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 Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LDLM
  25
  26 #ifdef __KERNEL__
  27 # include <linux/slab.h>
  28 # include <linux/module.h>
  29 # include <linux/lustre_dlm.h>
  30 #else
  31 # include <liblustre.h>
  32 # include <linux/kp30.h>
  33 #endif
  34
  35 #include <linux/obd_class.h>
  36 #include "ldlm_internal.h"
  37
  38 //struct lustre_lock ldlm_everything_lock;
  39
  40 /* lock types */
  41 char *ldlm_lockname[] = {
  42         [0] "--",
  43         [LCK_EX] "EX",
  44         [LCK_PW] "PW",
  45         [LCK_PR] "PR",
  46         [LCK_CW] "CW",
  47         [LCK_CR] "CR",
  48         [LCK_NL] "NL"
  49 };
  50 char *ldlm_typename[] = {
  51         [LDLM_PLAIN] "PLN",
  52         [LDLM_EXTENT] "EXT",
  53         [LDLM_FLOCK] "FLK",
  54 };
  55
  56 char *ldlm_it2str(int it)
  57 {
  58         switch (it) {
  59         case IT_OPEN:
  60                 return "open";
  61         case IT_CREAT:
  62                 return "creat";
  63         case (IT_OPEN | IT_CREAT):
  64                 return "open|creat";
  65         case IT_READDIR:
  66                 return "readdir";
  67         case IT_GETATTR:
  68                 return "getattr";
  69         case IT_LOOKUP:
  70                 return "lookup";
  71         case IT_UNLINK:
  72                 return "unlink";
  73         case IT_GETXATTR:
  74                 return "getxattr";
  75         default:
  76                 CERROR("Unknown intent %d\n", it);
  77                 return "UNKNOWN";
  78         }
  79 }
  80
  81 extern kmem_cache_t *ldlm_lock_slab;
  82 struct lustre_lock ldlm_handle_lock;
  83
  84 static ldlm_processing_policy ldlm_processing_policy_table[] = {
  85         [LDLM_PLAIN] ldlm_process_plain_lock,
  86         [LDLM_EXTENT] ldlm_process_extent_lock,
  87 #ifdef __KERNEL__
  88         [LDLM_FLOCK] ldlm_process_flock_lock,
  89 #endif
  90 };
  91
  92 static ldlm_res_policy ldlm_intent_policy_func;
  93
  94 void ldlm_register_intent(ldlm_res_policy arg)
  95 {
  96         ldlm_intent_policy_func = arg;
  97 }
  98
  99 void ldlm_unregister_intent(void)
 100 {
 101         ldlm_intent_policy_func = NULL;
 102 }
 103
 104 /*
 105  * REFCOUNTED LOCK OBJECTS
 106  */
 107
 108
 109 /*
 110  * Lock refcounts, during creation:
 111  *   - one special one for allocation, dec'd only once in destroy
 112  *   - one for being a lock that's in-use
 113  *   - one for the addref associated with a new lock
 114  */
 115 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
 116 {
 117         atomic_inc(&lock->l_refc);
 118         return lock;
 119 }
 120
 121 void ldlm_lock_put(struct ldlm_lock *lock)
 122 {
 123         ENTRY;
 124
 125         if (atomic_dec_and_test(&lock->l_refc)) {
 126                 struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 127
 128                 l_lock(&ns->ns_lock);
 129                 LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing");
 130                 LASSERT(lock->l_destroyed);
 131                 LASSERT(list_empty(&lock->l_res_link));
 132
 133                 spin_lock(&ns->ns_counter_lock);
 134                 ns->ns_locks--;
 135                 spin_unlock(&ns->ns_counter_lock);
 136
 137                 ldlm_resource_putref(lock->l_resource);
 138                 lock->l_resource = NULL;
 139                 if (lock->l_export)
 140                         class_export_put(lock->l_export);
 141
 142                 if (lock->l_parent)
 143                         LDLM_LOCK_PUT(lock->l_parent);
 144
 145                 OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
 146                 l_unlock(&ns->ns_lock);
 147         }
 148
 149         EXIT;
 150 }
 151
 152 void ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
 153 {
 154         ENTRY;
 155         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 156         if (!list_empty(&lock->l_lru)) {
 157                 list_del_init(&lock->l_lru);
 158                 lock->l_resource->lr_namespace->ns_nr_unused--;
 159                 LASSERT(lock->l_resource->lr_namespace->ns_nr_unused >= 0);
 160         }
 161         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 162         EXIT;
 163 }
 164
 165 /* This used to have a 'strict' flact, which recovery would use to mark an
 166  * in-use lock as needing-to-die.  Lest I am ever tempted to put it back, I
 167  * shall explain why it's gone: with the new hash table scheme, once you call
 168  * ldlm_lock_destroy, you can never drop your final references on this lock.
 169  * Because it's not in the hash table anymore.  -phil */
 170 void ldlm_lock_destroy(struct ldlm_lock *lock)
 171 {
 172         ENTRY;
 173         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 174
 175         if (!list_empty(&lock->l_children)) {
 176                 LDLM_ERROR(lock, "still has children (%p)!",
 177                            lock->l_children.next);
 178                 ldlm_lock_dump(D_ERROR, lock, 0);
 179                 LBUG();
 180         }
 181         if (lock->l_readers || lock->l_writers) {
 182                 LDLM_ERROR(lock, "lock still has references");
 183                 ldlm_lock_dump(D_ERROR, lock, 0);
 184                 LBUG();
 185         }
 186
 187         if (!list_empty(&lock->l_res_link)) {
 188                 ldlm_lock_dump(D_ERROR, lock, 0);
 189                 LBUG();
 190         }
 191
 192         if (lock->l_destroyed) {
 193                 LASSERT(list_empty(&lock->l_lru));
 194                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 195                 EXIT;
 196                 return;
 197         }
 198         lock->l_destroyed = 1;
 199
 200         list_del_init(&lock->l_export_chain);
 201         ldlm_lock_remove_from_lru(lock);
 202         class_handle_unhash(&lock->l_handle);
 203
 204 #if 0
 205         /* Wake anyone waiting for this lock */
 206         /* FIXME: I should probably add yet another flag, instead of using
 207          * l_export to only call this on clients */
 208         if (lock->l_export)
 209                 class_export_put(lock->l_export);
 210         lock->l_export = NULL;
 211         if (lock->l_export && lock->l_completion_ast)
 212                 lock->l_completion_ast(lock, 0);
 213 #endif
 214
 215         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 216         LDLM_LOCK_PUT(lock);
 217         EXIT;
 218 }
 219
 220 /* this is called by portals_handle2object with the handle lock taken */
 221 static void lock_handle_addref(void *lock)
 222 {
 223         LDLM_LOCK_GET((struct ldlm_lock *)lock);
 224 }
 225
 226 /*
 227  * usage: pass in a resource on which you have done ldlm_resource_get
 228  *        pass in a parent lock on which you have done a ldlm_lock_get
 229  *        after return, ldlm_*_put the resource and parent
 230  * returns: lock with refcount 1
 231  */
 232 static struct ldlm_lock *ldlm_lock_new(struct ldlm_lock *parent,
 233                                        struct ldlm_resource *resource)
 234 {
 235         struct ldlm_lock *lock;
 236         ENTRY;
 237
 238         if (resource == NULL)
 239                 LBUG();
 240
 241         OBD_SLAB_ALLOC(lock, ldlm_lock_slab, SLAB_NOFS, sizeof(*lock));
 242         if (lock == NULL)
 243                 RETURN(NULL);
 244
 245         lock->l_resource = ldlm_resource_getref(resource);
 246
 247         atomic_set(&lock->l_refc, 2);
 248         INIT_LIST_HEAD(&lock->l_children);
 249         INIT_LIST_HEAD(&lock->l_res_link);
 250         INIT_LIST_HEAD(&lock->l_lru);
 251         INIT_LIST_HEAD(&lock->l_export_chain);
 252         INIT_LIST_HEAD(&lock->l_pending_chain);
 253         init_waitqueue_head(&lock->l_waitq);
 254
 255         spin_lock(&resource->lr_namespace->ns_counter_lock);
 256         resource->lr_namespace->ns_locks++;
 257         spin_unlock(&resource->lr_namespace->ns_counter_lock);
 258
 259         if (parent != NULL) {
 260                 l_lock(&parent->l_resource->lr_namespace->ns_lock);
 261                 lock->l_parent = LDLM_LOCK_GET(parent);
 262                 list_add(&lock->l_childof, &parent->l_children);
 263                 l_unlock(&parent->l_resource->lr_namespace->ns_lock);
 264         }
 265
 266         INIT_LIST_HEAD(&lock->l_handle.h_link);
 267         class_handle_hash(&lock->l_handle, lock_handle_addref);
 268
 269         RETURN(lock);
 270 }
 271
 272 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
 273                               struct ldlm_res_id new_resid)
 274 {
 275         struct ldlm_resource *oldres = lock->l_resource;
 276         ENTRY;
 277
 278         l_lock(&ns->ns_lock);
 279         if (memcmp(&new_resid, &lock->l_resource->lr_name,
 280                    sizeof(lock->l_resource->lr_name)) == 0) {
 281                 /* Nothing to do */
 282                 l_unlock(&ns->ns_lock);
 283                 RETURN(0);
 284         }
 285
 286         LASSERT(new_resid.name[0] != 0);
 287
 288         /* This function assumes that the lock isn't on any lists */
 289         LASSERT(list_empty(&lock->l_res_link));
 290
 291         lock->l_resource = ldlm_resource_get(ns, NULL, new_resid,
 292                                              lock->l_resource->lr_type, 1);
 293         if (lock->l_resource == NULL) {
 294                 LBUG();
 295                 RETURN(-ENOMEM);
 296         }
 297
 298         /* ...and the flowers are still standing! */
 299         ldlm_resource_putref(oldres);
 300
 301         l_unlock(&ns->ns_lock);
 302         RETURN(0);
 303 }
 304
 305 /*
 306  *  HANDLES
 307  */
 308
 309 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 310 {
 311         lockh->cookie = lock->l_handle.h_cookie;
 312 }
 313
 314 /* if flags: atomically get the lock and set the flags.
 315  *           Return NULL if flag already set
 316  */
 317
 318 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *handle, int flags)
 319 {
 320         struct ldlm_namespace *ns;
 321         struct ldlm_lock *lock = NULL, *retval = NULL;
 322         ENTRY;
 323
 324         LASSERT(handle);
 325
 326         lock = class_handle2object(handle->cookie);
 327         if (lock == NULL)
 328                 RETURN(NULL);
 329
 330         LASSERT(lock->l_resource != NULL);
 331         ns = lock->l_resource->lr_namespace;
 332         LASSERT(ns != NULL);
 333
 334         l_lock(&ns->ns_lock);
 335
 336         /* It's unlikely but possible that someone marked the lock as
 337          * destroyed after we did handle2object on it */
 338         if (lock->l_destroyed) {
 339                 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
 340                 LDLM_LOCK_PUT(lock);
 341                 GOTO(out, retval);
 342         }
 343
 344         if (flags && (lock->l_flags & flags)) {
 345                 LDLM_LOCK_PUT(lock);
 346                 GOTO(out, retval);
 347         }
 348
 349         if (flags)
 350                 lock->l_flags |= flags;
 351
 352         retval = lock;
 353         EXIT;
 354  out:
 355         l_unlock(&ns->ns_lock);
 356         return retval;
 357 }
 358
 359 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
 360                                       struct lustre_handle *handle)
 361 {
 362         struct ldlm_lock *retval = NULL;
 363
 364         l_lock(&ns->ns_lock);
 365         retval = __ldlm_handle2lock(handle, 0);
 366         l_unlock(&ns->ns_lock);
 367
 368         return retval;
 369 }
 370
 371 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 372 {
 373         ldlm_res2desc(lock->l_resource, &desc->l_resource);
 374         desc->l_req_mode = lock->l_req_mode;
 375         desc->l_granted_mode = lock->l_granted_mode;
 376         memcpy(&desc->l_policy_data, &lock->l_policy_data,
 377                sizeof(desc->l_policy_data));
 378         memcpy(desc->l_version, lock->l_version, sizeof(desc->l_version));
 379 }
 380
 381 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
 382                             void *data, int datalen)
 383 {
 384         struct ldlm_ast_work *w;
 385         ENTRY;
 386
 387         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 388         if (new && (lock->l_flags & LDLM_FL_AST_SENT))
 389                 GOTO(out, 0);
 390
 391         CDEBUG(D_OTHER, "lock %p incompatible; sending blocking AST.\n", lock);
 392
 393         OBD_ALLOC(w, sizeof(*w));
 394         if (!w) {
 395                 LBUG();
 396                 GOTO(out, 0);
 397         }
 398
 399         w->w_data = data;
 400         w->w_datalen = datalen;
 401         if (new) {
 402                 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
 403                 lock->l_flags |= LDLM_FL_AST_SENT;
 404                 /* If the enqueuing client said so, tell the AST recipient to
 405                  * discard dirty data, rather than writing back. */
 406                 if (new->l_flags & LDLM_AST_DISCARD_DATA)
 407                         lock->l_flags |= LDLM_FL_DISCARD_DATA;
 408                 w->w_blocking = 1;
 409                 ldlm_lock2desc(new, &w->w_desc);
 410         }
 411
 412         w->w_lock = LDLM_LOCK_GET(lock);
 413         list_add(&w->w_list, lock->l_resource->lr_tmp);
 414         EXIT;
 415  out:
 416         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 417 }
 418
 419 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 420 {
 421         struct ldlm_lock *lock;
 422
 423         lock = ldlm_handle2lock(lockh);
 424         ldlm_lock_addref_internal(lock, mode);
 425         LDLM_LOCK_PUT(lock);
 426 }
 427
 428 /* only called for local locks */
 429 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 430 {
 431         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 432         ldlm_lock_remove_from_lru(lock);
 433         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 434                 lock->l_readers++;
 435         else
 436                 lock->l_writers++;
 437         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 438         LDLM_LOCK_GET(lock);
 439         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 440 }
 441
 442 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
 443 {
 444         struct ldlm_namespace *ns;
 445         ENTRY;
 446
 447         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 448         ns = lock->l_resource->lr_namespace;
 449         l_lock(&ns->ns_lock);
 450         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR) {
 451                 LASSERT(lock->l_readers > 0);
 452                 lock->l_readers--;
 453         } else {
 454                 LASSERT(lock->l_writers > 0);
 455                 lock->l_writers--;
 456         }
 457
 458         if (lock->l_flags & LDLM_FL_LOCAL &&
 459             !lock->l_readers && !lock->l_writers) {
 460                 /* If this is a local lock on a server namespace and this was
 461                  * the last reference, cancel the lock. */
 462                 CDEBUG(D_INFO, "forcing cancel of local lock\n");
 463                 lock->l_flags |= LDLM_FL_CBPENDING;
 464         }
 465
 466         if (!lock->l_readers && !lock->l_writers &&
 467             (lock->l_flags & LDLM_FL_CBPENDING)) {
 468                 /* If we received a blocked AST and this was the last reference,
 469                  * run the callback. */
 470                 if (!ns->ns_client && lock->l_export)
 471                         CERROR("FL_CBPENDING set on non-local lock--just a "
 472                                "warning\n");
 473
 474                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 475                 l_unlock(&ns->ns_lock);
 476
 477                 l_check_no_ns_lock(ns);
 478                 /* FIXME: need a real 'desc' here */
 479                 if (lock->l_blocking_ast != NULL)
 480                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
 481                                              LDLM_CB_BLOCKING);
 482         } else if (ns->ns_client && !lock->l_readers && !lock->l_writers) {
 483                 /* If this is a client-side namespace and this was the last
 484                  * reference, put it on the LRU. */
 485                 LASSERT(list_empty(&lock->l_lru));
 486                 LASSERT(ns->ns_nr_unused >= 0);
 487                 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
 488                 ns->ns_nr_unused++;
 489                 l_unlock(&ns->ns_lock);
 490                 ldlm_cancel_lru(ns);
 491         } else {
 492                 l_unlock(&ns->ns_lock);
 493         }
 494
 495         LDLM_LOCK_PUT(lock);    /* matches the ldlm_lock_get in addref */
 496
 497         EXIT;
 498 }
 499
 500 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 501 {
 502         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 503         LASSERT(lock != NULL);
 504         ldlm_lock_decref_internal(lock, mode);
 505         LDLM_LOCK_PUT(lock);
 506 }
 507
 508 /* This will drop a lock reference and mark it for destruction, but will not
 509  * necessarily cancel the lock before returning. */
 510 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
 511 {
 512         struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
 513         ENTRY;
 514
 515         LASSERT(lock != NULL);
 516
 517         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 518         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 519         lock->l_flags |= LDLM_FL_CBPENDING;
 520         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 521         ldlm_lock_decref_internal(lock, mode);
 522         LDLM_LOCK_PUT(lock);
 523 }
 524
 525 /* NOTE: called by
 526  *  - ldlm_lock_enqueue
 527  *  - ldlm_reprocess_queue
 528  *  - ldlm_lock_convert
 529  */
 530 void ldlm_grant_lock(struct ldlm_lock *lock, void *data, int datalen,
 531                      int run_ast)
 532 {
 533         struct ldlm_resource *res = lock->l_resource;
 534         ENTRY;
 535
 536         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 537         lock->l_granted_mode = lock->l_req_mode;
 538         ldlm_resource_add_lock(res, &res->lr_granted, lock);
 539
 540         if (lock->l_granted_mode < res->lr_most_restr)
 541                 res->lr_most_restr = lock->l_granted_mode;
 542
 543         if (run_ast && lock->l_completion_ast != NULL)
 544                 ldlm_add_ast_work_item(lock, NULL, data, datalen);
 545
 546         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 547         EXIT;
 548 }
 549
 550 /* returns a referenced lock or NULL.  See the flag descriptions below, in the
 551  * comment above ldlm_lock_match */
 552 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
 553                                       struct ldlm_extent *extent,
 554                                       struct ldlm_lock *old_lock, int flags)
 555 {
 556         struct ldlm_lock *lock;
 557         struct list_head *tmp;
 558
 559         list_for_each(tmp, queue) {
 560                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 561
 562                 if (lock == old_lock)
 563                         break;
 564
 565                 /* llite sometimes wants to match locks that will be
 566                  * canceled when their users drop, but we allow it to match
 567                  * if it passes in CBPENDING and the lock still has users.
 568                  * this is generally only going to be used by children
 569                  * whose parents already hold a lock so forward progress
 570                  * can still happen. */
 571                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 572                     !(flags & LDLM_FL_CBPENDING))
 573                         continue;
 574                 if (lock->l_flags & LDLM_FL_CBPENDING &&
 575                     lock->l_readers == 0 && lock->l_writers == 0)
 576                         continue;
 577
 578                 if (lock->l_req_mode != mode)
 579                         continue;
 580
 581                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 582                     (lock->l_policy_data.l_extent.start > extent->start ||
 583                      lock->l_policy_data.l_extent.end < extent->end))
 584                         continue;
 585
 586                 if (lock->l_destroyed)
 587                         continue;
 588
 589                 if ((flags & LDLM_FL_LOCAL_ONLY) &&
 590                     !(lock->l_flags & LDLM_FL_LOCAL))
 591                         continue;
 592
 593                 ldlm_lock_addref_internal(lock, mode);
 594                 return lock;
 595         }
 596
 597         return NULL;
 598 }
 599
 600 /* Can be called in two ways:
 601  *
 602  * If 'ns' is NULL, then lockh describes an existing lock that we want to look
 603  * for a duplicate of.
 604  *
 605  * Otherwise, all of the fields must be filled in, to match against.
 606  *
 607  * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
 608  *     server (ie, connh is NULL)
 609  * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
 610  *     list will be considered
 611  * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
 612  *     to be canceled can still be matched as long as they still have reader
 613  *     or writer refernces
 614  *
 615  * Returns 1 if it finds an already-existing lock that is compatible; in this
 616  * case, lockh is filled in with a addref()ed lock
 617  */
 618 int ldlm_lock_match(struct ldlm_namespace *ns, int flags,
 619                     struct ldlm_res_id *res_id, __u32 type, void *cookie,
 620                     int cookielen, ldlm_mode_t mode,
 621                     struct lustre_handle *lockh)
 622 {
 623         struct ldlm_resource *res;
 624         struct ldlm_lock *lock, *old_lock = NULL;
 625         int rc = 0;
 626         ENTRY;
 627
 628         if (ns == NULL) {
 629                 old_lock = ldlm_handle2lock(lockh);
 630                 LASSERT(old_lock);
 631
 632                 ns = old_lock->l_resource->lr_namespace;
 633                 res_id = &old_lock->l_resource->lr_name;
 634                 type = old_lock->l_resource->lr_type;
 635                 mode = old_lock->l_req_mode;
 636         }
 637
 638         res = ldlm_resource_get(ns, NULL, *res_id, type, 0);
 639         if (res == NULL) {
 640                 LASSERT(old_lock == NULL);
 641                 RETURN(0);
 642         }
 643
 644         l_lock(&ns->ns_lock);
 645
 646         lock = search_queue(&res->lr_granted, mode, cookie, old_lock, flags);
 647         if (lock != NULL)
 648                 GOTO(out, rc = 1);
 649         if (flags & LDLM_FL_BLOCK_GRANTED)
 650                 GOTO(out, rc = 0);
 651         lock = search_queue(&res->lr_converting, mode, cookie, old_lock, flags);
 652         if (lock != NULL)
 653                 GOTO(out, rc = 1);
 654         lock = search_queue(&res->lr_waiting, mode, cookie, old_lock, flags);
 655         if (lock != NULL)
 656                 GOTO(out, rc = 1);
 657
 658         EXIT;
 659  out:
 660         ldlm_resource_putref(res);
 661         l_unlock(&ns->ns_lock);
 662
 663         if (lock) {
 664                 ldlm_lock2handle(lock, lockh);
 665                 if (lock->l_completion_ast)
 666                         lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC,
 667                                                NULL);
 668         }
 669         if (rc)
 670                 LDLM_DEBUG(lock, "matched");
 671         else
 672                 LDLM_DEBUG_NOLOCK("not matched");
 673
 674         if (old_lock)
 675                 LDLM_LOCK_PUT(old_lock);
 676
 677         return rc;
 678 }
 679
 680 /* Returns a referenced lock */
 681 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
 682                                    struct lustre_handle *parent_lock_handle,
 683                                    struct ldlm_res_id res_id, __u32 type,
 684                                    ldlm_mode_t mode,
 685                                    ldlm_blocking_callback blocking,
 686                                    ldlm_completion_callback completion,
 687                                    void *data)
 688 {
 689         struct ldlm_resource *res, *parent_res = NULL;
 690         struct ldlm_lock *lock, *parent_lock = NULL;
 691         ENTRY;
 692
 693         if (parent_lock_handle) {
 694                 parent_lock = ldlm_handle2lock(parent_lock_handle);
 695                 if (parent_lock)
 696                         parent_res = parent_lock->l_resource;
 697         }
 698
 699         res = ldlm_resource_get(ns, parent_res, res_id, type, 1);
 700         if (res == NULL)
 701                 RETURN(NULL);
 702
 703         lock = ldlm_lock_new(parent_lock, res);
 704         ldlm_resource_putref(res);
 705         if (parent_lock != NULL)
 706                 LDLM_LOCK_PUT(parent_lock);
 707
 708         if (lock == NULL)
 709                 RETURN(NULL);
 710
 711         lock->l_req_mode = mode;
 712         lock->l_ast_data = data;
 713         lock->l_blocking_ast = blocking;
 714         lock->l_completion_ast = completion;
 715
 716         RETURN(lock);
 717 }
 718
 719 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
 720                                struct ldlm_lock **lockp,
 721                                void *cookie, int cookie_len, int *flags)
 722 {
 723         struct ldlm_lock *lock = *lockp;
 724         struct ldlm_resource *res = lock->l_resource;
 725         int local = res->lr_namespace->ns_client;
 726         ldlm_processing_policy policy;
 727         ldlm_error_t rc = ELDLM_OK;
 728         ENTRY;
 729
 730         if (res->lr_type != LDLM_PLAIN)
 731                 memcpy(&lock->l_policy_data, cookie, cookie_len);
 732
 733         /* policies are not executed on the client or during replay */
 734         if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
 735             && !local && ldlm_intent_policy_func) {
 736                 rc = ldlm_intent_policy_func(ns, lockp, cookie,
 737                                              lock->l_req_mode, *flags, NULL);
 738                 if (rc == ELDLM_LOCK_REPLACED) {
 739                         /* The lock that was returned has already been granted,
 740                          * and placed into lockp.  Destroy the old one and our
 741                          * work here is done. */
 742                         ldlm_lock_destroy(lock);
 743                         LDLM_LOCK_PUT(lock);
 744                         *flags |= LDLM_FL_LOCK_CHANGED;
 745                         RETURN(0);
 746                 } else if (rc == ELDLM_LOCK_ABORTED ||
 747                            (rc == 0 && (*flags & LDLM_FL_INTENT_ONLY))) {
 748                         ldlm_lock_destroy(lock);
 749                         RETURN(rc);
 750                 }
 751                 LASSERT(rc == ELDLM_OK);
 752         }
 753
 754         l_lock(&ns->ns_lock);
 755         if (local && lock->l_req_mode == lock->l_granted_mode) {
 756                 /* The server returned a blocked lock, but it was granted before
 757                  * we got a chance to actually enqueue it.  We don't need to do
 758                  * anything else. */
 759                 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
 760                             LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
 761                 GOTO(out, ELDLM_OK);
 762         }
 763
 764         /* Some flags from the enqueue want to make it into the AST, via the
 765          * lock's l_flags. */
 766         lock->l_flags |= (*flags & LDLM_AST_DISCARD_DATA);
 767
 768         /* This distinction between local lock trees is very important; a client
 769          * namespace only has information about locks taken by that client, and
 770          * thus doesn't have enough information to decide for itself if it can
 771          * be granted (below).  In this case, we do exactly what the server
 772          * tells us to do, as dictated by the 'flags'.
 773          *
 774          * We do exactly the same thing during recovery, when the server is
 775          * more or less trusting the clients not to lie.
 776          *
 777          * FIXME (bug 268): Detect obvious lies by checking compatibility in
 778          * granted/converting queues. */
 779         ldlm_resource_unlink_lock(lock);
 780         if (local) {
 781                 if (*flags & LDLM_FL_BLOCK_CONV)
 782                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
 783                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
 784                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
 785                 else
 786                         ldlm_grant_lock(lock, NULL, 0, 0);
 787                 GOTO(out, ELDLM_OK);
 788         } else if (*flags & LDLM_FL_REPLAY) {
 789                 if (*flags & LDLM_FL_BLOCK_CONV) {
 790                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
 791                         GOTO(out, ELDLM_OK);
 792                 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
 793                         ldlm_resource_add_lock(res, &res->lr_waiting, lock);
 794                         GOTO(out, ELDLM_OK);
 795                 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
 796                         ldlm_grant_lock(lock, NULL, 0, 0);
 797                         GOTO(out, ELDLM_OK);
 798                 }
 799                 /* If no flags, fall through to normal enqueue path. */
 800         }
 801
 802         policy = ldlm_processing_policy_table[res->lr_type];
 803         policy(lock, flags, 1, &rc);
 804         EXIT;
 805 out:
 806         l_unlock(&ns->ns_lock);
 807         return rc;
 808 }
 809
 810 /* Must be called with namespace taken: queue is waiting or converting. */
 811 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue)
 812 {
 813         struct list_head *tmp, *pos;
 814         ldlm_processing_policy policy;
 815         int flags;
 816         int rc = LDLM_ITER_CONTINUE;
 817         ldlm_error_t err;
 818         ENTRY;
 819
 820         policy = ldlm_processing_policy_table[res->lr_type];
 821         LASSERT(policy);
 822
 823         list_for_each_safe(tmp, pos, queue) {
 824                 struct ldlm_lock *pending;
 825                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
 826
 827                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
 828
 829                 flags = 0;
 830                 rc = policy(pending, &flags, 0, &err);
 831                 if (rc != LDLM_ITER_CONTINUE)
 832                         break;
 833         }
 834
 835         RETURN(rc);
 836 }
 837
 838 int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list)
 839 {
 840         struct list_head *tmp, *pos;
 841         int rc, retval = 0;
 842         ENTRY;
 843
 844         l_check_no_ns_lock(ns);
 845
 846         list_for_each_safe(tmp, pos, rpc_list) {
 847                 struct ldlm_ast_work *w =
 848                         list_entry(tmp, struct ldlm_ast_work, w_list);
 849
 850                 /* It's possible to receive a completion AST before we've set
 851                  * the l_completion_ast pointer: either because the AST arrived
 852                  * before the reply, or simply because there's a small race
 853                  * window between receiving the reply and finishing the local
 854                  * enqueue. (bug 842)
 855                  *
 856                  * This can't happen with the blocking_ast, however, because we
 857                  * will never call the local blocking_ast until we drop our
 858                  * reader/writer reference, which we won't do until we get the
 859                  * reply and finish enqueueing. */
 860                 LASSERT(w->w_lock != NULL);
 861                 if (w->w_blocking) {
 862                         LASSERT(w->w_lock->l_blocking_ast != NULL);
 863                         rc = w->w_lock->l_blocking_ast
 864                                 (w->w_lock, &w->w_desc, w->w_data,
 865                                  LDLM_CB_BLOCKING);
 866                 } else if (w->w_lock->l_completion_ast != NULL) {
 867                         LASSERT(w->w_lock->l_completion_ast != NULL);
 868                         rc = w->w_lock->l_completion_ast(w->w_lock, w->w_flags,
 869                                                          w->w_data);
 870                 } else {
 871                         rc = 0;
 872                 }
 873                 if (rc == -ERESTART)
 874                         retval = rc;
 875                 else if (rc)
 876                         CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
 877                                "disconnect client\n");
 878                 LDLM_LOCK_PUT(w->w_lock);
 879                 list_del(&w->w_list);
 880                 OBD_FREE(w, sizeof(*w));
 881         }
 882         RETURN(retval);
 883 }
 884
 885 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
 886 {
 887         ldlm_reprocess_all(res);
 888         return LDLM_ITER_CONTINUE;
 889 }
 890
 891 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
 892 {
 893         int i, rc;
 894
 895         l_lock(&ns->ns_lock);
 896         for (i = 0; i < RES_HASH_SIZE; i++) {
 897                 struct list_head *tmp, *next;
 898                 list_for_each_safe(tmp, next, &(ns->ns_hash[i])) {
 899                         struct ldlm_resource *res =
 900                                 list_entry(tmp, struct ldlm_resource, lr_hash);
 901
 902                         ldlm_resource_getref(res);
 903                         l_unlock(&ns->ns_lock);
 904                         rc = reprocess_one_queue(res, NULL);
 905                         l_lock(&ns->ns_lock);
 906                         next = tmp->next;
 907                         ldlm_resource_putref(res);
 908                         if (rc == LDLM_ITER_STOP)
 909                                 GOTO(out, rc);
 910                 }
 911         }
 912  out:
 913         l_unlock(&ns->ns_lock);
 914         EXIT;
 915 }
 916
 917 void ldlm_reprocess_all(struct ldlm_resource *res)
 918 {
 919         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 920         int rc;
 921         ENTRY;
 922
 923         /* Local lock trees don't get reprocessed. */
 924         if (res->lr_namespace->ns_client) {
 925                 EXIT;
 926                 return;
 927         }
 928
 929  restart:
 930         l_lock(&res->lr_namespace->ns_lock);
 931         res->lr_tmp = &rpc_list;
 932
 933         rc = ldlm_reprocess_queue(res, &res->lr_converting);
 934         if (rc == LDLM_ITER_CONTINUE)
 935                 ldlm_reprocess_queue(res, &res->lr_waiting);
 936
 937         res->lr_tmp = NULL;
 938         l_unlock(&res->lr_namespace->ns_lock);
 939
 940         rc = ldlm_run_ast_work(res->lr_namespace, &rpc_list);
 941         if (rc == -ERESTART) {
 942                 LASSERT(list_empty(&rpc_list));
 943                 goto restart;
 944         }
 945         EXIT;
 946 }
 947
 948 void ldlm_cancel_callback(struct ldlm_lock *lock)
 949 {
 950         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 951         if (!(lock->l_flags & LDLM_FL_CANCEL)) {
 952                 lock->l_flags |= LDLM_FL_CANCEL;
 953                 if (lock->l_blocking_ast) {
 954                         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 955                         // l_check_no_ns_lock(lock->l_resource->lr_namespace);
 956                         lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
 957                                              LDLM_CB_CANCELING);
 958                         return;
 959                 } else {
 960                         LDLM_DEBUG(lock, "no blocking ast");
 961                 }
 962         }
 963         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 964 }
 965
 966 void ldlm_lock_cancel(struct ldlm_lock *lock)
 967 {
 968         struct ldlm_resource *res;
 969         struct ldlm_namespace *ns;
 970         ENTRY;
 971
 972         /* There's no race between calling this and taking the ns lock below;
 973          * a lock can only be put on the waiting list once, because it can only
 974          * issue a blocking AST once. */
 975         ldlm_del_waiting_lock(lock);
 976
 977         res = lock->l_resource;
 978         ns = res->lr_namespace;
 979
 980         l_lock(&ns->ns_lock);
 981         /* Please do not, no matter how tempting, remove this LBUG without
 982          * talking to me first. -phik */
 983         if (lock->l_readers || lock->l_writers) {
 984                 LDLM_DEBUG(lock, "lock still has references");
 985                 ldlm_lock_dump(D_OTHER, lock, 0);
 986                 LBUG();
 987         }
 988
 989         ldlm_cancel_callback(lock);
 990
 991         ldlm_resource_unlink_lock(lock);
 992         ldlm_lock_destroy(lock);
 993         l_unlock(&ns->ns_lock);
 994         EXIT;
 995 }
 996
 997 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
 998 {
 999         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1000         ENTRY;
1001
1002         if (lock == NULL)
1003                 RETURN(-EINVAL);
1004
1005         lock->l_ast_data = data;
1006         LDLM_LOCK_PUT(lock);
1007         RETURN(0);
1008 }
1009
1010 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1011 {
1012         struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
1013         struct ldlm_lock *lock;
1014         struct ldlm_resource *res;
1015
1016         l_lock(&ns->ns_lock);
1017         while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
1018                 lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
1019                                   struct ldlm_lock, l_export_chain);
1020                 res = ldlm_resource_getref(lock->l_resource);
1021                 LDLM_DEBUG(lock, "export %p", exp);
1022                 ldlm_lock_cancel(lock);
1023                 l_unlock(&ns->ns_lock);
1024                 ldlm_reprocess_all(res);
1025                 ldlm_resource_putref(res);
1026                 l_lock(&ns->ns_lock);
1027         }
1028         l_unlock(&ns->ns_lock);
1029 }
1030
1031 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1032                                         int *flags)
1033 {
1034         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1035         struct ldlm_resource *res;
1036         struct ldlm_namespace *ns;
1037         int granted = 0;
1038         ENTRY;
1039
1040         LBUG();
1041
1042         res = lock->l_resource;
1043         ns = res->lr_namespace;
1044
1045         l_lock(&ns->ns_lock);
1046
1047         lock->l_req_mode = new_mode;
1048         ldlm_resource_unlink_lock(lock);
1049
1050         /* If this is a local resource, put it on the appropriate list. */
1051         if (res->lr_namespace->ns_client) {
1052                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1053                         ldlm_resource_add_lock(res, &res->lr_converting, lock);
1054                 } else {
1055                         /* This should never happen, because of the way the
1056                          * server handles conversions. */
1057                         LBUG();
1058
1059                         res->lr_tmp = &rpc_list;
1060                         ldlm_grant_lock(lock, NULL, 0, 0);
1061                         res->lr_tmp = NULL;
1062                         granted = 1;
1063                         /* FIXME: completion handling not with ns_lock held ! */
1064                         if (lock->l_completion_ast)
1065                                 lock->l_completion_ast(lock, 0, NULL);
1066                 }
1067         } else {
1068                 /* FIXME: We should try the conversion right away and possibly
1069                  * return success without the need for an extra AST */
1070                 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1071                 *flags |= LDLM_FL_BLOCK_CONV;
1072         }
1073
1074         l_unlock(&ns->ns_lock);
1075
1076         if (granted)
1077                 ldlm_run_ast_work(ns, &rpc_list);
1078         RETURN(res);
1079 }
1080
1081 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1082 {
1083         char str[PTL_NALFMT_SIZE];
1084         struct obd_device *obd = NULL;
1085
1086         if (!((portal_debug | D_ERROR) & level))
1087                 return;
1088
1089         if (RES_VERSION_SIZE != 4)
1090                 LBUG();
1091
1092         if (!lock) {
1093                 CDEBUG(level, "  NULL LDLM lock\n");
1094                 return;
1095         }
1096
1097         CDEBUG(level,
1098                "  -- Lock dump: %p/"LPX64" (%x %x %x %x) (rc: %d) (pos: %d)\n",
1099                lock, lock->l_handle.h_cookie, lock->l_version[0],
1100                lock->l_version[1], lock->l_version[2], lock->l_version[3],
1101                atomic_read(&lock->l_refc), pos);
1102         if (lock->l_conn_export != NULL)
1103                 obd = lock->l_conn_export->exp_obd;
1104         if (lock->l_export && lock->l_export->exp_connection) {
1105                 CDEBUG(level, "  Node: NID "LPX64" (%s) on %s (rhandle: "LPX64")\n",
1106                        lock->l_export->exp_connection->c_peer.peer_nid,
1107                        portals_nid2str(lock->l_export->exp_connection->c_peer.peer_ni->pni_number,
1108                                        lock->l_export->exp_connection->c_peer.peer_nid, str),
1109                        lock->l_export->exp_connection->c_peer.peer_ni->pni_name,
1110                        lock->l_remote_handle.cookie);
1111         } else if (obd == NULL) {
1112                 CDEBUG(level, "  Node: local\n");
1113         } else {
1114                 struct obd_import *imp = obd->u.cli.cl_import;
1115                 CDEBUG(level, "  Node: NID "LPX64" (%s) on %s (rhandle: "LPX64")\n",
1116                        imp->imp_connection->c_peer.peer_nid,
1117                        portals_nid2str(imp->imp_connection->c_peer.peer_ni->pni_number,
1118                                        imp->imp_connection->c_peer.peer_nid, str),
1119                        imp->imp_connection->c_peer.peer_ni->pni_name,
1120                        lock->l_remote_handle.cookie);
1121         }
1122         CDEBUG(level, "  Resource: %p ("LPU64"/"LPU64")\n", lock->l_resource,
1123                lock->l_resource->lr_name.name[0],
1124                lock->l_resource->lr_name.name[1]);
1125         CDEBUG(level, "  Req mode: %d, grant mode: %d, rc: %u, read: %d, "
1126                "write: %d\n", (int)lock->l_req_mode, (int)lock->l_granted_mode,
1127                atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers);
1128         if (lock->l_resource->lr_type == LDLM_EXTENT)
1129                 CDEBUG(level, "  Extent: "LPU64" -> "LPU64"\n",
1130                        lock->l_policy_data.l_extent.start,
1131                        lock->l_policy_data.l_extent.end);
1132         else if (lock->l_resource->lr_type == LDLM_FLOCK)
1133                 CDEBUG(level, "  Pid: %d Extent: "LPU64" -> "LPU64"\n",
1134                        lock->l_policy_data.l_flock.pid,
1135                        lock->l_policy_data.l_flock.start,
1136                        lock->l_policy_data.l_flock.end);
1137 }
1138
1139 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1140 {
1141         struct ldlm_lock *lock;
1142
1143         lock = ldlm_handle2lock(lockh);
1144         if (lock == NULL)
1145                 return;
1146
1147         ldlm_lock_dump(D_OTHER, lock, 0);
1148
1149         LDLM_LOCK_PUT(lock);
1150 }