blob: ed553c60de827e0ebad24e3501e0e00d21c82cfc [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>
Tao Mae80de362011-05-23 10:36:43 +080032#include <linux/blkdev.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080033
Mark Fashehccd979b2005-12-15 14:31:24 -080034#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"
Tao Mabcbbb242009-08-18 11:29:12 +080052#include "refcounttree.h"
Tao Maa09d09b2011-02-22 08:21:10 +080053#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080054
55#include "buffer_head_io.h"
56
Tao Ma853a3a12009-08-18 11:22:18 +080057enum ocfs2_contig_type {
58 CONTIG_NONE = 0,
59 CONTIG_LEFT,
60 CONTIG_RIGHT,
61 CONTIG_LEFTRIGHT,
62};
Tao Mae7d4cb62008-08-18 17:38:44 +080063
Tao Ma853a3a12009-08-18 11:22:18 +080064static enum ocfs2_contig_type
65 ocfs2_extent_rec_contig(struct super_block *sb,
66 struct ocfs2_extent_rec *ext,
67 struct ocfs2_extent_rec *insert_rec);
Joel Becker1625f8a2008-08-21 17:11:10 -070068/*
69 * Operations for a specific extent tree type.
70 *
71 * To implement an on-disk btree (extent tree) type in ocfs2, add
72 * an ocfs2_extent_tree_operations structure and the matching
Joel Becker8d6220d2008-08-22 12:46:09 -070073 * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it
Joel Becker1625f8a2008-08-21 17:11:10 -070074 * for the allocation portion of the extent tree.
75 */
Tao Mae7d4cb62008-08-18 17:38:44 +080076struct ocfs2_extent_tree_operations {
Joel Becker1625f8a2008-08-21 17:11:10 -070077 /*
78 * last_eb_blk is the block number of the right most leaf extent
79 * block. Most on-disk structures containing an extent tree store
80 * this value for fast access. The ->eo_set_last_eb_blk() and
81 * ->eo_get_last_eb_blk() operations access this value. They are
82 * both required.
83 */
Joel Becker35dc0aa2008-08-20 16:25:06 -070084 void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
85 u64 blkno);
86 u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -070087
88 /*
89 * The on-disk structure usually keeps track of how many total
90 * clusters are stored in this extent tree. This function updates
91 * that value. new_clusters is the delta, and must be
92 * added to the total. Required.
93 */
Joel Becker6136ca52009-02-12 19:32:43 -080094 void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -070095 u32 new_clusters);
Joel Becker1625f8a2008-08-21 17:11:10 -070096
97 /*
Joel Becker92ba4702009-02-13 03:18:34 -080098 * If this extent tree is supported by an extent map, insert
99 * a record into the map.
100 */
101 void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
102 struct ocfs2_extent_rec *rec);
103
104 /*
Joel Becker4c911ee2009-02-13 02:50:12 -0800105 * If this extent tree is supported by an extent map, truncate the
106 * map to clusters,
107 */
108 void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
109 u32 clusters);
110
111 /*
Joel Becker1625f8a2008-08-21 17:11:10 -0700112 * If ->eo_insert_check() exists, it is called before rec is
113 * inserted into the extent tree. It is optional.
114 */
Joel Becker6136ca52009-02-12 19:32:43 -0800115 int (*eo_insert_check)(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700116 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800117 int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
Joel Becker0ce10102008-08-20 17:19:50 -0700118
Joel Becker1625f8a2008-08-21 17:11:10 -0700119 /*
120 * --------------------------------------------------------------
121 * The remaining are internal to ocfs2_extent_tree and don't have
122 * accessor functions
123 */
124
125 /*
126 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
127 * It is required.
128 */
Joel Becker0ce10102008-08-20 17:19:50 -0700129 void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -0700130
131 /*
132 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
133 * it exists. If it does not, et->et_max_leaf_clusters is set
134 * to 0 (unlimited). Optional.
135 */
Joel Becker6136ca52009-02-12 19:32:43 -0800136 void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
Tao Ma853a3a12009-08-18 11:22:18 +0800137
138 /*
139 * ->eo_extent_contig test whether the 2 ocfs2_extent_rec
140 * are contiguous or not. Optional. Don't need to set it if use
141 * ocfs2_extent_rec as the tree leaf.
142 */
143 enum ocfs2_contig_type
144 (*eo_extent_contig)(struct ocfs2_extent_tree *et,
145 struct ocfs2_extent_rec *ext,
146 struct ocfs2_extent_rec *insert_rec);
Tao Mae7d4cb62008-08-18 17:38:44 +0800147};
148
Joel Beckerf99b9b72008-08-20 19:36:33 -0700149
150/*
151 * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
152 * in the methods.
153 */
154static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
155static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
156 u64 blkno);
Joel Becker6136ca52009-02-12 19:32:43 -0800157static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700158 u32 clusters);
Joel Becker92ba4702009-02-13 03:18:34 -0800159static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
160 struct ocfs2_extent_rec *rec);
Joel Becker4c911ee2009-02-13 02:50:12 -0800161static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
162 u32 clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800163static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700164 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800165static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700166static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
167static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
168 .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk,
169 .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk,
170 .eo_update_clusters = ocfs2_dinode_update_clusters,
Joel Becker92ba4702009-02-13 03:18:34 -0800171 .eo_extent_map_insert = ocfs2_dinode_extent_map_insert,
Joel Becker4c911ee2009-02-13 02:50:12 -0800172 .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700173 .eo_insert_check = ocfs2_dinode_insert_check,
174 .eo_sanity_check = ocfs2_dinode_sanity_check,
175 .eo_fill_root_el = ocfs2_dinode_fill_root_el,
Tao Mae7d4cb62008-08-18 17:38:44 +0800176};
177
178static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
179 u64 blkno)
180{
Joel Beckerea5efa12008-08-20 16:57:27 -0700181 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800182
Joel Beckerf99b9b72008-08-20 19:36:33 -0700183 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800184 di->i_last_eb_blk = cpu_to_le64(blkno);
185}
186
187static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
188{
Joel Beckerea5efa12008-08-20 16:57:27 -0700189 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800190
Joel Beckerf99b9b72008-08-20 19:36:33 -0700191 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800192 return le64_to_cpu(di->i_last_eb_blk);
193}
194
Joel Becker6136ca52009-02-12 19:32:43 -0800195static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +0800196 u32 clusters)
197{
Joel Becker6136ca52009-02-12 19:32:43 -0800198 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
Joel Beckerea5efa12008-08-20 16:57:27 -0700199 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800200
201 le32_add_cpu(&di->i_clusters, clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800202 spin_lock(&oi->ip_lock);
203 oi->ip_clusters = le32_to_cpu(di->i_clusters);
204 spin_unlock(&oi->ip_lock);
Tao Mae7d4cb62008-08-18 17:38:44 +0800205}
206
Joel Becker92ba4702009-02-13 03:18:34 -0800207static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
208 struct ocfs2_extent_rec *rec)
209{
210 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
211
212 ocfs2_extent_map_insert_rec(inode, rec);
213}
214
Joel Becker4c911ee2009-02-13 02:50:12 -0800215static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
216 u32 clusters)
217{
218 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
219
220 ocfs2_extent_map_trunc(inode, clusters);
221}
222
Joel Becker6136ca52009-02-12 19:32:43 -0800223static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700224 struct ocfs2_extent_rec *rec)
225{
Joel Becker6136ca52009-02-12 19:32:43 -0800226 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
227 struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
Joel Becker1e61ee72008-08-20 18:32:45 -0700228
Joel Becker6136ca52009-02-12 19:32:43 -0800229 BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
Joel Becker1e61ee72008-08-20 18:32:45 -0700230 mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
Joel Becker6136ca52009-02-12 19:32:43 -0800231 (oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
Joel Becker1e61ee72008-08-20 18:32:45 -0700232 "Device %s, asking for sparse allocation: inode %llu, "
233 "cpos %u, clusters %u\n",
234 osb->dev_str,
Joel Becker6136ca52009-02-12 19:32:43 -0800235 (unsigned long long)oi->ip_blkno,
236 rec->e_cpos, oi->ip_clusters);
Joel Becker1e61ee72008-08-20 18:32:45 -0700237
238 return 0;
239}
240
Joel Becker6136ca52009-02-12 19:32:43 -0800241static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800242{
Joel Becker10995aa2008-11-13 14:49:12 -0800243 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800244
Joel Beckerf99b9b72008-08-20 19:36:33 -0700245 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Joel Becker10995aa2008-11-13 14:49:12 -0800246 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
Tao Mae7d4cb62008-08-18 17:38:44 +0800247
Joel Becker10995aa2008-11-13 14:49:12 -0800248 return 0;
Tao Mae7d4cb62008-08-18 17:38:44 +0800249}
250
Joel Beckerf99b9b72008-08-20 19:36:33 -0700251static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
252{
253 struct ocfs2_dinode *di = et->et_object;
254
255 et->et_root_el = &di->id2.i_list;
256}
257
Tao Mae7d4cb62008-08-18 17:38:44 +0800258
Joel Becker0ce10102008-08-20 17:19:50 -0700259static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
260{
Joel Becker2a50a742008-12-09 14:24:33 -0800261 struct ocfs2_xattr_value_buf *vb = et->et_object;
Joel Becker0ce10102008-08-20 17:19:50 -0700262
Joel Becker2a50a742008-12-09 14:24:33 -0800263 et->et_root_el = &vb->vb_xv->xr_list;
Joel Becker0ce10102008-08-20 17:19:50 -0700264}
265
Tao Maf56654c2008-08-18 17:38:48 +0800266static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
267 u64 blkno)
268{
Joel Becker2a50a742008-12-09 14:24:33 -0800269 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800270
Joel Becker2a50a742008-12-09 14:24:33 -0800271 vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
Tao Maf56654c2008-08-18 17:38:48 +0800272}
273
274static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
275{
Joel Becker2a50a742008-12-09 14:24:33 -0800276 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800277
Joel Becker2a50a742008-12-09 14:24:33 -0800278 return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
Tao Maf56654c2008-08-18 17:38:48 +0800279}
280
Joel Becker6136ca52009-02-12 19:32:43 -0800281static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
Tao Maf56654c2008-08-18 17:38:48 +0800282 u32 clusters)
283{
Joel Becker2a50a742008-12-09 14:24:33 -0800284 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800285
Joel Becker2a50a742008-12-09 14:24:33 -0800286 le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
Tao Maf56654c2008-08-18 17:38:48 +0800287}
288
Joel Becker1a09f552008-08-20 17:44:24 -0700289static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700290 .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk,
291 .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk,
292 .eo_update_clusters = ocfs2_xattr_value_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700293 .eo_fill_root_el = ocfs2_xattr_value_fill_root_el,
Tao Maf56654c2008-08-18 17:38:48 +0800294};
295
Joel Becker0ce10102008-08-20 17:19:50 -0700296static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
297{
298 struct ocfs2_xattr_block *xb = et->et_object;
299
300 et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
301}
302
Joel Becker6136ca52009-02-12 19:32:43 -0800303static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
Joel Becker943cced2008-08-20 17:31:10 -0700304{
Joel Becker6136ca52009-02-12 19:32:43 -0800305 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Joel Becker943cced2008-08-20 17:31:10 -0700306 et->et_max_leaf_clusters =
Joel Becker6136ca52009-02-12 19:32:43 -0800307 ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
Joel Becker943cced2008-08-20 17:31:10 -0700308}
309
Tao Maba492612008-08-18 17:38:49 +0800310static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
311 u64 blkno)
312{
Joel Beckerea5efa12008-08-20 16:57:27 -0700313 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800314 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
315
316 xt->xt_last_eb_blk = cpu_to_le64(blkno);
317}
318
319static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
320{
Joel Beckerea5efa12008-08-20 16:57:27 -0700321 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800322 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
323
324 return le64_to_cpu(xt->xt_last_eb_blk);
325}
326
Joel Becker6136ca52009-02-12 19:32:43 -0800327static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
Tao Maba492612008-08-18 17:38:49 +0800328 u32 clusters)
329{
Joel Beckerea5efa12008-08-20 16:57:27 -0700330 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800331
332 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
333}
334
Tao Maba492612008-08-18 17:38:49 +0800335static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700336 .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk,
337 .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk,
338 .eo_update_clusters = ocfs2_xattr_tree_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700339 .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el,
Joel Becker943cced2008-08-20 17:31:10 -0700340 .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
Tao Maba492612008-08-18 17:38:49 +0800341};
342
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800343static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
344 u64 blkno)
345{
346 struct ocfs2_dx_root_block *dx_root = et->et_object;
347
348 dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
349}
350
351static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
352{
353 struct ocfs2_dx_root_block *dx_root = et->et_object;
354
355 return le64_to_cpu(dx_root->dr_last_eb_blk);
356}
357
Joel Becker6136ca52009-02-12 19:32:43 -0800358static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800359 u32 clusters)
360{
361 struct ocfs2_dx_root_block *dx_root = et->et_object;
362
363 le32_add_cpu(&dx_root->dr_clusters, clusters);
364}
365
Joel Becker6136ca52009-02-12 19:32:43 -0800366static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800367{
368 struct ocfs2_dx_root_block *dx_root = et->et_object;
369
370 BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));
371
372 return 0;
373}
374
375static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
376{
377 struct ocfs2_dx_root_block *dx_root = et->et_object;
378
379 et->et_root_el = &dx_root->dr_list;
380}
381
382static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
383 .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk,
384 .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk,
385 .eo_update_clusters = ocfs2_dx_root_update_clusters,
386 .eo_sanity_check = ocfs2_dx_root_sanity_check,
387 .eo_fill_root_el = ocfs2_dx_root_fill_root_el,
388};
389
Tao Mafe924412009-08-18 11:22:25 +0800390static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et)
391{
392 struct ocfs2_refcount_block *rb = et->et_object;
393
394 et->et_root_el = &rb->rf_list;
395}
396
397static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
398 u64 blkno)
399{
400 struct ocfs2_refcount_block *rb = et->et_object;
401
402 rb->rf_last_eb_blk = cpu_to_le64(blkno);
403}
404
405static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
406{
407 struct ocfs2_refcount_block *rb = et->et_object;
408
409 return le64_to_cpu(rb->rf_last_eb_blk);
410}
411
412static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et,
413 u32 clusters)
414{
415 struct ocfs2_refcount_block *rb = et->et_object;
416
417 le32_add_cpu(&rb->rf_clusters, clusters);
418}
419
420static enum ocfs2_contig_type
421ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et,
422 struct ocfs2_extent_rec *ext,
423 struct ocfs2_extent_rec *insert_rec)
424{
425 return CONTIG_NONE;
426}
427
428static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
429 .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk,
430 .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk,
431 .eo_update_clusters = ocfs2_refcount_tree_update_clusters,
432 .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el,
433 .eo_extent_contig = ocfs2_refcount_tree_extent_contig,
434};
435
Joel Becker8d6220d2008-08-22 12:46:09 -0700436static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800437 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700438 struct buffer_head *bh,
Joel Becker13723d02008-10-17 19:25:01 -0700439 ocfs2_journal_access_func access,
Joel Becker8d6220d2008-08-22 12:46:09 -0700440 void *obj,
441 struct ocfs2_extent_tree_operations *ops)
Tao Mae7d4cb62008-08-18 17:38:44 +0800442{
Joel Becker1a09f552008-08-20 17:44:24 -0700443 et->et_ops = ops;
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700444 et->et_root_bh = bh;
Joel Becker5e404e92009-02-13 03:54:22 -0800445 et->et_ci = ci;
Joel Becker13723d02008-10-17 19:25:01 -0700446 et->et_root_journal_access = access;
Joel Beckerea5efa12008-08-20 16:57:27 -0700447 if (!obj)
448 obj = (void *)bh->b_data;
449 et->et_object = obj;
Tao Mae7d4cb62008-08-18 17:38:44 +0800450
Joel Becker0ce10102008-08-20 17:19:50 -0700451 et->et_ops->eo_fill_root_el(et);
Joel Becker943cced2008-08-20 17:31:10 -0700452 if (!et->et_ops->eo_fill_max_leaf_clusters)
453 et->et_max_leaf_clusters = 0;
454 else
Joel Becker6136ca52009-02-12 19:32:43 -0800455 et->et_ops->eo_fill_max_leaf_clusters(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800456}
457
Joel Becker8d6220d2008-08-22 12:46:09 -0700458void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800459 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700460 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700461{
Joel Becker5e404e92009-02-13 03:54:22 -0800462 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di,
Joel Becker13723d02008-10-17 19:25:01 -0700463 NULL, &ocfs2_dinode_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700464}
465
Joel Becker8d6220d2008-08-22 12:46:09 -0700466void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800467 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700468 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700469{
Joel Becker5e404e92009-02-13 03:54:22 -0800470 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb,
Joel Becker13723d02008-10-17 19:25:01 -0700471 NULL, &ocfs2_xattr_tree_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700472}
473
Joel Becker8d6220d2008-08-22 12:46:09 -0700474void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800475 struct ocfs2_caching_info *ci,
Joel Becker2a50a742008-12-09 14:24:33 -0800476 struct ocfs2_xattr_value_buf *vb)
Tao Mae7d4cb62008-08-18 17:38:44 +0800477{
Joel Becker5e404e92009-02-13 03:54:22 -0800478 __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb,
Joel Becker8d6220d2008-08-22 12:46:09 -0700479 &ocfs2_xattr_value_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800480}
481
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800482void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800483 struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800484 struct buffer_head *bh)
485{
Joel Becker5e404e92009-02-13 03:54:22 -0800486 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800487 NULL, &ocfs2_dx_root_et_ops);
488}
489
Tao Mafe924412009-08-18 11:22:25 +0800490void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
491 struct ocfs2_caching_info *ci,
492 struct buffer_head *bh)
493{
494 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb,
495 NULL, &ocfs2_refcount_tree_et_ops);
496}
497
Joel Becker35dc0aa2008-08-20 16:25:06 -0700498static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
499 u64 new_last_eb_blk)
Tao Mae7d4cb62008-08-18 17:38:44 +0800500{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700501 et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
Tao Mae7d4cb62008-08-18 17:38:44 +0800502}
503
Joel Becker35dc0aa2008-08-20 16:25:06 -0700504static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800505{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700506 return et->et_ops->eo_get_last_eb_blk(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800507}
508
Joel Becker6136ca52009-02-12 19:32:43 -0800509static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -0700510 u32 clusters)
Tao Mae7d4cb62008-08-18 17:38:44 +0800511{
Joel Becker6136ca52009-02-12 19:32:43 -0800512 et->et_ops->eo_update_clusters(et, clusters);
Joel Becker35dc0aa2008-08-20 16:25:06 -0700513}
514
Joel Becker92ba4702009-02-13 03:18:34 -0800515static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
516 struct ocfs2_extent_rec *rec)
517{
518 if (et->et_ops->eo_extent_map_insert)
519 et->et_ops->eo_extent_map_insert(et, rec);
520}
521
Joel Becker4c911ee2009-02-13 02:50:12 -0800522static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
523 u32 clusters)
524{
525 if (et->et_ops->eo_extent_map_truncate)
526 et->et_ops->eo_extent_map_truncate(et, clusters);
527}
528
Joel Becker13723d02008-10-17 19:25:01 -0700529static inline int ocfs2_et_root_journal_access(handle_t *handle,
Joel Becker13723d02008-10-17 19:25:01 -0700530 struct ocfs2_extent_tree *et,
531 int type)
532{
Joel Beckerd9a0a1f2009-02-12 17:32:34 -0800533 return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700534 type);
535}
536
Tao Ma853a3a12009-08-18 11:22:18 +0800537static inline enum ocfs2_contig_type
538 ocfs2_et_extent_contig(struct ocfs2_extent_tree *et,
539 struct ocfs2_extent_rec *rec,
540 struct ocfs2_extent_rec *insert_rec)
541{
542 if (et->et_ops->eo_extent_contig)
543 return et->et_ops->eo_extent_contig(et, rec, insert_rec);
544
545 return ocfs2_extent_rec_contig(
546 ocfs2_metadata_cache_get_super(et->et_ci),
547 rec, insert_rec);
548}
549
Joel Becker6136ca52009-02-12 19:32:43 -0800550static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700551 struct ocfs2_extent_rec *rec)
552{
553 int ret = 0;
554
555 if (et->et_ops->eo_insert_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800556 ret = et->et_ops->eo_insert_check(et, rec);
Joel Becker1e61ee72008-08-20 18:32:45 -0700557 return ret;
558}
559
Joel Becker6136ca52009-02-12 19:32:43 -0800560static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
Joel Becker35dc0aa2008-08-20 16:25:06 -0700561{
Joel Becker1e61ee72008-08-20 18:32:45 -0700562 int ret = 0;
563
564 if (et->et_ops->eo_sanity_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800565 ret = et->et_ops->eo_sanity_check(et);
Joel Becker1e61ee72008-08-20 18:32:45 -0700566 return ret;
Tao Mae7d4cb62008-08-18 17:38:44 +0800567}
568
Mark Fasheh59a5e412007-06-22 15:52:36 -0700569static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
570 struct ocfs2_extent_block *eb);
Joel Beckerd401dc12009-02-13 02:24:10 -0800571static void ocfs2_adjust_rightmost_records(handle_t *handle,
572 struct ocfs2_extent_tree *et,
Tao Ma6b791bc2009-06-12 14:18:36 +0800573 struct ocfs2_path *path,
574 struct ocfs2_extent_rec *insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -0800575/*
576 * Reset the actual path elements so that we can re-use the structure
577 * to build another path. Generally, this involves freeing the buffer
578 * heads.
579 */
Tao Mae2e9f602009-08-18 11:22:34 +0800580void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
Mark Fashehdcd05382007-01-16 11:32:23 -0800581{
582 int i, start = 0, depth = 0;
583 struct ocfs2_path_item *node;
584
585 if (keep_root)
586 start = 1;
587
588 for(i = start; i < path_num_items(path); i++) {
589 node = &path->p_node[i];
590
591 brelse(node->bh);
592 node->bh = NULL;
593 node->el = NULL;
594 }
595
596 /*
597 * Tree depth may change during truncate, or insert. If we're
598 * keeping the root extent list, then make sure that our path
599 * structure reflects the proper depth.
600 */
601 if (keep_root)
602 depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
Joel Becker13723d02008-10-17 19:25:01 -0700603 else
604 path_root_access(path) = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -0800605
606 path->p_tree_depth = depth;
607}
608
Tao Mae2e9f602009-08-18 11:22:34 +0800609void ocfs2_free_path(struct ocfs2_path *path)
Mark Fashehdcd05382007-01-16 11:32:23 -0800610{
611 if (path) {
612 ocfs2_reinit_path(path, 0);
613 kfree(path);
614 }
615}
616
617/*
Mark Fasheh328d5752007-06-18 10:48:04 -0700618 * All the elements of src into dest. After this call, src could be freed
619 * without affecting dest.
620 *
621 * Both paths should have the same root. Any non-root elements of dest
622 * will be freed.
623 */
624static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
625{
626 int i;
627
628 BUG_ON(path_root_bh(dest) != path_root_bh(src));
629 BUG_ON(path_root_el(dest) != path_root_el(src));
Joel Becker13723d02008-10-17 19:25:01 -0700630 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fasheh328d5752007-06-18 10:48:04 -0700631
632 ocfs2_reinit_path(dest, 1);
633
634 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
635 dest->p_node[i].bh = src->p_node[i].bh;
636 dest->p_node[i].el = src->p_node[i].el;
637
638 if (dest->p_node[i].bh)
639 get_bh(dest->p_node[i].bh);
640 }
641}
642
643/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800644 * Make the *dest path the same as src and re-initialize src path to
645 * have a root only.
646 */
647static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
648{
649 int i;
650
651 BUG_ON(path_root_bh(dest) != path_root_bh(src));
Joel Becker13723d02008-10-17 19:25:01 -0700652 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fashehdcd05382007-01-16 11:32:23 -0800653
654 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
655 brelse(dest->p_node[i].bh);
656
657 dest->p_node[i].bh = src->p_node[i].bh;
658 dest->p_node[i].el = src->p_node[i].el;
659
660 src->p_node[i].bh = NULL;
661 src->p_node[i].el = NULL;
662 }
663}
664
665/*
666 * Insert an extent block at given index.
667 *
668 * This will not take an additional reference on eb_bh.
669 */
670static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
671 struct buffer_head *eb_bh)
672{
673 struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
674
675 /*
676 * Right now, no root bh is an extent block, so this helps
677 * catch code errors with dinode trees. The assertion can be
678 * safely removed if we ever need to insert extent block
679 * structures at the root.
680 */
681 BUG_ON(index == 0);
682
683 path->p_node[index].bh = eb_bh;
684 path->p_node[index].el = &eb->h_list;
685}
686
687static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700688 struct ocfs2_extent_list *root_el,
689 ocfs2_journal_access_func access)
Mark Fashehdcd05382007-01-16 11:32:23 -0800690{
691 struct ocfs2_path *path;
692
693 BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
694
695 path = kzalloc(sizeof(*path), GFP_NOFS);
696 if (path) {
697 path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
698 get_bh(root_bh);
699 path_root_bh(path) = root_bh;
700 path_root_el(path) = root_el;
Joel Becker13723d02008-10-17 19:25:01 -0700701 path_root_access(path) = access;
Mark Fashehdcd05382007-01-16 11:32:23 -0800702 }
703
704 return path;
705}
706
Tao Mae2e9f602009-08-18 11:22:34 +0800707struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
Joel Beckerffdd7a52008-10-17 22:32:01 -0700708{
Joel Becker13723d02008-10-17 19:25:01 -0700709 return ocfs2_new_path(path_root_bh(path), path_root_el(path),
710 path_root_access(path));
Joel Beckerffdd7a52008-10-17 22:32:01 -0700711}
712
Tao Mae2e9f602009-08-18 11:22:34 +0800713struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
Joel Beckerffdd7a52008-10-17 22:32:01 -0700714{
Joel Becker13723d02008-10-17 19:25:01 -0700715 return ocfs2_new_path(et->et_root_bh, et->et_root_el,
716 et->et_root_journal_access);
717}
718
719/*
720 * Journal the buffer at depth idx. All idx>0 are extent_blocks,
721 * otherwise it's the root_access function.
722 *
723 * I don't like the way this function's name looks next to
724 * ocfs2_journal_access_path(), but I don't have a better one.
725 */
Tao Mae2e9f602009-08-18 11:22:34 +0800726int ocfs2_path_bh_journal_access(handle_t *handle,
727 struct ocfs2_caching_info *ci,
728 struct ocfs2_path *path,
729 int idx)
Joel Becker13723d02008-10-17 19:25:01 -0700730{
731 ocfs2_journal_access_func access = path_root_access(path);
732
733 if (!access)
734 access = ocfs2_journal_access;
735
736 if (idx)
737 access = ocfs2_journal_access_eb;
738
Joel Becker0cf2f762009-02-12 16:41:25 -0800739 return access(handle, ci, path->p_node[idx].bh,
Joel Becker13723d02008-10-17 19:25:01 -0700740 OCFS2_JOURNAL_ACCESS_WRITE);
Joel Beckerffdd7a52008-10-17 22:32:01 -0700741}
742
Mark Fashehdcd05382007-01-16 11:32:23 -0800743/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800744 * Convenience function to journal all components in a path.
745 */
Tao Mae2e9f602009-08-18 11:22:34 +0800746int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
747 handle_t *handle,
748 struct ocfs2_path *path)
Mark Fashehdcd05382007-01-16 11:32:23 -0800749{
750 int i, ret = 0;
751
752 if (!path)
753 goto out;
754
755 for(i = 0; i < path_num_items(path); i++) {
Joel Becker0cf2f762009-02-12 16:41:25 -0800756 ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -0800757 if (ret < 0) {
758 mlog_errno(ret);
759 goto out;
760 }
761 }
762
763out:
764 return ret;
765}
766
Mark Fasheh328d5752007-06-18 10:48:04 -0700767/*
768 * Return the index of the extent record which contains cluster #v_cluster.
769 * -1 is returned if it was not found.
770 *
771 * Should work fine on interior and exterior nodes.
772 */
773int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
774{
775 int ret = -1;
776 int i;
777 struct ocfs2_extent_rec *rec;
778 u32 rec_end, rec_start, clusters;
779
780 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
781 rec = &el->l_recs[i];
782
783 rec_start = le32_to_cpu(rec->e_cpos);
784 clusters = ocfs2_rec_clusters(el, rec);
785
786 rec_end = rec_start + clusters;
787
788 if (v_cluster >= rec_start && v_cluster < rec_end) {
789 ret = i;
790 break;
791 }
792 }
793
794 return ret;
795}
796
Mark Fashehe48edee2007-03-07 16:46:57 -0800797/*
798 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
Tao Ma853a3a12009-08-18 11:22:18 +0800799 * ocfs2_extent_rec_contig only work properly against leaf nodes!
Mark Fashehe48edee2007-03-07 16:46:57 -0800800 */
Mark Fashehdcd05382007-01-16 11:32:23 -0800801static int ocfs2_block_extent_contig(struct super_block *sb,
802 struct ocfs2_extent_rec *ext,
803 u64 blkno)
Mark Fashehccd979b2005-12-15 14:31:24 -0800804{
Mark Fashehe48edee2007-03-07 16:46:57 -0800805 u64 blk_end = le64_to_cpu(ext->e_blkno);
806
807 blk_end += ocfs2_clusters_to_blocks(sb,
808 le16_to_cpu(ext->e_leaf_clusters));
809
810 return blkno == blk_end;
Mark Fashehccd979b2005-12-15 14:31:24 -0800811}
812
Mark Fashehdcd05382007-01-16 11:32:23 -0800813static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
814 struct ocfs2_extent_rec *right)
815{
Mark Fashehe48edee2007-03-07 16:46:57 -0800816 u32 left_range;
817
818 left_range = le32_to_cpu(left->e_cpos) +
819 le16_to_cpu(left->e_leaf_clusters);
820
821 return (left_range == le32_to_cpu(right->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -0800822}
823
824static enum ocfs2_contig_type
Tao Ma853a3a12009-08-18 11:22:18 +0800825 ocfs2_extent_rec_contig(struct super_block *sb,
826 struct ocfs2_extent_rec *ext,
827 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -0800828{
829 u64 blkno = le64_to_cpu(insert_rec->e_blkno);
830
Mark Fasheh328d5752007-06-18 10:48:04 -0700831 /*
832 * Refuse to coalesce extent records with different flag
833 * fields - we don't want to mix unwritten extents with user
834 * data.
835 */
836 if (ext->e_flags != insert_rec->e_flags)
837 return CONTIG_NONE;
838
Mark Fashehdcd05382007-01-16 11:32:23 -0800839 if (ocfs2_extents_adjacent(ext, insert_rec) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800840 ocfs2_block_extent_contig(sb, ext, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800841 return CONTIG_RIGHT;
842
843 blkno = le64_to_cpu(ext->e_blkno);
844 if (ocfs2_extents_adjacent(insert_rec, ext) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800845 ocfs2_block_extent_contig(sb, insert_rec, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800846 return CONTIG_LEFT;
847
848 return CONTIG_NONE;
849}
850
851/*
852 * NOTE: We can have pretty much any combination of contiguousness and
853 * appending.
854 *
855 * The usefulness of APPEND_TAIL is more in that it lets us know that
856 * we'll have to update the path to that leaf.
857 */
858enum ocfs2_append_type {
859 APPEND_NONE = 0,
860 APPEND_TAIL,
861};
862
Mark Fasheh328d5752007-06-18 10:48:04 -0700863enum ocfs2_split_type {
864 SPLIT_NONE = 0,
865 SPLIT_LEFT,
866 SPLIT_RIGHT,
867};
868
Mark Fashehdcd05382007-01-16 11:32:23 -0800869struct ocfs2_insert_type {
Mark Fasheh328d5752007-06-18 10:48:04 -0700870 enum ocfs2_split_type ins_split;
Mark Fashehdcd05382007-01-16 11:32:23 -0800871 enum ocfs2_append_type ins_appending;
872 enum ocfs2_contig_type ins_contig;
873 int ins_contig_index;
Mark Fashehdcd05382007-01-16 11:32:23 -0800874 int ins_tree_depth;
875};
876
Mark Fasheh328d5752007-06-18 10:48:04 -0700877struct ocfs2_merge_ctxt {
878 enum ocfs2_contig_type c_contig_type;
879 int c_has_empty_extent;
880 int c_split_covers_rec;
Mark Fasheh328d5752007-06-18 10:48:04 -0700881};
882
Joel Becker5e965812008-11-13 14:49:16 -0800883static int ocfs2_validate_extent_block(struct super_block *sb,
884 struct buffer_head *bh)
885{
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700886 int rc;
Joel Becker5e965812008-11-13 14:49:16 -0800887 struct ocfs2_extent_block *eb =
888 (struct ocfs2_extent_block *)bh->b_data;
889
Tao Maa09d09b2011-02-22 08:21:10 +0800890 trace_ocfs2_validate_extent_block((unsigned long long)bh->b_blocknr);
Joel Becker970e4932008-11-13 14:49:19 -0800891
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700892 BUG_ON(!buffer_uptodate(bh));
893
894 /*
895 * If the ecc fails, we return the error but otherwise
896 * leave the filesystem running. We know any error is
897 * local to this block.
898 */
899 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
Joel Becker13723d02008-10-17 19:25:01 -0700900 if (rc) {
901 mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
902 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700903 return rc;
Joel Becker13723d02008-10-17 19:25:01 -0700904 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700905
906 /*
907 * Errors after here are fatal.
908 */
909
Joel Becker5e965812008-11-13 14:49:16 -0800910 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
911 ocfs2_error(sb,
912 "Extent block #%llu has bad signature %.*s",
913 (unsigned long long)bh->b_blocknr, 7,
914 eb->h_signature);
915 return -EINVAL;
916 }
917
918 if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
919 ocfs2_error(sb,
920 "Extent block #%llu has an invalid h_blkno "
921 "of %llu",
922 (unsigned long long)bh->b_blocknr,
923 (unsigned long long)le64_to_cpu(eb->h_blkno));
924 return -EINVAL;
925 }
926
927 if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
928 ocfs2_error(sb,
929 "Extent block #%llu has an invalid "
930 "h_fs_generation of #%u",
931 (unsigned long long)bh->b_blocknr,
932 le32_to_cpu(eb->h_fs_generation));
933 return -EINVAL;
934 }
935
936 return 0;
937}
938
Joel Becker3d03a302009-02-12 17:49:26 -0800939int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
Joel Becker5e965812008-11-13 14:49:16 -0800940 struct buffer_head **bh)
941{
942 int rc;
943 struct buffer_head *tmp = *bh;
944
Joel Becker3d03a302009-02-12 17:49:26 -0800945 rc = ocfs2_read_block(ci, eb_blkno, &tmp,
Joel Becker970e4932008-11-13 14:49:19 -0800946 ocfs2_validate_extent_block);
Joel Becker5e965812008-11-13 14:49:16 -0800947
948 /* If ocfs2_read_block() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -0800949 if (!rc && !*bh)
Joel Becker5e965812008-11-13 14:49:16 -0800950 *bh = tmp;
951
Joel Becker5e965812008-11-13 14:49:16 -0800952 return rc;
953}
954
955
Mark Fashehccd979b2005-12-15 14:31:24 -0800956/*
957 * How many free extents have we got before we need more meta data?
958 */
959int ocfs2_num_free_extents(struct ocfs2_super *osb,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700960 struct ocfs2_extent_tree *et)
Mark Fashehccd979b2005-12-15 14:31:24 -0800961{
962 int retval;
Tao Mae7d4cb62008-08-18 17:38:44 +0800963 struct ocfs2_extent_list *el = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800964 struct ocfs2_extent_block *eb;
965 struct buffer_head *eb_bh = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +0800966 u64 last_eb_blk = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800967
Joel Beckerf99b9b72008-08-20 19:36:33 -0700968 el = et->et_root_el;
969 last_eb_blk = ocfs2_et_get_last_eb_blk(et);
Mark Fashehccd979b2005-12-15 14:31:24 -0800970
Tao Mae7d4cb62008-08-18 17:38:44 +0800971 if (last_eb_blk) {
Joel Becker3d03a302009-02-12 17:49:26 -0800972 retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
973 &eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800974 if (retval < 0) {
975 mlog_errno(retval);
976 goto bail;
977 }
978 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
979 el = &eb->h_list;
Tao Mae7d4cb62008-08-18 17:38:44 +0800980 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800981
982 BUG_ON(el->l_tree_depth != 0);
983
984 retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
985bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700986 brelse(eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800987
Tao Maa09d09b2011-02-22 08:21:10 +0800988 trace_ocfs2_num_free_extents(retval);
Mark Fashehccd979b2005-12-15 14:31:24 -0800989 return retval;
990}
991
992/* expects array to already be allocated
993 *
994 * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
995 * l_count for you
996 */
Joel Becker42a5a7a2009-02-12 18:49:19 -0800997static int ocfs2_create_new_meta_bhs(handle_t *handle,
998 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -0800999 int wanted,
1000 struct ocfs2_alloc_context *meta_ac,
1001 struct buffer_head *bhs[])
1002{
1003 int count, status, i;
1004 u16 suballoc_bit_start;
1005 u32 num_got;
Joel Becker2b6cb572010-03-26 10:09:15 +08001006 u64 suballoc_loc, first_blkno;
Joel Becker42a5a7a2009-02-12 18:49:19 -08001007 struct ocfs2_super *osb =
1008 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001009 struct ocfs2_extent_block *eb;
1010
Mark Fashehccd979b2005-12-15 14:31:24 -08001011 count = 0;
1012 while (count < wanted) {
Joel Becker1ed9b772010-05-06 13:59:06 +08001013 status = ocfs2_claim_metadata(handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001014 meta_ac,
1015 wanted - count,
Joel Becker2b6cb572010-03-26 10:09:15 +08001016 &suballoc_loc,
Mark Fashehccd979b2005-12-15 14:31:24 -08001017 &suballoc_bit_start,
1018 &num_got,
1019 &first_blkno);
1020 if (status < 0) {
1021 mlog_errno(status);
1022 goto bail;
1023 }
1024
1025 for(i = count; i < (num_got + count); i++) {
1026 bhs[i] = sb_getblk(osb->sb, first_blkno);
1027 if (bhs[i] == NULL) {
1028 status = -EIO;
1029 mlog_errno(status);
1030 goto bail;
1031 }
Joel Becker42a5a7a2009-02-12 18:49:19 -08001032 ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001033
Joel Becker42a5a7a2009-02-12 18:49:19 -08001034 status = ocfs2_journal_access_eb(handle, et->et_ci,
1035 bhs[i],
Joel Becker13723d02008-10-17 19:25:01 -07001036 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001037 if (status < 0) {
1038 mlog_errno(status);
1039 goto bail;
1040 }
1041
1042 memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
1043 eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
1044 /* Ok, setup the minimal stuff here. */
1045 strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
1046 eb->h_blkno = cpu_to_le64(first_blkno);
1047 eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
Tiger Yangb89c5422010-01-25 14:11:06 +08001048 eb->h_suballoc_slot =
1049 cpu_to_le16(meta_ac->ac_alloc_slot);
Joel Becker2b6cb572010-03-26 10:09:15 +08001050 eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
Mark Fashehccd979b2005-12-15 14:31:24 -08001051 eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1052 eb->h_list.l_count =
1053 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
1054
1055 suballoc_bit_start++;
1056 first_blkno++;
1057
1058 /* We'll also be dirtied by the caller, so
1059 * this isn't absolutely necessary. */
Joel Beckerec20cec2010-03-19 14:13:52 -07001060 ocfs2_journal_dirty(handle, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001061 }
1062
1063 count += num_got;
1064 }
1065
1066 status = 0;
1067bail:
1068 if (status < 0) {
1069 for(i = 0; i < wanted; i++) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001070 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001071 bhs[i] = NULL;
1072 }
Tao Mac1e8d352011-03-07 16:43:21 +08001073 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001074 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001075 return status;
1076}
1077
1078/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001079 * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
1080 *
1081 * Returns the sum of the rightmost extent rec logical offset and
1082 * cluster count.
1083 *
1084 * ocfs2_add_branch() uses this to determine what logical cluster
1085 * value should be populated into the leftmost new branch records.
1086 *
1087 * ocfs2_shift_tree_depth() uses this to determine the # clusters
1088 * value for the new topmost tree record.
1089 */
1090static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el)
1091{
1092 int i;
1093
1094 i = le16_to_cpu(el->l_next_free_rec) - 1;
1095
1096 return le32_to_cpu(el->l_recs[i].e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001097 ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08001098}
1099
1100/*
Tao Ma6b791bc2009-06-12 14:18:36 +08001101 * Change range of the branches in the right most path according to the leaf
1102 * extent block's rightmost record.
1103 */
1104static int ocfs2_adjust_rightmost_branch(handle_t *handle,
Tao Ma6b791bc2009-06-12 14:18:36 +08001105 struct ocfs2_extent_tree *et)
1106{
1107 int status;
1108 struct ocfs2_path *path = NULL;
1109 struct ocfs2_extent_list *el;
1110 struct ocfs2_extent_rec *rec;
1111
1112 path = ocfs2_new_path_from_et(et);
1113 if (!path) {
1114 status = -ENOMEM;
1115 return status;
1116 }
1117
Joel Beckerfacdb772009-02-12 18:08:48 -08001118 status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
Tao Ma6b791bc2009-06-12 14:18:36 +08001119 if (status < 0) {
1120 mlog_errno(status);
1121 goto out;
1122 }
1123
Tao Mac901fb02010-04-26 14:34:57 +08001124 status = ocfs2_extend_trans(handle, path_num_items(path));
Tao Ma6b791bc2009-06-12 14:18:36 +08001125 if (status < 0) {
1126 mlog_errno(status);
1127 goto out;
1128 }
1129
Joel Beckerd401dc12009-02-13 02:24:10 -08001130 status = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma6b791bc2009-06-12 14:18:36 +08001131 if (status < 0) {
1132 mlog_errno(status);
1133 goto out;
1134 }
1135
1136 el = path_leaf_el(path);
1137 rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1];
1138
Joel Beckerd401dc12009-02-13 02:24:10 -08001139 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Tao Ma6b791bc2009-06-12 14:18:36 +08001140
1141out:
1142 ocfs2_free_path(path);
1143 return status;
1144}
1145
1146/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001147 * Add an entire tree branch to our inode. eb_bh is the extent block
Joel Beckerd401dc12009-02-13 02:24:10 -08001148 * to start at, if we don't want to start the branch at the root
Mark Fashehccd979b2005-12-15 14:31:24 -08001149 * structure.
1150 *
1151 * last_eb_bh is required as we have to update it's next_leaf pointer
1152 * for the new last extent block.
1153 *
1154 * the new branch will be 'empty' in the sense that every block will
Mark Fashehe48edee2007-03-07 16:46:57 -08001155 * contain a single record with cluster count == 0.
Mark Fashehccd979b2005-12-15 14:31:24 -08001156 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001157static int ocfs2_add_branch(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001158 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001159 struct buffer_head *eb_bh,
Mark Fasheh328d5752007-06-18 10:48:04 -07001160 struct buffer_head **last_eb_bh,
Mark Fashehccd979b2005-12-15 14:31:24 -08001161 struct ocfs2_alloc_context *meta_ac)
1162{
1163 int status, new_blocks, i;
1164 u64 next_blkno, new_last_eb_blk;
1165 struct buffer_head *bh;
1166 struct buffer_head **new_eb_bhs = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001167 struct ocfs2_extent_block *eb;
1168 struct ocfs2_extent_list *eb_el;
1169 struct ocfs2_extent_list *el;
Tao Ma6b791bc2009-06-12 14:18:36 +08001170 u32 new_cpos, root_end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001171
Mark Fasheh328d5752007-06-18 10:48:04 -07001172 BUG_ON(!last_eb_bh || !*last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001173
Mark Fashehccd979b2005-12-15 14:31:24 -08001174 if (eb_bh) {
1175 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1176 el = &eb->h_list;
1177 } else
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001178 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001179
1180 /* we never add a branch to a leaf. */
1181 BUG_ON(!el->l_tree_depth);
1182
1183 new_blocks = le16_to_cpu(el->l_tree_depth);
1184
Tao Ma6b791bc2009-06-12 14:18:36 +08001185 eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
1186 new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
1187 root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
1188
1189 /*
1190 * If there is a gap before the root end and the real end
1191 * of the righmost leaf block, we need to remove the gap
1192 * between new_cpos and root_end first so that the tree
1193 * is consistent after we add a new branch(it will start
1194 * from new_cpos).
1195 */
1196 if (root_end > new_cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08001197 trace_ocfs2_adjust_rightmost_branch(
1198 (unsigned long long)
1199 ocfs2_metadata_cache_owner(et->et_ci),
1200 root_end, new_cpos);
1201
Joel Beckerd401dc12009-02-13 02:24:10 -08001202 status = ocfs2_adjust_rightmost_branch(handle, et);
Tao Ma6b791bc2009-06-12 14:18:36 +08001203 if (status) {
1204 mlog_errno(status);
1205 goto bail;
1206 }
1207 }
1208
Mark Fashehccd979b2005-12-15 14:31:24 -08001209 /* allocate the number of new eb blocks we need */
1210 new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
1211 GFP_KERNEL);
1212 if (!new_eb_bhs) {
1213 status = -ENOMEM;
1214 mlog_errno(status);
1215 goto bail;
1216 }
1217
Joel Becker42a5a7a2009-02-12 18:49:19 -08001218 status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
Mark Fashehccd979b2005-12-15 14:31:24 -08001219 meta_ac, new_eb_bhs);
1220 if (status < 0) {
1221 mlog_errno(status);
1222 goto bail;
1223 }
1224
1225 /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
1226 * linked with the rest of the tree.
1227 * conversly, new_eb_bhs[0] is the new bottommost leaf.
1228 *
1229 * when we leave the loop, new_last_eb_blk will point to the
1230 * newest leaf, and next_blkno will point to the topmost extent
1231 * block. */
1232 next_blkno = new_last_eb_blk = 0;
1233 for(i = 0; i < new_blocks; i++) {
1234 bh = new_eb_bhs[i];
1235 eb = (struct ocfs2_extent_block *) bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001236 /* ocfs2_create_new_meta_bhs() should create it right! */
1237 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001238 eb_el = &eb->h_list;
1239
Joel Beckerd401dc12009-02-13 02:24:10 -08001240 status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
Joel Becker13723d02008-10-17 19:25:01 -07001241 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001242 if (status < 0) {
1243 mlog_errno(status);
1244 goto bail;
1245 }
1246
1247 eb->h_next_leaf_blk = 0;
1248 eb_el->l_tree_depth = cpu_to_le16(i);
1249 eb_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehdcd05382007-01-16 11:32:23 -08001250 /*
1251 * This actually counts as an empty extent as
1252 * c_clusters == 0
1253 */
1254 eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001255 eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehe48edee2007-03-07 16:46:57 -08001256 /*
1257 * eb_el isn't always an interior node, but even leaf
1258 * nodes want a zero'd flags and reserved field so
1259 * this gets the whole 32 bits regardless of use.
1260 */
1261 eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001262 if (!eb_el->l_tree_depth)
1263 new_last_eb_blk = le64_to_cpu(eb->h_blkno);
1264
Joel Beckerec20cec2010-03-19 14:13:52 -07001265 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001266 next_blkno = le64_to_cpu(eb->h_blkno);
1267 }
1268
1269 /* This is a bit hairy. We want to update up to three blocks
1270 * here without leaving any of them in an inconsistent state
1271 * in case of error. We don't have to worry about
1272 * journal_dirty erroring as it won't unless we've aborted the
1273 * handle (in which case we would never be here) so reserving
1274 * the write with journal_access is all we need to do. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001275 status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001276 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001277 if (status < 0) {
1278 mlog_errno(status);
1279 goto bail;
1280 }
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001281 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001282 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001283 if (status < 0) {
1284 mlog_errno(status);
1285 goto bail;
1286 }
1287 if (eb_bh) {
Joel Beckerd401dc12009-02-13 02:24:10 -08001288 status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001289 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001290 if (status < 0) {
1291 mlog_errno(status);
1292 goto bail;
1293 }
1294 }
1295
1296 /* Link the new branch into the rest of the tree (el will
Tao Mae7d4cb62008-08-18 17:38:44 +08001297 * either be on the root_bh, or the extent block passed in. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001298 i = le16_to_cpu(el->l_next_free_rec);
1299 el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08001300 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001301 el->l_recs[i].e_int_clusters = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 le16_add_cpu(&el->l_next_free_rec, 1);
1303
1304 /* fe needs a new last extent block pointer, as does the
1305 * next_leaf on the previously last-extent-block. */
Joel Becker35dc0aa2008-08-20 16:25:06 -07001306 ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
Mark Fashehccd979b2005-12-15 14:31:24 -08001307
Mark Fasheh328d5752007-06-18 10:48:04 -07001308 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001309 eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
1310
Joel Beckerec20cec2010-03-19 14:13:52 -07001311 ocfs2_journal_dirty(handle, *last_eb_bh);
1312 ocfs2_journal_dirty(handle, et->et_root_bh);
1313 if (eb_bh)
1314 ocfs2_journal_dirty(handle, eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001315
Mark Fasheh328d5752007-06-18 10:48:04 -07001316 /*
1317 * Some callers want to track the rightmost leaf so pass it
1318 * back here.
1319 */
1320 brelse(*last_eb_bh);
1321 get_bh(new_eb_bhs[0]);
1322 *last_eb_bh = new_eb_bhs[0];
1323
Mark Fashehccd979b2005-12-15 14:31:24 -08001324 status = 0;
1325bail:
1326 if (new_eb_bhs) {
1327 for (i = 0; i < new_blocks; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001328 brelse(new_eb_bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001329 kfree(new_eb_bhs);
1330 }
1331
Mark Fashehccd979b2005-12-15 14:31:24 -08001332 return status;
1333}
1334
1335/*
1336 * adds another level to the allocation tree.
1337 * returns back the new extent block so you can add a branch to it
1338 * after this call.
1339 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001340static int ocfs2_shift_tree_depth(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001341 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001342 struct ocfs2_alloc_context *meta_ac,
1343 struct buffer_head **ret_new_eb_bh)
1344{
1345 int status, i;
Mark Fashehdcd05382007-01-16 11:32:23 -08001346 u32 new_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001347 struct buffer_head *new_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001348 struct ocfs2_extent_block *eb;
Tao Mae7d4cb62008-08-18 17:38:44 +08001349 struct ocfs2_extent_list *root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001350 struct ocfs2_extent_list *eb_el;
1351
Joel Becker42a5a7a2009-02-12 18:49:19 -08001352 status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
Mark Fashehccd979b2005-12-15 14:31:24 -08001353 &new_eb_bh);
1354 if (status < 0) {
1355 mlog_errno(status);
1356 goto bail;
1357 }
1358
1359 eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001360 /* ocfs2_create_new_meta_bhs() should create it right! */
1361 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001362
1363 eb_el = &eb->h_list;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001364 root_el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001365
Joel Beckerd401dc12009-02-13 02:24:10 -08001366 status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001367 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001368 if (status < 0) {
1369 mlog_errno(status);
1370 goto bail;
1371 }
1372
Tao Mae7d4cb62008-08-18 17:38:44 +08001373 /* copy the root extent list data into the new extent block */
1374 eb_el->l_tree_depth = root_el->l_tree_depth;
1375 eb_el->l_next_free_rec = root_el->l_next_free_rec;
1376 for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1377 eb_el->l_recs[i] = root_el->l_recs[i];
Mark Fashehccd979b2005-12-15 14:31:24 -08001378
Joel Beckerec20cec2010-03-19 14:13:52 -07001379 ocfs2_journal_dirty(handle, new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001380
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001381 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001382 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001383 if (status < 0) {
1384 mlog_errno(status);
1385 goto bail;
1386 }
1387
Mark Fashehdcd05382007-01-16 11:32:23 -08001388 new_clusters = ocfs2_sum_rightmost_rec(eb_el);
1389
Tao Mae7d4cb62008-08-18 17:38:44 +08001390 /* update root_bh now */
1391 le16_add_cpu(&root_el->l_tree_depth, 1);
1392 root_el->l_recs[0].e_cpos = 0;
1393 root_el->l_recs[0].e_blkno = eb->h_blkno;
1394 root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
1395 for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1396 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
1397 root_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001398
1399 /* If this is our 1st tree depth shift, then last_eb_blk
1400 * becomes the allocated extent block */
Tao Mae7d4cb62008-08-18 17:38:44 +08001401 if (root_el->l_tree_depth == cpu_to_le16(1))
Joel Becker35dc0aa2008-08-20 16:25:06 -07001402 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001403
Joel Beckerec20cec2010-03-19 14:13:52 -07001404 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001405
1406 *ret_new_eb_bh = new_eb_bh;
1407 new_eb_bh = NULL;
1408 status = 0;
1409bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001410 brelse(new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001411
Mark Fashehccd979b2005-12-15 14:31:24 -08001412 return status;
1413}
1414
1415/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001416 * Should only be called when there is no space left in any of the
1417 * leaf nodes. What we want to do is find the lowest tree depth
1418 * non-leaf extent block with room for new records. There are three
1419 * valid results of this search:
1420 *
1421 * 1) a lowest extent block is found, then we pass it back in
1422 * *lowest_eb_bh and return '0'
1423 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001424 * 2) the search fails to find anything, but the root_el has room. We
Mark Fashehccd979b2005-12-15 14:31:24 -08001425 * pass NULL back in *lowest_eb_bh, but still return '0'
1426 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001427 * 3) the search fails to find anything AND the root_el is full, in
Mark Fashehccd979b2005-12-15 14:31:24 -08001428 * which case we return > 0
1429 *
1430 * return status < 0 indicates an error.
1431 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001432static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001433 struct buffer_head **target_bh)
1434{
1435 int status = 0, i;
1436 u64 blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08001437 struct ocfs2_extent_block *eb;
1438 struct ocfs2_extent_list *el;
1439 struct buffer_head *bh = NULL;
1440 struct buffer_head *lowest_bh = NULL;
1441
Mark Fashehccd979b2005-12-15 14:31:24 -08001442 *target_bh = NULL;
1443
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001444 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001445
1446 while(le16_to_cpu(el->l_tree_depth) > 1) {
1447 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08001448 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1449 "Owner %llu has empty "
Mark Fashehccd979b2005-12-15 14:31:24 -08001450 "extent list (next_free_rec == 0)",
Joel Becker3d03a302009-02-12 17:49:26 -08001451 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001452 status = -EIO;
1453 goto bail;
1454 }
1455 i = le16_to_cpu(el->l_next_free_rec) - 1;
1456 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1457 if (!blkno) {
Joel Becker3d03a302009-02-12 17:49:26 -08001458 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1459 "Owner %llu has extent "
Mark Fashehccd979b2005-12-15 14:31:24 -08001460 "list where extent # %d has no physical "
1461 "block start",
Joel Becker3d03a302009-02-12 17:49:26 -08001462 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
Mark Fashehccd979b2005-12-15 14:31:24 -08001463 status = -EIO;
1464 goto bail;
1465 }
1466
Mark Fasheha81cb882008-10-07 14:25:16 -07001467 brelse(bh);
1468 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001469
Joel Becker3d03a302009-02-12 17:49:26 -08001470 status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001471 if (status < 0) {
1472 mlog_errno(status);
1473 goto bail;
1474 }
1475
1476 eb = (struct ocfs2_extent_block *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001477 el = &eb->h_list;
1478
1479 if (le16_to_cpu(el->l_next_free_rec) <
1480 le16_to_cpu(el->l_count)) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001481 brelse(lowest_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001482 lowest_bh = bh;
1483 get_bh(lowest_bh);
1484 }
1485 }
1486
1487 /* If we didn't find one and the fe doesn't have any room,
1488 * then return '1' */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001489 el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001490 if (!lowest_bh && (el->l_next_free_rec == el->l_count))
Mark Fashehccd979b2005-12-15 14:31:24 -08001491 status = 1;
1492
1493 *target_bh = lowest_bh;
1494bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001495 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001496
Mark Fashehccd979b2005-12-15 14:31:24 -08001497 return status;
1498}
1499
Mark Fashehe48edee2007-03-07 16:46:57 -08001500/*
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001501 * Grow a b-tree so that it has more records.
1502 *
1503 * We might shift the tree depth in which case existing paths should
1504 * be considered invalid.
1505 *
1506 * Tree depth after the grow is returned via *final_depth.
Mark Fasheh328d5752007-06-18 10:48:04 -07001507 *
1508 * *last_eb_bh will be updated by ocfs2_add_branch().
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001509 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001510static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
1511 int *final_depth, struct buffer_head **last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001512 struct ocfs2_alloc_context *meta_ac)
1513{
1514 int ret, shift;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001515 struct ocfs2_extent_list *el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001516 int depth = le16_to_cpu(el->l_tree_depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001517 struct buffer_head *bh = NULL;
1518
1519 BUG_ON(meta_ac == NULL);
1520
Joel Beckerd401dc12009-02-13 02:24:10 -08001521 shift = ocfs2_find_branch_target(et, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001522 if (shift < 0) {
1523 ret = shift;
1524 mlog_errno(ret);
1525 goto out;
1526 }
1527
1528 /* We traveled all the way to the bottom of the allocation tree
1529 * and didn't find room for any more extents - we need to add
1530 * another tree level */
1531 if (shift) {
1532 BUG_ON(bh);
Tao Maa09d09b2011-02-22 08:21:10 +08001533 trace_ocfs2_grow_tree(
1534 (unsigned long long)
1535 ocfs2_metadata_cache_owner(et->et_ci),
1536 depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001537
1538 /* ocfs2_shift_tree_depth will return us a buffer with
1539 * the new extent block (so we can pass that to
1540 * ocfs2_add_branch). */
Joel Beckerd401dc12009-02-13 02:24:10 -08001541 ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001542 if (ret < 0) {
1543 mlog_errno(ret);
1544 goto out;
1545 }
1546 depth++;
Mark Fasheh328d5752007-06-18 10:48:04 -07001547 if (depth == 1) {
1548 /*
1549 * Special case: we have room now if we shifted from
1550 * tree_depth 0, so no more work needs to be done.
1551 *
1552 * We won't be calling add_branch, so pass
1553 * back *last_eb_bh as the new leaf. At depth
1554 * zero, it should always be null so there's
1555 * no reason to brelse.
1556 */
1557 BUG_ON(*last_eb_bh);
1558 get_bh(bh);
1559 *last_eb_bh = bh;
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001560 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07001561 }
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001562 }
1563
1564 /* call ocfs2_add_branch to add the final part of the tree with
1565 * the new data. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001566 ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001567 meta_ac);
1568 if (ret < 0) {
1569 mlog_errno(ret);
1570 goto out;
1571 }
1572
1573out:
1574 if (final_depth)
1575 *final_depth = depth;
1576 brelse(bh);
1577 return ret;
1578}
1579
1580/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001581 * This function will discard the rightmost extent record.
1582 */
1583static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
1584{
1585 int next_free = le16_to_cpu(el->l_next_free_rec);
1586 int count = le16_to_cpu(el->l_count);
1587 unsigned int num_bytes;
1588
1589 BUG_ON(!next_free);
1590 /* This will cause us to go off the end of our extent list. */
1591 BUG_ON(next_free >= count);
1592
1593 num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
1594
1595 memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
1596}
1597
1598static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
1599 struct ocfs2_extent_rec *insert_rec)
1600{
1601 int i, insert_index, next_free, has_empty, num_bytes;
1602 u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
1603 struct ocfs2_extent_rec *rec;
1604
1605 next_free = le16_to_cpu(el->l_next_free_rec);
1606 has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
1607
1608 BUG_ON(!next_free);
1609
1610 /* The tree code before us didn't allow enough room in the leaf. */
Julia Lawallb1f35502008-03-04 15:21:05 -08001611 BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
Mark Fashehdcd05382007-01-16 11:32:23 -08001612
1613 /*
1614 * The easiest way to approach this is to just remove the
1615 * empty extent and temporarily decrement next_free.
1616 */
1617 if (has_empty) {
1618 /*
1619 * If next_free was 1 (only an empty extent), this
1620 * loop won't execute, which is fine. We still want
1621 * the decrement above to happen.
1622 */
1623 for(i = 0; i < (next_free - 1); i++)
1624 el->l_recs[i] = el->l_recs[i+1];
1625
1626 next_free--;
1627 }
1628
1629 /*
1630 * Figure out what the new record index should be.
1631 */
1632 for(i = 0; i < next_free; i++) {
1633 rec = &el->l_recs[i];
1634
1635 if (insert_cpos < le32_to_cpu(rec->e_cpos))
1636 break;
1637 }
1638 insert_index = i;
1639
Tao Maa09d09b2011-02-22 08:21:10 +08001640 trace_ocfs2_rotate_leaf(insert_cpos, insert_index,
1641 has_empty, next_free,
1642 le16_to_cpu(el->l_count));
Mark Fashehdcd05382007-01-16 11:32:23 -08001643
1644 BUG_ON(insert_index < 0);
1645 BUG_ON(insert_index >= le16_to_cpu(el->l_count));
1646 BUG_ON(insert_index > next_free);
1647
1648 /*
1649 * No need to memmove if we're just adding to the tail.
1650 */
1651 if (insert_index != next_free) {
1652 BUG_ON(next_free >= le16_to_cpu(el->l_count));
1653
1654 num_bytes = next_free - insert_index;
1655 num_bytes *= sizeof(struct ocfs2_extent_rec);
1656 memmove(&el->l_recs[insert_index + 1],
1657 &el->l_recs[insert_index],
1658 num_bytes);
1659 }
1660
1661 /*
1662 * Either we had an empty extent, and need to re-increment or
1663 * there was no empty extent on a non full rightmost leaf node,
1664 * in which case we still need to increment.
1665 */
1666 next_free++;
1667 el->l_next_free_rec = cpu_to_le16(next_free);
1668 /*
1669 * Make sure none of the math above just messed up our tree.
1670 */
1671 BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
1672
1673 el->l_recs[insert_index] = *insert_rec;
1674
1675}
1676
Mark Fasheh328d5752007-06-18 10:48:04 -07001677static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
1678{
1679 int size, num_recs = le16_to_cpu(el->l_next_free_rec);
1680
1681 BUG_ON(num_recs == 0);
1682
1683 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
1684 num_recs--;
1685 size = num_recs * sizeof(struct ocfs2_extent_rec);
1686 memmove(&el->l_recs[0], &el->l_recs[1], size);
1687 memset(&el->l_recs[num_recs], 0,
1688 sizeof(struct ocfs2_extent_rec));
1689 el->l_next_free_rec = cpu_to_le16(num_recs);
1690 }
1691}
1692
Mark Fashehdcd05382007-01-16 11:32:23 -08001693/*
1694 * Create an empty extent record .
1695 *
1696 * l_next_free_rec may be updated.
1697 *
1698 * If an empty extent already exists do nothing.
1699 */
1700static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
1701{
1702 int next_free = le16_to_cpu(el->l_next_free_rec);
1703
Mark Fashehe48edee2007-03-07 16:46:57 -08001704 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
1705
Mark Fashehdcd05382007-01-16 11:32:23 -08001706 if (next_free == 0)
1707 goto set_and_inc;
1708
1709 if (ocfs2_is_empty_extent(&el->l_recs[0]))
1710 return;
1711
1712 mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
1713 "Asked to create an empty extent in a full list:\n"
1714 "count = %u, tree depth = %u",
1715 le16_to_cpu(el->l_count),
1716 le16_to_cpu(el->l_tree_depth));
1717
1718 ocfs2_shift_records_right(el);
1719
1720set_and_inc:
1721 le16_add_cpu(&el->l_next_free_rec, 1);
1722 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
1723}
1724
1725/*
1726 * For a rotation which involves two leaf nodes, the "root node" is
1727 * the lowest level tree node which contains a path to both leafs. This
1728 * resulting set of information can be used to form a complete "subtree"
1729 *
1730 * This function is passed two full paths from the dinode down to a
1731 * pair of adjacent leaves. It's task is to figure out which path
1732 * index contains the subtree root - this can be the root index itself
1733 * in a worst-case rotation.
1734 *
1735 * The array index of the subtree root is passed back.
1736 */
Tao Ma38a04e42009-11-30 14:32:19 +08001737int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
1738 struct ocfs2_path *left,
1739 struct ocfs2_path *right)
Mark Fashehdcd05382007-01-16 11:32:23 -08001740{
1741 int i = 0;
1742
1743 /*
1744 * Check that the caller passed in two paths from the same tree.
1745 */
1746 BUG_ON(path_root_bh(left) != path_root_bh(right));
1747
1748 do {
1749 i++;
1750
1751 /*
1752 * The caller didn't pass two adjacent paths.
1753 */
1754 mlog_bug_on_msg(i > left->p_tree_depth,
Joel Becker7dc02802009-02-12 19:20:13 -08001755 "Owner %llu, left depth %u, right depth %u\n"
Mark Fashehdcd05382007-01-16 11:32:23 -08001756 "left leaf blk %llu, right leaf blk %llu\n",
Joel Becker7dc02802009-02-12 19:20:13 -08001757 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
1758 left->p_tree_depth, right->p_tree_depth,
Mark Fashehdcd05382007-01-16 11:32:23 -08001759 (unsigned long long)path_leaf_bh(left)->b_blocknr,
1760 (unsigned long long)path_leaf_bh(right)->b_blocknr);
1761 } while (left->p_node[i].bh->b_blocknr ==
1762 right->p_node[i].bh->b_blocknr);
1763
1764 return i - 1;
1765}
1766
1767typedef void (path_insert_t)(void *, struct buffer_head *);
1768
1769/*
1770 * Traverse a btree path in search of cpos, starting at root_el.
1771 *
1772 * This code can be called with a cpos larger than the tree, in which
1773 * case it will return the rightmost path.
1774 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001775static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
Mark Fashehdcd05382007-01-16 11:32:23 -08001776 struct ocfs2_extent_list *root_el, u32 cpos,
1777 path_insert_t *func, void *data)
1778{
1779 int i, ret = 0;
1780 u32 range;
1781 u64 blkno;
1782 struct buffer_head *bh = NULL;
1783 struct ocfs2_extent_block *eb;
1784 struct ocfs2_extent_list *el;
1785 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08001786
1787 el = root_el;
1788 while (el->l_tree_depth) {
1789 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001790 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1791 "Owner %llu has empty extent list at "
Mark Fashehdcd05382007-01-16 11:32:23 -08001792 "depth %u\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001793 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001794 le16_to_cpu(el->l_tree_depth));
1795 ret = -EROFS;
1796 goto out;
1797
1798 }
1799
1800 for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
1801 rec = &el->l_recs[i];
1802
1803 /*
1804 * In the case that cpos is off the allocation
1805 * tree, this should just wind up returning the
1806 * rightmost record.
1807 */
1808 range = le32_to_cpu(rec->e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001809 ocfs2_rec_clusters(el, rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08001810 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
1811 break;
1812 }
1813
1814 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1815 if (blkno == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001816 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1817 "Owner %llu has bad blkno in extent list "
Mark Fashehdcd05382007-01-16 11:32:23 -08001818 "at depth %u (index %d)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001819 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001820 le16_to_cpu(el->l_tree_depth), i);
1821 ret = -EROFS;
1822 goto out;
1823 }
1824
1825 brelse(bh);
1826 bh = NULL;
Joel Beckerfacdb772009-02-12 18:08:48 -08001827 ret = ocfs2_read_extent_block(ci, blkno, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001828 if (ret) {
1829 mlog_errno(ret);
1830 goto out;
1831 }
1832
1833 eb = (struct ocfs2_extent_block *) bh->b_data;
1834 el = &eb->h_list;
Mark Fashehdcd05382007-01-16 11:32:23 -08001835
1836 if (le16_to_cpu(el->l_next_free_rec) >
1837 le16_to_cpu(el->l_count)) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001838 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1839 "Owner %llu has bad count in extent list "
Mark Fashehdcd05382007-01-16 11:32:23 -08001840 "at block %llu (next free=%u, count=%u)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001841 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001842 (unsigned long long)bh->b_blocknr,
1843 le16_to_cpu(el->l_next_free_rec),
1844 le16_to_cpu(el->l_count));
1845 ret = -EROFS;
1846 goto out;
1847 }
1848
1849 if (func)
1850 func(data, bh);
1851 }
1852
1853out:
1854 /*
1855 * Catch any trailing bh that the loop didn't handle.
1856 */
1857 brelse(bh);
1858
1859 return ret;
1860}
1861
1862/*
1863 * Given an initialized path (that is, it has a valid root extent
1864 * list), this function will traverse the btree in search of the path
1865 * which would contain cpos.
1866 *
1867 * The path traveled is recorded in the path structure.
1868 *
1869 * Note that this will not do any comparisons on leaf node extent
1870 * records, so it will work fine in the case that we just added a tree
1871 * branch.
1872 */
1873struct find_path_data {
1874 int index;
1875 struct ocfs2_path *path;
1876};
1877static void find_path_ins(void *data, struct buffer_head *bh)
1878{
1879 struct find_path_data *fp = data;
1880
1881 get_bh(bh);
1882 ocfs2_path_insert_eb(fp->path, fp->index, bh);
1883 fp->index++;
1884}
Tao Mae2e9f602009-08-18 11:22:34 +08001885int ocfs2_find_path(struct ocfs2_caching_info *ci,
1886 struct ocfs2_path *path, u32 cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08001887{
1888 struct find_path_data data;
1889
1890 data.index = 1;
1891 data.path = path;
Joel Beckerfacdb772009-02-12 18:08:48 -08001892 return __ocfs2_find_path(ci, path_root_el(path), cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08001893 find_path_ins, &data);
1894}
1895
1896static void find_leaf_ins(void *data, struct buffer_head *bh)
1897{
1898 struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
1899 struct ocfs2_extent_list *el = &eb->h_list;
1900 struct buffer_head **ret = data;
1901
1902 /* We want to retain only the leaf block. */
1903 if (le16_to_cpu(el->l_tree_depth) == 0) {
1904 get_bh(bh);
1905 *ret = bh;
1906 }
1907}
1908/*
1909 * Find the leaf block in the tree which would contain cpos. No
1910 * checking of the actual leaf is done.
1911 *
1912 * Some paths want to call this instead of allocating a path structure
1913 * and calling ocfs2_find_path().
1914 *
1915 * This function doesn't handle non btree extent lists.
1916 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001917int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
1918 struct ocfs2_extent_list *root_el, u32 cpos,
1919 struct buffer_head **leaf_bh)
Mark Fashehdcd05382007-01-16 11:32:23 -08001920{
1921 int ret;
1922 struct buffer_head *bh = NULL;
1923
Joel Beckerfacdb772009-02-12 18:08:48 -08001924 ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001925 if (ret) {
1926 mlog_errno(ret);
1927 goto out;
1928 }
1929
1930 *leaf_bh = bh;
1931out:
1932 return ret;
1933}
1934
1935/*
1936 * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
1937 *
1938 * Basically, we've moved stuff around at the bottom of the tree and
1939 * we need to fix up the extent records above the changes to reflect
1940 * the new changes.
1941 *
1942 * left_rec: the record on the left.
1943 * left_child_el: is the child list pointed to by left_rec
1944 * right_rec: the record to the right of left_rec
1945 * right_child_el: is the child list pointed to by right_rec
1946 *
1947 * By definition, this only works on interior nodes.
1948 */
1949static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
1950 struct ocfs2_extent_list *left_child_el,
1951 struct ocfs2_extent_rec *right_rec,
1952 struct ocfs2_extent_list *right_child_el)
1953{
1954 u32 left_clusters, right_end;
1955
1956 /*
1957 * Interior nodes never have holes. Their cpos is the cpos of
1958 * the leftmost record in their child list. Their cluster
1959 * count covers the full theoretical range of their child list
1960 * - the range between their cpos and the cpos of the record
1961 * immediately to their right.
1962 */
1963 left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
Tao Ma82e12642009-07-23 08:12:58 +08001964 if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
1965 BUG_ON(right_child_el->l_tree_depth);
Mark Fasheh328d5752007-06-18 10:48:04 -07001966 BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
1967 left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
1968 }
Mark Fashehdcd05382007-01-16 11:32:23 -08001969 left_clusters -= le32_to_cpu(left_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001970 left_rec->e_int_clusters = cpu_to_le32(left_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001971
1972 /*
1973 * Calculate the rightmost cluster count boundary before
Mark Fashehe48edee2007-03-07 16:46:57 -08001974 * moving cpos - we will need to adjust clusters after
Mark Fashehdcd05382007-01-16 11:32:23 -08001975 * updating e_cpos to keep the same highest cluster count.
1976 */
1977 right_end = le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001978 right_end += le32_to_cpu(right_rec->e_int_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001979
1980 right_rec->e_cpos = left_rec->e_cpos;
1981 le32_add_cpu(&right_rec->e_cpos, left_clusters);
1982
1983 right_end -= le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001984 right_rec->e_int_clusters = cpu_to_le32(right_end);
Mark Fashehdcd05382007-01-16 11:32:23 -08001985}
1986
1987/*
1988 * Adjust the adjacent root node records involved in a
1989 * rotation. left_el_blkno is passed in as a key so that we can easily
1990 * find it's index in the root list.
1991 */
1992static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
1993 struct ocfs2_extent_list *left_el,
1994 struct ocfs2_extent_list *right_el,
1995 u64 left_el_blkno)
1996{
1997 int i;
1998
1999 BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
2000 le16_to_cpu(left_el->l_tree_depth));
2001
2002 for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
2003 if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
2004 break;
2005 }
2006
2007 /*
2008 * The path walking code should have never returned a root and
2009 * two paths which are not adjacent.
2010 */
2011 BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
2012
2013 ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
2014 &root_el->l_recs[i + 1], right_el);
2015}
2016
2017/*
2018 * We've changed a leaf block (in right_path) and need to reflect that
2019 * change back up the subtree.
2020 *
2021 * This happens in multiple places:
2022 * - When we've moved an extent record from the left path leaf to the right
2023 * path leaf to make room for an empty extent in the left path leaf.
2024 * - When our insert into the right path leaf is at the leftmost edge
2025 * and requires an update of the path immediately to it's left. This
2026 * can occur at the end of some types of rotation and appending inserts.
Tao Ma677b9752008-01-30 14:21:05 +08002027 * - When we've adjusted the last extent record in the left path leaf and the
2028 * 1st extent record in the right path leaf during cross extent block merge.
Mark Fashehdcd05382007-01-16 11:32:23 -08002029 */
Joel Becker4619c732009-02-12 19:02:36 -08002030static void ocfs2_complete_edge_insert(handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -08002031 struct ocfs2_path *left_path,
2032 struct ocfs2_path *right_path,
2033 int subtree_index)
2034{
Joel Beckerec20cec2010-03-19 14:13:52 -07002035 int i, idx;
Mark Fashehdcd05382007-01-16 11:32:23 -08002036 struct ocfs2_extent_list *el, *left_el, *right_el;
2037 struct ocfs2_extent_rec *left_rec, *right_rec;
2038 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2039
2040 /*
2041 * Update the counts and position values within all the
2042 * interior nodes to reflect the leaf rotation we just did.
2043 *
2044 * The root node is handled below the loop.
2045 *
2046 * We begin the loop with right_el and left_el pointing to the
2047 * leaf lists and work our way up.
2048 *
2049 * NOTE: within this loop, left_el and right_el always refer
2050 * to the *child* lists.
2051 */
2052 left_el = path_leaf_el(left_path);
2053 right_el = path_leaf_el(right_path);
2054 for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
Tao Maa09d09b2011-02-22 08:21:10 +08002055 trace_ocfs2_complete_edge_insert(i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002056
2057 /*
2058 * One nice property of knowing that all of these
2059 * nodes are below the root is that we only deal with
2060 * the leftmost right node record and the rightmost
2061 * left node record.
2062 */
2063 el = left_path->p_node[i].el;
2064 idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
2065 left_rec = &el->l_recs[idx];
2066
2067 el = right_path->p_node[i].el;
2068 right_rec = &el->l_recs[0];
2069
2070 ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
2071 right_el);
2072
Joel Beckerec20cec2010-03-19 14:13:52 -07002073 ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
2074 ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002075
2076 /*
2077 * Setup our list pointers now so that the current
2078 * parents become children in the next iteration.
2079 */
2080 left_el = left_path->p_node[i].el;
2081 right_el = right_path->p_node[i].el;
2082 }
2083
2084 /*
2085 * At the root node, adjust the two adjacent records which
2086 * begin our path to the leaves.
2087 */
2088
2089 el = left_path->p_node[subtree_index].el;
2090 left_el = left_path->p_node[subtree_index + 1].el;
2091 right_el = right_path->p_node[subtree_index + 1].el;
2092
2093 ocfs2_adjust_root_records(el, left_el, right_el,
2094 left_path->p_node[subtree_index + 1].bh->b_blocknr);
2095
2096 root_bh = left_path->p_node[subtree_index].bh;
2097
Joel Beckerec20cec2010-03-19 14:13:52 -07002098 ocfs2_journal_dirty(handle, root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002099}
2100
Joel Becker5c601ab2009-02-12 19:10:13 -08002101static int ocfs2_rotate_subtree_right(handle_t *handle,
2102 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08002103 struct ocfs2_path *left_path,
2104 struct ocfs2_path *right_path,
2105 int subtree_index)
2106{
2107 int ret, i;
2108 struct buffer_head *right_leaf_bh;
2109 struct buffer_head *left_leaf_bh = NULL;
2110 struct buffer_head *root_bh;
2111 struct ocfs2_extent_list *right_el, *left_el;
2112 struct ocfs2_extent_rec move_rec;
2113
2114 left_leaf_bh = path_leaf_bh(left_path);
2115 left_el = path_leaf_el(left_path);
2116
2117 if (left_el->l_next_free_rec != left_el->l_count) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002118 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002119 "Inode %llu has non-full interior leaf node %llu"
2120 "(next free = %u)",
Joel Becker5c601ab2009-02-12 19:10:13 -08002121 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002122 (unsigned long long)left_leaf_bh->b_blocknr,
2123 le16_to_cpu(left_el->l_next_free_rec));
2124 return -EROFS;
2125 }
2126
2127 /*
2128 * This extent block may already have an empty record, so we
2129 * return early if so.
2130 */
2131 if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
2132 return 0;
2133
2134 root_bh = left_path->p_node[subtree_index].bh;
2135 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2136
Joel Becker5c601ab2009-02-12 19:10:13 -08002137 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002138 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002139 if (ret) {
2140 mlog_errno(ret);
2141 goto out;
2142 }
2143
2144 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002145 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002146 right_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002147 if (ret) {
2148 mlog_errno(ret);
2149 goto out;
2150 }
2151
Joel Becker5c601ab2009-02-12 19:10:13 -08002152 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002153 left_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002154 if (ret) {
2155 mlog_errno(ret);
2156 goto out;
2157 }
2158 }
2159
2160 right_leaf_bh = path_leaf_bh(right_path);
2161 right_el = path_leaf_el(right_path);
2162
2163 /* This is a code error, not a disk corruption. */
2164 mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
2165 "because rightmost leaf block %llu is empty\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002166 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002167 (unsigned long long)right_leaf_bh->b_blocknr);
2168
2169 ocfs2_create_empty_extent(right_el);
2170
Joel Beckerec20cec2010-03-19 14:13:52 -07002171 ocfs2_journal_dirty(handle, right_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002172
2173 /* Do the copy now. */
2174 i = le16_to_cpu(left_el->l_next_free_rec) - 1;
2175 move_rec = left_el->l_recs[i];
2176 right_el->l_recs[0] = move_rec;
2177
2178 /*
2179 * Clear out the record we just copied and shift everything
2180 * over, leaving an empty extent in the left leaf.
2181 *
2182 * We temporarily subtract from next_free_rec so that the
2183 * shift will lose the tail record (which is now defunct).
2184 */
2185 le16_add_cpu(&left_el->l_next_free_rec, -1);
2186 ocfs2_shift_records_right(left_el);
2187 memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2188 le16_add_cpu(&left_el->l_next_free_rec, 1);
2189
Joel Beckerec20cec2010-03-19 14:13:52 -07002190 ocfs2_journal_dirty(handle, left_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002191
Joel Becker4619c732009-02-12 19:02:36 -08002192 ocfs2_complete_edge_insert(handle, left_path, right_path,
2193 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002194
2195out:
2196 return ret;
2197}
2198
2199/*
2200 * Given a full path, determine what cpos value would return us a path
2201 * containing the leaf immediately to the left of the current one.
2202 *
2203 * Will return zero if the path passed in is already the leftmost path.
2204 */
Tristan Yeee149a72010-05-11 17:54:44 +08002205int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
2206 struct ocfs2_path *path, u32 *cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08002207{
2208 int i, j, ret = 0;
2209 u64 blkno;
2210 struct ocfs2_extent_list *el;
2211
Mark Fashehe48edee2007-03-07 16:46:57 -08002212 BUG_ON(path->p_tree_depth == 0);
2213
Mark Fashehdcd05382007-01-16 11:32:23 -08002214 *cpos = 0;
2215
2216 blkno = path_leaf_bh(path)->b_blocknr;
2217
2218 /* Start at the tree node just above the leaf and work our way up. */
2219 i = path->p_tree_depth - 1;
2220 while (i >= 0) {
2221 el = path->p_node[i].el;
2222
2223 /*
2224 * Find the extent record just before the one in our
2225 * path.
2226 */
2227 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2228 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2229 if (j == 0) {
2230 if (i == 0) {
2231 /*
2232 * We've determined that the
2233 * path specified is already
2234 * the leftmost one - return a
2235 * cpos of zero.
2236 */
2237 goto out;
2238 }
2239 /*
2240 * The leftmost record points to our
2241 * leaf - we need to travel up the
2242 * tree one level.
2243 */
2244 goto next_node;
2245 }
2246
2247 *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08002248 *cpos = *cpos + ocfs2_rec_clusters(el,
2249 &el->l_recs[j - 1]);
2250 *cpos = *cpos - 1;
Mark Fashehdcd05382007-01-16 11:32:23 -08002251 goto out;
2252 }
2253 }
2254
2255 /*
2256 * If we got here, we never found a valid node where
2257 * the tree indicated one should be.
2258 */
2259 ocfs2_error(sb,
2260 "Invalid extent tree at extent block %llu\n",
2261 (unsigned long long)blkno);
2262 ret = -EROFS;
2263 goto out;
2264
2265next_node:
2266 blkno = path->p_node[i].bh->b_blocknr;
2267 i--;
2268 }
2269
2270out:
2271 return ret;
2272}
2273
Mark Fasheh328d5752007-06-18 10:48:04 -07002274/*
2275 * Extend the transaction by enough credits to complete the rotation,
2276 * and still leave at least the original number of credits allocated
2277 * to this transaction.
2278 */
Mark Fashehdcd05382007-01-16 11:32:23 -08002279static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
Mark Fasheh328d5752007-06-18 10:48:04 -07002280 int op_credits,
Mark Fashehdcd05382007-01-16 11:32:23 -08002281 struct ocfs2_path *path)
2282{
Tao Mac901fb02010-04-26 14:34:57 +08002283 int ret = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002284 int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002285
Tao Mac901fb02010-04-26 14:34:57 +08002286 if (handle->h_buffer_credits < credits)
Tao Mac18b8122009-08-18 11:44:07 +08002287 ret = ocfs2_extend_trans(handle,
2288 credits - handle->h_buffer_credits);
Tao Mac18b8122009-08-18 11:44:07 +08002289
Tao Mac901fb02010-04-26 14:34:57 +08002290 return ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08002291}
2292
2293/*
2294 * Trap the case where we're inserting into the theoretical range past
2295 * the _actual_ left leaf range. Otherwise, we'll rotate a record
2296 * whose cpos is less than ours into the right leaf.
2297 *
2298 * It's only necessary to look at the rightmost record of the left
2299 * leaf because the logic that calls us should ensure that the
2300 * theoretical ranges in the path components above the leaves are
2301 * correct.
2302 */
2303static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
2304 u32 insert_cpos)
2305{
2306 struct ocfs2_extent_list *left_el;
2307 struct ocfs2_extent_rec *rec;
2308 int next_free;
2309
2310 left_el = path_leaf_el(left_path);
2311 next_free = le16_to_cpu(left_el->l_next_free_rec);
2312 rec = &left_el->l_recs[next_free - 1];
2313
2314 if (insert_cpos > le32_to_cpu(rec->e_cpos))
2315 return 1;
2316 return 0;
2317}
2318
Mark Fasheh328d5752007-06-18 10:48:04 -07002319static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
2320{
2321 int next_free = le16_to_cpu(el->l_next_free_rec);
2322 unsigned int range;
2323 struct ocfs2_extent_rec *rec;
2324
2325 if (next_free == 0)
2326 return 0;
2327
2328 rec = &el->l_recs[0];
2329 if (ocfs2_is_empty_extent(rec)) {
2330 /* Empty list. */
2331 if (next_free == 1)
2332 return 0;
2333 rec = &el->l_recs[1];
2334 }
2335
2336 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2337 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
2338 return 1;
2339 return 0;
2340}
2341
Mark Fashehdcd05382007-01-16 11:32:23 -08002342/*
2343 * Rotate all the records in a btree right one record, starting at insert_cpos.
2344 *
2345 * The path to the rightmost leaf should be passed in.
2346 *
2347 * The array is assumed to be large enough to hold an entire path (tree depth).
2348 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002349 * Upon successful return from this function:
Mark Fashehdcd05382007-01-16 11:32:23 -08002350 *
2351 * - The 'right_path' array will contain a path to the leaf block
2352 * whose range contains e_cpos.
2353 * - That leaf block will have a single empty extent in list index 0.
2354 * - In the case that the rotation requires a post-insert update,
2355 * *ret_left_path will contain a valid path which can be passed to
2356 * ocfs2_insert_path().
2357 */
Joel Becker1bbf0b82009-02-12 19:42:08 -08002358static int ocfs2_rotate_tree_right(handle_t *handle,
Joel Becker5c601ab2009-02-12 19:10:13 -08002359 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002360 enum ocfs2_split_type split,
Mark Fashehdcd05382007-01-16 11:32:23 -08002361 u32 insert_cpos,
2362 struct ocfs2_path *right_path,
2363 struct ocfs2_path **ret_left_path)
2364{
Mark Fasheh328d5752007-06-18 10:48:04 -07002365 int ret, start, orig_credits = handle->h_buffer_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002366 u32 cpos;
2367 struct ocfs2_path *left_path = NULL;
Joel Becker5c601ab2009-02-12 19:10:13 -08002368 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehdcd05382007-01-16 11:32:23 -08002369
2370 *ret_left_path = NULL;
2371
Joel Beckerffdd7a52008-10-17 22:32:01 -07002372 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002373 if (!left_path) {
2374 ret = -ENOMEM;
2375 mlog_errno(ret);
2376 goto out;
2377 }
2378
Joel Becker5c601ab2009-02-12 19:10:13 -08002379 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002380 if (ret) {
2381 mlog_errno(ret);
2382 goto out;
2383 }
2384
Tao Maa09d09b2011-02-22 08:21:10 +08002385 trace_ocfs2_rotate_tree_right(
2386 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2387 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002388
2389 /*
2390 * What we want to do here is:
2391 *
2392 * 1) Start with the rightmost path.
2393 *
2394 * 2) Determine a path to the leaf block directly to the left
2395 * of that leaf.
2396 *
2397 * 3) Determine the 'subtree root' - the lowest level tree node
2398 * which contains a path to both leaves.
2399 *
2400 * 4) Rotate the subtree.
2401 *
2402 * 5) Find the next subtree by considering the left path to be
2403 * the new right path.
2404 *
2405 * The check at the top of this while loop also accepts
2406 * insert_cpos == cpos because cpos is only a _theoretical_
2407 * value to get us the left path - insert_cpos might very well
2408 * be filling that hole.
2409 *
2410 * Stop at a cpos of '0' because we either started at the
2411 * leftmost branch (i.e., a tree with one branch and a
2412 * rotation inside of it), or we've gone as far as we can in
2413 * rotating subtrees.
2414 */
2415 while (cpos && insert_cpos <= cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08002416 trace_ocfs2_rotate_tree_right(
2417 (unsigned long long)
2418 ocfs2_metadata_cache_owner(et->et_ci),
2419 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002420
Joel Becker5c601ab2009-02-12 19:10:13 -08002421 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002422 if (ret) {
2423 mlog_errno(ret);
2424 goto out;
2425 }
2426
2427 mlog_bug_on_msg(path_leaf_bh(left_path) ==
2428 path_leaf_bh(right_path),
Joel Becker5c601ab2009-02-12 19:10:13 -08002429 "Owner %llu: error during insert of %u "
Mark Fashehdcd05382007-01-16 11:32:23 -08002430 "(left path cpos %u) results in two identical "
2431 "paths ending at %llu\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002432 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2433 insert_cpos, cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08002434 (unsigned long long)
2435 path_leaf_bh(left_path)->b_blocknr);
2436
Mark Fasheh328d5752007-06-18 10:48:04 -07002437 if (split == SPLIT_NONE &&
2438 ocfs2_rotate_requires_path_adjustment(left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002439 insert_cpos)) {
Mark Fashehdcd05382007-01-16 11:32:23 -08002440
2441 /*
2442 * We've rotated the tree as much as we
2443 * should. The rest is up to
2444 * ocfs2_insert_path() to complete, after the
2445 * record insertion. We indicate this
2446 * situation by returning the left path.
2447 *
2448 * The reason we don't adjust the records here
2449 * before the record insert is that an error
2450 * later might break the rule where a parent
2451 * record e_cpos will reflect the actual
2452 * e_cpos of the 1st nonempty record of the
2453 * child list.
2454 */
2455 *ret_left_path = left_path;
2456 goto out_ret_path;
2457 }
2458
Joel Becker7dc02802009-02-12 19:20:13 -08002459 start = ocfs2_find_subtree_root(et, left_path, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002460
Tao Maa09d09b2011-02-22 08:21:10 +08002461 trace_ocfs2_rotate_subtree(start,
2462 (unsigned long long)
2463 right_path->p_node[start].bh->b_blocknr,
2464 right_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08002465
2466 ret = ocfs2_extend_rotate_transaction(handle, start,
Mark Fasheh328d5752007-06-18 10:48:04 -07002467 orig_credits, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002468 if (ret) {
2469 mlog_errno(ret);
2470 goto out;
2471 }
2472
Joel Becker5c601ab2009-02-12 19:10:13 -08002473 ret = ocfs2_rotate_subtree_right(handle, et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002474 right_path, start);
2475 if (ret) {
2476 mlog_errno(ret);
2477 goto out;
2478 }
2479
Mark Fasheh328d5752007-06-18 10:48:04 -07002480 if (split != SPLIT_NONE &&
2481 ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
2482 insert_cpos)) {
2483 /*
2484 * A rotate moves the rightmost left leaf
2485 * record over to the leftmost right leaf
2486 * slot. If we're doing an extent split
2487 * instead of a real insert, then we have to
2488 * check that the extent to be split wasn't
2489 * just moved over. If it was, then we can
2490 * exit here, passing left_path back -
2491 * ocfs2_split_extent() is smart enough to
2492 * search both leaves.
2493 */
2494 *ret_left_path = left_path;
2495 goto out_ret_path;
2496 }
2497
Mark Fashehdcd05382007-01-16 11:32:23 -08002498 /*
2499 * There is no need to re-read the next right path
2500 * as we know that it'll be our current left
2501 * path. Optimize by copying values instead.
2502 */
2503 ocfs2_mv_path(right_path, left_path);
2504
Joel Becker5c601ab2009-02-12 19:10:13 -08002505 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002506 if (ret) {
2507 mlog_errno(ret);
2508 goto out;
2509 }
2510 }
2511
2512out:
2513 ocfs2_free_path(left_path);
2514
2515out_ret_path:
2516 return ret;
2517}
2518
Joel Becker09106ba2009-02-12 19:43:57 -08002519static int ocfs2_update_edge_lengths(handle_t *handle,
2520 struct ocfs2_extent_tree *et,
Tao Ma3c5e1062009-07-21 15:42:05 +08002521 int subtree_index, struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002522{
Tao Ma3c5e1062009-07-21 15:42:05 +08002523 int i, idx, ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002524 struct ocfs2_extent_rec *rec;
2525 struct ocfs2_extent_list *el;
2526 struct ocfs2_extent_block *eb;
2527 u32 range;
2528
Tao Ma3c5e1062009-07-21 15:42:05 +08002529 /*
2530 * In normal tree rotation process, we will never touch the
2531 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
2532 * doesn't reserve the credits for them either.
2533 *
2534 * But we do have a special case here which will update the rightmost
2535 * records for all the bh in the path.
2536 * So we have to allocate extra credits and access them.
2537 */
Tao Mac901fb02010-04-26 14:34:57 +08002538 ret = ocfs2_extend_trans(handle, subtree_index);
Tao Ma3c5e1062009-07-21 15:42:05 +08002539 if (ret) {
2540 mlog_errno(ret);
2541 goto out;
2542 }
2543
Joel Becker09106ba2009-02-12 19:43:57 -08002544 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002545 if (ret) {
2546 mlog_errno(ret);
2547 goto out;
2548 }
2549
Mark Fasheh328d5752007-06-18 10:48:04 -07002550 /* Path should always be rightmost. */
2551 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2552 BUG_ON(eb->h_next_leaf_blk != 0ULL);
2553
2554 el = &eb->h_list;
2555 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2556 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2557 rec = &el->l_recs[idx];
2558 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2559
2560 for (i = 0; i < path->p_tree_depth; i++) {
2561 el = path->p_node[i].el;
2562 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2563 rec = &el->l_recs[idx];
2564
2565 rec->e_int_clusters = cpu_to_le32(range);
2566 le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2567
2568 ocfs2_journal_dirty(handle, path->p_node[i].bh);
2569 }
Tao Ma3c5e1062009-07-21 15:42:05 +08002570out:
2571 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002572}
2573
Joel Becker6641b0c2009-02-12 18:57:52 -08002574static void ocfs2_unlink_path(handle_t *handle,
2575 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002576 struct ocfs2_cached_dealloc_ctxt *dealloc,
2577 struct ocfs2_path *path, int unlink_start)
2578{
2579 int ret, i;
2580 struct ocfs2_extent_block *eb;
2581 struct ocfs2_extent_list *el;
2582 struct buffer_head *bh;
2583
2584 for(i = unlink_start; i < path_num_items(path); i++) {
2585 bh = path->p_node[i].bh;
2586
2587 eb = (struct ocfs2_extent_block *)bh->b_data;
2588 /*
2589 * Not all nodes might have had their final count
2590 * decremented by the caller - handle this here.
2591 */
2592 el = &eb->h_list;
2593 if (le16_to_cpu(el->l_next_free_rec) > 1) {
2594 mlog(ML_ERROR,
2595 "Inode %llu, attempted to remove extent block "
2596 "%llu with %u records\n",
Joel Becker6641b0c2009-02-12 18:57:52 -08002597 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07002598 (unsigned long long)le64_to_cpu(eb->h_blkno),
2599 le16_to_cpu(el->l_next_free_rec));
2600
2601 ocfs2_journal_dirty(handle, bh);
Joel Becker6641b0c2009-02-12 18:57:52 -08002602 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002603 continue;
2604 }
2605
2606 el->l_next_free_rec = 0;
2607 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2608
2609 ocfs2_journal_dirty(handle, bh);
2610
2611 ret = ocfs2_cache_extent_block_free(dealloc, eb);
2612 if (ret)
2613 mlog_errno(ret);
2614
Joel Becker6641b0c2009-02-12 18:57:52 -08002615 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002616 }
2617}
2618
Joel Becker6641b0c2009-02-12 18:57:52 -08002619static void ocfs2_unlink_subtree(handle_t *handle,
2620 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002621 struct ocfs2_path *left_path,
2622 struct ocfs2_path *right_path,
2623 int subtree_index,
2624 struct ocfs2_cached_dealloc_ctxt *dealloc)
2625{
2626 int i;
2627 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2628 struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
2629 struct ocfs2_extent_list *el;
2630 struct ocfs2_extent_block *eb;
2631
2632 el = path_leaf_el(left_path);
2633
2634 eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2635
2636 for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2637 if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2638 break;
2639
2640 BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2641
2642 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2643 le16_add_cpu(&root_el->l_next_free_rec, -1);
2644
2645 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2646 eb->h_next_leaf_blk = 0;
2647
2648 ocfs2_journal_dirty(handle, root_bh);
2649 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2650
Joel Becker6641b0c2009-02-12 18:57:52 -08002651 ocfs2_unlink_path(handle, et, dealloc, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002652 subtree_index + 1);
2653}
2654
Joel Becker1e2dd632009-02-12 19:45:28 -08002655static int ocfs2_rotate_subtree_left(handle_t *handle,
2656 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002657 struct ocfs2_path *left_path,
2658 struct ocfs2_path *right_path,
2659 int subtree_index,
2660 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker1e2dd632009-02-12 19:45:28 -08002661 int *deleted)
Mark Fasheh328d5752007-06-18 10:48:04 -07002662{
2663 int ret, i, del_right_subtree = 0, right_has_empty = 0;
Tao Mae7d4cb62008-08-18 17:38:44 +08002664 struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002665 struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2666 struct ocfs2_extent_block *eb;
2667
2668 *deleted = 0;
2669
2670 right_leaf_el = path_leaf_el(right_path);
2671 left_leaf_el = path_leaf_el(left_path);
2672 root_bh = left_path->p_node[subtree_index].bh;
2673 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2674
2675 if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2676 return 0;
2677
2678 eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2679 if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2680 /*
2681 * It's legal for us to proceed if the right leaf is
2682 * the rightmost one and it has an empty extent. There
2683 * are two cases to handle - whether the leaf will be
2684 * empty after removal or not. If the leaf isn't empty
2685 * then just remove the empty extent up front. The
2686 * next block will handle empty leaves by flagging
2687 * them for unlink.
2688 *
2689 * Non rightmost leaves will throw -EAGAIN and the
2690 * caller can manually move the subtree and retry.
2691 */
2692
2693 if (eb->h_next_leaf_blk != 0ULL)
2694 return -EAGAIN;
2695
2696 if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002697 ret = ocfs2_journal_access_eb(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002698 path_leaf_bh(right_path),
2699 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002700 if (ret) {
2701 mlog_errno(ret);
2702 goto out;
2703 }
2704
2705 ocfs2_remove_empty_extent(right_leaf_el);
2706 } else
2707 right_has_empty = 1;
2708 }
2709
2710 if (eb->h_next_leaf_blk == 0ULL &&
2711 le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2712 /*
2713 * We have to update i_last_eb_blk during the meta
2714 * data delete.
2715 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08002716 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07002717 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002718 if (ret) {
2719 mlog_errno(ret);
2720 goto out;
2721 }
2722
2723 del_right_subtree = 1;
2724 }
2725
2726 /*
2727 * Getting here with an empty extent in the right path implies
2728 * that it's the rightmost path and will be deleted.
2729 */
2730 BUG_ON(right_has_empty && !del_right_subtree);
2731
Joel Becker1e2dd632009-02-12 19:45:28 -08002732 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002733 subtree_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07002734 if (ret) {
2735 mlog_errno(ret);
2736 goto out;
2737 }
2738
2739 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002740 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002741 right_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002742 if (ret) {
2743 mlog_errno(ret);
2744 goto out;
2745 }
2746
Joel Becker1e2dd632009-02-12 19:45:28 -08002747 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002748 left_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002749 if (ret) {
2750 mlog_errno(ret);
2751 goto out;
2752 }
2753 }
2754
2755 if (!right_has_empty) {
2756 /*
2757 * Only do this if we're moving a real
2758 * record. Otherwise, the action is delayed until
2759 * after removal of the right path in which case we
2760 * can do a simple shift to remove the empty extent.
2761 */
2762 ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2763 memset(&right_leaf_el->l_recs[0], 0,
2764 sizeof(struct ocfs2_extent_rec));
2765 }
2766 if (eb->h_next_leaf_blk == 0ULL) {
2767 /*
2768 * Move recs over to get rid of empty extent, decrease
2769 * next_free. This is allowed to remove the last
2770 * extent in our leaf (setting l_next_free_rec to
2771 * zero) - the delete code below won't care.
2772 */
2773 ocfs2_remove_empty_extent(right_leaf_el);
2774 }
2775
Joel Beckerec20cec2010-03-19 14:13:52 -07002776 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2777 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07002778
2779 if (del_right_subtree) {
Joel Becker6641b0c2009-02-12 18:57:52 -08002780 ocfs2_unlink_subtree(handle, et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002781 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08002782 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08002783 left_path);
2784 if (ret) {
2785 mlog_errno(ret);
2786 goto out;
2787 }
Mark Fasheh328d5752007-06-18 10:48:04 -07002788
2789 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07002790 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07002791
2792 /*
2793 * Removal of the extent in the left leaf was skipped
2794 * above so we could delete the right path
2795 * 1st.
2796 */
2797 if (right_has_empty)
2798 ocfs2_remove_empty_extent(left_leaf_el);
2799
Joel Beckerec20cec2010-03-19 14:13:52 -07002800 ocfs2_journal_dirty(handle, et_root_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002801
2802 *deleted = 1;
2803 } else
Joel Becker4619c732009-02-12 19:02:36 -08002804 ocfs2_complete_edge_insert(handle, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002805 subtree_index);
2806
2807out:
2808 return ret;
2809}
2810
2811/*
2812 * Given a full path, determine what cpos value would return us a path
2813 * containing the leaf immediately to the right of the current one.
2814 *
2815 * Will return zero if the path passed in is already the rightmost path.
2816 *
2817 * This looks similar, but is subtly different to
2818 * ocfs2_find_cpos_for_left_leaf().
2819 */
Tao Ma38a04e42009-11-30 14:32:19 +08002820int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2821 struct ocfs2_path *path, u32 *cpos)
Mark Fasheh328d5752007-06-18 10:48:04 -07002822{
2823 int i, j, ret = 0;
2824 u64 blkno;
2825 struct ocfs2_extent_list *el;
2826
2827 *cpos = 0;
2828
2829 if (path->p_tree_depth == 0)
2830 return 0;
2831
2832 blkno = path_leaf_bh(path)->b_blocknr;
2833
2834 /* Start at the tree node just above the leaf and work our way up. */
2835 i = path->p_tree_depth - 1;
2836 while (i >= 0) {
2837 int next_free;
2838
2839 el = path->p_node[i].el;
2840
2841 /*
2842 * Find the extent record just after the one in our
2843 * path.
2844 */
2845 next_free = le16_to_cpu(el->l_next_free_rec);
2846 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2847 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2848 if (j == (next_free - 1)) {
2849 if (i == 0) {
2850 /*
2851 * We've determined that the
2852 * path specified is already
2853 * the rightmost one - return a
2854 * cpos of zero.
2855 */
2856 goto out;
2857 }
2858 /*
2859 * The rightmost record points to our
2860 * leaf - we need to travel up the
2861 * tree one level.
2862 */
2863 goto next_node;
2864 }
2865
2866 *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2867 goto out;
2868 }
2869 }
2870
2871 /*
2872 * If we got here, we never found a valid node where
2873 * the tree indicated one should be.
2874 */
2875 ocfs2_error(sb,
2876 "Invalid extent tree at extent block %llu\n",
2877 (unsigned long long)blkno);
2878 ret = -EROFS;
2879 goto out;
2880
2881next_node:
2882 blkno = path->p_node[i].bh->b_blocknr;
2883 i--;
2884 }
2885
2886out:
2887 return ret;
2888}
2889
Joel Becker70f18c02009-02-13 02:09:31 -08002890static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2891 struct ocfs2_extent_tree *et,
Joel Becker13723d02008-10-17 19:25:01 -07002892 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002893{
2894 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07002895 struct buffer_head *bh = path_leaf_bh(path);
2896 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002897
2898 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2899 return 0;
2900
Joel Becker70f18c02009-02-13 02:09:31 -08002901 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Joel Becker13723d02008-10-17 19:25:01 -07002902 path_num_items(path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07002903 if (ret) {
2904 mlog_errno(ret);
2905 goto out;
2906 }
2907
2908 ocfs2_remove_empty_extent(el);
Joel Beckerec20cec2010-03-19 14:13:52 -07002909 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002910
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
Tao Maa09d09b2011-02-22 08:21:10 +08002964 trace_ocfs2_rotate_subtree(subtree_root,
Mark Fasheh328d5752007-06-18 10:48:04 -07002965 (unsigned long long)
2966 right_path->p_node[subtree_root].bh->b_blocknr,
2967 right_path->p_tree_depth);
2968
2969 ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
2970 orig_credits, left_path);
2971 if (ret) {
2972 mlog_errno(ret);
2973 goto out;
2974 }
2975
Mark Fashehe8aed342007-12-03 16:43:01 -08002976 /*
2977 * Caller might still want to make changes to the
2978 * tree root, so re-add it to the journal here.
2979 */
Joel Beckere46f74d2009-02-12 19:47:43 -08002980 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002981 left_path, 0);
Mark Fashehe8aed342007-12-03 16:43:01 -08002982 if (ret) {
2983 mlog_errno(ret);
2984 goto out;
2985 }
2986
Joel Becker1e2dd632009-02-12 19:45:28 -08002987 ret = ocfs2_rotate_subtree_left(handle, et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002988 right_path, subtree_root,
Joel Becker1e2dd632009-02-12 19:45:28 -08002989 dealloc, &deleted);
Mark Fasheh328d5752007-06-18 10:48:04 -07002990 if (ret == -EAGAIN) {
2991 /*
2992 * The rotation has to temporarily stop due to
2993 * the right subtree having an empty
2994 * extent. Pass it back to the caller for a
2995 * fixup.
2996 */
2997 *empty_extent_path = right_path;
2998 right_path = NULL;
2999 goto out;
3000 }
3001 if (ret) {
3002 mlog_errno(ret);
3003 goto out;
3004 }
3005
3006 /*
3007 * The subtree rotate might have removed records on
3008 * the rightmost edge. If so, then rotation is
3009 * complete.
3010 */
3011 if (deleted)
3012 break;
3013
3014 ocfs2_mv_path(left_path, right_path);
3015
Joel Beckere46f74d2009-02-12 19:47:43 -08003016 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003017 &right_cpos);
3018 if (ret) {
3019 mlog_errno(ret);
3020 goto out;
3021 }
3022 }
3023
3024out:
3025 ocfs2_free_path(right_path);
3026 ocfs2_free_path(left_path);
3027
3028 return ret;
3029}
3030
Joel Becker70f18c02009-02-13 02:09:31 -08003031static int ocfs2_remove_rightmost_path(handle_t *handle,
3032 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003033 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003034 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003035{
3036 int ret, subtree_index;
3037 u32 cpos;
3038 struct ocfs2_path *left_path = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003039 struct ocfs2_extent_block *eb;
3040 struct ocfs2_extent_list *el;
3041
Mark Fasheh328d5752007-06-18 10:48:04 -07003042
Joel Becker6136ca52009-02-12 19:32:43 -08003043 ret = ocfs2_et_sanity_check(et);
Tao Mae7d4cb62008-08-18 17:38:44 +08003044 if (ret)
3045 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003046 /*
3047 * There's two ways we handle this depending on
3048 * whether path is the only existing one.
3049 */
3050 ret = ocfs2_extend_rotate_transaction(handle, 0,
3051 handle->h_buffer_credits,
3052 path);
3053 if (ret) {
3054 mlog_errno(ret);
3055 goto out;
3056 }
3057
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003058 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003059 if (ret) {
3060 mlog_errno(ret);
3061 goto out;
3062 }
3063
Joel Becker3d03a302009-02-12 17:49:26 -08003064 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3065 path, &cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003066 if (ret) {
3067 mlog_errno(ret);
3068 goto out;
3069 }
3070
3071 if (cpos) {
3072 /*
3073 * We have a path to the left of this one - it needs
3074 * an update too.
3075 */
Joel Beckerffdd7a52008-10-17 22:32:01 -07003076 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003077 if (!left_path) {
3078 ret = -ENOMEM;
3079 mlog_errno(ret);
3080 goto out;
3081 }
3082
Joel Beckerfacdb772009-02-12 18:08:48 -08003083 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003084 if (ret) {
3085 mlog_errno(ret);
3086 goto out;
3087 }
3088
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003089 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003090 if (ret) {
3091 mlog_errno(ret);
3092 goto out;
3093 }
3094
Joel Becker7dc02802009-02-12 19:20:13 -08003095 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003096
Joel Becker6641b0c2009-02-12 18:57:52 -08003097 ocfs2_unlink_subtree(handle, et, left_path, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003098 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08003099 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08003100 left_path);
3101 if (ret) {
3102 mlog_errno(ret);
3103 goto out;
3104 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003105
3106 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07003107 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003108 } else {
3109 /*
3110 * 'path' is also the leftmost path which
3111 * means it must be the only one. This gets
3112 * handled differently because we want to
Joel Becker70f18c02009-02-13 02:09:31 -08003113 * revert the root back to having extents
Mark Fasheh328d5752007-06-18 10:48:04 -07003114 * in-line.
3115 */
Joel Becker6641b0c2009-02-12 18:57:52 -08003116 ocfs2_unlink_path(handle, et, dealloc, path, 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003117
Joel Beckerce1d9ea2008-08-20 16:30:07 -07003118 el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003119 el->l_tree_depth = 0;
3120 el->l_next_free_rec = 0;
3121 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3122
Joel Becker35dc0aa2008-08-20 16:25:06 -07003123 ocfs2_et_set_last_eb_blk(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003124 }
3125
3126 ocfs2_journal_dirty(handle, path_root_bh(path));
3127
3128out:
3129 ocfs2_free_path(left_path);
3130 return ret;
3131}
3132
3133/*
3134 * Left rotation of btree records.
3135 *
3136 * In many ways, this is (unsurprisingly) the opposite of right
3137 * rotation. We start at some non-rightmost path containing an empty
3138 * extent in the leaf block. The code works its way to the rightmost
3139 * path by rotating records to the left in every subtree.
3140 *
3141 * This is used by any code which reduces the number of extent records
3142 * in a leaf. After removal, an empty record should be placed in the
3143 * leftmost list position.
3144 *
3145 * This won't handle a length update of the rightmost path records if
3146 * the rightmost tree leaf record is removed so the caller is
3147 * responsible for detecting and correcting that.
3148 */
Joel Becker70f18c02009-02-13 02:09:31 -08003149static int ocfs2_rotate_tree_left(handle_t *handle,
3150 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003151 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003152 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003153{
3154 int ret, orig_credits = handle->h_buffer_credits;
3155 struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3156 struct ocfs2_extent_block *eb;
3157 struct ocfs2_extent_list *el;
3158
3159 el = path_leaf_el(path);
3160 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3161 return 0;
3162
3163 if (path->p_tree_depth == 0) {
3164rightmost_no_delete:
3165 /*
Tao Mae7d4cb62008-08-18 17:38:44 +08003166 * Inline extents. This is trivially handled, so do
Mark Fasheh328d5752007-06-18 10:48:04 -07003167 * it up front.
3168 */
Joel Becker70f18c02009-02-13 02:09:31 -08003169 ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003170 if (ret)
3171 mlog_errno(ret);
3172 goto out;
3173 }
3174
3175 /*
3176 * Handle rightmost branch now. There's several cases:
3177 * 1) simple rotation leaving records in there. That's trivial.
3178 * 2) rotation requiring a branch delete - there's no more
3179 * records left. Two cases of this:
3180 * a) There are branches to the left.
3181 * b) This is also the leftmost (the only) branch.
3182 *
3183 * 1) is handled via ocfs2_rotate_rightmost_leaf_left()
3184 * 2a) we need the left branch so that we can update it with the unlink
Joel Becker70f18c02009-02-13 02:09:31 -08003185 * 2b) we need to bring the root back to inline extents.
Mark Fasheh328d5752007-06-18 10:48:04 -07003186 */
3187
3188 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3189 el = &eb->h_list;
3190 if (eb->h_next_leaf_blk == 0) {
3191 /*
3192 * This gets a bit tricky if we're going to delete the
3193 * rightmost path. Get the other cases out of the way
3194 * 1st.
3195 */
3196 if (le16_to_cpu(el->l_next_free_rec) > 1)
3197 goto rightmost_no_delete;
3198
3199 if (le16_to_cpu(el->l_next_free_rec) == 0) {
3200 ret = -EIO;
Joel Becker70f18c02009-02-13 02:09:31 -08003201 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3202 "Owner %llu has empty extent block at %llu",
3203 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07003204 (unsigned long long)le64_to_cpu(eb->h_blkno));
3205 goto out;
3206 }
3207
3208 /*
3209 * XXX: The caller can not trust "path" any more after
3210 * this as it will have been deleted. What do we do?
3211 *
3212 * In theory the rotate-for-merge code will never get
3213 * here because it'll always ask for a rotate in a
3214 * nonempty list.
3215 */
3216
Joel Becker70f18c02009-02-13 02:09:31 -08003217 ret = ocfs2_remove_rightmost_path(handle, et, path,
3218 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003219 if (ret)
3220 mlog_errno(ret);
3221 goto out;
3222 }
3223
3224 /*
3225 * Now we can loop, remembering the path we get from -EAGAIN
3226 * and restarting from there.
3227 */
3228try_rotate:
Joel Beckere46f74d2009-02-12 19:47:43 -08003229 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3230 dealloc, &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003231 if (ret && ret != -EAGAIN) {
3232 mlog_errno(ret);
3233 goto out;
3234 }
3235
3236 while (ret == -EAGAIN) {
3237 tmp_path = restart_path;
3238 restart_path = NULL;
3239
Joel Beckere46f74d2009-02-12 19:47:43 -08003240 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07003241 tmp_path, dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08003242 &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003243 if (ret && ret != -EAGAIN) {
3244 mlog_errno(ret);
3245 goto out;
3246 }
3247
3248 ocfs2_free_path(tmp_path);
3249 tmp_path = NULL;
3250
3251 if (ret == 0)
3252 goto try_rotate;
3253 }
3254
3255out:
3256 ocfs2_free_path(tmp_path);
3257 ocfs2_free_path(restart_path);
3258 return ret;
3259}
3260
3261static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3262 int index)
3263{
3264 struct ocfs2_extent_rec *rec = &el->l_recs[index];
3265 unsigned int size;
3266
3267 if (rec->e_leaf_clusters == 0) {
3268 /*
3269 * We consumed all of the merged-from record. An empty
3270 * extent cannot exist anywhere but the 1st array
3271 * position, so move things over if the merged-from
3272 * record doesn't occupy that position.
3273 *
3274 * This creates a new empty extent so the caller
3275 * should be smart enough to have removed any existing
3276 * ones.
3277 */
3278 if (index > 0) {
3279 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3280 size = index * sizeof(struct ocfs2_extent_rec);
3281 memmove(&el->l_recs[1], &el->l_recs[0], size);
3282 }
3283
3284 /*
3285 * Always memset - the caller doesn't check whether it
3286 * created an empty extent, so there could be junk in
3287 * the other fields.
3288 */
3289 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3290 }
3291}
3292
Joel Becker4fe82c32009-02-13 02:16:08 -08003293static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003294 struct ocfs2_path *left_path,
3295 struct ocfs2_path **ret_right_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07003296{
3297 int ret;
Tao Ma677b9752008-01-30 14:21:05 +08003298 u32 right_cpos;
3299 struct ocfs2_path *right_path = NULL;
3300 struct ocfs2_extent_list *left_el;
3301
3302 *ret_right_path = NULL;
3303
3304 /* This function shouldn't be called for non-trees. */
3305 BUG_ON(left_path->p_tree_depth == 0);
3306
3307 left_el = path_leaf_el(left_path);
3308 BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3309
Joel Becker4fe82c32009-02-13 02:16:08 -08003310 ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3311 left_path, &right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003312 if (ret) {
3313 mlog_errno(ret);
3314 goto out;
3315 }
3316
3317 /* This function shouldn't be called for the rightmost leaf. */
3318 BUG_ON(right_cpos == 0);
3319
Joel Beckerffdd7a52008-10-17 22:32:01 -07003320 right_path = ocfs2_new_path_from_path(left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003321 if (!right_path) {
3322 ret = -ENOMEM;
3323 mlog_errno(ret);
3324 goto out;
3325 }
3326
Joel Becker4fe82c32009-02-13 02:16:08 -08003327 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003328 if (ret) {
3329 mlog_errno(ret);
3330 goto out;
3331 }
3332
3333 *ret_right_path = right_path;
3334out:
3335 if (ret)
3336 ocfs2_free_path(right_path);
3337 return ret;
3338}
3339
3340/*
3341 * Remove split_rec clusters from the record at index and merge them
3342 * onto the beginning of the record "next" to it.
3343 * For index < l_count - 1, the next means the extent rec at index + 1.
3344 * For index == l_count - 1, the "next" means the 1st extent rec of the
3345 * next extent block.
3346 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003347static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003348 handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003349 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003350 struct ocfs2_extent_rec *split_rec,
3351 int index)
3352{
3353 int ret, next_free, i;
Mark Fasheh328d5752007-06-18 10:48:04 -07003354 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3355 struct ocfs2_extent_rec *left_rec;
3356 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003357 struct ocfs2_extent_list *right_el;
3358 struct ocfs2_path *right_path = NULL;
3359 int subtree_index = 0;
3360 struct ocfs2_extent_list *el = path_leaf_el(left_path);
3361 struct buffer_head *bh = path_leaf_bh(left_path);
3362 struct buffer_head *root_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003363
3364 BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
Mark Fasheh328d5752007-06-18 10:48:04 -07003365 left_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003366
Al Viro9d8df6a2008-05-21 06:32:11 +01003367 if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
Tao Ma677b9752008-01-30 14:21:05 +08003368 le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3369 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003370 ret = ocfs2_get_right_path(et, left_path, &right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003371 if (ret) {
3372 mlog_errno(ret);
3373 goto out;
3374 }
3375
3376 right_el = path_leaf_el(right_path);
3377 next_free = le16_to_cpu(right_el->l_next_free_rec);
3378 BUG_ON(next_free <= 0);
3379 right_rec = &right_el->l_recs[0];
3380 if (ocfs2_is_empty_extent(right_rec)) {
Al Viro9d8df6a2008-05-21 06:32:11 +01003381 BUG_ON(next_free <= 1);
Tao Ma677b9752008-01-30 14:21:05 +08003382 right_rec = &right_el->l_recs[1];
3383 }
3384
3385 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3386 le16_to_cpu(left_rec->e_leaf_clusters) !=
3387 le32_to_cpu(right_rec->e_cpos));
3388
Joel Becker7dc02802009-02-12 19:20:13 -08003389 subtree_index = ocfs2_find_subtree_root(et, left_path,
3390 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003391
3392 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3393 handle->h_buffer_credits,
3394 right_path);
3395 if (ret) {
3396 mlog_errno(ret);
3397 goto out;
3398 }
3399
3400 root_bh = left_path->p_node[subtree_index].bh;
3401 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3402
Joel Becker7dc02802009-02-12 19:20:13 -08003403 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003404 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003405 if (ret) {
3406 mlog_errno(ret);
3407 goto out;
3408 }
3409
3410 for (i = subtree_index + 1;
3411 i < path_num_items(right_path); i++) {
Joel Becker7dc02802009-02-12 19:20:13 -08003412 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003413 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003414 if (ret) {
3415 mlog_errno(ret);
3416 goto out;
3417 }
3418
Joel Becker7dc02802009-02-12 19:20:13 -08003419 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003420 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003421 if (ret) {
3422 mlog_errno(ret);
3423 goto out;
3424 }
3425 }
3426
3427 } else {
3428 BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3429 right_rec = &el->l_recs[index + 1];
3430 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003431
Joel Becker7dc02802009-02-12 19:20:13 -08003432 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
Joel Becker13723d02008-10-17 19:25:01 -07003433 path_num_items(left_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003434 if (ret) {
3435 mlog_errno(ret);
3436 goto out;
3437 }
3438
3439 le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3440
3441 le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3442 le64_add_cpu(&right_rec->e_blkno,
Joel Becker7dc02802009-02-12 19:20:13 -08003443 -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3444 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003445 le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3446
3447 ocfs2_cleanup_merge(el, index);
3448
Joel Beckerec20cec2010-03-19 14:13:52 -07003449 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003450 if (right_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003451 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Joel Becker4619c732009-02-12 19:02:36 -08003452 ocfs2_complete_edge_insert(handle, left_path, right_path,
3453 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003454 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003455out:
Tao Ma677b9752008-01-30 14:21:05 +08003456 if (right_path)
3457 ocfs2_free_path(right_path);
3458 return ret;
3459}
3460
Joel Becker4fe82c32009-02-13 02:16:08 -08003461static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003462 struct ocfs2_path *right_path,
3463 struct ocfs2_path **ret_left_path)
3464{
3465 int ret;
3466 u32 left_cpos;
3467 struct ocfs2_path *left_path = NULL;
3468
3469 *ret_left_path = NULL;
3470
3471 /* This function shouldn't be called for non-trees. */
3472 BUG_ON(right_path->p_tree_depth == 0);
3473
Joel Becker4fe82c32009-02-13 02:16:08 -08003474 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003475 right_path, &left_cpos);
3476 if (ret) {
3477 mlog_errno(ret);
3478 goto out;
3479 }
3480
3481 /* This function shouldn't be called for the leftmost leaf. */
3482 BUG_ON(left_cpos == 0);
3483
Joel Beckerffdd7a52008-10-17 22:32:01 -07003484 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003485 if (!left_path) {
3486 ret = -ENOMEM;
3487 mlog_errno(ret);
3488 goto out;
3489 }
3490
Joel Becker4fe82c32009-02-13 02:16:08 -08003491 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003492 if (ret) {
3493 mlog_errno(ret);
3494 goto out;
3495 }
3496
3497 *ret_left_path = left_path;
3498out:
3499 if (ret)
3500 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003501 return ret;
3502}
3503
3504/*
3505 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003506 * onto the tail of the record "before" it.
3507 * For index > 0, the "before" means the extent rec at index - 1.
3508 *
3509 * For index == 0, the "before" means the last record of the previous
3510 * extent block. And there is also a situation that we may need to
3511 * remove the rightmost leaf extent block in the right_path and change
3512 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003513 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003514static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003515 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003516 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003517 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003518 struct ocfs2_cached_dealloc_ctxt *dealloc,
3519 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003520{
Tao Ma677b9752008-01-30 14:21:05 +08003521 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003522 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3523 struct ocfs2_extent_rec *left_rec;
3524 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003525 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3526 struct buffer_head *bh = path_leaf_bh(right_path);
3527 struct buffer_head *root_bh = NULL;
3528 struct ocfs2_path *left_path = NULL;
3529 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003530
Tao Ma677b9752008-01-30 14:21:05 +08003531 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003532
Mark Fasheh328d5752007-06-18 10:48:04 -07003533 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003534 if (index == 0) {
3535 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003536 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003537 if (ret) {
3538 mlog_errno(ret);
3539 goto out;
3540 }
3541
3542 left_el = path_leaf_el(left_path);
3543 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3544 le16_to_cpu(left_el->l_count));
3545
3546 left_rec = &left_el->l_recs[
3547 le16_to_cpu(left_el->l_next_free_rec) - 1];
3548 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3549 le16_to_cpu(left_rec->e_leaf_clusters) !=
3550 le32_to_cpu(split_rec->e_cpos));
3551
Joel Becker7dc02802009-02-12 19:20:13 -08003552 subtree_index = ocfs2_find_subtree_root(et, left_path,
3553 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003554
3555 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3556 handle->h_buffer_credits,
3557 left_path);
3558 if (ret) {
3559 mlog_errno(ret);
3560 goto out;
3561 }
3562
3563 root_bh = left_path->p_node[subtree_index].bh;
3564 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3565
Joel Becker4fe82c32009-02-13 02:16:08 -08003566 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003567 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003568 if (ret) {
3569 mlog_errno(ret);
3570 goto out;
3571 }
3572
3573 for (i = subtree_index + 1;
3574 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003575 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003576 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003577 if (ret) {
3578 mlog_errno(ret);
3579 goto out;
3580 }
3581
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 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003584 if (ret) {
3585 mlog_errno(ret);
3586 goto out;
3587 }
3588 }
3589 } else {
3590 left_rec = &el->l_recs[index - 1];
3591 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3592 has_empty_extent = 1;
3593 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003594
Joel Becker4fe82c32009-02-13 02:16:08 -08003595 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003596 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003597 if (ret) {
3598 mlog_errno(ret);
3599 goto out;
3600 }
3601
3602 if (has_empty_extent && index == 1) {
3603 /*
3604 * The easy case - we can just plop the record right in.
3605 */
3606 *left_rec = *split_rec;
3607
3608 has_empty_extent = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003609 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003610 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003611
3612 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3613 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003614 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3615 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003616 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3617
3618 ocfs2_cleanup_merge(el, index);
3619
Joel Beckerec20cec2010-03-19 14:13:52 -07003620 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003621 if (left_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003622 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
Tao Ma677b9752008-01-30 14:21:05 +08003623
3624 /*
3625 * In the situation that the right_rec is empty and the extent
3626 * block is empty also, ocfs2_complete_edge_insert can't handle
3627 * it and we need to delete the right extent block.
3628 */
3629 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3630 le16_to_cpu(el->l_next_free_rec) == 1) {
3631
Joel Becker70f18c02009-02-13 02:09:31 -08003632 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003633 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003634 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003635 if (ret) {
3636 mlog_errno(ret);
3637 goto out;
3638 }
3639
3640 /* Now the rightmost extent block has been deleted.
3641 * So we use the new rightmost path.
3642 */
3643 ocfs2_mv_path(right_path, left_path);
3644 left_path = NULL;
3645 } else
Joel Becker4619c732009-02-12 19:02:36 -08003646 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003647 right_path, subtree_index);
3648 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003649out:
Tao Ma677b9752008-01-30 14:21:05 +08003650 if (left_path)
3651 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003652 return ret;
3653}
3654
Joel Beckerc495dd22009-02-13 02:19:11 -08003655static int ocfs2_try_to_merge_extent(handle_t *handle,
3656 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003657 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003658 int split_index,
3659 struct ocfs2_extent_rec *split_rec,
3660 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003661 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003662{
Tao Mao518d7262007-08-28 17:25:35 -07003663 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003664 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003665 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3666
3667 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3668
Tao Mao518d7262007-08-28 17:25:35 -07003669 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
3670 /*
3671 * The merge code will need to create an empty
3672 * extent to take the place of the newly
3673 * emptied slot. Remove any pre-existing empty
3674 * extents - having more than one in a leaf is
3675 * illegal.
3676 */
Joel Becker70f18c02009-02-13 02:09:31 -08003677 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003678 if (ret) {
3679 mlog_errno(ret);
3680 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003681 }
Tao Mao518d7262007-08-28 17:25:35 -07003682 split_index--;
3683 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003684 }
3685
3686 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3687 /*
3688 * Left-right contig implies this.
3689 */
3690 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003691
3692 /*
3693 * Since the leftright insert always covers the entire
3694 * extent, this call will delete the insert record
3695 * entirely, resulting in an empty extent record added to
3696 * the extent block.
3697 *
3698 * Since the adding of an empty extent shifts
3699 * everything back to the right, there's no need to
3700 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003701 *
3702 * When the split_index is zero, we need to merge it to the
3703 * prevoius extent block. It is more efficient and easier
3704 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003705 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003706 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003707 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003708 if (ret) {
3709 mlog_errno(ret);
3710 goto out;
3711 }
3712
3713 /*
3714 * We can only get this from logic error above.
3715 */
3716 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3717
Tao Ma677b9752008-01-30 14:21:05 +08003718 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003719 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003720 if (ret) {
3721 mlog_errno(ret);
3722 goto out;
3723 }
Tao Ma677b9752008-01-30 14:21:05 +08003724
Mark Fasheh328d5752007-06-18 10:48:04 -07003725 rec = &el->l_recs[split_index];
3726
3727 /*
3728 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003729 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003730 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003731 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3732 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003733
Mark Fasheh328d5752007-06-18 10:48:04 -07003734 if (ret) {
3735 mlog_errno(ret);
3736 goto out;
3737 }
3738
Joel Becker70f18c02009-02-13 02:09:31 -08003739 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003740 /*
3741 * Error from this last rotate is not critical, so
3742 * print but don't bubble it up.
3743 */
3744 if (ret)
3745 mlog_errno(ret);
3746 ret = 0;
3747 } else {
3748 /*
3749 * Merge a record to the left or right.
3750 *
3751 * 'contig_type' is relative to the existing record,
3752 * so for example, if we're "right contig", it's to
3753 * the record on the left (hence the left merge).
3754 */
3755 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003756 ret = ocfs2_merge_rec_left(path, handle, et,
3757 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003758 split_index);
3759 if (ret) {
3760 mlog_errno(ret);
3761 goto out;
3762 }
3763 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003764 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003765 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003766 split_index);
3767 if (ret) {
3768 mlog_errno(ret);
3769 goto out;
3770 }
3771 }
3772
3773 if (ctxt->c_split_covers_rec) {
3774 /*
3775 * The merge may have left an empty extent in
3776 * our leaf. Try to rotate it away.
3777 */
Joel Becker70f18c02009-02-13 02:09:31 -08003778 ret = ocfs2_rotate_tree_left(handle, et, path,
3779 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003780 if (ret)
3781 mlog_errno(ret);
3782 ret = 0;
3783 }
3784 }
3785
3786out:
3787 return ret;
3788}
3789
3790static void ocfs2_subtract_from_rec(struct super_block *sb,
3791 enum ocfs2_split_type split,
3792 struct ocfs2_extent_rec *rec,
3793 struct ocfs2_extent_rec *split_rec)
3794{
3795 u64 len_blocks;
3796
3797 len_blocks = ocfs2_clusters_to_blocks(sb,
3798 le16_to_cpu(split_rec->e_leaf_clusters));
3799
3800 if (split == SPLIT_LEFT) {
3801 /*
3802 * Region is on the left edge of the existing
3803 * record.
3804 */
3805 le32_add_cpu(&rec->e_cpos,
3806 le16_to_cpu(split_rec->e_leaf_clusters));
3807 le64_add_cpu(&rec->e_blkno, len_blocks);
3808 le16_add_cpu(&rec->e_leaf_clusters,
3809 -le16_to_cpu(split_rec->e_leaf_clusters));
3810 } else {
3811 /*
3812 * Region is on the right edge of the existing
3813 * record.
3814 */
3815 le16_add_cpu(&rec->e_leaf_clusters,
3816 -le16_to_cpu(split_rec->e_leaf_clusters));
3817 }
3818}
3819
Mark Fashehdcd05382007-01-16 11:32:23 -08003820/*
3821 * Do the final bits of extent record insertion at the target leaf
3822 * list. If this leaf is part of an allocation tree, it is assumed
3823 * that the tree above has been prepared.
3824 */
Joel Beckerd5628622009-02-13 02:54:36 -08003825static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3826 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003827 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003828 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003829{
3830 int i = insert->ins_contig_index;
3831 unsigned int range;
3832 struct ocfs2_extent_rec *rec;
3833
Mark Fashehe48edee2007-03-07 16:46:57 -08003834 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003835
Mark Fasheh328d5752007-06-18 10:48:04 -07003836 if (insert->ins_split != SPLIT_NONE) {
3837 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3838 BUG_ON(i == -1);
3839 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003840 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3841 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003842 insert_rec);
3843 goto rotate;
3844 }
3845
Mark Fashehdcd05382007-01-16 11:32:23 -08003846 /*
3847 * Contiguous insert - either left or right.
3848 */
3849 if (insert->ins_contig != CONTIG_NONE) {
3850 rec = &el->l_recs[i];
3851 if (insert->ins_contig == CONTIG_LEFT) {
3852 rec->e_blkno = insert_rec->e_blkno;
3853 rec->e_cpos = insert_rec->e_cpos;
3854 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003855 le16_add_cpu(&rec->e_leaf_clusters,
3856 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003857 return;
3858 }
3859
3860 /*
3861 * Handle insert into an empty leaf.
3862 */
3863 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3864 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3865 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3866 el->l_recs[0] = *insert_rec;
3867 el->l_next_free_rec = cpu_to_le16(1);
3868 return;
3869 }
3870
3871 /*
3872 * Appending insert.
3873 */
3874 if (insert->ins_appending == APPEND_TAIL) {
3875 i = le16_to_cpu(el->l_next_free_rec) - 1;
3876 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003877 range = le32_to_cpu(rec->e_cpos)
3878 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003879 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3880
3881 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3882 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003883 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003884 "rec.cpos %u, rec.clusters %u, "
3885 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003886 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003887 le16_to_cpu(el->l_tree_depth),
3888 le16_to_cpu(el->l_count),
3889 le16_to_cpu(el->l_next_free_rec),
3890 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003891 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003892 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003893 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003894 i++;
3895 el->l_recs[i] = *insert_rec;
3896 le16_add_cpu(&el->l_next_free_rec, 1);
3897 return;
3898 }
3899
Mark Fasheh328d5752007-06-18 10:48:04 -07003900rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003901 /*
3902 * Ok, we have to rotate.
3903 *
3904 * At this point, it is safe to assume that inserting into an
3905 * empty leaf and appending to a leaf have both been handled
3906 * above.
3907 *
3908 * This leaf needs to have space, either by the empty 1st
3909 * extent record, or by virtue of an l_next_rec < l_count.
3910 */
3911 ocfs2_rotate_leaf(el, insert_rec);
3912}
3913
Joel Beckerd401dc12009-02-13 02:24:10 -08003914static void ocfs2_adjust_rightmost_records(handle_t *handle,
3915 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003916 struct ocfs2_path *path,
3917 struct ocfs2_extent_rec *insert_rec)
3918{
3919 int ret, i, next_free;
3920 struct buffer_head *bh;
3921 struct ocfs2_extent_list *el;
3922 struct ocfs2_extent_rec *rec;
3923
3924 /*
3925 * Update everything except the leaf block.
3926 */
3927 for (i = 0; i < path->p_tree_depth; i++) {
3928 bh = path->p_node[i].bh;
3929 el = path->p_node[i].el;
3930
3931 next_free = le16_to_cpu(el->l_next_free_rec);
3932 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003933 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3934 "Owner %llu has a bad extent list",
3935 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003936 ret = -EIO;
3937 return;
3938 }
3939
3940 rec = &el->l_recs[next_free - 1];
3941
3942 rec->e_int_clusters = insert_rec->e_cpos;
3943 le32_add_cpu(&rec->e_int_clusters,
3944 le16_to_cpu(insert_rec->e_leaf_clusters));
3945 le32_add_cpu(&rec->e_int_clusters,
3946 -le32_to_cpu(rec->e_cpos));
3947
Joel Beckerec20cec2010-03-19 14:13:52 -07003948 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07003949 }
3950}
3951
Joel Beckerd401dc12009-02-13 02:24:10 -08003952static int ocfs2_append_rec_to_path(handle_t *handle,
3953 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003954 struct ocfs2_extent_rec *insert_rec,
3955 struct ocfs2_path *right_path,
3956 struct ocfs2_path **ret_left_path)
3957{
Mark Fasheh328d5752007-06-18 10:48:04 -07003958 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003959 struct ocfs2_extent_list *el;
3960 struct ocfs2_path *left_path = NULL;
3961
3962 *ret_left_path = NULL;
3963
3964 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003965 * This shouldn't happen for non-trees. The extent rec cluster
3966 * count manipulation below only works for interior nodes.
3967 */
3968 BUG_ON(right_path->p_tree_depth == 0);
3969
3970 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08003971 * If our appending insert is at the leftmost edge of a leaf,
3972 * then we might need to update the rightmost records of the
3973 * neighboring path.
3974 */
3975 el = path_leaf_el(right_path);
3976 next_free = le16_to_cpu(el->l_next_free_rec);
3977 if (next_free == 0 ||
3978 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
3979 u32 left_cpos;
3980
Joel Beckerd401dc12009-02-13 02:24:10 -08003981 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3982 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003983 if (ret) {
3984 mlog_errno(ret);
3985 goto out;
3986 }
3987
Tao Maa09d09b2011-02-22 08:21:10 +08003988 trace_ocfs2_append_rec_to_path(
3989 (unsigned long long)
3990 ocfs2_metadata_cache_owner(et->et_ci),
3991 le32_to_cpu(insert_rec->e_cpos),
3992 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003993
3994 /*
3995 * No need to worry if the append is already in the
3996 * leftmost leaf.
3997 */
3998 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07003999 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004000 if (!left_path) {
4001 ret = -ENOMEM;
4002 mlog_errno(ret);
4003 goto out;
4004 }
4005
Joel Beckerd401dc12009-02-13 02:24:10 -08004006 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004007 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004008 if (ret) {
4009 mlog_errno(ret);
4010 goto out;
4011 }
4012
4013 /*
4014 * ocfs2_insert_path() will pass the left_path to the
4015 * journal for us.
4016 */
4017 }
4018 }
4019
Joel Beckerd401dc12009-02-13 02:24:10 -08004020 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004021 if (ret) {
4022 mlog_errno(ret);
4023 goto out;
4024 }
4025
Joel Beckerd401dc12009-02-13 02:24:10 -08004026 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004027
4028 *ret_left_path = left_path;
4029 ret = 0;
4030out:
4031 if (ret != 0)
4032 ocfs2_free_path(left_path);
4033
4034 return ret;
4035}
4036
Joel Beckerc38e52b2009-02-13 02:56:23 -08004037static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004038 struct ocfs2_path *left_path,
4039 struct ocfs2_path *right_path,
4040 struct ocfs2_extent_rec *split_rec,
4041 enum ocfs2_split_type split)
4042{
4043 int index;
4044 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4045 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4046 struct ocfs2_extent_rec *rec, *tmprec;
4047
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004048 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004049 if (left_path)
4050 left_el = path_leaf_el(left_path);
4051
4052 el = right_el;
4053 insert_el = right_el;
4054 index = ocfs2_search_extent_list(el, cpos);
4055 if (index != -1) {
4056 if (index == 0 && left_path) {
4057 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4058
4059 /*
4060 * This typically means that the record
4061 * started in the left path but moved to the
4062 * right as a result of rotation. We either
4063 * move the existing record to the left, or we
4064 * do the later insert there.
4065 *
4066 * In this case, the left path should always
4067 * exist as the rotate code will have passed
4068 * it back for a post-insert update.
4069 */
4070
4071 if (split == SPLIT_LEFT) {
4072 /*
4073 * It's a left split. Since we know
4074 * that the rotate code gave us an
4075 * empty extent in the left path, we
4076 * can just do the insert there.
4077 */
4078 insert_el = left_el;
4079 } else {
4080 /*
4081 * Right split - we have to move the
4082 * existing record over to the left
4083 * leaf. The insert will be into the
4084 * newly created empty extent in the
4085 * right leaf.
4086 */
4087 tmprec = &right_el->l_recs[index];
4088 ocfs2_rotate_leaf(left_el, tmprec);
4089 el = left_el;
4090
4091 memset(tmprec, 0, sizeof(*tmprec));
4092 index = ocfs2_search_extent_list(left_el, cpos);
4093 BUG_ON(index == -1);
4094 }
4095 }
4096 } else {
4097 BUG_ON(!left_path);
4098 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4099 /*
4100 * Left path is easy - we can just allow the insert to
4101 * happen.
4102 */
4103 el = left_el;
4104 insert_el = left_el;
4105 index = ocfs2_search_extent_list(el, cpos);
4106 BUG_ON(index == -1);
4107 }
4108
4109 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004110 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4111 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004112 ocfs2_rotate_leaf(insert_el, split_rec);
4113}
4114
Mark Fashehdcd05382007-01-16 11:32:23 -08004115/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004116 * This function only does inserts on an allocation b-tree. For tree
4117 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004118 *
4119 * right_path is the path we want to do the actual insert
4120 * in. left_path should only be passed in if we need to update that
4121 * portion of the tree after an edge insert.
4122 */
Joel Becker3505bec2009-02-13 02:57:58 -08004123static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004124 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004125 struct ocfs2_path *left_path,
4126 struct ocfs2_path *right_path,
4127 struct ocfs2_extent_rec *insert_rec,
4128 struct ocfs2_insert_type *insert)
4129{
4130 int ret, subtree_index;
4131 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004132
Mark Fashehdcd05382007-01-16 11:32:23 -08004133 if (left_path) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004134 /*
4135 * There's a chance that left_path got passed back to
4136 * us without being accounted for in the
4137 * journal. Extend our transaction here to be sure we
4138 * can change those blocks.
4139 */
Tao Mac901fb02010-04-26 14:34:57 +08004140 ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08004141 if (ret < 0) {
4142 mlog_errno(ret);
4143 goto out;
4144 }
4145
Joel Becker7dc02802009-02-12 19:20:13 -08004146 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004147 if (ret < 0) {
4148 mlog_errno(ret);
4149 goto out;
4150 }
4151 }
4152
Mark Fashehe8aed342007-12-03 16:43:01 -08004153 /*
4154 * Pass both paths to the journal. The majority of inserts
4155 * will be touching all components anyway.
4156 */
Joel Becker7dc02802009-02-12 19:20:13 -08004157 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004158 if (ret < 0) {
4159 mlog_errno(ret);
4160 goto out;
4161 }
4162
Mark Fasheh328d5752007-06-18 10:48:04 -07004163 if (insert->ins_split != SPLIT_NONE) {
4164 /*
4165 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004166 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004167 * function sort it all out.
4168 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004169 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004170 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004171
4172 /*
4173 * Split might have modified either leaf and we don't
4174 * have a guarantee that the later edge insert will
4175 * dirty this for us.
4176 */
4177 if (left_path)
Joel Beckerec20cec2010-03-19 14:13:52 -07004178 ocfs2_journal_dirty(handle,
4179 path_leaf_bh(left_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07004180 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004181 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4182 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004183
Joel Beckerec20cec2010-03-19 14:13:52 -07004184 ocfs2_journal_dirty(handle, leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004185
4186 if (left_path) {
4187 /*
4188 * The rotate code has indicated that we need to fix
4189 * up portions of the tree after the insert.
4190 *
4191 * XXX: Should we extend the transaction here?
4192 */
Joel Becker7dc02802009-02-12 19:20:13 -08004193 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004194 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004195 ocfs2_complete_edge_insert(handle, left_path, right_path,
4196 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004197 }
4198
4199 ret = 0;
4200out:
4201 return ret;
4202}
4203
Joel Becker3505bec2009-02-13 02:57:58 -08004204static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004205 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004206 struct ocfs2_extent_rec *insert_rec,
4207 struct ocfs2_insert_type *type)
4208{
4209 int ret, rotate = 0;
4210 u32 cpos;
4211 struct ocfs2_path *right_path = NULL;
4212 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004213 struct ocfs2_extent_list *el;
4214
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004215 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004216
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004217 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004218 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004219 if (ret) {
4220 mlog_errno(ret);
4221 goto out;
4222 }
4223
4224 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004225 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004226 goto out_update_clusters;
4227 }
4228
Joel Beckerffdd7a52008-10-17 22:32:01 -07004229 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004230 if (!right_path) {
4231 ret = -ENOMEM;
4232 mlog_errno(ret);
4233 goto out;
4234 }
4235
4236 /*
4237 * Determine the path to start with. Rotations need the
4238 * rightmost path, everything else can go directly to the
4239 * target leaf.
4240 */
4241 cpos = le32_to_cpu(insert_rec->e_cpos);
4242 if (type->ins_appending == APPEND_NONE &&
4243 type->ins_contig == CONTIG_NONE) {
4244 rotate = 1;
4245 cpos = UINT_MAX;
4246 }
4247
Joel Beckerfacdb772009-02-12 18:08:48 -08004248 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004249 if (ret) {
4250 mlog_errno(ret);
4251 goto out;
4252 }
4253
4254 /*
4255 * Rotations and appends need special treatment - they modify
4256 * parts of the tree's above them.
4257 *
4258 * Both might pass back a path immediate to the left of the
4259 * one being inserted to. This will be cause
4260 * ocfs2_insert_path() to modify the rightmost records of
4261 * left_path to account for an edge insert.
4262 *
4263 * XXX: When modifying this code, keep in mind that an insert
4264 * can wind up skipping both of these two special cases...
4265 */
4266 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004267 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004268 le32_to_cpu(insert_rec->e_cpos),
4269 right_path, &left_path);
4270 if (ret) {
4271 mlog_errno(ret);
4272 goto out;
4273 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004274
4275 /*
4276 * ocfs2_rotate_tree_right() might have extended the
4277 * transaction without re-journaling our tree root.
4278 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004279 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004280 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004281 if (ret) {
4282 mlog_errno(ret);
4283 goto out;
4284 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004285 } else if (type->ins_appending == APPEND_TAIL
4286 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004287 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004288 right_path, &left_path);
4289 if (ret) {
4290 mlog_errno(ret);
4291 goto out;
4292 }
4293 }
4294
Joel Becker3505bec2009-02-13 02:57:58 -08004295 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004296 insert_rec, type);
4297 if (ret) {
4298 mlog_errno(ret);
4299 goto out;
4300 }
4301
4302out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004303 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004304 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004305 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004306
Joel Beckerec20cec2010-03-19 14:13:52 -07004307 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004308
4309out:
4310 ocfs2_free_path(left_path);
4311 ocfs2_free_path(right_path);
4312
4313 return ret;
4314}
4315
Mark Fasheh328d5752007-06-18 10:48:04 -07004316static enum ocfs2_contig_type
Joel Beckera2970292009-02-13 03:09:54 -08004317ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
4318 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004319 struct ocfs2_extent_list *el, int index,
4320 struct ocfs2_extent_rec *split_rec)
4321{
Tao Maad5a4d72008-01-30 14:21:32 +08004322 int status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004323 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004324 u32 left_cpos, right_cpos;
4325 struct ocfs2_extent_rec *rec = NULL;
4326 struct ocfs2_extent_list *new_el;
4327 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4328 struct buffer_head *bh;
4329 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004330 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004331
4332 if (index > 0) {
4333 rec = &el->l_recs[index - 1];
4334 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004335 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004336 if (status)
4337 goto out;
4338
4339 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004340 left_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004341 if (!left_path)
4342 goto out;
4343
Joel Beckera2970292009-02-13 03:09:54 -08004344 status = ocfs2_find_path(et->et_ci, left_path,
4345 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004346 if (status)
4347 goto out;
4348
4349 new_el = path_leaf_el(left_path);
4350
4351 if (le16_to_cpu(new_el->l_next_free_rec) !=
4352 le16_to_cpu(new_el->l_count)) {
4353 bh = path_leaf_bh(left_path);
4354 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004355 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004356 "Extent block #%llu has an "
4357 "invalid l_next_free_rec of "
4358 "%d. It should have "
4359 "matched the l_count of %d",
4360 (unsigned long long)le64_to_cpu(eb->h_blkno),
4361 le16_to_cpu(new_el->l_next_free_rec),
4362 le16_to_cpu(new_el->l_count));
4363 status = -EINVAL;
Tao Maad5a4d72008-01-30 14:21:32 +08004364 goto out;
4365 }
4366 rec = &new_el->l_recs[
4367 le16_to_cpu(new_el->l_next_free_rec) - 1];
4368 }
4369 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004370
4371 /*
4372 * We're careful to check for an empty extent record here -
4373 * the merge code will know what to do if it sees one.
4374 */
Tao Maad5a4d72008-01-30 14:21:32 +08004375 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004376 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4377 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4378 ret = CONTIG_RIGHT;
4379 } else {
Tao Ma853a3a12009-08-18 11:22:18 +08004380 ret = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004381 }
4382 }
4383
Tao Maad5a4d72008-01-30 14:21:32 +08004384 rec = NULL;
4385 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4386 rec = &el->l_recs[index + 1];
4387 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4388 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004389 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004390 if (status)
4391 goto out;
4392
4393 if (right_cpos == 0)
4394 goto out;
4395
Joel Beckerffdd7a52008-10-17 22:32:01 -07004396 right_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004397 if (!right_path)
4398 goto out;
4399
Joel Beckera2970292009-02-13 03:09:54 -08004400 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004401 if (status)
4402 goto out;
4403
4404 new_el = path_leaf_el(right_path);
4405 rec = &new_el->l_recs[0];
4406 if (ocfs2_is_empty_extent(rec)) {
4407 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4408 bh = path_leaf_bh(right_path);
4409 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004410 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004411 "Extent block #%llu has an "
4412 "invalid l_next_free_rec of %d",
4413 (unsigned long long)le64_to_cpu(eb->h_blkno),
4414 le16_to_cpu(new_el->l_next_free_rec));
4415 status = -EINVAL;
Tao Maad5a4d72008-01-30 14:21:32 +08004416 goto out;
4417 }
4418 rec = &new_el->l_recs[1];
4419 }
4420 }
4421
4422 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004423 enum ocfs2_contig_type contig_type;
4424
Tao Ma853a3a12009-08-18 11:22:18 +08004425 contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004426
4427 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4428 ret = CONTIG_LEFTRIGHT;
4429 else if (ret == CONTIG_NONE)
4430 ret = contig_type;
4431 }
4432
Tao Maad5a4d72008-01-30 14:21:32 +08004433out:
4434 if (left_path)
4435 ocfs2_free_path(left_path);
4436 if (right_path)
4437 ocfs2_free_path(right_path);
4438
Mark Fasheh328d5752007-06-18 10:48:04 -07004439 return ret;
4440}
4441
Joel Becker1ef61b32009-02-13 03:12:33 -08004442static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004443 struct ocfs2_insert_type *insert,
4444 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004445 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004446{
4447 int i;
4448 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4449
Mark Fashehe48edee2007-03-07 16:46:57 -08004450 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4451
Mark Fashehdcd05382007-01-16 11:32:23 -08004452 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Tao Ma853a3a12009-08-18 11:22:18 +08004453 contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4454 insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004455 if (contig_type != CONTIG_NONE) {
4456 insert->ins_contig_index = i;
4457 break;
4458 }
4459 }
4460 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004461
4462 if (insert->ins_contig != CONTIG_NONE) {
4463 struct ocfs2_extent_rec *rec =
4464 &el->l_recs[insert->ins_contig_index];
4465 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4466 le16_to_cpu(insert_rec->e_leaf_clusters);
4467
4468 /*
4469 * Caller might want us to limit the size of extents, don't
4470 * calculate contiguousness if we might exceed that limit.
4471 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004472 if (et->et_max_leaf_clusters &&
4473 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004474 insert->ins_contig = CONTIG_NONE;
4475 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004476}
4477
4478/*
4479 * This should only be called against the righmost leaf extent list.
4480 *
4481 * ocfs2_figure_appending_type() will figure out whether we'll have to
4482 * insert at the tail of the rightmost leaf.
4483 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004484 * This should also work against the root extent list for tree's with 0
4485 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004486 * then the logic here makes sense.
4487 */
4488static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4489 struct ocfs2_extent_list *el,
4490 struct ocfs2_extent_rec *insert_rec)
4491{
4492 int i;
4493 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4494 struct ocfs2_extent_rec *rec;
4495
4496 insert->ins_appending = APPEND_NONE;
4497
Mark Fashehe48edee2007-03-07 16:46:57 -08004498 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004499
4500 if (!el->l_next_free_rec)
4501 goto set_tail_append;
4502
4503 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4504 /* Were all records empty? */
4505 if (le16_to_cpu(el->l_next_free_rec) == 1)
4506 goto set_tail_append;
4507 }
4508
4509 i = le16_to_cpu(el->l_next_free_rec) - 1;
4510 rec = &el->l_recs[i];
4511
Mark Fashehe48edee2007-03-07 16:46:57 -08004512 if (cpos >=
4513 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004514 goto set_tail_append;
4515
4516 return;
4517
4518set_tail_append:
4519 insert->ins_appending = APPEND_TAIL;
4520}
4521
4522/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004523 * Helper function called at the beginning of an insert.
Mark Fashehdcd05382007-01-16 11:32:23 -08004524 *
4525 * This computes a few things that are commonly used in the process of
4526 * inserting into the btree:
4527 * - Whether the new extent is contiguous with an existing one.
4528 * - The current tree depth.
4529 * - Whether the insert is an appending one.
4530 * - The total # of free records in the tree.
4531 *
4532 * All of the information is stored on the ocfs2_insert_type
4533 * structure.
4534 */
Joel Becker627961b2009-02-13 03:14:38 -08004535static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004536 struct buffer_head **last_eb_bh,
4537 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004538 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004539 struct ocfs2_insert_type *insert)
4540{
4541 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004542 struct ocfs2_extent_block *eb;
4543 struct ocfs2_extent_list *el;
4544 struct ocfs2_path *path = NULL;
4545 struct buffer_head *bh = NULL;
4546
Mark Fasheh328d5752007-06-18 10:48:04 -07004547 insert->ins_split = SPLIT_NONE;
4548
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004549 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004550 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4551
4552 if (el->l_tree_depth) {
4553 /*
4554 * If we have tree depth, we read in the
4555 * rightmost extent block ahead of time as
4556 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4557 * may want it later.
4558 */
Joel Becker3d03a302009-02-12 17:49:26 -08004559 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004560 ocfs2_et_get_last_eb_blk(et),
4561 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004562 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08004563 mlog_errno(ret);
Mark Fashehdcd05382007-01-16 11:32:23 -08004564 goto out;
4565 }
4566 eb = (struct ocfs2_extent_block *) bh->b_data;
4567 el = &eb->h_list;
4568 }
4569
4570 /*
4571 * Unless we have a contiguous insert, we'll need to know if
4572 * there is room left in our allocation tree for another
4573 * extent record.
4574 *
4575 * XXX: This test is simplistic, we can search for empty
4576 * extent records too.
4577 */
Tao Maoc77534f2007-08-28 17:22:33 -07004578 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004579 le16_to_cpu(el->l_next_free_rec);
4580
4581 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004582 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004583 ocfs2_figure_appending_type(insert, el, insert_rec);
4584 return 0;
4585 }
4586
Joel Beckerffdd7a52008-10-17 22:32:01 -07004587 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004588 if (!path) {
4589 ret = -ENOMEM;
4590 mlog_errno(ret);
4591 goto out;
4592 }
4593
4594 /*
4595 * In the case that we're inserting past what the tree
4596 * currently accounts for, ocfs2_find_path() will return for
4597 * us the rightmost tree path. This is accounted for below in
4598 * the appending code.
4599 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004600 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004601 if (ret) {
4602 mlog_errno(ret);
4603 goto out;
4604 }
4605
4606 el = path_leaf_el(path);
4607
4608 /*
4609 * Now that we have the path, there's two things we want to determine:
4610 * 1) Contiguousness (also set contig_index if this is so)
4611 *
4612 * 2) Are we doing an append? We can trivially break this up
4613 * into two types of appends: simple record append, or a
4614 * rotate inside the tail leaf.
4615 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004616 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004617
4618 /*
4619 * The insert code isn't quite ready to deal with all cases of
4620 * left contiguousness. Specifically, if it's an insert into
4621 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004622 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004623 * left. For now, just catch that case and fool the layers
4624 * above us. This works just fine for tree_depth == 0, which
4625 * is why we allow that above.
4626 */
4627 if (insert->ins_contig == CONTIG_LEFT &&
4628 insert->ins_contig_index == 0)
4629 insert->ins_contig = CONTIG_NONE;
4630
4631 /*
4632 * Ok, so we can simply compare against last_eb to figure out
4633 * whether the path doesn't exist. This will only happen in
4634 * the case that we're doing a tail append, so maybe we can
4635 * take advantage of that information somehow.
4636 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004637 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004638 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004639 /*
4640 * Ok, ocfs2_find_path() returned us the rightmost
4641 * tree path. This might be an appending insert. There are
4642 * two cases:
4643 * 1) We're doing a true append at the tail:
4644 * -This might even be off the end of the leaf
4645 * 2) We're "appending" by rotating in the tail
4646 */
4647 ocfs2_figure_appending_type(insert, el, insert_rec);
4648 }
4649
4650out:
4651 ocfs2_free_path(path);
4652
4653 if (ret == 0)
4654 *last_eb_bh = bh;
4655 else
4656 brelse(bh);
4657 return ret;
4658}
4659
4660/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004661 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004662 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004663 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004664 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004665int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004666 struct ocfs2_extent_tree *et,
4667 u32 cpos,
4668 u64 start_blk,
4669 u32 new_clusters,
4670 u8 flags,
4671 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004672{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004673 int status;
Tao Maoc77534f2007-08-28 17:22:33 -07004674 int uninitialized_var(free_records);
Mark Fashehccd979b2005-12-15 14:31:24 -08004675 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004676 struct ocfs2_insert_type insert = {0, };
4677 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004678
Tao Maa09d09b2011-02-22 08:21:10 +08004679 trace_ocfs2_insert_extent_start(
4680 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4681 cpos, new_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08004682
Mark Fashehe48edee2007-03-07 16:46:57 -08004683 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004684 rec.e_cpos = cpu_to_le32(cpos);
4685 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004686 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004687 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004688 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004689 if (status) {
4690 mlog_errno(status);
4691 goto bail;
4692 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004693
Joel Becker627961b2009-02-13 03:14:38 -08004694 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004695 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004696 if (status < 0) {
4697 mlog_errno(status);
4698 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004699 }
4700
Tao Maa09d09b2011-02-22 08:21:10 +08004701 trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4702 insert.ins_contig_index, free_records,
4703 insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004704
Tao Maoc77534f2007-08-28 17:22:33 -07004705 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004706 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004707 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004708 meta_ac);
4709 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004710 mlog_errno(status);
4711 goto bail;
4712 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004713 }
4714
Mark Fashehdcd05382007-01-16 11:32:23 -08004715 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004716 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004717 if (status < 0)
4718 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004719 else
4720 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004721
4722bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004723 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004724
Tao Maf56654c2008-08-18 17:38:48 +08004725 return status;
4726}
4727
Tao Ma0eb8d472008-08-18 17:38:45 +08004728/*
4729 * Allcate and add clusters into the extent b-tree.
4730 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004731 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d472008-08-18 17:38:45 +08004732 * it is not limited to the file storage. Any extent tree can use this
4733 * function if it implements the proper ocfs2_extent_tree.
4734 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004735int ocfs2_add_clusters_in_btree(handle_t *handle,
4736 struct ocfs2_extent_tree *et,
Tao Ma0eb8d472008-08-18 17:38:45 +08004737 u32 *logical_offset,
4738 u32 clusters_to_add,
4739 int mark_unwritten,
Tao Ma0eb8d472008-08-18 17:38:45 +08004740 struct ocfs2_alloc_context *data_ac,
4741 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004742 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d472008-08-18 17:38:45 +08004743{
Tao Maa09d09b2011-02-22 08:21:10 +08004744 int status = 0, err = 0;
Tao Ma0eb8d472008-08-18 17:38:45 +08004745 int free_extents;
4746 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4747 u32 bit_off, num_bits;
4748 u64 block;
4749 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004750 struct ocfs2_super *osb =
4751 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d472008-08-18 17:38:45 +08004752
4753 BUG_ON(!clusters_to_add);
4754
4755 if (mark_unwritten)
4756 flags = OCFS2_EXT_UNWRITTEN;
4757
Joel Becker3d03a302009-02-12 17:49:26 -08004758 free_extents = ocfs2_num_free_extents(osb, et);
Tao Ma0eb8d472008-08-18 17:38:45 +08004759 if (free_extents < 0) {
4760 status = free_extents;
4761 mlog_errno(status);
4762 goto leave;
4763 }
4764
4765 /* there are two cases which could cause us to EAGAIN in the
4766 * we-need-more-metadata case:
4767 * 1) we haven't reserved *any*
4768 * 2) we are so fragmented, we've needed to add metadata too
4769 * many times. */
4770 if (!free_extents && !meta_ac) {
Tao Maa09d09b2011-02-22 08:21:10 +08004771 err = -1;
Tao Ma0eb8d472008-08-18 17:38:45 +08004772 status = -EAGAIN;
4773 reason = RESTART_META;
4774 goto leave;
4775 } else if ((!free_extents)
4776 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004777 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Maa09d09b2011-02-22 08:21:10 +08004778 err = -2;
Tao Ma0eb8d472008-08-18 17:38:45 +08004779 status = -EAGAIN;
4780 reason = RESTART_META;
4781 goto leave;
4782 }
4783
Joel Becker1ed9b772010-05-06 13:59:06 +08004784 status = __ocfs2_claim_clusters(handle, data_ac, 1,
Tao Ma0eb8d472008-08-18 17:38:45 +08004785 clusters_to_add, &bit_off, &num_bits);
4786 if (status < 0) {
4787 if (status != -ENOSPC)
4788 mlog_errno(status);
4789 goto leave;
4790 }
4791
4792 BUG_ON(num_bits > clusters_to_add);
4793
Joel Becker13723d02008-10-17 19:25:01 -07004794 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004795 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004796 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d472008-08-18 17:38:45 +08004797 if (status < 0) {
4798 mlog_errno(status);
4799 goto leave;
4800 }
4801
4802 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Maa09d09b2011-02-22 08:21:10 +08004803 trace_ocfs2_add_clusters_in_btree(
4804 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4805 bit_off, num_bits);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004806 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004807 num_bits, flags, meta_ac);
Tao Ma0eb8d472008-08-18 17:38:45 +08004808 if (status < 0) {
4809 mlog_errno(status);
4810 goto leave;
4811 }
4812
Joel Beckerec20cec2010-03-19 14:13:52 -07004813 ocfs2_journal_dirty(handle, et->et_root_bh);
Tao Ma0eb8d472008-08-18 17:38:45 +08004814
4815 clusters_to_add -= num_bits;
4816 *logical_offset += num_bits;
4817
4818 if (clusters_to_add) {
Tao Maa09d09b2011-02-22 08:21:10 +08004819 err = clusters_to_add;
Tao Ma0eb8d472008-08-18 17:38:45 +08004820 status = -EAGAIN;
4821 reason = RESTART_TRANS;
4822 }
4823
4824leave:
Tao Ma0eb8d472008-08-18 17:38:45 +08004825 if (reason_ret)
4826 *reason_ret = reason;
Tao Maa09d09b2011-02-22 08:21:10 +08004827 trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
Tao Ma0eb8d472008-08-18 17:38:45 +08004828 return status;
4829}
4830
Mark Fasheh328d5752007-06-18 10:48:04 -07004831static void ocfs2_make_right_split_rec(struct super_block *sb,
4832 struct ocfs2_extent_rec *split_rec,
4833 u32 cpos,
4834 struct ocfs2_extent_rec *rec)
4835{
4836 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4837 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4838
4839 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4840
4841 split_rec->e_cpos = cpu_to_le32(cpos);
4842 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4843
4844 split_rec->e_blkno = rec->e_blkno;
4845 le64_add_cpu(&split_rec->e_blkno,
4846 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4847
4848 split_rec->e_flags = rec->e_flags;
4849}
4850
Joel Beckerd2311292009-02-13 03:43:22 -08004851static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004852 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004853 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004854 struct buffer_head **last_eb_bh,
4855 int split_index,
4856 struct ocfs2_extent_rec *orig_split_rec,
4857 struct ocfs2_alloc_context *meta_ac)
4858{
4859 int ret = 0, depth;
4860 unsigned int insert_range, rec_range, do_leftright = 0;
4861 struct ocfs2_extent_rec tmprec;
4862 struct ocfs2_extent_list *rightmost_el;
4863 struct ocfs2_extent_rec rec;
4864 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4865 struct ocfs2_insert_type insert;
4866 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004867
4868leftright:
4869 /*
4870 * Store a copy of the record on the stack - it might move
4871 * around as the tree is manipulated below.
4872 */
4873 rec = path_leaf_el(path)->l_recs[split_index];
4874
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004875 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004876
4877 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4878 if (depth) {
4879 BUG_ON(!(*last_eb_bh));
4880 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4881 rightmost_el = &eb->h_list;
4882 }
4883
4884 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4885 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004886 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004887 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004888 if (ret) {
4889 mlog_errno(ret);
4890 goto out;
4891 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004892 }
4893
4894 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4895 insert.ins_appending = APPEND_NONE;
4896 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004897 insert.ins_tree_depth = depth;
4898
4899 insert_range = le32_to_cpu(split_rec.e_cpos) +
4900 le16_to_cpu(split_rec.e_leaf_clusters);
4901 rec_range = le32_to_cpu(rec.e_cpos) +
4902 le16_to_cpu(rec.e_leaf_clusters);
4903
4904 if (split_rec.e_cpos == rec.e_cpos) {
4905 insert.ins_split = SPLIT_LEFT;
4906 } else if (insert_range == rec_range) {
4907 insert.ins_split = SPLIT_RIGHT;
4908 } else {
4909 /*
4910 * Left/right split. We fake this as a right split
4911 * first and then make a second pass as a left split.
4912 */
4913 insert.ins_split = SPLIT_RIGHT;
4914
Joel Beckerd2311292009-02-13 03:43:22 -08004915 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4916 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004917
4918 split_rec = tmprec;
4919
4920 BUG_ON(do_leftright);
4921 do_leftright = 1;
4922 }
4923
Joel Becker3505bec2009-02-13 02:57:58 -08004924 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004925 if (ret) {
4926 mlog_errno(ret);
4927 goto out;
4928 }
4929
4930 if (do_leftright == 1) {
4931 u32 cpos;
4932 struct ocfs2_extent_list *el;
4933
4934 do_leftright++;
4935 split_rec = *orig_split_rec;
4936
4937 ocfs2_reinit_path(path, 1);
4938
4939 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004940 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004941 if (ret) {
4942 mlog_errno(ret);
4943 goto out;
4944 }
4945
4946 el = path_leaf_el(path);
4947 split_index = ocfs2_search_extent_list(el, cpos);
4948 goto leftright;
4949 }
4950out:
4951
4952 return ret;
4953}
4954
Joel Beckerf3868d02009-02-17 19:46:04 -08004955static int ocfs2_replace_extent_rec(handle_t *handle,
4956 struct ocfs2_extent_tree *et,
Tao Ma47be12e2009-01-09 07:32:48 +08004957 struct ocfs2_path *path,
4958 struct ocfs2_extent_list *el,
4959 int split_index,
4960 struct ocfs2_extent_rec *split_rec)
4961{
4962 int ret;
4963
Joel Beckerf3868d02009-02-17 19:46:04 -08004964 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Tao Ma47be12e2009-01-09 07:32:48 +08004965 path_num_items(path) - 1);
4966 if (ret) {
4967 mlog_errno(ret);
4968 goto out;
4969 }
4970
4971 el->l_recs[split_index] = *split_rec;
4972
4973 ocfs2_journal_dirty(handle, path_leaf_bh(path));
4974out:
4975 return ret;
4976}
4977
Mark Fasheh328d5752007-06-18 10:48:04 -07004978/*
Tao Ma555936b2009-08-18 11:22:21 +08004979 * Split part or all of the extent record at split_index in the leaf
4980 * pointed to by path. Merge with the contiguous extent record if needed.
Mark Fasheh328d5752007-06-18 10:48:04 -07004981 *
4982 * Care is taken to handle contiguousness so as to not grow the tree.
4983 *
4984 * meta_ac is not strictly necessary - we only truly need it if growth
4985 * of the tree is required. All other cases will degrade into a less
4986 * optimal tree layout.
4987 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004988 * last_eb_bh should be the rightmost leaf block for any extent
4989 * btree. Since a split may grow the tree or a merge might shrink it,
4990 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07004991 *
4992 * This code is optimized for readability - several passes might be
4993 * made over certain portions of the tree. All of those blocks will
4994 * have been brought into cache (and pinned via the journal), so the
4995 * extra overhead is not expressed in terms of disk reads.
4996 */
Tao Mae2e9f602009-08-18 11:22:34 +08004997int ocfs2_split_extent(handle_t *handle,
4998 struct ocfs2_extent_tree *et,
4999 struct ocfs2_path *path,
5000 int split_index,
5001 struct ocfs2_extent_rec *split_rec,
5002 struct ocfs2_alloc_context *meta_ac,
5003 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005004{
5005 int ret = 0;
5006 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005007 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005008 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5009 struct ocfs2_merge_ctxt ctxt;
5010 struct ocfs2_extent_list *rightmost_el;
5011
Mark Fasheh328d5752007-06-18 10:48:04 -07005012 if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5013 ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5014 (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5015 ret = -EIO;
5016 mlog_errno(ret);
5017 goto out;
5018 }
5019
Joel Beckera2970292009-02-13 03:09:54 -08005020 ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
Mark Fasheh328d5752007-06-18 10:48:04 -07005021 split_index,
5022 split_rec);
5023
5024 /*
5025 * The core merge / split code wants to know how much room is
Joel Beckera1cf0762009-02-13 03:45:49 -08005026 * left in this allocation tree, so we pass the
Mark Fasheh328d5752007-06-18 10:48:04 -07005027 * rightmost extent list.
5028 */
5029 if (path->p_tree_depth) {
5030 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07005031
Joel Becker3d03a302009-02-12 17:49:26 -08005032 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005033 ocfs2_et_get_last_eb_blk(et),
5034 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005035 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08005036 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07005037 goto out;
5038 }
5039
5040 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fasheh328d5752007-06-18 10:48:04 -07005041 rightmost_el = &eb->h_list;
5042 } else
5043 rightmost_el = path_root_el(path);
5044
Mark Fasheh328d5752007-06-18 10:48:04 -07005045 if (rec->e_cpos == split_rec->e_cpos &&
5046 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5047 ctxt.c_split_covers_rec = 1;
5048 else
5049 ctxt.c_split_covers_rec = 0;
5050
5051 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5052
Tao Maa09d09b2011-02-22 08:21:10 +08005053 trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5054 ctxt.c_has_empty_extent,
5055 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005056
5057 if (ctxt.c_contig_type == CONTIG_NONE) {
5058 if (ctxt.c_split_covers_rec)
Joel Beckerf3868d02009-02-17 19:46:04 -08005059 ret = ocfs2_replace_extent_rec(handle, et, path, el,
Tao Ma47be12e2009-01-09 07:32:48 +08005060 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005061 else
Joel Beckerd2311292009-02-13 03:43:22 -08005062 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005063 &last_eb_bh, split_index,
5064 split_rec, meta_ac);
5065 if (ret)
5066 mlog_errno(ret);
5067 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005068 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005069 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005070 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005071 if (ret)
5072 mlog_errno(ret);
5073 }
5074
Mark Fasheh328d5752007-06-18 10:48:04 -07005075out:
5076 brelse(last_eb_bh);
5077 return ret;
5078}
5079
5080/*
Tao Ma555936b2009-08-18 11:22:21 +08005081 * Change the flags of the already-existing extent at cpos for len clusters.
5082 *
5083 * new_flags: the flags we want to set.
5084 * clear_flags: the flags we want to clear.
5085 * phys: the new physical offset we want this new extent starts from.
Mark Fasheh328d5752007-06-18 10:48:04 -07005086 *
5087 * If the existing extent is larger than the request, initiate a
5088 * split. An attempt will be made at merging with adjacent extents.
5089 *
5090 * The caller is responsible for passing down meta_ac if we'll need it.
5091 */
Tao Ma1aa75fe2009-08-18 11:28:39 +08005092int ocfs2_change_extent_flag(handle_t *handle,
5093 struct ocfs2_extent_tree *et,
5094 u32 cpos, u32 len, u32 phys,
5095 struct ocfs2_alloc_context *meta_ac,
5096 struct ocfs2_cached_dealloc_ctxt *dealloc,
5097 int new_flags, int clear_flags)
Mark Fasheh328d5752007-06-18 10:48:04 -07005098{
5099 int ret, index;
Tao Ma555936b2009-08-18 11:22:21 +08005100 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5101 u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
Mark Fasheh328d5752007-06-18 10:48:04 -07005102 struct ocfs2_extent_rec split_rec;
5103 struct ocfs2_path *left_path = NULL;
5104 struct ocfs2_extent_list *el;
Tao Ma555936b2009-08-18 11:22:21 +08005105 struct ocfs2_extent_rec *rec;
Mark Fasheh328d5752007-06-18 10:48:04 -07005106
Joel Beckerffdd7a52008-10-17 22:32:01 -07005107 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005108 if (!left_path) {
5109 ret = -ENOMEM;
5110 mlog_errno(ret);
5111 goto out;
5112 }
5113
Joel Beckerfacdb772009-02-12 18:08:48 -08005114 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005115 if (ret) {
5116 mlog_errno(ret);
5117 goto out;
5118 }
5119 el = path_leaf_el(left_path);
5120
5121 index = ocfs2_search_extent_list(el, cpos);
5122 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
Tao Ma555936b2009-08-18 11:22:21 +08005123 ocfs2_error(sb,
5124 "Owner %llu has an extent at cpos %u which can no "
Mark Fasheh328d5752007-06-18 10:48:04 -07005125 "longer be found.\n",
Tao Ma555936b2009-08-18 11:22:21 +08005126 (unsigned long long)
5127 ocfs2_metadata_cache_owner(et->et_ci), cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005128 ret = -EROFS;
5129 goto out;
5130 }
5131
Tao Ma555936b2009-08-18 11:22:21 +08005132 ret = -EIO;
5133 rec = &el->l_recs[index];
5134 if (new_flags && (rec->e_flags & new_flags)) {
5135 mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
5136 "extent that already had them",
5137 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5138 new_flags);
5139 goto out;
5140 }
5141
5142 if (clear_flags && !(rec->e_flags & clear_flags)) {
5143 mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
5144 "extent that didn't have them",
5145 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5146 clear_flags);
5147 goto out;
5148 }
5149
Mark Fasheh328d5752007-06-18 10:48:04 -07005150 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5151 split_rec.e_cpos = cpu_to_le32(cpos);
5152 split_rec.e_leaf_clusters = cpu_to_le16(len);
5153 split_rec.e_blkno = cpu_to_le64(start_blkno);
Tao Ma555936b2009-08-18 11:22:21 +08005154 split_rec.e_flags = rec->e_flags;
5155 if (new_flags)
5156 split_rec.e_flags |= new_flags;
5157 if (clear_flags)
5158 split_rec.e_flags &= ~clear_flags;
Mark Fasheh328d5752007-06-18 10:48:04 -07005159
Tao Mae2e9f602009-08-18 11:22:34 +08005160 ret = ocfs2_split_extent(handle, et, left_path,
5161 index, &split_rec, meta_ac,
5162 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005163 if (ret)
5164 mlog_errno(ret);
5165
5166out:
5167 ocfs2_free_path(left_path);
5168 return ret;
Tao Ma555936b2009-08-18 11:22:21 +08005169
5170}
5171
5172/*
5173 * Mark the already-existing extent at cpos as written for len clusters.
5174 * This removes the unwritten extent flag.
5175 *
5176 * If the existing extent is larger than the request, initiate a
5177 * split. An attempt will be made at merging with adjacent extents.
5178 *
5179 * The caller is responsible for passing down meta_ac if we'll need it.
5180 */
5181int ocfs2_mark_extent_written(struct inode *inode,
5182 struct ocfs2_extent_tree *et,
5183 handle_t *handle, u32 cpos, u32 len, u32 phys,
5184 struct ocfs2_alloc_context *meta_ac,
5185 struct ocfs2_cached_dealloc_ctxt *dealloc)
5186{
5187 int ret;
5188
Tao Maa09d09b2011-02-22 08:21:10 +08005189 trace_ocfs2_mark_extent_written(
5190 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5191 cpos, len, phys);
Tao Ma555936b2009-08-18 11:22:21 +08005192
5193 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
5194 ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
5195 "that are being written to, but the feature bit "
5196 "is not set in the super block.",
5197 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5198 ret = -EROFS;
5199 goto out;
5200 }
5201
5202 /*
5203 * XXX: This should be fixed up so that we just re-insert the
5204 * next extent records.
5205 */
5206 ocfs2_et_extent_map_truncate(et, 0);
5207
5208 ret = ocfs2_change_extent_flag(handle, et, cpos,
5209 len, phys, meta_ac, dealloc,
5210 0, OCFS2_EXT_UNWRITTEN);
5211 if (ret)
5212 mlog_errno(ret);
5213
5214out:
5215 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07005216}
5217
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005218static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5219 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005220 int index, u32 new_range,
5221 struct ocfs2_alloc_context *meta_ac)
5222{
Tao Mac901fb02010-04-26 14:34:57 +08005223 int ret, depth, credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005224 struct buffer_head *last_eb_bh = NULL;
5225 struct ocfs2_extent_block *eb;
5226 struct ocfs2_extent_list *rightmost_el, *el;
5227 struct ocfs2_extent_rec split_rec;
5228 struct ocfs2_extent_rec *rec;
5229 struct ocfs2_insert_type insert;
5230
5231 /*
5232 * Setup the record to split before we grow the tree.
5233 */
5234 el = path_leaf_el(path);
5235 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005236 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5237 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005238
5239 depth = path->p_tree_depth;
5240 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005241 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005242 ocfs2_et_get_last_eb_blk(et),
5243 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005244 if (ret < 0) {
5245 mlog_errno(ret);
5246 goto out;
5247 }
5248
5249 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5250 rightmost_el = &eb->h_list;
5251 } else
5252 rightmost_el = path_leaf_el(path);
5253
Tao Mac901fb02010-04-26 14:34:57 +08005254 credits = path->p_tree_depth +
5255 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005256 ret = ocfs2_extend_trans(handle, credits);
5257 if (ret) {
5258 mlog_errno(ret);
5259 goto out;
5260 }
5261
5262 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5263 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005264 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005265 meta_ac);
5266 if (ret) {
5267 mlog_errno(ret);
5268 goto out;
5269 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005270 }
5271
5272 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5273 insert.ins_appending = APPEND_NONE;
5274 insert.ins_contig = CONTIG_NONE;
5275 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005276 insert.ins_tree_depth = depth;
5277
Joel Becker3505bec2009-02-13 02:57:58 -08005278 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005279 if (ret)
5280 mlog_errno(ret);
5281
5282out:
5283 brelse(last_eb_bh);
5284 return ret;
5285}
5286
Joel Becker043beeb2009-02-13 02:42:30 -08005287static int ocfs2_truncate_rec(handle_t *handle,
5288 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005289 struct ocfs2_path *path, int index,
5290 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005291 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005292{
5293 int ret;
5294 u32 left_cpos, rec_range, trunc_range;
5295 int wants_rotate = 0, is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005296 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005297 struct ocfs2_path *left_path = NULL;
5298 struct ocfs2_extent_list *el = path_leaf_el(path);
5299 struct ocfs2_extent_rec *rec;
5300 struct ocfs2_extent_block *eb;
5301
5302 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Joel Becker70f18c02009-02-13 02:09:31 -08005303 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005304 if (ret) {
5305 mlog_errno(ret);
5306 goto out;
5307 }
5308
5309 index--;
5310 }
5311
5312 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5313 path->p_tree_depth) {
5314 /*
5315 * Check whether this is the rightmost tree record. If
5316 * we remove all of this record or part of its right
5317 * edge then an update of the record lengths above it
5318 * will be required.
5319 */
5320 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5321 if (eb->h_next_leaf_blk == 0)
5322 is_rightmost_tree_rec = 1;
5323 }
5324
5325 rec = &el->l_recs[index];
5326 if (index == 0 && path->p_tree_depth &&
5327 le32_to_cpu(rec->e_cpos) == cpos) {
5328 /*
5329 * Changing the leftmost offset (via partial or whole
5330 * record truncate) of an interior (or rightmost) path
5331 * means we have to update the subtree that is formed
5332 * by this leaf and the one to it's left.
5333 *
5334 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005335 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005336 * 2) The leaf is rightmost and will be empty after
5337 * we remove the extent record - the rotate code
5338 * knows how to update the newly formed edge.
5339 */
5340
Joel Becker043beeb2009-02-13 02:42:30 -08005341 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005342 if (ret) {
5343 mlog_errno(ret);
5344 goto out;
5345 }
5346
5347 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005348 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005349 if (!left_path) {
5350 ret = -ENOMEM;
5351 mlog_errno(ret);
5352 goto out;
5353 }
5354
Joel Beckerfacdb772009-02-12 18:08:48 -08005355 ret = ocfs2_find_path(et->et_ci, left_path,
5356 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005357 if (ret) {
5358 mlog_errno(ret);
5359 goto out;
5360 }
5361 }
5362 }
5363
5364 ret = ocfs2_extend_rotate_transaction(handle, 0,
5365 handle->h_buffer_credits,
5366 path);
5367 if (ret) {
5368 mlog_errno(ret);
5369 goto out;
5370 }
5371
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005372 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005373 if (ret) {
5374 mlog_errno(ret);
5375 goto out;
5376 }
5377
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005378 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005379 if (ret) {
5380 mlog_errno(ret);
5381 goto out;
5382 }
5383
5384 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5385 trunc_range = cpos + len;
5386
5387 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5388 int next_free;
5389
5390 memset(rec, 0, sizeof(*rec));
5391 ocfs2_cleanup_merge(el, index);
5392 wants_rotate = 1;
5393
5394 next_free = le16_to_cpu(el->l_next_free_rec);
5395 if (is_rightmost_tree_rec && next_free > 1) {
5396 /*
5397 * We skip the edge update if this path will
5398 * be deleted by the rotate code.
5399 */
5400 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005401 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005402 rec);
5403 }
5404 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5405 /* Remove leftmost portion of the record. */
5406 le32_add_cpu(&rec->e_cpos, len);
5407 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5408 le16_add_cpu(&rec->e_leaf_clusters, -len);
5409 } else if (rec_range == trunc_range) {
5410 /* Remove rightmost portion of the record */
5411 le16_add_cpu(&rec->e_leaf_clusters, -len);
5412 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005413 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005414 } else {
5415 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005416 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5417 "(%u, %u)\n",
5418 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005419 le32_to_cpu(rec->e_cpos),
5420 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5421 BUG();
5422 }
5423
5424 if (left_path) {
5425 int subtree_index;
5426
Joel Becker7dc02802009-02-12 19:20:13 -08005427 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005428 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005429 subtree_index);
5430 }
5431
5432 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5433
Joel Becker70f18c02009-02-13 02:09:31 -08005434 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005435 if (ret) {
5436 mlog_errno(ret);
5437 goto out;
5438 }
5439
5440out:
5441 ocfs2_free_path(left_path);
5442 return ret;
5443}
5444
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005445int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005446 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005447 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005448 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005449 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005450{
5451 int ret, index;
5452 u32 rec_range, trunc_range;
5453 struct ocfs2_extent_rec *rec;
5454 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005455 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005456
Joel Becker4c911ee2009-02-13 02:50:12 -08005457 /*
5458 * XXX: Why are we truncating to 0 instead of wherever this
5459 * affects us?
5460 */
5461 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005462
Joel Beckerffdd7a52008-10-17 22:32:01 -07005463 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005464 if (!path) {
5465 ret = -ENOMEM;
5466 mlog_errno(ret);
5467 goto out;
5468 }
5469
Joel Beckerfacdb772009-02-12 18:08:48 -08005470 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005471 if (ret) {
5472 mlog_errno(ret);
5473 goto out;
5474 }
5475
5476 el = path_leaf_el(path);
5477 index = ocfs2_search_extent_list(el, cpos);
5478 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005479 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5480 "Owner %llu has an extent at cpos %u which can no "
Mark Fashehd0c7d702007-07-03 13:27:22 -07005481 "longer be found.\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005482 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5483 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005484 ret = -EROFS;
5485 goto out;
5486 }
5487
5488 /*
5489 * We have 3 cases of extent removal:
5490 * 1) Range covers the entire extent rec
5491 * 2) Range begins or ends on one edge of the extent rec
5492 * 3) Range is in the middle of the extent rec (no shared edges)
5493 *
5494 * For case 1 we remove the extent rec and left rotate to
5495 * fill the hole.
5496 *
5497 * For case 2 we just shrink the existing extent rec, with a
5498 * tree update if the shrinking edge is also the edge of an
5499 * extent block.
5500 *
5501 * For case 3 we do a right split to turn the extent rec into
5502 * something case 2 can handle.
5503 */
5504 rec = &el->l_recs[index];
5505 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5506 trunc_range = cpos + len;
5507
5508 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5509
Tao Maa09d09b2011-02-22 08:21:10 +08005510 trace_ocfs2_remove_extent(
5511 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5512 cpos, len, index, le32_to_cpu(rec->e_cpos),
5513 ocfs2_rec_clusters(el, rec));
Mark Fashehd0c7d702007-07-03 13:27:22 -07005514
5515 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005516 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5517 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005518 if (ret) {
5519 mlog_errno(ret);
5520 goto out;
5521 }
5522 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005523 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005524 trunc_range, meta_ac);
5525 if (ret) {
5526 mlog_errno(ret);
5527 goto out;
5528 }
5529
5530 /*
5531 * The split could have manipulated the tree enough to
5532 * move the record location, so we have to look for it again.
5533 */
5534 ocfs2_reinit_path(path, 1);
5535
Joel Beckerfacdb772009-02-12 18:08:48 -08005536 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005537 if (ret) {
5538 mlog_errno(ret);
5539 goto out;
5540 }
5541
5542 el = path_leaf_el(path);
5543 index = ocfs2_search_extent_list(el, cpos);
5544 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005545 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5546 "Owner %llu: split at cpos %u lost record.",
5547 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005548 cpos);
5549 ret = -EROFS;
5550 goto out;
5551 }
5552
5553 /*
5554 * Double check our values here. If anything is fishy,
5555 * it's easier to catch it at the top level.
5556 */
5557 rec = &el->l_recs[index];
5558 rec_range = le32_to_cpu(rec->e_cpos) +
5559 ocfs2_rec_clusters(el, rec);
5560 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005561 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5562 "Owner %llu: error after split at cpos %u"
Mark Fashehd0c7d702007-07-03 13:27:22 -07005563 "trunc len %u, existing record is (%u,%u)",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005564 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005565 cpos, len, le32_to_cpu(rec->e_cpos),
5566 ocfs2_rec_clusters(el, rec));
5567 ret = -EROFS;
5568 goto out;
5569 }
5570
Joel Becker043beeb2009-02-13 02:42:30 -08005571 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5572 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005573 if (ret) {
5574 mlog_errno(ret);
5575 goto out;
5576 }
5577 }
5578
5579out:
5580 ocfs2_free_path(path);
5581 return ret;
5582}
5583
Tristan Ye78f94672010-05-11 17:54:42 +08005584/*
5585 * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5586 * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5587 * number to reserve some extra blocks, and it only handles meta
5588 * data allocations.
5589 *
5590 * Currently, only ocfs2_remove_btree_range() uses it for truncating
5591 * and punching holes.
5592 */
5593static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5594 struct ocfs2_extent_tree *et,
5595 u32 extents_to_split,
5596 struct ocfs2_alloc_context **ac,
5597 int extra_blocks)
5598{
5599 int ret = 0, num_free_extents;
5600 unsigned int max_recs_needed = 2 * extents_to_split;
5601 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5602
5603 *ac = NULL;
5604
5605 num_free_extents = ocfs2_num_free_extents(osb, et);
5606 if (num_free_extents < 0) {
5607 ret = num_free_extents;
5608 mlog_errno(ret);
5609 goto out;
5610 }
5611
5612 if (!num_free_extents ||
5613 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5614 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5615
5616 if (extra_blocks) {
5617 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5618 if (ret < 0) {
5619 if (ret != -ENOSPC)
5620 mlog_errno(ret);
5621 goto out;
5622 }
5623 }
5624
5625out:
5626 if (ret) {
5627 if (*ac) {
5628 ocfs2_free_alloc_context(*ac);
5629 *ac = NULL;
5630 }
5631 }
5632
5633 return ret;
5634}
5635
Mark Fashehfecc0112008-11-12 15:16:38 -08005636int ocfs2_remove_btree_range(struct inode *inode,
5637 struct ocfs2_extent_tree *et,
Tristan Ye78f94672010-05-11 17:54:42 +08005638 u32 cpos, u32 phys_cpos, u32 len, int flags,
5639 struct ocfs2_cached_dealloc_ctxt *dealloc,
5640 u64 refcount_loc)
Mark Fashehfecc0112008-11-12 15:16:38 -08005641{
Tristan Ye78f94672010-05-11 17:54:42 +08005642 int ret, credits = 0, extra_blocks = 0;
Mark Fashehfecc0112008-11-12 15:16:38 -08005643 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5645 struct inode *tl_inode = osb->osb_tl_inode;
5646 handle_t *handle;
5647 struct ocfs2_alloc_context *meta_ac = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08005648 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehfecc0112008-11-12 15:16:38 -08005649
Tristan Ye78f94672010-05-11 17:54:42 +08005650 if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
5651 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
5652 OCFS2_HAS_REFCOUNT_FL));
5653
5654 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5655 &ref_tree, NULL);
5656 if (ret) {
5657 mlog_errno(ret);
5658 goto out;
5659 }
5660
5661 ret = ocfs2_prepare_refcount_change_for_del(inode,
5662 refcount_loc,
5663 phys_blkno,
5664 len,
5665 &credits,
5666 &extra_blocks);
5667 if (ret < 0) {
5668 mlog_errno(ret);
5669 goto out;
5670 }
5671 }
5672
5673 ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5674 extra_blocks);
Mark Fashehfecc0112008-11-12 15:16:38 -08005675 if (ret) {
5676 mlog_errno(ret);
5677 return ret;
5678 }
5679
5680 mutex_lock(&tl_inode->i_mutex);
5681
5682 if (ocfs2_truncate_log_needs_flush(osb)) {
5683 ret = __ocfs2_flush_truncate_log(osb);
5684 if (ret < 0) {
5685 mlog_errno(ret);
5686 goto out;
5687 }
5688 }
5689
Tristan Ye78f94672010-05-11 17:54:42 +08005690 handle = ocfs2_start_trans(osb,
5691 ocfs2_remove_extent_credits(osb->sb) + credits);
Mark Fashehfecc0112008-11-12 15:16:38 -08005692 if (IS_ERR(handle)) {
5693 ret = PTR_ERR(handle);
5694 mlog_errno(ret);
5695 goto out;
5696 }
5697
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005698 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005699 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005700 if (ret) {
5701 mlog_errno(ret);
5702 goto out;
5703 }
5704
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005705 dquot_free_space_nodirty(inode,
Mark Fashehfd4ef232009-01-29 15:06:21 -08005706 ocfs2_clusters_to_bytes(inode->i_sb, len));
5707
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005708 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005709 if (ret) {
5710 mlog_errno(ret);
5711 goto out_commit;
5712 }
5713
Joel Becker6136ca52009-02-12 19:32:43 -08005714 ocfs2_et_update_clusters(et, -len);
Mark Fashehfecc0112008-11-12 15:16:38 -08005715
Joel Beckerec20cec2010-03-19 14:13:52 -07005716 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehfecc0112008-11-12 15:16:38 -08005717
Tristan Ye78f94672010-05-11 17:54:42 +08005718 if (phys_blkno) {
5719 if (flags & OCFS2_EXT_REFCOUNTED)
5720 ret = ocfs2_decrease_refcount(inode, handle,
5721 ocfs2_blocks_to_clusters(osb->sb,
5722 phys_blkno),
5723 len, meta_ac,
5724 dealloc, 1);
5725 else
5726 ret = ocfs2_truncate_log_append(osb, handle,
5727 phys_blkno, len);
5728 if (ret)
5729 mlog_errno(ret);
5730
5731 }
Mark Fashehfecc0112008-11-12 15:16:38 -08005732
5733out_commit:
5734 ocfs2_commit_trans(osb, handle);
5735out:
5736 mutex_unlock(&tl_inode->i_mutex);
5737
5738 if (meta_ac)
5739 ocfs2_free_alloc_context(meta_ac);
5740
Tristan Ye78f94672010-05-11 17:54:42 +08005741 if (ref_tree)
5742 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5743
Mark Fashehfecc0112008-11-12 15:16:38 -08005744 return ret;
5745}
5746
Mark Fasheh063c4562007-07-03 13:34:11 -07005747int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005748{
5749 struct buffer_head *tl_bh = osb->osb_tl_bh;
5750 struct ocfs2_dinode *di;
5751 struct ocfs2_truncate_log *tl;
5752
5753 di = (struct ocfs2_dinode *) tl_bh->b_data;
5754 tl = &di->id2.i_dealloc;
5755
5756 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5757 "slot %d, invalid truncate log parameters: used = "
5758 "%u, count = %u\n", osb->slot_num,
5759 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5760 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5761}
5762
5763static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5764 unsigned int new_start)
5765{
5766 unsigned int tail_index;
5767 unsigned int current_tail;
5768
5769 /* No records, nothing to coalesce */
5770 if (!le16_to_cpu(tl->tl_used))
5771 return 0;
5772
5773 tail_index = le16_to_cpu(tl->tl_used) - 1;
5774 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5775 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5776
5777 return current_tail == new_start;
5778}
5779
Mark Fasheh063c4562007-07-03 13:34:11 -07005780int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5781 handle_t *handle,
5782 u64 start_blk,
5783 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005784{
5785 int status, index;
5786 unsigned int start_cluster, tl_count;
5787 struct inode *tl_inode = osb->osb_tl_inode;
5788 struct buffer_head *tl_bh = osb->osb_tl_bh;
5789 struct ocfs2_dinode *di;
5790 struct ocfs2_truncate_log *tl;
5791
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005792 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005793
5794 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5795
5796 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005797
Joel Becker10995aa2008-11-13 14:49:12 -08005798 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5799 * by the underlying call to ocfs2_read_inode_block(), so any
5800 * corruption is a code bug */
5801 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5802
5803 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005804 tl_count = le16_to_cpu(tl->tl_count);
5805 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5806 tl_count == 0,
Mark Fashehb0697052006-03-03 10:24:33 -08005807 "Truncate record count on #%llu invalid "
5808 "wanted %u, actual %u\n",
5809 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005810 ocfs2_truncate_recs_per_inode(osb->sb),
5811 le16_to_cpu(tl->tl_count));
5812
5813 /* Caller should have known to flush before calling us. */
5814 index = le16_to_cpu(tl->tl_used);
5815 if (index >= tl_count) {
5816 status = -ENOSPC;
5817 mlog_errno(status);
5818 goto bail;
5819 }
5820
Joel Becker0cf2f762009-02-12 16:41:25 -08005821 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005822 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005823 if (status < 0) {
5824 mlog_errno(status);
5825 goto bail;
5826 }
5827
Tao Maa09d09b2011-02-22 08:21:10 +08005828 trace_ocfs2_truncate_log_append(
5829 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5830 start_cluster, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005831 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5832 /*
5833 * Move index back to the record we are coalescing with.
5834 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5835 */
5836 index--;
5837
5838 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
Tao Maa09d09b2011-02-22 08:21:10 +08005839 trace_ocfs2_truncate_log_append(
5840 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5841 index, le32_to_cpu(tl->tl_recs[index].t_start),
5842 num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005843 } else {
5844 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5845 tl->tl_used = cpu_to_le16(index + 1);
5846 }
5847 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5848
Joel Beckerec20cec2010-03-19 14:13:52 -07005849 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005850
Tao Ma50308d82010-11-04 15:14:11 +08005851 osb->truncated_clusters += num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08005852bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005853 return status;
5854}
5855
5856static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07005857 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08005858 struct inode *data_alloc_inode,
5859 struct buffer_head *data_alloc_bh)
5860{
5861 int status = 0;
5862 int i;
5863 unsigned int num_clusters;
5864 u64 start_blk;
5865 struct ocfs2_truncate_rec rec;
5866 struct ocfs2_dinode *di;
5867 struct ocfs2_truncate_log *tl;
5868 struct inode *tl_inode = osb->osb_tl_inode;
5869 struct buffer_head *tl_bh = osb->osb_tl_bh;
5870
Mark Fashehccd979b2005-12-15 14:31:24 -08005871 di = (struct ocfs2_dinode *) tl_bh->b_data;
5872 tl = &di->id2.i_dealloc;
5873 i = le16_to_cpu(tl->tl_used) - 1;
5874 while (i >= 0) {
5875 /* Caller has given us at least enough credits to
5876 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005877 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005878 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005879 if (status < 0) {
5880 mlog_errno(status);
5881 goto bail;
5882 }
5883
5884 tl->tl_used = cpu_to_le16(i);
5885
Joel Beckerec20cec2010-03-19 14:13:52 -07005886 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005887
5888 /* TODO: Perhaps we can calculate the bulk of the
5889 * credits up front rather than extending like
5890 * this. */
5891 status = ocfs2_extend_trans(handle,
5892 OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5893 if (status < 0) {
5894 mlog_errno(status);
5895 goto bail;
5896 }
5897
5898 rec = tl->tl_recs[i];
5899 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5900 le32_to_cpu(rec.t_start));
5901 num_clusters = le32_to_cpu(rec.t_clusters);
5902
5903 /* if start_blk is not set, we ignore the record as
5904 * invalid. */
5905 if (start_blk) {
Tao Maa09d09b2011-02-22 08:21:10 +08005906 trace_ocfs2_replay_truncate_records(
5907 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5908 i, le32_to_cpu(rec.t_start), num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005909
5910 status = ocfs2_free_clusters(handle, data_alloc_inode,
5911 data_alloc_bh, start_blk,
5912 num_clusters);
5913 if (status < 0) {
5914 mlog_errno(status);
5915 goto bail;
5916 }
5917 }
5918 i--;
5919 }
5920
Tao Ma50308d82010-11-04 15:14:11 +08005921 osb->truncated_clusters = 0;
5922
Mark Fashehccd979b2005-12-15 14:31:24 -08005923bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005924 return status;
5925}
5926
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005927/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005928int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005929{
5930 int status;
5931 unsigned int num_to_flush;
Mark Fasheh1fabe142006-10-09 18:11:45 -07005932 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005933 struct inode *tl_inode = osb->osb_tl_inode;
5934 struct inode *data_alloc_inode = NULL;
5935 struct buffer_head *tl_bh = osb->osb_tl_bh;
5936 struct buffer_head *data_alloc_bh = NULL;
5937 struct ocfs2_dinode *di;
5938 struct ocfs2_truncate_log *tl;
5939
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005940 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005941
5942 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005943
Joel Becker10995aa2008-11-13 14:49:12 -08005944 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5945 * by the underlying call to ocfs2_read_inode_block(), so any
5946 * corruption is a code bug */
5947 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5948
5949 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005950 num_to_flush = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08005951 trace_ocfs2_flush_truncate_log(
5952 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5953 num_to_flush);
Mark Fashehccd979b2005-12-15 14:31:24 -08005954 if (!num_to_flush) {
5955 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005956 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005957 }
5958
5959 data_alloc_inode = ocfs2_get_system_file_inode(osb,
5960 GLOBAL_BITMAP_SYSTEM_INODE,
5961 OCFS2_INVALID_SLOT);
5962 if (!data_alloc_inode) {
5963 status = -EINVAL;
5964 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005965 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005966 }
5967
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005968 mutex_lock(&data_alloc_inode->i_mutex);
5969
Mark Fashehe63aecb62007-10-18 15:30:42 -07005970 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08005971 if (status < 0) {
5972 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005973 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08005974 }
5975
Mark Fasheh65eff9c2006-10-09 17:26:22 -07005976 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005977 if (IS_ERR(handle)) {
5978 status = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005979 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005980 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08005981 }
5982
5983 status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
5984 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005985 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08005986 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08005987
Mark Fasheh02dc1af2006-10-09 16:48:10 -07005988 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005989
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005990out_unlock:
5991 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07005992 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08005993
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005994out_mutex:
5995 mutex_unlock(&data_alloc_inode->i_mutex);
5996 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08005997
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005998out:
Mark Fashehccd979b2005-12-15 14:31:24 -08005999 return status;
6000}
6001
6002int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6003{
6004 int status;
6005 struct inode *tl_inode = osb->osb_tl_inode;
6006
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006007 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006008 status = __ocfs2_flush_truncate_log(osb);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006009 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006010
6011 return status;
6012}
6013
David Howellsc4028952006-11-22 14:57:56 +00006014static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08006015{
6016 int status;
David Howellsc4028952006-11-22 14:57:56 +00006017 struct ocfs2_super *osb =
6018 container_of(work, struct ocfs2_super,
6019 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08006020
Mark Fashehccd979b2005-12-15 14:31:24 -08006021 status = ocfs2_flush_truncate_log(osb);
6022 if (status < 0)
6023 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08006024 else
Tiger Yangb89c5422010-01-25 14:11:06 +08006025 ocfs2_init_steal_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08006026}
6027
6028#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6029void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6030 int cancel)
6031{
6032 if (osb->osb_tl_inode) {
6033 /* We want to push off log flushes while truncates are
6034 * still running. */
6035 if (cancel)
6036 cancel_delayed_work(&osb->osb_truncate_log_wq);
6037
6038 queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
6039 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6040 }
6041}
6042
6043static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6044 int slot_num,
6045 struct inode **tl_inode,
6046 struct buffer_head **tl_bh)
6047{
6048 int status;
6049 struct inode *inode = NULL;
6050 struct buffer_head *bh = NULL;
6051
6052 inode = ocfs2_get_system_file_inode(osb,
6053 TRUNCATE_LOG_SYSTEM_INODE,
6054 slot_num);
6055 if (!inode) {
6056 status = -EINVAL;
6057 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6058 goto bail;
6059 }
6060
Joel Beckerb657c952008-11-13 14:49:11 -08006061 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006062 if (status < 0) {
6063 iput(inode);
6064 mlog_errno(status);
6065 goto bail;
6066 }
6067
6068 *tl_inode = inode;
6069 *tl_bh = bh;
6070bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08006071 return status;
6072}
6073
6074/* called during the 1st stage of node recovery. we stamp a clean
6075 * truncate log and pass back a copy for processing later. if the
6076 * truncate log does not require processing, a *tl_copy is set to
6077 * NULL. */
6078int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6079 int slot_num,
6080 struct ocfs2_dinode **tl_copy)
6081{
6082 int status;
6083 struct inode *tl_inode = NULL;
6084 struct buffer_head *tl_bh = NULL;
6085 struct ocfs2_dinode *di;
6086 struct ocfs2_truncate_log *tl;
6087
6088 *tl_copy = NULL;
6089
Tao Maa09d09b2011-02-22 08:21:10 +08006090 trace_ocfs2_begin_truncate_log_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08006091
6092 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6093 if (status < 0) {
6094 mlog_errno(status);
6095 goto bail;
6096 }
6097
6098 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006099
Joel Becker10995aa2008-11-13 14:49:12 -08006100 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6101 * validated by the underlying call to ocfs2_read_inode_block(),
6102 * so any corruption is a code bug */
6103 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6104
6105 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006106 if (le16_to_cpu(tl->tl_used)) {
Tao Maa09d09b2011-02-22 08:21:10 +08006107 trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
Mark Fashehccd979b2005-12-15 14:31:24 -08006108
6109 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6110 if (!(*tl_copy)) {
6111 status = -ENOMEM;
6112 mlog_errno(status);
6113 goto bail;
6114 }
6115
6116 /* Assuming the write-out below goes well, this copy
6117 * will be passed back to recovery for processing. */
6118 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6119
6120 /* All we need to do to clear the truncate log is set
6121 * tl_used. */
6122 tl->tl_used = 0;
6123
Joel Becker13723d02008-10-17 19:25:01 -07006124 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006125 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006126 if (status < 0) {
6127 mlog_errno(status);
6128 goto bail;
6129 }
6130 }
6131
6132bail:
6133 if (tl_inode)
6134 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006135 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006136
6137 if (status < 0 && (*tl_copy)) {
6138 kfree(*tl_copy);
6139 *tl_copy = NULL;
Tao Mac1e8d352011-03-07 16:43:21 +08006140 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006141 }
6142
Mark Fashehccd979b2005-12-15 14:31:24 -08006143 return status;
6144}
6145
6146int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6147 struct ocfs2_dinode *tl_copy)
6148{
6149 int status = 0;
6150 int i;
6151 unsigned int clusters, num_recs, start_cluster;
6152 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006153 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006154 struct inode *tl_inode = osb->osb_tl_inode;
6155 struct ocfs2_truncate_log *tl;
6156
Mark Fashehccd979b2005-12-15 14:31:24 -08006157 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6158 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6159 return -EINVAL;
6160 }
6161
6162 tl = &tl_copy->id2.i_dealloc;
6163 num_recs = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006164 trace_ocfs2_complete_truncate_log_recovery(
6165 (unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6166 num_recs);
Mark Fashehccd979b2005-12-15 14:31:24 -08006167
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006168 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006169 for(i = 0; i < num_recs; i++) {
6170 if (ocfs2_truncate_log_needs_flush(osb)) {
6171 status = __ocfs2_flush_truncate_log(osb);
6172 if (status < 0) {
6173 mlog_errno(status);
6174 goto bail_up;
6175 }
6176 }
6177
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006178 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006179 if (IS_ERR(handle)) {
6180 status = PTR_ERR(handle);
6181 mlog_errno(status);
6182 goto bail_up;
6183 }
6184
6185 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6186 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6187 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6188
6189 status = ocfs2_truncate_log_append(osb, handle,
6190 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006191 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006192 if (status < 0) {
6193 mlog_errno(status);
6194 goto bail_up;
6195 }
6196 }
6197
6198bail_up:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006199 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006200
Mark Fashehccd979b2005-12-15 14:31:24 -08006201 return status;
6202}
6203
6204void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6205{
6206 int status;
6207 struct inode *tl_inode = osb->osb_tl_inode;
6208
Mark Fashehccd979b2005-12-15 14:31:24 -08006209 if (tl_inode) {
6210 cancel_delayed_work(&osb->osb_truncate_log_wq);
6211 flush_workqueue(ocfs2_wq);
6212
6213 status = ocfs2_flush_truncate_log(osb);
6214 if (status < 0)
6215 mlog_errno(status);
6216
6217 brelse(osb->osb_tl_bh);
6218 iput(osb->osb_tl_inode);
6219 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006220}
6221
6222int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6223{
6224 int status;
6225 struct inode *tl_inode = NULL;
6226 struct buffer_head *tl_bh = NULL;
6227
Mark Fashehccd979b2005-12-15 14:31:24 -08006228 status = ocfs2_get_truncate_log_info(osb,
6229 osb->slot_num,
6230 &tl_inode,
6231 &tl_bh);
6232 if (status < 0)
6233 mlog_errno(status);
6234
6235 /* ocfs2_truncate_log_shutdown keys on the existence of
6236 * osb->osb_tl_inode so we don't set any of the osb variables
6237 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006238 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6239 ocfs2_truncate_log_worker);
Mark Fashehccd979b2005-12-15 14:31:24 -08006240 osb->osb_tl_bh = tl_bh;
6241 osb->osb_tl_inode = tl_inode;
6242
Mark Fashehccd979b2005-12-15 14:31:24 -08006243 return status;
6244}
6245
Mark Fasheh2b604352007-06-22 15:45:27 -07006246/*
6247 * Delayed de-allocation of suballocator blocks.
6248 *
6249 * Some sets of block de-allocations might involve multiple suballocator inodes.
6250 *
6251 * The locking for this can get extremely complicated, especially when
6252 * the suballocator inodes to delete from aren't known until deep
6253 * within an unrelated codepath.
6254 *
6255 * ocfs2_extent_block structures are a good example of this - an inode
6256 * btree could have been grown by any number of nodes each allocating
6257 * out of their own suballoc inode.
6258 *
6259 * These structures allow the delay of block de-allocation until a
6260 * later time, when locking of multiple cluster inodes won't cause
6261 * deadlock.
6262 */
6263
6264/*
Tao Ma2891d292008-11-12 08:26:58 +08006265 * Describe a single bit freed from a suballocator. For the block
6266 * suballocators, it represents one block. For the global cluster
6267 * allocator, it represents some clusters and free_bit indicates
6268 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006269 */
6270struct ocfs2_cached_block_free {
6271 struct ocfs2_cached_block_free *free_next;
Tao Ma74380c42010-03-22 14:20:18 +08006272 u64 free_bg;
Mark Fasheh2b604352007-06-22 15:45:27 -07006273 u64 free_blk;
6274 unsigned int free_bit;
6275};
6276
6277struct ocfs2_per_slot_free_list {
6278 struct ocfs2_per_slot_free_list *f_next_suballocator;
6279 int f_inode_type;
6280 int f_slot;
6281 struct ocfs2_cached_block_free *f_first;
6282};
6283
Tao Ma2891d292008-11-12 08:26:58 +08006284static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6285 int sysfile_type,
6286 int slot,
6287 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006288{
6289 int ret;
6290 u64 bg_blkno;
6291 handle_t *handle;
6292 struct inode *inode;
6293 struct buffer_head *di_bh = NULL;
6294 struct ocfs2_cached_block_free *tmp;
6295
6296 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6297 if (!inode) {
6298 ret = -EINVAL;
6299 mlog_errno(ret);
6300 goto out;
6301 }
6302
6303 mutex_lock(&inode->i_mutex);
6304
Mark Fashehe63aecb62007-10-18 15:30:42 -07006305 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006306 if (ret) {
6307 mlog_errno(ret);
6308 goto out_mutex;
6309 }
6310
6311 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6312 if (IS_ERR(handle)) {
6313 ret = PTR_ERR(handle);
6314 mlog_errno(ret);
6315 goto out_unlock;
6316 }
6317
6318 while (head) {
Tao Ma74380c42010-03-22 14:20:18 +08006319 if (head->free_bg)
6320 bg_blkno = head->free_bg;
6321 else
6322 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6323 head->free_bit);
Tao Maa09d09b2011-02-22 08:21:10 +08006324 trace_ocfs2_free_cached_blocks(
6325 (unsigned long long)head->free_blk, head->free_bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006326
6327 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6328 head->free_bit, bg_blkno, 1);
6329 if (ret) {
6330 mlog_errno(ret);
6331 goto out_journal;
6332 }
6333
6334 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6335 if (ret) {
6336 mlog_errno(ret);
6337 goto out_journal;
6338 }
6339
6340 tmp = head;
6341 head = head->free_next;
6342 kfree(tmp);
6343 }
6344
6345out_journal:
6346 ocfs2_commit_trans(osb, handle);
6347
6348out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006349 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006350 brelse(di_bh);
6351out_mutex:
6352 mutex_unlock(&inode->i_mutex);
6353 iput(inode);
6354out:
6355 while(head) {
6356 /* Premature exit may have left some dangling items. */
6357 tmp = head;
6358 head = head->free_next;
6359 kfree(tmp);
6360 }
6361
6362 return ret;
6363}
6364
Tao Ma2891d292008-11-12 08:26:58 +08006365int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6366 u64 blkno, unsigned int bit)
6367{
6368 int ret = 0;
6369 struct ocfs2_cached_block_free *item;
6370
Tao Ma74380c42010-03-22 14:20:18 +08006371 item = kzalloc(sizeof(*item), GFP_NOFS);
Tao Ma2891d292008-11-12 08:26:58 +08006372 if (item == NULL) {
6373 ret = -ENOMEM;
6374 mlog_errno(ret);
6375 return ret;
6376 }
6377
Tao Maa09d09b2011-02-22 08:21:10 +08006378 trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
Tao Ma2891d292008-11-12 08:26:58 +08006379
6380 item->free_blk = blkno;
6381 item->free_bit = bit;
6382 item->free_next = ctxt->c_global_allocator;
6383
6384 ctxt->c_global_allocator = item;
6385 return ret;
6386}
6387
6388static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6389 struct ocfs2_cached_block_free *head)
6390{
6391 struct ocfs2_cached_block_free *tmp;
6392 struct inode *tl_inode = osb->osb_tl_inode;
6393 handle_t *handle;
6394 int ret = 0;
6395
6396 mutex_lock(&tl_inode->i_mutex);
6397
6398 while (head) {
6399 if (ocfs2_truncate_log_needs_flush(osb)) {
6400 ret = __ocfs2_flush_truncate_log(osb);
6401 if (ret < 0) {
6402 mlog_errno(ret);
6403 break;
6404 }
6405 }
6406
6407 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6408 if (IS_ERR(handle)) {
6409 ret = PTR_ERR(handle);
6410 mlog_errno(ret);
6411 break;
6412 }
6413
6414 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6415 head->free_bit);
6416
6417 ocfs2_commit_trans(osb, handle);
6418 tmp = head;
6419 head = head->free_next;
6420 kfree(tmp);
6421
6422 if (ret < 0) {
6423 mlog_errno(ret);
6424 break;
6425 }
6426 }
6427
6428 mutex_unlock(&tl_inode->i_mutex);
6429
6430 while (head) {
6431 /* Premature exit may have left some dangling items. */
6432 tmp = head;
6433 head = head->free_next;
6434 kfree(tmp);
6435 }
6436
6437 return ret;
6438}
6439
Mark Fasheh2b604352007-06-22 15:45:27 -07006440int ocfs2_run_deallocs(struct ocfs2_super *osb,
6441 struct ocfs2_cached_dealloc_ctxt *ctxt)
6442{
6443 int ret = 0, ret2;
6444 struct ocfs2_per_slot_free_list *fl;
6445
6446 if (!ctxt)
6447 return 0;
6448
6449 while (ctxt->c_first_suballocator) {
6450 fl = ctxt->c_first_suballocator;
6451
6452 if (fl->f_first) {
Tao Maa09d09b2011-02-22 08:21:10 +08006453 trace_ocfs2_run_deallocs(fl->f_inode_type,
6454 fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006455 ret2 = ocfs2_free_cached_blocks(osb,
6456 fl->f_inode_type,
6457 fl->f_slot,
6458 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006459 if (ret2)
6460 mlog_errno(ret2);
6461 if (!ret)
6462 ret = ret2;
6463 }
6464
6465 ctxt->c_first_suballocator = fl->f_next_suballocator;
6466 kfree(fl);
6467 }
6468
Tao Ma2891d292008-11-12 08:26:58 +08006469 if (ctxt->c_global_allocator) {
6470 ret2 = ocfs2_free_cached_clusters(osb,
6471 ctxt->c_global_allocator);
6472 if (ret2)
6473 mlog_errno(ret2);
6474 if (!ret)
6475 ret = ret2;
6476
6477 ctxt->c_global_allocator = NULL;
6478 }
6479
Mark Fasheh2b604352007-06-22 15:45:27 -07006480 return ret;
6481}
6482
6483static struct ocfs2_per_slot_free_list *
6484ocfs2_find_per_slot_free_list(int type,
6485 int slot,
6486 struct ocfs2_cached_dealloc_ctxt *ctxt)
6487{
6488 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6489
6490 while (fl) {
6491 if (fl->f_inode_type == type && fl->f_slot == slot)
6492 return fl;
6493
6494 fl = fl->f_next_suballocator;
6495 }
6496
6497 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6498 if (fl) {
6499 fl->f_inode_type = type;
6500 fl->f_slot = slot;
6501 fl->f_first = NULL;
6502 fl->f_next_suballocator = ctxt->c_first_suballocator;
6503
6504 ctxt->c_first_suballocator = fl;
6505 }
6506 return fl;
6507}
6508
Tao Ma1823cb02009-08-18 11:24:49 +08006509int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
Tao Ma74380c42010-03-22 14:20:18 +08006510 int type, int slot, u64 suballoc,
6511 u64 blkno, unsigned int bit)
Mark Fasheh2b604352007-06-22 15:45:27 -07006512{
6513 int ret;
6514 struct ocfs2_per_slot_free_list *fl;
6515 struct ocfs2_cached_block_free *item;
6516
6517 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6518 if (fl == NULL) {
6519 ret = -ENOMEM;
6520 mlog_errno(ret);
6521 goto out;
6522 }
6523
Tao Ma74380c42010-03-22 14:20:18 +08006524 item = kzalloc(sizeof(*item), GFP_NOFS);
Mark Fasheh2b604352007-06-22 15:45:27 -07006525 if (item == NULL) {
6526 ret = -ENOMEM;
6527 mlog_errno(ret);
6528 goto out;
6529 }
6530
Tao Maa09d09b2011-02-22 08:21:10 +08006531 trace_ocfs2_cache_block_dealloc(type, slot,
6532 (unsigned long long)suballoc,
6533 (unsigned long long)blkno, bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006534
Tao Ma74380c42010-03-22 14:20:18 +08006535 item->free_bg = suballoc;
Mark Fasheh2b604352007-06-22 15:45:27 -07006536 item->free_blk = blkno;
6537 item->free_bit = bit;
6538 item->free_next = fl->f_first;
6539
6540 fl->f_first = item;
6541
6542 ret = 0;
6543out:
6544 return ret;
6545}
6546
Mark Fasheh59a5e412007-06-22 15:52:36 -07006547static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6548 struct ocfs2_extent_block *eb)
6549{
6550 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6551 le16_to_cpu(eb->h_suballoc_slot),
Tao Ma74380c42010-03-22 14:20:18 +08006552 le64_to_cpu(eb->h_suballoc_loc),
Mark Fasheh59a5e412007-06-22 15:52:36 -07006553 le64_to_cpu(eb->h_blkno),
6554 le16_to_cpu(eb->h_suballoc_bit));
6555}
6556
Joel Becker2b4e30f2008-09-03 20:03:41 -07006557static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006558{
6559 set_buffer_uptodate(bh);
6560 mark_buffer_dirty(bh);
6561 return 0;
6562}
6563
Tao Ma6f70fa52009-08-25 08:05:12 +08006564void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6565 unsigned int from, unsigned int to,
6566 struct page *page, int zero, u64 *phys)
Mark Fasheh1d410a62007-09-07 14:20:45 -07006567{
6568 int ret, partial = 0;
6569
6570 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6571 if (ret)
6572 mlog_errno(ret);
6573
6574 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006575 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006576
6577 /*
6578 * Need to set the buffers we zero'd into uptodate
6579 * here if they aren't - ocfs2_map_page_blocks()
6580 * might've skipped some
6581 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006582 ret = walk_page_buffers(handle, page_buffers(page),
6583 from, to, &partial,
6584 ocfs2_zero_func);
6585 if (ret < 0)
6586 mlog_errno(ret);
6587 else if (ocfs2_should_order_data(inode)) {
6588 ret = ocfs2_jbd2_file_inode(handle, inode);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006589 if (ret < 0)
6590 mlog_errno(ret);
6591 }
6592
6593 if (!partial)
6594 SetPageUptodate(page);
6595
6596 flush_dcache_page(page);
6597}
6598
Mark Fasheh35edec12007-07-06 14:41:18 -07006599static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6600 loff_t end, struct page **pages,
6601 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006602{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006603 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006604 struct page *page;
6605 unsigned int from, to = PAGE_CACHE_SIZE;
6606 struct super_block *sb = inode->i_sb;
6607
6608 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6609
6610 if (numpages == 0)
6611 goto out;
6612
Mark Fasheh35edec12007-07-06 14:41:18 -07006613 to = PAGE_CACHE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006614 for(i = 0; i < numpages; i++) {
6615 page = pages[i];
6616
Mark Fasheh35edec12007-07-06 14:41:18 -07006617 from = start & (PAGE_CACHE_SIZE - 1);
6618 if ((end >> PAGE_CACHE_SHIFT) == page->index)
6619 to = end & (PAGE_CACHE_SIZE - 1);
6620
Mark Fasheh60b11392007-02-16 11:46:50 -08006621 BUG_ON(from > PAGE_CACHE_SIZE);
6622 BUG_ON(to > PAGE_CACHE_SIZE);
6623
Mark Fasheh1d410a62007-09-07 14:20:45 -07006624 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6625 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006626
Mark Fasheh35edec12007-07-06 14:41:18 -07006627 start = (page->index + 1) << PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006628 }
6629out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006630 if (pages)
6631 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006632}
6633
Tao Ma6f70fa52009-08-25 08:05:12 +08006634int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6635 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006636{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006637 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006638 struct address_space *mapping = inode->i_mapping;
6639 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006640 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006641
Mark Fasheh35edec12007-07-06 14:41:18 -07006642 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006643
Mark Fasheh1d410a62007-09-07 14:20:45 -07006644 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006645 last_page_bytes = PAGE_ALIGN(end);
6646 index = start >> PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006647 do {
Jan Kara9b4c0ff2010-08-24 14:28:03 +02006648 pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fasheh60b11392007-02-16 11:46:50 -08006649 if (!pages[numpages]) {
6650 ret = -ENOMEM;
6651 mlog_errno(ret);
6652 goto out;
6653 }
6654
6655 numpages++;
6656 index++;
Mark Fasheh35edec12007-07-06 14:41:18 -07006657 } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006658
6659out:
6660 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006661 if (pages)
6662 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006663 numpages = 0;
6664 }
6665
6666 *num = numpages;
6667
6668 return ret;
6669}
6670
Tao Ma6f70fa52009-08-25 08:05:12 +08006671static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6672 struct page **pages, int *num)
6673{
6674 struct super_block *sb = inode->i_sb;
6675
6676 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6677 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6678
6679 return ocfs2_grab_pages(inode, start, end, pages, num);
6680}
6681
Mark Fasheh60b11392007-02-16 11:46:50 -08006682/*
6683 * Zero the area past i_size but still within an allocated
6684 * cluster. This avoids exposing nonzero data on subsequent file
6685 * extends.
6686 *
6687 * We need to call this before i_size is updated on the inode because
6688 * otherwise block_write_full_page() will skip writeout of pages past
6689 * i_size. The new_i_size parameter is passed for this reason.
6690 */
Mark Fasheh35edec12007-07-06 14:41:18 -07006691int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6692 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08006693{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006694 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08006695 struct page **pages = NULL;
6696 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006697 unsigned int ext_flags;
6698 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08006699
6700 /*
6701 * File systems which don't support sparse files zero on every
6702 * extend.
6703 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07006704 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08006705 return 0;
6706
Mark Fasheh1d410a62007-09-07 14:20:45 -07006707 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08006708 sizeof(struct page *), GFP_NOFS);
6709 if (pages == NULL) {
6710 ret = -ENOMEM;
6711 mlog_errno(ret);
6712 goto out;
6713 }
6714
Mark Fasheh1d410a62007-09-07 14:20:45 -07006715 if (range_start == range_end)
6716 goto out;
6717
6718 ret = ocfs2_extent_map_get_blocks(inode,
6719 range_start >> sb->s_blocksize_bits,
6720 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08006721 if (ret) {
6722 mlog_errno(ret);
6723 goto out;
6724 }
6725
Mark Fasheh1d410a62007-09-07 14:20:45 -07006726 /*
6727 * Tail is a hole, or is marked unwritten. In either case, we
6728 * can count on read and write to return/push zero's.
6729 */
6730 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08006731 goto out;
6732
Mark Fasheh1d410a62007-09-07 14:20:45 -07006733 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6734 &numpages);
6735 if (ret) {
6736 mlog_errno(ret);
6737 goto out;
6738 }
6739
Mark Fasheh35edec12007-07-06 14:41:18 -07006740 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6741 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08006742
6743 /*
6744 * Initiate writeout of the pages we zero'd here. We don't
6745 * wait on them - the truncate_inode_pages() call later will
6746 * do that for us.
6747 */
Christoph Hellwig2cfd30a2009-09-23 15:04:02 +02006748 ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6749 range_end - 1);
Mark Fasheh60b11392007-02-16 11:46:50 -08006750 if (ret)
6751 mlog_errno(ret);
6752
6753out:
6754 if (pages)
6755 kfree(pages);
6756
6757 return ret;
6758}
6759
Tiger Yangfdd77702008-08-18 17:08:55 +08006760static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
6761 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006762{
6763 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08006764 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006765
Tiger Yangfdd77702008-08-18 17:08:55 +08006766 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
6767 memset(&di->id2, 0, blocksize -
6768 offsetof(struct ocfs2_dinode, id2) -
6769 xattrsize);
6770 else
6771 memset(&di->id2, 0, blocksize -
6772 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006773}
6774
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006775void ocfs2_dinode_new_extent_list(struct inode *inode,
6776 struct ocfs2_dinode *di)
6777{
Tiger Yangfdd77702008-08-18 17:08:55 +08006778 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006779 di->id2.i_list.l_tree_depth = 0;
6780 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08006781 di->id2.i_list.l_count = cpu_to_le16(
6782 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006783}
6784
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006785void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
6786{
6787 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6788 struct ocfs2_inline_data *idata = &di->id2.i_data;
6789
6790 spin_lock(&oi->ip_lock);
6791 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
6792 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6793 spin_unlock(&oi->ip_lock);
6794
6795 /*
6796 * We clear the entire i_data structure here so that all
6797 * fields can be properly initialized.
6798 */
Tiger Yangfdd77702008-08-18 17:08:55 +08006799 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006800
Tiger Yangfdd77702008-08-18 17:08:55 +08006801 idata->id_count = cpu_to_le16(
6802 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006803}
6804
6805int ocfs2_convert_inline_data_to_extents(struct inode *inode,
6806 struct buffer_head *di_bh)
6807{
6808 int ret, i, has_data, num_pages = 0;
6809 handle_t *handle;
6810 u64 uninitialized_var(block);
6811 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6812 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6813 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006814 struct ocfs2_alloc_context *data_ac = NULL;
6815 struct page **pages = NULL;
6816 loff_t end = osb->s_clustersize;
Joel Beckerf99b9b72008-08-20 19:36:33 -07006817 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02006818 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006819
6820 has_data = i_size_read(inode) ? 1 : 0;
6821
6822 if (has_data) {
6823 pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
6824 sizeof(struct page *), GFP_NOFS);
6825 if (pages == NULL) {
6826 ret = -ENOMEM;
6827 mlog_errno(ret);
6828 goto out;
6829 }
6830
6831 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
6832 if (ret) {
6833 mlog_errno(ret);
6834 goto out;
6835 }
6836 }
6837
Jan Karaa90714c2008-10-09 19:38:40 +02006838 handle = ocfs2_start_trans(osb,
6839 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006840 if (IS_ERR(handle)) {
6841 ret = PTR_ERR(handle);
6842 mlog_errno(ret);
6843 goto out_unlock;
6844 }
6845
Joel Becker0cf2f762009-02-12 16:41:25 -08006846 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07006847 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006848 if (ret) {
6849 mlog_errno(ret);
6850 goto out_commit;
6851 }
6852
6853 if (has_data) {
6854 u32 bit_off, num;
6855 unsigned int page_end;
6856 u64 phys;
6857
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006858 ret = dquot_alloc_space_nodirty(inode,
6859 ocfs2_clusters_to_bytes(osb->sb, 1));
6860 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02006861 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02006862 did_quota = 1;
6863
Mark Fasheh4fe370a2009-12-07 13:15:40 -08006864 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
6865
Joel Becker1ed9b772010-05-06 13:59:06 +08006866 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006867 &num);
6868 if (ret) {
6869 mlog_errno(ret);
6870 goto out_commit;
6871 }
6872
6873 /*
6874 * Save two copies, one for insert, and one that can
6875 * be changed by ocfs2_map_and_dirty_page() below.
6876 */
6877 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
6878
6879 /*
6880 * Non sparse file systems zero on extend, so no need
6881 * to do that now.
6882 */
6883 if (!ocfs2_sparse_alloc(osb) &&
6884 PAGE_CACHE_SIZE < osb->s_clustersize)
6885 end = PAGE_CACHE_SIZE;
6886
6887 ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
6888 if (ret) {
6889 mlog_errno(ret);
6890 goto out_commit;
6891 }
6892
6893 /*
6894 * This should populate the 1st page for us and mark
6895 * it up to date.
6896 */
6897 ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
6898 if (ret) {
6899 mlog_errno(ret);
6900 goto out_commit;
6901 }
6902
6903 page_end = PAGE_CACHE_SIZE;
6904 if (PAGE_CACHE_SIZE > osb->s_clustersize)
6905 page_end = osb->s_clustersize;
6906
6907 for (i = 0; i < num_pages; i++)
6908 ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
6909 pages[i], i > 0, &phys);
6910 }
6911
6912 spin_lock(&oi->ip_lock);
6913 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
6914 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6915 spin_unlock(&oi->ip_lock);
6916
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006917 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006918
6919 ocfs2_journal_dirty(handle, di_bh);
6920
6921 if (has_data) {
6922 /*
6923 * An error at this point should be extremely rare. If
6924 * this proves to be false, we could always re-build
6925 * the in-inode data from our pages.
6926 */
Joel Becker5e404e92009-02-13 03:54:22 -08006927 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08006928 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006929 if (ret) {
6930 mlog_errno(ret);
6931 goto out_commit;
6932 }
6933
6934 inode->i_blocks = ocfs2_inode_sector_count(inode);
6935 }
6936
6937out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02006938 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006939 dquot_free_space_nodirty(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02006940 ocfs2_clusters_to_bytes(osb->sb, 1));
6941
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006942 ocfs2_commit_trans(osb, handle);
6943
6944out_unlock:
6945 if (data_ac)
6946 ocfs2_free_alloc_context(data_ac);
6947
6948out:
6949 if (pages) {
6950 ocfs2_unlock_and_free_pages(pages, num_pages);
6951 kfree(pages);
6952 }
6953
6954 return ret;
6955}
6956
Mark Fashehccd979b2005-12-15 14:31:24 -08006957/*
6958 * It is expected, that by the time you call this function,
6959 * inode->i_size and fe->i_size have been adjusted.
6960 *
6961 * WARNING: This will kfree the truncate context
6962 */
6963int ocfs2_commit_truncate(struct ocfs2_super *osb,
6964 struct inode *inode,
Tristan Ye78f94672010-05-11 17:54:42 +08006965 struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08006966{
Tristan Ye78f94672010-05-11 17:54:42 +08006967 int status = 0, i, flags = 0;
6968 u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
Tao Mabcbbb242009-08-18 11:29:12 +08006969 u64 blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08006970 struct ocfs2_extent_list *el;
Tristan Ye78f94672010-05-11 17:54:42 +08006971 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08006972 struct ocfs2_path *path = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08006973 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
6974 struct ocfs2_extent_list *root_el = &(di->id2.i_list);
6975 u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
6976 struct ocfs2_extent_tree et;
6977 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006978
Tristan Ye78f94672010-05-11 17:54:42 +08006979 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
6980 ocfs2_init_dealloc_ctxt(&dealloc);
6981
Mark Fashehdcd05382007-01-16 11:32:23 -08006982 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08006983 i_size_read(inode));
6984
Tristan Ye78f94672010-05-11 17:54:42 +08006985 path = ocfs2_new_path(di_bh, &di->id2.i_list,
Joel Becker13723d02008-10-17 19:25:01 -07006986 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08006987 if (!path) {
6988 status = -ENOMEM;
6989 mlog_errno(status);
6990 goto bail;
6991 }
Mark Fasheh83418972007-04-23 18:53:12 -07006992
6993 ocfs2_extent_map_trunc(inode, new_highest_cpos);
6994
Mark Fashehccd979b2005-12-15 14:31:24 -08006995start:
Mark Fashehdcd05382007-01-16 11:32:23 -08006996 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08006997 * Check that we still have allocation to delete.
6998 */
6999 if (OCFS2_I(inode)->ip_clusters == 0) {
7000 status = 0;
7001 goto bail;
7002 }
7003
7004 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007005 * Truncate always works against the rightmost tree branch.
7006 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007007 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007008 if (status) {
7009 mlog_errno(status);
7010 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007011 }
7012
Tao Maa09d09b2011-02-22 08:21:10 +08007013 trace_ocfs2_commit_truncate(
7014 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7015 new_highest_cpos,
7016 OCFS2_I(inode)->ip_clusters,
7017 path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08007018
7019 /*
7020 * By now, el will point to the extent list on the bottom most
7021 * portion of this tree. Only the tail record is considered in
7022 * each pass.
7023 *
7024 * We handle the following cases, in order:
7025 * - empty extent: delete the remaining branch
7026 * - remove the entire record
7027 * - remove a partial record
7028 * - no record needs to be removed (truncate has completed)
7029 */
7030 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007031 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7032 ocfs2_error(inode->i_sb,
7033 "Inode %llu has empty extent block at %llu\n",
7034 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7035 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7036 status = -EROFS;
7037 goto bail;
7038 }
7039
Mark Fashehccd979b2005-12-15 14:31:24 -08007040 i = le16_to_cpu(el->l_next_free_rec) - 1;
Tristan Ye78f94672010-05-11 17:54:42 +08007041 rec = &el->l_recs[i];
7042 flags = rec->e_flags;
7043 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7044
7045 if (i == 0 && ocfs2_is_empty_extent(rec)) {
7046 /*
7047 * Lower levels depend on this never happening, but it's best
7048 * to check it up here before changing the tree.
7049 */
7050 if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
7051 ocfs2_error(inode->i_sb, "Inode %lu has an empty "
7052 "extent record, depth %u\n", inode->i_ino,
7053 le16_to_cpu(root_el->l_tree_depth));
7054 status = -EROFS;
7055 goto bail;
7056 }
7057 trunc_cpos = le32_to_cpu(rec->e_cpos);
7058 trunc_len = 0;
7059 blkno = 0;
7060 } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7061 /*
7062 * Truncate entire record.
7063 */
7064 trunc_cpos = le32_to_cpu(rec->e_cpos);
7065 trunc_len = ocfs2_rec_clusters(el, rec);
7066 blkno = le64_to_cpu(rec->e_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007067 } else if (range > new_highest_cpos) {
Tristan Ye78f94672010-05-11 17:54:42 +08007068 /*
7069 * Partial truncate. it also should be
7070 * the last truncate we're doing.
7071 */
7072 trunc_cpos = new_highest_cpos;
7073 trunc_len = range - new_highest_cpos;
7074 coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7075 blkno = le64_to_cpu(rec->e_blkno) +
7076 ocfs2_clusters_to_blocks(inode->i_sb, coff);
Mark Fashehdcd05382007-01-16 11:32:23 -08007077 } else {
Tristan Ye78f94672010-05-11 17:54:42 +08007078 /*
7079 * Truncate completed, leave happily.
7080 */
Mark Fashehdcd05382007-01-16 11:32:23 -08007081 status = 0;
7082 goto bail;
7083 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007084
Tristan Ye78f94672010-05-11 17:54:42 +08007085 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007086
Tristan Ye78f94672010-05-11 17:54:42 +08007087 status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7088 phys_cpos, trunc_len, flags, &dealloc,
7089 refcount_loc);
Mark Fashehccd979b2005-12-15 14:31:24 -08007090 if (status < 0) {
7091 mlog_errno(status);
7092 goto bail;
7093 }
7094
Mark Fashehdcd05382007-01-16 11:32:23 -08007095 ocfs2_reinit_path(path, 1);
7096
7097 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007098 * The check above will catch the case where we've truncated
7099 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007100 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007101 goto start;
7102
Mark Fashehccd979b2005-12-15 14:31:24 -08007103bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08007104
7105 ocfs2_schedule_truncate_log_flush(osb, 1);
7106
Tristan Ye78f94672010-05-11 17:54:42 +08007107 ocfs2_run_deallocs(osb, &dealloc);
Mark Fasheh59a5e412007-06-22 15:52:36 -07007108
Mark Fashehdcd05382007-01-16 11:32:23 -08007109 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007110
Mark Fashehccd979b2005-12-15 14:31:24 -08007111 return status;
7112}
7113
Mark Fashehccd979b2005-12-15 14:31:24 -08007114/*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007115 * 'start' is inclusive, 'end' is not.
7116 */
7117int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7118 unsigned int start, unsigned int end, int trunc)
7119{
7120 int ret;
7121 unsigned int numbytes;
7122 handle_t *handle;
7123 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7124 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7125 struct ocfs2_inline_data *idata = &di->id2.i_data;
7126
7127 if (end > i_size_read(inode))
7128 end = i_size_read(inode);
7129
7130 BUG_ON(start >= end);
7131
7132 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7133 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7134 !ocfs2_supports_inline_data(osb)) {
7135 ocfs2_error(inode->i_sb,
7136 "Inline data flags for inode %llu don't agree! "
7137 "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
7138 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7139 le16_to_cpu(di->i_dyn_features),
7140 OCFS2_I(inode)->ip_dyn_features,
7141 osb->s_feature_incompat);
7142 ret = -EROFS;
7143 goto out;
7144 }
7145
7146 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7147 if (IS_ERR(handle)) {
7148 ret = PTR_ERR(handle);
7149 mlog_errno(ret);
7150 goto out;
7151 }
7152
Joel Becker0cf2f762009-02-12 16:41:25 -08007153 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007154 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007155 if (ret) {
7156 mlog_errno(ret);
7157 goto out_commit;
7158 }
7159
7160 numbytes = end - start;
7161 memset(idata->id_data + start, 0, numbytes);
7162
7163 /*
7164 * No need to worry about the data page here - it's been
7165 * truncated already and inline data doesn't need it for
7166 * pushing zero's to disk, so we'll let readpage pick it up
7167 * later.
7168 */
7169 if (trunc) {
7170 i_size_write(inode, start);
7171 di->i_size = cpu_to_le64(start);
7172 }
7173
7174 inode->i_blocks = ocfs2_inode_sector_count(inode);
7175 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7176
7177 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7178 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7179
7180 ocfs2_journal_dirty(handle, di_bh);
7181
7182out_commit:
7183 ocfs2_commit_trans(osb, handle);
7184
7185out:
7186 return ret;
7187}
Tao Mae80de362011-05-23 10:36:43 +08007188
7189static int ocfs2_trim_extent(struct super_block *sb,
7190 struct ocfs2_group_desc *gd,
7191 u32 start, u32 count)
7192{
7193 u64 discard, bcount;
7194
7195 bcount = ocfs2_clusters_to_blocks(sb, count);
7196 discard = le64_to_cpu(gd->bg_blkno) +
7197 ocfs2_clusters_to_blocks(sb, start);
7198
Tao Ma10fca352011-05-23 10:36:45 +08007199 trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7200
Tao Mae80de362011-05-23 10:36:43 +08007201 return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7202}
7203
7204static int ocfs2_trim_group(struct super_block *sb,
7205 struct ocfs2_group_desc *gd,
7206 u32 start, u32 max, u32 minbits)
7207{
7208 int ret = 0, count = 0, next;
7209 void *bitmap = gd->bg_bitmap;
7210
7211 if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7212 return 0;
7213
Tao Ma10fca352011-05-23 10:36:45 +08007214 trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7215 start, max, minbits);
7216
Tao Mae80de362011-05-23 10:36:43 +08007217 while (start < max) {
7218 start = ocfs2_find_next_zero_bit(bitmap, max, start);
7219 if (start >= max)
7220 break;
7221 next = ocfs2_find_next_bit(bitmap, max, start);
7222
7223 if ((next - start) >= minbits) {
7224 ret = ocfs2_trim_extent(sb, gd,
7225 start, next - start);
7226 if (ret < 0) {
7227 mlog_errno(ret);
7228 break;
7229 }
7230 count += next - start;
7231 }
7232 start = next + 1;
7233
7234 if (fatal_signal_pending(current)) {
7235 count = -ERESTARTSYS;
7236 break;
7237 }
7238
7239 if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7240 break;
7241 }
7242
7243 if (ret < 0)
7244 count = ret;
7245
7246 return count;
7247}
7248
7249int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7250{
7251 struct ocfs2_super *osb = OCFS2_SB(sb);
7252 u64 start, len, trimmed, first_group, last_group, group;
7253 int ret, cnt;
7254 u32 first_bit, last_bit, minlen;
7255 struct buffer_head *main_bm_bh = NULL;
7256 struct inode *main_bm_inode = NULL;
7257 struct buffer_head *gd_bh = NULL;
7258 struct ocfs2_dinode *main_bm;
7259 struct ocfs2_group_desc *gd = NULL;
7260
7261 start = range->start >> osb->s_clustersize_bits;
7262 len = range->len >> osb->s_clustersize_bits;
7263 minlen = range->minlen >> osb->s_clustersize_bits;
7264 trimmed = 0;
7265
7266 if (!len) {
7267 range->len = 0;
7268 return 0;
7269 }
7270
7271 if (minlen >= osb->bitmap_cpg)
7272 return -EINVAL;
7273
7274 main_bm_inode = ocfs2_get_system_file_inode(osb,
7275 GLOBAL_BITMAP_SYSTEM_INODE,
7276 OCFS2_INVALID_SLOT);
7277 if (!main_bm_inode) {
7278 ret = -EIO;
7279 mlog_errno(ret);
7280 goto out;
7281 }
7282
7283 mutex_lock(&main_bm_inode->i_mutex);
7284
7285 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7286 if (ret < 0) {
7287 mlog_errno(ret);
7288 goto out_mutex;
7289 }
7290 main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7291
7292 if (start >= le32_to_cpu(main_bm->i_clusters)) {
7293 ret = -EINVAL;
7294 goto out_unlock;
7295 }
7296
7297 if (start + len > le32_to_cpu(main_bm->i_clusters))
7298 len = le32_to_cpu(main_bm->i_clusters) - start;
7299
Tao Ma10fca352011-05-23 10:36:45 +08007300 trace_ocfs2_trim_fs(start, len, minlen);
7301
Tao Mae80de362011-05-23 10:36:43 +08007302 /* Determine first and last group to examine based on start and len */
7303 first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7304 if (first_group == osb->first_cluster_group_blkno)
7305 first_bit = start;
7306 else
7307 first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
7308 last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
7309 last_bit = osb->bitmap_cpg;
7310
7311 for (group = first_group; group <= last_group;) {
7312 if (first_bit + len >= osb->bitmap_cpg)
7313 last_bit = osb->bitmap_cpg;
7314 else
7315 last_bit = first_bit + len;
7316
7317 ret = ocfs2_read_group_descriptor(main_bm_inode,
7318 main_bm, group,
7319 &gd_bh);
7320 if (ret < 0) {
7321 mlog_errno(ret);
7322 break;
7323 }
7324
7325 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
7326 cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
7327 brelse(gd_bh);
7328 gd_bh = NULL;
7329 if (cnt < 0) {
7330 ret = cnt;
7331 mlog_errno(ret);
7332 break;
7333 }
7334
7335 trimmed += cnt;
7336 len -= osb->bitmap_cpg - first_bit;
7337 first_bit = 0;
7338 if (group == osb->first_cluster_group_blkno)
7339 group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7340 else
7341 group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7342 }
7343 range->len = trimmed * sb->s_blocksize;
7344out_unlock:
7345 ocfs2_inode_unlock(main_bm_inode, 0);
7346 brelse(main_bm_bh);
7347out_mutex:
7348 mutex_unlock(&main_bm_inode->i_mutex);
7349 iput(main_bm_inode);
7350out:
7351 return ret;
7352}