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