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