fs/ocfs2/dlm/dlmlock.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * dlmlock.c
   4  *
   5  * underlying calls for lock creation
   6  *
   7  * Copyright (C) 2004 Oracle.  All rights reserved.
   8  */
   9
  10
  11 #include <linux/module.h>
  12 #include <linux/fs.h>
  13 #include <linux/types.h>
  14 #include <linux/slab.h>
  15 #include <linux/highmem.h>
  16 #include <linux/init.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/random.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/socket.h>
  21 #include <linux/inet.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/delay.h>
  24
  25
  26 #include "../cluster/heartbeat.h"
  27 #include "../cluster/nodemanager.h"
  28 #include "../cluster/tcp.h"
  29
  30 #include "dlmapi.h"
  31 #include "dlmcommon.h"
  32
  33 #include "dlmconvert.h"
  34
  35 #define MLOG_MASK_PREFIX ML_DLM
  36 #include "../cluster/masklog.h"
  37
  38 static struct kmem_cache *dlm_lock_cache;
  39
  40 static DEFINE_SPINLOCK(dlm_cookie_lock);
  41 static u64 dlm_next_cookie = 1;
  42
  43 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
  44                                                struct dlm_lock_resource *res,
  45                                                struct dlm_lock *lock, int flags);
  46 static void dlm_init_lock(struct dlm_lock *newlock, int type,
  47                           u8 node, u64 cookie);
  48 static void dlm_lock_release(struct kref *kref);
  49 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
  50
  51 int dlm_init_lock_cache(void)
  52 {
  53         dlm_lock_cache = kmem_cache_create("o2dlm_lock",
  54                                            sizeof(struct dlm_lock),
  55                                            0, SLAB_HWCACHE_ALIGN, NULL);
  56         if (dlm_lock_cache == NULL)
  57                 return -ENOMEM;
  58         return 0;
  59 }
  60
  61 void dlm_destroy_lock_cache(void)
  62 {
  63         kmem_cache_destroy(dlm_lock_cache);
  64 }
  65
  66 /* Tell us whether we can grant a new lock request.
  67  * locking:
  68  *   caller needs:  res->spinlock
  69  *   taken:         none
  70  *   held on exit:  none
  71  * returns: 1 if the lock can be granted, 0 otherwise.
  72  */
  73 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
  74                                   struct dlm_lock *lock)
  75 {
  76         struct dlm_lock *tmplock;
  77
  78         list_for_each_entry(tmplock, &res->granted, list) {
  79                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
  80                         return 0;
  81         }
  82
  83         list_for_each_entry(tmplock, &res->converting, list) {
  84                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
  85                         return 0;
  86                 if (!dlm_lock_compatible(tmplock->ml.convert_type,
  87                                          lock->ml.type))
  88                         return 0;
  89         }
  90
  91         return 1;
  92 }
  93
  94 /* performs lock creation at the lockres master site
  95  * locking:
  96  *   caller needs:  none
  97  *   taken:         takes and drops res->spinlock
  98  *   held on exit:  none
  99  * returns: DLM_NORMAL, DLM_NOTQUEUED
 100  */
 101 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
 102                                       struct dlm_lock_resource *res,
 103                                       struct dlm_lock *lock, int flags)
 104 {
 105         int call_ast = 0, kick_thread = 0;
 106         enum dlm_status status = DLM_NORMAL;
 107
 108         mlog(0, "type=%d\n", lock->ml.type);
 109
 110         spin_lock(&res->spinlock);
 111         /* if called from dlm_create_lock_handler, need to
 112          * ensure it will not sleep in dlm_wait_on_lockres */
 113         status = __dlm_lockres_state_to_status(res);
 114         if (status != DLM_NORMAL &&
 115             lock->ml.node != dlm->node_num) {
 116                 /* erf.  state changed after lock was dropped. */
 117                 spin_unlock(&res->spinlock);
 118                 dlm_error(status);
 119                 return status;
 120         }
 121         __dlm_wait_on_lockres(res);
 122         __dlm_lockres_reserve_ast(res);
 123
 124         if (dlm_can_grant_new_lock(res, lock)) {
 125                 mlog(0, "I can grant this lock right away\n");
 126                 /* got it right away */
 127                 lock->lksb->status = DLM_NORMAL;
 128                 status = DLM_NORMAL;
 129                 dlm_lock_get(lock);
 130                 list_add_tail(&lock->list, &res->granted);
 131
 132                 /* for the recovery lock, we can't allow the ast
 133                  * to be queued since the dlmthread is already
 134                  * frozen.  but the recovery lock is always locked
 135                  * with LKM_NOQUEUE so we do not need the ast in
 136                  * this special case */
 137                 if (!dlm_is_recovery_lock(res->lockname.name,
 138                                           res->lockname.len)) {
 139                         kick_thread = 1;
 140                         call_ast = 1;
 141                 } else {
 142                         mlog(0, "%s: returning DLM_NORMAL to "
 143                              "node %u for reco lock\n", dlm->name,
 144                              lock->ml.node);
 145                 }
 146         } else {
 147                 /* for NOQUEUE request, unless we get the
 148                  * lock right away, return DLM_NOTQUEUED */
 149                 if (flags & LKM_NOQUEUE) {
 150                         status = DLM_NOTQUEUED;
 151                         if (dlm_is_recovery_lock(res->lockname.name,
 152                                                  res->lockname.len)) {
 153                                 mlog(0, "%s: returning NOTQUEUED to "
 154                                      "node %u for reco lock\n", dlm->name,
 155                                      lock->ml.node);
 156                         }
 157                 } else {
 158                         status = DLM_NORMAL;
 159                         dlm_lock_get(lock);
 160                         list_add_tail(&lock->list, &res->blocked);
 161                         kick_thread = 1;
 162                 }
 163         }
 164
 165         spin_unlock(&res->spinlock);
 166         wake_up(&res->wq);
 167
 168         /* either queue the ast or release it */
 169         if (call_ast)
 170                 dlm_queue_ast(dlm, lock);
 171         else
 172                 dlm_lockres_release_ast(dlm, res);
 173
 174         dlm_lockres_calc_usage(dlm, res);
 175         if (kick_thread)
 176                 dlm_kick_thread(dlm, res);
 177
 178         return status;
 179 }
 180
 181 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
 182                              struct dlm_lock *lock)
 183 {
 184         /* remove from local queue if it failed */
 185         list_del_init(&lock->list);
 186         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 187 }
 188
 189
 190 /*
 191  * locking:
 192  *   caller needs:  none
 193  *   taken:         takes and drops res->spinlock
 194  *   held on exit:  none
 195  * returns: DLM_DENIED, DLM_RECOVERING, or net status
 196  */
 197 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 198                                       struct dlm_lock_resource *res,
 199                                       struct dlm_lock *lock, int flags)
 200 {
 201         enum dlm_status status = DLM_DENIED;
 202         int lockres_changed = 1;
 203
 204         mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
 205              lock->ml.type, res->lockname.len,
 206              res->lockname.name, flags);
 207
 208         /*
 209          * Wait if resource is getting recovered, remastered, etc.
 210          * If the resource was remastered and new owner is self, then exit.
 211          */
 212         spin_lock(&res->spinlock);
 213         __dlm_wait_on_lockres(res);
 214         if (res->owner == dlm->node_num) {
 215                 spin_unlock(&res->spinlock);
 216                 return DLM_RECOVERING;
 217         }
 218         res->state |= DLM_LOCK_RES_IN_PROGRESS;
 219
 220         /* add lock to local (secondary) queue */
 221         dlm_lock_get(lock);
 222         list_add_tail(&lock->list, &res->blocked);
 223         lock->lock_pending = 1;
 224         spin_unlock(&res->spinlock);
 225
 226         /* spec seems to say that you will get DLM_NORMAL when the lock
 227          * has been queued, meaning we need to wait for a reply here. */
 228         status = dlm_send_remote_lock_request(dlm, res, lock, flags);
 229
 230         spin_lock(&res->spinlock);
 231         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 232         lock->lock_pending = 0;
 233         if (status != DLM_NORMAL) {
 234                 if (status == DLM_RECOVERING &&
 235                     dlm_is_recovery_lock(res->lockname.name,
 236                                          res->lockname.len)) {
 237                         /* recovery lock was mastered by dead node.
 238                          * we need to have calc_usage shoot down this
 239                          * lockres and completely remaster it. */
 240                         mlog(0, "%s: recovery lock was owned by "
 241                              "dead node %u, remaster it now.\n",
 242                              dlm->name, res->owner);
 243                 } else if (status != DLM_NOTQUEUED) {
 244                         /*
 245                          * DO NOT call calc_usage, as this would unhash
 246                          * the remote lockres before we ever get to use
 247                          * it.  treat as if we never made any change to
 248                          * the lockres.
 249                          */
 250                         lockres_changed = 0;
 251                         dlm_error(status);
 252                 }
 253                 dlm_revert_pending_lock(res, lock);
 254                 dlm_lock_put(lock);
 255         } else if (dlm_is_recovery_lock(res->lockname.name,
 256                                         res->lockname.len)) {
 257                 /* special case for the $RECOVERY lock.
 258                  * there will never be an AST delivered to put
 259                  * this lock on the proper secondary queue
 260                  * (granted), so do it manually. */
 261                 mlog(0, "%s: $RECOVERY lock for this node (%u) is "
 262                      "mastered by %u; got lock, manually granting (no ast)\n",
 263                      dlm->name, dlm->node_num, res->owner);
 264                 list_move_tail(&lock->list, &res->granted);
 265         }
 266         spin_unlock(&res->spinlock);
 267
 268         if (lockres_changed)
 269                 dlm_lockres_calc_usage(dlm, res);
 270
 271         wake_up(&res->wq);
 272         return status;
 273 }
 274
 275
 276 /* for remote lock creation.
 277  * locking:
 278  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
 279  *   taken:         none
 280  *   held on exit:  none
 281  * returns: DLM_NOLOCKMGR, or net status
 282  */
 283 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
 284                                                struct dlm_lock_resource *res,
 285                                                struct dlm_lock *lock, int flags)
 286 {
 287         struct dlm_create_lock create;
 288         int tmpret, status = 0;
 289         enum dlm_status ret;
 290
 291         memset(&create, 0, sizeof(create));
 292         create.node_idx = dlm->node_num;
 293         create.requested_type = lock->ml.type;
 294         create.cookie = lock->ml.cookie;
 295         create.namelen = res->lockname.len;
 296         create.flags = cpu_to_be32(flags);
 297         memcpy(create.name, res->lockname.name, create.namelen);
 298
 299         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
 300                                     sizeof(create), res->owner, &status);
 301         if (tmpret >= 0) {
 302                 ret = status;
 303                 if (ret == DLM_REJECTED) {
 304                         mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
 305                              "owned by node %u. That node is coming back up "
 306                              "currently.\n", dlm->name, create.namelen,
 307                              create.name, res->owner);
 308                         dlm_print_one_lock_resource(res);
 309                         BUG();
 310                 }
 311         } else {
 312                 mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
 313                      "node %u\n", dlm->name, create.namelen, create.name,
 314                      tmpret, res->owner);
 315                 if (dlm_is_host_down(tmpret))
 316                         ret = DLM_RECOVERING;
 317                 else
 318                         ret = dlm_err_to_dlm_status(tmpret);
 319         }
 320
 321         return ret;
 322 }
 323
 324 void dlm_lock_get(struct dlm_lock *lock)
 325 {
 326         kref_get(&lock->lock_refs);
 327 }
 328
 329 void dlm_lock_put(struct dlm_lock *lock)
 330 {
 331         kref_put(&lock->lock_refs, dlm_lock_release);
 332 }
 333
 334 static void dlm_lock_release(struct kref *kref)
 335 {
 336         struct dlm_lock *lock;
 337
 338         lock = container_of(kref, struct dlm_lock, lock_refs);
 339
 340         BUG_ON(!list_empty(&lock->list));
 341         BUG_ON(!list_empty(&lock->ast_list));
 342         BUG_ON(!list_empty(&lock->bast_list));
 343         BUG_ON(lock->ast_pending);
 344         BUG_ON(lock->bast_pending);
 345
 346         dlm_lock_detach_lockres(lock);
 347
 348         if (lock->lksb_kernel_allocated) {
 349                 mlog(0, "freeing kernel-allocated lksb\n");
 350                 kfree(lock->lksb);
 351         }
 352         kmem_cache_free(dlm_lock_cache, lock);
 353 }
 354
 355 /* associate a lock with it's lockres, getting a ref on the lockres */
 356 void dlm_lock_attach_lockres(struct dlm_lock *lock,
 357                              struct dlm_lock_resource *res)
 358 {
 359         dlm_lockres_get(res);
 360         lock->lockres = res;
 361 }
 362
 363 /* drop ref on lockres, if there is still one associated with lock */
 364 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
 365 {
 366         struct dlm_lock_resource *res;
 367
 368         res = lock->lockres;
 369         if (res) {
 370                 lock->lockres = NULL;
 371                 mlog(0, "removing lock's lockres reference\n");
 372                 dlm_lockres_put(res);
 373         }
 374 }
 375
 376 static void dlm_init_lock(struct dlm_lock *newlock, int type,
 377                           u8 node, u64 cookie)
 378 {
 379         INIT_LIST_HEAD(&newlock->list);
 380         INIT_LIST_HEAD(&newlock->ast_list);
 381         INIT_LIST_HEAD(&newlock->bast_list);
 382         spin_lock_init(&newlock->spinlock);
 383         newlock->ml.type = type;
 384         newlock->ml.convert_type = LKM_IVMODE;
 385         newlock->ml.highest_blocked = LKM_IVMODE;
 386         newlock->ml.node = node;
 387         newlock->ml.pad1 = 0;
 388         newlock->ml.list = 0;
 389         newlock->ml.flags = 0;
 390         newlock->ast = NULL;
 391         newlock->bast = NULL;
 392         newlock->astdata = NULL;
 393         newlock->ml.cookie = cpu_to_be64(cookie);
 394         newlock->ast_pending = 0;
 395         newlock->bast_pending = 0;
 396         newlock->convert_pending = 0;
 397         newlock->lock_pending = 0;
 398         newlock->unlock_pending = 0;
 399         newlock->cancel_pending = 0;
 400         newlock->lksb_kernel_allocated = 0;
 401
 402         kref_init(&newlock->lock_refs);
 403 }
 404
 405 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 406                                struct dlm_lockstatus *lksb)
 407 {
 408         struct dlm_lock *lock;
 409         int kernel_allocated = 0;
 410
 411         lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
 412         if (!lock)
 413                 return NULL;
 414
 415         if (!lksb) {
 416                 /* zero memory only if kernel-allocated */
 417                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 418                 if (!lksb) {
 419                         kmem_cache_free(dlm_lock_cache, lock);
 420                         return NULL;
 421                 }
 422                 kernel_allocated = 1;
 423         }
 424
 425         dlm_init_lock(lock, type, node, cookie);
 426         if (kernel_allocated)
 427                 lock->lksb_kernel_allocated = 1;
 428         lock->lksb = lksb;
 429         lksb->lockid = lock;
 430         return lock;
 431 }
 432
 433 /* handler for lock creation net message
 434  * locking:
 435  *   caller needs:  none
 436  *   taken:         takes and drops res->spinlock
 437  *   held on exit:  none
 438  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
 439  */
 440 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 441                             void **ret_data)
 442 {
 443         struct dlm_ctxt *dlm = data;
 444         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
 445         struct dlm_lock_resource *res = NULL;
 446         struct dlm_lock *newlock = NULL;
 447         struct dlm_lockstatus *lksb = NULL;
 448         enum dlm_status status = DLM_NORMAL;
 449         char *name;
 450         unsigned int namelen;
 451
 452         BUG_ON(!dlm);
 453
 454         if (!dlm_grab(dlm))
 455                 return DLM_REJECTED;
 456
 457         name = create->name;
 458         namelen = create->namelen;
 459         status = DLM_REJECTED;
 460         if (!dlm_domain_fully_joined(dlm)) {
 461                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
 462                      "sending a create_lock message for lock %.*s!\n",
 463                      dlm->name, create->node_idx, namelen, name);
 464                 dlm_error(status);
 465                 goto leave;
 466         }
 467
 468         status = DLM_IVBUFLEN;
 469         if (namelen > DLM_LOCKID_NAME_MAX) {
 470                 dlm_error(status);
 471                 goto leave;
 472         }
 473
 474         status = DLM_SYSERR;
 475         newlock = dlm_new_lock(create->requested_type,
 476                                create->node_idx,
 477                                be64_to_cpu(create->cookie), NULL);
 478         if (!newlock) {
 479                 dlm_error(status);
 480                 goto leave;
 481         }
 482
 483         lksb = newlock->lksb;
 484
 485         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
 486                 lksb->flags |= DLM_LKSB_GET_LVB;
 487                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
 488         }
 489
 490         status = DLM_IVLOCKID;
 491         res = dlm_lookup_lockres(dlm, name, namelen);
 492         if (!res) {
 493                 dlm_error(status);
 494                 goto leave;
 495         }
 496
 497         spin_lock(&res->spinlock);
 498         status = __dlm_lockres_state_to_status(res);
 499         spin_unlock(&res->spinlock);
 500
 501         if (status != DLM_NORMAL) {
 502                 mlog(0, "lockres recovering/migrating/in-progress\n");
 503                 goto leave;
 504         }
 505
 506         dlm_lock_attach_lockres(newlock, res);
 507
 508         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
 509 leave:
 510         if (status != DLM_NORMAL)
 511                 if (newlock)
 512                         dlm_lock_put(newlock);
 513
 514         if (res)
 515                 dlm_lockres_put(res);
 516
 517         dlm_put(dlm);
 518
 519         return status;
 520 }
 521
 522
 523 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
 524 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 525 {
 526         u64 tmpnode = node_num;
 527
 528         /* shift single byte of node num into top 8 bits */
 529         tmpnode <<= 56;
 530
 531         spin_lock(&dlm_cookie_lock);
 532         *cookie = (dlm_next_cookie | tmpnode);
 533         if (++dlm_next_cookie & 0xff00000000000000ull) {
 534                 mlog(0, "This node's cookie will now wrap!\n");
 535                 dlm_next_cookie = 1;
 536         }
 537         spin_unlock(&dlm_cookie_lock);
 538 }
 539
 540 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 541                         struct dlm_lockstatus *lksb, int flags,
 542                         const char *name, int namelen, dlm_astlockfunc_t *ast,
 543                         void *data, dlm_bastlockfunc_t *bast)
 544 {
 545         enum dlm_status status;
 546         struct dlm_lock_resource *res = NULL;
 547         struct dlm_lock *lock = NULL;
 548         int convert = 0, recovery = 0;
 549
 550         /* yes this function is a mess.
 551          * TODO: clean this up.  lots of common code in the
 552          *       lock and convert paths, especially in the retry blocks */
 553         if (!lksb) {
 554                 dlm_error(DLM_BADARGS);
 555                 return DLM_BADARGS;
 556         }
 557
 558         status = DLM_BADPARAM;
 559         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
 560                 dlm_error(status);
 561                 goto error;
 562         }
 563
 564         if (flags & ~LKM_VALID_FLAGS) {
 565                 dlm_error(status);
 566                 goto error;
 567         }
 568
 569         convert = (flags & LKM_CONVERT);
 570         recovery = (flags & LKM_RECOVERY);
 571
 572         if (recovery &&
 573             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 574                 dlm_error(status);
 575                 goto error;
 576         }
 577         if (convert && (flags & LKM_LOCAL)) {
 578                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
 579                 goto error;
 580         }
 581
 582         if (convert) {
 583                 /* CONVERT request */
 584
 585                 /* if converting, must pass in a valid dlm_lock */
 586                 lock = lksb->lockid;
 587                 if (!lock) {
 588                         mlog(ML_ERROR, "NULL lock pointer in convert "
 589                              "request\n");
 590                         goto error;
 591                 }
 592
 593                 res = lock->lockres;
 594                 if (!res) {
 595                         mlog(ML_ERROR, "NULL lockres pointer in convert "
 596                              "request\n");
 597                         goto error;
 598                 }
 599                 dlm_lockres_get(res);
 600
 601                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
 602                  * static after the original lock call.  convert requests will
 603                  * ensure that everything is the same, or return DLM_BADARGS.
 604                  * this means that DLM_DENIED_NOASTS will never be returned.
 605                  */
 606                 if (lock->lksb != lksb || lock->ast != ast ||
 607                     lock->bast != bast || lock->astdata != data) {
 608                         status = DLM_BADARGS;
 609                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
 610                              "astdata=%p\n", lksb, ast, bast, data);
 611                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
 612                              "astdata=%p\n", lock->lksb, lock->ast,
 613                              lock->bast, lock->astdata);
 614                         goto error;
 615                 }
 616 retry_convert:
 617                 dlm_wait_for_recovery(dlm);
 618
 619                 if (res->owner == dlm->node_num)
 620                         status = dlmconvert_master(dlm, res, lock, flags, mode);
 621                 else
 622                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
 623                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 624                     status == DLM_FORWARD) {
 625                         /* for now, see how this works without sleeping
 626                          * and just retry right away.  I suspect the reco
 627                          * or migration will complete fast enough that
 628                          * no waiting will be necessary */
 629                         mlog(0, "retrying convert with migration/recovery/"
 630                              "in-progress\n");
 631                         msleep(100);
 632                         goto retry_convert;
 633                 }
 634         } else {
 635                 u64 tmpcookie;
 636
 637                 /* LOCK request */
 638                 status = DLM_BADARGS;
 639                 if (!name) {
 640                         dlm_error(status);
 641                         goto error;
 642                 }
 643
 644                 status = DLM_IVBUFLEN;
 645                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 646                         dlm_error(status);
 647                         goto error;
 648                 }
 649
 650                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
 651                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
 652                 if (!lock) {
 653                         dlm_error(status);
 654                         goto error;
 655                 }
 656
 657                 if (!recovery)
 658                         dlm_wait_for_recovery(dlm);
 659
 660                 /* find or create the lock resource */
 661                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
 662                 if (!res) {
 663                         status = DLM_IVLOCKID;
 664                         dlm_error(status);
 665                         goto error;
 666                 }
 667
 668                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
 669                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
 670
 671                 dlm_lock_attach_lockres(lock, res);
 672                 lock->ast = ast;
 673                 lock->bast = bast;
 674                 lock->astdata = data;
 675
 676 retry_lock:
 677                 if (flags & LKM_VALBLK) {
 678                         mlog(0, "LKM_VALBLK passed by caller\n");
 679
 680                         /* LVB requests for non PR, PW or EX locks are
 681                          * ignored. */
 682                         if (mode < LKM_PRMODE)
 683                                 flags &= ~LKM_VALBLK;
 684                         else {
 685                                 flags |= LKM_GET_LVB;
 686                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
 687                         }
 688                 }
 689
 690                 if (res->owner == dlm->node_num)
 691                         status = dlmlock_master(dlm, res, lock, flags);
 692                 else
 693                         status = dlmlock_remote(dlm, res, lock, flags);
 694
 695                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 696                     status == DLM_FORWARD) {
 697                         msleep(100);
 698                         if (recovery) {
 699                                 if (status != DLM_RECOVERING)
 700                                         goto retry_lock;
 701                                 /* wait to see the node go down, then
 702                                  * drop down and allow the lockres to
 703                                  * get cleaned up.  need to remaster. */
 704                                 dlm_wait_for_node_death(dlm, res->owner,
 705                                                 DLM_NODE_DEATH_WAIT_MAX);
 706                         } else {
 707                                 dlm_wait_for_recovery(dlm);
 708                                 goto retry_lock;
 709                         }
 710                 }
 711
 712                 /* Inflight taken in dlm_get_lock_resource() is dropped here */
 713                 spin_lock(&res->spinlock);
 714                 dlm_lockres_drop_inflight_ref(dlm, res);
 715                 spin_unlock(&res->spinlock);
 716
 717                 dlm_lockres_calc_usage(dlm, res);
 718                 dlm_kick_thread(dlm, res);
 719
 720                 if (status != DLM_NORMAL) {
 721                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 722                         if (status != DLM_NOTQUEUED)
 723                                 dlm_error(status);
 724                         goto error;
 725                 }
 726         }
 727
 728 error:
 729         if (status != DLM_NORMAL) {
 730                 if (lock && !convert)
 731                         dlm_lock_put(lock);
 732                 // this is kind of unnecessary
 733                 lksb->status = status;
 734         }
 735
 736         /* put lockres ref from the convert path
 737          * or from dlm_get_lock_resource */
 738         if (res)
 739                 dlm_lockres_put(res);
 740
 741         return status;
 742 }
 743 EXPORT_SYMBOL_GPL(dlmlock);