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