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 lustre_lock ldlm_handle_lock;
  36 extern struct list_head ldlm_namespace_list;
  37 extern int (*mds_reint_p)(int offset, struct ptlrpc_request *req);
  38 extern int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req);
  39
  40 inline unsigned long round_timeout(unsigned long timeout)
  41 {
  42         return ((timeout / HZ) + 1) * HZ;
  43 }
  44
  45 static struct list_head waiting_locks_list;
  46 static spinlock_t waiting_locks_spinlock;
  47 static struct timer_list waiting_locks_timer;
  48 static int ldlm_already_setup = 0;
  49
  50 static void waiting_locks_callback(unsigned long unused)
  51 {
  52         struct list_head *liter, *n;
  53
  54         spin_lock_bh(&waiting_locks_spinlock);
  55         list_for_each_safe(liter, n, &waiting_locks_list) {
  56                 struct ldlm_lock *l = list_entry(liter, struct ldlm_lock,
  57                                                  l_pending_chain);
  58                 if (l->l_callback_timeout > jiffies)
  59                         break;
  60                 LDLM_DEBUG(l, "timer expired, recovering exp %p on conn %p",
  61                            l->l_export, l->l_export->exp_connection);
  62                 recovd_conn_fail(l->l_export->exp_connection);
  63         }
  64         spin_unlock_bh(&waiting_locks_spinlock);
  65 }
  66
  67 /*
  68  * Indicate that we're waiting for a client to call us back cancelling a given
  69  * lock.  We add it to the pending-callback chain, and schedule the lock-timeout
  70  * timer to fire appropriately.  (We round up to the next second, to avoid
  71  * floods of timer firings during periods of high lock contention and traffic).
  72  */
  73 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
  74 {
  75         unsigned long timeout_rounded;
  76         ENTRY;
  77
  78         LASSERT(list_empty(&lock->l_pending_chain));
  79
  80         spin_lock_bh(&waiting_locks_spinlock);
  81         lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2);
  82
  83         timeout_rounded = round_timeout(lock->l_callback_timeout);
  84
  85         if (timeout_rounded < waiting_locks_timer.expires ||
  86             !timer_pending(&waiting_locks_timer)) {
  87                 mod_timer(&waiting_locks_timer, timeout_rounded);
  88         }
  89         list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
  90         spin_unlock_bh(&waiting_locks_spinlock);
  91         RETURN(1);
  92 }
  93
  94 /*
  95  * Remove a lock from the pending list, likely because it had its cancellation
  96  * callback arrive without incident.  This adjusts the lock-timeout timer if
  97  * needed.  Returns 0 if the lock wasn't pending after all, 1 if it was.
  98  */
  99 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 100 {
 101         struct list_head *list_next;
 102
 103         ENTRY;
 104
 105         spin_lock_bh(&waiting_locks_spinlock);
 106
 107         if (list_empty(&lock->l_pending_chain)) {
 108                 spin_unlock_bh(&waiting_locks_spinlock);
 109                 RETURN(0);
 110         }
 111
 112         list_next = lock->l_pending_chain.next;
 113         if (lock->l_pending_chain.prev == &waiting_locks_list) {
 114                 /* Removing the head of the list, adjust timer. */
 115                 if (list_next == &waiting_locks_list) {
 116                         /* No more, just cancel. */
 117                         del_timer(&waiting_locks_timer);
 118                 } else {
 119                         struct ldlm_lock *next;
 120                         next = list_entry(list_next, struct ldlm_lock,
 121                                           l_pending_chain);
 122                         mod_timer(&waiting_locks_timer,
 123                                   round_timeout(next->l_callback_timeout));
 124                 }
 125         }
 126         list_del_init(&lock->l_pending_chain);
 127         spin_unlock_bh(&waiting_locks_spinlock);
 128         RETURN(1);
 129 }
 130
 131 static int ldlm_server_blocking_ast(struct ldlm_lock *lock,
 132                                     struct ldlm_lock_desc *desc,
 133                                     void *data, __u32 data_len, int flag)
 134 {
 135         struct ldlm_request *body;
 136         struct ptlrpc_request *req;
 137         int rc = 0, size = sizeof(*body);
 138         ENTRY;
 139
 140         if (flag == LDLM_CB_CANCELING) {
 141                 /* Don't need to do anything here. */
 142                 RETURN(0);
 143         }
 144
 145         LASSERT(lock);
 146
 147         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 148         if (lock->l_destroyed) {
 149                 /* What's the point? */
 150                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 151                 RETURN(0);
 152         }
 153
 154         req = ptlrpc_prep_req(&lock->l_export->exp_ldlm_data.led_import,
 155                               LDLM_BL_CALLBACK, 1, &size, NULL);
 156         if (!req)
 157                 RETURN(-ENOMEM);
 158
 159         body = lustre_msg_buf(req->rq_reqmsg, 0);
 160         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 161                sizeof(body->lock_handle1));
 162         memcpy(&body->lock_desc, desc, sizeof(*desc));
 163
 164         LDLM_DEBUG(lock, "server preparing blocking AST");
 165         req->rq_replen = 0; /* no reply needed */
 166
 167         ldlm_add_waiting_lock(lock);
 168         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 169
 170         (void)ptl_send_rpc(req);
 171
 172         /* not waiting for reply */
 173         ptlrpc_req_finished(req);
 174
 175         RETURN(rc);
 176 }
 177
 178 static int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags)
 179 {
 180         struct ldlm_request *body;
 181         struct ptlrpc_request *req;
 182         int rc = 0, size = sizeof(*body);
 183         ENTRY;
 184
 185         if (lock == NULL) {
 186                 LBUG();
 187                 RETURN(-EINVAL);
 188         }
 189
 190         req = ptlrpc_prep_req(&lock->l_export->exp_ldlm_data.led_import,
 191                               LDLM_CP_CALLBACK, 1, &size, NULL);
 192         if (!req)
 193                 RETURN(-ENOMEM);
 194
 195         body = lustre_msg_buf(req->rq_reqmsg, 0);
 196         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 197                sizeof(body->lock_handle1));
 198         body->lock_flags = flags;
 199         ldlm_lock2desc(lock, &body->lock_desc);
 200
 201         LDLM_DEBUG(lock, "server preparing completion AST");
 202         req->rq_replen = 0; /* no reply needed */
 203
 204         (void)ptl_send_rpc(req);
 205
 206         /* not waiting for reply */
 207         ptlrpc_req_finished(req);
 208
 209         RETURN(rc);
 210 }
 211
 212 int ldlm_handle_enqueue(struct ptlrpc_request *req)
 213 {
 214         struct obd_device *obddev = req->rq_export->exp_obd;
 215         struct ldlm_reply *dlm_rep;
 216         struct ldlm_request *dlm_req;
 217         int rc, size = sizeof(*dlm_rep), cookielen = 0;
 218         __u32 flags;
 219         ldlm_error_t err;
 220         struct ldlm_lock *lock = NULL;
 221         void *cookie = NULL;
 222         ENTRY;
 223
 224         LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
 225
 226         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 227         flags = dlm_req->lock_flags;
 228         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN &&
 229             (flags & LDLM_FL_HAS_INTENT)) {
 230                 /* In this case, the reply buffer is allocated deep in
 231                  * local_lock_enqueue by the policy function. */
 232                 cookie = req;
 233                 cookielen = sizeof(*req);
 234         } else {
 235                 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
 236                                      &req->rq_repmsg);
 237                 if (rc) {
 238                         CERROR("out of memory\n");
 239                         RETURN(-ENOMEM);
 240                 }
 241                 if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) {
 242                         cookie = &dlm_req->lock_desc.l_extent;
 243                         cookielen = sizeof(struct ldlm_extent);
 244                 }
 245         }
 246
 247         /* XXX notice that this lock has no callback data: of course the
 248            export would be exactly what we may want to use here... */
 249         lock = ldlm_lock_create(obddev->obd_namespace,
 250                                 &dlm_req->lock_handle2,
 251                                 dlm_req->lock_desc.l_resource.lr_name,
 252                                 dlm_req->lock_desc.l_resource.lr_type,
 253                                 dlm_req->lock_desc.l_req_mode, NULL, 0);
 254         if (!lock)
 255                 GOTO(out, err = -ENOMEM);
 256
 257         memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
 258                sizeof(lock->l_remote_handle));
 259         LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
 260
 261         LASSERT(req->rq_export);
 262         lock->l_export = req->rq_export;
 263         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 264         list_add(&lock->l_export_chain,
 265                  &lock->l_export->exp_ldlm_data.led_held_locks);
 266         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 267
 268         err = ldlm_lock_enqueue(lock, cookie, cookielen, &flags,
 269                                 ldlm_server_completion_ast,
 270                                 ldlm_server_blocking_ast);
 271         if (err != ELDLM_OK)
 272                 GOTO(out, err);
 273
 274         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 275         dlm_rep->lock_flags = flags;
 276
 277         ldlm_lock2handle(lock, &dlm_rep->lock_handle);
 278         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT)
 279                 memcpy(&dlm_rep->lock_extent, &lock->l_extent,
 280                        sizeof(lock->l_extent));
 281         if (dlm_rep->lock_flags & LDLM_FL_LOCK_CHANGED) {
 282                 memcpy(dlm_rep->lock_resource_name, lock->l_resource->lr_name,
 283                        sizeof(dlm_rep->lock_resource_name));
 284                 dlm_rep->lock_mode = lock->l_req_mode;
 285         }
 286
 287         EXIT;
 288  out:
 289         if (lock)
 290                 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
 291                            "(err=%d)", err);
 292         req->rq_status = err;
 293
 294         if (lock) {
 295                 if (!err)
 296                         ldlm_reprocess_all(lock->l_resource);
 297                 LDLM_LOCK_PUT(lock);
 298         }
 299         LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p)", lock);
 300
 301         return 0;
 302 }
 303
 304 int ldlm_handle_convert(struct ptlrpc_request *req)
 305 {
 306         struct ldlm_request *dlm_req;
 307         struct ldlm_reply *dlm_rep;
 308         struct ldlm_lock *lock;
 309         int rc, size = sizeof(*dlm_rep);
 310         ENTRY;
 311
 312         rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
 313         if (rc) {
 314                 CERROR("out of memory\n");
 315                 RETURN(-ENOMEM);
 316         }
 317         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 318         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
 319         dlm_rep->lock_flags = dlm_req->lock_flags;
 320
 321         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 322         if (!lock) {
 323                 req->rq_status = EINVAL;
 324         } else {
 325                 LDLM_DEBUG(lock, "server-side convert handler START");
 326                 ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode,
 327                                   &dlm_rep->lock_flags);
 328                 if (ldlm_del_waiting_lock(lock))
 329                         CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock);
 330                 req->rq_status = 0;
 331         }
 332
 333         if (lock) {
 334                 ldlm_reprocess_all(lock->l_resource);
 335                 LDLM_DEBUG(lock, "server-side convert handler END");
 336                 LDLM_LOCK_PUT(lock);
 337         } else
 338                 LDLM_DEBUG_NOLOCK("server-side convert handler END");
 339
 340         RETURN(0);
 341 }
 342
 343 int ldlm_handle_cancel(struct ptlrpc_request *req)
 344 {
 345         struct ldlm_request *dlm_req;
 346         struct ldlm_lock *lock;
 347         int rc;
 348         ENTRY;
 349
 350         rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
 351         if (rc) {
 352                 CERROR("out of memory\n");
 353                 RETURN(-ENOMEM);
 354         }
 355         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 356         if (!dlm_req) {
 357                 CERROR("bad request buffer for cancel\n");
 358                 RETURN(-EINVAL);
 359         }
 360
 361         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 362         if (!lock) {
 363                 LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock (lock "
 364                                   "%p)", (void *)(unsigned long)
 365                                   dlm_req->lock_handle1.addr);
 366                 req->rq_status = ESTALE;
 367         } else {
 368                 LDLM_DEBUG(lock, "server-side cancel handler START");
 369                 ldlm_lock_cancel(lock);
 370                 if (ldlm_del_waiting_lock(lock))
 371                         CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock);
 372                 req->rq_status = 0;
 373         }
 374
 375         if (ptlrpc_reply(req->rq_svc, req) != 0)
 376                 LBUG();
 377
 378         if (lock) {
 379                 ldlm_reprocess_all(lock->l_resource);
 380                 LDLM_DEBUG(lock, "server-side cancel handler END");
 381                 LDLM_LOCK_PUT(lock);
 382         }
 383
 384         RETURN(0);
 385 }
 386
 387 static int ldlm_handle_bl_callback(struct ptlrpc_request *req)
 388 {
 389         struct ldlm_request *dlm_req;
 390         struct ldlm_lock *lock;
 391         int do_ast;
 392         ENTRY;
 393
 394         OBD_FAIL_RETURN(OBD_FAIL_OSC_LOCK_BL_AST, 0);
 395
 396         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 397
 398         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 399         if (!lock) {
 400                 CERROR("blocking callback on lock "LPX64" - lock disappeared\n",
 401                        dlm_req->lock_handle1.addr);
 402                 RETURN(0);
 403         }
 404
 405         LDLM_DEBUG(lock, "client blocking AST callback handler START");
 406
 407         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 408         lock->l_flags |= LDLM_FL_CBPENDING;
 409         do_ast = (!lock->l_readers && !lock->l_writers);
 410         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 411
 412         if (do_ast) {
 413                 LDLM_DEBUG(lock, "already unused, calling "
 414                            "callback (%p)", lock->l_blocking_ast);
 415                 if (lock->l_blocking_ast != NULL) {
 416                         lock->l_blocking_ast(lock, &dlm_req->lock_desc,
 417                                              lock->l_data, lock->l_data_len,
 418                                              LDLM_CB_BLOCKING);
 419                 }
 420         } else
 421                 LDLM_DEBUG(lock, "Lock still has references, will be"
 422                            " cancelled later");
 423
 424         LDLM_DEBUG(lock, "client blocking callback handler END");
 425         LDLM_LOCK_PUT(lock);
 426         RETURN(0);
 427 }
 428
 429 static int ldlm_handle_cp_callback(struct ptlrpc_request *req)
 430 {
 431         struct list_head ast_list = LIST_HEAD_INIT(ast_list);
 432         struct ldlm_request *dlm_req;
 433         struct ldlm_lock *lock;
 434         ENTRY;
 435
 436         OBD_FAIL_RETURN(OBD_FAIL_OSC_LOCK_CP_AST, 0);
 437
 438         dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 439
 440         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 441         if (!lock) {
 442                 CERROR("completion callback on lock "LPX64" - lock "
 443                        "disappeared\n", dlm_req->lock_handle1.addr);
 444                 RETURN(0);
 445         }
 446
 447         LDLM_DEBUG(lock, "client completion callback handler START");
 448
 449         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 450
 451         /* If we receive the completion AST before the actual enqueue returned,
 452          * then we might need to switch lock modes, resources, or extents. */
 453         if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
 454                 lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
 455                 LDLM_DEBUG(lock, "completion AST, new lock mode");
 456         }
 457         if (lock->l_resource->lr_type == LDLM_EXTENT)
 458                 memcpy(&lock->l_extent, &dlm_req->lock_desc.l_extent,
 459                        sizeof(lock->l_extent));
 460         ldlm_resource_unlink_lock(lock);
 461         if (memcmp(dlm_req->lock_desc.l_resource.lr_name,
 462                    lock->l_resource->lr_name,
 463                    sizeof(__u64) * RES_NAME_SIZE) != 0) {
 464                 ldlm_lock_change_resource(lock,
 465                                          dlm_req->lock_desc.l_resource.lr_name);
 466                 LDLM_DEBUG(lock, "completion AST, new resource");
 467         }
 468         lock->l_resource->lr_tmp = &ast_list;
 469         ldlm_grant_lock(lock);
 470         lock->l_resource->lr_tmp = NULL;
 471         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 472         LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
 473         LDLM_LOCK_PUT(lock);
 474
 475         ldlm_run_ast_work(&ast_list);
 476
 477         LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
 478                           lock);
 479         RETURN(0);
 480 }
 481
 482 static int ldlm_callback_handler(struct ptlrpc_request *req)
 483 {
 484         int rc;
 485         ENTRY;
 486
 487         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 488         if (rc) {
 489                 CERROR("lustre_ldlm: Invalid request: %d\n", rc);
 490                 RETURN(rc);
 491         }
 492
 493         if (req->rq_export == NULL) {
 494                 struct ldlm_request *dlm_req;
 495
 496                 CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
 497                        req->rq_reqmsg->opc, req->rq_request_portal,
 498                        req->rq_reply_portal);
 499                 CERROR("--> export addr: "LPX64", cookie: "LPX64"\n",
 500                        req->rq_reqmsg->addr, req->rq_reqmsg->cookie);
 501                 dlm_req = lustre_msg_buf(req->rq_reqmsg, 0);
 502                 CERROR("--> lock addr: "LPX64", cookie: "LPX64"\n",
 503                        dlm_req->lock_handle1.addr,dlm_req->lock_handle1.cookie);
 504                 CERROR("--> ignoring this error as a temporary workaround!  "
 505                        "beware!\n");
 506                 //RETURN(-ENOTCONN);
 507         }
 508
 509         switch (req->rq_reqmsg->opc) {
 510         case LDLM_BL_CALLBACK:
 511                 CDEBUG(D_INODE, "blocking ast\n");
 512                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
 513                 rc = ldlm_handle_bl_callback(req);
 514                 RETURN(rc);
 515         case LDLM_CP_CALLBACK:
 516                 CDEBUG(D_INODE, "completion ast\n");
 517                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0);
 518                 rc = ldlm_handle_cp_callback(req);
 519                 RETURN(rc);
 520
 521         default:
 522                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
 523                 RETURN(-EINVAL);
 524         }
 525
 526         RETURN(0);
 527 }
 528
 529
 530 static int ldlm_cancel_handler(struct ptlrpc_request *req)
 531 {
 532         int rc;
 533         ENTRY;
 534
 535         rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
 536         if (rc) {
 537                 CERROR("lustre_ldlm: Invalid request: %d\n", rc);
 538                 RETURN(rc);
 539         }
 540
 541         if (req->rq_export == NULL) {
 542                 CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
 543                        req->rq_reqmsg->opc, req->rq_request_portal,
 544                        req->rq_reply_portal);
 545                 CERROR("--> export addr: "LPX64", cookie: "LPX64"\n",
 546                        req->rq_reqmsg->addr, req->rq_reqmsg->cookie);
 547                 CERROR("--> ignoring this error as a temporary workaround!  "
 548                        "beware!\n");
 549                 //RETURN(-ENOTCONN);
 550         }
 551
 552         switch (req->rq_reqmsg->opc) {
 553
 554         /* XXX FIXME move this back to mds/handler.c, bug 625069 */
 555         case LDLM_CANCEL:
 556                 CDEBUG(D_INODE, "cancel\n");
 557                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
 558                 rc = ldlm_handle_cancel(req);
 559                 if (rc)
 560                         break;
 561                 RETURN(0);
 562
 563         default:
 564                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
 565                 RETURN(-EINVAL);
 566         }
 567
 568         RETURN(0);
 569 }
 570
 571
 572 static int ldlm_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len,
 573                           void *karg, void *uarg)
 574 {
 575         struct obd_device *obddev = class_conn2obd(conn);
 576         struct ptlrpc_connection *connection;
 577         int err = 0;
 578         ENTRY;
 579
 580         if (_IOC_TYPE(cmd) != IOC_LDLM_TYPE || _IOC_NR(cmd) < IOC_LDLM_MIN_NR ||
 581             _IOC_NR(cmd) > IOC_LDLM_MAX_NR) {
 582                 CDEBUG(D_IOCTL, "invalid ioctl (type %ld, nr %ld, size %ld)\n",
 583                        _IOC_TYPE(cmd), _IOC_NR(cmd), _IOC_SIZE(cmd));
 584                 RETURN(-EINVAL);
 585         }
 586
 587         OBD_ALLOC(obddev->u.ldlm.ldlm_client,
 588                   sizeof(*obddev->u.ldlm.ldlm_client));
 589         connection = ptlrpc_uuid_to_connection("ldlm");
 590         if (!connection)
 591                 CERROR("No LDLM UUID found: assuming ldlm is local.\n");
 592
 593         switch (cmd) {
 594         case IOC_LDLM_TEST:
 595                 err = ldlm_test(obddev, conn);
 596                 CERROR("-- done err %d\n", err);
 597                 GOTO(out, err);
 598         case IOC_LDLM_DUMP:
 599                 ldlm_dump_all_namespaces();
 600                 GOTO(out, err);
 601         default:
 602                 GOTO(out, err = -EINVAL);
 603         }
 604
 605  out:
 606         if (connection)
 607                 ptlrpc_put_connection(connection);
 608         OBD_FREE(obddev->u.ldlm.ldlm_client,
 609                  sizeof(*obddev->u.ldlm.ldlm_client));
 610         return err;
 611 }
 612
 613 static int ldlm_setup(struct obd_device *obddev, obd_count len, void *buf)
 614 {
 615         struct ldlm_obd *ldlm = &obddev->u.ldlm;
 616         int rc, i;
 617         ENTRY;
 618
 619         if (ldlm_already_setup)
 620                 RETURN(-EALREADY);
 621
 622         MOD_INC_USE_COUNT;
 623
 624         rc = ldlm_proc_setup(obddev);
 625         if (rc != 0)
 626                 GOTO(out_dec, rc);
 627
 628         ldlm->ldlm_cb_service =
 629                 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
 630                                 LDLM_MAXREQSIZE, LDLM_CB_REQUEST_PORTAL,
 631                                 LDLM_CB_REPLY_PORTAL, "self",
 632                                 ldlm_callback_handler, "ldlm_cbd");
 633
 634         if (!ldlm->ldlm_cb_service) {
 635                 CERROR("failed to start service\n");
 636                 GOTO(out_proc, rc = -ENOMEM);
 637         }
 638
 639         ldlm->ldlm_cancel_service =
 640                 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
 641                                 LDLM_MAXREQSIZE, LDLM_CANCEL_REQUEST_PORTAL,
 642                                 LDLM_CANCEL_REPLY_PORTAL, "self",
 643                                 ldlm_cancel_handler, "ldlm_canceld");
 644
 645         if (!ldlm->ldlm_cancel_service) {
 646                 CERROR("failed to start service\n");
 647                 GOTO(out_proc, rc = -ENOMEM);
 648         }
 649
 650         for (i = 0; i < LDLM_NUM_THREADS; i++) {
 651                 char name[32];
 652                 sprintf(name, "ldlm_cn_%02d", i);
 653                 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cancel_service,
 654                                          name);
 655                 if (rc) {
 656                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
 657                         LBUG();
 658                         GOTO(out_thread, rc);
 659                 }
 660         }
 661
 662         for (i = 0; i < LDLM_NUM_THREADS; i++) {
 663                 char name[32];
 664                 sprintf(name, "ldlm_cb_%02d", i);
 665                 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cb_service, name);
 666                 if (rc) {
 667                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
 668                         LBUG();
 669                         GOTO(out_thread, rc);
 670                 }
 671         }
 672
 673         INIT_LIST_HEAD(&waiting_locks_list);
 674         spin_lock_init(&waiting_locks_spinlock);
 675         waiting_locks_timer.function = waiting_locks_callback;
 676         waiting_locks_timer.data = 0;
 677         init_timer(&waiting_locks_timer);
 678
 679         ldlm_already_setup = 1;
 680
 681         RETURN(0);
 682
 683  out_thread:
 684         ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
 685         ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
 686         ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
 687         ptlrpc_unregister_service(ldlm->ldlm_cb_service);
 688
 689  out_proc:
 690         ldlm_proc_cleanup(obddev);
 691
 692  out_dec:
 693         MOD_DEC_USE_COUNT;
 694         return rc;
 695 }
 696
 697 static int ldlm_cleanup(struct obd_device *obddev)
 698 {
 699         struct ldlm_obd *ldlm = &obddev->u.ldlm;
 700         ENTRY;
 701
 702         if (!list_empty(&ldlm_namespace_list)) {
 703                 CERROR("ldlm still has namespaces; clean these up first.\n");
 704                 RETURN(-EBUSY);
 705         }
 706
 707         ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
 708         ptlrpc_unregister_service(ldlm->ldlm_cb_service);
 709         ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
 710         ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
 711         ldlm_proc_cleanup(obddev);
 712
 713         ldlm_already_setup = 0;
 714         MOD_DEC_USE_COUNT;
 715         RETURN(0);
 716 }
 717
 718 static int ldlm_connect(struct lustre_handle *conn, struct obd_device *src,
 719                         obd_uuid_t cluuid, struct recovd_obd *recovd,
 720                         ptlrpc_recovery_cb_t recover)
 721 {
 722         return class_connect(conn, src, cluuid);
 723 }
 724
 725 struct obd_ops ldlm_obd_ops = {
 726         o_iocontrol:   ldlm_iocontrol,
 727         o_setup:       ldlm_setup,
 728         o_cleanup:     ldlm_cleanup,
 729         o_connect:     ldlm_connect,
 730         o_disconnect:  class_disconnect
 731 };
 732
 733 static int __init ldlm_init(void)
 734 {
 735         int rc = class_register_type(&ldlm_obd_ops, 0, OBD_LDLM_DEVICENAME);
 736         if (rc != 0)
 737                 return rc;
 738
 739         ldlm_resource_slab = kmem_cache_create("ldlm_resources",
 740                                                sizeof(struct ldlm_resource), 0,
 741                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 742         if (ldlm_resource_slab == NULL)
 743                 return -ENOMEM;
 744
 745         ldlm_lock_slab = kmem_cache_create("ldlm_locks",
 746                                            sizeof(struct ldlm_lock), 0,
 747                                            SLAB_HWCACHE_ALIGN, NULL, NULL);
 748         if (ldlm_lock_slab == NULL) {
 749                 kmem_cache_destroy(ldlm_resource_slab);
 750                 return -ENOMEM;
 751         }
 752
 753         l_lock_init(&ldlm_handle_lock);
 754
 755         return 0;
 756 }
 757
 758 static void __exit ldlm_exit(void)
 759 {
 760         class_unregister_type(OBD_LDLM_DEVICENAME);
 761         if (kmem_cache_destroy(ldlm_resource_slab) != 0)
 762                 CERROR("couldn't free ldlm resource slab\n");
 763         if (kmem_cache_destroy(ldlm_lock_slab) != 0)
 764                 CERROR("couldn't free ldlm lock slab\n");
 765 }
 766
 767 EXPORT_SYMBOL(ldlm_completion_ast);
 768 EXPORT_SYMBOL(ldlm_handle_enqueue);
 769 EXPORT_SYMBOL(ldlm_handle_cancel);
 770 EXPORT_SYMBOL(ldlm_handle_convert);
 771 EXPORT_SYMBOL(ldlm_register_intent);
 772 EXPORT_SYMBOL(ldlm_unregister_intent);
 773 EXPORT_SYMBOL(ldlm_lockname);
 774 EXPORT_SYMBOL(ldlm_typename);
 775 EXPORT_SYMBOL(__ldlm_handle2lock);
 776 EXPORT_SYMBOL(ldlm_lock2handle);
 777 EXPORT_SYMBOL(ldlm_lock_put);
 778 EXPORT_SYMBOL(ldlm_lock_match);
 779 EXPORT_SYMBOL(ldlm_lock_addref);
 780 EXPORT_SYMBOL(ldlm_lock_decref);
 781 EXPORT_SYMBOL(ldlm_lock_change_resource);
 782 EXPORT_SYMBOL(ldlm_lock_set_data);
 783 EXPORT_SYMBOL(ldlm_cli_convert);
 784 EXPORT_SYMBOL(ldlm_cli_enqueue);
 785 EXPORT_SYMBOL(ldlm_cli_cancel);
 786 EXPORT_SYMBOL(ldlm_cli_cancel_unused);
 787 EXPORT_SYMBOL(ldlm_match_or_enqueue);
 788 EXPORT_SYMBOL(ldlm_it2str);
 789 EXPORT_SYMBOL(ldlm_test);
 790 EXPORT_SYMBOL(ldlm_regression_start);
 791 EXPORT_SYMBOL(ldlm_regression_stop);
 792 EXPORT_SYMBOL(ldlm_lock_dump);
 793 EXPORT_SYMBOL(ldlm_namespace_new);
 794 EXPORT_SYMBOL(ldlm_namespace_cleanup);
 795 EXPORT_SYMBOL(ldlm_namespace_free);
 796 EXPORT_SYMBOL(ldlm_namespace_dump);
 797 EXPORT_SYMBOL(ldlm_cancel_locks_for_export);
 798 EXPORT_SYMBOL(ldlm_replay_locks);
 799 EXPORT_SYMBOL(ldlm_resource_foreach);
 800 EXPORT_SYMBOL(ldlm_namespace_foreach);
 801 EXPORT_SYMBOL(l_lock);
 802 EXPORT_SYMBOL(l_unlock);
 803
 804 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 805 MODULE_DESCRIPTION("Lustre Lock Management Module v0.1");
 806 MODULE_LICENSE("GPL");
 807
 808 module_init(ldlm_init);
 809 module_exit(ldlm_exit);