lustre/ldlm/ldlm_lockd.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  *   Author: Peter Braam <braam@clusterfs.com>
   6  *   Author: Phil Schwan <phil@clusterfs.com>
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define EXPORT_SYMTAB
  25 #define DEBUG_SUBSYSTEM S_LDLM
  26
  27 #include <linux/module.h>
  28 #include <linux/slab.h>
  29 #include <linux/lustre_dlm.h>
  30 #include <linux/init.h>
  31 #include <linux/obd_class.h>
  32
  33 extern kmem_cache_t *ldlm_resource_slab;
  34 extern kmem_cache_t *ldlm_lock_slab;
  35 extern struct list_head ldlm_namespace_list;
  36 extern int (*mds_reint_p)(int offset, struct ptlrpc_request *req);
  37 extern int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req);
  38
  39 inline unsigned long round_timeout(unsigned long timeout)
  40 {
  41         return ((timeout / HZ) + 1) * HZ;
  42 }
  43
  44 static struct list_head waiting_locks_list;
  45 static spinlock_t waiting_locks_spinlock;
  46 static struct timer_list waiting_locks_timer;
  47 static int ldlm_already_setup = 0;
  48
  49 static void waiting_locks_callback(unsigned long unused)
  50 {
  51         struct list_head *liter, *n;
  52
  53         spin_lock_bh(&waiting_locks_spinlock);
  54         list_for_each_safe(liter, n, &waiting_locks_list) {
  55                 struct ldlm_lock *l = list_entry(liter, struct ldlm_lock,
  56                                                  l_pending_chain);
  57                 if (l->l_callback_timeout > jiffies)
  58                         break;
  59                 LDLM_DEBUG(l, "timer expired, recovering exp %p on conn %p",
  60                            l->l_export, l->l_export->exp_connection);
  61                 recovd_conn_fail(l->l_export->exp_connection);
  62         }
  63         spin_unlock_bh(&waiting_locks_spinlock);
  64 }
  65
  66 /*
  67  * Indicate that we're waiting for a client to call us back cancelling a given
  68  * lock.  We add it to the pending-callback chain, and schedule the lock-timeout
  69  * timer to fire appropriately.  (We round up to the next second, to avoid
  70  * floods of timer firings during periods of high lock contention and traffic).
  71  */
  72 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
  73 {
  74         unsigned long timeout_rounded;
  75         ENTRY;
  76
  77         LASSERT(list_empty(&lock->l_pending_chain));
  78
  79         spin_lock_bh(&waiting_locks_spinlock);
  80         lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2);
  81
  82         timeout_rounded = round_timeout(lock->l_callback_timeout);
  83
  84         if (timeout_rounded < waiting_locks_timer.expires ||
  85             !timer_pending(&waiting_locks_timer)) {
  86                 mod_timer(&waiting_locks_timer, timeout_rounded);
  87         }
  88         list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
  89         spin_unlock_bh(&waiting_locks_spinlock);
  90         RETURN(1);
  91 }
  92
  93 /*
  94  * Remove a lock from the pending list, likely because it had its cancellation
  95  * callback arrive without incident.  This adjusts the lock-timeout timer if
  96  * needed.  Returns 0 if the lock wasn't pending after all, 1 if it was.
  97  */
  98 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
  99 {
 100         struct list_head *list_next;
 101
 102         ENTRY;
 103
 104         spin_lock_bh(&waiting_locks_spinlock);
 105
 106         if (list_empty(&lock->l_pending_chain)) {
 107                 spin_unlock_bh(&waiting_locks_spinlock);
 108                 RETURN(0);
 109         }
 110
 111         list_next = lock->l_pending_chain.next;
 112         if (lock->l_pending_chain.prev == &waiting_locks_list) {
 113                 /* Removing the head of the list, adjust timer. */
 114                 if (list_next == &waiting_locks_list) {
 115                         /* No more, just cancel. */
 116                         del_timer(&waiting_locks_timer);
 117                 } else {
 118                         struct ldlm_lock *next;
 119                         next = list_entry(list_next, struct ldlm_lock,
 120                                           l_pending_chain);
 121                         mod_timer(&waiting_locks_timer,
 122                                   round_timeout(next->l_callback_timeout));
 123                 }
 124         }
 125         list_del_init(&lock->l_pending_chain);
 126         spin_unlock_bh(&waiting_locks_spinlock);
 127         RETURN(1);
 128 }
 129
 130 static int ldlm_server_blocking_ast(struct ldlm_lock *lock,
 131                                     struct ldlm_lock_desc *desc,
 132                                     void *data, __u32 data_len, int flag)
 133 {
 134         struct ldlm_request *body;
 135         struct ptlrpc_request *req;
 136         int rc = 0, size = sizeof(*body);
 137         ENTRY;
 138
 139         if (flag == LDLM_CB_CANCELING) {
 140                 /* Don't need to do anything here. */
 141                 RETURN(0);
 142         }
 143
 144         req = ptlrpc_prep_req(&lock->l_export->exp_ldlm_data.led_import,
 145                               LDLM_BL_CALLBACK, 1, &size, NULL);
 146         if (!req)
 147                 RETURN(-ENOMEM);
 148
 149         body = lustre_msg_buf(req->rq_reqmsg, 0);
 150         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 151                sizeof(body->lock_handle1));
 152         memcpy(&body->lock_desc, desc, sizeof(*desc));
 153
 154         LDLM_DEBUG(lock, "server preparing blocking AST");
 155         req->rq_replen = 0; /* no reply needed */
 156
 157         ldlm_add_waiting_lock(lock);
 158         (void)ptl_send_rpc(req);
 159
 160         /* not waiting for reply */
 161         ptlrpc_req_finished(req);
 162
 163         RETURN(rc);
 164 }
 165
 166 static int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags)
 167 {
 168         struct ldlm_request *body;
 169         struct ptlrpc_request *req;
 170         int rc = 0, size = sizeof(*body);
 171         ENTRY;
 172
 173         if (lock == NULL) {
 174                 LBUG();
 175                 RETURN(-EINVAL);
 176         }
 177
 178         req = ptlrpc_prep_req(&lock->l_export->exp_ldlm_data.led_import,
 179                               LDLM_CP_CALLBACK, 1, &size, NULL);
 180         if (!req)
 181                 RETURN(-ENOMEM);
 182
 183         body = lustre_msg_buf(req->rq_reqmsg, 0);
 184         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 185                sizeof(body->lock_handle1));
 186         body->lock_flags = flags;
 187         ldlm_lock2desc(lock, &body->lock_desc);
 188
 189         LDLM_DEBUG(lock, "server preparing completion AST");
 190         req->rq_replen = 0; /* no reply needed */
 191
 192         (void)ptl_send_rpc(req);
 193
 194         /* not waiting for reply */
 195         ptlrpc_req_finished(req);
 196
 197         RETURN(rc);
 198 }
 199
 200 int ldlm_handle_enqueue(struct ptlrpc_request *req)
 201 {
 202         struct obd_device *obddev = req->rq_export->exp_obd;
 203         struct ldlm_reply *dlm_rep;
 204         struct ldlm_request *dlm_req;
 205         int rc, size = sizeof(*dlm_rep), cookielen = 0;
 206         __u32 flags;
 207         ldlm_error_t err;
 208         struct ldlm_lock *lock = NULL;
 209         void *cookie = NULL;
 210         ENTRY;
 211
 212         LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
 213
 214         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 215         flags = dlm_req->lock_flags;
 216         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN &&
 217             (flags & LDLM_FL_HAS_INTENT)) {
 218                 /* In this case, the reply buffer is allocated deep in
 219                  * local_lock_enqueue by the policy function. */
 220                 cookie = req;
 221                 cookielen = sizeof(*req);
 222         } else {
 223                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 224                                      &req->rq_repmsg);
 225                 if (rc) {
 226                         CERROR("out of memory\n");
 227                         RETURN(-ENOMEM);
 228                 }
 229                 if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) {
 230                         cookie = &dlm_req->lock_desc.l_extent;
 231                         cookielen = sizeof(struct ldlm_extent);
 232                 }
 233         }
 234
 235         /* XXX notice that this lock has no callback data: of course the
 236            export would be exactly what we may want to use here... */
 237         lock = ldlm_lock_create(obddev->obd_namespace,
 238                                 &dlm_req->lock_handle2,
 239                                 dlm_req->lock_desc.l_resource.lr_name,
 240                                 dlm_req->lock_desc.l_resource.lr_type,
 241                                 dlm_req->lock_desc.l_req_mode, NULL, 0);
 242         if (!lock)
 243                 GOTO(out, err = -ENOMEM);
 244
 245         memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
 246                sizeof(lock->l_remote_handle));
 247         LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
 248
 249         err = ldlm_lock_enqueue(lock, cookie, cookielen, &flags,
 250                                 ldlm_server_completion_ast,
 251                                 ldlm_server_blocking_ast);
 252         if (err != ELDLM_OK)
 253                 GOTO(out, err);
 254
 255         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 256         dlm_rep->lock_flags = flags;
 257
 258         ldlm_lock2handle(lock, &dlm_rep->lock_handle);
 259         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT)
 260                 memcpy(&dlm_rep->lock_extent, &lock->l_extent,
 261                        sizeof(lock->l_extent));
 262         if (dlm_rep->lock_flags & LDLM_FL_LOCK_CHANGED) {
 263                 memcpy(dlm_rep->lock_resource_name, lock->l_resource->lr_name,
 264                        sizeof(dlm_rep->lock_resource_name));
 265                 dlm_rep->lock_mode = lock->l_req_mode;
 266         }
 267
 268         lock->l_export = req->rq_export;
 269         if (lock->l_export) {
 270                 l_lock(&lock->l_resource->lr_namespace->ns_lock);
 271                 list_add(&lock->l_export_chain,
 272                          &lock->l_export->exp_ldlm_data.led_held_locks);
 273                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 274         }
 275
 276         EXIT;
 277  out:
 278         if (lock)
 279                 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
 280                            "(err=%d)", err);
 281         req->rq_status = err;
 282
 283         if (lock) {
 284                 if (!err)
 285                         ldlm_reprocess_all(lock->l_resource);
 286                 LDLM_LOCK_PUT(lock);
 287         }
 288         LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p)", lock);
 289
 290         return 0;
 291 }
 292
 293 int ldlm_handle_convert(struct ptlrpc_request *req)
 294 {
 295         struct ldlm_request *dlm_req;
 296         struct ldlm_reply *dlm_rep;
 297         struct ldlm_lock *lock;
 298         int rc, size = sizeof(*dlm_rep);
 299         ENTRY;
 300
 301         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 302         if (rc) {
 303                 CERROR("out of memory\n");
 304                 RETURN(-ENOMEM);
 305         }
 306         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 307         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 308         dlm_rep->lock_flags = dlm_req->lock_flags;
 309
 310         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 311         if (!lock) {
 312                 req->rq_status = EINVAL;
 313         } else {
 314                 LDLM_DEBUG(lock, "server-side convert handler START");
 315                 ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode,
 316                                   &dlm_rep->lock_flags);
 317                 if (ldlm_del_waiting_lock(lock))
 318                         CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock);
 319                 req->rq_status = 0;
 320         }
 321
 322         if (lock) {
 323                 ldlm_reprocess_all(lock->l_resource);
 324                 LDLM_DEBUG(lock, "server-side convert handler END");
 325                 LDLM_LOCK_PUT(lock);
 326         } else
 327                 LDLM_DEBUG_NOLOCK("server-side convert handler END");
 328
 329         RETURN(0);
 330 }
 331
 332 int ldlm_handle_cancel(struct ptlrpc_request *req)
 333 {
 334         struct ldlm_request *dlm_req;
 335         struct ldlm_lock *lock;
 336         int rc;
 337         ENTRY;
 338
 339         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 340         if (rc) {
 341                 CERROR("out of memory\n");
 342                 RETURN(-ENOMEM);
 343         }
 344         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 345         if (!dlm_req) {
 346                 CERROR("bad request buffer for cancel\n");
 347                 RETURN(-EINVAL);
 348         }
 349
 350         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 351         if (!lock) {
 352                 LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock (lock "
 353                                   "%p)", (void *)(unsigned long)
 354                                   dlm_req->lock_handle1.addr);
 355                 req->rq_status = ESTALE;
 356         } else {
 357                 LDLM_DEBUG(lock, "server-side cancel handler START");
 358                 ldlm_lock_cancel(lock);
 359                 if (ldlm_del_waiting_lock(lock))
 360                         CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock);
 361                 req->rq_status = 0;
 362         }
 363
 364         if (ptlrpc_reply(req->rq_svc, req) != 0)
 365                 LBUG();
 366
 367         if (lock) {
 368                 ldlm_reprocess_all(lock->l_resource);
 369                 LDLM_DEBUG(lock, "server-side cancel handler END");
 370                 LDLM_LOCK_PUT(lock);
 371         }
 372
 373         RETURN(0);
 374 }
 375
 376 static int ldlm_handle_bl_callback(struct ptlrpc_request *req)
 377 {
 378         struct ldlm_request *dlm_req;
 379         struct ldlm_lock *lock;
 380         int do_ast;
 381         ENTRY;
 382
 383         OBD_FAIL_RETURN(OBD_FAIL_OSC_LOCK_BL_AST, 0);
 384
 385         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 386
 387         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 388         if (!lock) {
 389                 CERROR("blocking callback on lock "LPX64" - lock disappeared\n",
 390                        dlm_req->lock_handle1.addr);
 391                 RETURN(0);
 392         }
 393
 394         LDLM_DEBUG(lock, "client blocking AST callback handler START");
 395
 396         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 397         lock->l_flags |= LDLM_FL_CBPENDING;
 398         do_ast = (!lock->l_readers && !lock->l_writers);
 399         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 400
 401         if (do_ast) {
 402                 LDLM_DEBUG(lock, "already unused, calling "
 403                            "callback (%p)", lock->l_blocking_ast);
 404                 if (lock->l_blocking_ast != NULL) {
 405                         lock->l_blocking_ast(lock, &dlm_req->lock_desc,
 406                                              lock->l_data, lock->l_data_len,
 407                                              LDLM_CB_BLOCKING);
 408                 }
 409         } else
 410                 LDLM_DEBUG(lock, "Lock still has references, will be"
 411                            " cancelled later");
 412
 413         LDLM_DEBUG(lock, "client blocking callback handler END");
 414         LDLM_LOCK_PUT(lock);
 415         RETURN(0);
 416 }
 417
 418 static int ldlm_handle_cp_callback(struct ptlrpc_request *req)
 419 {
 420         struct list_head ast_list = LIST_HEAD_INIT(ast_list);
 421         struct ldlm_request *dlm_req;
 422         struct ldlm_lock *lock;
 423         ENTRY;
 424
 425         OBD_FAIL_RETURN(OBD_FAIL_OSC_LOCK_CP_AST, 0);
 426
 427         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 428
 429         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 430         if (!lock) {
 431                 CERROR("completion callback on lock "LPX64" - lock "
 432                        "disappeared\n", dlm_req->lock_handle1.addr);
 433                 RETURN(0);
 434         }
 435
 436         LDLM_DEBUG(lock, "client completion callback handler START");
 437
 438         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 439
 440         /* If we receive the completion AST before the actual enqueue returned,
 441          * then we might need to switch lock modes, resources, or extents. */
 442         if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
 443                 lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
 444                 LDLM_DEBUG(lock, "completion AST, new lock mode");
 445         }
 446         if (lock->l_resource->lr_type == LDLM_EXTENT)
 447                 memcpy(&lock->l_extent, &dlm_req->lock_desc.l_extent,
 448                        sizeof(lock->l_extent));
 449         ldlm_resource_unlink_lock(lock);
 450         if (memcmp(dlm_req->lock_desc.l_resource.lr_name,
 451                    lock->l_resource->lr_name,
 452                    sizeof(__u64) * RES_NAME_SIZE) != 0) {
 453                 ldlm_lock_change_resource(lock,
 454                                          dlm_req->lock_desc.l_resource.lr_name);
 455                 LDLM_DEBUG(lock, "completion AST, new resource");
 456         }
 457         lock->l_resource->lr_tmp = &ast_list;
 458         ldlm_grant_lock(lock);
 459         lock->l_resource->lr_tmp = NULL;
 460         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 461         LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
 462         LDLM_LOCK_PUT(lock);
 463
 464         ldlm_run_ast_work(&ast_list);
 465
 466         LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
 467                           lock);
 468         RETURN(0);
 469 }
 470
 471 static int ldlm_callback_handler(struct ptlrpc_request *req)
 472 {
 473         int rc;
 474         ENTRY;
 475
 476         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 477         if (rc) {
 478                 CERROR("lustre_ldlm: Invalid request: %d\n", rc);
 479                 RETURN(rc);
 480         }
 481
 482         if (req->rq_export == NULL) {
 483                 CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
 484                        req->rq_reqmsg->opc, req->rq_request_portal,
 485                        req->rq_reply_portal);
 486                 CERROR("--> export addr: "LPX64", cookie: "LPX64"\n",
 487                        req->rq_reqmsg->addr, req->rq_reqmsg->cookie);
 488                 CERROR("--> ignoring this error as a temporary workaround!  "
 489                        "beware!\n");
 490                 //RETURN(-ENOTCONN);
 491         }
 492
 493         switch (req->rq_reqmsg->opc) {
 494         case LDLM_BL_CALLBACK:
 495                 CDEBUG(D_INODE, "blocking ast\n");
 496                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
 497                 rc = ldlm_handle_bl_callback(req);
 498                 RETURN(rc);
 499         case LDLM_CP_CALLBACK:
 500                 CDEBUG(D_INODE, "completion ast\n");
 501                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0);
 502                 rc = ldlm_handle_cp_callback(req);
 503                 RETURN(rc);
 504
 505         default:
 506                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
 507                 RETURN(-EINVAL);
 508         }
 509
 510         RETURN(0);
 511 }
 512
 513
 514 static int ldlm_cancel_handler(struct ptlrpc_request *req)
 515 {
 516         int rc;
 517         ENTRY;
 518
 519         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 520         if (rc) {
 521                 CERROR("lustre_ldlm: Invalid request: %d\n", rc);
 522                 RETURN(rc);
 523         }
 524
 525         if (req->rq_export == NULL) {
 526                 CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
 527                        req->rq_reqmsg->opc, req->rq_request_portal,
 528                        req->rq_reply_portal);
 529                 CERROR("--> export addr: "LPX64", cookie: "LPX64"\n",
 530                        req->rq_reqmsg->addr, req->rq_reqmsg->cookie);
 531                 CERROR("--> ignoring this error as a temporary workaround!  "
 532                        "beware!\n");
 533                 //RETURN(-ENOTCONN);
 534         }
 535
 536         switch (req->rq_reqmsg->opc) {
 537
 538         /* XXX FIXME move this back to mds/handler.c, bug 625069 */
 539         case LDLM_CANCEL:
 540                 CDEBUG(D_INODE, "cancel\n");
 541                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
 542                 rc = ldlm_handle_cancel(req);
 543                 if (rc)
 544                         break;
 545                 RETURN(0);
 546
 547         default:
 548                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
 549                 RETURN(-EINVAL);
 550         }
 551
 552         RETURN(0);
 553 }
 554
 555
 556 static int ldlm_iocontrol(long cmd, struct lustre_handle *conn, int len,
 557                           void *karg, void *uarg)
 558 {
 559         struct obd_device *obddev = class_conn2obd(conn);
 560         struct ptlrpc_connection *connection;
 561         int err = 0;
 562         ENTRY;
 563
 564         if (_IOC_TYPE(cmd) != IOC_LDLM_TYPE || _IOC_NR(cmd) < IOC_LDLM_MIN_NR ||
 565             _IOC_NR(cmd) > IOC_LDLM_MAX_NR) {
 566                 CDEBUG(D_IOCTL, "invalid ioctl (type %ld, nr %ld, size %ld)\n",
 567                        _IOC_TYPE(cmd), _IOC_NR(cmd), _IOC_SIZE(cmd));
 568                 RETURN(-EINVAL);
 569         }
 570
 571         OBD_ALLOC(obddev->u.ldlm.ldlm_client,
 572                   sizeof(*obddev->u.ldlm.ldlm_client));
 573         connection = ptlrpc_uuid_to_connection("ldlm");
 574         if (!connection)
 575                 CERROR("No LDLM UUID found: assuming ldlm is local.\n");
 576
 577         switch (cmd) {
 578         case IOC_LDLM_TEST:
 579                 err = ldlm_test(obddev, conn);
 580                 CERROR("-- done err %d\n", err);
 581                 GOTO(out, err);
 582         case IOC_LDLM_DUMP:
 583                 ldlm_dump_all_namespaces();
 584                 GOTO(out, err);
 585         default:
 586                 GOTO(out, err = -EINVAL);
 587         }
 588
 589  out:
 590         if (connection)
 591                 ptlrpc_put_connection(connection);
 592         OBD_FREE(obddev->u.ldlm.ldlm_client,
 593                  sizeof(*obddev->u.ldlm.ldlm_client));
 594         return err;
 595 }
 596
 597 static int ldlm_setup(struct obd_device *obddev, obd_count len, void *buf)
 598 {
 599         struct ldlm_obd *ldlm = &obddev->u.ldlm;
 600         int rc, i;
 601         ENTRY;
 602
 603         if (ldlm_already_setup)
 604                 RETURN(-EALREADY);
 605
 606         MOD_INC_USE_COUNT;
 607
 608         rc = ldlm_proc_setup(obddev);
 609         if (rc != 0)
 610                 GOTO(out_dec, rc);
 611
 612         ldlm->ldlm_cb_service =
 613                 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
 614                                 LDLM_MAXREQSIZE, LDLM_CB_REQUEST_PORTAL,
 615                                 LDLM_CB_REPLY_PORTAL, "self",
 616                                 ldlm_callback_handler, "ldlm_cbd");
 617
 618         if (!ldlm->ldlm_cb_service) {
 619                 CERROR("failed to start service\n");
 620                 GOTO(out_proc, rc = -ENOMEM);
 621         }
 622
 623         ldlm->ldlm_cancel_service =
 624                 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
 625                                 LDLM_MAXREQSIZE, LDLM_CANCEL_REQUEST_PORTAL,
 626                                 LDLM_CANCEL_REPLY_PORTAL, "self",
 627                                 ldlm_cancel_handler, "ldlm_canceld");
 628
 629         if (!ldlm->ldlm_cancel_service) {
 630                 CERROR("failed to start service\n");
 631                 GOTO(out_proc, rc = -ENOMEM);
 632         }
 633
 634         for (i = 0; i < LDLM_NUM_THREADS; i++) {
 635                 char name[32];
 636                 sprintf(name, "ldlm_cn_%02d", i);
 637                 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cancel_service,
 638                                          name);
 639                 if (rc) {
 640                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
 641                         LBUG();
 642                         GOTO(out_thread, rc);
 643                 }
 644         }
 645
 646         for (i = 0; i < LDLM_NUM_THREADS; i++) {
 647                 char name[32];
 648                 sprintf(name, "ldlm_cb_%02d", i);
 649                 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cb_service, name);
 650                 if (rc) {
 651                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
 652                         LBUG();
 653                         GOTO(out_thread, rc);
 654                 }
 655         }
 656
 657         INIT_LIST_HEAD(&waiting_locks_list);
 658         spin_lock_init(&waiting_locks_spinlock);
 659         waiting_locks_timer.function = waiting_locks_callback;
 660         waiting_locks_timer.data = 0;
 661         init_timer(&waiting_locks_timer);
 662
 663         ldlm_already_setup = 1;
 664
 665         RETURN(0);
 666
 667  out_thread:
 668         ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
 669         ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
 670         ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
 671         ptlrpc_unregister_service(ldlm->ldlm_cb_service);
 672
 673  out_proc:
 674         ldlm_proc_cleanup(obddev);
 675
 676  out_dec:
 677         MOD_DEC_USE_COUNT;
 678         return rc;
 679 }
 680
 681 static int ldlm_cleanup(struct obd_device *obddev)
 682 {
 683         struct ldlm_obd *ldlm = &obddev->u.ldlm;
 684         ENTRY;
 685
 686         if (!list_empty(&ldlm_namespace_list)) {
 687                 CERROR("ldlm still has namespaces; clean these up first.\n");
 688                 RETURN(-EBUSY);
 689         }
 690
 691         ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
 692         ptlrpc_unregister_service(ldlm->ldlm_cb_service);
 693         ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
 694         ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
 695         ldlm_proc_cleanup(obddev);
 696
 697         ldlm_already_setup = 0;
 698         MOD_DEC_USE_COUNT;
 699         RETURN(0);
 700 }
 701
 702 static int ldlm_connect(struct lustre_handle *conn, struct obd_device *src,
 703                         obd_uuid_t cluuid, struct recovd_obd *recovd,
 704                         ptlrpc_recovery_cb_t recover)
 705 {
 706         return class_connect(conn, src, cluuid);
 707 }
 708
 709 struct obd_ops ldlm_obd_ops = {
 710         o_iocontrol:   ldlm_iocontrol,
 711         o_setup:       ldlm_setup,
 712         o_cleanup:     ldlm_cleanup,
 713         o_connect:     ldlm_connect,
 714         o_disconnect:  class_disconnect
 715 };
 716
 717 static int __init ldlm_init(void)
 718 {
 719         int rc = class_register_type(&ldlm_obd_ops, 0, OBD_LDLM_DEVICENAME);
 720         if (rc != 0)
 721                 return rc;
 722
 723         ldlm_resource_slab = kmem_cache_create("ldlm_resources",
 724                                                sizeof(struct ldlm_resource), 0,
 725                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 726         if (ldlm_resource_slab == NULL)
 727                 return -ENOMEM;
 728
 729         ldlm_lock_slab = kmem_cache_create("ldlm_locks",
 730                                            sizeof(struct ldlm_lock), 0,
 731                                            SLAB_HWCACHE_ALIGN, NULL, NULL);
 732         if (ldlm_lock_slab == NULL) {
 733                 kmem_cache_destroy(ldlm_resource_slab);
 734                 return -ENOMEM;
 735         }
 736
 737         return 0;
 738 }
 739
 740 static void __exit ldlm_exit(void)
 741 {
 742         class_unregister_type(OBD_LDLM_DEVICENAME);
 743         if (kmem_cache_destroy(ldlm_resource_slab) != 0)
 744                 CERROR("couldn't free ldlm resource slab\n");
 745         if (kmem_cache_destroy(ldlm_lock_slab) != 0)
 746                 CERROR("couldn't free ldlm lock slab\n");
 747 }
 748
 749 EXPORT_SYMBOL(ldlm_completion_ast);
 750 EXPORT_SYMBOL(ldlm_handle_enqueue);
 751 EXPORT_SYMBOL(ldlm_handle_cancel);
 752 EXPORT_SYMBOL(ldlm_handle_convert);
 753 EXPORT_SYMBOL(ldlm_register_intent);
 754 EXPORT_SYMBOL(ldlm_unregister_intent);
 755 EXPORT_SYMBOL(ldlm_lockname);
 756 EXPORT_SYMBOL(ldlm_typename);
 757 EXPORT_SYMBOL(__ldlm_handle2lock);
 758 EXPORT_SYMBOL(ldlm_lock2handle);
 759 EXPORT_SYMBOL(ldlm_lock_put);
 760 EXPORT_SYMBOL(ldlm_lock_match);
 761 EXPORT_SYMBOL(ldlm_lock_addref);
 762 EXPORT_SYMBOL(ldlm_lock_decref);
 763 EXPORT_SYMBOL(ldlm_lock_change_resource);
 764 EXPORT_SYMBOL(ldlm_lock_set_data);
 765 EXPORT_SYMBOL(ldlm_cli_convert);
 766 EXPORT_SYMBOL(ldlm_cli_enqueue);
 767 EXPORT_SYMBOL(ldlm_cli_cancel);
 768 EXPORT_SYMBOL(ldlm_cli_cancel_unused);
 769 EXPORT_SYMBOL(ldlm_match_or_enqueue);
 770 EXPORT_SYMBOL(ldlm_it2str);
 771 EXPORT_SYMBOL(ldlm_test);
 772 EXPORT_SYMBOL(ldlm_regression_start);
 773 EXPORT_SYMBOL(ldlm_regression_stop);
 774 EXPORT_SYMBOL(ldlm_lock_dump);
 775 EXPORT_SYMBOL(ldlm_namespace_new);
 776 EXPORT_SYMBOL(ldlm_namespace_cleanup);
 777 EXPORT_SYMBOL(ldlm_namespace_free);
 778 EXPORT_SYMBOL(ldlm_namespace_dump);
 779 EXPORT_SYMBOL(ldlm_cancel_locks_for_export);
 780 EXPORT_SYMBOL(ldlm_replay_locks);
 781 EXPORT_SYMBOL(ldlm_resource_foreach);
 782 EXPORT_SYMBOL(ldlm_namespace_foreach);
 783 EXPORT_SYMBOL(l_lock);
 784 EXPORT_SYMBOL(l_unlock);
 785
 786 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 787 MODULE_DESCRIPTION("Lustre Lock Management Module v0.1");
 788 MODULE_LICENSE("GPL");
 789
 790 module_init(ldlm_init);
 791 module_exit(ldlm_exit);