/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
- * Author: Peter Braam <braam@clusterfs.com>
- * Author: Phil Schwan <phil@clusterfs.com>
+ * GPL HEADER START
*
- * This file is part of Lustre, http://www.lustre.org.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * Lustre is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see [sun.com URL with a
+ * copy of GPLv2].
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2003 Hewlett-Packard Development Company LP.
+ * Developed under the sponsorship of the US Government under
+ * Subcontract No. B514193
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
*/
#define DEBUG_SUBSYSTEM S_LDLM
#ifdef __KERNEL__
-#include <linux/lustre_dlm.h>
-#include <linux/obd_support.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_lib.h>
-#include <portals/list.h>
+#include <lustre_dlm.h>
+#include <obd_support.h>
+#include <obd_class.h>
+#include <lustre_lib.h>
+#include <libcfs/list.h>
#else
#include <liblustre.h>
+#include <obd_class.h>
#endif
#include "ldlm_internal.h"
#define l_flock_waitq l_lru
-static struct list_head ldlm_flock_waitq = LIST_HEAD_INIT(ldlm_flock_waitq);
+/**
+ * Wait queue for Posix lock deadlock detection, added with
+ * ldlm_lock::l_flock_waitq.
+ */
+static CFS_LIST_HEAD(ldlm_flock_waitq);
+/**
+ * Lock protecting access to ldlm_flock_waitq.
+ */
+spinlock_t ldlm_flock_waitq_lock = SPIN_LOCK_UNLOCKED;
+
+int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+ void *data, int flag);
/**
* list_for_remaining_safe - iterate over the remaining entries in a list
LDLM_DEBUG(lock, "ldlm_flock_destroy(mode: %d, flags: 0x%x)",
mode, flags);
+ /* Safe to not lock here, since it should be empty anyway */
+ LASSERT(list_empty(&lock->l_flock_waitq));
+
list_del_init(&lock->l_res_link);
if (flags == LDLM_FL_WAIT_NOREPROC) {
/* client side - set a flag to prevent sending a CANCEL */
ldlm_lock_decref_internal(lock, mode);
}
- ldlm_lock_destroy(lock);
+ ldlm_lock_destroy_nolock(lock);
EXIT;
}
pid_t blocking_pid = blocking_lock->l_policy_data.l_flock.pid;
struct ldlm_lock *lock;
+ spin_lock(&ldlm_flock_waitq_lock);
restart:
list_for_each_entry(lock, &ldlm_flock_waitq, l_flock_waitq) {
if ((lock->l_policy_data.l_flock.pid != blocking_pid) ||
blocking_pid = lock->l_policy_data.l_flock.blocking_pid;
blocking_export = (struct obd_export *)(long)
lock->l_policy_data.l_flock.blocking_export;
- if (blocking_pid == req_pid && blocking_export == req_export)
+ if (blocking_pid == req_pid && blocking_export == req_export) {
+ spin_unlock(&ldlm_flock_waitq_lock);
return 1;
+ }
goto restart;
}
+ spin_unlock(&ldlm_flock_waitq_lock);
return 0;
}
int
ldlm_process_flock_lock(struct ldlm_lock *req, int *flags, int first_enq,
- ldlm_error_t *err)
+ ldlm_error_t *err, struct list_head *work_list)
{
struct ldlm_resource *res = req->l_resource;
struct ldlm_namespace *ns = res->lr_namespace;
struct ldlm_lock *new = req;
struct ldlm_lock *new2 = NULL;
ldlm_mode_t mode = req->l_req_mode;
+ int local = ns_is_client(ns);
int added = (mode == LCK_NL);
int overlaps = 0;
ENTRY;
*err = ELDLM_OK;
- /* No blocking ASTs are sent for Posix file & record locks */
- req->l_blocking_ast = NULL;
+ if (local) {
+ /* No blocking ASTs are sent to the clients for
+ * Posix file & record locks */
+ req->l_blocking_ast = NULL;
+ } else {
+ /* Called on the server for lock cancels. */
+ req->l_blocking_ast = ldlm_flock_blocking_ast;
+ }
if ((*flags == LDLM_FL_WAIT_NOREPROC) || (mode == LCK_NL)) {
/* This loop determines where this processes locks start
}
}
} else {
+ lockmode_verify(mode);
+
/* This loop determines if there are existing locks
* that conflict with the new lock request. */
list_for_each(tmp, &res->lr_granted) {
/* locks are compatible, overlap doesn't matter */
if (lockmode_compat(lock->l_granted_mode, mode))
continue;
-
+
if (!ldlm_flocks_overlap(lock, req))
continue;
(long)(void *)lock->l_export;
LASSERT(list_empty(&req->l_flock_waitq));
+ spin_lock(&ldlm_flock_waitq_lock);
list_add_tail(&req->l_flock_waitq, &ldlm_flock_waitq);
+ spin_unlock(&ldlm_flock_waitq_lock);
ldlm_resource_add_lock(res, &res->lr_waiting, req);
*flags |= LDLM_FL_BLOCK_GRANTED;
RETURN(LDLM_ITER_STOP);
}
+ /* In case we had slept on this lock request take it off of the
+ * deadlock detection waitq. */
+ spin_lock(&ldlm_flock_waitq_lock);
+ list_del_init(&req->l_flock_waitq);
+ spin_unlock(&ldlm_flock_waitq_lock);
+
/* Scan the locks owned by this process that overlap this request.
* We may have to merge or split existing locks. */
* overflow and underflow. */
if ((new->l_policy_data.l_flock.start >
(lock->l_policy_data.l_flock.end + 1))
- && (lock->l_policy_data.l_flock.end != ~0))
+ && (lock->l_policy_data.l_flock.end !=
+ OBD_OBJECT_EOF))
continue;
if ((new->l_policy_data.l_flock.end <
/* XXX - if ldlm_lock_new() can sleep we should
* release the ns_lock, allocate the new lock,
* and restart processing this lock. */
- new2 = ldlm_lock_create(ns, NULL, res->lr_name, LDLM_FLOCK,
- lock->l_granted_mode, NULL, NULL, NULL);
+ new2 = ldlm_lock_create(ns, &res->lr_name, LDLM_FLOCK,
+ lock->l_granted_mode, NULL, NULL, NULL,
+ NULL, 0);
if (!new2) {
ldlm_flock_destroy(req, lock->l_granted_mode, *flags);
*err = -ENOLCK;
new2->l_conn_export = lock->l_conn_export;
if (lock->l_export != NULL) {
new2->l_export = class_export_get(lock->l_export);
+ spin_lock(&new2->l_export->exp_ldlm_data.led_lock);
list_add(&new2->l_export_chain,
&new2->l_export->exp_ldlm_data.led_held_locks);
+ spin_unlock(&new2->l_export->exp_ldlm_data.led_lock);
}
if (*flags == LDLM_FL_WAIT_NOREPROC)
ldlm_lock_addref_internal(new2, lock->l_granted_mode);
break;
}
+ /* At this point we're granting the lock request. */
+ req->l_granted_mode = req->l_req_mode;
+
/* Add req to the granted queue before calling ldlm_reprocess_all(). */
if (!added) {
- req->l_granted_mode = req->l_req_mode;
list_del_init(&req->l_res_link);
/* insert new lock before ownlocks in list. */
ldlm_resource_add_lock(res, ownlocks, req);
if (*flags != LDLM_FL_WAIT_NOREPROC) {
if (first_enq) {
/* If this is an unlock, reprocess the waitq and
- * send completions ASTs for locks that can now be
+ * send completions ASTs for locks that can now be
* granted. The only problem with doing this
* reprocessing here is that the completion ASTs for
* newly granted locks will be sent before the unlock
* but only once because first_enq will be false from
* ldlm_reprocess_queue. */
if ((mode == LCK_NL) && overlaps) {
- struct list_head rpc_list
- = LIST_HEAD_INIT(rpc_list);
+ CFS_LIST_HEAD(rpc_list);
int rc;
restart:
- res->lr_tmp = &rpc_list;
- ldlm_reprocess_queue(res, &res->lr_waiting);
- res->lr_tmp = NULL;
-
- l_unlock(&ns->ns_lock);
- rc = ldlm_run_ast_work(res->lr_namespace,
- &rpc_list);
- l_lock(&ns->ns_lock);
+ ldlm_reprocess_queue(res, &res->lr_waiting,
+ &rpc_list);
+
+ unlock_res(res);
+ rc = ldlm_run_ast_work(&rpc_list, LDLM_WORK_BL_AST);
+ lock_res(res);
if (rc == -ERESTART)
GOTO(restart, -ERESTART);
}
} else {
LASSERT(req->l_completion_ast);
- ldlm_add_ast_work_item(req, NULL, NULL, 0);
+ ldlm_add_ast_work_item(req, NULL, work_list);
}
}
if (added)
ldlm_flock_destroy(req, mode, *flags);
- ldlm_resource_dump(res);
+ ldlm_resource_dump(D_OTHER, res);
RETURN(LDLM_ITER_CONTINUE);
}
ENTRY;
lock = ((struct ldlm_flock_wait_data *)data)->fwd_lock;
+
+ /* take lock off the deadlock detection waitq. */
+ spin_lock(&ldlm_flock_waitq_lock);
+ list_del_init(&lock->l_flock_waitq);
+ spin_unlock(&ldlm_flock_waitq_lock);
+
+ /* client side - set flag to prevent lock from being put on lru list */
+ lock->l_flags |= LDLM_FL_CBPENDING;
+
ldlm_lock_decref_internal(lock, lock->l_req_mode);
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh);
- CDEBUG(D_DLMTRACE, "ldlm_cli_cancel: %d\n", rc);
+ if (rc != ELDLM_OK)
+ CERROR("ldlm_cli_cancel: %d\n", rc);
+
EXIT;
}
ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data)
{
struct ldlm_namespace *ns;
- struct file_lock *getlk = lock->l_ast_data;
+ cfs_flock_t *getlk = lock->l_ast_data;
struct ldlm_flock_wait_data fwd;
- unsigned long irqflags;
struct obd_device *obd;
struct obd_import *imp = NULL;
ldlm_error_t err;
CDEBUG(D_DLMTRACE, "flags: 0x%x data: %p getlk: %p\n",
flags, data, getlk);
- LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
-
- if (flags == 0) {
- wake_up(&lock->l_waitq);
+ /* Import invalidation. We need to actually release the lock
+ * references being held, so that it can go away. No point in
+ * holding the lock even if app still believes it has it, since
+ * server already dropped it anyway. Only for granted locks too. */
+ lock_res_and_lock(lock);
+ if ((lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) ==
+ (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) {
+ unlock_res_and_lock(lock);
+ if (lock->l_req_mode == lock->l_granted_mode &&
+ lock->l_granted_mode != LCK_NL)
+ ldlm_lock_decref_internal(lock, lock->l_req_mode);
RETURN(0);
}
+ unlock_res_and_lock(lock);
+
+ LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
LDLM_FL_BLOCK_CONV)))
LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
"sleeping");
- ldlm_lock_dump(D_OTHER, lock, 0);
-
fwd.fwd_lock = lock;
obd = class_exp2obd(lock->l_conn_export);
imp = obd->u.cli.cl_import;
if (imp != NULL) {
- spin_lock_irqsave(&imp->imp_lock, irqflags);
+ spin_lock(&imp->imp_lock);
fwd.fwd_generation = imp->imp_generation;
- spin_unlock_irqrestore(&imp->imp_lock, irqflags);
+ spin_unlock(&imp->imp_lock);
}
- lwi = LWI_TIMEOUT_INTR(0,NULL,ldlm_flock_interrupted_wait,&fwd);
+ lwi = LWI_TIMEOUT_INTR(0, NULL, ldlm_flock_interrupted_wait, &fwd);
/* Go to sleep until the lock is granted. */
rc = l_wait_event(lock->l_waitq,
((lock->l_req_mode == lock->l_granted_mode) ||
lock->l_destroyed), &lwi);
- if (rc) {
- LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
- rc);
- RETURN(rc);
- }
-
- LASSERT(!(lock->l_destroyed));
+ LDLM_DEBUG(lock, "client-side enqueue waking up: rc = %d", rc);
+ RETURN(rc);
granted:
-
- LDLM_DEBUG(lock, "client-side enqueue waking up");
+ LDLM_DEBUG(lock, "client-side enqueue granted");
ns = lock->l_resource->lr_namespace;
- l_lock(&ns->ns_lock);
+ lock_res(lock->l_resource);
+
+ /* before flock's complete ast gets here, the flock
+ * can possibly be freed by another thread
+ */
+ if (lock->l_destroyed) {
+ LDLM_DEBUG(lock, "already destroyed by another thread");
+ unlock_res(lock->l_resource);
+ RETURN(0);
+ }
- /* take data off of deadlock detection waitq. */
+ /* take lock off the deadlock detection waitq. */
+ spin_lock(&ldlm_flock_waitq_lock);
list_del_init(&lock->l_flock_waitq);
+ spin_unlock(&ldlm_flock_waitq_lock);
/* ldlm_lock_enqueue() has already placed lock on the granted list. */
list_del_init(&lock->l_res_link);
/* fcntl(F_GETLK) request */
/* The old mode was saved in getlk->fl_type so that if the mode
* in the lock changes we can decref the approprate refcount. */
- ldlm_flock_destroy(lock, getlk->fl_type, LDLM_FL_WAIT_NOREPROC);
+ ldlm_flock_destroy(lock, cfs_flock_type(getlk), LDLM_FL_WAIT_NOREPROC);
switch (lock->l_granted_mode) {
case LCK_PR:
- getlk->fl_type = F_RDLCK;
+ cfs_flock_set_type(getlk, F_RDLCK);
break;
case LCK_PW:
- getlk->fl_type = F_WRLCK;
+ cfs_flock_set_type(getlk, F_WRLCK);
break;
default:
- getlk->fl_type = F_UNLCK;
+ cfs_flock_set_type(getlk, F_UNLCK);
}
- getlk->fl_pid = lock->l_policy_data.l_flock.pid;
- getlk->fl_start = lock->l_policy_data.l_flock.start;
- getlk->fl_end = lock->l_policy_data.l_flock.end;
+ cfs_flock_set_pid(getlk, (pid_t)lock->l_policy_data.l_flock.pid);
+ cfs_flock_set_start(getlk, (loff_t)lock->l_policy_data.l_flock.start);
+ cfs_flock_set_end(getlk, (loff_t)lock->l_policy_data.l_flock.end);
} else {
+ int noreproc = LDLM_FL_WAIT_NOREPROC;
+
/* We need to reprocess the lock to do merges or splits
- * with existing locks owne by this process. */
- flags = LDLM_FL_WAIT_NOREPROC;
- ldlm_process_flock_lock(lock, &flags, 1, &err);
+ * with existing locks owned by this process. */
+ ldlm_process_flock_lock(lock, &noreproc, 1, &err, NULL);
+ if (flags == 0)
+ cfs_waitq_signal(&lock->l_waitq);
}
- l_unlock(&ns->ns_lock);
+ unlock_res(lock->l_resource);
+ RETURN(0);
+}
+EXPORT_SYMBOL(ldlm_flock_completion_ast);
+
+int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+ void *data, int flag)
+{
+ struct ldlm_namespace *ns;
+ ENTRY;
+
+ LASSERT(lock);
+ LASSERT(flag == LDLM_CB_CANCELING);
+
+ ns = lock->l_resource->lr_namespace;
+
+ /* take lock off the deadlock detection waitq. */
+ spin_lock(&ldlm_flock_waitq_lock);
+ list_del_init(&lock->l_flock_waitq);
+ spin_unlock(&ldlm_flock_waitq_lock);
RETURN(0);
}