blob: bf9357123bc58781156b003a58d080a74fce5933 [file] [log] [blame]
Thomas Gleixner328970d2019-05-24 12:04:05 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Masahiro Yamadafa60ce22021-05-06 18:06:44 -07002/*
Mark Fashehccd979b2005-12-15 14:31:24 -08003 * alloc.c
4 *
5 * Extent allocs and frees
6 *
7 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
Mark Fashehccd979b2005-12-15 14:31:24 -08008 */
9
10#include <linux/fs.h>
11#include <linux/types.h>
12#include <linux/slab.h>
13#include <linux/highmem.h>
Mark Fasheh60b11392007-02-16 11:46:50 -080014#include <linux/swap.h>
Jan Karaa90714c2008-10-09 19:38:40 +020015#include <linux/quotaops.h>
Tao Mae80de362011-05-23 10:36:43 +080016#include <linux/blkdev.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010017#include <linux/sched/signal.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080018
Mark Fashehccd979b2005-12-15 14:31:24 -080019#include <cluster/masklog.h>
20
21#include "ocfs2.h"
22
23#include "alloc.h"
Mark Fasheh60b11392007-02-16 11:46:50 -080024#include "aops.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070025#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080026#include "dlmglue.h"
27#include "extent_map.h"
28#include "inode.h"
29#include "journal.h"
30#include "localalloc.h"
31#include "suballoc.h"
32#include "sysfile.h"
33#include "file.h"
34#include "super.h"
35#include "uptodate.h"
Joel Becker2a50a742008-12-09 14:24:33 -080036#include "xattr.h"
Tao Mabcbbb242009-08-18 11:29:12 +080037#include "refcounttree.h"
Tao Maa09d09b2011-02-22 08:21:10 +080038#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039
40#include "buffer_head_io.h"
41
Tao Ma853a3a12009-08-18 11:22:18 +080042enum ocfs2_contig_type {
43 CONTIG_NONE = 0,
44 CONTIG_LEFT,
45 CONTIG_RIGHT,
46 CONTIG_LEFTRIGHT,
47};
Tao Mae7d4cb62008-08-18 17:38:44 +080048
Tao Ma853a3a12009-08-18 11:22:18 +080049static enum ocfs2_contig_type
50 ocfs2_extent_rec_contig(struct super_block *sb,
51 struct ocfs2_extent_rec *ext,
52 struct ocfs2_extent_rec *insert_rec);
Joel Becker1625f8a2008-08-21 17:11:10 -070053/*
54 * Operations for a specific extent tree type.
55 *
56 * To implement an on-disk btree (extent tree) type in ocfs2, add
57 * an ocfs2_extent_tree_operations structure and the matching
Joel Becker8d6220d2008-08-22 12:46:09 -070058 * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it
Joel Becker1625f8a2008-08-21 17:11:10 -070059 * for the allocation portion of the extent tree.
60 */
Tao Mae7d4cb62008-08-18 17:38:44 +080061struct ocfs2_extent_tree_operations {
Joel Becker1625f8a2008-08-21 17:11:10 -070062 /*
63 * last_eb_blk is the block number of the right most leaf extent
64 * block. Most on-disk structures containing an extent tree store
65 * this value for fast access. The ->eo_set_last_eb_blk() and
66 * ->eo_get_last_eb_blk() operations access this value. They are
67 * both required.
68 */
Joel Becker35dc0aa2008-08-20 16:25:06 -070069 void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
70 u64 blkno);
71 u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -070072
73 /*
74 * The on-disk structure usually keeps track of how many total
75 * clusters are stored in this extent tree. This function updates
76 * that value. new_clusters is the delta, and must be
77 * added to the total. Required.
78 */
Joel Becker6136ca52009-02-12 19:32:43 -080079 void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -070080 u32 new_clusters);
Joel Becker1625f8a2008-08-21 17:11:10 -070081
82 /*
Joel Becker92ba4702009-02-13 03:18:34 -080083 * If this extent tree is supported by an extent map, insert
84 * a record into the map.
85 */
86 void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
87 struct ocfs2_extent_rec *rec);
88
89 /*
Joel Becker4c911ee2009-02-13 02:50:12 -080090 * If this extent tree is supported by an extent map, truncate the
91 * map to clusters,
92 */
93 void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
94 u32 clusters);
95
96 /*
Joel Becker1625f8a2008-08-21 17:11:10 -070097 * If ->eo_insert_check() exists, it is called before rec is
98 * inserted into the extent tree. It is optional.
99 */
Joel Becker6136ca52009-02-12 19:32:43 -0800100 int (*eo_insert_check)(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700101 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800102 int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
Joel Becker0ce10102008-08-20 17:19:50 -0700103
Joel Becker1625f8a2008-08-21 17:11:10 -0700104 /*
105 * --------------------------------------------------------------
106 * The remaining are internal to ocfs2_extent_tree and don't have
107 * accessor functions
108 */
109
110 /*
111 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
112 * It is required.
113 */
Joel Becker0ce10102008-08-20 17:19:50 -0700114 void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
Joel Becker1625f8a2008-08-21 17:11:10 -0700115
116 /*
117 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
118 * it exists. If it does not, et->et_max_leaf_clusters is set
119 * to 0 (unlimited). Optional.
120 */
Joel Becker6136ca52009-02-12 19:32:43 -0800121 void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
Tao Ma853a3a12009-08-18 11:22:18 +0800122
123 /*
124 * ->eo_extent_contig test whether the 2 ocfs2_extent_rec
125 * are contiguous or not. Optional. Don't need to set it if use
126 * ocfs2_extent_rec as the tree leaf.
127 */
128 enum ocfs2_contig_type
129 (*eo_extent_contig)(struct ocfs2_extent_tree *et,
130 struct ocfs2_extent_rec *ext,
131 struct ocfs2_extent_rec *insert_rec);
Tao Mae7d4cb62008-08-18 17:38:44 +0800132};
133
Joel Beckerf99b9b72008-08-20 19:36:33 -0700134
135/*
136 * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
137 * in the methods.
138 */
139static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
140static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
141 u64 blkno);
Joel Becker6136ca52009-02-12 19:32:43 -0800142static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700143 u32 clusters);
Joel Becker92ba4702009-02-13 03:18:34 -0800144static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
145 struct ocfs2_extent_rec *rec);
Joel Becker4c911ee2009-02-13 02:50:12 -0800146static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
147 u32 clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800148static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700149 struct ocfs2_extent_rec *rec);
Joel Becker6136ca52009-02-12 19:32:43 -0800150static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700151static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
Changwei Ge71a36942018-01-31 16:15:06 -0800152
153static int ocfs2_reuse_blk_from_dealloc(handle_t *handle,
154 struct ocfs2_extent_tree *et,
155 struct buffer_head **new_eb_bh,
156 int blk_wanted, int *blk_given);
157static int ocfs2_is_dealloc_empty(struct ocfs2_extent_tree *et);
158
Julia Lawall9e62dc02016-01-14 15:17:21 -0800159static const struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
Joel Beckerf99b9b72008-08-20 19:36:33 -0700160 .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk,
161 .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk,
162 .eo_update_clusters = ocfs2_dinode_update_clusters,
Joel Becker92ba4702009-02-13 03:18:34 -0800163 .eo_extent_map_insert = ocfs2_dinode_extent_map_insert,
Joel Becker4c911ee2009-02-13 02:50:12 -0800164 .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700165 .eo_insert_check = ocfs2_dinode_insert_check,
166 .eo_sanity_check = ocfs2_dinode_sanity_check,
167 .eo_fill_root_el = ocfs2_dinode_fill_root_el,
Tao Mae7d4cb62008-08-18 17:38:44 +0800168};
169
170static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
171 u64 blkno)
172{
Joel Beckerea5efa12008-08-20 16:57:27 -0700173 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800174
Joel Beckerf99b9b72008-08-20 19:36:33 -0700175 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800176 di->i_last_eb_blk = cpu_to_le64(blkno);
177}
178
179static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
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 return le64_to_cpu(di->i_last_eb_blk);
185}
186
Joel Becker6136ca52009-02-12 19:32:43 -0800187static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +0800188 u32 clusters)
189{
Joel Becker6136ca52009-02-12 19:32:43 -0800190 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
Joel Beckerea5efa12008-08-20 16:57:27 -0700191 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800192
193 le32_add_cpu(&di->i_clusters, clusters);
Joel Becker6136ca52009-02-12 19:32:43 -0800194 spin_lock(&oi->ip_lock);
195 oi->ip_clusters = le32_to_cpu(di->i_clusters);
196 spin_unlock(&oi->ip_lock);
Tao Mae7d4cb62008-08-18 17:38:44 +0800197}
198
Joel Becker92ba4702009-02-13 03:18:34 -0800199static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
200 struct ocfs2_extent_rec *rec)
201{
202 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
203
204 ocfs2_extent_map_insert_rec(inode, rec);
205}
206
Joel Becker4c911ee2009-02-13 02:50:12 -0800207static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
208 u32 clusters)
209{
210 struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
211
212 ocfs2_extent_map_trunc(inode, clusters);
213}
214
Joel Becker6136ca52009-02-12 19:32:43 -0800215static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700216 struct ocfs2_extent_rec *rec)
217{
Joel Becker6136ca52009-02-12 19:32:43 -0800218 struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
219 struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
Joel Becker1e61ee72008-08-20 18:32:45 -0700220
Joel Becker6136ca52009-02-12 19:32:43 -0800221 BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
Joel Becker1e61ee72008-08-20 18:32:45 -0700222 mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
Joel Becker6136ca52009-02-12 19:32:43 -0800223 (oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
Joel Becker1e61ee72008-08-20 18:32:45 -0700224 "Device %s, asking for sparse allocation: inode %llu, "
225 "cpos %u, clusters %u\n",
226 osb->dev_str,
Joel Becker6136ca52009-02-12 19:32:43 -0800227 (unsigned long long)oi->ip_blkno,
228 rec->e_cpos, oi->ip_clusters);
Joel Becker1e61ee72008-08-20 18:32:45 -0700229
230 return 0;
231}
232
Joel Becker6136ca52009-02-12 19:32:43 -0800233static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800234{
Joel Becker10995aa2008-11-13 14:49:12 -0800235 struct ocfs2_dinode *di = et->et_object;
Tao Mae7d4cb62008-08-18 17:38:44 +0800236
Joel Beckerf99b9b72008-08-20 19:36:33 -0700237 BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
Joel Becker10995aa2008-11-13 14:49:12 -0800238 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
Tao Mae7d4cb62008-08-18 17:38:44 +0800239
Joel Becker10995aa2008-11-13 14:49:12 -0800240 return 0;
Tao Mae7d4cb62008-08-18 17:38:44 +0800241}
242
Joel Beckerf99b9b72008-08-20 19:36:33 -0700243static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
244{
245 struct ocfs2_dinode *di = et->et_object;
246
247 et->et_root_el = &di->id2.i_list;
248}
249
Tao Mae7d4cb62008-08-18 17:38:44 +0800250
Joel Becker0ce10102008-08-20 17:19:50 -0700251static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
252{
Joel Becker2a50a742008-12-09 14:24:33 -0800253 struct ocfs2_xattr_value_buf *vb = et->et_object;
Joel Becker0ce10102008-08-20 17:19:50 -0700254
Joel Becker2a50a742008-12-09 14:24:33 -0800255 et->et_root_el = &vb->vb_xv->xr_list;
Joel Becker0ce10102008-08-20 17:19:50 -0700256}
257
Tao Maf56654c2008-08-18 17:38:48 +0800258static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
259 u64 blkno)
260{
Joel Becker2a50a742008-12-09 14:24:33 -0800261 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800262
Joel Becker2a50a742008-12-09 14:24:33 -0800263 vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
Tao Maf56654c2008-08-18 17:38:48 +0800264}
265
266static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
267{
Joel Becker2a50a742008-12-09 14:24:33 -0800268 struct ocfs2_xattr_value_buf *vb = et->et_object;
Tao Maf56654c2008-08-18 17:38:48 +0800269
Joel Becker2a50a742008-12-09 14:24:33 -0800270 return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
Tao Maf56654c2008-08-18 17:38:48 +0800271}
272
Joel Becker6136ca52009-02-12 19:32:43 -0800273static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
Tao Maf56654c2008-08-18 17:38:48 +0800274 u32 clusters)
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 le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
Tao Maf56654c2008-08-18 17:38:48 +0800279}
280
Julia Lawall9e62dc02016-01-14 15:17:21 -0800281static const struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700282 .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk,
283 .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk,
284 .eo_update_clusters = ocfs2_xattr_value_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700285 .eo_fill_root_el = ocfs2_xattr_value_fill_root_el,
Tao Maf56654c2008-08-18 17:38:48 +0800286};
287
Joel Becker0ce10102008-08-20 17:19:50 -0700288static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
289{
290 struct ocfs2_xattr_block *xb = et->et_object;
291
292 et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
293}
294
Joel Becker6136ca52009-02-12 19:32:43 -0800295static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
Joel Becker943cced2008-08-20 17:31:10 -0700296{
Joel Becker6136ca52009-02-12 19:32:43 -0800297 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Joel Becker943cced2008-08-20 17:31:10 -0700298 et->et_max_leaf_clusters =
Joel Becker6136ca52009-02-12 19:32:43 -0800299 ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
Joel Becker943cced2008-08-20 17:31:10 -0700300}
301
Tao Maba492612008-08-18 17:38:49 +0800302static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
303 u64 blkno)
304{
Joel Beckerea5efa12008-08-20 16:57:27 -0700305 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800306 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
307
308 xt->xt_last_eb_blk = cpu_to_le64(blkno);
309}
310
311static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
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 return le64_to_cpu(xt->xt_last_eb_blk);
317}
318
Joel Becker6136ca52009-02-12 19:32:43 -0800319static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
Tao Maba492612008-08-18 17:38:49 +0800320 u32 clusters)
321{
Joel Beckerea5efa12008-08-20 16:57:27 -0700322 struct ocfs2_xattr_block *xb = et->et_object;
Tao Maba492612008-08-18 17:38:49 +0800323
324 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
325}
326
Julia Lawall9e62dc02016-01-14 15:17:21 -0800327static const struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
Joel Becker35dc0aa2008-08-20 16:25:06 -0700328 .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk,
329 .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk,
330 .eo_update_clusters = ocfs2_xattr_tree_update_clusters,
Joel Becker0ce10102008-08-20 17:19:50 -0700331 .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el,
Joel Becker943cced2008-08-20 17:31:10 -0700332 .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
Tao Maba492612008-08-18 17:38:49 +0800333};
334
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800335static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
336 u64 blkno)
337{
338 struct ocfs2_dx_root_block *dx_root = et->et_object;
339
340 dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
341}
342
343static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
344{
345 struct ocfs2_dx_root_block *dx_root = et->et_object;
346
347 return le64_to_cpu(dx_root->dr_last_eb_blk);
348}
349
Joel Becker6136ca52009-02-12 19:32:43 -0800350static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800351 u32 clusters)
352{
353 struct ocfs2_dx_root_block *dx_root = et->et_object;
354
355 le32_add_cpu(&dx_root->dr_clusters, clusters);
356}
357
Joel Becker6136ca52009-02-12 19:32:43 -0800358static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800359{
360 struct ocfs2_dx_root_block *dx_root = et->et_object;
361
362 BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));
363
364 return 0;
365}
366
367static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
368{
369 struct ocfs2_dx_root_block *dx_root = et->et_object;
370
371 et->et_root_el = &dx_root->dr_list;
372}
373
Julia Lawall9e62dc02016-01-14 15:17:21 -0800374static const struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800375 .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk,
376 .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk,
377 .eo_update_clusters = ocfs2_dx_root_update_clusters,
378 .eo_sanity_check = ocfs2_dx_root_sanity_check,
379 .eo_fill_root_el = ocfs2_dx_root_fill_root_el,
380};
381
Tao Mafe924412009-08-18 11:22:25 +0800382static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et)
383{
384 struct ocfs2_refcount_block *rb = et->et_object;
385
386 et->et_root_el = &rb->rf_list;
387}
388
389static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
390 u64 blkno)
391{
392 struct ocfs2_refcount_block *rb = et->et_object;
393
394 rb->rf_last_eb_blk = cpu_to_le64(blkno);
395}
396
397static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
398{
399 struct ocfs2_refcount_block *rb = et->et_object;
400
401 return le64_to_cpu(rb->rf_last_eb_blk);
402}
403
404static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et,
405 u32 clusters)
406{
407 struct ocfs2_refcount_block *rb = et->et_object;
408
409 le32_add_cpu(&rb->rf_clusters, clusters);
410}
411
412static enum ocfs2_contig_type
413ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et,
414 struct ocfs2_extent_rec *ext,
415 struct ocfs2_extent_rec *insert_rec)
416{
417 return CONTIG_NONE;
418}
419
Julia Lawall9e62dc02016-01-14 15:17:21 -0800420static const struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
Tao Mafe924412009-08-18 11:22:25 +0800421 .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk,
422 .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk,
423 .eo_update_clusters = ocfs2_refcount_tree_update_clusters,
424 .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el,
425 .eo_extent_contig = ocfs2_refcount_tree_extent_contig,
426};
427
Joel Becker8d6220d2008-08-22 12:46:09 -0700428static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800429 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700430 struct buffer_head *bh,
Joel Becker13723d02008-10-17 19:25:01 -0700431 ocfs2_journal_access_func access,
Joel Becker8d6220d2008-08-22 12:46:09 -0700432 void *obj,
Julia Lawall9e62dc02016-01-14 15:17:21 -0800433 const struct ocfs2_extent_tree_operations *ops)
Tao Mae7d4cb62008-08-18 17:38:44 +0800434{
Joel Becker1a09f552008-08-20 17:44:24 -0700435 et->et_ops = ops;
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700436 et->et_root_bh = bh;
Joel Becker5e404e92009-02-13 03:54:22 -0800437 et->et_ci = ci;
Joel Becker13723d02008-10-17 19:25:01 -0700438 et->et_root_journal_access = access;
Joel Beckerea5efa12008-08-20 16:57:27 -0700439 if (!obj)
440 obj = (void *)bh->b_data;
441 et->et_object = obj;
Changwei Ge71a36942018-01-31 16:15:06 -0800442 et->et_dealloc = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +0800443
Joel Becker0ce10102008-08-20 17:19:50 -0700444 et->et_ops->eo_fill_root_el(et);
Joel Becker943cced2008-08-20 17:31:10 -0700445 if (!et->et_ops->eo_fill_max_leaf_clusters)
446 et->et_max_leaf_clusters = 0;
447 else
Joel Becker6136ca52009-02-12 19:32:43 -0800448 et->et_ops->eo_fill_max_leaf_clusters(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800449}
450
Joel Becker8d6220d2008-08-22 12:46:09 -0700451void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800452 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700453 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700454{
Joel Becker5e404e92009-02-13 03:54:22 -0800455 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di,
Joel Becker13723d02008-10-17 19:25:01 -0700456 NULL, &ocfs2_dinode_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700457}
458
Joel Becker8d6220d2008-08-22 12:46:09 -0700459void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800460 struct ocfs2_caching_info *ci,
Joel Becker8d6220d2008-08-22 12:46:09 -0700461 struct buffer_head *bh)
Joel Becker1a09f552008-08-20 17:44:24 -0700462{
Joel Becker5e404e92009-02-13 03:54:22 -0800463 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb,
Joel Becker13723d02008-10-17 19:25:01 -0700464 NULL, &ocfs2_xattr_tree_et_ops);
Joel Becker1a09f552008-08-20 17:44:24 -0700465}
466
Joel Becker8d6220d2008-08-22 12:46:09 -0700467void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800468 struct ocfs2_caching_info *ci,
Joel Becker2a50a742008-12-09 14:24:33 -0800469 struct ocfs2_xattr_value_buf *vb)
Tao Mae7d4cb62008-08-18 17:38:44 +0800470{
Joel Becker5e404e92009-02-13 03:54:22 -0800471 __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb,
Joel Becker8d6220d2008-08-22 12:46:09 -0700472 &ocfs2_xattr_value_et_ops);
Tao Mae7d4cb62008-08-18 17:38:44 +0800473}
474
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800475void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
Joel Becker5e404e92009-02-13 03:54:22 -0800476 struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800477 struct buffer_head *bh)
478{
Joel Becker5e404e92009-02-13 03:54:22 -0800479 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800480 NULL, &ocfs2_dx_root_et_ops);
481}
482
Tao Mafe924412009-08-18 11:22:25 +0800483void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
484 struct ocfs2_caching_info *ci,
485 struct buffer_head *bh)
486{
487 __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb,
488 NULL, &ocfs2_refcount_tree_et_ops);
489}
490
Joel Becker35dc0aa2008-08-20 16:25:06 -0700491static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
492 u64 new_last_eb_blk)
Tao Mae7d4cb62008-08-18 17:38:44 +0800493{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700494 et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
Tao Mae7d4cb62008-08-18 17:38:44 +0800495}
496
Joel Becker35dc0aa2008-08-20 16:25:06 -0700497static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
Tao Mae7d4cb62008-08-18 17:38:44 +0800498{
Joel Beckerce1d9ea2008-08-20 16:30:07 -0700499 return et->et_ops->eo_get_last_eb_blk(et);
Tao Mae7d4cb62008-08-18 17:38:44 +0800500}
501
Joel Becker6136ca52009-02-12 19:32:43 -0800502static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
Joel Becker35dc0aa2008-08-20 16:25:06 -0700503 u32 clusters)
Tao Mae7d4cb62008-08-18 17:38:44 +0800504{
Joel Becker6136ca52009-02-12 19:32:43 -0800505 et->et_ops->eo_update_clusters(et, clusters);
Joel Becker35dc0aa2008-08-20 16:25:06 -0700506}
507
Joel Becker92ba4702009-02-13 03:18:34 -0800508static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
509 struct ocfs2_extent_rec *rec)
510{
511 if (et->et_ops->eo_extent_map_insert)
512 et->et_ops->eo_extent_map_insert(et, rec);
513}
514
Joel Becker4c911ee2009-02-13 02:50:12 -0800515static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
516 u32 clusters)
517{
518 if (et->et_ops->eo_extent_map_truncate)
519 et->et_ops->eo_extent_map_truncate(et, clusters);
520}
521
Joel Becker13723d02008-10-17 19:25:01 -0700522static inline int ocfs2_et_root_journal_access(handle_t *handle,
Joel Becker13723d02008-10-17 19:25:01 -0700523 struct ocfs2_extent_tree *et,
524 int type)
525{
Joel Beckerd9a0a1f2009-02-12 17:32:34 -0800526 return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700527 type);
528}
529
Tao Ma853a3a12009-08-18 11:22:18 +0800530static inline enum ocfs2_contig_type
531 ocfs2_et_extent_contig(struct ocfs2_extent_tree *et,
532 struct ocfs2_extent_rec *rec,
533 struct ocfs2_extent_rec *insert_rec)
534{
535 if (et->et_ops->eo_extent_contig)
536 return et->et_ops->eo_extent_contig(et, rec, insert_rec);
537
538 return ocfs2_extent_rec_contig(
539 ocfs2_metadata_cache_get_super(et->et_ci),
540 rec, insert_rec);
541}
542
Joel Becker6136ca52009-02-12 19:32:43 -0800543static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
Joel Becker1e61ee72008-08-20 18:32:45 -0700544 struct ocfs2_extent_rec *rec)
545{
546 int ret = 0;
547
548 if (et->et_ops->eo_insert_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800549 ret = et->et_ops->eo_insert_check(et, rec);
Joel Becker1e61ee72008-08-20 18:32:45 -0700550 return ret;
551}
552
Joel Becker6136ca52009-02-12 19:32:43 -0800553static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
Joel Becker35dc0aa2008-08-20 16:25:06 -0700554{
Joel Becker1e61ee72008-08-20 18:32:45 -0700555 int ret = 0;
556
557 if (et->et_ops->eo_sanity_check)
Joel Becker6136ca52009-02-12 19:32:43 -0800558 ret = et->et_ops->eo_sanity_check(et);
Joel Becker1e61ee72008-08-20 18:32:45 -0700559 return ret;
Tao Mae7d4cb62008-08-18 17:38:44 +0800560}
561
Mark Fasheh59a5e412007-06-22 15:52:36 -0700562static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
563 struct ocfs2_extent_block *eb);
Joel Beckerd401dc12009-02-13 02:24:10 -0800564static void ocfs2_adjust_rightmost_records(handle_t *handle,
565 struct ocfs2_extent_tree *et,
Tao Ma6b791bc2009-06-12 14:18:36 +0800566 struct ocfs2_path *path,
567 struct ocfs2_extent_rec *insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -0800568/*
569 * Reset the actual path elements so that we can re-use the structure
570 * to build another path. Generally, this involves freeing the buffer
571 * heads.
572 */
Tao Mae2e9f602009-08-18 11:22:34 +0800573void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
Mark Fashehdcd05382007-01-16 11:32:23 -0800574{
575 int i, start = 0, depth = 0;
576 struct ocfs2_path_item *node;
577
578 if (keep_root)
579 start = 1;
580
581 for(i = start; i < path_num_items(path); i++) {
582 node = &path->p_node[i];
583
584 brelse(node->bh);
585 node->bh = NULL;
586 node->el = NULL;
587 }
588
589 /*
590 * Tree depth may change during truncate, or insert. If we're
591 * keeping the root extent list, then make sure that our path
592 * structure reflects the proper depth.
593 */
594 if (keep_root)
595 depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
Joel Becker13723d02008-10-17 19:25:01 -0700596 else
597 path_root_access(path) = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -0800598
599 path->p_tree_depth = depth;
600}
601
Tao Mae2e9f602009-08-18 11:22:34 +0800602void ocfs2_free_path(struct ocfs2_path *path)
Mark Fashehdcd05382007-01-16 11:32:23 -0800603{
604 if (path) {
605 ocfs2_reinit_path(path, 0);
606 kfree(path);
607 }
608}
609
610/*
Mark Fasheh328d5752007-06-18 10:48:04 -0700611 * All the elements of src into dest. After this call, src could be freed
612 * without affecting dest.
613 *
614 * Both paths should have the same root. Any non-root elements of dest
615 * will be freed.
616 */
617static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
618{
619 int i;
620
621 BUG_ON(path_root_bh(dest) != path_root_bh(src));
622 BUG_ON(path_root_el(dest) != path_root_el(src));
Joel Becker13723d02008-10-17 19:25:01 -0700623 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fasheh328d5752007-06-18 10:48:04 -0700624
625 ocfs2_reinit_path(dest, 1);
626
627 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
628 dest->p_node[i].bh = src->p_node[i].bh;
629 dest->p_node[i].el = src->p_node[i].el;
630
631 if (dest->p_node[i].bh)
632 get_bh(dest->p_node[i].bh);
633 }
634}
635
636/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800637 * Make the *dest path the same as src and re-initialize src path to
638 * have a root only.
639 */
640static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
641{
642 int i;
643
644 BUG_ON(path_root_bh(dest) != path_root_bh(src));
Joel Becker13723d02008-10-17 19:25:01 -0700645 BUG_ON(path_root_access(dest) != path_root_access(src));
Mark Fashehdcd05382007-01-16 11:32:23 -0800646
647 for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
648 brelse(dest->p_node[i].bh);
649
650 dest->p_node[i].bh = src->p_node[i].bh;
651 dest->p_node[i].el = src->p_node[i].el;
652
653 src->p_node[i].bh = NULL;
654 src->p_node[i].el = NULL;
655 }
656}
657
658/*
659 * Insert an extent block at given index.
660 *
661 * This will not take an additional reference on eb_bh.
662 */
663static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
664 struct buffer_head *eb_bh)
665{
666 struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
667
668 /*
669 * Right now, no root bh is an extent block, so this helps
670 * catch code errors with dinode trees. The assertion can be
671 * safely removed if we ever need to insert extent block
672 * structures at the root.
673 */
674 BUG_ON(index == 0);
675
676 path->p_node[index].bh = eb_bh;
677 path->p_node[index].el = &eb->h_list;
678}
679
680static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700681 struct ocfs2_extent_list *root_el,
682 ocfs2_journal_access_func access)
Mark Fashehdcd05382007-01-16 11:32:23 -0800683{
684 struct ocfs2_path *path;
685
686 BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
687
688 path = kzalloc(sizeof(*path), GFP_NOFS);
689 if (path) {
690 path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
691 get_bh(root_bh);
692 path_root_bh(path) = root_bh;
693 path_root_el(path) = root_el;
Joel Becker13723d02008-10-17 19:25:01 -0700694 path_root_access(path) = access;
Mark Fashehdcd05382007-01-16 11:32:23 -0800695 }
696
697 return path;
698}
699
Tao Mae2e9f602009-08-18 11:22:34 +0800700struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
Joel Beckerffdd7a52008-10-17 22:32:01 -0700701{
Joel Becker13723d02008-10-17 19:25:01 -0700702 return ocfs2_new_path(path_root_bh(path), path_root_el(path),
703 path_root_access(path));
Joel Beckerffdd7a52008-10-17 22:32:01 -0700704}
705
Tao Mae2e9f602009-08-18 11:22:34 +0800706struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
Joel Beckerffdd7a52008-10-17 22:32:01 -0700707{
Joel Becker13723d02008-10-17 19:25:01 -0700708 return ocfs2_new_path(et->et_root_bh, et->et_root_el,
709 et->et_root_journal_access);
710}
711
712/*
713 * Journal the buffer at depth idx. All idx>0 are extent_blocks,
714 * otherwise it's the root_access function.
715 *
716 * I don't like the way this function's name looks next to
717 * ocfs2_journal_access_path(), but I don't have a better one.
718 */
Tao Mae2e9f602009-08-18 11:22:34 +0800719int ocfs2_path_bh_journal_access(handle_t *handle,
720 struct ocfs2_caching_info *ci,
721 struct ocfs2_path *path,
722 int idx)
Joel Becker13723d02008-10-17 19:25:01 -0700723{
724 ocfs2_journal_access_func access = path_root_access(path);
725
726 if (!access)
727 access = ocfs2_journal_access;
728
729 if (idx)
730 access = ocfs2_journal_access_eb;
731
Joel Becker0cf2f762009-02-12 16:41:25 -0800732 return access(handle, ci, path->p_node[idx].bh,
Joel Becker13723d02008-10-17 19:25:01 -0700733 OCFS2_JOURNAL_ACCESS_WRITE);
Joel Beckerffdd7a52008-10-17 22:32:01 -0700734}
735
Mark Fashehdcd05382007-01-16 11:32:23 -0800736/*
Mark Fashehdcd05382007-01-16 11:32:23 -0800737 * Convenience function to journal all components in a path.
738 */
Tao Mae2e9f602009-08-18 11:22:34 +0800739int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
740 handle_t *handle,
741 struct ocfs2_path *path)
Mark Fashehdcd05382007-01-16 11:32:23 -0800742{
743 int i, ret = 0;
744
745 if (!path)
746 goto out;
747
748 for(i = 0; i < path_num_items(path); i++) {
Joel Becker0cf2f762009-02-12 16:41:25 -0800749 ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -0800750 if (ret < 0) {
751 mlog_errno(ret);
752 goto out;
753 }
754 }
755
756out:
757 return ret;
758}
759
Mark Fasheh328d5752007-06-18 10:48:04 -0700760/*
761 * Return the index of the extent record which contains cluster #v_cluster.
762 * -1 is returned if it was not found.
763 *
764 * Should work fine on interior and exterior nodes.
765 */
766int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
767{
768 int ret = -1;
769 int i;
770 struct ocfs2_extent_rec *rec;
771 u32 rec_end, rec_start, clusters;
772
773 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
774 rec = &el->l_recs[i];
775
776 rec_start = le32_to_cpu(rec->e_cpos);
777 clusters = ocfs2_rec_clusters(el, rec);
778
779 rec_end = rec_start + clusters;
780
781 if (v_cluster >= rec_start && v_cluster < rec_end) {
782 ret = i;
783 break;
784 }
785 }
786
787 return ret;
788}
789
Mark Fashehe48edee2007-03-07 16:46:57 -0800790/*
791 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
Tao Ma853a3a12009-08-18 11:22:18 +0800792 * ocfs2_extent_rec_contig only work properly against leaf nodes!
Mark Fashehe48edee2007-03-07 16:46:57 -0800793 */
Mark Fashehdcd05382007-01-16 11:32:23 -0800794static int ocfs2_block_extent_contig(struct super_block *sb,
795 struct ocfs2_extent_rec *ext,
796 u64 blkno)
Mark Fashehccd979b2005-12-15 14:31:24 -0800797{
Mark Fashehe48edee2007-03-07 16:46:57 -0800798 u64 blk_end = le64_to_cpu(ext->e_blkno);
799
800 blk_end += ocfs2_clusters_to_blocks(sb,
801 le16_to_cpu(ext->e_leaf_clusters));
802
803 return blkno == blk_end;
Mark Fashehccd979b2005-12-15 14:31:24 -0800804}
805
Mark Fashehdcd05382007-01-16 11:32:23 -0800806static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
807 struct ocfs2_extent_rec *right)
808{
Mark Fashehe48edee2007-03-07 16:46:57 -0800809 u32 left_range;
810
811 left_range = le32_to_cpu(left->e_cpos) +
812 le16_to_cpu(left->e_leaf_clusters);
813
814 return (left_range == le32_to_cpu(right->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -0800815}
816
817static enum ocfs2_contig_type
Tao Ma853a3a12009-08-18 11:22:18 +0800818 ocfs2_extent_rec_contig(struct super_block *sb,
819 struct ocfs2_extent_rec *ext,
820 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -0800821{
822 u64 blkno = le64_to_cpu(insert_rec->e_blkno);
823
Mark Fasheh328d5752007-06-18 10:48:04 -0700824 /*
825 * Refuse to coalesce extent records with different flag
826 * fields - we don't want to mix unwritten extents with user
827 * data.
828 */
829 if (ext->e_flags != insert_rec->e_flags)
830 return CONTIG_NONE;
831
Mark Fashehdcd05382007-01-16 11:32:23 -0800832 if (ocfs2_extents_adjacent(ext, insert_rec) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800833 ocfs2_block_extent_contig(sb, ext, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800834 return CONTIG_RIGHT;
835
836 blkno = le64_to_cpu(ext->e_blkno);
837 if (ocfs2_extents_adjacent(insert_rec, ext) &&
Joel Beckerb4a17652009-02-13 03:07:09 -0800838 ocfs2_block_extent_contig(sb, insert_rec, blkno))
Mark Fashehdcd05382007-01-16 11:32:23 -0800839 return CONTIG_LEFT;
840
841 return CONTIG_NONE;
842}
843
844/*
845 * NOTE: We can have pretty much any combination of contiguousness and
846 * appending.
847 *
848 * The usefulness of APPEND_TAIL is more in that it lets us know that
849 * we'll have to update the path to that leaf.
850 */
851enum ocfs2_append_type {
852 APPEND_NONE = 0,
853 APPEND_TAIL,
854};
855
Mark Fasheh328d5752007-06-18 10:48:04 -0700856enum ocfs2_split_type {
857 SPLIT_NONE = 0,
858 SPLIT_LEFT,
859 SPLIT_RIGHT,
860};
861
Mark Fashehdcd05382007-01-16 11:32:23 -0800862struct ocfs2_insert_type {
Mark Fasheh328d5752007-06-18 10:48:04 -0700863 enum ocfs2_split_type ins_split;
Mark Fashehdcd05382007-01-16 11:32:23 -0800864 enum ocfs2_append_type ins_appending;
865 enum ocfs2_contig_type ins_contig;
866 int ins_contig_index;
Mark Fashehdcd05382007-01-16 11:32:23 -0800867 int ins_tree_depth;
868};
869
Mark Fasheh328d5752007-06-18 10:48:04 -0700870struct ocfs2_merge_ctxt {
871 enum ocfs2_contig_type c_contig_type;
872 int c_has_empty_extent;
873 int c_split_covers_rec;
Mark Fasheh328d5752007-06-18 10:48:04 -0700874};
875
Joel Becker5e965812008-11-13 14:49:16 -0800876static int ocfs2_validate_extent_block(struct super_block *sb,
877 struct buffer_head *bh)
878{
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700879 int rc;
Joel Becker5e965812008-11-13 14:49:16 -0800880 struct ocfs2_extent_block *eb =
881 (struct ocfs2_extent_block *)bh->b_data;
882
Tao Maa09d09b2011-02-22 08:21:10 +0800883 trace_ocfs2_validate_extent_block((unsigned long long)bh->b_blocknr);
Joel Becker970e4932008-11-13 14:49:19 -0800884
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700885 BUG_ON(!buffer_uptodate(bh));
886
887 /*
888 * If the ecc fails, we return the error but otherwise
889 * leave the filesystem running. We know any error is
890 * local to this block.
891 */
892 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
Joel Becker13723d02008-10-17 19:25:01 -0700893 if (rc) {
894 mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
895 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700896 return rc;
Joel Becker13723d02008-10-17 19:25:01 -0700897 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -0700898
899 /*
900 * Errors after here are fatal.
901 */
902
Joel Becker5e965812008-11-13 14:49:16 -0800903 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700904 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700905 "Extent block #%llu has bad signature %.*s\n",
906 (unsigned long long)bh->b_blocknr, 7,
907 eb->h_signature);
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700908 goto bail;
Joel Becker5e965812008-11-13 14:49:16 -0800909 }
910
911 if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700912 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700913 "Extent block #%llu has an invalid h_blkno of %llu\n",
914 (unsigned long long)bh->b_blocknr,
915 (unsigned long long)le64_to_cpu(eb->h_blkno));
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700916 goto bail;
Joel Becker5e965812008-11-13 14:49:16 -0800917 }
918
wangyan229ba1f2018-08-17 15:44:27 -0700919 if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation)
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700920 rc = ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -0700921 "Extent block #%llu has an invalid h_fs_generation of #%u\n",
922 (unsigned long long)bh->b_blocknr,
923 le32_to_cpu(eb->h_fs_generation));
Goldwyn Rodrigues17a5b9a2015-09-04 15:44:17 -0700924bail:
925 return rc;
Joel Becker5e965812008-11-13 14:49:16 -0800926}
927
Joel Becker3d03a302009-02-12 17:49:26 -0800928int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
Joel Becker5e965812008-11-13 14:49:16 -0800929 struct buffer_head **bh)
930{
931 int rc;
932 struct buffer_head *tmp = *bh;
933
Joel Becker3d03a302009-02-12 17:49:26 -0800934 rc = ocfs2_read_block(ci, eb_blkno, &tmp,
Joel Becker970e4932008-11-13 14:49:19 -0800935 ocfs2_validate_extent_block);
Joel Becker5e965812008-11-13 14:49:16 -0800936
937 /* If ocfs2_read_block() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -0800938 if (!rc && !*bh)
Joel Becker5e965812008-11-13 14:49:16 -0800939 *bh = tmp;
940
Joel Becker5e965812008-11-13 14:49:16 -0800941 return rc;
942}
943
944
Mark Fashehccd979b2005-12-15 14:31:24 -0800945/*
946 * How many free extents have we got before we need more meta data?
947 */
Jun Piao964f14a2017-09-06 16:19:11 -0700948int ocfs2_num_free_extents(struct ocfs2_extent_tree *et)
Mark Fashehccd979b2005-12-15 14:31:24 -0800949{
950 int retval;
Tao Mae7d4cb62008-08-18 17:38:44 +0800951 struct ocfs2_extent_list *el = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800952 struct ocfs2_extent_block *eb;
953 struct buffer_head *eb_bh = NULL;
Tao Mae7d4cb62008-08-18 17:38:44 +0800954 u64 last_eb_blk = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800955
Joel Beckerf99b9b72008-08-20 19:36:33 -0700956 el = et->et_root_el;
957 last_eb_blk = ocfs2_et_get_last_eb_blk(et);
Mark Fashehccd979b2005-12-15 14:31:24 -0800958
Tao Mae7d4cb62008-08-18 17:38:44 +0800959 if (last_eb_blk) {
Joel Becker3d03a302009-02-12 17:49:26 -0800960 retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
961 &eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 if (retval < 0) {
963 mlog_errno(retval);
964 goto bail;
965 }
966 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
967 el = &eb->h_list;
Tao Mae7d4cb62008-08-18 17:38:44 +0800968 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800969
970 BUG_ON(el->l_tree_depth != 0);
971
972 retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
973bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700974 brelse(eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800975
Tao Maa09d09b2011-02-22 08:21:10 +0800976 trace_ocfs2_num_free_extents(retval);
Mark Fashehccd979b2005-12-15 14:31:24 -0800977 return retval;
978}
979
980/* expects array to already be allocated
981 *
982 * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
983 * l_count for you
984 */
Joel Becker42a5a7a2009-02-12 18:49:19 -0800985static int ocfs2_create_new_meta_bhs(handle_t *handle,
986 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -0800987 int wanted,
988 struct ocfs2_alloc_context *meta_ac,
989 struct buffer_head *bhs[])
990{
991 int count, status, i;
992 u16 suballoc_bit_start;
993 u32 num_got;
Joel Becker2b6cb572010-03-26 10:09:15 +0800994 u64 suballoc_loc, first_blkno;
Joel Becker42a5a7a2009-02-12 18:49:19 -0800995 struct ocfs2_super *osb =
996 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -0800997 struct ocfs2_extent_block *eb;
998
Mark Fashehccd979b2005-12-15 14:31:24 -0800999 count = 0;
1000 while (count < wanted) {
Joel Becker1ed9b772010-05-06 13:59:06 +08001001 status = ocfs2_claim_metadata(handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001002 meta_ac,
1003 wanted - count,
Joel Becker2b6cb572010-03-26 10:09:15 +08001004 &suballoc_loc,
Mark Fashehccd979b2005-12-15 14:31:24 -08001005 &suballoc_bit_start,
1006 &num_got,
1007 &first_blkno);
1008 if (status < 0) {
1009 mlog_errno(status);
1010 goto bail;
1011 }
1012
1013 for(i = count; i < (num_got + count); i++) {
1014 bhs[i] = sb_getblk(osb->sb, first_blkno);
1015 if (bhs[i] == NULL) {
Rui Xiang7391a292013-11-12 15:06:54 -08001016 status = -ENOMEM;
Mark Fashehccd979b2005-12-15 14:31:24 -08001017 mlog_errno(status);
1018 goto bail;
1019 }
Joel Becker42a5a7a2009-02-12 18:49:19 -08001020 ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001021
Joel Becker42a5a7a2009-02-12 18:49:19 -08001022 status = ocfs2_journal_access_eb(handle, et->et_ci,
1023 bhs[i],
Joel Becker13723d02008-10-17 19:25:01 -07001024 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001025 if (status < 0) {
1026 mlog_errno(status);
1027 goto bail;
1028 }
1029
1030 memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
1031 eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
1032 /* Ok, setup the minimal stuff here. */
1033 strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
1034 eb->h_blkno = cpu_to_le64(first_blkno);
1035 eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
Tiger Yangb89c5422010-01-25 14:11:06 +08001036 eb->h_suballoc_slot =
1037 cpu_to_le16(meta_ac->ac_alloc_slot);
Joel Becker2b6cb572010-03-26 10:09:15 +08001038 eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
Mark Fashehccd979b2005-12-15 14:31:24 -08001039 eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1040 eb->h_list.l_count =
1041 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
1042
1043 suballoc_bit_start++;
1044 first_blkno++;
1045
1046 /* We'll also be dirtied by the caller, so
1047 * this isn't absolutely necessary. */
Joel Beckerec20cec2010-03-19 14:13:52 -07001048 ocfs2_journal_dirty(handle, bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001049 }
1050
1051 count += num_got;
1052 }
1053
1054 status = 0;
1055bail:
1056 if (status < 0) {
1057 for(i = 0; i < wanted; i++) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001058 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001059 bhs[i] = NULL;
1060 }
1061 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001062 return status;
1063}
1064
1065/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001066 * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
1067 *
1068 * Returns the sum of the rightmost extent rec logical offset and
1069 * cluster count.
1070 *
1071 * ocfs2_add_branch() uses this to determine what logical cluster
1072 * value should be populated into the leftmost new branch records.
1073 *
1074 * ocfs2_shift_tree_depth() uses this to determine the # clusters
1075 * value for the new topmost tree record.
1076 */
1077static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el)
1078{
1079 int i;
1080
1081 i = le16_to_cpu(el->l_next_free_rec) - 1;
1082
1083 return le32_to_cpu(el->l_recs[i].e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001084 ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fashehdcd05382007-01-16 11:32:23 -08001085}
1086
1087/*
Tao Ma6b791bc2009-06-12 14:18:36 +08001088 * Change range of the branches in the right most path according to the leaf
1089 * extent block's rightmost record.
1090 */
1091static int ocfs2_adjust_rightmost_branch(handle_t *handle,
Tao Ma6b791bc2009-06-12 14:18:36 +08001092 struct ocfs2_extent_tree *et)
1093{
1094 int status;
1095 struct ocfs2_path *path = NULL;
1096 struct ocfs2_extent_list *el;
1097 struct ocfs2_extent_rec *rec;
1098
1099 path = ocfs2_new_path_from_et(et);
1100 if (!path) {
1101 status = -ENOMEM;
1102 return status;
1103 }
1104
Joel Beckerfacdb772009-02-12 18:08:48 -08001105 status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
Tao Ma6b791bc2009-06-12 14:18:36 +08001106 if (status < 0) {
1107 mlog_errno(status);
1108 goto out;
1109 }
1110
Tao Mac901fb02010-04-26 14:34:57 +08001111 status = ocfs2_extend_trans(handle, path_num_items(path));
Tao Ma6b791bc2009-06-12 14:18:36 +08001112 if (status < 0) {
1113 mlog_errno(status);
1114 goto out;
1115 }
1116
Joel Beckerd401dc12009-02-13 02:24:10 -08001117 status = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma6b791bc2009-06-12 14:18:36 +08001118 if (status < 0) {
1119 mlog_errno(status);
1120 goto out;
1121 }
1122
1123 el = path_leaf_el(path);
Al Viro3a251f02012-04-13 12:22:00 -04001124 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1];
Tao Ma6b791bc2009-06-12 14:18:36 +08001125
Joel Beckerd401dc12009-02-13 02:24:10 -08001126 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Tao Ma6b791bc2009-06-12 14:18:36 +08001127
1128out:
1129 ocfs2_free_path(path);
1130 return status;
1131}
1132
1133/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001134 * Add an entire tree branch to our inode. eb_bh is the extent block
Joel Beckerd401dc12009-02-13 02:24:10 -08001135 * to start at, if we don't want to start the branch at the root
Mark Fashehccd979b2005-12-15 14:31:24 -08001136 * structure.
1137 *
1138 * last_eb_bh is required as we have to update it's next_leaf pointer
1139 * for the new last extent block.
1140 *
1141 * the new branch will be 'empty' in the sense that every block will
Mark Fashehe48edee2007-03-07 16:46:57 -08001142 * contain a single record with cluster count == 0.
Mark Fashehccd979b2005-12-15 14:31:24 -08001143 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001144static int ocfs2_add_branch(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001145 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001146 struct buffer_head *eb_bh,
Mark Fasheh328d5752007-06-18 10:48:04 -07001147 struct buffer_head **last_eb_bh,
Mark Fashehccd979b2005-12-15 14:31:24 -08001148 struct ocfs2_alloc_context *meta_ac)
1149{
Changwei Ge71a36942018-01-31 16:15:06 -08001150 int status, new_blocks, i, block_given = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001151 u64 next_blkno, new_last_eb_blk;
1152 struct buffer_head *bh;
1153 struct buffer_head **new_eb_bhs = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001154 struct ocfs2_extent_block *eb;
1155 struct ocfs2_extent_list *eb_el;
1156 struct ocfs2_extent_list *el;
Tao Ma6b791bc2009-06-12 14:18:36 +08001157 u32 new_cpos, root_end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001158
Mark Fasheh328d5752007-06-18 10:48:04 -07001159 BUG_ON(!last_eb_bh || !*last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001160
Mark Fashehccd979b2005-12-15 14:31:24 -08001161 if (eb_bh) {
1162 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1163 el = &eb->h_list;
1164 } else
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001165 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001166
1167 /* we never add a branch to a leaf. */
1168 BUG_ON(!el->l_tree_depth);
1169
1170 new_blocks = le16_to_cpu(el->l_tree_depth);
1171
Tao Ma6b791bc2009-06-12 14:18:36 +08001172 eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
1173 new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
1174 root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
1175
1176 /*
1177 * If there is a gap before the root end and the real end
1178 * of the righmost leaf block, we need to remove the gap
1179 * between new_cpos and root_end first so that the tree
1180 * is consistent after we add a new branch(it will start
1181 * from new_cpos).
1182 */
1183 if (root_end > new_cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08001184 trace_ocfs2_adjust_rightmost_branch(
1185 (unsigned long long)
1186 ocfs2_metadata_cache_owner(et->et_ci),
1187 root_end, new_cpos);
1188
Joel Beckerd401dc12009-02-13 02:24:10 -08001189 status = ocfs2_adjust_rightmost_branch(handle, et);
Tao Ma6b791bc2009-06-12 14:18:36 +08001190 if (status) {
1191 mlog_errno(status);
1192 goto bail;
1193 }
1194 }
1195
Mark Fashehccd979b2005-12-15 14:31:24 -08001196 /* allocate the number of new eb blocks we need */
1197 new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
1198 GFP_KERNEL);
1199 if (!new_eb_bhs) {
1200 status = -ENOMEM;
1201 mlog_errno(status);
1202 goto bail;
1203 }
1204
Changwei Ge71a36942018-01-31 16:15:06 -08001205 /* Firstyly, try to reuse dealloc since we have already estimated how
1206 * many extent blocks we may use.
1207 */
1208 if (!ocfs2_is_dealloc_empty(et)) {
1209 status = ocfs2_reuse_blk_from_dealloc(handle, et,
1210 new_eb_bhs, new_blocks,
1211 &block_given);
1212 if (status < 0) {
1213 mlog_errno(status);
1214 goto bail;
1215 }
1216 }
1217
1218 BUG_ON(block_given > new_blocks);
1219
1220 if (block_given < new_blocks) {
1221 BUG_ON(!meta_ac);
1222 status = ocfs2_create_new_meta_bhs(handle, et,
1223 new_blocks - block_given,
1224 meta_ac,
1225 &new_eb_bhs[block_given]);
1226 if (status < 0) {
1227 mlog_errno(status);
1228 goto bail;
1229 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001230 }
1231
1232 /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
1233 * linked with the rest of the tree.
1234 * conversly, new_eb_bhs[0] is the new bottommost leaf.
1235 *
1236 * when we leave the loop, new_last_eb_blk will point to the
1237 * newest leaf, and next_blkno will point to the topmost extent
1238 * block. */
1239 next_blkno = new_last_eb_blk = 0;
1240 for(i = 0; i < new_blocks; i++) {
1241 bh = new_eb_bhs[i];
1242 eb = (struct ocfs2_extent_block *) bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001243 /* ocfs2_create_new_meta_bhs() should create it right! */
1244 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001245 eb_el = &eb->h_list;
1246
Joel Beckerd401dc12009-02-13 02:24:10 -08001247 status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
Joel Becker13723d02008-10-17 19:25:01 -07001248 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001249 if (status < 0) {
1250 mlog_errno(status);
1251 goto bail;
1252 }
1253
1254 eb->h_next_leaf_blk = 0;
1255 eb_el->l_tree_depth = cpu_to_le16(i);
1256 eb_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehdcd05382007-01-16 11:32:23 -08001257 /*
1258 * This actually counts as an empty extent as
1259 * c_clusters == 0
1260 */
1261 eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001262 eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehe48edee2007-03-07 16:46:57 -08001263 /*
1264 * eb_el isn't always an interior node, but even leaf
1265 * nodes want a zero'd flags and reserved field so
1266 * this gets the whole 32 bits regardless of use.
1267 */
1268 eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001269 if (!eb_el->l_tree_depth)
1270 new_last_eb_blk = le64_to_cpu(eb->h_blkno);
1271
Joel Beckerec20cec2010-03-19 14:13:52 -07001272 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001273 next_blkno = le64_to_cpu(eb->h_blkno);
1274 }
1275
1276 /* This is a bit hairy. We want to update up to three blocks
1277 * here without leaving any of them in an inconsistent state
1278 * in case of error. We don't have to worry about
1279 * journal_dirty erroring as it won't unless we've aborted the
1280 * handle (in which case we would never be here) so reserving
1281 * the write with journal_access is all we need to do. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001282 status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001283 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001284 if (status < 0) {
1285 mlog_errno(status);
1286 goto bail;
1287 }
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001288 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001289 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001290 if (status < 0) {
1291 mlog_errno(status);
1292 goto bail;
1293 }
1294 if (eb_bh) {
Joel Beckerd401dc12009-02-13 02:24:10 -08001295 status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001296 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001297 if (status < 0) {
1298 mlog_errno(status);
1299 goto bail;
1300 }
1301 }
1302
1303 /* Link the new branch into the rest of the tree (el will
Tao Mae7d4cb62008-08-18 17:38:44 +08001304 * either be on the root_bh, or the extent block passed in. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001305 i = le16_to_cpu(el->l_next_free_rec);
1306 el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08001307 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001308 el->l_recs[i].e_int_clusters = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001309 le16_add_cpu(&el->l_next_free_rec, 1);
1310
1311 /* fe needs a new last extent block pointer, as does the
1312 * next_leaf on the previously last-extent-block. */
Joel Becker35dc0aa2008-08-20 16:25:06 -07001313 ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
Mark Fashehccd979b2005-12-15 14:31:24 -08001314
Mark Fasheh328d5752007-06-18 10:48:04 -07001315 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001316 eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
1317
Joel Beckerec20cec2010-03-19 14:13:52 -07001318 ocfs2_journal_dirty(handle, *last_eb_bh);
1319 ocfs2_journal_dirty(handle, et->et_root_bh);
1320 if (eb_bh)
1321 ocfs2_journal_dirty(handle, eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001322
Mark Fasheh328d5752007-06-18 10:48:04 -07001323 /*
1324 * Some callers want to track the rightmost leaf so pass it
1325 * back here.
1326 */
1327 brelse(*last_eb_bh);
1328 get_bh(new_eb_bhs[0]);
1329 *last_eb_bh = new_eb_bhs[0];
1330
Mark Fashehccd979b2005-12-15 14:31:24 -08001331 status = 0;
1332bail:
1333 if (new_eb_bhs) {
1334 for (i = 0; i < new_blocks; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001335 brelse(new_eb_bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001336 kfree(new_eb_bhs);
1337 }
1338
Mark Fashehccd979b2005-12-15 14:31:24 -08001339 return status;
1340}
1341
1342/*
1343 * adds another level to the allocation tree.
1344 * returns back the new extent block so you can add a branch to it
1345 * after this call.
1346 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001347static int ocfs2_shift_tree_depth(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08001348 struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001349 struct ocfs2_alloc_context *meta_ac,
1350 struct buffer_head **ret_new_eb_bh)
1351{
Changwei Ge71a36942018-01-31 16:15:06 -08001352 int status, i, block_given = 0;
Mark Fashehdcd05382007-01-16 11:32:23 -08001353 u32 new_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001354 struct buffer_head *new_eb_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001355 struct ocfs2_extent_block *eb;
Tao Mae7d4cb62008-08-18 17:38:44 +08001356 struct ocfs2_extent_list *root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001357 struct ocfs2_extent_list *eb_el;
1358
Changwei Ge71a36942018-01-31 16:15:06 -08001359 if (!ocfs2_is_dealloc_empty(et)) {
1360 status = ocfs2_reuse_blk_from_dealloc(handle, et,
1361 &new_eb_bh, 1,
1362 &block_given);
1363 } else if (meta_ac) {
1364 status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
1365 &new_eb_bh);
1366
1367 } else {
1368 BUG();
1369 }
1370
Mark Fashehccd979b2005-12-15 14:31:24 -08001371 if (status < 0) {
1372 mlog_errno(status);
1373 goto bail;
1374 }
1375
1376 eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
Joel Becker5e965812008-11-13 14:49:16 -08001377 /* ocfs2_create_new_meta_bhs() should create it right! */
1378 BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
Mark Fashehccd979b2005-12-15 14:31:24 -08001379
1380 eb_el = &eb->h_list;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001381 root_el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001382
Joel Beckerd401dc12009-02-13 02:24:10 -08001383 status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
Joel Becker13723d02008-10-17 19:25:01 -07001384 OCFS2_JOURNAL_ACCESS_CREATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001385 if (status < 0) {
1386 mlog_errno(status);
1387 goto bail;
1388 }
1389
Tao Mae7d4cb62008-08-18 17:38:44 +08001390 /* copy the root extent list data into the new extent block */
1391 eb_el->l_tree_depth = root_el->l_tree_depth;
1392 eb_el->l_next_free_rec = root_el->l_next_free_rec;
1393 for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1394 eb_el->l_recs[i] = root_el->l_recs[i];
Mark Fashehccd979b2005-12-15 14:31:24 -08001395
Joel Beckerec20cec2010-03-19 14:13:52 -07001396 ocfs2_journal_dirty(handle, new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001397
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08001398 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07001399 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001400 if (status < 0) {
1401 mlog_errno(status);
1402 goto bail;
1403 }
1404
Mark Fashehdcd05382007-01-16 11:32:23 -08001405 new_clusters = ocfs2_sum_rightmost_rec(eb_el);
1406
Tao Mae7d4cb62008-08-18 17:38:44 +08001407 /* update root_bh now */
1408 le16_add_cpu(&root_el->l_tree_depth, 1);
1409 root_el->l_recs[0].e_cpos = 0;
1410 root_el->l_recs[0].e_blkno = eb->h_blkno;
1411 root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
1412 for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1413 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
1414 root_el->l_next_free_rec = cpu_to_le16(1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001415
1416 /* If this is our 1st tree depth shift, then last_eb_blk
1417 * becomes the allocated extent block */
Tao Mae7d4cb62008-08-18 17:38:44 +08001418 if (root_el->l_tree_depth == cpu_to_le16(1))
Joel Becker35dc0aa2008-08-20 16:25:06 -07001419 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001420
Joel Beckerec20cec2010-03-19 14:13:52 -07001421 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001422
1423 *ret_new_eb_bh = new_eb_bh;
1424 new_eb_bh = NULL;
1425 status = 0;
1426bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001427 brelse(new_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001428
Mark Fashehccd979b2005-12-15 14:31:24 -08001429 return status;
1430}
1431
1432/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001433 * Should only be called when there is no space left in any of the
1434 * leaf nodes. What we want to do is find the lowest tree depth
1435 * non-leaf extent block with room for new records. There are three
1436 * valid results of this search:
1437 *
1438 * 1) a lowest extent block is found, then we pass it back in
1439 * *lowest_eb_bh and return '0'
1440 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001441 * 2) the search fails to find anything, but the root_el has room. We
Mark Fashehccd979b2005-12-15 14:31:24 -08001442 * pass NULL back in *lowest_eb_bh, but still return '0'
1443 *
Tao Mae7d4cb62008-08-18 17:38:44 +08001444 * 3) the search fails to find anything AND the root_el is full, in
Mark Fashehccd979b2005-12-15 14:31:24 -08001445 * which case we return > 0
1446 *
1447 * return status < 0 indicates an error.
1448 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001449static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
Mark Fashehccd979b2005-12-15 14:31:24 -08001450 struct buffer_head **target_bh)
1451{
1452 int status = 0, i;
1453 u64 blkno;
Mark Fashehccd979b2005-12-15 14:31:24 -08001454 struct ocfs2_extent_block *eb;
1455 struct ocfs2_extent_list *el;
1456 struct buffer_head *bh = NULL;
1457 struct buffer_head *lowest_bh = NULL;
1458
Mark Fashehccd979b2005-12-15 14:31:24 -08001459 *target_bh = NULL;
1460
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001461 el = et->et_root_el;
Mark Fashehccd979b2005-12-15 14:31:24 -08001462
1463 while(le16_to_cpu(el->l_tree_depth) > 1) {
1464 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Jun Piao93f59202018-08-17 15:44:24 -07001465 status = ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1466 "Owner %llu has empty extent list (next_free_rec == 0)\n",
1467 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fashehccd979b2005-12-15 14:31:24 -08001468 goto bail;
1469 }
1470 i = le16_to_cpu(el->l_next_free_rec) - 1;
1471 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1472 if (!blkno) {
Jun Piao93f59202018-08-17 15:44:24 -07001473 status = ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1474 "Owner %llu has extent list where extent # %d has no physical block start\n",
1475 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
Mark Fashehccd979b2005-12-15 14:31:24 -08001476 goto bail;
1477 }
1478
Mark Fasheha81cb882008-10-07 14:25:16 -07001479 brelse(bh);
1480 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001481
Joel Becker3d03a302009-02-12 17:49:26 -08001482 status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001483 if (status < 0) {
1484 mlog_errno(status);
1485 goto bail;
1486 }
1487
1488 eb = (struct ocfs2_extent_block *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08001489 el = &eb->h_list;
1490
1491 if (le16_to_cpu(el->l_next_free_rec) <
1492 le16_to_cpu(el->l_count)) {
Mark Fasheha81cb882008-10-07 14:25:16 -07001493 brelse(lowest_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001494 lowest_bh = bh;
1495 get_bh(lowest_bh);
1496 }
1497 }
1498
1499 /* If we didn't find one and the fe doesn't have any room,
1500 * then return '1' */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001501 el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001502 if (!lowest_bh && (el->l_next_free_rec == el->l_count))
Mark Fashehccd979b2005-12-15 14:31:24 -08001503 status = 1;
1504
1505 *target_bh = lowest_bh;
1506bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001507 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001508
Mark Fashehccd979b2005-12-15 14:31:24 -08001509 return status;
1510}
1511
Mark Fashehe48edee2007-03-07 16:46:57 -08001512/*
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001513 * Grow a b-tree so that it has more records.
1514 *
1515 * We might shift the tree depth in which case existing paths should
1516 * be considered invalid.
1517 *
1518 * Tree depth after the grow is returned via *final_depth.
Mark Fasheh328d5752007-06-18 10:48:04 -07001519 *
1520 * *last_eb_bh will be updated by ocfs2_add_branch().
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001521 */
Joel Beckerd401dc12009-02-13 02:24:10 -08001522static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
1523 int *final_depth, struct buffer_head **last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001524 struct ocfs2_alloc_context *meta_ac)
1525{
1526 int ret, shift;
Joel Beckerce1d9ea2008-08-20 16:30:07 -07001527 struct ocfs2_extent_list *el = et->et_root_el;
Tao Mae7d4cb62008-08-18 17:38:44 +08001528 int depth = le16_to_cpu(el->l_tree_depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001529 struct buffer_head *bh = NULL;
1530
Changwei Ge71a36942018-01-31 16:15:06 -08001531 BUG_ON(meta_ac == NULL && ocfs2_is_dealloc_empty(et));
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001532
Joel Beckerd401dc12009-02-13 02:24:10 -08001533 shift = ocfs2_find_branch_target(et, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001534 if (shift < 0) {
1535 ret = shift;
1536 mlog_errno(ret);
1537 goto out;
1538 }
1539
1540 /* We traveled all the way to the bottom of the allocation tree
1541 * and didn't find room for any more extents - we need to add
1542 * another tree level */
1543 if (shift) {
1544 BUG_ON(bh);
Tao Maa09d09b2011-02-22 08:21:10 +08001545 trace_ocfs2_grow_tree(
1546 (unsigned long long)
1547 ocfs2_metadata_cache_owner(et->et_ci),
1548 depth);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001549
1550 /* ocfs2_shift_tree_depth will return us a buffer with
1551 * the new extent block (so we can pass that to
1552 * ocfs2_add_branch). */
Joel Beckerd401dc12009-02-13 02:24:10 -08001553 ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001554 if (ret < 0) {
1555 mlog_errno(ret);
1556 goto out;
1557 }
1558 depth++;
Mark Fasheh328d5752007-06-18 10:48:04 -07001559 if (depth == 1) {
1560 /*
1561 * Special case: we have room now if we shifted from
1562 * tree_depth 0, so no more work needs to be done.
1563 *
1564 * We won't be calling add_branch, so pass
1565 * back *last_eb_bh as the new leaf. At depth
1566 * zero, it should always be null so there's
1567 * no reason to brelse.
1568 */
1569 BUG_ON(*last_eb_bh);
1570 get_bh(bh);
1571 *last_eb_bh = bh;
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001572 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07001573 }
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001574 }
1575
1576 /* call ocfs2_add_branch to add the final part of the tree with
1577 * the new data. */
Joel Beckerd401dc12009-02-13 02:24:10 -08001578 ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001579 meta_ac);
wangyan229ba1f2018-08-17 15:44:27 -07001580 if (ret < 0)
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001581 mlog_errno(ret);
Mark Fashehc3afcbb2007-05-29 14:28:51 -07001582
1583out:
1584 if (final_depth)
1585 *final_depth = depth;
1586 brelse(bh);
1587 return ret;
1588}
1589
1590/*
Mark Fashehdcd05382007-01-16 11:32:23 -08001591 * This function will discard the rightmost extent record.
1592 */
1593static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
1594{
1595 int next_free = le16_to_cpu(el->l_next_free_rec);
1596 int count = le16_to_cpu(el->l_count);
1597 unsigned int num_bytes;
1598
1599 BUG_ON(!next_free);
1600 /* This will cause us to go off the end of our extent list. */
1601 BUG_ON(next_free >= count);
1602
1603 num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
1604
1605 memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
1606}
1607
1608static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
1609 struct ocfs2_extent_rec *insert_rec)
1610{
1611 int i, insert_index, next_free, has_empty, num_bytes;
1612 u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
1613 struct ocfs2_extent_rec *rec;
1614
1615 next_free = le16_to_cpu(el->l_next_free_rec);
1616 has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
1617
1618 BUG_ON(!next_free);
1619
1620 /* The tree code before us didn't allow enough room in the leaf. */
Julia Lawallb1f35502008-03-04 15:21:05 -08001621 BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
Mark Fashehdcd05382007-01-16 11:32:23 -08001622
1623 /*
1624 * The easiest way to approach this is to just remove the
1625 * empty extent and temporarily decrement next_free.
1626 */
1627 if (has_empty) {
1628 /*
1629 * If next_free was 1 (only an empty extent), this
1630 * loop won't execute, which is fine. We still want
1631 * the decrement above to happen.
1632 */
1633 for(i = 0; i < (next_free - 1); i++)
1634 el->l_recs[i] = el->l_recs[i+1];
1635
1636 next_free--;
1637 }
1638
1639 /*
1640 * Figure out what the new record index should be.
1641 */
1642 for(i = 0; i < next_free; i++) {
1643 rec = &el->l_recs[i];
1644
1645 if (insert_cpos < le32_to_cpu(rec->e_cpos))
1646 break;
1647 }
1648 insert_index = i;
1649
Tao Maa09d09b2011-02-22 08:21:10 +08001650 trace_ocfs2_rotate_leaf(insert_cpos, insert_index,
1651 has_empty, next_free,
1652 le16_to_cpu(el->l_count));
Mark Fashehdcd05382007-01-16 11:32:23 -08001653
1654 BUG_ON(insert_index < 0);
1655 BUG_ON(insert_index >= le16_to_cpu(el->l_count));
1656 BUG_ON(insert_index > next_free);
1657
1658 /*
1659 * No need to memmove if we're just adding to the tail.
1660 */
1661 if (insert_index != next_free) {
1662 BUG_ON(next_free >= le16_to_cpu(el->l_count));
1663
1664 num_bytes = next_free - insert_index;
1665 num_bytes *= sizeof(struct ocfs2_extent_rec);
1666 memmove(&el->l_recs[insert_index + 1],
1667 &el->l_recs[insert_index],
1668 num_bytes);
1669 }
1670
1671 /*
1672 * Either we had an empty extent, and need to re-increment or
1673 * there was no empty extent on a non full rightmost leaf node,
1674 * in which case we still need to increment.
1675 */
1676 next_free++;
1677 el->l_next_free_rec = cpu_to_le16(next_free);
1678 /*
1679 * Make sure none of the math above just messed up our tree.
1680 */
1681 BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
1682
1683 el->l_recs[insert_index] = *insert_rec;
1684
1685}
1686
Mark Fasheh328d5752007-06-18 10:48:04 -07001687static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
1688{
1689 int size, num_recs = le16_to_cpu(el->l_next_free_rec);
1690
1691 BUG_ON(num_recs == 0);
1692
1693 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
1694 num_recs--;
1695 size = num_recs * sizeof(struct ocfs2_extent_rec);
1696 memmove(&el->l_recs[0], &el->l_recs[1], size);
1697 memset(&el->l_recs[num_recs], 0,
1698 sizeof(struct ocfs2_extent_rec));
1699 el->l_next_free_rec = cpu_to_le16(num_recs);
1700 }
1701}
1702
Mark Fashehdcd05382007-01-16 11:32:23 -08001703/*
1704 * Create an empty extent record .
1705 *
1706 * l_next_free_rec may be updated.
1707 *
1708 * If an empty extent already exists do nothing.
1709 */
1710static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
1711{
1712 int next_free = le16_to_cpu(el->l_next_free_rec);
1713
Mark Fashehe48edee2007-03-07 16:46:57 -08001714 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
1715
Mark Fashehdcd05382007-01-16 11:32:23 -08001716 if (next_free == 0)
1717 goto set_and_inc;
1718
1719 if (ocfs2_is_empty_extent(&el->l_recs[0]))
1720 return;
1721
1722 mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
1723 "Asked to create an empty extent in a full list:\n"
1724 "count = %u, tree depth = %u",
1725 le16_to_cpu(el->l_count),
1726 le16_to_cpu(el->l_tree_depth));
1727
1728 ocfs2_shift_records_right(el);
1729
1730set_and_inc:
1731 le16_add_cpu(&el->l_next_free_rec, 1);
1732 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
1733}
1734
1735/*
1736 * For a rotation which involves two leaf nodes, the "root node" is
1737 * the lowest level tree node which contains a path to both leafs. This
1738 * resulting set of information can be used to form a complete "subtree"
1739 *
1740 * This function is passed two full paths from the dinode down to a
1741 * pair of adjacent leaves. It's task is to figure out which path
1742 * index contains the subtree root - this can be the root index itself
1743 * in a worst-case rotation.
1744 *
1745 * The array index of the subtree root is passed back.
1746 */
Tao Ma38a04e42009-11-30 14:32:19 +08001747int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
1748 struct ocfs2_path *left,
1749 struct ocfs2_path *right)
Mark Fashehdcd05382007-01-16 11:32:23 -08001750{
1751 int i = 0;
1752
1753 /*
1754 * Check that the caller passed in two paths from the same tree.
1755 */
1756 BUG_ON(path_root_bh(left) != path_root_bh(right));
1757
1758 do {
1759 i++;
1760
1761 /*
1762 * The caller didn't pass two adjacent paths.
1763 */
1764 mlog_bug_on_msg(i > left->p_tree_depth,
Joel Becker7dc02802009-02-12 19:20:13 -08001765 "Owner %llu, left depth %u, right depth %u\n"
Mark Fashehdcd05382007-01-16 11:32:23 -08001766 "left leaf blk %llu, right leaf blk %llu\n",
Joel Becker7dc02802009-02-12 19:20:13 -08001767 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
1768 left->p_tree_depth, right->p_tree_depth,
Mark Fashehdcd05382007-01-16 11:32:23 -08001769 (unsigned long long)path_leaf_bh(left)->b_blocknr,
1770 (unsigned long long)path_leaf_bh(right)->b_blocknr);
1771 } while (left->p_node[i].bh->b_blocknr ==
1772 right->p_node[i].bh->b_blocknr);
1773
1774 return i - 1;
1775}
1776
1777typedef void (path_insert_t)(void *, struct buffer_head *);
1778
1779/*
1780 * Traverse a btree path in search of cpos, starting at root_el.
1781 *
1782 * This code can be called with a cpos larger than the tree, in which
1783 * case it will return the rightmost path.
1784 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001785static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
Mark Fashehdcd05382007-01-16 11:32:23 -08001786 struct ocfs2_extent_list *root_el, u32 cpos,
1787 path_insert_t *func, void *data)
1788{
1789 int i, ret = 0;
1790 u32 range;
1791 u64 blkno;
1792 struct buffer_head *bh = NULL;
1793 struct ocfs2_extent_block *eb;
1794 struct ocfs2_extent_list *el;
1795 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08001796
1797 el = root_el;
1798 while (el->l_tree_depth) {
1799 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001800 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001801 "Owner %llu has empty extent list at depth %u\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001802 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001803 le16_to_cpu(el->l_tree_depth));
1804 ret = -EROFS;
1805 goto out;
1806
1807 }
1808
1809 for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
1810 rec = &el->l_recs[i];
1811
1812 /*
1813 * In the case that cpos is off the allocation
1814 * tree, this should just wind up returning the
1815 * rightmost record.
1816 */
1817 range = le32_to_cpu(rec->e_cpos) +
Mark Fashehe48edee2007-03-07 16:46:57 -08001818 ocfs2_rec_clusters(el, rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08001819 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
1820 break;
1821 }
1822
1823 blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1824 if (blkno == 0) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001825 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001826 "Owner %llu has bad blkno in extent list at depth %u (index %d)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001827 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001828 le16_to_cpu(el->l_tree_depth), i);
1829 ret = -EROFS;
1830 goto out;
1831 }
1832
1833 brelse(bh);
1834 bh = NULL;
Joel Beckerfacdb772009-02-12 18:08:48 -08001835 ret = ocfs2_read_extent_block(ci, blkno, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001836 if (ret) {
1837 mlog_errno(ret);
1838 goto out;
1839 }
1840
1841 eb = (struct ocfs2_extent_block *) bh->b_data;
1842 el = &eb->h_list;
Mark Fashehdcd05382007-01-16 11:32:23 -08001843
1844 if (le16_to_cpu(el->l_next_free_rec) >
1845 le16_to_cpu(el->l_count)) {
Joel Beckerfacdb772009-02-12 18:08:48 -08001846 ocfs2_error(ocfs2_metadata_cache_get_super(ci),
Joe Perches7ecef142015-09-04 15:44:51 -07001847 "Owner %llu has bad count in extent list at block %llu (next free=%u, count=%u)\n",
Joel Beckerfacdb772009-02-12 18:08:48 -08001848 (unsigned long long)ocfs2_metadata_cache_owner(ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08001849 (unsigned long long)bh->b_blocknr,
1850 le16_to_cpu(el->l_next_free_rec),
1851 le16_to_cpu(el->l_count));
1852 ret = -EROFS;
1853 goto out;
1854 }
1855
1856 if (func)
1857 func(data, bh);
1858 }
1859
1860out:
1861 /*
1862 * Catch any trailing bh that the loop didn't handle.
1863 */
1864 brelse(bh);
1865
1866 return ret;
1867}
1868
1869/*
1870 * Given an initialized path (that is, it has a valid root extent
1871 * list), this function will traverse the btree in search of the path
1872 * which would contain cpos.
1873 *
1874 * The path traveled is recorded in the path structure.
1875 *
1876 * Note that this will not do any comparisons on leaf node extent
1877 * records, so it will work fine in the case that we just added a tree
1878 * branch.
1879 */
1880struct find_path_data {
1881 int index;
1882 struct ocfs2_path *path;
1883};
1884static void find_path_ins(void *data, struct buffer_head *bh)
1885{
1886 struct find_path_data *fp = data;
1887
1888 get_bh(bh);
1889 ocfs2_path_insert_eb(fp->path, fp->index, bh);
1890 fp->index++;
1891}
Tao Mae2e9f602009-08-18 11:22:34 +08001892int ocfs2_find_path(struct ocfs2_caching_info *ci,
1893 struct ocfs2_path *path, u32 cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08001894{
1895 struct find_path_data data;
1896
1897 data.index = 1;
1898 data.path = path;
Joel Beckerfacdb772009-02-12 18:08:48 -08001899 return __ocfs2_find_path(ci, path_root_el(path), cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08001900 find_path_ins, &data);
1901}
1902
1903static void find_leaf_ins(void *data, struct buffer_head *bh)
1904{
1905 struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
1906 struct ocfs2_extent_list *el = &eb->h_list;
1907 struct buffer_head **ret = data;
1908
1909 /* We want to retain only the leaf block. */
1910 if (le16_to_cpu(el->l_tree_depth) == 0) {
1911 get_bh(bh);
1912 *ret = bh;
1913 }
1914}
1915/*
1916 * Find the leaf block in the tree which would contain cpos. No
1917 * checking of the actual leaf is done.
1918 *
1919 * Some paths want to call this instead of allocating a path structure
1920 * and calling ocfs2_find_path().
1921 *
1922 * This function doesn't handle non btree extent lists.
1923 */
Joel Beckerfacdb772009-02-12 18:08:48 -08001924int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
1925 struct ocfs2_extent_list *root_el, u32 cpos,
1926 struct buffer_head **leaf_bh)
Mark Fashehdcd05382007-01-16 11:32:23 -08001927{
1928 int ret;
1929 struct buffer_head *bh = NULL;
1930
Joel Beckerfacdb772009-02-12 18:08:48 -08001931 ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08001932 if (ret) {
1933 mlog_errno(ret);
1934 goto out;
1935 }
1936
1937 *leaf_bh = bh;
1938out:
1939 return ret;
1940}
1941
1942/*
1943 * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
1944 *
1945 * Basically, we've moved stuff around at the bottom of the tree and
1946 * we need to fix up the extent records above the changes to reflect
1947 * the new changes.
1948 *
1949 * left_rec: the record on the left.
Mark Fashehdcd05382007-01-16 11:32:23 -08001950 * right_rec: the record to the right of left_rec
1951 * right_child_el: is the child list pointed to by right_rec
1952 *
1953 * By definition, this only works on interior nodes.
1954 */
1955static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08001956 struct ocfs2_extent_rec *right_rec,
1957 struct ocfs2_extent_list *right_child_el)
1958{
1959 u32 left_clusters, right_end;
1960
1961 /*
1962 * Interior nodes never have holes. Their cpos is the cpos of
1963 * the leftmost record in their child list. Their cluster
1964 * count covers the full theoretical range of their child list
1965 * - the range between their cpos and the cpos of the record
1966 * immediately to their right.
1967 */
1968 left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
Tao Ma82e12642009-07-23 08:12:58 +08001969 if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
1970 BUG_ON(right_child_el->l_tree_depth);
Mark Fasheh328d5752007-06-18 10:48:04 -07001971 BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
1972 left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
1973 }
Mark Fashehdcd05382007-01-16 11:32:23 -08001974 left_clusters -= le32_to_cpu(left_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001975 left_rec->e_int_clusters = cpu_to_le32(left_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001976
1977 /*
1978 * Calculate the rightmost cluster count boundary before
Mark Fashehe48edee2007-03-07 16:46:57 -08001979 * moving cpos - we will need to adjust clusters after
Mark Fashehdcd05382007-01-16 11:32:23 -08001980 * updating e_cpos to keep the same highest cluster count.
1981 */
1982 right_end = le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001983 right_end += le32_to_cpu(right_rec->e_int_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08001984
1985 right_rec->e_cpos = left_rec->e_cpos;
1986 le32_add_cpu(&right_rec->e_cpos, left_clusters);
1987
1988 right_end -= le32_to_cpu(right_rec->e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08001989 right_rec->e_int_clusters = cpu_to_le32(right_end);
Mark Fashehdcd05382007-01-16 11:32:23 -08001990}
1991
1992/*
1993 * Adjust the adjacent root node records involved in a
1994 * rotation. left_el_blkno is passed in as a key so that we can easily
1995 * find it's index in the root list.
1996 */
1997static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
1998 struct ocfs2_extent_list *left_el,
1999 struct ocfs2_extent_list *right_el,
2000 u64 left_el_blkno)
2001{
2002 int i;
2003
2004 BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
2005 le16_to_cpu(left_el->l_tree_depth));
2006
2007 for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
2008 if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
2009 break;
2010 }
2011
2012 /*
2013 * The path walking code should have never returned a root and
2014 * two paths which are not adjacent.
2015 */
2016 BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
2017
Jun Piao964f14a2017-09-06 16:19:11 -07002018 ocfs2_adjust_adjacent_records(&root_el->l_recs[i],
Mark Fashehdcd05382007-01-16 11:32:23 -08002019 &root_el->l_recs[i + 1], right_el);
2020}
2021
2022/*
2023 * We've changed a leaf block (in right_path) and need to reflect that
2024 * change back up the subtree.
2025 *
2026 * This happens in multiple places:
2027 * - When we've moved an extent record from the left path leaf to the right
2028 * path leaf to make room for an empty extent in the left path leaf.
2029 * - When our insert into the right path leaf is at the leftmost edge
2030 * and requires an update of the path immediately to it's left. This
2031 * can occur at the end of some types of rotation and appending inserts.
Tao Ma677b9752008-01-30 14:21:05 +08002032 * - When we've adjusted the last extent record in the left path leaf and the
2033 * 1st extent record in the right path leaf during cross extent block merge.
Mark Fashehdcd05382007-01-16 11:32:23 -08002034 */
Joel Becker4619c732009-02-12 19:02:36 -08002035static void ocfs2_complete_edge_insert(handle_t *handle,
Mark Fashehdcd05382007-01-16 11:32:23 -08002036 struct ocfs2_path *left_path,
2037 struct ocfs2_path *right_path,
2038 int subtree_index)
2039{
Joel Beckerec20cec2010-03-19 14:13:52 -07002040 int i, idx;
Mark Fashehdcd05382007-01-16 11:32:23 -08002041 struct ocfs2_extent_list *el, *left_el, *right_el;
2042 struct ocfs2_extent_rec *left_rec, *right_rec;
Colin Ian Kingf0188442022-01-14 14:03:45 -08002043 struct buffer_head *root_bh;
Mark Fashehdcd05382007-01-16 11:32:23 -08002044
2045 /*
2046 * Update the counts and position values within all the
2047 * interior nodes to reflect the leaf rotation we just did.
2048 *
2049 * The root node is handled below the loop.
2050 *
2051 * We begin the loop with right_el and left_el pointing to the
2052 * leaf lists and work our way up.
2053 *
2054 * NOTE: within this loop, left_el and right_el always refer
2055 * to the *child* lists.
2056 */
2057 left_el = path_leaf_el(left_path);
2058 right_el = path_leaf_el(right_path);
2059 for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
Tao Maa09d09b2011-02-22 08:21:10 +08002060 trace_ocfs2_complete_edge_insert(i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002061
2062 /*
2063 * One nice property of knowing that all of these
2064 * nodes are below the root is that we only deal with
2065 * the leftmost right node record and the rightmost
2066 * left node record.
2067 */
2068 el = left_path->p_node[i].el;
2069 idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
2070 left_rec = &el->l_recs[idx];
2071
2072 el = right_path->p_node[i].el;
2073 right_rec = &el->l_recs[0];
2074
Jun Piao964f14a2017-09-06 16:19:11 -07002075 ocfs2_adjust_adjacent_records(left_rec, right_rec, right_el);
Mark Fashehdcd05382007-01-16 11:32:23 -08002076
Joel Beckerec20cec2010-03-19 14:13:52 -07002077 ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
2078 ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002079
2080 /*
2081 * Setup our list pointers now so that the current
2082 * parents become children in the next iteration.
2083 */
2084 left_el = left_path->p_node[i].el;
2085 right_el = right_path->p_node[i].el;
2086 }
2087
2088 /*
2089 * At the root node, adjust the two adjacent records which
2090 * begin our path to the leaves.
2091 */
2092
2093 el = left_path->p_node[subtree_index].el;
2094 left_el = left_path->p_node[subtree_index + 1].el;
2095 right_el = right_path->p_node[subtree_index + 1].el;
2096
2097 ocfs2_adjust_root_records(el, left_el, right_el,
2098 left_path->p_node[subtree_index + 1].bh->b_blocknr);
2099
2100 root_bh = left_path->p_node[subtree_index].bh;
2101
Joel Beckerec20cec2010-03-19 14:13:52 -07002102 ocfs2_journal_dirty(handle, root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002103}
2104
Joel Becker5c601ab2009-02-12 19:10:13 -08002105static int ocfs2_rotate_subtree_right(handle_t *handle,
2106 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08002107 struct ocfs2_path *left_path,
2108 struct ocfs2_path *right_path,
2109 int subtree_index)
2110{
2111 int ret, i;
2112 struct buffer_head *right_leaf_bh;
2113 struct buffer_head *left_leaf_bh = NULL;
2114 struct buffer_head *root_bh;
2115 struct ocfs2_extent_list *right_el, *left_el;
2116 struct ocfs2_extent_rec move_rec;
2117
2118 left_leaf_bh = path_leaf_bh(left_path);
2119 left_el = path_leaf_el(left_path);
2120
2121 if (left_el->l_next_free_rec != left_el->l_count) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002122 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07002123 "Inode %llu has non-full interior leaf node %llu (next free = %u)\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002124 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002125 (unsigned long long)left_leaf_bh->b_blocknr,
2126 le16_to_cpu(left_el->l_next_free_rec));
2127 return -EROFS;
2128 }
2129
2130 /*
2131 * This extent block may already have an empty record, so we
2132 * return early if so.
2133 */
2134 if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
2135 return 0;
2136
2137 root_bh = left_path->p_node[subtree_index].bh;
2138 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2139
Joel Becker5c601ab2009-02-12 19:10:13 -08002140 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002141 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002142 if (ret) {
2143 mlog_errno(ret);
2144 goto out;
2145 }
2146
2147 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker5c601ab2009-02-12 19:10:13 -08002148 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002149 right_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002150 if (ret) {
2151 mlog_errno(ret);
2152 goto out;
2153 }
2154
Joel Becker5c601ab2009-02-12 19:10:13 -08002155 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002156 left_path, i);
Mark Fashehdcd05382007-01-16 11:32:23 -08002157 if (ret) {
2158 mlog_errno(ret);
2159 goto out;
2160 }
2161 }
2162
2163 right_leaf_bh = path_leaf_bh(right_path);
2164 right_el = path_leaf_el(right_path);
2165
2166 /* This is a code error, not a disk corruption. */
2167 mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
2168 "because rightmost leaf block %llu is empty\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002169 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08002170 (unsigned long long)right_leaf_bh->b_blocknr);
2171
2172 ocfs2_create_empty_extent(right_el);
2173
Joel Beckerec20cec2010-03-19 14:13:52 -07002174 ocfs2_journal_dirty(handle, right_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002175
2176 /* Do the copy now. */
2177 i = le16_to_cpu(left_el->l_next_free_rec) - 1;
2178 move_rec = left_el->l_recs[i];
2179 right_el->l_recs[0] = move_rec;
2180
2181 /*
2182 * Clear out the record we just copied and shift everything
2183 * over, leaving an empty extent in the left leaf.
2184 *
2185 * We temporarily subtract from next_free_rec so that the
2186 * shift will lose the tail record (which is now defunct).
2187 */
2188 le16_add_cpu(&left_el->l_next_free_rec, -1);
2189 ocfs2_shift_records_right(left_el);
2190 memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2191 le16_add_cpu(&left_el->l_next_free_rec, 1);
2192
Joel Beckerec20cec2010-03-19 14:13:52 -07002193 ocfs2_journal_dirty(handle, left_leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08002194
Joel Becker4619c732009-02-12 19:02:36 -08002195 ocfs2_complete_edge_insert(handle, left_path, right_path,
2196 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08002197
2198out:
2199 return ret;
2200}
2201
2202/*
2203 * Given a full path, determine what cpos value would return us a path
2204 * containing the leaf immediately to the left of the current one.
2205 *
2206 * Will return zero if the path passed in is already the leftmost path.
2207 */
Tristan Yeee149a72010-05-11 17:54:44 +08002208int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
2209 struct ocfs2_path *path, u32 *cpos)
Mark Fashehdcd05382007-01-16 11:32:23 -08002210{
2211 int i, j, ret = 0;
2212 u64 blkno;
2213 struct ocfs2_extent_list *el;
2214
Mark Fashehe48edee2007-03-07 16:46:57 -08002215 BUG_ON(path->p_tree_depth == 0);
2216
Mark Fashehdcd05382007-01-16 11:32:23 -08002217 *cpos = 0;
2218
2219 blkno = path_leaf_bh(path)->b_blocknr;
2220
2221 /* Start at the tree node just above the leaf and work our way up. */
2222 i = path->p_tree_depth - 1;
2223 while (i >= 0) {
2224 el = path->p_node[i].el;
2225
2226 /*
2227 * Find the extent record just before the one in our
2228 * path.
2229 */
2230 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2231 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2232 if (j == 0) {
2233 if (i == 0) {
2234 /*
2235 * We've determined that the
2236 * path specified is already
2237 * the leftmost one - return a
2238 * cpos of zero.
2239 */
2240 goto out;
2241 }
2242 /*
2243 * The leftmost record points to our
2244 * leaf - we need to travel up the
2245 * tree one level.
2246 */
2247 goto next_node;
2248 }
2249
2250 *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
Mark Fashehe48edee2007-03-07 16:46:57 -08002251 *cpos = *cpos + ocfs2_rec_clusters(el,
2252 &el->l_recs[j - 1]);
2253 *cpos = *cpos - 1;
Mark Fashehdcd05382007-01-16 11:32:23 -08002254 goto out;
2255 }
2256 }
2257
2258 /*
2259 * If we got here, we never found a valid node where
2260 * the tree indicated one should be.
2261 */
Joe Perches7ecef142015-09-04 15:44:51 -07002262 ocfs2_error(sb, "Invalid extent tree at extent block %llu\n",
Mark Fashehdcd05382007-01-16 11:32:23 -08002263 (unsigned long long)blkno);
2264 ret = -EROFS;
2265 goto out;
2266
2267next_node:
2268 blkno = path->p_node[i].bh->b_blocknr;
2269 i--;
2270 }
2271
2272out:
2273 return ret;
2274}
2275
Mark Fasheh328d5752007-06-18 10:48:04 -07002276/*
2277 * Extend the transaction by enough credits to complete the rotation,
2278 * and still leave at least the original number of credits allocated
2279 * to this transaction.
2280 */
Mark Fashehdcd05382007-01-16 11:32:23 -08002281static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
Mark Fasheh328d5752007-06-18 10:48:04 -07002282 int op_credits,
Mark Fashehdcd05382007-01-16 11:32:23 -08002283 struct ocfs2_path *path)
2284{
Tao Mac901fb02010-04-26 14:34:57 +08002285 int ret = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002286 int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
Mark Fashehdcd05382007-01-16 11:32:23 -08002287
Jan Kara9797a902019-11-05 17:44:18 +01002288 if (jbd2_handle_buffer_credits(handle) < credits)
Tao Mac18b8122009-08-18 11:44:07 +08002289 ret = ocfs2_extend_trans(handle,
Jan Kara9797a902019-11-05 17:44:18 +01002290 credits - jbd2_handle_buffer_credits(handle));
Tao Mac18b8122009-08-18 11:44:07 +08002291
Tao Mac901fb02010-04-26 14:34:57 +08002292 return ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08002293}
2294
2295/*
2296 * Trap the case where we're inserting into the theoretical range past
2297 * the _actual_ left leaf range. Otherwise, we'll rotate a record
2298 * whose cpos is less than ours into the right leaf.
2299 *
2300 * It's only necessary to look at the rightmost record of the left
2301 * leaf because the logic that calls us should ensure that the
2302 * theoretical ranges in the path components above the leaves are
2303 * correct.
2304 */
2305static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
2306 u32 insert_cpos)
2307{
2308 struct ocfs2_extent_list *left_el;
2309 struct ocfs2_extent_rec *rec;
2310 int next_free;
2311
2312 left_el = path_leaf_el(left_path);
2313 next_free = le16_to_cpu(left_el->l_next_free_rec);
2314 rec = &left_el->l_recs[next_free - 1];
2315
2316 if (insert_cpos > le32_to_cpu(rec->e_cpos))
2317 return 1;
2318 return 0;
2319}
2320
Mark Fasheh328d5752007-06-18 10:48:04 -07002321static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
2322{
2323 int next_free = le16_to_cpu(el->l_next_free_rec);
2324 unsigned int range;
2325 struct ocfs2_extent_rec *rec;
2326
2327 if (next_free == 0)
2328 return 0;
2329
2330 rec = &el->l_recs[0];
2331 if (ocfs2_is_empty_extent(rec)) {
2332 /* Empty list. */
2333 if (next_free == 1)
2334 return 0;
2335 rec = &el->l_recs[1];
2336 }
2337
2338 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2339 if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
2340 return 1;
2341 return 0;
2342}
2343
Mark Fashehdcd05382007-01-16 11:32:23 -08002344/*
2345 * Rotate all the records in a btree right one record, starting at insert_cpos.
2346 *
2347 * The path to the rightmost leaf should be passed in.
2348 *
2349 * The array is assumed to be large enough to hold an entire path (tree depth).
2350 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002351 * Upon successful return from this function:
Mark Fashehdcd05382007-01-16 11:32:23 -08002352 *
2353 * - The 'right_path' array will contain a path to the leaf block
2354 * whose range contains e_cpos.
2355 * - That leaf block will have a single empty extent in list index 0.
2356 * - In the case that the rotation requires a post-insert update,
2357 * *ret_left_path will contain a valid path which can be passed to
2358 * ocfs2_insert_path().
2359 */
Joel Becker1bbf0b82009-02-12 19:42:08 -08002360static int ocfs2_rotate_tree_right(handle_t *handle,
Joel Becker5c601ab2009-02-12 19:10:13 -08002361 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002362 enum ocfs2_split_type split,
Mark Fashehdcd05382007-01-16 11:32:23 -08002363 u32 insert_cpos,
2364 struct ocfs2_path *right_path,
2365 struct ocfs2_path **ret_left_path)
2366{
Jan Kara9797a902019-11-05 17:44:18 +01002367 int ret, start, orig_credits = jbd2_handle_buffer_credits(handle);
Mark Fashehdcd05382007-01-16 11:32:23 -08002368 u32 cpos;
2369 struct ocfs2_path *left_path = NULL;
Joel Becker5c601ab2009-02-12 19:10:13 -08002370 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehdcd05382007-01-16 11:32:23 -08002371
2372 *ret_left_path = NULL;
2373
Joel Beckerffdd7a52008-10-17 22:32:01 -07002374 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002375 if (!left_path) {
2376 ret = -ENOMEM;
2377 mlog_errno(ret);
2378 goto out;
2379 }
2380
Joel Becker5c601ab2009-02-12 19:10:13 -08002381 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002382 if (ret) {
2383 mlog_errno(ret);
2384 goto out;
2385 }
2386
Tao Maa09d09b2011-02-22 08:21:10 +08002387 trace_ocfs2_rotate_tree_right(
2388 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2389 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002390
2391 /*
2392 * What we want to do here is:
2393 *
2394 * 1) Start with the rightmost path.
2395 *
2396 * 2) Determine a path to the leaf block directly to the left
2397 * of that leaf.
2398 *
2399 * 3) Determine the 'subtree root' - the lowest level tree node
2400 * which contains a path to both leaves.
2401 *
2402 * 4) Rotate the subtree.
2403 *
2404 * 5) Find the next subtree by considering the left path to be
2405 * the new right path.
2406 *
2407 * The check at the top of this while loop also accepts
2408 * insert_cpos == cpos because cpos is only a _theoretical_
2409 * value to get us the left path - insert_cpos might very well
2410 * be filling that hole.
2411 *
2412 * Stop at a cpos of '0' because we either started at the
2413 * leftmost branch (i.e., a tree with one branch and a
2414 * rotation inside of it), or we've gone as far as we can in
2415 * rotating subtrees.
2416 */
2417 while (cpos && insert_cpos <= cpos) {
Tao Maa09d09b2011-02-22 08:21:10 +08002418 trace_ocfs2_rotate_tree_right(
2419 (unsigned long long)
2420 ocfs2_metadata_cache_owner(et->et_ci),
2421 insert_cpos, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002422
Joel Becker5c601ab2009-02-12 19:10:13 -08002423 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002424 if (ret) {
2425 mlog_errno(ret);
2426 goto out;
2427 }
2428
2429 mlog_bug_on_msg(path_leaf_bh(left_path) ==
2430 path_leaf_bh(right_path),
Joel Becker5c601ab2009-02-12 19:10:13 -08002431 "Owner %llu: error during insert of %u "
Mark Fashehdcd05382007-01-16 11:32:23 -08002432 "(left path cpos %u) results in two identical "
2433 "paths ending at %llu\n",
Joel Becker5c601ab2009-02-12 19:10:13 -08002434 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2435 insert_cpos, cpos,
Mark Fashehdcd05382007-01-16 11:32:23 -08002436 (unsigned long long)
2437 path_leaf_bh(left_path)->b_blocknr);
2438
Mark Fasheh328d5752007-06-18 10:48:04 -07002439 if (split == SPLIT_NONE &&
2440 ocfs2_rotate_requires_path_adjustment(left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002441 insert_cpos)) {
Mark Fashehdcd05382007-01-16 11:32:23 -08002442
2443 /*
2444 * We've rotated the tree as much as we
2445 * should. The rest is up to
2446 * ocfs2_insert_path() to complete, after the
2447 * record insertion. We indicate this
2448 * situation by returning the left path.
2449 *
2450 * The reason we don't adjust the records here
2451 * before the record insert is that an error
2452 * later might break the rule where a parent
2453 * record e_cpos will reflect the actual
2454 * e_cpos of the 1st nonempty record of the
2455 * child list.
2456 */
2457 *ret_left_path = left_path;
2458 goto out_ret_path;
2459 }
2460
Joel Becker7dc02802009-02-12 19:20:13 -08002461 start = ocfs2_find_subtree_root(et, left_path, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002462
Tao Maa09d09b2011-02-22 08:21:10 +08002463 trace_ocfs2_rotate_subtree(start,
2464 (unsigned long long)
2465 right_path->p_node[start].bh->b_blocknr,
2466 right_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08002467
2468 ret = ocfs2_extend_rotate_transaction(handle, start,
Mark Fasheh328d5752007-06-18 10:48:04 -07002469 orig_credits, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08002470 if (ret) {
2471 mlog_errno(ret);
2472 goto out;
2473 }
2474
Joel Becker5c601ab2009-02-12 19:10:13 -08002475 ret = ocfs2_rotate_subtree_right(handle, et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08002476 right_path, start);
2477 if (ret) {
2478 mlog_errno(ret);
2479 goto out;
2480 }
2481
Mark Fasheh328d5752007-06-18 10:48:04 -07002482 if (split != SPLIT_NONE &&
2483 ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
2484 insert_cpos)) {
2485 /*
2486 * A rotate moves the rightmost left leaf
2487 * record over to the leftmost right leaf
2488 * slot. If we're doing an extent split
2489 * instead of a real insert, then we have to
2490 * check that the extent to be split wasn't
2491 * just moved over. If it was, then we can
2492 * exit here, passing left_path back -
2493 * ocfs2_split_extent() is smart enough to
2494 * search both leaves.
2495 */
2496 *ret_left_path = left_path;
2497 goto out_ret_path;
2498 }
2499
Mark Fashehdcd05382007-01-16 11:32:23 -08002500 /*
2501 * There is no need to re-read the next right path
2502 * as we know that it'll be our current left
2503 * path. Optimize by copying values instead.
2504 */
2505 ocfs2_mv_path(right_path, left_path);
2506
Joel Becker5c601ab2009-02-12 19:10:13 -08002507 ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08002508 if (ret) {
2509 mlog_errno(ret);
2510 goto out;
2511 }
2512 }
2513
2514out:
2515 ocfs2_free_path(left_path);
2516
2517out_ret_path:
2518 return ret;
2519}
2520
Joel Becker09106ba2009-02-12 19:43:57 -08002521static int ocfs2_update_edge_lengths(handle_t *handle,
2522 struct ocfs2_extent_tree *et,
Jun Piao964f14a2017-09-06 16:19:11 -07002523 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002524{
Tao Ma3c5e1062009-07-21 15:42:05 +08002525 int i, idx, ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002526 struct ocfs2_extent_rec *rec;
2527 struct ocfs2_extent_list *el;
2528 struct ocfs2_extent_block *eb;
2529 u32 range;
2530
Joel Becker09106ba2009-02-12 19:43:57 -08002531 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002532 if (ret) {
2533 mlog_errno(ret);
2534 goto out;
2535 }
2536
Mark Fasheh328d5752007-06-18 10:48:04 -07002537 /* Path should always be rightmost. */
2538 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2539 BUG_ON(eb->h_next_leaf_blk != 0ULL);
2540
2541 el = &eb->h_list;
2542 BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2543 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2544 rec = &el->l_recs[idx];
2545 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2546
2547 for (i = 0; i < path->p_tree_depth; i++) {
2548 el = path->p_node[i].el;
2549 idx = le16_to_cpu(el->l_next_free_rec) - 1;
2550 rec = &el->l_recs[idx];
2551
2552 rec->e_int_clusters = cpu_to_le32(range);
2553 le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2554
2555 ocfs2_journal_dirty(handle, path->p_node[i].bh);
2556 }
Tao Ma3c5e1062009-07-21 15:42:05 +08002557out:
2558 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07002559}
2560
Joel Becker6641b0c2009-02-12 18:57:52 -08002561static void ocfs2_unlink_path(handle_t *handle,
2562 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002563 struct ocfs2_cached_dealloc_ctxt *dealloc,
2564 struct ocfs2_path *path, int unlink_start)
2565{
2566 int ret, i;
2567 struct ocfs2_extent_block *eb;
2568 struct ocfs2_extent_list *el;
2569 struct buffer_head *bh;
2570
2571 for(i = unlink_start; i < path_num_items(path); i++) {
2572 bh = path->p_node[i].bh;
2573
2574 eb = (struct ocfs2_extent_block *)bh->b_data;
2575 /*
2576 * Not all nodes might have had their final count
2577 * decremented by the caller - handle this here.
2578 */
2579 el = &eb->h_list;
2580 if (le16_to_cpu(el->l_next_free_rec) > 1) {
2581 mlog(ML_ERROR,
2582 "Inode %llu, attempted to remove extent block "
2583 "%llu with %u records\n",
Joel Becker6641b0c2009-02-12 18:57:52 -08002584 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fasheh328d5752007-06-18 10:48:04 -07002585 (unsigned long long)le64_to_cpu(eb->h_blkno),
2586 le16_to_cpu(el->l_next_free_rec));
2587
2588 ocfs2_journal_dirty(handle, bh);
Joel Becker6641b0c2009-02-12 18:57:52 -08002589 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002590 continue;
2591 }
2592
2593 el->l_next_free_rec = 0;
2594 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2595
2596 ocfs2_journal_dirty(handle, bh);
2597
2598 ret = ocfs2_cache_extent_block_free(dealloc, eb);
2599 if (ret)
2600 mlog_errno(ret);
2601
Joel Becker6641b0c2009-02-12 18:57:52 -08002602 ocfs2_remove_from_cache(et->et_ci, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002603 }
2604}
2605
Joel Becker6641b0c2009-02-12 18:57:52 -08002606static void ocfs2_unlink_subtree(handle_t *handle,
2607 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002608 struct ocfs2_path *left_path,
2609 struct ocfs2_path *right_path,
2610 int subtree_index,
2611 struct ocfs2_cached_dealloc_ctxt *dealloc)
2612{
2613 int i;
2614 struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2615 struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
Mark Fasheh328d5752007-06-18 10:48:04 -07002616 struct ocfs2_extent_block *eb;
2617
Mark Fasheh328d5752007-06-18 10:48:04 -07002618 eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2619
2620 for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2621 if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2622 break;
2623
2624 BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2625
2626 memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2627 le16_add_cpu(&root_el->l_next_free_rec, -1);
2628
2629 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2630 eb->h_next_leaf_blk = 0;
2631
2632 ocfs2_journal_dirty(handle, root_bh);
2633 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2634
Joel Becker6641b0c2009-02-12 18:57:52 -08002635 ocfs2_unlink_path(handle, et, dealloc, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002636 subtree_index + 1);
2637}
2638
Joel Becker1e2dd632009-02-12 19:45:28 -08002639static int ocfs2_rotate_subtree_left(handle_t *handle,
2640 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07002641 struct ocfs2_path *left_path,
2642 struct ocfs2_path *right_path,
2643 int subtree_index,
2644 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker1e2dd632009-02-12 19:45:28 -08002645 int *deleted)
Mark Fasheh328d5752007-06-18 10:48:04 -07002646{
2647 int ret, i, del_right_subtree = 0, right_has_empty = 0;
Tao Mae7d4cb62008-08-18 17:38:44 +08002648 struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002649 struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2650 struct ocfs2_extent_block *eb;
2651
2652 *deleted = 0;
2653
2654 right_leaf_el = path_leaf_el(right_path);
2655 left_leaf_el = path_leaf_el(left_path);
2656 root_bh = left_path->p_node[subtree_index].bh;
2657 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2658
2659 if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2660 return 0;
2661
2662 eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2663 if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2664 /*
2665 * It's legal for us to proceed if the right leaf is
2666 * the rightmost one and it has an empty extent. There
2667 * are two cases to handle - whether the leaf will be
2668 * empty after removal or not. If the leaf isn't empty
2669 * then just remove the empty extent up front. The
2670 * next block will handle empty leaves by flagging
2671 * them for unlink.
2672 *
2673 * Non rightmost leaves will throw -EAGAIN and the
2674 * caller can manually move the subtree and retry.
2675 */
2676
2677 if (eb->h_next_leaf_blk != 0ULL)
2678 return -EAGAIN;
2679
2680 if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002681 ret = ocfs2_journal_access_eb(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002682 path_leaf_bh(right_path),
2683 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002684 if (ret) {
2685 mlog_errno(ret);
2686 goto out;
2687 }
2688
2689 ocfs2_remove_empty_extent(right_leaf_el);
2690 } else
2691 right_has_empty = 1;
2692 }
2693
2694 if (eb->h_next_leaf_blk == 0ULL &&
2695 le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2696 /*
2697 * We have to update i_last_eb_blk during the meta
2698 * data delete.
2699 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08002700 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07002701 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh328d5752007-06-18 10:48:04 -07002702 if (ret) {
2703 mlog_errno(ret);
2704 goto out;
2705 }
2706
2707 del_right_subtree = 1;
2708 }
2709
2710 /*
2711 * Getting here with an empty extent in the right path implies
2712 * that it's the rightmost path and will be deleted.
2713 */
2714 BUG_ON(right_has_empty && !del_right_subtree);
2715
Joel Becker1e2dd632009-02-12 19:45:28 -08002716 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07002717 subtree_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07002718 if (ret) {
2719 mlog_errno(ret);
2720 goto out;
2721 }
2722
2723 for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
Joel Becker1e2dd632009-02-12 19:45:28 -08002724 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002725 right_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002726 if (ret) {
2727 mlog_errno(ret);
2728 goto out;
2729 }
2730
Joel Becker1e2dd632009-02-12 19:45:28 -08002731 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002732 left_path, i);
Mark Fasheh328d5752007-06-18 10:48:04 -07002733 if (ret) {
2734 mlog_errno(ret);
2735 goto out;
2736 }
2737 }
2738
2739 if (!right_has_empty) {
2740 /*
2741 * Only do this if we're moving a real
2742 * record. Otherwise, the action is delayed until
2743 * after removal of the right path in which case we
2744 * can do a simple shift to remove the empty extent.
2745 */
2746 ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2747 memset(&right_leaf_el->l_recs[0], 0,
2748 sizeof(struct ocfs2_extent_rec));
2749 }
2750 if (eb->h_next_leaf_blk == 0ULL) {
2751 /*
2752 * Move recs over to get rid of empty extent, decrease
2753 * next_free. This is allowed to remove the last
2754 * extent in our leaf (setting l_next_free_rec to
2755 * zero) - the delete code below won't care.
2756 */
2757 ocfs2_remove_empty_extent(right_leaf_el);
2758 }
2759
Joel Beckerec20cec2010-03-19 14:13:52 -07002760 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2761 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07002762
2763 if (del_right_subtree) {
Joel Becker6641b0c2009-02-12 18:57:52 -08002764 ocfs2_unlink_subtree(handle, et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002765 subtree_index, dealloc);
Jun Piao964f14a2017-09-06 16:19:11 -07002766 ret = ocfs2_update_edge_lengths(handle, et, left_path);
Tao Ma3c5e1062009-07-21 15:42:05 +08002767 if (ret) {
2768 mlog_errno(ret);
2769 goto out;
2770 }
Mark Fasheh328d5752007-06-18 10:48:04 -07002771
2772 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07002773 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07002774
2775 /*
2776 * Removal of the extent in the left leaf was skipped
2777 * above so we could delete the right path
2778 * 1st.
2779 */
2780 if (right_has_empty)
2781 ocfs2_remove_empty_extent(left_leaf_el);
2782
Joel Beckerec20cec2010-03-19 14:13:52 -07002783 ocfs2_journal_dirty(handle, et_root_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002784
2785 *deleted = 1;
2786 } else
Joel Becker4619c732009-02-12 19:02:36 -08002787 ocfs2_complete_edge_insert(handle, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002788 subtree_index);
2789
2790out:
2791 return ret;
2792}
2793
2794/*
2795 * Given a full path, determine what cpos value would return us a path
2796 * containing the leaf immediately to the right of the current one.
2797 *
2798 * Will return zero if the path passed in is already the rightmost path.
2799 *
2800 * This looks similar, but is subtly different to
2801 * ocfs2_find_cpos_for_left_leaf().
2802 */
Tao Ma38a04e42009-11-30 14:32:19 +08002803int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2804 struct ocfs2_path *path, u32 *cpos)
Mark Fasheh328d5752007-06-18 10:48:04 -07002805{
2806 int i, j, ret = 0;
2807 u64 blkno;
2808 struct ocfs2_extent_list *el;
2809
2810 *cpos = 0;
2811
2812 if (path->p_tree_depth == 0)
2813 return 0;
2814
2815 blkno = path_leaf_bh(path)->b_blocknr;
2816
2817 /* Start at the tree node just above the leaf and work our way up. */
2818 i = path->p_tree_depth - 1;
2819 while (i >= 0) {
2820 int next_free;
2821
2822 el = path->p_node[i].el;
2823
2824 /*
2825 * Find the extent record just after the one in our
2826 * path.
2827 */
2828 next_free = le16_to_cpu(el->l_next_free_rec);
2829 for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2830 if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2831 if (j == (next_free - 1)) {
2832 if (i == 0) {
2833 /*
2834 * We've determined that the
2835 * path specified is already
2836 * the rightmost one - return a
2837 * cpos of zero.
2838 */
2839 goto out;
2840 }
2841 /*
2842 * The rightmost record points to our
2843 * leaf - we need to travel up the
2844 * tree one level.
2845 */
2846 goto next_node;
2847 }
2848
2849 *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2850 goto out;
2851 }
2852 }
2853
2854 /*
2855 * If we got here, we never found a valid node where
2856 * the tree indicated one should be.
2857 */
Joe Perches7ecef142015-09-04 15:44:51 -07002858 ocfs2_error(sb, "Invalid extent tree at extent block %llu\n",
Mark Fasheh328d5752007-06-18 10:48:04 -07002859 (unsigned long long)blkno);
2860 ret = -EROFS;
2861 goto out;
2862
2863next_node:
2864 blkno = path->p_node[i].bh->b_blocknr;
2865 i--;
2866 }
2867
2868out:
2869 return ret;
2870}
2871
Joel Becker70f18c02009-02-13 02:09:31 -08002872static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2873 struct ocfs2_extent_tree *et,
Joel Becker13723d02008-10-17 19:25:01 -07002874 struct ocfs2_path *path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002875{
2876 int ret;
Joel Becker13723d02008-10-17 19:25:01 -07002877 struct buffer_head *bh = path_leaf_bh(path);
2878 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002879
2880 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2881 return 0;
2882
Joel Becker70f18c02009-02-13 02:09:31 -08002883 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Joel Becker13723d02008-10-17 19:25:01 -07002884 path_num_items(path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07002885 if (ret) {
2886 mlog_errno(ret);
2887 goto out;
2888 }
2889
2890 ocfs2_remove_empty_extent(el);
Joel Beckerec20cec2010-03-19 14:13:52 -07002891 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07002892
2893out:
2894 return ret;
2895}
2896
Joel Beckere46f74d2009-02-12 19:47:43 -08002897static int __ocfs2_rotate_tree_left(handle_t *handle,
2898 struct ocfs2_extent_tree *et,
2899 int orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07002900 struct ocfs2_path *path,
2901 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08002902 struct ocfs2_path **empty_extent_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07002903{
2904 int ret, subtree_root, deleted;
2905 u32 right_cpos;
2906 struct ocfs2_path *left_path = NULL;
2907 struct ocfs2_path *right_path = NULL;
Joel Beckere46f74d2009-02-12 19:47:43 -08002908 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fasheh328d5752007-06-18 10:48:04 -07002909
Xue jiufei099768b2015-06-24 16:55:09 -07002910 if (!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])))
2911 return 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07002912
2913 *empty_extent_path = NULL;
2914
Joel Beckere46f74d2009-02-12 19:47:43 -08002915 ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002916 if (ret) {
2917 mlog_errno(ret);
2918 goto out;
2919 }
2920
Joel Beckerffdd7a52008-10-17 22:32:01 -07002921 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002922 if (!left_path) {
2923 ret = -ENOMEM;
2924 mlog_errno(ret);
2925 goto out;
2926 }
2927
2928 ocfs2_cp_path(left_path, path);
2929
Joel Beckerffdd7a52008-10-17 22:32:01 -07002930 right_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07002931 if (!right_path) {
2932 ret = -ENOMEM;
2933 mlog_errno(ret);
2934 goto out;
2935 }
2936
2937 while (right_cpos) {
Joel Beckerfacdb772009-02-12 18:08:48 -08002938 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07002939 if (ret) {
2940 mlog_errno(ret);
2941 goto out;
2942 }
2943
Joel Becker7dc02802009-02-12 19:20:13 -08002944 subtree_root = ocfs2_find_subtree_root(et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002945 right_path);
2946
Tao Maa09d09b2011-02-22 08:21:10 +08002947 trace_ocfs2_rotate_subtree(subtree_root,
Mark Fasheh328d5752007-06-18 10:48:04 -07002948 (unsigned long long)
2949 right_path->p_node[subtree_root].bh->b_blocknr,
2950 right_path->p_tree_depth);
2951
Xue jiufei17215982016-03-25 14:21:41 -07002952 ret = ocfs2_extend_rotate_transaction(handle, 0,
Mark Fasheh328d5752007-06-18 10:48:04 -07002953 orig_credits, left_path);
2954 if (ret) {
2955 mlog_errno(ret);
2956 goto out;
2957 }
2958
Mark Fashehe8aed342007-12-03 16:43:01 -08002959 /*
2960 * Caller might still want to make changes to the
2961 * tree root, so re-add it to the journal here.
2962 */
Joel Beckere46f74d2009-02-12 19:47:43 -08002963 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07002964 left_path, 0);
Mark Fashehe8aed342007-12-03 16:43:01 -08002965 if (ret) {
2966 mlog_errno(ret);
2967 goto out;
2968 }
2969
Joel Becker1e2dd632009-02-12 19:45:28 -08002970 ret = ocfs2_rotate_subtree_left(handle, et, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07002971 right_path, subtree_root,
Joel Becker1e2dd632009-02-12 19:45:28 -08002972 dealloc, &deleted);
Mark Fasheh328d5752007-06-18 10:48:04 -07002973 if (ret == -EAGAIN) {
2974 /*
2975 * The rotation has to temporarily stop due to
2976 * the right subtree having an empty
2977 * extent. Pass it back to the caller for a
2978 * fixup.
2979 */
2980 *empty_extent_path = right_path;
2981 right_path = NULL;
2982 goto out;
2983 }
2984 if (ret) {
2985 mlog_errno(ret);
2986 goto out;
2987 }
2988
2989 /*
2990 * The subtree rotate might have removed records on
2991 * the rightmost edge. If so, then rotation is
2992 * complete.
2993 */
2994 if (deleted)
2995 break;
2996
2997 ocfs2_mv_path(left_path, right_path);
2998
Joel Beckere46f74d2009-02-12 19:47:43 -08002999 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003000 &right_cpos);
3001 if (ret) {
3002 mlog_errno(ret);
3003 goto out;
3004 }
3005 }
3006
3007out:
3008 ocfs2_free_path(right_path);
3009 ocfs2_free_path(left_path);
3010
3011 return ret;
3012}
3013
Joel Becker70f18c02009-02-13 02:09:31 -08003014static int ocfs2_remove_rightmost_path(handle_t *handle,
3015 struct ocfs2_extent_tree *et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003016 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003017 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003018{
3019 int ret, subtree_index;
3020 u32 cpos;
3021 struct ocfs2_path *left_path = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003022 struct ocfs2_extent_block *eb;
3023 struct ocfs2_extent_list *el;
3024
Joel Becker6136ca52009-02-12 19:32:43 -08003025 ret = ocfs2_et_sanity_check(et);
Tao Mae7d4cb62008-08-18 17:38:44 +08003026 if (ret)
3027 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003028
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003029 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003030 if (ret) {
3031 mlog_errno(ret);
3032 goto out;
3033 }
3034
Joel Becker3d03a302009-02-12 17:49:26 -08003035 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3036 path, &cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003037 if (ret) {
3038 mlog_errno(ret);
3039 goto out;
3040 }
3041
3042 if (cpos) {
3043 /*
3044 * We have a path to the left of this one - it needs
3045 * an update too.
3046 */
Joel Beckerffdd7a52008-10-17 22:32:01 -07003047 left_path = ocfs2_new_path_from_path(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003048 if (!left_path) {
3049 ret = -ENOMEM;
3050 mlog_errno(ret);
3051 goto out;
3052 }
3053
Joel Beckerfacdb772009-02-12 18:08:48 -08003054 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07003055 if (ret) {
3056 mlog_errno(ret);
3057 goto out;
3058 }
3059
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08003060 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003061 if (ret) {
3062 mlog_errno(ret);
3063 goto out;
3064 }
3065
Joel Becker7dc02802009-02-12 19:20:13 -08003066 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003067
Joel Becker6641b0c2009-02-12 18:57:52 -08003068 ocfs2_unlink_subtree(handle, et, left_path, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003069 subtree_index, dealloc);
Jun Piao964f14a2017-09-06 16:19:11 -07003070 ret = ocfs2_update_edge_lengths(handle, et, left_path);
Tao Ma3c5e1062009-07-21 15:42:05 +08003071 if (ret) {
3072 mlog_errno(ret);
3073 goto out;
3074 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003075
3076 eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
Joel Becker35dc0aa2008-08-20 16:25:06 -07003077 ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003078 } else {
3079 /*
3080 * 'path' is also the leftmost path which
3081 * means it must be the only one. This gets
3082 * handled differently because we want to
Joel Becker70f18c02009-02-13 02:09:31 -08003083 * revert the root back to having extents
Mark Fasheh328d5752007-06-18 10:48:04 -07003084 * in-line.
3085 */
Joel Becker6641b0c2009-02-12 18:57:52 -08003086 ocfs2_unlink_path(handle, et, dealloc, path, 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003087
Joel Beckerce1d9ea2008-08-20 16:30:07 -07003088 el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003089 el->l_tree_depth = 0;
3090 el->l_next_free_rec = 0;
3091 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3092
Joel Becker35dc0aa2008-08-20 16:25:06 -07003093 ocfs2_et_set_last_eb_blk(et, 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003094 }
3095
3096 ocfs2_journal_dirty(handle, path_root_bh(path));
3097
3098out:
3099 ocfs2_free_path(left_path);
3100 return ret;
3101}
3102
Xue jiufeid0c97d52015-09-04 15:44:48 -07003103static int ocfs2_remove_rightmost_empty_extent(struct ocfs2_super *osb,
3104 struct ocfs2_extent_tree *et,
3105 struct ocfs2_path *path,
3106 struct ocfs2_cached_dealloc_ctxt *dealloc)
3107{
3108 handle_t *handle;
3109 int ret;
3110 int credits = path->p_tree_depth * 2 + 1;
3111
3112 handle = ocfs2_start_trans(osb, credits);
3113 if (IS_ERR(handle)) {
3114 ret = PTR_ERR(handle);
3115 mlog_errno(ret);
3116 return ret;
3117 }
3118
3119 ret = ocfs2_remove_rightmost_path(handle, et, path, dealloc);
3120 if (ret)
3121 mlog_errno(ret);
3122
3123 ocfs2_commit_trans(osb, handle);
3124 return ret;
3125}
3126
Mark Fasheh328d5752007-06-18 10:48:04 -07003127/*
3128 * Left rotation of btree records.
3129 *
3130 * In many ways, this is (unsurprisingly) the opposite of right
3131 * rotation. We start at some non-rightmost path containing an empty
3132 * extent in the leaf block. The code works its way to the rightmost
3133 * path by rotating records to the left in every subtree.
3134 *
3135 * This is used by any code which reduces the number of extent records
3136 * in a leaf. After removal, an empty record should be placed in the
3137 * leftmost list position.
3138 *
3139 * This won't handle a length update of the rightmost path records if
3140 * the rightmost tree leaf record is removed so the caller is
3141 * responsible for detecting and correcting that.
3142 */
Joel Becker70f18c02009-02-13 02:09:31 -08003143static int ocfs2_rotate_tree_left(handle_t *handle,
3144 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003145 struct ocfs2_path *path,
Joel Becker70f18c02009-02-13 02:09:31 -08003146 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07003147{
Jan Kara9797a902019-11-05 17:44:18 +01003148 int ret, orig_credits = jbd2_handle_buffer_credits(handle);
Mark Fasheh328d5752007-06-18 10:48:04 -07003149 struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3150 struct ocfs2_extent_block *eb;
3151 struct ocfs2_extent_list *el;
3152
3153 el = path_leaf_el(path);
3154 if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3155 return 0;
3156
3157 if (path->p_tree_depth == 0) {
3158rightmost_no_delete:
3159 /*
Tao Mae7d4cb62008-08-18 17:38:44 +08003160 * Inline extents. This is trivially handled, so do
Mark Fasheh328d5752007-06-18 10:48:04 -07003161 * it up front.
3162 */
Joel Becker70f18c02009-02-13 02:09:31 -08003163 ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003164 if (ret)
3165 mlog_errno(ret);
3166 goto out;
3167 }
3168
3169 /*
3170 * Handle rightmost branch now. There's several cases:
3171 * 1) simple rotation leaving records in there. That's trivial.
3172 * 2) rotation requiring a branch delete - there's no more
3173 * records left. Two cases of this:
3174 * a) There are branches to the left.
3175 * b) This is also the leftmost (the only) branch.
3176 *
3177 * 1) is handled via ocfs2_rotate_rightmost_leaf_left()
3178 * 2a) we need the left branch so that we can update it with the unlink
Joel Becker70f18c02009-02-13 02:09:31 -08003179 * 2b) we need to bring the root back to inline extents.
Mark Fasheh328d5752007-06-18 10:48:04 -07003180 */
3181
3182 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3183 el = &eb->h_list;
3184 if (eb->h_next_leaf_blk == 0) {
3185 /*
3186 * This gets a bit tricky if we're going to delete the
3187 * rightmost path. Get the other cases out of the way
3188 * 1st.
3189 */
3190 if (le16_to_cpu(el->l_next_free_rec) > 1)
3191 goto rightmost_no_delete;
3192
3193 if (le16_to_cpu(el->l_next_free_rec) == 0) {
Jun Piao93f59202018-08-17 15:44:24 -07003194 ret = ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3195 "Owner %llu has empty extent block at %llu\n",
3196 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
3197 (unsigned long long)le64_to_cpu(eb->h_blkno));
Mark Fasheh328d5752007-06-18 10:48:04 -07003198 goto out;
3199 }
3200
3201 /*
3202 * XXX: The caller can not trust "path" any more after
3203 * this as it will have been deleted. What do we do?
3204 *
3205 * In theory the rotate-for-merge code will never get
3206 * here because it'll always ask for a rotate in a
3207 * nonempty list.
3208 */
3209
Joel Becker70f18c02009-02-13 02:09:31 -08003210 ret = ocfs2_remove_rightmost_path(handle, et, path,
3211 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003212 if (ret)
3213 mlog_errno(ret);
3214 goto out;
3215 }
3216
3217 /*
3218 * Now we can loop, remembering the path we get from -EAGAIN
3219 * and restarting from there.
3220 */
3221try_rotate:
Joel Beckere46f74d2009-02-12 19:47:43 -08003222 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3223 dealloc, &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003224 if (ret && ret != -EAGAIN) {
3225 mlog_errno(ret);
3226 goto out;
3227 }
3228
3229 while (ret == -EAGAIN) {
3230 tmp_path = restart_path;
3231 restart_path = NULL;
3232
Joel Beckere46f74d2009-02-12 19:47:43 -08003233 ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
Mark Fasheh328d5752007-06-18 10:48:04 -07003234 tmp_path, dealloc,
Joel Beckere46f74d2009-02-12 19:47:43 -08003235 &restart_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003236 if (ret && ret != -EAGAIN) {
3237 mlog_errno(ret);
3238 goto out;
3239 }
3240
3241 ocfs2_free_path(tmp_path);
3242 tmp_path = NULL;
3243
3244 if (ret == 0)
3245 goto try_rotate;
3246 }
3247
3248out:
3249 ocfs2_free_path(tmp_path);
3250 ocfs2_free_path(restart_path);
3251 return ret;
3252}
3253
3254static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3255 int index)
3256{
3257 struct ocfs2_extent_rec *rec = &el->l_recs[index];
3258 unsigned int size;
3259
3260 if (rec->e_leaf_clusters == 0) {
3261 /*
3262 * We consumed all of the merged-from record. An empty
3263 * extent cannot exist anywhere but the 1st array
3264 * position, so move things over if the merged-from
3265 * record doesn't occupy that position.
3266 *
3267 * This creates a new empty extent so the caller
3268 * should be smart enough to have removed any existing
3269 * ones.
3270 */
3271 if (index > 0) {
3272 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3273 size = index * sizeof(struct ocfs2_extent_rec);
3274 memmove(&el->l_recs[1], &el->l_recs[0], size);
3275 }
3276
3277 /*
3278 * Always memset - the caller doesn't check whether it
3279 * created an empty extent, so there could be junk in
3280 * the other fields.
3281 */
3282 memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3283 }
3284}
3285
Joel Becker4fe82c32009-02-13 02:16:08 -08003286static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003287 struct ocfs2_path *left_path,
3288 struct ocfs2_path **ret_right_path)
Mark Fasheh328d5752007-06-18 10:48:04 -07003289{
3290 int ret;
Tao Ma677b9752008-01-30 14:21:05 +08003291 u32 right_cpos;
3292 struct ocfs2_path *right_path = NULL;
3293 struct ocfs2_extent_list *left_el;
3294
3295 *ret_right_path = NULL;
3296
3297 /* This function shouldn't be called for non-trees. */
3298 BUG_ON(left_path->p_tree_depth == 0);
3299
3300 left_el = path_leaf_el(left_path);
3301 BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3302
Joel Becker4fe82c32009-02-13 02:16:08 -08003303 ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3304 left_path, &right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003305 if (ret) {
3306 mlog_errno(ret);
3307 goto out;
3308 }
3309
3310 /* This function shouldn't be called for the rightmost leaf. */
3311 BUG_ON(right_cpos == 0);
3312
Joel Beckerffdd7a52008-10-17 22:32:01 -07003313 right_path = ocfs2_new_path_from_path(left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003314 if (!right_path) {
3315 ret = -ENOMEM;
3316 mlog_errno(ret);
3317 goto out;
3318 }
3319
Joel Becker4fe82c32009-02-13 02:16:08 -08003320 ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003321 if (ret) {
3322 mlog_errno(ret);
3323 goto out;
3324 }
3325
3326 *ret_right_path = right_path;
3327out:
3328 if (ret)
3329 ocfs2_free_path(right_path);
3330 return ret;
3331}
3332
3333/*
3334 * Remove split_rec clusters from the record at index and merge them
3335 * onto the beginning of the record "next" to it.
3336 * For index < l_count - 1, the next means the extent rec at index + 1.
3337 * For index == l_count - 1, the "next" means the 1st extent rec of the
3338 * next extent block.
3339 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003340static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003341 handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003342 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003343 struct ocfs2_extent_rec *split_rec,
3344 int index)
3345{
3346 int ret, next_free, i;
Mark Fasheh328d5752007-06-18 10:48:04 -07003347 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3348 struct ocfs2_extent_rec *left_rec;
3349 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003350 struct ocfs2_extent_list *right_el;
3351 struct ocfs2_path *right_path = NULL;
3352 int subtree_index = 0;
3353 struct ocfs2_extent_list *el = path_leaf_el(left_path);
3354 struct buffer_head *bh = path_leaf_bh(left_path);
3355 struct buffer_head *root_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07003356
3357 BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
Mark Fasheh328d5752007-06-18 10:48:04 -07003358 left_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003359
Al Viro9d8df6a2008-05-21 06:32:11 +01003360 if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
Tao Ma677b9752008-01-30 14:21:05 +08003361 le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3362 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003363 ret = ocfs2_get_right_path(et, left_path, &right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003364 if (ret) {
3365 mlog_errno(ret);
Markus Elfring629a3b52015-04-14 15:42:53 -07003366 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003367 }
3368
3369 right_el = path_leaf_el(right_path);
3370 next_free = le16_to_cpu(right_el->l_next_free_rec);
3371 BUG_ON(next_free <= 0);
3372 right_rec = &right_el->l_recs[0];
3373 if (ocfs2_is_empty_extent(right_rec)) {
Al Viro9d8df6a2008-05-21 06:32:11 +01003374 BUG_ON(next_free <= 1);
Tao Ma677b9752008-01-30 14:21:05 +08003375 right_rec = &right_el->l_recs[1];
3376 }
3377
3378 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3379 le16_to_cpu(left_rec->e_leaf_clusters) !=
3380 le32_to_cpu(right_rec->e_cpos));
3381
Joel Becker7dc02802009-02-12 19:20:13 -08003382 subtree_index = ocfs2_find_subtree_root(et, left_path,
3383 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003384
3385 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
Jan Kara9797a902019-11-05 17:44:18 +01003386 jbd2_handle_buffer_credits(handle),
3387 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003388 if (ret) {
3389 mlog_errno(ret);
3390 goto out;
3391 }
3392
3393 root_bh = left_path->p_node[subtree_index].bh;
3394 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3395
Joel Becker7dc02802009-02-12 19:20:13 -08003396 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003397 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003398 if (ret) {
3399 mlog_errno(ret);
3400 goto out;
3401 }
3402
3403 for (i = subtree_index + 1;
3404 i < path_num_items(right_path); i++) {
Joel Becker7dc02802009-02-12 19:20:13 -08003405 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003406 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003407 if (ret) {
3408 mlog_errno(ret);
3409 goto out;
3410 }
3411
Joel Becker7dc02802009-02-12 19:20:13 -08003412 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003413 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003414 if (ret) {
3415 mlog_errno(ret);
3416 goto out;
3417 }
3418 }
3419
3420 } else {
3421 BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3422 right_rec = &el->l_recs[index + 1];
3423 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003424
Joel Becker7dc02802009-02-12 19:20:13 -08003425 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
Joel Becker13723d02008-10-17 19:25:01 -07003426 path_num_items(left_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003427 if (ret) {
3428 mlog_errno(ret);
3429 goto out;
3430 }
3431
3432 le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3433
3434 le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3435 le64_add_cpu(&right_rec->e_blkno,
Joel Becker7dc02802009-02-12 19:20:13 -08003436 -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3437 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003438 le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3439
3440 ocfs2_cleanup_merge(el, index);
3441
Joel Beckerec20cec2010-03-19 14:13:52 -07003442 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003443 if (right_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003444 ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
Joel Becker4619c732009-02-12 19:02:36 -08003445 ocfs2_complete_edge_insert(handle, left_path, right_path,
3446 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003447 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003448out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003449 ocfs2_free_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003450 return ret;
3451}
3452
Joel Becker4fe82c32009-02-13 02:16:08 -08003453static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003454 struct ocfs2_path *right_path,
3455 struct ocfs2_path **ret_left_path)
3456{
3457 int ret;
3458 u32 left_cpos;
3459 struct ocfs2_path *left_path = NULL;
3460
3461 *ret_left_path = NULL;
3462
3463 /* This function shouldn't be called for non-trees. */
3464 BUG_ON(right_path->p_tree_depth == 0);
3465
Joel Becker4fe82c32009-02-13 02:16:08 -08003466 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
Tao Ma677b9752008-01-30 14:21:05 +08003467 right_path, &left_cpos);
3468 if (ret) {
3469 mlog_errno(ret);
3470 goto out;
3471 }
3472
3473 /* This function shouldn't be called for the leftmost leaf. */
3474 BUG_ON(left_cpos == 0);
3475
Joel Beckerffdd7a52008-10-17 22:32:01 -07003476 left_path = ocfs2_new_path_from_path(right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003477 if (!left_path) {
3478 ret = -ENOMEM;
3479 mlog_errno(ret);
3480 goto out;
3481 }
3482
Joel Becker4fe82c32009-02-13 02:16:08 -08003483 ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
Tao Ma677b9752008-01-30 14:21:05 +08003484 if (ret) {
3485 mlog_errno(ret);
3486 goto out;
3487 }
3488
3489 *ret_left_path = left_path;
3490out:
3491 if (ret)
3492 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003493 return ret;
3494}
3495
3496/*
3497 * Remove split_rec clusters from the record at index and merge them
Tao Ma677b9752008-01-30 14:21:05 +08003498 * onto the tail of the record "before" it.
3499 * For index > 0, the "before" means the extent rec at index - 1.
3500 *
3501 * For index == 0, the "before" means the last record of the previous
3502 * extent block. And there is also a situation that we may need to
3503 * remove the rightmost leaf extent block in the right_path and change
3504 * the right path to indicate the new rightmost path.
Mark Fasheh328d5752007-06-18 10:48:04 -07003505 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003506static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003507 handle_t *handle,
Joel Becker4fe82c32009-02-13 02:16:08 -08003508 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003509 struct ocfs2_extent_rec *split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003510 struct ocfs2_cached_dealloc_ctxt *dealloc,
3511 int index)
Mark Fasheh328d5752007-06-18 10:48:04 -07003512{
Tao Ma677b9752008-01-30 14:21:05 +08003513 int ret, i, subtree_index = 0, has_empty_extent = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07003514 unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3515 struct ocfs2_extent_rec *left_rec;
3516 struct ocfs2_extent_rec *right_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003517 struct ocfs2_extent_list *el = path_leaf_el(right_path);
3518 struct buffer_head *bh = path_leaf_bh(right_path);
3519 struct buffer_head *root_bh = NULL;
3520 struct ocfs2_path *left_path = NULL;
3521 struct ocfs2_extent_list *left_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07003522
Tao Ma677b9752008-01-30 14:21:05 +08003523 BUG_ON(index < 0);
Mark Fasheh328d5752007-06-18 10:48:04 -07003524
Mark Fasheh328d5752007-06-18 10:48:04 -07003525 right_rec = &el->l_recs[index];
Tao Ma677b9752008-01-30 14:21:05 +08003526 if (index == 0) {
3527 /* we meet with a cross extent block merge. */
Joel Becker4fe82c32009-02-13 02:16:08 -08003528 ret = ocfs2_get_left_path(et, right_path, &left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003529 if (ret) {
3530 mlog_errno(ret);
Markus Elfring992ef6e2015-04-14 15:42:51 -07003531 return ret;
Tao Ma677b9752008-01-30 14:21:05 +08003532 }
3533
3534 left_el = path_leaf_el(left_path);
3535 BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3536 le16_to_cpu(left_el->l_count));
3537
3538 left_rec = &left_el->l_recs[
3539 le16_to_cpu(left_el->l_next_free_rec) - 1];
3540 BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3541 le16_to_cpu(left_rec->e_leaf_clusters) !=
3542 le32_to_cpu(split_rec->e_cpos));
3543
Joel Becker7dc02802009-02-12 19:20:13 -08003544 subtree_index = ocfs2_find_subtree_root(et, left_path,
3545 right_path);
Tao Ma677b9752008-01-30 14:21:05 +08003546
3547 ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
Jan Kara9797a902019-11-05 17:44:18 +01003548 jbd2_handle_buffer_credits(handle),
3549 left_path);
Tao Ma677b9752008-01-30 14:21:05 +08003550 if (ret) {
3551 mlog_errno(ret);
3552 goto out;
3553 }
3554
3555 root_bh = left_path->p_node[subtree_index].bh;
3556 BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3557
Joel Becker4fe82c32009-02-13 02:16:08 -08003558 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Joel Becker13723d02008-10-17 19:25:01 -07003559 subtree_index);
Tao Ma677b9752008-01-30 14:21:05 +08003560 if (ret) {
3561 mlog_errno(ret);
3562 goto out;
3563 }
3564
3565 for (i = subtree_index + 1;
3566 i < path_num_items(right_path); i++) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003567 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003568 right_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003569 if (ret) {
3570 mlog_errno(ret);
3571 goto out;
3572 }
3573
Joel Becker4fe82c32009-02-13 02:16:08 -08003574 ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
Joel Becker13723d02008-10-17 19:25:01 -07003575 left_path, i);
Tao Ma677b9752008-01-30 14:21:05 +08003576 if (ret) {
3577 mlog_errno(ret);
3578 goto out;
3579 }
3580 }
3581 } else {
3582 left_rec = &el->l_recs[index - 1];
3583 if (ocfs2_is_empty_extent(&el->l_recs[0]))
3584 has_empty_extent = 1;
3585 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003586
Joel Becker4fe82c32009-02-13 02:16:08 -08003587 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
Tao Ma9047bea2009-01-05 14:45:24 +08003588 path_num_items(right_path) - 1);
Mark Fasheh328d5752007-06-18 10:48:04 -07003589 if (ret) {
3590 mlog_errno(ret);
3591 goto out;
3592 }
3593
3594 if (has_empty_extent && index == 1) {
3595 /*
3596 * The easy case - we can just plop the record right in.
3597 */
3598 *left_rec = *split_rec;
Tao Ma677b9752008-01-30 14:21:05 +08003599 } else
Mark Fasheh328d5752007-06-18 10:48:04 -07003600 le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
Mark Fasheh328d5752007-06-18 10:48:04 -07003601
3602 le32_add_cpu(&right_rec->e_cpos, split_clusters);
3603 le64_add_cpu(&right_rec->e_blkno,
Joel Becker4fe82c32009-02-13 02:16:08 -08003604 ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3605 split_clusters));
Mark Fasheh328d5752007-06-18 10:48:04 -07003606 le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3607
3608 ocfs2_cleanup_merge(el, index);
3609
Joel Beckerec20cec2010-03-19 14:13:52 -07003610 ocfs2_journal_dirty(handle, bh);
Tao Ma677b9752008-01-30 14:21:05 +08003611 if (left_path) {
Joel Beckerec20cec2010-03-19 14:13:52 -07003612 ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
Tao Ma677b9752008-01-30 14:21:05 +08003613
3614 /*
3615 * In the situation that the right_rec is empty and the extent
3616 * block is empty also, ocfs2_complete_edge_insert can't handle
3617 * it and we need to delete the right extent block.
3618 */
3619 if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3620 le16_to_cpu(el->l_next_free_rec) == 1) {
Xue jiufei17215982016-03-25 14:21:41 -07003621 /* extend credit for ocfs2_remove_rightmost_path */
3622 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01003623 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07003624 right_path);
3625 if (ret) {
3626 mlog_errno(ret);
3627 goto out;
3628 }
Tao Ma677b9752008-01-30 14:21:05 +08003629
Joel Becker70f18c02009-02-13 02:09:31 -08003630 ret = ocfs2_remove_rightmost_path(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08003631 right_path,
Joel Becker70f18c02009-02-13 02:09:31 -08003632 dealloc);
Tao Ma677b9752008-01-30 14:21:05 +08003633 if (ret) {
3634 mlog_errno(ret);
3635 goto out;
3636 }
3637
3638 /* Now the rightmost extent block has been deleted.
3639 * So we use the new rightmost path.
3640 */
3641 ocfs2_mv_path(right_path, left_path);
3642 left_path = NULL;
3643 } else
Joel Becker4619c732009-02-12 19:02:36 -08003644 ocfs2_complete_edge_insert(handle, left_path,
Tao Ma677b9752008-01-30 14:21:05 +08003645 right_path, subtree_index);
3646 }
Mark Fasheh328d5752007-06-18 10:48:04 -07003647out:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07003648 ocfs2_free_path(left_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003649 return ret;
3650}
3651
Joel Beckerc495dd22009-02-13 02:19:11 -08003652static int ocfs2_try_to_merge_extent(handle_t *handle,
3653 struct ocfs2_extent_tree *et,
Tao Ma677b9752008-01-30 14:21:05 +08003654 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07003655 int split_index,
3656 struct ocfs2_extent_rec *split_rec,
3657 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Beckerc495dd22009-02-13 02:19:11 -08003658 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07003659{
Tao Mao518d7262007-08-28 17:25:35 -07003660 int ret = 0;
Tao Ma677b9752008-01-30 14:21:05 +08003661 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fasheh328d5752007-06-18 10:48:04 -07003662 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3663
3664 BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3665
Tao Mao518d7262007-08-28 17:25:35 -07003666 if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
Xue jiufei17215982016-03-25 14:21:41 -07003667 /* extend credit for ocfs2_remove_rightmost_path */
3668 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01003669 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07003670 path);
3671 if (ret) {
3672 mlog_errno(ret);
3673 goto out;
3674 }
Tao Mao518d7262007-08-28 17:25:35 -07003675 /*
3676 * The merge code will need to create an empty
3677 * extent to take the place of the newly
3678 * emptied slot. Remove any pre-existing empty
3679 * extents - having more than one in a leaf is
3680 * illegal.
3681 */
Joel Becker70f18c02009-02-13 02:09:31 -08003682 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Tao Mao518d7262007-08-28 17:25:35 -07003683 if (ret) {
3684 mlog_errno(ret);
3685 goto out;
Mark Fasheh328d5752007-06-18 10:48:04 -07003686 }
Tao Mao518d7262007-08-28 17:25:35 -07003687 split_index--;
3688 rec = &el->l_recs[split_index];
Mark Fasheh328d5752007-06-18 10:48:04 -07003689 }
3690
3691 if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3692 /*
3693 * Left-right contig implies this.
3694 */
3695 BUG_ON(!ctxt->c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07003696
3697 /*
3698 * Since the leftright insert always covers the entire
3699 * extent, this call will delete the insert record
3700 * entirely, resulting in an empty extent record added to
3701 * the extent block.
3702 *
3703 * Since the adding of an empty extent shifts
3704 * everything back to the right, there's no need to
3705 * update split_index here.
Tao Ma677b9752008-01-30 14:21:05 +08003706 *
3707 * When the split_index is zero, we need to merge it to the
3708 * prevoius extent block. It is more efficient and easier
3709 * if we do merge_right first and merge_left later.
Mark Fasheh328d5752007-06-18 10:48:04 -07003710 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003711 ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
Tao Ma677b9752008-01-30 14:21:05 +08003712 split_index);
Mark Fasheh328d5752007-06-18 10:48:04 -07003713 if (ret) {
3714 mlog_errno(ret);
3715 goto out;
3716 }
3717
3718 /*
3719 * We can only get this from logic error above.
3720 */
3721 BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3722
Xue jiufei17215982016-03-25 14:21:41 -07003723 /* extend credit for ocfs2_remove_rightmost_path */
3724 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01003725 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07003726 path);
3727 if (ret) {
3728 mlog_errno(ret);
3729 goto out;
3730 }
3731
Tao Ma677b9752008-01-30 14:21:05 +08003732 /* The merge left us with an empty extent, remove it. */
Joel Becker70f18c02009-02-13 02:09:31 -08003733 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003734 if (ret) {
3735 mlog_errno(ret);
3736 goto out;
3737 }
Tao Ma677b9752008-01-30 14:21:05 +08003738
Mark Fasheh328d5752007-06-18 10:48:04 -07003739 rec = &el->l_recs[split_index];
3740
3741 /*
3742 * Note that we don't pass split_rec here on purpose -
Tao Ma677b9752008-01-30 14:21:05 +08003743 * we've merged it into the rec already.
Mark Fasheh328d5752007-06-18 10:48:04 -07003744 */
Joel Becker4fe82c32009-02-13 02:16:08 -08003745 ret = ocfs2_merge_rec_left(path, handle, et, rec,
3746 dealloc, split_index);
Tao Ma677b9752008-01-30 14:21:05 +08003747
Mark Fasheh328d5752007-06-18 10:48:04 -07003748 if (ret) {
3749 mlog_errno(ret);
3750 goto out;
3751 }
3752
Xue jiufei17215982016-03-25 14:21:41 -07003753 /* extend credit for ocfs2_remove_rightmost_path */
3754 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01003755 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07003756 path);
3757 if (ret) {
3758 mlog_errno(ret);
3759 goto out;
3760 }
3761
Joel Becker70f18c02009-02-13 02:09:31 -08003762 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003763 /*
3764 * Error from this last rotate is not critical, so
3765 * print but don't bubble it up.
3766 */
3767 if (ret)
3768 mlog_errno(ret);
3769 ret = 0;
3770 } else {
3771 /*
3772 * Merge a record to the left or right.
3773 *
3774 * 'contig_type' is relative to the existing record,
3775 * so for example, if we're "right contig", it's to
3776 * the record on the left (hence the left merge).
3777 */
3778 if (ctxt->c_contig_type == CONTIG_RIGHT) {
Joel Becker4fe82c32009-02-13 02:16:08 -08003779 ret = ocfs2_merge_rec_left(path, handle, et,
3780 split_rec, dealloc,
Mark Fasheh328d5752007-06-18 10:48:04 -07003781 split_index);
3782 if (ret) {
3783 mlog_errno(ret);
3784 goto out;
3785 }
3786 } else {
Joel Becker4fe82c32009-02-13 02:16:08 -08003787 ret = ocfs2_merge_rec_right(path, handle,
Joel Becker7dc02802009-02-12 19:20:13 -08003788 et, split_rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003789 split_index);
3790 if (ret) {
3791 mlog_errno(ret);
3792 goto out;
3793 }
3794 }
3795
3796 if (ctxt->c_split_covers_rec) {
Xue jiufei17215982016-03-25 14:21:41 -07003797 /* extend credit for ocfs2_remove_rightmost_path */
3798 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01003799 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07003800 path);
3801 if (ret) {
3802 mlog_errno(ret);
3803 ret = 0;
3804 goto out;
3805 }
3806
Mark Fasheh328d5752007-06-18 10:48:04 -07003807 /*
3808 * The merge may have left an empty extent in
3809 * our leaf. Try to rotate it away.
3810 */
Joel Becker70f18c02009-02-13 02:09:31 -08003811 ret = ocfs2_rotate_tree_left(handle, et, path,
3812 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07003813 if (ret)
3814 mlog_errno(ret);
3815 ret = 0;
3816 }
3817 }
3818
3819out:
3820 return ret;
3821}
3822
3823static void ocfs2_subtract_from_rec(struct super_block *sb,
3824 enum ocfs2_split_type split,
3825 struct ocfs2_extent_rec *rec,
3826 struct ocfs2_extent_rec *split_rec)
3827{
3828 u64 len_blocks;
3829
3830 len_blocks = ocfs2_clusters_to_blocks(sb,
3831 le16_to_cpu(split_rec->e_leaf_clusters));
3832
3833 if (split == SPLIT_LEFT) {
3834 /*
3835 * Region is on the left edge of the existing
3836 * record.
3837 */
3838 le32_add_cpu(&rec->e_cpos,
3839 le16_to_cpu(split_rec->e_leaf_clusters));
3840 le64_add_cpu(&rec->e_blkno, len_blocks);
3841 le16_add_cpu(&rec->e_leaf_clusters,
3842 -le16_to_cpu(split_rec->e_leaf_clusters));
3843 } else {
3844 /*
3845 * Region is on the right edge of the existing
3846 * record.
3847 */
3848 le16_add_cpu(&rec->e_leaf_clusters,
3849 -le16_to_cpu(split_rec->e_leaf_clusters));
3850 }
3851}
3852
Mark Fashehdcd05382007-01-16 11:32:23 -08003853/*
3854 * Do the final bits of extent record insertion at the target leaf
3855 * list. If this leaf is part of an allocation tree, it is assumed
3856 * that the tree above has been prepared.
3857 */
Joel Beckerd5628622009-02-13 02:54:36 -08003858static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3859 struct ocfs2_extent_rec *insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08003860 struct ocfs2_extent_list *el,
Joel Beckerd5628622009-02-13 02:54:36 -08003861 struct ocfs2_insert_type *insert)
Mark Fashehdcd05382007-01-16 11:32:23 -08003862{
3863 int i = insert->ins_contig_index;
3864 unsigned int range;
3865 struct ocfs2_extent_rec *rec;
3866
Mark Fashehe48edee2007-03-07 16:46:57 -08003867 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08003868
Mark Fasheh328d5752007-06-18 10:48:04 -07003869 if (insert->ins_split != SPLIT_NONE) {
3870 i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3871 BUG_ON(i == -1);
3872 rec = &el->l_recs[i];
Joel Beckerd5628622009-02-13 02:54:36 -08003873 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3874 insert->ins_split, rec,
Mark Fasheh328d5752007-06-18 10:48:04 -07003875 insert_rec);
3876 goto rotate;
3877 }
3878
Mark Fashehdcd05382007-01-16 11:32:23 -08003879 /*
3880 * Contiguous insert - either left or right.
3881 */
3882 if (insert->ins_contig != CONTIG_NONE) {
3883 rec = &el->l_recs[i];
3884 if (insert->ins_contig == CONTIG_LEFT) {
3885 rec->e_blkno = insert_rec->e_blkno;
3886 rec->e_cpos = insert_rec->e_cpos;
3887 }
Mark Fashehe48edee2007-03-07 16:46:57 -08003888 le16_add_cpu(&rec->e_leaf_clusters,
3889 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003890 return;
3891 }
3892
3893 /*
3894 * Handle insert into an empty leaf.
3895 */
3896 if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3897 ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3898 ocfs2_is_empty_extent(&el->l_recs[0]))) {
3899 el->l_recs[0] = *insert_rec;
3900 el->l_next_free_rec = cpu_to_le16(1);
3901 return;
3902 }
3903
3904 /*
3905 * Appending insert.
3906 */
3907 if (insert->ins_appending == APPEND_TAIL) {
3908 i = le16_to_cpu(el->l_next_free_rec) - 1;
3909 rec = &el->l_recs[i];
Mark Fashehe48edee2007-03-07 16:46:57 -08003910 range = le32_to_cpu(rec->e_cpos)
3911 + le16_to_cpu(rec->e_leaf_clusters);
Mark Fashehdcd05382007-01-16 11:32:23 -08003912 BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3913
3914 mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3915 le16_to_cpu(el->l_count),
Joel Beckerd5628622009-02-13 02:54:36 -08003916 "owner %llu, depth %u, count %u, next free %u, "
Mark Fashehdcd05382007-01-16 11:32:23 -08003917 "rec.cpos %u, rec.clusters %u, "
3918 "insert.cpos %u, insert.clusters %u\n",
Joel Beckerd5628622009-02-13 02:54:36 -08003919 ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehdcd05382007-01-16 11:32:23 -08003920 le16_to_cpu(el->l_tree_depth),
3921 le16_to_cpu(el->l_count),
3922 le16_to_cpu(el->l_next_free_rec),
3923 le32_to_cpu(el->l_recs[i].e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003924 le16_to_cpu(el->l_recs[i].e_leaf_clusters),
Mark Fashehdcd05382007-01-16 11:32:23 -08003925 le32_to_cpu(insert_rec->e_cpos),
Mark Fashehe48edee2007-03-07 16:46:57 -08003926 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08003927 i++;
3928 el->l_recs[i] = *insert_rec;
3929 le16_add_cpu(&el->l_next_free_rec, 1);
3930 return;
3931 }
3932
Mark Fasheh328d5752007-06-18 10:48:04 -07003933rotate:
Mark Fashehdcd05382007-01-16 11:32:23 -08003934 /*
3935 * Ok, we have to rotate.
3936 *
3937 * At this point, it is safe to assume that inserting into an
3938 * empty leaf and appending to a leaf have both been handled
3939 * above.
3940 *
3941 * This leaf needs to have space, either by the empty 1st
wangyan41f4dc82020-04-01 21:03:38 -07003942 * extent record, or by virtue of an l_next_free_rec < l_count.
Mark Fashehdcd05382007-01-16 11:32:23 -08003943 */
3944 ocfs2_rotate_leaf(el, insert_rec);
3945}
3946
Joel Beckerd401dc12009-02-13 02:24:10 -08003947static void ocfs2_adjust_rightmost_records(handle_t *handle,
3948 struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07003949 struct ocfs2_path *path,
3950 struct ocfs2_extent_rec *insert_rec)
3951{
Changwei Ged22aa612018-01-31 16:14:55 -08003952 int i, next_free;
Mark Fasheh328d5752007-06-18 10:48:04 -07003953 struct buffer_head *bh;
3954 struct ocfs2_extent_list *el;
3955 struct ocfs2_extent_rec *rec;
3956
3957 /*
3958 * Update everything except the leaf block.
3959 */
3960 for (i = 0; i < path->p_tree_depth; i++) {
3961 bh = path->p_node[i].bh;
3962 el = path->p_node[i].el;
3963
3964 next_free = le16_to_cpu(el->l_next_free_rec);
3965 if (next_free == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08003966 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07003967 "Owner %llu has a bad extent list\n",
Joel Beckerd401dc12009-02-13 02:24:10 -08003968 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
Mark Fasheh328d5752007-06-18 10:48:04 -07003969 return;
3970 }
3971
3972 rec = &el->l_recs[next_free - 1];
3973
3974 rec->e_int_clusters = insert_rec->e_cpos;
3975 le32_add_cpu(&rec->e_int_clusters,
3976 le16_to_cpu(insert_rec->e_leaf_clusters));
3977 le32_add_cpu(&rec->e_int_clusters,
3978 -le32_to_cpu(rec->e_cpos));
3979
Joel Beckerec20cec2010-03-19 14:13:52 -07003980 ocfs2_journal_dirty(handle, bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07003981 }
3982}
3983
Joel Beckerd401dc12009-02-13 02:24:10 -08003984static int ocfs2_append_rec_to_path(handle_t *handle,
3985 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08003986 struct ocfs2_extent_rec *insert_rec,
3987 struct ocfs2_path *right_path,
3988 struct ocfs2_path **ret_left_path)
3989{
Mark Fasheh328d5752007-06-18 10:48:04 -07003990 int ret, next_free;
Mark Fashehdcd05382007-01-16 11:32:23 -08003991 struct ocfs2_extent_list *el;
3992 struct ocfs2_path *left_path = NULL;
3993
3994 *ret_left_path = NULL;
3995
3996 /*
Mark Fashehe48edee2007-03-07 16:46:57 -08003997 * This shouldn't happen for non-trees. The extent rec cluster
3998 * count manipulation below only works for interior nodes.
3999 */
4000 BUG_ON(right_path->p_tree_depth == 0);
4001
4002 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08004003 * If our appending insert is at the leftmost edge of a leaf,
4004 * then we might need to update the rightmost records of the
4005 * neighboring path.
4006 */
4007 el = path_leaf_el(right_path);
4008 next_free = le16_to_cpu(el->l_next_free_rec);
4009 if (next_free == 0 ||
4010 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
4011 u32 left_cpos;
4012
Joel Beckerd401dc12009-02-13 02:24:10 -08004013 ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
4014 right_path, &left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004015 if (ret) {
4016 mlog_errno(ret);
4017 goto out;
4018 }
4019
Tao Maa09d09b2011-02-22 08:21:10 +08004020 trace_ocfs2_append_rec_to_path(
4021 (unsigned long long)
4022 ocfs2_metadata_cache_owner(et->et_ci),
4023 le32_to_cpu(insert_rec->e_cpos),
4024 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004025
4026 /*
4027 * No need to worry if the append is already in the
4028 * leftmost leaf.
4029 */
4030 if (left_cpos) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004031 left_path = ocfs2_new_path_from_path(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004032 if (!left_path) {
4033 ret = -ENOMEM;
4034 mlog_errno(ret);
4035 goto out;
4036 }
4037
Joel Beckerd401dc12009-02-13 02:24:10 -08004038 ret = ocfs2_find_path(et->et_ci, left_path,
Joel Beckerfacdb772009-02-12 18:08:48 -08004039 left_cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004040 if (ret) {
4041 mlog_errno(ret);
4042 goto out;
4043 }
4044
4045 /*
4046 * ocfs2_insert_path() will pass the left_path to the
4047 * journal for us.
4048 */
4049 }
4050 }
4051
Joel Beckerd401dc12009-02-13 02:24:10 -08004052 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004053 if (ret) {
4054 mlog_errno(ret);
4055 goto out;
4056 }
4057
Joel Beckerd401dc12009-02-13 02:24:10 -08004058 ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004059
4060 *ret_left_path = left_path;
4061 ret = 0;
4062out:
4063 if (ret != 0)
4064 ocfs2_free_path(left_path);
4065
4066 return ret;
4067}
4068
Joel Beckerc38e52b2009-02-13 02:56:23 -08004069static void ocfs2_split_record(struct ocfs2_extent_tree *et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004070 struct ocfs2_path *left_path,
4071 struct ocfs2_path *right_path,
4072 struct ocfs2_extent_rec *split_rec,
4073 enum ocfs2_split_type split)
4074{
4075 int index;
4076 u32 cpos = le32_to_cpu(split_rec->e_cpos);
4077 struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4078 struct ocfs2_extent_rec *rec, *tmprec;
4079
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08004080 right_el = path_leaf_el(right_path);
Mark Fasheh328d5752007-06-18 10:48:04 -07004081 if (left_path)
4082 left_el = path_leaf_el(left_path);
4083
4084 el = right_el;
4085 insert_el = right_el;
4086 index = ocfs2_search_extent_list(el, cpos);
4087 if (index != -1) {
4088 if (index == 0 && left_path) {
4089 BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4090
4091 /*
4092 * This typically means that the record
4093 * started in the left path but moved to the
4094 * right as a result of rotation. We either
4095 * move the existing record to the left, or we
4096 * do the later insert there.
4097 *
4098 * In this case, the left path should always
4099 * exist as the rotate code will have passed
4100 * it back for a post-insert update.
4101 */
4102
4103 if (split == SPLIT_LEFT) {
4104 /*
4105 * It's a left split. Since we know
4106 * that the rotate code gave us an
4107 * empty extent in the left path, we
4108 * can just do the insert there.
4109 */
4110 insert_el = left_el;
4111 } else {
4112 /*
4113 * Right split - we have to move the
4114 * existing record over to the left
4115 * leaf. The insert will be into the
4116 * newly created empty extent in the
4117 * right leaf.
4118 */
4119 tmprec = &right_el->l_recs[index];
4120 ocfs2_rotate_leaf(left_el, tmprec);
4121 el = left_el;
4122
4123 memset(tmprec, 0, sizeof(*tmprec));
4124 index = ocfs2_search_extent_list(left_el, cpos);
4125 BUG_ON(index == -1);
4126 }
4127 }
4128 } else {
4129 BUG_ON(!left_path);
4130 BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4131 /*
4132 * Left path is easy - we can just allow the insert to
4133 * happen.
4134 */
4135 el = left_el;
4136 insert_el = left_el;
4137 index = ocfs2_search_extent_list(el, cpos);
4138 BUG_ON(index == -1);
4139 }
4140
4141 rec = &el->l_recs[index];
Joel Beckerc38e52b2009-02-13 02:56:23 -08004142 ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4143 split, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004144 ocfs2_rotate_leaf(insert_el, split_rec);
4145}
4146
Mark Fashehdcd05382007-01-16 11:32:23 -08004147/*
Tao Mae7d4cb62008-08-18 17:38:44 +08004148 * This function only does inserts on an allocation b-tree. For tree
4149 * depth = 0, ocfs2_insert_at_leaf() is called directly.
Mark Fashehdcd05382007-01-16 11:32:23 -08004150 *
4151 * right_path is the path we want to do the actual insert
4152 * in. left_path should only be passed in if we need to update that
4153 * portion of the tree after an edge insert.
4154 */
Joel Becker3505bec2009-02-13 02:57:58 -08004155static int ocfs2_insert_path(handle_t *handle,
Joel Becker7dc02802009-02-12 19:20:13 -08004156 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004157 struct ocfs2_path *left_path,
4158 struct ocfs2_path *right_path,
4159 struct ocfs2_extent_rec *insert_rec,
4160 struct ocfs2_insert_type *insert)
4161{
4162 int ret, subtree_index;
4163 struct buffer_head *leaf_bh = path_leaf_bh(right_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004164
Mark Fashehdcd05382007-01-16 11:32:23 -08004165 if (left_path) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004166 /*
4167 * There's a chance that left_path got passed back to
4168 * us without being accounted for in the
4169 * journal. Extend our transaction here to be sure we
4170 * can change those blocks.
4171 */
Tao Mac901fb02010-04-26 14:34:57 +08004172 ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08004173 if (ret < 0) {
4174 mlog_errno(ret);
4175 goto out;
4176 }
4177
Joel Becker7dc02802009-02-12 19:20:13 -08004178 ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
Mark Fashehdcd05382007-01-16 11:32:23 -08004179 if (ret < 0) {
4180 mlog_errno(ret);
4181 goto out;
4182 }
4183 }
4184
Mark Fashehe8aed342007-12-03 16:43:01 -08004185 /*
4186 * Pass both paths to the journal. The majority of inserts
4187 * will be touching all components anyway.
4188 */
Joel Becker7dc02802009-02-12 19:20:13 -08004189 ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
Mark Fashehe8aed342007-12-03 16:43:01 -08004190 if (ret < 0) {
4191 mlog_errno(ret);
4192 goto out;
4193 }
4194
Mark Fasheh328d5752007-06-18 10:48:04 -07004195 if (insert->ins_split != SPLIT_NONE) {
4196 /*
4197 * We could call ocfs2_insert_at_leaf() for some types
Joe Perchesc78bad12008-02-03 17:33:42 +02004198 * of splits, but it's easier to just let one separate
Mark Fasheh328d5752007-06-18 10:48:04 -07004199 * function sort it all out.
4200 */
Joel Beckerc38e52b2009-02-13 02:56:23 -08004201 ocfs2_split_record(et, left_path, right_path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004202 insert_rec, insert->ins_split);
Mark Fashehe8aed342007-12-03 16:43:01 -08004203
4204 /*
4205 * Split might have modified either leaf and we don't
4206 * have a guarantee that the later edge insert will
4207 * dirty this for us.
4208 */
4209 if (left_path)
Joel Beckerec20cec2010-03-19 14:13:52 -07004210 ocfs2_journal_dirty(handle,
4211 path_leaf_bh(left_path));
Mark Fasheh328d5752007-06-18 10:48:04 -07004212 } else
Joel Beckerd5628622009-02-13 02:54:36 -08004213 ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4214 insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004215
Joel Beckerec20cec2010-03-19 14:13:52 -07004216 ocfs2_journal_dirty(handle, leaf_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004217
4218 if (left_path) {
4219 /*
4220 * The rotate code has indicated that we need to fix
4221 * up portions of the tree after the insert.
4222 *
4223 * XXX: Should we extend the transaction here?
4224 */
Joel Becker7dc02802009-02-12 19:20:13 -08004225 subtree_index = ocfs2_find_subtree_root(et, left_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004226 right_path);
Joel Becker4619c732009-02-12 19:02:36 -08004227 ocfs2_complete_edge_insert(handle, left_path, right_path,
4228 subtree_index);
Mark Fashehdcd05382007-01-16 11:32:23 -08004229 }
4230
4231 ret = 0;
4232out:
4233 return ret;
4234}
4235
Joel Becker3505bec2009-02-13 02:57:58 -08004236static int ocfs2_do_insert_extent(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004237 struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004238 struct ocfs2_extent_rec *insert_rec,
4239 struct ocfs2_insert_type *type)
4240{
4241 int ret, rotate = 0;
4242 u32 cpos;
4243 struct ocfs2_path *right_path = NULL;
4244 struct ocfs2_path *left_path = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004245 struct ocfs2_extent_list *el;
4246
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004247 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004248
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004249 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004250 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehdcd05382007-01-16 11:32:23 -08004251 if (ret) {
4252 mlog_errno(ret);
4253 goto out;
4254 }
4255
4256 if (le16_to_cpu(el->l_tree_depth) == 0) {
Joel Beckerd5628622009-02-13 02:54:36 -08004257 ocfs2_insert_at_leaf(et, insert_rec, el, type);
Mark Fashehdcd05382007-01-16 11:32:23 -08004258 goto out_update_clusters;
4259 }
4260
Joel Beckerffdd7a52008-10-17 22:32:01 -07004261 right_path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004262 if (!right_path) {
4263 ret = -ENOMEM;
4264 mlog_errno(ret);
4265 goto out;
4266 }
4267
4268 /*
4269 * Determine the path to start with. Rotations need the
4270 * rightmost path, everything else can go directly to the
4271 * target leaf.
4272 */
4273 cpos = le32_to_cpu(insert_rec->e_cpos);
4274 if (type->ins_appending == APPEND_NONE &&
4275 type->ins_contig == CONTIG_NONE) {
4276 rotate = 1;
4277 cpos = UINT_MAX;
4278 }
4279
Joel Beckerfacdb772009-02-12 18:08:48 -08004280 ret = ocfs2_find_path(et->et_ci, right_path, cpos);
Mark Fashehdcd05382007-01-16 11:32:23 -08004281 if (ret) {
4282 mlog_errno(ret);
4283 goto out;
4284 }
4285
4286 /*
4287 * Rotations and appends need special treatment - they modify
4288 * parts of the tree's above them.
4289 *
4290 * Both might pass back a path immediate to the left of the
4291 * one being inserted to. This will be cause
4292 * ocfs2_insert_path() to modify the rightmost records of
4293 * left_path to account for an edge insert.
4294 *
4295 * XXX: When modifying this code, keep in mind that an insert
4296 * can wind up skipping both of these two special cases...
4297 */
4298 if (rotate) {
Joel Becker1bbf0b82009-02-12 19:42:08 -08004299 ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
Mark Fashehdcd05382007-01-16 11:32:23 -08004300 le32_to_cpu(insert_rec->e_cpos),
4301 right_path, &left_path);
4302 if (ret) {
4303 mlog_errno(ret);
4304 goto out;
4305 }
Mark Fashehe8aed342007-12-03 16:43:01 -08004306
4307 /*
4308 * ocfs2_rotate_tree_right() might have extended the
4309 * transaction without re-journaling our tree root.
4310 */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004311 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004312 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehe8aed342007-12-03 16:43:01 -08004313 if (ret) {
4314 mlog_errno(ret);
4315 goto out;
4316 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004317 } else if (type->ins_appending == APPEND_TAIL
4318 && type->ins_contig != CONTIG_LEFT) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004319 ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
Mark Fashehdcd05382007-01-16 11:32:23 -08004320 right_path, &left_path);
4321 if (ret) {
4322 mlog_errno(ret);
4323 goto out;
4324 }
4325 }
4326
Joel Becker3505bec2009-02-13 02:57:58 -08004327 ret = ocfs2_insert_path(handle, et, left_path, right_path,
Mark Fashehdcd05382007-01-16 11:32:23 -08004328 insert_rec, type);
4329 if (ret) {
4330 mlog_errno(ret);
4331 goto out;
4332 }
4333
4334out_update_clusters:
Mark Fasheh328d5752007-06-18 10:48:04 -07004335 if (type->ins_split == SPLIT_NONE)
Joel Becker6136ca52009-02-12 19:32:43 -08004336 ocfs2_et_update_clusters(et,
Joel Becker35dc0aa2008-08-20 16:25:06 -07004337 le16_to_cpu(insert_rec->e_leaf_clusters));
Mark Fashehdcd05382007-01-16 11:32:23 -08004338
Joel Beckerec20cec2010-03-19 14:13:52 -07004339 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004340
4341out:
4342 ocfs2_free_path(left_path);
4343 ocfs2_free_path(right_path);
4344
4345 return ret;
4346}
4347
Xue jiufei9f99ad02015-06-24 16:55:07 -07004348static int ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
Joel Beckera2970292009-02-13 03:09:54 -08004349 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004350 struct ocfs2_extent_list *el, int index,
Xue jiufei9f99ad02015-06-24 16:55:07 -07004351 struct ocfs2_extent_rec *split_rec,
4352 struct ocfs2_merge_ctxt *ctxt)
Mark Fasheh328d5752007-06-18 10:48:04 -07004353{
Xue jiufei9f99ad02015-06-24 16:55:07 -07004354 int status = 0;
Mark Fasheh328d5752007-06-18 10:48:04 -07004355 enum ocfs2_contig_type ret = CONTIG_NONE;
Tao Maad5a4d72008-01-30 14:21:32 +08004356 u32 left_cpos, right_cpos;
4357 struct ocfs2_extent_rec *rec = NULL;
4358 struct ocfs2_extent_list *new_el;
4359 struct ocfs2_path *left_path = NULL, *right_path = NULL;
4360 struct buffer_head *bh;
4361 struct ocfs2_extent_block *eb;
Joel Beckera2970292009-02-13 03:09:54 -08004362 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Tao Maad5a4d72008-01-30 14:21:32 +08004363
4364 if (index > 0) {
4365 rec = &el->l_recs[index - 1];
4366 } else if (path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004367 status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004368 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004369 goto exit;
Tao Maad5a4d72008-01-30 14:21:32 +08004370
4371 if (left_cpos != 0) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07004372 left_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004373 if (!left_path) {
4374 status = -ENOMEM;
4375 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004376 goto exit;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004377 }
Tao Maad5a4d72008-01-30 14:21:32 +08004378
Joel Beckera2970292009-02-13 03:09:54 -08004379 status = ocfs2_find_path(et->et_ci, left_path,
4380 left_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004381 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004382 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004383
4384 new_el = path_leaf_el(left_path);
4385
4386 if (le16_to_cpu(new_el->l_next_free_rec) !=
4387 le16_to_cpu(new_el->l_count)) {
4388 bh = path_leaf_bh(left_path);
4389 eb = (struct ocfs2_extent_block *)bh->b_data;
Jun Piao93f59202018-08-17 15:44:24 -07004390 status = ocfs2_error(sb,
4391 "Extent block #%llu has an invalid l_next_free_rec of %d. It should have matched the l_count of %d\n",
4392 (unsigned long long)le64_to_cpu(eb->h_blkno),
4393 le16_to_cpu(new_el->l_next_free_rec),
4394 le16_to_cpu(new_el->l_count));
Markus Elfring06a269cc2015-04-14 15:42:48 -07004395 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004396 }
4397 rec = &new_el->l_recs[
4398 le16_to_cpu(new_el->l_next_free_rec) - 1];
4399 }
4400 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004401
4402 /*
4403 * We're careful to check for an empty extent record here -
4404 * the merge code will know what to do if it sees one.
4405 */
Tao Maad5a4d72008-01-30 14:21:32 +08004406 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004407 if (index == 1 && ocfs2_is_empty_extent(rec)) {
4408 if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4409 ret = CONTIG_RIGHT;
4410 } else {
Tao Ma853a3a12009-08-18 11:22:18 +08004411 ret = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004412 }
4413 }
4414
Tao Maad5a4d72008-01-30 14:21:32 +08004415 rec = NULL;
4416 if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4417 rec = &el->l_recs[index + 1];
4418 else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4419 path->p_tree_depth > 0) {
Joel Beckera2970292009-02-13 03:09:54 -08004420 status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004421 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004422 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004423
4424 if (right_cpos == 0)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004425 goto free_left_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004426
Joel Beckerffdd7a52008-10-17 22:32:01 -07004427 right_path = ocfs2_new_path_from_path(path);
Xue jiufei9f99ad02015-06-24 16:55:07 -07004428 if (!right_path) {
4429 status = -ENOMEM;
4430 mlog_errno(status);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004431 goto free_left_path;
Xue jiufei9f99ad02015-06-24 16:55:07 -07004432 }
Tao Maad5a4d72008-01-30 14:21:32 +08004433
Joel Beckera2970292009-02-13 03:09:54 -08004434 status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
Tao Maad5a4d72008-01-30 14:21:32 +08004435 if (status)
Markus Elfring06a269cc2015-04-14 15:42:48 -07004436 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004437
4438 new_el = path_leaf_el(right_path);
4439 rec = &new_el->l_recs[0];
4440 if (ocfs2_is_empty_extent(rec)) {
4441 if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4442 bh = path_leaf_bh(right_path);
4443 eb = (struct ocfs2_extent_block *)bh->b_data;
Jun Piao93f59202018-08-17 15:44:24 -07004444 status = ocfs2_error(sb,
4445 "Extent block #%llu has an invalid l_next_free_rec of %d\n",
4446 (unsigned long long)le64_to_cpu(eb->h_blkno),
4447 le16_to_cpu(new_el->l_next_free_rec));
Markus Elfring06a269cc2015-04-14 15:42:48 -07004448 goto free_right_path;
Tao Maad5a4d72008-01-30 14:21:32 +08004449 }
4450 rec = &new_el->l_recs[1];
4451 }
4452 }
4453
4454 if (rec) {
Mark Fasheh328d5752007-06-18 10:48:04 -07004455 enum ocfs2_contig_type contig_type;
4456
Tao Ma853a3a12009-08-18 11:22:18 +08004457 contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004458
4459 if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4460 ret = CONTIG_LEFTRIGHT;
4461 else if (ret == CONTIG_NONE)
4462 ret = contig_type;
4463 }
4464
Markus Elfring06a269cc2015-04-14 15:42:48 -07004465free_right_path:
Markus Elfringfd90d4df2015-04-14 15:42:42 -07004466 ocfs2_free_path(right_path);
Markus Elfring06a269cc2015-04-14 15:42:48 -07004467free_left_path:
4468 ocfs2_free_path(left_path);
4469exit:
Xue jiufei9f99ad02015-06-24 16:55:07 -07004470 if (status == 0)
4471 ctxt->c_contig_type = ret;
4472
4473 return status;
Mark Fasheh328d5752007-06-18 10:48:04 -07004474}
4475
Joel Becker1ef61b32009-02-13 03:12:33 -08004476static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004477 struct ocfs2_insert_type *insert,
4478 struct ocfs2_extent_list *el,
Joel Becker1ef61b32009-02-13 03:12:33 -08004479 struct ocfs2_extent_rec *insert_rec)
Mark Fashehdcd05382007-01-16 11:32:23 -08004480{
4481 int i;
4482 enum ocfs2_contig_type contig_type = CONTIG_NONE;
4483
Mark Fashehe48edee2007-03-07 16:46:57 -08004484 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4485
Mark Fashehdcd05382007-01-16 11:32:23 -08004486 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
Tao Ma853a3a12009-08-18 11:22:18 +08004487 contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4488 insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004489 if (contig_type != CONTIG_NONE) {
4490 insert->ins_contig_index = i;
4491 break;
4492 }
4493 }
4494 insert->ins_contig = contig_type;
Tao Maca12b7c2008-08-18 17:38:52 +08004495
4496 if (insert->ins_contig != CONTIG_NONE) {
4497 struct ocfs2_extent_rec *rec =
4498 &el->l_recs[insert->ins_contig_index];
4499 unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4500 le16_to_cpu(insert_rec->e_leaf_clusters);
4501
4502 /*
4503 * Caller might want us to limit the size of extents, don't
4504 * calculate contiguousness if we might exceed that limit.
4505 */
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004506 if (et->et_max_leaf_clusters &&
4507 (len > et->et_max_leaf_clusters))
Tao Maca12b7c2008-08-18 17:38:52 +08004508 insert->ins_contig = CONTIG_NONE;
4509 }
Mark Fashehdcd05382007-01-16 11:32:23 -08004510}
4511
4512/*
4513 * This should only be called against the righmost leaf extent list.
4514 *
4515 * ocfs2_figure_appending_type() will figure out whether we'll have to
4516 * insert at the tail of the rightmost leaf.
4517 *
Tao Mae7d4cb62008-08-18 17:38:44 +08004518 * This should also work against the root extent list for tree's with 0
4519 * depth. If we consider the root extent list to be the rightmost leaf node
Mark Fashehdcd05382007-01-16 11:32:23 -08004520 * then the logic here makes sense.
4521 */
4522static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4523 struct ocfs2_extent_list *el,
4524 struct ocfs2_extent_rec *insert_rec)
4525{
4526 int i;
4527 u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4528 struct ocfs2_extent_rec *rec;
4529
4530 insert->ins_appending = APPEND_NONE;
4531
Mark Fashehe48edee2007-03-07 16:46:57 -08004532 BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
Mark Fashehdcd05382007-01-16 11:32:23 -08004533
4534 if (!el->l_next_free_rec)
4535 goto set_tail_append;
4536
4537 if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4538 /* Were all records empty? */
4539 if (le16_to_cpu(el->l_next_free_rec) == 1)
4540 goto set_tail_append;
4541 }
4542
4543 i = le16_to_cpu(el->l_next_free_rec) - 1;
4544 rec = &el->l_recs[i];
4545
Mark Fashehe48edee2007-03-07 16:46:57 -08004546 if (cpos >=
4547 (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
Mark Fashehdcd05382007-01-16 11:32:23 -08004548 goto set_tail_append;
4549
4550 return;
4551
4552set_tail_append:
4553 insert->ins_appending = APPEND_TAIL;
4554}
4555
4556/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004557 * Helper function called at the beginning of an insert.
Mark Fashehdcd05382007-01-16 11:32:23 -08004558 *
4559 * This computes a few things that are commonly used in the process of
4560 * inserting into the btree:
4561 * - Whether the new extent is contiguous with an existing one.
4562 * - The current tree depth.
4563 * - Whether the insert is an appending one.
4564 * - The total # of free records in the tree.
4565 *
4566 * All of the information is stored on the ocfs2_insert_type
4567 * structure.
4568 */
Joel Becker627961b2009-02-13 03:14:38 -08004569static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
Mark Fashehdcd05382007-01-16 11:32:23 -08004570 struct buffer_head **last_eb_bh,
4571 struct ocfs2_extent_rec *insert_rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004572 int *free_records,
Mark Fashehdcd05382007-01-16 11:32:23 -08004573 struct ocfs2_insert_type *insert)
4574{
4575 int ret;
Mark Fashehdcd05382007-01-16 11:32:23 -08004576 struct ocfs2_extent_block *eb;
4577 struct ocfs2_extent_list *el;
4578 struct ocfs2_path *path = NULL;
4579 struct buffer_head *bh = NULL;
4580
Mark Fasheh328d5752007-06-18 10:48:04 -07004581 insert->ins_split = SPLIT_NONE;
4582
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004583 el = et->et_root_el;
Mark Fashehdcd05382007-01-16 11:32:23 -08004584 insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4585
4586 if (el->l_tree_depth) {
4587 /*
4588 * If we have tree depth, we read in the
4589 * rightmost extent block ahead of time as
4590 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4591 * may want it later.
4592 */
Joel Becker3d03a302009-02-12 17:49:26 -08004593 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08004594 ocfs2_et_get_last_eb_blk(et),
4595 &bh);
Mark Fashehdcd05382007-01-16 11:32:23 -08004596 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08004597 mlog_errno(ret);
Mark Fashehdcd05382007-01-16 11:32:23 -08004598 goto out;
4599 }
4600 eb = (struct ocfs2_extent_block *) bh->b_data;
4601 el = &eb->h_list;
4602 }
4603
4604 /*
4605 * Unless we have a contiguous insert, we'll need to know if
4606 * there is room left in our allocation tree for another
4607 * extent record.
4608 *
4609 * XXX: This test is simplistic, we can search for empty
4610 * extent records too.
4611 */
Tao Maoc77534f2007-08-28 17:22:33 -07004612 *free_records = le16_to_cpu(el->l_count) -
Mark Fashehdcd05382007-01-16 11:32:23 -08004613 le16_to_cpu(el->l_next_free_rec);
4614
4615 if (!insert->ins_tree_depth) {
Joel Becker1ef61b32009-02-13 03:12:33 -08004616 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004617 ocfs2_figure_appending_type(insert, el, insert_rec);
4618 return 0;
4619 }
4620
Joel Beckerffdd7a52008-10-17 22:32:01 -07004621 path = ocfs2_new_path_from_et(et);
Mark Fashehdcd05382007-01-16 11:32:23 -08004622 if (!path) {
4623 ret = -ENOMEM;
4624 mlog_errno(ret);
4625 goto out;
4626 }
4627
4628 /*
4629 * In the case that we're inserting past what the tree
4630 * currently accounts for, ocfs2_find_path() will return for
4631 * us the rightmost tree path. This is accounted for below in
4632 * the appending code.
4633 */
Joel Beckerfacdb772009-02-12 18:08:48 -08004634 ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
Mark Fashehdcd05382007-01-16 11:32:23 -08004635 if (ret) {
4636 mlog_errno(ret);
4637 goto out;
4638 }
4639
4640 el = path_leaf_el(path);
4641
4642 /*
4643 * Now that we have the path, there's two things we want to determine:
4644 * 1) Contiguousness (also set contig_index if this is so)
4645 *
4646 * 2) Are we doing an append? We can trivially break this up
4647 * into two types of appends: simple record append, or a
4648 * rotate inside the tail leaf.
4649 */
Joel Becker1ef61b32009-02-13 03:12:33 -08004650 ocfs2_figure_contig_type(et, insert, el, insert_rec);
Mark Fashehdcd05382007-01-16 11:32:23 -08004651
4652 /*
4653 * The insert code isn't quite ready to deal with all cases of
4654 * left contiguousness. Specifically, if it's an insert into
4655 * the 1st record in a leaf, it will require the adjustment of
Mark Fashehe48edee2007-03-07 16:46:57 -08004656 * cluster count on the last record of the path directly to it's
Mark Fashehdcd05382007-01-16 11:32:23 -08004657 * left. For now, just catch that case and fool the layers
4658 * above us. This works just fine for tree_depth == 0, which
4659 * is why we allow that above.
4660 */
4661 if (insert->ins_contig == CONTIG_LEFT &&
4662 insert->ins_contig_index == 0)
4663 insert->ins_contig = CONTIG_NONE;
4664
4665 /*
4666 * Ok, so we can simply compare against last_eb to figure out
4667 * whether the path doesn't exist. This will only happen in
4668 * the case that we're doing a tail append, so maybe we can
4669 * take advantage of that information somehow.
4670 */
Joel Becker35dc0aa2008-08-20 16:25:06 -07004671 if (ocfs2_et_get_last_eb_blk(et) ==
Tao Mae7d4cb62008-08-18 17:38:44 +08004672 path_leaf_bh(path)->b_blocknr) {
Mark Fashehdcd05382007-01-16 11:32:23 -08004673 /*
4674 * Ok, ocfs2_find_path() returned us the rightmost
4675 * tree path. This might be an appending insert. There are
4676 * two cases:
4677 * 1) We're doing a true append at the tail:
4678 * -This might even be off the end of the leaf
4679 * 2) We're "appending" by rotating in the tail
4680 */
4681 ocfs2_figure_appending_type(insert, el, insert_rec);
4682 }
4683
4684out:
4685 ocfs2_free_path(path);
4686
4687 if (ret == 0)
4688 *last_eb_bh = bh;
4689 else
4690 brelse(bh);
4691 return ret;
4692}
4693
4694/*
Joel Beckercc79d8c2009-02-13 03:24:43 -08004695 * Insert an extent into a btree.
Mark Fashehdcd05382007-01-16 11:32:23 -08004696 *
Joel Beckercc79d8c2009-02-13 03:24:43 -08004697 * The caller needs to update the owning btree's cluster count.
Mark Fashehdcd05382007-01-16 11:32:23 -08004698 */
Joel Beckercc79d8c2009-02-13 03:24:43 -08004699int ocfs2_insert_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004700 struct ocfs2_extent_tree *et,
4701 u32 cpos,
4702 u64 start_blk,
4703 u32 new_clusters,
4704 u8 flags,
4705 struct ocfs2_alloc_context *meta_ac)
Mark Fashehccd979b2005-12-15 14:31:24 -08004706{
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004707 int status;
Kees Cook3f649ab2020-06-03 13:09:38 -07004708 int free_records;
Mark Fashehccd979b2005-12-15 14:31:24 -08004709 struct buffer_head *last_eb_bh = NULL;
Mark Fashehdcd05382007-01-16 11:32:23 -08004710 struct ocfs2_insert_type insert = {0, };
4711 struct ocfs2_extent_rec rec;
Mark Fashehccd979b2005-12-15 14:31:24 -08004712
Tao Maa09d09b2011-02-22 08:21:10 +08004713 trace_ocfs2_insert_extent_start(
4714 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4715 cpos, new_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08004716
Mark Fashehe48edee2007-03-07 16:46:57 -08004717 memset(&rec, 0, sizeof(rec));
Mark Fashehdcd05382007-01-16 11:32:23 -08004718 rec.e_cpos = cpu_to_le32(cpos);
4719 rec.e_blkno = cpu_to_le64(start_blk);
Mark Fashehe48edee2007-03-07 16:46:57 -08004720 rec.e_leaf_clusters = cpu_to_le16(new_clusters);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08004721 rec.e_flags = flags;
Joel Becker6136ca52009-02-12 19:32:43 -08004722 status = ocfs2_et_insert_check(et, &rec);
Joel Becker1e61ee72008-08-20 18:32:45 -07004723 if (status) {
4724 mlog_errno(status);
4725 goto bail;
4726 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004727
Joel Becker627961b2009-02-13 03:14:38 -08004728 status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
Tao Maoc77534f2007-08-28 17:22:33 -07004729 &free_records, &insert);
Mark Fashehdcd05382007-01-16 11:32:23 -08004730 if (status < 0) {
4731 mlog_errno(status);
4732 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08004733 }
4734
Tao Maa09d09b2011-02-22 08:21:10 +08004735 trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4736 insert.ins_contig_index, free_records,
4737 insert.ins_tree_depth);
Mark Fashehccd979b2005-12-15 14:31:24 -08004738
Tao Maoc77534f2007-08-28 17:22:33 -07004739 if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004740 status = ocfs2_grow_tree(handle, et,
Mark Fasheh328d5752007-06-18 10:48:04 -07004741 &insert.ins_tree_depth, &last_eb_bh,
Mark Fashehc3afcbb2007-05-29 14:28:51 -07004742 meta_ac);
4743 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08004744 mlog_errno(status);
4745 goto bail;
4746 }
Mark Fashehccd979b2005-12-15 14:31:24 -08004747 }
4748
Mark Fashehdcd05382007-01-16 11:32:23 -08004749 /* Finally, we can add clusters. This might rotate the tree for us. */
Joel Becker3505bec2009-02-13 02:57:58 -08004750 status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
Mark Fashehccd979b2005-12-15 14:31:24 -08004751 if (status < 0)
4752 mlog_errno(status);
Joel Becker92ba4702009-02-13 03:18:34 -08004753 else
4754 ocfs2_et_extent_map_insert(et, &rec);
Mark Fashehccd979b2005-12-15 14:31:24 -08004755
4756bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07004757 brelse(last_eb_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08004758
Tao Maf56654c2008-08-18 17:38:48 +08004759 return status;
4760}
4761
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004762/*
4763 * Allcate and add clusters into the extent b-tree.
4764 * The new clusters(clusters_to_add) will be inserted at logical_offset.
Joel Beckerf99b9b72008-08-20 19:36:33 -07004765 * The extent b-tree's root is specified by et, and
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004766 * it is not limited to the file storage. Any extent tree can use this
4767 * function if it implements the proper ocfs2_extent_tree.
4768 */
Joel Beckercbee7e12009-02-13 03:34:15 -08004769int ocfs2_add_clusters_in_btree(handle_t *handle,
4770 struct ocfs2_extent_tree *et,
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004771 u32 *logical_offset,
4772 u32 clusters_to_add,
4773 int mark_unwritten,
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004774 struct ocfs2_alloc_context *data_ac,
4775 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004776 enum ocfs2_alloc_restarted *reason_ret)
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004777{
Tao Maa09d09b2011-02-22 08:21:10 +08004778 int status = 0, err = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004779 int need_free = 0;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004780 int free_extents;
4781 enum ocfs2_alloc_restarted reason = RESTART_NONE;
4782 u32 bit_off, num_bits;
4783 u64 block;
4784 u8 flags = 0;
Joel Beckercbee7e12009-02-13 03:34:15 -08004785 struct ocfs2_super *osb =
4786 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004787
4788 BUG_ON(!clusters_to_add);
4789
4790 if (mark_unwritten)
4791 flags = OCFS2_EXT_UNWRITTEN;
4792
Jun Piao964f14a2017-09-06 16:19:11 -07004793 free_extents = ocfs2_num_free_extents(et);
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004794 if (free_extents < 0) {
4795 status = free_extents;
4796 mlog_errno(status);
4797 goto leave;
4798 }
4799
4800 /* there are two cases which could cause us to EAGAIN in the
4801 * we-need-more-metadata case:
4802 * 1) we haven't reserved *any*
4803 * 2) we are so fragmented, we've needed to add metadata too
4804 * many times. */
4805 if (!free_extents && !meta_ac) {
Tao Maa09d09b2011-02-22 08:21:10 +08004806 err = -1;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004807 status = -EAGAIN;
4808 reason = RESTART_META;
4809 goto leave;
4810 } else if ((!free_extents)
4811 && (ocfs2_alloc_context_bits_left(meta_ac)
Joel Beckerf99b9b72008-08-20 19:36:33 -07004812 < ocfs2_extend_meta_needed(et->et_root_el))) {
Tao Maa09d09b2011-02-22 08:21:10 +08004813 err = -2;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004814 status = -EAGAIN;
4815 reason = RESTART_META;
4816 goto leave;
4817 }
4818
Joel Becker1ed9b772010-05-06 13:59:06 +08004819 status = __ocfs2_claim_clusters(handle, data_ac, 1,
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004820 clusters_to_add, &bit_off, &num_bits);
4821 if (status < 0) {
4822 if (status != -ENOSPC)
4823 mlog_errno(status);
4824 goto leave;
4825 }
4826
4827 BUG_ON(num_bits > clusters_to_add);
4828
Joel Becker13723d02008-10-17 19:25:01 -07004829 /* reserve our write early -- insert_extent may update the tree root */
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08004830 status = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07004831 OCFS2_JOURNAL_ACCESS_WRITE);
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004832 if (status < 0) {
4833 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004834 need_free = 1;
4835 goto bail;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004836 }
4837
4838 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Tao Maa09d09b2011-02-22 08:21:10 +08004839 trace_ocfs2_add_clusters_in_btree(
4840 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4841 bit_off, num_bits);
Joel Beckercc79d8c2009-02-13 03:24:43 -08004842 status = ocfs2_insert_extent(handle, et, *logical_offset, block,
Joel Beckerf99b9b72008-08-20 19:36:33 -07004843 num_bits, flags, meta_ac);
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004844 if (status < 0) {
4845 mlog_errno(status);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004846 need_free = 1;
4847 goto bail;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004848 }
4849
Joel Beckerec20cec2010-03-19 14:13:52 -07004850 ocfs2_journal_dirty(handle, et->et_root_bh);
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004851
4852 clusters_to_add -= num_bits;
4853 *logical_offset += num_bits;
4854
4855 if (clusters_to_add) {
Tao Maa09d09b2011-02-22 08:21:10 +08004856 err = clusters_to_add;
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004857 status = -EAGAIN;
4858 reason = RESTART_TRANS;
4859 }
4860
Zongxun Wangfb951eb2014-02-06 12:04:20 -08004861bail:
4862 if (need_free) {
4863 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
4864 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
4865 bit_off, num_bits);
4866 else
4867 ocfs2_free_clusters(handle,
4868 data_ac->ac_inode,
4869 data_ac->ac_bh,
4870 ocfs2_clusters_to_blocks(osb->sb, bit_off),
4871 num_bits);
4872 }
4873
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004874leave:
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004875 if (reason_ret)
4876 *reason_ret = reason;
Tao Maa09d09b2011-02-22 08:21:10 +08004877 trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
Tao Ma0eb8d47e2008-08-18 17:38:45 +08004878 return status;
4879}
4880
Mark Fasheh328d5752007-06-18 10:48:04 -07004881static void ocfs2_make_right_split_rec(struct super_block *sb,
4882 struct ocfs2_extent_rec *split_rec,
4883 u32 cpos,
4884 struct ocfs2_extent_rec *rec)
4885{
4886 u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4887 u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4888
4889 memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4890
4891 split_rec->e_cpos = cpu_to_le32(cpos);
4892 split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4893
4894 split_rec->e_blkno = rec->e_blkno;
4895 le64_add_cpu(&split_rec->e_blkno,
4896 ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4897
4898 split_rec->e_flags = rec->e_flags;
4899}
4900
Joel Beckerd2311292009-02-13 03:43:22 -08004901static int ocfs2_split_and_insert(handle_t *handle,
Tao Mae7d4cb62008-08-18 17:38:44 +08004902 struct ocfs2_extent_tree *et,
Joel Beckerd2311292009-02-13 03:43:22 -08004903 struct ocfs2_path *path,
Mark Fasheh328d5752007-06-18 10:48:04 -07004904 struct buffer_head **last_eb_bh,
4905 int split_index,
4906 struct ocfs2_extent_rec *orig_split_rec,
4907 struct ocfs2_alloc_context *meta_ac)
4908{
4909 int ret = 0, depth;
4910 unsigned int insert_range, rec_range, do_leftright = 0;
4911 struct ocfs2_extent_rec tmprec;
4912 struct ocfs2_extent_list *rightmost_el;
4913 struct ocfs2_extent_rec rec;
4914 struct ocfs2_extent_rec split_rec = *orig_split_rec;
4915 struct ocfs2_insert_type insert;
4916 struct ocfs2_extent_block *eb;
Mark Fasheh328d5752007-06-18 10:48:04 -07004917
4918leftright:
4919 /*
4920 * Store a copy of the record on the stack - it might move
4921 * around as the tree is manipulated below.
4922 */
4923 rec = path_leaf_el(path)->l_recs[split_index];
4924
Joel Beckerce1d9ea2008-08-20 16:30:07 -07004925 rightmost_el = et->et_root_el;
Mark Fasheh328d5752007-06-18 10:48:04 -07004926
4927 depth = le16_to_cpu(rightmost_el->l_tree_depth);
4928 if (depth) {
4929 BUG_ON(!(*last_eb_bh));
4930 eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4931 rightmost_el = &eb->h_list;
4932 }
4933
4934 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4935 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08004936 ret = ocfs2_grow_tree(handle, et,
Tao Mae7d4cb62008-08-18 17:38:44 +08004937 &depth, last_eb_bh, meta_ac);
Mark Fasheh328d5752007-06-18 10:48:04 -07004938 if (ret) {
4939 mlog_errno(ret);
4940 goto out;
4941 }
Mark Fasheh328d5752007-06-18 10:48:04 -07004942 }
4943
4944 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4945 insert.ins_appending = APPEND_NONE;
4946 insert.ins_contig = CONTIG_NONE;
Mark Fasheh328d5752007-06-18 10:48:04 -07004947 insert.ins_tree_depth = depth;
4948
4949 insert_range = le32_to_cpu(split_rec.e_cpos) +
4950 le16_to_cpu(split_rec.e_leaf_clusters);
4951 rec_range = le32_to_cpu(rec.e_cpos) +
4952 le16_to_cpu(rec.e_leaf_clusters);
4953
4954 if (split_rec.e_cpos == rec.e_cpos) {
4955 insert.ins_split = SPLIT_LEFT;
4956 } else if (insert_range == rec_range) {
4957 insert.ins_split = SPLIT_RIGHT;
4958 } else {
4959 /*
4960 * Left/right split. We fake this as a right split
4961 * first and then make a second pass as a left split.
4962 */
4963 insert.ins_split = SPLIT_RIGHT;
4964
Joel Beckerd2311292009-02-13 03:43:22 -08004965 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4966 &tmprec, insert_range, &rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07004967
4968 split_rec = tmprec;
4969
4970 BUG_ON(do_leftright);
4971 do_leftright = 1;
4972 }
4973
Joel Becker3505bec2009-02-13 02:57:58 -08004974 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fasheh328d5752007-06-18 10:48:04 -07004975 if (ret) {
4976 mlog_errno(ret);
4977 goto out;
4978 }
4979
4980 if (do_leftright == 1) {
4981 u32 cpos;
4982 struct ocfs2_extent_list *el;
4983
4984 do_leftright++;
4985 split_rec = *orig_split_rec;
4986
4987 ocfs2_reinit_path(path, 1);
4988
4989 cpos = le32_to_cpu(split_rec.e_cpos);
Joel Beckerfacdb772009-02-12 18:08:48 -08004990 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07004991 if (ret) {
4992 mlog_errno(ret);
4993 goto out;
4994 }
4995
4996 el = path_leaf_el(path);
4997 split_index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07004998 if (split_index == -1) {
4999 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005000 "Owner %llu has an extent at cpos %u which can no longer be found\n",
5001 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5002 cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005003 ret = -EROFS;
5004 goto out;
5005 }
Mark Fasheh328d5752007-06-18 10:48:04 -07005006 goto leftright;
5007 }
5008out:
5009
5010 return ret;
5011}
5012
Joel Beckerf3868d02009-02-17 19:46:04 -08005013static int ocfs2_replace_extent_rec(handle_t *handle,
5014 struct ocfs2_extent_tree *et,
Tao Ma47be12e2009-01-09 07:32:48 +08005015 struct ocfs2_path *path,
5016 struct ocfs2_extent_list *el,
5017 int split_index,
5018 struct ocfs2_extent_rec *split_rec)
5019{
5020 int ret;
5021
Joel Beckerf3868d02009-02-17 19:46:04 -08005022 ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
Tao Ma47be12e2009-01-09 07:32:48 +08005023 path_num_items(path) - 1);
5024 if (ret) {
5025 mlog_errno(ret);
5026 goto out;
5027 }
5028
5029 el->l_recs[split_index] = *split_rec;
5030
5031 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5032out:
5033 return ret;
5034}
5035
Mark Fasheh328d5752007-06-18 10:48:04 -07005036/*
Tao Ma555936b2009-08-18 11:22:21 +08005037 * Split part or all of the extent record at split_index in the leaf
5038 * pointed to by path. Merge with the contiguous extent record if needed.
Mark Fasheh328d5752007-06-18 10:48:04 -07005039 *
5040 * Care is taken to handle contiguousness so as to not grow the tree.
5041 *
5042 * meta_ac is not strictly necessary - we only truly need it if growth
5043 * of the tree is required. All other cases will degrade into a less
5044 * optimal tree layout.
5045 *
Tao Mae7d4cb62008-08-18 17:38:44 +08005046 * last_eb_bh should be the rightmost leaf block for any extent
5047 * btree. Since a split may grow the tree or a merge might shrink it,
5048 * the caller cannot trust the contents of that buffer after this call.
Mark Fasheh328d5752007-06-18 10:48:04 -07005049 *
5050 * This code is optimized for readability - several passes might be
5051 * made over certain portions of the tree. All of those blocks will
5052 * have been brought into cache (and pinned via the journal), so the
5053 * extra overhead is not expressed in terms of disk reads.
5054 */
Tao Mae2e9f602009-08-18 11:22:34 +08005055int ocfs2_split_extent(handle_t *handle,
5056 struct ocfs2_extent_tree *et,
5057 struct ocfs2_path *path,
5058 int split_index,
5059 struct ocfs2_extent_rec *split_rec,
5060 struct ocfs2_alloc_context *meta_ac,
5061 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh328d5752007-06-18 10:48:04 -07005062{
5063 int ret = 0;
5064 struct ocfs2_extent_list *el = path_leaf_el(path);
Mark Fashehe8aed342007-12-03 16:43:01 -08005065 struct buffer_head *last_eb_bh = NULL;
Mark Fasheh328d5752007-06-18 10:48:04 -07005066 struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5067 struct ocfs2_merge_ctxt ctxt;
Mark Fasheh328d5752007-06-18 10:48:04 -07005068
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) {
Joel Becker3d03a302009-02-12 17:49:26 -08005092 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005093 ocfs2_et_get_last_eb_blk(et),
5094 &last_eb_bh);
Mark Fasheh328d5752007-06-18 10:48:04 -07005095 if (ret) {
Tao Mac1e8d352011-03-07 16:43:21 +08005096 mlog_errno(ret);
Mark Fasheh328d5752007-06-18 10:48:04 -07005097 goto out;
5098 }
Changwei Ged22aa612018-01-31 16:14:55 -08005099 }
Mark Fasheh328d5752007-06-18 10:48:04 -07005100
Mark Fasheh328d5752007-06-18 10:48:04 -07005101 if (rec->e_cpos == split_rec->e_cpos &&
5102 rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5103 ctxt.c_split_covers_rec = 1;
5104 else
5105 ctxt.c_split_covers_rec = 0;
5106
5107 ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5108
Tao Maa09d09b2011-02-22 08:21:10 +08005109 trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5110 ctxt.c_has_empty_extent,
5111 ctxt.c_split_covers_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005112
5113 if (ctxt.c_contig_type == CONTIG_NONE) {
5114 if (ctxt.c_split_covers_rec)
Joel Beckerf3868d02009-02-17 19:46:04 -08005115 ret = ocfs2_replace_extent_rec(handle, et, path, el,
Tao Ma47be12e2009-01-09 07:32:48 +08005116 split_index, split_rec);
Mark Fasheh328d5752007-06-18 10:48:04 -07005117 else
Joel Beckerd2311292009-02-13 03:43:22 -08005118 ret = ocfs2_split_and_insert(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005119 &last_eb_bh, split_index,
5120 split_rec, meta_ac);
5121 if (ret)
5122 mlog_errno(ret);
5123 } else {
Joel Beckerc495dd22009-02-13 02:19:11 -08005124 ret = ocfs2_try_to_merge_extent(handle, et, path,
Mark Fasheh328d5752007-06-18 10:48:04 -07005125 split_index, split_rec,
Joel Beckerc495dd22009-02-13 02:19:11 -08005126 dealloc, &ctxt);
Mark Fasheh328d5752007-06-18 10:48:04 -07005127 if (ret)
5128 mlog_errno(ret);
5129 }
5130
Mark Fasheh328d5752007-06-18 10:48:04 -07005131out:
5132 brelse(last_eb_bh);
5133 return ret;
5134}
5135
5136/*
Tao Ma555936b2009-08-18 11:22:21 +08005137 * Change the flags of the already-existing extent at cpos for len clusters.
5138 *
5139 * new_flags: the flags we want to set.
5140 * clear_flags: the flags we want to clear.
5141 * phys: the new physical offset we want this new extent starts from.
Mark Fasheh328d5752007-06-18 10:48:04 -07005142 *
5143 * If the existing extent is larger than the request, initiate a
5144 * split. An attempt will be made at merging with adjacent extents.
5145 *
5146 * The caller is responsible for passing down meta_ac if we'll need it.
5147 */
Tao Ma1aa75fe2009-08-18 11:28:39 +08005148int ocfs2_change_extent_flag(handle_t *handle,
5149 struct ocfs2_extent_tree *et,
5150 u32 cpos, u32 len, u32 phys,
5151 struct ocfs2_alloc_context *meta_ac,
5152 struct ocfs2_cached_dealloc_ctxt *dealloc,
5153 int new_flags, int clear_flags)
Mark Fasheh328d5752007-06-18 10:48:04 -07005154{
5155 int ret, index;
Tao Ma555936b2009-08-18 11:22:21 +08005156 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5157 u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
Mark Fasheh328d5752007-06-18 10:48:04 -07005158 struct ocfs2_extent_rec split_rec;
5159 struct ocfs2_path *left_path = NULL;
5160 struct ocfs2_extent_list *el;
Tao Ma555936b2009-08-18 11:22:21 +08005161 struct ocfs2_extent_rec *rec;
Mark Fasheh328d5752007-06-18 10:48:04 -07005162
Joel Beckerffdd7a52008-10-17 22:32:01 -07005163 left_path = ocfs2_new_path_from_et(et);
Mark Fasheh328d5752007-06-18 10:48:04 -07005164 if (!left_path) {
5165 ret = -ENOMEM;
5166 mlog_errno(ret);
5167 goto out;
5168 }
5169
Joel Beckerfacdb772009-02-12 18:08:48 -08005170 ret = ocfs2_find_path(et->et_ci, left_path, cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005171 if (ret) {
5172 mlog_errno(ret);
5173 goto out;
5174 }
5175 el = path_leaf_el(left_path);
5176
5177 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005178 if (index == -1) {
Tao Ma555936b2009-08-18 11:22:21 +08005179 ocfs2_error(sb,
Joe Perches7ecef142015-09-04 15:44:51 -07005180 "Owner %llu has an extent at cpos %u which can no longer be found\n",
5181 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5182 cpos);
Mark Fasheh328d5752007-06-18 10:48:04 -07005183 ret = -EROFS;
5184 goto out;
5185 }
5186
Tao Ma555936b2009-08-18 11:22:21 +08005187 ret = -EIO;
5188 rec = &el->l_recs[index];
5189 if (new_flags && (rec->e_flags & new_flags)) {
5190 mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
Darrick J. Wong86544fb2016-11-09 14:13:09 -08005191 "extent that already had them\n",
Tao Ma555936b2009-08-18 11:22:21 +08005192 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5193 new_flags);
5194 goto out;
5195 }
5196
5197 if (clear_flags && !(rec->e_flags & clear_flags)) {
5198 mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
Darrick J. Wong86544fb2016-11-09 14:13:09 -08005199 "extent that didn't have them\n",
Tao Ma555936b2009-08-18 11:22:21 +08005200 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5201 clear_flags);
5202 goto out;
5203 }
5204
Mark Fasheh328d5752007-06-18 10:48:04 -07005205 memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5206 split_rec.e_cpos = cpu_to_le32(cpos);
5207 split_rec.e_leaf_clusters = cpu_to_le16(len);
5208 split_rec.e_blkno = cpu_to_le64(start_blkno);
Tao Ma555936b2009-08-18 11:22:21 +08005209 split_rec.e_flags = rec->e_flags;
5210 if (new_flags)
5211 split_rec.e_flags |= new_flags;
5212 if (clear_flags)
5213 split_rec.e_flags &= ~clear_flags;
Mark Fasheh328d5752007-06-18 10:48:04 -07005214
Tao Mae2e9f602009-08-18 11:22:34 +08005215 ret = ocfs2_split_extent(handle, et, left_path,
5216 index, &split_rec, meta_ac,
5217 dealloc);
Mark Fasheh328d5752007-06-18 10:48:04 -07005218 if (ret)
5219 mlog_errno(ret);
5220
5221out:
5222 ocfs2_free_path(left_path);
5223 return ret;
Tao Ma555936b2009-08-18 11:22:21 +08005224
5225}
5226
5227/*
5228 * Mark the already-existing extent at cpos as written for len clusters.
5229 * This removes the unwritten extent flag.
5230 *
5231 * If the existing extent is larger than the request, initiate a
5232 * split. An attempt will be made at merging with adjacent extents.
5233 *
5234 * The caller is responsible for passing down meta_ac if we'll need it.
5235 */
5236int ocfs2_mark_extent_written(struct inode *inode,
5237 struct ocfs2_extent_tree *et,
5238 handle_t *handle, u32 cpos, u32 len, u32 phys,
5239 struct ocfs2_alloc_context *meta_ac,
5240 struct ocfs2_cached_dealloc_ctxt *dealloc)
5241{
5242 int ret;
5243
Tao Maa09d09b2011-02-22 08:21:10 +08005244 trace_ocfs2_mark_extent_written(
5245 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5246 cpos, len, phys);
Tao Ma555936b2009-08-18 11:22:21 +08005247
5248 if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
Joe Perches7ecef142015-09-04 15:44:51 -07005249 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 +08005250 (unsigned long long)OCFS2_I(inode)->ip_blkno);
5251 ret = -EROFS;
5252 goto out;
5253 }
5254
5255 /*
5256 * XXX: This should be fixed up so that we just re-insert the
5257 * next extent records.
5258 */
5259 ocfs2_et_extent_map_truncate(et, 0);
5260
5261 ret = ocfs2_change_extent_flag(handle, et, cpos,
5262 len, phys, meta_ac, dealloc,
5263 0, OCFS2_EXT_UNWRITTEN);
5264 if (ret)
5265 mlog_errno(ret);
5266
5267out:
5268 return ret;
Mark Fasheh328d5752007-06-18 10:48:04 -07005269}
5270
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005271static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5272 struct ocfs2_path *path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005273 int index, u32 new_range,
5274 struct ocfs2_alloc_context *meta_ac)
5275{
Tao Mac901fb02010-04-26 14:34:57 +08005276 int ret, depth, credits;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005277 struct buffer_head *last_eb_bh = NULL;
5278 struct ocfs2_extent_block *eb;
5279 struct ocfs2_extent_list *rightmost_el, *el;
5280 struct ocfs2_extent_rec split_rec;
5281 struct ocfs2_extent_rec *rec;
5282 struct ocfs2_insert_type insert;
5283
5284 /*
5285 * Setup the record to split before we grow the tree.
5286 */
5287 el = path_leaf_el(path);
5288 rec = &el->l_recs[index];
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005289 ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5290 &split_rec, new_range, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005291
5292 depth = path->p_tree_depth;
5293 if (depth > 0) {
Joel Becker3d03a302009-02-12 17:49:26 -08005294 ret = ocfs2_read_extent_block(et->et_ci,
Joel Becker5e965812008-11-13 14:49:16 -08005295 ocfs2_et_get_last_eb_blk(et),
5296 &last_eb_bh);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005297 if (ret < 0) {
5298 mlog_errno(ret);
5299 goto out;
5300 }
5301
5302 eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5303 rightmost_el = &eb->h_list;
5304 } else
5305 rightmost_el = path_leaf_el(path);
5306
Tao Mac901fb02010-04-26 14:34:57 +08005307 credits = path->p_tree_depth +
5308 ocfs2_extend_meta_needed(et->et_root_el);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005309 ret = ocfs2_extend_trans(handle, credits);
5310 if (ret) {
5311 mlog_errno(ret);
5312 goto out;
5313 }
5314
5315 if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5316 le16_to_cpu(rightmost_el->l_count)) {
Joel Beckerd401dc12009-02-13 02:24:10 -08005317 ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005318 meta_ac);
5319 if (ret) {
5320 mlog_errno(ret);
5321 goto out;
5322 }
Mark Fashehd0c7d702007-07-03 13:27:22 -07005323 }
5324
5325 memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5326 insert.ins_appending = APPEND_NONE;
5327 insert.ins_contig = CONTIG_NONE;
5328 insert.ins_split = SPLIT_RIGHT;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005329 insert.ins_tree_depth = depth;
5330
Joel Becker3505bec2009-02-13 02:57:58 -08005331 ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005332 if (ret)
5333 mlog_errno(ret);
5334
5335out:
5336 brelse(last_eb_bh);
5337 return ret;
5338}
5339
Joel Becker043beeb2009-02-13 02:42:30 -08005340static int ocfs2_truncate_rec(handle_t *handle,
5341 struct ocfs2_extent_tree *et,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005342 struct ocfs2_path *path, int index,
5343 struct ocfs2_cached_dealloc_ctxt *dealloc,
Joel Becker043beeb2009-02-13 02:42:30 -08005344 u32 cpos, u32 len)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005345{
5346 int ret;
5347 u32 left_cpos, rec_range, trunc_range;
piaojunc14688e2016-05-19 17:09:47 -07005348 int is_rightmost_tree_rec = 0;
Joel Becker043beeb2009-02-13 02:42:30 -08005349 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005350 struct ocfs2_path *left_path = NULL;
5351 struct ocfs2_extent_list *el = path_leaf_el(path);
5352 struct ocfs2_extent_rec *rec;
5353 struct ocfs2_extent_block *eb;
5354
5355 if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
Xue jiufei17215982016-03-25 14:21:41 -07005356 /* extend credit for ocfs2_remove_rightmost_path */
5357 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01005358 jbd2_handle_buffer_credits(handle),
Xue jiufei17215982016-03-25 14:21:41 -07005359 path);
5360 if (ret) {
5361 mlog_errno(ret);
5362 goto out;
5363 }
5364
Joel Becker70f18c02009-02-13 02:09:31 -08005365 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005366 if (ret) {
5367 mlog_errno(ret);
5368 goto out;
5369 }
5370
5371 index--;
5372 }
5373
5374 if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5375 path->p_tree_depth) {
5376 /*
5377 * Check whether this is the rightmost tree record. If
5378 * we remove all of this record or part of its right
5379 * edge then an update of the record lengths above it
5380 * will be required.
5381 */
5382 eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5383 if (eb->h_next_leaf_blk == 0)
5384 is_rightmost_tree_rec = 1;
5385 }
5386
5387 rec = &el->l_recs[index];
5388 if (index == 0 && path->p_tree_depth &&
5389 le32_to_cpu(rec->e_cpos) == cpos) {
5390 /*
5391 * Changing the leftmost offset (via partial or whole
5392 * record truncate) of an interior (or rightmost) path
5393 * means we have to update the subtree that is formed
5394 * by this leaf and the one to it's left.
5395 *
5396 * There are two cases we can skip:
Joel Becker043beeb2009-02-13 02:42:30 -08005397 * 1) Path is the leftmost one in our btree.
Mark Fashehd0c7d702007-07-03 13:27:22 -07005398 * 2) The leaf is rightmost and will be empty after
5399 * we remove the extent record - the rotate code
5400 * knows how to update the newly formed edge.
5401 */
5402
Joel Becker043beeb2009-02-13 02:42:30 -08005403 ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005404 if (ret) {
5405 mlog_errno(ret);
5406 goto out;
5407 }
5408
5409 if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
Joel Beckerffdd7a52008-10-17 22:32:01 -07005410 left_path = ocfs2_new_path_from_path(path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005411 if (!left_path) {
5412 ret = -ENOMEM;
5413 mlog_errno(ret);
5414 goto out;
5415 }
5416
Joel Beckerfacdb772009-02-12 18:08:48 -08005417 ret = ocfs2_find_path(et->et_ci, left_path,
5418 left_cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005419 if (ret) {
5420 mlog_errno(ret);
5421 goto out;
5422 }
5423 }
5424 }
5425
5426 ret = ocfs2_extend_rotate_transaction(handle, 0,
Jan Kara9797a902019-11-05 17:44:18 +01005427 jbd2_handle_buffer_credits(handle),
5428 path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005429 if (ret) {
5430 mlog_errno(ret);
5431 goto out;
5432 }
5433
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005434 ret = ocfs2_journal_access_path(et->et_ci, handle, path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005435 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, left_path);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005441 if (ret) {
5442 mlog_errno(ret);
5443 goto out;
5444 }
5445
5446 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5447 trunc_range = cpos + len;
5448
5449 if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5450 int next_free;
5451
5452 memset(rec, 0, sizeof(*rec));
5453 ocfs2_cleanup_merge(el, index);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005454
5455 next_free = le16_to_cpu(el->l_next_free_rec);
5456 if (is_rightmost_tree_rec && next_free > 1) {
5457 /*
5458 * We skip the edge update if this path will
5459 * be deleted by the rotate code.
5460 */
5461 rec = &el->l_recs[next_free - 1];
Joel Beckerd401dc12009-02-13 02:24:10 -08005462 ocfs2_adjust_rightmost_records(handle, et, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005463 rec);
5464 }
5465 } else if (le32_to_cpu(rec->e_cpos) == cpos) {
5466 /* Remove leftmost portion of the record. */
5467 le32_add_cpu(&rec->e_cpos, len);
5468 le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5469 le16_add_cpu(&rec->e_leaf_clusters, -len);
5470 } else if (rec_range == trunc_range) {
5471 /* Remove rightmost portion of the record */
5472 le16_add_cpu(&rec->e_leaf_clusters, -len);
5473 if (is_rightmost_tree_rec)
Joel Beckerd401dc12009-02-13 02:24:10 -08005474 ocfs2_adjust_rightmost_records(handle, et, path, rec);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005475 } else {
5476 /* Caller should have trapped this. */
Joel Becker043beeb2009-02-13 02:42:30 -08005477 mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5478 "(%u, %u)\n",
5479 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005480 le32_to_cpu(rec->e_cpos),
5481 le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5482 BUG();
5483 }
5484
5485 if (left_path) {
5486 int subtree_index;
5487
Joel Becker7dc02802009-02-12 19:20:13 -08005488 subtree_index = ocfs2_find_subtree_root(et, left_path, path);
Joel Becker4619c732009-02-12 19:02:36 -08005489 ocfs2_complete_edge_insert(handle, left_path, path,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005490 subtree_index);
5491 }
5492
5493 ocfs2_journal_dirty(handle, path_leaf_bh(path));
5494
Joel Becker70f18c02009-02-13 02:09:31 -08005495 ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
wangyan229ba1f2018-08-17 15:44:27 -07005496 if (ret)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005497 mlog_errno(ret);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005498
5499out:
5500 ocfs2_free_path(left_path);
5501 return ret;
5502}
5503
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005504int ocfs2_remove_extent(handle_t *handle,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005505 struct ocfs2_extent_tree *et,
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005506 u32 cpos, u32 len,
Mark Fasheh063c4562007-07-03 13:34:11 -07005507 struct ocfs2_alloc_context *meta_ac,
Joel Beckerf99b9b72008-08-20 19:36:33 -07005508 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005509{
5510 int ret, index;
5511 u32 rec_range, trunc_range;
5512 struct ocfs2_extent_rec *rec;
5513 struct ocfs2_extent_list *el;
Tao Mae7d4cb62008-08-18 17:38:44 +08005514 struct ocfs2_path *path = NULL;
Mark Fashehd0c7d702007-07-03 13:27:22 -07005515
Joel Becker4c911ee2009-02-13 02:50:12 -08005516 /*
5517 * XXX: Why are we truncating to 0 instead of wherever this
5518 * affects us?
5519 */
5520 ocfs2_et_extent_map_truncate(et, 0);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005521
Joel Beckerffdd7a52008-10-17 22:32:01 -07005522 path = ocfs2_new_path_from_et(et);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005523 if (!path) {
5524 ret = -ENOMEM;
5525 mlog_errno(ret);
5526 goto out;
5527 }
5528
Joel Beckerfacdb772009-02-12 18:08:48 -08005529 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005530 if (ret) {
5531 mlog_errno(ret);
5532 goto out;
5533 }
5534
5535 el = path_leaf_el(path);
5536 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005537 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005538 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005539 "Owner %llu has an extent at cpos %u which can no longer be found\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005540 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5541 cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005542 ret = -EROFS;
5543 goto out;
5544 }
5545
5546 /*
5547 * We have 3 cases of extent removal:
5548 * 1) Range covers the entire extent rec
5549 * 2) Range begins or ends on one edge of the extent rec
5550 * 3) Range is in the middle of the extent rec (no shared edges)
5551 *
5552 * For case 1 we remove the extent rec and left rotate to
5553 * fill the hole.
5554 *
5555 * For case 2 we just shrink the existing extent rec, with a
5556 * tree update if the shrinking edge is also the edge of an
5557 * extent block.
5558 *
5559 * For case 3 we do a right split to turn the extent rec into
5560 * something case 2 can handle.
5561 */
5562 rec = &el->l_recs[index];
5563 rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5564 trunc_range = cpos + len;
5565
5566 BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5567
Tao Maa09d09b2011-02-22 08:21:10 +08005568 trace_ocfs2_remove_extent(
5569 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5570 cpos, len, index, le32_to_cpu(rec->e_cpos),
5571 ocfs2_rec_clusters(el, rec));
Mark Fashehd0c7d702007-07-03 13:27:22 -07005572
5573 if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
Joel Becker043beeb2009-02-13 02:42:30 -08005574 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5575 cpos, len);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005576 if (ret) {
5577 mlog_errno(ret);
5578 goto out;
5579 }
5580 } else {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005581 ret = ocfs2_split_tree(handle, et, path, index,
Mark Fashehd0c7d702007-07-03 13:27:22 -07005582 trunc_range, meta_ac);
5583 if (ret) {
5584 mlog_errno(ret);
5585 goto out;
5586 }
5587
5588 /*
5589 * The split could have manipulated the tree enough to
5590 * move the record location, so we have to look for it again.
5591 */
5592 ocfs2_reinit_path(path, 1);
5593
Joel Beckerfacdb772009-02-12 18:08:48 -08005594 ret = ocfs2_find_path(et->et_ci, path, cpos);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005595 if (ret) {
5596 mlog_errno(ret);
5597 goto out;
5598 }
5599
5600 el = path_leaf_el(path);
5601 index = ocfs2_search_extent_list(el, cpos);
Yingtai Xie981035b2014-08-06 16:03:54 -07005602 if (index == -1) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005603 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005604 "Owner %llu: split at cpos %u lost record\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005605 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005606 cpos);
5607 ret = -EROFS;
5608 goto out;
5609 }
5610
5611 /*
5612 * Double check our values here. If anything is fishy,
5613 * it's easier to catch it at the top level.
5614 */
5615 rec = &el->l_recs[index];
5616 rec_range = le32_to_cpu(rec->e_cpos) +
5617 ocfs2_rec_clusters(el, rec);
5618 if (rec_range != trunc_range) {
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005619 ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
Joe Perches7ecef142015-09-04 15:44:51 -07005620 "Owner %llu: error after split at cpos %u trunc len %u, existing record is (%u,%u)\n",
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005621 (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
Mark Fashehd0c7d702007-07-03 13:27:22 -07005622 cpos, len, le32_to_cpu(rec->e_cpos),
5623 ocfs2_rec_clusters(el, rec));
5624 ret = -EROFS;
5625 goto out;
5626 }
5627
Joel Becker043beeb2009-02-13 02:42:30 -08005628 ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5629 cpos, len);
wangyan229ba1f2018-08-17 15:44:27 -07005630 if (ret)
Mark Fashehd0c7d702007-07-03 13:27:22 -07005631 mlog_errno(ret);
Mark Fashehd0c7d702007-07-03 13:27:22 -07005632 }
5633
5634out:
5635 ocfs2_free_path(path);
5636 return ret;
5637}
5638
Tristan Ye78f94672010-05-11 17:54:42 +08005639/*
5640 * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5641 * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5642 * number to reserve some extra blocks, and it only handles meta
5643 * data allocations.
5644 *
5645 * Currently, only ocfs2_remove_btree_range() uses it for truncating
5646 * and punching holes.
5647 */
5648static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5649 struct ocfs2_extent_tree *et,
5650 u32 extents_to_split,
5651 struct ocfs2_alloc_context **ac,
5652 int extra_blocks)
5653{
5654 int ret = 0, num_free_extents;
5655 unsigned int max_recs_needed = 2 * extents_to_split;
5656 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5657
5658 *ac = NULL;
5659
Jun Piao964f14a2017-09-06 16:19:11 -07005660 num_free_extents = ocfs2_num_free_extents(et);
Tristan Ye78f94672010-05-11 17:54:42 +08005661 if (num_free_extents < 0) {
5662 ret = num_free_extents;
5663 mlog_errno(ret);
5664 goto out;
5665 }
5666
5667 if (!num_free_extents ||
5668 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5669 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5670
5671 if (extra_blocks) {
5672 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5673 if (ret < 0) {
5674 if (ret != -ENOSPC)
5675 mlog_errno(ret);
Tristan Ye78f94672010-05-11 17:54:42 +08005676 }
5677 }
5678
5679out:
5680 if (ret) {
5681 if (*ac) {
5682 ocfs2_free_alloc_context(*ac);
5683 *ac = NULL;
5684 }
5685 }
5686
5687 return ret;
5688}
5689
Mark Fashehfecc0112008-11-12 15:16:38 -08005690int ocfs2_remove_btree_range(struct inode *inode,
5691 struct ocfs2_extent_tree *et,
Tristan Ye78f94672010-05-11 17:54:42 +08005692 u32 cpos, u32 phys_cpos, u32 len, int flags,
5693 struct ocfs2_cached_dealloc_ctxt *dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08005694 u64 refcount_loc, bool refcount_tree_locked)
Mark Fashehfecc0112008-11-12 15:16:38 -08005695{
Tristan Ye78f94672010-05-11 17:54:42 +08005696 int ret, credits = 0, extra_blocks = 0;
Mark Fashehfecc0112008-11-12 15:16:38 -08005697 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5698 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5699 struct inode *tl_inode = osb->osb_tl_inode;
5700 handle_t *handle;
5701 struct ocfs2_alloc_context *meta_ac = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08005702 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehfecc0112008-11-12 15:16:38 -08005703
Tristan Ye78f94672010-05-11 17:54:42 +08005704 if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
Darrick J. Wong84e40082016-11-09 14:13:09 -08005705 BUG_ON(!ocfs2_is_refcount_inode(inode));
Tristan Ye78f94672010-05-11 17:54:42 +08005706
Junxiao Bif62f12b2014-12-18 16:17:32 -08005707 if (!refcount_tree_locked) {
5708 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5709 &ref_tree, NULL);
5710 if (ret) {
5711 mlog_errno(ret);
5712 goto bail;
5713 }
Tristan Ye78f94672010-05-11 17:54:42 +08005714 }
5715
5716 ret = ocfs2_prepare_refcount_change_for_del(inode,
5717 refcount_loc,
5718 phys_blkno,
5719 len,
5720 &credits,
5721 &extra_blocks);
5722 if (ret < 0) {
5723 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005724 goto bail;
Tristan Ye78f94672010-05-11 17:54:42 +08005725 }
5726 }
5727
5728 ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5729 extra_blocks);
Mark Fashehfecc0112008-11-12 15:16:38 -08005730 if (ret) {
5731 mlog_errno(ret);
Joseph Qi33add0e2013-07-03 15:00:52 -07005732 goto bail;
Mark Fashehfecc0112008-11-12 15:16:38 -08005733 }
5734
Al Viro59551022016-01-22 15:40:57 -05005735 inode_lock(tl_inode);
Mark Fashehfecc0112008-11-12 15:16:38 -08005736
5737 if (ocfs2_truncate_log_needs_flush(osb)) {
5738 ret = __ocfs2_flush_truncate_log(osb);
5739 if (ret < 0) {
5740 mlog_errno(ret);
5741 goto out;
5742 }
5743 }
5744
Tristan Ye78f94672010-05-11 17:54:42 +08005745 handle = ocfs2_start_trans(osb,
5746 ocfs2_remove_extent_credits(osb->sb) + credits);
Mark Fashehfecc0112008-11-12 15:16:38 -08005747 if (IS_ERR(handle)) {
5748 ret = PTR_ERR(handle);
5749 mlog_errno(ret);
5750 goto out;
5751 }
5752
Joel Beckerd9a0a1f2009-02-12 17:32:34 -08005753 ret = ocfs2_et_root_journal_access(handle, et,
Joel Becker13723d02008-10-17 19:25:01 -07005754 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehfecc0112008-11-12 15:16:38 -08005755 if (ret) {
5756 mlog_errno(ret);
Wengang Wangb8a0ae52011-10-12 15:22:15 +08005757 goto out_commit;
Mark Fashehfecc0112008-11-12 15:16:38 -08005758 }
5759
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005760 dquot_free_space_nodirty(inode,
Mark Fashehfd4ef232009-01-29 15:06:21 -08005761 ocfs2_clusters_to_bytes(inode->i_sb, len));
5762
Joel Beckerdbdcf6a2009-02-13 03:41:26 -08005763 ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
Mark Fashehfecc0112008-11-12 15:16:38 -08005764 if (ret) {
5765 mlog_errno(ret);
5766 goto out_commit;
5767 }
5768
Joel Becker6136ca52009-02-12 19:32:43 -08005769 ocfs2_et_update_clusters(et, -len);
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07005770 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fashehfecc0112008-11-12 15:16:38 -08005771
Joel Beckerec20cec2010-03-19 14:13:52 -07005772 ocfs2_journal_dirty(handle, et->et_root_bh);
Mark Fashehfecc0112008-11-12 15:16:38 -08005773
Tristan Ye78f94672010-05-11 17:54:42 +08005774 if (phys_blkno) {
5775 if (flags & OCFS2_EXT_REFCOUNTED)
5776 ret = ocfs2_decrease_refcount(inode, handle,
5777 ocfs2_blocks_to_clusters(osb->sb,
5778 phys_blkno),
5779 len, meta_ac,
5780 dealloc, 1);
5781 else
5782 ret = ocfs2_truncate_log_append(osb, handle,
5783 phys_blkno, len);
5784 if (ret)
5785 mlog_errno(ret);
5786
5787 }
Mark Fashehfecc0112008-11-12 15:16:38 -08005788
5789out_commit:
5790 ocfs2_commit_trans(osb, handle);
5791out:
Al Viro59551022016-01-22 15:40:57 -05005792 inode_unlock(tl_inode);
Joseph Qi33add0e2013-07-03 15:00:52 -07005793bail:
Mark Fashehfecc0112008-11-12 15:16:38 -08005794 if (meta_ac)
5795 ocfs2_free_alloc_context(meta_ac);
5796
Tristan Ye78f94672010-05-11 17:54:42 +08005797 if (ref_tree)
5798 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5799
Mark Fashehfecc0112008-11-12 15:16:38 -08005800 return ret;
5801}
5802
Mark Fasheh063c4562007-07-03 13:34:11 -07005803int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005804{
5805 struct buffer_head *tl_bh = osb->osb_tl_bh;
5806 struct ocfs2_dinode *di;
5807 struct ocfs2_truncate_log *tl;
5808
5809 di = (struct ocfs2_dinode *) tl_bh->b_data;
5810 tl = &di->id2.i_dealloc;
5811
5812 mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5813 "slot %d, invalid truncate log parameters: used = "
5814 "%u, count = %u\n", osb->slot_num,
5815 le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5816 return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5817}
5818
5819static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5820 unsigned int new_start)
5821{
5822 unsigned int tail_index;
5823 unsigned int current_tail;
5824
5825 /* No records, nothing to coalesce */
5826 if (!le16_to_cpu(tl->tl_used))
5827 return 0;
5828
5829 tail_index = le16_to_cpu(tl->tl_used) - 1;
5830 current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5831 current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5832
5833 return current_tail == new_start;
5834}
5835
Mark Fasheh063c4562007-07-03 13:34:11 -07005836int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5837 handle_t *handle,
5838 u64 start_blk,
5839 unsigned int num_clusters)
Mark Fashehccd979b2005-12-15 14:31:24 -08005840{
5841 int status, index;
5842 unsigned int start_cluster, tl_count;
5843 struct inode *tl_inode = osb->osb_tl_inode;
5844 struct buffer_head *tl_bh = osb->osb_tl_bh;
5845 struct ocfs2_dinode *di;
5846 struct ocfs2_truncate_log *tl;
5847
Al Viro59551022016-01-22 15:40:57 -05005848 BUG_ON(inode_trylock(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08005849
5850 start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5851
5852 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08005853
Joel Becker10995aa2008-11-13 14:49:12 -08005854 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
5855 * by the underlying call to ocfs2_read_inode_block(), so any
5856 * corruption is a code bug */
5857 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5858
5859 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08005860 tl_count = le16_to_cpu(tl->tl_count);
5861 mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5862 tl_count == 0,
Mark Fashehb06970532006-03-03 10:24:33 -08005863 "Truncate record count on #%llu invalid "
5864 "wanted %u, actual %u\n",
5865 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08005866 ocfs2_truncate_recs_per_inode(osb->sb),
5867 le16_to_cpu(tl->tl_count));
5868
5869 /* Caller should have known to flush before calling us. */
5870 index = le16_to_cpu(tl->tl_used);
5871 if (index >= tl_count) {
5872 status = -ENOSPC;
5873 mlog_errno(status);
5874 goto bail;
5875 }
5876
Joel Becker0cf2f762009-02-12 16:41:25 -08005877 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005878 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005879 if (status < 0) {
5880 mlog_errno(status);
5881 goto bail;
5882 }
5883
Tao Maa09d09b2011-02-22 08:21:10 +08005884 trace_ocfs2_truncate_log_append(
5885 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5886 start_cluster, num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005887 if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5888 /*
5889 * Move index back to the record we are coalescing with.
5890 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5891 */
5892 index--;
5893
5894 num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
Tao Maa09d09b2011-02-22 08:21:10 +08005895 trace_ocfs2_truncate_log_append(
5896 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5897 index, le32_to_cpu(tl->tl_recs[index].t_start),
5898 num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005899 } else {
5900 tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5901 tl->tl_used = cpu_to_le16(index + 1);
5902 }
5903 tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5904
Joel Beckerec20cec2010-03-19 14:13:52 -07005905 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005906
Tao Ma50308d82010-11-04 15:14:11 +08005907 osb->truncated_clusters += num_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08005908bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005909 return status;
5910}
5911
5912static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
Mark Fashehccd979b2005-12-15 14:31:24 -08005913 struct inode *data_alloc_inode,
5914 struct buffer_head *data_alloc_bh)
5915{
5916 int status = 0;
5917 int i;
5918 unsigned int num_clusters;
5919 u64 start_blk;
5920 struct ocfs2_truncate_rec rec;
5921 struct ocfs2_dinode *di;
5922 struct ocfs2_truncate_log *tl;
5923 struct inode *tl_inode = osb->osb_tl_inode;
5924 struct buffer_head *tl_bh = osb->osb_tl_bh;
Junxiao Bi2b0ad002016-09-19 14:44:21 -07005925 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08005926
Mark Fashehccd979b2005-12-15 14:31:24 -08005927 di = (struct ocfs2_dinode *) tl_bh->b_data;
5928 tl = &di->id2.i_dealloc;
5929 i = le16_to_cpu(tl->tl_used) - 1;
5930 while (i >= 0) {
Junxiao Bi2b0ad002016-09-19 14:44:21 -07005931 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5932 if (IS_ERR(handle)) {
5933 status = PTR_ERR(handle);
5934 mlog_errno(status);
5935 goto bail;
5936 }
5937
Mark Fashehccd979b2005-12-15 14:31:24 -08005938 /* Caller has given us at least enough credits to
5939 * update the truncate log dinode */
Joel Becker0cf2f762009-02-12 16:41:25 -08005940 status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
Joel Becker13723d02008-10-17 19:25:01 -07005941 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08005942 if (status < 0) {
Chenyuan Miae3fab52021-11-05 13:34:45 -07005943 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005944 mlog_errno(status);
5945 goto bail;
5946 }
5947
5948 tl->tl_used = cpu_to_le16(i);
5949
Joel Beckerec20cec2010-03-19 14:13:52 -07005950 ocfs2_journal_dirty(handle, tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08005951
Mark Fashehccd979b2005-12-15 14:31:24 -08005952 rec = tl->tl_recs[i];
5953 start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5954 le32_to_cpu(rec.t_start));
5955 num_clusters = le32_to_cpu(rec.t_clusters);
5956
5957 /* if start_blk is not set, we ignore the record as
5958 * invalid. */
5959 if (start_blk) {
Tao Maa09d09b2011-02-22 08:21:10 +08005960 trace_ocfs2_replay_truncate_records(
5961 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5962 i, le32_to_cpu(rec.t_start), num_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08005963
5964 status = ocfs2_free_clusters(handle, data_alloc_inode,
5965 data_alloc_bh, start_blk,
5966 num_clusters);
5967 if (status < 0) {
Chenyuan Miae3fab52021-11-05 13:34:45 -07005968 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005969 mlog_errno(status);
5970 goto bail;
5971 }
5972 }
Xue jiufei102c2592016-03-25 14:21:44 -07005973
Junxiao Bi2b0ad002016-09-19 14:44:21 -07005974 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08005975 i--;
5976 }
5977
Tao Ma50308d82010-11-04 15:14:11 +08005978 osb->truncated_clusters = 0;
5979
Mark Fashehccd979b2005-12-15 14:31:24 -08005980bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08005981 return status;
5982}
5983
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08005984/* Expects you to already be holding tl_inode->i_mutex */
Mark Fasheh063c4562007-07-03 13:34:11 -07005985int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
Mark Fashehccd979b2005-12-15 14:31:24 -08005986{
5987 int status;
5988 unsigned int num_to_flush;
Mark Fashehccd979b2005-12-15 14:31:24 -08005989 struct inode *tl_inode = osb->osb_tl_inode;
5990 struct inode *data_alloc_inode = NULL;
5991 struct buffer_head *tl_bh = osb->osb_tl_bh;
5992 struct buffer_head *data_alloc_bh = NULL;
5993 struct ocfs2_dinode *di;
5994 struct ocfs2_truncate_log *tl;
Changwei Ged7283b392019-09-23 15:33:40 -07005995 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -08005996
Al Viro59551022016-01-22 15:40:57 -05005997 BUG_ON(inode_trylock(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08005998
5999 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006000
Joel Becker10995aa2008-11-13 14:49:12 -08006001 /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated
6002 * by the underlying call to ocfs2_read_inode_block(), so any
6003 * corruption is a code bug */
6004 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6005
6006 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006007 num_to_flush = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006008 trace_ocfs2_flush_truncate_log(
6009 (unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
6010 num_to_flush);
Mark Fashehccd979b2005-12-15 14:31:24 -08006011 if (!num_to_flush) {
6012 status = 0;
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006013 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006014 }
6015
Randy Dunlap679edeb2020-10-13 16:48:21 -07006016 /* Appending truncate log(TA) and flushing truncate log(TF) are
Changwei Ged7283b392019-09-23 15:33:40 -07006017 * two separated transactions. They can be both committed but not
6018 * checkpointed. If crash occurs then, both two transaction will be
6019 * replayed with several already released to global bitmap clusters.
6020 * Then truncate log will be replayed resulting in cluster double free.
6021 */
6022 jbd2_journal_lock_updates(journal->j_journal);
Leah Rumancik01d5d962021-05-18 15:13:25 +00006023 status = jbd2_journal_flush(journal->j_journal, 0);
Changwei Ged7283b392019-09-23 15:33:40 -07006024 jbd2_journal_unlock_updates(journal->j_journal);
6025 if (status < 0) {
6026 mlog_errno(status);
6027 goto out;
6028 }
6029
Mark Fashehccd979b2005-12-15 14:31:24 -08006030 data_alloc_inode = ocfs2_get_system_file_inode(osb,
6031 GLOBAL_BITMAP_SYSTEM_INODE,
6032 OCFS2_INVALID_SLOT);
6033 if (!data_alloc_inode) {
6034 status = -EINVAL;
6035 mlog(ML_ERROR, "Could not get bitmap inode!\n");
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006036 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08006037 }
6038
Al Viro59551022016-01-22 15:40:57 -05006039 inode_lock(data_alloc_inode);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006040
Mark Fashehe63aecb62007-10-18 15:30:42 -07006041 status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006042 if (status < 0) {
6043 mlog_errno(status);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006044 goto out_mutex;
Mark Fashehccd979b2005-12-15 14:31:24 -08006045 }
6046
Junxiao Bi2b0ad002016-09-19 14:44:21 -07006047 status = ocfs2_replay_truncate_records(osb, data_alloc_inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08006048 data_alloc_bh);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006049 if (status < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -08006050 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006051
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006052 brelse(data_alloc_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07006053 ocfs2_inode_unlock(data_alloc_inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08006054
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006055out_mutex:
Al Viro59551022016-01-22 15:40:57 -05006056 inode_unlock(data_alloc_inode);
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006057 iput(data_alloc_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006058
Mark Fashehe08dc8b2006-10-05 15:58:48 -07006059out:
Mark Fashehccd979b2005-12-15 14:31:24 -08006060 return status;
6061}
6062
6063int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6064{
6065 int status;
6066 struct inode *tl_inode = osb->osb_tl_inode;
6067
Al Viro59551022016-01-22 15:40:57 -05006068 inode_lock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006069 status = __ocfs2_flush_truncate_log(osb);
Al Viro59551022016-01-22 15:40:57 -05006070 inode_unlock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006071
6072 return status;
6073}
6074
David Howellsc4028952006-11-22 14:57:56 +00006075static void ocfs2_truncate_log_worker(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08006076{
6077 int status;
David Howellsc4028952006-11-22 14:57:56 +00006078 struct ocfs2_super *osb =
6079 container_of(work, struct ocfs2_super,
6080 osb_truncate_log_wq.work);
Mark Fashehccd979b2005-12-15 14:31:24 -08006081
Mark Fashehccd979b2005-12-15 14:31:24 -08006082 status = ocfs2_flush_truncate_log(osb);
6083 if (status < 0)
6084 mlog_errno(status);
Tao Ma4d0ddb22008-03-05 16:11:46 +08006085 else
Tiger Yangb89c5422010-01-25 14:11:06 +08006086 ocfs2_init_steal_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08006087}
6088
6089#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6090void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6091 int cancel)
6092{
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006093 if (osb->osb_tl_inode &&
6094 atomic_read(&osb->osb_tl_disable) == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006095 /* We want to push off log flushes while truncates are
6096 * still running. */
6097 if (cancel)
6098 cancel_delayed_work(&osb->osb_truncate_log_wq);
6099
jiangyiwen35ddf782016-03-25 14:21:32 -07006100 queue_delayed_work(osb->ocfs2_wq, &osb->osb_truncate_log_wq,
Mark Fashehccd979b2005-12-15 14:31:24 -08006101 OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6102 }
6103}
6104
Eric Ren2070ad12016-08-02 14:02:10 -07006105/*
6106 * Try to flush truncate logs if we can free enough clusters from it.
6107 * As for return value, "< 0" means error, "0" no space and "1" means
6108 * we have freed enough spaces and let the caller try to allocate again.
6109 */
6110int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
6111 unsigned int needed)
6112{
6113 tid_t target;
6114 int ret = 0;
6115 unsigned int truncated_clusters;
6116
6117 inode_lock(osb->osb_tl_inode);
6118 truncated_clusters = osb->truncated_clusters;
6119 inode_unlock(osb->osb_tl_inode);
6120
6121 /*
6122 * Check whether we can succeed in allocating if we free
6123 * the truncate log.
6124 */
6125 if (truncated_clusters < needed)
6126 goto out;
6127
6128 ret = ocfs2_flush_truncate_log(osb);
6129 if (ret) {
6130 mlog_errno(ret);
6131 goto out;
6132 }
6133
6134 if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
6135 jbd2_log_wait_commit(osb->journal->j_journal, target);
6136 ret = 1;
6137 }
6138out:
6139 return ret;
6140}
6141
Mark Fashehccd979b2005-12-15 14:31:24 -08006142static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6143 int slot_num,
6144 struct inode **tl_inode,
6145 struct buffer_head **tl_bh)
6146{
6147 int status;
6148 struct inode *inode = NULL;
6149 struct buffer_head *bh = NULL;
6150
6151 inode = ocfs2_get_system_file_inode(osb,
6152 TRUNCATE_LOG_SYSTEM_INODE,
6153 slot_num);
6154 if (!inode) {
6155 status = -EINVAL;
6156 mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6157 goto bail;
6158 }
6159
Joel Beckerb657c952008-11-13 14:49:11 -08006160 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006161 if (status < 0) {
6162 iput(inode);
6163 mlog_errno(status);
6164 goto bail;
6165 }
6166
6167 *tl_inode = inode;
6168 *tl_bh = bh;
6169bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08006170 return status;
6171}
6172
6173/* called during the 1st stage of node recovery. we stamp a clean
6174 * truncate log and pass back a copy for processing later. if the
6175 * truncate log does not require processing, a *tl_copy is set to
6176 * NULL. */
6177int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6178 int slot_num,
6179 struct ocfs2_dinode **tl_copy)
6180{
6181 int status;
6182 struct inode *tl_inode = NULL;
6183 struct buffer_head *tl_bh = NULL;
6184 struct ocfs2_dinode *di;
6185 struct ocfs2_truncate_log *tl;
6186
6187 *tl_copy = NULL;
6188
Tao Maa09d09b2011-02-22 08:21:10 +08006189 trace_ocfs2_begin_truncate_log_recovery(slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08006190
6191 status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6192 if (status < 0) {
6193 mlog_errno(status);
6194 goto bail;
6195 }
6196
6197 di = (struct ocfs2_dinode *) tl_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08006198
Joel Becker10995aa2008-11-13 14:49:12 -08006199 /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's
6200 * validated by the underlying call to ocfs2_read_inode_block(),
6201 * so any corruption is a code bug */
6202 BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6203
6204 tl = &di->id2.i_dealloc;
Mark Fashehccd979b2005-12-15 14:31:24 -08006205 if (le16_to_cpu(tl->tl_used)) {
Tao Maa09d09b2011-02-22 08:21:10 +08006206 trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
Mark Fashehccd979b2005-12-15 14:31:24 -08006207
Fuqian Huangd8b2fa62019-07-11 20:53:19 -07006208 /*
6209 * Assuming the write-out below goes well, this copy will be
6210 * passed back to recovery for processing.
6211 */
6212 *tl_copy = kmemdup(tl_bh->b_data, tl_bh->b_size, GFP_KERNEL);
Mark Fashehccd979b2005-12-15 14:31:24 -08006213 if (!(*tl_copy)) {
6214 status = -ENOMEM;
6215 mlog_errno(status);
6216 goto bail;
6217 }
6218
Mark Fashehccd979b2005-12-15 14:31:24 -08006219 /* All we need to do to clear the truncate log is set
6220 * tl_used. */
6221 tl->tl_used = 0;
6222
Joel Becker13723d02008-10-17 19:25:01 -07006223 ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08006224 status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08006225 if (status < 0) {
6226 mlog_errno(status);
6227 goto bail;
6228 }
6229 }
6230
6231bail:
Joseph Qi72865d92016-01-14 15:17:27 -08006232 iput(tl_inode);
Mark Fasheha81cb882008-10-07 14:25:16 -07006233 brelse(tl_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08006234
Joseph Qi46359292015-09-04 15:44:54 -07006235 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08006236 kfree(*tl_copy);
6237 *tl_copy = NULL;
Tao Mac1e8d352011-03-07 16:43:21 +08006238 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08006239 }
6240
Mark Fashehccd979b2005-12-15 14:31:24 -08006241 return status;
6242}
6243
6244int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6245 struct ocfs2_dinode *tl_copy)
6246{
6247 int status = 0;
6248 int i;
6249 unsigned int clusters, num_recs, start_cluster;
6250 u64 start_blk;
Mark Fasheh1fabe142006-10-09 18:11:45 -07006251 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08006252 struct inode *tl_inode = osb->osb_tl_inode;
6253 struct ocfs2_truncate_log *tl;
6254
Mark Fashehccd979b2005-12-15 14:31:24 -08006255 if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6256 mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6257 return -EINVAL;
6258 }
6259
6260 tl = &tl_copy->id2.i_dealloc;
6261 num_recs = le16_to_cpu(tl->tl_used);
Tao Maa09d09b2011-02-22 08:21:10 +08006262 trace_ocfs2_complete_truncate_log_recovery(
6263 (unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6264 num_recs);
Mark Fashehccd979b2005-12-15 14:31:24 -08006265
Al Viro59551022016-01-22 15:40:57 -05006266 inode_lock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006267 for(i = 0; i < num_recs; i++) {
6268 if (ocfs2_truncate_log_needs_flush(osb)) {
6269 status = __ocfs2_flush_truncate_log(osb);
6270 if (status < 0) {
6271 mlog_errno(status);
6272 goto bail_up;
6273 }
6274 }
6275
Mark Fasheh65eff9c2006-10-09 17:26:22 -07006276 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
Mark Fashehccd979b2005-12-15 14:31:24 -08006277 if (IS_ERR(handle)) {
6278 status = PTR_ERR(handle);
6279 mlog_errno(status);
6280 goto bail_up;
6281 }
6282
6283 clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6284 start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6285 start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6286
6287 status = ocfs2_truncate_log_append(osb, handle,
6288 start_blk, clusters);
Mark Fasheh02dc1af2006-10-09 16:48:10 -07006289 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08006290 if (status < 0) {
6291 mlog_errno(status);
6292 goto bail_up;
6293 }
6294 }
6295
6296bail_up:
Al Viro59551022016-01-22 15:40:57 -05006297 inode_unlock(tl_inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08006298
Mark Fashehccd979b2005-12-15 14:31:24 -08006299 return status;
6300}
6301
6302void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6303{
6304 int status;
6305 struct inode *tl_inode = osb->osb_tl_inode;
6306
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006307 atomic_set(&osb->osb_tl_disable, 1);
6308
Mark Fashehccd979b2005-12-15 14:31:24 -08006309 if (tl_inode) {
6310 cancel_delayed_work(&osb->osb_truncate_log_wq);
jiangyiwen35ddf782016-03-25 14:21:32 -07006311 flush_workqueue(osb->ocfs2_wq);
Mark Fashehccd979b2005-12-15 14:31:24 -08006312
6313 status = ocfs2_flush_truncate_log(osb);
6314 if (status < 0)
6315 mlog_errno(status);
6316
6317 brelse(osb->osb_tl_bh);
6318 iput(osb->osb_tl_inode);
6319 }
Mark Fashehccd979b2005-12-15 14:31:24 -08006320}
6321
6322int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6323{
6324 int status;
6325 struct inode *tl_inode = NULL;
6326 struct buffer_head *tl_bh = NULL;
6327
Mark Fashehccd979b2005-12-15 14:31:24 -08006328 status = ocfs2_get_truncate_log_info(osb,
6329 osb->slot_num,
6330 &tl_inode,
6331 &tl_bh);
6332 if (status < 0)
6333 mlog_errno(status);
6334
6335 /* ocfs2_truncate_log_shutdown keys on the existence of
6336 * osb->osb_tl_inode so we don't set any of the osb variables
6337 * until we're sure all is well. */
David Howellsc4028952006-11-22 14:57:56 +00006338 INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6339 ocfs2_truncate_log_worker);
Xue jiufeia9e9aca2014-06-04 16:06:13 -07006340 atomic_set(&osb->osb_tl_disable, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08006341 osb->osb_tl_bh = tl_bh;
6342 osb->osb_tl_inode = tl_inode;
6343
Mark Fashehccd979b2005-12-15 14:31:24 -08006344 return status;
6345}
6346
Mark Fasheh2b604352007-06-22 15:45:27 -07006347/*
6348 * Delayed de-allocation of suballocator blocks.
6349 *
6350 * Some sets of block de-allocations might involve multiple suballocator inodes.
6351 *
6352 * The locking for this can get extremely complicated, especially when
6353 * the suballocator inodes to delete from aren't known until deep
6354 * within an unrelated codepath.
6355 *
6356 * ocfs2_extent_block structures are a good example of this - an inode
6357 * btree could have been grown by any number of nodes each allocating
6358 * out of their own suballoc inode.
6359 *
6360 * These structures allow the delay of block de-allocation until a
6361 * later time, when locking of multiple cluster inodes won't cause
6362 * deadlock.
6363 */
6364
6365/*
Tao Ma2891d292008-11-12 08:26:58 +08006366 * Describe a single bit freed from a suballocator. For the block
6367 * suballocators, it represents one block. For the global cluster
6368 * allocator, it represents some clusters and free_bit indicates
6369 * clusters number.
Mark Fasheh2b604352007-06-22 15:45:27 -07006370 */
6371struct ocfs2_cached_block_free {
6372 struct ocfs2_cached_block_free *free_next;
Tao Ma74380c42010-03-22 14:20:18 +08006373 u64 free_bg;
Mark Fasheh2b604352007-06-22 15:45:27 -07006374 u64 free_blk;
6375 unsigned int free_bit;
6376};
6377
6378struct ocfs2_per_slot_free_list {
6379 struct ocfs2_per_slot_free_list *f_next_suballocator;
6380 int f_inode_type;
6381 int f_slot;
6382 struct ocfs2_cached_block_free *f_first;
6383};
6384
Tao Ma2891d292008-11-12 08:26:58 +08006385static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6386 int sysfile_type,
6387 int slot,
6388 struct ocfs2_cached_block_free *head)
Mark Fasheh2b604352007-06-22 15:45:27 -07006389{
6390 int ret;
6391 u64 bg_blkno;
6392 handle_t *handle;
6393 struct inode *inode;
6394 struct buffer_head *di_bh = NULL;
6395 struct ocfs2_cached_block_free *tmp;
6396
6397 inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6398 if (!inode) {
6399 ret = -EINVAL;
6400 mlog_errno(ret);
6401 goto out;
6402 }
6403
Al Viro59551022016-01-22 15:40:57 -05006404 inode_lock(inode);
Mark Fasheh2b604352007-06-22 15:45:27 -07006405
Mark Fashehe63aecb62007-10-18 15:30:42 -07006406 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006407 if (ret) {
6408 mlog_errno(ret);
6409 goto out_mutex;
6410 }
6411
Mark Fasheh2b604352007-06-22 15:45:27 -07006412 while (head) {
Tao Ma74380c42010-03-22 14:20:18 +08006413 if (head->free_bg)
6414 bg_blkno = head->free_bg;
6415 else
6416 bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6417 head->free_bit);
Junxiao Bid5bf1412016-09-19 14:44:24 -07006418 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6419 if (IS_ERR(handle)) {
6420 ret = PTR_ERR(handle);
6421 mlog_errno(ret);
6422 goto out_unlock;
6423 }
6424
Tao Maa09d09b2011-02-22 08:21:10 +08006425 trace_ocfs2_free_cached_blocks(
6426 (unsigned long long)head->free_blk, head->free_bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006427
6428 ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6429 head->free_bit, bg_blkno, 1);
Junxiao Bid5bf1412016-09-19 14:44:24 -07006430 if (ret)
Mark Fasheh2b604352007-06-22 15:45:27 -07006431 mlog_errno(ret);
Mark Fasheh2b604352007-06-22 15:45:27 -07006432
Junxiao Bid5bf1412016-09-19 14:44:24 -07006433 ocfs2_commit_trans(osb, handle);
Mark Fasheh2b604352007-06-22 15:45:27 -07006434
6435 tmp = head;
6436 head = head->free_next;
6437 kfree(tmp);
6438 }
6439
Mark Fasheh2b604352007-06-22 15:45:27 -07006440out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07006441 ocfs2_inode_unlock(inode, 1);
Mark Fasheh2b604352007-06-22 15:45:27 -07006442 brelse(di_bh);
6443out_mutex:
Al Viro59551022016-01-22 15:40:57 -05006444 inode_unlock(inode);
Mark Fasheh2b604352007-06-22 15:45:27 -07006445 iput(inode);
6446out:
6447 while(head) {
6448 /* Premature exit may have left some dangling items. */
6449 tmp = head;
6450 head = head->free_next;
6451 kfree(tmp);
6452 }
6453
6454 return ret;
6455}
6456
Tao Ma2891d292008-11-12 08:26:58 +08006457int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6458 u64 blkno, unsigned int bit)
6459{
6460 int ret = 0;
6461 struct ocfs2_cached_block_free *item;
6462
Tao Ma74380c42010-03-22 14:20:18 +08006463 item = kzalloc(sizeof(*item), GFP_NOFS);
Tao Ma2891d292008-11-12 08:26:58 +08006464 if (item == NULL) {
6465 ret = -ENOMEM;
6466 mlog_errno(ret);
6467 return ret;
6468 }
6469
Tao Maa09d09b2011-02-22 08:21:10 +08006470 trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
Tao Ma2891d292008-11-12 08:26:58 +08006471
6472 item->free_blk = blkno;
6473 item->free_bit = bit;
6474 item->free_next = ctxt->c_global_allocator;
6475
6476 ctxt->c_global_allocator = item;
6477 return ret;
6478}
6479
6480static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6481 struct ocfs2_cached_block_free *head)
6482{
6483 struct ocfs2_cached_block_free *tmp;
6484 struct inode *tl_inode = osb->osb_tl_inode;
6485 handle_t *handle;
6486 int ret = 0;
6487
Al Viro59551022016-01-22 15:40:57 -05006488 inode_lock(tl_inode);
Tao Ma2891d292008-11-12 08:26:58 +08006489
6490 while (head) {
6491 if (ocfs2_truncate_log_needs_flush(osb)) {
6492 ret = __ocfs2_flush_truncate_log(osb);
6493 if (ret < 0) {
6494 mlog_errno(ret);
6495 break;
6496 }
6497 }
6498
6499 handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6500 if (IS_ERR(handle)) {
6501 ret = PTR_ERR(handle);
6502 mlog_errno(ret);
6503 break;
6504 }
6505
6506 ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6507 head->free_bit);
6508
6509 ocfs2_commit_trans(osb, handle);
6510 tmp = head;
6511 head = head->free_next;
6512 kfree(tmp);
6513
6514 if (ret < 0) {
6515 mlog_errno(ret);
6516 break;
6517 }
6518 }
6519
Al Viro59551022016-01-22 15:40:57 -05006520 inode_unlock(tl_inode);
Tao Ma2891d292008-11-12 08:26:58 +08006521
6522 while (head) {
6523 /* Premature exit may have left some dangling items. */
6524 tmp = head;
6525 head = head->free_next;
6526 kfree(tmp);
6527 }
6528
6529 return ret;
6530}
6531
Mark Fasheh2b604352007-06-22 15:45:27 -07006532int ocfs2_run_deallocs(struct ocfs2_super *osb,
6533 struct ocfs2_cached_dealloc_ctxt *ctxt)
6534{
6535 int ret = 0, ret2;
6536 struct ocfs2_per_slot_free_list *fl;
6537
6538 if (!ctxt)
6539 return 0;
6540
6541 while (ctxt->c_first_suballocator) {
6542 fl = ctxt->c_first_suballocator;
6543
6544 if (fl->f_first) {
Tao Maa09d09b2011-02-22 08:21:10 +08006545 trace_ocfs2_run_deallocs(fl->f_inode_type,
6546 fl->f_slot);
Tao Ma2891d292008-11-12 08:26:58 +08006547 ret2 = ocfs2_free_cached_blocks(osb,
6548 fl->f_inode_type,
6549 fl->f_slot,
6550 fl->f_first);
Mark Fasheh2b604352007-06-22 15:45:27 -07006551 if (ret2)
6552 mlog_errno(ret2);
6553 if (!ret)
6554 ret = ret2;
6555 }
6556
6557 ctxt->c_first_suballocator = fl->f_next_suballocator;
6558 kfree(fl);
6559 }
6560
Tao Ma2891d292008-11-12 08:26:58 +08006561 if (ctxt->c_global_allocator) {
6562 ret2 = ocfs2_free_cached_clusters(osb,
6563 ctxt->c_global_allocator);
6564 if (ret2)
6565 mlog_errno(ret2);
6566 if (!ret)
6567 ret = ret2;
6568
6569 ctxt->c_global_allocator = NULL;
6570 }
6571
Mark Fasheh2b604352007-06-22 15:45:27 -07006572 return ret;
6573}
6574
6575static struct ocfs2_per_slot_free_list *
6576ocfs2_find_per_slot_free_list(int type,
6577 int slot,
6578 struct ocfs2_cached_dealloc_ctxt *ctxt)
6579{
6580 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6581
6582 while (fl) {
6583 if (fl->f_inode_type == type && fl->f_slot == slot)
6584 return fl;
6585
6586 fl = fl->f_next_suballocator;
6587 }
6588
6589 fl = kmalloc(sizeof(*fl), GFP_NOFS);
6590 if (fl) {
6591 fl->f_inode_type = type;
6592 fl->f_slot = slot;
6593 fl->f_first = NULL;
6594 fl->f_next_suballocator = ctxt->c_first_suballocator;
6595
6596 ctxt->c_first_suballocator = fl;
6597 }
6598 return fl;
6599}
6600
Changwei Ge71a36942018-01-31 16:15:06 -08006601static struct ocfs2_per_slot_free_list *
6602ocfs2_find_preferred_free_list(int type,
6603 int preferred_slot,
6604 int *real_slot,
6605 struct ocfs2_cached_dealloc_ctxt *ctxt)
6606{
6607 struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6608
6609 while (fl) {
6610 if (fl->f_inode_type == type && fl->f_slot == preferred_slot) {
6611 *real_slot = fl->f_slot;
6612 return fl;
6613 }
6614
6615 fl = fl->f_next_suballocator;
6616 }
6617
6618 /* If we can't find any free list matching preferred slot, just use
6619 * the first one.
6620 */
6621 fl = ctxt->c_first_suballocator;
6622 *real_slot = fl->f_slot;
6623
6624 return fl;
6625}
6626
6627/* Return Value 1 indicates empty */
6628static int ocfs2_is_dealloc_empty(struct ocfs2_extent_tree *et)
6629{
6630 struct ocfs2_per_slot_free_list *fl = NULL;
6631
6632 if (!et->et_dealloc)
6633 return 1;
6634
6635 fl = et->et_dealloc->c_first_suballocator;
6636 if (!fl)
6637 return 1;
6638
6639 if (!fl->f_first)
6640 return 1;
6641
6642 return 0;
6643}
6644
6645/* If extent was deleted from tree due to extent rotation and merging, and
6646 * no metadata is reserved ahead of time. Try to reuse some extents
6647 * just deleted. This is only used to reuse extent blocks.
6648 * It is supposed to find enough extent blocks in dealloc if our estimation
6649 * on metadata is accurate.
6650 */
6651static int ocfs2_reuse_blk_from_dealloc(handle_t *handle,
6652 struct ocfs2_extent_tree *et,
6653 struct buffer_head **new_eb_bh,
6654 int blk_wanted, int *blk_given)
6655{
6656 int i, status = 0, real_slot;
6657 struct ocfs2_cached_dealloc_ctxt *dealloc;
6658 struct ocfs2_per_slot_free_list *fl;
6659 struct ocfs2_cached_block_free *bf;
6660 struct ocfs2_extent_block *eb;
6661 struct ocfs2_super *osb =
6662 OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
6663
6664 *blk_given = 0;
6665
6666 /* If extent tree doesn't have a dealloc, this is not faulty. Just
6667 * tell upper caller dealloc can't provide any block and it should
6668 * ask for alloc to claim more space.
6669 */
6670 dealloc = et->et_dealloc;
6671 if (!dealloc)
6672 goto bail;
6673
6674 for (i = 0; i < blk_wanted; i++) {
6675 /* Prefer to use local slot */
6676 fl = ocfs2_find_preferred_free_list(EXTENT_ALLOC_SYSTEM_INODE,
6677 osb->slot_num, &real_slot,
6678 dealloc);
6679 /* If no more block can be reused, we should claim more
6680 * from alloc. Just return here normally.
6681 */
6682 if (!fl) {
6683 status = 0;
6684 break;
6685 }
6686
6687 bf = fl->f_first;
6688 fl->f_first = bf->free_next;
6689
6690 new_eb_bh[i] = sb_getblk(osb->sb, bf->free_blk);
6691 if (new_eb_bh[i] == NULL) {
6692 status = -ENOMEM;
6693 mlog_errno(status);
6694 goto bail;
6695 }
6696
6697 mlog(0, "Reusing block(%llu) from "
6698 "dealloc(local slot:%d, real slot:%d)\n",
6699 bf->free_blk, osb->slot_num, real_slot);
6700
6701 ocfs2_set_new_buffer_uptodate(et->et_ci, new_eb_bh[i]);
6702
6703 status = ocfs2_journal_access_eb(handle, et->et_ci,
6704 new_eb_bh[i],
6705 OCFS2_JOURNAL_ACCESS_CREATE);
6706 if (status < 0) {
6707 mlog_errno(status);
6708 goto bail;
6709 }
6710
6711 memset(new_eb_bh[i]->b_data, 0, osb->sb->s_blocksize);
6712 eb = (struct ocfs2_extent_block *) new_eb_bh[i]->b_data;
6713
6714 /* We can't guarantee that buffer head is still cached, so
6715 * polutlate the extent block again.
6716 */
6717 strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
6718 eb->h_blkno = cpu_to_le64(bf->free_blk);
6719 eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
6720 eb->h_suballoc_slot = cpu_to_le16(real_slot);
6721 eb->h_suballoc_loc = cpu_to_le64(bf->free_bg);
6722 eb->h_suballoc_bit = cpu_to_le16(bf->free_bit);
6723 eb->h_list.l_count =
6724 cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
6725
6726 /* We'll also be dirtied by the caller, so
6727 * this isn't absolutely necessary.
6728 */
6729 ocfs2_journal_dirty(handle, new_eb_bh[i]);
6730
6731 if (!fl->f_first) {
6732 dealloc->c_first_suballocator = fl->f_next_suballocator;
6733 kfree(fl);
6734 }
6735 kfree(bf);
6736 }
6737
6738 *blk_given = i;
6739
6740bail:
6741 if (unlikely(status < 0)) {
6742 for (i = 0; i < blk_wanted; i++)
6743 brelse(new_eb_bh[i]);
6744 }
6745
6746 return status;
6747}
6748
Tao Ma1823cb02009-08-18 11:24:49 +08006749int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
Tao Ma74380c42010-03-22 14:20:18 +08006750 int type, int slot, u64 suballoc,
6751 u64 blkno, unsigned int bit)
Mark Fasheh2b604352007-06-22 15:45:27 -07006752{
6753 int ret;
6754 struct ocfs2_per_slot_free_list *fl;
6755 struct ocfs2_cached_block_free *item;
6756
6757 fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6758 if (fl == NULL) {
6759 ret = -ENOMEM;
6760 mlog_errno(ret);
6761 goto out;
6762 }
6763
Tao Ma74380c42010-03-22 14:20:18 +08006764 item = kzalloc(sizeof(*item), GFP_NOFS);
Mark Fasheh2b604352007-06-22 15:45:27 -07006765 if (item == NULL) {
6766 ret = -ENOMEM;
6767 mlog_errno(ret);
6768 goto out;
6769 }
6770
Tao Maa09d09b2011-02-22 08:21:10 +08006771 trace_ocfs2_cache_block_dealloc(type, slot,
6772 (unsigned long long)suballoc,
6773 (unsigned long long)blkno, bit);
Mark Fasheh2b604352007-06-22 15:45:27 -07006774
Tao Ma74380c42010-03-22 14:20:18 +08006775 item->free_bg = suballoc;
Mark Fasheh2b604352007-06-22 15:45:27 -07006776 item->free_blk = blkno;
6777 item->free_bit = bit;
6778 item->free_next = fl->f_first;
6779
6780 fl->f_first = item;
6781
6782 ret = 0;
6783out:
6784 return ret;
6785}
6786
Mark Fasheh59a5e412007-06-22 15:52:36 -07006787static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6788 struct ocfs2_extent_block *eb)
6789{
6790 return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6791 le16_to_cpu(eb->h_suballoc_slot),
Tao Ma74380c42010-03-22 14:20:18 +08006792 le64_to_cpu(eb->h_suballoc_loc),
Mark Fasheh59a5e412007-06-22 15:52:36 -07006793 le64_to_cpu(eb->h_blkno),
6794 le16_to_cpu(eb->h_suballoc_bit));
6795}
6796
Joel Becker2b4e30f2008-09-03 20:03:41 -07006797static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
Mark Fasheh60b11392007-02-16 11:46:50 -08006798{
6799 set_buffer_uptodate(bh);
6800 mark_buffer_dirty(bh);
6801 return 0;
6802}
6803
Tao Ma6f70fa52009-08-25 08:05:12 +08006804void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6805 unsigned int from, unsigned int to,
6806 struct page *page, int zero, u64 *phys)
Mark Fasheh1d410a62007-09-07 14:20:45 -07006807{
6808 int ret, partial = 0;
Joseph Qibbd0f322019-09-23 15:33:08 -07006809 loff_t start_byte = ((loff_t)page->index << PAGE_SHIFT) + from;
6810 loff_t length = to - from;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006811
6812 ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6813 if (ret)
6814 mlog_errno(ret);
6815
6816 if (zero)
Christoph Lametereebd2aa2008-02-04 22:28:29 -08006817 zero_user_segment(page, from, to);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006818
6819 /*
6820 * Need to set the buffers we zero'd into uptodate
6821 * here if they aren't - ocfs2_map_page_blocks()
6822 * might've skipped some
6823 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07006824 ret = walk_page_buffers(handle, page_buffers(page),
6825 from, to, &partial,
6826 ocfs2_zero_func);
6827 if (ret < 0)
6828 mlog_errno(ret);
6829 else if (ocfs2_should_order_data(inode)) {
Joseph Qibbd0f322019-09-23 15:33:08 -07006830 ret = ocfs2_jbd2_inode_add_write(handle, inode,
6831 start_byte, length);
Mark Fasheh1d410a62007-09-07 14:20:45 -07006832 if (ret < 0)
6833 mlog_errno(ret);
6834 }
6835
6836 if (!partial)
6837 SetPageUptodate(page);
6838
6839 flush_dcache_page(page);
6840}
6841
Mark Fasheh35edec12007-07-06 14:41:18 -07006842static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6843 loff_t end, struct page **pages,
6844 int numpages, u64 phys, handle_t *handle)
Mark Fasheh60b11392007-02-16 11:46:50 -08006845{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006846 int i;
Mark Fasheh60b11392007-02-16 11:46:50 -08006847 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006848 unsigned int from, to = PAGE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006849 struct super_block *sb = inode->i_sb;
6850
6851 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6852
6853 if (numpages == 0)
6854 goto out;
6855
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006856 to = PAGE_SIZE;
Mark Fasheh60b11392007-02-16 11:46:50 -08006857 for(i = 0; i < numpages; i++) {
6858 page = pages[i];
6859
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006860 from = start & (PAGE_SIZE - 1);
6861 if ((end >> PAGE_SHIFT) == page->index)
6862 to = end & (PAGE_SIZE - 1);
Mark Fasheh35edec12007-07-06 14:41:18 -07006863
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006864 BUG_ON(from > PAGE_SIZE);
6865 BUG_ON(to > PAGE_SIZE);
Mark Fasheh60b11392007-02-16 11:46:50 -08006866
Mark Fasheh1d410a62007-09-07 14:20:45 -07006867 ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6868 &phys);
Mark Fasheh60b11392007-02-16 11:46:50 -08006869
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006870 start = (page->index + 1) << PAGE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006871 }
6872out:
Mark Fasheh1d410a62007-09-07 14:20:45 -07006873 if (pages)
6874 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006875}
6876
Tao Ma6f70fa52009-08-25 08:05:12 +08006877int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6878 struct page **pages, int *num)
Mark Fasheh60b11392007-02-16 11:46:50 -08006879{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006880 int numpages, ret = 0;
Mark Fasheh60b11392007-02-16 11:46:50 -08006881 struct address_space *mapping = inode->i_mapping;
6882 unsigned long index;
Mark Fasheh35edec12007-07-06 14:41:18 -07006883 loff_t last_page_bytes;
Mark Fasheh60b11392007-02-16 11:46:50 -08006884
Mark Fasheh35edec12007-07-06 14:41:18 -07006885 BUG_ON(start > end);
Mark Fasheh60b11392007-02-16 11:46:50 -08006886
Mark Fasheh1d410a62007-09-07 14:20:45 -07006887 numpages = 0;
Mark Fasheh35edec12007-07-06 14:41:18 -07006888 last_page_bytes = PAGE_ALIGN(end);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006889 index = start >> PAGE_SHIFT;
Mark Fasheh60b11392007-02-16 11:46:50 -08006890 do {
Jan Kara9b4c0ff2010-08-24 14:28:03 +02006891 pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
Mark Fasheh60b11392007-02-16 11:46:50 -08006892 if (!pages[numpages]) {
6893 ret = -ENOMEM;
6894 mlog_errno(ret);
6895 goto out;
6896 }
6897
6898 numpages++;
6899 index++;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006900 } while (index < (last_page_bytes >> PAGE_SHIFT));
Mark Fasheh60b11392007-02-16 11:46:50 -08006901
6902out:
6903 if (ret != 0) {
Mark Fasheh1d410a62007-09-07 14:20:45 -07006904 if (pages)
6905 ocfs2_unlock_and_free_pages(pages, numpages);
Mark Fasheh60b11392007-02-16 11:46:50 -08006906 numpages = 0;
6907 }
6908
6909 *num = numpages;
6910
6911 return ret;
6912}
6913
Tao Ma6f70fa52009-08-25 08:05:12 +08006914static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6915 struct page **pages, int *num)
6916{
6917 struct super_block *sb = inode->i_sb;
6918
6919 BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6920 (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6921
6922 return ocfs2_grab_pages(inode, start, end, pages, num);
6923}
6924
Mark Fasheh60b11392007-02-16 11:46:50 -08006925/*
Jan Karac7c14a32021-11-05 13:34:58 -07006926 * Zero partial cluster for a hole punch or truncate. This avoids exposing
6927 * nonzero data on subsequent file extends.
Mark Fasheh60b11392007-02-16 11:46:50 -08006928 *
6929 * We need to call this before i_size is updated on the inode because
6930 * otherwise block_write_full_page() will skip writeout of pages past
Jan Karac7c14a32021-11-05 13:34:58 -07006931 * i_size.
Mark Fasheh60b11392007-02-16 11:46:50 -08006932 */
Mark Fasheh35edec12007-07-06 14:41:18 -07006933int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6934 u64 range_start, u64 range_end)
Mark Fasheh60b11392007-02-16 11:46:50 -08006935{
Mark Fasheh1d410a62007-09-07 14:20:45 -07006936 int ret = 0, numpages;
Mark Fasheh60b11392007-02-16 11:46:50 -08006937 struct page **pages = NULL;
6938 u64 phys;
Mark Fasheh1d410a62007-09-07 14:20:45 -07006939 unsigned int ext_flags;
6940 struct super_block *sb = inode->i_sb;
Mark Fasheh60b11392007-02-16 11:46:50 -08006941
6942 /*
6943 * File systems which don't support sparse files zero on every
6944 * extend.
6945 */
Mark Fasheh1d410a62007-09-07 14:20:45 -07006946 if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
Mark Fasheh60b11392007-02-16 11:46:50 -08006947 return 0;
6948
Jan Karac7c14a32021-11-05 13:34:58 -07006949 /*
6950 * Avoid zeroing pages fully beyond current i_size. It is pointless as
6951 * underlying blocks of those pages should be already zeroed out and
6952 * page writeback will skip them anyway.
6953 */
6954 range_end = min_t(u64, range_end, i_size_read(inode));
6955 if (range_start >= range_end)
6956 return 0;
6957
Mark Fasheh1d410a62007-09-07 14:20:45 -07006958 pages = kcalloc(ocfs2_pages_per_cluster(sb),
Mark Fasheh60b11392007-02-16 11:46:50 -08006959 sizeof(struct page *), GFP_NOFS);
6960 if (pages == NULL) {
6961 ret = -ENOMEM;
6962 mlog_errno(ret);
6963 goto out;
6964 }
6965
Mark Fasheh1d410a62007-09-07 14:20:45 -07006966 ret = ocfs2_extent_map_get_blocks(inode,
6967 range_start >> sb->s_blocksize_bits,
6968 &phys, NULL, &ext_flags);
Mark Fasheh60b11392007-02-16 11:46:50 -08006969 if (ret) {
6970 mlog_errno(ret);
6971 goto out;
6972 }
6973
Mark Fasheh1d410a62007-09-07 14:20:45 -07006974 /*
6975 * Tail is a hole, or is marked unwritten. In either case, we
6976 * can count on read and write to return/push zero's.
6977 */
6978 if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
Mark Fasheh60b11392007-02-16 11:46:50 -08006979 goto out;
6980
Mark Fasheh1d410a62007-09-07 14:20:45 -07006981 ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6982 &numpages);
6983 if (ret) {
6984 mlog_errno(ret);
6985 goto out;
6986 }
6987
Mark Fasheh35edec12007-07-06 14:41:18 -07006988 ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6989 numpages, phys, handle);
Mark Fasheh60b11392007-02-16 11:46:50 -08006990
6991 /*
6992 * Initiate writeout of the pages we zero'd here. We don't
6993 * wait on them - the truncate_inode_pages() call later will
6994 * do that for us.
6995 */
Christoph Hellwig2cfd30a2009-09-23 15:04:02 +02006996 ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6997 range_end - 1);
Mark Fasheh60b11392007-02-16 11:46:50 -08006998 if (ret)
6999 mlog_errno(ret);
7000
7001out:
Tim Gardnerd787ab02013-02-21 16:42:44 -08007002 kfree(pages);
Mark Fasheh60b11392007-02-16 11:46:50 -08007003
7004 return ret;
7005}
7006
Tiger Yangfdd77702008-08-18 17:08:55 +08007007static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
7008 struct ocfs2_dinode *di)
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007009{
7010 unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
Tiger Yangfdd77702008-08-18 17:08:55 +08007011 unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007012
Tiger Yangfdd77702008-08-18 17:08:55 +08007013 if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
7014 memset(&di->id2, 0, blocksize -
7015 offsetof(struct ocfs2_dinode, id2) -
7016 xattrsize);
7017 else
7018 memset(&di->id2, 0, blocksize -
7019 offsetof(struct ocfs2_dinode, id2));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007020}
7021
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007022void ocfs2_dinode_new_extent_list(struct inode *inode,
7023 struct ocfs2_dinode *di)
7024{
Tiger Yangfdd77702008-08-18 17:08:55 +08007025 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007026 di->id2.i_list.l_tree_depth = 0;
7027 di->id2.i_list.l_next_free_rec = 0;
Tiger Yangfdd77702008-08-18 17:08:55 +08007028 di->id2.i_list.l_count = cpu_to_le16(
7029 ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007030}
7031
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007032void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
7033{
7034 struct ocfs2_inode_info *oi = OCFS2_I(inode);
7035 struct ocfs2_inline_data *idata = &di->id2.i_data;
7036
7037 spin_lock(&oi->ip_lock);
7038 oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
7039 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
7040 spin_unlock(&oi->ip_lock);
7041
7042 /*
7043 * We clear the entire i_data structure here so that all
7044 * fields can be properly initialized.
7045 */
Tiger Yangfdd77702008-08-18 17:08:55 +08007046 ocfs2_zero_dinode_id2_with_xattr(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007047
Tiger Yangfdd77702008-08-18 17:08:55 +08007048 idata->id_count = cpu_to_le16(
7049 ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007050}
7051
7052int ocfs2_convert_inline_data_to_extents(struct inode *inode,
7053 struct buffer_head *di_bh)
7054{
Jan Kara5314454e2021-10-18 15:15:39 -07007055 int ret, has_data, num_pages = 0;
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007056 int need_free = 0;
7057 u32 bit_off, num;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007058 handle_t *handle;
Kees Cook3f649ab2020-06-03 13:09:38 -07007059 u64 block;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007060 struct ocfs2_inode_info *oi = OCFS2_I(inode);
7061 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7062 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007063 struct ocfs2_alloc_context *data_ac = NULL;
Jan Kara5314454e2021-10-18 15:15:39 -07007064 struct page *page = NULL;
Joel Beckerf99b9b72008-08-20 19:36:33 -07007065 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +02007066 int did_quota = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007067
7068 has_data = i_size_read(inode) ? 1 : 0;
7069
7070 if (has_data) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007071 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
7072 if (ret) {
7073 mlog_errno(ret);
Jan Kara5314454e2021-10-18 15:15:39 -07007074 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007075 }
7076 }
7077
Jan Karaa90714c2008-10-09 19:38:40 +02007078 handle = ocfs2_start_trans(osb,
7079 ocfs2_inline_to_extents_credits(osb->sb));
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007080 if (IS_ERR(handle)) {
7081 ret = PTR_ERR(handle);
7082 mlog_errno(ret);
alex chen15eba0f2015-02-10 14:09:02 -08007083 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007084 }
7085
Joel Becker0cf2f762009-02-12 16:41:25 -08007086 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007087 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007088 if (ret) {
7089 mlog_errno(ret);
7090 goto out_commit;
7091 }
7092
7093 if (has_data) {
Jan Kara5314454e2021-10-18 15:15:39 -07007094 unsigned int page_end = min_t(unsigned, PAGE_SIZE,
7095 osb->s_clustersize);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007096 u64 phys;
7097
Christoph Hellwig5dd40562010-03-03 09:05:00 -05007098 ret = dquot_alloc_space_nodirty(inode,
7099 ocfs2_clusters_to_bytes(osb->sb, 1));
7100 if (ret)
Jan Karaa90714c2008-10-09 19:38:40 +02007101 goto out_commit;
Jan Karaa90714c2008-10-09 19:38:40 +02007102 did_quota = 1;
7103
piaojund324cd42018-04-05 16:18:37 -07007104 data_ac->ac_resv = &oi->ip_la_data_resv;
Mark Fasheh4fe370a2009-12-07 13:15:40 -08007105
Joel Becker1ed9b772010-05-06 13:59:06 +08007106 ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007107 &num);
7108 if (ret) {
7109 mlog_errno(ret);
7110 goto out_commit;
7111 }
7112
7113 /*
7114 * Save two copies, one for insert, and one that can
7115 * be changed by ocfs2_map_and_dirty_page() below.
7116 */
7117 block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
7118
Jan Kara5314454e2021-10-18 15:15:39 -07007119 ret = ocfs2_grab_eof_pages(inode, 0, page_end, &page,
7120 &num_pages);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007121 if (ret) {
7122 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007123 need_free = 1;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007124 goto out_commit;
7125 }
7126
7127 /*
7128 * This should populate the 1st page for us and mark
7129 * it up to date.
7130 */
Jan Kara5314454e2021-10-18 15:15:39 -07007131 ret = ocfs2_read_inline_data(inode, page, di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007132 if (ret) {
7133 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007134 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08007135 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007136 }
7137
Jan Kara5314454e2021-10-18 15:15:39 -07007138 ocfs2_map_and_dirty_page(inode, handle, 0, page_end, page, 0,
7139 &phys);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007140 }
7141
7142 spin_lock(&oi->ip_lock);
7143 oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
7144 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
7145 spin_unlock(&oi->ip_lock);
7146
Darrick J. Wong2931cdc2014-04-03 14:46:48 -07007147 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh5b6a3a22007-09-13 16:33:54 -07007148 ocfs2_dinode_new_extent_list(inode, di);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007149
7150 ocfs2_journal_dirty(handle, di_bh);
7151
7152 if (has_data) {
7153 /*
7154 * An error at this point should be extremely rare. If
7155 * this proves to be false, we could always re-build
7156 * the in-inode data from our pages.
7157 */
Joel Becker5e404e92009-02-13 03:54:22 -08007158 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
Joel Beckercc79d8c2009-02-13 03:24:43 -08007159 ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007160 if (ret) {
7161 mlog_errno(ret);
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007162 need_free = 1;
alex chen15eba0f2015-02-10 14:09:02 -08007163 goto out_unlock;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007164 }
7165
7166 inode->i_blocks = ocfs2_inode_sector_count(inode);
7167 }
7168
alex chen15eba0f2015-02-10 14:09:02 -08007169out_unlock:
Jan Kara5314454e2021-10-18 15:15:39 -07007170 if (page)
7171 ocfs2_unlock_and_free_pages(&page, num_pages);
alex chen15eba0f2015-02-10 14:09:02 -08007172
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007173out_commit:
Jan Karaa90714c2008-10-09 19:38:40 +02007174 if (ret < 0 && did_quota)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05007175 dquot_free_space_nodirty(inode,
Jan Karaa90714c2008-10-09 19:38:40 +02007176 ocfs2_clusters_to_bytes(osb->sb, 1));
7177
Zongxun Wangfb951eb2014-02-06 12:04:20 -08007178 if (need_free) {
7179 if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
7180 ocfs2_free_local_alloc_bits(osb, handle, data_ac,
7181 bit_off, num);
7182 else
7183 ocfs2_free_clusters(handle,
7184 data_ac->ac_inode,
7185 data_ac->ac_bh,
7186 ocfs2_clusters_to_blocks(osb->sb, bit_off),
7187 num);
7188 }
7189
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007190 ocfs2_commit_trans(osb, handle);
7191
alex chen15eba0f2015-02-10 14:09:02 -08007192out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007193 if (data_ac)
7194 ocfs2_free_alloc_context(data_ac);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007195 return ret;
7196}
7197
Mark Fashehccd979b2005-12-15 14:31:24 -08007198/*
7199 * It is expected, that by the time you call this function,
7200 * inode->i_size and fe->i_size have been adjusted.
7201 *
7202 * WARNING: This will kfree the truncate context
7203 */
7204int ocfs2_commit_truncate(struct ocfs2_super *osb,
7205 struct inode *inode,
Tristan Ye78f94672010-05-11 17:54:42 +08007206 struct buffer_head *di_bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08007207{
Tristan Ye78f94672010-05-11 17:54:42 +08007208 int status = 0, i, flags = 0;
7209 u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
Tao Mabcbbb242009-08-18 11:29:12 +08007210 u64 blkno = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08007211 struct ocfs2_extent_list *el;
Tristan Ye78f94672010-05-11 17:54:42 +08007212 struct ocfs2_extent_rec *rec;
Mark Fashehdcd05382007-01-16 11:32:23 -08007213 struct ocfs2_path *path = NULL;
Tristan Ye78f94672010-05-11 17:54:42 +08007214 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7215 struct ocfs2_extent_list *root_el = &(di->id2.i_list);
7216 u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
7217 struct ocfs2_extent_tree et;
7218 struct ocfs2_cached_dealloc_ctxt dealloc;
Junxiao Bif62f12b2014-12-18 16:17:32 -08007219 struct ocfs2_refcount_tree *ref_tree = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08007220
Tristan Ye78f94672010-05-11 17:54:42 +08007221 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7222 ocfs2_init_dealloc_ctxt(&dealloc);
7223
Mark Fashehdcd05382007-01-16 11:32:23 -08007224 new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
Mark Fashehccd979b2005-12-15 14:31:24 -08007225 i_size_read(inode));
7226
Tristan Ye78f94672010-05-11 17:54:42 +08007227 path = ocfs2_new_path(di_bh, &di->id2.i_list,
Joel Becker13723d02008-10-17 19:25:01 -07007228 ocfs2_journal_access_di);
Mark Fashehdcd05382007-01-16 11:32:23 -08007229 if (!path) {
7230 status = -ENOMEM;
7231 mlog_errno(status);
7232 goto bail;
7233 }
Mark Fasheh83418972007-04-23 18:53:12 -07007234
7235 ocfs2_extent_map_trunc(inode, new_highest_cpos);
7236
Mark Fashehccd979b2005-12-15 14:31:24 -08007237start:
Mark Fashehdcd05382007-01-16 11:32:23 -08007238 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007239 * Check that we still have allocation to delete.
7240 */
7241 if (OCFS2_I(inode)->ip_clusters == 0) {
7242 status = 0;
7243 goto bail;
7244 }
7245
7246 /*
Mark Fashehdcd05382007-01-16 11:32:23 -08007247 * Truncate always works against the rightmost tree branch.
7248 */
Joel Beckerfacdb772009-02-12 18:08:48 -08007249 status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
Mark Fashehdcd05382007-01-16 11:32:23 -08007250 if (status) {
7251 mlog_errno(status);
7252 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08007253 }
7254
Tao Maa09d09b2011-02-22 08:21:10 +08007255 trace_ocfs2_commit_truncate(
7256 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7257 new_highest_cpos,
7258 OCFS2_I(inode)->ip_clusters,
7259 path->p_tree_depth);
Mark Fashehdcd05382007-01-16 11:32:23 -08007260
7261 /*
7262 * By now, el will point to the extent list on the bottom most
7263 * portion of this tree. Only the tail record is considered in
7264 * each pass.
7265 *
7266 * We handle the following cases, in order:
7267 * - empty extent: delete the remaining branch
7268 * - remove the entire record
7269 * - remove a partial record
7270 * - no record needs to be removed (truncate has completed)
7271 */
7272 el = path_leaf_el(path);
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007273 if (le16_to_cpu(el->l_next_free_rec) == 0) {
7274 ocfs2_error(inode->i_sb,
7275 "Inode %llu has empty extent block at %llu\n",
7276 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7277 (unsigned long long)path_leaf_bh(path)->b_blocknr);
7278 status = -EROFS;
7279 goto bail;
7280 }
7281
Mark Fashehccd979b2005-12-15 14:31:24 -08007282 i = le16_to_cpu(el->l_next_free_rec) - 1;
Tristan Ye78f94672010-05-11 17:54:42 +08007283 rec = &el->l_recs[i];
7284 flags = rec->e_flags;
7285 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7286
7287 if (i == 0 && ocfs2_is_empty_extent(rec)) {
7288 /*
7289 * Lower levels depend on this never happening, but it's best
7290 * to check it up here before changing the tree.
7291 */
7292 if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
Xue jiufeid0c97d52015-09-04 15:44:48 -07007293 mlog(ML_ERROR, "Inode %lu has an empty "
Tristan Ye78f94672010-05-11 17:54:42 +08007294 "extent record, depth %u\n", inode->i_ino,
7295 le16_to_cpu(root_el->l_tree_depth));
Xue jiufeid0c97d52015-09-04 15:44:48 -07007296 status = ocfs2_remove_rightmost_empty_extent(osb,
7297 &et, path, &dealloc);
7298 if (status) {
7299 mlog_errno(status);
7300 goto bail;
7301 }
7302
7303 ocfs2_reinit_path(path, 1);
7304 goto start;
7305 } else {
7306 trunc_cpos = le32_to_cpu(rec->e_cpos);
7307 trunc_len = 0;
7308 blkno = 0;
Tristan Ye78f94672010-05-11 17:54:42 +08007309 }
Tristan Ye78f94672010-05-11 17:54:42 +08007310 } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7311 /*
7312 * Truncate entire record.
7313 */
7314 trunc_cpos = le32_to_cpu(rec->e_cpos);
7315 trunc_len = ocfs2_rec_clusters(el, rec);
7316 blkno = le64_to_cpu(rec->e_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007317 } else if (range > new_highest_cpos) {
Tristan Ye78f94672010-05-11 17:54:42 +08007318 /*
7319 * Partial truncate. it also should be
7320 * the last truncate we're doing.
7321 */
7322 trunc_cpos = new_highest_cpos;
7323 trunc_len = range - new_highest_cpos;
7324 coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7325 blkno = le64_to_cpu(rec->e_blkno) +
7326 ocfs2_clusters_to_blocks(inode->i_sb, coff);
Mark Fashehdcd05382007-01-16 11:32:23 -08007327 } else {
Tristan Ye78f94672010-05-11 17:54:42 +08007328 /*
7329 * Truncate completed, leave happily.
7330 */
Mark Fashehdcd05382007-01-16 11:32:23 -08007331 status = 0;
7332 goto bail;
7333 }
Mark Fashehccd979b2005-12-15 14:31:24 -08007334
Tristan Ye78f94672010-05-11 17:54:42 +08007335 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -08007336
Junxiao Bif62f12b2014-12-18 16:17:32 -08007337 if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) {
7338 status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
7339 &ref_tree, NULL);
7340 if (status) {
7341 mlog_errno(status);
7342 goto bail;
7343 }
7344 }
7345
Tristan Ye78f94672010-05-11 17:54:42 +08007346 status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7347 phys_cpos, trunc_len, flags, &dealloc,
Junxiao Bif62f12b2014-12-18 16:17:32 -08007348 refcount_loc, true);
Mark Fashehccd979b2005-12-15 14:31:24 -08007349 if (status < 0) {
7350 mlog_errno(status);
7351 goto bail;
7352 }
7353
Mark Fashehdcd05382007-01-16 11:32:23 -08007354 ocfs2_reinit_path(path, 1);
7355
7356 /*
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007357 * The check above will catch the case where we've truncated
7358 * away all allocation.
Mark Fashehdcd05382007-01-16 11:32:23 -08007359 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08007360 goto start;
7361
Mark Fashehccd979b2005-12-15 14:31:24 -08007362bail:
Junxiao Bif62f12b2014-12-18 16:17:32 -08007363 if (ref_tree)
7364 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08007365
7366 ocfs2_schedule_truncate_log_flush(osb, 1);
7367
Tristan Ye78f94672010-05-11 17:54:42 +08007368 ocfs2_run_deallocs(osb, &dealloc);
Mark Fasheh59a5e412007-06-22 15:52:36 -07007369
Mark Fashehdcd05382007-01-16 11:32:23 -08007370 ocfs2_free_path(path);
Mark Fashehccd979b2005-12-15 14:31:24 -08007371
Mark Fashehccd979b2005-12-15 14:31:24 -08007372 return status;
7373}
7374
Mark Fashehccd979b2005-12-15 14:31:24 -08007375/*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007376 * 'start' is inclusive, 'end' is not.
7377 */
7378int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7379 unsigned int start, unsigned int end, int trunc)
7380{
7381 int ret;
7382 unsigned int numbytes;
7383 handle_t *handle;
7384 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7385 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7386 struct ocfs2_inline_data *idata = &di->id2.i_data;
7387
Changwei Ge783fda852020-04-10 14:32:38 -07007388 /* No need to punch hole beyond i_size. */
7389 if (start >= i_size_read(inode))
7390 return 0;
7391
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007392 if (end > i_size_read(inode))
7393 end = i_size_read(inode);
7394
Younger Liud62e74b2014-02-10 14:25:51 -08007395 BUG_ON(start > end);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007396
7397 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7398 !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7399 !ocfs2_supports_inline_data(osb)) {
7400 ocfs2_error(inode->i_sb,
Joe Perches7ecef142015-09-04 15:44:51 -07007401 "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 -07007402 (unsigned long long)OCFS2_I(inode)->ip_blkno,
7403 le16_to_cpu(di->i_dyn_features),
7404 OCFS2_I(inode)->ip_dyn_features,
7405 osb->s_feature_incompat);
7406 ret = -EROFS;
7407 goto out;
7408 }
7409
7410 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7411 if (IS_ERR(handle)) {
7412 ret = PTR_ERR(handle);
7413 mlog_errno(ret);
7414 goto out;
7415 }
7416
Joel Becker0cf2f762009-02-12 16:41:25 -08007417 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -07007418 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007419 if (ret) {
7420 mlog_errno(ret);
7421 goto out_commit;
7422 }
7423
7424 numbytes = end - start;
7425 memset(idata->id_data + start, 0, numbytes);
7426
7427 /*
7428 * No need to worry about the data page here - it's been
7429 * truncated already and inline data doesn't need it for
7430 * pushing zero's to disk, so we'll let readpage pick it up
7431 * later.
7432 */
7433 if (trunc) {
7434 i_size_write(inode, start);
7435 di->i_size = cpu_to_le64(start);
7436 }
7437
7438 inode->i_blocks = ocfs2_inode_sector_count(inode);
Deepa Dinamani078cd822016-09-14 07:48:04 -07007439 inode->i_ctime = inode->i_mtime = current_time(inode);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007440
7441 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7442 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7443
Darrick J. Wong6fdb7022014-04-03 14:47:08 -07007444 ocfs2_update_inode_fsync_trans(handle, inode, 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07007445 ocfs2_journal_dirty(handle, di_bh);
7446
7447out_commit:
7448 ocfs2_commit_trans(osb, handle);
7449
7450out:
7451 return ret;
7452}
Tao Mae80de362011-05-23 10:36:43 +08007453
7454static int ocfs2_trim_extent(struct super_block *sb,
7455 struct ocfs2_group_desc *gd,
Ashish Samant105ddc92017-11-02 15:59:37 -07007456 u64 group, u32 start, u32 count)
Tao Mae80de362011-05-23 10:36:43 +08007457{
7458 u64 discard, bcount;
Ashish Samant105ddc92017-11-02 15:59:37 -07007459 struct ocfs2_super *osb = OCFS2_SB(sb);
Tao Mae80de362011-05-23 10:36:43 +08007460
7461 bcount = ocfs2_clusters_to_blocks(sb, count);
Ashish Samant105ddc92017-11-02 15:59:37 -07007462 discard = ocfs2_clusters_to_blocks(sb, start);
7463
7464 /*
7465 * For the first cluster group, the gd->bg_blkno is not at the start
7466 * of the group, but at an offset from the start. If we add it while
7467 * calculating discard for first group, we will wrongly start fstrim a
7468 * few blocks after the desried start block and the range can cross
7469 * over into the next cluster group. So, add it only if this is not
7470 * the first cluster group.
7471 */
7472 if (group != osb->first_cluster_group_blkno)
7473 discard += le64_to_cpu(gd->bg_blkno);
Tao Mae80de362011-05-23 10:36:43 +08007474
Tao Ma10fca352011-05-23 10:36:45 +08007475 trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7476
Tao Mae80de362011-05-23 10:36:43 +08007477 return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7478}
7479
7480static int ocfs2_trim_group(struct super_block *sb,
Ashish Samant105ddc92017-11-02 15:59:37 -07007481 struct ocfs2_group_desc *gd, u64 group,
Tao Mae80de362011-05-23 10:36:43 +08007482 u32 start, u32 max, u32 minbits)
7483{
7484 int ret = 0, count = 0, next;
7485 void *bitmap = gd->bg_bitmap;
7486
7487 if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7488 return 0;
7489
Tao Ma10fca352011-05-23 10:36:45 +08007490 trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7491 start, max, minbits);
7492
Tao Mae80de362011-05-23 10:36:43 +08007493 while (start < max) {
7494 start = ocfs2_find_next_zero_bit(bitmap, max, start);
7495 if (start >= max)
7496 break;
7497 next = ocfs2_find_next_bit(bitmap, max, start);
7498
7499 if ((next - start) >= minbits) {
Ashish Samant105ddc92017-11-02 15:59:37 -07007500 ret = ocfs2_trim_extent(sb, gd, group,
Tao Mae80de362011-05-23 10:36:43 +08007501 start, next - start);
7502 if (ret < 0) {
7503 mlog_errno(ret);
7504 break;
7505 }
7506 count += next - start;
7507 }
7508 start = next + 1;
7509
7510 if (fatal_signal_pending(current)) {
7511 count = -ERESTARTSYS;
7512 break;
7513 }
7514
7515 if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7516 break;
7517 }
7518
7519 if (ret < 0)
7520 count = ret;
7521
7522 return count;
7523}
7524
Gang He5500ab42019-03-05 15:41:45 -08007525static
7526int ocfs2_trim_mainbm(struct super_block *sb, struct fstrim_range *range)
Tao Mae80de362011-05-23 10:36:43 +08007527{
7528 struct ocfs2_super *osb = OCFS2_SB(sb);
Gang He5500ab42019-03-05 15:41:45 -08007529 u64 start, len, trimmed = 0, first_group, last_group = 0, group = 0;
Tao Mae80de362011-05-23 10:36:43 +08007530 int ret, cnt;
7531 u32 first_bit, last_bit, minlen;
7532 struct buffer_head *main_bm_bh = NULL;
7533 struct inode *main_bm_inode = NULL;
7534 struct buffer_head *gd_bh = NULL;
7535 struct ocfs2_dinode *main_bm;
7536 struct ocfs2_group_desc *gd = NULL;
7537
7538 start = range->start >> osb->s_clustersize_bits;
7539 len = range->len >> osb->s_clustersize_bits;
7540 minlen = range->minlen >> osb->s_clustersize_bits;
Tao Mae80de362011-05-23 10:36:43 +08007541
Jie Liuaa897622014-01-21 15:48:35 -08007542 if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
Tao Mae80de362011-05-23 10:36:43 +08007543 return -EINVAL;
7544
Gang He5500ab42019-03-05 15:41:45 -08007545 trace_ocfs2_trim_mainbm(start, len, minlen);
7546
7547next_group:
Tao Mae80de362011-05-23 10:36:43 +08007548 main_bm_inode = ocfs2_get_system_file_inode(osb,
7549 GLOBAL_BITMAP_SYSTEM_INODE,
7550 OCFS2_INVALID_SLOT);
7551 if (!main_bm_inode) {
7552 ret = -EIO;
7553 mlog_errno(ret);
7554 goto out;
7555 }
7556
Al Viro59551022016-01-22 15:40:57 -05007557 inode_lock(main_bm_inode);
Tao Mae80de362011-05-23 10:36:43 +08007558
7559 ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7560 if (ret < 0) {
7561 mlog_errno(ret);
7562 goto out_mutex;
7563 }
7564 main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7565
Gang He5500ab42019-03-05 15:41:45 -08007566 /*
7567 * Do some check before trim the first group.
7568 */
7569 if (!group) {
7570 if (start >= le32_to_cpu(main_bm->i_clusters)) {
7571 ret = -EINVAL;
Gang He637dd202018-01-31 16:15:13 -08007572 goto out_unlock;
7573 }
7574
Gang He5500ab42019-03-05 15:41:45 -08007575 if (start + len > le32_to_cpu(main_bm->i_clusters))
7576 len = le32_to_cpu(main_bm->i_clusters) - start;
Gang He637dd202018-01-31 16:15:13 -08007577
Gang He5500ab42019-03-05 15:41:45 -08007578 /*
7579 * Determine first and last group to examine based on
7580 * start and len
7581 */
7582 first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7583 if (first_group == osb->first_cluster_group_blkno)
7584 first_bit = start;
7585 else
7586 first_bit = start - ocfs2_blocks_to_clusters(sb,
7587 first_group);
7588 last_group = ocfs2_which_cluster_group(main_bm_inode,
7589 start + len - 1);
7590 group = first_group;
Gang He637dd202018-01-31 16:15:13 -08007591 }
7592
Gang He5500ab42019-03-05 15:41:45 -08007593 do {
Tao Mae80de362011-05-23 10:36:43 +08007594 if (first_bit + len >= osb->bitmap_cpg)
7595 last_bit = osb->bitmap_cpg;
7596 else
7597 last_bit = first_bit + len;
7598
7599 ret = ocfs2_read_group_descriptor(main_bm_inode,
7600 main_bm, group,
7601 &gd_bh);
7602 if (ret < 0) {
7603 mlog_errno(ret);
7604 break;
7605 }
7606
7607 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
Ashish Samant105ddc92017-11-02 15:59:37 -07007608 cnt = ocfs2_trim_group(sb, gd, group,
7609 first_bit, last_bit, minlen);
Tao Mae80de362011-05-23 10:36:43 +08007610 brelse(gd_bh);
7611 gd_bh = NULL;
7612 if (cnt < 0) {
7613 ret = cnt;
7614 mlog_errno(ret);
7615 break;
7616 }
7617
7618 trimmed += cnt;
7619 len -= osb->bitmap_cpg - first_bit;
7620 first_bit = 0;
7621 if (group == osb->first_cluster_group_blkno)
7622 group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7623 else
7624 group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
Gang He5500ab42019-03-05 15:41:45 -08007625 } while (0);
Gang He637dd202018-01-31 16:15:13 -08007626
Tao Mae80de362011-05-23 10:36:43 +08007627out_unlock:
7628 ocfs2_inode_unlock(main_bm_inode, 0);
7629 brelse(main_bm_bh);
Gang He5500ab42019-03-05 15:41:45 -08007630 main_bm_bh = NULL;
Tao Mae80de362011-05-23 10:36:43 +08007631out_mutex:
Al Viro59551022016-01-22 15:40:57 -05007632 inode_unlock(main_bm_inode);
Tao Mae80de362011-05-23 10:36:43 +08007633 iput(main_bm_inode);
Gang He5500ab42019-03-05 15:41:45 -08007634
7635 /*
7636 * If all the groups trim are not done or failed, but we should release
7637 * main_bm related locks for avoiding the current IO starve, then go to
7638 * trim the next group
7639 */
Gang He8dd71b22020-10-13 16:48:24 -07007640 if (ret >= 0 && group <= last_group) {
7641 cond_resched();
Gang He5500ab42019-03-05 15:41:45 -08007642 goto next_group;
Gang He8dd71b22020-10-13 16:48:24 -07007643 }
Tao Mae80de362011-05-23 10:36:43 +08007644out:
Gang He5500ab42019-03-05 15:41:45 -08007645 range->len = trimmed * sb->s_blocksize;
7646 return ret;
7647}
7648
7649int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7650{
7651 int ret;
7652 struct ocfs2_super *osb = OCFS2_SB(sb);
7653 struct ocfs2_trim_fs_info info, *pinfo = NULL;
7654
7655 ocfs2_trim_fs_lock_res_init(osb);
7656
7657 trace_ocfs2_trim_fs(range->start, range->len, range->minlen);
7658
7659 ret = ocfs2_trim_fs_lock(osb, NULL, 1);
7660 if (ret < 0) {
7661 if (ret != -EAGAIN) {
7662 mlog_errno(ret);
7663 ocfs2_trim_fs_lock_res_uninit(osb);
7664 return ret;
7665 }
7666
7667 mlog(ML_NOTICE, "Wait for trim on device (%s) to "
7668 "finish, which is running from another node.\n",
7669 osb->dev_str);
7670 ret = ocfs2_trim_fs_lock(osb, &info, 0);
7671 if (ret < 0) {
7672 mlog_errno(ret);
7673 ocfs2_trim_fs_lock_res_uninit(osb);
7674 return ret;
7675 }
7676
7677 if (info.tf_valid && info.tf_success &&
7678 info.tf_start == range->start &&
7679 info.tf_len == range->len &&
7680 info.tf_minlen == range->minlen) {
7681 /* Avoid sending duplicated trim to a shared device */
7682 mlog(ML_NOTICE, "The same trim on device (%s) was "
7683 "just done from node (%u), return.\n",
7684 osb->dev_str, info.tf_nodenum);
7685 range->len = info.tf_trimlen;
7686 goto out;
7687 }
7688 }
7689
7690 info.tf_nodenum = osb->node_num;
7691 info.tf_start = range->start;
7692 info.tf_len = range->len;
7693 info.tf_minlen = range->minlen;
7694
7695 ret = ocfs2_trim_mainbm(sb, range);
7696
7697 info.tf_trimlen = range->len;
7698 info.tf_success = (ret < 0 ? 0 : 1);
7699 pinfo = &info;
7700out:
7701 ocfs2_trim_fs_unlock(osb, pinfo);
7702 ocfs2_trim_fs_lock_res_uninit(osb);
Tao Mae80de362011-05-23 10:36:43 +08007703 return ret;
7704}