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