lustre/ldlm/ldlm_lock.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   5  *   Author: Peter Braam <braam@clusterfs.com>
   6  *   Author: Phil Schwan <phil@clusterfs.com>
   7  *
   8  *   This file is part of the Lustre file system, http://www.lustre.org
   9  *   Lustre is a trademark of Cluster File Systems, Inc.
  10  *
  11  *   You may have signed or agreed to another license before downloading
  12  *   this software.  If so, you are bound by the terms and conditions
  13  *   of that agreement, and the following does not apply to you.  See the
  14  *   LICENSE file included with this distribution for more information.
  15  *
  16  *   If you did not agree to a different license, then this copy of Lustre
  17  *   is open source software; you can redistribute it and/or modify it
  18  *   under the terms of version 2 of the GNU General Public License as
  19  *   published by the Free Software Foundation.
  20  *
  21  *   In either case, Lustre is distributed in the hope that it will be
  22  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  23  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  24  *   license text for more details.
  25  */
  26
  27 #define DEBUG_SUBSYSTEM S_LDLM
  28
  29 #ifdef __KERNEL__
  30 # include <libcfs/libcfs.h>
  31 # include <linux/lustre_intent.h>
  32 #else
  33 # include <liblustre.h>
  34 # include <libcfs/kp30.h>
  35 #endif
  36
  37 #include <obd_class.h>
  38 #include "ldlm_internal.h"
  39
  40 //struct lustre_lock ldlm_everything_lock;
  41
  42 /* lock's skip list pointers fix mode */
  43 #define LDLM_JOIN_NONE          0
  44 #define LDLM_MODE_JOIN_RIGHT    1
  45 #define LDLM_MODE_JOIN_LEFT     (1 << 1)
  46 #define LDLM_POLICY_JOIN_RIGHT  (1 << 2)
  47 #define LDLM_POLICY_JOIN_LEFT   (1 << 3)
  48
  49 /* lock types */
  50 char *ldlm_lockname[] = {
  51         [0] "--",
  52         [LCK_EX] "EX",
  53         [LCK_PW] "PW",
  54         [LCK_PR] "PR",
  55         [LCK_CW] "CW",
  56         [LCK_CR] "CR",
  57         [LCK_NL] "NL",
  58         [LCK_GROUP] "GROUP"
  59 };
  60
  61 char *ldlm_typename[] = {
  62         [LDLM_PLAIN] "PLN",
  63         [LDLM_EXTENT] "EXT",
  64         [LDLM_FLOCK] "FLK",
  65         [LDLM_IBITS] "IBT",
  66 };
  67
  68 char *ldlm_it2str(int it)
  69 {
  70         switch (it) {
  71         case IT_OPEN:
  72                 return "open";
  73         case IT_CREAT:
  74                 return "creat";
  75         case (IT_OPEN | IT_CREAT):
  76                 return "open|creat";
  77         case IT_READDIR:
  78                 return "readdir";
  79         case IT_GETATTR:
  80                 return "getattr";
  81         case IT_LOOKUP:
  82                 return "lookup";
  83         case IT_UNLINK:
  84                 return "unlink";
  85         case IT_GETXATTR:
  86                 return "getxattr";
  87         default:
  88                 CERROR("Unknown intent %d\n", it);
  89                 return "UNKNOWN";
  90         }
  91 }
  92
  93 extern cfs_mem_cache_t *ldlm_lock_slab;
  94
  95 static ldlm_processing_policy ldlm_processing_policy_table[] = {
  96         [LDLM_PLAIN] ldlm_process_plain_lock,
  97         [LDLM_EXTENT] ldlm_process_extent_lock,
  98 #ifdef __KERNEL__
  99         [LDLM_FLOCK] ldlm_process_flock_lock,
 100 #endif
 101         [LDLM_IBITS] ldlm_process_inodebits_lock,
 102 };
 103
 104 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
 105 {
 106         return ldlm_processing_policy_table[res->lr_type];
 107 }
 108
 109 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
 110 {
 111         ns->ns_policy = arg;
 112 }
 113
 114 /*
 115  * REFCOUNTED LOCK OBJECTS
 116  */
 117
 118
 119 /*
 120  * Lock refcounts, during creation:
 121  *   - one special one for allocation, dec'd only once in destroy
 122  *   - one for being a lock that's in-use
 123  *   - one for the addref associated with a new lock
 124  */
 125 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 126 {
 127         atomic_inc(&lock->l_refc);
 128         return lock;
 129 }
 130
 131 static void ldlm_lock_free(struct ldlm_lock *lock, size_t size)
 132 {
 133         LASSERT(size == sizeof(*lock));
 134         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
 135 }
 136
 137 void ldlm_lock_put(struct ldlm_lock *lock)
 138 {
 139         ENTRY;
 140
 141         LASSERT(lock->l_resource != LP_POISON);
 142         LASSERT(atomic_read(&lock->l_refc) > 0);
 143         if (atomic_dec_and_test(&lock->l_refc)) {
 144                 struct ldlm_resource *res;
 145
 146                 LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing it.");
 147
 148                 res = lock->l_resource;
 149                 LASSERT(lock->l_destroyed);
 150                 LASSERT(list_empty(&lock->l_res_link));
 151                 LASSERT(list_empty(&lock->l_pending_chain));
 152
 153                 atomic_dec(&res->lr_namespace->ns_locks);
 154                 ldlm_resource_putref(res);
 155                 lock->l_resource = NULL;
 156                 if (lock->l_export) {
 157                         class_export_put(lock->l_export);
 158                         lock->l_export = NULL;
 159                 }
 160
 161                 if (lock->l_lvb_data != NULL)
 162                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
 163
 164                 OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
 165                                 ldlm_lock_free);
 166         }
 167
 168         EXIT;
 169 }
 170
 171 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
 172 {
 173         int rc = 0;
 174         if (!list_empty(&lock->l_lru)) {
 175                 struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 176                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 177                 list_del_init(&lock->l_lru);
 178                 ns->ns_nr_unused--;
 179                 LASSERT(ns->ns_nr_unused >= 0);
 180                 rc = 1;
 181         }
 182         return rc;
 183 }
 184
 185 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 186 {
 187         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 188         int rc;
 189         ENTRY;
 190         spin_lock(&ns->ns_unused_lock);
 191         rc = ldlm_lock_remove_from_lru_nolock(lock);
 192         spin_unlock(&ns->ns_unused_lock);
 193         EXIT;
 194         return rc;
 195 }
 196
 197 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
 198 {
 199         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 200         lock->l_last_used = cfs_time_current();
 201         LASSERT(list_empty(&lock->l_lru));
 202         list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 203         LASSERT(ns->ns_nr_unused >= 0);
 204         ns->ns_nr_unused++;
 205 }
 206
 207 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
 208 {
 209         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 210         ENTRY;
 211         spin_lock(&ns->ns_unused_lock);
 212         ldlm_lock_add_to_lru_nolock(lock);
 213         spin_unlock(&ns->ns_unused_lock);
 214         EXIT;
 215 }
 216
 217 void ldlm_lock_touch_in_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         if (!list_empty(&lock->l_lru)) {
 223                 ldlm_lock_remove_from_lru_nolock(lock);
 224                 ldlm_lock_add_to_lru_nolock(lock);
 225         }
 226         spin_unlock(&ns->ns_unused_lock);
 227         EXIT;
 228 }
 229
 230 /* This used to have a 'strict' flag, which recovery would use to mark an
 231  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 232  * shall explain why it's gone: with the new hash table scheme, once you call
 233  * ldlm_lock_destroy, you can never drop your final references on this lock.
 234  * Because it's not in the hash table anymore.  -phil */
 235 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
 236 {
 237         ENTRY;
 238
 239         if (lock->l_readers || lock->l_writers) {
 240                 LDLM_ERROR(lock, "lock still has references");
 241                 ldlm_lock_dump(D_ERROR, lock, 0);
 242                 LBUG();
 243         }
 244
 245         if (!list_empty(&lock->l_res_link)) {
 246                 LDLM_ERROR(lock, "lock still on resource");
 247                 ldlm_lock_dump(D_ERROR, lock, 0);
 248                 LBUG();
 249         }
 250
 251         if (lock->l_destroyed) {
 252                 LASSERT(list_empty(&lock->l_lru));
 253                 EXIT;
 254                 return 0;
 255         }
 256         lock->l_destroyed = 1;
 257
 258         if (lock->l_export)
 259                 spin_lock(&lock->l_export->exp_ldlm_data.led_lock);
 260         list_del_init(&lock->l_export_chain);
 261         if (lock->l_export)
 262                 spin_unlock(&lock->l_export->exp_ldlm_data.led_lock);
 263
 264         ldlm_lock_remove_from_lru(lock);
 265         class_handle_unhash(&lock->l_handle);
 266
 267 #if 0
 268         /* Wake anyone waiting for this lock */
 269         /* FIXME: I should probably add yet another flag, instead of using
 270          * l_export to only call this on clients */
 271         if (lock->l_export)
 272                 class_export_put(lock->l_export);
 273         lock->l_export = NULL;
 274         if (lock->l_export && lock->l_completion_ast)
 275                 lock->l_completion_ast(lock, 0);
 276 #endif
 277         EXIT;
 278         return 1;
 279 }
 280
 281 void ldlm_lock_destroy(struct ldlm_lock *lock)
 282 {
 283         int first;
 284         ENTRY;
 285         lock_res_and_lock(lock);
 286         first = ldlm_lock_destroy_internal(lock);
 287         unlock_res_and_lock(lock);
 288
 289         /* drop reference from hashtable only for first destroy */
 290         if (first)
 291                 LDLM_LOCK_PUT(lock);
 292         EXIT;
 293 }
 294
 295 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
 296 {
 297         int first;
 298         ENTRY;
 299         first = ldlm_lock_destroy_internal(lock);
 300         /* drop reference from hashtable only for first destroy */
 301         if (first)
 302                 LDLM_LOCK_PUT(lock);
 303         EXIT;
 304 }
 305
 306 /* this is called by portals_handle2object with the handle lock taken */
 307 static void lock_handle_addref(void *lock)
 308 {
 309         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 310 }
 311
 312 /*
 313  * usage: pass in a resource on which you have done ldlm_resource_get
 314  *        pass in a parent lock on which you have done a ldlm_lock_get
 315  *        after return, ldlm_*_put the resource and parent
 316  * returns: lock with refcount 2 - one for current caller and one for remote
 317  */
 318 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
 319 {
 320         struct ldlm_lock *lock;
 321         ENTRY;
 322
 323         if (resource == NULL)
 324                 LBUG();
 325
 326         OBD_SLAB_ALLOC(lock, ldlm_lock_slab, CFS_ALLOC_IO, sizeof(*lock));
 327         if (lock == NULL)
 328                 RETURN(NULL);
 329
 330         spin_lock_init(&lock->l_lock);
 331         lock->l_resource = ldlm_resource_getref(resource);
 332
 333         atomic_set(&lock->l_refc, 2);
 334         CFS_INIT_LIST_HEAD(&lock->l_res_link);
 335         CFS_INIT_LIST_HEAD(&lock->l_lru);
 336         CFS_INIT_LIST_HEAD(&lock->l_export_chain);
 337         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
 338         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
 339         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
 340         cfs_waitq_init(&lock->l_waitq);
 341         lock->l_blocking_lock = NULL;
 342         lock->l_sl_mode.prev = NULL;
 343         lock->l_sl_mode.next = NULL;
 344         lock->l_sl_policy.prev = NULL;
 345         lock->l_sl_policy.next = NULL;
 346
 347         atomic_inc(&resource->lr_namespace->ns_locks);
 348         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
 349         class_handle_hash(&lock->l_handle, lock_handle_addref);
 350
 351         CFS_INIT_LIST_HEAD(&lock->l_extents_list);
 352         spin_lock_init(&lock->l_extents_list_lock);
 353         CFS_INIT_LIST_HEAD(&lock->l_cache_locks_list);
 354
 355         RETURN(lock);
 356 }
 357
 358 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 359                               struct ldlm_res_id new_resid)
 360 {
 361         struct ldlm_resource *oldres = lock->l_resource;
 362         struct ldlm_resource *newres;
 363         int type;
 364         ENTRY;
 365
 366         LASSERT(ns_is_client(ns));
 367
 368         lock_res_and_lock(lock);
 369         if (memcmp(&new_resid, &lock->l_resource->lr_name,
 370                    sizeof(lock->l_resource->lr_name)) == 0) {
 371                 /* Nothing to do */
 372                 unlock_res_and_lock(lock);
 373                 RETURN(0);
 374         }
 375
 376         LASSERT(new_resid.name[0] != 0);
 377
 378         /* This function assumes that the lock isn't on any lists */
 379         LASSERT(list_empty(&lock->l_res_link));
 380
 381         type = oldres->lr_type;
 382         unlock_res_and_lock(lock);
 383
 384         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 385         if (newres == NULL) {
 386                 LBUG();
 387                 RETURN(-ENOMEM);
 388         }
 389
 390         lock_res_and_lock(lock);
 391         LASSERT(memcmp(&new_resid, &lock->l_resource->lr_name,
 392                        sizeof(lock->l_resource->lr_name)) != 0);
 393         lock_res(newres);
 394         lock->l_resource = newres;
 395         unlock_res(oldres);
 396         unlock_res_and_lock(lock);
 397
 398         /* ...and the flowers are still standing! */
 399         ldlm_resource_putref(oldres);
 400
 401         RETURN(0);
 402 }
 403
 404 /*
 405  *  HANDLES
 406  */
 407
 408 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 409 {
 410         lockh->cookie = lock->l_handle.h_cookie;
 411 }
 412
 413 /* if flags: atomically get the lock and set the flags.
 414  *           Return NULL if flag already set
 415  */
 416
 417 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *handle, int flags)
 418 {
 419         struct ldlm_namespace *ns;
 420         struct ldlm_lock *lock = NULL, *retval = NULL;
 421         ENTRY;
 422
 423         LASSERT(handle);
 424
 425         lock = class_handle2object(handle->cookie);
 426         if (lock == NULL)
 427                 RETURN(NULL);
 428
 429         LASSERT(lock->l_resource != NULL);
 430         ns = lock->l_resource->lr_namespace;
 431         LASSERT(ns != NULL);
 432
 433         lock_res_and_lock(lock);
 434
 435         /* It's unlikely but possible that someone marked the lock as
 436          * destroyed after we did handle2object on it */
 437         if (lock->l_destroyed) {
 438                 unlock_res_and_lock(lock);
 439                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 440                 LDLM_LOCK_PUT(lock);
 441                 GOTO(out, retval);
 442         }
 443
 444         if (flags && (lock->l_flags & flags)) {
 445                 unlock_res_and_lock(lock);
 446                 LDLM_LOCK_PUT(lock);
 447                 GOTO(out, retval);
 448         }
 449
 450         if (flags)
 451                 lock->l_flags |= flags;
 452
 453         unlock_res_and_lock(lock);
 454         retval = lock;
 455         EXIT;
 456  out:
 457         return retval;
 458 }
 459
 460 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
 461                                       struct lustre_handle *handle)
 462 {
 463         struct ldlm_lock *retval = NULL;
 464         retval = __ldlm_handle2lock(handle, 0);
 465         return retval;
 466 }
 467
 468 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 469 {
 470         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
 471         /* INODEBITS_INTEROP: If the other side does not support
 472          * inodebits, reply with a plain lock descriptor.
 473          */
 474         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
 475             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
 476                 struct ldlm_resource res = *lock->l_resource;
 477
 478                 /* Make sure all the right bits are set in this lock we
 479                    are going to pass to client */
 480                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
 481                          (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
 482                          "Inappropriate inode lock bits during "
 483                          "conversion " LPU64 "\n",
 484                          lock->l_policy_data.l_inodebits.bits);
 485                 res.lr_type = LDLM_PLAIN;
 486                 ldlm_res2desc(&res, &desc->l_resource);
 487                 /* Convert "new" lock mode to something old client can
 488                    understand */
 489                 if ((lock->l_req_mode == LCK_CR) ||
 490                     (lock->l_req_mode == LCK_CW))
 491                         desc->l_req_mode = LCK_PR;
 492                 else
 493                         desc->l_req_mode = lock->l_req_mode;
 494                 if ((lock->l_granted_mode == LCK_CR) ||
 495                     (lock->l_granted_mode == LCK_CW)) {
 496                         desc->l_granted_mode = LCK_PR;
 497                 } else {
 498                         /* We never grant PW/EX locks to clients */
 499                         LASSERT((lock->l_granted_mode != LCK_PW) &&
 500                                 (lock->l_granted_mode != LCK_EX));
 501                         desc->l_granted_mode = lock->l_granted_mode;
 502                 }
 503
 504                 /* We do not copy policy here, because there is no
 505                    policy for plain locks */
 506         } else {
 507                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 508                 desc->l_req_mode = lock->l_req_mode;
 509                 desc->l_granted_mode = lock->l_granted_mode;
 510                 desc->l_policy_data = lock->l_policy_data;
 511         }
 512 }
 513
 514 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 515                            struct list_head *work_list)
 516 {
 517         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
 518                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 519                 lock->l_flags |= LDLM_FL_AST_SENT;
 520                 /* If the enqueuing client said so, tell the AST recipient to
 521                  * discard dirty data, rather than writing back. */
 522                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 523                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 524                 LASSERT(list_empty(&lock->l_bl_ast));
 525                 list_add(&lock->l_bl_ast, work_list);
 526                 LDLM_LOCK_GET(lock);
 527                 LASSERT(lock->l_blocking_lock == NULL);
 528                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
 529         }
 530 }
 531
 532 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
 533 {
 534         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
 535                 lock->l_flags |= LDLM_FL_CP_REQD;
 536                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
 537                 LASSERT(list_empty(&lock->l_cp_ast));
 538                 list_add(&lock->l_cp_ast, work_list);
 539                 LDLM_LOCK_GET(lock);
 540         }
 541 }
 542
 543 /* must be called with lr_lock held */
 544 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 545                                 struct list_head *work_list)
 546 {
 547         ENTRY;
 548         check_res_locked(lock->l_resource);
 549         if (new)
 550                 ldlm_add_bl_work_item(lock, new, work_list);
 551         else
 552                 ldlm_add_cp_work_item(lock, work_list);
 553         EXIT;
 554 }
 555
 556 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 557 {
 558         struct ldlm_lock *lock;
 559
 560         lock = ldlm_handle2lock(lockh);
 561         LASSERT(lock != NULL);
 562         ldlm_lock_addref_internal(lock, mode);
 563         LDLM_LOCK_PUT(lock);
 564 }
 565
 566 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 567 {
 568         ldlm_lock_remove_from_lru(lock);
 569         if (mode & (LCK_NL | LCK_CR | LCK_PR))
 570                 lock->l_readers++;
 571         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
 572                 lock->l_writers++;
 573         LDLM_LOCK_GET(lock);
 574         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 575 }
 576
 577 /* only called for local locks */
 578 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 579 {
 580         lock_res_and_lock(lock);
 581         ldlm_lock_addref_internal_nolock(lock, mode);
 582         unlock_res_and_lock(lock);
 583 }
 584
 585 /* only called in ldlm_flock_destroy and for local locks.
 586  * for LDLM_FLOCK type locks, l_blocking_ast is null, and
 587  * ldlm_lock_remove_from_lru() does nothing, it is safe
 588  * for ldlm_flock_destroy usage by dropping some code */
 589 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 590 {
 591         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 592         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 593                 LASSERT(lock->l_readers > 0);
 594                 lock->l_readers--;
 595         }
 596         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
 597                 LASSERT(lock->l_writers > 0);
 598                 lock->l_writers--;
 599         }
 600
 601         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 602 }
 603
 604 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 605 {
 606         struct ldlm_namespace *ns;
 607         ENTRY;
 608
 609         lock_res_and_lock(lock);
 610
 611         ns = lock->l_resource->lr_namespace;
 612
 613         ldlm_lock_decref_internal_nolock(lock, mode);
 614
 615         if (lock->l_flags & LDLM_FL_LOCAL &&
 616             !lock->l_readers && !lock->l_writers) {
 617                 /* If this is a local lock on a server namespace and this was
 618                  * the last reference, cancel the lock. */
 619                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 620                 lock->l_flags |= LDLM_FL_CBPENDING;
 621         }
 622
 623         if (!lock->l_readers && !lock->l_writers &&
 624             (lock->l_flags & LDLM_FL_CBPENDING)) {
 625                 /* If we received a blocked AST and this was the last reference,
 626                  * run the callback. */
 627                 if (ns_is_server(ns) && lock->l_export)
 628                         CERROR("FL_CBPENDING set on non-local lock--just a "
 629                                "warning\n");
 630
 631                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 632
 633                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
 634                 ldlm_lock_remove_from_lru(lock);
 635                 unlock_res_and_lock(lock);
 636                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
 637                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
 638                         ldlm_handle_bl_callback(ns, NULL, lock);
 639         } else if (ns_is_client(ns) &&
 640                    !lock->l_readers && !lock->l_writers &&
 641                    !(lock->l_flags & LDLM_FL_NO_LRU)) {
 642                 /* If this is a client-side namespace and this was the last
 643                  * reference, put it on the LRU. */
 644                 ldlm_lock_add_to_lru(lock);
 645                 unlock_res_and_lock(lock);
 646                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
 647                  * are not supported by the server, otherwise, it is done on
 648                  * enqueue. */
 649                 if (!exp_connect_cancelset(lock->l_conn_export) &&
 650                     !ns_connect_lru_resize(ns))
 651                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
 652         } else {
 653                 unlock_res_and_lock(lock);
 654         }
 655
 656         EXIT;
 657 }
 658
 659 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 660 {
 661         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 662         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
 663         ldlm_lock_decref_internal(lock, mode);
 664         LDLM_LOCK_PUT(lock);
 665 }
 666
 667 /* This will drop a lock reference and mark it for destruction, but will not
 668  * necessarily cancel the lock before returning. */
 669 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 670 {
 671         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 672         ENTRY;
 673
 674         LASSERT(lock != NULL);
 675
 676         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 677         lock_res_and_lock(lock);
 678         lock->l_flags |= LDLM_FL_CBPENDING;
 679         unlock_res_and_lock(lock);
 680         ldlm_lock_decref_internal(lock, mode);
 681         LDLM_LOCK_PUT(lock);
 682 }
 683
 684 /*
 685  * search_granted_lock
 686  *
 687  * Description:
 688  *      Finds a position to insert the new lock.
 689  * Parameters:
 690  *      queue [input]:  the granted list where search acts on;
 691  *      req [input]:    the lock whose position to be located;
 692  *      lockp [output]: the position where the lock should be inserted before, or
 693  *                      NULL indicating @req should be appended to @queue.
 694  * Return Values:
 695  *      Bit-masks combination of following values indicating in which way the
 696  *      lock need to be inserted.
 697  *      - LDLM_JOIN_NONE:       noting about skip list needs to be fixed;
 698  *      - LDLM_MODE_JOIN_RIGHT: @req needs join right becoming the head of a
 699  *                              mode group;
 700  *      - LDLM_POLICY_JOIN_RIGHT: @req needs join right becoming the head of
 701  *                                a policy group.
 702  * NOTE: called by
 703  *  - ldlm_grant_lock_with_skiplist
 704  */
 705 static int search_granted_lock(struct list_head *queue,
 706                         struct ldlm_lock *req,
 707                         struct ldlm_lock **lockp)
 708 {
 709         struct list_head *tmp, *tmp_tail;
 710         struct ldlm_lock *lock, *mode_head_lock;
 711         int rc = LDLM_JOIN_NONE;
 712         ENTRY;
 713
 714         list_for_each(tmp, queue) {
 715                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 716
 717                 if (lock->l_req_mode != req->l_req_mode) {
 718                         if (LDLM_SL_HEAD(&lock->l_sl_mode))
 719                                 tmp = &list_entry(lock->l_sl_mode.next,
 720                                                   struct ldlm_lock,
 721                                                   l_sl_mode)->l_res_link;
 722                         continue;
 723                 }
 724
 725                 /* found the same mode group */
 726                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
 727                         *lockp = lock;
 728                         rc = LDLM_MODE_JOIN_RIGHT;
 729                         GOTO(out, rc);
 730                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
 731                         tmp_tail = tmp;
 732                         if (LDLM_SL_HEAD(&lock->l_sl_mode))
 733                                 tmp_tail = &list_entry(lock->l_sl_mode.next,
 734                                                        struct ldlm_lock,
 735                                                        l_sl_mode)->l_res_link;
 736                         mode_head_lock = lock;
 737                         for (;;) {
 738                                 if (lock->l_policy_data.l_inodebits.bits ==
 739                                     req->l_policy_data.l_inodebits.bits) {
 740                                         /* matched policy lock is found */
 741                                         *lockp = lock;
 742                                         rc |= LDLM_POLICY_JOIN_RIGHT;
 743
 744                                         /* if the policy group head is also a
 745                                          * mode group head or a single mode
 746                                          * group lock */
 747                                         if (LDLM_SL_HEAD(&lock->l_sl_mode) ||
 748                                             (tmp == tmp_tail &&
 749                                              LDLM_SL_EMPTY(&lock->l_sl_mode)))
 750                                                 rc |= LDLM_MODE_JOIN_RIGHT;
 751                                         GOTO(out, rc);
 752                                 }
 753
 754                                 if (LDLM_SL_HEAD(&lock->l_sl_policy))
 755                                         tmp = &list_entry(lock->l_sl_policy.next,
 756                                                           struct ldlm_lock,
 757                                                           l_sl_policy)->l_res_link;
 758
 759                                 if (tmp == tmp_tail)
 760                                         break;
 761                                 else
 762                                         tmp = tmp->next;
 763                                 lock = list_entry(tmp, struct ldlm_lock,
 764                                                   l_res_link);
 765                         }  /* for all locks in the matched mode group */
 766
 767                         /* no matched policy group is found, insert before
 768                          * the mode group head lock */
 769                         *lockp = mode_head_lock;
 770                         rc = LDLM_MODE_JOIN_RIGHT;
 771                         GOTO(out, rc);
 772                 } else {
 773                         LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
 774                         LBUG();
 775                 }
 776         }
 777
 778         /* no matched mode group is found, append to the end */
 779         *lockp = NULL;
 780         rc = LDLM_JOIN_NONE;
 781         EXIT;
 782 out:
 783         return rc;
 784 }
 785
 786 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
 787                                        struct ldlm_lock *lockp,
 788                                        int join)
 789 {
 790         struct ldlm_resource *res = lock->l_resource;
 791         ENTRY;
 792
 793         LASSERT(lockp || join == LDLM_JOIN_NONE);
 794
 795         check_res_locked(res);
 796
 797         ldlm_resource_dump(D_OTHER, res);
 798         CDEBUG(D_OTHER, "About to add this lock:\n");
 799         ldlm_lock_dump(D_OTHER, lock, 0);
 800
 801         if (lock->l_destroyed) {
 802                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 803                 return;
 804         }
 805
 806         LASSERT(list_empty(&lock->l_res_link));
 807
 808         if (!lockp)
 809                 list_add_tail(&lock->l_res_link, &lock->l_resource->lr_granted);
 810         else if ((join & LDLM_MODE_JOIN_LEFT) || (join & LDLM_POLICY_JOIN_LEFT))
 811                 list_add(&lock->l_res_link, &lockp->l_res_link);
 812         else
 813                 list_add_tail(&lock->l_res_link, &lockp->l_res_link);
 814
 815         /* fix skip lists */
 816         if (join & LDLM_MODE_JOIN_RIGHT) {
 817                 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_mode));
 818                 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
 819                         lock->l_sl_mode.next = &lockp->l_sl_mode;
 820                         lockp->l_sl_mode.prev = &lock->l_sl_mode;
 821                 } else if (LDLM_SL_HEAD(&lockp->l_sl_mode)) {
 822                         lock->l_sl_mode.next = lockp->l_sl_mode.next;
 823                         lockp->l_sl_mode.next = NULL;
 824                         lock->l_sl_mode.next->prev = &lock->l_sl_mode;
 825                 }
 826         } else if (join & LDLM_MODE_JOIN_LEFT) {
 827                 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_mode));
 828                 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
 829                         lock->l_sl_mode.prev = &lockp->l_sl_mode;
 830                         lockp->l_sl_mode.next = &lock->l_sl_mode;
 831                 } else if (LDLM_SL_TAIL(&lockp->l_sl_mode)) {
 832                         lock->l_sl_mode.prev = lockp->l_sl_mode.prev;
 833                         lockp->l_sl_mode.prev = NULL;
 834                         lock->l_sl_mode.prev->next = &lock->l_sl_mode;
 835                 }
 836         }
 837
 838         if (join & LDLM_POLICY_JOIN_RIGHT) {
 839                 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_policy));
 840                 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
 841                         lock->l_sl_policy.next = &lockp->l_sl_policy;
 842                         lockp->l_sl_policy.prev = &lock->l_sl_policy;
 843                 } else if (LDLM_SL_HEAD(&lockp->l_sl_policy)) {
 844                         lock->l_sl_policy.next = lockp->l_sl_policy.next;
 845                         lockp->l_sl_policy.next = NULL;
 846                         lock->l_sl_policy.next->prev = &lock->l_sl_policy;
 847                 }
 848         } else if (join & LDLM_POLICY_JOIN_LEFT) {
 849                 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_policy));
 850                 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
 851                         lock->l_sl_policy.prev = &lockp->l_sl_policy;
 852                         lockp->l_sl_policy.next = &lock->l_sl_policy;
 853                 } else if (LDLM_SL_TAIL(&lockp->l_sl_policy)) {
 854                         lock->l_sl_policy.prev = lockp->l_sl_policy.prev;
 855                         lockp->l_sl_policy.prev = NULL;
 856                         lock->l_sl_policy.prev->next = &lock->l_sl_policy;
 857                 }
 858         }
 859
 860         EXIT;
 861 }
 862
 863 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
 864 {
 865         int join = LDLM_JOIN_NONE;
 866         struct ldlm_lock *lockp = NULL;
 867         ENTRY;
 868
 869         LASSERT(lock->l_req_mode == lock->l_granted_mode);
 870
 871         join = search_granted_lock(&lock->l_resource->lr_granted, lock, &lockp);
 872         ldlm_granted_list_add_lock(lock, lockp, join);
 873         EXIT;
 874 }
 875
 876 /* NOTE: called by
 877  *  - ldlm_lock_enqueue
 878  *  - ldlm_reprocess_queue
 879  *  - ldlm_lock_convert
 880  *
 881  * must be called with lr_lock held
 882  */
 883 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
 884 {
 885         struct ldlm_resource *res = lock->l_resource;
 886         ENTRY;
 887
 888         check_res_locked(res);
 889
 890         lock->l_granted_mode = lock->l_req_mode;
 891         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
 892                 ldlm_grant_lock_with_skiplist(lock);
 893         else
 894                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
 895
 896         if (lock->l_granted_mode < res->lr_most_restr)
 897                 res->lr_most_restr = lock->l_granted_mode;
 898
 899         if (work_list && lock->l_completion_ast != NULL)
 900                 ldlm_add_ast_work_item(lock, NULL, work_list);
 901
 902         ldlm_pool_add(&res->lr_namespace->ns_pool, lock);
 903         EXIT;
 904 }
 905
 906 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
 907  * comment above ldlm_lock_match */
 908 static struct ldlm_lock *search_queue(struct list_head *queue,
 909                                       ldlm_mode_t *mode,
 910                                       ldlm_policy_data_t *policy,
 911                                       struct ldlm_lock *old_lock, int flags)
 912 {
 913         struct ldlm_lock *lock;
 914         struct list_head *tmp;
 915
 916         list_for_each(tmp, queue) {
 917                 ldlm_mode_t match;
 918
 919                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 920
 921                 if (lock == old_lock)
 922                         continue;
 923
 924                 /* llite sometimes wants to match locks that will be
 925                  * canceled when their users drop, but we allow it to match
 926                  * if it passes in CBPENDING and the lock still has users.
 927                  * this is generally only going to be used by children
 928                  * whose parents already hold a lock so forward progress
 929                  * can still happen. */
 930                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 931                     !(flags & LDLM_FL_CBPENDING))
 932                         continue;
 933                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 934                     lock->l_readers == 0 && lock->l_writers == 0)
 935                         continue;
 936
 937                 if (!(lock->l_req_mode & *mode))
 938                         continue;
 939
 940                 match = lock->l_req_mode;
 941                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 942                     (lock->l_policy_data.l_extent.start >
 943                      policy->l_extent.start ||
 944                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
 945                         continue;
 946
 947                 if (unlikely(match == LCK_GROUP) &&
 948                     lock->l_resource->lr_type == LDLM_EXTENT &&
 949                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
 950                         continue;
 951
 952                 /* We match if we have existing lock with same or wider set
 953                    of bits. */
 954                 if (lock->l_resource->lr_type == LDLM_IBITS &&
 955                      ((lock->l_policy_data.l_inodebits.bits &
 956                       policy->l_inodebits.bits) !=
 957                       policy->l_inodebits.bits))
 958                         continue;
 959
 960                 if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
 961                         continue;
 962
 963                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
 964                     !(lock->l_flags & LDLM_FL_LOCAL))
 965                         continue;
 966
 967                 if (flags & LDLM_FL_TEST_LOCK) {
 968                         LDLM_LOCK_GET(lock);
 969                         ldlm_lock_touch_in_lru(lock);
 970                 } else {
 971                         ldlm_lock_addref_internal_nolock(lock, match);
 972                 }
 973                 *mode = match;
 974                 return lock;
 975         }
 976
 977         return NULL;
 978 }
 979
 980 void ldlm_lock_allow_match(struct ldlm_lock *lock)
 981 {
 982         lock_res_and_lock(lock);
 983         lock->l_flags |= LDLM_FL_LVB_READY;
 984         cfs_waitq_signal(&lock->l_waitq);
 985         unlock_res_and_lock(lock);
 986 }
 987
 988 /* Can be called in two ways:
 989  *
 990  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 991  * for a duplicate of.
 992  *
 993  * Otherwise, all of the fields must be filled in, to match against.
 994  *
 995  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
 996  *     server (ie, connh is NULL)
 997  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
 998  *     list will be considered
 999  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1000  *     to be canceled can still be matched as long as they still have reader
1001  *     or writer refernces
1002  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1003  *     just tell us if we would have matched.
1004  *
1005  * Returns 1 if it finds an already-existing lock that is compatible; in this
1006  * case, lockh is filled in with a addref()ed lock
1007  */
1008 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1009                             struct ldlm_res_id *res_id, ldlm_type_t type,
1010                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
1011                             struct lustre_handle *lockh)
1012 {
1013         struct ldlm_resource *res;
1014         struct ldlm_lock *lock, *old_lock = NULL;
1015         int rc = 0;
1016         ENTRY;
1017
1018         if (ns == NULL) {
1019                 old_lock = ldlm_handle2lock(lockh);
1020                 LASSERT(old_lock);
1021
1022                 ns = old_lock->l_resource->lr_namespace;
1023                 res_id = &old_lock->l_resource->lr_name;
1024                 type = old_lock->l_resource->lr_type;
1025                 mode = old_lock->l_req_mode;
1026         }
1027
1028         res = ldlm_resource_get(ns, NULL, *res_id, type, 0);
1029         if (res == NULL) {
1030                 LASSERT(old_lock == NULL);
1031                 RETURN(0);
1032         }
1033
1034         lock_res(res);
1035
1036         lock = search_queue(&res->lr_granted, &mode, policy, old_lock, flags);
1037         if (lock != NULL)
1038                 GOTO(out, rc = 1);
1039         if (flags & LDLM_FL_BLOCK_GRANTED)
1040                 GOTO(out, rc = 0);
1041         lock = search_queue(&res->lr_converting, &mode, policy, old_lock, flags);
1042         if (lock != NULL)
1043                 GOTO(out, rc = 1);
1044         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock, flags);
1045         if (lock != NULL)
1046                 GOTO(out, rc = 1);
1047
1048         EXIT;
1049  out:
1050         unlock_res(res);
1051         ldlm_resource_putref(res);
1052
1053         if (lock) {
1054                 ldlm_lock2handle(lock, lockh);
1055                 if ((flags & LDLM_FL_LVB_READY) && (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1056                         struct l_wait_info lwi;
1057                         if (lock->l_completion_ast) {
1058                                 int err = lock->l_completion_ast(lock,
1059                                                           LDLM_FL_WAIT_NOREPROC,
1060                                                                  NULL);
1061                                 if (err) {
1062                                         if (flags & LDLM_FL_TEST_LOCK)
1063                                                 LDLM_LOCK_PUT(lock);
1064                                         else
1065                                                 ldlm_lock_decref_internal(lock, mode);
1066                                         rc = 0;
1067                                         goto out2;
1068                                 }
1069                         }
1070
1071                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
1072                                                LWI_ON_SIGNAL_NOOP, NULL);
1073
1074                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1075                         l_wait_event(lock->l_waitq,
1076                                      (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1077                 }
1078         }
1079  out2:
1080         if (rc) {
1081                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1082                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1083                                 res_id->name[2] : policy->l_extent.start,
1084                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1085                                 res_id->name[3] : policy->l_extent.end);
1086         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1087                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1088                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1089                                   type, mode, res_id->name[0], res_id->name[1],
1090                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1091                                         res_id->name[2] :policy->l_extent.start,
1092                                 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1093                                         res_id->name[3] : policy->l_extent.end);
1094         }
1095         if (old_lock)
1096                 LDLM_LOCK_PUT(old_lock);
1097         if (flags & LDLM_FL_TEST_LOCK && rc)
1098                 LDLM_LOCK_PUT(lock);
1099
1100         return rc ? mode : 0;
1101 }
1102
1103 /* Returns a referenced lock */
1104 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1105                                    struct ldlm_res_id res_id, ldlm_type_t type,
1106                                    ldlm_mode_t mode,
1107                                    ldlm_blocking_callback blocking,
1108                                    ldlm_completion_callback completion,
1109                                    ldlm_glimpse_callback glimpse,
1110                                    void *data, __u32 lvb_len)
1111 {
1112         struct ldlm_lock *lock;
1113         struct ldlm_resource *res;
1114         ENTRY;
1115
1116         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1117         if (res == NULL)
1118                 RETURN(NULL);
1119
1120         lock = ldlm_lock_new(res);
1121         ldlm_resource_putref(res);
1122
1123         if (lock == NULL)
1124                 RETURN(NULL);
1125
1126         lock->l_req_mode = mode;
1127         lock->l_ast_data = data;
1128         lock->l_blocking_ast = blocking;
1129         lock->l_completion_ast = completion;
1130         lock->l_glimpse_ast = glimpse;
1131         lock->l_pid = cfs_curproc_pid();
1132
1133         if (lvb_len) {
1134                 lock->l_lvb_len = lvb_len;
1135                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1136                 if (lock->l_lvb_data == NULL) {
1137                         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1138                         RETURN(NULL);
1139                 }
1140         }
1141
1142         RETURN(lock);
1143 }
1144
1145 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1146                                struct ldlm_lock **lockp,
1147                                void *cookie, int *flags)
1148 {
1149         struct ldlm_lock *lock = *lockp;
1150         struct ldlm_resource *res = lock->l_resource;
1151         int local = ns_is_client(res->lr_namespace);
1152         ldlm_processing_policy policy;
1153         ldlm_error_t rc = ELDLM_OK;
1154         ENTRY;
1155
1156         do_gettimeofday(&lock->l_enqueued_time);
1157         /* policies are not executed on the client or during replay */
1158         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1159             && !local && ns->ns_policy) {
1160                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1161                                    NULL);
1162                 if (rc == ELDLM_LOCK_REPLACED) {
1163                         /* The lock that was returned has already been granted,
1164                          * and placed into lockp.  If it's not the same as the
1165                          * one we passed in, then destroy the old one and our
1166                          * work here is done. */
1167                         if (lock != *lockp) {
1168                                 ldlm_lock_destroy(lock);
1169                                 LDLM_LOCK_PUT(lock);
1170                         }
1171                         *flags |= LDLM_FL_LOCK_CHANGED;
1172                         RETURN(0);
1173                 } else if (rc != ELDLM_OK ||
1174                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1175                         ldlm_lock_destroy(lock);
1176                         RETURN(rc);
1177                 }
1178         }
1179
1180         lock_res_and_lock(lock);
1181         if (local && lock->l_req_mode == lock->l_granted_mode) {
1182                 /* The server returned a blocked lock, but it was granted before
1183                  * we got a chance to actually enqueue it.  We don't need to do
1184                  * anything else. */
1185                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1186                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1187                 GOTO(out, ELDLM_OK);
1188         }
1189
1190         /* Some flags from the enqueue want to make it into the AST, via the
1191          * lock's l_flags. */
1192         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1193
1194         /* This distinction between local lock trees is very important; a client
1195          * namespace only has information about locks taken by that client, and
1196          * thus doesn't have enough information to decide for itself if it can
1197          * be granted (below).  In this case, we do exactly what the server
1198          * tells us to do, as dictated by the 'flags'.
1199          *
1200          * We do exactly the same thing during recovery, when the server is
1201          * more or less trusting the clients not to lie.
1202          *
1203          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1204          * granted/converting queues. */
1205         ldlm_resource_unlink_lock(lock);
1206         if (local) {
1207                 if (*flags & LDLM_FL_BLOCK_CONV)
1208                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1209                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1210                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1211                 else
1212                         ldlm_grant_lock(lock, NULL);
1213                 GOTO(out, ELDLM_OK);
1214         } else if (*flags & LDLM_FL_REPLAY) {
1215                 if (*flags & LDLM_FL_BLOCK_CONV) {
1216                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1217                         GOTO(out, ELDLM_OK);
1218                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1219                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1220                         GOTO(out, ELDLM_OK);
1221                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1222                         ldlm_grant_lock(lock, NULL);
1223                         GOTO(out, ELDLM_OK);
1224                 }
1225                 /* If no flags, fall through to normal enqueue path. */
1226         }
1227
1228         policy = ldlm_processing_policy_table[res->lr_type];
1229         policy(lock, flags, 1, &rc, NULL);
1230         GOTO(out, rc);
1231 out:
1232         unlock_res_and_lock(lock);
1233         return rc;
1234 }
1235
1236 /* Must be called with namespace taken: queue is waiting or converting. */
1237 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1238                          struct list_head *work_list)
1239 {
1240         struct list_head *tmp, *pos;
1241         ldlm_processing_policy policy;
1242         int flags;
1243         int rc = LDLM_ITER_CONTINUE;
1244         ldlm_error_t err;
1245         ENTRY;
1246
1247         check_res_locked(res);
1248
1249         policy = ldlm_processing_policy_table[res->lr_type];
1250         LASSERT(policy);
1251
1252         list_for_each_safe(tmp, pos, queue) {
1253                 struct ldlm_lock *pending;
1254                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1255
1256                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1257
1258                 flags = 0;
1259                 rc = policy(pending, &flags, 0, &err, work_list);
1260                 if (rc != LDLM_ITER_CONTINUE)
1261                         break;
1262         }
1263
1264         RETURN(rc);
1265 }
1266
1267 /* Helper function for pair ldlm_run_{bl,cp}_ast_work().
1268  *
1269  * Send an existing rpc set specified by @arg->set and then
1270  * destroy it. Create new one if @do_create flag is set. */
1271 static void
1272 ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create)
1273 {
1274         int rc;
1275
1276         rc = ptlrpc_set_wait(arg->set);
1277         if (arg->type == LDLM_BL_CALLBACK)
1278                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2);
1279         ptlrpc_set_destroy(arg->set);
1280
1281         if (do_create)
1282                 arg->set = ptlrpc_prep_set();
1283 }
1284
1285 int ldlm_run_bl_ast_work(struct list_head *rpc_list)
1286 {
1287         struct ldlm_cb_set_arg arg;
1288         struct list_head *tmp, *pos;
1289         struct ldlm_lock_desc d;
1290         int ast_count;
1291         int rc = 0;
1292         ENTRY;
1293
1294         arg.set = ptlrpc_prep_set();
1295         atomic_set(&arg.restart, 0);
1296         arg.type = LDLM_BL_CALLBACK;
1297
1298         ast_count = 0;
1299         list_for_each_safe(tmp, pos, rpc_list) {
1300                 struct ldlm_lock *lock =
1301                         list_entry(tmp, struct ldlm_lock, l_bl_ast);
1302
1303                 /* nobody should touch l_bl_ast */
1304                 lock_res_and_lock(lock);
1305                 list_del_init(&lock->l_bl_ast);
1306
1307                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1308                 LASSERT(lock->l_bl_ast_run == 0);
1309                 LASSERT(lock->l_blocking_lock);
1310                 lock->l_bl_ast_run++;
1311                 unlock_res_and_lock(lock);
1312
1313                 ldlm_lock2desc(lock->l_blocking_lock, &d);
1314
1315                 LDLM_LOCK_PUT(lock->l_blocking_lock);
1316                 lock->l_blocking_lock = NULL;
1317                 rc = lock->l_blocking_ast(lock, &d, (void *)&arg,
1318                                           LDLM_CB_BLOCKING);
1319                 LDLM_LOCK_PUT(lock);
1320                 ast_count++;
1321
1322                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1323                  * and create a new set for requests that remained in
1324                  * @rpc_list */
1325                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1326                         ldlm_send_and_maybe_create_set(&arg, 1);
1327                         ast_count = 0;
1328                 }
1329         }
1330
1331         if (ast_count > 0)
1332                 ldlm_send_and_maybe_create_set(&arg, 0);
1333         else
1334                 /* In case when number of ASTs is multiply of
1335                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1336                  * @arg.set must be destroyed here, otherwise we get
1337                  * write memory leaking. */
1338                 ptlrpc_set_destroy(arg.set);
1339
1340         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1341 }
1342
1343 int ldlm_run_cp_ast_work(struct list_head *rpc_list)
1344 {
1345         struct ldlm_cb_set_arg arg;
1346         struct list_head *tmp, *pos;
1347         int ast_count;
1348         int rc = 0;
1349         ENTRY;
1350
1351         arg.set = ptlrpc_prep_set();
1352         atomic_set(&arg.restart, 0);
1353         arg.type = LDLM_CP_CALLBACK;
1354
1355         /* It's possible to receive a completion AST before we've set
1356          * the l_completion_ast pointer: either because the AST arrived
1357          * before the reply, or simply because there's a small race
1358          * window between receiving the reply and finishing the local
1359          * enqueue. (bug 842)
1360          *
1361          * This can't happen with the blocking_ast, however, because we
1362          * will never call the local blocking_ast until we drop our
1363          * reader/writer reference, which we won't do until we get the
1364          * reply and finish enqueueing. */
1365
1366         ast_count = 0;
1367         list_for_each_safe(tmp, pos, rpc_list) {
1368                 struct ldlm_lock *lock =
1369                         list_entry(tmp, struct ldlm_lock, l_cp_ast);
1370                 ldlm_completion_callback completion_callback;
1371
1372                 /* nobody should touch l_cp_ast */
1373                 lock_res_and_lock(lock);
1374                 list_del_init(&lock->l_cp_ast);
1375                 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1376                 /* save l_completion_ast since it can be changed by
1377                  * mds_intent_policy(), see bug 14225 */
1378                 completion_callback = lock->l_completion_ast;
1379                 lock->l_flags &= ~LDLM_FL_CP_REQD;
1380                 unlock_res_and_lock(lock);
1381
1382                 if (completion_callback != NULL) {
1383                         rc = completion_callback(lock, 0, (void *)&arg);
1384                         ast_count++;
1385                 }
1386                 LDLM_LOCK_PUT(lock);
1387
1388                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1389                  * and create a new set for requests that remained in
1390                  * @rpc_list */
1391                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1392                         ldlm_send_and_maybe_create_set(&arg, 1);
1393                         ast_count = 0;
1394                 }
1395         }
1396
1397         if (ast_count > 0)
1398                 ldlm_send_and_maybe_create_set(&arg, 0);
1399         else
1400                 /* In case when number of ASTs is multiply of
1401                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1402                  * @arg.set must be destroyed here, otherwise we get
1403                  * write memory leaking. */
1404                 ptlrpc_set_destroy(arg.set);
1405
1406         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1407 }
1408
1409 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1410 {
1411         ldlm_reprocess_all(res);
1412         return LDLM_ITER_CONTINUE;
1413 }
1414
1415 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1416 {
1417         struct list_head *tmp;
1418         int i, rc;
1419
1420         ENTRY;
1421         spin_lock(&ns->ns_hash_lock);
1422         for (i = 0; i < RES_HASH_SIZE; i++) {
1423                 tmp = ns->ns_hash[i].next;
1424                 while (tmp != &(ns->ns_hash[i])) {
1425                         struct ldlm_resource *res =
1426                                 list_entry(tmp, struct ldlm_resource, lr_hash);
1427
1428                         ldlm_resource_getref(res);
1429                         spin_unlock(&ns->ns_hash_lock);
1430
1431                         rc = reprocess_one_queue(res, NULL);
1432
1433                         spin_lock(&ns->ns_hash_lock);
1434                         tmp = tmp->next;
1435                         ldlm_resource_putref_locked(res);
1436
1437                         if (rc == LDLM_ITER_STOP)
1438                                 GOTO(out, rc);
1439                 }
1440         }
1441  out:
1442         spin_unlock(&ns->ns_hash_lock);
1443         EXIT;
1444 }
1445
1446 void ldlm_reprocess_all(struct ldlm_resource *res)
1447 {
1448         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1449         int rc;
1450         ENTRY;
1451
1452         /* Local lock trees don't get reprocessed. */
1453         if (ns_is_client(res->lr_namespace)) {
1454                 EXIT;
1455                 return;
1456         }
1457
1458  restart:
1459         lock_res(res);
1460         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1461         if (rc == LDLM_ITER_CONTINUE)
1462                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1463         unlock_res(res);
1464
1465         rc = ldlm_run_cp_ast_work(&rpc_list);
1466         if (rc == -ERESTART) {
1467                 LASSERT(list_empty(&rpc_list));
1468                 goto restart;
1469         }
1470         EXIT;
1471 }
1472
1473 void ldlm_cancel_callback(struct ldlm_lock *lock)
1474 {
1475         check_res_locked(lock->l_resource);
1476         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1477                 lock->l_flags |= LDLM_FL_CANCEL;
1478                 if (lock->l_blocking_ast) {
1479                         // l_check_no_ns_lock(ns);
1480                         unlock_res_and_lock(lock);
1481                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1482                                              LDLM_CB_CANCELING);
1483                         lock_res_and_lock(lock);
1484                 } else {
1485                         LDLM_DEBUG(lock, "no blocking ast");
1486                 }
1487         }
1488         lock->l_flags |= LDLM_FL_BL_DONE;
1489 }
1490
1491 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1492 {
1493         struct ldlm_lock *lock;
1494
1495         if (req->l_resource->lr_type != LDLM_PLAIN &&
1496             req->l_resource->lr_type != LDLM_IBITS)
1497                 return;
1498
1499         if (LDLM_SL_HEAD(&req->l_sl_mode)) {
1500                 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1501                                   l_res_link);
1502                 if (req->l_sl_mode.next == &lock->l_sl_mode) {
1503                         lock->l_sl_mode.prev = NULL;
1504                 } else {
1505                         lock->l_sl_mode.next = req->l_sl_mode.next;
1506                         lock->l_sl_mode.next->prev = &lock->l_sl_mode;
1507                 }
1508                 req->l_sl_mode.next = NULL;
1509         } else if (LDLM_SL_TAIL(&req->l_sl_mode)) {
1510                 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1511                                   l_res_link);
1512                 if (req->l_sl_mode.prev == &lock->l_sl_mode) {
1513                         lock->l_sl_mode.next = NULL;
1514                 } else {
1515                         lock->l_sl_mode.prev = req->l_sl_mode.prev;
1516                         lock->l_sl_mode.prev->next = &lock->l_sl_mode;
1517                 }
1518                 req->l_sl_mode.prev = NULL;
1519         }
1520
1521         if (LDLM_SL_HEAD(&req->l_sl_policy)) {
1522                 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1523                                   l_res_link);
1524                 if (req->l_sl_policy.next == &lock->l_sl_policy) {
1525                         lock->l_sl_policy.prev = NULL;
1526                 } else {
1527                         lock->l_sl_policy.next = req->l_sl_policy.next;
1528                         lock->l_sl_policy.next->prev = &lock->l_sl_policy;
1529                 }
1530                 req->l_sl_policy.next = NULL;
1531         } else if (LDLM_SL_TAIL(&req->l_sl_policy)) {
1532                 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1533                                   l_res_link);
1534                 if (req->l_sl_policy.prev == &lock->l_sl_policy) {
1535                         lock->l_sl_policy.next = NULL;
1536                 } else {
1537                         lock->l_sl_policy.prev = req->l_sl_policy.prev;
1538                         lock->l_sl_policy.prev->next = &lock->l_sl_policy;
1539                 }
1540                 req->l_sl_policy.prev = NULL;
1541         }
1542 }
1543
1544 void ldlm_lock_cancel(struct ldlm_lock *lock)
1545 {
1546         struct ldlm_resource *res;
1547         struct ldlm_namespace *ns;
1548         ENTRY;
1549
1550         lock_res_and_lock(lock);
1551
1552         res = lock->l_resource;
1553         ns = res->lr_namespace;
1554
1555         /* Please do not, no matter how tempting, remove this LBUG without
1556          * talking to me first. -phik */
1557         if (lock->l_readers || lock->l_writers) {
1558                 LDLM_ERROR(lock, "lock still has references");
1559                 LBUG();
1560         }
1561
1562         ldlm_del_waiting_lock(lock);
1563
1564         /* Releases res lock */
1565         ldlm_cancel_callback(lock);
1566
1567         /* Yes, second time, just in case it was added again while we were
1568            running with no res lock in ldlm_cancel_callback */
1569         ldlm_del_waiting_lock(lock);
1570         ldlm_resource_unlink_lock(lock);
1571         ldlm_lock_destroy_nolock(lock);
1572
1573         if (lock->l_granted_mode == lock->l_req_mode)
1574                 ldlm_pool_del(&ns->ns_pool, lock);
1575
1576         /* Make sure we will not be called again for same lock what is possible
1577          * if not to zero out lock->l_granted_mode */
1578         lock->l_granted_mode = 0;
1579         unlock_res_and_lock(lock);
1580
1581         EXIT;
1582 }
1583
1584 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1585 {
1586         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1587         ENTRY;
1588
1589         if (lock == NULL)
1590                 RETURN(-EINVAL);
1591
1592         lock->l_ast_data = data;
1593         LDLM_LOCK_PUT(lock);
1594         RETURN(0);
1595 }
1596
1597 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1598 {
1599         struct ldlm_lock *lock;
1600         struct ldlm_resource *res;
1601
1602         spin_lock(&exp->exp_ldlm_data.led_lock);
1603         while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
1604                 lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
1605                                   struct ldlm_lock, l_export_chain);
1606                 res = ldlm_resource_getref(lock->l_resource);
1607                 LDLM_LOCK_GET(lock);
1608                 spin_unlock(&exp->exp_ldlm_data.led_lock);
1609
1610                 LDLM_DEBUG(lock, "export %p", exp);
1611                 ldlm_res_lvbo_update(res, NULL, 0, 1);
1612
1613                 ldlm_lock_cancel(lock);
1614                 ldlm_reprocess_all(res);
1615
1616                 ldlm_resource_putref(res);
1617                 LDLM_LOCK_PUT(lock);
1618                 spin_lock(&exp->exp_ldlm_data.led_lock);
1619         }
1620         spin_unlock(&exp->exp_ldlm_data.led_lock);
1621 }
1622
1623 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1624                                         __u32 *flags)
1625 {
1626         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1627         struct ldlm_resource *res;
1628         struct ldlm_namespace *ns;
1629         int granted = 0;
1630         int old_mode, rc;
1631         struct ldlm_lock *mark_lock = NULL;
1632         int join= LDLM_JOIN_NONE;
1633         ldlm_error_t err;
1634         ENTRY;
1635
1636         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1637                 *flags |= LDLM_FL_BLOCK_GRANTED;
1638                 RETURN(lock->l_resource);
1639         }
1640
1641         LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
1642                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1643
1644         lock_res_and_lock(lock);
1645
1646         res = lock->l_resource;
1647         ns = res->lr_namespace;
1648
1649         old_mode = lock->l_req_mode;
1650         lock->l_req_mode = new_mode;
1651         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1652                 /* remember the lock position where the lock might be
1653                  * added back to the granted list later and also
1654                  * remember the join mode for skiplist fixing. */
1655                 if (LDLM_SL_HEAD(&lock->l_sl_mode))
1656                         join = LDLM_MODE_JOIN_RIGHT;
1657                 else if (LDLM_SL_TAIL(&lock->l_sl_mode))
1658                         join = LDLM_MODE_JOIN_LEFT;
1659                 if (LDLM_SL_HEAD(&lock->l_sl_policy))
1660                         join |= LDLM_POLICY_JOIN_RIGHT;
1661                 else if (LDLM_SL_TAIL(&lock->l_sl_policy))
1662                         join |= LDLM_POLICY_JOIN_LEFT;
1663
1664                 LASSERT(!((join & LDLM_MODE_JOIN_RIGHT) &&
1665                           (join & LDLM_POLICY_JOIN_LEFT)));
1666                 LASSERT(!((join & LDLM_MODE_JOIN_LEFT) &&
1667                           (join & LDLM_POLICY_JOIN_RIGHT)));
1668
1669                 if ((join & LDLM_MODE_JOIN_LEFT) ||
1670                     (join & LDLM_POLICY_JOIN_LEFT))
1671                         mark_lock = list_entry(lock->l_res_link.prev,
1672                                                struct ldlm_lock, l_res_link);
1673                 else if (lock->l_res_link.next != &res->lr_granted)
1674                         mark_lock = list_entry(lock->l_res_link.next,
1675                                                struct ldlm_lock, l_res_link);
1676         }
1677         ldlm_resource_unlink_lock(lock);
1678
1679         /* If this is a local resource, put it on the appropriate list. */
1680         if (ns_is_client(res->lr_namespace)) {
1681                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1682                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1683                 } else {
1684                         /* This should never happen, because of the way the
1685                          * server handles conversions. */
1686                         LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1687                                    *flags);
1688                         LBUG();
1689
1690                         ldlm_grant_lock(lock, &rpc_list);
1691                         granted = 1;
1692                         /* FIXME: completion handling not with ns_lock held ! */
1693                         if (lock->l_completion_ast)
1694                                 lock->l_completion_ast(lock, 0, NULL);
1695                 }
1696         } else {
1697                 int pflags = 0;
1698                 ldlm_processing_policy policy;
1699                 policy = ldlm_processing_policy_table[res->lr_type];
1700                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1701                 if (rc == LDLM_ITER_STOP) {
1702                         lock->l_req_mode = old_mode;
1703                         ldlm_granted_list_add_lock(lock, mark_lock, join);
1704                         res = NULL;
1705                 } else {
1706                         *flags |= LDLM_FL_BLOCK_GRANTED;
1707                         granted = 1;
1708                 }
1709         }
1710         unlock_res_and_lock(lock);
1711
1712         if (granted)
1713                 ldlm_run_cp_ast_work(&rpc_list);
1714         RETURN(res);
1715 }
1716
1717 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1718 {
1719         struct obd_device *obd = NULL;
1720
1721         if (!((libcfs_debug | D_ERROR) & level))
1722                 return;
1723
1724         if (!lock) {
1725                 CDEBUG(level, "  NULL LDLM lock\n");
1726                 return;
1727         }
1728
1729         CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1730                lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1731                pos, lock->l_pid);
1732         if (lock->l_conn_export != NULL)
1733                 obd = lock->l_conn_export->exp_obd;
1734         if (lock->l_export && lock->l_export->exp_connection) {
1735                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1736                      libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1737                      lock->l_remote_handle.cookie);
1738         } else if (obd == NULL) {
1739                 CDEBUG(level, "  Node: local\n");
1740         } else {
1741                 struct obd_import *imp = obd->u.cli.cl_import;
1742                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1743                        libcfs_nid2str(imp->imp_connection->c_peer.nid),
1744                        lock->l_remote_handle.cookie);
1745         }
1746         CDEBUG(level, "  Resource: %p ("LPU64"/"LPU64")\n", lock->l_resource,
1747                lock->l_resource->lr_name.name[0],
1748                lock->l_resource->lr_name.name[1]);
1749         CDEBUG(level, "  Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1750                "write: %d flags: %#x\n", ldlm_lockname[lock->l_req_mode],
1751                ldlm_lockname[lock->l_granted_mode],
1752                atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1753                lock->l_flags);
1754         if (lock->l_resource->lr_type == LDLM_EXTENT)
1755                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64
1756                        " (req "LPU64"-"LPU64")\n",
1757                        lock->l_policy_data.l_extent.start,
1758                        lock->l_policy_data.l_extent.end,
1759                        lock->l_req_extent.start, lock->l_req_extent.end);
1760         else if (lock->l_resource->lr_type == LDLM_FLOCK)
1761                 CDEBUG(level, "  Pid: %d Extent: "LPU64" -> "LPU64"\n",
1762                        lock->l_policy_data.l_flock.pid,
1763                        lock->l_policy_data.l_flock.start,
1764                        lock->l_policy_data.l_flock.end);
1765        else if (lock->l_resource->lr_type == LDLM_IBITS)
1766                 CDEBUG(level, "  Bits: "LPX64"\n",
1767                        lock->l_policy_data.l_inodebits.bits);
1768 }
1769
1770 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1771 {
1772         struct ldlm_lock *lock;
1773
1774         if (!((libcfs_debug | D_ERROR) & level))
1775                 return;
1776
1777         lock = ldlm_handle2lock(lockh);
1778         if (lock == NULL)
1779                 return;
1780
1781         ldlm_lock_dump(D_OTHER, lock, 0);
1782
1783         LDLM_LOCK_PUT(lock);
1784 }
1785
1786 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1787                       struct libcfs_debug_msg_data *data, const char *fmt,
1788                       ...)
1789 {
1790         va_list args;
1791         cfs_debug_limit_state_t *cdls = data->msg_cdls;
1792
1793         va_start(args, fmt);
1794         if (lock->l_resource == NULL) {
1795                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1796                                    data->msg_fn, data->msg_line, fmt, args,
1797                                    " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1798                                    "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: "
1799                                    LPX64" expref: %d pid: %u\n", lock,
1800                                    lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1801                                    lock->l_readers, lock->l_writers,
1802                                    ldlm_lockname[lock->l_granted_mode],
1803                                    ldlm_lockname[lock->l_req_mode],
1804                                    lock->l_flags, lock->l_remote_handle.cookie,
1805                                    lock->l_export ?
1806                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1807                                    lock->l_pid);
1808                 va_end(args);
1809                 return;
1810         }
1811
1812         switch (lock->l_resource->lr_type) {
1813         case LDLM_EXTENT:
1814                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1815                                    data->msg_fn, data->msg_line, fmt, args,
1816                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1817                                    "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1818                                    "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64
1819                                     " expref: %d pid: %u\n",
1820                                     lock->l_resource->lr_namespace->ns_name, lock,
1821                                     lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1822                                     lock->l_readers, lock->l_writers,
1823                                     ldlm_lockname[lock->l_granted_mode],
1824                                     ldlm_lockname[lock->l_req_mode],
1825                                     lock->l_resource->lr_name.name[0],
1826                                     lock->l_resource->lr_name.name[1],
1827                                     atomic_read(&lock->l_resource->lr_refcount),
1828                                     ldlm_typename[lock->l_resource->lr_type],
1829                                     lock->l_policy_data.l_extent.start,
1830                                     lock->l_policy_data.l_extent.end,
1831                                     lock->l_req_extent.start, lock->l_req_extent.end,
1832                                     lock->l_flags, lock->l_remote_handle.cookie,
1833                                     lock->l_export ?
1834                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1835                                     lock->l_pid);
1836                 break;
1837         case LDLM_FLOCK:
1838                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1839                                    data->msg_fn, data->msg_line, fmt, args,
1840                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1841                                    "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1842                                    "["LPU64"->"LPU64"] flags: %x remote: "LPX64
1843                                    " expref: %d pid: %u\n",
1844                                    lock->l_resource->lr_namespace->ns_name, lock,
1845                                    lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1846                                    lock->l_readers, lock->l_writers,
1847                                    ldlm_lockname[lock->l_granted_mode],
1848                                    ldlm_lockname[lock->l_req_mode],
1849                                    lock->l_resource->lr_name.name[0],
1850                                    lock->l_resource->lr_name.name[1],
1851                                    atomic_read(&lock->l_resource->lr_refcount),
1852                                    ldlm_typename[lock->l_resource->lr_type],
1853                                    lock->l_policy_data.l_flock.pid,
1854                                    lock->l_policy_data.l_flock.start,
1855                                    lock->l_policy_data.l_flock.end,
1856                                    lock->l_flags, lock->l_remote_handle.cookie,
1857                                    lock->l_export ?
1858                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1859                                    lock->l_pid);
1860                 break;
1861         case LDLM_IBITS:
1862                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1863                                    data->msg_fn, data->msg_line, fmt, args,
1864                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1865                                    "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1866                                    "flags: %x remote: "LPX64" expref: %d "
1867                                    "pid %u\n",
1868                                    lock->l_resource->lr_namespace->ns_name,
1869                                    lock, lock->l_handle.h_cookie,
1870                                    atomic_read (&lock->l_refc),
1871                                    lock->l_readers, lock->l_writers,
1872                                    ldlm_lockname[lock->l_granted_mode],
1873                                    ldlm_lockname[lock->l_req_mode],
1874                                    lock->l_resource->lr_name.name[0],
1875                                    lock->l_resource->lr_name.name[1],
1876                                    lock->l_policy_data.l_inodebits.bits,
1877                                    atomic_read(&lock->l_resource->lr_refcount),
1878                                    ldlm_typename[lock->l_resource->lr_type],
1879                                    lock->l_flags, lock->l_remote_handle.cookie,
1880                                    lock->l_export ?
1881                                         atomic_read(&lock->l_export->exp_refcount) : -99,
1882                                    lock->l_pid);
1883                 break;
1884         default:
1885                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1886                                    data->msg_fn, data->msg_line, fmt, args,
1887                                    " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1888                                    "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x "
1889                                    "remote: "LPX64" expref: %d pid: %u\n",
1890                                    lock->l_resource->lr_namespace->ns_name,
1891                                    lock, lock->l_handle.h_cookie,
1892                                    atomic_read (&lock->l_refc),
1893                                    lock->l_readers, lock->l_writers,
1894                                    ldlm_lockname[lock->l_granted_mode],
1895                                    ldlm_lockname[lock->l_req_mode],
1896                                    lock->l_resource->lr_name.name[0],
1897                                    lock->l_resource->lr_name.name[1],
1898                                    atomic_read(&lock->l_resource->lr_refcount),
1899                                    ldlm_typename[lock->l_resource->lr_type],
1900                                    lock->l_flags, lock->l_remote_handle.cookie,
1901                                    lock->l_export ?
1902                                          atomic_read(&lock->l_export->exp_refcount) : -99,
1903                                    lock->l_pid);
1904                 break;
1905         }
1906         va_end(args);
1907 }
1908 EXPORT_SYMBOL(_ldlm_lock_debug);