blob: a3ded88718c93eec4bf9767018826c960285a07d [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);
Julia Lawall9e62dc02016-01-14 15:17:21 -0800167static const struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
Joel Beckerf99b9b72008-08-20 19:36:33 -0700168 .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
Julia Lawall9e62dc02016-01-14 15:17:21 -0800289static const 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
Julia Lawall9e62dc02016-01-14 15:17:21 -0800335static const 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
Julia Lawall9e62dc02016-01-14 15:17:21 -0800382static const struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800383 .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
Julia Lawall9e62dc02016-01-14 15:17:21 -0800428static const struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
Tao Mafe924412009-08-18 11:22:25 +0800429 .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,
Julia Lawall9e62dc02016-01-14 15:17:21 -0800441 const 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)) {
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700911 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700912 "Extent block #%llu has bad signature %.*s\n",
913 (unsigned long long)bh->b_blocknr, 7,
914 eb->h_signature);
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700915 goto bail;
Joel Becker5e965812008-11-13 14:49:16 -0800916 }
917
918 if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700919 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700920 "Extent block #%llu has an invalid h_blkno of %llu\n",
921 (unsigned long long)bh->b_blocknr,
922 (unsigned long long)le64_to_cpu(eb->h_blkno));
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700923 goto bail;
Joel Becker5e965812008-11-13 14:49:16 -0800924 }
925
926 if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700927 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700928 "Extent block #%llu has an invalid h_fs_generation of #%u\n",
929 (unsigned long long)bh->b_blocknr,
930 le32_to_cpu(eb->h_fs_generation));
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700931 goto bail;
Joel Becker5e965812008-11-13 14:49:16 -0800932 }
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700933bail:
934 return rc;
Joel Becker5e965812008-11-13 14:49:16 -0800935}
936
Joel Becker3d03a302009-02-12 17:49:26 -0800937int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
Joel Becker5e965812008-11-13 14:49:16 -0800938 struct buffer_head **bh)
939{
940 int rc;
941 struct buffer_head *tmp = *bh;
942
Joel Becker3d03a302009-02-12 17:49:26 -0800943 rc = ocfs2_read_block(ci, eb_blkno, &tmp,
Joel Becker970e4932008-11-13 14:49:19 -0800944 ocfs2_validate_extent_block);
Joel Becker5e965812008-11-13 14:49:16 -0800945
946 /* If ocfs2_read_block() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -0800947 if (!rc && !*bh)
Joel Becker5e965812008-11-13 14:49:16 -0800948 *bh = tmp;
949
Joel Becker5e965812008-11-13 14:49:16 -0800950 return rc;
951}
952
953
Mark Fashehccd979b2005-12-15 14:31:24 -0800954/*
955 * How many free extents have we got before we need more meta data?
956 */
957int ocfs2_num_free_extents(struct ocfs2_super *osb,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700958 struct ocfs2_extent_tree *et)
Mark Fashehccd979b2005-12-15 14:31:24 -0800959{
960 int retval;
Tao Mae7d4cb62008-08-18 17:38:44 +0800961 struct ocfs2_extent_list *el = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 struct ocfs2_extent_block *eb;
963 struct buffer_head *eb_bh = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +0800964 u64 last_eb_blk = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800965
Joel Beckerf99b9b72008-08-20 19:36:33 -0700966 el = et->et_root_el;
967 last_eb_blk = ocfs2_et_get_last_eb_blk(et);
Mark Fashehccd979b2005-12-15 14:31:24 -0800968
Tao Mae7d4cb62008-08-18 17:38:44 +0800969 if (last_eb_blk) {
Joel Becker3d03a302009-02-12 17:49:26 -0800970 retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
971 &eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800972 if (retval < 0) {
973 mlog_errno(retval);
974 goto bail;
975 }
976 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
977 el = &eb->h_list;
Tao Mae7d4cb62008-08-18 17:38:44 +0800978 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800979
980 BUG_ON(el->l_tree_depth != 0);
981
982 retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
983bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700984 brelse(eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800985
Tao Maa09d09b2011-02-22 08:21:10 +0800986 trace_ocfs2_num_free_extents(retval);
Mark Fashehccd979b2005-12-15 14:31:24 -0800987 return retval;
988}
989
990/* expects array to already be allocated
991 *
992 * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
993 * l_count for you
994 */
Joel Becker42a5a7a2009-02-12 18:49:19 -0800995static int ocfs2_create_new_meta_bhs(handle_t *handle,
996 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -0800997 int wanted,
998 struct ocfs2_alloc_context *meta_ac,
999 struct buffer_head *bhs[])
1000{
1001 int count, status, i;
1002 u16 suballoc_bit_start;
1003 u32 num_got;
Joel Becker2b6cb572010-03-26 10:09:15 +08001004 u64 suballoc_loc, first_blkno;
Joel Becker42a5a7a2009-02-12 18:49:19 -08001005 struct ocfs2_super *osb =
1006 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001007 struct ocfs2_extent_block *eb;
1008
Mark Fashehccd979b2005-12-15 14:31:24 -08001009 count = 0;
1010 while (count < wanted) {
Joel Becker1ed9b772010-05-06 13:59:06 +08001011 status = ocfs2_claim_metadata(handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001012 meta_ac,
1013 wanted - count,
Joel Becker2b6cb572010-03-26 10:09:15 +08001014 &suballoc_loc,
Mark Fashehccd979b2005-12-15 14:31:24 -08001015 &suballoc_bit_start,
1016 &num_got,
1017 &first_blkno);
1018 if (status < 0) {
1019 mlog_errno(status);
1020 goto bail;
1021 }
1022
1023 for(i = count; i < (num_got + count); i++) {
1024 bhs[i] = sb_getblk(osb->sb, first_blkno);
1025 if (bhs[i] == NULL) {
Rui Xiang7391a292013-11-12 15:06:54 -08001026 status = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -08001027 mlog_errno(status);
1028 goto bail;
1029 }
Joel Becker42a5a7a2009-02-12 18:49:19 -08001030 ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001031
Joel Becker42a5a7a2009-02-12 18:49:19 -08001032 status = ocfs2_journal_access_eb(handle, et->et_ci,
1033 bhs[i],
Joel Becker13723d02008-10-17 19:25:01 -07001034 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001035 if (status < 0) {
1036 mlog_errno(status);
1037 goto bail;
1038 }
1039
1040 memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
1041 eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
1042 /* Ok, setup the minimal stuff here. */
1043 strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
1044 eb->h_blkno = cpu_to_le64(first_blkno);
1045 eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
Tiger Yangb89c5422010-01-25 14:11:06 +08001046 eb->h_suballoc_slot =
1047 cpu_to_le16(meta_ac->ac_alloc_slot);
Joel Becker2b6cb572010-03-26 10:09:15 +08001048 eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
Mark Fashehccd979b2005-12-15 14:31:24 -08001049 eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1050 eb->h_list.l_count =
1051 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
1052
1053 suballoc_bit_start++;
1054 first_blkno++;
1055
1056 /* We'll also be dirtied by the caller, so
1057 * this isn't absolutely necessary. */
Joel Beckerec20cec2010-03-19 14:13:52 -07001058 ocfs2_journal_dirty(handle, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001059 }
1060
1061 count += num_got;
1062 }
1063
1064 status = 0;
1065bail:
1066 if (status < 0) {
1067 for(i = 0; i < wanted; i++) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001068 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001069 bhs[i] = NULL;
1070 }
Tao Mac1e8d352011-03-07 16:43:21 +08001071 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001072 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001073 return status;
1074}
1075
1076/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001077 * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
1078 *
1079 * Returns the sum of the rightmost extent rec logical offset and
1080 * cluster count.
1081 *
1082 * ocfs2_add_branch() uses this to determine what logical cluster
1083 * value should be populated into the leftmost new branch records.
1084 *
1085 * ocfs2_shift_tree_depth() uses this to determine the # clusters
1086 * value for the new topmost tree record.
1087 */
1088static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el)
1089{
1090 int i;
1091
1092 i = le16_to_cpu(el->l_next_free_rec) - 1;
1093
1094 return le32_to_cpu(el->l_recs[i].e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001095 ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08001096}
1097
1098/*
Tao Ma6b791bc2009-06-12 14:18:36 +08001099 * Change range of the branches in the right most path according to the leaf
1100 * extent block's rightmost record.
1101 */
1102static int ocfs2_adjust_rightmost_branch(handle_t *handle,
Tao Ma6b791bc2009-06-12 14:18:36 +08001103 struct ocfs2_extent_tree *et)
1104{
1105 int status;
1106 struct ocfs2_path *path = NULL;
1107 struct ocfs2_extent_list *el;
1108 struct ocfs2_extent_rec *rec;
1109
1110 path = ocfs2_new_path_from_et(et);
1111 if (!path) {
1112 status = -ENOMEM;
1113 return status;
1114 }
1115
Joel Beckerfacdb772009-02-12 18:08:48 -08001116 status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
Tao Ma6b791bc2009-06-12 14:18:36 +08001117 if (status < 0) {
1118 mlog_errno(status);
1119 goto out;
1120 }
1121
Tao Mac901fb02010-04-26 14:34:57 +08001122 status = ocfs2_extend_trans(handle, path_num_items(path));
Tao Ma6b791bc2009-06-12 14:18:36 +08001123 if (status < 0) {
1124 mlog_errno(status);
1125 goto out;
1126 }
1127
Joel Beckerd401dc12009-02-13 02:24:10 -08001128 status = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma6b791bc2009-06-12 14:18:36 +08001129 if (status < 0) {
1130 mlog_errno(status);
1131 goto out;
1132 }
1133
1134 el = path_leaf_el(path);
Al Viro3a251f02012-04-13 12:22:00 -04001135 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1];
Tao Ma6b791bc2009-06-12 14:18:36 +08001136
Joel Beckerd401dc12009-02-13 02:24:10 -08001137 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Tao Ma6b791bc2009-06-12 14:18:36 +08001138
1139out:
1140 ocfs2_free_path(path);
1141 return status;
1142}
1143
1144/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001145 * Add an entire tree branch to our inode. eb_bh is the extent block
Joel Beckerd401dc12009-02-13 02:24:10 -08001146 * to start at, if we don't want to start the branch at the root
Mark Fashehccd979b2005-12-15 14:31:24 -08001147 * structure.
1148 *
1149 * last_eb_bh is required as we have to update it's next_leaf pointer
1150 * for the new last extent block.
1151 *
1152 * the new branch will be 'empty' in the sense that every block will
Mark Fashehe48edee2007-03-07 16:46:57 -08001153 * contain a single record with cluster count == 0.
Mark Fashehccd979b2005-12-15 14:31:24 -08001154 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001155static int ocfs2_add_branch(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001156 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001157 struct buffer_head *eb_bh,
Mark Fasheh328d5752007-06-18 10:48:04 -07001158 struct buffer_head **last_eb_bh,
Mark Fashehccd979b2005-12-15 14:31:24 -08001159 struct ocfs2_alloc_context *meta_ac)
1160{
1161 int status, new_blocks, i;
1162 u64 next_blkno, new_last_eb_blk;
1163 struct buffer_head *bh;
1164 struct buffer_head **new_eb_bhs = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001165 struct ocfs2_extent_block *eb;
1166 struct ocfs2_extent_list *eb_el;
1167 struct ocfs2_extent_list *el;
Tao Ma6b791bc2009-06-12 14:18:36 +08001168 u32 new_cpos, root_end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001169
Mark Fasheh328d5752007-06-18 10:48:04 -07001170 BUG_ON(!last_eb_bh || !*last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001171
Mark Fashehccd979b2005-12-15 14:31:24 -08001172 if (eb_bh) {
1173 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1174 el = &eb->h_list;
1175 } else
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001176 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001177
1178 /* we never add a branch to a leaf. */
1179 BUG_ON(!el->l_tree_depth);
1180
1181 new_blocks = le16_to_cpu(el->l_tree_depth);
1182
Tao Ma6b791bc2009-06-12 14:18:36 +08001183 eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
1184 new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
1185 root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
1186
1187 /*
1188 * If there is a gap before the root end and the real end
1189 * of the righmost leaf block, we need to remove the gap
1190 * between new_cpos and root_end first so that the tree
1191 * is consistent after we add a new branch(it will start
1192 * from new_cpos).
1193 */
1194 if (root_end > new_cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08001195 trace_ocfs2_adjust_rightmost_branch(
1196 (unsigned long long)
1197 ocfs2_metadata_cache_owner(et->et_ci),
1198 root_end, new_cpos);
1199
Joel Beckerd401dc12009-02-13 02:24:10 -08001200 status = ocfs2_adjust_rightmost_branch(handle, et);
Tao Ma6b791bc2009-06-12 14:18:36 +08001201 if (status) {
1202 mlog_errno(status);
1203 goto bail;
1204 }
1205 }
1206
Mark Fashehccd979b2005-12-15 14:31:24 -08001207 /* allocate the number of new eb blocks we need */
1208 new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
1209 GFP_KERNEL);
1210 if (!new_eb_bhs) {
1211 status = -ENOMEM;
1212 mlog_errno(status);
1213 goto bail;
1214 }
1215
Joel Becker42a5a7a2009-02-12 18:49:19 -08001216 status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
Mark Fashehccd979b2005-12-15 14:31:24 -08001217 meta_ac, new_eb_bhs);
1218 if (status < 0) {
1219 mlog_errno(status);
1220 goto bail;
1221 }
1222
1223 /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
1224 * linked with the rest of the tree.
1225 * conversly, new_eb_bhs[0] is the new bottommost leaf.
1226 *
1227 * when we leave the loop, new_last_eb_blk will point to the
1228 * newest leaf, and next_blkno will point to the topmost extent
1229 * block. */
1230 next_blkno = new_last_eb_blk = 0;
1231 for(i = 0; i < new_blocks; i++) {
1232 bh = new_eb_bhs[i];
1233 eb = (struct ocfs2_extent_block *) bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001234 /* ocfs2_create_new_meta_bhs() should create it right! */
1235 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001236 eb_el = &eb->h_list;
1237
Joel Beckerd401dc12009-02-13 02:24:10 -08001238 status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
Joel Becker13723d02008-10-17 19:25:01 -07001239 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001240 if (status < 0) {
1241 mlog_errno(status);
1242 goto bail;
1243 }
1244
1245 eb->h_next_leaf_blk = 0;
1246 eb_el->l_tree_depth = cpu_to_le16(i);
1247 eb_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehdcd05382007-01-16 11:32:23 -08001248 /*
1249 * This actually counts as an empty extent as
1250 * c_clusters == 0
1251 */
1252 eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001253 eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehe48edee2007-03-07 16:46:57 -08001254 /*
1255 * eb_el isn't always an interior node, but even leaf
1256 * nodes want a zero'd flags and reserved field so
1257 * this gets the whole 32 bits regardless of use.
1258 */
1259 eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001260 if (!eb_el->l_tree_depth)
1261 new_last_eb_blk = le64_to_cpu(eb->h_blkno);
1262
Joel Beckerec20cec2010-03-19 14:13:52 -07001263 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001264 next_blkno = le64_to_cpu(eb->h_blkno);
1265 }
1266
1267 /* This is a bit hairy. We want to update up to three blocks
1268 * here without leaving any of them in an inconsistent state
1269 * in case of error. We don't have to worry about
1270 * journal_dirty erroring as it won't unless we've aborted the
1271 * handle (in which case we would never be here) so reserving
1272 * the write with journal_access is all we need to do. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001273 status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001274 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001275 if (status < 0) {
1276 mlog_errno(status);
1277 goto bail;
1278 }
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001279 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001280 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001281 if (status < 0) {
1282 mlog_errno(status);
1283 goto bail;
1284 }
1285 if (eb_bh) {
Joel Beckerd401dc12009-02-13 02:24:10 -08001286 status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001287 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001288 if (status < 0) {
1289 mlog_errno(status);
1290 goto bail;
1291 }
1292 }
1293
1294 /* Link the new branch into the rest of the tree (el will
Tao Mae7d4cb62008-08-18 17:38:44 +08001295 * either be on the root_bh, or the extent block passed in. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001296 i = le16_to_cpu(el->l_next_free_rec);
1297 el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08001298 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001299 el->l_recs[i].e_int_clusters = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001300 le16_add_cpu(&el->l_next_free_rec, 1);
1301
1302 /* fe needs a new last extent block pointer, as does the
1303 * next_leaf on the previously last-extent-block. */
Joel Becker35dc0aa2008-08-20 16:25:06 -07001304 ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
Mark Fashehccd979b2005-12-15 14:31:24 -08001305
Mark Fasheh328d5752007-06-18 10:48:04 -07001306 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001307 eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
1308
Joel Beckerec20cec2010-03-19 14:13:52 -07001309 ocfs2_journal_dirty(handle, *last_eb_bh);
1310 ocfs2_journal_dirty(handle, et->et_root_bh);
1311 if (eb_bh)
1312 ocfs2_journal_dirty(handle, eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001313
Mark Fasheh328d5752007-06-18 10:48:04 -07001314 /*
1315 * Some callers want to track the rightmost leaf so pass it
1316 * back here.
1317 */
1318 brelse(*last_eb_bh);
1319 get_bh(new_eb_bhs[0]);
1320 *last_eb_bh = new_eb_bhs[0];
1321
Mark Fashehccd979b2005-12-15 14:31:24 -08001322 status = 0;
1323bail:
1324 if (new_eb_bhs) {
1325 for (i = 0; i < new_blocks; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001326 brelse(new_eb_bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001327 kfree(new_eb_bhs);
1328 }
1329
Mark Fashehccd979b2005-12-15 14:31:24 -08001330 return status;
1331}
1332
1333/*
1334 * adds another level to the allocation tree.
1335 * returns back the new extent block so you can add a branch to it
1336 * after this call.
1337 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001338static int ocfs2_shift_tree_depth(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001339 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001340 struct ocfs2_alloc_context *meta_ac,
1341 struct buffer_head **ret_new_eb_bh)
1342{
1343 int status, i;
Mark Fashehdcd05382007-01-16 11:32:23 -08001344 u32 new_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001345 struct buffer_head *new_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001346 struct ocfs2_extent_block *eb;
Tao Mae7d4cb62008-08-18 17:38:44 +08001347 struct ocfs2_extent_list *root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001348 struct ocfs2_extent_list *eb_el;
1349
Joel Becker42a5a7a2009-02-12 18:49:19 -08001350 status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
Mark Fashehccd979b2005-12-15 14:31:24 -08001351 &new_eb_bh);
1352 if (status < 0) {
1353 mlog_errno(status);
1354 goto bail;
1355 }
1356
1357 eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001358 /* ocfs2_create_new_meta_bhs() should create it right! */
1359 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001360
1361 eb_el = &eb->h_list;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001362 root_el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001363
Joel Beckerd401dc12009-02-13 02:24:10 -08001364 status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001365 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001366 if (status < 0) {
1367 mlog_errno(status);
1368 goto bail;
1369 }
1370
Tao Mae7d4cb62008-08-18 17:38:44 +08001371 /* copy the root extent list data into the new extent block */
1372 eb_el->l_tree_depth = root_el->l_tree_depth;
1373 eb_el->l_next_free_rec = root_el->l_next_free_rec;
1374 for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1375 eb_el->l_recs[i] = root_el->l_recs[i];
Mark Fashehccd979b2005-12-15 14:31:24 -08001376
Joel Beckerec20cec2010-03-19 14:13:52 -07001377 ocfs2_journal_dirty(handle, new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001378
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001379 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001380 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001381 if (status < 0) {
1382 mlog_errno(status);
1383 goto bail;
1384 }
1385
Mark Fashehdcd05382007-01-16 11:32:23 -08001386 new_clusters = ocfs2_sum_rightmost_rec(eb_el);
1387
Tao Mae7d4cb62008-08-18 17:38:44 +08001388 /* update root_bh now */
1389 le16_add_cpu(&root_el->l_tree_depth, 1);
1390 root_el->l_recs[0].e_cpos = 0;
1391 root_el->l_recs[0].e_blkno = eb->h_blkno;
1392 root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
1393 for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1394 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
1395 root_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001396
1397 /* If this is our 1st tree depth shift, then last_eb_blk
1398 * becomes the allocated extent block */
Tao Mae7d4cb62008-08-18 17:38:44 +08001399 if (root_el->l_tree_depth == cpu_to_le16(1))
Joel Becker35dc0aa2008-08-20 16:25:06 -07001400 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001401
Joel Beckerec20cec2010-03-19 14:13:52 -07001402 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001403
1404 *ret_new_eb_bh = new_eb_bh;
1405 new_eb_bh = NULL;
1406 status = 0;
1407bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001408 brelse(new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001409
Mark Fashehccd979b2005-12-15 14:31:24 -08001410 return status;
1411}
1412
1413/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001414 * Should only be called when there is no space left in any of the
1415 * leaf nodes. What we want to do is find the lowest tree depth
1416 * non-leaf extent block with room for new records. There are three
1417 * valid results of this search:
1418 *
1419 * 1) a lowest extent block is found, then we pass it back in
1420 * *lowest_eb_bh and return '0'
1421 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001422 * 2) the search fails to find anything, but the root_el has room. We
Mark Fashehccd979b2005-12-15 14:31:24 -08001423 * pass NULL back in *lowest_eb_bh, but still return '0'
1424 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001425 * 3) the search fails to find anything AND the root_el is full, in
Mark Fashehccd979b2005-12-15 14:31:24 -08001426 * which case we return > 0
1427 *
1428 * return status < 0 indicates an error.
1429 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001430static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001431 struct buffer_head **target_bh)
1432{
1433 int status = 0, i;
1434 u64 blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08001435 struct ocfs2_extent_block *eb;
1436 struct ocfs2_extent_list *el;
1437 struct buffer_head *bh = NULL;
1438 struct buffer_head *lowest_bh = NULL;
1439
Mark Fashehccd979b2005-12-15 14:31:24 -08001440 *target_bh = NULL;
1441
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001442 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001443
1444 while(le16_to_cpu(el->l_tree_depth) > 1) {
1445 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08001446 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001447 "Owner %llu has empty extent list (next_free_rec == 0)\n",
Joel Becker3d03a302009-02-12 17:49:26 -08001448 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001449 status = -EIO;
1450 goto bail;
1451 }
1452 i = le16_to_cpu(el->l_next_free_rec) - 1;
1453 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1454 if (!blkno) {
Joel Becker3d03a302009-02-12 17:49:26 -08001455 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001456 "Owner %llu has extent list where extent # %d has no physical block start\n",
Joel Becker3d03a302009-02-12 17:49:26 -08001457 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
Mark Fashehccd979b2005-12-15 14:31:24 -08001458 status = -EIO;
1459 goto bail;
1460 }
1461
Mark Fasheha81cb882008-10-07 14:25:16 -07001462 brelse(bh);
1463 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001464
Joel Becker3d03a302009-02-12 17:49:26 -08001465 status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001466 if (status < 0) {
1467 mlog_errno(status);
1468 goto bail;
1469 }
1470
1471 eb = (struct ocfs2_extent_block *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001472 el = &eb->h_list;
1473
1474 if (le16_to_cpu(el->l_next_free_rec) <
1475 le16_to_cpu(el->l_count)) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001476 brelse(lowest_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001477 lowest_bh = bh;
1478 get_bh(lowest_bh);
1479 }
1480 }
1481
1482 /* If we didn't find one and the fe doesn't have any room,
1483 * then return '1' */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001484 el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001485 if (!lowest_bh && (el->l_next_free_rec == el->l_count))
Mark Fashehccd979b2005-12-15 14:31:24 -08001486 status = 1;
1487
1488 *target_bh = lowest_bh;
1489bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001490 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001491
Mark Fashehccd979b2005-12-15 14:31:24 -08001492 return status;
1493}
1494
Mark Fashehe48edee2007-03-07 16:46:57 -08001495/*
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001496 * Grow a b-tree so that it has more records.
1497 *
1498 * We might shift the tree depth in which case existing paths should
1499 * be considered invalid.
1500 *
1501 * Tree depth after the grow is returned via *final_depth.
Mark Fasheh328d5752007-06-18 10:48:04 -07001502 *
1503 * *last_eb_bh will be updated by ocfs2_add_branch().
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001504 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001505static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
1506 int *final_depth, struct buffer_head **last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001507 struct ocfs2_alloc_context *meta_ac)
1508{
1509 int ret, shift;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001510 struct ocfs2_extent_list *el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001511 int depth = le16_to_cpu(el->l_tree_depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001512 struct buffer_head *bh = NULL;
1513
1514 BUG_ON(meta_ac == NULL);
1515
Joel Beckerd401dc12009-02-13 02:24:10 -08001516 shift = ocfs2_find_branch_target(et, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001517 if (shift < 0) {
1518 ret = shift;
1519 mlog_errno(ret);
1520 goto out;
1521 }
1522
1523 /* We traveled all the way to the bottom of the allocation tree
1524 * and didn't find room for any more extents - we need to add
1525 * another tree level */
1526 if (shift) {
1527 BUG_ON(bh);
Tao Maa09d09b2011-02-22 08:21:10 +08001528 trace_ocfs2_grow_tree(
1529 (unsigned long long)
1530 ocfs2_metadata_cache_owner(et->et_ci),
1531 depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001532
1533 /* ocfs2_shift_tree_depth will return us a buffer with
1534 * the new extent block (so we can pass that to
1535 * ocfs2_add_branch). */
Joel Beckerd401dc12009-02-13 02:24:10 -08001536 ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001537 if (ret < 0) {
1538 mlog_errno(ret);
1539 goto out;
1540 }
1541 depth++;
Mark Fasheh328d5752007-06-18 10:48:04 -07001542 if (depth == 1) {
1543 /*
1544 * Special case: we have room now if we shifted from
1545 * tree_depth 0, so no more work needs to be done.
1546 *
1547 * We won't be calling add_branch, so pass
1548 * back *last_eb_bh as the new leaf. At depth
1549 * zero, it should always be null so there's
1550 * no reason to brelse.
1551 */
1552 BUG_ON(*last_eb_bh);
1553 get_bh(bh);
1554 *last_eb_bh = bh;
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001555 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07001556 }
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001557 }
1558
1559 /* call ocfs2_add_branch to add the final part of the tree with
1560 * the new data. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001561 ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001562 meta_ac);
1563 if (ret < 0) {
1564 mlog_errno(ret);
1565 goto out;
1566 }
1567
1568out:
1569 if (final_depth)
1570 *final_depth = depth;
1571 brelse(bh);
1572 return ret;
1573}
1574
1575/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001576 * This function will discard the rightmost extent record.
1577 */
1578static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
1579{
1580 int next_free = le16_to_cpu(el->l_next_free_rec);
1581 int count = le16_to_cpu(el->l_count);
1582 unsigned int num_bytes;
1583
1584 BUG_ON(!next_free);
1585 /* This will cause us to go off the end of our extent list. */
1586 BUG_ON(next_free >= count);
1587
1588 num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
1589
1590 memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
1591}
1592
1593static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
1594 struct ocfs2_extent_rec *insert_rec)
1595{
1596 int i, insert_index, next_free, has_empty, num_bytes;
1597 u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
1598 struct ocfs2_extent_rec *rec;
1599
1600 next_free = le16_to_cpu(el->l_next_free_rec);
1601 has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
1602
1603 BUG_ON(!next_free);
1604
1605 /* The tree code before us didn't allow enough room in the leaf. */
Julia Lawallb1f35502008-03-04 15:21:05 -08001606 BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
Mark Fashehdcd05382007-01-16 11:32:23 -08001607
1608 /*
1609 * The easiest way to approach this is to just remove the
1610 * empty extent and temporarily decrement next_free.
1611 */
1612 if (has_empty) {
1613 /*
1614 * If next_free was 1 (only an empty extent), this
1615 * loop won't execute, which is fine. We still want
1616 * the decrement above to happen.
1617 */
1618 for(i = 0; i < (next_free - 1); i++)
1619 el->l_recs[i] = el->l_recs[i+1];
1620
1621 next_free--;
1622 }
1623
1624 /*
1625 * Figure out what the new record index should be.
1626 */
1627 for(i = 0; i < next_free; i++) {
1628 rec = &el->l_recs[i];
1629
1630 if (insert_cpos < le32_to_cpu(rec->e_cpos))
1631 break;
1632 }
1633 insert_index = i;
1634
Tao Maa09d09b2011-02-22 08:21:10 +08001635 trace_ocfs2_rotate_leaf(insert_cpos, insert_index,
1636 has_empty, next_free,
1637 le16_to_cpu(el->l_count));
Mark Fashehdcd05382007-01-16 11:32:23 -08001638
1639 BUG_ON(insert_index < 0);
1640 BUG_ON(insert_index >= le16_to_cpu(el->l_count));
1641 BUG_ON(insert_index > next_free);
1642
1643 /*
1644 * No need to memmove if we're just adding to the tail.
1645 */
1646 if (insert_index != next_free) {
1647 BUG_ON(next_free >= le16_to_cpu(el->l_count));
1648
1649 num_bytes = next_free - insert_index;
1650 num_bytes *= sizeof(struct ocfs2_extent_rec);
1651 memmove(&el->l_recs[insert_index + 1],
1652 &el->l_recs[insert_index],
1653 num_bytes);
1654 }
1655
1656 /*
1657 * Either we had an empty extent, and need to re-increment or
1658 * there was no empty extent on a non full rightmost leaf node,
1659 * in which case we still need to increment.
1660 */
1661 next_free++;
1662 el->l_next_free_rec = cpu_to_le16(next_free);
1663 /*
1664 * Make sure none of the math above just messed up our tree.
1665 */
1666 BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
1667
1668 el->l_recs[insert_index] = *insert_rec;
1669
1670}
1671
Mark Fasheh328d5752007-06-18 10:48:04 -07001672static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
1673{
1674 int size, num_recs = le16_to_cpu(el->l_next_free_rec);
1675
1676 BUG_ON(num_recs == 0);
1677
1678 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
1679 num_recs--;
1680 size = num_recs * sizeof(struct ocfs2_extent_rec);
1681 memmove(&el->l_recs[0], &el->l_recs[1], size);
1682 memset(&el->l_recs[num_recs], 0,
1683 sizeof(struct ocfs2_extent_rec));
1684 el->l_next_free_rec = cpu_to_le16(num_recs);
1685 }
1686}
1687
Mark Fashehdcd05382007-01-16 11:32:23 -08001688/*
1689 * Create an empty extent record .
1690 *
1691 * l_next_free_rec may be updated.
1692 *
1693 * If an empty extent already exists do nothing.
1694 */
1695static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
1696{
1697 int next_free = le16_to_cpu(el->l_next_free_rec);
1698
Mark Fashehe48edee2007-03-07 16:46:57 -08001699 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
1700
Mark Fashehdcd05382007-01-16 11:32:23 -08001701 if (next_free == 0)
1702 goto set_and_inc;
1703
1704 if (ocfs2_is_empty_extent(&el->l_recs[0]))
1705 return;
1706
1707 mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
1708 "Asked to create an empty extent in a full list:\n"
1709 "count = %u, tree depth = %u",
1710 le16_to_cpu(el->l_count),
1711 le16_to_cpu(el->l_tree_depth));
1712
1713 ocfs2_shift_records_right(el);
1714
1715set_and_inc:
1716 le16_add_cpu(&el->l_next_free_rec, 1);
1717 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
1718}
1719
1720/*
1721 * For a rotation which involves two leaf nodes, the "root node" is
1722 * the lowest level tree node which contains a path to both leafs. This
1723 * resulting set of information can be used to form a complete "subtree"
1724 *
1725 * This function is passed two full paths from the dinode down to a
1726 * pair of adjacent leaves. It's task is to figure out which path
1727 * index contains the subtree root - this can be the root index itself
1728 * in a worst-case rotation.
1729 *
1730 * The array index of the subtree root is passed back.
1731 */
Tao Ma38a04e42009-11-30 14:32:19 +08001732int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
1733 struct ocfs2_path *left,
1734 struct ocfs2_path *right)
Mark Fashehdcd05382007-01-16 11:32:23 -08001735{
1736 int i = 0;
1737
1738 /*
1739 * Check that the caller passed in two paths from the same tree.
1740 */
1741 BUG_ON(path_root_bh(left) != path_root_bh(right));
1742
1743 do {
1744 i++;
1745
1746 /*
1747 * The caller didn't pass two adjacent paths.
1748 */
1749 mlog_bug_on_msg(i > left->p_tree_depth,
Joel Becker7dc02802009-02-12 19:20:13 -08001750 "Owner %llu, left depth %u, right depth %u\n"
Mark Fashehdcd05382007-01-16 11:32:23 -08001751 "left leaf blk %llu, right leaf blk %llu\n",
Joel Becker7dc02802009-02-12 19:20:13 -08001752 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
1753 left->p_tree_depth, right->p_tree_depth,
Mark Fashehdcd05382007-01-16 11:32:23 -08001754 (unsigned long long)path_leaf_bh(left)->b_blocknr,
1755 (unsigned long long)path_leaf_bh(right)->b_blocknr);
1756 } while (left->p_node[i].bh->b_blocknr ==
1757 right->p_node[i].bh->b_blocknr);
1758
1759 return i - 1;
1760}
1761
1762typedef void (path_insert_t)(void *, struct buffer_head *);
1763
1764/*
1765 * Traverse a btree path in search of cpos, starting at root_el.
1766 *
1767 * This code can be called with a cpos larger than the tree, in which
1768 * case it will return the rightmost path.
1769 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001770static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
Mark Fashehdcd05382007-01-16 11:32:23 -08001771 struct ocfs2_extent_list *root_el, u32 cpos,
1772 path_insert_t *func, void *data)
1773{
1774 int i, ret = 0;
1775 u32 range;
1776 u64 blkno;
1777 struct buffer_head *bh = NULL;
1778 struct ocfs2_extent_block *eb;
1779 struct ocfs2_extent_list *el;
1780 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08001781
1782 el = root_el;
1783 while (el->l_tree_depth) {
1784 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001785 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001786 "Owner %llu has empty extent list at depth %u\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001787 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001788 le16_to_cpu(el->l_tree_depth));
1789 ret = -EROFS;
1790 goto out;
1791
1792 }
1793
1794 for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
1795 rec = &el->l_recs[i];
1796
1797 /*
1798 * In the case that cpos is off the allocation
1799 * tree, this should just wind up returning the
1800 * rightmost record.
1801 */
1802 range = le32_to_cpu(rec->e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001803 ocfs2_rec_clusters(el, rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08001804 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
1805 break;
1806 }
1807
1808 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1809 if (blkno == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001810 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001811 "Owner %llu has bad blkno in extent list at depth %u (index %d)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001812 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001813 le16_to_cpu(el->l_tree_depth), i);
1814 ret = -EROFS;
1815 goto out;
1816 }
1817
1818 brelse(bh);
1819 bh = NULL;
Joel Beckerfacdb772009-02-12 18:08:48 -08001820 ret = ocfs2_read_extent_block(ci, blkno, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001821 if (ret) {
1822 mlog_errno(ret);
1823 goto out;
1824 }
1825
1826 eb = (struct ocfs2_extent_block *) bh->b_data;
1827 el = &eb->h_list;
Mark Fashehdcd05382007-01-16 11:32:23 -08001828
1829 if (le16_to_cpu(el->l_next_free_rec) >
1830 le16_to_cpu(el->l_count)) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001831 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001832 "Owner %llu has bad count in extent list at block %llu (next free=%u, count=%u)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001833 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001834 (unsigned long long)bh->b_blocknr,
1835 le16_to_cpu(el->l_next_free_rec),
1836 le16_to_cpu(el->l_count));
1837 ret = -EROFS;
1838 goto out;
1839 }
1840
1841 if (func)
1842 func(data, bh);
1843 }
1844
1845out:
1846 /*
1847 * Catch any trailing bh that the loop didn't handle.
1848 */
1849 brelse(bh);
1850
1851 return ret;
1852}
1853
1854/*
1855 * Given an initialized path (that is, it has a valid root extent
1856 * list), this function will traverse the btree in search of the path
1857 * which would contain cpos.
1858 *
1859 * The path traveled is recorded in the path structure.
1860 *
1861 * Note that this will not do any comparisons on leaf node extent
1862 * records, so it will work fine in the case that we just added a tree
1863 * branch.
1864 */
1865struct find_path_data {
1866 int index;
1867 struct ocfs2_path *path;
1868};
1869static void find_path_ins(void *data, struct buffer_head *bh)
1870{
1871 struct find_path_data *fp = data;
1872
1873 get_bh(bh);
1874 ocfs2_path_insert_eb(fp->path, fp->index, bh);
1875 fp->index++;
1876}
Tao Mae2e9f602009-08-18 11:22:34 +08001877int ocfs2_find_path(struct ocfs2_caching_info *ci,
1878 struct ocfs2_path *path, u32 cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08001879{
1880 struct find_path_data data;
1881
1882 data.index = 1;
1883 data.path = path;
Joel Beckerfacdb772009-02-12 18:08:48 -08001884 return __ocfs2_find_path(ci, path_root_el(path), cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08001885 find_path_ins, &data);
1886}
1887
1888static void find_leaf_ins(void *data, struct buffer_head *bh)
1889{
1890 struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
1891 struct ocfs2_extent_list *el = &eb->h_list;
1892 struct buffer_head **ret = data;
1893
1894 /* We want to retain only the leaf block. */
1895 if (le16_to_cpu(el->l_tree_depth) == 0) {
1896 get_bh(bh);
1897 *ret = bh;
1898 }
1899}
1900/*
1901 * Find the leaf block in the tree which would contain cpos. No
1902 * checking of the actual leaf is done.
1903 *
1904 * Some paths want to call this instead of allocating a path structure
1905 * and calling ocfs2_find_path().
1906 *
1907 * This function doesn't handle non btree extent lists.
1908 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001909int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
1910 struct ocfs2_extent_list *root_el, u32 cpos,
1911 struct buffer_head **leaf_bh)
Mark Fashehdcd05382007-01-16 11:32:23 -08001912{
1913 int ret;
1914 struct buffer_head *bh = NULL;
1915
Joel Beckerfacdb772009-02-12 18:08:48 -08001916 ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001917 if (ret) {
1918 mlog_errno(ret);
1919 goto out;
1920 }
1921
1922 *leaf_bh = bh;
1923out:
1924 return ret;
1925}
1926
1927/*
1928 * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
1929 *
1930 * Basically, we've moved stuff around at the bottom of the tree and
1931 * we need to fix up the extent records above the changes to reflect
1932 * the new changes.
1933 *
1934 * left_rec: the record on the left.
1935 * left_child_el: is the child list pointed to by left_rec
1936 * right_rec: the record to the right of left_rec
1937 * right_child_el: is the child list pointed to by right_rec
1938 *
1939 * By definition, this only works on interior nodes.
1940 */
1941static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
1942 struct ocfs2_extent_list *left_child_el,
1943 struct ocfs2_extent_rec *right_rec,
1944 struct ocfs2_extent_list *right_child_el)
1945{
1946 u32 left_clusters, right_end;
1947
1948 /*
1949 * Interior nodes never have holes. Their cpos is the cpos of
1950 * the leftmost record in their child list. Their cluster
1951 * count covers the full theoretical range of their child list
1952 * - the range between their cpos and the cpos of the record
1953 * immediately to their right.
1954 */
1955 left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
Tao Ma82e12642009-07-23 08:12:58 +08001956 if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
1957 BUG_ON(right_child_el->l_tree_depth);
Mark Fasheh328d5752007-06-18 10:48:04 -07001958 BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
1959 left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
1960 }
Mark Fashehdcd05382007-01-16 11:32:23 -08001961 left_clusters -= le32_to_cpu(left_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001962 left_rec->e_int_clusters = cpu_to_le32(left_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001963
1964 /*
1965 * Calculate the rightmost cluster count boundary before
Mark Fashehe48edee2007-03-07 16:46:57 -08001966 * moving cpos - we will need to adjust clusters after
Mark Fashehdcd05382007-01-16 11:32:23 -08001967 * updating e_cpos to keep the same highest cluster count.
1968 */
1969 right_end = le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001970 right_end += le32_to_cpu(right_rec->e_int_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001971
1972 right_rec->e_cpos = left_rec->e_cpos;
1973 le32_add_cpu(&right_rec->e_cpos, left_clusters);
1974
1975 right_end -= le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001976 right_rec->e_int_clusters = cpu_to_le32(right_end);
Mark Fashehdcd05382007-01-16 11:32:23 -08001977}
1978
1979/*
1980 * Adjust the adjacent root node records involved in a
1981 * rotation. left_el_blkno is passed in as a key so that we can easily
1982 * find it's index in the root list.
1983 */
1984static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
1985 struct ocfs2_extent_list *left_el,
1986 struct ocfs2_extent_list *right_el,
1987 u64 left_el_blkno)
1988{
1989 int i;
1990
1991 BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
1992 le16_to_cpu(left_el->l_tree_depth));
1993
1994 for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
1995 if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
1996 break;
1997 }
1998
1999 /*
2000 * The path walking code should have never returned a root and
2001 * two paths which are not adjacent.
2002 */
2003 BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
2004
2005 ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
2006 &root_el->l_recs[i + 1], right_el);
2007}
2008
2009/*
2010 * We've changed a leaf block (in right_path) and need to reflect that
2011 * change back up the subtree.
2012 *
2013 * This happens in multiple places:
2014 * - When we've moved an extent record from the left path leaf to the right
2015 * path leaf to make room for an empty extent in the left path leaf.
2016 * - When our insert into the right path leaf is at the leftmost edge
2017 * and requires an update of the path immediately to it's left. This
2018 * can occur at the end of some types of rotation and appending inserts.
Tao Ma677b9752008-01-30 14:21:05 +08002019 * - When we've adjusted the last extent record in the left path leaf and the
2020 * 1st extent record in the right path leaf during cross extent block merge.
Mark Fashehdcd05382007-01-16 11:32:23 -08002021 */
Joel Becker4619c732009-02-12 19:02:36 -08002022static void ocfs2_complete_edge_insert(handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -08002023 struct ocfs2_path *left_path,
2024 struct ocfs2_path *right_path,
2025 int subtree_index)
2026{
Joel Beckerec20cec2010-03-19 14:13:52 -07002027 int i, idx;
Mark Fashehdcd05382007-01-16 11:32:23 -08002028 struct ocfs2_extent_list *el, *left_el, *right_el;
2029 struct ocfs2_extent_rec *left_rec, *right_rec;
2030 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2031
2032 /*
2033 * Update the counts and position values within all the
2034 * interior nodes to reflect the leaf rotation we just did.
2035 *
2036 * The root node is handled below the loop.
2037 *
2038 * We begin the loop with right_el and left_el pointing to the
2039 * leaf lists and work our way up.
2040 *
2041 * NOTE: within this loop, left_el and right_el always refer
2042 * to the *child* lists.
2043 */
2044 left_el = path_leaf_el(left_path);
2045 right_el = path_leaf_el(right_path);
2046 for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
Tao Maa09d09b2011-02-22 08:21:10 +08002047 trace_ocfs2_complete_edge_insert(i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002048
2049 /*
2050 * One nice property of knowing that all of these
2051 * nodes are below the root is that we only deal with
2052 * the leftmost right node record and the rightmost
2053 * left node record.
2054 */
2055 el = left_path->p_node[i].el;
2056 idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
2057 left_rec = &el->l_recs[idx];
2058
2059 el = right_path->p_node[i].el;
2060 right_rec = &el->l_recs[0];
2061
2062 ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
2063 right_el);
2064
Joel Beckerec20cec2010-03-19 14:13:52 -07002065 ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
2066 ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002067
2068 /*
2069 * Setup our list pointers now so that the current
2070 * parents become children in the next iteration.
2071 */
2072 left_el = left_path->p_node[i].el;
2073 right_el = right_path->p_node[i].el;
2074 }
2075
2076 /*
2077 * At the root node, adjust the two adjacent records which
2078 * begin our path to the leaves.
2079 */
2080
2081 el = left_path->p_node[subtree_index].el;
2082 left_el = left_path->p_node[subtree_index + 1].el;
2083 right_el = right_path->p_node[subtree_index + 1].el;
2084
2085 ocfs2_adjust_root_records(el, left_el, right_el,
2086 left_path->p_node[subtree_index + 1].bh->b_blocknr);
2087
2088 root_bh = left_path->p_node[subtree_index].bh;
2089
Joel Beckerec20cec2010-03-19 14:13:52 -07002090 ocfs2_journal_dirty(handle, root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002091}
2092
Joel Becker5c601ab2009-02-12 19:10:13 -08002093static int ocfs2_rotate_subtree_right(handle_t *handle,
2094 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08002095 struct ocfs2_path *left_path,
2096 struct ocfs2_path *right_path,
2097 int subtree_index)
2098{
2099 int ret, i;
2100 struct buffer_head *right_leaf_bh;
2101 struct buffer_head *left_leaf_bh = NULL;
2102 struct buffer_head *root_bh;
2103 struct ocfs2_extent_list *right_el, *left_el;
2104 struct ocfs2_extent_rec move_rec;
2105
2106 left_leaf_bh = path_leaf_bh(left_path);
2107 left_el = path_leaf_el(left_path);
2108
2109 if (left_el->l_next_free_rec != left_el->l_count) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002110 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07002111 "Inode %llu has non-full interior leaf node %llu (next free = %u)\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002112 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002113 (unsigned long long)left_leaf_bh->b_blocknr,
2114 le16_to_cpu(left_el->l_next_free_rec));
2115 return -EROFS;
2116 }
2117
2118 /*
2119 * This extent block may already have an empty record, so we
2120 * return early if so.
2121 */
2122 if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
2123 return 0;
2124
2125 root_bh = left_path->p_node[subtree_index].bh;
2126 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2127
Joel Becker5c601ab2009-02-12 19:10:13 -08002128 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002129 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002130 if (ret) {
2131 mlog_errno(ret);
2132 goto out;
2133 }
2134
2135 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002136 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002137 right_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002138 if (ret) {
2139 mlog_errno(ret);
2140 goto out;
2141 }
2142
Joel Becker5c601ab2009-02-12 19:10:13 -08002143 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002144 left_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002145 if (ret) {
2146 mlog_errno(ret);
2147 goto out;
2148 }
2149 }
2150
2151 right_leaf_bh = path_leaf_bh(right_path);
2152 right_el = path_leaf_el(right_path);
2153
2154 /* This is a code error, not a disk corruption. */
2155 mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
2156 "because rightmost leaf block %llu is empty\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002157 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002158 (unsigned long long)right_leaf_bh->b_blocknr);
2159
2160 ocfs2_create_empty_extent(right_el);
2161
Joel Beckerec20cec2010-03-19 14:13:52 -07002162 ocfs2_journal_dirty(handle, right_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002163
2164 /* Do the copy now. */
2165 i = le16_to_cpu(left_el->l_next_free_rec) - 1;
2166 move_rec = left_el->l_recs[i];
2167 right_el->l_recs[0] = move_rec;
2168
2169 /*
2170 * Clear out the record we just copied and shift everything
2171 * over, leaving an empty extent in the left leaf.
2172 *
2173 * We temporarily subtract from next_free_rec so that the
2174 * shift will lose the tail record (which is now defunct).
2175 */
2176 le16_add_cpu(&left_el->l_next_free_rec, -1);
2177 ocfs2_shift_records_right(left_el);
2178 memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2179 le16_add_cpu(&left_el->l_next_free_rec, 1);
2180
Joel Beckerec20cec2010-03-19 14:13:52 -07002181 ocfs2_journal_dirty(handle, left_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002182
Joel Becker4619c732009-02-12 19:02:36 -08002183 ocfs2_complete_edge_insert(handle, left_path, right_path,
2184 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002185
2186out:
2187 return ret;
2188}
2189
2190/*
2191 * Given a full path, determine what cpos value would return us a path
2192 * containing the leaf immediately to the left of the current one.
2193 *
2194 * Will return zero if the path passed in is already the leftmost path.
2195 */
Tristan Yeee149a72010-05-11 17:54:44 +08002196int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
2197 struct ocfs2_path *path, u32 *cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08002198{
2199 int i, j, ret = 0;
2200 u64 blkno;
2201 struct ocfs2_extent_list *el;
2202
Mark Fashehe48edee2007-03-07 16:46:57 -08002203 BUG_ON(path->p_tree_depth == 0);
2204
Mark Fashehdcd05382007-01-16 11:32:23 -08002205 *cpos = 0;
2206
2207 blkno = path_leaf_bh(path)->b_blocknr;
2208
2209 /* Start at the tree node just above the leaf and work our way up. */
2210 i = path->p_tree_depth - 1;
2211 while (i >= 0) {
2212 el = path->p_node[i].el;
2213
2214 /*
2215 * Find the extent record just before the one in our
2216 * path.
2217 */
2218 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2219 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2220 if (j == 0) {
2221 if (i == 0) {
2222 /*
2223 * We've determined that the
2224 * path specified is already
2225 * the leftmost one - return a
2226 * cpos of zero.
2227 */
2228 goto out;
2229 }
2230 /*
2231 * The leftmost record points to our
2232 * leaf - we need to travel up the
2233 * tree one level.
2234 */
2235 goto next_node;
2236 }
2237
2238 *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08002239 *cpos = *cpos + ocfs2_rec_clusters(el,
2240 &el->l_recs[j - 1]);
2241 *cpos = *cpos - 1;
Mark Fashehdcd05382007-01-16 11:32:23 -08002242 goto out;
2243 }
2244 }
2245
2246 /*
2247 * If we got here, we never found a valid node where
2248 * the tree indicated one should be.
2249 */
Joe Perches7ecef142015-09-04 15:44:51 -07002250 ocfs2_error(sb, "Invalid extent tree at extent block %llu\n",
Mark Fashehdcd05382007-01-16 11:32:23 -08002251 (unsigned long long)blkno);
2252 ret = -EROFS;
2253 goto out;
2254
2255next_node:
2256 blkno = path->p_node[i].bh->b_blocknr;
2257 i--;
2258 }
2259
2260out:
2261 return ret;
2262}
2263
Mark Fasheh328d5752007-06-18 10:48:04 -07002264/*
2265 * Extend the transaction by enough credits to complete the rotation,
2266 * and still leave at least the original number of credits allocated
2267 * to this transaction.
2268 */
Mark Fashehdcd05382007-01-16 11:32:23 -08002269static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
Mark Fasheh328d5752007-06-18 10:48:04 -07002270 int op_credits,
Mark Fashehdcd05382007-01-16 11:32:23 -08002271 struct ocfs2_path *path)
2272{
Tao Mac901fb02010-04-26 14:34:57 +08002273 int ret = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002274 int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002275
Tao Mac901fb02010-04-26 14:34:57 +08002276 if (handle->h_buffer_credits < credits)
Tao Mac18b8122009-08-18 11:44:07 +08002277 ret = ocfs2_extend_trans(handle,
2278 credits - handle->h_buffer_credits);
Tao Mac18b8122009-08-18 11:44:07 +08002279
Tao Mac901fb02010-04-26 14:34:57 +08002280 return ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08002281}
2282
2283/*
2284 * Trap the case where we're inserting into the theoretical range past
2285 * the _actual_ left leaf range. Otherwise, we'll rotate a record
2286 * whose cpos is less than ours into the right leaf.
2287 *
2288 * It's only necessary to look at the rightmost record of the left
2289 * leaf because the logic that calls us should ensure that the
2290 * theoretical ranges in the path components above the leaves are
2291 * correct.
2292 */
2293static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
2294 u32 insert_cpos)
2295{
2296 struct ocfs2_extent_list *left_el;
2297 struct ocfs2_extent_rec *rec;
2298 int next_free;
2299
2300 left_el = path_leaf_el(left_path);
2301 next_free = le16_to_cpu(left_el->l_next_free_rec);
2302 rec = &left_el->l_recs[next_free - 1];
2303
2304 if (insert_cpos > le32_to_cpu(rec->e_cpos))
2305 return 1;
2306 return 0;
2307}
2308
Mark Fasheh328d5752007-06-18 10:48:04 -07002309static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
2310{
2311 int next_free = le16_to_cpu(el->l_next_free_rec);
2312 unsigned int range;
2313 struct ocfs2_extent_rec *rec;
2314
2315 if (next_free == 0)
2316 return 0;
2317
2318 rec = &el->l_recs[0];
2319 if (ocfs2_is_empty_extent(rec)) {
2320 /* Empty list. */
2321 if (next_free == 1)
2322 return 0;
2323 rec = &el->l_recs[1];
2324 }
2325
2326 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2327 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
2328 return 1;
2329 return 0;
2330}
2331
Mark Fashehdcd05382007-01-16 11:32:23 -08002332/*
2333 * Rotate all the records in a btree right one record, starting at insert_cpos.
2334 *
2335 * The path to the rightmost leaf should be passed in.
2336 *
2337 * The array is assumed to be large enough to hold an entire path (tree depth).
2338 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002339 * Upon successful return from this function:
Mark Fashehdcd05382007-01-16 11:32:23 -08002340 *
2341 * - The 'right_path' array will contain a path to the leaf block
2342 * whose range contains e_cpos.
2343 * - That leaf block will have a single empty extent in list index 0.
2344 * - In the case that the rotation requires a post-insert update,
2345 * *ret_left_path will contain a valid path which can be passed to
2346 * ocfs2_insert_path().
2347 */
Joel Becker1bbf0b82009-02-12 19:42:08 -08002348static int ocfs2_rotate_tree_right(handle_t *handle,
Joel Becker5c601ab2009-02-12 19:10:13 -08002349 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002350 enum ocfs2_split_type split,
Mark Fashehdcd05382007-01-16 11:32:23 -08002351 u32 insert_cpos,
2352 struct ocfs2_path *right_path,
2353 struct ocfs2_path **ret_left_path)
2354{
Mark Fasheh328d5752007-06-18 10:48:04 -07002355 int ret, start, orig_credits = handle->h_buffer_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002356 u32 cpos;
2357 struct ocfs2_path *left_path = NULL;
Joel Becker5c601ab2009-02-12 19:10:13 -08002358 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehdcd05382007-01-16 11:32:23 -08002359
2360 *ret_left_path = NULL;
2361
Joel Beckerffdd7a52008-10-17 22:32:01 -07002362 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002363 if (!left_path) {
2364 ret = -ENOMEM;
2365 mlog_errno(ret);
2366 goto out;
2367 }
2368
Joel Becker5c601ab2009-02-12 19:10:13 -08002369 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002370 if (ret) {
2371 mlog_errno(ret);
2372 goto out;
2373 }
2374
Tao Maa09d09b2011-02-22 08:21:10 +08002375 trace_ocfs2_rotate_tree_right(
2376 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2377 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002378
2379 /*
2380 * What we want to do here is:
2381 *
2382 * 1) Start with the rightmost path.
2383 *
2384 * 2) Determine a path to the leaf block directly to the left
2385 * of that leaf.
2386 *
2387 * 3) Determine the 'subtree root' - the lowest level tree node
2388 * which contains a path to both leaves.
2389 *
2390 * 4) Rotate the subtree.
2391 *
2392 * 5) Find the next subtree by considering the left path to be
2393 * the new right path.
2394 *
2395 * The check at the top of this while loop also accepts
2396 * insert_cpos == cpos because cpos is only a _theoretical_
2397 * value to get us the left path - insert_cpos might very well
2398 * be filling that hole.
2399 *
2400 * Stop at a cpos of '0' because we either started at the
2401 * leftmost branch (i.e., a tree with one branch and a
2402 * rotation inside of it), or we've gone as far as we can in
2403 * rotating subtrees.
2404 */
2405 while (cpos && insert_cpos <= cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08002406 trace_ocfs2_rotate_tree_right(
2407 (unsigned long long)
2408 ocfs2_metadata_cache_owner(et->et_ci),
2409 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002410
Joel Becker5c601ab2009-02-12 19:10:13 -08002411 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002412 if (ret) {
2413 mlog_errno(ret);
2414 goto out;
2415 }
2416
2417 mlog_bug_on_msg(path_leaf_bh(left_path) ==
2418 path_leaf_bh(right_path),
Joel Becker5c601ab2009-02-12 19:10:13 -08002419 "Owner %llu: error during insert of %u "
Mark Fashehdcd05382007-01-16 11:32:23 -08002420 "(left path cpos %u) results in two identical "
2421 "paths ending at %llu\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002422 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2423 insert_cpos, cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08002424 (unsigned long long)
2425 path_leaf_bh(left_path)->b_blocknr);
2426
Mark Fasheh328d5752007-06-18 10:48:04 -07002427 if (split == SPLIT_NONE &&
2428 ocfs2_rotate_requires_path_adjustment(left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002429 insert_cpos)) {
Mark Fashehdcd05382007-01-16 11:32:23 -08002430
2431 /*
2432 * We've rotated the tree as much as we
2433 * should. The rest is up to
2434 * ocfs2_insert_path() to complete, after the
2435 * record insertion. We indicate this
2436 * situation by returning the left path.
2437 *
2438 * The reason we don't adjust the records here
2439 * before the record insert is that an error
2440 * later might break the rule where a parent
2441 * record e_cpos will reflect the actual
2442 * e_cpos of the 1st nonempty record of the
2443 * child list.
2444 */
2445 *ret_left_path = left_path;
2446 goto out_ret_path;
2447 }
2448
Joel Becker7dc02802009-02-12 19:20:13 -08002449 start = ocfs2_find_subtree_root(et, left_path, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002450
Tao Maa09d09b2011-02-22 08:21:10 +08002451 trace_ocfs2_rotate_subtree(start,
2452 (unsigned long long)
2453 right_path->p_node[start].bh->b_blocknr,
2454 right_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08002455
2456 ret = ocfs2_extend_rotate_transaction(handle, start,
Mark Fasheh328d5752007-06-18 10:48:04 -07002457 orig_credits, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002458 if (ret) {
2459 mlog_errno(ret);
2460 goto out;
2461 }
2462
Joel Becker5c601ab2009-02-12 19:10:13 -08002463 ret = ocfs2_rotate_subtree_right(handle, et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002464 right_path, start);
2465 if (ret) {
2466 mlog_errno(ret);
2467 goto out;
2468 }
2469
Mark Fasheh328d5752007-06-18 10:48:04 -07002470 if (split != SPLIT_NONE &&
2471 ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
2472 insert_cpos)) {
2473 /*
2474 * A rotate moves the rightmost left leaf
2475 * record over to the leftmost right leaf
2476 * slot. If we're doing an extent split
2477 * instead of a real insert, then we have to
2478 * check that the extent to be split wasn't
2479 * just moved over. If it was, then we can
2480 * exit here, passing left_path back -
2481 * ocfs2_split_extent() is smart enough to
2482 * search both leaves.
2483 */
2484 *ret_left_path = left_path;
2485 goto out_ret_path;
2486 }
2487
Mark Fashehdcd05382007-01-16 11:32:23 -08002488 /*
2489 * There is no need to re-read the next right path
2490 * as we know that it'll be our current left
2491 * path. Optimize by copying values instead.
2492 */
2493 ocfs2_mv_path(right_path, left_path);
2494
Joel Becker5c601ab2009-02-12 19:10:13 -08002495 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002496 if (ret) {
2497 mlog_errno(ret);
2498 goto out;
2499 }
2500 }
2501
2502out:
2503 ocfs2_free_path(left_path);
2504
2505out_ret_path:
2506 return ret;
2507}
2508
Joel Becker09106ba2009-02-12 19:43:57 -08002509static int ocfs2_update_edge_lengths(handle_t *handle,
2510 struct ocfs2_extent_tree *et,
Tao Ma3c5e1062009-07-21 15:42:05 +08002511 int subtree_index, struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002512{
Tao Ma3c5e1062009-07-21 15:42:05 +08002513 int i, idx, ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002514 struct ocfs2_extent_rec *rec;
2515 struct ocfs2_extent_list *el;
2516 struct ocfs2_extent_block *eb;
2517 u32 range;
2518
Tao Ma3c5e1062009-07-21 15:42:05 +08002519 /*
2520 * In normal tree rotation process, we will never touch the
2521 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
2522 * doesn't reserve the credits for them either.
2523 *
2524 * But we do have a special case here which will update the rightmost
2525 * records for all the bh in the path.
2526 * So we have to allocate extra credits and access them.
2527 */
Tao Mac901fb02010-04-26 14:34:57 +08002528 ret = ocfs2_extend_trans(handle, subtree_index);
Tao Ma3c5e1062009-07-21 15:42:05 +08002529 if (ret) {
2530 mlog_errno(ret);
2531 goto out;
2532 }
2533
Joel Becker09106ba2009-02-12 19:43:57 -08002534 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002535 if (ret) {
2536 mlog_errno(ret);
2537 goto out;
2538 }
2539
Mark Fasheh328d5752007-06-18 10:48:04 -07002540 /* Path should always be rightmost. */
2541 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2542 BUG_ON(eb->h_next_leaf_blk != 0ULL);
2543
2544 el = &eb->h_list;
2545 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2546 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2547 rec = &el->l_recs[idx];
2548 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2549
2550 for (i = 0; i < path->p_tree_depth; i++) {
2551 el = path->p_node[i].el;
2552 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2553 rec = &el->l_recs[idx];
2554
2555 rec->e_int_clusters = cpu_to_le32(range);
2556 le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2557
2558 ocfs2_journal_dirty(handle, path->p_node[i].bh);
2559 }
Tao Ma3c5e1062009-07-21 15:42:05 +08002560out:
2561 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002562}
2563
Joel Becker6641b0c2009-02-12 18:57:52 -08002564static void ocfs2_unlink_path(handle_t *handle,
2565 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002566 struct ocfs2_cached_dealloc_ctxt *dealloc,
2567 struct ocfs2_path *path, int unlink_start)
2568{
2569 int ret, i;
2570 struct ocfs2_extent_block *eb;
2571 struct ocfs2_extent_list *el;
2572 struct buffer_head *bh;
2573
2574 for(i = unlink_start; i < path_num_items(path); i++) {
2575 bh = path->p_node[i].bh;
2576
2577 eb = (struct ocfs2_extent_block *)bh->b_data;
2578 /*
2579 * Not all nodes might have had their final count
2580 * decremented by the caller - handle this here.
2581 */
2582 el = &eb->h_list;
2583 if (le16_to_cpu(el->l_next_free_rec) > 1) {
2584 mlog(ML_ERROR,
2585 "Inode %llu, attempted to remove extent block "
2586 "%llu with %u records\n",
Joel Becker6641b0c2009-02-12 18:57:52 -08002587 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07002588 (unsigned long long)le64_to_cpu(eb->h_blkno),
2589 le16_to_cpu(el->l_next_free_rec));
2590
2591 ocfs2_journal_dirty(handle, bh);
Joel Becker6641b0c2009-02-12 18:57:52 -08002592 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002593 continue;
2594 }
2595
2596 el->l_next_free_rec = 0;
2597 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2598
2599 ocfs2_journal_dirty(handle, bh);
2600
2601 ret = ocfs2_cache_extent_block_free(dealloc, eb);
2602 if (ret)
2603 mlog_errno(ret);
2604
Joel Becker6641b0c2009-02-12 18:57:52 -08002605 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002606 }
2607}
2608
Joel Becker6641b0c2009-02-12 18:57:52 -08002609static void ocfs2_unlink_subtree(handle_t *handle,
2610 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002611 struct ocfs2_path *left_path,
2612 struct ocfs2_path *right_path,
2613 int subtree_index,
2614 struct ocfs2_cached_dealloc_ctxt *dealloc)
2615{
2616 int i;
2617 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2618 struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
2619 struct ocfs2_extent_list *el;
2620 struct ocfs2_extent_block *eb;
2621
2622 el = path_leaf_el(left_path);
2623
2624 eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2625
2626 for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2627 if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2628 break;
2629
2630 BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2631
2632 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2633 le16_add_cpu(&root_el->l_next_free_rec, -1);
2634
2635 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2636 eb->h_next_leaf_blk = 0;
2637
2638 ocfs2_journal_dirty(handle, root_bh);
2639 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2640
Joel Becker6641b0c2009-02-12 18:57:52 -08002641 ocfs2_unlink_path(handle, et, dealloc, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002642 subtree_index + 1);
2643}
2644
Joel Becker1e2dd632009-02-12 19:45:28 -08002645static int ocfs2_rotate_subtree_left(handle_t *handle,
2646 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002647 struct ocfs2_path *left_path,
2648 struct ocfs2_path *right_path,
2649 int subtree_index,
2650 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker1e2dd632009-02-12 19:45:28 -08002651 int *deleted)
Mark Fasheh328d5752007-06-18 10:48:04 -07002652{
2653 int ret, i, del_right_subtree = 0, right_has_empty = 0;
Tao Mae7d4cb62008-08-18 17:38:44 +08002654 struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002655 struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2656 struct ocfs2_extent_block *eb;
2657
2658 *deleted = 0;
2659
2660 right_leaf_el = path_leaf_el(right_path);
2661 left_leaf_el = path_leaf_el(left_path);
2662 root_bh = left_path->p_node[subtree_index].bh;
2663 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2664
2665 if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2666 return 0;
2667
2668 eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2669 if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2670 /*
2671 * It's legal for us to proceed if the right leaf is
2672 * the rightmost one and it has an empty extent. There
2673 * are two cases to handle - whether the leaf will be
2674 * empty after removal or not. If the leaf isn't empty
2675 * then just remove the empty extent up front. The
2676 * next block will handle empty leaves by flagging
2677 * them for unlink.
2678 *
2679 * Non rightmost leaves will throw -EAGAIN and the
2680 * caller can manually move the subtree and retry.
2681 */
2682
2683 if (eb->h_next_leaf_blk != 0ULL)
2684 return -EAGAIN;
2685
2686 if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002687 ret = ocfs2_journal_access_eb(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002688 path_leaf_bh(right_path),
2689 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002690 if (ret) {
2691 mlog_errno(ret);
2692 goto out;
2693 }
2694
2695 ocfs2_remove_empty_extent(right_leaf_el);
2696 } else
2697 right_has_empty = 1;
2698 }
2699
2700 if (eb->h_next_leaf_blk == 0ULL &&
2701 le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2702 /*
2703 * We have to update i_last_eb_blk during the meta
2704 * data delete.
2705 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08002706 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07002707 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002708 if (ret) {
2709 mlog_errno(ret);
2710 goto out;
2711 }
2712
2713 del_right_subtree = 1;
2714 }
2715
2716 /*
2717 * Getting here with an empty extent in the right path implies
2718 * that it's the rightmost path and will be deleted.
2719 */
2720 BUG_ON(right_has_empty && !del_right_subtree);
2721
Joel Becker1e2dd632009-02-12 19:45:28 -08002722 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002723 subtree_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07002724 if (ret) {
2725 mlog_errno(ret);
2726 goto out;
2727 }
2728
2729 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002730 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002731 right_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002732 if (ret) {
2733 mlog_errno(ret);
2734 goto out;
2735 }
2736
Joel Becker1e2dd632009-02-12 19:45:28 -08002737 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002738 left_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002739 if (ret) {
2740 mlog_errno(ret);
2741 goto out;
2742 }
2743 }
2744
2745 if (!right_has_empty) {
2746 /*
2747 * Only do this if we're moving a real
2748 * record. Otherwise, the action is delayed until
2749 * after removal of the right path in which case we
2750 * can do a simple shift to remove the empty extent.
2751 */
2752 ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2753 memset(&right_leaf_el->l_recs[0], 0,
2754 sizeof(struct ocfs2_extent_rec));
2755 }
2756 if (eb->h_next_leaf_blk == 0ULL) {
2757 /*
2758 * Move recs over to get rid of empty extent, decrease
2759 * next_free. This is allowed to remove the last
2760 * extent in our leaf (setting l_next_free_rec to
2761 * zero) - the delete code below won't care.
2762 */
2763 ocfs2_remove_empty_extent(right_leaf_el);
2764 }
2765
Joel Beckerec20cec2010-03-19 14:13:52 -07002766 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2767 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07002768
2769 if (del_right_subtree) {
Joel Becker6641b0c2009-02-12 18:57:52 -08002770 ocfs2_unlink_subtree(handle, et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002771 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08002772 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08002773 left_path);
2774 if (ret) {
2775 mlog_errno(ret);
2776 goto out;
2777 }
Mark Fasheh328d5752007-06-18 10:48:04 -07002778
2779 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07002780 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07002781
2782 /*
2783 * Removal of the extent in the left leaf was skipped
2784 * above so we could delete the right path
2785 * 1st.
2786 */
2787 if (right_has_empty)
2788 ocfs2_remove_empty_extent(left_leaf_el);
2789
Joel Beckerec20cec2010-03-19 14:13:52 -07002790 ocfs2_journal_dirty(handle, et_root_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002791
2792 *deleted = 1;
2793 } else
Joel Becker4619c732009-02-12 19:02:36 -08002794 ocfs2_complete_edge_insert(handle, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002795 subtree_index);
2796
2797out:
2798 return ret;
2799}
2800
2801/*
2802 * Given a full path, determine what cpos value would return us a path
2803 * containing the leaf immediately to the right of the current one.
2804 *
2805 * Will return zero if the path passed in is already the rightmost path.
2806 *
2807 * This looks similar, but is subtly different to
2808 * ocfs2_find_cpos_for_left_leaf().
2809 */
Tao Ma38a04e42009-11-30 14:32:19 +08002810int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2811 struct ocfs2_path *path, u32 *cpos)
Mark Fasheh328d5752007-06-18 10:48:04 -07002812{
2813 int i, j, ret = 0;
2814 u64 blkno;
2815 struct ocfs2_extent_list *el;
2816
2817 *cpos = 0;
2818
2819 if (path->p_tree_depth == 0)
2820 return 0;
2821
2822 blkno = path_leaf_bh(path)->b_blocknr;
2823
2824 /* Start at the tree node just above the leaf and work our way up. */
2825 i = path->p_tree_depth - 1;
2826 while (i >= 0) {
2827 int next_free;
2828
2829 el = path->p_node[i].el;
2830
2831 /*
2832 * Find the extent record just after the one in our
2833 * path.
2834 */
2835 next_free = le16_to_cpu(el->l_next_free_rec);
2836 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2837 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2838 if (j == (next_free - 1)) {
2839 if (i == 0) {
2840 /*
2841 * We've determined that the
2842 * path specified is already
2843 * the rightmost one - return a
2844 * cpos of zero.
2845 */
2846 goto out;
2847 }
2848 /*
2849 * The rightmost record points to our
2850 * leaf - we need to travel up the
2851 * tree one level.
2852 */
2853 goto next_node;
2854 }
2855
2856 *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2857 goto out;
2858 }
2859 }
2860
2861 /*
2862 * If we got here, we never found a valid node where
2863 * the tree indicated one should be.
2864 */
Joe Perches7ecef142015-09-04 15:44:51 -07002865 ocfs2_error(sb, "Invalid extent tree at extent block %llu\n",
Mark Fasheh328d5752007-06-18 10:48:04 -07002866 (unsigned long long)blkno);
2867 ret = -EROFS;
2868 goto out;
2869
2870next_node:
2871 blkno = path->p_node[i].bh->b_blocknr;
2872 i--;
2873 }
2874
2875out:
2876 return ret;
2877}
2878
Joel Becker70f18c02009-02-13 02:09:31 -08002879static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2880 struct ocfs2_extent_tree *et,
Joel Becker13723d02008-10-17 19:25:01 -07002881 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002882{
2883 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07002884 struct buffer_head *bh = path_leaf_bh(path);
2885 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002886
2887 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2888 return 0;
2889
Joel Becker70f18c02009-02-13 02:09:31 -08002890 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Joel Becker13723d02008-10-17 19:25:01 -07002891 path_num_items(path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07002892 if (ret) {
2893 mlog_errno(ret);
2894 goto out;
2895 }
2896
2897 ocfs2_remove_empty_extent(el);
Joel Beckerec20cec2010-03-19 14:13:52 -07002898 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002899
2900out:
2901 return ret;
2902}
2903
Joel Beckere46f74d2009-02-12 19:47:43 -08002904static int __ocfs2_rotate_tree_left(handle_t *handle,
2905 struct ocfs2_extent_tree *et,
2906 int orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07002907 struct ocfs2_path *path,
2908 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08002909 struct ocfs2_path **empty_extent_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002910{
2911 int ret, subtree_root, deleted;
2912 u32 right_cpos;
2913 struct ocfs2_path *left_path = NULL;
2914 struct ocfs2_path *right_path = NULL;
Joel Beckere46f74d2009-02-12 19:47:43 -08002915 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fasheh328d5752007-06-18 10:48:04 -07002916
Xue jiufei099768b2015-06-24 16:55:09 -07002917 if (!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])))
2918 return 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002919
2920 *empty_extent_path = NULL;
2921
Joel Beckere46f74d2009-02-12 19:47:43 -08002922 ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002923 if (ret) {
2924 mlog_errno(ret);
2925 goto out;
2926 }
2927
Joel Beckerffdd7a52008-10-17 22:32:01 -07002928 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002929 if (!left_path) {
2930 ret = -ENOMEM;
2931 mlog_errno(ret);
2932 goto out;
2933 }
2934
2935 ocfs2_cp_path(left_path, path);
2936
Joel Beckerffdd7a52008-10-17 22:32:01 -07002937 right_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002938 if (!right_path) {
2939 ret = -ENOMEM;
2940 mlog_errno(ret);
2941 goto out;
2942 }
2943
2944 while (right_cpos) {
Joel Beckerfacdb772009-02-12 18:08:48 -08002945 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002946 if (ret) {
2947 mlog_errno(ret);
2948 goto out;
2949 }
2950
Joel Becker7dc02802009-02-12 19:20:13 -08002951 subtree_root = ocfs2_find_subtree_root(et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002952 right_path);
2953
Tao Maa09d09b2011-02-22 08:21:10 +08002954 trace_ocfs2_rotate_subtree(subtree_root,
Mark Fasheh328d5752007-06-18 10:48:04 -07002955 (unsigned long long)
2956 right_path->p_node[subtree_root].bh->b_blocknr,
2957 right_path->p_tree_depth);
2958
2959 ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
2960 orig_credits, left_path);
2961 if (ret) {
2962 mlog_errno(ret);
2963 goto out;
2964 }
2965
Mark Fashehe8aed342007-12-03 16:43:01 -08002966 /*
2967 * Caller might still want to make changes to the
2968 * tree root, so re-add it to the journal here.
2969 */
Joel Beckere46f74d2009-02-12 19:47:43 -08002970 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002971 left_path, 0);
Mark Fashehe8aed342007-12-03 16:43:01 -08002972 if (ret) {
2973 mlog_errno(ret);
2974 goto out;
2975 }
2976
Joel Becker1e2dd632009-02-12 19:45:28 -08002977 ret = ocfs2_rotate_subtree_left(handle, et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002978 right_path, subtree_root,
Joel Becker1e2dd632009-02-12 19:45:28 -08002979 dealloc, &deleted);
Mark Fasheh328d5752007-06-18 10:48:04 -07002980 if (ret == -EAGAIN) {
2981 /*
2982 * The rotation has to temporarily stop due to
2983 * the right subtree having an empty
2984 * extent. Pass it back to the caller for a
2985 * fixup.
2986 */
2987 *empty_extent_path = right_path;
2988 right_path = NULL;
2989 goto out;
2990 }
2991 if (ret) {
2992 mlog_errno(ret);
2993 goto out;
2994 }
2995
2996 /*
2997 * The subtree rotate might have removed records on
2998 * the rightmost edge. If so, then rotation is
2999 * complete.
3000 */
3001 if (deleted)
3002 break;
3003
3004 ocfs2_mv_path(left_path, right_path);
3005
Joel Beckere46f74d2009-02-12 19:47:43 -08003006 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003007 &right_cpos);
3008 if (ret) {
3009 mlog_errno(ret);
3010 goto out;
3011 }
3012 }
3013
3014out:
3015 ocfs2_free_path(right_path);
3016 ocfs2_free_path(left_path);
3017
3018 return ret;
3019}
3020
Joel Becker70f18c02009-02-13 02:09:31 -08003021static int ocfs2_remove_rightmost_path(handle_t *handle,
3022 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003023 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003024 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003025{
3026 int ret, subtree_index;
3027 u32 cpos;
3028 struct ocfs2_path *left_path = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003029 struct ocfs2_extent_block *eb;
3030 struct ocfs2_extent_list *el;
3031
Mark Fasheh328d5752007-06-18 10:48:04 -07003032
Joel Becker6136ca52009-02-12 19:32:43 -08003033 ret = ocfs2_et_sanity_check(et);
Tao Mae7d4cb62008-08-18 17:38:44 +08003034 if (ret)
3035 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003036 /*
3037 * There's two ways we handle this depending on
3038 * whether path is the only existing one.
3039 */
3040 ret = ocfs2_extend_rotate_transaction(handle, 0,
3041 handle->h_buffer_credits,
3042 path);
3043 if (ret) {
3044 mlog_errno(ret);
3045 goto out;
3046 }
3047
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003048 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003049 if (ret) {
3050 mlog_errno(ret);
3051 goto out;
3052 }
3053
Joel Becker3d03a302009-02-12 17:49:26 -08003054 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3055 path, &cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003056 if (ret) {
3057 mlog_errno(ret);
3058 goto out;
3059 }
3060
3061 if (cpos) {
3062 /*
3063 * We have a path to the left of this one - it needs
3064 * an update too.
3065 */
Joel Beckerffdd7a52008-10-17 22:32:01 -07003066 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003067 if (!left_path) {
3068 ret = -ENOMEM;
3069 mlog_errno(ret);
3070 goto out;
3071 }
3072
Joel Beckerfacdb772009-02-12 18:08:48 -08003073 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003074 if (ret) {
3075 mlog_errno(ret);
3076 goto out;
3077 }
3078
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003079 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003080 if (ret) {
3081 mlog_errno(ret);
3082 goto out;
3083 }
3084
Joel Becker7dc02802009-02-12 19:20:13 -08003085 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003086
Joel Becker6641b0c2009-02-12 18:57:52 -08003087 ocfs2_unlink_subtree(handle, et, left_path, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003088 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08003089 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08003090 left_path);
3091 if (ret) {
3092 mlog_errno(ret);
3093 goto out;
3094 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003095
3096 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07003097 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003098 } else {
3099 /*
3100 * 'path' is also the leftmost path which
3101 * means it must be the only one. This gets
3102 * handled differently because we want to
Joel Becker70f18c02009-02-13 02:09:31 -08003103 * revert the root back to having extents
Mark Fasheh328d5752007-06-18 10:48:04 -07003104 * in-line.
3105 */
Joel Becker6641b0c2009-02-12 18:57:52 -08003106 ocfs2_unlink_path(handle, et, dealloc, path, 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003107
Joel Beckerce1d9ea2008-08-20 16:30:07 -07003108 el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003109 el->l_tree_depth = 0;
3110 el->l_next_free_rec = 0;
3111 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3112
Joel Becker35dc0aa2008-08-20 16:25:06 -07003113 ocfs2_et_set_last_eb_blk(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003114 }
3115
3116 ocfs2_journal_dirty(handle, path_root_bh(path));
3117
3118out:
3119 ocfs2_free_path(left_path);
3120 return ret;
3121}
3122
Xue jiufeid0c97d52015-09-04 15:44:48 -07003123static int ocfs2_remove_rightmost_empty_extent(struct ocfs2_super *osb,
3124 struct ocfs2_extent_tree *et,
3125 struct ocfs2_path *path,
3126 struct ocfs2_cached_dealloc_ctxt *dealloc)
3127{
3128 handle_t *handle;
3129 int ret;
3130 int credits = path->p_tree_depth * 2 + 1;
3131
3132 handle = ocfs2_start_trans(osb, credits);
3133 if (IS_ERR(handle)) {
3134 ret = PTR_ERR(handle);
3135 mlog_errno(ret);
3136 return ret;
3137 }
3138
3139 ret = ocfs2_remove_rightmost_path(handle, et, path, dealloc);
3140 if (ret)
3141 mlog_errno(ret);
3142
3143 ocfs2_commit_trans(osb, handle);
3144 return ret;
3145}
3146
Mark Fasheh328d5752007-06-18 10:48:04 -07003147/*
3148 * Left rotation of btree records.
3149 *
3150 * In many ways, this is (unsurprisingly) the opposite of right
3151 * rotation. We start at some non-rightmost path containing an empty
3152 * extent in the leaf block. The code works its way to the rightmost
3153 * path by rotating records to the left in every subtree.
3154 *
3155 * This is used by any code which reduces the number of extent records
3156 * in a leaf. After removal, an empty record should be placed in the
3157 * leftmost list position.
3158 *
3159 * This won't handle a length update of the rightmost path records if
3160 * the rightmost tree leaf record is removed so the caller is
3161 * responsible for detecting and correcting that.
3162 */
Joel Becker70f18c02009-02-13 02:09:31 -08003163static int ocfs2_rotate_tree_left(handle_t *handle,
3164 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003165 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003166 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003167{
3168 int ret, orig_credits = handle->h_buffer_credits;
3169 struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3170 struct ocfs2_extent_block *eb;
3171 struct ocfs2_extent_list *el;
3172
3173 el = path_leaf_el(path);
3174 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3175 return 0;
3176
3177 if (path->p_tree_depth == 0) {
3178rightmost_no_delete:
3179 /*
Tao Mae7d4cb62008-08-18 17:38:44 +08003180 * Inline extents. This is trivially handled, so do
Mark Fasheh328d5752007-06-18 10:48:04 -07003181 * it up front.
3182 */
Joel Becker70f18c02009-02-13 02:09:31 -08003183 ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003184 if (ret)
3185 mlog_errno(ret);
3186 goto out;
3187 }
3188
3189 /*
3190 * Handle rightmost branch now. There's several cases:
3191 * 1) simple rotation leaving records in there. That's trivial.
3192 * 2) rotation requiring a branch delete - there's no more
3193 * records left. Two cases of this:
3194 * a) There are branches to the left.
3195 * b) This is also the leftmost (the only) branch.
3196 *
3197 * 1) is handled via ocfs2_rotate_rightmost_leaf_left()
3198 * 2a) we need the left branch so that we can update it with the unlink
Joel Becker70f18c02009-02-13 02:09:31 -08003199 * 2b) we need to bring the root back to inline extents.
Mark Fasheh328d5752007-06-18 10:48:04 -07003200 */
3201
3202 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3203 el = &eb->h_list;
3204 if (eb->h_next_leaf_blk == 0) {
3205 /*
3206 * This gets a bit tricky if we're going to delete the
3207 * rightmost path. Get the other cases out of the way
3208 * 1st.
3209 */
3210 if (le16_to_cpu(el->l_next_free_rec) > 1)
3211 goto rightmost_no_delete;
3212
3213 if (le16_to_cpu(el->l_next_free_rec) == 0) {
3214 ret = -EIO;
Joel Becker70f18c02009-02-13 02:09:31 -08003215 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07003216 "Owner %llu has empty extent block at %llu\n",
Joel Becker70f18c02009-02-13 02:09:31 -08003217 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07003218 (unsigned long long)le64_to_cpu(eb->h_blkno));
3219 goto out;
3220 }
3221
3222 /*
3223 * XXX: The caller can not trust "path" any more after
3224 * this as it will have been deleted. What do we do?
3225 *
3226 * In theory the rotate-for-merge code will never get
3227 * here because it'll always ask for a rotate in a
3228 * nonempty list.
3229 */
3230
Joel Becker70f18c02009-02-13 02:09:31 -08003231 ret = ocfs2_remove_rightmost_path(handle, et, path,
3232 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003233 if (ret)
3234 mlog_errno(ret);
3235 goto out;
3236 }
3237
3238 /*
3239 * Now we can loop, remembering the path we get from -EAGAIN
3240 * and restarting from there.
3241 */
3242try_rotate:
Joel Beckere46f74d2009-02-12 19:47:43 -08003243 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3244 dealloc, &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003245 if (ret && ret != -EAGAIN) {
3246 mlog_errno(ret);
3247 goto out;
3248 }
3249
3250 while (ret == -EAGAIN) {
3251 tmp_path = restart_path;
3252 restart_path = NULL;
3253
Joel Beckere46f74d2009-02-12 19:47:43 -08003254 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07003255 tmp_path, dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08003256 &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003257 if (ret && ret != -EAGAIN) {
3258 mlog_errno(ret);
3259 goto out;
3260 }
3261
3262 ocfs2_free_path(tmp_path);
3263 tmp_path = NULL;
3264
3265 if (ret == 0)
3266 goto try_rotate;
3267 }
3268
3269out:
3270 ocfs2_free_path(tmp_path);
3271 ocfs2_free_path(restart_path);
3272 return ret;
3273}
3274
3275static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3276 int index)
3277{
3278 struct ocfs2_extent_rec *rec = &el->l_recs[index];
3279 unsigned int size;
3280
3281 if (rec->e_leaf_clusters == 0) {
3282 /*
3283 * We consumed all of the merged-from record. An empty
3284 * extent cannot exist anywhere but the 1st array
3285 * position, so move things over if the merged-from
3286 * record doesn't occupy that position.
3287 *
3288 * This creates a new empty extent so the caller
3289 * should be smart enough to have removed any existing
3290 * ones.
3291 */
3292 if (index > 0) {
3293 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3294 size = index * sizeof(struct ocfs2_extent_rec);
3295 memmove(&el->l_recs[1], &el->l_recs[0], size);
3296 }
3297
3298 /*
3299 * Always memset - the caller doesn't check whether it
3300 * created an empty extent, so there could be junk in
3301 * the other fields.
3302 */
3303 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3304 }
3305}
3306
Joel Becker4fe82c32009-02-13 02:16:08 -08003307static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003308 struct ocfs2_path *left_path,
3309 struct ocfs2_path **ret_right_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07003310{
3311 int ret;
Tao Ma677b9752008-01-30 14:21:05 +08003312 u32 right_cpos;
3313 struct ocfs2_path *right_path = NULL;
3314 struct ocfs2_extent_list *left_el;
3315
3316 *ret_right_path = NULL;
3317
3318 /* This function shouldn't be called for non-trees. */
3319 BUG_ON(left_path->p_tree_depth == 0);
3320
3321 left_el = path_leaf_el(left_path);
3322 BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3323
Joel Becker4fe82c32009-02-13 02:16:08 -08003324 ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3325 left_path, &right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003326 if (ret) {
3327 mlog_errno(ret);
3328 goto out;
3329 }
3330
3331 /* This function shouldn't be called for the rightmost leaf. */
3332 BUG_ON(right_cpos == 0);
3333
Joel Beckerffdd7a52008-10-17 22:32:01 -07003334 right_path = ocfs2_new_path_from_path(left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003335 if (!right_path) {
3336 ret = -ENOMEM;
3337 mlog_errno(ret);
3338 goto out;
3339 }
3340
Joel Becker4fe82c32009-02-13 02:16:08 -08003341 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003342 if (ret) {
3343 mlog_errno(ret);
3344 goto out;
3345 }
3346
3347 *ret_right_path = right_path;
3348out:
3349 if (ret)
3350 ocfs2_free_path(right_path);
3351 return ret;
3352}
3353
3354/*
3355 * Remove split_rec clusters from the record at index and merge them
3356 * onto the beginning of the record "next" to it.
3357 * For index < l_count - 1, the next means the extent rec at index + 1.
3358 * For index == l_count - 1, the "next" means the 1st extent rec of the
3359 * next extent block.
3360 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003361static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003362 handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003363 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003364 struct ocfs2_extent_rec *split_rec,
3365 int index)
3366{
3367 int ret, next_free, i;
Mark Fasheh328d5752007-06-18 10:48:04 -07003368 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3369 struct ocfs2_extent_rec *left_rec;
3370 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003371 struct ocfs2_extent_list *right_el;
3372 struct ocfs2_path *right_path = NULL;
3373 int subtree_index = 0;
3374 struct ocfs2_extent_list *el = path_leaf_el(left_path);
3375 struct buffer_head *bh = path_leaf_bh(left_path);
3376 struct buffer_head *root_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003377
3378 BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
Mark Fasheh328d5752007-06-18 10:48:04 -07003379 left_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003380
Al Viro9d8df6a2008-05-21 06:32:11 +01003381 if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
Tao Ma677b9752008-01-30 14:21:05 +08003382 le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3383 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003384 ret = ocfs2_get_right_path(et, left_path, &right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003385 if (ret) {
3386 mlog_errno(ret);
Markus Elfring629a3b52015-04-14 15:42:53 -07003387 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003388 }
3389
3390 right_el = path_leaf_el(right_path);
3391 next_free = le16_to_cpu(right_el->l_next_free_rec);
3392 BUG_ON(next_free <= 0);
3393 right_rec = &right_el->l_recs[0];
3394 if (ocfs2_is_empty_extent(right_rec)) {
Al Viro9d8df6a2008-05-21 06:32:11 +01003395 BUG_ON(next_free <= 1);
Tao Ma677b9752008-01-30 14:21:05 +08003396 right_rec = &right_el->l_recs[1];
3397 }
3398
3399 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3400 le16_to_cpu(left_rec->e_leaf_clusters) !=
3401 le32_to_cpu(right_rec->e_cpos));
3402
Joel Becker7dc02802009-02-12 19:20:13 -08003403 subtree_index = ocfs2_find_subtree_root(et, left_path,
3404 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003405
3406 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3407 handle->h_buffer_credits,
3408 right_path);
3409 if (ret) {
3410 mlog_errno(ret);
3411 goto out;
3412 }
3413
3414 root_bh = left_path->p_node[subtree_index].bh;
3415 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3416
Joel Becker7dc02802009-02-12 19:20:13 -08003417 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003418 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003419 if (ret) {
3420 mlog_errno(ret);
3421 goto out;
3422 }
3423
3424 for (i = subtree_index + 1;
3425 i < path_num_items(right_path); i++) {
Joel Becker7dc02802009-02-12 19:20:13 -08003426 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003427 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003428 if (ret) {
3429 mlog_errno(ret);
3430 goto out;
3431 }
3432
Joel Becker7dc02802009-02-12 19:20:13 -08003433 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003434 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003435 if (ret) {
3436 mlog_errno(ret);
3437 goto out;
3438 }
3439 }
3440
3441 } else {
3442 BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3443 right_rec = &el->l_recs[index + 1];
3444 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003445
Joel Becker7dc02802009-02-12 19:20:13 -08003446 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
Joel Becker13723d02008-10-17 19:25:01 -07003447 path_num_items(left_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003448 if (ret) {
3449 mlog_errno(ret);
3450 goto out;
3451 }
3452
3453 le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3454
3455 le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3456 le64_add_cpu(&right_rec->e_blkno,
Joel Becker7dc02802009-02-12 19:20:13 -08003457 -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3458 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003459 le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3460
3461 ocfs2_cleanup_merge(el, index);
3462
Joel Beckerec20cec2010-03-19 14:13:52 -07003463 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003464 if (right_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003465 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Joel Becker4619c732009-02-12 19:02:36 -08003466 ocfs2_complete_edge_insert(handle, left_path, right_path,
3467 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003468 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003469out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003470 ocfs2_free_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003471 return ret;
3472}
3473
Joel Becker4fe82c32009-02-13 02:16:08 -08003474static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003475 struct ocfs2_path *right_path,
3476 struct ocfs2_path **ret_left_path)
3477{
3478 int ret;
3479 u32 left_cpos;
3480 struct ocfs2_path *left_path = NULL;
3481
3482 *ret_left_path = NULL;
3483
3484 /* This function shouldn't be called for non-trees. */
3485 BUG_ON(right_path->p_tree_depth == 0);
3486
Joel Becker4fe82c32009-02-13 02:16:08 -08003487 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003488 right_path, &left_cpos);
3489 if (ret) {
3490 mlog_errno(ret);
3491 goto out;
3492 }
3493
3494 /* This function shouldn't be called for the leftmost leaf. */
3495 BUG_ON(left_cpos == 0);
3496
Joel Beckerffdd7a52008-10-17 22:32:01 -07003497 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003498 if (!left_path) {
3499 ret = -ENOMEM;
3500 mlog_errno(ret);
3501 goto out;
3502 }
3503
Joel Becker4fe82c32009-02-13 02:16:08 -08003504 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003505 if (ret) {
3506 mlog_errno(ret);
3507 goto out;
3508 }
3509
3510 *ret_left_path = left_path;
3511out:
3512 if (ret)
3513 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003514 return ret;
3515}
3516
3517/*
3518 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003519 * onto the tail of the record "before" it.
3520 * For index > 0, the "before" means the extent rec at index - 1.
3521 *
3522 * For index == 0, the "before" means the last record of the previous
3523 * extent block. And there is also a situation that we may need to
3524 * remove the rightmost leaf extent block in the right_path and change
3525 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003526 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003527static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003528 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003529 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003530 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003531 struct ocfs2_cached_dealloc_ctxt *dealloc,
3532 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003533{
Tao Ma677b9752008-01-30 14:21:05 +08003534 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003535 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3536 struct ocfs2_extent_rec *left_rec;
3537 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003538 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3539 struct buffer_head *bh = path_leaf_bh(right_path);
3540 struct buffer_head *root_bh = NULL;
3541 struct ocfs2_path *left_path = NULL;
3542 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003543
Tao Ma677b9752008-01-30 14:21:05 +08003544 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003545
Mark Fasheh328d5752007-06-18 10:48:04 -07003546 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003547 if (index == 0) {
3548 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003549 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003550 if (ret) {
3551 mlog_errno(ret);
Markus Elfring992ef6e2015-04-14 15:42:51 -07003552 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003553 }
3554
3555 left_el = path_leaf_el(left_path);
3556 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3557 le16_to_cpu(left_el->l_count));
3558
3559 left_rec = &left_el->l_recs[
3560 le16_to_cpu(left_el->l_next_free_rec) - 1];
3561 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3562 le16_to_cpu(left_rec->e_leaf_clusters) !=
3563 le32_to_cpu(split_rec->e_cpos));
3564
Joel Becker7dc02802009-02-12 19:20:13 -08003565 subtree_index = ocfs2_find_subtree_root(et, left_path,
3566 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003567
3568 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3569 handle->h_buffer_credits,
3570 left_path);
3571 if (ret) {
3572 mlog_errno(ret);
3573 goto out;
3574 }
3575
3576 root_bh = left_path->p_node[subtree_index].bh;
3577 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3578
Joel Becker4fe82c32009-02-13 02:16:08 -08003579 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003580 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003581 if (ret) {
3582 mlog_errno(ret);
3583 goto out;
3584 }
3585
3586 for (i = subtree_index + 1;
3587 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003588 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003589 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003590 if (ret) {
3591 mlog_errno(ret);
3592 goto out;
3593 }
3594
Joel Becker4fe82c32009-02-13 02:16:08 -08003595 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003596 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003597 if (ret) {
3598 mlog_errno(ret);
3599 goto out;
3600 }
3601 }
3602 } else {
3603 left_rec = &el->l_recs[index - 1];
3604 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3605 has_empty_extent = 1;
3606 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003607
Joel Becker4fe82c32009-02-13 02:16:08 -08003608 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003609 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003610 if (ret) {
3611 mlog_errno(ret);
3612 goto out;
3613 }
3614
3615 if (has_empty_extent && index == 1) {
3616 /*
3617 * The easy case - we can just plop the record right in.
3618 */
3619 *left_rec = *split_rec;
3620
3621 has_empty_extent = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003622 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003623 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003624
3625 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3626 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003627 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3628 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003629 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3630
3631 ocfs2_cleanup_merge(el, index);
3632
Joel Beckerec20cec2010-03-19 14:13:52 -07003633 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003634 if (left_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003635 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
Tao Ma677b9752008-01-30 14:21:05 +08003636
3637 /*
3638 * In the situation that the right_rec is empty and the extent
3639 * block is empty also, ocfs2_complete_edge_insert can't handle
3640 * it and we need to delete the right extent block.
3641 */
3642 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3643 le16_to_cpu(el->l_next_free_rec) == 1) {
3644
Joel Becker70f18c02009-02-13 02:09:31 -08003645 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003646 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003647 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003648 if (ret) {
3649 mlog_errno(ret);
3650 goto out;
3651 }
3652
3653 /* Now the rightmost extent block has been deleted.
3654 * So we use the new rightmost path.
3655 */
3656 ocfs2_mv_path(right_path, left_path);
3657 left_path = NULL;
3658 } else
Joel Becker4619c732009-02-12 19:02:36 -08003659 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003660 right_path, subtree_index);
3661 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003662out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003663 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003664 return ret;
3665}
3666
Joel Beckerc495dd22009-02-13 02:19:11 -08003667static int ocfs2_try_to_merge_extent(handle_t *handle,
3668 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003669 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003670 int split_index,
3671 struct ocfs2_extent_rec *split_rec,
3672 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003673 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003674{
Tao Mao518d7262007-08-28 17:25:35 -07003675 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003676 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003677 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3678
3679 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3680
Tao Mao518d7262007-08-28 17:25:35 -07003681 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
3682 /*
3683 * The merge code will need to create an empty
3684 * extent to take the place of the newly
3685 * emptied slot. Remove any pre-existing empty
3686 * extents - having more than one in a leaf is
3687 * illegal.
3688 */
Joel Becker70f18c02009-02-13 02:09:31 -08003689 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003690 if (ret) {
3691 mlog_errno(ret);
3692 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003693 }
Tao Mao518d7262007-08-28 17:25:35 -07003694 split_index--;
3695 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003696 }
3697
3698 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3699 /*
3700 * Left-right contig implies this.
3701 */
3702 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003703
3704 /*
3705 * Since the leftright insert always covers the entire
3706 * extent, this call will delete the insert record
3707 * entirely, resulting in an empty extent record added to
3708 * the extent block.
3709 *
3710 * Since the adding of an empty extent shifts
3711 * everything back to the right, there's no need to
3712 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003713 *
3714 * When the split_index is zero, we need to merge it to the
3715 * prevoius extent block. It is more efficient and easier
3716 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003717 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003718 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003719 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003720 if (ret) {
3721 mlog_errno(ret);
3722 goto out;
3723 }
3724
3725 /*
3726 * We can only get this from logic error above.
3727 */
3728 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3729
Tao Ma677b9752008-01-30 14:21:05 +08003730 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003731 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003732 if (ret) {
3733 mlog_errno(ret);
3734 goto out;
3735 }
Tao Ma677b9752008-01-30 14:21:05 +08003736
Mark Fasheh328d5752007-06-18 10:48:04 -07003737 rec = &el->l_recs[split_index];
3738
3739 /*
3740 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003741 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003742 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003743 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3744 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003745
Mark Fasheh328d5752007-06-18 10:48:04 -07003746 if (ret) {
3747 mlog_errno(ret);
3748 goto out;
3749 }
3750
Joel Becker70f18c02009-02-13 02:09:31 -08003751 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003752 /*
3753 * Error from this last rotate is not critical, so
3754 * print but don't bubble it up.
3755 */
3756 if (ret)
3757 mlog_errno(ret);
3758 ret = 0;
3759 } else {
3760 /*
3761 * Merge a record to the left or right.
3762 *
3763 * 'contig_type' is relative to the existing record,
3764 * so for example, if we're "right contig", it's to
3765 * the record on the left (hence the left merge).
3766 */
3767 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003768 ret = ocfs2_merge_rec_left(path, handle, et,
3769 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003770 split_index);
3771 if (ret) {
3772 mlog_errno(ret);
3773 goto out;
3774 }
3775 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003776 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003777 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003778 split_index);
3779 if (ret) {
3780 mlog_errno(ret);
3781 goto out;
3782 }
3783 }
3784
3785 if (ctxt->c_split_covers_rec) {
3786 /*
3787 * The merge may have left an empty extent in
3788 * our leaf. Try to rotate it away.
3789 */
Joel Becker70f18c02009-02-13 02:09:31 -08003790 ret = ocfs2_rotate_tree_left(handle, et, path,
3791 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003792 if (ret)
3793 mlog_errno(ret);
3794 ret = 0;
3795 }
3796 }
3797
3798out:
3799 return ret;
3800}
3801
3802static void ocfs2_subtract_from_rec(struct super_block *sb,
3803 enum ocfs2_split_type split,
3804 struct ocfs2_extent_rec *rec,
3805 struct ocfs2_extent_rec *split_rec)
3806{
3807 u64 len_blocks;
3808
3809 len_blocks = ocfs2_clusters_to_blocks(sb,
3810 le16_to_cpu(split_rec->e_leaf_clusters));
3811
3812 if (split == SPLIT_LEFT) {
3813 /*
3814 * Region is on the left edge of the existing
3815 * record.
3816 */
3817 le32_add_cpu(&rec->e_cpos,
3818 le16_to_cpu(split_rec->e_leaf_clusters));
3819 le64_add_cpu(&rec->e_blkno, len_blocks);
3820 le16_add_cpu(&rec->e_leaf_clusters,
3821 -le16_to_cpu(split_rec->e_leaf_clusters));
3822 } else {
3823 /*
3824 * Region is on the right edge of the existing
3825 * record.
3826 */
3827 le16_add_cpu(&rec->e_leaf_clusters,
3828 -le16_to_cpu(split_rec->e_leaf_clusters));
3829 }
3830}
3831
Mark Fashehdcd05382007-01-16 11:32:23 -08003832/*
3833 * Do the final bits of extent record insertion at the target leaf
3834 * list. If this leaf is part of an allocation tree, it is assumed
3835 * that the tree above has been prepared.
3836 */
Joel Beckerd5628622009-02-13 02:54:36 -08003837static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3838 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003839 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003840 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003841{
3842 int i = insert->ins_contig_index;
3843 unsigned int range;
3844 struct ocfs2_extent_rec *rec;
3845
Mark Fashehe48edee2007-03-07 16:46:57 -08003846 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003847
Mark Fasheh328d5752007-06-18 10:48:04 -07003848 if (insert->ins_split != SPLIT_NONE) {
3849 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3850 BUG_ON(i == -1);
3851 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003852 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3853 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003854 insert_rec);
3855 goto rotate;
3856 }
3857
Mark Fashehdcd05382007-01-16 11:32:23 -08003858 /*
3859 * Contiguous insert - either left or right.
3860 */
3861 if (insert->ins_contig != CONTIG_NONE) {
3862 rec = &el->l_recs[i];
3863 if (insert->ins_contig == CONTIG_LEFT) {
3864 rec->e_blkno = insert_rec->e_blkno;
3865 rec->e_cpos = insert_rec->e_cpos;
3866 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003867 le16_add_cpu(&rec->e_leaf_clusters,
3868 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003869 return;
3870 }
3871
3872 /*
3873 * Handle insert into an empty leaf.
3874 */
3875 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3876 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3877 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3878 el->l_recs[0] = *insert_rec;
3879 el->l_next_free_rec = cpu_to_le16(1);
3880 return;
3881 }
3882
3883 /*
3884 * Appending insert.
3885 */
3886 if (insert->ins_appending == APPEND_TAIL) {
3887 i = le16_to_cpu(el->l_next_free_rec) - 1;
3888 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003889 range = le32_to_cpu(rec->e_cpos)
3890 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003891 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3892
3893 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3894 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003895 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003896 "rec.cpos %u, rec.clusters %u, "
3897 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003898 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003899 le16_to_cpu(el->l_tree_depth),
3900 le16_to_cpu(el->l_count),
3901 le16_to_cpu(el->l_next_free_rec),
3902 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003903 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003904 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003905 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003906 i++;
3907 el->l_recs[i] = *insert_rec;
3908 le16_add_cpu(&el->l_next_free_rec, 1);
3909 return;
3910 }
3911
Mark Fasheh328d5752007-06-18 10:48:04 -07003912rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003913 /*
3914 * Ok, we have to rotate.
3915 *
3916 * At this point, it is safe to assume that inserting into an
3917 * empty leaf and appending to a leaf have both been handled
3918 * above.
3919 *
3920 * This leaf needs to have space, either by the empty 1st
3921 * extent record, or by virtue of an l_next_rec < l_count.
3922 */
3923 ocfs2_rotate_leaf(el, insert_rec);
3924}
3925
Joel Beckerd401dc12009-02-13 02:24:10 -08003926static void ocfs2_adjust_rightmost_records(handle_t *handle,
3927 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003928 struct ocfs2_path *path,
3929 struct ocfs2_extent_rec *insert_rec)
3930{
3931 int ret, i, next_free;
3932 struct buffer_head *bh;
3933 struct ocfs2_extent_list *el;
3934 struct ocfs2_extent_rec *rec;
3935
3936 /*
3937 * Update everything except the leaf block.
3938 */
3939 for (i = 0; i < path->p_tree_depth; i++) {
3940 bh = path->p_node[i].bh;
3941 el = path->p_node[i].el;
3942
3943 next_free = le16_to_cpu(el->l_next_free_rec);
3944 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003945 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07003946 "Owner %llu has a bad extent list\n",
Joel Beckerd401dc12009-02-13 02:24:10 -08003947 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003948 ret = -EIO;
3949 return;
3950 }
3951
3952 rec = &el->l_recs[next_free - 1];
3953
3954 rec->e_int_clusters = insert_rec->e_cpos;
3955 le32_add_cpu(&rec->e_int_clusters,
3956 le16_to_cpu(insert_rec->e_leaf_clusters));
3957 le32_add_cpu(&rec->e_int_clusters,
3958 -le32_to_cpu(rec->e_cpos));
3959
Joel Beckerec20cec2010-03-19 14:13:52 -07003960 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07003961 }
3962}
3963
Joel Beckerd401dc12009-02-13 02:24:10 -08003964static int ocfs2_append_rec_to_path(handle_t *handle,
3965 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003966 struct ocfs2_extent_rec *insert_rec,
3967 struct ocfs2_path *right_path,
3968 struct ocfs2_path **ret_left_path)
3969{
Mark Fasheh328d5752007-06-18 10:48:04 -07003970 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003971 struct ocfs2_extent_list *el;
3972 struct ocfs2_path *left_path = NULL;
3973
3974 *ret_left_path = NULL;
3975
3976 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003977 * This shouldn't happen for non-trees. The extent rec cluster
3978 * count manipulation below only works for interior nodes.
3979 */
3980 BUG_ON(right_path->p_tree_depth == 0);
3981
3982 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08003983 * If our appending insert is at the leftmost edge of a leaf,
3984 * then we might need to update the rightmost records of the
3985 * neighboring path.
3986 */
3987 el = path_leaf_el(right_path);
3988 next_free = le16_to_cpu(el->l_next_free_rec);
3989 if (next_free == 0 ||
3990 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
3991 u32 left_cpos;
3992
Joel Beckerd401dc12009-02-13 02:24:10 -08003993 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3994 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08003995 if (ret) {
3996 mlog_errno(ret);
3997 goto out;
3998 }
3999
Tao Maa09d09b2011-02-22 08:21:10 +08004000 trace_ocfs2_append_rec_to_path(
4001 (unsigned long long)
4002 ocfs2_metadata_cache_owner(et->et_ci),
4003 le32_to_cpu(insert_rec->e_cpos),
4004 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004005
4006 /*
4007 * No need to worry if the append is already in the
4008 * leftmost leaf.
4009 */
4010 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004011 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004012 if (!left_path) {
4013 ret = -ENOMEM;
4014 mlog_errno(ret);
4015 goto out;
4016 }
4017
Joel Beckerd401dc12009-02-13 02:24:10 -08004018 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004019 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004020 if (ret) {
4021 mlog_errno(ret);
4022 goto out;
4023 }
4024
4025 /*
4026 * ocfs2_insert_path() will pass the left_path to the
4027 * journal for us.
4028 */
4029 }
4030 }
4031
Joel Beckerd401dc12009-02-13 02:24:10 -08004032 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004033 if (ret) {
4034 mlog_errno(ret);
4035 goto out;
4036 }
4037
Joel Beckerd401dc12009-02-13 02:24:10 -08004038 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004039
4040 *ret_left_path = left_path;
4041 ret = 0;
4042out:
4043 if (ret != 0)
4044 ocfs2_free_path(left_path);
4045
4046 return ret;
4047}
4048
Joel Beckerc38e52b2009-02-13 02:56:23 -08004049static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004050 struct ocfs2_path *left_path,
4051 struct ocfs2_path *right_path,
4052 struct ocfs2_extent_rec *split_rec,
4053 enum ocfs2_split_type split)
4054{
4055 int index;
4056 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4057 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4058 struct ocfs2_extent_rec *rec, *tmprec;
4059
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004060 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004061 if (left_path)
4062 left_el = path_leaf_el(left_path);
4063
4064 el = right_el;
4065 insert_el = right_el;
4066 index = ocfs2_search_extent_list(el, cpos);
4067 if (index != -1) {
4068 if (index == 0 && left_path) {
4069 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4070
4071 /*
4072 * This typically means that the record
4073 * started in the left path but moved to the
4074 * right as a result of rotation. We either
4075 * move the existing record to the left, or we
4076 * do the later insert there.
4077 *
4078 * In this case, the left path should always
4079 * exist as the rotate code will have passed
4080 * it back for a post-insert update.
4081 */
4082
4083 if (split == SPLIT_LEFT) {
4084 /*
4085 * It's a left split. Since we know
4086 * that the rotate code gave us an
4087 * empty extent in the left path, we
4088 * can just do the insert there.
4089 */
4090 insert_el = left_el;
4091 } else {
4092 /*
4093 * Right split - we have to move the
4094 * existing record over to the left
4095 * leaf. The insert will be into the
4096 * newly created empty extent in the
4097 * right leaf.
4098 */
4099 tmprec = &right_el->l_recs[index];
4100 ocfs2_rotate_leaf(left_el, tmprec);
4101 el = left_el;
4102
4103 memset(tmprec, 0, sizeof(*tmprec));
4104 index = ocfs2_search_extent_list(left_el, cpos);
4105 BUG_ON(index == -1);
4106 }
4107 }
4108 } else {
4109 BUG_ON(!left_path);
4110 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4111 /*
4112 * Left path is easy - we can just allow the insert to
4113 * happen.
4114 */
4115 el = left_el;
4116 insert_el = left_el;
4117 index = ocfs2_search_extent_list(el, cpos);
4118 BUG_ON(index == -1);
4119 }
4120
4121 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004122 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4123 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004124 ocfs2_rotate_leaf(insert_el, split_rec);
4125}
4126
Mark Fashehdcd05382007-01-16 11:32:23 -08004127/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004128 * This function only does inserts on an allocation b-tree. For tree
4129 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004130 *
4131 * right_path is the path we want to do the actual insert
4132 * in. left_path should only be passed in if we need to update that
4133 * portion of the tree after an edge insert.
4134 */
Joel Becker3505bec2009-02-13 02:57:58 -08004135static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004136 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004137 struct ocfs2_path *left_path,
4138 struct ocfs2_path *right_path,
4139 struct ocfs2_extent_rec *insert_rec,
4140 struct ocfs2_insert_type *insert)
4141{
4142 int ret, subtree_index;
4143 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004144
Mark Fashehdcd05382007-01-16 11:32:23 -08004145 if (left_path) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004146 /*
4147 * There's a chance that left_path got passed back to
4148 * us without being accounted for in the
4149 * journal. Extend our transaction here to be sure we
4150 * can change those blocks.
4151 */
Tao Mac901fb02010-04-26 14:34:57 +08004152 ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08004153 if (ret < 0) {
4154 mlog_errno(ret);
4155 goto out;
4156 }
4157
Joel Becker7dc02802009-02-12 19:20:13 -08004158 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004159 if (ret < 0) {
4160 mlog_errno(ret);
4161 goto out;
4162 }
4163 }
4164
Mark Fashehe8aed342007-12-03 16:43:01 -08004165 /*
4166 * Pass both paths to the journal. The majority of inserts
4167 * will be touching all components anyway.
4168 */
Joel Becker7dc02802009-02-12 19:20:13 -08004169 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004170 if (ret < 0) {
4171 mlog_errno(ret);
4172 goto out;
4173 }
4174
Mark Fasheh328d5752007-06-18 10:48:04 -07004175 if (insert->ins_split != SPLIT_NONE) {
4176 /*
4177 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004178 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004179 * function sort it all out.
4180 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004181 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004182 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004183
4184 /*
4185 * Split might have modified either leaf and we don't
4186 * have a guarantee that the later edge insert will
4187 * dirty this for us.
4188 */
4189 if (left_path)
Joel Beckerec20cec2010-03-19 14:13:52 -07004190 ocfs2_journal_dirty(handle,
4191 path_leaf_bh(left_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07004192 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004193 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4194 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004195
Joel Beckerec20cec2010-03-19 14:13:52 -07004196 ocfs2_journal_dirty(handle, leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004197
4198 if (left_path) {
4199 /*
4200 * The rotate code has indicated that we need to fix
4201 * up portions of the tree after the insert.
4202 *
4203 * XXX: Should we extend the transaction here?
4204 */
Joel Becker7dc02802009-02-12 19:20:13 -08004205 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004206 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004207 ocfs2_complete_edge_insert(handle, left_path, right_path,
4208 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004209 }
4210
4211 ret = 0;
4212out:
4213 return ret;
4214}
4215
Joel Becker3505bec2009-02-13 02:57:58 -08004216static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004217 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004218 struct ocfs2_extent_rec *insert_rec,
4219 struct ocfs2_insert_type *type)
4220{
4221 int ret, rotate = 0;
4222 u32 cpos;
4223 struct ocfs2_path *right_path = NULL;
4224 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004225 struct ocfs2_extent_list *el;
4226
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004227 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004228
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004229 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004230 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004231 if (ret) {
4232 mlog_errno(ret);
4233 goto out;
4234 }
4235
4236 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004237 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004238 goto out_update_clusters;
4239 }
4240
Joel Beckerffdd7a52008-10-17 22:32:01 -07004241 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004242 if (!right_path) {
4243 ret = -ENOMEM;
4244 mlog_errno(ret);
4245 goto out;
4246 }
4247
4248 /*
4249 * Determine the path to start with. Rotations need the
4250 * rightmost path, everything else can go directly to the
4251 * target leaf.
4252 */
4253 cpos = le32_to_cpu(insert_rec->e_cpos);
4254 if (type->ins_appending == APPEND_NONE &&
4255 type->ins_contig == CONTIG_NONE) {
4256 rotate = 1;
4257 cpos = UINT_MAX;
4258 }
4259
Joel Beckerfacdb772009-02-12 18:08:48 -08004260 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004261 if (ret) {
4262 mlog_errno(ret);
4263 goto out;
4264 }
4265
4266 /*
4267 * Rotations and appends need special treatment - they modify
4268 * parts of the tree's above them.
4269 *
4270 * Both might pass back a path immediate to the left of the
4271 * one being inserted to. This will be cause
4272 * ocfs2_insert_path() to modify the rightmost records of
4273 * left_path to account for an edge insert.
4274 *
4275 * XXX: When modifying this code, keep in mind that an insert
4276 * can wind up skipping both of these two special cases...
4277 */
4278 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004279 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004280 le32_to_cpu(insert_rec->e_cpos),
4281 right_path, &left_path);
4282 if (ret) {
4283 mlog_errno(ret);
4284 goto out;
4285 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004286
4287 /*
4288 * ocfs2_rotate_tree_right() might have extended the
4289 * transaction without re-journaling our tree root.
4290 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004291 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004292 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004293 if (ret) {
4294 mlog_errno(ret);
4295 goto out;
4296 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004297 } else if (type->ins_appending == APPEND_TAIL
4298 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004299 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004300 right_path, &left_path);
4301 if (ret) {
4302 mlog_errno(ret);
4303 goto out;
4304 }
4305 }
4306
Joel Becker3505bec2009-02-13 02:57:58 -08004307 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004308 insert_rec, type);
4309 if (ret) {
4310 mlog_errno(ret);
4311 goto out;
4312 }
4313
4314out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004315 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004316 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004317 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004318
Joel Beckerec20cec2010-03-19 14:13:52 -07004319 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004320
4321out:
4322 ocfs2_free_path(left_path);
4323 ocfs2_free_path(right_path);
4324
4325 return ret;
4326}
4327
Xue jiufei9f99ad02015-06-24 16:55:07 -07004328static int ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
Joel Beckera2970292009-02-13 03:09:54 -08004329 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004330 struct ocfs2_extent_list *el, int index,
Xue jiufei9f99ad02015-06-24 16:55:07 -07004331 struct ocfs2_extent_rec *split_rec,
4332 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07004333{
Xue jiufei9f99ad02015-06-24 16:55:07 -07004334 int status = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07004335 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004336 u32 left_cpos, right_cpos;
4337 struct ocfs2_extent_rec *rec = NULL;
4338 struct ocfs2_extent_list *new_el;
4339 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4340 struct buffer_head *bh;
4341 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004342 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004343
4344 if (index > 0) {
4345 rec = &el->l_recs[index - 1];
4346 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004347 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004348 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004349 goto exit;
Tao Maad5a4d72008-01-30 14:21:32 +08004350
4351 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004352 left_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004353 if (!left_path) {
4354 status = -ENOMEM;
4355 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004356 goto exit;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004357 }
Tao Maad5a4d72008-01-30 14:21:32 +08004358
Joel Beckera2970292009-02-13 03:09:54 -08004359 status = ocfs2_find_path(et->et_ci, left_path,
4360 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004361 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004362 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004363
4364 new_el = path_leaf_el(left_path);
4365
4366 if (le16_to_cpu(new_el->l_next_free_rec) !=
4367 le16_to_cpu(new_el->l_count)) {
4368 bh = path_leaf_bh(left_path);
4369 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004370 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07004371 "Extent block #%llu has an invalid l_next_free_rec of %d. It should have matched the l_count of %d\n",
Joel Becker5e965812008-11-13 14:49:16 -08004372 (unsigned long long)le64_to_cpu(eb->h_blkno),
4373 le16_to_cpu(new_el->l_next_free_rec),
4374 le16_to_cpu(new_el->l_count));
4375 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004376 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004377 }
4378 rec = &new_el->l_recs[
4379 le16_to_cpu(new_el->l_next_free_rec) - 1];
4380 }
4381 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004382
4383 /*
4384 * We're careful to check for an empty extent record here -
4385 * the merge code will know what to do if it sees one.
4386 */
Tao Maad5a4d72008-01-30 14:21:32 +08004387 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004388 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4389 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4390 ret = CONTIG_RIGHT;
4391 } else {
Tao Ma853a3a12009-08-18 11:22:18 +08004392 ret = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004393 }
4394 }
4395
Tao Maad5a4d72008-01-30 14:21:32 +08004396 rec = NULL;
4397 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4398 rec = &el->l_recs[index + 1];
4399 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4400 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004401 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004402 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004403 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004404
4405 if (right_cpos == 0)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004406 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004407
Joel Beckerffdd7a52008-10-17 22:32:01 -07004408 right_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004409 if (!right_path) {
4410 status = -ENOMEM;
4411 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004412 goto free_left_path;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004413 }
Tao Maad5a4d72008-01-30 14:21:32 +08004414
Joel Beckera2970292009-02-13 03:09:54 -08004415 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004416 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004417 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004418
4419 new_el = path_leaf_el(right_path);
4420 rec = &new_el->l_recs[0];
4421 if (ocfs2_is_empty_extent(rec)) {
4422 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4423 bh = path_leaf_bh(right_path);
4424 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004425 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07004426 "Extent block #%llu has an invalid l_next_free_rec of %d\n",
Joel Becker5e965812008-11-13 14:49:16 -08004427 (unsigned long long)le64_to_cpu(eb->h_blkno),
4428 le16_to_cpu(new_el->l_next_free_rec));
4429 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004430 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004431 }
4432 rec = &new_el->l_recs[1];
4433 }
4434 }
4435
4436 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004437 enum ocfs2_contig_type contig_type;
4438
Tao Ma853a3a12009-08-18 11:22:18 +08004439 contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004440
4441 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4442 ret = CONTIG_LEFTRIGHT;
4443 else if (ret == CONTIG_NONE)
4444 ret = contig_type;
4445 }
4446
Markus Elfring06a269cc2015-04-14 15:42:48 -07004447free_right_path:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07004448 ocfs2_free_path(right_path);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004449free_left_path:
4450 ocfs2_free_path(left_path);
4451exit:
Xue jiufei9f99ad02015-06-24 16:55:07 -07004452 if (status == 0)
4453 ctxt->c_contig_type = ret;
4454
4455 return status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004456}
4457
Joel Becker1ef61b32009-02-13 03:12:33 -08004458static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004459 struct ocfs2_insert_type *insert,
4460 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004461 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004462{
4463 int i;
4464 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4465
Mark Fashehe48edee2007-03-07 16:46:57 -08004466 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4467
Mark Fashehdcd05382007-01-16 11:32:23 -08004468 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Tao Ma853a3a12009-08-18 11:22:18 +08004469 contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4470 insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004471 if (contig_type != CONTIG_NONE) {
4472 insert->ins_contig_index = i;
4473 break;
4474 }
4475 }
4476 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004477
4478 if (insert->ins_contig != CONTIG_NONE) {
4479 struct ocfs2_extent_rec *rec =
4480 &el->l_recs[insert->ins_contig_index];
4481 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4482 le16_to_cpu(insert_rec->e_leaf_clusters);
4483
4484 /*
4485 * Caller might want us to limit the size of extents, don't
4486 * calculate contiguousness if we might exceed that limit.
4487 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004488 if (et->et_max_leaf_clusters &&
4489 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004490 insert->ins_contig = CONTIG_NONE;
4491 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004492}
4493
4494/*
4495 * This should only be called against the righmost leaf extent list.
4496 *
4497 * ocfs2_figure_appending_type() will figure out whether we'll have to
4498 * insert at the tail of the rightmost leaf.
4499 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004500 * This should also work against the root extent list for tree's with 0
4501 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004502 * then the logic here makes sense.
4503 */
4504static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4505 struct ocfs2_extent_list *el,
4506 struct ocfs2_extent_rec *insert_rec)
4507{
4508 int i;
4509 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4510 struct ocfs2_extent_rec *rec;
4511
4512 insert->ins_appending = APPEND_NONE;
4513
Mark Fashehe48edee2007-03-07 16:46:57 -08004514 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004515
4516 if (!el->l_next_free_rec)
4517 goto set_tail_append;
4518
4519 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4520 /* Were all records empty? */
4521 if (le16_to_cpu(el->l_next_free_rec) == 1)
4522 goto set_tail_append;
4523 }
4524
4525 i = le16_to_cpu(el->l_next_free_rec) - 1;
4526 rec = &el->l_recs[i];
4527
Mark Fashehe48edee2007-03-07 16:46:57 -08004528 if (cpos >=
4529 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004530 goto set_tail_append;
4531
4532 return;
4533
4534set_tail_append:
4535 insert->ins_appending = APPEND_TAIL;
4536}
4537
4538/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004539 * Helper function called at the beginning of an insert.
Mark Fashehdcd05382007-01-16 11:32:23 -08004540 *
4541 * This computes a few things that are commonly used in the process of
4542 * inserting into the btree:
4543 * - Whether the new extent is contiguous with an existing one.
4544 * - The current tree depth.
4545 * - Whether the insert is an appending one.
4546 * - The total # of free records in the tree.
4547 *
4548 * All of the information is stored on the ocfs2_insert_type
4549 * structure.
4550 */
Joel Becker627961b2009-02-13 03:14:38 -08004551static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004552 struct buffer_head **last_eb_bh,
4553 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004554 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004555 struct ocfs2_insert_type *insert)
4556{
4557 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004558 struct ocfs2_extent_block *eb;
4559 struct ocfs2_extent_list *el;
4560 struct ocfs2_path *path = NULL;
4561 struct buffer_head *bh = NULL;
4562
Mark Fasheh328d5752007-06-18 10:48:04 -07004563 insert->ins_split = SPLIT_NONE;
4564
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004565 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004566 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4567
4568 if (el->l_tree_depth) {
4569 /*
4570 * If we have tree depth, we read in the
4571 * rightmost extent block ahead of time as
4572 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4573 * may want it later.
4574 */
Joel Becker3d03a302009-02-12 17:49:26 -08004575 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004576 ocfs2_et_get_last_eb_blk(et),
4577 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004578 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08004579 mlog_errno(ret);
Mark Fashehdcd05382007-01-16 11:32:23 -08004580 goto out;
4581 }
4582 eb = (struct ocfs2_extent_block *) bh->b_data;
4583 el = &eb->h_list;
4584 }
4585
4586 /*
4587 * Unless we have a contiguous insert, we'll need to know if
4588 * there is room left in our allocation tree for another
4589 * extent record.
4590 *
4591 * XXX: This test is simplistic, we can search for empty
4592 * extent records too.
4593 */
Tao Maoc77534f2007-08-28 17:22:33 -07004594 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004595 le16_to_cpu(el->l_next_free_rec);
4596
4597 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004598 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004599 ocfs2_figure_appending_type(insert, el, insert_rec);
4600 return 0;
4601 }
4602
Joel Beckerffdd7a52008-10-17 22:32:01 -07004603 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004604 if (!path) {
4605 ret = -ENOMEM;
4606 mlog_errno(ret);
4607 goto out;
4608 }
4609
4610 /*
4611 * In the case that we're inserting past what the tree
4612 * currently accounts for, ocfs2_find_path() will return for
4613 * us the rightmost tree path. This is accounted for below in
4614 * the appending code.
4615 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004616 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004617 if (ret) {
4618 mlog_errno(ret);
4619 goto out;
4620 }
4621
4622 el = path_leaf_el(path);
4623
4624 /*
4625 * Now that we have the path, there's two things we want to determine:
4626 * 1) Contiguousness (also set contig_index if this is so)
4627 *
4628 * 2) Are we doing an append? We can trivially break this up
4629 * into two types of appends: simple record append, or a
4630 * rotate inside the tail leaf.
4631 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004632 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004633
4634 /*
4635 * The insert code isn't quite ready to deal with all cases of
4636 * left contiguousness. Specifically, if it's an insert into
4637 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004638 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004639 * left. For now, just catch that case and fool the layers
4640 * above us. This works just fine for tree_depth == 0, which
4641 * is why we allow that above.
4642 */
4643 if (insert->ins_contig == CONTIG_LEFT &&
4644 insert->ins_contig_index == 0)
4645 insert->ins_contig = CONTIG_NONE;
4646
4647 /*
4648 * Ok, so we can simply compare against last_eb to figure out
4649 * whether the path doesn't exist. This will only happen in
4650 * the case that we're doing a tail append, so maybe we can
4651 * take advantage of that information somehow.
4652 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004653 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004654 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004655 /*
4656 * Ok, ocfs2_find_path() returned us the rightmost
4657 * tree path. This might be an appending insert. There are
4658 * two cases:
4659 * 1) We're doing a true append at the tail:
4660 * -This might even be off the end of the leaf
4661 * 2) We're "appending" by rotating in the tail
4662 */
4663 ocfs2_figure_appending_type(insert, el, insert_rec);
4664 }
4665
4666out:
4667 ocfs2_free_path(path);
4668
4669 if (ret == 0)
4670 *last_eb_bh = bh;
4671 else
4672 brelse(bh);
4673 return ret;
4674}
4675
4676/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004677 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004678 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004679 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004680 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004681int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004682 struct ocfs2_extent_tree *et,
4683 u32 cpos,
4684 u64 start_blk,
4685 u32 new_clusters,
4686 u8 flags,
4687 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004688{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004689 int status;
Tao Maoc77534f2007-08-28 17:22:33 -07004690 int uninitialized_var(free_records);
Mark Fashehccd979b2005-12-15 14:31:24 -08004691 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004692 struct ocfs2_insert_type insert = {0, };
4693 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004694
Tao Maa09d09b2011-02-22 08:21:10 +08004695 trace_ocfs2_insert_extent_start(
4696 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4697 cpos, new_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08004698
Mark Fashehe48edee2007-03-07 16:46:57 -08004699 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004700 rec.e_cpos = cpu_to_le32(cpos);
4701 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004702 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004703 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004704 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004705 if (status) {
4706 mlog_errno(status);
4707 goto bail;
4708 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004709
Joel Becker627961b2009-02-13 03:14:38 -08004710 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004711 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004712 if (status < 0) {
4713 mlog_errno(status);
4714 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004715 }
4716
Tao Maa09d09b2011-02-22 08:21:10 +08004717 trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4718 insert.ins_contig_index, free_records,
4719 insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004720
Tao Maoc77534f2007-08-28 17:22:33 -07004721 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004722 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004723 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004724 meta_ac);
4725 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004726 mlog_errno(status);
4727 goto bail;
4728 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004729 }
4730
Mark Fashehdcd05382007-01-16 11:32:23 -08004731 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004732 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004733 if (status < 0)
4734 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004735 else
4736 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004737
4738bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004739 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004740
Tao Maf56654c2008-08-18 17:38:48 +08004741 return status;
4742}
4743
Tao Ma0eb8d472008-08-18 17:38:45 +08004744/*
4745 * Allcate and add clusters into the extent b-tree.
4746 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004747 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d472008-08-18 17:38:45 +08004748 * it is not limited to the file storage. Any extent tree can use this
4749 * function if it implements the proper ocfs2_extent_tree.
4750 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004751int ocfs2_add_clusters_in_btree(handle_t *handle,
4752 struct ocfs2_extent_tree *et,
Tao Ma0eb8d472008-08-18 17:38:45 +08004753 u32 *logical_offset,
4754 u32 clusters_to_add,
4755 int mark_unwritten,
Tao Ma0eb8d472008-08-18 17:38:45 +08004756 struct ocfs2_alloc_context *data_ac,
4757 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004758 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d472008-08-18 17:38:45 +08004759{
Tao Maa09d09b2011-02-22 08:21:10 +08004760 int status = 0, err = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004761 int need_free = 0;
Tao Ma0eb8d472008-08-18 17:38:45 +08004762 int free_extents;
4763 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4764 u32 bit_off, num_bits;
4765 u64 block;
4766 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004767 struct ocfs2_super *osb =
4768 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d472008-08-18 17:38:45 +08004769
4770 BUG_ON(!clusters_to_add);
4771
4772 if (mark_unwritten)
4773 flags = OCFS2_EXT_UNWRITTEN;
4774
Joel Becker3d03a302009-02-12 17:49:26 -08004775 free_extents = ocfs2_num_free_extents(osb, et);
Tao Ma0eb8d472008-08-18 17:38:45 +08004776 if (free_extents < 0) {
4777 status = free_extents;
4778 mlog_errno(status);
4779 goto leave;
4780 }
4781
4782 /* there are two cases which could cause us to EAGAIN in the
4783 * we-need-more-metadata case:
4784 * 1) we haven't reserved *any*
4785 * 2) we are so fragmented, we've needed to add metadata too
4786 * many times. */
4787 if (!free_extents && !meta_ac) {
Tao Maa09d09b2011-02-22 08:21:10 +08004788 err = -1;
Tao Ma0eb8d472008-08-18 17:38:45 +08004789 status = -EAGAIN;
4790 reason = RESTART_META;
4791 goto leave;
4792 } else if ((!free_extents)
4793 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004794 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Maa09d09b2011-02-22 08:21:10 +08004795 err = -2;
Tao Ma0eb8d472008-08-18 17:38:45 +08004796 status = -EAGAIN;
4797 reason = RESTART_META;
4798 goto leave;
4799 }
4800
Joel Becker1ed9b772010-05-06 13:59:06 +08004801 status = __ocfs2_claim_clusters(handle, data_ac, 1,
Tao Ma0eb8d472008-08-18 17:38:45 +08004802 clusters_to_add, &bit_off, &num_bits);
4803 if (status < 0) {
4804 if (status != -ENOSPC)
4805 mlog_errno(status);
4806 goto leave;
4807 }
4808
4809 BUG_ON(num_bits > clusters_to_add);
4810
Joel Becker13723d02008-10-17 19:25:01 -07004811 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004812 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004813 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d472008-08-18 17:38:45 +08004814 if (status < 0) {
4815 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004816 need_free = 1;
4817 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004818 }
4819
4820 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Maa09d09b2011-02-22 08:21:10 +08004821 trace_ocfs2_add_clusters_in_btree(
4822 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4823 bit_off, num_bits);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004824 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004825 num_bits, flags, meta_ac);
Tao Ma0eb8d472008-08-18 17:38:45 +08004826 if (status < 0) {
4827 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004828 need_free = 1;
4829 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004830 }
4831
Joel Beckerec20cec2010-03-19 14:13:52 -07004832 ocfs2_journal_dirty(handle, et->et_root_bh);
Tao Ma0eb8d472008-08-18 17:38:45 +08004833
4834 clusters_to_add -= num_bits;
4835 *logical_offset += num_bits;
4836
4837 if (clusters_to_add) {
Tao Maa09d09b2011-02-22 08:21:10 +08004838 err = clusters_to_add;
Tao Ma0eb8d472008-08-18 17:38:45 +08004839 status = -EAGAIN;
4840 reason = RESTART_TRANS;
4841 }
4842
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004843bail:
4844 if (need_free) {
4845 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
4846 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
4847 bit_off, num_bits);
4848 else
4849 ocfs2_free_clusters(handle,
4850 data_ac->ac_inode,
4851 data_ac->ac_bh,
4852 ocfs2_clusters_to_blocks(osb->sb, bit_off),
4853 num_bits);
4854 }
4855
Tao Ma0eb8d472008-08-18 17:38:45 +08004856leave:
Tao Ma0eb8d472008-08-18 17:38:45 +08004857 if (reason_ret)
4858 *reason_ret = reason;
Tao Maa09d09b2011-02-22 08:21:10 +08004859 trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
Tao Ma0eb8d472008-08-18 17:38:45 +08004860 return status;
4861}
4862
Mark Fasheh328d5752007-06-18 10:48:04 -07004863static void ocfs2_make_right_split_rec(struct super_block *sb,
4864 struct ocfs2_extent_rec *split_rec,
4865 u32 cpos,
4866 struct ocfs2_extent_rec *rec)
4867{
4868 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4869 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4870
4871 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4872
4873 split_rec->e_cpos = cpu_to_le32(cpos);
4874 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4875
4876 split_rec->e_blkno = rec->e_blkno;
4877 le64_add_cpu(&split_rec->e_blkno,
4878 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4879
4880 split_rec->e_flags = rec->e_flags;
4881}
4882
Joel Beckerd2311292009-02-13 03:43:22 -08004883static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004884 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004885 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004886 struct buffer_head **last_eb_bh,
4887 int split_index,
4888 struct ocfs2_extent_rec *orig_split_rec,
4889 struct ocfs2_alloc_context *meta_ac)
4890{
4891 int ret = 0, depth;
4892 unsigned int insert_range, rec_range, do_leftright = 0;
4893 struct ocfs2_extent_rec tmprec;
4894 struct ocfs2_extent_list *rightmost_el;
4895 struct ocfs2_extent_rec rec;
4896 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4897 struct ocfs2_insert_type insert;
4898 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004899
4900leftright:
4901 /*
4902 * Store a copy of the record on the stack - it might move
4903 * around as the tree is manipulated below.
4904 */
4905 rec = path_leaf_el(path)->l_recs[split_index];
4906
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004907 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004908
4909 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4910 if (depth) {
4911 BUG_ON(!(*last_eb_bh));
4912 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4913 rightmost_el = &eb->h_list;
4914 }
4915
4916 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4917 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004918 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004919 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004920 if (ret) {
4921 mlog_errno(ret);
4922 goto out;
4923 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004924 }
4925
4926 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4927 insert.ins_appending = APPEND_NONE;
4928 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004929 insert.ins_tree_depth = depth;
4930
4931 insert_range = le32_to_cpu(split_rec.e_cpos) +
4932 le16_to_cpu(split_rec.e_leaf_clusters);
4933 rec_range = le32_to_cpu(rec.e_cpos) +
4934 le16_to_cpu(rec.e_leaf_clusters);
4935
4936 if (split_rec.e_cpos == rec.e_cpos) {
4937 insert.ins_split = SPLIT_LEFT;
4938 } else if (insert_range == rec_range) {
4939 insert.ins_split = SPLIT_RIGHT;
4940 } else {
4941 /*
4942 * Left/right split. We fake this as a right split
4943 * first and then make a second pass as a left split.
4944 */
4945 insert.ins_split = SPLIT_RIGHT;
4946
Joel Beckerd2311292009-02-13 03:43:22 -08004947 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4948 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004949
4950 split_rec = tmprec;
4951
4952 BUG_ON(do_leftright);
4953 do_leftright = 1;
4954 }
4955
Joel Becker3505bec2009-02-13 02:57:58 -08004956 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004957 if (ret) {
4958 mlog_errno(ret);
4959 goto out;
4960 }
4961
4962 if (do_leftright == 1) {
4963 u32 cpos;
4964 struct ocfs2_extent_list *el;
4965
4966 do_leftright++;
4967 split_rec = *orig_split_rec;
4968
4969 ocfs2_reinit_path(path, 1);
4970
4971 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004972 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004973 if (ret) {
4974 mlog_errno(ret);
4975 goto out;
4976 }
4977
4978 el = path_leaf_el(path);
4979 split_index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07004980 if (split_index == -1) {
4981 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07004982 "Owner %llu has an extent at cpos %u which can no longer be found\n",
4983 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4984 cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07004985 ret = -EROFS;
4986 goto out;
4987 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004988 goto leftright;
4989 }
4990out:
4991
4992 return ret;
4993}
4994
Joel Beckerf3868d02009-02-17 19:46:04 -08004995static int ocfs2_replace_extent_rec(handle_t *handle,
4996 struct ocfs2_extent_tree *et,
Tao Ma47be12e2009-01-09 07:32:48 +08004997 struct ocfs2_path *path,
4998 struct ocfs2_extent_list *el,
4999 int split_index,
5000 struct ocfs2_extent_rec *split_rec)
5001{
5002 int ret;
5003
Joel Beckerf3868d02009-02-17 19:46:04 -08005004 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Tao Ma47be12e2009-01-09 07:32:48 +08005005 path_num_items(path) - 1);
5006 if (ret) {
5007 mlog_errno(ret);
5008 goto out;
5009 }
5010
5011 el->l_recs[split_index] = *split_rec;
5012
5013 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5014out:
5015 return ret;
5016}
5017
Mark Fasheh328d5752007-06-18 10:48:04 -07005018/*
Tao Ma555936b2009-08-18 11:22:21 +08005019 * Split part or all of the extent record at split_index in the leaf
5020 * pointed to by path. Merge with the contiguous extent record if needed.
Mark Fasheh328d5752007-06-18 10:48:04 -07005021 *
5022 * Care is taken to handle contiguousness so as to not grow the tree.
5023 *
5024 * meta_ac is not strictly necessary - we only truly need it if growth
5025 * of the tree is required. All other cases will degrade into a less
5026 * optimal tree layout.
5027 *
Tao Mae7d4cb62008-08-18 17:38:44 +08005028 * last_eb_bh should be the rightmost leaf block for any extent
5029 * btree. Since a split may grow the tree or a merge might shrink it,
5030 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07005031 *
5032 * This code is optimized for readability - several passes might be
5033 * made over certain portions of the tree. All of those blocks will
5034 * have been brought into cache (and pinned via the journal), so the
5035 * extra overhead is not expressed in terms of disk reads.
5036 */
Tao Mae2e9f602009-08-18 11:22:34 +08005037int ocfs2_split_extent(handle_t *handle,
5038 struct ocfs2_extent_tree *et,
5039 struct ocfs2_path *path,
5040 int split_index,
5041 struct ocfs2_extent_rec *split_rec,
5042 struct ocfs2_alloc_context *meta_ac,
5043 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005044{
5045 int ret = 0;
5046 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005047 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005048 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5049 struct ocfs2_merge_ctxt ctxt;
5050 struct ocfs2_extent_list *rightmost_el;
5051
Mark Fasheh328d5752007-06-18 10:48:04 -07005052 if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5053 ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5054 (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5055 ret = -EIO;
5056 mlog_errno(ret);
5057 goto out;
5058 }
5059
Xue jiufei9f99ad02015-06-24 16:55:07 -07005060 ret = ocfs2_figure_merge_contig_type(et, path, el,
5061 split_index,
5062 split_rec,
5063 &ctxt);
5064 if (ret) {
5065 mlog_errno(ret);
5066 goto out;
5067 }
Mark Fasheh328d5752007-06-18 10:48:04 -07005068
5069 /*
5070 * The core merge / split code wants to know how much room is
Joel Beckera1cf0762009-02-13 03:45:49 -08005071 * left in this allocation tree, so we pass the
Mark Fasheh328d5752007-06-18 10:48:04 -07005072 * rightmost extent list.
5073 */
5074 if (path->p_tree_depth) {
5075 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07005076
Joel Becker3d03a302009-02-12 17:49:26 -08005077 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005078 ocfs2_et_get_last_eb_blk(et),
5079 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005080 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08005081 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07005082 goto out;
5083 }
5084
5085 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fasheh328d5752007-06-18 10:48:04 -07005086 rightmost_el = &eb->h_list;
5087 } else
5088 rightmost_el = path_root_el(path);
5089
Mark Fasheh328d5752007-06-18 10:48:04 -07005090 if (rec->e_cpos == split_rec->e_cpos &&
5091 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5092 ctxt.c_split_covers_rec = 1;
5093 else
5094 ctxt.c_split_covers_rec = 0;
5095
5096 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5097
Tao Maa09d09b2011-02-22 08:21:10 +08005098 trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5099 ctxt.c_has_empty_extent,
5100 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005101
5102 if (ctxt.c_contig_type == CONTIG_NONE) {
5103 if (ctxt.c_split_covers_rec)
Joel Beckerf3868d02009-02-17 19:46:04 -08005104 ret = ocfs2_replace_extent_rec(handle, et, path, el,
Tao Ma47be12e2009-01-09 07:32:48 +08005105 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005106 else
Joel Beckerd2311292009-02-13 03:43:22 -08005107 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005108 &last_eb_bh, split_index,
5109 split_rec, meta_ac);
5110 if (ret)
5111 mlog_errno(ret);
5112 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005113 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005114 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005115 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005116 if (ret)
5117 mlog_errno(ret);
5118 }
5119
Mark Fasheh328d5752007-06-18 10:48:04 -07005120out:
5121 brelse(last_eb_bh);
5122 return ret;
5123}
5124
5125/*
Tao Ma555936b2009-08-18 11:22:21 +08005126 * Change the flags of the already-existing extent at cpos for len clusters.
5127 *
5128 * new_flags: the flags we want to set.
5129 * clear_flags: the flags we want to clear.
5130 * phys: the new physical offset we want this new extent starts from.
Mark Fasheh328d5752007-06-18 10:48:04 -07005131 *
5132 * If the existing extent is larger than the request, initiate a
5133 * split. An attempt will be made at merging with adjacent extents.
5134 *
5135 * The caller is responsible for passing down meta_ac if we'll need it.
5136 */
Tao Ma1aa75fe2009-08-18 11:28:39 +08005137int ocfs2_change_extent_flag(handle_t *handle,
5138 struct ocfs2_extent_tree *et,
5139 u32 cpos, u32 len, u32 phys,
5140 struct ocfs2_alloc_context *meta_ac,
5141 struct ocfs2_cached_dealloc_ctxt *dealloc,
5142 int new_flags, int clear_flags)
Mark Fasheh328d5752007-06-18 10:48:04 -07005143{
5144 int ret, index;
Tao Ma555936b2009-08-18 11:22:21 +08005145 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5146 u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
Mark Fasheh328d5752007-06-18 10:48:04 -07005147 struct ocfs2_extent_rec split_rec;
5148 struct ocfs2_path *left_path = NULL;
5149 struct ocfs2_extent_list *el;
Tao Ma555936b2009-08-18 11:22:21 +08005150 struct ocfs2_extent_rec *rec;
Mark Fasheh328d5752007-06-18 10:48:04 -07005151
Joel Beckerffdd7a52008-10-17 22:32:01 -07005152 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005153 if (!left_path) {
5154 ret = -ENOMEM;
5155 mlog_errno(ret);
5156 goto out;
5157 }
5158
Joel Beckerfacdb772009-02-12 18:08:48 -08005159 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005160 if (ret) {
5161 mlog_errno(ret);
5162 goto out;
5163 }
5164 el = path_leaf_el(left_path);
5165
5166 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005167 if (index == -1) {
Tao Ma555936b2009-08-18 11:22:21 +08005168 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07005169 "Owner %llu has an extent at cpos %u which can no longer be found\n",
5170 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5171 cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005172 ret = -EROFS;
5173 goto out;
5174 }
5175
Tao Ma555936b2009-08-18 11:22:21 +08005176 ret = -EIO;
5177 rec = &el->l_recs[index];
5178 if (new_flags && (rec->e_flags & new_flags)) {
5179 mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
5180 "extent that already had them",
5181 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5182 new_flags);
5183 goto out;
5184 }
5185
5186 if (clear_flags && !(rec->e_flags & clear_flags)) {
5187 mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
5188 "extent that didn't have them",
5189 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5190 clear_flags);
5191 goto out;
5192 }
5193
Mark Fasheh328d5752007-06-18 10:48:04 -07005194 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5195 split_rec.e_cpos = cpu_to_le32(cpos);
5196 split_rec.e_leaf_clusters = cpu_to_le16(len);
5197 split_rec.e_blkno = cpu_to_le64(start_blkno);
Tao Ma555936b2009-08-18 11:22:21 +08005198 split_rec.e_flags = rec->e_flags;
5199 if (new_flags)
5200 split_rec.e_flags |= new_flags;
5201 if (clear_flags)
5202 split_rec.e_flags &= ~clear_flags;
Mark Fasheh328d5752007-06-18 10:48:04 -07005203
Tao Mae2e9f602009-08-18 11:22:34 +08005204 ret = ocfs2_split_extent(handle, et, left_path,
5205 index, &split_rec, meta_ac,
5206 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005207 if (ret)
5208 mlog_errno(ret);
5209
5210out:
5211 ocfs2_free_path(left_path);
5212 return ret;
Tao Ma555936b2009-08-18 11:22:21 +08005213
5214}
5215
5216/*
5217 * Mark the already-existing extent at cpos as written for len clusters.
5218 * This removes the unwritten extent flag.
5219 *
5220 * If the existing extent is larger than the request, initiate a
5221 * split. An attempt will be made at merging with adjacent extents.
5222 *
5223 * The caller is responsible for passing down meta_ac if we'll need it.
5224 */
5225int ocfs2_mark_extent_written(struct inode *inode,
5226 struct ocfs2_extent_tree *et,
5227 handle_t *handle, u32 cpos, u32 len, u32 phys,
5228 struct ocfs2_alloc_context *meta_ac,
5229 struct ocfs2_cached_dealloc_ctxt *dealloc)
5230{
5231 int ret;
5232
Tao Maa09d09b2011-02-22 08:21:10 +08005233 trace_ocfs2_mark_extent_written(
5234 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5235 cpos, len, phys);
Tao Ma555936b2009-08-18 11:22:21 +08005236
5237 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
Joe Perches7ecef142015-09-04 15:44:51 -07005238 ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents that are being written to, but the feature bit is not set in the super block\n",
Tao Ma555936b2009-08-18 11:22:21 +08005239 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5240 ret = -EROFS;
5241 goto out;
5242 }
5243
5244 /*
5245 * XXX: This should be fixed up so that we just re-insert the
5246 * next extent records.
5247 */
5248 ocfs2_et_extent_map_truncate(et, 0);
5249
5250 ret = ocfs2_change_extent_flag(handle, et, cpos,
5251 len, phys, meta_ac, dealloc,
5252 0, OCFS2_EXT_UNWRITTEN);
5253 if (ret)
5254 mlog_errno(ret);
5255
5256out:
5257 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07005258}
5259
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005260static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5261 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005262 int index, u32 new_range,
5263 struct ocfs2_alloc_context *meta_ac)
5264{
Tao Mac901fb02010-04-26 14:34:57 +08005265 int ret, depth, credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005266 struct buffer_head *last_eb_bh = NULL;
5267 struct ocfs2_extent_block *eb;
5268 struct ocfs2_extent_list *rightmost_el, *el;
5269 struct ocfs2_extent_rec split_rec;
5270 struct ocfs2_extent_rec *rec;
5271 struct ocfs2_insert_type insert;
5272
5273 /*
5274 * Setup the record to split before we grow the tree.
5275 */
5276 el = path_leaf_el(path);
5277 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005278 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5279 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005280
5281 depth = path->p_tree_depth;
5282 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005283 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005284 ocfs2_et_get_last_eb_blk(et),
5285 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005286 if (ret < 0) {
5287 mlog_errno(ret);
5288 goto out;
5289 }
5290
5291 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5292 rightmost_el = &eb->h_list;
5293 } else
5294 rightmost_el = path_leaf_el(path);
5295
Tao Mac901fb02010-04-26 14:34:57 +08005296 credits = path->p_tree_depth +
5297 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005298 ret = ocfs2_extend_trans(handle, credits);
5299 if (ret) {
5300 mlog_errno(ret);
5301 goto out;
5302 }
5303
5304 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5305 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005306 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005307 meta_ac);
5308 if (ret) {
5309 mlog_errno(ret);
5310 goto out;
5311 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005312 }
5313
5314 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5315 insert.ins_appending = APPEND_NONE;
5316 insert.ins_contig = CONTIG_NONE;
5317 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005318 insert.ins_tree_depth = depth;
5319
Joel Becker3505bec2009-02-13 02:57:58 -08005320 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005321 if (ret)
5322 mlog_errno(ret);
5323
5324out:
5325 brelse(last_eb_bh);
5326 return ret;
5327}
5328
Joel Becker043beeb2009-02-13 02:42:30 -08005329static int ocfs2_truncate_rec(handle_t *handle,
5330 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005331 struct ocfs2_path *path, int index,
5332 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005333 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005334{
5335 int ret;
5336 u32 left_cpos, rec_range, trunc_range;
5337 int wants_rotate = 0, is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005338 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005339 struct ocfs2_path *left_path = NULL;
5340 struct ocfs2_extent_list *el = path_leaf_el(path);
5341 struct ocfs2_extent_rec *rec;
5342 struct ocfs2_extent_block *eb;
5343
5344 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Joel Becker70f18c02009-02-13 02:09:31 -08005345 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005346 if (ret) {
5347 mlog_errno(ret);
5348 goto out;
5349 }
5350
5351 index--;
5352 }
5353
5354 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5355 path->p_tree_depth) {
5356 /*
5357 * Check whether this is the rightmost tree record. If
5358 * we remove all of this record or part of its right
5359 * edge then an update of the record lengths above it
5360 * will be required.
5361 */
5362 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5363 if (eb->h_next_leaf_blk == 0)
5364 is_rightmost_tree_rec = 1;
5365 }
5366
5367 rec = &el->l_recs[index];
5368 if (index == 0 && path->p_tree_depth &&
5369 le32_to_cpu(rec->e_cpos) == cpos) {
5370 /*
5371 * Changing the leftmost offset (via partial or whole
5372 * record truncate) of an interior (or rightmost) path
5373 * means we have to update the subtree that is formed
5374 * by this leaf and the one to it's left.
5375 *
5376 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005377 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005378 * 2) The leaf is rightmost and will be empty after
5379 * we remove the extent record - the rotate code
5380 * knows how to update the newly formed edge.
5381 */
5382
Joel Becker043beeb2009-02-13 02:42:30 -08005383 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005384 if (ret) {
5385 mlog_errno(ret);
5386 goto out;
5387 }
5388
5389 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005390 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005391 if (!left_path) {
5392 ret = -ENOMEM;
5393 mlog_errno(ret);
5394 goto out;
5395 }
5396
Joel Beckerfacdb772009-02-12 18:08:48 -08005397 ret = ocfs2_find_path(et->et_ci, left_path,
5398 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005399 if (ret) {
5400 mlog_errno(ret);
5401 goto out;
5402 }
5403 }
5404 }
5405
5406 ret = ocfs2_extend_rotate_transaction(handle, 0,
5407 handle->h_buffer_credits,
5408 path);
5409 if (ret) {
5410 mlog_errno(ret);
5411 goto out;
5412 }
5413
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005414 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005415 if (ret) {
5416 mlog_errno(ret);
5417 goto out;
5418 }
5419
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005420 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005421 if (ret) {
5422 mlog_errno(ret);
5423 goto out;
5424 }
5425
5426 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5427 trunc_range = cpos + len;
5428
5429 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5430 int next_free;
5431
5432 memset(rec, 0, sizeof(*rec));
5433 ocfs2_cleanup_merge(el, index);
5434 wants_rotate = 1;
5435
5436 next_free = le16_to_cpu(el->l_next_free_rec);
5437 if (is_rightmost_tree_rec && next_free > 1) {
5438 /*
5439 * We skip the edge update if this path will
5440 * be deleted by the rotate code.
5441 */
5442 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005443 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005444 rec);
5445 }
5446 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5447 /* Remove leftmost portion of the record. */
5448 le32_add_cpu(&rec->e_cpos, len);
5449 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5450 le16_add_cpu(&rec->e_leaf_clusters, -len);
5451 } else if (rec_range == trunc_range) {
5452 /* Remove rightmost portion of the record */
5453 le16_add_cpu(&rec->e_leaf_clusters, -len);
5454 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005455 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005456 } else {
5457 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005458 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5459 "(%u, %u)\n",
5460 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005461 le32_to_cpu(rec->e_cpos),
5462 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5463 BUG();
5464 }
5465
5466 if (left_path) {
5467 int subtree_index;
5468
Joel Becker7dc02802009-02-12 19:20:13 -08005469 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005470 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005471 subtree_index);
5472 }
5473
5474 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5475
Joel Becker70f18c02009-02-13 02:09:31 -08005476 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005477 if (ret) {
5478 mlog_errno(ret);
5479 goto out;
5480 }
5481
5482out:
5483 ocfs2_free_path(left_path);
5484 return ret;
5485}
5486
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005487int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005488 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005489 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005490 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005491 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005492{
5493 int ret, index;
5494 u32 rec_range, trunc_range;
5495 struct ocfs2_extent_rec *rec;
5496 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005497 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005498
Joel Becker4c911ee2009-02-13 02:50:12 -08005499 /*
5500 * XXX: Why are we truncating to 0 instead of wherever this
5501 * affects us?
5502 */
5503 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005504
Joel Beckerffdd7a52008-10-17 22:32:01 -07005505 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005506 if (!path) {
5507 ret = -ENOMEM;
5508 mlog_errno(ret);
5509 goto out;
5510 }
5511
Joel Beckerfacdb772009-02-12 18:08:48 -08005512 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005513 if (ret) {
5514 mlog_errno(ret);
5515 goto out;
5516 }
5517
5518 el = path_leaf_el(path);
5519 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005520 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005521 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005522 "Owner %llu has an extent at cpos %u which can no longer be found\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005523 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5524 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005525 ret = -EROFS;
5526 goto out;
5527 }
5528
5529 /*
5530 * We have 3 cases of extent removal:
5531 * 1) Range covers the entire extent rec
5532 * 2) Range begins or ends on one edge of the extent rec
5533 * 3) Range is in the middle of the extent rec (no shared edges)
5534 *
5535 * For case 1 we remove the extent rec and left rotate to
5536 * fill the hole.
5537 *
5538 * For case 2 we just shrink the existing extent rec, with a
5539 * tree update if the shrinking edge is also the edge of an
5540 * extent block.
5541 *
5542 * For case 3 we do a right split to turn the extent rec into
5543 * something case 2 can handle.
5544 */
5545 rec = &el->l_recs[index];
5546 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5547 trunc_range = cpos + len;
5548
5549 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5550
Tao Maa09d09b2011-02-22 08:21:10 +08005551 trace_ocfs2_remove_extent(
5552 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5553 cpos, len, index, le32_to_cpu(rec->e_cpos),
5554 ocfs2_rec_clusters(el, rec));
Mark Fashehd0c7d702007-07-03 13:27:22 -07005555
5556 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005557 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5558 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005559 if (ret) {
5560 mlog_errno(ret);
5561 goto out;
5562 }
5563 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005564 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005565 trunc_range, meta_ac);
5566 if (ret) {
5567 mlog_errno(ret);
5568 goto out;
5569 }
5570
5571 /*
5572 * The split could have manipulated the tree enough to
5573 * move the record location, so we have to look for it again.
5574 */
5575 ocfs2_reinit_path(path, 1);
5576
Joel Beckerfacdb772009-02-12 18:08:48 -08005577 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005578 if (ret) {
5579 mlog_errno(ret);
5580 goto out;
5581 }
5582
5583 el = path_leaf_el(path);
5584 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005585 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005586 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005587 "Owner %llu: split at cpos %u lost record\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005588 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005589 cpos);
5590 ret = -EROFS;
5591 goto out;
5592 }
5593
5594 /*
5595 * Double check our values here. If anything is fishy,
5596 * it's easier to catch it at the top level.
5597 */
5598 rec = &el->l_recs[index];
5599 rec_range = le32_to_cpu(rec->e_cpos) +
5600 ocfs2_rec_clusters(el, rec);
5601 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005602 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005603 "Owner %llu: error after split at cpos %u trunc len %u, existing record is (%u,%u)\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005604 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005605 cpos, len, le32_to_cpu(rec->e_cpos),
5606 ocfs2_rec_clusters(el, rec));
5607 ret = -EROFS;
5608 goto out;
5609 }
5610
Joel Becker043beeb2009-02-13 02:42:30 -08005611 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5612 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005613 if (ret) {
5614 mlog_errno(ret);
5615 goto out;
5616 }
5617 }
5618
5619out:
5620 ocfs2_free_path(path);
5621 return ret;
5622}
5623
Tristan Ye78f94672010-05-11 17:54:42 +08005624/*
5625 * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5626 * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5627 * number to reserve some extra blocks, and it only handles meta
5628 * data allocations.
5629 *
5630 * Currently, only ocfs2_remove_btree_range() uses it for truncating
5631 * and punching holes.
5632 */
5633static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5634 struct ocfs2_extent_tree *et,
5635 u32 extents_to_split,
5636 struct ocfs2_alloc_context **ac,
5637 int extra_blocks)
5638{
5639 int ret = 0, num_free_extents;
5640 unsigned int max_recs_needed = 2 * extents_to_split;
5641 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5642
5643 *ac = NULL;
5644
5645 num_free_extents = ocfs2_num_free_extents(osb, et);
5646 if (num_free_extents < 0) {
5647 ret = num_free_extents;
5648 mlog_errno(ret);
5649 goto out;
5650 }
5651
5652 if (!num_free_extents ||
5653 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5654 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5655
5656 if (extra_blocks) {
5657 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5658 if (ret < 0) {
5659 if (ret != -ENOSPC)
5660 mlog_errno(ret);
5661 goto out;
5662 }
5663 }
5664
5665out:
5666 if (ret) {
5667 if (*ac) {
5668 ocfs2_free_alloc_context(*ac);
5669 *ac = NULL;
5670 }
5671 }
5672
5673 return ret;
5674}
5675
Mark Fashehfecc0112008-11-12 15:16:38 -08005676int ocfs2_remove_btree_range(struct inode *inode,
5677 struct ocfs2_extent_tree *et,
Tristan Ye78f94672010-05-11 17:54:42 +08005678 u32 cpos, u32 phys_cpos, u32 len, int flags,
5679 struct ocfs2_cached_dealloc_ctxt *dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08005680 u64 refcount_loc, bool refcount_tree_locked)
Mark Fashehfecc0112008-11-12 15:16:38 -08005681{
Tristan Ye78f94672010-05-11 17:54:42 +08005682 int ret, credits = 0, extra_blocks = 0;
Mark Fashehfecc0112008-11-12 15:16:38 -08005683 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5684 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5685 struct inode *tl_inode = osb->osb_tl_inode;
5686 handle_t *handle;
5687 struct ocfs2_alloc_context *meta_ac = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08005688 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehfecc0112008-11-12 15:16:38 -08005689
Tristan Ye78f94672010-05-11 17:54:42 +08005690 if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
5691 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
5692 OCFS2_HAS_REFCOUNT_FL));
5693
Junxiao Bif62f12b2014-12-18 16:17:32 -08005694 if (!refcount_tree_locked) {
5695 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5696 &ref_tree, NULL);
5697 if (ret) {
5698 mlog_errno(ret);
5699 goto bail;
5700 }
Tristan Ye78f94672010-05-11 17:54:42 +08005701 }
5702
5703 ret = ocfs2_prepare_refcount_change_for_del(inode,
5704 refcount_loc,
5705 phys_blkno,
5706 len,
5707 &credits,
5708 &extra_blocks);
5709 if (ret < 0) {
5710 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005711 goto bail;
Tristan Ye78f94672010-05-11 17:54:42 +08005712 }
5713 }
5714
5715 ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5716 extra_blocks);
Mark Fashehfecc0112008-11-12 15:16:38 -08005717 if (ret) {
5718 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005719 goto bail;
Mark Fashehfecc0112008-11-12 15:16:38 -08005720 }
5721
5722 mutex_lock(&tl_inode->i_mutex);
5723
5724 if (ocfs2_truncate_log_needs_flush(osb)) {
5725 ret = __ocfs2_flush_truncate_log(osb);
5726 if (ret < 0) {
5727 mlog_errno(ret);
5728 goto out;
5729 }
5730 }
5731
Tristan Ye78f94672010-05-11 17:54:42 +08005732 handle = ocfs2_start_trans(osb,
5733 ocfs2_remove_extent_credits(osb->sb) + credits);
Mark Fashehfecc0112008-11-12 15:16:38 -08005734 if (IS_ERR(handle)) {
5735 ret = PTR_ERR(handle);
5736 mlog_errno(ret);
5737 goto out;
5738 }
5739
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005740 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005741 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005742 if (ret) {
5743 mlog_errno(ret);
Wengang Wangb8a0ae52011-10-12 15:22:15 +08005744 goto out_commit;
Mark Fashehfecc0112008-11-12 15:16:38 -08005745 }
5746
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005747 dquot_free_space_nodirty(inode,
Mark Fashehfd4ef232009-01-29 15:06:21 -08005748 ocfs2_clusters_to_bytes(inode->i_sb, len));
5749
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005750 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005751 if (ret) {
5752 mlog_errno(ret);
5753 goto out_commit;
5754 }
5755
Joel Becker6136ca52009-02-12 19:32:43 -08005756 ocfs2_et_update_clusters(et, -len);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07005757 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fashehfecc0112008-11-12 15:16:38 -08005758
Joel Beckerec20cec2010-03-19 14:13:52 -07005759 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehfecc0112008-11-12 15:16:38 -08005760
Tristan Ye78f94672010-05-11 17:54:42 +08005761 if (phys_blkno) {
5762 if (flags & OCFS2_EXT_REFCOUNTED)
5763 ret = ocfs2_decrease_refcount(inode, handle,
5764 ocfs2_blocks_to_clusters(osb->sb,
5765 phys_blkno),
5766 len, meta_ac,
5767 dealloc, 1);
5768 else
5769 ret = ocfs2_truncate_log_append(osb, handle,
5770 phys_blkno, len);
5771 if (ret)
5772 mlog_errno(ret);
5773
5774 }
Mark Fashehfecc0112008-11-12 15:16:38 -08005775
5776out_commit:
5777 ocfs2_commit_trans(osb, handle);
5778out:
5779 mutex_unlock(&tl_inode->i_mutex);
Joseph Qi33add0e2013-07-03 15:00:52 -07005780bail:
Mark Fashehfecc0112008-11-12 15:16:38 -08005781 if (meta_ac)
5782 ocfs2_free_alloc_context(meta_ac);
5783
Tristan Ye78f94672010-05-11 17:54:42 +08005784 if (ref_tree)
5785 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5786
Mark Fashehfecc0112008-11-12 15:16:38 -08005787 return ret;
5788}
5789
Mark Fasheh063c4562007-07-03 13:34:11 -07005790int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005791{
5792 struct buffer_head *tl_bh = osb->osb_tl_bh;
5793 struct ocfs2_dinode *di;
5794 struct ocfs2_truncate_log *tl;
5795
5796 di = (struct ocfs2_dinode *) tl_bh->b_data;
5797 tl = &di->id2.i_dealloc;
5798
5799 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5800 "slot %d, invalid truncate log parameters: used = "
5801 "%u, count = %u\n", osb->slot_num,
5802 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5803 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5804}
5805
5806static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5807 unsigned int new_start)
5808{
5809 unsigned int tail_index;
5810 unsigned int current_tail;
5811
5812 /* No records, nothing to coalesce */
5813 if (!le16_to_cpu(tl->tl_used))
5814 return 0;
5815
5816 tail_index = le16_to_cpu(tl->tl_used) - 1;
5817 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5818 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5819
5820 return current_tail == new_start;
5821}
5822
Mark Fasheh063c4562007-07-03 13:34:11 -07005823int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5824 handle_t *handle,
5825 u64 start_blk,
5826 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005827{
5828 int status, index;
5829 unsigned int start_cluster, tl_count;
5830 struct inode *tl_inode = osb->osb_tl_inode;
5831 struct buffer_head *tl_bh = osb->osb_tl_bh;
5832 struct ocfs2_dinode *di;
5833 struct ocfs2_truncate_log *tl;
5834
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005835 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005836
5837 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5838
5839 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005840
Joel Becker10995aa2008-11-13 14:49:12 -08005841 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5842 * by the underlying call to ocfs2_read_inode_block(), so any
5843 * corruption is a code bug */
5844 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5845
5846 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005847 tl_count = le16_to_cpu(tl->tl_count);
5848 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5849 tl_count == 0,
Mark Fashehb06970532006-03-03 10:24:33 -08005850 "Truncate record count on #%llu invalid "
5851 "wanted %u, actual %u\n",
5852 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005853 ocfs2_truncate_recs_per_inode(osb->sb),
5854 le16_to_cpu(tl->tl_count));
5855
5856 /* Caller should have known to flush before calling us. */
5857 index = le16_to_cpu(tl->tl_used);
5858 if (index >= tl_count) {
5859 status = -ENOSPC;
5860 mlog_errno(status);
5861 goto bail;
5862 }
5863
Joel Becker0cf2f762009-02-12 16:41:25 -08005864 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005865 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005866 if (status < 0) {
5867 mlog_errno(status);
5868 goto bail;
5869 }
5870
Tao Maa09d09b2011-02-22 08:21:10 +08005871 trace_ocfs2_truncate_log_append(
5872 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5873 start_cluster, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005874 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5875 /*
5876 * Move index back to the record we are coalescing with.
5877 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5878 */
5879 index--;
5880
5881 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
Tao Maa09d09b2011-02-22 08:21:10 +08005882 trace_ocfs2_truncate_log_append(
5883 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5884 index, le32_to_cpu(tl->tl_recs[index].t_start),
5885 num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005886 } else {
5887 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5888 tl->tl_used = cpu_to_le16(index + 1);
5889 }
5890 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5891
Joel Beckerec20cec2010-03-19 14:13:52 -07005892 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005893
Tao Ma50308d82010-11-04 15:14:11 +08005894 osb->truncated_clusters += num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08005895bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005896 return status;
5897}
5898
5899static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07005900 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08005901 struct inode *data_alloc_inode,
5902 struct buffer_head *data_alloc_bh)
5903{
5904 int status = 0;
5905 int i;
5906 unsigned int num_clusters;
5907 u64 start_blk;
5908 struct ocfs2_truncate_rec rec;
5909 struct ocfs2_dinode *di;
5910 struct ocfs2_truncate_log *tl;
5911 struct inode *tl_inode = osb->osb_tl_inode;
5912 struct buffer_head *tl_bh = osb->osb_tl_bh;
5913
Mark Fashehccd979b2005-12-15 14:31:24 -08005914 di = (struct ocfs2_dinode *) tl_bh->b_data;
5915 tl = &di->id2.i_dealloc;
5916 i = le16_to_cpu(tl->tl_used) - 1;
5917 while (i >= 0) {
5918 /* Caller has given us at least enough credits to
5919 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005920 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005921 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005922 if (status < 0) {
5923 mlog_errno(status);
5924 goto bail;
5925 }
5926
5927 tl->tl_used = cpu_to_le16(i);
5928
Joel Beckerec20cec2010-03-19 14:13:52 -07005929 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005930
5931 /* TODO: Perhaps we can calculate the bulk of the
5932 * credits up front rather than extending like
5933 * this. */
5934 status = ocfs2_extend_trans(handle,
5935 OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5936 if (status < 0) {
5937 mlog_errno(status);
5938 goto bail;
5939 }
5940
5941 rec = tl->tl_recs[i];
5942 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5943 le32_to_cpu(rec.t_start));
5944 num_clusters = le32_to_cpu(rec.t_clusters);
5945
5946 /* if start_blk is not set, we ignore the record as
5947 * invalid. */
5948 if (start_blk) {
Tao Maa09d09b2011-02-22 08:21:10 +08005949 trace_ocfs2_replay_truncate_records(
5950 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5951 i, le32_to_cpu(rec.t_start), num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005952
5953 status = ocfs2_free_clusters(handle, data_alloc_inode,
5954 data_alloc_bh, start_blk,
5955 num_clusters);
5956 if (status < 0) {
5957 mlog_errno(status);
5958 goto bail;
5959 }
5960 }
5961 i--;
5962 }
5963
Tao Ma50308d82010-11-04 15:14:11 +08005964 osb->truncated_clusters = 0;
5965
Mark Fashehccd979b2005-12-15 14:31:24 -08005966bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005967 return status;
5968}
5969
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005970/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005971int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005972{
5973 int status;
5974 unsigned int num_to_flush;
Mark Fasheh1fabe142006-10-09 18:11:45 -07005975 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005976 struct inode *tl_inode = osb->osb_tl_inode;
5977 struct inode *data_alloc_inode = NULL;
5978 struct buffer_head *tl_bh = osb->osb_tl_bh;
5979 struct buffer_head *data_alloc_bh = NULL;
5980 struct ocfs2_dinode *di;
5981 struct ocfs2_truncate_log *tl;
5982
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005983 BUG_ON(mutex_trylock(&tl_inode->i_mutex));
Mark Fashehccd979b2005-12-15 14:31:24 -08005984
5985 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005986
Joel Becker10995aa2008-11-13 14:49:12 -08005987 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5988 * by the underlying call to ocfs2_read_inode_block(), so any
5989 * corruption is a code bug */
5990 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5991
5992 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005993 num_to_flush = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08005994 trace_ocfs2_flush_truncate_log(
5995 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5996 num_to_flush);
Mark Fashehccd979b2005-12-15 14:31:24 -08005997 if (!num_to_flush) {
5998 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07005999 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006000 }
6001
6002 data_alloc_inode = ocfs2_get_system_file_inode(osb,
6003 GLOBAL_BITMAP_SYSTEM_INODE,
6004 OCFS2_INVALID_SLOT);
6005 if (!data_alloc_inode) {
6006 status = -EINVAL;
6007 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006008 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006009 }
6010
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006011 mutex_lock(&data_alloc_inode->i_mutex);
6012
Mark Fashehe63aecb62007-10-18 15:30:42 -07006013 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006014 if (status < 0) {
6015 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006016 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08006017 }
6018
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006019 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006020 if (IS_ERR(handle)) {
6021 status = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006022 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006023 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08006024 }
6025
6026 status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
6027 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006028 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08006029 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006030
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006031 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006032
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006033out_unlock:
6034 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07006035 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006036
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006037out_mutex:
6038 mutex_unlock(&data_alloc_inode->i_mutex);
6039 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006040
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006041out:
Mark Fashehccd979b2005-12-15 14:31:24 -08006042 return status;
6043}
6044
6045int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6046{
6047 int status;
6048 struct inode *tl_inode = osb->osb_tl_inode;
6049
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006050 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006051 status = __ocfs2_flush_truncate_log(osb);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006052 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006053
6054 return status;
6055}
6056
David Howellsc4028952006-11-22 14:57:56 +00006057static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08006058{
6059 int status;
David Howellsc4028952006-11-22 14:57:56 +00006060 struct ocfs2_super *osb =
6061 container_of(work, struct ocfs2_super,
6062 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08006063
Mark Fashehccd979b2005-12-15 14:31:24 -08006064 status = ocfs2_flush_truncate_log(osb);
6065 if (status < 0)
6066 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08006067 else
Tiger Yangb89c5422010-01-25 14:11:06 +08006068 ocfs2_init_steal_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08006069}
6070
6071#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6072void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6073 int cancel)
6074{
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006075 if (osb->osb_tl_inode &&
6076 atomic_read(&osb->osb_tl_disable) == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006077 /* We want to push off log flushes while truncates are
6078 * still running. */
6079 if (cancel)
6080 cancel_delayed_work(&osb->osb_truncate_log_wq);
6081
6082 queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
6083 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6084 }
6085}
6086
6087static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6088 int slot_num,
6089 struct inode **tl_inode,
6090 struct buffer_head **tl_bh)
6091{
6092 int status;
6093 struct inode *inode = NULL;
6094 struct buffer_head *bh = NULL;
6095
6096 inode = ocfs2_get_system_file_inode(osb,
6097 TRUNCATE_LOG_SYSTEM_INODE,
6098 slot_num);
6099 if (!inode) {
6100 status = -EINVAL;
6101 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6102 goto bail;
6103 }
6104
Joel Beckerb657c952008-11-13 14:49:11 -08006105 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006106 if (status < 0) {
6107 iput(inode);
6108 mlog_errno(status);
6109 goto bail;
6110 }
6111
6112 *tl_inode = inode;
6113 *tl_bh = bh;
6114bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08006115 return status;
6116}
6117
6118/* called during the 1st stage of node recovery. we stamp a clean
6119 * truncate log and pass back a copy for processing later. if the
6120 * truncate log does not require processing, a *tl_copy is set to
6121 * NULL. */
6122int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6123 int slot_num,
6124 struct ocfs2_dinode **tl_copy)
6125{
6126 int status;
6127 struct inode *tl_inode = NULL;
6128 struct buffer_head *tl_bh = NULL;
6129 struct ocfs2_dinode *di;
6130 struct ocfs2_truncate_log *tl;
6131
6132 *tl_copy = NULL;
6133
Tao Maa09d09b2011-02-22 08:21:10 +08006134 trace_ocfs2_begin_truncate_log_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08006135
6136 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6137 if (status < 0) {
6138 mlog_errno(status);
6139 goto bail;
6140 }
6141
6142 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006143
Joel Becker10995aa2008-11-13 14:49:12 -08006144 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6145 * validated by the underlying call to ocfs2_read_inode_block(),
6146 * so any corruption is a code bug */
6147 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6148
6149 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006150 if (le16_to_cpu(tl->tl_used)) {
Tao Maa09d09b2011-02-22 08:21:10 +08006151 trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
Mark Fashehccd979b2005-12-15 14:31:24 -08006152
6153 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6154 if (!(*tl_copy)) {
6155 status = -ENOMEM;
6156 mlog_errno(status);
6157 goto bail;
6158 }
6159
6160 /* Assuming the write-out below goes well, this copy
6161 * will be passed back to recovery for processing. */
6162 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6163
6164 /* All we need to do to clear the truncate log is set
6165 * tl_used. */
6166 tl->tl_used = 0;
6167
Joel Becker13723d02008-10-17 19:25:01 -07006168 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006169 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006170 if (status < 0) {
6171 mlog_errno(status);
6172 goto bail;
6173 }
6174 }
6175
6176bail:
Joseph Qi72865d92016-01-14 15:17:27 -08006177 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006178 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006179
Joseph Qi46359292015-09-04 15:44:54 -07006180 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006181 kfree(*tl_copy);
6182 *tl_copy = NULL;
Tao Mac1e8d352011-03-07 16:43:21 +08006183 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006184 }
6185
Mark Fashehccd979b2005-12-15 14:31:24 -08006186 return status;
6187}
6188
6189int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6190 struct ocfs2_dinode *tl_copy)
6191{
6192 int status = 0;
6193 int i;
6194 unsigned int clusters, num_recs, start_cluster;
6195 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006196 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006197 struct inode *tl_inode = osb->osb_tl_inode;
6198 struct ocfs2_truncate_log *tl;
6199
Mark Fashehccd979b2005-12-15 14:31:24 -08006200 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6201 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6202 return -EINVAL;
6203 }
6204
6205 tl = &tl_copy->id2.i_dealloc;
6206 num_recs = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006207 trace_ocfs2_complete_truncate_log_recovery(
6208 (unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6209 num_recs);
Mark Fashehccd979b2005-12-15 14:31:24 -08006210
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006211 mutex_lock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006212 for(i = 0; i < num_recs; i++) {
6213 if (ocfs2_truncate_log_needs_flush(osb)) {
6214 status = __ocfs2_flush_truncate_log(osb);
6215 if (status < 0) {
6216 mlog_errno(status);
6217 goto bail_up;
6218 }
6219 }
6220
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006221 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006222 if (IS_ERR(handle)) {
6223 status = PTR_ERR(handle);
6224 mlog_errno(status);
6225 goto bail_up;
6226 }
6227
6228 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6229 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6230 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6231
6232 status = ocfs2_truncate_log_append(osb, handle,
6233 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006234 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006235 if (status < 0) {
6236 mlog_errno(status);
6237 goto bail_up;
6238 }
6239 }
6240
6241bail_up:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08006242 mutex_unlock(&tl_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08006243
Mark Fashehccd979b2005-12-15 14:31:24 -08006244 return status;
6245}
6246
6247void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6248{
6249 int status;
6250 struct inode *tl_inode = osb->osb_tl_inode;
6251
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006252 atomic_set(&osb->osb_tl_disable, 1);
6253
Mark Fashehccd979b2005-12-15 14:31:24 -08006254 if (tl_inode) {
6255 cancel_delayed_work(&osb->osb_truncate_log_wq);
6256 flush_workqueue(ocfs2_wq);
6257
6258 status = ocfs2_flush_truncate_log(osb);
6259 if (status < 0)
6260 mlog_errno(status);
6261
6262 brelse(osb->osb_tl_bh);
6263 iput(osb->osb_tl_inode);
6264 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006265}
6266
6267int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6268{
6269 int status;
6270 struct inode *tl_inode = NULL;
6271 struct buffer_head *tl_bh = NULL;
6272
Mark Fashehccd979b2005-12-15 14:31:24 -08006273 status = ocfs2_get_truncate_log_info(osb,
6274 osb->slot_num,
6275 &tl_inode,
6276 &tl_bh);
6277 if (status < 0)
6278 mlog_errno(status);
6279
6280 /* ocfs2_truncate_log_shutdown keys on the existence of
6281 * osb->osb_tl_inode so we don't set any of the osb variables
6282 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006283 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6284 ocfs2_truncate_log_worker);
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006285 atomic_set(&osb->osb_tl_disable, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08006286 osb->osb_tl_bh = tl_bh;
6287 osb->osb_tl_inode = tl_inode;
6288
Mark Fashehccd979b2005-12-15 14:31:24 -08006289 return status;
6290}
6291
Mark Fasheh2b604352007-06-22 15:45:27 -07006292/*
6293 * Delayed de-allocation of suballocator blocks.
6294 *
6295 * Some sets of block de-allocations might involve multiple suballocator inodes.
6296 *
6297 * The locking for this can get extremely complicated, especially when
6298 * the suballocator inodes to delete from aren't known until deep
6299 * within an unrelated codepath.
6300 *
6301 * ocfs2_extent_block structures are a good example of this - an inode
6302 * btree could have been grown by any number of nodes each allocating
6303 * out of their own suballoc inode.
6304 *
6305 * These structures allow the delay of block de-allocation until a
6306 * later time, when locking of multiple cluster inodes won't cause
6307 * deadlock.
6308 */
6309
6310/*
Tao Ma2891d292008-11-12 08:26:58 +08006311 * Describe a single bit freed from a suballocator. For the block
6312 * suballocators, it represents one block. For the global cluster
6313 * allocator, it represents some clusters and free_bit indicates
6314 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006315 */
6316struct ocfs2_cached_block_free {
6317 struct ocfs2_cached_block_free *free_next;
Tao Ma74380c42010-03-22 14:20:18 +08006318 u64 free_bg;
Mark Fasheh2b604352007-06-22 15:45:27 -07006319 u64 free_blk;
6320 unsigned int free_bit;
6321};
6322
6323struct ocfs2_per_slot_free_list {
6324 struct ocfs2_per_slot_free_list *f_next_suballocator;
6325 int f_inode_type;
6326 int f_slot;
6327 struct ocfs2_cached_block_free *f_first;
6328};
6329
Tao Ma2891d292008-11-12 08:26:58 +08006330static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6331 int sysfile_type,
6332 int slot,
6333 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006334{
6335 int ret;
6336 u64 bg_blkno;
6337 handle_t *handle;
6338 struct inode *inode;
6339 struct buffer_head *di_bh = NULL;
6340 struct ocfs2_cached_block_free *tmp;
6341
6342 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6343 if (!inode) {
6344 ret = -EINVAL;
6345 mlog_errno(ret);
6346 goto out;
6347 }
6348
6349 mutex_lock(&inode->i_mutex);
6350
Mark Fashehe63aecb62007-10-18 15:30:42 -07006351 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006352 if (ret) {
6353 mlog_errno(ret);
6354 goto out_mutex;
6355 }
6356
6357 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6358 if (IS_ERR(handle)) {
6359 ret = PTR_ERR(handle);
6360 mlog_errno(ret);
6361 goto out_unlock;
6362 }
6363
6364 while (head) {
Tao Ma74380c42010-03-22 14:20:18 +08006365 if (head->free_bg)
6366 bg_blkno = head->free_bg;
6367 else
6368 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6369 head->free_bit);
Tao Maa09d09b2011-02-22 08:21:10 +08006370 trace_ocfs2_free_cached_blocks(
6371 (unsigned long long)head->free_blk, head->free_bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006372
6373 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6374 head->free_bit, bg_blkno, 1);
6375 if (ret) {
6376 mlog_errno(ret);
6377 goto out_journal;
6378 }
6379
6380 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6381 if (ret) {
6382 mlog_errno(ret);
6383 goto out_journal;
6384 }
6385
6386 tmp = head;
6387 head = head->free_next;
6388 kfree(tmp);
6389 }
6390
6391out_journal:
6392 ocfs2_commit_trans(osb, handle);
6393
6394out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006395 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006396 brelse(di_bh);
6397out_mutex:
6398 mutex_unlock(&inode->i_mutex);
6399 iput(inode);
6400out:
6401 while(head) {
6402 /* Premature exit may have left some dangling items. */
6403 tmp = head;
6404 head = head->free_next;
6405 kfree(tmp);
6406 }
6407
6408 return ret;
6409}
6410
Tao Ma2891d292008-11-12 08:26:58 +08006411int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6412 u64 blkno, unsigned int bit)
6413{
6414 int ret = 0;
6415 struct ocfs2_cached_block_free *item;
6416
Tao Ma74380c42010-03-22 14:20:18 +08006417 item = kzalloc(sizeof(*item), GFP_NOFS);
Tao Ma2891d292008-11-12 08:26:58 +08006418 if (item == NULL) {
6419 ret = -ENOMEM;
6420 mlog_errno(ret);
6421 return ret;
6422 }
6423
Tao Maa09d09b2011-02-22 08:21:10 +08006424 trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
Tao Ma2891d292008-11-12 08:26:58 +08006425
6426 item->free_blk = blkno;
6427 item->free_bit = bit;
6428 item->free_next = ctxt->c_global_allocator;
6429
6430 ctxt->c_global_allocator = item;
6431 return ret;
6432}
6433
6434static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6435 struct ocfs2_cached_block_free *head)
6436{
6437 struct ocfs2_cached_block_free *tmp;
6438 struct inode *tl_inode = osb->osb_tl_inode;
6439 handle_t *handle;
6440 int ret = 0;
6441
6442 mutex_lock(&tl_inode->i_mutex);
6443
6444 while (head) {
6445 if (ocfs2_truncate_log_needs_flush(osb)) {
6446 ret = __ocfs2_flush_truncate_log(osb);
6447 if (ret < 0) {
6448 mlog_errno(ret);
6449 break;
6450 }
6451 }
6452
6453 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6454 if (IS_ERR(handle)) {
6455 ret = PTR_ERR(handle);
6456 mlog_errno(ret);
6457 break;
6458 }
6459
6460 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6461 head->free_bit);
6462
6463 ocfs2_commit_trans(osb, handle);
6464 tmp = head;
6465 head = head->free_next;
6466 kfree(tmp);
6467
6468 if (ret < 0) {
6469 mlog_errno(ret);
6470 break;
6471 }
6472 }
6473
6474 mutex_unlock(&tl_inode->i_mutex);
6475
6476 while (head) {
6477 /* Premature exit may have left some dangling items. */
6478 tmp = head;
6479 head = head->free_next;
6480 kfree(tmp);
6481 }
6482
6483 return ret;
6484}
6485
Mark Fasheh2b604352007-06-22 15:45:27 -07006486int ocfs2_run_deallocs(struct ocfs2_super *osb,
6487 struct ocfs2_cached_dealloc_ctxt *ctxt)
6488{
6489 int ret = 0, ret2;
6490 struct ocfs2_per_slot_free_list *fl;
6491
6492 if (!ctxt)
6493 return 0;
6494
6495 while (ctxt->c_first_suballocator) {
6496 fl = ctxt->c_first_suballocator;
6497
6498 if (fl->f_first) {
Tao Maa09d09b2011-02-22 08:21:10 +08006499 trace_ocfs2_run_deallocs(fl->f_inode_type,
6500 fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006501 ret2 = ocfs2_free_cached_blocks(osb,
6502 fl->f_inode_type,
6503 fl->f_slot,
6504 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006505 if (ret2)
6506 mlog_errno(ret2);
6507 if (!ret)
6508 ret = ret2;
6509 }
6510
6511 ctxt->c_first_suballocator = fl->f_next_suballocator;
6512 kfree(fl);
6513 }
6514
Tao Ma2891d292008-11-12 08:26:58 +08006515 if (ctxt->c_global_allocator) {
6516 ret2 = ocfs2_free_cached_clusters(osb,
6517 ctxt->c_global_allocator);
6518 if (ret2)
6519 mlog_errno(ret2);
6520 if (!ret)
6521 ret = ret2;
6522
6523 ctxt->c_global_allocator = NULL;
6524 }
6525
Mark Fasheh2b604352007-06-22 15:45:27 -07006526 return ret;
6527}
6528
6529static struct ocfs2_per_slot_free_list *
6530ocfs2_find_per_slot_free_list(int type,
6531 int slot,
6532 struct ocfs2_cached_dealloc_ctxt *ctxt)
6533{
6534 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6535
6536 while (fl) {
6537 if (fl->f_inode_type == type && fl->f_slot == slot)
6538 return fl;
6539
6540 fl = fl->f_next_suballocator;
6541 }
6542
6543 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6544 if (fl) {
6545 fl->f_inode_type = type;
6546 fl->f_slot = slot;
6547 fl->f_first = NULL;
6548 fl->f_next_suballocator = ctxt->c_first_suballocator;
6549
6550 ctxt->c_first_suballocator = fl;
6551 }
6552 return fl;
6553}
6554
Tao Ma1823cb02009-08-18 11:24:49 +08006555int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
Tao Ma74380c42010-03-22 14:20:18 +08006556 int type, int slot, u64 suballoc,
6557 u64 blkno, unsigned int bit)
Mark Fasheh2b604352007-06-22 15:45:27 -07006558{
6559 int ret;
6560 struct ocfs2_per_slot_free_list *fl;
6561 struct ocfs2_cached_block_free *item;
6562
6563 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6564 if (fl == NULL) {
6565 ret = -ENOMEM;
6566 mlog_errno(ret);
6567 goto out;
6568 }
6569
Tao Ma74380c42010-03-22 14:20:18 +08006570 item = kzalloc(sizeof(*item), GFP_NOFS);
Mark Fasheh2b604352007-06-22 15:45:27 -07006571 if (item == NULL) {
6572 ret = -ENOMEM;
6573 mlog_errno(ret);
6574 goto out;
6575 }
6576
Tao Maa09d09b2011-02-22 08:21:10 +08006577 trace_ocfs2_cache_block_dealloc(type, slot,
6578 (unsigned long long)suballoc,
6579 (unsigned long long)blkno, bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006580
Tao Ma74380c42010-03-22 14:20:18 +08006581 item->free_bg = suballoc;
Mark Fasheh2b604352007-06-22 15:45:27 -07006582 item->free_blk = blkno;
6583 item->free_bit = bit;
6584 item->free_next = fl->f_first;
6585
6586 fl->f_first = item;
6587
6588 ret = 0;
6589out:
6590 return ret;
6591}
6592
Mark Fasheh59a5e412007-06-22 15:52:36 -07006593static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6594 struct ocfs2_extent_block *eb)
6595{
6596 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6597 le16_to_cpu(eb->h_suballoc_slot),
Tao Ma74380c42010-03-22 14:20:18 +08006598 le64_to_cpu(eb->h_suballoc_loc),
Mark Fasheh59a5e412007-06-22 15:52:36 -07006599 le64_to_cpu(eb->h_blkno),
6600 le16_to_cpu(eb->h_suballoc_bit));
6601}
6602
Joel Becker2b4e30f2008-09-03 20:03:41 -07006603static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006604{
6605 set_buffer_uptodate(bh);
6606 mark_buffer_dirty(bh);
6607 return 0;
6608}
6609
Tao Ma6f70fa52009-08-25 08:05:12 +08006610void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6611 unsigned int from, unsigned int to,
6612 struct page *page, int zero, u64 *phys)
Mark Fasheh1d410a62007-09-07 14:20:45 -07006613{
6614 int ret, partial = 0;
6615
6616 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6617 if (ret)
6618 mlog_errno(ret);
6619
6620 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006621 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006622
6623 /*
6624 * Need to set the buffers we zero'd into uptodate
6625 * here if they aren't - ocfs2_map_page_blocks()
6626 * might've skipped some
6627 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006628 ret = walk_page_buffers(handle, page_buffers(page),
6629 from, to, &partial,
6630 ocfs2_zero_func);
6631 if (ret < 0)
6632 mlog_errno(ret);
6633 else if (ocfs2_should_order_data(inode)) {
6634 ret = ocfs2_jbd2_file_inode(handle, inode);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006635 if (ret < 0)
6636 mlog_errno(ret);
6637 }
6638
6639 if (!partial)
6640 SetPageUptodate(page);
6641
6642 flush_dcache_page(page);
6643}
6644
Mark Fasheh35edec12007-07-06 14:41:18 -07006645static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6646 loff_t end, struct page **pages,
6647 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006648{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006649 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006650 struct page *page;
6651 unsigned int from, to = PAGE_CACHE_SIZE;
6652 struct super_block *sb = inode->i_sb;
6653
6654 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6655
6656 if (numpages == 0)
6657 goto out;
6658
Mark Fasheh35edec12007-07-06 14:41:18 -07006659 to = PAGE_CACHE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006660 for(i = 0; i < numpages; i++) {
6661 page = pages[i];
6662
Mark Fasheh35edec12007-07-06 14:41:18 -07006663 from = start & (PAGE_CACHE_SIZE - 1);
6664 if ((end >> PAGE_CACHE_SHIFT) == page->index)
6665 to = end & (PAGE_CACHE_SIZE - 1);
6666
Mark Fasheh60b11392007-02-16 11:46:50 -08006667 BUG_ON(from > PAGE_CACHE_SIZE);
6668 BUG_ON(to > PAGE_CACHE_SIZE);
6669
Mark Fasheh1d410a62007-09-07 14:20:45 -07006670 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6671 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006672
Mark Fasheh35edec12007-07-06 14:41:18 -07006673 start = (page->index + 1) << PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006674 }
6675out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006676 if (pages)
6677 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006678}
6679
Tao Ma6f70fa52009-08-25 08:05:12 +08006680int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6681 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006682{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006683 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006684 struct address_space *mapping = inode->i_mapping;
6685 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006686 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006687
Mark Fasheh35edec12007-07-06 14:41:18 -07006688 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006689
Mark Fasheh1d410a62007-09-07 14:20:45 -07006690 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006691 last_page_bytes = PAGE_ALIGN(end);
6692 index = start >> PAGE_CACHE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006693 do {
Jan Kara9b4c0ff2010-08-24 14:28:03 +02006694 pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fasheh60b11392007-02-16 11:46:50 -08006695 if (!pages[numpages]) {
6696 ret = -ENOMEM;
6697 mlog_errno(ret);
6698 goto out;
6699 }
6700
6701 numpages++;
6702 index++;
Mark Fasheh35edec12007-07-06 14:41:18 -07006703 } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006704
6705out:
6706 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006707 if (pages)
6708 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006709 numpages = 0;
6710 }
6711
6712 *num = numpages;
6713
6714 return ret;
6715}
6716
Tao Ma6f70fa52009-08-25 08:05:12 +08006717static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6718 struct page **pages, int *num)
6719{
6720 struct super_block *sb = inode->i_sb;
6721
6722 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6723 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6724
6725 return ocfs2_grab_pages(inode, start, end, pages, num);
6726}
6727
Mark Fasheh60b11392007-02-16 11:46:50 -08006728/*
6729 * Zero the area past i_size but still within an allocated
6730 * cluster. This avoids exposing nonzero data on subsequent file
6731 * extends.
6732 *
6733 * We need to call this before i_size is updated on the inode because
6734 * otherwise block_write_full_page() will skip writeout of pages past
6735 * i_size. The new_i_size parameter is passed for this reason.
6736 */
Mark Fasheh35edec12007-07-06 14:41:18 -07006737int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6738 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08006739{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006740 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08006741 struct page **pages = NULL;
6742 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006743 unsigned int ext_flags;
6744 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08006745
6746 /*
6747 * File systems which don't support sparse files zero on every
6748 * extend.
6749 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07006750 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08006751 return 0;
6752
Mark Fasheh1d410a62007-09-07 14:20:45 -07006753 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08006754 sizeof(struct page *), GFP_NOFS);
6755 if (pages == NULL) {
6756 ret = -ENOMEM;
6757 mlog_errno(ret);
6758 goto out;
6759 }
6760
Mark Fasheh1d410a62007-09-07 14:20:45 -07006761 if (range_start == range_end)
6762 goto out;
6763
6764 ret = ocfs2_extent_map_get_blocks(inode,
6765 range_start >> sb->s_blocksize_bits,
6766 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08006767 if (ret) {
6768 mlog_errno(ret);
6769 goto out;
6770 }
6771
Mark Fasheh1d410a62007-09-07 14:20:45 -07006772 /*
6773 * Tail is a hole, or is marked unwritten. In either case, we
6774 * can count on read and write to return/push zero's.
6775 */
6776 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08006777 goto out;
6778
Mark Fasheh1d410a62007-09-07 14:20:45 -07006779 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6780 &numpages);
6781 if (ret) {
6782 mlog_errno(ret);
6783 goto out;
6784 }
6785
Mark Fasheh35edec12007-07-06 14:41:18 -07006786 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6787 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08006788
6789 /*
6790 * Initiate writeout of the pages we zero'd here. We don't
6791 * wait on them - the truncate_inode_pages() call later will
6792 * do that for us.
6793 */
Christoph Hellwig2cfd30a2009-09-23 15:04:02 +02006794 ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6795 range_end - 1);
Mark Fasheh60b11392007-02-16 11:46:50 -08006796 if (ret)
6797 mlog_errno(ret);
6798
6799out:
Tim Gardnerd787ab02013-02-21 16:42:44 -08006800 kfree(pages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006801
6802 return ret;
6803}
6804
Tiger Yangfdd77702008-08-18 17:08:55 +08006805static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
6806 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006807{
6808 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08006809 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006810
Tiger Yangfdd77702008-08-18 17:08:55 +08006811 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
6812 memset(&di->id2, 0, blocksize -
6813 offsetof(struct ocfs2_dinode, id2) -
6814 xattrsize);
6815 else
6816 memset(&di->id2, 0, blocksize -
6817 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006818}
6819
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006820void ocfs2_dinode_new_extent_list(struct inode *inode,
6821 struct ocfs2_dinode *di)
6822{
Tiger Yangfdd77702008-08-18 17:08:55 +08006823 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006824 di->id2.i_list.l_tree_depth = 0;
6825 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08006826 di->id2.i_list.l_count = cpu_to_le16(
6827 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006828}
6829
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006830void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
6831{
6832 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6833 struct ocfs2_inline_data *idata = &di->id2.i_data;
6834
6835 spin_lock(&oi->ip_lock);
6836 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
6837 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6838 spin_unlock(&oi->ip_lock);
6839
6840 /*
6841 * We clear the entire i_data structure here so that all
6842 * fields can be properly initialized.
6843 */
Tiger Yangfdd77702008-08-18 17:08:55 +08006844 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006845
Tiger Yangfdd77702008-08-18 17:08:55 +08006846 idata->id_count = cpu_to_le16(
6847 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006848}
6849
6850int ocfs2_convert_inline_data_to_extents(struct inode *inode,
6851 struct buffer_head *di_bh)
6852{
6853 int ret, i, has_data, num_pages = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006854 int need_free = 0;
6855 u32 bit_off, num;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006856 handle_t *handle;
6857 u64 uninitialized_var(block);
6858 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6859 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6860 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006861 struct ocfs2_alloc_context *data_ac = NULL;
6862 struct page **pages = NULL;
6863 loff_t end = osb->s_clustersize;
Joel Beckerf99b9b72008-08-20 19:36:33 -07006864 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02006865 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006866
6867 has_data = i_size_read(inode) ? 1 : 0;
6868
6869 if (has_data) {
6870 pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
6871 sizeof(struct page *), GFP_NOFS);
6872 if (pages == NULL) {
6873 ret = -ENOMEM;
6874 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006875 return ret;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006876 }
6877
6878 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
6879 if (ret) {
6880 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006881 goto free_pages;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006882 }
6883 }
6884
Jan Karaa90714c2008-10-09 19:38:40 +02006885 handle = ocfs2_start_trans(osb,
6886 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006887 if (IS_ERR(handle)) {
6888 ret = PTR_ERR(handle);
6889 mlog_errno(ret);
alex chen15eba0f2015-02-10 14:09:02 -08006890 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006891 }
6892
Joel Becker0cf2f762009-02-12 16:41:25 -08006893 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07006894 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006895 if (ret) {
6896 mlog_errno(ret);
6897 goto out_commit;
6898 }
6899
6900 if (has_data) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006901 unsigned int page_end;
6902 u64 phys;
6903
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006904 ret = dquot_alloc_space_nodirty(inode,
6905 ocfs2_clusters_to_bytes(osb->sb, 1));
6906 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02006907 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02006908 did_quota = 1;
6909
Mark Fasheh4fe370a2009-12-07 13:15:40 -08006910 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
6911
Joel Becker1ed9b772010-05-06 13:59:06 +08006912 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006913 &num);
6914 if (ret) {
6915 mlog_errno(ret);
6916 goto out_commit;
6917 }
6918
6919 /*
6920 * Save two copies, one for insert, and one that can
6921 * be changed by ocfs2_map_and_dirty_page() below.
6922 */
6923 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
6924
6925 /*
6926 * Non sparse file systems zero on extend, so no need
6927 * to do that now.
6928 */
6929 if (!ocfs2_sparse_alloc(osb) &&
6930 PAGE_CACHE_SIZE < osb->s_clustersize)
6931 end = PAGE_CACHE_SIZE;
6932
6933 ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
6934 if (ret) {
6935 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006936 need_free = 1;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006937 goto out_commit;
6938 }
6939
6940 /*
6941 * This should populate the 1st page for us and mark
6942 * it up to date.
6943 */
6944 ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
6945 if (ret) {
6946 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006947 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08006948 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006949 }
6950
6951 page_end = PAGE_CACHE_SIZE;
6952 if (PAGE_CACHE_SIZE > osb->s_clustersize)
6953 page_end = osb->s_clustersize;
6954
6955 for (i = 0; i < num_pages; i++)
6956 ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
6957 pages[i], i > 0, &phys);
6958 }
6959
6960 spin_lock(&oi->ip_lock);
6961 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
6962 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6963 spin_unlock(&oi->ip_lock);
6964
Darrick J. Wong2931cdc2014-04-03 14:46:48 -07006965 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006966 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006967
6968 ocfs2_journal_dirty(handle, di_bh);
6969
6970 if (has_data) {
6971 /*
6972 * An error at this point should be extremely rare. If
6973 * this proves to be false, we could always re-build
6974 * the in-inode data from our pages.
6975 */
Joel Becker5e404e92009-02-13 03:54:22 -08006976 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08006977 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006978 if (ret) {
6979 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006980 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08006981 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006982 }
6983
6984 inode->i_blocks = ocfs2_inode_sector_count(inode);
6985 }
6986
alex chen15eba0f2015-02-10 14:09:02 -08006987out_unlock:
6988 if (pages)
6989 ocfs2_unlock_and_free_pages(pages, num_pages);
6990
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006991out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02006992 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006993 dquot_free_space_nodirty(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02006994 ocfs2_clusters_to_bytes(osb->sb, 1));
6995
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006996 if (need_free) {
6997 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
6998 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
6999 bit_off, num);
7000 else
7001 ocfs2_free_clusters(handle,
7002 data_ac->ac_inode,
7003 data_ac->ac_bh,
7004 ocfs2_clusters_to_blocks(osb->sb, bit_off),
7005 num);
7006 }
7007
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007008 ocfs2_commit_trans(osb, handle);
7009
alex chen15eba0f2015-02-10 14:09:02 -08007010out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007011 if (data_ac)
7012 ocfs2_free_alloc_context(data_ac);
Markus Elfring3cc79b72015-04-14 15:42:45 -07007013free_pages:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07007014 kfree(pages);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007015 return ret;
7016}
7017
Mark Fashehccd979b2005-12-15 14:31:24 -08007018/*
7019 * It is expected, that by the time you call this function,
7020 * inode->i_size and fe->i_size have been adjusted.
7021 *
7022 * WARNING: This will kfree the truncate context
7023 */
7024int ocfs2_commit_truncate(struct ocfs2_super *osb,
7025 struct inode *inode,
Tristan Ye78f94672010-05-11 17:54:42 +08007026 struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08007027{
Tristan Ye78f94672010-05-11 17:54:42 +08007028 int status = 0, i, flags = 0;
7029 u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
Tao Mabcbbb242009-08-18 11:29:12 +08007030 u64 blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08007031 struct ocfs2_extent_list *el;
Tristan Ye78f94672010-05-11 17:54:42 +08007032 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08007033 struct ocfs2_path *path = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08007034 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7035 struct ocfs2_extent_list *root_el = &(di->id2.i_list);
7036 u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
7037 struct ocfs2_extent_tree et;
7038 struct ocfs2_cached_dealloc_ctxt dealloc;
Junxiao Bif62f12b2014-12-18 16:17:32 -08007039 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007040
Tristan Ye78f94672010-05-11 17:54:42 +08007041 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7042 ocfs2_init_dealloc_ctxt(&dealloc);
7043
Mark Fashehdcd05382007-01-16 11:32:23 -08007044 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08007045 i_size_read(inode));
7046
Tristan Ye78f94672010-05-11 17:54:42 +08007047 path = ocfs2_new_path(di_bh, &di->id2.i_list,
Joel Becker13723d02008-10-17 19:25:01 -07007048 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08007049 if (!path) {
7050 status = -ENOMEM;
7051 mlog_errno(status);
7052 goto bail;
7053 }
Mark Fasheh83418972007-04-23 18:53:12 -07007054
7055 ocfs2_extent_map_trunc(inode, new_highest_cpos);
7056
Mark Fashehccd979b2005-12-15 14:31:24 -08007057start:
Mark Fashehdcd05382007-01-16 11:32:23 -08007058 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007059 * Check that we still have allocation to delete.
7060 */
7061 if (OCFS2_I(inode)->ip_clusters == 0) {
7062 status = 0;
7063 goto bail;
7064 }
7065
7066 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007067 * Truncate always works against the rightmost tree branch.
7068 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007069 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007070 if (status) {
7071 mlog_errno(status);
7072 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007073 }
7074
Tao Maa09d09b2011-02-22 08:21:10 +08007075 trace_ocfs2_commit_truncate(
7076 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7077 new_highest_cpos,
7078 OCFS2_I(inode)->ip_clusters,
7079 path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08007080
7081 /*
7082 * By now, el will point to the extent list on the bottom most
7083 * portion of this tree. Only the tail record is considered in
7084 * each pass.
7085 *
7086 * We handle the following cases, in order:
7087 * - empty extent: delete the remaining branch
7088 * - remove the entire record
7089 * - remove a partial record
7090 * - no record needs to be removed (truncate has completed)
7091 */
7092 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007093 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7094 ocfs2_error(inode->i_sb,
7095 "Inode %llu has empty extent block at %llu\n",
7096 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7097 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7098 status = -EROFS;
7099 goto bail;
7100 }
7101
Mark Fashehccd979b2005-12-15 14:31:24 -08007102 i = le16_to_cpu(el->l_next_free_rec) - 1;
Tristan Ye78f94672010-05-11 17:54:42 +08007103 rec = &el->l_recs[i];
7104 flags = rec->e_flags;
7105 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7106
7107 if (i == 0 && ocfs2_is_empty_extent(rec)) {
7108 /*
7109 * Lower levels depend on this never happening, but it's best
7110 * to check it up here before changing the tree.
7111 */
7112 if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
Xue jiufeid0c97d52015-09-04 15:44:48 -07007113 mlog(ML_ERROR, "Inode %lu has an empty "
Tristan Ye78f94672010-05-11 17:54:42 +08007114 "extent record, depth %u\n", inode->i_ino,
7115 le16_to_cpu(root_el->l_tree_depth));
Xue jiufeid0c97d52015-09-04 15:44:48 -07007116 status = ocfs2_remove_rightmost_empty_extent(osb,
7117 &et, path, &dealloc);
7118 if (status) {
7119 mlog_errno(status);
7120 goto bail;
7121 }
7122
7123 ocfs2_reinit_path(path, 1);
7124 goto start;
7125 } else {
7126 trunc_cpos = le32_to_cpu(rec->e_cpos);
7127 trunc_len = 0;
7128 blkno = 0;
Tristan Ye78f94672010-05-11 17:54:42 +08007129 }
Tristan Ye78f94672010-05-11 17:54:42 +08007130 } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7131 /*
7132 * Truncate entire record.
7133 */
7134 trunc_cpos = le32_to_cpu(rec->e_cpos);
7135 trunc_len = ocfs2_rec_clusters(el, rec);
7136 blkno = le64_to_cpu(rec->e_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007137 } else if (range > new_highest_cpos) {
Tristan Ye78f94672010-05-11 17:54:42 +08007138 /*
7139 * Partial truncate. it also should be
7140 * the last truncate we're doing.
7141 */
7142 trunc_cpos = new_highest_cpos;
7143 trunc_len = range - new_highest_cpos;
7144 coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7145 blkno = le64_to_cpu(rec->e_blkno) +
7146 ocfs2_clusters_to_blocks(inode->i_sb, coff);
Mark Fashehdcd05382007-01-16 11:32:23 -08007147 } else {
Tristan Ye78f94672010-05-11 17:54:42 +08007148 /*
7149 * Truncate completed, leave happily.
7150 */
Mark Fashehdcd05382007-01-16 11:32:23 -08007151 status = 0;
7152 goto bail;
7153 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007154
Tristan Ye78f94672010-05-11 17:54:42 +08007155 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007156
Junxiao Bif62f12b2014-12-18 16:17:32 -08007157 if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) {
7158 status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
7159 &ref_tree, NULL);
7160 if (status) {
7161 mlog_errno(status);
7162 goto bail;
7163 }
7164 }
7165
Tristan Ye78f94672010-05-11 17:54:42 +08007166 status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7167 phys_cpos, trunc_len, flags, &dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08007168 refcount_loc, true);
Mark Fashehccd979b2005-12-15 14:31:24 -08007169 if (status < 0) {
7170 mlog_errno(status);
7171 goto bail;
7172 }
7173
Mark Fashehdcd05382007-01-16 11:32:23 -08007174 ocfs2_reinit_path(path, 1);
7175
7176 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007177 * The check above will catch the case where we've truncated
7178 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007179 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007180 goto start;
7181
Mark Fashehccd979b2005-12-15 14:31:24 -08007182bail:
Junxiao Bif62f12b2014-12-18 16:17:32 -08007183 if (ref_tree)
7184 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08007185
7186 ocfs2_schedule_truncate_log_flush(osb, 1);
7187
Tristan Ye78f94672010-05-11 17:54:42 +08007188 ocfs2_run_deallocs(osb, &dealloc);
Mark Fasheh59a5e412007-06-22 15:52:36 -07007189
Mark Fashehdcd05382007-01-16 11:32:23 -08007190 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007191
Mark Fashehccd979b2005-12-15 14:31:24 -08007192 return status;
7193}
7194
Mark Fashehccd979b2005-12-15 14:31:24 -08007195/*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007196 * 'start' is inclusive, 'end' is not.
7197 */
7198int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7199 unsigned int start, unsigned int end, int trunc)
7200{
7201 int ret;
7202 unsigned int numbytes;
7203 handle_t *handle;
7204 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7205 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7206 struct ocfs2_inline_data *idata = &di->id2.i_data;
7207
7208 if (end > i_size_read(inode))
7209 end = i_size_read(inode);
7210
Younger Liud62e74b2014-02-10 14:25:51 -08007211 BUG_ON(start > end);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007212
7213 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7214 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7215 !ocfs2_supports_inline_data(osb)) {
7216 ocfs2_error(inode->i_sb,
Joe Perches7ecef142015-09-04 15:44:51 -07007217 "Inline data flags for inode %llu don't agree! Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007218 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7219 le16_to_cpu(di->i_dyn_features),
7220 OCFS2_I(inode)->ip_dyn_features,
7221 osb->s_feature_incompat);
7222 ret = -EROFS;
7223 goto out;
7224 }
7225
7226 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7227 if (IS_ERR(handle)) {
7228 ret = PTR_ERR(handle);
7229 mlog_errno(ret);
7230 goto out;
7231 }
7232
Joel Becker0cf2f762009-02-12 16:41:25 -08007233 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007234 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007235 if (ret) {
7236 mlog_errno(ret);
7237 goto out_commit;
7238 }
7239
7240 numbytes = end - start;
7241 memset(idata->id_data + start, 0, numbytes);
7242
7243 /*
7244 * No need to worry about the data page here - it's been
7245 * truncated already and inline data doesn't need it for
7246 * pushing zero's to disk, so we'll let readpage pick it up
7247 * later.
7248 */
7249 if (trunc) {
7250 i_size_write(inode, start);
7251 di->i_size = cpu_to_le64(start);
7252 }
7253
7254 inode->i_blocks = ocfs2_inode_sector_count(inode);
7255 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7256
7257 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7258 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7259
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07007260 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007261 ocfs2_journal_dirty(handle, di_bh);
7262
7263out_commit:
7264 ocfs2_commit_trans(osb, handle);
7265
7266out:
7267 return ret;
7268}
Tao Mae80de362011-05-23 10:36:43 +08007269
7270static int ocfs2_trim_extent(struct super_block *sb,
7271 struct ocfs2_group_desc *gd,
7272 u32 start, u32 count)
7273{
7274 u64 discard, bcount;
7275
7276 bcount = ocfs2_clusters_to_blocks(sb, count);
7277 discard = le64_to_cpu(gd->bg_blkno) +
7278 ocfs2_clusters_to_blocks(sb, start);
7279
Tao Ma10fca352011-05-23 10:36:45 +08007280 trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7281
Tao Mae80de362011-05-23 10:36:43 +08007282 return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7283}
7284
7285static int ocfs2_trim_group(struct super_block *sb,
7286 struct ocfs2_group_desc *gd,
7287 u32 start, u32 max, u32 minbits)
7288{
7289 int ret = 0, count = 0, next;
7290 void *bitmap = gd->bg_bitmap;
7291
7292 if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7293 return 0;
7294
Tao Ma10fca352011-05-23 10:36:45 +08007295 trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7296 start, max, minbits);
7297
Tao Mae80de362011-05-23 10:36:43 +08007298 while (start < max) {
7299 start = ocfs2_find_next_zero_bit(bitmap, max, start);
7300 if (start >= max)
7301 break;
7302 next = ocfs2_find_next_bit(bitmap, max, start);
7303
7304 if ((next - start) >= minbits) {
7305 ret = ocfs2_trim_extent(sb, gd,
7306 start, next - start);
7307 if (ret < 0) {
7308 mlog_errno(ret);
7309 break;
7310 }
7311 count += next - start;
7312 }
7313 start = next + 1;
7314
7315 if (fatal_signal_pending(current)) {
7316 count = -ERESTARTSYS;
7317 break;
7318 }
7319
7320 if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7321 break;
7322 }
7323
7324 if (ret < 0)
7325 count = ret;
7326
7327 return count;
7328}
7329
7330int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7331{
7332 struct ocfs2_super *osb = OCFS2_SB(sb);
7333 u64 start, len, trimmed, first_group, last_group, group;
7334 int ret, cnt;
7335 u32 first_bit, last_bit, minlen;
7336 struct buffer_head *main_bm_bh = NULL;
7337 struct inode *main_bm_inode = NULL;
7338 struct buffer_head *gd_bh = NULL;
7339 struct ocfs2_dinode *main_bm;
7340 struct ocfs2_group_desc *gd = NULL;
7341
7342 start = range->start >> osb->s_clustersize_bits;
7343 len = range->len >> osb->s_clustersize_bits;
7344 minlen = range->minlen >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007345
Jie Liuaa897622014-01-21 15:48:35 -08007346 if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
Tao Mae80de362011-05-23 10:36:43 +08007347 return -EINVAL;
7348
7349 main_bm_inode = ocfs2_get_system_file_inode(osb,
7350 GLOBAL_BITMAP_SYSTEM_INODE,
7351 OCFS2_INVALID_SLOT);
7352 if (!main_bm_inode) {
7353 ret = -EIO;
7354 mlog_errno(ret);
7355 goto out;
7356 }
7357
7358 mutex_lock(&main_bm_inode->i_mutex);
7359
7360 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7361 if (ret < 0) {
7362 mlog_errno(ret);
7363 goto out_mutex;
7364 }
7365 main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7366
7367 if (start >= le32_to_cpu(main_bm->i_clusters)) {
7368 ret = -EINVAL;
7369 goto out_unlock;
7370 }
7371
Jie Liuaa897622014-01-21 15:48:35 -08007372 len = range->len >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007373 if (start + len > le32_to_cpu(main_bm->i_clusters))
7374 len = le32_to_cpu(main_bm->i_clusters) - start;
7375
Tao Ma10fca352011-05-23 10:36:45 +08007376 trace_ocfs2_trim_fs(start, len, minlen);
7377
Tao Mae80de362011-05-23 10:36:43 +08007378 /* Determine first and last group to examine based on start and len */
7379 first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7380 if (first_group == osb->first_cluster_group_blkno)
7381 first_bit = start;
7382 else
7383 first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
7384 last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
7385 last_bit = osb->bitmap_cpg;
7386
Jie Liuaa897622014-01-21 15:48:35 -08007387 trimmed = 0;
Tao Mae80de362011-05-23 10:36:43 +08007388 for (group = first_group; group <= last_group;) {
7389 if (first_bit + len >= osb->bitmap_cpg)
7390 last_bit = osb->bitmap_cpg;
7391 else
7392 last_bit = first_bit + len;
7393
7394 ret = ocfs2_read_group_descriptor(main_bm_inode,
7395 main_bm, group,
7396 &gd_bh);
7397 if (ret < 0) {
7398 mlog_errno(ret);
7399 break;
7400 }
7401
7402 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
7403 cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
7404 brelse(gd_bh);
7405 gd_bh = NULL;
7406 if (cnt < 0) {
7407 ret = cnt;
7408 mlog_errno(ret);
7409 break;
7410 }
7411
7412 trimmed += cnt;
7413 len -= osb->bitmap_cpg - first_bit;
7414 first_bit = 0;
7415 if (group == osb->first_cluster_group_blkno)
7416 group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7417 else
7418 group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7419 }
7420 range->len = trimmed * sb->s_blocksize;
7421out_unlock:
7422 ocfs2_inode_unlock(main_bm_inode, 0);
7423 brelse(main_bm_bh);
7424out_mutex:
7425 mutex_unlock(&main_bm_inode->i_mutex);
7426 iput(main_bm_inode);
7427out:
7428 return ret;
7429}