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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ptlrpc/sec.c
  37  *
  38  * Author: Eric Mei <ericm@clusterfs.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_SEC
  42
  43 #include <libcfs/libcfs.h>
  44 #ifndef __KERNEL__
  45 #include <liblustre.h>
  46 #include <libcfs/list.h>
  47 #else
  48 #include <linux/crypto.h>
  49 #include <linux/key.h>
  50 #endif
  51
  52 #include <obd.h>
  53 #include <obd_class.h>
  54 #include <obd_support.h>
  55 #include <lustre_net.h>
  56 #include <lustre_import.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_sec.h>
  59
  60 #include "ptlrpc_internal.h"
  61
  62 /***********************************************
  63  * policy registers                            *
  64  ***********************************************/
  65
  66 static cfs_rwlock_t policy_lock;
  67 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  68         NULL,
  69 };
  70
  71 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  72 {
  73         __u16 number = policy->sp_policy;
  74
  75         LASSERT(policy->sp_name);
  76         LASSERT(policy->sp_cops);
  77         LASSERT(policy->sp_sops);
  78
  79         if (number >= SPTLRPC_POLICY_MAX)
  80                 return -EINVAL;
  81
  82         cfs_write_lock(&policy_lock);
  83         if (unlikely(policies[number])) {
  84                 cfs_write_unlock(&policy_lock);
  85                 return -EALREADY;
  86         }
  87         policies[number] = policy;
  88         cfs_write_unlock(&policy_lock);
  89
  90         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  91         return 0;
  92 }
  93 EXPORT_SYMBOL(sptlrpc_register_policy);
  94
  95 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  96 {
  97         __u16 number = policy->sp_policy;
  98
  99         LASSERT(number < SPTLRPC_POLICY_MAX);
 100
 101         cfs_write_lock(&policy_lock);
 102         if (unlikely(policies[number] == NULL)) {
 103                 cfs_write_unlock(&policy_lock);
 104                 CERROR("%s: already unregistered\n", policy->sp_name);
 105                 return -EINVAL;
 106         }
 107
 108         LASSERT(policies[number] == policy);
 109         policies[number] = NULL;
 110         cfs_write_unlock(&policy_lock);
 111
 112         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 113         return 0;
 114 }
 115 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 116
 117 static
 118 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
 119 {
 120         static CFS_DEFINE_MUTEX(load_mutex);
 121         static cfs_atomic_t       loaded = CFS_ATOMIC_INIT(0);
 122         struct ptlrpc_sec_policy *policy;
 123         __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
 124         __u16                     flag = 0;
 125
 126         if (number >= SPTLRPC_POLICY_MAX)
 127                 return NULL;
 128
 129         while (1) {
 130                 cfs_read_lock(&policy_lock);
 131                 policy = policies[number];
 132                 if (policy && !cfs_try_module_get(policy->sp_owner))
 133                         policy = NULL;
 134                 if (policy == NULL)
 135                         flag = cfs_atomic_read(&loaded);
 136                 cfs_read_unlock(&policy_lock);
 137
 138                 if (policy != NULL || flag != 0 ||
 139                     number != SPTLRPC_POLICY_GSS)
 140                         break;
 141
 142                 /* try to load gss module, once */
 143                 cfs_mutex_lock(&load_mutex);
 144                 if (cfs_atomic_read(&loaded) == 0) {
 145                         if (cfs_request_module("ptlrpc_gss") == 0)
 146                                 CDEBUG(D_SEC,
 147                                        "module ptlrpc_gss loaded on demand\n");
 148                         else
 149                                 CERROR("Unable to load module ptlrpc_gss\n");
 150
 151                         cfs_atomic_set(&loaded, 1);
 152                 }
 153                 cfs_mutex_unlock(&load_mutex);
 154         }
 155
 156         return policy;
 157 }
 158
 159 __u32 sptlrpc_name2flavor_base(const char *name)
 160 {
 161         if (!strcmp(name, "null"))
 162                 return SPTLRPC_FLVR_NULL;
 163         if (!strcmp(name, "plain"))
 164                 return SPTLRPC_FLVR_PLAIN;
 165         if (!strcmp(name, "krb5n"))
 166                 return SPTLRPC_FLVR_KRB5N;
 167         if (!strcmp(name, "krb5a"))
 168                 return SPTLRPC_FLVR_KRB5A;
 169         if (!strcmp(name, "krb5i"))
 170                 return SPTLRPC_FLVR_KRB5I;
 171         if (!strcmp(name, "krb5p"))
 172                 return SPTLRPC_FLVR_KRB5P;
 173
 174         return SPTLRPC_FLVR_INVALID;
 175 }
 176 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 177
 178 const char *sptlrpc_flavor2name_base(__u32 flvr)
 179 {
 180         __u32   base = SPTLRPC_FLVR_BASE(flvr);
 181
 182         if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
 183                 return "null";
 184         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
 185                 return "plain";
 186         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
 187                 return "krb5n";
 188         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
 189                 return "krb5a";
 190         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
 191                 return "krb5i";
 192         else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
 193                 return "krb5p";
 194
 195         CERROR("invalid wire flavor 0x%x\n", flvr);
 196         return "invalid";
 197 }
 198 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 199
 200 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
 201                                char *buf, int bufsize)
 202 {
 203         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
 204                 snprintf(buf, bufsize, "hash:%s",
 205                          sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
 206         else
 207                 snprintf(buf, bufsize, "%s",
 208                          sptlrpc_flavor2name_base(sf->sf_rpc));
 209
 210         buf[bufsize - 1] = '\0';
 211         return buf;
 212 }
 213 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
 214
 215 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
 216 {
 217         snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
 218
 219         /*
 220          * currently we don't support customized bulk specification for
 221          * flavors other than plain
 222          */
 223         if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
 224                 char bspec[16];
 225
 226                 bspec[0] = '-';
 227                 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
 228                 strncat(buf, bspec, bufsize);
 229         }
 230
 231         buf[bufsize - 1] = '\0';
 232         return buf;
 233 }
 234 EXPORT_SYMBOL(sptlrpc_flavor2name);
 235
 236 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
 237 {
 238         buf[0] = '\0';
 239
 240         if (flags & PTLRPC_SEC_FL_REVERSE)
 241                 strncat(buf, "reverse,", bufsize);
 242         if (flags & PTLRPC_SEC_FL_ROOTONLY)
 243                 strncat(buf, "rootonly,", bufsize);
 244         if (flags & PTLRPC_SEC_FL_UDESC)
 245                 strncat(buf, "udesc,", bufsize);
 246         if (flags & PTLRPC_SEC_FL_BULK)
 247                 strncat(buf, "bulk,", bufsize);
 248         if (buf[0] == '\0')
 249                 strncat(buf, "-,", bufsize);
 250
 251         buf[bufsize - 1] = '\0';
 252         return buf;
 253 }
 254 EXPORT_SYMBOL(sptlrpc_secflags2str);
 255
 256 /**************************************************
 257  * client context APIs                            *
 258  **************************************************/
 259
 260 static
 261 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 262 {
 263         struct vfs_cred vcred;
 264         int create = 1, remove_dead = 1;
 265
 266         LASSERT(sec);
 267         LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
 268
 269         if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
 270                                      PTLRPC_SEC_FL_ROOTONLY)) {
 271                 vcred.vc_uid = 0;
 272                 vcred.vc_gid = 0;
 273                 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
 274                         create = 0;
 275                         remove_dead = 0;
 276                 }
 277         } else {
 278                 vcred.vc_uid = cfs_curproc_uid();
 279                 vcred.vc_gid = cfs_curproc_gid();
 280         }
 281
 282         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
 283                                                    create, remove_dead);
 284 }
 285
 286 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 287 {
 288         cfs_atomic_inc(&ctx->cc_refcount);
 289         return ctx;
 290 }
 291 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
 292
 293 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 294 {
 295         struct ptlrpc_sec *sec = ctx->cc_sec;
 296
 297         LASSERT(sec);
 298         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 299
 300         if (!cfs_atomic_dec_and_test(&ctx->cc_refcount))
 301                 return;
 302
 303         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 304 }
 305 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 306
 307 /**
 308  * Expire the client context immediately.
 309  *
 310  * \pre Caller must hold at least 1 reference on the \a ctx.
 311  */
 312 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 313 {
 314         LASSERT(ctx->cc_ops->die);
 315         ctx->cc_ops->die(ctx, 0);
 316 }
 317 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 318
 319 /**
 320  * To wake up the threads who are waiting for this client context. Called
 321  * after some status change happened on \a ctx.
 322  */
 323 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 324 {
 325         struct ptlrpc_request *req, *next;
 326
 327         cfs_spin_lock(&ctx->cc_lock);
 328         cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list,
 329                                      rq_ctx_chain) {
 330                 cfs_list_del_init(&req->rq_ctx_chain);
 331                 ptlrpc_client_wake_req(req);
 332         }
 333         cfs_spin_unlock(&ctx->cc_lock);
 334 }
 335 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 336
 337 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 338 {
 339         LASSERT(ctx->cc_ops);
 340
 341         if (ctx->cc_ops->display == NULL)
 342                 return 0;
 343
 344         return ctx->cc_ops->display(ctx, buf, bufsize);
 345 }
 346
 347 static int import_sec_check_expire(struct obd_import *imp)
 348 {
 349         int     adapt = 0;
 350
 351         cfs_spin_lock(&imp->imp_lock);
 352         if (imp->imp_sec_expire &&
 353             imp->imp_sec_expire < cfs_time_current_sec()) {
 354                 adapt = 1;
 355                 imp->imp_sec_expire = 0;
 356         }
 357         cfs_spin_unlock(&imp->imp_lock);
 358
 359         if (!adapt)
 360                 return 0;
 361
 362         CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
 363         return sptlrpc_import_sec_adapt(imp, NULL, 0);
 364 }
 365
 366 static int import_sec_validate_get(struct obd_import *imp,
 367                                    struct ptlrpc_sec **sec)
 368 {
 369         int     rc;
 370
 371         if (unlikely(imp->imp_sec_expire)) {
 372                 rc = import_sec_check_expire(imp);
 373                 if (rc)
 374                         return rc;
 375         }
 376
 377         *sec = sptlrpc_import_sec_ref(imp);
 378         if (*sec == NULL) {
 379                 CERROR("import %p (%s) with no sec\n",
 380                        imp, ptlrpc_import_state_name(imp->imp_state));
 381                 return -EACCES;
 382         }
 383
 384         if (unlikely((*sec)->ps_dying)) {
 385                 CERROR("attempt to use dying sec %p\n", sec);
 386                 sptlrpc_sec_put(*sec);
 387                 return -EACCES;
 388         }
 389
 390         return 0;
 391 }
 392
 393 /**
 394  * Given a \a req, find or allocate a appropriate context for it.
 395  * \pre req->rq_cli_ctx == NULL.
 396  *
 397  * \retval 0 succeed, and req->rq_cli_ctx is set.
 398  * \retval -ev error number, and req->rq_cli_ctx == NULL.
 399  */
 400 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 401 {
 402         struct obd_import *imp = req->rq_import;
 403         struct ptlrpc_sec *sec;
 404         int                rc;
 405         ENTRY;
 406
 407         LASSERT(!req->rq_cli_ctx);
 408         LASSERT(imp);
 409
 410         rc = import_sec_validate_get(imp, &sec);
 411         if (rc)
 412                 RETURN(rc);
 413
 414         req->rq_cli_ctx = get_my_ctx(sec);
 415
 416         sptlrpc_sec_put(sec);
 417
 418         if (!req->rq_cli_ctx) {
 419                 CERROR("req %p: fail to get context\n", req);
 420                 RETURN(-ENOMEM);
 421         }
 422
 423         RETURN(0);
 424 }
 425
 426 /**
 427  * Drop the context for \a req.
 428  * \pre req->rq_cli_ctx != NULL.
 429  * \post req->rq_cli_ctx == NULL.
 430  *
 431  * If \a sync == 0, this function should return quickly without sleep;
 432  * otherwise it might trigger and wait for the whole process of sending
 433  * an context-destroying rpc to server.
 434  */
 435 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 436 {
 437         ENTRY;
 438
 439         LASSERT(req);
 440         LASSERT(req->rq_cli_ctx);
 441
 442         /* request might be asked to release earlier while still
 443          * in the context waiting list.
 444          */
 445         if (!cfs_list_empty(&req->rq_ctx_chain)) {
 446                 cfs_spin_lock(&req->rq_cli_ctx->cc_lock);
 447                 cfs_list_del_init(&req->rq_ctx_chain);
 448                 cfs_spin_unlock(&req->rq_cli_ctx->cc_lock);
 449         }
 450
 451         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
 452         req->rq_cli_ctx = NULL;
 453         EXIT;
 454 }
 455
 456 static
 457 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 458                            struct ptlrpc_cli_ctx *oldctx,
 459                            struct ptlrpc_cli_ctx *newctx)
 460 {
 461         struct sptlrpc_flavor   old_flvr;
 462         char                   *reqmsg = NULL; /* to workaround old gcc */
 463         int                     reqmsg_size;
 464         int                     rc = 0;
 465
 466         LASSERT(req->rq_reqmsg);
 467         LASSERT(req->rq_reqlen);
 468         LASSERT(req->rq_replen);
 469
 470         CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
 471                "switch sec %p(%s) -> %p(%s)\n", req,
 472                oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
 473                newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
 474                oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
 475                newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 476
 477         /* save flavor */
 478         old_flvr = req->rq_flvr;
 479
 480         /* save request message */
 481         reqmsg_size = req->rq_reqlen;
 482         if (reqmsg_size != 0) {
 483                 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
 484                 if (reqmsg == NULL)
 485                         return -ENOMEM;
 486                 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
 487         }
 488
 489         /* release old req/rep buf */
 490         req->rq_cli_ctx = oldctx;
 491         sptlrpc_cli_free_reqbuf(req);
 492         sptlrpc_cli_free_repbuf(req);
 493         req->rq_cli_ctx = newctx;
 494
 495         /* recalculate the flavor */
 496         sptlrpc_req_set_flavor(req, 0);
 497
 498         /* alloc new request buffer
 499          * we don't need to alloc reply buffer here, leave it to the
 500          * rest procedure of ptlrpc */
 501         if (reqmsg_size != 0) {
 502                 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
 503                 if (!rc) {
 504                         LASSERT(req->rq_reqmsg);
 505                         memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
 506                 } else {
 507                         CWARN("failed to alloc reqbuf: %d\n", rc);
 508                         req->rq_flvr = old_flvr;
 509                 }
 510
 511                 OBD_FREE_LARGE(reqmsg, reqmsg_size);
 512         }
 513         return rc;
 514 }
 515
 516 /**
 517  * If current context of \a req is dead somehow, e.g. we just switched flavor
 518  * thus marked original contexts dead, we'll find a new context for it. if
 519  * no switch is needed, \a req will end up with the same context.
 520  *
 521  * \note a request must have a context, to keep other parts of code happy.
 522  * In any case of failure during the switching, we must restore the old one.
 523  */
 524 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 525 {
 526         struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
 527         struct ptlrpc_cli_ctx *newctx;
 528         int                    rc;
 529         ENTRY;
 530
 531         LASSERT(oldctx);
 532
 533         sptlrpc_cli_ctx_get(oldctx);
 534         sptlrpc_req_put_ctx(req, 0);
 535
 536         rc = sptlrpc_req_get_ctx(req);
 537         if (unlikely(rc)) {
 538                 LASSERT(!req->rq_cli_ctx);
 539
 540                 /* restore old ctx */
 541                 req->rq_cli_ctx = oldctx;
 542                 RETURN(rc);
 543         }
 544
 545         newctx = req->rq_cli_ctx;
 546         LASSERT(newctx);
 547
 548         if (unlikely(newctx == oldctx &&
 549                      cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
 550                 /*
 551                  * still get the old dead ctx, usually means system too busy
 552                  */
 553                 CDEBUG(D_SEC,
 554                        "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
 555                        newctx, newctx->cc_flags);
 556
 557                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
 558                                                    CFS_HZ);
 559         } else {
 560                 /*
 561                  * it's possible newctx == oldctx if we're switching
 562                  * subflavor with the same sec.
 563                  */
 564                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
 565                 if (rc) {
 566                         /* restore old ctx */
 567                         sptlrpc_req_put_ctx(req, 0);
 568                         req->rq_cli_ctx = oldctx;
 569                         RETURN(rc);
 570                 }
 571
 572                 LASSERT(req->rq_cli_ctx == newctx);
 573         }
 574
 575         sptlrpc_cli_ctx_put(oldctx, 1);
 576         RETURN(0);
 577 }
 578 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
 579
 580 static
 581 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 582 {
 583         if (cli_ctx_is_refreshed(ctx))
 584                 return 1;
 585         return 0;
 586 }
 587
 588 static
 589 int ctx_refresh_timeout(void *data)
 590 {
 591         struct ptlrpc_request *req = data;
 592         int rc;
 593
 594         /* conn_cnt is needed in expire_one_request */
 595         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 596
 597         rc = ptlrpc_expire_one_request(req, 1);
 598         /* if we started recovery, we should mark this ctx dead; otherwise
 599          * in case of lgssd died nobody would retire this ctx, following
 600          * connecting will still find the same ctx thus cause deadlock.
 601          * there's an assumption that expire time of the request should be
 602          * later than the context refresh expire time.
 603          */
 604         if (rc == 0)
 605                 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
 606         return rc;
 607 }
 608
 609 static
 610 void ctx_refresh_interrupt(void *data)
 611 {
 612         struct ptlrpc_request *req = data;
 613
 614         cfs_spin_lock(&req->rq_lock);
 615         req->rq_intr = 1;
 616         cfs_spin_unlock(&req->rq_lock);
 617 }
 618
 619 static
 620 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 621 {
 622         cfs_spin_lock(&ctx->cc_lock);
 623         if (!cfs_list_empty(&req->rq_ctx_chain))
 624                 cfs_list_del_init(&req->rq_ctx_chain);
 625         cfs_spin_unlock(&ctx->cc_lock);
 626 }
 627
 628 /**
 629  * To refresh the context of \req, if it's not up-to-date.
 630  * \param timeout
 631  * - < 0: don't wait
 632  * - = 0: wait until success or fatal error occur
 633  * - > 0: timeout value (in seconds)
 634  *
 635  * The status of the context could be subject to be changed by other threads
 636  * at any time. We allow this race, but once we return with 0, the caller will
 637  * suppose it's uptodated and keep using it until the owning rpc is done.
 638  *
 639  * \retval 0 only if the context is uptodated.
 640  * \retval -ev error number.
 641  */
 642 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 643 {
 644         struct ptlrpc_cli_ctx  *ctx = req->rq_cli_ctx;
 645         struct ptlrpc_sec      *sec;
 646         struct l_wait_info      lwi;
 647         int                     rc;
 648         ENTRY;
 649
 650         LASSERT(ctx);
 651
 652         if (req->rq_ctx_init || req->rq_ctx_fini)
 653                 RETURN(0);
 654
 655         /*
 656          * during the process a request's context might change type even
 657          * (e.g. from gss ctx to null ctx), so each loop we need to re-check
 658          * everything
 659          */
 660 again:
 661         rc = import_sec_validate_get(req->rq_import, &sec);
 662         if (rc)
 663                 RETURN(rc);
 664
 665         if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 666                 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
 667                       req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
 668                 req_off_ctx_list(req, ctx);
 669                 sptlrpc_req_replace_dead_ctx(req);
 670                 ctx = req->rq_cli_ctx;
 671         }
 672         sptlrpc_sec_put(sec);
 673
 674         if (cli_ctx_is_eternal(ctx))
 675                 RETURN(0);
 676
 677         if (unlikely(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
 678                 LASSERT(ctx->cc_ops->refresh);
 679                 ctx->cc_ops->refresh(ctx);
 680         }
 681         LASSERT(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 682
 683         LASSERT(ctx->cc_ops->validate);
 684         if (ctx->cc_ops->validate(ctx) == 0) {
 685                 req_off_ctx_list(req, ctx);
 686                 RETURN(0);
 687         }
 688
 689         if (unlikely(cfs_test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
 690                 cfs_spin_lock(&req->rq_lock);
 691                 req->rq_err = 1;
 692                 cfs_spin_unlock(&req->rq_lock);
 693                 req_off_ctx_list(req, ctx);
 694                 RETURN(-EPERM);
 695         }
 696
 697         /*
 698          * There's a subtle issue for resending RPCs, suppose following
 699          * situation:
 700          *  1. the request was sent to server.
 701          *  2. recovery was kicked start, after finished the request was
 702          *     marked as resent.
 703          *  3. resend the request.
 704          *  4. old reply from server received, we accept and verify the reply.
 705          *     this has to be success, otherwise the error will be aware
 706          *     by application.
 707          *  5. new reply from server received, dropped by LNet.
 708          *
 709          * Note the xid of old & new request is the same. We can't simply
 710          * change xid for the resent request because the server replies on
 711          * it for reply reconstruction.
 712          *
 713          * Commonly the original context should be uptodate because we
 714          * have a expiry nice time; server will keep its context because
 715          * we at least hold a ref of old context which prevent context
 716          * destroying RPC being sent. So server still can accept the request
 717          * and finish the RPC. But if that's not the case:
 718          *  1. If server side context has been trimmed, a NO_CONTEXT will
 719          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 720          *     context by force.
 721          *  2. Current context never be refreshed, then we are fine: we
 722          *     never really send request with old context before.
 723          */
 724         if (cfs_test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
 725             unlikely(req->rq_reqmsg) &&
 726             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 727                 req_off_ctx_list(req, ctx);
 728                 RETURN(0);
 729         }
 730
 731         if (unlikely(cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 732                 req_off_ctx_list(req, ctx);
 733                 /*
 734                  * don't switch ctx if import was deactivated
 735                  */
 736                 if (req->rq_import->imp_deactive) {
 737                         cfs_spin_lock(&req->rq_lock);
 738                         req->rq_err = 1;
 739                         cfs_spin_unlock(&req->rq_lock);
 740                         RETURN(-EINTR);
 741                 }
 742
 743                 rc = sptlrpc_req_replace_dead_ctx(req);
 744                 if (rc) {
 745                         LASSERT(ctx == req->rq_cli_ctx);
 746                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
 747                                 req, ctx, rc);
 748                         cfs_spin_lock(&req->rq_lock);
 749                         req->rq_err = 1;
 750                         cfs_spin_unlock(&req->rq_lock);
 751                         RETURN(rc);
 752                 }
 753
 754                 ctx = req->rq_cli_ctx;
 755                 goto again;
 756         }
 757
 758         /*
 759          * Now we're sure this context is during upcall, add myself into
 760          * waiting list
 761          */
 762         cfs_spin_lock(&ctx->cc_lock);
 763         if (cfs_list_empty(&req->rq_ctx_chain))
 764                 cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 765         cfs_spin_unlock(&ctx->cc_lock);
 766
 767         if (timeout < 0)
 768                 RETURN(-EWOULDBLOCK);
 769
 770         /* Clear any flags that may be present from previous sends */
 771         LASSERT(req->rq_receiving_reply == 0);
 772         cfs_spin_lock(&req->rq_lock);
 773         req->rq_err = 0;
 774         req->rq_timedout = 0;
 775         req->rq_resend = 0;
 776         req->rq_restart = 0;
 777         cfs_spin_unlock(&req->rq_lock);
 778
 779         lwi = LWI_TIMEOUT_INTR(timeout * CFS_HZ, ctx_refresh_timeout,
 780                                ctx_refresh_interrupt, req);
 781         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 782
 783         /*
 784          * following cases could lead us here:
 785          * - successfully refreshed;
 786          * - interrupted;
 787          * - timedout, and we don't want recover from the failure;
 788          * - timedout, and waked up upon recovery finished;
 789          * - someone else mark this ctx dead by force;
 790          * - someone invalidate the req and call ptlrpc_client_wake_req(),
 791          *   e.g. ptlrpc_abort_inflight();
 792          */
 793         if (!cli_ctx_is_refreshed(ctx)) {
 794                 /* timed out or interruptted */
 795                 req_off_ctx_list(req, ctx);
 796
 797                 LASSERT(rc != 0);
 798                 RETURN(rc);
 799         }
 800
 801         goto again;
 802 }
 803
 804 /**
 805  * Initialize flavor settings for \a req, according to \a opcode.
 806  *
 807  * \note this could be called in two situations:
 808  * - new request from ptlrpc_pre_req(), with proper @opcode
 809  * - old request which changed ctx in the middle, with @opcode == 0
 810  */
 811 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 812 {
 813         struct ptlrpc_sec *sec;
 814
 815         LASSERT(req->rq_import);
 816         LASSERT(req->rq_cli_ctx);
 817         LASSERT(req->rq_cli_ctx->cc_sec);
 818         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 819
 820         /* special security flags accoding to opcode */
 821         switch (opcode) {
 822         case OST_READ:
 823         case MDS_READPAGE:
 824         case MGS_CONFIG_READ:
 825         case OBD_IDX_READ:
 826                 req->rq_bulk_read = 1;
 827                 break;
 828         case OST_WRITE:
 829         case MDS_WRITEPAGE:
 830                 req->rq_bulk_write = 1;
 831                 break;
 832         case SEC_CTX_INIT:
 833                 req->rq_ctx_init = 1;
 834                 break;
 835         case SEC_CTX_FINI:
 836                 req->rq_ctx_fini = 1;
 837                 break;
 838         case 0:
 839                 /* init/fini rpc won't be resend, so can't be here */
 840                 LASSERT(req->rq_ctx_init == 0);
 841                 LASSERT(req->rq_ctx_fini == 0);
 842
 843                 /* cleanup flags, which should be recalculated */
 844                 req->rq_pack_udesc = 0;
 845                 req->rq_pack_bulk = 0;
 846                 break;
 847         }
 848
 849         sec = req->rq_cli_ctx->cc_sec;
 850
 851         cfs_spin_lock(&sec->ps_lock);
 852         req->rq_flvr = sec->ps_flvr;
 853         cfs_spin_unlock(&sec->ps_lock);
 854
 855         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
 856          * destruction rpc */
 857         if (unlikely(req->rq_ctx_init))
 858                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
 859         else if (unlikely(req->rq_ctx_fini))
 860                 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 861
 862         /* user descriptor flag, null security can't do it anyway */
 863         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
 864             (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
 865                 req->rq_pack_udesc = 1;
 866
 867         /* bulk security flag */
 868         if ((req->rq_bulk_read || req->rq_bulk_write) &&
 869             sptlrpc_flavor_has_bulk(&req->rq_flvr))
 870                 req->rq_pack_bulk = 1;
 871 }
 872
 873 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 874 {
 875         if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
 876                 return;
 877
 878         LASSERT(req->rq_clrbuf);
 879         if (req->rq_pool || !req->rq_reqbuf)
 880                 return;
 881
 882         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
 883         req->rq_reqbuf = NULL;
 884         req->rq_reqbuf_len = 0;
 885 }
 886
 887 /**
 888  * Given an import \a imp, check whether current user has a valid context
 889  * or not. We may create a new context and try to refresh it, and try
 890  * repeatedly try in case of non-fatal errors. Return 0 means success.
 891  */
 892 int sptlrpc_import_check_ctx(struct obd_import *imp)
 893 {
 894         struct ptlrpc_sec     *sec;
 895         struct ptlrpc_cli_ctx *ctx;
 896         struct ptlrpc_request *req = NULL;
 897         int rc;
 898         ENTRY;
 899
 900         cfs_might_sleep();
 901
 902         sec = sptlrpc_import_sec_ref(imp);
 903         ctx = get_my_ctx(sec);
 904         sptlrpc_sec_put(sec);
 905
 906         if (!ctx)
 907                 RETURN(-ENOMEM);
 908
 909         if (cli_ctx_is_eternal(ctx) ||
 910             ctx->cc_ops->validate(ctx) == 0) {
 911                 sptlrpc_cli_ctx_put(ctx, 1);
 912                 RETURN(0);
 913         }
 914
 915         if (cli_ctx_is_error(ctx)) {
 916                 sptlrpc_cli_ctx_put(ctx, 1);
 917                 RETURN(-EACCES);
 918         }
 919
 920         OBD_ALLOC_PTR(req);
 921         if (!req)
 922                 RETURN(-ENOMEM);
 923
 924         cfs_spin_lock_init(&req->rq_lock);
 925         cfs_atomic_set(&req->rq_refcount, 10000);
 926         CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
 927         cfs_waitq_init(&req->rq_reply_waitq);
 928         cfs_waitq_init(&req->rq_set_waitq);
 929         req->rq_import = imp;
 930         req->rq_flvr = sec->ps_flvr;
 931         req->rq_cli_ctx = ctx;
 932
 933         rc = sptlrpc_req_refresh_ctx(req, 0);
 934         LASSERT(cfs_list_empty(&req->rq_ctx_chain));
 935         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
 936         OBD_FREE_PTR(req);
 937
 938         RETURN(rc);
 939 }
 940
 941 /**
 942  * Used by ptlrpc client, to perform the pre-defined security transformation
 943  * upon the request message of \a req. After this function called,
 944  * req->rq_reqmsg is still accessible as clear text.
 945  */
 946 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 947 {
 948         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 949         int rc = 0;
 950         ENTRY;
 951
 952         LASSERT(ctx);
 953         LASSERT(ctx->cc_sec);
 954         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
 955
 956         /* we wrap bulk request here because now we can be sure
 957          * the context is uptodate.
 958          */
 959         if (req->rq_bulk) {
 960                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
 961                 if (rc)
 962                         RETURN(rc);
 963         }
 964
 965         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
 966         case SPTLRPC_SVC_NULL:
 967         case SPTLRPC_SVC_AUTH:
 968         case SPTLRPC_SVC_INTG:
 969                 LASSERT(ctx->cc_ops->sign);
 970                 rc = ctx->cc_ops->sign(ctx, req);
 971                 break;
 972         case SPTLRPC_SVC_PRIV:
 973                 LASSERT(ctx->cc_ops->seal);
 974                 rc = ctx->cc_ops->seal(ctx, req);
 975                 break;
 976         default:
 977                 LBUG();
 978         }
 979
 980         if (rc == 0) {
 981                 LASSERT(req->rq_reqdata_len);
 982                 LASSERT(req->rq_reqdata_len % 8 == 0);
 983                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
 984         }
 985
 986         RETURN(rc);
 987 }
 988
 989 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
 990 {
 991         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 992         int                    rc;
 993         ENTRY;
 994
 995         LASSERT(ctx);
 996         LASSERT(ctx->cc_sec);
 997         LASSERT(req->rq_repbuf);
 998         LASSERT(req->rq_repdata);
 999         LASSERT(req->rq_repmsg == NULL);
1000
1001         req->rq_rep_swab_mask = 0;
1002
1003         rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1004         switch (rc) {
1005         case 1:
1006                 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1007         case 0:
1008                 break;
1009         default:
1010                 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1011                 RETURN(-EPROTO);
1012         }
1013
1014         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1015                 CERROR("replied data length %d too small\n",
1016                        req->rq_repdata_len);
1017                 RETURN(-EPROTO);
1018         }
1019
1020         if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1021             SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1022                 CERROR("reply policy %u doesn't match request policy %u\n",
1023                        SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1024                        SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1025                 RETURN(-EPROTO);
1026         }
1027
1028         switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1029         case SPTLRPC_SVC_NULL:
1030         case SPTLRPC_SVC_AUTH:
1031         case SPTLRPC_SVC_INTG:
1032                 LASSERT(ctx->cc_ops->verify);
1033                 rc = ctx->cc_ops->verify(ctx, req);
1034                 break;
1035         case SPTLRPC_SVC_PRIV:
1036                 LASSERT(ctx->cc_ops->unseal);
1037                 rc = ctx->cc_ops->unseal(ctx, req);
1038                 break;
1039         default:
1040                 LBUG();
1041         }
1042         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1043
1044         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1045             !req->rq_ctx_init)
1046                 req->rq_rep_swab_mask = 0;
1047         RETURN(rc);
1048 }
1049
1050 /**
1051  * Used by ptlrpc client, to perform security transformation upon the reply
1052  * message of \a req. After return successfully, req->rq_repmsg points to
1053  * the reply message in clear text.
1054  *
1055  * \pre the reply buffer should have been un-posted from LNet, so nothing is
1056  * going to change.
1057  */
1058 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1059 {
1060         LASSERT(req->rq_repbuf);
1061         LASSERT(req->rq_repdata == NULL);
1062         LASSERT(req->rq_repmsg == NULL);
1063         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1064
1065         if (req->rq_reply_off == 0 &&
1066             (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1067                 CERROR("real reply with offset 0\n");
1068                 return -EPROTO;
1069         }
1070
1071         if (req->rq_reply_off % 8 != 0) {
1072                 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1073                 return -EPROTO;
1074         }
1075
1076         req->rq_repdata = (struct lustre_msg *)
1077                                 (req->rq_repbuf + req->rq_reply_off);
1078         req->rq_repdata_len = req->rq_nob_received;
1079
1080         return do_cli_unwrap_reply(req);
1081 }
1082
1083 /**
1084  * Used by ptlrpc client, to perform security transformation upon the early
1085  * reply message of \a req. We expect the rq_reply_off is 0, and
1086  * rq_nob_received is the early reply size.
1087  *
1088  * Because the receive buffer might be still posted, the reply data might be
1089  * changed at any time, no matter we're holding rq_lock or not. For this reason
1090  * we allocate a separate ptlrpc_request and reply buffer for early reply
1091  * processing.
1092  *
1093  * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1094  * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1095  * \a *req_ret to release it.
1096  * \retval -ev error number, and \a req_ret will not be set.
1097  */
1098 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1099                                    struct ptlrpc_request **req_ret)
1100 {
1101         struct ptlrpc_request  *early_req;
1102         char                   *early_buf;
1103         int                     early_bufsz, early_size;
1104         int                     rc;
1105         ENTRY;
1106
1107         OBD_ALLOC_PTR(early_req);
1108         if (early_req == NULL)
1109                 RETURN(-ENOMEM);
1110
1111         early_size = req->rq_nob_received;
1112         early_bufsz = size_roundup_power2(early_size);
1113         OBD_ALLOC_LARGE(early_buf, early_bufsz);
1114         if (early_buf == NULL)
1115                 GOTO(err_req, rc = -ENOMEM);
1116
1117         /* sanity checkings and copy data out, do it inside spinlock */
1118         cfs_spin_lock(&req->rq_lock);
1119
1120         if (req->rq_replied) {
1121                 cfs_spin_unlock(&req->rq_lock);
1122                 GOTO(err_buf, rc = -EALREADY);
1123         }
1124
1125         LASSERT(req->rq_repbuf);
1126         LASSERT(req->rq_repdata == NULL);
1127         LASSERT(req->rq_repmsg == NULL);
1128
1129         if (req->rq_reply_off != 0) {
1130                 CERROR("early reply with offset %u\n", req->rq_reply_off);
1131                 cfs_spin_unlock(&req->rq_lock);
1132                 GOTO(err_buf, rc = -EPROTO);
1133         }
1134
1135         if (req->rq_nob_received != early_size) {
1136                 /* even another early arrived the size should be the same */
1137                 CERROR("data size has changed from %u to %u\n",
1138                        early_size, req->rq_nob_received);
1139                 cfs_spin_unlock(&req->rq_lock);
1140                 GOTO(err_buf, rc = -EINVAL);
1141         }
1142
1143         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1144                 CERROR("early reply length %d too small\n",
1145                        req->rq_nob_received);
1146                 cfs_spin_unlock(&req->rq_lock);
1147                 GOTO(err_buf, rc = -EALREADY);
1148         }
1149
1150         memcpy(early_buf, req->rq_repbuf, early_size);
1151         cfs_spin_unlock(&req->rq_lock);
1152
1153         cfs_spin_lock_init(&early_req->rq_lock);
1154         early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1155         early_req->rq_flvr = req->rq_flvr;
1156         early_req->rq_repbuf = early_buf;
1157         early_req->rq_repbuf_len = early_bufsz;
1158         early_req->rq_repdata = (struct lustre_msg *) early_buf;
1159         early_req->rq_repdata_len = early_size;
1160         early_req->rq_early = 1;
1161         early_req->rq_reqmsg = req->rq_reqmsg;
1162
1163         rc = do_cli_unwrap_reply(early_req);
1164         if (rc) {
1165                 DEBUG_REQ(D_ADAPTTO, early_req,
1166                           "error %d unwrap early reply", rc);
1167                 GOTO(err_ctx, rc);
1168         }
1169
1170         LASSERT(early_req->rq_repmsg);
1171         *req_ret = early_req;
1172         RETURN(0);
1173
1174 err_ctx:
1175         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1176 err_buf:
1177         OBD_FREE_LARGE(early_buf, early_bufsz);
1178 err_req:
1179         OBD_FREE_PTR(early_req);
1180         RETURN(rc);
1181 }
1182
1183 /**
1184  * Used by ptlrpc client, to release a processed early reply \a early_req.
1185  *
1186  * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1187  */
1188 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1189 {
1190         LASSERT(early_req->rq_repbuf);
1191         LASSERT(early_req->rq_repdata);
1192         LASSERT(early_req->rq_repmsg);
1193
1194         sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1195         OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1196         OBD_FREE_PTR(early_req);
1197 }
1198
1199 /**************************************************
1200  * sec ID                                         *
1201  **************************************************/
1202
1203 /*
1204  * "fixed" sec (e.g. null) use sec_id < 0
1205  */
1206 static cfs_atomic_t sptlrpc_sec_id = CFS_ATOMIC_INIT(1);
1207
1208 int sptlrpc_get_next_secid(void)
1209 {
1210         return cfs_atomic_inc_return(&sptlrpc_sec_id);
1211 }
1212 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1213
1214 /**************************************************
1215  * client side high-level security APIs           *
1216  **************************************************/
1217
1218 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1219                                    int grace, int force)
1220 {
1221         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1222
1223         LASSERT(policy->sp_cops);
1224         LASSERT(policy->sp_cops->flush_ctx_cache);
1225
1226         return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1227 }
1228
1229 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1230 {
1231         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1232
1233         LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1234         LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1235         LASSERT(policy->sp_cops->destroy_sec);
1236
1237         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1238
1239         policy->sp_cops->destroy_sec(sec);
1240         sptlrpc_policy_put(policy);
1241 }
1242
1243 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1244 {
1245         sec_cop_destroy_sec(sec);
1246 }
1247 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1248
1249 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1250 {
1251         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1252
1253         if (sec->ps_policy->sp_cops->kill_sec) {
1254                 sec->ps_policy->sp_cops->kill_sec(sec);
1255
1256                 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1257         }
1258 }
1259
1260 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1261 {
1262         if (sec)
1263                 cfs_atomic_inc(&sec->ps_refcount);
1264
1265         return sec;
1266 }
1267 EXPORT_SYMBOL(sptlrpc_sec_get);
1268
1269 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1270 {
1271         if (sec) {
1272                 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1273
1274                 if (cfs_atomic_dec_and_test(&sec->ps_refcount)) {
1275                         sptlrpc_gc_del_sec(sec);
1276                         sec_cop_destroy_sec(sec);
1277                 }
1278         }
1279 }
1280 EXPORT_SYMBOL(sptlrpc_sec_put);
1281
1282 /*
1283  * policy module is responsible for taking refrence of import
1284  */
1285 static
1286 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1287                                        struct ptlrpc_svc_ctx *svc_ctx,
1288                                        struct sptlrpc_flavor *sf,
1289                                        enum lustre_sec_part sp)
1290 {
1291         struct ptlrpc_sec_policy *policy;
1292         struct ptlrpc_sec        *sec;
1293         char                      str[32];
1294         ENTRY;
1295
1296         if (svc_ctx) {
1297                 LASSERT(imp->imp_dlm_fake == 1);
1298
1299                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1300                        imp->imp_obd->obd_type->typ_name,
1301                        imp->imp_obd->obd_name,
1302                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1303
1304                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1305                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1306         } else {
1307                 LASSERT(imp->imp_dlm_fake == 0);
1308
1309                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1310                        imp->imp_obd->obd_type->typ_name,
1311                        imp->imp_obd->obd_name,
1312                        sptlrpc_flavor2name(sf, str, sizeof(str)));
1313
1314                 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1315                 if (!policy) {
1316                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1317                         RETURN(NULL);
1318                 }
1319         }
1320
1321         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1322         if (sec) {
1323                 cfs_atomic_inc(&sec->ps_refcount);
1324
1325                 sec->ps_part = sp;
1326
1327                 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1328                         sptlrpc_gc_add_sec(sec);
1329         } else {
1330                 sptlrpc_policy_put(policy);
1331         }
1332
1333         RETURN(sec);
1334 }
1335
1336 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1337 {
1338         struct ptlrpc_sec *sec;
1339
1340         cfs_spin_lock(&imp->imp_lock);
1341         sec = sptlrpc_sec_get(imp->imp_sec);
1342         cfs_spin_unlock(&imp->imp_lock);
1343
1344         return sec;
1345 }
1346 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1347
1348 static void sptlrpc_import_sec_install(struct obd_import *imp,
1349                                        struct ptlrpc_sec *sec)
1350 {
1351         struct ptlrpc_sec *old_sec;
1352
1353         LASSERT_ATOMIC_POS(&sec->ps_refcount);
1354
1355         cfs_spin_lock(&imp->imp_lock);
1356         old_sec = imp->imp_sec;
1357         imp->imp_sec = sec;
1358         cfs_spin_unlock(&imp->imp_lock);
1359
1360         if (old_sec) {
1361                 sptlrpc_sec_kill(old_sec);
1362
1363                 /* balance the ref taken by this import */
1364                 sptlrpc_sec_put(old_sec);
1365         }
1366 }
1367
1368 static inline
1369 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1370 {
1371         return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1372 }
1373
1374 static inline
1375 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1376 {
1377         *dst = *src;
1378 }
1379
1380 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1381                                              struct ptlrpc_sec *sec,
1382                                              struct sptlrpc_flavor *sf)
1383 {
1384         char    str1[32], str2[32];
1385
1386         if (sec->ps_flvr.sf_flags != sf->sf_flags)
1387                 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1388                        sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1389                                             str1, sizeof(str1)),
1390                        sptlrpc_secflags2str(sf->sf_flags,
1391                                             str2, sizeof(str2)));
1392
1393         cfs_spin_lock(&sec->ps_lock);
1394         flavor_copy(&sec->ps_flvr, sf);
1395         cfs_spin_unlock(&sec->ps_lock);
1396 }
1397
1398 /**
1399  * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1400  * configuration. Upon called, imp->imp_sec may or may not be NULL.
1401  *
1402  *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1403  *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1404  */
1405 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1406                              struct ptlrpc_svc_ctx *svc_ctx,
1407                              struct sptlrpc_flavor *flvr)
1408 {
1409         struct ptlrpc_connection   *conn;
1410         struct sptlrpc_flavor       sf;
1411         struct ptlrpc_sec          *sec, *newsec;
1412         enum lustre_sec_part        sp;
1413         char                        str[24];
1414         int                         rc = 0;
1415         ENTRY;
1416
1417         cfs_might_sleep();
1418
1419         if (imp == NULL)
1420                 RETURN(0);
1421
1422         conn = imp->imp_connection;
1423
1424         if (svc_ctx == NULL) {
1425                 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1426                 /*
1427                  * normal import, determine flavor from rule set, except
1428                  * for mgc the flavor is predetermined.
1429                  */
1430                 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1431                         sf = cliobd->cl_flvr_mgc;
1432                 else
1433                         sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1434                                                    cliobd->cl_sp_to,
1435                                                    &cliobd->cl_target_uuid,
1436                                                    conn->c_self, &sf);
1437
1438                 sp = imp->imp_obd->u.cli.cl_sp_me;
1439         } else {
1440                 /* reverse import, determine flavor from incoming reqeust */
1441                 sf = *flvr;
1442
1443                 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1444                         sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1445                                       PTLRPC_SEC_FL_ROOTONLY;
1446
1447                 sp = sptlrpc_target_sec_part(imp->imp_obd);
1448         }
1449
1450         sec = sptlrpc_import_sec_ref(imp);
1451         if (sec) {
1452                 char    str2[24];
1453
1454                 if (flavor_equal(&sf, &sec->ps_flvr))
1455                         GOTO(out, rc);
1456
1457                 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1458                        imp->imp_obd->obd_name,
1459                        obd_uuid2str(&conn->c_remote_uuid),
1460                        sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1461                        sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1462
1463                 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1464                     SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1465                     SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1466                     SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1467                         sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1468                         GOTO(out, rc);
1469                 }
1470         } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1471                    SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1472                 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1473                        imp->imp_obd->obd_name,
1474                        obd_uuid2str(&conn->c_remote_uuid),
1475                        LNET_NIDNET(conn->c_self),
1476                        sptlrpc_flavor2name(&sf, str, sizeof(str)));
1477         }
1478
1479         cfs_mutex_lock(&imp->imp_sec_mutex);
1480
1481         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1482         if (newsec) {
1483                 sptlrpc_import_sec_install(imp, newsec);
1484         } else {
1485                 CERROR("import %s->%s: failed to create new sec\n",
1486                        imp->imp_obd->obd_name,
1487                        obd_uuid2str(&conn->c_remote_uuid));
1488                 rc = -EPERM;
1489         }
1490
1491         cfs_mutex_unlock(&imp->imp_sec_mutex);
1492 out:
1493         sptlrpc_sec_put(sec);
1494         RETURN(rc);
1495 }
1496
1497 void sptlrpc_import_sec_put(struct obd_import *imp)
1498 {
1499         if (imp->imp_sec) {
1500                 sptlrpc_sec_kill(imp->imp_sec);
1501
1502                 sptlrpc_sec_put(imp->imp_sec);
1503                 imp->imp_sec = NULL;
1504         }
1505 }
1506
1507 static void import_flush_ctx_common(struct obd_import *imp,
1508                                     uid_t uid, int grace, int force)
1509 {
1510         struct ptlrpc_sec *sec;
1511
1512         if (imp == NULL)
1513                 return;
1514
1515         sec = sptlrpc_import_sec_ref(imp);
1516         if (sec == NULL)
1517                 return;
1518
1519         sec_cop_flush_ctx_cache(sec, uid, grace, force);
1520         sptlrpc_sec_put(sec);
1521 }
1522
1523 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1524 {
1525         /* it's important to use grace mode, see explain in
1526          * sptlrpc_req_refresh_ctx() */
1527         import_flush_ctx_common(imp, 0, 1, 1);
1528 }
1529
1530 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1531 {
1532         import_flush_ctx_common(imp, cfs_curproc_uid(), 1, 1);
1533 }
1534 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1535
1536 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1537 {
1538         import_flush_ctx_common(imp, -1, 1, 1);
1539 }
1540 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1541
1542 /**
1543  * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1544  * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1545  */
1546 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1547 {
1548         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1549         struct ptlrpc_sec_policy *policy;
1550         int rc;
1551
1552         LASSERT(ctx);
1553         LASSERT(ctx->cc_sec);
1554         LASSERT(ctx->cc_sec->ps_policy);
1555         LASSERT(req->rq_reqmsg == NULL);
1556         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1557
1558         policy = ctx->cc_sec->ps_policy;
1559         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1560         if (!rc) {
1561                 LASSERT(req->rq_reqmsg);
1562                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1563
1564                 /* zeroing preallocated buffer */
1565                 if (req->rq_pool)
1566                         memset(req->rq_reqmsg, 0, msgsize);
1567         }
1568
1569         return rc;
1570 }
1571
1572 /**
1573  * Used by ptlrpc client to free request buffer of \a req. After this
1574  * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1575  */
1576 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1577 {
1578         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1579         struct ptlrpc_sec_policy *policy;
1580
1581         LASSERT(ctx);
1582         LASSERT(ctx->cc_sec);
1583         LASSERT(ctx->cc_sec->ps_policy);
1584         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1585
1586         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1587                 return;
1588
1589         policy = ctx->cc_sec->ps_policy;
1590         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1591         req->rq_reqmsg = NULL;
1592 }
1593
1594 /*
1595  * NOTE caller must guarantee the buffer size is enough for the enlargement
1596  */
1597 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1598                                   int segment, int newsize)
1599 {
1600         void   *src, *dst;
1601         int     oldsize, oldmsg_size, movesize;
1602
1603         LASSERT(segment < msg->lm_bufcount);
1604         LASSERT(msg->lm_buflens[segment] <= newsize);
1605
1606         if (msg->lm_buflens[segment] == newsize)
1607                 return;
1608
1609         /* nothing to do if we are enlarging the last segment */
1610         if (segment == msg->lm_bufcount - 1) {
1611                 msg->lm_buflens[segment] = newsize;
1612                 return;
1613         }
1614
1615         oldsize = msg->lm_buflens[segment];
1616
1617         src = lustre_msg_buf(msg, segment + 1, 0);
1618         msg->lm_buflens[segment] = newsize;
1619         dst = lustre_msg_buf(msg, segment + 1, 0);
1620         msg->lm_buflens[segment] = oldsize;
1621
1622         /* move from segment + 1 to end segment */
1623         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1624         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1625         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1626         LASSERT(movesize >= 0);
1627
1628         if (movesize)
1629                 memmove(dst, src, movesize);
1630
1631         /* note we don't clear the ares where old data live, not secret */
1632
1633         /* finally set new segment size */
1634         msg->lm_buflens[segment] = newsize;
1635 }
1636 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1637
1638 /**
1639  * Used by ptlrpc client to enlarge the \a segment of request message pointed
1640  * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1641  * preserved after the enlargement. this must be called after original request
1642  * buffer being allocated.
1643  *
1644  * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1645  * so caller should refresh its local pointers if needed.
1646  */
1647 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1648                                int segment, int newsize)
1649 {
1650         struct ptlrpc_cli_ctx    *ctx = req->rq_cli_ctx;
1651         struct ptlrpc_sec_cops   *cops;
1652         struct lustre_msg        *msg = req->rq_reqmsg;
1653
1654         LASSERT(ctx);
1655         LASSERT(msg);
1656         LASSERT(msg->lm_bufcount > segment);
1657         LASSERT(msg->lm_buflens[segment] <= newsize);
1658
1659         if (msg->lm_buflens[segment] == newsize)
1660                 return 0;
1661
1662         cops = ctx->cc_sec->ps_policy->sp_cops;
1663         LASSERT(cops->enlarge_reqbuf);
1664         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1665 }
1666 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1667
1668 /**
1669  * Used by ptlrpc client to allocate reply buffer of \a req.
1670  *
1671  * \note After this, req->rq_repmsg is still not accessible.
1672  */
1673 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1674 {
1675         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1676         struct ptlrpc_sec_policy *policy;
1677         ENTRY;
1678
1679         LASSERT(ctx);
1680         LASSERT(ctx->cc_sec);
1681         LASSERT(ctx->cc_sec->ps_policy);
1682
1683         if (req->rq_repbuf)
1684                 RETURN(0);
1685
1686         policy = ctx->cc_sec->ps_policy;
1687         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1688 }
1689
1690 /**
1691  * Used by ptlrpc client to free reply buffer of \a req. After this
1692  * req->rq_repmsg is set to NULL and should not be accessed anymore.
1693  */
1694 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1695 {
1696         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1697         struct ptlrpc_sec_policy *policy;
1698         ENTRY;
1699
1700         LASSERT(ctx);
1701         LASSERT(ctx->cc_sec);
1702         LASSERT(ctx->cc_sec->ps_policy);
1703         LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1704
1705         if (req->rq_repbuf == NULL)
1706                 return;
1707         LASSERT(req->rq_repbuf_len);
1708
1709         policy = ctx->cc_sec->ps_policy;
1710         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1711         req->rq_repmsg = NULL;
1712         EXIT;
1713 }
1714
1715 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1716                                 struct ptlrpc_cli_ctx *ctx)
1717 {
1718         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1719
1720         if (!policy->sp_cops->install_rctx)
1721                 return 0;
1722         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1723 }
1724
1725 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1726                                 struct ptlrpc_svc_ctx *ctx)
1727 {
1728         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1729
1730         if (!policy->sp_sops->install_rctx)
1731                 return 0;
1732         return policy->sp_sops->install_rctx(imp, ctx);
1733 }
1734
1735 /****************************************
1736  * server side security                 *
1737  ****************************************/
1738
1739 static int flavor_allowed(struct sptlrpc_flavor *exp,
1740                           struct ptlrpc_request *req)
1741 {
1742         struct sptlrpc_flavor *flvr = &req->rq_flvr;
1743
1744         if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1745                 return 1;
1746
1747         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1748             SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1749             SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1750             SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1751                 return 1;
1752
1753         return 0;
1754 }
1755
1756 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
1757
1758 /**
1759  * Given an export \a exp, check whether the flavor of incoming \a req
1760  * is allowed by the export \a exp. Main logic is about taking care of
1761  * changing configurations. Return 0 means success.
1762  */
1763 int sptlrpc_target_export_check(struct obd_export *exp,
1764                                 struct ptlrpc_request *req)
1765 {
1766         struct sptlrpc_flavor   flavor;
1767
1768         if (exp == NULL)
1769                 return 0;
1770
1771         /* client side export has no imp_reverse, skip
1772          * FIXME maybe we should check flavor this as well??? */
1773         if (exp->exp_imp_reverse == NULL)
1774                 return 0;
1775
1776         /* don't care about ctx fini rpc */
1777         if (req->rq_ctx_fini)
1778                 return 0;
1779
1780         cfs_spin_lock(&exp->exp_lock);
1781
1782         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1783          * the first req with the new flavor, then treat it as current flavor,
1784          * adapt reverse sec according to it.
1785          * note the first rpc with new flavor might not be with root ctx, in
1786          * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1787         if (unlikely(exp->exp_flvr_changed) &&
1788             flavor_allowed(&exp->exp_flvr_old[1], req)) {
1789                 /* make the new flavor as "current", and old ones as
1790                  * about-to-expire */
1791                 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1792                        exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1793                 flavor = exp->exp_flvr_old[1];
1794                 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1795                 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1796                 exp->exp_flvr_old[0] = exp->exp_flvr;
1797                 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1798                                           EXP_FLVR_UPDATE_EXPIRE;
1799                 exp->exp_flvr = flavor;
1800
1801                 /* flavor change finished */
1802                 exp->exp_flvr_changed = 0;
1803                 LASSERT(exp->exp_flvr_adapt == 1);
1804
1805                 /* if it's gss, we only interested in root ctx init */
1806                 if (req->rq_auth_gss &&
1807                     !(req->rq_ctx_init &&
1808                       (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1809                        req->rq_auth_usr_ost))) {
1810                         cfs_spin_unlock(&exp->exp_lock);
1811                         CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1812                                req->rq_auth_gss, req->rq_ctx_init,
1813                                req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1814                                req->rq_auth_usr_ost);
1815                         return 0;
1816                 }
1817
1818                 exp->exp_flvr_adapt = 0;
1819                 cfs_spin_unlock(&exp->exp_lock);
1820
1821                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1822                                                 req->rq_svc_ctx, &flavor);
1823         }
1824
1825         /* if it equals to the current flavor, we accept it, but need to
1826          * dealing with reverse sec/ctx */
1827         if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1828                 /* most cases should return here, we only interested in
1829                  * gss root ctx init */
1830                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1831                     (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1832                      !req->rq_auth_usr_ost)) {
1833                         cfs_spin_unlock(&exp->exp_lock);
1834                         return 0;
1835                 }
1836
1837                 /* if flavor just changed, we should not proceed, just leave
1838                  * it and current flavor will be discovered and replaced
1839                  * shortly, and let _this_ rpc pass through */
1840                 if (exp->exp_flvr_changed) {
1841                         LASSERT(exp->exp_flvr_adapt);
1842                         cfs_spin_unlock(&exp->exp_lock);
1843                         return 0;
1844                 }
1845
1846                 if (exp->exp_flvr_adapt) {
1847                         exp->exp_flvr_adapt = 0;
1848                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1849                                exp, exp->exp_flvr.sf_rpc,
1850                                exp->exp_flvr_old[0].sf_rpc,
1851                                exp->exp_flvr_old[1].sf_rpc);
1852                         flavor = exp->exp_flvr;
1853                         cfs_spin_unlock(&exp->exp_lock);
1854
1855                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1856                                                         req->rq_svc_ctx,
1857                                                         &flavor);
1858                 } else {
1859                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1860                                "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1861                                exp->exp_flvr_old[0].sf_rpc,
1862                                exp->exp_flvr_old[1].sf_rpc);
1863                         cfs_spin_unlock(&exp->exp_lock);
1864
1865                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1866                                                            req->rq_svc_ctx);
1867                 }
1868         }
1869
1870         if (exp->exp_flvr_expire[0]) {
1871                 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1872                         if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1873                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1874                                        "middle one ("CFS_DURATION_T")\n", exp,
1875                                        exp->exp_flvr.sf_rpc,
1876                                        exp->exp_flvr_old[0].sf_rpc,
1877                                        exp->exp_flvr_old[1].sf_rpc,
1878                                        exp->exp_flvr_expire[0] -
1879                                                 cfs_time_current_sec());
1880                                 cfs_spin_unlock(&exp->exp_lock);
1881                                 return 0;
1882                         }
1883                 } else {
1884                         CDEBUG(D_SEC, "mark middle expired\n");
1885                         exp->exp_flvr_expire[0] = 0;
1886                 }
1887                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1888                        exp->exp_flvr.sf_rpc,
1889                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1890                        req->rq_flvr.sf_rpc);
1891         }
1892
1893         /* now it doesn't match the current flavor, the only chance we can
1894          * accept it is match the old flavors which is not expired. */
1895         if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1896                 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1897                         if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1898                                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1899                                        "oldest one ("CFS_DURATION_T")\n", exp,
1900                                        exp->exp_flvr.sf_rpc,
1901                                        exp->exp_flvr_old[0].sf_rpc,
1902                                        exp->exp_flvr_old[1].sf_rpc,
1903                                        exp->exp_flvr_expire[1] -
1904                                                 cfs_time_current_sec());
1905                                 cfs_spin_unlock(&exp->exp_lock);
1906                                 return 0;
1907                         }
1908                 } else {
1909                         CDEBUG(D_SEC, "mark oldest expired\n");
1910                         exp->exp_flvr_expire[1] = 0;
1911                 }
1912                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1913                        exp, exp->exp_flvr.sf_rpc,
1914                        exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1915                        req->rq_flvr.sf_rpc);
1916         } else {
1917                 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1918                        exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1919                        exp->exp_flvr_old[1].sf_rpc);
1920         }
1921
1922         cfs_spin_unlock(&exp->exp_lock);
1923
1924         CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1925               "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1926               exp, exp->exp_obd->obd_name,
1927               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1928               req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1929               req->rq_flvr.sf_rpc,
1930               exp->exp_flvr.sf_rpc,
1931               exp->exp_flvr_old[0].sf_rpc,
1932               exp->exp_flvr_expire[0] ?
1933               (unsigned long) (exp->exp_flvr_expire[0] -
1934                                cfs_time_current_sec()) : 0,
1935               exp->exp_flvr_old[1].sf_rpc,
1936               exp->exp_flvr_expire[1] ?
1937               (unsigned long) (exp->exp_flvr_expire[1] -
1938                                cfs_time_current_sec()) : 0);
1939         return -EACCES;
1940 }
1941 EXPORT_SYMBOL(sptlrpc_target_export_check);
1942
1943 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1944                                       struct sptlrpc_rule_set *rset)
1945 {
1946         struct obd_export       *exp;
1947         struct sptlrpc_flavor    new_flvr;
1948
1949         LASSERT(obd);
1950
1951         cfs_spin_lock(&obd->obd_dev_lock);
1952
1953         cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1954                 if (exp->exp_connection == NULL)
1955                         continue;
1956
1957                 /* note if this export had just been updated flavor
1958                  * (exp_flvr_changed == 1), this will override the
1959                  * previous one. */
1960                 cfs_spin_lock(&exp->exp_lock);
1961                 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1962                                              exp->exp_connection->c_peer.nid,
1963                                              &new_flvr);
1964                 if (exp->exp_flvr_changed ||
1965                     !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1966                         exp->exp_flvr_old[1] = new_flvr;
1967                         exp->exp_flvr_expire[1] = 0;
1968                         exp->exp_flvr_changed = 1;
1969                         exp->exp_flvr_adapt = 1;
1970
1971                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1972                                exp, sptlrpc_part2name(exp->exp_sp_peer),
1973                                exp->exp_flvr.sf_rpc,
1974                                exp->exp_flvr_old[1].sf_rpc);
1975                 }
1976                 cfs_spin_unlock(&exp->exp_lock);
1977         }
1978
1979         cfs_spin_unlock(&obd->obd_dev_lock);
1980 }
1981 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1982
1983 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1984 {
1985         /* peer's claim is unreliable unless gss is being used */
1986         if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1987                 return svc_rc;
1988
1989         switch (req->rq_sp_from) {
1990         case LUSTRE_SP_CLI:
1991                 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1992                         DEBUG_REQ(D_ERROR, req, "faked source CLI");
1993                         svc_rc = SECSVC_DROP;
1994                 }
1995                 break;
1996         case LUSTRE_SP_MDT:
1997                 if (!req->rq_auth_usr_mdt) {
1998                         DEBUG_REQ(D_ERROR, req, "faked source MDT");
1999                         svc_rc = SECSVC_DROP;
2000                 }
2001                 break;
2002         case LUSTRE_SP_OST:
2003                 if (!req->rq_auth_usr_ost) {
2004                         DEBUG_REQ(D_ERROR, req, "faked source OST");
2005                         svc_rc = SECSVC_DROP;
2006                 }
2007                 break;
2008         case LUSTRE_SP_MGS:
2009         case LUSTRE_SP_MGC:
2010                 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2011                     !req->rq_auth_usr_ost) {
2012                         DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2013                         svc_rc = SECSVC_DROP;
2014                 }
2015                 break;
2016         case LUSTRE_SP_ANY:
2017         default:
2018                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2019                 svc_rc = SECSVC_DROP;
2020         }
2021
2022         return svc_rc;
2023 }
2024
2025 /**
2026  * Used by ptlrpc server, to perform transformation upon request message of
2027  * incoming \a req. This must be the first thing to do with a incoming
2028  * request in ptlrpc layer.
2029  *
2030  * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2031  * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2032  * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2033  * reply message has been prepared.
2034  * \retval SECSVC_DROP failed, this request should be dropped.
2035  */
2036 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2037 {
2038         struct ptlrpc_sec_policy *policy;
2039         struct lustre_msg        *msg = req->rq_reqbuf;
2040         int                       rc;
2041         ENTRY;
2042
2043         LASSERT(msg);
2044         LASSERT(req->rq_reqmsg == NULL);
2045         LASSERT(req->rq_repmsg == NULL);
2046         LASSERT(req->rq_svc_ctx == NULL);
2047
2048         req->rq_req_swab_mask = 0;
2049
2050         rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2051         switch (rc) {
2052         case 1:
2053                 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2054         case 0:
2055                 break;
2056         default:
2057                 CERROR("error unpacking request from %s x"LPU64"\n",
2058                        libcfs_id2str(req->rq_peer), req->rq_xid);
2059                 RETURN(SECSVC_DROP);
2060         }
2061
2062         req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2063         req->rq_sp_from = LUSTRE_SP_ANY;
2064         req->rq_auth_uid = INVALID_UID;
2065         req->rq_auth_mapped_uid = INVALID_UID;
2066
2067         policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2068         if (!policy) {
2069                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2070                 RETURN(SECSVC_DROP);
2071         }
2072
2073         LASSERT(policy->sp_sops->accept);
2074         rc = policy->sp_sops->accept(req);
2075         sptlrpc_policy_put(policy);
2076         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2077         LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2078
2079         /*
2080          * if it's not null flavor (which means embedded packing msg),
2081          * reset the swab mask for the comming inner msg unpacking.
2082          */
2083         if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2084                 req->rq_req_swab_mask = 0;
2085
2086         /* sanity check for the request source */
2087         rc = sptlrpc_svc_check_from(req, rc);
2088         RETURN(rc);
2089 }
2090
2091 /**
2092  * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2093  * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2094  * a buffer of \a msglen size.
2095  */
2096 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2097 {
2098         struct ptlrpc_sec_policy *policy;
2099         struct ptlrpc_reply_state *rs;
2100         int rc;
2101         ENTRY;
2102
2103         LASSERT(req->rq_svc_ctx);
2104         LASSERT(req->rq_svc_ctx->sc_policy);
2105
2106         policy = req->rq_svc_ctx->sc_policy;
2107         LASSERT(policy->sp_sops->alloc_rs);
2108
2109         rc = policy->sp_sops->alloc_rs(req, msglen);
2110         if (unlikely(rc == -ENOMEM)) {
2111                 /* failed alloc, try emergency pool */
2112                 rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_svcpt);
2113                 if (rs == NULL)
2114                         RETURN(-ENOMEM);
2115
2116                 req->rq_reply_state = rs;
2117                 rc = policy->sp_sops->alloc_rs(req, msglen);
2118                 if (rc) {
2119                         lustre_put_emerg_rs(rs);
2120                         req->rq_reply_state = NULL;
2121                 }
2122         }
2123
2124         LASSERT(rc != 0 ||
2125                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2126
2127         RETURN(rc);
2128 }
2129
2130 /**
2131  * Used by ptlrpc server, to perform transformation upon reply message.
2132  *
2133  * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2134  * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2135  */
2136 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2137 {
2138         struct ptlrpc_sec_policy *policy;
2139         int rc;
2140         ENTRY;
2141
2142         LASSERT(req->rq_svc_ctx);
2143         LASSERT(req->rq_svc_ctx->sc_policy);
2144
2145         policy = req->rq_svc_ctx->sc_policy;
2146         LASSERT(policy->sp_sops->authorize);
2147
2148         rc = policy->sp_sops->authorize(req);
2149         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2150
2151         RETURN(rc);
2152 }
2153
2154 /**
2155  * Used by ptlrpc server, to free reply_state.
2156  */
2157 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2158 {
2159         struct ptlrpc_sec_policy *policy;
2160         unsigned int prealloc;
2161         ENTRY;
2162
2163         LASSERT(rs->rs_svc_ctx);
2164         LASSERT(rs->rs_svc_ctx->sc_policy);
2165
2166         policy = rs->rs_svc_ctx->sc_policy;
2167         LASSERT(policy->sp_sops->free_rs);
2168
2169         prealloc = rs->rs_prealloc;
2170         policy->sp_sops->free_rs(rs);
2171
2172         if (prealloc)
2173                 lustre_put_emerg_rs(rs);
2174         EXIT;
2175 }
2176
2177 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2178 {
2179         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2180
2181         if (ctx != NULL)
2182                 cfs_atomic_inc(&ctx->sc_refcount);
2183 }
2184
2185 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2186 {
2187         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2188
2189         if (ctx == NULL)
2190                 return;
2191
2192         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2193         if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) {
2194                 if (ctx->sc_policy->sp_sops->free_ctx)
2195                         ctx->sc_policy->sp_sops->free_ctx(ctx);
2196         }
2197         req->rq_svc_ctx = NULL;
2198 }
2199
2200 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2201 {
2202         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2203
2204         if (ctx == NULL)
2205                 return;
2206
2207         LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2208         if (ctx->sc_policy->sp_sops->invalidate_ctx)
2209                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2210 }
2211 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2212
2213 /****************************************
2214  * bulk security                        *
2215  ****************************************/
2216
2217 /**
2218  * Perform transformation upon bulk data pointed by \a desc. This is called
2219  * before transforming the request message.
2220  */
2221 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2222                           struct ptlrpc_bulk_desc *desc)
2223 {
2224         struct ptlrpc_cli_ctx *ctx;
2225
2226         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2227
2228         if (!req->rq_pack_bulk)
2229                 return 0;
2230
2231         ctx = req->rq_cli_ctx;
2232         if (ctx->cc_ops->wrap_bulk)
2233                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2234         return 0;
2235 }
2236 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2237
2238 /**
2239  * This is called after unwrap the reply message.
2240  * return nob of actual plain text size received, or error code.
2241  */
2242 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2243                                  struct ptlrpc_bulk_desc *desc,
2244                                  int nob)
2245 {
2246         struct ptlrpc_cli_ctx  *ctx;
2247         int                     rc;
2248
2249         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2250
2251         if (!req->rq_pack_bulk)
2252                 return desc->bd_nob_transferred;
2253
2254         ctx = req->rq_cli_ctx;
2255         if (ctx->cc_ops->unwrap_bulk) {
2256                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2257                 if (rc < 0)
2258                         return rc;
2259         }
2260         return desc->bd_nob_transferred;
2261 }
2262 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2263
2264 /**
2265  * This is called after unwrap the reply message.
2266  * return 0 for success or error code.
2267  */
2268 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2269                                   struct ptlrpc_bulk_desc *desc)
2270 {
2271         struct ptlrpc_cli_ctx  *ctx;
2272         int                     rc;
2273
2274         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2275
2276         if (!req->rq_pack_bulk)
2277                 return 0;
2278
2279         ctx = req->rq_cli_ctx;
2280         if (ctx->cc_ops->unwrap_bulk) {
2281                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2282                 if (rc < 0)
2283                         return rc;
2284         }
2285
2286         /*
2287          * if everything is going right, nob should equals to nob_transferred.
2288          * in case of privacy mode, nob_transferred needs to be adjusted.
2289          */
2290         if (desc->bd_nob != desc->bd_nob_transferred) {
2291                 CERROR("nob %d doesn't match transferred nob %d",
2292                        desc->bd_nob, desc->bd_nob_transferred);
2293                 return -EPROTO;
2294         }
2295
2296         return 0;
2297 }
2298 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2299
2300 #ifdef HAVE_SERVER_SUPPORT
2301 /**
2302  * Performe transformation upon outgoing bulk read.
2303  */
2304 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2305                           struct ptlrpc_bulk_desc *desc)
2306 {
2307         struct ptlrpc_svc_ctx *ctx;
2308
2309         LASSERT(req->rq_bulk_read);
2310
2311         if (!req->rq_pack_bulk)
2312                 return 0;
2313
2314         ctx = req->rq_svc_ctx;
2315         if (ctx->sc_policy->sp_sops->wrap_bulk)
2316                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2317
2318         return 0;
2319 }
2320 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2321
2322 /**
2323  * Performe transformation upon incoming bulk write.
2324  */
2325 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2326                             struct ptlrpc_bulk_desc *desc)
2327 {
2328         struct ptlrpc_svc_ctx *ctx;
2329         int                    rc;
2330
2331         LASSERT(req->rq_bulk_write);
2332
2333         /*
2334          * if it's in privacy mode, transferred should >= expected; otherwise
2335          * transferred should == expected.
2336          */
2337         if (desc->bd_nob_transferred < desc->bd_nob ||
2338             (desc->bd_nob_transferred > desc->bd_nob &&
2339              SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2340              SPTLRPC_BULK_SVC_PRIV)) {
2341                 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2342                           desc->bd_nob_transferred, desc->bd_nob);
2343                 return -ETIMEDOUT;
2344         }
2345
2346         if (!req->rq_pack_bulk)
2347                 return 0;
2348
2349         ctx = req->rq_svc_ctx;
2350         if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2351                 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2352                 if (rc)
2353                         CERROR("error unwrap bulk: %d\n", rc);
2354         }
2355
2356         /* return 0 to allow reply be sent */
2357         return 0;
2358 }
2359 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2360
2361 /**
2362  * Prepare buffers for incoming bulk write.
2363  */
2364 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2365                           struct ptlrpc_bulk_desc *desc)
2366 {
2367         struct ptlrpc_svc_ctx *ctx;
2368
2369         LASSERT(req->rq_bulk_write);
2370
2371         if (!req->rq_pack_bulk)
2372                 return 0;
2373
2374         ctx = req->rq_svc_ctx;
2375         if (ctx->sc_policy->sp_sops->prep_bulk)
2376                 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2377
2378         return 0;
2379 }
2380 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2381
2382 #endif /* HAVE_SERVER_SUPPORT */
2383
2384 /****************************************
2385  * user descriptor helpers              *
2386  ****************************************/
2387
2388 int sptlrpc_current_user_desc_size(void)
2389 {
2390         int ngroups;
2391
2392 #ifdef __KERNEL__
2393         ngroups = current_ngroups;
2394
2395         if (ngroups > LUSTRE_MAX_GROUPS)
2396                 ngroups = LUSTRE_MAX_GROUPS;
2397 #else
2398         ngroups = 0;
2399 #endif
2400         return sptlrpc_user_desc_size(ngroups);
2401 }
2402 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2403
2404 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2405 {
2406         struct ptlrpc_user_desc *pud;
2407
2408         pud = lustre_msg_buf(msg, offset, 0);
2409
2410         pud->pud_uid = cfs_curproc_uid();
2411         pud->pud_gid = cfs_curproc_gid();
2412         pud->pud_fsuid = cfs_curproc_fsuid();
2413         pud->pud_fsgid = cfs_curproc_fsgid();
2414         pud->pud_cap = cfs_curproc_cap_pack();
2415         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2416
2417 #ifdef __KERNEL__
2418         task_lock(current);
2419         if (pud->pud_ngroups > current_ngroups)
2420                 pud->pud_ngroups = current_ngroups;
2421         memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2422                pud->pud_ngroups * sizeof(__u32));
2423         task_unlock(current);
2424 #endif
2425
2426         return 0;
2427 }
2428 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2429
2430 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2431 {
2432         struct ptlrpc_user_desc *pud;
2433         int                      i;
2434
2435         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2436         if (!pud)
2437                 return -EINVAL;
2438
2439         if (swabbed) {
2440                 __swab32s(&pud->pud_uid);
2441                 __swab32s(&pud->pud_gid);
2442                 __swab32s(&pud->pud_fsuid);
2443                 __swab32s(&pud->pud_fsgid);
2444                 __swab32s(&pud->pud_cap);
2445                 __swab32s(&pud->pud_ngroups);
2446         }
2447
2448         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2449                 CERROR("%u groups is too large\n", pud->pud_ngroups);
2450                 return -EINVAL;
2451         }
2452
2453         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2454             msg->lm_buflens[offset]) {
2455                 CERROR("%u groups are claimed but bufsize only %u\n",
2456                        pud->pud_ngroups, msg->lm_buflens[offset]);
2457                 return -EINVAL;
2458         }
2459
2460         if (swabbed) {
2461                 for (i = 0; i < pud->pud_ngroups; i++)
2462                         __swab32s(&pud->pud_groups[i]);
2463         }
2464
2465         return 0;
2466 }
2467 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2468
2469 /****************************************
2470  * misc helpers                         *
2471  ****************************************/
2472
2473 const char * sec2target_str(struct ptlrpc_sec *sec)
2474 {
2475         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2476                 return "*";
2477         if (sec_is_reverse(sec))
2478                 return "c";
2479         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2480 }
2481 EXPORT_SYMBOL(sec2target_str);
2482
2483 /*
2484  * return true if the bulk data is protected
2485  */
2486 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2487 {
2488         switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2489         case SPTLRPC_BULK_SVC_INTG:
2490         case SPTLRPC_BULK_SVC_PRIV:
2491                 return 1;
2492         default:
2493                 return 0;
2494         }
2495 }
2496 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2497
2498 /****************************************
2499  * crypto API helper/alloc blkciper     *
2500  ****************************************/
2501
2502 /****************************************
2503  * initialize/finalize                  *
2504  ****************************************/
2505
2506 int sptlrpc_init(void)
2507 {
2508         int rc;
2509
2510         cfs_rwlock_init(&policy_lock);
2511
2512         rc = sptlrpc_gc_init();
2513         if (rc)
2514                 goto out;
2515
2516         rc = sptlrpc_conf_init();
2517         if (rc)
2518                 goto out_gc;
2519
2520         rc = sptlrpc_enc_pool_init();
2521         if (rc)
2522                 goto out_conf;
2523
2524         rc = sptlrpc_null_init();
2525         if (rc)
2526                 goto out_pool;
2527
2528         rc = sptlrpc_plain_init();
2529         if (rc)
2530                 goto out_null;
2531
2532         rc = sptlrpc_lproc_init();
2533         if (rc)
2534                 goto out_plain;
2535
2536         return 0;
2537
2538 out_plain:
2539         sptlrpc_plain_fini();
2540 out_null:
2541         sptlrpc_null_fini();
2542 out_pool:
2543         sptlrpc_enc_pool_fini();
2544 out_conf:
2545         sptlrpc_conf_fini();
2546 out_gc:
2547         sptlrpc_gc_fini();
2548 out:
2549         return rc;
2550 }
2551
2552 void sptlrpc_fini(void)
2553 {
2554         sptlrpc_lproc_fini();
2555         sptlrpc_plain_fini();
2556         sptlrpc_null_fini();
2557         sptlrpc_enc_pool_fini();
2558         sptlrpc_conf_fini();
2559         sptlrpc_gc_fini();
2560 }