lustre/ldlm/ldlm_lock.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ldlm/ldlm_lock.c
  37  *
  38  * Author: Peter Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LDLM
  43
  44 #ifdef __KERNEL__
  45 # include <libcfs/libcfs.h>
  46 # include <linux/lustre_intent.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <obd_class.h>
  52 #include "ldlm_internal.h"
  53
  54 /* lock types */
  55 char *ldlm_lockname[] = {
  56         [0] "--",
  57         [LCK_EX] "EX",
  58         [LCK_PW] "PW",
  59         [LCK_PR] "PR",
  60         [LCK_CW] "CW",
  61         [LCK_CR] "CR",
  62         [LCK_NL] "NL",
  63         [LCK_GROUP] "GROUP",
  64         [LCK_COS] "COS"
  65 };
  66
  67 char *ldlm_typename[] = {
  68         [LDLM_PLAIN] "PLN",
  69         [LDLM_EXTENT] "EXT",
  70         [LDLM_FLOCK] "FLK",
  71         [LDLM_IBITS] "IBT",
  72 };
  73
  74 static ldlm_policy_wire_to_local_t ldlm_policy_wire18_to_local[] = {
  75         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local,
  76         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local,
  77         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire18_to_local,
  78         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local,
  79 };
  80
  81 static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = {
  82         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local,
  83         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local,
  84         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire21_to_local,
  85         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local,
  86 };
  87
  88 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
  89         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_local_to_wire,
  90         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_local_to_wire,
  91         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_local_to_wire,
  92         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_local_to_wire,
  93 };
  94
  95 /**
  96  * Converts lock policy from local format to on the wire lock_desc format
  97  */
  98 void ldlm_convert_policy_to_wire(ldlm_type_t type,
  99                                  const ldlm_policy_data_t *lpolicy,
 100                                  ldlm_wire_policy_data_t *wpolicy)
 101 {
 102         ldlm_policy_local_to_wire_t convert;
 103
 104         convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
 105
 106         convert(lpolicy, wpolicy);
 107 }
 108
 109 /**
 110  * Converts lock policy from on the wire lock_desc format to local format
 111  */
 112 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
 113                                   const ldlm_wire_policy_data_t *wpolicy,
 114                                   ldlm_policy_data_t *lpolicy)
 115 {
 116         ldlm_policy_wire_to_local_t convert;
 117         int new_client;
 118
 119         /** some badnes for 2.0.0 clients, but 2.0.0 isn't supported */
 120         new_client = (exp->exp_connect_flags & OBD_CONNECT_FULL20) != 0;
 121         if (new_client)
 122                convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE];
 123         else
 124                convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE];
 125
 126         convert(wpolicy, lpolicy);
 127 }
 128
 129 char *ldlm_it2str(int it)
 130 {
 131         switch (it) {
 132         case IT_OPEN:
 133                 return "open";
 134         case IT_CREAT:
 135                 return "creat";
 136         case (IT_OPEN | IT_CREAT):
 137                 return "open|creat";
 138         case IT_READDIR:
 139                 return "readdir";
 140         case IT_GETATTR:
 141                 return "getattr";
 142         case IT_LOOKUP:
 143                 return "lookup";
 144         case IT_UNLINK:
 145                 return "unlink";
 146         case IT_GETXATTR:
 147                 return "getxattr";
 148         case IT_LAYOUT:
 149                 return "layout";
 150         default:
 151                 CERROR("Unknown intent %d\n", it);
 152                 return "UNKNOWN";
 153         }
 154 }
 155
 156 extern cfs_mem_cache_t *ldlm_lock_slab;
 157
 158 #ifdef HAVE_SERVER_SUPPORT
 159 static ldlm_processing_policy ldlm_processing_policy_table[] = {
 160         [LDLM_PLAIN] ldlm_process_plain_lock,
 161         [LDLM_EXTENT] ldlm_process_extent_lock,
 162 # ifdef __KERNEL__
 163         [LDLM_FLOCK] ldlm_process_flock_lock,
 164 # endif
 165         [LDLM_IBITS] ldlm_process_inodebits_lock,
 166 };
 167
 168 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
 169 {
 170         return ldlm_processing_policy_table[res->lr_type];
 171 }
 172 #endif /* HAVE_SERVER_SUPPORT */
 173
 174 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
 175 {
 176         ns->ns_policy = arg;
 177 }
 178
 179 /*
 180  * REFCOUNTED LOCK OBJECTS
 181  */
 182
 183
 184 /*
 185  * Lock refcounts, during creation:
 186  *   - one special one for allocation, dec'd only once in destroy
 187  *   - one for being a lock that's in-use
 188  *   - one for the addref associated with a new lock
 189  */
 190 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 191 {
 192         cfs_atomic_inc(&lock->l_refc);
 193         return lock;
 194 }
 195
 196 void ldlm_lock_put(struct ldlm_lock *lock)
 197 {
 198         ENTRY;
 199
 200         LASSERT(lock->l_resource != LP_POISON);
 201         LASSERT(cfs_atomic_read(&lock->l_refc) > 0);
 202         if (cfs_atomic_dec_and_test(&lock->l_refc)) {
 203                 struct ldlm_resource *res;
 204
 205                 LDLM_DEBUG(lock,
 206                            "final lock_put on destroyed lock, freeing it.");
 207
 208                 res = lock->l_resource;
 209                 LASSERT(lock->l_destroyed);
 210                 LASSERT(cfs_list_empty(&lock->l_res_link));
 211                 LASSERT(cfs_list_empty(&lock->l_pending_chain));
 212
 213                 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
 214                                      LDLM_NSS_LOCKS);
 215                 lu_ref_del(&res->lr_reference, "lock", lock);
 216                 ldlm_resource_putref(res);
 217                 lock->l_resource = NULL;
 218                 if (lock->l_export) {
 219                         class_export_lock_put(lock->l_export, lock);
 220                         lock->l_export = NULL;
 221                 }
 222
 223                 if (lock->l_lvb_data != NULL)
 224                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
 225
 226                 ldlm_interval_free(ldlm_interval_detach(lock));
 227                 lu_ref_fini(&lock->l_reference);
 228                 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
 229         }
 230
 231         EXIT;
 232 }
 233
 234 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
 235 {
 236         int rc = 0;
 237         if (!cfs_list_empty(&lock->l_lru)) {
 238                 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 239
 240                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 241                 cfs_list_del_init(&lock->l_lru);
 242                 if (lock->l_flags & LDLM_FL_SKIPPED)
 243                         lock->l_flags &= ~LDLM_FL_SKIPPED;
 244                 LASSERT(ns->ns_nr_unused > 0);
 245                 ns->ns_nr_unused--;
 246                 rc = 1;
 247         }
 248         return rc;
 249 }
 250
 251 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 252 {
 253         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 254         int rc;
 255
 256         ENTRY;
 257         if (lock->l_ns_srv) {
 258                 LASSERT(cfs_list_empty(&lock->l_lru));
 259                 RETURN(0);
 260         }
 261
 262         cfs_spin_lock(&ns->ns_lock);
 263         rc = ldlm_lock_remove_from_lru_nolock(lock);
 264         cfs_spin_unlock(&ns->ns_lock);
 265         EXIT;
 266         return rc;
 267 }
 268
 269 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
 270 {
 271         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 272
 273         lock->l_last_used = cfs_time_current();
 274         LASSERT(cfs_list_empty(&lock->l_lru));
 275         LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 276         cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 277         LASSERT(ns->ns_nr_unused >= 0);
 278         ns->ns_nr_unused++;
 279 }
 280
 281 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
 282 {
 283         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 284
 285         ENTRY;
 286         cfs_spin_lock(&ns->ns_lock);
 287         ldlm_lock_add_to_lru_nolock(lock);
 288         cfs_spin_unlock(&ns->ns_lock);
 289         EXIT;
 290 }
 291
 292 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
 293 {
 294         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 295
 296         ENTRY;
 297         if (lock->l_ns_srv) {
 298                 LASSERT(cfs_list_empty(&lock->l_lru));
 299                 EXIT;
 300                 return;
 301         }
 302
 303         cfs_spin_lock(&ns->ns_lock);
 304         if (!cfs_list_empty(&lock->l_lru)) {
 305                 ldlm_lock_remove_from_lru_nolock(lock);
 306                 ldlm_lock_add_to_lru_nolock(lock);
 307         }
 308         cfs_spin_unlock(&ns->ns_lock);
 309         EXIT;
 310 }
 311
 312 /* This used to have a 'strict' flag, which recovery would use to mark an
 313  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 314  * shall explain why it's gone: with the new hash table scheme, once you call
 315  * ldlm_lock_destroy, you can never drop your final references on this lock.
 316  * Because it's not in the hash table anymore.  -phil */
 317 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
 318 {
 319         ENTRY;
 320
 321         if (lock->l_readers || lock->l_writers) {
 322                 LDLM_ERROR(lock, "lock still has references");
 323                 LBUG();
 324         }
 325
 326         if (!cfs_list_empty(&lock->l_res_link)) {
 327                 LDLM_ERROR(lock, "lock still on resource");
 328                 LBUG();
 329         }
 330
 331         if (lock->l_destroyed) {
 332                 LASSERT(cfs_list_empty(&lock->l_lru));
 333                 EXIT;
 334                 return 0;
 335         }
 336         lock->l_destroyed = 1;
 337
 338         if (lock->l_export && lock->l_export->exp_lock_hash &&
 339             !cfs_hlist_unhashed(&lock->l_exp_hash))
 340                 cfs_hash_del(lock->l_export->exp_lock_hash,
 341                              &lock->l_remote_handle, &lock->l_exp_hash);
 342
 343         ldlm_lock_remove_from_lru(lock);
 344         class_handle_unhash(&lock->l_handle);
 345
 346 #if 0
 347         /* Wake anyone waiting for this lock */
 348         /* FIXME: I should probably add yet another flag, instead of using
 349          * l_export to only call this on clients */
 350         if (lock->l_export)
 351                 class_export_put(lock->l_export);
 352         lock->l_export = NULL;
 353         if (lock->l_export && lock->l_completion_ast)
 354                 lock->l_completion_ast(lock, 0);
 355 #endif
 356         EXIT;
 357         return 1;
 358 }
 359
 360 void ldlm_lock_destroy(struct ldlm_lock *lock)
 361 {
 362         int first;
 363         ENTRY;
 364         lock_res_and_lock(lock);
 365         first = ldlm_lock_destroy_internal(lock);
 366         unlock_res_and_lock(lock);
 367
 368         /* drop reference from hashtable only for first destroy */
 369         if (first) {
 370                 lu_ref_del(&lock->l_reference, "hash", lock);
 371                 LDLM_LOCK_RELEASE(lock);
 372         }
 373         EXIT;
 374 }
 375
 376 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
 377 {
 378         int first;
 379         ENTRY;
 380         first = ldlm_lock_destroy_internal(lock);
 381         /* drop reference from hashtable only for first destroy */
 382         if (first) {
 383                 lu_ref_del(&lock->l_reference, "hash", lock);
 384                 LDLM_LOCK_RELEASE(lock);
 385         }
 386         EXIT;
 387 }
 388
 389 /* this is called by portals_handle2object with the handle lock taken */
 390 static void lock_handle_addref(void *lock)
 391 {
 392         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 393 }
 394
 395 static void lock_handle_free(void *lock, int size)
 396 {
 397         LASSERT(size == sizeof(struct ldlm_lock));
 398         OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
 399 }
 400
 401 struct portals_handle_ops lock_handle_ops = {
 402         .hop_addref = lock_handle_addref,
 403         .hop_free   = lock_handle_free,
 404 };
 405
 406 /*
 407  * usage: pass in a resource on which you have done ldlm_resource_get
 408  *        new lock will take over the refcount.
 409  * returns: lock with refcount 2 - one for current caller and one for remote
 410  */
 411 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
 412 {
 413         struct ldlm_lock *lock;
 414         ENTRY;
 415
 416         if (resource == NULL)
 417                 LBUG();
 418
 419         OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO);
 420         if (lock == NULL)
 421                 RETURN(NULL);
 422
 423         cfs_spin_lock_init(&lock->l_lock);
 424         lock->l_resource = resource;
 425         lu_ref_add(&resource->lr_reference, "lock", lock);
 426
 427         cfs_atomic_set(&lock->l_refc, 2);
 428         CFS_INIT_LIST_HEAD(&lock->l_res_link);
 429         CFS_INIT_LIST_HEAD(&lock->l_lru);
 430         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
 431         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
 432         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
 433         CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
 434         cfs_waitq_init(&lock->l_waitq);
 435         lock->l_blocking_lock = NULL;
 436         CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
 437         CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
 438         CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
 439
 440         lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
 441                              LDLM_NSS_LOCKS);
 442         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
 443         class_handle_hash(&lock->l_handle, &lock_handle_ops);
 444
 445         lu_ref_init(&lock->l_reference);
 446         lu_ref_add(&lock->l_reference, "hash", lock);
 447         lock->l_callback_timeout = 0;
 448
 449 #if LUSTRE_TRACKS_LOCK_EXP_REFS
 450         CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
 451         lock->l_exp_refs_nr = 0;
 452         lock->l_exp_refs_target = NULL;
 453 #endif
 454         CFS_INIT_LIST_HEAD(&lock->l_exp_list);
 455
 456         RETURN(lock);
 457 }
 458
 459 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 460                               const struct ldlm_res_id *new_resid)
 461 {
 462         struct ldlm_resource *oldres = lock->l_resource;
 463         struct ldlm_resource *newres;
 464         int type;
 465         ENTRY;
 466
 467         LASSERT(ns_is_client(ns));
 468
 469         lock_res_and_lock(lock);
 470         if (memcmp(new_resid, &lock->l_resource->lr_name,
 471                    sizeof(lock->l_resource->lr_name)) == 0) {
 472                 /* Nothing to do */
 473                 unlock_res_and_lock(lock);
 474                 RETURN(0);
 475         }
 476
 477         LASSERT(new_resid->name[0] != 0);
 478
 479         /* This function assumes that the lock isn't on any lists */
 480         LASSERT(cfs_list_empty(&lock->l_res_link));
 481
 482         type = oldres->lr_type;
 483         unlock_res_and_lock(lock);
 484
 485         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 486         if (newres == NULL)
 487                 RETURN(-ENOMEM);
 488
 489         lu_ref_add(&newres->lr_reference, "lock", lock);
 490         /*
 491          * To flip the lock from the old to the new resource, lock, oldres and
 492          * newres have to be locked. Resource spin-locks are nested within
 493          * lock->l_lock, and are taken in the memory address order to avoid
 494          * dead-locks.
 495          */
 496         cfs_spin_lock(&lock->l_lock);
 497         oldres = lock->l_resource;
 498         if (oldres < newres) {
 499                 lock_res(oldres);
 500                 lock_res_nested(newres, LRT_NEW);
 501         } else {
 502                 lock_res(newres);
 503                 lock_res_nested(oldres, LRT_NEW);
 504         }
 505         LASSERT(memcmp(new_resid, &oldres->lr_name,
 506                        sizeof oldres->lr_name) != 0);
 507         lock->l_resource = newres;
 508         unlock_res(oldres);
 509         unlock_res_and_lock(lock);
 510
 511         /* ...and the flowers are still standing! */
 512         lu_ref_del(&oldres->lr_reference, "lock", lock);
 513         ldlm_resource_putref(oldres);
 514
 515         RETURN(0);
 516 }
 517
 518 /*
 519  *  HANDLES
 520  */
 521
 522 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
 523 {
 524         lockh->cookie = lock->l_handle.h_cookie;
 525 }
 526
 527 /* if flags: atomically get the lock and set the flags.
 528  *           Return NULL if flag already set
 529  */
 530
 531 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
 532                                      int flags)
 533 {
 534         struct ldlm_lock *lock;
 535         ENTRY;
 536
 537         LASSERT(handle);
 538
 539         lock = class_handle2object(handle->cookie);
 540         if (lock == NULL)
 541                 RETURN(NULL);
 542
 543         /* It's unlikely but possible that someone marked the lock as
 544          * destroyed after we did handle2object on it */
 545         if (flags == 0 && !lock->l_destroyed) {
 546                 lu_ref_add(&lock->l_reference, "handle", cfs_current());
 547                 RETURN(lock);
 548         }
 549
 550         lock_res_and_lock(lock);
 551
 552         LASSERT(lock->l_resource != NULL);
 553
 554         lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
 555         if (unlikely(lock->l_destroyed)) {
 556                 unlock_res_and_lock(lock);
 557                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 558                 LDLM_LOCK_PUT(lock);
 559                 RETURN(NULL);
 560         }
 561
 562         if (flags && (lock->l_flags & flags)) {
 563                 unlock_res_and_lock(lock);
 564                 LDLM_LOCK_PUT(lock);
 565                 RETURN(NULL);
 566         }
 567
 568         if (flags)
 569                 lock->l_flags |= flags;
 570
 571         unlock_res_and_lock(lock);
 572         RETURN(lock);
 573 }
 574
 575 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 576 {
 577         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
 578         /* INODEBITS_INTEROP: If the other side does not support
 579          * inodebits, reply with a plain lock descriptor.
 580          */
 581         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
 582             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
 583                 /* Make sure all the right bits are set in this lock we
 584                    are going to pass to client */
 585                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
 586                          (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 587                           MDS_INODELOCK_LAYOUT),
 588                          "Inappropriate inode lock bits during "
 589                          "conversion " LPU64 "\n",
 590                          lock->l_policy_data.l_inodebits.bits);
 591
 592                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 593                 desc->l_resource.lr_type = LDLM_PLAIN;
 594
 595                 /* Convert "new" lock mode to something old client can
 596                    understand */
 597                 if ((lock->l_req_mode == LCK_CR) ||
 598                     (lock->l_req_mode == LCK_CW))
 599                         desc->l_req_mode = LCK_PR;
 600                 else
 601                         desc->l_req_mode = lock->l_req_mode;
 602                 if ((lock->l_granted_mode == LCK_CR) ||
 603                     (lock->l_granted_mode == LCK_CW)) {
 604                         desc->l_granted_mode = LCK_PR;
 605                 } else {
 606                         /* We never grant PW/EX locks to clients */
 607                         LASSERT((lock->l_granted_mode != LCK_PW) &&
 608                                 (lock->l_granted_mode != LCK_EX));
 609                         desc->l_granted_mode = lock->l_granted_mode;
 610                 }
 611
 612                 /* We do not copy policy here, because there is no
 613                    policy for plain locks */
 614         } else {
 615                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 616                 desc->l_req_mode = lock->l_req_mode;
 617                 desc->l_granted_mode = lock->l_granted_mode;
 618                 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
 619                                             &lock->l_policy_data,
 620                                             &desc->l_policy_data);
 621         }
 622 }
 623
 624 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 625                            cfs_list_t *work_list)
 626 {
 627         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
 628                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 629                 lock->l_flags |= LDLM_FL_AST_SENT;
 630                 /* If the enqueuing client said so, tell the AST recipient to
 631                  * discard dirty data, rather than writing back. */
 632                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 633                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 634                 LASSERT(cfs_list_empty(&lock->l_bl_ast));
 635                 cfs_list_add(&lock->l_bl_ast, work_list);
 636                 LDLM_LOCK_GET(lock);
 637                 LASSERT(lock->l_blocking_lock == NULL);
 638                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
 639         }
 640 }
 641
 642 void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list)
 643 {
 644         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
 645                 lock->l_flags |= LDLM_FL_CP_REQD;
 646                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
 647                 LASSERT(cfs_list_empty(&lock->l_cp_ast));
 648                 cfs_list_add(&lock->l_cp_ast, work_list);
 649                 LDLM_LOCK_GET(lock);
 650         }
 651 }
 652
 653 /* must be called with lr_lock held */
 654 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 655                             cfs_list_t *work_list)
 656 {
 657         ENTRY;
 658         check_res_locked(lock->l_resource);
 659         if (new)
 660                 ldlm_add_bl_work_item(lock, new, work_list);
 661         else
 662                 ldlm_add_cp_work_item(lock, work_list);
 663         EXIT;
 664 }
 665
 666 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 667 {
 668         struct ldlm_lock *lock;
 669
 670         lock = ldlm_handle2lock(lockh);
 671         LASSERT(lock != NULL);
 672         ldlm_lock_addref_internal(lock, mode);
 673         LDLM_LOCK_PUT(lock);
 674 }
 675
 676 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 677 {
 678         ldlm_lock_remove_from_lru(lock);
 679         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 680                 lock->l_readers++;
 681                 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
 682         }
 683         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
 684                 lock->l_writers++;
 685                 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
 686         }
 687         LDLM_LOCK_GET(lock);
 688         lu_ref_add_atomic(&lock->l_reference, "user", lock);
 689         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 690 }
 691
 692 /**
 693  * Attempts to addref a lock, and fails if lock is already LDLM_FL_CBPENDING
 694  * or destroyed.
 695  *
 696  * \retval 0 success, lock was addref-ed
 697  *
 698  * \retval -EAGAIN lock is being canceled.
 699  */
 700 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
 701 {
 702         struct ldlm_lock *lock;
 703         int               result;
 704
 705         result = -EAGAIN;
 706         lock = ldlm_handle2lock(lockh);
 707         if (lock != NULL) {
 708                 lock_res_and_lock(lock);
 709                 if (lock->l_readers != 0 || lock->l_writers != 0 ||
 710                     !(lock->l_flags & LDLM_FL_CBPENDING)) {
 711                         ldlm_lock_addref_internal_nolock(lock, mode);
 712                         result = 0;
 713                 }
 714                 unlock_res_and_lock(lock);
 715                 LDLM_LOCK_PUT(lock);
 716         }
 717         return result;
 718 }
 719
 720 /* only called for local locks */
 721 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 722 {
 723         lock_res_and_lock(lock);
 724         ldlm_lock_addref_internal_nolock(lock, mode);
 725         unlock_res_and_lock(lock);
 726 }
 727
 728 /* only called in ldlm_flock_destroy and for local locks.
 729  *  * for LDLM_FLOCK type locks, l_blocking_ast is null, and
 730  *   * ldlm_lock_remove_from_lru() does nothing, it is safe
 731  *    * for ldlm_flock_destroy usage by dropping some code */
 732 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 733 {
 734         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 735         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 736                 LASSERT(lock->l_readers > 0);
 737                 lu_ref_del(&lock->l_reference, "reader", lock);
 738                 lock->l_readers--;
 739         }
 740         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
 741                 LASSERT(lock->l_writers > 0);
 742                 lu_ref_del(&lock->l_reference, "writer", lock);
 743                 lock->l_writers--;
 744         }
 745
 746         lu_ref_del(&lock->l_reference, "user", lock);
 747         LDLM_LOCK_RELEASE(lock);    /* matches the LDLM_LOCK_GET() in addref */
 748 }
 749
 750 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 751 {
 752         struct ldlm_namespace *ns;
 753         ENTRY;
 754
 755         lock_res_and_lock(lock);
 756
 757         ns = ldlm_lock_to_ns(lock);
 758
 759         ldlm_lock_decref_internal_nolock(lock, mode);
 760
 761         if (lock->l_flags & LDLM_FL_LOCAL &&
 762             !lock->l_readers && !lock->l_writers) {
 763                 /* If this is a local lock on a server namespace and this was
 764                  * the last reference, cancel the lock. */
 765                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 766                 lock->l_flags |= LDLM_FL_CBPENDING;
 767         }
 768
 769         if (!lock->l_readers && !lock->l_writers &&
 770             (lock->l_flags & LDLM_FL_CBPENDING)) {
 771                 /* If we received a blocked AST and this was the last reference,
 772                  * run the callback. */
 773                 if (lock->l_ns_srv && lock->l_export)
 774                         CERROR("FL_CBPENDING set on non-local lock--just a "
 775                                "warning\n");
 776
 777                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 778
 779                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
 780                 ldlm_lock_remove_from_lru(lock);
 781                 unlock_res_and_lock(lock);
 782
 783                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
 784                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
 785
 786                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
 787                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
 788                         ldlm_handle_bl_callback(ns, NULL, lock);
 789         } else if (ns_is_client(ns) &&
 790                    !lock->l_readers && !lock->l_writers &&
 791                    !(lock->l_flags & LDLM_FL_NO_LRU) &&
 792                    !(lock->l_flags & LDLM_FL_BL_AST)) {
 793
 794                 LDLM_DEBUG(lock, "add lock into lru list");
 795
 796                 /* If this is a client-side namespace and this was the last
 797                  * reference, put it on the LRU. */
 798                 ldlm_lock_add_to_lru(lock);
 799                 unlock_res_and_lock(lock);
 800
 801                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
 802                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
 803
 804                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
 805                  * are not supported by the server, otherwise, it is done on
 806                  * enqueue. */
 807                 if (!exp_connect_cancelset(lock->l_conn_export) &&
 808                     !ns_connect_lru_resize(ns))
 809                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
 810         } else {
 811                 LDLM_DEBUG(lock, "do not add lock into lru list");
 812                 unlock_res_and_lock(lock);
 813         }
 814
 815         EXIT;
 816 }
 817
 818 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 819 {
 820         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 821         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
 822         ldlm_lock_decref_internal(lock, mode);
 823         LDLM_LOCK_PUT(lock);
 824 }
 825
 826 /* This will drop a lock reference and mark it for destruction, but will not
 827  * necessarily cancel the lock before returning. */
 828 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 829 {
 830         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 831         ENTRY;
 832
 833         LASSERT(lock != NULL);
 834
 835         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 836         lock_res_and_lock(lock);
 837         lock->l_flags |= LDLM_FL_CBPENDING;
 838         unlock_res_and_lock(lock);
 839         ldlm_lock_decref_internal(lock, mode);
 840         LDLM_LOCK_PUT(lock);
 841 }
 842
 843 struct sl_insert_point {
 844         cfs_list_t *res_link;
 845         cfs_list_t *mode_link;
 846         cfs_list_t *policy_link;
 847 };
 848
 849 /*
 850  * search_granted_lock
 851  *
 852  * Description:
 853  *      Finds a position to insert the new lock.
 854  * Parameters:
 855  *      queue [input]:  the granted list where search acts on;
 856  *      req [input]:    the lock whose position to be located;
 857  *      prev [output]:  positions within 3 lists to insert @req to
 858  * Return Value:
 859  *      filled @prev
 860  * NOTE: called by
 861  *  - ldlm_grant_lock_with_skiplist
 862  */
 863 static void search_granted_lock(cfs_list_t *queue,
 864                                 struct ldlm_lock *req,
 865                                 struct sl_insert_point *prev)
 866 {
 867         cfs_list_t *tmp;
 868         struct ldlm_lock *lock, *mode_end, *policy_end;
 869         ENTRY;
 870
 871         cfs_list_for_each(tmp, queue) {
 872                 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
 873
 874                 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
 875                                           struct ldlm_lock, l_sl_mode);
 876
 877                 if (lock->l_req_mode != req->l_req_mode) {
 878                         /* jump to last lock of mode group */
 879                         tmp = &mode_end->l_res_link;
 880                         continue;
 881                 }
 882
 883                 /* suitable mode group is found */
 884                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
 885                         /* insert point is last lock of the mode group */
 886                         prev->res_link = &mode_end->l_res_link;
 887                         prev->mode_link = &mode_end->l_sl_mode;
 888                         prev->policy_link = &req->l_sl_policy;
 889                         EXIT;
 890                         return;
 891                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
 892                         for (;;) {
 893                                 policy_end =
 894                                         cfs_list_entry(lock->l_sl_policy.prev,
 895                                                        struct ldlm_lock,
 896                                                        l_sl_policy);
 897
 898                                 if (lock->l_policy_data.l_inodebits.bits ==
 899                                     req->l_policy_data.l_inodebits.bits) {
 900                                         /* insert point is last lock of
 901                                          * the policy group */
 902                                         prev->res_link =
 903                                                 &policy_end->l_res_link;
 904                                         prev->mode_link =
 905                                                 &policy_end->l_sl_mode;
 906                                         prev->policy_link =
 907                                                 &policy_end->l_sl_policy;
 908                                         EXIT;
 909                                         return;
 910                                 }
 911
 912                                 if (policy_end == mode_end)
 913                                         /* done with mode group */
 914                                         break;
 915
 916                                 /* go to next policy group within mode group */
 917                                 tmp = policy_end->l_res_link.next;
 918                                 lock = cfs_list_entry(tmp, struct ldlm_lock,
 919                                                       l_res_link);
 920                         }  /* loop over policy groups within the mode group */
 921
 922                         /* insert point is last lock of the mode group,
 923                          * new policy group is started */
 924                         prev->res_link = &mode_end->l_res_link;
 925                         prev->mode_link = &mode_end->l_sl_mode;
 926                         prev->policy_link = &req->l_sl_policy;
 927                         EXIT;
 928                         return;
 929                 } else {
 930                         LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
 931                         LBUG();
 932                 }
 933         }
 934
 935         /* insert point is last lock on the queue,
 936          * new mode group and new policy group are started */
 937         prev->res_link = queue->prev;
 938         prev->mode_link = &req->l_sl_mode;
 939         prev->policy_link = &req->l_sl_policy;
 940         EXIT;
 941         return;
 942 }
 943
 944 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
 945                                        struct sl_insert_point *prev)
 946 {
 947         struct ldlm_resource *res = lock->l_resource;
 948         ENTRY;
 949
 950         check_res_locked(res);
 951
 952         ldlm_resource_dump(D_INFO, res);
 953         LDLM_DEBUG(lock, "About to add lock:");
 954
 955         if (lock->l_destroyed) {
 956                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 957                 return;
 958         }
 959
 960         LASSERT(cfs_list_empty(&lock->l_res_link));
 961         LASSERT(cfs_list_empty(&lock->l_sl_mode));
 962         LASSERT(cfs_list_empty(&lock->l_sl_policy));
 963
 964         cfs_list_add(&lock->l_res_link, prev->res_link);
 965         cfs_list_add(&lock->l_sl_mode, prev->mode_link);
 966         cfs_list_add(&lock->l_sl_policy, prev->policy_link);
 967
 968         EXIT;
 969 }
 970
 971 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
 972 {
 973         struct sl_insert_point prev;
 974         ENTRY;
 975
 976         LASSERT(lock->l_req_mode == lock->l_granted_mode);
 977
 978         search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
 979         ldlm_granted_list_add_lock(lock, &prev);
 980         EXIT;
 981 }
 982
 983 /* NOTE: called by
 984  *  - ldlm_lock_enqueue
 985  *  - ldlm_reprocess_queue
 986  *  - ldlm_lock_convert
 987  *
 988  * must be called with lr_lock held
 989  */
 990 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
 991 {
 992         struct ldlm_resource *res = lock->l_resource;
 993         ENTRY;
 994
 995         check_res_locked(res);
 996
 997         lock->l_granted_mode = lock->l_req_mode;
 998         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
 999                 ldlm_grant_lock_with_skiplist(lock);
1000         else if (res->lr_type == LDLM_EXTENT)
1001                 ldlm_extent_add_lock(res, lock);
1002         else
1003                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1004
1005         if (lock->l_granted_mode < res->lr_most_restr)
1006                 res->lr_most_restr = lock->l_granted_mode;
1007
1008         if (work_list && lock->l_completion_ast != NULL)
1009                 ldlm_add_ast_work_item(lock, NULL, work_list);
1010
1011         ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1012         EXIT;
1013 }
1014
1015 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
1016  * comment above ldlm_lock_match */
1017 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1018                                       ldlm_mode_t *mode,
1019                                       ldlm_policy_data_t *policy,
1020                                       struct ldlm_lock *old_lock,
1021                                       int flags, int unref)
1022 {
1023         struct ldlm_lock *lock;
1024         cfs_list_t       *tmp;
1025
1026         cfs_list_for_each(tmp, queue) {
1027                 ldlm_mode_t match;
1028
1029                 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1030
1031                 if (lock == old_lock)
1032                         break;
1033
1034                 /* llite sometimes wants to match locks that will be
1035                  * canceled when their users drop, but we allow it to match
1036                  * if it passes in CBPENDING and the lock still has users.
1037                  * this is generally only going to be used by children
1038                  * whose parents already hold a lock so forward progress
1039                  * can still happen. */
1040                 if (lock->l_flags & LDLM_FL_CBPENDING &&
1041                     !(flags & LDLM_FL_CBPENDING))
1042                         continue;
1043                 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
1044                     lock->l_readers == 0 && lock->l_writers == 0)
1045                         continue;
1046
1047                 if (!(lock->l_req_mode & *mode))
1048                         continue;
1049                 match = lock->l_req_mode;
1050
1051                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1052                     (lock->l_policy_data.l_extent.start >
1053                      policy->l_extent.start ||
1054                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
1055                         continue;
1056
1057                 if (unlikely(match == LCK_GROUP) &&
1058                     lock->l_resource->lr_type == LDLM_EXTENT &&
1059                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1060                         continue;
1061
1062                 /* We match if we have existing lock with same or wider set
1063                    of bits. */
1064                 if (lock->l_resource->lr_type == LDLM_IBITS &&
1065                      ((lock->l_policy_data.l_inodebits.bits &
1066                       policy->l_inodebits.bits) !=
1067                       policy->l_inodebits.bits))
1068                         continue;
1069
1070                 if (!unref &&
1071                     (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1072                      lock->l_failed))
1073                         continue;
1074
1075                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1076                     !(lock->l_flags & LDLM_FL_LOCAL))
1077                         continue;
1078
1079                 if (flags & LDLM_FL_TEST_LOCK) {
1080                         LDLM_LOCK_GET(lock);
1081                         ldlm_lock_touch_in_lru(lock);
1082                 } else {
1083                         ldlm_lock_addref_internal_nolock(lock, match);
1084                 }
1085                 *mode = match;
1086                 return lock;
1087         }
1088
1089         return NULL;
1090 }
1091
1092 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1093 {
1094         if (!lock->l_failed) {
1095                 lock->l_failed = 1;
1096                 cfs_waitq_broadcast(&lock->l_waitq);
1097         }
1098 }
1099 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1100
1101 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1102 {
1103         lock_res_and_lock(lock);
1104         ldlm_lock_fail_match_locked(lock);
1105         unlock_res_and_lock(lock);
1106 }
1107 EXPORT_SYMBOL(ldlm_lock_fail_match);
1108
1109 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1110 {
1111         lock->l_flags |= LDLM_FL_LVB_READY;
1112         cfs_waitq_broadcast(&lock->l_waitq);
1113 }
1114
1115 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1116 {
1117         lock_res_and_lock(lock);
1118         ldlm_lock_allow_match_locked(lock);
1119         unlock_res_and_lock(lock);
1120 }
1121
1122 /* Can be called in two ways:
1123  *
1124  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1125  * for a duplicate of.
1126  *
1127  * Otherwise, all of the fields must be filled in, to match against.
1128  *
1129  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1130  *     server (ie, connh is NULL)
1131  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1132  *     list will be considered
1133  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1134  *     to be canceled can still be matched as long as they still have reader
1135  *     or writer refernces
1136  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1137  *     just tell us if we would have matched.
1138  *
1139  * Returns 1 if it finds an already-existing lock that is compatible; in this
1140  * case, lockh is filled in with a addref()ed lock
1141  *
1142  * we also check security context, if that failed we simply return 0 (to keep
1143  * caller code unchanged), the context failure will be discovered by caller
1144  * sometime later.
1145  */
1146 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1147                             const struct ldlm_res_id *res_id, ldlm_type_t type,
1148                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
1149                             struct lustre_handle *lockh, int unref)
1150 {
1151         struct ldlm_resource *res;
1152         struct ldlm_lock *lock, *old_lock = NULL;
1153         int rc = 0;
1154         ENTRY;
1155
1156         if (ns == NULL) {
1157                 old_lock = ldlm_handle2lock(lockh);
1158                 LASSERT(old_lock);
1159
1160                 ns = ldlm_lock_to_ns(old_lock);
1161                 res_id = &old_lock->l_resource->lr_name;
1162                 type = old_lock->l_resource->lr_type;
1163                 mode = old_lock->l_req_mode;
1164         }
1165
1166         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1167         if (res == NULL) {
1168                 LASSERT(old_lock == NULL);
1169                 RETURN(0);
1170         }
1171
1172         LDLM_RESOURCE_ADDREF(res);
1173         lock_res(res);
1174
1175         lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1176                             flags, unref);
1177         if (lock != NULL)
1178                 GOTO(out, rc = 1);
1179         if (flags & LDLM_FL_BLOCK_GRANTED)
1180                 GOTO(out, rc = 0);
1181         lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1182                             flags, unref);
1183         if (lock != NULL)
1184                 GOTO(out, rc = 1);
1185         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1186                             flags, unref);
1187         if (lock != NULL)
1188                 GOTO(out, rc = 1);
1189
1190         EXIT;
1191  out:
1192         unlock_res(res);
1193         LDLM_RESOURCE_DELREF(res);
1194         ldlm_resource_putref(res);
1195
1196         if (lock) {
1197                 ldlm_lock2handle(lock, lockh);
1198                 if ((flags & LDLM_FL_LVB_READY) &&
1199                     (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1200                         struct l_wait_info lwi;
1201                         if (lock->l_completion_ast) {
1202                                 int err = lock->l_completion_ast(lock,
1203                                                           LDLM_FL_WAIT_NOREPROC,
1204                                                                  NULL);
1205                                 if (err) {
1206                                         if (flags & LDLM_FL_TEST_LOCK)
1207                                                 LDLM_LOCK_RELEASE(lock);
1208                                         else
1209                                                 ldlm_lock_decref_internal(lock,
1210                                                                           mode);
1211                                         rc = 0;
1212                                         goto out2;
1213                                 }
1214                         }
1215
1216                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1217                                                NULL, LWI_ON_SIGNAL_NOOP, NULL);
1218
1219                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1220                         l_wait_event(lock->l_waitq,
1221                                      lock->l_flags & LDLM_FL_LVB_READY ||
1222                                      lock->l_failed,
1223                                      &lwi);
1224                         if (!(lock->l_flags & LDLM_FL_LVB_READY)) {
1225                                 if (flags & LDLM_FL_TEST_LOCK)
1226                                         LDLM_LOCK_RELEASE(lock);
1227                                 else
1228                                         ldlm_lock_decref_internal(lock, mode);
1229                                 rc = 0;
1230                         }
1231                 }
1232         }
1233  out2:
1234         if (rc) {
1235                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1236                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1237                                 res_id->name[2] : policy->l_extent.start,
1238                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1239                                 res_id->name[3] : policy->l_extent.end);
1240
1241                 /* check user's security context */
1242                 if (lock->l_conn_export &&
1243                     sptlrpc_import_check_ctx(
1244                                 class_exp2cliimp(lock->l_conn_export))) {
1245                         if (!(flags & LDLM_FL_TEST_LOCK))
1246                                 ldlm_lock_decref_internal(lock, mode);
1247                         rc = 0;
1248                 }
1249
1250                 if (flags & LDLM_FL_TEST_LOCK)
1251                         LDLM_LOCK_RELEASE(lock);
1252
1253         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1254                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1255                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1256                                   type, mode, res_id->name[0], res_id->name[1],
1257                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1258                                         res_id->name[2] :policy->l_extent.start,
1259                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1260                                         res_id->name[3] : policy->l_extent.end);
1261         }
1262         if (old_lock)
1263                 LDLM_LOCK_PUT(old_lock);
1264
1265         return rc ? mode : 0;
1266 }
1267
1268 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1269                                         __u64 *bits)
1270 {
1271         struct ldlm_lock *lock;
1272         ldlm_mode_t mode = 0;
1273         ENTRY;
1274
1275         lock = ldlm_handle2lock(lockh);
1276         if (lock != NULL) {
1277                 lock_res_and_lock(lock);
1278                 if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1279                     lock->l_failed)
1280                         GOTO(out, mode);
1281
1282                 if (lock->l_flags & LDLM_FL_CBPENDING &&
1283                     lock->l_readers == 0 && lock->l_writers == 0)
1284                         GOTO(out, mode);
1285
1286                 if (bits)
1287                         *bits = lock->l_policy_data.l_inodebits.bits;
1288                 mode = lock->l_granted_mode;
1289                 ldlm_lock_addref_internal_nolock(lock, mode);
1290         }
1291
1292         EXIT;
1293
1294 out:
1295         if (lock != NULL) {
1296                 unlock_res_and_lock(lock);
1297                 LDLM_LOCK_PUT(lock);
1298         }
1299         return mode;
1300 }
1301 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1302
1303 /* Returns a referenced lock */
1304 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1305                                    const struct ldlm_res_id *res_id,
1306                                    ldlm_type_t type,
1307                                    ldlm_mode_t mode,
1308                                    const struct ldlm_callback_suite *cbs,
1309                                    void *data, __u32 lvb_len)
1310 {
1311         struct ldlm_lock *lock;
1312         struct ldlm_resource *res;
1313         ENTRY;
1314
1315         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1316         if (res == NULL)
1317                 RETURN(NULL);
1318
1319         lock = ldlm_lock_new(res);
1320
1321         if (lock == NULL)
1322                 RETURN(NULL);
1323
1324         lock->l_req_mode = mode;
1325         lock->l_ast_data = data;
1326         lock->l_pid = cfs_curproc_pid();
1327         lock->l_ns_srv = !!ns_is_server(ns);
1328         if (cbs) {
1329                 lock->l_blocking_ast = cbs->lcs_blocking;
1330                 lock->l_completion_ast = cbs->lcs_completion;
1331                 lock->l_glimpse_ast = cbs->lcs_glimpse;
1332                 lock->l_weigh_ast = cbs->lcs_weigh;
1333         }
1334
1335         lock->l_tree_node = NULL;
1336         /* if this is the extent lock, allocate the interval tree node */
1337         if (type == LDLM_EXTENT) {
1338                 if (ldlm_interval_alloc(lock) == NULL)
1339                         GOTO(out, 0);
1340         }
1341
1342         if (lvb_len) {
1343                 lock->l_lvb_len = lvb_len;
1344                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1345                 if (lock->l_lvb_data == NULL)
1346                         GOTO(out, 0);
1347         }
1348
1349         if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1350                 GOTO(out, 0);
1351
1352         RETURN(lock);
1353
1354 out:
1355         ldlm_lock_destroy(lock);
1356         LDLM_LOCK_RELEASE(lock);
1357         return NULL;
1358 }
1359
1360 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1361                                struct ldlm_lock **lockp,
1362                                void *cookie, int *flags)
1363 {
1364         struct ldlm_lock *lock = *lockp;
1365         struct ldlm_resource *res = lock->l_resource;
1366         int local = ns_is_client(ldlm_res_to_ns(res));
1367 #ifdef HAVE_SERVER_SUPPORT
1368         ldlm_processing_policy policy;
1369 #endif
1370         ldlm_error_t rc = ELDLM_OK;
1371         struct ldlm_interval *node = NULL;
1372         ENTRY;
1373
1374         lock->l_last_activity = cfs_time_current_sec();
1375         /* policies are not executed on the client or during replay */
1376         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1377             && !local && ns->ns_policy) {
1378                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1379                                    NULL);
1380                 if (rc == ELDLM_LOCK_REPLACED) {
1381                         /* The lock that was returned has already been granted,
1382                          * and placed into lockp.  If it's not the same as the
1383                          * one we passed in, then destroy the old one and our
1384                          * work here is done. */
1385                         if (lock != *lockp) {
1386                                 ldlm_lock_destroy(lock);
1387                                 LDLM_LOCK_RELEASE(lock);
1388                         }
1389                         *flags |= LDLM_FL_LOCK_CHANGED;
1390                         RETURN(0);
1391                 } else if (rc != ELDLM_OK ||
1392                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1393                         ldlm_lock_destroy(lock);
1394                         RETURN(rc);
1395                 }
1396         }
1397
1398         /* For a replaying lock, it might be already in granted list. So
1399          * unlinking the lock will cause the interval node to be freed, we
1400          * have to allocate the interval node early otherwise we can't regrant
1401          * this lock in the future. - jay */
1402         if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1403                 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1404
1405         lock_res_and_lock(lock);
1406         if (local && lock->l_req_mode == lock->l_granted_mode) {
1407                 /* The server returned a blocked lock, but it was granted
1408                  * before we got a chance to actually enqueue it.  We don't
1409                  * need to do anything else. */
1410                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1411                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1412                 GOTO(out, ELDLM_OK);
1413         }
1414
1415         ldlm_resource_unlink_lock(lock);
1416         if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1417                 if (node == NULL) {
1418                         ldlm_lock_destroy_nolock(lock);
1419                         GOTO(out, rc = -ENOMEM);
1420                 }
1421
1422                 CFS_INIT_LIST_HEAD(&node->li_group);
1423                 ldlm_interval_attach(node, lock);
1424                 node = NULL;
1425         }
1426
1427         /* Some flags from the enqueue want to make it into the AST, via the
1428          * lock's l_flags. */
1429         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1430
1431         /* This distinction between local lock trees is very important; a client
1432          * namespace only has information about locks taken by that client, and
1433          * thus doesn't have enough information to decide for itself if it can
1434          * be granted (below).  In this case, we do exactly what the server
1435          * tells us to do, as dictated by the 'flags'.
1436          *
1437          * We do exactly the same thing during recovery, when the server is
1438          * more or less trusting the clients not to lie.
1439          *
1440          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1441          * granted/converting queues. */
1442         if (local) {
1443                 if (*flags & LDLM_FL_BLOCK_CONV)
1444                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1445                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1446                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1447                 else
1448                         ldlm_grant_lock(lock, NULL);
1449                 GOTO(out, ELDLM_OK);
1450 #ifdef HAVE_SERVER_SUPPORT
1451         } else if (*flags & LDLM_FL_REPLAY) {
1452                 if (*flags & LDLM_FL_BLOCK_CONV) {
1453                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1454                         GOTO(out, ELDLM_OK);
1455                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1456                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1457                         GOTO(out, ELDLM_OK);
1458                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1459                         ldlm_grant_lock(lock, NULL);
1460                         GOTO(out, ELDLM_OK);
1461                 }
1462                 /* If no flags, fall through to normal enqueue path. */
1463         }
1464
1465         policy = ldlm_processing_policy_table[res->lr_type];
1466         policy(lock, flags, 1, &rc, NULL);
1467         GOTO(out, rc);
1468 #else
1469         } else {
1470                 CERROR("This is client-side-only module, cannot handle "
1471                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
1472                 LBUG();
1473         }
1474 #endif
1475
1476 out:
1477         unlock_res_and_lock(lock);
1478         if (node)
1479                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1480         return rc;
1481 }
1482
1483 #ifdef HAVE_SERVER_SUPPORT
1484 /* Must be called with namespace taken: queue is waiting or converting. */
1485 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1486                          cfs_list_t *work_list)
1487 {
1488         cfs_list_t *tmp, *pos;
1489         ldlm_processing_policy policy;
1490         int flags;
1491         int rc = LDLM_ITER_CONTINUE;
1492         ldlm_error_t err;
1493         ENTRY;
1494
1495         check_res_locked(res);
1496
1497         policy = ldlm_processing_policy_table[res->lr_type];
1498         LASSERT(policy);
1499
1500         cfs_list_for_each_safe(tmp, pos, queue) {
1501                 struct ldlm_lock *pending;
1502                 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1503
1504                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1505
1506                 flags = 0;
1507                 rc = policy(pending, &flags, 0, &err, work_list);
1508                 if (rc != LDLM_ITER_CONTINUE)
1509                         break;
1510         }
1511
1512         RETURN(rc);
1513 }
1514 #endif
1515
1516 static int
1517 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1518 {
1519         struct ldlm_cb_set_arg *arg = opaq;
1520         struct ldlm_lock_desc   d;
1521         int                     rc;
1522         struct ldlm_lock       *lock;
1523         ENTRY;
1524
1525         if (cfs_list_empty(arg->list))
1526                 RETURN(-ENOENT);
1527
1528         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1529
1530         /* nobody should touch l_bl_ast */
1531         lock_res_and_lock(lock);
1532         cfs_list_del_init(&lock->l_bl_ast);
1533
1534         LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1535         LASSERT(lock->l_bl_ast_run == 0);
1536         LASSERT(lock->l_blocking_lock);
1537         lock->l_bl_ast_run++;
1538         unlock_res_and_lock(lock);
1539
1540         ldlm_lock2desc(lock->l_blocking_lock, &d);
1541
1542         rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1543         LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1544         lock->l_blocking_lock = NULL;
1545         LDLM_LOCK_RELEASE(lock);
1546
1547         RETURN(rc);
1548 }
1549
1550 static int
1551 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1552 {
1553         struct ldlm_cb_set_arg  *arg = opaq;
1554         int                      rc = 0;
1555         struct ldlm_lock        *lock;
1556         ldlm_completion_callback completion_callback;
1557         ENTRY;
1558
1559         if (cfs_list_empty(arg->list))
1560                 RETURN(-ENOENT);
1561
1562         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1563
1564         /* It's possible to receive a completion AST before we've set
1565          * the l_completion_ast pointer: either because the AST arrived
1566          * before the reply, or simply because there's a small race
1567          * window between receiving the reply and finishing the local
1568          * enqueue. (bug 842)
1569          *
1570          * This can't happen with the blocking_ast, however, because we
1571          * will never call the local blocking_ast until we drop our
1572          * reader/writer reference, which we won't do until we get the
1573          * reply and finish enqueueing. */
1574
1575         /* nobody should touch l_cp_ast */
1576         lock_res_and_lock(lock);
1577         cfs_list_del_init(&lock->l_cp_ast);
1578         LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1579         /* save l_completion_ast since it can be changed by
1580          * mds_intent_policy(), see bug 14225 */
1581         completion_callback = lock->l_completion_ast;
1582         lock->l_flags &= ~LDLM_FL_CP_REQD;
1583         unlock_res_and_lock(lock);
1584
1585         if (completion_callback != NULL)
1586                 rc = completion_callback(lock, 0, (void *)arg);
1587         LDLM_LOCK_RELEASE(lock);
1588
1589         RETURN(rc);
1590 }
1591
1592 static int
1593 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1594 {
1595         struct ldlm_cb_set_arg *arg = opaq;
1596         struct ldlm_lock_desc   desc;
1597         int                     rc;
1598         struct ldlm_lock       *lock;
1599         ENTRY;
1600
1601         if (cfs_list_empty(arg->list))
1602                 RETURN(-ENOENT);
1603
1604         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1605         cfs_list_del_init(&lock->l_rk_ast);
1606
1607         /* the desc just pretend to exclusive */
1608         ldlm_lock2desc(lock, &desc);
1609         desc.l_req_mode = LCK_EX;
1610         desc.l_granted_mode = 0;
1611
1612         rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1613         LDLM_LOCK_RELEASE(lock);
1614
1615         RETURN(rc);
1616 }
1617
1618 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1619                       ldlm_desc_ast_t ast_type)
1620 {
1621         struct ldlm_cb_set_arg *arg;
1622         set_producer_func       work_ast_lock;
1623         int                     rc;
1624
1625         if (cfs_list_empty(rpc_list))
1626                 RETURN(0);
1627
1628         OBD_ALLOC_PTR(arg);
1629         if (arg == NULL)
1630                 RETURN(-ENOMEM);
1631
1632         cfs_atomic_set(&arg->restart, 0);
1633         arg->list = rpc_list;
1634
1635         switch (ast_type) {
1636                 case LDLM_WORK_BL_AST:
1637                         arg->type = LDLM_BL_CALLBACK;
1638                         work_ast_lock = ldlm_work_bl_ast_lock;
1639                         break;
1640                 case LDLM_WORK_CP_AST:
1641                         arg->type = LDLM_CP_CALLBACK;
1642                         work_ast_lock = ldlm_work_cp_ast_lock;
1643                         break;
1644                 case LDLM_WORK_REVOKE_AST:
1645                         arg->type = LDLM_BL_CALLBACK;
1646                         work_ast_lock = ldlm_work_revoke_ast_lock;
1647                         break;
1648                 default:
1649                         LBUG();
1650         }
1651
1652         /* We create a ptlrpc request set with flow control extension.
1653          * This request set will use the work_ast_lock function to produce new
1654          * requests and will send a new request each time one completes in order
1655          * to keep the number of requests in flight to ns_max_parallel_ast */
1656         arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
1657                                      work_ast_lock, arg);
1658         if (arg->set == NULL)
1659                 GOTO(out, rc = -ENOMEM);
1660
1661         ptlrpc_set_wait(arg->set);
1662         ptlrpc_set_destroy(arg->set);
1663
1664         rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0;
1665         GOTO(out, rc);
1666 out:
1667         OBD_FREE_PTR(arg);
1668         return rc;
1669 }
1670
1671 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1672 {
1673         ldlm_reprocess_all(res);
1674         return LDLM_ITER_CONTINUE;
1675 }
1676
1677 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1678                               cfs_hlist_node_t *hnode, void *arg)
1679 {
1680         struct ldlm_resource *res = cfs_hash_object(hs, hnode);
1681         int    rc;
1682
1683         rc = reprocess_one_queue(res, arg);
1684
1685         return rc == LDLM_ITER_STOP;
1686 }
1687
1688 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1689 {
1690         ENTRY;
1691
1692         if (ns != NULL) {
1693                 cfs_hash_for_each_nolock(ns->ns_rs_hash,
1694                                          ldlm_reprocess_res, NULL);
1695         }
1696         EXIT;
1697 }
1698
1699 void ldlm_reprocess_all(struct ldlm_resource *res)
1700 {
1701         CFS_LIST_HEAD(rpc_list);
1702
1703 #ifdef HAVE_SERVER_SUPPORT
1704         int rc;
1705         ENTRY;
1706         /* Local lock trees don't get reprocessed. */
1707         if (ns_is_client(ldlm_res_to_ns(res))) {
1708                 EXIT;
1709                 return;
1710         }
1711
1712 restart:
1713         lock_res(res);
1714         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1715         if (rc == LDLM_ITER_CONTINUE)
1716                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1717         unlock_res(res);
1718
1719         rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
1720                                LDLM_WORK_CP_AST);
1721         if (rc == -ERESTART) {
1722                 LASSERT(cfs_list_empty(&rpc_list));
1723                 goto restart;
1724         }
1725 #else
1726         ENTRY;
1727         if (!ns_is_client(ldlm_res_to_ns(res))) {
1728                 CERROR("This is client-side-only module, cannot handle "
1729                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
1730                 LBUG();
1731         }
1732 #endif
1733         EXIT;
1734 }
1735
1736 void ldlm_cancel_callback(struct ldlm_lock *lock)
1737 {
1738         check_res_locked(lock->l_resource);
1739         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1740                 lock->l_flags |= LDLM_FL_CANCEL;
1741                 if (lock->l_blocking_ast) {
1742                         // l_check_no_ns_lock(ns);
1743                         unlock_res_and_lock(lock);
1744                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1745                                              LDLM_CB_CANCELING);
1746                         lock_res_and_lock(lock);
1747                 } else {
1748                         LDLM_DEBUG(lock, "no blocking ast");
1749                 }
1750         }
1751         lock->l_flags |= LDLM_FL_BL_DONE;
1752 }
1753
1754 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1755 {
1756         if (req->l_resource->lr_type != LDLM_PLAIN &&
1757             req->l_resource->lr_type != LDLM_IBITS)
1758                 return;
1759
1760         cfs_list_del_init(&req->l_sl_policy);
1761         cfs_list_del_init(&req->l_sl_mode);
1762 }
1763
1764 void ldlm_lock_cancel(struct ldlm_lock *lock)
1765 {
1766         struct ldlm_resource *res;
1767         struct ldlm_namespace *ns;
1768         ENTRY;
1769
1770         lock_res_and_lock(lock);
1771
1772         res = lock->l_resource;
1773         ns  = ldlm_res_to_ns(res);
1774
1775         /* Please do not, no matter how tempting, remove this LBUG without
1776          * talking to me first. -phik */
1777         if (lock->l_readers || lock->l_writers) {
1778                 LDLM_ERROR(lock, "lock still has references");
1779                 LBUG();
1780         }
1781
1782         ldlm_del_waiting_lock(lock);
1783
1784         /* Releases cancel callback. */
1785         ldlm_cancel_callback(lock);
1786
1787         /* Yes, second time, just in case it was added again while we were
1788            running with no res lock in ldlm_cancel_callback */
1789         ldlm_del_waiting_lock(lock);
1790         ldlm_resource_unlink_lock(lock);
1791         ldlm_lock_destroy_nolock(lock);
1792
1793         if (lock->l_granted_mode == lock->l_req_mode)
1794                 ldlm_pool_del(&ns->ns_pool, lock);
1795
1796         /* Make sure we will not be called again for same lock what is possible
1797          * if not to zero out lock->l_granted_mode */
1798         lock->l_granted_mode = LCK_MINMODE;
1799         unlock_res_and_lock(lock);
1800
1801         EXIT;
1802 }
1803
1804 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1805 {
1806         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1807         int rc = -EINVAL;
1808         ENTRY;
1809
1810         if (lock) {
1811                 if (lock->l_ast_data == NULL)
1812                         lock->l_ast_data = data;
1813                 if (lock->l_ast_data == data)
1814                         rc = 0;
1815                 LDLM_LOCK_PUT(lock);
1816         }
1817         RETURN(rc);
1818 }
1819 EXPORT_SYMBOL(ldlm_lock_set_data);
1820
1821 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1822                                     cfs_hlist_node_t *hnode, void *data)
1823
1824 {
1825         struct obd_export    *exp  = data;
1826         struct ldlm_lock     *lock = cfs_hash_object(hs, hnode);
1827         struct ldlm_resource *res;
1828
1829         res = ldlm_resource_getref(lock->l_resource);
1830         LDLM_LOCK_GET(lock);
1831
1832         LDLM_DEBUG(lock, "export %p", exp);
1833         ldlm_res_lvbo_update(res, NULL, 1);
1834         ldlm_lock_cancel(lock);
1835         ldlm_reprocess_all(res);
1836         ldlm_resource_putref(res);
1837         LDLM_LOCK_RELEASE(lock);
1838         return 0;
1839 }
1840
1841 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1842 {
1843         cfs_hash_for_each_empty(exp->exp_lock_hash,
1844                                 ldlm_cancel_locks_for_export_cb, exp);
1845 }
1846
1847 /**
1848  * Downgrade an exclusive lock.
1849  *
1850  * A fast variant of ldlm_lock_convert for convertion of exclusive
1851  * locks. The convertion is always successful.
1852  *
1853  * \param lock A lock to convert
1854  * \param new_mode new lock mode
1855  */
1856 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
1857 {
1858         ENTRY;
1859
1860         LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
1861         LASSERT(new_mode == LCK_COS);
1862
1863         lock_res_and_lock(lock);
1864         ldlm_resource_unlink_lock(lock);
1865         /*
1866          * Remove the lock from pool as it will be added again in
1867          * ldlm_grant_lock() called below.
1868          */
1869         ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
1870
1871         lock->l_req_mode = new_mode;
1872         ldlm_grant_lock(lock, NULL);
1873         unlock_res_and_lock(lock);
1874         ldlm_reprocess_all(lock->l_resource);
1875
1876         EXIT;
1877 }
1878
1879 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1880                                         __u32 *flags)
1881 {
1882         CFS_LIST_HEAD(rpc_list);
1883         struct ldlm_resource *res;
1884         struct ldlm_namespace *ns;
1885         int granted = 0;
1886         int old_mode;
1887         struct sl_insert_point prev;
1888         struct ldlm_interval *node;
1889         ENTRY;
1890
1891         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1892                 *flags |= LDLM_FL_BLOCK_GRANTED;
1893                 RETURN(lock->l_resource);
1894         }
1895
1896         /* I can't check the type of lock here because the bitlock of lock
1897          * is not held here, so do the allocation blindly. -jay */
1898         OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1899         if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
1900                 RETURN(NULL);
1901
1902         LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
1903                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1904
1905         lock_res_and_lock(lock);
1906
1907         res = lock->l_resource;
1908         ns  = ldlm_res_to_ns(res);
1909
1910         old_mode = lock->l_req_mode;
1911         lock->l_req_mode = new_mode;
1912         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1913                 /* remember the lock position where the lock might be
1914                  * added back to the granted list later and also
1915                  * remember the join mode for skiplist fixing. */
1916                 prev.res_link = lock->l_res_link.prev;
1917                 prev.mode_link = lock->l_sl_mode.prev;
1918                 prev.policy_link = lock->l_sl_policy.prev;
1919                 ldlm_resource_unlink_lock(lock);
1920         } else {
1921                 ldlm_resource_unlink_lock(lock);
1922                 if (res->lr_type == LDLM_EXTENT) {
1923                         /* FIXME: ugly code, I have to attach the lock to a
1924                          * interval node again since perhaps it will be granted
1925                          * soon */
1926                         CFS_INIT_LIST_HEAD(&node->li_group);
1927                         ldlm_interval_attach(node, lock);
1928                         node = NULL;
1929                 }
1930         }
1931
1932         /*
1933          * Remove old lock from the pool before adding the lock with new
1934          * mode below in ->policy()
1935          */
1936         ldlm_pool_del(&ns->ns_pool, lock);
1937
1938         /* If this is a local resource, put it on the appropriate list. */
1939         if (ns_is_client(ldlm_res_to_ns(res))) {
1940                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1941                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1942                 } else {
1943                         /* This should never happen, because of the way the
1944                          * server handles conversions. */
1945                         LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1946                                    *flags);
1947                         LBUG();
1948
1949                         ldlm_grant_lock(lock, &rpc_list);
1950                         granted = 1;
1951                         /* FIXME: completion handling not with lr_lock held ! */
1952                         if (lock->l_completion_ast)
1953                                 lock->l_completion_ast(lock, 0, NULL);
1954                 }
1955 #ifdef HAVE_SERVER_SUPPORT
1956         } else {
1957                 int rc;
1958                 ldlm_error_t err;
1959                 int pflags = 0;
1960                 ldlm_processing_policy policy;
1961                 policy = ldlm_processing_policy_table[res->lr_type];
1962                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1963                 if (rc == LDLM_ITER_STOP) {
1964                         lock->l_req_mode = old_mode;
1965                         if (res->lr_type == LDLM_EXTENT)
1966                                 ldlm_extent_add_lock(res, lock);
1967                         else
1968                                 ldlm_granted_list_add_lock(lock, &prev);
1969
1970                         res = NULL;
1971                 } else {
1972                         *flags |= LDLM_FL_BLOCK_GRANTED;
1973                         granted = 1;
1974                 }
1975         }
1976 #else
1977         } else {
1978                 CERROR("This is client-side-only module, cannot handle "
1979                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
1980                 LBUG();
1981         }
1982 #endif
1983         unlock_res_and_lock(lock);
1984
1985         if (granted)
1986                 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
1987         if (node)
1988                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1989         RETURN(res);
1990 }
1991
1992 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1993 {
1994         struct ldlm_lock *lock;
1995
1996         if (!((libcfs_debug | D_ERROR) & level))
1997                 return;
1998
1999         lock = ldlm_handle2lock(lockh);
2000         if (lock == NULL)
2001                 return;
2002
2003         LDLM_DEBUG_LIMIT(level, lock, "###");
2004
2005         LDLM_LOCK_PUT(lock);
2006 }
2007
2008 void _ldlm_lock_debug(struct ldlm_lock *lock,
2009                       struct libcfs_debug_msg_data *msgdata,
2010                       const char *fmt, ...)
2011 {
2012         va_list args;
2013         struct obd_export *exp = lock->l_export;
2014         struct ldlm_resource *resource = lock->l_resource;
2015         char *nid = "local";
2016
2017         va_start(args, fmt);
2018
2019         if (exp && exp->exp_connection) {
2020                 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2021         } else if (exp && exp->exp_obd != NULL) {
2022                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2023                 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2024         }
2025
2026         if (resource == NULL) {
2027                 libcfs_debug_vmsg2(msgdata, fmt, args,
2028                        " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2029                        "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2030                        "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2031                        lock,
2032                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2033                        lock->l_readers, lock->l_writers,
2034                        ldlm_lockname[lock->l_granted_mode],
2035                        ldlm_lockname[lock->l_req_mode],
2036                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2037                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2038                        lock->l_pid, lock->l_callback_timeout);
2039                 va_end(args);
2040                 return;
2041         }
2042
2043         switch (resource->lr_type) {
2044         case LDLM_EXTENT:
2045                 libcfs_debug_vmsg2(msgdata, fmt, args,
2046                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2047                        "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
2048                        "] (req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote:"
2049                        " "LPX64" expref: %d pid: %u timeout %lu\n",
2050                        ldlm_lock_to_ns_name(lock), lock,
2051                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2052                        lock->l_readers, lock->l_writers,
2053                        ldlm_lockname[lock->l_granted_mode],
2054                        ldlm_lockname[lock->l_req_mode],
2055                        resource->lr_name.name[0],
2056                        resource->lr_name.name[1],
2057                        cfs_atomic_read(&resource->lr_refcount),
2058                        ldlm_typename[resource->lr_type],
2059                        lock->l_policy_data.l_extent.start,
2060                        lock->l_policy_data.l_extent.end,
2061                        lock->l_req_extent.start, lock->l_req_extent.end,
2062                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2063                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2064                        lock->l_pid, lock->l_callback_timeout);
2065                 break;
2066
2067         case LDLM_FLOCK:
2068                 libcfs_debug_vmsg2(msgdata, fmt, args,
2069                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2070                        "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
2071                        "["LPU64"->"LPU64"] flags: "LPX64" nid: %s remote: "LPX64
2072                        " expref: %d pid: %u timeout: %lu\n",
2073                        ldlm_lock_to_ns_name(lock), lock,
2074                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2075                        lock->l_readers, lock->l_writers,
2076                        ldlm_lockname[lock->l_granted_mode],
2077                        ldlm_lockname[lock->l_req_mode],
2078                        resource->lr_name.name[0],
2079                        resource->lr_name.name[1],
2080                        cfs_atomic_read(&resource->lr_refcount),
2081                        ldlm_typename[resource->lr_type],
2082                        lock->l_policy_data.l_flock.pid,
2083                        lock->l_policy_data.l_flock.start,
2084                        lock->l_policy_data.l_flock.end,
2085                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2086                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2087                        lock->l_pid, lock->l_callback_timeout);
2088                 break;
2089
2090         case LDLM_IBITS:
2091                 libcfs_debug_vmsg2(msgdata, fmt, args,
2092                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2093                        "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
2094                        "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2095                        "pid: %u timeout: %lu\n",
2096                        ldlm_lock_to_ns_name(lock),
2097                        lock, lock->l_handle.h_cookie,
2098                        cfs_atomic_read (&lock->l_refc),
2099                        lock->l_readers, lock->l_writers,
2100                        ldlm_lockname[lock->l_granted_mode],
2101                        ldlm_lockname[lock->l_req_mode],
2102                        resource->lr_name.name[0],
2103                        resource->lr_name.name[1],
2104                        lock->l_policy_data.l_inodebits.bits,
2105                        cfs_atomic_read(&resource->lr_refcount),
2106                        ldlm_typename[resource->lr_type],
2107                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2108                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2109                        lock->l_pid, lock->l_callback_timeout);
2110                 break;
2111
2112         default:
2113                 libcfs_debug_vmsg2(msgdata, fmt, args,
2114                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2115                        "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
2116                        "nid: %s remote: "LPX64" expref: %d pid: %u timeout %lu"
2117                        "\n",
2118                        ldlm_lock_to_ns_name(lock),
2119                        lock, lock->l_handle.h_cookie,
2120                        cfs_atomic_read (&lock->l_refc),
2121                        lock->l_readers, lock->l_writers,
2122                        ldlm_lockname[lock->l_granted_mode],
2123                        ldlm_lockname[lock->l_req_mode],
2124                        resource->lr_name.name[0],
2125                        resource->lr_name.name[1],
2126                        cfs_atomic_read(&resource->lr_refcount),
2127                        ldlm_typename[resource->lr_type],
2128                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2129                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2130                        lock->l_pid, lock->l_callback_timeout);
2131                 break;
2132         }
2133         va_end(args);
2134 }
2135 EXPORT_SYMBOL(_ldlm_lock_debug);