4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2012, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ldlm/ldlm_extent.c
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
39 * This file contains implementation of EXTENT lock type
41 * EXTENT lock type is for locking a contiguous range of values, represented
42 * by 64-bit starting and ending offsets (inclusive). There are several extent
43 * lock modes, some of which may be mutually incompatible. Extent locks are
44 * considered incompatible if their modes are incompatible and their extents
45 * intersect. See the lock mode compatibility matrix in lustre_dlm.h.
48 #define DEBUG_SUBSYSTEM S_LDLM
49 #include <linux/libcfs/libcfs.h>
50 #include <lustre_dlm.h>
51 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <lustre_lib.h>
55 #include "ldlm_internal.h"
57 /* When a lock is cancelled by a client, the KMS may undergo change if this
58 * is the "highest lock". This function returns the new KMS value.
59 * Caller must hold lr_lock already.
61 * NB: A lock on [x,y] protects a KMS of up to y + 1 bytes!
63 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms)
65 struct ldlm_resource *res = lock->l_resource;
66 struct list_head *tmp;
67 struct ldlm_lock *lck;
70 /* don't let another thread in ldlm_extent_shift_kms race in
71 * just after we finish and take our lock into account in its
72 * calculation of the kms
74 ldlm_set_kms_ignore(lock);
76 list_for_each(tmp, &res->lr_granted) {
77 lck = list_entry(tmp, struct ldlm_lock, l_res_link);
79 if (ldlm_is_kms_ignore(lck))
82 if (lck->l_policy_data.l_extent.end >= old_kms)
85 /* This extent _has_ to be smaller than old_kms (checked above)
86 * so kms can only ever be smaller or the same as old_kms.
88 if (lck->l_policy_data.l_extent.end + 1 > kms)
89 kms = lck->l_policy_data.l_extent.end + 1;
91 LASSERTF(kms <= old_kms, "kms %llu old_kms %llu\n", kms, old_kms);
95 EXPORT_SYMBOL(ldlm_extent_shift_kms);
97 struct kmem_cache *ldlm_interval_slab;
99 /* interval tree, for LDLM_EXTENT. */
100 static void ldlm_interval_attach(struct ldlm_interval *n, struct ldlm_lock *l)
102 LASSERT(!l->l_tree_node);
103 LASSERT(l->l_resource->lr_type == LDLM_EXTENT);
105 list_add_tail(&l->l_sl_policy, &n->li_group);
109 struct ldlm_interval *ldlm_interval_alloc(struct ldlm_lock *lock)
111 struct ldlm_interval *node;
113 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
114 node = kmem_cache_zalloc(ldlm_interval_slab, GFP_NOFS);
118 INIT_LIST_HEAD(&node->li_group);
119 ldlm_interval_attach(node, lock);
123 void ldlm_interval_free(struct ldlm_interval *node)
126 LASSERT(list_empty(&node->li_group));
127 LASSERT(!interval_is_intree(&node->li_node));
128 kmem_cache_free(ldlm_interval_slab, node);
132 struct ldlm_interval *ldlm_interval_detach(struct ldlm_lock *l)
134 struct ldlm_interval *n = l->l_tree_node;
139 LASSERT(!list_empty(&n->li_group));
140 l->l_tree_node = NULL;
141 list_del_init(&l->l_sl_policy);
143 return list_empty(&n->li_group) ? n : NULL;
146 static inline int lock_mode_to_index(enum ldlm_mode mode)
151 LASSERT(is_power_of_2(mode));
152 for (index = -1; mode; index++)
154 LASSERT(index < LCK_MODE_NUM);
158 /** Add newly granted lock into interval tree for the resource. */
159 void ldlm_extent_add_lock(struct ldlm_resource *res,
160 struct ldlm_lock *lock)
162 struct interval_node *found, **root;
163 struct ldlm_interval *node;
164 struct ldlm_extent *extent;
167 LASSERT(lock->l_granted_mode == lock->l_req_mode);
169 node = lock->l_tree_node;
171 LASSERT(!interval_is_intree(&node->li_node));
173 idx = lock_mode_to_index(lock->l_granted_mode);
174 LASSERT(lock->l_granted_mode == 1 << idx);
175 LASSERT(lock->l_granted_mode == res->lr_itree[idx].lit_mode);
177 /* node extent initialize */
178 extent = &lock->l_policy_data.l_extent;
179 rc = interval_set(&node->li_node, extent->start, extent->end);
182 root = &res->lr_itree[idx].lit_root;
183 found = interval_insert(&node->li_node, root);
184 if (found) { /* The policy group found. */
185 struct ldlm_interval *tmp;
187 tmp = ldlm_interval_detach(lock);
188 ldlm_interval_free(tmp);
189 ldlm_interval_attach(to_ldlm_interval(found), lock);
191 res->lr_itree[idx].lit_size++;
193 /* even though we use interval tree to manage the extent lock, we also
194 * add the locks into grant list, for debug purpose, ..
196 ldlm_resource_add_lock(res, &res->lr_granted, lock);
198 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_GRANT_CHECK)) {
199 struct ldlm_lock *lck;
201 list_for_each_entry_reverse(lck, &res->lr_granted,
205 if (lockmode_compat(lck->l_granted_mode,
206 lock->l_granted_mode))
208 if (ldlm_extent_overlap(&lck->l_req_extent,
209 &lock->l_req_extent)) {
210 CDEBUG(D_ERROR, "granting conflicting lock %p %p\n",
212 ldlm_resource_dump(D_ERROR, res);
219 /** Remove cancelled lock from resource interval tree. */
220 void ldlm_extent_unlink_lock(struct ldlm_lock *lock)
222 struct ldlm_resource *res = lock->l_resource;
223 struct ldlm_interval *node = lock->l_tree_node;
224 struct ldlm_interval_tree *tree;
227 if (!node || !interval_is_intree(&node->li_node)) /* duplicate unlink */
230 idx = lock_mode_to_index(lock->l_granted_mode);
231 LASSERT(lock->l_granted_mode == 1 << idx);
232 tree = &res->lr_itree[idx];
234 LASSERT(tree->lit_root); /* assure the tree is not null */
237 node = ldlm_interval_detach(lock);
239 interval_erase(&node->li_node, &tree->lit_root);
240 ldlm_interval_free(node);
244 void ldlm_extent_policy_wire_to_local(const union ldlm_wire_policy_data *wpolicy,
245 union ldlm_policy_data *lpolicy)
247 lpolicy->l_extent.start = wpolicy->l_extent.start;
248 lpolicy->l_extent.end = wpolicy->l_extent.end;
249 lpolicy->l_extent.gid = wpolicy->l_extent.gid;
252 void ldlm_extent_policy_local_to_wire(const union ldlm_policy_data *lpolicy,
253 union ldlm_wire_policy_data *wpolicy)
255 memset(wpolicy, 0, sizeof(*wpolicy));
256 wpolicy->l_extent.start = lpolicy->l_extent.start;
257 wpolicy->l_extent.end = lpolicy->l_extent.end;
258 wpolicy->l_extent.gid = lpolicy->l_extent.gid;