lustre/ptlrpc/sec.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/ptlrpc/sec.c
  33  *
  34  * Author: Eric Mei <ericm@clusterfs.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_SEC
  38
  39 #include <linux/user_namespace.h>
  40 #include <linux/uidgid.h>
  41 #include <linux/crypto.h>
  42 #include <linux/key.h>
  43
  44 #include <libcfs/libcfs.h>
  45 #include <obd.h>
  46 #include <obd_class.h>
  47 #include <obd_support.h>
  48 #include <lustre_net.h>
  49 #include <lustre_import.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_sec.h>
  52
  53 #include "ptlrpc_internal.h"
  54
  55 static int send_sepol;
  56 module_param(send_sepol, int, 0644);
  57 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
  58
  59 /*
  60  * policy registers
  61  */
  62
  63 static rwlock_t policy_lock;
  64 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  65         NULL,
  66 };
  67
  68 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  69 {
  70         __u16 number = policy->sp_policy;
  71
  72         LASSERT(policy->sp_name);
  73         LASSERT(policy->sp_cops);
  74         LASSERT(policy->sp_sops);
  75
  76         if (number >= SPTLRPC_POLICY_MAX)
  77                 return -EINVAL;
  78
  79         write_lock(&policy_lock);
  80         if (unlikely(policies[number])) {
  81                 write_unlock(&policy_lock);
  82                 return -EALREADY;
  83         }
  84         policies[number] = policy;
  85         write_unlock(&policy_lock);
  86
  87         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  88         return 0;
  89 }
  90 EXPORT_SYMBOL(sptlrpc_register_policy);
  91
  92 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  93 {
  94         __u16 number = policy->sp_policy;
  95
  96         LASSERT(number < SPTLRPC_POLICY_MAX);
  97
  98         write_lock(&policy_lock);
  99         if (unlikely(policies[number] == NULL)) {
 100                 write_unlock(&policy_lock);
 101                 CERROR("%s: already unregistered\n", policy->sp_name);
 102                 return -EINVAL;
 103         }
 104
 105         LASSERT(policies[number] == policy);
 106         policies[number] = NULL;
 107         write_unlock(&policy_lock);
 108
 109         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 110         return 0;
 111 }
 112 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 113
 114 static
 115 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
 116 {
 117         static DEFINE_MUTEX(load_mutex);
 118         static atomic_t           loaded = ATOMIC_INIT(0);
 119         struct ptlrpc_sec_policy *policy;
 120         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 121         __u16                     flag = 0;
 122
 123         if (number >= SPTLRPC_POLICY_MAX)
 124                 return NULL;
 125
 126         while (1) {
 127                 read_lock(&policy_lock);
 128                 policy = policies[number];
 129                 if (policy && !try_module_get(policy->sp_owner))
 130                         policy = NULL;
 131                 if (policy == NULL)
 132                         flag = atomic_read(&loaded);
 133                 read_unlock(&policy_lock);
 134
 135                 if (policy != NULL || flag != 0 ||
 136                     number != SPTLRPC_POLICY_GSS)
 137                         break;
 138
 139                 /* try to load gss module, once */
 140                 mutex_lock(&load_mutex);
 141                 if (atomic_read(&loaded) == 0) {
 142                         if (request_module("ptlrpc_gss") == 0)
 143                                 CDEBUG(D_SEC,
 144                                        "module ptlrpc_gss loaded on demand\n");
 145                         else
 146                                 CERROR("Unable to load module ptlrpc_gss\n");
 147
 148                         atomic_set(&loaded, 1);
 149                 }
 150                 mutex_unlock(&load_mutex);
 151         }
 152
 153         return policy;
 154 }
 155
 156 __u32 sptlrpc_name2flavor_base(const char *name)
 157 {
 158         if (!strcmp(name, "null"))
 159                 return SPTLRPC_FLVR_NULL;
 160         if (!strcmp(name, "plain"))
 161                 return SPTLRPC_FLVR_PLAIN;
 162         if (!strcmp(name, "gssnull"))
 163                 return SPTLRPC_FLVR_GSSNULL;
 164         if (!strcmp(name, "krb5n"))
 165                 return SPTLRPC_FLVR_KRB5N;
 166         if (!strcmp(name, "krb5a"))
 167                 return SPTLRPC_FLVR_KRB5A;
 168         if (!strcmp(name, "krb5i"))
 169                 return SPTLRPC_FLVR_KRB5I;
 170         if (!strcmp(name, "krb5p"))
 171                 return SPTLRPC_FLVR_KRB5P;
 172         if (!strcmp(name, "skn"))
 173                 return SPTLRPC_FLVR_SKN;
 174         if (!strcmp(name, "ska"))
 175                 return SPTLRPC_FLVR_SKA;
 176         if (!strcmp(name, "ski"))
 177                 return SPTLRPC_FLVR_SKI;
 178         if (!strcmp(name, "skpi"))
 179                 return SPTLRPC_FLVR_SKPI;
 180
 181         return SPTLRPC_FLVR_INVALID;
 182 }
 183 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 184
 185 const char *sptlrpc_flavor2name_base(__u32 flvr)
 186 {
 187         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 188
 189         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 190                 return "null";
 191         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 192                 return "plain";
 193         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
 194                 return "gssnull";
 195         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 196                 return "krb5n";
 197         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 198                 return "krb5a";
 199         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 200                 return "krb5i";
 201         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 202                 return "krb5p";
 203         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
 204                 return "skn";
 205         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
 206                 return "ska";
 207         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
 208                 return "ski";
 209         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
 210                 return "skpi";
 211
 212         CERROR("invalid wire flavor 0x%x\n", flvr);
 213         return "invalid";
 214 }
 215 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 216
 217 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 218                                char *buf, int bufsize)
 219 {
 220         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 221                 snprintf(buf, bufsize, "hash:%s",
 222                         sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 223         else
 224                 snprintf(buf, bufsize, "%s",
 225                         sptlrpc_flavor2name_base(sf->sf_rpc));
 226
 227         buf[bufsize - 1] = '\0';
 228         return buf;
 229 }
 230 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 231
 232 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 233 {
 234         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 235
 236         /*
 237          * currently we don't support customized bulk specification for
 238          * flavors other than plain
 239          */
 240         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 241                 char bspec[16];
 242
 243                 bspec[0] = '-';
 244                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 245                 strncat(buf, bspec, bufsize);
 246         }
 247
 248         buf[bufsize - 1] = '\0';
 249         return buf;
 250 }
 251 EXPORT_SYMBOL(sptlrpc_flavor2name);
 252
 253 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 254 {
 255         buf[0] = '\0';
 256
 257         if (flags & PTLRPC_SEC_FL_REVERSE)
 258                 strlcat(buf, "reverse,", bufsize);
 259         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 260                 strlcat(buf, "rootonly,", bufsize);
 261         if (flags & PTLRPC_SEC_FL_UDESC)
 262                 strlcat(buf, "udesc,", bufsize);
 263         if (flags & PTLRPC_SEC_FL_BULK)
 264                 strlcat(buf, "bulk,", bufsize);
 265         if (buf[0] == '\0')
 266                 strlcat(buf, "-,", bufsize);
 267
 268         return buf;
 269 }
 270 EXPORT_SYMBOL(sptlrpc_secflags2str);
 271
 272 /*
 273  * client context APIs
 274  */
 275
 276 static
 277 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 278 {
 279         struct vfs_cred vcred;
 280         int create = 1, remove_dead = 1;
 281
 282         LASSERT(sec);
 283         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 284
 285         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 286                                      PTLRPC_SEC_FL_ROOTONLY)) {
 287                 vcred.vc_uid = 0;
 288                 vcred.vc_gid = 0;
 289                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 290                         create = 0;
 291                         remove_dead = 0;
 292                 }
 293         } else {
 294                 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
 295                 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
 296         }
 297
 298         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
 299                                                    remove_dead);
 300 }
 301
 302 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 303 {
 304         atomic_inc(&ctx->cc_refcount);
 305         return ctx;
 306 }
 307 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 308
 309 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 310 {
 311         struct ptlrpc_sec *sec = ctx->cc_sec;
 312
 313         LASSERT(sec);
 314         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 315
 316         if (!atomic_dec_and_test(&ctx->cc_refcount))
 317                 return;
 318
 319         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 320 }
 321 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 322
 323 /**
 324  * Expire the client context immediately.
 325  *
 326  * \pre Caller must hold at least 1 reference on the \a ctx.
 327  */
 328 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 329 {
 330         LASSERT(ctx->cc_ops->die);
 331         ctx->cc_ops->die(ctx, 0);
 332 }
 333 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 334
 335 /**
 336  * To wake up the threads who are waiting for this client context. Called
 337  * after some status change happened on \a ctx.
 338  */
 339 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 340 {
 341         struct ptlrpc_request *req, *next;
 342
 343         spin_lock(&ctx->cc_lock);
 344         list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 345                                      rq_ctx_chain) {
 346                 list_del_init(&req->rq_ctx_chain);
 347                 ptlrpc_client_wake_req(req);
 348         }
 349         spin_unlock(&ctx->cc_lock);
 350 }
 351 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 352
 353 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 354 {
 355         LASSERT(ctx->cc_ops);
 356
 357         if (ctx->cc_ops->display == NULL)
 358                 return 0;
 359
 360         return ctx->cc_ops->display(ctx, buf, bufsize);
 361 }
 362
 363 static int import_sec_check_expire(struct obd_import *imp)
 364 {
 365         int adapt = 0;
 366
 367         write_lock(&imp->imp_sec_lock);
 368         if (imp->imp_sec_expire &&
 369             imp->imp_sec_expire < ktime_get_real_seconds()) {
 370                 adapt = 1;
 371                 imp->imp_sec_expire = 0;
 372         }
 373         write_unlock(&imp->imp_sec_lock);
 374
 375         if (!adapt)
 376                 return 0;
 377
 378         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 379         return sptlrpc_import_sec_adapt(imp, NULL, NULL);
 380 }
 381
 382 /**
 383  * Get and validate the client side ptlrpc security facilities from
 384  * \a imp. There is a race condition on client reconnect when the import is
 385  * being destroyed while there are outstanding client bound requests. In
 386  * this case do not output any error messages if import secuity is not
 387  * found.
 388  *
 389  * \param[in] imp obd import associated with client
 390  * \param[out] sec client side ptlrpc security
 391  *
 392  * \retval 0 if security retrieved successfully
 393  * \retval -ve errno if there was a problem
 394  */
 395 static int import_sec_validate_get(struct obd_import *imp,
 396                                    struct ptlrpc_sec **sec)
 397 {
 398         int rc;
 399
 400         if (unlikely(imp->imp_sec_expire)) {
 401                 rc = import_sec_check_expire(imp);
 402                 if (rc)
 403                         return rc;
 404         }
 405
 406         *sec = sptlrpc_import_sec_ref(imp);
 407         if (*sec == NULL) {
 408                 CERROR("import %p (%s) with no sec\n",
 409                         imp, ptlrpc_import_state_name(imp->imp_state));
 410                 return -EACCES;
 411         }
 412
 413         if (unlikely((*sec)->ps_dying)) {
 414                 CERROR("attempt to use dying sec %p\n", sec);
 415                 sptlrpc_sec_put(*sec);
 416                 return -EACCES;
 417         }
 418
 419         return 0;
 420 }
 421
 422 /**
 423  * Given a \a req, find or allocate an appropriate context for it.
 424  * \pre req->rq_cli_ctx == NULL.
 425  *
 426  * \retval 0 succeed, and req->rq_cli_ctx is set.
 427  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 428  */
 429 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 430 {
 431         struct obd_import *imp = req->rq_import;
 432         struct ptlrpc_sec *sec;
 433         int rc;
 434
 435         ENTRY;
 436
 437         LASSERT(!req->rq_cli_ctx);
 438         LASSERT(imp);
 439
 440         rc = import_sec_validate_get(imp, &sec);
 441         if (rc)
 442                 RETURN(rc);
 443
 444         req->rq_cli_ctx = get_my_ctx(sec);
 445
 446         sptlrpc_sec_put(sec);
 447
 448         if (!req->rq_cli_ctx) {
 449                 CERROR("req %p: fail to get context\n", req);
 450                 RETURN(-ECONNREFUSED);
 451         }
 452
 453         RETURN(0);
 454 }
 455
 456 /**
 457  * Drop the context for \a req.
 458  * \pre req->rq_cli_ctx != NULL.
 459  * \post req->rq_cli_ctx == NULL.
 460  *
 461  * If \a sync == 0, this function should return quickly without sleep;
 462  * otherwise it might trigger and wait for the whole process of sending
 463  * an context-destroying rpc to server.
 464  */
 465 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 466 {
 467         ENTRY;
 468
 469         LASSERT(req);
 470         LASSERT(req->rq_cli_ctx);
 471
 472         /*
 473          * request might be asked to release earlier while still
 474          * in the context waiting list.
 475          */
 476         if (!list_empty(&req->rq_ctx_chain)) {
 477                 spin_lock(&req->rq_cli_ctx->cc_lock);
 478                 list_del_init(&req->rq_ctx_chain);
 479                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 480         }
 481
 482         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 483         req->rq_cli_ctx = NULL;
 484         EXIT;
 485 }
 486
 487 static
 488 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 489                            struct ptlrpc_cli_ctx *oldctx,
 490                            struct ptlrpc_cli_ctx *newctx)
 491 {
 492         struct sptlrpc_flavor   old_flvr;
 493         char *reqmsg = NULL; /* to workaround old gcc */
 494         int reqmsg_size;
 495         int rc = 0;
 496
 497         LASSERT(req->rq_reqmsg);
 498         LASSERT(req->rq_reqlen);
 499         LASSERT(req->rq_replen);
 500
 501         CDEBUG(D_SEC,
 502                "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
 503                req, oldctx, oldctx->cc_vcred.vc_uid,
 504                sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
 505                sec2target_str(newctx->cc_sec), oldctx->cc_sec,
 506                oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
 507                newctx->cc_sec->ps_policy->sp_name);
 508
 509         /* save flavor */
 510         old_flvr = req->rq_flvr;
 511
 512         /* save request message */
 513         reqmsg_size = req->rq_reqlen;
 514         if (reqmsg_size != 0) {
 515                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 516                 if (reqmsg == NULL)
 517                         return -ENOMEM;
 518                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 519         }
 520
 521         /* release old req/rep buf */
 522         req->rq_cli_ctx = oldctx;
 523         sptlrpc_cli_free_reqbuf(req);
 524         sptlrpc_cli_free_repbuf(req);
 525         req->rq_cli_ctx = newctx;
 526
 527         /* recalculate the flavor */
 528         sptlrpc_req_set_flavor(req, 0);
 529
 530         /*
 531          * alloc new request buffer
 532          * we don't need to alloc reply buffer here, leave it to the
 533          * rest procedure of ptlrpc
 534          */
 535         if (reqmsg_size != 0) {
 536                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 537                 if (!rc) {
 538                         LASSERT(req->rq_reqmsg);
 539                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 540                 } else {
 541                         CWARN("failed to alloc reqbuf: %d\n", rc);
 542                         req->rq_flvr = old_flvr;
 543                 }
 544
 545                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 546         }
 547         return rc;
 548 }
 549
 550 /**
 551  * If current context of \a req is dead somehow, e.g. we just switched flavor
 552  * thus marked original contexts dead, we'll find a new context for it. if
 553  * no switch is needed, \a req will end up with the same context.
 554  *
 555  * \note a request must have a context, to keep other parts of code happy.
 556  * In any case of failure during the switching, we must restore the old one.
 557  */
 558 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 559 {
 560         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 561         struct ptlrpc_cli_ctx *newctx;
 562         int rc;
 563
 564         ENTRY;
 565
 566         LASSERT(oldctx);
 567
 568         sptlrpc_cli_ctx_get(oldctx);
 569         sptlrpc_req_put_ctx(req, 0);
 570
 571         rc = sptlrpc_req_get_ctx(req);
 572         if (unlikely(rc)) {
 573                 LASSERT(!req->rq_cli_ctx);
 574
 575                 /* restore old ctx */
 576                 req->rq_cli_ctx = oldctx;
 577                 RETURN(rc);
 578         }
 579
 580         newctx = req->rq_cli_ctx;
 581         LASSERT(newctx);
 582
 583         if (unlikely(newctx == oldctx &&
 584                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 585                 /*
 586                  * still get the old dead ctx, usually means system too busy
 587                  */
 588                 CDEBUG(D_SEC,
 589                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 590                        newctx, newctx->cc_flags);
 591
 592                 schedule_timeout_interruptible(cfs_time_seconds(1));
 593         } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
 594                             == 0)) {
 595                 /*
 596                  * new ctx not up to date yet
 597                  */
 598                 CDEBUG(D_SEC,
 599                        "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
 600                        newctx, newctx->cc_flags);
 601         } else {
 602                 /*
 603                  * it's possible newctx == oldctx if we're switching
 604                  * subflavor with the same sec.
 605                  */
 606                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 607                 if (rc) {
 608                         /* restore old ctx */
 609                         sptlrpc_req_put_ctx(req, 0);
 610                         req->rq_cli_ctx = oldctx;
 611                         RETURN(rc);
 612                 }
 613
 614                 LASSERT(req->rq_cli_ctx == newctx);
 615         }
 616
 617         sptlrpc_cli_ctx_put(oldctx, 1);
 618         RETURN(0);
 619 }
 620 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 621
 622 static
 623 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 624 {
 625         if (cli_ctx_is_refreshed(ctx))
 626                 return 1;
 627         return 0;
 628 }
 629
 630 static
 631 void ctx_refresh_interrupt(struct ptlrpc_request *req)
 632 {
 633
 634         spin_lock(&req->rq_lock);
 635         req->rq_intr = 1;
 636         spin_unlock(&req->rq_lock);
 637 }
 638
 639 static
 640 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 641 {
 642         spin_lock(&ctx->cc_lock);
 643         if (!list_empty(&req->rq_ctx_chain))
 644                 list_del_init(&req->rq_ctx_chain);
 645         spin_unlock(&ctx->cc_lock);
 646 }
 647
 648 /**
 649  * To refresh the context of \req, if it's not up-to-date.
 650  * \param timeout
 651  * - == 0: do not wait
 652  * - == MAX_SCHEDULE_TIMEOUT: wait indefinitely
 653  * - > 0: not supported
 654  *
 655  * The status of the context could be subject to be changed by other threads
 656  * at any time. We allow this race, but once we return with 0, the caller will
 657  * suppose it's uptodated and keep using it until the owning rpc is done.
 658  *
 659  * \retval 0 only if the context is uptodated.
 660  * \retval -ev error number.
 661  */
 662 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 663 {
 664         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 665         struct ptlrpc_sec *sec;
 666         int rc;
 667
 668         ENTRY;
 669
 670         LASSERT(ctx);
 671
 672         if (req->rq_ctx_init || req->rq_ctx_fini)
 673                 RETURN(0);
 674
 675         if (timeout != 0 && timeout != MAX_SCHEDULE_TIMEOUT) {
 676                 CERROR("req %p: invalid timeout %lu\n", req, timeout);
 677                 RETURN(-EINVAL);
 678         }
 679
 680         /*
 681          * during the process a request's context might change type even
 682          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 683          * everything
 684          */
 685 again:
 686         rc = import_sec_validate_get(req->rq_import, &sec);
 687         if (rc)
 688                 RETURN(rc);
 689
 690         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 691                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 692                        req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 693                 req_off_ctx_list(req, ctx);
 694                 sptlrpc_req_replace_dead_ctx(req);
 695                 ctx = req->rq_cli_ctx;
 696         }
 697         sptlrpc_sec_put(sec);
 698
 699         if (cli_ctx_is_eternal(ctx))
 700                 RETURN(0);
 701
 702         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 703                 if (ctx->cc_ops->refresh)
 704                         ctx->cc_ops->refresh(ctx);
 705         }
 706         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 707
 708         LASSERT(ctx->cc_ops->validate);
 709         if (ctx->cc_ops->validate(ctx) == 0) {
 710                 req_off_ctx_list(req, ctx);
 711                 RETURN(0);
 712         }
 713
 714         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 715                 spin_lock(&req->rq_lock);
 716                 req->rq_err = 1;
 717                 spin_unlock(&req->rq_lock);
 718                 req_off_ctx_list(req, ctx);
 719                 RETURN(-EPERM);
 720         }
 721
 722         /*
 723          * There's a subtle issue for resending RPCs, suppose following
 724          * situation:
 725          *  1. the request was sent to server.
 726          *  2. recovery was kicked start, after finished the request was
 727          *     marked as resent.
 728          *  3. resend the request.
 729          *  4. old reply from server received, we accept and verify the reply.
 730          *     this has to be success, otherwise the error will be aware
 731          *     by application.
 732          *  5. new reply from server received, dropped by LNet.
 733          *
 734          * Note the xid of old & new request is the same. We can't simply
 735          * change xid for the resent request because the server replies on
 736          * it for reply reconstruction.
 737          *
 738          * Commonly the original context should be uptodate because we
 739          * have an expiry nice time; server will keep its context because
 740          * we at least hold a ref of old context which prevent context
 741          * from destroying RPC being sent. So server still can accept the
 742          * request and finish the RPC. But if that's not the case:
 743          *  1. If server side context has been trimmed, a NO_CONTEXT will
 744          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 745          *     context by force.
 746          *  2. Current context never be refreshed, then we are fine: we
 747          *     never really send request with old context before.
 748          */
 749         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 750             unlikely(req->rq_reqmsg) &&
 751             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 752                 req_off_ctx_list(req, ctx);
 753                 RETURN(0);
 754         }
 755
 756         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 757                 req_off_ctx_list(req, ctx);
 758                 /*
 759                  * don't switch ctx if import was deactivated
 760                  */
 761                 if (req->rq_import->imp_deactive) {
 762                         spin_lock(&req->rq_lock);
 763                         req->rq_err = 1;
 764                         spin_unlock(&req->rq_lock);
 765                         RETURN(-EINTR);
 766                 }
 767
 768                 rc = sptlrpc_req_replace_dead_ctx(req);
 769                 if (rc) {
 770                         LASSERT(ctx == req->rq_cli_ctx);
 771                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 772                                req, ctx, rc);
 773                         spin_lock(&req->rq_lock);
 774                         req->rq_err = 1;
 775                         spin_unlock(&req->rq_lock);
 776                         RETURN(rc);
 777                 }
 778
 779                 ctx = req->rq_cli_ctx;
 780                 goto again;
 781         }
 782
 783         /*
 784          * Now we're sure this context is during upcall, add myself into
 785          * waiting list
 786          */
 787         spin_lock(&ctx->cc_lock);
 788         if (list_empty(&req->rq_ctx_chain))
 789                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 790         spin_unlock(&ctx->cc_lock);
 791
 792         if (timeout == 0)
 793                 RETURN(-EWOULDBLOCK);
 794
 795         /* Clear any flags that may be present from previous sends */
 796         LASSERT(req->rq_receiving_reply == 0);
 797         spin_lock(&req->rq_lock);
 798         req->rq_err = 0;
 799         req->rq_timedout = 0;
 800         req->rq_resend = 0;
 801         req->rq_restart = 0;
 802         spin_unlock(&req->rq_lock);
 803
 804         /* by now we know that timeout value is MAX_SCHEDULE_TIMEOUT,
 805          * so wait indefinitely with non-fatal signals blocked
 806          */
 807         if (l_wait_event_abortable(req->rq_reply_waitq,
 808                                    ctx_check_refresh(ctx)) == -ERESTARTSYS) {
 809                 rc = -EINTR;
 810                 ctx_refresh_interrupt(req);
 811         }
 812
 813         /*
 814          * following cases could lead us here:
 815          * - successfully refreshed;
 816          * - interrupted;
 817          * - timedout, and we don't want recover from the failure;
 818          * - timedout, and waked up upon recovery finished;
 819          * - someone else mark this ctx dead by force;
 820          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 821          *   e.g. ptlrpc_abort_inflight();
 822          */
 823         if (!cli_ctx_is_refreshed(ctx)) {
 824                 /* timed out or interruptted */
 825                 req_off_ctx_list(req, ctx);
 826
 827                 LASSERT(rc != 0);
 828                 RETURN(rc);
 829         }
 830
 831         goto again;
 832 }
 833
 834 /* Bring ptlrpc_sec context up-to-date */
 835 int sptlrpc_export_update_ctx(struct obd_export *exp)
 836 {
 837         struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
 838         struct ptlrpc_sec *sec = NULL;
 839         struct ptlrpc_cli_ctx *ctx = NULL;
 840         int rc = 0;
 841
 842         if (imp)
 843                 sec = sptlrpc_import_sec_ref(imp);
 844         if (sec) {
 845                 ctx = get_my_ctx(sec);
 846                 sptlrpc_sec_put(sec);
 847         }
 848
 849         if (ctx) {
 850                 if (ctx->cc_ops->refresh)
 851                         rc = ctx->cc_ops->refresh(ctx);
 852                 sptlrpc_cli_ctx_put(ctx, 1);
 853         }
 854         return rc;
 855 }
 856
 857 /**
 858  * Initialize flavor settings for \a req, according to \a opcode.
 859  *
 860  * \note this could be called in two situations:
 861  * - new request from ptlrpc_pre_req(), with proper @opcode
 862  * - old request which changed ctx in the middle, with @opcode == 0
 863  */
 864 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 865 {
 866         struct ptlrpc_sec *sec;
 867
 868         LASSERT(req->rq_import);
 869         LASSERT(req->rq_cli_ctx);
 870         LASSERT(req->rq_cli_ctx->cc_sec);
 871         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 872
 873         /* special security flags according to opcode */
 874         switch (opcode) {
 875         case OST_READ:
 876         case MDS_READPAGE:
 877         case MGS_CONFIG_READ:
 878         case OBD_IDX_READ:
 879                 req->rq_bulk_read = 1;
 880                 break;
 881         case OST_WRITE:
 882         case MDS_WRITEPAGE:
 883                 req->rq_bulk_write = 1;
 884                 break;
 885         case SEC_CTX_INIT:
 886                 req->rq_ctx_init = 1;
 887                 break;
 888         case SEC_CTX_FINI:
 889                 req->rq_ctx_fini = 1;
 890                 break;
 891         case 0:
 892                 /* init/fini rpc won't be resend, so can't be here */
 893                 LASSERT(req->rq_ctx_init == 0);
 894                 LASSERT(req->rq_ctx_fini == 0);
 895
 896                 /* cleanup flags, which should be recalculated */
 897                 req->rq_pack_udesc = 0;
 898                 req->rq_pack_bulk = 0;
 899                 break;
 900         }
 901
 902         sec = req->rq_cli_ctx->cc_sec;
 903
 904         spin_lock(&sec->ps_lock);
 905         req->rq_flvr = sec->ps_flvr;
 906         spin_unlock(&sec->ps_lock);
 907
 908         /*
 909          * force SVC_NULL for context initiation rpc, SVC_INTG for context
 910          * destruction rpc
 911          */
 912         if (unlikely(req->rq_ctx_init))
 913                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 914         else if (unlikely(req->rq_ctx_fini))
 915                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 916
 917         /* user descriptor flag, null security can't do it anyway */
 918         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 919             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 920                 req->rq_pack_udesc = 1;
 921
 922         /* bulk security flag */
 923         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 924             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 925                 req->rq_pack_bulk = 1;
 926 }
 927
 928 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 929 {
 930         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 931                 return;
 932
 933         LASSERT(req->rq_clrbuf);
 934         if (req->rq_pool || !req->rq_reqbuf)
 935                 return;
 936
 937         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 938         req->rq_reqbuf = NULL;
 939         req->rq_reqbuf_len = 0;
 940 }
 941
 942 /**
 943  * Given an import \a imp, check whether current user has a valid context
 944  * or not. We may create a new context and try to refresh it, and try
 945  * repeatedly try in case of non-fatal errors. Return 0 means success.
 946  */
 947 int sptlrpc_import_check_ctx(struct obd_import *imp)
 948 {
 949         struct ptlrpc_sec     *sec;
 950         struct ptlrpc_cli_ctx *ctx;
 951         struct ptlrpc_request *req = NULL;
 952         int rc;
 953
 954         ENTRY;
 955
 956         might_sleep();
 957
 958         sec = sptlrpc_import_sec_ref(imp);
 959         ctx = get_my_ctx(sec);
 960         sptlrpc_sec_put(sec);
 961
 962         if (!ctx)
 963                 RETURN(-ENOMEM);
 964
 965         if (cli_ctx_is_eternal(ctx) ||
 966             ctx->cc_ops->validate(ctx) == 0) {
 967                 sptlrpc_cli_ctx_put(ctx, 1);
 968                 RETURN(0);
 969         }
 970
 971         if (cli_ctx_is_error(ctx)) {
 972                 sptlrpc_cli_ctx_put(ctx, 1);
 973                 RETURN(-EACCES);
 974         }
 975
 976         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 977         if (!req)
 978                 RETURN(-ENOMEM);
 979
 980         ptlrpc_cli_req_init(req);
 981         atomic_set(&req->rq_refcount, 10000);
 982
 983         req->rq_import = imp;
 984         req->rq_flvr = sec->ps_flvr;
 985         req->rq_cli_ctx = ctx;
 986
 987         rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
 988         LASSERT(list_empty(&req->rq_ctx_chain));
 989         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 990         ptlrpc_request_cache_free(req);
 991
 992         RETURN(rc);
 993 }
 994
 995 /**
 996  * Used by ptlrpc client, to perform the pre-defined security transformation
 997  * upon the request message of \a req. After this function called,
 998  * req->rq_reqmsg is still accessible as clear text.
 999  */
1000 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1001 {
1002         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1003         int rc = 0;
1004
1005         ENTRY;
1006
1007         LASSERT(ctx);
1008         LASSERT(ctx->cc_sec);
1009         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1010
1011         /*
1012          * we wrap bulk request here because now we can be sure
1013          * the context is uptodate.
1014          */
1015         if (req->rq_bulk) {
1016                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1017                 if (rc)
1018                         RETURN(rc);
1019         }
1020
1021         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1022         case SPTLRPC_SVC_NULL:
1023         case SPTLRPC_SVC_AUTH:
1024         case SPTLRPC_SVC_INTG:
1025                 LASSERT(ctx->cc_ops->sign);
1026                 rc = ctx->cc_ops->sign(ctx, req);
1027                 break;
1028         case SPTLRPC_SVC_PRIV:
1029                 LASSERT(ctx->cc_ops->seal);
1030                 rc = ctx->cc_ops->seal(ctx, req);
1031                 break;
1032         default:
1033                 LBUG();
1034         }
1035
1036         if (rc == 0) {
1037                 LASSERT(req->rq_reqdata_len);
1038                 LASSERT(req->rq_reqdata_len % 8 == 0);
1039                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1040         }
1041
1042         RETURN(rc);
1043 }
1044
1045 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1046 {
1047         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1048         int rc;
1049
1050         ENTRY;
1051
1052         LASSERT(ctx);
1053         LASSERT(ctx->cc_sec);
1054         LASSERT(req->rq_repbuf);
1055         LASSERT(req->rq_repdata);
1056         LASSERT(req->rq_repmsg == NULL);
1057
1058         req->rq_rep_swab_mask = 0;
1059
1060         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1061         switch (rc) {
1062         case 1:
1063                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1064         case 0:
1065                 break;
1066         default:
1067                 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1068                 RETURN(-EPROTO);
1069         }
1070
1071         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1072                 CERROR("replied data length %d too small\n",
1073                        req->rq_repdata_len);
1074                 RETURN(-EPROTO);
1075         }
1076
1077         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1078             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1079                 CERROR("reply policy %u doesn't match request policy %u\n",
1080                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1081                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1082                 RETURN(-EPROTO);
1083         }
1084
1085         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1086         case SPTLRPC_SVC_NULL:
1087         case SPTLRPC_SVC_AUTH:
1088         case SPTLRPC_SVC_INTG:
1089                 LASSERT(ctx->cc_ops->verify);
1090                 rc = ctx->cc_ops->verify(ctx, req);
1091                 break;
1092         case SPTLRPC_SVC_PRIV:
1093                 LASSERT(ctx->cc_ops->unseal);
1094                 rc = ctx->cc_ops->unseal(ctx, req);
1095                 break;
1096         default:
1097                 LBUG();
1098         }
1099         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1100
1101         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1102             !req->rq_ctx_init)
1103                 req->rq_rep_swab_mask = 0;
1104         RETURN(rc);
1105 }
1106
1107 /**
1108  * Used by ptlrpc client, to perform security transformation upon the reply
1109  * message of \a req. After return successfully, req->rq_repmsg points to
1110  * the reply message in clear text.
1111  *
1112  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1113  * going to change.
1114  */
1115 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1116 {
1117         LASSERT(req->rq_repbuf);
1118         LASSERT(req->rq_repdata == NULL);
1119         LASSERT(req->rq_repmsg == NULL);
1120         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1121
1122         if (req->rq_reply_off == 0 &&
1123             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1124                 CERROR("real reply with offset 0\n");
1125                 return -EPROTO;
1126         }
1127
1128         if (req->rq_reply_off % 8 != 0) {
1129                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1130                 return -EPROTO;
1131         }
1132
1133         req->rq_repdata = (struct lustre_msg *)
1134                                 (req->rq_repbuf + req->rq_reply_off);
1135         req->rq_repdata_len = req->rq_nob_received;
1136
1137         return do_cli_unwrap_reply(req);
1138 }
1139
1140 /**
1141  * Used by ptlrpc client, to perform security transformation upon the early
1142  * reply message of \a req. We expect the rq_reply_off is 0, and
1143  * rq_nob_received is the early reply size.
1144  *
1145  * Because the receive buffer might be still posted, the reply data might be
1146  * changed at any time, no matter we're holding rq_lock or not. For this reason
1147  * we allocate a separate ptlrpc_request and reply buffer for early reply
1148  * processing.
1149  *
1150  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1151  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1152  * \a *req_ret to release it.
1153  * \retval -ev error number, and \a req_ret will not be set.
1154  */
1155 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1156                                    struct ptlrpc_request **req_ret)
1157 {
1158         struct ptlrpc_request *early_req;
1159         char *early_buf;
1160         int early_bufsz, early_size;
1161         int rc;
1162
1163         ENTRY;
1164
1165         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1166         if (early_req == NULL)
1167                 RETURN(-ENOMEM);
1168
1169         ptlrpc_cli_req_init(early_req);
1170
1171         early_size = req->rq_nob_received;
1172         early_bufsz = size_roundup_power2(early_size);
1173         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1174         if (early_buf == NULL)
1175                 GOTO(err_req, rc = -ENOMEM);
1176
1177         /* sanity checkings and copy data out, do it inside spinlock */
1178         spin_lock(&req->rq_lock);
1179
1180         if (req->rq_replied) {
1181                 spin_unlock(&req->rq_lock);
1182                 GOTO(err_buf, rc = -EALREADY);
1183         }
1184
1185         LASSERT(req->rq_repbuf);
1186         LASSERT(req->rq_repdata == NULL);
1187         LASSERT(req->rq_repmsg == NULL);
1188
1189         if (req->rq_reply_off != 0) {
1190                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1191                 spin_unlock(&req->rq_lock);
1192                 GOTO(err_buf, rc = -EPROTO);
1193         }
1194
1195         if (req->rq_nob_received != early_size) {
1196                 /* even another early arrived the size should be the same */
1197                 CERROR("data size has changed from %u to %u\n",
1198                        early_size, req->rq_nob_received);
1199                 spin_unlock(&req->rq_lock);
1200                 GOTO(err_buf, rc = -EINVAL);
1201         }
1202
1203         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1204                 CERROR("early reply length %d too small\n",
1205                        req->rq_nob_received);
1206                 spin_unlock(&req->rq_lock);
1207                 GOTO(err_buf, rc = -EALREADY);
1208         }
1209
1210         memcpy(early_buf, req->rq_repbuf, early_size);
1211         spin_unlock(&req->rq_lock);
1212
1213         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1214         early_req->rq_flvr = req->rq_flvr;
1215         early_req->rq_repbuf = early_buf;
1216         early_req->rq_repbuf_len = early_bufsz;
1217         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1218         early_req->rq_repdata_len = early_size;
1219         early_req->rq_early = 1;
1220         early_req->rq_reqmsg = req->rq_reqmsg;
1221
1222         rc = do_cli_unwrap_reply(early_req);
1223         if (rc) {
1224                 DEBUG_REQ(D_ADAPTTO, early_req,
1225                           "unwrap early reply: rc = %d", rc);
1226                 GOTO(err_ctx, rc);
1227         }
1228
1229         LASSERT(early_req->rq_repmsg);
1230         *req_ret = early_req;
1231         RETURN(0);
1232
1233 err_ctx:
1234         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1235 err_buf:
1236         OBD_FREE_LARGE(early_buf, early_bufsz);
1237 err_req:
1238         ptlrpc_request_cache_free(early_req);
1239         RETURN(rc);
1240 }
1241
1242 /**
1243  * Used by ptlrpc client, to release a processed early reply \a early_req.
1244  *
1245  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1246  */
1247 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1248 {
1249         LASSERT(early_req->rq_repbuf);
1250         LASSERT(early_req->rq_repdata);
1251         LASSERT(early_req->rq_repmsg);
1252
1253         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1254         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1255         ptlrpc_request_cache_free(early_req);
1256 }
1257
1258 /**************************************************
1259  * sec ID                                         *
1260  **************************************************/
1261
1262 /*
1263  * "fixed" sec (e.g. null) use sec_id < 0
1264  */
1265 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1266
1267 int sptlrpc_get_next_secid(void)
1268 {
1269         return atomic_inc_return(&sptlrpc_sec_id);
1270 }
1271 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1272
1273 /*
1274  * client side high-level security APIs
1275  */
1276
1277 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1278                                    int grace, int force)
1279 {
1280         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1281
1282         LASSERT(policy->sp_cops);
1283         LASSERT(policy->sp_cops->flush_ctx_cache);
1284
1285         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1286 }
1287
1288 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1289 {
1290         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1291
1292         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1293         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1294         LASSERT(policy->sp_cops->destroy_sec);
1295
1296         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1297
1298         policy->sp_cops->destroy_sec(sec);
1299         sptlrpc_policy_put(policy);
1300 }
1301
1302 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1303 {
1304         sec_cop_destroy_sec(sec);
1305 }
1306 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1307
1308 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1309 {
1310         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1311
1312         if (sec->ps_policy->sp_cops->kill_sec) {
1313                 sec->ps_policy->sp_cops->kill_sec(sec);
1314
1315                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1316         }
1317 }
1318
1319 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1320 {
1321         if (sec)
1322                 atomic_inc(&sec->ps_refcount);
1323
1324         return sec;
1325 }
1326 EXPORT_SYMBOL(sptlrpc_sec_get);
1327
1328 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1329 {
1330         if (sec) {
1331                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1332
1333                 if (atomic_dec_and_test(&sec->ps_refcount)) {
1334                         sptlrpc_gc_del_sec(sec);
1335                         sec_cop_destroy_sec(sec);
1336                 }
1337         }
1338 }
1339 EXPORT_SYMBOL(sptlrpc_sec_put);
1340
1341 /*
1342  * policy module is responsible for taking refrence of import
1343  */
1344 static
1345 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1346                                        struct ptlrpc_svc_ctx *svc_ctx,
1347                                        struct sptlrpc_flavor *sf,
1348                                        enum lustre_sec_part sp)
1349 {
1350         struct ptlrpc_sec_policy *policy;
1351         struct ptlrpc_sec *sec;
1352         char str[32];
1353
1354         ENTRY;
1355
1356         if (svc_ctx) {
1357                 LASSERT(imp->imp_dlm_fake == 1);
1358
1359                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1360                        imp->imp_obd->obd_type->typ_name,
1361                        imp->imp_obd->obd_name,
1362                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1363
1364                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1365                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1366         } else {
1367                 LASSERT(imp->imp_dlm_fake == 0);
1368
1369                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1370                        imp->imp_obd->obd_type->typ_name,
1371                        imp->imp_obd->obd_name,
1372                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1373
1374                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1375                 if (!policy) {
1376                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1377                         RETURN(NULL);
1378                 }
1379         }
1380
1381         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1382         if (sec) {
1383                 atomic_inc(&sec->ps_refcount);
1384
1385                 sec->ps_part = sp;
1386
1387                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1388                         sptlrpc_gc_add_sec(sec);
1389         } else {
1390                 sptlrpc_policy_put(policy);
1391         }
1392
1393         RETURN(sec);
1394 }
1395
1396 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1397 {
1398         struct ptlrpc_sec *sec;
1399
1400         read_lock(&imp->imp_sec_lock);
1401         sec = sptlrpc_sec_get(imp->imp_sec);
1402         read_unlock(&imp->imp_sec_lock);
1403
1404         return sec;
1405 }
1406 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1407
1408 static void sptlrpc_import_sec_install(struct obd_import *imp,
1409                                        struct ptlrpc_sec *sec)
1410 {
1411         struct ptlrpc_sec *old_sec;
1412
1413         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1414
1415         write_lock(&imp->imp_sec_lock);
1416         old_sec = imp->imp_sec;
1417         imp->imp_sec = sec;
1418         write_unlock(&imp->imp_sec_lock);
1419
1420         if (old_sec) {
1421                 sptlrpc_sec_kill(old_sec);
1422
1423                 /* balance the ref taken by this import */
1424                 sptlrpc_sec_put(old_sec);
1425         }
1426 }
1427
1428 static inline
1429 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1430 {
1431         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1432 }
1433
1434 static inline
1435 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1436 {
1437         *dst = *src;
1438 }
1439
1440 /**
1441  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1442  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1443  *
1444  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1445  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1446  */
1447 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1448                              struct ptlrpc_svc_ctx *svc_ctx,
1449                              struct sptlrpc_flavor *flvr)
1450 {
1451         struct ptlrpc_connection *conn;
1452         struct sptlrpc_flavor sf;
1453         struct ptlrpc_sec *sec, *newsec;
1454         enum lustre_sec_part sp;
1455         char str[24];
1456         int rc = 0;
1457
1458         ENTRY;
1459
1460         might_sleep();
1461
1462         if (imp == NULL)
1463                 RETURN(0);
1464
1465         conn = imp->imp_connection;
1466
1467         if (svc_ctx == NULL) {
1468                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1469                 /*
1470                  * normal import, determine flavor from rule set, except
1471                  * for mgc the flavor is predetermined.
1472                  */
1473                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1474                         sf = cliobd->cl_flvr_mgc;
1475                 else
1476                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1477                                                    cliobd->cl_sp_to,
1478                                                    &cliobd->cl_target_uuid,
1479                                                    conn->c_self, &sf);
1480
1481                 sp = imp->imp_obd->u.cli.cl_sp_me;
1482         } else {
1483                 /* reverse import, determine flavor from incoming reqeust */
1484                 sf = *flvr;
1485
1486                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1487                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1488                                       PTLRPC_SEC_FL_ROOTONLY;
1489
1490                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1491         }
1492
1493         sec = sptlrpc_import_sec_ref(imp);
1494         if (sec) {
1495                 char str2[24];
1496
1497                 if (flavor_equal(&sf, &sec->ps_flvr))
1498                         GOTO(out, rc);
1499
1500                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1501                        imp->imp_obd->obd_name,
1502                        obd_uuid2str(&conn->c_remote_uuid),
1503                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1504                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1505         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1506                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1507                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1508                        imp->imp_obd->obd_name,
1509                        obd_uuid2str(&conn->c_remote_uuid),
1510                        LNET_NIDNET(conn->c_self),
1511                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1512         }
1513
1514         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1515         if (newsec) {
1516                 sptlrpc_import_sec_install(imp, newsec);
1517         } else {
1518                 CERROR("import %s->%s: failed to create new sec\n",
1519                        imp->imp_obd->obd_name,
1520                        obd_uuid2str(&conn->c_remote_uuid));
1521                 rc = -EPERM;
1522         }
1523
1524 out:
1525         sptlrpc_sec_put(sec);
1526         RETURN(rc);
1527 }
1528
1529 void sptlrpc_import_sec_put(struct obd_import *imp)
1530 {
1531         if (imp->imp_sec) {
1532                 sptlrpc_sec_kill(imp->imp_sec);
1533
1534                 sptlrpc_sec_put(imp->imp_sec);
1535                 imp->imp_sec = NULL;
1536         }
1537 }
1538
1539 static void import_flush_ctx_common(struct obd_import *imp,
1540                                     uid_t uid, int grace, int force)
1541 {
1542         struct ptlrpc_sec *sec;
1543
1544         if (imp == NULL)
1545                 return;
1546
1547         sec = sptlrpc_import_sec_ref(imp);
1548         if (sec == NULL)
1549                 return;
1550
1551         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1552         sptlrpc_sec_put(sec);
1553 }
1554
1555 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1556 {
1557         /*
1558          * it's important to use grace mode, see explain in
1559          * sptlrpc_req_refresh_ctx()
1560          */
1561         import_flush_ctx_common(imp, 0, 1, 1);
1562 }
1563
1564 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1565 {
1566         import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1567                                 1, 1);
1568 }
1569 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1570
1571 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1572 {
1573         import_flush_ctx_common(imp, -1, 1, 1);
1574 }
1575 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1576
1577 /**
1578  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1579  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1580  */
1581 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1582 {
1583         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1584         struct ptlrpc_sec_policy *policy;
1585         int rc;
1586
1587         LASSERT(ctx);
1588         LASSERT(ctx->cc_sec);
1589         LASSERT(ctx->cc_sec->ps_policy);
1590         LASSERT(req->rq_reqmsg == NULL);
1591         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1592
1593         policy = ctx->cc_sec->ps_policy;
1594         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1595         if (!rc) {
1596                 LASSERT(req->rq_reqmsg);
1597                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1598
1599                 /* zeroing preallocated buffer */
1600                 if (req->rq_pool)
1601                         memset(req->rq_reqmsg, 0, msgsize);
1602         }
1603
1604         return rc;
1605 }
1606
1607 /**
1608  * Used by ptlrpc client to free request buffer of \a req. After this
1609  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1610  */
1611 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1612 {
1613         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1614         struct ptlrpc_sec_policy *policy;
1615
1616         LASSERT(ctx);
1617         LASSERT(ctx->cc_sec);
1618         LASSERT(ctx->cc_sec->ps_policy);
1619         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1620
1621         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1622                 return;
1623
1624         policy = ctx->cc_sec->ps_policy;
1625         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1626         req->rq_reqmsg = NULL;
1627 }
1628
1629 /*
1630  * NOTE caller must guarantee the buffer size is enough for the enlargement
1631  */
1632 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1633                                   int segment, int newsize)
1634 {
1635         void *src, *dst;
1636         int oldsize, oldmsg_size, movesize;
1637
1638         LASSERT(segment < msg->lm_bufcount);
1639         LASSERT(msg->lm_buflens[segment] <= newsize);
1640
1641         if (msg->lm_buflens[segment] == newsize)
1642                 return;
1643
1644         /* nothing to do if we are enlarging the last segment */
1645         if (segment == msg->lm_bufcount - 1) {
1646                 msg->lm_buflens[segment] = newsize;
1647                 return;
1648         }
1649
1650         oldsize = msg->lm_buflens[segment];
1651
1652         src = lustre_msg_buf(msg, segment + 1, 0);
1653         msg->lm_buflens[segment] = newsize;
1654         dst = lustre_msg_buf(msg, segment + 1, 0);
1655         msg->lm_buflens[segment] = oldsize;
1656
1657         /* move from segment + 1 to end segment */
1658         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1659         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1660         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1661         LASSERT(movesize >= 0);
1662
1663         if (movesize)
1664                 memmove(dst, src, movesize);
1665
1666         /* note we don't clear the ares where old data live, not secret */
1667
1668         /* finally set new segment size */
1669         msg->lm_buflens[segment] = newsize;
1670 }
1671 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1672
1673 /**
1674  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1675  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1676  * preserved after the enlargement. this must be called after original request
1677  * buffer being allocated.
1678  *
1679  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1680  * so caller should refresh its local pointers if needed.
1681  */
1682 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1683                                const struct req_msg_field *field,
1684                                int newsize)
1685 {
1686         struct req_capsule *pill = &req->rq_pill;
1687         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1688         struct ptlrpc_sec_cops *cops;
1689         struct lustre_msg *msg = req->rq_reqmsg;
1690         int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1691
1692         LASSERT(ctx);
1693         LASSERT(msg);
1694         LASSERT(msg->lm_bufcount > segment);
1695         LASSERT(msg->lm_buflens[segment] <= newsize);
1696
1697         if (msg->lm_buflens[segment] == newsize)
1698                 return 0;
1699
1700         cops = ctx->cc_sec->ps_policy->sp_cops;
1701         LASSERT(cops->enlarge_reqbuf);
1702         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1703 }
1704 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1705
1706 /**
1707  * Used by ptlrpc client to allocate reply buffer of \a req.
1708  *
1709  * \note After this, req->rq_repmsg is still not accessible.
1710  */
1711 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1712 {
1713         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1714         struct ptlrpc_sec_policy *policy;
1715
1716         ENTRY;
1717
1718         LASSERT(ctx);
1719         LASSERT(ctx->cc_sec);
1720         LASSERT(ctx->cc_sec->ps_policy);
1721
1722         if (req->rq_repbuf)
1723                 RETURN(0);
1724
1725         policy = ctx->cc_sec->ps_policy;
1726         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1727 }
1728
1729 /**
1730  * Used by ptlrpc client to free reply buffer of \a req. After this
1731  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1732  */
1733 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1734 {
1735         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1736         struct ptlrpc_sec_policy *policy;
1737
1738         ENTRY;
1739
1740         LASSERT(ctx);
1741         LASSERT(ctx->cc_sec);
1742         LASSERT(ctx->cc_sec->ps_policy);
1743         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1744
1745         if (req->rq_repbuf == NULL)
1746                 return;
1747         LASSERT(req->rq_repbuf_len);
1748
1749         policy = ctx->cc_sec->ps_policy;
1750         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1751         req->rq_repmsg = NULL;
1752         EXIT;
1753 }
1754 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1755
1756 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1757                                 struct ptlrpc_cli_ctx *ctx)
1758 {
1759         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1760
1761         if (!policy->sp_cops->install_rctx)
1762                 return 0;
1763         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1764 }
1765
1766 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1767                                 struct ptlrpc_svc_ctx *ctx)
1768 {
1769         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1770
1771         if (!policy->sp_sops->install_rctx)
1772                 return 0;
1773         return policy->sp_sops->install_rctx(imp, ctx);
1774 }
1775
1776 /* Get SELinux policy info from userspace */
1777 static int sepol_helper(struct obd_import *imp)
1778 {
1779         char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1780         char *argv[] = {
1781                 [0] = "/usr/sbin/l_getsepol",
1782                 [1] = "-o",
1783                 [2] = NULL,         /* obd type */
1784                 [3] = "-n",
1785                 [4] = NULL,         /* obd name */
1786                 [5] = "-t",
1787                 [6] = mtime_str,    /* policy mtime */
1788                 [7] = "-m",
1789                 [8] = mode_str,     /* enforcing mode */
1790                 [9] = NULL
1791         };
1792         char *envp[] = {
1793                 [0] = "HOME=/",
1794                 [1] = "PATH=/sbin:/usr/sbin",
1795                 [2] = NULL
1796         };
1797         signed short ret;
1798         int rc = 0;
1799
1800         if (imp == NULL || imp->imp_obd == NULL ||
1801             imp->imp_obd->obd_type == NULL) {
1802                 rc = -EINVAL;
1803         } else {
1804                 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1805                 argv[4] = imp->imp_obd->obd_name;
1806                 spin_lock(&imp->imp_sec->ps_lock);
1807                 if (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1808                     imp->imp_sec->ps_sepol[0] == '\0') {
1809                         /* ps_sepol has not been initialized */
1810                         argv[5] = NULL;
1811                         argv[7] = NULL;
1812                 } else {
1813                         time64_t mtime_ms;
1814
1815                         mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1816                         snprintf(mtime_str, sizeof(mtime_str), "%lld",
1817                                  mtime_ms / MSEC_PER_SEC);
1818                         mode_str[0] = imp->imp_sec->ps_sepol[0];
1819                 }
1820                 spin_unlock(&imp->imp_sec->ps_lock);
1821                 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1822                 rc = ret>>8;
1823         }
1824
1825         return rc;
1826 }
1827
1828 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1829 {
1830         ktime_t checknext;
1831
1832         if (send_sepol == 0 || !selinux_is_enabled())
1833                 return 0;
1834
1835         if (send_sepol == -1)
1836                 /* send_sepol == -1 means fetch sepol status every time */
1837                 return 1;
1838
1839         spin_lock(&imp_sec->ps_lock);
1840         checknext = imp_sec->ps_sepol_checknext;
1841         spin_unlock(&imp_sec->ps_lock);
1842
1843         /* next check is too far in time, please update */
1844         if (ktime_after(checknext,
1845                         ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1846                 goto setnext;
1847
1848         if (ktime_before(ktime_get(), checknext))
1849                 /* too early to fetch sepol status */
1850                 return 0;
1851
1852 setnext:
1853         /* define new sepol_checknext time */
1854         spin_lock(&imp_sec->ps_lock);
1855         imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1856                                                 ktime_set(send_sepol, 0));
1857         spin_unlock(&imp_sec->ps_lock);
1858
1859         return 1;
1860 }
1861
1862 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1863 {
1864         struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1865         int rc = 0;
1866
1867         ENTRY;
1868
1869         (req->rq_sepol)[0] = '\0';
1870
1871 #ifndef HAVE_SELINUX
1872         if (unlikely(send_sepol != 0))
1873                 CDEBUG(D_SEC,
1874                        "Client cannot report SELinux status, it was not built against libselinux.\n");
1875         RETURN(0);
1876 #endif
1877
1878         if (send_sepol == 0 || !selinux_is_enabled())
1879                 RETURN(0);
1880
1881         if (imp_sec == NULL)
1882                 RETURN(-EINVAL);
1883
1884         /* Retrieve SELinux status info */
1885         if (sptlrpc_sepol_needs_check(imp_sec))
1886                 rc = sepol_helper(req->rq_import);
1887         if (likely(rc == 0)) {
1888                 spin_lock(&imp_sec->ps_lock);
1889                 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1890                        sizeof(req->rq_sepol));
1891                 spin_unlock(&imp_sec->ps_lock);
1892         }
1893
1894         RETURN(rc);
1895 }
1896 EXPORT_SYMBOL(sptlrpc_get_sepol);
1897
1898 /*
1899  * server side security
1900  */
1901
1902 static int flavor_allowed(struct sptlrpc_flavor *exp,
1903                           struct ptlrpc_request *req)
1904 {
1905         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1906
1907         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1908                 return 1;
1909
1910         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1911             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1912             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1913             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1914                 return 1;
1915
1916         return 0;
1917 }
1918
1919 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1920
1921 /**
1922  * Given an export \a exp, check whether the flavor of incoming \a req
1923  * is allowed by the export \a exp. Main logic is about taking care of
1924  * changing configurations. Return 0 means success.
1925  */
1926 int sptlrpc_target_export_check(struct obd_export *exp,
1927                                 struct ptlrpc_request *req)
1928 {
1929         struct sptlrpc_flavor   flavor;
1930
1931         if (exp == NULL)
1932                 return 0;
1933
1934         /*
1935          * client side export has no imp_reverse, skip
1936          * FIXME maybe we should check flavor this as well???
1937          */
1938         if (exp->exp_imp_reverse == NULL)
1939                 return 0;
1940
1941         /* don't care about ctx fini rpc */
1942         if (req->rq_ctx_fini)
1943                 return 0;
1944
1945         spin_lock(&exp->exp_lock);
1946
1947         /*
1948          * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1949          * the first req with the new flavor, then treat it as current flavor,
1950          * adapt reverse sec according to it.
1951          * note the first rpc with new flavor might not be with root ctx, in
1952          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1953          */
1954         if (unlikely(exp->exp_flvr_changed) &&
1955             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1956                 /*
1957                  * make the new flavor as "current", and old ones as
1958                  * about-to-expire
1959                  */
1960                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1961                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1962                 flavor = exp->exp_flvr_old[1];
1963                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1964                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1965                 exp->exp_flvr_old[0] = exp->exp_flvr;
1966                 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1967                                           EXP_FLVR_UPDATE_EXPIRE;
1968                 exp->exp_flvr = flavor;
1969
1970                 /* flavor change finished */
1971                 exp->exp_flvr_changed = 0;
1972                 LASSERT(exp->exp_flvr_adapt == 1);
1973
1974                 /* if it's gss, we only interested in root ctx init */
1975                 if (req->rq_auth_gss &&
1976                     !(req->rq_ctx_init &&
1977                     (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1978                     req->rq_auth_usr_ost))) {
1979                         spin_unlock(&exp->exp_lock);
1980                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1981                                req->rq_auth_gss, req->rq_ctx_init,
1982                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1983                                req->rq_auth_usr_ost);
1984                         return 0;
1985                 }
1986
1987                 exp->exp_flvr_adapt = 0;
1988                 spin_unlock(&exp->exp_lock);
1989
1990                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1991                                                 req->rq_svc_ctx, &flavor);
1992         }
1993
1994         /*
1995          * if it equals to the current flavor, we accept it, but need to
1996          * dealing with reverse sec/ctx
1997          */
1998         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1999                 /*
2000                  * most cases should return here, we only interested in
2001                  * gss root ctx init
2002                  */
2003                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2004                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2005                      !req->rq_auth_usr_ost)) {
2006                         spin_unlock(&exp->exp_lock);
2007                         return 0;
2008                 }
2009
2010                 /*
2011                  * if flavor just changed, we should not proceed, just leave
2012                  * it and current flavor will be discovered and replaced
2013                  * shortly, and let _this_ rpc pass through
2014                  */
2015                 if (exp->exp_flvr_changed) {
2016                         LASSERT(exp->exp_flvr_adapt);
2017                         spin_unlock(&exp->exp_lock);
2018                         return 0;
2019                 }
2020
2021                 if (exp->exp_flvr_adapt) {
2022                         exp->exp_flvr_adapt = 0;
2023                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2024                                exp, exp->exp_flvr.sf_rpc,
2025                                exp->exp_flvr_old[0].sf_rpc,
2026                                exp->exp_flvr_old[1].sf_rpc);
2027                         flavor = exp->exp_flvr;
2028                         spin_unlock(&exp->exp_lock);
2029
2030                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2031                                                         req->rq_svc_ctx,
2032                                                         &flavor);
2033                 } else {
2034                         CDEBUG(D_SEC,
2035                                "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2036                                exp, exp->exp_flvr.sf_rpc,
2037                                exp->exp_flvr_old[0].sf_rpc,
2038                                exp->exp_flvr_old[1].sf_rpc);
2039                         spin_unlock(&exp->exp_lock);
2040
2041                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2042                                                            req->rq_svc_ctx);
2043                 }
2044         }
2045
2046         if (exp->exp_flvr_expire[0]) {
2047                 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2048                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2049                                 CDEBUG(D_SEC,
2050                                        "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2051                                        exp, exp->exp_flvr.sf_rpc,
2052                                        exp->exp_flvr_old[0].sf_rpc,
2053                                        exp->exp_flvr_old[1].sf_rpc,
2054                                        (s64)(exp->exp_flvr_expire[0] -
2055                                              ktime_get_real_seconds()));
2056                                 spin_unlock(&exp->exp_lock);
2057                                 return 0;
2058                         }
2059                 } else {
2060                         CDEBUG(D_SEC, "mark middle expired\n");
2061                         exp->exp_flvr_expire[0] = 0;
2062                 }
2063                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2064                        exp->exp_flvr.sf_rpc,
2065                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2066                        req->rq_flvr.sf_rpc);
2067         }
2068
2069         /*
2070          * now it doesn't match the current flavor, the only chance we can
2071          * accept it is match the old flavors which is not expired.
2072          */
2073         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2074                 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2075                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2076                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2077                                        exp,
2078                                        exp->exp_flvr.sf_rpc,
2079                                        exp->exp_flvr_old[0].sf_rpc,
2080                                        exp->exp_flvr_old[1].sf_rpc,
2081                                        (s64)(exp->exp_flvr_expire[1] -
2082                                        ktime_get_real_seconds()));
2083                                 spin_unlock(&exp->exp_lock);
2084                                 return 0;
2085                         }
2086                 } else {
2087                         CDEBUG(D_SEC, "mark oldest expired\n");
2088                         exp->exp_flvr_expire[1] = 0;
2089                 }
2090                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2091                        exp, exp->exp_flvr.sf_rpc,
2092                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2093                        req->rq_flvr.sf_rpc);
2094         } else {
2095                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2096                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2097                        exp->exp_flvr_old[1].sf_rpc);
2098         }
2099
2100         spin_unlock(&exp->exp_lock);
2101
2102         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2103               exp, exp->exp_obd->obd_name,
2104               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2105               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2106               req->rq_flvr.sf_rpc,
2107               exp->exp_flvr.sf_rpc,
2108               exp->exp_flvr_old[0].sf_rpc,
2109               exp->exp_flvr_expire[0] ?
2110               (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2111               exp->exp_flvr_old[1].sf_rpc,
2112               exp->exp_flvr_expire[1] ?
2113               (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2114         return -EACCES;
2115 }
2116 EXPORT_SYMBOL(sptlrpc_target_export_check);
2117
2118 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2119                                       struct sptlrpc_rule_set *rset)
2120 {
2121         struct obd_export *exp;
2122         struct sptlrpc_flavor new_flvr;
2123
2124         LASSERT(obd);
2125
2126         spin_lock(&obd->obd_dev_lock);
2127
2128         list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2129                 if (exp->exp_connection == NULL)
2130                         continue;
2131
2132                 /*
2133                  * note if this export had just been updated flavor
2134                  * (exp_flvr_changed == 1), this will override the
2135                  * previous one.
2136                  */
2137                 spin_lock(&exp->exp_lock);
2138                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2139                                              exp->exp_connection->c_peer.nid,
2140                                              &new_flvr);
2141                 if (exp->exp_flvr_changed ||
2142                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2143                         exp->exp_flvr_old[1] = new_flvr;
2144                         exp->exp_flvr_expire[1] = 0;
2145                         exp->exp_flvr_changed = 1;
2146                         exp->exp_flvr_adapt = 1;
2147
2148                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2149                                exp, sptlrpc_part2name(exp->exp_sp_peer),
2150                                exp->exp_flvr.sf_rpc,
2151                                exp->exp_flvr_old[1].sf_rpc);
2152                 }
2153                 spin_unlock(&exp->exp_lock);
2154         }
2155
2156         spin_unlock(&obd->obd_dev_lock);
2157 }
2158 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2159
2160 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2161 {
2162         /* peer's claim is unreliable unless gss is being used */
2163         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2164                 return svc_rc;
2165
2166         switch (req->rq_sp_from) {
2167         case LUSTRE_SP_CLI:
2168                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2169                         /* The below message is checked in sanity-sec test_33 */
2170                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
2171                         svc_rc = SECSVC_DROP;
2172                 }
2173                 break;
2174         case LUSTRE_SP_MDT:
2175                 if (!req->rq_auth_usr_mdt) {
2176                         /* The below message is checked in sanity-sec test_33 */
2177                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
2178                         svc_rc = SECSVC_DROP;
2179                 }
2180                 break;
2181         case LUSTRE_SP_OST:
2182                 if (!req->rq_auth_usr_ost) {
2183                         /* The below message is checked in sanity-sec test_33 */
2184                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2185                         svc_rc = SECSVC_DROP;
2186                 }
2187                 break;
2188         case LUSTRE_SP_MGS:
2189         case LUSTRE_SP_MGC:
2190                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2191                     !req->rq_auth_usr_ost) {
2192                         /* The below message is checked in sanity-sec test_33 */
2193                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2194                         svc_rc = SECSVC_DROP;
2195                 }
2196                 break;
2197         case LUSTRE_SP_ANY:
2198         default:
2199                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2200                 svc_rc = SECSVC_DROP;
2201         }
2202
2203         return svc_rc;
2204 }
2205
2206 /**
2207  * Used by ptlrpc server, to perform transformation upon request message of
2208  * incoming \a req. This must be the first thing to do with an incoming
2209  * request in ptlrpc layer.
2210  *
2211  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2212  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2213  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2214  * reply message has been prepared.
2215  * \retval SECSVC_DROP failed, this request should be dropped.
2216  */
2217 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2218 {
2219         struct ptlrpc_sec_policy *policy;
2220         struct lustre_msg *msg = req->rq_reqbuf;
2221         int rc;
2222
2223         ENTRY;
2224
2225         LASSERT(msg);
2226         LASSERT(req->rq_reqmsg == NULL);
2227         LASSERT(req->rq_repmsg == NULL);
2228         LASSERT(req->rq_svc_ctx == NULL);
2229
2230         req->rq_req_swab_mask = 0;
2231
2232         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2233         switch (rc) {
2234         case 1:
2235                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2236         case 0:
2237                 break;
2238         default:
2239                 CERROR("error unpacking request from %s x%llu\n",
2240                        libcfs_id2str(req->rq_peer), req->rq_xid);
2241                 RETURN(SECSVC_DROP);
2242         }
2243
2244         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2245         req->rq_sp_from = LUSTRE_SP_ANY;
2246         req->rq_auth_uid = -1; /* set to INVALID_UID */
2247         req->rq_auth_mapped_uid = -1;
2248
2249         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2250         if (!policy) {
2251                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2252                 RETURN(SECSVC_DROP);
2253         }
2254
2255         LASSERT(policy->sp_sops->accept);
2256         rc = policy->sp_sops->accept(req);
2257         sptlrpc_policy_put(policy);
2258         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2259         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2260
2261         /*
2262          * if it's not null flavor (which means embedded packing msg),
2263          * reset the swab mask for the comming inner msg unpacking.
2264          */
2265         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2266                 req->rq_req_swab_mask = 0;
2267
2268         /* sanity check for the request source */
2269         rc = sptlrpc_svc_check_from(req, rc);
2270         RETURN(rc);
2271 }
2272
2273 /**
2274  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2275  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2276  * a buffer of \a msglen size.
2277  */
2278 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2279 {
2280         struct ptlrpc_sec_policy *policy;
2281         struct ptlrpc_reply_state *rs;
2282         int rc;
2283
2284         ENTRY;
2285
2286         LASSERT(req->rq_svc_ctx);
2287         LASSERT(req->rq_svc_ctx->sc_policy);
2288
2289         policy = req->rq_svc_ctx->sc_policy;
2290         LASSERT(policy->sp_sops->alloc_rs);
2291
2292         rc = policy->sp_sops->alloc_rs(req, msglen);
2293         if (unlikely(rc == -ENOMEM)) {
2294                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2295
2296                 if (svcpt->scp_service->srv_max_reply_size <
2297                    msglen + sizeof(struct ptlrpc_reply_state)) {
2298                         /* Just return failure if the size is too big */
2299                         CERROR("size of message is too big (%zd), %d allowed\n",
2300                                 msglen + sizeof(struct ptlrpc_reply_state),
2301                                 svcpt->scp_service->srv_max_reply_size);
2302                         RETURN(-ENOMEM);
2303                 }
2304
2305                 /* failed alloc, try emergency pool */
2306                 rs = lustre_get_emerg_rs(svcpt);
2307                 if (rs == NULL)
2308                         RETURN(-ENOMEM);
2309
2310                 req->rq_reply_state = rs;
2311                 rc = policy->sp_sops->alloc_rs(req, msglen);
2312                 if (rc) {
2313                         lustre_put_emerg_rs(rs);
2314                         req->rq_reply_state = NULL;
2315                 }
2316         }
2317
2318         LASSERT(rc != 0 ||
2319                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2320
2321         RETURN(rc);
2322 }
2323
2324 /**
2325  * Used by ptlrpc server, to perform transformation upon reply message.
2326  *
2327  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2328  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2329  */
2330 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2331 {
2332         struct ptlrpc_sec_policy *policy;
2333         int rc;
2334
2335         ENTRY;
2336
2337         LASSERT(req->rq_svc_ctx);
2338         LASSERT(req->rq_svc_ctx->sc_policy);
2339
2340         policy = req->rq_svc_ctx->sc_policy;
2341         LASSERT(policy->sp_sops->authorize);
2342
2343         rc = policy->sp_sops->authorize(req);
2344         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2345
2346         RETURN(rc);
2347 }
2348
2349 /**
2350  * Used by ptlrpc server, to free reply_state.
2351  */
2352 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2353 {
2354         struct ptlrpc_sec_policy *policy;
2355         unsigned int prealloc;
2356
2357         ENTRY;
2358
2359         LASSERT(rs->rs_svc_ctx);
2360         LASSERT(rs->rs_svc_ctx->sc_policy);
2361
2362         policy = rs->rs_svc_ctx->sc_policy;
2363         LASSERT(policy->sp_sops->free_rs);
2364
2365         prealloc = rs->rs_prealloc;
2366         policy->sp_sops->free_rs(rs);
2367
2368         if (prealloc)
2369                 lustre_put_emerg_rs(rs);
2370         EXIT;
2371 }
2372
2373 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2374 {
2375         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2376
2377         if (ctx != NULL)
2378                 atomic_inc(&ctx->sc_refcount);
2379 }
2380
2381 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2382 {
2383         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2384
2385         if (ctx == NULL)
2386                 return;
2387
2388         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2389         if (atomic_dec_and_test(&ctx->sc_refcount)) {
2390                 if (ctx->sc_policy->sp_sops->free_ctx)
2391                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2392         }
2393         req->rq_svc_ctx = NULL;
2394 }
2395
2396 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2397 {
2398         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2399
2400         if (ctx == NULL)
2401                 return;
2402
2403         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2404         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2405                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2406 }
2407 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2408
2409 /*
2410  * bulk security
2411  */
2412
2413 /**
2414  * Perform transformation upon bulk data pointed by \a desc. This is called
2415  * before transforming the request message.
2416  */
2417 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2418                           struct ptlrpc_bulk_desc *desc)
2419 {
2420         struct ptlrpc_cli_ctx *ctx;
2421
2422         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2423
2424         if (!req->rq_pack_bulk)
2425                 return 0;
2426
2427         ctx = req->rq_cli_ctx;
2428         if (ctx->cc_ops->wrap_bulk)
2429                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2430         return 0;
2431 }
2432 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2433
2434 /**
2435  * This is called after unwrap the reply message.
2436  * return nob of actual plain text size received, or error code.
2437  */
2438 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2439                                  struct ptlrpc_bulk_desc *desc,
2440                                  int nob)
2441 {
2442         struct ptlrpc_cli_ctx *ctx;
2443         int rc;
2444
2445         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2446
2447         if (!req->rq_pack_bulk)
2448                 return desc->bd_nob_transferred;
2449
2450         ctx = req->rq_cli_ctx;
2451         if (ctx->cc_ops->unwrap_bulk) {
2452                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2453                 if (rc < 0)
2454                         return rc;
2455         }
2456         return desc->bd_nob_transferred;
2457 }
2458 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2459
2460 /**
2461  * This is called after unwrap the reply message.
2462  * return 0 for success or error code.
2463  */
2464 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2465                                   struct ptlrpc_bulk_desc *desc)
2466 {
2467         struct ptlrpc_cli_ctx *ctx;
2468         int rc;
2469
2470         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2471
2472         if (!req->rq_pack_bulk)
2473                 return 0;
2474
2475         ctx = req->rq_cli_ctx;
2476         if (ctx->cc_ops->unwrap_bulk) {
2477                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2478                 if (rc < 0)
2479                         return rc;
2480         }
2481
2482         /*
2483          * if everything is going right, nob should equals to nob_transferred.
2484          * in case of privacy mode, nob_transferred needs to be adjusted.
2485          */
2486         if (desc->bd_nob != desc->bd_nob_transferred) {
2487                 CERROR("nob %d doesn't match transferred nob %d\n",
2488                        desc->bd_nob, desc->bd_nob_transferred);
2489                 return -EPROTO;
2490         }
2491
2492         return 0;
2493 }
2494 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2495
2496 #ifdef HAVE_SERVER_SUPPORT
2497 /**
2498  * Performe transformation upon outgoing bulk read.
2499  */
2500 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2501                           struct ptlrpc_bulk_desc *desc)
2502 {
2503         struct ptlrpc_svc_ctx *ctx;
2504
2505         LASSERT(req->rq_bulk_read);
2506
2507         if (!req->rq_pack_bulk)
2508                 return 0;
2509
2510         ctx = req->rq_svc_ctx;
2511         if (ctx->sc_policy->sp_sops->wrap_bulk)
2512                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2513
2514         return 0;
2515 }
2516 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2517
2518 /**
2519  * Performe transformation upon incoming bulk write.
2520  */
2521 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2522                             struct ptlrpc_bulk_desc *desc)
2523 {
2524         struct ptlrpc_svc_ctx *ctx;
2525         int rc;
2526
2527         LASSERT(req->rq_bulk_write);
2528
2529         /*
2530          * if it's in privacy mode, transferred should >= expected; otherwise
2531          * transferred should == expected.
2532          */
2533         if (desc->bd_nob_transferred < desc->bd_nob ||
2534             (desc->bd_nob_transferred > desc->bd_nob &&
2535              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2536              SPTLRPC_BULK_SVC_PRIV)) {
2537                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2538                           desc->bd_nob_transferred, desc->bd_nob);
2539                 return -ETIMEDOUT;
2540         }
2541
2542         if (!req->rq_pack_bulk)
2543                 return 0;
2544
2545         ctx = req->rq_svc_ctx;
2546         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2547                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2548                 if (rc)
2549                         CERROR("error unwrap bulk: %d\n", rc);
2550         }
2551
2552         /* return 0 to allow reply be sent */
2553         return 0;
2554 }
2555 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2556
2557 /**
2558  * Prepare buffers for incoming bulk write.
2559  */
2560 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2561                           struct ptlrpc_bulk_desc *desc)
2562 {
2563         struct ptlrpc_svc_ctx *ctx;
2564
2565         LASSERT(req->rq_bulk_write);
2566
2567         if (!req->rq_pack_bulk)
2568                 return 0;
2569
2570         ctx = req->rq_svc_ctx;
2571         if (ctx->sc_policy->sp_sops->prep_bulk)
2572                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2573
2574         return 0;
2575 }
2576 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2577
2578 #endif /* HAVE_SERVER_SUPPORT */
2579
2580 /*
2581  * user descriptor helpers
2582  */
2583
2584 int sptlrpc_current_user_desc_size(void)
2585 {
2586         int ngroups;
2587
2588         ngroups = current_cred()->group_info->ngroups;
2589
2590         if (ngroups > LUSTRE_MAX_GROUPS)
2591                 ngroups = LUSTRE_MAX_GROUPS;
2592         return sptlrpc_user_desc_size(ngroups);
2593 }
2594 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2595
2596 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2597 {
2598         struct ptlrpc_user_desc *pud;
2599         int ngroups;
2600
2601         pud = lustre_msg_buf(msg, offset, 0);
2602
2603         pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2604         pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2605         pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2606         pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2607         pud->pud_cap = cfs_curproc_cap_pack();
2608         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2609
2610         task_lock(current);
2611         ngroups = current_cred()->group_info->ngroups;
2612         if (pud->pud_ngroups > ngroups)
2613                 pud->pud_ngroups = ngroups;
2614 #ifdef HAVE_GROUP_INFO_GID
2615         memcpy(pud->pud_groups, current_cred()->group_info->gid,
2616                pud->pud_ngroups * sizeof(__u32));
2617 #else /* !HAVE_GROUP_INFO_GID */
2618         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2619                pud->pud_ngroups * sizeof(__u32));
2620 #endif /* HAVE_GROUP_INFO_GID */
2621         task_unlock(current);
2622
2623         return 0;
2624 }
2625 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2626
2627 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2628 {
2629         struct ptlrpc_user_desc *pud;
2630         int i;
2631
2632         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2633         if (!pud)
2634                 return -EINVAL;
2635
2636         if (swabbed) {
2637                 __swab32s(&pud->pud_uid);
2638                 __swab32s(&pud->pud_gid);
2639                 __swab32s(&pud->pud_fsuid);
2640                 __swab32s(&pud->pud_fsgid);
2641                 __swab32s(&pud->pud_cap);
2642                 __swab32s(&pud->pud_ngroups);
2643         }
2644
2645         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2646                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2647                 return -EINVAL;
2648         }
2649
2650         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2651             msg->lm_buflens[offset]) {
2652                 CERROR("%u groups are claimed but bufsize only %u\n",
2653                        pud->pud_ngroups, msg->lm_buflens[offset]);
2654                 return -EINVAL;
2655         }
2656
2657         if (swabbed) {
2658                 for (i = 0; i < pud->pud_ngroups; i++)
2659                         __swab32s(&pud->pud_groups[i]);
2660         }
2661
2662         return 0;
2663 }
2664 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2665
2666 /*
2667  * misc helpers
2668  */
2669
2670 const char *sec2target_str(struct ptlrpc_sec *sec)
2671 {
2672         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2673                 return "*";
2674         if (sec_is_reverse(sec))
2675                 return "c";
2676         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2677 }
2678 EXPORT_SYMBOL(sec2target_str);
2679
2680 /*
2681  * return true if the bulk data is protected
2682  */
2683 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2684 {
2685         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2686         case SPTLRPC_BULK_SVC_INTG:
2687         case SPTLRPC_BULK_SVC_PRIV:
2688                 return 1;
2689         default:
2690                 return 0;
2691         }
2692 }
2693 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2694
2695 /*
2696  * crypto API helper/alloc blkciper
2697  */
2698
2699 /*
2700  * initialize/finalize
2701  */
2702
2703 int sptlrpc_init(void)
2704 {
2705         int rc;
2706
2707         rwlock_init(&policy_lock);
2708
2709         rc = sptlrpc_gc_init();
2710         if (rc)
2711                 goto out;
2712
2713         rc = sptlrpc_conf_init();
2714         if (rc)
2715                 goto out_gc;
2716
2717         rc = sptlrpc_enc_pool_init();
2718         if (rc)
2719                 goto out_conf;
2720
2721         rc = sptlrpc_null_init();
2722         if (rc)
2723                 goto out_pool;
2724
2725         rc = sptlrpc_plain_init();
2726         if (rc)
2727                 goto out_null;
2728
2729         rc = sptlrpc_lproc_init();
2730         if (rc)
2731                 goto out_plain;
2732
2733         return 0;
2734
2735 out_plain:
2736         sptlrpc_plain_fini();
2737 out_null:
2738         sptlrpc_null_fini();
2739 out_pool:
2740         sptlrpc_enc_pool_fini();
2741 out_conf:
2742         sptlrpc_conf_fini();
2743 out_gc:
2744         sptlrpc_gc_fini();
2745 out:
2746         return rc;
2747 }
2748
2749 void sptlrpc_fini(void)
2750 {
2751         sptlrpc_lproc_fini();
2752         sptlrpc_plain_fini();
2753         sptlrpc_null_fini();
2754         sptlrpc_enc_pool_fini();
2755         sptlrpc_conf_fini();
2756         sptlrpc_gc_fini();
2757 }