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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/ptlrpc/sec.c
  33  *
  34  * Author: Eric Mei <ericm@clusterfs.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_SEC
  38
  39 #include <linux/user_namespace.h>
  40 #ifdef HAVE_UIDGID_HEADER
  41 # include <linux/uidgid.h>
  42 #endif
  43 #include <linux/crypto.h>
  44 #include <linux/key.h>
  45
  46 #include <libcfs/libcfs.h>
  47 #include <obd.h>
  48 #include <obd_class.h>
  49 #include <obd_support.h>
  50 #include <lustre_net.h>
  51 #include <lustre_import.h>
  52 #include <lustre_dlm.h>
  53 #include <lustre_sec.h>
  54
  55 #include "ptlrpc_internal.h"
  56
  57 static int send_sepol;
  58 module_param(send_sepol, int, 0644);
  59 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
  60
  61 /*
  62  * policy registers
  63  */
  64
  65 static rwlock_t policy_lock;
  66 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  67         NULL,
  68 };
  69
  70 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  71 {
  72         __u16 number = policy->sp_policy;
  73
  74         LASSERT(policy->sp_name);
  75         LASSERT(policy->sp_cops);
  76         LASSERT(policy->sp_sops);
  77
  78         if (number >= SPTLRPC_POLICY_MAX)
  79                 return -EINVAL;
  80
  81         write_lock(&policy_lock);
  82         if (unlikely(policies[number])) {
  83                 write_unlock(&policy_lock);
  84                 return -EALREADY;
  85         }
  86         policies[number] = policy;
  87         write_unlock(&policy_lock);
  88
  89         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  90         return 0;
  91 }
  92 EXPORT_SYMBOL(sptlrpc_register_policy);
  93
  94 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  95 {
  96         __u16 number = policy->sp_policy;
  97
  98         LASSERT(number < SPTLRPC_POLICY_MAX);
  99
 100         write_lock(&policy_lock);
 101         if (unlikely(policies[number] == NULL)) {
 102                 write_unlock(&policy_lock);
 103                 CERROR("%s: already unregistered\n", policy->sp_name);
 104                 return -EINVAL;
 105         }
 106
 107         LASSERT(policies[number] == policy);
 108         policies[number] = NULL;
 109         write_unlock(&policy_lock);
 110
 111         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 112         return 0;
 113 }
 114 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 115
 116 static
 117 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
 118 {
 119         static DEFINE_MUTEX(load_mutex);
 120         static atomic_t           loaded = ATOMIC_INIT(0);
 121         struct ptlrpc_sec_policy *policy;
 122         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 123         __u16                     flag = 0;
 124
 125         if (number >= SPTLRPC_POLICY_MAX)
 126                 return NULL;
 127
 128         while (1) {
 129                 read_lock(&policy_lock);
 130                 policy = policies[number];
 131                 if (policy && !try_module_get(policy->sp_owner))
 132                         policy = NULL;
 133                 if (policy == NULL)
 134                         flag = atomic_read(&loaded);
 135                 read_unlock(&policy_lock);
 136
 137                 if (policy != NULL || flag != 0 ||
 138                     number != SPTLRPC_POLICY_GSS)
 139                         break;
 140
 141                 /* try to load gss module, once */
 142                 mutex_lock(&load_mutex);
 143                 if (atomic_read(&loaded) == 0) {
 144                         if (request_module("ptlrpc_gss") == 0)
 145                                 CDEBUG(D_SEC,
 146                                        "module ptlrpc_gss loaded on demand\n");
 147                         else
 148                                 CERROR("Unable to load module ptlrpc_gss\n");
 149
 150                         atomic_set(&loaded, 1);
 151                 }
 152                 mutex_unlock(&load_mutex);
 153         }
 154
 155         return policy;
 156 }
 157
 158 __u32 sptlrpc_name2flavor_base(const char *name)
 159 {
 160         if (!strcmp(name, "null"))
 161                 return SPTLRPC_FLVR_NULL;
 162         if (!strcmp(name, "plain"))
 163                 return SPTLRPC_FLVR_PLAIN;
 164         if (!strcmp(name, "gssnull"))
 165                 return SPTLRPC_FLVR_GSSNULL;
 166         if (!strcmp(name, "krb5n"))
 167                 return SPTLRPC_FLVR_KRB5N;
 168         if (!strcmp(name, "krb5a"))
 169                 return SPTLRPC_FLVR_KRB5A;
 170         if (!strcmp(name, "krb5i"))
 171                 return SPTLRPC_FLVR_KRB5I;
 172         if (!strcmp(name, "krb5p"))
 173                 return SPTLRPC_FLVR_KRB5P;
 174         if (!strcmp(name, "skn"))
 175                 return SPTLRPC_FLVR_SKN;
 176         if (!strcmp(name, "ska"))
 177                 return SPTLRPC_FLVR_SKA;
 178         if (!strcmp(name, "ski"))
 179                 return SPTLRPC_FLVR_SKI;
 180         if (!strcmp(name, "skpi"))
 181                 return SPTLRPC_FLVR_SKPI;
 182
 183         return SPTLRPC_FLVR_INVALID;
 184 }
 185 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 186
 187 const char *sptlrpc_flavor2name_base(__u32 flvr)
 188 {
 189         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 190
 191         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 192                 return "null";
 193         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 194                 return "plain";
 195         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
 196                 return "gssnull";
 197         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 198                 return "krb5n";
 199         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 200                 return "krb5a";
 201         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 202                 return "krb5i";
 203         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 204                 return "krb5p";
 205         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
 206                 return "skn";
 207         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
 208                 return "ska";
 209         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
 210                 return "ski";
 211         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
 212                 return "skpi";
 213
 214         CERROR("invalid wire flavor 0x%x\n", flvr);
 215         return "invalid";
 216 }
 217 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 218
 219 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 220                                char *buf, int bufsize)
 221 {
 222         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 223                 snprintf(buf, bufsize, "hash:%s",
 224                         sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 225         else
 226                 snprintf(buf, bufsize, "%s",
 227                         sptlrpc_flavor2name_base(sf->sf_rpc));
 228
 229         buf[bufsize - 1] = '\0';
 230         return buf;
 231 }
 232 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 233
 234 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 235 {
 236         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 237
 238         /*
 239          * currently we don't support customized bulk specification for
 240          * flavors other than plain
 241          */
 242         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 243                 char bspec[16];
 244
 245                 bspec[0] = '-';
 246                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 247                 strncat(buf, bspec, bufsize);
 248         }
 249
 250         buf[bufsize - 1] = '\0';
 251         return buf;
 252 }
 253 EXPORT_SYMBOL(sptlrpc_flavor2name);
 254
 255 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 256 {
 257         buf[0] = '\0';
 258
 259         if (flags & PTLRPC_SEC_FL_REVERSE)
 260                 strlcat(buf, "reverse,", bufsize);
 261         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 262                 strlcat(buf, "rootonly,", bufsize);
 263         if (flags & PTLRPC_SEC_FL_UDESC)
 264                 strlcat(buf, "udesc,", bufsize);
 265         if (flags & PTLRPC_SEC_FL_BULK)
 266                 strlcat(buf, "bulk,", bufsize);
 267         if (buf[0] == '\0')
 268                 strlcat(buf, "-,", bufsize);
 269
 270         return buf;
 271 }
 272 EXPORT_SYMBOL(sptlrpc_secflags2str);
 273
 274 /*
 275  * client context APIs
 276  */
 277
 278 static
 279 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 280 {
 281         struct vfs_cred vcred;
 282         int create = 1, remove_dead = 1;
 283
 284         LASSERT(sec);
 285         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 286
 287         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 288                                      PTLRPC_SEC_FL_ROOTONLY)) {
 289                 vcred.vc_uid = 0;
 290                 vcred.vc_gid = 0;
 291                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 292                         create = 0;
 293                         remove_dead = 0;
 294                 }
 295         } else {
 296                 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
 297                 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
 298         }
 299
 300         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
 301                                                    remove_dead);
 302 }
 303
 304 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 305 {
 306         atomic_inc(&ctx->cc_refcount);
 307         return ctx;
 308 }
 309 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 310
 311 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 312 {
 313         struct ptlrpc_sec *sec = ctx->cc_sec;
 314
 315         LASSERT(sec);
 316         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 317
 318         if (!atomic_dec_and_test(&ctx->cc_refcount))
 319                 return;
 320
 321         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 322 }
 323 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 324
 325 /**
 326  * Expire the client context immediately.
 327  *
 328  * \pre Caller must hold at least 1 reference on the \a ctx.
 329  */
 330 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 331 {
 332         LASSERT(ctx->cc_ops->die);
 333         ctx->cc_ops->die(ctx, 0);
 334 }
 335 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 336
 337 /**
 338  * To wake up the threads who are waiting for this client context. Called
 339  * after some status change happened on \a ctx.
 340  */
 341 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 342 {
 343         struct ptlrpc_request *req, *next;
 344
 345         spin_lock(&ctx->cc_lock);
 346         list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 347                                      rq_ctx_chain) {
 348                 list_del_init(&req->rq_ctx_chain);
 349                 ptlrpc_client_wake_req(req);
 350         }
 351         spin_unlock(&ctx->cc_lock);
 352 }
 353 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 354
 355 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 356 {
 357         LASSERT(ctx->cc_ops);
 358
 359         if (ctx->cc_ops->display == NULL)
 360                 return 0;
 361
 362         return ctx->cc_ops->display(ctx, buf, bufsize);
 363 }
 364
 365 static int import_sec_check_expire(struct obd_import *imp)
 366 {
 367         int adapt = 0;
 368
 369         spin_lock(&imp->imp_lock);
 370         if (imp->imp_sec_expire &&
 371             imp->imp_sec_expire < ktime_get_real_seconds()) {
 372                 adapt = 1;
 373                 imp->imp_sec_expire = 0;
 374         }
 375         spin_unlock(&imp->imp_lock);
 376
 377         if (!adapt)
 378                 return 0;
 379
 380         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 381         return sptlrpc_import_sec_adapt(imp, NULL, NULL);
 382 }
 383
 384 /**
 385  * Get and validate the client side ptlrpc security facilities from
 386  * \a imp. There is a race condition on client reconnect when the import is
 387  * being destroyed while there are outstanding client bound requests. In
 388  * this case do not output any error messages if import secuity is not
 389  * found.
 390  *
 391  * \param[in] imp obd import associated with client
 392  * \param[out] sec client side ptlrpc security
 393  *
 394  * \retval 0 if security retrieved successfully
 395  * \retval -ve errno if there was a problem
 396  */
 397 static int import_sec_validate_get(struct obd_import *imp,
 398                                    struct ptlrpc_sec **sec)
 399 {
 400         int rc;
 401
 402         if (unlikely(imp->imp_sec_expire)) {
 403                 rc = import_sec_check_expire(imp);
 404                 if (rc)
 405                         return rc;
 406         }
 407
 408         *sec = sptlrpc_import_sec_ref(imp);
 409         if (*sec == NULL) {
 410                 CERROR("import %p (%s) with no sec\n",
 411                         imp, ptlrpc_import_state_name(imp->imp_state));
 412                 return -EACCES;
 413         }
 414
 415         if (unlikely((*sec)->ps_dying)) {
 416                 CERROR("attempt to use dying sec %p\n", sec);
 417                 sptlrpc_sec_put(*sec);
 418                 return -EACCES;
 419         }
 420
 421         return 0;
 422 }
 423
 424 /**
 425  * Given a \a req, find or allocate an appropriate context for it.
 426  * \pre req->rq_cli_ctx == NULL.
 427  *
 428  * \retval 0 succeed, and req->rq_cli_ctx is set.
 429  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 430  */
 431 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 432 {
 433         struct obd_import *imp = req->rq_import;
 434         struct ptlrpc_sec *sec;
 435         int rc;
 436
 437         ENTRY;
 438
 439         LASSERT(!req->rq_cli_ctx);
 440         LASSERT(imp);
 441
 442         rc = import_sec_validate_get(imp, &sec);
 443         if (rc)
 444                 RETURN(rc);
 445
 446         req->rq_cli_ctx = get_my_ctx(sec);
 447
 448         sptlrpc_sec_put(sec);
 449
 450         if (!req->rq_cli_ctx) {
 451                 CERROR("req %p: fail to get context\n", req);
 452                 RETURN(-ECONNREFUSED);
 453         }
 454
 455         RETURN(0);
 456 }
 457
 458 /**
 459  * Drop the context for \a req.
 460  * \pre req->rq_cli_ctx != NULL.
 461  * \post req->rq_cli_ctx == NULL.
 462  *
 463  * If \a sync == 0, this function should return quickly without sleep;
 464  * otherwise it might trigger and wait for the whole process of sending
 465  * an context-destroying rpc to server.
 466  */
 467 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 468 {
 469         ENTRY;
 470
 471         LASSERT(req);
 472         LASSERT(req->rq_cli_ctx);
 473
 474         /*
 475          * request might be asked to release earlier while still
 476          * in the context waiting list.
 477          */
 478         if (!list_empty(&req->rq_ctx_chain)) {
 479                 spin_lock(&req->rq_cli_ctx->cc_lock);
 480                 list_del_init(&req->rq_ctx_chain);
 481                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 482         }
 483
 484         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 485         req->rq_cli_ctx = NULL;
 486         EXIT;
 487 }
 488
 489 static
 490 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 491                            struct ptlrpc_cli_ctx *oldctx,
 492                            struct ptlrpc_cli_ctx *newctx)
 493 {
 494         struct sptlrpc_flavor   old_flvr;
 495         char *reqmsg = NULL; /* to workaround old gcc */
 496         int reqmsg_size;
 497         int rc = 0;
 498
 499         LASSERT(req->rq_reqmsg);
 500         LASSERT(req->rq_reqlen);
 501         LASSERT(req->rq_replen);
 502
 503         CDEBUG(D_SEC,
 504                "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
 505                req, oldctx, oldctx->cc_vcred.vc_uid,
 506                sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
 507                sec2target_str(newctx->cc_sec), oldctx->cc_sec,
 508                oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
 509                newctx->cc_sec->ps_policy->sp_name);
 510
 511         /* save flavor */
 512         old_flvr = req->rq_flvr;
 513
 514         /* save request message */
 515         reqmsg_size = req->rq_reqlen;
 516         if (reqmsg_size != 0) {
 517                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 518                 if (reqmsg == NULL)
 519                         return -ENOMEM;
 520                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 521         }
 522
 523         /* release old req/rep buf */
 524         req->rq_cli_ctx = oldctx;
 525         sptlrpc_cli_free_reqbuf(req);
 526         sptlrpc_cli_free_repbuf(req);
 527         req->rq_cli_ctx = newctx;
 528
 529         /* recalculate the flavor */
 530         sptlrpc_req_set_flavor(req, 0);
 531
 532         /*
 533          * alloc new request buffer
 534          * we don't need to alloc reply buffer here, leave it to the
 535          * rest procedure of ptlrpc
 536          */
 537         if (reqmsg_size != 0) {
 538                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 539                 if (!rc) {
 540                         LASSERT(req->rq_reqmsg);
 541                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 542                 } else {
 543                         CWARN("failed to alloc reqbuf: %d\n", rc);
 544                         req->rq_flvr = old_flvr;
 545                 }
 546
 547                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 548         }
 549         return rc;
 550 }
 551
 552 /**
 553  * If current context of \a req is dead somehow, e.g. we just switched flavor
 554  * thus marked original contexts dead, we'll find a new context for it. if
 555  * no switch is needed, \a req will end up with the same context.
 556  *
 557  * \note a request must have a context, to keep other parts of code happy.
 558  * In any case of failure during the switching, we must restore the old one.
 559  */
 560 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 561 {
 562         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 563         struct ptlrpc_cli_ctx *newctx;
 564         int rc;
 565
 566         ENTRY;
 567
 568         LASSERT(oldctx);
 569
 570         sptlrpc_cli_ctx_get(oldctx);
 571         sptlrpc_req_put_ctx(req, 0);
 572
 573         rc = sptlrpc_req_get_ctx(req);
 574         if (unlikely(rc)) {
 575                 LASSERT(!req->rq_cli_ctx);
 576
 577                 /* restore old ctx */
 578                 req->rq_cli_ctx = oldctx;
 579                 RETURN(rc);
 580         }
 581
 582         newctx = req->rq_cli_ctx;
 583         LASSERT(newctx);
 584
 585         if (unlikely(newctx == oldctx &&
 586                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 587                 /*
 588                  * still get the old dead ctx, usually means system too busy
 589                  */
 590                 CDEBUG(D_SEC,
 591                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 592                        newctx, newctx->cc_flags);
 593
 594                 set_current_state(TASK_INTERRUPTIBLE);
 595                 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
 596         } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
 597                             == 0)) {
 598                 /*
 599                  * new ctx not up to date yet
 600                  */
 601                 CDEBUG(D_SEC,
 602                        "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
 603                        newctx, newctx->cc_flags);
 604         } else {
 605                 /*
 606                  * it's possible newctx == oldctx if we're switching
 607                  * subflavor with the same sec.
 608                  */
 609                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 610                 if (rc) {
 611                         /* restore old ctx */
 612                         sptlrpc_req_put_ctx(req, 0);
 613                         req->rq_cli_ctx = oldctx;
 614                         RETURN(rc);
 615                 }
 616
 617                 LASSERT(req->rq_cli_ctx == newctx);
 618         }
 619
 620         sptlrpc_cli_ctx_put(oldctx, 1);
 621         RETURN(0);
 622 }
 623 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 624
 625 static
 626 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 627 {
 628         if (cli_ctx_is_refreshed(ctx))
 629                 return 1;
 630         return 0;
 631 }
 632
 633 static
 634 int ctx_refresh_timeout(void *data)
 635 {
 636         struct ptlrpc_request *req = data;
 637         int rc;
 638
 639         /* conn_cnt is needed in expire_one_request */
 640         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 641
 642         rc = ptlrpc_expire_one_request(req, 1);
 643         /*
 644          * if we started recovery, we should mark this ctx dead; otherwise
 645          * in case of lgssd died nobody would retire this ctx, following
 646          * connecting will still find the same ctx thus cause deadlock.
 647          * there's an assumption that expire time of the request should be
 648          * later than the context refresh expire time.
 649          */
 650         if (rc == 0)
 651                 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
 652         return rc;
 653 }
 654
 655 static
 656 void ctx_refresh_interrupt(void *data)
 657 {
 658         struct ptlrpc_request *req = data;
 659
 660         spin_lock(&req->rq_lock);
 661         req->rq_intr = 1;
 662         spin_unlock(&req->rq_lock);
 663 }
 664
 665 static
 666 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 667 {
 668         spin_lock(&ctx->cc_lock);
 669         if (!list_empty(&req->rq_ctx_chain))
 670                 list_del_init(&req->rq_ctx_chain);
 671         spin_unlock(&ctx->cc_lock);
 672 }
 673
 674 /**
 675  * To refresh the context of \req, if it's not up-to-date.
 676  * \param timeout
 677  * - < 0: don't wait
 678  * - = 0: wait until success or fatal error occur
 679  * - > 0: timeout value (in seconds)
 680  *
 681  * The status of the context could be subject to be changed by other threads
 682  * at any time. We allow this race, but once we return with 0, the caller will
 683  * suppose it's uptodated and keep using it until the owning rpc is done.
 684  *
 685  * \retval 0 only if the context is uptodated.
 686  * \retval -ev error number.
 687  */
 688 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 689 {
 690         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 691         struct ptlrpc_sec *sec;
 692         struct l_wait_info lwi;
 693         int rc;
 694
 695         ENTRY;
 696
 697         LASSERT(ctx);
 698
 699         if (req->rq_ctx_init || req->rq_ctx_fini)
 700                 RETURN(0);
 701
 702         /*
 703          * during the process a request's context might change type even
 704          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 705          * everything
 706          */
 707 again:
 708         rc = import_sec_validate_get(req->rq_import, &sec);
 709         if (rc)
 710                 RETURN(rc);
 711
 712         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 713                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 714                        req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 715                 req_off_ctx_list(req, ctx);
 716                 sptlrpc_req_replace_dead_ctx(req);
 717                 ctx = req->rq_cli_ctx;
 718         }
 719         sptlrpc_sec_put(sec);
 720
 721         if (cli_ctx_is_eternal(ctx))
 722                 RETURN(0);
 723
 724         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 725                 LASSERT(ctx->cc_ops->refresh);
 726                 ctx->cc_ops->refresh(ctx);
 727         }
 728         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 729
 730         LASSERT(ctx->cc_ops->validate);
 731         if (ctx->cc_ops->validate(ctx) == 0) {
 732                 req_off_ctx_list(req, ctx);
 733                 RETURN(0);
 734         }
 735
 736         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 737                 spin_lock(&req->rq_lock);
 738                 req->rq_err = 1;
 739                 spin_unlock(&req->rq_lock);
 740                 req_off_ctx_list(req, ctx);
 741                 RETURN(-EPERM);
 742         }
 743
 744         /*
 745          * There's a subtle issue for resending RPCs, suppose following
 746          * situation:
 747          *  1. the request was sent to server.
 748          *  2. recovery was kicked start, after finished the request was
 749          *     marked as resent.
 750          *  3. resend the request.
 751          *  4. old reply from server received, we accept and verify the reply.
 752          *     this has to be success, otherwise the error will be aware
 753          *     by application.
 754          *  5. new reply from server received, dropped by LNet.
 755          *
 756          * Note the xid of old & new request is the same. We can't simply
 757          * change xid for the resent request because the server replies on
 758          * it for reply reconstruction.
 759          *
 760          * Commonly the original context should be uptodate because we
 761          * have an expiry nice time; server will keep its context because
 762          * we at least hold a ref of old context which prevent context
 763          * from destroying RPC being sent. So server still can accept the
 764          * request and finish the RPC. But if that's not the case:
 765          *  1. If server side context has been trimmed, a NO_CONTEXT will
 766          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 767          *     context by force.
 768          *  2. Current context never be refreshed, then we are fine: we
 769          *     never really send request with old context before.
 770          */
 771         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 772             unlikely(req->rq_reqmsg) &&
 773             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 774                 req_off_ctx_list(req, ctx);
 775                 RETURN(0);
 776         }
 777
 778         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 779                 req_off_ctx_list(req, ctx);
 780                 /*
 781                  * don't switch ctx if import was deactivated
 782                  */
 783                 if (req->rq_import->imp_deactive) {
 784                         spin_lock(&req->rq_lock);
 785                         req->rq_err = 1;
 786                         spin_unlock(&req->rq_lock);
 787                         RETURN(-EINTR);
 788                 }
 789
 790                 rc = sptlrpc_req_replace_dead_ctx(req);
 791                 if (rc) {
 792                         LASSERT(ctx == req->rq_cli_ctx);
 793                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 794                                req, ctx, rc);
 795                         spin_lock(&req->rq_lock);
 796                         req->rq_err = 1;
 797                         spin_unlock(&req->rq_lock);
 798                         RETURN(rc);
 799                 }
 800
 801                 ctx = req->rq_cli_ctx;
 802                 goto again;
 803         }
 804
 805         /*
 806          * Now we're sure this context is during upcall, add myself into
 807          * waiting list
 808          */
 809         spin_lock(&ctx->cc_lock);
 810         if (list_empty(&req->rq_ctx_chain))
 811                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 812         spin_unlock(&ctx->cc_lock);
 813
 814         if (timeout < 0)
 815                 RETURN(-EWOULDBLOCK);
 816
 817         /* Clear any flags that may be present from previous sends */
 818         LASSERT(req->rq_receiving_reply == 0);
 819         spin_lock(&req->rq_lock);
 820         req->rq_err = 0;
 821         req->rq_timedout = 0;
 822         req->rq_resend = 0;
 823         req->rq_restart = 0;
 824         spin_unlock(&req->rq_lock);
 825
 826         lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
 827                                ctx_refresh_timeout,
 828                                ctx_refresh_interrupt, req);
 829         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 830
 831         /*
 832          * following cases could lead us here:
 833          * - successfully refreshed;
 834          * - interrupted;
 835          * - timedout, and we don't want recover from the failure;
 836          * - timedout, and waked up upon recovery finished;
 837          * - someone else mark this ctx dead by force;
 838          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 839          *   e.g. ptlrpc_abort_inflight();
 840          */
 841         if (!cli_ctx_is_refreshed(ctx)) {
 842                 /* timed out or interruptted */
 843                 req_off_ctx_list(req, ctx);
 844
 845                 LASSERT(rc != 0);
 846                 RETURN(rc);
 847         }
 848
 849         goto again;
 850 }
 851
 852 /**
 853  * Initialize flavor settings for \a req, according to \a opcode.
 854  *
 855  * \note this could be called in two situations:
 856  * - new request from ptlrpc_pre_req(), with proper @opcode
 857  * - old request which changed ctx in the middle, with @opcode == 0
 858  */
 859 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 860 {
 861         struct ptlrpc_sec *sec;
 862
 863         LASSERT(req->rq_import);
 864         LASSERT(req->rq_cli_ctx);
 865         LASSERT(req->rq_cli_ctx->cc_sec);
 866         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 867
 868         /* special security flags according to opcode */
 869         switch (opcode) {
 870         case OST_READ:
 871         case MDS_READPAGE:
 872         case MGS_CONFIG_READ:
 873         case OBD_IDX_READ:
 874                 req->rq_bulk_read = 1;
 875                 break;
 876         case OST_WRITE:
 877         case MDS_WRITEPAGE:
 878                 req->rq_bulk_write = 1;
 879                 break;
 880         case SEC_CTX_INIT:
 881                 req->rq_ctx_init = 1;
 882                 break;
 883         case SEC_CTX_FINI:
 884                 req->rq_ctx_fini = 1;
 885                 break;
 886         case 0:
 887                 /* init/fini rpc won't be resend, so can't be here */
 888                 LASSERT(req->rq_ctx_init == 0);
 889                 LASSERT(req->rq_ctx_fini == 0);
 890
 891                 /* cleanup flags, which should be recalculated */
 892                 req->rq_pack_udesc = 0;
 893                 req->rq_pack_bulk = 0;
 894                 break;
 895         }
 896
 897         sec = req->rq_cli_ctx->cc_sec;
 898
 899         spin_lock(&sec->ps_lock);
 900         req->rq_flvr = sec->ps_flvr;
 901         spin_unlock(&sec->ps_lock);
 902
 903         /*
 904          * force SVC_NULL for context initiation rpc, SVC_INTG for context
 905          * destruction rpc
 906          */
 907         if (unlikely(req->rq_ctx_init))
 908                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 909         else if (unlikely(req->rq_ctx_fini))
 910                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 911
 912         /* user descriptor flag, null security can't do it anyway */
 913         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 914             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 915                 req->rq_pack_udesc = 1;
 916
 917         /* bulk security flag */
 918         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 919             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 920                 req->rq_pack_bulk = 1;
 921 }
 922
 923 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 924 {
 925         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 926                 return;
 927
 928         LASSERT(req->rq_clrbuf);
 929         if (req->rq_pool || !req->rq_reqbuf)
 930                 return;
 931
 932         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 933         req->rq_reqbuf = NULL;
 934         req->rq_reqbuf_len = 0;
 935 }
 936
 937 /**
 938  * Given an import \a imp, check whether current user has a valid context
 939  * or not. We may create a new context and try to refresh it, and try
 940  * repeatedly try in case of non-fatal errors. Return 0 means success.
 941  */
 942 int sptlrpc_import_check_ctx(struct obd_import *imp)
 943 {
 944         struct ptlrpc_sec     *sec;
 945         struct ptlrpc_cli_ctx *ctx;
 946         struct ptlrpc_request *req = NULL;
 947         int rc;
 948
 949         ENTRY;
 950
 951         might_sleep();
 952
 953         sec = sptlrpc_import_sec_ref(imp);
 954         ctx = get_my_ctx(sec);
 955         sptlrpc_sec_put(sec);
 956
 957         if (!ctx)
 958                 RETURN(-ENOMEM);
 959
 960         if (cli_ctx_is_eternal(ctx) ||
 961             ctx->cc_ops->validate(ctx) == 0) {
 962                 sptlrpc_cli_ctx_put(ctx, 1);
 963                 RETURN(0);
 964         }
 965
 966         if (cli_ctx_is_error(ctx)) {
 967                 sptlrpc_cli_ctx_put(ctx, 1);
 968                 RETURN(-EACCES);
 969         }
 970
 971         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 972         if (!req)
 973                 RETURN(-ENOMEM);
 974
 975         ptlrpc_cli_req_init(req);
 976         atomic_set(&req->rq_refcount, 10000);
 977
 978         req->rq_import = imp;
 979         req->rq_flvr = sec->ps_flvr;
 980         req->rq_cli_ctx = ctx;
 981
 982         rc = sptlrpc_req_refresh_ctx(req, 0);
 983         LASSERT(list_empty(&req->rq_ctx_chain));
 984         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 985         ptlrpc_request_cache_free(req);
 986
 987         RETURN(rc);
 988 }
 989
 990 /**
 991  * Used by ptlrpc client, to perform the pre-defined security transformation
 992  * upon the request message of \a req. After this function called,
 993  * req->rq_reqmsg is still accessible as clear text.
 994  */
 995 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 996 {
 997         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 998         int rc = 0;
 999
1000         ENTRY;
1001
1002         LASSERT(ctx);
1003         LASSERT(ctx->cc_sec);
1004         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1005
1006         /*
1007          * we wrap bulk request here because now we can be sure
1008          * the context is uptodate.
1009          */
1010         if (req->rq_bulk) {
1011                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1012                 if (rc)
1013                         RETURN(rc);
1014         }
1015
1016         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1017         case SPTLRPC_SVC_NULL:
1018         case SPTLRPC_SVC_AUTH:
1019         case SPTLRPC_SVC_INTG:
1020                 LASSERT(ctx->cc_ops->sign);
1021                 rc = ctx->cc_ops->sign(ctx, req);
1022                 break;
1023         case SPTLRPC_SVC_PRIV:
1024                 LASSERT(ctx->cc_ops->seal);
1025                 rc = ctx->cc_ops->seal(ctx, req);
1026                 break;
1027         default:
1028                 LBUG();
1029         }
1030
1031         if (rc == 0) {
1032                 LASSERT(req->rq_reqdata_len);
1033                 LASSERT(req->rq_reqdata_len % 8 == 0);
1034                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1035         }
1036
1037         RETURN(rc);
1038 }
1039
1040 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1041 {
1042         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1043         int rc;
1044
1045         ENTRY;
1046
1047         LASSERT(ctx);
1048         LASSERT(ctx->cc_sec);
1049         LASSERT(req->rq_repbuf);
1050         LASSERT(req->rq_repdata);
1051         LASSERT(req->rq_repmsg == NULL);
1052
1053         req->rq_rep_swab_mask = 0;
1054
1055         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1056         switch (rc) {
1057         case 1:
1058                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1059         case 0:
1060                 break;
1061         default:
1062                 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1063                 RETURN(-EPROTO);
1064         }
1065
1066         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1067                 CERROR("replied data length %d too small\n",
1068                        req->rq_repdata_len);
1069                 RETURN(-EPROTO);
1070         }
1071
1072         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1073             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1074                 CERROR("reply policy %u doesn't match request policy %u\n",
1075                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1076                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1077                 RETURN(-EPROTO);
1078         }
1079
1080         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1081         case SPTLRPC_SVC_NULL:
1082         case SPTLRPC_SVC_AUTH:
1083         case SPTLRPC_SVC_INTG:
1084                 LASSERT(ctx->cc_ops->verify);
1085                 rc = ctx->cc_ops->verify(ctx, req);
1086                 break;
1087         case SPTLRPC_SVC_PRIV:
1088                 LASSERT(ctx->cc_ops->unseal);
1089                 rc = ctx->cc_ops->unseal(ctx, req);
1090                 break;
1091         default:
1092                 LBUG();
1093         }
1094         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1095
1096         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1097             !req->rq_ctx_init)
1098                 req->rq_rep_swab_mask = 0;
1099         RETURN(rc);
1100 }
1101
1102 /**
1103  * Used by ptlrpc client, to perform security transformation upon the reply
1104  * message of \a req. After return successfully, req->rq_repmsg points to
1105  * the reply message in clear text.
1106  *
1107  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1108  * going to change.
1109  */
1110 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1111 {
1112         LASSERT(req->rq_repbuf);
1113         LASSERT(req->rq_repdata == NULL);
1114         LASSERT(req->rq_repmsg == NULL);
1115         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1116
1117         if (req->rq_reply_off == 0 &&
1118             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1119                 CERROR("real reply with offset 0\n");
1120                 return -EPROTO;
1121         }
1122
1123         if (req->rq_reply_off % 8 != 0) {
1124                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1125                 return -EPROTO;
1126         }
1127
1128         req->rq_repdata = (struct lustre_msg *)
1129                                 (req->rq_repbuf + req->rq_reply_off);
1130         req->rq_repdata_len = req->rq_nob_received;
1131
1132         return do_cli_unwrap_reply(req);
1133 }
1134
1135 /**
1136  * Used by ptlrpc client, to perform security transformation upon the early
1137  * reply message of \a req. We expect the rq_reply_off is 0, and
1138  * rq_nob_received is the early reply size.
1139  *
1140  * Because the receive buffer might be still posted, the reply data might be
1141  * changed at any time, no matter we're holding rq_lock or not. For this reason
1142  * we allocate a separate ptlrpc_request and reply buffer for early reply
1143  * processing.
1144  *
1145  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1146  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1147  * \a *req_ret to release it.
1148  * \retval -ev error number, and \a req_ret will not be set.
1149  */
1150 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1151                                    struct ptlrpc_request **req_ret)
1152 {
1153         struct ptlrpc_request *early_req;
1154         char *early_buf;
1155         int early_bufsz, early_size;
1156         int rc;
1157
1158         ENTRY;
1159
1160         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1161         if (early_req == NULL)
1162                 RETURN(-ENOMEM);
1163
1164         ptlrpc_cli_req_init(early_req);
1165
1166         early_size = req->rq_nob_received;
1167         early_bufsz = size_roundup_power2(early_size);
1168         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1169         if (early_buf == NULL)
1170                 GOTO(err_req, rc = -ENOMEM);
1171
1172         /* sanity checkings and copy data out, do it inside spinlock */
1173         spin_lock(&req->rq_lock);
1174
1175         if (req->rq_replied) {
1176                 spin_unlock(&req->rq_lock);
1177                 GOTO(err_buf, rc = -EALREADY);
1178         }
1179
1180         LASSERT(req->rq_repbuf);
1181         LASSERT(req->rq_repdata == NULL);
1182         LASSERT(req->rq_repmsg == NULL);
1183
1184         if (req->rq_reply_off != 0) {
1185                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1186                 spin_unlock(&req->rq_lock);
1187                 GOTO(err_buf, rc = -EPROTO);
1188         }
1189
1190         if (req->rq_nob_received != early_size) {
1191                 /* even another early arrived the size should be the same */
1192                 CERROR("data size has changed from %u to %u\n",
1193                        early_size, req->rq_nob_received);
1194                 spin_unlock(&req->rq_lock);
1195                 GOTO(err_buf, rc = -EINVAL);
1196         }
1197
1198         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1199                 CERROR("early reply length %d too small\n",
1200                        req->rq_nob_received);
1201                 spin_unlock(&req->rq_lock);
1202                 GOTO(err_buf, rc = -EALREADY);
1203         }
1204
1205         memcpy(early_buf, req->rq_repbuf, early_size);
1206         spin_unlock(&req->rq_lock);
1207
1208         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1209         early_req->rq_flvr = req->rq_flvr;
1210         early_req->rq_repbuf = early_buf;
1211         early_req->rq_repbuf_len = early_bufsz;
1212         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1213         early_req->rq_repdata_len = early_size;
1214         early_req->rq_early = 1;
1215         early_req->rq_reqmsg = req->rq_reqmsg;
1216
1217         rc = do_cli_unwrap_reply(early_req);
1218         if (rc) {
1219                 DEBUG_REQ(D_ADAPTTO, early_req,
1220                           "error %d unwrap early reply", rc);
1221                 GOTO(err_ctx, rc);
1222         }
1223
1224         LASSERT(early_req->rq_repmsg);
1225         *req_ret = early_req;
1226         RETURN(0);
1227
1228 err_ctx:
1229         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1230 err_buf:
1231         OBD_FREE_LARGE(early_buf, early_bufsz);
1232 err_req:
1233         ptlrpc_request_cache_free(early_req);
1234         RETURN(rc);
1235 }
1236
1237 /**
1238  * Used by ptlrpc client, to release a processed early reply \a early_req.
1239  *
1240  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1241  */
1242 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1243 {
1244         LASSERT(early_req->rq_repbuf);
1245         LASSERT(early_req->rq_repdata);
1246         LASSERT(early_req->rq_repmsg);
1247
1248         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1249         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1250         ptlrpc_request_cache_free(early_req);
1251 }
1252
1253 /**************************************************
1254  * sec ID                                         *
1255  **************************************************/
1256
1257 /*
1258  * "fixed" sec (e.g. null) use sec_id < 0
1259  */
1260 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1261
1262 int sptlrpc_get_next_secid(void)
1263 {
1264         return atomic_inc_return(&sptlrpc_sec_id);
1265 }
1266 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1267
1268 /*
1269  * client side high-level security APIs
1270  */
1271
1272 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1273                                    int grace, int force)
1274 {
1275         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1276
1277         LASSERT(policy->sp_cops);
1278         LASSERT(policy->sp_cops->flush_ctx_cache);
1279
1280         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1281 }
1282
1283 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1284 {
1285         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1286
1287         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1288         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1289         LASSERT(policy->sp_cops->destroy_sec);
1290
1291         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1292
1293         policy->sp_cops->destroy_sec(sec);
1294         sptlrpc_policy_put(policy);
1295 }
1296
1297 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1298 {
1299         sec_cop_destroy_sec(sec);
1300 }
1301 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1302
1303 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1304 {
1305         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1306
1307         if (sec->ps_policy->sp_cops->kill_sec) {
1308                 sec->ps_policy->sp_cops->kill_sec(sec);
1309
1310                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1311         }
1312 }
1313
1314 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1315 {
1316         if (sec)
1317                 atomic_inc(&sec->ps_refcount);
1318
1319         return sec;
1320 }
1321 EXPORT_SYMBOL(sptlrpc_sec_get);
1322
1323 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1324 {
1325         if (sec) {
1326                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1327
1328                 if (atomic_dec_and_test(&sec->ps_refcount)) {
1329                         sptlrpc_gc_del_sec(sec);
1330                         sec_cop_destroy_sec(sec);
1331                 }
1332         }
1333 }
1334 EXPORT_SYMBOL(sptlrpc_sec_put);
1335
1336 /*
1337  * policy module is responsible for taking refrence of import
1338  */
1339 static
1340 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1341                                        struct ptlrpc_svc_ctx *svc_ctx,
1342                                        struct sptlrpc_flavor *sf,
1343                                        enum lustre_sec_part sp)
1344 {
1345         struct ptlrpc_sec_policy *policy;
1346         struct ptlrpc_sec *sec;
1347         char str[32];
1348
1349         ENTRY;
1350
1351         if (svc_ctx) {
1352                 LASSERT(imp->imp_dlm_fake == 1);
1353
1354                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1355                        imp->imp_obd->obd_type->typ_name,
1356                        imp->imp_obd->obd_name,
1357                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1358
1359                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1360                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1361         } else {
1362                 LASSERT(imp->imp_dlm_fake == 0);
1363
1364                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1365                        imp->imp_obd->obd_type->typ_name,
1366                        imp->imp_obd->obd_name,
1367                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1368
1369                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1370                 if (!policy) {
1371                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1372                         RETURN(NULL);
1373                 }
1374         }
1375
1376         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1377         if (sec) {
1378                 atomic_inc(&sec->ps_refcount);
1379
1380                 sec->ps_part = sp;
1381
1382                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1383                         sptlrpc_gc_add_sec(sec);
1384         } else {
1385                 sptlrpc_policy_put(policy);
1386         }
1387
1388         RETURN(sec);
1389 }
1390
1391 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1392 {
1393         struct ptlrpc_sec *sec;
1394
1395         spin_lock(&imp->imp_lock);
1396         sec = sptlrpc_sec_get(imp->imp_sec);
1397         spin_unlock(&imp->imp_lock);
1398
1399         return sec;
1400 }
1401 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1402
1403 static void sptlrpc_import_sec_install(struct obd_import *imp,
1404                                        struct ptlrpc_sec *sec)
1405 {
1406         struct ptlrpc_sec *old_sec;
1407
1408         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1409
1410         spin_lock(&imp->imp_lock);
1411         old_sec = imp->imp_sec;
1412         imp->imp_sec = sec;
1413         spin_unlock(&imp->imp_lock);
1414
1415         if (old_sec) {
1416                 sptlrpc_sec_kill(old_sec);
1417
1418                 /* balance the ref taken by this import */
1419                 sptlrpc_sec_put(old_sec);
1420         }
1421 }
1422
1423 static inline
1424 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1425 {
1426         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1427 }
1428
1429 static inline
1430 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1431 {
1432         *dst = *src;
1433 }
1434
1435 /**
1436  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1437  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1438  *
1439  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1440  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1441  */
1442 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1443                              struct ptlrpc_svc_ctx *svc_ctx,
1444                              struct sptlrpc_flavor *flvr)
1445 {
1446         struct ptlrpc_connection *conn;
1447         struct sptlrpc_flavor sf;
1448         struct ptlrpc_sec *sec, *newsec;
1449         enum lustre_sec_part sp;
1450         char str[24];
1451         int rc = 0;
1452
1453         ENTRY;
1454
1455         might_sleep();
1456
1457         if (imp == NULL)
1458                 RETURN(0);
1459
1460         conn = imp->imp_connection;
1461
1462         if (svc_ctx == NULL) {
1463                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1464                 /*
1465                  * normal import, determine flavor from rule set, except
1466                  * for mgc the flavor is predetermined.
1467                  */
1468                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1469                         sf = cliobd->cl_flvr_mgc;
1470                 else
1471                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1472                                                    cliobd->cl_sp_to,
1473                                                    &cliobd->cl_target_uuid,
1474                                                    conn->c_self, &sf);
1475
1476                 sp = imp->imp_obd->u.cli.cl_sp_me;
1477         } else {
1478                 /* reverse import, determine flavor from incoming reqeust */
1479                 sf = *flvr;
1480
1481                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1482                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1483                                       PTLRPC_SEC_FL_ROOTONLY;
1484
1485                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1486         }
1487
1488         sec = sptlrpc_import_sec_ref(imp);
1489         if (sec) {
1490                 char str2[24];
1491
1492                 if (flavor_equal(&sf, &sec->ps_flvr))
1493                         GOTO(out, rc);
1494
1495                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1496                        imp->imp_obd->obd_name,
1497                        obd_uuid2str(&conn->c_remote_uuid),
1498                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1499                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1500         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1501                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1502                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1503                        imp->imp_obd->obd_name,
1504                        obd_uuid2str(&conn->c_remote_uuid),
1505                        LNET_NIDNET(conn->c_self),
1506                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1507         }
1508
1509         mutex_lock(&imp->imp_sec_mutex);
1510
1511         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1512         if (newsec) {
1513                 sptlrpc_import_sec_install(imp, newsec);
1514         } else {
1515                 CERROR("import %s->%s: failed to create new sec\n",
1516                        imp->imp_obd->obd_name,
1517                        obd_uuid2str(&conn->c_remote_uuid));
1518                 rc = -EPERM;
1519         }
1520
1521         mutex_unlock(&imp->imp_sec_mutex);
1522 out:
1523         sptlrpc_sec_put(sec);
1524         RETURN(rc);
1525 }
1526
1527 void sptlrpc_import_sec_put(struct obd_import *imp)
1528 {
1529         if (imp->imp_sec) {
1530                 sptlrpc_sec_kill(imp->imp_sec);
1531
1532                 sptlrpc_sec_put(imp->imp_sec);
1533                 imp->imp_sec = NULL;
1534         }
1535 }
1536
1537 static void import_flush_ctx_common(struct obd_import *imp,
1538                                     uid_t uid, int grace, int force)
1539 {
1540         struct ptlrpc_sec *sec;
1541
1542         if (imp == NULL)
1543                 return;
1544
1545         sec = sptlrpc_import_sec_ref(imp);
1546         if (sec == NULL)
1547                 return;
1548
1549         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1550         sptlrpc_sec_put(sec);
1551 }
1552
1553 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1554 {
1555         /*
1556          * it's important to use grace mode, see explain in
1557          * sptlrpc_req_refresh_ctx()
1558          */
1559         import_flush_ctx_common(imp, 0, 1, 1);
1560 }
1561
1562 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1563 {
1564         import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1565                                 1, 1);
1566 }
1567 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1568
1569 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1570 {
1571         import_flush_ctx_common(imp, -1, 1, 1);
1572 }
1573 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1574
1575 /**
1576  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1577  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1578  */
1579 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1580 {
1581         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1582         struct ptlrpc_sec_policy *policy;
1583         int rc;
1584
1585         LASSERT(ctx);
1586         LASSERT(ctx->cc_sec);
1587         LASSERT(ctx->cc_sec->ps_policy);
1588         LASSERT(req->rq_reqmsg == NULL);
1589         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1590
1591         policy = ctx->cc_sec->ps_policy;
1592         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1593         if (!rc) {
1594                 LASSERT(req->rq_reqmsg);
1595                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1596
1597                 /* zeroing preallocated buffer */
1598                 if (req->rq_pool)
1599                         memset(req->rq_reqmsg, 0, msgsize);
1600         }
1601
1602         return rc;
1603 }
1604
1605 /**
1606  * Used by ptlrpc client to free request buffer of \a req. After this
1607  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1608  */
1609 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1610 {
1611         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1612         struct ptlrpc_sec_policy *policy;
1613
1614         LASSERT(ctx);
1615         LASSERT(ctx->cc_sec);
1616         LASSERT(ctx->cc_sec->ps_policy);
1617         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1618
1619         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1620                 return;
1621
1622         policy = ctx->cc_sec->ps_policy;
1623         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1624         req->rq_reqmsg = NULL;
1625 }
1626
1627 /*
1628  * NOTE caller must guarantee the buffer size is enough for the enlargement
1629  */
1630 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1631                                   int segment, int newsize)
1632 {
1633         void *src, *dst;
1634         int oldsize, oldmsg_size, movesize;
1635
1636         LASSERT(segment < msg->lm_bufcount);
1637         LASSERT(msg->lm_buflens[segment] <= newsize);
1638
1639         if (msg->lm_buflens[segment] == newsize)
1640                 return;
1641
1642         /* nothing to do if we are enlarging the last segment */
1643         if (segment == msg->lm_bufcount - 1) {
1644                 msg->lm_buflens[segment] = newsize;
1645                 return;
1646         }
1647
1648         oldsize = msg->lm_buflens[segment];
1649
1650         src = lustre_msg_buf(msg, segment + 1, 0);
1651         msg->lm_buflens[segment] = newsize;
1652         dst = lustre_msg_buf(msg, segment + 1, 0);
1653         msg->lm_buflens[segment] = oldsize;
1654
1655         /* move from segment + 1 to end segment */
1656         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1657         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1658         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1659         LASSERT(movesize >= 0);
1660
1661         if (movesize)
1662                 memmove(dst, src, movesize);
1663
1664         /* note we don't clear the ares where old data live, not secret */
1665
1666         /* finally set new segment size */
1667         msg->lm_buflens[segment] = newsize;
1668 }
1669 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1670
1671 /**
1672  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1673  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1674  * preserved after the enlargement. this must be called after original request
1675  * buffer being allocated.
1676  *
1677  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1678  * so caller should refresh its local pointers if needed.
1679  */
1680 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1681                                const struct req_msg_field *field,
1682                                int newsize)
1683 {
1684         struct req_capsule *pill = &req->rq_pill;
1685         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1686         struct ptlrpc_sec_cops *cops;
1687         struct lustre_msg *msg = req->rq_reqmsg;
1688         int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1689
1690         LASSERT(ctx);
1691         LASSERT(msg);
1692         LASSERT(msg->lm_bufcount > segment);
1693         LASSERT(msg->lm_buflens[segment] <= newsize);
1694
1695         if (msg->lm_buflens[segment] == newsize)
1696                 return 0;
1697
1698         cops = ctx->cc_sec->ps_policy->sp_cops;
1699         LASSERT(cops->enlarge_reqbuf);
1700         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1701 }
1702 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1703
1704 /**
1705  * Used by ptlrpc client to allocate reply buffer of \a req.
1706  *
1707  * \note After this, req->rq_repmsg is still not accessible.
1708  */
1709 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1710 {
1711         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1712         struct ptlrpc_sec_policy *policy;
1713
1714         ENTRY;
1715
1716         LASSERT(ctx);
1717         LASSERT(ctx->cc_sec);
1718         LASSERT(ctx->cc_sec->ps_policy);
1719
1720         if (req->rq_repbuf)
1721                 RETURN(0);
1722
1723         policy = ctx->cc_sec->ps_policy;
1724         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1725 }
1726
1727 /**
1728  * Used by ptlrpc client to free reply buffer of \a req. After this
1729  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1730  */
1731 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1732 {
1733         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1734         struct ptlrpc_sec_policy *policy;
1735
1736         ENTRY;
1737
1738         LASSERT(ctx);
1739         LASSERT(ctx->cc_sec);
1740         LASSERT(ctx->cc_sec->ps_policy);
1741         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1742
1743         if (req->rq_repbuf == NULL)
1744                 return;
1745         LASSERT(req->rq_repbuf_len);
1746
1747         policy = ctx->cc_sec->ps_policy;
1748         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1749         req->rq_repmsg = NULL;
1750         EXIT;
1751 }
1752 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1753
1754 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1755                                 struct ptlrpc_cli_ctx *ctx)
1756 {
1757         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1758
1759         if (!policy->sp_cops->install_rctx)
1760                 return 0;
1761         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1762 }
1763
1764 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1765                                 struct ptlrpc_svc_ctx *ctx)
1766 {
1767         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1768
1769         if (!policy->sp_sops->install_rctx)
1770                 return 0;
1771         return policy->sp_sops->install_rctx(imp, ctx);
1772 }
1773
1774 /* Get SELinux policy info from userspace */
1775 static int sepol_helper(struct obd_import *imp)
1776 {
1777         char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1778         char *argv[] = {
1779                 [0] = "/usr/sbin/l_getsepol",
1780                 [1] = "-o",
1781                 [2] = NULL,         /* obd type */
1782                 [3] = "-n",
1783                 [4] = NULL,         /* obd name */
1784                 [5] = "-t",
1785                 [6] = mtime_str,    /* policy mtime */
1786                 [7] = "-m",
1787                 [8] = mode_str,     /* enforcing mode */
1788                 [9] = NULL
1789         };
1790         char *envp[] = {
1791                 [0] = "HOME=/",
1792                 [1] = "PATH=/sbin:/usr/sbin",
1793                 [2] = NULL
1794         };
1795         signed short ret;
1796         int rc = 0;
1797
1798         if (imp == NULL || imp->imp_obd == NULL ||
1799             imp->imp_obd->obd_type == NULL) {
1800                 rc = -EINVAL;
1801         } else {
1802                 argv[2] = imp->imp_obd->obd_type->typ_name;
1803                 argv[4] = imp->imp_obd->obd_name;
1804                 spin_lock(&imp->imp_sec->ps_lock);
1805                 if (imp->imp_sec->ps_sepol_mtime == 0 &&
1806                     imp->imp_sec->ps_sepol[0] == '\0') {
1807                         /* ps_sepol has not been initialized */
1808                         argv[5] = NULL;
1809                         argv[7] = NULL;
1810                 } else {
1811                         snprintf(mtime_str, sizeof(mtime_str), "%lu",
1812                                  imp->imp_sec->ps_sepol_mtime);
1813                         mode_str[0] = imp->imp_sec->ps_sepol[0];
1814                 }
1815                 spin_unlock(&imp->imp_sec->ps_lock);
1816                 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1817                 rc = ret>>8;
1818         }
1819
1820         return rc;
1821 }
1822
1823 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1824 {
1825         ktime_t checknext;
1826
1827         if (send_sepol == 0 || !selinux_is_enabled())
1828                 return 0;
1829
1830         if (send_sepol == -1)
1831                 /* send_sepol == -1 means fetch sepol status every time */
1832                 return 1;
1833
1834         spin_lock(&imp_sec->ps_lock);
1835         checknext = imp_sec->ps_sepol_checknext;
1836         spin_unlock(&imp_sec->ps_lock);
1837
1838         /* next check is too far in time, please update */
1839         if (ktime_after(checknext,
1840                         ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1841                 goto setnext;
1842
1843         if (ktime_before(ktime_get(), checknext))
1844                 /* too early to fetch sepol status */
1845                 return 0;
1846
1847 setnext:
1848         /* define new sepol_checknext time */
1849         spin_lock(&imp_sec->ps_lock);
1850         imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1851                                                 ktime_set(send_sepol, 0));
1852         spin_unlock(&imp_sec->ps_lock);
1853
1854         return 1;
1855 }
1856
1857 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1858 {
1859         struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1860         int rc = 0;
1861
1862         ENTRY;
1863
1864         (req->rq_sepol)[0] = '\0';
1865
1866 #ifndef HAVE_SELINUX
1867         if (unlikely(send_sepol != 0))
1868                 CDEBUG(D_SEC,
1869                        "Client cannot report SELinux status, it was not built against libselinux.\n");
1870         RETURN(0);
1871 #endif
1872
1873         if (send_sepol == 0 || !selinux_is_enabled())
1874                 RETURN(0);
1875
1876         if (imp_sec == NULL)
1877                 RETURN(-EINVAL);
1878
1879         /* Retrieve SELinux status info */
1880         if (sptlrpc_sepol_needs_check(imp_sec))
1881                 rc = sepol_helper(req->rq_import);
1882         if (likely(rc == 0)) {
1883                 spin_lock(&imp_sec->ps_lock);
1884                 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1885                        sizeof(req->rq_sepol));
1886                 spin_unlock(&imp_sec->ps_lock);
1887         }
1888
1889         RETURN(rc);
1890 }
1891 EXPORT_SYMBOL(sptlrpc_get_sepol);
1892
1893 /*
1894  * server side security
1895  */
1896
1897 static int flavor_allowed(struct sptlrpc_flavor *exp,
1898                           struct ptlrpc_request *req)
1899 {
1900         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1901
1902         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1903                 return 1;
1904
1905         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1906             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1907             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1908             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1909                 return 1;
1910
1911         return 0;
1912 }
1913
1914 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1915
1916 /**
1917  * Given an export \a exp, check whether the flavor of incoming \a req
1918  * is allowed by the export \a exp. Main logic is about taking care of
1919  * changing configurations. Return 0 means success.
1920  */
1921 int sptlrpc_target_export_check(struct obd_export *exp,
1922                                 struct ptlrpc_request *req)
1923 {
1924         struct sptlrpc_flavor   flavor;
1925
1926         if (exp == NULL)
1927                 return 0;
1928
1929         /*
1930          * client side export has no imp_reverse, skip
1931          * FIXME maybe we should check flavor this as well???
1932          */
1933         if (exp->exp_imp_reverse == NULL)
1934                 return 0;
1935
1936         /* don't care about ctx fini rpc */
1937         if (req->rq_ctx_fini)
1938                 return 0;
1939
1940         spin_lock(&exp->exp_lock);
1941
1942         /*
1943          * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1944          * the first req with the new flavor, then treat it as current flavor,
1945          * adapt reverse sec according to it.
1946          * note the first rpc with new flavor might not be with root ctx, in
1947          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1948          */
1949         if (unlikely(exp->exp_flvr_changed) &&
1950             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1951                 /*
1952                  * make the new flavor as "current", and old ones as
1953                  * about-to-expire
1954                  */
1955                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1956                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1957                 flavor = exp->exp_flvr_old[1];
1958                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1959                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1960                 exp->exp_flvr_old[0] = exp->exp_flvr;
1961                 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1962                                           EXP_FLVR_UPDATE_EXPIRE;
1963                 exp->exp_flvr = flavor;
1964
1965                 /* flavor change finished */
1966                 exp->exp_flvr_changed = 0;
1967                 LASSERT(exp->exp_flvr_adapt == 1);
1968
1969                 /* if it's gss, we only interested in root ctx init */
1970                 if (req->rq_auth_gss &&
1971                     !(req->rq_ctx_init &&
1972                     (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1973                     req->rq_auth_usr_ost))) {
1974                         spin_unlock(&exp->exp_lock);
1975                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1976                                req->rq_auth_gss, req->rq_ctx_init,
1977                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1978                                req->rq_auth_usr_ost);
1979                         return 0;
1980                 }
1981
1982                 exp->exp_flvr_adapt = 0;
1983                 spin_unlock(&exp->exp_lock);
1984
1985                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1986                                                 req->rq_svc_ctx, &flavor);
1987         }
1988
1989         /*
1990          * if it equals to the current flavor, we accept it, but need to
1991          * dealing with reverse sec/ctx
1992          */
1993         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1994                 /*
1995                  * most cases should return here, we only interested in
1996                  * gss root ctx init
1997                  */
1998                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1999                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2000                      !req->rq_auth_usr_ost)) {
2001                         spin_unlock(&exp->exp_lock);
2002                         return 0;
2003                 }
2004
2005                 /*
2006                  * if flavor just changed, we should not proceed, just leave
2007                  * it and current flavor will be discovered and replaced
2008                  * shortly, and let _this_ rpc pass through
2009                  */
2010                 if (exp->exp_flvr_changed) {
2011                         LASSERT(exp->exp_flvr_adapt);
2012                         spin_unlock(&exp->exp_lock);
2013                         return 0;
2014                 }
2015
2016                 if (exp->exp_flvr_adapt) {
2017                         exp->exp_flvr_adapt = 0;
2018                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2019                                exp, exp->exp_flvr.sf_rpc,
2020                                exp->exp_flvr_old[0].sf_rpc,
2021                                exp->exp_flvr_old[1].sf_rpc);
2022                         flavor = exp->exp_flvr;
2023                         spin_unlock(&exp->exp_lock);
2024
2025                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2026                                                         req->rq_svc_ctx,
2027                                                         &flavor);
2028                 } else {
2029                         CDEBUG(D_SEC,
2030                                "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2031                                exp, exp->exp_flvr.sf_rpc,
2032                                exp->exp_flvr_old[0].sf_rpc,
2033                                exp->exp_flvr_old[1].sf_rpc);
2034                         spin_unlock(&exp->exp_lock);
2035
2036                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2037                                                            req->rq_svc_ctx);
2038                 }
2039         }
2040
2041         if (exp->exp_flvr_expire[0]) {
2042                 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2043                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2044                                 CDEBUG(D_SEC,
2045                                        "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2046                                        exp, exp->exp_flvr.sf_rpc,
2047                                        exp->exp_flvr_old[0].sf_rpc,
2048                                        exp->exp_flvr_old[1].sf_rpc,
2049                                        (s64)(exp->exp_flvr_expire[0] -
2050                                              ktime_get_real_seconds()));
2051                                 spin_unlock(&exp->exp_lock);
2052                                 return 0;
2053                         }
2054                 } else {
2055                         CDEBUG(D_SEC, "mark middle expired\n");
2056                         exp->exp_flvr_expire[0] = 0;
2057                 }
2058                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2059                        exp->exp_flvr.sf_rpc,
2060                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2061                        req->rq_flvr.sf_rpc);
2062         }
2063
2064         /*
2065          * now it doesn't match the current flavor, the only chance we can
2066          * accept it is match the old flavors which is not expired.
2067          */
2068         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2069                 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2070                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2071                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2072                                        exp,
2073                                        exp->exp_flvr.sf_rpc,
2074                                        exp->exp_flvr_old[0].sf_rpc,
2075                                        exp->exp_flvr_old[1].sf_rpc,
2076                                        (s64)(exp->exp_flvr_expire[1] -
2077                                        ktime_get_real_seconds()));
2078                                 spin_unlock(&exp->exp_lock);
2079                                 return 0;
2080                         }
2081                 } else {
2082                         CDEBUG(D_SEC, "mark oldest expired\n");
2083                         exp->exp_flvr_expire[1] = 0;
2084                 }
2085                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2086                        exp, exp->exp_flvr.sf_rpc,
2087                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2088                        req->rq_flvr.sf_rpc);
2089         } else {
2090                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2091                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2092                        exp->exp_flvr_old[1].sf_rpc);
2093         }
2094
2095         spin_unlock(&exp->exp_lock);
2096
2097         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2098               exp, exp->exp_obd->obd_name,
2099               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2100               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2101               req->rq_flvr.sf_rpc,
2102               exp->exp_flvr.sf_rpc,
2103               exp->exp_flvr_old[0].sf_rpc,
2104               exp->exp_flvr_expire[0] ?
2105               (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2106               exp->exp_flvr_old[1].sf_rpc,
2107               exp->exp_flvr_expire[1] ?
2108               (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2109         return -EACCES;
2110 }
2111 EXPORT_SYMBOL(sptlrpc_target_export_check);
2112
2113 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2114                                       struct sptlrpc_rule_set *rset)
2115 {
2116         struct obd_export *exp;
2117         struct sptlrpc_flavor new_flvr;
2118
2119         LASSERT(obd);
2120
2121         spin_lock(&obd->obd_dev_lock);
2122
2123         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2124                 if (exp->exp_connection == NULL)
2125                         continue;
2126
2127                 /*
2128                  * note if this export had just been updated flavor
2129                  * (exp_flvr_changed == 1), this will override the
2130                  * previous one.
2131                  */
2132                 spin_lock(&exp->exp_lock);
2133                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2134                                              exp->exp_connection->c_peer.nid,
2135                                              &new_flvr);
2136                 if (exp->exp_flvr_changed ||
2137                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2138                         exp->exp_flvr_old[1] = new_flvr;
2139                         exp->exp_flvr_expire[1] = 0;
2140                         exp->exp_flvr_changed = 1;
2141                         exp->exp_flvr_adapt = 1;
2142
2143                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2144                                exp, sptlrpc_part2name(exp->exp_sp_peer),
2145                                exp->exp_flvr.sf_rpc,
2146                                exp->exp_flvr_old[1].sf_rpc);
2147                 }
2148                 spin_unlock(&exp->exp_lock);
2149         }
2150
2151         spin_unlock(&obd->obd_dev_lock);
2152 }
2153 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2154
2155 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2156 {
2157         /* peer's claim is unreliable unless gss is being used */
2158         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2159                 return svc_rc;
2160
2161         switch (req->rq_sp_from) {
2162         case LUSTRE_SP_CLI:
2163                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2164                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
2165                         svc_rc = SECSVC_DROP;
2166                 }
2167                 break;
2168         case LUSTRE_SP_MDT:
2169                 if (!req->rq_auth_usr_mdt) {
2170                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
2171                         svc_rc = SECSVC_DROP;
2172                 }
2173                 break;
2174         case LUSTRE_SP_OST:
2175                 if (!req->rq_auth_usr_ost) {
2176                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2177                         svc_rc = SECSVC_DROP;
2178                 }
2179                 break;
2180         case LUSTRE_SP_MGS:
2181         case LUSTRE_SP_MGC:
2182                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2183                     !req->rq_auth_usr_ost) {
2184                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2185                         svc_rc = SECSVC_DROP;
2186                 }
2187                 break;
2188         case LUSTRE_SP_ANY:
2189         default:
2190                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2191                 svc_rc = SECSVC_DROP;
2192         }
2193
2194         return svc_rc;
2195 }
2196
2197 /**
2198  * Used by ptlrpc server, to perform transformation upon request message of
2199  * incoming \a req. This must be the first thing to do with an incoming
2200  * request in ptlrpc layer.
2201  *
2202  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2203  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2204  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2205  * reply message has been prepared.
2206  * \retval SECSVC_DROP failed, this request should be dropped.
2207  */
2208 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2209 {
2210         struct ptlrpc_sec_policy *policy;
2211         struct lustre_msg *msg = req->rq_reqbuf;
2212         int rc;
2213
2214         ENTRY;
2215
2216         LASSERT(msg);
2217         LASSERT(req->rq_reqmsg == NULL);
2218         LASSERT(req->rq_repmsg == NULL);
2219         LASSERT(req->rq_svc_ctx == NULL);
2220
2221         req->rq_req_swab_mask = 0;
2222
2223         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2224         switch (rc) {
2225         case 1:
2226                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2227         case 0:
2228                 break;
2229         default:
2230                 CERROR("error unpacking request from %s x%llu\n",
2231                        libcfs_id2str(req->rq_peer), req->rq_xid);
2232                 RETURN(SECSVC_DROP);
2233         }
2234
2235         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2236         req->rq_sp_from = LUSTRE_SP_ANY;
2237         req->rq_auth_uid = -1; /* set to INVALID_UID */
2238         req->rq_auth_mapped_uid = -1;
2239
2240         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2241         if (!policy) {
2242                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2243                 RETURN(SECSVC_DROP);
2244         }
2245
2246         LASSERT(policy->sp_sops->accept);
2247         rc = policy->sp_sops->accept(req);
2248         sptlrpc_policy_put(policy);
2249         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2250         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2251
2252         /*
2253          * if it's not null flavor (which means embedded packing msg),
2254          * reset the swab mask for the comming inner msg unpacking.
2255          */
2256         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2257                 req->rq_req_swab_mask = 0;
2258
2259         /* sanity check for the request source */
2260         rc = sptlrpc_svc_check_from(req, rc);
2261         RETURN(rc);
2262 }
2263
2264 /**
2265  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2266  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2267  * a buffer of \a msglen size.
2268  */
2269 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2270 {
2271         struct ptlrpc_sec_policy *policy;
2272         struct ptlrpc_reply_state *rs;
2273         int rc;
2274
2275         ENTRY;
2276
2277         LASSERT(req->rq_svc_ctx);
2278         LASSERT(req->rq_svc_ctx->sc_policy);
2279
2280         policy = req->rq_svc_ctx->sc_policy;
2281         LASSERT(policy->sp_sops->alloc_rs);
2282
2283         rc = policy->sp_sops->alloc_rs(req, msglen);
2284         if (unlikely(rc == -ENOMEM)) {
2285                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2286
2287                 if (svcpt->scp_service->srv_max_reply_size <
2288                    msglen + sizeof(struct ptlrpc_reply_state)) {
2289                         /* Just return failure if the size is too big */
2290                         CERROR("size of message is too big (%zd), %d allowed\n",
2291                                 msglen + sizeof(struct ptlrpc_reply_state),
2292                                 svcpt->scp_service->srv_max_reply_size);
2293                         RETURN(-ENOMEM);
2294                 }
2295
2296                 /* failed alloc, try emergency pool */
2297                 rs = lustre_get_emerg_rs(svcpt);
2298                 if (rs == NULL)
2299                         RETURN(-ENOMEM);
2300
2301                 req->rq_reply_state = rs;
2302                 rc = policy->sp_sops->alloc_rs(req, msglen);
2303                 if (rc) {
2304                         lustre_put_emerg_rs(rs);
2305                         req->rq_reply_state = NULL;
2306                 }
2307         }
2308
2309         LASSERT(rc != 0 ||
2310                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2311
2312         RETURN(rc);
2313 }
2314
2315 /**
2316  * Used by ptlrpc server, to perform transformation upon reply message.
2317  *
2318  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2319  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2320  */
2321 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2322 {
2323         struct ptlrpc_sec_policy *policy;
2324         int rc;
2325
2326         ENTRY;
2327
2328         LASSERT(req->rq_svc_ctx);
2329         LASSERT(req->rq_svc_ctx->sc_policy);
2330
2331         policy = req->rq_svc_ctx->sc_policy;
2332         LASSERT(policy->sp_sops->authorize);
2333
2334         rc = policy->sp_sops->authorize(req);
2335         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2336
2337         RETURN(rc);
2338 }
2339
2340 /**
2341  * Used by ptlrpc server, to free reply_state.
2342  */
2343 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2344 {
2345         struct ptlrpc_sec_policy *policy;
2346         unsigned int prealloc;
2347
2348         ENTRY;
2349
2350         LASSERT(rs->rs_svc_ctx);
2351         LASSERT(rs->rs_svc_ctx->sc_policy);
2352
2353         policy = rs->rs_svc_ctx->sc_policy;
2354         LASSERT(policy->sp_sops->free_rs);
2355
2356         prealloc = rs->rs_prealloc;
2357         policy->sp_sops->free_rs(rs);
2358
2359         if (prealloc)
2360                 lustre_put_emerg_rs(rs);
2361         EXIT;
2362 }
2363
2364 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2365 {
2366         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2367
2368         if (ctx != NULL)
2369                 atomic_inc(&ctx->sc_refcount);
2370 }
2371
2372 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2373 {
2374         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2375
2376         if (ctx == NULL)
2377                 return;
2378
2379         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2380         if (atomic_dec_and_test(&ctx->sc_refcount)) {
2381                 if (ctx->sc_policy->sp_sops->free_ctx)
2382                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2383         }
2384         req->rq_svc_ctx = NULL;
2385 }
2386
2387 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2388 {
2389         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2390
2391         if (ctx == NULL)
2392                 return;
2393
2394         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2395         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2396                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2397 }
2398 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2399
2400 /*
2401  * bulk security
2402  */
2403
2404 /**
2405  * Perform transformation upon bulk data pointed by \a desc. This is called
2406  * before transforming the request message.
2407  */
2408 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2409                           struct ptlrpc_bulk_desc *desc)
2410 {
2411         struct ptlrpc_cli_ctx *ctx;
2412
2413         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2414
2415         if (!req->rq_pack_bulk)
2416                 return 0;
2417
2418         ctx = req->rq_cli_ctx;
2419         if (ctx->cc_ops->wrap_bulk)
2420                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2421         return 0;
2422 }
2423 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2424
2425 /**
2426  * This is called after unwrap the reply message.
2427  * return nob of actual plain text size received, or error code.
2428  */
2429 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2430                                  struct ptlrpc_bulk_desc *desc,
2431                                  int nob)
2432 {
2433         struct ptlrpc_cli_ctx *ctx;
2434         int rc;
2435
2436         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2437
2438         if (!req->rq_pack_bulk)
2439                 return desc->bd_nob_transferred;
2440
2441         ctx = req->rq_cli_ctx;
2442         if (ctx->cc_ops->unwrap_bulk) {
2443                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2444                 if (rc < 0)
2445                         return rc;
2446         }
2447         return desc->bd_nob_transferred;
2448 }
2449 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2450
2451 /**
2452  * This is called after unwrap the reply message.
2453  * return 0 for success or error code.
2454  */
2455 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2456                                   struct ptlrpc_bulk_desc *desc)
2457 {
2458         struct ptlrpc_cli_ctx *ctx;
2459         int rc;
2460
2461         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2462
2463         if (!req->rq_pack_bulk)
2464                 return 0;
2465
2466         ctx = req->rq_cli_ctx;
2467         if (ctx->cc_ops->unwrap_bulk) {
2468                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2469                 if (rc < 0)
2470                         return rc;
2471         }
2472
2473         /*
2474          * if everything is going right, nob should equals to nob_transferred.
2475          * in case of privacy mode, nob_transferred needs to be adjusted.
2476          */
2477         if (desc->bd_nob != desc->bd_nob_transferred) {
2478                 CERROR("nob %d doesn't match transferred nob %d\n",
2479                        desc->bd_nob, desc->bd_nob_transferred);
2480                 return -EPROTO;
2481         }
2482
2483         return 0;
2484 }
2485 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2486
2487 #ifdef HAVE_SERVER_SUPPORT
2488 /**
2489  * Performe transformation upon outgoing bulk read.
2490  */
2491 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2492                           struct ptlrpc_bulk_desc *desc)
2493 {
2494         struct ptlrpc_svc_ctx *ctx;
2495
2496         LASSERT(req->rq_bulk_read);
2497
2498         if (!req->rq_pack_bulk)
2499                 return 0;
2500
2501         ctx = req->rq_svc_ctx;
2502         if (ctx->sc_policy->sp_sops->wrap_bulk)
2503                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2504
2505         return 0;
2506 }
2507 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2508
2509 /**
2510  * Performe transformation upon incoming bulk write.
2511  */
2512 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2513                             struct ptlrpc_bulk_desc *desc)
2514 {
2515         struct ptlrpc_svc_ctx *ctx;
2516         int rc;
2517
2518         LASSERT(req->rq_bulk_write);
2519
2520         /*
2521          * if it's in privacy mode, transferred should >= expected; otherwise
2522          * transferred should == expected.
2523          */
2524         if (desc->bd_nob_transferred < desc->bd_nob ||
2525             (desc->bd_nob_transferred > desc->bd_nob &&
2526              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2527              SPTLRPC_BULK_SVC_PRIV)) {
2528                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2529                           desc->bd_nob_transferred, desc->bd_nob);
2530                 return -ETIMEDOUT;
2531         }
2532
2533         if (!req->rq_pack_bulk)
2534                 return 0;
2535
2536         ctx = req->rq_svc_ctx;
2537         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2538                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2539                 if (rc)
2540                         CERROR("error unwrap bulk: %d\n", rc);
2541         }
2542
2543         /* return 0 to allow reply be sent */
2544         return 0;
2545 }
2546 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2547
2548 /**
2549  * Prepare buffers for incoming bulk write.
2550  */
2551 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2552                           struct ptlrpc_bulk_desc *desc)
2553 {
2554         struct ptlrpc_svc_ctx *ctx;
2555
2556         LASSERT(req->rq_bulk_write);
2557
2558         if (!req->rq_pack_bulk)
2559                 return 0;
2560
2561         ctx = req->rq_svc_ctx;
2562         if (ctx->sc_policy->sp_sops->prep_bulk)
2563                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2564
2565         return 0;
2566 }
2567 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2568
2569 #endif /* HAVE_SERVER_SUPPORT */
2570
2571 /*
2572  * user descriptor helpers
2573  */
2574
2575 int sptlrpc_current_user_desc_size(void)
2576 {
2577         int ngroups;
2578
2579         ngroups = current_ngroups;
2580
2581         if (ngroups > LUSTRE_MAX_GROUPS)
2582                 ngroups = LUSTRE_MAX_GROUPS;
2583         return sptlrpc_user_desc_size(ngroups);
2584 }
2585 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2586
2587 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2588 {
2589         struct ptlrpc_user_desc *pud;
2590
2591         pud = lustre_msg_buf(msg, offset, 0);
2592
2593         pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2594         pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2595         pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2596         pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2597         pud->pud_cap = cfs_curproc_cap_pack();
2598         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2599
2600         task_lock(current);
2601         if (pud->pud_ngroups > current_ngroups)
2602                 pud->pud_ngroups = current_ngroups;
2603 #ifdef HAVE_GROUP_INFO_GID
2604         memcpy(pud->pud_groups, current_cred()->group_info->gid,
2605                pud->pud_ngroups * sizeof(__u32));
2606 #else /* !HAVE_GROUP_INFO_GID */
2607         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2608                pud->pud_ngroups * sizeof(__u32));
2609 #endif /* HAVE_GROUP_INFO_GID */
2610         task_unlock(current);
2611
2612         return 0;
2613 }
2614 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2615
2616 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2617 {
2618         struct ptlrpc_user_desc *pud;
2619         int i;
2620
2621         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2622         if (!pud)
2623                 return -EINVAL;
2624
2625         if (swabbed) {
2626                 __swab32s(&pud->pud_uid);
2627                 __swab32s(&pud->pud_gid);
2628                 __swab32s(&pud->pud_fsuid);
2629                 __swab32s(&pud->pud_fsgid);
2630                 __swab32s(&pud->pud_cap);
2631                 __swab32s(&pud->pud_ngroups);
2632         }
2633
2634         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2635                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2636                 return -EINVAL;
2637         }
2638
2639         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2640             msg->lm_buflens[offset]) {
2641                 CERROR("%u groups are claimed but bufsize only %u\n",
2642                        pud->pud_ngroups, msg->lm_buflens[offset]);
2643                 return -EINVAL;
2644         }
2645
2646         if (swabbed) {
2647                 for (i = 0; i < pud->pud_ngroups; i++)
2648                         __swab32s(&pud->pud_groups[i]);
2649         }
2650
2651         return 0;
2652 }
2653 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2654
2655 /*
2656  * misc helpers
2657  */
2658
2659 const char *sec2target_str(struct ptlrpc_sec *sec)
2660 {
2661         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2662                 return "*";
2663         if (sec_is_reverse(sec))
2664                 return "c";
2665         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2666 }
2667 EXPORT_SYMBOL(sec2target_str);
2668
2669 /*
2670  * return true if the bulk data is protected
2671  */
2672 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2673 {
2674         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2675         case SPTLRPC_BULK_SVC_INTG:
2676         case SPTLRPC_BULK_SVC_PRIV:
2677                 return 1;
2678         default:
2679                 return 0;
2680         }
2681 }
2682 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2683
2684 /*
2685  * crypto API helper/alloc blkciper
2686  */
2687
2688 /*
2689  * initialize/finalize
2690  */
2691
2692 int sptlrpc_init(void)
2693 {
2694         int rc;
2695
2696         rwlock_init(&policy_lock);
2697
2698         rc = sptlrpc_gc_init();
2699         if (rc)
2700                 goto out;
2701
2702         rc = sptlrpc_conf_init();
2703         if (rc)
2704                 goto out_gc;
2705
2706         rc = sptlrpc_enc_pool_init();
2707         if (rc)
2708                 goto out_conf;
2709
2710         rc = sptlrpc_null_init();
2711         if (rc)
2712                 goto out_pool;
2713
2714         rc = sptlrpc_plain_init();
2715         if (rc)
2716                 goto out_null;
2717
2718         rc = sptlrpc_lproc_init();
2719         if (rc)
2720                 goto out_plain;
2721
2722         return 0;
2723
2724 out_plain:
2725         sptlrpc_plain_fini();
2726 out_null:
2727         sptlrpc_null_fini();
2728 out_pool:
2729         sptlrpc_enc_pool_fini();
2730 out_conf:
2731         sptlrpc_conf_fini();
2732 out_gc:
2733         sptlrpc_gc_fini();
2734 out:
2735         return rc;
2736 }
2737
2738 void sptlrpc_fini(void)
2739 {
2740         sptlrpc_lproc_fini();
2741         sptlrpc_plain_fini();
2742         sptlrpc_null_fini();
2743         sptlrpc_enc_pool_fini();
2744         sptlrpc_conf_fini();
2745         sptlrpc_gc_fini();
2746 }