drivers/staging/lustre/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, 2012, 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 "../../include/linux/libcfs/libcfs.h"
  40 #include <linux/crypto.h>
  41 #include <linux/key.h>
  42
  43 #include "../include/obd.h"
  44 #include "../include/obd_class.h"
  45 #include "../include/obd_support.h"
  46 #include "../include/lustre_net.h"
  47 #include "../include/lustre_import.h"
  48 #include "../include/lustre_dlm.h"
  49 #include "../include/lustre_sec.h"
  50
  51 #include "ptlrpc_internal.h"
  52
  53 /***********************************************
  54  * policy registers                         *
  55  ***********************************************/
  56
  57 static rwlock_t policy_lock;
  58 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  59         NULL,
  60 };
  61
  62 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  63 {
  64         __u16 number = policy->sp_policy;
  65
  66         LASSERT(policy->sp_name);
  67         LASSERT(policy->sp_cops);
  68         LASSERT(policy->sp_sops);
  69
  70         if (number >= SPTLRPC_POLICY_MAX)
  71                 return -EINVAL;
  72
  73         write_lock(&policy_lock);
  74         if (unlikely(policies[number])) {
  75                 write_unlock(&policy_lock);
  76                 return -EALREADY;
  77         }
  78         policies[number] = policy;
  79         write_unlock(&policy_lock);
  80
  81         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  82         return 0;
  83 }
  84 EXPORT_SYMBOL(sptlrpc_register_policy);
  85
  86 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  87 {
  88         __u16 number = policy->sp_policy;
  89
  90         LASSERT(number < SPTLRPC_POLICY_MAX);
  91
  92         write_lock(&policy_lock);
  93         if (unlikely(!policies[number])) {
  94                 write_unlock(&policy_lock);
  95                 CERROR("%s: already unregistered\n", policy->sp_name);
  96                 return -EINVAL;
  97         }
  98
  99         LASSERT(policies[number] == policy);
 100         policies[number] = NULL;
 101         write_unlock(&policy_lock);
 102
 103         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 104         return 0;
 105 }
 106 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 107
 108 static
 109 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
 110 {
 111         static DEFINE_MUTEX(load_mutex);
 112         static atomic_t loaded = ATOMIC_INIT(0);
 113         struct ptlrpc_sec_policy *policy;
 114         __u16 number = SPTLRPC_FLVR_POLICY(flavor);
 115         __u16 flag = 0;
 116
 117         if (number >= SPTLRPC_POLICY_MAX)
 118                 return NULL;
 119
 120         while (1) {
 121                 read_lock(&policy_lock);
 122                 policy = policies[number];
 123                 if (policy && !try_module_get(policy->sp_owner))
 124                         policy = NULL;
 125                 if (!policy)
 126                         flag = atomic_read(&loaded);
 127                 read_unlock(&policy_lock);
 128
 129                 if (policy || flag != 0 ||
 130                     number != SPTLRPC_POLICY_GSS)
 131                         break;
 132
 133                 /* try to load gss module, once */
 134                 mutex_lock(&load_mutex);
 135                 if (atomic_read(&loaded) == 0) {
 136                         if (request_module("ptlrpc_gss") == 0)
 137                                 CDEBUG(D_SEC,
 138                                        "module ptlrpc_gss loaded on demand\n");
 139                         else
 140                                 CERROR("Unable to load module ptlrpc_gss\n");
 141
 142                         atomic_set(&loaded, 1);
 143                 }
 144                 mutex_unlock(&load_mutex);
 145         }
 146
 147         return policy;
 148 }
 149
 150 __u32 sptlrpc_name2flavor_base(const char *name)
 151 {
 152         if (!strcmp(name, "null"))
 153                 return SPTLRPC_FLVR_NULL;
 154         if (!strcmp(name, "plain"))
 155                 return SPTLRPC_FLVR_PLAIN;
 156         if (!strcmp(name, "krb5n"))
 157                 return SPTLRPC_FLVR_KRB5N;
 158         if (!strcmp(name, "krb5a"))
 159                 return SPTLRPC_FLVR_KRB5A;
 160         if (!strcmp(name, "krb5i"))
 161                 return SPTLRPC_FLVR_KRB5I;
 162         if (!strcmp(name, "krb5p"))
 163                 return SPTLRPC_FLVR_KRB5P;
 164
 165         return SPTLRPC_FLVR_INVALID;
 166 }
 167 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 168
 169 const char *sptlrpc_flavor2name_base(__u32 flvr)
 170 {
 171         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 172
 173         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 174                 return "null";
 175         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 176                 return "plain";
 177         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 178                 return "krb5n";
 179         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 180                 return "krb5a";
 181         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 182                 return "krb5i";
 183         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 184                 return "krb5p";
 185
 186         CERROR("invalid wire flavor 0x%x\n", flvr);
 187         return "invalid";
 188 }
 189 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 190
 191 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 192                                char *buf, int bufsize)
 193 {
 194         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 195                 snprintf(buf, bufsize, "hash:%s",
 196                          sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 197         else
 198                 snprintf(buf, bufsize, "%s",
 199                          sptlrpc_flavor2name_base(sf->sf_rpc));
 200
 201         buf[bufsize - 1] = '\0';
 202         return buf;
 203 }
 204 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 205
 206 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 207 {
 208         strlcpy(buf, sptlrpc_flavor2name_base(sf->sf_rpc), bufsize);
 209
 210         /*
 211          * currently we don't support customized bulk specification for
 212          * flavors other than plain
 213          */
 214         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 215                 char bspec[16];
 216
 217                 bspec[0] = '-';
 218                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 219                 strlcat(buf, bspec, bufsize);
 220         }
 221
 222         return buf;
 223 }
 224 EXPORT_SYMBOL(sptlrpc_flavor2name);
 225
 226 static char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 227 {
 228         buf[0] = '\0';
 229
 230         if (flags & PTLRPC_SEC_FL_REVERSE)
 231                 strlcat(buf, "reverse,", bufsize);
 232         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 233                 strlcat(buf, "rootonly,", bufsize);
 234         if (flags & PTLRPC_SEC_FL_UDESC)
 235                 strlcat(buf, "udesc,", bufsize);
 236         if (flags & PTLRPC_SEC_FL_BULK)
 237                 strlcat(buf, "bulk,", bufsize);
 238         if (buf[0] == '\0')
 239                 strlcat(buf, "-,", bufsize);
 240
 241         return buf;
 242 }
 243
 244 /**************************************************
 245  * client context APIs                      *
 246  **************************************************/
 247
 248 static
 249 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 250 {
 251         struct vfs_cred vcred;
 252         int create = 1, remove_dead = 1;
 253
 254         LASSERT(sec);
 255         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 256
 257         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 258                                      PTLRPC_SEC_FL_ROOTONLY)) {
 259                 vcred.vc_uid = 0;
 260                 vcred.vc_gid = 0;
 261                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 262                         create = 0;
 263                         remove_dead = 0;
 264                 }
 265         } else {
 266                 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
 267                 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
 268         }
 269
 270         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
 271                                                    create, remove_dead);
 272 }
 273
 274 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 275 {
 276         atomic_inc(&ctx->cc_refcount);
 277         return ctx;
 278 }
 279 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 280
 281 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 282 {
 283         struct ptlrpc_sec *sec = ctx->cc_sec;
 284
 285         LASSERT(sec);
 286         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 287
 288         if (!atomic_dec_and_test(&ctx->cc_refcount))
 289                 return;
 290
 291         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 292 }
 293 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 294
 295 static int import_sec_check_expire(struct obd_import *imp)
 296 {
 297         int adapt = 0;
 298
 299         spin_lock(&imp->imp_lock);
 300         if (imp->imp_sec_expire &&
 301             imp->imp_sec_expire < ktime_get_real_seconds()) {
 302                 adapt = 1;
 303                 imp->imp_sec_expire = 0;
 304         }
 305         spin_unlock(&imp->imp_lock);
 306
 307         if (!adapt)
 308                 return 0;
 309
 310         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 311         return sptlrpc_import_sec_adapt(imp, NULL, NULL);
 312 }
 313
 314 /**
 315  * Get and validate the client side ptlrpc security facilities from
 316  * \a imp. There is a race condition on client reconnect when the import is
 317  * being destroyed while there are outstanding client bound requests. In
 318  * this case do not output any error messages if import secuity is not
 319  * found.
 320  *
 321  * \param[in] imp obd import associated with client
 322  * \param[out] sec client side ptlrpc security
 323  *
 324  * \retval 0 if security retrieved successfully
 325  * \retval -ve errno if there was a problem
 326  */
 327 static int import_sec_validate_get(struct obd_import *imp,
 328                                    struct ptlrpc_sec **sec)
 329 {
 330         int rc;
 331
 332         if (unlikely(imp->imp_sec_expire)) {
 333                 rc = import_sec_check_expire(imp);
 334                 if (rc)
 335                         return rc;
 336         }
 337
 338         *sec = sptlrpc_import_sec_ref(imp);
 339         /* Only output an error when the import is still active */
 340         if (!*sec) {
 341                 if (list_empty(&imp->imp_zombie_chain))
 342                         CERROR("import %p (%s) with no sec\n",
 343                                imp, ptlrpc_import_state_name(imp->imp_state));
 344                 return -EACCES;
 345         }
 346
 347         if (unlikely((*sec)->ps_dying)) {
 348                 CERROR("attempt to use dying sec %p\n", sec);
 349                 sptlrpc_sec_put(*sec);
 350                 return -EACCES;
 351         }
 352
 353         return 0;
 354 }
 355
 356 /**
 357  * Given a \a req, find or allocate a appropriate context for it.
 358  * \pre req->rq_cli_ctx == NULL.
 359  *
 360  * \retval 0 succeed, and req->rq_cli_ctx is set.
 361  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 362  */
 363 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 364 {
 365         struct obd_import *imp = req->rq_import;
 366         struct ptlrpc_sec *sec;
 367         int             rc;
 368
 369         LASSERT(!req->rq_cli_ctx);
 370         LASSERT(imp);
 371
 372         rc = import_sec_validate_get(imp, &sec);
 373         if (rc)
 374                 return rc;
 375
 376         req->rq_cli_ctx = get_my_ctx(sec);
 377
 378         sptlrpc_sec_put(sec);
 379
 380         if (!req->rq_cli_ctx) {
 381                 CERROR("req %p: fail to get context\n", req);
 382                 return -ENOMEM;
 383         }
 384
 385         return 0;
 386 }
 387
 388 /**
 389  * Drop the context for \a req.
 390  * \pre req->rq_cli_ctx != NULL.
 391  * \post req->rq_cli_ctx == NULL.
 392  *
 393  * If \a sync == 0, this function should return quickly without sleep;
 394  * otherwise it might trigger and wait for the whole process of sending
 395  * an context-destroying rpc to server.
 396  */
 397 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 398 {
 399         LASSERT(req);
 400         LASSERT(req->rq_cli_ctx);
 401
 402         /* request might be asked to release earlier while still
 403          * in the context waiting list.
 404          */
 405         if (!list_empty(&req->rq_ctx_chain)) {
 406                 spin_lock(&req->rq_cli_ctx->cc_lock);
 407                 list_del_init(&req->rq_ctx_chain);
 408                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 409         }
 410
 411         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 412         req->rq_cli_ctx = NULL;
 413 }
 414
 415 static
 416 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 417                            struct ptlrpc_cli_ctx *oldctx,
 418                            struct ptlrpc_cli_ctx *newctx)
 419 {
 420         struct sptlrpc_flavor old_flvr;
 421         char *reqmsg = NULL; /* to workaround old gcc */
 422         int reqmsg_size;
 423         int rc = 0;
 424
 425         LASSERT(req->rq_reqmsg);
 426         LASSERT(req->rq_reqlen);
 427         LASSERT(req->rq_replen);
 428
 429         CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
 430                req,
 431                oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
 432                newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
 433                oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
 434                newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 435
 436         /* save flavor */
 437         old_flvr = req->rq_flvr;
 438
 439         /* save request message */
 440         reqmsg_size = req->rq_reqlen;
 441         if (reqmsg_size != 0) {
 442                 reqmsg = libcfs_kvzalloc(reqmsg_size, GFP_NOFS);
 443                 if (!reqmsg)
 444                         return -ENOMEM;
 445                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 446         }
 447
 448         /* release old req/rep buf */
 449         req->rq_cli_ctx = oldctx;
 450         sptlrpc_cli_free_reqbuf(req);
 451         sptlrpc_cli_free_repbuf(req);
 452         req->rq_cli_ctx = newctx;
 453
 454         /* recalculate the flavor */
 455         sptlrpc_req_set_flavor(req, 0);
 456
 457         /* alloc new request buffer
 458          * we don't need to alloc reply buffer here, leave it to the
 459          * rest procedure of ptlrpc
 460          */
 461         if (reqmsg_size != 0) {
 462                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 463                 if (!rc) {
 464                         LASSERT(req->rq_reqmsg);
 465                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 466                 } else {
 467                         CWARN("failed to alloc reqbuf: %d\n", rc);
 468                         req->rq_flvr = old_flvr;
 469                 }
 470
 471                 kvfree(reqmsg);
 472         }
 473         return rc;
 474 }
 475
 476 /**
 477  * If current context of \a req is dead somehow, e.g. we just switched flavor
 478  * thus marked original contexts dead, we'll find a new context for it. if
 479  * no switch is needed, \a req will end up with the same context.
 480  *
 481  * \note a request must have a context, to keep other parts of code happy.
 482  * In any case of failure during the switching, we must restore the old one.
 483  */
 484 static int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 485 {
 486         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 487         struct ptlrpc_cli_ctx *newctx;
 488         int rc;
 489
 490         LASSERT(oldctx);
 491
 492         sptlrpc_cli_ctx_get(oldctx);
 493         sptlrpc_req_put_ctx(req, 0);
 494
 495         rc = sptlrpc_req_get_ctx(req);
 496         if (unlikely(rc)) {
 497                 LASSERT(!req->rq_cli_ctx);
 498
 499                 /* restore old ctx */
 500                 req->rq_cli_ctx = oldctx;
 501                 return rc;
 502         }
 503
 504         newctx = req->rq_cli_ctx;
 505         LASSERT(newctx);
 506
 507         if (unlikely(newctx == oldctx &&
 508                      test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 509                 /*
 510                  * still get the old dead ctx, usually means system too busy
 511                  */
 512                 CDEBUG(D_SEC,
 513                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 514                        newctx, newctx->cc_flags);
 515
 516                 set_current_state(TASK_INTERRUPTIBLE);
 517                 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
 518         } else {
 519                 /*
 520                  * it's possible newctx == oldctx if we're switching
 521                  * subflavor with the same sec.
 522                  */
 523                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 524                 if (rc) {
 525                         /* restore old ctx */
 526                         sptlrpc_req_put_ctx(req, 0);
 527                         req->rq_cli_ctx = oldctx;
 528                         return rc;
 529                 }
 530
 531                 LASSERT(req->rq_cli_ctx == newctx);
 532         }
 533
 534         sptlrpc_cli_ctx_put(oldctx, 1);
 535         return 0;
 536 }
 537
 538 static
 539 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 540 {
 541         if (cli_ctx_is_refreshed(ctx))
 542                 return 1;
 543         return 0;
 544 }
 545
 546 static
 547 int ctx_refresh_timeout(void *data)
 548 {
 549         struct ptlrpc_request *req = data;
 550         int rc;
 551
 552         /* conn_cnt is needed in expire_one_request */
 553         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 554
 555         rc = ptlrpc_expire_one_request(req, 1);
 556         /* if we started recovery, we should mark this ctx dead; otherwise
 557          * in case of lgssd died nobody would retire this ctx, following
 558          * connecting will still find the same ctx thus cause deadlock.
 559          * there's an assumption that expire time of the request should be
 560          * later than the context refresh expire time.
 561          */
 562         if (rc == 0)
 563                 req->rq_cli_ctx->cc_ops->force_die(req->rq_cli_ctx, 0);
 564         return rc;
 565 }
 566
 567 static
 568 void ctx_refresh_interrupt(void *data)
 569 {
 570         struct ptlrpc_request *req = data;
 571
 572         spin_lock(&req->rq_lock);
 573         req->rq_intr = 1;
 574         spin_unlock(&req->rq_lock);
 575 }
 576
 577 static
 578 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 579 {
 580         spin_lock(&ctx->cc_lock);
 581         if (!list_empty(&req->rq_ctx_chain))
 582                 list_del_init(&req->rq_ctx_chain);
 583         spin_unlock(&ctx->cc_lock);
 584 }
 585
 586 /**
 587  * To refresh the context of \req, if it's not up-to-date.
 588  * \param timeout
 589  * - < 0: don't wait
 590  * - = 0: wait until success or fatal error occur
 591  * - > 0: timeout value (in seconds)
 592  *
 593  * The status of the context could be subject to be changed by other threads
 594  * at any time. We allow this race, but once we return with 0, the caller will
 595  * suppose it's uptodated and keep using it until the owning rpc is done.
 596  *
 597  * \retval 0 only if the context is uptodated.
 598  * \retval -ev error number.
 599  */
 600 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 601 {
 602         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 603         struct ptlrpc_sec *sec;
 604         struct l_wait_info lwi;
 605         int rc;
 606
 607         LASSERT(ctx);
 608
 609         if (req->rq_ctx_init || req->rq_ctx_fini)
 610                 return 0;
 611
 612         /*
 613          * during the process a request's context might change type even
 614          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 615          * everything
 616          */
 617 again:
 618         rc = import_sec_validate_get(req->rq_import, &sec);
 619         if (rc)
 620                 return rc;
 621
 622         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 623                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 624                        req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 625                 req_off_ctx_list(req, ctx);
 626                 sptlrpc_req_replace_dead_ctx(req);
 627                 ctx = req->rq_cli_ctx;
 628         }
 629         sptlrpc_sec_put(sec);
 630
 631         if (cli_ctx_is_eternal(ctx))
 632                 return 0;
 633
 634         if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 635                 LASSERT(ctx->cc_ops->refresh);
 636                 ctx->cc_ops->refresh(ctx);
 637         }
 638         LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 639
 640         LASSERT(ctx->cc_ops->validate);
 641         if (ctx->cc_ops->validate(ctx) == 0) {
 642                 req_off_ctx_list(req, ctx);
 643                 return 0;
 644         }
 645
 646         if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 647                 spin_lock(&req->rq_lock);
 648                 req->rq_err = 1;
 649                 spin_unlock(&req->rq_lock);
 650                 req_off_ctx_list(req, ctx);
 651                 return -EPERM;
 652         }
 653
 654         /*
 655          * There's a subtle issue for resending RPCs, suppose following
 656          * situation:
 657          *  1. the request was sent to server.
 658          *  2. recovery was kicked start, after finished the request was
 659          *     marked as resent.
 660          *  3. resend the request.
 661          *  4. old reply from server received, we accept and verify the reply.
 662          *     this has to be success, otherwise the error will be aware
 663          *     by application.
 664          *  5. new reply from server received, dropped by LNet.
 665          *
 666          * Note the xid of old & new request is the same. We can't simply
 667          * change xid for the resent request because the server replies on
 668          * it for reply reconstruction.
 669          *
 670          * Commonly the original context should be uptodate because we
 671          * have a expiry nice time; server will keep its context because
 672          * we at least hold a ref of old context which prevent context
 673          * destroying RPC being sent. So server still can accept the request
 674          * and finish the RPC. But if that's not the case:
 675          *  1. If server side context has been trimmed, a NO_CONTEXT will
 676          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 677          *     context by force.
 678          *  2. Current context never be refreshed, then we are fine: we
 679          *     never really send request with old context before.
 680          */
 681         if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 682             unlikely(req->rq_reqmsg) &&
 683             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 684                 req_off_ctx_list(req, ctx);
 685                 return 0;
 686         }
 687
 688         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 689                 req_off_ctx_list(req, ctx);
 690                 /*
 691                  * don't switch ctx if import was deactivated
 692                  */
 693                 if (req->rq_import->imp_deactive) {
 694                         spin_lock(&req->rq_lock);
 695                         req->rq_err = 1;
 696                         spin_unlock(&req->rq_lock);
 697                         return -EINTR;
 698                 }
 699
 700                 rc = sptlrpc_req_replace_dead_ctx(req);
 701                 if (rc) {
 702                         LASSERT(ctx == req->rq_cli_ctx);
 703                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 704                                req, ctx, rc);
 705                         spin_lock(&req->rq_lock);
 706                         req->rq_err = 1;
 707                         spin_unlock(&req->rq_lock);
 708                         return rc;
 709                 }
 710
 711                 ctx = req->rq_cli_ctx;
 712                 goto again;
 713         }
 714
 715         /*
 716          * Now we're sure this context is during upcall, add myself into
 717          * waiting list
 718          */
 719         spin_lock(&ctx->cc_lock);
 720         if (list_empty(&req->rq_ctx_chain))
 721                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 722         spin_unlock(&ctx->cc_lock);
 723
 724         if (timeout < 0)
 725                 return -EWOULDBLOCK;
 726
 727         /* Clear any flags that may be present from previous sends */
 728         LASSERT(req->rq_receiving_reply == 0);
 729         spin_lock(&req->rq_lock);
 730         req->rq_err = 0;
 731         req->rq_timedout = 0;
 732         req->rq_resend = 0;
 733         req->rq_restart = 0;
 734         spin_unlock(&req->rq_lock);
 735
 736         lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
 737                                ctx_refresh_timeout, ctx_refresh_interrupt,
 738                                req);
 739         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 740
 741         /*
 742          * following cases could lead us here:
 743          * - successfully refreshed;
 744          * - interrupted;
 745          * - timedout, and we don't want recover from the failure;
 746          * - timedout, and waked up upon recovery finished;
 747          * - someone else mark this ctx dead by force;
 748          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 749          *   e.g. ptlrpc_abort_inflight();
 750          */
 751         if (!cli_ctx_is_refreshed(ctx)) {
 752                 /* timed out or interrupted */
 753                 req_off_ctx_list(req, ctx);
 754
 755                 LASSERT(rc != 0);
 756                 return rc;
 757         }
 758
 759         goto again;
 760 }
 761
 762 /**
 763  * Initialize flavor settings for \a req, according to \a opcode.
 764  *
 765  * \note this could be called in two situations:
 766  * - new request from ptlrpc_pre_req(), with proper @opcode
 767  * - old request which changed ctx in the middle, with @opcode == 0
 768  */
 769 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 770 {
 771         struct ptlrpc_sec *sec;
 772
 773         LASSERT(req->rq_import);
 774         LASSERT(req->rq_cli_ctx);
 775         LASSERT(req->rq_cli_ctx->cc_sec);
 776         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 777
 778         /* special security flags according to opcode */
 779         switch (opcode) {
 780         case OST_READ:
 781         case MDS_READPAGE:
 782         case MGS_CONFIG_READ:
 783         case OBD_IDX_READ:
 784                 req->rq_bulk_read = 1;
 785                 break;
 786         case OST_WRITE:
 787         case MDS_WRITEPAGE:
 788                 req->rq_bulk_write = 1;
 789                 break;
 790         case SEC_CTX_INIT:
 791                 req->rq_ctx_init = 1;
 792                 break;
 793         case SEC_CTX_FINI:
 794                 req->rq_ctx_fini = 1;
 795                 break;
 796         case 0:
 797                 /* init/fini rpc won't be resend, so can't be here */
 798                 LASSERT(req->rq_ctx_init == 0);
 799                 LASSERT(req->rq_ctx_fini == 0);
 800
 801                 /* cleanup flags, which should be recalculated */
 802                 req->rq_pack_udesc = 0;
 803                 req->rq_pack_bulk = 0;
 804                 break;
 805         }
 806
 807         sec = req->rq_cli_ctx->cc_sec;
 808
 809         spin_lock(&sec->ps_lock);
 810         req->rq_flvr = sec->ps_flvr;
 811         spin_unlock(&sec->ps_lock);
 812
 813         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
 814          * destruction rpc
 815          */
 816         if (unlikely(req->rq_ctx_init))
 817                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 818         else if (unlikely(req->rq_ctx_fini))
 819                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 820
 821         /* user descriptor flag, null security can't do it anyway */
 822         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 823             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 824                 req->rq_pack_udesc = 1;
 825
 826         /* bulk security flag */
 827         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 828             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 829                 req->rq_pack_bulk = 1;
 830 }
 831
 832 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 833 {
 834         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 835                 return;
 836
 837         LASSERT(req->rq_clrbuf);
 838         if (req->rq_pool || !req->rq_reqbuf)
 839                 return;
 840
 841         kfree(req->rq_reqbuf);
 842         req->rq_reqbuf = NULL;
 843         req->rq_reqbuf_len = 0;
 844 }
 845
 846 /**
 847  * Given an import \a imp, check whether current user has a valid context
 848  * or not. We may create a new context and try to refresh it, and try
 849  * repeatedly try in case of non-fatal errors. Return 0 means success.
 850  */
 851 int sptlrpc_import_check_ctx(struct obd_import *imp)
 852 {
 853         struct ptlrpc_sec *sec;
 854         struct ptlrpc_cli_ctx *ctx;
 855         struct ptlrpc_request *req = NULL;
 856         int rc;
 857
 858         might_sleep();
 859
 860         sec = sptlrpc_import_sec_ref(imp);
 861         ctx = get_my_ctx(sec);
 862         sptlrpc_sec_put(sec);
 863
 864         if (!ctx)
 865                 return -ENOMEM;
 866
 867         if (cli_ctx_is_eternal(ctx) ||
 868             ctx->cc_ops->validate(ctx) == 0) {
 869                 sptlrpc_cli_ctx_put(ctx, 1);
 870                 return 0;
 871         }
 872
 873         if (cli_ctx_is_error(ctx)) {
 874                 sptlrpc_cli_ctx_put(ctx, 1);
 875                 return -EACCES;
 876         }
 877
 878         req = ptlrpc_request_cache_alloc(GFP_NOFS);
 879         if (!req)
 880                 return -ENOMEM;
 881
 882         ptlrpc_cli_req_init(req);
 883         atomic_set(&req->rq_refcount, 10000);
 884
 885         req->rq_import = imp;
 886         req->rq_flvr = sec->ps_flvr;
 887         req->rq_cli_ctx = ctx;
 888
 889         rc = sptlrpc_req_refresh_ctx(req, 0);
 890         LASSERT(list_empty(&req->rq_ctx_chain));
 891         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 892         ptlrpc_request_cache_free(req);
 893
 894         return rc;
 895 }
 896
 897 /**
 898  * Used by ptlrpc client, to perform the pre-defined security transformation
 899  * upon the request message of \a req. After this function called,
 900  * req->rq_reqmsg is still accessible as clear text.
 901  */
 902 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 903 {
 904         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 905         int rc = 0;
 906
 907         LASSERT(ctx);
 908         LASSERT(ctx->cc_sec);
 909         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
 910
 911         /* we wrap bulk request here because now we can be sure
 912          * the context is uptodate.
 913          */
 914         if (req->rq_bulk) {
 915                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
 916                 if (rc)
 917                         return rc;
 918         }
 919
 920         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 921         case SPTLRPC_SVC_NULL:
 922         case SPTLRPC_SVC_AUTH:
 923         case SPTLRPC_SVC_INTG:
 924                 LASSERT(ctx->cc_ops->sign);
 925                 rc = ctx->cc_ops->sign(ctx, req);
 926                 break;
 927         case SPTLRPC_SVC_PRIV:
 928                 LASSERT(ctx->cc_ops->seal);
 929                 rc = ctx->cc_ops->seal(ctx, req);
 930                 break;
 931         default:
 932                 LBUG();
 933         }
 934
 935         if (rc == 0) {
 936                 LASSERT(req->rq_reqdata_len);
 937                 LASSERT(req->rq_reqdata_len % 8 == 0);
 938                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
 939         }
 940
 941         return rc;
 942 }
 943
 944 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
 945 {
 946         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 947         int rc;
 948
 949         LASSERT(ctx);
 950         LASSERT(ctx->cc_sec);
 951         LASSERT(req->rq_repbuf);
 952         LASSERT(req->rq_repdata);
 953         LASSERT(!req->rq_repmsg);
 954
 955         req->rq_rep_swab_mask = 0;
 956
 957         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
 958         switch (rc) {
 959         case 1:
 960                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
 961         case 0:
 962                 break;
 963         default:
 964                 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
 965                 return -EPROTO;
 966         }
 967
 968         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
 969                 CERROR("replied data length %d too small\n",
 970                        req->rq_repdata_len);
 971                 return -EPROTO;
 972         }
 973
 974         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
 975             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
 976                 CERROR("reply policy %u doesn't match request policy %u\n",
 977                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
 978                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
 979                 return -EPROTO;
 980         }
 981
 982         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 983         case SPTLRPC_SVC_NULL:
 984         case SPTLRPC_SVC_AUTH:
 985         case SPTLRPC_SVC_INTG:
 986                 LASSERT(ctx->cc_ops->verify);
 987                 rc = ctx->cc_ops->verify(ctx, req);
 988                 break;
 989         case SPTLRPC_SVC_PRIV:
 990                 LASSERT(ctx->cc_ops->unseal);
 991                 rc = ctx->cc_ops->unseal(ctx, req);
 992                 break;
 993         default:
 994                 LBUG();
 995         }
 996         LASSERT(rc || req->rq_repmsg || req->rq_resend);
 997
 998         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
 999             !req->rq_ctx_init)
1000                 req->rq_rep_swab_mask = 0;
1001         return rc;
1002 }
1003
1004 /**
1005  * Used by ptlrpc client, to perform security transformation upon the reply
1006  * message of \a req. After return successfully, req->rq_repmsg points to
1007  * the reply message in clear text.
1008  *
1009  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1010  * going to change.
1011  */
1012 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1013 {
1014         LASSERT(req->rq_repbuf);
1015         LASSERT(!req->rq_repdata);
1016         LASSERT(!req->rq_repmsg);
1017         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1018
1019         if (req->rq_reply_off == 0 &&
1020             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1021                 CERROR("real reply with offset 0\n");
1022                 return -EPROTO;
1023         }
1024
1025         if (req->rq_reply_off % 8 != 0) {
1026                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1027                 return -EPROTO;
1028         }
1029
1030         req->rq_repdata = (struct lustre_msg *)
1031                                 (req->rq_repbuf + req->rq_reply_off);
1032         req->rq_repdata_len = req->rq_nob_received;
1033
1034         return do_cli_unwrap_reply(req);
1035 }
1036
1037 /**
1038  * Used by ptlrpc client, to perform security transformation upon the early
1039  * reply message of \a req. We expect the rq_reply_off is 0, and
1040  * rq_nob_received is the early reply size.
1041  *
1042  * Because the receive buffer might be still posted, the reply data might be
1043  * changed at any time, no matter we're holding rq_lock or not. For this reason
1044  * we allocate a separate ptlrpc_request and reply buffer for early reply
1045  * processing.
1046  *
1047  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1048  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1049  * \a *req_ret to release it.
1050  * \retval -ev error number, and \a req_ret will not be set.
1051  */
1052 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1053                                    struct ptlrpc_request **req_ret)
1054 {
1055         struct ptlrpc_request *early_req;
1056         char *early_buf;
1057         int early_bufsz, early_size;
1058         int rc;
1059
1060         early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1061         if (!early_req)
1062                 return -ENOMEM;
1063
1064         ptlrpc_cli_req_init(early_req);
1065
1066         early_size = req->rq_nob_received;
1067         early_bufsz = size_roundup_power2(early_size);
1068         early_buf = libcfs_kvzalloc(early_bufsz, GFP_NOFS);
1069         if (!early_buf) {
1070                 rc = -ENOMEM;
1071                 goto err_req;
1072         }
1073
1074         /* sanity checkings and copy data out, do it inside spinlock */
1075         spin_lock(&req->rq_lock);
1076
1077         if (req->rq_replied) {
1078                 spin_unlock(&req->rq_lock);
1079                 rc = -EALREADY;
1080                 goto err_buf;
1081         }
1082
1083         LASSERT(req->rq_repbuf);
1084         LASSERT(!req->rq_repdata);
1085         LASSERT(!req->rq_repmsg);
1086
1087         if (req->rq_reply_off != 0) {
1088                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1089                 spin_unlock(&req->rq_lock);
1090                 rc = -EPROTO;
1091                 goto err_buf;
1092         }
1093
1094         if (req->rq_nob_received != early_size) {
1095                 /* even another early arrived the size should be the same */
1096                 CERROR("data size has changed from %u to %u\n",
1097                        early_size, req->rq_nob_received);
1098                 spin_unlock(&req->rq_lock);
1099                 rc = -EINVAL;
1100                 goto err_buf;
1101         }
1102
1103         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1104                 CERROR("early reply length %d too small\n",
1105                        req->rq_nob_received);
1106                 spin_unlock(&req->rq_lock);
1107                 rc = -EALREADY;
1108                 goto err_buf;
1109         }
1110
1111         memcpy(early_buf, req->rq_repbuf, early_size);
1112         spin_unlock(&req->rq_lock);
1113
1114         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1115         early_req->rq_flvr = req->rq_flvr;
1116         early_req->rq_repbuf = early_buf;
1117         early_req->rq_repbuf_len = early_bufsz;
1118         early_req->rq_repdata = (struct lustre_msg *)early_buf;
1119         early_req->rq_repdata_len = early_size;
1120         early_req->rq_early = 1;
1121         early_req->rq_reqmsg = req->rq_reqmsg;
1122
1123         rc = do_cli_unwrap_reply(early_req);
1124         if (rc) {
1125                 DEBUG_REQ(D_ADAPTTO, early_req,
1126                           "error %d unwrap early reply", rc);
1127                 goto err_ctx;
1128         }
1129
1130         LASSERT(early_req->rq_repmsg);
1131         *req_ret = early_req;
1132         return 0;
1133
1134 err_ctx:
1135         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1136 err_buf:
1137         kvfree(early_buf);
1138 err_req:
1139         ptlrpc_request_cache_free(early_req);
1140         return rc;
1141 }
1142
1143 /**
1144  * Used by ptlrpc client, to release a processed early reply \a early_req.
1145  *
1146  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1147  */
1148 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1149 {
1150         LASSERT(early_req->rq_repbuf);
1151         LASSERT(early_req->rq_repdata);
1152         LASSERT(early_req->rq_repmsg);
1153
1154         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1155         kvfree(early_req->rq_repbuf);
1156         ptlrpc_request_cache_free(early_req);
1157 }
1158
1159 /**************************************************
1160  * sec ID                                        *
1161  **************************************************/
1162
1163 /*
1164  * "fixed" sec (e.g. null) use sec_id < 0
1165  */
1166 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1167
1168 int sptlrpc_get_next_secid(void)
1169 {
1170         return atomic_inc_return(&sptlrpc_sec_id);
1171 }
1172 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1173
1174 /**************************************************
1175  * client side high-level security APIs    *
1176  **************************************************/
1177
1178 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1179                                    int grace, int force)
1180 {
1181         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1182
1183         LASSERT(policy->sp_cops);
1184         LASSERT(policy->sp_cops->flush_ctx_cache);
1185
1186         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1187 }
1188
1189 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1190 {
1191         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1192
1193         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1194         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1195         LASSERT(policy->sp_cops->destroy_sec);
1196
1197         CDEBUG(D_SEC, "%s@%p: being destroyed\n", sec->ps_policy->sp_name, sec);
1198
1199         policy->sp_cops->destroy_sec(sec);
1200         sptlrpc_policy_put(policy);
1201 }
1202
1203 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1204 {
1205         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1206
1207         if (sec->ps_policy->sp_cops->kill_sec) {
1208                 sec->ps_policy->sp_cops->kill_sec(sec);
1209
1210                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1211         }
1212 }
1213
1214 static struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1215 {
1216         if (sec)
1217                 atomic_inc(&sec->ps_refcount);
1218
1219         return sec;
1220 }
1221
1222 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1223 {
1224         if (sec) {
1225                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1226
1227                 if (atomic_dec_and_test(&sec->ps_refcount)) {
1228                         sptlrpc_gc_del_sec(sec);
1229                         sec_cop_destroy_sec(sec);
1230                 }
1231         }
1232 }
1233 EXPORT_SYMBOL(sptlrpc_sec_put);
1234
1235 /*
1236  * policy module is responsible for taking reference of import
1237  */
1238 static
1239 struct ptlrpc_sec *sptlrpc_sec_create(struct obd_import *imp,
1240                                       struct ptlrpc_svc_ctx *svc_ctx,
1241                                       struct sptlrpc_flavor *sf,
1242                                       enum lustre_sec_part sp)
1243 {
1244         struct ptlrpc_sec_policy *policy;
1245         struct ptlrpc_sec *sec;
1246         char str[32];
1247
1248         if (svc_ctx) {
1249                 LASSERT(imp->imp_dlm_fake == 1);
1250
1251                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1252                        imp->imp_obd->obd_type->typ_name,
1253                        imp->imp_obd->obd_name,
1254                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1255
1256                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1257                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1258         } else {
1259                 LASSERT(imp->imp_dlm_fake == 0);
1260
1261                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1262                        imp->imp_obd->obd_type->typ_name,
1263                        imp->imp_obd->obd_name,
1264                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1265
1266                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1267                 if (!policy) {
1268                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1269                         return NULL;
1270                 }
1271         }
1272
1273         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1274         if (sec) {
1275                 atomic_inc(&sec->ps_refcount);
1276
1277                 sec->ps_part = sp;
1278
1279                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1280                         sptlrpc_gc_add_sec(sec);
1281         } else {
1282                 sptlrpc_policy_put(policy);
1283         }
1284
1285         return sec;
1286 }
1287
1288 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1289 {
1290         struct ptlrpc_sec *sec;
1291
1292         spin_lock(&imp->imp_lock);
1293         sec = sptlrpc_sec_get(imp->imp_sec);
1294         spin_unlock(&imp->imp_lock);
1295
1296         return sec;
1297 }
1298 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1299
1300 static void sptlrpc_import_sec_install(struct obd_import *imp,
1301                                        struct ptlrpc_sec *sec)
1302 {
1303         struct ptlrpc_sec *old_sec;
1304
1305         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1306
1307         spin_lock(&imp->imp_lock);
1308         old_sec = imp->imp_sec;
1309         imp->imp_sec = sec;
1310         spin_unlock(&imp->imp_lock);
1311
1312         if (old_sec) {
1313                 sptlrpc_sec_kill(old_sec);
1314
1315                 /* balance the ref taken by this import */
1316                 sptlrpc_sec_put(old_sec);
1317         }
1318 }
1319
1320 static inline
1321 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1322 {
1323         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1324 }
1325
1326 static inline
1327 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1328 {
1329         *dst = *src;
1330 }
1331
1332 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1333                                              struct ptlrpc_sec *sec,
1334                                              struct sptlrpc_flavor *sf)
1335 {
1336         char str1[32], str2[32];
1337
1338         if (sec->ps_flvr.sf_flags != sf->sf_flags)
1339                 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1340                        sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1341                                             str1, sizeof(str1)),
1342                        sptlrpc_secflags2str(sf->sf_flags,
1343                                             str2, sizeof(str2)));
1344
1345         spin_lock(&sec->ps_lock);
1346         flavor_copy(&sec->ps_flvr, sf);
1347         spin_unlock(&sec->ps_lock);
1348 }
1349
1350 /**
1351  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1352  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1353  *
1354  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1355  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1356  */
1357 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1358                              struct ptlrpc_svc_ctx *svc_ctx,
1359                              struct sptlrpc_flavor *flvr)
1360 {
1361         struct ptlrpc_connection *conn;
1362         struct sptlrpc_flavor sf;
1363         struct ptlrpc_sec *sec, *newsec;
1364         enum lustre_sec_part sp;
1365         char str[24];
1366         int rc = 0;
1367
1368         might_sleep();
1369
1370         if (!imp)
1371                 return 0;
1372
1373         conn = imp->imp_connection;
1374
1375         if (!svc_ctx) {
1376                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1377                 /*
1378                  * normal import, determine flavor from rule set, except
1379                  * for mgc the flavor is predetermined.
1380                  */
1381                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1382                         sf = cliobd->cl_flvr_mgc;
1383                 else
1384                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1385                                                    cliobd->cl_sp_to,
1386                                                    &cliobd->cl_target_uuid,
1387                                                    conn->c_self, &sf);
1388
1389                 sp = imp->imp_obd->u.cli.cl_sp_me;
1390         } else {
1391                 /* reverse import, determine flavor from incoming request */
1392                 sf = *flvr;
1393
1394                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1395                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1396                                       PTLRPC_SEC_FL_ROOTONLY;
1397
1398                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1399         }
1400
1401         sec = sptlrpc_import_sec_ref(imp);
1402         if (sec) {
1403                 char str2[24];
1404
1405                 if (flavor_equal(&sf, &sec->ps_flvr))
1406                         goto out;
1407
1408                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1409                        imp->imp_obd->obd_name,
1410                        obd_uuid2str(&conn->c_remote_uuid),
1411                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1412                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1413
1414                 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1415                     SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1416                     SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1417                     SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1418                         sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1419                         goto out;
1420                 }
1421         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1422                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1423                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1424                        imp->imp_obd->obd_name,
1425                        obd_uuid2str(&conn->c_remote_uuid),
1426                        LNET_NIDNET(conn->c_self),
1427                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1428         }
1429
1430         mutex_lock(&imp->imp_sec_mutex);
1431
1432         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1433         if (newsec) {
1434                 sptlrpc_import_sec_install(imp, newsec);
1435         } else {
1436                 CERROR("import %s->%s: failed to create new sec\n",
1437                        imp->imp_obd->obd_name,
1438                        obd_uuid2str(&conn->c_remote_uuid));
1439                 rc = -EPERM;
1440         }
1441
1442         mutex_unlock(&imp->imp_sec_mutex);
1443 out:
1444         sptlrpc_sec_put(sec);
1445         return rc;
1446 }
1447
1448 void sptlrpc_import_sec_put(struct obd_import *imp)
1449 {
1450         if (imp->imp_sec) {
1451                 sptlrpc_sec_kill(imp->imp_sec);
1452
1453                 sptlrpc_sec_put(imp->imp_sec);
1454                 imp->imp_sec = NULL;
1455         }
1456 }
1457
1458 static void import_flush_ctx_common(struct obd_import *imp,
1459                                     uid_t uid, int grace, int force)
1460 {
1461         struct ptlrpc_sec *sec;
1462
1463         if (!imp)
1464                 return;
1465
1466         sec = sptlrpc_import_sec_ref(imp);
1467         if (!sec)
1468                 return;
1469
1470         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1471         sptlrpc_sec_put(sec);
1472 }
1473
1474 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1475 {
1476         import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1477                                 1, 1);
1478 }
1479 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1480
1481 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1482 {
1483         import_flush_ctx_common(imp, -1, 1, 1);
1484 }
1485 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1486
1487 /**
1488  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1489  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1490  */
1491 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1492 {
1493         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1494         struct ptlrpc_sec_policy *policy;
1495         int rc;
1496
1497         LASSERT(ctx);
1498         LASSERT(ctx->cc_sec);
1499         LASSERT(ctx->cc_sec->ps_policy);
1500         LASSERT(!req->rq_reqmsg);
1501         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1502
1503         policy = ctx->cc_sec->ps_policy;
1504         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1505         if (!rc) {
1506                 LASSERT(req->rq_reqmsg);
1507                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1508
1509                 /* zeroing preallocated buffer */
1510                 if (req->rq_pool)
1511                         memset(req->rq_reqmsg, 0, msgsize);
1512         }
1513
1514         return rc;
1515 }
1516
1517 /**
1518  * Used by ptlrpc client to free request buffer of \a req. After this
1519  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1520  */
1521 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1522 {
1523         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1524         struct ptlrpc_sec_policy *policy;
1525
1526         LASSERT(ctx);
1527         LASSERT(ctx->cc_sec);
1528         LASSERT(ctx->cc_sec->ps_policy);
1529         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1530
1531         if (!req->rq_reqbuf && !req->rq_clrbuf)
1532                 return;
1533
1534         policy = ctx->cc_sec->ps_policy;
1535         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1536         req->rq_reqmsg = NULL;
1537 }
1538
1539 /*
1540  * NOTE caller must guarantee the buffer size is enough for the enlargement
1541  */
1542 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1543                                   int segment, int newsize)
1544 {
1545         void *src, *dst;
1546         int oldsize, oldmsg_size, movesize;
1547
1548         LASSERT(segment < msg->lm_bufcount);
1549         LASSERT(msg->lm_buflens[segment] <= newsize);
1550
1551         if (msg->lm_buflens[segment] == newsize)
1552                 return;
1553
1554         /* nothing to do if we are enlarging the last segment */
1555         if (segment == msg->lm_bufcount - 1) {
1556                 msg->lm_buflens[segment] = newsize;
1557                 return;
1558         }
1559
1560         oldsize = msg->lm_buflens[segment];
1561
1562         src = lustre_msg_buf(msg, segment + 1, 0);
1563         msg->lm_buflens[segment] = newsize;
1564         dst = lustre_msg_buf(msg, segment + 1, 0);
1565         msg->lm_buflens[segment] = oldsize;
1566
1567         /* move from segment + 1 to end segment */
1568         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1569         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1570         movesize = oldmsg_size - ((unsigned long)src - (unsigned long)msg);
1571         LASSERT(movesize >= 0);
1572
1573         if (movesize)
1574                 memmove(dst, src, movesize);
1575
1576         /* note we don't clear the ares where old data live, not secret */
1577
1578         /* finally set new segment size */
1579         msg->lm_buflens[segment] = newsize;
1580 }
1581 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1582
1583 /**
1584  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1585  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1586  * preserved after the enlargement. this must be called after original request
1587  * buffer being allocated.
1588  *
1589  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1590  * so caller should refresh its local pointers if needed.
1591  */
1592 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1593                                int segment, int newsize)
1594 {
1595         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1596         struct ptlrpc_sec_cops *cops;
1597         struct lustre_msg *msg = req->rq_reqmsg;
1598
1599         LASSERT(ctx);
1600         LASSERT(msg);
1601         LASSERT(msg->lm_bufcount > segment);
1602         LASSERT(msg->lm_buflens[segment] <= newsize);
1603
1604         if (msg->lm_buflens[segment] == newsize)
1605                 return 0;
1606
1607         cops = ctx->cc_sec->ps_policy->sp_cops;
1608         LASSERT(cops->enlarge_reqbuf);
1609         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1610 }
1611 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1612
1613 /**
1614  * Used by ptlrpc client to allocate reply buffer of \a req.
1615  *
1616  * \note After this, req->rq_repmsg is still not accessible.
1617  */
1618 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1619 {
1620         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1621         struct ptlrpc_sec_policy *policy;
1622
1623         LASSERT(ctx);
1624         LASSERT(ctx->cc_sec);
1625         LASSERT(ctx->cc_sec->ps_policy);
1626
1627         if (req->rq_repbuf)
1628                 return 0;
1629
1630         policy = ctx->cc_sec->ps_policy;
1631         return policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize);
1632 }
1633
1634 /**
1635  * Used by ptlrpc client to free reply buffer of \a req. After this
1636  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1637  */
1638 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1639 {
1640         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1641         struct ptlrpc_sec_policy *policy;
1642
1643         LASSERT(ctx);
1644         LASSERT(ctx->cc_sec);
1645         LASSERT(ctx->cc_sec->ps_policy);
1646         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1647
1648         if (!req->rq_repbuf)
1649                 return;
1650         LASSERT(req->rq_repbuf_len);
1651
1652         policy = ctx->cc_sec->ps_policy;
1653         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1654         req->rq_repmsg = NULL;
1655 }
1656
1657 static int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1658                                        struct ptlrpc_svc_ctx *ctx)
1659 {
1660         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1661
1662         if (!policy->sp_sops->install_rctx)
1663                 return 0;
1664         return policy->sp_sops->install_rctx(imp, ctx);
1665 }
1666
1667 /****************************************
1668  * server side security          *
1669  ****************************************/
1670
1671 static int flavor_allowed(struct sptlrpc_flavor *exp,
1672                           struct ptlrpc_request *req)
1673 {
1674         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1675
1676         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1677                 return 1;
1678
1679         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1680             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1681             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1682             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1683                 return 1;
1684
1685         return 0;
1686 }
1687
1688 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1689
1690 /**
1691  * Given an export \a exp, check whether the flavor of incoming \a req
1692  * is allowed by the export \a exp. Main logic is about taking care of
1693  * changing configurations. Return 0 means success.
1694  */
1695 int sptlrpc_target_export_check(struct obd_export *exp,
1696                                 struct ptlrpc_request *req)
1697 {
1698         struct sptlrpc_flavor flavor;
1699
1700         if (!exp)
1701                 return 0;
1702
1703         /* client side export has no imp_reverse, skip
1704          * FIXME maybe we should check flavor this as well???
1705          */
1706         if (!exp->exp_imp_reverse)
1707                 return 0;
1708
1709         /* don't care about ctx fini rpc */
1710         if (req->rq_ctx_fini)
1711                 return 0;
1712
1713         spin_lock(&exp->exp_lock);
1714
1715         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1716          * the first req with the new flavor, then treat it as current flavor,
1717          * adapt reverse sec according to it.
1718          * note the first rpc with new flavor might not be with root ctx, in
1719          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1720          */
1721         if (unlikely(exp->exp_flvr_changed) &&
1722             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1723                 /* make the new flavor as "current", and old ones as
1724                  * about-to-expire
1725                  */
1726                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1727                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1728                 flavor = exp->exp_flvr_old[1];
1729                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1730                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1731                 exp->exp_flvr_old[0] = exp->exp_flvr;
1732                 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1733                                           EXP_FLVR_UPDATE_EXPIRE;
1734                 exp->exp_flvr = flavor;
1735
1736                 /* flavor change finished */
1737                 exp->exp_flvr_changed = 0;
1738                 LASSERT(exp->exp_flvr_adapt == 1);
1739
1740                 /* if it's gss, we only interested in root ctx init */
1741                 if (req->rq_auth_gss &&
1742                     !(req->rq_ctx_init &&
1743                       (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1744                        req->rq_auth_usr_ost))) {
1745                         spin_unlock(&exp->exp_lock);
1746                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1747                                req->rq_auth_gss, req->rq_ctx_init,
1748                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1749                                req->rq_auth_usr_ost);
1750                         return 0;
1751                 }
1752
1753                 exp->exp_flvr_adapt = 0;
1754                 spin_unlock(&exp->exp_lock);
1755
1756                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1757                                                 req->rq_svc_ctx, &flavor);
1758         }
1759
1760         /* if it equals to the current flavor, we accept it, but need to
1761          * dealing with reverse sec/ctx
1762          */
1763         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1764                 /* most cases should return here, we only interested in
1765                  * gss root ctx init
1766                  */
1767                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1768                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1769                      !req->rq_auth_usr_ost)) {
1770                         spin_unlock(&exp->exp_lock);
1771                         return 0;
1772                 }
1773
1774                 /* if flavor just changed, we should not proceed, just leave
1775                  * it and current flavor will be discovered and replaced
1776                  * shortly, and let _this_ rpc pass through
1777                  */
1778                 if (exp->exp_flvr_changed) {
1779                         LASSERT(exp->exp_flvr_adapt);
1780                         spin_unlock(&exp->exp_lock);
1781                         return 0;
1782                 }
1783
1784                 if (exp->exp_flvr_adapt) {
1785                         exp->exp_flvr_adapt = 0;
1786                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1787                                exp, exp->exp_flvr.sf_rpc,
1788                                exp->exp_flvr_old[0].sf_rpc,
1789                                exp->exp_flvr_old[1].sf_rpc);
1790                         flavor = exp->exp_flvr;
1791                         spin_unlock(&exp->exp_lock);
1792
1793                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1794                                                         req->rq_svc_ctx,
1795                                                         &flavor);
1796                 } else {
1797                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
1798                                exp, exp->exp_flvr.sf_rpc,
1799                                exp->exp_flvr_old[0].sf_rpc,
1800                                exp->exp_flvr_old[1].sf_rpc);
1801                         spin_unlock(&exp->exp_lock);
1802
1803                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1804                                                            req->rq_svc_ctx);
1805                 }
1806         }
1807
1808         if (exp->exp_flvr_expire[0]) {
1809                 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
1810                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1811                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
1812                                        exp->exp_flvr.sf_rpc,
1813                                        exp->exp_flvr_old[0].sf_rpc,
1814                                        exp->exp_flvr_old[1].sf_rpc,
1815                                        (s64)(exp->exp_flvr_expire[0] -
1816                                        ktime_get_real_seconds()));
1817                                 spin_unlock(&exp->exp_lock);
1818                                 return 0;
1819                         }
1820                 } else {
1821                         CDEBUG(D_SEC, "mark middle expired\n");
1822                         exp->exp_flvr_expire[0] = 0;
1823                 }
1824                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1825                        exp->exp_flvr.sf_rpc,
1826                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1827                        req->rq_flvr.sf_rpc);
1828         }
1829
1830         /* now it doesn't match the current flavor, the only chance we can
1831          * accept it is match the old flavors which is not expired.
1832          */
1833         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1834                 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
1835                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1836                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
1837                                        exp,
1838                                        exp->exp_flvr.sf_rpc,
1839                                        exp->exp_flvr_old[0].sf_rpc,
1840                                        exp->exp_flvr_old[1].sf_rpc,
1841                                        (s64)(exp->exp_flvr_expire[1] -
1842                                        ktime_get_real_seconds()));
1843                                 spin_unlock(&exp->exp_lock);
1844                                 return 0;
1845                         }
1846                 } else {
1847                         CDEBUG(D_SEC, "mark oldest expired\n");
1848                         exp->exp_flvr_expire[1] = 0;
1849                 }
1850                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1851                        exp, exp->exp_flvr.sf_rpc,
1852                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1853                        req->rq_flvr.sf_rpc);
1854         } else {
1855                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1856                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1857                        exp->exp_flvr_old[1].sf_rpc);
1858         }
1859
1860         spin_unlock(&exp->exp_lock);
1861
1862         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
1863               exp, exp->exp_obd->obd_name,
1864               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1865               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1866               req->rq_flvr.sf_rpc,
1867               exp->exp_flvr.sf_rpc,
1868               exp->exp_flvr_old[0].sf_rpc,
1869               exp->exp_flvr_expire[0] ?
1870               (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
1871               exp->exp_flvr_old[1].sf_rpc,
1872               exp->exp_flvr_expire[1] ?
1873               (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
1874         return -EACCES;
1875 }
1876 EXPORT_SYMBOL(sptlrpc_target_export_check);
1877
1878 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1879 {
1880         /* peer's claim is unreliable unless gss is being used */
1881         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1882                 return svc_rc;
1883
1884         switch (req->rq_sp_from) {
1885         case LUSTRE_SP_CLI:
1886                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1887                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
1888                         svc_rc = SECSVC_DROP;
1889                 }
1890                 break;
1891         case LUSTRE_SP_MDT:
1892                 if (!req->rq_auth_usr_mdt) {
1893                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
1894                         svc_rc = SECSVC_DROP;
1895                 }
1896                 break;
1897         case LUSTRE_SP_OST:
1898                 if (!req->rq_auth_usr_ost) {
1899                         DEBUG_REQ(D_ERROR, req, "faked source OST");
1900                         svc_rc = SECSVC_DROP;
1901                 }
1902                 break;
1903         case LUSTRE_SP_MGS:
1904         case LUSTRE_SP_MGC:
1905                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1906                     !req->rq_auth_usr_ost) {
1907                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
1908                         svc_rc = SECSVC_DROP;
1909                 }
1910                 break;
1911         case LUSTRE_SP_ANY:
1912         default:
1913                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
1914                 svc_rc = SECSVC_DROP;
1915         }
1916
1917         return svc_rc;
1918 }
1919
1920 /**
1921  * Used by ptlrpc server, to perform transformation upon request message of
1922  * incoming \a req. This must be the first thing to do with a incoming
1923  * request in ptlrpc layer.
1924  *
1925  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
1926  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
1927  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
1928  * reply message has been prepared.
1929  * \retval SECSVC_DROP failed, this request should be dropped.
1930  */
1931 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
1932 {
1933         struct ptlrpc_sec_policy *policy;
1934         struct lustre_msg *msg = req->rq_reqbuf;
1935         int rc;
1936
1937         LASSERT(msg);
1938         LASSERT(!req->rq_reqmsg);
1939         LASSERT(!req->rq_repmsg);
1940         LASSERT(!req->rq_svc_ctx);
1941
1942         req->rq_req_swab_mask = 0;
1943
1944         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
1945         switch (rc) {
1946         case 1:
1947                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1948         case 0:
1949                 break;
1950         default:
1951                 CERROR("error unpacking request from %s x%llu\n",
1952                        libcfs_id2str(req->rq_peer), req->rq_xid);
1953                 return SECSVC_DROP;
1954         }
1955
1956         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
1957         req->rq_sp_from = LUSTRE_SP_ANY;
1958         req->rq_auth_uid = -1;
1959         req->rq_auth_mapped_uid = -1;
1960
1961         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
1962         if (!policy) {
1963                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
1964                 return SECSVC_DROP;
1965         }
1966
1967         LASSERT(policy->sp_sops->accept);
1968         rc = policy->sp_sops->accept(req);
1969         sptlrpc_policy_put(policy);
1970         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
1971         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
1972
1973         /*
1974          * if it's not null flavor (which means embedded packing msg),
1975          * reset the swab mask for the coming inner msg unpacking.
1976          */
1977         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
1978                 req->rq_req_swab_mask = 0;
1979
1980         /* sanity check for the request source */
1981         rc = sptlrpc_svc_check_from(req, rc);
1982         return rc;
1983 }
1984
1985 /**
1986  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
1987  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
1988  * a buffer of \a msglen size.
1989  */
1990 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
1991 {
1992         struct ptlrpc_sec_policy *policy;
1993         struct ptlrpc_reply_state *rs;
1994         int rc;
1995
1996         LASSERT(req->rq_svc_ctx);
1997         LASSERT(req->rq_svc_ctx->sc_policy);
1998
1999         policy = req->rq_svc_ctx->sc_policy;
2000         LASSERT(policy->sp_sops->alloc_rs);
2001
2002         rc = policy->sp_sops->alloc_rs(req, msglen);
2003         if (unlikely(rc == -ENOMEM)) {
2004                 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2005
2006                 if (svcpt->scp_service->srv_max_reply_size <
2007                    msglen + sizeof(struct ptlrpc_reply_state)) {
2008                         /* Just return failure if the size is too big */
2009                         CERROR("size of message is too big (%zd), %d allowed\n",
2010                                msglen + sizeof(struct ptlrpc_reply_state),
2011                                svcpt->scp_service->srv_max_reply_size);
2012                         return -ENOMEM;
2013                 }
2014
2015                 /* failed alloc, try emergency pool */
2016                 rs = lustre_get_emerg_rs(svcpt);
2017                 if (!rs)
2018                         return -ENOMEM;
2019
2020                 req->rq_reply_state = rs;
2021                 rc = policy->sp_sops->alloc_rs(req, msglen);
2022                 if (rc) {
2023                         lustre_put_emerg_rs(rs);
2024                         req->rq_reply_state = NULL;
2025                 }
2026         }
2027
2028         LASSERT(rc != 0 ||
2029                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2030
2031         return rc;
2032 }
2033
2034 /**
2035  * Used by ptlrpc server, to perform transformation upon reply message.
2036  *
2037  * \post req->rq_reply_off is set to appropriate server-controlled reply offset.
2038  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2039  */
2040 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2041 {
2042         struct ptlrpc_sec_policy *policy;
2043         int rc;
2044
2045         LASSERT(req->rq_svc_ctx);
2046         LASSERT(req->rq_svc_ctx->sc_policy);
2047
2048         policy = req->rq_svc_ctx->sc_policy;
2049         LASSERT(policy->sp_sops->authorize);
2050
2051         rc = policy->sp_sops->authorize(req);
2052         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2053
2054         return rc;
2055 }
2056
2057 /**
2058  * Used by ptlrpc server, to free reply_state.
2059  */
2060 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2061 {
2062         struct ptlrpc_sec_policy *policy;
2063         unsigned int prealloc;
2064
2065         LASSERT(rs->rs_svc_ctx);
2066         LASSERT(rs->rs_svc_ctx->sc_policy);
2067
2068         policy = rs->rs_svc_ctx->sc_policy;
2069         LASSERT(policy->sp_sops->free_rs);
2070
2071         prealloc = rs->rs_prealloc;
2072         policy->sp_sops->free_rs(rs);
2073
2074         if (prealloc)
2075                 lustre_put_emerg_rs(rs);
2076 }
2077
2078 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2079 {
2080         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2081
2082         if (ctx)
2083                 atomic_inc(&ctx->sc_refcount);
2084 }
2085
2086 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2087 {
2088         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2089
2090         if (!ctx)
2091                 return;
2092
2093         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2094         if (atomic_dec_and_test(&ctx->sc_refcount)) {
2095                 if (ctx->sc_policy->sp_sops->free_ctx)
2096                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2097         }
2098         req->rq_svc_ctx = NULL;
2099 }
2100
2101 /****************************************
2102  * bulk security                        *
2103  ****************************************/
2104
2105 /**
2106  * Perform transformation upon bulk data pointed by \a desc. This is called
2107  * before transforming the request message.
2108  */
2109 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2110                           struct ptlrpc_bulk_desc *desc)
2111 {
2112         struct ptlrpc_cli_ctx *ctx;
2113
2114         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2115
2116         if (!req->rq_pack_bulk)
2117                 return 0;
2118
2119         ctx = req->rq_cli_ctx;
2120         if (ctx->cc_ops->wrap_bulk)
2121                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2122         return 0;
2123 }
2124 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2125
2126 /**
2127  * This is called after unwrap the reply message.
2128  * return nob of actual plain text size received, or error code.
2129  */
2130 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2131                                  struct ptlrpc_bulk_desc *desc,
2132                                  int nob)
2133 {
2134         struct ptlrpc_cli_ctx *ctx;
2135         int rc;
2136
2137         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2138
2139         if (!req->rq_pack_bulk)
2140                 return desc->bd_nob_transferred;
2141
2142         ctx = req->rq_cli_ctx;
2143         if (ctx->cc_ops->unwrap_bulk) {
2144                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2145                 if (rc < 0)
2146                         return rc;
2147         }
2148         return desc->bd_nob_transferred;
2149 }
2150 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2151
2152 /**
2153  * This is called after unwrap the reply message.
2154  * return 0 for success or error code.
2155  */
2156 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2157                                   struct ptlrpc_bulk_desc *desc)
2158 {
2159         struct ptlrpc_cli_ctx *ctx;
2160         int rc;
2161
2162         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2163
2164         if (!req->rq_pack_bulk)
2165                 return 0;
2166
2167         ctx = req->rq_cli_ctx;
2168         if (ctx->cc_ops->unwrap_bulk) {
2169                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2170                 if (rc < 0)
2171                         return rc;
2172         }
2173
2174         /*
2175          * if everything is going right, nob should equals to nob_transferred.
2176          * in case of privacy mode, nob_transferred needs to be adjusted.
2177          */
2178         if (desc->bd_nob != desc->bd_nob_transferred) {
2179                 CERROR("nob %d doesn't match transferred nob %d\n",
2180                        desc->bd_nob, desc->bd_nob_transferred);
2181                 return -EPROTO;
2182         }
2183
2184         return 0;
2185 }
2186 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2187
2188 /****************************************
2189  * user descriptor helpers            *
2190  ****************************************/
2191
2192 int sptlrpc_current_user_desc_size(void)
2193 {
2194         int ngroups;
2195
2196         ngroups = current_ngroups;
2197
2198         if (ngroups > LUSTRE_MAX_GROUPS)
2199                 ngroups = LUSTRE_MAX_GROUPS;
2200         return sptlrpc_user_desc_size(ngroups);
2201 }
2202 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2203
2204 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2205 {
2206         struct ptlrpc_user_desc *pud;
2207
2208         pud = lustre_msg_buf(msg, offset, 0);
2209
2210         if (!pud)
2211                 return -EINVAL;
2212
2213         pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2214         pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2215         pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2216         pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2217         pud->pud_cap = cfs_curproc_cap_pack();
2218         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2219
2220         task_lock(current);
2221         if (pud->pud_ngroups > current_ngroups)
2222                 pud->pud_ngroups = current_ngroups;
2223         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2224                pud->pud_ngroups * sizeof(__u32));
2225         task_unlock(current);
2226
2227         return 0;
2228 }
2229 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2230
2231 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2232 {
2233         struct ptlrpc_user_desc *pud;
2234         int i;
2235
2236         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2237         if (!pud)
2238                 return -EINVAL;
2239
2240         if (swabbed) {
2241                 __swab32s(&pud->pud_uid);
2242                 __swab32s(&pud->pud_gid);
2243                 __swab32s(&pud->pud_fsuid);
2244                 __swab32s(&pud->pud_fsgid);
2245                 __swab32s(&pud->pud_cap);
2246                 __swab32s(&pud->pud_ngroups);
2247         }
2248
2249         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2250                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2251                 return -EINVAL;
2252         }
2253
2254         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2255             msg->lm_buflens[offset]) {
2256                 CERROR("%u groups are claimed but bufsize only %u\n",
2257                        pud->pud_ngroups, msg->lm_buflens[offset]);
2258                 return -EINVAL;
2259         }
2260
2261         if (swabbed) {
2262                 for (i = 0; i < pud->pud_ngroups; i++)
2263                         __swab32s(&pud->pud_groups[i]);
2264         }
2265
2266         return 0;
2267 }
2268 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2269
2270 /****************************************
2271  * misc helpers                  *
2272  ****************************************/
2273
2274 const char *sec2target_str(struct ptlrpc_sec *sec)
2275 {
2276         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2277                 return "*";
2278         if (sec_is_reverse(sec))
2279                 return "c";
2280         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2281 }
2282 EXPORT_SYMBOL(sec2target_str);
2283
2284 /*
2285  * return true if the bulk data is protected
2286  */
2287 bool sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2288 {
2289         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2290         case SPTLRPC_BULK_SVC_INTG:
2291         case SPTLRPC_BULK_SVC_PRIV:
2292                 return true;
2293         default:
2294                 return false;
2295         }
2296 }
2297 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2298
2299 /****************************************
2300  * crypto API helper/alloc blkciper     *
2301  ****************************************/
2302
2303 /****************************************
2304  * initialize/finalize            *
2305  ****************************************/
2306
2307 int sptlrpc_init(void)
2308 {
2309         int rc;
2310
2311         rwlock_init(&policy_lock);
2312
2313         rc = sptlrpc_gc_init();
2314         if (rc)
2315                 goto out;
2316
2317         rc = sptlrpc_conf_init();
2318         if (rc)
2319                 goto out_gc;
2320
2321         rc = sptlrpc_enc_pool_init();
2322         if (rc)
2323                 goto out_conf;
2324
2325         rc = sptlrpc_null_init();
2326         if (rc)
2327                 goto out_pool;
2328
2329         rc = sptlrpc_plain_init();
2330         if (rc)
2331                 goto out_null;
2332
2333         rc = sptlrpc_lproc_init();
2334         if (rc)
2335                 goto out_plain;
2336
2337         return 0;
2338
2339 out_plain:
2340         sptlrpc_plain_fini();
2341 out_null:
2342         sptlrpc_null_fini();
2343 out_pool:
2344         sptlrpc_enc_pool_fini();
2345 out_conf:
2346         sptlrpc_conf_fini();
2347 out_gc:
2348         sptlrpc_gc_fini();
2349 out:
2350         return rc;
2351 }
2352
2353 void sptlrpc_fini(void)
2354 {
2355         sptlrpc_lproc_fini();
2356         sptlrpc_plain_fini();
2357         sptlrpc_null_fini();
2358         sptlrpc_enc_pool_fini();
2359         sptlrpc_conf_fini();
2360         sptlrpc_gc_fini();
2361 }