]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/resize.c |
ac27a0ec | 3 | * |
617ba13b | 4 | * Support for resizing an ext4 filesystem while it is mounted. |
ac27a0ec DK |
5 | * |
6 | * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com> | |
7 | * | |
8 | * This could probably be made into a module, because it is not often in use. | |
9 | */ | |
10 | ||
11 | ||
617ba13b | 12 | #define EXT4FS_DEBUG |
ac27a0ec | 13 | |
ac27a0ec DK |
14 | #include <linux/errno.h> |
15 | #include <linux/slab.h> | |
16 | ||
3dcf5451 | 17 | #include "ext4_jbd2.h" |
ac27a0ec DK |
18 | |
19 | #define outside(b, first, last) ((b) < (first) || (b) >= (last)) | |
20 | #define inside(b, first, last) ((b) >= (first) && (b) < (last)) | |
21 | ||
22 | static int verify_group_input(struct super_block *sb, | |
617ba13b | 23 | struct ext4_new_group_data *input) |
ac27a0ec | 24 | { |
617ba13b MC |
25 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
26 | struct ext4_super_block *es = sbi->s_es; | |
bd81d8ee | 27 | ext4_fsblk_t start = ext4_blocks_count(es); |
617ba13b | 28 | ext4_fsblk_t end = start + input->blocks_count; |
fd2d4291 | 29 | ext4_group_t group = input->group; |
617ba13b MC |
30 | ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group; |
31 | unsigned overhead = ext4_bg_has_super(sb, group) ? | |
32 | (1 + ext4_bg_num_gdb(sb, group) + | |
ac27a0ec | 33 | le16_to_cpu(es->s_reserved_gdt_blocks)) : 0; |
617ba13b | 34 | ext4_fsblk_t metaend = start + overhead; |
ac27a0ec | 35 | struct buffer_head *bh = NULL; |
3a5b2ecd | 36 | ext4_grpblk_t free_blocks_count, offset; |
ac27a0ec DK |
37 | int err = -EINVAL; |
38 | ||
39 | input->free_blocks_count = free_blocks_count = | |
40 | input->blocks_count - 2 - overhead - sbi->s_itb_per_group; | |
41 | ||
42 | if (test_opt(sb, DEBUG)) | |
617ba13b | 43 | printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks " |
ac27a0ec | 44 | "(%d free, %u reserved)\n", |
617ba13b | 45 | ext4_bg_has_super(sb, input->group) ? "normal" : |
ac27a0ec DK |
46 | "no-super", input->group, input->blocks_count, |
47 | free_blocks_count, input->reserved_blocks); | |
48 | ||
3a5b2ecd | 49 | ext4_get_group_no_and_offset(sb, start, NULL, &offset); |
ac27a0ec | 50 | if (group != sbi->s_groups_count) |
12062ddd | 51 | ext4_warning(sb, "Cannot add at group %u (only %u groups)", |
ac27a0ec | 52 | input->group, sbi->s_groups_count); |
3a5b2ecd | 53 | else if (offset != 0) |
12062ddd | 54 | ext4_warning(sb, "Last group not full"); |
ac27a0ec | 55 | else if (input->reserved_blocks > input->blocks_count / 5) |
12062ddd | 56 | ext4_warning(sb, "Reserved blocks too high (%u)", |
ac27a0ec DK |
57 | input->reserved_blocks); |
58 | else if (free_blocks_count < 0) | |
12062ddd | 59 | ext4_warning(sb, "Bad blocks count %u", |
ac27a0ec DK |
60 | input->blocks_count); |
61 | else if (!(bh = sb_bread(sb, end - 1))) | |
12062ddd | 62 | ext4_warning(sb, "Cannot read last block (%llu)", |
ac27a0ec DK |
63 | end - 1); |
64 | else if (outside(input->block_bitmap, start, end)) | |
12062ddd | 65 | ext4_warning(sb, "Block bitmap not in group (block %llu)", |
1939e49a | 66 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 67 | else if (outside(input->inode_bitmap, start, end)) |
12062ddd | 68 | ext4_warning(sb, "Inode bitmap not in group (block %llu)", |
1939e49a | 69 | (unsigned long long)input->inode_bitmap); |
ac27a0ec | 70 | else if (outside(input->inode_table, start, end) || |
2b2d6d01 | 71 | outside(itend - 1, start, end)) |
12062ddd | 72 | ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)", |
1939e49a | 73 | (unsigned long long)input->inode_table, itend - 1); |
ac27a0ec | 74 | else if (input->inode_bitmap == input->block_bitmap) |
12062ddd | 75 | ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)", |
1939e49a | 76 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 77 | else if (inside(input->block_bitmap, input->inode_table, itend)) |
12062ddd ES |
78 | ext4_warning(sb, "Block bitmap (%llu) in inode table " |
79 | "(%llu-%llu)", | |
1939e49a RD |
80 | (unsigned long long)input->block_bitmap, |
81 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 82 | else if (inside(input->inode_bitmap, input->inode_table, itend)) |
12062ddd ES |
83 | ext4_warning(sb, "Inode bitmap (%llu) in inode table " |
84 | "(%llu-%llu)", | |
1939e49a RD |
85 | (unsigned long long)input->inode_bitmap, |
86 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 87 | else if (inside(input->block_bitmap, start, metaend)) |
12062ddd | 88 | ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
89 | (unsigned long long)input->block_bitmap, |
90 | start, metaend - 1); | |
ac27a0ec | 91 | else if (inside(input->inode_bitmap, start, metaend)) |
12062ddd | 92 | ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
93 | (unsigned long long)input->inode_bitmap, |
94 | start, metaend - 1); | |
ac27a0ec | 95 | else if (inside(input->inode_table, start, metaend) || |
2b2d6d01 | 96 | inside(itend - 1, start, metaend)) |
12062ddd ES |
97 | ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table " |
98 | "(%llu-%llu)", | |
1939e49a RD |
99 | (unsigned long long)input->inode_table, |
100 | itend - 1, start, metaend - 1); | |
ac27a0ec DK |
101 | else |
102 | err = 0; | |
103 | brelse(bh); | |
104 | ||
105 | return err; | |
106 | } | |
107 | ||
108 | static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, | |
617ba13b | 109 | ext4_fsblk_t blk) |
ac27a0ec DK |
110 | { |
111 | struct buffer_head *bh; | |
112 | int err; | |
113 | ||
114 | bh = sb_getblk(sb, blk); | |
115 | if (!bh) | |
116 | return ERR_PTR(-EIO); | |
617ba13b | 117 | if ((err = ext4_journal_get_write_access(handle, bh))) { |
ac27a0ec DK |
118 | brelse(bh); |
119 | bh = ERR_PTR(err); | |
120 | } else { | |
121 | lock_buffer(bh); | |
122 | memset(bh->b_data, 0, sb->s_blocksize); | |
123 | set_buffer_uptodate(bh); | |
124 | unlock_buffer(bh); | |
125 | } | |
126 | ||
127 | return bh; | |
128 | } | |
129 | ||
14904107 ES |
130 | /* |
131 | * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA. | |
132 | * If that fails, restart the transaction & regain write access for the | |
133 | * buffer head which is used for block_bitmap modifications. | |
134 | */ | |
135 | static int extend_or_restart_transaction(handle_t *handle, int thresh, | |
136 | struct buffer_head *bh) | |
137 | { | |
138 | int err; | |
139 | ||
0390131b | 140 | if (ext4_handle_has_enough_credits(handle, thresh)) |
14904107 ES |
141 | return 0; |
142 | ||
143 | err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA); | |
144 | if (err < 0) | |
145 | return err; | |
146 | if (err) { | |
147 | if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) | |
148 | return err; | |
2b2d6d01 | 149 | if ((err = ext4_journal_get_write_access(handle, bh))) |
14904107 | 150 | return err; |
2b2d6d01 | 151 | } |
14904107 ES |
152 | |
153 | return 0; | |
154 | } | |
155 | ||
ac27a0ec DK |
156 | /* |
157 | * Set up the block and inode bitmaps, and the inode table for the new group. | |
158 | * This doesn't need to be part of the main transaction, since we are only | |
159 | * changing blocks outside the actual filesystem. We still do journaling to | |
160 | * ensure the recovery is correct in case of a failure just after resize. | |
161 | * If any part of this fails, we simply abort the resize. | |
162 | */ | |
163 | static int setup_new_group_blocks(struct super_block *sb, | |
617ba13b | 164 | struct ext4_new_group_data *input) |
ac27a0ec | 165 | { |
617ba13b MC |
166 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
167 | ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group); | |
168 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec | 169 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0; |
617ba13b | 170 | unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group); |
ac27a0ec DK |
171 | struct buffer_head *bh; |
172 | handle_t *handle; | |
617ba13b MC |
173 | ext4_fsblk_t block; |
174 | ext4_grpblk_t bit; | |
ac27a0ec DK |
175 | int i; |
176 | int err = 0, err2; | |
177 | ||
14904107 ES |
178 | /* This transaction may be extended/restarted along the way */ |
179 | handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA); | |
180 | ||
ac27a0ec DK |
181 | if (IS_ERR(handle)) |
182 | return PTR_ERR(handle); | |
183 | ||
32ed5058 | 184 | mutex_lock(&sbi->s_resize_lock); |
ac27a0ec DK |
185 | if (input->group != sbi->s_groups_count) { |
186 | err = -EBUSY; | |
187 | goto exit_journal; | |
188 | } | |
189 | ||
190 | if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) { | |
191 | err = PTR_ERR(bh); | |
192 | goto exit_journal; | |
193 | } | |
194 | ||
617ba13b | 195 | if (ext4_bg_has_super(sb, input->group)) { |
c549a95d | 196 | ext4_debug("mark backup superblock %#04llx (+0)\n", start); |
617ba13b | 197 | ext4_set_bit(0, bh->b_data); |
ac27a0ec DK |
198 | } |
199 | ||
200 | /* Copy all of the GDT blocks into the backup in this group */ | |
201 | for (i = 0, bit = 1, block = start + 1; | |
202 | i < gdblocks; i++, block++, bit++) { | |
203 | struct buffer_head *gdb; | |
204 | ||
c549a95d | 205 | ext4_debug("update backup group %#04llx (+%d)\n", block, bit); |
ac27a0ec | 206 | |
14904107 ES |
207 | if ((err = extend_or_restart_transaction(handle, 1, bh))) |
208 | goto exit_bh; | |
209 | ||
ac27a0ec DK |
210 | gdb = sb_getblk(sb, block); |
211 | if (!gdb) { | |
212 | err = -EIO; | |
213 | goto exit_bh; | |
214 | } | |
617ba13b | 215 | if ((err = ext4_journal_get_write_access(handle, gdb))) { |
ac27a0ec DK |
216 | brelse(gdb); |
217 | goto exit_bh; | |
218 | } | |
5b615287 ES |
219 | lock_buffer(gdb); |
220 | memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size); | |
ac27a0ec | 221 | set_buffer_uptodate(gdb); |
5b615287 | 222 | unlock_buffer(gdb); |
0390131b | 223 | ext4_handle_dirty_metadata(handle, NULL, gdb); |
617ba13b | 224 | ext4_set_bit(bit, bh->b_data); |
ac27a0ec DK |
225 | brelse(gdb); |
226 | } | |
227 | ||
228 | /* Zero out all of the reserved backup group descriptor table blocks */ | |
229 | for (i = 0, bit = gdblocks + 1, block = start + bit; | |
230 | i < reserved_gdb; i++, block++, bit++) { | |
231 | struct buffer_head *gdb; | |
232 | ||
c549a95d | 233 | ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit); |
ac27a0ec | 234 | |
14904107 ES |
235 | if ((err = extend_or_restart_transaction(handle, 1, bh))) |
236 | goto exit_bh; | |
237 | ||
ac27a0ec | 238 | if (IS_ERR(gdb = bclean(handle, sb, block))) { |
c09eef30 | 239 | err = PTR_ERR(gdb); |
ac27a0ec DK |
240 | goto exit_bh; |
241 | } | |
0390131b | 242 | ext4_handle_dirty_metadata(handle, NULL, gdb); |
617ba13b | 243 | ext4_set_bit(bit, bh->b_data); |
ac27a0ec DK |
244 | brelse(gdb); |
245 | } | |
c549a95d | 246 | ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap, |
ac27a0ec | 247 | input->block_bitmap - start); |
617ba13b | 248 | ext4_set_bit(input->block_bitmap - start, bh->b_data); |
c549a95d | 249 | ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap, |
ac27a0ec | 250 | input->inode_bitmap - start); |
617ba13b | 251 | ext4_set_bit(input->inode_bitmap - start, bh->b_data); |
ac27a0ec DK |
252 | |
253 | /* Zero out all of the inode table blocks */ | |
254 | for (i = 0, block = input->inode_table, bit = block - start; | |
255 | i < sbi->s_itb_per_group; i++, bit++, block++) { | |
256 | struct buffer_head *it; | |
257 | ||
c549a95d | 258 | ext4_debug("clear inode block %#04llx (+%d)\n", block, bit); |
14904107 ES |
259 | |
260 | if ((err = extend_or_restart_transaction(handle, 1, bh))) | |
261 | goto exit_bh; | |
262 | ||
ac27a0ec DK |
263 | if (IS_ERR(it = bclean(handle, sb, block))) { |
264 | err = PTR_ERR(it); | |
265 | goto exit_bh; | |
266 | } | |
0390131b | 267 | ext4_handle_dirty_metadata(handle, NULL, it); |
ac27a0ec | 268 | brelse(it); |
617ba13b | 269 | ext4_set_bit(bit, bh->b_data); |
ac27a0ec | 270 | } |
14904107 ES |
271 | |
272 | if ((err = extend_or_restart_transaction(handle, 2, bh))) | |
273 | goto exit_bh; | |
274 | ||
648f5879 | 275 | mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, bh->b_data); |
0390131b | 276 | ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec | 277 | brelse(bh); |
ac27a0ec | 278 | /* Mark unused entries in inode bitmap used */ |
c549a95d | 279 | ext4_debug("clear inode bitmap %#04llx (+%llu)\n", |
ac27a0ec DK |
280 | input->inode_bitmap, input->inode_bitmap - start); |
281 | if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) { | |
282 | err = PTR_ERR(bh); | |
283 | goto exit_journal; | |
284 | } | |
285 | ||
648f5879 | 286 | mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
ac27a0ec | 287 | bh->b_data); |
0390131b | 288 | ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
289 | exit_bh: |
290 | brelse(bh); | |
291 | ||
292 | exit_journal: | |
32ed5058 | 293 | mutex_unlock(&sbi->s_resize_lock); |
617ba13b | 294 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
295 | err = err2; |
296 | ||
297 | return err; | |
298 | } | |
299 | ||
300 | /* | |
301 | * Iterate through the groups which hold BACKUP superblock/GDT copies in an | |
617ba13b | 302 | * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before |
ac27a0ec DK |
303 | * calling this for the first time. In a sparse filesystem it will be the |
304 | * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... | |
305 | * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... | |
306 | */ | |
617ba13b | 307 | static unsigned ext4_list_backups(struct super_block *sb, unsigned *three, |
ac27a0ec DK |
308 | unsigned *five, unsigned *seven) |
309 | { | |
310 | unsigned *min = three; | |
311 | int mult = 3; | |
312 | unsigned ret; | |
313 | ||
617ba13b MC |
314 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
315 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
ac27a0ec DK |
316 | ret = *min; |
317 | *min += 1; | |
318 | return ret; | |
319 | } | |
320 | ||
321 | if (*five < *min) { | |
322 | min = five; | |
323 | mult = 5; | |
324 | } | |
325 | if (*seven < *min) { | |
326 | min = seven; | |
327 | mult = 7; | |
328 | } | |
329 | ||
330 | ret = *min; | |
331 | *min *= mult; | |
332 | ||
333 | return ret; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Check that all of the backup GDT blocks are held in the primary GDT block. | |
338 | * It is assumed that they are stored in group order. Returns the number of | |
339 | * groups in current filesystem that have BACKUPS, or -ve error code. | |
340 | */ | |
341 | static int verify_reserved_gdb(struct super_block *sb, | |
342 | struct buffer_head *primary) | |
343 | { | |
617ba13b | 344 | const ext4_fsblk_t blk = primary->b_blocknr; |
fd2d4291 | 345 | const ext4_group_t end = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
346 | unsigned three = 1; |
347 | unsigned five = 5; | |
348 | unsigned seven = 7; | |
349 | unsigned grp; | |
350 | __le32 *p = (__le32 *)primary->b_data; | |
351 | int gdbackups = 0; | |
352 | ||
617ba13b | 353 | while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { |
bd81d8ee LV |
354 | if (le32_to_cpu(*p++) != |
355 | grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ | |
12062ddd | 356 | ext4_warning(sb, "reserved GDT %llu" |
2ae02107 | 357 | " missing grp %d (%llu)", |
ac27a0ec | 358 | blk, grp, |
bd81d8ee LV |
359 | grp * |
360 | (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + | |
361 | blk); | |
ac27a0ec DK |
362 | return -EINVAL; |
363 | } | |
617ba13b | 364 | if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) |
ac27a0ec DK |
365 | return -EFBIG; |
366 | } | |
367 | ||
368 | return gdbackups; | |
369 | } | |
370 | ||
371 | /* | |
372 | * Called when we need to bring a reserved group descriptor table block into | |
373 | * use from the resize inode. The primary copy of the new GDT block currently | |
374 | * is an indirect block (under the double indirect block in the resize inode). | |
375 | * The new backup GDT blocks will be stored as leaf blocks in this indirect | |
376 | * block, in group order. Even though we know all the block numbers we need, | |
377 | * we check to ensure that the resize inode has actually reserved these blocks. | |
378 | * | |
379 | * Don't need to update the block bitmaps because the blocks are still in use. | |
380 | * | |
381 | * We get all of the error cases out of the way, so that we are sure to not | |
382 | * fail once we start modifying the data on disk, because JBD has no rollback. | |
383 | */ | |
384 | static int add_new_gdb(handle_t *handle, struct inode *inode, | |
617ba13b | 385 | struct ext4_new_group_data *input, |
ac27a0ec DK |
386 | struct buffer_head **primary) |
387 | { | |
388 | struct super_block *sb = inode->i_sb; | |
617ba13b MC |
389 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
390 | unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb); | |
391 | ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; | |
ac27a0ec DK |
392 | struct buffer_head **o_group_desc, **n_group_desc; |
393 | struct buffer_head *dind; | |
394 | int gdbackups; | |
617ba13b | 395 | struct ext4_iloc iloc; |
ac27a0ec DK |
396 | __le32 *data; |
397 | int err; | |
398 | ||
399 | if (test_opt(sb, DEBUG)) | |
400 | printk(KERN_DEBUG | |
617ba13b | 401 | "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", |
ac27a0ec DK |
402 | gdb_num); |
403 | ||
af5bc92d TT |
404 | /* |
405 | * If we are not using the primary superblock/GDT copy don't resize, | |
2b2d6d01 TT |
406 | * because the user tools have no way of handling this. Probably a |
407 | * bad time to do it anyways. | |
408 | */ | |
617ba13b MC |
409 | if (EXT4_SB(sb)->s_sbh->b_blocknr != |
410 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) { | |
12062ddd | 411 | ext4_warning(sb, "won't resize using backup superblock at %llu", |
617ba13b | 412 | (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr); |
ac27a0ec DK |
413 | return -EPERM; |
414 | } | |
415 | ||
416 | *primary = sb_bread(sb, gdblock); | |
417 | if (!*primary) | |
418 | return -EIO; | |
419 | ||
420 | if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) { | |
421 | err = gdbackups; | |
422 | goto exit_bh; | |
423 | } | |
424 | ||
617ba13b | 425 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
ac27a0ec DK |
426 | dind = sb_bread(sb, le32_to_cpu(*data)); |
427 | if (!dind) { | |
428 | err = -EIO; | |
429 | goto exit_bh; | |
430 | } | |
431 | ||
432 | data = (__le32 *)dind->b_data; | |
617ba13b | 433 | if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { |
12062ddd | 434 | ext4_warning(sb, "new group %u GDT block %llu not reserved", |
ac27a0ec DK |
435 | input->group, gdblock); |
436 | err = -EINVAL; | |
437 | goto exit_dind; | |
438 | } | |
439 | ||
617ba13b | 440 | if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh))) |
ac27a0ec DK |
441 | goto exit_dind; |
442 | ||
617ba13b | 443 | if ((err = ext4_journal_get_write_access(handle, *primary))) |
ac27a0ec DK |
444 | goto exit_sbh; |
445 | ||
617ba13b | 446 | if ((err = ext4_journal_get_write_access(handle, dind))) |
ac27a0ec DK |
447 | goto exit_primary; |
448 | ||
617ba13b MC |
449 | /* ext4_reserve_inode_write() gets a reference on the iloc */ |
450 | if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) | |
ac27a0ec DK |
451 | goto exit_dindj; |
452 | ||
453 | n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *), | |
216553c4 | 454 | GFP_NOFS); |
ac27a0ec DK |
455 | if (!n_group_desc) { |
456 | err = -ENOMEM; | |
12062ddd | 457 | ext4_warning(sb, |
ac27a0ec DK |
458 | "not enough memory for %lu groups", gdb_num + 1); |
459 | goto exit_inode; | |
460 | } | |
461 | ||
462 | /* | |
463 | * Finally, we have all of the possible failures behind us... | |
464 | * | |
465 | * Remove new GDT block from inode double-indirect block and clear out | |
466 | * the new GDT block for use (which also "frees" the backup GDT blocks | |
467 | * from the reserved inode). We don't need to change the bitmaps for | |
468 | * these blocks, because they are marked as in-use from being in the | |
469 | * reserved inode, and will become GDT blocks (primary and backup). | |
470 | */ | |
617ba13b | 471 | data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; |
0390131b | 472 | ext4_handle_dirty_metadata(handle, NULL, dind); |
ac27a0ec DK |
473 | brelse(dind); |
474 | inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9; | |
617ba13b | 475 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec | 476 | memset((*primary)->b_data, 0, sb->s_blocksize); |
0390131b | 477 | ext4_handle_dirty_metadata(handle, NULL, *primary); |
ac27a0ec | 478 | |
617ba13b | 479 | o_group_desc = EXT4_SB(sb)->s_group_desc; |
ac27a0ec | 480 | memcpy(n_group_desc, o_group_desc, |
617ba13b | 481 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); |
ac27a0ec | 482 | n_group_desc[gdb_num] = *primary; |
617ba13b MC |
483 | EXT4_SB(sb)->s_group_desc = n_group_desc; |
484 | EXT4_SB(sb)->s_gdb_count++; | |
ac27a0ec DK |
485 | kfree(o_group_desc); |
486 | ||
e8546d06 | 487 | le16_add_cpu(&es->s_reserved_gdt_blocks, -1); |
0390131b | 488 | ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
489 | |
490 | return 0; | |
491 | ||
492 | exit_inode: | |
2b2d6d01 | 493 | /* ext4_journal_release_buffer(handle, iloc.bh); */ |
ac27a0ec DK |
494 | brelse(iloc.bh); |
495 | exit_dindj: | |
2b2d6d01 | 496 | /* ext4_journal_release_buffer(handle, dind); */ |
ac27a0ec | 497 | exit_primary: |
2b2d6d01 | 498 | /* ext4_journal_release_buffer(handle, *primary); */ |
ac27a0ec | 499 | exit_sbh: |
2b2d6d01 | 500 | /* ext4_journal_release_buffer(handle, *primary); */ |
ac27a0ec DK |
501 | exit_dind: |
502 | brelse(dind); | |
503 | exit_bh: | |
504 | brelse(*primary); | |
505 | ||
617ba13b | 506 | ext4_debug("leaving with error %d\n", err); |
ac27a0ec DK |
507 | return err; |
508 | } | |
509 | ||
510 | /* | |
511 | * Called when we are adding a new group which has a backup copy of each of | |
512 | * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. | |
513 | * We need to add these reserved backup GDT blocks to the resize inode, so | |
514 | * that they are kept for future resizing and not allocated to files. | |
515 | * | |
516 | * Each reserved backup GDT block will go into a different indirect block. | |
517 | * The indirect blocks are actually the primary reserved GDT blocks, | |
518 | * so we know in advance what their block numbers are. We only get the | |
519 | * double-indirect block to verify it is pointing to the primary reserved | |
520 | * GDT blocks so we don't overwrite a data block by accident. The reserved | |
521 | * backup GDT blocks are stored in their reserved primary GDT block. | |
522 | */ | |
523 | static int reserve_backup_gdb(handle_t *handle, struct inode *inode, | |
617ba13b | 524 | struct ext4_new_group_data *input) |
ac27a0ec DK |
525 | { |
526 | struct super_block *sb = inode->i_sb; | |
617ba13b | 527 | int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); |
ac27a0ec DK |
528 | struct buffer_head **primary; |
529 | struct buffer_head *dind; | |
617ba13b MC |
530 | struct ext4_iloc iloc; |
531 | ext4_fsblk_t blk; | |
ac27a0ec DK |
532 | __le32 *data, *end; |
533 | int gdbackups = 0; | |
534 | int res, i; | |
535 | int err; | |
536 | ||
216553c4 | 537 | primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS); |
ac27a0ec DK |
538 | if (!primary) |
539 | return -ENOMEM; | |
540 | ||
617ba13b | 541 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
ac27a0ec DK |
542 | dind = sb_bread(sb, le32_to_cpu(*data)); |
543 | if (!dind) { | |
544 | err = -EIO; | |
545 | goto exit_free; | |
546 | } | |
547 | ||
617ba13b | 548 | blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; |
94460093 JB |
549 | data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % |
550 | EXT4_ADDR_PER_BLOCK(sb)); | |
617ba13b | 551 | end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); |
ac27a0ec DK |
552 | |
553 | /* Get each reserved primary GDT block and verify it holds backups */ | |
554 | for (res = 0; res < reserved_gdb; res++, blk++) { | |
555 | if (le32_to_cpu(*data) != blk) { | |
12062ddd | 556 | ext4_warning(sb, "reserved block %llu" |
ac27a0ec DK |
557 | " not at offset %ld", |
558 | blk, | |
559 | (long)(data - (__le32 *)dind->b_data)); | |
560 | err = -EINVAL; | |
561 | goto exit_bh; | |
562 | } | |
563 | primary[res] = sb_bread(sb, blk); | |
564 | if (!primary[res]) { | |
565 | err = -EIO; | |
566 | goto exit_bh; | |
567 | } | |
568 | if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) { | |
569 | brelse(primary[res]); | |
570 | err = gdbackups; | |
571 | goto exit_bh; | |
572 | } | |
573 | if (++data >= end) | |
574 | data = (__le32 *)dind->b_data; | |
575 | } | |
576 | ||
577 | for (i = 0; i < reserved_gdb; i++) { | |
617ba13b | 578 | if ((err = ext4_journal_get_write_access(handle, primary[i]))) { |
ac27a0ec DK |
579 | /* |
580 | int j; | |
581 | for (j = 0; j < i; j++) | |
617ba13b | 582 | ext4_journal_release_buffer(handle, primary[j]); |
ac27a0ec DK |
583 | */ |
584 | goto exit_bh; | |
585 | } | |
586 | } | |
587 | ||
617ba13b | 588 | if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) |
ac27a0ec DK |
589 | goto exit_bh; |
590 | ||
591 | /* | |
592 | * Finally we can add each of the reserved backup GDT blocks from | |
593 | * the new group to its reserved primary GDT block. | |
594 | */ | |
617ba13b | 595 | blk = input->group * EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
596 | for (i = 0; i < reserved_gdb; i++) { |
597 | int err2; | |
598 | data = (__le32 *)primary[i]->b_data; | |
599 | /* printk("reserving backup %lu[%u] = %lu\n", | |
600 | primary[i]->b_blocknr, gdbackups, | |
601 | blk + primary[i]->b_blocknr); */ | |
602 | data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); | |
0390131b | 603 | err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); |
ac27a0ec DK |
604 | if (!err) |
605 | err = err2; | |
606 | } | |
607 | inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9; | |
617ba13b | 608 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
609 | |
610 | exit_bh: | |
611 | while (--res >= 0) | |
612 | brelse(primary[res]); | |
613 | brelse(dind); | |
614 | ||
615 | exit_free: | |
616 | kfree(primary); | |
617 | ||
618 | return err; | |
619 | } | |
620 | ||
621 | /* | |
617ba13b | 622 | * Update the backup copies of the ext4 metadata. These don't need to be part |
ac27a0ec DK |
623 | * of the main resize transaction, because e2fsck will re-write them if there |
624 | * is a problem (basically only OOM will cause a problem). However, we | |
625 | * _should_ update the backups if possible, in case the primary gets trashed | |
626 | * for some reason and we need to run e2fsck from a backup superblock. The | |
627 | * important part is that the new block and inode counts are in the backup | |
628 | * superblocks, and the location of the new group metadata in the GDT backups. | |
629 | * | |
32ed5058 TT |
630 | * We do not need take the s_resize_lock for this, because these |
631 | * blocks are not otherwise touched by the filesystem code when it is | |
632 | * mounted. We don't need to worry about last changing from | |
633 | * sbi->s_groups_count, because the worst that can happen is that we | |
634 | * do not copy the full number of backups at this time. The resize | |
635 | * which changed s_groups_count will backup again. | |
ac27a0ec DK |
636 | */ |
637 | static void update_backups(struct super_block *sb, | |
638 | int blk_off, char *data, int size) | |
639 | { | |
617ba13b | 640 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
fd2d4291 | 641 | const ext4_group_t last = sbi->s_groups_count; |
617ba13b | 642 | const int bpg = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
643 | unsigned three = 1; |
644 | unsigned five = 5; | |
645 | unsigned seven = 7; | |
fd2d4291 | 646 | ext4_group_t group; |
ac27a0ec DK |
647 | int rest = sb->s_blocksize - size; |
648 | handle_t *handle; | |
649 | int err = 0, err2; | |
650 | ||
617ba13b | 651 | handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA); |
ac27a0ec DK |
652 | if (IS_ERR(handle)) { |
653 | group = 1; | |
654 | err = PTR_ERR(handle); | |
655 | goto exit_err; | |
656 | } | |
657 | ||
617ba13b | 658 | while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) { |
ac27a0ec DK |
659 | struct buffer_head *bh; |
660 | ||
661 | /* Out of journal space, and can't get more - abort - so sad */ | |
0390131b FM |
662 | if (ext4_handle_valid(handle) && |
663 | handle->h_buffer_credits == 0 && | |
617ba13b MC |
664 | ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) && |
665 | (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) | |
ac27a0ec DK |
666 | break; |
667 | ||
668 | bh = sb_getblk(sb, group * bpg + blk_off); | |
669 | if (!bh) { | |
670 | err = -EIO; | |
671 | break; | |
672 | } | |
617ba13b | 673 | ext4_debug("update metadata backup %#04lx\n", |
ac27a0ec | 674 | (unsigned long)bh->b_blocknr); |
617ba13b | 675 | if ((err = ext4_journal_get_write_access(handle, bh))) |
ac27a0ec DK |
676 | break; |
677 | lock_buffer(bh); | |
678 | memcpy(bh->b_data, data, size); | |
679 | if (rest) | |
680 | memset(bh->b_data + size, 0, rest); | |
681 | set_buffer_uptodate(bh); | |
682 | unlock_buffer(bh); | |
0390131b | 683 | ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
684 | brelse(bh); |
685 | } | |
617ba13b | 686 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
687 | err = err2; |
688 | ||
689 | /* | |
690 | * Ugh! Need to have e2fsck write the backup copies. It is too | |
691 | * late to revert the resize, we shouldn't fail just because of | |
692 | * the backup copies (they are only needed in case of corruption). | |
693 | * | |
694 | * However, if we got here we have a journal problem too, so we | |
695 | * can't really start a transaction to mark the superblock. | |
696 | * Chicken out and just set the flag on the hope it will be written | |
697 | * to disk, and if not - we will simply wait until next fsck. | |
698 | */ | |
699 | exit_err: | |
700 | if (err) { | |
12062ddd | 701 | ext4_warning(sb, "can't update backup for group %u (err %d), " |
ac27a0ec | 702 | "forcing fsck on next reboot", group, err); |
617ba13b MC |
703 | sbi->s_mount_state &= ~EXT4_VALID_FS; |
704 | sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); | |
ac27a0ec DK |
705 | mark_buffer_dirty(sbi->s_sbh); |
706 | } | |
707 | } | |
708 | ||
709 | /* Add group descriptor data to an existing or new group descriptor block. | |
710 | * Ensure we handle all possible error conditions _before_ we start modifying | |
711 | * the filesystem, because we cannot abort the transaction and not have it | |
712 | * write the data to disk. | |
713 | * | |
714 | * If we are on a GDT block boundary, we need to get the reserved GDT block. | |
715 | * Otherwise, we may need to add backup GDT blocks for a sparse group. | |
716 | * | |
717 | * We only need to hold the superblock lock while we are actually adding | |
718 | * in the new group's counts to the superblock. Prior to that we have | |
719 | * not really "added" the group at all. We re-check that we are still | |
720 | * adding in the last group in case things have changed since verifying. | |
721 | */ | |
617ba13b | 722 | int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) |
ac27a0ec | 723 | { |
617ba13b MC |
724 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
725 | struct ext4_super_block *es = sbi->s_es; | |
726 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec DK |
727 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; |
728 | struct buffer_head *primary = NULL; | |
617ba13b | 729 | struct ext4_group_desc *gdp; |
ac27a0ec DK |
730 | struct inode *inode = NULL; |
731 | handle_t *handle; | |
732 | int gdb_off, gdb_num; | |
733 | int err, err2; | |
734 | ||
617ba13b MC |
735 | gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb); |
736 | gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 737 | |
617ba13b MC |
738 | if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb, |
739 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
12062ddd | 740 | ext4_warning(sb, "Can't resize non-sparse filesystem further"); |
ac27a0ec DK |
741 | return -EPERM; |
742 | } | |
743 | ||
bd81d8ee LV |
744 | if (ext4_blocks_count(es) + input->blocks_count < |
745 | ext4_blocks_count(es)) { | |
12062ddd | 746 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
747 | return -EINVAL; |
748 | } | |
749 | ||
617ba13b | 750 | if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < |
ac27a0ec | 751 | le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 752 | ext4_warning(sb, "inodes_count overflow"); |
ac27a0ec DK |
753 | return -EINVAL; |
754 | } | |
755 | ||
756 | if (reserved_gdb || gdb_off == 0) { | |
617ba13b | 757 | if (!EXT4_HAS_COMPAT_FEATURE(sb, |
37609fd5 JB |
758 | EXT4_FEATURE_COMPAT_RESIZE_INODE) |
759 | || !le16_to_cpu(es->s_reserved_gdt_blocks)) { | |
12062ddd | 760 | ext4_warning(sb, |
ac27a0ec DK |
761 | "No reserved GDT blocks, can't resize"); |
762 | return -EPERM; | |
763 | } | |
1d1fe1ee DH |
764 | inode = ext4_iget(sb, EXT4_RESIZE_INO); |
765 | if (IS_ERR(inode)) { | |
12062ddd | 766 | ext4_warning(sb, "Error opening resize inode"); |
1d1fe1ee | 767 | return PTR_ERR(inode); |
ac27a0ec DK |
768 | } |
769 | } | |
770 | ||
920313a7 | 771 | |
ac27a0ec DK |
772 | if ((err = verify_group_input(sb, input))) |
773 | goto exit_put; | |
774 | ||
775 | if ((err = setup_new_group_blocks(sb, input))) | |
776 | goto exit_put; | |
777 | ||
778 | /* | |
779 | * We will always be modifying at least the superblock and a GDT | |
780 | * block. If we are adding a group past the last current GDT block, | |
781 | * we will also modify the inode and the dindirect block. If we | |
782 | * are adding a group with superblock/GDT backups we will also | |
783 | * modify each of the reserved GDT dindirect blocks. | |
784 | */ | |
617ba13b MC |
785 | handle = ext4_journal_start_sb(sb, |
786 | ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec DK |
787 | 3 + reserved_gdb : 4); |
788 | if (IS_ERR(handle)) { | |
789 | err = PTR_ERR(handle); | |
790 | goto exit_put; | |
791 | } | |
792 | ||
32ed5058 | 793 | mutex_lock(&sbi->s_resize_lock); |
ac27a0ec | 794 | if (input->group != sbi->s_groups_count) { |
12062ddd | 795 | ext4_warning(sb, "multiple resizers run on filesystem!"); |
ac27a0ec DK |
796 | err = -EBUSY; |
797 | goto exit_journal; | |
798 | } | |
799 | ||
617ba13b | 800 | if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh))) |
ac27a0ec DK |
801 | goto exit_journal; |
802 | ||
2b2d6d01 TT |
803 | /* |
804 | * We will only either add reserved group blocks to a backup group | |
805 | * or remove reserved blocks for the first group in a new group block. | |
806 | * Doing both would be mean more complex code, and sane people don't | |
807 | * use non-sparse filesystems anymore. This is already checked above. | |
808 | */ | |
ac27a0ec DK |
809 | if (gdb_off) { |
810 | primary = sbi->s_group_desc[gdb_num]; | |
617ba13b | 811 | if ((err = ext4_journal_get_write_access(handle, primary))) |
ac27a0ec DK |
812 | goto exit_journal; |
813 | ||
617ba13b | 814 | if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) && |
ac27a0ec DK |
815 | (err = reserve_backup_gdb(handle, inode, input))) |
816 | goto exit_journal; | |
817 | } else if ((err = add_new_gdb(handle, inode, input, &primary))) | |
818 | goto exit_journal; | |
819 | ||
2b2d6d01 TT |
820 | /* |
821 | * OK, now we've set up the new group. Time to make it active. | |
822 | * | |
32ed5058 | 823 | * We do not lock all allocations via s_resize_lock |
2b2d6d01 TT |
824 | * so we have to be safe wrt. concurrent accesses the group |
825 | * data. So we need to be careful to set all of the relevant | |
826 | * group descriptor data etc. *before* we enable the group. | |
827 | * | |
828 | * The key field here is sbi->s_groups_count: as long as | |
829 | * that retains its old value, nobody is going to access the new | |
830 | * group. | |
831 | * | |
832 | * So first we update all the descriptor metadata for the new | |
833 | * group; then we update the total disk blocks count; then we | |
834 | * update the groups count to enable the group; then finally we | |
835 | * update the free space counts so that the system can start | |
836 | * using the new disk blocks. | |
837 | */ | |
ac27a0ec DK |
838 | |
839 | /* Update group descriptor block for new group */ | |
2856922c FB |
840 | gdp = (struct ext4_group_desc *)((char *)primary->b_data + |
841 | gdb_off * EXT4_DESC_SIZE(sb)); | |
ac27a0ec | 842 | |
fdff73f0 | 843 | memset(gdp, 0, EXT4_DESC_SIZE(sb)); |
8fadc143 AR |
844 | ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */ |
845 | ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */ | |
846 | ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */ | |
560671a0 AK |
847 | ext4_free_blks_set(sb, gdp, input->free_blocks_count); |
848 | ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); | |
fdff73f0 | 849 | gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED); |
717d50e4 | 850 | gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp); |
ac27a0ec | 851 | |
5f21b0e6 FB |
852 | /* |
853 | * We can allocate memory for mb_alloc based on the new group | |
854 | * descriptor | |
855 | */ | |
920313a7 | 856 | err = ext4_mb_add_groupinfo(sb, input->group, gdp); |
08c3a813 | 857 | if (err) |
c2ea3fde TT |
858 | goto exit_journal; |
859 | ||
ac27a0ec DK |
860 | /* |
861 | * Make the new blocks and inodes valid next. We do this before | |
862 | * increasing the group count so that once the group is enabled, | |
863 | * all of its blocks and inodes are already valid. | |
864 | * | |
865 | * We always allocate group-by-group, then block-by-block or | |
866 | * inode-by-inode within a group, so enabling these | |
867 | * blocks/inodes before the group is live won't actually let us | |
868 | * allocate the new space yet. | |
869 | */ | |
bd81d8ee | 870 | ext4_blocks_count_set(es, ext4_blocks_count(es) + |
ac27a0ec | 871 | input->blocks_count); |
e8546d06 | 872 | le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec DK |
873 | |
874 | /* | |
875 | * We need to protect s_groups_count against other CPUs seeing | |
876 | * inconsistent state in the superblock. | |
877 | * | |
878 | * The precise rules we use are: | |
879 | * | |
32ed5058 | 880 | * * Writers of s_groups_count *must* hold s_resize_lock |
ac27a0ec DK |
881 | * AND |
882 | * * Writers must perform a smp_wmb() after updating all dependent | |
883 | * data and before modifying the groups count | |
884 | * | |
32ed5058 | 885 | * * Readers must hold s_resize_lock over the access |
ac27a0ec DK |
886 | * OR |
887 | * * Readers must perform an smp_rmb() after reading the groups count | |
888 | * and before reading any dependent data. | |
889 | * | |
890 | * NB. These rules can be relaxed when checking the group count | |
891 | * while freeing data, as we can only allocate from a block | |
892 | * group after serialising against the group count, and we can | |
893 | * only then free after serialising in turn against that | |
894 | * allocation. | |
895 | */ | |
896 | smp_wmb(); | |
897 | ||
898 | /* Update the global fs size fields */ | |
899 | sbi->s_groups_count++; | |
900 | ||
0390131b | 901 | ext4_handle_dirty_metadata(handle, NULL, primary); |
ac27a0ec DK |
902 | |
903 | /* Update the reserved block counts only once the new group is | |
904 | * active. */ | |
bd81d8ee | 905 | ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + |
ac27a0ec DK |
906 | input->reserved_blocks); |
907 | ||
908 | /* Update the free space counts */ | |
aa0dff2d | 909 | percpu_counter_add(&sbi->s_freeblocks_counter, |
ac27a0ec | 910 | input->free_blocks_count); |
aa0dff2d | 911 | percpu_counter_add(&sbi->s_freeinodes_counter, |
617ba13b | 912 | EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec | 913 | |
c62a11fd FB |
914 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { |
915 | ext4_group_t flex_group; | |
916 | flex_group = ext4_flex_group(sbi, input->group); | |
9f24e420 TT |
917 | atomic_add(input->free_blocks_count, |
918 | &sbi->s_flex_groups[flex_group].free_blocks); | |
919 | atomic_add(EXT4_INODES_PER_GROUP(sb), | |
920 | &sbi->s_flex_groups[flex_group].free_inodes); | |
c62a11fd FB |
921 | } |
922 | ||
0390131b | 923 | ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); |
ac27a0ec DK |
924 | sb->s_dirt = 1; |
925 | ||
926 | exit_journal: | |
32ed5058 | 927 | mutex_unlock(&sbi->s_resize_lock); |
617ba13b | 928 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
929 | err = err2; |
930 | if (!err) { | |
931 | update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, | |
617ba13b | 932 | sizeof(struct ext4_super_block)); |
ac27a0ec DK |
933 | update_backups(sb, primary->b_blocknr, primary->b_data, |
934 | primary->b_size); | |
935 | } | |
936 | exit_put: | |
937 | iput(inode); | |
938 | return err; | |
617ba13b | 939 | } /* ext4_group_add */ |
ac27a0ec | 940 | |
2b2d6d01 TT |
941 | /* |
942 | * Extend the filesystem to the new number of blocks specified. This entry | |
ac27a0ec DK |
943 | * point is only used to extend the current filesystem to the end of the last |
944 | * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" | |
945 | * for emergencies (because it has no dependencies on reserved blocks). | |
946 | * | |
617ba13b | 947 | * If we _really_ wanted, we could use default values to call ext4_group_add() |
ac27a0ec DK |
948 | * allow the "remount" trick to work for arbitrary resizing, assuming enough |
949 | * GDT blocks are reserved to grow to the desired size. | |
950 | */ | |
617ba13b MC |
951 | int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, |
952 | ext4_fsblk_t n_blocks_count) | |
ac27a0ec | 953 | { |
617ba13b | 954 | ext4_fsblk_t o_blocks_count; |
fd2d4291 | 955 | ext4_group_t o_groups_count; |
617ba13b MC |
956 | ext4_grpblk_t last; |
957 | ext4_grpblk_t add; | |
af5bc92d | 958 | struct buffer_head *bh; |
ac27a0ec DK |
959 | handle_t *handle; |
960 | int err; | |
5f21b0e6 | 961 | ext4_group_t group; |
ac27a0ec DK |
962 | |
963 | /* We don't need to worry about locking wrt other resizers just | |
964 | * yet: we're going to revalidate es->s_blocks_count after | |
32ed5058 | 965 | * taking the s_resize_lock below. */ |
bd81d8ee | 966 | o_blocks_count = ext4_blocks_count(es); |
617ba13b | 967 | o_groups_count = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
968 | |
969 | if (test_opt(sb, DEBUG)) | |
2ae02107 | 970 | printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n", |
ac27a0ec DK |
971 | o_blocks_count, n_blocks_count); |
972 | ||
973 | if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) | |
974 | return 0; | |
975 | ||
976 | if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { | |
617ba13b | 977 | printk(KERN_ERR "EXT4-fs: filesystem on %s:" |
2ae02107 | 978 | " too large to resize to %llu blocks safely\n", |
ac27a0ec DK |
979 | sb->s_id, n_blocks_count); |
980 | if (sizeof(sector_t) < 8) | |
12062ddd | 981 | ext4_warning(sb, "CONFIG_LBDAF not enabled"); |
ac27a0ec DK |
982 | return -EINVAL; |
983 | } | |
984 | ||
985 | if (n_blocks_count < o_blocks_count) { | |
12062ddd | 986 | ext4_warning(sb, "can't shrink FS - resize aborted"); |
ac27a0ec DK |
987 | return -EBUSY; |
988 | } | |
989 | ||
990 | /* Handle the remaining blocks in the last group only. */ | |
5f21b0e6 | 991 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); |
ac27a0ec DK |
992 | |
993 | if (last == 0) { | |
12062ddd | 994 | ext4_warning(sb, "need to use ext2online to resize further"); |
ac27a0ec DK |
995 | return -EPERM; |
996 | } | |
997 | ||
617ba13b | 998 | add = EXT4_BLOCKS_PER_GROUP(sb) - last; |
ac27a0ec DK |
999 | |
1000 | if (o_blocks_count + add < o_blocks_count) { | |
12062ddd | 1001 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1002 | return -EINVAL; |
1003 | } | |
1004 | ||
1005 | if (o_blocks_count + add > n_blocks_count) | |
1006 | add = n_blocks_count - o_blocks_count; | |
1007 | ||
1008 | if (o_blocks_count + add < n_blocks_count) | |
12062ddd | 1009 | ext4_warning(sb, "will only finish group (%llu blocks, %u new)", |
ac27a0ec DK |
1010 | o_blocks_count + add, add); |
1011 | ||
1012 | /* See if the device is actually as big as what was requested */ | |
2b2d6d01 | 1013 | bh = sb_bread(sb, o_blocks_count + add - 1); |
ac27a0ec | 1014 | if (!bh) { |
12062ddd | 1015 | ext4_warning(sb, "can't read last block, resize aborted"); |
ac27a0ec DK |
1016 | return -ENOSPC; |
1017 | } | |
1018 | brelse(bh); | |
1019 | ||
1020 | /* We will update the superblock, one block bitmap, and | |
617ba13b | 1021 | * one group descriptor via ext4_free_blocks(). |
ac27a0ec | 1022 | */ |
617ba13b | 1023 | handle = ext4_journal_start_sb(sb, 3); |
ac27a0ec DK |
1024 | if (IS_ERR(handle)) { |
1025 | err = PTR_ERR(handle); | |
12062ddd | 1026 | ext4_warning(sb, "error %d on journal start", err); |
ac27a0ec DK |
1027 | goto exit_put; |
1028 | } | |
1029 | ||
32ed5058 | 1030 | mutex_lock(&EXT4_SB(sb)->s_resize_lock); |
bd81d8ee | 1031 | if (o_blocks_count != ext4_blocks_count(es)) { |
12062ddd | 1032 | ext4_warning(sb, "multiple resizers run on filesystem!"); |
32ed5058 | 1033 | mutex_unlock(&EXT4_SB(sb)->s_resize_lock); |
5606bf5d | 1034 | ext4_journal_stop(handle); |
ac27a0ec DK |
1035 | err = -EBUSY; |
1036 | goto exit_put; | |
1037 | } | |
1038 | ||
617ba13b MC |
1039 | if ((err = ext4_journal_get_write_access(handle, |
1040 | EXT4_SB(sb)->s_sbh))) { | |
12062ddd | 1041 | ext4_warning(sb, "error %d on journal write access", err); |
32ed5058 | 1042 | mutex_unlock(&EXT4_SB(sb)->s_resize_lock); |
617ba13b | 1043 | ext4_journal_stop(handle); |
ac27a0ec DK |
1044 | goto exit_put; |
1045 | } | |
bd81d8ee | 1046 | ext4_blocks_count_set(es, o_blocks_count + add); |
0390131b | 1047 | ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
ac27a0ec | 1048 | sb->s_dirt = 1; |
32ed5058 | 1049 | mutex_unlock(&EXT4_SB(sb)->s_resize_lock); |
c549a95d | 1050 | ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, |
ac27a0ec | 1051 | o_blocks_count + add); |
e21675d4 AK |
1052 | /* We add the blocks to the bitmap and set the group need init bit */ |
1053 | ext4_add_groupblocks(handle, sb, o_blocks_count, add); | |
2ae02107 | 1054 | ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, |
ac27a0ec | 1055 | o_blocks_count + add); |
617ba13b | 1056 | if ((err = ext4_journal_stop(handle))) |
ac27a0ec | 1057 | goto exit_put; |
5f21b0e6 | 1058 | |
ac27a0ec | 1059 | if (test_opt(sb, DEBUG)) |
bd81d8ee LV |
1060 | printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n", |
1061 | ext4_blocks_count(es)); | |
617ba13b MC |
1062 | update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es, |
1063 | sizeof(struct ext4_super_block)); | |
ac27a0ec DK |
1064 | exit_put: |
1065 | return err; | |
617ba13b | 1066 | } /* ext4_group_extend */ |