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/init.h>
  34 #include <linux/sysctl.h>
  35 #include <linux/random.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/socket.h>
  38 #include <linux/inet.h>
  39 #include <linux/spinlock.h>
  40 #include <linux/delay.h>
  41
  42
  43 #include "cluster/heartbeat.h"
  44 #include "cluster/nodemanager.h"
  45 #include "cluster/tcp.h"
  46
  47 #include "dlmapi.h"
  48 #include "dlmcommon.h"
  49
  50 #include "dlmconvert.h"
  51
  52 #define MLOG_MASK_PREFIX ML_DLM
  53 #include "cluster/masklog.h"
  54
  55 static struct kmem_cache *dlm_lock_cache = NULL;
  56
  57 static DEFINE_SPINLOCK(dlm_cookie_lock);
  58 static u64 dlm_next_cookie = 1;
  59
  60 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
  61                                                struct dlm_lock_resource *res,
  62                                                struct dlm_lock *lock, int flags);
  63 static void dlm_init_lock(struct dlm_lock *newlock, int type,
  64                           u8 node, u64 cookie);
  65 static void dlm_lock_release(struct kref *kref);
  66 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
  67
  68 int dlm_init_lock_cache(void)
  69 {
  70         dlm_lock_cache = kmem_cache_create("o2dlm_lock",
  71                                            sizeof(struct dlm_lock),
  72                                            0, SLAB_HWCACHE_ALIGN, NULL);
  73         if (dlm_lock_cache == NULL)
  74                 return -ENOMEM;
  75         return 0;
  76 }
  77
  78 void dlm_destroy_lock_cache(void)
  79 {
  80         if (dlm_lock_cache)
  81                 kmem_cache_destroy(dlm_lock_cache);
  82 }
  83
  84 /* Tell us whether we can grant a new lock request.
  85  * locking:
  86  *   caller needs:  res->spinlock
  87  *   taken:         none
  88  *   held on exit:  none
  89  * returns: 1 if the lock can be granted, 0 otherwise.
  90  */
  91 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
  92                                   struct dlm_lock *lock)
  93 {
  94         struct list_head *iter;
  95         struct dlm_lock *tmplock;
  96
  97         list_for_each(iter, &res->granted) {
  98                 tmplock = list_entry(iter, struct dlm_lock, list);
  99
 100                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
 101                         return 0;
 102         }
 103
 104         list_for_each(iter, &res->converting) {
 105                 tmplock = list_entry(iter, struct dlm_lock, list);
 106
 107                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
 108                         return 0;
 109                 if (!dlm_lock_compatible(tmplock->ml.convert_type,
 110                                          lock->ml.type))
 111                         return 0;
 112         }
 113
 114         return 1;
 115 }
 116
 117 /* performs lock creation at the lockres master site
 118  * locking:
 119  *   caller needs:  none
 120  *   taken:         takes and drops res->spinlock
 121  *   held on exit:  none
 122  * returns: DLM_NORMAL, DLM_NOTQUEUED
 123  */
 124 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
 125                                       struct dlm_lock_resource *res,
 126                                       struct dlm_lock *lock, int flags)
 127 {
 128         int call_ast = 0, kick_thread = 0;
 129         enum dlm_status status = DLM_NORMAL;
 130
 131         mlog(0, "type=%d\n", lock->ml.type);
 132
 133         spin_lock(&res->spinlock);
 134         /* if called from dlm_create_lock_handler, need to
 135          * ensure it will not sleep in dlm_wait_on_lockres */
 136         status = __dlm_lockres_state_to_status(res);
 137         if (status != DLM_NORMAL &&
 138             lock->ml.node != dlm->node_num) {
 139                 /* erf.  state changed after lock was dropped. */
 140                 spin_unlock(&res->spinlock);
 141                 dlm_error(status);
 142                 return status;
 143         }
 144         __dlm_wait_on_lockres(res);
 145         __dlm_lockres_reserve_ast(res);
 146
 147         if (dlm_can_grant_new_lock(res, lock)) {
 148                 mlog(0, "I can grant this lock right away\n");
 149                 /* got it right away */
 150                 lock->lksb->status = DLM_NORMAL;
 151                 status = DLM_NORMAL;
 152                 dlm_lock_get(lock);
 153                 list_add_tail(&lock->list, &res->granted);
 154
 155                 /* for the recovery lock, we can't allow the ast
 156                  * to be queued since the dlmthread is already
 157                  * frozen.  but the recovery lock is always locked
 158                  * with LKM_NOQUEUE so we do not need the ast in
 159                  * this special case */
 160                 if (!dlm_is_recovery_lock(res->lockname.name,
 161                                           res->lockname.len)) {
 162                         kick_thread = 1;
 163                         call_ast = 1;
 164                 } else {
 165                         mlog(0, "%s: returning DLM_NORMAL to "
 166                              "node %u for reco lock\n", dlm->name,
 167                              lock->ml.node);
 168                 }
 169         } else {
 170                 /* for NOQUEUE request, unless we get the
 171                  * lock right away, return DLM_NOTQUEUED */
 172                 if (flags & LKM_NOQUEUE) {
 173                         status = DLM_NOTQUEUED;
 174                         if (dlm_is_recovery_lock(res->lockname.name,
 175                                                  res->lockname.len)) {
 176                                 mlog(0, "%s: returning NOTQUEUED to "
 177                                      "node %u for reco lock\n", dlm->name,
 178                                      lock->ml.node);
 179                         }
 180                 } else {
 181                         dlm_lock_get(lock);
 182                         list_add_tail(&lock->list, &res->blocked);
 183                         kick_thread = 1;
 184                 }
 185         }
 186         /* reduce the inflight count, this may result in the lockres
 187          * being purged below during calc_usage */
 188         if (lock->ml.node == dlm->node_num)
 189                 dlm_lockres_drop_inflight_ref(dlm, res);
 190
 191         spin_unlock(&res->spinlock);
 192         wake_up(&res->wq);
 193
 194         /* either queue the ast or release it */
 195         if (call_ast)
 196                 dlm_queue_ast(dlm, lock);
 197         else
 198                 dlm_lockres_release_ast(dlm, res);
 199
 200         dlm_lockres_calc_usage(dlm, res);
 201         if (kick_thread)
 202                 dlm_kick_thread(dlm, res);
 203
 204         return status;
 205 }
 206
 207 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
 208                              struct dlm_lock *lock)
 209 {
 210         /* remove from local queue if it failed */
 211         list_del_init(&lock->list);
 212         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 213 }
 214
 215
 216 /*
 217  * locking:
 218  *   caller needs:  none
 219  *   taken:         takes and drops res->spinlock
 220  *   held on exit:  none
 221  * returns: DLM_DENIED, DLM_RECOVERING, or net status
 222  */
 223 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 224                                       struct dlm_lock_resource *res,
 225                                       struct dlm_lock *lock, int flags)
 226 {
 227         enum dlm_status status = DLM_DENIED;
 228         int lockres_changed = 1;
 229
 230         mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
 231              lock->ml.type, res->lockname.len,
 232              res->lockname.name, flags);
 233
 234         spin_lock(&res->spinlock);
 235
 236         /* will exit this call with spinlock held */
 237         __dlm_wait_on_lockres(res);
 238         res->state |= DLM_LOCK_RES_IN_PROGRESS;
 239
 240         /* add lock to local (secondary) queue */
 241         dlm_lock_get(lock);
 242         list_add_tail(&lock->list, &res->blocked);
 243         lock->lock_pending = 1;
 244         spin_unlock(&res->spinlock);
 245
 246         /* spec seems to say that you will get DLM_NORMAL when the lock
 247          * has been queued, meaning we need to wait for a reply here. */
 248         status = dlm_send_remote_lock_request(dlm, res, lock, flags);
 249
 250         spin_lock(&res->spinlock);
 251         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 252         lock->lock_pending = 0;
 253         if (status != DLM_NORMAL) {
 254                 if (status == DLM_RECOVERING &&
 255                     dlm_is_recovery_lock(res->lockname.name,
 256                                          res->lockname.len)) {
 257                         /* recovery lock was mastered by dead node.
 258                          * we need to have calc_usage shoot down this
 259                          * lockres and completely remaster it. */
 260                         mlog(0, "%s: recovery lock was owned by "
 261                              "dead node %u, remaster it now.\n",
 262                              dlm->name, res->owner);
 263                 } else if (status != DLM_NOTQUEUED) {
 264                         /*
 265                          * DO NOT call calc_usage, as this would unhash
 266                          * the remote lockres before we ever get to use
 267                          * it.  treat as if we never made any change to
 268                          * the lockres.
 269                          */
 270                         lockres_changed = 0;
 271                         dlm_error(status);
 272                 }
 273                 dlm_revert_pending_lock(res, lock);
 274                 dlm_lock_put(lock);
 275         } else if (dlm_is_recovery_lock(res->lockname.name,
 276                                         res->lockname.len)) {
 277                 /* special case for the $RECOVERY lock.
 278                  * there will never be an AST delivered to put
 279                  * this lock on the proper secondary queue
 280                  * (granted), so do it manually. */
 281                 mlog(0, "%s: $RECOVERY lock for this node (%u) is "
 282                      "mastered by %u; got lock, manually granting (no ast)\n",
 283                      dlm->name, dlm->node_num, res->owner);
 284                 list_move_tail(&lock->list, &res->granted);
 285         }
 286         spin_unlock(&res->spinlock);
 287
 288         if (lockres_changed)
 289                 dlm_lockres_calc_usage(dlm, res);
 290
 291         wake_up(&res->wq);
 292         return status;
 293 }
 294
 295
 296 /* for remote lock creation.
 297  * locking:
 298  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
 299  *   taken:         none
 300  *   held on exit:  none
 301  * returns: DLM_NOLOCKMGR, or net status
 302  */
 303 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
 304                                                struct dlm_lock_resource *res,
 305                                                struct dlm_lock *lock, int flags)
 306 {
 307         struct dlm_create_lock create;
 308         int tmpret, status = 0;
 309         enum dlm_status ret;
 310
 311         memset(&create, 0, sizeof(create));
 312         create.node_idx = dlm->node_num;
 313         create.requested_type = lock->ml.type;
 314         create.cookie = lock->ml.cookie;
 315         create.namelen = res->lockname.len;
 316         create.flags = cpu_to_be32(flags);
 317         memcpy(create.name, res->lockname.name, create.namelen);
 318
 319         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
 320                                     sizeof(create), res->owner, &status);
 321         if (tmpret >= 0) {
 322                 ret = status;
 323                 if (ret == DLM_REJECTED) {
 324                         mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
 325                              "owned by node %u. That node is coming back up "
 326                              "currently.\n", dlm->name, create.namelen,
 327                              create.name, res->owner);
 328                         dlm_print_one_lock_resource(res);
 329                         BUG();
 330                 }
 331         } else {
 332                 mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
 333                      "node %u\n", dlm->name, create.namelen, create.name,
 334                      tmpret, res->owner);
 335                 if (dlm_is_host_down(tmpret))
 336                         ret = DLM_RECOVERING;
 337                 else
 338                         ret = dlm_err_to_dlm_status(tmpret);
 339         }
 340
 341         return ret;
 342 }
 343
 344 void dlm_lock_get(struct dlm_lock *lock)
 345 {
 346         kref_get(&lock->lock_refs);
 347 }
 348
 349 void dlm_lock_put(struct dlm_lock *lock)
 350 {
 351         kref_put(&lock->lock_refs, dlm_lock_release);
 352 }
 353
 354 static void dlm_lock_release(struct kref *kref)
 355 {
 356         struct dlm_lock *lock;
 357
 358         lock = container_of(kref, struct dlm_lock, lock_refs);
 359
 360         BUG_ON(!list_empty(&lock->list));
 361         BUG_ON(!list_empty(&lock->ast_list));
 362         BUG_ON(!list_empty(&lock->bast_list));
 363         BUG_ON(lock->ast_pending);
 364         BUG_ON(lock->bast_pending);
 365
 366         dlm_lock_detach_lockres(lock);
 367
 368         if (lock->lksb_kernel_allocated) {
 369                 mlog(0, "freeing kernel-allocated lksb\n");
 370                 kfree(lock->lksb);
 371         }
 372         kmem_cache_free(dlm_lock_cache, lock);
 373 }
 374
 375 /* associate a lock with it's lockres, getting a ref on the lockres */
 376 void dlm_lock_attach_lockres(struct dlm_lock *lock,
 377                              struct dlm_lock_resource *res)
 378 {
 379         dlm_lockres_get(res);
 380         lock->lockres = res;
 381 }
 382
 383 /* drop ref on lockres, if there is still one associated with lock */
 384 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
 385 {
 386         struct dlm_lock_resource *res;
 387
 388         res = lock->lockres;
 389         if (res) {
 390                 lock->lockres = NULL;
 391                 mlog(0, "removing lock's lockres reference\n");
 392                 dlm_lockres_put(res);
 393         }
 394 }
 395
 396 static void dlm_init_lock(struct dlm_lock *newlock, int type,
 397                           u8 node, u64 cookie)
 398 {
 399         INIT_LIST_HEAD(&newlock->list);
 400         INIT_LIST_HEAD(&newlock->ast_list);
 401         INIT_LIST_HEAD(&newlock->bast_list);
 402         spin_lock_init(&newlock->spinlock);
 403         newlock->ml.type = type;
 404         newlock->ml.convert_type = LKM_IVMODE;
 405         newlock->ml.highest_blocked = LKM_IVMODE;
 406         newlock->ml.node = node;
 407         newlock->ml.pad1 = 0;
 408         newlock->ml.list = 0;
 409         newlock->ml.flags = 0;
 410         newlock->ast = NULL;
 411         newlock->bast = NULL;
 412         newlock->astdata = NULL;
 413         newlock->ml.cookie = cpu_to_be64(cookie);
 414         newlock->ast_pending = 0;
 415         newlock->bast_pending = 0;
 416         newlock->convert_pending = 0;
 417         newlock->lock_pending = 0;
 418         newlock->unlock_pending = 0;
 419         newlock->cancel_pending = 0;
 420         newlock->lksb_kernel_allocated = 0;
 421
 422         kref_init(&newlock->lock_refs);
 423 }
 424
 425 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 426                                struct dlm_lockstatus *lksb)
 427 {
 428         struct dlm_lock *lock;
 429         int kernel_allocated = 0;
 430
 431         lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
 432         if (!lock)
 433                 return NULL;
 434
 435         if (!lksb) {
 436                 /* zero memory only if kernel-allocated */
 437                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 438                 if (!lksb) {
 439                         kfree(lock);
 440                         return NULL;
 441                 }
 442                 kernel_allocated = 1;
 443         }
 444
 445         dlm_init_lock(lock, type, node, cookie);
 446         if (kernel_allocated)
 447                 lock->lksb_kernel_allocated = 1;
 448         lock->lksb = lksb;
 449         lksb->lockid = lock;
 450         return lock;
 451 }
 452
 453 /* handler for lock creation net message
 454  * locking:
 455  *   caller needs:  none
 456  *   taken:         takes and drops res->spinlock
 457  *   held on exit:  none
 458  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
 459  */
 460 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 461                             void **ret_data)
 462 {
 463         struct dlm_ctxt *dlm = data;
 464         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
 465         struct dlm_lock_resource *res = NULL;
 466         struct dlm_lock *newlock = NULL;
 467         struct dlm_lockstatus *lksb = NULL;
 468         enum dlm_status status = DLM_NORMAL;
 469         char *name;
 470         unsigned int namelen;
 471
 472         BUG_ON(!dlm);
 473
 474         if (!dlm_grab(dlm))
 475                 return DLM_REJECTED;
 476
 477         name = create->name;
 478         namelen = create->namelen;
 479         status = DLM_REJECTED;
 480         if (!dlm_domain_fully_joined(dlm)) {
 481                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
 482                      "sending a create_lock message for lock %.*s!\n",
 483                      dlm->name, create->node_idx, namelen, name);
 484                 dlm_error(status);
 485                 goto leave;
 486         }
 487
 488         status = DLM_IVBUFLEN;
 489         if (namelen > DLM_LOCKID_NAME_MAX) {
 490                 dlm_error(status);
 491                 goto leave;
 492         }
 493
 494         status = DLM_SYSERR;
 495         newlock = dlm_new_lock(create->requested_type,
 496                                create->node_idx,
 497                                be64_to_cpu(create->cookie), NULL);
 498         if (!newlock) {
 499                 dlm_error(status);
 500                 goto leave;
 501         }
 502
 503         lksb = newlock->lksb;
 504
 505         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
 506                 lksb->flags |= DLM_LKSB_GET_LVB;
 507                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
 508         }
 509
 510         status = DLM_IVLOCKID;
 511         res = dlm_lookup_lockres(dlm, name, namelen);
 512         if (!res) {
 513                 dlm_error(status);
 514                 goto leave;
 515         }
 516
 517         spin_lock(&res->spinlock);
 518         status = __dlm_lockres_state_to_status(res);
 519         spin_unlock(&res->spinlock);
 520
 521         if (status != DLM_NORMAL) {
 522                 mlog(0, "lockres recovering/migrating/in-progress\n");
 523                 goto leave;
 524         }
 525
 526         dlm_lock_attach_lockres(newlock, res);
 527
 528         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
 529 leave:
 530         if (status != DLM_NORMAL)
 531                 if (newlock)
 532                         dlm_lock_put(newlock);
 533
 534         if (res)
 535                 dlm_lockres_put(res);
 536
 537         dlm_put(dlm);
 538
 539         return status;
 540 }
 541
 542
 543 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
 544 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 545 {
 546         u64 tmpnode = node_num;
 547
 548         /* shift single byte of node num into top 8 bits */
 549         tmpnode <<= 56;
 550
 551         spin_lock(&dlm_cookie_lock);
 552         *cookie = (dlm_next_cookie | tmpnode);
 553         if (++dlm_next_cookie & 0xff00000000000000ull) {
 554                 mlog(0, "This node's cookie will now wrap!\n");
 555                 dlm_next_cookie = 1;
 556         }
 557         spin_unlock(&dlm_cookie_lock);
 558 }
 559
 560 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 561                         struct dlm_lockstatus *lksb, int flags,
 562                         const char *name, int namelen, dlm_astlockfunc_t *ast,
 563                         void *data, dlm_bastlockfunc_t *bast)
 564 {
 565         enum dlm_status status;
 566         struct dlm_lock_resource *res = NULL;
 567         struct dlm_lock *lock = NULL;
 568         int convert = 0, recovery = 0;
 569
 570         /* yes this function is a mess.
 571          * TODO: clean this up.  lots of common code in the
 572          *       lock and convert paths, especially in the retry blocks */
 573         if (!lksb) {
 574                 dlm_error(DLM_BADARGS);
 575                 return DLM_BADARGS;
 576         }
 577
 578         status = DLM_BADPARAM;
 579         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
 580                 dlm_error(status);
 581                 goto error;
 582         }
 583
 584         if (flags & ~LKM_VALID_FLAGS) {
 585                 dlm_error(status);
 586                 goto error;
 587         }
 588
 589         convert = (flags & LKM_CONVERT);
 590         recovery = (flags & LKM_RECOVERY);
 591
 592         if (recovery &&
 593             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 594                 dlm_error(status);
 595                 goto error;
 596         }
 597         if (convert && (flags & LKM_LOCAL)) {
 598                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
 599                 goto error;
 600         }
 601
 602         if (convert) {
 603                 /* CONVERT request */
 604
 605                 /* if converting, must pass in a valid dlm_lock */
 606                 lock = lksb->lockid;
 607                 if (!lock) {
 608                         mlog(ML_ERROR, "NULL lock pointer in convert "
 609                              "request\n");
 610                         goto error;
 611                 }
 612
 613                 res = lock->lockres;
 614                 if (!res) {
 615                         mlog(ML_ERROR, "NULL lockres pointer in convert "
 616                              "request\n");
 617                         goto error;
 618                 }
 619                 dlm_lockres_get(res);
 620
 621                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
 622                  * static after the original lock call.  convert requests will
 623                  * ensure that everything is the same, or return DLM_BADARGS.
 624                  * this means that DLM_DENIED_NOASTS will never be returned.
 625                  */
 626                 if (lock->lksb != lksb || lock->ast != ast ||
 627                     lock->bast != bast || lock->astdata != data) {
 628                         status = DLM_BADARGS;
 629                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
 630                              "astdata=%p\n", lksb, ast, bast, data);
 631                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
 632                              "astdata=%p\n", lock->lksb, lock->ast,
 633                              lock->bast, lock->astdata);
 634                         goto error;
 635                 }
 636 retry_convert:
 637                 dlm_wait_for_recovery(dlm);
 638
 639                 if (res->owner == dlm->node_num)
 640                         status = dlmconvert_master(dlm, res, lock, flags, mode);
 641                 else
 642                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
 643                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 644                     status == DLM_FORWARD) {
 645                         /* for now, see how this works without sleeping
 646                          * and just retry right away.  I suspect the reco
 647                          * or migration will complete fast enough that
 648                          * no waiting will be necessary */
 649                         mlog(0, "retrying convert with migration/recovery/"
 650                              "in-progress\n");
 651                         msleep(100);
 652                         goto retry_convert;
 653                 }
 654         } else {
 655                 u64 tmpcookie;
 656
 657                 /* LOCK request */
 658                 status = DLM_BADARGS;
 659                 if (!name) {
 660                         dlm_error(status);
 661                         goto error;
 662                 }
 663
 664                 status = DLM_IVBUFLEN;
 665                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 666                         dlm_error(status);
 667                         goto error;
 668                 }
 669
 670                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
 671                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
 672                 if (!lock) {
 673                         dlm_error(status);
 674                         goto error;
 675                 }
 676
 677                 if (!recovery)
 678                         dlm_wait_for_recovery(dlm);
 679
 680                 /* find or create the lock resource */
 681                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
 682                 if (!res) {
 683                         status = DLM_IVLOCKID;
 684                         dlm_error(status);
 685                         goto error;
 686                 }
 687
 688                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
 689                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
 690
 691                 dlm_lock_attach_lockres(lock, res);
 692                 lock->ast = ast;
 693                 lock->bast = bast;
 694                 lock->astdata = data;
 695
 696 retry_lock:
 697                 if (flags & LKM_VALBLK) {
 698                         mlog(0, "LKM_VALBLK passed by caller\n");
 699
 700                         /* LVB requests for non PR, PW or EX locks are
 701                          * ignored. */
 702                         if (mode < LKM_PRMODE)
 703                                 flags &= ~LKM_VALBLK;
 704                         else {
 705                                 flags |= LKM_GET_LVB;
 706                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
 707                         }
 708                 }
 709
 710                 if (res->owner == dlm->node_num)
 711                         status = dlmlock_master(dlm, res, lock, flags);
 712                 else
 713                         status = dlmlock_remote(dlm, res, lock, flags);
 714
 715                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 716                     status == DLM_FORWARD) {
 717                         mlog(0, "retrying lock with migration/"
 718                              "recovery/in progress\n");
 719                         msleep(100);
 720                         /* no waiting for dlm_reco_thread */
 721                         if (recovery) {
 722                                 if (status != DLM_RECOVERING)
 723                                         goto retry_lock;
 724
 725                                 mlog(0, "%s: got RECOVERING "
 726                                      "for $RECOVERY lock, master "
 727                                      "was %u\n", dlm->name,
 728                                      res->owner);
 729                                 /* wait to see the node go down, then
 730                                  * drop down and allow the lockres to
 731                                  * get cleaned up.  need to remaster. */
 732                                 dlm_wait_for_node_death(dlm, res->owner,
 733                                                 DLM_NODE_DEATH_WAIT_MAX);
 734                         } else {
 735                                 dlm_wait_for_recovery(dlm);
 736                                 goto retry_lock;
 737                         }
 738                 }
 739
 740                 if (status != DLM_NORMAL) {
 741                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 742                         if (status != DLM_NOTQUEUED)
 743                                 dlm_error(status);
 744                         goto error;
 745                 }
 746         }
 747
 748 error:
 749         if (status != DLM_NORMAL) {
 750                 if (lock && !convert)
 751                         dlm_lock_put(lock);
 752                 // this is kind of unnecessary
 753                 lksb->status = status;
 754         }
 755
 756         /* put lockres ref from the convert path
 757          * or from dlm_get_lock_resource */
 758         if (res)
 759                 dlm_lockres_put(res);
 760
 761         return status;
 762 }
 763 EXPORT_SYMBOL_GPL(dlmlock);