]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/ext3/resize.c | |
3 | * | |
4 | * Support for resizing an ext3 filesystem while it is mounted. | |
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 | #include <linux/config.h> | |
12 | ||
13 | #define EXT3FS_DEBUG | |
14 | ||
15 | #include <linux/sched.h> | |
16 | #include <linux/smp_lock.h> | |
17 | #include <linux/ext3_jbd.h> | |
18 | ||
19 | #include <linux/errno.h> | |
20 | #include <linux/slab.h> | |
21 | ||
22 | ||
23 | #define outside(b, first, last) ((b) < (first) || (b) >= (last)) | |
24 | #define inside(b, first, last) ((b) >= (first) && (b) < (last)) | |
25 | ||
26 | static int verify_group_input(struct super_block *sb, | |
27 | struct ext3_new_group_data *input) | |
28 | { | |
29 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
30 | struct ext3_super_block *es = sbi->s_es; | |
31 | unsigned start = le32_to_cpu(es->s_blocks_count); | |
32 | unsigned end = start + input->blocks_count; | |
33 | unsigned group = input->group; | |
34 | unsigned itend = input->inode_table + EXT3_SB(sb)->s_itb_per_group; | |
35 | unsigned overhead = ext3_bg_has_super(sb, group) ? | |
36 | (1 + ext3_bg_num_gdb(sb, group) + | |
37 | le16_to_cpu(es->s_reserved_gdt_blocks)) : 0; | |
38 | unsigned metaend = start + overhead; | |
39 | struct buffer_head *bh = NULL; | |
40 | int free_blocks_count; | |
41 | int err = -EINVAL; | |
42 | ||
43 | input->free_blocks_count = free_blocks_count = | |
44 | input->blocks_count - 2 - overhead - sbi->s_itb_per_group; | |
45 | ||
46 | if (test_opt(sb, DEBUG)) | |
47 | printk(KERN_DEBUG "EXT3-fs: adding %s group %u: %u blocks " | |
48 | "(%d free, %u reserved)\n", | |
49 | ext3_bg_has_super(sb, input->group) ? "normal" : | |
50 | "no-super", input->group, input->blocks_count, | |
51 | free_blocks_count, input->reserved_blocks); | |
52 | ||
53 | if (group != sbi->s_groups_count) | |
54 | ext3_warning(sb, __FUNCTION__, | |
55 | "Cannot add at group %u (only %lu groups)", | |
56 | input->group, sbi->s_groups_count); | |
57 | else if ((start - le32_to_cpu(es->s_first_data_block)) % | |
58 | EXT3_BLOCKS_PER_GROUP(sb)) | |
59 | ext3_warning(sb, __FUNCTION__, "Last group not full"); | |
60 | else if (input->reserved_blocks > input->blocks_count / 5) | |
61 | ext3_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)", | |
62 | input->reserved_blocks); | |
63 | else if (free_blocks_count < 0) | |
64 | ext3_warning(sb, __FUNCTION__, "Bad blocks count %u", | |
65 | input->blocks_count); | |
66 | else if (!(bh = sb_bread(sb, end - 1))) | |
67 | ext3_warning(sb, __FUNCTION__, "Cannot read last block (%u)", | |
68 | end - 1); | |
69 | else if (outside(input->block_bitmap, start, end)) | |
70 | ext3_warning(sb, __FUNCTION__, | |
71 | "Block bitmap not in group (block %u)", | |
72 | input->block_bitmap); | |
73 | else if (outside(input->inode_bitmap, start, end)) | |
74 | ext3_warning(sb, __FUNCTION__, | |
75 | "Inode bitmap not in group (block %u)", | |
76 | input->inode_bitmap); | |
77 | else if (outside(input->inode_table, start, end) || | |
78 | outside(itend - 1, start, end)) | |
79 | ext3_warning(sb, __FUNCTION__, | |
80 | "Inode table not in group (blocks %u-%u)", | |
81 | input->inode_table, itend - 1); | |
82 | else if (input->inode_bitmap == input->block_bitmap) | |
83 | ext3_warning(sb, __FUNCTION__, | |
84 | "Block bitmap same as inode bitmap (%u)", | |
85 | input->block_bitmap); | |
86 | else if (inside(input->block_bitmap, input->inode_table, itend)) | |
87 | ext3_warning(sb, __FUNCTION__, | |
88 | "Block bitmap (%u) in inode table (%u-%u)", | |
89 | input->block_bitmap, input->inode_table, itend-1); | |
90 | else if (inside(input->inode_bitmap, input->inode_table, itend)) | |
91 | ext3_warning(sb, __FUNCTION__, | |
92 | "Inode bitmap (%u) in inode table (%u-%u)", | |
93 | input->inode_bitmap, input->inode_table, itend-1); | |
94 | else if (inside(input->block_bitmap, start, metaend)) | |
95 | ext3_warning(sb, __FUNCTION__, | |
96 | "Block bitmap (%u) in GDT table (%u-%u)", | |
97 | input->block_bitmap, start, metaend - 1); | |
98 | else if (inside(input->inode_bitmap, start, metaend)) | |
99 | ext3_warning(sb, __FUNCTION__, | |
100 | "Inode bitmap (%u) in GDT table (%u-%u)", | |
101 | input->inode_bitmap, start, metaend - 1); | |
102 | else if (inside(input->inode_table, start, metaend) || | |
103 | inside(itend - 1, start, metaend)) | |
104 | ext3_warning(sb, __FUNCTION__, | |
105 | "Inode table (%u-%u) overlaps GDT table (%u-%u)", | |
106 | input->inode_table, itend - 1, start, metaend - 1); | |
107 | else | |
108 | err = 0; | |
109 | brelse(bh); | |
110 | ||
111 | return err; | |
112 | } | |
113 | ||
114 | static struct buffer_head *bclean(handle_t *handle, struct super_block *sb, | |
115 | unsigned long blk) | |
116 | { | |
117 | struct buffer_head *bh; | |
118 | int err; | |
119 | ||
120 | bh = sb_getblk(sb, blk); | |
2973dfdb GOC |
121 | if (!bh) |
122 | return ERR_PTR(-EIO); | |
1da177e4 LT |
123 | if ((err = ext3_journal_get_write_access(handle, bh))) { |
124 | brelse(bh); | |
125 | bh = ERR_PTR(err); | |
126 | } else { | |
127 | lock_buffer(bh); | |
128 | memset(bh->b_data, 0, sb->s_blocksize); | |
129 | set_buffer_uptodate(bh); | |
130 | unlock_buffer(bh); | |
131 | } | |
132 | ||
133 | return bh; | |
134 | } | |
135 | ||
136 | /* | |
137 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
138 | * need to use it within a single byte (to ensure we get endianness right). | |
139 | * We can use memset for the rest of the bitmap as there are no other users. | |
140 | */ | |
141 | static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) | |
142 | { | |
143 | int i; | |
144 | ||
145 | if (start_bit >= end_bit) | |
146 | return; | |
147 | ||
148 | ext3_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
149 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
150 | ext3_set_bit(i, bitmap); | |
151 | if (i < end_bit) | |
152 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
153 | } | |
154 | ||
155 | /* | |
156 | * Set up the block and inode bitmaps, and the inode table for the new group. | |
157 | * This doesn't need to be part of the main transaction, since we are only | |
158 | * changing blocks outside the actual filesystem. We still do journaling to | |
159 | * ensure the recovery is correct in case of a failure just after resize. | |
160 | * If any part of this fails, we simply abort the resize. | |
161 | */ | |
162 | static int setup_new_group_blocks(struct super_block *sb, | |
163 | struct ext3_new_group_data *input) | |
164 | { | |
165 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
166 | unsigned long start = input->group * sbi->s_blocks_per_group + | |
167 | le32_to_cpu(sbi->s_es->s_first_data_block); | |
168 | int reserved_gdb = ext3_bg_has_super(sb, input->group) ? | |
169 | le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0; | |
170 | unsigned long gdblocks = ext3_bg_num_gdb(sb, input->group); | |
171 | struct buffer_head *bh; | |
172 | handle_t *handle; | |
173 | unsigned long block; | |
174 | int bit; | |
175 | int i; | |
176 | int err = 0, err2; | |
177 | ||
178 | handle = ext3_journal_start_sb(sb, reserved_gdb + gdblocks + | |
179 | 2 + sbi->s_itb_per_group); | |
180 | if (IS_ERR(handle)) | |
181 | return PTR_ERR(handle); | |
182 | ||
183 | lock_super(sb); | |
184 | if (input->group != sbi->s_groups_count) { | |
185 | err = -EBUSY; | |
186 | goto exit_journal; | |
187 | } | |
188 | ||
189 | if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) { | |
190 | err = PTR_ERR(bh); | |
191 | goto exit_journal; | |
192 | } | |
193 | ||
194 | if (ext3_bg_has_super(sb, input->group)) { | |
195 | ext3_debug("mark backup superblock %#04lx (+0)\n", start); | |
196 | ext3_set_bit(0, bh->b_data); | |
197 | } | |
198 | ||
199 | /* Copy all of the GDT blocks into the backup in this group */ | |
200 | for (i = 0, bit = 1, block = start + 1; | |
201 | i < gdblocks; i++, block++, bit++) { | |
202 | struct buffer_head *gdb; | |
203 | ||
204 | ext3_debug("update backup group %#04lx (+%d)\n", block, bit); | |
205 | ||
206 | gdb = sb_getblk(sb, block); | |
2973dfdb GOC |
207 | if (!gdb) { |
208 | err = -EIO; | |
209 | goto exit_bh; | |
210 | } | |
1da177e4 LT |
211 | if ((err = ext3_journal_get_write_access(handle, gdb))) { |
212 | brelse(gdb); | |
213 | goto exit_bh; | |
214 | } | |
215 | lock_buffer(bh); | |
216 | memcpy(gdb->b_data, sbi->s_group_desc[i], bh->b_size); | |
217 | set_buffer_uptodate(gdb); | |
218 | unlock_buffer(bh); | |
219 | ext3_journal_dirty_metadata(handle, gdb); | |
220 | ext3_set_bit(bit, bh->b_data); | |
221 | brelse(gdb); | |
222 | } | |
223 | ||
224 | /* Zero out all of the reserved backup group descriptor table blocks */ | |
225 | for (i = 0, bit = gdblocks + 1, block = start + bit; | |
226 | i < reserved_gdb; i++, block++, bit++) { | |
227 | struct buffer_head *gdb; | |
228 | ||
229 | ext3_debug("clear reserved block %#04lx (+%d)\n", block, bit); | |
230 | ||
231 | if (IS_ERR(gdb = bclean(handle, sb, block))) { | |
232 | err = PTR_ERR(bh); | |
233 | goto exit_bh; | |
234 | } | |
235 | ext3_journal_dirty_metadata(handle, gdb); | |
236 | ext3_set_bit(bit, bh->b_data); | |
237 | brelse(gdb); | |
238 | } | |
239 | ext3_debug("mark block bitmap %#04x (+%ld)\n", input->block_bitmap, | |
240 | input->block_bitmap - start); | |
241 | ext3_set_bit(input->block_bitmap - start, bh->b_data); | |
242 | ext3_debug("mark inode bitmap %#04x (+%ld)\n", input->inode_bitmap, | |
243 | input->inode_bitmap - start); | |
244 | ext3_set_bit(input->inode_bitmap - start, bh->b_data); | |
245 | ||
246 | /* Zero out all of the inode table blocks */ | |
247 | for (i = 0, block = input->inode_table, bit = block - start; | |
248 | i < sbi->s_itb_per_group; i++, bit++, block++) { | |
249 | struct buffer_head *it; | |
250 | ||
8bdac5d1 | 251 | ext3_debug("clear inode block %#04lx (+%d)\n", block, bit); |
1da177e4 LT |
252 | if (IS_ERR(it = bclean(handle, sb, block))) { |
253 | err = PTR_ERR(it); | |
254 | goto exit_bh; | |
255 | } | |
256 | ext3_journal_dirty_metadata(handle, it); | |
257 | brelse(it); | |
258 | ext3_set_bit(bit, bh->b_data); | |
259 | } | |
260 | mark_bitmap_end(input->blocks_count, EXT3_BLOCKS_PER_GROUP(sb), | |
261 | bh->b_data); | |
262 | ext3_journal_dirty_metadata(handle, bh); | |
263 | brelse(bh); | |
264 | ||
265 | /* Mark unused entries in inode bitmap used */ | |
266 | ext3_debug("clear inode bitmap %#04x (+%ld)\n", | |
267 | input->inode_bitmap, input->inode_bitmap - start); | |
268 | if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) { | |
269 | err = PTR_ERR(bh); | |
270 | goto exit_journal; | |
271 | } | |
272 | ||
273 | mark_bitmap_end(EXT3_INODES_PER_GROUP(sb), EXT3_BLOCKS_PER_GROUP(sb), | |
274 | bh->b_data); | |
275 | ext3_journal_dirty_metadata(handle, bh); | |
276 | exit_bh: | |
277 | brelse(bh); | |
278 | ||
279 | exit_journal: | |
280 | unlock_super(sb); | |
281 | if ((err2 = ext3_journal_stop(handle)) && !err) | |
282 | err = err2; | |
283 | ||
284 | return err; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Iterate through the groups which hold BACKUP superblock/GDT copies in an | |
289 | * ext3 filesystem. The counters should be initialized to 1, 5, and 7 before | |
290 | * calling this for the first time. In a sparse filesystem it will be the | |
291 | * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ... | |
292 | * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ... | |
293 | */ | |
294 | static unsigned ext3_list_backups(struct super_block *sb, unsigned *three, | |
295 | unsigned *five, unsigned *seven) | |
296 | { | |
297 | unsigned *min = three; | |
298 | int mult = 3; | |
299 | unsigned ret; | |
300 | ||
301 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, | |
302 | EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
303 | ret = *min; | |
304 | *min += 1; | |
305 | return ret; | |
306 | } | |
307 | ||
308 | if (*five < *min) { | |
309 | min = five; | |
310 | mult = 5; | |
311 | } | |
312 | if (*seven < *min) { | |
313 | min = seven; | |
314 | mult = 7; | |
315 | } | |
316 | ||
317 | ret = *min; | |
318 | *min *= mult; | |
319 | ||
320 | return ret; | |
321 | } | |
322 | ||
323 | /* | |
324 | * Check that all of the backup GDT blocks are held in the primary GDT block. | |
325 | * It is assumed that they are stored in group order. Returns the number of | |
326 | * groups in current filesystem that have BACKUPS, or -ve error code. | |
327 | */ | |
328 | static int verify_reserved_gdb(struct super_block *sb, | |
329 | struct buffer_head *primary) | |
330 | { | |
331 | const unsigned long blk = primary->b_blocknr; | |
332 | const unsigned long end = EXT3_SB(sb)->s_groups_count; | |
333 | unsigned three = 1; | |
334 | unsigned five = 5; | |
335 | unsigned seven = 7; | |
336 | unsigned grp; | |
337 | __u32 *p = (__u32 *)primary->b_data; | |
338 | int gdbackups = 0; | |
339 | ||
340 | while ((grp = ext3_list_backups(sb, &three, &five, &seven)) < end) { | |
341 | if (le32_to_cpu(*p++) != grp * EXT3_BLOCKS_PER_GROUP(sb) + blk){ | |
342 | ext3_warning(sb, __FUNCTION__, | |
343 | "reserved GDT %ld missing grp %d (%ld)\n", | |
344 | blk, grp, | |
345 | grp * EXT3_BLOCKS_PER_GROUP(sb) + blk); | |
346 | return -EINVAL; | |
347 | } | |
348 | if (++gdbackups > EXT3_ADDR_PER_BLOCK(sb)) | |
349 | return -EFBIG; | |
350 | } | |
351 | ||
352 | return gdbackups; | |
353 | } | |
354 | ||
355 | /* | |
356 | * Called when we need to bring a reserved group descriptor table block into | |
357 | * use from the resize inode. The primary copy of the new GDT block currently | |
358 | * is an indirect block (under the double indirect block in the resize inode). | |
359 | * The new backup GDT blocks will be stored as leaf blocks in this indirect | |
360 | * block, in group order. Even though we know all the block numbers we need, | |
361 | * we check to ensure that the resize inode has actually reserved these blocks. | |
362 | * | |
363 | * Don't need to update the block bitmaps because the blocks are still in use. | |
364 | * | |
365 | * We get all of the error cases out of the way, so that we are sure to not | |
366 | * fail once we start modifying the data on disk, because JBD has no rollback. | |
367 | */ | |
368 | static int add_new_gdb(handle_t *handle, struct inode *inode, | |
369 | struct ext3_new_group_data *input, | |
370 | struct buffer_head **primary) | |
371 | { | |
372 | struct super_block *sb = inode->i_sb; | |
373 | struct ext3_super_block *es = EXT3_SB(sb)->s_es; | |
374 | unsigned long gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb); | |
375 | unsigned long gdblock = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num; | |
376 | struct buffer_head **o_group_desc, **n_group_desc; | |
377 | struct buffer_head *dind; | |
378 | int gdbackups; | |
379 | struct ext3_iloc iloc; | |
380 | __u32 *data; | |
381 | int err; | |
382 | ||
383 | if (test_opt(sb, DEBUG)) | |
384 | printk(KERN_DEBUG | |
385 | "EXT3-fs: ext3_add_new_gdb: adding group block %lu\n", | |
386 | gdb_num); | |
387 | ||
388 | /* | |
389 | * If we are not using the primary superblock/GDT copy don't resize, | |
390 | * because the user tools have no way of handling this. Probably a | |
391 | * bad time to do it anyways. | |
392 | */ | |
393 | if (EXT3_SB(sb)->s_sbh->b_blocknr != | |
394 | le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) { | |
395 | ext3_warning(sb, __FUNCTION__, | |
396 | "won't resize using backup superblock at %llu\n", | |
397 | (unsigned long long)EXT3_SB(sb)->s_sbh->b_blocknr); | |
398 | return -EPERM; | |
399 | } | |
400 | ||
401 | *primary = sb_bread(sb, gdblock); | |
402 | if (!*primary) | |
403 | return -EIO; | |
404 | ||
405 | if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) { | |
406 | err = gdbackups; | |
407 | goto exit_bh; | |
408 | } | |
409 | ||
410 | data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK; | |
411 | dind = sb_bread(sb, le32_to_cpu(*data)); | |
412 | if (!dind) { | |
413 | err = -EIO; | |
414 | goto exit_bh; | |
415 | } | |
416 | ||
417 | data = (__u32 *)dind->b_data; | |
418 | if (le32_to_cpu(data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)]) != gdblock) { | |
419 | ext3_warning(sb, __FUNCTION__, | |
420 | "new group %u GDT block %lu not reserved\n", | |
421 | input->group, gdblock); | |
422 | err = -EINVAL; | |
423 | goto exit_dind; | |
424 | } | |
425 | ||
426 | if ((err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh))) | |
427 | goto exit_dind; | |
428 | ||
429 | if ((err = ext3_journal_get_write_access(handle, *primary))) | |
430 | goto exit_sbh; | |
431 | ||
432 | if ((err = ext3_journal_get_write_access(handle, dind))) | |
433 | goto exit_primary; | |
434 | ||
435 | /* ext3_reserve_inode_write() gets a reference on the iloc */ | |
436 | if ((err = ext3_reserve_inode_write(handle, inode, &iloc))) | |
437 | goto exit_dindj; | |
438 | ||
439 | n_group_desc = (struct buffer_head **)kmalloc((gdb_num + 1) * | |
440 | sizeof(struct buffer_head *), GFP_KERNEL); | |
441 | if (!n_group_desc) { | |
442 | err = -ENOMEM; | |
443 | ext3_warning (sb, __FUNCTION__, | |
444 | "not enough memory for %lu groups", gdb_num + 1); | |
445 | goto exit_inode; | |
446 | } | |
447 | ||
448 | /* | |
449 | * Finally, we have all of the possible failures behind us... | |
450 | * | |
451 | * Remove new GDT block from inode double-indirect block and clear out | |
452 | * the new GDT block for use (which also "frees" the backup GDT blocks | |
453 | * from the reserved inode). We don't need to change the bitmaps for | |
454 | * these blocks, because they are marked as in-use from being in the | |
455 | * reserved inode, and will become GDT blocks (primary and backup). | |
456 | */ | |
457 | data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)] = 0; | |
458 | ext3_journal_dirty_metadata(handle, dind); | |
459 | brelse(dind); | |
460 | inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9; | |
461 | ext3_mark_iloc_dirty(handle, inode, &iloc); | |
462 | memset((*primary)->b_data, 0, sb->s_blocksize); | |
463 | ext3_journal_dirty_metadata(handle, *primary); | |
464 | ||
465 | o_group_desc = EXT3_SB(sb)->s_group_desc; | |
466 | memcpy(n_group_desc, o_group_desc, | |
467 | EXT3_SB(sb)->s_gdb_count * sizeof(struct buffer_head *)); | |
468 | n_group_desc[gdb_num] = *primary; | |
469 | EXT3_SB(sb)->s_group_desc = n_group_desc; | |
470 | EXT3_SB(sb)->s_gdb_count++; | |
471 | kfree(o_group_desc); | |
472 | ||
473 | es->s_reserved_gdt_blocks = | |
474 | cpu_to_le16(le16_to_cpu(es->s_reserved_gdt_blocks) - 1); | |
475 | ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); | |
476 | ||
477 | return 0; | |
478 | ||
479 | exit_inode: | |
480 | //ext3_journal_release_buffer(handle, iloc.bh); | |
481 | brelse(iloc.bh); | |
482 | exit_dindj: | |
483 | //ext3_journal_release_buffer(handle, dind); | |
484 | exit_primary: | |
485 | //ext3_journal_release_buffer(handle, *primary); | |
486 | exit_sbh: | |
487 | //ext3_journal_release_buffer(handle, *primary); | |
488 | exit_dind: | |
489 | brelse(dind); | |
490 | exit_bh: | |
491 | brelse(*primary); | |
492 | ||
493 | ext3_debug("leaving with error %d\n", err); | |
494 | return err; | |
495 | } | |
496 | ||
497 | /* | |
498 | * Called when we are adding a new group which has a backup copy of each of | |
499 | * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks. | |
500 | * We need to add these reserved backup GDT blocks to the resize inode, so | |
501 | * that they are kept for future resizing and not allocated to files. | |
502 | * | |
503 | * Each reserved backup GDT block will go into a different indirect block. | |
504 | * The indirect blocks are actually the primary reserved GDT blocks, | |
505 | * so we know in advance what their block numbers are. We only get the | |
506 | * double-indirect block to verify it is pointing to the primary reserved | |
507 | * GDT blocks so we don't overwrite a data block by accident. The reserved | |
508 | * backup GDT blocks are stored in their reserved primary GDT block. | |
509 | */ | |
510 | static int reserve_backup_gdb(handle_t *handle, struct inode *inode, | |
511 | struct ext3_new_group_data *input) | |
512 | { | |
513 | struct super_block *sb = inode->i_sb; | |
514 | int reserved_gdb =le16_to_cpu(EXT3_SB(sb)->s_es->s_reserved_gdt_blocks); | |
515 | struct buffer_head **primary; | |
516 | struct buffer_head *dind; | |
517 | struct ext3_iloc iloc; | |
518 | unsigned long blk; | |
519 | __u32 *data, *end; | |
520 | int gdbackups = 0; | |
521 | int res, i; | |
522 | int err; | |
523 | ||
524 | primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL); | |
525 | if (!primary) | |
526 | return -ENOMEM; | |
527 | ||
528 | data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK; | |
529 | dind = sb_bread(sb, le32_to_cpu(*data)); | |
530 | if (!dind) { | |
531 | err = -EIO; | |
532 | goto exit_free; | |
533 | } | |
534 | ||
535 | blk = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + EXT3_SB(sb)->s_gdb_count; | |
536 | data = (__u32 *)dind->b_data + EXT3_SB(sb)->s_gdb_count; | |
537 | end = (__u32 *)dind->b_data + EXT3_ADDR_PER_BLOCK(sb); | |
538 | ||
539 | /* Get each reserved primary GDT block and verify it holds backups */ | |
540 | for (res = 0; res < reserved_gdb; res++, blk++) { | |
541 | if (le32_to_cpu(*data) != blk) { | |
542 | ext3_warning(sb, __FUNCTION__, | |
543 | "reserved block %lu not at offset %ld\n", | |
544 | blk, (long)(data - (__u32 *)dind->b_data)); | |
545 | err = -EINVAL; | |
546 | goto exit_bh; | |
547 | } | |
548 | primary[res] = sb_bread(sb, blk); | |
549 | if (!primary[res]) { | |
550 | err = -EIO; | |
551 | goto exit_bh; | |
552 | } | |
553 | if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) { | |
554 | brelse(primary[res]); | |
555 | err = gdbackups; | |
556 | goto exit_bh; | |
557 | } | |
558 | if (++data >= end) | |
559 | data = (__u32 *)dind->b_data; | |
560 | } | |
561 | ||
562 | for (i = 0; i < reserved_gdb; i++) { | |
563 | if ((err = ext3_journal_get_write_access(handle, primary[i]))) { | |
564 | /* | |
565 | int j; | |
566 | for (j = 0; j < i; j++) | |
567 | ext3_journal_release_buffer(handle, primary[j]); | |
568 | */ | |
569 | goto exit_bh; | |
570 | } | |
571 | } | |
572 | ||
573 | if ((err = ext3_reserve_inode_write(handle, inode, &iloc))) | |
574 | goto exit_bh; | |
575 | ||
576 | /* | |
577 | * Finally we can add each of the reserved backup GDT blocks from | |
578 | * the new group to its reserved primary GDT block. | |
579 | */ | |
580 | blk = input->group * EXT3_BLOCKS_PER_GROUP(sb); | |
581 | for (i = 0; i < reserved_gdb; i++) { | |
582 | int err2; | |
583 | data = (__u32 *)primary[i]->b_data; | |
584 | /* printk("reserving backup %lu[%u] = %lu\n", | |
585 | primary[i]->b_blocknr, gdbackups, | |
586 | blk + primary[i]->b_blocknr); */ | |
587 | data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr); | |
588 | err2 = ext3_journal_dirty_metadata(handle, primary[i]); | |
589 | if (!err) | |
590 | err = err2; | |
591 | } | |
592 | inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9; | |
593 | ext3_mark_iloc_dirty(handle, inode, &iloc); | |
594 | ||
595 | exit_bh: | |
596 | while (--res >= 0) | |
597 | brelse(primary[res]); | |
598 | brelse(dind); | |
599 | ||
600 | exit_free: | |
601 | kfree(primary); | |
602 | ||
603 | return err; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Update the backup copies of the ext3 metadata. These don't need to be part | |
608 | * of the main resize transaction, because e2fsck will re-write them if there | |
609 | * is a problem (basically only OOM will cause a problem). However, we | |
610 | * _should_ update the backups if possible, in case the primary gets trashed | |
611 | * for some reason and we need to run e2fsck from a backup superblock. The | |
612 | * important part is that the new block and inode counts are in the backup | |
613 | * superblocks, and the location of the new group metadata in the GDT backups. | |
614 | * | |
615 | * We do not need lock_super() for this, because these blocks are not | |
616 | * otherwise touched by the filesystem code when it is mounted. We don't | |
617 | * need to worry about last changing from sbi->s_groups_count, because the | |
618 | * worst that can happen is that we do not copy the full number of backups | |
619 | * at this time. The resize which changed s_groups_count will backup again. | |
620 | */ | |
621 | static void update_backups(struct super_block *sb, | |
622 | int blk_off, char *data, int size) | |
623 | { | |
624 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
625 | const unsigned long last = sbi->s_groups_count; | |
626 | const int bpg = EXT3_BLOCKS_PER_GROUP(sb); | |
627 | unsigned three = 1; | |
628 | unsigned five = 5; | |
629 | unsigned seven = 7; | |
630 | unsigned group; | |
631 | int rest = sb->s_blocksize - size; | |
632 | handle_t *handle; | |
633 | int err = 0, err2; | |
634 | ||
635 | handle = ext3_journal_start_sb(sb, EXT3_MAX_TRANS_DATA); | |
636 | if (IS_ERR(handle)) { | |
637 | group = 1; | |
638 | err = PTR_ERR(handle); | |
639 | goto exit_err; | |
640 | } | |
641 | ||
642 | while ((group = ext3_list_backups(sb, &three, &five, &seven)) < last) { | |
643 | struct buffer_head *bh; | |
644 | ||
645 | /* Out of journal space, and can't get more - abort - so sad */ | |
646 | if (handle->h_buffer_credits == 0 && | |
647 | ext3_journal_extend(handle, EXT3_MAX_TRANS_DATA) && | |
648 | (err = ext3_journal_restart(handle, EXT3_MAX_TRANS_DATA))) | |
649 | break; | |
650 | ||
651 | bh = sb_getblk(sb, group * bpg + blk_off); | |
2973dfdb GOC |
652 | if (!bh) { |
653 | err = -EIO; | |
654 | break; | |
655 | } | |
8bdac5d1 GOC |
656 | ext3_debug("update metadata backup %#04lx\n", |
657 | (unsigned long)bh->b_blocknr); | |
1da177e4 LT |
658 | if ((err = ext3_journal_get_write_access(handle, bh))) |
659 | break; | |
660 | lock_buffer(bh); | |
661 | memcpy(bh->b_data, data, size); | |
662 | if (rest) | |
663 | memset(bh->b_data + size, 0, rest); | |
664 | set_buffer_uptodate(bh); | |
665 | unlock_buffer(bh); | |
666 | ext3_journal_dirty_metadata(handle, bh); | |
667 | brelse(bh); | |
668 | } | |
669 | if ((err2 = ext3_journal_stop(handle)) && !err) | |
670 | err = err2; | |
671 | ||
672 | /* | |
673 | * Ugh! Need to have e2fsck write the backup copies. It is too | |
674 | * late to revert the resize, we shouldn't fail just because of | |
675 | * the backup copies (they are only needed in case of corruption). | |
676 | * | |
677 | * However, if we got here we have a journal problem too, so we | |
678 | * can't really start a transaction to mark the superblock. | |
679 | * Chicken out and just set the flag on the hope it will be written | |
680 | * to disk, and if not - we will simply wait until next fsck. | |
681 | */ | |
682 | exit_err: | |
683 | if (err) { | |
684 | ext3_warning(sb, __FUNCTION__, | |
685 | "can't update backup for group %d (err %d), " | |
686 | "forcing fsck on next reboot\n", group, err); | |
687 | sbi->s_mount_state &= ~EXT3_VALID_FS; | |
688 | sbi->s_es->s_state &= ~cpu_to_le16(EXT3_VALID_FS); | |
689 | mark_buffer_dirty(sbi->s_sbh); | |
690 | } | |
691 | } | |
692 | ||
693 | /* Add group descriptor data to an existing or new group descriptor block. | |
694 | * Ensure we handle all possible error conditions _before_ we start modifying | |
695 | * the filesystem, because we cannot abort the transaction and not have it | |
696 | * write the data to disk. | |
697 | * | |
698 | * If we are on a GDT block boundary, we need to get the reserved GDT block. | |
699 | * Otherwise, we may need to add backup GDT blocks for a sparse group. | |
700 | * | |
701 | * We only need to hold the superblock lock while we are actually adding | |
702 | * in the new group's counts to the superblock. Prior to that we have | |
703 | * not really "added" the group at all. We re-check that we are still | |
704 | * adding in the last group in case things have changed since verifying. | |
705 | */ | |
706 | int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input) | |
707 | { | |
708 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
709 | struct ext3_super_block *es = sbi->s_es; | |
710 | int reserved_gdb = ext3_bg_has_super(sb, input->group) ? | |
711 | le16_to_cpu(es->s_reserved_gdt_blocks) : 0; | |
712 | struct buffer_head *primary = NULL; | |
713 | struct ext3_group_desc *gdp; | |
714 | struct inode *inode = NULL; | |
715 | handle_t *handle; | |
716 | int gdb_off, gdb_num; | |
717 | int err, err2; | |
718 | ||
719 | gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb); | |
720 | gdb_off = input->group % EXT3_DESC_PER_BLOCK(sb); | |
721 | ||
722 | if (gdb_off == 0 && !EXT3_HAS_RO_COMPAT_FEATURE(sb, | |
723 | EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) { | |
724 | ext3_warning(sb, __FUNCTION__, | |
725 | "Can't resize non-sparse filesystem further\n"); | |
726 | return -EPERM; | |
727 | } | |
728 | ||
729 | if (reserved_gdb || gdb_off == 0) { | |
730 | if (!EXT3_HAS_COMPAT_FEATURE(sb, | |
731 | EXT3_FEATURE_COMPAT_RESIZE_INODE)){ | |
732 | ext3_warning(sb, __FUNCTION__, | |
733 | "No reserved GDT blocks, can't resize\n"); | |
734 | return -EPERM; | |
735 | } | |
736 | inode = iget(sb, EXT3_RESIZE_INO); | |
737 | if (!inode || is_bad_inode(inode)) { | |
738 | ext3_warning(sb, __FUNCTION__, | |
739 | "Error opening resize inode\n"); | |
740 | iput(inode); | |
741 | return -ENOENT; | |
742 | } | |
743 | } | |
744 | ||
745 | if ((err = verify_group_input(sb, input))) | |
746 | goto exit_put; | |
747 | ||
748 | if ((err = setup_new_group_blocks(sb, input))) | |
749 | goto exit_put; | |
750 | ||
751 | /* | |
752 | * We will always be modifying at least the superblock and a GDT | |
753 | * block. If we are adding a group past the last current GDT block, | |
754 | * we will also modify the inode and the dindirect block. If we | |
755 | * are adding a group with superblock/GDT backups we will also | |
756 | * modify each of the reserved GDT dindirect blocks. | |
757 | */ | |
758 | handle = ext3_journal_start_sb(sb, | |
759 | ext3_bg_has_super(sb, input->group) ? | |
760 | 3 + reserved_gdb : 4); | |
761 | if (IS_ERR(handle)) { | |
762 | err = PTR_ERR(handle); | |
763 | goto exit_put; | |
764 | } | |
765 | ||
766 | lock_super(sb); | |
767 | if (input->group != EXT3_SB(sb)->s_groups_count) { | |
768 | ext3_warning(sb, __FUNCTION__, | |
769 | "multiple resizers run on filesystem!\n"); | |
aa877b3d | 770 | err = -EBUSY; |
1da177e4 LT |
771 | goto exit_journal; |
772 | } | |
773 | ||
774 | if ((err = ext3_journal_get_write_access(handle, sbi->s_sbh))) | |
775 | goto exit_journal; | |
776 | ||
777 | /* | |
778 | * We will only either add reserved group blocks to a backup group | |
779 | * or remove reserved blocks for the first group in a new group block. | |
780 | * Doing both would be mean more complex code, and sane people don't | |
781 | * use non-sparse filesystems anymore. This is already checked above. | |
782 | */ | |
783 | if (gdb_off) { | |
784 | primary = sbi->s_group_desc[gdb_num]; | |
785 | if ((err = ext3_journal_get_write_access(handle, primary))) | |
786 | goto exit_journal; | |
787 | ||
788 | if (reserved_gdb && ext3_bg_num_gdb(sb, input->group) && | |
789 | (err = reserve_backup_gdb(handle, inode, input))) | |
790 | goto exit_journal; | |
791 | } else if ((err = add_new_gdb(handle, inode, input, &primary))) | |
792 | goto exit_journal; | |
793 | ||
794 | /* | |
795 | * OK, now we've set up the new group. Time to make it active. | |
796 | * | |
797 | * Current kernels don't lock all allocations via lock_super(), | |
798 | * so we have to be safe wrt. concurrent accesses the group | |
799 | * data. So we need to be careful to set all of the relevant | |
800 | * group descriptor data etc. *before* we enable the group. | |
801 | * | |
802 | * The key field here is EXT3_SB(sb)->s_groups_count: as long as | |
803 | * that retains its old value, nobody is going to access the new | |
804 | * group. | |
805 | * | |
806 | * So first we update all the descriptor metadata for the new | |
807 | * group; then we update the total disk blocks count; then we | |
808 | * update the groups count to enable the group; then finally we | |
809 | * update the free space counts so that the system can start | |
810 | * using the new disk blocks. | |
811 | */ | |
812 | ||
813 | /* Update group descriptor block for new group */ | |
814 | gdp = (struct ext3_group_desc *)primary->b_data + gdb_off; | |
815 | ||
816 | gdp->bg_block_bitmap = cpu_to_le32(input->block_bitmap); | |
817 | gdp->bg_inode_bitmap = cpu_to_le32(input->inode_bitmap); | |
818 | gdp->bg_inode_table = cpu_to_le32(input->inode_table); | |
819 | gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count); | |
820 | gdp->bg_free_inodes_count = cpu_to_le16(EXT3_INODES_PER_GROUP(sb)); | |
821 | ||
822 | /* | |
823 | * Make the new blocks and inodes valid next. We do this before | |
824 | * increasing the group count so that once the group is enabled, | |
825 | * all of its blocks and inodes are already valid. | |
826 | * | |
827 | * We always allocate group-by-group, then block-by-block or | |
828 | * inode-by-inode within a group, so enabling these | |
829 | * blocks/inodes before the group is live won't actually let us | |
830 | * allocate the new space yet. | |
831 | */ | |
832 | es->s_blocks_count = cpu_to_le32(le32_to_cpu(es->s_blocks_count) + | |
833 | input->blocks_count); | |
834 | es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) + | |
835 | EXT3_INODES_PER_GROUP(sb)); | |
836 | ||
837 | /* | |
838 | * We need to protect s_groups_count against other CPUs seeing | |
839 | * inconsistent state in the superblock. | |
840 | * | |
841 | * The precise rules we use are: | |
842 | * | |
843 | * * Writers of s_groups_count *must* hold lock_super | |
844 | * AND | |
845 | * * Writers must perform a smp_wmb() after updating all dependent | |
846 | * data and before modifying the groups count | |
847 | * | |
848 | * * Readers must hold lock_super() over the access | |
849 | * OR | |
850 | * * Readers must perform an smp_rmb() after reading the groups count | |
851 | * and before reading any dependent data. | |
852 | * | |
853 | * NB. These rules can be relaxed when checking the group count | |
854 | * while freeing data, as we can only allocate from a block | |
855 | * group after serialising against the group count, and we can | |
856 | * only then free after serialising in turn against that | |
857 | * allocation. | |
858 | */ | |
859 | smp_wmb(); | |
860 | ||
861 | /* Update the global fs size fields */ | |
862 | EXT3_SB(sb)->s_groups_count++; | |
863 | ||
864 | ext3_journal_dirty_metadata(handle, primary); | |
865 | ||
866 | /* Update the reserved block counts only once the new group is | |
867 | * active. */ | |
868 | es->s_r_blocks_count = cpu_to_le32(le32_to_cpu(es->s_r_blocks_count) + | |
869 | input->reserved_blocks); | |
870 | ||
871 | /* Update the free space counts */ | |
872 | percpu_counter_mod(&sbi->s_freeblocks_counter, | |
873 | input->free_blocks_count); | |
874 | percpu_counter_mod(&sbi->s_freeinodes_counter, | |
875 | EXT3_INODES_PER_GROUP(sb)); | |
876 | ||
877 | ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); | |
878 | sb->s_dirt = 1; | |
879 | ||
880 | exit_journal: | |
881 | unlock_super(sb); | |
882 | if ((err2 = ext3_journal_stop(handle)) && !err) | |
883 | err = err2; | |
884 | if (!err) { | |
885 | update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es, | |
886 | sizeof(struct ext3_super_block)); | |
887 | update_backups(sb, primary->b_blocknr, primary->b_data, | |
888 | primary->b_size); | |
889 | } | |
890 | exit_put: | |
891 | iput(inode); | |
892 | return err; | |
893 | } /* ext3_group_add */ | |
894 | ||
895 | /* Extend the filesystem to the new number of blocks specified. This entry | |
896 | * point is only used to extend the current filesystem to the end of the last | |
897 | * existing group. It can be accessed via ioctl, or by "remount,resize=<size>" | |
898 | * for emergencies (because it has no dependencies on reserved blocks). | |
899 | * | |
900 | * If we _really_ wanted, we could use default values to call ext3_group_add() | |
901 | * allow the "remount" trick to work for arbitrary resizing, assuming enough | |
902 | * GDT blocks are reserved to grow to the desired size. | |
903 | */ | |
904 | int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es, | |
905 | unsigned long n_blocks_count) | |
906 | { | |
907 | unsigned long o_blocks_count; | |
908 | unsigned long o_groups_count; | |
909 | unsigned long last; | |
910 | int add; | |
911 | struct buffer_head * bh; | |
912 | handle_t *handle; | |
913 | int err, freed_blocks; | |
914 | ||
915 | /* We don't need to worry about locking wrt other resizers just | |
916 | * yet: we're going to revalidate es->s_blocks_count after | |
917 | * taking lock_super() below. */ | |
918 | o_blocks_count = le32_to_cpu(es->s_blocks_count); | |
919 | o_groups_count = EXT3_SB(sb)->s_groups_count; | |
920 | ||
921 | if (test_opt(sb, DEBUG)) | |
922 | printk(KERN_DEBUG "EXT3-fs: extending last group from %lu to %lu blocks\n", | |
923 | o_blocks_count, n_blocks_count); | |
924 | ||
925 | if (n_blocks_count == 0 || n_blocks_count == o_blocks_count) | |
926 | return 0; | |
927 | ||
928 | if (n_blocks_count < o_blocks_count) { | |
929 | ext3_warning(sb, __FUNCTION__, | |
930 | "can't shrink FS - resize aborted"); | |
931 | return -EBUSY; | |
932 | } | |
933 | ||
934 | /* Handle the remaining blocks in the last group only. */ | |
935 | last = (o_blocks_count - le32_to_cpu(es->s_first_data_block)) % | |
936 | EXT3_BLOCKS_PER_GROUP(sb); | |
937 | ||
938 | if (last == 0) { | |
939 | ext3_warning(sb, __FUNCTION__, | |
940 | "need to use ext2online to resize further\n"); | |
941 | return -EPERM; | |
942 | } | |
943 | ||
944 | add = EXT3_BLOCKS_PER_GROUP(sb) - last; | |
945 | ||
946 | if (o_blocks_count + add > n_blocks_count) | |
947 | add = n_blocks_count - o_blocks_count; | |
948 | ||
949 | if (o_blocks_count + add < n_blocks_count) | |
950 | ext3_warning(sb, __FUNCTION__, | |
951 | "will only finish group (%lu blocks, %u new)", | |
952 | o_blocks_count + add, add); | |
953 | ||
954 | /* See if the device is actually as big as what was requested */ | |
955 | bh = sb_bread(sb, o_blocks_count + add -1); | |
956 | if (!bh) { | |
957 | ext3_warning(sb, __FUNCTION__, | |
958 | "can't read last block, resize aborted"); | |
959 | return -ENOSPC; | |
960 | } | |
961 | brelse(bh); | |
962 | ||
963 | /* We will update the superblock, one block bitmap, and | |
964 | * one group descriptor via ext3_free_blocks(). | |
965 | */ | |
966 | handle = ext3_journal_start_sb(sb, 3); | |
967 | if (IS_ERR(handle)) { | |
968 | err = PTR_ERR(handle); | |
969 | ext3_warning(sb, __FUNCTION__, "error %d on journal start",err); | |
970 | goto exit_put; | |
971 | } | |
972 | ||
973 | lock_super(sb); | |
974 | if (o_blocks_count != le32_to_cpu(es->s_blocks_count)) { | |
975 | ext3_warning(sb, __FUNCTION__, | |
976 | "multiple resizers run on filesystem!\n"); | |
977 | err = -EBUSY; | |
978 | goto exit_put; | |
979 | } | |
980 | ||
981 | if ((err = ext3_journal_get_write_access(handle, | |
982 | EXT3_SB(sb)->s_sbh))) { | |
983 | ext3_warning(sb, __FUNCTION__, | |
984 | "error %d on journal write access", err); | |
985 | unlock_super(sb); | |
986 | ext3_journal_stop(handle); | |
987 | goto exit_put; | |
988 | } | |
989 | es->s_blocks_count = cpu_to_le32(o_blocks_count + add); | |
990 | ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); | |
991 | sb->s_dirt = 1; | |
992 | unlock_super(sb); | |
993 | ext3_debug("freeing blocks %ld through %ld\n", o_blocks_count, | |
994 | o_blocks_count + add); | |
995 | ext3_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks); | |
996 | ext3_debug("freed blocks %ld through %ld\n", o_blocks_count, | |
997 | o_blocks_count + add); | |
998 | if ((err = ext3_journal_stop(handle))) | |
999 | goto exit_put; | |
1000 | if (test_opt(sb, DEBUG)) | |
1001 | printk(KERN_DEBUG "EXT3-fs: extended group to %u blocks\n", | |
1002 | le32_to_cpu(es->s_blocks_count)); | |
1003 | update_backups(sb, EXT3_SB(sb)->s_sbh->b_blocknr, (char *)es, | |
1004 | sizeof(struct ext3_super_block)); | |
1005 | exit_put: | |
1006 | return err; | |
1007 | } /* ext3_group_extend */ |