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 <libcfs/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 /* LDLM state */
  51
  52 static struct ldlm_state *ldlm_state;
  53
  54 inline unsigned long round_timeout(unsigned long timeout)
  55 {
  56         return ((timeout / HZ) + 1) * HZ;
  57 }
  58
  59 #ifdef __KERNEL__
  60 /* XXX should this be per-ldlm? */
  61 static struct list_head waiting_locks_list;
  62 static spinlock_t waiting_locks_spinlock;
  63 static struct timer_list waiting_locks_timer;
  64
  65 static struct expired_lock_thread {
  66         wait_queue_head_t         elt_waitq;
  67         int                       elt_state;
  68         struct list_head          elt_expired_locks;
  69         spinlock_t                elt_lock;
  70 } expired_lock_thread;
  71 #endif
  72
  73 #if !defined(ENOTSUPP)
  74 #  define ENOTSUPP 524
  75 #endif
  76
  77 #define ELT_STOPPED   0
  78 #define ELT_READY     1
  79 #define ELT_TERMINATE 2
  80
  81 struct ldlm_bl_pool {
  82         spinlock_t              blp_lock;
  83         struct list_head        blp_list;
  84         wait_queue_head_t       blp_waitq;
  85         atomic_t                blp_num_threads;
  86         struct completion       blp_comp;
  87 };
  88
  89 struct ldlm_bl_work_item {
  90         struct list_head        blwi_entry;
  91         struct ldlm_namespace   *blwi_ns;
  92         struct ldlm_lock_desc   blwi_ld;
  93         struct ldlm_lock        *blwi_lock;
  94 };
  95
  96 #ifdef __KERNEL__
  97
  98 static inline int have_expired_locks(void)
  99 {
 100         int need_to_run;
 101
 102         spin_lock_bh(&expired_lock_thread.elt_lock);
 103         need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks);
 104         spin_unlock_bh(&expired_lock_thread.elt_lock);
 105
 106         RETURN(need_to_run);
 107 }
 108
 109 static int expired_lock_main(void *arg)
 110 {
 111         struct list_head *expired = &expired_lock_thread.elt_expired_locks;
 112         struct l_wait_info lwi = { 0 };
 113         unsigned long flags;
 114
 115         ENTRY;
 116         lock_kernel();
 117         kportal_daemonize("ldlm_elt");
 118
 119         SIGNAL_MASK_LOCK(current, flags);
 120         sigfillset(&current->blocked);
 121         RECALC_SIGPENDING;
 122         SIGNAL_MASK_UNLOCK(current, flags);
 123
 124         unlock_kernel();
 125
 126         expired_lock_thread.elt_state = ELT_READY;
 127         wake_up(&expired_lock_thread.elt_waitq);
 128
 129         while (1) {
 130                 l_wait_event(expired_lock_thread.elt_waitq,
 131                              have_expired_locks() ||
 132                              expired_lock_thread.elt_state == ELT_TERMINATE,
 133                              &lwi);
 134
 135                 spin_lock_bh(&expired_lock_thread.elt_lock);
 136                 while (!list_empty(expired)) {
 137                         struct obd_export *export;
 138                         struct ldlm_lock *lock;
 139
 140                         lock = list_entry(expired->next, struct ldlm_lock,
 141                                           l_pending_chain);
 142                         if ((void *)lock < LP_POISON + PAGE_SIZE &&
 143                             (void *)lock >= LP_POISON) {
 144                                 CERROR("free lock on elt list %p\n", lock);
 145                                 LBUG();
 146                         }
 147                         list_del_init(&lock->l_pending_chain);
 148                         if ((void *)lock->l_export < LP_POISON + PAGE_SIZE &&
 149                             (void *)lock->l_export >= LP_POISON + PAGE_SIZE) {
 150                                 CERROR("lock with free export on elt list %p\n",
 151                                        export);
 152                                 lock->l_export = NULL;
 153                                 LDLM_ERROR(lock, "free export\n");
 154                                 continue;
 155                         }
 156                         export = class_export_get(lock->l_export);
 157                         spin_unlock_bh(&expired_lock_thread.elt_lock);
 158
 159                         ptlrpc_fail_export(export);
 160                         class_export_put(export);
 161                         spin_lock_bh(&expired_lock_thread.elt_lock);
 162                 }
 163                 spin_unlock_bh(&expired_lock_thread.elt_lock);
 164
 165                 if (expired_lock_thread.elt_state == ELT_TERMINATE)
 166                         break;
 167         }
 168
 169         expired_lock_thread.elt_state = ELT_STOPPED;
 170         wake_up(&expired_lock_thread.elt_waitq);
 171         RETURN(0);
 172 }
 173
 174 static void waiting_locks_callback(unsigned long unused)
 175 {
 176         struct ldlm_lock *lock;
 177         char str[PTL_NALFMT_SIZE];
 178
 179         if (obd_dump_on_timeout)
 180                 portals_debug_dumplog();
 181
 182         spin_lock_bh(&waiting_locks_spinlock);
 183         while (!list_empty(&waiting_locks_list)) {
 184                 lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
 185                                   l_pending_chain);
 186
 187                 if ((lock->l_callback_timeout > jiffies) ||
 188                     (lock->l_req_mode == LCK_GROUP))
 189                         break;
 190
 191                 LDLM_ERROR(lock, "lock callback timer expired: evicting client "
 192                            "%s@%s nid %s ",
 193                            lock->l_export->exp_client_uuid.uuid,
 194                            lock->l_export->exp_connection->c_remote_uuid.uuid,
 195                            ptlrpc_peernid2str(&lock->l_export->exp_connection->c_peer, str));
 196
 197                 spin_lock_bh(&expired_lock_thread.elt_lock);
 198                 list_del(&lock->l_pending_chain);
 199                 list_add(&lock->l_pending_chain,
 200                          &expired_lock_thread.elt_expired_locks);
 201                 spin_unlock_bh(&expired_lock_thread.elt_lock);
 202                 wake_up(&expired_lock_thread.elt_waitq);
 203         }
 204
 205         /*
 206          * Make sure the timer will fire again if we have any locks
 207          * left.
 208          */
 209         if (!list_empty(&waiting_locks_list)) {
 210                 unsigned long timeout_rounded;
 211                 lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
 212                                   l_pending_chain);
 213                 timeout_rounded = round_timeout(lock->l_callback_timeout);
 214                 mod_timer(&waiting_locks_timer, timeout_rounded);
 215         }
 216         spin_unlock_bh(&waiting_locks_spinlock);
 217 }
 218
 219 /*
 220  * Indicate that we're waiting for a client to call us back cancelling a given
 221  * lock.  We add it to the pending-callback chain, and schedule the lock-timeout
 222  * timer to fire appropriately.  (We round up to the next second, to avoid
 223  * floods of timer firings during periods of high lock contention and traffic).
 224  */
 225 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
 226 {
 227         unsigned long timeout_rounded;
 228
 229         spin_lock_bh(&waiting_locks_spinlock);
 230         if (!list_empty(&lock->l_pending_chain)) {
 231                 LDLM_DEBUG(lock, "not re-adding to wait list");
 232                 spin_unlock_bh(&waiting_locks_spinlock);
 233                 return 0;
 234         }
 235         LDLM_DEBUG(lock, "adding to wait list");
 236
 237         lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2);
 238
 239         timeout_rounded = round_timeout(lock->l_callback_timeout);
 240
 241         if (timeout_rounded < waiting_locks_timer.expires ||
 242             !timer_pending(&waiting_locks_timer)) {
 243                 mod_timer(&waiting_locks_timer, timeout_rounded);
 244         }
 245         list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
 246         spin_unlock_bh(&waiting_locks_spinlock);
 247         return 1;
 248 }
 249
 250 /*
 251  * Remove a lock from the pending list, likely because it had its cancellation
 252  * callback arrive without incident.  This adjusts the lock-timeout timer if
 253  * needed.  Returns 0 if the lock wasn't pending after all, 1 if it was.
 254  */
 255 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 256 {
 257         struct list_head *list_next;
 258
 259         if (lock->l_export == NULL) {
 260                 /* We don't have a "waiting locks list" on clients. */
 261                 LDLM_DEBUG(lock, "client lock: no-op");
 262                 return 0;
 263         }
 264
 265         spin_lock_bh(&waiting_locks_spinlock);
 266
 267         if (list_empty(&lock->l_pending_chain)) {
 268                 spin_unlock_bh(&waiting_locks_spinlock);
 269                 LDLM_DEBUG(lock, "wasn't waiting");
 270                 return 0;
 271         }
 272
 273         list_next = lock->l_pending_chain.next;
 274         if (lock->l_pending_chain.prev == &waiting_locks_list) {
 275                 /* Removing the head of the list, adjust timer. */
 276                 if (list_next == &waiting_locks_list) {
 277                         /* No more, just cancel. */
 278                         del_timer(&waiting_locks_timer);
 279                 } else {
 280                         struct ldlm_lock *next;
 281                         next = list_entry(list_next, struct ldlm_lock,
 282                                           l_pending_chain);
 283                         mod_timer(&waiting_locks_timer,
 284                                   round_timeout(next->l_callback_timeout));
 285                 }
 286         }
 287
 288         spin_lock_bh(&expired_lock_thread.elt_lock);
 289         list_del_init(&lock->l_pending_chain);
 290         spin_unlock_bh(&expired_lock_thread.elt_lock);
 291
 292         spin_unlock_bh(&waiting_locks_spinlock);
 293         LDLM_DEBUG(lock, "removed");
 294         return 1;
 295 }
 296
 297 #else /* !__KERNEL__ */
 298
 299 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
 300 {
 301         RETURN(1);
 302 }
 303
 304 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 305 {
 306         RETURN(0);
 307 }
 308
 309 #endif /* __KERNEL__ */
 310
 311 static void ldlm_failed_ast(struct ldlm_lock *lock, int rc,const char *ast_type)
 312 {
 313         struct ptlrpc_connection *conn = lock->l_export->exp_connection;
 314         char str[PTL_NALFMT_SIZE];
 315
 316         LDLM_ERROR(lock, "%s AST failed (%d): evicting client %s@%s NID "LPX64
 317                    " (%s)", ast_type, rc, lock->l_export->exp_client_uuid.uuid,
 318                    conn->c_remote_uuid.uuid, conn->c_peer.peer_id.nid,
 319                    ptlrpc_peernid2str(&conn->c_peer, str));
 320
 321         if (obd_dump_on_timeout)
 322                 portals_debug_dumplog();
 323         ptlrpc_fail_export(lock->l_export);
 324 }
 325
 326 static int ldlm_handle_ast_error(struct ldlm_lock *lock,
 327                                  struct ptlrpc_request *req, int rc,
 328                                  const char *ast_type)
 329 {
 330         struct ptlrpc_peer *peer = &req->rq_import->imp_connection->c_peer;
 331         char str[PTL_NALFMT_SIZE];
 332
 333         if (rc == -ETIMEDOUT || rc == -EINTR || rc == -ENOTCONN) {
 334                 LASSERT(lock->l_export);
 335                 if (lock->l_export->exp_libclient) {
 336                         LDLM_DEBUG(lock, "%s AST to liblustre client (nid %s)"
 337                                    " timeout, just cancelling lock", ast_type,
 338                                    ptlrpc_peernid2str(peer, str));
 339                         ldlm_lock_cancel(lock);
 340                         rc = -ERESTART;
 341                 } else {
 342                         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 343                         ldlm_del_waiting_lock(lock);
 344                         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 345                         ldlm_failed_ast(lock, rc, ast_type);
 346                 }
 347         } else if (rc) {
 348                 if (rc == -EINVAL)
 349                         LDLM_DEBUG(lock, "client (nid %s) returned %d"
 350                                    " from %s AST - normal race",
 351                                    ptlrpc_peernid2str(peer, str),
 352                                    req->rq_repmsg->status, ast_type);
 353                 else
 354                         LDLM_ERROR(lock, "client (nid %s) returned %d "
 355                                    "from %s AST", ptlrpc_peernid2str(peer, str),
 356                                    (req->rq_repmsg != NULL) ?
 357                                    req->rq_repmsg->status : 0, ast_type);
 358                 ldlm_lock_cancel(lock);
 359                 /* Server-side AST functions are called from ldlm_reprocess_all,
 360                  * which needs to be told to please restart its reprocessing. */
 361                 rc = -ERESTART;
 362         }
 363
 364         return rc;
 365 }
 366
 367 int ldlm_server_blocking_ast(struct ldlm_lock *lock,
 368                              struct ldlm_lock_desc *desc,
 369                              void *data, int flag)
 370 {
 371         struct ldlm_request *body;
 372         struct ptlrpc_request *req;
 373         int rc = 0, size = sizeof(*body);
 374         ENTRY;
 375
 376         if (flag == LDLM_CB_CANCELING) {
 377                 /* Don't need to do anything here. */
 378                 RETURN(0);
 379         }
 380
 381         LASSERT(lock);
 382
 383         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 384         if (lock->l_granted_mode != lock->l_req_mode) {
 385                 /* this blocking AST will be communicated as part of the
 386                  * completion AST instead */
 387                 LDLM_DEBUG(lock, "lock not granted, not sending blocking AST");
 388                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 389                 RETURN(0);
 390         }
 391
 392         if (lock->l_destroyed) {
 393                 /* What's the point? */
 394                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 395                 RETURN(0);
 396         }
 397
 398 #if 0
 399         if (LTIME_S(CURRENT_TIME) - lock->l_export->exp_last_request_time > 30){
 400                 ldlm_failed_ast(lock, -ETIMEDOUT, "Not-attempted blocking");
 401                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 402                 RETURN(-ETIMEDOUT);
 403         }
 404 #endif
 405
 406         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 407                               LUSTRE_DLM_VERSION, LDLM_BL_CALLBACK,
 408                               1, &size, NULL);
 409         if (req == NULL) {
 410                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 411                 RETURN(-ENOMEM);
 412         }
 413
 414         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 415         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 416                sizeof(body->lock_handle1));
 417         memcpy(&body->lock_desc, desc, sizeof(*desc));
 418         body->lock_flags |= (lock->l_flags & LDLM_AST_FLAGS);
 419
 420         LDLM_DEBUG(lock, "server preparing blocking AST");
 421         req->rq_replen = lustre_msg_size(0, NULL);
 422
 423         if (lock->l_granted_mode == lock->l_req_mode)
 424                 ldlm_add_waiting_lock(lock);
 425         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 426
 427         req->rq_send_state = LUSTRE_IMP_FULL;
 428         req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */
 429         rc = ptlrpc_queue_wait(req);
 430         if (rc != 0)
 431                 rc = ldlm_handle_ast_error(lock, req, rc, "blocking");
 432
 433         ptlrpc_req_finished(req);
 434
 435         RETURN(rc);
 436 }
 437
 438 /* XXX copied from ptlrpc/service.c */
 439 static long timeval_sub(struct timeval *large, struct timeval *small)
 440 {
 441         return (large->tv_sec - small->tv_sec) * 1000000 +
 442                 (large->tv_usec - small->tv_usec);
 443 }
 444
 445 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data)
 446 {
 447         struct ldlm_request *body;
 448         struct ptlrpc_request *req;
 449         struct timeval granted_time;
 450         long total_enqueue_wait;
 451         int rc = 0, size[2] = {sizeof(*body)}, buffers = 1;
 452         ENTRY;
 453
 454         LASSERT(lock != NULL);
 455
 456         do_gettimeofday(&granted_time);
 457         total_enqueue_wait = timeval_sub(&granted_time, &lock->l_enqueued_time);
 458
 459         if (total_enqueue_wait / 1000000 > obd_timeout)
 460                 LDLM_ERROR(lock, "enqueue wait took %ldus", total_enqueue_wait);
 461
 462         down(&lock->l_resource->lr_lvb_sem);
 463         if (lock->l_resource->lr_lvb_len) {
 464                 buffers = 2;
 465                 size[1] = lock->l_resource->lr_lvb_len;
 466         }
 467         up(&lock->l_resource->lr_lvb_sem);
 468
 469         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 470                               LUSTRE_DLM_VERSION, LDLM_CP_CALLBACK,
 471                               buffers, size, NULL);
 472         if (req == NULL)
 473                 RETURN(-ENOMEM);
 474
 475         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
 476         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 477                sizeof(body->lock_handle1));
 478         body->lock_flags = flags;
 479         ldlm_lock2desc(lock, &body->lock_desc);
 480
 481         if (buffers == 2) {
 482                 void *lvb;
 483
 484                 down(&lock->l_resource->lr_lvb_sem);
 485                 lvb = lustre_msg_buf(req->rq_reqmsg, 1,
 486                                      lock->l_resource->lr_lvb_len);
 487
 488                 memcpy(lvb, lock->l_resource->lr_lvb_data,
 489                        lock->l_resource->lr_lvb_len);
 490                 up(&lock->l_resource->lr_lvb_sem);
 491         }
 492
 493         LDLM_DEBUG(lock, "server preparing completion AST (after %ldus wait)",
 494                    total_enqueue_wait);
 495         req->rq_replen = lustre_msg_size(0, NULL);
 496
 497         req->rq_send_state = LUSTRE_IMP_FULL;
 498         req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */
 499
 500         /* We only send real blocking ASTs after the lock is granted */
 501         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 502         if (lock->l_flags & LDLM_FL_AST_SENT) {
 503                 body->lock_flags |= LDLM_FL_AST_SENT;
 504                 ldlm_add_waiting_lock(lock); /* start the lock-timeout clock */
 505         }
 506         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 507
 508         rc = ptlrpc_queue_wait(req);
 509         if (rc != 0)
 510                 rc = ldlm_handle_ast_error(lock, req, rc, "completion");
 511
 512         ptlrpc_req_finished(req);
 513
 514         RETURN(rc);
 515 }
 516
 517 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data)
 518 {
 519         struct ldlm_resource *res = lock->l_resource;
 520         struct ldlm_request *body;
 521         struct ptlrpc_request *req;
 522         int rc = 0, size = sizeof(*body);
 523         ENTRY;
 524
 525         LASSERT(lock != NULL);
 526
 527         req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse,
 528                               LUSTRE_DLM_VERSION, LDLM_GL_CALLBACK,
 529                               1, &size, NULL);
 530         if (req == NULL)
 531                 RETURN(-ENOMEM);
 532
 533         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
 534         memcpy(&body->lock_handle1, &lock->l_remote_handle,
 535                sizeof(body->lock_handle1));
 536         ldlm_lock2desc(lock, &body->lock_desc);
 537
 538         down(&lock->l_resource->lr_lvb_sem);
 539         size = lock->l_resource->lr_lvb_len;
 540         up(&lock->l_resource->lr_lvb_sem);
 541         req->rq_replen = lustre_msg_size(1, &size);
 542
 543         req->rq_send_state = LUSTRE_IMP_FULL;
 544         req->rq_timeout = 2; /* 2 second timeout for initial AST reply */
 545
 546         rc = ptlrpc_queue_wait(req);
 547         if (rc == -ELDLM_NO_LOCK_DATA)
 548                 LDLM_DEBUG(lock, "lost race - client has a lock but no inode");
 549         else if (rc != 0)
 550                 rc = ldlm_handle_ast_error(lock, req, rc, "glimpse");
 551         else
 552                 rc = res->lr_namespace->ns_lvbo->lvbo_update
 553                         (res, req->rq_repmsg, 0, 1);
 554         ptlrpc_req_finished(req);
 555         RETURN(rc);
 556 }
 557
 558 static struct ldlm_lock *
 559 find_existing_lock(struct obd_export *exp, struct lustre_handle *remote_hdl)
 560 {
 561         struct obd_device *obd = exp->exp_obd;
 562         struct list_head *iter;
 563
 564         l_lock(&obd->obd_namespace->ns_lock);
 565         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
 566                 struct ldlm_lock *lock;
 567                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
 568                 if (lock->l_remote_handle.cookie == remote_hdl->cookie) {
 569                         LDLM_LOCK_GET(lock);
 570                         l_unlock(&obd->obd_namespace->ns_lock);
 571                         return lock;
 572                 }
 573         }
 574         l_unlock(&obd->obd_namespace->ns_lock);
 575         return NULL;
 576 }
 577
 578
 579 int ldlm_handle_enqueue(struct ptlrpc_request *req,
 580                         ldlm_completion_callback completion_callback,
 581                         ldlm_blocking_callback blocking_callback,
 582                         ldlm_glimpse_callback glimpse_callback)
 583 {
 584         struct obd_device *obddev = req->rq_export->exp_obd;
 585         struct ldlm_reply *dlm_rep;
 586         struct ldlm_request *dlm_req;
 587         int rc = 0, size[2] = {sizeof(*dlm_rep)};
 588         __u32 flags;
 589         ldlm_error_t err = ELDLM_OK;
 590         struct ldlm_lock *lock = NULL;
 591         void *cookie = NULL;
 592         ENTRY;
 593
 594         LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
 595
 596         dlm_req = lustre_swab_reqbuf (req, MDS_REQ_INTENT_LOCKREQ_OFF,
 597                                       sizeof (*dlm_req),
 598                                       lustre_swab_ldlm_request);
 599         if (dlm_req == NULL) {
 600                 CERROR ("Can't unpack dlm_req\n");
 601                 GOTO(out, rc = -EFAULT);
 602         }
 603
 604         flags = dlm_req->lock_flags;
 605
 606         LASSERT(req->rq_export);
 607
 608         if (flags & LDLM_FL_REPLAY) {
 609                 lock = find_existing_lock(req->rq_export,
 610                                           &dlm_req->lock_handle1);
 611                 if (lock != NULL) {
 612                         DEBUG_REQ(D_HA, req, "found existing lock cookie "LPX64,
 613                                   lock->l_handle.h_cookie);
 614                         GOTO(existing_lock, rc = 0);
 615                 }
 616         }
 617
 618         /* The lock's callback data might be set in the policy function */
 619         lock = ldlm_lock_create(obddev->obd_namespace, &dlm_req->lock_handle2,
 620                                 dlm_req->lock_desc.l_resource.lr_name,
 621                                 dlm_req->lock_desc.l_resource.lr_type,
 622                                 dlm_req->lock_desc.l_req_mode,
 623                                 blocking_callback, completion_callback,
 624                                 glimpse_callback, NULL, 0);
 625         if (!lock)
 626                 GOTO(out, rc = -ENOMEM);
 627
 628         do_gettimeofday(&lock->l_enqueued_time);
 629         memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
 630                sizeof(lock->l_remote_handle));
 631         LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
 632
 633         LASSERT(req->rq_export);
 634         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_BLOCKED, obd_timeout * 2);
 635         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 636         if (req->rq_export->exp_failed) {
 637                 LDLM_ERROR(lock,"lock on destroyed export %p\n",req->rq_export);
 638                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 639                 GOTO(out, err = -ENOTCONN);
 640         }
 641         lock->l_export = class_export_get(req->rq_export);
 642
 643         list_add(&lock->l_export_chain,
 644                  &lock->l_export->exp_ldlm_data.led_held_locks);
 645         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 646
 647 existing_lock:
 648
 649         if (flags & LDLM_FL_HAS_INTENT) {
 650                 /* In this case, the reply buffer is allocated deep in
 651                  * local_lock_enqueue by the policy function. */
 652                 cookie = req;
 653         } else {
 654                 int buffers = 1;
 655                 down(&lock->l_resource->lr_lvb_sem);
 656                 if (lock->l_resource->lr_lvb_len) {
 657                         size[1] = lock->l_resource->lr_lvb_len;
 658                         buffers = 2;
 659                 }
 660                 up(&lock->l_resource->lr_lvb_sem);
 661                 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR))
 662                         GOTO(out, rc = -ENOMEM);
 663
 664                 rc = lustre_pack_reply(req, buffers, size, NULL);
 665                 if (rc)
 666                         GOTO(out, rc);
 667         }
 668
 669         if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN)
 670                 memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data,
 671                        sizeof(ldlm_policy_data_t));
 672         if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT)
 673                 memcpy(&lock->l_req_extent, &lock->l_policy_data.l_extent,
 674                        sizeof(lock->l_req_extent));
 675
 676         err = ldlm_lock_enqueue(obddev->obd_namespace, &lock, cookie, &flags);
 677         if (err)
 678                 GOTO(out, err);
 679
 680         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
 681         dlm_rep->lock_flags = flags;
 682
 683         ldlm_lock2desc(lock, &dlm_rep->lock_desc);
 684         ldlm_lock2handle(lock, &dlm_rep->lock_handle);
 685
 686         /* We never send a blocking AST until the lock is granted, but
 687          * we can tell it right now */
 688         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 689         if (lock->l_flags & LDLM_FL_AST_SENT) {
 690                 dlm_rep->lock_flags |= LDLM_FL_AST_SENT;
 691                 if (lock->l_granted_mode == lock->l_req_mode)
 692                         ldlm_add_waiting_lock(lock);
 693         }
 694         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 695
 696         EXIT;
 697  out:
 698         req->rq_status = err;
 699         if (req->rq_reply_state == NULL) {
 700                 err = lustre_pack_reply(req, 0, NULL, NULL);
 701                 if (rc == 0)
 702                         rc = err;
 703                 req->rq_status = rc;
 704         }
 705
 706         /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
 707          * ldlm_reprocess_all.  If this moves, revisit that code. -phil */
 708         if (lock) {
 709                 l_lock(&lock->l_resource->lr_namespace->ns_lock);
 710                 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
 711                            "(err=%d, rc=%d)", err, rc);
 712                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 713
 714                 if (rc == 0) {
 715                         down(&lock->l_resource->lr_lvb_sem);
 716                         size[1] = lock->l_resource->lr_lvb_len;
 717                         if (size[1] > 0) {
 718                                 void *lvb = lustre_msg_buf(req->rq_repmsg,
 719                                                            1, size[1]);
 720                                 LASSERTF(lvb != NULL, "req %p, lock %p\n",
 721                                          req, lock);
 722
 723                                 memcpy(lvb, lock->l_resource->lr_lvb_data,
 724                                        size[1]);
 725                         }
 726                         up(&lock->l_resource->lr_lvb_sem);
 727                 } else {
 728                         ldlm_lock_destroy(lock);
 729                 }
 730
 731                 if (!err && dlm_req->lock_desc.l_resource.lr_type != LDLM_FLOCK)
 732                         ldlm_reprocess_all(lock->l_resource);
 733                 LDLM_LOCK_PUT(lock);
 734         }
 735         LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p, rc %d)",
 736                           lock, rc);
 737
 738         return rc;
 739 }
 740
 741 int ldlm_handle_convert(struct ptlrpc_request *req)
 742 {
 743         struct ldlm_request *dlm_req;
 744         struct ldlm_reply *dlm_rep;
 745         struct ldlm_lock *lock;
 746         int rc, size = sizeof(*dlm_rep);
 747         ENTRY;
 748
 749         dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
 750                                      lustre_swab_ldlm_request);
 751         if (dlm_req == NULL) {
 752                 CERROR ("Can't unpack dlm_req\n");
 753                 RETURN (-EFAULT);
 754         }
 755
 756         rc = lustre_pack_reply(req, 1, &size, NULL);
 757         if (rc) {
 758                 CERROR("out of memory\n");
 759                 RETURN(-ENOMEM);
 760         }
 761         dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
 762         dlm_rep->lock_flags = dlm_req->lock_flags;
 763
 764         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 765         if (!lock) {
 766                 req->rq_status = EINVAL;
 767         } else {
 768                 void *res = NULL;
 769
 770                 l_lock(&lock->l_resource->lr_namespace->ns_lock);
 771                 LDLM_DEBUG(lock, "server-side convert handler START");
 772                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 773
 774                 res = ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode,
 775                                         &dlm_rep->lock_flags);
 776                 if (res) {
 777                         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 778                         if (ldlm_del_waiting_lock(lock))
 779                                 CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock);
 780                         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 781                         req->rq_status = 0;
 782                 } else {
 783                         req->rq_status = EDEADLOCK;
 784                 }
 785         }
 786
 787         if (lock) {
 788                 if (!req->rq_status)
 789                         ldlm_reprocess_all(lock->l_resource);
 790                 l_lock(&lock->l_resource->lr_namespace->ns_lock);
 791                 LDLM_DEBUG(lock, "server-side convert handler END");
 792                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 793                 LDLM_LOCK_PUT(lock);
 794         } else
 795                 LDLM_DEBUG_NOLOCK("server-side convert handler END");
 796
 797         RETURN(0);
 798 }
 799
 800 int ldlm_handle_cancel(struct ptlrpc_request *req)
 801 {
 802         struct ldlm_request *dlm_req;
 803         struct ldlm_lock *lock;
 804         struct ldlm_resource *res;
 805         int rc;
 806         ENTRY;
 807
 808         dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
 809                                       lustre_swab_ldlm_request);
 810         if (dlm_req == NULL) {
 811                 CERROR("bad request buffer for cancel\n");
 812                 RETURN(-EFAULT);
 813         }
 814
 815         rc = lustre_pack_reply(req, 0, NULL, NULL);
 816         if (rc) {
 817                 CERROR("out of memory\n");
 818                 RETURN(-ENOMEM);
 819         }
 820
 821         lock = ldlm_handle2lock(&dlm_req->lock_handle1);
 822         if (!lock) {
 823                 CERROR("received cancel for unknown lock cookie "LPX64
 824                        " from client %s id %s\n",
 825                        dlm_req->lock_handle1.cookie,
 826                        req->rq_export->exp_client_uuid.uuid,
 827                        req->rq_peerstr);
 828                 LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock "
 829                                   "(cookie "LPU64")",
 830                                   dlm_req->lock_handle1.cookie);
 831                 req->rq_status = ESTALE;
 832         } else {
 833                 LDLM_DEBUG(lock, "server-side cancel handler START");
 834                 res = lock->l_resource;
 835                 if (res && res->lr_namespace->ns_lvbo &&
 836                     res->lr_namespace->ns_lvbo->lvbo_update) {
 837                         (void)res->lr_namespace->ns_lvbo->lvbo_update
 838                                 (res, NULL, 0, 0);
 839                                 //(res, req->rq_reqmsg, 1);
 840                 }
 841
 842                 l_lock(&res->lr_namespace->ns_lock);
 843                 ldlm_lock_cancel(lock);
 844                 if (ldlm_del_waiting_lock(lock))
 845                         CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock);
 846                 l_unlock(&res->lr_namespace->ns_lock);
 847                 req->rq_status = rc;
 848         }
 849
 850         if (ptlrpc_reply(req) != 0)
 851                 LBUG();
 852
 853         if (lock) {
 854                 ldlm_reprocess_all(lock->l_resource);
 855                 l_lock(&lock->l_resource->lr_namespace->ns_lock);
 856                 LDLM_DEBUG(lock, "server-side cancel handler END");
 857                 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 858                 LDLM_LOCK_PUT(lock);
 859         }
 860
 861         RETURN(0);
 862 }
 863
 864 void ldlm_handle_bl_callback(struct ldlm_namespace *ns,
 865                              struct ldlm_lock_desc *ld, struct ldlm_lock *lock)
 866 {
 867         int do_ast;
 868         ENTRY;
 869
 870         l_lock(&ns->ns_lock);
 871         LDLM_DEBUG(lock, "client blocking AST callback handler START");
 872
 873         lock->l_flags |= LDLM_FL_CBPENDING;
 874         do_ast = (!lock->l_readers && !lock->l_writers);
 875
 876         if (do_ast) {
 877                 LDLM_DEBUG(lock, "already unused, calling "
 878                            "callback (%p)", lock->l_blocking_ast);
 879                 if (lock->l_blocking_ast != NULL) {
 880                         l_unlock(&ns->ns_lock);
 881                         l_check_no_ns_lock(ns);
 882                         lock->l_blocking_ast(lock, ld, lock->l_ast_data,
 883                                              LDLM_CB_BLOCKING);
 884                         l_lock(&ns->ns_lock);
 885                 }
 886         } else {
 887                 LDLM_DEBUG(lock, "Lock still has references, will be"
 888                            " cancelled later");
 889         }
 890
 891         LDLM_DEBUG(lock, "client blocking callback handler END");
 892         l_unlock(&ns->ns_lock);
 893         LDLM_LOCK_PUT(lock);
 894         EXIT;
 895 }
 896
 897 static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
 898                                     struct ldlm_namespace *ns,
 899                                     struct ldlm_request *dlm_req,
 900                                     struct ldlm_lock *lock)
 901 {
 902         LIST_HEAD(ast_list);
 903         ENTRY;
 904
 905         l_lock(&ns->ns_lock);
 906         LDLM_DEBUG(lock, "client completion callback handler START");
 907
 908         /* If we receive the completion AST before the actual enqueue returned,
 909          * then we might need to switch lock modes, resources, or extents. */
 910         if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
 911                 lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
 912                 LDLM_DEBUG(lock, "completion AST, new lock mode");
 913         }
 914
 915         if (lock->l_resource->lr_type != LDLM_PLAIN) {
 916                 memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data,
 917                        sizeof(lock->l_policy_data));
 918                 LDLM_DEBUG(lock, "completion AST, new policy data");
 919         }
 920
 921         ldlm_resource_unlink_lock(lock);
 922         if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
 923                    &lock->l_resource->lr_name,
 924                    sizeof(lock->l_resource->lr_name)) != 0) {
 925                 ldlm_lock_change_resource(ns, lock,
 926                                          dlm_req->lock_desc.l_resource.lr_name);
 927                 LDLM_DEBUG(lock, "completion AST, new resource");
 928         }
 929
 930         if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
 931                 lock->l_flags |= LDLM_FL_CBPENDING;
 932                 LDLM_DEBUG(lock, "completion AST includes blocking AST");
 933         }
 934
 935         if (lock->l_lvb_len) {
 936                 void *lvb;
 937                 lvb = lustre_swab_reqbuf(req, 1, lock->l_lvb_len,
 938                                          lock->l_lvb_swabber);
 939                 if (lvb == NULL) {
 940                         LDLM_ERROR(lock, "completion AST did not contain "
 941                                    "expected LVB!");
 942                 } else {
 943                         memcpy(lock->l_lvb_data, lvb, lock->l_lvb_len);
 944                 }
 945         }
 946
 947         lock->l_resource->lr_tmp = &ast_list;
 948         ldlm_grant_lock(lock, req, sizeof(*req), 1);
 949         lock->l_resource->lr_tmp = NULL;
 950         LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
 951         l_unlock(&ns->ns_lock);
 952         LDLM_LOCK_PUT(lock);
 953
 954         ldlm_run_ast_work(ns, &ast_list);
 955
 956         LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
 957                           lock);
 958         EXIT;
 959 }
 960
 961 static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
 962                                     struct ldlm_namespace *ns,
 963                                     struct ldlm_request *dlm_req,
 964                                     struct ldlm_lock *lock)
 965 {
 966         int rc = -ENOSYS;
 967         ENTRY;
 968
 969         l_lock(&ns->ns_lock);
 970         LDLM_DEBUG(lock, "client glimpse AST callback handler");
 971
 972         if (lock->l_glimpse_ast != NULL) {
 973                 l_unlock(&ns->ns_lock);
 974                 l_check_no_ns_lock(ns);
 975                 rc = lock->l_glimpse_ast(lock, req);
 976                 l_lock(&ns->ns_lock);
 977         }
 978
 979         if (req->rq_repmsg != NULL) {
 980                 ptlrpc_reply(req);
 981         } else {
 982                 req->rq_status = rc;
 983                 ptlrpc_error(req);
 984         }
 985
 986         l_unlock(&ns->ns_lock);
 987         if (lock->l_granted_mode == LCK_PW &&
 988             !lock->l_readers && !lock->l_writers &&
 989             time_after(jiffies, lock->l_last_used + 10 * HZ)) {
 990                 if (ldlm_bl_to_thread(ns, NULL, lock))
 991                         ldlm_handle_bl_callback(ns, NULL, lock);
 992
 993                 EXIT;
 994                 return;
 995         }
 996         LDLM_LOCK_PUT(lock);
 997         EXIT;
 998 }
 999
1000 static int ldlm_callback_reply(struct ptlrpc_request *req, int rc)
1001 {
1002         req->rq_status = rc;
1003         if (req->rq_reply_state == NULL) {
1004                 rc = lustre_pack_reply(req, 0, NULL, NULL);
1005                 if (rc)
1006                         return rc;
1007         }
1008         return ptlrpc_reply(req);
1009 }
1010
1011 int ldlm_bl_to_thread(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
1012                       struct ldlm_lock *lock)
1013 {
1014 #ifdef __KERNEL__
1015         struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
1016         struct ldlm_bl_work_item *blwi;
1017         ENTRY;
1018
1019         OBD_ALLOC(blwi, sizeof(*blwi));
1020         if (blwi == NULL)
1021                 RETURN(-ENOMEM);
1022
1023         blwi->blwi_ns = ns;
1024         if (ld != NULL)
1025                 blwi->blwi_ld = *ld;
1026         blwi->blwi_lock = lock;
1027
1028         spin_lock(&blp->blp_lock);
1029         list_add_tail(&blwi->blwi_entry, &blp->blp_list);
1030         wake_up(&blp->blp_waitq);
1031         spin_unlock(&blp->blp_lock);
1032
1033         RETURN(0);
1034 #else
1035         RETURN(-ENOSYS);
1036 #endif
1037
1038 }
1039
1040 static int ldlm_msg_check_version(struct lustre_msg *msg)
1041 {
1042         int rc;
1043
1044         switch (msg->opc) {
1045         case LDLM_ENQUEUE:
1046         case LDLM_CONVERT:
1047         case LDLM_CANCEL:
1048         case LDLM_BL_CALLBACK:
1049         case LDLM_CP_CALLBACK:
1050         case LDLM_GL_CALLBACK:
1051                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1052                 if (rc)
1053                         CERROR("bad opc %u version %08x, expecting %08x\n",
1054                                msg->opc, msg->version, LUSTRE_DLM_VERSION);
1055                 break;
1056         case OBD_LOG_CANCEL:
1057         case LLOG_ORIGIN_HANDLE_OPEN:
1058         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
1059         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
1060         case LLOG_ORIGIN_HANDLE_READ_HEADER:
1061         case LLOG_ORIGIN_HANDLE_CLOSE:
1062         case LLOG_CATINFO:
1063                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1064                 if (rc)
1065                         CERROR("bad opc %u version %08x, expecting %08x\n",
1066                                msg->opc, msg->version, LUSTRE_LOG_VERSION);
1067                 break;
1068         default:
1069                 CERROR("LDLM unknown opcode %d\n", msg->opc);
1070                 rc = -ENOTSUPP;
1071                 break;
1072         }
1073
1074         return rc;
1075 }
1076
1077 static int ldlm_callback_handler(struct ptlrpc_request *req)
1078 {
1079         struct ldlm_namespace *ns;
1080         struct ldlm_request *dlm_req;
1081         struct ldlm_lock *lock;
1082         int rc;
1083         ENTRY;
1084
1085         rc = ldlm_msg_check_version(req->rq_reqmsg);
1086         if (rc) {
1087                 CERROR("LDLM_CB drop mal-formed request\n");
1088                 RETURN(rc);
1089         }
1090
1091         /* Requests arrive in sender's byte order.  The ptlrpc service
1092          * handler has already checked and, if necessary, byte-swapped the
1093          * incoming request message body, but I am responsible for the
1094          * message buffers. */
1095
1096         if (req->rq_export == NULL) {
1097                 struct ldlm_request *dlm_req;
1098
1099                 CDEBUG(D_RPCTRACE, "operation %d from %s with bad "
1100                        "export cookie "LPX64"; this is "
1101                        "normal if this node rebooted with a lock held\n",
1102                        req->rq_reqmsg->opc,
1103                        req->rq_peerstr,
1104                        req->rq_reqmsg->handle.cookie);
1105                 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
1106                                              lustre_swab_ldlm_request);
1107                 if (dlm_req != NULL)
1108                         CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n",
1109                                dlm_req->lock_handle1.cookie);
1110
1111                 ldlm_callback_reply(req, -ENOTCONN);
1112                 RETURN(0);
1113         }
1114
1115         LASSERT(req->rq_export != NULL);
1116         LASSERT(req->rq_export->exp_obd != NULL);
1117
1118         switch(req->rq_reqmsg->opc) {
1119         case LDLM_BL_CALLBACK:
1120                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
1121                 break;
1122         case LDLM_CP_CALLBACK:
1123                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0);
1124                 break;
1125         case LDLM_GL_CALLBACK:
1126                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_GL_CALLBACK, 0);
1127                 break;
1128         case OBD_LOG_CANCEL:
1129                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1130                 rc = llog_origin_handle_cancel(req);
1131                 ldlm_callback_reply(req, rc);
1132                 RETURN(0);
1133         case LLOG_ORIGIN_HANDLE_OPEN:
1134                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1135                 rc = llog_origin_handle_open(req);
1136                 ldlm_callback_reply(req, rc);
1137                 RETURN(0);
1138         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
1139                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1140                 rc = llog_origin_handle_next_block(req);
1141                 ldlm_callback_reply(req, rc);
1142                 RETURN(0);
1143         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
1144                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1145                 rc = llog_origin_handle_prev_block(req);
1146                 ldlm_callback_reply(req, rc);
1147                 RETURN(0);
1148         case LLOG_ORIGIN_HANDLE_READ_HEADER:
1149                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1150                 rc = llog_origin_handle_read_header(req);
1151                 ldlm_callback_reply(req, rc);
1152                 RETURN(0);
1153         case LLOG_ORIGIN_HANDLE_CLOSE:
1154                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0);
1155                 rc = llog_origin_handle_close(req);
1156                 ldlm_callback_reply(req, rc);
1157                 RETURN(0);
1158         default:
1159                 CERROR("unknown opcode %u\n", req->rq_reqmsg->opc);
1160                 ldlm_callback_reply(req, -EPROTO);
1161                 RETURN(0);
1162         }
1163
1164         ns = req->rq_export->exp_obd->obd_namespace;
1165         LASSERT(ns != NULL);
1166
1167         dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
1168                                       lustre_swab_ldlm_request);
1169         if (dlm_req == NULL) {
1170                 CERROR ("can't unpack dlm_req\n");
1171                 ldlm_callback_reply (req, -EPROTO);
1172                 RETURN (0);
1173         }
1174
1175         lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle1);
1176         if (!lock) {
1177                 LDLM_DEBUG_NOLOCK("callback on lock "LPX64" - lock "
1178                                   "disappeared\n",dlm_req->lock_handle1.cookie);
1179                 ldlm_callback_reply(req, -EINVAL);
1180                 RETURN(0);
1181         }
1182
1183         /* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */
1184         lock->l_flags |= (dlm_req->lock_flags & LDLM_AST_FLAGS);
1185
1186         /* We want the ost thread to get this reply so that it can respond
1187          * to ost requests (write cache writeback) that might be triggered
1188          * in the callback.
1189          *
1190          * But we'd also like to be able to indicate in the reply that we're
1191          * cancelling right now, because it's unused, or have an intent result
1192          * in the reply, so we might have to push the responsibility for sending
1193          * the reply down into the AST handlers, alas. */
1194
1195         switch (req->rq_reqmsg->opc) {
1196         case LDLM_BL_CALLBACK:
1197                 CDEBUG(D_INODE, "blocking ast\n");
1198                 ldlm_callback_reply(req, 0);
1199                 if (ldlm_bl_to_thread(ns, &dlm_req->lock_desc, lock))
1200                         ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock);
1201
1202                 break;
1203         case LDLM_CP_CALLBACK:
1204                 CDEBUG(D_INODE, "completion ast\n");
1205                 ldlm_callback_reply(req, 0);
1206                 ldlm_handle_cp_callback(req, ns, dlm_req, lock);
1207                 break;
1208         case LDLM_GL_CALLBACK:
1209                 CDEBUG(D_INODE, "glimpse ast\n");
1210                 ldlm_handle_gl_callback(req, ns, dlm_req, lock);
1211                 break;
1212         default:
1213                 LBUG();                         /* checked above */
1214         }
1215
1216         RETURN(0);
1217 }
1218
1219 static int ldlm_cancel_handler(struct ptlrpc_request *req)
1220 {
1221         int rc = 0;
1222         ENTRY;
1223
1224         rc = ldlm_msg_check_version(req->rq_reqmsg);
1225         if (rc) {
1226                 CERROR("LDLM_CL drop mal-formed request\n");
1227                 RETURN(rc);
1228         }
1229
1230         /* Requests arrive in sender's byte order.  The ptlrpc service
1231          * handler has already checked and, if necessary, byte-swapped the
1232          * incoming request message body, but I am responsible for the
1233          * message buffers. */
1234
1235         if (req->rq_export == NULL) {
1236                 struct ldlm_request *dlm_req;
1237                 CERROR("operation %d with bad export from %s\n",
1238                        req->rq_reqmsg->opc,
1239                        req->rq_peerstr);
1240                 CERROR("--> export cookie: "LPX64"\n",
1241                        req->rq_reqmsg->handle.cookie);
1242                 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
1243                                              lustre_swab_ldlm_request);
1244                 if (dlm_req != NULL)
1245                         ldlm_lock_dump_handle(D_ERROR, &dlm_req->lock_handle1);
1246                 ldlm_callback_reply(req, -ENOTCONN);
1247                 RETURN(0);
1248         }
1249
1250         switch (req->rq_reqmsg->opc) {
1251
1252         /* XXX FIXME move this back to mds/handler.c, bug 249 */
1253         case LDLM_CANCEL:
1254                 CDEBUG(D_INODE, "cancel\n");
1255                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1256                 rc = ldlm_handle_cancel(req);
1257                 break;
1258         default:
1259                 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
1260                 ldlm_callback_reply(req, -EINVAL);
1261         }
1262
1263         RETURN(rc);
1264 }
1265
1266 #ifdef __KERNEL__
1267 static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp)
1268 {
1269         struct ldlm_bl_work_item *blwi = NULL;
1270
1271         spin_lock(&blp->blp_lock);
1272         if (!list_empty(&blp->blp_list)) {
1273                 blwi = list_entry(blp->blp_list.next, struct ldlm_bl_work_item,
1274                                   blwi_entry);
1275                 list_del(&blwi->blwi_entry);
1276         }
1277         spin_unlock(&blp->blp_lock);
1278
1279         return blwi;
1280 }
1281
1282 struct ldlm_bl_thread_data {
1283         int                     bltd_num;
1284         struct ldlm_bl_pool     *bltd_blp;
1285 };
1286
1287 static int ldlm_bl_thread_main(void *arg)
1288 {
1289         struct ldlm_bl_thread_data *bltd = arg;
1290         struct ldlm_bl_pool *blp = bltd->bltd_blp;
1291         unsigned long flags;
1292         ENTRY;
1293
1294         /* XXX boiler-plate */
1295         {
1296                 char name[sizeof(current->comm)];
1297                 snprintf(name, sizeof(name) - 1, "ldlm_bl_%02d",
1298                          bltd->bltd_num);
1299                 kportal_daemonize(name);
1300         }
1301         SIGNAL_MASK_LOCK(current, flags);
1302         sigfillset(&current->blocked);
1303         RECALC_SIGPENDING;
1304         SIGNAL_MASK_UNLOCK(current, flags);
1305
1306         atomic_inc(&blp->blp_num_threads);
1307         complete(&blp->blp_comp);
1308
1309         while(1) {
1310                 struct l_wait_info lwi = { 0 };
1311                 struct ldlm_bl_work_item *blwi = NULL;
1312
1313                 l_wait_event_exclusive(blp->blp_waitq,
1314                                        (blwi = ldlm_bl_get_work(blp)) != NULL,
1315                                        &lwi);
1316
1317                 if (blwi->blwi_ns == NULL)
1318                         break;
1319
1320                 ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld,
1321                                         blwi->blwi_lock);
1322                 OBD_FREE(blwi, sizeof(*blwi));
1323         }
1324
1325         atomic_dec(&blp->blp_num_threads);
1326         complete(&blp->blp_comp);
1327         RETURN(0);
1328 }
1329
1330 #endif
1331
1332 static int ldlm_setup(void);
1333 static int ldlm_cleanup(int force);
1334
1335 int ldlm_get_ref(void)
1336 {
1337         int rc = 0;
1338         down(&ldlm_ref_sem);
1339         if (++ldlm_refcount == 1) {
1340                 rc = ldlm_setup();
1341                 if (rc)
1342                         ldlm_refcount--;
1343         }
1344         up(&ldlm_ref_sem);
1345
1346         RETURN(rc);
1347 }
1348
1349 void ldlm_put_ref(int force)
1350 {
1351         down(&ldlm_ref_sem);
1352         if (ldlm_refcount == 1) {
1353                 int rc = ldlm_cleanup(force);
1354                 if (rc)
1355                         CERROR("ldlm_cleanup failed: %d\n", rc);
1356                 else
1357                         ldlm_refcount--;
1358         } else {
1359                 ldlm_refcount--;
1360         }
1361         up(&ldlm_ref_sem);
1362
1363         EXIT;
1364 }
1365
1366 static int ldlm_setup(void)
1367 {
1368         struct ldlm_bl_pool *blp;
1369         int rc = 0;
1370 #ifdef __KERNEL__
1371         int i;
1372 #endif
1373         ENTRY;
1374
1375         if (ldlm_state != NULL)
1376                 RETURN(-EALREADY);
1377
1378         OBD_ALLOC(ldlm_state, sizeof(*ldlm_state));
1379         if (ldlm_state == NULL)
1380                 RETURN(-ENOMEM);
1381
1382 #ifdef __KERNEL__
1383         rc = ldlm_proc_setup();
1384         if (rc != 0)
1385                 GOTO(out_free, rc);
1386 #endif
1387
1388         ldlm_state->ldlm_cb_service =
1389                 ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE,
1390                                 LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1391                                 1500, ldlm_callback_handler, "ldlm_cbd",
1392                                 ldlm_svc_proc_dir);
1393
1394         if (!ldlm_state->ldlm_cb_service) {
1395                 CERROR("failed to start service\n");
1396                 GOTO(out_proc, rc = -ENOMEM);
1397         }
1398
1399         ldlm_state->ldlm_cancel_service =
1400                 ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE,
1401                                 LDLM_CANCEL_REQUEST_PORTAL,
1402                                 LDLM_CANCEL_REPLY_PORTAL, 30000,
1403                                 ldlm_cancel_handler, "ldlm_canceld",
1404                                 ldlm_svc_proc_dir);
1405
1406         if (!ldlm_state->ldlm_cancel_service) {
1407                 CERROR("failed to start service\n");
1408                 GOTO(out_proc, rc = -ENOMEM);
1409         }
1410
1411         OBD_ALLOC(blp, sizeof(*blp));
1412         if (blp == NULL)
1413                 GOTO(out_proc, rc = -ENOMEM);
1414         ldlm_state->ldlm_bl_pool = blp;
1415
1416         atomic_set(&blp->blp_num_threads, 0);
1417         init_waitqueue_head(&blp->blp_waitq);
1418         spin_lock_init(&blp->blp_lock);
1419
1420         INIT_LIST_HEAD(&blp->blp_list);
1421
1422 #ifdef __KERNEL__
1423         for (i = 0; i < LDLM_NUM_THREADS; i++) {
1424                 struct ldlm_bl_thread_data bltd = {
1425                         .bltd_num = i,
1426                         .bltd_blp = blp,
1427                 };
1428                 init_completion(&blp->blp_comp);
1429                 rc = kernel_thread(ldlm_bl_thread_main, &bltd, 0);
1430                 if (rc < 0) {
1431                         CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
1432                         GOTO(out_thread, rc);
1433                 }
1434                 wait_for_completion(&blp->blp_comp);
1435         }
1436
1437         rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cancel_service,
1438                                     LDLM_NUM_THREADS, "ldlm_cn");
1439         if (rc)
1440                 GOTO(out_thread, rc);
1441
1442         rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cb_service,
1443                                     LDLM_NUM_THREADS, "ldlm_cb");
1444         if (rc)
1445                 GOTO(out_thread, rc);
1446
1447         INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
1448         spin_lock_init(&expired_lock_thread.elt_lock);
1449         expired_lock_thread.elt_state = ELT_STOPPED;
1450         init_waitqueue_head(&expired_lock_thread.elt_waitq);
1451
1452         rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS);
1453         if (rc < 0) {
1454                 CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
1455                 GOTO(out_thread, rc);
1456         }
1457
1458         wait_event(expired_lock_thread.elt_waitq,
1459                    expired_lock_thread.elt_state == ELT_READY);
1460
1461         INIT_LIST_HEAD(&waiting_locks_list);
1462         spin_lock_init(&waiting_locks_spinlock);
1463         waiting_locks_timer.function = waiting_locks_callback;
1464         waiting_locks_timer.data = 0;
1465         init_timer(&waiting_locks_timer);
1466 #endif
1467
1468         RETURN(0);
1469
1470 #ifdef __KERNEL__
1471  out_thread:
1472         ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
1473         ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
1474 #endif
1475
1476  out_proc:
1477 #ifdef __KERNEL__
1478         ldlm_proc_cleanup();
1479  out_free:
1480 #endif
1481         OBD_FREE(ldlm_state, sizeof(*ldlm_state));
1482         ldlm_state = NULL;
1483         return rc;
1484 }
1485
1486 static int ldlm_cleanup(int force)
1487 {
1488 #ifdef __KERNEL__
1489         struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
1490 #endif
1491         ENTRY;
1492
1493         if (!list_empty(&ldlm_namespace_list)) {
1494                 CERROR("ldlm still has namespaces; clean these up first.\n");
1495                 ldlm_dump_all_namespaces(D_DLMTRACE);
1496                 RETURN(-EBUSY);
1497         }
1498
1499 #ifdef __KERNEL__
1500         while (atomic_read(&blp->blp_num_threads) > 0) {
1501                 struct ldlm_bl_work_item blwi = { .blwi_ns = NULL };
1502
1503                 init_completion(&blp->blp_comp);
1504
1505                 spin_lock(&blp->blp_lock);
1506                 list_add_tail(&blwi.blwi_entry, &blp->blp_list);
1507                 wake_up(&blp->blp_waitq);
1508                 spin_unlock(&blp->blp_lock);
1509
1510                 wait_for_completion(&blp->blp_comp);
1511         }
1512         OBD_FREE(blp, sizeof(*blp));
1513
1514         ptlrpc_stop_all_threads(ldlm_state->ldlm_cb_service);
1515         ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
1516         ptlrpc_stop_all_threads(ldlm_state->ldlm_cancel_service);
1517         ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
1518         ldlm_proc_cleanup();
1519
1520         expired_lock_thread.elt_state = ELT_TERMINATE;
1521         wake_up(&expired_lock_thread.elt_waitq);
1522         wait_event(expired_lock_thread.elt_waitq,
1523                    expired_lock_thread.elt_state == ELT_STOPPED);
1524 #else
1525         ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
1526         ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
1527 #endif
1528
1529         OBD_FREE(ldlm_state, sizeof(*ldlm_state));
1530         ldlm_state = NULL;
1531
1532         RETURN(0);
1533 }
1534
1535 int __init ldlm_init(void)
1536 {
1537         ldlm_resource_slab = kmem_cache_create("ldlm_resources",
1538                                                sizeof(struct ldlm_resource), 0,
1539                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1540         if (ldlm_resource_slab == NULL)
1541                 return -ENOMEM;
1542
1543         ldlm_lock_slab = kmem_cache_create("ldlm_locks",
1544                                            sizeof(struct ldlm_lock), 0,
1545                                            SLAB_HWCACHE_ALIGN, NULL, NULL);
1546         if (ldlm_lock_slab == NULL) {
1547                 kmem_cache_destroy(ldlm_resource_slab);
1548                 return -ENOMEM;
1549         }
1550
1551         l_lock_init(&ldlm_handle_lock);
1552
1553         return 0;
1554 }
1555
1556 void __exit ldlm_exit(void)
1557 {
1558         if ( ldlm_refcount )
1559                 CERROR("ldlm_refcount is %d in ldlm_exit!\n", ldlm_refcount);
1560         LASSERTF(kmem_cache_destroy(ldlm_resource_slab) == 0,
1561                  "couldn't free ldlm resource slab\n");
1562         LASSERTF(kmem_cache_destroy(ldlm_lock_slab) == 0,
1563                  "couldn't free ldlm lock slab\n");
1564
1565 }
1566
1567 /* ldlm_flock.c */
1568 EXPORT_SYMBOL(ldlm_flock_completion_ast);
1569
1570 /* ldlm_extent.c */
1571 EXPORT_SYMBOL(ldlm_extent_shift_kms);
1572
1573 /* ldlm_lock.c */
1574 EXPORT_SYMBOL(ldlm_get_processing_policy);
1575 EXPORT_SYMBOL(ldlm_lock2desc);
1576 EXPORT_SYMBOL(ldlm_register_intent);
1577 EXPORT_SYMBOL(ldlm_lockname);
1578 EXPORT_SYMBOL(ldlm_typename);
1579 EXPORT_SYMBOL(ldlm_lock2handle);
1580 EXPORT_SYMBOL(__ldlm_handle2lock);
1581 EXPORT_SYMBOL(ldlm_lock_get);
1582 EXPORT_SYMBOL(ldlm_lock_put);
1583 EXPORT_SYMBOL(ldlm_lock_match);
1584 EXPORT_SYMBOL(ldlm_lock_cancel);
1585 EXPORT_SYMBOL(ldlm_lock_addref);
1586 EXPORT_SYMBOL(ldlm_lock_decref);
1587 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
1588 EXPORT_SYMBOL(ldlm_lock_change_resource);
1589 EXPORT_SYMBOL(ldlm_lock_set_data);
1590 EXPORT_SYMBOL(ldlm_it2str);
1591 EXPORT_SYMBOL(ldlm_lock_dump);
1592 EXPORT_SYMBOL(ldlm_lock_dump_handle);
1593 EXPORT_SYMBOL(ldlm_cancel_locks_for_export);
1594 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
1595 EXPORT_SYMBOL(ldlm_lock_allow_match);
1596
1597 /* ldlm_request.c */
1598 EXPORT_SYMBOL(ldlm_completion_ast);
1599 EXPORT_SYMBOL(ldlm_expired_completion_wait);
1600 EXPORT_SYMBOL(ldlm_cli_convert);
1601 EXPORT_SYMBOL(ldlm_cli_enqueue);
1602 EXPORT_SYMBOL(ldlm_cli_cancel);
1603 EXPORT_SYMBOL(ldlm_cli_cancel_unused);
1604 EXPORT_SYMBOL(ldlm_replay_locks);
1605 EXPORT_SYMBOL(ldlm_resource_foreach);
1606 EXPORT_SYMBOL(ldlm_namespace_foreach);
1607 EXPORT_SYMBOL(ldlm_namespace_foreach_res);
1608 EXPORT_SYMBOL(ldlm_change_cbdata);
1609
1610 /* ldlm_lockd.c */
1611 EXPORT_SYMBOL(ldlm_server_blocking_ast);
1612 EXPORT_SYMBOL(ldlm_server_completion_ast);
1613 EXPORT_SYMBOL(ldlm_server_glimpse_ast);
1614 EXPORT_SYMBOL(ldlm_handle_enqueue);
1615 EXPORT_SYMBOL(ldlm_handle_cancel);
1616 EXPORT_SYMBOL(ldlm_handle_convert);
1617 EXPORT_SYMBOL(ldlm_del_waiting_lock);
1618 EXPORT_SYMBOL(ldlm_get_ref);
1619 EXPORT_SYMBOL(ldlm_put_ref);
1620
1621 #if 0
1622 /* ldlm_test.c */
1623 EXPORT_SYMBOL(ldlm_test);
1624 EXPORT_SYMBOL(ldlm_regression_start);
1625 EXPORT_SYMBOL(ldlm_regression_stop);
1626 #endif
1627
1628 /* ldlm_resource.c */
1629 EXPORT_SYMBOL(ldlm_namespace_new);
1630 EXPORT_SYMBOL(ldlm_namespace_cleanup);
1631 EXPORT_SYMBOL(ldlm_namespace_free);
1632 EXPORT_SYMBOL(ldlm_namespace_dump);
1633 EXPORT_SYMBOL(ldlm_dump_all_namespaces);
1634 EXPORT_SYMBOL(ldlm_resource_get);
1635 EXPORT_SYMBOL(ldlm_resource_putref);
1636
1637 /* l_lock.c */
1638 EXPORT_SYMBOL(l_lock);
1639 EXPORT_SYMBOL(l_unlock);
1640
1641 /* ldlm_lib.c */
1642 EXPORT_SYMBOL(client_import_add_conn);
1643 EXPORT_SYMBOL(client_import_del_conn);
1644 EXPORT_SYMBOL(client_obd_setup);
1645 EXPORT_SYMBOL(client_obd_cleanup);
1646 EXPORT_SYMBOL(client_connect_import);
1647 EXPORT_SYMBOL(client_disconnect_export);
1648 EXPORT_SYMBOL(target_start_recovery_thread);
1649 EXPORT_SYMBOL(target_stop_recovery_thread);
1650 EXPORT_SYMBOL(target_handle_connect);
1651 EXPORT_SYMBOL(target_cleanup_recovery);
1652 EXPORT_SYMBOL(target_destroy_export);
1653 EXPORT_SYMBOL(target_cancel_recovery_timer);
1654 EXPORT_SYMBOL(target_send_reply);
1655 EXPORT_SYMBOL(target_queue_recovery_request);
1656 EXPORT_SYMBOL(target_handle_ping);
1657 EXPORT_SYMBOL(target_handle_disconnect);
1658 EXPORT_SYMBOL(target_queue_final_reply);