ati_pcigart: fix printk format warning
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext4 / resize.c
blob3cfc343c41b53e3eae9e3d635519c653f600acc6
1 /*
2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
9 */
12 #define EXT4FS_DEBUG
14 #include <linux/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
20 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
22 static int verify_group_input(struct super_block *sb,
23 struct ext4_new_group_data *input)
25 struct ext4_sb_info *sbi = EXT4_SB(sb);
26 struct ext4_super_block *es = sbi->s_es;
27 ext4_fsblk_t start = ext4_blocks_count(es);
28 ext4_fsblk_t end = start + input->blocks_count;
29 ext4_group_t group = input->group;
30 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
31 unsigned overhead = ext4_bg_has_super(sb, group) ?
32 (1 + ext4_bg_num_gdb(sb, group) +
33 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
34 ext4_fsblk_t metaend = start + overhead;
35 struct buffer_head *bh = NULL;
36 ext4_grpblk_t free_blocks_count, offset;
37 int err = -EINVAL;
39 input->free_blocks_count = free_blocks_count =
40 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
42 if (test_opt(sb, DEBUG))
43 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
44 "(%d free, %u reserved)\n",
45 ext4_bg_has_super(sb, input->group) ? "normal" :
46 "no-super", input->group, input->blocks_count,
47 free_blocks_count, input->reserved_blocks);
49 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
50 if (group != sbi->s_groups_count)
51 ext4_warning(sb, __func__,
52 "Cannot add at group %u (only %u groups)",
53 input->group, sbi->s_groups_count);
54 else if (offset != 0)
55 ext4_warning(sb, __func__, "Last group not full");
56 else if (input->reserved_blocks > input->blocks_count / 5)
57 ext4_warning(sb, __func__, "Reserved blocks too high (%u)",
58 input->reserved_blocks);
59 else if (free_blocks_count < 0)
60 ext4_warning(sb, __func__, "Bad blocks count %u",
61 input->blocks_count);
62 else if (!(bh = sb_bread(sb, end - 1)))
63 ext4_warning(sb, __func__,
64 "Cannot read last block (%llu)",
65 end - 1);
66 else if (outside(input->block_bitmap, start, end))
67 ext4_warning(sb, __func__,
68 "Block bitmap not in group (block %llu)",
69 (unsigned long long)input->block_bitmap);
70 else if (outside(input->inode_bitmap, start, end))
71 ext4_warning(sb, __func__,
72 "Inode bitmap not in group (block %llu)",
73 (unsigned long long)input->inode_bitmap);
74 else if (outside(input->inode_table, start, end) ||
75 outside(itend - 1, start, end))
76 ext4_warning(sb, __func__,
77 "Inode table not in group (blocks %llu-%llu)",
78 (unsigned long long)input->inode_table, itend - 1);
79 else if (input->inode_bitmap == input->block_bitmap)
80 ext4_warning(sb, __func__,
81 "Block bitmap same as inode bitmap (%llu)",
82 (unsigned long long)input->block_bitmap);
83 else if (inside(input->block_bitmap, input->inode_table, itend))
84 ext4_warning(sb, __func__,
85 "Block bitmap (%llu) in inode table (%llu-%llu)",
86 (unsigned long long)input->block_bitmap,
87 (unsigned long long)input->inode_table, itend - 1);
88 else if (inside(input->inode_bitmap, input->inode_table, itend))
89 ext4_warning(sb, __func__,
90 "Inode bitmap (%llu) in inode table (%llu-%llu)",
91 (unsigned long long)input->inode_bitmap,
92 (unsigned long long)input->inode_table, itend - 1);
93 else if (inside(input->block_bitmap, start, metaend))
94 ext4_warning(sb, __func__,
95 "Block bitmap (%llu) in GDT table"
96 " (%llu-%llu)",
97 (unsigned long long)input->block_bitmap,
98 start, metaend - 1);
99 else if (inside(input->inode_bitmap, start, metaend))
100 ext4_warning(sb, __func__,
101 "Inode bitmap (%llu) in GDT table"
102 " (%llu-%llu)",
103 (unsigned long long)input->inode_bitmap,
104 start, metaend - 1);
105 else if (inside(input->inode_table, start, metaend) ||
106 inside(itend - 1, start, metaend))
107 ext4_warning(sb, __func__,
108 "Inode table (%llu-%llu) overlaps"
109 "GDT table (%llu-%llu)",
110 (unsigned long long)input->inode_table,
111 itend - 1, start, metaend - 1);
112 else
113 err = 0;
114 brelse(bh);
116 return err;
119 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
120 ext4_fsblk_t blk)
122 struct buffer_head *bh;
123 int err;
125 bh = sb_getblk(sb, blk);
126 if (!bh)
127 return ERR_PTR(-EIO);
128 if ((err = ext4_journal_get_write_access(handle, bh))) {
129 brelse(bh);
130 bh = ERR_PTR(err);
131 } else {
132 lock_buffer(bh);
133 memset(bh->b_data, 0, sb->s_blocksize);
134 set_buffer_uptodate(bh);
135 unlock_buffer(bh);
138 return bh;
142 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
143 * If that fails, restart the transaction & regain write access for the
144 * buffer head which is used for block_bitmap modifications.
146 static int extend_or_restart_transaction(handle_t *handle, int thresh,
147 struct buffer_head *bh)
149 int err;
151 if (ext4_handle_has_enough_credits(handle, thresh))
152 return 0;
154 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
155 if (err < 0)
156 return err;
157 if (err) {
158 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
159 return err;
160 if ((err = ext4_journal_get_write_access(handle, bh)))
161 return err;
164 return 0;
168 * Set up the block and inode bitmaps, and the inode table for the new group.
169 * This doesn't need to be part of the main transaction, since we are only
170 * changing blocks outside the actual filesystem. We still do journaling to
171 * ensure the recovery is correct in case of a failure just after resize.
172 * If any part of this fails, we simply abort the resize.
174 static int setup_new_group_blocks(struct super_block *sb,
175 struct ext4_new_group_data *input)
177 struct ext4_sb_info *sbi = EXT4_SB(sb);
178 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
179 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
180 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
181 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
182 struct buffer_head *bh;
183 handle_t *handle;
184 ext4_fsblk_t block;
185 ext4_grpblk_t bit;
186 int i;
187 int err = 0, err2;
189 /* This transaction may be extended/restarted along the way */
190 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
192 if (IS_ERR(handle))
193 return PTR_ERR(handle);
195 mutex_lock(&sbi->s_resize_lock);
196 if (input->group != sbi->s_groups_count) {
197 err = -EBUSY;
198 goto exit_journal;
201 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
202 err = PTR_ERR(bh);
203 goto exit_journal;
206 if (ext4_bg_has_super(sb, input->group)) {
207 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
208 ext4_set_bit(0, bh->b_data);
211 /* Copy all of the GDT blocks into the backup in this group */
212 for (i = 0, bit = 1, block = start + 1;
213 i < gdblocks; i++, block++, bit++) {
214 struct buffer_head *gdb;
216 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
218 if ((err = extend_or_restart_transaction(handle, 1, bh)))
219 goto exit_bh;
221 gdb = sb_getblk(sb, block);
222 if (!gdb) {
223 err = -EIO;
224 goto exit_bh;
226 if ((err = ext4_journal_get_write_access(handle, gdb))) {
227 brelse(gdb);
228 goto exit_bh;
230 lock_buffer(gdb);
231 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
232 set_buffer_uptodate(gdb);
233 unlock_buffer(gdb);
234 ext4_handle_dirty_metadata(handle, NULL, gdb);
235 ext4_set_bit(bit, bh->b_data);
236 brelse(gdb);
239 /* Zero out all of the reserved backup group descriptor table blocks */
240 for (i = 0, bit = gdblocks + 1, block = start + bit;
241 i < reserved_gdb; i++, block++, bit++) {
242 struct buffer_head *gdb;
244 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit);
246 if ((err = extend_or_restart_transaction(handle, 1, bh)))
247 goto exit_bh;
249 if (IS_ERR(gdb = bclean(handle, sb, block))) {
250 err = PTR_ERR(bh);
251 goto exit_bh;
253 ext4_handle_dirty_metadata(handle, NULL, gdb);
254 ext4_set_bit(bit, bh->b_data);
255 brelse(gdb);
257 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
258 input->block_bitmap - start);
259 ext4_set_bit(input->block_bitmap - start, bh->b_data);
260 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
261 input->inode_bitmap - start);
262 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
264 /* Zero out all of the inode table blocks */
265 for (i = 0, block = input->inode_table, bit = block - start;
266 i < sbi->s_itb_per_group; i++, bit++, block++) {
267 struct buffer_head *it;
269 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit);
271 if ((err = extend_or_restart_transaction(handle, 1, bh)))
272 goto exit_bh;
274 if (IS_ERR(it = bclean(handle, sb, block))) {
275 err = PTR_ERR(it);
276 goto exit_bh;
278 ext4_handle_dirty_metadata(handle, NULL, it);
279 brelse(it);
280 ext4_set_bit(bit, bh->b_data);
283 if ((err = extend_or_restart_transaction(handle, 2, bh)))
284 goto exit_bh;
286 mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, bh->b_data);
287 ext4_handle_dirty_metadata(handle, NULL, bh);
288 brelse(bh);
289 /* Mark unused entries in inode bitmap used */
290 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
291 input->inode_bitmap, input->inode_bitmap - start);
292 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
293 err = PTR_ERR(bh);
294 goto exit_journal;
297 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
298 bh->b_data);
299 ext4_handle_dirty_metadata(handle, NULL, bh);
300 exit_bh:
301 brelse(bh);
303 exit_journal:
304 mutex_unlock(&sbi->s_resize_lock);
305 if ((err2 = ext4_journal_stop(handle)) && !err)
306 err = err2;
308 return err;
312 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
313 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
314 * calling this for the first time. In a sparse filesystem it will be the
315 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
316 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
318 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
319 unsigned *five, unsigned *seven)
321 unsigned *min = three;
322 int mult = 3;
323 unsigned ret;
325 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
326 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
327 ret = *min;
328 *min += 1;
329 return ret;
332 if (*five < *min) {
333 min = five;
334 mult = 5;
336 if (*seven < *min) {
337 min = seven;
338 mult = 7;
341 ret = *min;
342 *min *= mult;
344 return ret;
348 * Check that all of the backup GDT blocks are held in the primary GDT block.
349 * It is assumed that they are stored in group order. Returns the number of
350 * groups in current filesystem that have BACKUPS, or -ve error code.
352 static int verify_reserved_gdb(struct super_block *sb,
353 struct buffer_head *primary)
355 const ext4_fsblk_t blk = primary->b_blocknr;
356 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
357 unsigned three = 1;
358 unsigned five = 5;
359 unsigned seven = 7;
360 unsigned grp;
361 __le32 *p = (__le32 *)primary->b_data;
362 int gdbackups = 0;
364 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
365 if (le32_to_cpu(*p++) !=
366 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
367 ext4_warning(sb, __func__,
368 "reserved GDT %llu"
369 " missing grp %d (%llu)",
370 blk, grp,
371 grp *
372 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
373 blk);
374 return -EINVAL;
376 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
377 return -EFBIG;
380 return gdbackups;
384 * Called when we need to bring a reserved group descriptor table block into
385 * use from the resize inode. The primary copy of the new GDT block currently
386 * is an indirect block (under the double indirect block in the resize inode).
387 * The new backup GDT blocks will be stored as leaf blocks in this indirect
388 * block, in group order. Even though we know all the block numbers we need,
389 * we check to ensure that the resize inode has actually reserved these blocks.
391 * Don't need to update the block bitmaps because the blocks are still in use.
393 * We get all of the error cases out of the way, so that we are sure to not
394 * fail once we start modifying the data on disk, because JBD has no rollback.
396 static int add_new_gdb(handle_t *handle, struct inode *inode,
397 struct ext4_new_group_data *input,
398 struct buffer_head **primary)
400 struct super_block *sb = inode->i_sb;
401 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
402 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
403 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
404 struct buffer_head **o_group_desc, **n_group_desc;
405 struct buffer_head *dind;
406 int gdbackups;
407 struct ext4_iloc iloc;
408 __le32 *data;
409 int err;
411 if (test_opt(sb, DEBUG))
412 printk(KERN_DEBUG
413 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
414 gdb_num);
417 * If we are not using the primary superblock/GDT copy don't resize,
418 * because the user tools have no way of handling this. Probably a
419 * bad time to do it anyways.
421 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
422 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
423 ext4_warning(sb, __func__,
424 "won't resize using backup superblock at %llu",
425 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
426 return -EPERM;
429 *primary = sb_bread(sb, gdblock);
430 if (!*primary)
431 return -EIO;
433 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
434 err = gdbackups;
435 goto exit_bh;
438 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
439 dind = sb_bread(sb, le32_to_cpu(*data));
440 if (!dind) {
441 err = -EIO;
442 goto exit_bh;
445 data = (__le32 *)dind->b_data;
446 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
447 ext4_warning(sb, __func__,
448 "new group %u GDT block %llu not reserved",
449 input->group, gdblock);
450 err = -EINVAL;
451 goto exit_dind;
454 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
455 goto exit_dind;
457 if ((err = ext4_journal_get_write_access(handle, *primary)))
458 goto exit_sbh;
460 if ((err = ext4_journal_get_write_access(handle, dind)))
461 goto exit_primary;
463 /* ext4_reserve_inode_write() gets a reference on the iloc */
464 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
465 goto exit_dindj;
467 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
468 GFP_NOFS);
469 if (!n_group_desc) {
470 err = -ENOMEM;
471 ext4_warning(sb, __func__,
472 "not enough memory for %lu groups", gdb_num + 1);
473 goto exit_inode;
477 * Finally, we have all of the possible failures behind us...
479 * Remove new GDT block from inode double-indirect block and clear out
480 * the new GDT block for use (which also "frees" the backup GDT blocks
481 * from the reserved inode). We don't need to change the bitmaps for
482 * these blocks, because they are marked as in-use from being in the
483 * reserved inode, and will become GDT blocks (primary and backup).
485 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
486 ext4_handle_dirty_metadata(handle, NULL, dind);
487 brelse(dind);
488 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
489 ext4_mark_iloc_dirty(handle, inode, &iloc);
490 memset((*primary)->b_data, 0, sb->s_blocksize);
491 ext4_handle_dirty_metadata(handle, NULL, *primary);
493 o_group_desc = EXT4_SB(sb)->s_group_desc;
494 memcpy(n_group_desc, o_group_desc,
495 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
496 n_group_desc[gdb_num] = *primary;
497 EXT4_SB(sb)->s_group_desc = n_group_desc;
498 EXT4_SB(sb)->s_gdb_count++;
499 kfree(o_group_desc);
501 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
502 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
504 return 0;
506 exit_inode:
507 /* ext4_journal_release_buffer(handle, iloc.bh); */
508 brelse(iloc.bh);
509 exit_dindj:
510 /* ext4_journal_release_buffer(handle, dind); */
511 exit_primary:
512 /* ext4_journal_release_buffer(handle, *primary); */
513 exit_sbh:
514 /* ext4_journal_release_buffer(handle, *primary); */
515 exit_dind:
516 brelse(dind);
517 exit_bh:
518 brelse(*primary);
520 ext4_debug("leaving with error %d\n", err);
521 return err;
525 * Called when we are adding a new group which has a backup copy of each of
526 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
527 * We need to add these reserved backup GDT blocks to the resize inode, so
528 * that they are kept for future resizing and not allocated to files.
530 * Each reserved backup GDT block will go into a different indirect block.
531 * The indirect blocks are actually the primary reserved GDT blocks,
532 * so we know in advance what their block numbers are. We only get the
533 * double-indirect block to verify it is pointing to the primary reserved
534 * GDT blocks so we don't overwrite a data block by accident. The reserved
535 * backup GDT blocks are stored in their reserved primary GDT block.
537 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
538 struct ext4_new_group_data *input)
540 struct super_block *sb = inode->i_sb;
541 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
542 struct buffer_head **primary;
543 struct buffer_head *dind;
544 struct ext4_iloc iloc;
545 ext4_fsblk_t blk;
546 __le32 *data, *end;
547 int gdbackups = 0;
548 int res, i;
549 int err;
551 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
552 if (!primary)
553 return -ENOMEM;
555 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
556 dind = sb_bread(sb, le32_to_cpu(*data));
557 if (!dind) {
558 err = -EIO;
559 goto exit_free;
562 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
563 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
564 EXT4_ADDR_PER_BLOCK(sb));
565 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
567 /* Get each reserved primary GDT block and verify it holds backups */
568 for (res = 0; res < reserved_gdb; res++, blk++) {
569 if (le32_to_cpu(*data) != blk) {
570 ext4_warning(sb, __func__,
571 "reserved block %llu"
572 " not at offset %ld",
573 blk,
574 (long)(data - (__le32 *)dind->b_data));
575 err = -EINVAL;
576 goto exit_bh;
578 primary[res] = sb_bread(sb, blk);
579 if (!primary[res]) {
580 err = -EIO;
581 goto exit_bh;
583 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
584 brelse(primary[res]);
585 err = gdbackups;
586 goto exit_bh;
588 if (++data >= end)
589 data = (__le32 *)dind->b_data;
592 for (i = 0; i < reserved_gdb; i++) {
593 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
595 int j;
596 for (j = 0; j < i; j++)
597 ext4_journal_release_buffer(handle, primary[j]);
599 goto exit_bh;
603 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
604 goto exit_bh;
607 * Finally we can add each of the reserved backup GDT blocks from
608 * the new group to its reserved primary GDT block.
610 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
611 for (i = 0; i < reserved_gdb; i++) {
612 int err2;
613 data = (__le32 *)primary[i]->b_data;
614 /* printk("reserving backup %lu[%u] = %lu\n",
615 primary[i]->b_blocknr, gdbackups,
616 blk + primary[i]->b_blocknr); */
617 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
618 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
619 if (!err)
620 err = err2;
622 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
623 ext4_mark_iloc_dirty(handle, inode, &iloc);
625 exit_bh:
626 while (--res >= 0)
627 brelse(primary[res]);
628 brelse(dind);
630 exit_free:
631 kfree(primary);
633 return err;
637 * Update the backup copies of the ext4 metadata. These don't need to be part
638 * of the main resize transaction, because e2fsck will re-write them if there
639 * is a problem (basically only OOM will cause a problem). However, we
640 * _should_ update the backups if possible, in case the primary gets trashed
641 * for some reason and we need to run e2fsck from a backup superblock. The
642 * important part is that the new block and inode counts are in the backup
643 * superblocks, and the location of the new group metadata in the GDT backups.
645 * We do not need take the s_resize_lock for this, because these
646 * blocks are not otherwise touched by the filesystem code when it is
647 * mounted. We don't need to worry about last changing from
648 * sbi->s_groups_count, because the worst that can happen is that we
649 * do not copy the full number of backups at this time. The resize
650 * which changed s_groups_count will backup again.
652 static void update_backups(struct super_block *sb,
653 int blk_off, char *data, int size)
655 struct ext4_sb_info *sbi = EXT4_SB(sb);
656 const ext4_group_t last = sbi->s_groups_count;
657 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
658 unsigned three = 1;
659 unsigned five = 5;
660 unsigned seven = 7;
661 ext4_group_t group;
662 int rest = sb->s_blocksize - size;
663 handle_t *handle;
664 int err = 0, err2;
666 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
667 if (IS_ERR(handle)) {
668 group = 1;
669 err = PTR_ERR(handle);
670 goto exit_err;
673 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
674 struct buffer_head *bh;
676 /* Out of journal space, and can't get more - abort - so sad */
677 if (ext4_handle_valid(handle) &&
678 handle->h_buffer_credits == 0 &&
679 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
680 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
681 break;
683 bh = sb_getblk(sb, group * bpg + blk_off);
684 if (!bh) {
685 err = -EIO;
686 break;
688 ext4_debug("update metadata backup %#04lx\n",
689 (unsigned long)bh->b_blocknr);
690 if ((err = ext4_journal_get_write_access(handle, bh)))
691 break;
692 lock_buffer(bh);
693 memcpy(bh->b_data, data, size);
694 if (rest)
695 memset(bh->b_data + size, 0, rest);
696 set_buffer_uptodate(bh);
697 unlock_buffer(bh);
698 ext4_handle_dirty_metadata(handle, NULL, bh);
699 brelse(bh);
701 if ((err2 = ext4_journal_stop(handle)) && !err)
702 err = err2;
705 * Ugh! Need to have e2fsck write the backup copies. It is too
706 * late to revert the resize, we shouldn't fail just because of
707 * the backup copies (they are only needed in case of corruption).
709 * However, if we got here we have a journal problem too, so we
710 * can't really start a transaction to mark the superblock.
711 * Chicken out and just set the flag on the hope it will be written
712 * to disk, and if not - we will simply wait until next fsck.
714 exit_err:
715 if (err) {
716 ext4_warning(sb, __func__,
717 "can't update backup for group %u (err %d), "
718 "forcing fsck on next reboot", group, err);
719 sbi->s_mount_state &= ~EXT4_VALID_FS;
720 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
721 mark_buffer_dirty(sbi->s_sbh);
725 /* Add group descriptor data to an existing or new group descriptor block.
726 * Ensure we handle all possible error conditions _before_ we start modifying
727 * the filesystem, because we cannot abort the transaction and not have it
728 * write the data to disk.
730 * If we are on a GDT block boundary, we need to get the reserved GDT block.
731 * Otherwise, we may need to add backup GDT blocks for a sparse group.
733 * We only need to hold the superblock lock while we are actually adding
734 * in the new group's counts to the superblock. Prior to that we have
735 * not really "added" the group at all. We re-check that we are still
736 * adding in the last group in case things have changed since verifying.
738 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
740 struct ext4_sb_info *sbi = EXT4_SB(sb);
741 struct ext4_super_block *es = sbi->s_es;
742 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
743 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
744 struct buffer_head *primary = NULL;
745 struct ext4_group_desc *gdp;
746 struct inode *inode = NULL;
747 handle_t *handle;
748 int gdb_off, gdb_num;
749 int err, err2;
751 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
752 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
754 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
755 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
756 ext4_warning(sb, __func__,
757 "Can't resize non-sparse filesystem further");
758 return -EPERM;
761 if (ext4_blocks_count(es) + input->blocks_count <
762 ext4_blocks_count(es)) {
763 ext4_warning(sb, __func__, "blocks_count overflow");
764 return -EINVAL;
767 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
768 le32_to_cpu(es->s_inodes_count)) {
769 ext4_warning(sb, __func__, "inodes_count overflow");
770 return -EINVAL;
773 if (reserved_gdb || gdb_off == 0) {
774 if (!EXT4_HAS_COMPAT_FEATURE(sb,
775 EXT4_FEATURE_COMPAT_RESIZE_INODE)
776 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
777 ext4_warning(sb, __func__,
778 "No reserved GDT blocks, can't resize");
779 return -EPERM;
781 inode = ext4_iget(sb, EXT4_RESIZE_INO);
782 if (IS_ERR(inode)) {
783 ext4_warning(sb, __func__,
784 "Error opening resize inode");
785 return PTR_ERR(inode);
790 if ((err = verify_group_input(sb, input)))
791 goto exit_put;
793 if ((err = setup_new_group_blocks(sb, input)))
794 goto exit_put;
797 * We will always be modifying at least the superblock and a GDT
798 * block. If we are adding a group past the last current GDT block,
799 * we will also modify the inode and the dindirect block. If we
800 * are adding a group with superblock/GDT backups we will also
801 * modify each of the reserved GDT dindirect blocks.
803 handle = ext4_journal_start_sb(sb,
804 ext4_bg_has_super(sb, input->group) ?
805 3 + reserved_gdb : 4);
806 if (IS_ERR(handle)) {
807 err = PTR_ERR(handle);
808 goto exit_put;
811 mutex_lock(&sbi->s_resize_lock);
812 if (input->group != sbi->s_groups_count) {
813 ext4_warning(sb, __func__,
814 "multiple resizers run on filesystem!");
815 err = -EBUSY;
816 goto exit_journal;
819 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
820 goto exit_journal;
823 * We will only either add reserved group blocks to a backup group
824 * or remove reserved blocks for the first group in a new group block.
825 * Doing both would be mean more complex code, and sane people don't
826 * use non-sparse filesystems anymore. This is already checked above.
828 if (gdb_off) {
829 primary = sbi->s_group_desc[gdb_num];
830 if ((err = ext4_journal_get_write_access(handle, primary)))
831 goto exit_journal;
833 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
834 (err = reserve_backup_gdb(handle, inode, input)))
835 goto exit_journal;
836 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
837 goto exit_journal;
840 * OK, now we've set up the new group. Time to make it active.
842 * We do not lock all allocations via s_resize_lock
843 * so we have to be safe wrt. concurrent accesses the group
844 * data. So we need to be careful to set all of the relevant
845 * group descriptor data etc. *before* we enable the group.
847 * The key field here is sbi->s_groups_count: as long as
848 * that retains its old value, nobody is going to access the new
849 * group.
851 * So first we update all the descriptor metadata for the new
852 * group; then we update the total disk blocks count; then we
853 * update the groups count to enable the group; then finally we
854 * update the free space counts so that the system can start
855 * using the new disk blocks.
858 /* Update group descriptor block for new group */
859 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
860 gdb_off * EXT4_DESC_SIZE(sb));
862 memset(gdp, 0, EXT4_DESC_SIZE(sb));
863 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
864 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
865 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
866 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
867 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
868 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
869 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
872 * We can allocate memory for mb_alloc based on the new group
873 * descriptor
875 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
876 if (err)
877 goto exit_journal;
880 * Make the new blocks and inodes valid next. We do this before
881 * increasing the group count so that once the group is enabled,
882 * all of its blocks and inodes are already valid.
884 * We always allocate group-by-group, then block-by-block or
885 * inode-by-inode within a group, so enabling these
886 * blocks/inodes before the group is live won't actually let us
887 * allocate the new space yet.
889 ext4_blocks_count_set(es, ext4_blocks_count(es) +
890 input->blocks_count);
891 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
894 * We need to protect s_groups_count against other CPUs seeing
895 * inconsistent state in the superblock.
897 * The precise rules we use are:
899 * * Writers of s_groups_count *must* hold s_resize_lock
900 * AND
901 * * Writers must perform a smp_wmb() after updating all dependent
902 * data and before modifying the groups count
904 * * Readers must hold s_resize_lock over the access
905 * OR
906 * * Readers must perform an smp_rmb() after reading the groups count
907 * and before reading any dependent data.
909 * NB. These rules can be relaxed when checking the group count
910 * while freeing data, as we can only allocate from a block
911 * group after serialising against the group count, and we can
912 * only then free after serialising in turn against that
913 * allocation.
915 smp_wmb();
917 /* Update the global fs size fields */
918 sbi->s_groups_count++;
920 ext4_handle_dirty_metadata(handle, NULL, primary);
922 /* Update the reserved block counts only once the new group is
923 * active. */
924 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
925 input->reserved_blocks);
927 /* Update the free space counts */
928 percpu_counter_add(&sbi->s_freeblocks_counter,
929 input->free_blocks_count);
930 percpu_counter_add(&sbi->s_freeinodes_counter,
931 EXT4_INODES_PER_GROUP(sb));
933 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
934 ext4_group_t flex_group;
935 flex_group = ext4_flex_group(sbi, input->group);
936 atomic_add(input->free_blocks_count,
937 &sbi->s_flex_groups[flex_group].free_blocks);
938 atomic_add(EXT4_INODES_PER_GROUP(sb),
939 &sbi->s_flex_groups[flex_group].free_inodes);
942 ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
943 sb->s_dirt = 1;
945 exit_journal:
946 mutex_unlock(&sbi->s_resize_lock);
947 if ((err2 = ext4_journal_stop(handle)) && !err)
948 err = err2;
949 if (!err) {
950 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
951 sizeof(struct ext4_super_block));
952 update_backups(sb, primary->b_blocknr, primary->b_data,
953 primary->b_size);
955 exit_put:
956 iput(inode);
957 return err;
958 } /* ext4_group_add */
961 * Extend the filesystem to the new number of blocks specified. This entry
962 * point is only used to extend the current filesystem to the end of the last
963 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
964 * for emergencies (because it has no dependencies on reserved blocks).
966 * If we _really_ wanted, we could use default values to call ext4_group_add()
967 * allow the "remount" trick to work for arbitrary resizing, assuming enough
968 * GDT blocks are reserved to grow to the desired size.
970 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
971 ext4_fsblk_t n_blocks_count)
973 ext4_fsblk_t o_blocks_count;
974 ext4_group_t o_groups_count;
975 ext4_grpblk_t last;
976 ext4_grpblk_t add;
977 struct buffer_head *bh;
978 handle_t *handle;
979 int err;
980 ext4_group_t group;
982 /* We don't need to worry about locking wrt other resizers just
983 * yet: we're going to revalidate es->s_blocks_count after
984 * taking the s_resize_lock below. */
985 o_blocks_count = ext4_blocks_count(es);
986 o_groups_count = EXT4_SB(sb)->s_groups_count;
988 if (test_opt(sb, DEBUG))
989 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
990 o_blocks_count, n_blocks_count);
992 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
993 return 0;
995 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
996 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
997 " too large to resize to %llu blocks safely\n",
998 sb->s_id, n_blocks_count);
999 if (sizeof(sector_t) < 8)
1000 ext4_warning(sb, __func__, "CONFIG_LBDAF not enabled");
1001 return -EINVAL;
1004 if (n_blocks_count < o_blocks_count) {
1005 ext4_warning(sb, __func__,
1006 "can't shrink FS - resize aborted");
1007 return -EBUSY;
1010 /* Handle the remaining blocks in the last group only. */
1011 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1013 if (last == 0) {
1014 ext4_warning(sb, __func__,
1015 "need to use ext2online to resize further");
1016 return -EPERM;
1019 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1021 if (o_blocks_count + add < o_blocks_count) {
1022 ext4_warning(sb, __func__, "blocks_count overflow");
1023 return -EINVAL;
1026 if (o_blocks_count + add > n_blocks_count)
1027 add = n_blocks_count - o_blocks_count;
1029 if (o_blocks_count + add < n_blocks_count)
1030 ext4_warning(sb, __func__,
1031 "will only finish group (%llu"
1032 " blocks, %u new)",
1033 o_blocks_count + add, add);
1035 /* See if the device is actually as big as what was requested */
1036 bh = sb_bread(sb, o_blocks_count + add - 1);
1037 if (!bh) {
1038 ext4_warning(sb, __func__,
1039 "can't read last block, resize aborted");
1040 return -ENOSPC;
1042 brelse(bh);
1044 /* We will update the superblock, one block bitmap, and
1045 * one group descriptor via ext4_free_blocks().
1047 handle = ext4_journal_start_sb(sb, 3);
1048 if (IS_ERR(handle)) {
1049 err = PTR_ERR(handle);
1050 ext4_warning(sb, __func__, "error %d on journal start", err);
1051 goto exit_put;
1054 mutex_lock(&EXT4_SB(sb)->s_resize_lock);
1055 if (o_blocks_count != ext4_blocks_count(es)) {
1056 ext4_warning(sb, __func__,
1057 "multiple resizers run on filesystem!");
1058 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1059 ext4_journal_stop(handle);
1060 err = -EBUSY;
1061 goto exit_put;
1064 if ((err = ext4_journal_get_write_access(handle,
1065 EXT4_SB(sb)->s_sbh))) {
1066 ext4_warning(sb, __func__,
1067 "error %d on journal write access", err);
1068 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1069 ext4_journal_stop(handle);
1070 goto exit_put;
1072 ext4_blocks_count_set(es, o_blocks_count + add);
1073 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1074 sb->s_dirt = 1;
1075 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1076 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1077 o_blocks_count + add);
1078 /* We add the blocks to the bitmap and set the group need init bit */
1079 ext4_add_groupblocks(handle, sb, o_blocks_count, add);
1080 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1081 o_blocks_count + add);
1082 if ((err = ext4_journal_stop(handle)))
1083 goto exit_put;
1085 if (test_opt(sb, DEBUG))
1086 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1087 ext4_blocks_count(es));
1088 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1089 sizeof(struct ext4_super_block));
1090 exit_put:
1091 return err;
1092 } /* ext4_group_extend */