blob: 8efb2f0a344763a2421204e935af1e09897239be [file] [log] [blame]
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001/*
2 * linux/fs/ext4/block_validity.c
3 *
4 * Copyright (C) 2009
5 * Theodore Ts'o (tytso@mit.edu)
6 *
7 * Track which blocks in the filesystem are metadata blocks that
8 * should never be used as data blocks by files or directories.
9 */
10
11#include <linux/time.h>
12#include <linux/fs.h>
13#include <linux/namei.h>
14#include <linux/quotaops.h>
15#include <linux/buffer_head.h>
16#include <linux/module.h>
17#include <linux/swap.h>
18#include <linux/pagemap.h>
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040019#include <linux/blkdev.h>
20#include <linux/mutex.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/slab.h>
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040022#include "ext4.h"
23
24struct ext4_system_zone {
25 struct rb_node node;
26 ext4_fsblk_t start_blk;
27 unsigned int count;
28};
29
30static struct kmem_cache *ext4_system_zone_cachep;
31
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040032int __init ext4_init_system_zone(void)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040033{
Theodore Ts'o16828082010-10-27 21:30:09 -040034 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040035 if (ext4_system_zone_cachep == NULL)
36 return -ENOMEM;
37 return 0;
38}
39
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040040void ext4_exit_system_zone(void)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040041{
42 kmem_cache_destroy(ext4_system_zone_cachep);
43}
44
45static inline int can_merge(struct ext4_system_zone *entry1,
46 struct ext4_system_zone *entry2)
47{
48 if ((entry1->start_blk + entry1->count) == entry2->start_blk)
49 return 1;
50 return 0;
51}
52
53/*
54 * Mark a range of blocks as belonging to the "system zone" --- that
55 * is, filesystem metadata blocks which should never be used by
56 * inodes.
57 */
58static int add_system_zone(struct ext4_sb_info *sbi,
59 ext4_fsblk_t start_blk,
60 unsigned int count)
61{
62 struct ext4_system_zone *new_entry = NULL, *entry;
63 struct rb_node **n = &sbi->system_blks.rb_node, *node;
64 struct rb_node *parent = NULL, *new_node = NULL;
65
66 while (*n) {
67 parent = *n;
68 entry = rb_entry(parent, struct ext4_system_zone, node);
69 if (start_blk < entry->start_blk)
70 n = &(*n)->rb_left;
71 else if (start_blk >= (entry->start_blk + entry->count))
72 n = &(*n)->rb_right;
73 else {
Theodore Ts'o60e66792010-05-17 07:00:00 -040074 if (start_blk + count > (entry->start_blk +
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040075 entry->count))
Theodore Ts'o60e66792010-05-17 07:00:00 -040076 entry->count = (start_blk + count -
Theodore Ts'o6fd058f2009-05-17 15:38:01 -040077 entry->start_blk);
78 new_node = *n;
79 new_entry = rb_entry(new_node, struct ext4_system_zone,
80 node);
81 break;
82 }
83 }
84
85 if (!new_entry) {
86 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
87 GFP_KERNEL);
88 if (!new_entry)
89 return -ENOMEM;
90 new_entry->start_blk = start_blk;
91 new_entry->count = count;
92 new_node = &new_entry->node;
93
94 rb_link_node(new_node, parent, n);
95 rb_insert_color(new_node, &sbi->system_blks);
96 }
97
98 /* Can we merge to the left? */
99 node = rb_prev(new_node);
100 if (node) {
101 entry = rb_entry(node, struct ext4_system_zone, node);
102 if (can_merge(entry, new_entry)) {
103 new_entry->start_blk = entry->start_blk;
104 new_entry->count += entry->count;
105 rb_erase(node, &sbi->system_blks);
106 kmem_cache_free(ext4_system_zone_cachep, entry);
107 }
108 }
109
110 /* Can we merge to the right? */
111 node = rb_next(new_node);
112 if (node) {
113 entry = rb_entry(node, struct ext4_system_zone, node);
114 if (can_merge(new_entry, entry)) {
115 new_entry->count += entry->count;
116 rb_erase(node, &sbi->system_blks);
117 kmem_cache_free(ext4_system_zone_cachep, entry);
118 }
119 }
120 return 0;
121}
122
123static void debug_print_tree(struct ext4_sb_info *sbi)
124{
125 struct rb_node *node;
126 struct ext4_system_zone *entry;
127 int first = 1;
128
129 printk(KERN_INFO "System zones: ");
130 node = rb_first(&sbi->system_blks);
131 while (node) {
132 entry = rb_entry(node, struct ext4_system_zone, node);
133 printk("%s%llu-%llu", first ? "" : ", ",
134 entry->start_blk, entry->start_blk + entry->count - 1);
135 first = 0;
136 node = rb_next(node);
137 }
138 printk("\n");
139}
140
141int ext4_setup_system_zone(struct super_block *sb)
142{
143 ext4_group_t ngroups = ext4_get_groups_count(sb);
144 struct ext4_sb_info *sbi = EXT4_SB(sb);
145 struct ext4_group_desc *gdp;
146 ext4_group_t i;
147 int flex_size = ext4_flex_bg_size(sbi);
148 int ret;
149
150 if (!test_opt(sb, BLOCK_VALIDITY)) {
151 if (EXT4_SB(sb)->system_blks.rb_node)
152 ext4_release_system_zone(sb);
153 return 0;
154 }
155 if (EXT4_SB(sb)->system_blks.rb_node)
156 return 0;
157
158 for (i=0; i < ngroups; i++) {
159 if (ext4_bg_has_super(sb, i) &&
160 ((i < 5) || ((i % flex_size) == 0)))
161 add_system_zone(sbi, ext4_group_first_block_no(sb, i),
Theodore Ts'o10329882009-11-15 15:29:56 -0500162 ext4_bg_num_gdb(sb, i) + 1);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400163 gdp = ext4_get_group_desc(sb, i, NULL);
164 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1);
165 if (ret)
166 return ret;
167 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1);
168 if (ret)
169 return ret;
170 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
171 sbi->s_itb_per_group);
172 if (ret)
173 return ret;
174 }
175
176 if (test_opt(sb, DEBUG))
177 debug_print_tree(EXT4_SB(sb));
178 return 0;
179}
180
181/* Called when the filesystem is unmounted */
182void ext4_release_system_zone(struct super_block *sb)
183{
184 struct rb_node *n = EXT4_SB(sb)->system_blks.rb_node;
185 struct rb_node *parent;
186 struct ext4_system_zone *entry;
187
188 while (n) {
189 /* Do the node's children first */
190 if (n->rb_left) {
191 n = n->rb_left;
192 continue;
193 }
194 if (n->rb_right) {
195 n = n->rb_right;
196 continue;
197 }
198 /*
199 * The node has no children; free it, and then zero
200 * out parent's link to it. Finally go to the
201 * beginning of the loop and try to free the parent
202 * node.
203 */
204 parent = rb_parent(n);
205 entry = rb_entry(n, struct ext4_system_zone, node);
206 kmem_cache_free(ext4_system_zone_cachep, entry);
207 if (!parent)
Venkatesh Pallipadi64e290e2010-03-04 22:25:21 -0500208 EXT4_SB(sb)->system_blks = RB_ROOT;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400209 else if (parent->rb_left == n)
210 parent->rb_left = NULL;
211 else if (parent->rb_right == n)
212 parent->rb_right = NULL;
213 n = parent;
214 }
Venkatesh Pallipadi64e290e2010-03-04 22:25:21 -0500215 EXT4_SB(sb)->system_blks = RB_ROOT;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400216}
217
218/*
219 * Returns 1 if the passed-in block region (start_blk,
220 * start_blk+count) is valid; 0 if some part of the block region
221 * overlaps with filesystem metadata blocks.
222 */
223int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
224 unsigned int count)
225{
226 struct ext4_system_zone *entry;
227 struct rb_node *n = sbi->system_blks.rb_node;
228
229 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
Theodore Ts'o1585d8d2009-11-22 20:48:34 -0500230 (start_blk + count < start_blk) ||
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400231 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
232 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400233 return 0;
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400234 }
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400235 while (n) {
236 entry = rb_entry(n, struct ext4_system_zone, node);
237 if (start_blk + count - 1 < entry->start_blk)
238 n = n->rb_left;
239 else if (start_blk >= (entry->start_blk + entry->count))
240 n = n->rb_right;
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400241 else {
242 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400243 return 0;
Theodore Ts'o1c13d5c2010-07-27 11:56:03 -0400244 }
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400245 }
246 return 1;
247}
248
Theodore Ts'o1f7d1e72011-06-27 19:16:02 -0400249int ext4_check_blockref(const char *function, unsigned int line,
250 struct inode *inode, __le32 *p, unsigned int max)
251{
252 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
253 __le32 *bref = p;
254 unsigned int blk;
255
256 while (bref < p+max) {
257 blk = le32_to_cpu(*bref++);
258 if (blk &&
259 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
260 blk, 1))) {
261 es->s_last_error_block = cpu_to_le64(blk);
262 ext4_error_inode(inode, function, line, blk,
263 "invalid block");
264 return -EIO;
265 }
266 }
267 return 0;
268}
Amir Goldsteindae1e522011-06-27 19:40:50 -0400269