1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
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.
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.
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.
25 #define DEBUG_SUBSYSTEM S_LDLM
28 # include <linux/module.h>
29 # include <linux/slab.h>
30 # include <linux/init.h>
31 # include <linux/wait.h>
33 # include <liblustre.h>
36 #include <linux/lustre_dlm.h>
37 #include <linux/obd_class.h>
38 extern kmem_cache_t *ldlm_resource_slab;
39 extern kmem_cache_t *ldlm_lock_slab;
40 extern struct lustre_lock ldlm_handle_lock;
41 extern struct list_head ldlm_namespace_list;
42 extern int (*mds_reint_p)(int offset, struct ptlrpc_request *req);
43 extern int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req);
45 static int ldlm_already_setup = 0;
49 inline unsigned long round_timeout(unsigned long timeout)
51 return ((timeout / HZ) + 1) * HZ;
54 /* XXX should this be per-ldlm? */
55 static struct list_head waiting_locks_list;
56 static spinlock_t waiting_locks_spinlock;
57 static struct timer_list waiting_locks_timer;
59 static struct expired_lock_thread {
60 wait_queue_head_t elt_waitq;
62 struct list_head elt_expired_locks;
64 } expired_lock_thread;
68 #define ELT_TERMINATE 2
70 static inline int have_expired_locks(void)
74 spin_lock_bh(&expired_lock_thread.elt_lock);
75 need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks);
76 spin_unlock_bh(&expired_lock_thread.elt_lock);
81 static int expired_lock_main(void *arg)
83 struct list_head *expired = &expired_lock_thread.elt_expired_locks;
84 struct l_wait_info lwi = { 0 };
89 kportal_daemonize("ldlm_elt");
91 SIGNAL_MASK_LOCK(current, flags);
92 sigfillset(¤t->blocked);
94 SIGNAL_MASK_UNLOCK(current, flags);
98 expired_lock_thread.elt_state = ELT_READY;
99 wake_up(&expired_lock_thread.elt_waitq);
102 l_wait_event(expired_lock_thread.elt_waitq,
103 have_expired_locks() ||
104 expired_lock_thread.elt_state == ELT_TERMINATE,
107 spin_lock_bh(&expired_lock_thread.elt_lock);
108 while (!list_empty(expired)) {
109 struct ldlm_lock *lock = list_entry(expired->next,
112 spin_unlock_bh(&expired_lock_thread.elt_lock);
114 ptlrpc_fail_export(lock->l_export);
116 spin_lock_bh(&expired_lock_thread.elt_lock);
118 spin_unlock_bh(&expired_lock_thread.elt_lock);
120 if (expired_lock_thread.elt_state == ELT_TERMINATE)
124 expired_lock_thread.elt_state = ELT_STOPPED;
125 wake_up(&expired_lock_thread.elt_waitq);
129 static void waiting_locks_callback(unsigned long unused)
131 struct ldlm_lock *lock;
133 spin_lock_bh(&waiting_locks_spinlock);
134 while (!list_empty(&waiting_locks_list)) {
135 lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
138 if (lock->l_callback_timeout > jiffies)
141 LDLM_ERROR(lock, "lock callback timer expired: evicting client "
143 lock->l_export->exp_client_uuid.uuid,
144 lock->l_export->exp_connection->c_remote_uuid.uuid,
145 lock->l_export->exp_connection->c_peer.peer_nid);
147 spin_lock_bh(&expired_lock_thread.elt_lock);
148 list_del(&lock->l_pending_chain);
149 list_add(&lock->l_pending_chain,
150 &expired_lock_thread.elt_expired_locks);
151 spin_unlock_bh(&expired_lock_thread.elt_lock);
152 wake_up(&expired_lock_thread.elt_waitq);
155 spin_unlock_bh(&waiting_locks_spinlock);
159 * Indicate that we're waiting for a client to call us back cancelling a given
160 * lock. We add it to the pending-callback chain, and schedule the lock-timeout
161 * timer to fire appropriately. (We round up to the next second, to avoid
162 * floods of timer firings during periods of high lock contention and traffic).
164 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
166 unsigned long timeout_rounded;
168 LDLM_DEBUG(lock, "adding to wait list");
169 LASSERT(list_empty(&lock->l_pending_chain));
171 spin_lock_bh(&waiting_locks_spinlock);
172 lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2);
174 timeout_rounded = round_timeout(lock->l_callback_timeout);
176 if (timeout_rounded < waiting_locks_timer.expires ||
177 !timer_pending(&waiting_locks_timer)) {
178 mod_timer(&waiting_locks_timer, timeout_rounded);
180 list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
181 spin_unlock_bh(&waiting_locks_spinlock);
182 /* We drop this ref when we get removed from the list. */
183 class_export_get(lock->l_export);
188 * Remove a lock from the pending list, likely because it had its cancellation
189 * callback arrive without incident. This adjusts the lock-timeout timer if
190 * needed. Returns 0 if the lock wasn't pending after all, 1 if it was.
192 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
194 struct list_head *list_next;
196 if (lock->l_export == NULL) {
197 /* We don't have a "waiting locks list" on clients. */
198 LDLM_DEBUG(lock, "client lock: no-op");
202 spin_lock_bh(&waiting_locks_spinlock);
204 if (list_empty(&lock->l_pending_chain)) {
205 spin_unlock_bh(&waiting_locks_spinlock);
206 LDLM_DEBUG(lock, "wasn't waiting");
210 list_next = lock->l_pending_chain.next;
211 if (lock->l_pending_chain.prev == &waiting_locks_list) {
212 /* Removing the head of the list, adjust timer. */
213 if (list_next == &waiting_locks_list) {
214 /* No more, just cancel. */
215 del_timer(&waiting_locks_timer);
217 struct ldlm_lock *next;
218 next = list_entry(list_next, struct ldlm_lock,
220 mod_timer(&waiting_locks_timer,
221 round_timeout(next->l_callback_timeout));
224 list_del_init(&lock->l_pending_chain);
225 spin_unlock_bh(&waiting_locks_spinlock);
226 /* We got this ref when we were added to the list. */
227 class_export_put(lock->l_export);
228 LDLM_DEBUG(lock, "removed");
232 #else /* !__KERNEL__ */
234 static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
239 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
244 #endif /* __KERNEL__ */
246 static inline void ldlm_failed_ast(struct ldlm_lock *lock, int rc,
249 CERROR("%s AST failed (%d) for res "LPU64"/"LPU64
250 ", mode %s: evicting client %s@%s NID "LPU64"\n",
252 lock->l_resource->lr_name.name[0],
253 lock->l_resource->lr_name.name[1],
254 ldlm_lockname[lock->l_granted_mode],
255 lock->l_export->exp_client_uuid.uuid,
256 lock->l_export->exp_connection->c_remote_uuid.uuid,
257 lock->l_export->exp_connection->c_peer.peer_nid);
258 ptlrpc_fail_export(lock->l_export);
261 int ldlm_server_blocking_ast(struct ldlm_lock *lock,
262 struct ldlm_lock_desc *desc,
263 void *data, int flag)
265 struct ldlm_request *body;
266 struct ptlrpc_request *req;
267 int rc = 0, size = sizeof(*body);
270 if (flag == LDLM_CB_CANCELING) {
271 /* Don't need to do anything here. */
277 l_lock(&lock->l_resource->lr_namespace->ns_lock);
278 /* XXX This is necessary because, with the lock re-tasking, we actually
279 * _can_ get called in here twice. (bug 830) */
280 if (!list_empty(&lock->l_pending_chain)) {
281 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
285 if (lock->l_destroyed) {
286 /* What's the point? */
287 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
292 if (LTIME_S(CURRENT_TIME) - lock->l_export->exp_last_request_time > 30){
293 ldlm_failed_ast(lock, -ETIMEDOUT, "Not-attempted blocking");
298 req = ptlrpc_prep_req(lock->l_export->exp_ldlm_data.led_import,
299 LDLM_BL_CALLBACK, 1, &size, NULL);
303 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
304 memcpy(&body->lock_handle1, &lock->l_remote_handle,
305 sizeof(body->lock_handle1));
306 memcpy(&body->lock_desc, desc, sizeof(*desc));
308 LDLM_DEBUG(lock, "server preparing blocking AST");
309 req->rq_replen = lustre_msg_size(0, NULL);
311 ldlm_add_waiting_lock(lock);
312 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
314 req->rq_level = LUSTRE_CONN_RECOVER;
315 req->rq_timeout = 2; /* 2 second timeout for initial AST reply */
316 rc = ptlrpc_queue_wait(req);
317 if (rc == -ETIMEDOUT || rc == -EINTR) {
318 ldlm_del_waiting_lock(lock);
319 ldlm_failed_ast(lock, rc, "blocking");
322 CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d "
323 "from blocking AST for lock %p--normal race\n",
324 req->rq_connection->c_peer.peer_nid,
325 req->rq_repmsg->status, lock);
326 else if (rc == -ENOTCONN)
327 CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d "
328 "from blocking AST for lock %p--this client was "
329 "probably rebooted while it held a lock, nothing"
330 " serious\n",req->rq_connection->c_peer.peer_nid,
331 req->rq_repmsg->status, lock);
333 CDEBUG(D_ERROR, "client (nid "LPU64") returned %d "
334 "from blocking AST for lock %p\n",
335 req->rq_connection->c_peer.peer_nid,
336 req->rq_repmsg->status, lock);
337 LDLM_DEBUG(lock, "client returned error %d from blocking AST",
339 ldlm_lock_cancel(lock);
340 /* Server-side AST functions are called from ldlm_reprocess_all,
341 * which needs to be told to please restart its reprocessing. */
345 ptlrpc_req_finished(req);
350 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data)
352 struct ldlm_request *body;
353 struct ptlrpc_request *req;
354 int rc = 0, size = sizeof(*body);
362 req = ptlrpc_prep_req(lock->l_export->exp_ldlm_data.led_import,
363 LDLM_CP_CALLBACK, 1, &size, NULL);
367 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
368 memcpy(&body->lock_handle1, &lock->l_remote_handle,
369 sizeof(body->lock_handle1));
370 body->lock_flags = flags;
371 ldlm_lock2desc(lock, &body->lock_desc);
373 LDLM_DEBUG(lock, "server preparing completion AST");
374 req->rq_replen = lustre_msg_size(0, NULL);
376 req->rq_level = LUSTRE_CONN_RECOVER;
377 req->rq_timeout = 2; /* 2 second timeout for initial AST reply */
378 rc = ptlrpc_queue_wait(req);
379 if (rc == -ETIMEDOUT || rc == -EINTR) {
380 ldlm_del_waiting_lock(lock);
381 ldlm_failed_ast(lock, rc, "completion");
383 CERROR("client returned %d from completion AST for lock %p\n",
384 req->rq_status, lock);
385 LDLM_DEBUG(lock, "client returned error %d from completion AST",
387 ldlm_lock_cancel(lock);
388 /* Server-side AST functions are called from ldlm_reprocess_all,
389 * which needs to be told to please restart its reprocessing. */
392 ptlrpc_req_finished(req);
397 int ldlm_handle_enqueue(struct ptlrpc_request *req,
398 ldlm_completion_callback completion_callback,
399 ldlm_blocking_callback blocking_callback)
401 struct obd_device *obddev = req->rq_export->exp_obd;
402 struct ldlm_reply *dlm_rep;
403 struct ldlm_request *dlm_req;
404 int rc, size = sizeof(*dlm_rep), cookielen = 0;
407 struct ldlm_lock *lock = NULL;
411 LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
413 dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
414 lustre_swab_ldlm_request);
415 if (dlm_req == NULL) {
416 CERROR ("Can't unpack dlm_req\n");
420 flags = dlm_req->lock_flags;
421 if (dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN &&
422 (flags & LDLM_FL_HAS_INTENT)) {
423 /* In this case, the reply buffer is allocated deep in
424 * local_lock_enqueue by the policy function. */
426 cookielen = sizeof(*req);
428 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
431 CERROR("out of memory\n");
434 if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) {
435 cookie = &dlm_req->lock_desc.l_extent;
436 cookielen = sizeof(struct ldlm_extent);
440 /* The lock's callback data might be set in the policy function */
441 lock = ldlm_lock_create(obddev->obd_namespace,
442 &dlm_req->lock_handle2,
443 dlm_req->lock_desc.l_resource.lr_name,
444 dlm_req->lock_desc.l_resource.lr_type,
445 dlm_req->lock_desc.l_req_mode,
446 blocking_callback, NULL);
448 GOTO(out, err = -ENOMEM);
450 memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
451 sizeof(lock->l_remote_handle));
452 LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
454 LASSERT(req->rq_export);
455 lock->l_export = req->rq_export;
456 l_lock(&lock->l_resource->lr_namespace->ns_lock);
457 list_add(&lock->l_export_chain,
458 &lock->l_export->exp_ldlm_data.led_held_locks);
459 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
461 err = ldlm_lock_enqueue(obddev->obd_namespace, &lock, cookie, cookielen,
462 &flags, completion_callback);
466 dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
467 dlm_rep->lock_flags = flags;
469 ldlm_lock2handle(lock, &dlm_rep->lock_handle);
470 if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT)
471 memcpy(&dlm_rep->lock_extent, &lock->l_extent,
472 sizeof(lock->l_extent));
473 if (dlm_rep->lock_flags & LDLM_FL_LOCK_CHANGED) {
474 memcpy(&dlm_rep->lock_resource_name, &lock->l_resource->lr_name,
475 sizeof(dlm_rep->lock_resource_name));
476 dlm_rep->lock_mode = lock->l_req_mode;
482 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
484 req->rq_status = err;
486 /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
487 * ldlm_reprocess_all. If this moves, revisit that code. -phil */
490 ldlm_reprocess_all(lock->l_resource);
493 LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p)", lock);
498 int ldlm_handle_convert(struct ptlrpc_request *req)
500 struct ldlm_request *dlm_req;
501 struct ldlm_reply *dlm_rep;
502 struct ldlm_lock *lock;
503 int rc, size = sizeof(*dlm_rep);
506 dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
507 lustre_swab_ldlm_request);
508 if (dlm_req == NULL) {
509 CERROR ("Can't unpack dlm_req\n");
513 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
515 CERROR("out of memory\n");
518 dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
519 dlm_rep->lock_flags = dlm_req->lock_flags;
521 lock = ldlm_handle2lock(&dlm_req->lock_handle1);
523 req->rq_status = EINVAL;
525 LDLM_DEBUG(lock, "server-side convert handler START");
526 ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode,
527 &dlm_rep->lock_flags);
528 if (ldlm_del_waiting_lock(lock))
529 CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock);
534 ldlm_reprocess_all(lock->l_resource);
535 LDLM_DEBUG(lock, "server-side convert handler END");
538 LDLM_DEBUG_NOLOCK("server-side convert handler END");
543 int ldlm_handle_cancel(struct ptlrpc_request *req)
545 struct ldlm_request *dlm_req;
546 struct ldlm_lock *lock;
550 dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
551 lustre_swab_ldlm_request);
552 if (dlm_req == NULL) {
553 CERROR("bad request buffer for cancel\n");
557 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
559 CERROR("out of memory\n");
563 lock = ldlm_handle2lock(&dlm_req->lock_handle1);
565 CERROR("received cancel for unknown lock cookie "LPX64
566 " from nid "LPU64"\n", dlm_req->lock_handle1.cookie,
567 req->rq_connection->c_peer.peer_nid);
568 LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock "
570 dlm_req->lock_handle1.cookie);
571 req->rq_status = ESTALE;
573 LDLM_DEBUG(lock, "server-side cancel handler START");
574 ldlm_lock_cancel(lock);
575 if (ldlm_del_waiting_lock(lock))
576 CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock);
580 if (ptlrpc_reply(req) != 0)
584 ldlm_reprocess_all(lock->l_resource);
585 LDLM_DEBUG(lock, "server-side cancel handler END");
592 static void ldlm_handle_bl_callback(struct ptlrpc_request *req,
593 struct ldlm_namespace *ns,
594 struct ldlm_request *dlm_req,
595 struct ldlm_lock *lock)
600 l_lock(&ns->ns_lock);
601 LDLM_DEBUG(lock, "client blocking AST callback handler START");
603 lock->l_flags |= LDLM_FL_CBPENDING;
604 do_ast = (!lock->l_readers && !lock->l_writers);
607 LDLM_DEBUG(lock, "already unused, calling "
608 "callback (%p)", lock->l_blocking_ast);
609 if (lock->l_blocking_ast != NULL) {
610 l_unlock(&ns->ns_lock);
611 lock->l_blocking_ast(lock, &dlm_req->lock_desc,
612 lock->l_data, LDLM_CB_BLOCKING);
613 l_lock(&ns->ns_lock);
616 LDLM_DEBUG(lock, "Lock still has references, will be"
620 LDLM_DEBUG(lock, "client blocking callback handler END");
621 l_unlock(&ns->ns_lock);
626 static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
627 struct ldlm_namespace *ns,
628 struct ldlm_request *dlm_req,
629 struct ldlm_lock *lock)
634 l_lock(&ns->ns_lock);
635 LDLM_DEBUG(lock, "client completion callback handler START");
637 /* If we receive the completion AST before the actual enqueue returned,
638 * then we might need to switch lock modes, resources, or extents. */
639 if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
640 lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
641 LDLM_DEBUG(lock, "completion AST, new lock mode");
643 if (lock->l_resource->lr_type == LDLM_EXTENT) {
644 memcpy(&lock->l_extent, &dlm_req->lock_desc.l_extent,
645 sizeof(lock->l_extent));
647 if ((lock->l_extent.end & ~PAGE_MASK) != ~PAGE_MASK) {
648 /* XXX Old versions of BA OST code have a fencepost bug
649 * which will cause them to grant a lock that's one
650 * byte too large. This can be safely removed after BA
651 * ships their next release -phik (02 Apr 2003) */
652 lock->l_extent.end--;
653 } else if ((lock->l_extent.start & ~PAGE_MASK) ==
655 lock->l_extent.start++;
659 ldlm_resource_unlink_lock(lock);
660 if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
661 &lock->l_resource->lr_name,
662 sizeof(lock->l_resource->lr_name)) != 0) {
663 ldlm_lock_change_resource(ns, lock,
664 dlm_req->lock_desc.l_resource.lr_name);
665 LDLM_DEBUG(lock, "completion AST, new resource");
667 lock->l_resource->lr_tmp = &ast_list;
668 ldlm_grant_lock(lock, req, sizeof(*req));
669 lock->l_resource->lr_tmp = NULL;
670 LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
671 l_unlock(&ns->ns_lock);
674 ldlm_run_ast_work(&ast_list);
676 LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
681 static int ldlm_callback_reply(struct ptlrpc_request *req, int rc)
684 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen,
688 return ptlrpc_reply(req);
691 static int ldlm_callback_handler(struct ptlrpc_request *req)
693 struct ldlm_namespace *ns;
694 struct ldlm_request *dlm_req;
695 struct ldlm_lock *lock;
698 /* Requests arrive in sender's byte order. The ptlrpc service
699 * handler has already checked and, if necessary, byte-swapped the
700 * incoming request message body, but I am responsible for the
701 * message buffers. */
703 if (req->rq_export == NULL) {
704 struct ldlm_request *dlm_req;
706 CDEBUG(D_RPCTRACE, "operation %d from nid "LPU64" with bad "
707 "export cookie "LPX64" (ptl req %d/rep %d); this is "
708 "normal if this node rebooted with a lock held\n",
709 req->rq_reqmsg->opc, req->rq_connection->c_peer.peer_nid,
710 req->rq_reqmsg->handle.cookie,
711 req->rq_request_portal, req->rq_reply_portal);
713 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
714 lustre_swab_ldlm_request);
716 CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n",
717 dlm_req->lock_handle1.cookie);
719 ldlm_callback_reply(req, -ENOTCONN);
723 if (req->rq_reqmsg->opc == LDLM_BL_CALLBACK) {
724 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
725 } else if (req->rq_reqmsg->opc == LDLM_CP_CALLBACK) {
726 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0);
728 ldlm_callback_reply(req, -EPROTO);
732 LASSERT(req->rq_export != NULL);
733 LASSERT(req->rq_export->exp_obd != NULL);
734 ns = req->rq_export->exp_obd->obd_namespace;
737 dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req),
738 lustre_swab_ldlm_request);
739 if (dlm_req == NULL) {
740 CERROR ("can't unpack dlm_req\n");
741 ldlm_callback_reply (req, -EPROTO);
745 lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle1);
747 CDEBUG(D_INODE, "callback on lock "LPX64" - lock disappeared\n",
748 dlm_req->lock_handle1.cookie);
749 ldlm_callback_reply(req, -EINVAL);
753 /* we want the ost thread to get this reply so that it can respond
754 * to ost requests (write cache writeback) that might be triggered
756 ldlm_callback_reply(req, 0);
758 switch (req->rq_reqmsg->opc) {
759 case LDLM_BL_CALLBACK:
760 CDEBUG(D_INODE, "blocking ast\n");
761 ldlm_handle_bl_callback(req, ns, dlm_req, lock);
763 case LDLM_CP_CALLBACK:
764 CDEBUG(D_INODE, "completion ast\n");
765 ldlm_handle_cp_callback(req, ns, dlm_req, lock);
768 LBUG(); /* checked above */
774 static int ldlm_cancel_handler(struct ptlrpc_request *req)
779 /* Requests arrive in sender's byte order. The ptlrpc service
780 * handler has already checked and, if necessary, byte-swapped the
781 * incoming request message body, but I am responsible for the
782 * message buffers. */
784 if (req->rq_export == NULL) {
785 struct ldlm_request *dlm_req;
786 CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
787 req->rq_reqmsg->opc, req->rq_request_portal,
788 req->rq_reply_portal);
789 CERROR("--> export cookie: "LPX64"\n",
790 req->rq_reqmsg->handle.cookie);
791 dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
792 lustre_swab_ldlm_request);
794 ldlm_lock_dump_handle(D_ERROR, &dlm_req->lock_handle1);
798 switch (req->rq_reqmsg->opc) {
800 /* XXX FIXME move this back to mds/handler.c, bug 249 */
802 CDEBUG(D_INODE, "cancel\n");
803 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
804 rc = ldlm_handle_cancel(req);
810 CERROR("invalid opcode %d\n", req->rq_reqmsg->opc);
817 static int ldlm_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len,
818 void *karg, void *uarg)
820 struct obd_device *obddev = class_conn2obd(conn);
821 struct ptlrpc_connection *connection;
822 struct obd_uuid uuid = { "ldlm" };
826 if (_IOC_TYPE(cmd) != IOC_LDLM_TYPE || _IOC_NR(cmd) < IOC_LDLM_MIN_NR ||
827 _IOC_NR(cmd) > IOC_LDLM_MAX_NR) {
828 CDEBUG(D_IOCTL, "invalid ioctl (type %d, nr %d, size %d)\n",
829 _IOC_TYPE(cmd), _IOC_NR(cmd), _IOC_SIZE(cmd));
833 OBD_ALLOC(obddev->u.ldlm.ldlm_client,
834 sizeof(*obddev->u.ldlm.ldlm_client));
835 connection = ptlrpc_uuid_to_connection(&uuid);
837 CERROR("No LDLM UUID found: assuming ldlm is local.\n");
841 //err = ldlm_test(obddev, conn);
843 CERROR("-- NO TESTS WERE RUN done err %d\n", err);
846 ldlm_dump_all_namespaces();
849 GOTO(out, err = -EINVAL);
854 ptlrpc_put_connection(connection);
855 OBD_FREE(obddev->u.ldlm.ldlm_client,
856 sizeof(*obddev->u.ldlm.ldlm_client));
860 static int ldlm_setup(struct obd_device *obddev, obd_count len, void *buf)
862 struct ldlm_obd *ldlm = &obddev->u.ldlm;
866 if (ldlm_already_setup)
869 rc = ldlm_proc_setup(obddev);
874 inter_module_register("ldlm_cli_cancel_unused", THIS_MODULE,
875 ldlm_cli_cancel_unused);
876 inter_module_register("ldlm_namespace_cleanup", THIS_MODULE,
877 ldlm_namespace_cleanup);
878 inter_module_register("ldlm_replay_locks", THIS_MODULE,
881 ldlm->ldlm_cb_service =
882 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
883 LDLM_MAXREQSIZE, LDLM_CB_REQUEST_PORTAL,
884 LDLM_CB_REPLY_PORTAL,
885 ldlm_callback_handler, "ldlm_cbd", obddev);
887 if (!ldlm->ldlm_cb_service) {
888 CERROR("failed to start service\n");
889 GOTO(out_proc, rc = -ENOMEM);
892 ldlm->ldlm_cancel_service =
893 ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
894 LDLM_MAXREQSIZE, LDLM_CANCEL_REQUEST_PORTAL,
895 LDLM_CANCEL_REPLY_PORTAL,
896 ldlm_cancel_handler, "ldlm_canceld", obddev);
898 if (!ldlm->ldlm_cancel_service) {
899 CERROR("failed to start service\n");
900 GOTO(out_proc, rc = -ENOMEM);
903 for (i = 0; i < LDLM_NUM_THREADS; i++) {
905 sprintf(name, "ldlm_cn_%02d", i);
906 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cancel_service,
909 CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
911 GOTO(out_thread, rc);
915 for (i = 0; i < LDLM_NUM_THREADS; i++) {
917 sprintf(name, "ldlm_cb_%02d", i);
918 rc = ptlrpc_start_thread(obddev, ldlm->ldlm_cb_service, name);
920 CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc);
922 GOTO(out_thread, rc);
926 INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
927 spin_lock_init(&expired_lock_thread.elt_lock);
928 expired_lock_thread.elt_state = ELT_STOPPED;
929 init_waitqueue_head(&expired_lock_thread.elt_waitq);
931 rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS);
933 CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
934 GOTO(out_thread, rc);
937 wait_event(expired_lock_thread.elt_waitq,
938 expired_lock_thread.elt_state == ELT_READY);
940 INIT_LIST_HEAD(&waiting_locks_list);
941 spin_lock_init(&waiting_locks_spinlock);
942 waiting_locks_timer.function = waiting_locks_callback;
943 waiting_locks_timer.data = 0;
944 init_timer(&waiting_locks_timer);
947 ldlm_already_setup = 1;
953 ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
954 ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
955 ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
956 ptlrpc_unregister_service(ldlm->ldlm_cb_service);
959 ldlm_proc_cleanup(obddev);
964 static int ldlm_cleanup(struct obd_device *obddev, int force, int failover)
966 struct ldlm_obd *ldlm = &obddev->u.ldlm;
969 if (!list_empty(&ldlm_namespace_list)) {
970 CERROR("ldlm still has namespaces; clean these up first.\n");
971 ldlm_dump_all_namespaces();
977 ptlrpc_put_ldlm_hooks();
978 } else if (ptlrpc_ldlm_hooks_referenced()) {
979 CERROR("Some connections weren't cleaned up; run lconf with "
980 "--force to forcibly unload.\n");
981 ptlrpc_dump_connections();
985 ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
986 ptlrpc_unregister_service(ldlm->ldlm_cb_service);
987 ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
988 ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
989 ldlm_proc_cleanup(obddev);
991 expired_lock_thread.elt_state = ELT_TERMINATE;
992 wake_up(&expired_lock_thread.elt_waitq);
993 wait_event(expired_lock_thread.elt_waitq,
994 expired_lock_thread.elt_state == ELT_STOPPED);
996 inter_module_unregister("ldlm_namespace_cleanup");
997 inter_module_unregister("ldlm_cli_cancel_unused");
998 inter_module_unregister("ldlm_replay_locks");
1001 ldlm_already_setup = 0;
1005 static int ldlm_connect(struct lustre_handle *conn, struct obd_device *src,
1006 struct obd_uuid *cluuid)
1008 return class_connect(conn, src, cluuid);
1011 struct obd_ops ldlm_obd_ops = {
1012 o_owner: THIS_MODULE,
1013 o_iocontrol: ldlm_iocontrol,
1014 o_setup: ldlm_setup,
1015 o_cleanup: ldlm_cleanup,
1016 o_connect: ldlm_connect,
1017 o_disconnect: class_disconnect
1020 int __init ldlm_init(void)
1022 int rc = class_register_type(&ldlm_obd_ops, 0, OBD_LDLM_DEVICENAME);
1026 ldlm_resource_slab = kmem_cache_create("ldlm_resources",
1027 sizeof(struct ldlm_resource), 0,
1028 SLAB_HWCACHE_ALIGN, NULL, NULL);
1029 if (ldlm_resource_slab == NULL)
1032 ldlm_lock_slab = kmem_cache_create("ldlm_locks",
1033 sizeof(struct ldlm_lock), 0,
1034 SLAB_HWCACHE_ALIGN, NULL, NULL);
1035 if (ldlm_lock_slab == NULL) {
1036 kmem_cache_destroy(ldlm_resource_slab);
1040 l_lock_init(&ldlm_handle_lock);
1045 static void __exit ldlm_exit(void)
1047 class_unregister_type(OBD_LDLM_DEVICENAME);
1048 if (kmem_cache_destroy(ldlm_resource_slab) != 0)
1049 CERROR("couldn't free ldlm resource slab\n");
1050 if (kmem_cache_destroy(ldlm_lock_slab) != 0)
1051 CERROR("couldn't free ldlm lock slab\n");
1055 EXPORT_SYMBOL(ldlm_lock2desc);
1056 EXPORT_SYMBOL(ldlm_register_intent);
1057 EXPORT_SYMBOL(ldlm_unregister_intent);
1058 EXPORT_SYMBOL(ldlm_lockname);
1059 EXPORT_SYMBOL(ldlm_typename);
1060 EXPORT_SYMBOL(ldlm_lock2handle);
1061 EXPORT_SYMBOL(__ldlm_handle2lock);
1062 EXPORT_SYMBOL(ldlm_lock_put);
1063 EXPORT_SYMBOL(ldlm_lock_match);
1064 EXPORT_SYMBOL(ldlm_lock_cancel);
1065 EXPORT_SYMBOL(ldlm_lock_addref);
1066 EXPORT_SYMBOL(ldlm_lock_decref);
1067 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
1068 EXPORT_SYMBOL(ldlm_lock_change_resource);
1069 EXPORT_SYMBOL(ldlm_lock_set_data);
1070 EXPORT_SYMBOL(ldlm_it2str);
1071 EXPORT_SYMBOL(ldlm_lock_dump);
1072 EXPORT_SYMBOL(ldlm_lock_dump_handle);
1073 EXPORT_SYMBOL(ldlm_cancel_locks_for_export);
1074 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
1076 /* ldlm_request.c */
1077 EXPORT_SYMBOL(ldlm_completion_ast);
1078 EXPORT_SYMBOL(ldlm_expired_completion_wait);
1079 EXPORT_SYMBOL(ldlm_cli_convert);
1080 EXPORT_SYMBOL(ldlm_cli_enqueue);
1081 EXPORT_SYMBOL(ldlm_cli_cancel);
1082 EXPORT_SYMBOL(ldlm_cli_cancel_unused);
1083 EXPORT_SYMBOL(ldlm_replay_locks);
1084 EXPORT_SYMBOL(ldlm_resource_foreach);
1085 EXPORT_SYMBOL(ldlm_namespace_foreach);
1086 EXPORT_SYMBOL(ldlm_namespace_foreach_res);
1089 EXPORT_SYMBOL(ldlm_server_blocking_ast);
1090 EXPORT_SYMBOL(ldlm_server_completion_ast);
1091 EXPORT_SYMBOL(ldlm_handle_enqueue);
1092 EXPORT_SYMBOL(ldlm_handle_cancel);
1093 EXPORT_SYMBOL(ldlm_handle_convert);
1094 EXPORT_SYMBOL(ldlm_del_waiting_lock);
1098 EXPORT_SYMBOL(ldlm_test);
1099 EXPORT_SYMBOL(ldlm_regression_start);
1100 EXPORT_SYMBOL(ldlm_regression_stop);
1103 /* ldlm_resource.c */
1104 EXPORT_SYMBOL(ldlm_namespace_new);
1105 EXPORT_SYMBOL(ldlm_namespace_cleanup);
1106 EXPORT_SYMBOL(ldlm_namespace_free);
1107 EXPORT_SYMBOL(ldlm_namespace_dump);
1110 EXPORT_SYMBOL(l_lock);
1111 EXPORT_SYMBOL(l_unlock);
1114 EXPORT_SYMBOL(client_import_connect);
1115 EXPORT_SYMBOL(client_import_disconnect);
1116 EXPORT_SYMBOL(target_abort_recovery);
1117 EXPORT_SYMBOL(target_handle_connect);
1118 EXPORT_SYMBOL(target_cancel_recovery_timer);
1119 EXPORT_SYMBOL(target_send_reply);
1120 EXPORT_SYMBOL(target_queue_recovery_request);
1121 EXPORT_SYMBOL(target_handle_ping);
1122 EXPORT_SYMBOL(target_handle_disconnect);
1123 EXPORT_SYMBOL(target_queue_final_reply);
1126 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1127 MODULE_DESCRIPTION("Lustre Lock Management Module v0.1");
1128 MODULE_LICENSE("GPL");
1130 module_init(ldlm_init);
1131 module_exit(ldlm_exit);