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  *
  31  * lustre/ptlrpc/sec.c
  32  *
  33  * Author: Eric Mei <ericm@clusterfs.com>
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_SEC
  37
  38 #include <linux/user_namespace.h>
  39 #include <linux/uidgid.h>
  40 #include <linux/crypto.h>
  41 #include <linux/key.h>
  42
  43 #include <libcfs/libcfs.h>
  44 #include <obd.h>
  45 #include <obd_class.h>
  46 #include <obd_support.h>
  47 #include <lustre_net.h>
  48 #include <lustre_import.h>
  49 #include <lustre_dlm.h>
  50 #include <lustre_sec.h>
  51
  52 #include "ptlrpc_internal.h"
  53
  54 static int send_sepol;
  55 module_param(send_sepol, int, 0644);
  56 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
  57
  58 /*
  59  * policy registers
  60  */
  61
  62 static rwlock_t policy_lock;
  63 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  64         NULL,
  65 };
  66
  67 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  68 {
  69         __u16 number = policy->sp_policy;
  70
  71         LASSERT(policy->sp_name);
  72         LASSERT(policy->sp_cops);
  73         LASSERT(policy->sp_sops);
  74
  75         if (number >= SPTLRPC_POLICY_MAX)
  76                 return -EINVAL;
  77
  78         write_lock(&policy_lock);
  79         if (unlikely(policies[number])) {
  80                 write_unlock(&policy_lock);
  81                 return -EALREADY;
  82         }
  83         policies[number] = policy;
  84         write_unlock(&policy_lock);
  85
  86         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  87         return 0;
  88 }
  89 EXPORT_SYMBOL(sptlrpc_register_policy);
  90
  91 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  92 {
  93         __u16 number = policy->sp_policy;
  94
  95         LASSERT(number < SPTLRPC_POLICY_MAX);
  96
  97         write_lock(&policy_lock);
  98         if (unlikely(policies[number] == NULL)) {
  99                 write_unlock(&policy_lock);
 100                 CERROR("%s: already unregistered\n", policy->sp_name);
 101                 return -EINVAL;
 102         }
 103
 104         LASSERT(policies[number] == policy);
 105         policies[number] = NULL;
 106         write_unlock(&policy_lock);
 107
 108         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 109         return 0;
 110 }
 111 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 112
 113 static
 114 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
 115 {
 116         static DEFINE_MUTEX(load_mutex);
 117         static atomic_t           loaded = ATOMIC_INIT(0);
 118         struct ptlrpc_sec_policy *policy;
 119         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 120         __u16                     flag = 0;
 121
 122         if (number >= SPTLRPC_POLICY_MAX)
 123                 return NULL;
 124
 125         while (1) {
 126                 read_lock(&policy_lock);
 127                 policy = policies[number];
 128                 if (policy && !try_module_get(policy->sp_owner))
 129                         policy = NULL;
 130                 if (policy == NULL)
 131                         flag = atomic_read(&loaded);
 132                 read_unlock(&policy_lock);
 133
 134                 if (policy != NULL || flag != 0 ||
 135                     number != SPTLRPC_POLICY_GSS)
 136                         break;
 137
 138                 /* try to load gss module, once */
 139                 mutex_lock(&load_mutex);
 140                 if (atomic_read(&loaded) == 0) {
 141                         if (request_module("ptlrpc_gss") == 0)
 142                                 CDEBUG(D_SEC,
 143                                        "module ptlrpc_gss loaded on demand\n");
 144                         else
 145                                 CERROR("Unable to load module ptlrpc_gss\n");
 146
 147                         atomic_set(&loaded, 1);
 148                 }
 149                 mutex_unlock(&load_mutex);
 150         }
 151
 152         return policy;
 153 }
 154
 155 __u32 sptlrpc_name2flavor_base(const char *name)
 156 {
 157         if (!strcmp(name, "null"))
 158                 return SPTLRPC_FLVR_NULL;
 159         if (!strcmp(name, "plain"))
 160                 return SPTLRPC_FLVR_PLAIN;
 161         if (!strcmp(name, "gssnull"))
 162                 return SPTLRPC_FLVR_GSSNULL;
 163         if (!strcmp(name, "krb5n"))
 164                 return SPTLRPC_FLVR_KRB5N;
 165         if (!strcmp(name, "krb5a"))
 166                 return SPTLRPC_FLVR_KRB5A;
 167         if (!strcmp(name, "krb5i"))
 168                 return SPTLRPC_FLVR_KRB5I;
 169         if (!strcmp(name, "krb5p"))
 170                 return SPTLRPC_FLVR_KRB5P;
 171         if (!strcmp(name, "skn"))
 172                 return SPTLRPC_FLVR_SKN;
 173         if (!strcmp(name, "ska"))
 174                 return SPTLRPC_FLVR_SKA;
 175         if (!strcmp(name, "ski"))
 176                 return SPTLRPC_FLVR_SKI;
 177         if (!strcmp(name, "skpi"))
 178                 return SPTLRPC_FLVR_SKPI;
 179
 180         return SPTLRPC_FLVR_INVALID;
 181 }
 182 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 183
 184 const char *sptlrpc_flavor2name_base(__u32 flvr)
 185 {
 186         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 187
 188         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 189                 return "null";
 190         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 191                 return "plain";
 192         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
 193                 return "gssnull";
 194         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 195                 return "krb5n";
 196         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 197                 return "krb5a";
 198         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 199                 return "krb5i";
 200         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 201                 return "krb5p";
 202         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
 203                 return "skn";
 204         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
 205                 return "ska";
 206         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
 207                 return "ski";
 208         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
 209                 return "skpi";
 210
 211         CERROR("invalid wire flavor 0x%x\n", flvr);
 212         return "invalid";
 213 }
 214 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 215
 216 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 217                                char *buf, int bufsize)
 218 {
 219         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 220                 snprintf(buf, bufsize, "hash:%s",
 221                         sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 222         else
 223                 snprintf(buf, bufsize, "%s",
 224                         sptlrpc_flavor2name_base(sf->sf_rpc));
 225
 226         buf[bufsize - 1] = '\0';
 227         return buf;
 228 }
 229 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 230
 231 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 232 {
 233         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 234
 235         /*
 236          * currently we don't support customized bulk specification for
 237          * flavors other than plain
 238          */
 239         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 240                 char bspec[16];
 241
 242                 bspec[0] = '-';
 243                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 244                 strncat(buf, bspec, bufsize);
 245         }
 246
 247         buf[bufsize - 1] = '\0';
 248         return buf;
 249 }
 250 EXPORT_SYMBOL(sptlrpc_flavor2name);
 251
 252 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 253 {
 254         buf[0] = '\0';
 255
 256         if (flags & PTLRPC_SEC_FL_REVERSE)
 257                 strlcat(buf, "reverse,", bufsize);
 258         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 259                 strlcat(buf, "rootonly,", bufsize);
 260         if (flags & PTLRPC_SEC_FL_UDESC)
 261                 strlcat(buf, "udesc,", bufsize);
 262         if (flags & PTLRPC_SEC_FL_BULK)
 263                 strlcat(buf, "bulk,", bufsize);
 264         if (buf[0] == '\0')
 265                 strlcat(buf, "-,", bufsize);
 266
 267         return buf;
 268 }
 269 EXPORT_SYMBOL(sptlrpc_secflags2str);
 270
 271 /*
 272  * client context APIs
 273  */
 274
 275 static
 276 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 277 {
 278         struct vfs_cred vcred;
 279         int create = 1, remove_dead = 1;
 280
 281         LASSERT(sec);
 282         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 283
 284         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 285                                      PTLRPC_SEC_FL_ROOTONLY)) {
 286                 vcred.vc_uid = 0;
 287                 vcred.vc_gid = 0;
 288                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 289                         create = 0;
 290                         remove_dead = 0;
 291                 }
 292         } else {
 293                 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
 294                 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
 295         }
 296
 297         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
 298                                                    remove_dead);
 299 }
 300
 301 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 302 {
 303         atomic_inc(&ctx->cc_refcount);
 304         return ctx;
 305 }
 306 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 307
 308 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 309 {
 310         struct ptlrpc_sec *sec = ctx->cc_sec;
 311
 312         LASSERT(sec);
 313         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 314
 315         if (!atomic_dec_and_test(&ctx->cc_refcount))
 316                 return;
 317
 318         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 319 }
 320 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 321
 322 /**
 323  * Expire the client context immediately.
 324  *
 325  * \pre Caller must hold at least 1 reference on the \a ctx.
 326  */
 327 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 328 {
 329         LASSERT(ctx->cc_ops->die);
 330         ctx->cc_ops->die(ctx, 0);
 331 }
 332 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 333
 334 /**
 335  * To wake up the threads who are waiting for this client context. Called
 336  * after some status change happened on \a ctx.
 337  */
 338 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 339 {
 340         struct ptlrpc_request *req, *next;
 341
 342         spin_lock(&ctx->cc_lock);
 343         list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 344                                      rq_ctx_chain) {
 345                 list_del_init(&req->rq_ctx_chain);
 346                 ptlrpc_client_wake_req(req);
 347         }
 348         spin_unlock(&ctx->cc_lock);
 349 }
 350 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 351
 352 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 353 {
 354         LASSERT(ctx->cc_ops);
 355
 356         if (ctx->cc_ops->display == NULL)
 357                 return 0;
 358
 359         return ctx->cc_ops->display(ctx, buf, bufsize);
 360 }
 361
 362 static int import_sec_check_expire(struct obd_import *imp)
 363 {
 364         int adapt = 0;
 365
 366         write_lock(&imp->imp_sec_lock);
 367         if (imp->imp_sec_expire &&
 368             imp->imp_sec_expire < ktime_get_real_seconds()) {
 369                 adapt = 1;
 370                 imp->imp_sec_expire = 0;
 371         }
 372         write_unlock(&imp->imp_sec_lock);
 373
 374         if (!adapt)
 375                 return 0;
 376
 377         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 378         return sptlrpc_import_sec_adapt(imp, NULL, NULL);
 379 }
 380
 381 /**
 382  * Get and validate the client side ptlrpc security facilities from
 383  * \a imp. There is a race condition on client reconnect when the import is
 384  * being destroyed while there are outstanding client bound requests. In
 385  * this case do not output any error messages if import secuity is not
 386  * found.
 387  *
 388  * \param[in] imp obd import associated with client
 389  * \param[out] sec client side ptlrpc security
 390  *
 391  * \retval 0 if security retrieved successfully
 392  * \retval -ve errno if there was a problem
 393  */
 394 static int import_sec_validate_get(struct obd_import *imp,
 395                                    struct ptlrpc_sec **sec)
 396 {
 397         int rc;
 398
 399         if (unlikely(imp->imp_sec_expire)) {
 400                 rc = import_sec_check_expire(imp);
 401                 if (rc)
 402                         return rc;
 403         }
 404
 405         *sec = sptlrpc_import_sec_ref(imp);
 406         if (*sec == NULL) {
 407                 /* Only output an error when the import is still active */
 408                 if (!test_bit(WORK_STRUCT_PENDING_BIT,
 409                               work_data_bits(&imp->imp_zombie_work)))
 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         CDEBUG(D_SEC,
 500                "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
 501                req, oldctx, oldctx->cc_vcred.vc_uid,
 502                sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
 503                sec2target_str(newctx->cc_sec), oldctx->cc_sec,
 504                oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
 505                newctx->cc_sec->ps_policy->sp_name);
 506
 507         /* save flavor */
 508         old_flvr = req->rq_flvr;
 509
 510         /* save request message */
 511         reqmsg_size = req->rq_reqlen;
 512         if (reqmsg_size != 0) {
 513                 LASSERT(req->rq_reqmsg);
 514                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 515                 if (reqmsg == NULL)
 516                         return -ENOMEM;
 517                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 518         }
 519
 520         /* release old req/rep buf */
 521         req->rq_cli_ctx = oldctx;
 522         sptlrpc_cli_free_reqbuf(req);
 523         sptlrpc_cli_free_repbuf(req);
 524         req->rq_cli_ctx = newctx;
 525
 526         /* recalculate the flavor */
 527         sptlrpc_req_set_flavor(req, 0);
 528
 529         /*
 530          * alloc new request buffer
 531          * we don't need to alloc reply buffer here, leave it to the
 532          * rest procedure of ptlrpc
 533          */
 534         if (reqmsg_size != 0) {
 535                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 536                 if (!rc) {
 537                         LASSERT(req->rq_reqmsg);
 538                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 539                 } else {
 540                         CWARN("failed to alloc reqbuf: %d\n", rc);
 541                         req->rq_flvr = old_flvr;
 542                 }
 543
 544                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 545         }
 546         return rc;
 547 }
 548
 549 /**
 550  * If current context of \a req is dead somehow, e.g. we just switched flavor
 551  * thus marked original contexts dead, we'll find a new context for it. if
 552  * no switch is needed, \a req will end up with the same context.
 553  *
 554  * \note a request must have a context, to keep other parts of code happy.
 555  * In any case of failure during the switching, we must restore the old one.
 556  */
 557 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 558 {
 559         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 560         struct ptlrpc_cli_ctx *newctx;
 561         int rc;
 562
 563         ENTRY;
 564
 565         LASSERT(oldctx);
 566
 567         sptlrpc_cli_ctx_get(oldctx);
 568         sptlrpc_req_put_ctx(req, 0);
 569
 570         rc = sptlrpc_req_get_ctx(req);
 571         if (unlikely(rc)) {
 572                 LASSERT(!req->rq_cli_ctx);
 573
 574                 /* restore old ctx */
 575                 req->rq_cli_ctx = oldctx;
 576                 RETURN(rc);
 577         }
 578
 579         newctx = req->rq_cli_ctx;
 580         LASSERT(newctx);
 581
 582         if (unlikely(newctx == oldctx &&
 583                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 584                 /*
 585                  * still get the old dead ctx, usually means system too busy
 586                  */
 587                 CDEBUG(D_SEC,
 588                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 589                        newctx, newctx->cc_flags);
 590
 591                 schedule_timeout_interruptible(cfs_time_seconds(1));
 592         } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
 593                             == 0)) {
 594                 /*
 595                  * new ctx not up to date yet
 596                  */
 597                 CDEBUG(D_SEC,
 598                        "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
 599                        newctx, newctx->cc_flags);
 600         } else {
 601                 /*
 602                  * it's possible newctx == oldctx if we're switching
 603                  * subflavor with the same sec.
 604                  */
 605                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 606                 if (rc) {
 607                         /* restore old ctx */
 608                         sptlrpc_req_put_ctx(req, 0);
 609                         req->rq_cli_ctx = oldctx;
 610                         RETURN(rc);
 611                 }
 612
 613                 LASSERT(req->rq_cli_ctx == newctx);
 614         }
 615
 616         sptlrpc_cli_ctx_put(oldctx, 1);
 617         RETURN(0);
 618 }
 619 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 620
 621 static
 622 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 623 {
 624         if (cli_ctx_is_refreshed(ctx))
 625                 return 1;
 626         return 0;
 627 }
 628
 629 static
 630 void ctx_refresh_interrupt(struct ptlrpc_request *req)
 631 {
 632
 633         spin_lock(&req->rq_lock);
 634         req->rq_intr = 1;
 635         spin_unlock(&req->rq_lock);
 636 }
 637
 638 static
 639 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 640 {
 641         spin_lock(&ctx->cc_lock);
 642         if (!list_empty(&req->rq_ctx_chain))
 643                 list_del_init(&req->rq_ctx_chain);
 644         spin_unlock(&ctx->cc_lock);
 645 }
 646
 647 /**
 648  * To refresh the context of \req, if it's not up-to-date.
 649  * \param timeout
 650  * - == 0: do not wait
 651  * - == MAX_SCHEDULE_TIMEOUT: wait indefinitely
 652  * - > 0: not supported
 653  *
 654  * The status of the context could be subject to be changed by other threads
 655  * at any time. We allow this race, but once we return with 0, the caller will
 656  * suppose it's uptodated and keep using it until the owning rpc is done.
 657  *
 658  * \retval 0 only if the context is uptodated.
 659  * \retval -ev error number.
 660  */
 661 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 662 {
 663         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 664         struct ptlrpc_sec *sec;
 665         int rc;
 666
 667         ENTRY;
 668
 669         LASSERT(ctx);
 670
 671         if (req->rq_ctx_init || req->rq_ctx_fini)
 672                 RETURN(0);
 673
 674         if (timeout != 0 && timeout != MAX_SCHEDULE_TIMEOUT) {
 675                 CERROR("req %p: invalid timeout %lu\n", req, timeout);
 676                 RETURN(-EINVAL);
 677         }
 678
 679         /*
 680          * during the process a request's context might change type even
 681          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 682          * everything
 683          */
 684 again:
 685         rc = import_sec_validate_get(req->rq_import, &sec);
 686         if (rc)
 687                 RETURN(rc);
 688
 689         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 690                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 691                        req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 692                 req_off_ctx_list(req, ctx);
 693                 sptlrpc_req_replace_dead_ctx(req);
 694                 ctx = req->rq_cli_ctx;
 695         }
 696         sptlrpc_sec_put(sec);
 697
 698         if (cli_ctx_is_eternal(ctx))
 699                 RETURN(0);
 700
 701         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 702                 if (ctx->cc_ops->refresh)
 703                         ctx->cc_ops->refresh(ctx);
 704         }
 705         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 706
 707         LASSERT(ctx->cc_ops->validate);
 708         if (ctx->cc_ops->validate(ctx) == 0) {
 709                 req_off_ctx_list(req, ctx);
 710                 RETURN(0);
 711         }
 712
 713         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 714                 spin_lock(&req->rq_lock);
 715                 req->rq_err = 1;
 716                 spin_unlock(&req->rq_lock);
 717                 req_off_ctx_list(req, ctx);
 718                 RETURN(-EPERM);
 719         }
 720
 721         /*
 722          * There's a subtle issue for resending RPCs, suppose following
 723          * situation:
 724          *  1. the request was sent to server.
 725          *  2. recovery was kicked start, after finished the request was
 726          *     marked as resent.
 727          *  3. resend the request.
 728          *  4. old reply from server received, we accept and verify the reply.
 729          *     this has to be success, otherwise the error will be aware
 730          *     by application.
 731          *  5. new reply from server received, dropped by LNet.
 732          *
 733          * Note the xid of old & new request is the same. We can't simply
 734          * change xid for the resent request because the server replies on
 735          * it for reply reconstruction.
 736          *
 737          * Commonly the original context should be uptodate because we
 738          * have an expiry nice time; server will keep its context because
 739          * we at least hold a ref of old context which prevent context
 740          * from destroying RPC being sent. So server still can accept the
 741          * request and finish the RPC. But if that's not the case:
 742          *  1. If server side context has been trimmed, a NO_CONTEXT will
 743          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 744          *     context by force.
 745          *  2. Current context never be refreshed, then we are fine: we
 746          *     never really send request with old context before.
 747          */
 748         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 749             unlikely(req->rq_reqmsg) &&
 750             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 751                 req_off_ctx_list(req, ctx);
 752                 RETURN(0);
 753         }
 754
 755         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 756                 req_off_ctx_list(req, ctx);
 757                 /*
 758                  * don't switch ctx if import was deactivated
 759                  */
 760                 if (req->rq_import->imp_deactive) {
 761                         spin_lock(&req->rq_lock);
 762                         req->rq_err = 1;
 763                         spin_unlock(&req->rq_lock);
 764                         RETURN(-EINTR);
 765                 }
 766
 767                 rc = sptlrpc_req_replace_dead_ctx(req);
 768                 if (rc) {
 769                         LASSERT(ctx == req->rq_cli_ctx);
 770                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 771                                req, ctx, rc);
 772                         spin_lock(&req->rq_lock);
 773                         req->rq_err = 1;
 774                         spin_unlock(&req->rq_lock);
 775                         RETURN(rc);
 776                 }
 777
 778                 ctx = req->rq_cli_ctx;
 779                 goto again;
 780         }
 781
 782         /*
 783          * Now we're sure this context is during upcall, add myself into
 784          * waiting list
 785          */
 786         spin_lock(&ctx->cc_lock);
 787         if (list_empty(&req->rq_ctx_chain))
 788                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 789         spin_unlock(&ctx->cc_lock);
 790
 791         if (timeout == 0)
 792                 RETURN(-EAGAIN);
 793
 794         /* Clear any flags that may be present from previous sends */
 795         LASSERT(req->rq_receiving_reply == 0);
 796         spin_lock(&req->rq_lock);
 797         req->rq_err = 0;
 798         req->rq_timedout = 0;
 799         req->rq_resend = 0;
 800         req->rq_restart = 0;
 801         spin_unlock(&req->rq_lock);
 802
 803         /* by now we know that timeout value is MAX_SCHEDULE_TIMEOUT,
 804          * so wait indefinitely with non-fatal signals blocked
 805          */
 806         if (l_wait_event_abortable(req->rq_reply_waitq,
 807                                    ctx_check_refresh(ctx)) == -ERESTARTSYS) {
 808                 rc = -EINTR;
 809                 ctx_refresh_interrupt(req);
 810         }
 811
 812         /*
 813          * following cases could lead us here:
 814          * - successfully refreshed;
 815          * - interrupted;
 816          * - timedout, and we don't want recover from the failure;
 817          * - timedout, and waked up upon recovery finished;
 818          * - someone else mark this ctx dead by force;
 819          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 820          *   e.g. ptlrpc_abort_inflight();
 821          */
 822         if (!cli_ctx_is_refreshed(ctx)) {
 823                 /* timed out or interruptted */
 824                 req_off_ctx_list(req, ctx);
 825
 826                 LASSERT(rc != 0);
 827                 RETURN(rc);
 828         }
 829
 830         goto again;
 831 }
 832
 833 /* Bring ptlrpc_sec context up-to-date */
 834 int sptlrpc_export_update_ctx(struct obd_export *exp)
 835 {
 836         struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
 837         struct ptlrpc_sec *sec = NULL;
 838         struct ptlrpc_cli_ctx *ctx = NULL;
 839         int rc = 0;
 840
 841         if (imp)
 842                 sec = sptlrpc_import_sec_ref(imp);
 843         if (sec) {
 844                 ctx = get_my_ctx(sec);
 845                 sptlrpc_sec_put(sec);
 846         }
 847
 848         if (ctx) {
 849                 if (ctx->cc_ops->refresh)
 850                         rc = ctx->cc_ops->refresh(ctx);
 851                 sptlrpc_cli_ctx_put(ctx, 1);
 852         }
 853         return rc;
 854 }
 855
 856 /**
 857  * Initialize flavor settings for \a req, according to \a opcode.
 858  *
 859  * \note this could be called in two situations:
 860  * - new request from ptlrpc_pre_req(), with proper @opcode
 861  * - old request which changed ctx in the middle, with @opcode == 0
 862  */
 863 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 864 {
 865         struct ptlrpc_sec *sec;
 866
 867         LASSERT(req->rq_import);
 868         LASSERT(req->rq_cli_ctx);
 869         LASSERT(req->rq_cli_ctx->cc_sec);
 870         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 871
 872         /* special security flags according to opcode */
 873         switch (opcode) {
 874         case OST_READ:
 875         case MDS_READPAGE:
 876         case MGS_CONFIG_READ:
 877         case OBD_IDX_READ:
 878                 req->rq_bulk_read = 1;
 879                 break;
 880         case OST_WRITE:
 881         case MDS_WRITEPAGE:
 882                 req->rq_bulk_write = 1;
 883                 break;
 884         case SEC_CTX_INIT:
 885                 req->rq_ctx_init = 1;
 886                 break;
 887         case SEC_CTX_FINI:
 888                 req->rq_ctx_fini = 1;
 889                 break;
 890         case 0:
 891                 /* init/fini rpc won't be resend, so can't be here */
 892                 LASSERT(req->rq_ctx_init == 0);
 893                 LASSERT(req->rq_ctx_fini == 0);
 894
 895                 /* cleanup flags, which should be recalculated */
 896                 req->rq_pack_udesc = 0;
 897                 req->rq_pack_bulk = 0;
 898                 break;
 899         }
 900
 901         sec = req->rq_cli_ctx->cc_sec;
 902
 903         spin_lock(&sec->ps_lock);
 904         req->rq_flvr = sec->ps_flvr;
 905         spin_unlock(&sec->ps_lock);
 906
 907         /*
 908          * force SVC_NULL for context initiation rpc, SVC_INTG for context
 909          * destruction rpc
 910          */
 911         if (unlikely(req->rq_ctx_init))
 912                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 913         else if (unlikely(req->rq_ctx_fini))
 914                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 915
 916         /* user descriptor flag, null security can't do it anyway */
 917         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 918             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 919                 req->rq_pack_udesc = 1;
 920
 921         /* bulk security flag */
 922         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 923             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 924                 req->rq_pack_bulk = 1;
 925 }
 926
 927 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 928 {
 929         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 930                 return;
 931
 932         LASSERT(req->rq_clrbuf);
 933         if (req->rq_pool || !req->rq_reqbuf)
 934                 return;
 935
 936         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 937         req->rq_reqbuf = NULL;
 938         req->rq_reqbuf_len = 0;
 939 }
 940
 941 /**
 942  * Given an import \a imp, check whether current user has a valid context
 943  * or not. We may create a new context and try to refresh it, and try
 944  * repeatedly try in case of non-fatal errors. Return 0 means success.
 945  */
 946 int sptlrpc_import_check_ctx(struct obd_import *imp)
 947 {
 948         struct ptlrpc_sec     *sec;
 949         struct ptlrpc_cli_ctx *ctx;
 950         struct ptlrpc_request *req = NULL;
 951         int rc;
 952
 953         ENTRY;
 954
 955         might_sleep();
 956
 957         sec = sptlrpc_import_sec_ref(imp);
 958         ctx = get_my_ctx(sec);
 959         sptlrpc_sec_put(sec);
 960
 961         if (!ctx)
 962                 RETURN(-ENOMEM);
 963
 964         if (cli_ctx_is_eternal(ctx) ||
 965             ctx->cc_ops->validate(ctx) == 0) {
 966                 sptlrpc_cli_ctx_put(ctx, 1);
 967                 RETURN(0);
 968         }
 969
 970         if (cli_ctx_is_error(ctx)) {
 971                 sptlrpc_cli_ctx_put(ctx, 1);
 972                 RETURN(-EACCES);
 973         }
 974
 975         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 976         if (!req)
 977                 RETURN(-ENOMEM);
 978
 979         ptlrpc_cli_req_init(req);
 980         atomic_set(&req->rq_refcount, 10000);
 981
 982         req->rq_import = imp;
 983         req->rq_flvr = sec->ps_flvr;
 984         req->rq_cli_ctx = ctx;
 985
 986         rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
 987         LASSERT(list_empty(&req->rq_ctx_chain));
 988         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 989         ptlrpc_request_cache_free(req);
 990
 991         RETURN(rc);
 992 }
 993
 994 /**
 995  * Used by ptlrpc client, to perform the pre-defined security transformation
 996  * upon the request message of \a req. After this function called,
 997  * req->rq_reqmsg is still accessible as clear text.
 998  */
 999 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1000 {
1001         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1002         int rc = 0;
1003
1004         ENTRY;
1005
1006         LASSERT(ctx);
1007         LASSERT(ctx->cc_sec);
1008         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1009
1010         /*
1011          * we wrap bulk request here because now we can be sure
1012          * the context is uptodate.
1013          */
1014         if (req->rq_bulk) {
1015                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1016                 if (rc)
1017                         RETURN(rc);
1018         }
1019
1020         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1021         case SPTLRPC_SVC_NULL:
1022         case SPTLRPC_SVC_AUTH:
1023         case SPTLRPC_SVC_INTG:
1024                 LASSERT(ctx->cc_ops->sign);
1025                 rc = ctx->cc_ops->sign(ctx, req);
1026                 break;
1027         case SPTLRPC_SVC_PRIV:
1028                 LASSERT(ctx->cc_ops->seal);
1029                 rc = ctx->cc_ops->seal(ctx, req);
1030                 break;
1031         default:
1032                 LBUG();
1033         }
1034
1035         if (rc == 0) {
1036                 LASSERT(req->rq_reqdata_len);
1037                 LASSERT(req->rq_reqdata_len % 8 == 0);
1038                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1039         }
1040
1041         RETURN(rc);
1042 }
1043
1044 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1045 {
1046         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1047         int rc;
1048
1049         ENTRY;
1050
1051         LASSERT(ctx);
1052         LASSERT(ctx->cc_sec);
1053         LASSERT(req->rq_repbuf);
1054         LASSERT(req->rq_repdata);
1055         LASSERT(req->rq_repmsg == NULL);
1056
1057         req->rq_rep_swab_mask = 0;
1058
1059         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1060         switch (rc) {
1061         case 1:
1062                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1063         case 0:
1064                 break;
1065         default:
1066                 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1067                 RETURN(-EPROTO);
1068         }
1069
1070         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1071                 CERROR("replied data length %d too small\n",
1072                        req->rq_repdata_len);
1073                 RETURN(-EPROTO);
1074         }
1075
1076         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1077             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1078                 CERROR("reply policy %u doesn't match request policy %u\n",
1079                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1080                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1081                 RETURN(-EPROTO);
1082         }
1083
1084         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1085         case SPTLRPC_SVC_NULL:
1086         case SPTLRPC_SVC_AUTH:
1087         case SPTLRPC_SVC_INTG:
1088                 LASSERT(ctx->cc_ops->verify);
1089                 rc = ctx->cc_ops->verify(ctx, req);
1090                 break;
1091         case SPTLRPC_SVC_PRIV:
1092                 LASSERT(ctx->cc_ops->unseal);
1093                 rc = ctx->cc_ops->unseal(ctx, req);
1094                 break;
1095         default:
1096                 LBUG();
1097         }
1098         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1099
1100         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1101             !req->rq_ctx_init)
1102                 req->rq_rep_swab_mask = 0;
1103         RETURN(rc);
1104 }
1105
1106 /**
1107  * Used by ptlrpc client, to perform security transformation upon the reply
1108  * message of \a req. After return successfully, req->rq_repmsg points to
1109  * the reply message in clear text.
1110  *
1111  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1112  * going to change.
1113  */
1114 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1115 {
1116         LASSERT(req->rq_repbuf);
1117         LASSERT(req->rq_repdata == NULL);
1118         LASSERT(req->rq_repmsg == NULL);
1119         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1120
1121         if (req->rq_reply_off == 0 &&
1122             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1123                 CERROR("real reply with offset 0\n");
1124                 return -EPROTO;
1125         }
1126
1127         if (req->rq_reply_off % 8 != 0) {
1128                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1129                 return -EPROTO;
1130         }
1131
1132         req->rq_repdata = (struct lustre_msg *)
1133                                 (req->rq_repbuf + req->rq_reply_off);
1134         req->rq_repdata_len = req->rq_nob_received;
1135
1136         return do_cli_unwrap_reply(req);
1137 }
1138
1139 /**
1140  * Used by ptlrpc client, to perform security transformation upon the early
1141  * reply message of \a req. We expect the rq_reply_off is 0, and
1142  * rq_nob_received is the early reply size.
1143  *
1144  * Because the receive buffer might be still posted, the reply data might be
1145  * changed at any time, no matter we're holding rq_lock or not. For this reason
1146  * we allocate a separate ptlrpc_request and reply buffer for early reply
1147  * processing.
1148  *
1149  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1150  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1151  * \a *req_ret to release it.
1152  * \retval -ev error number, and \a req_ret will not be set.
1153  */
1154 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1155                                    struct ptlrpc_request **req_ret)
1156 {
1157         struct ptlrpc_request *early_req;
1158         char *early_buf;
1159         int early_bufsz, early_size;
1160         int rc;
1161
1162         ENTRY;
1163
1164         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1165         if (early_req == NULL)
1166                 RETURN(-ENOMEM);
1167
1168         ptlrpc_cli_req_init(early_req);
1169
1170         early_size = req->rq_nob_received;
1171         early_bufsz = size_roundup_power2(early_size);
1172         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1173         if (early_buf == NULL)
1174                 GOTO(err_req, rc = -ENOMEM);
1175
1176         /* sanity checkings and copy data out, do it inside spinlock */
1177         spin_lock(&req->rq_lock);
1178
1179         if (req->rq_replied) {
1180                 spin_unlock(&req->rq_lock);
1181                 GOTO(err_buf, rc = -EALREADY);
1182         }
1183
1184         LASSERT(req->rq_repbuf);
1185         LASSERT(req->rq_repdata == NULL);
1186         LASSERT(req->rq_repmsg == NULL);
1187
1188         if (req->rq_reply_off != 0) {
1189                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1190                 spin_unlock(&req->rq_lock);
1191                 GOTO(err_buf, rc = -EPROTO);
1192         }
1193
1194         if (req->rq_nob_received != early_size) {
1195                 /* even another early arrived the size should be the same */
1196                 CERROR("data size has changed from %u to %u\n",
1197                        early_size, req->rq_nob_received);
1198                 spin_unlock(&req->rq_lock);
1199                 GOTO(err_buf, rc = -EINVAL);
1200         }
1201
1202         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1203                 CERROR("early reply length %d too small\n",
1204                        req->rq_nob_received);
1205                 spin_unlock(&req->rq_lock);
1206                 GOTO(err_buf, rc = -EALREADY);
1207         }
1208
1209         memcpy(early_buf, req->rq_repbuf, early_size);
1210         spin_unlock(&req->rq_lock);
1211
1212         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1213         early_req->rq_flvr = req->rq_flvr;
1214         early_req->rq_repbuf = early_buf;
1215         early_req->rq_repbuf_len = early_bufsz;
1216         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1217         early_req->rq_repdata_len = early_size;
1218         early_req->rq_early = 1;
1219         early_req->rq_reqmsg = req->rq_reqmsg;
1220
1221         rc = do_cli_unwrap_reply(early_req);
1222         if (rc) {
1223                 DEBUG_REQ(D_ADAPTTO, early_req,
1224                           "unwrap early reply: rc = %d", rc);
1225                 GOTO(err_ctx, rc);
1226         }
1227
1228         LASSERT(early_req->rq_repmsg);
1229         *req_ret = early_req;
1230         RETURN(0);
1231
1232 err_ctx:
1233         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1234 err_buf:
1235         OBD_FREE_LARGE(early_buf, early_bufsz);
1236 err_req:
1237         ptlrpc_request_cache_free(early_req);
1238         RETURN(rc);
1239 }
1240
1241 /**
1242  * Used by ptlrpc client, to release a processed early reply \a early_req.
1243  *
1244  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1245  */
1246 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1247 {
1248         LASSERT(early_req->rq_repbuf);
1249         LASSERT(early_req->rq_repdata);
1250         LASSERT(early_req->rq_repmsg);
1251
1252         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1253         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1254         ptlrpc_request_cache_free(early_req);
1255 }
1256
1257 /**************************************************
1258  * sec ID                                         *
1259  **************************************************/
1260
1261 /*
1262  * "fixed" sec (e.g. null) use sec_id < 0
1263  */
1264 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1265
1266 int sptlrpc_get_next_secid(void)
1267 {
1268         return atomic_inc_return(&sptlrpc_sec_id);
1269 }
1270 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1271
1272 /*
1273  * client side high-level security APIs
1274  */
1275
1276 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1277                                    int grace, int force)
1278 {
1279         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1280
1281         LASSERT(policy->sp_cops);
1282         LASSERT(policy->sp_cops->flush_ctx_cache);
1283
1284         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1285 }
1286
1287 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1288 {
1289         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1290
1291         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1292         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1293         LASSERT(policy->sp_cops->destroy_sec);
1294
1295         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1296
1297         policy->sp_cops->destroy_sec(sec);
1298         sptlrpc_policy_put(policy);
1299 }
1300
1301 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1302 {
1303         sec_cop_destroy_sec(sec);
1304 }
1305 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1306
1307 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1308 {
1309         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1310
1311         if (sec->ps_policy->sp_cops->kill_sec) {
1312                 sec->ps_policy->sp_cops->kill_sec(sec);
1313
1314                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1315         }
1316 }
1317
1318 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1319 {
1320         if (sec)
1321                 atomic_inc(&sec->ps_refcount);
1322
1323         return sec;
1324 }
1325 EXPORT_SYMBOL(sptlrpc_sec_get);
1326
1327 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1328 {
1329         if (sec) {
1330                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1331
1332                 if (atomic_dec_and_test(&sec->ps_refcount)) {
1333                         sptlrpc_gc_del_sec(sec);
1334                         sec_cop_destroy_sec(sec);
1335                 }
1336         }
1337 }
1338 EXPORT_SYMBOL(sptlrpc_sec_put);
1339
1340 /*
1341  * policy module is responsible for taking refrence of import
1342  */
1343 static
1344 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1345                                        struct ptlrpc_svc_ctx *svc_ctx,
1346                                        struct sptlrpc_flavor *sf,
1347                                        enum lustre_sec_part sp)
1348 {
1349         struct ptlrpc_sec_policy *policy;
1350         struct ptlrpc_sec *sec;
1351         char str[32];
1352
1353         ENTRY;
1354
1355         if (svc_ctx) {
1356                 LASSERT(imp->imp_dlm_fake == 1);
1357
1358                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1359                        imp->imp_obd->obd_type->typ_name,
1360                        imp->imp_obd->obd_name,
1361                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1362
1363                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1364                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1365         } else {
1366                 LASSERT(imp->imp_dlm_fake == 0);
1367
1368                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1369                        imp->imp_obd->obd_type->typ_name,
1370                        imp->imp_obd->obd_name,
1371                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1372
1373                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1374                 if (!policy) {
1375                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1376                         RETURN(NULL);
1377                 }
1378         }
1379
1380         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1381         if (sec) {
1382                 atomic_inc(&sec->ps_refcount);
1383
1384                 sec->ps_part = sp;
1385
1386                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1387                         sptlrpc_gc_add_sec(sec);
1388         } else {
1389                 sptlrpc_policy_put(policy);
1390         }
1391
1392         RETURN(sec);
1393 }
1394
1395 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1396 {
1397         struct ptlrpc_sec *sec;
1398
1399         read_lock(&imp->imp_sec_lock);
1400         sec = sptlrpc_sec_get(imp->imp_sec);
1401         read_unlock(&imp->imp_sec_lock);
1402
1403         return sec;
1404 }
1405 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1406
1407 static void sptlrpc_import_sec_install(struct obd_import *imp,
1408                                        struct ptlrpc_sec *sec)
1409 {
1410         struct ptlrpc_sec *old_sec;
1411
1412         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1413
1414         write_lock(&imp->imp_sec_lock);
1415         old_sec = imp->imp_sec;
1416         imp->imp_sec = sec;
1417         write_unlock(&imp->imp_sec_lock);
1418
1419         if (old_sec) {
1420                 sptlrpc_sec_kill(old_sec);
1421
1422                 /* balance the ref taken by this import */
1423                 sptlrpc_sec_put(old_sec);
1424         }
1425 }
1426
1427 static inline
1428 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1429 {
1430         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1431 }
1432
1433 static inline
1434 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1435 {
1436         *dst = *src;
1437 }
1438
1439 /**
1440  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1441  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1442  *
1443  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1444  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1445  */
1446 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1447                              struct ptlrpc_svc_ctx *svc_ctx,
1448                              struct sptlrpc_flavor *flvr)
1449 {
1450         struct ptlrpc_connection *conn;
1451         struct sptlrpc_flavor sf;
1452         struct ptlrpc_sec *sec, *newsec;
1453         enum lustre_sec_part sp;
1454         char str[24];
1455         int rc = 0;
1456
1457         ENTRY;
1458
1459         might_sleep();
1460
1461         if (imp == NULL)
1462                 RETURN(0);
1463
1464         conn = imp->imp_connection;
1465
1466         if (svc_ctx == NULL) {
1467                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1468                 /*
1469                  * normal import, determine flavor from rule set, except
1470                  * for mgc the flavor is predetermined.
1471                  */
1472                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1473                         sf = cliobd->cl_flvr_mgc;
1474                 else
1475                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1476                                                    cliobd->cl_sp_to,
1477                                                    &cliobd->cl_target_uuid,
1478                                                    conn->c_self, &sf);
1479
1480                 sp = imp->imp_obd->u.cli.cl_sp_me;
1481         } else {
1482                 /* reverse import, determine flavor from incoming reqeust */
1483                 sf = *flvr;
1484
1485                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1486                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1487                                       PTLRPC_SEC_FL_ROOTONLY;
1488
1489                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1490         }
1491
1492         sec = sptlrpc_import_sec_ref(imp);
1493         if (sec) {
1494                 char str2[24];
1495
1496                 if (flavor_equal(&sf, &sec->ps_flvr))
1497                         GOTO(out, rc);
1498
1499                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1500                        imp->imp_obd->obd_name,
1501                        obd_uuid2str(&conn->c_remote_uuid),
1502                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1503                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1504         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1505                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1506                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1507                        imp->imp_obd->obd_name,
1508                        obd_uuid2str(&conn->c_remote_uuid),
1509                        LNET_NIDNET(conn->c_self),
1510                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1511         }
1512
1513         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1514         if (newsec) {
1515                 sptlrpc_import_sec_install(imp, newsec);
1516         } else {
1517                 CERROR("import %s->%s: failed to create new sec\n",
1518                        imp->imp_obd->obd_name,
1519                        obd_uuid2str(&conn->c_remote_uuid));
1520                 rc = -EPERM;
1521         }
1522
1523 out:
1524         sptlrpc_sec_put(sec);
1525         RETURN(rc);
1526 }
1527
1528 void sptlrpc_import_sec_put(struct obd_import *imp)
1529 {
1530         if (imp->imp_sec) {
1531                 sptlrpc_sec_kill(imp->imp_sec);
1532
1533                 sptlrpc_sec_put(imp->imp_sec);
1534                 imp->imp_sec = NULL;
1535         }
1536 }
1537
1538 static void import_flush_ctx_common(struct obd_import *imp,
1539                                     uid_t uid, int grace, int force)
1540 {
1541         struct ptlrpc_sec *sec;
1542
1543         if (imp == NULL)
1544                 return;
1545
1546         sec = sptlrpc_import_sec_ref(imp);
1547         if (sec == NULL)
1548                 return;
1549
1550         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1551         sptlrpc_sec_put(sec);
1552 }
1553
1554 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1555 {
1556         /*
1557          * it's important to use grace mode, see explain in
1558          * sptlrpc_req_refresh_ctx()
1559          */
1560         import_flush_ctx_common(imp, 0, 1, 1);
1561 }
1562
1563 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1564 {
1565         import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1566                                 1, 1);
1567 }
1568 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1569
1570 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1571 {
1572         import_flush_ctx_common(imp, -1, 1, 1);
1573 }
1574 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1575
1576 /**
1577  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1578  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1579  */
1580 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1581 {
1582         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1583         struct ptlrpc_sec_policy *policy;
1584         int rc;
1585
1586         LASSERT(ctx);
1587         LASSERT(ctx->cc_sec);
1588         LASSERT(ctx->cc_sec->ps_policy);
1589         LASSERT(req->rq_reqmsg == NULL);
1590         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1591
1592         policy = ctx->cc_sec->ps_policy;
1593         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1594         if (!rc) {
1595                 LASSERT(req->rq_reqmsg);
1596                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1597
1598                 /* zeroing preallocated buffer */
1599                 if (req->rq_pool)
1600                         memset(req->rq_reqmsg, 0, msgsize);
1601         }
1602
1603         return rc;
1604 }
1605
1606 /**
1607  * Used by ptlrpc client to free request buffer of \a req. After this
1608  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1609  */
1610 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1611 {
1612         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1613         struct ptlrpc_sec_policy *policy;
1614
1615         LASSERT(ctx);
1616         LASSERT(ctx->cc_sec);
1617         LASSERT(ctx->cc_sec->ps_policy);
1618         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1619
1620         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1621                 return;
1622
1623         policy = ctx->cc_sec->ps_policy;
1624         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1625         req->rq_reqmsg = NULL;
1626 }
1627
1628 /*
1629  * NOTE caller must guarantee the buffer size is enough for the enlargement
1630  */
1631 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1632                                   int segment, int newsize)
1633 {
1634         void *src, *dst;
1635         int oldsize, oldmsg_size, movesize;
1636
1637         LASSERT(segment < msg->lm_bufcount);
1638         LASSERT(msg->lm_buflens[segment] <= newsize);
1639
1640         if (msg->lm_buflens[segment] == newsize)
1641                 return;
1642
1643         /* nothing to do if we are enlarging the last segment */
1644         if (segment == msg->lm_bufcount - 1) {
1645                 msg->lm_buflens[segment] = newsize;
1646                 return;
1647         }
1648
1649         oldsize = msg->lm_buflens[segment];
1650
1651         src = lustre_msg_buf(msg, segment + 1, 0);
1652         msg->lm_buflens[segment] = newsize;
1653         dst = lustre_msg_buf(msg, segment + 1, 0);
1654         msg->lm_buflens[segment] = oldsize;
1655
1656         /* move from segment + 1 to end segment */
1657         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1658         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1659         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1660         LASSERT(movesize >= 0);
1661
1662         if (movesize)
1663                 memmove(dst, src, movesize);
1664
1665         /* note we don't clear the ares where old data live, not secret */
1666
1667         /* finally set new segment size */
1668         msg->lm_buflens[segment] = newsize;
1669 }
1670 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1671
1672 /**
1673  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1674  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1675  * preserved after the enlargement. this must be called after original request
1676  * buffer being allocated.
1677  *
1678  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1679  * so caller should refresh its local pointers if needed.
1680  */
1681 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1682                                const struct req_msg_field *field,
1683                                int newsize)
1684 {
1685         struct req_capsule *pill = &req->rq_pill;
1686         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1687         struct ptlrpc_sec_cops *cops;
1688         struct lustre_msg *msg = req->rq_reqmsg;
1689         int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1690
1691         LASSERT(ctx);
1692         LASSERT(msg);
1693         LASSERT(msg->lm_bufcount > segment);
1694         LASSERT(msg->lm_buflens[segment] <= newsize);
1695
1696         if (msg->lm_buflens[segment] == newsize)
1697                 return 0;
1698
1699         cops = ctx->cc_sec->ps_policy->sp_cops;
1700         LASSERT(cops->enlarge_reqbuf);
1701         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1702 }
1703 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1704
1705 /**
1706  * Used by ptlrpc client to allocate reply buffer of \a req.
1707  *
1708  * \note After this, req->rq_repmsg is still not accessible.
1709  */
1710 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1711 {
1712         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1713         struct ptlrpc_sec_policy *policy;
1714
1715         ENTRY;
1716
1717         LASSERT(ctx);
1718         LASSERT(ctx->cc_sec);
1719         LASSERT(ctx->cc_sec->ps_policy);
1720
1721         if (req->rq_repbuf)
1722                 RETURN(0);
1723
1724         policy = ctx->cc_sec->ps_policy;
1725         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1726 }
1727
1728 /**
1729  * Used by ptlrpc client to free reply buffer of \a req. After this
1730  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1731  */
1732 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1733 {
1734         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1735         struct ptlrpc_sec_policy *policy;
1736
1737         ENTRY;
1738
1739         LASSERT(ctx);
1740         LASSERT(ctx->cc_sec);
1741         LASSERT(ctx->cc_sec->ps_policy);
1742         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1743
1744         if (req->rq_repbuf == NULL)
1745                 return;
1746         LASSERT(req->rq_repbuf_len);
1747
1748         policy = ctx->cc_sec->ps_policy;
1749         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1750         req->rq_repmsg = NULL;
1751         EXIT;
1752 }
1753 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1754
1755 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1756                                 struct ptlrpc_cli_ctx *ctx)
1757 {
1758         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1759
1760         if (!policy->sp_cops->install_rctx)
1761                 return 0;
1762         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1763 }
1764
1765 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1766                                 struct ptlrpc_svc_ctx *ctx)
1767 {
1768         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1769
1770         if (!policy->sp_sops->install_rctx)
1771                 return 0;
1772         return policy->sp_sops->install_rctx(imp, ctx);
1773 }
1774
1775 /* Get SELinux policy info from userspace */
1776 static int sepol_helper(struct obd_import *imp)
1777 {
1778         char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1779         char *argv[] = {
1780                 [0] = "/usr/sbin/l_getsepol",
1781                 [1] = "-o",
1782                 [2] = NULL,         /* obd type */
1783                 [3] = "-n",
1784                 [4] = NULL,         /* obd name */
1785                 [5] = "-t",
1786                 [6] = mtime_str,    /* policy mtime */
1787                 [7] = "-m",
1788                 [8] = mode_str,     /* enforcing mode */
1789                 [9] = NULL
1790         };
1791         char *envp[] = {
1792                 [0] = "HOME=/",
1793                 [1] = "PATH=/sbin:/usr/sbin",
1794                 [2] = NULL
1795         };
1796         signed short ret;
1797         int rc = 0;
1798
1799         if (imp == NULL || imp->imp_obd == NULL ||
1800             imp->imp_obd->obd_type == NULL) {
1801                 rc = -EINVAL;
1802         } else {
1803                 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1804                 argv[4] = imp->imp_obd->obd_name;
1805                 spin_lock(&imp->imp_sec->ps_lock);
1806                 if (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1807                     imp->imp_sec->ps_sepol[0] == '\0') {
1808                         /* ps_sepol has not been initialized */
1809                         argv[5] = NULL;
1810                         argv[7] = NULL;
1811                 } else {
1812                         time64_t mtime_ms;
1813
1814                         mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1815                         snprintf(mtime_str, sizeof(mtime_str), "%lld",
1816                                  mtime_ms / MSEC_PER_SEC);
1817                         mode_str[0] = imp->imp_sec->ps_sepol[0];
1818                 }
1819                 spin_unlock(&imp->imp_sec->ps_lock);
1820                 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1821                 rc = ret>>8;
1822         }
1823
1824         return rc;
1825 }
1826
1827 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1828 {
1829         ktime_t checknext;
1830
1831         if (send_sepol == 0 || !selinux_is_enabled())
1832                 return 0;
1833
1834         if (send_sepol == -1)
1835                 /* send_sepol == -1 means fetch sepol status every time */
1836                 return 1;
1837
1838         spin_lock(&imp_sec->ps_lock);
1839         checknext = imp_sec->ps_sepol_checknext;
1840         spin_unlock(&imp_sec->ps_lock);
1841
1842         /* next check is too far in time, please update */
1843         if (ktime_after(checknext,
1844                         ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1845                 goto setnext;
1846
1847         if (ktime_before(ktime_get(), checknext))
1848                 /* too early to fetch sepol status */
1849                 return 0;
1850
1851 setnext:
1852         /* define new sepol_checknext time */
1853         spin_lock(&imp_sec->ps_lock);
1854         imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1855                                                 ktime_set(send_sepol, 0));
1856         spin_unlock(&imp_sec->ps_lock);
1857
1858         return 1;
1859 }
1860
1861 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1862 {
1863         struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1864         int rc = 0;
1865
1866         ENTRY;
1867
1868         (req->rq_sepol)[0] = '\0';
1869
1870 #ifndef HAVE_SELINUX
1871         if (unlikely(send_sepol != 0))
1872                 CDEBUG(D_SEC,
1873                        "Client cannot report SELinux status, it was not built against libselinux.\n");
1874         RETURN(0);
1875 #endif
1876
1877         if (send_sepol == 0 || !selinux_is_enabled())
1878                 RETURN(0);
1879
1880         if (imp_sec == NULL)
1881                 RETURN(-EINVAL);
1882
1883         /* Retrieve SELinux status info */
1884         if (sptlrpc_sepol_needs_check(imp_sec))
1885                 rc = sepol_helper(req->rq_import);
1886         if (likely(rc == 0)) {
1887                 spin_lock(&imp_sec->ps_lock);
1888                 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1889                        sizeof(req->rq_sepol));
1890                 spin_unlock(&imp_sec->ps_lock);
1891         }
1892
1893         RETURN(rc);
1894 }
1895 EXPORT_SYMBOL(sptlrpc_get_sepol);
1896
1897 /*
1898  * server side security
1899  */
1900
1901 static int flavor_allowed(struct sptlrpc_flavor *exp,
1902                           struct ptlrpc_request *req)
1903 {
1904         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1905
1906         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1907                 return 1;
1908
1909         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1910             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1911             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1912             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1913                 return 1;
1914
1915         return 0;
1916 }
1917
1918 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1919
1920 /**
1921  * Given an export \a exp, check whether the flavor of incoming \a req
1922  * is allowed by the export \a exp. Main logic is about taking care of
1923  * changing configurations. Return 0 means success.
1924  */
1925 int sptlrpc_target_export_check(struct obd_export *exp,
1926                                 struct ptlrpc_request *req)
1927 {
1928         struct sptlrpc_flavor   flavor;
1929
1930         if (exp == NULL)
1931                 return 0;
1932
1933         /*
1934          * client side export has no imp_reverse, skip
1935          * FIXME maybe we should check flavor this as well???
1936          */
1937         if (exp->exp_imp_reverse == NULL)
1938                 return 0;
1939
1940         /* don't care about ctx fini rpc */
1941         if (req->rq_ctx_fini)
1942                 return 0;
1943
1944         spin_lock(&exp->exp_lock);
1945
1946         /*
1947          * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1948          * the first req with the new flavor, then treat it as current flavor,
1949          * adapt reverse sec according to it.
1950          * note the first rpc with new flavor might not be with root ctx, in
1951          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1952          */
1953         if (unlikely(exp->exp_flvr_changed) &&
1954             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1955                 /*
1956                  * make the new flavor as "current", and old ones as
1957                  * about-to-expire
1958                  */
1959                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1960                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1961                 flavor = exp->exp_flvr_old[1];
1962                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1963                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1964                 exp->exp_flvr_old[0] = exp->exp_flvr;
1965                 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1966                                           EXP_FLVR_UPDATE_EXPIRE;
1967                 exp->exp_flvr = flavor;
1968
1969                 /* flavor change finished */
1970                 exp->exp_flvr_changed = 0;
1971                 LASSERT(exp->exp_flvr_adapt == 1);
1972
1973                 /* if it's gss, we only interested in root ctx init */
1974                 if (req->rq_auth_gss &&
1975                     !(req->rq_ctx_init &&
1976                     (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1977                     req->rq_auth_usr_ost))) {
1978                         spin_unlock(&exp->exp_lock);
1979                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1980                                req->rq_auth_gss, req->rq_ctx_init,
1981                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1982                                req->rq_auth_usr_ost);
1983                         return 0;
1984                 }
1985
1986                 exp->exp_flvr_adapt = 0;
1987                 spin_unlock(&exp->exp_lock);
1988
1989                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1990                                                 req->rq_svc_ctx, &flavor);
1991         }
1992
1993         /*
1994          * if it equals to the current flavor, we accept it, but need to
1995          * dealing with reverse sec/ctx
1996          */
1997         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1998                 /*
1999                  * most cases should return here, we only interested in
2000                  * gss root ctx init
2001                  */
2002                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2003                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2004                      !req->rq_auth_usr_ost)) {
2005                         spin_unlock(&exp->exp_lock);
2006                         return 0;
2007                 }
2008
2009                 /*
2010                  * if flavor just changed, we should not proceed, just leave
2011                  * it and current flavor will be discovered and replaced
2012                  * shortly, and let _this_ rpc pass through
2013                  */
2014                 if (exp->exp_flvr_changed) {
2015                         LASSERT(exp->exp_flvr_adapt);
2016                         spin_unlock(&exp->exp_lock);
2017                         return 0;
2018                 }
2019
2020                 if (exp->exp_flvr_adapt) {
2021                         exp->exp_flvr_adapt = 0;
2022                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2023                                exp, exp->exp_flvr.sf_rpc,
2024                                exp->exp_flvr_old[0].sf_rpc,
2025                                exp->exp_flvr_old[1].sf_rpc);
2026                         flavor = exp->exp_flvr;
2027                         spin_unlock(&exp->exp_lock);
2028
2029                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2030                                                         req->rq_svc_ctx,
2031                                                         &flavor);
2032                 } else {
2033                         CDEBUG(D_SEC,
2034                                "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2035                                exp, exp->exp_flvr.sf_rpc,
2036                                exp->exp_flvr_old[0].sf_rpc,
2037                                exp->exp_flvr_old[1].sf_rpc);
2038                         spin_unlock(&exp->exp_lock);
2039
2040                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2041                                                            req->rq_svc_ctx);
2042                 }
2043         }
2044
2045         if (exp->exp_flvr_expire[0]) {
2046                 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2047                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2048                                 CDEBUG(D_SEC,
2049                                        "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2050                                        exp, exp->exp_flvr.sf_rpc,
2051                                        exp->exp_flvr_old[0].sf_rpc,
2052                                        exp->exp_flvr_old[1].sf_rpc,
2053                                        (s64)(exp->exp_flvr_expire[0] -
2054                                              ktime_get_real_seconds()));
2055                                 spin_unlock(&exp->exp_lock);
2056                                 return 0;
2057                         }
2058                 } else {
2059                         CDEBUG(D_SEC, "mark middle expired\n");
2060                         exp->exp_flvr_expire[0] = 0;
2061                 }
2062                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2063                        exp->exp_flvr.sf_rpc,
2064                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2065                        req->rq_flvr.sf_rpc);
2066         }
2067
2068         /*
2069          * now it doesn't match the current flavor, the only chance we can
2070          * accept it is match the old flavors which is not expired.
2071          */
2072         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2073                 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2074                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2075                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2076                                        exp,
2077                                        exp->exp_flvr.sf_rpc,
2078                                        exp->exp_flvr_old[0].sf_rpc,
2079                                        exp->exp_flvr_old[1].sf_rpc,
2080                                        (s64)(exp->exp_flvr_expire[1] -
2081                                        ktime_get_real_seconds()));
2082                                 spin_unlock(&exp->exp_lock);
2083                                 return 0;
2084                         }
2085                 } else {
2086                         CDEBUG(D_SEC, "mark oldest expired\n");
2087                         exp->exp_flvr_expire[1] = 0;
2088                 }
2089                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2090                        exp, exp->exp_flvr.sf_rpc,
2091                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2092                        req->rq_flvr.sf_rpc);
2093         } else {
2094                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2095                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2096                        exp->exp_flvr_old[1].sf_rpc);
2097         }
2098
2099         spin_unlock(&exp->exp_lock);
2100
2101         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2102               exp, exp->exp_obd->obd_name,
2103               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2104               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2105               req->rq_flvr.sf_rpc,
2106               exp->exp_flvr.sf_rpc,
2107               exp->exp_flvr_old[0].sf_rpc,
2108               exp->exp_flvr_expire[0] ?
2109               (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2110               exp->exp_flvr_old[1].sf_rpc,
2111               exp->exp_flvr_expire[1] ?
2112               (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2113         return -EACCES;
2114 }
2115 EXPORT_SYMBOL(sptlrpc_target_export_check);
2116
2117 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2118                                       struct sptlrpc_rule_set *rset)
2119 {
2120         struct obd_export *exp;
2121         struct sptlrpc_flavor new_flvr;
2122
2123         LASSERT(obd);
2124
2125         spin_lock(&obd->obd_dev_lock);
2126
2127         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2128                 if (exp->exp_connection == NULL)
2129                         continue;
2130
2131                 /*
2132                  * note if this export had just been updated flavor
2133                  * (exp_flvr_changed == 1), this will override the
2134                  * previous one.
2135                  */
2136                 spin_lock(&exp->exp_lock);
2137                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2138                                              exp->exp_connection->c_peer.nid,
2139                                              &new_flvr);
2140                 if (exp->exp_flvr_changed ||
2141                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2142                         exp->exp_flvr_old[1] = new_flvr;
2143                         exp->exp_flvr_expire[1] = 0;
2144                         exp->exp_flvr_changed = 1;
2145                         exp->exp_flvr_adapt = 1;
2146
2147                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2148                                exp, sptlrpc_part2name(exp->exp_sp_peer),
2149                                exp->exp_flvr.sf_rpc,
2150                                exp->exp_flvr_old[1].sf_rpc);
2151                 }
2152                 spin_unlock(&exp->exp_lock);
2153         }
2154
2155         spin_unlock(&obd->obd_dev_lock);
2156 }
2157 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2158
2159 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2160 {
2161         /* peer's claim is unreliable unless gss is being used */
2162         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2163                 return svc_rc;
2164
2165         switch (req->rq_sp_from) {
2166         case LUSTRE_SP_CLI:
2167                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2168                         /* The below message is checked in sanity-sec test_33 */
2169                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
2170                         svc_rc = SECSVC_DROP;
2171                 }
2172                 break;
2173         case LUSTRE_SP_MDT:
2174                 if (!req->rq_auth_usr_mdt) {
2175                         /* The below message is checked in sanity-sec test_33 */
2176                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
2177                         svc_rc = SECSVC_DROP;
2178                 }
2179                 break;
2180         case LUSTRE_SP_OST:
2181                 if (!req->rq_auth_usr_ost) {
2182                         /* The below message is checked in sanity-sec test_33 */
2183                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2184                         svc_rc = SECSVC_DROP;
2185                 }
2186                 break;
2187         case LUSTRE_SP_MGS:
2188         case LUSTRE_SP_MGC:
2189                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2190                     !req->rq_auth_usr_ost) {
2191                         /* The below message is checked in sanity-sec test_33 */
2192                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2193                         svc_rc = SECSVC_DROP;
2194                 }
2195                 break;
2196         case LUSTRE_SP_ANY:
2197         default:
2198                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2199                 svc_rc = SECSVC_DROP;
2200         }
2201
2202         return svc_rc;
2203 }
2204
2205 /**
2206  * Used by ptlrpc server, to perform transformation upon request message of
2207  * incoming \a req. This must be the first thing to do with an incoming
2208  * request in ptlrpc layer.
2209  *
2210  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2211  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2212  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2213  * reply message has been prepared.
2214  * \retval SECSVC_DROP failed, this request should be dropped.
2215  */
2216 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2217 {
2218         struct ptlrpc_sec_policy *policy;
2219         struct lustre_msg *msg = req->rq_reqbuf;
2220         int rc;
2221
2222         ENTRY;
2223
2224         LASSERT(msg);
2225         LASSERT(req->rq_reqmsg == NULL);
2226         LASSERT(req->rq_repmsg == NULL);
2227         LASSERT(req->rq_svc_ctx == NULL);
2228
2229         req->rq_req_swab_mask = 0;
2230
2231         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2232         switch (rc) {
2233         case 1:
2234                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2235         case 0:
2236                 break;
2237         default:
2238                 CERROR("error unpacking request from %s x%llu\n",
2239                        libcfs_id2str(req->rq_peer), req->rq_xid);
2240                 RETURN(SECSVC_DROP);
2241         }
2242
2243         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2244         req->rq_sp_from = LUSTRE_SP_ANY;
2245         req->rq_auth_uid = -1; /* set to INVALID_UID */
2246         req->rq_auth_mapped_uid = -1;
2247
2248         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2249         if (!policy) {
2250                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2251                 RETURN(SECSVC_DROP);
2252         }
2253
2254         LASSERT(policy->sp_sops->accept);
2255         rc = policy->sp_sops->accept(req);
2256         sptlrpc_policy_put(policy);
2257         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2258         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2259
2260         /*
2261          * if it's not null flavor (which means embedded packing msg),
2262          * reset the swab mask for the comming inner msg unpacking.
2263          */
2264         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2265                 req->rq_req_swab_mask = 0;
2266
2267         /* sanity check for the request source */
2268         rc = sptlrpc_svc_check_from(req, rc);
2269         RETURN(rc);
2270 }
2271
2272 /**
2273  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2274  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2275  * a buffer of \a msglen size.
2276  */
2277 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2278 {
2279         struct ptlrpc_sec_policy *policy;
2280         struct ptlrpc_reply_state *rs;
2281         int rc;
2282
2283         ENTRY;
2284
2285         LASSERT(req->rq_svc_ctx);
2286         LASSERT(req->rq_svc_ctx->sc_policy);
2287
2288         policy = req->rq_svc_ctx->sc_policy;
2289         LASSERT(policy->sp_sops->alloc_rs);
2290
2291         rc = policy->sp_sops->alloc_rs(req, msglen);
2292         if (unlikely(rc == -ENOMEM)) {
2293                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2294
2295                 if (svcpt->scp_service->srv_max_reply_size <
2296                    msglen + sizeof(struct ptlrpc_reply_state)) {
2297                         /* Just return failure if the size is too big */
2298                         CERROR("size of message is too big (%zd), %d allowed\n",
2299                                 msglen + sizeof(struct ptlrpc_reply_state),
2300                                 svcpt->scp_service->srv_max_reply_size);
2301                         RETURN(-ENOMEM);
2302                 }
2303
2304                 /* failed alloc, try emergency pool */
2305                 rs = lustre_get_emerg_rs(svcpt);
2306                 if (rs == NULL)
2307                         RETURN(-ENOMEM);
2308
2309                 req->rq_reply_state = rs;
2310                 rc = policy->sp_sops->alloc_rs(req, msglen);
2311                 if (rc) {
2312                         lustre_put_emerg_rs(rs);
2313                         req->rq_reply_state = NULL;
2314                 }
2315         }
2316
2317         LASSERT(rc != 0 ||
2318                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2319
2320         RETURN(rc);
2321 }
2322
2323 /**
2324  * Used by ptlrpc server, to perform transformation upon reply message.
2325  *
2326  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2327  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2328  */
2329 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2330 {
2331         struct ptlrpc_sec_policy *policy;
2332         int rc;
2333
2334         ENTRY;
2335
2336         LASSERT(req->rq_svc_ctx);
2337         LASSERT(req->rq_svc_ctx->sc_policy);
2338
2339         policy = req->rq_svc_ctx->sc_policy;
2340         LASSERT(policy->sp_sops->authorize);
2341
2342         rc = policy->sp_sops->authorize(req);
2343         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2344
2345         RETURN(rc);
2346 }
2347
2348 /**
2349  * Used by ptlrpc server, to free reply_state.
2350  */
2351 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2352 {
2353         struct ptlrpc_sec_policy *policy;
2354         unsigned int prealloc;
2355
2356         ENTRY;
2357
2358         LASSERT(rs->rs_svc_ctx);
2359         LASSERT(rs->rs_svc_ctx->sc_policy);
2360
2361         policy = rs->rs_svc_ctx->sc_policy;
2362         LASSERT(policy->sp_sops->free_rs);
2363
2364         prealloc = rs->rs_prealloc;
2365         policy->sp_sops->free_rs(rs);
2366
2367         if (prealloc)
2368                 lustre_put_emerg_rs(rs);
2369         EXIT;
2370 }
2371
2372 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2373 {
2374         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2375
2376         if (ctx != NULL)
2377                 atomic_inc(&ctx->sc_refcount);
2378 }
2379
2380 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2381 {
2382         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2383
2384         if (ctx == NULL)
2385                 return;
2386
2387         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2388         if (atomic_dec_and_test(&ctx->sc_refcount)) {
2389                 if (ctx->sc_policy->sp_sops->free_ctx)
2390                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2391         }
2392         req->rq_svc_ctx = NULL;
2393 }
2394
2395 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2396 {
2397         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2398
2399         if (ctx == NULL)
2400                 return;
2401
2402         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2403         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2404                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2405 }
2406 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2407
2408 /*
2409  * bulk security
2410  */
2411
2412 /**
2413  * Perform transformation upon bulk data pointed by \a desc. This is called
2414  * before transforming the request message.
2415  */
2416 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2417                           struct ptlrpc_bulk_desc *desc)
2418 {
2419         struct ptlrpc_cli_ctx *ctx;
2420
2421         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2422
2423         if (!req->rq_pack_bulk)
2424                 return 0;
2425
2426         ctx = req->rq_cli_ctx;
2427         if (ctx->cc_ops->wrap_bulk)
2428                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2429         return 0;
2430 }
2431 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2432
2433 /**
2434  * This is called after unwrap the reply message.
2435  * return nob of actual plain text size received, or error code.
2436  */
2437 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2438                                  struct ptlrpc_bulk_desc *desc,
2439                                  int nob)
2440 {
2441         struct ptlrpc_cli_ctx *ctx;
2442         int rc;
2443
2444         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2445
2446         if (!req->rq_pack_bulk)
2447                 return desc->bd_nob_transferred;
2448
2449         ctx = req->rq_cli_ctx;
2450         if (ctx->cc_ops->unwrap_bulk) {
2451                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2452                 if (rc < 0)
2453                         return rc;
2454         }
2455         return desc->bd_nob_transferred;
2456 }
2457 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2458
2459 /**
2460  * This is called after unwrap the reply message.
2461  * return 0 for success or error code.
2462  */
2463 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2464                                   struct ptlrpc_bulk_desc *desc)
2465 {
2466         struct ptlrpc_cli_ctx *ctx;
2467         int rc;
2468
2469         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2470
2471         if (!req->rq_pack_bulk)
2472                 return 0;
2473
2474         ctx = req->rq_cli_ctx;
2475         if (ctx->cc_ops->unwrap_bulk) {
2476                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2477                 if (rc < 0)
2478                         return rc;
2479         }
2480
2481         /*
2482          * if everything is going right, nob should equals to nob_transferred.
2483          * in case of privacy mode, nob_transferred needs to be adjusted.
2484          */
2485         if (desc->bd_nob != desc->bd_nob_transferred) {
2486                 CERROR("nob %d doesn't match transferred nob %d\n",
2487                        desc->bd_nob, desc->bd_nob_transferred);
2488                 return -EPROTO;
2489         }
2490
2491         return 0;
2492 }
2493 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2494
2495 #ifdef HAVE_SERVER_SUPPORT
2496 /**
2497  * Performe transformation upon outgoing bulk read.
2498  */
2499 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2500                           struct ptlrpc_bulk_desc *desc)
2501 {
2502         struct ptlrpc_svc_ctx *ctx;
2503
2504         LASSERT(req->rq_bulk_read);
2505
2506         if (!req->rq_pack_bulk)
2507                 return 0;
2508
2509         ctx = req->rq_svc_ctx;
2510         if (ctx->sc_policy->sp_sops->wrap_bulk)
2511                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2512
2513         return 0;
2514 }
2515 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2516
2517 /**
2518  * Performe transformation upon incoming bulk write.
2519  */
2520 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2521                             struct ptlrpc_bulk_desc *desc)
2522 {
2523         struct ptlrpc_svc_ctx *ctx;
2524         int rc;
2525
2526         LASSERT(req->rq_bulk_write);
2527
2528         /*
2529          * if it's in privacy mode, transferred should >= expected; otherwise
2530          * transferred should == expected.
2531          */
2532         if (desc->bd_nob_transferred < desc->bd_nob ||
2533             (desc->bd_nob_transferred > desc->bd_nob &&
2534              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2535              SPTLRPC_BULK_SVC_PRIV)) {
2536                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2537                           desc->bd_nob_transferred, desc->bd_nob);
2538                 return -ETIMEDOUT;
2539         }
2540
2541         if (!req->rq_pack_bulk)
2542                 return 0;
2543
2544         ctx = req->rq_svc_ctx;
2545         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2546                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2547                 if (rc)
2548                         CERROR("error unwrap bulk: %d\n", rc);
2549         }
2550
2551         /* return 0 to allow reply be sent */
2552         return 0;
2553 }
2554 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2555
2556 /**
2557  * Prepare buffers for incoming bulk write.
2558  */
2559 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2560                           struct ptlrpc_bulk_desc *desc)
2561 {
2562         struct ptlrpc_svc_ctx *ctx;
2563
2564         LASSERT(req->rq_bulk_write);
2565
2566         if (!req->rq_pack_bulk)
2567                 return 0;
2568
2569         ctx = req->rq_svc_ctx;
2570         if (ctx->sc_policy->sp_sops->prep_bulk)
2571                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2572
2573         return 0;
2574 }
2575 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2576
2577 #endif /* HAVE_SERVER_SUPPORT */
2578
2579 /*
2580  * user descriptor helpers
2581  */
2582
2583 int sptlrpc_current_user_desc_size(void)
2584 {
2585         int ngroups;
2586
2587         ngroups = current_cred()->group_info->ngroups;
2588
2589         if (ngroups > LUSTRE_MAX_GROUPS)
2590                 ngroups = LUSTRE_MAX_GROUPS;
2591         return sptlrpc_user_desc_size(ngroups);
2592 }
2593 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2594
2595 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2596 {
2597         struct ptlrpc_user_desc *pud;
2598         int ngroups;
2599
2600         pud = lustre_msg_buf(msg, offset, 0);
2601
2602         pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2603         pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2604         pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2605         pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2606         pud->pud_cap = cfs_curproc_cap_pack();
2607         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2608
2609         task_lock(current);
2610         ngroups = current_cred()->group_info->ngroups;
2611         if (pud->pud_ngroups > ngroups)
2612                 pud->pud_ngroups = ngroups;
2613 #ifdef HAVE_GROUP_INFO_GID
2614         memcpy(pud->pud_groups, current_cred()->group_info->gid,
2615                pud->pud_ngroups * sizeof(__u32));
2616 #else /* !HAVE_GROUP_INFO_GID */
2617         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2618                pud->pud_ngroups * sizeof(__u32));
2619 #endif /* HAVE_GROUP_INFO_GID */
2620         task_unlock(current);
2621
2622         return 0;
2623 }
2624 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2625
2626 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2627 {
2628         struct ptlrpc_user_desc *pud;
2629         int i;
2630
2631         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2632         if (!pud)
2633                 return -EINVAL;
2634
2635         if (swabbed) {
2636                 __swab32s(&pud->pud_uid);
2637                 __swab32s(&pud->pud_gid);
2638                 __swab32s(&pud->pud_fsuid);
2639                 __swab32s(&pud->pud_fsgid);
2640                 __swab32s(&pud->pud_cap);
2641                 __swab32s(&pud->pud_ngroups);
2642         }
2643
2644         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2645                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2646                 return -EINVAL;
2647         }
2648
2649         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2650             msg->lm_buflens[offset]) {
2651                 CERROR("%u groups are claimed but bufsize only %u\n",
2652                        pud->pud_ngroups, msg->lm_buflens[offset]);
2653                 return -EINVAL;
2654         }
2655
2656         if (swabbed) {
2657                 for (i = 0; i < pud->pud_ngroups; i++)
2658                         __swab32s(&pud->pud_groups[i]);
2659         }
2660
2661         return 0;
2662 }
2663 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2664
2665 /*
2666  * misc helpers
2667  */
2668
2669 const char *sec2target_str(struct ptlrpc_sec *sec)
2670 {
2671         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2672                 return "*";
2673         if (sec_is_reverse(sec))
2674                 return "c";
2675         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2676 }
2677 EXPORT_SYMBOL(sec2target_str);
2678
2679 /*
2680  * return true if the bulk data is protected
2681  */
2682 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2683 {
2684         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2685         case SPTLRPC_BULK_SVC_INTG:
2686         case SPTLRPC_BULK_SVC_PRIV:
2687                 return 1;
2688         default:
2689                 return 0;
2690         }
2691 }
2692 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2693
2694 /*
2695  * crypto API helper/alloc blkciper
2696  */
2697
2698 /*
2699  * initialize/finalize
2700  */
2701
2702 int sptlrpc_init(void)
2703 {
2704         int rc;
2705
2706         rwlock_init(&policy_lock);
2707
2708         rc = sptlrpc_gc_init();
2709         if (rc)
2710                 goto out;
2711
2712         rc = sptlrpc_conf_init();
2713         if (rc)
2714                 goto out_gc;
2715
2716         rc = sptlrpc_enc_pool_init();
2717         if (rc)
2718                 goto out_conf;
2719
2720         rc = sptlrpc_null_init();
2721         if (rc)
2722                 goto out_pool;
2723
2724         rc = sptlrpc_plain_init();
2725         if (rc)
2726                 goto out_null;
2727
2728         rc = sptlrpc_lproc_init();
2729         if (rc)
2730                 goto out_plain;
2731
2732         return 0;
2733
2734 out_plain:
2735         sptlrpc_plain_fini();
2736 out_null:
2737         sptlrpc_null_fini();
2738 out_pool:
2739         sptlrpc_enc_pool_fini();
2740 out_conf:
2741         sptlrpc_conf_fini();
2742 out_gc:
2743         sptlrpc_gc_fini();
2744 out:
2745         return rc;
2746 }
2747
2748 void sptlrpc_fini(void)
2749 {
2750         sptlrpc_lproc_fini();
2751         sptlrpc_plain_fini();
2752         sptlrpc_null_fini();
2753         sptlrpc_enc_pool_fini();
2754         sptlrpc_conf_fini();
2755         sptlrpc_gc_fini();
2756 }