lustre/ldlm/ldlm_resource.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *   Author: Phil Schwan <phil@clusterfs.com>
   6  *   Author: Peter Braam <braam@clusterfs.com>
   7  *
   8  *   This file is part of the Lustre file system, http://www.lustre.org
   9  *   Lustre is a trademark of Cluster File Systems, Inc.
  10  *
  11  *   You may have signed or agreed to another license before downloading
  12  *   this software.  If so, you are bound by the terms and conditions
  13  *   of that agreement, and the following does not apply to you.  See the
  14  *   LICENSE file included with this distribution for more information.
  15  *
  16  *   If you did not agree to a different license, then this copy of Lustre
  17  *   is open source software; you can redistribute it and/or modify it
  18  *   under the terms of version 2 of the GNU General Public License as
  19  *   published by the Free Software Foundation.
  20  *
  21  *   In either case, Lustre is distributed in the hope that it will be
  22  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  23  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  24  *   license text for more details.
  25  */
  26
  27 #define DEBUG_SUBSYSTEM S_LDLM
  28 #ifdef __KERNEL__
  29 # include <lustre_dlm.h>
  30 #else
  31 # include <liblustre.h>
  32 #endif
  33
  34 #include <obd_class.h>
  35 #include "ldlm_internal.h"
  36
  37 cfs_mem_cache_t *ldlm_resource_slab, *ldlm_lock_slab;
  38
  39 atomic_t ldlm_srv_namespace_nr = ATOMIC_INIT(0);
  40 atomic_t ldlm_cli_namespace_nr = ATOMIC_INIT(0);
  41
  42 struct semaphore ldlm_srv_namespace_lock;
  43 struct list_head ldlm_srv_namespace_list =
  44         CFS_LIST_HEAD_INIT(ldlm_srv_namespace_list);
  45
  46 struct semaphore ldlm_cli_namespace_lock;
  47 struct list_head ldlm_cli_namespace_list =
  48         CFS_LIST_HEAD_INIT(ldlm_cli_namespace_list);
  49
  50 cfs_proc_dir_entry_t *ldlm_type_proc_dir = NULL;
  51 cfs_proc_dir_entry_t *ldlm_ns_proc_dir = NULL;
  52 cfs_proc_dir_entry_t *ldlm_svc_proc_dir = NULL;
  53
  54 #ifdef LPROCFS
  55 static int ldlm_proc_dump_ns(struct file *file, const char *buffer,
  56                              unsigned long count, void *data)
  57 {
  58         ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
  59         ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
  60         RETURN(count);
  61 }
  62
  63 int ldlm_proc_setup(void)
  64 {
  65         int rc;
  66         struct lprocfs_vars list[] = {
  67                 { "dump_namespaces", NULL, ldlm_proc_dump_ns, NULL },
  68                 { NULL }};
  69         ENTRY;
  70         LASSERT(ldlm_ns_proc_dir == NULL);
  71
  72         ldlm_type_proc_dir = lprocfs_register(OBD_LDLM_DEVICENAME,
  73                                               proc_lustre_root,
  74                                               NULL, NULL);
  75         if (IS_ERR(ldlm_type_proc_dir)) {
  76                 CERROR("LProcFS failed in ldlm-init\n");
  77                 rc = PTR_ERR(ldlm_type_proc_dir);
  78                 GOTO(err, rc);
  79         }
  80
  81         ldlm_ns_proc_dir = lprocfs_register("namespaces",
  82                                             ldlm_type_proc_dir,
  83                                             NULL, NULL);
  84         if (IS_ERR(ldlm_ns_proc_dir)) {
  85                 CERROR("LProcFS failed in ldlm-init\n");
  86                 rc = PTR_ERR(ldlm_ns_proc_dir);
  87                 GOTO(err_type, rc);
  88         }
  89
  90         ldlm_svc_proc_dir = lprocfs_register("services",
  91                                             ldlm_type_proc_dir,
  92                                             NULL, NULL);
  93         if (IS_ERR(ldlm_svc_proc_dir)) {
  94                 CERROR("LProcFS failed in ldlm-init\n");
  95                 rc = PTR_ERR(ldlm_svc_proc_dir);
  96                 GOTO(err_ns, rc);
  97         }
  98
  99         rc = lprocfs_add_vars(ldlm_type_proc_dir, list, NULL);
 100
 101         RETURN(0);
 102
 103 err_ns:
 104         lprocfs_remove(&ldlm_ns_proc_dir);
 105 err_type:
 106         lprocfs_remove(&ldlm_type_proc_dir);
 107 err:
 108         ldlm_svc_proc_dir = NULL;
 109         RETURN(rc);
 110 }
 111
 112 void ldlm_proc_cleanup(void)
 113 {
 114         if (ldlm_svc_proc_dir)
 115                 lprocfs_remove(&ldlm_svc_proc_dir);
 116
 117         if (ldlm_ns_proc_dir)
 118                 lprocfs_remove(&ldlm_ns_proc_dir);
 119
 120         if (ldlm_type_proc_dir)
 121                 lprocfs_remove(&ldlm_type_proc_dir);
 122 }
 123
 124 static int lprocfs_rd_lru_size(char *page, char **start, off_t off,
 125                                int count, int *eof, void *data)
 126 {
 127         struct ldlm_namespace *ns = data;
 128         __u32 *nr = &ns->ns_max_unused;
 129
 130         if (ns_connect_lru_resize(ns))
 131                 nr = &ns->ns_nr_unused;
 132         return lprocfs_rd_uint(page, start, off, count, eof, nr);
 133 }
 134
 135 static int lprocfs_wr_lru_size(struct file *file, const char *buffer,
 136                                unsigned long count, void *data)
 137 {
 138         struct ldlm_namespace *ns = data;
 139         char dummy[MAX_STRING_SIZE + 1], *end;
 140         unsigned long tmp;
 141         int lru_resize;
 142
 143         dummy[MAX_STRING_SIZE] = '\0';
 144         if (copy_from_user(dummy, buffer, MAX_STRING_SIZE))
 145                 return -EFAULT;
 146
 147         if (count == 6 && memcmp(dummy, "clear", 5) == 0) {
 148                 CDEBUG(D_DLMTRACE,
 149                        "dropping all unused locks from namespace %s\n",
 150                        ns->ns_name);
 151                 if (ns_connect_lru_resize(ns)) {
 152                         int canceled, unused  = ns->ns_nr_unused;
 153
 154                         /* Try to cancel all @ns_nr_unused locks. */
 155                         canceled = ldlm_cancel_lru(ns, unused, LDLM_SYNC);
 156                         if (canceled < unused) {
 157                                 CERROR("not all requested locks are canceled, "
 158                                        "requested: %d, canceled: %d\n", unused,
 159                                        canceled);
 160                                 return -EINVAL;
 161                         }
 162                 } else {
 163                         tmp = ns->ns_max_unused;
 164                         ns->ns_max_unused = 0;
 165                         ldlm_cancel_lru(ns, 0, LDLM_SYNC);
 166                         ns->ns_max_unused = tmp;
 167                 }
 168                 return count;
 169         }
 170
 171         tmp = simple_strtoul(dummy, &end, 0);
 172         if (dummy == end) {
 173                 CERROR("invalid value written\n");
 174                 return -EINVAL;
 175         }
 176         lru_resize = (tmp == 0);
 177
 178         if (ns_connect_lru_resize(ns)) {
 179                 if (!lru_resize)
 180                         ns->ns_max_unused = (unsigned int)tmp;
 181
 182                 if (tmp > ns->ns_nr_unused)
 183                         tmp = ns->ns_nr_unused;
 184                 tmp = ns->ns_nr_unused - tmp;
 185
 186                 CDEBUG(D_DLMTRACE, "changing namespace %s unused locks from %u to %u\n",
 187                        ns->ns_name, ns->ns_nr_unused, (unsigned int)tmp);
 188                 ldlm_cancel_lru(ns, (unsigned int)tmp, LDLM_ASYNC);
 189
 190                 if (!lru_resize) {
 191                         CDEBUG(D_DLMTRACE, "disable lru_resize for namespace %s\n",
 192                                ns->ns_name);
 193                         ns->ns_connect_flags &= ~OBD_CONNECT_LRU_RESIZE;
 194                 }
 195         } else {
 196                 CDEBUG(D_DLMTRACE, "changing namespace %s max_unused from %u to %u\n",
 197                        ns->ns_name, ns->ns_max_unused, (unsigned int)tmp);
 198                 ns->ns_max_unused = (unsigned int)tmp;
 199                 ldlm_cancel_lru(ns, 0, LDLM_ASYNC);
 200
 201                 /* Make sure that originally lru resize was supported before
 202                  * turning it on here. */
 203                 if (lru_resize &&
 204                     (ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) {
 205                         CDEBUG(D_DLMTRACE, "enable lru_resize for namespace %s\n",
 206                                ns->ns_name);
 207                         ns->ns_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208                 }
 209         }
 210
 211         return count;
 212 }
 213
 214 void ldlm_proc_namespace(struct ldlm_namespace *ns)
 215 {
 216         struct lprocfs_vars lock_vars[2];
 217         char lock_name[MAX_STRING_SIZE + 1];
 218
 219         LASSERT(ns != NULL);
 220         LASSERT(ns->ns_name != NULL);
 221
 222         lock_name[MAX_STRING_SIZE] = '\0';
 223
 224         memset(lock_vars, 0, sizeof(lock_vars));
 225         lock_vars[0].name = lock_name;
 226
 227         snprintf(lock_name, MAX_STRING_SIZE, "%s/resource_count", ns->ns_name);
 228         lock_vars[0].data = &ns->ns_refcount;
 229         lock_vars[0].read_fptr = lprocfs_rd_atomic;
 230         lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
 231
 232         snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_count", ns->ns_name);
 233         lock_vars[0].data = &ns->ns_locks;
 234         lock_vars[0].read_fptr = lprocfs_rd_atomic;
 235         lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
 236
 237         if (ns_is_client(ns)) {
 238                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_unused_count",
 239                          ns->ns_name);
 240                 lock_vars[0].data = &ns->ns_nr_unused;
 241                 lock_vars[0].read_fptr = lprocfs_rd_uint;
 242                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
 243
 244                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_size",
 245                          ns->ns_name);
 246                 lock_vars[0].data = ns;
 247                 lock_vars[0].read_fptr = lprocfs_rd_lru_size;
 248                 lock_vars[0].write_fptr = lprocfs_wr_lru_size;
 249                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
 250
 251                 snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_max_age",
 252                          ns->ns_name);
 253                 lock_vars[0].data = &ns->ns_max_age;
 254                 lock_vars[0].read_fptr = lprocfs_rd_uint;
 255                 lock_vars[0].write_fptr = lprocfs_wr_uint;
 256                 lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
 257
 258         }
 259 }
 260 #undef MAX_STRING_SIZE
 261 #else
 262 #define ldlm_proc_namespace(ns) do {} while (0)
 263 #endif /* LPROCFS */
 264
 265 struct ldlm_namespace *ldlm_namespace_new(char *name, ldlm_side_t client,
 266                                           ldlm_appetite_t apt)
 267 {
 268         struct ldlm_namespace *ns = NULL;
 269         struct list_head *bucket;
 270         int rc, idx, namelen;
 271         ENTRY;
 272
 273         rc = ldlm_get_ref();
 274         if (rc) {
 275                 CERROR("ldlm_get_ref failed: %d\n", rc);
 276                 RETURN(NULL);
 277         }
 278
 279         OBD_ALLOC_PTR(ns);
 280         if (!ns)
 281                 GOTO(out_ref, NULL);
 282
 283         OBD_VMALLOC(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
 284         if (!ns->ns_hash)
 285                 GOTO(out_ns, NULL);
 286
 287         ns->ns_appetite = apt;
 288         namelen = strlen(name);
 289         OBD_ALLOC(ns->ns_name, namelen + 1);
 290         if (!ns->ns_name)
 291                 GOTO(out_hash, NULL);
 292
 293         strcpy(ns->ns_name, name);
 294
 295         CFS_INIT_LIST_HEAD(&ns->ns_root_list);
 296         ns->ns_refcount = 0;
 297         ns->ns_client = client;
 298         spin_lock_init(&ns->ns_hash_lock);
 299         atomic_set(&ns->ns_locks, 0);
 300         ns->ns_resources = 0;
 301         cfs_waitq_init(&ns->ns_waitq);
 302
 303         for (bucket = ns->ns_hash + RES_HASH_SIZE - 1; bucket >= ns->ns_hash;
 304              bucket--)
 305                 CFS_INIT_LIST_HEAD(bucket);
 306
 307         CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
 308         ns->ns_nr_unused = 0;
 309         ns->ns_max_unused = LDLM_DEFAULT_LRU_SIZE;
 310         ns->ns_max_age = LDLM_DEFAULT_MAX_ALIVE;
 311         spin_lock_init(&ns->ns_unused_lock);
 312         ns->ns_orig_connect_flags = 0;
 313         ns->ns_connect_flags = 0;
 314         ldlm_proc_namespace(ns);
 315
 316         idx = atomic_read(ldlm_namespace_nr(client));
 317         rc = ldlm_pool_init(&ns->ns_pool, ns, idx, client);
 318         if (rc) {
 319                 CERROR("Can't initialize lock pool, rc %d\n", rc);
 320                 GOTO(out_proc, rc);
 321         }
 322
 323         mutex_down(ldlm_namespace_lock(client));
 324         list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
 325         atomic_inc(ldlm_namespace_nr(client));
 326         mutex_up(ldlm_namespace_lock(client));
 327
 328         RETURN(ns);
 329 out_proc:
 330         ldlm_namespace_cleanup(ns, 0);
 331         OBD_FREE(ns->ns_name, namelen + 1);
 332 out_hash:
 333         OBD_VFREE(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
 334 out_ns:
 335         OBD_FREE_PTR(ns);
 336 out_ref:
 337         ldlm_put_ref(0);
 338         RETURN(NULL);
 339 }
 340
 341 extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
 342
 343 /* If flags contains FL_LOCAL_ONLY, don't try to tell the server, just cleanup.
 344  * This is currently only used for recovery, and we make certain assumptions
 345  * as a result--notably, that we shouldn't cancel locks with refs. -phil
 346  *
 347  * Called with the ns_lock held. */
 348 static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
 349                              int flags)
 350 {
 351         struct list_head *tmp;
 352         int rc = 0, client = ns_is_client(res->lr_namespace);
 353         int local_only = (flags & LDLM_FL_LOCAL_ONLY);
 354         ENTRY;
 355
 356
 357         do {
 358                 struct ldlm_lock *lock = NULL;
 359
 360                 /* first, we look for non-cleaned-yet lock
 361                  * all cleaned locks are marked by CLEANED flag */
 362                 lock_res(res);
 363                 list_for_each(tmp, q) {
 364                         lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 365                         if (lock->l_flags & LDLM_FL_CLEANED) {
 366                                 lock = NULL;
 367                                 continue;
 368                         }
 369                         LDLM_LOCK_GET(lock);
 370                         lock->l_flags |= LDLM_FL_CLEANED;
 371                         break;
 372                 }
 373
 374                 if (lock == NULL) {
 375                         unlock_res(res);
 376                         break;
 377                 }
 378
 379                 /* Set CBPENDING so nothing in the cancellation path
 380                  * can match this lock */
 381                 lock->l_flags |= LDLM_FL_CBPENDING;
 382                 lock->l_flags |= LDLM_FL_FAILED;
 383                 lock->l_flags |= flags;
 384
 385                 /* ... without sending a CANCEL message for local_only. */
 386                 if (local_only)
 387                         lock->l_flags |= LDLM_FL_LOCAL_ONLY;
 388
 389                 if (local_only && (lock->l_readers || lock->l_writers)) {
 390                         /* This is a little bit gross, but much better than the
 391                          * alternative: pretend that we got a blocking AST from
 392                          * the server, so that when the lock is decref'd, it
 393                          * will go away ... */
 394                         unlock_res(res);
 395                         LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
 396                         if (lock->l_completion_ast)
 397                                 lock->l_completion_ast(lock, 0, NULL);
 398                         LDLM_LOCK_PUT(lock);
 399                         continue;
 400                 }
 401
 402                 if (client) {
 403                         struct lustre_handle lockh;
 404
 405                         unlock_res(res);
 406                         ldlm_lock2handle(lock, &lockh);
 407                         rc = ldlm_cli_cancel(&lockh);
 408                         if (rc)
 409                                 CERROR("ldlm_cli_cancel: %d\n", rc);
 410                 } else {
 411                         ldlm_resource_unlink_lock(lock);
 412                         unlock_res(res);
 413                         LDLM_DEBUG(lock, "Freeing a lock still held by a "
 414                                    "client node");
 415                         ldlm_lock_destroy(lock);
 416                 }
 417                 LDLM_LOCK_PUT(lock);
 418         } while (1);
 419
 420         EXIT;
 421 }
 422
 423 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags)
 424 {
 425         struct list_head *tmp;
 426         int i;
 427
 428         if (ns == NULL) {
 429                 CDEBUG(D_INFO, "NULL ns, skipping cleanup\n");
 430                 return ELDLM_OK;
 431         }
 432
 433         for (i = 0; i < RES_HASH_SIZE; i++) {
 434                 spin_lock(&ns->ns_hash_lock);
 435                 tmp = ns->ns_hash[i].next;
 436                 while (tmp != &(ns->ns_hash[i])) {
 437                         struct ldlm_resource *res;
 438                         res = list_entry(tmp, struct ldlm_resource, lr_hash);
 439                         ldlm_resource_getref(res);
 440                         spin_unlock(&ns->ns_hash_lock);
 441
 442                         cleanup_resource(res, &res->lr_granted, flags);
 443                         cleanup_resource(res, &res->lr_converting, flags);
 444                         cleanup_resource(res, &res->lr_waiting, flags);
 445
 446                         spin_lock(&ns->ns_hash_lock);
 447                         tmp  = tmp->next;
 448
 449                         /* XXX: former stuff caused issues in case of race
 450                          * between ldlm_namespace_cleanup() and lockd() when
 451                          * client gets blocking ast when lock gets distracted by
 452                          * server. This is 1_4 branch solution, let's see how
 453                          * will it behave. */
 454                         if (!ldlm_resource_putref_locked(res))
 455                                 CDEBUG(D_INFO,
 456                                        "Namespace %s resource refcount nonzero "
 457                                        "(%d) after lock cleanup; forcing cleanup.\n",
 458                                        ns->ns_name, atomic_read(&res->lr_refcount));
 459                 }
 460                 spin_unlock(&ns->ns_hash_lock);
 461         }
 462
 463         return ELDLM_OK;
 464 }
 465
 466 /* Cleanup, but also free, the namespace */
 467 int ldlm_namespace_free(struct ldlm_namespace *ns, int force)
 468 {
 469         ENTRY;
 470         if (!ns)
 471                 RETURN(ELDLM_OK);
 472
 473         mutex_down(ldlm_namespace_lock(ns->ns_client));
 474         list_del(&ns->ns_list_chain);
 475         atomic_dec(ldlm_namespace_nr(ns->ns_client));
 476         ldlm_pool_fini(&ns->ns_pool);
 477         mutex_up(ldlm_namespace_lock(ns->ns_client));
 478
 479         /* At shutdown time, don't call the cancellation callback */
 480         ldlm_namespace_cleanup(ns, 0);
 481
 482 #ifdef LPROCFS
 483         {
 484                 struct proc_dir_entry *dir;
 485                 dir = lprocfs_srch(ldlm_ns_proc_dir, ns->ns_name);
 486                 if (dir == NULL) {
 487                         CERROR("dlm namespace %s has no procfs dir?\n",
 488                                ns->ns_name);
 489                 } else {
 490                         lprocfs_remove(&dir);
 491                 }
 492         }
 493 #endif
 494
 495         if (ns->ns_refcount > 0) {
 496                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
 497                 int rc;
 498                 CDEBUG(D_DLMTRACE,
 499                        "dlm namespace %s free waiting on refcount %d\n",
 500                        ns->ns_name, ns->ns_refcount);
 501                 rc = l_wait_event(ns->ns_waitq,
 502                                   ns->ns_refcount == 0, &lwi);
 503                 if (ns->ns_refcount)
 504                         LCONSOLE_ERROR_MSG(0x139, "Lock manager: wait for %s "
 505                                            "namespace cleanup aborted with %d "
 506                                            "resources in use. (%d)\nI'm going "
 507                                            "to try to clean up anyway, but I "
 508                                            "might need a reboot of this node.\n",
 509                                             ns->ns_name, (int) ns->ns_refcount,
 510                                             rc);
 511                 CDEBUG(D_DLMTRACE,
 512                        "dlm namespace %s free done waiting\n", ns->ns_name);
 513         }
 514
 515         OBD_VFREE(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
 516         OBD_FREE(ns->ns_name, strlen(ns->ns_name) + 1);
 517         OBD_FREE_PTR(ns);
 518
 519         ldlm_put_ref(force);
 520
 521         RETURN(ELDLM_OK);
 522 }
 523
 524 void ldlm_namespace_get_nolock(struct ldlm_namespace *ns)
 525 {
 526         LASSERT(ns->ns_refcount >= 0);
 527         ns->ns_refcount++;
 528 }
 529
 530 void ldlm_namespace_get(struct ldlm_namespace *ns)
 531 {
 532         spin_lock(&ns->ns_hash_lock);
 533         ldlm_namespace_get_nolock(ns);
 534         spin_unlock(&ns->ns_hash_lock);
 535 }
 536
 537 void ldlm_namespace_put_nolock(struct ldlm_namespace *ns, int wakeup)
 538 {
 539         LASSERT(ns->ns_refcount > 0);
 540         ns->ns_refcount--;
 541         if (ns->ns_refcount == 0 && wakeup)
 542                 wake_up(&ns->ns_waitq);
 543 }
 544
 545 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup)
 546 {
 547         spin_lock(&ns->ns_hash_lock);
 548         ldlm_namespace_put_nolock(ns, wakeup);
 549         spin_unlock(&ns->ns_hash_lock);
 550 }
 551
 552 /* Should be called under ldlm_namespace_lock(client) taken */
 553 void ldlm_namespace_move(struct ldlm_namespace *ns, ldlm_side_t client)
 554 {
 555         LASSERT(!list_empty(&ns->ns_list_chain));
 556         LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
 557         list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
 558 }
 559
 560 /* Should be called under ldlm_namespace_lock(client) taken */
 561 struct ldlm_namespace *ldlm_namespace_first(ldlm_side_t client)
 562 {
 563         LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
 564         LASSERT(!list_empty(ldlm_namespace_list(client)));
 565         return container_of(ldlm_namespace_list(client)->next,
 566                 struct ldlm_namespace, ns_list_chain);
 567 }
 568 static __u32 ldlm_hash_fn(struct ldlm_resource *parent,
 569                           const struct ldlm_res_id *name)
 570 {
 571         __u32 hash = 0;
 572         int i;
 573
 574         for (i = 0; i < RES_NAME_SIZE; i++)
 575                 hash += name->name[i];
 576
 577         hash += (__u32)((unsigned long)parent >> 4);
 578
 579         return (hash & RES_HASH_MASK);
 580 }
 581
 582 static struct ldlm_resource *ldlm_resource_new(void)
 583 {
 584         struct ldlm_resource *res;
 585
 586         OBD_SLAB_ALLOC(res, ldlm_resource_slab, CFS_ALLOC_IO, sizeof *res);
 587         if (res == NULL)
 588                 return NULL;
 589
 590         memset(res, 0, sizeof(*res));
 591
 592         CFS_INIT_LIST_HEAD(&res->lr_children);
 593         CFS_INIT_LIST_HEAD(&res->lr_childof);
 594         CFS_INIT_LIST_HEAD(&res->lr_granted);
 595         CFS_INIT_LIST_HEAD(&res->lr_converting);
 596         CFS_INIT_LIST_HEAD(&res->lr_waiting);
 597         atomic_set(&res->lr_refcount, 1);
 598         spin_lock_init(&res->lr_lock);
 599
 600         /* one who creates the resource must unlock
 601          * the semaphore after lvb initialization */
 602         init_MUTEX_LOCKED(&res->lr_lvb_sem);
 603
 604         return res;
 605 }
 606
 607 /* must be called with hash lock held */
 608 static struct ldlm_resource *
 609 ldlm_resource_find(struct ldlm_namespace *ns, const struct ldlm_res_id *name,
 610                    __u32 hash)
 611 {
 612         struct list_head *bucket, *tmp;
 613         struct ldlm_resource *res;
 614
 615         LASSERT_SPIN_LOCKED(&ns->ns_hash_lock);
 616         bucket = ns->ns_hash + hash;
 617
 618         list_for_each(tmp, bucket) {
 619                 res = list_entry(tmp, struct ldlm_resource, lr_hash);
 620                 if (memcmp(&res->lr_name, name, sizeof(res->lr_name)) == 0)
 621                         return res;
 622         }
 623
 624         return NULL;
 625 }
 626
 627 /* Args: locked namespace
 628  * Returns: newly-allocated, referenced, unlocked resource */
 629 static struct ldlm_resource *
 630 ldlm_resource_add(struct ldlm_namespace *ns, struct ldlm_resource *parent,
 631                   const struct ldlm_res_id *name, __u32 hash, ldlm_type_t type)
 632 {
 633         struct list_head *bucket;
 634         struct ldlm_resource *res, *old_res;
 635         ENTRY;
 636
 637         LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
 638                  "type: %d\n", type);
 639
 640         res = ldlm_resource_new();
 641         if (!res)
 642                 RETURN(NULL);
 643
 644         res->lr_name = *name;
 645         res->lr_namespace = ns;
 646         res->lr_type = type;
 647         res->lr_most_restr = LCK_NL;
 648
 649         spin_lock(&ns->ns_hash_lock);
 650         old_res = ldlm_resource_find(ns, name, hash);
 651         if (old_res) {
 652                 /* someone won the race and added the resource before */
 653                 ldlm_resource_getref(old_res);
 654                 spin_unlock(&ns->ns_hash_lock);
 655                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
 656                 /* synchronize WRT resource creation */
 657                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
 658                         down(&old_res->lr_lvb_sem);
 659                         up(&old_res->lr_lvb_sem);
 660                 }
 661                 RETURN(old_res);
 662         }
 663
 664         /* we won! let's add the resource */
 665         bucket = ns->ns_hash + hash;
 666         list_add(&res->lr_hash, bucket);
 667         ns->ns_resources++;
 668         ldlm_namespace_get_nolock(ns);
 669
 670         if (parent == NULL) {
 671                 list_add(&res->lr_childof, &ns->ns_root_list);
 672         } else {
 673                 res->lr_parent = parent;
 674                 list_add(&res->lr_childof, &parent->lr_children);
 675         }
 676         spin_unlock(&ns->ns_hash_lock);
 677
 678         if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
 679                 int rc;
 680
 681                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CREATE_RESOURCE, 2);
 682                 rc = ns->ns_lvbo->lvbo_init(res);
 683                 if (rc)
 684                         CERROR("lvbo_init failed for resource "
 685                                LPU64": rc %d\n", name->name[0], rc);
 686                 /* we create resource with locked lr_lvb_sem */
 687                 up(&res->lr_lvb_sem);
 688         }
 689
 690         RETURN(res);
 691 }
 692
 693 /* Args: unlocked namespace
 694  * Locks: takes and releases ns->ns_lock and res->lr_lock
 695  * Returns: referenced, unlocked ldlm_resource or NULL */
 696 struct ldlm_resource *
 697 ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
 698                   const struct ldlm_res_id *name, ldlm_type_t type, int create)
 699 {
 700         __u32 hash = ldlm_hash_fn(parent, name);
 701         struct ldlm_resource *res = NULL;
 702         ENTRY;
 703
 704         LASSERT(ns != NULL);
 705         LASSERT(ns->ns_hash != NULL);
 706         LASSERT(name->name[0] != 0);
 707
 708         spin_lock(&ns->ns_hash_lock);
 709         res = ldlm_resource_find(ns, name, hash);
 710         if (res) {
 711                 ldlm_resource_getref(res);
 712                 spin_unlock(&ns->ns_hash_lock);
 713                 /* synchronize WRT resource creation */
 714                 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
 715                         down(&res->lr_lvb_sem);
 716                         up(&res->lr_lvb_sem);
 717                 }
 718                 RETURN(res);
 719         }
 720         spin_unlock(&ns->ns_hash_lock);
 721
 722         if (create == 0)
 723                 RETURN(NULL);
 724
 725         res = ldlm_resource_add(ns, parent, name, hash, type);
 726         RETURN(res);
 727 }
 728
 729 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
 730 {
 731         LASSERT(res != NULL);
 732         LASSERT(res != LP_POISON);
 733         atomic_inc(&res->lr_refcount);
 734         CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
 735                atomic_read(&res->lr_refcount));
 736         return res;
 737 }
 738
 739 void __ldlm_resource_putref_final(struct ldlm_resource *res)
 740 {
 741         struct ldlm_namespace *ns = res->lr_namespace;
 742
 743         LASSERT_SPIN_LOCKED(&ns->ns_hash_lock);
 744
 745         if (!list_empty(&res->lr_granted)) {
 746                 ldlm_resource_dump(D_ERROR, res);
 747                 LBUG();
 748         }
 749
 750         if (!list_empty(&res->lr_converting)) {
 751                 ldlm_resource_dump(D_ERROR, res);
 752                 LBUG();
 753         }
 754
 755         if (!list_empty(&res->lr_waiting)) {
 756                 ldlm_resource_dump(D_ERROR, res);
 757                 LBUG();
 758         }
 759
 760         if (!list_empty(&res->lr_children)) {
 761                 ldlm_resource_dump(D_ERROR, res);
 762                 LBUG();
 763         }
 764
 765         /* Pass 0 here to not wake ->ns_waitq up yet, we will do it few
 766          * lines below when all children are freed. */
 767         ldlm_namespace_put_nolock(ns, 0);
 768         list_del_init(&res->lr_hash);
 769         list_del_init(&res->lr_childof);
 770
 771         ns->ns_resources--;
 772         if (ns->ns_resources == 0)
 773                 wake_up(&ns->ns_waitq);
 774 }
 775
 776 /* Returns 1 if the resource was freed, 0 if it remains. */
 777 int ldlm_resource_putref(struct ldlm_resource *res)
 778 {
 779         struct ldlm_namespace *ns = res->lr_namespace;
 780         int rc = 0;
 781         ENTRY;
 782
 783         CDEBUG(D_INFO, "putref res: %p count: %d\n", res,
 784                atomic_read(&res->lr_refcount) - 1);
 785         LASSERTF(atomic_read(&res->lr_refcount) > 0, "%d",
 786                  atomic_read(&res->lr_refcount));
 787         LASSERTF(atomic_read(&res->lr_refcount) < LI_POISON, "%d",
 788                  atomic_read(&res->lr_refcount));
 789
 790         if (atomic_dec_and_lock(&res->lr_refcount, &ns->ns_hash_lock)) {
 791                 __ldlm_resource_putref_final(res);
 792                 spin_unlock(&ns->ns_hash_lock);
 793                 if (res->lr_lvb_data)
 794                         OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
 795                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
 796                 rc = 1;
 797         }
 798
 799         RETURN(rc);
 800 }
 801
 802 /* Returns 1 if the resource was freed, 0 if it remains. */
 803 int ldlm_resource_putref_locked(struct ldlm_resource *res)
 804 {
 805         int rc = 0;
 806         ENTRY;
 807
 808         CDEBUG(D_INFO, "putref res: %p count: %d\n", res,
 809                atomic_read(&res->lr_refcount) - 1);
 810         LASSERT(atomic_read(&res->lr_refcount) > 0);
 811         LASSERT(atomic_read(&res->lr_refcount) < LI_POISON);
 812
 813         LASSERT(atomic_read(&res->lr_refcount) >= 0);
 814         if (atomic_dec_and_test(&res->lr_refcount)) {
 815                 __ldlm_resource_putref_final(res);
 816                 if (res->lr_lvb_data)
 817                         OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
 818                 OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
 819                 rc = 1;
 820         }
 821
 822         RETURN(rc);
 823 }
 824
 825 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
 826                             struct ldlm_lock *lock)
 827 {
 828         check_res_locked(res);
 829
 830         ldlm_resource_dump(D_OTHER, res);
 831         CDEBUG(D_OTHER, "About to add this lock:\n");
 832         ldlm_lock_dump(D_OTHER, lock, 0);
 833
 834         if (lock->l_destroyed) {
 835                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 836                 return;
 837         }
 838
 839         LASSERT(list_empty(&lock->l_res_link));
 840
 841         list_add_tail(&lock->l_res_link, head);
 842 }
 843
 844 void ldlm_resource_insert_lock_after(struct ldlm_lock *original,
 845                                      struct ldlm_lock *new)
 846 {
 847         struct ldlm_resource *res = original->l_resource;
 848
 849         check_res_locked(res);
 850
 851         ldlm_resource_dump(D_OTHER, res);
 852         CDEBUG(D_OTHER, "About to insert this lock after %p:\n", original);
 853         ldlm_lock_dump(D_OTHER, new, 0);
 854
 855         if (new->l_destroyed) {
 856                 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
 857                 goto out;
 858         }
 859
 860         LASSERT(list_empty(&new->l_res_link));
 861
 862         list_add(&new->l_res_link, &original->l_res_link);
 863  out:;
 864 }
 865
 866 void ldlm_resource_unlink_lock(struct ldlm_lock *lock)
 867 {
 868         check_res_locked(lock->l_resource);
 869         ldlm_unlink_lock_skiplist(lock);
 870         list_del_init(&lock->l_res_link);
 871 }
 872
 873 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
 874 {
 875         desc->lr_type = res->lr_type;
 876         desc->lr_name = res->lr_name;
 877 }
 878
 879 void ldlm_dump_all_namespaces(ldlm_side_t client, int level)
 880 {
 881         struct list_head *tmp;
 882
 883         if (!((libcfs_debug | D_ERROR) & level))
 884                 return;
 885
 886         mutex_down(ldlm_namespace_lock(client));
 887
 888         list_for_each(tmp, ldlm_namespace_list(client)) {
 889                 struct ldlm_namespace *ns;
 890                 ns = list_entry(tmp, struct ldlm_namespace, ns_list_chain);
 891                 ldlm_namespace_dump(level, ns);
 892         }
 893
 894         mutex_up(ldlm_namespace_lock(client));
 895 }
 896
 897 void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
 898 {
 899         struct list_head *tmp;
 900
 901         if (!((libcfs_debug | D_ERROR) & level))
 902                 return;
 903
 904         CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
 905                ns->ns_name, ns->ns_refcount,
 906                ns_is_client(ns) ? "client" : "server");
 907
 908         if (cfs_time_before(cfs_time_current(), ns->ns_next_dump))
 909                 return;
 910
 911         spin_lock(&ns->ns_hash_lock);
 912         tmp = ns->ns_root_list.next;
 913         while (tmp != &ns->ns_root_list) {
 914                 struct ldlm_resource *res;
 915                 res = list_entry(tmp, struct ldlm_resource, lr_childof);
 916
 917                 ldlm_resource_getref(res);
 918                 spin_unlock(&ns->ns_hash_lock);
 919
 920                 lock_res(res);
 921                 ldlm_resource_dump(level, res);
 922                 unlock_res(res);
 923
 924                 spin_lock(&ns->ns_hash_lock);
 925                 tmp = tmp->next;
 926                 ldlm_resource_putref_locked(res);
 927         }
 928         ns->ns_next_dump = cfs_time_shift(10);
 929         spin_unlock(&ns->ns_hash_lock);
 930 }
 931
 932 void ldlm_resource_dump(int level, struct ldlm_resource *res)
 933 {
 934         struct list_head *tmp;
 935         int pos;
 936
 937         CLASSERT(RES_NAME_SIZE == 4);
 938
 939         if (!((libcfs_debug | D_ERROR) & level))
 940                 return;
 941
 942         CDEBUG(level, "--- Resource: %p ("LPU64"/"LPU64"/"LPU64"/"LPU64
 943                ") (rc: %d)\n", res, res->lr_name.name[0], res->lr_name.name[1],
 944                res->lr_name.name[2], res->lr_name.name[3],
 945                atomic_read(&res->lr_refcount));
 946
 947         if (!list_empty(&res->lr_granted)) {
 948                 pos = 0;
 949                 CDEBUG(level, "Granted locks:\n");
 950                 list_for_each(tmp, &res->lr_granted) {
 951                         struct ldlm_lock *lock;
 952                         lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 953                         ldlm_lock_dump(level, lock, ++pos);
 954                 }
 955         }
 956         if (!list_empty(&res->lr_converting)) {
 957                 pos = 0;
 958                 CDEBUG(level, "Converting locks:\n");
 959                 list_for_each(tmp, &res->lr_converting) {
 960                         struct ldlm_lock *lock;
 961                         lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 962                         ldlm_lock_dump(level, lock, ++pos);
 963                 }
 964         }
 965         if (!list_empty(&res->lr_waiting)) {
 966                 pos = 0;
 967                 CDEBUG(level, "Waiting locks:\n");
 968                 list_for_each(tmp, &res->lr_waiting) {
 969                         struct ldlm_lock *lock;
 970                         lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 971                         ldlm_lock_dump(level, lock, ++pos);
 972                 }
 973         }
 974 }