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 | ||
617ba13b | 13 | #define EXT4FS_DEBUG |
ac27a0ec | 14 | |
ac27a0ec DK |
15 | #include <linux/errno.h> |
16 | #include <linux/slab.h> | |
17 | ||
3dcf5451 | 18 | #include "ext4_jbd2.h" |
ac27a0ec | 19 | |
8f82f840 YY |
20 | int ext4_resize_begin(struct super_block *sb) |
21 | { | |
ba48e66e | 22 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8f82f840 YY |
23 | int ret = 0; |
24 | ||
25 | if (!capable(CAP_SYS_RESOURCE)) | |
26 | return -EPERM; | |
27 | ||
011fa994 TT |
28 | /* |
29 | * If we are not using the primary superblock/GDT copy don't resize, | |
30 | * because the user tools have no way of handling this. Probably a | |
31 | * bad time to do it anyways. | |
32 | */ | |
ba48e66e | 33 | if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) != |
011fa994 TT |
34 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) { |
35 | ext4_warning(sb, "won't resize using backup superblock at %llu", | |
36 | (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr); | |
37 | return -EPERM; | |
38 | } | |
39 | ||
ce723c31 YY |
40 | /* |
41 | * We are not allowed to do online-resizing on a filesystem mounted | |
42 | * with error, because it can destroy the filesystem easily. | |
43 | */ | |
44 | if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { | |
45 | ext4_warning(sb, "There are errors in the filesystem, " | |
8d2ae1cb | 46 | "so online resizing is not allowed"); |
ce723c31 YY |
47 | return -EPERM; |
48 | } | |
49 | ||
9549a168 TT |
50 | if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING, |
51 | &EXT4_SB(sb)->s_ext4_flags)) | |
8f82f840 YY |
52 | ret = -EBUSY; |
53 | ||
54 | return ret; | |
55 | } | |
56 | ||
57 | void ext4_resize_end(struct super_block *sb) | |
58 | { | |
9549a168 | 59 | clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags); |
4e857c58 | 60 | smp_mb__after_atomic(); |
8f82f840 YY |
61 | } |
62 | ||
01f795f9 YY |
63 | static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb, |
64 | ext4_group_t group) { | |
65 | return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) << | |
66 | EXT4_DESC_PER_BLOCK_BITS(sb); | |
67 | } | |
68 | ||
69 | static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb, | |
70 | ext4_group_t group) { | |
71 | group = ext4_meta_bg_first_group(sb, group); | |
72 | return ext4_group_first_block_no(sb, group); | |
73 | } | |
74 | ||
75 | static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb, | |
76 | ext4_group_t group) { | |
77 | ext4_grpblk_t overhead; | |
78 | overhead = ext4_bg_num_gdb(sb, group); | |
79 | if (ext4_bg_has_super(sb, group)) | |
80 | overhead += 1 + | |
81 | le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); | |
82 | return overhead; | |
83 | } | |
84 | ||
ac27a0ec DK |
85 | #define outside(b, first, last) ((b) < (first) || (b) >= (last)) |
86 | #define inside(b, first, last) ((b) >= (first) && (b) < (last)) | |
87 | ||
88 | static int verify_group_input(struct super_block *sb, | |
617ba13b | 89 | struct ext4_new_group_data *input) |
ac27a0ec | 90 | { |
617ba13b MC |
91 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
92 | struct ext4_super_block *es = sbi->s_es; | |
bd81d8ee | 93 | ext4_fsblk_t start = ext4_blocks_count(es); |
617ba13b | 94 | ext4_fsblk_t end = start + input->blocks_count; |
fd2d4291 | 95 | ext4_group_t group = input->group; |
617ba13b | 96 | ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group; |
b302ef2d TT |
97 | unsigned overhead; |
98 | ext4_fsblk_t metaend; | |
ac27a0ec | 99 | struct buffer_head *bh = NULL; |
3a5b2ecd | 100 | ext4_grpblk_t free_blocks_count, offset; |
ac27a0ec DK |
101 | int err = -EINVAL; |
102 | ||
b302ef2d TT |
103 | if (group != sbi->s_groups_count) { |
104 | ext4_warning(sb, "Cannot add at group %u (only %u groups)", | |
105 | input->group, sbi->s_groups_count); | |
106 | return -EINVAL; | |
107 | } | |
108 | ||
109 | overhead = ext4_group_overhead_blocks(sb, group); | |
110 | metaend = start + overhead; | |
ac27a0ec DK |
111 | input->free_blocks_count = free_blocks_count = |
112 | input->blocks_count - 2 - overhead - sbi->s_itb_per_group; | |
113 | ||
114 | if (test_opt(sb, DEBUG)) | |
617ba13b | 115 | printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks " |
ac27a0ec | 116 | "(%d free, %u reserved)\n", |
617ba13b | 117 | ext4_bg_has_super(sb, input->group) ? "normal" : |
ac27a0ec DK |
118 | "no-super", input->group, input->blocks_count, |
119 | free_blocks_count, input->reserved_blocks); | |
120 | ||
3a5b2ecd | 121 | ext4_get_group_no_and_offset(sb, start, NULL, &offset); |
b302ef2d | 122 | if (offset != 0) |
12062ddd | 123 | ext4_warning(sb, "Last group not full"); |
ac27a0ec | 124 | else if (input->reserved_blocks > input->blocks_count / 5) |
12062ddd | 125 | ext4_warning(sb, "Reserved blocks too high (%u)", |
ac27a0ec DK |
126 | input->reserved_blocks); |
127 | else if (free_blocks_count < 0) | |
12062ddd | 128 | ext4_warning(sb, "Bad blocks count %u", |
ac27a0ec | 129 | input->blocks_count); |
9da1f6d0 TT |
130 | else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) { |
131 | err = PTR_ERR(bh); | |
132 | bh = NULL; | |
12062ddd | 133 | ext4_warning(sb, "Cannot read last block (%llu)", |
ac27a0ec | 134 | end - 1); |
9da1f6d0 | 135 | } else if (outside(input->block_bitmap, start, end)) |
12062ddd | 136 | ext4_warning(sb, "Block bitmap not in group (block %llu)", |
1939e49a | 137 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 138 | else if (outside(input->inode_bitmap, start, end)) |
12062ddd | 139 | ext4_warning(sb, "Inode bitmap not in group (block %llu)", |
1939e49a | 140 | (unsigned long long)input->inode_bitmap); |
ac27a0ec | 141 | else if (outside(input->inode_table, start, end) || |
2b2d6d01 | 142 | outside(itend - 1, start, end)) |
12062ddd | 143 | ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)", |
1939e49a | 144 | (unsigned long long)input->inode_table, itend - 1); |
ac27a0ec | 145 | else if (input->inode_bitmap == input->block_bitmap) |
12062ddd | 146 | ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)", |
1939e49a | 147 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 148 | else if (inside(input->block_bitmap, input->inode_table, itend)) |
12062ddd ES |
149 | ext4_warning(sb, "Block bitmap (%llu) in inode table " |
150 | "(%llu-%llu)", | |
1939e49a RD |
151 | (unsigned long long)input->block_bitmap, |
152 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 153 | else if (inside(input->inode_bitmap, input->inode_table, itend)) |
12062ddd ES |
154 | ext4_warning(sb, "Inode bitmap (%llu) in inode table " |
155 | "(%llu-%llu)", | |
1939e49a RD |
156 | (unsigned long long)input->inode_bitmap, |
157 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 158 | else if (inside(input->block_bitmap, start, metaend)) |
12062ddd | 159 | ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
160 | (unsigned long long)input->block_bitmap, |
161 | start, metaend - 1); | |
ac27a0ec | 162 | else if (inside(input->inode_bitmap, start, metaend)) |
12062ddd | 163 | ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
164 | (unsigned long long)input->inode_bitmap, |
165 | start, metaend - 1); | |
ac27a0ec | 166 | else if (inside(input->inode_table, start, metaend) || |
2b2d6d01 | 167 | inside(itend - 1, start, metaend)) |
12062ddd ES |
168 | ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table " |
169 | "(%llu-%llu)", | |
1939e49a RD |
170 | (unsigned long long)input->inode_table, |
171 | itend - 1, start, metaend - 1); | |
ac27a0ec DK |
172 | else |
173 | err = 0; | |
174 | brelse(bh); | |
175 | ||
176 | return err; | |
177 | } | |
178 | ||
28c7bac0 YY |
179 | /* |
180 | * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex | |
181 | * group each time. | |
182 | */ | |
183 | struct ext4_new_flex_group_data { | |
184 | struct ext4_new_group_data *groups; /* new_group_data for groups | |
185 | in the flex group */ | |
186 | __u16 *bg_flags; /* block group flags of groups | |
187 | in @groups */ | |
188 | ext4_group_t count; /* number of groups in @groups | |
189 | */ | |
190 | }; | |
191 | ||
192 | /* | |
193 | * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of | |
194 | * @flexbg_size. | |
195 | * | |
196 | * Returns NULL on failure otherwise address of the allocated structure. | |
197 | */ | |
198 | static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size) | |
199 | { | |
200 | struct ext4_new_flex_group_data *flex_gd; | |
201 | ||
202 | flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS); | |
203 | if (flex_gd == NULL) | |
204 | goto out3; | |
205 | ||
46901760 | 206 | if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data)) |
967ac8af | 207 | goto out2; |
28c7bac0 YY |
208 | flex_gd->count = flexbg_size; |
209 | ||
210 | flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) * | |
211 | flexbg_size, GFP_NOFS); | |
212 | if (flex_gd->groups == NULL) | |
213 | goto out2; | |
214 | ||
215 | flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS); | |
216 | if (flex_gd->bg_flags == NULL) | |
217 | goto out1; | |
218 | ||
219 | return flex_gd; | |
220 | ||
221 | out1: | |
222 | kfree(flex_gd->groups); | |
223 | out2: | |
224 | kfree(flex_gd); | |
225 | out3: | |
226 | return NULL; | |
227 | } | |
228 | ||
229 | static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd) | |
230 | { | |
231 | kfree(flex_gd->bg_flags); | |
232 | kfree(flex_gd->groups); | |
233 | kfree(flex_gd); | |
234 | } | |
235 | ||
3fbea4b3 YY |
236 | /* |
237 | * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps | |
238 | * and inode tables for a flex group. | |
239 | * | |
240 | * This function is used by 64bit-resize. Note that this function allocates | |
241 | * group tables from the 1st group of groups contained by @flexgd, which may | |
242 | * be a partial of a flex group. | |
243 | * | |
244 | * @sb: super block of fs to which the groups belongs | |
03c1c290 YY |
245 | * |
246 | * Returns 0 on a successful allocation of the metadata blocks in the | |
247 | * block group. | |
3fbea4b3 | 248 | */ |
03c1c290 | 249 | static int ext4_alloc_group_tables(struct super_block *sb, |
3fbea4b3 YY |
250 | struct ext4_new_flex_group_data *flex_gd, |
251 | int flexbg_size) | |
252 | { | |
253 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
3fbea4b3 YY |
254 | ext4_fsblk_t start_blk; |
255 | ext4_fsblk_t last_blk; | |
256 | ext4_group_t src_group; | |
257 | ext4_group_t bb_index = 0; | |
258 | ext4_group_t ib_index = 0; | |
259 | ext4_group_t it_index = 0; | |
260 | ext4_group_t group; | |
261 | ext4_group_t last_group; | |
262 | unsigned overhead; | |
b93c9535 | 263 | __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0; |
3fbea4b3 YY |
264 | |
265 | BUG_ON(flex_gd->count == 0 || group_data == NULL); | |
266 | ||
267 | src_group = group_data[0].group; | |
268 | last_group = src_group + flex_gd->count - 1; | |
269 | ||
270 | BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) != | |
271 | (last_group & ~(flexbg_size - 1)))); | |
272 | next_group: | |
273 | group = group_data[0].group; | |
03c1c290 YY |
274 | if (src_group >= group_data[0].group + flex_gd->count) |
275 | return -ENOSPC; | |
3fbea4b3 YY |
276 | start_blk = ext4_group_first_block_no(sb, src_group); |
277 | last_blk = start_blk + group_data[src_group - group].blocks_count; | |
278 | ||
01f795f9 | 279 | overhead = ext4_group_overhead_blocks(sb, src_group); |
3fbea4b3 YY |
280 | |
281 | start_blk += overhead; | |
282 | ||
3fbea4b3 YY |
283 | /* We collect contiguous blocks as much as possible. */ |
284 | src_group++; | |
01f795f9 YY |
285 | for (; src_group <= last_group; src_group++) { |
286 | overhead = ext4_group_overhead_blocks(sb, src_group); | |
b93c9535 | 287 | if (overhead == 0) |
3fbea4b3 YY |
288 | last_blk += group_data[src_group - group].blocks_count; |
289 | else | |
290 | break; | |
01f795f9 | 291 | } |
3fbea4b3 YY |
292 | |
293 | /* Allocate block bitmaps */ | |
294 | for (; bb_index < flex_gd->count; bb_index++) { | |
295 | if (start_blk >= last_blk) | |
296 | goto next_group; | |
297 | group_data[bb_index].block_bitmap = start_blk++; | |
bd86298e | 298 | group = ext4_get_group_number(sb, start_blk - 1); |
3fbea4b3 YY |
299 | group -= group_data[0].group; |
300 | group_data[group].free_blocks_count--; | |
b93c9535 | 301 | flex_gd->bg_flags[group] &= uninit_mask; |
3fbea4b3 YY |
302 | } |
303 | ||
304 | /* Allocate inode bitmaps */ | |
305 | for (; ib_index < flex_gd->count; ib_index++) { | |
306 | if (start_blk >= last_blk) | |
307 | goto next_group; | |
308 | group_data[ib_index].inode_bitmap = start_blk++; | |
bd86298e | 309 | group = ext4_get_group_number(sb, start_blk - 1); |
3fbea4b3 YY |
310 | group -= group_data[0].group; |
311 | group_data[group].free_blocks_count--; | |
b93c9535 | 312 | flex_gd->bg_flags[group] &= uninit_mask; |
3fbea4b3 YY |
313 | } |
314 | ||
315 | /* Allocate inode tables */ | |
316 | for (; it_index < flex_gd->count; it_index++) { | |
b93c9535 TT |
317 | unsigned int itb = EXT4_SB(sb)->s_itb_per_group; |
318 | ext4_fsblk_t next_group_start; | |
319 | ||
320 | if (start_blk + itb > last_blk) | |
3fbea4b3 YY |
321 | goto next_group; |
322 | group_data[it_index].inode_table = start_blk; | |
b93c9535 TT |
323 | group = ext4_get_group_number(sb, start_blk); |
324 | next_group_start = ext4_group_first_block_no(sb, group + 1); | |
3fbea4b3 | 325 | group -= group_data[0].group; |
3fbea4b3 | 326 | |
b93c9535 TT |
327 | if (start_blk + itb > next_group_start) { |
328 | flex_gd->bg_flags[group + 1] &= uninit_mask; | |
329 | overhead = start_blk + itb - next_group_start; | |
330 | group_data[group + 1].free_blocks_count -= overhead; | |
331 | itb -= overhead; | |
332 | } | |
333 | ||
334 | group_data[group].free_blocks_count -= itb; | |
335 | flex_gd->bg_flags[group] &= uninit_mask; | |
3fbea4b3 YY |
336 | start_blk += EXT4_SB(sb)->s_itb_per_group; |
337 | } | |
338 | ||
339 | if (test_opt(sb, DEBUG)) { | |
340 | int i; | |
341 | group = group_data[0].group; | |
342 | ||
343 | printk(KERN_DEBUG "EXT4-fs: adding a flex group with " | |
344 | "%d groups, flexbg size is %d:\n", flex_gd->count, | |
345 | flexbg_size); | |
346 | ||
347 | for (i = 0; i < flex_gd->count; i++) { | |
348 | printk(KERN_DEBUG "adding %s group %u: %u " | |
349 | "blocks (%d free)\n", | |
350 | ext4_bg_has_super(sb, group + i) ? "normal" : | |
351 | "no-super", group + i, | |
352 | group_data[i].blocks_count, | |
353 | group_data[i].free_blocks_count); | |
354 | } | |
355 | } | |
03c1c290 | 356 | return 0; |
3fbea4b3 YY |
357 | } |
358 | ||
ac27a0ec | 359 | static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, |
617ba13b | 360 | ext4_fsblk_t blk) |
ac27a0ec DK |
361 | { |
362 | struct buffer_head *bh; | |
363 | int err; | |
364 | ||
365 | bh = sb_getblk(sb, blk); | |
aebf0243 | 366 | if (unlikely(!bh)) |
860d21e2 | 367 | return ERR_PTR(-ENOMEM); |
5d601255 | 368 | BUFFER_TRACE(bh, "get_write_access"); |
617ba13b | 369 | if ((err = ext4_journal_get_write_access(handle, bh))) { |
ac27a0ec DK |
370 | brelse(bh); |
371 | bh = ERR_PTR(err); | |
372 | } else { | |
ac27a0ec DK |
373 | memset(bh->b_data, 0, sb->s_blocksize); |
374 | set_buffer_uptodate(bh); | |
ac27a0ec DK |
375 | } |
376 | ||
377 | return bh; | |
378 | } | |
379 | ||
14904107 ES |
380 | /* |
381 | * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA. | |
382 | * If that fails, restart the transaction & regain write access for the | |
383 | * buffer head which is used for block_bitmap modifications. | |
384 | */ | |
6d40bc5a | 385 | static int extend_or_restart_transaction(handle_t *handle, int thresh) |
14904107 ES |
386 | { |
387 | int err; | |
388 | ||
0390131b | 389 | if (ext4_handle_has_enough_credits(handle, thresh)) |
14904107 ES |
390 | return 0; |
391 | ||
392 | err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA); | |
393 | if (err < 0) | |
394 | return err; | |
395 | if (err) { | |
6d40bc5a YY |
396 | err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA); |
397 | if (err) | |
14904107 | 398 | return err; |
2b2d6d01 | 399 | } |
14904107 ES |
400 | |
401 | return 0; | |
402 | } | |
403 | ||
33afdcc5 YY |
404 | /* |
405 | * set_flexbg_block_bitmap() mark @count blocks starting from @block used. | |
406 | * | |
407 | * Helper function for ext4_setup_new_group_blocks() which set . | |
408 | * | |
409 | * @sb: super block | |
410 | * @handle: journal handle | |
411 | * @flex_gd: flex group data | |
412 | */ | |
413 | static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle, | |
414 | struct ext4_new_flex_group_data *flex_gd, | |
415 | ext4_fsblk_t block, ext4_group_t count) | |
416 | { | |
417 | ext4_group_t count2; | |
418 | ||
419 | ext4_debug("mark blocks [%llu/%u] used\n", block, count); | |
420 | for (count2 = count; count > 0; count -= count2, block += count2) { | |
421 | ext4_fsblk_t start; | |
422 | struct buffer_head *bh; | |
423 | ext4_group_t group; | |
424 | int err; | |
425 | ||
bd86298e | 426 | group = ext4_get_group_number(sb, block); |
33afdcc5 YY |
427 | start = ext4_group_first_block_no(sb, group); |
428 | group -= flex_gd->groups[0].group; | |
429 | ||
3d2660d0 | 430 | count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start); |
33afdcc5 YY |
431 | if (count2 > count) |
432 | count2 = count; | |
433 | ||
434 | if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) { | |
435 | BUG_ON(flex_gd->count > 1); | |
436 | continue; | |
437 | } | |
438 | ||
439 | err = extend_or_restart_transaction(handle, 1); | |
440 | if (err) | |
441 | return err; | |
442 | ||
443 | bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap); | |
aebf0243 | 444 | if (unlikely(!bh)) |
860d21e2 | 445 | return -ENOMEM; |
33afdcc5 | 446 | |
5d601255 | 447 | BUFFER_TRACE(bh, "get_write_access"); |
33afdcc5 | 448 | err = ext4_journal_get_write_access(handle, bh); |
feddbc01 VA |
449 | if (err) { |
450 | brelse(bh); | |
33afdcc5 | 451 | return err; |
feddbc01 | 452 | } |
33afdcc5 YY |
453 | ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block, |
454 | block - start, count2); | |
455 | ext4_set_bits(bh->b_data, block - start, count2); | |
456 | ||
457 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
feddbc01 | 458 | brelse(bh); |
33afdcc5 YY |
459 | if (unlikely(err)) |
460 | return err; | |
33afdcc5 YY |
461 | } |
462 | ||
463 | return 0; | |
464 | } | |
465 | ||
466 | /* | |
467 | * Set up the block and inode bitmaps, and the inode table for the new groups. | |
468 | * This doesn't need to be part of the main transaction, since we are only | |
469 | * changing blocks outside the actual filesystem. We still do journaling to | |
470 | * ensure the recovery is correct in case of a failure just after resize. | |
471 | * If any part of this fails, we simply abort the resize. | |
472 | * | |
473 | * setup_new_flex_group_blocks handles a flex group as follow: | |
474 | * 1. copy super block and GDT, and initialize group tables if necessary. | |
475 | * In this step, we only set bits in blocks bitmaps for blocks taken by | |
476 | * super block and GDT. | |
477 | * 2. allocate group tables in block bitmaps, that is, set bits in block | |
478 | * bitmap for blocks taken by group tables. | |
479 | */ | |
480 | static int setup_new_flex_group_blocks(struct super_block *sb, | |
481 | struct ext4_new_flex_group_data *flex_gd) | |
482 | { | |
483 | int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group}; | |
484 | ext4_fsblk_t start; | |
485 | ext4_fsblk_t block; | |
486 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
487 | struct ext4_super_block *es = sbi->s_es; | |
488 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
489 | __u16 *bg_flags = flex_gd->bg_flags; | |
490 | handle_t *handle; | |
491 | ext4_group_t group, count; | |
492 | struct buffer_head *bh = NULL; | |
493 | int reserved_gdb, i, j, err = 0, err2; | |
01f795f9 | 494 | int meta_bg; |
33afdcc5 YY |
495 | |
496 | BUG_ON(!flex_gd->count || !group_data || | |
497 | group_data[0].group != sbi->s_groups_count); | |
498 | ||
499 | reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); | |
e2b911c5 | 500 | meta_bg = ext4_has_feature_meta_bg(sb); |
33afdcc5 YY |
501 | |
502 | /* This transaction may be extended/restarted along the way */ | |
9924a92a | 503 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); |
33afdcc5 YY |
504 | if (IS_ERR(handle)) |
505 | return PTR_ERR(handle); | |
506 | ||
507 | group = group_data[0].group; | |
508 | for (i = 0; i < flex_gd->count; i++, group++) { | |
509 | unsigned long gdblocks; | |
01f795f9 | 510 | ext4_grpblk_t overhead; |
33afdcc5 YY |
511 | |
512 | gdblocks = ext4_bg_num_gdb(sb, group); | |
513 | start = ext4_group_first_block_no(sb, group); | |
514 | ||
01f795f9 | 515 | if (meta_bg == 0 && !ext4_bg_has_super(sb, group)) |
6df935ad YY |
516 | goto handle_itb; |
517 | ||
01f795f9 YY |
518 | if (meta_bg == 1) { |
519 | ext4_group_t first_group; | |
520 | first_group = ext4_meta_bg_first_group(sb, group); | |
521 | if (first_group != group + 1 && | |
522 | first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1) | |
523 | goto handle_itb; | |
524 | } | |
525 | ||
526 | block = start + ext4_bg_has_super(sb, group); | |
33afdcc5 | 527 | /* Copy all of the GDT blocks into the backup in this group */ |
01f795f9 | 528 | for (j = 0; j < gdblocks; j++, block++) { |
33afdcc5 YY |
529 | struct buffer_head *gdb; |
530 | ||
531 | ext4_debug("update backup group %#04llx\n", block); | |
532 | err = extend_or_restart_transaction(handle, 1); | |
533 | if (err) | |
534 | goto out; | |
535 | ||
536 | gdb = sb_getblk(sb, block); | |
aebf0243 | 537 | if (unlikely(!gdb)) { |
860d21e2 | 538 | err = -ENOMEM; |
33afdcc5 YY |
539 | goto out; |
540 | } | |
541 | ||
5d601255 | 542 | BUFFER_TRACE(gdb, "get_write_access"); |
33afdcc5 YY |
543 | err = ext4_journal_get_write_access(handle, gdb); |
544 | if (err) { | |
545 | brelse(gdb); | |
546 | goto out; | |
547 | } | |
548 | memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data, | |
549 | gdb->b_size); | |
550 | set_buffer_uptodate(gdb); | |
551 | ||
552 | err = ext4_handle_dirty_metadata(handle, NULL, gdb); | |
553 | if (unlikely(err)) { | |
554 | brelse(gdb); | |
555 | goto out; | |
556 | } | |
557 | brelse(gdb); | |
558 | } | |
559 | ||
560 | /* Zero out all of the reserved backup group descriptor | |
561 | * table blocks | |
562 | */ | |
563 | if (ext4_bg_has_super(sb, group)) { | |
564 | err = sb_issue_zeroout(sb, gdblocks + start + 1, | |
565 | reserved_gdb, GFP_NOFS); | |
566 | if (err) | |
567 | goto out; | |
568 | } | |
569 | ||
6df935ad | 570 | handle_itb: |
33afdcc5 YY |
571 | /* Initialize group tables of the grop @group */ |
572 | if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED)) | |
573 | goto handle_bb; | |
574 | ||
575 | /* Zero out all of the inode table blocks */ | |
576 | block = group_data[i].inode_table; | |
577 | ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", | |
578 | block, sbi->s_itb_per_group); | |
579 | err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, | |
580 | GFP_NOFS); | |
581 | if (err) | |
582 | goto out; | |
583 | ||
584 | handle_bb: | |
585 | if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT) | |
586 | goto handle_ib; | |
587 | ||
588 | /* Initialize block bitmap of the @group */ | |
589 | block = group_data[i].block_bitmap; | |
590 | err = extend_or_restart_transaction(handle, 1); | |
591 | if (err) | |
592 | goto out; | |
593 | ||
594 | bh = bclean(handle, sb, block); | |
595 | if (IS_ERR(bh)) { | |
596 | err = PTR_ERR(bh); | |
597 | goto out; | |
598 | } | |
01f795f9 YY |
599 | overhead = ext4_group_overhead_blocks(sb, group); |
600 | if (overhead != 0) { | |
33afdcc5 YY |
601 | ext4_debug("mark backup superblock %#04llx (+0)\n", |
602 | start); | |
01f795f9 | 603 | ext4_set_bits(bh->b_data, 0, overhead); |
33afdcc5 YY |
604 | } |
605 | ext4_mark_bitmap_end(group_data[i].blocks_count, | |
606 | sb->s_blocksize * 8, bh->b_data); | |
607 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
4e7e558e | 608 | brelse(bh); |
33afdcc5 YY |
609 | if (err) |
610 | goto out; | |
33afdcc5 YY |
611 | |
612 | handle_ib: | |
613 | if (bg_flags[i] & EXT4_BG_INODE_UNINIT) | |
614 | continue; | |
615 | ||
616 | /* Initialize inode bitmap of the @group */ | |
617 | block = group_data[i].inode_bitmap; | |
618 | err = extend_or_restart_transaction(handle, 1); | |
619 | if (err) | |
620 | goto out; | |
621 | /* Mark unused entries in inode bitmap used */ | |
622 | bh = bclean(handle, sb, block); | |
623 | if (IS_ERR(bh)) { | |
624 | err = PTR_ERR(bh); | |
625 | goto out; | |
626 | } | |
627 | ||
628 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), | |
629 | sb->s_blocksize * 8, bh->b_data); | |
630 | err = ext4_handle_dirty_metadata(handle, NULL, bh); | |
4e7e558e | 631 | brelse(bh); |
33afdcc5 YY |
632 | if (err) |
633 | goto out; | |
33afdcc5 | 634 | } |
33afdcc5 YY |
635 | |
636 | /* Mark group tables in block bitmap */ | |
637 | for (j = 0; j < GROUP_TABLE_COUNT; j++) { | |
638 | count = group_table_count[j]; | |
639 | start = (&group_data[0].block_bitmap)[j]; | |
640 | block = start; | |
641 | for (i = 1; i < flex_gd->count; i++) { | |
642 | block += group_table_count[j]; | |
643 | if (block == (&group_data[i].block_bitmap)[j]) { | |
644 | count += group_table_count[j]; | |
645 | continue; | |
646 | } | |
647 | err = set_flexbg_block_bitmap(sb, handle, | |
648 | flex_gd, start, count); | |
649 | if (err) | |
650 | goto out; | |
651 | count = group_table_count[j]; | |
b93c9535 | 652 | start = (&group_data[i].block_bitmap)[j]; |
33afdcc5 YY |
653 | block = start; |
654 | } | |
655 | ||
656 | if (count) { | |
657 | err = set_flexbg_block_bitmap(sb, handle, | |
658 | flex_gd, start, count); | |
659 | if (err) | |
660 | goto out; | |
661 | } | |
662 | } | |
663 | ||
664 | out: | |
33afdcc5 YY |
665 | err2 = ext4_journal_stop(handle); |
666 | if (err2 && !err) | |
667 | err = err2; | |
668 | ||
669 | return err; | |
670 | } | |
671 | ||
ac27a0ec DK |
672 | /* |
673 | * Iterate through the groups which hold BACKUP superblock/GDT copies in an | |
617ba13b | 674 | * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before |
ac27a0ec DK |
675 | * calling this for the first time. In a sparse filesystem it will be the |
676 | * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... | |
677 | * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... | |
678 | */ | |
617ba13b | 679 | static unsigned ext4_list_backups(struct super_block *sb, unsigned *three, |
ac27a0ec DK |
680 | unsigned *five, unsigned *seven) |
681 | { | |
682 | unsigned *min = three; | |
683 | int mult = 3; | |
684 | unsigned ret; | |
685 | ||
e2b911c5 | 686 | if (!ext4_has_feature_sparse_super(sb)) { |
ac27a0ec DK |
687 | ret = *min; |
688 | *min += 1; | |
689 | return ret; | |
690 | } | |
691 | ||
692 | if (*five < *min) { | |
693 | min = five; | |
694 | mult = 5; | |
695 | } | |
696 | if (*seven < *min) { | |
697 | min = seven; | |
698 | mult = 7; | |
699 | } | |
700 | ||
701 | ret = *min; | |
702 | *min *= mult; | |
703 | ||
704 | return ret; | |
705 | } | |
706 | ||
707 | /* | |
708 | * Check that all of the backup GDT blocks are held in the primary GDT block. | |
709 | * It is assumed that they are stored in group order. Returns the number of | |
710 | * groups in current filesystem that have BACKUPS, or -ve error code. | |
711 | */ | |
712 | static int verify_reserved_gdb(struct super_block *sb, | |
c72df9f9 | 713 | ext4_group_t end, |
ac27a0ec DK |
714 | struct buffer_head *primary) |
715 | { | |
617ba13b | 716 | const ext4_fsblk_t blk = primary->b_blocknr; |
ac27a0ec DK |
717 | unsigned three = 1; |
718 | unsigned five = 5; | |
719 | unsigned seven = 7; | |
720 | unsigned grp; | |
721 | __le32 *p = (__le32 *)primary->b_data; | |
722 | int gdbackups = 0; | |
723 | ||
617ba13b | 724 | while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { |
bd81d8ee LV |
725 | if (le32_to_cpu(*p++) != |
726 | grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ | |
12062ddd | 727 | ext4_warning(sb, "reserved GDT %llu" |
2ae02107 | 728 | " missing grp %d (%llu)", |
ac27a0ec | 729 | blk, grp, |
bd81d8ee LV |
730 | grp * |
731 | (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + | |
732 | blk); | |
ac27a0ec DK |
733 | return -EINVAL; |
734 | } | |
617ba13b | 735 | if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) |
ac27a0ec DK |
736 | return -EFBIG; |
737 | } | |
738 | ||
739 | return gdbackups; | |
740 | } | |
741 | ||
742 | /* | |
743 | * Called when we need to bring a reserved group descriptor table block into | |
744 | * use from the resize inode. The primary copy of the new GDT block currently | |
745 | * is an indirect block (under the double indirect block in the resize inode). | |
746 | * The new backup GDT blocks will be stored as leaf blocks in this indirect | |
747 | * block, in group order. Even though we know all the block numbers we need, | |
748 | * we check to ensure that the resize inode has actually reserved these blocks. | |
749 | * | |
750 | * Don't need to update the block bitmaps because the blocks are still in use. | |
751 | * | |
752 | * We get all of the error cases out of the way, so that we are sure to not | |
753 | * fail once we start modifying the data on disk, because JBD has no rollback. | |
754 | */ | |
755 | static int add_new_gdb(handle_t *handle, struct inode *inode, | |
2f919710 | 756 | ext4_group_t group) |
ac27a0ec DK |
757 | { |
758 | struct super_block *sb = inode->i_sb; | |
617ba13b | 759 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
2f919710 | 760 | unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); |
617ba13b | 761 | ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; |
9da1f6d0 TT |
762 | struct buffer_head **o_group_desc, **n_group_desc = NULL; |
763 | struct buffer_head *dind = NULL; | |
764 | struct buffer_head *gdb_bh = NULL; | |
ac27a0ec | 765 | int gdbackups; |
9da1f6d0 | 766 | struct ext4_iloc iloc = { .bh = NULL }; |
ac27a0ec DK |
767 | __le32 *data; |
768 | int err; | |
769 | ||
770 | if (test_opt(sb, DEBUG)) | |
771 | printk(KERN_DEBUG | |
617ba13b | 772 | "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", |
ac27a0ec DK |
773 | gdb_num); |
774 | ||
9da1f6d0 TT |
775 | gdb_bh = ext4_sb_bread(sb, gdblock, 0); |
776 | if (IS_ERR(gdb_bh)) | |
777 | return PTR_ERR(gdb_bh); | |
ac27a0ec | 778 | |
c72df9f9 | 779 | gdbackups = verify_reserved_gdb(sb, group, gdb_bh); |
2f919710 | 780 | if (gdbackups < 0) { |
ac27a0ec | 781 | err = gdbackups; |
9da1f6d0 | 782 | goto errout; |
ac27a0ec DK |
783 | } |
784 | ||
617ba13b | 785 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
9da1f6d0 TT |
786 | dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); |
787 | if (IS_ERR(dind)) { | |
788 | err = PTR_ERR(dind); | |
789 | dind = NULL; | |
790 | goto errout; | |
ac27a0ec DK |
791 | } |
792 | ||
793 | data = (__le32 *)dind->b_data; | |
617ba13b | 794 | if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { |
12062ddd | 795 | ext4_warning(sb, "new group %u GDT block %llu not reserved", |
2f919710 | 796 | group, gdblock); |
ac27a0ec | 797 | err = -EINVAL; |
9da1f6d0 | 798 | goto errout; |
ac27a0ec DK |
799 | } |
800 | ||
5d601255 | 801 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
b4097142 TT |
802 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
803 | if (unlikely(err)) | |
9da1f6d0 | 804 | goto errout; |
ac27a0ec | 805 | |
5d601255 | 806 | BUFFER_TRACE(gdb_bh, "get_write_access"); |
2f919710 | 807 | err = ext4_journal_get_write_access(handle, gdb_bh); |
b4097142 | 808 | if (unlikely(err)) |
9da1f6d0 | 809 | goto errout; |
ac27a0ec | 810 | |
5d601255 | 811 | BUFFER_TRACE(dind, "get_write_access"); |
b4097142 TT |
812 | err = ext4_journal_get_write_access(handle, dind); |
813 | if (unlikely(err)) | |
814 | ext4_std_error(sb, err); | |
ac27a0ec | 815 | |
617ba13b | 816 | /* ext4_reserve_inode_write() gets a reference on the iloc */ |
b4097142 TT |
817 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
818 | if (unlikely(err)) | |
9da1f6d0 | 819 | goto errout; |
ac27a0ec | 820 | |
f18a5f21 TT |
821 | n_group_desc = ext4_kvmalloc((gdb_num + 1) * |
822 | sizeof(struct buffer_head *), | |
823 | GFP_NOFS); | |
ac27a0ec DK |
824 | if (!n_group_desc) { |
825 | err = -ENOMEM; | |
f18a5f21 TT |
826 | ext4_warning(sb, "not enough memory for %lu groups", |
827 | gdb_num + 1); | |
9da1f6d0 | 828 | goto errout; |
ac27a0ec DK |
829 | } |
830 | ||
831 | /* | |
832 | * Finally, we have all of the possible failures behind us... | |
833 | * | |
834 | * Remove new GDT block from inode double-indirect block and clear out | |
835 | * the new GDT block for use (which also "frees" the backup GDT blocks | |
836 | * from the reserved inode). We don't need to change the bitmaps for | |
837 | * these blocks, because they are marked as in-use from being in the | |
838 | * reserved inode, and will become GDT blocks (primary and backup). | |
839 | */ | |
617ba13b | 840 | data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; |
b4097142 TT |
841 | err = ext4_handle_dirty_metadata(handle, NULL, dind); |
842 | if (unlikely(err)) { | |
843 | ext4_std_error(sb, err); | |
9da1f6d0 | 844 | goto errout; |
b4097142 | 845 | } |
ac27a0ec | 846 | inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9; |
617ba13b | 847 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
2f919710 YY |
848 | memset(gdb_bh->b_data, 0, sb->s_blocksize); |
849 | err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); | |
b4097142 TT |
850 | if (unlikely(err)) { |
851 | ext4_std_error(sb, err); | |
9da1f6d0 | 852 | goto errout; |
b4097142 TT |
853 | } |
854 | brelse(dind); | |
ac27a0ec | 855 | |
617ba13b | 856 | o_group_desc = EXT4_SB(sb)->s_group_desc; |
ac27a0ec | 857 | memcpy(n_group_desc, o_group_desc, |
617ba13b | 858 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); |
2f919710 | 859 | n_group_desc[gdb_num] = gdb_bh; |
617ba13b MC |
860 | EXT4_SB(sb)->s_group_desc = n_group_desc; |
861 | EXT4_SB(sb)->s_gdb_count++; | |
b93b41d4 | 862 | kvfree(o_group_desc); |
ac27a0ec | 863 | |
e8546d06 | 864 | le16_add_cpu(&es->s_reserved_gdt_blocks, -1); |
b50924c2 | 865 | err = ext4_handle_dirty_super(handle, sb); |
b4097142 TT |
866 | if (err) |
867 | ext4_std_error(sb, err); | |
b4097142 | 868 | return err; |
9da1f6d0 | 869 | errout: |
b93b41d4 | 870 | kvfree(n_group_desc); |
ac27a0ec | 871 | brelse(iloc.bh); |
ac27a0ec | 872 | brelse(dind); |
2f919710 | 873 | brelse(gdb_bh); |
ac27a0ec | 874 | |
617ba13b | 875 | ext4_debug("leaving with error %d\n", err); |
ac27a0ec DK |
876 | return err; |
877 | } | |
878 | ||
01f795f9 YY |
879 | /* |
880 | * add_new_gdb_meta_bg is the sister of add_new_gdb. | |
881 | */ | |
882 | static int add_new_gdb_meta_bg(struct super_block *sb, | |
883 | handle_t *handle, ext4_group_t group) { | |
884 | ext4_fsblk_t gdblock; | |
885 | struct buffer_head *gdb_bh; | |
886 | struct buffer_head **o_group_desc, **n_group_desc; | |
887 | unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
888 | int err; | |
889 | ||
890 | gdblock = ext4_meta_bg_first_block_no(sb, group) + | |
891 | ext4_bg_has_super(sb, group); | |
9da1f6d0 TT |
892 | gdb_bh = ext4_sb_bread(sb, gdblock, 0); |
893 | if (IS_ERR(gdb_bh)) | |
894 | return PTR_ERR(gdb_bh); | |
01f795f9 YY |
895 | n_group_desc = ext4_kvmalloc((gdb_num + 1) * |
896 | sizeof(struct buffer_head *), | |
897 | GFP_NOFS); | |
898 | if (!n_group_desc) { | |
64176ffd | 899 | brelse(gdb_bh); |
01f795f9 YY |
900 | err = -ENOMEM; |
901 | ext4_warning(sb, "not enough memory for %lu groups", | |
902 | gdb_num + 1); | |
903 | return err; | |
904 | } | |
905 | ||
906 | o_group_desc = EXT4_SB(sb)->s_group_desc; | |
907 | memcpy(n_group_desc, o_group_desc, | |
908 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); | |
909 | n_group_desc[gdb_num] = gdb_bh; | |
7a82651d LC |
910 | |
911 | BUFFER_TRACE(gdb_bh, "get_write_access"); | |
912 | err = ext4_journal_get_write_access(handle, gdb_bh); | |
913 | if (err) { | |
914 | kvfree(n_group_desc); | |
915 | brelse(gdb_bh); | |
916 | return err; | |
917 | } | |
918 | ||
01f795f9 YY |
919 | EXT4_SB(sb)->s_group_desc = n_group_desc; |
920 | EXT4_SB(sb)->s_gdb_count++; | |
b93b41d4 | 921 | kvfree(o_group_desc); |
01f795f9 YY |
922 | return err; |
923 | } | |
924 | ||
ac27a0ec DK |
925 | /* |
926 | * Called when we are adding a new group which has a backup copy of each of | |
927 | * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. | |
928 | * We need to add these reserved backup GDT blocks to the resize inode, so | |
929 | * that they are kept for future resizing and not allocated to files. | |
930 | * | |
931 | * Each reserved backup GDT block will go into a different indirect block. | |
932 | * The indirect blocks are actually the primary reserved GDT blocks, | |
933 | * so we know in advance what their block numbers are. We only get the | |
934 | * double-indirect block to verify it is pointing to the primary reserved | |
935 | * GDT blocks so we don't overwrite a data block by accident. The reserved | |
936 | * backup GDT blocks are stored in their reserved primary GDT block. | |
937 | */ | |
938 | static int reserve_backup_gdb(handle_t *handle, struct inode *inode, | |
668f4dc5 | 939 | ext4_group_t group) |
ac27a0ec DK |
940 | { |
941 | struct super_block *sb = inode->i_sb; | |
617ba13b | 942 | int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); |
ac27a0ec DK |
943 | struct buffer_head **primary; |
944 | struct buffer_head *dind; | |
617ba13b MC |
945 | struct ext4_iloc iloc; |
946 | ext4_fsblk_t blk; | |
ac27a0ec DK |
947 | __le32 *data, *end; |
948 | int gdbackups = 0; | |
949 | int res, i; | |
950 | int err; | |
951 | ||
216553c4 | 952 | primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS); |
ac27a0ec DK |
953 | if (!primary) |
954 | return -ENOMEM; | |
955 | ||
617ba13b | 956 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
9da1f6d0 TT |
957 | dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0); |
958 | if (IS_ERR(dind)) { | |
959 | err = PTR_ERR(dind); | |
960 | dind = NULL; | |
ac27a0ec DK |
961 | goto exit_free; |
962 | } | |
963 | ||
617ba13b | 964 | blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; |
94460093 JB |
965 | data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % |
966 | EXT4_ADDR_PER_BLOCK(sb)); | |
617ba13b | 967 | end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); |
ac27a0ec DK |
968 | |
969 | /* Get each reserved primary GDT block and verify it holds backups */ | |
970 | for (res = 0; res < reserved_gdb; res++, blk++) { | |
971 | if (le32_to_cpu(*data) != blk) { | |
12062ddd | 972 | ext4_warning(sb, "reserved block %llu" |
ac27a0ec DK |
973 | " not at offset %ld", |
974 | blk, | |
975 | (long)(data - (__le32 *)dind->b_data)); | |
976 | err = -EINVAL; | |
977 | goto exit_bh; | |
978 | } | |
9da1f6d0 TT |
979 | primary[res] = ext4_sb_bread(sb, blk, 0); |
980 | if (IS_ERR(primary[res])) { | |
981 | err = PTR_ERR(primary[res]); | |
982 | primary[res] = NULL; | |
ac27a0ec DK |
983 | goto exit_bh; |
984 | } | |
c72df9f9 YY |
985 | gdbackups = verify_reserved_gdb(sb, group, primary[res]); |
986 | if (gdbackups < 0) { | |
ac27a0ec DK |
987 | brelse(primary[res]); |
988 | err = gdbackups; | |
989 | goto exit_bh; | |
990 | } | |
991 | if (++data >= end) | |
992 | data = (__le32 *)dind->b_data; | |
993 | } | |
994 | ||
995 | for (i = 0; i < reserved_gdb; i++) { | |
5d601255 | 996 | BUFFER_TRACE(primary[i], "get_write_access"); |
37be2f59 | 997 | if ((err = ext4_journal_get_write_access(handle, primary[i]))) |
ac27a0ec | 998 | goto exit_bh; |
ac27a0ec DK |
999 | } |
1000 | ||
617ba13b | 1001 | if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) |
ac27a0ec DK |
1002 | goto exit_bh; |
1003 | ||
1004 | /* | |
1005 | * Finally we can add each of the reserved backup GDT blocks from | |
1006 | * the new group to its reserved primary GDT block. | |
1007 | */ | |
668f4dc5 | 1008 | blk = group * EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1009 | for (i = 0; i < reserved_gdb; i++) { |
1010 | int err2; | |
1011 | data = (__le32 *)primary[i]->b_data; | |
1012 | /* printk("reserving backup %lu[%u] = %lu\n", | |
1013 | primary[i]->b_blocknr, gdbackups, | |
1014 | blk + primary[i]->b_blocknr); */ | |
1015 | data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); | |
0390131b | 1016 | err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); |
ac27a0ec DK |
1017 | if (!err) |
1018 | err = err2; | |
1019 | } | |
1020 | inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9; | |
617ba13b | 1021 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
1022 | |
1023 | exit_bh: | |
1024 | while (--res >= 0) | |
1025 | brelse(primary[res]); | |
1026 | brelse(dind); | |
1027 | ||
1028 | exit_free: | |
1029 | kfree(primary); | |
1030 | ||
1031 | return err; | |
1032 | } | |
1033 | ||
1034 | /* | |
617ba13b | 1035 | * Update the backup copies of the ext4 metadata. These don't need to be part |
ac27a0ec DK |
1036 | * of the main resize transaction, because e2fsck will re-write them if there |
1037 | * is a problem (basically only OOM will cause a problem). However, we | |
1038 | * _should_ update the backups if possible, in case the primary gets trashed | |
1039 | * for some reason and we need to run e2fsck from a backup superblock. The | |
1040 | * important part is that the new block and inode counts are in the backup | |
1041 | * superblocks, and the location of the new group metadata in the GDT backups. | |
1042 | * | |
32ed5058 TT |
1043 | * We do not need take the s_resize_lock for this, because these |
1044 | * blocks are not otherwise touched by the filesystem code when it is | |
1045 | * mounted. We don't need to worry about last changing from | |
1046 | * sbi->s_groups_count, because the worst that can happen is that we | |
1047 | * do not copy the full number of backups at this time. The resize | |
1048 | * which changed s_groups_count will backup again. | |
ac27a0ec | 1049 | */ |
904dad47 | 1050 | static void update_backups(struct super_block *sb, sector_t blk_off, char *data, |
01f795f9 | 1051 | int size, int meta_bg) |
ac27a0ec | 1052 | { |
617ba13b | 1053 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
01f795f9 | 1054 | ext4_group_t last; |
617ba13b | 1055 | const int bpg = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
1056 | unsigned three = 1; |
1057 | unsigned five = 5; | |
1058 | unsigned seven = 7; | |
01f795f9 | 1059 | ext4_group_t group = 0; |
ac27a0ec DK |
1060 | int rest = sb->s_blocksize - size; |
1061 | handle_t *handle; | |
1062 | int err = 0, err2; | |
1063 | ||
9924a92a | 1064 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA); |
ac27a0ec DK |
1065 | if (IS_ERR(handle)) { |
1066 | group = 1; | |
1067 | err = PTR_ERR(handle); | |
1068 | goto exit_err; | |
1069 | } | |
1070 | ||
01f795f9 YY |
1071 | if (meta_bg == 0) { |
1072 | group = ext4_list_backups(sb, &three, &five, &seven); | |
1073 | last = sbi->s_groups_count; | |
1074 | } else { | |
904dad47 | 1075 | group = ext4_get_group_number(sb, blk_off) + 1; |
01f795f9 YY |
1076 | last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2); |
1077 | } | |
1078 | ||
1079 | while (group < sbi->s_groups_count) { | |
ac27a0ec | 1080 | struct buffer_head *bh; |
01f795f9 | 1081 | ext4_fsblk_t backup_block; |
ac27a0ec DK |
1082 | |
1083 | /* Out of journal space, and can't get more - abort - so sad */ | |
0390131b FM |
1084 | if (ext4_handle_valid(handle) && |
1085 | handle->h_buffer_credits == 0 && | |
617ba13b MC |
1086 | ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) && |
1087 | (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) | |
ac27a0ec DK |
1088 | break; |
1089 | ||
01f795f9 | 1090 | if (meta_bg == 0) |
9378c676 | 1091 | backup_block = ((ext4_fsblk_t)group) * bpg + blk_off; |
01f795f9 YY |
1092 | else |
1093 | backup_block = (ext4_group_first_block_no(sb, group) + | |
1094 | ext4_bg_has_super(sb, group)); | |
1095 | ||
1096 | bh = sb_getblk(sb, backup_block); | |
aebf0243 | 1097 | if (unlikely(!bh)) { |
860d21e2 | 1098 | err = -ENOMEM; |
ac27a0ec DK |
1099 | break; |
1100 | } | |
01f795f9 YY |
1101 | ext4_debug("update metadata backup %llu(+%llu)\n", |
1102 | backup_block, backup_block - | |
1103 | ext4_group_first_block_no(sb, group)); | |
5d601255 | 1104 | BUFFER_TRACE(bh, "get_write_access"); |
9910b257 VA |
1105 | if ((err = ext4_journal_get_write_access(handle, bh))) { |
1106 | brelse(bh); | |
ac27a0ec | 1107 | break; |
9910b257 | 1108 | } |
ac27a0ec DK |
1109 | lock_buffer(bh); |
1110 | memcpy(bh->b_data, data, size); | |
1111 | if (rest) | |
1112 | memset(bh->b_data + size, 0, rest); | |
1113 | set_buffer_uptodate(bh); | |
1114 | unlock_buffer(bh); | |
b4097142 TT |
1115 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
1116 | if (unlikely(err)) | |
1117 | ext4_std_error(sb, err); | |
ac27a0ec | 1118 | brelse(bh); |
01f795f9 YY |
1119 | |
1120 | if (meta_bg == 0) | |
1121 | group = ext4_list_backups(sb, &three, &five, &seven); | |
1122 | else if (group == last) | |
1123 | break; | |
1124 | else | |
1125 | group = last; | |
ac27a0ec | 1126 | } |
617ba13b | 1127 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
1128 | err = err2; |
1129 | ||
1130 | /* | |
1131 | * Ugh! Need to have e2fsck write the backup copies. It is too | |
1132 | * late to revert the resize, we shouldn't fail just because of | |
1133 | * the backup copies (they are only needed in case of corruption). | |
1134 | * | |
1135 | * However, if we got here we have a journal problem too, so we | |
1136 | * can't really start a transaction to mark the superblock. | |
1137 | * Chicken out and just set the flag on the hope it will be written | |
1138 | * to disk, and if not - we will simply wait until next fsck. | |
1139 | */ | |
1140 | exit_err: | |
1141 | if (err) { | |
12062ddd | 1142 | ext4_warning(sb, "can't update backup for group %u (err %d), " |
ac27a0ec | 1143 | "forcing fsck on next reboot", group, err); |
617ba13b MC |
1144 | sbi->s_mount_state &= ~EXT4_VALID_FS; |
1145 | sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); | |
ac27a0ec DK |
1146 | mark_buffer_dirty(sbi->s_sbh); |
1147 | } | |
1148 | } | |
1149 | ||
bb08c1e7 YY |
1150 | /* |
1151 | * ext4_add_new_descs() adds @count group descriptor of groups | |
1152 | * starting at @group | |
1153 | * | |
1154 | * @handle: journal handle | |
1155 | * @sb: super block | |
1156 | * @group: the group no. of the first group desc to be added | |
1157 | * @resize_inode: the resize inode | |
1158 | * @count: number of group descriptors to be added | |
1159 | */ | |
1160 | static int ext4_add_new_descs(handle_t *handle, struct super_block *sb, | |
1161 | ext4_group_t group, struct inode *resize_inode, | |
1162 | ext4_group_t count) | |
1163 | { | |
1164 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1165 | struct ext4_super_block *es = sbi->s_es; | |
1166 | struct buffer_head *gdb_bh; | |
1167 | int i, gdb_off, gdb_num, err = 0; | |
01f795f9 | 1168 | int meta_bg; |
bb08c1e7 | 1169 | |
e2b911c5 | 1170 | meta_bg = ext4_has_feature_meta_bg(sb); |
bb08c1e7 YY |
1171 | for (i = 0; i < count; i++, group++) { |
1172 | int reserved_gdb = ext4_bg_has_super(sb, group) ? | |
1173 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; | |
1174 | ||
1175 | gdb_off = group % EXT4_DESC_PER_BLOCK(sb); | |
1176 | gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
1177 | ||
1178 | /* | |
1179 | * We will only either add reserved group blocks to a backup group | |
1180 | * or remove reserved blocks for the first group in a new group block. | |
1181 | * Doing both would be mean more complex code, and sane people don't | |
1182 | * use non-sparse filesystems anymore. This is already checked above. | |
1183 | */ | |
1184 | if (gdb_off) { | |
1185 | gdb_bh = sbi->s_group_desc[gdb_num]; | |
5d601255 | 1186 | BUFFER_TRACE(gdb_bh, "get_write_access"); |
bb08c1e7 YY |
1187 | err = ext4_journal_get_write_access(handle, gdb_bh); |
1188 | ||
1189 | if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group)) | |
1190 | err = reserve_backup_gdb(handle, resize_inode, group); | |
01f795f9 YY |
1191 | } else if (meta_bg != 0) { |
1192 | err = add_new_gdb_meta_bg(sb, handle, group); | |
1193 | } else { | |
bb08c1e7 | 1194 | err = add_new_gdb(handle, resize_inode, group); |
01f795f9 | 1195 | } |
bb08c1e7 YY |
1196 | if (err) |
1197 | break; | |
1198 | } | |
1199 | return err; | |
1200 | } | |
1201 | ||
41a246d1 DW |
1202 | static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block) |
1203 | { | |
1204 | struct buffer_head *bh = sb_getblk(sb, block); | |
aebf0243 | 1205 | if (unlikely(!bh)) |
41a246d1 | 1206 | return NULL; |
7f1468d1 DM |
1207 | if (!bh_uptodate_or_lock(bh)) { |
1208 | if (bh_submit_read(bh) < 0) { | |
1209 | brelse(bh); | |
1210 | return NULL; | |
1211 | } | |
41a246d1 | 1212 | } |
41a246d1 DW |
1213 | |
1214 | return bh; | |
1215 | } | |
1216 | ||
1217 | static int ext4_set_bitmap_checksums(struct super_block *sb, | |
1218 | ext4_group_t group, | |
1219 | struct ext4_group_desc *gdp, | |
1220 | struct ext4_new_group_data *group_data) | |
1221 | { | |
1222 | struct buffer_head *bh; | |
1223 | ||
9aa5d32b | 1224 | if (!ext4_has_metadata_csum(sb)) |
41a246d1 DW |
1225 | return 0; |
1226 | ||
1227 | bh = ext4_get_bitmap(sb, group_data->inode_bitmap); | |
1228 | if (!bh) | |
1229 | return -EIO; | |
1230 | ext4_inode_bitmap_csum_set(sb, group, gdp, bh, | |
1231 | EXT4_INODES_PER_GROUP(sb) / 8); | |
1232 | brelse(bh); | |
1233 | ||
fa77dcfa DW |
1234 | bh = ext4_get_bitmap(sb, group_data->block_bitmap); |
1235 | if (!bh) | |
1236 | return -EIO; | |
79f1ba49 | 1237 | ext4_block_bitmap_csum_set(sb, group, gdp, bh); |
fa77dcfa DW |
1238 | brelse(bh); |
1239 | ||
41a246d1 DW |
1240 | return 0; |
1241 | } | |
1242 | ||
083f5b24 YY |
1243 | /* |
1244 | * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg | |
1245 | */ | |
1246 | static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb, | |
1247 | struct ext4_new_flex_group_data *flex_gd) | |
1248 | { | |
1249 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
1250 | struct ext4_group_desc *gdp; | |
1251 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1252 | struct buffer_head *gdb_bh; | |
1253 | ext4_group_t group; | |
1254 | __u16 *bg_flags = flex_gd->bg_flags; | |
1255 | int i, gdb_off, gdb_num, err = 0; | |
1256 | ||
1257 | ||
1258 | for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) { | |
1259 | group = group_data->group; | |
1260 | ||
1261 | gdb_off = group % EXT4_DESC_PER_BLOCK(sb); | |
1262 | gdb_num = group / EXT4_DESC_PER_BLOCK(sb); | |
1263 | ||
1264 | /* | |
1265 | * get_write_access() has been called on gdb_bh by ext4_add_new_desc(). | |
1266 | */ | |
1267 | gdb_bh = sbi->s_group_desc[gdb_num]; | |
1268 | /* Update group descriptor block for new group */ | |
2716b802 | 1269 | gdp = (struct ext4_group_desc *)(gdb_bh->b_data + |
083f5b24 YY |
1270 | gdb_off * EXT4_DESC_SIZE(sb)); |
1271 | ||
1272 | memset(gdp, 0, EXT4_DESC_SIZE(sb)); | |
1273 | ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap); | |
1274 | ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap); | |
41a246d1 DW |
1275 | err = ext4_set_bitmap_checksums(sb, group, gdp, group_data); |
1276 | if (err) { | |
1277 | ext4_std_error(sb, err); | |
1278 | break; | |
1279 | } | |
1280 | ||
083f5b24 YY |
1281 | ext4_inode_table_set(sb, gdp, group_data->inode_table); |
1282 | ext4_free_group_clusters_set(sb, gdp, | |
810da240 | 1283 | EXT4_NUM_B2C(sbi, group_data->free_blocks_count)); |
083f5b24 | 1284 | ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); |
93f90526 TT |
1285 | if (ext4_has_group_desc_csum(sb)) |
1286 | ext4_itable_unused_set(sb, gdp, | |
1287 | EXT4_INODES_PER_GROUP(sb)); | |
083f5b24 | 1288 | gdp->bg_flags = cpu_to_le16(*bg_flags); |
feb0ab32 | 1289 | ext4_group_desc_csum_set(sb, group, gdp); |
083f5b24 YY |
1290 | |
1291 | err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh); | |
1292 | if (unlikely(err)) { | |
1293 | ext4_std_error(sb, err); | |
1294 | break; | |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * We can allocate memory for mb_alloc based on the new group | |
1299 | * descriptor | |
1300 | */ | |
1301 | err = ext4_mb_add_groupinfo(sb, group, gdp); | |
1302 | if (err) | |
1303 | break; | |
1304 | } | |
1305 | return err; | |
1306 | } | |
1307 | ||
2e10e2f2 YY |
1308 | /* |
1309 | * ext4_update_super() updates the super block so that the newly added | |
1310 | * groups can be seen by the filesystem. | |
1311 | * | |
1312 | * @sb: super block | |
1313 | * @flex_gd: new added groups | |
1314 | */ | |
1315 | static void ext4_update_super(struct super_block *sb, | |
1316 | struct ext4_new_flex_group_data *flex_gd) | |
1317 | { | |
1318 | ext4_fsblk_t blocks_count = 0; | |
1319 | ext4_fsblk_t free_blocks = 0; | |
1320 | ext4_fsblk_t reserved_blocks = 0; | |
1321 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
1322 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1323 | struct ext4_super_block *es = sbi->s_es; | |
8a991849 | 1324 | int i; |
2e10e2f2 YY |
1325 | |
1326 | BUG_ON(flex_gd->count == 0 || group_data == NULL); | |
1327 | /* | |
1328 | * Make the new blocks and inodes valid next. We do this before | |
1329 | * increasing the group count so that once the group is enabled, | |
1330 | * all of its blocks and inodes are already valid. | |
1331 | * | |
1332 | * We always allocate group-by-group, then block-by-block or | |
1333 | * inode-by-inode within a group, so enabling these | |
1334 | * blocks/inodes before the group is live won't actually let us | |
1335 | * allocate the new space yet. | |
1336 | */ | |
1337 | for (i = 0; i < flex_gd->count; i++) { | |
1338 | blocks_count += group_data[i].blocks_count; | |
1339 | free_blocks += group_data[i].free_blocks_count; | |
1340 | } | |
1341 | ||
1342 | reserved_blocks = ext4_r_blocks_count(es) * 100; | |
01f795f9 | 1343 | reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es)); |
2e10e2f2 YY |
1344 | reserved_blocks *= blocks_count; |
1345 | do_div(reserved_blocks, 100); | |
1346 | ||
1347 | ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count); | |
636d7e2e | 1348 | ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks); |
2e10e2f2 YY |
1349 | le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) * |
1350 | flex_gd->count); | |
636d7e2e DW |
1351 | le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) * |
1352 | flex_gd->count); | |
2e10e2f2 | 1353 | |
01f795f9 | 1354 | ext4_debug("free blocks count %llu", ext4_free_blocks_count(es)); |
2e10e2f2 YY |
1355 | /* |
1356 | * We need to protect s_groups_count against other CPUs seeing | |
1357 | * inconsistent state in the superblock. | |
1358 | * | |
1359 | * The precise rules we use are: | |
1360 | * | |
1361 | * * Writers must perform a smp_wmb() after updating all | |
1362 | * dependent data and before modifying the groups count | |
1363 | * | |
1364 | * * Readers must perform an smp_rmb() after reading the groups | |
1365 | * count and before reading any dependent data. | |
1366 | * | |
1367 | * NB. These rules can be relaxed when checking the group count | |
1368 | * while freeing data, as we can only allocate from a block | |
1369 | * group after serialising against the group count, and we can | |
1370 | * only then free after serialising in turn against that | |
1371 | * allocation. | |
1372 | */ | |
1373 | smp_wmb(); | |
1374 | ||
1375 | /* Update the global fs size fields */ | |
1376 | sbi->s_groups_count += flex_gd->count; | |
c5c72d81 TT |
1377 | sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count, |
1378 | (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb))); | |
2e10e2f2 YY |
1379 | |
1380 | /* Update the reserved block counts only once the new group is | |
1381 | * active. */ | |
1382 | ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + | |
1383 | reserved_blocks); | |
1384 | ||
1385 | /* Update the free space counts */ | |
1386 | percpu_counter_add(&sbi->s_freeclusters_counter, | |
810da240 | 1387 | EXT4_NUM_B2C(sbi, free_blocks)); |
2e10e2f2 YY |
1388 | percpu_counter_add(&sbi->s_freeinodes_counter, |
1389 | EXT4_INODES_PER_GROUP(sb) * flex_gd->count); | |
1390 | ||
01f795f9 YY |
1391 | ext4_debug("free blocks count %llu", |
1392 | percpu_counter_read(&sbi->s_freeclusters_counter)); | |
e2b911c5 | 1393 | if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) { |
2e10e2f2 YY |
1394 | ext4_group_t flex_group; |
1395 | flex_group = ext4_flex_group(sbi, group_data[0].group); | |
90ba983f TT |
1396 | atomic64_add(EXT4_NUM_B2C(sbi, free_blocks), |
1397 | &sbi->s_flex_groups[flex_group].free_clusters); | |
2e10e2f2 YY |
1398 | atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count, |
1399 | &sbi->s_flex_groups[flex_group].free_inodes); | |
1400 | } | |
1401 | ||
952fc18e TT |
1402 | /* |
1403 | * Update the fs overhead information | |
1404 | */ | |
1405 | ext4_calculate_overhead(sb); | |
1406 | ||
2e10e2f2 YY |
1407 | if (test_opt(sb, DEBUG)) |
1408 | printk(KERN_DEBUG "EXT4-fs: added group %u:" | |
1409 | "%llu blocks(%llu free %llu reserved)\n", flex_gd->count, | |
1410 | blocks_count, free_blocks, reserved_blocks); | |
1411 | } | |
1412 | ||
4bac1f8c YY |
1413 | /* Add a flex group to an fs. Ensure we handle all possible error conditions |
1414 | * _before_ we start modifying the filesystem, because we cannot abort the | |
1415 | * transaction and not have it write the data to disk. | |
1416 | */ | |
1417 | static int ext4_flex_group_add(struct super_block *sb, | |
1418 | struct inode *resize_inode, | |
1419 | struct ext4_new_flex_group_data *flex_gd) | |
1420 | { | |
1421 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1422 | struct ext4_super_block *es = sbi->s_es; | |
1423 | ext4_fsblk_t o_blocks_count; | |
1424 | ext4_grpblk_t last; | |
1425 | ext4_group_t group; | |
1426 | handle_t *handle; | |
1427 | unsigned reserved_gdb; | |
1428 | int err = 0, err2 = 0, credit; | |
1429 | ||
1430 | BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags); | |
1431 | ||
1432 | reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks); | |
1433 | o_blocks_count = ext4_blocks_count(es); | |
1434 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); | |
1435 | BUG_ON(last); | |
1436 | ||
1437 | err = setup_new_flex_group_blocks(sb, flex_gd); | |
1438 | if (err) | |
1439 | goto exit; | |
1440 | /* | |
1441 | * We will always be modifying at least the superblock and GDT | |
2c869b26 | 1442 | * blocks. If we are adding a group past the last current GDT block, |
4bac1f8c YY |
1443 | * we will also modify the inode and the dindirect block. If we |
1444 | * are adding a group with superblock/GDT backups we will also | |
1445 | * modify each of the reserved GDT dindirect blocks. | |
1446 | */ | |
2c869b26 JK |
1447 | credit = 3; /* sb, resize inode, resize inode dindirect */ |
1448 | /* GDT blocks */ | |
1449 | credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb)); | |
1450 | credit += reserved_gdb; /* Reserved GDT dindirect blocks */ | |
9924a92a | 1451 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit); |
4bac1f8c YY |
1452 | if (IS_ERR(handle)) { |
1453 | err = PTR_ERR(handle); | |
1454 | goto exit; | |
1455 | } | |
1456 | ||
5d601255 | 1457 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
4bac1f8c YY |
1458 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
1459 | if (err) | |
1460 | goto exit_journal; | |
1461 | ||
1462 | group = flex_gd->groups[0].group; | |
1463 | BUG_ON(group != EXT4_SB(sb)->s_groups_count); | |
1464 | err = ext4_add_new_descs(handle, sb, group, | |
1465 | resize_inode, flex_gd->count); | |
1466 | if (err) | |
1467 | goto exit_journal; | |
1468 | ||
1469 | err = ext4_setup_new_descs(handle, sb, flex_gd); | |
1470 | if (err) | |
1471 | goto exit_journal; | |
1472 | ||
1473 | ext4_update_super(sb, flex_gd); | |
1474 | ||
1475 | err = ext4_handle_dirty_super(handle, sb); | |
1476 | ||
1477 | exit_journal: | |
1478 | err2 = ext4_journal_stop(handle); | |
1479 | if (!err) | |
1480 | err = err2; | |
1481 | ||
1482 | if (!err) { | |
2ebd1704 YY |
1483 | int gdb_num = group / EXT4_DESC_PER_BLOCK(sb); |
1484 | int gdb_num_end = ((group + flex_gd->count - 1) / | |
1485 | EXT4_DESC_PER_BLOCK(sb)); | |
e2b911c5 | 1486 | int meta_bg = ext4_has_feature_meta_bg(sb); |
0acdb887 | 1487 | sector_t old_gdb = 0; |
2ebd1704 | 1488 | |
4bac1f8c | 1489 | update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, |
01f795f9 | 1490 | sizeof(struct ext4_super_block), 0); |
2ebd1704 | 1491 | for (; gdb_num <= gdb_num_end; gdb_num++) { |
4bac1f8c | 1492 | struct buffer_head *gdb_bh; |
2ebd1704 | 1493 | |
4bac1f8c | 1494 | gdb_bh = sbi->s_group_desc[gdb_num]; |
0acdb887 TM |
1495 | if (old_gdb == gdb_bh->b_blocknr) |
1496 | continue; | |
4bac1f8c | 1497 | update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data, |
01f795f9 | 1498 | gdb_bh->b_size, meta_bg); |
0acdb887 | 1499 | old_gdb = gdb_bh->b_blocknr; |
4bac1f8c YY |
1500 | } |
1501 | } | |
1502 | exit: | |
1503 | return err; | |
1504 | } | |
1505 | ||
19c5246d YY |
1506 | static int ext4_setup_next_flex_gd(struct super_block *sb, |
1507 | struct ext4_new_flex_group_data *flex_gd, | |
1508 | ext4_fsblk_t n_blocks_count, | |
1509 | unsigned long flexbg_size) | |
1510 | { | |
1511 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
1512 | struct ext4_new_group_data *group_data = flex_gd->groups; | |
1513 | ext4_fsblk_t o_blocks_count; | |
1514 | ext4_group_t n_group; | |
1515 | ext4_group_t group; | |
1516 | ext4_group_t last_group; | |
1517 | ext4_grpblk_t last; | |
1518 | ext4_grpblk_t blocks_per_group; | |
1519 | unsigned long i; | |
1520 | ||
1521 | blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb); | |
1522 | ||
1523 | o_blocks_count = ext4_blocks_count(es); | |
1524 | ||
1525 | if (o_blocks_count == n_blocks_count) | |
1526 | return 0; | |
1527 | ||
1528 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); | |
1529 | BUG_ON(last); | |
1530 | ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); | |
1531 | ||
1532 | last_group = group | (flexbg_size - 1); | |
1533 | if (last_group > n_group) | |
1534 | last_group = n_group; | |
1535 | ||
1536 | flex_gd->count = last_group - group + 1; | |
1537 | ||
1538 | for (i = 0; i < flex_gd->count; i++) { | |
1539 | int overhead; | |
1540 | ||
1541 | group_data[i].group = group + i; | |
1542 | group_data[i].blocks_count = blocks_per_group; | |
01f795f9 | 1543 | overhead = ext4_group_overhead_blocks(sb, group + i); |
19c5246d | 1544 | group_data[i].free_blocks_count = blocks_per_group - overhead; |
7f511862 | 1545 | if (ext4_has_group_desc_csum(sb)) { |
19c5246d YY |
1546 | flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT | |
1547 | EXT4_BG_INODE_UNINIT; | |
7f511862 TT |
1548 | if (!test_opt(sb, INIT_INODE_TABLE)) |
1549 | flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED; | |
1550 | } else | |
19c5246d YY |
1551 | flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED; |
1552 | } | |
1553 | ||
feb0ab32 | 1554 | if (last_group == n_group && ext4_has_group_desc_csum(sb)) |
19c5246d YY |
1555 | /* We need to initialize block bitmap of last group. */ |
1556 | flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT; | |
1557 | ||
1558 | if ((last_group == n_group) && (last != blocks_per_group - 1)) { | |
1559 | group_data[i - 1].blocks_count = last + 1; | |
1560 | group_data[i - 1].free_blocks_count -= blocks_per_group- | |
1561 | last - 1; | |
1562 | } | |
1563 | ||
1564 | return 1; | |
1565 | } | |
1566 | ||
ac27a0ec DK |
1567 | /* Add group descriptor data to an existing or new group descriptor block. |
1568 | * Ensure we handle all possible error conditions _before_ we start modifying | |
1569 | * the filesystem, because we cannot abort the transaction and not have it | |
1570 | * write the data to disk. | |
1571 | * | |
1572 | * If we are on a GDT block boundary, we need to get the reserved GDT block. | |
1573 | * Otherwise, we may need to add backup GDT blocks for a sparse group. | |
1574 | * | |
1575 | * We only need to hold the superblock lock while we are actually adding | |
1576 | * in the new group's counts to the superblock. Prior to that we have | |
1577 | * not really "added" the group at all. We re-check that we are still | |
1578 | * adding in the last group in case things have changed since verifying. | |
1579 | */ | |
617ba13b | 1580 | int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) |
ac27a0ec | 1581 | { |
61f296cc | 1582 | struct ext4_new_flex_group_data flex_gd; |
617ba13b MC |
1583 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
1584 | struct ext4_super_block *es = sbi->s_es; | |
1585 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec | 1586 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; |
ac27a0ec | 1587 | struct inode *inode = NULL; |
03b40e34 | 1588 | int gdb_off; |
61f296cc YY |
1589 | int err; |
1590 | __u16 bg_flags = 0; | |
ac27a0ec | 1591 | |
617ba13b | 1592 | gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); |
ac27a0ec | 1593 | |
e2b911c5 | 1594 | if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) { |
12062ddd | 1595 | ext4_warning(sb, "Can't resize non-sparse filesystem further"); |
ac27a0ec DK |
1596 | return -EPERM; |
1597 | } | |
1598 | ||
bd81d8ee LV |
1599 | if (ext4_blocks_count(es) + input->blocks_count < |
1600 | ext4_blocks_count(es)) { | |
12062ddd | 1601 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1602 | return -EINVAL; |
1603 | } | |
1604 | ||
617ba13b | 1605 | if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < |
ac27a0ec | 1606 | le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 1607 | ext4_warning(sb, "inodes_count overflow"); |
ac27a0ec DK |
1608 | return -EINVAL; |
1609 | } | |
1610 | ||
1611 | if (reserved_gdb || gdb_off == 0) { | |
0bf8b3fd | 1612 | if (!ext4_has_feature_resize_inode(sb) || |
e2b911c5 | 1613 | !le16_to_cpu(es->s_reserved_gdt_blocks)) { |
12062ddd | 1614 | ext4_warning(sb, |
ac27a0ec DK |
1615 | "No reserved GDT blocks, can't resize"); |
1616 | return -EPERM; | |
1617 | } | |
1d1fe1ee DH |
1618 | inode = ext4_iget(sb, EXT4_RESIZE_INO); |
1619 | if (IS_ERR(inode)) { | |
12062ddd | 1620 | ext4_warning(sb, "Error opening resize inode"); |
1d1fe1ee | 1621 | return PTR_ERR(inode); |
ac27a0ec DK |
1622 | } |
1623 | } | |
1624 | ||
920313a7 | 1625 | |
61f296cc | 1626 | err = verify_group_input(sb, input); |
08c3a813 | 1627 | if (err) |
61f296cc | 1628 | goto out; |
ac27a0ec | 1629 | |
117fff10 TT |
1630 | err = ext4_alloc_flex_bg_array(sb, input->group + 1); |
1631 | if (err) | |
7f511862 | 1632 | goto out; |
117fff10 | 1633 | |
28623c2f TT |
1634 | err = ext4_mb_alloc_groupinfo(sb, input->group + 1); |
1635 | if (err) | |
1636 | goto out; | |
1637 | ||
61f296cc YY |
1638 | flex_gd.count = 1; |
1639 | flex_gd.groups = input; | |
1640 | flex_gd.bg_flags = &bg_flags; | |
1641 | err = ext4_flex_group_add(sb, inode, &flex_gd); | |
1642 | out: | |
ac27a0ec DK |
1643 | iput(inode); |
1644 | return err; | |
617ba13b | 1645 | } /* ext4_group_add */ |
ac27a0ec | 1646 | |
18e31438 YY |
1647 | /* |
1648 | * extend a group without checking assuming that checking has been done. | |
1649 | */ | |
1650 | static int ext4_group_extend_no_check(struct super_block *sb, | |
1651 | ext4_fsblk_t o_blocks_count, ext4_grpblk_t add) | |
1652 | { | |
1653 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; | |
1654 | handle_t *handle; | |
1655 | int err = 0, err2; | |
1656 | ||
1657 | /* We will update the superblock, one block bitmap, and | |
1658 | * one group descriptor via ext4_group_add_blocks(). | |
1659 | */ | |
9924a92a | 1660 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3); |
18e31438 YY |
1661 | if (IS_ERR(handle)) { |
1662 | err = PTR_ERR(handle); | |
1663 | ext4_warning(sb, "error %d on journal start", err); | |
1664 | return err; | |
1665 | } | |
1666 | ||
5d601255 | 1667 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
18e31438 YY |
1668 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
1669 | if (err) { | |
1670 | ext4_warning(sb, "error %d on journal write access", err); | |
1671 | goto errout; | |
1672 | } | |
1673 | ||
1674 | ext4_blocks_count_set(es, o_blocks_count + add); | |
636d7e2e | 1675 | ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add); |
18e31438 YY |
1676 | ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, |
1677 | o_blocks_count + add); | |
1678 | /* We add the blocks to the bitmap and set the group need init bit */ | |
1679 | err = ext4_group_add_blocks(handle, sb, o_blocks_count, add); | |
1680 | if (err) | |
1681 | goto errout; | |
1682 | ext4_handle_dirty_super(handle, sb); | |
1683 | ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, | |
1684 | o_blocks_count + add); | |
1685 | errout: | |
1686 | err2 = ext4_journal_stop(handle); | |
1687 | if (err2 && !err) | |
1688 | err = err2; | |
1689 | ||
1690 | if (!err) { | |
1691 | if (test_opt(sb, DEBUG)) | |
1692 | printk(KERN_DEBUG "EXT4-fs: extended group to %llu " | |
1693 | "blocks\n", ext4_blocks_count(es)); | |
6ca792ed | 1694 | update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, |
01f795f9 | 1695 | (char *)es, sizeof(struct ext4_super_block), 0); |
18e31438 YY |
1696 | } |
1697 | return err; | |
1698 | } | |
1699 | ||
2b2d6d01 TT |
1700 | /* |
1701 | * Extend the filesystem to the new number of blocks specified. This entry | |
ac27a0ec DK |
1702 | * point is only used to extend the current filesystem to the end of the last |
1703 | * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" | |
1704 | * for emergencies (because it has no dependencies on reserved blocks). | |
1705 | * | |
617ba13b | 1706 | * If we _really_ wanted, we could use default values to call ext4_group_add() |
ac27a0ec DK |
1707 | * allow the "remount" trick to work for arbitrary resizing, assuming enough |
1708 | * GDT blocks are reserved to grow to the desired size. | |
1709 | */ | |
617ba13b MC |
1710 | int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, |
1711 | ext4_fsblk_t n_blocks_count) | |
ac27a0ec | 1712 | { |
617ba13b | 1713 | ext4_fsblk_t o_blocks_count; |
617ba13b MC |
1714 | ext4_grpblk_t last; |
1715 | ext4_grpblk_t add; | |
af5bc92d | 1716 | struct buffer_head *bh; |
d89651c8 | 1717 | int err; |
5f21b0e6 | 1718 | ext4_group_t group; |
ac27a0ec | 1719 | |
bd81d8ee | 1720 | o_blocks_count = ext4_blocks_count(es); |
ac27a0ec DK |
1721 | |
1722 | if (test_opt(sb, DEBUG)) | |
92b97816 TT |
1723 | ext4_msg(sb, KERN_DEBUG, |
1724 | "extending last group from %llu to %llu blocks", | |
1725 | o_blocks_count, n_blocks_count); | |
ac27a0ec DK |
1726 | |
1727 | if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) | |
1728 | return 0; | |
1729 | ||
1730 | if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { | |
92b97816 TT |
1731 | ext4_msg(sb, KERN_ERR, |
1732 | "filesystem too large to resize to %llu blocks safely", | |
1733 | n_blocks_count); | |
ac27a0ec | 1734 | if (sizeof(sector_t) < 8) |
12062ddd | 1735 | ext4_warning(sb, "CONFIG_LBDAF not enabled"); |
ac27a0ec DK |
1736 | return -EINVAL; |
1737 | } | |
1738 | ||
1739 | if (n_blocks_count < o_blocks_count) { | |
12062ddd | 1740 | ext4_warning(sb, "can't shrink FS - resize aborted"); |
8f82f840 | 1741 | return -EINVAL; |
ac27a0ec DK |
1742 | } |
1743 | ||
1744 | /* Handle the remaining blocks in the last group only. */ | |
5f21b0e6 | 1745 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); |
ac27a0ec DK |
1746 | |
1747 | if (last == 0) { | |
12062ddd | 1748 | ext4_warning(sb, "need to use ext2online to resize further"); |
ac27a0ec DK |
1749 | return -EPERM; |
1750 | } | |
1751 | ||
617ba13b | 1752 | add = EXT4_BLOCKS_PER_GROUP(sb) - last; |
ac27a0ec DK |
1753 | |
1754 | if (o_blocks_count + add < o_blocks_count) { | |
12062ddd | 1755 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1756 | return -EINVAL; |
1757 | } | |
1758 | ||
1759 | if (o_blocks_count + add > n_blocks_count) | |
1760 | add = n_blocks_count - o_blocks_count; | |
1761 | ||
1762 | if (o_blocks_count + add < n_blocks_count) | |
12062ddd | 1763 | ext4_warning(sb, "will only finish group (%llu blocks, %u new)", |
ac27a0ec DK |
1764 | o_blocks_count + add, add); |
1765 | ||
1766 | /* See if the device is actually as big as what was requested */ | |
2b2d6d01 | 1767 | bh = sb_bread(sb, o_blocks_count + add - 1); |
ac27a0ec | 1768 | if (!bh) { |
12062ddd | 1769 | ext4_warning(sb, "can't read last block, resize aborted"); |
ac27a0ec DK |
1770 | return -ENOSPC; |
1771 | } | |
1772 | brelse(bh); | |
1773 | ||
d89651c8 | 1774 | err = ext4_group_extend_no_check(sb, o_blocks_count, add); |
ac27a0ec | 1775 | return err; |
617ba13b | 1776 | } /* ext4_group_extend */ |
19c5246d | 1777 | |
1c6bd717 TT |
1778 | |
1779 | static int num_desc_blocks(struct super_block *sb, ext4_group_t groups) | |
1780 | { | |
1781 | return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb); | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Release the resize inode and drop the resize_inode feature if there | |
1786 | * are no more reserved gdt blocks, and then convert the file system | |
1787 | * to enable meta_bg | |
1788 | */ | |
1789 | static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode) | |
1790 | { | |
1791 | handle_t *handle; | |
1792 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1793 | struct ext4_super_block *es = sbi->s_es; | |
59e31c15 | 1794 | struct ext4_inode_info *ei = EXT4_I(inode); |
1c6bd717 TT |
1795 | ext4_fsblk_t nr; |
1796 | int i, ret, err = 0; | |
1797 | int credits = 1; | |
1798 | ||
1799 | ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg"); | |
59e31c15 | 1800 | if (inode) { |
1c6bd717 TT |
1801 | if (es->s_reserved_gdt_blocks) { |
1802 | ext4_error(sb, "Unexpected non-zero " | |
1803 | "s_reserved_gdt_blocks"); | |
1804 | return -EPERM; | |
1805 | } | |
1c6bd717 TT |
1806 | |
1807 | /* Do a quick sanity check of the resize inode */ | |
1808 | if (inode->i_blocks != 1 << (inode->i_blkbits - 9)) | |
1809 | goto invalid_resize_inode; | |
1810 | for (i = 0; i < EXT4_N_BLOCKS; i++) { | |
1811 | if (i == EXT4_DIND_BLOCK) { | |
1812 | if (ei->i_data[i]) | |
1813 | continue; | |
1814 | else | |
1815 | goto invalid_resize_inode; | |
1816 | } | |
1817 | if (ei->i_data[i]) | |
1818 | goto invalid_resize_inode; | |
1819 | } | |
1820 | credits += 3; /* block bitmap, bg descriptor, resize inode */ | |
1821 | } | |
1822 | ||
9924a92a | 1823 | handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits); |
1c6bd717 TT |
1824 | if (IS_ERR(handle)) |
1825 | return PTR_ERR(handle); | |
1826 | ||
5d601255 | 1827 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
1c6bd717 TT |
1828 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
1829 | if (err) | |
1830 | goto errout; | |
1831 | ||
e2b911c5 DW |
1832 | ext4_clear_feature_resize_inode(sb); |
1833 | ext4_set_feature_meta_bg(sb); | |
1c6bd717 TT |
1834 | sbi->s_es->s_first_meta_bg = |
1835 | cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count)); | |
1836 | ||
1837 | err = ext4_handle_dirty_super(handle, sb); | |
1838 | if (err) { | |
1839 | ext4_std_error(sb, err); | |
1840 | goto errout; | |
1841 | } | |
1842 | ||
1843 | if (inode) { | |
1844 | nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]); | |
1845 | ext4_free_blocks(handle, inode, NULL, nr, 1, | |
1846 | EXT4_FREE_BLOCKS_METADATA | | |
1847 | EXT4_FREE_BLOCKS_FORGET); | |
1848 | ei->i_data[EXT4_DIND_BLOCK] = 0; | |
1849 | inode->i_blocks = 0; | |
1850 | ||
1851 | err = ext4_mark_inode_dirty(handle, inode); | |
1852 | if (err) | |
1853 | ext4_std_error(sb, err); | |
1854 | } | |
1855 | ||
1856 | errout: | |
1857 | ret = ext4_journal_stop(handle); | |
1858 | if (!err) | |
1859 | err = ret; | |
1860 | return ret; | |
1861 | ||
1862 | invalid_resize_inode: | |
1863 | ext4_error(sb, "corrupted/inconsistent resize inode"); | |
1864 | return -EINVAL; | |
1865 | } | |
1866 | ||
19c5246d YY |
1867 | /* |
1868 | * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count | |
1869 | * | |
1870 | * @sb: super block of the fs to be resized | |
1871 | * @n_blocks_count: the number of blocks resides in the resized fs | |
1872 | */ | |
1873 | int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) | |
1874 | { | |
1875 | struct ext4_new_flex_group_data *flex_gd = NULL; | |
1876 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1877 | struct ext4_super_block *es = sbi->s_es; | |
1878 | struct buffer_head *bh; | |
01f795f9 YY |
1879 | struct inode *resize_inode = NULL; |
1880 | ext4_grpblk_t add, offset; | |
19c5246d YY |
1881 | unsigned long n_desc_blocks; |
1882 | unsigned long o_desc_blocks; | |
01f795f9 YY |
1883 | ext4_group_t o_group; |
1884 | ext4_group_t n_group; | |
1885 | ext4_fsblk_t o_blocks_count; | |
1c6bd717 | 1886 | ext4_fsblk_t n_blocks_count_retry = 0; |
4da4a56e | 1887 | unsigned long last_update_time = 0; |
117fff10 | 1888 | int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex; |
01f795f9 | 1889 | int meta_bg; |
19c5246d | 1890 | |
59e31c15 TT |
1891 | /* See if the device is actually as big as what was requested */ |
1892 | bh = sb_bread(sb, n_blocks_count - 1); | |
1893 | if (!bh) { | |
1894 | ext4_warning(sb, "can't read last block, resize aborted"); | |
1895 | return -ENOSPC; | |
1896 | } | |
1897 | brelse(bh); | |
1898 | ||
1c6bd717 | 1899 | retry: |
19c5246d YY |
1900 | o_blocks_count = ext4_blocks_count(es); |
1901 | ||
59e31c15 TT |
1902 | ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu " |
1903 | "to %llu blocks", o_blocks_count, n_blocks_count); | |
19c5246d YY |
1904 | |
1905 | if (n_blocks_count < o_blocks_count) { | |
1906 | /* On-line shrinking not supported */ | |
1907 | ext4_warning(sb, "can't shrink FS - resize aborted"); | |
1908 | return -EINVAL; | |
1909 | } | |
1910 | ||
1911 | if (n_blocks_count == o_blocks_count) | |
1912 | /* Nothing need to do */ | |
1913 | return 0; | |
1914 | ||
bd86298e | 1915 | n_group = ext4_get_group_number(sb, n_blocks_count - 1); |
f3233cb2 | 1916 | if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) { |
3f8a6411 TT |
1917 | ext4_warning(sb, "resize would cause inodes_count overflow"); |
1918 | return -EINVAL; | |
1919 | } | |
a0ade1de | 1920 | ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset); |
19c5246d | 1921 | |
1c6bd717 TT |
1922 | n_desc_blocks = num_desc_blocks(sb, n_group + 1); |
1923 | o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count); | |
19c5246d | 1924 | |
e2b911c5 | 1925 | meta_bg = ext4_has_feature_meta_bg(sb); |
19c5246d | 1926 | |
e2b911c5 | 1927 | if (ext4_has_feature_resize_inode(sb)) { |
01f795f9 YY |
1928 | if (meta_bg) { |
1929 | ext4_error(sb, "resize_inode and meta_bg enabled " | |
1930 | "simultaneously"); | |
1931 | return -EINVAL; | |
1932 | } | |
1c6bd717 TT |
1933 | if (n_desc_blocks > o_desc_blocks + |
1934 | le16_to_cpu(es->s_reserved_gdt_blocks)) { | |
1935 | n_blocks_count_retry = n_blocks_count; | |
1936 | n_desc_blocks = o_desc_blocks + | |
1937 | le16_to_cpu(es->s_reserved_gdt_blocks); | |
1938 | n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb); | |
aec51758 | 1939 | n_blocks_count = (ext4_fsblk_t)n_group * |
803b8b94 JK |
1940 | EXT4_BLOCKS_PER_GROUP(sb) + |
1941 | le32_to_cpu(es->s_first_data_block); | |
1c6bd717 | 1942 | n_group--; /* set to last group number */ |
01f795f9 | 1943 | } |
1c6bd717 TT |
1944 | |
1945 | if (!resize_inode) | |
1946 | resize_inode = ext4_iget(sb, EXT4_RESIZE_INO); | |
01f795f9 YY |
1947 | if (IS_ERR(resize_inode)) { |
1948 | ext4_warning(sb, "Error opening resize inode"); | |
1949 | return PTR_ERR(resize_inode); | |
1950 | } | |
1c6bd717 TT |
1951 | } |
1952 | ||
59e31c15 | 1953 | if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) { |
1c6bd717 TT |
1954 | err = ext4_convert_meta_bg(sb, resize_inode); |
1955 | if (err) | |
1956 | goto out; | |
1957 | if (resize_inode) { | |
1958 | iput(resize_inode); | |
1959 | resize_inode = NULL; | |
1960 | } | |
1961 | if (n_blocks_count_retry) { | |
1962 | n_blocks_count = n_blocks_count_retry; | |
1963 | n_blocks_count_retry = 0; | |
1964 | goto retry; | |
1965 | } | |
19c5246d YY |
1966 | } |
1967 | ||
6a4d7b58 TT |
1968 | /* |
1969 | * Make sure the last group has enough space so that it's | |
1970 | * guaranteed to have enough space for all metadata blocks | |
1971 | * that it might need to hold. (We might not need to store | |
1972 | * the inode table blocks in the last block group, but there | |
1973 | * will be cases where this might be needed.) | |
1974 | */ | |
1975 | if ((ext4_group_first_block_no(sb, n_group) + | |
1976 | ext4_group_overhead_blocks(sb, n_group) + 2 + | |
1977 | sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) { | |
1978 | n_blocks_count = ext4_group_first_block_no(sb, n_group); | |
1979 | n_group--; | |
1980 | n_blocks_count_retry = 0; | |
1981 | if (resize_inode) { | |
1982 | iput(resize_inode); | |
1983 | resize_inode = NULL; | |
1984 | } | |
1985 | goto retry; | |
1986 | } | |
1987 | ||
a0ade1de LC |
1988 | /* extend the last group */ |
1989 | if (n_group == o_group) | |
1990 | add = n_blocks_count - o_blocks_count; | |
1991 | else | |
1992 | add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1); | |
1993 | if (add > 0) { | |
19c5246d YY |
1994 | err = ext4_group_extend_no_check(sb, o_blocks_count, add); |
1995 | if (err) | |
1996 | goto out; | |
1997 | } | |
1998 | ||
d7574ad0 | 1999 | if (ext4_blocks_count(es) == n_blocks_count) |
19c5246d YY |
2000 | goto out; |
2001 | ||
117fff10 TT |
2002 | err = ext4_alloc_flex_bg_array(sb, n_group + 1); |
2003 | if (err) | |
65b1ce49 | 2004 | goto out; |
117fff10 | 2005 | |
28623c2f TT |
2006 | err = ext4_mb_alloc_groupinfo(sb, n_group + 1); |
2007 | if (err) | |
2008 | goto out; | |
2009 | ||
19c5246d YY |
2010 | flex_gd = alloc_flex_gd(flexbg_size); |
2011 | if (flex_gd == NULL) { | |
2012 | err = -ENOMEM; | |
2013 | goto out; | |
2014 | } | |
2015 | ||
2016 | /* Add flex groups. Note that a regular group is a | |
2017 | * flex group with 1 group. | |
2018 | */ | |
2019 | while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count, | |
2020 | flexbg_size)) { | |
4da4a56e TT |
2021 | if (jiffies - last_update_time > HZ * 10) { |
2022 | if (last_update_time) | |
2023 | ext4_msg(sb, KERN_INFO, | |
2024 | "resized to %llu blocks", | |
2025 | ext4_blocks_count(es)); | |
2026 | last_update_time = jiffies; | |
2027 | } | |
03c1c290 YY |
2028 | if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0) |
2029 | break; | |
19c5246d YY |
2030 | err = ext4_flex_group_add(sb, resize_inode, flex_gd); |
2031 | if (unlikely(err)) | |
2032 | break; | |
2033 | } | |
2034 | ||
1c6bd717 TT |
2035 | if (!err && n_blocks_count_retry) { |
2036 | n_blocks_count = n_blocks_count_retry; | |
2037 | n_blocks_count_retry = 0; | |
2038 | free_flex_gd(flex_gd); | |
2039 | flex_gd = NULL; | |
3caa7b62 VA |
2040 | if (resize_inode) { |
2041 | iput(resize_inode); | |
2042 | resize_inode = NULL; | |
2043 | } | |
1c6bd717 TT |
2044 | goto retry; |
2045 | } | |
2046 | ||
19c5246d YY |
2047 | out: |
2048 | if (flex_gd) | |
2049 | free_flex_gd(flex_gd); | |
01f795f9 YY |
2050 | if (resize_inode != NULL) |
2051 | iput(resize_inode); | |
ce1bdbed LC |
2052 | if (err) |
2053 | ext4_warning(sb, "error (%d) occurred during " | |
2054 | "file system resize", err); | |
2055 | ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", | |
2056 | ext4_blocks_count(es)); | |
19c5246d YY |
2057 | return err; |
2058 | } |