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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 # include <libcfs/kp30.h>
  50 #endif
  51
  52 #include <obd_class.h>
  53 #include "ldlm_internal.h"
  54
  55 //struct lustre_lock ldlm_everything_lock;
  56
  57 /* lock types */
  58 char *ldlm_lockname[] = {
  59         [0] "--",
  60         [LCK_EX] "EX",
  61         [LCK_PW] "PW",
  62         [LCK_PR] "PR",
  63         [LCK_CW] "CW",
  64         [LCK_CR] "CR",
  65         [LCK_NL] "NL",
  66         [LCK_GROUP] "GROUP"
  67 };
  68
  69 char *ldlm_typename[] = {
  70         [LDLM_PLAIN] "PLN",
  71         [LDLM_EXTENT] "EXT",
  72         [LDLM_FLOCK] "FLK",
  73         [LDLM_IBITS] "IBT",
  74 };
  75
  76 char *ldlm_it2str(int it)
  77 {
  78         switch (it) {
  79         case IT_OPEN:
  80                 return "open";
  81         case IT_CREAT:
  82                 return "creat";
  83         case (IT_OPEN | IT_CREAT):
  84                 return "open|creat";
  85         case IT_READDIR:
  86                 return "readdir";
  87         case IT_GETATTR:
  88                 return "getattr";
  89         case IT_LOOKUP:
  90                 return "lookup";
  91         case IT_UNLINK:
  92                 return "unlink";
  93         case IT_GETXATTR:
  94                 return "getxattr";
  95         default:
  96                 CERROR("Unknown intent %d\n", it);
  97                 return "UNKNOWN";
  98         }
  99 }
 100
 101 extern cfs_mem_cache_t *ldlm_lock_slab;
 102
 103 static ldlm_processing_policy ldlm_processing_policy_table[] = {
 104         [LDLM_PLAIN] ldlm_process_plain_lock,
 105         [LDLM_EXTENT] ldlm_process_extent_lock,
 106 #ifdef __KERNEL__
 107         [LDLM_FLOCK] ldlm_process_flock_lock,
 108 #endif
 109         [LDLM_IBITS] ldlm_process_inodebits_lock,
 110 };
 111
 112 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
 113 {
 114         return ldlm_processing_policy_table[res->lr_type];
 115 }
 116
 117 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
 118 {
 119         ns->ns_policy = arg;
 120 }
 121
 122 /*
 123  * REFCOUNTED LOCK OBJECTS
 124  */
 125
 126
 127 /*
 128  * Lock refcounts, during creation:
 129  *   - one special one for allocation, dec'd only once in destroy
 130  *   - one for being a lock that's in-use
 131  *   - one for the addref associated with a new lock
 132  */
 133 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 134 {
 135         atomic_inc(&lock->l_refc);
 136         return lock;
 137 }
 138
 139 static void ldlm_lock_free(struct ldlm_lock *lock, size_t size)
 140 {
 141         LASSERT(size == sizeof(*lock));
 142         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
 143 }
 144
 145 void ldlm_lock_put(struct ldlm_lock *lock)
 146 {
 147         ENTRY;
 148
 149         LASSERT(lock->l_resource != LP_POISON);
 150         LASSERT(atomic_read(&lock->l_refc) > 0);
 151         if (atomic_dec_and_test(&lock->l_refc)) {
 152                 struct ldlm_resource *res;
 153
 154                 LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing it.");
 155
 156                 res = lock->l_resource;
 157                 LASSERT(lock->l_destroyed);
 158                 LASSERT(list_empty(&lock->l_res_link));
 159                 LASSERT(list_empty(&lock->l_pending_chain));
 160
 161                 atomic_dec(&res->lr_namespace->ns_locks);
 162                 ldlm_resource_putref(res);
 163                 lock->l_resource = NULL;
 164                 if (lock->l_export) {
 165                         class_export_put(lock->l_export);
 166                         lock->l_export = NULL;
 167                 }
 168
 169                 if (lock->l_lvb_data != NULL)
 170                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
 171
 172                 ldlm_interval_free(ldlm_interval_detach(lock));
 173                 OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
 174                                 ldlm_lock_free);
 175         }
 176
 177         EXIT;
 178 }
 179
 180 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
 181 {
 182         int rc = 0;
 183         if (!list_empty(&lock->l_lru)) {
 184                 struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 185                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 186                 list_del_init(&lock->l_lru);
 187                 ns->ns_nr_unused--;
 188                 LASSERT(ns->ns_nr_unused >= 0);
 189                 rc = 1;
 190         }
 191         return rc;
 192 }
 193
 194 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 195 {
 196         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 197         int rc;
 198         ENTRY;
 199         spin_lock(&ns->ns_unused_lock);
 200         rc = ldlm_lock_remove_from_lru_nolock(lock);
 201         spin_unlock(&ns->ns_unused_lock);
 202         EXIT;
 203         return rc;
 204 }
 205
 206 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
 207 {
 208         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 209         lock->l_last_used = cfs_time_current();
 210         LASSERT(list_empty(&lock->l_lru));
 211         LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 212         list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 213         LASSERT(ns->ns_nr_unused >= 0);
 214         ns->ns_nr_unused++;
 215 }
 216
 217 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
 218 {
 219         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 220         ENTRY;
 221         spin_lock(&ns->ns_unused_lock);
 222         ldlm_lock_add_to_lru_nolock(lock);
 223         spin_unlock(&ns->ns_unused_lock);
 224         EXIT;
 225 }
 226
 227 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
 228 {
 229         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 230         ENTRY;
 231         spin_lock(&ns->ns_unused_lock);
 232         if (!list_empty(&lock->l_lru)) {
 233                 ldlm_lock_remove_from_lru_nolock(lock);
 234                 ldlm_lock_add_to_lru_nolock(lock);
 235         }
 236         spin_unlock(&ns->ns_unused_lock);
 237         EXIT;
 238 }
 239
 240 /* This used to have a 'strict' flag, which recovery would use to mark an
 241  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 242  * shall explain why it's gone: with the new hash table scheme, once you call
 243  * ldlm_lock_destroy, you can never drop your final references on this lock.
 244  * Because it's not in the hash table anymore.  -phil */
 245 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
 246 {
 247         ENTRY;
 248
 249         if (lock->l_readers || lock->l_writers) {
 250                 LDLM_ERROR(lock, "lock still has references");
 251                 ldlm_lock_dump(D_ERROR, lock, 0);
 252                 LBUG();
 253         }
 254
 255         if (!list_empty(&lock->l_res_link)) {
 256                 LDLM_ERROR(lock, "lock still on resource");
 257                 ldlm_lock_dump(D_ERROR, lock, 0);
 258                 LBUG();
 259         }
 260
 261         if (lock->l_destroyed) {
 262                 LASSERT(list_empty(&lock->l_lru));
 263                 EXIT;
 264                 return 0;
 265         }
 266         lock->l_destroyed = 1;
 267
 268         if (lock->l_export && lock->l_export->exp_lock_hash)
 269                 lustre_hash_del(lock->l_export->exp_lock_hash,
 270                                 &lock->l_remote_handle, &lock->l_exp_hash);
 271
 272         ldlm_lock_remove_from_lru(lock);
 273         class_handle_unhash(&lock->l_handle);
 274
 275 #if 0
 276         /* Wake anyone waiting for this lock */
 277         /* FIXME: I should probably add yet another flag, instead of using
 278          * l_export to only call this on clients */
 279         if (lock->l_export)
 280                 class_export_put(lock->l_export);
 281         lock->l_export = NULL;
 282         if (lock->l_export && lock->l_completion_ast)
 283                 lock->l_completion_ast(lock, 0);
 284 #endif
 285         EXIT;
 286         return 1;
 287 }
 288
 289 void ldlm_lock_destroy(struct ldlm_lock *lock)
 290 {
 291         int first;
 292         ENTRY;
 293         lock_res_and_lock(lock);
 294         first = ldlm_lock_destroy_internal(lock);
 295         unlock_res_and_lock(lock);
 296
 297         /* drop reference from hashtable only for first destroy */
 298         if (first)
 299                 LDLM_LOCK_PUT(lock);
 300         EXIT;
 301 }
 302
 303 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
 304 {
 305         int first;
 306         ENTRY;
 307         first = ldlm_lock_destroy_internal(lock);
 308         /* drop reference from hashtable only for first destroy */
 309         if (first)
 310                 LDLM_LOCK_PUT(lock);
 311         EXIT;
 312 }
 313
 314 /* this is called by portals_handle2object with the handle lock taken */
 315 static void lock_handle_addref(void *lock)
 316 {
 317         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 318 }
 319
 320 /*
 321  * usage: pass in a resource on which you have done ldlm_resource_get
 322  *        pass in a parent lock on which you have done a ldlm_lock_get
 323  *        after return, ldlm_*_put the resource and parent
 324  * returns: lock with refcount 2 - one for current caller and one for remote
 325  */
 326 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
 327 {
 328         struct ldlm_lock *lock;
 329         ENTRY;
 330
 331         if (resource == NULL)
 332                 LBUG();
 333
 334         OBD_SLAB_ALLOC(lock, ldlm_lock_slab, CFS_ALLOC_IO, sizeof(*lock));
 335         if (lock == NULL)
 336                 RETURN(NULL);
 337
 338         spin_lock_init(&lock->l_lock);
 339         lock->l_resource = ldlm_resource_getref(resource);
 340
 341         atomic_set(&lock->l_refc, 2);
 342         CFS_INIT_LIST_HEAD(&lock->l_res_link);
 343         CFS_INIT_LIST_HEAD(&lock->l_lru);
 344         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
 345         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
 346         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
 347         cfs_waitq_init(&lock->l_waitq);
 348         lock->l_blocking_lock = NULL;
 349         CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
 350         CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
 351         CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
 352
 353         atomic_inc(&resource->lr_namespace->ns_locks);
 354         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
 355         class_handle_hash(&lock->l_handle, lock_handle_addref);
 356
 357         CFS_INIT_LIST_HEAD(&lock->l_extents_list);
 358         spin_lock_init(&lock->l_extents_list_lock);
 359         CFS_INIT_LIST_HEAD(&lock->l_cache_locks_list);
 360         lock->l_callback_timeout = 0;
 361
 362         RETURN(lock);
 363 }
 364
 365 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 366                               struct ldlm_res_id new_resid)
 367 {
 368         struct ldlm_resource *oldres = lock->l_resource;
 369         struct ldlm_resource *newres;
 370         int type;
 371         ENTRY;
 372
 373         LASSERT(ns_is_client(ns));
 374
 375         lock_res_and_lock(lock);
 376         if (memcmp(&new_resid, &lock->l_resource->lr_name,
 377                    sizeof(lock->l_resource->lr_name)) == 0) {
 378                 /* Nothing to do */
 379                 unlock_res_and_lock(lock);
 380                 RETURN(0);
 381         }
 382
 383         LASSERT(new_resid.name[0] != 0);
 384
 385         /* This function assumes that the lock isn't on any lists */
 386         LASSERT(list_empty(&lock->l_res_link));
 387
 388         type = oldres->lr_type;
 389         unlock_res_and_lock(lock);
 390
 391         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 392         if (newres == NULL)
 393                 RETURN(-ENOMEM);
 394
 395         lock_res_and_lock(lock);
 396         LASSERT(memcmp(&new_resid, &lock->l_resource->lr_name,
 397                        sizeof(lock->l_resource->lr_name)) != 0);
 398         lock_res(newres);
 399         lock->l_resource = newres;
 400         unlock_res(oldres);
 401         unlock_res_and_lock(lock);
 402
 403         /* ...and the flowers are still standing! */
 404         ldlm_resource_putref(oldres);
 405
 406         RETURN(0);
 407 }
 408
 409 /*
 410  *  HANDLES
 411  */
 412
 413 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 414 {
 415         lockh->cookie = lock->l_handle.h_cookie;
 416 }
 417
 418 /* if flags: atomically get the lock and set the flags.
 419  *           Return NULL if flag already set
 420  */
 421
 422 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *handle, int flags)
 423 {
 424         struct ldlm_namespace *ns;
 425         struct ldlm_lock *lock = NULL, *retval = NULL;
 426         ENTRY;
 427
 428         LASSERT(handle);
 429
 430         lock = class_handle2object(handle->cookie);
 431         if (lock == NULL)
 432                 RETURN(NULL);
 433
 434         LASSERT(lock->l_resource != NULL);
 435         ns = lock->l_resource->lr_namespace;
 436         LASSERT(ns != NULL);
 437
 438         lock_res_and_lock(lock);
 439
 440         /* It's unlikely but possible that someone marked the lock as
 441          * destroyed after we did handle2object on it */
 442         if (lock->l_destroyed) {
 443                 unlock_res_and_lock(lock);
 444                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 445                 LDLM_LOCK_PUT(lock);
 446                 GOTO(out, retval);
 447         }
 448
 449         if (flags && (lock->l_flags & flags)) {
 450                 unlock_res_and_lock(lock);
 451                 LDLM_LOCK_PUT(lock);
 452                 GOTO(out, retval);
 453         }
 454
 455         if (flags)
 456                 lock->l_flags |= flags;
 457
 458         unlock_res_and_lock(lock);
 459         retval = lock;
 460         EXIT;
 461  out:
 462         return retval;
 463 }
 464
 465 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
 466                                       struct lustre_handle *handle)
 467 {
 468         struct ldlm_lock *retval = NULL;
 469         retval = __ldlm_handle2lock(handle, 0);
 470         return retval;
 471 }
 472
 473 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 474 {
 475         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
 476         /* INODEBITS_INTEROP: If the other side does not support
 477          * inodebits, reply with a plain lock descriptor.
 478          */
 479         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
 480             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
 481                 struct ldlm_resource res = *lock->l_resource;
 482
 483                 /* Make sure all the right bits are set in this lock we
 484                    are going to pass to client */
 485                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
 486                          (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
 487                          "Inappropriate inode lock bits during "
 488                          "conversion " LPU64 "\n",
 489                          lock->l_policy_data.l_inodebits.bits);
 490                 res.lr_type = LDLM_PLAIN;
 491                 ldlm_res2desc(&res, &desc->l_resource);
 492                 /* Convert "new" lock mode to something old client can
 493                    understand */
 494                 if ((lock->l_req_mode == LCK_CR) ||
 495                     (lock->l_req_mode == LCK_CW))
 496                         desc->l_req_mode = LCK_PR;
 497                 else
 498                         desc->l_req_mode = lock->l_req_mode;
 499                 if ((lock->l_granted_mode == LCK_CR) ||
 500                     (lock->l_granted_mode == LCK_CW)) {
 501                         desc->l_granted_mode = LCK_PR;
 502                 } else {
 503                         /* We never grant PW/EX locks to clients */
 504                         LASSERT((lock->l_granted_mode != LCK_PW) &&
 505                                 (lock->l_granted_mode != LCK_EX));
 506                         desc->l_granted_mode = lock->l_granted_mode;
 507                 }
 508
 509                 /* We do not copy policy here, because there is no
 510                    policy for plain locks */
 511         } else {
 512                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 513                 desc->l_req_mode = lock->l_req_mode;
 514                 desc->l_granted_mode = lock->l_granted_mode;
 515                 desc->l_policy_data = lock->l_policy_data;
 516         }
 517 }
 518
 519 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 520                            struct list_head *work_list)
 521 {
 522         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
 523                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 524                 lock->l_flags |= LDLM_FL_AST_SENT;
 525                 /* If the enqueuing client said so, tell the AST recipient to
 526                  * discard dirty data, rather than writing back. */
 527                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 528                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 529                 LASSERT(list_empty(&lock->l_bl_ast));
 530                 list_add(&lock->l_bl_ast, work_list);
 531                 LDLM_LOCK_GET(lock);
 532                 LASSERT(lock->l_blocking_lock == NULL);
 533                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
 534         }
 535 }
 536
 537 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
 538 {
 539         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
 540                 lock->l_flags |= LDLM_FL_CP_REQD;
 541                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
 542                 LASSERT(list_empty(&lock->l_cp_ast));
 543                 list_add(&lock->l_cp_ast, work_list);
 544                 LDLM_LOCK_GET(lock);
 545         }
 546 }
 547
 548 /* must be called with lr_lock held */
 549 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 550                                 struct list_head *work_list)
 551 {
 552         ENTRY;
 553         check_res_locked(lock->l_resource);
 554         if (new)
 555                 ldlm_add_bl_work_item(lock, new, work_list);
 556         else
 557                 ldlm_add_cp_work_item(lock, work_list);
 558         EXIT;
 559 }
 560
 561 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 562 {
 563         struct ldlm_lock *lock;
 564
 565         lock = ldlm_handle2lock(lockh);
 566         LASSERT(lock != NULL);
 567         ldlm_lock_addref_internal(lock, mode);
 568         LDLM_LOCK_PUT(lock);
 569 }
 570
 571 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 572 {
 573         ldlm_lock_remove_from_lru(lock);
 574         if (mode & (LCK_NL | LCK_CR | LCK_PR))
 575                 lock->l_readers++;
 576         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
 577                 lock->l_writers++;
 578         LDLM_LOCK_GET(lock);
 579         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 580 }
 581
 582 /* only called for local locks */
 583 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 584 {
 585         lock_res_and_lock(lock);
 586         ldlm_lock_addref_internal_nolock(lock, mode);
 587         unlock_res_and_lock(lock);
 588 }
 589
 590 /* only called in ldlm_flock_destroy and for local locks.
 591  * for LDLM_FLOCK type locks, l_blocking_ast is null, and
 592  * ldlm_lock_remove_from_lru() does nothing, it is safe
 593  * for ldlm_flock_destroy usage by dropping some code */
 594 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 595 {
 596         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 597         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 598                 LASSERT(lock->l_readers > 0);
 599                 lock->l_readers--;
 600         }
 601         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
 602                 LASSERT(lock->l_writers > 0);
 603                 lock->l_writers--;
 604         }
 605
 606         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 607 }
 608
 609 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 610 {
 611         struct ldlm_namespace *ns;
 612         ENTRY;
 613
 614         lock_res_and_lock(lock);
 615
 616         ns = lock->l_resource->lr_namespace;
 617
 618         ldlm_lock_decref_internal_nolock(lock, mode);
 619
 620         if (lock->l_flags & LDLM_FL_LOCAL &&
 621             !lock->l_readers && !lock->l_writers) {
 622                 /* If this is a local lock on a server namespace and this was
 623                  * the last reference, cancel the lock. */
 624                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 625                 lock->l_flags |= LDLM_FL_CBPENDING;
 626         }
 627
 628         if (!lock->l_readers && !lock->l_writers &&
 629             (lock->l_flags & LDLM_FL_CBPENDING)) {
 630                 /* If we received a blocked AST and this was the last reference,
 631                  * run the callback. */
 632                 if (ns_is_server(ns) && lock->l_export)
 633                         CERROR("FL_CBPENDING set on non-local lock--just a "
 634                                "warning\n");
 635
 636                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 637
 638                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
 639                 ldlm_lock_remove_from_lru(lock);
 640                 unlock_res_and_lock(lock);
 641
 642                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
 643                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
 644
 645                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
 646                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
 647                         ldlm_handle_bl_callback(ns, NULL, lock);
 648         } else if (ns_is_client(ns) &&
 649                    !lock->l_readers && !lock->l_writers &&
 650                    !(lock->l_flags & LDLM_FL_NO_LRU) &&
 651                    !(lock->l_flags & LDLM_FL_BL_AST)) {
 652                 /* If this is a client-side namespace and this was the last
 653                  * reference, put it on the LRU. */
 654                 ldlm_lock_add_to_lru(lock);
 655                 unlock_res_and_lock(lock);
 656                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
 657                  * are not supported by the server, otherwise, it is done on
 658                  * enqueue. */
 659                 if (!exp_connect_cancelset(lock->l_conn_export) &&
 660                     !ns_connect_lru_resize(ns))
 661                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
 662         } else {
 663                 unlock_res_and_lock(lock);
 664         }
 665
 666         EXIT;
 667 }
 668
 669 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 670 {
 671         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 672         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
 673         ldlm_lock_decref_internal(lock, mode);
 674         LDLM_LOCK_PUT(lock);
 675 }
 676
 677 /* This will drop a lock reference and mark it for destruction, but will not
 678  * necessarily cancel the lock before returning. */
 679 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 680 {
 681         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 682         ENTRY;
 683
 684         LASSERT(lock != NULL);
 685
 686         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 687         lock_res_and_lock(lock);
 688         lock->l_flags |= LDLM_FL_CBPENDING;
 689         unlock_res_and_lock(lock);
 690         ldlm_lock_decref_internal(lock, mode);
 691         LDLM_LOCK_PUT(lock);
 692 }
 693
 694 struct sl_insert_point {
 695         struct list_head *res_link;
 696         struct list_head *mode_link;
 697         struct list_head *policy_link;
 698 };
 699
 700 /*
 701  * search_granted_lock
 702  *
 703  * Description:
 704  *      Finds a position to insert the new lock.
 705  * Parameters:
 706  *      queue [input]:  the granted list where search acts on;
 707  *      req [input]:    the lock whose position to be located;
 708  *      prev [output]:  positions within 3 lists to insert @req to
 709  * Return Value:
 710  *      filled @prev
 711  * NOTE: called by
 712  *  - ldlm_grant_lock_with_skiplist
 713  */
 714 static void search_granted_lock(struct list_head *queue,
 715                                 struct ldlm_lock *req,
 716                                 struct sl_insert_point *prev)
 717 {
 718         struct list_head *tmp;
 719         struct ldlm_lock *lock, *mode_end, *policy_end;
 720         ENTRY;
 721
 722         list_for_each(tmp, queue) {
 723                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 724
 725                 mode_end = list_entry(lock->l_sl_mode.prev, struct ldlm_lock,
 726                                       l_sl_mode);
 727
 728                 if (lock->l_req_mode != req->l_req_mode) {
 729                         /* jump to last lock of mode group */
 730                         tmp = &mode_end->l_res_link;
 731                         continue;
 732                 }
 733
 734                 /* suitable mode group is found */
 735                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
 736                         /* insert point is last lock of the mode group */
 737                         prev->res_link = &mode_end->l_res_link;
 738                         prev->mode_link = &mode_end->l_sl_mode;
 739                         prev->policy_link = &req->l_sl_policy;
 740                         EXIT;
 741                         return;
 742                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
 743                         for (;;) {
 744                                 policy_end = list_entry(lock->l_sl_policy.prev,
 745                                                         struct ldlm_lock,
 746                                                         l_sl_policy);
 747
 748                                 if (lock->l_policy_data.l_inodebits.bits ==
 749                                     req->l_policy_data.l_inodebits.bits) {
 750                                         /* insert point is last lock of
 751                                          * the policy group */
 752                                         prev->res_link =
 753                                                 &policy_end->l_res_link;
 754                                         prev->mode_link =
 755                                                 &policy_end->l_sl_mode;
 756                                         prev->policy_link =
 757                                                 &policy_end->l_sl_policy;
 758                                         EXIT;
 759                                         return;
 760                                 }
 761
 762                                 if (policy_end == mode_end)
 763                                         /* done with mode group */
 764                                         break;
 765
 766                                 /* jump to next policy group within the mode group */
 767                                 tmp = policy_end->l_res_link.next;
 768                                 lock = list_entry(tmp, struct ldlm_lock,
 769                                                   l_res_link);
 770                         }  /* loop over policy groups within the mode group */
 771
 772                         /* insert point is last lock of the mode group,
 773                          * new policy group is started */
 774                         prev->res_link = &mode_end->l_res_link;
 775                         prev->mode_link = &mode_end->l_sl_mode;
 776                         prev->policy_link = &req->l_sl_policy;
 777                         EXIT;
 778                         return;
 779                 } else {
 780                         LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
 781                         LBUG();
 782                 }
 783         }
 784
 785         /* insert point is last lock on the queue,
 786          * new mode group and new policy group are started */
 787         prev->res_link = queue->prev;
 788         prev->mode_link = &req->l_sl_mode;
 789         prev->policy_link = &req->l_sl_policy;
 790         EXIT;
 791         return;
 792 }
 793
 794 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
 795                                        struct sl_insert_point *prev)
 796 {
 797         struct ldlm_resource *res = lock->l_resource;
 798         ENTRY;
 799
 800         check_res_locked(res);
 801
 802         ldlm_resource_dump(D_INFO, res);
 803         CDEBUG(D_OTHER, "About to add this lock:\n");
 804         ldlm_lock_dump(D_OTHER, lock, 0);
 805
 806         if (lock->l_destroyed) {
 807                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 808                 return;
 809         }
 810
 811         LASSERT(list_empty(&lock->l_res_link));
 812         LASSERT(list_empty(&lock->l_sl_mode));
 813         LASSERT(list_empty(&lock->l_sl_policy));
 814
 815         list_add(&lock->l_res_link, prev->res_link);
 816         list_add(&lock->l_sl_mode, prev->mode_link);
 817         list_add(&lock->l_sl_policy, prev->policy_link);
 818
 819         EXIT;
 820 }
 821
 822 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
 823 {
 824         struct sl_insert_point prev;
 825         ENTRY;
 826
 827         LASSERT(lock->l_req_mode == lock->l_granted_mode);
 828
 829         search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
 830         ldlm_granted_list_add_lock(lock, &prev);
 831         EXIT;
 832 }
 833
 834 /* NOTE: called by
 835  *  - ldlm_lock_enqueue
 836  *  - ldlm_reprocess_queue
 837  *  - ldlm_lock_convert
 838  *
 839  * must be called with lr_lock held
 840  */
 841 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
 842 {
 843         struct ldlm_resource *res = lock->l_resource;
 844         ENTRY;
 845
 846         check_res_locked(res);
 847
 848         lock->l_granted_mode = lock->l_req_mode;
 849         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
 850                 ldlm_grant_lock_with_skiplist(lock);
 851         else if (res->lr_type == LDLM_EXTENT)
 852                 ldlm_extent_add_lock(res, lock);
 853         else
 854                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
 855
 856         if (lock->l_granted_mode < res->lr_most_restr)
 857                 res->lr_most_restr = lock->l_granted_mode;
 858
 859         if (work_list && lock->l_completion_ast != NULL)
 860                 ldlm_add_ast_work_item(lock, NULL, work_list);
 861
 862         ldlm_pool_add(&res->lr_namespace->ns_pool, lock);
 863         EXIT;
 864 }
 865
 866 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
 867  * comment above ldlm_lock_match */
 868 static struct ldlm_lock *search_queue(struct list_head *queue,
 869                                       ldlm_mode_t *mode,
 870                                       ldlm_policy_data_t *policy,
 871                                       struct ldlm_lock *old_lock, int flags)
 872 {
 873         struct ldlm_lock *lock;
 874         struct list_head *tmp;
 875
 876         list_for_each(tmp, queue) {
 877                 ldlm_mode_t match;
 878
 879                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 880
 881                 if (lock == old_lock)
 882                         break;
 883
 884                 /* llite sometimes wants to match locks that will be
 885                  * canceled when their users drop, but we allow it to match
 886                  * if it passes in CBPENDING and the lock still has users.
 887                  * this is generally only going to be used by children
 888                  * whose parents already hold a lock so forward progress
 889                  * can still happen. */
 890                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 891                     !(flags & LDLM_FL_CBPENDING))
 892                         continue;
 893                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 894                     lock->l_readers == 0 && lock->l_writers == 0)
 895                         continue;
 896
 897                 if (!(lock->l_req_mode & *mode))
 898                         continue;
 899
 900                 match = lock->l_req_mode;
 901                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 902                     (lock->l_policy_data.l_extent.start >
 903                      policy->l_extent.start ||
 904                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
 905                         continue;
 906
 907                 if (unlikely(match == LCK_GROUP) &&
 908                     lock->l_resource->lr_type == LDLM_EXTENT &&
 909                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
 910                         continue;
 911
 912                 /* We match if we have existing lock with same or wider set
 913                    of bits. */
 914                 if (lock->l_resource->lr_type == LDLM_IBITS &&
 915                      ((lock->l_policy_data.l_inodebits.bits &
 916                       policy->l_inodebits.bits) !=
 917                       policy->l_inodebits.bits))
 918                         continue;
 919
 920                 if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
 921                         continue;
 922
 923                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
 924                     !(lock->l_flags & LDLM_FL_LOCAL))
 925                         continue;
 926
 927                 if (flags & LDLM_FL_TEST_LOCK) {
 928                         LDLM_LOCK_GET(lock);
 929                         ldlm_lock_touch_in_lru(lock);
 930                 } else {
 931                         ldlm_lock_addref_internal_nolock(lock, match);
 932                 }
 933                 *mode = match;
 934                 return lock;
 935         }
 936
 937         return NULL;
 938 }
 939
 940 void ldlm_lock_allow_match(struct ldlm_lock *lock)
 941 {
 942         lock_res_and_lock(lock);
 943         lock->l_flags |= LDLM_FL_LVB_READY;
 944         cfs_waitq_signal(&lock->l_waitq);
 945         unlock_res_and_lock(lock);
 946 }
 947
 948 int ldlm_lock_fast_match(struct ldlm_lock *lock, int rw,
 949                          obd_off start, obd_off end,
 950                          void **cookie)
 951 {
 952         LASSERT(rw == OBD_BRW_READ || rw == OBD_BRW_WRITE);
 953
 954         if (!lock)
 955                 return 0;
 956
 957         lock_res_and_lock(lock);
 958         /* check if granted mode is compatible */
 959         if (rw == OBD_BRW_WRITE &&
 960             !(lock->l_granted_mode & (LCK_PW|LCK_GROUP)))
 961                 goto no_match;
 962
 963         /* does the lock cover the region we would like to access? */
 964         if ((lock->l_policy_data.l_extent.start > start) ||
 965             (lock->l_policy_data.l_extent.end < end))
 966                 goto no_match;
 967
 968         /* if we received a blocking callback and the lock is no longer
 969          * referenced, don't use it */
 970         if ((lock->l_flags & LDLM_FL_CBPENDING) &&
 971             !lock->l_writers && !lock->l_readers)
 972                 goto no_match;
 973
 974         ldlm_lock_addref_internal_nolock(lock, rw == OBD_BRW_WRITE ? LCK_PW : LCK_PR);
 975         unlock_res_and_lock(lock);
 976         *cookie = (void *)lock;
 977         return 1; /* avoid using rc for stack relief */
 978
 979 no_match:
 980         unlock_res_and_lock(lock);
 981         return 0;
 982 }
 983
 984 void ldlm_lock_fast_release(void *cookie, int rw)
 985 {
 986         struct ldlm_lock *lock = (struct ldlm_lock *)cookie;
 987
 988         LASSERT(lock != NULL);
 989         LASSERT(rw == OBD_BRW_READ || rw == OBD_BRW_WRITE);
 990         LASSERT(rw == OBD_BRW_READ || (lock->l_granted_mode & (LCK_PW | LCK_GROUP)));
 991         ldlm_lock_decref_internal(lock, rw == OBD_BRW_WRITE ? LCK_PW : LCK_PR);
 992 }
 993
 994 /* Can be called in two ways:
 995  *
 996  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 997  * for a duplicate of.
 998  *
 999  * Otherwise, all of the fields must be filled in, to match against.
1000  *
1001  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1002  *     server (ie, connh is NULL)
1003  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1004  *     list will be considered
1005  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1006  *     to be canceled can still be matched as long as they still have reader
1007  *     or writer refernces
1008  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1009  *     just tell us if we would have matched.
1010  *
1011  * Returns 1 if it finds an already-existing lock that is compatible; in this
1012  * case, lockh is filled in with a addref()ed lock
1013  */
1014 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1015                             struct ldlm_res_id *res_id, ldlm_type_t type,
1016                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
1017                             struct lustre_handle *lockh)
1018 {
1019         struct ldlm_resource *res;
1020         struct ldlm_lock *lock, *old_lock = NULL;
1021         int rc = 0;
1022         ENTRY;
1023
1024         if (ns == NULL) {
1025                 old_lock = ldlm_handle2lock(lockh);
1026                 LASSERT(old_lock);
1027
1028                 ns = old_lock->l_resource->lr_namespace;
1029                 res_id = &old_lock->l_resource->lr_name;
1030                 type = old_lock->l_resource->lr_type;
1031                 mode = old_lock->l_req_mode;
1032         }
1033
1034         res = ldlm_resource_get(ns, NULL, *res_id, type, 0);
1035         if (res == NULL) {
1036                 LASSERT(old_lock == NULL);
1037                 RETURN(0);
1038         }
1039
1040         lock_res(res);
1041
1042         lock = search_queue(&res->lr_granted, &mode, policy, old_lock, flags);
1043         if (lock != NULL)
1044                 GOTO(out, rc = 1);
1045         if (flags & LDLM_FL_BLOCK_GRANTED)
1046                 GOTO(out, rc = 0);
1047         lock = search_queue(&res->lr_converting, &mode, policy, old_lock, flags);
1048         if (lock != NULL)
1049                 GOTO(out, rc = 1);
1050         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock, flags);
1051         if (lock != NULL)
1052                 GOTO(out, rc = 1);
1053
1054         EXIT;
1055  out:
1056         unlock_res(res);
1057         ldlm_resource_putref(res);
1058
1059         if (lock) {
1060                 ldlm_lock2handle(lock, lockh);
1061                 if ((flags & LDLM_FL_LVB_READY) && (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1062                         struct l_wait_info lwi;
1063                         if (lock->l_completion_ast) {
1064                                 int err = lock->l_completion_ast(lock,
1065                                                           LDLM_FL_WAIT_NOREPROC,
1066                                                                  NULL);
1067                                 if (err) {
1068                                         if (flags & LDLM_FL_TEST_LOCK)
1069                                                 LDLM_LOCK_PUT(lock);
1070                                         else
1071                                                 ldlm_lock_decref_internal(lock, mode);
1072                                         rc = 0;
1073                                         goto out2;
1074                                 }
1075                         }
1076
1077                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
1078                                                LWI_ON_SIGNAL_NOOP, NULL);
1079
1080                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1081                         l_wait_event(lock->l_waitq,
1082                                      (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1083                 }
1084         }
1085  out2:
1086         if (rc) {
1087                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1088                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1089                                 res_id->name[2] : policy->l_extent.start,
1090                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1091                                 res_id->name[3] : policy->l_extent.end);
1092         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1093                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1094                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1095                                   type, mode, res_id->name[0], res_id->name[1],
1096                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1097                                         res_id->name[2] :policy->l_extent.start,
1098                                 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1099                                         res_id->name[3] : policy->l_extent.end);
1100         }
1101         if (old_lock)
1102                 LDLM_LOCK_PUT(old_lock);
1103         if (flags & LDLM_FL_TEST_LOCK && rc)
1104                 LDLM_LOCK_PUT(lock);
1105
1106         return rc ? mode : 0;
1107 }
1108
1109 /* Returns a referenced lock */
1110 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1111                                    struct ldlm_res_id res_id, ldlm_type_t type,
1112                                    ldlm_mode_t mode,
1113                                    ldlm_blocking_callback blocking,
1114                                    ldlm_completion_callback completion,
1115                                    ldlm_glimpse_callback glimpse,
1116                                    void *data, __u32 lvb_len)
1117 {
1118         struct ldlm_lock *lock;
1119         struct ldlm_resource *res;
1120         ENTRY;
1121
1122         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1123         if (res == NULL)
1124                 RETURN(NULL);
1125
1126         lock = ldlm_lock_new(res);
1127         ldlm_resource_putref(res);
1128
1129         if (lock == NULL)
1130                 RETURN(NULL);
1131
1132         lock->l_req_mode = mode;
1133         lock->l_ast_data = data;
1134         lock->l_blocking_ast = blocking;
1135         lock->l_completion_ast = completion;
1136         lock->l_glimpse_ast = glimpse;
1137         lock->l_pid = cfs_curproc_pid();
1138
1139         lock->l_tree_node = NULL;
1140         /* if this is the extent lock, allocate the interval tree node */
1141         if (type == LDLM_EXTENT) {
1142                 if (ldlm_interval_alloc(lock) == NULL)
1143                         GOTO(out, 0);
1144         }
1145
1146         if (lvb_len) {
1147                 lock->l_lvb_len = lvb_len;
1148                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1149                 if (lock->l_lvb_data == NULL)
1150                         GOTO(out, 0);
1151         }
1152
1153         RETURN(lock);
1154
1155 out:
1156         if (lock->l_lvb_data)
1157                 OBD_FREE(lock->l_lvb_data, lvb_len);
1158         ldlm_interval_free(ldlm_interval_detach(lock));
1159         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1160         return NULL;
1161 }
1162
1163 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1164                                struct ldlm_lock **lockp,
1165                                void *cookie, int *flags)
1166 {
1167         struct ldlm_lock *lock = *lockp;
1168         struct ldlm_resource *res = lock->l_resource;
1169         int local = ns_is_client(res->lr_namespace);
1170         ldlm_processing_policy policy;
1171         ldlm_error_t rc = ELDLM_OK;
1172         struct ldlm_interval *node = NULL;
1173         ENTRY;
1174
1175         do_gettimeofday(&lock->l_enqueued_time);
1176         /* policies are not executed on the client or during replay */
1177         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1178             && !local && ns->ns_policy) {
1179                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1180                                    NULL);
1181                 if (rc == ELDLM_LOCK_REPLACED) {
1182                         /* The lock that was returned has already been granted,
1183                          * and placed into lockp.  If it's not the same as the
1184                          * one we passed in, then destroy the old one and our
1185                          * work here is done. */
1186                         if (lock != *lockp) {
1187                                 ldlm_lock_destroy(lock);
1188                                 LDLM_LOCK_PUT(lock);
1189                         }
1190                         *flags |= LDLM_FL_LOCK_CHANGED;
1191                         RETURN(0);
1192                 } else if (rc != ELDLM_OK ||
1193                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1194                         ldlm_lock_destroy(lock);
1195                         RETURN(rc);
1196                 }
1197         }
1198
1199         /* For a replaying lock, it might be already in granted list. So
1200          * unlinking the lock will cause the interval node to be freed, we
1201          * have to allocate the interval node early otherwise we can't regrant
1202          * this lock in the future. - jay */
1203         if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1204                 OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO,
1205                                sizeof(*node));
1206
1207         lock_res_and_lock(lock);
1208         if (local && lock->l_req_mode == lock->l_granted_mode) {
1209                 /* The server returned a blocked lock, but it was granted
1210                  * before we got a chance to actually enqueue it.  We don't
1211                  * need to do anything else. */
1212                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1213                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1214                 GOTO(out, ELDLM_OK);
1215         }
1216
1217         ldlm_resource_unlink_lock(lock);
1218         if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1219                 LASSERT(!local && (*flags & LDLM_FL_REPLAY));
1220                 if (node == NULL) {
1221                         ldlm_lock_destroy_nolock(lock);
1222                         GOTO(out, rc = -ENOMEM);
1223                 }
1224
1225                 CFS_INIT_LIST_HEAD(&node->li_group);
1226                 ldlm_interval_attach(node, lock);
1227                 node = NULL;
1228         }
1229
1230         /* Some flags from the enqueue want to make it into the AST, via the
1231          * lock's l_flags. */
1232         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1233
1234         /* This distinction between local lock trees is very important; a client
1235          * namespace only has information about locks taken by that client, and
1236          * thus doesn't have enough information to decide for itself if it can
1237          * be granted (below).  In this case, we do exactly what the server
1238          * tells us to do, as dictated by the 'flags'.
1239          *
1240          * We do exactly the same thing during recovery, when the server is
1241          * more or less trusting the clients not to lie.
1242          *
1243          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1244          * granted/converting queues. */
1245         if (local) {
1246                 if (*flags & LDLM_FL_BLOCK_CONV)
1247                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1248                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1249                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1250                 else
1251                         ldlm_grant_lock(lock, NULL);
1252                 GOTO(out, ELDLM_OK);
1253         } else if (*flags & LDLM_FL_REPLAY) {
1254                 if (*flags & LDLM_FL_BLOCK_CONV) {
1255                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1256                         GOTO(out, ELDLM_OK);
1257                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1258                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1259                         GOTO(out, ELDLM_OK);
1260                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1261                         ldlm_grant_lock(lock, NULL);
1262                         GOTO(out, ELDLM_OK);
1263                 }
1264                 /* If no flags, fall through to normal enqueue path. */
1265         }
1266
1267         policy = ldlm_processing_policy_table[res->lr_type];
1268         policy(lock, flags, 1, &rc, NULL);
1269         GOTO(out, rc);
1270 out:
1271         unlock_res_and_lock(lock);
1272         if (node)
1273                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1274         return rc;
1275 }
1276
1277 /* Must be called with namespace taken: queue is waiting or converting. */
1278 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1279                          struct list_head *work_list)
1280 {
1281         struct list_head *tmp, *pos;
1282         ldlm_processing_policy policy;
1283         int flags;
1284         int rc = LDLM_ITER_CONTINUE;
1285         ldlm_error_t err;
1286         ENTRY;
1287
1288         check_res_locked(res);
1289
1290         policy = ldlm_processing_policy_table[res->lr_type];
1291         LASSERT(policy);
1292
1293         list_for_each_safe(tmp, pos, queue) {
1294                 struct ldlm_lock *pending;
1295                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1296
1297                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1298
1299                 flags = 0;
1300                 rc = policy(pending, &flags, 0, &err, work_list);
1301                 if (rc != LDLM_ITER_CONTINUE)
1302                         break;
1303         }
1304
1305         RETURN(rc);
1306 }
1307
1308 /* Helper function for pair ldlm_run_{bl,cp}_ast_work().
1309  *
1310  * Send an existing rpc set specified by @arg->set and then
1311  * destroy it. Create new one if @do_create flag is set. */
1312 static void
1313 ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create)
1314 {
1315         int rc;
1316
1317         rc = ptlrpc_set_wait(arg->set);
1318         if (arg->type == LDLM_BL_CALLBACK)
1319                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2);
1320         ptlrpc_set_destroy(arg->set);
1321
1322         if (do_create)
1323                 arg->set = ptlrpc_prep_set();
1324 }
1325
1326 int ldlm_run_bl_ast_work(struct list_head *rpc_list)
1327 {
1328         struct ldlm_cb_set_arg arg;
1329         struct list_head *tmp, *pos;
1330         struct ldlm_lock_desc d;
1331         int ast_count;
1332         int rc = 0;
1333         ENTRY;
1334
1335         arg.set = ptlrpc_prep_set();
1336         atomic_set(&arg.restart, 0);
1337         arg.type = LDLM_BL_CALLBACK;
1338
1339         ast_count = 0;
1340         list_for_each_safe(tmp, pos, rpc_list) {
1341                 struct ldlm_lock *lock =
1342                         list_entry(tmp, struct ldlm_lock, l_bl_ast);
1343
1344                 /* nobody should touch l_bl_ast */
1345                 lock_res_and_lock(lock);
1346                 list_del_init(&lock->l_bl_ast);
1347
1348                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1349                 LASSERT(lock->l_bl_ast_run == 0);
1350                 LASSERT(lock->l_blocking_lock);
1351                 lock->l_bl_ast_run++;
1352                 unlock_res_and_lock(lock);
1353
1354                 ldlm_lock2desc(lock->l_blocking_lock, &d);
1355
1356                 LDLM_LOCK_PUT(lock->l_blocking_lock);
1357                 lock->l_blocking_lock = NULL;
1358                 rc = lock->l_blocking_ast(lock, &d, (void *)&arg,
1359                                           LDLM_CB_BLOCKING);
1360                 LDLM_LOCK_PUT(lock);
1361                 ast_count++;
1362
1363                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1364                  * and create a new set for requests that remained in
1365                  * @rpc_list */
1366                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1367                         ldlm_send_and_maybe_create_set(&arg, 1);
1368                         ast_count = 0;
1369                 }
1370         }
1371
1372         if (ast_count > 0)
1373                 ldlm_send_and_maybe_create_set(&arg, 0);
1374         else
1375                 /* In case when number of ASTs is multiply of
1376                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1377                  * @arg.set must be destroyed here, otherwise we get
1378                  * write memory leaking. */
1379                 ptlrpc_set_destroy(arg.set);
1380
1381         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1382 }
1383
1384 int ldlm_run_cp_ast_work(struct list_head *rpc_list)
1385 {
1386         struct ldlm_cb_set_arg arg;
1387         struct list_head *tmp, *pos;
1388         int ast_count;
1389         int rc = 0;
1390         ENTRY;
1391
1392         arg.set = ptlrpc_prep_set();
1393         atomic_set(&arg.restart, 0);
1394         arg.type = LDLM_CP_CALLBACK;
1395
1396         /* It's possible to receive a completion AST before we've set
1397          * the l_completion_ast pointer: either because the AST arrived
1398          * before the reply, or simply because there's a small race
1399          * window between receiving the reply and finishing the local
1400          * enqueue. (bug 842)
1401          *
1402          * This can't happen with the blocking_ast, however, because we
1403          * will never call the local blocking_ast until we drop our
1404          * reader/writer reference, which we won't do until we get the
1405          * reply and finish enqueueing. */
1406
1407         ast_count = 0;
1408         list_for_each_safe(tmp, pos, rpc_list) {
1409                 struct ldlm_lock *lock =
1410                         list_entry(tmp, struct ldlm_lock, l_cp_ast);
1411                 ldlm_completion_callback completion_callback;
1412
1413                 /* nobody should touch l_cp_ast */
1414                 lock_res_and_lock(lock);
1415                 list_del_init(&lock->l_cp_ast);
1416                 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1417                 /* save l_completion_ast since it can be changed by
1418                  * mds_intent_policy(), see bug 14225 */
1419                 completion_callback = lock->l_completion_ast;
1420                 lock->l_flags &= ~LDLM_FL_CP_REQD;
1421                 unlock_res_and_lock(lock);
1422
1423                 if (completion_callback != NULL) {
1424                         rc = completion_callback(lock, 0, (void *)&arg);
1425                         ast_count++;
1426                 }
1427                 LDLM_LOCK_PUT(lock);
1428
1429                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1430                  * and create a new set for requests that remained in
1431                  * @rpc_list */
1432                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1433                         ldlm_send_and_maybe_create_set(&arg, 1);
1434                         ast_count = 0;
1435                 }
1436         }
1437
1438         if (ast_count > 0)
1439                 ldlm_send_and_maybe_create_set(&arg, 0);
1440         else
1441                 /* In case when number of ASTs is multiply of
1442                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1443                  * @arg.set must be destroyed here, otherwise we get
1444                  * write memory leaking. */
1445                 ptlrpc_set_destroy(arg.set);
1446
1447         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1448 }
1449
1450 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1451 {
1452         ldlm_reprocess_all(res);
1453         return LDLM_ITER_CONTINUE;
1454 }
1455
1456 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1457 {
1458         struct list_head *tmp;
1459         int i, rc;
1460
1461         ENTRY;
1462         spin_lock(&ns->ns_hash_lock);
1463         for (i = 0; i < RES_HASH_SIZE; i++) {
1464                 tmp = ns->ns_hash[i].next;
1465                 while (tmp != &(ns->ns_hash[i])) {
1466                         struct ldlm_resource *res =
1467                                 list_entry(tmp, struct ldlm_resource, lr_hash);
1468
1469                         ldlm_resource_getref(res);
1470                         spin_unlock(&ns->ns_hash_lock);
1471
1472                         rc = reprocess_one_queue(res, NULL);
1473
1474                         spin_lock(&ns->ns_hash_lock);
1475                         tmp = tmp->next;
1476                         ldlm_resource_putref_locked(res);
1477
1478                         if (rc == LDLM_ITER_STOP)
1479                                 GOTO(out, rc);
1480                 }
1481         }
1482  out:
1483         spin_unlock(&ns->ns_hash_lock);
1484         EXIT;
1485 }
1486
1487 void ldlm_reprocess_all(struct ldlm_resource *res)
1488 {
1489         CFS_LIST_HEAD(rpc_list);
1490         int rc;
1491         ENTRY;
1492
1493         /* Local lock trees don't get reprocessed. */
1494         if (ns_is_client(res->lr_namespace)) {
1495                 EXIT;
1496                 return;
1497         }
1498
1499  restart:
1500         lock_res(res);
1501         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1502         if (rc == LDLM_ITER_CONTINUE)
1503                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1504         unlock_res(res);
1505
1506         rc = ldlm_run_cp_ast_work(&rpc_list);
1507         if (rc == -ERESTART) {
1508                 LASSERT(list_empty(&rpc_list));
1509                 goto restart;
1510         }
1511         EXIT;
1512 }
1513
1514 void ldlm_cancel_callback(struct ldlm_lock *lock)
1515 {
1516         check_res_locked(lock->l_resource);
1517         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1518                 lock->l_flags |= LDLM_FL_CANCEL;
1519                 if (lock->l_blocking_ast) {
1520                         // l_check_no_ns_lock(ns);
1521                         unlock_res_and_lock(lock);
1522                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1523                                              LDLM_CB_CANCELING);
1524                         lock_res_and_lock(lock);
1525                 } else {
1526                         LDLM_DEBUG(lock, "no blocking ast");
1527                 }
1528         }
1529         lock->l_flags |= LDLM_FL_BL_DONE;
1530 }
1531
1532 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1533 {
1534         if (req->l_resource->lr_type != LDLM_PLAIN &&
1535             req->l_resource->lr_type != LDLM_IBITS)
1536                 return;
1537
1538         list_del_init(&req->l_sl_policy);
1539         list_del_init(&req->l_sl_mode);
1540 }
1541
1542 void ldlm_lock_cancel(struct ldlm_lock *lock)
1543 {
1544         struct ldlm_resource *res;
1545         struct ldlm_namespace *ns;
1546         ENTRY;
1547
1548         lock_res_and_lock(lock);
1549
1550         res = lock->l_resource;
1551         ns = res->lr_namespace;
1552
1553         /* Please do not, no matter how tempting, remove this LBUG without
1554          * talking to me first. -phik */
1555         if (lock->l_readers || lock->l_writers) {
1556                 LDLM_ERROR(lock, "lock still has references");
1557                 LBUG();
1558         }
1559
1560         ldlm_del_waiting_lock(lock);
1561
1562         /* Releases res lock */
1563         ldlm_cancel_callback(lock);
1564
1565         /* Yes, second time, just in case it was added again while we were
1566            running with no res lock in ldlm_cancel_callback */
1567         ldlm_del_waiting_lock(lock);
1568         ldlm_resource_unlink_lock(lock);
1569         ldlm_lock_destroy_nolock(lock);
1570
1571         if (lock->l_granted_mode == lock->l_req_mode)
1572                 ldlm_pool_del(&ns->ns_pool, lock);
1573
1574         /* Make sure we will not be called again for same lock what is possible
1575          * if not to zero out lock->l_granted_mode */
1576         lock->l_granted_mode = 0;
1577         unlock_res_and_lock(lock);
1578
1579         EXIT;
1580 }
1581
1582 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1583 {
1584         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1585         ENTRY;
1586
1587         if (lock == NULL)
1588                 RETURN(-EINVAL);
1589
1590         lock->l_ast_data = data;
1591         LDLM_LOCK_PUT(lock);
1592         RETURN(0);
1593 }
1594
1595 void ldlm_cancel_locks_for_export_cb(void *obj, void *data)
1596 {
1597         struct obd_export     *exp = data;
1598         struct ldlm_lock      *lock = obj;
1599         struct ldlm_resource  *res;
1600
1601         res = ldlm_resource_getref(lock->l_resource);
1602         LDLM_LOCK_GET(lock);
1603
1604         LDLM_DEBUG(lock, "export %p", exp);
1605         ldlm_res_lvbo_update(res, NULL, 0, 1);
1606
1607         ldlm_lock_cancel(lock);
1608         ldlm_reprocess_all(res);
1609
1610         ldlm_resource_putref(res);
1611         LDLM_LOCK_PUT(lock);
1612 }
1613
1614 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1615 {
1616         lustre_hash_for_each_empty(exp->exp_lock_hash,
1617                                    ldlm_cancel_locks_for_export_cb, exp);
1618 }
1619
1620 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1621                                         __u32 *flags)
1622 {
1623         CFS_LIST_HEAD(rpc_list);
1624         struct ldlm_resource *res;
1625         struct ldlm_namespace *ns;
1626         int granted = 0;
1627         int old_mode, rc;
1628         struct sl_insert_point prev;
1629         ldlm_error_t err;
1630         struct ldlm_interval *node;
1631         ENTRY;
1632
1633         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1634                 *flags |= LDLM_FL_BLOCK_GRANTED;
1635                 RETURN(lock->l_resource);
1636         }
1637
1638         /* I can't check the type of lock here because the bitlock of lock
1639          * is not held here, so do the allocation blindly. -jay */
1640         OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO, sizeof(*node));
1641         if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
1642                 RETURN(NULL);
1643
1644         LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
1645                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1646
1647         lock_res_and_lock(lock);
1648
1649         res = lock->l_resource;
1650         ns = res->lr_namespace;
1651
1652         old_mode = lock->l_req_mode;
1653         lock->l_req_mode = new_mode;
1654         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1655                 /* remember the lock position where the lock might be
1656                  * added back to the granted list later and also
1657                  * remember the join mode for skiplist fixing. */
1658                 prev.res_link = lock->l_res_link.prev;
1659                 prev.mode_link = lock->l_sl_mode.prev;
1660                 prev.policy_link = lock->l_sl_policy.prev;
1661                 ldlm_resource_unlink_lock(lock);
1662         } else {
1663                 ldlm_resource_unlink_lock(lock);
1664                 if (res->lr_type == LDLM_EXTENT) {
1665                         /* FIXME: ugly code, I have to attach the lock to a
1666                          * interval node again since perhaps it will be granted
1667                          * soon */
1668                         CFS_INIT_LIST_HEAD(&node->li_group);
1669                         ldlm_interval_attach(node, lock);
1670                         node = NULL;
1671                 }
1672         }
1673
1674         /* If this is a local resource, put it on the appropriate list. */
1675         if (ns_is_client(res->lr_namespace)) {
1676                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1677                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1678                 } else {
1679                         /* This should never happen, because of the way the
1680                          * server handles conversions. */
1681                         LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1682                                    *flags);
1683                         LBUG();
1684
1685                         ldlm_grant_lock(lock, &rpc_list);
1686                         granted = 1;
1687                         /* FIXME: completion handling not with ns_lock held ! */
1688                         if (lock->l_completion_ast)
1689                                 lock->l_completion_ast(lock, 0, NULL);
1690                 }
1691         } else {
1692                 int pflags = 0;
1693                 ldlm_processing_policy policy;
1694                 policy = ldlm_processing_policy_table[res->lr_type];
1695                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1696                 if (rc == LDLM_ITER_STOP) {
1697                         lock->l_req_mode = old_mode;
1698                         if (res->lr_type == LDLM_EXTENT)
1699                                 ldlm_extent_add_lock(res, lock);
1700                         else
1701                                 ldlm_granted_list_add_lock(lock, &prev);
1702
1703                         res = NULL;
1704                 } else {
1705                         *flags |= LDLM_FL_BLOCK_GRANTED;
1706                         granted = 1;
1707                 }
1708         }
1709         unlock_res_and_lock(lock);
1710
1711         if (granted)
1712                 ldlm_run_cp_ast_work(&rpc_list);
1713         if (node)
1714                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1715         RETURN(res);
1716 }
1717
1718 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1719 {
1720         struct obd_device *obd = NULL;
1721
1722         if (!((libcfs_debug | D_ERROR) & level))
1723                 return;
1724
1725         if (!lock) {
1726                 CDEBUG(level, "  NULL LDLM lock\n");
1727                 return;
1728         }
1729
1730         CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1731                lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1732                pos, lock->l_pid);
1733         if (lock->l_conn_export != NULL)
1734                 obd = lock->l_conn_export->exp_obd;
1735         if (lock->l_export && lock->l_export->exp_connection) {
1736                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1737                      libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1738                      lock->l_remote_handle.cookie);
1739         } else if (obd == NULL) {
1740                 CDEBUG(level, "  Node: local\n");
1741         } else {
1742                 struct obd_import *imp = obd->u.cli.cl_import;
1743                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1744                        libcfs_nid2str(imp->imp_connection->c_peer.nid),
1745                        lock->l_remote_handle.cookie);
1746         }
1747         CDEBUG(level, "  Resource: %p ("LPU64"/"LPU64")\n", lock->l_resource,
1748                lock->l_resource->lr_name.name[0],
1749                lock->l_resource->lr_name.name[1]);
1750         CDEBUG(level, "  Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1751                "write: %d flags: "LPX64"\n", ldlm_lockname[lock->l_req_mode],
1752                ldlm_lockname[lock->l_granted_mode],
1753                atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1754                lock->l_flags);
1755         if (lock->l_resource->lr_type == LDLM_EXTENT)
1756                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64
1757                        " (req "LPU64"-"LPU64")\n",
1758                        lock->l_policy_data.l_extent.start,
1759                        lock->l_policy_data.l_extent.end,
1760                        lock->l_req_extent.start, lock->l_req_extent.end);
1761         else if (lock->l_resource->lr_type == LDLM_FLOCK)
1762                 CDEBUG(level, "  Pid: %d Extent: "LPU64" -> "LPU64"\n",
1763                        lock->l_policy_data.l_flock.pid,
1764                        lock->l_policy_data.l_flock.start,
1765                        lock->l_policy_data.l_flock.end);
1766        else if (lock->l_resource->lr_type == LDLM_IBITS)
1767                 CDEBUG(level, "  Bits: "LPX64"\n",
1768                        lock->l_policy_data.l_inodebits.bits);
1769 }
1770
1771 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1772 {
1773         struct ldlm_lock *lock;
1774
1775         if (!((libcfs_debug | D_ERROR) & level))
1776                 return;
1777
1778         lock = ldlm_handle2lock(lockh);
1779         if (lock == NULL)
1780                 return;
1781
1782         ldlm_lock_dump(D_OTHER, lock, 0);
1783
1784         LDLM_LOCK_PUT(lock);
1785 }
1786
1787 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1788                       struct libcfs_debug_msg_data *data, const char *fmt,
1789                       ...)
1790 {
1791         va_list args;
1792         cfs_debug_limit_state_t *cdls = data->msg_cdls;
1793
1794         va_start(args, fmt);
1795         if (lock->l_resource == NULL) {
1796                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1797                                    data->msg_fn, data->msg_line, fmt, args,
1798                                    " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1799                                    "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" remote: "
1800                                    LPX64" expref: %d pid: %u timeout: %lu\n", lock,
1801                                    lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1802                                    lock->l_readers, lock->l_writers,
1803                                    ldlm_lockname[lock->l_granted_mode],
1804                                    ldlm_lockname[lock->l_req_mode],
1805                                    lock->l_flags, lock->l_remote_handle.cookie,
1806                                    lock->l_export ?
1807                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1808                                    lock->l_pid, lock->l_callback_timeout);
1809                 va_end(args);
1810                 return;
1811         }
1812
1813         switch (lock->l_resource->lr_type) {
1814         case LDLM_EXTENT:
1815                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1816                                    data->msg_fn, data->msg_line, fmt, args,
1817                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1818                                    "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1819                                    "] (req "LPU64"->"LPU64") flags: "LPX64" remote: "LPX64
1820                                     " expref: %d pid: %u timeout %lu\n",
1821                                     lock->l_resource->lr_namespace->ns_name, lock,
1822                                     lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1823                                     lock->l_readers, lock->l_writers,
1824                                     ldlm_lockname[lock->l_granted_mode],
1825                                     ldlm_lockname[lock->l_req_mode],
1826                                     lock->l_resource->lr_name.name[0],
1827                                     lock->l_resource->lr_name.name[1],
1828                                     atomic_read(&lock->l_resource->lr_refcount),
1829                                     ldlm_typename[lock->l_resource->lr_type],
1830                                     lock->l_policy_data.l_extent.start,
1831                                     lock->l_policy_data.l_extent.end,
1832                                     lock->l_req_extent.start, lock->l_req_extent.end,
1833                                     lock->l_flags, lock->l_remote_handle.cookie,
1834                                     lock->l_export ?
1835                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1836                                     lock->l_pid, lock->l_callback_timeout);
1837                 break;
1838         case LDLM_FLOCK:
1839                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1840                                    data->msg_fn, data->msg_line, fmt, args,
1841                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1842                                    "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1843                                    "["LPU64"->"LPU64"] flags: "LPX64" remote: "LPX64
1844                                    " expref: %d pid: %u timeout: %lu\n",
1845                                    lock->l_resource->lr_namespace->ns_name, lock,
1846                                    lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1847                                    lock->l_readers, lock->l_writers,
1848                                    ldlm_lockname[lock->l_granted_mode],
1849                                    ldlm_lockname[lock->l_req_mode],
1850                                    lock->l_resource->lr_name.name[0],
1851                                    lock->l_resource->lr_name.name[1],
1852                                    atomic_read(&lock->l_resource->lr_refcount),
1853                                    ldlm_typename[lock->l_resource->lr_type],
1854                                    lock->l_policy_data.l_flock.pid,
1855                                    lock->l_policy_data.l_flock.start,
1856                                    lock->l_policy_data.l_flock.end,
1857                                    lock->l_flags, lock->l_remote_handle.cookie,
1858                                    lock->l_export ?
1859                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1860                                    lock->l_pid, lock->l_callback_timeout);
1861                 break;
1862         case LDLM_IBITS:
1863                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1864                                    data->msg_fn, data->msg_line, fmt, args,
1865                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1866                                    "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1867                                    "flags: "LPX64" remote: "LPX64" expref: %d "
1868                                    "pid: %u timeout: %lu\n",
1869                                    lock->l_resource->lr_namespace->ns_name,
1870                                    lock, lock->l_handle.h_cookie,
1871                                    atomic_read (&lock->l_refc),
1872                                    lock->l_readers, lock->l_writers,
1873                                    ldlm_lockname[lock->l_granted_mode],
1874                                    ldlm_lockname[lock->l_req_mode],
1875                                    lock->l_resource->lr_name.name[0],
1876                                    lock->l_resource->lr_name.name[1],
1877                                    lock->l_policy_data.l_inodebits.bits,
1878                                    atomic_read(&lock->l_resource->lr_refcount),
1879                                    ldlm_typename[lock->l_resource->lr_type],
1880                                    lock->l_flags, lock->l_remote_handle.cookie,
1881                                    lock->l_export ?
1882                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1883                                    lock->l_pid, lock->l_callback_timeout);
1884                 break;
1885         default:
1886                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1887                                    data->msg_fn, data->msg_line, fmt, args,
1888                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1889                                    "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
1890                                    "remote: "LPX64" expref: %d pid: %u timeout %lu\n",
1891                                    lock->l_resource->lr_namespace->ns_name,
1892                                    lock, lock->l_handle.h_cookie,
1893                                    atomic_read (&lock->l_refc),
1894                                    lock->l_readers, lock->l_writers,
1895                                    ldlm_lockname[lock->l_granted_mode],
1896                                    ldlm_lockname[lock->l_req_mode],
1897                                    lock->l_resource->lr_name.name[0],
1898                                    lock->l_resource->lr_name.name[1],
1899                                    atomic_read(&lock->l_resource->lr_refcount),
1900                                    ldlm_typename[lock->l_resource->lr_type],
1901                                    lock->l_flags, lock->l_remote_handle.cookie,
1902                                    lock->l_export ?
1903                                          atomic_read(&lock->l_export->exp_refcount) : -99,
1904                                    lock->l_pid, lock->l_callback_timeout);
1905                 break;
1906         }
1907         va_end(args);
1908 }
1909 EXPORT_SYMBOL(_ldlm_lock_debug);