Pileus Git - ~andy/linux/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/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                         status = DLM_NORMAL;
 182                         dlm_lock_get(lock);
 183                         list_add_tail(&lock->list, &res->blocked);
 184                         kick_thread = 1;
 185                 }
 186         }
 187
 188         spin_unlock(&res->spinlock);
 189         wake_up(&res->wq);
 190
 191         /* either queue the ast or release it */
 192         if (call_ast)
 193                 dlm_queue_ast(dlm, lock);
 194         else
 195                 dlm_lockres_release_ast(dlm, res);
 196
 197         dlm_lockres_calc_usage(dlm, res);
 198         if (kick_thread)
 199                 dlm_kick_thread(dlm, res);
 200
 201         return status;
 202 }
 203
 204 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
 205                              struct dlm_lock *lock)
 206 {
 207         /* remove from local queue if it failed */
 208         list_del_init(&lock->list);
 209         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 210 }
 211
 212
 213 /*
 214  * locking:
 215  *   caller needs:  none
 216  *   taken:         takes and drops res->spinlock
 217  *   held on exit:  none
 218  * returns: DLM_DENIED, DLM_RECOVERING, or net status
 219  */
 220 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 221                                       struct dlm_lock_resource *res,
 222                                       struct dlm_lock *lock, int flags)
 223 {
 224         enum dlm_status status = DLM_DENIED;
 225         int lockres_changed = 1;
 226
 227         mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
 228              lock->ml.type, res->lockname.len,
 229              res->lockname.name, flags);
 230
 231         /*
 232          * Wait if resource is getting recovered, remastered, etc.
 233          * If the resource was remastered and new owner is self, then exit.
 234          */
 235         spin_lock(&res->spinlock);
 236         __dlm_wait_on_lockres(res);
 237         if (res->owner == dlm->node_num) {
 238                 spin_unlock(&res->spinlock);
 239                 return DLM_RECOVERING;
 240         }
 241         res->state |= DLM_LOCK_RES_IN_PROGRESS;
 242
 243         /* add lock to local (secondary) queue */
 244         dlm_lock_get(lock);
 245         list_add_tail(&lock->list, &res->blocked);
 246         lock->lock_pending = 1;
 247         spin_unlock(&res->spinlock);
 248
 249         /* spec seems to say that you will get DLM_NORMAL when the lock
 250          * has been queued, meaning we need to wait for a reply here. */
 251         status = dlm_send_remote_lock_request(dlm, res, lock, flags);
 252
 253         spin_lock(&res->spinlock);
 254         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 255         lock->lock_pending = 0;
 256         if (status != DLM_NORMAL) {
 257                 if (status == DLM_RECOVERING &&
 258                     dlm_is_recovery_lock(res->lockname.name,
 259                                          res->lockname.len)) {
 260                         /* recovery lock was mastered by dead node.
 261                          * we need to have calc_usage shoot down this
 262                          * lockres and completely remaster it. */
 263                         mlog(0, "%s: recovery lock was owned by "
 264                              "dead node %u, remaster it now.\n",
 265                              dlm->name, res->owner);
 266                 } else if (status != DLM_NOTQUEUED) {
 267                         /*
 268                          * DO NOT call calc_usage, as this would unhash
 269                          * the remote lockres before we ever get to use
 270                          * it.  treat as if we never made any change to
 271                          * the lockres.
 272                          */
 273                         lockres_changed = 0;
 274                         dlm_error(status);
 275                 }
 276                 dlm_revert_pending_lock(res, lock);
 277                 dlm_lock_put(lock);
 278         } else if (dlm_is_recovery_lock(res->lockname.name,
 279                                         res->lockname.len)) {
 280                 /* special case for the $RECOVERY lock.
 281                  * there will never be an AST delivered to put
 282                  * this lock on the proper secondary queue
 283                  * (granted), so do it manually. */
 284                 mlog(0, "%s: $RECOVERY lock for this node (%u) is "
 285                      "mastered by %u; got lock, manually granting (no ast)\n",
 286                      dlm->name, dlm->node_num, res->owner);
 287                 list_move_tail(&lock->list, &res->granted);
 288         }
 289         spin_unlock(&res->spinlock);
 290
 291         if (lockres_changed)
 292                 dlm_lockres_calc_usage(dlm, res);
 293
 294         wake_up(&res->wq);
 295         return status;
 296 }
 297
 298
 299 /* for remote lock creation.
 300  * locking:
 301  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
 302  *   taken:         none
 303  *   held on exit:  none
 304  * returns: DLM_NOLOCKMGR, or net status
 305  */
 306 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
 307                                                struct dlm_lock_resource *res,
 308                                                struct dlm_lock *lock, int flags)
 309 {
 310         struct dlm_create_lock create;
 311         int tmpret, status = 0;
 312         enum dlm_status ret;
 313
 314         memset(&create, 0, sizeof(create));
 315         create.node_idx = dlm->node_num;
 316         create.requested_type = lock->ml.type;
 317         create.cookie = lock->ml.cookie;
 318         create.namelen = res->lockname.len;
 319         create.flags = cpu_to_be32(flags);
 320         memcpy(create.name, res->lockname.name, create.namelen);
 321
 322         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
 323                                     sizeof(create), res->owner, &status);
 324         if (tmpret >= 0) {
 325                 ret = status;
 326                 if (ret == DLM_REJECTED) {
 327                         mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
 328                              "owned by node %u. That node is coming back up "
 329                              "currently.\n", dlm->name, create.namelen,
 330                              create.name, res->owner);
 331                         dlm_print_one_lock_resource(res);
 332                         BUG();
 333                 }
 334         } else {
 335                 mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
 336                      "node %u\n", dlm->name, create.namelen, create.name,
 337                      tmpret, res->owner);
 338                 if (dlm_is_host_down(tmpret))
 339                         ret = DLM_RECOVERING;
 340                 else
 341                         ret = dlm_err_to_dlm_status(tmpret);
 342         }
 343
 344         return ret;
 345 }
 346
 347 void dlm_lock_get(struct dlm_lock *lock)
 348 {
 349         kref_get(&lock->lock_refs);
 350 }
 351
 352 void dlm_lock_put(struct dlm_lock *lock)
 353 {
 354         kref_put(&lock->lock_refs, dlm_lock_release);
 355 }
 356
 357 static void dlm_lock_release(struct kref *kref)
 358 {
 359         struct dlm_lock *lock;
 360
 361         lock = container_of(kref, struct dlm_lock, lock_refs);
 362
 363         BUG_ON(!list_empty(&lock->list));
 364         BUG_ON(!list_empty(&lock->ast_list));
 365         BUG_ON(!list_empty(&lock->bast_list));
 366         BUG_ON(lock->ast_pending);
 367         BUG_ON(lock->bast_pending);
 368
 369         dlm_lock_detach_lockres(lock);
 370
 371         if (lock->lksb_kernel_allocated) {
 372                 mlog(0, "freeing kernel-allocated lksb\n");
 373                 kfree(lock->lksb);
 374         }
 375         kmem_cache_free(dlm_lock_cache, lock);
 376 }
 377
 378 /* associate a lock with it's lockres, getting a ref on the lockres */
 379 void dlm_lock_attach_lockres(struct dlm_lock *lock,
 380                              struct dlm_lock_resource *res)
 381 {
 382         dlm_lockres_get(res);
 383         lock->lockres = res;
 384 }
 385
 386 /* drop ref on lockres, if there is still one associated with lock */
 387 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
 388 {
 389         struct dlm_lock_resource *res;
 390
 391         res = lock->lockres;
 392         if (res) {
 393                 lock->lockres = NULL;
 394                 mlog(0, "removing lock's lockres reference\n");
 395                 dlm_lockres_put(res);
 396         }
 397 }
 398
 399 static void dlm_init_lock(struct dlm_lock *newlock, int type,
 400                           u8 node, u64 cookie)
 401 {
 402         INIT_LIST_HEAD(&newlock->list);
 403         INIT_LIST_HEAD(&newlock->ast_list);
 404         INIT_LIST_HEAD(&newlock->bast_list);
 405         spin_lock_init(&newlock->spinlock);
 406         newlock->ml.type = type;
 407         newlock->ml.convert_type = LKM_IVMODE;
 408         newlock->ml.highest_blocked = LKM_IVMODE;
 409         newlock->ml.node = node;
 410         newlock->ml.pad1 = 0;
 411         newlock->ml.list = 0;
 412         newlock->ml.flags = 0;
 413         newlock->ast = NULL;
 414         newlock->bast = NULL;
 415         newlock->astdata = NULL;
 416         newlock->ml.cookie = cpu_to_be64(cookie);
 417         newlock->ast_pending = 0;
 418         newlock->bast_pending = 0;
 419         newlock->convert_pending = 0;
 420         newlock->lock_pending = 0;
 421         newlock->unlock_pending = 0;
 422         newlock->cancel_pending = 0;
 423         newlock->lksb_kernel_allocated = 0;
 424
 425         kref_init(&newlock->lock_refs);
 426 }
 427
 428 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 429                                struct dlm_lockstatus *lksb)
 430 {
 431         struct dlm_lock *lock;
 432         int kernel_allocated = 0;
 433
 434         lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
 435         if (!lock)
 436                 return NULL;
 437
 438         if (!lksb) {
 439                 /* zero memory only if kernel-allocated */
 440                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 441                 if (!lksb) {
 442                         kmem_cache_free(dlm_lock_cache, lock);
 443                         return NULL;
 444                 }
 445                 kernel_allocated = 1;
 446         }
 447
 448         dlm_init_lock(lock, type, node, cookie);
 449         if (kernel_allocated)
 450                 lock->lksb_kernel_allocated = 1;
 451         lock->lksb = lksb;
 452         lksb->lockid = lock;
 453         return lock;
 454 }
 455
 456 /* handler for lock creation net message
 457  * locking:
 458  *   caller needs:  none
 459  *   taken:         takes and drops res->spinlock
 460  *   held on exit:  none
 461  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
 462  */
 463 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 464                             void **ret_data)
 465 {
 466         struct dlm_ctxt *dlm = data;
 467         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
 468         struct dlm_lock_resource *res = NULL;
 469         struct dlm_lock *newlock = NULL;
 470         struct dlm_lockstatus *lksb = NULL;
 471         enum dlm_status status = DLM_NORMAL;
 472         char *name;
 473         unsigned int namelen;
 474
 475         BUG_ON(!dlm);
 476
 477         if (!dlm_grab(dlm))
 478                 return DLM_REJECTED;
 479
 480         name = create->name;
 481         namelen = create->namelen;
 482         status = DLM_REJECTED;
 483         if (!dlm_domain_fully_joined(dlm)) {
 484                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
 485                      "sending a create_lock message for lock %.*s!\n",
 486                      dlm->name, create->node_idx, namelen, name);
 487                 dlm_error(status);
 488                 goto leave;
 489         }
 490
 491         status = DLM_IVBUFLEN;
 492         if (namelen > DLM_LOCKID_NAME_MAX) {
 493                 dlm_error(status);
 494                 goto leave;
 495         }
 496
 497         status = DLM_SYSERR;
 498         newlock = dlm_new_lock(create->requested_type,
 499                                create->node_idx,
 500                                be64_to_cpu(create->cookie), NULL);
 501         if (!newlock) {
 502                 dlm_error(status);
 503                 goto leave;
 504         }
 505
 506         lksb = newlock->lksb;
 507
 508         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
 509                 lksb->flags |= DLM_LKSB_GET_LVB;
 510                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
 511         }
 512
 513         status = DLM_IVLOCKID;
 514         res = dlm_lookup_lockres(dlm, name, namelen);
 515         if (!res) {
 516                 dlm_error(status);
 517                 goto leave;
 518         }
 519
 520         spin_lock(&res->spinlock);
 521         status = __dlm_lockres_state_to_status(res);
 522         spin_unlock(&res->spinlock);
 523
 524         if (status != DLM_NORMAL) {
 525                 mlog(0, "lockres recovering/migrating/in-progress\n");
 526                 goto leave;
 527         }
 528
 529         dlm_lock_attach_lockres(newlock, res);
 530
 531         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
 532 leave:
 533         if (status != DLM_NORMAL)
 534                 if (newlock)
 535                         dlm_lock_put(newlock);
 536
 537         if (res)
 538                 dlm_lockres_put(res);
 539
 540         dlm_put(dlm);
 541
 542         return status;
 543 }
 544
 545
 546 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
 547 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 548 {
 549         u64 tmpnode = node_num;
 550
 551         /* shift single byte of node num into top 8 bits */
 552         tmpnode <<= 56;
 553
 554         spin_lock(&dlm_cookie_lock);
 555         *cookie = (dlm_next_cookie | tmpnode);
 556         if (++dlm_next_cookie & 0xff00000000000000ull) {
 557                 mlog(0, "This node's cookie will now wrap!\n");
 558                 dlm_next_cookie = 1;
 559         }
 560         spin_unlock(&dlm_cookie_lock);
 561 }
 562
 563 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 564                         struct dlm_lockstatus *lksb, int flags,
 565                         const char *name, int namelen, dlm_astlockfunc_t *ast,
 566                         void *data, dlm_bastlockfunc_t *bast)
 567 {
 568         enum dlm_status status;
 569         struct dlm_lock_resource *res = NULL;
 570         struct dlm_lock *lock = NULL;
 571         int convert = 0, recovery = 0;
 572
 573         /* yes this function is a mess.
 574          * TODO: clean this up.  lots of common code in the
 575          *       lock and convert paths, especially in the retry blocks */
 576         if (!lksb) {
 577                 dlm_error(DLM_BADARGS);
 578                 return DLM_BADARGS;
 579         }
 580
 581         status = DLM_BADPARAM;
 582         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
 583                 dlm_error(status);
 584                 goto error;
 585         }
 586
 587         if (flags & ~LKM_VALID_FLAGS) {
 588                 dlm_error(status);
 589                 goto error;
 590         }
 591
 592         convert = (flags & LKM_CONVERT);
 593         recovery = (flags & LKM_RECOVERY);
 594
 595         if (recovery &&
 596             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 597                 dlm_error(status);
 598                 goto error;
 599         }
 600         if (convert && (flags & LKM_LOCAL)) {
 601                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
 602                 goto error;
 603         }
 604
 605         if (convert) {
 606                 /* CONVERT request */
 607
 608                 /* if converting, must pass in a valid dlm_lock */
 609                 lock = lksb->lockid;
 610                 if (!lock) {
 611                         mlog(ML_ERROR, "NULL lock pointer in convert "
 612                              "request\n");
 613                         goto error;
 614                 }
 615
 616                 res = lock->lockres;
 617                 if (!res) {
 618                         mlog(ML_ERROR, "NULL lockres pointer in convert "
 619                              "request\n");
 620                         goto error;
 621                 }
 622                 dlm_lockres_get(res);
 623
 624                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
 625                  * static after the original lock call.  convert requests will
 626                  * ensure that everything is the same, or return DLM_BADARGS.
 627                  * this means that DLM_DENIED_NOASTS will never be returned.
 628                  */
 629                 if (lock->lksb != lksb || lock->ast != ast ||
 630                     lock->bast != bast || lock->astdata != data) {
 631                         status = DLM_BADARGS;
 632                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
 633                              "astdata=%p\n", lksb, ast, bast, data);
 634                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
 635                              "astdata=%p\n", lock->lksb, lock->ast,
 636                              lock->bast, lock->astdata);
 637                         goto error;
 638                 }
 639 retry_convert:
 640                 dlm_wait_for_recovery(dlm);
 641
 642                 if (res->owner == dlm->node_num)
 643                         status = dlmconvert_master(dlm, res, lock, flags, mode);
 644                 else
 645                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
 646                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 647                     status == DLM_FORWARD) {
 648                         /* for now, see how this works without sleeping
 649                          * and just retry right away.  I suspect the reco
 650                          * or migration will complete fast enough that
 651                          * no waiting will be necessary */
 652                         mlog(0, "retrying convert with migration/recovery/"
 653                              "in-progress\n");
 654                         msleep(100);
 655                         goto retry_convert;
 656                 }
 657         } else {
 658                 u64 tmpcookie;
 659
 660                 /* LOCK request */
 661                 status = DLM_BADARGS;
 662                 if (!name) {
 663                         dlm_error(status);
 664                         goto error;
 665                 }
 666
 667                 status = DLM_IVBUFLEN;
 668                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 669                         dlm_error(status);
 670                         goto error;
 671                 }
 672
 673                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
 674                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
 675                 if (!lock) {
 676                         dlm_error(status);
 677                         goto error;
 678                 }
 679
 680                 if (!recovery)
 681                         dlm_wait_for_recovery(dlm);
 682
 683                 /* find or create the lock resource */
 684                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
 685                 if (!res) {
 686                         status = DLM_IVLOCKID;
 687                         dlm_error(status);
 688                         goto error;
 689                 }
 690
 691                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
 692                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
 693
 694                 dlm_lock_attach_lockres(lock, res);
 695                 lock->ast = ast;
 696                 lock->bast = bast;
 697                 lock->astdata = data;
 698
 699 retry_lock:
 700                 if (flags & LKM_VALBLK) {
 701                         mlog(0, "LKM_VALBLK passed by caller\n");
 702
 703                         /* LVB requests for non PR, PW or EX locks are
 704                          * ignored. */
 705                         if (mode < LKM_PRMODE)
 706                                 flags &= ~LKM_VALBLK;
 707                         else {
 708                                 flags |= LKM_GET_LVB;
 709                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
 710                         }
 711                 }
 712
 713                 if (res->owner == dlm->node_num)
 714                         status = dlmlock_master(dlm, res, lock, flags);
 715                 else
 716                         status = dlmlock_remote(dlm, res, lock, flags);
 717
 718                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 719                     status == DLM_FORWARD) {
 720                         msleep(100);
 721                         if (recovery) {
 722                                 if (status != DLM_RECOVERING)
 723                                         goto retry_lock;
 724                                 /* wait to see the node go down, then
 725                                  * drop down and allow the lockres to
 726                                  * get cleaned up.  need to remaster. */
 727                                 dlm_wait_for_node_death(dlm, res->owner,
 728                                                 DLM_NODE_DEATH_WAIT_MAX);
 729                         } else {
 730                                 dlm_wait_for_recovery(dlm);
 731                                 goto retry_lock;
 732                         }
 733                 }
 734
 735                 /* Inflight taken in dlm_get_lock_resource() is dropped here */
 736                 spin_lock(&res->spinlock);
 737                 dlm_lockres_drop_inflight_ref(dlm, res);
 738                 spin_unlock(&res->spinlock);
 739
 740                 dlm_lockres_calc_usage(dlm, res);
 741                 dlm_kick_thread(dlm, res);
 742
 743                 if (status != DLM_NORMAL) {
 744                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 745                         if (status != DLM_NOTQUEUED)
 746                                 dlm_error(status);
 747                         goto error;
 748                 }
 749         }
 750
 751 error:
 752         if (status != DLM_NORMAL) {
 753                 if (lock && !convert)
 754                         dlm_lock_put(lock);
 755                 // this is kind of unnecessary
 756                 lksb->status = status;
 757         }
 758
 759         /* put lockres ref from the convert path
 760          * or from dlm_get_lock_resource */
 761         if (res)
 762                 dlm_lockres_put(res);
 763
 764         return status;
 765 }
 766 EXPORT_SYMBOL_GPL(dlmlock);