lustre/ldlm/ldlm_lock.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (c) 2002 Cluster File Systems, Inc.
   5  *   Author: Peter Braam <braam@clusterfs.com>
   6  *   Author: Phil Schwan <phil@clusterfs.com>
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LDLM
  25
  26 #include <linux/slab.h>
  27 #include <linux/module.h>
  28 #include <linux/random.h>
  29 #include <linux/lustre_dlm.h>
  30 #include <linux/lustre_mds.h>
  31 #include <linux/obd_class.h>
  32
  33 //struct lustre_lock ldlm_everything_lock;
  34
  35 /* lock types */
  36 char *ldlm_lockname[] = {
  37         [0] "--",
  38         [LCK_EX] "EX",
  39         [LCK_PW] "PW",
  40         [LCK_PR] "PR",
  41         [LCK_CW] "CW",
  42         [LCK_CR] "CR",
  43         [LCK_NL] "NL"
  44 };
  45 char *ldlm_typename[] = {
  46         [LDLM_PLAIN] "PLN",
  47         [LDLM_EXTENT] "EXT",
  48 };
  49
  50 char *ldlm_it2str(int it)
  51 {
  52         switch (it) {
  53         case IT_OPEN:
  54                 return "open";
  55         case IT_CREAT:
  56                 return "creat";
  57         case (IT_OPEN | IT_CREAT):
  58                 return "open|creat";
  59         case IT_MKDIR:
  60                 return "mkdir";
  61         case IT_LINK:
  62                 return "link";
  63         case IT_LINK2:
  64                 return "link2";
  65         case IT_SYMLINK:
  66                 return "symlink";
  67         case IT_UNLINK:
  68                 return "unlink";
  69         case IT_RMDIR:
  70                 return "rmdir";
  71         case IT_RENAME:
  72                 return "rename";
  73         case IT_RENAME2:
  74                 return "rename2";
  75         case IT_READDIR:
  76                 return "readdir";
  77         case IT_GETATTR:
  78                 return "getattr";
  79         case IT_SETATTR:
  80                 return "setattr";
  81         case IT_READLINK:
  82                 return "readlink";
  83         case IT_MKNOD:
  84                 return "mknod";
  85         case IT_LOOKUP:
  86                 return "lookup";
  87         default:
  88                 CERROR("Unknown intent %d\n", it);
  89                 return "UNKNOWN";
  90         }
  91 }
  92
  93 extern kmem_cache_t *ldlm_lock_slab;
  94
  95 static int ldlm_plain_compat(struct ldlm_lock *a, struct ldlm_lock *b);
  96
  97 ldlm_res_compat ldlm_res_compat_table[] = {
  98         [LDLM_PLAIN] ldlm_plain_compat,
  99         [LDLM_EXTENT] ldlm_extent_compat,
 100 };
 101
 102 static ldlm_res_policy ldlm_intent_policy_func;
 103
 104 static int ldlm_plain_policy(struct ldlm_lock *lock, void *req_cookie,
 105                              ldlm_mode_t mode, int flags, void *data)
 106 {
 107         if ((flags & LDLM_FL_HAS_INTENT) && ldlm_intent_policy_func) {
 108                 return ldlm_intent_policy_func(lock, req_cookie, mode, flags,
 109                                                data);
 110         }
 111
 112         return ELDLM_OK;
 113 }
 114
 115 ldlm_res_policy ldlm_res_policy_table[] = {
 116         [LDLM_PLAIN] ldlm_plain_policy,
 117         [LDLM_EXTENT] ldlm_extent_policy,
 118 };
 119
 120 void ldlm_register_intent(ldlm_res_policy arg)
 121 {
 122         ldlm_intent_policy_func = arg;
 123 }
 124
 125 void ldlm_unregister_intent(void)
 126 {
 127         ldlm_intent_policy_func = NULL;
 128 }
 129
 130 /*
 131  * REFCOUNTED LOCK OBJECTS
 132  */
 133
 134
 135 /*
 136  * Lock refcounts, during creation:
 137  *   - one special one for allocation, dec'd only once in destroy
 138  *   - one for being a lock that's in-use
 139  *   - one for the addref associated with a new lock
 140  */
 141 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 142 {
 143         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 144         lock->l_refc++;
 145         ldlm_resource_getref(lock->l_resource);
 146         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 147         return lock;
 148 }
 149
 150 void ldlm_lock_put(struct ldlm_lock *lock)
 151 {
 152         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 153         ENTRY;
 154
 155         l_lock(&ns->ns_lock);
 156         lock->l_refc--;
 157         //LDLM_DEBUG(lock, "after refc--");
 158         if (lock->l_refc < 0)
 159                 LBUG();
 160
 161         if (ldlm_resource_put(lock->l_resource)) {
 162                 LASSERT(lock->l_refc == 0);
 163                 lock->l_resource = NULL;
 164         }
 165         if (lock->l_parent)
 166                 LDLM_LOCK_PUT(lock->l_parent);
 167
 168         if (lock->l_refc == 0 && (lock->l_flags & LDLM_FL_DESTROYED)) {
 169                 l_unlock(&ns->ns_lock);
 170                 LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing");
 171
 172                 //spin_lock(&ldlm_handle_lock);
 173                 spin_lock(&ns->ns_counter_lock);
 174                 ns->ns_locks--;
 175                 spin_unlock(&ns->ns_counter_lock);
 176
 177                 lock->l_resource = NULL;
 178                 lock->l_random = DEAD_HANDLE_MAGIC;
 179                 if (lock->l_export && lock->l_export->exp_connection)
 180                         ptlrpc_put_connection(lock->l_export->exp_connection);
 181                 kmem_cache_free(ldlm_lock_slab, lock);
 182                 //spin_unlock(&ldlm_handle_lock);
 183                 CDEBUG(D_MALLOC, "kfreed 'lock': %d at %p (tot 0).\n",
 184                        sizeof(*lock), lock);
 185         } else
 186                 l_unlock(&ns->ns_lock);
 187
 188         EXIT;
 189 }
 190
 191 void ldlm_lock_destroy(struct ldlm_lock *lock)
 192 {
 193         ENTRY;
 194         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 195
 196         if (!list_empty(&lock->l_children)) {
 197                 LDLM_DEBUG(lock, "still has children (%p)!",
 198                            lock->l_children.next);
 199                 ldlm_lock_dump(lock);
 200                 LBUG();
 201         }
 202         if (lock->l_readers || lock->l_writers) {
 203                 LDLM_DEBUG(lock, "lock still has references");
 204                 ldlm_lock_dump(lock);
 205         }
 206
 207         if (!list_empty(&lock->l_res_link)) {
 208                 ldlm_lock_dump(lock);
 209                 LBUG();
 210         }
 211
 212         if (lock->l_flags & LDLM_FL_DESTROYED) {
 213                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 214                 EXIT;
 215                 return;
 216         }
 217
 218         list_del_init(&lock->l_lru);
 219         list_del(&lock->l_export_chain);
 220         lock->l_export = NULL;
 221         lock->l_flags |= LDLM_FL_DESTROYED;
 222
 223         /* Wake anyone waiting for this lock */
 224         /* FIXME: I should probably add yet another flag, instead of using
 225          * l_export to only call this on clients */
 226         if (lock->l_export && lock->l_completion_ast)
 227                 lock->l_completion_ast(lock, 0);
 228
 229         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 230         LDLM_LOCK_PUT(lock);
 231         EXIT;
 232 }
 233
 234 /*
 235    usage: pass in a resource on which you have done get
 236           pass in a parent lock on which you have done a get
 237           do not put the resource or the parent
 238    returns: lock with refcount 1
 239 */
 240 static struct ldlm_lock *ldlm_lock_new(struct ldlm_lock *parent,
 241                                        struct ldlm_resource *resource)
 242 {
 243         struct ldlm_lock *lock;
 244         ENTRY;
 245
 246         if (resource == NULL)
 247                 LBUG();
 248
 249         lock = kmem_cache_alloc(ldlm_lock_slab, SLAB_KERNEL);
 250         if (lock == NULL)
 251                 RETURN(NULL);
 252
 253         memset(lock, 0, sizeof(*lock));
 254         get_random_bytes(&lock->l_random, sizeof(__u64));
 255
 256         lock->l_resource = resource;
 257         /* this refcount matches the one of the resource passed
 258            in which is not being put away */
 259         lock->l_refc = 1;
 260         INIT_LIST_HEAD(&lock->l_children);
 261         INIT_LIST_HEAD(&lock->l_res_link);
 262         INIT_LIST_HEAD(&lock->l_lru);
 263         INIT_LIST_HEAD(&lock->l_export_chain);
 264         INIT_LIST_HEAD(&lock->l_pending_chain);
 265         init_waitqueue_head(&lock->l_waitq);
 266
 267         spin_lock(&resource->lr_namespace->ns_counter_lock);
 268         resource->lr_namespace->ns_locks++;
 269         spin_unlock(&resource->lr_namespace->ns_counter_lock);
 270
 271         if (parent != NULL) {
 272                 l_lock(&parent->l_resource->lr_namespace->ns_lock);
 273                 lock->l_parent = parent;
 274                 list_add(&lock->l_childof, &parent->l_children);
 275                 l_unlock(&parent->l_resource->lr_namespace->ns_lock);
 276         }
 277
 278         CDEBUG(D_MALLOC, "kmalloced 'lock': %d at "
 279                "%p (tot %d).\n", sizeof(*lock), lock, 1);
 280         /* this is the extra refcount, to prevent the lock from evaporating */
 281         LDLM_LOCK_GET(lock);
 282         RETURN(lock);
 283 }
 284
 285 int ldlm_lock_change_resource(struct ldlm_lock *lock, __u64 new_resid[3])
 286 {
 287         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 288         struct ldlm_resource *oldres = lock->l_resource;
 289         int type, i;
 290         ENTRY;
 291
 292         l_lock(&ns->ns_lock);
 293         if (memcmp(new_resid, lock->l_resource->lr_name,
 294                    sizeof(lock->l_resource->lr_name)) == 0) {
 295                 /* Nothing to do */
 296                 l_unlock(&ns->ns_lock);
 297                 RETURN(0);
 298         }
 299
 300         type = lock->l_resource->lr_type;
 301         if (new_resid[0] == 0)
 302                 LBUG();
 303         lock->l_resource = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 304         if (lock->l_resource == NULL) {
 305                 LBUG();
 306                 RETURN(-ENOMEM);
 307         }
 308
 309         /* move references over */
 310         for (i = 0; i < lock->l_refc; i++) {
 311                 int rc;
 312                 ldlm_resource_getref(lock->l_resource);
 313                 rc = ldlm_resource_put(oldres);
 314                 if (rc == 1 && i != lock->l_refc - 1)
 315                         LBUG();
 316         }
 317         /* compensate for the initial get above.. */
 318         ldlm_resource_put(lock->l_resource);
 319
 320         l_unlock(&ns->ns_lock);
 321         RETURN(0);
 322 }
 323
 324 /*
 325  *  HANDLES
 326  */
 327
 328 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 329 {
 330         lockh->addr = (__u64) (unsigned long)lock;
 331         lockh->cookie = lock->l_random;
 332 }
 333
 334 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *handle,
 335                                      int strict)
 336 {
 337         struct ldlm_lock *lock = NULL, *retval = NULL;
 338         ENTRY;
 339
 340         if (!handle || !handle->addr)
 341                 RETURN(NULL);
 342
 343         //spin_lock(&ldlm_handle_lock);
 344         lock = (struct ldlm_lock *)(unsigned long)(handle->addr);
 345         if (!kmem_cache_validate(ldlm_lock_slab, (void *)lock)) {
 346                 CERROR("bogus lock %p\n", lock);
 347                 GOTO(out2, retval);
 348         }
 349
 350         if (lock->l_random != handle->cookie) {
 351                 CERROR("bogus cookie: lock %p has "LPX64" vs. handle "LPX64"\n",
 352                        lock, lock->l_random, handle->cookie);
 353                 GOTO(out2, NULL);
 354         }
 355         if (!lock->l_resource) {
 356                 CERROR("trying to lock bogus resource: lock %p\n", lock);
 357                 LDLM_DEBUG(lock, "ldlm_handle2lock(%p)", lock);
 358                 GOTO(out2, retval);
 359         }
 360         if (!lock->l_resource->lr_namespace) {
 361                 CERROR("trying to lock bogus namespace: lock %p\n", lock);
 362                 LDLM_DEBUG(lock, "ldlm_handle2lock(%p)", lock);
 363                 GOTO(out2, retval);
 364         }
 365
 366         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 367         if (strict && lock->l_flags & LDLM_FL_DESTROYED) {
 368                 CERROR("lock already destroyed: lock %p\n", lock);
 369                 LDLM_DEBUG(lock, "ldlm_handle2lock(%p)", lock);
 370                 GOTO(out, NULL);
 371         }
 372
 373         retval = LDLM_LOCK_GET(lock);
 374         if (!retval)
 375                 CERROR("lock disappeared below us!!! %p\n", lock);
 376         EXIT;
 377  out:
 378         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 379  out2:
 380         //spin_unlock(&ldlm_handle_lock);
 381         return retval;
 382 }
 383
 384 static int ldlm_plain_compat(struct ldlm_lock *a, struct ldlm_lock *b)
 385 {
 386         return lockmode_compat(a->l_req_mode, b->l_req_mode);
 387 }
 388
 389 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 390 {
 391         ldlm_res2desc(lock->l_resource, &desc->l_resource);
 392         desc->l_req_mode = lock->l_req_mode;
 393         desc->l_granted_mode = lock->l_granted_mode;
 394         memcpy(&desc->l_extent, &lock->l_extent, sizeof(desc->l_extent));
 395         memcpy(desc->l_version, lock->l_version, sizeof(desc->l_version));
 396 }
 397
 398 static void ldlm_add_ast_work_item(struct ldlm_lock *lock,
 399                                    struct ldlm_lock *new)
 400 {
 401         struct ldlm_ast_work *w;
 402         ENTRY;
 403
 404         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 405         if (new && (lock->l_flags & LDLM_FL_AST_SENT))
 406                 GOTO(out, 0);
 407
 408         OBD_ALLOC(w, sizeof(*w));
 409         if (!w) {
 410                 LBUG();
 411                 GOTO(out, 0);
 412         }
 413
 414         if (new) {
 415                 lock->l_flags |= LDLM_FL_AST_SENT;
 416                 w->w_blocking = 1;
 417                 ldlm_lock2desc(new, &w->w_desc);
 418         }
 419
 420         w->w_lock = LDLM_LOCK_GET(lock);
 421         list_add(&w->w_list, lock->l_resource->lr_tmp);
 422       out:
 423         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 424         return;
 425 }
 426
 427 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 428 {
 429         struct ldlm_lock *lock;
 430
 431         lock = ldlm_handle2lock(lockh);
 432         ldlm_lock_addref_internal(lock, mode);
 433         LDLM_LOCK_PUT(lock);
 434 }
 435
 436 /* only called for local locks */
 437 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 438 {
 439         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 440
 441         if (!list_empty(&lock->l_lru)) {
 442                 list_del_init(&lock->l_lru);
 443                 lock->l_resource->lr_namespace->ns_nr_unused--;
 444                 LASSERT(lock->l_resource->lr_namespace->ns_nr_unused >= 0);
 445         }
 446
 447         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 448                 lock->l_readers++;
 449         else
 450                 lock->l_writers++;
 451         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 452         LDLM_LOCK_GET(lock);
 453         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 454 }
 455
 456 /* Args: unlocked lock */
 457 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
 458                                     __u64 *res_id, int flags);
 459
 460 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 461 {
 462         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 463         struct ldlm_namespace *ns;
 464         ENTRY;
 465
 466         if (lock == NULL)
 467                 LBUG();
 468
 469         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 470         ns = lock->l_resource->lr_namespace;
 471         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 472         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 473                 lock->l_readers--;
 474         else
 475                 lock->l_writers--;
 476
 477         /* If we received a blocked AST and this was the last reference,
 478          * run the callback. */
 479         if (!lock->l_readers && !lock->l_writers &&
 480             (lock->l_flags & LDLM_FL_CBPENDING)) {
 481                 if (!lock->l_resource->lr_namespace->ns_client &&
 482                     lock->l_export)
 483                         CERROR("FL_CBPENDING set on non-local lock--just a "
 484                                "warning\n");
 485
 486                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 487                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 488
 489                 /* FIXME: need a real 'desc' here */
 490                 lock->l_blocking_ast(lock, NULL, lock->l_data,
 491                                      lock->l_data_len, LDLM_CB_BLOCKING);
 492         } else if (!lock->l_readers && !lock->l_writers) {
 493                 LASSERT(list_empty(&lock->l_lru));
 494                 LASSERT(ns->ns_nr_unused >= 0);
 495                 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 496                 ns->ns_nr_unused++;
 497                 if (ns->ns_client && ns->ns_nr_unused >= ns->ns_max_unused) {
 498                         CDEBUG(D_DLMTRACE, "%d unused (max %d), cancelling "
 499                                "LRU\n", ns->ns_nr_unused, ns->ns_max_unused);
 500                         ldlm_cli_cancel_unused_resource
 501                                 (ns, lock->l_resource->lr_name, LDLM_FL_REDUCE);
 502                 }
 503                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 504         } else
 505                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 506
 507         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 508         LDLM_LOCK_PUT(lock);    /* matches the handle2lock above */
 509
 510         EXIT;
 511 }
 512
 513 static int ldlm_lock_compat_list(struct ldlm_lock *lock, int send_cbs,
 514                                  struct list_head *queue)
 515 {
 516         struct list_head *tmp, *pos;
 517         int rc = 1;
 518
 519         list_for_each_safe(tmp, pos, queue) {
 520                 struct ldlm_lock *child;
 521                 ldlm_res_compat compat;
 522
 523                 child = list_entry(tmp, struct ldlm_lock, l_res_link);
 524                 if (lock == child)
 525                         continue;
 526
 527                 compat = ldlm_res_compat_table[child->l_resource->lr_type];
 528                 if (compat && compat(child, lock)) {
 529                         CDEBUG(D_OTHER, "compat function succeded, next.\n");
 530                         continue;
 531                 }
 532                 if (lockmode_compat(child->l_granted_mode, lock->l_req_mode)) {
 533                         CDEBUG(D_OTHER, "lock modes are compatible, next.\n");
 534                         continue;
 535                 }
 536
 537                 rc = 0;
 538
 539                 if (send_cbs && child->l_blocking_ast != NULL) {
 540                         CDEBUG(D_OTHER, "lock %p incompatible; sending "
 541                                "blocking AST.\n", child);
 542                         ldlm_add_ast_work_item(child, lock);
 543                 }
 544         }
 545
 546         return rc;
 547 }
 548
 549 static int ldlm_lock_compat(struct ldlm_lock *lock, int send_cbs)
 550 {
 551         int rc;
 552         ENTRY;
 553
 554         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 555         rc = ldlm_lock_compat_list(lock, send_cbs,
 556                                    &lock->l_resource->lr_granted);
 557         /* FIXME: should we be sending ASTs to converting? */
 558         if (rc)
 559                 rc = ldlm_lock_compat_list
 560                         (lock, send_cbs, &lock->l_resource->lr_converting);
 561
 562         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 563         RETURN(rc);
 564 }
 565
 566 /* NOTE: called by
 567    - ldlm_handle_enqueuque - resource
 568 */
 569 void ldlm_grant_lock(struct ldlm_lock *lock)
 570 {
 571         struct ldlm_resource *res = lock->l_resource;
 572         ENTRY;
 573
 574         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 575         ldlm_resource_add_lock(res, &res->lr_granted, lock);
 576         lock->l_granted_mode = lock->l_req_mode;
 577
 578         if (lock->l_granted_mode < res->lr_most_restr)
 579                 res->lr_most_restr = lock->l_granted_mode;
 580
 581         if (lock->l_completion_ast) {
 582                 ldlm_add_ast_work_item(lock, NULL);
 583         }
 584         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 585         EXIT;
 586 }
 587
 588 /* returns a referenced lock or NULL */
 589 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
 590                                       struct ldlm_extent *extent,
 591                                       struct ldlm_lock *old_lock)
 592 {
 593         struct ldlm_lock *lock;
 594         struct list_head *tmp;
 595
 596         list_for_each(tmp, queue) {
 597                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 598
 599                 if (lock == old_lock)
 600                         continue;
 601
 602                 if (lock->l_flags & (LDLM_FL_CBPENDING | LDLM_FL_DESTROYED))
 603                         continue;
 604
 605                 /* lock_convert() takes the resource lock, so we're sure that
 606                  * req_mode and lr_type won't change beneath us */
 607                 if (lock->l_req_mode != mode)
 608                         continue;
 609
 610                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 611                     (lock->l_extent.start > extent->start ||
 612                      lock->l_extent.end < extent->end))
 613                         continue;
 614
 615                 ldlm_lock_addref_internal(lock, mode);
 616                 return lock;
 617         }
 618
 619         return NULL;
 620 }
 621
 622 /* Can be called in two ways:
 623  *
 624  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 625  * for a duplicate of.
 626  *
 627  * Otherwise, all of the fields must be filled in, to match against.
 628  *
 629  * Returns 1 if it finds an already-existing lock that is compatible; in this
 630  * case, lockh is filled in with a addref()ed lock
 631  */
 632 int ldlm_lock_match(struct ldlm_namespace *ns, __u64 *res_id, __u32 type,
 633                     void *cookie, int cookielen, ldlm_mode_t mode,
 634                     struct lustre_handle *lockh)
 635 {
 636         struct ldlm_resource *res;
 637         struct ldlm_lock *lock, *old_lock = NULL;
 638         int rc = 0;
 639         ENTRY;
 640
 641         if (ns == NULL) {
 642                 old_lock = ldlm_handle2lock(lockh);
 643                 LASSERT(old_lock);
 644
 645                 ns = old_lock->l_resource->lr_namespace;
 646                 res_id = old_lock->l_resource->lr_name;
 647                 type = old_lock->l_resource->lr_type;
 648                 mode = old_lock->l_req_mode;
 649         }
 650
 651         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
 652         if (res == NULL) {
 653                 LASSERT(old_lock == NULL);
 654                 RETURN(0);
 655         }
 656
 657         ns = res->lr_namespace;
 658         l_lock(&ns->ns_lock);
 659
 660         if ((lock = search_queue(&res->lr_granted, mode, cookie, old_lock)))
 661                 GOTO(out, rc = 1);
 662         if ((lock = search_queue(&res->lr_converting, mode, cookie, old_lock)))
 663                 GOTO(out, rc = 1);
 664         if ((lock = search_queue(&res->lr_waiting, mode, cookie, old_lock)))
 665                 GOTO(out, rc = 1);
 666
 667         EXIT;
 668        out:
 669         ldlm_resource_put(res);
 670         l_unlock(&ns->ns_lock);
 671
 672         if (lock) {
 673                 ldlm_lock2handle(lock, lockh);
 674                 if (lock->l_completion_ast)
 675                         lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC);
 676         }
 677         if (rc)
 678                 LDLM_DEBUG(lock, "matched");
 679         else
 680                 LDLM_DEBUG_NOLOCK("not matched");
 681
 682         if (old_lock)
 683                 LDLM_LOCK_PUT(old_lock);
 684
 685         return rc;
 686 }
 687
 688 /* Returns a referenced lock */
 689 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
 690                                    struct lustre_handle *parent_lock_handle,
 691                                    __u64 * res_id, __u32 type,
 692                                    ldlm_mode_t mode, void *data, __u32 data_len)
 693 {
 694         struct ldlm_resource *res, *parent_res = NULL;
 695         struct ldlm_lock *lock, *parent_lock;
 696
 697         parent_lock = ldlm_handle2lock(parent_lock_handle);
 698         if (parent_lock)
 699                 parent_res = parent_lock->l_resource;
 700
 701         res = ldlm_resource_get(ns, parent_res, res_id, type, 1);
 702         if (res == NULL)
 703                 RETURN(NULL);
 704
 705         lock = ldlm_lock_new(parent_lock, res);
 706         if (lock == NULL) {
 707                 ldlm_resource_put(res);
 708                 RETURN(NULL);
 709         }
 710
 711         lock->l_req_mode = mode;
 712         lock->l_data = data;
 713         lock->l_data_len = data_len;
 714
 715         return lock;
 716 }
 717
 718 /* Must be called with lock->l_lock and lock->l_resource->lr_lock not held */
 719 ldlm_error_t ldlm_lock_enqueue(struct ldlm_lock * lock,
 720                                void *cookie, int cookie_len,
 721                                int *flags,
 722                                ldlm_completion_callback completion,
 723                                ldlm_blocking_callback blocking)
 724 {
 725         struct ldlm_resource *res;
 726         int local;
 727         ldlm_res_policy policy;
 728         ENTRY;
 729
 730         res = lock->l_resource;
 731         lock->l_blocking_ast = blocking;
 732
 733         if (res->lr_type == LDLM_EXTENT)
 734                 memcpy(&lock->l_extent, cookie, sizeof(lock->l_extent));
 735
 736         /* policies are not executed on the client or during replay */
 737         local = res->lr_namespace->ns_client;
 738         if (!local && !(*flags & LDLM_FL_REPLAY) &&
 739             (policy = ldlm_res_policy_table[res->lr_type])) {
 740                 int rc;
 741                 rc = policy(lock, cookie, lock->l_req_mode, *flags, NULL);
 742
 743                 if (rc == ELDLM_LOCK_CHANGED) {
 744                         res = lock->l_resource;
 745                         *flags |= LDLM_FL_LOCK_CHANGED;
 746                 } else if (rc == ELDLM_LOCK_ABORTED) {
 747                         ldlm_lock_destroy(lock);
 748                         RETURN(rc);
 749                 }
 750         }
 751
 752         l_lock(&res->lr_namespace->ns_lock);
 753         if (local && lock->l_req_mode == lock->l_granted_mode) {
 754                 /* The server returned a blocked lock, but it was granted before
 755                  * we got a chance to actually enqueue it.  We don't need to do
 756                  * anything else. */
 757                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
 758                           LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
 759                 GOTO(out, ELDLM_OK);
 760         }
 761
 762         /* This distinction between local lock trees is very important; a client
 763          * namespace only has information about locks taken by that client, and
 764          * thus doesn't have enough information to decide for itself if it can
 765          * be granted (below).  In this case, we do exactly what the server
 766          * tells us to do, as dictated by the 'flags'.
 767          *
 768          * We do exactly the same thing during recovery, when the server is
 769          * more or less trusting the clients not to lie.
 770          *
 771          * FIXME (bug 629283): Detect obvious lies by checking compatibility in
 772          * granted/converting queues. */
 773         ldlm_resource_unlink_lock(lock);
 774         if (local || (*flags & LDLM_FL_REPLAY)) {
 775                 if (*flags & LDLM_FL_BLOCK_CONV)
 776                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 777                                                lock);
 778                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
 779                         ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 780                 else
 781                         ldlm_grant_lock(lock);
 782                 GOTO(out, ELDLM_OK);
 783         }
 784
 785         /* FIXME: We may want to optimize by checking lr_most_restr */
 786         if (!list_empty(&res->lr_converting)) {
 787                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 788                 *flags |= LDLM_FL_BLOCK_CONV;
 789                 GOTO(out, ELDLM_OK);
 790         }
 791         if (!list_empty(&res->lr_waiting)) {
 792                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 793                 *flags |= LDLM_FL_BLOCK_WAIT;
 794                 GOTO(out, ELDLM_OK);
 795         }
 796         if (!ldlm_lock_compat(lock, 0)) {
 797                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 798                 *flags |= LDLM_FL_BLOCK_GRANTED;
 799                 GOTO(out, ELDLM_OK);
 800         }
 801
 802         ldlm_grant_lock(lock);
 803         EXIT;
 804       out:
 805         l_unlock(&res->lr_namespace->ns_lock);
 806         /* Don't set 'completion_ast' until here so that if the lock is granted
 807          * immediately we don't do an unnecessary completion call. */
 808         lock->l_completion_ast = completion;
 809         return ELDLM_OK;
 810 }
 811
 812 /* Must be called with namespace taken: queue is waiting or converting. */
 813 static int ldlm_reprocess_queue(struct ldlm_resource *res,
 814                                 struct list_head *queue)
 815 {
 816         struct list_head *tmp, *pos;
 817         ENTRY;
 818
 819         list_for_each_safe(tmp, pos, queue) {
 820                 struct ldlm_lock *pending;
 821                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
 822
 823                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
 824
 825                 if (!ldlm_lock_compat(pending, 1))
 826                         RETURN(1);
 827
 828                 list_del_init(&pending->l_res_link);
 829                 ldlm_grant_lock(pending);
 830         }
 831
 832         RETURN(0);
 833 }
 834
 835 void ldlm_run_ast_work(struct list_head *rpc_list)
 836 {
 837         struct list_head *tmp, *pos;
 838         int rc;
 839         ENTRY;
 840
 841         list_for_each_safe(tmp, pos, rpc_list) {
 842                 struct ldlm_ast_work *w =
 843                         list_entry(tmp, struct ldlm_ast_work, w_list);
 844
 845                 if (w->w_blocking)
 846                         rc = w->w_lock->l_blocking_ast
 847                                 (w->w_lock, &w->w_desc, w->w_data,
 848                                  w->w_datalen, LDLM_CB_BLOCKING);
 849                 else
 850                         rc = w->w_lock->l_completion_ast(w->w_lock, w->w_flags);
 851                 if (rc)
 852                         CERROR("Failed AST - should clean & disconnect "
 853                                "client\n");
 854                 LDLM_LOCK_PUT(w->w_lock);
 855                 list_del(&w->w_list);
 856                 OBD_FREE(w, sizeof(*w));
 857         }
 858         EXIT;
 859 }
 860
 861 /* Must be called with resource->lr_lock not taken. */
 862 void ldlm_reprocess_all(struct ldlm_resource *res)
 863 {
 864         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 865         ENTRY;
 866
 867         /* Local lock trees don't get reprocessed. */
 868         if (res->lr_namespace->ns_client) {
 869                 EXIT;
 870                 return;
 871         }
 872
 873         l_lock(&res->lr_namespace->ns_lock);
 874         res->lr_tmp = &rpc_list;
 875
 876         ldlm_reprocess_queue(res, &res->lr_converting);
 877         if (list_empty(&res->lr_converting))
 878                 ldlm_reprocess_queue(res, &res->lr_waiting);
 879
 880         res->lr_tmp = NULL;
 881         l_unlock(&res->lr_namespace->ns_lock);
 882
 883         ldlm_run_ast_work(&rpc_list);
 884         EXIT;
 885 }
 886
 887 void ldlm_cancel_callback(struct ldlm_lock *lock)
 888 {
 889         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 890         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
 891                 lock->l_flags |= LDLM_FL_CANCEL;
 892                 lock->l_blocking_ast(lock, NULL, lock->l_data,
 893                                      lock->l_data_len, LDLM_CB_CANCELING);
 894         }
 895         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 896 }
 897
 898 void ldlm_lock_cancel(struct ldlm_lock *lock)
 899 {
 900         struct ldlm_resource *res;
 901         struct ldlm_namespace *ns;
 902         ENTRY;
 903
 904         res = lock->l_resource;
 905         ns = res->lr_namespace;
 906
 907         l_lock(&ns->ns_lock);
 908         if (lock->l_readers || lock->l_writers)
 909                 CDEBUG(D_INFO, "lock still has references (%d readers, %d "
 910                        "writers)\n", lock->l_readers, lock->l_writers);
 911
 912         ldlm_cancel_callback(lock);
 913
 914         ldlm_del_waiting_lock(lock);
 915         ldlm_resource_unlink_lock(lock);
 916         ldlm_lock_destroy(lock);
 917         l_unlock(&ns->ns_lock);
 918         EXIT;
 919 }
 920
 921 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data, int datalen)
 922 {
 923         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 924         ENTRY;
 925
 926         if (lock == NULL)
 927                 RETURN(-EINVAL);
 928
 929         lock->l_data = data;
 930         lock->l_data_len = datalen;
 931
 932         LDLM_LOCK_PUT(lock);
 933
 934         RETURN(0);
 935 }
 936
 937 void ldlm_cancel_locks_for_export(struct obd_export *exp)
 938 {
 939         struct list_head *iter, *n; /* MUST BE CALLED "n"! */
 940
 941         list_for_each_safe(iter, n, &exp->exp_ldlm_data.led_held_locks) {
 942                 struct ldlm_lock *lock;
 943                 struct ldlm_resource *res;
 944                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
 945                 res = ldlm_resource_getref(lock->l_resource);
 946                 LDLM_DEBUG(lock, "export %p", exp);
 947                 ldlm_lock_cancel(lock);
 948                 ldlm_reprocess_all(res);
 949                 ldlm_resource_put(res);
 950         }
 951 }
 952
 953 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
 954                                         int *flags)
 955 {
 956         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 957         struct ldlm_resource *res;
 958         struct ldlm_namespace *ns;
 959         int granted = 0;
 960         ENTRY;
 961
 962         res = lock->l_resource;
 963         ns = res->lr_namespace;
 964
 965         l_lock(&ns->ns_lock);
 966
 967         lock->l_req_mode = new_mode;
 968         ldlm_resource_unlink_lock(lock);
 969
 970         /* If this is a local resource, put it on the appropriate list. */
 971         if (res->lr_namespace->ns_client) {
 972                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED))
 973                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 974                                                lock);
 975                 else {
 976                         /* This should never happen, because of the way the
 977                          * server handles conversions. */
 978                         LBUG();
 979
 980                         res->lr_tmp = &rpc_list;
 981                         ldlm_grant_lock(lock);
 982                         res->lr_tmp = NULL;
 983                         granted = 1;
 984                         /* FIXME: completion handling not with ns_lock held ! */
 985                         if (lock->l_completion_ast)
 986                                 lock->l_completion_ast(lock, 0);
 987                 }
 988         } else {
 989                 /* FIXME: We should try the conversion right away and possibly
 990                  * return success without the need for an extra AST */
 991                 ldlm_resource_add_lock(res, res->lr_converting.prev, lock);
 992                 *flags |= LDLM_FL_BLOCK_CONV;
 993         }
 994
 995         l_unlock(&ns->ns_lock);
 996
 997         if (granted)
 998                 ldlm_run_ast_work(&rpc_list);
 999         RETURN(res);
1000 }
1001
1002 void ldlm_lock_dump(struct ldlm_lock *lock)
1003 {
1004         char ver[128];
1005
1006         if (!(portal_debug & D_OTHER))
1007                 return;
1008
1009         if (RES_VERSION_SIZE != 4)
1010                 LBUG();
1011
1012         if (!lock) {
1013                 CDEBUG(D_OTHER, "  NULL LDLM lock\n");
1014                 return;
1015         }
1016
1017         snprintf(ver, sizeof(ver), "%x %x %x %x",
1018                  lock->l_version[0], lock->l_version[1],
1019                  lock->l_version[2], lock->l_version[3]);
1020
1021         CDEBUG(D_OTHER, "  -- Lock dump: %p (%s)\n", lock, ver);
1022         if (lock->l_export && lock->l_export->exp_connection)
1023                 CDEBUG(D_OTHER, "  Node: NID %x (rhandle: "LPX64")\n",
1024                        lock->l_export->exp_connection->c_peer.peer_nid,
1025                        lock->l_remote_handle.addr);
1026         else
1027                 CDEBUG(D_OTHER, "  Node: local\n");
1028         CDEBUG(D_OTHER, "  Parent: %p\n", lock->l_parent);
1029         CDEBUG(D_OTHER, "  Resource: %p ("LPD64")\n", lock->l_resource,
1030                lock->l_resource->lr_name[0]);
1031         CDEBUG(D_OTHER, "  Requested mode: %d, granted mode: %d\n",
1032                (int)lock->l_req_mode, (int)lock->l_granted_mode);
1033         CDEBUG(D_OTHER, "  Readers: %u ; Writers; %u\n",
1034                lock->l_readers, lock->l_writers);
1035         if (lock->l_resource->lr_type == LDLM_EXTENT)
1036                 CDEBUG(D_OTHER, "  Extent: %Lu -> %Lu\n",
1037                        (unsigned long long)lock->l_extent.start,
1038                        (unsigned long long)lock->l_extent.end);
1039 }