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, 2003 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 #ifndef EXPORT_SYMTAB
  25 # define EXPORT_SYMTAB
  26 #endif
  27 #define DEBUG_SUBSYSTEM S_LDLM
  28
  29 #ifdef __KERNEL__
  30 # include <linux/module.h>
  31 # include <linux/slab.h>
  32 # include <linux/init.h>
  33 # include <linux/wait.h>
  34 #else
  35 # include <liblustre.h>
  36 #endif
  37
  38 #include <linux/lustre_dlm.h>
  39 #include <linux/obd_class.h>
  40 #include <portals/list.h>
  41 #include "ldlm_internal.h"
  42
  43 extern kmem_cache_t *ldlm_resource_slab;
  44 extern kmem_cache_t *ldlm_lock_slab;
  45 extern struct lustre_lock ldlm_handle_lock;
  46 extern struct list_head ldlm_namespace_list;
  47
  48 static DECLARE_MUTEX(ldlm_ref_sem);
  49 static int ldlm_refcount;
  50
  51 /* LDLM state */
  52
  53 static struct ldlm_state *ldlm_state;
  54
  55 inline unsigned long round_timeout(unsigned long timeout)
  56 {
  57         return ((timeout / HZ) + 1) * HZ;
  58 }
  59
  60 #ifdef __KERNEL__
  61 /* XXX should this be per-ldlm? */
  62 static struct list_head waiting_locks_list;
  63 static spinlock_t waiting_locks_spinlock;
  64 static struct timer_list waiting_locks_timer;
  65
  66 static struct expired_lock_thread {
  67         wait_queue_head_t         elt_waitq;
  68         int                       elt_state;
  69         struct list_head          elt_expired_locks;
  70         spinlock_t                elt_lock;
  71 } expired_lock_thread;
  72 #endif
  73
  74 #define ELT_STOPPED   0
  75 #define ELT_READY     1
  76 #define ELT_TERMINATE 2
  77
  78 struct ldlm_bl_pool {
  79         spinlock_t              blp_lock;
  80         struct list_head        blp_list;
  81         wait_queue_head_t       blp_waitq;
  82         atomic_t                blp_num_threads;
  83         struct completion       blp_comp;
  84 };
  85
  86 struct ldlm_bl_work_item {
  87         struct list_head        blwi_entry;
  88         struct ldlm_namespace   *blwi_ns;
  89         struct ldlm_lock_desc   blwi_ld;
  90         struct ldlm_lock        *blwi_lock;
  91 };
  92
  93 #ifdef __KERNEL__
  94
  95 static inline int have_expired_locks(void)
  96 {
  97         int need_to_run;
  98
  99         spin_lock_bh(&expired_lock_thread.elt_lock);
 100         need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks);
 101         spin_unlock_bh(&expired_lock_thread.elt_lock);
 102
 103         RETURN(need_to_run);
 104 }
 105
 106 static int expired_lock_main(void *arg)
 107 {
 108         struct list_head *expired = &expired_lock_thread.elt_expired_locks;
 109         struct l_wait_info lwi = { 0 };
 110         unsigned long flags;
 111
 112         ENTRY;
 113         lock_kernel();
 114         kportal_daemonize("ldlm_elt");
 115
 116         SIGNAL_MASK_LOCK(current, flags);
 117         sigfillset(&current->blocked);
 118         RECALC_SIGPENDING;
 119         SIGNAL_MASK_UNLOCK(current, flags);
 120
 121         unlock_kernel();
 122
 123         expired_lock_thread.elt_state = ELT_READY;
 124         wake_up(&expired_lock_thread.elt_waitq);
 125
 126         while (1) {
 127                 l_wait_event(expired_lock_thread.elt_waitq,
 128                              have_expired_locks() ||
 129                              expired_lock_thread.elt_state == ELT_TERMINATE,
 130                              &lwi);
 131
 132                 spin_lock_bh(&expired_lock_thread.elt_lock);
 133                 while (!list_empty(expired)) {
 134                         struct obd_export *export;
 135                         struct ldlm_lock *lock;
 136
 137                         lock = list_entry(expired->next, struct ldlm_lock,
 138                                           l_pending_chain);
 139                         if ((void *)lock < LP_POISON + PAGE_SIZE &&
 140                             (void *)lock >= LP_POISON) {
 141                                 CERROR("free lock on elt list %p\n", lock);
 142                                 LBUG();
 143                         }
 144                         list_del_init(&lock->l_pending_chain);
 145                         if ((void *)lock->l_export < LP_POISON + PAGE_SIZE &&
 146                             (void *)lock->l_export >= LP_POISON + PAGE_SIZE) {
 147                                 CERROR("lock with free export on elt list %p\n",
 148                                        export);
 149                                 lock->l_export = NULL;
 150                                 LDLM_ERROR(lock, "free export\n");
 151                                 continue;
 152                         }
 153                         export = class_export_get(lock->l_export);
 154                         spin_unlock_bh(&expired_lock_thread.elt_lock);
 155
 156                         ptlrpc_fail_export(export);
 157                         class_export_put(export);
 158                         spin_lock_bh(&expired_lock_thread.elt_lock);
 159                 }
 160                 spin_unlock_bh(&expired_lock_thread.elt_lock);
 161
 162                 if (expired_lock_thread.elt_state == ELT_TERMINATE)
 163                         break;
 164         }
 165
 166         expired_lock_thread.elt_state = ELT_STOPPED;
 167         wake_up(&expired_lock_thread.elt_waitq);
 168         RETURN(0);
 169 }
 170
 171 static void waiting_locks_callback(unsigned long unused)
 172 {
 173         struct ldlm_lock *lock;
 174         char str[PTL_NALFMT_SIZE];
 175
 176         spin_lock_bh(&waiting_locks_spinlock);
 177         while (!list_empty(&waiting_locks_list)) {
 178                 lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
 179                                   l_pending_chain);
 180
 181                 if (lock->l_callback_timeout > jiffies)
 182                         break;
 183
 184                 LDLM_ERROR(lock, "lock callback timer expired: evicting client "
 185                            "%s@%s nid %s ",
 186                            lock->l_export->exp_client_uuid.uuid,
 187                            lock->l_export->exp_connection->c_remote_uuid.uuid,
 188                            ptlrpc_peernid2str(&lock->l_export->exp_connection->c_peer, str));
 189
 190                 spin_lock_bh(&expired_lock_thread.elt_lock);
 191                 list_del(&lock->l_pending_chain);
 192                 list_add(&lock->l_pending_chain,
 193                          &expired_lock_thread.elt_expired_locks);
 194                 spin_unlock_bh(&expired_lock_thread.elt_lock);
 195                 wake_up(&expired_lock_thread.elt_waitq);
 196         }
 197
 198         /*
 199          * Make sure the timer will fire again if we have any locks
 200          * left.
 201          */
 202         if (!list_empty(&waiting_locks_list)) {
 203                 unsigned long timeout_rounded;
 204                 lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
 205                                   l_pending_chain);
 206                 timeout_rounded = round_timeout(lock->l_callback_timeout);
 207                 mod_timer(&waiting_locks_timer, timeout_rounded);
 208         }
 209         spin_unlock_bh(&waiting_locks_spinlock);
 210 }
 211
 212 /*
 213  * Indicate that we're waiting for a client to call us back cancelling a given
 214  * lock.  We add it to the pending-callback chain, and schedule the lock-timeout
 215  * timer to fire appropriately.  (We round up to the next second, to avoid
 216  * floods of timer firings during periods of high lock contention and traffic).
 217  */
 218 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
 219 {
 220         unsigned long timeout_rounded;
 221
 222         spin_lock_bh(&waiting_locks_spinlock);
 223         if (!list_empty(&lock->l_pending_chain)) {
 224                 LDLM_DEBUG(lock, "not re-adding to wait list");
 225                 spin_unlock_bh(&waiting_locks_spinlock);
 226                 return 0;
 227         }
 228         LDLM_DEBUG(lock, "adding to wait list");
 229
 230         lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2);
 231
 232         timeout_rounded = round_timeout(lock->l_callback_timeout);
 233
 234         if (timeout_rounded < waiting_locks_timer.expires ||
 235             !timer_pending(&waiting_locks_timer)) {
 236                 mod_timer(&waiting_locks_timer, timeout_rounded);
 237         }
 238         list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
 239         spin_unlock_bh(&waiting_locks_spinlock);
 240         return 1;
 241 }
 242
 243 /*
 244  * Remove a lock from the pending list, likely because it had its cancellation
 245  * callback arrive without incident.  This adjusts the lock-timeout timer if
 246  * needed.  Returns 0 if the lock wasn't pending after all, 1 if it was.
 247  */
 248 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 249 {
 250         struct list_head *list_next;
 251
 252         if (lock->l_export == NULL) {
 253                 /* We don't have a "waiting locks list" on clients. */
 254                 LDLM_DEBUG(lock, "client lock: no-op");
 255                 return 0;
 256         }
 257
 258         spin_lock_bh(&waiting_locks_spinlock);
 259
 260         if (list_empty(&lock->l_pending_chain)) {
 261                 spin_unlock_bh(&waiting_locks_spinlock);
 262                 LDLM_DEBUG(lock, "wasn't waiting");
 263                 return 0;
 264         }
 265
 266         list_next = lock->l_pending_chain.next;
 267         if (lock->l_pending_chain.prev == &waiting_locks_list) {
 268                 /* Removing the head of the list, adjust timer. */
 269                 if (list_next == &waiting_locks_list) {
 270                         /* No more, just cancel. */
 271                         del_timer(&waiting_locks_timer);
 272                 } else {
 273                         struct ldlm_lock *next;
 274                         next = list_entry(list_next, struct ldlm_lock,
 275                                           l_pending_chain);
 276                         mod_timer(&waiting_locks_timer,
 277                                   round_timeout(next->l_callback_timeout));
 278                 }
 279         }
 280
 281         /* the lock could already be expired, get the elt_lock also */
 282         spin_lock_bh(&expired_lock_thread.elt_lock);
 283         list_del_init(&lock->l_pending_chain);
 284         spin_unlock_bh(&expired_lock_thread.elt_lock);
 285
 286         spin_unlock_bh(&waiting_locks_spinlock);
 287         LDLM_DEBUG(lock, "removed");
 288         return 1;
 289 }
 290
 291 #else /* !__KERNEL__ */
 292
 293 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
 294 {
 295         RETURN(1);
 296 }
 297
 298 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 299 {
 300         RETURN(0);
 301 }
 302
 303 #endif /* __KERNEL__ */
 304
 305 static void ldlm_failed_ast(struct ldlm_lock *lock, int rc,const char *ast_type)
 306 {
 307         struct ptlrpc_connection *conn = lock->l_export->exp_connection;
 308         char str[PTL_NALFMT_SIZE];
 309
 310         LDLM_ERROR(lock, "%s AST failed (%d): evicting client %s@%s NID "LPX64
 311                    " (%s)", ast_type, rc, lock->l_export->exp_client_uuid.uuid,
 312                    conn->c_remote_uuid.uuid, conn->c_peer.peer_id.nid,
 313                    ptlrpc_peernid2str(&conn->c_peer, str));
 314
 315         ptlrpc_fail_export(lock->l_export);
 316 }
 317
 318 static int ldlm_handle_ast_error(struct ldlm_lock *lock,
 319                                  struct ptlrpc_request *req, int rc,
 320                                  const char *ast_type)
 321 {
 322         struct ptlrpc_peer *peer = &req->rq_import->imp_connection->c_peer;
 323         char str[PTL_NALFMT_SIZE];
 324
 325         if (rc == -ETIMEDOUT || rc == -EINTR || rc == -ENOTCONN) {
 326                 LASSERT(lock->l_export);
 327                 if (lock->l_export->exp_libclient) {
 328                         LDLM_DEBUG(lock, "%s AST to liblustre client (nid %s)"
 329                                    " timeout, just cancelling lock", ast_type,
 330                                    ptlrpc_peernid2str(peer, str));
 331                         ldlm_lock_cancel(lock);
 332                         rc = -ERESTART;
 333                 } else {
 334                         ldlm_del_waiting_lock(lock);
 335                         ldlm_failed_ast(lock, rc, ast_type);
 336                 }
 337         } else if (rc) {
 338                 if (rc == -EINVAL)
 339                         LDLM_DEBUG(lock, "client (nid %s) returned %d"
 340                                    " from %s AST - normal race",
 341                                    ptlrpc_peernid2str(peer, str),
 342                                    req->rq_repmsg->status, ast_type);
 343                 else
 344                         LDLM_ERROR(lock, "client (nid %s) returned %d "
 345                                    "from %s AST", ptlrpc_peernid2str(peer, str),
 346                                    (req->rq_repmsg != NULL) ?
 347                                    req->rq_repmsg->status : 0, ast_type);
 348                 ldlm_lock_cancel(lock);
 349                 /* Server-side AST functions are called from ldlm_reprocess_all,
 350                  * which needs to be told to please restart its reprocessing. */
 351                 rc = -ERESTART;
 352         }
 353
 354         return rc;
 355 }
 356
 357 int ldlm_server_blocking_ast(struct ldlm_lock *lock,
 358                              struct ldlm_lock_desc *desc,
 359                              void *data, int flag)
 360 {
 361         struct ldlm_request *body;
 362         struct ptlrpc_request *req;
 363         int rc = 0, size = sizeof(*body);
 364         ENTRY;
 365
 366         if (flag == LDLM_CB_CANCELING) {
 367                 /* Don't need to do anything here. */
 368                 RETURN(0);
 369         }
 370
 371         LASSERT(lock);
 372
 373         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 374         if (lock->l_granted_mode != lock->l_req_mode) {
 375                 /* this blocking AST will be communicated as part of the
 376                  * completion AST instead */
 377                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 378                 LDLM_DEBUG(lock, "lock not granted, not sending blocking AST");
 379                 RETURN(0);
 380         }
 381
 382         if (lock->l_destroyed) {
 383                 /* What's the point? */
 384                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 385                 RETURN(0);
 386         }
 387
 388 #if 0
 389         if (CURRENT_SECONDS - lock->l_export->exp_last_request_time > 30){
 390                 ldlm_failed_ast(lock, -ETIMEDOUT, "Not-attempted blocking");
 391                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 392                 RETURN(-ETIMEDOUT);
 393         }
 394 #endif
 395
 396         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 397                               LDLM_BL_CALLBACK, 1, &size, NULL);
 398         if (req == NULL) {
 399                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 400                 RETURN(-ENOMEM);
 401         }
 402
 403         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 404         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 405                sizeof(body->lock_handle1));
 406         memcpy(&body->lock_desc, desc, sizeof(*desc));
 407         body->lock_flags |= (lock->l_flags & LDLM_AST_FLAGS);
 408
 409         LDLM_DEBUG(lock, "server preparing blocking AST");
 410         req->rq_replen = lustre_msg_size(0, NULL);
 411
 412         if (lock->l_granted_mode == lock->l_req_mode)
 413                 ldlm_add_waiting_lock(lock);
 414         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 415
 416         req->rq_send_state = LUSTRE_IMP_FULL;
 417         req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */
 418         rc = ptlrpc_queue_wait(req);
 419         if (rc != 0)
 420                 rc = ldlm_handle_ast_error(lock, req, rc, "blocking");
 421
 422         ptlrpc_req_finished(req);
 423
 424         RETURN(rc);
 425 }
 426
 427 /* XXX copied from ptlrpc/service.c */
 428 static long timeval_sub(struct timeval *large, struct timeval *small)
 429 {
 430         return (large->tv_sec - small->tv_sec) * 1000000 +
 431                 (large->tv_usec - small->tv_usec);
 432 }
 433
 434 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data)
 435 {
 436         struct ldlm_request *body;
 437         struct ptlrpc_request *req;
 438         struct timeval granted_time;
 439         long total_enqueue_wait;
 440         int rc = 0, size[2] = {sizeof(*body)}, buffers = 1;
 441         ENTRY;
 442
 443         LASSERT(lock != NULL);
 444
 445         do_gettimeofday(&granted_time);
 446         total_enqueue_wait = timeval_sub(&granted_time, &lock->l_enqueued_time);
 447
 448         if (total_enqueue_wait / 1000000 > obd_timeout)
 449                 LDLM_ERROR(lock, "enqueue wait took %luus", total_enqueue_wait);
 450
 451         down(&lock->l_resource->lr_lvb_sem);
 452         if (lock->l_resource->lr_lvb_len) {
 453                 buffers = 2;
 454                 size[1] = lock->l_resource->lr_lvb_len;
 455         }
 456         up(&lock->l_resource->lr_lvb_sem);
 457
 458         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 459                               LDLM_CP_CALLBACK, buffers, size, NULL);
 460         if (req == NULL)
 461                 RETURN(-ENOMEM);
 462
 463         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 464         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 465                sizeof(body->lock_handle1));
 466         body->lock_flags = flags;
 467         ldlm_lock2desc(lock, &body->lock_desc);
 468
 469         if (buffers == 2) {
 470                 void *lvb;
 471
 472                 down(&lock->l_resource->lr_lvb_sem);
 473                 lvb = lustre_msg_buf(req->rq_reqmsg, 1,
 474                                      lock->l_resource->lr_lvb_len);
 475                 memcpy(lvb, lock->l_resource->lr_lvb_data,
 476                        lock->l_resource->lr_lvb_len);
 477                 up(&lock->l_resource->lr_lvb_sem);
 478         }
 479
 480         LDLM_DEBUG(lock, "server preparing completion AST (after %ldus wait)",
 481                    total_enqueue_wait);
 482         req->rq_replen = lustre_msg_size(0, NULL);
 483
 484         req->rq_send_state = LUSTRE_IMP_FULL;
 485         req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */
 486
 487         /* We only send real blocking ASTs after the lock is granted */
 488         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 489         if (lock->l_flags & LDLM_FL_AST_SENT) {
 490                 body->lock_flags |= LDLM_FL_AST_SENT;
 491                 ldlm_add_waiting_lock(lock); /* start the lock-timeout clock */
 492         }
 493         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 494
 495         rc = ptlrpc_queue_wait(req);
 496         if (rc != 0)
 497                 rc = ldlm_handle_ast_error(lock, req, rc, "completion");
 498
 499         ptlrpc_req_finished(req);
 500
 501         RETURN(rc);
 502 }
 503
 504 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data)
 505 {
 506         struct ldlm_resource *res = lock->l_resource;
 507         struct ldlm_request *body;
 508         struct ptlrpc_request *req;
 509         int rc = 0, size = sizeof(*body);
 510         ENTRY;
 511
 512         LASSERT(lock != NULL);
 513
 514         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 515                               LDLM_GL_CALLBACK, 1, &size, NULL);
 516         if (req == NULL)
 517                 RETURN(-ENOMEM);
 518
 519         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 520         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 521                sizeof(body->lock_handle1));
 522         ldlm_lock2desc(lock, &body->lock_desc);
 523
 524         down(&lock->l_resource->lr_lvb_sem);
 525         size = lock->l_resource->lr_lvb_len;
 526         up(&lock->l_resource->lr_lvb_sem);
 527         req->rq_replen = lustre_msg_size(1, &size);
 528
 529         req->rq_send_state = LUSTRE_IMP_FULL;
 530         req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */
 531
 532         rc = ptlrpc_queue_wait(req);
 533         if (rc == -ELDLM_NO_LOCK_DATA)
 534                 LDLM_DEBUG(lock, "lost race - client has a lock but no inode");
 535         else if (rc != 0)
 536                 rc = ldlm_handle_ast_error(lock, req, rc, "glimpse");
 537         else
 538                 rc = res->lr_namespace->ns_lvbo->lvbo_update
 539                         (res, req->rq_repmsg, 0, 1);
 540         ptlrpc_req_finished(req);
 541         RETURN(rc);
 542 }
 543
 544 static struct ldlm_lock *
 545 find_existing_lock(struct obd_export *exp, struct lustre_handle *remote_hdl)
 546 {
 547         struct obd_device *obd = exp->exp_obd;
 548         struct list_head *iter;
 549
 550         l_lock(&obd->obd_namespace->ns_lock);
 551         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
 552                 struct ldlm_lock *lock;
 553                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
 554                 if (lock->l_remote_handle.cookie == remote_hdl->cookie) {
 555                         LDLM_LOCK_GET(lock);
 556                         l_unlock(&obd->obd_namespace->ns_lock);
 557                         return lock;
 558                 }
 559         }
 560         l_unlock(&obd->obd_namespace->ns_lock);
 561         return NULL;
 562 }
 563
 564
 565 int ldlm_handle_enqueue(struct ptlrpc_request *req,
 566                         ldlm_completion_callback completion_callback,
 567                         ldlm_blocking_callback blocking_callback,
 568                         ldlm_glimpse_callback glimpse_callback)
 569 {
 570         struct obd_device *obddev = req->rq_export->exp_obd;
 571         struct ldlm_reply *dlm_rep;
 572         struct ldlm_request *dlm_req;
 573         int rc = 0, size[2] = {sizeof(*dlm_rep)};
 574         __u32 flags;
 575         ldlm_error_t err = ELDLM_OK;
 576         struct ldlm_lock *lock = NULL;
 577         void *cookie = NULL;
 578         ENTRY;
 579
 580         LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
 581
 582         dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
 583                                       lustre_swab_ldlm_request);
 584         if (dlm_req == NULL) {
 585                 CERROR ("Can't unpack dlm_req\n");
 586                 GOTO(out, rc = -EFAULT);
 587         }
 588
 589         flags = dlm_req->lock_flags;
 590
 591         LASSERT(req->rq_export);
 592
 593         if (flags & LDLM_FL_REPLAY) {
 594                 lock = find_existing_lock(req->rq_export,
 595                                           &dlm_req->lock_handle1);
 596                 if (lock != NULL) {
 597                         DEBUG_REQ(D_HA, req, "found existing lock cookie "LPX64,
 598                                   lock->l_handle.h_cookie);
 599                         GOTO(existing_lock, rc = 0);
 600                 }
 601         }
 602
 603         /* The lock's callback data might be set in the policy function */
 604         lock = ldlm_lock_create(obddev->obd_namespace, &dlm_req->lock_handle2,
 605                                 dlm_req->lock_desc.l_resource.lr_name,
 606                                 dlm_req->lock_desc.l_resource.lr_type,
 607                                 dlm_req->lock_desc.l_req_mode,
 608                                 blocking_callback, completion_callback,
 609                                 glimpse_callback, NULL, 0);
 610         if (!lock)
 611                 GOTO(out, rc = -ENOMEM);
 612
 613         do_gettimeofday(&lock->l_enqueued_time);
 614         memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
 615                sizeof(lock->l_remote_handle));
 616         LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
 617
 618         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_BLOCKED, obd_timeout * 2);
 619         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 620         if (req->rq_export->exp_failed) {
 621                 LDLM_ERROR(lock,"lock on destroyed export %p\n",req->rq_export);
 622                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 623                 GOTO(out, rc = -ENOTCONN);
 624         }
 625         lock->l_export = class_export_get(req->rq_export);
 626         list_add(&lock->l_export_chain,
 627                  &lock->l_export->exp_ldlm_data.led_held_locks);
 628         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 629
 630 existing_lock:
 631
 632         if (flags & LDLM_FL_HAS_INTENT) {
 633                 /* In this case, the reply buffer is allocated deep in
 634                  * local_lock_enqueue by the policy function. */
 635                 cookie = req;
 636         } else {
 637                 int buffers = 1;
 638
 639                 down(&lock->l_resource->lr_lvb_sem);
 640                 if (lock->l_resource->lr_lvb_len) {
 641                         size[1] = lock->l_resource->lr_lvb_len;
 642                         buffers = 2;
 643                 }
 644                 up(&lock->l_resource->lr_lvb_sem);
 645
 646                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR))
 647                         GOTO(out, rc = -ENOMEM);
 648
 649                 rc = lustre_pack_reply(req, buffers, size, NULL);
 650                 if (rc)
 651                         GOTO(out, rc);
 652         }
 653
 654         if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN)
 655                 memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data,
 656                        sizeof(ldlm_policy_data_t));
 657         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT)
 658                 memcpy(&lock->l_req_extent, &lock->l_policy_data.l_extent,
 659                        sizeof(lock->l_req_extent));
 660
 661         err = ldlm_lock_enqueue(obddev->obd_namespace, &lock, cookie, &flags);
 662         if (err)
 663                 GOTO(out, err);
 664
 665         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
 666         dlm_rep->lock_flags = flags;
 667
 668         ldlm_lock2desc(lock, &dlm_rep->lock_desc);
 669         ldlm_lock2handle(lock, &dlm_rep->lock_handle);
 670
 671         /* We never send a blocking AST until the lock is granted, but
 672          * we can tell it right now */
 673         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 674         if (lock->l_flags & LDLM_FL_AST_SENT) {
 675                 dlm_rep->lock_flags |= LDLM_FL_AST_SENT;
 676                 if (lock->l_granted_mode == lock->l_req_mode)
 677                         ldlm_add_waiting_lock(lock);
 678         }
 679         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 680
 681         EXIT;
 682  out:
 683         req->rq_status = err;
 684         if (req->rq_reply_state == NULL) {
 685                 err = lustre_pack_reply(req, 0, NULL, NULL);
 686                 if (rc == 0)
 687                         rc = err;
 688                 req->rq_status = rc;
 689         }
 690
 691         /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
 692          * ldlm_reprocess_all.  If this moves, revisit that code. -phil */
 693         if (lock) {
 694                 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
 695                            "(err=%d, rc=%d)", err, rc);
 696
 697                 if (rc == 0) {
 698                         down(&lock->l_resource->lr_lvb_sem);
 699                         size[1] = lock->l_resource->lr_lvb_len;
 700                         if (size[1] > 0) {
 701                                 void *lvb = lustre_msg_buf(req->rq_repmsg,
 702                                                            1, size[1]);
 703                                 LASSERTF(lvb != NULL, "req %p, lock %p\n",
 704                                          req, lock);
 705
 706                                 memcpy(lvb, lock->l_resource->lr_lvb_data,
 707                                        size[1]);
 708                         }
 709                         up(&lock->l_resource->lr_lvb_sem);
 710                 } else {
 711                         ldlm_lock_destroy(lock);
 712                 }
 713
 714                 if (!err && dlm_req->lock_desc.l_resource.lr_type != LDLM_FLOCK)
 715                         ldlm_reprocess_all(lock->l_resource);
 716                 LDLM_LOCK_PUT(lock);
 717         }
 718         LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p, rc %d)",
 719                           lock, rc);
 720
 721         return rc;
 722 }
 723
 724 int ldlm_handle_convert(struct ptlrpc_request *req)
 725 {
 726         struct ldlm_request *dlm_req;
 727         struct ldlm_reply *dlm_rep;
 728         struct ldlm_lock *lock;
 729         int rc, size = sizeof(*dlm_rep);
 730         ENTRY;
 731
 732         dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
 733                                       lustre_swab_ldlm_request);
 734         if (dlm_req == NULL) {
 735                 CERROR ("Can't unpack dlm_req\n");
 736                 RETURN (-EFAULT);
 737         }
 738
 739         rc = lustre_pack_reply(req, 1, &size, NULL);
 740         if (rc) {
 741                 CERROR("out of memory\n");
 742                 RETURN(-ENOMEM);
 743         }
 744         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
 745         dlm_rep->lock_flags = dlm_req->lock_flags;
 746
 747         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 748         if (!lock) {
 749                 req->rq_status = EINVAL;
 750         } else {
 751                 LDLM_DEBUG(lock, "server-side convert handler START");
 752                 ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode,
 753                                   &dlm_rep->lock_flags);
 754                 if (ldlm_del_waiting_lock(lock))
 755                         CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock);
 756                 req->rq_status = 0;
 757         }
 758
 759         if (lock) {
 760                 ldlm_reprocess_all(lock->l_resource);
 761                 LDLM_DEBUG(lock, "server-side convert handler END");
 762                 LDLM_LOCK_PUT(lock);
 763         } else
 764                 LDLM_DEBUG_NOLOCK("server-side convert handler END");
 765
 766         RETURN(0);
 767 }
 768
 769 int ldlm_handle_cancel(struct ptlrpc_request *req)
 770 {
 771         struct ldlm_request *dlm_req;
 772         struct ldlm_lock *lock;
 773         struct ldlm_resource *res;
 774         int rc;
 775         ENTRY;
 776
 777         dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
 778                                       lustre_swab_ldlm_request);
 779         if (dlm_req == NULL) {
 780                 CERROR("bad request buffer for cancel\n");
 781                 RETURN(-EFAULT);
 782         }
 783
 784         rc = lustre_pack_reply(req, 0, NULL, NULL);
 785         if (rc) {
 786                 CERROR("out of memory\n");
 787                 RETURN(-ENOMEM);
 788         }
 789
 790         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 791         if (!lock) {
 792                 CERROR("received cancel for unknown lock cookie "LPX64
 793                        " from client %s id %s\n",
 794                        dlm_req->lock_handle1.cookie,
 795                        req->rq_export->exp_client_uuid.uuid,
 796                        req->rq_peerstr);
 797                 LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock "
 798                                   "(cookie "LPU64")",
 799                                   dlm_req->lock_handle1.cookie);
 800                 req->rq_status = ESTALE;
 801         } else {
 802                 LDLM_DEBUG(lock, "server-side cancel handler START");
 803                 res = lock->l_resource;
 804                 if (res && res->lr_namespace->ns_lvbo &&
 805                     res->lr_namespace->ns_lvbo->lvbo_update) {
 806                         (void)res->lr_namespace->ns_lvbo->lvbo_update
 807                                 (res, NULL, 0, 0);
 808                                 //(res, req->rq_reqmsg, 1);
 809                 }
 810
 811                 ldlm_lock_cancel(lock);
 812                 if (ldlm_del_waiting_lock(lock))
 813                         CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock);
 814                 req->rq_status = rc;
 815         }
 816
 817         if (ptlrpc_reply(req) != 0)
 818                 LBUG();
 819
 820         if (lock) {
 821                 ldlm_reprocess_all(lock->l_resource);
 822                 LDLM_DEBUG(lock, "server-side cancel handler END");
 823                 LDLM_LOCK_PUT(lock);
 824         }
 825
 826         RETURN(0);
 827 }
 828
 829 void ldlm_handle_bl_callback(struct ldlm_namespace *ns,
 830                              struct ldlm_lock_desc *ld, struct ldlm_lock *lock)
 831 {
 832         int do_ast;
 833         ENTRY;
 834
 835         l_lock(&ns->ns_lock);
 836         LDLM_DEBUG(lock, "client blocking AST callback handler START");
 837
 838         lock->l_flags |= LDLM_FL_CBPENDING;
 839         do_ast = (!lock->l_readers && !lock->l_writers);
 840
 841         if (do_ast) {
 842                 LDLM_DEBUG(lock, "already unused, calling "
 843                            "callback (%p)", lock->l_blocking_ast);
 844                 if (lock->l_blocking_ast != NULL) {
 845                         l_unlock(&ns->ns_lock);
 846                         l_check_no_ns_lock(ns);
 847                         lock->l_blocking_ast(lock, ld, lock->l_ast_data,
 848                                              LDLM_CB_BLOCKING);
 849                         l_lock(&ns->ns_lock);
 850                 }
 851         } else {
 852                 LDLM_DEBUG(lock, "Lock still has references, will be"
 853                            " cancelled later");
 854         }
 855
 856         LDLM_DEBUG(lock, "client blocking callback handler END");
 857         l_unlock(&ns->ns_lock);
 858         LDLM_LOCK_PUT(lock);
 859         EXIT;
 860 }
 861
 862 static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
 863                                     struct ldlm_namespace *ns,
 864                                     struct ldlm_request *dlm_req,
 865                                     struct ldlm_lock *lock)
 866 {
 867         LIST_HEAD(ast_list);
 868         ENTRY;
 869
 870         l_lock(&ns->ns_lock);
 871         LDLM_DEBUG(lock, "client completion callback handler START");
 872
 873         /* If we receive the completion AST before the actual enqueue returned,
 874          * then we might need to switch lock modes, resources, or extents. */
 875         if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
 876                 lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
 877                 LDLM_DEBUG(lock, "completion AST, new lock mode");
 878         }
 879
 880         if (lock->l_resource->lr_type != LDLM_PLAIN) {
 881                 memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data,
 882                        sizeof(lock->l_policy_data));
 883                 LDLM_DEBUG(lock, "completion AST, new policy data");
 884         }
 885
 886         ldlm_resource_unlink_lock(lock);
 887         if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
 888                    &lock->l_resource->lr_name,
 889                    sizeof(lock->l_resource->lr_name)) != 0) {
 890                 ldlm_lock_change_resource(ns, lock,
 891                                          dlm_req->lock_desc.l_resource.lr_name);
 892                 LDLM_DEBUG(lock, "completion AST, new resource");
 893         }
 894
 895         if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
 896                 lock->l_flags |= LDLM_FL_CBPENDING;
 897                 LDLM_DEBUG(lock, "completion AST includes blocking AST");
 898         }
 899
 900         if (lock->l_lvb_len) {
 901                 void *lvb;
 902                 lvb = lustre_swab_reqbuf(req, 1, lock->l_lvb_len,
 903                                          lock->l_lvb_swabber);
 904                 if (lvb == NULL) {
 905                         LDLM_ERROR(lock, "completion AST did not contain "
 906                                    "expected LVB!");
 907                 } else {
 908                         memcpy(lock->l_lvb_data, lvb, lock->l_lvb_len);
 909                 }
 910         }
 911
 912         lock->l_resource->lr_tmp = &ast_list;
 913         ldlm_grant_lock(lock, req, sizeof(*req), 1);
 914         lock->l_resource->lr_tmp = NULL;
 915         LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
 916         l_unlock(&ns->ns_lock);
 917         LDLM_LOCK_PUT(lock);
 918
 919         ldlm_run_ast_work(ns, &ast_list);
 920
 921         LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
 922                           lock);
 923         EXIT;
 924 }
 925
 926 static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
 927                                     struct ldlm_namespace *ns,
 928                                     struct ldlm_request *dlm_req,
 929                                     struct ldlm_lock *lock)
 930 {
 931         int rc = -ENOSYS;
 932         ENTRY;
 933
 934         l_lock(&ns->ns_lock);
 935         LDLM_DEBUG(lock, "client glimpse AST callback handler");
 936
 937         if (lock->l_glimpse_ast != NULL) {
 938                 l_unlock(&ns->ns_lock);
 939                 l_check_no_ns_lock(ns);
 940                 rc = lock->l_glimpse_ast(lock, req);
 941                 l_lock(&ns->ns_lock);
 942         }
 943
 944         if (req->rq_repmsg != NULL) {
 945                 ptlrpc_reply(req);
 946         } else {
 947                 req->rq_status = rc;
 948                 ptlrpc_error(req);
 949         }
 950
 951         l_unlock(&ns->ns_lock);
 952         if (lock->l_granted_mode == LCK_PW &&
 953             !lock->l_readers && !lock->l_writers &&
 954             time_after(jiffies, lock->l_last_used + 10 * HZ)) {
 955                 if (ldlm_bl_to_thread(ns, NULL, lock))
 956                         ldlm_handle_bl_callback(ns, NULL, lock);
 957                 EXIT;
 958                 return;
 959         }
 960
 961         LDLM_LOCK_PUT(lock);
 962         EXIT;
 963 }
 964
 965 static int ldlm_callback_reply(struct ptlrpc_request *req, int rc)
 966 {
 967         req->rq_status = rc;
 968         if (req->rq_reply_state == NULL) {
 969                 rc = lustre_pack_reply(req, 0, NULL, NULL);
 970                 if (rc)
 971                         return rc;
 972         }
 973         return ptlrpc_reply(req);
 974 }
 975
 976 int ldlm_bl_to_thread(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
 977                       struct ldlm_lock *lock)
 978 {
 979 #ifdef __KERNEL__
 980         struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
 981         struct ldlm_bl_work_item *blwi;
 982         ENTRY;
 983
 984         OBD_ALLOC(blwi, sizeof(*blwi));
 985         if (blwi == NULL)
 986                 RETURN(-ENOMEM);
 987
 988         blwi->blwi_ns = ns;
 989         if (ld != NULL)
 990                 blwi->blwi_ld = *ld;
 991         blwi->blwi_lock = lock;
 992
 993         spin_lock(&blp->blp_lock);
 994         list_add_tail(&blwi->blwi_entry, &blp->blp_list);
 995         wake_up(&blp->blp_waitq);
 996         spin_unlock(&blp->blp_lock);
 997
 998         RETURN(0);
 999 #else
1000         RETURN(-ENOSYS);
1001 #endif
1002 }
1003
1004 static int ldlm_callback_handler(struct ptlrpc_request *req)
1005 {
1006         struct ldlm_namespace *ns;
1007         struct ldlm_request *dlm_req;
1008         struct ldlm_lock *lock;
1009         int rc;
1010         ENTRY;
1011
1012         /* Requests arrive in sender's byte order.  The ptlrpc service
1013          * handler has already checked and, if necessary, byte-swapped the
1014          * incoming request message body, but I am responsible for the
1015          * message buffers. */
1016
1017         if (req->rq_export == NULL) {
1018                 struct ldlm_request *dlm_req;
1019
1020                 CDEBUG(D_RPCTRACE, "operation %d from %s with bad "
1021                        "export cookie "LPX64"; this is "
1022                        "normal if this node rebooted with a lock held\n",
1023                        req->rq_reqmsg->opc,
1024                        req->rq_peerstr,
1025                        req->rq_reqmsg->handle.cookie);
1026
1027                 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
1028                                              lustre_swab_ldlm_request);
1029                 if (dlm_req != NULL)
1030                         CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n",
1031                                dlm_req->lock_handle1.cookie);
1032
1033                 ldlm_callback_reply(req, -ENOTCONN);
1034                 RETURN(0);
1035         }
1036
1037         LASSERT(req->rq_export != NULL);
1038         LASSERT(req->rq_export->exp_obd != NULL);
1039
1040         switch(req->rq_reqmsg->opc) {
1041         case LDLM_BL_CALLBACK:
1042                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
1043                 break;
1044         case LDLM_CP_CALLBACK:
1045                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0);
1046                 break;
1047         case LDLM_GL_CALLBACK:
1048                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_GL_CALLBACK, 0);
1049                 break;
1050         case OBD_LOG_CANCEL:
1051                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1052                 rc = llog_origin_handle_cancel(req);
1053                 ldlm_callback_reply(req, rc);
1054                 RETURN(0);
1055         case LLOG_ORIGIN_HANDLE_CREATE:
1056                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1057                 rc = llog_origin_handle_create(req);
1058                 ldlm_callback_reply(req, rc);
1059                 RETURN(0);
1060         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
1061                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1062                 rc = llog_origin_handle_next_block(req);
1063                 ldlm_callback_reply(req, rc);
1064                 RETURN(0);
1065         case LLOG_ORIGIN_HANDLE_READ_HEADER:
1066                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1067                 rc = llog_origin_handle_read_header(req);
1068                 ldlm_callback_reply(req, rc);
1069                 RETURN(0);
1070         case LLOG_ORIGIN_HANDLE_CLOSE:
1071                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1072                 rc = llog_origin_handle_close(req);
1073                 ldlm_callback_reply(req, rc);
1074                 RETURN(0);
1075         default:
1076                 CERROR("unknown opcode %u\n", req->rq_reqmsg->opc);
1077                 ldlm_callback_reply(req, -EPROTO);
1078                 RETURN(0);
1079         }
1080
1081         ns = req->rq_export->exp_obd->obd_namespace;
1082         LASSERT(ns != NULL);
1083
1084         dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
1085                                       lustre_swab_ldlm_request);
1086         if (dlm_req == NULL) {
1087                 CERROR ("can't unpack dlm_req\n");
1088                 ldlm_callback_reply (req, -EPROTO);
1089                 RETURN (0);
1090         }
1091
1092         lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle1);
1093         if (!lock) {
1094                 CDEBUG(D_INODE, "callback on lock "LPX64" - lock disappeared\n",
1095                        dlm_req->lock_handle1.cookie);
1096                 ldlm_callback_reply(req, -EINVAL);
1097                 RETURN(0);
1098         }
1099
1100         /* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */
1101         lock->l_flags |= (dlm_req->lock_flags & LDLM_AST_FLAGS);
1102
1103         /* We want the ost thread to get this reply so that it can respond
1104          * to ost requests (write cache writeback) that might be triggered
1105          * in the callback.
1106          *
1107          * But we'd also like to be able to indicate in the reply that we're
1108          * cancelling right now, because it's unused, or have an intent result
1109          * in the reply, so we might have to push the responsibility for sending
1110          * the reply down into the AST handlers, alas. */
1111
1112         switch (req->rq_reqmsg->opc) {
1113         case LDLM_BL_CALLBACK:
1114                 CDEBUG(D_INODE, "blocking ast\n");
1115                 ldlm_callback_reply(req, 0);
1116                 if (ldlm_bl_to_thread(ns, &dlm_req->lock_desc, lock))
1117                         ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock);
1118                 break;
1119         case LDLM_CP_CALLBACK:
1120                 CDEBUG(D_INODE, "completion ast\n");
1121                 ldlm_callback_reply(req, 0);
1122                 ldlm_handle_cp_callback(req, ns, dlm_req, lock);
1123                 break;
1124         case LDLM_GL_CALLBACK:
1125                 CDEBUG(D_INODE, "glimpse ast\n");
1126                 ldlm_handle_gl_callback(req, ns, dlm_req, lock);
1127                 break;
1128         default:
1129                 LBUG();                         /* checked above */
1130         }
1131
1132         RETURN(0);
1133 }
1134
1135 static int ldlm_cancel_handler(struct ptlrpc_request *req)
1136 {
1137         int rc;
1138         ENTRY;
1139
1140         /* Requests arrive in sender's byte order.  The ptlrpc service
1141          * handler has already checked and, if necessary, byte-swapped the
1142          * incoming request message body, but I am responsible for the
1143          * message buffers. */
1144
1145         if (req->rq_export == NULL) {
1146                 struct ldlm_request *dlm_req;
1147                 CERROR("operation %d with bad export from %s\n",
1148                        req->rq_reqmsg->opc,
1149                        req->rq_peerstr);
1150                 CERROR("--> export cookie: "LPX64"\n",
1151                        req->rq_reqmsg->handle.cookie);
1152                 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
1153                                              lustre_swab_ldlm_request);
1154                 if (dlm_req != NULL)
1155                         ldlm_lock_dump_handle(D_ERROR, &dlm_req->lock_handle1);
1156                 ldlm_callback_reply(req, -ENOTCONN);
1157                 RETURN(0);
1158         }
1159
1160         switch (req->rq_reqmsg->opc) {
1161
1162         /* XXX FIXME move this back to mds/handler.c, bug 249 */
1163         case LDLM_CANCEL:
1164                 CDEBUG(D_INODE, "cancel\n");
1165                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1166                 rc = ldlm_handle_cancel(req);
1167                 if (rc)
1168                         break;
1169                 RETURN(0);
1170
1171         default:
1172                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
1173                 ldlm_callback_reply(req, -EINVAL);
1174         }
1175
1176         RETURN(0);
1177 }
1178
1179 #ifdef __KERNEL__
1180 static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp)
1181 {
1182         struct ldlm_bl_work_item *blwi = NULL;
1183
1184         spin_lock(&blp->blp_lock);
1185         if (!list_empty(&blp->blp_list)) {
1186                 blwi = list_entry(blp->blp_list.next, struct ldlm_bl_work_item,
1187                                   blwi_entry);
1188                 list_del(&blwi->blwi_entry);
1189         }
1190         spin_unlock(&blp->blp_lock);
1191
1192         return blwi;
1193 }
1194
1195 struct ldlm_bl_thread_data {
1196         int                     bltd_num;
1197         struct ldlm_bl_pool     *bltd_blp;
1198 };
1199
1200 static int ldlm_bl_thread_main(void *arg)
1201 {
1202         struct ldlm_bl_thread_data *bltd = arg;
1203         struct ldlm_bl_pool *blp = bltd->bltd_blp;
1204         unsigned long flags;
1205         ENTRY;
1206
1207         /* XXX boiler-plate */
1208         {
1209                 char name[sizeof(current->comm)];
1210                 snprintf(name, sizeof(name) - 1, "ldlm_bl_%02d",
1211                          bltd->bltd_num);
1212                 kportal_daemonize(name);
1213         }
1214         SIGNAL_MASK_LOCK(current, flags);
1215         sigfillset(&current->blocked);
1216         RECALC_SIGPENDING;
1217         SIGNAL_MASK_UNLOCK(current, flags);
1218
1219         atomic_inc(&blp->blp_num_threads);
1220         complete(&blp->blp_comp);
1221
1222         while(1) {
1223                 struct l_wait_info lwi = { 0 };
1224                 struct ldlm_bl_work_item *blwi = NULL;
1225
1226                 l_wait_event_exclusive(blp->blp_waitq,
1227                                        (blwi = ldlm_bl_get_work(blp)) != NULL,
1228                                        &lwi);
1229
1230                 if (blwi->blwi_ns == NULL)
1231                         break;
1232
1233                 ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld,
1234                                         blwi->blwi_lock);
1235                 OBD_FREE(blwi, sizeof(*blwi));
1236         }
1237
1238         atomic_dec(&blp->blp_num_threads);
1239         complete(&blp->blp_comp);
1240         RETURN(0);
1241 }
1242
1243 #endif
1244
1245 static int ldlm_setup(void);
1246 static int ldlm_cleanup(int force);
1247
1248 int ldlm_get_ref(void)
1249 {
1250         int rc = 0;
1251         down(&ldlm_ref_sem);
1252         if (++ldlm_refcount == 1) {
1253                 rc = ldlm_setup();
1254                 if (rc)
1255                         ldlm_refcount--;
1256         }
1257         up(&ldlm_ref_sem);
1258
1259         RETURN(rc);
1260 }
1261
1262 void ldlm_put_ref(int force)
1263 {
1264         down(&ldlm_ref_sem);
1265         if (ldlm_refcount == 1) {
1266                 int rc = ldlm_cleanup(force);
1267                 if (rc)
1268                         CERROR("ldlm_cleanup failed: %d\n", rc);
1269                 else
1270                         ldlm_refcount--;
1271         } else {
1272                 ldlm_refcount--;
1273         }
1274         up(&ldlm_ref_sem);
1275
1276         EXIT;
1277 }
1278
1279 static int ldlm_setup(void)
1280 {
1281         struct ldlm_bl_pool *blp;
1282         int rc = 0;
1283 #ifdef __KERNEL__
1284         int i;
1285 #endif
1286         ENTRY;
1287
1288         if (ldlm_state != NULL)
1289                 RETURN(-EALREADY);
1290
1291         OBD_ALLOC(ldlm_state, sizeof(*ldlm_state));
1292         if (ldlm_state == NULL)
1293                 RETURN(-ENOMEM);
1294
1295 #ifdef __KERNEL__
1296         rc = ldlm_proc_setup();
1297         if (rc != 0)
1298                 GOTO(out_free, rc);
1299 #endif
1300
1301         ldlm_state->ldlm_cb_service =
1302                 ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE,
1303                                 LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1304                                 1500, ldlm_callback_handler, "ldlm_cbd",
1305                                 ldlm_svc_proc_dir);
1306
1307         if (!ldlm_state->ldlm_cb_service) {
1308                 CERROR("failed to start service\n");
1309                 GOTO(out_proc, rc = -ENOMEM);
1310         }
1311
1312         ldlm_state->ldlm_cancel_service =
1313                 ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE,
1314                                 LDLM_CANCEL_REQUEST_PORTAL,
1315                                 LDLM_CANCEL_REPLY_PORTAL, 30000,
1316                                 ldlm_cancel_handler, "ldlm_canceld",
1317                                 ldlm_svc_proc_dir);
1318
1319         if (!ldlm_state->ldlm_cancel_service) {
1320                 CERROR("failed to start service\n");
1321                 GOTO(out_proc, rc = -ENOMEM);
1322         }
1323
1324         OBD_ALLOC(blp, sizeof(*blp));
1325         if (blp == NULL)
1326                 GOTO(out_proc, rc = -ENOMEM);
1327         ldlm_state->ldlm_bl_pool = blp;
1328
1329         atomic_set(&blp->blp_num_threads, 0);
1330         init_waitqueue_head(&blp->blp_waitq);
1331         spin_lock_init(&blp->blp_lock);
1332
1333         INIT_LIST_HEAD(&blp->blp_list);
1334
1335 #ifdef __KERNEL__
1336         for (i = 0; i < LDLM_NUM_THREADS; i++) {
1337                 struct ldlm_bl_thread_data bltd = {
1338                         .bltd_num = i,
1339                         .bltd_blp = blp,
1340                 };
1341                 init_completion(&blp->blp_comp);
1342                 rc = kernel_thread(ldlm_bl_thread_main, &bltd, 0);
1343                 if (rc < 0) {
1344                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
1345                         GOTO(out_thread, rc);
1346                 }
1347                 wait_for_completion(&blp->blp_comp);
1348         }
1349
1350         rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cancel_service,
1351                                     LDLM_NUM_THREADS, "ldlm_cn");
1352         if (rc)
1353                 GOTO(out_thread, rc);
1354
1355         rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cb_service,
1356                                     LDLM_NUM_THREADS, "ldlm_cb");
1357         if (rc)
1358                 GOTO(out_thread, rc);
1359
1360         INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
1361         spin_lock_init(&expired_lock_thread.elt_lock);
1362         expired_lock_thread.elt_state = ELT_STOPPED;
1363         init_waitqueue_head(&expired_lock_thread.elt_waitq);
1364
1365         rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS);
1366         if (rc < 0) {
1367                 CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
1368                 GOTO(out_thread, rc);
1369         }
1370
1371         wait_event(expired_lock_thread.elt_waitq,
1372                    expired_lock_thread.elt_state == ELT_READY);
1373
1374         INIT_LIST_HEAD(&waiting_locks_list);
1375         spin_lock_init(&waiting_locks_spinlock);
1376         waiting_locks_timer.function = waiting_locks_callback;
1377         waiting_locks_timer.data = 0;
1378         init_timer(&waiting_locks_timer);
1379 #endif
1380
1381         RETURN(0);
1382
1383 #ifdef __KERNEL__
1384  out_thread:
1385         ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
1386         ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
1387 #endif
1388
1389  out_proc:
1390 #ifdef __KERNEL__
1391         ldlm_proc_cleanup();
1392  out_free:
1393 #endif
1394         OBD_FREE(ldlm_state, sizeof(*ldlm_state));
1395         ldlm_state = NULL;
1396         return rc;
1397 }
1398
1399 static int ldlm_cleanup(int force)
1400 {
1401 #ifdef __KERNEL__
1402         struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
1403 #endif
1404         ENTRY;
1405
1406         if (!list_empty(&ldlm_namespace_list)) {
1407                 CERROR("ldlm still has namespaces; clean these up first.\n");
1408                 ldlm_dump_all_namespaces(D_DLMTRACE);
1409                 RETURN(-EBUSY);
1410         }
1411
1412 #ifdef __KERNEL__
1413         while (atomic_read(&blp->blp_num_threads) > 0) {
1414                 struct ldlm_bl_work_item blwi = { .blwi_ns = NULL };
1415
1416                 init_completion(&blp->blp_comp);
1417
1418                 spin_lock(&blp->blp_lock);
1419                 list_add_tail(&blwi.blwi_entry, &blp->blp_list);
1420                 wake_up(&blp->blp_waitq);
1421                 spin_unlock(&blp->blp_lock);
1422
1423                 wait_for_completion(&blp->blp_comp);
1424         }
1425         OBD_FREE(blp, sizeof(*blp));
1426
1427         ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
1428         ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
1429         ldlm_proc_cleanup();
1430
1431         expired_lock_thread.elt_state = ELT_TERMINATE;
1432         wake_up(&expired_lock_thread.elt_waitq);
1433         wait_event(expired_lock_thread.elt_waitq,
1434                    expired_lock_thread.elt_state == ELT_STOPPED);
1435
1436 #endif
1437
1438         OBD_FREE(ldlm_state, sizeof(*ldlm_state));
1439         ldlm_state = NULL;
1440
1441         RETURN(0);
1442 }
1443
1444 int __init ldlm_init(void)
1445 {
1446         ldlm_resource_slab = kmem_cache_create("ldlm_resources",
1447                                                sizeof(struct ldlm_resource), 0,
1448                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1449         if (ldlm_resource_slab == NULL)
1450                 return -ENOMEM;
1451
1452         ldlm_lock_slab = kmem_cache_create("ldlm_locks",
1453                                            sizeof(struct ldlm_lock), 0,
1454                                            SLAB_HWCACHE_ALIGN, NULL, NULL);
1455         if (ldlm_lock_slab == NULL) {
1456                 kmem_cache_destroy(ldlm_resource_slab);
1457                 return -ENOMEM;
1458         }
1459
1460         l_lock_init(&ldlm_handle_lock);
1461
1462         return 0;
1463 }
1464
1465 void __exit ldlm_exit(void)
1466 {
1467         if ( ldlm_refcount )
1468                 CERROR("ldlm_refcount is %d in ldlm_exit!\n", ldlm_refcount);
1469         LASSERTF(kmem_cache_destroy(ldlm_resource_slab) == 0,
1470                  "couldn't free ldlm resource slab\n");
1471         LASSERTF(kmem_cache_destroy(ldlm_lock_slab) == 0,
1472                  "couldn't free ldlm lock slab\n");
1473 }
1474
1475 /* ldlm_flock.c */
1476 EXPORT_SYMBOL(ldlm_flock_completion_ast);
1477
1478 /* ldlm_extent.c */
1479 EXPORT_SYMBOL(ldlm_extent_shift_kms);
1480
1481 /* ldlm_lock.c */
1482 EXPORT_SYMBOL(ldlm_get_processing_policy);
1483 EXPORT_SYMBOL(ldlm_lock2desc);
1484 EXPORT_SYMBOL(ldlm_register_intent);
1485 EXPORT_SYMBOL(ldlm_lockname);
1486 EXPORT_SYMBOL(ldlm_typename);
1487 EXPORT_SYMBOL(ldlm_lock2handle);
1488 EXPORT_SYMBOL(__ldlm_handle2lock);
1489 EXPORT_SYMBOL(ldlm_lock_get);
1490 EXPORT_SYMBOL(ldlm_lock_put);
1491 EXPORT_SYMBOL(ldlm_lock_match);
1492 EXPORT_SYMBOL(ldlm_lock_cancel);
1493 EXPORT_SYMBOL(ldlm_lock_addref);
1494 EXPORT_SYMBOL(ldlm_lock_decref);
1495 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
1496 EXPORT_SYMBOL(ldlm_lock_change_resource);
1497 EXPORT_SYMBOL(ldlm_lock_set_data);
1498 EXPORT_SYMBOL(ldlm_it2str);
1499 EXPORT_SYMBOL(ldlm_lock_dump);
1500 EXPORT_SYMBOL(ldlm_lock_dump_handle);
1501 EXPORT_SYMBOL(ldlm_cancel_locks_for_export);
1502 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
1503 EXPORT_SYMBOL(ldlm_lock_allow_match);
1504
1505 /* ldlm_request.c */
1506 EXPORT_SYMBOL(ldlm_completion_ast);
1507 EXPORT_SYMBOL(ldlm_expired_completion_wait);
1508 EXPORT_SYMBOL(ldlm_cli_convert);
1509 EXPORT_SYMBOL(ldlm_cli_enqueue);
1510 EXPORT_SYMBOL(ldlm_cli_cancel);
1511 EXPORT_SYMBOL(ldlm_cli_cancel_unused);
1512 EXPORT_SYMBOL(ldlm_replay_locks);
1513 EXPORT_SYMBOL(ldlm_resource_foreach);
1514 EXPORT_SYMBOL(ldlm_namespace_foreach);
1515 EXPORT_SYMBOL(ldlm_namespace_foreach_res);
1516 EXPORT_SYMBOL(ldlm_change_cbdata);
1517
1518 /* ldlm_lockd.c */
1519 EXPORT_SYMBOL(ldlm_server_blocking_ast);
1520 EXPORT_SYMBOL(ldlm_server_completion_ast);
1521 EXPORT_SYMBOL(ldlm_server_glimpse_ast);
1522 EXPORT_SYMBOL(ldlm_handle_enqueue);
1523 EXPORT_SYMBOL(ldlm_handle_cancel);
1524 EXPORT_SYMBOL(ldlm_handle_convert);
1525 EXPORT_SYMBOL(ldlm_del_waiting_lock);
1526 EXPORT_SYMBOL(ldlm_get_ref);
1527 EXPORT_SYMBOL(ldlm_put_ref);
1528
1529 #if 0
1530 /* ldlm_test.c */
1531 EXPORT_SYMBOL(ldlm_test);
1532 EXPORT_SYMBOL(ldlm_regression_start);
1533 EXPORT_SYMBOL(ldlm_regression_stop);
1534 #endif
1535
1536 /* ldlm_resource.c */
1537 EXPORT_SYMBOL(ldlm_namespace_new);
1538 EXPORT_SYMBOL(ldlm_namespace_cleanup);
1539 EXPORT_SYMBOL(ldlm_namespace_free);
1540 EXPORT_SYMBOL(ldlm_namespace_dump);
1541 EXPORT_SYMBOL(ldlm_dump_all_namespaces);
1542 EXPORT_SYMBOL(ldlm_resource_get);
1543 EXPORT_SYMBOL(ldlm_resource_putref);
1544
1545 /* l_lock.c */
1546 EXPORT_SYMBOL(l_lock);
1547 EXPORT_SYMBOL(l_unlock);
1548
1549 /* ldlm_lib.c */
1550 EXPORT_SYMBOL(client_obd_setup);
1551 EXPORT_SYMBOL(client_obd_cleanup);
1552 EXPORT_SYMBOL(client_connect_import);
1553 EXPORT_SYMBOL(client_disconnect_export);
1554 EXPORT_SYMBOL(target_abort_recovery);
1555 EXPORT_SYMBOL(target_cleanup_recovery);
1556 EXPORT_SYMBOL(target_handle_connect);
1557 EXPORT_SYMBOL(target_destroy_export);
1558 EXPORT_SYMBOL(target_cancel_recovery_timer);
1559 EXPORT_SYMBOL(target_send_reply);
1560 EXPORT_SYMBOL(target_queue_recovery_request);
1561 EXPORT_SYMBOL(target_handle_ping);
1562 EXPORT_SYMBOL(target_handle_disconnect);
1563 EXPORT_SYMBOL(target_queue_final_reply);