lustre/ptlrpc/sec.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2004-2006 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 #define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_SEC
  26
  27 #include <libcfs/libcfs.h>
  28 #ifndef __KERNEL__
  29 #include <liblustre.h>
  30 #include <libcfs/list.h>
  31 #else
  32 #include <linux/crypto.h>
  33 #endif
  34
  35 #include <obd.h>
  36 #include <obd_class.h>
  37 #include <obd_support.h>
  38 #include <lustre_net.h>
  39 #include <lustre_import.h>
  40 #include <lustre_dlm.h>
  41 #include <lustre_sec.h>
  42
  43 #include "ptlrpc_internal.h"
  44
  45 static void sptlrpc_sec_destroy(struct ptlrpc_sec *sec);
  46 static int sptlrpc_sec_destroy_ctx(struct ptlrpc_sec *sec,
  47                                    struct ptlrpc_cli_ctx *ctx);
  48 static void sptlrpc_ctx_refresh(struct ptlrpc_cli_ctx *ctx);
  49
  50 /***********************************************
  51  * policy registers                            *
  52  ***********************************************/
  53
  54 static rwlock_t policy_lock = RW_LOCK_UNLOCKED;
  55 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
  56         NULL,
  57 };
  58
  59 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
  60 {
  61         __u32 number = policy->sp_policy;
  62
  63         LASSERT(policy->sp_name);
  64         LASSERT(policy->sp_cops);
  65         LASSERT(policy->sp_sops);
  66
  67         if (number >= SPTLRPC_POLICY_MAX)
  68                 return -EINVAL;
  69
  70         write_lock(&policy_lock);
  71         if (unlikely(policies[number])) {
  72                 write_unlock(&policy_lock);
  73                 return -EALREADY;
  74         }
  75         policies[number] = policy;
  76         write_unlock(&policy_lock);
  77
  78         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
  79         return 0;
  80 }
  81 EXPORT_SYMBOL(sptlrpc_register_policy);
  82
  83 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
  84 {
  85         __u32 number = policy->sp_policy;
  86
  87         LASSERT(number < SPTLRPC_POLICY_MAX);
  88
  89         write_lock(&policy_lock);
  90         if (unlikely(policies[number] == NULL)) {
  91                 write_unlock(&policy_lock);
  92                 CERROR("%s: already unregistered\n", policy->sp_name);
  93                 return -EINVAL;
  94         }
  95
  96         LASSERT(policies[number] == policy);
  97         policies[number] = NULL;
  98         write_unlock(&policy_lock);
  99
 100         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
 101         return 0;
 102 }
 103 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 104
 105 static
 106 struct ptlrpc_sec_policy * sptlrpc_flavor2policy(ptlrpc_sec_flavor_t flavor)
 107 {
 108         static DECLARE_MUTEX(load_mutex);
 109         static atomic_t         loaded = ATOMIC_INIT(0);
 110         struct                  ptlrpc_sec_policy *policy;
 111         __u32                   number = SEC_FLAVOR_POLICY(flavor), flag = 0;
 112
 113         if (number >= SPTLRPC_POLICY_MAX)
 114                 return NULL;
 115
 116 again:
 117         read_lock(&policy_lock);
 118         policy = policies[number];
 119         if (policy && !try_module_get(policy->sp_owner))
 120                 policy = NULL;
 121         if (policy == NULL)
 122                 flag = atomic_read(&loaded);
 123         read_unlock(&policy_lock);
 124
 125 #ifdef CONFIG_KMOD
 126         /* if failure, try to load gss module, once */
 127         if (unlikely(policy == NULL) &&
 128             number == SPTLRPC_POLICY_GSS && flag == 0) {
 129                 mutex_down(&load_mutex);
 130                 if (atomic_read(&loaded) == 0) {
 131                         if (request_module("ptlrpc_gss") != 0)
 132                                 CERROR("Unable to load module ptlrpc_gss\n");
 133                         else
 134                                 CWARN("module ptlrpc_gss loaded\n");
 135
 136                         atomic_set(&loaded, 1);
 137                 }
 138                 mutex_up(&load_mutex);
 139
 140                 goto again;
 141         }
 142 #endif
 143
 144         return policy;
 145 }
 146
 147 ptlrpc_sec_flavor_t sptlrpc_name2flavor(const char *name)
 148 {
 149         if (!strcmp(name, "null"))
 150                 return SPTLRPC_FLVR_NULL;
 151         if (!strcmp(name, "plain"))
 152                 return SPTLRPC_FLVR_PLAIN;
 153         if (!strcmp(name, "krb5"))
 154                 return SPTLRPC_FLVR_KRB5;
 155         if (!strcmp(name, "krb5i"))
 156                 return SPTLRPC_FLVR_KRB5I;
 157         if (!strcmp(name, "krb5p"))
 158                 return SPTLRPC_FLVR_KRB5P;
 159
 160         return SPTLRPC_FLVR_INVALID;
 161 }
 162 EXPORT_SYMBOL(sptlrpc_name2flavor);
 163
 164 char *sptlrpc_flavor2name(ptlrpc_sec_flavor_t flavor)
 165 {
 166         switch (flavor) {
 167         case SPTLRPC_FLVR_NULL:
 168                 return "null";
 169         case SPTLRPC_FLVR_PLAIN:
 170                 return "plain";
 171         case SPTLRPC_FLVR_KRB5:
 172                 return "krb5";
 173         case SPTLRPC_FLVR_KRB5I:
 174                 return "krb5i";
 175         case SPTLRPC_FLVR_KRB5P:
 176                 return "krb5p";
 177         default:
 178                 CERROR("invalid flavor 0x%x(p%u,s%u,v%u)\n", flavor,
 179                        SEC_FLAVOR_POLICY(flavor), SEC_FLAVOR_SUBPOLICY(flavor),
 180                        SEC_FLAVOR_SVC(flavor));
 181         }
 182         return "UNKNOWN";
 183 }
 184 EXPORT_SYMBOL(sptlrpc_flavor2name);
 185
 186 /***********************************************
 187  * context helpers                             *
 188  * internal APIs                               *
 189  * cache management                            *
 190  ***********************************************/
 191
 192 static inline
 193 unsigned long ctx_status(struct ptlrpc_cli_ctx *ctx)
 194 {
 195         smp_mb();
 196         return (ctx->cc_flags & PTLRPC_CTX_STATUS_MASK);
 197 }
 198
 199 static inline
 200 int ctx_is_uptodate(struct ptlrpc_cli_ctx *ctx)
 201 {
 202         return (ctx_status(ctx) == PTLRPC_CTX_UPTODATE);
 203 }
 204
 205 static inline
 206 int ctx_is_refreshed(struct ptlrpc_cli_ctx *ctx)
 207 {
 208         return (ctx_status(ctx) != 0);
 209 }
 210
 211 static inline
 212 int ctx_is_dead(struct ptlrpc_cli_ctx *ctx)
 213 {
 214         smp_mb();
 215         return ((ctx->cc_flags & (PTLRPC_CTX_DEAD | PTLRPC_CTX_ERROR)) != 0);
 216 }
 217
 218 static inline
 219 int ctx_is_eternal(struct ptlrpc_cli_ctx *ctx)
 220 {
 221         smp_mb();
 222         return ((ctx->cc_flags & PTLRPC_CTX_ETERNAL) != 0);
 223 }
 224
 225 static
 226 int ctx_expire(struct ptlrpc_cli_ctx *ctx)
 227 {
 228         LASSERT(atomic_read(&ctx->cc_refcount));
 229
 230         if (!test_and_set_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags)) {
 231                 cfs_time_t now = cfs_time_current_sec();
 232
 233                 smp_mb();
 234                 clear_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags);
 235
 236                 if (ctx->cc_expire && cfs_time_aftereq(now, ctx->cc_expire))
 237                         CWARN("ctx %p(%u->%s): get expired (%lds exceeds)\n",
 238                               ctx, ctx->cc_vcred.vc_uid,
 239                               sec2target_str(ctx->cc_sec),
 240                               cfs_time_sub(now, ctx->cc_expire));
 241                 else
 242                         CWARN("ctx %p(%u->%s): force to die (%lds remains)\n",
 243                               ctx, ctx->cc_vcred.vc_uid,
 244                               sec2target_str(ctx->cc_sec),
 245                               ctx->cc_expire == 0 ? 0 :
 246                               cfs_time_sub(ctx->cc_expire, now));
 247
 248                 return 1;
 249         }
 250         return 0;
 251 }
 252
 253 static
 254 void ctx_enhash(struct ptlrpc_cli_ctx *ctx, struct hlist_head *hash)
 255 {
 256         set_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags);
 257         atomic_inc(&ctx->cc_refcount);
 258         hlist_add_head(&ctx->cc_hash, hash);
 259 }
 260
 261 static
 262 void ctx_unhash(struct ptlrpc_cli_ctx *ctx, struct hlist_head *freelist)
 263 {
 264         LASSERT_SPIN_LOCKED(&ctx->cc_sec->ps_lock);
 265         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 266         LASSERT(test_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags));
 267         LASSERT(!hlist_unhashed(&ctx->cc_hash));
 268
 269         clear_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags);
 270
 271         if (atomic_dec_and_test(&ctx->cc_refcount)) {
 272                 __hlist_del(&ctx->cc_hash);
 273                 hlist_add_head(&ctx->cc_hash, freelist);
 274         } else
 275                 hlist_del_init(&ctx->cc_hash);
 276 }
 277
 278 /*
 279  * return 1 if the context is dead.
 280  */
 281 static
 282 int ctx_check_death(struct ptlrpc_cli_ctx *ctx, struct hlist_head *freelist)
 283 {
 284         if (unlikely(ctx_is_dead(ctx)))
 285                 goto unhash;
 286
 287         /* expire is 0 means never expire. a newly created gss context
 288          * which during upcall also has 0 expiration
 289          */
 290         smp_mb();
 291         if (ctx->cc_expire == 0)
 292                 return 0;
 293
 294         /* check real expiration */
 295         smp_mb();
 296         if (cfs_time_after(ctx->cc_expire, cfs_time_current_sec()))
 297                 return 0;
 298
 299         ctx_expire(ctx);
 300
 301 unhash:
 302         if (freelist)
 303                 ctx_unhash(ctx, freelist);
 304
 305         return 1;
 306 }
 307
 308 static inline
 309 int ctx_check_death_locked(struct ptlrpc_cli_ctx *ctx,
 310                            struct hlist_head *freelist)
 311 {
 312         LASSERT(ctx->cc_sec);
 313         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 314         LASSERT_SPIN_LOCKED(&ctx->cc_sec->ps_lock);
 315         LASSERT(test_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags));
 316
 317         return ctx_check_death(ctx, freelist);
 318 }
 319
 320 static
 321 int ctx_check_uptodate(struct ptlrpc_cli_ctx *ctx)
 322 {
 323         LASSERT(ctx->cc_sec);
 324         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 325
 326         if (!ctx_check_death(ctx, NULL) && ctx_is_uptodate(ctx))
 327                 return 1;
 328         return 0;
 329 }
 330
 331 static inline
 332 int ctx_match(struct ptlrpc_cli_ctx *ctx, struct vfs_cred *vcred)
 333 {
 334         /* a little bit optimization for null policy */
 335         if (!ctx->cc_ops->match)
 336                 return 1;
 337
 338         return ctx->cc_ops->match(ctx, vcred);
 339 }
 340
 341 static
 342 void ctx_list_destroy(struct hlist_head *head)
 343 {
 344         struct ptlrpc_cli_ctx *ctx;
 345
 346         while (!hlist_empty(head)) {
 347                 ctx = hlist_entry(head->first, struct ptlrpc_cli_ctx, cc_hash);
 348
 349                 LASSERT(atomic_read(&ctx->cc_refcount) == 0);
 350                 LASSERT(test_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags) == 0);
 351
 352                 hlist_del_init(&ctx->cc_hash);
 353                 sptlrpc_sec_destroy_ctx(ctx->cc_sec, ctx);
 354         }
 355 }
 356
 357 static
 358 void ctx_cache_gc(struct ptlrpc_sec *sec, struct hlist_head *freelist)
 359 {
 360         struct ptlrpc_cli_ctx *ctx;
 361         struct hlist_node *pos, *next;
 362         int i;
 363         ENTRY;
 364
 365         CDEBUG(D_SEC, "do gc on sec %s@%p\n", sec->ps_policy->sp_name, sec);
 366
 367         for (i = 0; i < sec->ps_ccache_size; i++) {
 368                 hlist_for_each_entry_safe(ctx, pos, next,
 369                                           &sec->ps_ccache[i], cc_hash)
 370                         ctx_check_death_locked(ctx, freelist);
 371         }
 372
 373         sec->ps_gc_next = cfs_time_current_sec() + sec->ps_gc_interval;
 374         EXIT;
 375 }
 376
 377 /*
 378  * @uid: which user. "-1" means flush all.
 379  * @grace: mark context DEAD, allow graceful destroy like notify
 380  *         server side, etc.
 381  * @force: also flush busy entries.
 382  *
 383  * return the number of busy context encountered.
 384  *
 385  * In any cases, never touch "eternal" contexts.
 386  */
 387 static
 388 int ctx_cache_flush(struct ptlrpc_sec *sec, uint64_t pag, uid_t uid,
 389                     int grace, int force)
 390 {
 391         struct ptlrpc_cli_ctx *ctx;
 392         struct hlist_node *pos, *next;
 393         HLIST_HEAD(freelist);
 394         int i, busy = 0;
 395         ENTRY;
 396
 397         might_sleep_if(grace);
 398
 399         spin_lock(&sec->ps_lock);
 400         for (i = 0; i < sec->ps_ccache_size; i++) {
 401                 hlist_for_each_entry_safe(ctx, pos, next,
 402                                           &sec->ps_ccache[i], cc_hash) {
 403                         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 404
 405                         if (ctx_is_eternal(ctx))
 406                                 continue;
 407
 408                         if (sec->ps_flags & PTLRPC_SEC_FL_PAG) {
 409                                 if (pag != -1 && pag != ctx->cc_vcred.vc_pag)
 410                                         continue;
 411                         } else {
 412                                 if (uid != -1 && uid != ctx->cc_vcred.vc_uid)
 413                                         continue;
 414                         }
 415
 416                         if (atomic_read(&ctx->cc_refcount) > 1) {
 417                                 busy++;
 418                                 if (!force)
 419                                         continue;
 420
 421                                 CWARN("flush busy(%d) ctx %p(%u->%s) by force, "
 422                                       "grace %d\n",
 423                                       atomic_read(&ctx->cc_refcount),
 424                                       ctx, ctx->cc_vcred.vc_uid,
 425                                       sec2target_str(ctx->cc_sec), grace);
 426                         }
 427                         ctx_unhash(ctx, &freelist);
 428
 429                         set_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags);
 430                         if (!grace)
 431                                 clear_bit(PTLRPC_CTX_UPTODATE_BIT,
 432                                           &ctx->cc_flags);
 433                 }
 434         }
 435         spin_unlock(&sec->ps_lock);
 436
 437         ctx_list_destroy(&freelist);
 438         RETURN(busy);
 439 }
 440
 441 static inline
 442 unsigned int ctx_hash_index(struct ptlrpc_sec *sec, __u64 key)
 443 {
 444         return (unsigned int) (key & (sec->ps_ccache_size - 1));
 445 }
 446
 447 /*
 448  * return matched context. If it's a newly created one, we also give the
 449  * first push to refresh. return NULL if error happens.
 450  */
 451 static
 452 struct ptlrpc_cli_ctx * ctx_cache_lookup(struct ptlrpc_sec *sec,
 453                                          struct vfs_cred *vcred,
 454                                          int create, int remove_dead)
 455 {
 456         struct ptlrpc_cli_ctx *ctx = NULL, *new = NULL;
 457         struct hlist_head *hash_head;
 458         struct hlist_node *pos, *next;
 459         HLIST_HEAD(freelist);
 460         unsigned int hash, gc = 0, found = 0;
 461         ENTRY;
 462
 463         might_sleep();
 464
 465         hash = ctx_hash_index(sec, vcred->vc_pag);
 466         LASSERT(hash < sec->ps_ccache_size);
 467         hash_head = &sec->ps_ccache[hash];
 468
 469 retry:
 470         spin_lock(&sec->ps_lock);
 471
 472         /* gc_next == 0 means never do gc */
 473         if (remove_dead && sec->ps_gc_next &&
 474             cfs_time_after(cfs_time_current_sec(), sec->ps_gc_next)) {
 475                 ctx_cache_gc(sec, &freelist);
 476                 gc = 1;
 477         }
 478
 479         hlist_for_each_entry_safe(ctx, pos, next, hash_head, cc_hash) {
 480                 if (gc == 0 &&
 481                     ctx_check_death_locked(ctx, remove_dead ? &freelist : NULL))
 482                         continue;
 483
 484                 if (ctx_match(ctx, vcred)) {
 485                         found = 1;
 486                         break;
 487                 }
 488         }
 489
 490         if (found) {
 491                 if (new && new != ctx) {
 492                         /* lost the race, just free it */
 493                         hlist_add_head(&new->cc_hash, &freelist);
 494                         new = NULL;
 495                 }
 496
 497                 /* hot node, move to head */
 498                 if (hash_head->first != &ctx->cc_hash) {
 499                         __hlist_del(&ctx->cc_hash);
 500                         hlist_add_head(&ctx->cc_hash, hash_head);
 501                 }
 502         } else {
 503                 /* don't allocate for reverse sec */
 504                 if (sec->ps_flags & PTLRPC_SEC_FL_REVERSE) {
 505                         spin_unlock(&sec->ps_lock);
 506                         RETURN(NULL);
 507                 }
 508
 509                 if (new) {
 510                         ctx_enhash(new, hash_head);
 511                         ctx = new;
 512                 } else if (create) {
 513                         spin_unlock(&sec->ps_lock);
 514                         new = sec->ps_policy->sp_cops->create_ctx(sec, vcred);
 515                         if (new) {
 516                                 atomic_inc(&sec->ps_busy);
 517                                 goto retry;
 518                         }
 519                 } else
 520                         ctx = NULL;
 521         }
 522
 523         /* hold a ref */
 524         if (ctx)
 525                 atomic_inc(&ctx->cc_refcount);
 526
 527         spin_unlock(&sec->ps_lock);
 528
 529         /* the allocator of the context must give the first push to refresh */
 530         if (new) {
 531                 LASSERT(new == ctx);
 532                 sptlrpc_ctx_refresh(new);
 533         }
 534
 535         ctx_list_destroy(&freelist);
 536         RETURN(ctx);
 537 }
 538
 539 static inline
 540 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 541 {
 542         struct vfs_cred vcred;
 543         int create = 1, remove_dead = 1;
 544
 545         if (sec->ps_flags & (PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY)) {
 546                 vcred.vc_pag = 0;
 547                 vcred.vc_uid = 0;
 548                 vcred.vc_gid = 0;
 549                 if (sec->ps_flags & PTLRPC_SEC_FL_REVERSE) {
 550                         create = 0;
 551                         remove_dead = 0;
 552                 }
 553         } else {
 554                 vcred.vc_pag = sec->ps_flags & PTLRPC_SEC_FL_PAG ?
 555                                CURRENT_PAG : current->uid;
 556                 vcred.vc_uid = current->uid;
 557                 vcred.vc_gid = current->gid;
 558                 /* don't distinguash root from others in pag mode */
 559                 if (vcred.vc_uid == 0)
 560                         vcred.vc_pag = 0;
 561         }
 562
 563         if (sec->ps_policy->sp_cops->lookup_ctx)
 564                 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred);
 565         else
 566                 return ctx_cache_lookup(sec, &vcred, create, remove_dead);
 567 }
 568
 569 /**************************************************
 570  * client context APIs                            *
 571  **************************************************/
 572
 573 static
 574 void sptlrpc_ctx_refresh(struct ptlrpc_cli_ctx *ctx)
 575 {
 576         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 577
 578         if (!ctx_is_refreshed(ctx) && ctx->cc_ops->refresh)
 579                 ctx->cc_ops->refresh(ctx);
 580 }
 581
 582 struct ptlrpc_cli_ctx *sptlrpc_ctx_get(struct ptlrpc_cli_ctx *ctx)
 583 {
 584         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 585         atomic_inc(&ctx->cc_refcount);
 586         return ctx;
 587 }
 588 EXPORT_SYMBOL(sptlrpc_ctx_get);
 589
 590 void sptlrpc_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
 591 {
 592         struct ptlrpc_sec *sec = ctx->cc_sec;
 593
 594         LASSERT(sec);
 595         LASSERT(atomic_read(&ctx->cc_refcount));
 596
 597         if (!atomic_dec_and_test(&ctx->cc_refcount))
 598                 return;
 599
 600         LASSERT(test_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags) == 0);
 601         LASSERT(hlist_unhashed(&ctx->cc_hash));
 602
 603         /* if required async, we must clear the UPTODATE bit to prevent extra
 604          * rpcs during destroy procedure.
 605          */
 606         if (!sync)
 607                 clear_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags);
 608
 609         /* destroy this context */
 610         if (!sptlrpc_sec_destroy_ctx(sec, ctx))
 611                 return;
 612
 613         CWARN("%s@%p: put last ctx, also destroy the sec\n",
 614               sec->ps_policy->sp_name, sec);
 615
 616         sptlrpc_sec_destroy(sec);
 617 }
 618 EXPORT_SYMBOL(sptlrpc_ctx_put);
 619
 620 /*
 621  * mark a ctx as DEAD, and pull it out from hash table.
 622  *
 623  * NOTE: the caller must hold at least 1 ref on the ctx.
 624  */
 625 void sptlrpc_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 626 {
 627         LASSERT(ctx->cc_sec);
 628         LASSERT(atomic_read(&ctx->cc_refcount) > 0);
 629
 630         ctx_expire(ctx);
 631
 632         spin_lock(&ctx->cc_sec->ps_lock);
 633
 634         if (test_and_clear_bit(PTLRPC_CTX_HASHED_BIT, &ctx->cc_flags)) {
 635                 LASSERT(!hlist_unhashed(&ctx->cc_hash));
 636                 LASSERT(atomic_read(&ctx->cc_refcount) > 1);
 637
 638                 hlist_del_init(&ctx->cc_hash);
 639                 if (atomic_dec_and_test(&ctx->cc_refcount))
 640                         LBUG();
 641         }
 642
 643         spin_unlock(&ctx->cc_sec->ps_lock);
 644 }
 645 EXPORT_SYMBOL(sptlrpc_ctx_expire);
 646
 647 void sptlrpc_ctx_replace(struct ptlrpc_sec *sec, struct ptlrpc_cli_ctx *new)
 648 {
 649         struct ptlrpc_cli_ctx *ctx;
 650         struct hlist_node *pos, *next;
 651         HLIST_HEAD(freelist);
 652         unsigned int hash;
 653         ENTRY;
 654
 655         hash = ctx_hash_index(sec, new->cc_vcred.vc_pag);
 656         LASSERT(hash < sec->ps_ccache_size);
 657
 658         spin_lock(&sec->ps_lock);
 659
 660         hlist_for_each_entry_safe(ctx, pos, next,
 661                                   &sec->ps_ccache[hash], cc_hash) {
 662                 if (!ctx_match(ctx, &new->cc_vcred))
 663                         continue;
 664
 665                 ctx_expire(ctx);
 666                 ctx_unhash(ctx, &freelist);
 667                 break;
 668         }
 669
 670         ctx_enhash(new, &sec->ps_ccache[hash]);
 671         atomic_inc(&sec->ps_busy);
 672
 673         spin_unlock(&sec->ps_lock);
 674
 675         ctx_list_destroy(&freelist);
 676         EXIT;
 677 }
 678 EXPORT_SYMBOL(sptlrpc_ctx_replace);
 679
 680 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
 681 {
 682         struct obd_import *imp = req->rq_import;
 683         ENTRY;
 684
 685         LASSERT(!req->rq_cli_ctx);
 686         LASSERT(imp);
 687
 688         req->rq_cli_ctx = get_my_ctx(imp->imp_sec);
 689
 690         if (!req->rq_cli_ctx) {
 691                 CERROR("req %p: fail to get context from cache\n", req);
 692                 RETURN(-ENOMEM);
 693         }
 694
 695         RETURN(0);
 696 }
 697
 698 void sptlrpc_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 699 {
 700         struct ptlrpc_request *req, *next;
 701
 702         spin_lock(&ctx->cc_lock);
 703         list_for_each_entry_safe(req, next, &ctx->cc_req_list, rq_ctx_chain) {
 704                 list_del_init(&req->rq_ctx_chain);
 705                 ptlrpc_wake_client_req(req);
 706         }
 707         spin_unlock(&ctx->cc_lock);
 708 }
 709 EXPORT_SYMBOL(sptlrpc_ctx_wakeup);
 710
 711 int sptlrpc_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
 712 {
 713         LASSERT(ctx->cc_ops);
 714
 715         if (ctx->cc_ops->display == NULL)
 716                 return 0;
 717
 718         return ctx->cc_ops->display(ctx, buf, bufsize);
 719 }
 720
 721 void sptlrpc_req_put_ctx(struct ptlrpc_request *req)
 722 {
 723         ENTRY;
 724
 725         LASSERT(req);
 726         LASSERT(req->rq_cli_ctx);
 727
 728         /* request might be asked to release earlier while still
 729          * in the context waiting list.
 730          */
 731         if (!list_empty(&req->rq_ctx_chain)) {
 732                 spin_lock(&req->rq_cli_ctx->cc_lock);
 733                 list_del_init(&req->rq_ctx_chain);
 734                 spin_unlock(&req->rq_cli_ctx->cc_lock);
 735         }
 736
 737         /* this could be called with spinlock hold, use async mode */
 738         sptlrpc_ctx_put(req->rq_cli_ctx, 0);
 739         req->rq_cli_ctx = NULL;
 740         EXIT;
 741 }
 742
 743 /*
 744  * request must have a context. if failed to get new context,
 745  * just restore the old one
 746  */
 747 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 748 {
 749         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
 750         int rc;
 751         ENTRY;
 752
 753         LASSERT(ctx);
 754         LASSERT(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags));
 755
 756         /* make sure not on context waiting list */
 757         spin_lock(&ctx->cc_lock);
 758         list_del_init(&req->rq_ctx_chain);
 759         spin_unlock(&ctx->cc_lock);
 760
 761         sptlrpc_ctx_get(ctx);
 762         sptlrpc_req_put_ctx(req);
 763         rc = sptlrpc_req_get_ctx(req);
 764         if (!rc) {
 765                 LASSERT(req->rq_cli_ctx);
 766                 LASSERT(req->rq_cli_ctx != ctx);
 767                 sptlrpc_ctx_put(ctx, 1);
 768         } else {
 769                 LASSERT(!req->rq_cli_ctx);
 770                 req->rq_cli_ctx = ctx;
 771         }
 772         RETURN(rc);
 773 }
 774
 775 static
 776 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
 777 {
 778         smp_mb();
 779         if (ctx_is_refreshed(ctx))
 780                 return 1;
 781         return 0;
 782 }
 783
 784 static
 785 int ctx_refresh_timeout(void *data)
 786 {
 787         struct ptlrpc_request *req = data;
 788         int rc;
 789
 790         /* conn_cnt is needed in expire_one_request */
 791         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 792
 793         rc = ptlrpc_expire_one_request(req);
 794         /* if we started recovery, we should mark this ctx dead; otherwise
 795          * in case of lgssd died nobody would retire this ctx, following
 796          * connecting will still find the same ctx thus cause deadlock.
 797          * there's an assumption that expire time of the request should be
 798          * later than the context refresh expire time.
 799          */
 800         if (rc == 0)
 801                 ctx_expire(req->rq_cli_ctx);
 802         return rc;
 803 }
 804
 805 static
 806 void ctx_refresh_interrupt(void *data)
 807 {
 808         /* do nothing */
 809 }
 810
 811 /*
 812  * the status of context could be subject to be changed by other threads at any
 813  * time. we allow this race. but once we return with 0, the caller will
 814  * suppose it's uptodated and keep using it until the affected rpc is done.
 815  *
 816  * @timeout:
 817  *    < 0  - don't wait
 818  *    = 0  - wait until success or fatal error occur
 819  *    > 0  - timeout value
 820  *
 821  * return 0 only if the context is uptodated.
 822  */
 823 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 824 {
 825         struct ptlrpc_cli_ctx  *ctx = req->rq_cli_ctx;
 826         struct l_wait_info      lwi;
 827         int                     rc;
 828         ENTRY;
 829
 830         LASSERT(ctx);
 831
 832         /* special ctxs */
 833         if (ctx_is_eternal(ctx) || req->rq_ctx_init || req->rq_ctx_fini)
 834                 RETURN(0);
 835
 836         /* reverse ctxs, don't refresh */
 837         if (ctx->cc_sec->ps_flags & PTLRPC_SEC_FL_REVERSE)
 838                 RETURN(0);
 839
 840         spin_lock(&ctx->cc_lock);
 841 again:
 842         if (ctx_check_uptodate(ctx)) {
 843                 if (!list_empty(&req->rq_ctx_chain))
 844                         list_del_init(&req->rq_ctx_chain);
 845                 spin_unlock(&ctx->cc_lock);
 846                 RETURN(0);
 847         }
 848
 849         if (test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags)) {
 850                 req->rq_err = 1;
 851                 if (!list_empty(&req->rq_ctx_chain))
 852                         list_del_init(&req->rq_ctx_chain);
 853                 spin_unlock(&ctx->cc_lock);
 854                 RETURN(-EPERM);
 855         }
 856
 857         /* This is subtle. For resent message we have to keep original
 858          * context to survive following situation:
 859          *  1. the request sent to server
 860          *  2. recovery was kick start
 861          *  3. recovery finished, the request marked as resent
 862          *  4. resend the request
 863          *  5. old reply from server received (because xid is the same)
 864          *  6. verify reply (has to be success)
 865          *  7. new reply from server received, lnet drop it
 866          *
 867          * Note we can't simply change xid for resent request because
 868          * server reply on it for reply reconstruction.
 869          *
 870          * Commonly the original context should be uptodate because we
 871          * have a expiry nice time; And server will keep their half part
 872          * context because we at least hold a ref of old context which
 873          * prevent the context detroy RPC be sent. So server still can
 874          * accept the request and finish RPC. Two cases:
 875          *  1. If server side context has been trimed, a NO_CONTEXT will
 876          *     be returned, gss_cli_ctx_verify/unseal will switch to new
 877          *     context by force.
 878          *  2. Current context never be refreshed, then we are fine: we
 879          *     never really send request with old context before.
 880          */
 881         if (test_bit(PTLRPC_CTX_UPTODATE, &ctx->cc_flags) &&
 882             req->rq_reqmsg &&
 883             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
 884                 if (!list_empty(&req->rq_ctx_chain))
 885                         list_del_init(&req->rq_ctx_chain);
 886                 spin_unlock(&ctx->cc_lock);
 887                 RETURN(0);
 888         }
 889
 890         if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
 891                 spin_unlock(&ctx->cc_lock);
 892
 893                 /* don't have to, but we don't want to release it too soon */
 894                 sptlrpc_ctx_get(ctx);
 895
 896                 rc = sptlrpc_req_replace_dead_ctx(req);
 897                 if (rc) {
 898                         LASSERT(ctx == req->rq_cli_ctx);
 899                         CERROR("req %p: failed to replace dead ctx %p\n",
 900                                 req, ctx);
 901                         req->rq_err = 1;
 902                         LASSERT(list_empty(&req->rq_ctx_chain));
 903                         sptlrpc_ctx_put(ctx, 1);
 904                         RETURN(-ENOMEM);
 905                 }
 906
 907                 LASSERT(ctx != req->rq_cli_ctx);
 908                 CWARN("req %p: replace dead ctx %p(%u->%s) => %p\n",
 909                       req, ctx, ctx->cc_vcred.vc_uid,
 910                       sec2target_str(ctx->cc_sec), req->rq_cli_ctx);
 911
 912                 sptlrpc_ctx_put(ctx, 1);
 913                 ctx = req->rq_cli_ctx;
 914                 LASSERT(list_empty(&req->rq_ctx_chain));
 915
 916                 spin_lock(&ctx->cc_lock);
 917                 goto again;
 918         }
 919
 920         /* Now we're sure this context is during upcall, add myself into
 921          * waiting list
 922          */
 923         if (list_empty(&req->rq_ctx_chain))
 924                 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
 925
 926         spin_unlock(&ctx->cc_lock);
 927
 928         if (timeout < 0) {
 929                 RETURN(-EWOULDBLOCK);
 930         }
 931
 932         /* Clear any flags that may be present from previous sends */
 933         LASSERT(req->rq_receiving_reply == 0);
 934         spin_lock(&req->rq_lock);
 935         req->rq_err = 0;
 936         req->rq_timedout = 0;
 937         req->rq_resend = 0;
 938         req->rq_restart = 0;
 939         spin_unlock(&req->rq_lock);
 940
 941         lwi = LWI_TIMEOUT_INTR(timeout == 0 ? LONG_MAX : timeout * HZ,
 942                                ctx_refresh_timeout, ctx_refresh_interrupt, req);
 943         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 944
 945         spin_lock(&ctx->cc_lock);
 946         /* five cases we are here:
 947          * 1. successfully refreshed;
 948          * 2. someone else mark this ctx dead by force;
 949          * 3. interruptted;
 950          * 4. timedout, and we don't want recover from the failure;
 951          * 5. timedout, and waked up upon recovery finished;
 952          */
 953         if (!ctx_is_refreshed(ctx)) {
 954                 /* timed out or interruptted */
 955                 list_del_init(&req->rq_ctx_chain);
 956                 spin_unlock(&ctx->cc_lock);
 957
 958                 LASSERT(rc != 0);
 959                 RETURN(rc);
 960         }
 961
 962         goto again;
 963 }
 964
 965 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 966 {
 967         struct sec_flavor_config *conf;
 968
 969         LASSERT(req->rq_import);
 970         LASSERT(req->rq_import->imp_sec);
 971         LASSERT(req->rq_cli_ctx);
 972         LASSERT(req->rq_cli_ctx->cc_sec);
 973         LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
 974
 975         /* special security flags accoding to opcode */
 976         switch (opcode) {
 977         case OST_READ:
 978                 req->rq_bulk_read = 1;
 979                 break;
 980         case OST_WRITE:
 981                 req->rq_bulk_write = 1;
 982                 break;
 983         case SEC_CTX_INIT:
 984                 req->rq_ctx_init = 1;
 985                 break;
 986         case SEC_CTX_FINI:
 987                 req->rq_ctx_fini = 1;
 988                 break;
 989         }
 990
 991         req->rq_sec_flavor = req->rq_cli_ctx->cc_sec->ps_flavor;
 992
 993         /* force SVC_NONE for context initiation rpc, SVC_AUTH for context
 994          * destruction rpc
 995          */
 996         if (unlikely(req->rq_ctx_init)) {
 997                 req->rq_sec_flavor = SEC_MAKE_RPC_FLAVOR(
 998                                 SEC_FLAVOR_POLICY(req->rq_sec_flavor),
 999                                 SEC_FLAVOR_SUBPOLICY(req->rq_sec_flavor),
1000                                 SEC_FLAVOR_SVC(SPTLRPC_SVC_NONE));
1001         } else if (unlikely(req->rq_ctx_fini)) {
1002                 req->rq_sec_flavor = SEC_MAKE_RPC_FLAVOR(
1003                                 SEC_FLAVOR_POLICY(req->rq_sec_flavor),
1004                                 SEC_FLAVOR_SUBPOLICY(req->rq_sec_flavor),
1005                                 SEC_FLAVOR_SVC(SPTLRPC_SVC_AUTH));
1006         }
1007
1008         conf = &req->rq_import->imp_obd->u.cli.cl_sec_conf;
1009
1010         /* user descriptor flag, except ROOTONLY which don't need, and
1011          * null security which can't
1012          */
1013         if ((conf->sfc_flags & PTLRPC_SEC_FL_ROOTONLY) == 0 &&
1014             req->rq_sec_flavor != SPTLRPC_FLVR_NULL)
1015                 req->rq_sec_flavor |= SEC_FLAVOR_FL_USER;
1016
1017         /* bulk security flag */
1018         if ((req->rq_bulk_read || req->rq_bulk_write) &&
1019             (conf->sfc_bulk_priv != BULK_PRIV_ALG_NULL ||
1020              conf->sfc_bulk_csum != BULK_CSUM_ALG_NULL))
1021                 req->rq_sec_flavor |= SEC_FLAVOR_FL_BULK;
1022 }
1023
1024 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
1025 {
1026         if (SEC_FLAVOR_SVC(req->rq_sec_flavor) != SPTLRPC_SVC_PRIV)
1027                 return;
1028
1029         LASSERT(req->rq_clrbuf);
1030         if (req->rq_pool || !req->rq_reqbuf)
1031                 return;
1032
1033         OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
1034         req->rq_reqbuf = NULL;
1035         req->rq_reqbuf_len = 0;
1036 }
1037
1038 /*
1039  * check whether current user have valid context for an import or not.
1040  * might repeatedly try in case of non-fatal errors.
1041  * return 0 on success, < 0 on failure
1042  */
1043 int sptlrpc_import_check_ctx(struct obd_import *imp)
1044 {
1045         struct ptlrpc_cli_ctx *ctx;
1046         struct ptlrpc_request *req = NULL;
1047         int rc;
1048         ENTRY;
1049
1050         might_sleep();
1051
1052         ctx = get_my_ctx(imp->imp_sec);
1053         if (!ctx)
1054                 RETURN(1);
1055
1056         if (ctx_is_eternal(ctx)) {
1057                 sptlrpc_ctx_put(ctx, 1);
1058                 RETURN(0);
1059         }
1060
1061         OBD_ALLOC_PTR(req);
1062         if (!req)
1063                 RETURN(-ENOMEM);
1064
1065         spin_lock_init(&req->rq_lock);
1066         atomic_set(&req->rq_refcount, 10000);
1067         INIT_LIST_HEAD(&req->rq_ctx_chain);
1068         init_waitqueue_head(&req->rq_reply_waitq);
1069         req->rq_import = imp;
1070         req->rq_cli_ctx = ctx;
1071
1072         rc = sptlrpc_req_refresh_ctx(req, 0);
1073         LASSERT(list_empty(&req->rq_ctx_chain));
1074         sptlrpc_ctx_put(req->rq_cli_ctx, 1);
1075         OBD_FREE_PTR(req);
1076
1077         RETURN(rc);
1078 }
1079
1080 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1081 {
1082         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1083         int rc = 0;
1084         ENTRY;
1085
1086         LASSERT(ctx);
1087         LASSERT(ctx->cc_sec);
1088         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1089
1090         /* we wrap bulk request here because now we can be sure
1091          * the context is uptodate.
1092          */
1093         if (req->rq_bulk) {
1094                 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1095                 if (rc)
1096                         RETURN(rc);
1097         }
1098
1099         switch (SEC_FLAVOR_SVC(req->rq_sec_flavor)) {
1100         case SPTLRPC_SVC_NONE:
1101         case SPTLRPC_SVC_AUTH:
1102                 LASSERT(ctx->cc_ops->sign);
1103                 rc = ctx->cc_ops->sign(ctx, req);
1104                 break;
1105         case SPTLRPC_SVC_PRIV:
1106                 LASSERT(ctx->cc_ops->seal);
1107                 rc = ctx->cc_ops->seal(ctx, req);
1108                 break;
1109         default:
1110                 LBUG();
1111         }
1112
1113         if (rc == 0) {
1114                 LASSERT(req->rq_reqdata_len);
1115                 LASSERT(req->rq_reqdata_len % 8 == 0);
1116                 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1117         }
1118
1119         RETURN(rc);
1120 }
1121
1122 /*
1123  * rq_nob_received is the actual received data length
1124  */
1125 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1126 {
1127         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1128         int rc;
1129         ENTRY;
1130
1131         LASSERT(ctx);
1132         LASSERT(ctx->cc_sec);
1133         LASSERT(ctx->cc_ops);
1134         LASSERT(req->rq_repbuf);
1135
1136         req->rq_repdata_len = req->rq_nob_received;
1137
1138         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1139                 CERROR("replied data length %d too small\n",
1140                        req->rq_nob_received);
1141                 RETURN(-EPROTO);
1142         }
1143
1144         if (req->rq_repbuf->lm_magic == LUSTRE_MSG_MAGIC_V1 ||
1145             req->rq_repbuf->lm_magic == LUSTRE_MSG_MAGIC_V1_SWABBED) {
1146                 /* it's must be null flavor, so our requets also should be
1147                  * in null flavor */
1148                 if (SEC_FLAVOR_POLICY(req->rq_sec_flavor) !=
1149                     SPTLRPC_POLICY_NULL) {
1150                         CERROR("request flavor is %x but reply with null\n",
1151                                req->rq_sec_flavor);
1152                         RETURN(-EPROTO);
1153                 }
1154         } else {
1155                 /* v2 message... */
1156                 ptlrpc_sec_flavor_t tmpf = req->rq_repbuf->lm_secflvr;
1157
1158                 if (req->rq_repbuf->lm_magic == LUSTRE_MSG_MAGIC_V2_SWABBED)
1159                         __swab32s(&tmpf);
1160
1161                 if (SEC_FLAVOR_POLICY(tmpf) !=
1162                     SEC_FLAVOR_POLICY(req->rq_sec_flavor)) {
1163                         CERROR("request policy %u while reply with %d\n",
1164                                SEC_FLAVOR_POLICY(req->rq_sec_flavor),
1165                                SEC_FLAVOR_POLICY(tmpf));
1166                         RETURN(-EPROTO);
1167                 }
1168
1169                 if ((SEC_FLAVOR_POLICY(req->rq_sec_flavor) !=
1170                      SPTLRPC_POLICY_NULL) &&
1171                     lustre_unpack_msg(req->rq_repbuf, req->rq_nob_received))
1172                         RETURN(-EPROTO);
1173         }
1174
1175         switch (SEC_FLAVOR_SVC(req->rq_sec_flavor)) {
1176         case SPTLRPC_SVC_NONE:
1177         case SPTLRPC_SVC_AUTH:
1178                 LASSERT(ctx->cc_ops->verify);
1179                 rc = ctx->cc_ops->verify(ctx, req);
1180                 break;
1181         case SPTLRPC_SVC_PRIV:
1182                 LASSERT(ctx->cc_ops->unseal);
1183                 rc = ctx->cc_ops->unseal(ctx, req);
1184                 break;
1185         default:
1186                 LBUG();
1187         }
1188
1189         LASSERT(rc || req->rq_repmsg || req->rq_resend);
1190         RETURN(rc);
1191 }
1192
1193 /**************************************************
1194  * security APIs                                  *
1195  **************************************************/
1196
1197 /*
1198  * let policy module to determine whether take refrence of
1199  * import or not.
1200  */
1201 static
1202 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1203                                        struct ptlrpc_svc_ctx *ctx,
1204                                        __u32 flavor,
1205                                        unsigned long flags)
1206 {
1207         struct ptlrpc_sec_policy *policy;
1208         struct ptlrpc_sec *sec;
1209         ENTRY;
1210
1211         flavor = SEC_FLAVOR_RPC(flavor);
1212
1213         if (ctx) {
1214                 LASSERT(imp->imp_dlm_fake == 1);
1215
1216                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1217                        imp->imp_obd->obd_type->typ_name,
1218                        imp->imp_obd->obd_name,
1219                        sptlrpc_flavor2name(flavor));
1220
1221                 policy = sptlrpc_policy_get(ctx->sc_policy);
1222                 flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1223         } else {
1224                 LASSERT(imp->imp_dlm_fake == 0);
1225
1226                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1227                        imp->imp_obd->obd_type->typ_name,
1228                        imp->imp_obd->obd_name,
1229                        sptlrpc_flavor2name(flavor));
1230
1231                 policy = sptlrpc_flavor2policy(flavor);
1232                 if (!policy) {
1233                         CERROR("invalid flavor 0x%x\n", flavor);
1234                         RETURN(NULL);
1235                 }
1236         }
1237
1238         sec = policy->sp_cops->create_sec(imp, ctx, flavor, flags);
1239         if (sec) {
1240                 atomic_inc(&sec->ps_refcount);
1241
1242                 /* take 1 busy count on behalf of sec itself,
1243                  * balanced in sptlrpc_set_put()
1244                  */
1245                 atomic_inc(&sec->ps_busy);
1246         } else
1247                 sptlrpc_policy_put(policy);
1248
1249         RETURN(sec);
1250 }
1251
1252 static
1253 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1254 {
1255         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1256
1257         LASSERT(policy);
1258         LASSERT(atomic_read(&sec->ps_refcount) == 0);
1259         LASSERT(atomic_read(&sec->ps_busy) == 0);
1260         LASSERT(policy->sp_cops->destroy_sec);
1261
1262         policy->sp_cops->destroy_sec(sec);
1263         sptlrpc_policy_put(policy);
1264 }
1265
1266 static
1267 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1268 {
1269         struct ptlrpc_sec_policy *policy = sec->ps_policy;
1270
1271         if (!atomic_dec_and_test(&sec->ps_refcount)) {
1272                 sptlrpc_policy_put(policy);
1273                 return;
1274         }
1275
1276         ctx_cache_flush(sec, -1, -1, 1, 1);
1277
1278         if (atomic_dec_and_test(&sec->ps_busy))
1279                 sptlrpc_sec_destroy(sec);
1280         else
1281                 CWARN("delay to destroy %s@%p: busy contexts\n",
1282                       policy->sp_name, sec);
1283 }
1284
1285 /*
1286  * return 1 means we should also destroy the sec structure.
1287  * normally return 0
1288  */
1289 static
1290 int sptlrpc_sec_destroy_ctx(struct ptlrpc_sec *sec,
1291                             struct ptlrpc_cli_ctx *ctx)
1292 {
1293         LASSERT(sec == ctx->cc_sec);
1294         LASSERT(atomic_read(&sec->ps_busy));
1295         LASSERT(atomic_read(&ctx->cc_refcount) == 0);
1296         LASSERT(hlist_unhashed(&ctx->cc_hash));
1297         LASSERT(list_empty(&ctx->cc_req_list));
1298         LASSERT(sec->ps_policy->sp_cops->destroy_ctx);
1299
1300         sec->ps_policy->sp_cops->destroy_ctx(sec, ctx);
1301
1302         if (atomic_dec_and_test(&sec->ps_busy)) {
1303                 LASSERT(atomic_read(&sec->ps_refcount) == 0);
1304                 return 1;
1305         }
1306
1307         return 0;
1308 }
1309
1310 /*
1311  * when complete successfully, req->rq_reqmsg should point to the
1312  * right place.
1313  */
1314 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1315 {
1316         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1317         struct ptlrpc_sec_policy *policy;
1318         int rc;
1319
1320         LASSERT(ctx);
1321         LASSERT(atomic_read(&ctx->cc_refcount));
1322         LASSERT(ctx->cc_sec);
1323         LASSERT(ctx->cc_sec->ps_policy);
1324         LASSERT(req->rq_reqmsg == NULL);
1325
1326         policy = ctx->cc_sec->ps_policy;
1327         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1328         if (!rc) {
1329                 LASSERT(req->rq_reqmsg);
1330                 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1331
1332                 /* zeroing preallocated buffer */
1333                 if (req->rq_pool)
1334                         memset(req->rq_reqmsg, 0, msgsize);
1335         }
1336
1337         return rc;
1338 }
1339
1340 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1341 {
1342         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1343         struct ptlrpc_sec_policy *policy;
1344
1345         LASSERT(ctx);
1346         LASSERT(atomic_read(&ctx->cc_refcount));
1347         LASSERT(ctx->cc_sec);
1348         LASSERT(ctx->cc_sec->ps_policy);
1349         LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1350
1351         policy = ctx->cc_sec->ps_policy;
1352         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1353 }
1354
1355 /*
1356  * NOTE caller must guarantee the buffer size is enough for the enlargement
1357  */
1358 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1359                                   int segment, int newsize)
1360 {
1361         void   *src, *dst;
1362         int     oldsize, oldmsg_size, movesize;
1363
1364         LASSERT(segment < msg->lm_bufcount);
1365         LASSERT(msg->lm_buflens[segment] <= newsize);
1366
1367         if (msg->lm_buflens[segment] == newsize)
1368                 return;
1369
1370         /* nothing to do if we are enlarging the last segment */
1371         if (segment == msg->lm_bufcount - 1) {
1372                 msg->lm_buflens[segment] = newsize;
1373                 return;
1374         }
1375
1376         oldsize = msg->lm_buflens[segment];
1377
1378         src = lustre_msg_buf(msg, segment + 1, 0);
1379         msg->lm_buflens[segment] = newsize;
1380         dst = lustre_msg_buf(msg, segment + 1, 0);
1381         msg->lm_buflens[segment] = oldsize;
1382
1383         /* move from segment + 1 to end segment */
1384         LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1385         oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1386         movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1387         LASSERT(movesize >= 0);
1388
1389         if (movesize)
1390                 memmove(dst, src, movesize);
1391
1392         /* note we don't clear the ares where old data live, not secret */
1393
1394         /* finally set new segment size */
1395         msg->lm_buflens[segment] = newsize;
1396 }
1397 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1398
1399 /*
1400  * enlarge @segment of upper message req->rq_reqmsg to @newsize, all data
1401  * will be preserved after enlargement. this must be called after rq_reqmsg has
1402  * been intialized at least.
1403  *
1404  * caller's attention: upon return, rq_reqmsg and rq_reqlen might have
1405  * been changed.
1406  */
1407 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1408                                int segment, int newsize)
1409 {
1410         struct ptlrpc_cli_ctx    *ctx = req->rq_cli_ctx;
1411         struct ptlrpc_sec_cops   *cops;
1412         struct lustre_msg        *msg = req->rq_reqmsg;
1413
1414         LASSERT(ctx);
1415         LASSERT(msg);
1416         LASSERT(msg->lm_bufcount > segment);
1417         LASSERT(msg->lm_buflens[segment] <= newsize);
1418
1419         if (msg->lm_buflens[segment] == newsize)
1420                 return 0;
1421
1422         cops = ctx->cc_sec->ps_policy->sp_cops;
1423         LASSERT(cops->enlarge_reqbuf);
1424         return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1425 }
1426 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1427
1428 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1429 {
1430         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1431         struct ptlrpc_sec_policy *policy;
1432         ENTRY;
1433
1434         LASSERT(ctx);
1435         LASSERT(atomic_read(&ctx->cc_refcount));
1436         LASSERT(ctx->cc_sec);
1437         LASSERT(ctx->cc_sec->ps_policy);
1438
1439         if (req->rq_repbuf)
1440                 RETURN(0);
1441
1442         policy = ctx->cc_sec->ps_policy;
1443         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1444 }
1445
1446 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1447 {
1448         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1449         struct ptlrpc_sec_policy *policy;
1450         ENTRY;
1451
1452         LASSERT(ctx);
1453         LASSERT(atomic_read(&ctx->cc_refcount));
1454         LASSERT(ctx->cc_sec);
1455         LASSERT(ctx->cc_sec->ps_policy);
1456         LASSERT(req->rq_repbuf);
1457
1458         policy = ctx->cc_sec->ps_policy;
1459         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1460         EXIT;
1461 }
1462
1463 int sptlrpc_import_get_sec(struct obd_import *imp,
1464                            struct ptlrpc_svc_ctx *ctx,
1465                            __u32 flavor,
1466                            unsigned long flags)
1467 {
1468         struct obd_device *obd = imp->imp_obd;
1469         ENTRY;
1470
1471         LASSERT(obd);
1472         LASSERT(obd->obd_type);
1473
1474         /* old sec might be still there in reconnecting */
1475         if (imp->imp_sec)
1476                 RETURN(0);
1477
1478         imp->imp_sec = sptlrpc_sec_create(imp, ctx, flavor, flags);
1479         if (!imp->imp_sec)
1480                 RETURN(-EINVAL);
1481
1482         RETURN(0);
1483 }
1484
1485 void sptlrpc_import_put_sec(struct obd_import *imp)
1486 {
1487         if (imp->imp_sec == NULL)
1488                 return;
1489
1490         sptlrpc_sec_put(imp->imp_sec);
1491         imp->imp_sec = NULL;
1492 }
1493
1494 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1495 {
1496         if (imp == NULL || imp->imp_sec == NULL)
1497                 return;
1498
1499         /* use 'grace' mode, it's crutial see explain in
1500          * sptlrpc_req_refresh_ctx()
1501          */
1502         ctx_cache_flush(imp->imp_sec, 0, 0, 1, 1);
1503 }
1504
1505 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1506 {
1507         uint64_t pag = cfs_current()->uid == 0 ? 0 : CURRENT_PAG;
1508
1509         if (imp == NULL || imp->imp_sec == NULL)
1510                 return;
1511
1512         ctx_cache_flush(imp->imp_sec, pag, cfs_current()->uid, 1, 1);
1513 }
1514 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1515
1516 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1517 {
1518         if (imp == NULL || imp->imp_sec == NULL)
1519                 return;
1520
1521         ctx_cache_flush(imp->imp_sec, -1, -1, 0, 1);
1522 }
1523 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1524
1525 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1526                                 struct ptlrpc_cli_ctx *ctx)
1527 {
1528         struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1529
1530         if (!policy->sp_cops->install_rctx)
1531                 return 0;
1532         return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1533 }
1534
1535 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1536                                 struct ptlrpc_svc_ctx *ctx)
1537 {
1538         struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1539
1540         if (!policy->sp_sops->install_rctx)
1541                 return 0;
1542         return policy->sp_sops->install_rctx(imp, ctx);
1543 }
1544
1545 /****************************************
1546  * server side security                 *
1547  ****************************************/
1548
1549 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
1550 {
1551         struct ptlrpc_sec_policy *policy;
1552         struct lustre_msg *msg = req->rq_reqbuf;
1553         int rc;
1554         ENTRY;
1555
1556         LASSERT(msg);
1557         LASSERT(req->rq_reqmsg == NULL);
1558         LASSERT(req->rq_repmsg == NULL);
1559
1560         /*
1561          * in any case we avoid to call unpack_msg() for request of null flavor
1562          * which will later be done by ptlrpc_server_handle_request().
1563          */
1564         if (req->rq_reqdata_len < sizeof(struct lustre_msg)) {
1565                 CERROR("request size %d too small\n", req->rq_reqdata_len);
1566                 RETURN(SECSVC_DROP);
1567         }
1568
1569         if (msg->lm_magic == LUSTRE_MSG_MAGIC_V1 ||
1570             msg->lm_magic == LUSTRE_MSG_MAGIC_V1_SWABBED) {
1571                 req->rq_sec_flavor = SPTLRPC_FLVR_NULL;
1572         } else {
1573                 req->rq_sec_flavor = msg->lm_secflvr;
1574
1575                 if (msg->lm_magic == LUSTRE_MSG_MAGIC_V2_SWABBED)
1576                         __swab32s(&req->rq_sec_flavor);
1577
1578                 if ((SEC_FLAVOR_POLICY(req->rq_sec_flavor) !=
1579                      SPTLRPC_POLICY_NULL) &&
1580                     lustre_unpack_msg(msg, req->rq_reqdata_len))
1581                         RETURN(SECSVC_DROP);
1582         }
1583
1584         policy = sptlrpc_flavor2policy(req->rq_sec_flavor);
1585         if (!policy) {
1586                 CERROR("unsupported security flavor %x\n", req->rq_sec_flavor);
1587                 RETURN(SECSVC_DROP);
1588         }
1589
1590         LASSERT(policy->sp_sops->accept);
1591         rc = policy->sp_sops->accept(req);
1592
1593         LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
1594         sptlrpc_policy_put(policy);
1595
1596         /* FIXME move to proper place */
1597         if (rc == SECSVC_OK) {
1598                 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
1599
1600                 if (opc == OST_WRITE)
1601                         req->rq_bulk_write = 1;
1602                 else if (opc == OST_READ)
1603                         req->rq_bulk_read = 1;
1604         }
1605
1606         RETURN(rc);
1607 }
1608
1609 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req,
1610                          int msglen)
1611 {
1612         struct ptlrpc_sec_policy *policy;
1613         struct ptlrpc_reply_state *rs;
1614         int rc;
1615         ENTRY;
1616
1617         LASSERT(req->rq_svc_ctx);
1618         LASSERT(req->rq_svc_ctx->sc_policy);
1619
1620         policy = req->rq_svc_ctx->sc_policy;
1621         LASSERT(policy->sp_sops->alloc_rs);
1622
1623         rc = policy->sp_sops->alloc_rs(req, msglen);
1624         if (unlikely(rc == -ENOMEM)) {
1625                 /* failed alloc, try emergency pool */
1626                 rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service);
1627                 if (rs == NULL)
1628                         RETURN(-ENOMEM);
1629
1630                 req->rq_reply_state = rs;
1631                 rc = policy->sp_sops->alloc_rs(req, msglen);
1632                 if (rc) {
1633                         lustre_put_emerg_rs(rs);
1634                         req->rq_reply_state = NULL;
1635                 }
1636         }
1637
1638         LASSERT(rc != 0 ||
1639                 (req->rq_reply_state && req->rq_reply_state->rs_msg));
1640
1641         RETURN(rc);
1642 }
1643
1644 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
1645 {
1646         struct ptlrpc_sec_policy *policy;
1647         int rc;
1648         ENTRY;
1649
1650         LASSERT(req->rq_svc_ctx);
1651         LASSERT(req->rq_svc_ctx->sc_policy);
1652
1653         policy = req->rq_svc_ctx->sc_policy;
1654         LASSERT(policy->sp_sops->authorize);
1655
1656         rc = policy->sp_sops->authorize(req);
1657         LASSERT(rc || req->rq_reply_state->rs_repdata_len);
1658
1659         RETURN(rc);
1660 }
1661
1662 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
1663 {
1664         struct ptlrpc_sec_policy *policy;
1665         unsigned int prealloc;
1666         ENTRY;
1667
1668         LASSERT(rs->rs_svc_ctx);
1669         LASSERT(rs->rs_svc_ctx->sc_policy);
1670
1671         policy = rs->rs_svc_ctx->sc_policy;
1672         LASSERT(policy->sp_sops->free_rs);
1673
1674         prealloc = rs->rs_prealloc;
1675         policy->sp_sops->free_rs(rs);
1676
1677         if (prealloc)
1678                 lustre_put_emerg_rs(rs);
1679         EXIT;
1680 }
1681
1682 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
1683 {
1684         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
1685
1686         if (ctx == NULL)
1687                 return;
1688
1689         LASSERT(atomic_read(&ctx->sc_refcount) > 0);
1690         atomic_inc(&ctx->sc_refcount);
1691 }
1692
1693 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
1694 {
1695         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
1696
1697         if (ctx == NULL)
1698                 return;
1699
1700         LASSERT(atomic_read(&ctx->sc_refcount) > 0);
1701         if (atomic_dec_and_test(&ctx->sc_refcount)) {
1702                 if (ctx->sc_policy->sp_sops->free_ctx)
1703                         ctx->sc_policy->sp_sops->free_ctx(ctx);
1704         }
1705         req->rq_svc_ctx = NULL;
1706 }
1707
1708 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
1709 {
1710         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
1711
1712         if (ctx == NULL)
1713                 return;
1714
1715         LASSERT(atomic_read(&ctx->sc_refcount) > 0);
1716         if (ctx->sc_policy->sp_sops->invalidate_ctx)
1717                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
1718 }
1719 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
1720
1721 /****************************************
1722  * bulk security                        *
1723  ****************************************/
1724
1725 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
1726                           struct ptlrpc_bulk_desc *desc)
1727 {
1728         struct ptlrpc_cli_ctx *ctx;
1729
1730         if (!SEC_FLAVOR_HAS_BULK(req->rq_sec_flavor))
1731                 return 0;
1732
1733         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
1734
1735         ctx = req->rq_cli_ctx;
1736         if (ctx->cc_ops->wrap_bulk)
1737                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
1738         return 0;
1739 }
1740 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
1741
1742 static
1743 void pga_to_bulk_desc(int nob, obd_count pg_count, struct brw_page **pga,
1744                       struct ptlrpc_bulk_desc *desc)
1745 {
1746         int i;
1747
1748         LASSERT(pga);
1749         LASSERT(*pga);
1750
1751         for (i = 0; i < pg_count && nob > 0; i++) {
1752 #ifdef __KERNEL__
1753                 desc->bd_iov[i].kiov_page = pga[i]->pg;
1754                 desc->bd_iov[i].kiov_len = pga[i]->count > nob ?
1755                                            nob : pga[i]->count;
1756                 desc->bd_iov[i].kiov_offset = pga[i]->off & ~CFS_PAGE_MASK;
1757 #else
1758 #warning FIXME for liblustre!
1759                 desc->bd_iov[i].iov_base = pga[i]->pg->addr;
1760                 desc->bd_iov[i].iov_len = pga[i]->count > nob ?
1761                                            nob : pga[i]->count;
1762 #endif
1763
1764                 desc->bd_iov_count++;
1765                 nob -= pga[i]->count;
1766         }
1767 }
1768
1769 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
1770                                  int nob, obd_count pg_count,
1771                                  struct brw_page **pga)
1772 {
1773         struct ptlrpc_bulk_desc *desc;
1774         struct ptlrpc_cli_ctx *ctx;
1775         int rc = 0;
1776
1777         if (!SEC_FLAVOR_HAS_BULK(req->rq_sec_flavor))
1778                 return 0;
1779
1780         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
1781
1782         OBD_ALLOC(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
1783         if (desc == NULL) {
1784                 CERROR("out of memory, can't verify bulk read data\n");
1785                 return -ENOMEM;
1786         }
1787
1788         pga_to_bulk_desc(nob, pg_count, pga, desc);
1789
1790         ctx = req->rq_cli_ctx;
1791         if (ctx->cc_ops->unwrap_bulk)
1792                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
1793
1794         OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
1795
1796         return rc;
1797 }
1798 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
1799
1800 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
1801                                   struct ptlrpc_bulk_desc *desc)
1802 {
1803         struct ptlrpc_cli_ctx *ctx;
1804
1805         if (!SEC_FLAVOR_HAS_BULK(req->rq_sec_flavor))
1806                 return 0;
1807
1808         LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
1809
1810         ctx = req->rq_cli_ctx;
1811         if (ctx->cc_ops->unwrap_bulk)
1812                 return ctx->cc_ops->unwrap_bulk(ctx, req, desc);
1813
1814         return 0;
1815 }
1816 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
1817
1818 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
1819                           struct ptlrpc_bulk_desc *desc)
1820 {
1821         struct ptlrpc_svc_ctx *ctx;
1822
1823         if (!SEC_FLAVOR_HAS_BULK(req->rq_sec_flavor))
1824                 return 0;
1825
1826         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
1827
1828         ctx = req->rq_svc_ctx;
1829         if (ctx->sc_policy->sp_sops->wrap_bulk)
1830                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
1831
1832         return 0;
1833 }
1834 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
1835
1836 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
1837                             struct ptlrpc_bulk_desc *desc)
1838 {
1839         struct ptlrpc_svc_ctx *ctx;
1840
1841         if (!SEC_FLAVOR_HAS_BULK(req->rq_sec_flavor))
1842                 return 0;
1843
1844         LASSERT(req->rq_bulk_read || req->rq_bulk_write);
1845
1846         ctx = req->rq_svc_ctx;
1847         if (ctx->sc_policy->sp_sops->unwrap_bulk);
1848                 return ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
1849
1850         return 0;
1851 }
1852 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
1853
1854
1855 /****************************************
1856  * user descriptor helpers              *
1857  ****************************************/
1858
1859 int sptlrpc_user_desc_size(void)
1860 {
1861 #ifdef __KERNEL__
1862         int ngroups = current_ngroups;
1863
1864         if (ngroups > LUSTRE_MAX_GROUPS)
1865                 ngroups = LUSTRE_MAX_GROUPS;
1866
1867         return sizeof(struct ptlrpc_user_desc) + ngroups * sizeof(__u32);
1868 #else
1869         return sizeof(struct ptlrpc_user_desc);
1870 #endif
1871 }
1872 EXPORT_SYMBOL(sptlrpc_user_desc_size);
1873
1874 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
1875 {
1876         struct ptlrpc_user_desc *pud;
1877
1878         pud = lustre_msg_buf(msg, offset, 0);
1879
1880         pud->pud_uid = cfs_current()->uid;
1881         pud->pud_gid = cfs_current()->gid;
1882         pud->pud_fsuid = cfs_current()->fsuid;
1883         pud->pud_fsgid = cfs_current()->fsgid;
1884         pud->pud_cap = cfs_current()->cap_effective;
1885         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
1886
1887 #ifdef __KERNEL__
1888         task_lock(current);
1889         if (pud->pud_ngroups > current_ngroups)
1890                 pud->pud_ngroups = current_ngroups;
1891         memcpy(pud->pud_groups, cfs_current()->group_info->blocks[0],
1892                pud->pud_ngroups * sizeof(__u32));
1893         task_unlock(current);
1894 #endif
1895
1896         return 0;
1897 }
1898 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
1899
1900 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset)
1901 {
1902         struct ptlrpc_user_desc *pud;
1903         int                      i;
1904
1905         pud = lustre_msg_buf(msg, offset, sizeof(*pud));
1906         if (!pud)
1907                 return -EINVAL;
1908
1909         if (lustre_msg_swabbed(msg)) {
1910                 __swab32s(&pud->pud_uid);
1911                 __swab32s(&pud->pud_gid);
1912                 __swab32s(&pud->pud_fsuid);
1913                 __swab32s(&pud->pud_fsgid);
1914                 __swab32s(&pud->pud_cap);
1915                 __swab32s(&pud->pud_ngroups);
1916         }
1917
1918         if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
1919                 CERROR("%u groups is too large\n", pud->pud_ngroups);
1920                 return -EINVAL;
1921         }
1922
1923         if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
1924             msg->lm_buflens[offset]) {
1925                 CERROR("%u groups are claimed but bufsize only %u\n",
1926                        pud->pud_ngroups, msg->lm_buflens[offset]);
1927                 return -EINVAL;
1928         }
1929
1930         if (lustre_msg_swabbed(msg)) {
1931                 for (i = 0; i < pud->pud_ngroups; i++)
1932                         __swab32s(&pud->pud_groups[i]);
1933         }
1934
1935         return 0;
1936 }
1937 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
1938
1939 /****************************************
1940  * user supplied flavor string parsing  *
1941  ****************************************/
1942
1943 static
1944 int get_default_flavor(enum lustre_part to_part, struct sec_flavor_config *conf)
1945 {
1946         conf->sfc_bulk_priv = BULK_PRIV_ALG_NULL;
1947         conf->sfc_bulk_csum = BULK_CSUM_ALG_NULL;
1948         conf->sfc_flags = 0;
1949
1950         switch (to_part) {
1951         case LUSTRE_MDT:
1952                 conf->sfc_rpc_flavor = SPTLRPC_FLVR_PLAIN;
1953                 return 0;
1954         case LUSTRE_OST:
1955                 conf->sfc_rpc_flavor = SPTLRPC_FLVR_NULL;
1956                 return 0;
1957         default:
1958                 CERROR("Unknown to lustre part %d, apply defaults\n", to_part);
1959                 conf->sfc_rpc_flavor = SPTLRPC_FLVR_NULL;
1960                 return -EINVAL;
1961         }
1962 }
1963
1964 static
1965 void get_flavor_by_rpc(__u32 rpc_flavor, struct sec_flavor_config *conf)
1966 {
1967         conf->sfc_rpc_flavor = rpc_flavor;
1968         conf->sfc_bulk_priv = BULK_PRIV_ALG_NULL;
1969         conf->sfc_bulk_csum = BULK_CSUM_ALG_NULL;
1970         conf->sfc_flags = 0;
1971
1972         switch (rpc_flavor) {
1973         case SPTLRPC_FLVR_NULL:
1974         case SPTLRPC_FLVR_PLAIN:
1975                 break;
1976         case SPTLRPC_FLVR_KRB5P:
1977                 conf->sfc_bulk_priv = BULK_PRIV_ALG_ARC4;
1978                 /* fall through */
1979         case SPTLRPC_FLVR_KRB5I:
1980                 conf->sfc_bulk_csum = BULK_CSUM_ALG_SHA1;
1981                 break;
1982         default:
1983                 LBUG();
1984         }
1985 }
1986
1987 static
1988 void get_flavor_by_rpc_bulk(__u32 rpc_flavor, int bulk_priv,
1989                             struct sec_flavor_config *conf)
1990 {
1991         if (bulk_priv)
1992                 conf->sfc_bulk_priv = BULK_PRIV_ALG_ARC4;
1993         else
1994                 conf->sfc_bulk_priv = BULK_PRIV_ALG_NULL;
1995
1996         switch (rpc_flavor) {
1997         case SPTLRPC_FLVR_PLAIN:
1998                 conf->sfc_bulk_csum = BULK_CSUM_ALG_MD5;
1999                 break;
2000         case SPTLRPC_FLVR_KRB5I:
2001         case SPTLRPC_FLVR_KRB5P:
2002                 conf->sfc_bulk_csum = BULK_CSUM_ALG_SHA1;
2003                 break;
2004         default:
2005                 LBUG();
2006         }
2007 }
2008
2009 static __u32 __flavors[] = {
2010         SPTLRPC_FLVR_NULL,
2011         SPTLRPC_FLVR_PLAIN,
2012         SPTLRPC_FLVR_KRB5I,
2013         SPTLRPC_FLVR_KRB5P,
2014 };
2015
2016 #define __nflavors      (sizeof(__flavors)/sizeof(__u32))
2017
2018 /*
2019  * flavor string format: rpc[-bulk{n|i|p}[:cksum/enc]]
2020  * for examples:
2021  *  null
2022  *  plain-bulki
2023  *  krb5p-bulkn
2024  *  krb5i-bulkp
2025  *  krb5i-bulkp:sha512/arc4
2026  */
2027 int sptlrpc_parse_flavor(enum lustre_part from_part, enum lustre_part to_part,
2028                          char *str, struct sec_flavor_config *conf)
2029 {
2030         char   *f, *bulk, *alg, *enc;
2031         char    buf[64];
2032         int     i, bulk_priv;
2033         ENTRY;
2034
2035         if (str == NULL) {
2036                 if (get_default_flavor(to_part, conf))
2037                         return -EINVAL;
2038                 goto set_flags;
2039         }
2040
2041         for (i = 0; i < __nflavors; i++) {
2042                 f = sptlrpc_flavor2name(__flavors[i]);
2043                 if (strncmp(str, f, strlen(f)) == 0)
2044                         break;
2045         }
2046
2047         if (i >= __nflavors)
2048                 GOTO(invalid, -EINVAL);
2049
2050         /* prepare local buffer thus we can modify it as we want */
2051         strncpy(buf, str, 64);
2052         buf[64 - 1] = '\0';
2053
2054         /* find bulk string */
2055         bulk = strchr(buf, '-');
2056         if (bulk)
2057                 *bulk++ = '\0';
2058
2059         /* now the first part must equal to rpc flavor name */
2060         if (strcmp(buf, f) != 0)
2061                 GOTO(invalid, -EINVAL);
2062
2063         get_flavor_by_rpc(__flavors[i], conf);
2064
2065         if (bulk == NULL)
2066                 goto set_flags;
2067
2068         /* null flavor should not have any suffix */
2069         if (__flavors[i] == SPTLRPC_FLVR_NULL)
2070                 GOTO(invalid, -EINVAL);
2071
2072         /* find bulk algorithm string */
2073         alg = strchr(bulk, ':');
2074         if (alg)
2075                 *alg++ = '\0';
2076
2077         /* verify bulk section */
2078         if (strcmp(bulk, "bulkn") == 0) {
2079                 conf->sfc_bulk_csum = BULK_CSUM_ALG_NULL;
2080                 conf->sfc_bulk_priv = BULK_PRIV_ALG_NULL;
2081                 goto set_flags;
2082         }
2083
2084         if (strcmp(bulk, "bulki") == 0)
2085                 bulk_priv = 0;
2086         else if (strcmp(bulk, "bulkp") == 0)
2087                 bulk_priv = 1;
2088         else
2089                 GOTO(invalid, -EINVAL);
2090
2091         /* plain policy dosen't support bulk encryption */
2092         if (bulk_priv && __flavors[i] == SPTLRPC_FLVR_PLAIN)
2093                 GOTO(invalid, -EINVAL);
2094
2095         get_flavor_by_rpc_bulk(__flavors[i], bulk_priv, conf);
2096
2097         if (alg == NULL)
2098                 goto set_flags;
2099
2100         /* find encryption algorithm string */
2101         enc = strchr(alg, '/');
2102         if (enc)
2103                 *enc++ = '\0';
2104
2105         /* bulk combination sanity check */
2106         if ((bulk_priv && enc == NULL) || (bulk_priv == 0 && enc))
2107                 GOTO(invalid, -EINVAL);
2108
2109         /* checksum algorithm */
2110         for (i = 0; i < BULK_CSUM_ALG_MAX; i++) {
2111                 if (strcmp(alg, sptlrpc_bulk_csum_alg2name(i)) == 0) {
2112                         conf->sfc_bulk_csum = i;
2113                         break;
2114                 }
2115         }
2116         if (i >= BULK_CSUM_ALG_MAX)
2117                 GOTO(invalid, -EINVAL);
2118
2119         /* privacy algorithm */
2120         if (enc) {
2121                 if (strcmp(enc, "arc4") != 0)
2122                         GOTO(invalid, -EINVAL);
2123                 conf->sfc_bulk_priv = BULK_PRIV_ALG_ARC4;
2124         }
2125
2126 set_flags:
2127         /* * set ROOTONLY flag:
2128          *   - to OST
2129          *   - from MDT to MDT
2130          * * set BULK flag for:
2131          *   - from CLI to OST
2132          */
2133         if (to_part == LUSTRE_OST ||
2134             (from_part == LUSTRE_MDT && to_part == LUSTRE_MDT))
2135                 conf->sfc_flags |= PTLRPC_SEC_FL_ROOTONLY;
2136         if (from_part == LUSTRE_CLI && to_part == LUSTRE_OST)
2137                 conf->sfc_flags |= PTLRPC_SEC_FL_BULK;
2138
2139 #ifdef __BIG_ENDIAN
2140         __swab32s(&conf->sfc_rpc_flavor);
2141         __swab32s(&conf->sfc_bulk_csum);
2142         __swab32s(&conf->sfc_bulk_priv);
2143         __swab32s(&conf->sfc_flags);
2144 #endif
2145         return 0;
2146 invalid:
2147         CERROR("invalid flavor string: %s\n", str);
2148         return -EINVAL;
2149 }
2150 EXPORT_SYMBOL(sptlrpc_parse_flavor);
2151
2152 /****************************************
2153  * misc helpers                         *
2154  ****************************************/
2155
2156 const char * sec2target_str(struct ptlrpc_sec *sec)
2157 {
2158         if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2159                 return "*";
2160         if (sec->ps_flags & PTLRPC_SEC_FL_REVERSE)
2161                 return "c";
2162         return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2163 }
2164 EXPORT_SYMBOL(sec2target_str);
2165
2166 /****************************************
2167  * initialize/finalize                  *
2168  ****************************************/
2169
2170 int sptlrpc_init(void)
2171 {
2172         int rc;
2173
2174         rc = sptlrpc_enc_pool_init();
2175         if (rc)
2176                 goto out;
2177
2178         rc = sptlrpc_null_init();
2179         if (rc)
2180                 goto out_pool;
2181
2182         rc = sptlrpc_plain_init();
2183         if (rc)
2184                 goto out_null;
2185
2186         rc = sptlrpc_lproc_init();
2187         if (rc)
2188                 goto out_plain;
2189
2190         return 0;
2191
2192 out_plain:
2193         sptlrpc_plain_fini();
2194 out_null:
2195         sptlrpc_null_fini();
2196 out_pool:
2197         sptlrpc_enc_pool_fini();
2198 out:
2199         return rc;
2200 }
2201
2202 void sptlrpc_fini(void)
2203 {
2204         sptlrpc_lproc_fini();
2205         sptlrpc_plain_fini();
2206         sptlrpc_null_fini();
2207         sptlrpc_enc_pool_fini();
2208 }