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 Cluster File Systems, Inc.
   5  *
   6  * This code is issued under the GNU General Public License.
   7  * See the file COPYING in this distribution
   8  *
   9  * by Cluster File Systems, Inc.
  10  */
  11
  12 #define DEBUG_SUBSYSTEM S_LDLM
  13
  14 #include <linux/lustre_dlm.h>
  15 #include <linux/obd_class.h>
  16
  17 kmem_cache_t *ldlm_resource_slab, *ldlm_lock_slab;
  18
  19 spinlock_t ldlm_namespace_lock = SPIN_LOCK_UNLOCKED;
  20 struct list_head ldlm_namespace_list = LIST_HEAD_INIT(ldlm_namespace_list);
  21 static struct proc_dir_entry *ldlm_ns_proc_dir = NULL;
  22
  23 int ldlm_proc_setup(struct obd_device *obd)
  24 {
  25         ENTRY;
  26
  27         if (obd->obd_proc_entry == NULL)
  28                 RETURN(-EINVAL);
  29
  30         ldlm_ns_proc_dir = proc_mkdir("namespaces", obd->obd_proc_entry);
  31         if (ldlm_ns_proc_dir == NULL) {
  32                 CERROR("Couldn't create /proc/lustre/ldlm/namespaces\n");
  33                 RETURN(-EPERM);
  34         }
  35         RETURN(0);
  36 }
  37
  38 void ldlm_proc_cleanup(struct obd_device *obd)
  39 {
  40         proc_lustre_remove_obd_entry("namespaces", obd);
  41 }
  42
  43 /* FIXME: This can go away when we start to really use lprocfs */
  44 static int lprocfs_ll_rd(char *page, char **start, off_t off,
  45                          int count, int *eof, void *data)
  46 {
  47         int len;
  48         __u64 *temp = (__u64 *)data;
  49
  50         len = snprintf(page, count, "%Lu\n", *temp);
  51
  52         return len;
  53 }
  54
  55 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 client)
  56 {
  57         struct ldlm_namespace *ns = NULL;
  58         struct list_head *bucket;
  59         struct proc_dir_entry *proc_entry;
  60
  61         OBD_ALLOC(ns, sizeof(*ns));
  62         if (!ns) {
  63                 LBUG();
  64                 GOTO(out, NULL);
  65         }
  66
  67         ns->ns_hash = vmalloc(sizeof(*ns->ns_hash) * RES_HASH_SIZE);
  68         if (!ns->ns_hash) {
  69                 LBUG();
  70                 GOTO(out, ns);
  71         }
  72         obd_memory += sizeof(*ns->ns_hash) * RES_HASH_SIZE;
  73
  74         OBD_ALLOC(ns->ns_name, strlen(name) + 1);
  75         if (!ns->ns_name) {
  76                 LBUG();
  77                 GOTO(out, ns);
  78         }
  79         strcpy(ns->ns_name, name);
  80
  81         INIT_LIST_HEAD(&ns->ns_root_list);
  82         ns->ns_refcount = 0;
  83         ns->ns_client = client;
  84         spin_lock_init(&ns->ns_counter_lock);
  85         ns->ns_locks = 0;
  86         ns->ns_resources = 0;
  87
  88         for (bucket = ns->ns_hash + RES_HASH_SIZE - 1; bucket >= ns->ns_hash;
  89              bucket--)
  90                 INIT_LIST_HEAD(bucket);
  91
  92         spin_lock(&ldlm_namespace_lock);
  93         list_add(&ns->ns_list_chain, &ldlm_namespace_list);
  94         spin_unlock(&ldlm_namespace_lock);
  95
  96         ns->ns_proc_dir = proc_mkdir(ns->ns_name, ldlm_ns_proc_dir);
  97         if (ns->ns_proc_dir == NULL)
  98                 CERROR("Unable to create proc directory for namespace.\n");
  99         proc_entry = create_proc_entry("resource_count", 0444, ns->ns_proc_dir);
 100         proc_entry->read_proc = lprocfs_ll_rd;
 101         proc_entry->data = &ns->ns_resources;
 102         proc_entry = create_proc_entry("lock_count", 0444, ns->ns_proc_dir);
 103         proc_entry->read_proc = lprocfs_ll_rd;
 104         proc_entry->data = &ns->ns_locks;
 105
 106         RETURN(ns);
 107
 108  out:
 109         if (ns && ns->ns_hash) {
 110                 vfree(ns->ns_hash);
 111                 obd_memory -= sizeof(*ns->ns_hash) * RES_HASH_SIZE;
 112         }
 113         if (ns && ns->ns_name)
 114                 OBD_FREE(ns->ns_name, strlen(name) + 1);
 115         if (ns)
 116                 OBD_FREE(ns, sizeof(*ns));
 117         return NULL;
 118 }
 119
 120 extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
 121
 122 /* If 'local_only' is true, don't try to tell the server, just cleanup. */
 123 static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
 124                              int local_only)
 125 {
 126         struct list_head *tmp, *pos;
 127         int rc = 0, client = res->lr_namespace->ns_client;
 128         ENTRY;
 129
 130         list_for_each_safe(tmp, pos, q) {
 131                 struct ldlm_lock *lock;
 132                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 133                 LDLM_LOCK_GET(lock);
 134
 135                 if (client) {
 136                         struct lustre_handle lockh;
 137                         ldlm_lock2handle(lock, &lockh);
 138                         if (!local_only) {
 139                                 rc = ldlm_cli_cancel(&lockh);
 140                                 if (rc)
 141                                         CERROR("ldlm_cli_cancel: %d\n", rc);
 142                         }
 143                         /* Force local cleanup on errors, too. */
 144                         if (local_only || rc != ELDLM_OK)
 145                                 ldlm_lock_cancel(lock);
 146                 } else {
 147                         LDLM_DEBUG(lock, "Freeing a lock still held by a "
 148                                    "client node.\n");
 149
 150                         ldlm_resource_unlink_lock(lock);
 151                         ldlm_lock_destroy(lock);
 152                 }
 153                 LDLM_LOCK_PUT(lock);
 154         }
 155 }
 156
 157 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int local_only)
 158 {
 159         int i;
 160
 161         l_lock(&ldlm_everything_lock);
 162         for (i = 0; i < RES_HASH_SIZE; i++) {
 163                 struct list_head *tmp, *pos;
 164                 list_for_each_safe(tmp, pos, &(ns->ns_hash[i])) {
 165                         struct ldlm_resource *res;
 166                         res = list_entry(tmp, struct ldlm_resource, lr_hash);
 167                         ldlm_resource_getref(res);
 168
 169                         cleanup_resource(res, &res->lr_granted, local_only);
 170                         cleanup_resource(res, &res->lr_converting, local_only);
 171                         cleanup_resource(res, &res->lr_waiting, local_only);
 172
 173                         /* XXX this is a bit counter-intuitive and should
 174                          * probably be cleaner: don't force cleanup if we're
 175                          * local_only (which is only used by recovery).  We
 176                          * probably still have outstanding lock refs which
 177                          * reference these resources. -phil */
 178                         if (!ldlm_resource_put(res) && !local_only) {
 179                                 CERROR("Resource refcount nonzero (%d) after "
 180                                        "lock cleanup; forcing cleanup.\n",
 181                                        atomic_read(&res->lr_refcount));
 182                                 ldlm_resource_dump(res);
 183                                 atomic_set(&res->lr_refcount, 1);
 184                                 ldlm_resource_put(res);
 185                         }
 186                 }
 187         }
 188         l_unlock(&ldlm_everything_lock);
 189
 190         return ELDLM_OK;
 191 }
 192
 193 /* Cleanup, but also free, the namespace */
 194 int ldlm_namespace_free(struct ldlm_namespace *ns)
 195 {
 196         if (!ns)
 197                 RETURN(ELDLM_OK);
 198
 199         spin_lock(&ldlm_namespace_lock);
 200         list_del(&ns->ns_list_chain);
 201         remove_proc_entry("resource_count", ns->ns_proc_dir);
 202         remove_proc_entry("lock_count", ns->ns_proc_dir);
 203         remove_proc_entry(ns->ns_name, ldlm_ns_proc_dir);
 204         spin_unlock(&ldlm_namespace_lock);
 205
 206         ldlm_namespace_cleanup(ns, 0);
 207
 208         vfree(ns->ns_hash /* , sizeof(*ns->ns_hash) * RES_HASH_SIZE */);
 209         obd_memory -= sizeof(*ns->ns_hash) * RES_HASH_SIZE;
 210         OBD_FREE(ns->ns_name, strlen(ns->ns_name) + 1);
 211         OBD_FREE(ns, sizeof(*ns));
 212
 213         return ELDLM_OK;
 214 }
 215
 216 int ldlm_client_free(struct obd_export *exp)
 217 {
 218         struct ldlm_export_data *led = &exp->exp_ldlm_data;
 219         ptlrpc_cleanup_client(&led->led_import);
 220         RETURN(0);
 221 }
 222
 223 static __u32 ldlm_hash_fn(struct ldlm_resource *parent, __u64 *name)
 224 {
 225         __u32 hash = 0;
 226         int i;
 227
 228         for (i = 0; i < RES_NAME_SIZE; i++)
 229                 hash += name[i];
 230
 231         hash += (__u32)((unsigned long)parent >> 4);
 232
 233         return (hash & RES_HASH_MASK);
 234 }
 235
 236 static struct ldlm_resource *ldlm_resource_new(void)
 237 {
 238         struct ldlm_resource *res;
 239
 240         res = kmem_cache_alloc(ldlm_resource_slab, SLAB_KERNEL);
 241         if (res == NULL) {
 242                 LBUG();
 243                 return NULL;
 244         }
 245         memset(res, 0, sizeof(*res));
 246
 247         INIT_LIST_HEAD(&res->lr_children);
 248         INIT_LIST_HEAD(&res->lr_childof);
 249         INIT_LIST_HEAD(&res->lr_granted);
 250         INIT_LIST_HEAD(&res->lr_converting);
 251         INIT_LIST_HEAD(&res->lr_waiting);
 252
 253         atomic_set(&res->lr_refcount, 1);
 254
 255         return res;
 256 }
 257
 258 /* Args: locked namespace
 259  * Returns: newly-allocated, referenced, unlocked resource */
 260 static struct ldlm_resource *ldlm_resource_add(struct ldlm_namespace *ns,
 261                                                struct ldlm_resource *parent,
 262                                                __u64 *name, __u32 type)
 263 {
 264         struct list_head *bucket;
 265         struct ldlm_resource *res;
 266         ENTRY;
 267
 268         if (type < LDLM_MIN_TYPE || type > LDLM_MAX_TYPE) {
 269                 LBUG();
 270                 RETURN(NULL);
 271         }
 272
 273         res = ldlm_resource_new();
 274         if (!res) {
 275                 LBUG();
 276                 RETURN(NULL);
 277         }
 278
 279         spin_lock(&ns->ns_counter_lock);
 280         ns->ns_resources++;
 281         spin_unlock(&ns->ns_counter_lock);
 282
 283         memcpy(res->lr_name, name, sizeof(res->lr_name));
 284         res->lr_namespace = ns;
 285         ns->ns_refcount++;
 286
 287         res->lr_type = type;
 288         res->lr_most_restr = LCK_NL;
 289
 290         bucket = ns->ns_hash + ldlm_hash_fn(parent, name);
 291         list_add(&res->lr_hash, bucket);
 292
 293         if (parent == NULL)
 294                 list_add(&res->lr_childof, &ns->ns_root_list);
 295         else {
 296                 res->lr_parent = parent;
 297                 list_add(&res->lr_childof, &parent->lr_children);
 298         }
 299
 300         RETURN(res);
 301 }
 302
 303 /* Args: unlocked namespace
 304  * Locks: takes and releases ns->ns_lock and res->lr_lock
 305  * Returns: referenced, unlocked ldlm_resource or NULL */
 306 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
 307                                         struct ldlm_resource *parent,
 308                                         __u64 *name, __u32 type, int create)
 309 {
 310         struct list_head *bucket;
 311         struct list_head *tmp = bucket;
 312         struct ldlm_resource *res = NULL;
 313         ENTRY;
 314
 315         if (ns == NULL || ns->ns_hash == NULL) {
 316                 LBUG();
 317                 RETURN(NULL);
 318         }
 319
 320         l_lock(&ldlm_everything_lock);
 321         bucket = ns->ns_hash + ldlm_hash_fn(parent, name);
 322
 323         list_for_each(tmp, bucket) {
 324                 struct ldlm_resource *chk;
 325                 chk = list_entry(tmp, struct ldlm_resource, lr_hash);
 326
 327                 if (memcmp(chk->lr_name, name, sizeof(chk->lr_name)) == 0) {
 328                         res = chk;
 329                         atomic_inc(&res->lr_refcount);
 330                         EXIT;
 331                         break;
 332                 }
 333         }
 334
 335         if (res == NULL && create)
 336                 res = ldlm_resource_add(ns, parent, name, type);
 337         l_unlock(&ldlm_everything_lock);
 338
 339         RETURN(res);
 340 }
 341
 342 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
 343 {
 344         atomic_inc(&res->lr_refcount);
 345         return res;
 346 }
 347
 348 /* Returns 1 if the resource was freed, 0 if it remains. */
 349 int ldlm_resource_put(struct ldlm_resource *res)
 350 {
 351         int rc = 0;
 352
 353         if (atomic_dec_and_test(&res->lr_refcount)) {
 354                 struct ldlm_namespace *ns = res->lr_namespace;
 355                 ENTRY;
 356
 357                 l_lock(&ldlm_everything_lock);
 358
 359                 if (atomic_read(&res->lr_refcount) != 0) {
 360                         /* We lost the race. */
 361                         l_unlock(&ldlm_everything_lock);
 362                         goto out;
 363                 }
 364
 365                 if (!list_empty(&res->lr_granted))
 366                         LBUG();
 367
 368                 if (!list_empty(&res->lr_converting))
 369                         LBUG();
 370
 371                 if (!list_empty(&res->lr_waiting))
 372                         LBUG();
 373
 374                 if (!list_empty(&res->lr_children))
 375                         LBUG();
 376
 377                 ns->ns_refcount--;
 378                 list_del(&res->lr_hash);
 379                 list_del(&res->lr_childof);
 380
 381                 kmem_cache_free(ldlm_resource_slab, res);
 382                 l_unlock(&ldlm_everything_lock);
 383
 384                 spin_lock(&ns->ns_counter_lock);
 385                 ns->ns_resources--;
 386                 spin_unlock(&ns->ns_counter_lock);
 387
 388                 rc = 1;
 389         } else {
 390                 ENTRY;
 391         out:
 392                 if (atomic_read(&res->lr_refcount) < 0)
 393                         LBUG();
 394         }
 395
 396         RETURN(rc);
 397 }
 398
 399 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
 400                             struct ldlm_lock *lock)
 401 {
 402         l_lock(&ldlm_everything_lock);
 403
 404         ldlm_resource_dump(res);
 405         ldlm_lock_dump(lock);
 406
 407         if (!list_empty(&lock->l_res_link))
 408                 LBUG();
 409
 410         list_add(&lock->l_res_link, head);
 411         l_unlock(&ldlm_everything_lock);
 412 }
 413
 414 void ldlm_resource_unlink_lock(struct ldlm_lock *lock)
 415 {
 416         l_lock(&ldlm_everything_lock);
 417         list_del_init(&lock->l_res_link);
 418         l_unlock(&ldlm_everything_lock);
 419 }
 420
 421 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
 422 {
 423         desc->lr_type = res->lr_type;
 424         memcpy(desc->lr_name, res->lr_name, sizeof(desc->lr_name));
 425         memcpy(desc->lr_version, res->lr_version, sizeof(desc->lr_version));
 426 }
 427
 428 void ldlm_dump_all_namespaces(void)
 429 {
 430         struct list_head *tmp;
 431
 432         spin_lock(&ldlm_namespace_lock);
 433
 434         list_for_each(tmp, &ldlm_namespace_list) {
 435                 struct ldlm_namespace *ns;
 436                 ns = list_entry(tmp, struct ldlm_namespace, ns_list_chain);
 437                 ldlm_namespace_dump(ns);
 438         }
 439
 440         spin_unlock(&ldlm_namespace_lock);
 441 }
 442
 443 void ldlm_namespace_dump(struct ldlm_namespace *ns)
 444 {
 445         struct list_head *tmp;
 446
 447         l_lock(&ldlm_everything_lock);
 448         CDEBUG(D_OTHER, "--- Namespace: %s (rc: %d, client: %d)\n", ns->ns_name,
 449                ns->ns_refcount, ns->ns_client);
 450
 451         list_for_each(tmp, &ns->ns_root_list) {
 452                 struct ldlm_resource *res;
 453                 res = list_entry(tmp, struct ldlm_resource, lr_childof);
 454
 455                 /* Once we have resources with children, this should really dump
 456                  * them recursively. */
 457                 ldlm_resource_dump(res);
 458         }
 459         l_unlock(&ldlm_everything_lock);
 460 }
 461
 462 void ldlm_resource_dump(struct ldlm_resource *res)
 463 {
 464         struct list_head *tmp;
 465         char name[256];
 466
 467         if (RES_NAME_SIZE != 3)
 468                 LBUG();
 469
 470         snprintf(name, sizeof(name), "%Lx %Lx %Lx",
 471                  (unsigned long long)res->lr_name[0],
 472                  (unsigned long long)res->lr_name[1],
 473                  (unsigned long long)res->lr_name[2]);
 474
 475         CDEBUG(D_OTHER, "--- Resource: %p (%s) (rc: %d)\n", res, name,
 476                atomic_read(&res->lr_refcount));
 477         CDEBUG(D_OTHER, "Namespace: %p (%s)\n", res->lr_namespace,
 478                res->lr_namespace->ns_name);
 479         CDEBUG(D_OTHER, "Parent: %p, root: %p\n", res->lr_parent, res->lr_root);
 480
 481         CDEBUG(D_OTHER, "Granted locks:\n");
 482         list_for_each(tmp, &res->lr_granted) {
 483                 struct ldlm_lock *lock;
 484                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 485                 ldlm_lock_dump(lock);
 486         }
 487
 488         CDEBUG(D_OTHER, "Converting locks:\n");
 489         list_for_each(tmp, &res->lr_converting) {
 490                 struct ldlm_lock *lock;
 491                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 492                 ldlm_lock_dump(lock);
 493         }
 494
 495         CDEBUG(D_OTHER, "Waiting locks:\n");
 496         list_for_each(tmp, &res->lr_waiting) {
 497                 struct ldlm_lock *lock;
 498                 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
 499                 ldlm_lock_dump(lock);
 500         }
 501 }