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/lustre_dlm.h>
  29 #include <linux/lustre_mds.h>
  30 #include <linux/obd_class.h>
  31
  32 //struct lustre_lock ldlm_everything_lock;
  33
  34 /* lock types */
  35 char *ldlm_lockname[] = {
  36         [0] "--",
  37         [LCK_EX] "EX",
  38         [LCK_PW] "PW",
  39         [LCK_PR] "PR",
  40         [LCK_CW] "CW",
  41         [LCK_CR] "CR",
  42         [LCK_NL] "NL"
  43 };
  44 char *ldlm_typename[] = {
  45         [LDLM_PLAIN] "PLN",
  46         [LDLM_EXTENT] "EXT",
  47 };
  48
  49 char *ldlm_it2str(int it)
  50 {
  51         switch (it) {
  52         case IT_OPEN:
  53                 return "open";
  54         case IT_CREAT:
  55                 return "creat";
  56         case (IT_OPEN | IT_CREAT):
  57                 return "open|creat";
  58         case IT_MKDIR:
  59                 return "mkdir";
  60         case IT_LINK:
  61                 return "link";
  62         case IT_LINK2:
  63                 return "link2";
  64         case IT_SYMLINK:
  65                 return "symlink";
  66         case IT_UNLINK:
  67                 return "unlink";
  68         case IT_RMDIR:
  69                 return "rmdir";
  70         case IT_RENAME:
  71                 return "rename";
  72         case IT_RENAME2:
  73                 return "rename2";
  74         case IT_READDIR:
  75                 return "readdir";
  76         case IT_GETATTR:
  77                 return "getattr";
  78         case IT_SETATTR:
  79                 return "setattr";
  80         case IT_READLINK:
  81                 return "readlink";
  82         case IT_MKNOD:
  83                 return "mknod";
  84         case IT_LOOKUP:
  85                 return "lookup";
  86         default:
  87                 CERROR("Unknown intent %d\n", it);
  88                 return "UNKNOWN";
  89         }
  90 }
  91
  92 extern kmem_cache_t *ldlm_lock_slab;
  93 struct lustre_lock ldlm_handle_lock;
  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_namespace *ns, struct ldlm_lock *lock,
 105                              void *req_cookie, ldlm_mode_t mode, int flags,
 106                              void *data)
 107 {
 108         if ((flags & LDLM_FL_HAS_INTENT) && ldlm_intent_policy_func) {
 109                 return ldlm_intent_policy_func(ns, lock, req_cookie, mode,
 110                                                flags, data);
 111         }
 112
 113         return ELDLM_OK;
 114 }
 115
 116 ldlm_res_policy ldlm_res_policy_table[] = {
 117         [LDLM_PLAIN] ldlm_plain_policy,
 118         [LDLM_EXTENT] ldlm_extent_policy,
 119 };
 120
 121 void ldlm_register_intent(ldlm_res_policy arg)
 122 {
 123         ldlm_intent_policy_func = arg;
 124 }
 125
 126 void ldlm_unregister_intent(void)
 127 {
 128         ldlm_intent_policy_func = NULL;
 129 }
 130
 131 /*
 132  * REFCOUNTED LOCK OBJECTS
 133  */
 134
 135
 136 /*
 137  * Lock refcounts, during creation:
 138  *   - one special one for allocation, dec'd only once in destroy
 139  *   - one for being a lock that's in-use
 140  *   - one for the addref associated with a new lock
 141  */
 142 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 143 {
 144         atomic_inc(&lock->l_refc);
 145         return lock;
 146 }
 147
 148 void ldlm_lock_put(struct ldlm_lock *lock)
 149 {
 150         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 151         ENTRY;
 152
 153         if (atomic_dec_and_test(&lock->l_refc)) {
 154                 l_lock(&ns->ns_lock);
 155                 LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing");
 156                 LASSERT(lock->l_destroyed);
 157                 LASSERT(list_empty(&lock->l_res_link));
 158
 159                 spin_lock(&ns->ns_counter_lock);
 160                 ns->ns_locks--;
 161                 spin_unlock(&ns->ns_counter_lock);
 162
 163                 ldlm_resource_putref(lock->l_resource);
 164                 lock->l_resource = NULL;
 165
 166                 if (lock->l_parent)
 167                         LDLM_LOCK_PUT(lock->l_parent);
 168
 169                 PORTAL_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
 170                 l_unlock(&ns->ns_lock);
 171         }
 172
 173         EXIT;
 174 }
 175
 176 void ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 177 {
 178         ENTRY;
 179         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 180         if (!list_empty(&lock->l_lru)) {
 181                 list_del_init(&lock->l_lru);
 182                 lock->l_resource->lr_namespace->ns_nr_unused--;
 183                 LASSERT(lock->l_resource->lr_namespace->ns_nr_unused >= 0);
 184         }
 185         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 186         EXIT;
 187 }
 188
 189 /* Only called with strict == 0 by recovery, to mark in-use locks as
 190  * should-be-destroyed */
 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(D_ERROR, lock);
 200                 LBUG();
 201         }
 202         if (lock->l_readers || lock->l_writers) {
 203                 LDLM_DEBUG(lock, "lock still has references");
 204                 ldlm_lock_dump(D_OTHER, lock);
 205         }
 206
 207         if (!list_empty(&lock->l_res_link)) {
 208                 ldlm_lock_dump(D_ERROR, lock);
 209                 LBUG();
 210         }
 211
 212         if (lock->l_destroyed) {
 213                 LASSERT(list_empty(&lock->l_lru));
 214                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 215                 EXIT;
 216                 return;
 217         }
 218         lock->l_destroyed = 1;
 219
 220         list_del_init(&lock->l_export_chain);
 221         ldlm_lock_remove_from_lru(lock);
 222         portals_handle_unhash(&lock->l_handle);
 223
 224 #if 0
 225         /* Wake anyone waiting for this lock */
 226         /* FIXME: I should probably add yet another flag, instead of using
 227          * l_export to only call this on clients */
 228         lock->l_export = NULL;
 229         if (lock->l_export && lock->l_completion_ast)
 230                 lock->l_completion_ast(lock, 0);
 231 #endif
 232
 233         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 234         LDLM_LOCK_PUT(lock);
 235         EXIT;
 236 }
 237
 238 /* this is called by portals_handle2object with the handle lock taken */
 239 static void lock_handle_addref(void *lock)
 240 {
 241         ldlm_lock_get(lock);
 242 }
 243
 244 /*
 245  * usage: pass in a resource on which you have done ldlm_resource_get
 246  *        pass in a parent lock on which you have done a ldlm_lock_get
 247  *        after return, ldlm_*_put the resource and parent
 248  * returns: lock with refcount 1
 249  */
 250 static struct ldlm_lock *ldlm_lock_new(struct ldlm_lock *parent,
 251                                        struct ldlm_resource *resource)
 252 {
 253         struct ldlm_lock *lock;
 254         ENTRY;
 255
 256         if (resource == NULL)
 257                 LBUG();
 258
 259         PORTAL_SLAB_ALLOC(lock, ldlm_lock_slab, sizeof(*lock));
 260         if (lock == NULL)
 261                 RETURN(NULL);
 262
 263         lock->l_resource = ldlm_resource_getref(resource);
 264
 265         atomic_set(&lock->l_refc, 2);
 266         INIT_LIST_HEAD(&lock->l_children);
 267         INIT_LIST_HEAD(&lock->l_res_link);
 268         INIT_LIST_HEAD(&lock->l_lru);
 269         INIT_LIST_HEAD(&lock->l_export_chain);
 270         INIT_LIST_HEAD(&lock->l_pending_chain);
 271         init_waitqueue_head(&lock->l_waitq);
 272
 273         spin_lock(&resource->lr_namespace->ns_counter_lock);
 274         resource->lr_namespace->ns_locks++;
 275         spin_unlock(&resource->lr_namespace->ns_counter_lock);
 276
 277         if (parent != NULL) {
 278                 l_lock(&parent->l_resource->lr_namespace->ns_lock);
 279                 lock->l_parent = LDLM_LOCK_GET(parent);
 280                 list_add(&lock->l_childof, &parent->l_children);
 281                 l_unlock(&parent->l_resource->lr_namespace->ns_lock);
 282         }
 283
 284         INIT_LIST_HEAD(&lock->l_handle.h_link);
 285         portals_handle_hash(&lock->l_handle, lock_handle_addref);
 286
 287         RETURN(lock);
 288 }
 289
 290 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 291                               __u64 new_resid[3])
 292 {
 293         struct ldlm_resource *oldres = lock->l_resource;
 294         ENTRY;
 295
 296         l_lock(&ns->ns_lock);
 297         if (memcmp(new_resid, lock->l_resource->lr_name,
 298                    sizeof(lock->l_resource->lr_name)) == 0) {
 299                 /* Nothing to do */
 300                 l_unlock(&ns->ns_lock);
 301                 RETURN(0);
 302         }
 303
 304         LASSERT(new_resid[0] != 0);
 305
 306         /* This function assumes that the lock isn't on any lists */
 307         LASSERT(list_empty(&lock->l_res_link));
 308
 309         lock->l_resource = ldlm_resource_get(ns, NULL, new_resid,
 310                                              lock->l_resource->lr_type, 1);
 311         if (lock->l_resource == NULL) {
 312                 LBUG();
 313                 RETURN(-ENOMEM);
 314         }
 315
 316         /* ...and the flowers are still standing! */
 317         ldlm_resource_putref(oldres);
 318
 319         l_unlock(&ns->ns_lock);
 320         RETURN(0);
 321 }
 322
 323 /*
 324  *  HANDLES
 325  */
 326
 327 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 328 {
 329         //lockh->addr = (__u64)(unsigned long)lock;
 330         memset(&lockh->addr, 0x69, sizeof(lockh->addr));
 331         lockh->cookie = lock->l_handle.h_cookie;
 332 }
 333
 334 /* if flags: atomically get the lock and set the flags.
 335  *           Return NULL if flag already set
 336  */
 337
 338 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *handle, int flags)
 339 {
 340         struct ldlm_lock *lock = NULL, *retval = NULL;
 341         ENTRY;
 342
 343         LASSERT(handle);
 344
 345         lock = portals_handle2object(handle->cookie);
 346         if (lock == NULL)
 347                 RETURN(NULL);
 348
 349         LASSERT(lock->l_resource != NULL);
 350         LASSERT(lock->l_resource->lr_namespace != NULL);
 351
 352         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 353
 354         /* It's unlikely but possible that someone marked the lock as
 355          * destroyed after we did handle2object on it */
 356         if (lock->l_destroyed) {
 357                 CERROR("lock already destroyed: lock %p\n", lock);
 358                 LDLM_LOCK_PUT(lock);
 359                 GOTO(out, retval);
 360         }
 361
 362         if (flags && (lock->l_flags & flags)) {
 363                 LDLM_LOCK_PUT(lock);
 364                 GOTO(out, retval);
 365         }
 366
 367         if (flags)
 368                 lock->l_flags |= flags;
 369
 370         retval = lock;
 371         EXIT;
 372  out:
 373         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 374         return retval;
 375 }
 376
 377 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
 378                                       struct lustre_handle *handle)
 379 {
 380         struct ldlm_lock *retval = NULL;
 381
 382         l_lock(&ns->ns_lock);
 383         retval = __ldlm_handle2lock(handle, 0);
 384         l_unlock(&ns->ns_lock);
 385
 386         return retval;
 387 }
 388
 389 static int ldlm_plain_compat(struct ldlm_lock *a, struct ldlm_lock *b)
 390 {
 391         return lockmode_compat(a->l_req_mode, b->l_req_mode);
 392 }
 393
 394 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 395 {
 396         ldlm_res2desc(lock->l_resource, &desc->l_resource);
 397         desc->l_req_mode = lock->l_req_mode;
 398         desc->l_granted_mode = lock->l_granted_mode;
 399         memcpy(&desc->l_extent, &lock->l_extent, sizeof(desc->l_extent));
 400         memcpy(desc->l_version, lock->l_version, sizeof(desc->l_version));
 401 }
 402
 403 static void ldlm_add_ast_work_item(struct ldlm_lock *lock,
 404                                    struct ldlm_lock *new)
 405 {
 406         struct ldlm_ast_work *w;
 407         ENTRY;
 408
 409         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 410         if (new && (lock->l_flags & LDLM_FL_AST_SENT))
 411                 GOTO(out, 0);
 412
 413         OBD_ALLOC(w, sizeof(*w));
 414         if (!w) {
 415                 LBUG();
 416                 GOTO(out, 0);
 417         }
 418
 419         if (new) {
 420                 lock->l_flags |= LDLM_FL_AST_SENT;
 421                 w->w_blocking = 1;
 422                 ldlm_lock2desc(new, &w->w_desc);
 423         }
 424
 425         w->w_lock = LDLM_LOCK_GET(lock);
 426         list_add(&w->w_list, lock->l_resource->lr_tmp);
 427       out:
 428         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 429         return;
 430 }
 431
 432 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 433 {
 434         struct ldlm_lock *lock;
 435
 436         lock = ldlm_handle2lock(lockh);
 437         ldlm_lock_addref_internal(lock, mode);
 438         LDLM_LOCK_PUT(lock);
 439 }
 440
 441 /* only called for local locks */
 442 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 443 {
 444         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 445         ldlm_lock_remove_from_lru(lock);
 446         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 447                 lock->l_readers++;
 448         else
 449                 lock->l_writers++;
 450         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 451         LDLM_LOCK_GET(lock);
 452         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 453 }
 454
 455 /* Args: unlocked lock */
 456 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
 457                                     __u64 *res_id, int flags);
 458
 459 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 460 {
 461         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 462         struct ldlm_namespace *ns;
 463         ENTRY;
 464
 465         if (lock == NULL)
 466                 LBUG();
 467
 468         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 469         ns = lock->l_resource->lr_namespace;
 470         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 471         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR) {
 472                 LASSERT(lock->l_readers > 0);
 473                 lock->l_readers--;
 474         } else {
 475                 LASSERT(lock->l_writers > 0);
 476                 lock->l_writers--;
 477         }
 478
 479         /* If we received a blocked AST and this was the last reference,
 480          * run the callback. */
 481         if (!lock->l_readers && !lock->l_writers &&
 482             (lock->l_flags & LDLM_FL_CBPENDING)) {
 483                 if (!lock->l_resource->lr_namespace->ns_client &&
 484                     lock->l_export)
 485                         CERROR("FL_CBPENDING set on non-local lock--just a "
 486                                "warning\n");
 487
 488                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 489                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 490
 491                 /* FIXME: need a real 'desc' here */
 492                 lock->l_blocking_ast(lock, NULL, lock->l_data,
 493                                      lock->l_data_len, LDLM_CB_BLOCKING);
 494         } else if (ns->ns_client && !lock->l_readers && !lock->l_writers) {
 495                 LASSERT(list_empty(&lock->l_lru));
 496                 LASSERT(ns->ns_nr_unused >= 0);
 497                 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 498                 ns->ns_nr_unused++;
 499                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 500                 ldlm_cancel_lru(ns);
 501         } else {
 502                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 503         }
 504
 505         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 506         LDLM_LOCK_PUT(lock);    /* matches the handle2lock above */
 507
 508         EXIT;
 509 }
 510
 511 static int ldlm_lock_compat_list(struct ldlm_lock *lock, int send_cbs,
 512                                  struct list_head *queue)
 513 {
 514         struct list_head *tmp, *pos;
 515         int rc = 1;
 516
 517         list_for_each_safe(tmp, pos, queue) {
 518                 struct ldlm_lock *child;
 519                 ldlm_res_compat compat;
 520
 521                 child = list_entry(tmp, struct ldlm_lock, l_res_link);
 522                 if (lock == child)
 523                         continue;
 524
 525                 compat = ldlm_res_compat_table[child->l_resource->lr_type];
 526                 if (compat && compat(child, lock)) {
 527                         CDEBUG(D_OTHER, "compat function succeded, next.\n");
 528                         continue;
 529                 }
 530                 if (lockmode_compat(child->l_granted_mode, lock->l_req_mode)) {
 531                         CDEBUG(D_OTHER, "lock modes are compatible, next.\n");
 532                         continue;
 533                 }
 534
 535                 rc = 0;
 536
 537                 if (send_cbs && child->l_blocking_ast != NULL) {
 538                         CDEBUG(D_OTHER, "lock %p incompatible; sending "
 539                                "blocking AST.\n", child);
 540                         ldlm_add_ast_work_item(child, lock);
 541                 }
 542         }
 543
 544         return rc;
 545 }
 546
 547 static int ldlm_lock_compat(struct ldlm_lock *lock, int send_cbs)
 548 {
 549         int rc;
 550         ENTRY;
 551
 552         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 553         rc = ldlm_lock_compat_list(lock, send_cbs,
 554                                    &lock->l_resource->lr_granted);
 555         /* FIXME: should we be sending ASTs to converting? */
 556         if (rc)
 557                 rc = ldlm_lock_compat_list
 558                         (lock, send_cbs, &lock->l_resource->lr_converting);
 559
 560         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 561         RETURN(rc);
 562 }
 563
 564 /* NOTE: called by
 565    - ldlm_handle_enqueuque - resource
 566 */
 567 void ldlm_grant_lock(struct ldlm_lock *lock)
 568 {
 569         struct ldlm_resource *res = lock->l_resource;
 570         ENTRY;
 571
 572         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 573         ldlm_resource_add_lock(res, &res->lr_granted, lock);
 574         lock->l_granted_mode = lock->l_req_mode;
 575
 576         if (lock->l_granted_mode < res->lr_most_restr)
 577                 res->lr_most_restr = lock->l_granted_mode;
 578
 579         if (lock->l_completion_ast) {
 580                 ldlm_add_ast_work_item(lock, NULL);
 581         }
 582         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 583         EXIT;
 584 }
 585
 586 /* returns a referenced lock or NULL */
 587 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
 588                                       struct ldlm_extent *extent,
 589                                       struct ldlm_lock *old_lock)
 590 {
 591         struct ldlm_lock *lock;
 592         struct list_head *tmp;
 593
 594         list_for_each(tmp, queue) {
 595                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 596
 597                 if (lock == old_lock)
 598                         continue;
 599
 600                 if (lock->l_flags & LDLM_FL_CBPENDING)
 601                         continue;
 602
 603                 if (lock->l_req_mode != mode)
 604                         continue;
 605
 606                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 607                     (lock->l_extent.start > extent->start ||
 608                      lock->l_extent.end < extent->end))
 609                         continue;
 610
 611                 if (lock->l_destroyed)
 612                         continue;
 613
 614                 ldlm_lock_addref_internal(lock, mode);
 615                 return lock;
 616         }
 617
 618         return NULL;
 619 }
 620
 621 /* Can be called in two ways:
 622  *
 623  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 624  * for a duplicate of.
 625  *
 626  * Otherwise, all of the fields must be filled in, to match against.
 627  *
 628  * Returns 1 if it finds an already-existing lock that is compatible; in this
 629  * case, lockh is filled in with a addref()ed lock
 630  */
 631 int ldlm_lock_match(struct ldlm_namespace *ns, __u64 *res_id, __u32 type,
 632                     void *cookie, int cookielen, ldlm_mode_t mode,
 633                     struct lustre_handle *lockh)
 634 {
 635         struct ldlm_resource *res;
 636         struct ldlm_lock *lock, *old_lock = NULL;
 637         int rc = 0;
 638         ENTRY;
 639
 640         if (ns == NULL) {
 641                 old_lock = ldlm_handle2lock(lockh);
 642                 LASSERT(old_lock);
 643
 644                 ns = old_lock->l_resource->lr_namespace;
 645                 res_id = old_lock->l_resource->lr_name;
 646                 type = old_lock->l_resource->lr_type;
 647                 mode = old_lock->l_req_mode;
 648         }
 649
 650         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
 651         if (res == NULL) {
 652                 LASSERT(old_lock == NULL);
 653                 RETURN(0);
 654         }
 655
 656         l_lock(&ns->ns_lock);
 657
 658         if ((lock = search_queue(&res->lr_granted, mode, cookie, old_lock)))
 659                 GOTO(out, rc = 1);
 660         if ((lock = search_queue(&res->lr_converting, mode, cookie, old_lock)))
 661                 GOTO(out, rc = 1);
 662         if ((lock = search_queue(&res->lr_waiting, mode, cookie, old_lock)))
 663                 GOTO(out, rc = 1);
 664
 665         EXIT;
 666        out:
 667         ldlm_resource_putref(res);
 668         l_unlock(&ns->ns_lock);
 669
 670         if (lock) {
 671                 ldlm_lock2handle(lock, lockh);
 672                 if (lock->l_completion_ast)
 673                         lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC);
 674         }
 675         if (rc)
 676                 LDLM_DEBUG(lock, "matched");
 677         else
 678                 LDLM_DEBUG_NOLOCK("not matched");
 679
 680         if (old_lock)
 681                 LDLM_LOCK_PUT(old_lock);
 682
 683         return rc;
 684 }
 685
 686 /* Returns a referenced lock */
 687 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
 688                                    struct lustre_handle *parent_lock_handle,
 689                                    __u64 * res_id, __u32 type,
 690                                    ldlm_mode_t mode, void *data, __u32 data_len)
 691 {
 692         struct ldlm_resource *res, *parent_res = NULL;
 693         struct ldlm_lock *lock, *parent_lock = NULL;
 694
 695         if (parent_lock_handle) {
 696                 parent_lock = ldlm_handle2lock(parent_lock_handle);
 697                 if (parent_lock)
 698                         parent_res = parent_lock->l_resource;
 699         }
 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         ldlm_resource_putref(res);
 707         if (parent_lock != NULL)
 708                 LDLM_LOCK_PUT(parent_lock);
 709
 710         if (lock == NULL)
 711                 RETURN(NULL);
 712
 713         lock->l_req_mode = mode;
 714         lock->l_data = data;
 715         lock->l_data_len = data_len;
 716
 717         return lock;
 718 }
 719
 720 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
 721                                struct ldlm_lock *lock,
 722                                void *cookie, int cookie_len,
 723                                int *flags,
 724                                ldlm_completion_callback completion,
 725                                ldlm_blocking_callback blocking)
 726 {
 727         struct ldlm_resource *res;
 728         int local;
 729         ldlm_res_policy policy;
 730         ENTRY;
 731
 732         res = lock->l_resource;
 733         lock->l_blocking_ast = blocking;
 734
 735         if (res->lr_type == LDLM_EXTENT)
 736                 memcpy(&lock->l_extent, cookie, sizeof(lock->l_extent));
 737
 738         /* policies are not executed on the client or during replay */
 739         local = res->lr_namespace->ns_client;
 740         if (!local && !(*flags & LDLM_FL_REPLAY) &&
 741             (policy = ldlm_res_policy_table[res->lr_type])) {
 742                 int rc;
 743                 rc = policy(ns, lock, cookie, lock->l_req_mode, *flags, NULL);
 744
 745                 if (rc == ELDLM_LOCK_CHANGED) {
 746                         res = lock->l_resource;
 747                         *flags |= LDLM_FL_LOCK_CHANGED;
 748                 } else if (rc == ELDLM_LOCK_ABORTED) {
 749                         ldlm_lock_destroy(lock);
 750                         RETURN(rc);
 751                 }
 752         }
 753
 754         l_lock(&ns->ns_lock);
 755         if (local && lock->l_req_mode == lock->l_granted_mode) {
 756                 /* The server returned a blocked lock, but it was granted before
 757                  * we got a chance to actually enqueue it.  We don't need to do
 758                  * anything else. */
 759                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
 760                           LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
 761                 GOTO(out, ELDLM_OK);
 762         }
 763
 764         /* This distinction between local lock trees is very important; a client
 765          * namespace only has information about locks taken by that client, and
 766          * thus doesn't have enough information to decide for itself if it can
 767          * be granted (below).  In this case, we do exactly what the server
 768          * tells us to do, as dictated by the 'flags'.
 769          *
 770          * We do exactly the same thing during recovery, when the server is
 771          * more or less trusting the clients not to lie.
 772          *
 773          * FIXME (bug 268): Detect obvious lies by checking compatibility in
 774          * granted/converting queues. */
 775         ldlm_resource_unlink_lock(lock);
 776         if (local) {
 777                 if (*flags & LDLM_FL_BLOCK_CONV)
 778                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 779                                                lock);
 780                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
 781                         ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 782                 else
 783                         ldlm_grant_lock(lock);
 784                 GOTO(out, ELDLM_OK);
 785         } else if (*flags & LDLM_FL_REPLAY) {
 786                 if (*flags & LDLM_FL_BLOCK_CONV) {
 787                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 788                                                lock);
 789                         GOTO(out, ELDLM_OK);
 790                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
 791                         ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 792                         GOTO(out, ELDLM_OK);
 793                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
 794                         ldlm_grant_lock(lock);
 795                         GOTO(out, ELDLM_OK);
 796                 }
 797                 /* If no flags, fall through to normal enqueue path. */
 798         }
 799
 800         /* FIXME: We may want to optimize by checking lr_most_restr */
 801         if (!list_empty(&res->lr_converting)) {
 802                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 803                 *flags |= LDLM_FL_BLOCK_CONV;
 804                 GOTO(out, ELDLM_OK);
 805         }
 806         if (!list_empty(&res->lr_waiting)) {
 807                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 808                 *flags |= LDLM_FL_BLOCK_WAIT;
 809                 GOTO(out, ELDLM_OK);
 810         }
 811         if (!ldlm_lock_compat(lock, 0)) {
 812                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 813                 *flags |= LDLM_FL_BLOCK_GRANTED;
 814                 GOTO(out, ELDLM_OK);
 815         }
 816
 817         ldlm_grant_lock(lock);
 818         EXIT;
 819       out:
 820         l_unlock(&ns->ns_lock);
 821         /* Don't set 'completion_ast' until here so that if the lock is granted
 822          * immediately we don't do an unnecessary completion call. */
 823         lock->l_completion_ast = completion;
 824         return ELDLM_OK;
 825 }
 826
 827 /* Must be called with namespace taken: queue is waiting or converting. */
 828 static int ldlm_reprocess_queue(struct ldlm_resource *res,
 829                                 struct list_head *queue)
 830 {
 831         struct list_head *tmp, *pos;
 832         ENTRY;
 833
 834         list_for_each_safe(tmp, pos, queue) {
 835                 struct ldlm_lock *pending;
 836                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
 837
 838                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
 839
 840                 if (!ldlm_lock_compat(pending, 1))
 841                         RETURN(1);
 842
 843                 list_del_init(&pending->l_res_link);
 844                 ldlm_grant_lock(pending);
 845         }
 846
 847         RETURN(0);
 848 }
 849
 850 int ldlm_run_ast_work(struct list_head *rpc_list)
 851 {
 852         struct list_head *tmp, *pos;
 853         int rc, retval = 0;
 854         ENTRY;
 855
 856         list_for_each_safe(tmp, pos, rpc_list) {
 857                 struct ldlm_ast_work *w =
 858                         list_entry(tmp, struct ldlm_ast_work, w_list);
 859
 860                 if (w->w_blocking)
 861                         rc = w->w_lock->l_blocking_ast
 862                                 (w->w_lock, &w->w_desc, w->w_data,
 863                                  w->w_datalen, LDLM_CB_BLOCKING);
 864                 else
 865                         rc = w->w_lock->l_completion_ast(w->w_lock, w->w_flags);
 866                 if (rc == -ERESTART)
 867                         retval = rc;
 868                 else if (rc)
 869                         CERROR("Failed AST - should clean & disconnect "
 870                                "client\n");
 871                 LDLM_LOCK_PUT(w->w_lock);
 872                 list_del(&w->w_list);
 873                 OBD_FREE(w, sizeof(*w));
 874         }
 875         RETURN(retval);
 876 }
 877
 878 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
 879 {
 880         ldlm_reprocess_all(res);
 881         return LDLM_ITER_CONTINUE;
 882 }
 883
 884 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
 885 {
 886         (void)ldlm_namespace_foreach_res(ns, reprocess_one_queue, NULL);
 887 }
 888
 889 /* Must be called with resource->lr_lock not taken. */
 890 void ldlm_reprocess_all(struct ldlm_resource *res)
 891 {
 892         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 893         int rc;
 894         ENTRY;
 895
 896         /* Local lock trees don't get reprocessed. */
 897         if (res->lr_namespace->ns_client) {
 898                 EXIT;
 899                 return;
 900         }
 901
 902  restart:
 903         l_lock(&res->lr_namespace->ns_lock);
 904         res->lr_tmp = &rpc_list;
 905
 906         ldlm_reprocess_queue(res, &res->lr_converting);
 907         if (list_empty(&res->lr_converting))
 908                 ldlm_reprocess_queue(res, &res->lr_waiting);
 909
 910         res->lr_tmp = NULL;
 911         l_unlock(&res->lr_namespace->ns_lock);
 912
 913         rc = ldlm_run_ast_work(&rpc_list);
 914         if (rc == -ERESTART)
 915                 goto restart;
 916         EXIT;
 917 }
 918
 919 void ldlm_cancel_callback(struct ldlm_lock *lock)
 920 {
 921         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 922         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
 923                 lock->l_flags |= LDLM_FL_CANCEL;
 924                 if (lock->l_blocking_ast)
 925                         lock->l_blocking_ast(lock, NULL, lock->l_data,
 926                                              lock->l_data_len,
 927                                              LDLM_CB_CANCELING);
 928                 else
 929                         LDLM_DEBUG(lock, "no blocking ast");
 930         }
 931         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 932 }
 933
 934 void ldlm_lock_cancel(struct ldlm_lock *lock)
 935 {
 936         struct ldlm_resource *res;
 937         struct ldlm_namespace *ns;
 938         ENTRY;
 939
 940         res = lock->l_resource;
 941         ns = res->lr_namespace;
 942
 943         l_lock(&ns->ns_lock);
 944         /* Please do not, no matter how tempting, remove this LBUG without
 945          * talking to me first. -phik */
 946         if (lock->l_readers || lock->l_writers) {
 947                 LDLM_DEBUG(lock, "lock still has references");
 948                 ldlm_lock_dump(D_OTHER, lock);
 949                 LBUG();
 950         }
 951
 952         ldlm_cancel_callback(lock);
 953
 954         ldlm_del_waiting_lock(lock);
 955         ldlm_resource_unlink_lock(lock);
 956         ldlm_lock_destroy(lock);
 957         l_unlock(&ns->ns_lock);
 958         EXIT;
 959 }
 960
 961 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data, int datalen)
 962 {
 963         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 964         ENTRY;
 965
 966         if (lock == NULL)
 967                 RETURN(-EINVAL);
 968
 969         lock->l_data = data;
 970         lock->l_data_len = datalen;
 971
 972         LDLM_LOCK_PUT(lock);
 973
 974         RETURN(0);
 975 }
 976
 977 void ldlm_cancel_locks_for_export(struct obd_export *exp)
 978 {
 979         struct list_head *iter, *n; /* MUST BE CALLED "n"! */
 980
 981         list_for_each_safe(iter, n, &exp->exp_ldlm_data.led_held_locks) {
 982                 struct ldlm_lock *lock;
 983                 struct ldlm_resource *res;
 984                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
 985                 res = ldlm_resource_getref(lock->l_resource);
 986                 LDLM_DEBUG(lock, "export %p", exp);
 987                 ldlm_lock_cancel(lock);
 988                 ldlm_reprocess_all(res);
 989                 ldlm_resource_putref(res);
 990         }
 991 }
 992
 993 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
 994                                         int *flags)
 995 {
 996         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 997         struct ldlm_resource *res;
 998         struct ldlm_namespace *ns;
 999         int granted = 0;
1000         ENTRY;
1001
1002         res = lock->l_resource;
1003         ns = res->lr_namespace;
1004
1005         l_lock(&ns->ns_lock);
1006
1007         lock->l_req_mode = new_mode;
1008         ldlm_resource_unlink_lock(lock);
1009
1010         /* If this is a local resource, put it on the appropriate list. */
1011         if (res->lr_namespace->ns_client) {
1012                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED))
1013                         ldlm_resource_add_lock(res, res->lr_converting.prev,
1014                                                lock);
1015                 else {
1016                         /* This should never happen, because of the way the
1017                          * server handles conversions. */
1018                         LBUG();
1019
1020                         res->lr_tmp = &rpc_list;
1021                         ldlm_grant_lock(lock);
1022                         res->lr_tmp = NULL;
1023                         granted = 1;
1024                         /* FIXME: completion handling not with ns_lock held ! */
1025                         if (lock->l_completion_ast)
1026                                 lock->l_completion_ast(lock, 0);
1027                 }
1028         } else {
1029                 /* FIXME: We should try the conversion right away and possibly
1030                  * return success without the need for an extra AST */
1031                 ldlm_resource_add_lock(res, res->lr_converting.prev, lock);
1032                 *flags |= LDLM_FL_BLOCK_CONV;
1033         }
1034
1035         l_unlock(&ns->ns_lock);
1036
1037         if (granted)
1038                 ldlm_run_ast_work(&rpc_list);
1039         RETURN(res);
1040 }
1041
1042 void ldlm_lock_dump(int level, struct ldlm_lock *lock)
1043 {
1044         char ver[128];
1045
1046         if (!(portal_debug & level))
1047                 return;
1048
1049         if (RES_VERSION_SIZE != 4)
1050                 LBUG();
1051
1052         if (!lock) {
1053                 CDEBUG(level, "  NULL LDLM lock\n");
1054                 return;
1055         }
1056
1057         snprintf(ver, sizeof(ver), "%x %x %x %x",
1058                  lock->l_version[0], lock->l_version[1],
1059                  lock->l_version[2], lock->l_version[3]);
1060
1061         CDEBUG(level, "  -- Lock dump: %p (%s)\n", lock, ver);
1062         if (lock->l_export && lock->l_export->exp_connection)
1063                 CDEBUG(level, "  Node: NID %x (rhandle: "LPX64")\n",
1064                        lock->l_export->exp_connection->c_peer.peer_nid,
1065                        lock->l_remote_handle.cookie);
1066         else
1067                 CDEBUG(level, "  Node: local\n");
1068         CDEBUG(level, "  Parent: %p\n", lock->l_parent);
1069         CDEBUG(level, "  Resource: %p ("LPD64")\n", lock->l_resource,
1070                lock->l_resource->lr_name[0]);
1071         CDEBUG(level, "  Requested mode: %d, granted mode: %d\n",
1072                (int)lock->l_req_mode, (int)lock->l_granted_mode);
1073         CDEBUG(level, "  Readers: %u ; Writers; %u\n",
1074                lock->l_readers, lock->l_writers);
1075         if (lock->l_resource->lr_type == LDLM_EXTENT)
1076                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64"\n",
1077                        lock->l_extent.start, lock->l_extent.end);
1078 }
1079
1080 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1081 {
1082         struct ldlm_lock *lock;
1083
1084         lock = ldlm_handle2lock(lockh);
1085         if (lock == NULL)
1086                 return;
1087
1088         ldlm_lock_dump(D_OTHER, lock);
1089
1090         LDLM_LOCK_PUT(lock);
1091 }