
2333          * transferred should == expected.
2334          */
2335         if (desc->bd_nob_transferred < desc->bd_nob ||
2336             (desc->bd_nob_transferred > desc->bd_nob &&
2337              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2338              SPTLRPC_BULK_SVC_PRIV)) {
2339                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2340                           desc->bd_nob_transferred, desc->bd_nob);
2341                 return -ETIMEDOUT;
2342         }
2343
2344         if (!req->rq_pack_bulk)
2345                 return 0;
2346
2347         ctx = req->rq_svc_ctx;
2348         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2349                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2350                 if (rc)
2351                         CERROR("error unwrap bulk: %d\n", rc);
2352         }
2353
2354         /* return 0 to allow reply be sent */
2355         return 0;
2356 }
2357 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2358
2359 /**
2360  * Prepare buffers for incoming bulk write.
2361  */
2362 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2363                           struct ptlrpc_bulk_desc *desc)
2364 {
2365         struct ptlrpc_svc_ctx *ctx;
2366
2367         LASSERT(req->rq_bulk_write);
2368
2369         if (!req->rq_pack_bulk)
2370                 return 0;
2371
2372         ctx = req->rq_svc_ctx;
2373         if (ctx->sc_policy->sp_sops->prep_bulk)
2374                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2375
2376         return 0;
2377 }
2378 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2379
2380 /****************************************
2381  * user descriptor helpers              *
2382  ****************************************/
2383
2384 int sptlrpc_current_user_desc_size(void)
2385 {
2386         int ngroups;
2387
2388 #ifdef __KERNEL__
2389         ngroups = current_ngroups;
2390
2391         if (ngroups > LUSTRE_MAX_GROUPS)
2392                 ngroups = LUSTRE_MAX_GROUPS;
2393 #else
2394         ngroups = 0;
2395 #endif
2396         return sptlrpc_user_desc_size(ngroups);
2397 }
2398 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2399
2400 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2401 {
2402         struct ptlrpc_user_desc *pud;
2403
2404         pud = lustre_msg_buf(msg, offset, 0);
2405
2406         pud->pud_uid = cfs_curproc_uid();
2407         pud->pud_gid = cfs_curproc_gid();
2408         pud->pud_fsuid = cfs_curproc_fsuid();
2409         pud->pud_fsgid = cfs_curproc_fsgid();
2410         pud->pud_cap = cfs_curproc_cap_pack();
2411         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2412
2413 #ifdef __KERNEL__
2414         task_lock(current);
2415         if (pud->pud_ngroups > current_ngroups)
2416                 pud->pud_ngroups = current_ngroups;
2417         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2418                pud->pud_ngroups * sizeof(__u32));
2419         task_unlock(current);
2420 #endif
2421
2422         return 0;
2423 }
2424 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2425
2426 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2427 {
2428         struct ptlrpc_user_desc *pud;
2429         int                      i;
2430
2431         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2432         if (!pud)
2433                 return -EINVAL;
2434
2435         if (swabbed) {
2436                 __swab32s(&pud->pud_uid);
2437                 __swab32s(&pud->pud_gid);
2438                 __swab32s(&pud->pud_fsuid);
2439                 __swab32s(&pud->pud_fsgid);
2440                 __swab32s(&pud->pud_cap);
2441                 __swab32s(&pud->pud_ngroups);
2442         }
2443
2444         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2445                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2446                 return -EINVAL;
2447         }
2448
2449         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2450             msg->lm_buflens[offset]) {
2451                 CERROR("%u groups are claimed but bufsize only %u\n",
2452                        pud->pud_ngroups, msg->lm_buflens[offset]);
2453                 return -EINVAL;
2454         }
2455
2456         if (swabbed) {
2457                 for (i = 0; i < pud->pud_ngroups; i++)
2458                         __swab32s(&pud->pud_groups[i]);
2459         }
2460
2461         return 0;
2462 }
2463 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2464
2465 /****************************************
2466  * misc helpers                         *
2467  ****************************************/
2468
2469 const char * sec2target_str(struct ptlrpc_sec *sec)
2470 {
2471         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2472                 return "*";
2473         if (sec_is_reverse(sec))
2474                 return "c";
2475         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2476 }
2477 EXPORT_SYMBOL(sec2target_str);
2478
2479 /*
2480  * return true if the bulk data is protected
2481  */
2482 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2483 {
2484         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2485         case SPTLRPC_BULK_SVC_INTG:
2486         case SPTLRPC_BULK_SVC_PRIV:
2487                 return 1;
2488         default:
2489                 return 0;
2490         }
2491 }
2492 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2493
2494 /****************************************
2495  * crypto API helper/alloc blkciper     *
2496  ****************************************/
2497
2498 /****************************************
2499  * initialize/finalize                  *
2500  ****************************************/
2501
2502 int sptlrpc_init(void)
2503 {
2504         int rc;
2505
2506         cfs_rwlock_init(&policy_lock);
2507
2508         rc = sptlrpc_gc_init();
2509         if (rc)
2510                 goto out;
2511
2512         rc = sptlrpc_conf_init();
2513         if (rc)
2514                 goto out_gc;
2515
2516         rc = sptlrpc_enc_pool_init();
2517         if (rc)
2518                 goto out_conf;
2519
2520         rc = sptlrpc_null_init();
2521         if (rc)
2522                 goto out_pool;
2523
2524         rc = sptlrpc_plain_init();
2525         if (rc)
2526                 goto out_null;
2527
2528         rc = sptlrpc_lproc_init();
2529         if (rc)
2530                 goto out_plain;
2531
2532         return 0;
2533
2534 out_plain:
2535         sptlrpc_plain_fini();
2536 out_null:
2537         sptlrpc_null_fini();
2538 out_pool:
2539         sptlrpc_enc_pool_fini();
2540 out_conf:
2541         sptlrpc_conf_fini();
2542 out_gc:
2543         sptlrpc_gc_fini();
2544 out:
2545         return rc;
2546 }
2547
2548 void sptlrpc_fini(void)
2549 {
2550         sptlrpc_lproc_fini();
2551         sptlrpc_plain_fini();
2552         sptlrpc_null_fini();
2553         sptlrpc_enc_pool_fini();
2554         sptlrpc_conf_fini();
2555         sptlrpc_gc_fini();
2556 }