aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4/resize.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ext4/resize.c')
-rw-r--r--fs/ext4/resize.c1045
1 files changed, 1045 insertions, 0 deletions
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
new file mode 100644
index 000000000000..1e9578052cd3
--- /dev/null
+++ b/fs/ext4/resize.c
@@ -0,0 +1,1045 @@
1/*
2 * linux/fs/ext4/resize.c
3 *
4 * Support for resizing an ext4 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
12#define EXT4FS_DEBUG
13
14#include <linux/sched.h>
15#include <linux/smp_lock.h>
16#include <linux/ext4_jbd2.h>
17
18#include <linux/errno.h>
19#include <linux/slab.h>
20
21
22#define outside(b, first, last) ((b) < (first) || (b) >= (last))
23#define inside(b, first, last) ((b) >= (first) && (b) < (last))
24
25static int verify_group_input(struct super_block *sb,
26 struct ext4_new_group_data *input)
27{
28 struct ext4_sb_info *sbi = EXT4_SB(sb);
29 struct ext4_super_block *es = sbi->s_es;
30 ext4_fsblk_t start = ext4_blocks_count(es);
31 ext4_fsblk_t end = start + input->blocks_count;
32 unsigned group = input->group;
33 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
34 unsigned overhead = ext4_bg_has_super(sb, group) ?
35 (1 + ext4_bg_num_gdb(sb, group) +
36 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
37 ext4_fsblk_t metaend = start + overhead;
38 struct buffer_head *bh = NULL;
39 ext4_grpblk_t free_blocks_count, offset;
40 int err = -EINVAL;
41
42 input->free_blocks_count = free_blocks_count =
43 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
44
45 if (test_opt(sb, DEBUG))
46 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
47 "(%d free, %u reserved)\n",
48 ext4_bg_has_super(sb, input->group) ? "normal" :
49 "no-super", input->group, input->blocks_count,
50 free_blocks_count, input->reserved_blocks);
51
52 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
53 if (group != sbi->s_groups_count)
54 ext4_warning(sb, __FUNCTION__,
55 "Cannot add at group %u (only %lu groups)",
56 input->group, sbi->s_groups_count);
57 else if (offset != 0)
58 ext4_warning(sb, __FUNCTION__, "Last group not full");
59 else if (input->reserved_blocks > input->blocks_count / 5)
60 ext4_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)",
61 input->reserved_blocks);
62 else if (free_blocks_count < 0)
63 ext4_warning(sb, __FUNCTION__, "Bad blocks count %u",
64 input->blocks_count);
65 else if (!(bh = sb_bread(sb, end - 1)))
66 ext4_warning(sb, __FUNCTION__,
67 "Cannot read last block (%llu)",
68 end - 1);
69 else if (outside(input->block_bitmap, start, end))
70 ext4_warning(sb, __FUNCTION__,
71 "Block bitmap not in group (block %llu)",
72 input->block_bitmap);
73 else if (outside(input->inode_bitmap, start, end))
74 ext4_warning(sb, __FUNCTION__,
75 "Inode bitmap not in group (block %llu)",
76 input->inode_bitmap);
77 else if (outside(input->inode_table, start, end) ||
78 outside(itend - 1, start, end))
79 ext4_warning(sb, __FUNCTION__,
80 "Inode table not in group (blocks %llu-%llu)",
81 input->inode_table, itend - 1);
82 else if (input->inode_bitmap == input->block_bitmap)
83 ext4_warning(sb, __FUNCTION__,
84 "Block bitmap same as inode bitmap (%llu)",
85 input->block_bitmap);
86 else if (inside(input->block_bitmap, input->inode_table, itend))
87 ext4_warning(sb, __FUNCTION__,
88 "Block bitmap (%llu) in inode table (%llu-%llu)",
89 input->block_bitmap, input->inode_table, itend-1);
90 else if (inside(input->inode_bitmap, input->inode_table, itend))
91 ext4_warning(sb, __FUNCTION__,
92 "Inode bitmap (%llu) in inode table (%llu-%llu)",
93 input->inode_bitmap, input->inode_table, itend-1);
94 else if (inside(input->block_bitmap, start, metaend))
95 ext4_warning(sb, __FUNCTION__,
96 "Block bitmap (%llu) in GDT table"
97 " (%llu-%llu)",
98 input->block_bitmap, start, metaend - 1);
99 else if (inside(input->inode_bitmap, start, metaend))
100 ext4_warning(sb, __FUNCTION__,
101 "Inode bitmap (%llu) in GDT table"
102 " (%llu-%llu)",
103 input->inode_bitmap, start, metaend - 1);
104 else if (inside(input->inode_table, start, metaend) ||
105 inside(itend - 1, start, metaend))
106 ext4_warning(sb, __FUNCTION__,
107 "Inode table (%llu-%llu) overlaps"
108 "GDT table (%llu-%llu)",
109 input->inode_table, itend - 1, start, metaend - 1);
110 else
111 err = 0;
112 brelse(bh);
113
114 return err;
115}
116
117static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
118 ext4_fsblk_t blk)
119{
120 struct buffer_head *bh;
121 int err;
122
123 bh = sb_getblk(sb, blk);
124 if (!bh)
125 return ERR_PTR(-EIO);
126 if ((err = ext4_journal_get_write_access(handle, bh))) {
127 brelse(bh);
128 bh = ERR_PTR(err);
129 } else {
130 lock_buffer(bh);
131 memset(bh->b_data, 0, sb->s_blocksize);
132 set_buffer_uptodate(bh);
133 unlock_buffer(bh);
134 }
135
136 return bh;
137}
138
139/*
140 * To avoid calling the atomic setbit hundreds or thousands of times, we only
141 * need to use it within a single byte (to ensure we get endianness right).
142 * We can use memset for the rest of the bitmap as there are no other users.
143 */
144static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
145{
146 int i;
147
148 if (start_bit >= end_bit)
149 return;
150
151 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
152 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
153 ext4_set_bit(i, bitmap);
154 if (i < end_bit)
155 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
156}
157
158/*
159 * Set up the block and inode bitmaps, and the inode table for the new group.
160 * This doesn't need to be part of the main transaction, since we are only
161 * changing blocks outside the actual filesystem. We still do journaling to
162 * ensure the recovery is correct in case of a failure just after resize.
163 * If any part of this fails, we simply abort the resize.
164 */
165static int setup_new_group_blocks(struct super_block *sb,
166 struct ext4_new_group_data *input)
167{
168 struct ext4_sb_info *sbi = EXT4_SB(sb);
169 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
170 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
171 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
172 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
173 struct buffer_head *bh;
174 handle_t *handle;
175 ext4_fsblk_t block;
176 ext4_grpblk_t bit;
177 int i;
178 int err = 0, err2;
179
180 handle = ext4_journal_start_sb(sb, reserved_gdb + gdblocks +
181 2 + sbi->s_itb_per_group);
182 if (IS_ERR(handle))
183 return PTR_ERR(handle);
184
185 lock_super(sb);
186 if (input->group != sbi->s_groups_count) {
187 err = -EBUSY;
188 goto exit_journal;
189 }
190
191 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
192 err = PTR_ERR(bh);
193 goto exit_journal;
194 }
195
196 if (ext4_bg_has_super(sb, input->group)) {
197 ext4_debug("mark backup superblock %#04lx (+0)\n", start);
198 ext4_set_bit(0, bh->b_data);
199 }
200
201 /* Copy all of the GDT blocks into the backup in this group */
202 for (i = 0, bit = 1, block = start + 1;
203 i < gdblocks; i++, block++, bit++) {
204 struct buffer_head *gdb;
205
206 ext4_debug("update backup group %#04lx (+%d)\n", block, bit);
207
208 gdb = sb_getblk(sb, block);
209 if (!gdb) {
210 err = -EIO;
211 goto exit_bh;
212 }
213 if ((err = ext4_journal_get_write_access(handle, gdb))) {
214 brelse(gdb);
215 goto exit_bh;
216 }
217 lock_buffer(bh);
218 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, bh->b_size);
219 set_buffer_uptodate(gdb);
220 unlock_buffer(bh);
221 ext4_journal_dirty_metadata(handle, gdb);
222 ext4_set_bit(bit, bh->b_data);
223 brelse(gdb);
224 }
225
226 /* Zero out all of the reserved backup group descriptor table blocks */
227 for (i = 0, bit = gdblocks + 1, block = start + bit;
228 i < reserved_gdb; i++, block++, bit++) {
229 struct buffer_head *gdb;
230
231 ext4_debug("clear reserved block %#04lx (+%d)\n", block, bit);
232
233 if (IS_ERR(gdb = bclean(handle, sb, block))) {
234 err = PTR_ERR(bh);
235 goto exit_bh;
236 }
237 ext4_journal_dirty_metadata(handle, gdb);
238 ext4_set_bit(bit, bh->b_data);
239 brelse(gdb);
240 }
241 ext4_debug("mark block bitmap %#04x (+%ld)\n", input->block_bitmap,
242 input->block_bitmap - start);
243 ext4_set_bit(input->block_bitmap - start, bh->b_data);
244 ext4_debug("mark inode bitmap %#04x (+%ld)\n", input->inode_bitmap,
245 input->inode_bitmap - start);
246 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
247
248 /* Zero out all of the inode table blocks */
249 for (i = 0, block = input->inode_table, bit = block - start;
250 i < sbi->s_itb_per_group; i++, bit++, block++) {
251 struct buffer_head *it;
252
253 ext4_debug("clear inode block %#04lx (+%d)\n", block, bit);
254 if (IS_ERR(it = bclean(handle, sb, block))) {
255 err = PTR_ERR(it);
256 goto exit_bh;
257 }
258 ext4_journal_dirty_metadata(handle, it);
259 brelse(it);
260 ext4_set_bit(bit, bh->b_data);
261 }
262 mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
263 bh->b_data);
264 ext4_journal_dirty_metadata(handle, bh);
265 brelse(bh);
266
267 /* Mark unused entries in inode bitmap used */
268 ext4_debug("clear inode bitmap %#04x (+%ld)\n",
269 input->inode_bitmap, input->inode_bitmap - start);
270 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
271 err = PTR_ERR(bh);
272 goto exit_journal;
273 }
274
275 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
276 bh->b_data);
277 ext4_journal_dirty_metadata(handle, bh);
278exit_bh:
279 brelse(bh);
280
281exit_journal:
282 unlock_super(sb);
283 if ((err2 = ext4_journal_stop(handle)) && !err)
284 err = err2;
285
286 return err;
287}
288
289
290/*
291 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
292 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
293 * calling this for the first time. In a sparse filesystem it will be the
294 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
295 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
296 */
297static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
298 unsigned *five, unsigned *seven)
299{
300 unsigned *min = three;
301 int mult = 3;
302 unsigned ret;
303
304 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
305 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
306 ret = *min;
307 *min += 1;
308 return ret;
309 }
310
311 if (*five < *min) {
312 min = five;
313 mult = 5;
314 }
315 if (*seven < *min) {
316 min = seven;
317 mult = 7;
318 }
319
320 ret = *min;
321 *min *= mult;
322
323 return ret;
324}
325
326/*
327 * Check that all of the backup GDT blocks are held in the primary GDT block.
328 * It is assumed that they are stored in group order. Returns the number of
329 * groups in current filesystem that have BACKUPS, or -ve error code.
330 */
331static int verify_reserved_gdb(struct super_block *sb,
332 struct buffer_head *primary)
333{
334 const ext4_fsblk_t blk = primary->b_blocknr;
335 const unsigned long end = EXT4_SB(sb)->s_groups_count;
336 unsigned three = 1;
337 unsigned five = 5;
338 unsigned seven = 7;
339 unsigned grp;
340 __le32 *p = (__le32 *)primary->b_data;
341 int gdbackups = 0;
342
343 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
344 if (le32_to_cpu(*p++) !=
345 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
346 ext4_warning(sb, __FUNCTION__,
347 "reserved GDT %llu"
348 " missing grp %d (%llu)",
349 blk, grp,
350 grp *
351 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
352 blk);
353 return -EINVAL;
354 }
355 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
356 return -EFBIG;
357 }
358
359 return gdbackups;
360}
361
362/*
363 * Called when we need to bring a reserved group descriptor table block into
364 * use from the resize inode. The primary copy of the new GDT block currently
365 * is an indirect block (under the double indirect block in the resize inode).
366 * The new backup GDT blocks will be stored as leaf blocks in this indirect
367 * block, in group order. Even though we know all the block numbers we need,
368 * we check to ensure that the resize inode has actually reserved these blocks.
369 *
370 * Don't need to update the block bitmaps because the blocks are still in use.
371 *
372 * We get all of the error cases out of the way, so that we are sure to not
373 * fail once we start modifying the data on disk, because JBD has no rollback.
374 */
375static int add_new_gdb(handle_t *handle, struct inode *inode,
376 struct ext4_new_group_data *input,
377 struct buffer_head **primary)
378{
379 struct super_block *sb = inode->i_sb;
380 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
381 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
382 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
383 struct buffer_head **o_group_desc, **n_group_desc;
384 struct buffer_head *dind;
385 int gdbackups;
386 struct ext4_iloc iloc;
387 __le32 *data;
388 int err;
389
390 if (test_opt(sb, DEBUG))
391 printk(KERN_DEBUG
392 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
393 gdb_num);
394
395 /*
396 * If we are not using the primary superblock/GDT copy don't resize,
397 * because the user tools have no way of handling this. Probably a
398 * bad time to do it anyways.
399 */
400 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
401 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
402 ext4_warning(sb, __FUNCTION__,
403 "won't resize using backup superblock at %llu",
404 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
405 return -EPERM;
406 }
407
408 *primary = sb_bread(sb, gdblock);
409 if (!*primary)
410 return -EIO;
411
412 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
413 err = gdbackups;
414 goto exit_bh;
415 }
416
417 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
418 dind = sb_bread(sb, le32_to_cpu(*data));
419 if (!dind) {
420 err = -EIO;
421 goto exit_bh;
422 }
423
424 data = (__le32 *)dind->b_data;
425 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
426 ext4_warning(sb, __FUNCTION__,
427 "new group %u GDT block %llu not reserved",
428 input->group, gdblock);
429 err = -EINVAL;
430 goto exit_dind;
431 }
432
433 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
434 goto exit_dind;
435
436 if ((err = ext4_journal_get_write_access(handle, *primary)))
437 goto exit_sbh;
438
439 if ((err = ext4_journal_get_write_access(handle, dind)))
440 goto exit_primary;
441
442 /* ext4_reserve_inode_write() gets a reference on the iloc */
443 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
444 goto exit_dindj;
445
446 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
447 GFP_KERNEL);
448 if (!n_group_desc) {
449 err = -ENOMEM;
450 ext4_warning (sb, __FUNCTION__,
451 "not enough memory for %lu groups", gdb_num + 1);
452 goto exit_inode;
453 }
454
455 /*
456 * Finally, we have all of the possible failures behind us...
457 *
458 * Remove new GDT block from inode double-indirect block and clear out
459 * the new GDT block for use (which also "frees" the backup GDT blocks
460 * from the reserved inode). We don't need to change the bitmaps for
461 * these blocks, because they are marked as in-use from being in the
462 * reserved inode, and will become GDT blocks (primary and backup).
463 */
464 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
465 ext4_journal_dirty_metadata(handle, dind);
466 brelse(dind);
467 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
468 ext4_mark_iloc_dirty(handle, inode, &iloc);
469 memset((*primary)->b_data, 0, sb->s_blocksize);
470 ext4_journal_dirty_metadata(handle, *primary);
471
472 o_group_desc = EXT4_SB(sb)->s_group_desc;
473 memcpy(n_group_desc, o_group_desc,
474 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
475 n_group_desc[gdb_num] = *primary;
476 EXT4_SB(sb)->s_group_desc = n_group_desc;
477 EXT4_SB(sb)->s_gdb_count++;
478 kfree(o_group_desc);
479
480 es->s_reserved_gdt_blocks =
481 cpu_to_le16(le16_to_cpu(es->s_reserved_gdt_blocks) - 1);
482 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
483
484 return 0;
485
486exit_inode:
487 //ext4_journal_release_buffer(handle, iloc.bh);
488 brelse(iloc.bh);
489exit_dindj:
490 //ext4_journal_release_buffer(handle, dind);
491exit_primary:
492 //ext4_journal_release_buffer(handle, *primary);
493exit_sbh:
494 //ext4_journal_release_buffer(handle, *primary);
495exit_dind:
496 brelse(dind);
497exit_bh:
498 brelse(*primary);
499
500 ext4_debug("leaving with error %d\n", err);
501 return err;
502}
503
504/*
505 * Called when we are adding a new group which has a backup copy of each of
506 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
507 * We need to add these reserved backup GDT blocks to the resize inode, so
508 * that they are kept for future resizing and not allocated to files.
509 *
510 * Each reserved backup GDT block will go into a different indirect block.
511 * The indirect blocks are actually the primary reserved GDT blocks,
512 * so we know in advance what their block numbers are. We only get the
513 * double-indirect block to verify it is pointing to the primary reserved
514 * GDT blocks so we don't overwrite a data block by accident. The reserved
515 * backup GDT blocks are stored in their reserved primary GDT block.
516 */
517static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
518 struct ext4_new_group_data *input)
519{
520 struct super_block *sb = inode->i_sb;
521 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
522 struct buffer_head **primary;
523 struct buffer_head *dind;
524 struct ext4_iloc iloc;
525 ext4_fsblk_t blk;
526 __le32 *data, *end;
527 int gdbackups = 0;
528 int res, i;
529 int err;
530
531 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL);
532 if (!primary)
533 return -ENOMEM;
534
535 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
536 dind = sb_bread(sb, le32_to_cpu(*data));
537 if (!dind) {
538 err = -EIO;
539 goto exit_free;
540 }
541
542 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
543 data = (__le32 *)dind->b_data + EXT4_SB(sb)->s_gdb_count;
544 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
545
546 /* Get each reserved primary GDT block and verify it holds backups */
547 for (res = 0; res < reserved_gdb; res++, blk++) {
548 if (le32_to_cpu(*data) != blk) {
549 ext4_warning(sb, __FUNCTION__,
550 "reserved block %llu"
551 " not at offset %ld",
552 blk,
553 (long)(data - (__le32 *)dind->b_data));
554 err = -EINVAL;
555 goto exit_bh;
556 }
557 primary[res] = sb_bread(sb, blk);
558 if (!primary[res]) {
559 err = -EIO;
560 goto exit_bh;
561 }
562 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
563 brelse(primary[res]);
564 err = gdbackups;
565 goto exit_bh;
566 }
567 if (++data >= end)
568 data = (__le32 *)dind->b_data;
569 }
570
571 for (i = 0; i < reserved_gdb; i++) {
572 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
573 /*
574 int j;
575 for (j = 0; j < i; j++)
576 ext4_journal_release_buffer(handle, primary[j]);
577 */
578 goto exit_bh;
579 }
580 }
581
582 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
583 goto exit_bh;
584
585 /*
586 * Finally we can add each of the reserved backup GDT blocks from
587 * the new group to its reserved primary GDT block.
588 */
589 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
590 for (i = 0; i < reserved_gdb; i++) {
591 int err2;
592 data = (__le32 *)primary[i]->b_data;
593 /* printk("reserving backup %lu[%u] = %lu\n",
594 primary[i]->b_blocknr, gdbackups,
595 blk + primary[i]->b_blocknr); */
596 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
597 err2 = ext4_journal_dirty_metadata(handle, primary[i]);
598 if (!err)
599 err = err2;
600 }
601 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
602 ext4_mark_iloc_dirty(handle, inode, &iloc);
603
604exit_bh:
605 while (--res >= 0)
606 brelse(primary[res]);
607 brelse(dind);
608
609exit_free:
610 kfree(primary);
611
612 return err;
613}
614
615/*
616 * Update the backup copies of the ext4 metadata. These don't need to be part
617 * of the main resize transaction, because e2fsck will re-write them if there
618 * is a problem (basically only OOM will cause a problem). However, we
619 * _should_ update the backups if possible, in case the primary gets trashed
620 * for some reason and we need to run e2fsck from a backup superblock. The
621 * important part is that the new block and inode counts are in the backup
622 * superblocks, and the location of the new group metadata in the GDT backups.
623 *
624 * We do not need lock_super() for this, because these blocks are not
625 * otherwise touched by the filesystem code when it is mounted. We don't
626 * need to worry about last changing from sbi->s_groups_count, because the
627 * worst that can happen is that we do not copy the full number of backups
628 * at this time. The resize which changed s_groups_count will backup again.
629 */
630static void update_backups(struct super_block *sb,
631 int blk_off, char *data, int size)
632{
633 struct ext4_sb_info *sbi = EXT4_SB(sb);
634 const unsigned long last = sbi->s_groups_count;
635 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
636 unsigned three = 1;
637 unsigned five = 5;
638 unsigned seven = 7;
639 unsigned group;
640 int rest = sb->s_blocksize - size;
641 handle_t *handle;
642 int err = 0, err2;
643
644 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
645 if (IS_ERR(handle)) {
646 group = 1;
647 err = PTR_ERR(handle);
648 goto exit_err;
649 }
650
651 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
652 struct buffer_head *bh;
653
654 /* Out of journal space, and can't get more - abort - so sad */
655 if (handle->h_buffer_credits == 0 &&
656 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
657 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
658 break;
659
660 bh = sb_getblk(sb, group * bpg + blk_off);
661 if (!bh) {
662 err = -EIO;
663 break;
664 }
665 ext4_debug("update metadata backup %#04lx\n",
666 (unsigned long)bh->b_blocknr);
667 if ((err = ext4_journal_get_write_access(handle, bh)))
668 break;
669 lock_buffer(bh);
670 memcpy(bh->b_data, data, size);
671 if (rest)
672 memset(bh->b_data + size, 0, rest);
673 set_buffer_uptodate(bh);
674 unlock_buffer(bh);
675 ext4_journal_dirty_metadata(handle, bh);
676 brelse(bh);
677 }
678 if ((err2 = ext4_journal_stop(handle)) && !err)
679 err = err2;
680
681 /*
682 * Ugh! Need to have e2fsck write the backup copies. It is too
683 * late to revert the resize, we shouldn't fail just because of
684 * the backup copies (they are only needed in case of corruption).
685 *
686 * However, if we got here we have a journal problem too, so we
687 * can't really start a transaction to mark the superblock.
688 * Chicken out and just set the flag on the hope it will be written
689 * to disk, and if not - we will simply wait until next fsck.
690 */
691exit_err:
692 if (err) {
693 ext4_warning(sb, __FUNCTION__,
694 "can't update backup for group %d (err %d), "
695 "forcing fsck on next reboot", group, err);
696 sbi->s_mount_state &= ~EXT4_VALID_FS;
697 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
698 mark_buffer_dirty(sbi->s_sbh);
699 }
700}
701
702/* Add group descriptor data to an existing or new group descriptor block.
703 * Ensure we handle all possible error conditions _before_ we start modifying
704 * the filesystem, because we cannot abort the transaction and not have it
705 * write the data to disk.
706 *
707 * If we are on a GDT block boundary, we need to get the reserved GDT block.
708 * Otherwise, we may need to add backup GDT blocks for a sparse group.
709 *
710 * We only need to hold the superblock lock while we are actually adding
711 * in the new group's counts to the superblock. Prior to that we have
712 * not really "added" the group at all. We re-check that we are still
713 * adding in the last group in case things have changed since verifying.
714 */
715int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
716{
717 struct ext4_sb_info *sbi = EXT4_SB(sb);
718 struct ext4_super_block *es = sbi->s_es;
719 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
720 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
721 struct buffer_head *primary = NULL;
722 struct ext4_group_desc *gdp;
723 struct inode *inode = NULL;
724 handle_t *handle;
725 int gdb_off, gdb_num;
726 int err, err2;
727
728 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
729 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
730
731 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
732 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
733 ext4_warning(sb, __FUNCTION__,
734 "Can't resize non-sparse filesystem further");
735 return -EPERM;
736 }
737
738 if (ext4_blocks_count(es) + input->blocks_count <
739 ext4_blocks_count(es)) {
740 ext4_warning(sb, __FUNCTION__, "blocks_count overflow\n");
741 return -EINVAL;
742 }
743
744 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
745 le32_to_cpu(es->s_inodes_count)) {
746 ext4_warning(sb, __FUNCTION__, "inodes_count overflow\n");
747 return -EINVAL;
748 }
749
750 if (reserved_gdb || gdb_off == 0) {
751 if (!EXT4_HAS_COMPAT_FEATURE(sb,
752 EXT4_FEATURE_COMPAT_RESIZE_INODE)){
753 ext4_warning(sb, __FUNCTION__,
754 "No reserved GDT blocks, can't resize");
755 return -EPERM;
756 }
757 inode = iget(sb, EXT4_RESIZE_INO);
758 if (!inode || is_bad_inode(inode)) {
759 ext4_warning(sb, __FUNCTION__,
760 "Error opening resize inode");
761 iput(inode);
762 return -ENOENT;
763 }
764 }
765
766 if ((err = verify_group_input(sb, input)))
767 goto exit_put;
768
769 if ((err = setup_new_group_blocks(sb, input)))
770 goto exit_put;
771
772 /*
773 * We will always be modifying at least the superblock and a GDT
774 * block. If we are adding a group past the last current GDT block,
775 * we will also modify the inode and the dindirect block. If we
776 * are adding a group with superblock/GDT backups we will also
777 * modify each of the reserved GDT dindirect blocks.
778 */
779 handle = ext4_journal_start_sb(sb,
780 ext4_bg_has_super(sb, input->group) ?
781 3 + reserved_gdb : 4);
782 if (IS_ERR(handle)) {
783 err = PTR_ERR(handle);
784 goto exit_put;
785 }
786
787 lock_super(sb);
788 if (input->group != sbi->s_groups_count) {
789 ext4_warning(sb, __FUNCTION__,
790 "multiple resizers run on filesystem!");
791 err = -EBUSY;
792 goto exit_journal;
793 }
794
795 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
796 goto exit_journal;
797
798 /*
799 * We will only either add reserved group blocks to a backup group
800 * or remove reserved blocks for the first group in a new group block.
801 * Doing both would be mean more complex code, and sane people don't
802 * use non-sparse filesystems anymore. This is already checked above.
803 */
804 if (gdb_off) {
805 primary = sbi->s_group_desc[gdb_num];
806 if ((err = ext4_journal_get_write_access(handle, primary)))
807 goto exit_journal;
808
809 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
810 (err = reserve_backup_gdb(handle, inode, input)))
811 goto exit_journal;
812 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
813 goto exit_journal;
814
815 /*
816 * OK, now we've set up the new group. Time to make it active.
817 *
818 * Current kernels don't lock all allocations via lock_super(),
819 * so we have to be safe wrt. concurrent accesses the group
820 * data. So we need to be careful to set all of the relevant
821 * group descriptor data etc. *before* we enable the group.
822 *
823 * The key field here is sbi->s_groups_count: as long as
824 * that retains its old value, nobody is going to access the new
825 * group.
826 *
827 * So first we update all the descriptor metadata for the new
828 * group; then we update the total disk blocks count; then we
829 * update the groups count to enable the group; then finally we
830 * update the free space counts so that the system can start
831 * using the new disk blocks.
832 */
833
834 /* Update group descriptor block for new group */
835 gdp = (struct ext4_group_desc *)primary->b_data + gdb_off;
836
837 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
838 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
839 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
840 gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
841 gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
842
843 /*
844 * Make the new blocks and inodes valid next. We do this before
845 * increasing the group count so that once the group is enabled,
846 * all of its blocks and inodes are already valid.
847 *
848 * We always allocate group-by-group, then block-by-block or
849 * inode-by-inode within a group, so enabling these
850 * blocks/inodes before the group is live won't actually let us
851 * allocate the new space yet.
852 */
853 ext4_blocks_count_set(es, ext4_blocks_count(es) +
854 input->blocks_count);
855 es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) +
856 EXT4_INODES_PER_GROUP(sb));
857
858 /*
859 * We need to protect s_groups_count against other CPUs seeing
860 * inconsistent state in the superblock.
861 *
862 * The precise rules we use are:
863 *
864 * * Writers of s_groups_count *must* hold lock_super
865 * AND
866 * * Writers must perform a smp_wmb() after updating all dependent
867 * data and before modifying the groups count
868 *
869 * * Readers must hold lock_super() over the access
870 * OR
871 * * Readers must perform an smp_rmb() after reading the groups count
872 * and before reading any dependent data.
873 *
874 * NB. These rules can be relaxed when checking the group count
875 * while freeing data, as we can only allocate from a block
876 * group after serialising against the group count, and we can
877 * only then free after serialising in turn against that
878 * allocation.
879 */
880 smp_wmb();
881
882 /* Update the global fs size fields */
883 sbi->s_groups_count++;
884
885 ext4_journal_dirty_metadata(handle, primary);
886
887 /* Update the reserved block counts only once the new group is
888 * active. */
889 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
890 input->reserved_blocks);
891
892 /* Update the free space counts */
893 percpu_counter_mod(&sbi->s_freeblocks_counter,
894 input->free_blocks_count);
895 percpu_counter_mod(&sbi->s_freeinodes_counter,
896 EXT4_INODES_PER_GROUP(sb));
897
898 ext4_journal_dirty_metadata(handle, sbi->s_sbh);
899 sb->s_dirt = 1;
900
901exit_journal:
902 unlock_super(sb);
903 if ((err2 = ext4_journal_stop(handle)) && !err)
904 err = err2;
905 if (!err) {
906 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
907 sizeof(struct ext4_super_block));
908 update_backups(sb, primary->b_blocknr, primary->b_data,
909 primary->b_size);
910 }
911exit_put:
912 iput(inode);
913 return err;
914} /* ext4_group_add */
915
916/* Extend the filesystem to the new number of blocks specified. This entry
917 * point is only used to extend the current filesystem to the end of the last
918 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
919 * for emergencies (because it has no dependencies on reserved blocks).
920 *
921 * If we _really_ wanted, we could use default values to call ext4_group_add()
922 * allow the "remount" trick to work for arbitrary resizing, assuming enough
923 * GDT blocks are reserved to grow to the desired size.
924 */
925int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
926 ext4_fsblk_t n_blocks_count)
927{
928 ext4_fsblk_t o_blocks_count;
929 unsigned long o_groups_count;
930 ext4_grpblk_t last;
931 ext4_grpblk_t add;
932 struct buffer_head * bh;
933 handle_t *handle;
934 int err;
935 unsigned long freed_blocks;
936
937 /* We don't need to worry about locking wrt other resizers just
938 * yet: we're going to revalidate es->s_blocks_count after
939 * taking lock_super() below. */
940 o_blocks_count = ext4_blocks_count(es);
941 o_groups_count = EXT4_SB(sb)->s_groups_count;
942
943 if (test_opt(sb, DEBUG))
944 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
945 o_blocks_count, n_blocks_count);
946
947 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
948 return 0;
949
950 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
951 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
952 " too large to resize to %llu blocks safely\n",
953 sb->s_id, n_blocks_count);
954 if (sizeof(sector_t) < 8)
955 ext4_warning(sb, __FUNCTION__,
956 "CONFIG_LBD not enabled\n");
957 return -EINVAL;
958 }
959
960 if (n_blocks_count < o_blocks_count) {
961 ext4_warning(sb, __FUNCTION__,
962 "can't shrink FS - resize aborted");
963 return -EBUSY;
964 }
965
966 /* Handle the remaining blocks in the last group only. */
967 ext4_get_group_no_and_offset(sb, o_blocks_count, NULL, &last);
968
969 if (last == 0) {
970 ext4_warning(sb, __FUNCTION__,
971 "need to use ext2online to resize further");
972 return -EPERM;
973 }
974
975 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
976
977 if (o_blocks_count + add < o_blocks_count) {
978 ext4_warning(sb, __FUNCTION__, "blocks_count overflow");
979 return -EINVAL;
980 }
981
982 if (o_blocks_count + add > n_blocks_count)
983 add = n_blocks_count - o_blocks_count;
984
985 if (o_blocks_count + add < n_blocks_count)
986 ext4_warning(sb, __FUNCTION__,
987 "will only finish group (%llu"
988 " blocks, %u new)",
989 o_blocks_count + add, add);
990
991 /* See if the device is actually as big as what was requested */
992 bh = sb_bread(sb, o_blocks_count + add -1);
993 if (!bh) {
994 ext4_warning(sb, __FUNCTION__,
995 "can't read last block, resize aborted");
996 return -ENOSPC;
997 }
998 brelse(bh);
999
1000 /* We will update the superblock, one block bitmap, and
1001 * one group descriptor via ext4_free_blocks().
1002 */
1003 handle = ext4_journal_start_sb(sb, 3);
1004 if (IS_ERR(handle)) {
1005 err = PTR_ERR(handle);
1006 ext4_warning(sb, __FUNCTION__, "error %d on journal start",err);
1007 goto exit_put;
1008 }
1009
1010 lock_super(sb);
1011 if (o_blocks_count != ext4_blocks_count(es)) {
1012 ext4_warning(sb, __FUNCTION__,
1013 "multiple resizers run on filesystem!");
1014 unlock_super(sb);
1015 err = -EBUSY;
1016 goto exit_put;
1017 }
1018
1019 if ((err = ext4_journal_get_write_access(handle,
1020 EXT4_SB(sb)->s_sbh))) {
1021 ext4_warning(sb, __FUNCTION__,
1022 "error %d on journal write access", err);
1023 unlock_super(sb);
1024 ext4_journal_stop(handle);
1025 goto exit_put;
1026 }
1027 ext4_blocks_count_set(es, o_blocks_count + add);
1028 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1029 sb->s_dirt = 1;
1030 unlock_super(sb);
1031 ext4_debug("freeing blocks %lu through %llu\n", o_blocks_count,
1032 o_blocks_count + add);
1033 ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
1034 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1035 o_blocks_count + add);
1036 if ((err = ext4_journal_stop(handle)))
1037 goto exit_put;
1038 if (test_opt(sb, DEBUG))
1039 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1040 ext4_blocks_count(es));
1041 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1042 sizeof(struct ext4_super_block));
1043exit_put:
1044 return err;
1045} /* ext4_group_extend */
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-02 09:44:40 -0500