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                 // successfully sent and received
 323                 ret = status;  // this is already a dlm_status
 324                 if (ret == DLM_REJECTED) {
 325                         mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
 326                              "no longer owned by %u.  that node is coming back "
 327                              "up currently.\n", dlm->name, create.namelen,
 328                              create.name, res->owner);
 329                         dlm_print_one_lock_resource(res);
 330                         BUG();
 331                 }
 332         } else {
 333                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
 334                      "node %u\n", tmpret, DLM_CREATE_LOCK_MSG, dlm->key,
 335                      res->owner);
 336                 if (dlm_is_host_down(tmpret)) {
 337                         ret = DLM_RECOVERING;
 338                         mlog(0, "node %u died so returning DLM_RECOVERING "
 339                              "from lock message!\n", res->owner);
 340                 } else {
 341                         ret = dlm_err_to_dlm_status(tmpret);
 342                 }
 343         }
 344
 345         return ret;
 346 }
 347
 348 void dlm_lock_get(struct dlm_lock *lock)
 349 {
 350         kref_get(&lock->lock_refs);
 351 }
 352
 353 void dlm_lock_put(struct dlm_lock *lock)
 354 {
 355         kref_put(&lock->lock_refs, dlm_lock_release);
 356 }
 357
 358 static void dlm_lock_release(struct kref *kref)
 359 {
 360         struct dlm_lock *lock;
 361
 362         lock = container_of(kref, struct dlm_lock, lock_refs);
 363
 364         BUG_ON(!list_empty(&lock->list));
 365         BUG_ON(!list_empty(&lock->ast_list));
 366         BUG_ON(!list_empty(&lock->bast_list));
 367         BUG_ON(lock->ast_pending);
 368         BUG_ON(lock->bast_pending);
 369
 370         dlm_lock_detach_lockres(lock);
 371
 372         if (lock->lksb_kernel_allocated) {
 373                 mlog(0, "freeing kernel-allocated lksb\n");
 374                 kfree(lock->lksb);
 375         }
 376         kmem_cache_free(dlm_lock_cache, lock);
 377 }
 378
 379 /* associate a lock with it's lockres, getting a ref on the lockres */
 380 void dlm_lock_attach_lockres(struct dlm_lock *lock,
 381                              struct dlm_lock_resource *res)
 382 {
 383         dlm_lockres_get(res);
 384         lock->lockres = res;
 385 }
 386
 387 /* drop ref on lockres, if there is still one associated with lock */
 388 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
 389 {
 390         struct dlm_lock_resource *res;
 391
 392         res = lock->lockres;
 393         if (res) {
 394                 lock->lockres = NULL;
 395                 mlog(0, "removing lock's lockres reference\n");
 396                 dlm_lockres_put(res);
 397         }
 398 }
 399
 400 static void dlm_init_lock(struct dlm_lock *newlock, int type,
 401                           u8 node, u64 cookie)
 402 {
 403         INIT_LIST_HEAD(&newlock->list);
 404         INIT_LIST_HEAD(&newlock->ast_list);
 405         INIT_LIST_HEAD(&newlock->bast_list);
 406         spin_lock_init(&newlock->spinlock);
 407         newlock->ml.type = type;
 408         newlock->ml.convert_type = LKM_IVMODE;
 409         newlock->ml.highest_blocked = LKM_IVMODE;
 410         newlock->ml.node = node;
 411         newlock->ml.pad1 = 0;
 412         newlock->ml.list = 0;
 413         newlock->ml.flags = 0;
 414         newlock->ast = NULL;
 415         newlock->bast = NULL;
 416         newlock->astdata = NULL;
 417         newlock->ml.cookie = cpu_to_be64(cookie);
 418         newlock->ast_pending = 0;
 419         newlock->bast_pending = 0;
 420         newlock->convert_pending = 0;
 421         newlock->lock_pending = 0;
 422         newlock->unlock_pending = 0;
 423         newlock->cancel_pending = 0;
 424         newlock->lksb_kernel_allocated = 0;
 425
 426         kref_init(&newlock->lock_refs);
 427 }
 428
 429 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 430                                struct dlm_lockstatus *lksb)
 431 {
 432         struct dlm_lock *lock;
 433         int kernel_allocated = 0;
 434
 435         lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
 436         if (!lock)
 437                 return NULL;
 438
 439         if (!lksb) {
 440                 /* zero memory only if kernel-allocated */
 441                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 442                 if (!lksb) {
 443                         kfree(lock);
 444                         return NULL;
 445                 }
 446                 kernel_allocated = 1;
 447         }
 448
 449         dlm_init_lock(lock, type, node, cookie);
 450         if (kernel_allocated)
 451                 lock->lksb_kernel_allocated = 1;
 452         lock->lksb = lksb;
 453         lksb->lockid = lock;
 454         return lock;
 455 }
 456
 457 /* handler for lock creation net message
 458  * locking:
 459  *   caller needs:  none
 460  *   taken:         takes and drops res->spinlock
 461  *   held on exit:  none
 462  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
 463  */
 464 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
 465                             void **ret_data)
 466 {
 467         struct dlm_ctxt *dlm = data;
 468         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
 469         struct dlm_lock_resource *res = NULL;
 470         struct dlm_lock *newlock = NULL;
 471         struct dlm_lockstatus *lksb = NULL;
 472         enum dlm_status status = DLM_NORMAL;
 473         char *name;
 474         unsigned int namelen;
 475
 476         BUG_ON(!dlm);
 477
 478         if (!dlm_grab(dlm))
 479                 return DLM_REJECTED;
 480
 481         name = create->name;
 482         namelen = create->namelen;
 483         status = DLM_REJECTED;
 484         if (!dlm_domain_fully_joined(dlm)) {
 485                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
 486                      "sending a create_lock message for lock %.*s!\n",
 487                      dlm->name, create->node_idx, namelen, name);
 488                 dlm_error(status);
 489                 goto leave;
 490         }
 491
 492         status = DLM_IVBUFLEN;
 493         if (namelen > DLM_LOCKID_NAME_MAX) {
 494                 dlm_error(status);
 495                 goto leave;
 496         }
 497
 498         status = DLM_SYSERR;
 499         newlock = dlm_new_lock(create->requested_type,
 500                                create->node_idx,
 501                                be64_to_cpu(create->cookie), NULL);
 502         if (!newlock) {
 503                 dlm_error(status);
 504                 goto leave;
 505         }
 506
 507         lksb = newlock->lksb;
 508
 509         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
 510                 lksb->flags |= DLM_LKSB_GET_LVB;
 511                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
 512         }
 513
 514         status = DLM_IVLOCKID;
 515         res = dlm_lookup_lockres(dlm, name, namelen);
 516         if (!res) {
 517                 dlm_error(status);
 518                 goto leave;
 519         }
 520
 521         spin_lock(&res->spinlock);
 522         status = __dlm_lockres_state_to_status(res);
 523         spin_unlock(&res->spinlock);
 524
 525         if (status != DLM_NORMAL) {
 526                 mlog(0, "lockres recovering/migrating/in-progress\n");
 527                 goto leave;
 528         }
 529
 530         dlm_lock_attach_lockres(newlock, res);
 531
 532         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
 533 leave:
 534         if (status != DLM_NORMAL)
 535                 if (newlock)
 536                         dlm_lock_put(newlock);
 537
 538         if (res)
 539                 dlm_lockres_put(res);
 540
 541         dlm_put(dlm);
 542
 543         return status;
 544 }
 545
 546
 547 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
 548 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 549 {
 550         u64 tmpnode = node_num;
 551
 552         /* shift single byte of node num into top 8 bits */
 553         tmpnode <<= 56;
 554
 555         spin_lock(&dlm_cookie_lock);
 556         *cookie = (dlm_next_cookie | tmpnode);
 557         if (++dlm_next_cookie & 0xff00000000000000ull) {
 558                 mlog(0, "This node's cookie will now wrap!\n");
 559                 dlm_next_cookie = 1;
 560         }
 561         spin_unlock(&dlm_cookie_lock);
 562 }
 563
 564 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 565                         struct dlm_lockstatus *lksb, int flags,
 566                         const char *name, int namelen, dlm_astlockfunc_t *ast,
 567                         void *data, dlm_bastlockfunc_t *bast)
 568 {
 569         enum dlm_status status;
 570         struct dlm_lock_resource *res = NULL;
 571         struct dlm_lock *lock = NULL;
 572         int convert = 0, recovery = 0;
 573
 574         /* yes this function is a mess.
 575          * TODO: clean this up.  lots of common code in the
 576          *       lock and convert paths, especially in the retry blocks */
 577         if (!lksb) {
 578                 dlm_error(DLM_BADARGS);
 579                 return DLM_BADARGS;
 580         }
 581
 582         status = DLM_BADPARAM;
 583         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
 584                 dlm_error(status);
 585                 goto error;
 586         }
 587
 588         if (flags & ~LKM_VALID_FLAGS) {
 589                 dlm_error(status);
 590                 goto error;
 591         }
 592
 593         convert = (flags & LKM_CONVERT);
 594         recovery = (flags & LKM_RECOVERY);
 595
 596         if (recovery &&
 597             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 598                 dlm_error(status);
 599                 goto error;
 600         }
 601         if (convert && (flags & LKM_LOCAL)) {
 602                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
 603                 goto error;
 604         }
 605
 606         if (convert) {
 607                 /* CONVERT request */
 608
 609                 /* if converting, must pass in a valid dlm_lock */
 610                 lock = lksb->lockid;
 611                 if (!lock) {
 612                         mlog(ML_ERROR, "NULL lock pointer in convert "
 613                              "request\n");
 614                         goto error;
 615                 }
 616
 617                 res = lock->lockres;
 618                 if (!res) {
 619                         mlog(ML_ERROR, "NULL lockres pointer in convert "
 620                              "request\n");
 621                         goto error;
 622                 }
 623                 dlm_lockres_get(res);
 624
 625                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
 626                  * static after the original lock call.  convert requests will
 627                  * ensure that everything is the same, or return DLM_BADARGS.
 628                  * this means that DLM_DENIED_NOASTS will never be returned.
 629                  */
 630                 if (lock->lksb != lksb || lock->ast != ast ||
 631                     lock->bast != bast || lock->astdata != data) {
 632                         status = DLM_BADARGS;
 633                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
 634                              "astdata=%p\n", lksb, ast, bast, data);
 635                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
 636                              "astdata=%p\n", lock->lksb, lock->ast,
 637                              lock->bast, lock->astdata);
 638                         goto error;
 639                 }
 640 retry_convert:
 641                 dlm_wait_for_recovery(dlm);
 642
 643                 if (res->owner == dlm->node_num)
 644                         status = dlmconvert_master(dlm, res, lock, flags, mode);
 645                 else
 646                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
 647                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 648                     status == DLM_FORWARD) {
 649                         /* for now, see how this works without sleeping
 650                          * and just retry right away.  I suspect the reco
 651                          * or migration will complete fast enough that
 652                          * no waiting will be necessary */
 653                         mlog(0, "retrying convert with migration/recovery/"
 654                              "in-progress\n");
 655                         msleep(100);
 656                         goto retry_convert;
 657                 }
 658         } else {
 659                 u64 tmpcookie;
 660
 661                 /* LOCK request */
 662                 status = DLM_BADARGS;
 663                 if (!name) {
 664                         dlm_error(status);
 665                         goto error;
 666                 }
 667
 668                 status = DLM_IVBUFLEN;
 669                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 670                         dlm_error(status);
 671                         goto error;
 672                 }
 673
 674                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
 675                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
 676                 if (!lock) {
 677                         dlm_error(status);
 678                         goto error;
 679                 }
 680
 681                 if (!recovery)
 682                         dlm_wait_for_recovery(dlm);
 683
 684                 /* find or create the lock resource */
 685                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
 686                 if (!res) {
 687                         status = DLM_IVLOCKID;
 688                         dlm_error(status);
 689                         goto error;
 690                 }
 691
 692                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
 693                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
 694
 695                 dlm_lock_attach_lockres(lock, res);
 696                 lock->ast = ast;
 697                 lock->bast = bast;
 698                 lock->astdata = data;
 699
 700 retry_lock:
 701                 if (flags & LKM_VALBLK) {
 702                         mlog(0, "LKM_VALBLK passed by caller\n");
 703
 704                         /* LVB requests for non PR, PW or EX locks are
 705                          * ignored. */
 706                         if (mode < LKM_PRMODE)
 707                                 flags &= ~LKM_VALBLK;
 708                         else {
 709                                 flags |= LKM_GET_LVB;
 710                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
 711                         }
 712                 }
 713
 714                 if (res->owner == dlm->node_num)
 715                         status = dlmlock_master(dlm, res, lock, flags);
 716                 else
 717                         status = dlmlock_remote(dlm, res, lock, flags);
 718
 719                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 720                     status == DLM_FORWARD) {
 721                         mlog(0, "retrying lock with migration/"
 722                              "recovery/in progress\n");
 723                         msleep(100);
 724                         /* no waiting for dlm_reco_thread */
 725                         if (recovery) {
 726                                 if (status != DLM_RECOVERING)
 727                                         goto retry_lock;
 728
 729                                 mlog(0, "%s: got RECOVERING "
 730                                      "for $RECOVERY lock, master "
 731                                      "was %u\n", dlm->name,
 732                                      res->owner);
 733                                 /* wait to see the node go down, then
 734                                  * drop down and allow the lockres to
 735                                  * get cleaned up.  need to remaster. */
 736                                 dlm_wait_for_node_death(dlm, res->owner,
 737                                                 DLM_NODE_DEATH_WAIT_MAX);
 738                         } else {
 739                                 dlm_wait_for_recovery(dlm);
 740                                 goto retry_lock;
 741                         }
 742                 }
 743
 744                 if (status != DLM_NORMAL) {
 745                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 746                         if (status != DLM_NOTQUEUED)
 747                                 dlm_error(status);
 748                         goto error;
 749                 }
 750         }
 751
 752 error:
 753         if (status != DLM_NORMAL) {
 754                 if (lock && !convert)
 755                         dlm_lock_put(lock);
 756                 // this is kind of unnecessary
 757                 lksb->status = status;
 758         }
 759
 760         /* put lockres ref from the convert path
 761          * or from dlm_get_lock_resource */
 762         if (res)
 763                 dlm_lockres_put(res);
 764
 765         return status;
 766 }
 767 EXPORT_SYMBOL_GPL(dlmlock);