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 | 18 | |
8f82f840 YY |
19 | int ext4_resize_begin(struct super_block *sb) |
20 | { | |
21 | int ret = 0; | |
22 | ||
23 | if (!capable(CAP_SYS_RESOURCE)) | |
24 | return -EPERM; | |
25 | ||
ce723c31 YY |
26 | /* |
27 | * We are not allowed to do online-resizing on a filesystem mounted | |
28 | * with error, because it can destroy the filesystem easily. | |
29 | */ | |
30 | if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) { | |
31 | ext4_warning(sb, "There are errors in the filesystem, " | |
32 | "so online resizing is not allowed\n"); | |
33 | return -EPERM; | |
34 | } | |
35 | ||
8f82f840 YY |
36 | if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags)) |
37 | ret = -EBUSY; | |
38 | ||
39 | return ret; | |
40 | } | |
41 | ||
42 | void ext4_resize_end(struct super_block *sb) | |
43 | { | |
44 | clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags); | |
45 | smp_mb__after_clear_bit(); | |
46 | } | |
47 | ||
ac27a0ec DK |
48 | #define outside(b, first, last) ((b) < (first) || (b) >= (last)) |
49 | #define inside(b, first, last) ((b) >= (first) && (b) < (last)) | |
50 | ||
51 | static int verify_group_input(struct super_block *sb, | |
617ba13b | 52 | struct ext4_new_group_data *input) |
ac27a0ec | 53 | { |
617ba13b MC |
54 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
55 | struct ext4_super_block *es = sbi->s_es; | |
bd81d8ee | 56 | ext4_fsblk_t start = ext4_blocks_count(es); |
617ba13b | 57 | ext4_fsblk_t end = start + input->blocks_count; |
fd2d4291 | 58 | ext4_group_t group = input->group; |
617ba13b MC |
59 | ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group; |
60 | unsigned overhead = ext4_bg_has_super(sb, group) ? | |
61 | (1 + ext4_bg_num_gdb(sb, group) + | |
ac27a0ec | 62 | le16_to_cpu(es->s_reserved_gdt_blocks)) : 0; |
617ba13b | 63 | ext4_fsblk_t metaend = start + overhead; |
ac27a0ec | 64 | struct buffer_head *bh = NULL; |
3a5b2ecd | 65 | ext4_grpblk_t free_blocks_count, offset; |
ac27a0ec DK |
66 | int err = -EINVAL; |
67 | ||
68 | input->free_blocks_count = free_blocks_count = | |
69 | input->blocks_count - 2 - overhead - sbi->s_itb_per_group; | |
70 | ||
71 | if (test_opt(sb, DEBUG)) | |
617ba13b | 72 | printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks " |
ac27a0ec | 73 | "(%d free, %u reserved)\n", |
617ba13b | 74 | ext4_bg_has_super(sb, input->group) ? "normal" : |
ac27a0ec DK |
75 | "no-super", input->group, input->blocks_count, |
76 | free_blocks_count, input->reserved_blocks); | |
77 | ||
3a5b2ecd | 78 | ext4_get_group_no_and_offset(sb, start, NULL, &offset); |
ac27a0ec | 79 | if (group != sbi->s_groups_count) |
12062ddd | 80 | ext4_warning(sb, "Cannot add at group %u (only %u groups)", |
ac27a0ec | 81 | input->group, sbi->s_groups_count); |
3a5b2ecd | 82 | else if (offset != 0) |
12062ddd | 83 | ext4_warning(sb, "Last group not full"); |
ac27a0ec | 84 | else if (input->reserved_blocks > input->blocks_count / 5) |
12062ddd | 85 | ext4_warning(sb, "Reserved blocks too high (%u)", |
ac27a0ec DK |
86 | input->reserved_blocks); |
87 | else if (free_blocks_count < 0) | |
12062ddd | 88 | ext4_warning(sb, "Bad blocks count %u", |
ac27a0ec DK |
89 | input->blocks_count); |
90 | else if (!(bh = sb_bread(sb, end - 1))) | |
12062ddd | 91 | ext4_warning(sb, "Cannot read last block (%llu)", |
ac27a0ec DK |
92 | end - 1); |
93 | else if (outside(input->block_bitmap, start, end)) | |
12062ddd | 94 | ext4_warning(sb, "Block bitmap not in group (block %llu)", |
1939e49a | 95 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 96 | else if (outside(input->inode_bitmap, start, end)) |
12062ddd | 97 | ext4_warning(sb, "Inode bitmap not in group (block %llu)", |
1939e49a | 98 | (unsigned long long)input->inode_bitmap); |
ac27a0ec | 99 | else if (outside(input->inode_table, start, end) || |
2b2d6d01 | 100 | outside(itend - 1, start, end)) |
12062ddd | 101 | ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)", |
1939e49a | 102 | (unsigned long long)input->inode_table, itend - 1); |
ac27a0ec | 103 | else if (input->inode_bitmap == input->block_bitmap) |
12062ddd | 104 | ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)", |
1939e49a | 105 | (unsigned long long)input->block_bitmap); |
ac27a0ec | 106 | else if (inside(input->block_bitmap, input->inode_table, itend)) |
12062ddd ES |
107 | ext4_warning(sb, "Block bitmap (%llu) in inode table " |
108 | "(%llu-%llu)", | |
1939e49a RD |
109 | (unsigned long long)input->block_bitmap, |
110 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 111 | else if (inside(input->inode_bitmap, input->inode_table, itend)) |
12062ddd ES |
112 | ext4_warning(sb, "Inode bitmap (%llu) in inode table " |
113 | "(%llu-%llu)", | |
1939e49a RD |
114 | (unsigned long long)input->inode_bitmap, |
115 | (unsigned long long)input->inode_table, itend - 1); | |
ac27a0ec | 116 | else if (inside(input->block_bitmap, start, metaend)) |
12062ddd | 117 | ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
118 | (unsigned long long)input->block_bitmap, |
119 | start, metaend - 1); | |
ac27a0ec | 120 | else if (inside(input->inode_bitmap, start, metaend)) |
12062ddd | 121 | ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)", |
1939e49a RD |
122 | (unsigned long long)input->inode_bitmap, |
123 | start, metaend - 1); | |
ac27a0ec | 124 | else if (inside(input->inode_table, start, metaend) || |
2b2d6d01 | 125 | inside(itend - 1, start, metaend)) |
12062ddd ES |
126 | ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table " |
127 | "(%llu-%llu)", | |
1939e49a RD |
128 | (unsigned long long)input->inode_table, |
129 | itend - 1, start, metaend - 1); | |
ac27a0ec DK |
130 | else |
131 | err = 0; | |
132 | brelse(bh); | |
133 | ||
134 | return err; | |
135 | } | |
136 | ||
137 | static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, | |
617ba13b | 138 | ext4_fsblk_t blk) |
ac27a0ec DK |
139 | { |
140 | struct buffer_head *bh; | |
141 | int err; | |
142 | ||
143 | bh = sb_getblk(sb, blk); | |
144 | if (!bh) | |
145 | return ERR_PTR(-EIO); | |
617ba13b | 146 | if ((err = ext4_journal_get_write_access(handle, bh))) { |
ac27a0ec DK |
147 | brelse(bh); |
148 | bh = ERR_PTR(err); | |
149 | } else { | |
150 | lock_buffer(bh); | |
151 | memset(bh->b_data, 0, sb->s_blocksize); | |
152 | set_buffer_uptodate(bh); | |
153 | unlock_buffer(bh); | |
154 | } | |
155 | ||
156 | return bh; | |
157 | } | |
158 | ||
14904107 ES |
159 | /* |
160 | * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA. | |
161 | * If that fails, restart the transaction & regain write access for the | |
162 | * buffer head which is used for block_bitmap modifications. | |
163 | */ | |
164 | static int extend_or_restart_transaction(handle_t *handle, int thresh, | |
165 | struct buffer_head *bh) | |
166 | { | |
167 | int err; | |
168 | ||
0390131b | 169 | if (ext4_handle_has_enough_credits(handle, thresh)) |
14904107 ES |
170 | return 0; |
171 | ||
172 | err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA); | |
173 | if (err < 0) | |
174 | return err; | |
175 | if (err) { | |
176 | if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) | |
177 | return err; | |
2b2d6d01 | 178 | if ((err = ext4_journal_get_write_access(handle, bh))) |
14904107 | 179 | return err; |
2b2d6d01 | 180 | } |
14904107 ES |
181 | |
182 | return 0; | |
183 | } | |
184 | ||
ac27a0ec DK |
185 | /* |
186 | * Set up the block and inode bitmaps, and the inode table for the new group. | |
187 | * This doesn't need to be part of the main transaction, since we are only | |
188 | * changing blocks outside the actual filesystem. We still do journaling to | |
189 | * ensure the recovery is correct in case of a failure just after resize. | |
190 | * If any part of this fails, we simply abort the resize. | |
191 | */ | |
192 | static int setup_new_group_blocks(struct super_block *sb, | |
617ba13b | 193 | struct ext4_new_group_data *input) |
ac27a0ec | 194 | { |
617ba13b MC |
195 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
196 | ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group); | |
197 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec | 198 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0; |
617ba13b | 199 | unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group); |
ac27a0ec DK |
200 | struct buffer_head *bh; |
201 | handle_t *handle; | |
617ba13b MC |
202 | ext4_fsblk_t block; |
203 | ext4_grpblk_t bit; | |
ac27a0ec DK |
204 | int i; |
205 | int err = 0, err2; | |
206 | ||
14904107 ES |
207 | /* This transaction may be extended/restarted along the way */ |
208 | handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA); | |
209 | ||
ac27a0ec DK |
210 | if (IS_ERR(handle)) |
211 | return PTR_ERR(handle); | |
212 | ||
8f82f840 | 213 | BUG_ON(input->group != sbi->s_groups_count); |
ac27a0ec DK |
214 | |
215 | if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) { | |
216 | err = PTR_ERR(bh); | |
217 | goto exit_journal; | |
218 | } | |
219 | ||
ac27a0ec DK |
220 | /* Copy all of the GDT blocks into the backup in this group */ |
221 | for (i = 0, bit = 1, block = start + 1; | |
222 | i < gdblocks; i++, block++, bit++) { | |
223 | struct buffer_head *gdb; | |
224 | ||
c549a95d | 225 | ext4_debug("update backup group %#04llx (+%d)\n", block, bit); |
ac27a0ec | 226 | |
14904107 ES |
227 | if ((err = extend_or_restart_transaction(handle, 1, bh))) |
228 | goto exit_bh; | |
229 | ||
ac27a0ec DK |
230 | gdb = sb_getblk(sb, block); |
231 | if (!gdb) { | |
232 | err = -EIO; | |
233 | goto exit_bh; | |
234 | } | |
617ba13b | 235 | if ((err = ext4_journal_get_write_access(handle, gdb))) { |
ac27a0ec DK |
236 | brelse(gdb); |
237 | goto exit_bh; | |
238 | } | |
5b615287 ES |
239 | lock_buffer(gdb); |
240 | memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size); | |
ac27a0ec | 241 | set_buffer_uptodate(gdb); |
5b615287 | 242 | unlock_buffer(gdb); |
b4097142 TT |
243 | err = ext4_handle_dirty_metadata(handle, NULL, gdb); |
244 | if (unlikely(err)) { | |
245 | brelse(gdb); | |
246 | goto exit_bh; | |
247 | } | |
ac27a0ec DK |
248 | brelse(gdb); |
249 | } | |
250 | ||
251 | /* Zero out all of the reserved backup group descriptor table blocks */ | |
da488945 | 252 | ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", |
a31437b8 LC |
253 | block, sbi->s_itb_per_group); |
254 | err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb, | |
a107e5a3 | 255 | GFP_NOFS); |
a31437b8 LC |
256 | if (err) |
257 | goto exit_bh; | |
c3e94d1d YY |
258 | |
259 | if (ext4_bg_has_super(sb, input->group)) { | |
260 | ext4_debug("mark backup group tables %#04llx (+0)\n", start); | |
261 | ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1); | |
262 | } | |
14904107 | 263 | |
c549a95d | 264 | ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap, |
ac27a0ec | 265 | input->block_bitmap - start); |
617ba13b | 266 | ext4_set_bit(input->block_bitmap - start, bh->b_data); |
c549a95d | 267 | ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap, |
ac27a0ec | 268 | input->inode_bitmap - start); |
617ba13b | 269 | ext4_set_bit(input->inode_bitmap - start, bh->b_data); |
ac27a0ec DK |
270 | |
271 | /* Zero out all of the inode table blocks */ | |
a31437b8 | 272 | block = input->inode_table; |
da488945 | 273 | ext4_debug("clear inode table blocks %#04llx -> %#04lx\n", |
a31437b8 | 274 | block, sbi->s_itb_per_group); |
a107e5a3 | 275 | err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS); |
a31437b8 LC |
276 | if (err) |
277 | goto exit_bh; | |
c3e94d1d YY |
278 | ext4_set_bits(bh->b_data, input->inode_table - start, |
279 | sbi->s_itb_per_group); | |
14904107 ES |
280 | |
281 | if ((err = extend_or_restart_transaction(handle, 2, bh))) | |
282 | goto exit_bh; | |
283 | ||
61d08673 TT |
284 | ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, |
285 | bh->b_data); | |
b4097142 TT |
286 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
287 | if (unlikely(err)) { | |
288 | ext4_std_error(sb, err); | |
289 | goto exit_bh; | |
290 | } | |
ac27a0ec | 291 | brelse(bh); |
ac27a0ec | 292 | /* Mark unused entries in inode bitmap used */ |
c549a95d | 293 | ext4_debug("clear inode bitmap %#04llx (+%llu)\n", |
ac27a0ec DK |
294 | input->inode_bitmap, input->inode_bitmap - start); |
295 | if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) { | |
296 | err = PTR_ERR(bh); | |
297 | goto exit_journal; | |
298 | } | |
299 | ||
61d08673 TT |
300 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
301 | bh->b_data); | |
b4097142 TT |
302 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
303 | if (unlikely(err)) | |
304 | ext4_std_error(sb, err); | |
ac27a0ec DK |
305 | exit_bh: |
306 | brelse(bh); | |
307 | ||
308 | exit_journal: | |
617ba13b | 309 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
310 | err = err2; |
311 | ||
312 | return err; | |
313 | } | |
314 | ||
315 | /* | |
316 | * Iterate through the groups which hold BACKUP superblock/GDT copies in an | |
617ba13b | 317 | * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before |
ac27a0ec DK |
318 | * calling this for the first time. In a sparse filesystem it will be the |
319 | * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... | |
320 | * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... | |
321 | */ | |
617ba13b | 322 | static unsigned ext4_list_backups(struct super_block *sb, unsigned *three, |
ac27a0ec DK |
323 | unsigned *five, unsigned *seven) |
324 | { | |
325 | unsigned *min = three; | |
326 | int mult = 3; | |
327 | unsigned ret; | |
328 | ||
617ba13b MC |
329 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
330 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
ac27a0ec DK |
331 | ret = *min; |
332 | *min += 1; | |
333 | return ret; | |
334 | } | |
335 | ||
336 | if (*five < *min) { | |
337 | min = five; | |
338 | mult = 5; | |
339 | } | |
340 | if (*seven < *min) { | |
341 | min = seven; | |
342 | mult = 7; | |
343 | } | |
344 | ||
345 | ret = *min; | |
346 | *min *= mult; | |
347 | ||
348 | return ret; | |
349 | } | |
350 | ||
351 | /* | |
352 | * Check that all of the backup GDT blocks are held in the primary GDT block. | |
353 | * It is assumed that they are stored in group order. Returns the number of | |
354 | * groups in current filesystem that have BACKUPS, or -ve error code. | |
355 | */ | |
356 | static int verify_reserved_gdb(struct super_block *sb, | |
357 | struct buffer_head *primary) | |
358 | { | |
617ba13b | 359 | const ext4_fsblk_t blk = primary->b_blocknr; |
fd2d4291 | 360 | const ext4_group_t end = EXT4_SB(sb)->s_groups_count; |
ac27a0ec DK |
361 | unsigned three = 1; |
362 | unsigned five = 5; | |
363 | unsigned seven = 7; | |
364 | unsigned grp; | |
365 | __le32 *p = (__le32 *)primary->b_data; | |
366 | int gdbackups = 0; | |
367 | ||
617ba13b | 368 | while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) { |
bd81d8ee LV |
369 | if (le32_to_cpu(*p++) != |
370 | grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){ | |
12062ddd | 371 | ext4_warning(sb, "reserved GDT %llu" |
2ae02107 | 372 | " missing grp %d (%llu)", |
ac27a0ec | 373 | blk, grp, |
bd81d8ee LV |
374 | grp * |
375 | (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) + | |
376 | blk); | |
ac27a0ec DK |
377 | return -EINVAL; |
378 | } | |
617ba13b | 379 | if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb)) |
ac27a0ec DK |
380 | return -EFBIG; |
381 | } | |
382 | ||
383 | return gdbackups; | |
384 | } | |
385 | ||
386 | /* | |
387 | * Called when we need to bring a reserved group descriptor table block into | |
388 | * use from the resize inode. The primary copy of the new GDT block currently | |
389 | * is an indirect block (under the double indirect block in the resize inode). | |
390 | * The new backup GDT blocks will be stored as leaf blocks in this indirect | |
391 | * block, in group order. Even though we know all the block numbers we need, | |
392 | * we check to ensure that the resize inode has actually reserved these blocks. | |
393 | * | |
394 | * Don't need to update the block bitmaps because the blocks are still in use. | |
395 | * | |
396 | * We get all of the error cases out of the way, so that we are sure to not | |
397 | * fail once we start modifying the data on disk, because JBD has no rollback. | |
398 | */ | |
399 | static int add_new_gdb(handle_t *handle, struct inode *inode, | |
617ba13b | 400 | struct ext4_new_group_data *input, |
ac27a0ec DK |
401 | struct buffer_head **primary) |
402 | { | |
403 | struct super_block *sb = inode->i_sb; | |
617ba13b MC |
404 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
405 | unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb); | |
406 | ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; | |
ac27a0ec DK |
407 | struct buffer_head **o_group_desc, **n_group_desc; |
408 | struct buffer_head *dind; | |
409 | int gdbackups; | |
617ba13b | 410 | struct ext4_iloc iloc; |
ac27a0ec DK |
411 | __le32 *data; |
412 | int err; | |
413 | ||
414 | if (test_opt(sb, DEBUG)) | |
415 | printk(KERN_DEBUG | |
617ba13b | 416 | "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n", |
ac27a0ec DK |
417 | gdb_num); |
418 | ||
af5bc92d TT |
419 | /* |
420 | * If we are not using the primary superblock/GDT copy don't resize, | |
2b2d6d01 TT |
421 | * because the user tools have no way of handling this. Probably a |
422 | * bad time to do it anyways. | |
423 | */ | |
617ba13b MC |
424 | if (EXT4_SB(sb)->s_sbh->b_blocknr != |
425 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) { | |
12062ddd | 426 | ext4_warning(sb, "won't resize using backup superblock at %llu", |
617ba13b | 427 | (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr); |
ac27a0ec DK |
428 | return -EPERM; |
429 | } | |
430 | ||
431 | *primary = sb_bread(sb, gdblock); | |
432 | if (!*primary) | |
433 | return -EIO; | |
434 | ||
435 | if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) { | |
436 | err = gdbackups; | |
437 | goto exit_bh; | |
438 | } | |
439 | ||
617ba13b | 440 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
ac27a0ec DK |
441 | dind = sb_bread(sb, le32_to_cpu(*data)); |
442 | if (!dind) { | |
443 | err = -EIO; | |
444 | goto exit_bh; | |
445 | } | |
446 | ||
447 | data = (__le32 *)dind->b_data; | |
617ba13b | 448 | if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) { |
12062ddd | 449 | ext4_warning(sb, "new group %u GDT block %llu not reserved", |
ac27a0ec DK |
450 | input->group, gdblock); |
451 | err = -EINVAL; | |
452 | goto exit_dind; | |
453 | } | |
454 | ||
b4097142 TT |
455 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
456 | if (unlikely(err)) | |
ac27a0ec DK |
457 | goto exit_dind; |
458 | ||
b4097142 TT |
459 | err = ext4_journal_get_write_access(handle, *primary); |
460 | if (unlikely(err)) | |
ac27a0ec DK |
461 | goto exit_sbh; |
462 | ||
b4097142 TT |
463 | err = ext4_journal_get_write_access(handle, dind); |
464 | if (unlikely(err)) | |
465 | ext4_std_error(sb, err); | |
ac27a0ec | 466 | |
617ba13b | 467 | /* ext4_reserve_inode_write() gets a reference on the iloc */ |
b4097142 TT |
468 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
469 | if (unlikely(err)) | |
ac27a0ec DK |
470 | goto exit_dindj; |
471 | ||
472 | n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *), | |
216553c4 | 473 | GFP_NOFS); |
ac27a0ec DK |
474 | if (!n_group_desc) { |
475 | err = -ENOMEM; | |
12062ddd | 476 | ext4_warning(sb, |
ac27a0ec DK |
477 | "not enough memory for %lu groups", gdb_num + 1); |
478 | goto exit_inode; | |
479 | } | |
480 | ||
481 | /* | |
482 | * Finally, we have all of the possible failures behind us... | |
483 | * | |
484 | * Remove new GDT block from inode double-indirect block and clear out | |
485 | * the new GDT block for use (which also "frees" the backup GDT blocks | |
486 | * from the reserved inode). We don't need to change the bitmaps for | |
487 | * these blocks, because they are marked as in-use from being in the | |
488 | * reserved inode, and will become GDT blocks (primary and backup). | |
489 | */ | |
617ba13b | 490 | data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0; |
b4097142 TT |
491 | err = ext4_handle_dirty_metadata(handle, NULL, dind); |
492 | if (unlikely(err)) { | |
493 | ext4_std_error(sb, err); | |
494 | goto exit_inode; | |
495 | } | |
ac27a0ec | 496 | inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9; |
617ba13b | 497 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec | 498 | memset((*primary)->b_data, 0, sb->s_blocksize); |
b4097142 TT |
499 | err = ext4_handle_dirty_metadata(handle, NULL, *primary); |
500 | if (unlikely(err)) { | |
501 | ext4_std_error(sb, err); | |
502 | goto exit_inode; | |
503 | } | |
504 | brelse(dind); | |
ac27a0ec | 505 | |
617ba13b | 506 | o_group_desc = EXT4_SB(sb)->s_group_desc; |
ac27a0ec | 507 | memcpy(n_group_desc, o_group_desc, |
617ba13b | 508 | EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); |
ac27a0ec | 509 | n_group_desc[gdb_num] = *primary; |
617ba13b MC |
510 | EXT4_SB(sb)->s_group_desc = n_group_desc; |
511 | EXT4_SB(sb)->s_gdb_count++; | |
ac27a0ec DK |
512 | kfree(o_group_desc); |
513 | ||
e8546d06 | 514 | le16_add_cpu(&es->s_reserved_gdt_blocks, -1); |
b4097142 TT |
515 | err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
516 | if (err) | |
517 | ext4_std_error(sb, err); | |
ac27a0ec | 518 | |
b4097142 | 519 | return err; |
ac27a0ec DK |
520 | |
521 | exit_inode: | |
537a0310 | 522 | /* ext4_handle_release_buffer(handle, iloc.bh); */ |
ac27a0ec DK |
523 | brelse(iloc.bh); |
524 | exit_dindj: | |
537a0310 | 525 | /* ext4_handle_release_buffer(handle, dind); */ |
ac27a0ec | 526 | exit_sbh: |
537a0310 | 527 | /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */ |
ac27a0ec DK |
528 | exit_dind: |
529 | brelse(dind); | |
530 | exit_bh: | |
531 | brelse(*primary); | |
532 | ||
617ba13b | 533 | ext4_debug("leaving with error %d\n", err); |
ac27a0ec DK |
534 | return err; |
535 | } | |
536 | ||
537 | /* | |
538 | * Called when we are adding a new group which has a backup copy of each of | |
539 | * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. | |
540 | * We need to add these reserved backup GDT blocks to the resize inode, so | |
541 | * that they are kept for future resizing and not allocated to files. | |
542 | * | |
543 | * Each reserved backup GDT block will go into a different indirect block. | |
544 | * The indirect blocks are actually the primary reserved GDT blocks, | |
545 | * so we know in advance what their block numbers are. We only get the | |
546 | * double-indirect block to verify it is pointing to the primary reserved | |
547 | * GDT blocks so we don't overwrite a data block by accident. The reserved | |
548 | * backup GDT blocks are stored in their reserved primary GDT block. | |
549 | */ | |
550 | static int reserve_backup_gdb(handle_t *handle, struct inode *inode, | |
617ba13b | 551 | struct ext4_new_group_data *input) |
ac27a0ec DK |
552 | { |
553 | struct super_block *sb = inode->i_sb; | |
617ba13b | 554 | int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks); |
ac27a0ec DK |
555 | struct buffer_head **primary; |
556 | struct buffer_head *dind; | |
617ba13b MC |
557 | struct ext4_iloc iloc; |
558 | ext4_fsblk_t blk; | |
ac27a0ec DK |
559 | __le32 *data, *end; |
560 | int gdbackups = 0; | |
561 | int res, i; | |
562 | int err; | |
563 | ||
216553c4 | 564 | primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS); |
ac27a0ec DK |
565 | if (!primary) |
566 | return -ENOMEM; | |
567 | ||
617ba13b | 568 | data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK; |
ac27a0ec DK |
569 | dind = sb_bread(sb, le32_to_cpu(*data)); |
570 | if (!dind) { | |
571 | err = -EIO; | |
572 | goto exit_free; | |
573 | } | |
574 | ||
617ba13b | 575 | blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count; |
94460093 JB |
576 | data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count % |
577 | EXT4_ADDR_PER_BLOCK(sb)); | |
617ba13b | 578 | end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb); |
ac27a0ec DK |
579 | |
580 | /* Get each reserved primary GDT block and verify it holds backups */ | |
581 | for (res = 0; res < reserved_gdb; res++, blk++) { | |
582 | if (le32_to_cpu(*data) != blk) { | |
12062ddd | 583 | ext4_warning(sb, "reserved block %llu" |
ac27a0ec DK |
584 | " not at offset %ld", |
585 | blk, | |
586 | (long)(data - (__le32 *)dind->b_data)); | |
587 | err = -EINVAL; | |
588 | goto exit_bh; | |
589 | } | |
590 | primary[res] = sb_bread(sb, blk); | |
591 | if (!primary[res]) { | |
592 | err = -EIO; | |
593 | goto exit_bh; | |
594 | } | |
595 | if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) { | |
596 | brelse(primary[res]); | |
597 | err = gdbackups; | |
598 | goto exit_bh; | |
599 | } | |
600 | if (++data >= end) | |
601 | data = (__le32 *)dind->b_data; | |
602 | } | |
603 | ||
604 | for (i = 0; i < reserved_gdb; i++) { | |
617ba13b | 605 | if ((err = ext4_journal_get_write_access(handle, primary[i]))) { |
ac27a0ec DK |
606 | /* |
607 | int j; | |
608 | for (j = 0; j < i; j++) | |
537a0310 | 609 | ext4_handle_release_buffer(handle, primary[j]); |
ac27a0ec DK |
610 | */ |
611 | goto exit_bh; | |
612 | } | |
613 | } | |
614 | ||
617ba13b | 615 | if ((err = ext4_reserve_inode_write(handle, inode, &iloc))) |
ac27a0ec DK |
616 | goto exit_bh; |
617 | ||
618 | /* | |
619 | * Finally we can add each of the reserved backup GDT blocks from | |
620 | * the new group to its reserved primary GDT block. | |
621 | */ | |
617ba13b | 622 | blk = input->group * EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
623 | for (i = 0; i < reserved_gdb; i++) { |
624 | int err2; | |
625 | data = (__le32 *)primary[i]->b_data; | |
626 | /* printk("reserving backup %lu[%u] = %lu\n", | |
627 | primary[i]->b_blocknr, gdbackups, | |
628 | blk + primary[i]->b_blocknr); */ | |
629 | data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); | |
0390131b | 630 | err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]); |
ac27a0ec DK |
631 | if (!err) |
632 | err = err2; | |
633 | } | |
634 | inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9; | |
617ba13b | 635 | ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
636 | |
637 | exit_bh: | |
638 | while (--res >= 0) | |
639 | brelse(primary[res]); | |
640 | brelse(dind); | |
641 | ||
642 | exit_free: | |
643 | kfree(primary); | |
644 | ||
645 | return err; | |
646 | } | |
647 | ||
648 | /* | |
617ba13b | 649 | * Update the backup copies of the ext4 metadata. These don't need to be part |
ac27a0ec DK |
650 | * of the main resize transaction, because e2fsck will re-write them if there |
651 | * is a problem (basically only OOM will cause a problem). However, we | |
652 | * _should_ update the backups if possible, in case the primary gets trashed | |
653 | * for some reason and we need to run e2fsck from a backup superblock. The | |
654 | * important part is that the new block and inode counts are in the backup | |
655 | * superblocks, and the location of the new group metadata in the GDT backups. | |
656 | * | |
32ed5058 TT |
657 | * We do not need take the s_resize_lock for this, because these |
658 | * blocks are not otherwise touched by the filesystem code when it is | |
659 | * mounted. We don't need to worry about last changing from | |
660 | * sbi->s_groups_count, because the worst that can happen is that we | |
661 | * do not copy the full number of backups at this time. The resize | |
662 | * which changed s_groups_count will backup again. | |
ac27a0ec DK |
663 | */ |
664 | static void update_backups(struct super_block *sb, | |
665 | int blk_off, char *data, int size) | |
666 | { | |
617ba13b | 667 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
fd2d4291 | 668 | const ext4_group_t last = sbi->s_groups_count; |
617ba13b | 669 | const int bpg = EXT4_BLOCKS_PER_GROUP(sb); |
ac27a0ec DK |
670 | unsigned three = 1; |
671 | unsigned five = 5; | |
672 | unsigned seven = 7; | |
fd2d4291 | 673 | ext4_group_t group; |
ac27a0ec DK |
674 | int rest = sb->s_blocksize - size; |
675 | handle_t *handle; | |
676 | int err = 0, err2; | |
677 | ||
617ba13b | 678 | handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA); |
ac27a0ec DK |
679 | if (IS_ERR(handle)) { |
680 | group = 1; | |
681 | err = PTR_ERR(handle); | |
682 | goto exit_err; | |
683 | } | |
684 | ||
617ba13b | 685 | while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) { |
ac27a0ec DK |
686 | struct buffer_head *bh; |
687 | ||
688 | /* Out of journal space, and can't get more - abort - so sad */ | |
0390131b FM |
689 | if (ext4_handle_valid(handle) && |
690 | handle->h_buffer_credits == 0 && | |
617ba13b MC |
691 | ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) && |
692 | (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA))) | |
ac27a0ec DK |
693 | break; |
694 | ||
695 | bh = sb_getblk(sb, group * bpg + blk_off); | |
696 | if (!bh) { | |
697 | err = -EIO; | |
698 | break; | |
699 | } | |
617ba13b | 700 | ext4_debug("update metadata backup %#04lx\n", |
ac27a0ec | 701 | (unsigned long)bh->b_blocknr); |
617ba13b | 702 | if ((err = ext4_journal_get_write_access(handle, bh))) |
ac27a0ec DK |
703 | break; |
704 | lock_buffer(bh); | |
705 | memcpy(bh->b_data, data, size); | |
706 | if (rest) | |
707 | memset(bh->b_data + size, 0, rest); | |
708 | set_buffer_uptodate(bh); | |
709 | unlock_buffer(bh); | |
b4097142 TT |
710 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
711 | if (unlikely(err)) | |
712 | ext4_std_error(sb, err); | |
ac27a0ec DK |
713 | brelse(bh); |
714 | } | |
617ba13b | 715 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
716 | err = err2; |
717 | ||
718 | /* | |
719 | * Ugh! Need to have e2fsck write the backup copies. It is too | |
720 | * late to revert the resize, we shouldn't fail just because of | |
721 | * the backup copies (they are only needed in case of corruption). | |
722 | * | |
723 | * However, if we got here we have a journal problem too, so we | |
724 | * can't really start a transaction to mark the superblock. | |
725 | * Chicken out and just set the flag on the hope it will be written | |
726 | * to disk, and if not - we will simply wait until next fsck. | |
727 | */ | |
728 | exit_err: | |
729 | if (err) { | |
12062ddd | 730 | ext4_warning(sb, "can't update backup for group %u (err %d), " |
ac27a0ec | 731 | "forcing fsck on next reboot", group, err); |
617ba13b MC |
732 | sbi->s_mount_state &= ~EXT4_VALID_FS; |
733 | sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS); | |
ac27a0ec DK |
734 | mark_buffer_dirty(sbi->s_sbh); |
735 | } | |
736 | } | |
737 | ||
738 | /* Add group descriptor data to an existing or new group descriptor block. | |
739 | * Ensure we handle all possible error conditions _before_ we start modifying | |
740 | * the filesystem, because we cannot abort the transaction and not have it | |
741 | * write the data to disk. | |
742 | * | |
743 | * If we are on a GDT block boundary, we need to get the reserved GDT block. | |
744 | * Otherwise, we may need to add backup GDT blocks for a sparse group. | |
745 | * | |
746 | * We only need to hold the superblock lock while we are actually adding | |
747 | * in the new group's counts to the superblock. Prior to that we have | |
748 | * not really "added" the group at all. We re-check that we are still | |
749 | * adding in the last group in case things have changed since verifying. | |
750 | */ | |
617ba13b | 751 | int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) |
ac27a0ec | 752 | { |
617ba13b MC |
753 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
754 | struct ext4_super_block *es = sbi->s_es; | |
755 | int reserved_gdb = ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec DK |
756 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; |
757 | struct buffer_head *primary = NULL; | |
617ba13b | 758 | struct ext4_group_desc *gdp; |
ac27a0ec DK |
759 | struct inode *inode = NULL; |
760 | handle_t *handle; | |
761 | int gdb_off, gdb_num; | |
762 | int err, err2; | |
763 | ||
617ba13b MC |
764 | gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb); |
765 | gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 766 | |
617ba13b MC |
767 | if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb, |
768 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
12062ddd | 769 | ext4_warning(sb, "Can't resize non-sparse filesystem further"); |
ac27a0ec DK |
770 | return -EPERM; |
771 | } | |
772 | ||
bd81d8ee LV |
773 | if (ext4_blocks_count(es) + input->blocks_count < |
774 | ext4_blocks_count(es)) { | |
12062ddd | 775 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
776 | return -EINVAL; |
777 | } | |
778 | ||
617ba13b | 779 | if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) < |
ac27a0ec | 780 | le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 781 | ext4_warning(sb, "inodes_count overflow"); |
ac27a0ec DK |
782 | return -EINVAL; |
783 | } | |
784 | ||
785 | if (reserved_gdb || gdb_off == 0) { | |
617ba13b | 786 | if (!EXT4_HAS_COMPAT_FEATURE(sb, |
37609fd5 JB |
787 | EXT4_FEATURE_COMPAT_RESIZE_INODE) |
788 | || !le16_to_cpu(es->s_reserved_gdt_blocks)) { | |
12062ddd | 789 | ext4_warning(sb, |
ac27a0ec DK |
790 | "No reserved GDT blocks, can't resize"); |
791 | return -EPERM; | |
792 | } | |
1d1fe1ee DH |
793 | inode = ext4_iget(sb, EXT4_RESIZE_INO); |
794 | if (IS_ERR(inode)) { | |
12062ddd | 795 | ext4_warning(sb, "Error opening resize inode"); |
1d1fe1ee | 796 | return PTR_ERR(inode); |
ac27a0ec DK |
797 | } |
798 | } | |
799 | ||
920313a7 | 800 | |
ac27a0ec DK |
801 | if ((err = verify_group_input(sb, input))) |
802 | goto exit_put; | |
803 | ||
804 | if ((err = setup_new_group_blocks(sb, input))) | |
805 | goto exit_put; | |
806 | ||
807 | /* | |
808 | * We will always be modifying at least the superblock and a GDT | |
809 | * block. If we are adding a group past the last current GDT block, | |
810 | * we will also modify the inode and the dindirect block. If we | |
811 | * are adding a group with superblock/GDT backups we will also | |
812 | * modify each of the reserved GDT dindirect blocks. | |
813 | */ | |
617ba13b MC |
814 | handle = ext4_journal_start_sb(sb, |
815 | ext4_bg_has_super(sb, input->group) ? | |
ac27a0ec DK |
816 | 3 + reserved_gdb : 4); |
817 | if (IS_ERR(handle)) { | |
818 | err = PTR_ERR(handle); | |
819 | goto exit_put; | |
820 | } | |
821 | ||
617ba13b | 822 | if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh))) |
ac27a0ec DK |
823 | goto exit_journal; |
824 | ||
2b2d6d01 TT |
825 | /* |
826 | * We will only either add reserved group blocks to a backup group | |
827 | * or remove reserved blocks for the first group in a new group block. | |
828 | * Doing both would be mean more complex code, and sane people don't | |
829 | * use non-sparse filesystems anymore. This is already checked above. | |
830 | */ | |
ac27a0ec DK |
831 | if (gdb_off) { |
832 | primary = sbi->s_group_desc[gdb_num]; | |
617ba13b | 833 | if ((err = ext4_journal_get_write_access(handle, primary))) |
ac27a0ec DK |
834 | goto exit_journal; |
835 | ||
617ba13b | 836 | if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) && |
ac27a0ec DK |
837 | (err = reserve_backup_gdb(handle, inode, input))) |
838 | goto exit_journal; | |
839 | } else if ((err = add_new_gdb(handle, inode, input, &primary))) | |
840 | goto exit_journal; | |
841 | ||
2b2d6d01 TT |
842 | /* |
843 | * OK, now we've set up the new group. Time to make it active. | |
844 | * | |
2b2d6d01 TT |
845 | * so we have to be safe wrt. concurrent accesses the group |
846 | * data. So we need to be careful to set all of the relevant | |
847 | * group descriptor data etc. *before* we enable the group. | |
848 | * | |
849 | * The key field here is sbi->s_groups_count: as long as | |
850 | * that retains its old value, nobody is going to access the new | |
851 | * group. | |
852 | * | |
853 | * So first we update all the descriptor metadata for the new | |
854 | * group; then we update the total disk blocks count; then we | |
855 | * update the groups count to enable the group; then finally we | |
856 | * update the free space counts so that the system can start | |
857 | * using the new disk blocks. | |
858 | */ | |
ac27a0ec DK |
859 | |
860 | /* Update group descriptor block for new group */ | |
2856922c FB |
861 | gdp = (struct ext4_group_desc *)((char *)primary->b_data + |
862 | gdb_off * EXT4_DESC_SIZE(sb)); | |
ac27a0ec | 863 | |
fdff73f0 | 864 | memset(gdp, 0, EXT4_DESC_SIZE(sb)); |
8fadc143 AR |
865 | ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */ |
866 | ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */ | |
867 | ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */ | |
560671a0 AK |
868 | ext4_free_blks_set(sb, gdp, input->free_blocks_count); |
869 | ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb)); | |
fdff73f0 | 870 | gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED); |
717d50e4 | 871 | gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp); |
ac27a0ec | 872 | |
5f21b0e6 FB |
873 | /* |
874 | * We can allocate memory for mb_alloc based on the new group | |
875 | * descriptor | |
876 | */ | |
920313a7 | 877 | err = ext4_mb_add_groupinfo(sb, input->group, gdp); |
08c3a813 | 878 | if (err) |
c2ea3fde TT |
879 | goto exit_journal; |
880 | ||
ac27a0ec DK |
881 | /* |
882 | * Make the new blocks and inodes valid next. We do this before | |
883 | * increasing the group count so that once the group is enabled, | |
884 | * all of its blocks and inodes are already valid. | |
885 | * | |
886 | * We always allocate group-by-group, then block-by-block or | |
887 | * inode-by-inode within a group, so enabling these | |
888 | * blocks/inodes before the group is live won't actually let us | |
889 | * allocate the new space yet. | |
890 | */ | |
bd81d8ee | 891 | ext4_blocks_count_set(es, ext4_blocks_count(es) + |
ac27a0ec | 892 | input->blocks_count); |
e8546d06 | 893 | le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec DK |
894 | |
895 | /* | |
896 | * We need to protect s_groups_count against other CPUs seeing | |
897 | * inconsistent state in the superblock. | |
898 | * | |
899 | * The precise rules we use are: | |
900 | * | |
ac27a0ec DK |
901 | * * Writers must perform a smp_wmb() after updating all dependent |
902 | * data and before modifying the groups count | |
903 | * | |
ac27a0ec DK |
904 | * * Readers must perform an smp_rmb() after reading the groups count |
905 | * and before reading any dependent data. | |
906 | * | |
907 | * NB. These rules can be relaxed when checking the group count | |
908 | * while freeing data, as we can only allocate from a block | |
909 | * group after serialising against the group count, and we can | |
910 | * only then free after serialising in turn against that | |
911 | * allocation. | |
912 | */ | |
913 | smp_wmb(); | |
914 | ||
915 | /* Update the global fs size fields */ | |
916 | sbi->s_groups_count++; | |
917 | ||
b4097142 TT |
918 | err = ext4_handle_dirty_metadata(handle, NULL, primary); |
919 | if (unlikely(err)) { | |
920 | ext4_std_error(sb, err); | |
921 | goto exit_journal; | |
922 | } | |
ac27a0ec DK |
923 | |
924 | /* Update the reserved block counts only once the new group is | |
925 | * active. */ | |
bd81d8ee | 926 | ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) + |
ac27a0ec DK |
927 | input->reserved_blocks); |
928 | ||
929 | /* Update the free space counts */ | |
aa0dff2d | 930 | percpu_counter_add(&sbi->s_freeblocks_counter, |
ac27a0ec | 931 | input->free_blocks_count); |
aa0dff2d | 932 | percpu_counter_add(&sbi->s_freeinodes_counter, |
617ba13b | 933 | EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec | 934 | |
42007efd ES |
935 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) && |
936 | sbi->s_log_groups_per_flex) { | |
c62a11fd FB |
937 | ext4_group_t flex_group; |
938 | flex_group = ext4_flex_group(sbi, input->group); | |
9f24e420 TT |
939 | atomic_add(input->free_blocks_count, |
940 | &sbi->s_flex_groups[flex_group].free_blocks); | |
941 | atomic_add(EXT4_INODES_PER_GROUP(sb), | |
942 | &sbi->s_flex_groups[flex_group].free_inodes); | |
c62a11fd FB |
943 | } |
944 | ||
a0375156 | 945 | ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
946 | |
947 | exit_journal: | |
617ba13b | 948 | if ((err2 = ext4_journal_stop(handle)) && !err) |
ac27a0ec DK |
949 | err = err2; |
950 | if (!err) { | |
951 | update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, | |
617ba13b | 952 | sizeof(struct ext4_super_block)); |
ac27a0ec DK |
953 | update_backups(sb, primary->b_blocknr, primary->b_data, |
954 | primary->b_size); | |
955 | } | |
956 | exit_put: | |
957 | iput(inode); | |
958 | return err; | |
617ba13b | 959 | } /* ext4_group_add */ |
ac27a0ec | 960 | |
2b2d6d01 TT |
961 | /* |
962 | * Extend the filesystem to the new number of blocks specified. This entry | |
ac27a0ec DK |
963 | * point is only used to extend the current filesystem to the end of the last |
964 | * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" | |
965 | * for emergencies (because it has no dependencies on reserved blocks). | |
966 | * | |
617ba13b | 967 | * If we _really_ wanted, we could use default values to call ext4_group_add() |
ac27a0ec DK |
968 | * allow the "remount" trick to work for arbitrary resizing, assuming enough |
969 | * GDT blocks are reserved to grow to the desired size. | |
970 | */ | |
617ba13b MC |
971 | int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, |
972 | ext4_fsblk_t n_blocks_count) | |
ac27a0ec | 973 | { |
617ba13b | 974 | ext4_fsblk_t o_blocks_count; |
617ba13b MC |
975 | ext4_grpblk_t last; |
976 | ext4_grpblk_t add; | |
af5bc92d | 977 | struct buffer_head *bh; |
ac27a0ec | 978 | handle_t *handle; |
cc7365df | 979 | int err, err2; |
5f21b0e6 | 980 | ext4_group_t group; |
ac27a0ec | 981 | |
bd81d8ee | 982 | o_blocks_count = ext4_blocks_count(es); |
ac27a0ec DK |
983 | |
984 | if (test_opt(sb, DEBUG)) | |
2b79b09d | 985 | printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n", |
ac27a0ec DK |
986 | o_blocks_count, n_blocks_count); |
987 | ||
988 | if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) | |
989 | return 0; | |
990 | ||
991 | if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) { | |
617ba13b | 992 | printk(KERN_ERR "EXT4-fs: filesystem on %s:" |
2ae02107 | 993 | " too large to resize to %llu blocks safely\n", |
ac27a0ec DK |
994 | sb->s_id, n_blocks_count); |
995 | if (sizeof(sector_t) < 8) | |
12062ddd | 996 | ext4_warning(sb, "CONFIG_LBDAF not enabled"); |
ac27a0ec DK |
997 | return -EINVAL; |
998 | } | |
999 | ||
1000 | if (n_blocks_count < o_blocks_count) { | |
12062ddd | 1001 | ext4_warning(sb, "can't shrink FS - resize aborted"); |
8f82f840 | 1002 | return -EINVAL; |
ac27a0ec DK |
1003 | } |
1004 | ||
1005 | /* Handle the remaining blocks in the last group only. */ | |
5f21b0e6 | 1006 | ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last); |
ac27a0ec DK |
1007 | |
1008 | if (last == 0) { | |
12062ddd | 1009 | ext4_warning(sb, "need to use ext2online to resize further"); |
ac27a0ec DK |
1010 | return -EPERM; |
1011 | } | |
1012 | ||
617ba13b | 1013 | add = EXT4_BLOCKS_PER_GROUP(sb) - last; |
ac27a0ec DK |
1014 | |
1015 | if (o_blocks_count + add < o_blocks_count) { | |
12062ddd | 1016 | ext4_warning(sb, "blocks_count overflow"); |
ac27a0ec DK |
1017 | return -EINVAL; |
1018 | } | |
1019 | ||
1020 | if (o_blocks_count + add > n_blocks_count) | |
1021 | add = n_blocks_count - o_blocks_count; | |
1022 | ||
1023 | if (o_blocks_count + add < n_blocks_count) | |
12062ddd | 1024 | ext4_warning(sb, "will only finish group (%llu blocks, %u new)", |
ac27a0ec DK |
1025 | o_blocks_count + add, add); |
1026 | ||
1027 | /* See if the device is actually as big as what was requested */ | |
2b2d6d01 | 1028 | bh = sb_bread(sb, o_blocks_count + add - 1); |
ac27a0ec | 1029 | if (!bh) { |
12062ddd | 1030 | ext4_warning(sb, "can't read last block, resize aborted"); |
ac27a0ec DK |
1031 | return -ENOSPC; |
1032 | } | |
1033 | brelse(bh); | |
1034 | ||
1035 | /* We will update the superblock, one block bitmap, and | |
617ba13b | 1036 | * one group descriptor via ext4_free_blocks(). |
ac27a0ec | 1037 | */ |
617ba13b | 1038 | handle = ext4_journal_start_sb(sb, 3); |
ac27a0ec DK |
1039 | if (IS_ERR(handle)) { |
1040 | err = PTR_ERR(handle); | |
12062ddd | 1041 | ext4_warning(sb, "error %d on journal start", err); |
ac27a0ec DK |
1042 | goto exit_put; |
1043 | } | |
1044 | ||
617ba13b MC |
1045 | if ((err = ext4_journal_get_write_access(handle, |
1046 | EXT4_SB(sb)->s_sbh))) { | |
12062ddd | 1047 | ext4_warning(sb, "error %d on journal write access", err); |
617ba13b | 1048 | ext4_journal_stop(handle); |
ac27a0ec DK |
1049 | goto exit_put; |
1050 | } | |
bd81d8ee | 1051 | ext4_blocks_count_set(es, o_blocks_count + add); |
c549a95d | 1052 | ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, |
ac27a0ec | 1053 | o_blocks_count + add); |
e21675d4 | 1054 | /* We add the blocks to the bitmap and set the group need init bit */ |
cc7365df | 1055 | err = ext4_group_add_blocks(handle, sb, o_blocks_count, add); |
a0375156 | 1056 | ext4_handle_dirty_super(handle, sb); |
2ae02107 | 1057 | ext4_debug("freed blocks %llu through %llu\n", o_blocks_count, |
ac27a0ec | 1058 | o_blocks_count + add); |
cc7365df YY |
1059 | err2 = ext4_journal_stop(handle); |
1060 | if (!err && err2) | |
1061 | err = err2; | |
1062 | ||
1063 | if (err) | |
ac27a0ec | 1064 | goto exit_put; |
5f21b0e6 | 1065 | |
ac27a0ec | 1066 | if (test_opt(sb, DEBUG)) |
bd81d8ee LV |
1067 | printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n", |
1068 | ext4_blocks_count(es)); | |
617ba13b MC |
1069 | update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es, |
1070 | sizeof(struct ext4_super_block)); | |
ac27a0ec DK |
1071 | exit_put: |
1072 | return err; | |
617ba13b | 1073 | } /* ext4_group_extend */ |