git.oblomov.eu Git - linux-2.6/blob - fs/ocfs2/dlm/dlmlock.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmlock.c
   5  *
   6  * underlying calls for lock creation
   7  *
   8  * Copyright (C) 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  *
  25  */
  26
  27
  28 #include <linux/module.h>
  29 #include <linux/fs.h>
  30 #include <linux/types.h>
  31 #include <linux/slab.h>
  32 #include <linux/highmem.h>
  33 #include <linux/utsname.h>
  34 #include <linux/init.h>
  35 #include <linux/sysctl.h>
  36 #include <linux/random.h>
  37 #include <linux/blkdev.h>
  38 #include <linux/socket.h>
  39 #include <linux/inet.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/delay.h>
  42
  43
  44 #include "cluster/heartbeat.h"
  45 #include "cluster/nodemanager.h"
  46 #include "cluster/tcp.h"
  47
  48 #include "dlmapi.h"
  49 #include "dlmcommon.h"
  50
  51 #include "dlmconvert.h"
  52
  53 #define MLOG_MASK_PREFIX ML_DLM
  54 #include "cluster/masklog.h"
  55
  56 static DEFINE_SPINLOCK(dlm_cookie_lock);
  57 static u64 dlm_next_cookie = 1;
  58
  59 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
  60                                                struct dlm_lock_resource *res,
  61                                                struct dlm_lock *lock, int flags);
  62 static void dlm_init_lock(struct dlm_lock *newlock, int type,
  63                           u8 node, u64 cookie);
  64 static void dlm_lock_release(struct kref *kref);
  65 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
  66
  67 /* Tell us whether we can grant a new lock request.
  68  * locking:
  69  *   caller needs:  res->spinlock
  70  *   taken:         none
  71  *   held on exit:  none
  72  * returns: 1 if the lock can be granted, 0 otherwise.
  73  */
  74 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
  75                                   struct dlm_lock *lock)
  76 {
  77         struct list_head *iter;
  78         struct dlm_lock *tmplock;
  79
  80         list_for_each(iter, &res->granted) {
  81                 tmplock = list_entry(iter, struct dlm_lock, list);
  82
  83                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
  84                         return 0;
  85         }
  86
  87         list_for_each(iter, &res->converting) {
  88                 tmplock = list_entry(iter, struct dlm_lock, list);
  89
  90                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
  91                         return 0;
  92         }
  93
  94         return 1;
  95 }
  96
  97 /* performs lock creation at the lockres master site
  98  * locking:
  99  *   caller needs:  none
 100  *   taken:         takes and drops res->spinlock
 101  *   held on exit:  none
 102  * returns: DLM_NORMAL, DLM_NOTQUEUED
 103  */
 104 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
 105                                       struct dlm_lock_resource *res,
 106                                       struct dlm_lock *lock, int flags)
 107 {
 108         int call_ast = 0, kick_thread = 0;
 109         enum dlm_status status = DLM_NORMAL;
 110
 111         mlog_entry("type=%d\n", lock->ml.type);
 112
 113         spin_lock(&res->spinlock);
 114         /* if called from dlm_create_lock_handler, need to
 115          * ensure it will not sleep in dlm_wait_on_lockres */
 116         status = __dlm_lockres_state_to_status(res);
 117         if (status != DLM_NORMAL &&
 118             lock->ml.node != dlm->node_num) {
 119                 /* erf.  state changed after lock was dropped. */
 120                 spin_unlock(&res->spinlock);
 121                 dlm_error(status);
 122                 return status;
 123         }
 124         __dlm_wait_on_lockres(res);
 125         __dlm_lockres_reserve_ast(res);
 126
 127         if (dlm_can_grant_new_lock(res, lock)) {
 128                 mlog(0, "I can grant this lock right away\n");
 129                 /* got it right away */
 130                 lock->lksb->status = DLM_NORMAL;
 131                 status = DLM_NORMAL;
 132                 dlm_lock_get(lock);
 133                 list_add_tail(&lock->list, &res->granted);
 134
 135                 /* for the recovery lock, we can't allow the ast
 136                  * to be queued since the dlmthread is already
 137                  * frozen.  but the recovery lock is always locked
 138                  * with LKM_NOQUEUE so we do not need the ast in
 139                  * this special case */
 140                 if (!dlm_is_recovery_lock(res->lockname.name,
 141                                           res->lockname.len)) {
 142                         kick_thread = 1;
 143                         call_ast = 1;
 144                 } else {
 145                         mlog(0, "%s: returning DLM_NORMAL to "
 146                              "node %u for reco lock\n", dlm->name,
 147                              lock->ml.node);
 148                 }
 149         } else {
 150                 /* for NOQUEUE request, unless we get the
 151                  * lock right away, return DLM_NOTQUEUED */
 152                 if (flags & LKM_NOQUEUE) {
 153                         status = DLM_NOTQUEUED;
 154                         if (dlm_is_recovery_lock(res->lockname.name,
 155                                                  res->lockname.len)) {
 156                                 mlog(0, "%s: returning NOTQUEUED to "
 157                                      "node %u for reco lock\n", dlm->name,
 158                                      lock->ml.node);
 159                         }
 160                 } else {
 161                         dlm_lock_get(lock);
 162                         list_add_tail(&lock->list, &res->blocked);
 163                         kick_thread = 1;
 164                 }
 165         }
 166         /* reduce the inflight count, this may result in the lockres
 167          * being purged below during calc_usage */
 168         if (lock->ml.node == dlm->node_num)
 169                 dlm_lockres_drop_inflight_ref(dlm, res);
 170
 171         spin_unlock(&res->spinlock);
 172         wake_up(&res->wq);
 173
 174         /* either queue the ast or release it */
 175         if (call_ast)
 176                 dlm_queue_ast(dlm, lock);
 177         else
 178                 dlm_lockres_release_ast(dlm, res);
 179
 180         dlm_lockres_calc_usage(dlm, res);
 181         if (kick_thread)
 182                 dlm_kick_thread(dlm, res);
 183
 184         return status;
 185 }
 186
 187 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
 188                              struct dlm_lock *lock)
 189 {
 190         /* remove from local queue if it failed */
 191         list_del_init(&lock->list);
 192         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 193 }
 194
 195
 196 /*
 197  * locking:
 198  *   caller needs:  none
 199  *   taken:         takes and drops res->spinlock
 200  *   held on exit:  none
 201  * returns: DLM_DENIED, DLM_RECOVERING, or net status
 202  */
 203 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 204                                       struct dlm_lock_resource *res,
 205                                       struct dlm_lock *lock, int flags)
 206 {
 207         enum dlm_status status = DLM_DENIED;
 208         int lockres_changed = 1;
 209
 210         mlog_entry("type=%d\n", lock->ml.type);
 211         mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
 212              res->lockname.name, flags);
 213
 214         spin_lock(&res->spinlock);
 215
 216         /* will exit this call with spinlock held */
 217         __dlm_wait_on_lockres(res);
 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         mlog_entry_void();
 292
 293         memset(&create, 0, sizeof(create));
 294         create.node_idx = dlm->node_num;
 295         create.requested_type = lock->ml.type;
 296         create.cookie = lock->ml.cookie;
 297         create.namelen = res->lockname.len;
 298         create.flags = cpu_to_be32(flags);
 299         memcpy(create.name, res->lockname.name, create.namelen);
 300
 301         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
 302                                     sizeof(create), res->owner, &status);
 303         if (tmpret >= 0) {
 304                 // successfully sent and received
 305                 ret = status;  // this is already a dlm_status
 306                 if (ret == DLM_REJECTED) {
 307                         mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
 308                              "no longer owned by %u.  that node is coming back "
 309                              "up currently.\n", dlm->name, create.namelen,
 310                              create.name, res->owner);
 311                         dlm_print_one_lock_resource(res);
 312                         BUG();
 313                 }
 314         } else {
 315                 mlog_errno(tmpret);
 316                 if (dlm_is_host_down(tmpret)) {
 317                         ret = DLM_RECOVERING;
 318                         mlog(0, "node %u died so returning DLM_RECOVERING "
 319                              "from lock message!\n", res->owner);
 320                 } else {
 321                         ret = dlm_err_to_dlm_status(tmpret);
 322                 }
 323         }
 324
 325         return ret;
 326 }
 327
 328 void dlm_lock_get(struct dlm_lock *lock)
 329 {
 330         kref_get(&lock->lock_refs);
 331 }
 332
 333 void dlm_lock_put(struct dlm_lock *lock)
 334 {
 335         kref_put(&lock->lock_refs, dlm_lock_release);
 336 }
 337
 338 static void dlm_lock_release(struct kref *kref)
 339 {
 340         struct dlm_lock *lock;
 341
 342         lock = container_of(kref, struct dlm_lock, lock_refs);
 343
 344         BUG_ON(!list_empty(&lock->list));
 345         BUG_ON(!list_empty(&lock->ast_list));
 346         BUG_ON(!list_empty(&lock->bast_list));
 347         BUG_ON(lock->ast_pending);
 348         BUG_ON(lock->bast_pending);
 349
 350         dlm_lock_detach_lockres(lock);
 351
 352         if (lock->lksb_kernel_allocated) {
 353                 mlog(0, "freeing kernel-allocated lksb\n");
 354                 kfree(lock->lksb);
 355         }
 356         kfree(lock);
 357 }
 358
 359 /* associate a lock with it's lockres, getting a ref on the lockres */
 360 void dlm_lock_attach_lockres(struct dlm_lock *lock,
 361                              struct dlm_lock_resource *res)
 362 {
 363         dlm_lockres_get(res);
 364         lock->lockres = res;
 365 }
 366
 367 /* drop ref on lockres, if there is still one associated with lock */
 368 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
 369 {
 370         struct dlm_lock_resource *res;
 371
 372         res = lock->lockres;
 373         if (res) {
 374                 lock->lockres = NULL;
 375                 mlog(0, "removing lock's lockres reference\n");
 376                 dlm_lockres_put(res);
 377         }
 378 }
 379
 380 static void dlm_init_lock(struct dlm_lock *newlock, int type,
 381                           u8 node, u64 cookie)
 382 {
 383         INIT_LIST_HEAD(&newlock->list);
 384         INIT_LIST_HEAD(&newlock->ast_list);
 385         INIT_LIST_HEAD(&newlock->bast_list);
 386         spin_lock_init(&newlock->spinlock);
 387         newlock->ml.type = type;
 388         newlock->ml.convert_type = LKM_IVMODE;
 389         newlock->ml.highest_blocked = LKM_IVMODE;
 390         newlock->ml.node = node;
 391         newlock->ml.pad1 = 0;
 392         newlock->ml.list = 0;
 393         newlock->ml.flags = 0;
 394         newlock->ast = NULL;
 395         newlock->bast = NULL;
 396         newlock->astdata = NULL;
 397         newlock->ml.cookie = cpu_to_be64(cookie);
 398         newlock->ast_pending = 0;
 399         newlock->bast_pending = 0;
 400         newlock->convert_pending = 0;
 401         newlock->lock_pending = 0;
 402         newlock->unlock_pending = 0;
 403         newlock->cancel_pending = 0;
 404         newlock->lksb_kernel_allocated = 0;
 405
 406         kref_init(&newlock->lock_refs);
 407 }
 408
 409 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 410                                struct dlm_lockstatus *lksb)
 411 {
 412         struct dlm_lock *lock;
 413         int kernel_allocated = 0;
 414
 415         lock = kzalloc(sizeof(*lock), GFP_NOFS);
 416         if (!lock)
 417                 return NULL;
 418
 419         if (!lksb) {
 420                 /* zero memory only if kernel-allocated */
 421                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 422                 if (!lksb) {
 423                         kfree(lock);
 424                         return NULL;
 425                 }
 426                 kernel_allocated = 1;
 427         }
 428
 429         dlm_init_lock(lock, type, node, cookie);
 430         if (kernel_allocated)
 431                 lock->lksb_kernel_allocated = 1;
 432         lock->lksb = lksb;
 433         lksb->lockid = lock;
 434         return lock;
 435 }
 436
 437 /* handler for lock creation net message
 438  * locking:
 439  *   caller needs:  none
 440  *   taken:         takes and drops res->spinlock
 441  *   held on exit:  none
 442  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
 443  */
 444 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 445                             void **ret_data)
 446 {
 447         struct dlm_ctxt *dlm = data;
 448         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
 449         struct dlm_lock_resource *res = NULL;
 450         struct dlm_lock *newlock = NULL;
 451         struct dlm_lockstatus *lksb = NULL;
 452         enum dlm_status status = DLM_NORMAL;
 453         char *name;
 454         unsigned int namelen;
 455
 456         BUG_ON(!dlm);
 457
 458         mlog_entry_void();
 459
 460         if (!dlm_grab(dlm))
 461                 return DLM_REJECTED;
 462
 463         name = create->name;
 464         namelen = create->namelen;
 465         status = DLM_REJECTED;
 466         if (!dlm_domain_fully_joined(dlm)) {
 467                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
 468                      "sending a create_lock message for lock %.*s!\n",
 469                      dlm->name, create->node_idx, namelen, name);
 470                 dlm_error(status);
 471                 goto leave;
 472         }
 473
 474         status = DLM_IVBUFLEN;
 475         if (namelen > DLM_LOCKID_NAME_MAX) {
 476                 dlm_error(status);
 477                 goto leave;
 478         }
 479
 480         status = DLM_SYSERR;
 481         newlock = dlm_new_lock(create->requested_type,
 482                                create->node_idx,
 483                                be64_to_cpu(create->cookie), NULL);
 484         if (!newlock) {
 485                 dlm_error(status);
 486                 goto leave;
 487         }
 488
 489         lksb = newlock->lksb;
 490
 491         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
 492                 lksb->flags |= DLM_LKSB_GET_LVB;
 493                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
 494         }
 495
 496         status = DLM_IVLOCKID;
 497         res = dlm_lookup_lockres(dlm, name, namelen);
 498         if (!res) {
 499                 dlm_error(status);
 500                 goto leave;
 501         }
 502
 503         spin_lock(&res->spinlock);
 504         status = __dlm_lockres_state_to_status(res);
 505         spin_unlock(&res->spinlock);
 506
 507         if (status != DLM_NORMAL) {
 508                 mlog(0, "lockres recovering/migrating/in-progress\n");
 509                 goto leave;
 510         }
 511
 512         dlm_lock_attach_lockres(newlock, res);
 513
 514         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
 515 leave:
 516         if (status != DLM_NORMAL)
 517                 if (newlock)
 518                         dlm_lock_put(newlock);
 519
 520         if (res)
 521                 dlm_lockres_put(res);
 522
 523         dlm_put(dlm);
 524
 525         return status;
 526 }
 527
 528
 529 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
 530 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 531 {
 532         u64 tmpnode = node_num;
 533
 534         /* shift single byte of node num into top 8 bits */
 535         tmpnode <<= 56;
 536
 537         spin_lock(&dlm_cookie_lock);
 538         *cookie = (dlm_next_cookie | tmpnode);
 539         if (++dlm_next_cookie & 0xff00000000000000ull) {
 540                 mlog(0, "This node's cookie will now wrap!\n");
 541                 dlm_next_cookie = 1;
 542         }
 543         spin_unlock(&dlm_cookie_lock);
 544 }
 545
 546 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 547                         struct dlm_lockstatus *lksb, int flags,
 548                         const char *name, int namelen, dlm_astlockfunc_t *ast,
 549                         void *data, dlm_bastlockfunc_t *bast)
 550 {
 551         enum dlm_status status;
 552         struct dlm_lock_resource *res = NULL;
 553         struct dlm_lock *lock = NULL;
 554         int convert = 0, recovery = 0;
 555
 556         /* yes this function is a mess.
 557          * TODO: clean this up.  lots of common code in the
 558          *       lock and convert paths, especially in the retry blocks */
 559         if (!lksb) {
 560                 dlm_error(DLM_BADARGS);
 561                 return DLM_BADARGS;
 562         }
 563
 564         status = DLM_BADPARAM;
 565         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
 566                 dlm_error(status);
 567                 goto error;
 568         }
 569
 570         if (flags & ~LKM_VALID_FLAGS) {
 571                 dlm_error(status);
 572                 goto error;
 573         }
 574
 575         convert = (flags & LKM_CONVERT);
 576         recovery = (flags & LKM_RECOVERY);
 577
 578         if (recovery &&
 579             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 580                 dlm_error(status);
 581                 goto error;
 582         }
 583         if (convert && (flags & LKM_LOCAL)) {
 584                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
 585                 goto error;
 586         }
 587
 588         if (convert) {
 589                 /* CONVERT request */
 590
 591                 /* if converting, must pass in a valid dlm_lock */
 592                 lock = lksb->lockid;
 593                 if (!lock) {
 594                         mlog(ML_ERROR, "NULL lock pointer in convert "
 595                              "request\n");
 596                         goto error;
 597                 }
 598
 599                 res = lock->lockres;
 600                 if (!res) {
 601                         mlog(ML_ERROR, "NULL lockres pointer in convert "
 602                              "request\n");
 603                         goto error;
 604                 }
 605                 dlm_lockres_get(res);
 606
 607                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
 608                  * static after the original lock call.  convert requests will
 609                  * ensure that everything is the same, or return DLM_BADARGS.
 610                  * this means that DLM_DENIED_NOASTS will never be returned.
 611                  */
 612                 if (lock->lksb != lksb || lock->ast != ast ||
 613                     lock->bast != bast || lock->astdata != data) {
 614                         status = DLM_BADARGS;
 615                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
 616                              "astdata=%p\n", lksb, ast, bast, data);
 617                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
 618                              "astdata=%p\n", lock->lksb, lock->ast,
 619                              lock->bast, lock->astdata);
 620                         goto error;
 621                 }
 622 retry_convert:
 623                 dlm_wait_for_recovery(dlm);
 624
 625                 if (res->owner == dlm->node_num)
 626                         status = dlmconvert_master(dlm, res, lock, flags, mode);
 627                 else
 628                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
 629                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 630                     status == DLM_FORWARD) {
 631                         /* for now, see how this works without sleeping
 632                          * and just retry right away.  I suspect the reco
 633                          * or migration will complete fast enough that
 634                          * no waiting will be necessary */
 635                         mlog(0, "retrying convert with migration/recovery/"
 636                              "in-progress\n");
 637                         msleep(100);
 638                         goto retry_convert;
 639                 }
 640         } else {
 641                 u64 tmpcookie;
 642
 643                 /* LOCK request */
 644                 status = DLM_BADARGS;
 645                 if (!name) {
 646                         dlm_error(status);
 647                         goto error;
 648                 }
 649
 650                 status = DLM_IVBUFLEN;
 651                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 652                         dlm_error(status);
 653                         goto error;
 654                 }
 655
 656                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
 657                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
 658                 if (!lock) {
 659                         dlm_error(status);
 660                         goto error;
 661                 }
 662
 663                 if (!recovery)
 664                         dlm_wait_for_recovery(dlm);
 665
 666                 /* find or create the lock resource */
 667                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
 668                 if (!res) {
 669                         status = DLM_IVLOCKID;
 670                         dlm_error(status);
 671                         goto error;
 672                 }
 673
 674                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
 675                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
 676
 677                 dlm_lock_attach_lockres(lock, res);
 678                 lock->ast = ast;
 679                 lock->bast = bast;
 680                 lock->astdata = data;
 681
 682 retry_lock:
 683                 if (flags & LKM_VALBLK) {
 684                         mlog(0, "LKM_VALBLK passed by caller\n");
 685
 686                         /* LVB requests for non PR, PW or EX locks are
 687                          * ignored. */
 688                         if (mode < LKM_PRMODE)
 689                                 flags &= ~LKM_VALBLK;
 690                         else {
 691                                 flags |= LKM_GET_LVB;
 692                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
 693                         }
 694                 }
 695
 696                 if (res->owner == dlm->node_num)
 697                         status = dlmlock_master(dlm, res, lock, flags);
 698                 else
 699                         status = dlmlock_remote(dlm, res, lock, flags);
 700
 701                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 702                     status == DLM_FORWARD) {
 703                         mlog(0, "retrying lock with migration/"
 704                              "recovery/in progress\n");
 705                         msleep(100);
 706                         /* no waiting for dlm_reco_thread */
 707                         if (recovery) {
 708                                 if (status != DLM_RECOVERING)
 709                                         goto retry_lock;
 710
 711                                 mlog(0, "%s: got RECOVERING "
 712                                      "for $RECOVERY lock, master "
 713                                      "was %u\n", dlm->name,
 714                                      res->owner);
 715                                 /* wait to see the node go down, then
 716                                  * drop down and allow the lockres to
 717                                  * get cleaned up.  need to remaster. */
 718                                 dlm_wait_for_node_death(dlm, res->owner,
 719                                                 DLM_NODE_DEATH_WAIT_MAX);
 720                         } else {
 721                                 dlm_wait_for_recovery(dlm);
 722                                 goto retry_lock;
 723                         }
 724                 }
 725
 726                 if (status != DLM_NORMAL) {
 727                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 728                         if (status != DLM_NOTQUEUED)
 729                                 dlm_error(status);
 730                         goto error;
 731                 }
 732         }
 733
 734 error:
 735         if (status != DLM_NORMAL) {
 736                 if (lock && !convert)
 737                         dlm_lock_put(lock);
 738                 // this is kind of unnecessary
 739                 lksb->status = status;
 740         }
 741
 742         /* put lockres ref from the convert path
 743          * or from dlm_get_lock_resource */
 744         if (res)
 745                 dlm_lockres_put(res);
 746
 747         return status;
 748 }
 749 EXPORT_SYMBOL_GPL(dlmlock);