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 Cluster File Systems, Inc.
   5  *
   6  * This code is issued under the GNU General Public License.
   7  * See the file COPYING in this distribution
   8  *
   9  * by Cluster File Systems, Inc.
  10  * authors, Peter Braam <braam@clusterfs.com> &
  11  * Phil Schwan <phil@clusterfs.com>
  12  */
  13
  14 #define DEBUG_SUBSYSTEM S_LDLM
  15
  16 #include <linux/slab.h>
  17 #include <linux/module.h>
  18 #include <linux/random.h>
  19 #include <linux/lustre_dlm.h>
  20 #include <linux/lustre_mds.h>
  21
  22 /* lock types */
  23 char *ldlm_lockname[] = {
  24         [0]      "--",
  25         [LCK_EX] "EX",
  26         [LCK_PW] "PW",
  27         [LCK_PR] "PR",
  28         [LCK_CW] "CW",
  29         [LCK_CR] "CR",
  30         [LCK_NL] "NL"
  31 };
  32 char *ldlm_typename[] = {
  33         [LDLM_PLAIN]     "PLN",
  34         [LDLM_EXTENT]    "EXT",
  35         [LDLM_MDSINTENT] "INT"
  36 };
  37
  38 extern kmem_cache_t *ldlm_lock_slab;
  39 int (*mds_reint_p)(int offset, struct ptlrpc_request *req) = NULL;
  40 int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req) = NULL;
  41
  42 static int ldlm_plain_compat(struct ldlm_lock *a, struct ldlm_lock *b);
  43 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
  44                               ldlm_mode_t mode, void *data);
  45
  46 ldlm_res_compat ldlm_res_compat_table [] = {
  47         [LDLM_PLAIN] ldlm_plain_compat,
  48         [LDLM_EXTENT] ldlm_extent_compat,
  49         [LDLM_MDSINTENT] ldlm_plain_compat
  50 };
  51
  52 ldlm_res_policy ldlm_res_policy_table [] = {
  53         [LDLM_PLAIN] NULL,
  54         [LDLM_EXTENT] ldlm_extent_policy,
  55         [LDLM_MDSINTENT] ldlm_intent_policy
  56 };
  57
  58
  59 /*
  60  * REFCOUNTED LOCK OBJECTS
  61  */
  62
  63
  64 /*
  65  * Lock refcounts, during creation:
  66  *   - one special one for allocation, dec'd only once in destroy
  67  *   - one for being a lock that's in-use
  68  *   - one for the addref associated with a new lock
  69  */
  70 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
  71 {
  72         l_lock(&lock->l_resource->lr_namespace->ns_lock);
  73         lock->l_refc++;
  74         ldlm_resource_getref(lock->l_resource);
  75         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
  76         return lock;
  77 }
  78
  79 void ldlm_lock_put(struct ldlm_lock *lock)
  80 {
  81         struct lustre_lock *nslock = &lock->l_resource->lr_namespace->ns_lock;
  82         ENTRY;
  83
  84         l_lock(nslock);
  85         lock->l_refc--;
  86         LDLM_DEBUG(lock, "after refc--");
  87         if (lock->l_refc < 0)
  88                 LBUG();
  89
  90         ldlm_resource_put(lock->l_resource);
  91         if (lock->l_parent)
  92                 ldlm_lock_put(lock->l_parent);
  93
  94         if (lock->l_refc == 0 && (lock->l_flags & LDLM_FL_DESTROYED)) {
  95                 if (lock->l_connection)
  96                         ptlrpc_put_connection(lock->l_connection);
  97                 CDEBUG(D_MALLOC, "kfreed 'lock': %d at %p (tot 1).\n",
  98                        sizeof(*lock), lock);
  99                 kmem_cache_free(ldlm_lock_slab, lock);
 100         }
 101         l_unlock(nslock);
 102         EXIT;
 103 }
 104
 105 void ldlm_lock_destroy(struct ldlm_lock *lock)
 106 {
 107         ENTRY;
 108         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 109
 110         if (!list_empty(&lock->l_children)) {
 111                 LDLM_DEBUG(lock, "still has children (%p)!",
 112                            lock->l_children.next);
 113                 ldlm_lock_dump(lock);
 114                 LBUG();
 115         }
 116         if (lock->l_readers || lock->l_writers) {
 117                 LDLM_DEBUG(lock, "lock still has references");
 118                 ldlm_lock_dump(lock);
 119                 LBUG();
 120         }
 121
 122         if (!list_empty(&lock->l_res_link)) {
 123                 ldlm_lock_dump(lock);
 124                 LBUG();
 125         }
 126
 127         if (lock->l_flags & LDLM_FL_DESTROYED) {
 128                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 129                 EXIT;
 130                 return;
 131         }
 132
 133         lock->l_flags = LDLM_FL_DESTROYED;
 134         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 135         ldlm_lock_put(lock);
 136         EXIT;
 137 }
 138
 139 /*
 140    usage: pass in a resource on which you have done get
 141           pass in a parent lock on which you have done a get
 142           do not put the resource or the parent
 143    returns: lock with refcount 1
 144 */
 145 static struct ldlm_lock *ldlm_lock_new(struct ldlm_lock *parent,
 146                                        struct ldlm_resource *resource)
 147 {
 148         struct ldlm_lock *lock;
 149         ENTRY;
 150
 151         if (resource == NULL)
 152                 LBUG();
 153
 154         lock = kmem_cache_alloc(ldlm_lock_slab, SLAB_KERNEL);
 155         if (lock == NULL)
 156                 RETURN(NULL);
 157         CDEBUG(D_MALLOC, "kmalloced 'lock': %d at "
 158                "%p (tot %d).\n", sizeof(*lock), lock, 1);
 159
 160         memset(lock, 0, sizeof(*lock));
 161         get_random_bytes(&lock->l_random, sizeof(__u64));
 162
 163         lock->l_resource = resource;
 164         /* this refcount matches the one of the resource passed
 165            in which is not being put away */
 166         lock->l_refc = 1;
 167         INIT_LIST_HEAD(&lock->l_children);
 168         INIT_LIST_HEAD(&lock->l_res_link);
 169         init_waitqueue_head(&lock->l_waitq);
 170
 171         if (parent != NULL) {
 172                 l_lock(&parent->l_resource->lr_namespace->ns_lock);
 173                 lock->l_parent = parent;
 174                 list_add(&lock->l_childof, &parent->l_children);
 175                 l_unlock(&parent->l_resource->lr_namespace->ns_lock);
 176         }
 177         /* this is the extra refcount, to prevent the lock
 178            evaporating */
 179         ldlm_lock_get(lock);
 180         RETURN(lock);
 181 }
 182
 183 int ldlm_lock_change_resource(struct ldlm_lock *lock, __u64 new_resid[3])
 184 {
 185         struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
 186         struct ldlm_resource *oldres = lock->l_resource;
 187         int type, i;
 188         ENTRY;
 189
 190         l_lock(&ns->ns_lock);
 191         type = lock->l_resource->lr_type;
 192
 193         lock->l_resource = ldlm_resource_get(ns, NULL, new_resid, type, 1);
 194         if (lock->l_resource == NULL) {
 195                 LBUG();
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* move references over */
 200         for (i = 0; i < lock->l_refc; i++) {
 201                 int rc;
 202                 ldlm_resource_getref(lock->l_resource);
 203                 rc = ldlm_resource_put(oldres);
 204                 if (rc == 1 && i != lock->l_refc - 1)
 205                         LBUG();
 206         }
 207         /* compensate for the initial get above.. */
 208         ldlm_resource_put(lock->l_resource);
 209
 210         l_unlock(&ns->ns_lock);
 211         RETURN(0);
 212 }
 213
 214 /*
 215  *  HANDLES
 216  */
 217
 218 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh)
 219 {
 220         lockh->addr = (__u64)(unsigned long)lock;
 221         lockh->cookie = lock->l_random;
 222 }
 223
 224 struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *handle)
 225 {
 226         struct ldlm_lock *lock = NULL;
 227         ENTRY;
 228
 229         if (!handle || !handle->addr)
 230                 RETURN(NULL);
 231
 232         lock = (struct ldlm_lock *)(unsigned long)(handle->addr);
 233         if (!kmem_cache_validate(ldlm_lock_slab, (void *)lock))
 234                 RETURN(NULL);
 235
 236         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 237         if (lock->l_random != handle->cookie)
 238                 GOTO(out, handle = NULL);
 239
 240         if (lock->l_flags & LDLM_FL_DESTROYED)
 241                 GOTO(out, handle = NULL);
 242
 243         ldlm_lock_get(lock);
 244         EXIT;
 245  out:
 246         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 247         return lock;
 248 }
 249
 250
 251
 252 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
 253                               ldlm_mode_t mode, void *data)
 254 {
 255         struct ptlrpc_request *req = req_cookie;
 256         int rc = 0;
 257         ENTRY;
 258
 259         if (!req_cookie)
 260                 RETURN(0);
 261
 262         if (req->rq_reqmsg->bufcount > 1) {
 263                 /* an intent needs to be considered */
 264                 struct ldlm_intent *it = lustre_msg_buf(req->rq_reqmsg, 1);
 265                 struct mds_body *mds_rep;
 266                 struct ldlm_reply *rep;
 267                 __u64 new_resid[3] = {0, 0, 0}, old_res;
 268                 int bufcount = -1, rc, size[3] = {sizeof(struct ldlm_reply),
 269                                                   sizeof(struct mds_body),
 270                                                   sizeof(struct obdo)};
 271
 272                 it->opc = NTOH__u64(it->opc);
 273
 274                 LDLM_DEBUG(lock, "intent policy, opc: %Ld", it->opc);
 275
 276                 /* prepare reply */
 277                 switch(it->opc) {
 278                 case IT_GETATTR:
 279                         /* Note that in the negative case you may be returning
 280                          * a file and its obdo */
 281                 case IT_CREAT:
 282                 case IT_CREAT|IT_OPEN:
 283                 case IT_MKDIR:
 284                 case IT_SYMLINK:
 285                 case IT_MKNOD:
 286                 case IT_LINK:
 287                 case IT_OPEN:
 288                 case IT_RENAME:
 289                         bufcount = 3;
 290                         break;
 291                 case IT_RENAME2:
 292                         bufcount = 5;
 293                         break;
 294                 case IT_UNLINK:
 295                         bufcount = 2;
 296                         size[1] = sizeof(struct obdo);
 297                         break;
 298                 case IT_RMDIR:
 299                         bufcount = 1;
 300                         break;
 301                 default:
 302                         LBUG();
 303                 }
 304
 305                 rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
 306                                      &req->rq_repmsg);
 307                 if (rc) {
 308                         rc = req->rq_status = -ENOMEM;
 309                         RETURN(rc);
 310                 }
 311
 312                 rep = lustre_msg_buf(req->rq_repmsg, 0);
 313                 rep->lock_policy_res1 = 1;
 314
 315                 /* execute policy */
 316                 switch (it->opc) {
 317                 case IT_CREAT:
 318                 case IT_CREAT|IT_OPEN:
 319                 case IT_MKDIR:
 320                 case IT_SETATTR:
 321                 case IT_SYMLINK:
 322                 case IT_MKNOD:
 323                 case IT_LINK:
 324                 case IT_UNLINK:
 325                 case IT_RMDIR:
 326                 case IT_RENAME2:
 327                         if (mds_reint_p == NULL)
 328                                 mds_reint_p =
 329                                         inter_module_get_request
 330                                         ("mds_reint", "mds");
 331                         if (IS_ERR(mds_reint_p)) {
 332                                 CERROR("MDSINTENT locks require the MDS "
 333                                        "module.\n");
 334                                 LBUG();
 335                                 RETURN(-EINVAL);
 336                         }
 337                         rc = mds_reint_p(2, req);
 338                         if (rc)
 339                                 LBUG();
 340                         break;
 341                 case IT_GETATTR:
 342                 case IT_READDIR:
 343                 case IT_RENAME:
 344                 case IT_OPEN:
 345                         if (mds_getattr_name_p == NULL)
 346                                 mds_getattr_name_p =
 347                                         inter_module_get_request
 348                                         ("mds_getattr_name", "mds");
 349                         if (IS_ERR(mds_getattr_name_p)) {
 350                                 CERROR("MDSINTENT locks require the MDS "
 351                                        "module.\n");
 352                                 LBUG();
 353                                 RETURN(-EINVAL);
 354                         }
 355                         rc = mds_getattr_name_p(2, req);
 356                         /* FIXME: we need to sit down and decide on who should
 357                          * set req->rq_status, who should return negative and
 358                          * positive return values, and what they all mean. */
 359                         if (rc || req->rq_status != 0) {
 360                                 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
 361                                 rep->lock_policy_res2 = req->rq_status;
 362                                 RETURN(ELDLM_LOCK_ABORTED);
 363                         }
 364                         break;
 365                 case IT_READDIR|IT_OPEN:
 366                         LBUG();
 367                         break;
 368                 default:
 369                         CERROR("Unhandled intent\n");
 370                         LBUG();
 371                 }
 372
 373                 if (it->opc == IT_UNLINK || it->opc == IT_RMDIR ||
 374                     it->opc == IT_RENAME || it->opc == IT_RENAME2)
 375                         RETURN(ELDLM_LOCK_ABORTED);
 376
 377                 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
 378                 rep->lock_policy_res2 = req->rq_status;
 379                 new_resid[0] = NTOH__u32(mds_rep->ino);
 380                 if (new_resid[0] == 0)
 381                         LBUG();
 382                 old_res = lock->l_resource->lr_name[0];
 383
 384                 CDEBUG(D_INFO, "remote intent: locking %d instead of"
 385                        "%ld\n", mds_rep->ino, (long)old_res);
 386
 387                 ldlm_lock_change_resource(lock, new_resid);
 388                 if (lock->l_resource == NULL) {
 389                         LBUG();
 390                         RETURN(-ENOMEM);
 391                 }
 392                 LDLM_DEBUG(lock, "intent policy, old res %ld",
 393                            (long)old_res);
 394                 RETURN(ELDLM_LOCK_CHANGED);
 395         } else {
 396                 int size = sizeof(struct ldlm_reply);
 397                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 398                                      &req->rq_repmsg);
 399                 if (rc) {
 400                         CERROR("out of memory\n");
 401                         LBUG();
 402                         RETURN(-ENOMEM);
 403                 }
 404         }
 405         RETURN(rc);
 406 }
 407
 408 static int ldlm_plain_compat(struct ldlm_lock *a, struct ldlm_lock *b)
 409 {
 410         return lockmode_compat(a->l_req_mode, b->l_req_mode);
 411 }
 412
 413 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
 414 {
 415         ldlm_res2desc(lock->l_resource, &desc->l_resource);
 416         desc->l_req_mode = lock->l_req_mode;
 417         desc->l_granted_mode = lock->l_granted_mode;
 418         memcpy(&desc->l_extent, &lock->l_extent, sizeof(desc->l_extent));
 419         memcpy(desc->l_version, lock->l_version, sizeof(desc->l_version));
 420 }
 421
 422 static void ldlm_add_ast_work_item(struct ldlm_lock *lock,
 423                                    struct ldlm_lock *new)
 424 {
 425         struct ldlm_ast_work *w;
 426         ENTRY;
 427
 428         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 429         if (new && (lock->l_flags & LDLM_FL_AST_SENT))
 430                 GOTO(out, 0);
 431
 432         OBD_ALLOC(w, sizeof(*w));
 433         if (!w) {
 434                 LBUG();
 435                 return;
 436         }
 437
 438         if (new) {
 439                 lock->l_flags |= LDLM_FL_AST_SENT;
 440                 w->w_blocking = 1;
 441                 ldlm_lock2desc(new, &w->w_desc);
 442         }
 443
 444         w->w_lock = ldlm_lock_get(lock);
 445         list_add(&w->w_list, lock->l_resource->lr_tmp);
 446  out:
 447         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 448         return;
 449 }
 450
 451 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
 452 {
 453         struct ldlm_lock *lock;
 454
 455         lock = ldlm_handle2lock(lockh);
 456         ldlm_lock_addref_internal(lock, mode);
 457         ldlm_lock_put(lock);
 458 }
 459
 460 /* only called for local locks */
 461 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
 462 {
 463         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 464         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 465                 lock->l_readers++;
 466         else
 467                 lock->l_writers++;
 468         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 469         ldlm_lock_get(lock);
 470         LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
 471 }
 472
 473 /* Args: unlocked lock */
 474 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
 475 {
 476         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 477         ENTRY;
 478
 479         if (lock == NULL)
 480                 LBUG();
 481
 482         LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
 483         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 484         if (mode == LCK_NL || mode == LCK_CR || mode == LCK_PR)
 485                 lock->l_readers--;
 486         else
 487                 lock->l_writers--;
 488
 489         /* If we received a blocked AST and this was the last reference,
 490          * run the callback. */
 491         if (!lock->l_readers && !lock->l_writers &&
 492             (lock->l_flags & LDLM_FL_CBPENDING)) {
 493                 struct lustre_handle lockh;
 494
 495                 if (!lock->l_resource->lr_namespace->ns_client) {
 496                         CERROR("LDLM_FL_CBPENDING set on non-local lock!\n");
 497                         LBUG();
 498                 }
 499
 500                 LDLM_DEBUG(lock, "final decref done on cbpending lock");
 501                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 502
 503                 ldlm_lock2handle(lock, &lockh);
 504                 /* FIXME: -1 is a really, really bad 'desc' */
 505                 lock->l_blocking_ast(&lockh, (void *)-1, lock->l_data,
 506                                      lock->l_data_len);
 507         } else
 508                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 509
 510         ldlm_lock_put(lock); /* matches the ldlm_lock_get in addref */
 511         ldlm_lock_put(lock); /* matches the handle2lock above */
 512
 513         EXIT;
 514 }
 515
 516 static int ldlm_lock_compat_list(struct ldlm_lock *lock, int send_cbs,
 517                              struct list_head *queue)
 518 {
 519         struct list_head *tmp, *pos;
 520         int rc = 1;
 521
 522         list_for_each_safe(tmp, pos, queue) {
 523                 struct ldlm_lock *child;
 524                 ldlm_res_compat compat;
 525
 526                 child = list_entry(tmp, struct ldlm_lock, l_res_link);
 527                 if (lock == child)
 528                         continue;
 529
 530                 compat = ldlm_res_compat_table[child->l_resource->lr_type];
 531                 if (compat && compat(child, lock)) {
 532                         CDEBUG(D_OTHER, "compat function succeded, next.\n");
 533                         continue;
 534                 }
 535                 if (lockmode_compat(child->l_granted_mode, lock->l_req_mode)) {
 536                         CDEBUG(D_OTHER, "lock modes are compatible, next.\n");
 537                         continue;
 538                 }
 539
 540                 rc = 0;
 541
 542                 if (send_cbs && child->l_blocking_ast != NULL) {
 543                         CDEBUG(D_OTHER, "incompatible; sending blocking "
 544                                "AST.\n");
 545                         ldlm_add_ast_work_item(child, lock);
 546                 }
 547         }
 548
 549         return rc;
 550 }
 551
 552 static int ldlm_lock_compat(struct ldlm_lock *lock, int send_cbs)
 553 {
 554         int rc;
 555         ENTRY;
 556
 557         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 558         rc = ldlm_lock_compat_list(lock, send_cbs,
 559                                    &lock->l_resource->lr_granted);
 560         /* FIXME: should we be sending ASTs to converting? */
 561         if (rc)
 562                 rc = ldlm_lock_compat_list
 563                         (lock, send_cbs, &lock->l_resource->lr_converting);
 564
 565         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 566         RETURN(rc);
 567 }
 568
 569 /* NOTE: called by
 570    - ldlm_handle_enqueuque - resource
 571 */
 572 void ldlm_grant_lock(struct ldlm_lock *lock)
 573 {
 574         struct ldlm_resource *res = lock->l_resource;
 575         ENTRY;
 576
 577         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 578         ldlm_resource_add_lock(res, &res->lr_granted, lock);
 579         lock->l_granted_mode = lock->l_req_mode;
 580
 581         if (lock->l_granted_mode < res->lr_most_restr)
 582                 res->lr_most_restr = lock->l_granted_mode;
 583
 584         if (lock->l_completion_ast) {
 585                 ldlm_add_ast_work_item(lock, NULL);
 586         }
 587         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 588         EXIT;
 589 }
 590
 591 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
 592                                       struct ldlm_extent *extent)
 593 {
 594         struct ldlm_lock *lock;
 595         struct list_head *tmp;
 596
 597         list_for_each(tmp, queue) {
 598                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 599
 600                 if (lock->l_flags & LDLM_FL_CBPENDING)
 601                         continue;
 602
 603                 /* lock_convert() takes the resource lock, so we're sure that
 604                  * req_mode, lr_type, and l_cookie won't change beneath us */
 605                 if (lock->l_req_mode != mode)
 606                         continue;
 607
 608                 if (lock->l_resource->lr_type == LDLM_EXTENT &&
 609                     (lock->l_extent.start > extent->start ||
 610                      lock->l_extent.end < extent->end))
 611                         continue;
 612
 613                 ldlm_lock_addref_internal(lock, mode);
 614                 return lock;
 615         }
 616
 617         return NULL;
 618 }
 619
 620 /* Must be called with no resource or lock locks held.
 621  *
 622  * Returns 1 if it finds an already-existing lock that is compatible; in this
 623  * case, lockh is filled in with a addref()ed lock */
 624 int ldlm_lock_match(struct ldlm_namespace *ns, __u64 *res_id, __u32 type,
 625                     void *cookie, int cookielen, ldlm_mode_t mode,
 626                     struct lustre_handle *lockh)
 627 {
 628         struct ldlm_resource *res;
 629         struct ldlm_lock *lock;
 630         int rc = 0;
 631         ENTRY;
 632
 633         res = ldlm_resource_get(ns, NULL, res_id, type, 0);
 634         if (res == NULL)
 635                 RETURN(0);
 636
 637         ns = res->lr_namespace;
 638         l_lock(&ns->ns_lock);
 639
 640         if ((lock = search_queue(&res->lr_granted, mode, cookie)))
 641                 GOTO(out, rc = 1);
 642         if ((lock = search_queue(&res->lr_converting, mode, cookie)))
 643                 GOTO(out, rc = 1);
 644         if ((lock = search_queue(&res->lr_waiting, mode, cookie)))
 645                 GOTO(out, rc = 1);
 646
 647         EXIT;
 648  out:
 649         ldlm_resource_put(res);
 650         l_unlock(&ns->ns_lock);
 651
 652         if (lock) {
 653                 ldlm_lock2handle(lock, lockh);
 654                 wait_event_interruptible(lock->l_waitq, lock->l_req_mode ==
 655                                          lock->l_granted_mode);
 656         }
 657         if (rc)
 658                 LDLM_DEBUG(lock, "matched");
 659         else
 660                 LDLM_DEBUG_NOLOCK("not matched");
 661         return rc;
 662 }
 663
 664 /*   Returns a referenced, lock */
 665 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
 666                                    struct lustre_handle *parent_lock_handle,
 667                                    __u64 *res_id, __u32 type,
 668                                    ldlm_mode_t mode,
 669                                    void *data,
 670                                    __u32 data_len)
 671 {
 672         struct ldlm_resource *res, *parent_res = NULL;
 673         struct ldlm_lock *lock, *parent_lock;
 674
 675         parent_lock = ldlm_handle2lock(parent_lock_handle);
 676         if (parent_lock)
 677                 parent_res = parent_lock->l_resource;
 678
 679         res = ldlm_resource_get(ns, parent_res, res_id, type, 1);
 680         if (res == NULL)
 681                 RETURN(NULL);
 682
 683         lock = ldlm_lock_new(parent_lock, res);
 684         if (lock == NULL) {
 685                 ldlm_resource_put(res);
 686                 RETURN(NULL);
 687         }
 688
 689         lock->l_req_mode = mode;
 690         lock->l_data = data;
 691         lock->l_data_len = data_len;
 692
 693         return lock;
 694 }
 695
 696 /* Must be called with lock->l_lock and lock->l_resource->lr_lock not held */
 697 ldlm_error_t ldlm_lock_enqueue(struct ldlm_lock *lock,
 698                                void *cookie, int cookie_len,
 699                                int *flags,
 700                                ldlm_lock_callback completion,
 701                                ldlm_lock_callback blocking)
 702 {
 703         struct ldlm_resource *res;
 704         int local;
 705         ldlm_res_policy policy;
 706         ENTRY;
 707
 708         res = lock->l_resource;
 709         lock->l_blocking_ast = blocking;
 710
 711         if (res->lr_type == LDLM_EXTENT)
 712                 memcpy(&lock->l_extent, cookie, sizeof(lock->l_extent));
 713
 714         /* policies are not executed on the client */
 715         local = res->lr_namespace->ns_client;
 716         if (!local && (policy = ldlm_res_policy_table[res->lr_type])) {
 717                 int rc;
 718                 rc = policy(lock, cookie, lock->l_req_mode, NULL);
 719
 720                 if (rc == ELDLM_LOCK_CHANGED) {
 721                         res = lock->l_resource;
 722                         *flags |= LDLM_FL_LOCK_CHANGED;
 723                 } else if (rc == ELDLM_LOCK_ABORTED) {
 724                         ldlm_lock_destroy(lock);
 725                         RETURN(rc);
 726                 }
 727         }
 728
 729         lock->l_cookie = cookie;
 730         lock->l_cookie_len = cookie_len;
 731
 732         if (local && lock->l_req_mode == lock->l_granted_mode) {
 733                 /* The server returned a blocked lock, but it was granted before
 734                  * we got a chance to actually enqueue it.  We don't need to do
 735                  * anything else. */
 736                 GOTO(out, ELDLM_OK);
 737         }
 738
 739         /* If this is a local resource, put it on the appropriate list. */
 740         /* FIXME: don't like this: can we call ldlm_resource_unlink_lock? */
 741         list_del_init(&lock->l_res_link);
 742         if (local) {
 743                 if (*flags & LDLM_FL_BLOCK_CONV)
 744                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 745                                                lock);
 746                 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
 747                         ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 748                 else
 749                         ldlm_grant_lock(lock);
 750                 GOTO(out, ELDLM_OK);
 751         }
 752
 753         /* FIXME: We may want to optimize by checking lr_most_restr */
 754         if (!list_empty(&res->lr_converting)) {
 755                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 756                 *flags |= LDLM_FL_BLOCK_CONV;
 757                 GOTO(out, ELDLM_OK);
 758         }
 759         if (!list_empty(&res->lr_waiting)) {
 760                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 761                 *flags |= LDLM_FL_BLOCK_WAIT;
 762                 GOTO(out, ELDLM_OK);
 763         }
 764         if (!ldlm_lock_compat(lock,0)) {
 765                 ldlm_resource_add_lock(res, res->lr_waiting.prev, lock);
 766                 *flags |= LDLM_FL_BLOCK_GRANTED;
 767                 GOTO(out, ELDLM_OK);
 768         }
 769
 770         ldlm_grant_lock(lock);
 771         EXIT;
 772  out:
 773         /* Don't set 'completion_ast' until here so that if the lock is granted
 774          * immediately we don't do an unnecessary completion call. */
 775         lock->l_completion_ast = completion;
 776         return ELDLM_OK;
 777 }
 778
 779 /* Must be called with namespace taken: queue is waiting or converting. */
 780 static int ldlm_reprocess_queue(struct ldlm_resource *res,
 781                                 struct list_head *queue)
 782 {
 783         struct list_head *tmp, *pos;
 784         ENTRY;
 785
 786         list_for_each_safe(tmp, pos, queue) {
 787                 struct ldlm_lock *pending;
 788                 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
 789
 790                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
 791
 792                 if (!ldlm_lock_compat(pending, 1))
 793                         RETURN(1);
 794
 795                 list_del_init(&pending->l_res_link);
 796                 ldlm_grant_lock(pending);
 797         }
 798
 799         RETURN(0);
 800 }
 801
 802 void ldlm_run_ast_work(struct list_head *rpc_list)
 803 {
 804         struct list_head *tmp, *pos;
 805         int rc;
 806         ENTRY;
 807
 808         list_for_each_safe(tmp, pos, rpc_list) {
 809                 struct ldlm_ast_work *w =
 810                         list_entry(tmp, struct ldlm_ast_work, w_list);
 811                 struct lustre_handle lockh;
 812
 813                 ldlm_lock2handle(w->w_lock, &lockh);
 814                 if (w->w_blocking)
 815                         rc = w->w_lock->l_blocking_ast
 816                                 (&lockh, &w->w_desc, w->w_data, w->w_datalen);
 817                 else
 818                         rc = w->w_lock->l_completion_ast
 819                                 (&lockh, NULL, w->w_data, w->w_datalen);
 820                 if (rc)
 821                         CERROR("Failed AST - should clean & disconnect "
 822                                "client\n");
 823                 ldlm_lock_put(w->w_lock);
 824                 list_del(&w->w_list);
 825                 OBD_FREE(w, sizeof(*w));
 826         }
 827         EXIT;
 828 }
 829
 830 /* Must be called with resource->lr_lock not taken. */
 831 void ldlm_reprocess_all(struct ldlm_resource *res)
 832 {
 833         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 834         ENTRY;
 835
 836         /* Local lock trees don't get reprocessed. */
 837         if (res->lr_namespace->ns_client) {
 838                 EXIT;
 839                 return;
 840         }
 841
 842         l_lock(&res->lr_namespace->ns_lock);
 843         res->lr_tmp = &rpc_list;
 844
 845         ldlm_reprocess_queue(res, &res->lr_converting);
 846         if (list_empty(&res->lr_converting))
 847                 ldlm_reprocess_queue(res, &res->lr_waiting);
 848
 849         res->lr_tmp = NULL;
 850         l_unlock(&res->lr_namespace->ns_lock);
 851
 852         ldlm_run_ast_work(&rpc_list);
 853         EXIT;
 854 }
 855
 856 /* Must be called with lock and lock->l_resource unlocked */
 857 void ldlm_lock_cancel(struct ldlm_lock *lock)
 858 {
 859         struct ldlm_resource *res;
 860         struct ldlm_namespace *ns;
 861         ENTRY;
 862
 863         res = lock->l_resource;
 864         ns = res->lr_namespace;
 865
 866         l_lock(&ns->ns_lock);
 867         if (lock->l_readers || lock->l_writers)
 868                 CDEBUG(D_INFO, "lock still has references (%d readers, %d "
 869                        "writers)\n", lock->l_readers, lock->l_writers);
 870
 871         ldlm_resource_unlink_lock(lock);
 872         ldlm_lock_destroy(lock);
 873         l_unlock(&ns->ns_lock);
 874 }
 875
 876 /* Must be called with lock and lock->l_resource unlocked */
 877 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
 878                                         int *flags)
 879 {
 880         struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 881         struct ldlm_resource *res;
 882         struct ldlm_namespace *ns;
 883         int granted = 0;
 884         ENTRY;
 885
 886         res = lock->l_resource;
 887         ns = res->lr_namespace;
 888
 889         l_lock(&ns->ns_lock);
 890
 891         lock->l_req_mode = new_mode;
 892         ldlm_resource_unlink_lock(lock);
 893
 894         /* If this is a local resource, put it on the appropriate list. */
 895         if (res->lr_namespace->ns_client) {
 896                 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED))
 897                         ldlm_resource_add_lock(res, res->lr_converting.prev,
 898                                                lock);
 899                 else {
 900                         res->lr_tmp = &rpc_list;
 901                         ldlm_grant_lock(lock);
 902                         res->lr_tmp = NULL;
 903                         granted = 1;
 904                         /* FIXME: completion handling not with ns_lock held ! */
 905                         wake_up(&lock->l_waitq);
 906                 }
 907         } else
 908                 list_add(&lock->l_res_link, res->lr_converting.prev);
 909
 910         l_unlock(&ns->ns_lock);
 911
 912         if (granted)
 913                 ldlm_run_ast_work(&rpc_list);
 914         RETURN(res);
 915 }
 916
 917 void ldlm_lock_dump(struct ldlm_lock *lock)
 918 {
 919         char ver[128];
 920
 921         if (!(portal_debug & D_OTHER))
 922                 return;
 923
 924         if (RES_VERSION_SIZE != 4)
 925                 LBUG();
 926
 927         if (!lock) {
 928                 CDEBUG(D_OTHER, "  NULL LDLM lock\n");
 929                 return;
 930         }
 931
 932         snprintf(ver, sizeof(ver), "%x %x %x %x",
 933                  lock->l_version[0], lock->l_version[1],
 934                  lock->l_version[2], lock->l_version[3]);
 935
 936         CDEBUG(D_OTHER, "  -- Lock dump: %p (%s)\n", lock, ver);
 937         CDEBUG(D_OTHER, "  Parent: %p\n", lock->l_parent);
 938         CDEBUG(D_OTHER, "  Resource: %p (%Ld)\n", lock->l_resource,
 939                lock->l_resource->lr_name[0]);
 940         CDEBUG(D_OTHER, "  Requested mode: %d, granted mode: %d\n",
 941                (int)lock->l_req_mode, (int)lock->l_granted_mode);
 942         CDEBUG(D_OTHER, "  Readers: %u ; Writers; %u\n",
 943                lock->l_readers, lock->l_writers);
 944         if (lock->l_resource->lr_type == LDLM_EXTENT)
 945                 CDEBUG(D_OTHER, "  Extent: %Lu -> %Lu\n",
 946                        (unsigned long long)lock->l_extent.start,
 947                        (unsigned long long)lock->l_extent.end);
 948 }