lustre/ldlm/ldlm_lock.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ldlm/ldlm_lock.c
  37  *
  38  * Author: Peter Braam <braam@clusterfs.com>
  39  * Author: Phil Schwan <phil@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LDLM
  43
  44 #ifdef __KERNEL__
  45 # include <libcfs/libcfs.h>
  46 # include <linux/lustre_intent.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <obd_class.h>
  52 #include "ldlm_internal.h"
  53
  54 /* lock types */
  55 char *ldlm_lockname[] = {
  56         [0] "--",
  57         [LCK_EX] "EX",
  58         [LCK_PW] "PW",
  59         [LCK_PR] "PR",
  60         [LCK_CW] "CW",
  61         [LCK_CR] "CR",
  62         [LCK_NL] "NL",
  63         [LCK_GROUP] "GROUP",
  64         [LCK_COS] "COS"
  65 };
  66 EXPORT_SYMBOL(ldlm_lockname);
  67
  68 char *ldlm_typename[] = {
  69         [LDLM_PLAIN] "PLN",
  70         [LDLM_EXTENT] "EXT",
  71         [LDLM_FLOCK] "FLK",
  72         [LDLM_IBITS] "IBT",
  73 };
  74 EXPORT_SYMBOL(ldlm_typename);
  75
  76 static ldlm_policy_wire_to_local_t ldlm_policy_wire18_to_local[] = {
  77         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local,
  78         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local,
  79         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire18_to_local,
  80         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local,
  81 };
  82
  83 static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = {
  84         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local,
  85         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local,
  86         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire21_to_local,
  87         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local,
  88 };
  89
  90 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
  91         [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_local_to_wire,
  92         [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_local_to_wire,
  93         [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_local_to_wire,
  94         [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_local_to_wire,
  95 };
  96
  97 /**
  98  * Converts lock policy from local format to on the wire lock_desc format
  99  */
 100 void ldlm_convert_policy_to_wire(ldlm_type_t type,
 101                                  const ldlm_policy_data_t *lpolicy,
 102                                  ldlm_wire_policy_data_t *wpolicy)
 103 {
 104         ldlm_policy_local_to_wire_t convert;
 105
 106         convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
 107
 108         convert(lpolicy, wpolicy);
 109 }
 110
 111 /**
 112  * Converts lock policy from on the wire lock_desc format to local format
 113  */
 114 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
 115                                   const ldlm_wire_policy_data_t *wpolicy,
 116                                   ldlm_policy_data_t *lpolicy)
 117 {
 118         ldlm_policy_wire_to_local_t convert;
 119         int new_client;
 120
 121         /** some badnes for 2.0.0 clients, but 2.0.0 isn't supported */
 122         new_client = (exp->exp_connect_flags & OBD_CONNECT_FULL20) != 0;
 123         if (new_client)
 124                convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE];
 125         else
 126                convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE];
 127
 128         convert(wpolicy, lpolicy);
 129 }
 130
 131 char *ldlm_it2str(int it)
 132 {
 133         switch (it) {
 134         case IT_OPEN:
 135                 return "open";
 136         case IT_CREAT:
 137                 return "creat";
 138         case (IT_OPEN | IT_CREAT):
 139                 return "open|creat";
 140         case IT_READDIR:
 141                 return "readdir";
 142         case IT_GETATTR:
 143                 return "getattr";
 144         case IT_LOOKUP:
 145                 return "lookup";
 146         case IT_UNLINK:
 147                 return "unlink";
 148         case IT_GETXATTR:
 149                 return "getxattr";
 150         case IT_LAYOUT:
 151                 return "layout";
 152         default:
 153                 CERROR("Unknown intent %d\n", it);
 154                 return "UNKNOWN";
 155         }
 156 }
 157 EXPORT_SYMBOL(ldlm_it2str);
 158
 159 extern cfs_mem_cache_t *ldlm_lock_slab;
 160
 161 #ifdef HAVE_SERVER_SUPPORT
 162 static ldlm_processing_policy ldlm_processing_policy_table[] = {
 163         [LDLM_PLAIN] ldlm_process_plain_lock,
 164         [LDLM_EXTENT] ldlm_process_extent_lock,
 165 # ifdef __KERNEL__
 166         [LDLM_FLOCK] ldlm_process_flock_lock,
 167 # endif
 168         [LDLM_IBITS] ldlm_process_inodebits_lock,
 169 };
 170
 171 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
 172 {
 173         return ldlm_processing_policy_table[res->lr_type];
 174 }
 175 EXPORT_SYMBOL(ldlm_get_processing_policy);
 176 #endif /* HAVE_SERVER_SUPPORT */
 177
 178 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
 179 {
 180         ns->ns_policy = arg;
 181 }
 182 EXPORT_SYMBOL(ldlm_register_intent);
 183
 184 /*
 185  * REFCOUNTED LOCK OBJECTS
 186  */
 187
 188
 189 /*
 190  * Lock refcounts, during creation:
 191  *   - one special one for allocation, dec'd only once in destroy
 192  *   - one for being a lock that's in-use
 193  *   - one for the addref associated with a new lock
 194  */
 195 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 196 {
 197         cfs_atomic_inc(&lock->l_refc);
 198         return lock;
 199 }
 200 EXPORT_SYMBOL(ldlm_lock_get);
 201
 202 void ldlm_lock_put(struct ldlm_lock *lock)
 203 {
 204         ENTRY;
 205
 206         LASSERT(lock->l_resource != LP_POISON);
 207         LASSERT(cfs_atomic_read(&lock->l_refc) > 0);
 208         if (cfs_atomic_dec_and_test(&lock->l_refc)) {
 209                 struct ldlm_resource *res;
 210
 211                 LDLM_DEBUG(lock,
 212                            "final lock_put on destroyed lock, freeing it.");
 213
 214                 res = lock->l_resource;
 215                 LASSERT(lock->l_destroyed);
 216                 LASSERT(cfs_list_empty(&lock->l_res_link));
 217                 LASSERT(cfs_list_empty(&lock->l_pending_chain));
 218
 219                 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
 220                                      LDLM_NSS_LOCKS);
 221                 lu_ref_del(&res->lr_reference, "lock", lock);
 222                 ldlm_resource_putref(res);
 223                 lock->l_resource = NULL;
 224                 if (lock->l_export) {
 225                         class_export_lock_put(lock->l_export, lock);
 226                         lock->l_export = NULL;
 227                 }
 228
 229                 if (lock->l_lvb_data != NULL)
 230                         OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
 231
 232                 ldlm_interval_free(ldlm_interval_detach(lock));
 233                 lu_ref_fini(&lock->l_reference);
 234                 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
 235         }
 236
 237         EXIT;
 238 }
 239 EXPORT_SYMBOL(ldlm_lock_put);
 240
 241 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
 242 {
 243         int rc = 0;
 244         if (!cfs_list_empty(&lock->l_lru)) {
 245                 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 246
 247                 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 248                 cfs_list_del_init(&lock->l_lru);
 249                 if (lock->l_flags & LDLM_FL_SKIPPED)
 250                         lock->l_flags &= ~LDLM_FL_SKIPPED;
 251                 LASSERT(ns->ns_nr_unused > 0);
 252                 ns->ns_nr_unused--;
 253                 rc = 1;
 254         }
 255         return rc;
 256 }
 257
 258 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 259 {
 260         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 261         int rc;
 262
 263         ENTRY;
 264         if (lock->l_ns_srv) {
 265                 LASSERT(cfs_list_empty(&lock->l_lru));
 266                 RETURN(0);
 267         }
 268
 269         spin_lock(&ns->ns_lock);
 270         rc = ldlm_lock_remove_from_lru_nolock(lock);
 271         spin_unlock(&ns->ns_lock);
 272         EXIT;
 273         return rc;
 274 }
 275
 276 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
 277 {
 278         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 279
 280         lock->l_last_used = cfs_time_current();
 281         LASSERT(cfs_list_empty(&lock->l_lru));
 282         LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
 283         cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 284         LASSERT(ns->ns_nr_unused >= 0);
 285         ns->ns_nr_unused++;
 286 }
 287
 288 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
 289 {
 290         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 291
 292         ENTRY;
 293         spin_lock(&ns->ns_lock);
 294         ldlm_lock_add_to_lru_nolock(lock);
 295         spin_unlock(&ns->ns_lock);
 296         EXIT;
 297 }
 298
 299 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
 300 {
 301         struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 302
 303         ENTRY;
 304         if (lock->l_ns_srv) {
 305                 LASSERT(cfs_list_empty(&lock->l_lru));
 306                 EXIT;
 307                 return;
 308         }
 309
 310         spin_lock(&ns->ns_lock);
 311         if (!cfs_list_empty(&lock->l_lru)) {
 312                 ldlm_lock_remove_from_lru_nolock(lock);
 313                 ldlm_lock_add_to_lru_nolock(lock);
 314         }
 315         spin_unlock(&ns->ns_lock);
 316         EXIT;
 317 }
 318
 319 /* This used to have a 'strict' flag, which recovery would use to mark an
 320  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 321  * shall explain why it's gone: with the new hash table scheme, once you call
 322  * ldlm_lock_destroy, you can never drop your final references on this lock.
 323  * Because it's not in the hash table anymore.  -phil */
 324 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
 325 {
 326         ENTRY;
 327
 328         if (lock->l_readers || lock->l_writers) {
 329                 LDLM_ERROR(lock, "lock still has references");
 330                 LBUG();
 331         }
 332
 333         if (!cfs_list_empty(&lock->l_res_link)) {
 334                 LDLM_ERROR(lock, "lock still on resource");
 335                 LBUG();
 336         }
 337
 338         if (lock->l_destroyed) {
 339                 LASSERT(cfs_list_empty(&lock->l_lru));
 340                 EXIT;
 341                 return 0;
 342         }
 343         lock->l_destroyed = 1;
 344
 345         if (lock->l_export && lock->l_export->exp_lock_hash) {
 346                 /* NB: it's safe to call cfs_hash_del() even lock isn't
 347                  * in exp_lock_hash. */
 348                 /* In the function below, .hs_keycmp resolves to
 349                  * ldlm_export_lock_keycmp() */
 350                 /* coverity[overrun-buffer-val] */
 351                 cfs_hash_del(lock->l_export->exp_lock_hash,
 352                              &lock->l_remote_handle, &lock->l_exp_hash);
 353         }
 354
 355         ldlm_lock_remove_from_lru(lock);
 356         class_handle_unhash(&lock->l_handle);
 357
 358 #if 0
 359         /* Wake anyone waiting for this lock */
 360         /* FIXME: I should probably add yet another flag, instead of using
 361          * l_export to only call this on clients */
 362         if (lock->l_export)
 363                 class_export_put(lock->l_export);
 364         lock->l_export = NULL;
 365         if (lock->l_export && lock->l_completion_ast)
 366                 lock->l_completion_ast(lock, 0);
 367 #endif
 368         EXIT;
 369         return 1;
 370 }
 371
 372 void ldlm_lock_destroy(struct ldlm_lock *lock)
 373 {
 374         int first;
 375         ENTRY;
 376         lock_res_and_lock(lock);
 377         first = ldlm_lock_destroy_internal(lock);
 378         unlock_res_and_lock(lock);
 379
 380         /* drop reference from hashtable only for first destroy */
 381         if (first) {
 382                 lu_ref_del(&lock->l_reference, "hash", lock);
 383                 LDLM_LOCK_RELEASE(lock);
 384         }
 385         EXIT;
 386 }
 387
 388 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
 389 {
 390         int first;
 391         ENTRY;
 392         first = ldlm_lock_destroy_internal(lock);
 393         /* drop reference from hashtable only for first destroy */
 394         if (first) {
 395                 lu_ref_del(&lock->l_reference, "hash", lock);
 396                 LDLM_LOCK_RELEASE(lock);
 397         }
 398         EXIT;
 399 }
 400
 401 /* this is called by portals_handle2object with the handle lock taken */
 402 static void lock_handle_addref(void *lock)
 403 {
 404         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 405 }
 406
 407 static void lock_handle_free(void *lock, int size)
 408 {
 409         LASSERT(size == sizeof(struct ldlm_lock));
 410         OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
 411 }
 412
 413 struct portals_handle_ops lock_handle_ops = {
 414         .hop_addref = lock_handle_addref,
 415         .hop_free   = lock_handle_free,
 416 };
 417
 418 /*
 419  * usage: pass in a resource on which you have done ldlm_resource_get
 420  *        new lock will take over the refcount.
 421  * returns: lock with refcount 2 - one for current caller and one for remote
 422  */
 423 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
 424 {
 425         struct ldlm_lock *lock;
 426         ENTRY;
 427
 428         if (resource == NULL)
 429                 LBUG();
 430
 431         OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO);
 432         if (lock == NULL)
 433                 RETURN(NULL);
 434
 435         spin_lock_init(&lock->l_lock);
 436         lock->l_resource = resource;
 437         lu_ref_add(&resource->lr_reference, "lock", lock);
 438
 439         cfs_atomic_set(&lock->l_refc, 2);
 440         CFS_INIT_LIST_HEAD(&lock->l_res_link);
 441         CFS_INIT_LIST_HEAD(&lock->l_lru);
 442         CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
 443         CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
 444         CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
 445         CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
 446         cfs_waitq_init(&lock->l_waitq);
 447         lock->l_blocking_lock = NULL;
 448         CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
 449         CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
 450         CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
 451         CFS_INIT_HLIST_NODE(&lock->l_exp_flock_hash);
 452
 453         lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
 454                              LDLM_NSS_LOCKS);
 455         CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
 456         class_handle_hash(&lock->l_handle, &lock_handle_ops);
 457
 458         lu_ref_init(&lock->l_reference);
 459         lu_ref_add(&lock->l_reference, "hash", lock);
 460         lock->l_callback_timeout = 0;
 461
 462 #if LUSTRE_TRACKS_LOCK_EXP_REFS
 463         CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
 464         lock->l_exp_refs_nr = 0;
 465         lock->l_exp_refs_target = NULL;
 466 #endif
 467         CFS_INIT_LIST_HEAD(&lock->l_exp_list);
 468
 469         RETURN(lock);
 470 }
 471
 472 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 473                               const struct ldlm_res_id *new_resid)
 474 {
 475         struct ldlm_resource *oldres = lock->l_resource;
 476         struct ldlm_resource *newres;
 477         int type;
 478         ENTRY;
 479
 480         LASSERT(ns_is_client(ns));
 481
 482         lock_res_and_lock(lock);
 483         if (memcmp(new_resid, &lock->l_resource->lr_name,
 484                    sizeof(lock->l_resource->lr_name)) == 0) {
 485                 /* Nothing to do */
 486                 unlock_res_and_lock(lock);
 487                 RETURN(0);
 488         }
 489
 490         LASSERT(new_resid->name[0] != 0);
 491
 492         /* This function assumes that the lock isn't on any lists */
 493         LASSERT(cfs_list_empty(&lock->l_res_link));
 494
 495         type = oldres->lr_type;
 496         unlock_res_and_lock(lock);
 497
 498         newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 499         if (newres == NULL)
 500                 RETURN(-ENOMEM);
 501
 502         lu_ref_add(&newres->lr_reference, "lock", lock);
 503         /*
 504          * To flip the lock from the old to the new resource, lock, oldres and
 505          * newres have to be locked. Resource spin-locks are nested within
 506          * lock->l_lock, and are taken in the memory address order to avoid
 507          * dead-locks.
 508          */
 509         spin_lock(&lock->l_lock);
 510         oldres = lock->l_resource;
 511         if (oldres < newres) {
 512                 lock_res(oldres);
 513                 lock_res_nested(newres, LRT_NEW);
 514         } else {
 515                 lock_res(newres);
 516                 lock_res_nested(oldres, LRT_NEW);
 517         }
 518         LASSERT(memcmp(new_resid, &oldres->lr_name,
 519                        sizeof oldres->lr_name) != 0);
 520         lock->l_resource = newres;
 521         unlock_res(oldres);
 522         unlock_res_and_lock(lock);
 523
 524         /* ...and the flowers are still standing! */
 525         lu_ref_del(&oldres->lr_reference, "lock", lock);
 526         ldlm_resource_putref(oldres);
 527
 528         RETURN(0);
 529 }
 530 EXPORT_SYMBOL(ldlm_lock_change_resource);
 531
 532 /*
 533  *  HANDLES
 534  */
 535
 536 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
 537 {
 538         lockh->cookie = lock->l_handle.h_cookie;
 539 }
 540 EXPORT_SYMBOL(ldlm_lock2handle);
 541
 542 /* if flags: atomically get the lock and set the flags.
 543  *           Return NULL if flag already set
 544  */
 545
 546 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
 547                                      __u64 flags)
 548 {
 549         struct ldlm_lock *lock;
 550         ENTRY;
 551
 552         LASSERT(handle);
 553
 554         lock = class_handle2object(handle->cookie);
 555         if (lock == NULL)
 556                 RETURN(NULL);
 557
 558         /* It's unlikely but possible that someone marked the lock as
 559          * destroyed after we did handle2object on it */
 560         if (flags == 0 && !lock->l_destroyed) {
 561                 lu_ref_add(&lock->l_reference, "handle", cfs_current());
 562                 RETURN(lock);
 563         }
 564
 565         lock_res_and_lock(lock);
 566
 567         LASSERT(lock->l_resource != NULL);
 568
 569         lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
 570         if (unlikely(lock->l_destroyed)) {
 571                 unlock_res_and_lock(lock);
 572                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 573                 LDLM_LOCK_PUT(lock);
 574                 RETURN(NULL);
 575         }
 576
 577         if (flags && (lock->l_flags & flags)) {
 578                 unlock_res_and_lock(lock);
 579                 LDLM_LOCK_PUT(lock);
 580                 RETURN(NULL);
 581         }
 582
 583         if (flags)
 584                 lock->l_flags |= flags;
 585
 586         unlock_res_and_lock(lock);
 587         RETURN(lock);
 588 }
 589 EXPORT_SYMBOL(__ldlm_handle2lock);
 590
 591 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 592 {
 593         struct obd_export *exp = lock->l_export?:lock->l_conn_export;
 594         /* INODEBITS_INTEROP: If the other side does not support
 595          * inodebits, reply with a plain lock descriptor.
 596          */
 597         if ((lock->l_resource->lr_type == LDLM_IBITS) &&
 598             (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
 599                 /* Make sure all the right bits are set in this lock we
 600                    are going to pass to client */
 601                 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
 602                          (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
 603                           MDS_INODELOCK_LAYOUT),
 604                          "Inappropriate inode lock bits during "
 605                          "conversion " LPU64 "\n",
 606                          lock->l_policy_data.l_inodebits.bits);
 607
 608                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 609                 desc->l_resource.lr_type = LDLM_PLAIN;
 610
 611                 /* Convert "new" lock mode to something old client can
 612                    understand */
 613                 if ((lock->l_req_mode == LCK_CR) ||
 614                     (lock->l_req_mode == LCK_CW))
 615                         desc->l_req_mode = LCK_PR;
 616                 else
 617                         desc->l_req_mode = lock->l_req_mode;
 618                 if ((lock->l_granted_mode == LCK_CR) ||
 619                     (lock->l_granted_mode == LCK_CW)) {
 620                         desc->l_granted_mode = LCK_PR;
 621                 } else {
 622                         /* We never grant PW/EX locks to clients */
 623                         LASSERT((lock->l_granted_mode != LCK_PW) &&
 624                                 (lock->l_granted_mode != LCK_EX));
 625                         desc->l_granted_mode = lock->l_granted_mode;
 626                 }
 627
 628                 /* We do not copy policy here, because there is no
 629                    policy for plain locks */
 630         } else {
 631                 ldlm_res2desc(lock->l_resource, &desc->l_resource);
 632                 desc->l_req_mode = lock->l_req_mode;
 633                 desc->l_granted_mode = lock->l_granted_mode;
 634                 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
 635                                             &lock->l_policy_data,
 636                                             &desc->l_policy_data);
 637         }
 638 }
 639 EXPORT_SYMBOL(ldlm_lock2desc);
 640
 641 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 642                            cfs_list_t *work_list)
 643 {
 644         if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
 645                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 646                 lock->l_flags |= LDLM_FL_AST_SENT;
 647                 /* If the enqueuing client said so, tell the AST recipient to
 648                  * discard dirty data, rather than writing back. */
 649                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 650                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 651                 LASSERT(cfs_list_empty(&lock->l_bl_ast));
 652                 cfs_list_add(&lock->l_bl_ast, work_list);
 653                 LDLM_LOCK_GET(lock);
 654                 LASSERT(lock->l_blocking_lock == NULL);
 655                 lock->l_blocking_lock = LDLM_LOCK_GET(new);
 656         }
 657 }
 658
 659 void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list)
 660 {
 661         if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
 662                 lock->l_flags |= LDLM_FL_CP_REQD;
 663                 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
 664                 LASSERT(cfs_list_empty(&lock->l_cp_ast));
 665                 cfs_list_add(&lock->l_cp_ast, work_list);
 666                 LDLM_LOCK_GET(lock);
 667         }
 668 }
 669
 670 /* must be called with lr_lock held */
 671 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 672                             cfs_list_t *work_list)
 673 {
 674         ENTRY;
 675         check_res_locked(lock->l_resource);
 676         if (new)
 677                 ldlm_add_bl_work_item(lock, new, work_list);
 678         else
 679                 ldlm_add_cp_work_item(lock, work_list);
 680         EXIT;
 681 }
 682
 683 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 684 {
 685         struct ldlm_lock *lock;
 686
 687         lock = ldlm_handle2lock(lockh);
 688         LASSERT(lock != NULL);
 689         ldlm_lock_addref_internal(lock, mode);
 690         LDLM_LOCK_PUT(lock);
 691 }
 692 EXPORT_SYMBOL(ldlm_lock_addref);
 693
 694 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 695 {
 696         ldlm_lock_remove_from_lru(lock);
 697         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 698                 lock->l_readers++;
 699                 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
 700         }
 701         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
 702                 lock->l_writers++;
 703                 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
 704         }
 705         LDLM_LOCK_GET(lock);
 706         lu_ref_add_atomic(&lock->l_reference, "user", lock);
 707         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 708 }
 709
 710 /**
 711  * Attempts to addref a lock, and fails if lock is already LDLM_FL_CBPENDING
 712  * or destroyed.
 713  *
 714  * \retval 0 success, lock was addref-ed
 715  *
 716  * \retval -EAGAIN lock is being canceled.
 717  */
 718 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
 719 {
 720         struct ldlm_lock *lock;
 721         int               result;
 722
 723         result = -EAGAIN;
 724         lock = ldlm_handle2lock(lockh);
 725         if (lock != NULL) {
 726                 lock_res_and_lock(lock);
 727                 if (lock->l_readers != 0 || lock->l_writers != 0 ||
 728                     !(lock->l_flags & LDLM_FL_CBPENDING)) {
 729                         ldlm_lock_addref_internal_nolock(lock, mode);
 730                         result = 0;
 731                 }
 732                 unlock_res_and_lock(lock);
 733                 LDLM_LOCK_PUT(lock);
 734         }
 735         return result;
 736 }
 737 EXPORT_SYMBOL(ldlm_lock_addref_try);
 738
 739 /* only called for local locks */
 740 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 741 {
 742         lock_res_and_lock(lock);
 743         ldlm_lock_addref_internal_nolock(lock, mode);
 744         unlock_res_and_lock(lock);
 745 }
 746
 747 /* only called in ldlm_flock_destroy and for local locks.
 748  *  * for LDLM_FLOCK type locks, l_blocking_ast is null, and
 749  *   * ldlm_lock_remove_from_lru() does nothing, it is safe
 750  *    * for ldlm_flock_destroy usage by dropping some code */
 751 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
 752 {
 753         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 754         if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
 755                 LASSERT(lock->l_readers > 0);
 756                 lu_ref_del(&lock->l_reference, "reader", lock);
 757                 lock->l_readers--;
 758         }
 759         if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
 760                 LASSERT(lock->l_writers > 0);
 761                 lu_ref_del(&lock->l_reference, "writer", lock);
 762                 lock->l_writers--;
 763         }
 764
 765         lu_ref_del(&lock->l_reference, "user", lock);
 766         LDLM_LOCK_RELEASE(lock);    /* matches the LDLM_LOCK_GET() in addref */
 767 }
 768
 769 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 770 {
 771         struct ldlm_namespace *ns;
 772         ENTRY;
 773
 774         lock_res_and_lock(lock);
 775
 776         ns = ldlm_lock_to_ns(lock);
 777
 778         ldlm_lock_decref_internal_nolock(lock, mode);
 779
 780         if (lock->l_flags & LDLM_FL_LOCAL &&
 781             !lock->l_readers && !lock->l_writers) {
 782                 /* If this is a local lock on a server namespace and this was
 783                  * the last reference, cancel the lock. */
 784                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 785                 lock->l_flags |= LDLM_FL_CBPENDING;
 786         }
 787
 788         if (!lock->l_readers && !lock->l_writers &&
 789             (lock->l_flags & LDLM_FL_CBPENDING)) {
 790                 /* If we received a blocked AST and this was the last reference,
 791                  * run the callback. */
 792                 if (lock->l_ns_srv && lock->l_export)
 793                         CERROR("FL_CBPENDING set on non-local lock--just a "
 794                                "warning\n");
 795
 796                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 797
 798                 LDLM_LOCK_GET(lock); /* dropped by bl thread */
 799                 ldlm_lock_remove_from_lru(lock);
 800                 unlock_res_and_lock(lock);
 801
 802                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
 803                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
 804
 805                 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
 806                     ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
 807                         ldlm_handle_bl_callback(ns, NULL, lock);
 808         } else if (ns_is_client(ns) &&
 809                    !lock->l_readers && !lock->l_writers &&
 810                    !(lock->l_flags & LDLM_FL_NO_LRU) &&
 811                    !(lock->l_flags & LDLM_FL_BL_AST)) {
 812
 813                 LDLM_DEBUG(lock, "add lock into lru list");
 814
 815                 /* If this is a client-side namespace and this was the last
 816                  * reference, put it on the LRU. */
 817                 ldlm_lock_add_to_lru(lock);
 818                 unlock_res_and_lock(lock);
 819
 820                 if (lock->l_flags & LDLM_FL_FAIL_LOC)
 821                         OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
 822
 823                 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
 824                  * are not supported by the server, otherwise, it is done on
 825                  * enqueue. */
 826                 if (!exp_connect_cancelset(lock->l_conn_export) &&
 827                     !ns_connect_lru_resize(ns))
 828                         ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
 829         } else {
 830                 LDLM_DEBUG(lock, "do not add lock into lru list");
 831                 unlock_res_and_lock(lock);
 832         }
 833
 834         EXIT;
 835 }
 836
 837 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 838 {
 839         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 840         LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
 841         ldlm_lock_decref_internal(lock, mode);
 842         LDLM_LOCK_PUT(lock);
 843 }
 844 EXPORT_SYMBOL(ldlm_lock_decref);
 845
 846 /* This will drop a lock reference and mark it for destruction, but will not
 847  * necessarily cancel the lock before returning. */
 848 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 849 {
 850         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 851         ENTRY;
 852
 853         LASSERT(lock != NULL);
 854
 855         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 856         lock_res_and_lock(lock);
 857         lock->l_flags |= LDLM_FL_CBPENDING;
 858         unlock_res_and_lock(lock);
 859         ldlm_lock_decref_internal(lock, mode);
 860         LDLM_LOCK_PUT(lock);
 861 }
 862 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
 863
 864 struct sl_insert_point {
 865         cfs_list_t *res_link;
 866         cfs_list_t *mode_link;
 867         cfs_list_t *policy_link;
 868 };
 869
 870 /*
 871  * search_granted_lock
 872  *
 873  * Description:
 874  *      Finds a position to insert the new lock.
 875  * Parameters:
 876  *      queue [input]:  the granted list where search acts on;
 877  *      req [input]:    the lock whose position to be located;
 878  *      prev [output]:  positions within 3 lists to insert @req to
 879  * Return Value:
 880  *      filled @prev
 881  * NOTE: called by
 882  *  - ldlm_grant_lock_with_skiplist
 883  */
 884 static void search_granted_lock(cfs_list_t *queue,
 885                                 struct ldlm_lock *req,
 886                                 struct sl_insert_point *prev)
 887 {
 888         cfs_list_t *tmp;
 889         struct ldlm_lock *lock, *mode_end, *policy_end;
 890         ENTRY;
 891
 892         cfs_list_for_each(tmp, queue) {
 893                 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
 894
 895                 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
 896                                           struct ldlm_lock, l_sl_mode);
 897
 898                 if (lock->l_req_mode != req->l_req_mode) {
 899                         /* jump to last lock of mode group */
 900                         tmp = &mode_end->l_res_link;
 901                         continue;
 902                 }
 903
 904                 /* suitable mode group is found */
 905                 if (lock->l_resource->lr_type == LDLM_PLAIN) {
 906                         /* insert point is last lock of the mode group */
 907                         prev->res_link = &mode_end->l_res_link;
 908                         prev->mode_link = &mode_end->l_sl_mode;
 909                         prev->policy_link = &req->l_sl_policy;
 910                         EXIT;
 911                         return;
 912                 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
 913                         for (;;) {
 914                                 policy_end =
 915                                         cfs_list_entry(lock->l_sl_policy.prev,
 916                                                        struct ldlm_lock,
 917                                                        l_sl_policy);
 918
 919                                 if (lock->l_policy_data.l_inodebits.bits ==
 920                                     req->l_policy_data.l_inodebits.bits) {
 921                                         /* insert point is last lock of
 922                                          * the policy group */
 923                                         prev->res_link =
 924                                                 &policy_end->l_res_link;
 925                                         prev->mode_link =
 926                                                 &policy_end->l_sl_mode;
 927                                         prev->policy_link =
 928                                                 &policy_end->l_sl_policy;
 929                                         EXIT;
 930                                         return;
 931                                 }
 932
 933                                 if (policy_end == mode_end)
 934                                         /* done with mode group */
 935                                         break;
 936
 937                                 /* go to next policy group within mode group */
 938                                 tmp = policy_end->l_res_link.next;
 939                                 lock = cfs_list_entry(tmp, struct ldlm_lock,
 940                                                       l_res_link);
 941                         }  /* loop over policy groups within the mode group */
 942
 943                         /* insert point is last lock of the mode group,
 944                          * new policy group is started */
 945                         prev->res_link = &mode_end->l_res_link;
 946                         prev->mode_link = &mode_end->l_sl_mode;
 947                         prev->policy_link = &req->l_sl_policy;
 948                         EXIT;
 949                         return;
 950                 } else {
 951                         LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
 952                         LBUG();
 953                 }
 954         }
 955
 956         /* insert point is last lock on the queue,
 957          * new mode group and new policy group are started */
 958         prev->res_link = queue->prev;
 959         prev->mode_link = &req->l_sl_mode;
 960         prev->policy_link = &req->l_sl_policy;
 961         EXIT;
 962         return;
 963 }
 964
 965 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
 966                                        struct sl_insert_point *prev)
 967 {
 968         struct ldlm_resource *res = lock->l_resource;
 969         ENTRY;
 970
 971         check_res_locked(res);
 972
 973         ldlm_resource_dump(D_INFO, res);
 974         LDLM_DEBUG(lock, "About to add lock:");
 975
 976         if (lock->l_destroyed) {
 977                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 978                 return;
 979         }
 980
 981         LASSERT(cfs_list_empty(&lock->l_res_link));
 982         LASSERT(cfs_list_empty(&lock->l_sl_mode));
 983         LASSERT(cfs_list_empty(&lock->l_sl_policy));
 984
 985         /*
 986          * lock->link == prev->link means lock is first starting the group.
 987          * Don't re-add to itself to suppress kernel warnings.
 988          */
 989         if (&lock->l_res_link != prev->res_link)
 990                 cfs_list_add(&lock->l_res_link, prev->res_link);
 991         if (&lock->l_sl_mode != prev->mode_link)
 992                 cfs_list_add(&lock->l_sl_mode, prev->mode_link);
 993         if (&lock->l_sl_policy != prev->policy_link)
 994                 cfs_list_add(&lock->l_sl_policy, prev->policy_link);
 995
 996         EXIT;
 997 }
 998
 999 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1000 {
1001         struct sl_insert_point prev;
1002         ENTRY;
1003
1004         LASSERT(lock->l_req_mode == lock->l_granted_mode);
1005
1006         search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1007         ldlm_granted_list_add_lock(lock, &prev);
1008         EXIT;
1009 }
1010
1011 /* NOTE: called by
1012  *  - ldlm_lock_enqueue
1013  *  - ldlm_reprocess_queue
1014  *  - ldlm_lock_convert
1015  *
1016  * must be called with lr_lock held
1017  */
1018 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
1019 {
1020         struct ldlm_resource *res = lock->l_resource;
1021         ENTRY;
1022
1023         check_res_locked(res);
1024
1025         lock->l_granted_mode = lock->l_req_mode;
1026         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1027                 ldlm_grant_lock_with_skiplist(lock);
1028         else if (res->lr_type == LDLM_EXTENT)
1029                 ldlm_extent_add_lock(res, lock);
1030         else
1031                 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1032
1033         if (lock->l_granted_mode < res->lr_most_restr)
1034                 res->lr_most_restr = lock->l_granted_mode;
1035
1036         if (work_list && lock->l_completion_ast != NULL)
1037                 ldlm_add_ast_work_item(lock, NULL, work_list);
1038
1039         ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1040         EXIT;
1041 }
1042
1043 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
1044  * comment above ldlm_lock_match */
1045 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1046                                       ldlm_mode_t *mode,
1047                                       ldlm_policy_data_t *policy,
1048                                       struct ldlm_lock *old_lock,
1049                                       __u64 flags, int unref)
1050 {
1051         struct ldlm_lock *lock;
1052         cfs_list_t       *tmp;
1053
1054         cfs_list_for_each(tmp, queue) {
1055                 ldlm_mode_t match;
1056
1057                 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1058
1059                 if (lock == old_lock)
1060                         break;
1061
1062                 /* llite sometimes wants to match locks that will be
1063                  * canceled when their users drop, but we allow it to match
1064                  * if it passes in CBPENDING and the lock still has users.
1065                  * this is generally only going to be used by children
1066                  * whose parents already hold a lock so forward progress
1067                  * can still happen. */
1068                 if (lock->l_flags & LDLM_FL_CBPENDING &&
1069                     !(flags & LDLM_FL_CBPENDING))
1070                         continue;
1071                 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
1072                     lock->l_readers == 0 && lock->l_writers == 0)
1073                         continue;
1074
1075                 if (!(lock->l_req_mode & *mode))
1076                         continue;
1077                 match = lock->l_req_mode;
1078
1079                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1080                     (lock->l_policy_data.l_extent.start >
1081                      policy->l_extent.start ||
1082                      lock->l_policy_data.l_extent.end < policy->l_extent.end))
1083                         continue;
1084
1085                 if (unlikely(match == LCK_GROUP) &&
1086                     lock->l_resource->lr_type == LDLM_EXTENT &&
1087                     lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1088                         continue;
1089
1090                 /* We match if we have existing lock with same or wider set
1091                    of bits. */
1092                 if (lock->l_resource->lr_type == LDLM_IBITS &&
1093                      ((lock->l_policy_data.l_inodebits.bits &
1094                       policy->l_inodebits.bits) !=
1095                       policy->l_inodebits.bits))
1096                         continue;
1097
1098                 if (!unref &&
1099                     (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1100                      lock->l_failed))
1101                         continue;
1102
1103                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1104                     !(lock->l_flags & LDLM_FL_LOCAL))
1105                         continue;
1106
1107                 if (flags & LDLM_FL_TEST_LOCK) {
1108                         LDLM_LOCK_GET(lock);
1109                         ldlm_lock_touch_in_lru(lock);
1110                 } else {
1111                         ldlm_lock_addref_internal_nolock(lock, match);
1112                 }
1113                 *mode = match;
1114                 return lock;
1115         }
1116
1117         return NULL;
1118 }
1119
1120 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1121 {
1122         if (!lock->l_failed) {
1123                 lock->l_failed = 1;
1124                 cfs_waitq_broadcast(&lock->l_waitq);
1125         }
1126 }
1127 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1128
1129 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1130 {
1131         lock_res_and_lock(lock);
1132         ldlm_lock_fail_match_locked(lock);
1133         unlock_res_and_lock(lock);
1134 }
1135 EXPORT_SYMBOL(ldlm_lock_fail_match);
1136
1137 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1138 {
1139         lock->l_flags |= LDLM_FL_LVB_READY;
1140         cfs_waitq_broadcast(&lock->l_waitq);
1141 }
1142 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1143
1144 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1145 {
1146         lock_res_and_lock(lock);
1147         ldlm_lock_allow_match_locked(lock);
1148         unlock_res_and_lock(lock);
1149 }
1150 EXPORT_SYMBOL(ldlm_lock_allow_match);
1151
1152 /* Can be called in two ways:
1153  *
1154  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1155  * for a duplicate of.
1156  *
1157  * Otherwise, all of the fields must be filled in, to match against.
1158  *
1159  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1160  *     server (ie, connh is NULL)
1161  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1162  *     list will be considered
1163  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1164  *     to be canceled can still be matched as long as they still have reader
1165  *     or writer refernces
1166  * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1167  *     just tell us if we would have matched.
1168  *
1169  * Returns 1 if it finds an already-existing lock that is compatible; in this
1170  * case, lockh is filled in with a addref()ed lock
1171  *
1172  * we also check security context, if that failed we simply return 0 (to keep
1173  * caller code unchanged), the context failure will be discovered by caller
1174  * sometime later.
1175  */
1176 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1177                             const struct ldlm_res_id *res_id, ldlm_type_t type,
1178                             ldlm_policy_data_t *policy, ldlm_mode_t mode,
1179                             struct lustre_handle *lockh, int unref)
1180 {
1181         struct ldlm_resource *res;
1182         struct ldlm_lock *lock, *old_lock = NULL;
1183         int rc = 0;
1184         ENTRY;
1185
1186         if (ns == NULL) {
1187                 old_lock = ldlm_handle2lock(lockh);
1188                 LASSERT(old_lock);
1189
1190                 ns = ldlm_lock_to_ns(old_lock);
1191                 res_id = &old_lock->l_resource->lr_name;
1192                 type = old_lock->l_resource->lr_type;
1193                 mode = old_lock->l_req_mode;
1194         }
1195
1196         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1197         if (res == NULL) {
1198                 LASSERT(old_lock == NULL);
1199                 RETURN(0);
1200         }
1201
1202         LDLM_RESOURCE_ADDREF(res);
1203         lock_res(res);
1204
1205         lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1206                             flags, unref);
1207         if (lock != NULL)
1208                 GOTO(out, rc = 1);
1209         if (flags & LDLM_FL_BLOCK_GRANTED)
1210                 GOTO(out, rc = 0);
1211         lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1212                             flags, unref);
1213         if (lock != NULL)
1214                 GOTO(out, rc = 1);
1215         lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1216                             flags, unref);
1217         if (lock != NULL)
1218                 GOTO(out, rc = 1);
1219
1220         EXIT;
1221  out:
1222         unlock_res(res);
1223         LDLM_RESOURCE_DELREF(res);
1224         ldlm_resource_putref(res);
1225
1226         if (lock) {
1227                 ldlm_lock2handle(lock, lockh);
1228                 if ((flags & LDLM_FL_LVB_READY) &&
1229                     (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1230                         struct l_wait_info lwi;
1231                         if (lock->l_completion_ast) {
1232                                 int err = lock->l_completion_ast(lock,
1233                                                           LDLM_FL_WAIT_NOREPROC,
1234                                                                  NULL);
1235                                 if (err) {
1236                                         if (flags & LDLM_FL_TEST_LOCK)
1237                                                 LDLM_LOCK_RELEASE(lock);
1238                                         else
1239                                                 ldlm_lock_decref_internal(lock,
1240                                                                           mode);
1241                                         rc = 0;
1242                                         goto out2;
1243                                 }
1244                         }
1245
1246                         lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1247                                                NULL, LWI_ON_SIGNAL_NOOP, NULL);
1248
1249                         /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1250                         l_wait_event(lock->l_waitq,
1251                                      lock->l_flags & LDLM_FL_LVB_READY ||
1252                                      lock->l_destroyed || lock->l_failed,
1253                                      &lwi);
1254                         if (!(lock->l_flags & LDLM_FL_LVB_READY)) {
1255                                 if (flags & LDLM_FL_TEST_LOCK)
1256                                         LDLM_LOCK_RELEASE(lock);
1257                                 else
1258                                         ldlm_lock_decref_internal(lock, mode);
1259                                 rc = 0;
1260                         }
1261                 }
1262         }
1263  out2:
1264         if (rc) {
1265                 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1266                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1267                                 res_id->name[2] : policy->l_extent.start,
1268                            (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1269                                 res_id->name[3] : policy->l_extent.end);
1270
1271                 /* check user's security context */
1272                 if (lock->l_conn_export &&
1273                     sptlrpc_import_check_ctx(
1274                                 class_exp2cliimp(lock->l_conn_export))) {
1275                         if (!(flags & LDLM_FL_TEST_LOCK))
1276                                 ldlm_lock_decref_internal(lock, mode);
1277                         rc = 0;
1278                 }
1279
1280                 if (flags & LDLM_FL_TEST_LOCK)
1281                         LDLM_LOCK_RELEASE(lock);
1282
1283         } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1284                 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1285                                   LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1286                                   type, mode, res_id->name[0], res_id->name[1],
1287                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1288                                         res_id->name[2] :policy->l_extent.start,
1289                                   (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1290                                         res_id->name[3] : policy->l_extent.end);
1291         }
1292         if (old_lock)
1293                 LDLM_LOCK_PUT(old_lock);
1294
1295         return rc ? mode : 0;
1296 }
1297 EXPORT_SYMBOL(ldlm_lock_match);
1298
1299 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1300                                         __u64 *bits)
1301 {
1302         struct ldlm_lock *lock;
1303         ldlm_mode_t mode = 0;
1304         ENTRY;
1305
1306         lock = ldlm_handle2lock(lockh);
1307         if (lock != NULL) {
1308                 lock_res_and_lock(lock);
1309                 if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1310                     lock->l_failed)
1311                         GOTO(out, mode);
1312
1313                 if (lock->l_flags & LDLM_FL_CBPENDING &&
1314                     lock->l_readers == 0 && lock->l_writers == 0)
1315                         GOTO(out, mode);
1316
1317                 if (bits)
1318                         *bits = lock->l_policy_data.l_inodebits.bits;
1319                 mode = lock->l_granted_mode;
1320                 ldlm_lock_addref_internal_nolock(lock, mode);
1321         }
1322
1323         EXIT;
1324
1325 out:
1326         if (lock != NULL) {
1327                 unlock_res_and_lock(lock);
1328                 LDLM_LOCK_PUT(lock);
1329         }
1330         return mode;
1331 }
1332 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1333
1334 /* Returns a referenced lock */
1335 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1336                                    const struct ldlm_res_id *res_id,
1337                                    ldlm_type_t type,
1338                                    ldlm_mode_t mode,
1339                                    const struct ldlm_callback_suite *cbs,
1340                                    void *data, __u32 lvb_len)
1341 {
1342         struct ldlm_lock *lock;
1343         struct ldlm_resource *res;
1344         ENTRY;
1345
1346         res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1347         if (res == NULL)
1348                 RETURN(NULL);
1349
1350         lock = ldlm_lock_new(res);
1351
1352         if (lock == NULL)
1353                 RETURN(NULL);
1354
1355         lock->l_req_mode = mode;
1356         lock->l_ast_data = data;
1357         lock->l_pid = cfs_curproc_pid();
1358         lock->l_ns_srv = !!ns_is_server(ns);
1359         if (cbs) {
1360                 lock->l_blocking_ast = cbs->lcs_blocking;
1361                 lock->l_completion_ast = cbs->lcs_completion;
1362                 lock->l_glimpse_ast = cbs->lcs_glimpse;
1363                 lock->l_weigh_ast = cbs->lcs_weigh;
1364         }
1365
1366         lock->l_tree_node = NULL;
1367         /* if this is the extent lock, allocate the interval tree node */
1368         if (type == LDLM_EXTENT) {
1369                 if (ldlm_interval_alloc(lock) == NULL)
1370                         GOTO(out, 0);
1371         }
1372
1373         if (lvb_len) {
1374                 lock->l_lvb_len = lvb_len;
1375                 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1376                 if (lock->l_lvb_data == NULL)
1377                         GOTO(out, 0);
1378         }
1379
1380         if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1381                 GOTO(out, 0);
1382
1383         RETURN(lock);
1384
1385 out:
1386         ldlm_lock_destroy(lock);
1387         LDLM_LOCK_RELEASE(lock);
1388         return NULL;
1389 }
1390
1391 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1392                                struct ldlm_lock **lockp,
1393                                void *cookie, __u64 *flags)
1394 {
1395         struct ldlm_lock *lock = *lockp;
1396         struct ldlm_resource *res = lock->l_resource;
1397         int local = ns_is_client(ldlm_res_to_ns(res));
1398 #ifdef HAVE_SERVER_SUPPORT
1399         ldlm_processing_policy policy;
1400 #endif
1401         ldlm_error_t rc = ELDLM_OK;
1402         struct ldlm_interval *node = NULL;
1403         ENTRY;
1404
1405         lock->l_last_activity = cfs_time_current_sec();
1406         /* policies are not executed on the client or during replay */
1407         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1408             && !local && ns->ns_policy) {
1409                 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1410                                    NULL);
1411                 if (rc == ELDLM_LOCK_REPLACED) {
1412                         /* The lock that was returned has already been granted,
1413                          * and placed into lockp.  If it's not the same as the
1414                          * one we passed in, then destroy the old one and our
1415                          * work here is done. */
1416                         if (lock != *lockp) {
1417                                 ldlm_lock_destroy(lock);
1418                                 LDLM_LOCK_RELEASE(lock);
1419                         }
1420                         *flags |= LDLM_FL_LOCK_CHANGED;
1421                         RETURN(0);
1422                 } else if (rc != ELDLM_OK ||
1423                            (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1424                         ldlm_lock_destroy(lock);
1425                         RETURN(rc);
1426                 }
1427         }
1428
1429         /* For a replaying lock, it might be already in granted list. So
1430          * unlinking the lock will cause the interval node to be freed, we
1431          * have to allocate the interval node early otherwise we can't regrant
1432          * this lock in the future. - jay */
1433         if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1434                 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1435
1436         lock_res_and_lock(lock);
1437         if (local && lock->l_req_mode == lock->l_granted_mode) {
1438                 /* The server returned a blocked lock, but it was granted
1439                  * before we got a chance to actually enqueue it.  We don't
1440                  * need to do anything else. */
1441                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1442                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1443                 GOTO(out, ELDLM_OK);
1444         }
1445
1446         ldlm_resource_unlink_lock(lock);
1447         if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1448                 if (node == NULL) {
1449                         ldlm_lock_destroy_nolock(lock);
1450                         GOTO(out, rc = -ENOMEM);
1451                 }
1452
1453                 CFS_INIT_LIST_HEAD(&node->li_group);
1454                 ldlm_interval_attach(node, lock);
1455                 node = NULL;
1456         }
1457
1458         /* Some flags from the enqueue want to make it into the AST, via the
1459          * lock's l_flags. */
1460         lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1461
1462         /* This distinction between local lock trees is very important; a client
1463          * namespace only has information about locks taken by that client, and
1464          * thus doesn't have enough information to decide for itself if it can
1465          * be granted (below).  In this case, we do exactly what the server
1466          * tells us to do, as dictated by the 'flags'.
1467          *
1468          * We do exactly the same thing during recovery, when the server is
1469          * more or less trusting the clients not to lie.
1470          *
1471          * FIXME (bug 268): Detect obvious lies by checking compatibility in
1472          * granted/converting queues. */
1473         if (local) {
1474                 if (*flags & LDLM_FL_BLOCK_CONV)
1475                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1476                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1477                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1478                 else
1479                         ldlm_grant_lock(lock, NULL);
1480                 GOTO(out, ELDLM_OK);
1481 #ifdef HAVE_SERVER_SUPPORT
1482         } else if (*flags & LDLM_FL_REPLAY) {
1483                 if (*flags & LDLM_FL_BLOCK_CONV) {
1484                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1485                         GOTO(out, ELDLM_OK);
1486                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1487                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1488                         GOTO(out, ELDLM_OK);
1489                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1490                         ldlm_grant_lock(lock, NULL);
1491                         GOTO(out, ELDLM_OK);
1492                 }
1493                 /* If no flags, fall through to normal enqueue path. */
1494         }
1495
1496         policy = ldlm_processing_policy_table[res->lr_type];
1497         policy(lock, flags, 1, &rc, NULL);
1498         GOTO(out, rc);
1499 #else
1500         } else {
1501                 CERROR("This is client-side-only module, cannot handle "
1502                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
1503                 LBUG();
1504         }
1505 #endif
1506
1507 out:
1508         unlock_res_and_lock(lock);
1509         if (node)
1510                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1511         return rc;
1512 }
1513
1514 #ifdef HAVE_SERVER_SUPPORT
1515 /* Must be called with namespace taken: queue is waiting or converting. */
1516 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1517                          cfs_list_t *work_list)
1518 {
1519         cfs_list_t *tmp, *pos;
1520         ldlm_processing_policy policy;
1521         __u64 flags;
1522         int rc = LDLM_ITER_CONTINUE;
1523         ldlm_error_t err;
1524         ENTRY;
1525
1526         check_res_locked(res);
1527
1528         policy = ldlm_processing_policy_table[res->lr_type];
1529         LASSERT(policy);
1530
1531         cfs_list_for_each_safe(tmp, pos, queue) {
1532                 struct ldlm_lock *pending;
1533                 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1534
1535                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1536
1537                 flags = 0;
1538                 rc = policy(pending, &flags, 0, &err, work_list);
1539                 if (rc != LDLM_ITER_CONTINUE)
1540                         break;
1541         }
1542
1543         RETURN(rc);
1544 }
1545 #endif
1546
1547 static int
1548 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1549 {
1550         struct ldlm_cb_set_arg *arg = opaq;
1551         struct ldlm_lock_desc   d;
1552         int                     rc;
1553         struct ldlm_lock       *lock;
1554         ENTRY;
1555
1556         if (cfs_list_empty(arg->list))
1557                 RETURN(-ENOENT);
1558
1559         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1560
1561         /* nobody should touch l_bl_ast */
1562         lock_res_and_lock(lock);
1563         cfs_list_del_init(&lock->l_bl_ast);
1564
1565         LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1566         LASSERT(lock->l_bl_ast_run == 0);
1567         LASSERT(lock->l_blocking_lock);
1568         lock->l_bl_ast_run++;
1569         unlock_res_and_lock(lock);
1570
1571         ldlm_lock2desc(lock->l_blocking_lock, &d);
1572
1573         rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1574         LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1575         lock->l_blocking_lock = NULL;
1576         LDLM_LOCK_RELEASE(lock);
1577
1578         RETURN(rc);
1579 }
1580
1581 static int
1582 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1583 {
1584         struct ldlm_cb_set_arg  *arg = opaq;
1585         int                      rc = 0;
1586         struct ldlm_lock        *lock;
1587         ldlm_completion_callback completion_callback;
1588         ENTRY;
1589
1590         if (cfs_list_empty(arg->list))
1591                 RETURN(-ENOENT);
1592
1593         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1594
1595         /* It's possible to receive a completion AST before we've set
1596          * the l_completion_ast pointer: either because the AST arrived
1597          * before the reply, or simply because there's a small race
1598          * window between receiving the reply and finishing the local
1599          * enqueue. (bug 842)
1600          *
1601          * This can't happen with the blocking_ast, however, because we
1602          * will never call the local blocking_ast until we drop our
1603          * reader/writer reference, which we won't do until we get the
1604          * reply and finish enqueueing. */
1605
1606         /* nobody should touch l_cp_ast */
1607         lock_res_and_lock(lock);
1608         cfs_list_del_init(&lock->l_cp_ast);
1609         LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1610         /* save l_completion_ast since it can be changed by
1611          * mds_intent_policy(), see bug 14225 */
1612         completion_callback = lock->l_completion_ast;
1613         lock->l_flags &= ~LDLM_FL_CP_REQD;
1614         unlock_res_and_lock(lock);
1615
1616         if (completion_callback != NULL)
1617                 rc = completion_callback(lock, 0, (void *)arg);
1618         LDLM_LOCK_RELEASE(lock);
1619
1620         RETURN(rc);
1621 }
1622
1623 static int
1624 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1625 {
1626         struct ldlm_cb_set_arg *arg = opaq;
1627         struct ldlm_lock_desc   desc;
1628         int                     rc;
1629         struct ldlm_lock       *lock;
1630         ENTRY;
1631
1632         if (cfs_list_empty(arg->list))
1633                 RETURN(-ENOENT);
1634
1635         lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1636         cfs_list_del_init(&lock->l_rk_ast);
1637
1638         /* the desc just pretend to exclusive */
1639         ldlm_lock2desc(lock, &desc);
1640         desc.l_req_mode = LCK_EX;
1641         desc.l_granted_mode = 0;
1642
1643         rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1644         LDLM_LOCK_RELEASE(lock);
1645
1646         RETURN(rc);
1647 }
1648
1649 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1650 {
1651         struct ldlm_cb_set_arg          *arg = opaq;
1652         struct ldlm_glimpse_work        *gl_work;
1653         struct ldlm_lock                *lock;
1654         int                              rc = 0;
1655         ENTRY;
1656
1657         if (cfs_list_empty(arg->list))
1658                 RETURN(-ENOENT);
1659
1660         gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work,
1661                                  gl_list);
1662         cfs_list_del_init(&gl_work->gl_list);
1663
1664         lock = gl_work->gl_lock;
1665
1666         /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1667         arg->gl_desc = gl_work->gl_desc;
1668
1669         /* invoke the actual glimpse callback */
1670         if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1671                 rc = 1;
1672
1673         LDLM_LOCK_RELEASE(lock);
1674
1675         if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1676                 OBD_FREE_PTR(gl_work);
1677
1678         RETURN(rc);
1679 }
1680
1681 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1682                       ldlm_desc_ast_t ast_type)
1683 {
1684         struct ldlm_cb_set_arg *arg;
1685         set_producer_func       work_ast_lock;
1686         int                     rc;
1687
1688         if (cfs_list_empty(rpc_list))
1689                 RETURN(0);
1690
1691         OBD_ALLOC_PTR(arg);
1692         if (arg == NULL)
1693                 RETURN(-ENOMEM);
1694
1695         cfs_atomic_set(&arg->restart, 0);
1696         arg->list = rpc_list;
1697
1698         switch (ast_type) {
1699                 case LDLM_WORK_BL_AST:
1700                         arg->type = LDLM_BL_CALLBACK;
1701                         work_ast_lock = ldlm_work_bl_ast_lock;
1702                         break;
1703                 case LDLM_WORK_CP_AST:
1704                         arg->type = LDLM_CP_CALLBACK;
1705                         work_ast_lock = ldlm_work_cp_ast_lock;
1706                         break;
1707                 case LDLM_WORK_REVOKE_AST:
1708                         arg->type = LDLM_BL_CALLBACK;
1709                         work_ast_lock = ldlm_work_revoke_ast_lock;
1710                         break;
1711                 case LDLM_WORK_GL_AST:
1712                         arg->type = LDLM_GL_CALLBACK;
1713                         work_ast_lock = ldlm_work_gl_ast_lock;
1714                         break;
1715                 default:
1716                         LBUG();
1717         }
1718
1719         /* We create a ptlrpc request set with flow control extension.
1720          * This request set will use the work_ast_lock function to produce new
1721          * requests and will send a new request each time one completes in order
1722          * to keep the number of requests in flight to ns_max_parallel_ast */
1723         arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
1724                                      work_ast_lock, arg);
1725         if (arg->set == NULL)
1726                 GOTO(out, rc = -ENOMEM);
1727
1728         ptlrpc_set_wait(arg->set);
1729         ptlrpc_set_destroy(arg->set);
1730
1731         rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0;
1732         GOTO(out, rc);
1733 out:
1734         OBD_FREE_PTR(arg);
1735         return rc;
1736 }
1737
1738 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1739 {
1740         ldlm_reprocess_all(res);
1741         return LDLM_ITER_CONTINUE;
1742 }
1743
1744 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1745                               cfs_hlist_node_t *hnode, void *arg)
1746 {
1747         struct ldlm_resource *res = cfs_hash_object(hs, hnode);
1748         int    rc;
1749
1750         rc = reprocess_one_queue(res, arg);
1751
1752         return rc == LDLM_ITER_STOP;
1753 }
1754
1755 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1756 {
1757         ENTRY;
1758
1759         if (ns != NULL) {
1760                 cfs_hash_for_each_nolock(ns->ns_rs_hash,
1761                                          ldlm_reprocess_res, NULL);
1762         }
1763         EXIT;
1764 }
1765 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
1766
1767 void ldlm_reprocess_all(struct ldlm_resource *res)
1768 {
1769         CFS_LIST_HEAD(rpc_list);
1770
1771 #ifdef HAVE_SERVER_SUPPORT
1772         int rc;
1773         ENTRY;
1774         /* Local lock trees don't get reprocessed. */
1775         if (ns_is_client(ldlm_res_to_ns(res))) {
1776                 EXIT;
1777                 return;
1778         }
1779
1780 restart:
1781         lock_res(res);
1782         rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1783         if (rc == LDLM_ITER_CONTINUE)
1784                 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1785         unlock_res(res);
1786
1787         rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
1788                                LDLM_WORK_CP_AST);
1789         if (rc == -ERESTART) {
1790                 LASSERT(cfs_list_empty(&rpc_list));
1791                 goto restart;
1792         }
1793 #else
1794         ENTRY;
1795         if (!ns_is_client(ldlm_res_to_ns(res))) {
1796                 CERROR("This is client-side-only module, cannot handle "
1797                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
1798                 LBUG();
1799         }
1800 #endif
1801         EXIT;
1802 }
1803
1804 void ldlm_cancel_callback(struct ldlm_lock *lock)
1805 {
1806         check_res_locked(lock->l_resource);
1807         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1808                 lock->l_flags |= LDLM_FL_CANCEL;
1809                 if (lock->l_blocking_ast) {
1810                         // l_check_no_ns_lock(ns);
1811                         unlock_res_and_lock(lock);
1812                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1813                                              LDLM_CB_CANCELING);
1814                         lock_res_and_lock(lock);
1815                 } else {
1816                         LDLM_DEBUG(lock, "no blocking ast");
1817                 }
1818         }
1819         lock->l_flags |= LDLM_FL_BL_DONE;
1820 }
1821
1822 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1823 {
1824         if (req->l_resource->lr_type != LDLM_PLAIN &&
1825             req->l_resource->lr_type != LDLM_IBITS)
1826                 return;
1827
1828         cfs_list_del_init(&req->l_sl_policy);
1829         cfs_list_del_init(&req->l_sl_mode);
1830 }
1831
1832 void ldlm_lock_cancel(struct ldlm_lock *lock)
1833 {
1834         struct ldlm_resource *res;
1835         struct ldlm_namespace *ns;
1836         ENTRY;
1837
1838         lock_res_and_lock(lock);
1839
1840         res = lock->l_resource;
1841         ns  = ldlm_res_to_ns(res);
1842
1843         /* Please do not, no matter how tempting, remove this LBUG without
1844          * talking to me first. -phik */
1845         if (lock->l_readers || lock->l_writers) {
1846                 LDLM_ERROR(lock, "lock still has references");
1847                 LBUG();
1848         }
1849
1850         if (lock->l_waited)
1851                 ldlm_del_waiting_lock(lock);
1852
1853         /* Releases cancel callback. */
1854         ldlm_cancel_callback(lock);
1855
1856         /* Yes, second time, just in case it was added again while we were
1857            running with no res lock in ldlm_cancel_callback */
1858         if (lock->l_waited)
1859                 ldlm_del_waiting_lock(lock);
1860
1861         ldlm_resource_unlink_lock(lock);
1862         ldlm_lock_destroy_nolock(lock);
1863
1864         if (lock->l_granted_mode == lock->l_req_mode)
1865                 ldlm_pool_del(&ns->ns_pool, lock);
1866
1867         /* Make sure we will not be called again for same lock what is possible
1868          * if not to zero out lock->l_granted_mode */
1869         lock->l_granted_mode = LCK_MINMODE;
1870         unlock_res_and_lock(lock);
1871
1872         EXIT;
1873 }
1874 EXPORT_SYMBOL(ldlm_lock_cancel);
1875
1876 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1877 {
1878         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1879         int rc = -EINVAL;
1880         ENTRY;
1881
1882         if (lock) {
1883                 if (lock->l_ast_data == NULL)
1884                         lock->l_ast_data = data;
1885                 if (lock->l_ast_data == data)
1886                         rc = 0;
1887                 LDLM_LOCK_PUT(lock);
1888         }
1889         RETURN(rc);
1890 }
1891 EXPORT_SYMBOL(ldlm_lock_set_data);
1892
1893 struct export_cl_data {
1894         struct obd_export       *ecl_exp;
1895         int                     ecl_loop;
1896 };
1897
1898 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1899                                     cfs_hlist_node_t *hnode, void *data)
1900
1901 {
1902         struct export_cl_data   *ecl = (struct export_cl_data *)data;
1903         struct obd_export       *exp  = ecl->ecl_exp;
1904         struct ldlm_lock     *lock = cfs_hash_object(hs, hnode);
1905         struct ldlm_resource *res;
1906
1907         res = ldlm_resource_getref(lock->l_resource);
1908         LDLM_LOCK_GET(lock);
1909
1910         LDLM_DEBUG(lock, "export %p", exp);
1911         ldlm_res_lvbo_update(res, NULL, 1);
1912         ldlm_lock_cancel(lock);
1913         ldlm_reprocess_all(res);
1914         ldlm_resource_putref(res);
1915         LDLM_LOCK_RELEASE(lock);
1916
1917         ecl->ecl_loop++;
1918         if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
1919                 CDEBUG(D_INFO,
1920                        "Cancel lock %p for export %p (loop %d), still have "
1921                        "%d locks left on hash table.\n",
1922                        lock, exp, ecl->ecl_loop,
1923                        cfs_atomic_read(&hs->hs_count));
1924         }
1925
1926         return 0;
1927 }
1928
1929 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1930 {
1931         struct export_cl_data   ecl = {
1932                 .ecl_exp        = exp,
1933                 .ecl_loop       = 0,
1934         };
1935
1936         cfs_hash_for_each_empty(exp->exp_lock_hash,
1937                                 ldlm_cancel_locks_for_export_cb, &ecl);
1938 }
1939
1940 /**
1941  * Downgrade an exclusive lock.
1942  *
1943  * A fast variant of ldlm_lock_convert for convertion of exclusive
1944  * locks. The convertion is always successful.
1945  *
1946  * \param lock A lock to convert
1947  * \param new_mode new lock mode
1948  */
1949 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
1950 {
1951         ENTRY;
1952
1953         LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
1954         LASSERT(new_mode == LCK_COS);
1955
1956         lock_res_and_lock(lock);
1957         ldlm_resource_unlink_lock(lock);
1958         /*
1959          * Remove the lock from pool as it will be added again in
1960          * ldlm_grant_lock() called below.
1961          */
1962         ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
1963
1964         lock->l_req_mode = new_mode;
1965         ldlm_grant_lock(lock, NULL);
1966         unlock_res_and_lock(lock);
1967         ldlm_reprocess_all(lock->l_resource);
1968
1969         EXIT;
1970 }
1971 EXPORT_SYMBOL(ldlm_lock_downgrade);
1972
1973 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1974                                         __u32 *flags)
1975 {
1976         CFS_LIST_HEAD(rpc_list);
1977         struct ldlm_resource *res;
1978         struct ldlm_namespace *ns;
1979         int granted = 0;
1980 #ifdef HAVE_SERVER_SUPPORT
1981         int old_mode;
1982         struct sl_insert_point prev;
1983 #endif
1984         struct ldlm_interval *node;
1985         ENTRY;
1986
1987         if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1988                 *flags |= LDLM_FL_BLOCK_GRANTED;
1989                 RETURN(lock->l_resource);
1990         }
1991
1992         /* I can't check the type of lock here because the bitlock of lock
1993          * is not held here, so do the allocation blindly. -jay */
1994         OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1995         if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
1996                 RETURN(NULL);
1997
1998         LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
1999                  "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2000
2001         lock_res_and_lock(lock);
2002
2003         res = lock->l_resource;
2004         ns  = ldlm_res_to_ns(res);
2005
2006 #ifdef HAVE_SERVER_SUPPORT
2007         old_mode = lock->l_req_mode;
2008 #endif
2009         lock->l_req_mode = new_mode;
2010         if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2011 #ifdef HAVE_SERVER_SUPPORT
2012                 /* remember the lock position where the lock might be
2013                  * added back to the granted list later and also
2014                  * remember the join mode for skiplist fixing. */
2015                 prev.res_link = lock->l_res_link.prev;
2016                 prev.mode_link = lock->l_sl_mode.prev;
2017                 prev.policy_link = lock->l_sl_policy.prev;
2018 #endif
2019                 ldlm_resource_unlink_lock(lock);
2020         } else {
2021                 ldlm_resource_unlink_lock(lock);
2022                 if (res->lr_type == LDLM_EXTENT) {
2023                         /* FIXME: ugly code, I have to attach the lock to a
2024                          * interval node again since perhaps it will be granted
2025                          * soon */
2026                         CFS_INIT_LIST_HEAD(&node->li_group);
2027                         ldlm_interval_attach(node, lock);
2028                         node = NULL;
2029                 }
2030         }
2031
2032         /*
2033          * Remove old lock from the pool before adding the lock with new
2034          * mode below in ->policy()
2035          */
2036         ldlm_pool_del(&ns->ns_pool, lock);
2037
2038         /* If this is a local resource, put it on the appropriate list. */
2039         if (ns_is_client(ldlm_res_to_ns(res))) {
2040                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2041                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
2042                 } else {
2043                         /* This should never happen, because of the way the
2044                          * server handles conversions. */
2045                         LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2046                                    *flags);
2047                         LBUG();
2048
2049                         ldlm_grant_lock(lock, &rpc_list);
2050                         granted = 1;
2051                         /* FIXME: completion handling not with lr_lock held ! */
2052                         if (lock->l_completion_ast)
2053                                 lock->l_completion_ast(lock, 0, NULL);
2054                 }
2055 #ifdef HAVE_SERVER_SUPPORT
2056         } else {
2057                 int rc;
2058                 ldlm_error_t err;
2059                 __u64 pflags = 0;
2060                 ldlm_processing_policy policy;
2061                 policy = ldlm_processing_policy_table[res->lr_type];
2062                 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2063                 if (rc == LDLM_ITER_STOP) {
2064                         lock->l_req_mode = old_mode;
2065                         if (res->lr_type == LDLM_EXTENT)
2066                                 ldlm_extent_add_lock(res, lock);
2067                         else
2068                                 ldlm_granted_list_add_lock(lock, &prev);
2069
2070                         res = NULL;
2071                 } else {
2072                         *flags |= LDLM_FL_BLOCK_GRANTED;
2073                         granted = 1;
2074                 }
2075         }
2076 #else
2077         } else {
2078                 CERROR("This is client-side-only module, cannot handle "
2079                        "LDLM_NAMESPACE_SERVER resource type lock.\n");
2080                 LBUG();
2081         }
2082 #endif
2083         unlock_res_and_lock(lock);
2084
2085         if (granted)
2086                 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2087         if (node)
2088                 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2089         RETURN(res);
2090 }
2091 EXPORT_SYMBOL(ldlm_lock_convert);
2092
2093 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2094 {
2095         struct ldlm_lock *lock;
2096
2097         if (!((libcfs_debug | D_ERROR) & level))
2098                 return;
2099
2100         lock = ldlm_handle2lock(lockh);
2101         if (lock == NULL)
2102                 return;
2103
2104         LDLM_DEBUG_LIMIT(level, lock, "###");
2105
2106         LDLM_LOCK_PUT(lock);
2107 }
2108 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2109
2110 void _ldlm_lock_debug(struct ldlm_lock *lock,
2111                       struct libcfs_debug_msg_data *msgdata,
2112                       const char *fmt, ...)
2113 {
2114         va_list args;
2115         struct obd_export *exp = lock->l_export;
2116         struct ldlm_resource *resource = lock->l_resource;
2117         char *nid = "local";
2118
2119         va_start(args, fmt);
2120
2121         if (exp && exp->exp_connection) {
2122                 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2123         } else if (exp && exp->exp_obd != NULL) {
2124                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2125                 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2126         }
2127
2128         if (resource == NULL) {
2129                 libcfs_debug_vmsg2(msgdata, fmt, args,
2130                        " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2131                        "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2132                        "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2133                        lock,
2134                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2135                        lock->l_readers, lock->l_writers,
2136                        ldlm_lockname[lock->l_granted_mode],
2137                        ldlm_lockname[lock->l_req_mode],
2138                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2139                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2140                        lock->l_pid, lock->l_callback_timeout);
2141                 va_end(args);
2142                 return;
2143         }
2144
2145         switch (resource->lr_type) {
2146         case LDLM_EXTENT:
2147                 libcfs_debug_vmsg2(msgdata, fmt, args,
2148                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2149                        "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
2150                        "] (req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote:"
2151                        " "LPX64" expref: %d pid: %u timeout %lu\n",
2152                        ldlm_lock_to_ns_name(lock), lock,
2153                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2154                        lock->l_readers, lock->l_writers,
2155                        ldlm_lockname[lock->l_granted_mode],
2156                        ldlm_lockname[lock->l_req_mode],
2157                        resource->lr_name.name[0],
2158                        resource->lr_name.name[1],
2159                        cfs_atomic_read(&resource->lr_refcount),
2160                        ldlm_typename[resource->lr_type],
2161                        lock->l_policy_data.l_extent.start,
2162                        lock->l_policy_data.l_extent.end,
2163                        lock->l_req_extent.start, lock->l_req_extent.end,
2164                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2165                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2166                        lock->l_pid, lock->l_callback_timeout);
2167                 break;
2168
2169         case LDLM_FLOCK:
2170                 libcfs_debug_vmsg2(msgdata, fmt, args,
2171                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2172                        "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
2173                        "["LPU64"->"LPU64"] flags: "LPX64" nid: %s remote: "LPX64
2174                        " expref: %d pid: %u timeout: %lu\n",
2175                        ldlm_lock_to_ns_name(lock), lock,
2176                        lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2177                        lock->l_readers, lock->l_writers,
2178                        ldlm_lockname[lock->l_granted_mode],
2179                        ldlm_lockname[lock->l_req_mode],
2180                        resource->lr_name.name[0],
2181                        resource->lr_name.name[1],
2182                        cfs_atomic_read(&resource->lr_refcount),
2183                        ldlm_typename[resource->lr_type],
2184                        lock->l_policy_data.l_flock.pid,
2185                        lock->l_policy_data.l_flock.start,
2186                        lock->l_policy_data.l_flock.end,
2187                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2188                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2189                        lock->l_pid, lock->l_callback_timeout);
2190                 break;
2191
2192         case LDLM_IBITS:
2193                 libcfs_debug_vmsg2(msgdata, fmt, args,
2194                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2195                        "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
2196                        "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2197                        "pid: %u timeout: %lu\n",
2198                        ldlm_lock_to_ns_name(lock),
2199                        lock, lock->l_handle.h_cookie,
2200                        cfs_atomic_read (&lock->l_refc),
2201                        lock->l_readers, lock->l_writers,
2202                        ldlm_lockname[lock->l_granted_mode],
2203                        ldlm_lockname[lock->l_req_mode],
2204                        resource->lr_name.name[0],
2205                        resource->lr_name.name[1],
2206                        lock->l_policy_data.l_inodebits.bits,
2207                        cfs_atomic_read(&resource->lr_refcount),
2208                        ldlm_typename[resource->lr_type],
2209                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2210                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2211                        lock->l_pid, lock->l_callback_timeout);
2212                 break;
2213
2214         default:
2215                 libcfs_debug_vmsg2(msgdata, fmt, args,
2216                        " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2217                        "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
2218                        "nid: %s remote: "LPX64" expref: %d pid: %u timeout %lu"
2219                        "\n",
2220                        ldlm_lock_to_ns_name(lock),
2221                        lock, lock->l_handle.h_cookie,
2222                        cfs_atomic_read (&lock->l_refc),
2223                        lock->l_readers, lock->l_writers,
2224                        ldlm_lockname[lock->l_granted_mode],
2225                        ldlm_lockname[lock->l_req_mode],
2226                        resource->lr_name.name[0],
2227                        resource->lr_name.name[1],
2228                        cfs_atomic_read(&resource->lr_refcount),
2229                        ldlm_typename[resource->lr_type],
2230                        lock->l_flags, nid, lock->l_remote_handle.cookie,
2231                        exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2232                        lock->l_pid, lock->l_callback_timeout);
2233                 break;
2234         }
2235         va_end(args);
2236 }
2237 EXPORT_SYMBOL(_ldlm_lock_debug);