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 # ifndef HAVE_VFS_INTENT_PATCHES
  32 # include <linux/lustre_intent.h>
  33 # endif
  34 #else
  35 # include <liblustre.h>
  36 #endif
  37
  38 #include <obd_class.h>
  39 #include "ldlm_internal.h"
  40
  41 /* lock types */
  42 char *ldlm_lockname[] = {
  43         [0] "--",
  44         [LCK_EX] "EX",
  45         [LCK_PW] "PW",
  46         [LCK_PR] "PR",
  47         [LCK_CW] "CW",
  48         [LCK_CR] "CR",
  49         [LCK_NL] "NL",
  50         [LCK_GROUP] "GROUP"
  51 };
  52
  53 char *ldlm_typename[] = {
  54         [LDLM_PLAIN] "PLN",
  55         [LDLM_EXTENT] "EXT",
  56         [LDLM_FLOCK] "FLK",
  57         [LDLM_IBITS] "IBT",
  58 };
  59
  60 char *ldlm_it2str(int it)
  61 {
  62         switch (it) {
  63         case IT_OPEN:
  64                 return "open";
  65         case IT_CREAT:
  66                 return "creat";
  67         case (IT_OPEN | IT_CREAT):
  68                 return "open|creat";
  69         case IT_READDIR:
  70                 return "readdir";
  71         case IT_GETATTR:
  72                 return "getattr";
  73         case IT_LOOKUP:
  74                 return "lookup";
  75         case IT_UNLINK:
  76                 return "unlink";
  77         case IT_GETXATTR:
  78                 return "getxattr";
  79         default:
  80                 CERROR("Unknown intent %d\n", it);
  81                 return "UNKNOWN";
  82         }
  83 }
  84
  85 extern cfs_mem_cache_t *ldlm_lock_slab;
  86
  87 static ldlm_processing_policy ldlm_processing_policy_table[] = {
  88         [LDLM_PLAIN] ldlm_process_plain_lock,
  89         [LDLM_EXTENT] ldlm_process_extent_lock,
  90 #ifdef __KERNEL__
  91         [LDLM_FLOCK] ldlm_process_flock_lock,
  92 #endif
  93         [LDLM_IBITS] ldlm_process_inodebits_lock,
  94 };
  95
  96 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
  97 {
  98         return ldlm_processing_policy_table[res->lr_type];
  99 }
 100
 101 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
 102 {
 103         ns->ns_policy = arg;
 104 }
 105
 106 /*
 107  * REFCOUNTED LOCK OBJECTS
 108  */
 109
 110
 111 /*
 112  * Lock refcounts, during creation:
 113  *   - one special one for allocation, dec'd only once in destroy
 114  *   - one for being a lock that's in-use
 115  *   - one for the addref associated with a new lock
 116  */
 117 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 118 {
 119         atomic_inc(&lock->l_refc);
 120         return lock;
 121 }
 122
 123 static void ldlm_lock_free(struct ldlm_lock *lock, size_t size)
 124 {
 125         LASSERT(size == sizeof(*lock));
 126         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
 127 }
 128
 129 void ldlm_lock_put(struct ldlm_lock *lock)
 130 {
 131         ENTRY;
 132
 133         LASSERT(lock->l_resource != LP_POISON);
 134         LASSERT(atomic_read(&lock->l_refc) > 0);
 135         if (atomic_dec_and_test(&lock->l_refc)) {
 136                 struct ldlm_resource *res;
 137
 138                 LDLM_DEBUG(lock,
 139                            "final lock_put on destroyed lock, freeing it.");
 140
 141                 res = lock->l_resource;
 142                 LASSERT(lock->l_destroyed);
 143                 LASSERT(list_empty(&lock->l_res_link));
 144                 LASSERT(list_empty(&lock->l_pending_chain));
 145
 146                 atomic_dec(&res->lr_namespace->ns_locks);
 147                 ldlm_resource_putref(res);
 148                 lock->l_resource = NULL;
 149                 if (lock->l_export) {
 150                         class_export_put(lock->l_export);
 151                         lock->l_export = NULL;
 152                 }
 153
 154                 if (lock->l_lvb_data != NULL)
 155                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
 156
 157                 ldlm_interval_free(ldlm_interval_detach(lock));
 158                 OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
 159                                 ldlm_lock_free);
 160         }
 161
 162         EXIT;
 163 }
 164
 165 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
 166 {
 167         int rc = 0;
 168         if (!list_empty(&lock->l_lru)) {
 169                 struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 170                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 171                 list_del_init(&lock->l_lru);
 172                 ns->ns_nr_unused--;
 173                 LASSERT(ns->ns_nr_unused >= 0);
 174                 rc = 1;
 175         }
 176         return rc;
 177 }
 178
 179 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 180 {
 181         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 182         int rc;
 183         ENTRY;
 184         spin_lock(&ns->ns_unused_lock);
 185         rc = ldlm_lock_remove_from_lru_nolock(lock);
 186         spin_unlock(&ns->ns_unused_lock);
 187         EXIT;
 188         return rc;
 189 }
 190
 191 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
 192 {
 193         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 194         lock->l_last_used = cfs_time_current();
 195         LASSERT(list_empty(&lock->l_lru));
 196         LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 197         list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 198         LASSERT(ns->ns_nr_unused >= 0);
 199         ns->ns_nr_unused++;
 200 }
 201
 202 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
 203 {
 204         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 205         ENTRY;
 206         spin_lock(&ns->ns_unused_lock);
 207         ldlm_lock_add_to_lru_nolock(lock);
 208         spin_unlock(&ns->ns_unused_lock);
 209         EXIT;
 210 }
 211
 212 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
 213 {
 214         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 215         ENTRY;
 216         spin_lock(&ns->ns_unused_lock);
 217         if (!list_empty(&lock->l_lru)) {
 218                 ldlm_lock_remove_from_lru_nolock(lock);
 219                 ldlm_lock_add_to_lru_nolock(lock);
 220         }
 221         spin_unlock(&ns->ns_unused_lock);
 222         EXIT;
 223 }
 224
 225 /* This used to have a 'strict' flag, which recovery would use to mark an
 226  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 227  * shall explain why it's gone: with the new hash table scheme, once you call
 228  * ldlm_lock_destroy, you can never drop your final references on this lock.
 229  * Because it's not in the hash table anymore.  -phil */
 230 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
 231 {
 232         ENTRY;
 233
 234         if (lock->l_readers || lock->l_writers) {
 235                 LDLM_ERROR(lock, "lock still has references");
 236                 ldlm_lock_dump(D_ERROR, lock, 0);
 237                 LBUG();
 238         }
 239
 240         if (!list_empty(&lock->l_res_link)) {
 241                 LDLM_ERROR(lock, "lock still on resource");
 242                 ldlm_lock_dump(D_ERROR, lock, 0);
 243                 LBUG();
 244         }
 245
 246         if (lock->l_destroyed) {
 247                 LASSERT(list_empty(&lock->l_lru));
 248                 EXIT;
 249                 return 0;
 250         }
 251         lock->l_destroyed = 1;
 252
 253         if (lock->l_export)
 254                 spin_lock(&lock->l_export->exp_ldlm_data.led_lock);
 255         list_del_init(&lock->l_export_chain);
 256         if (lock->l_export)
 257                 spin_unlock(&lock->l_export->exp_ldlm_data.led_lock);
 258
 259         ldlm_lock_remove_from_lru(lock);
 260         class_handle_unhash(&lock->l_handle);
 261
 262 #if 0
 263         /* Wake anyone waiting for this lock */
 264         /* FIXME: I should probably add yet another flag, instead of using
 265          * l_export to only call this on clients */
 266         if (lock->l_export)
 267                 class_export_put(lock->l_export);
 268         lock->l_export = NULL;
 269         if (lock->l_export && lock->l_completion_ast)
 270                 lock->l_completion_ast(lock, 0);
 271 #endif
 272         EXIT;
 273         return 1;
 274 }
 275
 276 void ldlm_lock_destroy(struct ldlm_lock *lock)
 277 {
 278         int first;
 279         ENTRY;
 280         lock_res_and_lock(lock);
 281         first = ldlm_lock_destroy_internal(lock);
 282         unlock_res_and_lock(lock);
 283
 284         /* drop reference from hashtable only for first destroy */
 285         if (first)
 286                 LDLM_LOCK_PUT(lock);
 287         EXIT;
 288 }
 289
 290 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
 291 {
 292         int first;
 293         ENTRY;
 294         first = ldlm_lock_destroy_internal(lock);
 295         /* drop reference from hashtable only for first destroy */
 296         if (first)
 297                 LDLM_LOCK_PUT(lock);
 298         EXIT;
 299 }
 300
 301 /* this is called by portals_handle2object with the handle lock taken */
 302 static void lock_handle_addref(void *lock)
 303 {
 304         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 305 }
 306
 307 /*
 308  * usage: pass in a resource on which you have done ldlm_resource_get
 309  *        pass in a parent lock on which you have done a ldlm_lock_get
 310  *        after return, ldlm_*_put the resource and parent
 311  * returns: lock with refcount 2 - one for current caller and one for remote
 312  */
 313 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
 314 {
 315         struct ldlm_lock *lock;
 316         ENTRY;
 317
 318         if (resource == NULL)
 319                 LBUG();
 320
 321         OBD_SLAB_ALLOC(lock, ldlm_lock_slab, CFS_ALLOC_IO, sizeof(*lock));
 322         if (lock == NULL)
 323                 RETURN(NULL);
 324
 325         spin_lock_init(&lock->l_lock);
 326         lock->l_resource = ldlm_resource_getref(resource);
 327
 328         atomic_set(&lock->l_refc, 2);
 329         CFS_INIT_LIST_HEAD(&lock->l_res_link);
 330         CFS_INIT_LIST_HEAD(&lock->l_lru);
 331         CFS_INIT_LIST_HEAD(&lock->l_export_chain);
 332         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
 333         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
 334         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
 335         cfs_waitq_init(&lock->l_waitq);
 336         lock->l_blocking_lock = NULL;
 337         CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
 338         CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
 339
 340         atomic_inc(&resource->lr_namespace->ns_locks);
 341         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
 342         class_handle_hash(&lock->l_handle, lock_handle_addref);
 343
 344         CFS_INIT_LIST_HEAD(&lock->l_extents_list);
 345         spin_lock_init(&lock->l_extents_list_lock);
 346         CFS_INIT_LIST_HEAD(&lock->l_cache_locks_list);
 347
 348         RETURN(lock);
 349 }
 350
 351 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 352                               const struct ldlm_res_id *new_resid)
 353 {
 354         struct ldlm_resource *oldres = lock->l_resource;
 355         struct ldlm_resource *newres;
 356         int type;
 357         ENTRY;
 358
 359         LASSERT(ns_is_client(ns));
 360
 361         lock_res_and_lock(lock);
 362         if (memcmp(new_resid, &lock->l_resource->lr_name,
 363                    sizeof(lock->l_resource->lr_name)) == 0) {
 364                 /* Nothing to do */
 365                 unlock_res_and_lock(lock);
 366                 RETURN(0);
 367         }
 368
 369         LASSERT(new_resid->name[0] != 0);
 370
 371         /* This function assumes that the lock isn't on any lists */
 372         LASSERT(list_empty(&lock->l_res_link));
 373
 374         type = oldres->lr_type;
 375         unlock_res_and_lock(lock);
 376
 377         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 378         if (newres == NULL)
 379                 RETURN(-ENOMEM);
 380
 381         lock_res_and_lock(lock);
 382         LASSERT(memcmp(new_resid, &lock->l_resource->lr_name,
 383                        sizeof(lock->l_resource->lr_name)) != 0);
 384         lock_res(newres);
 385         lock->l_resource = newres;
 386         unlock_res(oldres);
 387         unlock_res_and_lock(lock);
 388
 389         /* ...and the flowers are still standing! */
 390         ldlm_resource_putref(oldres);
 391
 392         RETURN(0);
 393 }
 394
 395 /*
 396  *  HANDLES
 397  */
 398
 399 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
 400 {
 401         lockh->cookie = lock->l_handle.h_cookie;
 402 }
 403
 404 /* if flags: atomically get the lock and set the flags.
 405  *           Return NULL if flag already set
 406  */
 407
 408 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
 409                                      int flags)
 410 {
 411         struct ldlm_namespace *ns;
 412         struct ldlm_lock *lock, *retval = NULL;
 413         ENTRY;
 414
 415         LASSERT(handle);
 416
 417         lock = class_handle2object(handle->cookie);
 418         if (lock == NULL)
 419                 RETURN(NULL);
 420
 421         LASSERT(lock->l_resource != NULL);
 422         ns = lock->l_resource->lr_namespace;
 423         LASSERT(ns != NULL);
 424
 425         lock_res_and_lock(lock);
 426
 427         /* It's unlikely but possible that someone marked the lock as
 428          * destroyed after we did handle2object on it */
 429         if (lock->l_destroyed) {
 430                 unlock_res_and_lock(lock);
 431                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 432                 LDLM_LOCK_PUT(lock);
 433                 GOTO(out, retval);
 434         }
 435
 436         if (flags && (lock->l_flags & flags)) {
 437                 unlock_res_and_lock(lock);
 438                 LDLM_LOCK_PUT(lock);
 439                 GOTO(out, retval);
 440         }
 441
 442         if (flags)
 443                 lock->l_flags |= flags;
 444
 445         unlock_res_and_lock(lock);
 446         retval = lock;
 447         EXIT;
 448  out:
 449         return retval;
 450 }
 451
 452 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
 453                                       const struct lustre_handle *handle)
 454 {
 455         struct ldlm_lock *retval = NULL;
 456         retval = __ldlm_handle2lock(handle, 0);
 457         return retval;
 458 }
 459
 460 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 461 {
 462         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
 463         /* INODEBITS_INTEROP: If the other side does not support
 464          * inodebits, reply with a plain lock descriptor.
 465          */
 466         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
 467             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
 468                 struct ldlm_resource res = *lock->l_resource;
 469
 470                 /* Make sure all the right bits are set in this lock we
 471                    are going to pass to client */
 472                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
 473                          (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
 474                          "Inappropriate inode lock bits during "
 475                          "conversion " LPU64 "\n",
 476                          lock->l_policy_data.l_inodebits.bits);
 477                 res.lr_type = LDLM_PLAIN;
 478                 ldlm_res2desc(&res, &desc->l_resource);
 479                 /* Convert "new" lock mode to something old client can
 480                    understand */
 481                 if ((lock->l_req_mode == LCK_CR) ||
 482                     (lock->l_req_mode == LCK_CW))
 483                         desc->l_req_mode = LCK_PR;
 484                 else
 485                         desc->l_req_mode = lock->l_req_mode;
 486                 if ((lock->l_granted_mode == LCK_CR) ||
 487                     (lock->l_granted_mode == LCK_CW)) {
 488                         desc->l_granted_mode = LCK_PR;
 489                 } else {
 490                         /* We never grant PW/EX locks to clients */
 491                         LASSERT((lock->l_granted_mode != LCK_PW) &&
 492                                 (lock->l_granted_mode != LCK_EX));
 493                         desc->l_granted_mode = lock->l_granted_mode;
 494                 }
 495
 496                 /* We do not copy policy here, because there is no
 497                    policy for plain locks */
 498         } else {
 499                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 500                 desc->l_req_mode = lock->l_req_mode;
 501                 desc->l_granted_mode = lock->l_granted_mode;
 502                 desc->l_policy_data = lock->l_policy_data;
 503         }
 504 }
 505
 506 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 507                            struct list_head *work_list)
 508 {
 509         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
 510                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 511                 lock->l_flags |= LDLM_FL_AST_SENT;
 512                 /* If the enqueuing client said so, tell the AST recipient to
 513                  * discard dirty data, rather than writing back. */
 514                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 515                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 516                 LASSERT(list_empty(&lock->l_bl_ast));
 517                 list_add(&lock->l_bl_ast, work_list);
 518                 LDLM_LOCK_GET(lock);
 519                 LASSERT(lock->l_blocking_lock == NULL);
 520                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
 521         }
 522 }
 523
 524 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
 525 {
 526         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
 527                 lock->l_flags |= LDLM_FL_CP_REQD;
 528                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
 529                 LASSERT(list_empty(&lock->l_cp_ast));
 530                 list_add(&lock->l_cp_ast, work_list);
 531                 LDLM_LOCK_GET(lock);
 532         }
 533 }
 534
 535 /* must be called with lr_lock held */
 536 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 537                                 struct list_head *work_list)
 538 {
 539         ENTRY;
 540         check_res_locked(lock->l_resource);
 541         if (new)
 542                 ldlm_add_bl_work_item(lock, new, work_list);
 543         else
 544                 ldlm_add_cp_work_item(lock, work_list);
 545         EXIT;
 546 }
 547
 548 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 549 {
 550         struct ldlm_lock *lock;
 551
 552         lock = ldlm_handle2lock(lockh);
 553         LASSERT(lock != NULL);
 554         ldlm_lock_addref_internal(lock, mode);
 555         LDLM_LOCK_PUT(lock);
 556 }
 557
 558 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 559 {
 560         ldlm_lock_remove_from_lru(lock);
 561         if (mode & (LCK_NL | LCK_CR | LCK_PR))
 562                 lock->l_readers++;
 563         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
 564                 lock->l_writers++;
 565         LDLM_LOCK_GET(lock);
 566         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 567 }
 568
 569 /* only called for local locks */
 570 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 571 {
 572         lock_res_and_lock(lock);
 573         ldlm_lock_addref_internal_nolock(lock, mode);
 574         unlock_res_and_lock(lock);
 575 }
 576
 577 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 578 {
 579         struct ldlm_namespace *ns;
 580         ENTRY;
 581
 582         lock_res_and_lock(lock);
 583
 584         ns = lock->l_resource->lr_namespace;
 585
 586         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 587         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 588                 LASSERT(lock->l_readers > 0);
 589                 lock->l_readers--;
 590         }
 591         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
 592                 LASSERT(lock->l_writers > 0);
 593                 lock->l_writers--;
 594         }
 595
 596         if (lock->l_flags & LDLM_FL_LOCAL &&
 597             !lock->l_readers && !lock->l_writers) {
 598                 /* If this is a local lock on a server namespace and this was
 599                  * the last reference, cancel the lock. */
 600                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 601                 lock->l_flags |= LDLM_FL_CBPENDING;
 602         }
 603
 604         if (!lock->l_readers && !lock->l_writers &&
 605             (lock->l_flags & LDLM_FL_CBPENDING)) {
 606                 /* If we received a blocked AST and this was the last reference,
 607                  * run the callback. */
 608                 if (ns_is_server(ns) && lock->l_export)
 609                         CERROR("FL_CBPENDING set on non-local lock--just a "
 610                                "warning\n");
 611
 612                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 613
 614                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
 615                 ldlm_lock_remove_from_lru(lock);
 616                 unlock_res_and_lock(lock);
 617                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
 618                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
 619                         ldlm_handle_bl_callback(ns, NULL, lock);
 620         } else if (ns_is_client(ns) &&
 621                    !lock->l_readers && !lock->l_writers &&
 622                    !(lock->l_flags & LDLM_FL_NO_LRU) &&
 623                    !(lock->l_flags & LDLM_FL_BL_AST)) {
 624                 /* If this is a client-side namespace and this was the last
 625                  * reference, put it on the LRU. */
 626                 ldlm_lock_add_to_lru(lock);
 627                 unlock_res_and_lock(lock);
 628                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
 629                  * are not supported by the server, otherwise, it is done on
 630                  * enqueue. */
 631                 if (!exp_connect_cancelset(lock->l_conn_export) &&
 632                     !ns_connect_lru_resize(ns))
 633                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
 634         } else {
 635                 unlock_res_and_lock(lock);
 636         }
 637
 638         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 639
 640         EXIT;
 641 }
 642
 643 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 644 {
 645         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 646         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
 647         ldlm_lock_decref_internal(lock, mode);
 648         LDLM_LOCK_PUT(lock);
 649 }
 650
 651 /* This will drop a lock reference and mark it for destruction, but will not
 652  * necessarily cancel the lock before returning. */
 653 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 654 {
 655         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 656         ENTRY;
 657
 658         LASSERT(lock != NULL);
 659
 660         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 661         lock_res_and_lock(lock);
 662         lock->l_flags |= LDLM_FL_CBPENDING;
 663         unlock_res_and_lock(lock);
 664         ldlm_lock_decref_internal(lock, mode);
 665         LDLM_LOCK_PUT(lock);
 666 }
 667
 668 struct sl_insert_point {
 669         struct list_head *res_link;
 670         struct list_head *mode_link;
 671         struct list_head *policy_link;
 672 };
 673
 674 /*
 675  * search_granted_lock
 676  *
 677  * Description:
 678  *      Finds a position to insert the new lock.
 679  * Parameters:
 680  *      queue [input]:  the granted list where search acts on;
 681  *      req [input]:    the lock whose position to be located;
 682  *      prev [output]:  positions within 3 lists to insert @req to
 683  * Return Value:
 684  *      filled @prev
 685  * NOTE: called by
 686  *  - ldlm_grant_lock_with_skiplist
 687  */
 688 static void search_granted_lock(struct list_head *queue,
 689                                 struct ldlm_lock *req,
 690                                 struct sl_insert_point *prev)
 691 {
 692         struct list_head *tmp;
 693         struct ldlm_lock *lock, *mode_end, *policy_end;
 694         ENTRY;
 695
 696         list_for_each(tmp, queue) {
 697                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 698
 699                 mode_end = list_entry(lock->l_sl_mode.prev, struct ldlm_lock,
 700                                       l_sl_mode);
 701
 702                 if (lock->l_req_mode != req->l_req_mode) {
 703                         /* jump to last lock of mode group */
 704                         tmp = &mode_end->l_res_link;
 705                         continue;
 706                 }
 707
 708                 /* suitable mode group is found */
 709                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
 710                         /* insert point is last lock of the mode group */
 711                         prev->res_link = &mode_end->l_res_link;
 712                         prev->mode_link = &mode_end->l_sl_mode;
 713                         prev->policy_link = &req->l_sl_policy;
 714                         EXIT;
 715                         return;
 716                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
 717                         for (;;) {
 718                                 policy_end = list_entry(lock->l_sl_policy.prev,
 719                                                         struct ldlm_lock,
 720                                                         l_sl_policy);
 721
 722                                 if (lock->l_policy_data.l_inodebits.bits ==
 723                                     req->l_policy_data.l_inodebits.bits) {
 724                                         /* insert point is last lock of
 725                                          * the policy group */
 726                                         prev->res_link =
 727                                                 &policy_end->l_res_link;
 728                                         prev->mode_link =
 729                                                 &policy_end->l_sl_mode;
 730                                         prev->policy_link =
 731                                                 &policy_end->l_sl_policy;
 732                                         EXIT;
 733                                         return;
 734                                 }
 735
 736                                 if (policy_end == mode_end)
 737                                         /* done with mode group */
 738                                         break;
 739
 740                                 /* jump to next policy group within the mode group */
 741                                 tmp = policy_end->l_res_link.next;
 742                                 lock = list_entry(tmp, struct ldlm_lock,
 743                                                   l_res_link);
 744                         }  /* loop over policy groups within the mode group */
 745
 746                         /* insert point is last lock of the mode group,
 747                          * new policy group is started */
 748                         prev->res_link = &mode_end->l_res_link;
 749                         prev->mode_link = &mode_end->l_sl_mode;
 750                         prev->policy_link = &req->l_sl_policy;
 751                         EXIT;
 752                         return;
 753                 } else {
 754                         LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
 755                         LBUG();
 756                 }
 757         }
 758
 759         /* insert point is last lock on the queue,
 760          * new mode group and new policy group are started */
 761         prev->res_link = queue->prev;
 762         prev->mode_link = &req->l_sl_mode;
 763         prev->policy_link = &req->l_sl_policy;
 764         EXIT;
 765         return;
 766 }
 767
 768 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
 769                                        struct sl_insert_point *prev)
 770 {
 771         struct ldlm_resource *res = lock->l_resource;
 772         ENTRY;
 773
 774         check_res_locked(res);
 775
 776         ldlm_resource_dump(D_OTHER, res);
 777         CDEBUG(D_OTHER, "About to add this lock:\n");
 778         ldlm_lock_dump(D_OTHER, lock, 0);
 779
 780         if (lock->l_destroyed) {
 781                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 782                 return;
 783         }
 784
 785         LASSERT(list_empty(&lock->l_res_link));
 786         LASSERT(list_empty(&lock->l_sl_mode));
 787         LASSERT(list_empty(&lock->l_sl_policy));
 788
 789         list_add(&lock->l_res_link, prev->res_link);
 790         list_add(&lock->l_sl_mode, prev->mode_link);
 791         list_add(&lock->l_sl_policy, prev->policy_link);
 792
 793         EXIT;
 794 }
 795
 796 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
 797 {
 798         struct sl_insert_point prev;
 799         ENTRY;
 800
 801         LASSERT(lock->l_req_mode == lock->l_granted_mode);
 802
 803         search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
 804         ldlm_granted_list_add_lock(lock, &prev);
 805         EXIT;
 806 }
 807
 808 /* NOTE: called by
 809  *  - ldlm_lock_enqueue
 810  *  - ldlm_reprocess_queue
 811  *  - ldlm_lock_convert
 812  *
 813  * must be called with lr_lock held
 814  */
 815 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
 816 {
 817         struct ldlm_resource *res = lock->l_resource;
 818         ENTRY;
 819
 820         check_res_locked(res);
 821
 822         lock->l_granted_mode = lock->l_req_mode;
 823         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
 824                 ldlm_grant_lock_with_skiplist(lock);
 825         else if (res->lr_type == LDLM_EXTENT)
 826                 ldlm_extent_add_lock(res, lock);
 827         else
 828                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
 829
 830         if (lock->l_granted_mode < res->lr_most_restr)
 831                 res->lr_most_restr = lock->l_granted_mode;
 832
 833         if (work_list && lock->l_completion_ast != NULL)
 834                 ldlm_add_ast_work_item(lock, NULL, work_list);
 835
 836         ldlm_pool_add(&res->lr_namespace->ns_pool, lock);
 837         EXIT;
 838 }
 839
 840 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
 841  * comment above ldlm_lock_match */
 842 static struct ldlm_lock *search_queue(struct list_head *queue,
 843                                       ldlm_mode_t *mode,
 844                                       ldlm_policy_data_t *policy,
 845                                       struct ldlm_lock *old_lock, int flags)
 846 {
 847         struct ldlm_lock *lock;
 848         struct list_head *tmp;
 849
 850         list_for_each(tmp, queue) {
 851                 ldlm_mode_t match;
 852
 853                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 854
 855                 if (lock == old_lock)
 856                         break;
 857
 858                 /* llite sometimes wants to match locks that will be
 859                  * canceled when their users drop, but we allow it to match
 860                  * if it passes in CBPENDING and the lock still has users.
 861                  * this is generally only going to be used by children
 862                  * whose parents already hold a lock so forward progress
 863                  * can still happen. */
 864                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 865                     !(flags & LDLM_FL_CBPENDING))
 866                         continue;
 867                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 868                     lock->l_readers == 0 && lock->l_writers == 0)
 869                         continue;
 870
 871                 if (!(lock->l_req_mode & *mode))
 872                         continue;
 873                 match = lock->l_req_mode;
 874
 875                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 876                     (lock->l_policy_data.l_extent.start >
 877                      policy->l_extent.start ||
 878                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
 879                         continue;
 880
 881                 if (unlikely(match == LCK_GROUP) &&
 882                     lock->l_resource->lr_type == LDLM_EXTENT &&
 883                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
 884                         continue;
 885
 886                 /* We match if we have existing lock with same or wider set
 887                    of bits. */
 888                 if (lock->l_resource->lr_type == LDLM_IBITS &&
 889                      ((lock->l_policy_data.l_inodebits.bits &
 890                       policy->l_inodebits.bits) !=
 891                       policy->l_inodebits.bits))
 892                         continue;
 893
 894                 if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
 895                         continue;
 896
 897                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
 898                     !(lock->l_flags & LDLM_FL_LOCAL))
 899                         continue;
 900
 901                 if (flags & LDLM_FL_TEST_LOCK) {
 902                         LDLM_LOCK_GET(lock);
 903                         ldlm_lock_touch_in_lru(lock);
 904                 } else {
 905                         ldlm_lock_addref_internal_nolock(lock, match);
 906                 }
 907                 *mode = match;
 908                 return lock;
 909         }
 910
 911         return NULL;
 912 }
 913
 914 void ldlm_lock_allow_match(struct ldlm_lock *lock)
 915 {
 916         lock_res_and_lock(lock);
 917         lock->l_flags |= LDLM_FL_LVB_READY;
 918         cfs_waitq_signal(&lock->l_waitq);
 919         unlock_res_and_lock(lock);
 920 }
 921
 922 int ldlm_lock_fast_match(struct ldlm_lock *lock, int rw,
 923                          obd_off start, obd_off end,
 924                          void **cookie)
 925 {
 926         LASSERT(rw == OBD_BRW_READ || rw == OBD_BRW_WRITE);
 927         /* should LCK_GROUP be handled in a special way? */
 928         if (lock && (rw == OBD_BRW_READ ||
 929                      (lock->l_granted_mode & (LCK_PW | LCK_GROUP))) &&
 930             (lock->l_policy_data.l_extent.start <= start) &&
 931             (lock->l_policy_data.l_extent.end >= end)) {
 932                 ldlm_lock_addref_internal(lock, rw == OBD_BRW_WRITE ? LCK_PW : LCK_PR);
 933                 *cookie = (void *)lock;
 934                 return 1; /* avoid using rc for stack relief */
 935         }
 936         return 0;
 937 }
 938
 939 void ldlm_lock_fast_release(void *cookie, int rw)
 940 {
 941         struct ldlm_lock *lock = (struct ldlm_lock *)cookie;
 942
 943         LASSERT(lock != NULL);
 944         LASSERT(rw == OBD_BRW_READ || rw == OBD_BRW_WRITE);
 945         LASSERT(rw == OBD_BRW_READ ||
 946                 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)));
 947         ldlm_lock_decref_internal(lock, rw == OBD_BRW_WRITE ? LCK_PW : LCK_PR);
 948 }
 949
 950 /* Can be called in two ways:
 951  *
 952  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 953  * for a duplicate of.
 954  *
 955  * Otherwise, all of the fields must be filled in, to match against.
 956  *
 957  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
 958  *     server (ie, connh is NULL)
 959  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
 960  *     list will be considered
 961  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
 962  *     to be canceled can still be matched as long as they still have reader
 963  *     or writer refernces
 964  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
 965  *     just tell us if we would have matched.
 966  *
 967  * Returns 1 if it finds an already-existing lock that is compatible; in this
 968  * case, lockh is filled in with a addref()ed lock
 969  *
 970  * we also check security context, if that failed we simply return 0 (to keep
 971  * caller code unchanged), the context failure will be discovered by caller
 972  * sometime later.
 973  */
 974 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
 975                             const struct ldlm_res_id *res_id, ldlm_type_t type,
 976                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
 977                             struct lustre_handle *lockh)
 978 {
 979         struct ldlm_resource *res;
 980         struct ldlm_lock *lock, *old_lock = NULL;
 981         int rc = 0;
 982         ENTRY;
 983
 984         if (ns == NULL) {
 985                 old_lock = ldlm_handle2lock(lockh);
 986                 LASSERT(old_lock);
 987
 988                 ns = old_lock->l_resource->lr_namespace;
 989                 res_id = &old_lock->l_resource->lr_name;
 990                 type = old_lock->l_resource->lr_type;
 991                 mode = old_lock->l_req_mode;
 992         }
 993
 994         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
 995         if (res == NULL) {
 996                 LASSERT(old_lock == NULL);
 997                 RETURN(0);
 998         }
 999
1000         lock_res(res);
1001
1002         lock = search_queue(&res->lr_granted, &mode, policy, old_lock, flags);
1003         if (lock != NULL)
1004                 GOTO(out, rc = 1);
1005         if (flags & LDLM_FL_BLOCK_GRANTED)
1006                 GOTO(out, rc = 0);
1007         lock = search_queue(&res->lr_converting, &mode, policy, old_lock, flags);
1008         if (lock != NULL)
1009                 GOTO(out, rc = 1);
1010         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock, flags);
1011         if (lock != NULL)
1012                 GOTO(out, rc = 1);
1013
1014         EXIT;
1015  out:
1016         unlock_res(res);
1017         ldlm_resource_putref(res);
1018
1019         if (lock) {
1020                 ldlm_lock2handle(lock, lockh);
1021                 if ((flags & LDLM_FL_LVB_READY) &&
1022                     (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1023                         struct l_wait_info lwi;
1024                         if (lock->l_completion_ast) {
1025                                 int err = lock->l_completion_ast(lock,
1026                                                           LDLM_FL_WAIT_NOREPROC,
1027                                                                  NULL);
1028                                 if (err) {
1029                                         if (flags & LDLM_FL_TEST_LOCK)
1030                                                 LDLM_LOCK_PUT(lock);
1031                                         else
1032                                                 ldlm_lock_decref_internal(lock, mode);
1033                                         rc = 0;
1034                                         goto out2;
1035                                 }
1036                         }
1037
1038                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
1039                                                LWI_ON_SIGNAL_NOOP, NULL);
1040
1041                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1042                         l_wait_event(lock->l_waitq,
1043                                      (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1044                 }
1045         }
1046  out2:
1047         if (rc) {
1048                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1049                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1050                                 res_id->name[2] : policy->l_extent.start,
1051                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1052                                 res_id->name[3] : policy->l_extent.end);
1053
1054                 /* check user's security context */
1055                 if (lock->l_conn_export &&
1056                     sptlrpc_import_check_ctx(
1057                                 class_exp2cliimp(lock->l_conn_export))) {
1058                         if (!(flags & LDLM_FL_TEST_LOCK))
1059                                 ldlm_lock_decref_internal(lock, mode);
1060                         rc = 0;
1061                 }
1062
1063                 if (flags & LDLM_FL_TEST_LOCK)
1064                         LDLM_LOCK_PUT(lock);
1065
1066         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1067                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1068                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1069                                   type, mode, res_id->name[0], res_id->name[1],
1070                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1071                                         res_id->name[2] :policy->l_extent.start,
1072                                 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1073                                         res_id->name[3] : policy->l_extent.end);
1074         }
1075         if (old_lock)
1076                 LDLM_LOCK_PUT(old_lock);
1077
1078         return rc ? mode : 0;
1079 }
1080
1081 /* Returns a referenced lock */
1082 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1083                                    const struct ldlm_res_id *res_id,
1084                                    ldlm_type_t type,
1085                                    ldlm_mode_t mode,
1086                                    ldlm_blocking_callback blocking,
1087                                    ldlm_completion_callback completion,
1088                                    ldlm_glimpse_callback glimpse,
1089                                    void *data, __u32 lvb_len)
1090 {
1091         struct ldlm_lock *lock;
1092         struct ldlm_resource *res;
1093         ENTRY;
1094
1095         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1096         if (res == NULL)
1097                 RETURN(NULL);
1098
1099         lock = ldlm_lock_new(res);
1100         ldlm_resource_putref(res);
1101
1102         if (lock == NULL)
1103                 RETURN(NULL);
1104
1105         lock->l_req_mode = mode;
1106         lock->l_ast_data = data;
1107         lock->l_blocking_ast = blocking;
1108         lock->l_completion_ast = completion;
1109         lock->l_glimpse_ast = glimpse;
1110         lock->l_pid = cfs_curproc_pid();
1111
1112         lock->l_tree_node = NULL;
1113         /* if this is the extent lock, allocate the interval tree node */
1114         if (type == LDLM_EXTENT) {
1115                 if (ldlm_interval_alloc(lock) == NULL)
1116                         GOTO(out, 0);
1117         }
1118
1119         if (lvb_len) {
1120                 lock->l_lvb_len = lvb_len;
1121                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1122                 if (lock->l_lvb_data == NULL)
1123                         GOTO(out, 0);
1124         }
1125
1126         RETURN(lock);
1127
1128 out:
1129         if (lock->l_lvb_data)
1130                 OBD_FREE(lock->l_lvb_data, lvb_len);
1131         ldlm_interval_free(ldlm_interval_detach(lock));
1132         OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1133         return NULL;
1134 }
1135
1136 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1137                                struct ldlm_lock **lockp,
1138                                void *cookie, int *flags)
1139 {
1140         struct ldlm_lock *lock = *lockp;
1141         struct ldlm_resource *res = lock->l_resource;
1142         int local = ns_is_client(res->lr_namespace);
1143         ldlm_processing_policy policy;
1144         ldlm_error_t rc = ELDLM_OK;
1145         struct ldlm_interval *node = NULL;
1146         ENTRY;
1147
1148         do_gettimeofday(&lock->l_enqueued_time);
1149         /* policies are not executed on the client or during replay */
1150         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1151             && !local && ns->ns_policy) {
1152                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1153                                    NULL);
1154                 if (rc == ELDLM_LOCK_REPLACED) {
1155                         /* The lock that was returned has already been granted,
1156                          * and placed into lockp.  If it's not the same as the
1157                          * one we passed in, then destroy the old one and our
1158                          * work here is done. */
1159                         if (lock != *lockp) {
1160                                 ldlm_lock_destroy(lock);
1161                                 LDLM_LOCK_PUT(lock);
1162                         }
1163                         *flags |= LDLM_FL_LOCK_CHANGED;
1164                         RETURN(0);
1165                 } else if (rc != ELDLM_OK ||
1166                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1167                         ldlm_lock_destroy(lock);
1168                         RETURN(rc);
1169                 }
1170         }
1171
1172         /* For a replaying lock, it might be already in granted list. So
1173          * unlinking the lock will cause the interval node to be freed, we
1174          * have to allocate the interval node early otherwise we can't regrant
1175          * this lock in the future. - jay */
1176         if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1177                 OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO,
1178                                sizeof(*node));
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
1183                  * before we got a chance to actually enqueue it.  We don't
1184                  * need to do 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         ldlm_resource_unlink_lock(lock);
1191         if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1192                 if (node == NULL) {
1193                         ldlm_lock_destroy_nolock(lock);
1194                         GOTO(out, rc = -ENOMEM);
1195                 }
1196
1197                 CFS_INIT_LIST_HEAD(&node->li_group);
1198                 ldlm_interval_attach(node, lock);
1199                 node = NULL;
1200         }
1201
1202         /* Some flags from the enqueue want to make it into the AST, via the
1203          * lock's l_flags. */
1204         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1205
1206         /* This distinction between local lock trees is very important; a client
1207          * namespace only has information about locks taken by that client, and
1208          * thus doesn't have enough information to decide for itself if it can
1209          * be granted (below).  In this case, we do exactly what the server
1210          * tells us to do, as dictated by the 'flags'.
1211          *
1212          * We do exactly the same thing during recovery, when the server is
1213          * more or less trusting the clients not to lie.
1214          *
1215          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1216          * granted/converting queues. */
1217         if (local) {
1218                 if (*flags & LDLM_FL_BLOCK_CONV)
1219                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1220                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1221                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1222                 else
1223                         ldlm_grant_lock(lock, NULL);
1224                 GOTO(out, ELDLM_OK);
1225         } else if (*flags & LDLM_FL_REPLAY) {
1226                 if (*flags & LDLM_FL_BLOCK_CONV) {
1227                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1228                         GOTO(out, ELDLM_OK);
1229                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1230                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1231                         GOTO(out, ELDLM_OK);
1232                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1233                         ldlm_grant_lock(lock, NULL);
1234                         GOTO(out, ELDLM_OK);
1235                 }
1236                 /* If no flags, fall through to normal enqueue path. */
1237         }
1238
1239         policy = ldlm_processing_policy_table[res->lr_type];
1240         policy(lock, flags, 1, &rc, NULL);
1241         GOTO(out, rc);
1242 out:
1243         unlock_res_and_lock(lock);
1244         if (node)
1245                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1246         return rc;
1247 }
1248
1249 /* Must be called with namespace taken: queue is waiting or converting. */
1250 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1251                          struct list_head *work_list)
1252 {
1253         struct list_head *tmp, *pos;
1254         ldlm_processing_policy policy;
1255         int flags;
1256         int rc = LDLM_ITER_CONTINUE;
1257         ldlm_error_t err;
1258         ENTRY;
1259
1260         check_res_locked(res);
1261
1262         policy = ldlm_processing_policy_table[res->lr_type];
1263         LASSERT(policy);
1264
1265         list_for_each_safe(tmp, pos, queue) {
1266                 struct ldlm_lock *pending;
1267                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1268
1269                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1270
1271                 flags = 0;
1272                 rc = policy(pending, &flags, 0, &err, work_list);
1273                 if (rc != LDLM_ITER_CONTINUE)
1274                         break;
1275         }
1276
1277         RETURN(rc);
1278 }
1279
1280 /* Helper function for ldlm_run_ast_work().
1281  *
1282  * Send an existing rpc set specified by @arg->set and then
1283  * destroy it. Create new one if @do_create flag is set. */
1284 static void
1285 ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create)
1286 {
1287         ENTRY;
1288
1289         ptlrpc_set_wait(arg->set);
1290         if (arg->type == LDLM_BL_CALLBACK)
1291                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2);
1292         ptlrpc_set_destroy(arg->set);
1293
1294         if (do_create)
1295                 arg->set = ptlrpc_prep_set();
1296
1297         EXIT;
1298 }
1299
1300 static int
1301 ldlm_work_bl_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1302 {
1303         struct ldlm_lock_desc d;
1304         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_bl_ast);
1305         ENTRY;
1306
1307         /* nobody should touch l_bl_ast */
1308         lock_res_and_lock(lock);
1309         list_del_init(&lock->l_bl_ast);
1310
1311         LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1312         LASSERT(lock->l_bl_ast_run == 0);
1313         LASSERT(lock->l_blocking_lock);
1314         lock->l_bl_ast_run++;
1315         unlock_res_and_lock(lock);
1316
1317         ldlm_lock2desc(lock->l_blocking_lock, &d);
1318
1319         LDLM_LOCK_PUT(lock->l_blocking_lock);
1320         lock->l_blocking_lock = NULL;
1321         lock->l_blocking_ast(lock, &d, (void *)arg,
1322                              LDLM_CB_BLOCKING);
1323         LDLM_LOCK_PUT(lock);
1324
1325         RETURN(1);
1326 }
1327
1328 static int
1329 ldlm_work_cp_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1330 {
1331         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_cp_ast);
1332         ldlm_completion_callback completion_callback;
1333         int rc = 0;
1334         ENTRY;
1335
1336         /* It's possible to receive a completion AST before we've set
1337          * the l_completion_ast pointer: either because the AST arrived
1338          * before the reply, or simply because there's a small race
1339          * window between receiving the reply and finishing the local
1340          * enqueue. (bug 842)
1341          *
1342          * This can't happen with the blocking_ast, however, because we
1343          * will never call the local blocking_ast until we drop our
1344          * reader/writer reference, which we won't do until we get the
1345          * reply and finish enqueueing. */
1346
1347         /* nobody should touch l_cp_ast */
1348         lock_res_and_lock(lock);
1349         list_del_init(&lock->l_cp_ast);
1350         LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1351         /* save l_completion_ast since it can be changed by
1352          * mds_intent_policy(), see bug 14225 */
1353         completion_callback = lock->l_completion_ast;
1354         lock->l_flags &= ~LDLM_FL_CP_REQD;
1355         unlock_res_and_lock(lock);
1356
1357         if (completion_callback != NULL) {
1358                 completion_callback(lock, 0, (void *)arg);
1359                 rc = 1;
1360         }
1361         LDLM_LOCK_PUT(lock);
1362
1363         RETURN(rc);
1364 }
1365
1366 static int
1367 ldlm_work_revoke_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
1368 {
1369         struct ldlm_lock_desc desc;
1370         struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_export_chain);
1371         ENTRY;
1372
1373         list_del_init(&lock->l_export_chain);
1374
1375         /* the desc just pretend to exclusive */
1376         ldlm_lock2desc(lock, &desc);
1377         desc.l_req_mode = LCK_EX;
1378         desc.l_granted_mode = 0;
1379
1380         lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1381         LDLM_LOCK_PUT(lock);
1382
1383         RETURN(1);
1384 }
1385
1386 int ldlm_run_ast_work(struct list_head *rpc_list, ldlm_desc_ast_t ast_type)
1387 {
1388         struct ldlm_cb_set_arg arg;
1389         struct list_head *tmp, *pos;
1390         int (*work_ast_lock)(struct list_head *tmp, struct ldlm_cb_set_arg *arg);
1391         int ast_count;
1392         ENTRY;
1393
1394         arg.set = ptlrpc_prep_set();
1395         atomic_set(&arg.restart, 0);
1396         switch (ast_type) {
1397         case LDLM_WORK_BL_AST:
1398                 arg.type = LDLM_BL_CALLBACK;
1399                 work_ast_lock = ldlm_work_bl_ast_lock;
1400                 break;
1401         case LDLM_WORK_CP_AST:
1402                 arg.type = LDLM_CP_CALLBACK;
1403                 work_ast_lock = ldlm_work_cp_ast_lock;
1404                 break;
1405         case LDLM_WORK_REVOKE_AST:
1406                 arg.type = LDLM_BL_CALLBACK;
1407                 work_ast_lock = ldlm_work_revoke_ast_lock;
1408                 break;
1409         default:
1410                 LBUG();
1411         }
1412
1413         ast_count = 0;
1414         list_for_each_safe(tmp, pos, rpc_list) {
1415                 ast_count += work_ast_lock(tmp, &arg);
1416
1417                 /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
1418                  * and create a new set for requests that remained in
1419                  * @rpc_list */
1420                 if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
1421                         ldlm_send_and_maybe_create_set(&arg, 1);
1422                         ast_count = 0;
1423                 }
1424         }
1425
1426         if (ast_count > 0)
1427                 ldlm_send_and_maybe_create_set(&arg, 0);
1428         else
1429                 /* In case when number of ASTs is multiply of
1430                  * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
1431                  * @arg.set must be destroyed here, otherwise we get
1432                  * write memory leaking. */
1433                 ptlrpc_set_destroy(arg.set);
1434
1435         RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
1436 }
1437
1438 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1439 {
1440         ldlm_reprocess_all(res);
1441         return LDLM_ITER_CONTINUE;
1442 }
1443
1444 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1445 {
1446         struct list_head *tmp;
1447         int i, rc;
1448
1449         if (ns == NULL)
1450                 return;
1451
1452         ENTRY;
1453         spin_lock(&ns->ns_hash_lock);
1454         for (i = 0; i < RES_HASH_SIZE; i++) {
1455                 tmp = ns->ns_hash[i].next;
1456                 while (tmp != &(ns->ns_hash[i])) {
1457                         struct ldlm_resource *res =
1458                                 list_entry(tmp, struct ldlm_resource, lr_hash);
1459
1460                         ldlm_resource_getref(res);
1461                         spin_unlock(&ns->ns_hash_lock);
1462
1463                         rc = reprocess_one_queue(res, NULL);
1464
1465                         spin_lock(&ns->ns_hash_lock);
1466                         tmp = tmp->next;
1467                         ldlm_resource_putref_locked(res);
1468
1469                         if (rc == LDLM_ITER_STOP)
1470                                 GOTO(out, rc);
1471                 }
1472         }
1473  out:
1474         spin_unlock(&ns->ns_hash_lock);
1475         EXIT;
1476 }
1477
1478 void ldlm_reprocess_all(struct ldlm_resource *res)
1479 {
1480         CFS_LIST_HEAD(rpc_list);
1481         int rc;
1482         ENTRY;
1483
1484         /* Local lock trees don't get reprocessed. */
1485         if (ns_is_client(res->lr_namespace)) {
1486                 EXIT;
1487                 return;
1488         }
1489
1490  restart:
1491         lock_res(res);
1492         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1493         if (rc == LDLM_ITER_CONTINUE)
1494                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1495         unlock_res(res);
1496
1497         rc = ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
1498         if (rc == -ERESTART) {
1499                 LASSERT(list_empty(&rpc_list));
1500                 goto restart;
1501         }
1502         EXIT;
1503 }
1504
1505 void ldlm_cancel_callback(struct ldlm_lock *lock)
1506 {
1507         check_res_locked(lock->l_resource);
1508         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1509                 lock->l_flags |= LDLM_FL_CANCEL;
1510                 if (lock->l_blocking_ast) {
1511                         // l_check_no_ns_lock(ns);
1512                         unlock_res_and_lock(lock);
1513                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1514                                              LDLM_CB_CANCELING);
1515                         lock_res_and_lock(lock);
1516                 } else {
1517                         LDLM_DEBUG(lock, "no blocking ast");
1518                 }
1519         }
1520         lock->l_flags |= LDLM_FL_BL_DONE;
1521 }
1522
1523 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1524 {
1525         if (req->l_resource->lr_type != LDLM_PLAIN &&
1526             req->l_resource->lr_type != LDLM_IBITS)
1527                 return;
1528
1529         list_del_init(&req->l_sl_policy);
1530         list_del_init(&req->l_sl_mode);
1531 }
1532
1533 void ldlm_lock_cancel(struct ldlm_lock *lock)
1534 {
1535         struct ldlm_resource *res;
1536         struct ldlm_namespace *ns;
1537         ENTRY;
1538
1539         lock_res_and_lock(lock);
1540
1541         res = lock->l_resource;
1542         ns = res->lr_namespace;
1543
1544         /* Please do not, no matter how tempting, remove this LBUG without
1545          * talking to me first. -phik */
1546         if (lock->l_readers || lock->l_writers) {
1547                 LDLM_ERROR(lock, "lock still has references");
1548                 LBUG();
1549         }
1550
1551         ldlm_del_waiting_lock(lock);
1552
1553         /* Releases cancel callback. */
1554         ldlm_cancel_callback(lock);
1555
1556         /* Yes, second time, just in case it was added again while we were
1557            running with no res lock in ldlm_cancel_callback */
1558         ldlm_del_waiting_lock(lock);
1559         ldlm_resource_unlink_lock(lock);
1560         ldlm_lock_destroy_nolock(lock);
1561
1562         if (lock->l_granted_mode == lock->l_req_mode)
1563                 ldlm_pool_del(&ns->ns_pool, lock);
1564
1565         /* Make sure we will not be called again for same lock what is possible
1566          * if not to zero out lock->l_granted_mode */
1567         lock->l_granted_mode = 0;
1568         unlock_res_and_lock(lock);
1569
1570         EXIT;
1571 }
1572
1573 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1574 {
1575         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1576         ENTRY;
1577
1578         if (lock == NULL)
1579                 RETURN(-EINVAL);
1580
1581         lock->l_ast_data = data;
1582         LDLM_LOCK_PUT(lock);
1583         RETURN(0);
1584 }
1585
1586 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1587 {
1588         struct ldlm_lock *lock;
1589         struct ldlm_resource *res;
1590
1591         spin_lock(&exp->exp_ldlm_data.led_lock);
1592         while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
1593                 lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
1594                                   struct ldlm_lock, l_export_chain);
1595                 res = ldlm_resource_getref(lock->l_resource);
1596                 LDLM_LOCK_GET(lock);
1597                 spin_unlock(&exp->exp_ldlm_data.led_lock);
1598
1599                 LDLM_DEBUG(lock, "export %p", exp);
1600                 ldlm_res_lvbo_update(res, NULL, 0, 1);
1601
1602                 ldlm_lock_cancel(lock);
1603                 ldlm_reprocess_all(res);
1604
1605                 ldlm_resource_putref(res);
1606                 LDLM_LOCK_PUT(lock);
1607                 spin_lock(&exp->exp_ldlm_data.led_lock);
1608         }
1609         spin_unlock(&exp->exp_ldlm_data.led_lock);
1610 }
1611
1612 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1613                                         __u32 *flags)
1614 {
1615         CFS_LIST_HEAD(rpc_list);
1616         struct ldlm_resource *res;
1617         struct ldlm_namespace *ns;
1618         int granted = 0;
1619         int old_mode, rc;
1620         struct sl_insert_point prev;
1621         ldlm_error_t err;
1622         struct ldlm_interval *node;
1623         ENTRY;
1624
1625         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1626                 *flags |= LDLM_FL_BLOCK_GRANTED;
1627                 RETURN(lock->l_resource);
1628         }
1629
1630         /* I can't check the type of lock here because the bitlock of lock
1631          * is not held here, so do the allocation blindly. -jay */
1632         OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO, sizeof(*node));
1633         if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
1634                 RETURN(NULL);
1635
1636         LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
1637                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1638
1639         lock_res_and_lock(lock);
1640
1641         res = lock->l_resource;
1642         ns = res->lr_namespace;
1643
1644         old_mode = lock->l_req_mode;
1645         lock->l_req_mode = new_mode;
1646         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1647                 /* remember the lock position where the lock might be
1648                  * added back to the granted list later and also
1649                  * remember the join mode for skiplist fixing. */
1650                 prev.res_link = lock->l_res_link.prev;
1651                 prev.mode_link = lock->l_sl_mode.prev;
1652                 prev.policy_link = lock->l_sl_policy.prev;
1653                 ldlm_resource_unlink_lock(lock);
1654         } else {
1655                 ldlm_resource_unlink_lock(lock);
1656                 if (res->lr_type == LDLM_EXTENT) {
1657                         /* FIXME: ugly code, I have to attach the lock to a
1658                          * interval node again since perhaps it will be granted
1659                          * soon */
1660                         CFS_INIT_LIST_HEAD(&node->li_group);
1661                         ldlm_interval_attach(node, lock);
1662                         node = NULL;
1663                 }
1664         }
1665
1666         /* If this is a local resource, put it on the appropriate list. */
1667         if (ns_is_client(res->lr_namespace)) {
1668                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1669                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1670                 } else {
1671                         /* This should never happen, because of the way the
1672                          * server handles conversions. */
1673                         LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1674                                    *flags);
1675                         LBUG();
1676
1677                         ldlm_grant_lock(lock, &rpc_list);
1678                         granted = 1;
1679                         /* FIXME: completion handling not with ns_lock held ! */
1680                         if (lock->l_completion_ast)
1681                                 lock->l_completion_ast(lock, 0, NULL);
1682                 }
1683         } else {
1684                 int pflags = 0;
1685                 ldlm_processing_policy policy;
1686                 policy = ldlm_processing_policy_table[res->lr_type];
1687                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1688                 if (rc == LDLM_ITER_STOP) {
1689                         lock->l_req_mode = old_mode;
1690                         if (res->lr_type == LDLM_EXTENT)
1691                                 ldlm_extent_add_lock(res, lock);
1692                         else
1693                                 ldlm_granted_list_add_lock(lock, &prev);
1694
1695                         res = NULL;
1696                 } else {
1697                         *flags |= LDLM_FL_BLOCK_GRANTED;
1698                         granted = 1;
1699                 }
1700         }
1701         unlock_res_and_lock(lock);
1702
1703         if (granted)
1704                 ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
1705         if (node)
1706                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1707         RETURN(res);
1708 }
1709
1710 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1711 {
1712         struct obd_device *obd = NULL;
1713
1714         if (!((libcfs_debug | D_ERROR) & level))
1715                 return;
1716
1717         if (!lock) {
1718                 CDEBUG(level, "  NULL LDLM lock\n");
1719                 return;
1720         }
1721
1722         CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1723                lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1724                pos, lock->l_pid);
1725         if (lock->l_conn_export != NULL)
1726                 obd = lock->l_conn_export->exp_obd;
1727         if (lock->l_export && lock->l_export->exp_connection) {
1728                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1729                      libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1730                      lock->l_remote_handle.cookie);
1731         } else if (obd == NULL) {
1732                 CDEBUG(level, "  Node: local\n");
1733         } else {
1734                 struct obd_import *imp = obd->u.cli.cl_import;
1735                 CDEBUG(level, "  Node: NID %s (rhandle: "LPX64")\n",
1736                        libcfs_nid2str(imp->imp_connection->c_peer.nid),
1737                        lock->l_remote_handle.cookie);
1738         }
1739         CDEBUG(level, "  Resource: %p ("LPU64"/"LPU64"/"LPU64")\n",
1740                   lock->l_resource,
1741                   lock->l_resource->lr_name.name[0],
1742                   lock->l_resource->lr_name.name[1],
1743                   lock->l_resource->lr_name.name[2]);
1744         CDEBUG(level, "  Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1745                "write: %d flags: %#x\n", ldlm_lockname[lock->l_req_mode],
1746                ldlm_lockname[lock->l_granted_mode],
1747                atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1748                lock->l_flags);
1749         if (lock->l_resource->lr_type == LDLM_EXTENT)
1750                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64
1751                        " (req "LPU64"-"LPU64")\n",
1752                        lock->l_policy_data.l_extent.start,
1753                        lock->l_policy_data.l_extent.end,
1754                        lock->l_req_extent.start, lock->l_req_extent.end);
1755         else if (lock->l_resource->lr_type == LDLM_FLOCK)
1756                 CDEBUG(level, "  Pid: %d Extent: "LPU64" -> "LPU64"\n",
1757                        lock->l_policy_data.l_flock.pid,
1758                        lock->l_policy_data.l_flock.start,
1759                        lock->l_policy_data.l_flock.end);
1760        else if (lock->l_resource->lr_type == LDLM_IBITS)
1761                 CDEBUG(level, "  Bits: "LPX64"\n",
1762                        lock->l_policy_data.l_inodebits.bits);
1763 }
1764
1765 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1766 {
1767         struct ldlm_lock *lock;
1768
1769         if (!((libcfs_debug | D_ERROR) & level))
1770                 return;
1771
1772         lock = ldlm_handle2lock(lockh);
1773         if (lock == NULL)
1774                 return;
1775
1776         ldlm_lock_dump(D_OTHER, lock, 0);
1777
1778         LDLM_LOCK_PUT(lock);
1779 }
1780
1781 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1782                       struct libcfs_debug_msg_data *data, const char *fmt,
1783                       ...)
1784 {
1785         va_list args;
1786         cfs_debug_limit_state_t *cdls = data->msg_cdls;
1787
1788         va_start(args, fmt);
1789
1790         if (lock->l_resource == NULL) {
1791                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1792                                    data->msg_fn, data->msg_line, fmt, args,
1793                        " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1794                        "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: "
1795                        LPX64" expref: %d pid: %u\n", lock,
1796                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1797                        lock->l_readers, lock->l_writers,
1798                        ldlm_lockname[lock->l_granted_mode],
1799                        ldlm_lockname[lock->l_req_mode],
1800                        lock->l_flags, lock->l_remote_handle.cookie,
1801                        lock->l_export ?
1802                        atomic_read(&lock->l_export->exp_refcount) : -99,
1803                        lock->l_pid);
1804                 va_end(args);
1805                 return;
1806         }
1807
1808         switch (lock->l_resource->lr_type) {
1809         case LDLM_EXTENT:
1810                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1811                                    data->msg_fn, data->msg_line, fmt, args,
1812                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1813                        "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1814                        "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64
1815                        " expref: %d pid: %u\n",
1816                        lock->l_resource->lr_namespace->ns_name, lock,
1817                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1818                        lock->l_readers, lock->l_writers,
1819                        ldlm_lockname[lock->l_granted_mode],
1820                        ldlm_lockname[lock->l_req_mode],
1821                        lock->l_resource->lr_name.name[0],
1822                        lock->l_resource->lr_name.name[1],
1823                        atomic_read(&lock->l_resource->lr_refcount),
1824                        ldlm_typename[lock->l_resource->lr_type],
1825                        lock->l_policy_data.l_extent.start,
1826                        lock->l_policy_data.l_extent.end,
1827                        lock->l_req_extent.start, lock->l_req_extent.end,
1828                        lock->l_flags, lock->l_remote_handle.cookie,
1829                        lock->l_export ?
1830                        atomic_read(&lock->l_export->exp_refcount) : -99,
1831                        lock->l_pid);
1832                 break;
1833
1834         case LDLM_FLOCK:
1835                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1836                                    data->msg_fn, data->msg_line, fmt, args,
1837                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1838                        "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1839                        "["LPU64"->"LPU64"] flags: %x remote: "LPX64
1840                        " expref: %d pid: %u\n",
1841                        lock->l_resource->lr_namespace->ns_name, lock,
1842                        lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1843                        lock->l_readers, lock->l_writers,
1844                        ldlm_lockname[lock->l_granted_mode],
1845                        ldlm_lockname[lock->l_req_mode],
1846                        lock->l_resource->lr_name.name[0],
1847                        lock->l_resource->lr_name.name[1],
1848                        atomic_read(&lock->l_resource->lr_refcount),
1849                        ldlm_typename[lock->l_resource->lr_type],
1850                        lock->l_policy_data.l_flock.pid,
1851                        lock->l_policy_data.l_flock.start,
1852                        lock->l_policy_data.l_flock.end,
1853                        lock->l_flags, lock->l_remote_handle.cookie,
1854                        lock->l_export ?
1855                        atomic_read(&lock->l_export->exp_refcount) : -99,
1856                        lock->l_pid);
1857                 break;
1858
1859         case LDLM_IBITS:
1860                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1861                                    data->msg_fn, data->msg_line, fmt, args,
1862                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1863                        "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1864                        "flags: %x remote: "LPX64" expref: %d "
1865                        "pid %u\n",
1866                        lock->l_resource->lr_namespace->ns_name,
1867                        lock, lock->l_handle.h_cookie,
1868                        atomic_read (&lock->l_refc),
1869                        lock->l_readers, lock->l_writers,
1870                        ldlm_lockname[lock->l_granted_mode],
1871                        ldlm_lockname[lock->l_req_mode],
1872                        lock->l_resource->lr_name.name[0],
1873                        lock->l_resource->lr_name.name[1],
1874                        lock->l_policy_data.l_inodebits.bits,
1875                        atomic_read(&lock->l_resource->lr_refcount),
1876                        ldlm_typename[lock->l_resource->lr_type],
1877                        lock->l_flags, lock->l_remote_handle.cookie,
1878                        lock->l_export ?
1879                        atomic_read(&lock->l_export->exp_refcount) : -99,
1880                        lock->l_pid);
1881                 break;
1882
1883         default:
1884                 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1885                                    data->msg_fn, data->msg_line, fmt, args,
1886                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1887                        "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x "
1888                        "remote: "LPX64" expref: %d pid: %u\n",
1889                        lock->l_resource->lr_namespace->ns_name,
1890                        lock, lock->l_handle.h_cookie,
1891                        atomic_read (&lock->l_refc),
1892                        lock->l_readers, lock->l_writers,
1893                        ldlm_lockname[lock->l_granted_mode],
1894                        ldlm_lockname[lock->l_req_mode],
1895                        lock->l_resource->lr_name.name[0],
1896                        lock->l_resource->lr_name.name[1],
1897                        atomic_read(&lock->l_resource->lr_refcount),
1898                        ldlm_typename[lock->l_resource->lr_type],
1899                        lock->l_flags, lock->l_remote_handle.cookie,
1900                        lock->l_export ?
1901                        atomic_read(&lock->l_export->exp_refcount) : -99,
1902                        lock->l_pid);
1903                 break;
1904         }
1905         va_end(args);
1906 }
1907 EXPORT_SYMBOL(_ldlm_lock_debug);