blob: 2d7f76e52c379cf04b7f820c610c171e491d4830 [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) {
Rui Xiang7391a292013-11-12 15:06:54 -08001028 status = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -08001029 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);
Al Viro3a251f02012-04-13 12:22:00 -04001137 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1];
Tao Ma6b791bc2009-06-12 14:18:36 +08001138
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);
Markus Elfring629a3b52015-04-14 15:42:53 -07003373 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003374 }
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:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003456 ocfs2_free_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003457 return ret;
3458}
3459
Joel Becker4fe82c32009-02-13 02:16:08 -08003460static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003461 struct ocfs2_path *right_path,
3462 struct ocfs2_path **ret_left_path)
3463{
3464 int ret;
3465 u32 left_cpos;
3466 struct ocfs2_path *left_path = NULL;
3467
3468 *ret_left_path = NULL;
3469
3470 /* This function shouldn't be called for non-trees. */
3471 BUG_ON(right_path->p_tree_depth == 0);
3472
Joel Becker4fe82c32009-02-13 02:16:08 -08003473 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003474 right_path, &left_cpos);
3475 if (ret) {
3476 mlog_errno(ret);
3477 goto out;
3478 }
3479
3480 /* This function shouldn't be called for the leftmost leaf. */
3481 BUG_ON(left_cpos == 0);
3482
Joel Beckerffdd7a52008-10-17 22:32:01 -07003483 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003484 if (!left_path) {
3485 ret = -ENOMEM;
3486 mlog_errno(ret);
3487 goto out;
3488 }
3489
Joel Becker4fe82c32009-02-13 02:16:08 -08003490 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003491 if (ret) {
3492 mlog_errno(ret);
3493 goto out;
3494 }
3495
3496 *ret_left_path = left_path;
3497out:
3498 if (ret)
3499 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003500 return ret;
3501}
3502
3503/*
3504 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003505 * onto the tail of the record "before" it.
3506 * For index > 0, the "before" means the extent rec at index - 1.
3507 *
3508 * For index == 0, the "before" means the last record of the previous
3509 * extent block. And there is also a situation that we may need to
3510 * remove the rightmost leaf extent block in the right_path and change
3511 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003512 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003513static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003514 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003515 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003516 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003517 struct ocfs2_cached_dealloc_ctxt *dealloc,
3518 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003519{
Tao Ma677b9752008-01-30 14:21:05 +08003520 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003521 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3522 struct ocfs2_extent_rec *left_rec;
3523 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003524 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3525 struct buffer_head *bh = path_leaf_bh(right_path);
3526 struct buffer_head *root_bh = NULL;
3527 struct ocfs2_path *left_path = NULL;
3528 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003529
Tao Ma677b9752008-01-30 14:21:05 +08003530 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003531
Mark Fasheh328d5752007-06-18 10:48:04 -07003532 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003533 if (index == 0) {
3534 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003535 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003536 if (ret) {
3537 mlog_errno(ret);
Markus Elfring992ef6e2015-04-14 15:42:51 -07003538 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003539 }
3540
3541 left_el = path_leaf_el(left_path);
3542 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3543 le16_to_cpu(left_el->l_count));
3544
3545 left_rec = &left_el->l_recs[
3546 le16_to_cpu(left_el->l_next_free_rec) - 1];
3547 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3548 le16_to_cpu(left_rec->e_leaf_clusters) !=
3549 le32_to_cpu(split_rec->e_cpos));
3550
Joel Becker7dc02802009-02-12 19:20:13 -08003551 subtree_index = ocfs2_find_subtree_root(et, left_path,
3552 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003553
3554 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3555 handle->h_buffer_credits,
3556 left_path);
3557 if (ret) {
3558 mlog_errno(ret);
3559 goto out;
3560 }
3561
3562 root_bh = left_path->p_node[subtree_index].bh;
3563 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3564
Joel Becker4fe82c32009-02-13 02:16:08 -08003565 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003566 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003567 if (ret) {
3568 mlog_errno(ret);
3569 goto out;
3570 }
3571
3572 for (i = subtree_index + 1;
3573 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003574 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003575 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003576 if (ret) {
3577 mlog_errno(ret);
3578 goto out;
3579 }
3580
Joel Becker4fe82c32009-02-13 02:16:08 -08003581 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003582 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003583 if (ret) {
3584 mlog_errno(ret);
3585 goto out;
3586 }
3587 }
3588 } else {
3589 left_rec = &el->l_recs[index - 1];
3590 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3591 has_empty_extent = 1;
3592 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003593
Joel Becker4fe82c32009-02-13 02:16:08 -08003594 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003595 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003596 if (ret) {
3597 mlog_errno(ret);
3598 goto out;
3599 }
3600
3601 if (has_empty_extent && index == 1) {
3602 /*
3603 * The easy case - we can just plop the record right in.
3604 */
3605 *left_rec = *split_rec;
3606
3607 has_empty_extent = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003608 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003609 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003610
3611 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3612 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003613 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3614 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003615 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3616
3617 ocfs2_cleanup_merge(el, index);
3618
Joel Beckerec20cec2010-03-19 14:13:52 -07003619 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003620 if (left_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003621 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
Tao Ma677b9752008-01-30 14:21:05 +08003622
3623 /*
3624 * In the situation that the right_rec is empty and the extent
3625 * block is empty also, ocfs2_complete_edge_insert can't handle
3626 * it and we need to delete the right extent block.
3627 */
3628 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3629 le16_to_cpu(el->l_next_free_rec) == 1) {
3630
Joel Becker70f18c02009-02-13 02:09:31 -08003631 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003632 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003633 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003634 if (ret) {
3635 mlog_errno(ret);
3636 goto out;
3637 }
3638
3639 /* Now the rightmost extent block has been deleted.
3640 * So we use the new rightmost path.
3641 */
3642 ocfs2_mv_path(right_path, left_path);
3643 left_path = NULL;
3644 } else
Joel Becker4619c732009-02-12 19:02:36 -08003645 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003646 right_path, subtree_index);
3647 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003648out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003649 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003650 return ret;
3651}
3652
Joel Beckerc495dd22009-02-13 02:19:11 -08003653static int ocfs2_try_to_merge_extent(handle_t *handle,
3654 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003655 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003656 int split_index,
3657 struct ocfs2_extent_rec *split_rec,
3658 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003659 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003660{
Tao Mao518d7262007-08-28 17:25:35 -07003661 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003662 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003663 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3664
3665 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3666
Tao Mao518d7262007-08-28 17:25:35 -07003667 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
3668 /*
3669 * The merge code will need to create an empty
3670 * extent to take the place of the newly
3671 * emptied slot. Remove any pre-existing empty
3672 * extents - having more than one in a leaf is
3673 * illegal.
3674 */
Joel Becker70f18c02009-02-13 02:09:31 -08003675 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003676 if (ret) {
3677 mlog_errno(ret);
3678 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003679 }
Tao Mao518d7262007-08-28 17:25:35 -07003680 split_index--;
3681 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003682 }
3683
3684 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3685 /*
3686 * Left-right contig implies this.
3687 */
3688 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003689
3690 /*
3691 * Since the leftright insert always covers the entire
3692 * extent, this call will delete the insert record
3693 * entirely, resulting in an empty extent record added to
3694 * the extent block.
3695 *
3696 * Since the adding of an empty extent shifts
3697 * everything back to the right, there's no need to
3698 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003699 *
3700 * When the split_index is zero, we need to merge it to the
3701 * prevoius extent block. It is more efficient and easier
3702 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003703 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003704 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003705 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003706 if (ret) {
3707 mlog_errno(ret);
3708 goto out;
3709 }
3710
3711 /*
3712 * We can only get this from logic error above.
3713 */
3714 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3715
Tao Ma677b9752008-01-30 14:21:05 +08003716 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003717 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003718 if (ret) {
3719 mlog_errno(ret);
3720 goto out;
3721 }
Tao Ma677b9752008-01-30 14:21:05 +08003722
Mark Fasheh328d5752007-06-18 10:48:04 -07003723 rec = &el->l_recs[split_index];
3724
3725 /*
3726 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003727 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003728 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003729 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3730 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003731
Mark Fasheh328d5752007-06-18 10:48:04 -07003732 if (ret) {
3733 mlog_errno(ret);
3734 goto out;
3735 }
3736
Joel Becker70f18c02009-02-13 02:09:31 -08003737 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003738 /*
3739 * Error from this last rotate is not critical, so
3740 * print but don't bubble it up.
3741 */
3742 if (ret)
3743 mlog_errno(ret);
3744 ret = 0;
3745 } else {
3746 /*
3747 * Merge a record to the left or right.
3748 *
3749 * 'contig_type' is relative to the existing record,
3750 * so for example, if we're "right contig", it's to
3751 * the record on the left (hence the left merge).
3752 */
3753 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003754 ret = ocfs2_merge_rec_left(path, handle, et,
3755 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003756 split_index);
3757 if (ret) {
3758 mlog_errno(ret);
3759 goto out;
3760 }
3761 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003762 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003763 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003764 split_index);
3765 if (ret) {
3766 mlog_errno(ret);
3767 goto out;
3768 }
3769 }
3770
3771 if (ctxt->c_split_covers_rec) {
3772 /*
3773 * The merge may have left an empty extent in
3774 * our leaf. Try to rotate it away.
3775 */
Joel Becker70f18c02009-02-13 02:09:31 -08003776 ret = ocfs2_rotate_tree_left(handle, et, path,
3777 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003778 if (ret)
3779 mlog_errno(ret);
3780 ret = 0;
3781 }
3782 }
3783
3784out:
3785 return ret;
3786}
3787
3788static void ocfs2_subtract_from_rec(struct super_block *sb,
3789 enum ocfs2_split_type split,
3790 struct ocfs2_extent_rec *rec,
3791 struct ocfs2_extent_rec *split_rec)
3792{
3793 u64 len_blocks;
3794
3795 len_blocks = ocfs2_clusters_to_blocks(sb,
3796 le16_to_cpu(split_rec->e_leaf_clusters));
3797
3798 if (split == SPLIT_LEFT) {
3799 /*
3800 * Region is on the left edge of the existing
3801 * record.
3802 */
3803 le32_add_cpu(&rec->e_cpos,
3804 le16_to_cpu(split_rec->e_leaf_clusters));
3805 le64_add_cpu(&rec->e_blkno, len_blocks);
3806 le16_add_cpu(&rec->e_leaf_clusters,
3807 -le16_to_cpu(split_rec->e_leaf_clusters));
3808 } else {
3809 /*
3810 * Region is on the right edge of the existing
3811 * record.
3812 */
3813 le16_add_cpu(&rec->e_leaf_clusters,
3814 -le16_to_cpu(split_rec->e_leaf_clusters));
3815 }
3816}
3817
Mark Fashehdcd05382007-01-16 11:32:23 -08003818/*
3819 * Do the final bits of extent record insertion at the target leaf
3820 * list. If this leaf is part of an allocation tree, it is assumed
3821 * that the tree above has been prepared.
3822 */
Joel Beckerd5628622009-02-13 02:54:36 -08003823static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3824 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003825 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003826 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003827{
3828 int i = insert->ins_contig_index;
3829 unsigned int range;
3830 struct ocfs2_extent_rec *rec;
3831
Mark Fashehe48edee2007-03-07 16:46:57 -08003832 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003833
Mark Fasheh328d5752007-06-18 10:48:04 -07003834 if (insert->ins_split != SPLIT_NONE) {
3835 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3836 BUG_ON(i == -1);
3837 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003838 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3839 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003840 insert_rec);
3841 goto rotate;
3842 }
3843
Mark Fashehdcd05382007-01-16 11:32:23 -08003844 /*
3845 * Contiguous insert - either left or right.
3846 */
3847 if (insert->ins_contig != CONTIG_NONE) {
3848 rec = &el->l_recs[i];
3849 if (insert->ins_contig == CONTIG_LEFT) {
3850 rec->e_blkno = insert_rec->e_blkno;
3851 rec->e_cpos = insert_rec->e_cpos;
3852 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003853 le16_add_cpu(&rec->e_leaf_clusters,
3854 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003855 return;
3856 }
3857
3858 /*
3859 * Handle insert into an empty leaf.
3860 */
3861 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3862 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3863 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3864 el->l_recs[0] = *insert_rec;
3865 el->l_next_free_rec = cpu_to_le16(1);
3866 return;
3867 }
3868
3869 /*
3870 * Appending insert.
3871 */
3872 if (insert->ins_appending == APPEND_TAIL) {
3873 i = le16_to_cpu(el->l_next_free_rec) - 1;
3874 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003875 range = le32_to_cpu(rec->e_cpos)
3876 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003877 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3878
3879 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3880 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003881 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003882 "rec.cpos %u, rec.clusters %u, "
3883 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003884 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003885 le16_to_cpu(el->l_tree_depth),
3886 le16_to_cpu(el->l_count),
3887 le16_to_cpu(el->l_next_free_rec),
3888 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003889 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003890 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003891 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003892 i++;
3893 el->l_recs[i] = *insert_rec;
3894 le16_add_cpu(&el->l_next_free_rec, 1);
3895 return;
3896 }
3897
Mark Fasheh328d5752007-06-18 10:48:04 -07003898rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003899 /*
3900 * Ok, we have to rotate.
3901 *
3902 * At this point, it is safe to assume that inserting into an
3903 * empty leaf and appending to a leaf have both been handled
3904 * above.
3905 *
3906 * This leaf needs to have space, either by the empty 1st
3907 * extent record, or by virtue of an l_next_rec < l_count.
3908 */
3909 ocfs2_rotate_leaf(el, insert_rec);
3910}
3911
Joel Beckerd401dc12009-02-13 02:24:10 -08003912static void ocfs2_adjust_rightmost_records(handle_t *handle,
3913 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003914 struct ocfs2_path *path,
3915 struct ocfs2_extent_rec *insert_rec)
3916{
3917 int ret, i, next_free;
3918 struct buffer_head *bh;
3919 struct ocfs2_extent_list *el;
3920 struct ocfs2_extent_rec *rec;
3921
3922 /*
3923 * Update everything except the leaf block.
3924 */
3925 for (i = 0; i < path->p_tree_depth; i++) {
3926 bh = path->p_node[i].bh;
3927 el = path->p_node[i].el;
3928
3929 next_free = le16_to_cpu(el->l_next_free_rec);
3930 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003931 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3932 "Owner %llu has a bad extent list",
3933 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003934 ret = -EIO;
3935 return;
3936 }
3937
3938 rec = &el->l_recs[next_free - 1];
3939
3940 rec->e_int_clusters = insert_rec->e_cpos;
3941 le32_add_cpu(&rec->e_int_clusters,
3942 le16_to_cpu(insert_rec->e_leaf_clusters));
3943 le32_add_cpu(&rec->e_int_clusters,
3944 -le32_to_cpu(rec->e_cpos));
3945
Joel Beckerec20cec2010-03-19 14:13:52 -07003946 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07003947 }
3948}
3949
Joel Beckerd401dc12009-02-13 02:24:10 -08003950static int ocfs2_append_rec_to_path(handle_t *handle,
3951 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003952 struct ocfs2_extent_rec *insert_rec,
3953 struct ocfs2_path *right_path,
3954 struct ocfs2_path **ret_left_path)
3955{
Mark Fasheh328d5752007-06-18 10:48:04 -07003956 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003957 struct ocfs2_extent_list *el;
3958 struct ocfs2_path *left_path = NULL;
3959
3960 *ret_left_path = NULL;
3961
3962 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003963 * This shouldn't happen for non-trees. The extent rec cluster
3964 * count manipulation below only works for interior nodes.
3965 */
3966 BUG_ON(right_path->p_tree_depth == 0);
3967
3968 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08003969 * If our appending insert is at the leftmost edge of a leaf,
3970 * then we might need to update the rightmost records of the
3971 * neighboring path.
3972 */
3973 el = path_leaf_el(right_path);
3974 next_free = le16_to_cpu(el->l_next_free_rec);
3975 if (next_free == 0 ||
3976 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
3977 u32 left_cpos;
3978
Joel Beckerd401dc12009-02-13 02:24:10 -08003979 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3980 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003981 if (ret) {
3982 mlog_errno(ret);
3983 goto out;
3984 }
3985
Tao Maa09d09b2011-02-22 08:21:10 +08003986 trace_ocfs2_append_rec_to_path(
3987 (unsigned long long)
3988 ocfs2_metadata_cache_owner(et->et_ci),
3989 le32_to_cpu(insert_rec->e_cpos),
3990 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003991
3992 /*
3993 * No need to worry if the append is already in the
3994 * leftmost leaf.
3995 */
3996 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07003997 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08003998 if (!left_path) {
3999 ret = -ENOMEM;
4000 mlog_errno(ret);
4001 goto out;
4002 }
4003
Joel Beckerd401dc12009-02-13 02:24:10 -08004004 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004005 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004006 if (ret) {
4007 mlog_errno(ret);
4008 goto out;
4009 }
4010
4011 /*
4012 * ocfs2_insert_path() will pass the left_path to the
4013 * journal for us.
4014 */
4015 }
4016 }
4017
Joel Beckerd401dc12009-02-13 02:24:10 -08004018 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004019 if (ret) {
4020 mlog_errno(ret);
4021 goto out;
4022 }
4023
Joel Beckerd401dc12009-02-13 02:24:10 -08004024 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004025
4026 *ret_left_path = left_path;
4027 ret = 0;
4028out:
4029 if (ret != 0)
4030 ocfs2_free_path(left_path);
4031
4032 return ret;
4033}
4034
Joel Beckerc38e52b2009-02-13 02:56:23 -08004035static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004036 struct ocfs2_path *left_path,
4037 struct ocfs2_path *right_path,
4038 struct ocfs2_extent_rec *split_rec,
4039 enum ocfs2_split_type split)
4040{
4041 int index;
4042 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4043 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4044 struct ocfs2_extent_rec *rec, *tmprec;
4045
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004046 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004047 if (left_path)
4048 left_el = path_leaf_el(left_path);
4049
4050 el = right_el;
4051 insert_el = right_el;
4052 index = ocfs2_search_extent_list(el, cpos);
4053 if (index != -1) {
4054 if (index == 0 && left_path) {
4055 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4056
4057 /*
4058 * This typically means that the record
4059 * started in the left path but moved to the
4060 * right as a result of rotation. We either
4061 * move the existing record to the left, or we
4062 * do the later insert there.
4063 *
4064 * In this case, the left path should always
4065 * exist as the rotate code will have passed
4066 * it back for a post-insert update.
4067 */
4068
4069 if (split == SPLIT_LEFT) {
4070 /*
4071 * It's a left split. Since we know
4072 * that the rotate code gave us an
4073 * empty extent in the left path, we
4074 * can just do the insert there.
4075 */
4076 insert_el = left_el;
4077 } else {
4078 /*
4079 * Right split - we have to move the
4080 * existing record over to the left
4081 * leaf. The insert will be into the
4082 * newly created empty extent in the
4083 * right leaf.
4084 */
4085 tmprec = &right_el->l_recs[index];
4086 ocfs2_rotate_leaf(left_el, tmprec);
4087 el = left_el;
4088
4089 memset(tmprec, 0, sizeof(*tmprec));
4090 index = ocfs2_search_extent_list(left_el, cpos);
4091 BUG_ON(index == -1);
4092 }
4093 }
4094 } else {
4095 BUG_ON(!left_path);
4096 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4097 /*
4098 * Left path is easy - we can just allow the insert to
4099 * happen.
4100 */
4101 el = left_el;
4102 insert_el = left_el;
4103 index = ocfs2_search_extent_list(el, cpos);
4104 BUG_ON(index == -1);
4105 }
4106
4107 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004108 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4109 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004110 ocfs2_rotate_leaf(insert_el, split_rec);
4111}
4112
Mark Fashehdcd05382007-01-16 11:32:23 -08004113/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004114 * This function only does inserts on an allocation b-tree. For tree
4115 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004116 *
4117 * right_path is the path we want to do the actual insert
4118 * in. left_path should only be passed in if we need to update that
4119 * portion of the tree after an edge insert.
4120 */
Joel Becker3505bec2009-02-13 02:57:58 -08004121static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004122 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004123 struct ocfs2_path *left_path,
4124 struct ocfs2_path *right_path,
4125 struct ocfs2_extent_rec *insert_rec,
4126 struct ocfs2_insert_type *insert)
4127{
4128 int ret, subtree_index;
4129 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004130
Mark Fashehdcd05382007-01-16 11:32:23 -08004131 if (left_path) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004132 /*
4133 * There's a chance that left_path got passed back to
4134 * us without being accounted for in the
4135 * journal. Extend our transaction here to be sure we
4136 * can change those blocks.
4137 */
Tao Mac901fb02010-04-26 14:34:57 +08004138 ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08004139 if (ret < 0) {
4140 mlog_errno(ret);
4141 goto out;
4142 }
4143
Joel Becker7dc02802009-02-12 19:20:13 -08004144 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004145 if (ret < 0) {
4146 mlog_errno(ret);
4147 goto out;
4148 }
4149 }
4150
Mark Fashehe8aed342007-12-03 16:43:01 -08004151 /*
4152 * Pass both paths to the journal. The majority of inserts
4153 * will be touching all components anyway.
4154 */
Joel Becker7dc02802009-02-12 19:20:13 -08004155 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004156 if (ret < 0) {
4157 mlog_errno(ret);
4158 goto out;
4159 }
4160
Mark Fasheh328d5752007-06-18 10:48:04 -07004161 if (insert->ins_split != SPLIT_NONE) {
4162 /*
4163 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004164 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004165 * function sort it all out.
4166 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004167 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004168 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004169
4170 /*
4171 * Split might have modified either leaf and we don't
4172 * have a guarantee that the later edge insert will
4173 * dirty this for us.
4174 */
4175 if (left_path)
Joel Beckerec20cec2010-03-19 14:13:52 -07004176 ocfs2_journal_dirty(handle,
4177 path_leaf_bh(left_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07004178 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004179 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4180 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004181
Joel Beckerec20cec2010-03-19 14:13:52 -07004182 ocfs2_journal_dirty(handle, leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004183
4184 if (left_path) {
4185 /*
4186 * The rotate code has indicated that we need to fix
4187 * up portions of the tree after the insert.
4188 *
4189 * XXX: Should we extend the transaction here?
4190 */
Joel Becker7dc02802009-02-12 19:20:13 -08004191 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004192 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004193 ocfs2_complete_edge_insert(handle, left_path, right_path,
4194 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004195 }
4196
4197 ret = 0;
4198out:
4199 return ret;
4200}
4201
Joel Becker3505bec2009-02-13 02:57:58 -08004202static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004203 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004204 struct ocfs2_extent_rec *insert_rec,
4205 struct ocfs2_insert_type *type)
4206{
4207 int ret, rotate = 0;
4208 u32 cpos;
4209 struct ocfs2_path *right_path = NULL;
4210 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004211 struct ocfs2_extent_list *el;
4212
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004213 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004214
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004215 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004216 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004217 if (ret) {
4218 mlog_errno(ret);
4219 goto out;
4220 }
4221
4222 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004223 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004224 goto out_update_clusters;
4225 }
4226
Joel Beckerffdd7a52008-10-17 22:32:01 -07004227 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004228 if (!right_path) {
4229 ret = -ENOMEM;
4230 mlog_errno(ret);
4231 goto out;
4232 }
4233
4234 /*
4235 * Determine the path to start with. Rotations need the
4236 * rightmost path, everything else can go directly to the
4237 * target leaf.
4238 */
4239 cpos = le32_to_cpu(insert_rec->e_cpos);
4240 if (type->ins_appending == APPEND_NONE &&
4241 type->ins_contig == CONTIG_NONE) {
4242 rotate = 1;
4243 cpos = UINT_MAX;
4244 }
4245
Joel Beckerfacdb772009-02-12 18:08:48 -08004246 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004247 if (ret) {
4248 mlog_errno(ret);
4249 goto out;
4250 }
4251
4252 /*
4253 * Rotations and appends need special treatment - they modify
4254 * parts of the tree's above them.
4255 *
4256 * Both might pass back a path immediate to the left of the
4257 * one being inserted to. This will be cause
4258 * ocfs2_insert_path() to modify the rightmost records of
4259 * left_path to account for an edge insert.
4260 *
4261 * XXX: When modifying this code, keep in mind that an insert
4262 * can wind up skipping both of these two special cases...
4263 */
4264 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004265 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004266 le32_to_cpu(insert_rec->e_cpos),
4267 right_path, &left_path);
4268 if (ret) {
4269 mlog_errno(ret);
4270 goto out;
4271 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004272
4273 /*
4274 * ocfs2_rotate_tree_right() might have extended the
4275 * transaction without re-journaling our tree root.
4276 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004277 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004278 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004279 if (ret) {
4280 mlog_errno(ret);
4281 goto out;
4282 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004283 } else if (type->ins_appending == APPEND_TAIL
4284 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004285 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004286 right_path, &left_path);
4287 if (ret) {
4288 mlog_errno(ret);
4289 goto out;
4290 }
4291 }
4292
Joel Becker3505bec2009-02-13 02:57:58 -08004293 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004294 insert_rec, type);
4295 if (ret) {
4296 mlog_errno(ret);
4297 goto out;
4298 }
4299
4300out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004301 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004302 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004303 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004304
Joel Beckerec20cec2010-03-19 14:13:52 -07004305 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004306
4307out:
4308 ocfs2_free_path(left_path);
4309 ocfs2_free_path(right_path);
4310
4311 return ret;
4312}
4313
Mark Fasheh328d5752007-06-18 10:48:04 -07004314static enum ocfs2_contig_type
Joel Beckera2970292009-02-13 03:09:54 -08004315ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
4316 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004317 struct ocfs2_extent_list *el, int index,
4318 struct ocfs2_extent_rec *split_rec)
4319{
Tao Maad5a4d72008-01-30 14:21:32 +08004320 int status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004321 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004322 u32 left_cpos, right_cpos;
4323 struct ocfs2_extent_rec *rec = NULL;
4324 struct ocfs2_extent_list *new_el;
4325 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4326 struct buffer_head *bh;
4327 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004328 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004329
4330 if (index > 0) {
4331 rec = &el->l_recs[index - 1];
4332 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004333 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004334 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004335 goto exit;
Tao Maad5a4d72008-01-30 14:21:32 +08004336
4337 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004338 left_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004339 if (!left_path)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004340 goto exit;
Tao Maad5a4d72008-01-30 14:21:32 +08004341
Joel Beckera2970292009-02-13 03:09:54 -08004342 status = ocfs2_find_path(et->et_ci, left_path,
4343 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004344 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004345 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004346
4347 new_el = path_leaf_el(left_path);
4348
4349 if (le16_to_cpu(new_el->l_next_free_rec) !=
4350 le16_to_cpu(new_el->l_count)) {
4351 bh = path_leaf_bh(left_path);
4352 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004353 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004354 "Extent block #%llu has an "
4355 "invalid l_next_free_rec of "
4356 "%d. It should have "
4357 "matched the l_count of %d",
4358 (unsigned long long)le64_to_cpu(eb->h_blkno),
4359 le16_to_cpu(new_el->l_next_free_rec),
4360 le16_to_cpu(new_el->l_count));
4361 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004362 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004363 }
4364 rec = &new_el->l_recs[
4365 le16_to_cpu(new_el->l_next_free_rec) - 1];
4366 }
4367 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004368
4369 /*
4370 * We're careful to check for an empty extent record here -
4371 * the merge code will know what to do if it sees one.
4372 */
Tao Maad5a4d72008-01-30 14:21:32 +08004373 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004374 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4375 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4376 ret = CONTIG_RIGHT;
4377 } else {
Tao Ma853a3a12009-08-18 11:22:18 +08004378 ret = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004379 }
4380 }
4381
Tao Maad5a4d72008-01-30 14:21:32 +08004382 rec = NULL;
4383 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4384 rec = &el->l_recs[index + 1];
4385 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4386 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004387 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004388 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004389 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004390
4391 if (right_cpos == 0)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004392 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004393
Joel Beckerffdd7a52008-10-17 22:32:01 -07004394 right_path = ocfs2_new_path_from_path(path);
Tao Maad5a4d72008-01-30 14:21:32 +08004395 if (!right_path)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004396 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004397
Joel Beckera2970292009-02-13 03:09:54 -08004398 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004399 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004400 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004401
4402 new_el = path_leaf_el(right_path);
4403 rec = &new_el->l_recs[0];
4404 if (ocfs2_is_empty_extent(rec)) {
4405 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4406 bh = path_leaf_bh(right_path);
4407 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004408 ocfs2_error(sb,
Joel Becker5e965812008-11-13 14:49:16 -08004409 "Extent block #%llu has an "
4410 "invalid l_next_free_rec of %d",
4411 (unsigned long long)le64_to_cpu(eb->h_blkno),
4412 le16_to_cpu(new_el->l_next_free_rec));
4413 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004414 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004415 }
4416 rec = &new_el->l_recs[1];
4417 }
4418 }
4419
4420 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004421 enum ocfs2_contig_type contig_type;
4422
Tao Ma853a3a12009-08-18 11:22:18 +08004423 contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004424
4425 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4426 ret = CONTIG_LEFTRIGHT;
4427 else if (ret == CONTIG_NONE)
4428 ret = contig_type;
4429 }
4430
Markus Elfring06a269cc2015-04-14 15:42:48 -07004431free_right_path:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07004432 ocfs2_free_path(right_path);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004433free_left_path:
4434 ocfs2_free_path(left_path);
4435exit:
Mark Fasheh328d5752007-06-18 10:48:04 -07004436 return ret;
4437}
4438
Joel Becker1ef61b32009-02-13 03:12:33 -08004439static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004440 struct ocfs2_insert_type *insert,
4441 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004442 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004443{
4444 int i;
4445 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4446
Mark Fashehe48edee2007-03-07 16:46:57 -08004447 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4448
Mark Fashehdcd05382007-01-16 11:32:23 -08004449 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Tao Ma853a3a12009-08-18 11:22:18 +08004450 contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4451 insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004452 if (contig_type != CONTIG_NONE) {
4453 insert->ins_contig_index = i;
4454 break;
4455 }
4456 }
4457 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004458
4459 if (insert->ins_contig != CONTIG_NONE) {
4460 struct ocfs2_extent_rec *rec =
4461 &el->l_recs[insert->ins_contig_index];
4462 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4463 le16_to_cpu(insert_rec->e_leaf_clusters);
4464
4465 /*
4466 * Caller might want us to limit the size of extents, don't
4467 * calculate contiguousness if we might exceed that limit.
4468 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004469 if (et->et_max_leaf_clusters &&
4470 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004471 insert->ins_contig = CONTIG_NONE;
4472 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004473}
4474
4475/*
4476 * This should only be called against the righmost leaf extent list.
4477 *
4478 * ocfs2_figure_appending_type() will figure out whether we'll have to
4479 * insert at the tail of the rightmost leaf.
4480 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004481 * This should also work against the root extent list for tree's with 0
4482 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004483 * then the logic here makes sense.
4484 */
4485static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4486 struct ocfs2_extent_list *el,
4487 struct ocfs2_extent_rec *insert_rec)
4488{
4489 int i;
4490 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4491 struct ocfs2_extent_rec *rec;
4492
4493 insert->ins_appending = APPEND_NONE;
4494
Mark Fashehe48edee2007-03-07 16:46:57 -08004495 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004496
4497 if (!el->l_next_free_rec)
4498 goto set_tail_append;
4499
4500 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4501 /* Were all records empty? */
4502 if (le16_to_cpu(el->l_next_free_rec) == 1)
4503 goto set_tail_append;
4504 }
4505
4506 i = le16_to_cpu(el->l_next_free_rec) - 1;
4507 rec = &el->l_recs[i];
4508
Mark Fashehe48edee2007-03-07 16:46:57 -08004509 if (cpos >=
4510 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004511 goto set_tail_append;
4512
4513 return;
4514
4515set_tail_append:
4516 insert->ins_appending = APPEND_TAIL;
4517}
4518
4519/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004520 * Helper function called at the beginning of an insert.
Mark Fashehdcd05382007-01-16 11:32:23 -08004521 *
4522 * This computes a few things that are commonly used in the process of
4523 * inserting into the btree:
4524 * - Whether the new extent is contiguous with an existing one.
4525 * - The current tree depth.
4526 * - Whether the insert is an appending one.
4527 * - The total # of free records in the tree.
4528 *
4529 * All of the information is stored on the ocfs2_insert_type
4530 * structure.
4531 */
Joel Becker627961b2009-02-13 03:14:38 -08004532static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004533 struct buffer_head **last_eb_bh,
4534 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004535 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004536 struct ocfs2_insert_type *insert)
4537{
4538 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004539 struct ocfs2_extent_block *eb;
4540 struct ocfs2_extent_list *el;
4541 struct ocfs2_path *path = NULL;
4542 struct buffer_head *bh = NULL;
4543
Mark Fasheh328d5752007-06-18 10:48:04 -07004544 insert->ins_split = SPLIT_NONE;
4545
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004546 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004547 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4548
4549 if (el->l_tree_depth) {
4550 /*
4551 * If we have tree depth, we read in the
4552 * rightmost extent block ahead of time as
4553 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4554 * may want it later.
4555 */
Joel Becker3d03a302009-02-12 17:49:26 -08004556 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004557 ocfs2_et_get_last_eb_blk(et),
4558 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004559 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08004560 mlog_errno(ret);
Mark Fashehdcd05382007-01-16 11:32:23 -08004561 goto out;
4562 }
4563 eb = (struct ocfs2_extent_block *) bh->b_data;
4564 el = &eb->h_list;
4565 }
4566
4567 /*
4568 * Unless we have a contiguous insert, we'll need to know if
4569 * there is room left in our allocation tree for another
4570 * extent record.
4571 *
4572 * XXX: This test is simplistic, we can search for empty
4573 * extent records too.
4574 */
Tao Maoc77534f2007-08-28 17:22:33 -07004575 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004576 le16_to_cpu(el->l_next_free_rec);
4577
4578 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004579 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004580 ocfs2_figure_appending_type(insert, el, insert_rec);
4581 return 0;
4582 }
4583
Joel Beckerffdd7a52008-10-17 22:32:01 -07004584 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004585 if (!path) {
4586 ret = -ENOMEM;
4587 mlog_errno(ret);
4588 goto out;
4589 }
4590
4591 /*
4592 * In the case that we're inserting past what the tree
4593 * currently accounts for, ocfs2_find_path() will return for
4594 * us the rightmost tree path. This is accounted for below in
4595 * the appending code.
4596 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004597 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004598 if (ret) {
4599 mlog_errno(ret);
4600 goto out;
4601 }
4602
4603 el = path_leaf_el(path);
4604
4605 /*
4606 * Now that we have the path, there's two things we want to determine:
4607 * 1) Contiguousness (also set contig_index if this is so)
4608 *
4609 * 2) Are we doing an append? We can trivially break this up
4610 * into two types of appends: simple record append, or a
4611 * rotate inside the tail leaf.
4612 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004613 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004614
4615 /*
4616 * The insert code isn't quite ready to deal with all cases of
4617 * left contiguousness. Specifically, if it's an insert into
4618 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004619 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004620 * left. For now, just catch that case and fool the layers
4621 * above us. This works just fine for tree_depth == 0, which
4622 * is why we allow that above.
4623 */
4624 if (insert->ins_contig == CONTIG_LEFT &&
4625 insert->ins_contig_index == 0)
4626 insert->ins_contig = CONTIG_NONE;
4627
4628 /*
4629 * Ok, so we can simply compare against last_eb to figure out
4630 * whether the path doesn't exist. This will only happen in
4631 * the case that we're doing a tail append, so maybe we can
4632 * take advantage of that information somehow.
4633 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004634 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004635 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004636 /*
4637 * Ok, ocfs2_find_path() returned us the rightmost
4638 * tree path. This might be an appending insert. There are
4639 * two cases:
4640 * 1) We're doing a true append at the tail:
4641 * -This might even be off the end of the leaf
4642 * 2) We're "appending" by rotating in the tail
4643 */
4644 ocfs2_figure_appending_type(insert, el, insert_rec);
4645 }
4646
4647out:
4648 ocfs2_free_path(path);
4649
4650 if (ret == 0)
4651 *last_eb_bh = bh;
4652 else
4653 brelse(bh);
4654 return ret;
4655}
4656
4657/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004658 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004659 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004660 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004661 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004662int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004663 struct ocfs2_extent_tree *et,
4664 u32 cpos,
4665 u64 start_blk,
4666 u32 new_clusters,
4667 u8 flags,
4668 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004669{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004670 int status;
Tao Maoc77534f2007-08-28 17:22:33 -07004671 int uninitialized_var(free_records);
Mark Fashehccd979b2005-12-15 14:31:24 -08004672 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004673 struct ocfs2_insert_type insert = {0, };
4674 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004675
Tao Maa09d09b2011-02-22 08:21:10 +08004676 trace_ocfs2_insert_extent_start(
4677 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4678 cpos, new_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08004679
Mark Fashehe48edee2007-03-07 16:46:57 -08004680 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004681 rec.e_cpos = cpu_to_le32(cpos);
4682 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004683 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004684 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004685 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004686 if (status) {
4687 mlog_errno(status);
4688 goto bail;
4689 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004690
Joel Becker627961b2009-02-13 03:14:38 -08004691 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004692 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004693 if (status < 0) {
4694 mlog_errno(status);
4695 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004696 }
4697
Tao Maa09d09b2011-02-22 08:21:10 +08004698 trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4699 insert.ins_contig_index, free_records,
4700 insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004701
Tao Maoc77534f2007-08-28 17:22:33 -07004702 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004703 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004704 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004705 meta_ac);
4706 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004707 mlog_errno(status);
4708 goto bail;
4709 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004710 }
4711
Mark Fashehdcd05382007-01-16 11:32:23 -08004712 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004713 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004714 if (status < 0)
4715 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004716 else
4717 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004718
4719bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004720 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004721
Tao Maf56654c2008-08-18 17:38:48 +08004722 return status;
4723}
4724
Tao Ma0eb8d472008-08-18 17:38:45 +08004725/*
4726 * Allcate and add clusters into the extent b-tree.
4727 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004728 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d472008-08-18 17:38:45 +08004729 * it is not limited to the file storage. Any extent tree can use this
4730 * function if it implements the proper ocfs2_extent_tree.
4731 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004732int ocfs2_add_clusters_in_btree(handle_t *handle,
4733 struct ocfs2_extent_tree *et,
Tao Ma0eb8d472008-08-18 17:38:45 +08004734 u32 *logical_offset,
4735 u32 clusters_to_add,
4736 int mark_unwritten,
Tao Ma0eb8d472008-08-18 17:38:45 +08004737 struct ocfs2_alloc_context *data_ac,
4738 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004739 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d472008-08-18 17:38:45 +08004740{
Tao Maa09d09b2011-02-22 08:21:10 +08004741 int status = 0, err = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004742 int need_free = 0;
Tao Ma0eb8d472008-08-18 17:38:45 +08004743 int free_extents;
4744 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4745 u32 bit_off, num_bits;
4746 u64 block;
4747 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004748 struct ocfs2_super *osb =
4749 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d472008-08-18 17:38:45 +08004750
4751 BUG_ON(!clusters_to_add);
4752
4753 if (mark_unwritten)
4754 flags = OCFS2_EXT_UNWRITTEN;
4755
Joel Becker3d03a302009-02-12 17:49:26 -08004756 free_extents = ocfs2_num_free_extents(osb, et);
Tao Ma0eb8d472008-08-18 17:38:45 +08004757 if (free_extents < 0) {
4758 status = free_extents;
4759 mlog_errno(status);
4760 goto leave;
4761 }
4762
4763 /* there are two cases which could cause us to EAGAIN in the
4764 * we-need-more-metadata case:
4765 * 1) we haven't reserved *any*
4766 * 2) we are so fragmented, we've needed to add metadata too
4767 * many times. */
4768 if (!free_extents && !meta_ac) {
Tao Maa09d09b2011-02-22 08:21:10 +08004769 err = -1;
Tao Ma0eb8d472008-08-18 17:38:45 +08004770 status = -EAGAIN;
4771 reason = RESTART_META;
4772 goto leave;
4773 } else if ((!free_extents)
4774 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004775 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Maa09d09b2011-02-22 08:21:10 +08004776 err = -2;
Tao Ma0eb8d472008-08-18 17:38:45 +08004777 status = -EAGAIN;
4778 reason = RESTART_META;
4779 goto leave;
4780 }
4781
Joel Becker1ed9b772010-05-06 13:59:06 +08004782 status = __ocfs2_claim_clusters(handle, data_ac, 1,
Tao Ma0eb8d472008-08-18 17:38:45 +08004783 clusters_to_add, &bit_off, &num_bits);
4784 if (status < 0) {
4785 if (status != -ENOSPC)
4786 mlog_errno(status);
4787 goto leave;
4788 }
4789
4790 BUG_ON(num_bits > clusters_to_add);
4791
Joel Becker13723d02008-10-17 19:25:01 -07004792 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004793 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004794 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d472008-08-18 17:38:45 +08004795 if (status < 0) {
4796 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004797 need_free = 1;
4798 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004799 }
4800
4801 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Maa09d09b2011-02-22 08:21:10 +08004802 trace_ocfs2_add_clusters_in_btree(
4803 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4804 bit_off, num_bits);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004805 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004806 num_bits, flags, meta_ac);
Tao Ma0eb8d472008-08-18 17:38:45 +08004807 if (status < 0) {
4808 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004809 need_free = 1;
4810 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004811 }
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
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004824bail:
4825 if (need_free) {
4826 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
4827 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
4828 bit_off, num_bits);
4829 else
4830 ocfs2_free_clusters(handle,
4831 data_ac->ac_inode,
4832 data_ac->ac_bh,
4833 ocfs2_clusters_to_blocks(osb->sb, bit_off),
4834 num_bits);
4835 }
4836
Tao Ma0eb8d472008-08-18 17:38:45 +08004837leave:
Tao Ma0eb8d472008-08-18 17:38:45 +08004838 if (reason_ret)
4839 *reason_ret = reason;
Tao Maa09d09b2011-02-22 08:21:10 +08004840 trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
Tao Ma0eb8d472008-08-18 17:38:45 +08004841 return status;
4842}
4843
Mark Fasheh328d5752007-06-18 10:48:04 -07004844static void ocfs2_make_right_split_rec(struct super_block *sb,
4845 struct ocfs2_extent_rec *split_rec,
4846 u32 cpos,
4847 struct ocfs2_extent_rec *rec)
4848{
4849 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4850 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4851
4852 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4853
4854 split_rec->e_cpos = cpu_to_le32(cpos);
4855 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4856
4857 split_rec->e_blkno = rec->e_blkno;
4858 le64_add_cpu(&split_rec->e_blkno,
4859 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4860
4861 split_rec->e_flags = rec->e_flags;
4862}
4863
Joel Beckerd2311292009-02-13 03:43:22 -08004864static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004865 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004866 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004867 struct buffer_head **last_eb_bh,
4868 int split_index,
4869 struct ocfs2_extent_rec *orig_split_rec,
4870 struct ocfs2_alloc_context *meta_ac)
4871{
4872 int ret = 0, depth;
4873 unsigned int insert_range, rec_range, do_leftright = 0;
4874 struct ocfs2_extent_rec tmprec;
4875 struct ocfs2_extent_list *rightmost_el;
4876 struct ocfs2_extent_rec rec;
4877 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4878 struct ocfs2_insert_type insert;
4879 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004880
4881leftright:
4882 /*
4883 * Store a copy of the record on the stack - it might move
4884 * around as the tree is manipulated below.
4885 */
4886 rec = path_leaf_el(path)->l_recs[split_index];
4887
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004888 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004889
4890 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4891 if (depth) {
4892 BUG_ON(!(*last_eb_bh));
4893 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4894 rightmost_el = &eb->h_list;
4895 }
4896
4897 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4898 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004899 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004900 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004901 if (ret) {
4902 mlog_errno(ret);
4903 goto out;
4904 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004905 }
4906
4907 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4908 insert.ins_appending = APPEND_NONE;
4909 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004910 insert.ins_tree_depth = depth;
4911
4912 insert_range = le32_to_cpu(split_rec.e_cpos) +
4913 le16_to_cpu(split_rec.e_leaf_clusters);
4914 rec_range = le32_to_cpu(rec.e_cpos) +
4915 le16_to_cpu(rec.e_leaf_clusters);
4916
4917 if (split_rec.e_cpos == rec.e_cpos) {
4918 insert.ins_split = SPLIT_LEFT;
4919 } else if (insert_range == rec_range) {
4920 insert.ins_split = SPLIT_RIGHT;
4921 } else {
4922 /*
4923 * Left/right split. We fake this as a right split
4924 * first and then make a second pass as a left split.
4925 */
4926 insert.ins_split = SPLIT_RIGHT;
4927
Joel Beckerd2311292009-02-13 03:43:22 -08004928 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4929 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004930
4931 split_rec = tmprec;
4932
4933 BUG_ON(do_leftright);
4934 do_leftright = 1;
4935 }
4936
Joel Becker3505bec2009-02-13 02:57:58 -08004937 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004938 if (ret) {
4939 mlog_errno(ret);
4940 goto out;
4941 }
4942
4943 if (do_leftright == 1) {
4944 u32 cpos;
4945 struct ocfs2_extent_list *el;
4946
4947 do_leftright++;
4948 split_rec = *orig_split_rec;
4949
4950 ocfs2_reinit_path(path, 1);
4951
4952 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004953 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004954 if (ret) {
4955 mlog_errno(ret);
4956 goto out;
4957 }
4958
4959 el = path_leaf_el(path);
4960 split_index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07004961 if (split_index == -1) {
4962 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
4963 "Owner %llu has an extent at cpos %u "
4964 "which can no longer be found.\n",
4965 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4966 cpos);
4967 ret = -EROFS;
4968 goto out;
4969 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004970 goto leftright;
4971 }
4972out:
4973
4974 return ret;
4975}
4976
Joel Beckerf3868d02009-02-17 19:46:04 -08004977static int ocfs2_replace_extent_rec(handle_t *handle,
4978 struct ocfs2_extent_tree *et,
Tao Ma47be12e2009-01-09 07:32:48 +08004979 struct ocfs2_path *path,
4980 struct ocfs2_extent_list *el,
4981 int split_index,
4982 struct ocfs2_extent_rec *split_rec)
4983{
4984 int ret;
4985
Joel Beckerf3868d02009-02-17 19:46:04 -08004986 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Tao Ma47be12e2009-01-09 07:32:48 +08004987 path_num_items(path) - 1);
4988 if (ret) {
4989 mlog_errno(ret);
4990 goto out;
4991 }
4992
4993 el->l_recs[split_index] = *split_rec;
4994
4995 ocfs2_journal_dirty(handle, path_leaf_bh(path));
4996out:
4997 return ret;
4998}
4999
Mark Fasheh328d5752007-06-18 10:48:04 -07005000/*
Tao Ma555936b2009-08-18 11:22:21 +08005001 * Split part or all of the extent record at split_index in the leaf
5002 * pointed to by path. Merge with the contiguous extent record if needed.
Mark Fasheh328d5752007-06-18 10:48:04 -07005003 *
5004 * Care is taken to handle contiguousness so as to not grow the tree.
5005 *
5006 * meta_ac is not strictly necessary - we only truly need it if growth
5007 * of the tree is required. All other cases will degrade into a less
5008 * optimal tree layout.
5009 *
Tao Mae7d4cb62008-08-18 17:38:44 +08005010 * last_eb_bh should be the rightmost leaf block for any extent
5011 * btree. Since a split may grow the tree or a merge might shrink it,
5012 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07005013 *
5014 * This code is optimized for readability - several passes might be
5015 * made over certain portions of the tree. All of those blocks will
5016 * have been brought into cache (and pinned via the journal), so the
5017 * extra overhead is not expressed in terms of disk reads.
5018 */
Tao Mae2e9f602009-08-18 11:22:34 +08005019int ocfs2_split_extent(handle_t *handle,
5020 struct ocfs2_extent_tree *et,
5021 struct ocfs2_path *path,
5022 int split_index,
5023 struct ocfs2_extent_rec *split_rec,
5024 struct ocfs2_alloc_context *meta_ac,
5025 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005026{
5027 int ret = 0;
5028 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005029 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005030 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5031 struct ocfs2_merge_ctxt ctxt;
5032 struct ocfs2_extent_list *rightmost_el;
5033
Mark Fasheh328d5752007-06-18 10:48:04 -07005034 if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5035 ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5036 (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5037 ret = -EIO;
5038 mlog_errno(ret);
5039 goto out;
5040 }
5041
Joel Beckera2970292009-02-13 03:09:54 -08005042 ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
Mark Fasheh328d5752007-06-18 10:48:04 -07005043 split_index,
5044 split_rec);
5045
5046 /*
5047 * The core merge / split code wants to know how much room is
Joel Beckera1cf0762009-02-13 03:45:49 -08005048 * left in this allocation tree, so we pass the
Mark Fasheh328d5752007-06-18 10:48:04 -07005049 * rightmost extent list.
5050 */
5051 if (path->p_tree_depth) {
5052 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07005053
Joel Becker3d03a302009-02-12 17:49:26 -08005054 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005055 ocfs2_et_get_last_eb_blk(et),
5056 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005057 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08005058 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07005059 goto out;
5060 }
5061
5062 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fasheh328d5752007-06-18 10:48:04 -07005063 rightmost_el = &eb->h_list;
5064 } else
5065 rightmost_el = path_root_el(path);
5066
Mark Fasheh328d5752007-06-18 10:48:04 -07005067 if (rec->e_cpos == split_rec->e_cpos &&
5068 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5069 ctxt.c_split_covers_rec = 1;
5070 else
5071 ctxt.c_split_covers_rec = 0;
5072
5073 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5074
Tao Maa09d09b2011-02-22 08:21:10 +08005075 trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5076 ctxt.c_has_empty_extent,
5077 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005078
5079 if (ctxt.c_contig_type == CONTIG_NONE) {
5080 if (ctxt.c_split_covers_rec)
Joel Beckerf3868d02009-02-17 19:46:04 -08005081 ret = ocfs2_replace_extent_rec(handle, et, path, el,
Tao Ma47be12e2009-01-09 07:32:48 +08005082 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005083 else
Joel Beckerd2311292009-02-13 03:43:22 -08005084 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005085 &last_eb_bh, split_index,
5086 split_rec, meta_ac);
5087 if (ret)
5088 mlog_errno(ret);
5089 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005090 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005091 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005092 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005093 if (ret)
5094 mlog_errno(ret);
5095 }
5096
Mark Fasheh328d5752007-06-18 10:48:04 -07005097out:
5098 brelse(last_eb_bh);
5099 return ret;
5100}
5101
5102/*
Tao Ma555936b2009-08-18 11:22:21 +08005103 * Change the flags of the already-existing extent at cpos for len clusters.
5104 *
5105 * new_flags: the flags we want to set.
5106 * clear_flags: the flags we want to clear.
5107 * phys: the new physical offset we want this new extent starts from.
Mark Fasheh328d5752007-06-18 10:48:04 -07005108 *
5109 * If the existing extent is larger than the request, initiate a
5110 * split. An attempt will be made at merging with adjacent extents.
5111 *
5112 * The caller is responsible for passing down meta_ac if we'll need it.
5113 */
Tao Ma1aa75fe2009-08-18 11:28:39 +08005114int ocfs2_change_extent_flag(handle_t *handle,
5115 struct ocfs2_extent_tree *et,
5116 u32 cpos, u32 len, u32 phys,
5117 struct ocfs2_alloc_context *meta_ac,
5118 struct ocfs2_cached_dealloc_ctxt *dealloc,
5119 int new_flags, int clear_flags)
Mark Fasheh328d5752007-06-18 10:48:04 -07005120{
5121 int ret, index;
Tao Ma555936b2009-08-18 11:22:21 +08005122 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5123 u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
Mark Fasheh328d5752007-06-18 10:48:04 -07005124 struct ocfs2_extent_rec split_rec;
5125 struct ocfs2_path *left_path = NULL;
5126 struct ocfs2_extent_list *el;
Tao Ma555936b2009-08-18 11:22:21 +08005127 struct ocfs2_extent_rec *rec;
Mark Fasheh328d5752007-06-18 10:48:04 -07005128
Joel Beckerffdd7a52008-10-17 22:32:01 -07005129 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005130 if (!left_path) {
5131 ret = -ENOMEM;
5132 mlog_errno(ret);
5133 goto out;
5134 }
5135
Joel Beckerfacdb772009-02-12 18:08:48 -08005136 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005137 if (ret) {
5138 mlog_errno(ret);
5139 goto out;
5140 }
5141 el = path_leaf_el(left_path);
5142
5143 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005144 if (index == -1) {
Tao Ma555936b2009-08-18 11:22:21 +08005145 ocfs2_error(sb,
5146 "Owner %llu has an extent at cpos %u which can no "
Mark Fasheh328d5752007-06-18 10:48:04 -07005147 "longer be found.\n",
Tao Ma555936b2009-08-18 11:22:21 +08005148 (unsigned long long)
5149 ocfs2_metadata_cache_owner(et->et_ci), cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005150 ret = -EROFS;
5151 goto out;
5152 }
5153
Tao Ma555936b2009-08-18 11:22:21 +08005154 ret = -EIO;
5155 rec = &el->l_recs[index];
5156 if (new_flags && (rec->e_flags & new_flags)) {
5157 mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
5158 "extent that already had them",
5159 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5160 new_flags);
5161 goto out;
5162 }
5163
5164 if (clear_flags && !(rec->e_flags & clear_flags)) {
5165 mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
5166 "extent that didn't have them",
5167 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5168 clear_flags);
5169 goto out;
5170 }
5171
Mark Fasheh328d5752007-06-18 10:48:04 -07005172 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5173 split_rec.e_cpos = cpu_to_le32(cpos);
5174 split_rec.e_leaf_clusters = cpu_to_le16(len);
5175 split_rec.e_blkno = cpu_to_le64(start_blkno);
Tao Ma555936b2009-08-18 11:22:21 +08005176 split_rec.e_flags = rec->e_flags;
5177 if (new_flags)
5178 split_rec.e_flags |= new_flags;
5179 if (clear_flags)
5180 split_rec.e_flags &= ~clear_flags;
Mark Fasheh328d5752007-06-18 10:48:04 -07005181
Tao Mae2e9f602009-08-18 11:22:34 +08005182 ret = ocfs2_split_extent(handle, et, left_path,
5183 index, &split_rec, meta_ac,
5184 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005185 if (ret)
5186 mlog_errno(ret);
5187
5188out:
5189 ocfs2_free_path(left_path);
5190 return ret;
Tao Ma555936b2009-08-18 11:22:21 +08005191
5192}
5193
5194/*
5195 * Mark the already-existing extent at cpos as written for len clusters.
5196 * This removes the unwritten extent flag.
5197 *
5198 * If the existing extent is larger than the request, initiate a
5199 * split. An attempt will be made at merging with adjacent extents.
5200 *
5201 * The caller is responsible for passing down meta_ac if we'll need it.
5202 */
5203int ocfs2_mark_extent_written(struct inode *inode,
5204 struct ocfs2_extent_tree *et,
5205 handle_t *handle, u32 cpos, u32 len, u32 phys,
5206 struct ocfs2_alloc_context *meta_ac,
5207 struct ocfs2_cached_dealloc_ctxt *dealloc)
5208{
5209 int ret;
5210
Tao Maa09d09b2011-02-22 08:21:10 +08005211 trace_ocfs2_mark_extent_written(
5212 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5213 cpos, len, phys);
Tao Ma555936b2009-08-18 11:22:21 +08005214
5215 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
5216 ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
5217 "that are being written to, but the feature bit "
5218 "is not set in the super block.",
5219 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5220 ret = -EROFS;
5221 goto out;
5222 }
5223
5224 /*
5225 * XXX: This should be fixed up so that we just re-insert the
5226 * next extent records.
5227 */
5228 ocfs2_et_extent_map_truncate(et, 0);
5229
5230 ret = ocfs2_change_extent_flag(handle, et, cpos,
5231 len, phys, meta_ac, dealloc,
5232 0, OCFS2_EXT_UNWRITTEN);
5233 if (ret)
5234 mlog_errno(ret);
5235
5236out:
5237 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07005238}
5239
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005240static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5241 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005242 int index, u32 new_range,
5243 struct ocfs2_alloc_context *meta_ac)
5244{
Tao Mac901fb02010-04-26 14:34:57 +08005245 int ret, depth, credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005246 struct buffer_head *last_eb_bh = NULL;
5247 struct ocfs2_extent_block *eb;
5248 struct ocfs2_extent_list *rightmost_el, *el;
5249 struct ocfs2_extent_rec split_rec;
5250 struct ocfs2_extent_rec *rec;
5251 struct ocfs2_insert_type insert;
5252
5253 /*
5254 * Setup the record to split before we grow the tree.
5255 */
5256 el = path_leaf_el(path);
5257 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005258 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5259 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005260
5261 depth = path->p_tree_depth;
5262 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005263 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005264 ocfs2_et_get_last_eb_blk(et),
5265 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005266 if (ret < 0) {
5267 mlog_errno(ret);
5268 goto out;
5269 }
5270
5271 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5272 rightmost_el = &eb->h_list;
5273 } else
5274 rightmost_el = path_leaf_el(path);
5275
Tao Mac901fb02010-04-26 14:34:57 +08005276 credits = path->p_tree_depth +
5277 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005278 ret = ocfs2_extend_trans(handle, credits);
5279 if (ret) {
5280 mlog_errno(ret);
5281 goto out;
5282 }
5283
5284 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5285 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005286 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005287 meta_ac);
5288 if (ret) {
5289 mlog_errno(ret);
5290 goto out;
5291 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005292 }
5293
5294 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5295 insert.ins_appending = APPEND_NONE;
5296 insert.ins_contig = CONTIG_NONE;
5297 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005298 insert.ins_tree_depth = depth;
5299
Joel Becker3505bec2009-02-13 02:57:58 -08005300 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005301 if (ret)
5302 mlog_errno(ret);
5303
5304out:
5305 brelse(last_eb_bh);
5306 return ret;
5307}
5308
Joel Becker043beeb2009-02-13 02:42:30 -08005309static int ocfs2_truncate_rec(handle_t *handle,
5310 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005311 struct ocfs2_path *path, int index,
5312 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005313 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005314{
5315 int ret;
5316 u32 left_cpos, rec_range, trunc_range;
5317 int wants_rotate = 0, is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005318 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005319 struct ocfs2_path *left_path = NULL;
5320 struct ocfs2_extent_list *el = path_leaf_el(path);
5321 struct ocfs2_extent_rec *rec;
5322 struct ocfs2_extent_block *eb;
5323
5324 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Joel Becker70f18c02009-02-13 02:09:31 -08005325 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005326 if (ret) {
5327 mlog_errno(ret);
5328 goto out;
5329 }
5330
5331 index--;
5332 }
5333
5334 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5335 path->p_tree_depth) {
5336 /*
5337 * Check whether this is the rightmost tree record. If
5338 * we remove all of this record or part of its right
5339 * edge then an update of the record lengths above it
5340 * will be required.
5341 */
5342 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5343 if (eb->h_next_leaf_blk == 0)
5344 is_rightmost_tree_rec = 1;
5345 }
5346
5347 rec = &el->l_recs[index];
5348 if (index == 0 && path->p_tree_depth &&
5349 le32_to_cpu(rec->e_cpos) == cpos) {
5350 /*
5351 * Changing the leftmost offset (via partial or whole
5352 * record truncate) of an interior (or rightmost) path
5353 * means we have to update the subtree that is formed
5354 * by this leaf and the one to it's left.
5355 *
5356 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005357 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005358 * 2) The leaf is rightmost and will be empty after
5359 * we remove the extent record - the rotate code
5360 * knows how to update the newly formed edge.
5361 */
5362
Joel Becker043beeb2009-02-13 02:42:30 -08005363 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005364 if (ret) {
5365 mlog_errno(ret);
5366 goto out;
5367 }
5368
5369 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005370 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005371 if (!left_path) {
5372 ret = -ENOMEM;
5373 mlog_errno(ret);
5374 goto out;
5375 }
5376
Joel Beckerfacdb772009-02-12 18:08:48 -08005377 ret = ocfs2_find_path(et->et_ci, left_path,
5378 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005379 if (ret) {
5380 mlog_errno(ret);
5381 goto out;
5382 }
5383 }
5384 }
5385
5386 ret = ocfs2_extend_rotate_transaction(handle, 0,
5387 handle->h_buffer_credits,
5388 path);
5389 if (ret) {
5390 mlog_errno(ret);
5391 goto out;
5392 }
5393
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005394 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005395 if (ret) {
5396 mlog_errno(ret);
5397 goto out;
5398 }
5399
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005400 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005401 if (ret) {
5402 mlog_errno(ret);
5403 goto out;
5404 }
5405
5406 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5407 trunc_range = cpos + len;
5408
5409 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5410 int next_free;
5411
5412 memset(rec, 0, sizeof(*rec));
5413 ocfs2_cleanup_merge(el, index);
5414 wants_rotate = 1;
5415
5416 next_free = le16_to_cpu(el->l_next_free_rec);
5417 if (is_rightmost_tree_rec && next_free > 1) {
5418 /*
5419 * We skip the edge update if this path will
5420 * be deleted by the rotate code.
5421 */
5422 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005423 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005424 rec);
5425 }
5426 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5427 /* Remove leftmost portion of the record. */
5428 le32_add_cpu(&rec->e_cpos, len);
5429 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5430 le16_add_cpu(&rec->e_leaf_clusters, -len);
5431 } else if (rec_range == trunc_range) {
5432 /* Remove rightmost portion of the record */
5433 le16_add_cpu(&rec->e_leaf_clusters, -len);
5434 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005435 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005436 } else {
5437 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005438 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5439 "(%u, %u)\n",
5440 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005441 le32_to_cpu(rec->e_cpos),
5442 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5443 BUG();
5444 }
5445
5446 if (left_path) {
5447 int subtree_index;
5448
Joel Becker7dc02802009-02-12 19:20:13 -08005449 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005450 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005451 subtree_index);
5452 }
5453
5454 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5455
Joel Becker70f18c02009-02-13 02:09:31 -08005456 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005457 if (ret) {
5458 mlog_errno(ret);
5459 goto out;
5460 }
5461
5462out:
5463 ocfs2_free_path(left_path);
5464 return ret;
5465}
5466
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005467int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005468 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005469 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005470 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005471 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005472{
5473 int ret, index;
5474 u32 rec_range, trunc_range;
5475 struct ocfs2_extent_rec *rec;
5476 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005477 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005478
Joel Becker4c911ee2009-02-13 02:50:12 -08005479 /*
5480 * XXX: Why are we truncating to 0 instead of wherever this
5481 * affects us?
5482 */
5483 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005484
Joel Beckerffdd7a52008-10-17 22:32:01 -07005485 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005486 if (!path) {
5487 ret = -ENOMEM;
5488 mlog_errno(ret);
5489 goto out;
5490 }
5491
Joel Beckerfacdb772009-02-12 18:08:48 -08005492 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005493 if (ret) {
5494 mlog_errno(ret);
5495 goto out;
5496 }
5497
5498 el = path_leaf_el(path);
5499 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005500 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005501 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5502 "Owner %llu has an extent at cpos %u which can no "
Mark Fashehd0c7d702007-07-03 13:27:22 -07005503 "longer be found.\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005504 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5505 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005506 ret = -EROFS;
5507 goto out;
5508 }
5509
5510 /*
5511 * We have 3 cases of extent removal:
5512 * 1) Range covers the entire extent rec
5513 * 2) Range begins or ends on one edge of the extent rec
5514 * 3) Range is in the middle of the extent rec (no shared edges)
5515 *
5516 * For case 1 we remove the extent rec and left rotate to
5517 * fill the hole.
5518 *
5519 * For case 2 we just shrink the existing extent rec, with a
5520 * tree update if the shrinking edge is also the edge of an
5521 * extent block.
5522 *
5523 * For case 3 we do a right split to turn the extent rec into
5524 * something case 2 can handle.
5525 */
5526 rec = &el->l_recs[index];
5527 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5528 trunc_range = cpos + len;
5529
5530 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5531
Tao Maa09d09b2011-02-22 08:21:10 +08005532 trace_ocfs2_remove_extent(
5533 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5534 cpos, len, index, le32_to_cpu(rec->e_cpos),
5535 ocfs2_rec_clusters(el, rec));
Mark Fashehd0c7d702007-07-03 13:27:22 -07005536
5537 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005538 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5539 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005540 if (ret) {
5541 mlog_errno(ret);
5542 goto out;
5543 }
5544 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005545 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005546 trunc_range, meta_ac);
5547 if (ret) {
5548 mlog_errno(ret);
5549 goto out;
5550 }
5551
5552 /*
5553 * The split could have manipulated the tree enough to
5554 * move the record location, so we have to look for it again.
5555 */
5556 ocfs2_reinit_path(path, 1);
5557
Joel Beckerfacdb772009-02-12 18:08:48 -08005558 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005559 if (ret) {
5560 mlog_errno(ret);
5561 goto out;
5562 }
5563
5564 el = path_leaf_el(path);
5565 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005566 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005567 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5568 "Owner %llu: split at cpos %u lost record.",
5569 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005570 cpos);
5571 ret = -EROFS;
5572 goto out;
5573 }
5574
5575 /*
5576 * Double check our values here. If anything is fishy,
5577 * it's easier to catch it at the top level.
5578 */
5579 rec = &el->l_recs[index];
5580 rec_range = le32_to_cpu(rec->e_cpos) +
5581 ocfs2_rec_clusters(el, rec);
5582 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005583 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5584 "Owner %llu: error after split at cpos %u"
Mark Fashehd0c7d702007-07-03 13:27:22 -07005585 "trunc len %u, existing record is (%u,%u)",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005586 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005587 cpos, len, le32_to_cpu(rec->e_cpos),
5588 ocfs2_rec_clusters(el, rec));
5589 ret = -EROFS;
5590 goto out;
5591 }
5592
Joel Becker043beeb2009-02-13 02:42:30 -08005593 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5594 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005595 if (ret) {
5596 mlog_errno(ret);
5597 goto out;
5598 }
5599 }
5600
5601out:
5602 ocfs2_free_path(path);
5603 return ret;
5604}
5605
Tristan Ye78f94672010-05-11 17:54:42 +08005606/*
5607 * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5608 * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5609 * number to reserve some extra blocks, and it only handles meta
5610 * data allocations.
5611 *
5612 * Currently, only ocfs2_remove_btree_range() uses it for truncating
5613 * and punching holes.
5614 */
5615static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5616 struct ocfs2_extent_tree *et,
5617 u32 extents_to_split,
5618 struct ocfs2_alloc_context **ac,
5619 int extra_blocks)
5620{
5621 int ret = 0, num_free_extents;
5622 unsigned int max_recs_needed = 2 * extents_to_split;
5623 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5624
5625 *ac = NULL;
5626
5627 num_free_extents = ocfs2_num_free_extents(osb, et);
5628 if (num_free_extents < 0) {
5629 ret = num_free_extents;
5630 mlog_errno(ret);
5631 goto out;
5632 }
5633
5634 if (!num_free_extents ||
5635 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5636 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5637
5638 if (extra_blocks) {
5639 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5640 if (ret < 0) {
5641 if (ret != -ENOSPC)
5642 mlog_errno(ret);
5643 goto out;
5644 }
5645 }
5646
5647out:
5648 if (ret) {
5649 if (*ac) {
5650 ocfs2_free_alloc_context(*ac);
5651 *ac = NULL;
5652 }
5653 }
5654
5655 return ret;
5656}
5657
Mark Fashehfecc0112008-11-12 15:16:38 -08005658int ocfs2_remove_btree_range(struct inode *inode,
5659 struct ocfs2_extent_tree *et,
Tristan Ye78f94672010-05-11 17:54:42 +08005660 u32 cpos, u32 phys_cpos, u32 len, int flags,
5661 struct ocfs2_cached_dealloc_ctxt *dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08005662 u64 refcount_loc, bool refcount_tree_locked)
Mark Fashehfecc0112008-11-12 15:16:38 -08005663{
Tristan Ye78f94672010-05-11 17:54:42 +08005664 int ret, credits = 0, extra_blocks = 0;
Mark Fashehfecc0112008-11-12 15:16:38 -08005665 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5666 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5667 struct inode *tl_inode = osb->osb_tl_inode;
5668 handle_t *handle;
5669 struct ocfs2_alloc_context *meta_ac = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08005670 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehfecc0112008-11-12 15:16:38 -08005671
Tristan Ye78f94672010-05-11 17:54:42 +08005672 if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
5673 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
5674 OCFS2_HAS_REFCOUNT_FL));
5675
Junxiao Bif62f12b2014-12-18 16:17:32 -08005676 if (!refcount_tree_locked) {
5677 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5678 &ref_tree, NULL);
5679 if (ret) {
5680 mlog_errno(ret);
5681 goto bail;
5682 }
Tristan Ye78f94672010-05-11 17:54:42 +08005683 }
5684
5685 ret = ocfs2_prepare_refcount_change_for_del(inode,
5686 refcount_loc,
5687 phys_blkno,
5688 len,
5689 &credits,
5690 &extra_blocks);
5691 if (ret < 0) {
5692 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005693 goto bail;
Tristan Ye78f94672010-05-11 17:54:42 +08005694 }
5695 }
5696
5697 ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5698 extra_blocks);
Mark Fashehfecc0112008-11-12 15:16:38 -08005699 if (ret) {
5700 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005701 goto bail;
Mark Fashehfecc0112008-11-12 15:16:38 -08005702 }
5703
5704 mutex_lock(&tl_inode->i_mutex);
5705
5706 if (ocfs2_truncate_log_needs_flush(osb)) {
5707 ret = __ocfs2_flush_truncate_log(osb);
5708 if (ret < 0) {
5709 mlog_errno(ret);
5710 goto out;
5711 }
5712 }
5713
Tristan Ye78f94672010-05-11 17:54:42 +08005714 handle = ocfs2_start_trans(osb,
5715 ocfs2_remove_extent_credits(osb->sb) + credits);
Mark Fashehfecc0112008-11-12 15:16:38 -08005716 if (IS_ERR(handle)) {
5717 ret = PTR_ERR(handle);
5718 mlog_errno(ret);
5719 goto out;
5720 }
5721
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005722 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005723 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005724 if (ret) {
5725 mlog_errno(ret);
Wengang Wangb8a0ae52011-10-12 15:22:15 +08005726 goto out_commit;
Mark Fashehfecc0112008-11-12 15:16:38 -08005727 }
5728
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005729 dquot_free_space_nodirty(inode,
Mark Fashehfd4ef232009-01-29 15:06:21 -08005730 ocfs2_clusters_to_bytes(inode->i_sb, len));
5731
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005732 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005733 if (ret) {
5734 mlog_errno(ret);
5735 goto out_commit;
5736 }
5737
Joel Becker6136ca52009-02-12 19:32:43 -08005738 ocfs2_et_update_clusters(et, -len);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07005739 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fashehfecc0112008-11-12 15:16:38 -08005740
Joel Beckerec20cec2010-03-19 14:13:52 -07005741 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehfecc0112008-11-12 15:16:38 -08005742
Tristan Ye78f94672010-05-11 17:54:42 +08005743 if (phys_blkno) {
5744 if (flags & OCFS2_EXT_REFCOUNTED)
5745 ret = ocfs2_decrease_refcount(inode, handle,
5746 ocfs2_blocks_to_clusters(osb->sb,
5747 phys_blkno),
5748 len, meta_ac,
5749 dealloc, 1);
5750 else
5751 ret = ocfs2_truncate_log_append(osb, handle,
5752 phys_blkno, len);
5753 if (ret)
5754 mlog_errno(ret);
5755
5756 }
Mark Fashehfecc0112008-11-12 15:16:38 -08005757
5758out_commit:
5759 ocfs2_commit_trans(osb, handle);
5760out:
5761 mutex_unlock(&tl_inode->i_mutex);
Joseph Qi33add0e2013-07-03 15:00:52 -07005762bail:
Mark Fashehfecc0112008-11-12 15:16:38 -08005763 if (meta_ac)
5764 ocfs2_free_alloc_context(meta_ac);
5765
Tristan Ye78f94672010-05-11 17:54:42 +08005766 if (ref_tree)
5767 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5768
Mark Fashehfecc0112008-11-12 15:16:38 -08005769 return ret;
5770}
5771
Mark Fasheh063c4562007-07-03 13:34:11 -07005772int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005773{
5774 struct buffer_head *tl_bh = osb->osb_tl_bh;
5775 struct ocfs2_dinode *di;
5776 struct ocfs2_truncate_log *tl;
5777
5778 di = (struct ocfs2_dinode *) tl_bh->b_data;
5779 tl = &di->id2.i_dealloc;
5780
5781 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5782 "slot %d, invalid truncate log parameters: used = "
5783 "%u, count = %u\n", osb->slot_num,
5784 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5785 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5786}
5787
5788static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5789 unsigned int new_start)
5790{
5791 unsigned int tail_index;
5792 unsigned int current_tail;
5793
5794 /* No records, nothing to coalesce */
5795 if (!le16_to_cpu(tl->tl_used))
5796 return 0;
5797
5798 tail_index = le16_to_cpu(tl->tl_used) - 1;
5799 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5800 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5801
5802 return current_tail == new_start;
5803}
5804
Mark Fasheh063c4562007-07-03 13:34:11 -07005805int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5806 handle_t *handle,
5807 u64 start_blk,
5808 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005809{
5810 int status, index;
5811 unsigned int start_cluster, tl_count;
5812 struct inode *tl_inode = osb->osb_tl_inode;
5813 struct buffer_head *tl_bh = osb->osb_tl_bh;
5814 struct ocfs2_dinode *di;
5815 struct ocfs2_truncate_log *tl;
5816
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005817 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005818
5819 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5820
5821 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005822
Joel Becker10995aa2008-11-13 14:49:12 -08005823 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5824 * by the underlying call to ocfs2_read_inode_block(), so any
5825 * corruption is a code bug */
5826 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5827
5828 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005829 tl_count = le16_to_cpu(tl->tl_count);
5830 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5831 tl_count == 0,
Mark Fashehb06970532006-03-03 10:24:33 -08005832 "Truncate record count on #%llu invalid "
5833 "wanted %u, actual %u\n",
5834 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005835 ocfs2_truncate_recs_per_inode(osb->sb),
5836 le16_to_cpu(tl->tl_count));
5837
5838 /* Caller should have known to flush before calling us. */
5839 index = le16_to_cpu(tl->tl_used);
5840 if (index >= tl_count) {
5841 status = -ENOSPC;
5842 mlog_errno(status);
5843 goto bail;
5844 }
5845
Joel Becker0cf2f762009-02-12 16:41:25 -08005846 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005847 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005848 if (status < 0) {
5849 mlog_errno(status);
5850 goto bail;
5851 }
5852
Tao Maa09d09b2011-02-22 08:21:10 +08005853 trace_ocfs2_truncate_log_append(
5854 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5855 start_cluster, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005856 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5857 /*
5858 * Move index back to the record we are coalescing with.
5859 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5860 */
5861 index--;
5862
5863 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
Tao Maa09d09b2011-02-22 08:21:10 +08005864 trace_ocfs2_truncate_log_append(
5865 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5866 index, le32_to_cpu(tl->tl_recs[index].t_start),
5867 num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005868 } else {
5869 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5870 tl->tl_used = cpu_to_le16(index + 1);
5871 }
5872 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5873
Joel Beckerec20cec2010-03-19 14:13:52 -07005874 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005875
Tao Ma50308d82010-11-04 15:14:11 +08005876 osb->truncated_clusters += num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08005877bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005878 return status;
5879}
5880
5881static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07005882 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08005883 struct inode *data_alloc_inode,
5884 struct buffer_head *data_alloc_bh)
5885{
5886 int status = 0;
5887 int i;
5888 unsigned int num_clusters;
5889 u64 start_blk;
5890 struct ocfs2_truncate_rec rec;
5891 struct ocfs2_dinode *di;
5892 struct ocfs2_truncate_log *tl;
5893 struct inode *tl_inode = osb->osb_tl_inode;
5894 struct buffer_head *tl_bh = osb->osb_tl_bh;
5895
Mark Fashehccd979b2005-12-15 14:31:24 -08005896 di = (struct ocfs2_dinode *) tl_bh->b_data;
5897 tl = &di->id2.i_dealloc;
5898 i = le16_to_cpu(tl->tl_used) - 1;
5899 while (i >= 0) {
5900 /* Caller has given us at least enough credits to
5901 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005902 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005903 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005904 if (status < 0) {
5905 mlog_errno(status);
5906 goto bail;
5907 }
5908
5909 tl->tl_used = cpu_to_le16(i);
5910
Joel Beckerec20cec2010-03-19 14:13:52 -07005911 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005912
5913 /* TODO: Perhaps we can calculate the bulk of the
5914 * credits up front rather than extending like
5915 * this. */
5916 status = ocfs2_extend_trans(handle,
5917 OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5918 if (status < 0) {
5919 mlog_errno(status);
5920 goto bail;
5921 }
5922
5923 rec = tl->tl_recs[i];
5924 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5925 le32_to_cpu(rec.t_start));
5926 num_clusters = le32_to_cpu(rec.t_clusters);
5927
5928 /* if start_blk is not set, we ignore the record as
5929 * invalid. */
5930 if (start_blk) {
Tao Maa09d09b2011-02-22 08:21:10 +08005931 trace_ocfs2_replay_truncate_records(
5932 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5933 i, le32_to_cpu(rec.t_start), num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005934
5935 status = ocfs2_free_clusters(handle, data_alloc_inode,
5936 data_alloc_bh, start_blk,
5937 num_clusters);
5938 if (status < 0) {
5939 mlog_errno(status);
5940 goto bail;
5941 }
5942 }
5943 i--;
5944 }
5945
Tao Ma50308d82010-11-04 15:14:11 +08005946 osb->truncated_clusters = 0;
5947
Mark Fashehccd979b2005-12-15 14:31:24 -08005948bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005949 return status;
5950}
5951
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005952/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005953int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005954{
5955 int status;
5956 unsigned int num_to_flush;
Mark Fasheh1fabe142006-10-09 18:11:45 -07005957 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005958 struct inode *tl_inode = osb->osb_tl_inode;
5959 struct inode *data_alloc_inode = NULL;
5960 struct buffer_head *tl_bh = osb->osb_tl_bh;
5961 struct buffer_head *data_alloc_bh = NULL;
5962 struct ocfs2_dinode *di;
5963 struct ocfs2_truncate_log *tl;
5964
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005965 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005966
5967 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005968
Joel Becker10995aa2008-11-13 14:49:12 -08005969 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5970 * by the underlying call to ocfs2_read_inode_block(), so any
5971 * corruption is a code bug */
5972 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5973
5974 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005975 num_to_flush = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08005976 trace_ocfs2_flush_truncate_log(
5977 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5978 num_to_flush);
Mark Fashehccd979b2005-12-15 14:31:24 -08005979 if (!num_to_flush) {
5980 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005981 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005982 }
5983
5984 data_alloc_inode = ocfs2_get_system_file_inode(osb,
5985 GLOBAL_BITMAP_SYSTEM_INODE,
5986 OCFS2_INVALID_SLOT);
5987 if (!data_alloc_inode) {
5988 status = -EINVAL;
5989 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005990 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08005991 }
5992
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005993 mutex_lock(&data_alloc_inode->i_mutex);
5994
Mark Fashehe63aecb62007-10-18 15:30:42 -07005995 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08005996 if (status < 0) {
5997 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005998 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08005999 }
6000
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006001 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006002 if (IS_ERR(handle)) {
6003 status = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006004 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006005 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08006006 }
6007
6008 status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
6009 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006010 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08006011 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006012
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006013 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006014
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006015out_unlock:
6016 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07006017 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006018
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006019out_mutex:
6020 mutex_unlock(&data_alloc_inode->i_mutex);
6021 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006022
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006023out:
Mark Fashehccd979b2005-12-15 14:31:24 -08006024 return status;
6025}
6026
6027int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6028{
6029 int status;
6030 struct inode *tl_inode = osb->osb_tl_inode;
6031
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006032 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006033 status = __ocfs2_flush_truncate_log(osb);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006034 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006035
6036 return status;
6037}
6038
David Howellsc4028952006-11-22 14:57:56 +00006039static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08006040{
6041 int status;
David Howellsc4028952006-11-22 14:57:56 +00006042 struct ocfs2_super *osb =
6043 container_of(work, struct ocfs2_super,
6044 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08006045
Mark Fashehccd979b2005-12-15 14:31:24 -08006046 status = ocfs2_flush_truncate_log(osb);
6047 if (status < 0)
6048 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08006049 else
Tiger Yangb89c5422010-01-25 14:11:06 +08006050 ocfs2_init_steal_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08006051}
6052
6053#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6054void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6055 int cancel)
6056{
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006057 if (osb->osb_tl_inode &&
6058 atomic_read(&osb->osb_tl_disable) == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006059 /* We want to push off log flushes while truncates are
6060 * still running. */
6061 if (cancel)
6062 cancel_delayed_work(&osb->osb_truncate_log_wq);
6063
6064 queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
6065 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6066 }
6067}
6068
6069static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6070 int slot_num,
6071 struct inode **tl_inode,
6072 struct buffer_head **tl_bh)
6073{
6074 int status;
6075 struct inode *inode = NULL;
6076 struct buffer_head *bh = NULL;
6077
6078 inode = ocfs2_get_system_file_inode(osb,
6079 TRUNCATE_LOG_SYSTEM_INODE,
6080 slot_num);
6081 if (!inode) {
6082 status = -EINVAL;
6083 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6084 goto bail;
6085 }
6086
Joel Beckerb657c952008-11-13 14:49:11 -08006087 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006088 if (status < 0) {
6089 iput(inode);
6090 mlog_errno(status);
6091 goto bail;
6092 }
6093
6094 *tl_inode = inode;
6095 *tl_bh = bh;
6096bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08006097 return status;
6098}
6099
6100/* called during the 1st stage of node recovery. we stamp a clean
6101 * truncate log and pass back a copy for processing later. if the
6102 * truncate log does not require processing, a *tl_copy is set to
6103 * NULL. */
6104int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6105 int slot_num,
6106 struct ocfs2_dinode **tl_copy)
6107{
6108 int status;
6109 struct inode *tl_inode = NULL;
6110 struct buffer_head *tl_bh = NULL;
6111 struct ocfs2_dinode *di;
6112 struct ocfs2_truncate_log *tl;
6113
6114 *tl_copy = NULL;
6115
Tao Maa09d09b2011-02-22 08:21:10 +08006116 trace_ocfs2_begin_truncate_log_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08006117
6118 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6119 if (status < 0) {
6120 mlog_errno(status);
6121 goto bail;
6122 }
6123
6124 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006125
Joel Becker10995aa2008-11-13 14:49:12 -08006126 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6127 * validated by the underlying call to ocfs2_read_inode_block(),
6128 * so any corruption is a code bug */
6129 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6130
6131 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006132 if (le16_to_cpu(tl->tl_used)) {
Tao Maa09d09b2011-02-22 08:21:10 +08006133 trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
Mark Fashehccd979b2005-12-15 14:31:24 -08006134
6135 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6136 if (!(*tl_copy)) {
6137 status = -ENOMEM;
6138 mlog_errno(status);
6139 goto bail;
6140 }
6141
6142 /* Assuming the write-out below goes well, this copy
6143 * will be passed back to recovery for processing. */
6144 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6145
6146 /* All we need to do to clear the truncate log is set
6147 * tl_used. */
6148 tl->tl_used = 0;
6149
Joel Becker13723d02008-10-17 19:25:01 -07006150 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006151 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006152 if (status < 0) {
6153 mlog_errno(status);
6154 goto bail;
6155 }
6156 }
6157
6158bail:
6159 if (tl_inode)
6160 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006161 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006162
6163 if (status < 0 && (*tl_copy)) {
6164 kfree(*tl_copy);
6165 *tl_copy = NULL;
Tao Mac1e8d352011-03-07 16:43:21 +08006166 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006167 }
6168
Mark Fashehccd979b2005-12-15 14:31:24 -08006169 return status;
6170}
6171
6172int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6173 struct ocfs2_dinode *tl_copy)
6174{
6175 int status = 0;
6176 int i;
6177 unsigned int clusters, num_recs, start_cluster;
6178 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006179 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006180 struct inode *tl_inode = osb->osb_tl_inode;
6181 struct ocfs2_truncate_log *tl;
6182
Mark Fashehccd979b2005-12-15 14:31:24 -08006183 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6184 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6185 return -EINVAL;
6186 }
6187
6188 tl = &tl_copy->id2.i_dealloc;
6189 num_recs = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006190 trace_ocfs2_complete_truncate_log_recovery(
6191 (unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6192 num_recs);
Mark Fashehccd979b2005-12-15 14:31:24 -08006193
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006194 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006195 for(i = 0; i < num_recs; i++) {
6196 if (ocfs2_truncate_log_needs_flush(osb)) {
6197 status = __ocfs2_flush_truncate_log(osb);
6198 if (status < 0) {
6199 mlog_errno(status);
6200 goto bail_up;
6201 }
6202 }
6203
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006204 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006205 if (IS_ERR(handle)) {
6206 status = PTR_ERR(handle);
6207 mlog_errno(status);
6208 goto bail_up;
6209 }
6210
6211 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6212 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6213 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6214
6215 status = ocfs2_truncate_log_append(osb, handle,
6216 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006217 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006218 if (status < 0) {
6219 mlog_errno(status);
6220 goto bail_up;
6221 }
6222 }
6223
6224bail_up:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006225 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006226
Mark Fashehccd979b2005-12-15 14:31:24 -08006227 return status;
6228}
6229
6230void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6231{
6232 int status;
6233 struct inode *tl_inode = osb->osb_tl_inode;
6234
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006235 atomic_set(&osb->osb_tl_disable, 1);
6236
Mark Fashehccd979b2005-12-15 14:31:24 -08006237 if (tl_inode) {
6238 cancel_delayed_work(&osb->osb_truncate_log_wq);
6239 flush_workqueue(ocfs2_wq);
6240
6241 status = ocfs2_flush_truncate_log(osb);
6242 if (status < 0)
6243 mlog_errno(status);
6244
6245 brelse(osb->osb_tl_bh);
6246 iput(osb->osb_tl_inode);
6247 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006248}
6249
6250int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6251{
6252 int status;
6253 struct inode *tl_inode = NULL;
6254 struct buffer_head *tl_bh = NULL;
6255
Mark Fashehccd979b2005-12-15 14:31:24 -08006256 status = ocfs2_get_truncate_log_info(osb,
6257 osb->slot_num,
6258 &tl_inode,
6259 &tl_bh);
6260 if (status < 0)
6261 mlog_errno(status);
6262
6263 /* ocfs2_truncate_log_shutdown keys on the existence of
6264 * osb->osb_tl_inode so we don't set any of the osb variables
6265 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006266 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6267 ocfs2_truncate_log_worker);
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006268 atomic_set(&osb->osb_tl_disable, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08006269 osb->osb_tl_bh = tl_bh;
6270 osb->osb_tl_inode = tl_inode;
6271
Mark Fashehccd979b2005-12-15 14:31:24 -08006272 return status;
6273}
6274
Mark Fasheh2b604352007-06-22 15:45:27 -07006275/*
6276 * Delayed de-allocation of suballocator blocks.
6277 *
6278 * Some sets of block de-allocations might involve multiple suballocator inodes.
6279 *
6280 * The locking for this can get extremely complicated, especially when
6281 * the suballocator inodes to delete from aren't known until deep
6282 * within an unrelated codepath.
6283 *
6284 * ocfs2_extent_block structures are a good example of this - an inode
6285 * btree could have been grown by any number of nodes each allocating
6286 * out of their own suballoc inode.
6287 *
6288 * These structures allow the delay of block de-allocation until a
6289 * later time, when locking of multiple cluster inodes won't cause
6290 * deadlock.
6291 */
6292
6293/*
Tao Ma2891d292008-11-12 08:26:58 +08006294 * Describe a single bit freed from a suballocator. For the block
6295 * suballocators, it represents one block. For the global cluster
6296 * allocator, it represents some clusters and free_bit indicates
6297 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006298 */
6299struct ocfs2_cached_block_free {
6300 struct ocfs2_cached_block_free *free_next;
Tao Ma74380c42010-03-22 14:20:18 +08006301 u64 free_bg;
Mark Fasheh2b604352007-06-22 15:45:27 -07006302 u64 free_blk;
6303 unsigned int free_bit;
6304};
6305
6306struct ocfs2_per_slot_free_list {
6307 struct ocfs2_per_slot_free_list *f_next_suballocator;
6308 int f_inode_type;
6309 int f_slot;
6310 struct ocfs2_cached_block_free *f_first;
6311};
6312
Tao Ma2891d292008-11-12 08:26:58 +08006313static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6314 int sysfile_type,
6315 int slot,
6316 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006317{
6318 int ret;
6319 u64 bg_blkno;
6320 handle_t *handle;
6321 struct inode *inode;
6322 struct buffer_head *di_bh = NULL;
6323 struct ocfs2_cached_block_free *tmp;
6324
6325 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6326 if (!inode) {
6327 ret = -EINVAL;
6328 mlog_errno(ret);
6329 goto out;
6330 }
6331
6332 mutex_lock(&inode->i_mutex);
6333
Mark Fashehe63aecb62007-10-18 15:30:42 -07006334 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006335 if (ret) {
6336 mlog_errno(ret);
6337 goto out_mutex;
6338 }
6339
6340 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6341 if (IS_ERR(handle)) {
6342 ret = PTR_ERR(handle);
6343 mlog_errno(ret);
6344 goto out_unlock;
6345 }
6346
6347 while (head) {
Tao Ma74380c42010-03-22 14:20:18 +08006348 if (head->free_bg)
6349 bg_blkno = head->free_bg;
6350 else
6351 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6352 head->free_bit);
Tao Maa09d09b2011-02-22 08:21:10 +08006353 trace_ocfs2_free_cached_blocks(
6354 (unsigned long long)head->free_blk, head->free_bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006355
6356 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6357 head->free_bit, bg_blkno, 1);
6358 if (ret) {
6359 mlog_errno(ret);
6360 goto out_journal;
6361 }
6362
6363 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6364 if (ret) {
6365 mlog_errno(ret);
6366 goto out_journal;
6367 }
6368
6369 tmp = head;
6370 head = head->free_next;
6371 kfree(tmp);
6372 }
6373
6374out_journal:
6375 ocfs2_commit_trans(osb, handle);
6376
6377out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006378 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006379 brelse(di_bh);
6380out_mutex:
6381 mutex_unlock(&inode->i_mutex);
6382 iput(inode);
6383out:
6384 while(head) {
6385 /* Premature exit may have left some dangling items. */
6386 tmp = head;
6387 head = head->free_next;
6388 kfree(tmp);
6389 }
6390
6391 return ret;
6392}
6393
Tao Ma2891d292008-11-12 08:26:58 +08006394int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6395 u64 blkno, unsigned int bit)
6396{
6397 int ret = 0;
6398 struct ocfs2_cached_block_free *item;
6399
Tao Ma74380c42010-03-22 14:20:18 +08006400 item = kzalloc(sizeof(*item), GFP_NOFS);
Tao Ma2891d292008-11-12 08:26:58 +08006401 if (item == NULL) {
6402 ret = -ENOMEM;
6403 mlog_errno(ret);
6404 return ret;
6405 }
6406
Tao Maa09d09b2011-02-22 08:21:10 +08006407 trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
Tao Ma2891d292008-11-12 08:26:58 +08006408
6409 item->free_blk = blkno;
6410 item->free_bit = bit;
6411 item->free_next = ctxt->c_global_allocator;
6412
6413 ctxt->c_global_allocator = item;
6414 return ret;
6415}
6416
6417static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6418 struct ocfs2_cached_block_free *head)
6419{
6420 struct ocfs2_cached_block_free *tmp;
6421 struct inode *tl_inode = osb->osb_tl_inode;
6422 handle_t *handle;
6423 int ret = 0;
6424
6425 mutex_lock(&tl_inode->i_mutex);
6426
6427 while (head) {
6428 if (ocfs2_truncate_log_needs_flush(osb)) {
6429 ret = __ocfs2_flush_truncate_log(osb);
6430 if (ret < 0) {
6431 mlog_errno(ret);
6432 break;
6433 }
6434 }
6435
6436 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6437 if (IS_ERR(handle)) {
6438 ret = PTR_ERR(handle);
6439 mlog_errno(ret);
6440 break;
6441 }
6442
6443 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6444 head->free_bit);
6445
6446 ocfs2_commit_trans(osb, handle);
6447 tmp = head;
6448 head = head->free_next;
6449 kfree(tmp);
6450
6451 if (ret < 0) {
6452 mlog_errno(ret);
6453 break;
6454 }
6455 }
6456
6457 mutex_unlock(&tl_inode->i_mutex);
6458
6459 while (head) {
6460 /* Premature exit may have left some dangling items. */
6461 tmp = head;
6462 head = head->free_next;
6463 kfree(tmp);
6464 }
6465
6466 return ret;
6467}
6468
Mark Fasheh2b604352007-06-22 15:45:27 -07006469int ocfs2_run_deallocs(struct ocfs2_super *osb,
6470 struct ocfs2_cached_dealloc_ctxt *ctxt)
6471{
6472 int ret = 0, ret2;
6473 struct ocfs2_per_slot_free_list *fl;
6474
6475 if (!ctxt)
6476 return 0;
6477
6478 while (ctxt->c_first_suballocator) {
6479 fl = ctxt->c_first_suballocator;
6480
6481 if (fl->f_first) {
Tao Maa09d09b2011-02-22 08:21:10 +08006482 trace_ocfs2_run_deallocs(fl->f_inode_type,
6483 fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006484 ret2 = ocfs2_free_cached_blocks(osb,
6485 fl->f_inode_type,
6486 fl->f_slot,
6487 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006488 if (ret2)
6489 mlog_errno(ret2);
6490 if (!ret)
6491 ret = ret2;
6492 }
6493
6494 ctxt->c_first_suballocator = fl->f_next_suballocator;
6495 kfree(fl);
6496 }
6497
Tao Ma2891d292008-11-12 08:26:58 +08006498 if (ctxt->c_global_allocator) {
6499 ret2 = ocfs2_free_cached_clusters(osb,
6500 ctxt->c_global_allocator);
6501 if (ret2)
6502 mlog_errno(ret2);
6503 if (!ret)
6504 ret = ret2;
6505
6506 ctxt->c_global_allocator = NULL;
6507 }
6508
Mark Fasheh2b604352007-06-22 15:45:27 -07006509 return ret;
6510}
6511
6512static struct ocfs2_per_slot_free_list *
6513ocfs2_find_per_slot_free_list(int type,
6514 int slot,
6515 struct ocfs2_cached_dealloc_ctxt *ctxt)
6516{
6517 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6518
6519 while (fl) {
6520 if (fl->f_inode_type == type && fl->f_slot == slot)
6521 return fl;
6522
6523 fl = fl->f_next_suballocator;
6524 }
6525
6526 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6527 if (fl) {
6528 fl->f_inode_type = type;
6529 fl->f_slot = slot;
6530 fl->f_first = NULL;
6531 fl->f_next_suballocator = ctxt->c_first_suballocator;
6532
6533 ctxt->c_first_suballocator = fl;
6534 }
6535 return fl;
6536}
6537
Tao Ma1823cb02009-08-18 11:24:49 +08006538int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
Tao Ma74380c42010-03-22 14:20:18 +08006539 int type, int slot, u64 suballoc,
6540 u64 blkno, unsigned int bit)
Mark Fasheh2b604352007-06-22 15:45:27 -07006541{
6542 int ret;
6543 struct ocfs2_per_slot_free_list *fl;
6544 struct ocfs2_cached_block_free *item;
6545
6546 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6547 if (fl == NULL) {
6548 ret = -ENOMEM;
6549 mlog_errno(ret);
6550 goto out;
6551 }
6552
Tao Ma74380c42010-03-22 14:20:18 +08006553 item = kzalloc(sizeof(*item), GFP_NOFS);
Mark Fasheh2b604352007-06-22 15:45:27 -07006554 if (item == NULL) {
6555 ret = -ENOMEM;
6556 mlog_errno(ret);
6557 goto out;
6558 }
6559
Tao Maa09d09b2011-02-22 08:21:10 +08006560 trace_ocfs2_cache_block_dealloc(type, slot,
6561 (unsigned long long)suballoc,
6562 (unsigned long long)blkno, bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006563
Tao Ma74380c42010-03-22 14:20:18 +08006564 item->free_bg = suballoc;
Mark Fasheh2b604352007-06-22 15:45:27 -07006565 item->free_blk = blkno;
6566 item->free_bit = bit;
6567 item->free_next = fl->f_first;
6568
6569 fl->f_first = item;
6570
6571 ret = 0;
6572out:
6573 return ret;
6574}
6575
Mark Fasheh59a5e412007-06-22 15:52:36 -07006576static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6577 struct ocfs2_extent_block *eb)
6578{
6579 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6580 le16_to_cpu(eb->h_suballoc_slot),
Tao Ma74380c42010-03-22 14:20:18 +08006581 le64_to_cpu(eb->h_suballoc_loc),
Mark Fasheh59a5e412007-06-22 15:52:36 -07006582 le64_to_cpu(eb->h_blkno),
6583 le16_to_cpu(eb->h_suballoc_bit));
6584}
6585
Joel Becker2b4e30f2008-09-03 20:03:41 -07006586static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006587{
6588 set_buffer_uptodate(bh);
6589 mark_buffer_dirty(bh);
6590 return 0;
6591}
6592
Tao Ma6f70fa52009-08-25 08:05:12 +08006593void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6594 unsigned int from, unsigned int to,
6595 struct page *page, int zero, u64 *phys)
Mark Fasheh1d410a62007-09-07 14:20:45 -07006596{
6597 int ret, partial = 0;
6598
6599 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6600 if (ret)
6601 mlog_errno(ret);
6602
6603 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006604 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006605
6606 /*
6607 * Need to set the buffers we zero'd into uptodate
6608 * here if they aren't - ocfs2_map_page_blocks()
6609 * might've skipped some
6610 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006611 ret = walk_page_buffers(handle, page_buffers(page),
6612 from, to, &partial,
6613 ocfs2_zero_func);
6614 if (ret < 0)
6615 mlog_errno(ret);
6616 else if (ocfs2_should_order_data(inode)) {
6617 ret = ocfs2_jbd2_file_inode(handle, inode);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006618 if (ret < 0)
6619 mlog_errno(ret);
6620 }
6621
6622 if (!partial)
6623 SetPageUptodate(page);
6624
6625 flush_dcache_page(page);
6626}
6627
Mark Fasheh35edec12007-07-06 14:41:18 -07006628static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6629 loff_t end, struct page **pages,
6630 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006631{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006632 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006633 struct page *page;
6634 unsigned int from, to = PAGE_CACHE_SIZE;
6635 struct super_block *sb = inode->i_sb;
6636
6637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6638
6639 if (numpages == 0)
6640 goto out;
6641
Mark Fasheh35edec12007-07-06 14:41:18 -07006642 to = PAGE_CACHE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006643 for(i = 0; i < numpages; i++) {
6644 page = pages[i];
6645
Mark Fasheh35edec12007-07-06 14:41:18 -07006646 from = start & (PAGE_CACHE_SIZE - 1);
6647 if ((end >> PAGE_CACHE_SHIFT) == page->index)
6648 to = end & (PAGE_CACHE_SIZE - 1);
6649
Mark Fasheh60b11392007-02-16 11:46:50 -08006650 BUG_ON(from > PAGE_CACHE_SIZE);
6651 BUG_ON(to > PAGE_CACHE_SIZE);
6652
Mark Fasheh1d410a62007-09-07 14:20:45 -07006653 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6654 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006655
Mark Fasheh35edec12007-07-06 14:41:18 -07006656 start = (page->index + 1) << PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006657 }
6658out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006659 if (pages)
6660 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006661}
6662
Tao Ma6f70fa52009-08-25 08:05:12 +08006663int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6664 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006665{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006666 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006667 struct address_space *mapping = inode->i_mapping;
6668 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006669 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006670
Mark Fasheh35edec12007-07-06 14:41:18 -07006671 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006672
Mark Fasheh1d410a62007-09-07 14:20:45 -07006673 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006674 last_page_bytes = PAGE_ALIGN(end);
6675 index = start >> PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006676 do {
Jan Kara9b4c0ff2010-08-24 14:28:03 +02006677 pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fasheh60b11392007-02-16 11:46:50 -08006678 if (!pages[numpages]) {
6679 ret = -ENOMEM;
6680 mlog_errno(ret);
6681 goto out;
6682 }
6683
6684 numpages++;
6685 index++;
Mark Fasheh35edec12007-07-06 14:41:18 -07006686 } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006687
6688out:
6689 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006690 if (pages)
6691 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006692 numpages = 0;
6693 }
6694
6695 *num = numpages;
6696
6697 return ret;
6698}
6699
Tao Ma6f70fa52009-08-25 08:05:12 +08006700static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6701 struct page **pages, int *num)
6702{
6703 struct super_block *sb = inode->i_sb;
6704
6705 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6706 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6707
6708 return ocfs2_grab_pages(inode, start, end, pages, num);
6709}
6710
Mark Fasheh60b11392007-02-16 11:46:50 -08006711/*
6712 * Zero the area past i_size but still within an allocated
6713 * cluster. This avoids exposing nonzero data on subsequent file
6714 * extends.
6715 *
6716 * We need to call this before i_size is updated on the inode because
6717 * otherwise block_write_full_page() will skip writeout of pages past
6718 * i_size. The new_i_size parameter is passed for this reason.
6719 */
Mark Fasheh35edec12007-07-06 14:41:18 -07006720int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6721 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08006722{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006723 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08006724 struct page **pages = NULL;
6725 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006726 unsigned int ext_flags;
6727 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08006728
6729 /*
6730 * File systems which don't support sparse files zero on every
6731 * extend.
6732 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07006733 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08006734 return 0;
6735
Mark Fasheh1d410a62007-09-07 14:20:45 -07006736 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08006737 sizeof(struct page *), GFP_NOFS);
6738 if (pages == NULL) {
6739 ret = -ENOMEM;
6740 mlog_errno(ret);
6741 goto out;
6742 }
6743
Mark Fasheh1d410a62007-09-07 14:20:45 -07006744 if (range_start == range_end)
6745 goto out;
6746
6747 ret = ocfs2_extent_map_get_blocks(inode,
6748 range_start >> sb->s_blocksize_bits,
6749 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08006750 if (ret) {
6751 mlog_errno(ret);
6752 goto out;
6753 }
6754
Mark Fasheh1d410a62007-09-07 14:20:45 -07006755 /*
6756 * Tail is a hole, or is marked unwritten. In either case, we
6757 * can count on read and write to return/push zero's.
6758 */
6759 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08006760 goto out;
6761
Mark Fasheh1d410a62007-09-07 14:20:45 -07006762 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6763 &numpages);
6764 if (ret) {
6765 mlog_errno(ret);
6766 goto out;
6767 }
6768
Mark Fasheh35edec12007-07-06 14:41:18 -07006769 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6770 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08006771
6772 /*
6773 * Initiate writeout of the pages we zero'd here. We don't
6774 * wait on them - the truncate_inode_pages() call later will
6775 * do that for us.
6776 */
Christoph Hellwig2cfd30a2009-09-23 15:04:02 +02006777 ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6778 range_end - 1);
Mark Fasheh60b11392007-02-16 11:46:50 -08006779 if (ret)
6780 mlog_errno(ret);
6781
6782out:
Tim Gardnerd787ab02013-02-21 16:42:44 -08006783 kfree(pages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006784
6785 return ret;
6786}
6787
Tiger Yangfdd77702008-08-18 17:08:55 +08006788static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
6789 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006790{
6791 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08006792 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006793
Tiger Yangfdd77702008-08-18 17:08:55 +08006794 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
6795 memset(&di->id2, 0, blocksize -
6796 offsetof(struct ocfs2_dinode, id2) -
6797 xattrsize);
6798 else
6799 memset(&di->id2, 0, blocksize -
6800 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006801}
6802
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006803void ocfs2_dinode_new_extent_list(struct inode *inode,
6804 struct ocfs2_dinode *di)
6805{
Tiger Yangfdd77702008-08-18 17:08:55 +08006806 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006807 di->id2.i_list.l_tree_depth = 0;
6808 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08006809 di->id2.i_list.l_count = cpu_to_le16(
6810 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006811}
6812
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006813void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
6814{
6815 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6816 struct ocfs2_inline_data *idata = &di->id2.i_data;
6817
6818 spin_lock(&oi->ip_lock);
6819 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
6820 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6821 spin_unlock(&oi->ip_lock);
6822
6823 /*
6824 * We clear the entire i_data structure here so that all
6825 * fields can be properly initialized.
6826 */
Tiger Yangfdd77702008-08-18 17:08:55 +08006827 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006828
Tiger Yangfdd77702008-08-18 17:08:55 +08006829 idata->id_count = cpu_to_le16(
6830 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006831}
6832
6833int ocfs2_convert_inline_data_to_extents(struct inode *inode,
6834 struct buffer_head *di_bh)
6835{
6836 int ret, i, has_data, num_pages = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006837 int need_free = 0;
6838 u32 bit_off, num;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006839 handle_t *handle;
6840 u64 uninitialized_var(block);
6841 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6842 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6843 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006844 struct ocfs2_alloc_context *data_ac = NULL;
6845 struct page **pages = NULL;
6846 loff_t end = osb->s_clustersize;
Joel Beckerf99b9b72008-08-20 19:36:33 -07006847 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02006848 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006849
6850 has_data = i_size_read(inode) ? 1 : 0;
6851
6852 if (has_data) {
6853 pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
6854 sizeof(struct page *), GFP_NOFS);
6855 if (pages == NULL) {
6856 ret = -ENOMEM;
6857 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006858 return ret;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006859 }
6860
6861 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
6862 if (ret) {
6863 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006864 goto free_pages;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006865 }
6866 }
6867
Jan Karaa90714c2008-10-09 19:38:40 +02006868 handle = ocfs2_start_trans(osb,
6869 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006870 if (IS_ERR(handle)) {
6871 ret = PTR_ERR(handle);
6872 mlog_errno(ret);
alex chen15eba0f2015-02-10 14:09:02 -08006873 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006874 }
6875
Joel Becker0cf2f762009-02-12 16:41:25 -08006876 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07006877 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006878 if (ret) {
6879 mlog_errno(ret);
6880 goto out_commit;
6881 }
6882
6883 if (has_data) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006884 unsigned int page_end;
6885 u64 phys;
6886
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006887 ret = dquot_alloc_space_nodirty(inode,
6888 ocfs2_clusters_to_bytes(osb->sb, 1));
6889 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02006890 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02006891 did_quota = 1;
6892
Mark Fasheh4fe370a2009-12-07 13:15:40 -08006893 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
6894
Joel Becker1ed9b772010-05-06 13:59:06 +08006895 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006896 &num);
6897 if (ret) {
6898 mlog_errno(ret);
6899 goto out_commit;
6900 }
6901
6902 /*
6903 * Save two copies, one for insert, and one that can
6904 * be changed by ocfs2_map_and_dirty_page() below.
6905 */
6906 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
6907
6908 /*
6909 * Non sparse file systems zero on extend, so no need
6910 * to do that now.
6911 */
6912 if (!ocfs2_sparse_alloc(osb) &&
6913 PAGE_CACHE_SIZE < osb->s_clustersize)
6914 end = PAGE_CACHE_SIZE;
6915
6916 ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
6917 if (ret) {
6918 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006919 need_free = 1;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006920 goto out_commit;
6921 }
6922
6923 /*
6924 * This should populate the 1st page for us and mark
6925 * it up to date.
6926 */
6927 ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
6928 if (ret) {
6929 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006930 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08006931 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006932 }
6933
6934 page_end = PAGE_CACHE_SIZE;
6935 if (PAGE_CACHE_SIZE > osb->s_clustersize)
6936 page_end = osb->s_clustersize;
6937
6938 for (i = 0; i < num_pages; i++)
6939 ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
6940 pages[i], i > 0, &phys);
6941 }
6942
6943 spin_lock(&oi->ip_lock);
6944 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
6945 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6946 spin_unlock(&oi->ip_lock);
6947
Darrick J. Wong2931cdc2014-04-03 14:46:48 -07006948 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006949 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006950
6951 ocfs2_journal_dirty(handle, di_bh);
6952
6953 if (has_data) {
6954 /*
6955 * An error at this point should be extremely rare. If
6956 * this proves to be false, we could always re-build
6957 * the in-inode data from our pages.
6958 */
Joel Becker5e404e92009-02-13 03:54:22 -08006959 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08006960 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006961 if (ret) {
6962 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006963 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08006964 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006965 }
6966
6967 inode->i_blocks = ocfs2_inode_sector_count(inode);
6968 }
6969
alex chen15eba0f2015-02-10 14:09:02 -08006970out_unlock:
6971 if (pages)
6972 ocfs2_unlock_and_free_pages(pages, num_pages);
6973
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006974out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02006975 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006976 dquot_free_space_nodirty(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02006977 ocfs2_clusters_to_bytes(osb->sb, 1));
6978
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006979 if (need_free) {
6980 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
6981 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
6982 bit_off, num);
6983 else
6984 ocfs2_free_clusters(handle,
6985 data_ac->ac_inode,
6986 data_ac->ac_bh,
6987 ocfs2_clusters_to_blocks(osb->sb, bit_off),
6988 num);
6989 }
6990
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006991 ocfs2_commit_trans(osb, handle);
6992
alex chen15eba0f2015-02-10 14:09:02 -08006993out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006994 if (data_ac)
6995 ocfs2_free_alloc_context(data_ac);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006996free_pages:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07006997 kfree(pages);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006998 return ret;
6999}
7000
Mark Fashehccd979b2005-12-15 14:31:24 -08007001/*
7002 * It is expected, that by the time you call this function,
7003 * inode->i_size and fe->i_size have been adjusted.
7004 *
7005 * WARNING: This will kfree the truncate context
7006 */
7007int ocfs2_commit_truncate(struct ocfs2_super *osb,
7008 struct inode *inode,
Tristan Ye78f94672010-05-11 17:54:42 +08007009 struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08007010{
Tristan Ye78f94672010-05-11 17:54:42 +08007011 int status = 0, i, flags = 0;
7012 u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
Tao Mabcbbb242009-08-18 11:29:12 +08007013 u64 blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08007014 struct ocfs2_extent_list *el;
Tristan Ye78f94672010-05-11 17:54:42 +08007015 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08007016 struct ocfs2_path *path = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08007017 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7018 struct ocfs2_extent_list *root_el = &(di->id2.i_list);
7019 u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
7020 struct ocfs2_extent_tree et;
7021 struct ocfs2_cached_dealloc_ctxt dealloc;
Junxiao Bif62f12b2014-12-18 16:17:32 -08007022 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007023
Tristan Ye78f94672010-05-11 17:54:42 +08007024 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7025 ocfs2_init_dealloc_ctxt(&dealloc);
7026
Mark Fashehdcd05382007-01-16 11:32:23 -08007027 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08007028 i_size_read(inode));
7029
Tristan Ye78f94672010-05-11 17:54:42 +08007030 path = ocfs2_new_path(di_bh, &di->id2.i_list,
Joel Becker13723d02008-10-17 19:25:01 -07007031 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08007032 if (!path) {
7033 status = -ENOMEM;
7034 mlog_errno(status);
7035 goto bail;
7036 }
Mark Fasheh83418972007-04-23 18:53:12 -07007037
7038 ocfs2_extent_map_trunc(inode, new_highest_cpos);
7039
Mark Fashehccd979b2005-12-15 14:31:24 -08007040start:
Mark Fashehdcd05382007-01-16 11:32:23 -08007041 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007042 * Check that we still have allocation to delete.
7043 */
7044 if (OCFS2_I(inode)->ip_clusters == 0) {
7045 status = 0;
7046 goto bail;
7047 }
7048
7049 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007050 * Truncate always works against the rightmost tree branch.
7051 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007052 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007053 if (status) {
7054 mlog_errno(status);
7055 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007056 }
7057
Tao Maa09d09b2011-02-22 08:21:10 +08007058 trace_ocfs2_commit_truncate(
7059 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7060 new_highest_cpos,
7061 OCFS2_I(inode)->ip_clusters,
7062 path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08007063
7064 /*
7065 * By now, el will point to the extent list on the bottom most
7066 * portion of this tree. Only the tail record is considered in
7067 * each pass.
7068 *
7069 * We handle the following cases, in order:
7070 * - empty extent: delete the remaining branch
7071 * - remove the entire record
7072 * - remove a partial record
7073 * - no record needs to be removed (truncate has completed)
7074 */
7075 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007076 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7077 ocfs2_error(inode->i_sb,
7078 "Inode %llu has empty extent block at %llu\n",
7079 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7080 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7081 status = -EROFS;
7082 goto bail;
7083 }
7084
Mark Fashehccd979b2005-12-15 14:31:24 -08007085 i = le16_to_cpu(el->l_next_free_rec) - 1;
Tristan Ye78f94672010-05-11 17:54:42 +08007086 rec = &el->l_recs[i];
7087 flags = rec->e_flags;
7088 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7089
7090 if (i == 0 && ocfs2_is_empty_extent(rec)) {
7091 /*
7092 * Lower levels depend on this never happening, but it's best
7093 * to check it up here before changing the tree.
7094 */
7095 if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
7096 ocfs2_error(inode->i_sb, "Inode %lu has an empty "
7097 "extent record, depth %u\n", inode->i_ino,
7098 le16_to_cpu(root_el->l_tree_depth));
7099 status = -EROFS;
7100 goto bail;
7101 }
7102 trunc_cpos = le32_to_cpu(rec->e_cpos);
7103 trunc_len = 0;
7104 blkno = 0;
7105 } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7106 /*
7107 * Truncate entire record.
7108 */
7109 trunc_cpos = le32_to_cpu(rec->e_cpos);
7110 trunc_len = ocfs2_rec_clusters(el, rec);
7111 blkno = le64_to_cpu(rec->e_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007112 } else if (range > new_highest_cpos) {
Tristan Ye78f94672010-05-11 17:54:42 +08007113 /*
7114 * Partial truncate. it also should be
7115 * the last truncate we're doing.
7116 */
7117 trunc_cpos = new_highest_cpos;
7118 trunc_len = range - new_highest_cpos;
7119 coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7120 blkno = le64_to_cpu(rec->e_blkno) +
7121 ocfs2_clusters_to_blocks(inode->i_sb, coff);
Mark Fashehdcd05382007-01-16 11:32:23 -08007122 } else {
Tristan Ye78f94672010-05-11 17:54:42 +08007123 /*
7124 * Truncate completed, leave happily.
7125 */
Mark Fashehdcd05382007-01-16 11:32:23 -08007126 status = 0;
7127 goto bail;
7128 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007129
Tristan Ye78f94672010-05-11 17:54:42 +08007130 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007131
Junxiao Bif62f12b2014-12-18 16:17:32 -08007132 if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) {
7133 status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
7134 &ref_tree, NULL);
7135 if (status) {
7136 mlog_errno(status);
7137 goto bail;
7138 }
7139 }
7140
Tristan Ye78f94672010-05-11 17:54:42 +08007141 status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7142 phys_cpos, trunc_len, flags, &dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08007143 refcount_loc, true);
Mark Fashehccd979b2005-12-15 14:31:24 -08007144 if (status < 0) {
7145 mlog_errno(status);
7146 goto bail;
7147 }
7148
Mark Fashehdcd05382007-01-16 11:32:23 -08007149 ocfs2_reinit_path(path, 1);
7150
7151 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007152 * The check above will catch the case where we've truncated
7153 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007154 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007155 goto start;
7156
Mark Fashehccd979b2005-12-15 14:31:24 -08007157bail:
Junxiao Bif62f12b2014-12-18 16:17:32 -08007158 if (ref_tree)
7159 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08007160
7161 ocfs2_schedule_truncate_log_flush(osb, 1);
7162
Tristan Ye78f94672010-05-11 17:54:42 +08007163 ocfs2_run_deallocs(osb, &dealloc);
Mark Fasheh59a5e412007-06-22 15:52:36 -07007164
Mark Fashehdcd05382007-01-16 11:32:23 -08007165 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007166
Mark Fashehccd979b2005-12-15 14:31:24 -08007167 return status;
7168}
7169
Mark Fashehccd979b2005-12-15 14:31:24 -08007170/*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007171 * 'start' is inclusive, 'end' is not.
7172 */
7173int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7174 unsigned int start, unsigned int end, int trunc)
7175{
7176 int ret;
7177 unsigned int numbytes;
7178 handle_t *handle;
7179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7180 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7181 struct ocfs2_inline_data *idata = &di->id2.i_data;
7182
7183 if (end > i_size_read(inode))
7184 end = i_size_read(inode);
7185
Younger Liud62e74b2014-02-10 14:25:51 -08007186 BUG_ON(start > end);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007187
7188 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7189 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7190 !ocfs2_supports_inline_data(osb)) {
7191 ocfs2_error(inode->i_sb,
7192 "Inline data flags for inode %llu don't agree! "
7193 "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
7194 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7195 le16_to_cpu(di->i_dyn_features),
7196 OCFS2_I(inode)->ip_dyn_features,
7197 osb->s_feature_incompat);
7198 ret = -EROFS;
7199 goto out;
7200 }
7201
7202 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7203 if (IS_ERR(handle)) {
7204 ret = PTR_ERR(handle);
7205 mlog_errno(ret);
7206 goto out;
7207 }
7208
Joel Becker0cf2f762009-02-12 16:41:25 -08007209 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007210 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007211 if (ret) {
7212 mlog_errno(ret);
7213 goto out_commit;
7214 }
7215
7216 numbytes = end - start;
7217 memset(idata->id_data + start, 0, numbytes);
7218
7219 /*
7220 * No need to worry about the data page here - it's been
7221 * truncated already and inline data doesn't need it for
7222 * pushing zero's to disk, so we'll let readpage pick it up
7223 * later.
7224 */
7225 if (trunc) {
7226 i_size_write(inode, start);
7227 di->i_size = cpu_to_le64(start);
7228 }
7229
7230 inode->i_blocks = ocfs2_inode_sector_count(inode);
7231 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7232
7233 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7234 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7235
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07007236 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007237 ocfs2_journal_dirty(handle, di_bh);
7238
7239out_commit:
7240 ocfs2_commit_trans(osb, handle);
7241
7242out:
7243 return ret;
7244}
Tao Mae80de362011-05-23 10:36:43 +08007245
7246static int ocfs2_trim_extent(struct super_block *sb,
7247 struct ocfs2_group_desc *gd,
7248 u32 start, u32 count)
7249{
7250 u64 discard, bcount;
7251
7252 bcount = ocfs2_clusters_to_blocks(sb, count);
7253 discard = le64_to_cpu(gd->bg_blkno) +
7254 ocfs2_clusters_to_blocks(sb, start);
7255
Tao Ma10fca352011-05-23 10:36:45 +08007256 trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7257
Tao Mae80de362011-05-23 10:36:43 +08007258 return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7259}
7260
7261static int ocfs2_trim_group(struct super_block *sb,
7262 struct ocfs2_group_desc *gd,
7263 u32 start, u32 max, u32 minbits)
7264{
7265 int ret = 0, count = 0, next;
7266 void *bitmap = gd->bg_bitmap;
7267
7268 if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7269 return 0;
7270
Tao Ma10fca352011-05-23 10:36:45 +08007271 trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7272 start, max, minbits);
7273
Tao Mae80de362011-05-23 10:36:43 +08007274 while (start < max) {
7275 start = ocfs2_find_next_zero_bit(bitmap, max, start);
7276 if (start >= max)
7277 break;
7278 next = ocfs2_find_next_bit(bitmap, max, start);
7279
7280 if ((next - start) >= minbits) {
7281 ret = ocfs2_trim_extent(sb, gd,
7282 start, next - start);
7283 if (ret < 0) {
7284 mlog_errno(ret);
7285 break;
7286 }
7287 count += next - start;
7288 }
7289 start = next + 1;
7290
7291 if (fatal_signal_pending(current)) {
7292 count = -ERESTARTSYS;
7293 break;
7294 }
7295
7296 if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7297 break;
7298 }
7299
7300 if (ret < 0)
7301 count = ret;
7302
7303 return count;
7304}
7305
7306int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7307{
7308 struct ocfs2_super *osb = OCFS2_SB(sb);
7309 u64 start, len, trimmed, first_group, last_group, group;
7310 int ret, cnt;
7311 u32 first_bit, last_bit, minlen;
7312 struct buffer_head *main_bm_bh = NULL;
7313 struct inode *main_bm_inode = NULL;
7314 struct buffer_head *gd_bh = NULL;
7315 struct ocfs2_dinode *main_bm;
7316 struct ocfs2_group_desc *gd = NULL;
7317
7318 start = range->start >> osb->s_clustersize_bits;
7319 len = range->len >> osb->s_clustersize_bits;
7320 minlen = range->minlen >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007321
Jie Liuaa897622014-01-21 15:48:35 -08007322 if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
Tao Mae80de362011-05-23 10:36:43 +08007323 return -EINVAL;
7324
7325 main_bm_inode = ocfs2_get_system_file_inode(osb,
7326 GLOBAL_BITMAP_SYSTEM_INODE,
7327 OCFS2_INVALID_SLOT);
7328 if (!main_bm_inode) {
7329 ret = -EIO;
7330 mlog_errno(ret);
7331 goto out;
7332 }
7333
7334 mutex_lock(&main_bm_inode->i_mutex);
7335
7336 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7337 if (ret < 0) {
7338 mlog_errno(ret);
7339 goto out_mutex;
7340 }
7341 main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7342
7343 if (start >= le32_to_cpu(main_bm->i_clusters)) {
7344 ret = -EINVAL;
7345 goto out_unlock;
7346 }
7347
Jie Liuaa897622014-01-21 15:48:35 -08007348 len = range->len >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007349 if (start + len > le32_to_cpu(main_bm->i_clusters))
7350 len = le32_to_cpu(main_bm->i_clusters) - start;
7351
Tao Ma10fca352011-05-23 10:36:45 +08007352 trace_ocfs2_trim_fs(start, len, minlen);
7353
Tao Mae80de362011-05-23 10:36:43 +08007354 /* Determine first and last group to examine based on start and len */
7355 first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7356 if (first_group == osb->first_cluster_group_blkno)
7357 first_bit = start;
7358 else
7359 first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
7360 last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
7361 last_bit = osb->bitmap_cpg;
7362
Jie Liuaa897622014-01-21 15:48:35 -08007363 trimmed = 0;
Tao Mae80de362011-05-23 10:36:43 +08007364 for (group = first_group; group <= last_group;) {
7365 if (first_bit + len >= osb->bitmap_cpg)
7366 last_bit = osb->bitmap_cpg;
7367 else
7368 last_bit = first_bit + len;
7369
7370 ret = ocfs2_read_group_descriptor(main_bm_inode,
7371 main_bm, group,
7372 &gd_bh);
7373 if (ret < 0) {
7374 mlog_errno(ret);
7375 break;
7376 }
7377
7378 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
7379 cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
7380 brelse(gd_bh);
7381 gd_bh = NULL;
7382 if (cnt < 0) {
7383 ret = cnt;
7384 mlog_errno(ret);
7385 break;
7386 }
7387
7388 trimmed += cnt;
7389 len -= osb->bitmap_cpg - first_bit;
7390 first_bit = 0;
7391 if (group == osb->first_cluster_group_blkno)
7392 group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7393 else
7394 group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7395 }
7396 range->len = trimmed * sb->s_blocksize;
7397out_unlock:
7398 ocfs2_inode_unlock(main_bm_inode, 0);
7399 brelse(main_bm_bh);
7400out_mutex:
7401 mutex_unlock(&main_bm_inode->i_mutex);
7402 iput(main_bm_inode);
7403out:
7404 return ret;
7405}