blob: 460c0cedab3a435d5203c753c51b2cedbb709a67 [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
Joel Becker09106ba2009-02-12 19:43:57 -08002519 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002520 if (ret) {
2521 mlog_errno(ret);
2522 goto out;
2523 }
2524
Mark Fasheh328d5752007-06-18 10:48:04 -07002525 /* Path should always be rightmost. */
2526 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2527 BUG_ON(eb->h_next_leaf_blk != 0ULL);
2528
2529 el = &eb->h_list;
2530 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2531 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2532 rec = &el->l_recs[idx];
2533 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2534
2535 for (i = 0; i < path->p_tree_depth; i++) {
2536 el = path->p_node[i].el;
2537 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2538 rec = &el->l_recs[idx];
2539
2540 rec->e_int_clusters = cpu_to_le32(range);
2541 le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2542
2543 ocfs2_journal_dirty(handle, path->p_node[i].bh);
2544 }
Tao Ma3c5e1062009-07-21 15:42:05 +08002545out:
2546 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002547}
2548
Joel Becker6641b0c2009-02-12 18:57:52 -08002549static void ocfs2_unlink_path(handle_t *handle,
2550 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002551 struct ocfs2_cached_dealloc_ctxt *dealloc,
2552 struct ocfs2_path *path, int unlink_start)
2553{
2554 int ret, i;
2555 struct ocfs2_extent_block *eb;
2556 struct ocfs2_extent_list *el;
2557 struct buffer_head *bh;
2558
2559 for(i = unlink_start; i < path_num_items(path); i++) {
2560 bh = path->p_node[i].bh;
2561
2562 eb = (struct ocfs2_extent_block *)bh->b_data;
2563 /*
2564 * Not all nodes might have had their final count
2565 * decremented by the caller - handle this here.
2566 */
2567 el = &eb->h_list;
2568 if (le16_to_cpu(el->l_next_free_rec) > 1) {
2569 mlog(ML_ERROR,
2570 "Inode %llu, attempted to remove extent block "
2571 "%llu with %u records\n",
Joel Becker6641b0c2009-02-12 18:57:52 -08002572 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07002573 (unsigned long long)le64_to_cpu(eb->h_blkno),
2574 le16_to_cpu(el->l_next_free_rec));
2575
2576 ocfs2_journal_dirty(handle, bh);
Joel Becker6641b0c2009-02-12 18:57:52 -08002577 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002578 continue;
2579 }
2580
2581 el->l_next_free_rec = 0;
2582 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2583
2584 ocfs2_journal_dirty(handle, bh);
2585
2586 ret = ocfs2_cache_extent_block_free(dealloc, eb);
2587 if (ret)
2588 mlog_errno(ret);
2589
Joel Becker6641b0c2009-02-12 18:57:52 -08002590 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002591 }
2592}
2593
Joel Becker6641b0c2009-02-12 18:57:52 -08002594static void ocfs2_unlink_subtree(handle_t *handle,
2595 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002596 struct ocfs2_path *left_path,
2597 struct ocfs2_path *right_path,
2598 int subtree_index,
2599 struct ocfs2_cached_dealloc_ctxt *dealloc)
2600{
2601 int i;
2602 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2603 struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
2604 struct ocfs2_extent_list *el;
2605 struct ocfs2_extent_block *eb;
2606
2607 el = path_leaf_el(left_path);
2608
2609 eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2610
2611 for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2612 if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2613 break;
2614
2615 BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2616
2617 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2618 le16_add_cpu(&root_el->l_next_free_rec, -1);
2619
2620 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2621 eb->h_next_leaf_blk = 0;
2622
2623 ocfs2_journal_dirty(handle, root_bh);
2624 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2625
Joel Becker6641b0c2009-02-12 18:57:52 -08002626 ocfs2_unlink_path(handle, et, dealloc, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002627 subtree_index + 1);
2628}
2629
Joel Becker1e2dd632009-02-12 19:45:28 -08002630static int ocfs2_rotate_subtree_left(handle_t *handle,
2631 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002632 struct ocfs2_path *left_path,
2633 struct ocfs2_path *right_path,
2634 int subtree_index,
2635 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker1e2dd632009-02-12 19:45:28 -08002636 int *deleted)
Mark Fasheh328d5752007-06-18 10:48:04 -07002637{
2638 int ret, i, del_right_subtree = 0, right_has_empty = 0;
Tao Mae7d4cb62008-08-18 17:38:44 +08002639 struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002640 struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2641 struct ocfs2_extent_block *eb;
2642
2643 *deleted = 0;
2644
2645 right_leaf_el = path_leaf_el(right_path);
2646 left_leaf_el = path_leaf_el(left_path);
2647 root_bh = left_path->p_node[subtree_index].bh;
2648 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2649
2650 if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2651 return 0;
2652
2653 eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2654 if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2655 /*
2656 * It's legal for us to proceed if the right leaf is
2657 * the rightmost one and it has an empty extent. There
2658 * are two cases to handle - whether the leaf will be
2659 * empty after removal or not. If the leaf isn't empty
2660 * then just remove the empty extent up front. The
2661 * next block will handle empty leaves by flagging
2662 * them for unlink.
2663 *
2664 * Non rightmost leaves will throw -EAGAIN and the
2665 * caller can manually move the subtree and retry.
2666 */
2667
2668 if (eb->h_next_leaf_blk != 0ULL)
2669 return -EAGAIN;
2670
2671 if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002672 ret = ocfs2_journal_access_eb(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002673 path_leaf_bh(right_path),
2674 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002675 if (ret) {
2676 mlog_errno(ret);
2677 goto out;
2678 }
2679
2680 ocfs2_remove_empty_extent(right_leaf_el);
2681 } else
2682 right_has_empty = 1;
2683 }
2684
2685 if (eb->h_next_leaf_blk == 0ULL &&
2686 le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2687 /*
2688 * We have to update i_last_eb_blk during the meta
2689 * data delete.
2690 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08002691 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07002692 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002693 if (ret) {
2694 mlog_errno(ret);
2695 goto out;
2696 }
2697
2698 del_right_subtree = 1;
2699 }
2700
2701 /*
2702 * Getting here with an empty extent in the right path implies
2703 * that it's the rightmost path and will be deleted.
2704 */
2705 BUG_ON(right_has_empty && !del_right_subtree);
2706
Joel Becker1e2dd632009-02-12 19:45:28 -08002707 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002708 subtree_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07002709 if (ret) {
2710 mlog_errno(ret);
2711 goto out;
2712 }
2713
2714 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002715 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002716 right_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002717 if (ret) {
2718 mlog_errno(ret);
2719 goto out;
2720 }
2721
Joel Becker1e2dd632009-02-12 19:45:28 -08002722 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002723 left_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002724 if (ret) {
2725 mlog_errno(ret);
2726 goto out;
2727 }
2728 }
2729
2730 if (!right_has_empty) {
2731 /*
2732 * Only do this if we're moving a real
2733 * record. Otherwise, the action is delayed until
2734 * after removal of the right path in which case we
2735 * can do a simple shift to remove the empty extent.
2736 */
2737 ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2738 memset(&right_leaf_el->l_recs[0], 0,
2739 sizeof(struct ocfs2_extent_rec));
2740 }
2741 if (eb->h_next_leaf_blk == 0ULL) {
2742 /*
2743 * Move recs over to get rid of empty extent, decrease
2744 * next_free. This is allowed to remove the last
2745 * extent in our leaf (setting l_next_free_rec to
2746 * zero) - the delete code below won't care.
2747 */
2748 ocfs2_remove_empty_extent(right_leaf_el);
2749 }
2750
Joel Beckerec20cec2010-03-19 14:13:52 -07002751 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2752 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07002753
2754 if (del_right_subtree) {
Joel Becker6641b0c2009-02-12 18:57:52 -08002755 ocfs2_unlink_subtree(handle, et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002756 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08002757 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08002758 left_path);
2759 if (ret) {
2760 mlog_errno(ret);
2761 goto out;
2762 }
Mark Fasheh328d5752007-06-18 10:48:04 -07002763
2764 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07002765 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07002766
2767 /*
2768 * Removal of the extent in the left leaf was skipped
2769 * above so we could delete the right path
2770 * 1st.
2771 */
2772 if (right_has_empty)
2773 ocfs2_remove_empty_extent(left_leaf_el);
2774
Joel Beckerec20cec2010-03-19 14:13:52 -07002775 ocfs2_journal_dirty(handle, et_root_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002776
2777 *deleted = 1;
2778 } else
Joel Becker4619c732009-02-12 19:02:36 -08002779 ocfs2_complete_edge_insert(handle, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002780 subtree_index);
2781
2782out:
2783 return ret;
2784}
2785
2786/*
2787 * Given a full path, determine what cpos value would return us a path
2788 * containing the leaf immediately to the right of the current one.
2789 *
2790 * Will return zero if the path passed in is already the rightmost path.
2791 *
2792 * This looks similar, but is subtly different to
2793 * ocfs2_find_cpos_for_left_leaf().
2794 */
Tao Ma38a04e42009-11-30 14:32:19 +08002795int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2796 struct ocfs2_path *path, u32 *cpos)
Mark Fasheh328d5752007-06-18 10:48:04 -07002797{
2798 int i, j, ret = 0;
2799 u64 blkno;
2800 struct ocfs2_extent_list *el;
2801
2802 *cpos = 0;
2803
2804 if (path->p_tree_depth == 0)
2805 return 0;
2806
2807 blkno = path_leaf_bh(path)->b_blocknr;
2808
2809 /* Start at the tree node just above the leaf and work our way up. */
2810 i = path->p_tree_depth - 1;
2811 while (i >= 0) {
2812 int next_free;
2813
2814 el = path->p_node[i].el;
2815
2816 /*
2817 * Find the extent record just after the one in our
2818 * path.
2819 */
2820 next_free = le16_to_cpu(el->l_next_free_rec);
2821 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2822 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2823 if (j == (next_free - 1)) {
2824 if (i == 0) {
2825 /*
2826 * We've determined that the
2827 * path specified is already
2828 * the rightmost one - return a
2829 * cpos of zero.
2830 */
2831 goto out;
2832 }
2833 /*
2834 * The rightmost record points to our
2835 * leaf - we need to travel up the
2836 * tree one level.
2837 */
2838 goto next_node;
2839 }
2840
2841 *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2842 goto out;
2843 }
2844 }
2845
2846 /*
2847 * If we got here, we never found a valid node where
2848 * the tree indicated one should be.
2849 */
Joe Perches7ecef142015-09-04 15:44:51 -07002850 ocfs2_error(sb, "Invalid extent tree at extent block %llu\n",
Mark Fasheh328d5752007-06-18 10:48:04 -07002851 (unsigned long long)blkno);
2852 ret = -EROFS;
2853 goto out;
2854
2855next_node:
2856 blkno = path->p_node[i].bh->b_blocknr;
2857 i--;
2858 }
2859
2860out:
2861 return ret;
2862}
2863
Joel Becker70f18c02009-02-13 02:09:31 -08002864static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2865 struct ocfs2_extent_tree *et,
Joel Becker13723d02008-10-17 19:25:01 -07002866 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002867{
2868 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07002869 struct buffer_head *bh = path_leaf_bh(path);
2870 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002871
2872 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2873 return 0;
2874
Joel Becker70f18c02009-02-13 02:09:31 -08002875 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Joel Becker13723d02008-10-17 19:25:01 -07002876 path_num_items(path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07002877 if (ret) {
2878 mlog_errno(ret);
2879 goto out;
2880 }
2881
2882 ocfs2_remove_empty_extent(el);
Joel Beckerec20cec2010-03-19 14:13:52 -07002883 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002884
2885out:
2886 return ret;
2887}
2888
Joel Beckere46f74d2009-02-12 19:47:43 -08002889static int __ocfs2_rotate_tree_left(handle_t *handle,
2890 struct ocfs2_extent_tree *et,
2891 int orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07002892 struct ocfs2_path *path,
2893 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08002894 struct ocfs2_path **empty_extent_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002895{
2896 int ret, subtree_root, deleted;
2897 u32 right_cpos;
2898 struct ocfs2_path *left_path = NULL;
2899 struct ocfs2_path *right_path = NULL;
Joel Beckere46f74d2009-02-12 19:47:43 -08002900 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fasheh328d5752007-06-18 10:48:04 -07002901
Xue jiufei099768b2015-06-24 16:55:09 -07002902 if (!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])))
2903 return 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002904
2905 *empty_extent_path = NULL;
2906
Joel Beckere46f74d2009-02-12 19:47:43 -08002907 ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002908 if (ret) {
2909 mlog_errno(ret);
2910 goto out;
2911 }
2912
Joel Beckerffdd7a52008-10-17 22:32:01 -07002913 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002914 if (!left_path) {
2915 ret = -ENOMEM;
2916 mlog_errno(ret);
2917 goto out;
2918 }
2919
2920 ocfs2_cp_path(left_path, path);
2921
Joel Beckerffdd7a52008-10-17 22:32:01 -07002922 right_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002923 if (!right_path) {
2924 ret = -ENOMEM;
2925 mlog_errno(ret);
2926 goto out;
2927 }
2928
2929 while (right_cpos) {
Joel Beckerfacdb772009-02-12 18:08:48 -08002930 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002931 if (ret) {
2932 mlog_errno(ret);
2933 goto out;
2934 }
2935
Joel Becker7dc02802009-02-12 19:20:13 -08002936 subtree_root = ocfs2_find_subtree_root(et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002937 right_path);
2938
Tao Maa09d09b2011-02-22 08:21:10 +08002939 trace_ocfs2_rotate_subtree(subtree_root,
Mark Fasheh328d5752007-06-18 10:48:04 -07002940 (unsigned long long)
2941 right_path->p_node[subtree_root].bh->b_blocknr,
2942 right_path->p_tree_depth);
2943
Xue jiufei17215982016-03-25 14:21:41 -07002944 ret = ocfs2_extend_rotate_transaction(handle, 0,
Mark Fasheh328d5752007-06-18 10:48:04 -07002945 orig_credits, left_path);
2946 if (ret) {
2947 mlog_errno(ret);
2948 goto out;
2949 }
2950
Mark Fashehe8aed342007-12-03 16:43:01 -08002951 /*
2952 * Caller might still want to make changes to the
2953 * tree root, so re-add it to the journal here.
2954 */
Joel Beckere46f74d2009-02-12 19:47:43 -08002955 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002956 left_path, 0);
Mark Fashehe8aed342007-12-03 16:43:01 -08002957 if (ret) {
2958 mlog_errno(ret);
2959 goto out;
2960 }
2961
Joel Becker1e2dd632009-02-12 19:45:28 -08002962 ret = ocfs2_rotate_subtree_left(handle, et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002963 right_path, subtree_root,
Joel Becker1e2dd632009-02-12 19:45:28 -08002964 dealloc, &deleted);
Mark Fasheh328d5752007-06-18 10:48:04 -07002965 if (ret == -EAGAIN) {
2966 /*
2967 * The rotation has to temporarily stop due to
2968 * the right subtree having an empty
2969 * extent. Pass it back to the caller for a
2970 * fixup.
2971 */
2972 *empty_extent_path = right_path;
2973 right_path = NULL;
2974 goto out;
2975 }
2976 if (ret) {
2977 mlog_errno(ret);
2978 goto out;
2979 }
2980
2981 /*
2982 * The subtree rotate might have removed records on
2983 * the rightmost edge. If so, then rotation is
2984 * complete.
2985 */
2986 if (deleted)
2987 break;
2988
2989 ocfs2_mv_path(left_path, right_path);
2990
Joel Beckere46f74d2009-02-12 19:47:43 -08002991 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002992 &right_cpos);
2993 if (ret) {
2994 mlog_errno(ret);
2995 goto out;
2996 }
2997 }
2998
2999out:
3000 ocfs2_free_path(right_path);
3001 ocfs2_free_path(left_path);
3002
3003 return ret;
3004}
3005
Joel Becker70f18c02009-02-13 02:09:31 -08003006static int ocfs2_remove_rightmost_path(handle_t *handle,
3007 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003008 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003009 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003010{
3011 int ret, subtree_index;
3012 u32 cpos;
3013 struct ocfs2_path *left_path = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003014 struct ocfs2_extent_block *eb;
3015 struct ocfs2_extent_list *el;
3016
Joel Becker6136ca52009-02-12 19:32:43 -08003017 ret = ocfs2_et_sanity_check(et);
Tao Mae7d4cb62008-08-18 17:38:44 +08003018 if (ret)
3019 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003020
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003021 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003022 if (ret) {
3023 mlog_errno(ret);
3024 goto out;
3025 }
3026
Joel Becker3d03a302009-02-12 17:49:26 -08003027 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3028 path, &cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003029 if (ret) {
3030 mlog_errno(ret);
3031 goto out;
3032 }
3033
3034 if (cpos) {
3035 /*
3036 * We have a path to the left of this one - it needs
3037 * an update too.
3038 */
Joel Beckerffdd7a52008-10-17 22:32:01 -07003039 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003040 if (!left_path) {
3041 ret = -ENOMEM;
3042 mlog_errno(ret);
3043 goto out;
3044 }
3045
Joel Beckerfacdb772009-02-12 18:08:48 -08003046 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003047 if (ret) {
3048 mlog_errno(ret);
3049 goto out;
3050 }
3051
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003052 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003053 if (ret) {
3054 mlog_errno(ret);
3055 goto out;
3056 }
3057
Joel Becker7dc02802009-02-12 19:20:13 -08003058 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003059
Joel Becker6641b0c2009-02-12 18:57:52 -08003060 ocfs2_unlink_subtree(handle, et, left_path, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003061 subtree_index, dealloc);
Joel Becker09106ba2009-02-12 19:43:57 -08003062 ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
Tao Ma3c5e1062009-07-21 15:42:05 +08003063 left_path);
3064 if (ret) {
3065 mlog_errno(ret);
3066 goto out;
3067 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003068
3069 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07003070 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003071 } else {
3072 /*
3073 * 'path' is also the leftmost path which
3074 * means it must be the only one. This gets
3075 * handled differently because we want to
Joel Becker70f18c02009-02-13 02:09:31 -08003076 * revert the root back to having extents
Mark Fasheh328d5752007-06-18 10:48:04 -07003077 * in-line.
3078 */
Joel Becker6641b0c2009-02-12 18:57:52 -08003079 ocfs2_unlink_path(handle, et, dealloc, path, 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003080
Joel Beckerce1d9ea2008-08-20 16:30:07 -07003081 el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003082 el->l_tree_depth = 0;
3083 el->l_next_free_rec = 0;
3084 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3085
Joel Becker35dc0aa2008-08-20 16:25:06 -07003086 ocfs2_et_set_last_eb_blk(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003087 }
3088
3089 ocfs2_journal_dirty(handle, path_root_bh(path));
3090
3091out:
3092 ocfs2_free_path(left_path);
3093 return ret;
3094}
3095
Xue jiufeid0c97d52015-09-04 15:44:48 -07003096static int ocfs2_remove_rightmost_empty_extent(struct ocfs2_super *osb,
3097 struct ocfs2_extent_tree *et,
3098 struct ocfs2_path *path,
3099 struct ocfs2_cached_dealloc_ctxt *dealloc)
3100{
3101 handle_t *handle;
3102 int ret;
3103 int credits = path->p_tree_depth * 2 + 1;
3104
3105 handle = ocfs2_start_trans(osb, credits);
3106 if (IS_ERR(handle)) {
3107 ret = PTR_ERR(handle);
3108 mlog_errno(ret);
3109 return ret;
3110 }
3111
3112 ret = ocfs2_remove_rightmost_path(handle, et, path, dealloc);
3113 if (ret)
3114 mlog_errno(ret);
3115
3116 ocfs2_commit_trans(osb, handle);
3117 return ret;
3118}
3119
Mark Fasheh328d5752007-06-18 10:48:04 -07003120/*
3121 * Left rotation of btree records.
3122 *
3123 * In many ways, this is (unsurprisingly) the opposite of right
3124 * rotation. We start at some non-rightmost path containing an empty
3125 * extent in the leaf block. The code works its way to the rightmost
3126 * path by rotating records to the left in every subtree.
3127 *
3128 * This is used by any code which reduces the number of extent records
3129 * in a leaf. After removal, an empty record should be placed in the
3130 * leftmost list position.
3131 *
3132 * This won't handle a length update of the rightmost path records if
3133 * the rightmost tree leaf record is removed so the caller is
3134 * responsible for detecting and correcting that.
3135 */
Joel Becker70f18c02009-02-13 02:09:31 -08003136static int ocfs2_rotate_tree_left(handle_t *handle,
3137 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003138 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003139 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003140{
3141 int ret, orig_credits = handle->h_buffer_credits;
3142 struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3143 struct ocfs2_extent_block *eb;
3144 struct ocfs2_extent_list *el;
3145
3146 el = path_leaf_el(path);
3147 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3148 return 0;
3149
3150 if (path->p_tree_depth == 0) {
3151rightmost_no_delete:
3152 /*
Tao Mae7d4cb62008-08-18 17:38:44 +08003153 * Inline extents. This is trivially handled, so do
Mark Fasheh328d5752007-06-18 10:48:04 -07003154 * it up front.
3155 */
Joel Becker70f18c02009-02-13 02:09:31 -08003156 ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003157 if (ret)
3158 mlog_errno(ret);
3159 goto out;
3160 }
3161
3162 /*
3163 * Handle rightmost branch now. There's several cases:
3164 * 1) simple rotation leaving records in there. That's trivial.
3165 * 2) rotation requiring a branch delete - there's no more
3166 * records left. Two cases of this:
3167 * a) There are branches to the left.
3168 * b) This is also the leftmost (the only) branch.
3169 *
3170 * 1) is handled via ocfs2_rotate_rightmost_leaf_left()
3171 * 2a) we need the left branch so that we can update it with the unlink
Joel Becker70f18c02009-02-13 02:09:31 -08003172 * 2b) we need to bring the root back to inline extents.
Mark Fasheh328d5752007-06-18 10:48:04 -07003173 */
3174
3175 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3176 el = &eb->h_list;
3177 if (eb->h_next_leaf_blk == 0) {
3178 /*
3179 * This gets a bit tricky if we're going to delete the
3180 * rightmost path. Get the other cases out of the way
3181 * 1st.
3182 */
3183 if (le16_to_cpu(el->l_next_free_rec) > 1)
3184 goto rightmost_no_delete;
3185
3186 if (le16_to_cpu(el->l_next_free_rec) == 0) {
3187 ret = -EIO;
Joel Becker70f18c02009-02-13 02:09:31 -08003188 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07003189 "Owner %llu has empty extent block at %llu\n",
Joel Becker70f18c02009-02-13 02:09:31 -08003190 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07003191 (unsigned long long)le64_to_cpu(eb->h_blkno));
3192 goto out;
3193 }
3194
3195 /*
3196 * XXX: The caller can not trust "path" any more after
3197 * this as it will have been deleted. What do we do?
3198 *
3199 * In theory the rotate-for-merge code will never get
3200 * here because it'll always ask for a rotate in a
3201 * nonempty list.
3202 */
3203
Joel Becker70f18c02009-02-13 02:09:31 -08003204 ret = ocfs2_remove_rightmost_path(handle, et, path,
3205 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003206 if (ret)
3207 mlog_errno(ret);
3208 goto out;
3209 }
3210
3211 /*
3212 * Now we can loop, remembering the path we get from -EAGAIN
3213 * and restarting from there.
3214 */
3215try_rotate:
Joel Beckere46f74d2009-02-12 19:47:43 -08003216 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3217 dealloc, &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003218 if (ret && ret != -EAGAIN) {
3219 mlog_errno(ret);
3220 goto out;
3221 }
3222
3223 while (ret == -EAGAIN) {
3224 tmp_path = restart_path;
3225 restart_path = NULL;
3226
Joel Beckere46f74d2009-02-12 19:47:43 -08003227 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07003228 tmp_path, dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08003229 &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003230 if (ret && ret != -EAGAIN) {
3231 mlog_errno(ret);
3232 goto out;
3233 }
3234
3235 ocfs2_free_path(tmp_path);
3236 tmp_path = NULL;
3237
3238 if (ret == 0)
3239 goto try_rotate;
3240 }
3241
3242out:
3243 ocfs2_free_path(tmp_path);
3244 ocfs2_free_path(restart_path);
3245 return ret;
3246}
3247
3248static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3249 int index)
3250{
3251 struct ocfs2_extent_rec *rec = &el->l_recs[index];
3252 unsigned int size;
3253
3254 if (rec->e_leaf_clusters == 0) {
3255 /*
3256 * We consumed all of the merged-from record. An empty
3257 * extent cannot exist anywhere but the 1st array
3258 * position, so move things over if the merged-from
3259 * record doesn't occupy that position.
3260 *
3261 * This creates a new empty extent so the caller
3262 * should be smart enough to have removed any existing
3263 * ones.
3264 */
3265 if (index > 0) {
3266 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3267 size = index * sizeof(struct ocfs2_extent_rec);
3268 memmove(&el->l_recs[1], &el->l_recs[0], size);
3269 }
3270
3271 /*
3272 * Always memset - the caller doesn't check whether it
3273 * created an empty extent, so there could be junk in
3274 * the other fields.
3275 */
3276 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3277 }
3278}
3279
Joel Becker4fe82c32009-02-13 02:16:08 -08003280static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003281 struct ocfs2_path *left_path,
3282 struct ocfs2_path **ret_right_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07003283{
3284 int ret;
Tao Ma677b9752008-01-30 14:21:05 +08003285 u32 right_cpos;
3286 struct ocfs2_path *right_path = NULL;
3287 struct ocfs2_extent_list *left_el;
3288
3289 *ret_right_path = NULL;
3290
3291 /* This function shouldn't be called for non-trees. */
3292 BUG_ON(left_path->p_tree_depth == 0);
3293
3294 left_el = path_leaf_el(left_path);
3295 BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3296
Joel Becker4fe82c32009-02-13 02:16:08 -08003297 ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3298 left_path, &right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003299 if (ret) {
3300 mlog_errno(ret);
3301 goto out;
3302 }
3303
3304 /* This function shouldn't be called for the rightmost leaf. */
3305 BUG_ON(right_cpos == 0);
3306
Joel Beckerffdd7a52008-10-17 22:32:01 -07003307 right_path = ocfs2_new_path_from_path(left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003308 if (!right_path) {
3309 ret = -ENOMEM;
3310 mlog_errno(ret);
3311 goto out;
3312 }
3313
Joel Becker4fe82c32009-02-13 02:16:08 -08003314 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003315 if (ret) {
3316 mlog_errno(ret);
3317 goto out;
3318 }
3319
3320 *ret_right_path = right_path;
3321out:
3322 if (ret)
3323 ocfs2_free_path(right_path);
3324 return ret;
3325}
3326
3327/*
3328 * Remove split_rec clusters from the record at index and merge them
3329 * onto the beginning of the record "next" to it.
3330 * For index < l_count - 1, the next means the extent rec at index + 1.
3331 * For index == l_count - 1, the "next" means the 1st extent rec of the
3332 * next extent block.
3333 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003334static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003335 handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003336 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003337 struct ocfs2_extent_rec *split_rec,
3338 int index)
3339{
3340 int ret, next_free, i;
Mark Fasheh328d5752007-06-18 10:48:04 -07003341 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3342 struct ocfs2_extent_rec *left_rec;
3343 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003344 struct ocfs2_extent_list *right_el;
3345 struct ocfs2_path *right_path = NULL;
3346 int subtree_index = 0;
3347 struct ocfs2_extent_list *el = path_leaf_el(left_path);
3348 struct buffer_head *bh = path_leaf_bh(left_path);
3349 struct buffer_head *root_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003350
3351 BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
Mark Fasheh328d5752007-06-18 10:48:04 -07003352 left_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003353
Al Viro9d8df6a2008-05-21 06:32:11 +01003354 if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
Tao Ma677b9752008-01-30 14:21:05 +08003355 le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3356 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003357 ret = ocfs2_get_right_path(et, left_path, &right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003358 if (ret) {
3359 mlog_errno(ret);
Markus Elfring629a3b52015-04-14 15:42:53 -07003360 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003361 }
3362
3363 right_el = path_leaf_el(right_path);
3364 next_free = le16_to_cpu(right_el->l_next_free_rec);
3365 BUG_ON(next_free <= 0);
3366 right_rec = &right_el->l_recs[0];
3367 if (ocfs2_is_empty_extent(right_rec)) {
Al Viro9d8df6a2008-05-21 06:32:11 +01003368 BUG_ON(next_free <= 1);
Tao Ma677b9752008-01-30 14:21:05 +08003369 right_rec = &right_el->l_recs[1];
3370 }
3371
3372 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3373 le16_to_cpu(left_rec->e_leaf_clusters) !=
3374 le32_to_cpu(right_rec->e_cpos));
3375
Joel Becker7dc02802009-02-12 19:20:13 -08003376 subtree_index = ocfs2_find_subtree_root(et, left_path,
3377 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003378
3379 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3380 handle->h_buffer_credits,
3381 right_path);
3382 if (ret) {
3383 mlog_errno(ret);
3384 goto out;
3385 }
3386
3387 root_bh = left_path->p_node[subtree_index].bh;
3388 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3389
Joel Becker7dc02802009-02-12 19:20:13 -08003390 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003391 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003392 if (ret) {
3393 mlog_errno(ret);
3394 goto out;
3395 }
3396
3397 for (i = subtree_index + 1;
3398 i < path_num_items(right_path); i++) {
Joel Becker7dc02802009-02-12 19:20:13 -08003399 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003400 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003401 if (ret) {
3402 mlog_errno(ret);
3403 goto out;
3404 }
3405
Joel Becker7dc02802009-02-12 19:20:13 -08003406 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003407 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003408 if (ret) {
3409 mlog_errno(ret);
3410 goto out;
3411 }
3412 }
3413
3414 } else {
3415 BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3416 right_rec = &el->l_recs[index + 1];
3417 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003418
Joel Becker7dc02802009-02-12 19:20:13 -08003419 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
Joel Becker13723d02008-10-17 19:25:01 -07003420 path_num_items(left_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003421 if (ret) {
3422 mlog_errno(ret);
3423 goto out;
3424 }
3425
3426 le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3427
3428 le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3429 le64_add_cpu(&right_rec->e_blkno,
Joel Becker7dc02802009-02-12 19:20:13 -08003430 -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3431 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003432 le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3433
3434 ocfs2_cleanup_merge(el, index);
3435
Joel Beckerec20cec2010-03-19 14:13:52 -07003436 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003437 if (right_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003438 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Joel Becker4619c732009-02-12 19:02:36 -08003439 ocfs2_complete_edge_insert(handle, left_path, right_path,
3440 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003441 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003442out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003443 ocfs2_free_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003444 return ret;
3445}
3446
Joel Becker4fe82c32009-02-13 02:16:08 -08003447static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003448 struct ocfs2_path *right_path,
3449 struct ocfs2_path **ret_left_path)
3450{
3451 int ret;
3452 u32 left_cpos;
3453 struct ocfs2_path *left_path = NULL;
3454
3455 *ret_left_path = NULL;
3456
3457 /* This function shouldn't be called for non-trees. */
3458 BUG_ON(right_path->p_tree_depth == 0);
3459
Joel Becker4fe82c32009-02-13 02:16:08 -08003460 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003461 right_path, &left_cpos);
3462 if (ret) {
3463 mlog_errno(ret);
3464 goto out;
3465 }
3466
3467 /* This function shouldn't be called for the leftmost leaf. */
3468 BUG_ON(left_cpos == 0);
3469
Joel Beckerffdd7a52008-10-17 22:32:01 -07003470 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003471 if (!left_path) {
3472 ret = -ENOMEM;
3473 mlog_errno(ret);
3474 goto out;
3475 }
3476
Joel Becker4fe82c32009-02-13 02:16:08 -08003477 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003478 if (ret) {
3479 mlog_errno(ret);
3480 goto out;
3481 }
3482
3483 *ret_left_path = left_path;
3484out:
3485 if (ret)
3486 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003487 return ret;
3488}
3489
3490/*
3491 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003492 * onto the tail of the record "before" it.
3493 * For index > 0, the "before" means the extent rec at index - 1.
3494 *
3495 * For index == 0, the "before" means the last record of the previous
3496 * extent block. And there is also a situation that we may need to
3497 * remove the rightmost leaf extent block in the right_path and change
3498 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003499 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003500static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003501 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003502 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003503 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003504 struct ocfs2_cached_dealloc_ctxt *dealloc,
3505 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003506{
Tao Ma677b9752008-01-30 14:21:05 +08003507 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003508 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3509 struct ocfs2_extent_rec *left_rec;
3510 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003511 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3512 struct buffer_head *bh = path_leaf_bh(right_path);
3513 struct buffer_head *root_bh = NULL;
3514 struct ocfs2_path *left_path = NULL;
3515 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003516
Tao Ma677b9752008-01-30 14:21:05 +08003517 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003518
Mark Fasheh328d5752007-06-18 10:48:04 -07003519 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003520 if (index == 0) {
3521 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003522 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003523 if (ret) {
3524 mlog_errno(ret);
Markus Elfring992ef6e2015-04-14 15:42:51 -07003525 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003526 }
3527
3528 left_el = path_leaf_el(left_path);
3529 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3530 le16_to_cpu(left_el->l_count));
3531
3532 left_rec = &left_el->l_recs[
3533 le16_to_cpu(left_el->l_next_free_rec) - 1];
3534 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3535 le16_to_cpu(left_rec->e_leaf_clusters) !=
3536 le32_to_cpu(split_rec->e_cpos));
3537
Joel Becker7dc02802009-02-12 19:20:13 -08003538 subtree_index = ocfs2_find_subtree_root(et, left_path,
3539 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003540
3541 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3542 handle->h_buffer_credits,
3543 left_path);
3544 if (ret) {
3545 mlog_errno(ret);
3546 goto out;
3547 }
3548
3549 root_bh = left_path->p_node[subtree_index].bh;
3550 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3551
Joel Becker4fe82c32009-02-13 02:16:08 -08003552 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003553 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003554 if (ret) {
3555 mlog_errno(ret);
3556 goto out;
3557 }
3558
3559 for (i = subtree_index + 1;
3560 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003561 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003562 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003563 if (ret) {
3564 mlog_errno(ret);
3565 goto out;
3566 }
3567
Joel Becker4fe82c32009-02-13 02:16:08 -08003568 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003569 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003570 if (ret) {
3571 mlog_errno(ret);
3572 goto out;
3573 }
3574 }
3575 } else {
3576 left_rec = &el->l_recs[index - 1];
3577 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3578 has_empty_extent = 1;
3579 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003580
Joel Becker4fe82c32009-02-13 02:16:08 -08003581 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003582 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003583 if (ret) {
3584 mlog_errno(ret);
3585 goto out;
3586 }
3587
3588 if (has_empty_extent && index == 1) {
3589 /*
3590 * The easy case - we can just plop the record right in.
3591 */
3592 *left_rec = *split_rec;
3593
3594 has_empty_extent = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003595 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003596 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003597
3598 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3599 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003600 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3601 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003602 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3603
3604 ocfs2_cleanup_merge(el, index);
3605
Joel Beckerec20cec2010-03-19 14:13:52 -07003606 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003607 if (left_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003608 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
Tao Ma677b9752008-01-30 14:21:05 +08003609
3610 /*
3611 * In the situation that the right_rec is empty and the extent
3612 * block is empty also, ocfs2_complete_edge_insert can't handle
3613 * it and we need to delete the right extent block.
3614 */
3615 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3616 le16_to_cpu(el->l_next_free_rec) == 1) {
Xue jiufei17215982016-03-25 14:21:41 -07003617 /* extend credit for ocfs2_remove_rightmost_path */
3618 ret = ocfs2_extend_rotate_transaction(handle, 0,
3619 handle->h_buffer_credits,
3620 right_path);
3621 if (ret) {
3622 mlog_errno(ret);
3623 goto out;
3624 }
Tao Ma677b9752008-01-30 14:21:05 +08003625
Joel Becker70f18c02009-02-13 02:09:31 -08003626 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003627 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003628 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003629 if (ret) {
3630 mlog_errno(ret);
3631 goto out;
3632 }
3633
3634 /* Now the rightmost extent block has been deleted.
3635 * So we use the new rightmost path.
3636 */
3637 ocfs2_mv_path(right_path, left_path);
3638 left_path = NULL;
3639 } else
Joel Becker4619c732009-02-12 19:02:36 -08003640 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003641 right_path, subtree_index);
3642 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003643out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003644 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003645 return ret;
3646}
3647
Joel Beckerc495dd22009-02-13 02:19:11 -08003648static int ocfs2_try_to_merge_extent(handle_t *handle,
3649 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003650 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003651 int split_index,
3652 struct ocfs2_extent_rec *split_rec,
3653 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003654 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003655{
Tao Mao518d7262007-08-28 17:25:35 -07003656 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003657 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003658 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3659
3660 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3661
Tao Mao518d7262007-08-28 17:25:35 -07003662 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
Xue jiufei17215982016-03-25 14:21:41 -07003663 /* extend credit for ocfs2_remove_rightmost_path */
3664 ret = ocfs2_extend_rotate_transaction(handle, 0,
3665 handle->h_buffer_credits,
3666 path);
3667 if (ret) {
3668 mlog_errno(ret);
3669 goto out;
3670 }
Tao Mao518d7262007-08-28 17:25:35 -07003671 /*
3672 * The merge code will need to create an empty
3673 * extent to take the place of the newly
3674 * emptied slot. Remove any pre-existing empty
3675 * extents - having more than one in a leaf is
3676 * illegal.
3677 */
Joel Becker70f18c02009-02-13 02:09:31 -08003678 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003679 if (ret) {
3680 mlog_errno(ret);
3681 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003682 }
Tao Mao518d7262007-08-28 17:25:35 -07003683 split_index--;
3684 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003685 }
3686
3687 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3688 /*
3689 * Left-right contig implies this.
3690 */
3691 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003692
3693 /*
3694 * Since the leftright insert always covers the entire
3695 * extent, this call will delete the insert record
3696 * entirely, resulting in an empty extent record added to
3697 * the extent block.
3698 *
3699 * Since the adding of an empty extent shifts
3700 * everything back to the right, there's no need to
3701 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003702 *
3703 * When the split_index is zero, we need to merge it to the
3704 * prevoius extent block. It is more efficient and easier
3705 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003706 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003707 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003708 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003709 if (ret) {
3710 mlog_errno(ret);
3711 goto out;
3712 }
3713
3714 /*
3715 * We can only get this from logic error above.
3716 */
3717 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3718
Xue jiufei17215982016-03-25 14:21:41 -07003719 /* extend credit for ocfs2_remove_rightmost_path */
3720 ret = ocfs2_extend_rotate_transaction(handle, 0,
3721 handle->h_buffer_credits,
3722 path);
3723 if (ret) {
3724 mlog_errno(ret);
3725 goto out;
3726 }
3727
Tao Ma677b9752008-01-30 14:21:05 +08003728 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003729 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003730 if (ret) {
3731 mlog_errno(ret);
3732 goto out;
3733 }
Tao Ma677b9752008-01-30 14:21:05 +08003734
Mark Fasheh328d5752007-06-18 10:48:04 -07003735 rec = &el->l_recs[split_index];
3736
3737 /*
3738 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003739 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003740 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003741 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3742 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003743
Mark Fasheh328d5752007-06-18 10:48:04 -07003744 if (ret) {
3745 mlog_errno(ret);
3746 goto out;
3747 }
3748
Xue jiufei17215982016-03-25 14:21:41 -07003749 /* extend credit for ocfs2_remove_rightmost_path */
3750 ret = ocfs2_extend_rotate_transaction(handle, 0,
3751 handle->h_buffer_credits,
3752 path);
3753 if (ret) {
3754 mlog_errno(ret);
3755 goto out;
3756 }
3757
Joel Becker70f18c02009-02-13 02:09:31 -08003758 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003759 /*
3760 * Error from this last rotate is not critical, so
3761 * print but don't bubble it up.
3762 */
3763 if (ret)
3764 mlog_errno(ret);
3765 ret = 0;
3766 } else {
3767 /*
3768 * Merge a record to the left or right.
3769 *
3770 * 'contig_type' is relative to the existing record,
3771 * so for example, if we're "right contig", it's to
3772 * the record on the left (hence the left merge).
3773 */
3774 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003775 ret = ocfs2_merge_rec_left(path, handle, et,
3776 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003777 split_index);
3778 if (ret) {
3779 mlog_errno(ret);
3780 goto out;
3781 }
3782 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003783 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003784 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003785 split_index);
3786 if (ret) {
3787 mlog_errno(ret);
3788 goto out;
3789 }
3790 }
3791
3792 if (ctxt->c_split_covers_rec) {
Xue jiufei17215982016-03-25 14:21:41 -07003793 /* extend credit for ocfs2_remove_rightmost_path */
3794 ret = ocfs2_extend_rotate_transaction(handle, 0,
3795 handle->h_buffer_credits,
3796 path);
3797 if (ret) {
3798 mlog_errno(ret);
3799 ret = 0;
3800 goto out;
3801 }
3802
Mark Fasheh328d5752007-06-18 10:48:04 -07003803 /*
3804 * The merge may have left an empty extent in
3805 * our leaf. Try to rotate it away.
3806 */
Joel Becker70f18c02009-02-13 02:09:31 -08003807 ret = ocfs2_rotate_tree_left(handle, et, path,
3808 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003809 if (ret)
3810 mlog_errno(ret);
3811 ret = 0;
3812 }
3813 }
3814
3815out:
3816 return ret;
3817}
3818
3819static void ocfs2_subtract_from_rec(struct super_block *sb,
3820 enum ocfs2_split_type split,
3821 struct ocfs2_extent_rec *rec,
3822 struct ocfs2_extent_rec *split_rec)
3823{
3824 u64 len_blocks;
3825
3826 len_blocks = ocfs2_clusters_to_blocks(sb,
3827 le16_to_cpu(split_rec->e_leaf_clusters));
3828
3829 if (split == SPLIT_LEFT) {
3830 /*
3831 * Region is on the left edge of the existing
3832 * record.
3833 */
3834 le32_add_cpu(&rec->e_cpos,
3835 le16_to_cpu(split_rec->e_leaf_clusters));
3836 le64_add_cpu(&rec->e_blkno, len_blocks);
3837 le16_add_cpu(&rec->e_leaf_clusters,
3838 -le16_to_cpu(split_rec->e_leaf_clusters));
3839 } else {
3840 /*
3841 * Region is on the right edge of the existing
3842 * record.
3843 */
3844 le16_add_cpu(&rec->e_leaf_clusters,
3845 -le16_to_cpu(split_rec->e_leaf_clusters));
3846 }
3847}
3848
Mark Fashehdcd05382007-01-16 11:32:23 -08003849/*
3850 * Do the final bits of extent record insertion at the target leaf
3851 * list. If this leaf is part of an allocation tree, it is assumed
3852 * that the tree above has been prepared.
3853 */
Joel Beckerd5628622009-02-13 02:54:36 -08003854static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3855 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003856 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003857 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003858{
3859 int i = insert->ins_contig_index;
3860 unsigned int range;
3861 struct ocfs2_extent_rec *rec;
3862
Mark Fashehe48edee2007-03-07 16:46:57 -08003863 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003864
Mark Fasheh328d5752007-06-18 10:48:04 -07003865 if (insert->ins_split != SPLIT_NONE) {
3866 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3867 BUG_ON(i == -1);
3868 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003869 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3870 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003871 insert_rec);
3872 goto rotate;
3873 }
3874
Mark Fashehdcd05382007-01-16 11:32:23 -08003875 /*
3876 * Contiguous insert - either left or right.
3877 */
3878 if (insert->ins_contig != CONTIG_NONE) {
3879 rec = &el->l_recs[i];
3880 if (insert->ins_contig == CONTIG_LEFT) {
3881 rec->e_blkno = insert_rec->e_blkno;
3882 rec->e_cpos = insert_rec->e_cpos;
3883 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003884 le16_add_cpu(&rec->e_leaf_clusters,
3885 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003886 return;
3887 }
3888
3889 /*
3890 * Handle insert into an empty leaf.
3891 */
3892 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3893 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3894 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3895 el->l_recs[0] = *insert_rec;
3896 el->l_next_free_rec = cpu_to_le16(1);
3897 return;
3898 }
3899
3900 /*
3901 * Appending insert.
3902 */
3903 if (insert->ins_appending == APPEND_TAIL) {
3904 i = le16_to_cpu(el->l_next_free_rec) - 1;
3905 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003906 range = le32_to_cpu(rec->e_cpos)
3907 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003908 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3909
3910 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3911 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003912 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003913 "rec.cpos %u, rec.clusters %u, "
3914 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003915 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003916 le16_to_cpu(el->l_tree_depth),
3917 le16_to_cpu(el->l_count),
3918 le16_to_cpu(el->l_next_free_rec),
3919 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003920 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003921 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003922 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003923 i++;
3924 el->l_recs[i] = *insert_rec;
3925 le16_add_cpu(&el->l_next_free_rec, 1);
3926 return;
3927 }
3928
Mark Fasheh328d5752007-06-18 10:48:04 -07003929rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003930 /*
3931 * Ok, we have to rotate.
3932 *
3933 * At this point, it is safe to assume that inserting into an
3934 * empty leaf and appending to a leaf have both been handled
3935 * above.
3936 *
3937 * This leaf needs to have space, either by the empty 1st
3938 * extent record, or by virtue of an l_next_rec < l_count.
3939 */
3940 ocfs2_rotate_leaf(el, insert_rec);
3941}
3942
Joel Beckerd401dc12009-02-13 02:24:10 -08003943static void ocfs2_adjust_rightmost_records(handle_t *handle,
3944 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003945 struct ocfs2_path *path,
3946 struct ocfs2_extent_rec *insert_rec)
3947{
3948 int ret, i, next_free;
3949 struct buffer_head *bh;
3950 struct ocfs2_extent_list *el;
3951 struct ocfs2_extent_rec *rec;
3952
3953 /*
3954 * Update everything except the leaf block.
3955 */
3956 for (i = 0; i < path->p_tree_depth; i++) {
3957 bh = path->p_node[i].bh;
3958 el = path->p_node[i].el;
3959
3960 next_free = le16_to_cpu(el->l_next_free_rec);
3961 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003962 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07003963 "Owner %llu has a bad extent list\n",
Joel Beckerd401dc12009-02-13 02:24:10 -08003964 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003965 ret = -EIO;
3966 return;
3967 }
3968
3969 rec = &el->l_recs[next_free - 1];
3970
3971 rec->e_int_clusters = insert_rec->e_cpos;
3972 le32_add_cpu(&rec->e_int_clusters,
3973 le16_to_cpu(insert_rec->e_leaf_clusters));
3974 le32_add_cpu(&rec->e_int_clusters,
3975 -le32_to_cpu(rec->e_cpos));
3976
Joel Beckerec20cec2010-03-19 14:13:52 -07003977 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07003978 }
3979}
3980
Joel Beckerd401dc12009-02-13 02:24:10 -08003981static int ocfs2_append_rec_to_path(handle_t *handle,
3982 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003983 struct ocfs2_extent_rec *insert_rec,
3984 struct ocfs2_path *right_path,
3985 struct ocfs2_path **ret_left_path)
3986{
Mark Fasheh328d5752007-06-18 10:48:04 -07003987 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003988 struct ocfs2_extent_list *el;
3989 struct ocfs2_path *left_path = NULL;
3990
3991 *ret_left_path = NULL;
3992
3993 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003994 * This shouldn't happen for non-trees. The extent rec cluster
3995 * count manipulation below only works for interior nodes.
3996 */
3997 BUG_ON(right_path->p_tree_depth == 0);
3998
3999 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08004000 * If our appending insert is at the leftmost edge of a leaf,
4001 * then we might need to update the rightmost records of the
4002 * neighboring path.
4003 */
4004 el = path_leaf_el(right_path);
4005 next_free = le16_to_cpu(el->l_next_free_rec);
4006 if (next_free == 0 ||
4007 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
4008 u32 left_cpos;
4009
Joel Beckerd401dc12009-02-13 02:24:10 -08004010 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
4011 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004012 if (ret) {
4013 mlog_errno(ret);
4014 goto out;
4015 }
4016
Tao Maa09d09b2011-02-22 08:21:10 +08004017 trace_ocfs2_append_rec_to_path(
4018 (unsigned long long)
4019 ocfs2_metadata_cache_owner(et->et_ci),
4020 le32_to_cpu(insert_rec->e_cpos),
4021 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004022
4023 /*
4024 * No need to worry if the append is already in the
4025 * leftmost leaf.
4026 */
4027 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004028 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004029 if (!left_path) {
4030 ret = -ENOMEM;
4031 mlog_errno(ret);
4032 goto out;
4033 }
4034
Joel Beckerd401dc12009-02-13 02:24:10 -08004035 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004036 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004037 if (ret) {
4038 mlog_errno(ret);
4039 goto out;
4040 }
4041
4042 /*
4043 * ocfs2_insert_path() will pass the left_path to the
4044 * journal for us.
4045 */
4046 }
4047 }
4048
Joel Beckerd401dc12009-02-13 02:24:10 -08004049 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004050 if (ret) {
4051 mlog_errno(ret);
4052 goto out;
4053 }
4054
Joel Beckerd401dc12009-02-13 02:24:10 -08004055 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004056
4057 *ret_left_path = left_path;
4058 ret = 0;
4059out:
4060 if (ret != 0)
4061 ocfs2_free_path(left_path);
4062
4063 return ret;
4064}
4065
Joel Beckerc38e52b2009-02-13 02:56:23 -08004066static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004067 struct ocfs2_path *left_path,
4068 struct ocfs2_path *right_path,
4069 struct ocfs2_extent_rec *split_rec,
4070 enum ocfs2_split_type split)
4071{
4072 int index;
4073 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4074 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4075 struct ocfs2_extent_rec *rec, *tmprec;
4076
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004077 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004078 if (left_path)
4079 left_el = path_leaf_el(left_path);
4080
4081 el = right_el;
4082 insert_el = right_el;
4083 index = ocfs2_search_extent_list(el, cpos);
4084 if (index != -1) {
4085 if (index == 0 && left_path) {
4086 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4087
4088 /*
4089 * This typically means that the record
4090 * started in the left path but moved to the
4091 * right as a result of rotation. We either
4092 * move the existing record to the left, or we
4093 * do the later insert there.
4094 *
4095 * In this case, the left path should always
4096 * exist as the rotate code will have passed
4097 * it back for a post-insert update.
4098 */
4099
4100 if (split == SPLIT_LEFT) {
4101 /*
4102 * It's a left split. Since we know
4103 * that the rotate code gave us an
4104 * empty extent in the left path, we
4105 * can just do the insert there.
4106 */
4107 insert_el = left_el;
4108 } else {
4109 /*
4110 * Right split - we have to move the
4111 * existing record over to the left
4112 * leaf. The insert will be into the
4113 * newly created empty extent in the
4114 * right leaf.
4115 */
4116 tmprec = &right_el->l_recs[index];
4117 ocfs2_rotate_leaf(left_el, tmprec);
4118 el = left_el;
4119
4120 memset(tmprec, 0, sizeof(*tmprec));
4121 index = ocfs2_search_extent_list(left_el, cpos);
4122 BUG_ON(index == -1);
4123 }
4124 }
4125 } else {
4126 BUG_ON(!left_path);
4127 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4128 /*
4129 * Left path is easy - we can just allow the insert to
4130 * happen.
4131 */
4132 el = left_el;
4133 insert_el = left_el;
4134 index = ocfs2_search_extent_list(el, cpos);
4135 BUG_ON(index == -1);
4136 }
4137
4138 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004139 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4140 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004141 ocfs2_rotate_leaf(insert_el, split_rec);
4142}
4143
Mark Fashehdcd05382007-01-16 11:32:23 -08004144/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004145 * This function only does inserts on an allocation b-tree. For tree
4146 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004147 *
4148 * right_path is the path we want to do the actual insert
4149 * in. left_path should only be passed in if we need to update that
4150 * portion of the tree after an edge insert.
4151 */
Joel Becker3505bec2009-02-13 02:57:58 -08004152static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004153 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004154 struct ocfs2_path *left_path,
4155 struct ocfs2_path *right_path,
4156 struct ocfs2_extent_rec *insert_rec,
4157 struct ocfs2_insert_type *insert)
4158{
4159 int ret, subtree_index;
4160 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004161
Mark Fashehdcd05382007-01-16 11:32:23 -08004162 if (left_path) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004163 /*
4164 * There's a chance that left_path got passed back to
4165 * us without being accounted for in the
4166 * journal. Extend our transaction here to be sure we
4167 * can change those blocks.
4168 */
Tao Mac901fb02010-04-26 14:34:57 +08004169 ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08004170 if (ret < 0) {
4171 mlog_errno(ret);
4172 goto out;
4173 }
4174
Joel Becker7dc02802009-02-12 19:20:13 -08004175 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004176 if (ret < 0) {
4177 mlog_errno(ret);
4178 goto out;
4179 }
4180 }
4181
Mark Fashehe8aed342007-12-03 16:43:01 -08004182 /*
4183 * Pass both paths to the journal. The majority of inserts
4184 * will be touching all components anyway.
4185 */
Joel Becker7dc02802009-02-12 19:20:13 -08004186 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004187 if (ret < 0) {
4188 mlog_errno(ret);
4189 goto out;
4190 }
4191
Mark Fasheh328d5752007-06-18 10:48:04 -07004192 if (insert->ins_split != SPLIT_NONE) {
4193 /*
4194 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004195 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004196 * function sort it all out.
4197 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004198 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004199 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004200
4201 /*
4202 * Split might have modified either leaf and we don't
4203 * have a guarantee that the later edge insert will
4204 * dirty this for us.
4205 */
4206 if (left_path)
Joel Beckerec20cec2010-03-19 14:13:52 -07004207 ocfs2_journal_dirty(handle,
4208 path_leaf_bh(left_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07004209 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004210 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4211 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004212
Joel Beckerec20cec2010-03-19 14:13:52 -07004213 ocfs2_journal_dirty(handle, leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004214
4215 if (left_path) {
4216 /*
4217 * The rotate code has indicated that we need to fix
4218 * up portions of the tree after the insert.
4219 *
4220 * XXX: Should we extend the transaction here?
4221 */
Joel Becker7dc02802009-02-12 19:20:13 -08004222 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004223 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004224 ocfs2_complete_edge_insert(handle, left_path, right_path,
4225 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004226 }
4227
4228 ret = 0;
4229out:
4230 return ret;
4231}
4232
Joel Becker3505bec2009-02-13 02:57:58 -08004233static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004234 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004235 struct ocfs2_extent_rec *insert_rec,
4236 struct ocfs2_insert_type *type)
4237{
4238 int ret, rotate = 0;
4239 u32 cpos;
4240 struct ocfs2_path *right_path = NULL;
4241 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004242 struct ocfs2_extent_list *el;
4243
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004244 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004245
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004246 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004247 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004248 if (ret) {
4249 mlog_errno(ret);
4250 goto out;
4251 }
4252
4253 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004254 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004255 goto out_update_clusters;
4256 }
4257
Joel Beckerffdd7a52008-10-17 22:32:01 -07004258 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004259 if (!right_path) {
4260 ret = -ENOMEM;
4261 mlog_errno(ret);
4262 goto out;
4263 }
4264
4265 /*
4266 * Determine the path to start with. Rotations need the
4267 * rightmost path, everything else can go directly to the
4268 * target leaf.
4269 */
4270 cpos = le32_to_cpu(insert_rec->e_cpos);
4271 if (type->ins_appending == APPEND_NONE &&
4272 type->ins_contig == CONTIG_NONE) {
4273 rotate = 1;
4274 cpos = UINT_MAX;
4275 }
4276
Joel Beckerfacdb772009-02-12 18:08:48 -08004277 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004278 if (ret) {
4279 mlog_errno(ret);
4280 goto out;
4281 }
4282
4283 /*
4284 * Rotations and appends need special treatment - they modify
4285 * parts of the tree's above them.
4286 *
4287 * Both might pass back a path immediate to the left of the
4288 * one being inserted to. This will be cause
4289 * ocfs2_insert_path() to modify the rightmost records of
4290 * left_path to account for an edge insert.
4291 *
4292 * XXX: When modifying this code, keep in mind that an insert
4293 * can wind up skipping both of these two special cases...
4294 */
4295 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004296 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004297 le32_to_cpu(insert_rec->e_cpos),
4298 right_path, &left_path);
4299 if (ret) {
4300 mlog_errno(ret);
4301 goto out;
4302 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004303
4304 /*
4305 * ocfs2_rotate_tree_right() might have extended the
4306 * transaction without re-journaling our tree root.
4307 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004308 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004309 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004310 if (ret) {
4311 mlog_errno(ret);
4312 goto out;
4313 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004314 } else if (type->ins_appending == APPEND_TAIL
4315 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004316 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004317 right_path, &left_path);
4318 if (ret) {
4319 mlog_errno(ret);
4320 goto out;
4321 }
4322 }
4323
Joel Becker3505bec2009-02-13 02:57:58 -08004324 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004325 insert_rec, type);
4326 if (ret) {
4327 mlog_errno(ret);
4328 goto out;
4329 }
4330
4331out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004332 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004333 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004334 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004335
Joel Beckerec20cec2010-03-19 14:13:52 -07004336 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004337
4338out:
4339 ocfs2_free_path(left_path);
4340 ocfs2_free_path(right_path);
4341
4342 return ret;
4343}
4344
Xue jiufei9f99ad02015-06-24 16:55:07 -07004345static int ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
Joel Beckera2970292009-02-13 03:09:54 -08004346 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004347 struct ocfs2_extent_list *el, int index,
Xue jiufei9f99ad02015-06-24 16:55:07 -07004348 struct ocfs2_extent_rec *split_rec,
4349 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07004350{
Xue jiufei9f99ad02015-06-24 16:55:07 -07004351 int status = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07004352 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004353 u32 left_cpos, right_cpos;
4354 struct ocfs2_extent_rec *rec = NULL;
4355 struct ocfs2_extent_list *new_el;
4356 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4357 struct buffer_head *bh;
4358 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004359 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004360
4361 if (index > 0) {
4362 rec = &el->l_recs[index - 1];
4363 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004364 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004365 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004366 goto exit;
Tao Maad5a4d72008-01-30 14:21:32 +08004367
4368 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004369 left_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004370 if (!left_path) {
4371 status = -ENOMEM;
4372 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004373 goto exit;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004374 }
Tao Maad5a4d72008-01-30 14:21:32 +08004375
Joel Beckera2970292009-02-13 03:09:54 -08004376 status = ocfs2_find_path(et->et_ci, left_path,
4377 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004378 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004379 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004380
4381 new_el = path_leaf_el(left_path);
4382
4383 if (le16_to_cpu(new_el->l_next_free_rec) !=
4384 le16_to_cpu(new_el->l_count)) {
4385 bh = path_leaf_bh(left_path);
4386 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004387 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07004388 "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 -08004389 (unsigned long long)le64_to_cpu(eb->h_blkno),
4390 le16_to_cpu(new_el->l_next_free_rec),
4391 le16_to_cpu(new_el->l_count));
4392 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004393 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004394 }
4395 rec = &new_el->l_recs[
4396 le16_to_cpu(new_el->l_next_free_rec) - 1];
4397 }
4398 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004399
4400 /*
4401 * We're careful to check for an empty extent record here -
4402 * the merge code will know what to do if it sees one.
4403 */
Tao Maad5a4d72008-01-30 14:21:32 +08004404 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004405 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4406 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4407 ret = CONTIG_RIGHT;
4408 } else {
Tao Ma853a3a12009-08-18 11:22:18 +08004409 ret = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004410 }
4411 }
4412
Tao Maad5a4d72008-01-30 14:21:32 +08004413 rec = NULL;
4414 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4415 rec = &el->l_recs[index + 1];
4416 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4417 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004418 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004419 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004420 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004421
4422 if (right_cpos == 0)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004423 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004424
Joel Beckerffdd7a52008-10-17 22:32:01 -07004425 right_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004426 if (!right_path) {
4427 status = -ENOMEM;
4428 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004429 goto free_left_path;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004430 }
Tao Maad5a4d72008-01-30 14:21:32 +08004431
Joel Beckera2970292009-02-13 03:09:54 -08004432 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004433 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004434 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004435
4436 new_el = path_leaf_el(right_path);
4437 rec = &new_el->l_recs[0];
4438 if (ocfs2_is_empty_extent(rec)) {
4439 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4440 bh = path_leaf_bh(right_path);
4441 eb = (struct ocfs2_extent_block *)bh->b_data;
Joel Beckera2970292009-02-13 03:09:54 -08004442 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07004443 "Extent block #%llu has an invalid l_next_free_rec of %d\n",
Joel Becker5e965812008-11-13 14:49:16 -08004444 (unsigned long long)le64_to_cpu(eb->h_blkno),
4445 le16_to_cpu(new_el->l_next_free_rec));
4446 status = -EINVAL;
Markus Elfring06a269cc2015-04-14 15:42:48 -07004447 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004448 }
4449 rec = &new_el->l_recs[1];
4450 }
4451 }
4452
4453 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004454 enum ocfs2_contig_type contig_type;
4455
Tao Ma853a3a12009-08-18 11:22:18 +08004456 contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004457
4458 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4459 ret = CONTIG_LEFTRIGHT;
4460 else if (ret == CONTIG_NONE)
4461 ret = contig_type;
4462 }
4463
Markus Elfring06a269cc2015-04-14 15:42:48 -07004464free_right_path:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07004465 ocfs2_free_path(right_path);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004466free_left_path:
4467 ocfs2_free_path(left_path);
4468exit:
Xue jiufei9f99ad02015-06-24 16:55:07 -07004469 if (status == 0)
4470 ctxt->c_contig_type = ret;
4471
4472 return status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004473}
4474
Joel Becker1ef61b32009-02-13 03:12:33 -08004475static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004476 struct ocfs2_insert_type *insert,
4477 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004478 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004479{
4480 int i;
4481 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4482
Mark Fashehe48edee2007-03-07 16:46:57 -08004483 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4484
Mark Fashehdcd05382007-01-16 11:32:23 -08004485 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Tao Ma853a3a12009-08-18 11:22:18 +08004486 contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4487 insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004488 if (contig_type != CONTIG_NONE) {
4489 insert->ins_contig_index = i;
4490 break;
4491 }
4492 }
4493 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004494
4495 if (insert->ins_contig != CONTIG_NONE) {
4496 struct ocfs2_extent_rec *rec =
4497 &el->l_recs[insert->ins_contig_index];
4498 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4499 le16_to_cpu(insert_rec->e_leaf_clusters);
4500
4501 /*
4502 * Caller might want us to limit the size of extents, don't
4503 * calculate contiguousness if we might exceed that limit.
4504 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004505 if (et->et_max_leaf_clusters &&
4506 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004507 insert->ins_contig = CONTIG_NONE;
4508 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004509}
4510
4511/*
4512 * This should only be called against the righmost leaf extent list.
4513 *
4514 * ocfs2_figure_appending_type() will figure out whether we'll have to
4515 * insert at the tail of the rightmost leaf.
4516 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004517 * This should also work against the root extent list for tree's with 0
4518 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004519 * then the logic here makes sense.
4520 */
4521static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4522 struct ocfs2_extent_list *el,
4523 struct ocfs2_extent_rec *insert_rec)
4524{
4525 int i;
4526 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4527 struct ocfs2_extent_rec *rec;
4528
4529 insert->ins_appending = APPEND_NONE;
4530
Mark Fashehe48edee2007-03-07 16:46:57 -08004531 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004532
4533 if (!el->l_next_free_rec)
4534 goto set_tail_append;
4535
4536 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4537 /* Were all records empty? */
4538 if (le16_to_cpu(el->l_next_free_rec) == 1)
4539 goto set_tail_append;
4540 }
4541
4542 i = le16_to_cpu(el->l_next_free_rec) - 1;
4543 rec = &el->l_recs[i];
4544
Mark Fashehe48edee2007-03-07 16:46:57 -08004545 if (cpos >=
4546 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004547 goto set_tail_append;
4548
4549 return;
4550
4551set_tail_append:
4552 insert->ins_appending = APPEND_TAIL;
4553}
4554
4555/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004556 * Helper function called at the beginning of an insert.
Mark Fashehdcd05382007-01-16 11:32:23 -08004557 *
4558 * This computes a few things that are commonly used in the process of
4559 * inserting into the btree:
4560 * - Whether the new extent is contiguous with an existing one.
4561 * - The current tree depth.
4562 * - Whether the insert is an appending one.
4563 * - The total # of free records in the tree.
4564 *
4565 * All of the information is stored on the ocfs2_insert_type
4566 * structure.
4567 */
Joel Becker627961b2009-02-13 03:14:38 -08004568static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004569 struct buffer_head **last_eb_bh,
4570 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004571 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004572 struct ocfs2_insert_type *insert)
4573{
4574 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004575 struct ocfs2_extent_block *eb;
4576 struct ocfs2_extent_list *el;
4577 struct ocfs2_path *path = NULL;
4578 struct buffer_head *bh = NULL;
4579
Mark Fasheh328d5752007-06-18 10:48:04 -07004580 insert->ins_split = SPLIT_NONE;
4581
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004582 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004583 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4584
4585 if (el->l_tree_depth) {
4586 /*
4587 * If we have tree depth, we read in the
4588 * rightmost extent block ahead of time as
4589 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4590 * may want it later.
4591 */
Joel Becker3d03a302009-02-12 17:49:26 -08004592 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004593 ocfs2_et_get_last_eb_blk(et),
4594 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004595 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08004596 mlog_errno(ret);
Mark Fashehdcd05382007-01-16 11:32:23 -08004597 goto out;
4598 }
4599 eb = (struct ocfs2_extent_block *) bh->b_data;
4600 el = &eb->h_list;
4601 }
4602
4603 /*
4604 * Unless we have a contiguous insert, we'll need to know if
4605 * there is room left in our allocation tree for another
4606 * extent record.
4607 *
4608 * XXX: This test is simplistic, we can search for empty
4609 * extent records too.
4610 */
Tao Maoc77534f2007-08-28 17:22:33 -07004611 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004612 le16_to_cpu(el->l_next_free_rec);
4613
4614 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004615 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004616 ocfs2_figure_appending_type(insert, el, insert_rec);
4617 return 0;
4618 }
4619
Joel Beckerffdd7a52008-10-17 22:32:01 -07004620 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004621 if (!path) {
4622 ret = -ENOMEM;
4623 mlog_errno(ret);
4624 goto out;
4625 }
4626
4627 /*
4628 * In the case that we're inserting past what the tree
4629 * currently accounts for, ocfs2_find_path() will return for
4630 * us the rightmost tree path. This is accounted for below in
4631 * the appending code.
4632 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004633 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004634 if (ret) {
4635 mlog_errno(ret);
4636 goto out;
4637 }
4638
4639 el = path_leaf_el(path);
4640
4641 /*
4642 * Now that we have the path, there's two things we want to determine:
4643 * 1) Contiguousness (also set contig_index if this is so)
4644 *
4645 * 2) Are we doing an append? We can trivially break this up
4646 * into two types of appends: simple record append, or a
4647 * rotate inside the tail leaf.
4648 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004649 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004650
4651 /*
4652 * The insert code isn't quite ready to deal with all cases of
4653 * left contiguousness. Specifically, if it's an insert into
4654 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004655 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004656 * left. For now, just catch that case and fool the layers
4657 * above us. This works just fine for tree_depth == 0, which
4658 * is why we allow that above.
4659 */
4660 if (insert->ins_contig == CONTIG_LEFT &&
4661 insert->ins_contig_index == 0)
4662 insert->ins_contig = CONTIG_NONE;
4663
4664 /*
4665 * Ok, so we can simply compare against last_eb to figure out
4666 * whether the path doesn't exist. This will only happen in
4667 * the case that we're doing a tail append, so maybe we can
4668 * take advantage of that information somehow.
4669 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004670 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004671 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004672 /*
4673 * Ok, ocfs2_find_path() returned us the rightmost
4674 * tree path. This might be an appending insert. There are
4675 * two cases:
4676 * 1) We're doing a true append at the tail:
4677 * -This might even be off the end of the leaf
4678 * 2) We're "appending" by rotating in the tail
4679 */
4680 ocfs2_figure_appending_type(insert, el, insert_rec);
4681 }
4682
4683out:
4684 ocfs2_free_path(path);
4685
4686 if (ret == 0)
4687 *last_eb_bh = bh;
4688 else
4689 brelse(bh);
4690 return ret;
4691}
4692
4693/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004694 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004695 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004696 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004697 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004698int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004699 struct ocfs2_extent_tree *et,
4700 u32 cpos,
4701 u64 start_blk,
4702 u32 new_clusters,
4703 u8 flags,
4704 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004705{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004706 int status;
Tao Maoc77534f2007-08-28 17:22:33 -07004707 int uninitialized_var(free_records);
Mark Fashehccd979b2005-12-15 14:31:24 -08004708 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004709 struct ocfs2_insert_type insert = {0, };
4710 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004711
Tao Maa09d09b2011-02-22 08:21:10 +08004712 trace_ocfs2_insert_extent_start(
4713 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4714 cpos, new_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08004715
Mark Fashehe48edee2007-03-07 16:46:57 -08004716 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004717 rec.e_cpos = cpu_to_le32(cpos);
4718 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004719 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004720 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004721 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004722 if (status) {
4723 mlog_errno(status);
4724 goto bail;
4725 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004726
Joel Becker627961b2009-02-13 03:14:38 -08004727 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004728 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004729 if (status < 0) {
4730 mlog_errno(status);
4731 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004732 }
4733
Tao Maa09d09b2011-02-22 08:21:10 +08004734 trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4735 insert.ins_contig_index, free_records,
4736 insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004737
Tao Maoc77534f2007-08-28 17:22:33 -07004738 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004739 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004740 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004741 meta_ac);
4742 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004743 mlog_errno(status);
4744 goto bail;
4745 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004746 }
4747
Mark Fashehdcd05382007-01-16 11:32:23 -08004748 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004749 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004750 if (status < 0)
4751 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004752 else
4753 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004754
4755bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004756 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004757
Tao Maf56654c2008-08-18 17:38:48 +08004758 return status;
4759}
4760
Tao Ma0eb8d472008-08-18 17:38:45 +08004761/*
4762 * Allcate and add clusters into the extent b-tree.
4763 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004764 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d472008-08-18 17:38:45 +08004765 * it is not limited to the file storage. Any extent tree can use this
4766 * function if it implements the proper ocfs2_extent_tree.
4767 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004768int ocfs2_add_clusters_in_btree(handle_t *handle,
4769 struct ocfs2_extent_tree *et,
Tao Ma0eb8d472008-08-18 17:38:45 +08004770 u32 *logical_offset,
4771 u32 clusters_to_add,
4772 int mark_unwritten,
Tao Ma0eb8d472008-08-18 17:38:45 +08004773 struct ocfs2_alloc_context *data_ac,
4774 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004775 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d472008-08-18 17:38:45 +08004776{
Tao Maa09d09b2011-02-22 08:21:10 +08004777 int status = 0, err = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004778 int need_free = 0;
Tao Ma0eb8d472008-08-18 17:38:45 +08004779 int free_extents;
4780 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4781 u32 bit_off, num_bits;
4782 u64 block;
4783 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004784 struct ocfs2_super *osb =
4785 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d472008-08-18 17:38:45 +08004786
4787 BUG_ON(!clusters_to_add);
4788
4789 if (mark_unwritten)
4790 flags = OCFS2_EXT_UNWRITTEN;
4791
Joel Becker3d03a302009-02-12 17:49:26 -08004792 free_extents = ocfs2_num_free_extents(osb, et);
Tao Ma0eb8d472008-08-18 17:38:45 +08004793 if (free_extents < 0) {
4794 status = free_extents;
4795 mlog_errno(status);
4796 goto leave;
4797 }
4798
4799 /* there are two cases which could cause us to EAGAIN in the
4800 * we-need-more-metadata case:
4801 * 1) we haven't reserved *any*
4802 * 2) we are so fragmented, we've needed to add metadata too
4803 * many times. */
4804 if (!free_extents && !meta_ac) {
Tao Maa09d09b2011-02-22 08:21:10 +08004805 err = -1;
Tao Ma0eb8d472008-08-18 17:38:45 +08004806 status = -EAGAIN;
4807 reason = RESTART_META;
4808 goto leave;
4809 } else if ((!free_extents)
4810 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004811 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Maa09d09b2011-02-22 08:21:10 +08004812 err = -2;
Tao Ma0eb8d472008-08-18 17:38:45 +08004813 status = -EAGAIN;
4814 reason = RESTART_META;
4815 goto leave;
4816 }
4817
Joel Becker1ed9b772010-05-06 13:59:06 +08004818 status = __ocfs2_claim_clusters(handle, data_ac, 1,
Tao Ma0eb8d472008-08-18 17:38:45 +08004819 clusters_to_add, &bit_off, &num_bits);
4820 if (status < 0) {
4821 if (status != -ENOSPC)
4822 mlog_errno(status);
4823 goto leave;
4824 }
4825
4826 BUG_ON(num_bits > clusters_to_add);
4827
Joel Becker13723d02008-10-17 19:25:01 -07004828 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004829 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004830 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d472008-08-18 17:38:45 +08004831 if (status < 0) {
4832 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004833 need_free = 1;
4834 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004835 }
4836
4837 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Maa09d09b2011-02-22 08:21:10 +08004838 trace_ocfs2_add_clusters_in_btree(
4839 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4840 bit_off, num_bits);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004841 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004842 num_bits, flags, meta_ac);
Tao Ma0eb8d472008-08-18 17:38:45 +08004843 if (status < 0) {
4844 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004845 need_free = 1;
4846 goto bail;
Tao Ma0eb8d472008-08-18 17:38:45 +08004847 }
4848
Joel Beckerec20cec2010-03-19 14:13:52 -07004849 ocfs2_journal_dirty(handle, et->et_root_bh);
Tao Ma0eb8d472008-08-18 17:38:45 +08004850
4851 clusters_to_add -= num_bits;
4852 *logical_offset += num_bits;
4853
4854 if (clusters_to_add) {
Tao Maa09d09b2011-02-22 08:21:10 +08004855 err = clusters_to_add;
Tao Ma0eb8d472008-08-18 17:38:45 +08004856 status = -EAGAIN;
4857 reason = RESTART_TRANS;
4858 }
4859
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004860bail:
4861 if (need_free) {
4862 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
4863 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
4864 bit_off, num_bits);
4865 else
4866 ocfs2_free_clusters(handle,
4867 data_ac->ac_inode,
4868 data_ac->ac_bh,
4869 ocfs2_clusters_to_blocks(osb->sb, bit_off),
4870 num_bits);
4871 }
4872
Tao Ma0eb8d472008-08-18 17:38:45 +08004873leave:
Tao Ma0eb8d472008-08-18 17:38:45 +08004874 if (reason_ret)
4875 *reason_ret = reason;
Tao Maa09d09b2011-02-22 08:21:10 +08004876 trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
Tao Ma0eb8d472008-08-18 17:38:45 +08004877 return status;
4878}
4879
Mark Fasheh328d5752007-06-18 10:48:04 -07004880static void ocfs2_make_right_split_rec(struct super_block *sb,
4881 struct ocfs2_extent_rec *split_rec,
4882 u32 cpos,
4883 struct ocfs2_extent_rec *rec)
4884{
4885 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4886 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4887
4888 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4889
4890 split_rec->e_cpos = cpu_to_le32(cpos);
4891 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4892
4893 split_rec->e_blkno = rec->e_blkno;
4894 le64_add_cpu(&split_rec->e_blkno,
4895 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4896
4897 split_rec->e_flags = rec->e_flags;
4898}
4899
Joel Beckerd2311292009-02-13 03:43:22 -08004900static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004901 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004902 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004903 struct buffer_head **last_eb_bh,
4904 int split_index,
4905 struct ocfs2_extent_rec *orig_split_rec,
4906 struct ocfs2_alloc_context *meta_ac)
4907{
4908 int ret = 0, depth;
4909 unsigned int insert_range, rec_range, do_leftright = 0;
4910 struct ocfs2_extent_rec tmprec;
4911 struct ocfs2_extent_list *rightmost_el;
4912 struct ocfs2_extent_rec rec;
4913 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4914 struct ocfs2_insert_type insert;
4915 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004916
4917leftright:
4918 /*
4919 * Store a copy of the record on the stack - it might move
4920 * around as the tree is manipulated below.
4921 */
4922 rec = path_leaf_el(path)->l_recs[split_index];
4923
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004924 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004925
4926 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4927 if (depth) {
4928 BUG_ON(!(*last_eb_bh));
4929 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4930 rightmost_el = &eb->h_list;
4931 }
4932
4933 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4934 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004935 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004936 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004937 if (ret) {
4938 mlog_errno(ret);
4939 goto out;
4940 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004941 }
4942
4943 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4944 insert.ins_appending = APPEND_NONE;
4945 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004946 insert.ins_tree_depth = depth;
4947
4948 insert_range = le32_to_cpu(split_rec.e_cpos) +
4949 le16_to_cpu(split_rec.e_leaf_clusters);
4950 rec_range = le32_to_cpu(rec.e_cpos) +
4951 le16_to_cpu(rec.e_leaf_clusters);
4952
4953 if (split_rec.e_cpos == rec.e_cpos) {
4954 insert.ins_split = SPLIT_LEFT;
4955 } else if (insert_range == rec_range) {
4956 insert.ins_split = SPLIT_RIGHT;
4957 } else {
4958 /*
4959 * Left/right split. We fake this as a right split
4960 * first and then make a second pass as a left split.
4961 */
4962 insert.ins_split = SPLIT_RIGHT;
4963
Joel Beckerd2311292009-02-13 03:43:22 -08004964 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4965 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004966
4967 split_rec = tmprec;
4968
4969 BUG_ON(do_leftright);
4970 do_leftright = 1;
4971 }
4972
Joel Becker3505bec2009-02-13 02:57:58 -08004973 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004974 if (ret) {
4975 mlog_errno(ret);
4976 goto out;
4977 }
4978
4979 if (do_leftright == 1) {
4980 u32 cpos;
4981 struct ocfs2_extent_list *el;
4982
4983 do_leftright++;
4984 split_rec = *orig_split_rec;
4985
4986 ocfs2_reinit_path(path, 1);
4987
4988 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004989 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004990 if (ret) {
4991 mlog_errno(ret);
4992 goto out;
4993 }
4994
4995 el = path_leaf_el(path);
4996 split_index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07004997 if (split_index == -1) {
4998 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07004999 "Owner %llu has an extent at cpos %u which can no longer be found\n",
5000 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5001 cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005002 ret = -EROFS;
5003 goto out;
5004 }
Mark Fasheh328d5752007-06-18 10:48:04 -07005005 goto leftright;
5006 }
5007out:
5008
5009 return ret;
5010}
5011
Joel Beckerf3868d02009-02-17 19:46:04 -08005012static int ocfs2_replace_extent_rec(handle_t *handle,
5013 struct ocfs2_extent_tree *et,
Tao Ma47be12e2009-01-09 07:32:48 +08005014 struct ocfs2_path *path,
5015 struct ocfs2_extent_list *el,
5016 int split_index,
5017 struct ocfs2_extent_rec *split_rec)
5018{
5019 int ret;
5020
Joel Beckerf3868d02009-02-17 19:46:04 -08005021 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Tao Ma47be12e2009-01-09 07:32:48 +08005022 path_num_items(path) - 1);
5023 if (ret) {
5024 mlog_errno(ret);
5025 goto out;
5026 }
5027
5028 el->l_recs[split_index] = *split_rec;
5029
5030 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5031out:
5032 return ret;
5033}
5034
Mark Fasheh328d5752007-06-18 10:48:04 -07005035/*
Tao Ma555936b2009-08-18 11:22:21 +08005036 * Split part or all of the extent record at split_index in the leaf
5037 * pointed to by path. Merge with the contiguous extent record if needed.
Mark Fasheh328d5752007-06-18 10:48:04 -07005038 *
5039 * Care is taken to handle contiguousness so as to not grow the tree.
5040 *
5041 * meta_ac is not strictly necessary - we only truly need it if growth
5042 * of the tree is required. All other cases will degrade into a less
5043 * optimal tree layout.
5044 *
Tao Mae7d4cb62008-08-18 17:38:44 +08005045 * last_eb_bh should be the rightmost leaf block for any extent
5046 * btree. Since a split may grow the tree or a merge might shrink it,
5047 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07005048 *
5049 * This code is optimized for readability - several passes might be
5050 * made over certain portions of the tree. All of those blocks will
5051 * have been brought into cache (and pinned via the journal), so the
5052 * extra overhead is not expressed in terms of disk reads.
5053 */
Tao Mae2e9f602009-08-18 11:22:34 +08005054int ocfs2_split_extent(handle_t *handle,
5055 struct ocfs2_extent_tree *et,
5056 struct ocfs2_path *path,
5057 int split_index,
5058 struct ocfs2_extent_rec *split_rec,
5059 struct ocfs2_alloc_context *meta_ac,
5060 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005061{
5062 int ret = 0;
5063 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005064 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005065 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5066 struct ocfs2_merge_ctxt ctxt;
5067 struct ocfs2_extent_list *rightmost_el;
5068
Mark Fasheh328d5752007-06-18 10:48:04 -07005069 if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5070 ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5071 (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5072 ret = -EIO;
5073 mlog_errno(ret);
5074 goto out;
5075 }
5076
Xue jiufei9f99ad02015-06-24 16:55:07 -07005077 ret = ocfs2_figure_merge_contig_type(et, path, el,
5078 split_index,
5079 split_rec,
5080 &ctxt);
5081 if (ret) {
5082 mlog_errno(ret);
5083 goto out;
5084 }
Mark Fasheh328d5752007-06-18 10:48:04 -07005085
5086 /*
5087 * The core merge / split code wants to know how much room is
Joel Beckera1cf0762009-02-13 03:45:49 -08005088 * left in this allocation tree, so we pass the
Mark Fasheh328d5752007-06-18 10:48:04 -07005089 * rightmost extent list.
5090 */
5091 if (path->p_tree_depth) {
5092 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07005093
Joel Becker3d03a302009-02-12 17:49:26 -08005094 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005095 ocfs2_et_get_last_eb_blk(et),
5096 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005097 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08005098 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07005099 goto out;
5100 }
5101
5102 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
Mark Fasheh328d5752007-06-18 10:48:04 -07005103 rightmost_el = &eb->h_list;
5104 } else
5105 rightmost_el = path_root_el(path);
5106
Mark Fasheh328d5752007-06-18 10:48:04 -07005107 if (rec->e_cpos == split_rec->e_cpos &&
5108 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5109 ctxt.c_split_covers_rec = 1;
5110 else
5111 ctxt.c_split_covers_rec = 0;
5112
5113 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5114
Tao Maa09d09b2011-02-22 08:21:10 +08005115 trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5116 ctxt.c_has_empty_extent,
5117 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005118
5119 if (ctxt.c_contig_type == CONTIG_NONE) {
5120 if (ctxt.c_split_covers_rec)
Joel Beckerf3868d02009-02-17 19:46:04 -08005121 ret = ocfs2_replace_extent_rec(handle, et, path, el,
Tao Ma47be12e2009-01-09 07:32:48 +08005122 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005123 else
Joel Beckerd2311292009-02-13 03:43:22 -08005124 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005125 &last_eb_bh, split_index,
5126 split_rec, meta_ac);
5127 if (ret)
5128 mlog_errno(ret);
5129 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005130 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005131 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005132 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005133 if (ret)
5134 mlog_errno(ret);
5135 }
5136
Mark Fasheh328d5752007-06-18 10:48:04 -07005137out:
5138 brelse(last_eb_bh);
5139 return ret;
5140}
5141
5142/*
Tao Ma555936b2009-08-18 11:22:21 +08005143 * Change the flags of the already-existing extent at cpos for len clusters.
5144 *
5145 * new_flags: the flags we want to set.
5146 * clear_flags: the flags we want to clear.
5147 * phys: the new physical offset we want this new extent starts from.
Mark Fasheh328d5752007-06-18 10:48:04 -07005148 *
5149 * If the existing extent is larger than the request, initiate a
5150 * split. An attempt will be made at merging with adjacent extents.
5151 *
5152 * The caller is responsible for passing down meta_ac if we'll need it.
5153 */
Tao Ma1aa75fe2009-08-18 11:28:39 +08005154int ocfs2_change_extent_flag(handle_t *handle,
5155 struct ocfs2_extent_tree *et,
5156 u32 cpos, u32 len, u32 phys,
5157 struct ocfs2_alloc_context *meta_ac,
5158 struct ocfs2_cached_dealloc_ctxt *dealloc,
5159 int new_flags, int clear_flags)
Mark Fasheh328d5752007-06-18 10:48:04 -07005160{
5161 int ret, index;
Tao Ma555936b2009-08-18 11:22:21 +08005162 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5163 u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
Mark Fasheh328d5752007-06-18 10:48:04 -07005164 struct ocfs2_extent_rec split_rec;
5165 struct ocfs2_path *left_path = NULL;
5166 struct ocfs2_extent_list *el;
Tao Ma555936b2009-08-18 11:22:21 +08005167 struct ocfs2_extent_rec *rec;
Mark Fasheh328d5752007-06-18 10:48:04 -07005168
Joel Beckerffdd7a52008-10-17 22:32:01 -07005169 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005170 if (!left_path) {
5171 ret = -ENOMEM;
5172 mlog_errno(ret);
5173 goto out;
5174 }
5175
Joel Beckerfacdb772009-02-12 18:08:48 -08005176 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005177 if (ret) {
5178 mlog_errno(ret);
5179 goto out;
5180 }
5181 el = path_leaf_el(left_path);
5182
5183 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005184 if (index == -1) {
Tao Ma555936b2009-08-18 11:22:21 +08005185 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07005186 "Owner %llu has an extent at cpos %u which can no longer be found\n",
5187 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5188 cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005189 ret = -EROFS;
5190 goto out;
5191 }
5192
Tao Ma555936b2009-08-18 11:22:21 +08005193 ret = -EIO;
5194 rec = &el->l_recs[index];
5195 if (new_flags && (rec->e_flags & new_flags)) {
5196 mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
5197 "extent that already had them",
5198 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5199 new_flags);
5200 goto out;
5201 }
5202
5203 if (clear_flags && !(rec->e_flags & clear_flags)) {
5204 mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
5205 "extent that didn't have them",
5206 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5207 clear_flags);
5208 goto out;
5209 }
5210
Mark Fasheh328d5752007-06-18 10:48:04 -07005211 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5212 split_rec.e_cpos = cpu_to_le32(cpos);
5213 split_rec.e_leaf_clusters = cpu_to_le16(len);
5214 split_rec.e_blkno = cpu_to_le64(start_blkno);
Tao Ma555936b2009-08-18 11:22:21 +08005215 split_rec.e_flags = rec->e_flags;
5216 if (new_flags)
5217 split_rec.e_flags |= new_flags;
5218 if (clear_flags)
5219 split_rec.e_flags &= ~clear_flags;
Mark Fasheh328d5752007-06-18 10:48:04 -07005220
Tao Mae2e9f602009-08-18 11:22:34 +08005221 ret = ocfs2_split_extent(handle, et, left_path,
5222 index, &split_rec, meta_ac,
5223 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005224 if (ret)
5225 mlog_errno(ret);
5226
5227out:
5228 ocfs2_free_path(left_path);
5229 return ret;
Tao Ma555936b2009-08-18 11:22:21 +08005230
5231}
5232
5233/*
5234 * Mark the already-existing extent at cpos as written for len clusters.
5235 * This removes the unwritten extent flag.
5236 *
5237 * If the existing extent is larger than the request, initiate a
5238 * split. An attempt will be made at merging with adjacent extents.
5239 *
5240 * The caller is responsible for passing down meta_ac if we'll need it.
5241 */
5242int ocfs2_mark_extent_written(struct inode *inode,
5243 struct ocfs2_extent_tree *et,
5244 handle_t *handle, u32 cpos, u32 len, u32 phys,
5245 struct ocfs2_alloc_context *meta_ac,
5246 struct ocfs2_cached_dealloc_ctxt *dealloc)
5247{
5248 int ret;
5249
Tao Maa09d09b2011-02-22 08:21:10 +08005250 trace_ocfs2_mark_extent_written(
5251 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5252 cpos, len, phys);
Tao Ma555936b2009-08-18 11:22:21 +08005253
5254 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
Joe Perches7ecef142015-09-04 15:44:51 -07005255 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 +08005256 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5257 ret = -EROFS;
5258 goto out;
5259 }
5260
5261 /*
5262 * XXX: This should be fixed up so that we just re-insert the
5263 * next extent records.
5264 */
5265 ocfs2_et_extent_map_truncate(et, 0);
5266
5267 ret = ocfs2_change_extent_flag(handle, et, cpos,
5268 len, phys, meta_ac, dealloc,
5269 0, OCFS2_EXT_UNWRITTEN);
5270 if (ret)
5271 mlog_errno(ret);
5272
5273out:
5274 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07005275}
5276
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005277static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5278 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005279 int index, u32 new_range,
5280 struct ocfs2_alloc_context *meta_ac)
5281{
Tao Mac901fb02010-04-26 14:34:57 +08005282 int ret, depth, credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005283 struct buffer_head *last_eb_bh = NULL;
5284 struct ocfs2_extent_block *eb;
5285 struct ocfs2_extent_list *rightmost_el, *el;
5286 struct ocfs2_extent_rec split_rec;
5287 struct ocfs2_extent_rec *rec;
5288 struct ocfs2_insert_type insert;
5289
5290 /*
5291 * Setup the record to split before we grow the tree.
5292 */
5293 el = path_leaf_el(path);
5294 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005295 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5296 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005297
5298 depth = path->p_tree_depth;
5299 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005300 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005301 ocfs2_et_get_last_eb_blk(et),
5302 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005303 if (ret < 0) {
5304 mlog_errno(ret);
5305 goto out;
5306 }
5307
5308 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5309 rightmost_el = &eb->h_list;
5310 } else
5311 rightmost_el = path_leaf_el(path);
5312
Tao Mac901fb02010-04-26 14:34:57 +08005313 credits = path->p_tree_depth +
5314 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005315 ret = ocfs2_extend_trans(handle, credits);
5316 if (ret) {
5317 mlog_errno(ret);
5318 goto out;
5319 }
5320
5321 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5322 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005323 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005324 meta_ac);
5325 if (ret) {
5326 mlog_errno(ret);
5327 goto out;
5328 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005329 }
5330
5331 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5332 insert.ins_appending = APPEND_NONE;
5333 insert.ins_contig = CONTIG_NONE;
5334 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005335 insert.ins_tree_depth = depth;
5336
Joel Becker3505bec2009-02-13 02:57:58 -08005337 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005338 if (ret)
5339 mlog_errno(ret);
5340
5341out:
5342 brelse(last_eb_bh);
5343 return ret;
5344}
5345
Joel Becker043beeb2009-02-13 02:42:30 -08005346static int ocfs2_truncate_rec(handle_t *handle,
5347 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005348 struct ocfs2_path *path, int index,
5349 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005350 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005351{
5352 int ret;
5353 u32 left_cpos, rec_range, trunc_range;
piaojunc14688e2016-05-19 17:09:47 -07005354 int is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005355 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005356 struct ocfs2_path *left_path = NULL;
5357 struct ocfs2_extent_list *el = path_leaf_el(path);
5358 struct ocfs2_extent_rec *rec;
5359 struct ocfs2_extent_block *eb;
5360
5361 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Xue jiufei17215982016-03-25 14:21:41 -07005362 /* extend credit for ocfs2_remove_rightmost_path */
5363 ret = ocfs2_extend_rotate_transaction(handle, 0,
5364 handle->h_buffer_credits,
5365 path);
5366 if (ret) {
5367 mlog_errno(ret);
5368 goto out;
5369 }
5370
Joel Becker70f18c02009-02-13 02:09:31 -08005371 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005372 if (ret) {
5373 mlog_errno(ret);
5374 goto out;
5375 }
5376
5377 index--;
5378 }
5379
5380 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5381 path->p_tree_depth) {
5382 /*
5383 * Check whether this is the rightmost tree record. If
5384 * we remove all of this record or part of its right
5385 * edge then an update of the record lengths above it
5386 * will be required.
5387 */
5388 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5389 if (eb->h_next_leaf_blk == 0)
5390 is_rightmost_tree_rec = 1;
5391 }
5392
5393 rec = &el->l_recs[index];
5394 if (index == 0 && path->p_tree_depth &&
5395 le32_to_cpu(rec->e_cpos) == cpos) {
5396 /*
5397 * Changing the leftmost offset (via partial or whole
5398 * record truncate) of an interior (or rightmost) path
5399 * means we have to update the subtree that is formed
5400 * by this leaf and the one to it's left.
5401 *
5402 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005403 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005404 * 2) The leaf is rightmost and will be empty after
5405 * we remove the extent record - the rotate code
5406 * knows how to update the newly formed edge.
5407 */
5408
Joel Becker043beeb2009-02-13 02:42:30 -08005409 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005410 if (ret) {
5411 mlog_errno(ret);
5412 goto out;
5413 }
5414
5415 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005416 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005417 if (!left_path) {
5418 ret = -ENOMEM;
5419 mlog_errno(ret);
5420 goto out;
5421 }
5422
Joel Beckerfacdb772009-02-12 18:08:48 -08005423 ret = ocfs2_find_path(et->et_ci, left_path,
5424 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005425 if (ret) {
5426 mlog_errno(ret);
5427 goto out;
5428 }
5429 }
5430 }
5431
5432 ret = ocfs2_extend_rotate_transaction(handle, 0,
5433 handle->h_buffer_credits,
5434 path);
5435 if (ret) {
5436 mlog_errno(ret);
5437 goto out;
5438 }
5439
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005440 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005441 if (ret) {
5442 mlog_errno(ret);
5443 goto out;
5444 }
5445
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005446 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005447 if (ret) {
5448 mlog_errno(ret);
5449 goto out;
5450 }
5451
5452 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5453 trunc_range = cpos + len;
5454
5455 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5456 int next_free;
5457
5458 memset(rec, 0, sizeof(*rec));
5459 ocfs2_cleanup_merge(el, index);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005460
5461 next_free = le16_to_cpu(el->l_next_free_rec);
5462 if (is_rightmost_tree_rec && next_free > 1) {
5463 /*
5464 * We skip the edge update if this path will
5465 * be deleted by the rotate code.
5466 */
5467 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005468 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005469 rec);
5470 }
5471 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5472 /* Remove leftmost portion of the record. */
5473 le32_add_cpu(&rec->e_cpos, len);
5474 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5475 le16_add_cpu(&rec->e_leaf_clusters, -len);
5476 } else if (rec_range == trunc_range) {
5477 /* Remove rightmost portion of the record */
5478 le16_add_cpu(&rec->e_leaf_clusters, -len);
5479 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005480 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005481 } else {
5482 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005483 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5484 "(%u, %u)\n",
5485 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005486 le32_to_cpu(rec->e_cpos),
5487 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5488 BUG();
5489 }
5490
5491 if (left_path) {
5492 int subtree_index;
5493
Joel Becker7dc02802009-02-12 19:20:13 -08005494 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005495 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005496 subtree_index);
5497 }
5498
5499 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5500
Joel Becker70f18c02009-02-13 02:09:31 -08005501 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005502 if (ret) {
5503 mlog_errno(ret);
5504 goto out;
5505 }
5506
5507out:
5508 ocfs2_free_path(left_path);
5509 return ret;
5510}
5511
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005512int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005513 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005514 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005515 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005516 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005517{
5518 int ret, index;
5519 u32 rec_range, trunc_range;
5520 struct ocfs2_extent_rec *rec;
5521 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005522 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005523
Joel Becker4c911ee2009-02-13 02:50:12 -08005524 /*
5525 * XXX: Why are we truncating to 0 instead of wherever this
5526 * affects us?
5527 */
5528 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005529
Joel Beckerffdd7a52008-10-17 22:32:01 -07005530 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005531 if (!path) {
5532 ret = -ENOMEM;
5533 mlog_errno(ret);
5534 goto out;
5535 }
5536
Joel Beckerfacdb772009-02-12 18:08:48 -08005537 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005538 if (ret) {
5539 mlog_errno(ret);
5540 goto out;
5541 }
5542
5543 el = path_leaf_el(path);
5544 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005545 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005546 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005547 "Owner %llu has an extent at cpos %u which can no longer be found\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005548 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5549 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005550 ret = -EROFS;
5551 goto out;
5552 }
5553
5554 /*
5555 * We have 3 cases of extent removal:
5556 * 1) Range covers the entire extent rec
5557 * 2) Range begins or ends on one edge of the extent rec
5558 * 3) Range is in the middle of the extent rec (no shared edges)
5559 *
5560 * For case 1 we remove the extent rec and left rotate to
5561 * fill the hole.
5562 *
5563 * For case 2 we just shrink the existing extent rec, with a
5564 * tree update if the shrinking edge is also the edge of an
5565 * extent block.
5566 *
5567 * For case 3 we do a right split to turn the extent rec into
5568 * something case 2 can handle.
5569 */
5570 rec = &el->l_recs[index];
5571 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5572 trunc_range = cpos + len;
5573
5574 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5575
Tao Maa09d09b2011-02-22 08:21:10 +08005576 trace_ocfs2_remove_extent(
5577 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5578 cpos, len, index, le32_to_cpu(rec->e_cpos),
5579 ocfs2_rec_clusters(el, rec));
Mark Fashehd0c7d702007-07-03 13:27:22 -07005580
5581 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005582 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5583 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005584 if (ret) {
5585 mlog_errno(ret);
5586 goto out;
5587 }
5588 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005589 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005590 trunc_range, meta_ac);
5591 if (ret) {
5592 mlog_errno(ret);
5593 goto out;
5594 }
5595
5596 /*
5597 * The split could have manipulated the tree enough to
5598 * move the record location, so we have to look for it again.
5599 */
5600 ocfs2_reinit_path(path, 1);
5601
Joel Beckerfacdb772009-02-12 18:08:48 -08005602 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005603 if (ret) {
5604 mlog_errno(ret);
5605 goto out;
5606 }
5607
5608 el = path_leaf_el(path);
5609 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005610 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005611 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005612 "Owner %llu: split at cpos %u lost record\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005613 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005614 cpos);
5615 ret = -EROFS;
5616 goto out;
5617 }
5618
5619 /*
5620 * Double check our values here. If anything is fishy,
5621 * it's easier to catch it at the top level.
5622 */
5623 rec = &el->l_recs[index];
5624 rec_range = le32_to_cpu(rec->e_cpos) +
5625 ocfs2_rec_clusters(el, rec);
5626 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005627 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005628 "Owner %llu: error after split at cpos %u trunc len %u, existing record is (%u,%u)\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005629 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005630 cpos, len, le32_to_cpu(rec->e_cpos),
5631 ocfs2_rec_clusters(el, rec));
5632 ret = -EROFS;
5633 goto out;
5634 }
5635
Joel Becker043beeb2009-02-13 02:42:30 -08005636 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5637 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005638 if (ret) {
5639 mlog_errno(ret);
5640 goto out;
5641 }
5642 }
5643
5644out:
5645 ocfs2_free_path(path);
5646 return ret;
5647}
5648
Tristan Ye78f94672010-05-11 17:54:42 +08005649/*
5650 * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5651 * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5652 * number to reserve some extra blocks, and it only handles meta
5653 * data allocations.
5654 *
5655 * Currently, only ocfs2_remove_btree_range() uses it for truncating
5656 * and punching holes.
5657 */
5658static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5659 struct ocfs2_extent_tree *et,
5660 u32 extents_to_split,
5661 struct ocfs2_alloc_context **ac,
5662 int extra_blocks)
5663{
5664 int ret = 0, num_free_extents;
5665 unsigned int max_recs_needed = 2 * extents_to_split;
5666 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5667
5668 *ac = NULL;
5669
5670 num_free_extents = ocfs2_num_free_extents(osb, et);
5671 if (num_free_extents < 0) {
5672 ret = num_free_extents;
5673 mlog_errno(ret);
5674 goto out;
5675 }
5676
5677 if (!num_free_extents ||
5678 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5679 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5680
5681 if (extra_blocks) {
5682 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5683 if (ret < 0) {
5684 if (ret != -ENOSPC)
5685 mlog_errno(ret);
5686 goto out;
5687 }
5688 }
5689
5690out:
5691 if (ret) {
5692 if (*ac) {
5693 ocfs2_free_alloc_context(*ac);
5694 *ac = NULL;
5695 }
5696 }
5697
5698 return ret;
5699}
5700
Mark Fashehfecc0112008-11-12 15:16:38 -08005701int ocfs2_remove_btree_range(struct inode *inode,
5702 struct ocfs2_extent_tree *et,
Tristan Ye78f94672010-05-11 17:54:42 +08005703 u32 cpos, u32 phys_cpos, u32 len, int flags,
5704 struct ocfs2_cached_dealloc_ctxt *dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08005705 u64 refcount_loc, bool refcount_tree_locked)
Mark Fashehfecc0112008-11-12 15:16:38 -08005706{
Tristan Ye78f94672010-05-11 17:54:42 +08005707 int ret, credits = 0, extra_blocks = 0;
Mark Fashehfecc0112008-11-12 15:16:38 -08005708 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5709 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5710 struct inode *tl_inode = osb->osb_tl_inode;
5711 handle_t *handle;
5712 struct ocfs2_alloc_context *meta_ac = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08005713 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehfecc0112008-11-12 15:16:38 -08005714
Tristan Ye78f94672010-05-11 17:54:42 +08005715 if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
5716 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
5717 OCFS2_HAS_REFCOUNT_FL));
5718
Junxiao Bif62f12b2014-12-18 16:17:32 -08005719 if (!refcount_tree_locked) {
5720 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5721 &ref_tree, NULL);
5722 if (ret) {
5723 mlog_errno(ret);
5724 goto bail;
5725 }
Tristan Ye78f94672010-05-11 17:54:42 +08005726 }
5727
5728 ret = ocfs2_prepare_refcount_change_for_del(inode,
5729 refcount_loc,
5730 phys_blkno,
5731 len,
5732 &credits,
5733 &extra_blocks);
5734 if (ret < 0) {
5735 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005736 goto bail;
Tristan Ye78f94672010-05-11 17:54:42 +08005737 }
5738 }
5739
5740 ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5741 extra_blocks);
Mark Fashehfecc0112008-11-12 15:16:38 -08005742 if (ret) {
5743 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005744 goto bail;
Mark Fashehfecc0112008-11-12 15:16:38 -08005745 }
5746
Al Viro59551022016-01-22 15:40:57 -05005747 inode_lock(tl_inode);
Mark Fashehfecc0112008-11-12 15:16:38 -08005748
5749 if (ocfs2_truncate_log_needs_flush(osb)) {
5750 ret = __ocfs2_flush_truncate_log(osb);
5751 if (ret < 0) {
5752 mlog_errno(ret);
5753 goto out;
5754 }
5755 }
5756
Tristan Ye78f94672010-05-11 17:54:42 +08005757 handle = ocfs2_start_trans(osb,
5758 ocfs2_remove_extent_credits(osb->sb) + credits);
Mark Fashehfecc0112008-11-12 15:16:38 -08005759 if (IS_ERR(handle)) {
5760 ret = PTR_ERR(handle);
5761 mlog_errno(ret);
5762 goto out;
5763 }
5764
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005765 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005766 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005767 if (ret) {
5768 mlog_errno(ret);
Wengang Wangb8a0ae52011-10-12 15:22:15 +08005769 goto out_commit;
Mark Fashehfecc0112008-11-12 15:16:38 -08005770 }
5771
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005772 dquot_free_space_nodirty(inode,
Mark Fashehfd4ef232009-01-29 15:06:21 -08005773 ocfs2_clusters_to_bytes(inode->i_sb, len));
5774
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005775 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005776 if (ret) {
5777 mlog_errno(ret);
5778 goto out_commit;
5779 }
5780
Joel Becker6136ca52009-02-12 19:32:43 -08005781 ocfs2_et_update_clusters(et, -len);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07005782 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fashehfecc0112008-11-12 15:16:38 -08005783
Joel Beckerec20cec2010-03-19 14:13:52 -07005784 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehfecc0112008-11-12 15:16:38 -08005785
Tristan Ye78f94672010-05-11 17:54:42 +08005786 if (phys_blkno) {
5787 if (flags & OCFS2_EXT_REFCOUNTED)
5788 ret = ocfs2_decrease_refcount(inode, handle,
5789 ocfs2_blocks_to_clusters(osb->sb,
5790 phys_blkno),
5791 len, meta_ac,
5792 dealloc, 1);
5793 else
5794 ret = ocfs2_truncate_log_append(osb, handle,
5795 phys_blkno, len);
5796 if (ret)
5797 mlog_errno(ret);
5798
5799 }
Mark Fashehfecc0112008-11-12 15:16:38 -08005800
5801out_commit:
5802 ocfs2_commit_trans(osb, handle);
5803out:
Al Viro59551022016-01-22 15:40:57 -05005804 inode_unlock(tl_inode);
Joseph Qi33add0e2013-07-03 15:00:52 -07005805bail:
Mark Fashehfecc0112008-11-12 15:16:38 -08005806 if (meta_ac)
5807 ocfs2_free_alloc_context(meta_ac);
5808
Tristan Ye78f94672010-05-11 17:54:42 +08005809 if (ref_tree)
5810 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5811
Mark Fashehfecc0112008-11-12 15:16:38 -08005812 return ret;
5813}
5814
Mark Fasheh063c4562007-07-03 13:34:11 -07005815int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005816{
5817 struct buffer_head *tl_bh = osb->osb_tl_bh;
5818 struct ocfs2_dinode *di;
5819 struct ocfs2_truncate_log *tl;
5820
5821 di = (struct ocfs2_dinode *) tl_bh->b_data;
5822 tl = &di->id2.i_dealloc;
5823
5824 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5825 "slot %d, invalid truncate log parameters: used = "
5826 "%u, count = %u\n", osb->slot_num,
5827 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5828 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5829}
5830
5831static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5832 unsigned int new_start)
5833{
5834 unsigned int tail_index;
5835 unsigned int current_tail;
5836
5837 /* No records, nothing to coalesce */
5838 if (!le16_to_cpu(tl->tl_used))
5839 return 0;
5840
5841 tail_index = le16_to_cpu(tl->tl_used) - 1;
5842 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5843 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5844
5845 return current_tail == new_start;
5846}
5847
Mark Fasheh063c4562007-07-03 13:34:11 -07005848int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5849 handle_t *handle,
5850 u64 start_blk,
5851 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005852{
5853 int status, index;
5854 unsigned int start_cluster, tl_count;
5855 struct inode *tl_inode = osb->osb_tl_inode;
5856 struct buffer_head *tl_bh = osb->osb_tl_bh;
5857 struct ocfs2_dinode *di;
5858 struct ocfs2_truncate_log *tl;
5859
Al Viro59551022016-01-22 15:40:57 -05005860 BUG_ON(inode_trylock(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08005861
5862 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5863
5864 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005865
Joel Becker10995aa2008-11-13 14:49:12 -08005866 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5867 * by the underlying call to ocfs2_read_inode_block(), so any
5868 * corruption is a code bug */
5869 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5870
5871 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005872 tl_count = le16_to_cpu(tl->tl_count);
5873 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5874 tl_count == 0,
Mark Fashehb06970532006-03-03 10:24:33 -08005875 "Truncate record count on #%llu invalid "
5876 "wanted %u, actual %u\n",
5877 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005878 ocfs2_truncate_recs_per_inode(osb->sb),
5879 le16_to_cpu(tl->tl_count));
5880
5881 /* Caller should have known to flush before calling us. */
5882 index = le16_to_cpu(tl->tl_used);
5883 if (index >= tl_count) {
5884 status = -ENOSPC;
5885 mlog_errno(status);
5886 goto bail;
5887 }
5888
Joel Becker0cf2f762009-02-12 16:41:25 -08005889 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005890 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005891 if (status < 0) {
5892 mlog_errno(status);
5893 goto bail;
5894 }
5895
Tao Maa09d09b2011-02-22 08:21:10 +08005896 trace_ocfs2_truncate_log_append(
5897 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5898 start_cluster, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005899 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5900 /*
5901 * Move index back to the record we are coalescing with.
5902 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5903 */
5904 index--;
5905
5906 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
Tao Maa09d09b2011-02-22 08:21:10 +08005907 trace_ocfs2_truncate_log_append(
5908 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5909 index, le32_to_cpu(tl->tl_recs[index].t_start),
5910 num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005911 } else {
5912 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5913 tl->tl_used = cpu_to_le16(index + 1);
5914 }
5915 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5916
Joel Beckerec20cec2010-03-19 14:13:52 -07005917 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005918
Tao Ma50308d82010-11-04 15:14:11 +08005919 osb->truncated_clusters += num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08005920bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005921 return status;
5922}
5923
5924static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fasheh1fabe142006-10-09 18:11:45 -07005925 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08005926 struct inode *data_alloc_inode,
5927 struct buffer_head *data_alloc_bh)
5928{
5929 int status = 0;
5930 int i;
5931 unsigned int num_clusters;
5932 u64 start_blk;
5933 struct ocfs2_truncate_rec rec;
5934 struct ocfs2_dinode *di;
5935 struct ocfs2_truncate_log *tl;
5936 struct inode *tl_inode = osb->osb_tl_inode;
5937 struct buffer_head *tl_bh = osb->osb_tl_bh;
5938
Mark Fashehccd979b2005-12-15 14:31:24 -08005939 di = (struct ocfs2_dinode *) tl_bh->b_data;
5940 tl = &di->id2.i_dealloc;
5941 i = le16_to_cpu(tl->tl_used) - 1;
5942 while (i >= 0) {
5943 /* Caller has given us at least enough credits to
5944 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005945 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005946 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005947 if (status < 0) {
5948 mlog_errno(status);
5949 goto bail;
5950 }
5951
5952 tl->tl_used = cpu_to_le16(i);
5953
Joel Beckerec20cec2010-03-19 14:13:52 -07005954 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005955
Mark Fashehccd979b2005-12-15 14:31:24 -08005956 rec = tl->tl_recs[i];
5957 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5958 le32_to_cpu(rec.t_start));
5959 num_clusters = le32_to_cpu(rec.t_clusters);
5960
5961 /* if start_blk is not set, we ignore the record as
5962 * invalid. */
5963 if (start_blk) {
Tao Maa09d09b2011-02-22 08:21:10 +08005964 trace_ocfs2_replay_truncate_records(
5965 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5966 i, le32_to_cpu(rec.t_start), num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005967
5968 status = ocfs2_free_clusters(handle, data_alloc_inode,
5969 data_alloc_bh, start_blk,
5970 num_clusters);
5971 if (status < 0) {
5972 mlog_errno(status);
5973 goto bail;
5974 }
5975 }
Xue jiufei102c2592016-03-25 14:21:44 -07005976
5977 status = ocfs2_extend_trans(handle,
5978 OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5979 if (status < 0) {
5980 mlog_errno(status);
5981 goto bail;
5982 }
Mark Fashehccd979b2005-12-15 14:31:24 -08005983 i--;
5984 }
5985
Tao Ma50308d82010-11-04 15:14:11 +08005986 osb->truncated_clusters = 0;
5987
Mark Fashehccd979b2005-12-15 14:31:24 -08005988bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005989 return status;
5990}
5991
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005992/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005993int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005994{
5995 int status;
5996 unsigned int num_to_flush;
Mark Fasheh1fabe142006-10-09 18:11:45 -07005997 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005998 struct inode *tl_inode = osb->osb_tl_inode;
5999 struct inode *data_alloc_inode = NULL;
6000 struct buffer_head *tl_bh = osb->osb_tl_bh;
6001 struct buffer_head *data_alloc_bh = NULL;
6002 struct ocfs2_dinode *di;
6003 struct ocfs2_truncate_log *tl;
6004
Al Viro59551022016-01-22 15:40:57 -05006005 BUG_ON(inode_trylock(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006006
6007 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006008
Joel Becker10995aa2008-11-13 14:49:12 -08006009 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
6010 * by the underlying call to ocfs2_read_inode_block(), so any
6011 * corruption is a code bug */
6012 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6013
6014 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006015 num_to_flush = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006016 trace_ocfs2_flush_truncate_log(
6017 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
6018 num_to_flush);
Mark Fashehccd979b2005-12-15 14:31:24 -08006019 if (!num_to_flush) {
6020 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006021 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006022 }
6023
6024 data_alloc_inode = ocfs2_get_system_file_inode(osb,
6025 GLOBAL_BITMAP_SYSTEM_INODE,
6026 OCFS2_INVALID_SLOT);
6027 if (!data_alloc_inode) {
6028 status = -EINVAL;
6029 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006030 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006031 }
6032
Al Viro59551022016-01-22 15:40:57 -05006033 inode_lock(data_alloc_inode);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006034
Mark Fashehe63aecb62007-10-18 15:30:42 -07006035 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006036 if (status < 0) {
6037 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006038 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08006039 }
6040
Xue jiufei102c2592016-03-25 14:21:44 -07006041 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
Mark Fashehccd979b2005-12-15 14:31:24 -08006042 if (IS_ERR(handle)) {
6043 status = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006044 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006045 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08006046 }
6047
6048 status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
6049 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006050 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08006051 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006052
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006053 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006054
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006055out_unlock:
6056 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07006057 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006058
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006059out_mutex:
Al Viro59551022016-01-22 15:40:57 -05006060 inode_unlock(data_alloc_inode);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006061 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006062
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006063out:
Mark Fashehccd979b2005-12-15 14:31:24 -08006064 return status;
6065}
6066
6067int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6068{
6069 int status;
6070 struct inode *tl_inode = osb->osb_tl_inode;
6071
Al Viro59551022016-01-22 15:40:57 -05006072 inode_lock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006073 status = __ocfs2_flush_truncate_log(osb);
Al Viro59551022016-01-22 15:40:57 -05006074 inode_unlock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006075
6076 return status;
6077}
6078
David Howellsc4028952006-11-22 14:57:56 +00006079static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08006080{
6081 int status;
David Howellsc4028952006-11-22 14:57:56 +00006082 struct ocfs2_super *osb =
6083 container_of(work, struct ocfs2_super,
6084 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08006085
Mark Fashehccd979b2005-12-15 14:31:24 -08006086 status = ocfs2_flush_truncate_log(osb);
6087 if (status < 0)
6088 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08006089 else
Tiger Yangb89c5422010-01-25 14:11:06 +08006090 ocfs2_init_steal_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08006091}
6092
6093#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6094void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6095 int cancel)
6096{
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006097 if (osb->osb_tl_inode &&
6098 atomic_read(&osb->osb_tl_disable) == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006099 /* We want to push off log flushes while truncates are
6100 * still running. */
6101 if (cancel)
6102 cancel_delayed_work(&osb->osb_truncate_log_wq);
6103
jiangyiwen35ddf782016-03-25 14:21:32 -07006104 queue_delayed_work(osb->ocfs2_wq, &osb->osb_truncate_log_wq,
Mark Fashehccd979b2005-12-15 14:31:24 -08006105 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6106 }
6107}
6108
6109static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6110 int slot_num,
6111 struct inode **tl_inode,
6112 struct buffer_head **tl_bh)
6113{
6114 int status;
6115 struct inode *inode = NULL;
6116 struct buffer_head *bh = NULL;
6117
6118 inode = ocfs2_get_system_file_inode(osb,
6119 TRUNCATE_LOG_SYSTEM_INODE,
6120 slot_num);
6121 if (!inode) {
6122 status = -EINVAL;
6123 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6124 goto bail;
6125 }
6126
Joel Beckerb657c952008-11-13 14:49:11 -08006127 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006128 if (status < 0) {
6129 iput(inode);
6130 mlog_errno(status);
6131 goto bail;
6132 }
6133
6134 *tl_inode = inode;
6135 *tl_bh = bh;
6136bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08006137 return status;
6138}
6139
6140/* called during the 1st stage of node recovery. we stamp a clean
6141 * truncate log and pass back a copy for processing later. if the
6142 * truncate log does not require processing, a *tl_copy is set to
6143 * NULL. */
6144int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6145 int slot_num,
6146 struct ocfs2_dinode **tl_copy)
6147{
6148 int status;
6149 struct inode *tl_inode = NULL;
6150 struct buffer_head *tl_bh = NULL;
6151 struct ocfs2_dinode *di;
6152 struct ocfs2_truncate_log *tl;
6153
6154 *tl_copy = NULL;
6155
Tao Maa09d09b2011-02-22 08:21:10 +08006156 trace_ocfs2_begin_truncate_log_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08006157
6158 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6159 if (status < 0) {
6160 mlog_errno(status);
6161 goto bail;
6162 }
6163
6164 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006165
Joel Becker10995aa2008-11-13 14:49:12 -08006166 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6167 * validated by the underlying call to ocfs2_read_inode_block(),
6168 * so any corruption is a code bug */
6169 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6170
6171 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006172 if (le16_to_cpu(tl->tl_used)) {
Tao Maa09d09b2011-02-22 08:21:10 +08006173 trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
Mark Fashehccd979b2005-12-15 14:31:24 -08006174
6175 *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6176 if (!(*tl_copy)) {
6177 status = -ENOMEM;
6178 mlog_errno(status);
6179 goto bail;
6180 }
6181
6182 /* Assuming the write-out below goes well, this copy
6183 * will be passed back to recovery for processing. */
6184 memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6185
6186 /* All we need to do to clear the truncate log is set
6187 * tl_used. */
6188 tl->tl_used = 0;
6189
Joel Becker13723d02008-10-17 19:25:01 -07006190 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006191 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006192 if (status < 0) {
6193 mlog_errno(status);
6194 goto bail;
6195 }
6196 }
6197
6198bail:
Joseph Qi72865d92016-01-14 15:17:27 -08006199 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006200 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006201
Joseph Qi46359292015-09-04 15:44:54 -07006202 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006203 kfree(*tl_copy);
6204 *tl_copy = NULL;
Tao Mac1e8d352011-03-07 16:43:21 +08006205 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006206 }
6207
Mark Fashehccd979b2005-12-15 14:31:24 -08006208 return status;
6209}
6210
6211int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6212 struct ocfs2_dinode *tl_copy)
6213{
6214 int status = 0;
6215 int i;
6216 unsigned int clusters, num_recs, start_cluster;
6217 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006218 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006219 struct inode *tl_inode = osb->osb_tl_inode;
6220 struct ocfs2_truncate_log *tl;
6221
Mark Fashehccd979b2005-12-15 14:31:24 -08006222 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6223 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6224 return -EINVAL;
6225 }
6226
6227 tl = &tl_copy->id2.i_dealloc;
6228 num_recs = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006229 trace_ocfs2_complete_truncate_log_recovery(
6230 (unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6231 num_recs);
Mark Fashehccd979b2005-12-15 14:31:24 -08006232
Al Viro59551022016-01-22 15:40:57 -05006233 inode_lock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006234 for(i = 0; i < num_recs; i++) {
6235 if (ocfs2_truncate_log_needs_flush(osb)) {
6236 status = __ocfs2_flush_truncate_log(osb);
6237 if (status < 0) {
6238 mlog_errno(status);
6239 goto bail_up;
6240 }
6241 }
6242
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006243 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006244 if (IS_ERR(handle)) {
6245 status = PTR_ERR(handle);
6246 mlog_errno(status);
6247 goto bail_up;
6248 }
6249
6250 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6251 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6252 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6253
6254 status = ocfs2_truncate_log_append(osb, handle,
6255 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006256 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006257 if (status < 0) {
6258 mlog_errno(status);
6259 goto bail_up;
6260 }
6261 }
6262
6263bail_up:
Al Viro59551022016-01-22 15:40:57 -05006264 inode_unlock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006265
Mark Fashehccd979b2005-12-15 14:31:24 -08006266 return status;
6267}
6268
6269void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6270{
6271 int status;
6272 struct inode *tl_inode = osb->osb_tl_inode;
6273
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006274 atomic_set(&osb->osb_tl_disable, 1);
6275
Mark Fashehccd979b2005-12-15 14:31:24 -08006276 if (tl_inode) {
6277 cancel_delayed_work(&osb->osb_truncate_log_wq);
jiangyiwen35ddf782016-03-25 14:21:32 -07006278 flush_workqueue(osb->ocfs2_wq);
Mark Fashehccd979b2005-12-15 14:31:24 -08006279
6280 status = ocfs2_flush_truncate_log(osb);
6281 if (status < 0)
6282 mlog_errno(status);
6283
6284 brelse(osb->osb_tl_bh);
6285 iput(osb->osb_tl_inode);
6286 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006287}
6288
6289int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6290{
6291 int status;
6292 struct inode *tl_inode = NULL;
6293 struct buffer_head *tl_bh = NULL;
6294
Mark Fashehccd979b2005-12-15 14:31:24 -08006295 status = ocfs2_get_truncate_log_info(osb,
6296 osb->slot_num,
6297 &tl_inode,
6298 &tl_bh);
6299 if (status < 0)
6300 mlog_errno(status);
6301
6302 /* ocfs2_truncate_log_shutdown keys on the existence of
6303 * osb->osb_tl_inode so we don't set any of the osb variables
6304 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006305 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6306 ocfs2_truncate_log_worker);
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006307 atomic_set(&osb->osb_tl_disable, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08006308 osb->osb_tl_bh = tl_bh;
6309 osb->osb_tl_inode = tl_inode;
6310
Mark Fashehccd979b2005-12-15 14:31:24 -08006311 return status;
6312}
6313
Mark Fasheh2b604352007-06-22 15:45:27 -07006314/*
6315 * Delayed de-allocation of suballocator blocks.
6316 *
6317 * Some sets of block de-allocations might involve multiple suballocator inodes.
6318 *
6319 * The locking for this can get extremely complicated, especially when
6320 * the suballocator inodes to delete from aren't known until deep
6321 * within an unrelated codepath.
6322 *
6323 * ocfs2_extent_block structures are a good example of this - an inode
6324 * btree could have been grown by any number of nodes each allocating
6325 * out of their own suballoc inode.
6326 *
6327 * These structures allow the delay of block de-allocation until a
6328 * later time, when locking of multiple cluster inodes won't cause
6329 * deadlock.
6330 */
6331
6332/*
Tao Ma2891d292008-11-12 08:26:58 +08006333 * Describe a single bit freed from a suballocator. For the block
6334 * suballocators, it represents one block. For the global cluster
6335 * allocator, it represents some clusters and free_bit indicates
6336 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006337 */
6338struct ocfs2_cached_block_free {
6339 struct ocfs2_cached_block_free *free_next;
Tao Ma74380c42010-03-22 14:20:18 +08006340 u64 free_bg;
Mark Fasheh2b604352007-06-22 15:45:27 -07006341 u64 free_blk;
6342 unsigned int free_bit;
6343};
6344
6345struct ocfs2_per_slot_free_list {
6346 struct ocfs2_per_slot_free_list *f_next_suballocator;
6347 int f_inode_type;
6348 int f_slot;
6349 struct ocfs2_cached_block_free *f_first;
6350};
6351
Tao Ma2891d292008-11-12 08:26:58 +08006352static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6353 int sysfile_type,
6354 int slot,
6355 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006356{
6357 int ret;
6358 u64 bg_blkno;
6359 handle_t *handle;
6360 struct inode *inode;
6361 struct buffer_head *di_bh = NULL;
6362 struct ocfs2_cached_block_free *tmp;
6363
6364 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6365 if (!inode) {
6366 ret = -EINVAL;
6367 mlog_errno(ret);
6368 goto out;
6369 }
6370
Al Viro59551022016-01-22 15:40:57 -05006371 inode_lock(inode);
Mark Fasheh2b604352007-06-22 15:45:27 -07006372
Mark Fashehe63aecb62007-10-18 15:30:42 -07006373 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006374 if (ret) {
6375 mlog_errno(ret);
6376 goto out_mutex;
6377 }
6378
6379 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6380 if (IS_ERR(handle)) {
6381 ret = PTR_ERR(handle);
6382 mlog_errno(ret);
6383 goto out_unlock;
6384 }
6385
6386 while (head) {
Tao Ma74380c42010-03-22 14:20:18 +08006387 if (head->free_bg)
6388 bg_blkno = head->free_bg;
6389 else
6390 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6391 head->free_bit);
Tao Maa09d09b2011-02-22 08:21:10 +08006392 trace_ocfs2_free_cached_blocks(
6393 (unsigned long long)head->free_blk, head->free_bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006394
6395 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6396 head->free_bit, bg_blkno, 1);
6397 if (ret) {
6398 mlog_errno(ret);
6399 goto out_journal;
6400 }
6401
6402 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6403 if (ret) {
6404 mlog_errno(ret);
6405 goto out_journal;
6406 }
6407
6408 tmp = head;
6409 head = head->free_next;
6410 kfree(tmp);
6411 }
6412
6413out_journal:
6414 ocfs2_commit_trans(osb, handle);
6415
6416out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006417 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006418 brelse(di_bh);
6419out_mutex:
Al Viro59551022016-01-22 15:40:57 -05006420 inode_unlock(inode);
Mark Fasheh2b604352007-06-22 15:45:27 -07006421 iput(inode);
6422out:
6423 while(head) {
6424 /* Premature exit may have left some dangling items. */
6425 tmp = head;
6426 head = head->free_next;
6427 kfree(tmp);
6428 }
6429
6430 return ret;
6431}
6432
Tao Ma2891d292008-11-12 08:26:58 +08006433int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6434 u64 blkno, unsigned int bit)
6435{
6436 int ret = 0;
6437 struct ocfs2_cached_block_free *item;
6438
Tao Ma74380c42010-03-22 14:20:18 +08006439 item = kzalloc(sizeof(*item), GFP_NOFS);
Tao Ma2891d292008-11-12 08:26:58 +08006440 if (item == NULL) {
6441 ret = -ENOMEM;
6442 mlog_errno(ret);
6443 return ret;
6444 }
6445
Tao Maa09d09b2011-02-22 08:21:10 +08006446 trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
Tao Ma2891d292008-11-12 08:26:58 +08006447
6448 item->free_blk = blkno;
6449 item->free_bit = bit;
6450 item->free_next = ctxt->c_global_allocator;
6451
6452 ctxt->c_global_allocator = item;
6453 return ret;
6454}
6455
6456static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6457 struct ocfs2_cached_block_free *head)
6458{
6459 struct ocfs2_cached_block_free *tmp;
6460 struct inode *tl_inode = osb->osb_tl_inode;
6461 handle_t *handle;
6462 int ret = 0;
6463
Al Viro59551022016-01-22 15:40:57 -05006464 inode_lock(tl_inode);
Tao Ma2891d292008-11-12 08:26:58 +08006465
6466 while (head) {
6467 if (ocfs2_truncate_log_needs_flush(osb)) {
6468 ret = __ocfs2_flush_truncate_log(osb);
6469 if (ret < 0) {
6470 mlog_errno(ret);
6471 break;
6472 }
6473 }
6474
6475 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6476 if (IS_ERR(handle)) {
6477 ret = PTR_ERR(handle);
6478 mlog_errno(ret);
6479 break;
6480 }
6481
6482 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6483 head->free_bit);
6484
6485 ocfs2_commit_trans(osb, handle);
6486 tmp = head;
6487 head = head->free_next;
6488 kfree(tmp);
6489
6490 if (ret < 0) {
6491 mlog_errno(ret);
6492 break;
6493 }
6494 }
6495
Al Viro59551022016-01-22 15:40:57 -05006496 inode_unlock(tl_inode);
Tao Ma2891d292008-11-12 08:26:58 +08006497
6498 while (head) {
6499 /* Premature exit may have left some dangling items. */
6500 tmp = head;
6501 head = head->free_next;
6502 kfree(tmp);
6503 }
6504
6505 return ret;
6506}
6507
Mark Fasheh2b604352007-06-22 15:45:27 -07006508int ocfs2_run_deallocs(struct ocfs2_super *osb,
6509 struct ocfs2_cached_dealloc_ctxt *ctxt)
6510{
6511 int ret = 0, ret2;
6512 struct ocfs2_per_slot_free_list *fl;
6513
6514 if (!ctxt)
6515 return 0;
6516
6517 while (ctxt->c_first_suballocator) {
6518 fl = ctxt->c_first_suballocator;
6519
6520 if (fl->f_first) {
Tao Maa09d09b2011-02-22 08:21:10 +08006521 trace_ocfs2_run_deallocs(fl->f_inode_type,
6522 fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006523 ret2 = ocfs2_free_cached_blocks(osb,
6524 fl->f_inode_type,
6525 fl->f_slot,
6526 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006527 if (ret2)
6528 mlog_errno(ret2);
6529 if (!ret)
6530 ret = ret2;
6531 }
6532
6533 ctxt->c_first_suballocator = fl->f_next_suballocator;
6534 kfree(fl);
6535 }
6536
Tao Ma2891d292008-11-12 08:26:58 +08006537 if (ctxt->c_global_allocator) {
6538 ret2 = ocfs2_free_cached_clusters(osb,
6539 ctxt->c_global_allocator);
6540 if (ret2)
6541 mlog_errno(ret2);
6542 if (!ret)
6543 ret = ret2;
6544
6545 ctxt->c_global_allocator = NULL;
6546 }
6547
Mark Fasheh2b604352007-06-22 15:45:27 -07006548 return ret;
6549}
6550
6551static struct ocfs2_per_slot_free_list *
6552ocfs2_find_per_slot_free_list(int type,
6553 int slot,
6554 struct ocfs2_cached_dealloc_ctxt *ctxt)
6555{
6556 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6557
6558 while (fl) {
6559 if (fl->f_inode_type == type && fl->f_slot == slot)
6560 return fl;
6561
6562 fl = fl->f_next_suballocator;
6563 }
6564
6565 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6566 if (fl) {
6567 fl->f_inode_type = type;
6568 fl->f_slot = slot;
6569 fl->f_first = NULL;
6570 fl->f_next_suballocator = ctxt->c_first_suballocator;
6571
6572 ctxt->c_first_suballocator = fl;
6573 }
6574 return fl;
6575}
6576
Tao Ma1823cb02009-08-18 11:24:49 +08006577int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
Tao Ma74380c42010-03-22 14:20:18 +08006578 int type, int slot, u64 suballoc,
6579 u64 blkno, unsigned int bit)
Mark Fasheh2b604352007-06-22 15:45:27 -07006580{
6581 int ret;
6582 struct ocfs2_per_slot_free_list *fl;
6583 struct ocfs2_cached_block_free *item;
6584
6585 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6586 if (fl == NULL) {
6587 ret = -ENOMEM;
6588 mlog_errno(ret);
6589 goto out;
6590 }
6591
Tao Ma74380c42010-03-22 14:20:18 +08006592 item = kzalloc(sizeof(*item), GFP_NOFS);
Mark Fasheh2b604352007-06-22 15:45:27 -07006593 if (item == NULL) {
6594 ret = -ENOMEM;
6595 mlog_errno(ret);
6596 goto out;
6597 }
6598
Tao Maa09d09b2011-02-22 08:21:10 +08006599 trace_ocfs2_cache_block_dealloc(type, slot,
6600 (unsigned long long)suballoc,
6601 (unsigned long long)blkno, bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006602
Tao Ma74380c42010-03-22 14:20:18 +08006603 item->free_bg = suballoc;
Mark Fasheh2b604352007-06-22 15:45:27 -07006604 item->free_blk = blkno;
6605 item->free_bit = bit;
6606 item->free_next = fl->f_first;
6607
6608 fl->f_first = item;
6609
6610 ret = 0;
6611out:
6612 return ret;
6613}
6614
Mark Fasheh59a5e412007-06-22 15:52:36 -07006615static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6616 struct ocfs2_extent_block *eb)
6617{
6618 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6619 le16_to_cpu(eb->h_suballoc_slot),
Tao Ma74380c42010-03-22 14:20:18 +08006620 le64_to_cpu(eb->h_suballoc_loc),
Mark Fasheh59a5e412007-06-22 15:52:36 -07006621 le64_to_cpu(eb->h_blkno),
6622 le16_to_cpu(eb->h_suballoc_bit));
6623}
6624
Joel Becker2b4e30f2008-09-03 20:03:41 -07006625static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006626{
6627 set_buffer_uptodate(bh);
6628 mark_buffer_dirty(bh);
6629 return 0;
6630}
6631
Tao Ma6f70fa52009-08-25 08:05:12 +08006632void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6633 unsigned int from, unsigned int to,
6634 struct page *page, int zero, u64 *phys)
Mark Fasheh1d410a62007-09-07 14:20:45 -07006635{
6636 int ret, partial = 0;
6637
6638 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6639 if (ret)
6640 mlog_errno(ret);
6641
6642 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006643 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006644
6645 /*
6646 * Need to set the buffers we zero'd into uptodate
6647 * here if they aren't - ocfs2_map_page_blocks()
6648 * might've skipped some
6649 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006650 ret = walk_page_buffers(handle, page_buffers(page),
6651 from, to, &partial,
6652 ocfs2_zero_func);
6653 if (ret < 0)
6654 mlog_errno(ret);
6655 else if (ocfs2_should_order_data(inode)) {
6656 ret = ocfs2_jbd2_file_inode(handle, inode);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006657 if (ret < 0)
6658 mlog_errno(ret);
6659 }
6660
6661 if (!partial)
6662 SetPageUptodate(page);
6663
6664 flush_dcache_page(page);
6665}
6666
Mark Fasheh35edec12007-07-06 14:41:18 -07006667static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6668 loff_t end, struct page **pages,
6669 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006670{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006671 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006672 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006673 unsigned int from, to = PAGE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006674 struct super_block *sb = inode->i_sb;
6675
6676 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6677
6678 if (numpages == 0)
6679 goto out;
6680
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006681 to = PAGE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006682 for(i = 0; i < numpages; i++) {
6683 page = pages[i];
6684
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006685 from = start & (PAGE_SIZE - 1);
6686 if ((end >> PAGE_SHIFT) == page->index)
6687 to = end & (PAGE_SIZE - 1);
Mark Fasheh35edec12007-07-06 14:41:18 -07006688
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006689 BUG_ON(from > PAGE_SIZE);
6690 BUG_ON(to > PAGE_SIZE);
Mark Fasheh60b11392007-02-16 11:46:50 -08006691
Mark Fasheh1d410a62007-09-07 14:20:45 -07006692 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6693 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006694
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006695 start = (page->index + 1) << PAGE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006696 }
6697out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006698 if (pages)
6699 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006700}
6701
Tao Ma6f70fa52009-08-25 08:05:12 +08006702int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6703 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006704{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006705 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006706 struct address_space *mapping = inode->i_mapping;
6707 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006708 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006709
Mark Fasheh35edec12007-07-06 14:41:18 -07006710 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006711
Mark Fasheh1d410a62007-09-07 14:20:45 -07006712 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006713 last_page_bytes = PAGE_ALIGN(end);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006714 index = start >> PAGE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006715 do {
Jan Kara9b4c0ff2010-08-24 14:28:03 +02006716 pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fasheh60b11392007-02-16 11:46:50 -08006717 if (!pages[numpages]) {
6718 ret = -ENOMEM;
6719 mlog_errno(ret);
6720 goto out;
6721 }
6722
6723 numpages++;
6724 index++;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006725 } while (index < (last_page_bytes >> PAGE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006726
6727out:
6728 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006729 if (pages)
6730 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006731 numpages = 0;
6732 }
6733
6734 *num = numpages;
6735
6736 return ret;
6737}
6738
Tao Ma6f70fa52009-08-25 08:05:12 +08006739static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6740 struct page **pages, int *num)
6741{
6742 struct super_block *sb = inode->i_sb;
6743
6744 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6745 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6746
6747 return ocfs2_grab_pages(inode, start, end, pages, num);
6748}
6749
Mark Fasheh60b11392007-02-16 11:46:50 -08006750/*
6751 * Zero the area past i_size but still within an allocated
6752 * cluster. This avoids exposing nonzero data on subsequent file
6753 * extends.
6754 *
6755 * We need to call this before i_size is updated on the inode because
6756 * otherwise block_write_full_page() will skip writeout of pages past
6757 * i_size. The new_i_size parameter is passed for this reason.
6758 */
Mark Fasheh35edec12007-07-06 14:41:18 -07006759int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6760 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08006761{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006762 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08006763 struct page **pages = NULL;
6764 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006765 unsigned int ext_flags;
6766 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08006767
6768 /*
6769 * File systems which don't support sparse files zero on every
6770 * extend.
6771 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07006772 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08006773 return 0;
6774
Mark Fasheh1d410a62007-09-07 14:20:45 -07006775 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08006776 sizeof(struct page *), GFP_NOFS);
6777 if (pages == NULL) {
6778 ret = -ENOMEM;
6779 mlog_errno(ret);
6780 goto out;
6781 }
6782
Mark Fasheh1d410a62007-09-07 14:20:45 -07006783 if (range_start == range_end)
6784 goto out;
6785
6786 ret = ocfs2_extent_map_get_blocks(inode,
6787 range_start >> sb->s_blocksize_bits,
6788 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08006789 if (ret) {
6790 mlog_errno(ret);
6791 goto out;
6792 }
6793
Mark Fasheh1d410a62007-09-07 14:20:45 -07006794 /*
6795 * Tail is a hole, or is marked unwritten. In either case, we
6796 * can count on read and write to return/push zero's.
6797 */
6798 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08006799 goto out;
6800
Mark Fasheh1d410a62007-09-07 14:20:45 -07006801 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6802 &numpages);
6803 if (ret) {
6804 mlog_errno(ret);
6805 goto out;
6806 }
6807
Mark Fasheh35edec12007-07-06 14:41:18 -07006808 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6809 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08006810
6811 /*
6812 * Initiate writeout of the pages we zero'd here. We don't
6813 * wait on them - the truncate_inode_pages() call later will
6814 * do that for us.
6815 */
Christoph Hellwig2cfd30a2009-09-23 15:04:02 +02006816 ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6817 range_end - 1);
Mark Fasheh60b11392007-02-16 11:46:50 -08006818 if (ret)
6819 mlog_errno(ret);
6820
6821out:
Tim Gardnerd787ab02013-02-21 16:42:44 -08006822 kfree(pages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006823
6824 return ret;
6825}
6826
Tiger Yangfdd77702008-08-18 17:08:55 +08006827static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
6828 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006829{
6830 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08006831 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006832
Tiger Yangfdd77702008-08-18 17:08:55 +08006833 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
6834 memset(&di->id2, 0, blocksize -
6835 offsetof(struct ocfs2_dinode, id2) -
6836 xattrsize);
6837 else
6838 memset(&di->id2, 0, blocksize -
6839 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006840}
6841
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006842void ocfs2_dinode_new_extent_list(struct inode *inode,
6843 struct ocfs2_dinode *di)
6844{
Tiger Yangfdd77702008-08-18 17:08:55 +08006845 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006846 di->id2.i_list.l_tree_depth = 0;
6847 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08006848 di->id2.i_list.l_count = cpu_to_le16(
6849 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006850}
6851
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006852void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
6853{
6854 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6855 struct ocfs2_inline_data *idata = &di->id2.i_data;
6856
6857 spin_lock(&oi->ip_lock);
6858 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
6859 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6860 spin_unlock(&oi->ip_lock);
6861
6862 /*
6863 * We clear the entire i_data structure here so that all
6864 * fields can be properly initialized.
6865 */
Tiger Yangfdd77702008-08-18 17:08:55 +08006866 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006867
Tiger Yangfdd77702008-08-18 17:08:55 +08006868 idata->id_count = cpu_to_le16(
6869 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006870}
6871
6872int ocfs2_convert_inline_data_to_extents(struct inode *inode,
6873 struct buffer_head *di_bh)
6874{
6875 int ret, i, has_data, num_pages = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006876 int need_free = 0;
6877 u32 bit_off, num;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006878 handle_t *handle;
6879 u64 uninitialized_var(block);
6880 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6881 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6882 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006883 struct ocfs2_alloc_context *data_ac = NULL;
6884 struct page **pages = NULL;
6885 loff_t end = osb->s_clustersize;
Joel Beckerf99b9b72008-08-20 19:36:33 -07006886 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02006887 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006888
6889 has_data = i_size_read(inode) ? 1 : 0;
6890
6891 if (has_data) {
6892 pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
6893 sizeof(struct page *), GFP_NOFS);
6894 if (pages == NULL) {
6895 ret = -ENOMEM;
6896 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006897 return ret;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006898 }
6899
6900 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
6901 if (ret) {
6902 mlog_errno(ret);
Markus Elfring3cc79b72015-04-14 15:42:45 -07006903 goto free_pages;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006904 }
6905 }
6906
Jan Karaa90714c2008-10-09 19:38:40 +02006907 handle = ocfs2_start_trans(osb,
6908 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006909 if (IS_ERR(handle)) {
6910 ret = PTR_ERR(handle);
6911 mlog_errno(ret);
alex chen15eba0f2015-02-10 14:09:02 -08006912 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006913 }
6914
Joel Becker0cf2f762009-02-12 16:41:25 -08006915 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07006916 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006917 if (ret) {
6918 mlog_errno(ret);
6919 goto out_commit;
6920 }
6921
6922 if (has_data) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006923 unsigned int page_end;
6924 u64 phys;
6925
Christoph Hellwig5dd40562010-03-03 09:05:00 -05006926 ret = dquot_alloc_space_nodirty(inode,
6927 ocfs2_clusters_to_bytes(osb->sb, 1));
6928 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02006929 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02006930 did_quota = 1;
6931
Mark Fasheh4fe370a2009-12-07 13:15:40 -08006932 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
6933
Joel Becker1ed9b772010-05-06 13:59:06 +08006934 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006935 &num);
6936 if (ret) {
6937 mlog_errno(ret);
6938 goto out_commit;
6939 }
6940
6941 /*
6942 * Save two copies, one for insert, and one that can
6943 * be changed by ocfs2_map_and_dirty_page() below.
6944 */
6945 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
6946
6947 /*
6948 * Non sparse file systems zero on extend, so no need
6949 * to do that now.
6950 */
6951 if (!ocfs2_sparse_alloc(osb) &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006952 PAGE_SIZE < osb->s_clustersize)
6953 end = PAGE_SIZE;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006954
6955 ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
6956 if (ret) {
6957 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006958 need_free = 1;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006959 goto out_commit;
6960 }
6961
6962 /*
6963 * This should populate the 1st page for us and mark
6964 * it up to date.
6965 */
6966 ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
6967 if (ret) {
6968 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08006969 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08006970 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006971 }
6972
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006973 page_end = PAGE_SIZE;
6974 if (PAGE_SIZE > osb->s_clustersize)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006975 page_end = osb->s_clustersize;
6976
6977 for (i = 0; i < num_pages; i++)
6978 ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
6979 pages[i], i > 0, &phys);
6980 }
6981
6982 spin_lock(&oi->ip_lock);
6983 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
6984 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6985 spin_unlock(&oi->ip_lock);
6986
Darrick J. Wong2931cdc2014-04-03 14:46:48 -07006987 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07006988 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07006989
6990 ocfs2_journal_dirty(handle, di_bh);
6991
6992 if (has_data) {
6993 /*
6994 * An error at this point should be extremely rare. If
6995 * this proves to be false, we could always re-build
6996 * the in-inode data from our pages.
6997 */
Joel Becker5e404e92009-02-13 03:54:22 -08006998 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08006999 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007000 if (ret) {
7001 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007002 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08007003 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007004 }
7005
7006 inode->i_blocks = ocfs2_inode_sector_count(inode);
7007 }
7008
alex chen15eba0f2015-02-10 14:09:02 -08007009out_unlock:
7010 if (pages)
7011 ocfs2_unlock_and_free_pages(pages, num_pages);
7012
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007013out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02007014 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05007015 dquot_free_space_nodirty(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02007016 ocfs2_clusters_to_bytes(osb->sb, 1));
7017
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007018 if (need_free) {
7019 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
7020 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
7021 bit_off, num);
7022 else
7023 ocfs2_free_clusters(handle,
7024 data_ac->ac_inode,
7025 data_ac->ac_bh,
7026 ocfs2_clusters_to_blocks(osb->sb, bit_off),
7027 num);
7028 }
7029
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007030 ocfs2_commit_trans(osb, handle);
7031
alex chen15eba0f2015-02-10 14:09:02 -08007032out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007033 if (data_ac)
7034 ocfs2_free_alloc_context(data_ac);
Markus Elfring3cc79b72015-04-14 15:42:45 -07007035free_pages:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07007036 kfree(pages);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007037 return ret;
7038}
7039
Mark Fashehccd979b2005-12-15 14:31:24 -08007040/*
7041 * It is expected, that by the time you call this function,
7042 * inode->i_size and fe->i_size have been adjusted.
7043 *
7044 * WARNING: This will kfree the truncate context
7045 */
7046int ocfs2_commit_truncate(struct ocfs2_super *osb,
7047 struct inode *inode,
Tristan Ye78f94672010-05-11 17:54:42 +08007048 struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08007049{
Tristan Ye78f94672010-05-11 17:54:42 +08007050 int status = 0, i, flags = 0;
7051 u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
Tao Mabcbbb242009-08-18 11:29:12 +08007052 u64 blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08007053 struct ocfs2_extent_list *el;
Tristan Ye78f94672010-05-11 17:54:42 +08007054 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08007055 struct ocfs2_path *path = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08007056 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7057 struct ocfs2_extent_list *root_el = &(di->id2.i_list);
7058 u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
7059 struct ocfs2_extent_tree et;
7060 struct ocfs2_cached_dealloc_ctxt dealloc;
Junxiao Bif62f12b2014-12-18 16:17:32 -08007061 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007062
Tristan Ye78f94672010-05-11 17:54:42 +08007063 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7064 ocfs2_init_dealloc_ctxt(&dealloc);
7065
Mark Fashehdcd05382007-01-16 11:32:23 -08007066 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08007067 i_size_read(inode));
7068
Tristan Ye78f94672010-05-11 17:54:42 +08007069 path = ocfs2_new_path(di_bh, &di->id2.i_list,
Joel Becker13723d02008-10-17 19:25:01 -07007070 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08007071 if (!path) {
7072 status = -ENOMEM;
7073 mlog_errno(status);
7074 goto bail;
7075 }
Mark Fasheh83418972007-04-23 18:53:12 -07007076
7077 ocfs2_extent_map_trunc(inode, new_highest_cpos);
7078
Mark Fashehccd979b2005-12-15 14:31:24 -08007079start:
Mark Fashehdcd05382007-01-16 11:32:23 -08007080 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007081 * Check that we still have allocation to delete.
7082 */
7083 if (OCFS2_I(inode)->ip_clusters == 0) {
7084 status = 0;
7085 goto bail;
7086 }
7087
7088 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007089 * Truncate always works against the rightmost tree branch.
7090 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007091 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007092 if (status) {
7093 mlog_errno(status);
7094 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007095 }
7096
Tao Maa09d09b2011-02-22 08:21:10 +08007097 trace_ocfs2_commit_truncate(
7098 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7099 new_highest_cpos,
7100 OCFS2_I(inode)->ip_clusters,
7101 path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08007102
7103 /*
7104 * By now, el will point to the extent list on the bottom most
7105 * portion of this tree. Only the tail record is considered in
7106 * each pass.
7107 *
7108 * We handle the following cases, in order:
7109 * - empty extent: delete the remaining branch
7110 * - remove the entire record
7111 * - remove a partial record
7112 * - no record needs to be removed (truncate has completed)
7113 */
7114 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007115 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7116 ocfs2_error(inode->i_sb,
7117 "Inode %llu has empty extent block at %llu\n",
7118 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7119 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7120 status = -EROFS;
7121 goto bail;
7122 }
7123
Mark Fashehccd979b2005-12-15 14:31:24 -08007124 i = le16_to_cpu(el->l_next_free_rec) - 1;
Tristan Ye78f94672010-05-11 17:54:42 +08007125 rec = &el->l_recs[i];
7126 flags = rec->e_flags;
7127 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7128
7129 if (i == 0 && ocfs2_is_empty_extent(rec)) {
7130 /*
7131 * Lower levels depend on this never happening, but it's best
7132 * to check it up here before changing the tree.
7133 */
7134 if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
Xue jiufeid0c97d52015-09-04 15:44:48 -07007135 mlog(ML_ERROR, "Inode %lu has an empty "
Tristan Ye78f94672010-05-11 17:54:42 +08007136 "extent record, depth %u\n", inode->i_ino,
7137 le16_to_cpu(root_el->l_tree_depth));
Xue jiufeid0c97d52015-09-04 15:44:48 -07007138 status = ocfs2_remove_rightmost_empty_extent(osb,
7139 &et, path, &dealloc);
7140 if (status) {
7141 mlog_errno(status);
7142 goto bail;
7143 }
7144
7145 ocfs2_reinit_path(path, 1);
7146 goto start;
7147 } else {
7148 trunc_cpos = le32_to_cpu(rec->e_cpos);
7149 trunc_len = 0;
7150 blkno = 0;
Tristan Ye78f94672010-05-11 17:54:42 +08007151 }
Tristan Ye78f94672010-05-11 17:54:42 +08007152 } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7153 /*
7154 * Truncate entire record.
7155 */
7156 trunc_cpos = le32_to_cpu(rec->e_cpos);
7157 trunc_len = ocfs2_rec_clusters(el, rec);
7158 blkno = le64_to_cpu(rec->e_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007159 } else if (range > new_highest_cpos) {
Tristan Ye78f94672010-05-11 17:54:42 +08007160 /*
7161 * Partial truncate. it also should be
7162 * the last truncate we're doing.
7163 */
7164 trunc_cpos = new_highest_cpos;
7165 trunc_len = range - new_highest_cpos;
7166 coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7167 blkno = le64_to_cpu(rec->e_blkno) +
7168 ocfs2_clusters_to_blocks(inode->i_sb, coff);
Mark Fashehdcd05382007-01-16 11:32:23 -08007169 } else {
Tristan Ye78f94672010-05-11 17:54:42 +08007170 /*
7171 * Truncate completed, leave happily.
7172 */
Mark Fashehdcd05382007-01-16 11:32:23 -08007173 status = 0;
7174 goto bail;
7175 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007176
Tristan Ye78f94672010-05-11 17:54:42 +08007177 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007178
Junxiao Bif62f12b2014-12-18 16:17:32 -08007179 if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) {
7180 status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
7181 &ref_tree, NULL);
7182 if (status) {
7183 mlog_errno(status);
7184 goto bail;
7185 }
7186 }
7187
Tristan Ye78f94672010-05-11 17:54:42 +08007188 status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7189 phys_cpos, trunc_len, flags, &dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08007190 refcount_loc, true);
Mark Fashehccd979b2005-12-15 14:31:24 -08007191 if (status < 0) {
7192 mlog_errno(status);
7193 goto bail;
7194 }
7195
Mark Fashehdcd05382007-01-16 11:32:23 -08007196 ocfs2_reinit_path(path, 1);
7197
7198 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007199 * The check above will catch the case where we've truncated
7200 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007201 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007202 goto start;
7203
Mark Fashehccd979b2005-12-15 14:31:24 -08007204bail:
Junxiao Bif62f12b2014-12-18 16:17:32 -08007205 if (ref_tree)
7206 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08007207
7208 ocfs2_schedule_truncate_log_flush(osb, 1);
7209
Tristan Ye78f94672010-05-11 17:54:42 +08007210 ocfs2_run_deallocs(osb, &dealloc);
Mark Fasheh59a5e412007-06-22 15:52:36 -07007211
Mark Fashehdcd05382007-01-16 11:32:23 -08007212 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007213
Mark Fashehccd979b2005-12-15 14:31:24 -08007214 return status;
7215}
7216
Mark Fashehccd979b2005-12-15 14:31:24 -08007217/*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007218 * 'start' is inclusive, 'end' is not.
7219 */
7220int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7221 unsigned int start, unsigned int end, int trunc)
7222{
7223 int ret;
7224 unsigned int numbytes;
7225 handle_t *handle;
7226 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7227 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7228 struct ocfs2_inline_data *idata = &di->id2.i_data;
7229
7230 if (end > i_size_read(inode))
7231 end = i_size_read(inode);
7232
Younger Liud62e74b2014-02-10 14:25:51 -08007233 BUG_ON(start > end);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007234
7235 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7236 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7237 !ocfs2_supports_inline_data(osb)) {
7238 ocfs2_error(inode->i_sb,
Joe Perches7ecef142015-09-04 15:44:51 -07007239 "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 -07007240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7241 le16_to_cpu(di->i_dyn_features),
7242 OCFS2_I(inode)->ip_dyn_features,
7243 osb->s_feature_incompat);
7244 ret = -EROFS;
7245 goto out;
7246 }
7247
7248 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7249 if (IS_ERR(handle)) {
7250 ret = PTR_ERR(handle);
7251 mlog_errno(ret);
7252 goto out;
7253 }
7254
Joel Becker0cf2f762009-02-12 16:41:25 -08007255 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007256 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007257 if (ret) {
7258 mlog_errno(ret);
7259 goto out_commit;
7260 }
7261
7262 numbytes = end - start;
7263 memset(idata->id_data + start, 0, numbytes);
7264
7265 /*
7266 * No need to worry about the data page here - it's been
7267 * truncated already and inline data doesn't need it for
7268 * pushing zero's to disk, so we'll let readpage pick it up
7269 * later.
7270 */
7271 if (trunc) {
7272 i_size_write(inode, start);
7273 di->i_size = cpu_to_le64(start);
7274 }
7275
7276 inode->i_blocks = ocfs2_inode_sector_count(inode);
7277 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7278
7279 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7280 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7281
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07007282 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007283 ocfs2_journal_dirty(handle, di_bh);
7284
7285out_commit:
7286 ocfs2_commit_trans(osb, handle);
7287
7288out:
7289 return ret;
7290}
Tao Mae80de362011-05-23 10:36:43 +08007291
7292static int ocfs2_trim_extent(struct super_block *sb,
7293 struct ocfs2_group_desc *gd,
7294 u32 start, u32 count)
7295{
7296 u64 discard, bcount;
7297
7298 bcount = ocfs2_clusters_to_blocks(sb, count);
7299 discard = le64_to_cpu(gd->bg_blkno) +
7300 ocfs2_clusters_to_blocks(sb, start);
7301
Tao Ma10fca352011-05-23 10:36:45 +08007302 trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7303
Tao Mae80de362011-05-23 10:36:43 +08007304 return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7305}
7306
7307static int ocfs2_trim_group(struct super_block *sb,
7308 struct ocfs2_group_desc *gd,
7309 u32 start, u32 max, u32 minbits)
7310{
7311 int ret = 0, count = 0, next;
7312 void *bitmap = gd->bg_bitmap;
7313
7314 if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7315 return 0;
7316
Tao Ma10fca352011-05-23 10:36:45 +08007317 trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7318 start, max, minbits);
7319
Tao Mae80de362011-05-23 10:36:43 +08007320 while (start < max) {
7321 start = ocfs2_find_next_zero_bit(bitmap, max, start);
7322 if (start >= max)
7323 break;
7324 next = ocfs2_find_next_bit(bitmap, max, start);
7325
7326 if ((next - start) >= minbits) {
7327 ret = ocfs2_trim_extent(sb, gd,
7328 start, next - start);
7329 if (ret < 0) {
7330 mlog_errno(ret);
7331 break;
7332 }
7333 count += next - start;
7334 }
7335 start = next + 1;
7336
7337 if (fatal_signal_pending(current)) {
7338 count = -ERESTARTSYS;
7339 break;
7340 }
7341
7342 if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7343 break;
7344 }
7345
7346 if (ret < 0)
7347 count = ret;
7348
7349 return count;
7350}
7351
7352int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7353{
7354 struct ocfs2_super *osb = OCFS2_SB(sb);
7355 u64 start, len, trimmed, first_group, last_group, group;
7356 int ret, cnt;
7357 u32 first_bit, last_bit, minlen;
7358 struct buffer_head *main_bm_bh = NULL;
7359 struct inode *main_bm_inode = NULL;
7360 struct buffer_head *gd_bh = NULL;
7361 struct ocfs2_dinode *main_bm;
7362 struct ocfs2_group_desc *gd = NULL;
7363
7364 start = range->start >> osb->s_clustersize_bits;
7365 len = range->len >> osb->s_clustersize_bits;
7366 minlen = range->minlen >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007367
Jie Liuaa897622014-01-21 15:48:35 -08007368 if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
Tao Mae80de362011-05-23 10:36:43 +08007369 return -EINVAL;
7370
7371 main_bm_inode = ocfs2_get_system_file_inode(osb,
7372 GLOBAL_BITMAP_SYSTEM_INODE,
7373 OCFS2_INVALID_SLOT);
7374 if (!main_bm_inode) {
7375 ret = -EIO;
7376 mlog_errno(ret);
7377 goto out;
7378 }
7379
Al Viro59551022016-01-22 15:40:57 -05007380 inode_lock(main_bm_inode);
Tao Mae80de362011-05-23 10:36:43 +08007381
7382 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7383 if (ret < 0) {
7384 mlog_errno(ret);
7385 goto out_mutex;
7386 }
7387 main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7388
7389 if (start >= le32_to_cpu(main_bm->i_clusters)) {
7390 ret = -EINVAL;
7391 goto out_unlock;
7392 }
7393
Jie Liuaa897622014-01-21 15:48:35 -08007394 len = range->len >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007395 if (start + len > le32_to_cpu(main_bm->i_clusters))
7396 len = le32_to_cpu(main_bm->i_clusters) - start;
7397
Tao Ma10fca352011-05-23 10:36:45 +08007398 trace_ocfs2_trim_fs(start, len, minlen);
7399
Tao Mae80de362011-05-23 10:36:43 +08007400 /* Determine first and last group to examine based on start and len */
7401 first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7402 if (first_group == osb->first_cluster_group_blkno)
7403 first_bit = start;
7404 else
7405 first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
7406 last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
7407 last_bit = osb->bitmap_cpg;
7408
Jie Liuaa897622014-01-21 15:48:35 -08007409 trimmed = 0;
Tao Mae80de362011-05-23 10:36:43 +08007410 for (group = first_group; group <= last_group;) {
7411 if (first_bit + len >= osb->bitmap_cpg)
7412 last_bit = osb->bitmap_cpg;
7413 else
7414 last_bit = first_bit + len;
7415
7416 ret = ocfs2_read_group_descriptor(main_bm_inode,
7417 main_bm, group,
7418 &gd_bh);
7419 if (ret < 0) {
7420 mlog_errno(ret);
7421 break;
7422 }
7423
7424 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
7425 cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
7426 brelse(gd_bh);
7427 gd_bh = NULL;
7428 if (cnt < 0) {
7429 ret = cnt;
7430 mlog_errno(ret);
7431 break;
7432 }
7433
7434 trimmed += cnt;
7435 len -= osb->bitmap_cpg - first_bit;
7436 first_bit = 0;
7437 if (group == osb->first_cluster_group_blkno)
7438 group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7439 else
7440 group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7441 }
7442 range->len = trimmed * sb->s_blocksize;
7443out_unlock:
7444 ocfs2_inode_unlock(main_bm_inode, 0);
7445 brelse(main_bm_bh);
7446out_mutex:
Al Viro59551022016-01-22 15:40:57 -05007447 inode_unlock(main_bm_inode);
Tao Mae80de362011-05-23 10:36:43 +08007448 iput(main_bm_inode);
7449out:
7450 return ret;
7451}