blob: 85cd2adcc82407e4c793bad9ceab5c1a9734e9b8 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * alloc.c
5 *
6 * Extent allocs and frees
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
Mark Fasheh60b11392007-02-16 11:46:50 -080030#include <linux/swap.h>
Jan Karaa90714c2008-10-09 19:38:40 +020031#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080032
33#define MLOG_MASK_PREFIX ML_DISK_ALLOC
34#include <cluster/masklog.h>
35
36#include "ocfs2.h"
37
38#include "alloc.h"
Mark Fasheh60b11392007-02-16 11:46:50 -080039#include "aops.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070040#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080041#include "dlmglue.h"
42#include "extent_map.h"
43#include "inode.h"
44#include "journal.h"
45#include "localalloc.h"
46#include "suballoc.h"
47#include "sysfile.h"
48#include "file.h"
49#include "super.h"
50#include "uptodate.h"
Joel Becker2a50a742008-12-09 14:24:33 -080051#include "xattr.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080052
53#include "buffer_head_io.h"
54
Tao Mae7d4cb62008-08-18 17:38:44 +080055
Joel Becker1625f8a2008-08-21 17:11:10 -070056/*
57 * Operations for a specific extent tree type.
58 *
59 * To implement an on-disk btree (extent tree) type in ocfs2, add
60 * an ocfs2_extent_tree_operations structure and the matching
Joel Becker8d6220d2008-08-22 12:46:09 -070061 * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it
Joel Becker1625f8a2008-08-21 17:11:10 -070062 * for the allocation portion of the extent tree.
63 */
Tao Mae7d4cb62008-08-18 17:38:44 +080064struct ocfs2_extent_tree_operations {
Joel Becker1625f8a2008-08-21 17:11:10 -070065 /*
66 * last_eb_blk is the block number of the right most leaf extent
67 * block. Most on-disk structures containing an extent tree store
68 * this value for fast access. The ->eo_set_last_eb_blk() and
69 * ->eo_get_last_eb_blk() operations access this value. They are
70 * both required.
71 */
Joel Becker35dc0aa2008-08-20 16:25:06 -070072 void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
73 u64 blkno);
74 u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -070075
76 /*
77 * The on-disk structure usually keeps track of how many total
78 * clusters are stored in this extent tree. This function updates
79 * that value. new_clusters is the delta, and must be
80 * added to the total. Required.
81 */
Joel Becker6136ca52009-02-12 19:32:43 -080082 void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -070083 u32 new_clusters);
Joel Becker1625f8a2008-08-21 17:11:10 -070084
85 /*
Joel Becker92ba4702009-02-13 03:18:34 -080086 * If this extent tree is supported by an extent map, insert
87 * a record into the map.
88 */
89 void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
90 struct ocfs2_extent_rec *rec);
91
92 /*
Joel Becker4c911ee2009-02-13 02:50:12 -080093 * If this extent tree is supported by an extent map, truncate the
94 * map to clusters,
95 */
96 void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
97 u32 clusters);
98
99 /*
Joel Becker1625f8a2008-08-21 17:11:10 -0700100 * If ->eo_insert_check() exists, it is called before rec is
101 * inserted into the extent tree. It is optional.
102 */
Joel Becker6136ca52009-02-12 19:32:43 -0800103 int (*eo_insert_check)(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700104 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800105 int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
Joel Becker0ce10102008-08-20 17:19:50 -0700106
Joel Becker1625f8a2008-08-21 17:11:10 -0700107 /*
108 * --------------------------------------------------------------
109 * The remaining are internal to ocfs2_extent_tree and don't have
110 * accessor functions
111 */
112
113 /*
114 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
115 * It is required.
116 */
Joel Becker0ce10102008-08-20 17:19:50 -0700117 void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -0700118
119 /*
120 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
121 * it exists. If it does not, et->et_max_leaf_clusters is set
122 * to 0 (unlimited). Optional.
123 */
Joel Becker6136ca52009-02-12 19:32:43 -0800124 void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800125};
126
Joel Beckerf99b9b72008-08-20 19:36:33 -0700127
128/*
129 * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
130 * in the methods.
131 */
132static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
133static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
134 u64 blkno);
Joel Becker6136ca52009-02-12 19:32:43 -0800135static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700136 u32 clusters);
Joel Becker92ba4702009-02-13 03:18:34 -0800137static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
138 struct ocfs2_extent_rec *rec);
Joel Becker4c911ee2009-02-13 02:50:12 -0800139static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
140 u32 clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800141static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700142 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800143static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700144static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
145static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
146 .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk,
147 .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk,
148 .eo_update_clusters = ocfs2_dinode_update_clusters,
Joel Becker92ba4702009-02-13 03:18:34 -0800149 .eo_extent_map_insert = ocfs2_dinode_extent_map_insert,
Joel Becker4c911ee2009-02-13 02:50:12 -0800150 .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700151 .eo_insert_check = ocfs2_dinode_insert_check,
152 .eo_sanity_check = ocfs2_dinode_sanity_check,
153 .eo_fill_root_el = ocfs2_dinode_fill_root_el,
Tao Mae7d4cb62008-08-18 17:38:44 +0800154};
155
156static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
157 u64 blkno)
158{
Joel Beckerea5efa12008-08-20 16:57:27 -0700159 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800160
Joel Beckerf99b9b72008-08-20 19:36:33 -0700161 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800162 di->i_last_eb_blk = cpu_to_le64(blkno);
163}
164
165static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
166{
Joel Beckerea5efa12008-08-20 16:57:27 -0700167 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800168
Joel Beckerf99b9b72008-08-20 19:36:33 -0700169 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800170 return le64_to_cpu(di->i_last_eb_blk);
171}
172
Joel Becker6136ca52009-02-12 19:32:43 -0800173static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +0800174 u32 clusters)
175{
Joel Becker6136ca52009-02-12 19:32:43 -0800176 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
Joel Beckerea5efa12008-08-20 16:57:27 -0700177 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800178
179 le32_add_cpu(&di->i_clusters, clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800180 spin_lock(&oi->ip_lock);
181 oi->ip_clusters = le32_to_cpu(di->i_clusters);
182 spin_unlock(&oi->ip_lock);
Tao Mae7d4cb62008-08-18 17:38:44 +0800183}
184
Joel Becker92ba4702009-02-13 03:18:34 -0800185static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
186 struct ocfs2_extent_rec *rec)
187{
188 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
189
190 ocfs2_extent_map_insert_rec(inode, rec);
191}
192
Joel Becker4c911ee2009-02-13 02:50:12 -0800193static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
194 u32 clusters)
195{
196 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
197
198 ocfs2_extent_map_trunc(inode, clusters);
199}
200
Joel Becker6136ca52009-02-12 19:32:43 -0800201static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700202 struct ocfs2_extent_rec *rec)
203{
Joel Becker6136ca52009-02-12 19:32:43 -0800204 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
205 struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
Joel Becker1e61ee72008-08-20 18:32:45 -0700206
Joel Becker6136ca52009-02-12 19:32:43 -0800207 BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
Joel Becker1e61ee72008-08-20 18:32:45 -0700208 mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
Joel Becker6136ca52009-02-12 19:32:43 -0800209 (oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
Joel Becker1e61ee72008-08-20 18:32:45 -0700210 "Device %s, asking for sparse allocation: inode %llu, "
211 "cpos %u, clusters %u\n",
212 osb->dev_str,
Joel Becker6136ca52009-02-12 19:32:43 -0800213 (unsigned long long)oi->ip_blkno,
214 rec->e_cpos, oi->ip_clusters);
Joel Becker1e61ee72008-08-20 18:32:45 -0700215
216 return 0;
217}
218
Joel Becker6136ca52009-02-12 19:32:43 -0800219static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800220{
Joel Becker10995aa2008-11-13 14:49:12 -0800221 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800222
Joel Beckerf99b9b72008-08-20 19:36:33 -0700223 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Joel Becker10995aa2008-11-13 14:49:12 -0800224 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
Tao Mae7d4cb62008-08-18 17:38:44 +0800225
Joel Becker10995aa2008-11-13 14:49:12 -0800226 return 0;
Tao Mae7d4cb62008-08-18 17:38:44 +0800227}
228
Joel Beckerf99b9b72008-08-20 19:36:33 -0700229static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
230{
231 struct ocfs2_dinode *di = et->et_object;
232
233 et->et_root_el = &di->id2.i_list;
234}
235
Tao Mae7d4cb62008-08-18 17:38:44 +0800236
Joel Becker0ce10102008-08-20 17:19:50 -0700237static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
238{
Joel Becker2a50a742008-12-09 14:24:33 -0800239 struct ocfs2_xattr_value_buf *vb = et->et_object;
Joel Becker0ce10102008-08-20 17:19:50 -0700240
Joel Becker2a50a742008-12-09 14:24:33 -0800241 et->et_root_el = &vb->vb_xv->xr_list;
Joel Becker0ce10102008-08-20 17:19:50 -0700242}
243
Tao Maf56654c2008-08-18 17:38:48 +0800244static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
245 u64 blkno)
246{
Joel Becker2a50a742008-12-09 14:24:33 -0800247 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800248
Joel Becker2a50a742008-12-09 14:24:33 -0800249 vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
Tao Maf56654c2008-08-18 17:38:48 +0800250}
251
252static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
253{
Joel Becker2a50a742008-12-09 14:24:33 -0800254 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800255
Joel Becker2a50a742008-12-09 14:24:33 -0800256 return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
Tao Maf56654c2008-08-18 17:38:48 +0800257}
258
Joel Becker6136ca52009-02-12 19:32:43 -0800259static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
Tao Maf56654c2008-08-18 17:38:48 +0800260 u32 clusters)
261{
Joel Becker2a50a742008-12-09 14:24:33 -0800262 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800263
Joel Becker2a50a742008-12-09 14:24:33 -0800264 le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
Tao Maf56654c2008-08-18 17:38:48 +0800265}
266
Joel Becker1a09f552008-08-20 17:44:24 -0700267static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700268 .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk,
269 .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk,
270 .eo_update_clusters = ocfs2_xattr_value_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700271 .eo_fill_root_el = ocfs2_xattr_value_fill_root_el,
Tao Maf56654c2008-08-18 17:38:48 +0800272};
273
Joel Becker0ce10102008-08-20 17:19:50 -0700274static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
275{
276 struct ocfs2_xattr_block *xb = et->et_object;
277
278 et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
279}
280
Joel Becker6136ca52009-02-12 19:32:43 -0800281static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
Joel Becker943cced2008-08-20 17:31:10 -0700282{
Joel Becker6136ca52009-02-12 19:32:43 -0800283 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Joel Becker943cced2008-08-20 17:31:10 -0700284 et->et_max_leaf_clusters =
Joel Becker6136ca52009-02-12 19:32:43 -0800285 ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
Joel Becker943cced2008-08-20 17:31:10 -0700286}
287
Tao Maba492612008-08-18 17:38:49 +0800288static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
289 u64 blkno)
290{
Joel Beckerea5efa12008-08-20 16:57:27 -0700291 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800292 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
293
294 xt->xt_last_eb_blk = cpu_to_le64(blkno);
295}
296
297static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
298{
Joel Beckerea5efa12008-08-20 16:57:27 -0700299 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800300 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
301
302 return le64_to_cpu(xt->xt_last_eb_blk);
303}
304
Joel Becker6136ca52009-02-12 19:32:43 -0800305static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
Tao Maba492612008-08-18 17:38:49 +0800306 u32 clusters)
307{
Joel Beckerea5efa12008-08-20 16:57:27 -0700308 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800309
310 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
311}
312
Tao Maba492612008-08-18 17:38:49 +0800313static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700314 .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk,
315 .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk,
316 .eo_update_clusters = ocfs2_xattr_tree_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700317 .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el,
Joel Becker943cced2008-08-20 17:31:10 -0700318 .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
Tao Maba492612008-08-18 17:38:49 +0800319};
320
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800321static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
322 u64 blkno)
323{
324 struct ocfs2_dx_root_block *dx_root = et->et_object;
325
326 dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
327}
328
329static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
330{
331 struct ocfs2_dx_root_block *dx_root = et->et_object;
332
333 return le64_to_cpu(dx_root->dr_last_eb_blk);
334}
335
Joel Becker6136ca52009-02-12 19:32:43 -0800336static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800337 u32 clusters)
338{
339 struct ocfs2_dx_root_block *dx_root = et->et_object;
340
341 le32_add_cpu(&dx_root->dr_clusters, clusters);
342}
343
Joel Becker6136ca52009-02-12 19:32:43 -0800344static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800345{
346 struct ocfs2_dx_root_block *dx_root = et->et_object;
347
348 BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));
349
350 return 0;
351}
352
353static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
354{
355 struct ocfs2_dx_root_block *dx_root = et->et_object;
356
357 et->et_root_el = &dx_root->dr_list;
358}
359
360static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
361 .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk,
362 .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk,
363 .eo_update_clusters = ocfs2_dx_root_update_clusters,
364 .eo_sanity_check = ocfs2_dx_root_sanity_check,
365 .eo_fill_root_el = ocfs2_dx_root_fill_root_el,
366};
367
Joel Becker8d6220d2008-08-22 12:46:09 -0700368static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
369 struct inode *inode,
370 struct buffer_head *bh,
Joel Becker13723d02008-10-17 19:25:01 -0700371 ocfs2_journal_access_func access,
Joel Becker8d6220d2008-08-22 12:46:09 -0700372 void *obj,
373 struct ocfs2_extent_tree_operations *ops)
Tao Mae7d4cb62008-08-18 17:38:44 +0800374{
Joel Becker1a09f552008-08-20 17:44:24 -0700375 et->et_ops = ops;
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700376 et->et_root_bh = bh;
Joel Beckerd9a0a1f2009-02-12 17:32:34 -0800377 et->et_ci = INODE_CACHE(inode);
Joel Becker13723d02008-10-17 19:25:01 -0700378 et->et_root_journal_access = access;
Joel Beckerea5efa12008-08-20 16:57:27 -0700379 if (!obj)
380 obj = (void *)bh->b_data;
381 et->et_object = obj;
Tao Mae7d4cb62008-08-18 17:38:44 +0800382
Joel Becker0ce10102008-08-20 17:19:50 -0700383 et->et_ops->eo_fill_root_el(et);
Joel Becker943cced2008-08-20 17:31:10 -0700384 if (!et->et_ops->eo_fill_max_leaf_clusters)
385 et->et_max_leaf_clusters = 0;
386 else
Joel Becker6136ca52009-02-12 19:32:43 -0800387 et->et_ops->eo_fill_max_leaf_clusters(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800388}
389
Joel Becker8d6220d2008-08-22 12:46:09 -0700390void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
391 struct inode *inode,
392 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700393{
Joel Becker13723d02008-10-17 19:25:01 -0700394 __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_di,
395 NULL, &ocfs2_dinode_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700396}
397
Joel Becker8d6220d2008-08-22 12:46:09 -0700398void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700399 struct inode *inode,
Joel Becker8d6220d2008-08-22 12:46:09 -0700400 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700401{
Joel Becker13723d02008-10-17 19:25:01 -0700402 __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_xb,
403 NULL, &ocfs2_xattr_tree_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700404}
405
Joel Becker8d6220d2008-08-22 12:46:09 -0700406void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
407 struct inode *inode,
Joel Becker2a50a742008-12-09 14:24:33 -0800408 struct ocfs2_xattr_value_buf *vb)
Tao Mae7d4cb62008-08-18 17:38:44 +0800409{
Joel Becker2a50a742008-12-09 14:24:33 -0800410 __ocfs2_init_extent_tree(et, inode, vb->vb_bh, vb->vb_access, vb,
Joel Becker8d6220d2008-08-22 12:46:09 -0700411 &ocfs2_xattr_value_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800412}
413
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800414void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
415 struct inode *inode,
416 struct buffer_head *bh)
417{
418 __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_dr,
419 NULL, &ocfs2_dx_root_et_ops);
420}
421
Joel Becker35dc0aa2008-08-20 16:25:06 -0700422static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
423 u64 new_last_eb_blk)
Tao Mae7d4cb62008-08-18 17:38:44 +0800424{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700425 et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
Tao Mae7d4cb62008-08-18 17:38:44 +0800426}
427
Joel Becker35dc0aa2008-08-20 16:25:06 -0700428static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800429{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700430 return et->et_ops->eo_get_last_eb_blk(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800431}
432
Joel Becker6136ca52009-02-12 19:32:43 -0800433static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -0700434 u32 clusters)
Tao Mae7d4cb62008-08-18 17:38:44 +0800435{
Joel Becker6136ca52009-02-12 19:32:43 -0800436 et->et_ops->eo_update_clusters(et, clusters);
Joel Becker35dc0aa2008-08-20 16:25:06 -0700437}
438
Joel Becker92ba4702009-02-13 03:18:34 -0800439static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
440 struct ocfs2_extent_rec *rec)
441{
442 if (et->et_ops->eo_extent_map_insert)
443 et->et_ops->eo_extent_map_insert(et, rec);
444}
445
Joel Becker4c911ee2009-02-13 02:50:12 -0800446static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
447 u32 clusters)
448{
449 if (et->et_ops->eo_extent_map_truncate)
450 et->et_ops->eo_extent_map_truncate(et, clusters);
451}
452
Joel Becker13723d02008-10-17 19:25:01 -0700453static inline int ocfs2_et_root_journal_access(handle_t *handle,
Joel Becker13723d02008-10-17 19:25:01 -0700454 struct ocfs2_extent_tree *et,
455 int type)
456{
Joel Beckerd9a0a1f2009-02-12 17:32:34 -0800457 return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700458 type);
459}
460
Joel Becker6136ca52009-02-12 19:32:43 -0800461static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700462 struct ocfs2_extent_rec *rec)
463{
464 int ret = 0;
465
466 if (et->et_ops->eo_insert_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800467 ret = et->et_ops->eo_insert_check(et, rec);
Joel Becker1e61ee72008-08-20 18:32:45 -0700468 return ret;
469}
470
Joel Becker6136ca52009-02-12 19:32:43 -0800471static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
Joel Becker35dc0aa2008-08-20 16:25:06 -0700472{
Joel Becker1e61ee72008-08-20 18:32:45 -0700473 int ret = 0;
474
475 if (et->et_ops->eo_sanity_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800476 ret = et->et_ops->eo_sanity_check(et);
Joel Becker1e61ee72008-08-20 18:32:45 -0700477 return ret;
Tao Mae7d4cb62008-08-18 17:38:44 +0800478}
479
Mark Fashehccd979b2005-12-15 14:31:24 -0800480static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
Mark Fasheh59a5e412007-06-22 15:52:36 -0700481static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
482 struct ocfs2_extent_block *eb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800483
Mark Fashehdcd05382007-01-16 11:32:23 -0800484/*
485 * Structures which describe a path through a btree, and functions to
486 * manipulate them.
487 *
488 * The idea here is to be as generic as possible with the tree
489 * manipulation code.
490 */
491struct ocfs2_path_item {
492 struct buffer_head *bh;
493 struct ocfs2_extent_list *el;
494};
495
496#define OCFS2_MAX_PATH_DEPTH 5
497
498struct ocfs2_path {
Joel Becker13723d02008-10-17 19:25:01 -0700499 int p_tree_depth;
500 ocfs2_journal_access_func p_root_access;
501 struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH];
Mark Fashehdcd05382007-01-16 11:32:23 -0800502};
503
504#define path_root_bh(_path) ((_path)->p_node[0].bh)
505#define path_root_el(_path) ((_path)->p_node[0].el)
Joel Becker13723d02008-10-17 19:25:01 -0700506#define path_root_access(_path)((_path)->p_root_access)
Mark Fashehdcd05382007-01-16 11:32:23 -0800507#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
508#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
509#define path_num_items(_path) ((_path)->p_tree_depth + 1)
510
Joel Beckerfacdb772009-02-12 18:08:48 -0800511static int ocfs2_find_path(struct ocfs2_caching_info *ci,
512 struct ocfs2_path *path, u32 cpos);
Joel Beckerd401dc12009-02-13 02:24:10 -0800513static void ocfs2_adjust_rightmost_records(handle_t *handle,
514 struct ocfs2_extent_tree *et,
Tao Ma6b791bc2009-06-12 14:18:36 +0800515 struct ocfs2_path *path,
516 struct ocfs2_extent_rec *insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -0800517/*
518 * Reset the actual path elements so that we can re-use the structure
519 * to build another path. Generally, this involves freeing the buffer
520 * heads.
521 */
522static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
523{
524 int i, start = 0, depth = 0;
525 struct ocfs2_path_item *node;
526
527 if (keep_root)
528 start = 1;
529
530 for(i = start; i < path_num_items(path); i++) {
531 node = &path->p_node[i];
532
533 brelse(node->bh);
534 node->bh = NULL;
535 node->el = NULL;
536 }
537
538 /*
539 * Tree depth may change during truncate, or insert. If we're
540 * keeping the root extent list, then make sure that our path
541 * structure reflects the proper depth.
542 */
543 if (keep_root)
544 depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
Joel Becker13723d02008-10-17 19:25:01 -0700545 else
546 path_root_access(path) = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -0800547
548 path->p_tree_depth = depth;
549}
550
551static void ocfs2_free_path(struct ocfs2_path *path)
552{
553 if (path) {
554 ocfs2_reinit_path(path, 0);
555 kfree(path);
556 }
557}
558
559/*
Mark Fasheh328d5752007-06-18 10:48:04 -0700560 * All the elements of src into dest. After this call, src could be freed
561 * without affecting dest.
562 *
563 * Both paths should have the same root. Any non-root elements of dest
564 * will be freed.
565 */
566static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
567{
568 int i;
569
570 BUG_ON(path_root_bh(dest) != path_root_bh(src));
571 BUG_ON(path_root_el(dest) != path_root_el(src));
Joel Becker13723d02008-10-17 19:25:01 -0700572 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fasheh328d5752007-06-18 10:48:04 -0700573
574 ocfs2_reinit_path(dest, 1);
575
576 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
577 dest->p_node[i].bh = src->p_node[i].bh;
578 dest->p_node[i].el = src->p_node[i].el;
579
580 if (dest->p_node[i].bh)
581 get_bh(dest->p_node[i].bh);
582 }
583}
584
585/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800586 * Make the *dest path the same as src and re-initialize src path to
587 * have a root only.
588 */
589static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
590{
591 int i;
592
593 BUG_ON(path_root_bh(dest) != path_root_bh(src));
Joel Becker13723d02008-10-17 19:25:01 -0700594 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fashehdcd05382007-01-16 11:32:23 -0800595
596 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
597 brelse(dest->p_node[i].bh);
598
599 dest->p_node[i].bh = src->p_node[i].bh;
600 dest->p_node[i].el = src->p_node[i].el;
601
602 src->p_node[i].bh = NULL;
603 src->p_node[i].el = NULL;
604 }
605}
606
607/*
608 * Insert an extent block at given index.
609 *
610 * This will not take an additional reference on eb_bh.
611 */
612static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
613 struct buffer_head *eb_bh)
614{
615 struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
616
617 /*
618 * Right now, no root bh is an extent block, so this helps
619 * catch code errors with dinode trees. The assertion can be
620 * safely removed if we ever need to insert extent block
621 * structures at the root.
622 */
623 BUG_ON(index == 0);
624
625 path->p_node[index].bh = eb_bh;
626 path->p_node[index].el = &eb->h_list;
627}
628
629static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700630 struct ocfs2_extent_list *root_el,
631 ocfs2_journal_access_func access)
Mark Fashehdcd05382007-01-16 11:32:23 -0800632{
633 struct ocfs2_path *path;
634
635 BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
636
637 path = kzalloc(sizeof(*path), GFP_NOFS);
638 if (path) {
639 path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
640 get_bh(root_bh);
641 path_root_bh(path) = root_bh;
642 path_root_el(path) = root_el;
Joel Becker13723d02008-10-17 19:25:01 -0700643 path_root_access(path) = access;
Mark Fashehdcd05382007-01-16 11:32:23 -0800644 }
645
646 return path;
647}
648
Joel Beckerffdd7a52008-10-17 22:32:01 -0700649static struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
650{
Joel Becker13723d02008-10-17 19:25:01 -0700651 return ocfs2_new_path(path_root_bh(path), path_root_el(path),
652 path_root_access(path));
Joel Beckerffdd7a52008-10-17 22:32:01 -0700653}
654
655static struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
656{
Joel Becker13723d02008-10-17 19:25:01 -0700657 return ocfs2_new_path(et->et_root_bh, et->et_root_el,
658 et->et_root_journal_access);
659}
660
661/*
662 * Journal the buffer at depth idx. All idx>0 are extent_blocks,
663 * otherwise it's the root_access function.
664 *
665 * I don't like the way this function's name looks next to
666 * ocfs2_journal_access_path(), but I don't have a better one.
667 */
668static int ocfs2_path_bh_journal_access(handle_t *handle,
Joel Becker0cf2f762009-02-12 16:41:25 -0800669 struct ocfs2_caching_info *ci,
Joel Becker13723d02008-10-17 19:25:01 -0700670 struct ocfs2_path *path,
671 int idx)
672{
673 ocfs2_journal_access_func access = path_root_access(path);
674
675 if (!access)
676 access = ocfs2_journal_access;
677
678 if (idx)
679 access = ocfs2_journal_access_eb;
680
Joel Becker0cf2f762009-02-12 16:41:25 -0800681 return access(handle, ci, path->p_node[idx].bh,
Joel Becker13723d02008-10-17 19:25:01 -0700682 OCFS2_JOURNAL_ACCESS_WRITE);
Joel Beckerffdd7a52008-10-17 22:32:01 -0700683}
684
Mark Fashehdcd05382007-01-16 11:32:23 -0800685/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800686 * Convenience function to journal all components in a path.
687 */
Joel Becker0cf2f762009-02-12 16:41:25 -0800688static int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
689 handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -0800690 struct ocfs2_path *path)
691{
692 int i, ret = 0;
693
694 if (!path)
695 goto out;
696
697 for(i = 0; i < path_num_items(path); i++) {
Joel Becker0cf2f762009-02-12 16:41:25 -0800698 ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -0800699 if (ret < 0) {
700 mlog_errno(ret);
701 goto out;
702 }
703 }
704
705out:
706 return ret;
707}
708
Mark Fasheh328d5752007-06-18 10:48:04 -0700709/*
710 * Return the index of the extent record which contains cluster #v_cluster.
711 * -1 is returned if it was not found.
712 *
713 * Should work fine on interior and exterior nodes.
714 */
715int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
716{
717 int ret = -1;
718 int i;
719 struct ocfs2_extent_rec *rec;
720 u32 rec_end, rec_start, clusters;
721
722 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
723 rec = &el->l_recs[i];
724
725 rec_start = le32_to_cpu(rec->e_cpos);
726 clusters = ocfs2_rec_clusters(el, rec);
727
728 rec_end = rec_start + clusters;
729
730 if (v_cluster >= rec_start && v_cluster < rec_end) {
731 ret = i;
732 break;
733 }
734 }
735
736 return ret;
737}
738
Mark Fashehdcd05382007-01-16 11:32:23 -0800739enum ocfs2_contig_type {
740 CONTIG_NONE = 0,
741 CONTIG_LEFT,
Mark Fasheh328d5752007-06-18 10:48:04 -0700742 CONTIG_RIGHT,
743 CONTIG_LEFTRIGHT,
Mark Fashehdcd05382007-01-16 11:32:23 -0800744};
745
Mark Fashehe48edee2007-03-07 16:46:57 -0800746
747/*
748 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
749 * ocfs2_extent_contig only work properly against leaf nodes!
750 */
Mark Fashehdcd05382007-01-16 11:32:23 -0800751static int ocfs2_block_extent_contig(struct super_block *sb,
752 struct ocfs2_extent_rec *ext,
753 u64 blkno)
Mark Fashehccd979b2005-12-15 14:31:24 -0800754{
Mark Fashehe48edee2007-03-07 16:46:57 -0800755 u64 blk_end = le64_to_cpu(ext->e_blkno);
756
757 blk_end += ocfs2_clusters_to_blocks(sb,
758 le16_to_cpu(ext->e_leaf_clusters));
759
760 return blkno == blk_end;
Mark Fashehccd979b2005-12-15 14:31:24 -0800761}
762
Mark Fashehdcd05382007-01-16 11:32:23 -0800763static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
764 struct ocfs2_extent_rec *right)
765{
Mark Fashehe48edee2007-03-07 16:46:57 -0800766 u32 left_range;
767
768 left_range = le32_to_cpu(left->e_cpos) +
769 le16_to_cpu(left->e_leaf_clusters);
770
771 return (left_range == le32_to_cpu(right->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -0800772}
773
774static enum ocfs2_contig_type
Joel Beckerb4a17652009-02-13 03:07:09 -0800775 ocfs2_extent_contig(struct super_block *sb,
Mark Fashehdcd05382007-01-16 11:32:23 -0800776 struct ocfs2_extent_rec *ext,
777 struct ocfs2_extent_rec *insert_rec)
778{
779 u64 blkno = le64_to_cpu(insert_rec->e_blkno);
780
Mark Fasheh328d5752007-06-18 10:48:04 -0700781 /*
782 * Refuse to coalesce extent records with different flag
783 * fields - we don't want to mix unwritten extents with user
784 * data.
785 */
786 if (ext->e_flags != insert_rec->e_flags)
787 return CONTIG_NONE;
788
Mark Fashehdcd05382007-01-16 11:32:23 -0800789 if (ocfs2_extents_adjacent(ext, insert_rec) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800790 ocfs2_block_extent_contig(sb, ext, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800791 return CONTIG_RIGHT;
792
793 blkno = le64_to_cpu(ext->e_blkno);
794 if (ocfs2_extents_adjacent(insert_rec, ext) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800795 ocfs2_block_extent_contig(sb, insert_rec, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800796 return CONTIG_LEFT;
797
798 return CONTIG_NONE;
799}
800
801/*
802 * NOTE: We can have pretty much any combination of contiguousness and
803 * appending.
804 *
805 * The usefulness of APPEND_TAIL is more in that it lets us know that
806 * we'll have to update the path to that leaf.
807 */
808enum ocfs2_append_type {
809 APPEND_NONE = 0,
810 APPEND_TAIL,
811};
812
Mark Fasheh328d5752007-06-18 10:48:04 -0700813enum ocfs2_split_type {
814 SPLIT_NONE = 0,
815 SPLIT_LEFT,
816 SPLIT_RIGHT,
817};
818
Mark Fashehdcd05382007-01-16 11:32:23 -0800819struct ocfs2_insert_type {
Mark Fasheh328d5752007-06-18 10:48:04 -0700820 enum ocfs2_split_type ins_split;
Mark Fashehdcd05382007-01-16 11:32:23 -0800821 enum ocfs2_append_type ins_appending;
822 enum ocfs2_contig_type ins_contig;
823 int ins_contig_index;
Mark Fashehdcd05382007-01-16 11:32:23 -0800824 int ins_tree_depth;
825};
826
Mark Fasheh328d5752007-06-18 10:48:04 -0700827struct ocfs2_merge_ctxt {
828 enum ocfs2_contig_type c_contig_type;
829 int c_has_empty_extent;
830 int c_split_covers_rec;
Mark Fasheh328d5752007-06-18 10:48:04 -0700831};
832
Joel Becker5e965812008-11-13 14:49:16 -0800833static int ocfs2_validate_extent_block(struct super_block *sb,
834 struct buffer_head *bh)
835{
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700836 int rc;
Joel Becker5e965812008-11-13 14:49:16 -0800837 struct ocfs2_extent_block *eb =
838 (struct ocfs2_extent_block *)bh->b_data;
839
Joel Becker970e4932008-11-13 14:49:19 -0800840 mlog(0, "Validating extent block %llu\n",
841 (unsigned long long)bh->b_blocknr);
842
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700843 BUG_ON(!buffer_uptodate(bh));
844
845 /*
846 * If the ecc fails, we return the error but otherwise
847 * leave the filesystem running. We know any error is
848 * local to this block.
849 */
850 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
Joel Becker13723d02008-10-17 19:25:01 -0700851 if (rc) {
852 mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
853 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700854 return rc;
Joel Becker13723d02008-10-17 19:25:01 -0700855 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700856
857 /*
858 * Errors after here are fatal.
859 */
860
Joel Becker5e965812008-11-13 14:49:16 -0800861 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
862 ocfs2_error(sb,
863 "Extent block #%llu has bad signature %.*s",
864 (unsigned long long)bh->b_blocknr, 7,
865 eb->h_signature);
866 return -EINVAL;
867 }
868
869 if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
870 ocfs2_error(sb,
871 "Extent block #%llu has an invalid h_blkno "
872 "of %llu",
873 (unsigned long long)bh->b_blocknr,
874 (unsigned long long)le64_to_cpu(eb->h_blkno));
875 return -EINVAL;
876 }
877
878 if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
879 ocfs2_error(sb,
880 "Extent block #%llu has an invalid "
881 "h_fs_generation of #%u",
882 (unsigned long long)bh->b_blocknr,
883 le32_to_cpu(eb->h_fs_generation));
884 return -EINVAL;
885 }
886
887 return 0;
888}
889
Joel Becker3d03a302009-02-12 17:49:26 -0800890int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
Joel Becker5e965812008-11-13 14:49:16 -0800891 struct buffer_head **bh)
892{
893 int rc;
894 struct buffer_head *tmp = *bh;
895
Joel Becker3d03a302009-02-12 17:49:26 -0800896 rc = ocfs2_read_block(ci, eb_blkno, &tmp,
Joel Becker970e4932008-11-13 14:49:19 -0800897 ocfs2_validate_extent_block);
Joel Becker5e965812008-11-13 14:49:16 -0800898
899 /* If ocfs2_read_block() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -0800900 if (!rc && !*bh)
Joel Becker5e965812008-11-13 14:49:16 -0800901 *bh = tmp;
902
Joel Becker5e965812008-11-13 14:49:16 -0800903 return rc;
904}
905
906
Mark Fashehccd979b2005-12-15 14:31:24 -0800907/*
908 * How many free extents have we got before we need more meta data?
909 */
910int ocfs2_num_free_extents(struct ocfs2_super *osb,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700911 struct ocfs2_extent_tree *et)
Mark Fashehccd979b2005-12-15 14:31:24 -0800912{
913 int retval;
Tao Mae7d4cb62008-08-18 17:38:44 +0800914 struct ocfs2_extent_list *el = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800915 struct ocfs2_extent_block *eb;
916 struct buffer_head *eb_bh = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +0800917 u64 last_eb_blk = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800918
919 mlog_entry_void();
920
Joel Beckerf99b9b72008-08-20 19:36:33 -0700921 el = et->et_root_el;
922 last_eb_blk = ocfs2_et_get_last_eb_blk(et);
Mark Fashehccd979b2005-12-15 14:31:24 -0800923
Tao Mae7d4cb62008-08-18 17:38:44 +0800924 if (last_eb_blk) {
Joel Becker3d03a302009-02-12 17:49:26 -0800925 retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
926 &eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800927 if (retval < 0) {
928 mlog_errno(retval);
929 goto bail;
930 }
931 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
932 el = &eb->h_list;
Tao Mae7d4cb62008-08-18 17:38:44 +0800933 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800934
935 BUG_ON(el->l_tree_depth != 0);
936
937 retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
938bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700939 brelse(eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800940
941 mlog_exit(retval);
942 return retval;
943}
944
945/* expects array to already be allocated
946 *
947 * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
948 * l_count for you
949 */
Joel Becker42a5a7a2009-02-12 18:49:19 -0800950static int ocfs2_create_new_meta_bhs(handle_t *handle,
951 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -0800952 int wanted,
953 struct ocfs2_alloc_context *meta_ac,
954 struct buffer_head *bhs[])
955{
956 int count, status, i;
957 u16 suballoc_bit_start;
958 u32 num_got;
959 u64 first_blkno;
Joel Becker42a5a7a2009-02-12 18:49:19 -0800960 struct ocfs2_super *osb =
961 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 struct ocfs2_extent_block *eb;
963
964 mlog_entry_void();
965
966 count = 0;
967 while (count < wanted) {
968 status = ocfs2_claim_metadata(osb,
969 handle,
970 meta_ac,
971 wanted - count,
972 &suballoc_bit_start,
973 &num_got,
974 &first_blkno);
975 if (status < 0) {
976 mlog_errno(status);
977 goto bail;
978 }
979
980 for(i = count; i < (num_got + count); i++) {
981 bhs[i] = sb_getblk(osb->sb, first_blkno);
982 if (bhs[i] == NULL) {
983 status = -EIO;
984 mlog_errno(status);
985 goto bail;
986 }
Joel Becker42a5a7a2009-02-12 18:49:19 -0800987 ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -0800988
Joel Becker42a5a7a2009-02-12 18:49:19 -0800989 status = ocfs2_journal_access_eb(handle, et->et_ci,
990 bhs[i],
Joel Becker13723d02008-10-17 19:25:01 -0700991 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800992 if (status < 0) {
993 mlog_errno(status);
994 goto bail;
995 }
996
997 memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
998 eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
999 /* Ok, setup the minimal stuff here. */
1000 strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
1001 eb->h_blkno = cpu_to_le64(first_blkno);
1002 eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
Mark Fashehccd979b2005-12-15 14:31:24 -08001003 eb->h_suballoc_slot = cpu_to_le16(osb->slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001004 eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1005 eb->h_list.l_count =
1006 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
1007
1008 suballoc_bit_start++;
1009 first_blkno++;
1010
1011 /* We'll also be dirtied by the caller, so
1012 * this isn't absolutely necessary. */
1013 status = ocfs2_journal_dirty(handle, bhs[i]);
1014 if (status < 0) {
1015 mlog_errno(status);
1016 goto bail;
1017 }
1018 }
1019
1020 count += num_got;
1021 }
1022
1023 status = 0;
1024bail:
1025 if (status < 0) {
1026 for(i = 0; i < wanted; i++) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001027 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001028 bhs[i] = NULL;
1029 }
1030 }
1031 mlog_exit(status);
1032 return status;
1033}
1034
1035/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001036 * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
1037 *
1038 * Returns the sum of the rightmost extent rec logical offset and
1039 * cluster count.
1040 *
1041 * ocfs2_add_branch() uses this to determine what logical cluster
1042 * value should be populated into the leftmost new branch records.
1043 *
1044 * ocfs2_shift_tree_depth() uses this to determine the # clusters
1045 * value for the new topmost tree record.
1046 */
1047static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el)
1048{
1049 int i;
1050
1051 i = le16_to_cpu(el->l_next_free_rec) - 1;
1052
1053 return le32_to_cpu(el->l_recs[i].e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001054 ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08001055}
1056
1057/*
Tao Ma6b791bc2009-06-12 14:18:36 +08001058 * Change range of the branches in the right most path according to the leaf
1059 * extent block's rightmost record.
1060 */
1061static int ocfs2_adjust_rightmost_branch(handle_t *handle,
Tao Ma6b791bc2009-06-12 14:18:36 +08001062 struct ocfs2_extent_tree *et)
1063{
1064 int status;
1065 struct ocfs2_path *path = NULL;
1066 struct ocfs2_extent_list *el;
1067 struct ocfs2_extent_rec *rec;
1068
1069 path = ocfs2_new_path_from_et(et);
1070 if (!path) {
1071 status = -ENOMEM;
1072 return status;
1073 }
1074
Joel Beckerfacdb772009-02-12 18:08:48 -08001075 status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
Tao Ma6b791bc2009-06-12 14:18:36 +08001076 if (status < 0) {
1077 mlog_errno(status);
1078 goto out;
1079 }
1080
1081 status = ocfs2_extend_trans(handle, path_num_items(path) +
1082 handle->h_buffer_credits);
1083 if (status < 0) {
1084 mlog_errno(status);
1085 goto out;
1086 }
1087
Joel Beckerd401dc12009-02-13 02:24:10 -08001088 status = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma6b791bc2009-06-12 14:18:36 +08001089 if (status < 0) {
1090 mlog_errno(status);
1091 goto out;
1092 }
1093
1094 el = path_leaf_el(path);
1095 rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1];
1096
Joel Beckerd401dc12009-02-13 02:24:10 -08001097 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Tao Ma6b791bc2009-06-12 14:18:36 +08001098
1099out:
1100 ocfs2_free_path(path);
1101 return status;
1102}
1103
1104/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001105 * Add an entire tree branch to our inode. eb_bh is the extent block
Joel Beckerd401dc12009-02-13 02:24:10 -08001106 * to start at, if we don't want to start the branch at the root
Mark Fashehccd979b2005-12-15 14:31:24 -08001107 * structure.
1108 *
1109 * last_eb_bh is required as we have to update it's next_leaf pointer
1110 * for the new last extent block.
1111 *
1112 * the new branch will be 'empty' in the sense that every block will
Mark Fashehe48edee2007-03-07 16:46:57 -08001113 * contain a single record with cluster count == 0.
Mark Fashehccd979b2005-12-15 14:31:24 -08001114 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001115static int ocfs2_add_branch(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001116 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001117 struct buffer_head *eb_bh,
Mark Fasheh328d5752007-06-18 10:48:04 -07001118 struct buffer_head **last_eb_bh,
Mark Fashehccd979b2005-12-15 14:31:24 -08001119 struct ocfs2_alloc_context *meta_ac)
1120{
1121 int status, new_blocks, i;
1122 u64 next_blkno, new_last_eb_blk;
1123 struct buffer_head *bh;
1124 struct buffer_head **new_eb_bhs = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001125 struct ocfs2_extent_block *eb;
1126 struct ocfs2_extent_list *eb_el;
1127 struct ocfs2_extent_list *el;
Tao Ma6b791bc2009-06-12 14:18:36 +08001128 u32 new_cpos, root_end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001129
1130 mlog_entry_void();
1131
Mark Fasheh328d5752007-06-18 10:48:04 -07001132 BUG_ON(!last_eb_bh || !*last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001133
Mark Fashehccd979b2005-12-15 14:31:24 -08001134 if (eb_bh) {
1135 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1136 el = &eb->h_list;
1137 } else
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001138 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001139
1140 /* we never add a branch to a leaf. */
1141 BUG_ON(!el->l_tree_depth);
1142
1143 new_blocks = le16_to_cpu(el->l_tree_depth);
1144
Tao Ma6b791bc2009-06-12 14:18:36 +08001145 eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
1146 new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
1147 root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
1148
1149 /*
1150 * If there is a gap before the root end and the real end
1151 * of the righmost leaf block, we need to remove the gap
1152 * between new_cpos and root_end first so that the tree
1153 * is consistent after we add a new branch(it will start
1154 * from new_cpos).
1155 */
1156 if (root_end > new_cpos) {
1157 mlog(0, "adjust the cluster end from %u to %u\n",
1158 root_end, new_cpos);
Joel Beckerd401dc12009-02-13 02:24:10 -08001159 status = ocfs2_adjust_rightmost_branch(handle, et);
Tao Ma6b791bc2009-06-12 14:18:36 +08001160 if (status) {
1161 mlog_errno(status);
1162 goto bail;
1163 }
1164 }
1165
Mark Fashehccd979b2005-12-15 14:31:24 -08001166 /* allocate the number of new eb blocks we need */
1167 new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
1168 GFP_KERNEL);
1169 if (!new_eb_bhs) {
1170 status = -ENOMEM;
1171 mlog_errno(status);
1172 goto bail;
1173 }
1174
Joel Becker42a5a7a2009-02-12 18:49:19 -08001175 status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
Mark Fashehccd979b2005-12-15 14:31:24 -08001176 meta_ac, new_eb_bhs);
1177 if (status < 0) {
1178 mlog_errno(status);
1179 goto bail;
1180 }
1181
1182 /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
1183 * linked with the rest of the tree.
1184 * conversly, new_eb_bhs[0] is the new bottommost leaf.
1185 *
1186 * when we leave the loop, new_last_eb_blk will point to the
1187 * newest leaf, and next_blkno will point to the topmost extent
1188 * block. */
1189 next_blkno = new_last_eb_blk = 0;
1190 for(i = 0; i < new_blocks; i++) {
1191 bh = new_eb_bhs[i];
1192 eb = (struct ocfs2_extent_block *) bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001193 /* ocfs2_create_new_meta_bhs() should create it right! */
1194 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001195 eb_el = &eb->h_list;
1196
Joel Beckerd401dc12009-02-13 02:24:10 -08001197 status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
Joel Becker13723d02008-10-17 19:25:01 -07001198 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001199 if (status < 0) {
1200 mlog_errno(status);
1201 goto bail;
1202 }
1203
1204 eb->h_next_leaf_blk = 0;
1205 eb_el->l_tree_depth = cpu_to_le16(i);
1206 eb_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehdcd05382007-01-16 11:32:23 -08001207 /*
1208 * This actually counts as an empty extent as
1209 * c_clusters == 0
1210 */
1211 eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001212 eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehe48edee2007-03-07 16:46:57 -08001213 /*
1214 * eb_el isn't always an interior node, but even leaf
1215 * nodes want a zero'd flags and reserved field so
1216 * this gets the whole 32 bits regardless of use.
1217 */
1218 eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001219 if (!eb_el->l_tree_depth)
1220 new_last_eb_blk = le64_to_cpu(eb->h_blkno);
1221
1222 status = ocfs2_journal_dirty(handle, bh);
1223 if (status < 0) {
1224 mlog_errno(status);
1225 goto bail;
1226 }
1227
1228 next_blkno = le64_to_cpu(eb->h_blkno);
1229 }
1230
1231 /* This is a bit hairy. We want to update up to three blocks
1232 * here without leaving any of them in an inconsistent state
1233 * in case of error. We don't have to worry about
1234 * journal_dirty erroring as it won't unless we've aborted the
1235 * handle (in which case we would never be here) so reserving
1236 * the write with journal_access is all we need to do. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001237 status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001238 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001239 if (status < 0) {
1240 mlog_errno(status);
1241 goto bail;
1242 }
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001243 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001244 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001245 if (status < 0) {
1246 mlog_errno(status);
1247 goto bail;
1248 }
1249 if (eb_bh) {
Joel Beckerd401dc12009-02-13 02:24:10 -08001250 status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001251 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001252 if (status < 0) {
1253 mlog_errno(status);
1254 goto bail;
1255 }
1256 }
1257
1258 /* Link the new branch into the rest of the tree (el will
Tao Mae7d4cb62008-08-18 17:38:44 +08001259 * either be on the root_bh, or the extent block passed in. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001260 i = le16_to_cpu(el->l_next_free_rec);
1261 el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08001262 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001263 el->l_recs[i].e_int_clusters = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001264 le16_add_cpu(&el->l_next_free_rec, 1);
1265
1266 /* fe needs a new last extent block pointer, as does the
1267 * next_leaf on the previously last-extent-block. */
Joel Becker35dc0aa2008-08-20 16:25:06 -07001268 ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
Mark Fashehccd979b2005-12-15 14:31:24 -08001269
Mark Fasheh328d5752007-06-18 10:48:04 -07001270 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001271 eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
1272
Mark Fasheh328d5752007-06-18 10:48:04 -07001273 status = ocfs2_journal_dirty(handle, *last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001274 if (status < 0)
1275 mlog_errno(status);
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001276 status = ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001277 if (status < 0)
1278 mlog_errno(status);
1279 if (eb_bh) {
1280 status = ocfs2_journal_dirty(handle, eb_bh);
1281 if (status < 0)
1282 mlog_errno(status);
1283 }
1284
Mark Fasheh328d5752007-06-18 10:48:04 -07001285 /*
1286 * Some callers want to track the rightmost leaf so pass it
1287 * back here.
1288 */
1289 brelse(*last_eb_bh);
1290 get_bh(new_eb_bhs[0]);
1291 *last_eb_bh = new_eb_bhs[0];
1292
Mark Fashehccd979b2005-12-15 14:31:24 -08001293 status = 0;
1294bail:
1295 if (new_eb_bhs) {
1296 for (i = 0; i < new_blocks; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001297 brelse(new_eb_bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001298 kfree(new_eb_bhs);
1299 }
1300
1301 mlog_exit(status);
1302 return status;
1303}
1304
1305/*
1306 * adds another level to the allocation tree.
1307 * returns back the new extent block so you can add a branch to it
1308 * after this call.
1309 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001310static int ocfs2_shift_tree_depth(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001311 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001312 struct ocfs2_alloc_context *meta_ac,
1313 struct buffer_head **ret_new_eb_bh)
1314{
1315 int status, i;
Mark Fashehdcd05382007-01-16 11:32:23 -08001316 u32 new_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001317 struct buffer_head *new_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001318 struct ocfs2_extent_block *eb;
Tao Mae7d4cb62008-08-18 17:38:44 +08001319 struct ocfs2_extent_list *root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001320 struct ocfs2_extent_list *eb_el;
1321
1322 mlog_entry_void();
1323
Joel Becker42a5a7a2009-02-12 18:49:19 -08001324 status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
Mark Fashehccd979b2005-12-15 14:31:24 -08001325 &new_eb_bh);
1326 if (status < 0) {
1327 mlog_errno(status);
1328 goto bail;
1329 }
1330
1331 eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001332 /* ocfs2_create_new_meta_bhs() should create it right! */
1333 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001334
1335 eb_el = &eb->h_list;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001336 root_el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001337
Joel Beckerd401dc12009-02-13 02:24:10 -08001338 status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001339 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001340 if (status < 0) {
1341 mlog_errno(status);
1342 goto bail;
1343 }
1344
Tao Mae7d4cb62008-08-18 17:38:44 +08001345 /* copy the root extent list data into the new extent block */
1346 eb_el->l_tree_depth = root_el->l_tree_depth;
1347 eb_el->l_next_free_rec = root_el->l_next_free_rec;
1348 for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1349 eb_el->l_recs[i] = root_el->l_recs[i];
Mark Fashehccd979b2005-12-15 14:31:24 -08001350
1351 status = ocfs2_journal_dirty(handle, new_eb_bh);
1352 if (status < 0) {
1353 mlog_errno(status);
1354 goto bail;
1355 }
1356
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001357 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001358 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001359 if (status < 0) {
1360 mlog_errno(status);
1361 goto bail;
1362 }
1363
Mark Fashehdcd05382007-01-16 11:32:23 -08001364 new_clusters = ocfs2_sum_rightmost_rec(eb_el);
1365
Tao Mae7d4cb62008-08-18 17:38:44 +08001366 /* update root_bh now */
1367 le16_add_cpu(&root_el->l_tree_depth, 1);
1368 root_el->l_recs[0].e_cpos = 0;
1369 root_el->l_recs[0].e_blkno = eb->h_blkno;
1370 root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
1371 for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1372 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
1373 root_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001374
1375 /* If this is our 1st tree depth shift, then last_eb_blk
1376 * becomes the allocated extent block */
Tao Mae7d4cb62008-08-18 17:38:44 +08001377 if (root_el->l_tree_depth == cpu_to_le16(1))
Joel Becker35dc0aa2008-08-20 16:25:06 -07001378 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001379
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001380 status = ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001381 if (status < 0) {
1382 mlog_errno(status);
1383 goto bail;
1384 }
1385
1386 *ret_new_eb_bh = new_eb_bh;
1387 new_eb_bh = NULL;
1388 status = 0;
1389bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001390 brelse(new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001391
1392 mlog_exit(status);
1393 return status;
1394}
1395
1396/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001397 * Should only be called when there is no space left in any of the
1398 * leaf nodes. What we want to do is find the lowest tree depth
1399 * non-leaf extent block with room for new records. There are three
1400 * valid results of this search:
1401 *
1402 * 1) a lowest extent block is found, then we pass it back in
1403 * *lowest_eb_bh and return '0'
1404 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001405 * 2) the search fails to find anything, but the root_el has room. We
Mark Fashehccd979b2005-12-15 14:31:24 -08001406 * pass NULL back in *lowest_eb_bh, but still return '0'
1407 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001408 * 3) the search fails to find anything AND the root_el is full, in
Mark Fashehccd979b2005-12-15 14:31:24 -08001409 * which case we return > 0
1410 *
1411 * return status < 0 indicates an error.
1412 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001413static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001414 struct buffer_head **target_bh)
1415{
1416 int status = 0, i;
1417 u64 blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08001418 struct ocfs2_extent_block *eb;
1419 struct ocfs2_extent_list *el;
1420 struct buffer_head *bh = NULL;
1421 struct buffer_head *lowest_bh = NULL;
1422
1423 mlog_entry_void();
1424
1425 *target_bh = NULL;
1426
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001427 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001428
1429 while(le16_to_cpu(el->l_tree_depth) > 1) {
1430 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08001431 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1432 "Owner %llu has empty "
Mark Fashehccd979b2005-12-15 14:31:24 -08001433 "extent list (next_free_rec == 0)",
Joel Becker3d03a302009-02-12 17:49:26 -08001434 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001435 status = -EIO;
1436 goto bail;
1437 }
1438 i = le16_to_cpu(el->l_next_free_rec) - 1;
1439 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1440 if (!blkno) {
Joel Becker3d03a302009-02-12 17:49:26 -08001441 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1442 "Owner %llu has extent "
Mark Fashehccd979b2005-12-15 14:31:24 -08001443 "list where extent # %d has no physical "
1444 "block start",
Joel Becker3d03a302009-02-12 17:49:26 -08001445 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
Mark Fashehccd979b2005-12-15 14:31:24 -08001446 status = -EIO;
1447 goto bail;
1448 }
1449
Mark Fasheha81cb882008-10-07 14:25:16 -07001450 brelse(bh);
1451 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001452
Joel Becker3d03a302009-02-12 17:49:26 -08001453 status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001454 if (status < 0) {
1455 mlog_errno(status);
1456 goto bail;
1457 }
1458
1459 eb = (struct ocfs2_extent_block *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001460 el = &eb->h_list;
1461
1462 if (le16_to_cpu(el->l_next_free_rec) <
1463 le16_to_cpu(el->l_count)) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001464 brelse(lowest_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001465 lowest_bh = bh;
1466 get_bh(lowest_bh);
1467 }
1468 }
1469
1470 /* If we didn't find one and the fe doesn't have any room,
1471 * then return '1' */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001472 el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001473 if (!lowest_bh && (el->l_next_free_rec == el->l_count))
Mark Fashehccd979b2005-12-15 14:31:24 -08001474 status = 1;
1475
1476 *target_bh = lowest_bh;
1477bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001478 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001479
1480 mlog_exit(status);
1481 return status;
1482}
1483
Mark Fashehe48edee2007-03-07 16:46:57 -08001484/*
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001485 * Grow a b-tree so that it has more records.
1486 *
1487 * We might shift the tree depth in which case existing paths should
1488 * be considered invalid.
1489 *
1490 * Tree depth after the grow is returned via *final_depth.
Mark Fasheh328d5752007-06-18 10:48:04 -07001491 *
1492 * *last_eb_bh will be updated by ocfs2_add_branch().
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001493 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001494static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
1495 int *final_depth, struct buffer_head **last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001496 struct ocfs2_alloc_context *meta_ac)
1497{
1498 int ret, shift;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001499 struct ocfs2_extent_list *el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001500 int depth = le16_to_cpu(el->l_tree_depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001501 struct buffer_head *bh = NULL;
1502
1503 BUG_ON(meta_ac == NULL);
1504
Joel Beckerd401dc12009-02-13 02:24:10 -08001505 shift = ocfs2_find_branch_target(et, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001506 if (shift < 0) {
1507 ret = shift;
1508 mlog_errno(ret);
1509 goto out;
1510 }
1511
1512 /* We traveled all the way to the bottom of the allocation tree
1513 * and didn't find room for any more extents - we need to add
1514 * another tree level */
1515 if (shift) {
1516 BUG_ON(bh);
1517 mlog(0, "need to shift tree depth (current = %d)\n", depth);
1518
1519 /* ocfs2_shift_tree_depth will return us a buffer with
1520 * the new extent block (so we can pass that to
1521 * ocfs2_add_branch). */
Joel Beckerd401dc12009-02-13 02:24:10 -08001522 ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001523 if (ret < 0) {
1524 mlog_errno(ret);
1525 goto out;
1526 }
1527 depth++;
Mark Fasheh328d5752007-06-18 10:48:04 -07001528 if (depth == 1) {
1529 /*
1530 * Special case: we have room now if we shifted from
1531 * tree_depth 0, so no more work needs to be done.
1532 *
1533 * We won't be calling add_branch, so pass
1534 * back *last_eb_bh as the new leaf. At depth
1535 * zero, it should always be null so there's
1536 * no reason to brelse.
1537 */
1538 BUG_ON(*last_eb_bh);
1539 get_bh(bh);
1540 *last_eb_bh = bh;
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001541 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07001542 }
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001543 }
1544
1545 /* call ocfs2_add_branch to add the final part of the tree with
1546 * the new data. */
1547 mlog(0, "add branch. bh = %p\n", bh);
Joel Beckerd401dc12009-02-13 02:24:10 -08001548 ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001549 meta_ac);
1550 if (ret < 0) {
1551 mlog_errno(ret);
1552 goto out;
1553 }
1554
1555out:
1556 if (final_depth)
1557 *final_depth = depth;
1558 brelse(bh);
1559 return ret;
1560}
1561
1562/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001563 * This function will discard the rightmost extent record.
1564 */
1565static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
1566{
1567 int next_free = le16_to_cpu(el->l_next_free_rec);
1568 int count = le16_to_cpu(el->l_count);
1569 unsigned int num_bytes;
1570
1571 BUG_ON(!next_free);
1572 /* This will cause us to go off the end of our extent list. */
1573 BUG_ON(next_free >= count);
1574
1575 num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
1576
1577 memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
1578}
1579
1580static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
1581 struct ocfs2_extent_rec *insert_rec)
1582{
1583 int i, insert_index, next_free, has_empty, num_bytes;
1584 u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
1585 struct ocfs2_extent_rec *rec;
1586
1587 next_free = le16_to_cpu(el->l_next_free_rec);
1588 has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
1589
1590 BUG_ON(!next_free);
1591
1592 /* The tree code before us didn't allow enough room in the leaf. */
Julia Lawallb1f35502008-03-04 15:21:05 -08001593 BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
Mark Fashehdcd05382007-01-16 11:32:23 -08001594
1595 /*
1596 * The easiest way to approach this is to just remove the
1597 * empty extent and temporarily decrement next_free.
1598 */
1599 if (has_empty) {
1600 /*
1601 * If next_free was 1 (only an empty extent), this
1602 * loop won't execute, which is fine. We still want
1603 * the decrement above to happen.
1604 */
1605 for(i = 0; i < (next_free - 1); i++)
1606 el->l_recs[i] = el->l_recs[i+1];
1607
1608 next_free--;
1609 }
1610
1611 /*
1612 * Figure out what the new record index should be.
1613 */
1614 for(i = 0; i < next_free; i++) {
1615 rec = &el->l_recs[i];
1616
1617 if (insert_cpos < le32_to_cpu(rec->e_cpos))
1618 break;
1619 }
1620 insert_index = i;
1621
1622 mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n",
1623 insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count));
1624
1625 BUG_ON(insert_index < 0);
1626 BUG_ON(insert_index >= le16_to_cpu(el->l_count));
1627 BUG_ON(insert_index > next_free);
1628
1629 /*
1630 * No need to memmove if we're just adding to the tail.
1631 */
1632 if (insert_index != next_free) {
1633 BUG_ON(next_free >= le16_to_cpu(el->l_count));
1634
1635 num_bytes = next_free - insert_index;
1636 num_bytes *= sizeof(struct ocfs2_extent_rec);
1637 memmove(&el->l_recs[insert_index + 1],
1638 &el->l_recs[insert_index],
1639 num_bytes);
1640 }
1641
1642 /*
1643 * Either we had an empty extent, and need to re-increment or
1644 * there was no empty extent on a non full rightmost leaf node,
1645 * in which case we still need to increment.
1646 */
1647 next_free++;
1648 el->l_next_free_rec = cpu_to_le16(next_free);
1649 /*
1650 * Make sure none of the math above just messed up our tree.
1651 */
1652 BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
1653
1654 el->l_recs[insert_index] = *insert_rec;
1655
1656}
1657
Mark Fasheh328d5752007-06-18 10:48:04 -07001658static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
1659{
1660 int size, num_recs = le16_to_cpu(el->l_next_free_rec);
1661
1662 BUG_ON(num_recs == 0);
1663
1664 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
1665 num_recs--;
1666 size = num_recs * sizeof(struct ocfs2_extent_rec);
1667 memmove(&el->l_recs[0], &el->l_recs[1], size);
1668 memset(&el->l_recs[num_recs], 0,
1669 sizeof(struct ocfs2_extent_rec));
1670 el->l_next_free_rec = cpu_to_le16(num_recs);
1671 }
1672}
1673
Mark Fashehdcd05382007-01-16 11:32:23 -08001674/*
1675 * Create an empty extent record .
1676 *
1677 * l_next_free_rec may be updated.
1678 *
1679 * If an empty extent already exists do nothing.
1680 */
1681static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
1682{
1683 int next_free = le16_to_cpu(el->l_next_free_rec);
1684
Mark Fashehe48edee2007-03-07 16:46:57 -08001685 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
1686
Mark Fashehdcd05382007-01-16 11:32:23 -08001687 if (next_free == 0)
1688 goto set_and_inc;
1689
1690 if (ocfs2_is_empty_extent(&el->l_recs[0]))
1691 return;
1692
1693 mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
1694 "Asked to create an empty extent in a full list:\n"
1695 "count = %u, tree depth = %u",
1696 le16_to_cpu(el->l_count),
1697 le16_to_cpu(el->l_tree_depth));
1698
1699 ocfs2_shift_records_right(el);
1700
1701set_and_inc:
1702 le16_add_cpu(&el->l_next_free_rec, 1);
1703 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
1704}
1705
1706/*
1707 * For a rotation which involves two leaf nodes, the "root node" is
1708 * the lowest level tree node which contains a path to both leafs. This
1709 * resulting set of information can be used to form a complete "subtree"
1710 *
1711 * This function is passed two full paths from the dinode down to a
1712 * pair of adjacent leaves. It's task is to figure out which path
1713 * index contains the subtree root - this can be the root index itself
1714 * in a worst-case rotation.
1715 *
1716 * The array index of the subtree root is passed back.
1717 */
Joel Becker7dc02802009-02-12 19:20:13 -08001718static int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08001719 struct ocfs2_path *left,
1720 struct ocfs2_path *right)
1721{
1722 int i = 0;
1723
1724 /*
1725 * Check that the caller passed in two paths from the same tree.
1726 */
1727 BUG_ON(path_root_bh(left) != path_root_bh(right));
1728
1729 do {
1730 i++;
1731
1732 /*
1733 * The caller didn't pass two adjacent paths.
1734 */
1735 mlog_bug_on_msg(i > left->p_tree_depth,
Joel Becker7dc02802009-02-12 19:20:13 -08001736 "Owner %llu, left depth %u, right depth %u\n"
Mark Fashehdcd05382007-01-16 11:32:23 -08001737 "left leaf blk %llu, right leaf blk %llu\n",
Joel Becker7dc02802009-02-12 19:20:13 -08001738 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
1739 left->p_tree_depth, right->p_tree_depth,
Mark Fashehdcd05382007-01-16 11:32:23 -08001740 (unsigned long long)path_leaf_bh(left)->b_blocknr,
1741 (unsigned long long)path_leaf_bh(right)->b_blocknr);
1742 } while (left->p_node[i].bh->b_blocknr ==
1743 right->p_node[i].bh->b_blocknr);
1744
1745 return i - 1;
1746}
1747
1748typedef void (path_insert_t)(void *, struct buffer_head *);
1749
1750/*
1751 * Traverse a btree path in search of cpos, starting at root_el.
1752 *
1753 * This code can be called with a cpos larger than the tree, in which
1754 * case it will return the rightmost path.
1755 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001756static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
Mark Fashehdcd05382007-01-16 11:32:23 -08001757 struct ocfs2_extent_list *root_el, u32 cpos,
1758 path_insert_t *func, void *data)
1759{
1760 int i, ret = 0;
1761 u32 range;
1762 u64 blkno;
1763 struct buffer_head *bh = NULL;
1764 struct ocfs2_extent_block *eb;
1765 struct ocfs2_extent_list *el;
1766 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08001767
1768 el = root_el;
1769 while (el->l_tree_depth) {
1770 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001771 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1772 "Owner %llu has empty extent list at "
Mark Fashehdcd05382007-01-16 11:32:23 -08001773 "depth %u\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001774 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001775 le16_to_cpu(el->l_tree_depth));
1776 ret = -EROFS;
1777 goto out;
1778
1779 }
1780
1781 for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
1782 rec = &el->l_recs[i];
1783
1784 /*
1785 * In the case that cpos is off the allocation
1786 * tree, this should just wind up returning the
1787 * rightmost record.
1788 */
1789 range = le32_to_cpu(rec->e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001790 ocfs2_rec_clusters(el, rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08001791 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
1792 break;
1793 }
1794
1795 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1796 if (blkno == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001797 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1798 "Owner %llu has bad blkno in extent list "
Mark Fashehdcd05382007-01-16 11:32:23 -08001799 "at depth %u (index %d)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001800 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001801 le16_to_cpu(el->l_tree_depth), i);
1802 ret = -EROFS;
1803 goto out;
1804 }
1805
1806 brelse(bh);
1807 bh = NULL;
Joel Beckerfacdb772009-02-12 18:08:48 -08001808 ret = ocfs2_read_extent_block(ci, blkno, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001809 if (ret) {
1810 mlog_errno(ret);
1811 goto out;
1812 }
1813
1814 eb = (struct ocfs2_extent_block *) bh->b_data;
1815 el = &eb->h_list;
Mark Fashehdcd05382007-01-16 11:32:23 -08001816
1817 if (le16_to_cpu(el->l_next_free_rec) >
1818 le16_to_cpu(el->l_count)) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001819 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1820 "Owner %llu has bad count in extent list "
Mark Fashehdcd05382007-01-16 11:32:23 -08001821 "at block %llu (next free=%u, count=%u)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001822 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001823 (unsigned long long)bh->b_blocknr,
1824 le16_to_cpu(el->l_next_free_rec),
1825 le16_to_cpu(el->l_count));
1826 ret = -EROFS;
1827 goto out;
1828 }
1829
1830 if (func)
1831 func(data, bh);
1832 }
1833
1834out:
1835 /*
1836 * Catch any trailing bh that the loop didn't handle.
1837 */
1838 brelse(bh);
1839
1840 return ret;
1841}
1842
1843/*
1844 * Given an initialized path (that is, it has a valid root extent
1845 * list), this function will traverse the btree in search of the path
1846 * which would contain cpos.
1847 *
1848 * The path traveled is recorded in the path structure.
1849 *
1850 * Note that this will not do any comparisons on leaf node extent
1851 * records, so it will work fine in the case that we just added a tree
1852 * branch.
1853 */
1854struct find_path_data {
1855 int index;
1856 struct ocfs2_path *path;
1857};
1858static void find_path_ins(void *data, struct buffer_head *bh)
1859{
1860 struct find_path_data *fp = data;
1861
1862 get_bh(bh);
1863 ocfs2_path_insert_eb(fp->path, fp->index, bh);
1864 fp->index++;
1865}
Joel Beckerfacdb772009-02-12 18:08:48 -08001866static int ocfs2_find_path(struct ocfs2_caching_info *ci,
1867 struct ocfs2_path *path, u32 cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08001868{
1869 struct find_path_data data;
1870
1871 data.index = 1;
1872 data.path = path;
Joel Beckerfacdb772009-02-12 18:08:48 -08001873 return __ocfs2_find_path(ci, path_root_el(path), cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08001874 find_path_ins, &data);
1875}
1876
1877static void find_leaf_ins(void *data, struct buffer_head *bh)
1878{
1879 struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
1880 struct ocfs2_extent_list *el = &eb->h_list;
1881 struct buffer_head **ret = data;
1882
1883 /* We want to retain only the leaf block. */
1884 if (le16_to_cpu(el->l_tree_depth) == 0) {
1885 get_bh(bh);
1886 *ret = bh;
1887 }
1888}
1889/*
1890 * Find the leaf block in the tree which would contain cpos. No
1891 * checking of the actual leaf is done.
1892 *
1893 * Some paths want to call this instead of allocating a path structure
1894 * and calling ocfs2_find_path().
1895 *
1896 * This function doesn't handle non btree extent lists.
1897 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001898int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
1899 struct ocfs2_extent_list *root_el, u32 cpos,
1900 struct buffer_head **leaf_bh)
Mark Fashehdcd05382007-01-16 11:32:23 -08001901{
1902 int ret;
1903 struct buffer_head *bh = NULL;
1904
Joel Beckerfacdb772009-02-12 18:08:48 -08001905 ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001906 if (ret) {
1907 mlog_errno(ret);
1908 goto out;
1909 }
1910
1911 *leaf_bh = bh;
1912out:
1913 return ret;
1914}
1915
1916/*
1917 * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
1918 *
1919 * Basically, we've moved stuff around at the bottom of the tree and
1920 * we need to fix up the extent records above the changes to reflect
1921 * the new changes.
1922 *
1923 * left_rec: the record on the left.
1924 * left_child_el: is the child list pointed to by left_rec
1925 * right_rec: the record to the right of left_rec
1926 * right_child_el: is the child list pointed to by right_rec
1927 *
1928 * By definition, this only works on interior nodes.
1929 */
1930static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
1931 struct ocfs2_extent_list *left_child_el,
1932 struct ocfs2_extent_rec *right_rec,
1933 struct ocfs2_extent_list *right_child_el)
1934{
1935 u32 left_clusters, right_end;
1936
1937 /*
1938 * Interior nodes never have holes. Their cpos is the cpos of
1939 * the leftmost record in their child list. Their cluster
1940 * count covers the full theoretical range of their child list
1941 * - the range between their cpos and the cpos of the record
1942 * immediately to their right.
1943 */
1944 left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
Tao Ma82e12642009-07-23 08:12:58 +08001945 if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
1946 BUG_ON(right_child_el->l_tree_depth);
Mark Fasheh328d5752007-06-18 10:48:04 -07001947 BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
1948 left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
1949 }
Mark Fashehdcd05382007-01-16 11:32:23 -08001950 left_clusters -= le32_to_cpu(left_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001951 left_rec->e_int_clusters = cpu_to_le32(left_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001952
1953 /*
1954 * Calculate the rightmost cluster count boundary before
Mark Fashehe48edee2007-03-07 16:46:57 -08001955 * moving cpos - we will need to adjust clusters after
Mark Fashehdcd05382007-01-16 11:32:23 -08001956 * updating e_cpos to keep the same highest cluster count.
1957 */
1958 right_end = le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001959 right_end += le32_to_cpu(right_rec->e_int_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001960
1961 right_rec->e_cpos = left_rec->e_cpos;
1962 le32_add_cpu(&right_rec->e_cpos, left_clusters);
1963
1964 right_end -= le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001965 right_rec->e_int_clusters = cpu_to_le32(right_end);
Mark Fashehdcd05382007-01-16 11:32:23 -08001966}
1967
1968/*
1969 * Adjust the adjacent root node records involved in a
1970 * rotation. left_el_blkno is passed in as a key so that we can easily
1971 * find it's index in the root list.
1972 */
1973static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
1974 struct ocfs2_extent_list *left_el,
1975 struct ocfs2_extent_list *right_el,
1976 u64 left_el_blkno)
1977{
1978 int i;
1979
1980 BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
1981 le16_to_cpu(left_el->l_tree_depth));
1982
1983 for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
1984 if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
1985 break;
1986 }
1987
1988 /*
1989 * The path walking code should have never returned a root and
1990 * two paths which are not adjacent.
1991 */
1992 BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
1993
1994 ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
1995 &root_el->l_recs[i + 1], right_el);
1996}
1997
1998/*
1999 * We've changed a leaf block (in right_path) and need to reflect that
2000 * change back up the subtree.
2001 *
2002 * This happens in multiple places:
2003 * - When we've moved an extent record from the left path leaf to the right
2004 * path leaf to make room for an empty extent in the left path leaf.
2005 * - When our insert into the right path leaf is at the leftmost edge
2006 * and requires an update of the path immediately to it's left. This
2007 * can occur at the end of some types of rotation and appending inserts.
Tao Ma677b9752008-01-30 14:21:05 +08002008 * - When we've adjusted the last extent record in the left path leaf and the
2009 * 1st extent record in the right path leaf during cross extent block merge.
Mark Fashehdcd05382007-01-16 11:32:23 -08002010 */
Joel Becker4619c732009-02-12 19:02:36 -08002011static void ocfs2_complete_edge_insert(handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -08002012 struct ocfs2_path *left_path,
2013 struct ocfs2_path *right_path,
2014 int subtree_index)
2015{
2016 int ret, i, idx;
2017 struct ocfs2_extent_list *el, *left_el, *right_el;
2018 struct ocfs2_extent_rec *left_rec, *right_rec;
2019 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2020
2021 /*
2022 * Update the counts and position values within all the
2023 * interior nodes to reflect the leaf rotation we just did.
2024 *
2025 * The root node is handled below the loop.
2026 *
2027 * We begin the loop with right_el and left_el pointing to the
2028 * leaf lists and work our way up.
2029 *
2030 * NOTE: within this loop, left_el and right_el always refer
2031 * to the *child* lists.
2032 */
2033 left_el = path_leaf_el(left_path);
2034 right_el = path_leaf_el(right_path);
2035 for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
2036 mlog(0, "Adjust records at index %u\n", i);
2037
2038 /*
2039 * One nice property of knowing that all of these
2040 * nodes are below the root is that we only deal with
2041 * the leftmost right node record and the rightmost
2042 * left node record.
2043 */
2044 el = left_path->p_node[i].el;
2045 idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
2046 left_rec = &el->l_recs[idx];
2047
2048 el = right_path->p_node[i].el;
2049 right_rec = &el->l_recs[0];
2050
2051 ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
2052 right_el);
2053
2054 ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
2055 if (ret)
2056 mlog_errno(ret);
2057
2058 ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
2059 if (ret)
2060 mlog_errno(ret);
2061
2062 /*
2063 * Setup our list pointers now so that the current
2064 * parents become children in the next iteration.
2065 */
2066 left_el = left_path->p_node[i].el;
2067 right_el = right_path->p_node[i].el;
2068 }
2069
2070 /*
2071 * At the root node, adjust the two adjacent records which
2072 * begin our path to the leaves.
2073 */
2074
2075 el = left_path->p_node[subtree_index].el;
2076 left_el = left_path->p_node[subtree_index + 1].el;
2077 right_el = right_path->p_node[subtree_index + 1].el;
2078
2079 ocfs2_adjust_root_records(el, left_el, right_el,
2080 left_path->p_node[subtree_index + 1].bh->b_blocknr);
2081
2082 root_bh = left_path->p_node[subtree_index].bh;
2083
2084 ret = ocfs2_journal_dirty(handle, root_bh);
2085 if (ret)
2086 mlog_errno(ret);
2087}
2088
Joel Becker5c601ab2009-02-12 19:10:13 -08002089static int ocfs2_rotate_subtree_right(handle_t *handle,
2090 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08002091 struct ocfs2_path *left_path,
2092 struct ocfs2_path *right_path,
2093 int subtree_index)
2094{
2095 int ret, i;
2096 struct buffer_head *right_leaf_bh;
2097 struct buffer_head *left_leaf_bh = NULL;
2098 struct buffer_head *root_bh;
2099 struct ocfs2_extent_list *right_el, *left_el;
2100 struct ocfs2_extent_rec move_rec;
2101
2102 left_leaf_bh = path_leaf_bh(left_path);
2103 left_el = path_leaf_el(left_path);
2104
2105 if (left_el->l_next_free_rec != left_el->l_count) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002106 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002107 "Inode %llu has non-full interior leaf node %llu"
2108 "(next free = %u)",
Joel Becker5c601ab2009-02-12 19:10:13 -08002109 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002110 (unsigned long long)left_leaf_bh->b_blocknr,
2111 le16_to_cpu(left_el->l_next_free_rec));
2112 return -EROFS;
2113 }
2114
2115 /*
2116 * This extent block may already have an empty record, so we
2117 * return early if so.
2118 */
2119 if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
2120 return 0;
2121
2122 root_bh = left_path->p_node[subtree_index].bh;
2123 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2124
Joel Becker5c601ab2009-02-12 19:10:13 -08002125 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002126 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002127 if (ret) {
2128 mlog_errno(ret);
2129 goto out;
2130 }
2131
2132 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002133 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002134 right_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002135 if (ret) {
2136 mlog_errno(ret);
2137 goto out;
2138 }
2139
Joel Becker5c601ab2009-02-12 19:10:13 -08002140 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002141 left_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002142 if (ret) {
2143 mlog_errno(ret);
2144 goto out;
2145 }
2146 }
2147
2148 right_leaf_bh = path_leaf_bh(right_path);
2149 right_el = path_leaf_el(right_path);
2150
2151 /* This is a code error, not a disk corruption. */
2152 mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
2153 "because rightmost leaf block %llu is empty\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002154 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002155 (unsigned long long)right_leaf_bh->b_blocknr);
2156
2157 ocfs2_create_empty_extent(right_el);
2158
2159 ret = ocfs2_journal_dirty(handle, right_leaf_bh);
2160 if (ret) {
2161 mlog_errno(ret);
2162 goto out;
2163 }
2164
2165 /* Do the copy now. */
2166 i = le16_to_cpu(left_el->l_next_free_rec) - 1;
2167 move_rec = left_el->l_recs[i];
2168 right_el->l_recs[0] = move_rec;
2169
2170 /*
2171 * Clear out the record we just copied and shift everything
2172 * over, leaving an empty extent in the left leaf.
2173 *
2174 * We temporarily subtract from next_free_rec so that the
2175 * shift will lose the tail record (which is now defunct).
2176 */
2177 le16_add_cpu(&left_el->l_next_free_rec, -1);
2178 ocfs2_shift_records_right(left_el);
2179 memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2180 le16_add_cpu(&left_el->l_next_free_rec, 1);
2181
2182 ret = ocfs2_journal_dirty(handle, left_leaf_bh);
2183 if (ret) {
2184 mlog_errno(ret);
2185 goto out;
2186 }
2187
Joel Becker4619c732009-02-12 19:02:36 -08002188 ocfs2_complete_edge_insert(handle, left_path, right_path,
2189 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002190
2191out:
2192 return ret;
2193}
2194
2195/*
2196 * Given a full path, determine what cpos value would return us a path
2197 * containing the leaf immediately to the left of the current one.
2198 *
2199 * Will return zero if the path passed in is already the leftmost path.
2200 */
2201static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
2202 struct ocfs2_path *path, u32 *cpos)
2203{
2204 int i, j, ret = 0;
2205 u64 blkno;
2206 struct ocfs2_extent_list *el;
2207
Mark Fashehe48edee2007-03-07 16:46:57 -08002208 BUG_ON(path->p_tree_depth == 0);
2209
Mark Fashehdcd05382007-01-16 11:32:23 -08002210 *cpos = 0;
2211
2212 blkno = path_leaf_bh(path)->b_blocknr;
2213
2214 /* Start at the tree node just above the leaf and work our way up. */
2215 i = path->p_tree_depth - 1;
2216 while (i >= 0) {
2217 el = path->p_node[i].el;
2218
2219 /*
2220 * Find the extent record just before the one in our
2221 * path.
2222 */
2223 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2224 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2225 if (j == 0) {
2226 if (i == 0) {
2227 /*
2228 * We've determined that the
2229 * path specified is already
2230 * the leftmost one - return a
2231 * cpos of zero.
2232 */
2233 goto out;
2234 }
2235 /*
2236 * The leftmost record points to our
2237 * leaf - we need to travel up the
2238 * tree one level.
2239 */
2240 goto next_node;
2241 }
2242
2243 *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08002244 *cpos = *cpos + ocfs2_rec_clusters(el,
2245 &el->l_recs[j - 1]);
2246 *cpos = *cpos - 1;
Mark Fashehdcd05382007-01-16 11:32:23 -08002247 goto out;
2248 }
2249 }
2250
2251 /*
2252 * If we got here, we never found a valid node where
2253 * the tree indicated one should be.
2254 */
2255 ocfs2_error(sb,
2256 "Invalid extent tree at extent block %llu\n",
2257 (unsigned long long)blkno);
2258 ret = -EROFS;
2259 goto out;
2260
2261next_node:
2262 blkno = path->p_node[i].bh->b_blocknr;
2263 i--;
2264 }
2265
2266out:
2267 return ret;
2268}
2269
Mark Fasheh328d5752007-06-18 10:48:04 -07002270/*
2271 * Extend the transaction by enough credits to complete the rotation,
2272 * and still leave at least the original number of credits allocated
2273 * to this transaction.
2274 */
Mark Fashehdcd05382007-01-16 11:32:23 -08002275static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
Mark Fasheh328d5752007-06-18 10:48:04 -07002276 int op_credits,
Mark Fashehdcd05382007-01-16 11:32:23 -08002277 struct ocfs2_path *path)
2278{
Mark Fasheh328d5752007-06-18 10:48:04 -07002279 int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002280
2281 if (handle->h_buffer_credits < credits)
2282 return ocfs2_extend_trans(handle, credits);
2283
2284 return 0;
2285}
2286
2287/*
2288 * Trap the case where we're inserting into the theoretical range past
2289 * the _actual_ left leaf range. Otherwise, we'll rotate a record
2290 * whose cpos is less than ours into the right leaf.
2291 *
2292 * It's only necessary to look at the rightmost record of the left
2293 * leaf because the logic that calls us should ensure that the
2294 * theoretical ranges in the path components above the leaves are
2295 * correct.
2296 */
2297static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
2298 u32 insert_cpos)
2299{
2300 struct ocfs2_extent_list *left_el;
2301 struct ocfs2_extent_rec *rec;
2302 int next_free;
2303
2304 left_el = path_leaf_el(left_path);
2305 next_free = le16_to_cpu(left_el->l_next_free_rec);
2306 rec = &left_el->l_recs[next_free - 1];
2307
2308 if (insert_cpos > le32_to_cpu(rec->e_cpos))
2309 return 1;
2310 return 0;
2311}
2312
Mark Fasheh328d5752007-06-18 10:48:04 -07002313static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
2314{
2315 int next_free = le16_to_cpu(el->l_next_free_rec);
2316 unsigned int range;
2317 struct ocfs2_extent_rec *rec;
2318
2319 if (next_free == 0)
2320 return 0;
2321
2322 rec = &el->l_recs[0];
2323 if (ocfs2_is_empty_extent(rec)) {
2324 /* Empty list. */
2325 if (next_free == 1)
2326 return 0;
2327 rec = &el->l_recs[1];
2328 }
2329
2330 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2331 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
2332 return 1;
2333 return 0;
2334}
2335
Mark Fashehdcd05382007-01-16 11:32:23 -08002336/*
2337 * Rotate all the records in a btree right one record, starting at insert_cpos.
2338 *
2339 * The path to the rightmost leaf should be passed in.
2340 *
2341 * The array is assumed to be large enough to hold an entire path (tree depth).
2342 *
2343 * Upon succesful return from this function:
2344 *
2345 * - The 'right_path' array will contain a path to the leaf block
2346 * whose range contains e_cpos.
2347 * - That leaf block will have a single empty extent in list index 0.
2348 * - In the case that the rotation requires a post-insert update,
2349 * *ret_left_path will contain a valid path which can be passed to
2350 * ocfs2_insert_path().
2351 */
Joel Becker1bbf0b82009-02-12 19:42:08 -08002352static int ocfs2_rotate_tree_right(handle_t *handle,
Joel Becker5c601ab2009-02-12 19:10:13 -08002353 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002354 enum ocfs2_split_type split,
Mark Fashehdcd05382007-01-16 11:32:23 -08002355 u32 insert_cpos,
2356 struct ocfs2_path *right_path,
2357 struct ocfs2_path **ret_left_path)
2358{
Mark Fasheh328d5752007-06-18 10:48:04 -07002359 int ret, start, orig_credits = handle->h_buffer_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002360 u32 cpos;
2361 struct ocfs2_path *left_path = NULL;
Joel Becker5c601ab2009-02-12 19:10:13 -08002362 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehdcd05382007-01-16 11:32:23 -08002363
2364 *ret_left_path = NULL;
2365
Joel Beckerffdd7a52008-10-17 22:32:01 -07002366 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002367 if (!left_path) {
2368 ret = -ENOMEM;
2369 mlog_errno(ret);
2370 goto out;
2371 }
2372
Joel Becker5c601ab2009-02-12 19:10:13 -08002373 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002374 if (ret) {
2375 mlog_errno(ret);
2376 goto out;
2377 }
2378
2379 mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos);
2380
2381 /*
2382 * What we want to do here is:
2383 *
2384 * 1) Start with the rightmost path.
2385 *
2386 * 2) Determine a path to the leaf block directly to the left
2387 * of that leaf.
2388 *
2389 * 3) Determine the 'subtree root' - the lowest level tree node
2390 * which contains a path to both leaves.
2391 *
2392 * 4) Rotate the subtree.
2393 *
2394 * 5) Find the next subtree by considering the left path to be
2395 * the new right path.
2396 *
2397 * The check at the top of this while loop also accepts
2398 * insert_cpos == cpos because cpos is only a _theoretical_
2399 * value to get us the left path - insert_cpos might very well
2400 * be filling that hole.
2401 *
2402 * Stop at a cpos of '0' because we either started at the
2403 * leftmost branch (i.e., a tree with one branch and a
2404 * rotation inside of it), or we've gone as far as we can in
2405 * rotating subtrees.
2406 */
2407 while (cpos && insert_cpos <= cpos) {
2408 mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n",
2409 insert_cpos, cpos);
2410
Joel Becker5c601ab2009-02-12 19:10:13 -08002411 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002412 if (ret) {
2413 mlog_errno(ret);
2414 goto out;
2415 }
2416
2417 mlog_bug_on_msg(path_leaf_bh(left_path) ==
2418 path_leaf_bh(right_path),
Joel Becker5c601ab2009-02-12 19:10:13 -08002419 "Owner %llu: error during insert of %u "
Mark Fashehdcd05382007-01-16 11:32:23 -08002420 "(left path cpos %u) results in two identical "
2421 "paths ending at %llu\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002422 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2423 insert_cpos, cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08002424 (unsigned long long)
2425 path_leaf_bh(left_path)->b_blocknr);
2426
Mark Fasheh328d5752007-06-18 10:48:04 -07002427 if (split == SPLIT_NONE &&
2428 ocfs2_rotate_requires_path_adjustment(left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002429 insert_cpos)) {
Mark Fashehdcd05382007-01-16 11:32:23 -08002430
2431 /*
2432 * We've rotated the tree as much as we
2433 * should. The rest is up to
2434 * ocfs2_insert_path() to complete, after the
2435 * record insertion. We indicate this
2436 * situation by returning the left path.
2437 *
2438 * The reason we don't adjust the records here
2439 * before the record insert is that an error
2440 * later might break the rule where a parent
2441 * record e_cpos will reflect the actual
2442 * e_cpos of the 1st nonempty record of the
2443 * child list.
2444 */
2445 *ret_left_path = left_path;
2446 goto out_ret_path;
2447 }
2448
Joel Becker7dc02802009-02-12 19:20:13 -08002449 start = ocfs2_find_subtree_root(et, left_path, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002450
2451 mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n",
2452 start,
2453 (unsigned long long) right_path->p_node[start].bh->b_blocknr,
2454 right_path->p_tree_depth);
2455
2456 ret = ocfs2_extend_rotate_transaction(handle, start,
Mark Fasheh328d5752007-06-18 10:48:04 -07002457 orig_credits, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002458 if (ret) {
2459 mlog_errno(ret);
2460 goto out;
2461 }
2462
Joel Becker5c601ab2009-02-12 19:10:13 -08002463 ret = ocfs2_rotate_subtree_right(handle, et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002464 right_path, start);
2465 if (ret) {
2466 mlog_errno(ret);
2467 goto out;
2468 }
2469
Mark Fasheh328d5752007-06-18 10:48:04 -07002470 if (split != SPLIT_NONE &&
2471 ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
2472 insert_cpos)) {
2473 /*
2474 * A rotate moves the rightmost left leaf
2475 * record over to the leftmost right leaf
2476 * slot. If we're doing an extent split
2477 * instead of a real insert, then we have to
2478 * check that the extent to be split wasn't
2479 * just moved over. If it was, then we can
2480 * exit here, passing left_path back -
2481 * ocfs2_split_extent() is smart enough to
2482 * search both leaves.
2483 */
2484 *ret_left_path = left_path;
2485 goto out_ret_path;
2486 }
2487
Mark Fashehdcd05382007-01-16 11:32:23 -08002488 /*
2489 * There is no need to re-read the next right path
2490 * as we know that it'll be our current left
2491 * path. Optimize by copying values instead.
2492 */
2493 ocfs2_mv_path(right_path, left_path);
2494
Joel Becker5c601ab2009-02-12 19:10:13 -08002495 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002496 if (ret) {
2497 mlog_errno(ret);
2498 goto out;
2499 }
2500 }
2501
2502out:
2503 ocfs2_free_path(left_path);
2504
2505out_ret_path:
2506 return ret;
2507}
2508
Joel Becker09106ba2009-02-12 19:43:57 -08002509static int ocfs2_update_edge_lengths(handle_t *handle,
2510 struct ocfs2_extent_tree *et,
Tao Ma3c5e1062009-07-21 15:42:05 +08002511 int subtree_index, struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002512{
Tao Ma3c5e1062009-07-21 15:42:05 +08002513 int i, idx, ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002514 struct ocfs2_extent_rec *rec;
2515 struct ocfs2_extent_list *el;
2516 struct ocfs2_extent_block *eb;
2517 u32 range;
2518
Tao Ma3c5e1062009-07-21 15:42:05 +08002519 /*
2520 * In normal tree rotation process, we will never touch the
2521 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
2522 * doesn't reserve the credits for them either.
2523 *
2524 * But we do have a special case here which will update the rightmost
2525 * records for all the bh in the path.
2526 * So we have to allocate extra credits and access them.
2527 */
2528 ret = ocfs2_extend_trans(handle,
2529 handle->h_buffer_credits + subtree_index);
2530 if (ret) {
2531 mlog_errno(ret);
2532 goto out;
2533 }
2534
Joel Becker09106ba2009-02-12 19:43:57 -08002535 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002536 if (ret) {
2537 mlog_errno(ret);
2538 goto out;
2539 }
2540
Mark Fasheh328d5752007-06-18 10:48:04 -07002541 /* Path should always be rightmost. */
2542 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2543 BUG_ON(eb->h_next_leaf_blk != 0ULL);
2544
2545 el = &eb->h_list;
2546 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2547 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2548 rec = &el->l_recs[idx];
2549 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2550
2551 for (i = 0; i < path->p_tree_depth; i++) {
2552 el = path->p_node[i].el;
2553 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2554 rec = &el->l_recs[idx];
2555
2556 rec->e_int_clusters = cpu_to_le32(range);
2557 le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2558
2559 ocfs2_journal_dirty(handle, path->p_node[i].bh);
2560 }
Tao Ma3c5e1062009-07-21 15:42:05 +08002561out:
2562 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002563}
2564
Joel Becker6641b0c2009-02-12 18:57:52 -08002565static void ocfs2_unlink_path(handle_t *handle,
2566 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002567 struct ocfs2_cached_dealloc_ctxt *dealloc,
2568 struct ocfs2_path *path, int unlink_start)
2569{
2570 int ret, i;
2571 struct ocfs2_extent_block *eb;
2572 struct ocfs2_extent_list *el;
2573 struct buffer_head *bh;
2574
2575 for(i = unlink_start; i < path_num_items(path); i++) {
2576 bh = path->p_node[i].bh;
2577
2578 eb = (struct ocfs2_extent_block *)bh->b_data;
2579 /*
2580 * Not all nodes might have had their final count
2581 * decremented by the caller - handle this here.
2582 */
2583 el = &eb->h_list;
2584 if (le16_to_cpu(el->l_next_free_rec) > 1) {
2585 mlog(ML_ERROR,
2586 "Inode %llu, attempted to remove extent block "
2587 "%llu with %u records\n",
Joel Becker6641b0c2009-02-12 18:57:52 -08002588 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07002589 (unsigned long long)le64_to_cpu(eb->h_blkno),
2590 le16_to_cpu(el->l_next_free_rec));
2591
2592 ocfs2_journal_dirty(handle, bh);
Joel Becker6641b0c2009-02-12 18:57:52 -08002593 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002594 continue;
2595 }
2596
2597 el->l_next_free_rec = 0;
2598 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2599
2600 ocfs2_journal_dirty(handle, bh);
2601
2602 ret = ocfs2_cache_extent_block_free(dealloc, eb);
2603 if (ret)
2604 mlog_errno(ret);
2605
Joel Becker6641b0c2009-02-12 18:57:52 -08002606 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002607 }
2608}
2609
Joel Becker6641b0c2009-02-12 18:57:52 -08002610static void ocfs2_unlink_subtree(handle_t *handle,
2611 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002612 struct ocfs2_path *left_path,
2613 struct ocfs2_path *right_path,
2614 int subtree_index,
2615 struct ocfs2_cached_dealloc_ctxt *dealloc)
2616{
2617 int i;
2618 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2619 struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
2620 struct ocfs2_extent_list *el;
2621 struct ocfs2_extent_block *eb;
2622
2623 el = path_leaf_el(left_path);
2624
2625 eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2626
2627 for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2628 if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2629 break;
2630
2631 BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2632
2633 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2634 le16_add_cpu(&root_el->l_next_free_rec, -1);
2635
2636 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2637 eb->h_next_leaf_blk = 0;
2638
2639 ocfs2_journal_dirty(handle, root_bh);
2640 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2641
Joel Becker6641b0c2009-02-12 18:57:52 -08002642 ocfs2_unlink_path(handle, et, dealloc, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002643 subtree_index + 1);
2644}
2645
Joel Becker1e2dd632009-02-12 19:45:28 -08002646static int ocfs2_rotate_subtree_left(handle_t *handle,
2647 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002648 struct ocfs2_path *left_path,
2649 struct ocfs2_path *right_path,
2650 int subtree_index,
2651 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker1e2dd632009-02-12 19:45:28 -08002652 int *deleted)
Mark Fasheh328d5752007-06-18 10:48:04 -07002653{
2654 int ret, i, del_right_subtree = 0, right_has_empty = 0;
Tao Mae7d4cb62008-08-18 17:38:44 +08002655 struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002656 struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2657 struct ocfs2_extent_block *eb;
2658
2659 *deleted = 0;
2660
2661 right_leaf_el = path_leaf_el(right_path);
2662 left_leaf_el = path_leaf_el(left_path);
2663 root_bh = left_path->p_node[subtree_index].bh;
2664 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2665
2666 if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2667 return 0;
2668
2669 eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2670 if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2671 /*
2672 * It's legal for us to proceed if the right leaf is
2673 * the rightmost one and it has an empty extent. There
2674 * are two cases to handle - whether the leaf will be
2675 * empty after removal or not. If the leaf isn't empty
2676 * then just remove the empty extent up front. The
2677 * next block will handle empty leaves by flagging
2678 * them for unlink.
2679 *
2680 * Non rightmost leaves will throw -EAGAIN and the
2681 * caller can manually move the subtree and retry.
2682 */
2683
2684 if (eb->h_next_leaf_blk != 0ULL)
2685 return -EAGAIN;
2686
2687 if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002688 ret = ocfs2_journal_access_eb(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002689 path_leaf_bh(right_path),
2690 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002691 if (ret) {
2692 mlog_errno(ret);
2693 goto out;
2694 }
2695
2696 ocfs2_remove_empty_extent(right_leaf_el);
2697 } else
2698 right_has_empty = 1;
2699 }
2700
2701 if (eb->h_next_leaf_blk == 0ULL &&
2702 le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2703 /*
2704 * We have to update i_last_eb_blk during the meta
2705 * data delete.
2706 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08002707 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07002708 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002709 if (ret) {
2710 mlog_errno(ret);
2711 goto out;
2712 }
2713
2714 del_right_subtree = 1;
2715 }
2716
2717 /*
2718 * Getting here with an empty extent in the right path implies
2719 * that it's the rightmost path and will be deleted.
2720 */
2721 BUG_ON(right_has_empty && !del_right_subtree);
2722
Joel Becker1e2dd632009-02-12 19:45:28 -08002723 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002724 subtree_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07002725 if (ret) {
2726 mlog_errno(ret);
2727 goto out;
2728 }
2729
2730 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002731 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002732 right_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002733 if (ret) {
2734 mlog_errno(ret);
2735 goto out;
2736 }
2737
Joel Becker1e2dd632009-02-12 19:45:28 -08002738 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002739 left_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002740 if (ret) {
2741 mlog_errno(ret);
2742 goto out;
2743 }
2744 }
2745
2746 if (!right_has_empty) {
2747 /*
2748 * Only do this if we're moving a real
2749 * record. Otherwise, the action is delayed until
2750 * after removal of the right path in which case we
2751 * can do a simple shift to remove the empty extent.
2752 */
2753 ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2754 memset(&right_leaf_el->l_recs[0], 0,
2755 sizeof(struct ocfs2_extent_rec));
2756 }
2757 if (eb->h_next_leaf_blk == 0ULL) {
2758 /*
2759 * Move recs over to get rid of empty extent, decrease
2760 * next_free. This is allowed to remove the last
2761 * extent in our leaf (setting l_next_free_rec to
2762 * zero) - the delete code below won't care.
2763 */
2764 ocfs2_remove_empty_extent(right_leaf_el);
2765 }
2766
2767 ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2768 if (ret)
2769 mlog_errno(ret);
2770 ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
2771 if (ret)
2772 mlog_errno(ret);
2773
2774 if (del_right_subtree) {
Joel Becker6641b0c2009-02-12 18:57:52 -08002775 ocfs2_unlink_subtree(handle, et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002776 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08002777 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08002778 left_path);
2779 if (ret) {
2780 mlog_errno(ret);
2781 goto out;
2782 }
Mark Fasheh328d5752007-06-18 10:48:04 -07002783
2784 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07002785 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07002786
2787 /*
2788 * Removal of the extent in the left leaf was skipped
2789 * above so we could delete the right path
2790 * 1st.
2791 */
2792 if (right_has_empty)
2793 ocfs2_remove_empty_extent(left_leaf_el);
2794
Tao Mae7d4cb62008-08-18 17:38:44 +08002795 ret = ocfs2_journal_dirty(handle, et_root_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002796 if (ret)
2797 mlog_errno(ret);
2798
2799 *deleted = 1;
2800 } else
Joel Becker4619c732009-02-12 19:02:36 -08002801 ocfs2_complete_edge_insert(handle, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002802 subtree_index);
2803
2804out:
2805 return ret;
2806}
2807
2808/*
2809 * Given a full path, determine what cpos value would return us a path
2810 * containing the leaf immediately to the right of the current one.
2811 *
2812 * Will return zero if the path passed in is already the rightmost path.
2813 *
2814 * This looks similar, but is subtly different to
2815 * ocfs2_find_cpos_for_left_leaf().
2816 */
2817static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2818 struct ocfs2_path *path, u32 *cpos)
2819{
2820 int i, j, ret = 0;
2821 u64 blkno;
2822 struct ocfs2_extent_list *el;
2823
2824 *cpos = 0;
2825
2826 if (path->p_tree_depth == 0)
2827 return 0;
2828
2829 blkno = path_leaf_bh(path)->b_blocknr;
2830
2831 /* Start at the tree node just above the leaf and work our way up. */
2832 i = path->p_tree_depth - 1;
2833 while (i >= 0) {
2834 int next_free;
2835
2836 el = path->p_node[i].el;
2837
2838 /*
2839 * Find the extent record just after the one in our
2840 * path.
2841 */
2842 next_free = le16_to_cpu(el->l_next_free_rec);
2843 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2844 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2845 if (j == (next_free - 1)) {
2846 if (i == 0) {
2847 /*
2848 * We've determined that the
2849 * path specified is already
2850 * the rightmost one - return a
2851 * cpos of zero.
2852 */
2853 goto out;
2854 }
2855 /*
2856 * The rightmost record points to our
2857 * leaf - we need to travel up the
2858 * tree one level.
2859 */
2860 goto next_node;
2861 }
2862
2863 *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2864 goto out;
2865 }
2866 }
2867
2868 /*
2869 * If we got here, we never found a valid node where
2870 * the tree indicated one should be.
2871 */
2872 ocfs2_error(sb,
2873 "Invalid extent tree at extent block %llu\n",
2874 (unsigned long long)blkno);
2875 ret = -EROFS;
2876 goto out;
2877
2878next_node:
2879 blkno = path->p_node[i].bh->b_blocknr;
2880 i--;
2881 }
2882
2883out:
2884 return ret;
2885}
2886
Joel Becker70f18c02009-02-13 02:09:31 -08002887static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2888 struct ocfs2_extent_tree *et,
Joel Becker13723d02008-10-17 19:25:01 -07002889 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002890{
2891 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07002892 struct buffer_head *bh = path_leaf_bh(path);
2893 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002894
2895 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2896 return 0;
2897
Joel Becker70f18c02009-02-13 02:09:31 -08002898 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Joel Becker13723d02008-10-17 19:25:01 -07002899 path_num_items(path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07002900 if (ret) {
2901 mlog_errno(ret);
2902 goto out;
2903 }
2904
2905 ocfs2_remove_empty_extent(el);
2906
2907 ret = ocfs2_journal_dirty(handle, bh);
2908 if (ret)
2909 mlog_errno(ret);
2910
2911out:
2912 return ret;
2913}
2914
Joel Beckere46f74d2009-02-12 19:47:43 -08002915static int __ocfs2_rotate_tree_left(handle_t *handle,
2916 struct ocfs2_extent_tree *et,
2917 int orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07002918 struct ocfs2_path *path,
2919 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08002920 struct ocfs2_path **empty_extent_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002921{
2922 int ret, subtree_root, deleted;
2923 u32 right_cpos;
2924 struct ocfs2_path *left_path = NULL;
2925 struct ocfs2_path *right_path = NULL;
Joel Beckere46f74d2009-02-12 19:47:43 -08002926 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fasheh328d5752007-06-18 10:48:04 -07002927
2928 BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])));
2929
2930 *empty_extent_path = NULL;
2931
Joel Beckere46f74d2009-02-12 19:47:43 -08002932 ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002933 if (ret) {
2934 mlog_errno(ret);
2935 goto out;
2936 }
2937
Joel Beckerffdd7a52008-10-17 22:32:01 -07002938 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002939 if (!left_path) {
2940 ret = -ENOMEM;
2941 mlog_errno(ret);
2942 goto out;
2943 }
2944
2945 ocfs2_cp_path(left_path, path);
2946
Joel Beckerffdd7a52008-10-17 22:32:01 -07002947 right_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002948 if (!right_path) {
2949 ret = -ENOMEM;
2950 mlog_errno(ret);
2951 goto out;
2952 }
2953
2954 while (right_cpos) {
Joel Beckerfacdb772009-02-12 18:08:48 -08002955 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002956 if (ret) {
2957 mlog_errno(ret);
2958 goto out;
2959 }
2960
Joel Becker7dc02802009-02-12 19:20:13 -08002961 subtree_root = ocfs2_find_subtree_root(et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002962 right_path);
2963
2964 mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n",
2965 subtree_root,
2966 (unsigned long long)
2967 right_path->p_node[subtree_root].bh->b_blocknr,
2968 right_path->p_tree_depth);
2969
2970 ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
2971 orig_credits, left_path);
2972 if (ret) {
2973 mlog_errno(ret);
2974 goto out;
2975 }
2976
Mark Fashehe8aed342007-12-03 16:43:01 -08002977 /*
2978 * Caller might still want to make changes to the
2979 * tree root, so re-add it to the journal here.
2980 */
Joel Beckere46f74d2009-02-12 19:47:43 -08002981 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002982 left_path, 0);
Mark Fashehe8aed342007-12-03 16:43:01 -08002983 if (ret) {
2984 mlog_errno(ret);
2985 goto out;
2986 }
2987
Joel Becker1e2dd632009-02-12 19:45:28 -08002988 ret = ocfs2_rotate_subtree_left(handle, et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002989 right_path, subtree_root,
Joel Becker1e2dd632009-02-12 19:45:28 -08002990 dealloc, &deleted);
Mark Fasheh328d5752007-06-18 10:48:04 -07002991 if (ret == -EAGAIN) {
2992 /*
2993 * The rotation has to temporarily stop due to
2994 * the right subtree having an empty
2995 * extent. Pass it back to the caller for a
2996 * fixup.
2997 */
2998 *empty_extent_path = right_path;
2999 right_path = NULL;
3000 goto out;
3001 }
3002 if (ret) {
3003 mlog_errno(ret);
3004 goto out;
3005 }
3006
3007 /*
3008 * The subtree rotate might have removed records on
3009 * the rightmost edge. If so, then rotation is
3010 * complete.
3011 */
3012 if (deleted)
3013 break;
3014
3015 ocfs2_mv_path(left_path, right_path);
3016
Joel Beckere46f74d2009-02-12 19:47:43 -08003017 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003018 &right_cpos);
3019 if (ret) {
3020 mlog_errno(ret);
3021 goto out;
3022 }
3023 }
3024
3025out:
3026 ocfs2_free_path(right_path);
3027 ocfs2_free_path(left_path);
3028
3029 return ret;
3030}
3031
Joel Becker70f18c02009-02-13 02:09:31 -08003032static int ocfs2_remove_rightmost_path(handle_t *handle,
3033 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003034 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003035 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003036{
3037 int ret, subtree_index;
3038 u32 cpos;
3039 struct ocfs2_path *left_path = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003040 struct ocfs2_extent_block *eb;
3041 struct ocfs2_extent_list *el;
3042
Mark Fasheh328d5752007-06-18 10:48:04 -07003043
Joel Becker6136ca52009-02-12 19:32:43 -08003044 ret = ocfs2_et_sanity_check(et);
Tao Mae7d4cb62008-08-18 17:38:44 +08003045 if (ret)
3046 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003047 /*
3048 * There's two ways we handle this depending on
3049 * whether path is the only existing one.
3050 */
3051 ret = ocfs2_extend_rotate_transaction(handle, 0,
3052 handle->h_buffer_credits,
3053 path);
3054 if (ret) {
3055 mlog_errno(ret);
3056 goto out;
3057 }
3058
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003059 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003060 if (ret) {
3061 mlog_errno(ret);
3062 goto out;
3063 }
3064
Joel Becker3d03a302009-02-12 17:49:26 -08003065 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3066 path, &cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003067 if (ret) {
3068 mlog_errno(ret);
3069 goto out;
3070 }
3071
3072 if (cpos) {
3073 /*
3074 * We have a path to the left of this one - it needs
3075 * an update too.
3076 */
Joel Beckerffdd7a52008-10-17 22:32:01 -07003077 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003078 if (!left_path) {
3079 ret = -ENOMEM;
3080 mlog_errno(ret);
3081 goto out;
3082 }
3083
Joel Beckerfacdb772009-02-12 18:08:48 -08003084 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003085 if (ret) {
3086 mlog_errno(ret);
3087 goto out;
3088 }
3089
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003090 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003091 if (ret) {
3092 mlog_errno(ret);
3093 goto out;
3094 }
3095
Joel Becker7dc02802009-02-12 19:20:13 -08003096 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003097
Joel Becker6641b0c2009-02-12 18:57:52 -08003098 ocfs2_unlink_subtree(handle, et, left_path, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003099 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08003100 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08003101 left_path);
3102 if (ret) {
3103 mlog_errno(ret);
3104 goto out;
3105 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003106
3107 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07003108 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003109 } else {
3110 /*
3111 * 'path' is also the leftmost path which
3112 * means it must be the only one. This gets
3113 * handled differently because we want to
Joel Becker70f18c02009-02-13 02:09:31 -08003114 * revert the root back to having extents
Mark Fasheh328d5752007-06-18 10:48:04 -07003115 * in-line.
3116 */
Joel Becker6641b0c2009-02-12 18:57:52 -08003117 ocfs2_unlink_path(handle, et, dealloc, path, 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003118
Joel Beckerce1d9ea2008-08-20 16:30:07 -07003119 el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003120 el->l_tree_depth = 0;
3121 el->l_next_free_rec = 0;
3122 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3123
Joel Becker35dc0aa2008-08-20 16:25:06 -07003124 ocfs2_et_set_last_eb_blk(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003125 }
3126
3127 ocfs2_journal_dirty(handle, path_root_bh(path));
3128
3129out:
3130 ocfs2_free_path(left_path);
3131 return ret;
3132}
3133
3134/*
3135 * Left rotation of btree records.
3136 *
3137 * In many ways, this is (unsurprisingly) the opposite of right
3138 * rotation. We start at some non-rightmost path containing an empty
3139 * extent in the leaf block. The code works its way to the rightmost
3140 * path by rotating records to the left in every subtree.
3141 *
3142 * This is used by any code which reduces the number of extent records
3143 * in a leaf. After removal, an empty record should be placed in the
3144 * leftmost list position.
3145 *
3146 * This won't handle a length update of the rightmost path records if
3147 * the rightmost tree leaf record is removed so the caller is
3148 * responsible for detecting and correcting that.
3149 */
Joel Becker70f18c02009-02-13 02:09:31 -08003150static int ocfs2_rotate_tree_left(handle_t *handle,
3151 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003152 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003153 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003154{
3155 int ret, orig_credits = handle->h_buffer_credits;
3156 struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3157 struct ocfs2_extent_block *eb;
3158 struct ocfs2_extent_list *el;
3159
3160 el = path_leaf_el(path);
3161 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3162 return 0;
3163
3164 if (path->p_tree_depth == 0) {
3165rightmost_no_delete:
3166 /*
Tao Mae7d4cb62008-08-18 17:38:44 +08003167 * Inline extents. This is trivially handled, so do
Mark Fasheh328d5752007-06-18 10:48:04 -07003168 * it up front.
3169 */
Joel Becker70f18c02009-02-13 02:09:31 -08003170 ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003171 if (ret)
3172 mlog_errno(ret);
3173 goto out;
3174 }
3175
3176 /*
3177 * Handle rightmost branch now. There's several cases:
3178 * 1) simple rotation leaving records in there. That's trivial.
3179 * 2) rotation requiring a branch delete - there's no more
3180 * records left. Two cases of this:
3181 * a) There are branches to the left.
3182 * b) This is also the leftmost (the only) branch.
3183 *
3184 * 1) is handled via ocfs2_rotate_rightmost_leaf_left()
3185 * 2a) we need the left branch so that we can update it with the unlink
Joel Becker70f18c02009-02-13 02:09:31 -08003186 * 2b) we need to bring the root back to inline extents.
Mark Fasheh328d5752007-06-18 10:48:04 -07003187 */
3188
3189 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3190 el = &eb->h_list;
3191 if (eb->h_next_leaf_blk == 0) {
3192 /*
3193 * This gets a bit tricky if we're going to delete the
3194 * rightmost path. Get the other cases out of the way
3195 * 1st.
3196 */
3197 if (le16_to_cpu(el->l_next_free_rec) > 1)
3198 goto rightmost_no_delete;
3199
3200 if (le16_to_cpu(el->l_next_free_rec) == 0) {
3201 ret = -EIO;
Joel Becker70f18c02009-02-13 02:09:31 -08003202 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3203 "Owner %llu has empty extent block at %llu",
3204 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07003205 (unsigned long long)le64_to_cpu(eb->h_blkno));
3206 goto out;
3207 }
3208
3209 /*
3210 * XXX: The caller can not trust "path" any more after
3211 * this as it will have been deleted. What do we do?
3212 *
3213 * In theory the rotate-for-merge code will never get
3214 * here because it'll always ask for a rotate in a
3215 * nonempty list.
3216 */
3217
Joel Becker70f18c02009-02-13 02:09:31 -08003218 ret = ocfs2_remove_rightmost_path(handle, et, path,
3219 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003220 if (ret)
3221 mlog_errno(ret);
3222 goto out;
3223 }
3224
3225 /*
3226 * Now we can loop, remembering the path we get from -EAGAIN
3227 * and restarting from there.
3228 */
3229try_rotate:
Joel Beckere46f74d2009-02-12 19:47:43 -08003230 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3231 dealloc, &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003232 if (ret && ret != -EAGAIN) {
3233 mlog_errno(ret);
3234 goto out;
3235 }
3236
3237 while (ret == -EAGAIN) {
3238 tmp_path = restart_path;
3239 restart_path = NULL;
3240
Joel Beckere46f74d2009-02-12 19:47:43 -08003241 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07003242 tmp_path, dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08003243 &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003244 if (ret && ret != -EAGAIN) {
3245 mlog_errno(ret);
3246 goto out;
3247 }
3248
3249 ocfs2_free_path(tmp_path);
3250 tmp_path = NULL;
3251
3252 if (ret == 0)
3253 goto try_rotate;
3254 }
3255
3256out:
3257 ocfs2_free_path(tmp_path);
3258 ocfs2_free_path(restart_path);
3259 return ret;
3260}
3261
3262static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3263 int index)
3264{
3265 struct ocfs2_extent_rec *rec = &el->l_recs[index];
3266 unsigned int size;
3267
3268 if (rec->e_leaf_clusters == 0) {
3269 /*
3270 * We consumed all of the merged-from record. An empty
3271 * extent cannot exist anywhere but the 1st array
3272 * position, so move things over if the merged-from
3273 * record doesn't occupy that position.
3274 *
3275 * This creates a new empty extent so the caller
3276 * should be smart enough to have removed any existing
3277 * ones.
3278 */
3279 if (index > 0) {
3280 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3281 size = index * sizeof(struct ocfs2_extent_rec);
3282 memmove(&el->l_recs[1], &el->l_recs[0], size);
3283 }
3284
3285 /*
3286 * Always memset - the caller doesn't check whether it
3287 * created an empty extent, so there could be junk in
3288 * the other fields.
3289 */
3290 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3291 }
3292}
3293
Joel Becker4fe82c32009-02-13 02:16:08 -08003294static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003295 struct ocfs2_path *left_path,
3296 struct ocfs2_path **ret_right_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07003297{
3298 int ret;
Tao Ma677b9752008-01-30 14:21:05 +08003299 u32 right_cpos;
3300 struct ocfs2_path *right_path = NULL;
3301 struct ocfs2_extent_list *left_el;
3302
3303 *ret_right_path = NULL;
3304
3305 /* This function shouldn't be called for non-trees. */
3306 BUG_ON(left_path->p_tree_depth == 0);
3307
3308 left_el = path_leaf_el(left_path);
3309 BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3310
Joel Becker4fe82c32009-02-13 02:16:08 -08003311 ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3312 left_path, &right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003313 if (ret) {
3314 mlog_errno(ret);
3315 goto out;
3316 }
3317
3318 /* This function shouldn't be called for the rightmost leaf. */
3319 BUG_ON(right_cpos == 0);
3320
Joel Beckerffdd7a52008-10-17 22:32:01 -07003321 right_path = ocfs2_new_path_from_path(left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003322 if (!right_path) {
3323 ret = -ENOMEM;
3324 mlog_errno(ret);
3325 goto out;
3326 }
3327
Joel Becker4fe82c32009-02-13 02:16:08 -08003328 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003329 if (ret) {
3330 mlog_errno(ret);
3331 goto out;
3332 }
3333
3334 *ret_right_path = right_path;
3335out:
3336 if (ret)
3337 ocfs2_free_path(right_path);
3338 return ret;
3339}
3340
3341/*
3342 * Remove split_rec clusters from the record at index and merge them
3343 * onto the beginning of the record "next" to it.
3344 * For index < l_count - 1, the next means the extent rec at index + 1.
3345 * For index == l_count - 1, the "next" means the 1st extent rec of the
3346 * next extent block.
3347 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003348static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003349 handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003350 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003351 struct ocfs2_extent_rec *split_rec,
3352 int index)
3353{
3354 int ret, next_free, i;
Mark Fasheh328d5752007-06-18 10:48:04 -07003355 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3356 struct ocfs2_extent_rec *left_rec;
3357 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003358 struct ocfs2_extent_list *right_el;
3359 struct ocfs2_path *right_path = NULL;
3360 int subtree_index = 0;
3361 struct ocfs2_extent_list *el = path_leaf_el(left_path);
3362 struct buffer_head *bh = path_leaf_bh(left_path);
3363 struct buffer_head *root_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003364
3365 BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
Mark Fasheh328d5752007-06-18 10:48:04 -07003366 left_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003367
Al Viro9d8df6a2008-05-21 06:32:11 +01003368 if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
Tao Ma677b9752008-01-30 14:21:05 +08003369 le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3370 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003371 ret = ocfs2_get_right_path(et, left_path, &right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003372 if (ret) {
3373 mlog_errno(ret);
3374 goto out;
3375 }
3376
3377 right_el = path_leaf_el(right_path);
3378 next_free = le16_to_cpu(right_el->l_next_free_rec);
3379 BUG_ON(next_free <= 0);
3380 right_rec = &right_el->l_recs[0];
3381 if (ocfs2_is_empty_extent(right_rec)) {
Al Viro9d8df6a2008-05-21 06:32:11 +01003382 BUG_ON(next_free <= 1);
Tao Ma677b9752008-01-30 14:21:05 +08003383 right_rec = &right_el->l_recs[1];
3384 }
3385
3386 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3387 le16_to_cpu(left_rec->e_leaf_clusters) !=
3388 le32_to_cpu(right_rec->e_cpos));
3389
Joel Becker7dc02802009-02-12 19:20:13 -08003390 subtree_index = ocfs2_find_subtree_root(et, left_path,
3391 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003392
3393 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3394 handle->h_buffer_credits,
3395 right_path);
3396 if (ret) {
3397 mlog_errno(ret);
3398 goto out;
3399 }
3400
3401 root_bh = left_path->p_node[subtree_index].bh;
3402 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3403
Joel Becker7dc02802009-02-12 19:20:13 -08003404 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003405 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003406 if (ret) {
3407 mlog_errno(ret);
3408 goto out;
3409 }
3410
3411 for (i = subtree_index + 1;
3412 i < path_num_items(right_path); i++) {
Joel Becker7dc02802009-02-12 19:20:13 -08003413 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003414 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003415 if (ret) {
3416 mlog_errno(ret);
3417 goto out;
3418 }
3419
Joel Becker7dc02802009-02-12 19:20:13 -08003420 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003421 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003422 if (ret) {
3423 mlog_errno(ret);
3424 goto out;
3425 }
3426 }
3427
3428 } else {
3429 BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3430 right_rec = &el->l_recs[index + 1];
3431 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003432
Joel Becker7dc02802009-02-12 19:20:13 -08003433 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
Joel Becker13723d02008-10-17 19:25:01 -07003434 path_num_items(left_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003435 if (ret) {
3436 mlog_errno(ret);
3437 goto out;
3438 }
3439
3440 le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3441
3442 le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3443 le64_add_cpu(&right_rec->e_blkno,
Joel Becker7dc02802009-02-12 19:20:13 -08003444 -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3445 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003446 le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3447
3448 ocfs2_cleanup_merge(el, index);
3449
3450 ret = ocfs2_journal_dirty(handle, bh);
3451 if (ret)
3452 mlog_errno(ret);
3453
Tao Ma677b9752008-01-30 14:21:05 +08003454 if (right_path) {
3455 ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
3456 if (ret)
3457 mlog_errno(ret);
3458
Joel Becker4619c732009-02-12 19:02:36 -08003459 ocfs2_complete_edge_insert(handle, left_path, right_path,
3460 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003461 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003462out:
Tao Ma677b9752008-01-30 14:21:05 +08003463 if (right_path)
3464 ocfs2_free_path(right_path);
3465 return ret;
3466}
3467
Joel Becker4fe82c32009-02-13 02:16:08 -08003468static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003469 struct ocfs2_path *right_path,
3470 struct ocfs2_path **ret_left_path)
3471{
3472 int ret;
3473 u32 left_cpos;
3474 struct ocfs2_path *left_path = NULL;
3475
3476 *ret_left_path = NULL;
3477
3478 /* This function shouldn't be called for non-trees. */
3479 BUG_ON(right_path->p_tree_depth == 0);
3480
Joel Becker4fe82c32009-02-13 02:16:08 -08003481 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003482 right_path, &left_cpos);
3483 if (ret) {
3484 mlog_errno(ret);
3485 goto out;
3486 }
3487
3488 /* This function shouldn't be called for the leftmost leaf. */
3489 BUG_ON(left_cpos == 0);
3490
Joel Beckerffdd7a52008-10-17 22:32:01 -07003491 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003492 if (!left_path) {
3493 ret = -ENOMEM;
3494 mlog_errno(ret);
3495 goto out;
3496 }
3497
Joel Becker4fe82c32009-02-13 02:16:08 -08003498 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003499 if (ret) {
3500 mlog_errno(ret);
3501 goto out;
3502 }
3503
3504 *ret_left_path = left_path;
3505out:
3506 if (ret)
3507 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003508 return ret;
3509}
3510
3511/*
3512 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003513 * onto the tail of the record "before" it.
3514 * For index > 0, the "before" means the extent rec at index - 1.
3515 *
3516 * For index == 0, the "before" means the last record of the previous
3517 * extent block. And there is also a situation that we may need to
3518 * remove the rightmost leaf extent block in the right_path and change
3519 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003520 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003521static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003522 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003523 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003524 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003525 struct ocfs2_cached_dealloc_ctxt *dealloc,
3526 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003527{
Tao Ma677b9752008-01-30 14:21:05 +08003528 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003529 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3530 struct ocfs2_extent_rec *left_rec;
3531 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003532 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3533 struct buffer_head *bh = path_leaf_bh(right_path);
3534 struct buffer_head *root_bh = NULL;
3535 struct ocfs2_path *left_path = NULL;
3536 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003537
Tao Ma677b9752008-01-30 14:21:05 +08003538 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003539
Mark Fasheh328d5752007-06-18 10:48:04 -07003540 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003541 if (index == 0) {
3542 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003543 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003544 if (ret) {
3545 mlog_errno(ret);
3546 goto out;
3547 }
3548
3549 left_el = path_leaf_el(left_path);
3550 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3551 le16_to_cpu(left_el->l_count));
3552
3553 left_rec = &left_el->l_recs[
3554 le16_to_cpu(left_el->l_next_free_rec) - 1];
3555 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3556 le16_to_cpu(left_rec->e_leaf_clusters) !=
3557 le32_to_cpu(split_rec->e_cpos));
3558
Joel Becker7dc02802009-02-12 19:20:13 -08003559 subtree_index = ocfs2_find_subtree_root(et, left_path,
3560 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003561
3562 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3563 handle->h_buffer_credits,
3564 left_path);
3565 if (ret) {
3566 mlog_errno(ret);
3567 goto out;
3568 }
3569
3570 root_bh = left_path->p_node[subtree_index].bh;
3571 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3572
Joel Becker4fe82c32009-02-13 02:16:08 -08003573 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003574 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003575 if (ret) {
3576 mlog_errno(ret);
3577 goto out;
3578 }
3579
3580 for (i = subtree_index + 1;
3581 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003582 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003583 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003584 if (ret) {
3585 mlog_errno(ret);
3586 goto out;
3587 }
3588
Joel Becker4fe82c32009-02-13 02:16:08 -08003589 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003590 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003591 if (ret) {
3592 mlog_errno(ret);
3593 goto out;
3594 }
3595 }
3596 } else {
3597 left_rec = &el->l_recs[index - 1];
3598 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3599 has_empty_extent = 1;
3600 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003601
Joel Becker4fe82c32009-02-13 02:16:08 -08003602 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003603 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003604 if (ret) {
3605 mlog_errno(ret);
3606 goto out;
3607 }
3608
3609 if (has_empty_extent && index == 1) {
3610 /*
3611 * The easy case - we can just plop the record right in.
3612 */
3613 *left_rec = *split_rec;
3614
3615 has_empty_extent = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003616 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003617 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003618
3619 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3620 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003621 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3622 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003623 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3624
3625 ocfs2_cleanup_merge(el, index);
3626
3627 ret = ocfs2_journal_dirty(handle, bh);
3628 if (ret)
3629 mlog_errno(ret);
3630
Tao Ma677b9752008-01-30 14:21:05 +08003631 if (left_path) {
3632 ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
3633 if (ret)
3634 mlog_errno(ret);
3635
3636 /*
3637 * In the situation that the right_rec is empty and the extent
3638 * block is empty also, ocfs2_complete_edge_insert can't handle
3639 * it and we need to delete the right extent block.
3640 */
3641 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3642 le16_to_cpu(el->l_next_free_rec) == 1) {
3643
Joel Becker70f18c02009-02-13 02:09:31 -08003644 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003645 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003646 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003647 if (ret) {
3648 mlog_errno(ret);
3649 goto out;
3650 }
3651
3652 /* Now the rightmost extent block has been deleted.
3653 * So we use the new rightmost path.
3654 */
3655 ocfs2_mv_path(right_path, left_path);
3656 left_path = NULL;
3657 } else
Joel Becker4619c732009-02-12 19:02:36 -08003658 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003659 right_path, subtree_index);
3660 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003661out:
Tao Ma677b9752008-01-30 14:21:05 +08003662 if (left_path)
3663 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003664 return ret;
3665}
3666
Joel Beckerc495dd22009-02-13 02:19:11 -08003667static int ocfs2_try_to_merge_extent(handle_t *handle,
3668 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003669 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003670 int split_index,
3671 struct ocfs2_extent_rec *split_rec,
3672 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003673 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003674{
Tao Mao518d7262007-08-28 17:25:35 -07003675 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003676 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003677 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3678
3679 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3680
Tao Mao518d7262007-08-28 17:25:35 -07003681 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
3682 /*
3683 * The merge code will need to create an empty
3684 * extent to take the place of the newly
3685 * emptied slot. Remove any pre-existing empty
3686 * extents - having more than one in a leaf is
3687 * illegal.
3688 */
Joel Becker70f18c02009-02-13 02:09:31 -08003689 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003690 if (ret) {
3691 mlog_errno(ret);
3692 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003693 }
Tao Mao518d7262007-08-28 17:25:35 -07003694 split_index--;
3695 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003696 }
3697
3698 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3699 /*
3700 * Left-right contig implies this.
3701 */
3702 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003703
3704 /*
3705 * Since the leftright insert always covers the entire
3706 * extent, this call will delete the insert record
3707 * entirely, resulting in an empty extent record added to
3708 * the extent block.
3709 *
3710 * Since the adding of an empty extent shifts
3711 * everything back to the right, there's no need to
3712 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003713 *
3714 * When the split_index is zero, we need to merge it to the
3715 * prevoius extent block. It is more efficient and easier
3716 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003717 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003718 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003719 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003720 if (ret) {
3721 mlog_errno(ret);
3722 goto out;
3723 }
3724
3725 /*
3726 * We can only get this from logic error above.
3727 */
3728 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3729
Tao Ma677b9752008-01-30 14:21:05 +08003730 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003731 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003732 if (ret) {
3733 mlog_errno(ret);
3734 goto out;
3735 }
Tao Ma677b9752008-01-30 14:21:05 +08003736
Mark Fasheh328d5752007-06-18 10:48:04 -07003737 rec = &el->l_recs[split_index];
3738
3739 /*
3740 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003741 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003742 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003743 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3744 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003745
Mark Fasheh328d5752007-06-18 10:48:04 -07003746 if (ret) {
3747 mlog_errno(ret);
3748 goto out;
3749 }
3750
Joel Becker70f18c02009-02-13 02:09:31 -08003751 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003752 /*
3753 * Error from this last rotate is not critical, so
3754 * print but don't bubble it up.
3755 */
3756 if (ret)
3757 mlog_errno(ret);
3758 ret = 0;
3759 } else {
3760 /*
3761 * Merge a record to the left or right.
3762 *
3763 * 'contig_type' is relative to the existing record,
3764 * so for example, if we're "right contig", it's to
3765 * the record on the left (hence the left merge).
3766 */
3767 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003768 ret = ocfs2_merge_rec_left(path, handle, et,
3769 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003770 split_index);
3771 if (ret) {
3772 mlog_errno(ret);
3773 goto out;
3774 }
3775 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003776 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003777 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003778 split_index);
3779 if (ret) {
3780 mlog_errno(ret);
3781 goto out;
3782 }
3783 }
3784
3785 if (ctxt->c_split_covers_rec) {
3786 /*
3787 * The merge may have left an empty extent in
3788 * our leaf. Try to rotate it away.
3789 */
Joel Becker70f18c02009-02-13 02:09:31 -08003790 ret = ocfs2_rotate_tree_left(handle, et, path,
3791 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003792 if (ret)
3793 mlog_errno(ret);
3794 ret = 0;
3795 }
3796 }
3797
3798out:
3799 return ret;
3800}
3801
3802static void ocfs2_subtract_from_rec(struct super_block *sb,
3803 enum ocfs2_split_type split,
3804 struct ocfs2_extent_rec *rec,
3805 struct ocfs2_extent_rec *split_rec)
3806{
3807 u64 len_blocks;
3808
3809 len_blocks = ocfs2_clusters_to_blocks(sb,
3810 le16_to_cpu(split_rec->e_leaf_clusters));
3811
3812 if (split == SPLIT_LEFT) {
3813 /*
3814 * Region is on the left edge of the existing
3815 * record.
3816 */
3817 le32_add_cpu(&rec->e_cpos,
3818 le16_to_cpu(split_rec->e_leaf_clusters));
3819 le64_add_cpu(&rec->e_blkno, len_blocks);
3820 le16_add_cpu(&rec->e_leaf_clusters,
3821 -le16_to_cpu(split_rec->e_leaf_clusters));
3822 } else {
3823 /*
3824 * Region is on the right edge of the existing
3825 * record.
3826 */
3827 le16_add_cpu(&rec->e_leaf_clusters,
3828 -le16_to_cpu(split_rec->e_leaf_clusters));
3829 }
3830}
3831
Mark Fashehdcd05382007-01-16 11:32:23 -08003832/*
3833 * Do the final bits of extent record insertion at the target leaf
3834 * list. If this leaf is part of an allocation tree, it is assumed
3835 * that the tree above has been prepared.
3836 */
Joel Beckerd5628622009-02-13 02:54:36 -08003837static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3838 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003839 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003840 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003841{
3842 int i = insert->ins_contig_index;
3843 unsigned int range;
3844 struct ocfs2_extent_rec *rec;
3845
Mark Fashehe48edee2007-03-07 16:46:57 -08003846 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003847
Mark Fasheh328d5752007-06-18 10:48:04 -07003848 if (insert->ins_split != SPLIT_NONE) {
3849 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3850 BUG_ON(i == -1);
3851 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003852 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3853 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003854 insert_rec);
3855 goto rotate;
3856 }
3857
Mark Fashehdcd05382007-01-16 11:32:23 -08003858 /*
3859 * Contiguous insert - either left or right.
3860 */
3861 if (insert->ins_contig != CONTIG_NONE) {
3862 rec = &el->l_recs[i];
3863 if (insert->ins_contig == CONTIG_LEFT) {
3864 rec->e_blkno = insert_rec->e_blkno;
3865 rec->e_cpos = insert_rec->e_cpos;
3866 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003867 le16_add_cpu(&rec->e_leaf_clusters,
3868 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003869 return;
3870 }
3871
3872 /*
3873 * Handle insert into an empty leaf.
3874 */
3875 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3876 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3877 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3878 el->l_recs[0] = *insert_rec;
3879 el->l_next_free_rec = cpu_to_le16(1);
3880 return;
3881 }
3882
3883 /*
3884 * Appending insert.
3885 */
3886 if (insert->ins_appending == APPEND_TAIL) {
3887 i = le16_to_cpu(el->l_next_free_rec) - 1;
3888 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003889 range = le32_to_cpu(rec->e_cpos)
3890 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003891 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3892
3893 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3894 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003895 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003896 "rec.cpos %u, rec.clusters %u, "
3897 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003898 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003899 le16_to_cpu(el->l_tree_depth),
3900 le16_to_cpu(el->l_count),
3901 le16_to_cpu(el->l_next_free_rec),
3902 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003903 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003904 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003905 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003906 i++;
3907 el->l_recs[i] = *insert_rec;
3908 le16_add_cpu(&el->l_next_free_rec, 1);
3909 return;
3910 }
3911
Mark Fasheh328d5752007-06-18 10:48:04 -07003912rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003913 /*
3914 * Ok, we have to rotate.
3915 *
3916 * At this point, it is safe to assume that inserting into an
3917 * empty leaf and appending to a leaf have both been handled
3918 * above.
3919 *
3920 * This leaf needs to have space, either by the empty 1st
3921 * extent record, or by virtue of an l_next_rec < l_count.
3922 */
3923 ocfs2_rotate_leaf(el, insert_rec);
3924}
3925
Joel Beckerd401dc12009-02-13 02:24:10 -08003926static void ocfs2_adjust_rightmost_records(handle_t *handle,
3927 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003928 struct ocfs2_path *path,
3929 struct ocfs2_extent_rec *insert_rec)
3930{
3931 int ret, i, next_free;
3932 struct buffer_head *bh;
3933 struct ocfs2_extent_list *el;
3934 struct ocfs2_extent_rec *rec;
3935
3936 /*
3937 * Update everything except the leaf block.
3938 */
3939 for (i = 0; i < path->p_tree_depth; i++) {
3940 bh = path->p_node[i].bh;
3941 el = path->p_node[i].el;
3942
3943 next_free = le16_to_cpu(el->l_next_free_rec);
3944 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003945 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3946 "Owner %llu has a bad extent list",
3947 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003948 ret = -EIO;
3949 return;
3950 }
3951
3952 rec = &el->l_recs[next_free - 1];
3953
3954 rec->e_int_clusters = insert_rec->e_cpos;
3955 le32_add_cpu(&rec->e_int_clusters,
3956 le16_to_cpu(insert_rec->e_leaf_clusters));
3957 le32_add_cpu(&rec->e_int_clusters,
3958 -le32_to_cpu(rec->e_cpos));
3959
3960 ret = ocfs2_journal_dirty(handle, bh);
3961 if (ret)
3962 mlog_errno(ret);
3963
3964 }
3965}
3966
Joel Beckerd401dc12009-02-13 02:24:10 -08003967static int ocfs2_append_rec_to_path(handle_t *handle,
3968 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003969 struct ocfs2_extent_rec *insert_rec,
3970 struct ocfs2_path *right_path,
3971 struct ocfs2_path **ret_left_path)
3972{
Mark Fasheh328d5752007-06-18 10:48:04 -07003973 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003974 struct ocfs2_extent_list *el;
3975 struct ocfs2_path *left_path = NULL;
3976
3977 *ret_left_path = NULL;
3978
3979 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003980 * This shouldn't happen for non-trees. The extent rec cluster
3981 * count manipulation below only works for interior nodes.
3982 */
3983 BUG_ON(right_path->p_tree_depth == 0);
3984
3985 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08003986 * If our appending insert is at the leftmost edge of a leaf,
3987 * then we might need to update the rightmost records of the
3988 * neighboring path.
3989 */
3990 el = path_leaf_el(right_path);
3991 next_free = le16_to_cpu(el->l_next_free_rec);
3992 if (next_free == 0 ||
3993 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
3994 u32 left_cpos;
3995
Joel Beckerd401dc12009-02-13 02:24:10 -08003996 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3997 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003998 if (ret) {
3999 mlog_errno(ret);
4000 goto out;
4001 }
4002
4003 mlog(0, "Append may need a left path update. cpos: %u, "
4004 "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
4005 left_cpos);
4006
4007 /*
4008 * No need to worry if the append is already in the
4009 * leftmost leaf.
4010 */
4011 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004012 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004013 if (!left_path) {
4014 ret = -ENOMEM;
4015 mlog_errno(ret);
4016 goto out;
4017 }
4018
Joel Beckerd401dc12009-02-13 02:24:10 -08004019 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004020 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004021 if (ret) {
4022 mlog_errno(ret);
4023 goto out;
4024 }
4025
4026 /*
4027 * ocfs2_insert_path() will pass the left_path to the
4028 * journal for us.
4029 */
4030 }
4031 }
4032
Joel Beckerd401dc12009-02-13 02:24:10 -08004033 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004034 if (ret) {
4035 mlog_errno(ret);
4036 goto out;
4037 }
4038
Joel Beckerd401dc12009-02-13 02:24:10 -08004039 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004040
4041 *ret_left_path = left_path;
4042 ret = 0;
4043out:
4044 if (ret != 0)
4045 ocfs2_free_path(left_path);
4046
4047 return ret;
4048}
4049
Joel Beckerc38e52b2009-02-13 02:56:23 -08004050static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004051 struct ocfs2_path *left_path,
4052 struct ocfs2_path *right_path,
4053 struct ocfs2_extent_rec *split_rec,
4054 enum ocfs2_split_type split)
4055{
4056 int index;
4057 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4058 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4059 struct ocfs2_extent_rec *rec, *tmprec;
4060
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004061 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004062 if (left_path)
4063 left_el = path_leaf_el(left_path);
4064
4065 el = right_el;
4066 insert_el = right_el;
4067 index = ocfs2_search_extent_list(el, cpos);
4068 if (index != -1) {
4069 if (index == 0 && left_path) {
4070 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4071
4072 /*
4073 * This typically means that the record
4074 * started in the left path but moved to the
4075 * right as a result of rotation. We either
4076 * move the existing record to the left, or we
4077 * do the later insert there.
4078 *
4079 * In this case, the left path should always
4080 * exist as the rotate code will have passed
4081 * it back for a post-insert update.
4082 */
4083
4084 if (split == SPLIT_LEFT) {
4085 /*
4086 * It's a left split. Since we know
4087 * that the rotate code gave us an
4088 * empty extent in the left path, we
4089 * can just do the insert there.
4090 */
4091 insert_el = left_el;
4092 } else {
4093 /*
4094 * Right split - we have to move the
4095 * existing record over to the left
4096 * leaf. The insert will be into the
4097 * newly created empty extent in the
4098 * right leaf.
4099 */
4100 tmprec = &right_el->l_recs[index];
4101 ocfs2_rotate_leaf(left_el, tmprec);
4102 el = left_el;
4103
4104 memset(tmprec, 0, sizeof(*tmprec));
4105 index = ocfs2_search_extent_list(left_el, cpos);
4106 BUG_ON(index == -1);
4107 }
4108 }
4109 } else {
4110 BUG_ON(!left_path);
4111 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4112 /*
4113 * Left path is easy - we can just allow the insert to
4114 * happen.
4115 */
4116 el = left_el;
4117 insert_el = left_el;
4118 index = ocfs2_search_extent_list(el, cpos);
4119 BUG_ON(index == -1);
4120 }
4121
4122 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004123 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4124 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004125 ocfs2_rotate_leaf(insert_el, split_rec);
4126}
4127
Mark Fashehdcd05382007-01-16 11:32:23 -08004128/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004129 * This function only does inserts on an allocation b-tree. For tree
4130 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004131 *
4132 * right_path is the path we want to do the actual insert
4133 * in. left_path should only be passed in if we need to update that
4134 * portion of the tree after an edge insert.
4135 */
Joel Becker3505bec2009-02-13 02:57:58 -08004136static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004137 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004138 struct ocfs2_path *left_path,
4139 struct ocfs2_path *right_path,
4140 struct ocfs2_extent_rec *insert_rec,
4141 struct ocfs2_insert_type *insert)
4142{
4143 int ret, subtree_index;
4144 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004145
Mark Fashehdcd05382007-01-16 11:32:23 -08004146 if (left_path) {
4147 int credits = handle->h_buffer_credits;
4148
4149 /*
4150 * There's a chance that left_path got passed back to
4151 * us without being accounted for in the
4152 * journal. Extend our transaction here to be sure we
4153 * can change those blocks.
4154 */
4155 credits += left_path->p_tree_depth;
4156
4157 ret = ocfs2_extend_trans(handle, credits);
4158 if (ret < 0) {
4159 mlog_errno(ret);
4160 goto out;
4161 }
4162
Joel Becker7dc02802009-02-12 19:20:13 -08004163 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004164 if (ret < 0) {
4165 mlog_errno(ret);
4166 goto out;
4167 }
4168 }
4169
Mark Fashehe8aed342007-12-03 16:43:01 -08004170 /*
4171 * Pass both paths to the journal. The majority of inserts
4172 * will be touching all components anyway.
4173 */
Joel Becker7dc02802009-02-12 19:20:13 -08004174 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004175 if (ret < 0) {
4176 mlog_errno(ret);
4177 goto out;
4178 }
4179
Mark Fasheh328d5752007-06-18 10:48:04 -07004180 if (insert->ins_split != SPLIT_NONE) {
4181 /*
4182 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004183 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004184 * function sort it all out.
4185 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004186 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004187 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004188
4189 /*
4190 * Split might have modified either leaf and we don't
4191 * have a guarantee that the later edge insert will
4192 * dirty this for us.
4193 */
4194 if (left_path)
4195 ret = ocfs2_journal_dirty(handle,
4196 path_leaf_bh(left_path));
4197 if (ret)
4198 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07004199 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004200 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4201 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004202
Mark Fashehdcd05382007-01-16 11:32:23 -08004203 ret = ocfs2_journal_dirty(handle, leaf_bh);
4204 if (ret)
4205 mlog_errno(ret);
4206
4207 if (left_path) {
4208 /*
4209 * The rotate code has indicated that we need to fix
4210 * up portions of the tree after the insert.
4211 *
4212 * XXX: Should we extend the transaction here?
4213 */
Joel Becker7dc02802009-02-12 19:20:13 -08004214 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004215 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004216 ocfs2_complete_edge_insert(handle, left_path, right_path,
4217 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004218 }
4219
4220 ret = 0;
4221out:
4222 return ret;
4223}
4224
Joel Becker3505bec2009-02-13 02:57:58 -08004225static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004226 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004227 struct ocfs2_extent_rec *insert_rec,
4228 struct ocfs2_insert_type *type)
4229{
4230 int ret, rotate = 0;
4231 u32 cpos;
4232 struct ocfs2_path *right_path = NULL;
4233 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004234 struct ocfs2_extent_list *el;
4235
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004236 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004237
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004238 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004239 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004240 if (ret) {
4241 mlog_errno(ret);
4242 goto out;
4243 }
4244
4245 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004246 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004247 goto out_update_clusters;
4248 }
4249
Joel Beckerffdd7a52008-10-17 22:32:01 -07004250 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004251 if (!right_path) {
4252 ret = -ENOMEM;
4253 mlog_errno(ret);
4254 goto out;
4255 }
4256
4257 /*
4258 * Determine the path to start with. Rotations need the
4259 * rightmost path, everything else can go directly to the
4260 * target leaf.
4261 */
4262 cpos = le32_to_cpu(insert_rec->e_cpos);
4263 if (type->ins_appending == APPEND_NONE &&
4264 type->ins_contig == CONTIG_NONE) {
4265 rotate = 1;
4266 cpos = UINT_MAX;
4267 }
4268
Joel Beckerfacdb772009-02-12 18:08:48 -08004269 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004270 if (ret) {
4271 mlog_errno(ret);
4272 goto out;
4273 }
4274
4275 /*
4276 * Rotations and appends need special treatment - they modify
4277 * parts of the tree's above them.
4278 *
4279 * Both might pass back a path immediate to the left of the
4280 * one being inserted to. This will be cause
4281 * ocfs2_insert_path() to modify the rightmost records of
4282 * left_path to account for an edge insert.
4283 *
4284 * XXX: When modifying this code, keep in mind that an insert
4285 * can wind up skipping both of these two special cases...
4286 */
4287 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004288 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004289 le32_to_cpu(insert_rec->e_cpos),
4290 right_path, &left_path);
4291 if (ret) {
4292 mlog_errno(ret);
4293 goto out;
4294 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004295
4296 /*
4297 * ocfs2_rotate_tree_right() might have extended the
4298 * transaction without re-journaling our tree root.
4299 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004300 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004301 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004302 if (ret) {
4303 mlog_errno(ret);
4304 goto out;
4305 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004306 } else if (type->ins_appending == APPEND_TAIL
4307 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004308 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004309 right_path, &left_path);
4310 if (ret) {
4311 mlog_errno(ret);
4312 goto out;
4313 }
4314 }
4315
Joel Becker3505bec2009-02-13 02:57:58 -08004316 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004317 insert_rec, type);
4318 if (ret) {
4319 mlog_errno(ret);
4320 goto out;
4321 }
4322
4323out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004324 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004325 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004326 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004327
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004328 ret = ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004329 if (ret)
4330 mlog_errno(ret);
4331
4332out:
4333 ocfs2_free_path(left_path);
4334 ocfs2_free_path(right_path);
4335
4336 return ret;
4337}
4338
Mark Fasheh328d5752007-06-18 10:48:04 -07004339static enum ocfs2_contig_type
Joel Beckera2970292009-02-13 03:09:54 -08004340ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
4341 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004342 struct ocfs2_extent_list *el, int index,
4343 struct ocfs2_extent_rec *split_rec)
4344{
Tao Maad5a4d72008-01-30 14:21:32 +08004345 int status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004346 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004347 u32 left_cpos, right_cpos;
4348 struct ocfs2_extent_rec *rec = NULL;
4349 struct ocfs2_extent_list *new_el;
4350 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4351 struct buffer_head *bh;
4352 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004353 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004354
4355 if (index > 0) {
4356 rec = &el->l_recs[index - 1];
4357 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004358 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004359 if (status)
4360 goto out;
4361
4362 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004363 left_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004364 if (!left_path)
4365 goto out;
4366
Joel Beckera2970292009-02-13 03:09:54 -08004367 status = ocfs2_find_path(et->et_ci, left_path,
4368 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004369 if (status)
4370 goto out;
4371
4372 new_el = path_leaf_el(left_path);
4373
4374 if (le16_to_cpu(new_el->l_next_free_rec) !=
4375 le16_to_cpu(new_el->l_count)) {
4376 bh = path_leaf_bh(left_path);
4377 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004378 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004379 "Extent block #%llu has an "
4380 "invalid l_next_free_rec of "
4381 "%d. It should have "
4382 "matched the l_count of %d",
4383 (unsigned long long)le64_to_cpu(eb->h_blkno),
4384 le16_to_cpu(new_el->l_next_free_rec),
4385 le16_to_cpu(new_el->l_count));
4386 status = -EINVAL;
Tao Maad5a4d72008-01-30 14:21:32 +08004387 goto out;
4388 }
4389 rec = &new_el->l_recs[
4390 le16_to_cpu(new_el->l_next_free_rec) - 1];
4391 }
4392 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004393
4394 /*
4395 * We're careful to check for an empty extent record here -
4396 * the merge code will know what to do if it sees one.
4397 */
Tao Maad5a4d72008-01-30 14:21:32 +08004398 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004399 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4400 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4401 ret = CONTIG_RIGHT;
4402 } else {
Joel Beckera2970292009-02-13 03:09:54 -08004403 ret = ocfs2_extent_contig(sb, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004404 }
4405 }
4406
Tao Maad5a4d72008-01-30 14:21:32 +08004407 rec = NULL;
4408 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4409 rec = &el->l_recs[index + 1];
4410 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4411 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004412 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004413 if (status)
4414 goto out;
4415
4416 if (right_cpos == 0)
4417 goto out;
4418
Joel Beckerffdd7a52008-10-17 22:32:01 -07004419 right_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004420 if (!right_path)
4421 goto out;
4422
Joel Beckera2970292009-02-13 03:09:54 -08004423 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004424 if (status)
4425 goto out;
4426
4427 new_el = path_leaf_el(right_path);
4428 rec = &new_el->l_recs[0];
4429 if (ocfs2_is_empty_extent(rec)) {
4430 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4431 bh = path_leaf_bh(right_path);
4432 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004433 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004434 "Extent block #%llu has an "
4435 "invalid l_next_free_rec of %d",
4436 (unsigned long long)le64_to_cpu(eb->h_blkno),
4437 le16_to_cpu(new_el->l_next_free_rec));
4438 status = -EINVAL;
Tao Maad5a4d72008-01-30 14:21:32 +08004439 goto out;
4440 }
4441 rec = &new_el->l_recs[1];
4442 }
4443 }
4444
4445 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004446 enum ocfs2_contig_type contig_type;
4447
Joel Beckera2970292009-02-13 03:09:54 -08004448 contig_type = ocfs2_extent_contig(sb, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004449
4450 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4451 ret = CONTIG_LEFTRIGHT;
4452 else if (ret == CONTIG_NONE)
4453 ret = contig_type;
4454 }
4455
Tao Maad5a4d72008-01-30 14:21:32 +08004456out:
4457 if (left_path)
4458 ocfs2_free_path(left_path);
4459 if (right_path)
4460 ocfs2_free_path(right_path);
4461
Mark Fasheh328d5752007-06-18 10:48:04 -07004462 return ret;
4463}
4464
Joel Becker1ef61b32009-02-13 03:12:33 -08004465static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004466 struct ocfs2_insert_type *insert,
4467 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004468 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004469{
4470 int i;
4471 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4472
Mark Fashehe48edee2007-03-07 16:46:57 -08004473 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4474
Mark Fashehdcd05382007-01-16 11:32:23 -08004475 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004476 contig_type = ocfs2_extent_contig(ocfs2_metadata_cache_get_super(et->et_ci),
4477 &el->l_recs[i], insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004478 if (contig_type != CONTIG_NONE) {
4479 insert->ins_contig_index = i;
4480 break;
4481 }
4482 }
4483 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004484
4485 if (insert->ins_contig != CONTIG_NONE) {
4486 struct ocfs2_extent_rec *rec =
4487 &el->l_recs[insert->ins_contig_index];
4488 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4489 le16_to_cpu(insert_rec->e_leaf_clusters);
4490
4491 /*
4492 * Caller might want us to limit the size of extents, don't
4493 * calculate contiguousness if we might exceed that limit.
4494 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004495 if (et->et_max_leaf_clusters &&
4496 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004497 insert->ins_contig = CONTIG_NONE;
4498 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004499}
4500
4501/*
4502 * This should only be called against the righmost leaf extent list.
4503 *
4504 * ocfs2_figure_appending_type() will figure out whether we'll have to
4505 * insert at the tail of the rightmost leaf.
4506 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004507 * This should also work against the root extent list for tree's with 0
4508 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004509 * then the logic here makes sense.
4510 */
4511static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4512 struct ocfs2_extent_list *el,
4513 struct ocfs2_extent_rec *insert_rec)
4514{
4515 int i;
4516 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4517 struct ocfs2_extent_rec *rec;
4518
4519 insert->ins_appending = APPEND_NONE;
4520
Mark Fashehe48edee2007-03-07 16:46:57 -08004521 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004522
4523 if (!el->l_next_free_rec)
4524 goto set_tail_append;
4525
4526 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4527 /* Were all records empty? */
4528 if (le16_to_cpu(el->l_next_free_rec) == 1)
4529 goto set_tail_append;
4530 }
4531
4532 i = le16_to_cpu(el->l_next_free_rec) - 1;
4533 rec = &el->l_recs[i];
4534
Mark Fashehe48edee2007-03-07 16:46:57 -08004535 if (cpos >=
4536 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004537 goto set_tail_append;
4538
4539 return;
4540
4541set_tail_append:
4542 insert->ins_appending = APPEND_TAIL;
4543}
4544
4545/*
4546 * Helper function called at the begining of an insert.
4547 *
4548 * This computes a few things that are commonly used in the process of
4549 * inserting into the btree:
4550 * - Whether the new extent is contiguous with an existing one.
4551 * - The current tree depth.
4552 * - Whether the insert is an appending one.
4553 * - The total # of free records in the tree.
4554 *
4555 * All of the information is stored on the ocfs2_insert_type
4556 * structure.
4557 */
Joel Becker627961b2009-02-13 03:14:38 -08004558static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004559 struct buffer_head **last_eb_bh,
4560 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004561 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004562 struct ocfs2_insert_type *insert)
4563{
4564 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004565 struct ocfs2_extent_block *eb;
4566 struct ocfs2_extent_list *el;
4567 struct ocfs2_path *path = NULL;
4568 struct buffer_head *bh = NULL;
4569
Mark Fasheh328d5752007-06-18 10:48:04 -07004570 insert->ins_split = SPLIT_NONE;
4571
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004572 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004573 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4574
4575 if (el->l_tree_depth) {
4576 /*
4577 * If we have tree depth, we read in the
4578 * rightmost extent block ahead of time as
4579 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4580 * may want it later.
4581 */
Joel Becker3d03a302009-02-12 17:49:26 -08004582 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004583 ocfs2_et_get_last_eb_blk(et),
4584 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004585 if (ret) {
4586 mlog_exit(ret);
4587 goto out;
4588 }
4589 eb = (struct ocfs2_extent_block *) bh->b_data;
4590 el = &eb->h_list;
4591 }
4592
4593 /*
4594 * Unless we have a contiguous insert, we'll need to know if
4595 * there is room left in our allocation tree for another
4596 * extent record.
4597 *
4598 * XXX: This test is simplistic, we can search for empty
4599 * extent records too.
4600 */
Tao Maoc77534f2007-08-28 17:22:33 -07004601 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004602 le16_to_cpu(el->l_next_free_rec);
4603
4604 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004605 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004606 ocfs2_figure_appending_type(insert, el, insert_rec);
4607 return 0;
4608 }
4609
Joel Beckerffdd7a52008-10-17 22:32:01 -07004610 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004611 if (!path) {
4612 ret = -ENOMEM;
4613 mlog_errno(ret);
4614 goto out;
4615 }
4616
4617 /*
4618 * In the case that we're inserting past what the tree
4619 * currently accounts for, ocfs2_find_path() will return for
4620 * us the rightmost tree path. This is accounted for below in
4621 * the appending code.
4622 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004623 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004624 if (ret) {
4625 mlog_errno(ret);
4626 goto out;
4627 }
4628
4629 el = path_leaf_el(path);
4630
4631 /*
4632 * Now that we have the path, there's two things we want to determine:
4633 * 1) Contiguousness (also set contig_index if this is so)
4634 *
4635 * 2) Are we doing an append? We can trivially break this up
4636 * into two types of appends: simple record append, or a
4637 * rotate inside the tail leaf.
4638 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004639 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004640
4641 /*
4642 * The insert code isn't quite ready to deal with all cases of
4643 * left contiguousness. Specifically, if it's an insert into
4644 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004645 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004646 * left. For now, just catch that case and fool the layers
4647 * above us. This works just fine for tree_depth == 0, which
4648 * is why we allow that above.
4649 */
4650 if (insert->ins_contig == CONTIG_LEFT &&
4651 insert->ins_contig_index == 0)
4652 insert->ins_contig = CONTIG_NONE;
4653
4654 /*
4655 * Ok, so we can simply compare against last_eb to figure out
4656 * whether the path doesn't exist. This will only happen in
4657 * the case that we're doing a tail append, so maybe we can
4658 * take advantage of that information somehow.
4659 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004660 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004661 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004662 /*
4663 * Ok, ocfs2_find_path() returned us the rightmost
4664 * tree path. This might be an appending insert. There are
4665 * two cases:
4666 * 1) We're doing a true append at the tail:
4667 * -This might even be off the end of the leaf
4668 * 2) We're "appending" by rotating in the tail
4669 */
4670 ocfs2_figure_appending_type(insert, el, insert_rec);
4671 }
4672
4673out:
4674 ocfs2_free_path(path);
4675
4676 if (ret == 0)
4677 *last_eb_bh = bh;
4678 else
4679 brelse(bh);
4680 return ret;
4681}
4682
4683/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004684 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004685 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004686 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004687 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004688int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004689 struct ocfs2_extent_tree *et,
4690 u32 cpos,
4691 u64 start_blk,
4692 u32 new_clusters,
4693 u8 flags,
4694 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004695{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004696 int status;
Tao Maoc77534f2007-08-28 17:22:33 -07004697 int uninitialized_var(free_records);
Mark Fashehccd979b2005-12-15 14:31:24 -08004698 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004699 struct ocfs2_insert_type insert = {0, };
4700 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004701
Joel Beckercc79d8c2009-02-13 03:24:43 -08004702 mlog(0, "add %u clusters at position %u to owner %llu\n",
4703 new_clusters, cpos,
4704 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08004705
Mark Fashehe48edee2007-03-07 16:46:57 -08004706 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004707 rec.e_cpos = cpu_to_le32(cpos);
4708 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004709 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004710 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004711 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004712 if (status) {
4713 mlog_errno(status);
4714 goto bail;
4715 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004716
Joel Becker627961b2009-02-13 03:14:38 -08004717 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004718 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004719 if (status < 0) {
4720 mlog_errno(status);
4721 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004722 }
4723
Mark Fashehdcd05382007-01-16 11:32:23 -08004724 mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
4725 "Insert.contig_index: %d, Insert.free_records: %d, "
4726 "Insert.tree_depth: %d\n",
4727 insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
Tao Maoc77534f2007-08-28 17:22:33 -07004728 free_records, insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004729
Tao Maoc77534f2007-08-28 17:22:33 -07004730 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004731 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004732 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004733 meta_ac);
4734 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004735 mlog_errno(status);
4736 goto bail;
4737 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004738 }
4739
Mark Fashehdcd05382007-01-16 11:32:23 -08004740 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004741 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004742 if (status < 0)
4743 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004744 else
4745 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004746
4747bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004748 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004749
Tao Maf56654c2008-08-18 17:38:48 +08004750 mlog_exit(status);
4751 return status;
4752}
4753
Tao Ma0eb8d472008-08-18 17:38:45 +08004754/*
4755 * Allcate and add clusters into the extent b-tree.
4756 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004757 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d472008-08-18 17:38:45 +08004758 * it is not limited to the file storage. Any extent tree can use this
4759 * function if it implements the proper ocfs2_extent_tree.
4760 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004761int ocfs2_add_clusters_in_btree(handle_t *handle,
4762 struct ocfs2_extent_tree *et,
Tao Ma0eb8d472008-08-18 17:38:45 +08004763 u32 *logical_offset,
4764 u32 clusters_to_add,
4765 int mark_unwritten,
Tao Ma0eb8d472008-08-18 17:38:45 +08004766 struct ocfs2_alloc_context *data_ac,
4767 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004768 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d472008-08-18 17:38:45 +08004769{
4770 int status = 0;
4771 int free_extents;
4772 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4773 u32 bit_off, num_bits;
4774 u64 block;
4775 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004776 struct ocfs2_super *osb =
4777 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d472008-08-18 17:38:45 +08004778
4779 BUG_ON(!clusters_to_add);
4780
4781 if (mark_unwritten)
4782 flags = OCFS2_EXT_UNWRITTEN;
4783
Joel Becker3d03a302009-02-12 17:49:26 -08004784 free_extents = ocfs2_num_free_extents(osb, et);
Tao Ma0eb8d472008-08-18 17:38:45 +08004785 if (free_extents < 0) {
4786 status = free_extents;
4787 mlog_errno(status);
4788 goto leave;
4789 }
4790
4791 /* there are two cases which could cause us to EAGAIN in the
4792 * we-need-more-metadata case:
4793 * 1) we haven't reserved *any*
4794 * 2) we are so fragmented, we've needed to add metadata too
4795 * many times. */
4796 if (!free_extents && !meta_ac) {
4797 mlog(0, "we haven't reserved any metadata!\n");
4798 status = -EAGAIN;
4799 reason = RESTART_META;
4800 goto leave;
4801 } else if ((!free_extents)
4802 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004803 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Ma0eb8d472008-08-18 17:38:45 +08004804 mlog(0, "filesystem is really fragmented...\n");
4805 status = -EAGAIN;
4806 reason = RESTART_META;
4807 goto leave;
4808 }
4809
4810 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
4811 clusters_to_add, &bit_off, &num_bits);
4812 if (status < 0) {
4813 if (status != -ENOSPC)
4814 mlog_errno(status);
4815 goto leave;
4816 }
4817
4818 BUG_ON(num_bits > clusters_to_add);
4819
Joel Becker13723d02008-10-17 19:25:01 -07004820 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004821 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004822 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d472008-08-18 17:38:45 +08004823 if (status < 0) {
4824 mlog_errno(status);
4825 goto leave;
4826 }
4827
4828 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Joel Beckercbee7e12009-02-13 03:34:15 -08004829 mlog(0, "Allocating %u clusters at block %u for owner %llu\n",
4830 num_bits, bit_off,
4831 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Joel Beckercc79d8c2009-02-13 03:24:43 -08004832 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004833 num_bits, flags, meta_ac);
Tao Ma0eb8d472008-08-18 17:38:45 +08004834 if (status < 0) {
4835 mlog_errno(status);
4836 goto leave;
4837 }
4838
Joel Beckerf99b9b72008-08-20 19:36:33 -07004839 status = ocfs2_journal_dirty(handle, et->et_root_bh);
Tao Ma0eb8d472008-08-18 17:38:45 +08004840 if (status < 0) {
4841 mlog_errno(status);
4842 goto leave;
4843 }
4844
4845 clusters_to_add -= num_bits;
4846 *logical_offset += num_bits;
4847
4848 if (clusters_to_add) {
4849 mlog(0, "need to alloc once more, wanted = %u\n",
4850 clusters_to_add);
4851 status = -EAGAIN;
4852 reason = RESTART_TRANS;
4853 }
4854
4855leave:
4856 mlog_exit(status);
4857 if (reason_ret)
4858 *reason_ret = reason;
4859 return status;
4860}
4861
Mark Fasheh328d5752007-06-18 10:48:04 -07004862static void ocfs2_make_right_split_rec(struct super_block *sb,
4863 struct ocfs2_extent_rec *split_rec,
4864 u32 cpos,
4865 struct ocfs2_extent_rec *rec)
4866{
4867 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4868 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4869
4870 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4871
4872 split_rec->e_cpos = cpu_to_le32(cpos);
4873 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4874
4875 split_rec->e_blkno = rec->e_blkno;
4876 le64_add_cpu(&split_rec->e_blkno,
4877 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4878
4879 split_rec->e_flags = rec->e_flags;
4880}
4881
Joel Beckerd2311292009-02-13 03:43:22 -08004882static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004883 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004884 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004885 struct buffer_head **last_eb_bh,
4886 int split_index,
4887 struct ocfs2_extent_rec *orig_split_rec,
4888 struct ocfs2_alloc_context *meta_ac)
4889{
4890 int ret = 0, depth;
4891 unsigned int insert_range, rec_range, do_leftright = 0;
4892 struct ocfs2_extent_rec tmprec;
4893 struct ocfs2_extent_list *rightmost_el;
4894 struct ocfs2_extent_rec rec;
4895 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4896 struct ocfs2_insert_type insert;
4897 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004898
4899leftright:
4900 /*
4901 * Store a copy of the record on the stack - it might move
4902 * around as the tree is manipulated below.
4903 */
4904 rec = path_leaf_el(path)->l_recs[split_index];
4905
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004906 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004907
4908 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4909 if (depth) {
4910 BUG_ON(!(*last_eb_bh));
4911 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4912 rightmost_el = &eb->h_list;
4913 }
4914
4915 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4916 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004917 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004918 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004919 if (ret) {
4920 mlog_errno(ret);
4921 goto out;
4922 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004923 }
4924
4925 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4926 insert.ins_appending = APPEND_NONE;
4927 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004928 insert.ins_tree_depth = depth;
4929
4930 insert_range = le32_to_cpu(split_rec.e_cpos) +
4931 le16_to_cpu(split_rec.e_leaf_clusters);
4932 rec_range = le32_to_cpu(rec.e_cpos) +
4933 le16_to_cpu(rec.e_leaf_clusters);
4934
4935 if (split_rec.e_cpos == rec.e_cpos) {
4936 insert.ins_split = SPLIT_LEFT;
4937 } else if (insert_range == rec_range) {
4938 insert.ins_split = SPLIT_RIGHT;
4939 } else {
4940 /*
4941 * Left/right split. We fake this as a right split
4942 * first and then make a second pass as a left split.
4943 */
4944 insert.ins_split = SPLIT_RIGHT;
4945
Joel Beckerd2311292009-02-13 03:43:22 -08004946 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4947 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004948
4949 split_rec = tmprec;
4950
4951 BUG_ON(do_leftright);
4952 do_leftright = 1;
4953 }
4954
Joel Becker3505bec2009-02-13 02:57:58 -08004955 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004956 if (ret) {
4957 mlog_errno(ret);
4958 goto out;
4959 }
4960
4961 if (do_leftright == 1) {
4962 u32 cpos;
4963 struct ocfs2_extent_list *el;
4964
4965 do_leftright++;
4966 split_rec = *orig_split_rec;
4967
4968 ocfs2_reinit_path(path, 1);
4969
4970 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004971 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004972 if (ret) {
4973 mlog_errno(ret);
4974 goto out;
4975 }
4976
4977 el = path_leaf_el(path);
4978 split_index = ocfs2_search_extent_list(el, cpos);
4979 goto leftright;
4980 }
4981out:
4982
4983 return ret;
4984}
4985
Tao Ma47be12e2009-01-09 07:32:48 +08004986static int ocfs2_replace_extent_rec(struct inode *inode,
4987 handle_t *handle,
4988 struct ocfs2_path *path,
4989 struct ocfs2_extent_list *el,
4990 int split_index,
4991 struct ocfs2_extent_rec *split_rec)
4992{
4993 int ret;
4994
Joel Becker0cf2f762009-02-12 16:41:25 -08004995 ret = ocfs2_path_bh_journal_access(handle, INODE_CACHE(inode), path,
Tao Ma47be12e2009-01-09 07:32:48 +08004996 path_num_items(path) - 1);
4997 if (ret) {
4998 mlog_errno(ret);
4999 goto out;
5000 }
5001
5002 el->l_recs[split_index] = *split_rec;
5003
5004 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5005out:
5006 return ret;
5007}
5008
Mark Fasheh328d5752007-06-18 10:48:04 -07005009/*
5010 * Mark part or all of the extent record at split_index in the leaf
5011 * pointed to by path as written. This removes the unwritten
5012 * extent flag.
5013 *
5014 * Care is taken to handle contiguousness so as to not grow the tree.
5015 *
5016 * meta_ac is not strictly necessary - we only truly need it if growth
5017 * of the tree is required. All other cases will degrade into a less
5018 * optimal tree layout.
5019 *
Tao Mae7d4cb62008-08-18 17:38:44 +08005020 * last_eb_bh should be the rightmost leaf block for any extent
5021 * btree. Since a split may grow the tree or a merge might shrink it,
5022 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07005023 *
5024 * This code is optimized for readability - several passes might be
5025 * made over certain portions of the tree. All of those blocks will
5026 * have been brought into cache (and pinned via the journal), so the
5027 * extra overhead is not expressed in terms of disk reads.
5028 */
5029static int __ocfs2_mark_extent_written(struct inode *inode,
Tao Mae7d4cb62008-08-18 17:38:44 +08005030 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07005031 handle_t *handle,
5032 struct ocfs2_path *path,
5033 int split_index,
5034 struct ocfs2_extent_rec *split_rec,
5035 struct ocfs2_alloc_context *meta_ac,
5036 struct ocfs2_cached_dealloc_ctxt *dealloc)
5037{
5038 int ret = 0;
5039 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005040 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005041 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5042 struct ocfs2_merge_ctxt ctxt;
5043 struct ocfs2_extent_list *rightmost_el;
5044
Roel Kluin3cf0c502007-10-27 00:20:36 +02005045 if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh328d5752007-06-18 10:48:04 -07005046 ret = -EIO;
5047 mlog_errno(ret);
5048 goto out;
5049 }
5050
5051 if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5052 ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5053 (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5054 ret = -EIO;
5055 mlog_errno(ret);
5056 goto out;
5057 }
5058
Joel Beckera2970292009-02-13 03:09:54 -08005059 ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
Mark Fasheh328d5752007-06-18 10:48:04 -07005060 split_index,
5061 split_rec);
5062
5063 /*
5064 * The core merge / split code wants to know how much room is
5065 * left in this inodes allocation tree, so we pass the
5066 * rightmost extent list.
5067 */
5068 if (path->p_tree_depth) {
5069 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07005070
Joel Becker3d03a302009-02-12 17:49:26 -08005071 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005072 ocfs2_et_get_last_eb_blk(et),
5073 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005074 if (ret) {
5075 mlog_exit(ret);
5076 goto out;
5077 }
5078
5079 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fasheh328d5752007-06-18 10:48:04 -07005080 rightmost_el = &eb->h_list;
5081 } else
5082 rightmost_el = path_root_el(path);
5083
Mark Fasheh328d5752007-06-18 10:48:04 -07005084 if (rec->e_cpos == split_rec->e_cpos &&
5085 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5086 ctxt.c_split_covers_rec = 1;
5087 else
5088 ctxt.c_split_covers_rec = 0;
5089
5090 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5091
Mark Fasheh015452b2007-09-12 10:21:22 -07005092 mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n",
5093 split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent,
5094 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005095
5096 if (ctxt.c_contig_type == CONTIG_NONE) {
5097 if (ctxt.c_split_covers_rec)
Tao Ma47be12e2009-01-09 07:32:48 +08005098 ret = ocfs2_replace_extent_rec(inode, handle,
5099 path, el,
5100 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005101 else
Joel Beckerd2311292009-02-13 03:43:22 -08005102 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005103 &last_eb_bh, split_index,
5104 split_rec, meta_ac);
5105 if (ret)
5106 mlog_errno(ret);
5107 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005108 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005109 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005110 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005111 if (ret)
5112 mlog_errno(ret);
5113 }
5114
Mark Fasheh328d5752007-06-18 10:48:04 -07005115out:
5116 brelse(last_eb_bh);
5117 return ret;
5118}
5119
5120/*
5121 * Mark the already-existing extent at cpos as written for len clusters.
5122 *
5123 * If the existing extent is larger than the request, initiate a
5124 * split. An attempt will be made at merging with adjacent extents.
5125 *
5126 * The caller is responsible for passing down meta_ac if we'll need it.
5127 */
Joel Beckerf99b9b72008-08-20 19:36:33 -07005128int ocfs2_mark_extent_written(struct inode *inode,
5129 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07005130 handle_t *handle, u32 cpos, u32 len, u32 phys,
5131 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005132 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005133{
5134 int ret, index;
5135 u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys);
5136 struct ocfs2_extent_rec split_rec;
5137 struct ocfs2_path *left_path = NULL;
5138 struct ocfs2_extent_list *el;
5139
5140 mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n",
5141 inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno);
5142
5143 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
5144 ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
5145 "that are being written to, but the feature bit "
5146 "is not set in the super block.",
5147 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5148 ret = -EROFS;
5149 goto out;
5150 }
5151
5152 /*
5153 * XXX: This should be fixed up so that we just re-insert the
5154 * next extent records.
5155 */
Joel Becker4c911ee2009-02-13 02:50:12 -08005156 ocfs2_et_extent_map_truncate(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07005157
Joel Beckerffdd7a52008-10-17 22:32:01 -07005158 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005159 if (!left_path) {
5160 ret = -ENOMEM;
5161 mlog_errno(ret);
5162 goto out;
5163 }
5164
Joel Beckerfacdb772009-02-12 18:08:48 -08005165 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005166 if (ret) {
5167 mlog_errno(ret);
5168 goto out;
5169 }
5170 el = path_leaf_el(left_path);
5171
5172 index = ocfs2_search_extent_list(el, cpos);
5173 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
5174 ocfs2_error(inode->i_sb,
5175 "Inode %llu has an extent at cpos %u which can no "
5176 "longer be found.\n",
5177 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
5178 ret = -EROFS;
5179 goto out;
5180 }
5181
5182 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5183 split_rec.e_cpos = cpu_to_le32(cpos);
5184 split_rec.e_leaf_clusters = cpu_to_le16(len);
5185 split_rec.e_blkno = cpu_to_le64(start_blkno);
5186 split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags;
5187 split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN;
5188
Joel Beckerf99b9b72008-08-20 19:36:33 -07005189 ret = __ocfs2_mark_extent_written(inode, et, handle, left_path,
Tao Mae7d4cb62008-08-18 17:38:44 +08005190 index, &split_rec, meta_ac,
5191 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005192 if (ret)
5193 mlog_errno(ret);
5194
5195out:
5196 ocfs2_free_path(left_path);
5197 return ret;
5198}
5199
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005200static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5201 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005202 int index, u32 new_range,
5203 struct ocfs2_alloc_context *meta_ac)
5204{
5205 int ret, depth, credits = handle->h_buffer_credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005206 struct buffer_head *last_eb_bh = NULL;
5207 struct ocfs2_extent_block *eb;
5208 struct ocfs2_extent_list *rightmost_el, *el;
5209 struct ocfs2_extent_rec split_rec;
5210 struct ocfs2_extent_rec *rec;
5211 struct ocfs2_insert_type insert;
5212
5213 /*
5214 * Setup the record to split before we grow the tree.
5215 */
5216 el = path_leaf_el(path);
5217 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005218 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5219 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005220
5221 depth = path->p_tree_depth;
5222 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005223 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005224 ocfs2_et_get_last_eb_blk(et),
5225 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005226 if (ret < 0) {
5227 mlog_errno(ret);
5228 goto out;
5229 }
5230
5231 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5232 rightmost_el = &eb->h_list;
5233 } else
5234 rightmost_el = path_leaf_el(path);
5235
Tao Ma811f9332008-08-18 17:38:43 +08005236 credits += path->p_tree_depth +
Joel Beckerce1d9ea2008-08-20 16:30:07 -07005237 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005238 ret = ocfs2_extend_trans(handle, credits);
5239 if (ret) {
5240 mlog_errno(ret);
5241 goto out;
5242 }
5243
5244 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5245 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005246 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005247 meta_ac);
5248 if (ret) {
5249 mlog_errno(ret);
5250 goto out;
5251 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005252 }
5253
5254 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5255 insert.ins_appending = APPEND_NONE;
5256 insert.ins_contig = CONTIG_NONE;
5257 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005258 insert.ins_tree_depth = depth;
5259
Joel Becker3505bec2009-02-13 02:57:58 -08005260 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005261 if (ret)
5262 mlog_errno(ret);
5263
5264out:
5265 brelse(last_eb_bh);
5266 return ret;
5267}
5268
Joel Becker043beeb2009-02-13 02:42:30 -08005269static int ocfs2_truncate_rec(handle_t *handle,
5270 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005271 struct ocfs2_path *path, int index,
5272 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005273 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005274{
5275 int ret;
5276 u32 left_cpos, rec_range, trunc_range;
5277 int wants_rotate = 0, is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005278 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005279 struct ocfs2_path *left_path = NULL;
5280 struct ocfs2_extent_list *el = path_leaf_el(path);
5281 struct ocfs2_extent_rec *rec;
5282 struct ocfs2_extent_block *eb;
5283
5284 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Joel Becker70f18c02009-02-13 02:09:31 -08005285 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005286 if (ret) {
5287 mlog_errno(ret);
5288 goto out;
5289 }
5290
5291 index--;
5292 }
5293
5294 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5295 path->p_tree_depth) {
5296 /*
5297 * Check whether this is the rightmost tree record. If
5298 * we remove all of this record or part of its right
5299 * edge then an update of the record lengths above it
5300 * will be required.
5301 */
5302 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5303 if (eb->h_next_leaf_blk == 0)
5304 is_rightmost_tree_rec = 1;
5305 }
5306
5307 rec = &el->l_recs[index];
5308 if (index == 0 && path->p_tree_depth &&
5309 le32_to_cpu(rec->e_cpos) == cpos) {
5310 /*
5311 * Changing the leftmost offset (via partial or whole
5312 * record truncate) of an interior (or rightmost) path
5313 * means we have to update the subtree that is formed
5314 * by this leaf and the one to it's left.
5315 *
5316 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005317 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005318 * 2) The leaf is rightmost and will be empty after
5319 * we remove the extent record - the rotate code
5320 * knows how to update the newly formed edge.
5321 */
5322
Joel Becker043beeb2009-02-13 02:42:30 -08005323 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005324 if (ret) {
5325 mlog_errno(ret);
5326 goto out;
5327 }
5328
5329 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005330 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005331 if (!left_path) {
5332 ret = -ENOMEM;
5333 mlog_errno(ret);
5334 goto out;
5335 }
5336
Joel Beckerfacdb772009-02-12 18:08:48 -08005337 ret = ocfs2_find_path(et->et_ci, left_path,
5338 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005339 if (ret) {
5340 mlog_errno(ret);
5341 goto out;
5342 }
5343 }
5344 }
5345
5346 ret = ocfs2_extend_rotate_transaction(handle, 0,
5347 handle->h_buffer_credits,
5348 path);
5349 if (ret) {
5350 mlog_errno(ret);
5351 goto out;
5352 }
5353
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005354 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005355 if (ret) {
5356 mlog_errno(ret);
5357 goto out;
5358 }
5359
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005360 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005361 if (ret) {
5362 mlog_errno(ret);
5363 goto out;
5364 }
5365
5366 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5367 trunc_range = cpos + len;
5368
5369 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5370 int next_free;
5371
5372 memset(rec, 0, sizeof(*rec));
5373 ocfs2_cleanup_merge(el, index);
5374 wants_rotate = 1;
5375
5376 next_free = le16_to_cpu(el->l_next_free_rec);
5377 if (is_rightmost_tree_rec && next_free > 1) {
5378 /*
5379 * We skip the edge update if this path will
5380 * be deleted by the rotate code.
5381 */
5382 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005383 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005384 rec);
5385 }
5386 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5387 /* Remove leftmost portion of the record. */
5388 le32_add_cpu(&rec->e_cpos, len);
5389 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5390 le16_add_cpu(&rec->e_leaf_clusters, -len);
5391 } else if (rec_range == trunc_range) {
5392 /* Remove rightmost portion of the record */
5393 le16_add_cpu(&rec->e_leaf_clusters, -len);
5394 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005395 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005396 } else {
5397 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005398 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5399 "(%u, %u)\n",
5400 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005401 le32_to_cpu(rec->e_cpos),
5402 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5403 BUG();
5404 }
5405
5406 if (left_path) {
5407 int subtree_index;
5408
Joel Becker7dc02802009-02-12 19:20:13 -08005409 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005410 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005411 subtree_index);
5412 }
5413
5414 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5415
Joel Becker70f18c02009-02-13 02:09:31 -08005416 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005417 if (ret) {
5418 mlog_errno(ret);
5419 goto out;
5420 }
5421
5422out:
5423 ocfs2_free_path(left_path);
5424 return ret;
5425}
5426
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005427int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005428 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005429 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005430 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005431 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005432{
5433 int ret, index;
5434 u32 rec_range, trunc_range;
5435 struct ocfs2_extent_rec *rec;
5436 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005437 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005438
Joel Becker4c911ee2009-02-13 02:50:12 -08005439 /*
5440 * XXX: Why are we truncating to 0 instead of wherever this
5441 * affects us?
5442 */
5443 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005444
Joel Beckerffdd7a52008-10-17 22:32:01 -07005445 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005446 if (!path) {
5447 ret = -ENOMEM;
5448 mlog_errno(ret);
5449 goto out;
5450 }
5451
Joel Beckerfacdb772009-02-12 18:08:48 -08005452 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005453 if (ret) {
5454 mlog_errno(ret);
5455 goto out;
5456 }
5457
5458 el = path_leaf_el(path);
5459 index = ocfs2_search_extent_list(el, cpos);
5460 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005461 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5462 "Owner %llu has an extent at cpos %u which can no "
Mark Fashehd0c7d702007-07-03 13:27:22 -07005463 "longer be found.\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005464 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5465 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005466 ret = -EROFS;
5467 goto out;
5468 }
5469
5470 /*
5471 * We have 3 cases of extent removal:
5472 * 1) Range covers the entire extent rec
5473 * 2) Range begins or ends on one edge of the extent rec
5474 * 3) Range is in the middle of the extent rec (no shared edges)
5475 *
5476 * For case 1 we remove the extent rec and left rotate to
5477 * fill the hole.
5478 *
5479 * For case 2 we just shrink the existing extent rec, with a
5480 * tree update if the shrinking edge is also the edge of an
5481 * extent block.
5482 *
5483 * For case 3 we do a right split to turn the extent rec into
5484 * something case 2 can handle.
5485 */
5486 rec = &el->l_recs[index];
5487 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5488 trunc_range = cpos + len;
5489
5490 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5491
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005492 mlog(0, "Owner %llu, remove (cpos %u, len %u). Existing index %d "
Mark Fashehd0c7d702007-07-03 13:27:22 -07005493 "(cpos %u, len %u)\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005494 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5495 cpos, len, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005496 le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec));
5497
5498 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005499 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5500 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005501 if (ret) {
5502 mlog_errno(ret);
5503 goto out;
5504 }
5505 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005506 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005507 trunc_range, meta_ac);
5508 if (ret) {
5509 mlog_errno(ret);
5510 goto out;
5511 }
5512
5513 /*
5514 * The split could have manipulated the tree enough to
5515 * move the record location, so we have to look for it again.
5516 */
5517 ocfs2_reinit_path(path, 1);
5518
Joel Beckerfacdb772009-02-12 18:08:48 -08005519 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005520 if (ret) {
5521 mlog_errno(ret);
5522 goto out;
5523 }
5524
5525 el = path_leaf_el(path);
5526 index = ocfs2_search_extent_list(el, cpos);
5527 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005528 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5529 "Owner %llu: split at cpos %u lost record.",
5530 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005531 cpos);
5532 ret = -EROFS;
5533 goto out;
5534 }
5535
5536 /*
5537 * Double check our values here. If anything is fishy,
5538 * it's easier to catch it at the top level.
5539 */
5540 rec = &el->l_recs[index];
5541 rec_range = le32_to_cpu(rec->e_cpos) +
5542 ocfs2_rec_clusters(el, rec);
5543 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005544 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5545 "Owner %llu: error after split at cpos %u"
Mark Fashehd0c7d702007-07-03 13:27:22 -07005546 "trunc len %u, existing record is (%u,%u)",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005547 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005548 cpos, len, le32_to_cpu(rec->e_cpos),
5549 ocfs2_rec_clusters(el, rec));
5550 ret = -EROFS;
5551 goto out;
5552 }
5553
Joel Becker043beeb2009-02-13 02:42:30 -08005554 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5555 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005556 if (ret) {
5557 mlog_errno(ret);
5558 goto out;
5559 }
5560 }
5561
5562out:
5563 ocfs2_free_path(path);
5564 return ret;
5565}
5566
Mark Fashehfecc0112008-11-12 15:16:38 -08005567int ocfs2_remove_btree_range(struct inode *inode,
5568 struct ocfs2_extent_tree *et,
5569 u32 cpos, u32 phys_cpos, u32 len,
5570 struct ocfs2_cached_dealloc_ctxt *dealloc)
5571{
5572 int ret;
5573 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5574 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5575 struct inode *tl_inode = osb->osb_tl_inode;
5576 handle_t *handle;
5577 struct ocfs2_alloc_context *meta_ac = NULL;
5578
5579 ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac);
5580 if (ret) {
5581 mlog_errno(ret);
5582 return ret;
5583 }
5584
5585 mutex_lock(&tl_inode->i_mutex);
5586
5587 if (ocfs2_truncate_log_needs_flush(osb)) {
5588 ret = __ocfs2_flush_truncate_log(osb);
5589 if (ret < 0) {
5590 mlog_errno(ret);
5591 goto out;
5592 }
5593 }
5594
Jan Karaa90714c2008-10-09 19:38:40 +02005595 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
Mark Fashehfecc0112008-11-12 15:16:38 -08005596 if (IS_ERR(handle)) {
5597 ret = PTR_ERR(handle);
5598 mlog_errno(ret);
5599 goto out;
5600 }
5601
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005602 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005603 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005604 if (ret) {
5605 mlog_errno(ret);
5606 goto out;
5607 }
5608
Mark Fashehfd4ef232009-01-29 15:06:21 -08005609 vfs_dq_free_space_nodirty(inode,
5610 ocfs2_clusters_to_bytes(inode->i_sb, len));
5611
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005612 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005613 if (ret) {
5614 mlog_errno(ret);
5615 goto out_commit;
5616 }
5617
Joel Becker6136ca52009-02-12 19:32:43 -08005618 ocfs2_et_update_clusters(et, -len);
Mark Fashehfecc0112008-11-12 15:16:38 -08005619
5620 ret = ocfs2_journal_dirty(handle, et->et_root_bh);
5621 if (ret) {
5622 mlog_errno(ret);
5623 goto out_commit;
5624 }
5625
5626 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
5627 if (ret)
5628 mlog_errno(ret);
5629
5630out_commit:
5631 ocfs2_commit_trans(osb, handle);
5632out:
5633 mutex_unlock(&tl_inode->i_mutex);
5634
5635 if (meta_ac)
5636 ocfs2_free_alloc_context(meta_ac);
5637
5638 return ret;
5639}
5640
Mark Fasheh063c4562007-07-03 13:34:11 -07005641int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005642{
5643 struct buffer_head *tl_bh = osb->osb_tl_bh;
5644 struct ocfs2_dinode *di;
5645 struct ocfs2_truncate_log *tl;
5646
5647 di = (struct ocfs2_dinode *) tl_bh->b_data;
5648 tl = &di->id2.i_dealloc;
5649
5650 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5651 "slot %d, invalid truncate log parameters: used = "
5652 "%u, count = %u\n", osb->slot_num,
5653 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5654 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5655}
5656
5657static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5658 unsigned int new_start)
5659{
5660 unsigned int tail_index;
5661 unsigned int current_tail;
5662
5663 /* No records, nothing to coalesce */
5664 if (!le16_to_cpu(tl->tl_used))
5665 return 0;
5666
5667 tail_index = le16_to_cpu(tl->tl_used) - 1;
5668 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5669 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5670
5671 return current_tail == new_start;
5672}
5673
Mark Fasheh063c4562007-07-03 13:34:11 -07005674int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5675 handle_t *handle,
5676 u64 start_blk,
5677 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005678{
5679 int status, index;
5680 unsigned int start_cluster, tl_count;
5681 struct inode *tl_inode = osb->osb_tl_inode;
5682 struct buffer_head *tl_bh = osb->osb_tl_bh;
5683 struct ocfs2_dinode *di;
5684 struct ocfs2_truncate_log *tl;
5685
Mark Fashehb0697052006-03-03 10:24:33 -08005686 mlog_entry("start_blk = %llu, num_clusters = %u\n",
5687 (unsigned long long)start_blk, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005688
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005689 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005690
5691 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5692
5693 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005694
Joel Becker10995aa2008-11-13 14:49:12 -08005695 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5696 * by the underlying call to ocfs2_read_inode_block(), so any
5697 * corruption is a code bug */
5698 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5699
5700 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005701 tl_count = le16_to_cpu(tl->tl_count);
5702 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5703 tl_count == 0,
Mark Fashehb0697052006-03-03 10:24:33 -08005704 "Truncate record count on #%llu invalid "
5705 "wanted %u, actual %u\n",
5706 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005707 ocfs2_truncate_recs_per_inode(osb->sb),
5708 le16_to_cpu(tl->tl_count));
5709
5710 /* Caller should have known to flush before calling us. */
5711 index = le16_to_cpu(tl->tl_used);
5712 if (index >= tl_count) {
5713 status = -ENOSPC;
5714 mlog_errno(status);
5715 goto bail;
5716 }
5717
Joel Becker0cf2f762009-02-12 16:41:25 -08005718 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005719 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005720 if (status < 0) {
5721 mlog_errno(status);
5722 goto bail;
5723 }
5724
5725 mlog(0, "Log truncate of %u clusters starting at cluster %u to "
Mark Fashehb0697052006-03-03 10:24:33 -08005726 "%llu (index = %d)\n", num_clusters, start_cluster,
5727 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index);
Mark Fashehccd979b2005-12-15 14:31:24 -08005728
5729 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5730 /*
5731 * Move index back to the record we are coalescing with.
5732 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5733 */
5734 index--;
5735
5736 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
5737 mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n",
5738 index, le32_to_cpu(tl->tl_recs[index].t_start),
5739 num_clusters);
5740 } else {
5741 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5742 tl->tl_used = cpu_to_le16(index + 1);
5743 }
5744 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5745
5746 status = ocfs2_journal_dirty(handle, tl_bh);
5747 if (status < 0) {
5748 mlog_errno(status);
5749 goto bail;
5750 }
5751
5752bail:
5753 mlog_exit(status);
5754 return status;
5755}
5756
5757static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07005758 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08005759 struct inode *data_alloc_inode,
5760 struct buffer_head *data_alloc_bh)
5761{
5762 int status = 0;
5763 int i;
5764 unsigned int num_clusters;
5765 u64 start_blk;
5766 struct ocfs2_truncate_rec rec;
5767 struct ocfs2_dinode *di;
5768 struct ocfs2_truncate_log *tl;
5769 struct inode *tl_inode = osb->osb_tl_inode;
5770 struct buffer_head *tl_bh = osb->osb_tl_bh;
5771
5772 mlog_entry_void();
5773
5774 di = (struct ocfs2_dinode *) tl_bh->b_data;
5775 tl = &di->id2.i_dealloc;
5776 i = le16_to_cpu(tl->tl_used) - 1;
5777 while (i >= 0) {
5778 /* Caller has given us at least enough credits to
5779 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005780 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005781 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005782 if (status < 0) {
5783 mlog_errno(status);
5784 goto bail;
5785 }
5786
5787 tl->tl_used = cpu_to_le16(i);
5788
5789 status = ocfs2_journal_dirty(handle, tl_bh);
5790 if (status < 0) {
5791 mlog_errno(status);
5792 goto bail;
5793 }
5794
5795 /* TODO: Perhaps we can calculate the bulk of the
5796 * credits up front rather than extending like
5797 * this. */
5798 status = ocfs2_extend_trans(handle,
5799 OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5800 if (status < 0) {
5801 mlog_errno(status);
5802 goto bail;
5803 }
5804
5805 rec = tl->tl_recs[i];
5806 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5807 le32_to_cpu(rec.t_start));
5808 num_clusters = le32_to_cpu(rec.t_clusters);
5809
5810 /* if start_blk is not set, we ignore the record as
5811 * invalid. */
5812 if (start_blk) {
5813 mlog(0, "free record %d, start = %u, clusters = %u\n",
5814 i, le32_to_cpu(rec.t_start), num_clusters);
5815
5816 status = ocfs2_free_clusters(handle, data_alloc_inode,
5817 data_alloc_bh, start_blk,
5818 num_clusters);
5819 if (status < 0) {
5820 mlog_errno(status);
5821 goto bail;
5822 }
5823 }
5824 i--;
5825 }
5826
5827bail:
5828 mlog_exit(status);
5829 return status;
5830}
5831
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005832/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005833int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005834{
5835 int status;
5836 unsigned int num_to_flush;
Mark Fasheh1fabe142006-10-09 18:11:45 -07005837 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005838 struct inode *tl_inode = osb->osb_tl_inode;
5839 struct inode *data_alloc_inode = NULL;
5840 struct buffer_head *tl_bh = osb->osb_tl_bh;
5841 struct buffer_head *data_alloc_bh = NULL;
5842 struct ocfs2_dinode *di;
5843 struct ocfs2_truncate_log *tl;
5844
5845 mlog_entry_void();
5846
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005847 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005848
5849 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005850
Joel Becker10995aa2008-11-13 14:49:12 -08005851 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5852 * by the underlying call to ocfs2_read_inode_block(), so any
5853 * corruption is a code bug */
5854 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5855
5856 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005857 num_to_flush = le16_to_cpu(tl->tl_used);
Mark Fashehb0697052006-03-03 10:24:33 -08005858 mlog(0, "Flush %u records from truncate log #%llu\n",
5859 num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08005860 if (!num_to_flush) {
5861 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005862 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005863 }
5864
5865 data_alloc_inode = ocfs2_get_system_file_inode(osb,
5866 GLOBAL_BITMAP_SYSTEM_INODE,
5867 OCFS2_INVALID_SLOT);
5868 if (!data_alloc_inode) {
5869 status = -EINVAL;
5870 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005871 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005872 }
5873
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005874 mutex_lock(&data_alloc_inode->i_mutex);
5875
Mark Fashehe63aecb62007-10-18 15:30:42 -07005876 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08005877 if (status < 0) {
5878 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005879 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08005880 }
5881
Mark Fasheh65eff9c2006-10-09 17:26:22 -07005882 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005883 if (IS_ERR(handle)) {
5884 status = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005885 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005886 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08005887 }
5888
5889 status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
5890 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005891 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08005892 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08005893
Mark Fasheh02dc1af2006-10-09 16:48:10 -07005894 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005895
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005896out_unlock:
5897 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07005898 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08005899
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005900out_mutex:
5901 mutex_unlock(&data_alloc_inode->i_mutex);
5902 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08005903
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005904out:
Mark Fashehccd979b2005-12-15 14:31:24 -08005905 mlog_exit(status);
5906 return status;
5907}
5908
5909int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
5910{
5911 int status;
5912 struct inode *tl_inode = osb->osb_tl_inode;
5913
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005914 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08005915 status = __ocfs2_flush_truncate_log(osb);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005916 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08005917
5918 return status;
5919}
5920
David Howellsc4028952006-11-22 14:57:56 +00005921static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08005922{
5923 int status;
David Howellsc4028952006-11-22 14:57:56 +00005924 struct ocfs2_super *osb =
5925 container_of(work, struct ocfs2_super,
5926 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08005927
5928 mlog_entry_void();
5929
5930 status = ocfs2_flush_truncate_log(osb);
5931 if (status < 0)
5932 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08005933 else
5934 ocfs2_init_inode_steal_slot(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08005935
5936 mlog_exit(status);
5937}
5938
5939#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
5940void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
5941 int cancel)
5942{
5943 if (osb->osb_tl_inode) {
5944 /* We want to push off log flushes while truncates are
5945 * still running. */
5946 if (cancel)
5947 cancel_delayed_work(&osb->osb_truncate_log_wq);
5948
5949 queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
5950 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
5951 }
5952}
5953
5954static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
5955 int slot_num,
5956 struct inode **tl_inode,
5957 struct buffer_head **tl_bh)
5958{
5959 int status;
5960 struct inode *inode = NULL;
5961 struct buffer_head *bh = NULL;
5962
5963 inode = ocfs2_get_system_file_inode(osb,
5964 TRUNCATE_LOG_SYSTEM_INODE,
5965 slot_num);
5966 if (!inode) {
5967 status = -EINVAL;
5968 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
5969 goto bail;
5970 }
5971
Joel Beckerb657c952008-11-13 14:49:11 -08005972 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005973 if (status < 0) {
5974 iput(inode);
5975 mlog_errno(status);
5976 goto bail;
5977 }
5978
5979 *tl_inode = inode;
5980 *tl_bh = bh;
5981bail:
5982 mlog_exit(status);
5983 return status;
5984}
5985
5986/* called during the 1st stage of node recovery. we stamp a clean
5987 * truncate log and pass back a copy for processing later. if the
5988 * truncate log does not require processing, a *tl_copy is set to
5989 * NULL. */
5990int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
5991 int slot_num,
5992 struct ocfs2_dinode **tl_copy)
5993{
5994 int status;
5995 struct inode *tl_inode = NULL;
5996 struct buffer_head *tl_bh = NULL;
5997 struct ocfs2_dinode *di;
5998 struct ocfs2_truncate_log *tl;
5999
6000 *tl_copy = NULL;
6001
6002 mlog(0, "recover truncate log from slot %d\n", slot_num);
6003
6004 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6005 if (status < 0) {
6006 mlog_errno(status);
6007 goto bail;
6008 }
6009
6010 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006011
Joel Becker10995aa2008-11-13 14:49:12 -08006012 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6013 * validated by the underlying call to ocfs2_read_inode_block(),
6014 * so any corruption is a code bug */
6015 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6016
6017 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006018 if (le16_to_cpu(tl->tl_used)) {
6019 mlog(0, "We'll have %u logs to recover\n",
6020 le16_to_cpu(tl->tl_used));
6021
6022 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6023 if (!(*tl_copy)) {
6024 status = -ENOMEM;
6025 mlog_errno(status);
6026 goto bail;
6027 }
6028
6029 /* Assuming the write-out below goes well, this copy
6030 * will be passed back to recovery for processing. */
6031 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6032
6033 /* All we need to do to clear the truncate log is set
6034 * tl_used. */
6035 tl->tl_used = 0;
6036
Joel Becker13723d02008-10-17 19:25:01 -07006037 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006038 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006039 if (status < 0) {
6040 mlog_errno(status);
6041 goto bail;
6042 }
6043 }
6044
6045bail:
6046 if (tl_inode)
6047 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006048 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006049
6050 if (status < 0 && (*tl_copy)) {
6051 kfree(*tl_copy);
6052 *tl_copy = NULL;
6053 }
6054
6055 mlog_exit(status);
6056 return status;
6057}
6058
6059int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6060 struct ocfs2_dinode *tl_copy)
6061{
6062 int status = 0;
6063 int i;
6064 unsigned int clusters, num_recs, start_cluster;
6065 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006066 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006067 struct inode *tl_inode = osb->osb_tl_inode;
6068 struct ocfs2_truncate_log *tl;
6069
6070 mlog_entry_void();
6071
6072 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6073 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6074 return -EINVAL;
6075 }
6076
6077 tl = &tl_copy->id2.i_dealloc;
6078 num_recs = le16_to_cpu(tl->tl_used);
Mark Fashehb0697052006-03-03 10:24:33 -08006079 mlog(0, "cleanup %u records from %llu\n", num_recs,
Mark Fasheh1ca1a112007-04-27 16:01:25 -07006080 (unsigned long long)le64_to_cpu(tl_copy->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08006081
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006082 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006083 for(i = 0; i < num_recs; i++) {
6084 if (ocfs2_truncate_log_needs_flush(osb)) {
6085 status = __ocfs2_flush_truncate_log(osb);
6086 if (status < 0) {
6087 mlog_errno(status);
6088 goto bail_up;
6089 }
6090 }
6091
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006092 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006093 if (IS_ERR(handle)) {
6094 status = PTR_ERR(handle);
6095 mlog_errno(status);
6096 goto bail_up;
6097 }
6098
6099 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6100 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6101 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6102
6103 status = ocfs2_truncate_log_append(osb, handle,
6104 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006105 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006106 if (status < 0) {
6107 mlog_errno(status);
6108 goto bail_up;
6109 }
6110 }
6111
6112bail_up:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006113 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006114
6115 mlog_exit(status);
6116 return status;
6117}
6118
6119void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6120{
6121 int status;
6122 struct inode *tl_inode = osb->osb_tl_inode;
6123
6124 mlog_entry_void();
6125
6126 if (tl_inode) {
6127 cancel_delayed_work(&osb->osb_truncate_log_wq);
6128 flush_workqueue(ocfs2_wq);
6129
6130 status = ocfs2_flush_truncate_log(osb);
6131 if (status < 0)
6132 mlog_errno(status);
6133
6134 brelse(osb->osb_tl_bh);
6135 iput(osb->osb_tl_inode);
6136 }
6137
6138 mlog_exit_void();
6139}
6140
6141int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6142{
6143 int status;
6144 struct inode *tl_inode = NULL;
6145 struct buffer_head *tl_bh = NULL;
6146
6147 mlog_entry_void();
6148
6149 status = ocfs2_get_truncate_log_info(osb,
6150 osb->slot_num,
6151 &tl_inode,
6152 &tl_bh);
6153 if (status < 0)
6154 mlog_errno(status);
6155
6156 /* ocfs2_truncate_log_shutdown keys on the existence of
6157 * osb->osb_tl_inode so we don't set any of the osb variables
6158 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006159 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6160 ocfs2_truncate_log_worker);
Mark Fashehccd979b2005-12-15 14:31:24 -08006161 osb->osb_tl_bh = tl_bh;
6162 osb->osb_tl_inode = tl_inode;
6163
6164 mlog_exit(status);
6165 return status;
6166}
6167
Mark Fasheh2b604352007-06-22 15:45:27 -07006168/*
6169 * Delayed de-allocation of suballocator blocks.
6170 *
6171 * Some sets of block de-allocations might involve multiple suballocator inodes.
6172 *
6173 * The locking for this can get extremely complicated, especially when
6174 * the suballocator inodes to delete from aren't known until deep
6175 * within an unrelated codepath.
6176 *
6177 * ocfs2_extent_block structures are a good example of this - an inode
6178 * btree could have been grown by any number of nodes each allocating
6179 * out of their own suballoc inode.
6180 *
6181 * These structures allow the delay of block de-allocation until a
6182 * later time, when locking of multiple cluster inodes won't cause
6183 * deadlock.
6184 */
6185
6186/*
Tao Ma2891d292008-11-12 08:26:58 +08006187 * Describe a single bit freed from a suballocator. For the block
6188 * suballocators, it represents one block. For the global cluster
6189 * allocator, it represents some clusters and free_bit indicates
6190 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006191 */
6192struct ocfs2_cached_block_free {
6193 struct ocfs2_cached_block_free *free_next;
6194 u64 free_blk;
6195 unsigned int free_bit;
6196};
6197
6198struct ocfs2_per_slot_free_list {
6199 struct ocfs2_per_slot_free_list *f_next_suballocator;
6200 int f_inode_type;
6201 int f_slot;
6202 struct ocfs2_cached_block_free *f_first;
6203};
6204
Tao Ma2891d292008-11-12 08:26:58 +08006205static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6206 int sysfile_type,
6207 int slot,
6208 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006209{
6210 int ret;
6211 u64 bg_blkno;
6212 handle_t *handle;
6213 struct inode *inode;
6214 struct buffer_head *di_bh = NULL;
6215 struct ocfs2_cached_block_free *tmp;
6216
6217 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6218 if (!inode) {
6219 ret = -EINVAL;
6220 mlog_errno(ret);
6221 goto out;
6222 }
6223
6224 mutex_lock(&inode->i_mutex);
6225
Mark Fashehe63aecb62007-10-18 15:30:42 -07006226 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006227 if (ret) {
6228 mlog_errno(ret);
6229 goto out_mutex;
6230 }
6231
6232 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6233 if (IS_ERR(handle)) {
6234 ret = PTR_ERR(handle);
6235 mlog_errno(ret);
6236 goto out_unlock;
6237 }
6238
6239 while (head) {
6240 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6241 head->free_bit);
6242 mlog(0, "Free bit: (bit %u, blkno %llu)\n",
6243 head->free_bit, (unsigned long long)head->free_blk);
6244
6245 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6246 head->free_bit, bg_blkno, 1);
6247 if (ret) {
6248 mlog_errno(ret);
6249 goto out_journal;
6250 }
6251
6252 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6253 if (ret) {
6254 mlog_errno(ret);
6255 goto out_journal;
6256 }
6257
6258 tmp = head;
6259 head = head->free_next;
6260 kfree(tmp);
6261 }
6262
6263out_journal:
6264 ocfs2_commit_trans(osb, handle);
6265
6266out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006267 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006268 brelse(di_bh);
6269out_mutex:
6270 mutex_unlock(&inode->i_mutex);
6271 iput(inode);
6272out:
6273 while(head) {
6274 /* Premature exit may have left some dangling items. */
6275 tmp = head;
6276 head = head->free_next;
6277 kfree(tmp);
6278 }
6279
6280 return ret;
6281}
6282
Tao Ma2891d292008-11-12 08:26:58 +08006283int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6284 u64 blkno, unsigned int bit)
6285{
6286 int ret = 0;
6287 struct ocfs2_cached_block_free *item;
6288
6289 item = kmalloc(sizeof(*item), GFP_NOFS);
6290 if (item == NULL) {
6291 ret = -ENOMEM;
6292 mlog_errno(ret);
6293 return ret;
6294 }
6295
6296 mlog(0, "Insert clusters: (bit %u, blk %llu)\n",
6297 bit, (unsigned long long)blkno);
6298
6299 item->free_blk = blkno;
6300 item->free_bit = bit;
6301 item->free_next = ctxt->c_global_allocator;
6302
6303 ctxt->c_global_allocator = item;
6304 return ret;
6305}
6306
6307static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6308 struct ocfs2_cached_block_free *head)
6309{
6310 struct ocfs2_cached_block_free *tmp;
6311 struct inode *tl_inode = osb->osb_tl_inode;
6312 handle_t *handle;
6313 int ret = 0;
6314
6315 mutex_lock(&tl_inode->i_mutex);
6316
6317 while (head) {
6318 if (ocfs2_truncate_log_needs_flush(osb)) {
6319 ret = __ocfs2_flush_truncate_log(osb);
6320 if (ret < 0) {
6321 mlog_errno(ret);
6322 break;
6323 }
6324 }
6325
6326 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6327 if (IS_ERR(handle)) {
6328 ret = PTR_ERR(handle);
6329 mlog_errno(ret);
6330 break;
6331 }
6332
6333 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6334 head->free_bit);
6335
6336 ocfs2_commit_trans(osb, handle);
6337 tmp = head;
6338 head = head->free_next;
6339 kfree(tmp);
6340
6341 if (ret < 0) {
6342 mlog_errno(ret);
6343 break;
6344 }
6345 }
6346
6347 mutex_unlock(&tl_inode->i_mutex);
6348
6349 while (head) {
6350 /* Premature exit may have left some dangling items. */
6351 tmp = head;
6352 head = head->free_next;
6353 kfree(tmp);
6354 }
6355
6356 return ret;
6357}
6358
Mark Fasheh2b604352007-06-22 15:45:27 -07006359int ocfs2_run_deallocs(struct ocfs2_super *osb,
6360 struct ocfs2_cached_dealloc_ctxt *ctxt)
6361{
6362 int ret = 0, ret2;
6363 struct ocfs2_per_slot_free_list *fl;
6364
6365 if (!ctxt)
6366 return 0;
6367
6368 while (ctxt->c_first_suballocator) {
6369 fl = ctxt->c_first_suballocator;
6370
6371 if (fl->f_first) {
6372 mlog(0, "Free items: (type %u, slot %d)\n",
6373 fl->f_inode_type, fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006374 ret2 = ocfs2_free_cached_blocks(osb,
6375 fl->f_inode_type,
6376 fl->f_slot,
6377 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006378 if (ret2)
6379 mlog_errno(ret2);
6380 if (!ret)
6381 ret = ret2;
6382 }
6383
6384 ctxt->c_first_suballocator = fl->f_next_suballocator;
6385 kfree(fl);
6386 }
6387
Tao Ma2891d292008-11-12 08:26:58 +08006388 if (ctxt->c_global_allocator) {
6389 ret2 = ocfs2_free_cached_clusters(osb,
6390 ctxt->c_global_allocator);
6391 if (ret2)
6392 mlog_errno(ret2);
6393 if (!ret)
6394 ret = ret2;
6395
6396 ctxt->c_global_allocator = NULL;
6397 }
6398
Mark Fasheh2b604352007-06-22 15:45:27 -07006399 return ret;
6400}
6401
6402static struct ocfs2_per_slot_free_list *
6403ocfs2_find_per_slot_free_list(int type,
6404 int slot,
6405 struct ocfs2_cached_dealloc_ctxt *ctxt)
6406{
6407 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6408
6409 while (fl) {
6410 if (fl->f_inode_type == type && fl->f_slot == slot)
6411 return fl;
6412
6413 fl = fl->f_next_suballocator;
6414 }
6415
6416 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6417 if (fl) {
6418 fl->f_inode_type = type;
6419 fl->f_slot = slot;
6420 fl->f_first = NULL;
6421 fl->f_next_suballocator = ctxt->c_first_suballocator;
6422
6423 ctxt->c_first_suballocator = fl;
6424 }
6425 return fl;
6426}
6427
6428static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6429 int type, int slot, u64 blkno,
6430 unsigned int bit)
6431{
6432 int ret;
6433 struct ocfs2_per_slot_free_list *fl;
6434 struct ocfs2_cached_block_free *item;
6435
6436 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6437 if (fl == NULL) {
6438 ret = -ENOMEM;
6439 mlog_errno(ret);
6440 goto out;
6441 }
6442
6443 item = kmalloc(sizeof(*item), GFP_NOFS);
6444 if (item == NULL) {
6445 ret = -ENOMEM;
6446 mlog_errno(ret);
6447 goto out;
6448 }
6449
6450 mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n",
6451 type, slot, bit, (unsigned long long)blkno);
6452
6453 item->free_blk = blkno;
6454 item->free_bit = bit;
6455 item->free_next = fl->f_first;
6456
6457 fl->f_first = item;
6458
6459 ret = 0;
6460out:
6461 return ret;
6462}
6463
Mark Fasheh59a5e412007-06-22 15:52:36 -07006464static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6465 struct ocfs2_extent_block *eb)
6466{
6467 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6468 le16_to_cpu(eb->h_suballoc_slot),
6469 le64_to_cpu(eb->h_blkno),
6470 le16_to_cpu(eb->h_suballoc_bit));
6471}
6472
Mark Fashehccd979b2005-12-15 14:31:24 -08006473/* This function will figure out whether the currently last extent
6474 * block will be deleted, and if it will, what the new last extent
6475 * block will be so we can update his h_next_leaf_blk field, as well
6476 * as the dinodes i_last_eb_blk */
Mark Fashehdcd05382007-01-16 11:32:23 -08006477static int ocfs2_find_new_last_ext_blk(struct inode *inode,
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006478 unsigned int clusters_to_del,
Mark Fashehdcd05382007-01-16 11:32:23 -08006479 struct ocfs2_path *path,
Mark Fashehccd979b2005-12-15 14:31:24 -08006480 struct buffer_head **new_last_eb)
6481{
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006482 int next_free, ret = 0;
Mark Fashehdcd05382007-01-16 11:32:23 -08006483 u32 cpos;
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006484 struct ocfs2_extent_rec *rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08006485 struct ocfs2_extent_block *eb;
6486 struct ocfs2_extent_list *el;
6487 struct buffer_head *bh = NULL;
6488
6489 *new_last_eb = NULL;
6490
Mark Fashehccd979b2005-12-15 14:31:24 -08006491 /* we have no tree, so of course, no last_eb. */
Mark Fashehdcd05382007-01-16 11:32:23 -08006492 if (!path->p_tree_depth)
6493 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006494
6495 /* trunc to zero special case - this makes tree_depth = 0
6496 * regardless of what it is. */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006497 if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
Mark Fashehdcd05382007-01-16 11:32:23 -08006498 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006499
Mark Fashehdcd05382007-01-16 11:32:23 -08006500 el = path_leaf_el(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08006501 BUG_ON(!el->l_next_free_rec);
6502
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006503 /*
6504 * Make sure that this extent list will actually be empty
6505 * after we clear away the data. We can shortcut out if
6506 * there's more than one non-empty extent in the
6507 * list. Otherwise, a check of the remaining extent is
6508 * necessary.
6509 */
6510 next_free = le16_to_cpu(el->l_next_free_rec);
6511 rec = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08006512 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006513 if (next_free > 2)
Mark Fashehdcd05382007-01-16 11:32:23 -08006514 goto out;
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006515
6516 /* We may have a valid extent in index 1, check it. */
6517 if (next_free == 2)
6518 rec = &el->l_recs[1];
6519
6520 /*
6521 * Fall through - no more nonempty extents, so we want
6522 * to delete this leaf.
6523 */
6524 } else {
6525 if (next_free > 1)
6526 goto out;
6527
6528 rec = &el->l_recs[0];
6529 }
6530
6531 if (rec) {
6532 /*
6533 * Check it we'll only be trimming off the end of this
6534 * cluster.
6535 */
Mark Fashehe48edee2007-03-07 16:46:57 -08006536 if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del)
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006537 goto out;
6538 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006539
Mark Fashehdcd05382007-01-16 11:32:23 -08006540 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
6541 if (ret) {
6542 mlog_errno(ret);
6543 goto out;
6544 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006545
Joel Beckerfacdb772009-02-12 18:08:48 -08006546 ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08006547 if (ret) {
6548 mlog_errno(ret);
6549 goto out;
6550 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006551
Mark Fashehdcd05382007-01-16 11:32:23 -08006552 eb = (struct ocfs2_extent_block *) bh->b_data;
6553 el = &eb->h_list;
Joel Becker5e965812008-11-13 14:49:16 -08006554
6555 /* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block().
6556 * Any corruption is a code bug. */
6557 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08006558
6559 *new_last_eb = bh;
6560 get_bh(*new_last_eb);
Mark Fashehdcd05382007-01-16 11:32:23 -08006561 mlog(0, "returning block %llu, (cpos: %u)\n",
6562 (unsigned long long)le64_to_cpu(eb->h_blkno), cpos);
6563out:
6564 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006565
Mark Fashehdcd05382007-01-16 11:32:23 -08006566 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08006567}
6568
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006569/*
6570 * Trim some clusters off the rightmost edge of a tree. Only called
6571 * during truncate.
6572 *
6573 * The caller needs to:
6574 * - start journaling of each path component.
6575 * - compute and fully set up any new last ext block
6576 */
6577static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path,
6578 handle_t *handle, struct ocfs2_truncate_context *tc,
6579 u32 clusters_to_del, u64 *delete_start)
6580{
6581 int ret, i, index = path->p_tree_depth;
6582 u32 new_edge = 0;
6583 u64 deleted_eb = 0;
6584 struct buffer_head *bh;
6585 struct ocfs2_extent_list *el;
6586 struct ocfs2_extent_rec *rec;
6587
6588 *delete_start = 0;
6589
6590 while (index >= 0) {
6591 bh = path->p_node[index].bh;
6592 el = path->p_node[index].el;
6593
6594 mlog(0, "traveling tree (index = %d, block = %llu)\n",
6595 index, (unsigned long long)bh->b_blocknr);
6596
6597 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
6598
6599 if (index !=
6600 (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) {
6601 ocfs2_error(inode->i_sb,
6602 "Inode %lu has invalid ext. block %llu",
6603 inode->i_ino,
6604 (unsigned long long)bh->b_blocknr);
6605 ret = -EROFS;
6606 goto out;
6607 }
6608
6609find_tail_record:
6610 i = le16_to_cpu(el->l_next_free_rec) - 1;
6611 rec = &el->l_recs[i];
6612
6613 mlog(0, "Extent list before: record %d: (%u, %u, %llu), "
6614 "next = %u\n", i, le32_to_cpu(rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08006615 ocfs2_rec_clusters(el, rec),
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006616 (unsigned long long)le64_to_cpu(rec->e_blkno),
6617 le16_to_cpu(el->l_next_free_rec));
6618
Mark Fashehe48edee2007-03-07 16:46:57 -08006619 BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006620
6621 if (le16_to_cpu(el->l_tree_depth) == 0) {
6622 /*
6623 * If the leaf block contains a single empty
6624 * extent and no records, we can just remove
6625 * the block.
6626 */
6627 if (i == 0 && ocfs2_is_empty_extent(rec)) {
6628 memset(rec, 0,
6629 sizeof(struct ocfs2_extent_rec));
6630 el->l_next_free_rec = cpu_to_le16(0);
6631
6632 goto delete;
6633 }
6634
6635 /*
6636 * Remove any empty extents by shifting things
6637 * left. That should make life much easier on
6638 * the code below. This condition is rare
6639 * enough that we shouldn't see a performance
6640 * hit.
6641 */
6642 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
6643 le16_add_cpu(&el->l_next_free_rec, -1);
6644
6645 for(i = 0;
6646 i < le16_to_cpu(el->l_next_free_rec); i++)
6647 el->l_recs[i] = el->l_recs[i + 1];
6648
6649 memset(&el->l_recs[i], 0,
6650 sizeof(struct ocfs2_extent_rec));
6651
6652 /*
6653 * We've modified our extent list. The
6654 * simplest way to handle this change
6655 * is to being the search from the
6656 * start again.
6657 */
6658 goto find_tail_record;
6659 }
6660
Mark Fashehe48edee2007-03-07 16:46:57 -08006661 le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006662
6663 /*
6664 * We'll use "new_edge" on our way back up the
6665 * tree to know what our rightmost cpos is.
6666 */
Mark Fashehe48edee2007-03-07 16:46:57 -08006667 new_edge = le16_to_cpu(rec->e_leaf_clusters);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006668 new_edge += le32_to_cpu(rec->e_cpos);
6669
6670 /*
6671 * The caller will use this to delete data blocks.
6672 */
6673 *delete_start = le64_to_cpu(rec->e_blkno)
6674 + ocfs2_clusters_to_blocks(inode->i_sb,
Mark Fashehe48edee2007-03-07 16:46:57 -08006675 le16_to_cpu(rec->e_leaf_clusters));
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006676
6677 /*
6678 * If it's now empty, remove this record.
6679 */
Mark Fashehe48edee2007-03-07 16:46:57 -08006680 if (le16_to_cpu(rec->e_leaf_clusters) == 0) {
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006681 memset(rec, 0,
6682 sizeof(struct ocfs2_extent_rec));
6683 le16_add_cpu(&el->l_next_free_rec, -1);
6684 }
6685 } else {
6686 if (le64_to_cpu(rec->e_blkno) == deleted_eb) {
6687 memset(rec, 0,
6688 sizeof(struct ocfs2_extent_rec));
6689 le16_add_cpu(&el->l_next_free_rec, -1);
6690
6691 goto delete;
6692 }
6693
6694 /* Can this actually happen? */
6695 if (le16_to_cpu(el->l_next_free_rec) == 0)
6696 goto delete;
6697
6698 /*
6699 * We never actually deleted any clusters
6700 * because our leaf was empty. There's no
6701 * reason to adjust the rightmost edge then.
6702 */
6703 if (new_edge == 0)
6704 goto delete;
6705
Mark Fashehe48edee2007-03-07 16:46:57 -08006706 rec->e_int_clusters = cpu_to_le32(new_edge);
6707 le32_add_cpu(&rec->e_int_clusters,
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006708 -le32_to_cpu(rec->e_cpos));
6709
6710 /*
6711 * A deleted child record should have been
6712 * caught above.
6713 */
Mark Fashehe48edee2007-03-07 16:46:57 -08006714 BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006715 }
6716
6717delete:
6718 ret = ocfs2_journal_dirty(handle, bh);
6719 if (ret) {
6720 mlog_errno(ret);
6721 goto out;
6722 }
6723
6724 mlog(0, "extent list container %llu, after: record %d: "
6725 "(%u, %u, %llu), next = %u.\n",
6726 (unsigned long long)bh->b_blocknr, i,
Mark Fashehe48edee2007-03-07 16:46:57 -08006727 le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec),
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006728 (unsigned long long)le64_to_cpu(rec->e_blkno),
6729 le16_to_cpu(el->l_next_free_rec));
6730
6731 /*
6732 * We must be careful to only attempt delete of an
6733 * extent block (and not the root inode block).
6734 */
6735 if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) {
6736 struct ocfs2_extent_block *eb =
6737 (struct ocfs2_extent_block *)bh->b_data;
6738
6739 /*
6740 * Save this for use when processing the
6741 * parent block.
6742 */
6743 deleted_eb = le64_to_cpu(eb->h_blkno);
6744
6745 mlog(0, "deleting this extent block.\n");
6746
Joel Becker8cb471e2009-02-10 20:00:41 -08006747 ocfs2_remove_from_cache(INODE_CACHE(inode), bh);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006748
Mark Fashehe48edee2007-03-07 16:46:57 -08006749 BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0]));
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006750 BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
6751 BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));
6752
Mark Fasheh59a5e412007-06-22 15:52:36 -07006753 ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb);
6754 /* An error here is not fatal. */
6755 if (ret < 0)
6756 mlog_errno(ret);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006757 } else {
6758 deleted_eb = 0;
6759 }
6760
6761 index--;
6762 }
6763
6764 ret = 0;
6765out:
6766 return ret;
6767}
6768
Mark Fashehccd979b2005-12-15 14:31:24 -08006769static int ocfs2_do_truncate(struct ocfs2_super *osb,
6770 unsigned int clusters_to_del,
6771 struct inode *inode,
6772 struct buffer_head *fe_bh,
Mark Fasheh1fabe142006-10-09 18:11:45 -07006773 handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -08006774 struct ocfs2_truncate_context *tc,
6775 struct ocfs2_path *path)
Mark Fashehccd979b2005-12-15 14:31:24 -08006776{
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006777 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08006778 struct ocfs2_dinode *fe;
Mark Fashehccd979b2005-12-15 14:31:24 -08006779 struct ocfs2_extent_block *last_eb = NULL;
6780 struct ocfs2_extent_list *el;
Mark Fashehccd979b2005-12-15 14:31:24 -08006781 struct buffer_head *last_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08006782 u64 delete_blk = 0;
6783
6784 fe = (struct ocfs2_dinode *) fe_bh->b_data;
6785
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006786 status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del,
Mark Fashehdcd05382007-01-16 11:32:23 -08006787 path, &last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006788 if (status < 0) {
6789 mlog_errno(status);
6790 goto bail;
6791 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006792
Mark Fashehdcd05382007-01-16 11:32:23 -08006793 /*
6794 * Each component will be touched, so we might as well journal
6795 * here to avoid having to handle errors later.
6796 */
Joel Becker0cf2f762009-02-12 16:41:25 -08006797 status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006798 if (status < 0) {
6799 mlog_errno(status);
6800 goto bail;
Mark Fashehdcd05382007-01-16 11:32:23 -08006801 }
6802
6803 if (last_eb_bh) {
Joel Becker0cf2f762009-02-12 16:41:25 -08006804 status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07006805 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08006806 if (status < 0) {
6807 mlog_errno(status);
6808 goto bail;
6809 }
6810
6811 last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
6812 }
6813
6814 el = &(fe->id2.i_list);
6815
6816 /*
6817 * Lower levels depend on this never happening, but it's best
6818 * to check it up here before changing the tree.
6819 */
Mark Fashehe48edee2007-03-07 16:46:57 -08006820 if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) {
Mark Fashehdcd05382007-01-16 11:32:23 -08006821 ocfs2_error(inode->i_sb,
6822 "Inode %lu has an empty extent record, depth %u\n",
6823 inode->i_ino, le16_to_cpu(el->l_tree_depth));
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006824 status = -EROFS;
Mark Fashehccd979b2005-12-15 14:31:24 -08006825 goto bail;
6826 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006827
Jan Karaa90714c2008-10-09 19:38:40 +02006828 vfs_dq_free_space_nodirty(inode,
6829 ocfs2_clusters_to_bytes(osb->sb, clusters_to_del));
Mark Fashehccd979b2005-12-15 14:31:24 -08006830 spin_lock(&OCFS2_I(inode)->ip_lock);
6831 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) -
6832 clusters_to_del;
6833 spin_unlock(&OCFS2_I(inode)->ip_lock);
6834 le32_add_cpu(&fe->i_clusters, -clusters_to_del);
Mark Fashehe535e2e2007-08-31 10:23:41 -07006835 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006836
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006837 status = ocfs2_trim_tree(inode, path, handle, tc,
6838 clusters_to_del, &delete_blk);
6839 if (status) {
6840 mlog_errno(status);
6841 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08006842 }
6843
Mark Fashehdcd05382007-01-16 11:32:23 -08006844 if (le32_to_cpu(fe->i_clusters) == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006845 /* trunc to zero is a special case. */
6846 el->l_tree_depth = 0;
6847 fe->i_last_eb_blk = 0;
6848 } else if (last_eb)
6849 fe->i_last_eb_blk = last_eb->h_blkno;
6850
6851 status = ocfs2_journal_dirty(handle, fe_bh);
6852 if (status < 0) {
6853 mlog_errno(status);
6854 goto bail;
6855 }
6856
6857 if (last_eb) {
6858 /* If there will be a new last extent block, then by
6859 * definition, there cannot be any leaves to the right of
6860 * him. */
Mark Fashehccd979b2005-12-15 14:31:24 -08006861 last_eb->h_next_leaf_blk = 0;
6862 status = ocfs2_journal_dirty(handle, last_eb_bh);
6863 if (status < 0) {
6864 mlog_errno(status);
6865 goto bail;
6866 }
6867 }
6868
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006869 if (delete_blk) {
6870 status = ocfs2_truncate_log_append(osb, handle, delete_blk,
6871 clusters_to_del);
Mark Fashehccd979b2005-12-15 14:31:24 -08006872 if (status < 0) {
6873 mlog_errno(status);
6874 goto bail;
6875 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006876 }
6877 status = 0;
6878bail:
Tao Ma60e2ec42009-08-12 14:42:47 +08006879 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006880 mlog_exit(status);
6881 return status;
6882}
6883
Joel Becker2b4e30f2008-09-03 20:03:41 -07006884static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006885{
6886 set_buffer_uptodate(bh);
6887 mark_buffer_dirty(bh);
6888 return 0;
6889}
6890
Mark Fasheh1d410a62007-09-07 14:20:45 -07006891static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6892 unsigned int from, unsigned int to,
6893 struct page *page, int zero, u64 *phys)
6894{
6895 int ret, partial = 0;
6896
6897 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6898 if (ret)
6899 mlog_errno(ret);
6900
6901 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006902 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006903
6904 /*
6905 * Need to set the buffers we zero'd into uptodate
6906 * here if they aren't - ocfs2_map_page_blocks()
6907 * might've skipped some
6908 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006909 ret = walk_page_buffers(handle, page_buffers(page),
6910 from, to, &partial,
6911 ocfs2_zero_func);
6912 if (ret < 0)
6913 mlog_errno(ret);
6914 else if (ocfs2_should_order_data(inode)) {
6915 ret = ocfs2_jbd2_file_inode(handle, inode);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006916 if (ret < 0)
6917 mlog_errno(ret);
6918 }
6919
6920 if (!partial)
6921 SetPageUptodate(page);
6922
6923 flush_dcache_page(page);
6924}
6925
Mark Fasheh35edec12007-07-06 14:41:18 -07006926static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6927 loff_t end, struct page **pages,
6928 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006929{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006930 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006931 struct page *page;
6932 unsigned int from, to = PAGE_CACHE_SIZE;
6933 struct super_block *sb = inode->i_sb;
6934
6935 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6936
6937 if (numpages == 0)
6938 goto out;
6939
Mark Fasheh35edec12007-07-06 14:41:18 -07006940 to = PAGE_CACHE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006941 for(i = 0; i < numpages; i++) {
6942 page = pages[i];
6943
Mark Fasheh35edec12007-07-06 14:41:18 -07006944 from = start & (PAGE_CACHE_SIZE - 1);
6945 if ((end >> PAGE_CACHE_SHIFT) == page->index)
6946 to = end & (PAGE_CACHE_SIZE - 1);
6947
Mark Fasheh60b11392007-02-16 11:46:50 -08006948 BUG_ON(from > PAGE_CACHE_SIZE);
6949 BUG_ON(to > PAGE_CACHE_SIZE);
6950
Mark Fasheh1d410a62007-09-07 14:20:45 -07006951 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6952 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006953
Mark Fasheh35edec12007-07-06 14:41:18 -07006954 start = (page->index + 1) << PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006955 }
6956out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006957 if (pages)
6958 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006959}
6960
Mark Fasheh35edec12007-07-06 14:41:18 -07006961static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
Mark Fasheh1d410a62007-09-07 14:20:45 -07006962 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006963{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006964 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006965 struct super_block *sb = inode->i_sb;
6966 struct address_space *mapping = inode->i_mapping;
6967 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006968 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006969
Mark Fasheh35edec12007-07-06 14:41:18 -07006970 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006971
Mark Fasheh35edec12007-07-06 14:41:18 -07006972 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6973 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6974
Mark Fasheh1d410a62007-09-07 14:20:45 -07006975 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006976 last_page_bytes = PAGE_ALIGN(end);
6977 index = start >> PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006978 do {
6979 pages[numpages] = grab_cache_page(mapping, index);
6980 if (!pages[numpages]) {
6981 ret = -ENOMEM;
6982 mlog_errno(ret);
6983 goto out;
6984 }
6985
6986 numpages++;
6987 index++;
Mark Fasheh35edec12007-07-06 14:41:18 -07006988 } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006989
6990out:
6991 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006992 if (pages)
6993 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006994 numpages = 0;
6995 }
6996
6997 *num = numpages;
6998
6999 return ret;
7000}
7001
7002/*
7003 * Zero the area past i_size but still within an allocated
7004 * cluster. This avoids exposing nonzero data on subsequent file
7005 * extends.
7006 *
7007 * We need to call this before i_size is updated on the inode because
7008 * otherwise block_write_full_page() will skip writeout of pages past
7009 * i_size. The new_i_size parameter is passed for this reason.
7010 */
Mark Fasheh35edec12007-07-06 14:41:18 -07007011int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
7012 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08007013{
Mark Fasheh1d410a62007-09-07 14:20:45 -07007014 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08007015 struct page **pages = NULL;
7016 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07007017 unsigned int ext_flags;
7018 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08007019
7020 /*
7021 * File systems which don't support sparse files zero on every
7022 * extend.
7023 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07007024 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08007025 return 0;
7026
Mark Fasheh1d410a62007-09-07 14:20:45 -07007027 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08007028 sizeof(struct page *), GFP_NOFS);
7029 if (pages == NULL) {
7030 ret = -ENOMEM;
7031 mlog_errno(ret);
7032 goto out;
7033 }
7034
Mark Fasheh1d410a62007-09-07 14:20:45 -07007035 if (range_start == range_end)
7036 goto out;
7037
7038 ret = ocfs2_extent_map_get_blocks(inode,
7039 range_start >> sb->s_blocksize_bits,
7040 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08007041 if (ret) {
7042 mlog_errno(ret);
7043 goto out;
7044 }
7045
Mark Fasheh1d410a62007-09-07 14:20:45 -07007046 /*
7047 * Tail is a hole, or is marked unwritten. In either case, we
7048 * can count on read and write to return/push zero's.
7049 */
7050 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08007051 goto out;
7052
Mark Fasheh1d410a62007-09-07 14:20:45 -07007053 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
7054 &numpages);
7055 if (ret) {
7056 mlog_errno(ret);
7057 goto out;
7058 }
7059
Mark Fasheh35edec12007-07-06 14:41:18 -07007060 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
7061 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08007062
7063 /*
7064 * Initiate writeout of the pages we zero'd here. We don't
7065 * wait on them - the truncate_inode_pages() call later will
7066 * do that for us.
7067 */
Mark Fasheh35edec12007-07-06 14:41:18 -07007068 ret = do_sync_mapping_range(inode->i_mapping, range_start,
7069 range_end - 1, SYNC_FILE_RANGE_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -08007070 if (ret)
7071 mlog_errno(ret);
7072
7073out:
7074 if (pages)
7075 kfree(pages);
7076
7077 return ret;
7078}
7079
Tiger Yangfdd77702008-08-18 17:08:55 +08007080static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
7081 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007082{
7083 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08007084 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007085
Tiger Yangfdd77702008-08-18 17:08:55 +08007086 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
7087 memset(&di->id2, 0, blocksize -
7088 offsetof(struct ocfs2_dinode, id2) -
7089 xattrsize);
7090 else
7091 memset(&di->id2, 0, blocksize -
7092 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007093}
7094
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007095void ocfs2_dinode_new_extent_list(struct inode *inode,
7096 struct ocfs2_dinode *di)
7097{
Tiger Yangfdd77702008-08-18 17:08:55 +08007098 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007099 di->id2.i_list.l_tree_depth = 0;
7100 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08007101 di->id2.i_list.l_count = cpu_to_le16(
7102 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007103}
7104
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007105void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
7106{
7107 struct ocfs2_inode_info *oi = OCFS2_I(inode);
7108 struct ocfs2_inline_data *idata = &di->id2.i_data;
7109
7110 spin_lock(&oi->ip_lock);
7111 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
7112 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
7113 spin_unlock(&oi->ip_lock);
7114
7115 /*
7116 * We clear the entire i_data structure here so that all
7117 * fields can be properly initialized.
7118 */
Tiger Yangfdd77702008-08-18 17:08:55 +08007119 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007120
Tiger Yangfdd77702008-08-18 17:08:55 +08007121 idata->id_count = cpu_to_le16(
7122 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007123}
7124
7125int ocfs2_convert_inline_data_to_extents(struct inode *inode,
7126 struct buffer_head *di_bh)
7127{
7128 int ret, i, has_data, num_pages = 0;
7129 handle_t *handle;
7130 u64 uninitialized_var(block);
7131 struct ocfs2_inode_info *oi = OCFS2_I(inode);
7132 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7133 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007134 struct ocfs2_alloc_context *data_ac = NULL;
7135 struct page **pages = NULL;
7136 loff_t end = osb->s_clustersize;
Joel Beckerf99b9b72008-08-20 19:36:33 -07007137 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02007138 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007139
7140 has_data = i_size_read(inode) ? 1 : 0;
7141
7142 if (has_data) {
7143 pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
7144 sizeof(struct page *), GFP_NOFS);
7145 if (pages == NULL) {
7146 ret = -ENOMEM;
7147 mlog_errno(ret);
7148 goto out;
7149 }
7150
7151 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
7152 if (ret) {
7153 mlog_errno(ret);
7154 goto out;
7155 }
7156 }
7157
Jan Karaa90714c2008-10-09 19:38:40 +02007158 handle = ocfs2_start_trans(osb,
7159 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007160 if (IS_ERR(handle)) {
7161 ret = PTR_ERR(handle);
7162 mlog_errno(ret);
7163 goto out_unlock;
7164 }
7165
Joel Becker0cf2f762009-02-12 16:41:25 -08007166 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007167 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007168 if (ret) {
7169 mlog_errno(ret);
7170 goto out_commit;
7171 }
7172
7173 if (has_data) {
7174 u32 bit_off, num;
7175 unsigned int page_end;
7176 u64 phys;
7177
Jan Karaa90714c2008-10-09 19:38:40 +02007178 if (vfs_dq_alloc_space_nodirty(inode,
7179 ocfs2_clusters_to_bytes(osb->sb, 1))) {
7180 ret = -EDQUOT;
7181 goto out_commit;
7182 }
7183 did_quota = 1;
7184
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007185 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
7186 &num);
7187 if (ret) {
7188 mlog_errno(ret);
7189 goto out_commit;
7190 }
7191
7192 /*
7193 * Save two copies, one for insert, and one that can
7194 * be changed by ocfs2_map_and_dirty_page() below.
7195 */
7196 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
7197
7198 /*
7199 * Non sparse file systems zero on extend, so no need
7200 * to do that now.
7201 */
7202 if (!ocfs2_sparse_alloc(osb) &&
7203 PAGE_CACHE_SIZE < osb->s_clustersize)
7204 end = PAGE_CACHE_SIZE;
7205
7206 ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
7207 if (ret) {
7208 mlog_errno(ret);
7209 goto out_commit;
7210 }
7211
7212 /*
7213 * This should populate the 1st page for us and mark
7214 * it up to date.
7215 */
7216 ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
7217 if (ret) {
7218 mlog_errno(ret);
7219 goto out_commit;
7220 }
7221
7222 page_end = PAGE_CACHE_SIZE;
7223 if (PAGE_CACHE_SIZE > osb->s_clustersize)
7224 page_end = osb->s_clustersize;
7225
7226 for (i = 0; i < num_pages; i++)
7227 ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
7228 pages[i], i > 0, &phys);
7229 }
7230
7231 spin_lock(&oi->ip_lock);
7232 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
7233 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
7234 spin_unlock(&oi->ip_lock);
7235
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007236 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007237
7238 ocfs2_journal_dirty(handle, di_bh);
7239
7240 if (has_data) {
7241 /*
7242 * An error at this point should be extremely rare. If
7243 * this proves to be false, we could always re-build
7244 * the in-inode data from our pages.
7245 */
Joel Becker8d6220d2008-08-22 12:46:09 -07007246 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08007247 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007248 if (ret) {
7249 mlog_errno(ret);
7250 goto out_commit;
7251 }
7252
7253 inode->i_blocks = ocfs2_inode_sector_count(inode);
7254 }
7255
7256out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02007257 if (ret < 0 && did_quota)
7258 vfs_dq_free_space_nodirty(inode,
7259 ocfs2_clusters_to_bytes(osb->sb, 1));
7260
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007261 ocfs2_commit_trans(osb, handle);
7262
7263out_unlock:
7264 if (data_ac)
7265 ocfs2_free_alloc_context(data_ac);
7266
7267out:
7268 if (pages) {
7269 ocfs2_unlock_and_free_pages(pages, num_pages);
7270 kfree(pages);
7271 }
7272
7273 return ret;
7274}
7275
Mark Fashehccd979b2005-12-15 14:31:24 -08007276/*
7277 * It is expected, that by the time you call this function,
7278 * inode->i_size and fe->i_size have been adjusted.
7279 *
7280 * WARNING: This will kfree the truncate context
7281 */
7282int ocfs2_commit_truncate(struct ocfs2_super *osb,
7283 struct inode *inode,
7284 struct buffer_head *fe_bh,
7285 struct ocfs2_truncate_context *tc)
7286{
7287 int status, i, credits, tl_sem = 0;
Mark Fashehdcd05382007-01-16 11:32:23 -08007288 u32 clusters_to_del, new_highest_cpos, range;
Mark Fashehccd979b2005-12-15 14:31:24 -08007289 struct ocfs2_extent_list *el;
Mark Fasheh1fabe142006-10-09 18:11:45 -07007290 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007291 struct inode *tl_inode = osb->osb_tl_inode;
Mark Fashehdcd05382007-01-16 11:32:23 -08007292 struct ocfs2_path *path = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +08007293 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08007294
7295 mlog_entry_void();
7296
Mark Fashehdcd05382007-01-16 11:32:23 -08007297 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08007298 i_size_read(inode));
7299
Joel Becker13723d02008-10-17 19:25:01 -07007300 path = ocfs2_new_path(fe_bh, &di->id2.i_list,
7301 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08007302 if (!path) {
7303 status = -ENOMEM;
7304 mlog_errno(status);
7305 goto bail;
7306 }
Mark Fasheh83418972007-04-23 18:53:12 -07007307
7308 ocfs2_extent_map_trunc(inode, new_highest_cpos);
7309
Mark Fashehccd979b2005-12-15 14:31:24 -08007310start:
Mark Fashehdcd05382007-01-16 11:32:23 -08007311 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007312 * Check that we still have allocation to delete.
7313 */
7314 if (OCFS2_I(inode)->ip_clusters == 0) {
7315 status = 0;
7316 goto bail;
7317 }
7318
7319 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007320 * Truncate always works against the rightmost tree branch.
7321 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007322 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007323 if (status) {
7324 mlog_errno(status);
7325 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007326 }
7327
Mark Fashehdcd05382007-01-16 11:32:23 -08007328 mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n",
7329 OCFS2_I(inode)->ip_clusters, path->p_tree_depth);
7330
7331 /*
7332 * By now, el will point to the extent list on the bottom most
7333 * portion of this tree. Only the tail record is considered in
7334 * each pass.
7335 *
7336 * We handle the following cases, in order:
7337 * - empty extent: delete the remaining branch
7338 * - remove the entire record
7339 * - remove a partial record
7340 * - no record needs to be removed (truncate has completed)
7341 */
7342 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007343 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7344 ocfs2_error(inode->i_sb,
7345 "Inode %llu has empty extent block at %llu\n",
7346 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7347 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7348 status = -EROFS;
7349 goto bail;
7350 }
7351
Mark Fashehccd979b2005-12-15 14:31:24 -08007352 i = le16_to_cpu(el->l_next_free_rec) - 1;
Mark Fashehdcd05382007-01-16 11:32:23 -08007353 range = le32_to_cpu(el->l_recs[i].e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08007354 ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08007355 if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) {
7356 clusters_to_del = 0;
7357 } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) {
Mark Fashehe48edee2007-03-07 16:46:57 -08007358 clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08007359 } else if (range > new_highest_cpos) {
Mark Fashehe48edee2007-03-07 16:46:57 -08007360 clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) +
Mark Fashehccd979b2005-12-15 14:31:24 -08007361 le32_to_cpu(el->l_recs[i].e_cpos)) -
Mark Fashehdcd05382007-01-16 11:32:23 -08007362 new_highest_cpos;
7363 } else {
7364 status = 0;
7365 goto bail;
7366 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007367
Mark Fashehdcd05382007-01-16 11:32:23 -08007368 mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n",
7369 clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr);
7370
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08007371 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08007372 tl_sem = 1;
7373 /* ocfs2_truncate_log_needs_flush guarantees us at least one
7374 * record is free for use. If there isn't any, we flush to get
7375 * an empty truncate log. */
7376 if (ocfs2_truncate_log_needs_flush(osb)) {
7377 status = __ocfs2_flush_truncate_log(osb);
7378 if (status < 0) {
7379 mlog_errno(status);
7380 goto bail;
7381 }
7382 }
7383
7384 credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
Mark Fashehdcd05382007-01-16 11:32:23 -08007385 (struct ocfs2_dinode *)fe_bh->b_data,
7386 el);
Mark Fasheh65eff9c2006-10-09 17:26:22 -07007387 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -08007388 if (IS_ERR(handle)) {
7389 status = PTR_ERR(handle);
7390 handle = NULL;
7391 mlog_errno(status);
7392 goto bail;
7393 }
7394
Mark Fashehdcd05382007-01-16 11:32:23 -08007395 status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle,
7396 tc, path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007397 if (status < 0) {
7398 mlog_errno(status);
7399 goto bail;
7400 }
7401
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08007402 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08007403 tl_sem = 0;
7404
Mark Fasheh02dc1af2006-10-09 16:48:10 -07007405 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08007406 handle = NULL;
7407
Mark Fashehdcd05382007-01-16 11:32:23 -08007408 ocfs2_reinit_path(path, 1);
7409
7410 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007411 * The check above will catch the case where we've truncated
7412 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007413 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007414 goto start;
7415
Mark Fashehccd979b2005-12-15 14:31:24 -08007416bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08007417
7418 ocfs2_schedule_truncate_log_flush(osb, 1);
7419
7420 if (tl_sem)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08007421 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08007422
7423 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -07007424 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08007425
Mark Fasheh59a5e412007-06-22 15:52:36 -07007426 ocfs2_run_deallocs(osb, &tc->tc_dealloc);
7427
Mark Fashehdcd05382007-01-16 11:32:23 -08007428 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007429
7430 /* This will drop the ext_alloc cluster lock for us */
7431 ocfs2_free_truncate_context(tc);
7432
7433 mlog_exit(status);
7434 return status;
7435}
7436
Mark Fashehccd979b2005-12-15 14:31:24 -08007437/*
Mark Fasheh59a5e412007-06-22 15:52:36 -07007438 * Expects the inode to already be locked.
Mark Fashehccd979b2005-12-15 14:31:24 -08007439 */
7440int ocfs2_prepare_truncate(struct ocfs2_super *osb,
7441 struct inode *inode,
7442 struct buffer_head *fe_bh,
7443 struct ocfs2_truncate_context **tc)
7444{
Mark Fasheh59a5e412007-06-22 15:52:36 -07007445 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08007446 unsigned int new_i_clusters;
7447 struct ocfs2_dinode *fe;
7448 struct ocfs2_extent_block *eb;
Mark Fashehccd979b2005-12-15 14:31:24 -08007449 struct buffer_head *last_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007450
7451 mlog_entry_void();
7452
7453 *tc = NULL;
7454
7455 new_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
7456 i_size_read(inode));
7457 fe = (struct ocfs2_dinode *) fe_bh->b_data;
7458
7459 mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size ="
Mark Fasheh1ca1a112007-04-27 16:01:25 -07007460 "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters,
7461 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -08007462
Robert P. J. Daycd861282006-12-13 00:34:52 -08007463 *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -08007464 if (!(*tc)) {
7465 status = -ENOMEM;
7466 mlog_errno(status);
7467 goto bail;
7468 }
Mark Fasheh59a5e412007-06-22 15:52:36 -07007469 ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc);
Mark Fashehccd979b2005-12-15 14:31:24 -08007470
Mark Fashehccd979b2005-12-15 14:31:24 -08007471 if (fe->id2.i_list.l_tree_depth) {
Joel Becker3d03a302009-02-12 17:49:26 -08007472 status = ocfs2_read_extent_block(INODE_CACHE(inode),
Joel Becker5e965812008-11-13 14:49:16 -08007473 le64_to_cpu(fe->i_last_eb_blk),
7474 &last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08007475 if (status < 0) {
7476 mlog_errno(status);
7477 goto bail;
7478 }
7479 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08007480 }
7481
7482 (*tc)->tc_last_eb_bh = last_eb_bh;
7483
Mark Fashehccd979b2005-12-15 14:31:24 -08007484 status = 0;
7485bail:
7486 if (status < 0) {
7487 if (*tc)
7488 ocfs2_free_truncate_context(*tc);
7489 *tc = NULL;
7490 }
7491 mlog_exit_void();
7492 return status;
7493}
7494
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007495/*
7496 * 'start' is inclusive, 'end' is not.
7497 */
7498int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7499 unsigned int start, unsigned int end, int trunc)
7500{
7501 int ret;
7502 unsigned int numbytes;
7503 handle_t *handle;
7504 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7505 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7506 struct ocfs2_inline_data *idata = &di->id2.i_data;
7507
7508 if (end > i_size_read(inode))
7509 end = i_size_read(inode);
7510
7511 BUG_ON(start >= end);
7512
7513 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7514 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7515 !ocfs2_supports_inline_data(osb)) {
7516 ocfs2_error(inode->i_sb,
7517 "Inline data flags for inode %llu don't agree! "
7518 "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
7519 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7520 le16_to_cpu(di->i_dyn_features),
7521 OCFS2_I(inode)->ip_dyn_features,
7522 osb->s_feature_incompat);
7523 ret = -EROFS;
7524 goto out;
7525 }
7526
7527 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7528 if (IS_ERR(handle)) {
7529 ret = PTR_ERR(handle);
7530 mlog_errno(ret);
7531 goto out;
7532 }
7533
Joel Becker0cf2f762009-02-12 16:41:25 -08007534 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007535 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007536 if (ret) {
7537 mlog_errno(ret);
7538 goto out_commit;
7539 }
7540
7541 numbytes = end - start;
7542 memset(idata->id_data + start, 0, numbytes);
7543
7544 /*
7545 * No need to worry about the data page here - it's been
7546 * truncated already and inline data doesn't need it for
7547 * pushing zero's to disk, so we'll let readpage pick it up
7548 * later.
7549 */
7550 if (trunc) {
7551 i_size_write(inode, start);
7552 di->i_size = cpu_to_le64(start);
7553 }
7554
7555 inode->i_blocks = ocfs2_inode_sector_count(inode);
7556 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7557
7558 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7559 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7560
7561 ocfs2_journal_dirty(handle, di_bh);
7562
7563out_commit:
7564 ocfs2_commit_trans(osb, handle);
7565
7566out:
7567 return ret;
7568}
7569
Mark Fashehccd979b2005-12-15 14:31:24 -08007570static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
7571{
Mark Fasheh59a5e412007-06-22 15:52:36 -07007572 /*
7573 * The caller is responsible for completing deallocation
7574 * before freeing the context.
7575 */
7576 if (tc->tc_dealloc.c_first_suballocator != NULL)
7577 mlog(ML_NOTICE,
7578 "Truncate completion has non-empty dealloc context\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08007579
Mark Fasheha81cb882008-10-07 14:25:16 -07007580 brelse(tc->tc_last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08007581
7582 kfree(tc);
7583}