Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ac27a0ec | 2 | /* |
617ba13b | 3 | * linux/fs/ext4/resize.c |
ac27a0ec | 4 | * |
617ba13b | 5 | * Support for resizing an ext4 filesystem while it is mounted. |
ac27a0ec DK |
6 | * |
7 | * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com> | |
8 | * | |
9 | * This could probably be made into a module, because it is not often in use. | |
10 | */ | |
11 | ||
12 | ||
ac27a0ec DK |
13 | #include <linux/errno.h> |
14 | #include <linux/slab.h> | |
a861fb9f | 15 | #include <linux/jiffies.h> |
ac27a0ec | 16 | |
3dcf5451 | 17 | #include "ext4_jbd2.h" |
ac27a0ec | 18 | |
1d0c3924 TT |
19 | struct ext4_rcu_ptr { |
20 | struct rcu_head rcu; | |
21 | void *ptr; | |
22 | }; | |
23 | ||
24 | static void ext4_rcu_ptr_callback(struct rcu_head *head) | |
25 | { | |
26 | struct ext4_rcu_ptr *ptr; | |
27 | ||
28 | ptr = container_of(head, struct ext4_rcu_ptr, rcu); | |
29 | kvfree(ptr->ptr); | |
30 | kfree(ptr); | |
31 | } | |
32 | ||
33 | void ext4_kvfree_array_rcu(void *to_free) | |
34 | { | |
35 | struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL); | |
36 | ||
37 | if (ptr) { | |
38 | ptr->ptr = to_free; | |
39 | call_rcu(&ptr->rcu, ext4_rcu_ptr_callback); | |
40 | return; | |
41 | } | |
42 | synchronize_rcu(); | |
43 | kvfree(to_free); | |
44 | } | |
45 | ||
8f82f840 YY |
46 | int ext4_resize_begin(struct super_block *sb) |
47 | { | |
5f8c1093 | 48 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8f82f840 YY |
49 | int ret = 0; |
50 | ||
51 | if (!capable(CAP_SYS_RESOURCE)) | |
52 | return -EPERM; | |
53 | ||
b55c3cd1 ZY |
54 | /* |
55 | * If the reserved GDT blocks is non-zero, the resize_inode feature | |
56 | * should always be set. | |
57 | */ | |
70cbfd25 | 58 | if (sbi->s_es->s_reserved_gdt_blocks && |
b55c3cd1 ZY |
59 | !ext4_has_feature_resize_inode(sb)) { |
60 | ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero"); | |
61 | return -EFSCORRUPTED; | |
62 | } | |
63 | ||
011fa994 TT |
64 | /* |
65 | * If we are not using the primary superblock/GDT copy don't resize, | |
66 | * because the user tools have no way of handling this. Probably a | |
67 | * bad time to do it anyways. | |
68 | */ | |
5f8c1093 | 69 | if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) != |
70cbfd25 | 70 | le32_to_cpu(sbi->s_es->s_first_data_block)) { |
011fa994 | 71 | ext4_warning(sb, "won't resize using backup superblock at %llu", |
70cbfd25 | 72 | (unsigned long long)sbi->s_sbh->b_blocknr); |
011fa994 TT |
73 | return -EPERM; |
74 | } | |
75 | ||
ce723c31 YY |
76 | /* |
77 | * We are not allowed to do online-resizing on a filesystem mounted | |
78 | * with error, because it can destroy the filesystem easily. | |
79 | */ | |
70cbfd25 | 80 | if (sbi->s_mount_state & EXT4_ERROR_FS) { |
ce723c31 | 81 | ext4_warning(sb, "There are errors in the filesystem, " |
8d2ae1cb | 82 | "so online resizing is not allowed"); |
ce723c31 YY |
83 | return -EPERM; |
84 | } | |
85 | ||
b1489186 JT |
86 | if (ext4_has_feature_sparse_super2(sb)) { |
87 | ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2"); | |
88 | return -EOPNOTSUPP; | |
89 | } | |
e9f9f61d | 90 | |
9549a168 | 91 | if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING, |
70cbfd25 | 92 | &sbi->s_ext4_flags)) |
8f82f840 YY |
93 | ret = -EBUSY; |
94 | ||
95 | return ret; | |
96 | } | |
97 | ||
827891a3 | 98 | int ext4_resize_end(struct super_block *sb, bool update_backups) |
8f82f840 | 99 | { |
9549a168 | 100 | clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags); |
4e857c58 | 101 | smp_mb__after_atomic(); |
827891a3 TT |
102 | if (update_backups) |
103 | return ext4_update_overhead(sb, true); | |
104 | return 0; | |
8f82f840 YY |
105 | } |
106 | ||
01f795f9 YY |
107 | static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb, |
108 | ext4_group_t group) { | |
109 | ext4_grpblk_t overhead; | |
110 | overhead = ext4_bg_num_gdb(sb, group); | |
111 | if (ext4_bg_has_super(sb, group)) | |
112 | overhead += 1 + | |
113 | le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); | |
114 | return overhead; | |
115 | } | |
116 | ||
ac27a0ec DK |
117 | #define outside(b, first, last) ((b) < (first) || (b) >= (last)) |
118 | #define inside(b, first, last) ((b) >= (first) && (b) < (last)) | |
119 | ||
120 | static int verify_group_input(struct super_block *sb, | |
617ba13b | 121 | struct ext4_new_group_data *input) |
ac27a0ec | 122 | { |
617ba13b MC |
123 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
124 | struct ext4_super_block *es = sbi->s_es; | |
bd81d8ee | 125 | ext4_fsblk_t start = ext4_blocks_count(es); |
617ba13b | 126 | ext4_fsblk_t end = start + input->blocks_count; |
fd2d4291 | 127 | ext4_group_t group = input->group; |
617ba13b | 128 | ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group; |
b302ef2d TT |
129 | unsigned overhead; |
130 | ext4_fsblk_t metaend; | |
ac27a0ec | 131 | struct buffer_head *bh = NULL; |
3a5b2ecd | 132 | ext4_grpblk_t free_blocks_count, offset; |
ac27a0ec DK |
133 | int err = -EINVAL; |
134 | ||
b302ef2d TT |
135 | if (group != sbi->s_groups_count) { |
136 | ext4_warning(sb, "Cannot add at group %u (only %u groups)", | |
137 | input->group, sbi->s_groups_count); | |
138 | return -EINVAL; | |
139 | } | |
140 | ||
141 | overhead = ext4_group_overhead_blocks(sb, group); | |
142 | metaend = start + overhead; | |
1fc1bd2d KS |
143 | free_blocks_count = input->blocks_count - 2 - overhead - |
144 | sbi->s_itb_per_group; | |
145 | input->free_clusters_count = EXT4_B2C(sbi, free_blocks_count); | |
ac27a0ec DK |
146 | |
147 | if (test_opt(sb, DEBUG)) | |
617ba13b | 148 | printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks " |
ac27a0ec | 149 | "(%d free, %u reserved)\n", |
617ba13b | 150 | ext4_bg_has_super(sb, input->group) ? "normal" : |
ac27a0ec DK |
151 | "no-super", input->group, input->blocks_count, |
152 | free_blocks_count, input->reserved_blocks); | |
153 | ||
3a5b2ecd | 154 | ext4_get_group_no_and_offset(sb, start, NULL, &offset); |
b302ef2d | 155 | if (offset != 0) |
12062ddd | 156 | ext4_warning(sb, "Last group not full"); |
ac27a0ec | 157 | else if (input->reserved_blocks > input->blocks_count / 5) |
12062ddd | 158 | ext4_warning(sb, "Reserved blocks too high (%u)", |
ac27a0ec DK |
159 | input->reserved_blocks); |
160 | else if (free_blocks_count < 0) | |
12062ddd | 161 | ext4_warning(sb, "Bad blocks count %u", |
ac27a0ec | 162 | input->blocks_count); |
fb265c9c TT |
163 | else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) { |
164 | err = PTR_ERR(bh); | |
165 | bh = NULL; | |
12062ddd | 166 | ext4_warning(sb, "Cannot read last block (%llu)", |
ac27a0ec | 167 | end - 1); |
fb265c9c | 168 | } else if (outside(input->block_bitmap, start, end)) |
12062ddd | 169 | ext4_warning(sb, "Block bitmap not in group (block %llu)", |
1939e49a | 170 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 171 | else if (outside(input->inode_bitmap, start, end)) |
12062ddd | 172 | ext4_warning(sb, "Inode bitmap not in group (block %llu)", |
1939e49a | 173 | (unsigned long long)input->inode_bitmap); |
ac27a0ec | 174 | else if (outside(input->inode_table, start, end) || |
2b2d6d01 | 175 | outside(itend - 1, start, end)) |
12062ddd | 176 | ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)", |
1939e49a | 177 | (unsigned long long)input->inode_table, itend - 1); |
ac27a0ec | 178 | else if (input->inode_bitmap == input->block_bitmap) |
12062ddd | 179 | ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)", |
1939e49a | 180 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 181 | else if (inside(input->block_bitmap, input->inode_table, itend)) |
12062ddd ES |
182 | ext4_warning(sb, "Block bitmap (%llu) in inode table " |
183 | "(%llu-%llu)", | |
1939e49a RD |
184 | (unsigned long long)input->block_bitmap, |
185 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 186 | else if (inside(input->inode_bitmap, input->inode_table, itend)) |
12062ddd ES |
187 | ext4_warning(sb, "Inode bitmap (%llu) in inode table " |
188 | "(%llu-%llu)", | |
1939e49a RD |
189 | (unsigned long long)input->inode_bitmap, |
190 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 191 | else if (inside(input->block_bitmap, start, metaend)) |
12062ddd | 192 | ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
193 | (unsigned long long)input->block_bitmap, |
194 | start, metaend - 1); | |
ac27a0ec | 195 | else if (inside(input->inode_bitmap, start, metaend)) |
12062ddd | 196 | ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
197 | (unsigned long long)input->inode_bitmap, |
198 | start, metaend - 1); | |
ac27a0ec | 199 | else if (inside(input->inode_table, start, metaend) || |
2b2d6d01 | 200 | inside(itend - 1, start, metaend)) |
12062ddd ES |
201 | ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table " |
202 | "(%llu-%llu)", | |
1939e49a RD |
203 | (unsigned long long)input->inode_table, |
204 | itend - 1, start, metaend - 1); | |
ac27a0ec DK |
205 | else |
206 | err = 0; | |
207 | brelse(bh); | |
208 | ||
209 | return err; | |
210 | } | |
211 | ||
28c7bac0 YY |
212 | /* |
213 | * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex | |
214 | * group each time. | |
215 | */ | |
216 | struct ext4_new_flex_group_data { | |
217 | struct ext4_new_group_data *groups; /* new_group_data for groups | |
218 | in the flex group */ | |
219 | __u16 *bg_flags; /* block group flags of groups | |
220 | in @groups */ | |
5d1935ac BL |
221 | ext4_group_t resize_bg; /* number of allocated |
222 | new_group_data */ | |
28c7bac0 YY |
223 | ext4_group_t count; /* number of groups in @groups |
224 | */ | |
225 | }; | |
226 | ||
5d1935ac BL |
227 | /* |
228 | * Avoiding memory allocation failures due to too many groups added each time. | |
229 | */ | |
230 | #define MAX_RESIZE_BG 16384 | |
231 | ||
28c7bac0 YY |
232 | /* |
233 | * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of | |
234 | * @flexbg_size. | |
235 | * | |
236 | * Returns NULL on failure otherwise address of the allocated structure. | |
237 | */ | |
665d3e0a BL |
238 | static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size, |
239 | ext4_group_t o_group, ext4_group_t n_group) | |
28c7bac0 | 240 | { |
665d3e0a | 241 | ext4_group_t last_group; |
28c7bac0 YY |
242 | struct ext4_new_flex_group_data *flex_gd; |
243 | ||
244 | flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS); | |
245 | if (flex_gd == NULL) | |
246 | goto out3; | |
247 | ||
5d1935ac BL |
248 | if (unlikely(flexbg_size > MAX_RESIZE_BG)) |
249 | flex_gd->resize_bg = MAX_RESIZE_BG; | |
250 | else | |
251 | flex_gd->resize_bg = flexbg_size; | |
252 | ||
665d3e0a BL |
253 | /* Avoid allocating large 'groups' array if not needed */ |
254 | last_group = o_group | (flex_gd->resize_bg - 1); | |
255 | if (n_group <= last_group) | |
256 | flex_gd->resize_bg = 1 << fls(n_group - o_group + 1); | |
257 | else if (n_group - last_group < flex_gd->resize_bg) | |
258 | flex_gd->resize_bg = 1 << max(fls(last_group - o_group + 1), | |
259 | fls(n_group - last_group)); | |
260 | ||
5d1935ac | 261 | flex_gd->groups = kmalloc_array(flex_gd->resize_bg, |
6da2ec56 KC |
262 | sizeof(struct ext4_new_group_data), |
263 | GFP_NOFS); | |
28c7bac0 YY |
264 | if (flex_gd->groups == NULL) |
265 | goto out2; | |
266 | ||
5d1935ac | 267 | flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16), |
6da2ec56 | 268 | GFP_NOFS); |
28c7bac0 YY |
269 | if (flex_gd->bg_flags == NULL) |
270 | goto out1; | |
271 | ||
272 | return flex_gd; | |
273 | ||
274 | out1: | |
275 | kfree(flex_gd->groups); | |
276 | out2: | |
277 | kfree(flex_gd); | |
278 | out3: | |
279 | return NULL; | |
280 | } | |
281 | ||
282 | static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd) | |
283 | { | |
284 | kfree(flex_gd->bg_flags); | |
285 | kfree(flex_gd->groups); | |
286 | kfree(flex_gd); | |
287 | } | |
288 | ||
3fbea4b3 YY |
289 | /* |
290 | * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps | |
291 | * and inode tables for a flex group. | |
292 | * | |
293 | * This function is used by 64bit-resize. Note that this function allocates | |
294 | * group tables from the 1st group of groups contained by @flexgd, which may | |
295 | * be a partial of a flex group. | |
296 | * | |
297 | * @sb: super block of fs to which the groups belongs | |
03c1c290 YY |
298 | * |
299 | * Returns 0 on a successful allocation of the metadata blocks in the | |
300 | * block group. | |
3fbea4b3 | 301 | */ |
03c1c290 | 302 | static int ext4_alloc_group_tables(struct super_block *sb, |
3fbea4b3 | 303 | struct ext4_new_flex_group_data *flex_gd, |
658a5234 | 304 | unsigned int flexbg_size) |
3fbea4b3 YY |
305 | { |
306 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
3fbea4b3 YY |
307 | ext4_fsblk_t start_blk; |
308 | ext4_fsblk_t last_blk; | |
309 | ext4_group_t src_group; | |
310 | ext4_group_t bb_index = 0; | |
311 | ext4_group_t ib_index = 0; | |
312 | ext4_group_t it_index = 0; | |
313 | ext4_group_t group; | |
314 | ext4_group_t last_group; | |
315 | unsigned overhead; | |
b93c9535 | 316 | __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0; |
d77147ff | 317 | int i; |
3fbea4b3 YY |
318 | |
319 | BUG_ON(flex_gd->count == 0 || group_data == NULL); | |
320 | ||
321 | src_group = group_data[0].group; | |
322 | last_group = src_group + flex_gd->count - 1; | |
323 | ||
324 | BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) != | |
325 | (last_group & ~(flexbg_size - 1)))); | |
326 | next_group: | |
327 | group = group_data[0].group; | |
03c1c290 YY |
328 | if (src_group >= group_data[0].group + flex_gd->count) |
329 | return -ENOSPC; | |
3fbea4b3 YY |
330 | start_blk = ext4_group_first_block_no(sb, src_group); |
331 | last_blk = start_blk + group_data[src_group - group].blocks_count; | |
332 | ||
01f795f9 | 333 | overhead = ext4_group_overhead_blocks(sb, src_group); |
3fbea4b3 YY |
334 | |
335 | start_blk += overhead; | |
336 | ||
3fbea4b3 YY |
337 | /* We collect contiguous blocks as much as possible. */ |
338 | src_group++; | |
01f795f9 YY |
339 | for (; src_group <= last_group; src_group++) { |
340 | overhead = ext4_group_overhead_blocks(sb, src_group); | |
b93c9535 | 341 | if (overhead == 0) |
3fbea4b3 YY |
342 | last_blk += group_data[src_group - group].blocks_count; |
343 | else | |
344 | break; | |
01f795f9 | 345 | } |
3fbea4b3 YY |
346 | |
347 | /* Allocate block bitmaps */ | |
348 | for (; bb_index < flex_gd->count; bb_index++) { | |
349 | if (start_blk >= last_blk) | |
350 | goto next_group; | |
351 | group_data[bb_index].block_bitmap = start_blk++; | |
bd86298e | 352 | group = ext4_get_group_number(sb, start_blk - 1); |
3fbea4b3 | 353 | group -= group_data[0].group; |
d77147ff | 354 | group_data[group].mdata_blocks++; |
b93c9535 | 355 | flex_gd->bg_flags[group] &= uninit_mask; |
3fbea4b3 YY |
356 | } |
357 | ||
358 | /* Allocate inode bitmaps */ | |
359 | for (; ib_index < flex_gd->count; ib_index++) { | |
360 | if (start_blk >= last_blk) | |
361 | goto next_group; | |
362 | group_data[ib_index].inode_bitmap = start_blk++; | |
bd86298e | 363 | group = ext4_get_group_number(sb, start_blk - 1); |
3fbea4b3 | 364 | group -= group_data[0].group; |
d77147ff | 365 | group_data[group].mdata_blocks++; |
b93c9535 | 366 | flex_gd->bg_flags[group] &= uninit_mask; |
3fbea4b3 YY |
367 | } |
368 | ||
369 | /* Allocate inode tables */ | |
370 | for (; it_index < flex_gd->count; it_index++) { | |
b93c9535 TT |
371 | unsigned int itb = EXT4_SB(sb)->s_itb_per_group; |
372 | ext4_fsblk_t next_group_start; | |
373 | ||
374 | if (start_blk + itb > last_blk) | |
3fbea4b3 YY |
375 | goto next_group; |
376 | group_data[it_index].inode_table = start_blk; | |
b93c9535 TT |
377 | group = ext4_get_group_number(sb, start_blk); |
378 | next_group_start = ext4_group_first_block_no(sb, group + 1); | |
3fbea4b3 | 379 | group -= group_data[0].group; |
3fbea4b3 | 380 | |
b93c9535 TT |
381 | if (start_blk + itb > next_group_start) { |
382 | flex_gd->bg_flags[group + 1] &= uninit_mask; | |
383 | overhead = start_blk + itb - next_group_start; | |
d77147ff | 384 | group_data[group + 1].mdata_blocks += overhead; |
b93c9535 TT |
385 | itb -= overhead; |
386 | } | |
387 | ||
d77147ff | 388 | group_data[group].mdata_blocks += itb; |
b93c9535 | 389 | flex_gd->bg_flags[group] &= uninit_mask; |
3fbea4b3 YY |
390 | start_blk += EXT4_SB(sb)->s_itb_per_group; |
391 | } | |
392 | ||
d77147ff | 393 | /* Update free clusters count to exclude metadata blocks */ |
394 | for (i = 0; i < flex_gd->count; i++) { | |
395 | group_data[i].free_clusters_count -= | |
396 | EXT4_NUM_B2C(EXT4_SB(sb), | |
397 | group_data[i].mdata_blocks); | |
398 | } | |
399 | ||
3fbea4b3 YY |
400 | if (test_opt(sb, DEBUG)) { |
401 | int i; | |
402 | group = group_data[0].group; | |
403 | ||
404 | printk(KERN_DEBUG "EXT4-fs: adding a flex group with " | |
658a5234 | 405 | "%u groups, flexbg size is %u:\n", flex_gd->count, |
3fbea4b3 YY |
406 | flexbg_size); |
407 | ||
408 | for (i = 0; i < flex_gd->count; i++) { | |
d77147ff | 409 | ext4_debug( |
658a5234 | 410 | "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n", |
3fbea4b3 YY |
411 | ext4_bg_has_super(sb, group + i) ? "normal" : |
412 | "no-super", group + i, | |
413 | group_data[i].blocks_count, | |
d77147ff | 414 | group_data[i].free_clusters_count, |
415 | group_data[i].mdata_blocks); | |
3fbea4b3 YY |
416 | } |
417 | } | |
03c1c290 | 418 | return 0; |
3fbea4b3 YY |
419 | } |
420 | ||
ac27a0ec | 421 | static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, |
617ba13b | 422 | ext4_fsblk_t blk) |
ac27a0ec DK |
423 | { |
424 | struct buffer_head *bh; | |
425 | int err; | |
426 | ||
427 | bh = sb_getblk(sb, blk); | |
aebf0243 | 428 | if (unlikely(!bh)) |
860d21e2 | 429 | return ERR_PTR(-ENOMEM); |
5d601255 | 430 | BUFFER_TRACE(bh, "get_write_access"); |
188c299e JK |
431 | err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE); |
432 | if (err) { | |
ac27a0ec DK |
433 | brelse(bh); |
434 | bh = ERR_PTR(err); | |
435 | } else { | |
ac27a0ec DK |
436 | memset(bh->b_data, 0, sb->s_blocksize); |
437 | set_buffer_uptodate(bh); | |
ac27a0ec DK |
438 | } |
439 | ||
440 | return bh; | |
441 | } | |
442 | ||
83448bdf JK |
443 | static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits) |
444 | { | |
445 | return ext4_journal_ensure_credits_fn(handle, credits, | |
446 | EXT4_MAX_TRANS_DATA, 0, 0); | |
447 | } | |
448 | ||
33afdcc5 | 449 | /* |
d77147ff | 450 | * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used. |
33afdcc5 YY |
451 | * |
452 | * Helper function for ext4_setup_new_group_blocks() which set . | |
453 | * | |
454 | * @sb: super block | |
455 | * @handle: journal handle | |
456 | * @flex_gd: flex group data | |
457 | */ | |
458 | static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle, | |
459 | struct ext4_new_flex_group_data *flex_gd, | |
d77147ff | 460 | ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster) |
33afdcc5 | 461 | { |
d77147ff | 462 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
463 | ext4_group_t count = last_cluster - first_cluster + 1; | |
33afdcc5 YY |
464 | ext4_group_t count2; |
465 | ||
d77147ff | 466 | ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster, |
467 | last_cluster); | |
248b45b6 | 468 | for (; count > 0; count -= count2, first_cluster += count2) { |
33afdcc5 YY |
469 | ext4_fsblk_t start; |
470 | struct buffer_head *bh; | |
471 | ext4_group_t group; | |
472 | int err; | |
473 | ||
d77147ff | 474 | group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster)); |
475 | start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group)); | |
33afdcc5 YY |
476 | group -= flex_gd->groups[0].group; |
477 | ||
d77147ff | 478 | count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start); |
33afdcc5 YY |
479 | if (count2 > count) |
480 | count2 = count; | |
481 | ||
482 | if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) { | |
483 | BUG_ON(flex_gd->count > 1); | |
484 | continue; | |
485 | } | |
486 | ||
83448bdf | 487 | err = ext4_resize_ensure_credits_batch(handle, 1); |
a4130367 | 488 | if (err < 0) |
33afdcc5 YY |
489 | return err; |
490 | ||
491 | bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap); | |
aebf0243 | 492 | if (unlikely(!bh)) |
860d21e2 | 493 | return -ENOMEM; |
33afdcc5 | 494 | |
5d601255 | 495 | BUFFER_TRACE(bh, "get_write_access"); |
188c299e JK |
496 | err = ext4_journal_get_write_access(handle, sb, bh, |
497 | EXT4_JTR_NONE); | |
cea57941 VA |
498 | if (err) { |
499 | brelse(bh); | |
33afdcc5 | 500 | return err; |
cea57941 | 501 | } |
d77147ff | 502 | ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", |
503 | first_cluster, first_cluster - start, count2); | |
123e3016 | 504 | mb_set_bits(bh->b_data, first_cluster - start, count2); |
33afdcc5 YY |
505 | |
506 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
cea57941 | 507 | brelse(bh); |
33afdcc5 YY |
508 | if (unlikely(err)) |
509 | return err; | |
33afdcc5 YY |
510 | } |
511 | ||
512 | return 0; | |
513 | } | |
514 | ||
515 | /* | |
516 | * Set up the block and inode bitmaps, and the inode table for the new groups. | |
517 | * This doesn't need to be part of the main transaction, since we are only | |
518 | * changing blocks outside the actual filesystem. We still do journaling to | |
519 | * ensure the recovery is correct in case of a failure just after resize. | |
520 | * If any part of this fails, we simply abort the resize. | |
521 | * | |
522 | * setup_new_flex_group_blocks handles a flex group as follow: | |
523 | * 1. copy super block and GDT, and initialize group tables if necessary. | |
524 | * In this step, we only set bits in blocks bitmaps for blocks taken by | |
525 | * super block and GDT. | |
526 | * 2. allocate group tables in block bitmaps, that is, set bits in block | |
527 | * bitmap for blocks taken by group tables. | |
528 | */ | |
529 | static int setup_new_flex_group_blocks(struct super_block *sb, | |
530 | struct ext4_new_flex_group_data *flex_gd) | |
531 | { | |
532 | int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group}; | |
533 | ext4_fsblk_t start; | |
534 | ext4_fsblk_t block; | |
535 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
536 | struct ext4_super_block *es = sbi->s_es; | |
537 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
538 | __u16 *bg_flags = flex_gd->bg_flags; | |
539 | handle_t *handle; | |
540 | ext4_group_t group, count; | |
541 | struct buffer_head *bh = NULL; | |
542 | int reserved_gdb, i, j, err = 0, err2; | |
01f795f9 | 543 | int meta_bg; |
33afdcc5 YY |
544 | |
545 | BUG_ON(!flex_gd->count || !group_data || | |
546 | group_data[0].group != sbi->s_groups_count); | |
547 | ||
548 | reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); | |
e2b911c5 | 549 | meta_bg = ext4_has_feature_meta_bg(sb); |
33afdcc5 YY |
550 | |
551 | /* This transaction may be extended/restarted along the way */ | |
9924a92a | 552 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); |
33afdcc5 YY |
553 | if (IS_ERR(handle)) |
554 | return PTR_ERR(handle); | |
555 | ||
556 | group = group_data[0].group; | |
557 | for (i = 0; i < flex_gd->count; i++, group++) { | |
558 | unsigned long gdblocks; | |
01f795f9 | 559 | ext4_grpblk_t overhead; |
33afdcc5 YY |
560 | |
561 | gdblocks = ext4_bg_num_gdb(sb, group); | |
562 | start = ext4_group_first_block_no(sb, group); | |
563 | ||
01f795f9 | 564 | if (meta_bg == 0 && !ext4_bg_has_super(sb, group)) |
6df935ad YY |
565 | goto handle_itb; |
566 | ||
40dd7953 KS |
567 | if (meta_bg == 1) |
568 | goto handle_itb; | |
01f795f9 YY |
569 | |
570 | block = start + ext4_bg_has_super(sb, group); | |
33afdcc5 | 571 | /* Copy all of the GDT blocks into the backup in this group */ |
01f795f9 | 572 | for (j = 0; j < gdblocks; j++, block++) { |
33afdcc5 YY |
573 | struct buffer_head *gdb; |
574 | ||
575 | ext4_debug("update backup group %#04llx\n", block); | |
83448bdf | 576 | err = ext4_resize_ensure_credits_batch(handle, 1); |
a4130367 | 577 | if (err < 0) |
33afdcc5 YY |
578 | goto out; |
579 | ||
580 | gdb = sb_getblk(sb, block); | |
aebf0243 | 581 | if (unlikely(!gdb)) { |
860d21e2 | 582 | err = -ENOMEM; |
33afdcc5 YY |
583 | goto out; |
584 | } | |
585 | ||
5d601255 | 586 | BUFFER_TRACE(gdb, "get_write_access"); |
188c299e JK |
587 | err = ext4_journal_get_write_access(handle, sb, gdb, |
588 | EXT4_JTR_NONE); | |
33afdcc5 YY |
589 | if (err) { |
590 | brelse(gdb); | |
591 | goto out; | |
592 | } | |
1d0c3924 TT |
593 | memcpy(gdb->b_data, sbi_array_rcu_deref(sbi, |
594 | s_group_desc, j)->b_data, gdb->b_size); | |
33afdcc5 YY |
595 | set_buffer_uptodate(gdb); |
596 | ||
597 | err = ext4_handle_dirty_metadata(handle, NULL, gdb); | |
598 | if (unlikely(err)) { | |
599 | brelse(gdb); | |
600 | goto out; | |
601 | } | |
602 | brelse(gdb); | |
603 | } | |
604 | ||
605 | /* Zero out all of the reserved backup group descriptor | |
606 | * table blocks | |
607 | */ | |
608 | if (ext4_bg_has_super(sb, group)) { | |
609 | err = sb_issue_zeroout(sb, gdblocks + start + 1, | |
610 | reserved_gdb, GFP_NOFS); | |
611 | if (err) | |
612 | goto out; | |
613 | } | |
614 | ||
6df935ad | 615 | handle_itb: |
e44fc921 | 616 | /* Initialize group tables of the group @group */ |
33afdcc5 YY |
617 | if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED)) |
618 | goto handle_bb; | |
619 | ||
620 | /* Zero out all of the inode table blocks */ | |
621 | block = group_data[i].inode_table; | |
622 | ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", | |
623 | block, sbi->s_itb_per_group); | |
624 | err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, | |
625 | GFP_NOFS); | |
626 | if (err) | |
627 | goto out; | |
628 | ||
629 | handle_bb: | |
630 | if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT) | |
631 | goto handle_ib; | |
632 | ||
633 | /* Initialize block bitmap of the @group */ | |
634 | block = group_data[i].block_bitmap; | |
83448bdf | 635 | err = ext4_resize_ensure_credits_batch(handle, 1); |
a4130367 | 636 | if (err < 0) |
33afdcc5 YY |
637 | goto out; |
638 | ||
639 | bh = bclean(handle, sb, block); | |
640 | if (IS_ERR(bh)) { | |
641 | err = PTR_ERR(bh); | |
642 | goto out; | |
643 | } | |
01f795f9 YY |
644 | overhead = ext4_group_overhead_blocks(sb, group); |
645 | if (overhead != 0) { | |
33afdcc5 YY |
646 | ext4_debug("mark backup superblock %#04llx (+0)\n", |
647 | start); | |
123e3016 | 648 | mb_set_bits(bh->b_data, 0, |
d77147ff | 649 | EXT4_NUM_B2C(sbi, overhead)); |
33afdcc5 | 650 | } |
d77147ff | 651 | ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count), |
33afdcc5 YY |
652 | sb->s_blocksize * 8, bh->b_data); |
653 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
9e402893 | 654 | brelse(bh); |
33afdcc5 YY |
655 | if (err) |
656 | goto out; | |
33afdcc5 YY |
657 | |
658 | handle_ib: | |
659 | if (bg_flags[i] & EXT4_BG_INODE_UNINIT) | |
660 | continue; | |
661 | ||
662 | /* Initialize inode bitmap of the @group */ | |
663 | block = group_data[i].inode_bitmap; | |
83448bdf | 664 | err = ext4_resize_ensure_credits_batch(handle, 1); |
a4130367 | 665 | if (err < 0) |
33afdcc5 YY |
666 | goto out; |
667 | /* Mark unused entries in inode bitmap used */ | |
668 | bh = bclean(handle, sb, block); | |
669 | if (IS_ERR(bh)) { | |
670 | err = PTR_ERR(bh); | |
671 | goto out; | |
672 | } | |
673 | ||
674 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), | |
675 | sb->s_blocksize * 8, bh->b_data); | |
676 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
9e402893 | 677 | brelse(bh); |
33afdcc5 YY |
678 | if (err) |
679 | goto out; | |
33afdcc5 | 680 | } |
33afdcc5 YY |
681 | |
682 | /* Mark group tables in block bitmap */ | |
683 | for (j = 0; j < GROUP_TABLE_COUNT; j++) { | |
684 | count = group_table_count[j]; | |
685 | start = (&group_data[0].block_bitmap)[j]; | |
686 | block = start; | |
687 | for (i = 1; i < flex_gd->count; i++) { | |
688 | block += group_table_count[j]; | |
689 | if (block == (&group_data[i].block_bitmap)[j]) { | |
690 | count += group_table_count[j]; | |
691 | continue; | |
692 | } | |
693 | err = set_flexbg_block_bitmap(sb, handle, | |
d77147ff | 694 | flex_gd, |
695 | EXT4_B2C(sbi, start), | |
696 | EXT4_B2C(sbi, | |
697 | start + count | |
698 | - 1)); | |
33afdcc5 YY |
699 | if (err) |
700 | goto out; | |
701 | count = group_table_count[j]; | |
b93c9535 | 702 | start = (&group_data[i].block_bitmap)[j]; |
33afdcc5 YY |
703 | block = start; |
704 | } | |
705 | ||
7d4cd3b4 KS |
706 | err = set_flexbg_block_bitmap(sb, handle, |
707 | flex_gd, | |
708 | EXT4_B2C(sbi, start), | |
709 | EXT4_B2C(sbi, | |
710 | start + count | |
711 | - 1)); | |
712 | if (err) | |
713 | goto out; | |
33afdcc5 YY |
714 | } |
715 | ||
716 | out: | |
33afdcc5 YY |
717 | err2 = ext4_journal_stop(handle); |
718 | if (err2 && !err) | |
719 | err = err2; | |
720 | ||
721 | return err; | |
722 | } | |
723 | ||
ac27a0ec DK |
724 | /* |
725 | * Iterate through the groups which hold BACKUP superblock/GDT copies in an | |
617ba13b | 726 | * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before |
ac27a0ec DK |
727 | * calling this for the first time. In a sparse filesystem it will be the |
728 | * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... | |
729 | * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... | |
730 | */ | |
bbc605cd LC |
731 | unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three, |
732 | unsigned int *five, unsigned int *seven) | |
ac27a0ec | 733 | { |
bbc605cd LC |
734 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
735 | unsigned int *min = three; | |
ac27a0ec | 736 | int mult = 3; |
bbc605cd LC |
737 | unsigned int ret; |
738 | ||
739 | if (ext4_has_feature_sparse_super2(sb)) { | |
740 | do { | |
741 | if (*min > 2) | |
742 | return UINT_MAX; | |
743 | ret = le32_to_cpu(es->s_backup_bgs[*min - 1]); | |
744 | *min += 1; | |
745 | } while (!ret); | |
746 | return ret; | |
747 | } | |
ac27a0ec | 748 | |
e2b911c5 | 749 | if (!ext4_has_feature_sparse_super(sb)) { |
ac27a0ec DK |
750 | ret = *min; |
751 | *min += 1; | |
752 | return ret; | |
753 | } | |
754 | ||
755 | if (*five < *min) { | |
756 | min = five; | |
757 | mult = 5; | |
758 | } | |
759 | if (*seven < *min) { | |
760 | min = seven; | |
761 | mult = 7; | |
762 | } | |
763 | ||
764 | ret = *min; | |
765 | *min *= mult; | |
766 | ||
767 | return ret; | |
768 | } | |
769 | ||
770 | /* | |
771 | * Check that all of the backup GDT blocks are held in the primary GDT block. | |
772 | * It is assumed that they are stored in group order. Returns the number of | |
773 | * groups in current filesystem that have BACKUPS, or -ve error code. | |
774 | */ | |
775 | static int verify_reserved_gdb(struct super_block *sb, | |
c72df9f9 | 776 | ext4_group_t end, |
ac27a0ec DK |
777 | struct buffer_head *primary) |
778 | { | |
617ba13b | 779 | const ext4_fsblk_t blk = primary->b_blocknr; |
ac27a0ec DK |
780 | unsigned three = 1; |
781 | unsigned five = 5; | |
782 | unsigned seven = 7; | |
783 | unsigned grp; | |
784 | __le32 *p = (__le32 *)primary->b_data; | |
785 | int gdbackups = 0; | |
786 | ||
617ba13b | 787 | while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { |
bd81d8ee LV |
788 | if (le32_to_cpu(*p++) != |
789 | grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ | |
12062ddd | 790 | ext4_warning(sb, "reserved GDT %llu" |
2ae02107 | 791 | " missing grp %d (%llu)", |
ac27a0ec | 792 | blk, grp, |
bd81d8ee LV |
793 | grp * |
794 | (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + | |
795 | blk); | |
ac27a0ec DK |
796 | return -EINVAL; |
797 | } | |
617ba13b | 798 | if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) |
ac27a0ec DK |
799 | return -EFBIG; |
800 | } | |
801 | ||
802 | return gdbackups; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Called when we need to bring a reserved group descriptor table block into | |
807 | * use from the resize inode. The primary copy of the new GDT block currently | |
808 | * is an indirect block (under the double indirect block in the resize inode). | |
809 | * The new backup GDT blocks will be stored as leaf blocks in this indirect | |
810 | * block, in group order. Even though we know all the block numbers we need, | |
811 | * we check to ensure that the resize inode has actually reserved these blocks. | |
812 | * | |
813 | * Don't need to update the block bitmaps because the blocks are still in use. | |
814 | * | |
815 | * We get all of the error cases out of the way, so that we are sure to not | |
816 | * fail once we start modifying the data on disk, because JBD has no rollback. | |
817 | */ | |
818 | static int add_new_gdb(handle_t *handle, struct inode *inode, | |
2f919710 | 819 | ext4_group_t group) |
ac27a0ec DK |
820 | { |
821 | struct super_block *sb = inode->i_sb; | |
617ba13b | 822 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
2f919710 | 823 | unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); |
617ba13b | 824 | ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; |
fb265c9c TT |
825 | struct buffer_head **o_group_desc, **n_group_desc = NULL; |
826 | struct buffer_head *dind = NULL; | |
827 | struct buffer_head *gdb_bh = NULL; | |
ac27a0ec | 828 | int gdbackups; |
fb265c9c | 829 | struct ext4_iloc iloc = { .bh = NULL }; |
ac27a0ec DK |
830 | __le32 *data; |
831 | int err; | |
832 | ||
833 | if (test_opt(sb, DEBUG)) | |
834 | printk(KERN_DEBUG | |
617ba13b | 835 | "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", |
ac27a0ec DK |
836 | gdb_num); |
837 | ||
fb265c9c TT |
838 | gdb_bh = ext4_sb_bread(sb, gdblock, 0); |
839 | if (IS_ERR(gdb_bh)) | |
840 | return PTR_ERR(gdb_bh); | |
ac27a0ec | 841 | |
c72df9f9 | 842 | gdbackups = verify_reserved_gdb(sb, group, gdb_bh); |
2f919710 | 843 | if (gdbackups < 0) { |
ac27a0ec | 844 | err = gdbackups; |
fb265c9c | 845 | goto errout; |
ac27a0ec DK |
846 | } |
847 | ||
617ba13b | 848 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
fb265c9c TT |
849 | dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); |
850 | if (IS_ERR(dind)) { | |
851 | err = PTR_ERR(dind); | |
852 | dind = NULL; | |
853 | goto errout; | |
ac27a0ec DK |
854 | } |
855 | ||
856 | data = (__le32 *)dind->b_data; | |
617ba13b | 857 | if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { |
12062ddd | 858 | ext4_warning(sb, "new group %u GDT block %llu not reserved", |
2f919710 | 859 | group, gdblock); |
ac27a0ec | 860 | err = -EINVAL; |
fb265c9c | 861 | goto errout; |
ac27a0ec DK |
862 | } |
863 | ||
5d601255 | 864 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
188c299e JK |
865 | err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh, |
866 | EXT4_JTR_NONE); | |
b4097142 | 867 | if (unlikely(err)) |
fb265c9c | 868 | goto errout; |
ac27a0ec | 869 | |
5d601255 | 870 | BUFFER_TRACE(gdb_bh, "get_write_access"); |
188c299e | 871 | err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE); |
b4097142 | 872 | if (unlikely(err)) |
fb265c9c | 873 | goto errout; |
ac27a0ec | 874 | |
5d601255 | 875 | BUFFER_TRACE(dind, "get_write_access"); |
188c299e | 876 | err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE); |
c9e87161 | 877 | if (unlikely(err)) { |
b4097142 | 878 | ext4_std_error(sb, err); |
c9e87161 DL |
879 | goto errout; |
880 | } | |
ac27a0ec | 881 | |
617ba13b | 882 | /* ext4_reserve_inode_write() gets a reference on the iloc */ |
b4097142 TT |
883 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
884 | if (unlikely(err)) | |
fb265c9c | 885 | goto errout; |
ac27a0ec | 886 | |
71b565ce TT |
887 | n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), |
888 | GFP_KERNEL); | |
ac27a0ec DK |
889 | if (!n_group_desc) { |
890 | err = -ENOMEM; | |
f18a5f21 TT |
891 | ext4_warning(sb, "not enough memory for %lu groups", |
892 | gdb_num + 1); | |
fb265c9c | 893 | goto errout; |
ac27a0ec DK |
894 | } |
895 | ||
896 | /* | |
897 | * Finally, we have all of the possible failures behind us... | |
898 | * | |
899 | * Remove new GDT block from inode double-indirect block and clear out | |
900 | * the new GDT block for use (which also "frees" the backup GDT blocks | |
901 | * from the reserved inode). We don't need to change the bitmaps for | |
902 | * these blocks, because they are marked as in-use from being in the | |
903 | * reserved inode, and will become GDT blocks (primary and backup). | |
904 | */ | |
617ba13b | 905 | data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; |
b4097142 TT |
906 | err = ext4_handle_dirty_metadata(handle, NULL, dind); |
907 | if (unlikely(err)) { | |
908 | ext4_std_error(sb, err); | |
fb265c9c | 909 | goto errout; |
b4097142 | 910 | } |
d77147ff | 911 | inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> |
912 | (9 - EXT4_SB(sb)->s_cluster_bits); | |
617ba13b | 913 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
2f919710 YY |
914 | memset(gdb_bh->b_data, 0, sb->s_blocksize); |
915 | err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); | |
b4097142 TT |
916 | if (unlikely(err)) { |
917 | ext4_std_error(sb, err); | |
8c380ab4 | 918 | iloc.bh = NULL; |
fb265c9c | 919 | goto errout; |
b4097142 TT |
920 | } |
921 | brelse(dind); | |
ac27a0ec | 922 | |
1d0c3924 TT |
923 | rcu_read_lock(); |
924 | o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc); | |
ac27a0ec | 925 | memcpy(n_group_desc, o_group_desc, |
617ba13b | 926 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); |
1d0c3924 | 927 | rcu_read_unlock(); |
2f919710 | 928 | n_group_desc[gdb_num] = gdb_bh; |
1d0c3924 | 929 | rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc); |
617ba13b | 930 | EXT4_SB(sb)->s_gdb_count++; |
1d0c3924 | 931 | ext4_kvfree_array_rcu(o_group_desc); |
ac27a0ec | 932 | |
05c2c00f | 933 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e8546d06 | 934 | le16_add_cpu(&es->s_reserved_gdt_blocks, -1); |
05c2c00f JK |
935 | ext4_superblock_csum_set(sb); |
936 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
a3f5cf14 | 937 | err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
b4097142 TT |
938 | if (err) |
939 | ext4_std_error(sb, err); | |
b4097142 | 940 | return err; |
fb265c9c | 941 | errout: |
b93b41d4 | 942 | kvfree(n_group_desc); |
ac27a0ec | 943 | brelse(iloc.bh); |
ac27a0ec | 944 | brelse(dind); |
2f919710 | 945 | brelse(gdb_bh); |
ac27a0ec | 946 | |
617ba13b | 947 | ext4_debug("leaving with error %d\n", err); |
ac27a0ec DK |
948 | return err; |
949 | } | |
950 | ||
01f795f9 | 951 | /* |
9dca529b KS |
952 | * If there is no available space in the existing block group descriptors for |
953 | * the new block group and there are no reserved block group descriptors, then | |
954 | * the meta_bg feature will get enabled, and es->s_first_meta_bg will get set | |
955 | * to the first block group that is managed using meta_bg and s_first_meta_bg | |
956 | * must be a multiple of EXT4_DESC_PER_BLOCK(sb). | |
957 | * This function will be called when first group of meta_bg is added to bring | |
958 | * new group descriptors block of new added meta_bg. | |
01f795f9 YY |
959 | */ |
960 | static int add_new_gdb_meta_bg(struct super_block *sb, | |
961 | handle_t *handle, ext4_group_t group) { | |
962 | ext4_fsblk_t gdblock; | |
963 | struct buffer_head *gdb_bh; | |
964 | struct buffer_head **o_group_desc, **n_group_desc; | |
965 | unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
966 | int err; | |
967 | ||
9dca529b KS |
968 | gdblock = ext4_group_first_block_no(sb, group) + |
969 | ext4_bg_has_super(sb, group); | |
fb265c9c TT |
970 | gdb_bh = ext4_sb_bread(sb, gdblock, 0); |
971 | if (IS_ERR(gdb_bh)) | |
972 | return PTR_ERR(gdb_bh); | |
71b565ce TT |
973 | n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *), |
974 | GFP_KERNEL); | |
01f795f9 | 975 | if (!n_group_desc) { |
61a9c11e | 976 | brelse(gdb_bh); |
01f795f9 YY |
977 | err = -ENOMEM; |
978 | ext4_warning(sb, "not enough memory for %lu groups", | |
979 | gdb_num + 1); | |
980 | return err; | |
981 | } | |
982 | ||
1d0c3924 TT |
983 | rcu_read_lock(); |
984 | o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc); | |
01f795f9 YY |
985 | memcpy(n_group_desc, o_group_desc, |
986 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); | |
1d0c3924 | 987 | rcu_read_unlock(); |
01f795f9 | 988 | n_group_desc[gdb_num] = gdb_bh; |
d64264d6 LC |
989 | |
990 | BUFFER_TRACE(gdb_bh, "get_write_access"); | |
188c299e | 991 | err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE); |
d64264d6 LC |
992 | if (err) { |
993 | kvfree(n_group_desc); | |
994 | brelse(gdb_bh); | |
995 | return err; | |
996 | } | |
997 | ||
1d0c3924 | 998 | rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc); |
01f795f9 | 999 | EXT4_SB(sb)->s_gdb_count++; |
1d0c3924 | 1000 | ext4_kvfree_array_rcu(o_group_desc); |
01f795f9 YY |
1001 | return err; |
1002 | } | |
1003 | ||
ac27a0ec DK |
1004 | /* |
1005 | * Called when we are adding a new group which has a backup copy of each of | |
1006 | * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. | |
1007 | * We need to add these reserved backup GDT blocks to the resize inode, so | |
1008 | * that they are kept for future resizing and not allocated to files. | |
1009 | * | |
1010 | * Each reserved backup GDT block will go into a different indirect block. | |
1011 | * The indirect blocks are actually the primary reserved GDT blocks, | |
1012 | * so we know in advance what their block numbers are. We only get the | |
1013 | * double-indirect block to verify it is pointing to the primary reserved | |
1014 | * GDT blocks so we don't overwrite a data block by accident. The reserved | |
1015 | * backup GDT blocks are stored in their reserved primary GDT block. | |
1016 | */ | |
1017 | static int reserve_backup_gdb(handle_t *handle, struct inode *inode, | |
668f4dc5 | 1018 | ext4_group_t group) |
ac27a0ec DK |
1019 | { |
1020 | struct super_block *sb = inode->i_sb; | |
617ba13b | 1021 | int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); |
d77147ff | 1022 | int cluster_bits = EXT4_SB(sb)->s_cluster_bits; |
ac27a0ec DK |
1023 | struct buffer_head **primary; |
1024 | struct buffer_head *dind; | |
617ba13b MC |
1025 | struct ext4_iloc iloc; |
1026 | ext4_fsblk_t blk; | |
ac27a0ec DK |
1027 | __le32 *data, *end; |
1028 | int gdbackups = 0; | |
1029 | int res, i; | |
1030 | int err; | |
1031 | ||
6da2ec56 | 1032 | primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS); |
ac27a0ec DK |
1033 | if (!primary) |
1034 | return -ENOMEM; | |
1035 | ||
617ba13b | 1036 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
fb265c9c TT |
1037 | dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); |
1038 | if (IS_ERR(dind)) { | |
1039 | err = PTR_ERR(dind); | |
1040 | dind = NULL; | |
ac27a0ec DK |
1041 | goto exit_free; |
1042 | } | |
1043 | ||
617ba13b | 1044 | blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; |
94460093 JB |
1045 | data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % |
1046 | EXT4_ADDR_PER_BLOCK(sb)); | |
617ba13b | 1047 | end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); |
ac27a0ec DK |
1048 | |
1049 | /* Get each reserved primary GDT block and verify it holds backups */ | |
1050 | for (res = 0; res < reserved_gdb; res++, blk++) { | |
1051 | if (le32_to_cpu(*data) != blk) { | |
12062ddd | 1052 | ext4_warning(sb, "reserved block %llu" |
ac27a0ec DK |
1053 | " not at offset %ld", |
1054 | blk, | |
1055 | (long)(data - (__le32 *)dind->b_data)); | |
1056 | err = -EINVAL; | |
1057 | goto exit_bh; | |
1058 | } | |
fb265c9c TT |
1059 | primary[res] = ext4_sb_bread(sb, blk, 0); |
1060 | if (IS_ERR(primary[res])) { | |
1061 | err = PTR_ERR(primary[res]); | |
1062 | primary[res] = NULL; | |
ac27a0ec DK |
1063 | goto exit_bh; |
1064 | } | |
c72df9f9 YY |
1065 | gdbackups = verify_reserved_gdb(sb, group, primary[res]); |
1066 | if (gdbackups < 0) { | |
ac27a0ec DK |
1067 | brelse(primary[res]); |
1068 | err = gdbackups; | |
1069 | goto exit_bh; | |
1070 | } | |
1071 | if (++data >= end) | |
1072 | data = (__le32 *)dind->b_data; | |
1073 | } | |
1074 | ||
1075 | for (i = 0; i < reserved_gdb; i++) { | |
5d601255 | 1076 | BUFFER_TRACE(primary[i], "get_write_access"); |
188c299e JK |
1077 | if ((err = ext4_journal_get_write_access(handle, sb, primary[i], |
1078 | EXT4_JTR_NONE))) | |
ac27a0ec | 1079 | goto exit_bh; |
ac27a0ec DK |
1080 | } |
1081 | ||
617ba13b | 1082 | if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) |
ac27a0ec DK |
1083 | goto exit_bh; |
1084 | ||
1085 | /* | |
1086 | * Finally we can add each of the reserved backup GDT blocks from | |
1087 | * the new group to its reserved primary GDT block. | |
1088 | */ | |
668f4dc5 | 1089 | blk = group * EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1090 | for (i = 0; i < reserved_gdb; i++) { |
1091 | int err2; | |
1092 | data = (__le32 *)primary[i]->b_data; | |
ac27a0ec | 1093 | data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); |
0390131b | 1094 | err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); |
ac27a0ec DK |
1095 | if (!err) |
1096 | err = err2; | |
1097 | } | |
d77147ff | 1098 | |
1099 | inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits); | |
617ba13b | 1100 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
1101 | |
1102 | exit_bh: | |
1103 | while (--res >= 0) | |
1104 | brelse(primary[res]); | |
1105 | brelse(dind); | |
1106 | ||
1107 | exit_free: | |
1108 | kfree(primary); | |
1109 | ||
1110 | return err; | |
1111 | } | |
1112 | ||
8f49ec60 BL |
1113 | static inline void ext4_set_block_group_nr(struct super_block *sb, char *data, |
1114 | ext4_group_t group) | |
1115 | { | |
1116 | struct ext4_super_block *es = (struct ext4_super_block *) data; | |
1117 | ||
1118 | es->s_block_group_nr = cpu_to_le16(group); | |
1119 | if (ext4_has_metadata_csum(sb)) | |
1120 | es->s_checksum = ext4_superblock_csum(sb, es); | |
1121 | } | |
1122 | ||
ac27a0ec | 1123 | /* |
617ba13b | 1124 | * Update the backup copies of the ext4 metadata. These don't need to be part |
ac27a0ec DK |
1125 | * of the main resize transaction, because e2fsck will re-write them if there |
1126 | * is a problem (basically only OOM will cause a problem). However, we | |
1127 | * _should_ update the backups if possible, in case the primary gets trashed | |
1128 | * for some reason and we need to run e2fsck from a backup superblock. The | |
1129 | * important part is that the new block and inode counts are in the backup | |
1130 | * superblocks, and the location of the new group metadata in the GDT backups. | |
1131 | * | |
32ed5058 TT |
1132 | * We do not need take the s_resize_lock for this, because these |
1133 | * blocks are not otherwise touched by the filesystem code when it is | |
1134 | * mounted. We don't need to worry about last changing from | |
1135 | * sbi->s_groups_count, because the worst that can happen is that we | |
1136 | * do not copy the full number of backups at this time. The resize | |
1137 | * which changed s_groups_count will backup again. | |
ac27a0ec | 1138 | */ |
904dad47 | 1139 | static void update_backups(struct super_block *sb, sector_t blk_off, char *data, |
01f795f9 | 1140 | int size, int meta_bg) |
ac27a0ec | 1141 | { |
617ba13b | 1142 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
01f795f9 | 1143 | ext4_group_t last; |
617ba13b | 1144 | const int bpg = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1145 | unsigned three = 1; |
1146 | unsigned five = 5; | |
1147 | unsigned seven = 7; | |
01f795f9 | 1148 | ext4_group_t group = 0; |
ac27a0ec DK |
1149 | int rest = sb->s_blocksize - size; |
1150 | handle_t *handle; | |
1151 | int err = 0, err2; | |
1152 | ||
9924a92a | 1153 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); |
ac27a0ec DK |
1154 | if (IS_ERR(handle)) { |
1155 | group = 1; | |
1156 | err = PTR_ERR(handle); | |
1157 | goto exit_err; | |
1158 | } | |
1159 | ||
01f795f9 YY |
1160 | if (meta_bg == 0) { |
1161 | group = ext4_list_backups(sb, &three, &five, &seven); | |
1162 | last = sbi->s_groups_count; | |
1163 | } else { | |
904dad47 | 1164 | group = ext4_get_group_number(sb, blk_off) + 1; |
01f795f9 YY |
1165 | last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2); |
1166 | } | |
1167 | ||
1168 | while (group < sbi->s_groups_count) { | |
ac27a0ec | 1169 | struct buffer_head *bh; |
01f795f9 | 1170 | ext4_fsblk_t backup_block; |
8f49ec60 BL |
1171 | int has_super = ext4_bg_has_super(sb, group); |
1172 | ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group); | |
ac27a0ec DK |
1173 | |
1174 | /* Out of journal space, and can't get more - abort - so sad */ | |
83448bdf | 1175 | err = ext4_resize_ensure_credits_batch(handle, 1); |
a4130367 | 1176 | if (err < 0) |
ac27a0ec DK |
1177 | break; |
1178 | ||
01f795f9 | 1179 | if (meta_bg == 0) |
9378c676 | 1180 | backup_block = ((ext4_fsblk_t)group) * bpg + blk_off; |
01f795f9 | 1181 | else |
8f49ec60 | 1182 | backup_block = first_block + has_super; |
01f795f9 YY |
1183 | |
1184 | bh = sb_getblk(sb, backup_block); | |
aebf0243 | 1185 | if (unlikely(!bh)) { |
860d21e2 | 1186 | err = -ENOMEM; |
ac27a0ec DK |
1187 | break; |
1188 | } | |
01f795f9 YY |
1189 | ext4_debug("update metadata backup %llu(+%llu)\n", |
1190 | backup_block, backup_block - | |
1191 | ext4_group_first_block_no(sb, group)); | |
5d601255 | 1192 | BUFFER_TRACE(bh, "get_write_access"); |
188c299e | 1193 | if ((err = ext4_journal_get_write_access(handle, sb, bh, |
9adac8b0 KS |
1194 | EXT4_JTR_NONE))) { |
1195 | brelse(bh); | |
ac27a0ec | 1196 | break; |
9adac8b0 | 1197 | } |
ac27a0ec DK |
1198 | lock_buffer(bh); |
1199 | memcpy(bh->b_data, data, size); | |
1200 | if (rest) | |
1201 | memset(bh->b_data + size, 0, rest); | |
8f49ec60 BL |
1202 | if (has_super && (backup_block == first_block)) |
1203 | ext4_set_block_group_nr(sb, bh->b_data, group); | |
ac27a0ec DK |
1204 | set_buffer_uptodate(bh); |
1205 | unlock_buffer(bh); | |
b4097142 TT |
1206 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
1207 | if (unlikely(err)) | |
1208 | ext4_std_error(sb, err); | |
ac27a0ec | 1209 | brelse(bh); |
01f795f9 YY |
1210 | |
1211 | if (meta_bg == 0) | |
1212 | group = ext4_list_backups(sb, &three, &five, &seven); | |
1213 | else if (group == last) | |
1214 | break; | |
1215 | else | |
1216 | group = last; | |
ac27a0ec | 1217 | } |
617ba13b | 1218 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
1219 | err = err2; |
1220 | ||
1221 | /* | |
1222 | * Ugh! Need to have e2fsck write the backup copies. It is too | |
1223 | * late to revert the resize, we shouldn't fail just because of | |
1224 | * the backup copies (they are only needed in case of corruption). | |
1225 | * | |
1226 | * However, if we got here we have a journal problem too, so we | |
1227 | * can't really start a transaction to mark the superblock. | |
1228 | * Chicken out and just set the flag on the hope it will be written | |
1229 | * to disk, and if not - we will simply wait until next fsck. | |
1230 | */ | |
1231 | exit_err: | |
1232 | if (err) { | |
12062ddd | 1233 | ext4_warning(sb, "can't update backup for group %u (err %d), " |
ac27a0ec | 1234 | "forcing fsck on next reboot", group, err); |
617ba13b MC |
1235 | sbi->s_mount_state &= ~EXT4_VALID_FS; |
1236 | sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); | |
ac27a0ec DK |
1237 | mark_buffer_dirty(sbi->s_sbh); |
1238 | } | |
1239 | } | |
1240 | ||
bb08c1e7 YY |
1241 | /* |
1242 | * ext4_add_new_descs() adds @count group descriptor of groups | |
1243 | * starting at @group | |
1244 | * | |
1245 | * @handle: journal handle | |
1246 | * @sb: super block | |
1247 | * @group: the group no. of the first group desc to be added | |
1248 | * @resize_inode: the resize inode | |
1249 | * @count: number of group descriptors to be added | |
1250 | */ | |
1251 | static int ext4_add_new_descs(handle_t *handle, struct super_block *sb, | |
1252 | ext4_group_t group, struct inode *resize_inode, | |
1253 | ext4_group_t count) | |
1254 | { | |
1255 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1256 | struct ext4_super_block *es = sbi->s_es; | |
1257 | struct buffer_head *gdb_bh; | |
1258 | int i, gdb_off, gdb_num, err = 0; | |
01f795f9 | 1259 | int meta_bg; |
bb08c1e7 | 1260 | |
e2b911c5 | 1261 | meta_bg = ext4_has_feature_meta_bg(sb); |
bb08c1e7 YY |
1262 | for (i = 0; i < count; i++, group++) { |
1263 | int reserved_gdb = ext4_bg_has_super(sb, group) ? | |
1264 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; | |
1265 | ||
1266 | gdb_off = group % EXT4_DESC_PER_BLOCK(sb); | |
1267 | gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
1268 | ||
1269 | /* | |
1270 | * We will only either add reserved group blocks to a backup group | |
1271 | * or remove reserved blocks for the first group in a new group block. | |
1272 | * Doing both would be mean more complex code, and sane people don't | |
1273 | * use non-sparse filesystems anymore. This is already checked above. | |
1274 | */ | |
1275 | if (gdb_off) { | |
1d0c3924 TT |
1276 | gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, |
1277 | gdb_num); | |
5d601255 | 1278 | BUFFER_TRACE(gdb_bh, "get_write_access"); |
188c299e JK |
1279 | err = ext4_journal_get_write_access(handle, sb, gdb_bh, |
1280 | EXT4_JTR_NONE); | |
bb08c1e7 YY |
1281 | |
1282 | if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group)) | |
1283 | err = reserve_backup_gdb(handle, resize_inode, group); | |
01f795f9 YY |
1284 | } else if (meta_bg != 0) { |
1285 | err = add_new_gdb_meta_bg(sb, handle, group); | |
1286 | } else { | |
bb08c1e7 | 1287 | err = add_new_gdb(handle, resize_inode, group); |
01f795f9 | 1288 | } |
bb08c1e7 YY |
1289 | if (err) |
1290 | break; | |
1291 | } | |
1292 | return err; | |
1293 | } | |
1294 | ||
41a246d1 DW |
1295 | static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block) |
1296 | { | |
1297 | struct buffer_head *bh = sb_getblk(sb, block); | |
aebf0243 | 1298 | if (unlikely(!bh)) |
41a246d1 | 1299 | return NULL; |
7f1468d1 | 1300 | if (!bh_uptodate_or_lock(bh)) { |
2d069c08 | 1301 | if (ext4_read_bh(bh, 0, NULL) < 0) { |
7f1468d1 DM |
1302 | brelse(bh); |
1303 | return NULL; | |
1304 | } | |
41a246d1 | 1305 | } |
41a246d1 DW |
1306 | |
1307 | return bh; | |
1308 | } | |
1309 | ||
1310 | static int ext4_set_bitmap_checksums(struct super_block *sb, | |
41a246d1 DW |
1311 | struct ext4_group_desc *gdp, |
1312 | struct ext4_new_group_data *group_data) | |
1313 | { | |
1314 | struct buffer_head *bh; | |
1315 | ||
9aa5d32b | 1316 | if (!ext4_has_metadata_csum(sb)) |
41a246d1 DW |
1317 | return 0; |
1318 | ||
1319 | bh = ext4_get_bitmap(sb, group_data->inode_bitmap); | |
1320 | if (!bh) | |
1321 | return -EIO; | |
4fd873c8 | 1322 | ext4_inode_bitmap_csum_set(sb, gdp, bh, |
41a246d1 DW |
1323 | EXT4_INODES_PER_GROUP(sb) / 8); |
1324 | brelse(bh); | |
1325 | ||
fa77dcfa DW |
1326 | bh = ext4_get_bitmap(sb, group_data->block_bitmap); |
1327 | if (!bh) | |
1328 | return -EIO; | |
1df9bde4 | 1329 | ext4_block_bitmap_csum_set(sb, gdp, bh); |
fa77dcfa DW |
1330 | brelse(bh); |
1331 | ||
41a246d1 DW |
1332 | return 0; |
1333 | } | |
1334 | ||
083f5b24 YY |
1335 | /* |
1336 | * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg | |
1337 | */ | |
1338 | static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb, | |
1339 | struct ext4_new_flex_group_data *flex_gd) | |
1340 | { | |
1341 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
1342 | struct ext4_group_desc *gdp; | |
1343 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1344 | struct buffer_head *gdb_bh; | |
1345 | ext4_group_t group; | |
1346 | __u16 *bg_flags = flex_gd->bg_flags; | |
1347 | int i, gdb_off, gdb_num, err = 0; | |
d77147ff | 1348 | |
083f5b24 YY |
1349 | |
1350 | for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) { | |
1351 | group = group_data->group; | |
1352 | ||
1353 | gdb_off = group % EXT4_DESC_PER_BLOCK(sb); | |
1354 | gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
1355 | ||
1356 | /* | |
1357 | * get_write_access() has been called on gdb_bh by ext4_add_new_desc(). | |
1358 | */ | |
1d0c3924 | 1359 | gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num); |
083f5b24 | 1360 | /* Update group descriptor block for new group */ |
2716b802 | 1361 | gdp = (struct ext4_group_desc *)(gdb_bh->b_data + |
083f5b24 YY |
1362 | gdb_off * EXT4_DESC_SIZE(sb)); |
1363 | ||
1364 | memset(gdp, 0, EXT4_DESC_SIZE(sb)); | |
1365 | ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap); | |
1366 | ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap); | |
1df9bde4 | 1367 | err = ext4_set_bitmap_checksums(sb, gdp, group_data); |
41a246d1 DW |
1368 | if (err) { |
1369 | ext4_std_error(sb, err); | |
1370 | break; | |
1371 | } | |
1372 | ||
083f5b24 YY |
1373 | ext4_inode_table_set(sb, gdp, group_data->inode_table); |
1374 | ext4_free_group_clusters_set(sb, gdp, | |
d77147ff | 1375 | group_data->free_clusters_count); |
083f5b24 | 1376 | ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); |
93f90526 TT |
1377 | if (ext4_has_group_desc_csum(sb)) |
1378 | ext4_itable_unused_set(sb, gdp, | |
1379 | EXT4_INODES_PER_GROUP(sb)); | |
083f5b24 | 1380 | gdp->bg_flags = cpu_to_le16(*bg_flags); |
feb0ab32 | 1381 | ext4_group_desc_csum_set(sb, group, gdp); |
083f5b24 YY |
1382 | |
1383 | err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); | |
1384 | if (unlikely(err)) { | |
1385 | ext4_std_error(sb, err); | |
1386 | break; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * We can allocate memory for mb_alloc based on the new group | |
1391 | * descriptor | |
1392 | */ | |
1393 | err = ext4_mb_add_groupinfo(sb, group, gdp); | |
1394 | if (err) | |
1395 | break; | |
1396 | } | |
1397 | return err; | |
1398 | } | |
1399 | ||
026d0d27 KO |
1400 | static void ext4_add_overhead(struct super_block *sb, |
1401 | const ext4_fsblk_t overhead) | |
1402 | { | |
1403 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1404 | struct ext4_super_block *es = sbi->s_es; | |
1405 | ||
1406 | sbi->s_overhead += overhead; | |
1407 | es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead); | |
1408 | smp_wmb(); | |
1409 | } | |
1410 | ||
2e10e2f2 YY |
1411 | /* |
1412 | * ext4_update_super() updates the super block so that the newly added | |
1413 | * groups can be seen by the filesystem. | |
1414 | * | |
1415 | * @sb: super block | |
1416 | * @flex_gd: new added groups | |
1417 | */ | |
1418 | static void ext4_update_super(struct super_block *sb, | |
1419 | struct ext4_new_flex_group_data *flex_gd) | |
1420 | { | |
1421 | ext4_fsblk_t blocks_count = 0; | |
1422 | ext4_fsblk_t free_blocks = 0; | |
1423 | ext4_fsblk_t reserved_blocks = 0; | |
1424 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
1425 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1426 | struct ext4_super_block *es = sbi->s_es; | |
8a991849 | 1427 | int i; |
2e10e2f2 YY |
1428 | |
1429 | BUG_ON(flex_gd->count == 0 || group_data == NULL); | |
1430 | /* | |
1431 | * Make the new blocks and inodes valid next. We do this before | |
1432 | * increasing the group count so that once the group is enabled, | |
1433 | * all of its blocks and inodes are already valid. | |
1434 | * | |
1435 | * We always allocate group-by-group, then block-by-block or | |
1436 | * inode-by-inode within a group, so enabling these | |
1437 | * blocks/inodes before the group is live won't actually let us | |
1438 | * allocate the new space yet. | |
1439 | */ | |
1440 | for (i = 0; i < flex_gd->count; i++) { | |
1441 | blocks_count += group_data[i].blocks_count; | |
d77147ff | 1442 | free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count); |
2e10e2f2 YY |
1443 | } |
1444 | ||
1445 | reserved_blocks = ext4_r_blocks_count(es) * 100; | |
01f795f9 | 1446 | reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es)); |
2e10e2f2 YY |
1447 | reserved_blocks *= blocks_count; |
1448 | do_div(reserved_blocks, 100); | |
1449 | ||
05c2c00f | 1450 | lock_buffer(sbi->s_sbh); |
2e10e2f2 | 1451 | ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count); |
636d7e2e | 1452 | ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks); |
2e10e2f2 YY |
1453 | le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) * |
1454 | flex_gd->count); | |
636d7e2e DW |
1455 | le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) * |
1456 | flex_gd->count); | |
2e10e2f2 | 1457 | |
01f795f9 | 1458 | ext4_debug("free blocks count %llu", ext4_free_blocks_count(es)); |
2e10e2f2 YY |
1459 | /* |
1460 | * We need to protect s_groups_count against other CPUs seeing | |
1461 | * inconsistent state in the superblock. | |
1462 | * | |
1463 | * The precise rules we use are: | |
1464 | * | |
1465 | * * Writers must perform a smp_wmb() after updating all | |
1466 | * dependent data and before modifying the groups count | |
1467 | * | |
1468 | * * Readers must perform an smp_rmb() after reading the groups | |
1469 | * count and before reading any dependent data. | |
1470 | * | |
1471 | * NB. These rules can be relaxed when checking the group count | |
1472 | * while freeing data, as we can only allocate from a block | |
1473 | * group after serialising against the group count, and we can | |
1474 | * only then free after serialising in turn against that | |
1475 | * allocation. | |
1476 | */ | |
1477 | smp_wmb(); | |
1478 | ||
1479 | /* Update the global fs size fields */ | |
1480 | sbi->s_groups_count += flex_gd->count; | |
c5c72d81 TT |
1481 | sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, |
1482 | (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); | |
2e10e2f2 YY |
1483 | |
1484 | /* Update the reserved block counts only once the new group is | |
1485 | * active. */ | |
1486 | ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + | |
1487 | reserved_blocks); | |
1488 | ||
1489 | /* Update the free space counts */ | |
1490 | percpu_counter_add(&sbi->s_freeclusters_counter, | |
810da240 | 1491 | EXT4_NUM_B2C(sbi, free_blocks)); |
2e10e2f2 YY |
1492 | percpu_counter_add(&sbi->s_freeinodes_counter, |
1493 | EXT4_INODES_PER_GROUP(sb) * flex_gd->count); | |
1494 | ||
01f795f9 YY |
1495 | ext4_debug("free blocks count %llu", |
1496 | percpu_counter_read(&sbi->s_freeclusters_counter)); | |
e2b911c5 | 1497 | if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) { |
2e10e2f2 | 1498 | ext4_group_t flex_group; |
7c990728 SJS |
1499 | struct flex_groups *fg; |
1500 | ||
2e10e2f2 | 1501 | flex_group = ext4_flex_group(sbi, group_data[0].group); |
7c990728 | 1502 | fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group); |
90ba983f | 1503 | atomic64_add(EXT4_NUM_B2C(sbi, free_blocks), |
7c990728 | 1504 | &fg->free_clusters); |
2e10e2f2 | 1505 | atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count, |
7c990728 | 1506 | &fg->free_inodes); |
2e10e2f2 YY |
1507 | } |
1508 | ||
952fc18e | 1509 | /* |
026d0d27 KO |
1510 | * Update the fs overhead information. |
1511 | * | |
1512 | * For bigalloc, if the superblock already has a properly calculated | |
1513 | * overhead, update it with a value based on numbers already computed | |
1514 | * above for the newly allocated capacity. | |
952fc18e | 1515 | */ |
026d0d27 KO |
1516 | if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0)) |
1517 | ext4_add_overhead(sb, | |
1518 | EXT4_NUM_B2C(sbi, blocks_count - free_blocks)); | |
1519 | else | |
1520 | ext4_calculate_overhead(sb); | |
de394a86 | 1521 | es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead); |
952fc18e | 1522 | |
a408f33e BL |
1523 | ext4_superblock_csum_set(sb); |
1524 | unlock_buffer(sbi->s_sbh); | |
2e10e2f2 YY |
1525 | if (test_opt(sb, DEBUG)) |
1526 | printk(KERN_DEBUG "EXT4-fs: added group %u:" | |
1527 | "%llu blocks(%llu free %llu reserved)\n", flex_gd->count, | |
1528 | blocks_count, free_blocks, reserved_blocks); | |
1529 | } | |
1530 | ||
4bac1f8c YY |
1531 | /* Add a flex group to an fs. Ensure we handle all possible error conditions |
1532 | * _before_ we start modifying the filesystem, because we cannot abort the | |
1533 | * transaction and not have it write the data to disk. | |
1534 | */ | |
1535 | static int ext4_flex_group_add(struct super_block *sb, | |
1536 | struct inode *resize_inode, | |
1537 | struct ext4_new_flex_group_data *flex_gd) | |
1538 | { | |
1539 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1540 | struct ext4_super_block *es = sbi->s_es; | |
1541 | ext4_fsblk_t o_blocks_count; | |
1542 | ext4_grpblk_t last; | |
1543 | ext4_group_t group; | |
1544 | handle_t *handle; | |
1545 | unsigned reserved_gdb; | |
1546 | int err = 0, err2 = 0, credit; | |
1547 | ||
1548 | BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags); | |
1549 | ||
1550 | reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); | |
1551 | o_blocks_count = ext4_blocks_count(es); | |
1552 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); | |
1553 | BUG_ON(last); | |
1554 | ||
1555 | err = setup_new_flex_group_blocks(sb, flex_gd); | |
1556 | if (err) | |
1557 | goto exit; | |
1558 | /* | |
1559 | * We will always be modifying at least the superblock and GDT | |
2c869b26 | 1560 | * blocks. If we are adding a group past the last current GDT block, |
4bac1f8c YY |
1561 | * we will also modify the inode and the dindirect block. If we |
1562 | * are adding a group with superblock/GDT backups we will also | |
1563 | * modify each of the reserved GDT dindirect blocks. | |
1564 | */ | |
2c869b26 JK |
1565 | credit = 3; /* sb, resize inode, resize inode dindirect */ |
1566 | /* GDT blocks */ | |
1567 | credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb)); | |
1568 | credit += reserved_gdb; /* Reserved GDT dindirect blocks */ | |
9924a92a | 1569 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit); |
4bac1f8c YY |
1570 | if (IS_ERR(handle)) { |
1571 | err = PTR_ERR(handle); | |
1572 | goto exit; | |
1573 | } | |
1574 | ||
5d601255 | 1575 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
188c299e JK |
1576 | err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh, |
1577 | EXT4_JTR_NONE); | |
4bac1f8c YY |
1578 | if (err) |
1579 | goto exit_journal; | |
1580 | ||
1581 | group = flex_gd->groups[0].group; | |
49598e04 | 1582 | BUG_ON(group != sbi->s_groups_count); |
4bac1f8c YY |
1583 | err = ext4_add_new_descs(handle, sb, group, |
1584 | resize_inode, flex_gd->count); | |
1585 | if (err) | |
1586 | goto exit_journal; | |
1587 | ||
1588 | err = ext4_setup_new_descs(handle, sb, flex_gd); | |
1589 | if (err) | |
1590 | goto exit_journal; | |
1591 | ||
1592 | ext4_update_super(sb, flex_gd); | |
1593 | ||
a3f5cf14 | 1594 | err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); |
4bac1f8c YY |
1595 | |
1596 | exit_journal: | |
1597 | err2 = ext4_journal_stop(handle); | |
1598 | if (!err) | |
1599 | err = err2; | |
1600 | ||
1601 | if (!err) { | |
2ebd1704 YY |
1602 | int gdb_num = group / EXT4_DESC_PER_BLOCK(sb); |
1603 | int gdb_num_end = ((group + flex_gd->count - 1) / | |
1604 | EXT4_DESC_PER_BLOCK(sb)); | |
a6b3bfe1 MH |
1605 | int meta_bg = ext4_has_feature_meta_bg(sb) && |
1606 | gdb_num >= le32_to_cpu(es->s_first_meta_bg); | |
31f13421 KS |
1607 | sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr - |
1608 | ext4_group_first_block_no(sb, 0); | |
2ebd1704 | 1609 | |
0aeaa255 BL |
1610 | update_backups(sb, ext4_group_first_block_no(sb, 0), |
1611 | (char *)es, sizeof(struct ext4_super_block), 0); | |
2ebd1704 | 1612 | for (; gdb_num <= gdb_num_end; gdb_num++) { |
4bac1f8c | 1613 | struct buffer_head *gdb_bh; |
2ebd1704 | 1614 | |
1d0c3924 TT |
1615 | gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, |
1616 | gdb_num); | |
31f13421 KS |
1617 | update_backups(sb, gdb_bh->b_blocknr - padding_blocks, |
1618 | gdb_bh->b_data, gdb_bh->b_size, meta_bg); | |
4bac1f8c YY |
1619 | } |
1620 | } | |
1621 | exit: | |
1622 | return err; | |
1623 | } | |
1624 | ||
19c5246d YY |
1625 | static int ext4_setup_next_flex_gd(struct super_block *sb, |
1626 | struct ext4_new_flex_group_data *flex_gd, | |
5d1935ac | 1627 | ext4_fsblk_t n_blocks_count) |
19c5246d | 1628 | { |
d77147ff | 1629 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
1630 | struct ext4_super_block *es = sbi->s_es; | |
19c5246d YY |
1631 | struct ext4_new_group_data *group_data = flex_gd->groups; |
1632 | ext4_fsblk_t o_blocks_count; | |
1633 | ext4_group_t n_group; | |
1634 | ext4_group_t group; | |
1635 | ext4_group_t last_group; | |
1636 | ext4_grpblk_t last; | |
d77147ff | 1637 | ext4_grpblk_t clusters_per_group; |
19c5246d YY |
1638 | unsigned long i; |
1639 | ||
d77147ff | 1640 | clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb); |
19c5246d YY |
1641 | |
1642 | o_blocks_count = ext4_blocks_count(es); | |
1643 | ||
1644 | if (o_blocks_count == n_blocks_count) | |
1645 | return 0; | |
1646 | ||
1647 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); | |
1648 | BUG_ON(last); | |
1649 | ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); | |
1650 | ||
5d1935ac | 1651 | last_group = group | (flex_gd->resize_bg - 1); |
19c5246d YY |
1652 | if (last_group > n_group) |
1653 | last_group = n_group; | |
1654 | ||
1655 | flex_gd->count = last_group - group + 1; | |
1656 | ||
1657 | for (i = 0; i < flex_gd->count; i++) { | |
1658 | int overhead; | |
1659 | ||
1660 | group_data[i].group = group + i; | |
d77147ff | 1661 | group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb); |
01f795f9 | 1662 | overhead = ext4_group_overhead_blocks(sb, group + i); |
d77147ff | 1663 | group_data[i].mdata_blocks = overhead; |
1664 | group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb); | |
7f511862 | 1665 | if (ext4_has_group_desc_csum(sb)) { |
19c5246d YY |
1666 | flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT | |
1667 | EXT4_BG_INODE_UNINIT; | |
7f511862 TT |
1668 | if (!test_opt(sb, INIT_INODE_TABLE)) |
1669 | flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED; | |
1670 | } else | |
19c5246d YY |
1671 | flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED; |
1672 | } | |
1673 | ||
feb0ab32 | 1674 | if (last_group == n_group && ext4_has_group_desc_csum(sb)) |
19c5246d YY |
1675 | /* We need to initialize block bitmap of last group. */ |
1676 | flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT; | |
1677 | ||
d77147ff | 1678 | if ((last_group == n_group) && (last != clusters_per_group - 1)) { |
1679 | group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1); | |
1680 | group_data[i - 1].free_clusters_count -= clusters_per_group - | |
1681 | last - 1; | |
19c5246d YY |
1682 | } |
1683 | ||
1684 | return 1; | |
1685 | } | |
1686 | ||
ac27a0ec DK |
1687 | /* Add group descriptor data to an existing or new group descriptor block. |
1688 | * Ensure we handle all possible error conditions _before_ we start modifying | |
1689 | * the filesystem, because we cannot abort the transaction and not have it | |
1690 | * write the data to disk. | |
1691 | * | |
1692 | * If we are on a GDT block boundary, we need to get the reserved GDT block. | |
1693 | * Otherwise, we may need to add backup GDT blocks for a sparse group. | |
1694 | * | |
1695 | * We only need to hold the superblock lock while we are actually adding | |
1696 | * in the new group's counts to the superblock. Prior to that we have | |
1697 | * not really "added" the group at all. We re-check that we are still | |
1698 | * adding in the last group in case things have changed since verifying. | |
1699 | */ | |
617ba13b | 1700 | int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) |
ac27a0ec | 1701 | { |
61f296cc | 1702 | struct ext4_new_flex_group_data flex_gd; |
617ba13b MC |
1703 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
1704 | struct ext4_super_block *es = sbi->s_es; | |
1705 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec | 1706 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; |
ac27a0ec | 1707 | struct inode *inode = NULL; |
03b40e34 | 1708 | int gdb_off; |
61f296cc YY |
1709 | int err; |
1710 | __u16 bg_flags = 0; | |
ac27a0ec | 1711 | |
617ba13b | 1712 | gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); |
ac27a0ec | 1713 | |
e2b911c5 | 1714 | if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) { |
12062ddd | 1715 | ext4_warning(sb, "Can't resize non-sparse filesystem further"); |
ac27a0ec DK |
1716 | return -EPERM; |
1717 | } | |
1718 | ||
bd81d8ee LV |
1719 | if (ext4_blocks_count(es) + input->blocks_count < |
1720 | ext4_blocks_count(es)) { | |
12062ddd | 1721 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1722 | return -EINVAL; |
1723 | } | |
1724 | ||
617ba13b | 1725 | if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < |
ac27a0ec | 1726 | le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 1727 | ext4_warning(sb, "inodes_count overflow"); |
ac27a0ec DK |
1728 | return -EINVAL; |
1729 | } | |
1730 | ||
1731 | if (reserved_gdb || gdb_off == 0) { | |
e647e291 | 1732 | if (!ext4_has_feature_resize_inode(sb) || |
e2b911c5 | 1733 | !le16_to_cpu(es->s_reserved_gdt_blocks)) { |
12062ddd | 1734 | ext4_warning(sb, |
ac27a0ec DK |
1735 | "No reserved GDT blocks, can't resize"); |
1736 | return -EPERM; | |
1737 | } | |
8a363970 | 1738 | inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL); |
1d1fe1ee | 1739 | if (IS_ERR(inode)) { |
12062ddd | 1740 | ext4_warning(sb, "Error opening resize inode"); |
1d1fe1ee | 1741 | return PTR_ERR(inode); |
ac27a0ec DK |
1742 | } |
1743 | } | |
1744 | ||
920313a7 | 1745 | |
61f296cc | 1746 | err = verify_group_input(sb, input); |
08c3a813 | 1747 | if (err) |
61f296cc | 1748 | goto out; |
ac27a0ec | 1749 | |
117fff10 TT |
1750 | err = ext4_alloc_flex_bg_array(sb, input->group + 1); |
1751 | if (err) | |
7f511862 | 1752 | goto out; |
117fff10 | 1753 | |
28623c2f TT |
1754 | err = ext4_mb_alloc_groupinfo(sb, input->group + 1); |
1755 | if (err) | |
1756 | goto out; | |
1757 | ||
61f296cc YY |
1758 | flex_gd.count = 1; |
1759 | flex_gd.groups = input; | |
1760 | flex_gd.bg_flags = &bg_flags; | |
1761 | err = ext4_flex_group_add(sb, inode, &flex_gd); | |
1762 | out: | |
ac27a0ec DK |
1763 | iput(inode); |
1764 | return err; | |
617ba13b | 1765 | } /* ext4_group_add */ |
ac27a0ec | 1766 | |
18e31438 YY |
1767 | /* |
1768 | * extend a group without checking assuming that checking has been done. | |
1769 | */ | |
1770 | static int ext4_group_extend_no_check(struct super_block *sb, | |
1771 | ext4_fsblk_t o_blocks_count, ext4_grpblk_t add) | |
1772 | { | |
1773 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
1774 | handle_t *handle; | |
1775 | int err = 0, err2; | |
1776 | ||
1777 | /* We will update the superblock, one block bitmap, and | |
1778 | * one group descriptor via ext4_group_add_blocks(). | |
1779 | */ | |
9924a92a | 1780 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3); |
18e31438 YY |
1781 | if (IS_ERR(handle)) { |
1782 | err = PTR_ERR(handle); | |
1783 | ext4_warning(sb, "error %d on journal start", err); | |
1784 | return err; | |
1785 | } | |
1786 | ||
5d601255 | 1787 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
188c299e JK |
1788 | err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh, |
1789 | EXT4_JTR_NONE); | |
18e31438 YY |
1790 | if (err) { |
1791 | ext4_warning(sb, "error %d on journal write access", err); | |
1792 | goto errout; | |
1793 | } | |
1794 | ||
05c2c00f | 1795 | lock_buffer(EXT4_SB(sb)->s_sbh); |
18e31438 | 1796 | ext4_blocks_count_set(es, o_blocks_count + add); |
636d7e2e | 1797 | ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add); |
05c2c00f JK |
1798 | ext4_superblock_csum_set(sb); |
1799 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
18e31438 YY |
1800 | ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, |
1801 | o_blocks_count + add); | |
1802 | /* We add the blocks to the bitmap and set the group need init bit */ | |
1803 | err = ext4_group_add_blocks(handle, sb, o_blocks_count, add); | |
1804 | if (err) | |
1805 | goto errout; | |
a3f5cf14 | 1806 | ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
18e31438 YY |
1807 | ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, |
1808 | o_blocks_count + add); | |
1809 | errout: | |
1810 | err2 = ext4_journal_stop(handle); | |
1811 | if (err2 && !err) | |
1812 | err = err2; | |
1813 | ||
1814 | if (!err) { | |
1815 | if (test_opt(sb, DEBUG)) | |
1816 | printk(KERN_DEBUG "EXT4-fs: extended group to %llu " | |
1817 | "blocks\n", ext4_blocks_count(es)); | |
0aeaa255 | 1818 | update_backups(sb, ext4_group_first_block_no(sb, 0), |
01f795f9 | 1819 | (char *)es, sizeof(struct ext4_super_block), 0); |
18e31438 YY |
1820 | } |
1821 | return err; | |
1822 | } | |
1823 | ||
2b2d6d01 TT |
1824 | /* |
1825 | * Extend the filesystem to the new number of blocks specified. This entry | |
ac27a0ec DK |
1826 | * point is only used to extend the current filesystem to the end of the last |
1827 | * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" | |
1828 | * for emergencies (because it has no dependencies on reserved blocks). | |
1829 | * | |
617ba13b | 1830 | * If we _really_ wanted, we could use default values to call ext4_group_add() |
ac27a0ec DK |
1831 | * allow the "remount" trick to work for arbitrary resizing, assuming enough |
1832 | * GDT blocks are reserved to grow to the desired size. | |
1833 | */ | |
617ba13b MC |
1834 | int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, |
1835 | ext4_fsblk_t n_blocks_count) | |
ac27a0ec | 1836 | { |
617ba13b | 1837 | ext4_fsblk_t o_blocks_count; |
617ba13b MC |
1838 | ext4_grpblk_t last; |
1839 | ext4_grpblk_t add; | |
af5bc92d | 1840 | struct buffer_head *bh; |
5f21b0e6 | 1841 | ext4_group_t group; |
ac27a0ec | 1842 | |
bd81d8ee | 1843 | o_blocks_count = ext4_blocks_count(es); |
ac27a0ec DK |
1844 | |
1845 | if (test_opt(sb, DEBUG)) | |
92b97816 TT |
1846 | ext4_msg(sb, KERN_DEBUG, |
1847 | "extending last group from %llu to %llu blocks", | |
1848 | o_blocks_count, n_blocks_count); | |
ac27a0ec DK |
1849 | |
1850 | if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) | |
1851 | return 0; | |
1852 | ||
1853 | if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { | |
92b97816 TT |
1854 | ext4_msg(sb, KERN_ERR, |
1855 | "filesystem too large to resize to %llu blocks safely", | |
1856 | n_blocks_count); | |
ac27a0ec DK |
1857 | return -EINVAL; |
1858 | } | |
1859 | ||
1860 | if (n_blocks_count < o_blocks_count) { | |
12062ddd | 1861 | ext4_warning(sb, "can't shrink FS - resize aborted"); |
8f82f840 | 1862 | return -EINVAL; |
ac27a0ec DK |
1863 | } |
1864 | ||
1865 | /* Handle the remaining blocks in the last group only. */ | |
5f21b0e6 | 1866 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); |
ac27a0ec DK |
1867 | |
1868 | if (last == 0) { | |
12062ddd | 1869 | ext4_warning(sb, "need to use ext2online to resize further"); |
ac27a0ec DK |
1870 | return -EPERM; |
1871 | } | |
1872 | ||
617ba13b | 1873 | add = EXT4_BLOCKS_PER_GROUP(sb) - last; |
ac27a0ec DK |
1874 | |
1875 | if (o_blocks_count + add < o_blocks_count) { | |
12062ddd | 1876 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1877 | return -EINVAL; |
1878 | } | |
1879 | ||
1880 | if (o_blocks_count + add > n_blocks_count) | |
1881 | add = n_blocks_count - o_blocks_count; | |
1882 | ||
1883 | if (o_blocks_count + add < n_blocks_count) | |
12062ddd | 1884 | ext4_warning(sb, "will only finish group (%llu blocks, %u new)", |
ac27a0ec DK |
1885 | o_blocks_count + add, add); |
1886 | ||
1887 | /* See if the device is actually as big as what was requested */ | |
0a846f49 | 1888 | bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0); |
1889 | if (IS_ERR(bh)) { | |
12062ddd | 1890 | ext4_warning(sb, "can't read last block, resize aborted"); |
ac27a0ec DK |
1891 | return -ENOSPC; |
1892 | } | |
1893 | brelse(bh); | |
1894 | ||
71df9683 | 1895 | return ext4_group_extend_no_check(sb, o_blocks_count, add); |
617ba13b | 1896 | } /* ext4_group_extend */ |
19c5246d | 1897 | |
1c6bd717 TT |
1898 | |
1899 | static int num_desc_blocks(struct super_block *sb, ext4_group_t groups) | |
1900 | { | |
1901 | return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb); | |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * Release the resize inode and drop the resize_inode feature if there | |
1906 | * are no more reserved gdt blocks, and then convert the file system | |
1907 | * to enable meta_bg | |
1908 | */ | |
1909 | static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode) | |
1910 | { | |
1911 | handle_t *handle; | |
1912 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1913 | struct ext4_super_block *es = sbi->s_es; | |
59e31c15 | 1914 | struct ext4_inode_info *ei = EXT4_I(inode); |
1c6bd717 TT |
1915 | ext4_fsblk_t nr; |
1916 | int i, ret, err = 0; | |
1917 | int credits = 1; | |
1918 | ||
1919 | ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg"); | |
59e31c15 | 1920 | if (inode) { |
1c6bd717 TT |
1921 | if (es->s_reserved_gdt_blocks) { |
1922 | ext4_error(sb, "Unexpected non-zero " | |
1923 | "s_reserved_gdt_blocks"); | |
1924 | return -EPERM; | |
1925 | } | |
1c6bd717 TT |
1926 | |
1927 | /* Do a quick sanity check of the resize inode */ | |
d77147ff | 1928 | if (inode->i_blocks != 1 << (inode->i_blkbits - |
1929 | (9 - sbi->s_cluster_bits))) | |
1c6bd717 TT |
1930 | goto invalid_resize_inode; |
1931 | for (i = 0; i < EXT4_N_BLOCKS; i++) { | |
1932 | if (i == EXT4_DIND_BLOCK) { | |
1933 | if (ei->i_data[i]) | |
1934 | continue; | |
1935 | else | |
1936 | goto invalid_resize_inode; | |
1937 | } | |
1938 | if (ei->i_data[i]) | |
1939 | goto invalid_resize_inode; | |
1940 | } | |
1941 | credits += 3; /* block bitmap, bg descriptor, resize inode */ | |
1942 | } | |
1943 | ||
9924a92a | 1944 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits); |
1c6bd717 TT |
1945 | if (IS_ERR(handle)) |
1946 | return PTR_ERR(handle); | |
1947 | ||
5d601255 | 1948 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
188c299e JK |
1949 | err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh, |
1950 | EXT4_JTR_NONE); | |
1c6bd717 TT |
1951 | if (err) |
1952 | goto errout; | |
1953 | ||
05c2c00f | 1954 | lock_buffer(sbi->s_sbh); |
e2b911c5 DW |
1955 | ext4_clear_feature_resize_inode(sb); |
1956 | ext4_set_feature_meta_bg(sb); | |
1c6bd717 TT |
1957 | sbi->s_es->s_first_meta_bg = |
1958 | cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count)); | |
05c2c00f JK |
1959 | ext4_superblock_csum_set(sb); |
1960 | unlock_buffer(sbi->s_sbh); | |
1c6bd717 | 1961 | |
a3f5cf14 | 1962 | err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); |
1c6bd717 TT |
1963 | if (err) { |
1964 | ext4_std_error(sb, err); | |
1965 | goto errout; | |
1966 | } | |
1967 | ||
1968 | if (inode) { | |
1969 | nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]); | |
1970 | ext4_free_blocks(handle, inode, NULL, nr, 1, | |
1971 | EXT4_FREE_BLOCKS_METADATA | | |
1972 | EXT4_FREE_BLOCKS_FORGET); | |
1973 | ei->i_data[EXT4_DIND_BLOCK] = 0; | |
1974 | inode->i_blocks = 0; | |
1975 | ||
1976 | err = ext4_mark_inode_dirty(handle, inode); | |
1977 | if (err) | |
1978 | ext4_std_error(sb, err); | |
1979 | } | |
1980 | ||
1981 | errout: | |
1982 | ret = ext4_journal_stop(handle); | |
48f15515 | 1983 | return err ? err : ret; |
1c6bd717 TT |
1984 | |
1985 | invalid_resize_inode: | |
1986 | ext4_error(sb, "corrupted/inconsistent resize inode"); | |
1987 | return -EINVAL; | |
1988 | } | |
1989 | ||
19c5246d YY |
1990 | /* |
1991 | * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count | |
1992 | * | |
1993 | * @sb: super block of the fs to be resized | |
1994 | * @n_blocks_count: the number of blocks resides in the resized fs | |
1995 | */ | |
1996 | int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) | |
1997 | { | |
1998 | struct ext4_new_flex_group_data *flex_gd = NULL; | |
1999 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2000 | struct ext4_super_block *es = sbi->s_es; | |
2001 | struct buffer_head *bh; | |
01f795f9 YY |
2002 | struct inode *resize_inode = NULL; |
2003 | ext4_grpblk_t add, offset; | |
19c5246d YY |
2004 | unsigned long n_desc_blocks; |
2005 | unsigned long o_desc_blocks; | |
01f795f9 YY |
2006 | ext4_group_t o_group; |
2007 | ext4_group_t n_group; | |
2008 | ext4_fsblk_t o_blocks_count; | |
1c6bd717 | 2009 | ext4_fsblk_t n_blocks_count_retry = 0; |
4da4a56e | 2010 | unsigned long last_update_time = 0; |
658a5234 | 2011 | int err = 0; |
01f795f9 | 2012 | int meta_bg; |
658a5234 | 2013 | unsigned int flexbg_size = ext4_flex_bg_size(sbi); |
19c5246d | 2014 | |
59e31c15 | 2015 | /* See if the device is actually as big as what was requested */ |
0a846f49 | 2016 | bh = ext4_sb_bread(sb, n_blocks_count - 1, 0); |
2017 | if (IS_ERR(bh)) { | |
59e31c15 TT |
2018 | ext4_warning(sb, "can't read last block, resize aborted"); |
2019 | return -ENOSPC; | |
2020 | } | |
2021 | brelse(bh); | |
2022 | ||
69cb8e9d KO |
2023 | /* |
2024 | * For bigalloc, trim the requested size to the nearest cluster | |
2025 | * boundary to avoid creating an unusable filesystem. We do this | |
2026 | * silently, instead of returning an error, to avoid breaking | |
2027 | * callers that blindly resize the filesystem to the full size of | |
2028 | * the underlying block device. | |
2029 | */ | |
2030 | if (ext4_has_feature_bigalloc(sb)) | |
2031 | n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1); | |
2032 | ||
1c6bd717 | 2033 | retry: |
19c5246d YY |
2034 | o_blocks_count = ext4_blocks_count(es); |
2035 | ||
59e31c15 TT |
2036 | ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu " |
2037 | "to %llu blocks", o_blocks_count, n_blocks_count); | |
19c5246d YY |
2038 | |
2039 | if (n_blocks_count < o_blocks_count) { | |
2040 | /* On-line shrinking not supported */ | |
2041 | ext4_warning(sb, "can't shrink FS - resize aborted"); | |
2042 | return -EINVAL; | |
2043 | } | |
2044 | ||
2045 | if (n_blocks_count == o_blocks_count) | |
2046 | /* Nothing need to do */ | |
2047 | return 0; | |
2048 | ||
bd86298e | 2049 | n_group = ext4_get_group_number(sb, n_blocks_count - 1); |
4f2f76f7 | 2050 | if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) { |
3f8a6411 TT |
2051 | ext4_warning(sb, "resize would cause inodes_count overflow"); |
2052 | return -EINVAL; | |
2053 | } | |
a0ade1de | 2054 | ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset); |
19c5246d | 2055 | |
1c6bd717 TT |
2056 | n_desc_blocks = num_desc_blocks(sb, n_group + 1); |
2057 | o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count); | |
19c5246d | 2058 | |
e2b911c5 | 2059 | meta_bg = ext4_has_feature_meta_bg(sb); |
19c5246d | 2060 | |
e2b911c5 | 2061 | if (ext4_has_feature_resize_inode(sb)) { |
01f795f9 YY |
2062 | if (meta_bg) { |
2063 | ext4_error(sb, "resize_inode and meta_bg enabled " | |
2064 | "simultaneously"); | |
2065 | return -EINVAL; | |
2066 | } | |
1c6bd717 TT |
2067 | if (n_desc_blocks > o_desc_blocks + |
2068 | le16_to_cpu(es->s_reserved_gdt_blocks)) { | |
2069 | n_blocks_count_retry = n_blocks_count; | |
2070 | n_desc_blocks = o_desc_blocks + | |
2071 | le16_to_cpu(es->s_reserved_gdt_blocks); | |
2072 | n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb); | |
aec51758 | 2073 | n_blocks_count = (ext4_fsblk_t)n_group * |
f96c3ac8 JK |
2074 | EXT4_BLOCKS_PER_GROUP(sb) + |
2075 | le32_to_cpu(es->s_first_data_block); | |
1c6bd717 | 2076 | n_group--; /* set to last group number */ |
01f795f9 | 2077 | } |
1c6bd717 TT |
2078 | |
2079 | if (!resize_inode) | |
8a363970 TT |
2080 | resize_inode = ext4_iget(sb, EXT4_RESIZE_INO, |
2081 | EXT4_IGET_SPECIAL); | |
01f795f9 YY |
2082 | if (IS_ERR(resize_inode)) { |
2083 | ext4_warning(sb, "Error opening resize inode"); | |
2084 | return PTR_ERR(resize_inode); | |
2085 | } | |
1c6bd717 TT |
2086 | } |
2087 | ||
07be778c | 2088 | if ((!resize_inode && !meta_bg && n_desc_blocks > o_desc_blocks) || n_blocks_count == o_blocks_count) { |
1c6bd717 TT |
2089 | err = ext4_convert_meta_bg(sb, resize_inode); |
2090 | if (err) | |
2091 | goto out; | |
2092 | if (resize_inode) { | |
2093 | iput(resize_inode); | |
2094 | resize_inode = NULL; | |
2095 | } | |
2096 | if (n_blocks_count_retry) { | |
2097 | n_blocks_count = n_blocks_count_retry; | |
2098 | n_blocks_count_retry = 0; | |
2099 | goto retry; | |
2100 | } | |
19c5246d YY |
2101 | } |
2102 | ||
f0a459de TT |
2103 | /* |
2104 | * Make sure the last group has enough space so that it's | |
2105 | * guaranteed to have enough space for all metadata blocks | |
2106 | * that it might need to hold. (We might not need to store | |
2107 | * the inode table blocks in the last block group, but there | |
2108 | * will be cases where this might be needed.) | |
2109 | */ | |
2110 | if ((ext4_group_first_block_no(sb, n_group) + | |
2111 | ext4_group_overhead_blocks(sb, n_group) + 2 + | |
2112 | sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) { | |
2113 | n_blocks_count = ext4_group_first_block_no(sb, n_group); | |
2114 | n_group--; | |
2115 | n_blocks_count_retry = 0; | |
2116 | if (resize_inode) { | |
2117 | iput(resize_inode); | |
2118 | resize_inode = NULL; | |
2119 | } | |
2120 | goto retry; | |
2121 | } | |
2122 | ||
a0ade1de LC |
2123 | /* extend the last group */ |
2124 | if (n_group == o_group) | |
2125 | add = n_blocks_count - o_blocks_count; | |
2126 | else | |
d77147ff | 2127 | add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1)); |
a0ade1de | 2128 | if (add > 0) { |
19c5246d YY |
2129 | err = ext4_group_extend_no_check(sb, o_blocks_count, add); |
2130 | if (err) | |
2131 | goto out; | |
2132 | } | |
2133 | ||
df3cb754 | 2134 | if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0) |
19c5246d YY |
2135 | goto out; |
2136 | ||
117fff10 TT |
2137 | err = ext4_alloc_flex_bg_array(sb, n_group + 1); |
2138 | if (err) | |
f348e224 | 2139 | goto out; |
117fff10 | 2140 | |
28623c2f TT |
2141 | err = ext4_mb_alloc_groupinfo(sb, n_group + 1); |
2142 | if (err) | |
2143 | goto out; | |
2144 | ||
665d3e0a | 2145 | flex_gd = alloc_flex_gd(flexbg_size, o_group, n_group); |
19c5246d YY |
2146 | if (flex_gd == NULL) { |
2147 | err = -ENOMEM; | |
2148 | goto out; | |
2149 | } | |
2150 | ||
2151 | /* Add flex groups. Note that a regular group is a | |
2152 | * flex group with 1 group. | |
2153 | */ | |
5d1935ac | 2154 | while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) { |
a861fb9f | 2155 | if (time_is_before_jiffies(last_update_time + HZ * 10)) { |
4da4a56e TT |
2156 | if (last_update_time) |
2157 | ext4_msg(sb, KERN_INFO, | |
2158 | "resized to %llu blocks", | |
2159 | ext4_blocks_count(es)); | |
2160 | last_update_time = jiffies; | |
2161 | } | |
03c1c290 YY |
2162 | if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0) |
2163 | break; | |
19c5246d YY |
2164 | err = ext4_flex_group_add(sb, resize_inode, flex_gd); |
2165 | if (unlikely(err)) | |
2166 | break; | |
2167 | } | |
2168 | ||
1c6bd717 TT |
2169 | if (!err && n_blocks_count_retry) { |
2170 | n_blocks_count = n_blocks_count_retry; | |
2171 | n_blocks_count_retry = 0; | |
2172 | free_flex_gd(flex_gd); | |
2173 | flex_gd = NULL; | |
db6aee62 VA |
2174 | if (resize_inode) { |
2175 | iput(resize_inode); | |
2176 | resize_inode = NULL; | |
2177 | } | |
1c6bd717 TT |
2178 | goto retry; |
2179 | } | |
2180 | ||
19c5246d YY |
2181 | out: |
2182 | if (flex_gd) | |
2183 | free_flex_gd(flex_gd); | |
01f795f9 YY |
2184 | if (resize_inode != NULL) |
2185 | iput(resize_inode); | |
6c732840 LC |
2186 | if (err) |
2187 | ext4_warning(sb, "error (%d) occurred during " | |
2188 | "file system resize", err); | |
2189 | ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", | |
2190 | ext4_blocks_count(es)); | |
19c5246d YY |
2191 | return err; |
2192 | } |