lustre/ptlrpc/sec.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ptlrpc/sec.c
  37  *
  38  * Author: Eric Mei <ericm@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_SEC
  42
  43 #include <libcfs/libcfs.h>
  44 #ifndef __KERNEL__
  45 #include <liblustre.h>
  46 #include <libcfs/list.h>
  47 #else
  48 #include <linux/crypto.h>
  49 #include <linux/key.h>
  50 #endif
  51
  52 #include <obd.h>
  53 #include <obd_class.h>
  54 #include <obd_support.h>
  55 #include <lustre_net.h>
  56 #include <lustre_import.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_sec.h>
  59
  60 #include "ptlrpc_internal.h"
  61
  62 /***********************************************
  63  * policy registers                            *
  64  ***********************************************/
  65
  66 static rwlock_t policy_lock;
  67 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  68         NULL,
  69 };
  70
  71 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  72 {
  73         __u16 number = policy->sp_policy;
  74
  75         LASSERT(policy->sp_name);
  76         LASSERT(policy->sp_cops);
  77         LASSERT(policy->sp_sops);
  78
  79         if (number >= SPTLRPC_POLICY_MAX)
  80                 return -EINVAL;
  81
  82         write_lock(&policy_lock);
  83         if (unlikely(policies[number])) {
  84                 write_unlock(&policy_lock);
  85                 return -EALREADY;
  86         }
  87         policies[number] = policy;
  88         write_unlock(&policy_lock);
  89
  90         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  91         return 0;
  92 }
  93 EXPORT_SYMBOL(sptlrpc_register_policy);
  94
  95 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  96 {
  97         __u16 number = policy->sp_policy;
  98
  99         LASSERT(number < SPTLRPC_POLICY_MAX);
 100
 101         write_lock(&policy_lock);
 102         if (unlikely(policies[number] == NULL)) {
 103                 write_unlock(&policy_lock);
 104                 CERROR("%s: already unregistered\n", policy->sp_name);
 105                 return -EINVAL;
 106         }
 107
 108         LASSERT(policies[number] == policy);
 109         policies[number] = NULL;
 110         write_unlock(&policy_lock);
 111
 112         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 113         return 0;
 114 }
 115 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 116
 117 static
 118 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
 119 {
 120         static DEFINE_MUTEX(load_mutex);
 121         static cfs_atomic_t       loaded = CFS_ATOMIC_INIT(0);
 122         struct ptlrpc_sec_policy *policy;
 123         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 124         __u16                     flag = 0;
 125
 126         if (number >= SPTLRPC_POLICY_MAX)
 127                 return NULL;
 128
 129         while (1) {
 130                 read_lock(&policy_lock);
 131                 policy = policies[number];
 132                 if (policy && !try_module_get(policy->sp_owner))
 133                         policy = NULL;
 134                 if (policy == NULL)
 135                         flag = cfs_atomic_read(&loaded);
 136                 read_unlock(&policy_lock);
 137
 138                 if (policy != NULL || flag != 0 ||
 139                     number != SPTLRPC_POLICY_GSS)
 140                         break;
 141
 142                 /* try to load gss module, once */
 143                 mutex_lock(&load_mutex);
 144                 if (cfs_atomic_read(&loaded) == 0) {
 145                         if (request_module("ptlrpc_gss") == 0)
 146                                 CDEBUG(D_SEC,
 147                                        "module ptlrpc_gss loaded on demand\n");
 148                         else
 149                                 CERROR("Unable to load module ptlrpc_gss\n");
 150
 151                         cfs_atomic_set(&loaded, 1);
 152                 }
 153                 mutex_unlock(&load_mutex);
 154         }
 155
 156         return policy;
 157 }
 158
 159 __u32 sptlrpc_name2flavor_base(const char *name)
 160 {
 161         if (!strcmp(name, "null"))
 162                 return SPTLRPC_FLVR_NULL;
 163         if (!strcmp(name, "plain"))
 164                 return SPTLRPC_FLVR_PLAIN;
 165         if (!strcmp(name, "gssnull"))
 166                 return SPTLRPC_FLVR_GSSNULL;
 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         if (!strcmp(name, "ski"))
 176                 return SPTLRPC_FLVR_SKI;
 177         if (!strcmp(name, "skpi"))
 178                 return SPTLRPC_FLVR_SKPI;
 179
 180         return SPTLRPC_FLVR_INVALID;
 181 }
 182 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 183
 184 const char *sptlrpc_flavor2name_base(__u32 flvr)
 185 {
 186         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 187
 188         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 189                 return "null";
 190         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 191                 return "plain";
 192         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
 193                 return "gssnull";
 194         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 195                 return "krb5n";
 196         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 197                 return "krb5a";
 198         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 199                 return "krb5i";
 200         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 201                 return "krb5p";
 202         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
 203                 return "ski";
 204         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
 205                 return "skpi";
 206
 207         CERROR("invalid wire flavor 0x%x\n", flvr);
 208         return "invalid";
 209 }
 210 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 211
 212 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 213                                char *buf, int bufsize)
 214 {
 215         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 216                 snprintf(buf, bufsize, "hash:%s",
 217                          sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 218         else
 219                 snprintf(buf, bufsize, "%s",
 220                          sptlrpc_flavor2name_base(sf->sf_rpc));
 221
 222         buf[bufsize - 1] = '\0';
 223         return buf;
 224 }
 225 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 226
 227 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 228 {
 229         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 230
 231         /*
 232          * currently we don't support customized bulk specification for
 233          * flavors other than plain
 234          */
 235         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 236                 char bspec[16];
 237
 238                 bspec[0] = '-';
 239                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 240                 strncat(buf, bspec, bufsize);
 241         }
 242
 243         buf[bufsize - 1] = '\0';
 244         return buf;
 245 }
 246 EXPORT_SYMBOL(sptlrpc_flavor2name);
 247
 248 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 249 {
 250         buf[0] = '\0';
 251
 252         if (flags & PTLRPC_SEC_FL_REVERSE)
 253                 strlcat(buf, "reverse,", bufsize);
 254         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 255                 strlcat(buf, "rootonly,", bufsize);
 256         if (flags & PTLRPC_SEC_FL_UDESC)
 257                 strlcat(buf, "udesc,", bufsize);
 258         if (flags & PTLRPC_SEC_FL_BULK)
 259                 strlcat(buf, "bulk,", bufsize);
 260         if (buf[0] == '\0')
 261                 strlcat(buf, "-,", bufsize);
 262
 263         return buf;
 264 }
 265 EXPORT_SYMBOL(sptlrpc_secflags2str);
 266
 267 /**************************************************
 268  * client context APIs                            *
 269  **************************************************/
 270
 271 static
 272 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 273 {
 274         struct vfs_cred vcred;
 275         int create = 1, remove_dead = 1;
 276
 277         LASSERT(sec);
 278         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 279
 280         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 281                                      PTLRPC_SEC_FL_ROOTONLY)) {
 282                 vcred.vc_uid = 0;
 283                 vcred.vc_gid = 0;
 284                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 285                         create = 0;
 286                         remove_dead = 0;
 287                 }
 288         } else {
 289                 vcred.vc_uid = current_uid();
 290                 vcred.vc_gid = current_gid();
 291         }
 292
 293         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
 294                                                    remove_dead);
 295 }
 296
 297 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 298 {
 299         cfs_atomic_inc(&ctx->cc_refcount);
 300         return ctx;
 301 }
 302 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 303
 304 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 305 {
 306         struct ptlrpc_sec *sec = ctx->cc_sec;
 307
 308         LASSERT(sec);
 309         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 310
 311         if (!cfs_atomic_dec_and_test(&ctx->cc_refcount))
 312                 return;
 313
 314         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 315 }
 316 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 317
 318 /**
 319  * Expire the client context immediately.
 320  *
 321  * \pre Caller must hold at least 1 reference on the \a ctx.
 322  */
 323 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 324 {
 325         LASSERT(ctx->cc_ops->die);
 326         ctx->cc_ops->die(ctx, 0);
 327 }
 328 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 329
 330 /**
 331  * To wake up the threads who are waiting for this client context. Called
 332  * after some status change happened on \a ctx.
 333  */
 334 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 335 {
 336         struct ptlrpc_request *req, *next;
 337
 338         spin_lock(&ctx->cc_lock);
 339         cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 340                                      rq_ctx_chain) {
 341                 cfs_list_del_init(&req->rq_ctx_chain);
 342                 ptlrpc_client_wake_req(req);
 343         }
 344         spin_unlock(&ctx->cc_lock);
 345 }
 346 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 347
 348 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 349 {
 350         LASSERT(ctx->cc_ops);
 351
 352         if (ctx->cc_ops->display == NULL)
 353                 return 0;
 354
 355         return ctx->cc_ops->display(ctx, buf, bufsize);
 356 }
 357
 358 static int import_sec_check_expire(struct obd_import *imp)
 359 {
 360         int     adapt = 0;
 361
 362         spin_lock(&imp->imp_lock);
 363         if (imp->imp_sec_expire &&
 364             imp->imp_sec_expire < cfs_time_current_sec()) {
 365                 adapt = 1;
 366                 imp->imp_sec_expire = 0;
 367         }
 368         spin_unlock(&imp->imp_lock);
 369
 370         if (!adapt)
 371                 return 0;
 372
 373         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 374         return sptlrpc_import_sec_adapt(imp, NULL, 0);
 375 }
 376
 377 static int import_sec_validate_get(struct obd_import *imp,
 378                                    struct ptlrpc_sec **sec)
 379 {
 380         int     rc;
 381
 382         if (unlikely(imp->imp_sec_expire)) {
 383                 rc = import_sec_check_expire(imp);
 384                 if (rc)
 385                         return rc;
 386         }
 387
 388         *sec = sptlrpc_import_sec_ref(imp);
 389         if (*sec == NULL) {
 390                 CERROR("import %p (%s) with no sec\n",
 391                        imp, ptlrpc_import_state_name(imp->imp_state));
 392                 return -EACCES;
 393         }
 394
 395         if (unlikely((*sec)->ps_dying)) {
 396                 CERROR("attempt to use dying sec %p\n", sec);
 397                 sptlrpc_sec_put(*sec);
 398                 return -EACCES;
 399         }
 400
 401         return 0;
 402 }
 403
 404 /**
 405  * Given a \a req, find or allocate a appropriate context for it.
 406  * \pre req->rq_cli_ctx == NULL.
 407  *
 408  * \retval 0 succeed, and req->rq_cli_ctx is set.
 409  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 410  */
 411 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 412 {
 413         struct obd_import *imp = req->rq_import;
 414         struct ptlrpc_sec *sec;
 415         int                rc;
 416         ENTRY;
 417
 418         LASSERT(!req->rq_cli_ctx);
 419         LASSERT(imp);
 420
 421         rc = import_sec_validate_get(imp, &sec);
 422         if (rc)
 423                 RETURN(rc);
 424
 425         req->rq_cli_ctx = get_my_ctx(sec);
 426
 427         sptlrpc_sec_put(sec);
 428
 429         if (!req->rq_cli_ctx) {
 430                 CERROR("req %p: fail to get context\n", req);
 431                 RETURN(-ENOMEM);
 432         }
 433
 434         RETURN(0);
 435 }
 436
 437 /**
 438  * Drop the context for \a req.
 439  * \pre req->rq_cli_ctx != NULL.
 440  * \post req->rq_cli_ctx == NULL.
 441  *
 442  * If \a sync == 0, this function should return quickly without sleep;
 443  * otherwise it might trigger and wait for the whole process of sending
 444  * an context-destroying rpc to server.
 445  */
 446 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 447 {
 448         ENTRY;
 449
 450         LASSERT(req);
 451         LASSERT(req->rq_cli_ctx);
 452
 453         /* request might be asked to release earlier while still
 454          * in the context waiting list.
 455          */
 456         if (!cfs_list_empty(&req->rq_ctx_chain)) {
 457                 spin_lock(&req->rq_cli_ctx->cc_lock);
 458                 cfs_list_del_init(&req->rq_ctx_chain);
 459                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 460         }
 461
 462         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 463         req->rq_cli_ctx = NULL;
 464         EXIT;
 465 }
 466
 467 static
 468 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 469                            struct ptlrpc_cli_ctx *oldctx,
 470                            struct ptlrpc_cli_ctx *newctx)
 471 {
 472         struct sptlrpc_flavor   old_flvr;
 473         char                   *reqmsg = NULL; /* to workaround old gcc */
 474         int                     reqmsg_size;
 475         int                     rc = 0;
 476
 477         LASSERT(req->rq_reqmsg);
 478         LASSERT(req->rq_reqlen);
 479         LASSERT(req->rq_replen);
 480
 481         CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
 482                "switch sec %p(%s) -> %p(%s)\n", req,
 483                oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
 484                newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
 485                oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
 486                newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 487
 488         /* save flavor */
 489         old_flvr = req->rq_flvr;
 490
 491         /* save request message */
 492         reqmsg_size = req->rq_reqlen;
 493         if (reqmsg_size != 0) {
 494                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 495                 if (reqmsg == NULL)
 496                         return -ENOMEM;
 497                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 498         }
 499
 500         /* release old req/rep buf */
 501         req->rq_cli_ctx = oldctx;
 502         sptlrpc_cli_free_reqbuf(req);
 503         sptlrpc_cli_free_repbuf(req);
 504         req->rq_cli_ctx = newctx;
 505
 506         /* recalculate the flavor */
 507         sptlrpc_req_set_flavor(req, 0);
 508
 509         /* alloc new request buffer
 510          * we don't need to alloc reply buffer here, leave it to the
 511          * rest procedure of ptlrpc */
 512         if (reqmsg_size != 0) {
 513                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 514                 if (!rc) {
 515                         LASSERT(req->rq_reqmsg);
 516                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 517                 } else {
 518                         CWARN("failed to alloc reqbuf: %d\n", rc);
 519                         req->rq_flvr = old_flvr;
 520                 }
 521
 522                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 523         }
 524         return rc;
 525 }
 526
 527 /**
 528  * If current context of \a req is dead somehow, e.g. we just switched flavor
 529  * thus marked original contexts dead, we'll find a new context for it. if
 530  * no switch is needed, \a req will end up with the same context.
 531  *
 532  * \note a request must have a context, to keep other parts of code happy.
 533  * In any case of failure during the switching, we must restore the old one.
 534  */
 535 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 536 {
 537         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 538         struct ptlrpc_cli_ctx *newctx;
 539         int                    rc;
 540         ENTRY;
 541
 542         LASSERT(oldctx);
 543
 544         sptlrpc_cli_ctx_get(oldctx);
 545         sptlrpc_req_put_ctx(req, 0);
 546
 547         rc = sptlrpc_req_get_ctx(req);
 548         if (unlikely(rc)) {
 549                 LASSERT(!req->rq_cli_ctx);
 550
 551                 /* restore old ctx */
 552                 req->rq_cli_ctx = oldctx;
 553                 RETURN(rc);
 554         }
 555
 556         newctx = req->rq_cli_ctx;
 557         LASSERT(newctx);
 558
 559         if (unlikely(newctx == oldctx &&
 560                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 561                 /*
 562                  * still get the old dead ctx, usually means system too busy
 563                  */
 564                 CDEBUG(D_SEC,
 565                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 566                        newctx, newctx->cc_flags);
 567
 568                 schedule_timeout_and_set_state(TASK_INTERRUPTIBLE,
 569                                                    HZ);
 570         } else {
 571                 /*
 572                  * it's possible newctx == oldctx if we're switching
 573                  * subflavor with the same sec.
 574                  */
 575                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 576                 if (rc) {
 577                         /* restore old ctx */
 578                         sptlrpc_req_put_ctx(req, 0);
 579                         req->rq_cli_ctx = oldctx;
 580                         RETURN(rc);
 581                 }
 582
 583                 LASSERT(req->rq_cli_ctx == newctx);
 584         }
 585
 586         sptlrpc_cli_ctx_put(oldctx, 1);
 587         RETURN(0);
 588 }
 589 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 590
 591 static
 592 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 593 {
 594         if (cli_ctx_is_refreshed(ctx))
 595                 return 1;
 596         return 0;
 597 }
 598
 599 static
 600 int ctx_refresh_timeout(void *data)
 601 {
 602         struct ptlrpc_request *req = data;
 603         int rc;
 604
 605         /* conn_cnt is needed in expire_one_request */
 606         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 607
 608         rc = ptlrpc_expire_one_request(req, 1);
 609         /* if we started recovery, we should mark this ctx dead; otherwise
 610          * in case of lgssd died nobody would retire this ctx, following
 611          * connecting will still find the same ctx thus cause deadlock.
 612          * there's an assumption that expire time of the request should be
 613          * later than the context refresh expire time.
 614          */
 615         if (rc == 0)
 616                 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
 617         return rc;
 618 }
 619
 620 static
 621 void ctx_refresh_interrupt(void *data)
 622 {
 623         struct ptlrpc_request *req = data;
 624
 625         spin_lock(&req->rq_lock);
 626         req->rq_intr = 1;
 627         spin_unlock(&req->rq_lock);
 628 }
 629
 630 static
 631 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 632 {
 633         spin_lock(&ctx->cc_lock);
 634         if (!cfs_list_empty(&req->rq_ctx_chain))
 635                 cfs_list_del_init(&req->rq_ctx_chain);
 636         spin_unlock(&ctx->cc_lock);
 637 }
 638
 639 /**
 640  * To refresh the context of \req, if it's not up-to-date.
 641  * \param timeout
 642  * - < 0: don't wait
 643  * - = 0: wait until success or fatal error occur
 644  * - > 0: timeout value (in seconds)
 645  *
 646  * The status of the context could be subject to be changed by other threads
 647  * at any time. We allow this race, but once we return with 0, the caller will
 648  * suppose it's uptodated and keep using it until the owning rpc is done.
 649  *
 650  * \retval 0 only if the context is uptodated.
 651  * \retval -ev error number.
 652  */
 653 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 654 {
 655         struct ptlrpc_cli_ctx  *ctx = req->rq_cli_ctx;
 656         struct ptlrpc_sec      *sec;
 657         struct l_wait_info      lwi;
 658         int                     rc;
 659         ENTRY;
 660
 661         LASSERT(ctx);
 662
 663         if (req->rq_ctx_init || req->rq_ctx_fini)
 664                 RETURN(0);
 665
 666         /*
 667          * during the process a request's context might change type even
 668          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 669          * everything
 670          */
 671 again:
 672         rc = import_sec_validate_get(req->rq_import, &sec);
 673         if (rc)
 674                 RETURN(rc);
 675
 676         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 677                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 678                       req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 679                 req_off_ctx_list(req, ctx);
 680                 sptlrpc_req_replace_dead_ctx(req);
 681                 ctx = req->rq_cli_ctx;
 682         }
 683         sptlrpc_sec_put(sec);
 684
 685         if (cli_ctx_is_eternal(ctx))
 686                 RETURN(0);
 687
 688         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 689                 LASSERT(ctx->cc_ops->refresh);
 690                 ctx->cc_ops->refresh(ctx);
 691         }
 692         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 693
 694         LASSERT(ctx->cc_ops->validate);
 695         if (ctx->cc_ops->validate(ctx) == 0) {
 696                 req_off_ctx_list(req, ctx);
 697                 RETURN(0);
 698         }
 699
 700         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 701                 spin_lock(&req->rq_lock);
 702                 req->rq_err = 1;
 703                 spin_unlock(&req->rq_lock);
 704                 req_off_ctx_list(req, ctx);
 705                 RETURN(-EPERM);
 706         }
 707
 708         /*
 709          * There's a subtle issue for resending RPCs, suppose following
 710          * situation:
 711          *  1. the request was sent to server.
 712          *  2. recovery was kicked start, after finished the request was
 713          *     marked as resent.
 714          *  3. resend the request.
 715          *  4. old reply from server received, we accept and verify the reply.
 716          *     this has to be success, otherwise the error will be aware
 717          *     by application.
 718          *  5. new reply from server received, dropped by LNet.
 719          *
 720          * Note the xid of old & new request is the same. We can't simply
 721          * change xid for the resent request because the server replies on
 722          * it for reply reconstruction.
 723          *
 724          * Commonly the original context should be uptodate because we
 725          * have a expiry nice time; server will keep its context because
 726          * we at least hold a ref of old context which prevent context
 727          * destroying RPC being sent. So server still can accept the request
 728          * and finish the RPC. But if that's not the case:
 729          *  1. If server side context has been trimmed, a NO_CONTEXT will
 730          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 731          *     context by force.
 732          *  2. Current context never be refreshed, then we are fine: we
 733          *     never really send request with old context before.
 734          */
 735         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 736             unlikely(req->rq_reqmsg) &&
 737             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 738                 req_off_ctx_list(req, ctx);
 739                 RETURN(0);
 740         }
 741
 742         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 743                 req_off_ctx_list(req, ctx);
 744                 /*
 745                  * don't switch ctx if import was deactivated
 746                  */
 747                 if (req->rq_import->imp_deactive) {
 748                         spin_lock(&req->rq_lock);
 749                         req->rq_err = 1;
 750                         spin_unlock(&req->rq_lock);
 751                         RETURN(-EINTR);
 752                 }
 753
 754                 rc = sptlrpc_req_replace_dead_ctx(req);
 755                 if (rc) {
 756                         LASSERT(ctx == req->rq_cli_ctx);
 757                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 758                                req, ctx, rc);
 759                         spin_lock(&req->rq_lock);
 760                         req->rq_err = 1;
 761                         spin_unlock(&req->rq_lock);
 762                         RETURN(rc);
 763                 }
 764
 765                 ctx = req->rq_cli_ctx;
 766                 goto again;
 767         }
 768
 769         /*
 770          * Now we're sure this context is during upcall, add myself into
 771          * waiting list
 772          */
 773         spin_lock(&ctx->cc_lock);
 774         if (cfs_list_empty(&req->rq_ctx_chain))
 775                 cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 776         spin_unlock(&ctx->cc_lock);
 777
 778         if (timeout < 0)
 779                 RETURN(-EWOULDBLOCK);
 780
 781         /* Clear any flags that may be present from previous sends */
 782         LASSERT(req->rq_receiving_reply == 0);
 783         spin_lock(&req->rq_lock);
 784         req->rq_err = 0;
 785         req->rq_timedout = 0;
 786         req->rq_resend = 0;
 787         req->rq_restart = 0;
 788         spin_unlock(&req->rq_lock);
 789
 790         lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout,
 791                                ctx_refresh_interrupt, req);
 792         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 793
 794         /*
 795          * following cases could lead us here:
 796          * - successfully refreshed;
 797          * - interrupted;
 798          * - timedout, and we don't want recover from the failure;
 799          * - timedout, and waked up upon recovery finished;
 800          * - someone else mark this ctx dead by force;
 801          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 802          *   e.g. ptlrpc_abort_inflight();
 803          */
 804         if (!cli_ctx_is_refreshed(ctx)) {
 805                 /* timed out or interruptted */
 806                 req_off_ctx_list(req, ctx);
 807
 808                 LASSERT(rc != 0);
 809                 RETURN(rc);
 810         }
 811
 812         goto again;
 813 }
 814
 815 /**
 816  * Initialize flavor settings for \a req, according to \a opcode.
 817  *
 818  * \note this could be called in two situations:
 819  * - new request from ptlrpc_pre_req(), with proper @opcode
 820  * - old request which changed ctx in the middle, with @opcode == 0
 821  */
 822 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 823 {
 824         struct ptlrpc_sec *sec;
 825
 826         LASSERT(req->rq_import);
 827         LASSERT(req->rq_cli_ctx);
 828         LASSERT(req->rq_cli_ctx->cc_sec);
 829         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 830
 831         /* special security flags accoding to opcode */
 832         switch (opcode) {
 833         case OST_READ:
 834         case MDS_READPAGE:
 835         case MGS_CONFIG_READ:
 836         case OBD_IDX_READ:
 837                 req->rq_bulk_read = 1;
 838                 break;
 839         case OST_WRITE:
 840         case MDS_WRITEPAGE:
 841                 req->rq_bulk_write = 1;
 842                 break;
 843         case SEC_CTX_INIT:
 844                 req->rq_ctx_init = 1;
 845                 break;
 846         case SEC_CTX_FINI:
 847                 req->rq_ctx_fini = 1;
 848                 break;
 849         case 0:
 850                 /* init/fini rpc won't be resend, so can't be here */
 851                 LASSERT(req->rq_ctx_init == 0);
 852                 LASSERT(req->rq_ctx_fini == 0);
 853
 854                 /* cleanup flags, which should be recalculated */
 855                 req->rq_pack_udesc = 0;
 856                 req->rq_pack_bulk = 0;
 857                 break;
 858         }
 859
 860         sec = req->rq_cli_ctx->cc_sec;
 861
 862         spin_lock(&sec->ps_lock);
 863         req->rq_flvr = sec->ps_flvr;
 864         spin_unlock(&sec->ps_lock);
 865
 866         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
 867          * destruction rpc */
 868         if (unlikely(req->rq_ctx_init))
 869                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 870         else if (unlikely(req->rq_ctx_fini))
 871                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 872
 873         /* user descriptor flag, null security can't do it anyway */
 874         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 875             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 876                 req->rq_pack_udesc = 1;
 877
 878         /* bulk security flag */
 879         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 880             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 881                 req->rq_pack_bulk = 1;
 882 }
 883
 884 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 885 {
 886         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 887                 return;
 888
 889         LASSERT(req->rq_clrbuf);
 890         if (req->rq_pool || !req->rq_reqbuf)
 891                 return;
 892
 893         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 894         req->rq_reqbuf = NULL;
 895         req->rq_reqbuf_len = 0;
 896 }
 897
 898 /**
 899  * Given an import \a imp, check whether current user has a valid context
 900  * or not. We may create a new context and try to refresh it, and try
 901  * repeatedly try in case of non-fatal errors. Return 0 means success.
 902  */
 903 int sptlrpc_import_check_ctx(struct obd_import *imp)
 904 {
 905         struct ptlrpc_sec     *sec;
 906         struct ptlrpc_cli_ctx *ctx;
 907         struct ptlrpc_request *req = NULL;
 908         int rc;
 909         ENTRY;
 910
 911         might_sleep();
 912
 913         sec = sptlrpc_import_sec_ref(imp);
 914         ctx = get_my_ctx(sec);
 915         sptlrpc_sec_put(sec);
 916
 917         if (!ctx)
 918                 RETURN(-ENOMEM);
 919
 920         if (cli_ctx_is_eternal(ctx) ||
 921             ctx->cc_ops->validate(ctx) == 0) {
 922                 sptlrpc_cli_ctx_put(ctx, 1);
 923                 RETURN(0);
 924         }
 925
 926         if (cli_ctx_is_error(ctx)) {
 927                 sptlrpc_cli_ctx_put(ctx, 1);
 928                 RETURN(-EACCES);
 929         }
 930
 931         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 932         if (!req)
 933                 RETURN(-ENOMEM);
 934
 935         spin_lock_init(&req->rq_lock);
 936         cfs_atomic_set(&req->rq_refcount, 10000);
 937         CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
 938         init_waitqueue_head(&req->rq_reply_waitq);
 939         init_waitqueue_head(&req->rq_set_waitq);
 940         req->rq_import = imp;
 941         req->rq_flvr = sec->ps_flvr;
 942         req->rq_cli_ctx = ctx;
 943
 944         rc = sptlrpc_req_refresh_ctx(req, 0);
 945         LASSERT(cfs_list_empty(&req->rq_ctx_chain));
 946         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 947         ptlrpc_request_cache_free(req);
 948
 949         RETURN(rc);
 950 }
 951
 952 /**
 953  * Used by ptlrpc client, to perform the pre-defined security transformation
 954  * upon the request message of \a req. After this function called,
 955  * req->rq_reqmsg is still accessible as clear text.
 956  */
 957 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 958 {
 959         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 960         int rc = 0;
 961         ENTRY;
 962
 963         LASSERT(ctx);
 964         LASSERT(ctx->cc_sec);
 965         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
 966
 967         /* we wrap bulk request here because now we can be sure
 968          * the context is uptodate.
 969          */
 970         if (req->rq_bulk) {
 971                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
 972                 if (rc)
 973                         RETURN(rc);
 974         }
 975
 976         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 977         case SPTLRPC_SVC_NULL:
 978         case SPTLRPC_SVC_AUTH:
 979         case SPTLRPC_SVC_INTG:
 980                 LASSERT(ctx->cc_ops->sign);
 981                 rc = ctx->cc_ops->sign(ctx, req);
 982                 break;
 983         case SPTLRPC_SVC_PRIV:
 984                 LASSERT(ctx->cc_ops->seal);
 985                 rc = ctx->cc_ops->seal(ctx, req);
 986                 break;
 987         default:
 988                 LBUG();
 989         }
 990
 991         if (rc == 0) {
 992                 LASSERT(req->rq_reqdata_len);
 993                 LASSERT(req->rq_reqdata_len % 8 == 0);
 994                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
 995         }
 996
 997         RETURN(rc);
 998 }
 999
1000 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1001 {
1002         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1003         int                    rc;
1004         ENTRY;
1005
1006         LASSERT(ctx);
1007         LASSERT(ctx->cc_sec);
1008         LASSERT(req->rq_repbuf);
1009         LASSERT(req->rq_repdata);
1010         LASSERT(req->rq_repmsg == NULL);
1011
1012         req->rq_rep_swab_mask = 0;
1013
1014         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1015         switch (rc) {
1016         case 1:
1017                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1018         case 0:
1019                 break;
1020         default:
1021                 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1022                 RETURN(-EPROTO);
1023         }
1024
1025         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1026                 CERROR("replied data length %d too small\n",
1027                        req->rq_repdata_len);
1028                 RETURN(-EPROTO);
1029         }
1030
1031         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1032             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1033                 CERROR("reply policy %u doesn't match request policy %u\n",
1034                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1035                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1036                 RETURN(-EPROTO);
1037         }
1038
1039         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1040         case SPTLRPC_SVC_NULL:
1041         case SPTLRPC_SVC_AUTH:
1042         case SPTLRPC_SVC_INTG:
1043                 LASSERT(ctx->cc_ops->verify);
1044                 rc = ctx->cc_ops->verify(ctx, req);
1045                 break;
1046         case SPTLRPC_SVC_PRIV:
1047                 LASSERT(ctx->cc_ops->unseal);
1048                 rc = ctx->cc_ops->unseal(ctx, req);
1049                 break;
1050         default:
1051                 LBUG();
1052         }
1053         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1054
1055         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1056             !req->rq_ctx_init)
1057                 req->rq_rep_swab_mask = 0;
1058         RETURN(rc);
1059 }
1060
1061 /**
1062  * Used by ptlrpc client, to perform security transformation upon the reply
1063  * message of \a req. After return successfully, req->rq_repmsg points to
1064  * the reply message in clear text.
1065  *
1066  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1067  * going to change.
1068  */
1069 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1070 {
1071         LASSERT(req->rq_repbuf);
1072         LASSERT(req->rq_repdata == NULL);
1073         LASSERT(req->rq_repmsg == NULL);
1074         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1075
1076         if (req->rq_reply_off == 0 &&
1077             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1078                 CERROR("real reply with offset 0\n");
1079                 return -EPROTO;
1080         }
1081
1082         if (req->rq_reply_off % 8 != 0) {
1083                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1084                 return -EPROTO;
1085         }
1086
1087         req->rq_repdata = (struct lustre_msg *)
1088                                 (req->rq_repbuf + req->rq_reply_off);
1089         req->rq_repdata_len = req->rq_nob_received;
1090
1091         return do_cli_unwrap_reply(req);
1092 }
1093
1094 /**
1095  * Used by ptlrpc client, to perform security transformation upon the early
1096  * reply message of \a req. We expect the rq_reply_off is 0, and
1097  * rq_nob_received is the early reply size.
1098  *
1099  * Because the receive buffer might be still posted, the reply data might be
1100  * changed at any time, no matter we're holding rq_lock or not. For this reason
1101  * we allocate a separate ptlrpc_request and reply buffer for early reply
1102  * processing.
1103  *
1104  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1105  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1106  * \a *req_ret to release it.
1107  * \retval -ev error number, and \a req_ret will not be set.
1108  */
1109 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1110                                    struct ptlrpc_request **req_ret)
1111 {
1112         struct ptlrpc_request  *early_req;
1113         char                   *early_buf;
1114         int                     early_bufsz, early_size;
1115         int                     rc;
1116         ENTRY;
1117
1118         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1119         if (early_req == NULL)
1120                 RETURN(-ENOMEM);
1121
1122         early_size = req->rq_nob_received;
1123         early_bufsz = size_roundup_power2(early_size);
1124         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1125         if (early_buf == NULL)
1126                 GOTO(err_req, rc = -ENOMEM);
1127
1128         /* sanity checkings and copy data out, do it inside spinlock */
1129         spin_lock(&req->rq_lock);
1130
1131         if (req->rq_replied) {
1132                 spin_unlock(&req->rq_lock);
1133                 GOTO(err_buf, rc = -EALREADY);
1134         }
1135
1136         LASSERT(req->rq_repbuf);
1137         LASSERT(req->rq_repdata == NULL);
1138         LASSERT(req->rq_repmsg == NULL);
1139
1140         if (req->rq_reply_off != 0) {
1141                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1142                 spin_unlock(&req->rq_lock);
1143                 GOTO(err_buf, rc = -EPROTO);
1144         }
1145
1146         if (req->rq_nob_received != early_size) {
1147                 /* even another early arrived the size should be the same */
1148                 CERROR("data size has changed from %u to %u\n",
1149                        early_size, req->rq_nob_received);
1150                 spin_unlock(&req->rq_lock);
1151                 GOTO(err_buf, rc = -EINVAL);
1152         }
1153
1154         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1155                 CERROR("early reply length %d too small\n",
1156                        req->rq_nob_received);
1157                 spin_unlock(&req->rq_lock);
1158                 GOTO(err_buf, rc = -EALREADY);
1159         }
1160
1161         memcpy(early_buf, req->rq_repbuf, early_size);
1162         spin_unlock(&req->rq_lock);
1163
1164         spin_lock_init(&early_req->rq_lock);
1165         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1166         early_req->rq_flvr = req->rq_flvr;
1167         early_req->rq_repbuf = early_buf;
1168         early_req->rq_repbuf_len = early_bufsz;
1169         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1170         early_req->rq_repdata_len = early_size;
1171         early_req->rq_early = 1;
1172         early_req->rq_reqmsg = req->rq_reqmsg;
1173
1174         rc = do_cli_unwrap_reply(early_req);
1175         if (rc) {
1176                 DEBUG_REQ(D_ADAPTTO, early_req,
1177                           "error %d unwrap early reply", rc);
1178                 GOTO(err_ctx, rc);
1179         }
1180
1181         LASSERT(early_req->rq_repmsg);
1182         *req_ret = early_req;
1183         RETURN(0);
1184
1185 err_ctx:
1186         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1187 err_buf:
1188         OBD_FREE_LARGE(early_buf, early_bufsz);
1189 err_req:
1190         ptlrpc_request_cache_free(early_req);
1191         RETURN(rc);
1192 }
1193
1194 /**
1195  * Used by ptlrpc client, to release a processed early reply \a early_req.
1196  *
1197  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1198  */
1199 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1200 {
1201         LASSERT(early_req->rq_repbuf);
1202         LASSERT(early_req->rq_repdata);
1203         LASSERT(early_req->rq_repmsg);
1204
1205         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1206         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1207         ptlrpc_request_cache_free(early_req);
1208 }
1209
1210 /**************************************************
1211  * sec ID                                         *
1212  **************************************************/
1213
1214 /*
1215  * "fixed" sec (e.g. null) use sec_id < 0
1216  */
1217 static cfs_atomic_t sptlrpc_sec_id = CFS_ATOMIC_INIT(1);
1218
1219 int sptlrpc_get_next_secid(void)
1220 {
1221         return cfs_atomic_inc_return(&sptlrpc_sec_id);
1222 }
1223 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1224
1225 /**************************************************
1226  * client side high-level security APIs           *
1227  **************************************************/
1228
1229 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1230                                    int grace, int force)
1231 {
1232         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1233
1234         LASSERT(policy->sp_cops);
1235         LASSERT(policy->sp_cops->flush_ctx_cache);
1236
1237         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1238 }
1239
1240 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1241 {
1242         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1243
1244         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1245         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1246         LASSERT(policy->sp_cops->destroy_sec);
1247
1248         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1249
1250         policy->sp_cops->destroy_sec(sec);
1251         sptlrpc_policy_put(policy);
1252 }
1253
1254 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1255 {
1256         sec_cop_destroy_sec(sec);
1257 }
1258 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1259
1260 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1261 {
1262         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1263
1264         if (sec->ps_policy->sp_cops->kill_sec) {
1265                 sec->ps_policy->sp_cops->kill_sec(sec);
1266
1267                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1268         }
1269 }
1270
1271 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1272 {
1273         if (sec)
1274                 cfs_atomic_inc(&sec->ps_refcount);
1275
1276         return sec;
1277 }
1278 EXPORT_SYMBOL(sptlrpc_sec_get);
1279
1280 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1281 {
1282         if (sec) {
1283                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1284
1285                 if (cfs_atomic_dec_and_test(&sec->ps_refcount)) {
1286                         sptlrpc_gc_del_sec(sec);
1287                         sec_cop_destroy_sec(sec);
1288                 }
1289         }
1290 }
1291 EXPORT_SYMBOL(sptlrpc_sec_put);
1292
1293 /*
1294  * policy module is responsible for taking refrence of import
1295  */
1296 static
1297 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1298                                        struct ptlrpc_svc_ctx *svc_ctx,
1299                                        struct sptlrpc_flavor *sf,
1300                                        enum lustre_sec_part sp)
1301 {
1302         struct ptlrpc_sec_policy *policy;
1303         struct ptlrpc_sec        *sec;
1304         char                      str[32];
1305         ENTRY;
1306
1307         if (svc_ctx) {
1308                 LASSERT(imp->imp_dlm_fake == 1);
1309
1310                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1311                        imp->imp_obd->obd_type->typ_name,
1312                        imp->imp_obd->obd_name,
1313                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1314
1315                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1316                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1317         } else {
1318                 LASSERT(imp->imp_dlm_fake == 0);
1319
1320                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1321                        imp->imp_obd->obd_type->typ_name,
1322                        imp->imp_obd->obd_name,
1323                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1324
1325                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1326                 if (!policy) {
1327                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1328                         RETURN(NULL);
1329                 }
1330         }
1331
1332         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1333         if (sec) {
1334                 cfs_atomic_inc(&sec->ps_refcount);
1335
1336                 sec->ps_part = sp;
1337
1338                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1339                         sptlrpc_gc_add_sec(sec);
1340         } else {
1341                 sptlrpc_policy_put(policy);
1342         }
1343
1344         RETURN(sec);
1345 }
1346
1347 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1348 {
1349         struct ptlrpc_sec *sec;
1350
1351         spin_lock(&imp->imp_lock);
1352         sec = sptlrpc_sec_get(imp->imp_sec);
1353         spin_unlock(&imp->imp_lock);
1354
1355         return sec;
1356 }
1357 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1358
1359 static void sptlrpc_import_sec_install(struct obd_import *imp,
1360                                        struct ptlrpc_sec *sec)
1361 {
1362         struct ptlrpc_sec *old_sec;
1363
1364         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1365
1366         spin_lock(&imp->imp_lock);
1367         old_sec = imp->imp_sec;
1368         imp->imp_sec = sec;
1369         spin_unlock(&imp->imp_lock);
1370
1371         if (old_sec) {
1372                 sptlrpc_sec_kill(old_sec);
1373
1374                 /* balance the ref taken by this import */
1375                 sptlrpc_sec_put(old_sec);
1376         }
1377 }
1378
1379 static inline
1380 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1381 {
1382         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1383 }
1384
1385 static inline
1386 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1387 {
1388         *dst = *src;
1389 }
1390
1391 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1392                                              struct ptlrpc_sec *sec,
1393                                              struct sptlrpc_flavor *sf)
1394 {
1395         char    str1[32], str2[32];
1396
1397         if (sec->ps_flvr.sf_flags != sf->sf_flags)
1398                 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1399                        sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1400                                             str1, sizeof(str1)),
1401                        sptlrpc_secflags2str(sf->sf_flags,
1402                                             str2, sizeof(str2)));
1403
1404         spin_lock(&sec->ps_lock);
1405         flavor_copy(&sec->ps_flvr, sf);
1406         spin_unlock(&sec->ps_lock);
1407 }
1408
1409 /**
1410  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1411  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1412  *
1413  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1414  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1415  */
1416 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1417                              struct ptlrpc_svc_ctx *svc_ctx,
1418                              struct sptlrpc_flavor *flvr)
1419 {
1420         struct ptlrpc_connection   *conn;
1421         struct sptlrpc_flavor       sf;
1422         struct ptlrpc_sec          *sec, *newsec;
1423         enum lustre_sec_part        sp;
1424         char                        str[24];
1425         int                         rc = 0;
1426         ENTRY;
1427
1428         might_sleep();
1429
1430         if (imp == NULL)
1431                 RETURN(0);
1432
1433         conn = imp->imp_connection;
1434
1435         if (svc_ctx == NULL) {
1436                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1437                 /*
1438                  * normal import, determine flavor from rule set, except
1439                  * for mgc the flavor is predetermined.
1440                  */
1441                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1442                         sf = cliobd->cl_flvr_mgc;
1443                 else
1444                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1445                                                    cliobd->cl_sp_to,
1446                                                    &cliobd->cl_target_uuid,
1447                                                    conn->c_self, &sf);
1448
1449                 sp = imp->imp_obd->u.cli.cl_sp_me;
1450         } else {
1451                 /* reverse import, determine flavor from incoming reqeust */
1452                 sf = *flvr;
1453
1454                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1455                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1456                                       PTLRPC_SEC_FL_ROOTONLY;
1457
1458                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1459         }
1460
1461         sec = sptlrpc_import_sec_ref(imp);
1462         if (sec) {
1463                 char    str2[24];
1464
1465                 if (flavor_equal(&sf, &sec->ps_flvr))
1466                         GOTO(out, rc);
1467
1468                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1469                        imp->imp_obd->obd_name,
1470                        obd_uuid2str(&conn->c_remote_uuid),
1471                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1472                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1473
1474                 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1475                     SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1476                     SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1477                     SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1478                         sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1479                         GOTO(out, rc);
1480                 }
1481         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1482                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1483                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1484                        imp->imp_obd->obd_name,
1485                        obd_uuid2str(&conn->c_remote_uuid),
1486                        LNET_NIDNET(conn->c_self),
1487                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1488         }
1489
1490         mutex_lock(&imp->imp_sec_mutex);
1491
1492         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1493         if (newsec) {
1494                 sptlrpc_import_sec_install(imp, newsec);
1495         } else {
1496                 CERROR("import %s->%s: failed to create new sec\n",
1497                        imp->imp_obd->obd_name,
1498                        obd_uuid2str(&conn->c_remote_uuid));
1499                 rc = -EPERM;
1500         }
1501
1502         mutex_unlock(&imp->imp_sec_mutex);
1503 out:
1504         sptlrpc_sec_put(sec);
1505         RETURN(rc);
1506 }
1507
1508 void sptlrpc_import_sec_put(struct obd_import *imp)
1509 {
1510         if (imp->imp_sec) {
1511                 sptlrpc_sec_kill(imp->imp_sec);
1512
1513                 sptlrpc_sec_put(imp->imp_sec);
1514                 imp->imp_sec = NULL;
1515         }
1516 }
1517
1518 static void import_flush_ctx_common(struct obd_import *imp,
1519                                     uid_t uid, int grace, int force)
1520 {
1521         struct ptlrpc_sec *sec;
1522
1523         if (imp == NULL)
1524                 return;
1525
1526         sec = sptlrpc_import_sec_ref(imp);
1527         if (sec == NULL)
1528                 return;
1529
1530         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1531         sptlrpc_sec_put(sec);
1532 }
1533
1534 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1535 {
1536         /* it's important to use grace mode, see explain in
1537          * sptlrpc_req_refresh_ctx() */
1538         import_flush_ctx_common(imp, 0, 1, 1);
1539 }
1540
1541 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1542 {
1543         import_flush_ctx_common(imp, current_uid(), 1, 1);
1544 }
1545 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1546
1547 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1548 {
1549         import_flush_ctx_common(imp, -1, 1, 1);
1550 }
1551 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1552
1553 /**
1554  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1555  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1556  */
1557 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1558 {
1559         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1560         struct ptlrpc_sec_policy *policy;
1561         int rc;
1562
1563         LASSERT(ctx);
1564         LASSERT(ctx->cc_sec);
1565         LASSERT(ctx->cc_sec->ps_policy);
1566         LASSERT(req->rq_reqmsg == NULL);
1567         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1568
1569         policy = ctx->cc_sec->ps_policy;
1570         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1571         if (!rc) {
1572                 LASSERT(req->rq_reqmsg);
1573                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1574
1575                 /* zeroing preallocated buffer */
1576                 if (req->rq_pool)
1577                         memset(req->rq_reqmsg, 0, msgsize);
1578         }
1579
1580         return rc;
1581 }
1582
1583 /**
1584  * Used by ptlrpc client to free request buffer of \a req. After this
1585  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1586  */
1587 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1588 {
1589         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1590         struct ptlrpc_sec_policy *policy;
1591
1592         LASSERT(ctx);
1593         LASSERT(ctx->cc_sec);
1594         LASSERT(ctx->cc_sec->ps_policy);
1595         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1596
1597         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1598                 return;
1599
1600         policy = ctx->cc_sec->ps_policy;
1601         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1602         req->rq_reqmsg = NULL;
1603 }
1604
1605 /*
1606  * NOTE caller must guarantee the buffer size is enough for the enlargement
1607  */
1608 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1609                                   int segment, int newsize)
1610 {
1611         void   *src, *dst;
1612         int     oldsize, oldmsg_size, movesize;
1613
1614         LASSERT(segment < msg->lm_bufcount);
1615         LASSERT(msg->lm_buflens[segment] <= newsize);
1616
1617         if (msg->lm_buflens[segment] == newsize)
1618                 return;
1619
1620         /* nothing to do if we are enlarging the last segment */
1621         if (segment == msg->lm_bufcount - 1) {
1622                 msg->lm_buflens[segment] = newsize;
1623                 return;
1624         }
1625
1626         oldsize = msg->lm_buflens[segment];
1627
1628         src = lustre_msg_buf(msg, segment + 1, 0);
1629         msg->lm_buflens[segment] = newsize;
1630         dst = lustre_msg_buf(msg, segment + 1, 0);
1631         msg->lm_buflens[segment] = oldsize;
1632
1633         /* move from segment + 1 to end segment */
1634         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1635         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1636         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1637         LASSERT(movesize >= 0);
1638
1639         if (movesize)
1640                 memmove(dst, src, movesize);
1641
1642         /* note we don't clear the ares where old data live, not secret */
1643
1644         /* finally set new segment size */
1645         msg->lm_buflens[segment] = newsize;
1646 }
1647 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1648
1649 /**
1650  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1651  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1652  * preserved after the enlargement. this must be called after original request
1653  * buffer being allocated.
1654  *
1655  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1656  * so caller should refresh its local pointers if needed.
1657  */
1658 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1659                                int segment, int newsize)
1660 {
1661         struct ptlrpc_cli_ctx    *ctx = req->rq_cli_ctx;
1662         struct ptlrpc_sec_cops   *cops;
1663         struct lustre_msg        *msg = req->rq_reqmsg;
1664
1665         LASSERT(ctx);
1666         LASSERT(msg);
1667         LASSERT(msg->lm_bufcount > segment);
1668         LASSERT(msg->lm_buflens[segment] <= newsize);
1669
1670         if (msg->lm_buflens[segment] == newsize)
1671                 return 0;
1672
1673         cops = ctx->cc_sec->ps_policy->sp_cops;
1674         LASSERT(cops->enlarge_reqbuf);
1675         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1676 }
1677 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1678
1679 /**
1680  * Used by ptlrpc client to allocate reply buffer of \a req.
1681  *
1682  * \note After this, req->rq_repmsg is still not accessible.
1683  */
1684 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1685 {
1686         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1687         struct ptlrpc_sec_policy *policy;
1688         ENTRY;
1689
1690         LASSERT(ctx);
1691         LASSERT(ctx->cc_sec);
1692         LASSERT(ctx->cc_sec->ps_policy);
1693
1694         if (req->rq_repbuf)
1695                 RETURN(0);
1696
1697         policy = ctx->cc_sec->ps_policy;
1698         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1699 }
1700
1701 /**
1702  * Used by ptlrpc client to free reply buffer of \a req. After this
1703  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1704  */
1705 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1706 {
1707         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1708         struct ptlrpc_sec_policy *policy;
1709         ENTRY;
1710
1711         LASSERT(ctx);
1712         LASSERT(ctx->cc_sec);
1713         LASSERT(ctx->cc_sec->ps_policy);
1714         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1715
1716         if (req->rq_repbuf == NULL)
1717                 return;
1718         LASSERT(req->rq_repbuf_len);
1719
1720         policy = ctx->cc_sec->ps_policy;
1721         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1722         req->rq_repmsg = NULL;
1723         EXIT;
1724 }
1725
1726 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1727                                 struct ptlrpc_cli_ctx *ctx)
1728 {
1729         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1730
1731         if (!policy->sp_cops->install_rctx)
1732                 return 0;
1733         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1734 }
1735
1736 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1737                                 struct ptlrpc_svc_ctx *ctx)
1738 {
1739         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1740
1741         if (!policy->sp_sops->install_rctx)
1742                 return 0;
1743         return policy->sp_sops->install_rctx(imp, ctx);
1744 }
1745
1746 /****************************************
1747  * server side security                 *
1748  ****************************************/
1749
1750 static int flavor_allowed(struct sptlrpc_flavor *exp,
1751                           struct ptlrpc_request *req)
1752 {
1753         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1754
1755         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1756                 return 1;
1757
1758         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1759             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1760             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1761             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1762                 return 1;
1763
1764         return 0;
1765 }
1766
1767 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1768
1769 /**
1770  * Given an export \a exp, check whether the flavor of incoming \a req
1771  * is allowed by the export \a exp. Main logic is about taking care of
1772  * changing configurations. Return 0 means success.
1773  */
1774 int sptlrpc_target_export_check(struct obd_export *exp,
1775                                 struct ptlrpc_request *req)
1776 {
1777         struct sptlrpc_flavor   flavor;
1778
1779         if (exp == NULL)
1780                 return 0;
1781
1782         /* client side export has no imp_reverse, skip
1783          * FIXME maybe we should check flavor this as well??? */
1784         if (exp->exp_imp_reverse == NULL)
1785                 return 0;
1786
1787         /* don't care about ctx fini rpc */
1788         if (req->rq_ctx_fini)
1789                 return 0;
1790
1791         spin_lock(&exp->exp_lock);
1792
1793         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1794          * the first req with the new flavor, then treat it as current flavor,
1795          * adapt reverse sec according to it.
1796          * note the first rpc with new flavor might not be with root ctx, in
1797          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1798         if (unlikely(exp->exp_flvr_changed) &&
1799             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1800                 /* make the new flavor as "current", and old ones as
1801                  * about-to-expire */
1802                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1803                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1804                 flavor = exp->exp_flvr_old[1];
1805                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1806                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1807                 exp->exp_flvr_old[0] = exp->exp_flvr;
1808                 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1809                                           EXP_FLVR_UPDATE_EXPIRE;
1810                 exp->exp_flvr = flavor;
1811
1812                 /* flavor change finished */
1813                 exp->exp_flvr_changed = 0;
1814                 LASSERT(exp->exp_flvr_adapt == 1);
1815
1816                 /* if it's gss, we only interested in root ctx init */
1817                 if (req->rq_auth_gss &&
1818                     !(req->rq_ctx_init &&
1819                       (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1820                        req->rq_auth_usr_ost))) {
1821                         spin_unlock(&exp->exp_lock);
1822                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1823                                req->rq_auth_gss, req->rq_ctx_init,
1824                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1825                                req->rq_auth_usr_ost);
1826                         return 0;
1827                 }
1828
1829                 exp->exp_flvr_adapt = 0;
1830                 spin_unlock(&exp->exp_lock);
1831
1832                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1833                                                 req->rq_svc_ctx, &flavor);
1834         }
1835
1836         /* if it equals to the current flavor, we accept it, but need to
1837          * dealing with reverse sec/ctx */
1838         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1839                 /* most cases should return here, we only interested in
1840                  * gss root ctx init */
1841                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1842                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1843                      !req->rq_auth_usr_ost)) {
1844                         spin_unlock(&exp->exp_lock);
1845                         return 0;
1846                 }
1847
1848                 /* if flavor just changed, we should not proceed, just leave
1849                  * it and current flavor will be discovered and replaced
1850                  * shortly, and let _this_ rpc pass through */
1851                 if (exp->exp_flvr_changed) {
1852                         LASSERT(exp->exp_flvr_adapt);
1853                         spin_unlock(&exp->exp_lock);
1854                         return 0;
1855                 }
1856
1857                 if (exp->exp_flvr_adapt) {
1858                         exp->exp_flvr_adapt = 0;
1859                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1860                                exp, exp->exp_flvr.sf_rpc,
1861                                exp->exp_flvr_old[0].sf_rpc,
1862                                exp->exp_flvr_old[1].sf_rpc);
1863                         flavor = exp->exp_flvr;
1864                         spin_unlock(&exp->exp_lock);
1865
1866                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1867                                                         req->rq_svc_ctx,
1868                                                         &flavor);
1869                 } else {
1870                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1871                                "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1872                                exp->exp_flvr_old[0].sf_rpc,
1873                                exp->exp_flvr_old[1].sf_rpc);
1874                         spin_unlock(&exp->exp_lock);
1875
1876                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1877                                                            req->rq_svc_ctx);
1878                 }
1879         }
1880
1881         if (exp->exp_flvr_expire[0]) {
1882                 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1883                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1884                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1885                                        "middle one ("CFS_DURATION_T")\n", exp,
1886                                        exp->exp_flvr.sf_rpc,
1887                                        exp->exp_flvr_old[0].sf_rpc,
1888                                        exp->exp_flvr_old[1].sf_rpc,
1889                                        exp->exp_flvr_expire[0] -
1890                                                 cfs_time_current_sec());
1891                                 spin_unlock(&exp->exp_lock);
1892                                 return 0;
1893                         }
1894                 } else {
1895                         CDEBUG(D_SEC, "mark middle expired\n");
1896                         exp->exp_flvr_expire[0] = 0;
1897                 }
1898                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1899                        exp->exp_flvr.sf_rpc,
1900                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1901                        req->rq_flvr.sf_rpc);
1902         }
1903
1904         /* now it doesn't match the current flavor, the only chance we can
1905          * accept it is match the old flavors which is not expired. */
1906         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1907                 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1908                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1909                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1910                                        "oldest one ("CFS_DURATION_T")\n", exp,
1911                                        exp->exp_flvr.sf_rpc,
1912                                        exp->exp_flvr_old[0].sf_rpc,
1913                                        exp->exp_flvr_old[1].sf_rpc,
1914                                        exp->exp_flvr_expire[1] -
1915                                                 cfs_time_current_sec());
1916                                 spin_unlock(&exp->exp_lock);
1917                                 return 0;
1918                         }
1919                 } else {
1920                         CDEBUG(D_SEC, "mark oldest expired\n");
1921                         exp->exp_flvr_expire[1] = 0;
1922                 }
1923                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1924                        exp, exp->exp_flvr.sf_rpc,
1925                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1926                        req->rq_flvr.sf_rpc);
1927         } else {
1928                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1929                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1930                        exp->exp_flvr_old[1].sf_rpc);
1931         }
1932
1933         spin_unlock(&exp->exp_lock);
1934
1935         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1936               "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1937               exp, exp->exp_obd->obd_name,
1938               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1939               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1940               req->rq_flvr.sf_rpc,
1941               exp->exp_flvr.sf_rpc,
1942               exp->exp_flvr_old[0].sf_rpc,
1943               exp->exp_flvr_expire[0] ?
1944               (unsigned long) (exp->exp_flvr_expire[0] -
1945                                cfs_time_current_sec()) : 0,
1946               exp->exp_flvr_old[1].sf_rpc,
1947               exp->exp_flvr_expire[1] ?
1948               (unsigned long) (exp->exp_flvr_expire[1] -
1949                                cfs_time_current_sec()) : 0);
1950         return -EACCES;
1951 }
1952 EXPORT_SYMBOL(sptlrpc_target_export_check);
1953
1954 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1955                                       struct sptlrpc_rule_set *rset)
1956 {
1957         struct obd_export       *exp;
1958         struct sptlrpc_flavor    new_flvr;
1959
1960         LASSERT(obd);
1961
1962         spin_lock(&obd->obd_dev_lock);
1963
1964         cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1965                 if (exp->exp_connection == NULL)
1966                         continue;
1967
1968                 /* note if this export had just been updated flavor
1969                  * (exp_flvr_changed == 1), this will override the
1970                  * previous one. */
1971                 spin_lock(&exp->exp_lock);
1972                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1973                                              exp->exp_connection->c_peer.nid,
1974                                              &new_flvr);
1975                 if (exp->exp_flvr_changed ||
1976                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1977                         exp->exp_flvr_old[1] = new_flvr;
1978                         exp->exp_flvr_expire[1] = 0;
1979                         exp->exp_flvr_changed = 1;
1980                         exp->exp_flvr_adapt = 1;
1981
1982                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1983                                exp, sptlrpc_part2name(exp->exp_sp_peer),
1984                                exp->exp_flvr.sf_rpc,
1985                                exp->exp_flvr_old[1].sf_rpc);
1986                 }
1987                 spin_unlock(&exp->exp_lock);
1988         }
1989
1990         spin_unlock(&obd->obd_dev_lock);
1991 }
1992 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1993
1994 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1995 {
1996         /* peer's claim is unreliable unless gss is being used */
1997         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1998                 return svc_rc;
1999
2000         switch (req->rq_sp_from) {
2001         case LUSTRE_SP_CLI:
2002                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2003                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
2004                         svc_rc = SECSVC_DROP;
2005                 }
2006                 break;
2007         case LUSTRE_SP_MDT:
2008                 if (!req->rq_auth_usr_mdt) {
2009                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
2010                         svc_rc = SECSVC_DROP;
2011                 }
2012                 break;
2013         case LUSTRE_SP_OST:
2014                 if (!req->rq_auth_usr_ost) {
2015                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2016                         svc_rc = SECSVC_DROP;
2017                 }
2018                 break;
2019         case LUSTRE_SP_MGS:
2020         case LUSTRE_SP_MGC:
2021                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2022                     !req->rq_auth_usr_ost) {
2023                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2024                         svc_rc = SECSVC_DROP;
2025                 }
2026                 break;
2027         case LUSTRE_SP_ANY:
2028         default:
2029                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2030                 svc_rc = SECSVC_DROP;
2031         }
2032
2033         return svc_rc;
2034 }
2035
2036 /**
2037  * Used by ptlrpc server, to perform transformation upon request message of
2038  * incoming \a req. This must be the first thing to do with a incoming
2039  * request in ptlrpc layer.
2040  *
2041  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2042  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2043  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2044  * reply message has been prepared.
2045  * \retval SECSVC_DROP failed, this request should be dropped.
2046  */
2047 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2048 {
2049         struct ptlrpc_sec_policy *policy;
2050         struct lustre_msg        *msg = req->rq_reqbuf;
2051         int                       rc;
2052         ENTRY;
2053
2054         LASSERT(msg);
2055         LASSERT(req->rq_reqmsg == NULL);
2056         LASSERT(req->rq_repmsg == NULL);
2057         LASSERT(req->rq_svc_ctx == NULL);
2058
2059         req->rq_req_swab_mask = 0;
2060
2061         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2062         switch (rc) {
2063         case 1:
2064                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2065         case 0:
2066                 break;
2067         default:
2068                 CERROR("error unpacking request from %s x"LPU64"\n",
2069                        libcfs_id2str(req->rq_peer), req->rq_xid);
2070                 RETURN(SECSVC_DROP);
2071         }
2072
2073         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2074         req->rq_sp_from = LUSTRE_SP_ANY;
2075         req->rq_auth_uid = INVALID_UID;
2076         req->rq_auth_mapped_uid = INVALID_UID;
2077
2078         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2079         if (!policy) {
2080                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2081                 RETURN(SECSVC_DROP);
2082         }
2083
2084         LASSERT(policy->sp_sops->accept);
2085         rc = policy->sp_sops->accept(req);
2086         sptlrpc_policy_put(policy);
2087         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2088         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2089
2090         /*
2091          * if it's not null flavor (which means embedded packing msg),
2092          * reset the swab mask for the comming inner msg unpacking.
2093          */
2094         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2095                 req->rq_req_swab_mask = 0;
2096
2097         /* sanity check for the request source */
2098         rc = sptlrpc_svc_check_from(req, rc);
2099         RETURN(rc);
2100 }
2101
2102 /**
2103  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2104  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2105  * a buffer of \a msglen size.
2106  */
2107 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2108 {
2109         struct ptlrpc_sec_policy *policy;
2110         struct ptlrpc_reply_state *rs;
2111         int rc;
2112         ENTRY;
2113
2114         LASSERT(req->rq_svc_ctx);
2115         LASSERT(req->rq_svc_ctx->sc_policy);
2116
2117         policy = req->rq_svc_ctx->sc_policy;
2118         LASSERT(policy->sp_sops->alloc_rs);
2119
2120         rc = policy->sp_sops->alloc_rs(req, msglen);
2121         if (unlikely(rc == -ENOMEM)) {
2122                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2123                 if (svcpt->scp_service->srv_max_reply_size <
2124                    msglen + sizeof(struct ptlrpc_reply_state)) {
2125                         /* Just return failure if the size is too big */
2126                         CERROR("size of message is too big (%zd), %d allowed",
2127                                 msglen + sizeof(struct ptlrpc_reply_state),
2128                                 svcpt->scp_service->srv_max_reply_size);
2129                         RETURN(-ENOMEM);
2130                 }
2131
2132                 /* failed alloc, try emergency pool */
2133                 rs = lustre_get_emerg_rs(svcpt);
2134                 if (rs == NULL)
2135                         RETURN(-ENOMEM);
2136
2137                 req->rq_reply_state = rs;
2138                 rc = policy->sp_sops->alloc_rs(req, msglen);
2139                 if (rc) {
2140                         lustre_put_emerg_rs(rs);
2141                         req->rq_reply_state = NULL;
2142                 }
2143         }
2144
2145         LASSERT(rc != 0 ||
2146                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2147
2148         RETURN(rc);
2149 }
2150
2151 /**
2152  * Used by ptlrpc server, to perform transformation upon reply message.
2153  *
2154  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2155  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2156  */
2157 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2158 {
2159         struct ptlrpc_sec_policy *policy;
2160         int rc;
2161         ENTRY;
2162
2163         LASSERT(req->rq_svc_ctx);
2164         LASSERT(req->rq_svc_ctx->sc_policy);
2165
2166         policy = req->rq_svc_ctx->sc_policy;
2167         LASSERT(policy->sp_sops->authorize);
2168
2169         rc = policy->sp_sops->authorize(req);
2170         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2171
2172         RETURN(rc);
2173 }
2174
2175 /**
2176  * Used by ptlrpc server, to free reply_state.
2177  */
2178 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2179 {
2180         struct ptlrpc_sec_policy *policy;
2181         unsigned int prealloc;
2182         ENTRY;
2183
2184         LASSERT(rs->rs_svc_ctx);
2185         LASSERT(rs->rs_svc_ctx->sc_policy);
2186
2187         policy = rs->rs_svc_ctx->sc_policy;
2188         LASSERT(policy->sp_sops->free_rs);
2189
2190         prealloc = rs->rs_prealloc;
2191         policy->sp_sops->free_rs(rs);
2192
2193         if (prealloc)
2194                 lustre_put_emerg_rs(rs);
2195         EXIT;
2196 }
2197
2198 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2199 {
2200         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2201
2202         if (ctx != NULL)
2203                 cfs_atomic_inc(&ctx->sc_refcount);
2204 }
2205
2206 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2207 {
2208         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2209
2210         if (ctx == NULL)
2211                 return;
2212
2213         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2214         if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) {
2215                 if (ctx->sc_policy->sp_sops->free_ctx)
2216                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2217         }
2218         req->rq_svc_ctx = NULL;
2219 }
2220
2221 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2222 {
2223         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2224
2225         if (ctx == NULL)
2226                 return;
2227
2228         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2229         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2230                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2231 }
2232 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2233
2234 /****************************************
2235  * bulk security                        *
2236  ****************************************/
2237
2238 /**
2239  * Perform transformation upon bulk data pointed by \a desc. This is called
2240  * before transforming the request message.
2241  */
2242 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2243                           struct ptlrpc_bulk_desc *desc)
2244 {
2245         struct ptlrpc_cli_ctx *ctx;
2246
2247         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2248
2249         if (!req->rq_pack_bulk)
2250                 return 0;
2251
2252         ctx = req->rq_cli_ctx;
2253         if (ctx->cc_ops->wrap_bulk)
2254                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2255         return 0;
2256 }
2257 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2258
2259 /**
2260  * This is called after unwrap the reply message.
2261  * return nob of actual plain text size received, or error code.
2262  */
2263 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2264                                  struct ptlrpc_bulk_desc *desc,
2265                                  int nob)
2266 {
2267         struct ptlrpc_cli_ctx  *ctx;
2268         int                     rc;
2269
2270         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2271
2272         if (!req->rq_pack_bulk)
2273                 return desc->bd_nob_transferred;
2274
2275         ctx = req->rq_cli_ctx;
2276         if (ctx->cc_ops->unwrap_bulk) {
2277                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2278                 if (rc < 0)
2279                         return rc;
2280         }
2281         return desc->bd_nob_transferred;
2282 }
2283 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2284
2285 /**
2286  * This is called after unwrap the reply message.
2287  * return 0 for success or error code.
2288  */
2289 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2290                                   struct ptlrpc_bulk_desc *desc)
2291 {
2292         struct ptlrpc_cli_ctx  *ctx;
2293         int                     rc;
2294
2295         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2296
2297         if (!req->rq_pack_bulk)
2298                 return 0;
2299
2300         ctx = req->rq_cli_ctx;
2301         if (ctx->cc_ops->unwrap_bulk) {
2302                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2303                 if (rc < 0)
2304                         return rc;
2305         }
2306
2307         /*
2308          * if everything is going right, nob should equals to nob_transferred.
2309          * in case of privacy mode, nob_transferred needs to be adjusted.
2310          */
2311         if (desc->bd_nob != desc->bd_nob_transferred) {
2312                 CERROR("nob %d doesn't match transferred nob %d",
2313                        desc->bd_nob, desc->bd_nob_transferred);
2314                 return -EPROTO;
2315         }
2316
2317         return 0;
2318 }
2319 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2320
2321 #ifdef HAVE_SERVER_SUPPORT
2322 /**
2323  * Performe transformation upon outgoing bulk read.
2324  */
2325 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2326                           struct ptlrpc_bulk_desc *desc)
2327 {
2328         struct ptlrpc_svc_ctx *ctx;
2329
2330         LASSERT(req->rq_bulk_read);
2331
2332         if (!req->rq_pack_bulk)
2333                 return 0;
2334
2335         ctx = req->rq_svc_ctx;
2336         if (ctx->sc_policy->sp_sops->wrap_bulk)
2337                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2338
2339         return 0;
2340 }
2341 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2342
2343 /**
2344  * Performe transformation upon incoming bulk write.
2345  */
2346 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2347                             struct ptlrpc_bulk_desc *desc)
2348 {
2349         struct ptlrpc_svc_ctx *ctx;
2350         int                    rc;
2351
2352         LASSERT(req->rq_bulk_write);
2353
2354         /*
2355          * if it's in privacy mode, transferred should >= expected; otherwise
2356          * transferred should == expected.
2357          */
2358         if (desc->bd_nob_transferred < desc->bd_nob ||
2359             (desc->bd_nob_transferred > desc->bd_nob &&
2360              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2361              SPTLRPC_BULK_SVC_PRIV)) {
2362                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2363                           desc->bd_nob_transferred, desc->bd_nob);
2364                 return -ETIMEDOUT;
2365         }
2366
2367         if (!req->rq_pack_bulk)
2368                 return 0;
2369
2370         ctx = req->rq_svc_ctx;
2371         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2372                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2373                 if (rc)
2374                         CERROR("error unwrap bulk: %d\n", rc);
2375         }
2376
2377         /* return 0 to allow reply be sent */
2378         return 0;
2379 }
2380 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2381
2382 /**
2383  * Prepare buffers for incoming bulk write.
2384  */
2385 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2386                           struct ptlrpc_bulk_desc *desc)
2387 {
2388         struct ptlrpc_svc_ctx *ctx;
2389
2390         LASSERT(req->rq_bulk_write);
2391
2392         if (!req->rq_pack_bulk)
2393                 return 0;
2394
2395         ctx = req->rq_svc_ctx;
2396         if (ctx->sc_policy->sp_sops->prep_bulk)
2397                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2398
2399         return 0;
2400 }
2401 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2402
2403 #endif /* HAVE_SERVER_SUPPORT */
2404
2405 /****************************************
2406  * user descriptor helpers              *
2407  ****************************************/
2408
2409 int sptlrpc_current_user_desc_size(void)
2410 {
2411         int ngroups;
2412
2413 #ifdef __KERNEL__
2414         ngroups = current_ngroups;
2415
2416         if (ngroups > LUSTRE_MAX_GROUPS)
2417                 ngroups = LUSTRE_MAX_GROUPS;
2418 #else
2419         ngroups = 0;
2420 #endif
2421         return sptlrpc_user_desc_size(ngroups);
2422 }
2423 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2424
2425 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2426 {
2427         struct ptlrpc_user_desc *pud;
2428
2429         pud = lustre_msg_buf(msg, offset, 0);
2430
2431         pud->pud_uid = current_uid();
2432         pud->pud_gid = current_gid();
2433         pud->pud_fsuid = current_fsuid();
2434         pud->pud_fsgid = current_fsgid();
2435         pud->pud_cap = cfs_curproc_cap_pack();
2436         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2437
2438 #ifdef __KERNEL__
2439         task_lock(current);
2440         if (pud->pud_ngroups > current_ngroups)
2441                 pud->pud_ngroups = current_ngroups;
2442         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2443                pud->pud_ngroups * sizeof(__u32));
2444         task_unlock(current);
2445 #endif
2446
2447         return 0;
2448 }
2449 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2450
2451 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2452 {
2453         struct ptlrpc_user_desc *pud;
2454         int                      i;
2455
2456         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2457         if (!pud)
2458                 return -EINVAL;
2459
2460         if (swabbed) {
2461                 __swab32s(&pud->pud_uid);
2462                 __swab32s(&pud->pud_gid);
2463                 __swab32s(&pud->pud_fsuid);
2464                 __swab32s(&pud->pud_fsgid);
2465                 __swab32s(&pud->pud_cap);
2466                 __swab32s(&pud->pud_ngroups);
2467         }
2468
2469         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2470                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2471                 return -EINVAL;
2472         }
2473
2474         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2475             msg->lm_buflens[offset]) {
2476                 CERROR("%u groups are claimed but bufsize only %u\n",
2477                        pud->pud_ngroups, msg->lm_buflens[offset]);
2478                 return -EINVAL;
2479         }
2480
2481         if (swabbed) {
2482                 for (i = 0; i < pud->pud_ngroups; i++)
2483                         __swab32s(&pud->pud_groups[i]);
2484         }
2485
2486         return 0;
2487 }
2488 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2489
2490 /****************************************
2491  * misc helpers                         *
2492  ****************************************/
2493
2494 const char * sec2target_str(struct ptlrpc_sec *sec)
2495 {
2496         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2497                 return "*";
2498         if (sec_is_reverse(sec))
2499                 return "c";
2500         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2501 }
2502 EXPORT_SYMBOL(sec2target_str);
2503
2504 /*
2505  * return true if the bulk data is protected
2506  */
2507 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2508 {
2509         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2510         case SPTLRPC_BULK_SVC_INTG:
2511         case SPTLRPC_BULK_SVC_PRIV:
2512                 return 1;
2513         default:
2514                 return 0;
2515         }
2516 }
2517 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2518
2519 /****************************************
2520  * crypto API helper/alloc blkciper     *
2521  ****************************************/
2522
2523 /****************************************
2524  * initialize/finalize                  *
2525  ****************************************/
2526
2527 int sptlrpc_init(void)
2528 {
2529         int rc;
2530
2531         rwlock_init(&policy_lock);
2532
2533         rc = sptlrpc_gc_init();
2534         if (rc)
2535                 goto out;
2536
2537         rc = sptlrpc_conf_init();
2538         if (rc)
2539                 goto out_gc;
2540
2541         rc = sptlrpc_enc_pool_init();
2542         if (rc)
2543                 goto out_conf;
2544
2545         rc = sptlrpc_null_init();
2546         if (rc)
2547                 goto out_pool;
2548
2549         rc = sptlrpc_plain_init();
2550         if (rc)
2551                 goto out_null;
2552
2553         rc = sptlrpc_lproc_init();
2554         if (rc)
2555                 goto out_plain;
2556
2557         return 0;
2558
2559 out_plain:
2560         sptlrpc_plain_fini();
2561 out_null:
2562         sptlrpc_null_fini();
2563 out_pool:
2564         sptlrpc_enc_pool_fini();
2565 out_conf:
2566         sptlrpc_conf_fini();
2567 out_gc:
2568         sptlrpc_gc_fini();
2569 out:
2570         return rc;
2571 }
2572
2573 void sptlrpc_fini(void)
2574 {
2575         sptlrpc_lproc_fini();
2576         sptlrpc_plain_fini();
2577         sptlrpc_null_fini();
2578         sptlrpc_enc_pool_fini();
2579         sptlrpc_conf_fini();
2580         sptlrpc_gc_fini();
2581 }