Loading...
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/errno.h>
15#include <linux/slab.h>
16
17#include "ext4_jbd2.h"
18
19int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
26 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
36 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
48#define outside(b, first, last) ((b) < (first) || (b) >= (last))
49#define inside(b, first, last) ((b) >= (first) && (b) < (last))
50
51static int verify_group_input(struct super_block *sb,
52 struct ext4_new_group_data *input)
53{
54 struct ext4_sb_info *sbi = EXT4_SB(sb);
55 struct ext4_super_block *es = sbi->s_es;
56 ext4_fsblk_t start = ext4_blocks_count(es);
57 ext4_fsblk_t end = start + input->blocks_count;
58 ext4_group_t group = input->group;
59 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 unsigned overhead = ext4_bg_has_super(sb, group) ?
61 (1 + ext4_bg_num_gdb(sb, group) +
62 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
63 ext4_fsblk_t metaend = start + overhead;
64 struct buffer_head *bh = NULL;
65 ext4_grpblk_t free_blocks_count, offset;
66 int err = -EINVAL;
67
68 input->free_blocks_count = free_blocks_count =
69 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
70
71 if (test_opt(sb, DEBUG))
72 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
73 "(%d free, %u reserved)\n",
74 ext4_bg_has_super(sb, input->group) ? "normal" :
75 "no-super", input->group, input->blocks_count,
76 free_blocks_count, input->reserved_blocks);
77
78 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
79 if (group != sbi->s_groups_count)
80 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
81 input->group, sbi->s_groups_count);
82 else if (offset != 0)
83 ext4_warning(sb, "Last group not full");
84 else if (input->reserved_blocks > input->blocks_count / 5)
85 ext4_warning(sb, "Reserved blocks too high (%u)",
86 input->reserved_blocks);
87 else if (free_blocks_count < 0)
88 ext4_warning(sb, "Bad blocks count %u",
89 input->blocks_count);
90 else if (!(bh = sb_bread(sb, end - 1)))
91 ext4_warning(sb, "Cannot read last block (%llu)",
92 end - 1);
93 else if (outside(input->block_bitmap, start, end))
94 ext4_warning(sb, "Block bitmap not in group (block %llu)",
95 (unsigned long long)input->block_bitmap);
96 else if (outside(input->inode_bitmap, start, end))
97 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
98 (unsigned long long)input->inode_bitmap);
99 else if (outside(input->inode_table, start, end) ||
100 outside(itend - 1, start, end))
101 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
102 (unsigned long long)input->inode_table, itend - 1);
103 else if (input->inode_bitmap == input->block_bitmap)
104 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
105 (unsigned long long)input->block_bitmap);
106 else if (inside(input->block_bitmap, input->inode_table, itend))
107 ext4_warning(sb, "Block bitmap (%llu) in inode table "
108 "(%llu-%llu)",
109 (unsigned long long)input->block_bitmap,
110 (unsigned long long)input->inode_table, itend - 1);
111 else if (inside(input->inode_bitmap, input->inode_table, itend))
112 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
113 "(%llu-%llu)",
114 (unsigned long long)input->inode_bitmap,
115 (unsigned long long)input->inode_table, itend - 1);
116 else if (inside(input->block_bitmap, start, metaend))
117 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 (unsigned long long)input->block_bitmap,
119 start, metaend - 1);
120 else if (inside(input->inode_bitmap, start, metaend))
121 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 (unsigned long long)input->inode_bitmap,
123 start, metaend - 1);
124 else if (inside(input->inode_table, start, metaend) ||
125 inside(itend - 1, start, metaend))
126 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
127 "(%llu-%llu)",
128 (unsigned long long)input->inode_table,
129 itend - 1, start, metaend - 1);
130 else
131 err = 0;
132 brelse(bh);
133
134 return err;
135}
136
137static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
138 ext4_fsblk_t blk)
139{
140 struct buffer_head *bh;
141 int err;
142
143 bh = sb_getblk(sb, blk);
144 if (!bh)
145 return ERR_PTR(-EIO);
146 if ((err = ext4_journal_get_write_access(handle, bh))) {
147 brelse(bh);
148 bh = ERR_PTR(err);
149 } else {
150 memset(bh->b_data, 0, sb->s_blocksize);
151 set_buffer_uptodate(bh);
152 }
153
154 return bh;
155}
156
157/*
158 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
159 * If that fails, restart the transaction & regain write access for the
160 * buffer head which is used for block_bitmap modifications.
161 */
162static int extend_or_restart_transaction(handle_t *handle, int thresh)
163{
164 int err;
165
166 if (ext4_handle_has_enough_credits(handle, thresh))
167 return 0;
168
169 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
170 if (err < 0)
171 return err;
172 if (err) {
173 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
174 if (err)
175 return err;
176 }
177
178 return 0;
179}
180
181/*
182 * Set up the block and inode bitmaps, and the inode table for the new group.
183 * This doesn't need to be part of the main transaction, since we are only
184 * changing blocks outside the actual filesystem. We still do journaling to
185 * ensure the recovery is correct in case of a failure just after resize.
186 * If any part of this fails, we simply abort the resize.
187 */
188static int setup_new_group_blocks(struct super_block *sb,
189 struct ext4_new_group_data *input)
190{
191 struct ext4_sb_info *sbi = EXT4_SB(sb);
192 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
193 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
194 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
195 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
196 struct buffer_head *bh;
197 handle_t *handle;
198 ext4_fsblk_t block;
199 ext4_grpblk_t bit;
200 int i;
201 int err = 0, err2;
202
203 /* This transaction may be extended/restarted along the way */
204 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
205
206 if (IS_ERR(handle))
207 return PTR_ERR(handle);
208
209 BUG_ON(input->group != sbi->s_groups_count);
210
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;
215
216 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
217 err = extend_or_restart_transaction(handle, 1);
218 if (err)
219 goto exit_journal;
220
221 gdb = sb_getblk(sb, block);
222 if (!gdb) {
223 err = -EIO;
224 goto exit_journal;
225 }
226 if ((err = ext4_journal_get_write_access(handle, gdb))) {
227 brelse(gdb);
228 goto exit_journal;
229 }
230 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
231 set_buffer_uptodate(gdb);
232 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
233 if (unlikely(err)) {
234 brelse(gdb);
235 goto exit_journal;
236 }
237 brelse(gdb);
238 }
239
240 /* Zero out all of the reserved backup group descriptor table blocks */
241 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
242 block, sbi->s_itb_per_group);
243 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
244 GFP_NOFS);
245 if (err)
246 goto exit_journal;
247
248 err = extend_or_restart_transaction(handle, 2);
249 if (err)
250 goto exit_journal;
251
252 bh = bclean(handle, sb, input->block_bitmap);
253 if (IS_ERR(bh)) {
254 err = PTR_ERR(bh);
255 goto exit_journal;
256 }
257
258 if (ext4_bg_has_super(sb, input->group)) {
259 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
260 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
261 }
262
263 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
264 input->block_bitmap - start);
265 ext4_set_bit(input->block_bitmap - start, bh->b_data);
266 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
267 input->inode_bitmap - start);
268 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
269
270 /* Zero out all of the inode table blocks */
271 block = input->inode_table;
272 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
273 block, sbi->s_itb_per_group);
274 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
275 if (err)
276 goto exit_bh;
277 ext4_set_bits(bh->b_data, input->inode_table - start,
278 sbi->s_itb_per_group);
279
280
281 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
282 bh->b_data);
283 err = ext4_handle_dirty_metadata(handle, NULL, bh);
284 if (unlikely(err)) {
285 ext4_std_error(sb, err);
286 goto exit_bh;
287 }
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;
295 }
296
297 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
298 bh->b_data);
299 err = ext4_handle_dirty_metadata(handle, NULL, bh);
300 if (unlikely(err))
301 ext4_std_error(sb, err);
302exit_bh:
303 brelse(bh);
304
305exit_journal:
306 if ((err2 = ext4_journal_stop(handle)) && !err)
307 err = err2;
308
309 return err;
310}
311
312/*
313 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
314 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
315 * calling this for the first time. In a sparse filesystem it will be the
316 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
317 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
318 */
319static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
320 unsigned *five, unsigned *seven)
321{
322 unsigned *min = three;
323 int mult = 3;
324 unsigned ret;
325
326 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
327 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
328 ret = *min;
329 *min += 1;
330 return ret;
331 }
332
333 if (*five < *min) {
334 min = five;
335 mult = 5;
336 }
337 if (*seven < *min) {
338 min = seven;
339 mult = 7;
340 }
341
342 ret = *min;
343 *min *= mult;
344
345 return ret;
346}
347
348/*
349 * Check that all of the backup GDT blocks are held in the primary GDT block.
350 * It is assumed that they are stored in group order. Returns the number of
351 * groups in current filesystem that have BACKUPS, or -ve error code.
352 */
353static int verify_reserved_gdb(struct super_block *sb,
354 struct buffer_head *primary)
355{
356 const ext4_fsblk_t blk = primary->b_blocknr;
357 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
358 unsigned three = 1;
359 unsigned five = 5;
360 unsigned seven = 7;
361 unsigned grp;
362 __le32 *p = (__le32 *)primary->b_data;
363 int gdbackups = 0;
364
365 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
366 if (le32_to_cpu(*p++) !=
367 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
368 ext4_warning(sb, "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;
375 }
376 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
377 return -EFBIG;
378 }
379
380 return gdbackups;
381}
382
383/*
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.
390 *
391 * Don't need to update the block bitmaps because the blocks are still in use.
392 *
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.
395 */
396static int add_new_gdb(handle_t *handle, struct inode *inode,
397 ext4_group_t group)
398{
399 struct super_block *sb = inode->i_sb;
400 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
401 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
402 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
403 struct buffer_head **o_group_desc, **n_group_desc;
404 struct buffer_head *dind;
405 struct buffer_head *gdb_bh;
406 int gdbackups;
407 struct ext4_iloc iloc;
408 __le32 *data;
409 int err;
410
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);
415
416 /*
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.
420 */
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, "won't resize using backup superblock at %llu",
424 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
425 return -EPERM;
426 }
427
428 gdb_bh = sb_bread(sb, gdblock);
429 if (!gdb_bh)
430 return -EIO;
431
432 gdbackups = verify_reserved_gdb(sb, gdb_bh);
433 if (gdbackups < 0) {
434 err = gdbackups;
435 goto exit_bh;
436 }
437
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;
443 }
444
445 data = (__le32 *)dind->b_data;
446 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
447 ext4_warning(sb, "new group %u GDT block %llu not reserved",
448 group, gdblock);
449 err = -EINVAL;
450 goto exit_dind;
451 }
452
453 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
454 if (unlikely(err))
455 goto exit_dind;
456
457 err = ext4_journal_get_write_access(handle, gdb_bh);
458 if (unlikely(err))
459 goto exit_sbh;
460
461 err = ext4_journal_get_write_access(handle, dind);
462 if (unlikely(err))
463 ext4_std_error(sb, err);
464
465 /* ext4_reserve_inode_write() gets a reference on the iloc */
466 err = ext4_reserve_inode_write(handle, inode, &iloc);
467 if (unlikely(err))
468 goto exit_dindj;
469
470 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
471 sizeof(struct buffer_head *),
472 GFP_NOFS);
473 if (!n_group_desc) {
474 err = -ENOMEM;
475 ext4_warning(sb, "not enough memory for %lu groups",
476 gdb_num + 1);
477 goto exit_inode;
478 }
479
480 /*
481 * Finally, we have all of the possible failures behind us...
482 *
483 * Remove new GDT block from inode double-indirect block and clear out
484 * the new GDT block for use (which also "frees" the backup GDT blocks
485 * from the reserved inode). We don't need to change the bitmaps for
486 * these blocks, because they are marked as in-use from being in the
487 * reserved inode, and will become GDT blocks (primary and backup).
488 */
489 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
490 err = ext4_handle_dirty_metadata(handle, NULL, dind);
491 if (unlikely(err)) {
492 ext4_std_error(sb, err);
493 goto exit_inode;
494 }
495 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
496 ext4_mark_iloc_dirty(handle, inode, &iloc);
497 memset(gdb_bh->b_data, 0, sb->s_blocksize);
498 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
499 if (unlikely(err)) {
500 ext4_std_error(sb, err);
501 goto exit_inode;
502 }
503 brelse(dind);
504
505 o_group_desc = EXT4_SB(sb)->s_group_desc;
506 memcpy(n_group_desc, o_group_desc,
507 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
508 n_group_desc[gdb_num] = gdb_bh;
509 EXT4_SB(sb)->s_group_desc = n_group_desc;
510 EXT4_SB(sb)->s_gdb_count++;
511 ext4_kvfree(o_group_desc);
512
513 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
514 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
515 if (err)
516 ext4_std_error(sb, err);
517
518 return err;
519
520exit_inode:
521 ext4_kvfree(n_group_desc);
522 /* ext4_handle_release_buffer(handle, iloc.bh); */
523 brelse(iloc.bh);
524exit_dindj:
525 /* ext4_handle_release_buffer(handle, dind); */
526exit_sbh:
527 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
528exit_dind:
529 brelse(dind);
530exit_bh:
531 brelse(gdb_bh);
532
533 ext4_debug("leaving with error %d\n", err);
534 return err;
535}
536
537/*
538 * Called when we are adding a new group which has a backup copy of each of
539 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
540 * We need to add these reserved backup GDT blocks to the resize inode, so
541 * that they are kept for future resizing and not allocated to files.
542 *
543 * Each reserved backup GDT block will go into a different indirect block.
544 * The indirect blocks are actually the primary reserved GDT blocks,
545 * so we know in advance what their block numbers are. We only get the
546 * double-indirect block to verify it is pointing to the primary reserved
547 * GDT blocks so we don't overwrite a data block by accident. The reserved
548 * backup GDT blocks are stored in their reserved primary GDT block.
549 */
550static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
551 ext4_group_t group)
552{
553 struct super_block *sb = inode->i_sb;
554 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
555 struct buffer_head **primary;
556 struct buffer_head *dind;
557 struct ext4_iloc iloc;
558 ext4_fsblk_t blk;
559 __le32 *data, *end;
560 int gdbackups = 0;
561 int res, i;
562 int err;
563
564 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
565 if (!primary)
566 return -ENOMEM;
567
568 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
569 dind = sb_bread(sb, le32_to_cpu(*data));
570 if (!dind) {
571 err = -EIO;
572 goto exit_free;
573 }
574
575 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
576 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
577 EXT4_ADDR_PER_BLOCK(sb));
578 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
579
580 /* Get each reserved primary GDT block and verify it holds backups */
581 for (res = 0; res < reserved_gdb; res++, blk++) {
582 if (le32_to_cpu(*data) != blk) {
583 ext4_warning(sb, "reserved block %llu"
584 " not at offset %ld",
585 blk,
586 (long)(data - (__le32 *)dind->b_data));
587 err = -EINVAL;
588 goto exit_bh;
589 }
590 primary[res] = sb_bread(sb, blk);
591 if (!primary[res]) {
592 err = -EIO;
593 goto exit_bh;
594 }
595 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
596 brelse(primary[res]);
597 err = gdbackups;
598 goto exit_bh;
599 }
600 if (++data >= end)
601 data = (__le32 *)dind->b_data;
602 }
603
604 for (i = 0; i < reserved_gdb; i++) {
605 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
606 /*
607 int j;
608 for (j = 0; j < i; j++)
609 ext4_handle_release_buffer(handle, primary[j]);
610 */
611 goto exit_bh;
612 }
613 }
614
615 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
616 goto exit_bh;
617
618 /*
619 * Finally we can add each of the reserved backup GDT blocks from
620 * the new group to its reserved primary GDT block.
621 */
622 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
623 for (i = 0; i < reserved_gdb; i++) {
624 int err2;
625 data = (__le32 *)primary[i]->b_data;
626 /* printk("reserving backup %lu[%u] = %lu\n",
627 primary[i]->b_blocknr, gdbackups,
628 blk + primary[i]->b_blocknr); */
629 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
630 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
631 if (!err)
632 err = err2;
633 }
634 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
635 ext4_mark_iloc_dirty(handle, inode, &iloc);
636
637exit_bh:
638 while (--res >= 0)
639 brelse(primary[res]);
640 brelse(dind);
641
642exit_free:
643 kfree(primary);
644
645 return err;
646}
647
648/*
649 * Update the backup copies of the ext4 metadata. These don't need to be part
650 * of the main resize transaction, because e2fsck will re-write them if there
651 * is a problem (basically only OOM will cause a problem). However, we
652 * _should_ update the backups if possible, in case the primary gets trashed
653 * for some reason and we need to run e2fsck from a backup superblock. The
654 * important part is that the new block and inode counts are in the backup
655 * superblocks, and the location of the new group metadata in the GDT backups.
656 *
657 * We do not need take the s_resize_lock for this, because these
658 * blocks are not otherwise touched by the filesystem code when it is
659 * mounted. We don't need to worry about last changing from
660 * sbi->s_groups_count, because the worst that can happen is that we
661 * do not copy the full number of backups at this time. The resize
662 * which changed s_groups_count will backup again.
663 */
664static void update_backups(struct super_block *sb,
665 int blk_off, char *data, int size)
666{
667 struct ext4_sb_info *sbi = EXT4_SB(sb);
668 const ext4_group_t last = sbi->s_groups_count;
669 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
670 unsigned three = 1;
671 unsigned five = 5;
672 unsigned seven = 7;
673 ext4_group_t group;
674 int rest = sb->s_blocksize - size;
675 handle_t *handle;
676 int err = 0, err2;
677
678 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
679 if (IS_ERR(handle)) {
680 group = 1;
681 err = PTR_ERR(handle);
682 goto exit_err;
683 }
684
685 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
686 struct buffer_head *bh;
687
688 /* Out of journal space, and can't get more - abort - so sad */
689 if (ext4_handle_valid(handle) &&
690 handle->h_buffer_credits == 0 &&
691 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
692 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
693 break;
694
695 bh = sb_getblk(sb, group * bpg + blk_off);
696 if (!bh) {
697 err = -EIO;
698 break;
699 }
700 ext4_debug("update metadata backup %#04lx\n",
701 (unsigned long)bh->b_blocknr);
702 if ((err = ext4_journal_get_write_access(handle, bh)))
703 break;
704 lock_buffer(bh);
705 memcpy(bh->b_data, data, size);
706 if (rest)
707 memset(bh->b_data + size, 0, rest);
708 set_buffer_uptodate(bh);
709 unlock_buffer(bh);
710 err = ext4_handle_dirty_metadata(handle, NULL, bh);
711 if (unlikely(err))
712 ext4_std_error(sb, err);
713 brelse(bh);
714 }
715 if ((err2 = ext4_journal_stop(handle)) && !err)
716 err = err2;
717
718 /*
719 * Ugh! Need to have e2fsck write the backup copies. It is too
720 * late to revert the resize, we shouldn't fail just because of
721 * the backup copies (they are only needed in case of corruption).
722 *
723 * However, if we got here we have a journal problem too, so we
724 * can't really start a transaction to mark the superblock.
725 * Chicken out and just set the flag on the hope it will be written
726 * to disk, and if not - we will simply wait until next fsck.
727 */
728exit_err:
729 if (err) {
730 ext4_warning(sb, "can't update backup for group %u (err %d), "
731 "forcing fsck on next reboot", group, err);
732 sbi->s_mount_state &= ~EXT4_VALID_FS;
733 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
734 mark_buffer_dirty(sbi->s_sbh);
735 }
736}
737
738/* Add group descriptor data to an existing or new group descriptor block.
739 * Ensure we handle all possible error conditions _before_ we start modifying
740 * the filesystem, because we cannot abort the transaction and not have it
741 * write the data to disk.
742 *
743 * If we are on a GDT block boundary, we need to get the reserved GDT block.
744 * Otherwise, we may need to add backup GDT blocks for a sparse group.
745 *
746 * We only need to hold the superblock lock while we are actually adding
747 * in the new group's counts to the superblock. Prior to that we have
748 * not really "added" the group at all. We re-check that we are still
749 * adding in the last group in case things have changed since verifying.
750 */
751int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
752{
753 struct ext4_sb_info *sbi = EXT4_SB(sb);
754 struct ext4_super_block *es = sbi->s_es;
755 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
756 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
757 struct buffer_head *primary = NULL;
758 struct ext4_group_desc *gdp;
759 struct inode *inode = NULL;
760 handle_t *handle;
761 int gdb_off, gdb_num;
762 int err, err2;
763
764 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
765 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
766
767 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
768 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
769 ext4_warning(sb, "Can't resize non-sparse filesystem further");
770 return -EPERM;
771 }
772
773 if (ext4_blocks_count(es) + input->blocks_count <
774 ext4_blocks_count(es)) {
775 ext4_warning(sb, "blocks_count overflow");
776 return -EINVAL;
777 }
778
779 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
780 le32_to_cpu(es->s_inodes_count)) {
781 ext4_warning(sb, "inodes_count overflow");
782 return -EINVAL;
783 }
784
785 if (reserved_gdb || gdb_off == 0) {
786 if (!EXT4_HAS_COMPAT_FEATURE(sb,
787 EXT4_FEATURE_COMPAT_RESIZE_INODE)
788 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
789 ext4_warning(sb,
790 "No reserved GDT blocks, can't resize");
791 return -EPERM;
792 }
793 inode = ext4_iget(sb, EXT4_RESIZE_INO);
794 if (IS_ERR(inode)) {
795 ext4_warning(sb, "Error opening resize inode");
796 return PTR_ERR(inode);
797 }
798 }
799
800
801 if ((err = verify_group_input(sb, input)))
802 goto exit_put;
803
804 if ((err = setup_new_group_blocks(sb, input)))
805 goto exit_put;
806
807 /*
808 * We will always be modifying at least the superblock and a GDT
809 * block. If we are adding a group past the last current GDT block,
810 * we will also modify the inode and the dindirect block. If we
811 * are adding a group with superblock/GDT backups we will also
812 * modify each of the reserved GDT dindirect blocks.
813 */
814 handle = ext4_journal_start_sb(sb,
815 ext4_bg_has_super(sb, input->group) ?
816 3 + reserved_gdb : 4);
817 if (IS_ERR(handle)) {
818 err = PTR_ERR(handle);
819 goto exit_put;
820 }
821
822 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
823 goto exit_journal;
824
825 /*
826 * We will only either add reserved group blocks to a backup group
827 * or remove reserved blocks for the first group in a new group block.
828 * Doing both would be mean more complex code, and sane people don't
829 * use non-sparse filesystems anymore. This is already checked above.
830 */
831 if (gdb_off) {
832 primary = sbi->s_group_desc[gdb_num];
833 if ((err = ext4_journal_get_write_access(handle, primary)))
834 goto exit_journal;
835
836 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
837 err = reserve_backup_gdb(handle, inode, input->group);
838 if (err)
839 goto exit_journal;
840 }
841 } else {
842 /*
843 * Note that we can access new group descriptor block safely
844 * only if add_new_gdb() succeeds.
845 */
846 err = add_new_gdb(handle, inode, input->group);
847 if (err)
848 goto exit_journal;
849 primary = sbi->s_group_desc[gdb_num];
850 }
851
852 /*
853 * OK, now we've set up the new group. Time to make it active.
854 *
855 * so we have to be safe wrt. concurrent accesses the group
856 * data. So we need to be careful to set all of the relevant
857 * group descriptor data etc. *before* we enable the group.
858 *
859 * The key field here is sbi->s_groups_count: as long as
860 * that retains its old value, nobody is going to access the new
861 * group.
862 *
863 * So first we update all the descriptor metadata for the new
864 * group; then we update the total disk blocks count; then we
865 * update the groups count to enable the group; then finally we
866 * update the free space counts so that the system can start
867 * using the new disk blocks.
868 */
869
870 /* Update group descriptor block for new group */
871 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
872 gdb_off * EXT4_DESC_SIZE(sb));
873
874 memset(gdp, 0, EXT4_DESC_SIZE(sb));
875 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
876 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
877 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
878 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
879 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
880 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
881 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
882
883 /*
884 * We can allocate memory for mb_alloc based on the new group
885 * descriptor
886 */
887 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
888 if (err)
889 goto exit_journal;
890
891 /*
892 * Make the new blocks and inodes valid next. We do this before
893 * increasing the group count so that once the group is enabled,
894 * all of its blocks and inodes are already valid.
895 *
896 * We always allocate group-by-group, then block-by-block or
897 * inode-by-inode within a group, so enabling these
898 * blocks/inodes before the group is live won't actually let us
899 * allocate the new space yet.
900 */
901 ext4_blocks_count_set(es, ext4_blocks_count(es) +
902 input->blocks_count);
903 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
904
905 /*
906 * We need to protect s_groups_count against other CPUs seeing
907 * inconsistent state in the superblock.
908 *
909 * The precise rules we use are:
910 *
911 * * Writers must perform a smp_wmb() after updating all dependent
912 * data and before modifying the groups count
913 *
914 * * Readers must perform an smp_rmb() after reading the groups count
915 * and before reading any dependent data.
916 *
917 * NB. These rules can be relaxed when checking the group count
918 * while freeing data, as we can only allocate from a block
919 * group after serialising against the group count, and we can
920 * only then free after serialising in turn against that
921 * allocation.
922 */
923 smp_wmb();
924
925 /* Update the global fs size fields */
926 sbi->s_groups_count++;
927
928 err = ext4_handle_dirty_metadata(handle, NULL, primary);
929 if (unlikely(err)) {
930 ext4_std_error(sb, err);
931 goto exit_journal;
932 }
933
934 /* Update the reserved block counts only once the new group is
935 * active. */
936 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
937 input->reserved_blocks);
938
939 /* Update the free space counts */
940 percpu_counter_add(&sbi->s_freeblocks_counter,
941 input->free_blocks_count);
942 percpu_counter_add(&sbi->s_freeinodes_counter,
943 EXT4_INODES_PER_GROUP(sb));
944
945 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
946 sbi->s_log_groups_per_flex) {
947 ext4_group_t flex_group;
948 flex_group = ext4_flex_group(sbi, input->group);
949 atomic_add(input->free_blocks_count,
950 &sbi->s_flex_groups[flex_group].free_blocks);
951 atomic_add(EXT4_INODES_PER_GROUP(sb),
952 &sbi->s_flex_groups[flex_group].free_inodes);
953 }
954
955 ext4_handle_dirty_super(handle, sb);
956
957exit_journal:
958 if ((err2 = ext4_journal_stop(handle)) && !err)
959 err = err2;
960 if (!err && primary) {
961 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
962 sizeof(struct ext4_super_block));
963 update_backups(sb, primary->b_blocknr, primary->b_data,
964 primary->b_size);
965 }
966exit_put:
967 iput(inode);
968 return err;
969} /* ext4_group_add */
970
971/*
972 * Extend the filesystem to the new number of blocks specified. This entry
973 * point is only used to extend the current filesystem to the end of the last
974 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
975 * for emergencies (because it has no dependencies on reserved blocks).
976 *
977 * If we _really_ wanted, we could use default values to call ext4_group_add()
978 * allow the "remount" trick to work for arbitrary resizing, assuming enough
979 * GDT blocks are reserved to grow to the desired size.
980 */
981int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
982 ext4_fsblk_t n_blocks_count)
983{
984 ext4_fsblk_t o_blocks_count;
985 ext4_grpblk_t last;
986 ext4_grpblk_t add;
987 struct buffer_head *bh;
988 handle_t *handle;
989 int err, err2;
990 ext4_group_t group;
991
992 o_blocks_count = ext4_blocks_count(es);
993
994 if (test_opt(sb, DEBUG))
995 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
996 o_blocks_count, n_blocks_count);
997
998 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
999 return 0;
1000
1001 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1002 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1003 " too large to resize to %llu blocks safely\n",
1004 sb->s_id, n_blocks_count);
1005 if (sizeof(sector_t) < 8)
1006 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1007 return -EINVAL;
1008 }
1009
1010 if (n_blocks_count < o_blocks_count) {
1011 ext4_warning(sb, "can't shrink FS - resize aborted");
1012 return -EINVAL;
1013 }
1014
1015 /* Handle the remaining blocks in the last group only. */
1016 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1017
1018 if (last == 0) {
1019 ext4_warning(sb, "need to use ext2online to resize further");
1020 return -EPERM;
1021 }
1022
1023 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1024
1025 if (o_blocks_count + add < o_blocks_count) {
1026 ext4_warning(sb, "blocks_count overflow");
1027 return -EINVAL;
1028 }
1029
1030 if (o_blocks_count + add > n_blocks_count)
1031 add = n_blocks_count - o_blocks_count;
1032
1033 if (o_blocks_count + add < n_blocks_count)
1034 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1035 o_blocks_count + add, add);
1036
1037 /* See if the device is actually as big as what was requested */
1038 bh = sb_bread(sb, o_blocks_count + add - 1);
1039 if (!bh) {
1040 ext4_warning(sb, "can't read last block, resize aborted");
1041 return -ENOSPC;
1042 }
1043 brelse(bh);
1044
1045 /* We will update the superblock, one block bitmap, and
1046 * one group descriptor via ext4_free_blocks().
1047 */
1048 handle = ext4_journal_start_sb(sb, 3);
1049 if (IS_ERR(handle)) {
1050 err = PTR_ERR(handle);
1051 ext4_warning(sb, "error %d on journal start", err);
1052 goto exit_put;
1053 }
1054
1055 if ((err = ext4_journal_get_write_access(handle,
1056 EXT4_SB(sb)->s_sbh))) {
1057 ext4_warning(sb, "error %d on journal write access", err);
1058 ext4_journal_stop(handle);
1059 goto exit_put;
1060 }
1061 ext4_blocks_count_set(es, o_blocks_count + add);
1062 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1063 o_blocks_count + add);
1064 /* We add the blocks to the bitmap and set the group need init bit */
1065 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1066 ext4_handle_dirty_super(handle, sb);
1067 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1068 o_blocks_count + add);
1069 err2 = ext4_journal_stop(handle);
1070 if (!err && err2)
1071 err = err2;
1072
1073 if (err)
1074 goto exit_put;
1075
1076 if (test_opt(sb, DEBUG))
1077 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1078 ext4_blocks_count(es));
1079 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1080 sizeof(struct ext4_super_block));
1081exit_put:
1082 return err;
1083} /* ext4_group_extend */
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/errno.h>
15#include <linux/slab.h>
16
17#include "ext4_jbd2.h"
18
19int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
26 /*
27 * If we are not using the primary superblock/GDT copy don't resize,
28 * because the user tools have no way of handling this. Probably a
29 * bad time to do it anyways.
30 */
31 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
32 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
33 ext4_warning(sb, "won't resize using backup superblock at %llu",
34 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
35 return -EPERM;
36 }
37
38 /*
39 * We are not allowed to do online-resizing on a filesystem mounted
40 * with error, because it can destroy the filesystem easily.
41 */
42 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
43 ext4_warning(sb, "There are errors in the filesystem, "
44 "so online resizing is not allowed\n");
45 return -EPERM;
46 }
47
48 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
49 ret = -EBUSY;
50
51 return ret;
52}
53
54void ext4_resize_end(struct super_block *sb)
55{
56 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
57 smp_mb__after_atomic();
58}
59
60static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
61 ext4_group_t group) {
62 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
63 EXT4_DESC_PER_BLOCK_BITS(sb);
64}
65
66static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
67 ext4_group_t group) {
68 group = ext4_meta_bg_first_group(sb, group);
69 return ext4_group_first_block_no(sb, group);
70}
71
72static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
73 ext4_group_t group) {
74 ext4_grpblk_t overhead;
75 overhead = ext4_bg_num_gdb(sb, group);
76 if (ext4_bg_has_super(sb, group))
77 overhead += 1 +
78 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
79 return overhead;
80}
81
82#define outside(b, first, last) ((b) < (first) || (b) >= (last))
83#define inside(b, first, last) ((b) >= (first) && (b) < (last))
84
85static int verify_group_input(struct super_block *sb,
86 struct ext4_new_group_data *input)
87{
88 struct ext4_sb_info *sbi = EXT4_SB(sb);
89 struct ext4_super_block *es = sbi->s_es;
90 ext4_fsblk_t start = ext4_blocks_count(es);
91 ext4_fsblk_t end = start + input->blocks_count;
92 ext4_group_t group = input->group;
93 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
94 unsigned overhead;
95 ext4_fsblk_t metaend;
96 struct buffer_head *bh = NULL;
97 ext4_grpblk_t free_blocks_count, offset;
98 int err = -EINVAL;
99
100 if (group != sbi->s_groups_count) {
101 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
102 input->group, sbi->s_groups_count);
103 return -EINVAL;
104 }
105
106 overhead = ext4_group_overhead_blocks(sb, group);
107 metaend = start + overhead;
108 input->free_blocks_count = free_blocks_count =
109 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
110
111 if (test_opt(sb, DEBUG))
112 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
113 "(%d free, %u reserved)\n",
114 ext4_bg_has_super(sb, input->group) ? "normal" :
115 "no-super", input->group, input->blocks_count,
116 free_blocks_count, input->reserved_blocks);
117
118 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
119 if (offset != 0)
120 ext4_warning(sb, "Last group not full");
121 else if (input->reserved_blocks > input->blocks_count / 5)
122 ext4_warning(sb, "Reserved blocks too high (%u)",
123 input->reserved_blocks);
124 else if (free_blocks_count < 0)
125 ext4_warning(sb, "Bad blocks count %u",
126 input->blocks_count);
127 else if (!(bh = sb_bread(sb, end - 1)))
128 ext4_warning(sb, "Cannot read last block (%llu)",
129 end - 1);
130 else if (outside(input->block_bitmap, start, end))
131 ext4_warning(sb, "Block bitmap not in group (block %llu)",
132 (unsigned long long)input->block_bitmap);
133 else if (outside(input->inode_bitmap, start, end))
134 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
135 (unsigned long long)input->inode_bitmap);
136 else if (outside(input->inode_table, start, end) ||
137 outside(itend - 1, start, end))
138 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
139 (unsigned long long)input->inode_table, itend - 1);
140 else if (input->inode_bitmap == input->block_bitmap)
141 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
142 (unsigned long long)input->block_bitmap);
143 else if (inside(input->block_bitmap, input->inode_table, itend))
144 ext4_warning(sb, "Block bitmap (%llu) in inode table "
145 "(%llu-%llu)",
146 (unsigned long long)input->block_bitmap,
147 (unsigned long long)input->inode_table, itend - 1);
148 else if (inside(input->inode_bitmap, input->inode_table, itend))
149 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
150 "(%llu-%llu)",
151 (unsigned long long)input->inode_bitmap,
152 (unsigned long long)input->inode_table, itend - 1);
153 else if (inside(input->block_bitmap, start, metaend))
154 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
155 (unsigned long long)input->block_bitmap,
156 start, metaend - 1);
157 else if (inside(input->inode_bitmap, start, metaend))
158 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
159 (unsigned long long)input->inode_bitmap,
160 start, metaend - 1);
161 else if (inside(input->inode_table, start, metaend) ||
162 inside(itend - 1, start, metaend))
163 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
164 "(%llu-%llu)",
165 (unsigned long long)input->inode_table,
166 itend - 1, start, metaend - 1);
167 else
168 err = 0;
169 brelse(bh);
170
171 return err;
172}
173
174/*
175 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
176 * group each time.
177 */
178struct ext4_new_flex_group_data {
179 struct ext4_new_group_data *groups; /* new_group_data for groups
180 in the flex group */
181 __u16 *bg_flags; /* block group flags of groups
182 in @groups */
183 ext4_group_t count; /* number of groups in @groups
184 */
185};
186
187/*
188 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
189 * @flexbg_size.
190 *
191 * Returns NULL on failure otherwise address of the allocated structure.
192 */
193static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
194{
195 struct ext4_new_flex_group_data *flex_gd;
196
197 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
198 if (flex_gd == NULL)
199 goto out3;
200
201 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
202 goto out2;
203 flex_gd->count = flexbg_size;
204
205 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
206 flexbg_size, GFP_NOFS);
207 if (flex_gd->groups == NULL)
208 goto out2;
209
210 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
211 if (flex_gd->bg_flags == NULL)
212 goto out1;
213
214 return flex_gd;
215
216out1:
217 kfree(flex_gd->groups);
218out2:
219 kfree(flex_gd);
220out3:
221 return NULL;
222}
223
224static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
225{
226 kfree(flex_gd->bg_flags);
227 kfree(flex_gd->groups);
228 kfree(flex_gd);
229}
230
231/*
232 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
233 * and inode tables for a flex group.
234 *
235 * This function is used by 64bit-resize. Note that this function allocates
236 * group tables from the 1st group of groups contained by @flexgd, which may
237 * be a partial of a flex group.
238 *
239 * @sb: super block of fs to which the groups belongs
240 *
241 * Returns 0 on a successful allocation of the metadata blocks in the
242 * block group.
243 */
244static int ext4_alloc_group_tables(struct super_block *sb,
245 struct ext4_new_flex_group_data *flex_gd,
246 int flexbg_size)
247{
248 struct ext4_new_group_data *group_data = flex_gd->groups;
249 ext4_fsblk_t start_blk;
250 ext4_fsblk_t last_blk;
251 ext4_group_t src_group;
252 ext4_group_t bb_index = 0;
253 ext4_group_t ib_index = 0;
254 ext4_group_t it_index = 0;
255 ext4_group_t group;
256 ext4_group_t last_group;
257 unsigned overhead;
258 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
259
260 BUG_ON(flex_gd->count == 0 || group_data == NULL);
261
262 src_group = group_data[0].group;
263 last_group = src_group + flex_gd->count - 1;
264
265 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
266 (last_group & ~(flexbg_size - 1))));
267next_group:
268 group = group_data[0].group;
269 if (src_group >= group_data[0].group + flex_gd->count)
270 return -ENOSPC;
271 start_blk = ext4_group_first_block_no(sb, src_group);
272 last_blk = start_blk + group_data[src_group - group].blocks_count;
273
274 overhead = ext4_group_overhead_blocks(sb, src_group);
275
276 start_blk += overhead;
277
278 /* We collect contiguous blocks as much as possible. */
279 src_group++;
280 for (; src_group <= last_group; src_group++) {
281 overhead = ext4_group_overhead_blocks(sb, src_group);
282 if (overhead == 0)
283 last_blk += group_data[src_group - group].blocks_count;
284 else
285 break;
286 }
287
288 /* Allocate block bitmaps */
289 for (; bb_index < flex_gd->count; bb_index++) {
290 if (start_blk >= last_blk)
291 goto next_group;
292 group_data[bb_index].block_bitmap = start_blk++;
293 group = ext4_get_group_number(sb, start_blk - 1);
294 group -= group_data[0].group;
295 group_data[group].free_blocks_count--;
296 flex_gd->bg_flags[group] &= uninit_mask;
297 }
298
299 /* Allocate inode bitmaps */
300 for (; ib_index < flex_gd->count; ib_index++) {
301 if (start_blk >= last_blk)
302 goto next_group;
303 group_data[ib_index].inode_bitmap = start_blk++;
304 group = ext4_get_group_number(sb, start_blk - 1);
305 group -= group_data[0].group;
306 group_data[group].free_blocks_count--;
307 flex_gd->bg_flags[group] &= uninit_mask;
308 }
309
310 /* Allocate inode tables */
311 for (; it_index < flex_gd->count; it_index++) {
312 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
313 ext4_fsblk_t next_group_start;
314
315 if (start_blk + itb > last_blk)
316 goto next_group;
317 group_data[it_index].inode_table = start_blk;
318 group = ext4_get_group_number(sb, start_blk);
319 next_group_start = ext4_group_first_block_no(sb, group + 1);
320 group -= group_data[0].group;
321
322 if (start_blk + itb > next_group_start) {
323 flex_gd->bg_flags[group + 1] &= uninit_mask;
324 overhead = start_blk + itb - next_group_start;
325 group_data[group + 1].free_blocks_count -= overhead;
326 itb -= overhead;
327 }
328
329 group_data[group].free_blocks_count -= itb;
330 flex_gd->bg_flags[group] &= uninit_mask;
331 start_blk += EXT4_SB(sb)->s_itb_per_group;
332 }
333
334 if (test_opt(sb, DEBUG)) {
335 int i;
336 group = group_data[0].group;
337
338 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
339 "%d groups, flexbg size is %d:\n", flex_gd->count,
340 flexbg_size);
341
342 for (i = 0; i < flex_gd->count; i++) {
343 printk(KERN_DEBUG "adding %s group %u: %u "
344 "blocks (%d free)\n",
345 ext4_bg_has_super(sb, group + i) ? "normal" :
346 "no-super", group + i,
347 group_data[i].blocks_count,
348 group_data[i].free_blocks_count);
349 }
350 }
351 return 0;
352}
353
354static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
355 ext4_fsblk_t blk)
356{
357 struct buffer_head *bh;
358 int err;
359
360 bh = sb_getblk(sb, blk);
361 if (unlikely(!bh))
362 return ERR_PTR(-ENOMEM);
363 BUFFER_TRACE(bh, "get_write_access");
364 if ((err = ext4_journal_get_write_access(handle, bh))) {
365 brelse(bh);
366 bh = ERR_PTR(err);
367 } else {
368 memset(bh->b_data, 0, sb->s_blocksize);
369 set_buffer_uptodate(bh);
370 }
371
372 return bh;
373}
374
375/*
376 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
377 * If that fails, restart the transaction & regain write access for the
378 * buffer head which is used for block_bitmap modifications.
379 */
380static int extend_or_restart_transaction(handle_t *handle, int thresh)
381{
382 int err;
383
384 if (ext4_handle_has_enough_credits(handle, thresh))
385 return 0;
386
387 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
388 if (err < 0)
389 return err;
390 if (err) {
391 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
392 if (err)
393 return err;
394 }
395
396 return 0;
397}
398
399/*
400 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
401 *
402 * Helper function for ext4_setup_new_group_blocks() which set .
403 *
404 * @sb: super block
405 * @handle: journal handle
406 * @flex_gd: flex group data
407 */
408static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
409 struct ext4_new_flex_group_data *flex_gd,
410 ext4_fsblk_t block, ext4_group_t count)
411{
412 ext4_group_t count2;
413
414 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
415 for (count2 = count; count > 0; count -= count2, block += count2) {
416 ext4_fsblk_t start;
417 struct buffer_head *bh;
418 ext4_group_t group;
419 int err;
420
421 group = ext4_get_group_number(sb, block);
422 start = ext4_group_first_block_no(sb, group);
423 group -= flex_gd->groups[0].group;
424
425 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
426 if (count2 > count)
427 count2 = count;
428
429 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
430 BUG_ON(flex_gd->count > 1);
431 continue;
432 }
433
434 err = extend_or_restart_transaction(handle, 1);
435 if (err)
436 return err;
437
438 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
439 if (unlikely(!bh))
440 return -ENOMEM;
441
442 BUFFER_TRACE(bh, "get_write_access");
443 err = ext4_journal_get_write_access(handle, bh);
444 if (err)
445 return err;
446 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
447 block - start, count2);
448 ext4_set_bits(bh->b_data, block - start, count2);
449
450 err = ext4_handle_dirty_metadata(handle, NULL, bh);
451 if (unlikely(err))
452 return err;
453 brelse(bh);
454 }
455
456 return 0;
457}
458
459/*
460 * Set up the block and inode bitmaps, and the inode table for the new groups.
461 * This doesn't need to be part of the main transaction, since we are only
462 * changing blocks outside the actual filesystem. We still do journaling to
463 * ensure the recovery is correct in case of a failure just after resize.
464 * If any part of this fails, we simply abort the resize.
465 *
466 * setup_new_flex_group_blocks handles a flex group as follow:
467 * 1. copy super block and GDT, and initialize group tables if necessary.
468 * In this step, we only set bits in blocks bitmaps for blocks taken by
469 * super block and GDT.
470 * 2. allocate group tables in block bitmaps, that is, set bits in block
471 * bitmap for blocks taken by group tables.
472 */
473static int setup_new_flex_group_blocks(struct super_block *sb,
474 struct ext4_new_flex_group_data *flex_gd)
475{
476 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
477 ext4_fsblk_t start;
478 ext4_fsblk_t block;
479 struct ext4_sb_info *sbi = EXT4_SB(sb);
480 struct ext4_super_block *es = sbi->s_es;
481 struct ext4_new_group_data *group_data = flex_gd->groups;
482 __u16 *bg_flags = flex_gd->bg_flags;
483 handle_t *handle;
484 ext4_group_t group, count;
485 struct buffer_head *bh = NULL;
486 int reserved_gdb, i, j, err = 0, err2;
487 int meta_bg;
488
489 BUG_ON(!flex_gd->count || !group_data ||
490 group_data[0].group != sbi->s_groups_count);
491
492 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
493 meta_bg = ext4_has_feature_meta_bg(sb);
494
495 /* This transaction may be extended/restarted along the way */
496 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
497 if (IS_ERR(handle))
498 return PTR_ERR(handle);
499
500 group = group_data[0].group;
501 for (i = 0; i < flex_gd->count; i++, group++) {
502 unsigned long gdblocks;
503 ext4_grpblk_t overhead;
504
505 gdblocks = ext4_bg_num_gdb(sb, group);
506 start = ext4_group_first_block_no(sb, group);
507
508 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
509 goto handle_itb;
510
511 if (meta_bg == 1) {
512 ext4_group_t first_group;
513 first_group = ext4_meta_bg_first_group(sb, group);
514 if (first_group != group + 1 &&
515 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
516 goto handle_itb;
517 }
518
519 block = start + ext4_bg_has_super(sb, group);
520 /* Copy all of the GDT blocks into the backup in this group */
521 for (j = 0; j < gdblocks; j++, block++) {
522 struct buffer_head *gdb;
523
524 ext4_debug("update backup group %#04llx\n", block);
525 err = extend_or_restart_transaction(handle, 1);
526 if (err)
527 goto out;
528
529 gdb = sb_getblk(sb, block);
530 if (unlikely(!gdb)) {
531 err = -ENOMEM;
532 goto out;
533 }
534
535 BUFFER_TRACE(gdb, "get_write_access");
536 err = ext4_journal_get_write_access(handle, gdb);
537 if (err) {
538 brelse(gdb);
539 goto out;
540 }
541 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
542 gdb->b_size);
543 set_buffer_uptodate(gdb);
544
545 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
546 if (unlikely(err)) {
547 brelse(gdb);
548 goto out;
549 }
550 brelse(gdb);
551 }
552
553 /* Zero out all of the reserved backup group descriptor
554 * table blocks
555 */
556 if (ext4_bg_has_super(sb, group)) {
557 err = sb_issue_zeroout(sb, gdblocks + start + 1,
558 reserved_gdb, GFP_NOFS);
559 if (err)
560 goto out;
561 }
562
563handle_itb:
564 /* Initialize group tables of the grop @group */
565 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
566 goto handle_bb;
567
568 /* Zero out all of the inode table blocks */
569 block = group_data[i].inode_table;
570 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
571 block, sbi->s_itb_per_group);
572 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
573 GFP_NOFS);
574 if (err)
575 goto out;
576
577handle_bb:
578 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
579 goto handle_ib;
580
581 /* Initialize block bitmap of the @group */
582 block = group_data[i].block_bitmap;
583 err = extend_or_restart_transaction(handle, 1);
584 if (err)
585 goto out;
586
587 bh = bclean(handle, sb, block);
588 if (IS_ERR(bh)) {
589 err = PTR_ERR(bh);
590 bh = NULL;
591 goto out;
592 }
593 overhead = ext4_group_overhead_blocks(sb, group);
594 if (overhead != 0) {
595 ext4_debug("mark backup superblock %#04llx (+0)\n",
596 start);
597 ext4_set_bits(bh->b_data, 0, overhead);
598 }
599 ext4_mark_bitmap_end(group_data[i].blocks_count,
600 sb->s_blocksize * 8, bh->b_data);
601 err = ext4_handle_dirty_metadata(handle, NULL, bh);
602 if (err)
603 goto out;
604 brelse(bh);
605
606handle_ib:
607 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
608 continue;
609
610 /* Initialize inode bitmap of the @group */
611 block = group_data[i].inode_bitmap;
612 err = extend_or_restart_transaction(handle, 1);
613 if (err)
614 goto out;
615 /* Mark unused entries in inode bitmap used */
616 bh = bclean(handle, sb, block);
617 if (IS_ERR(bh)) {
618 err = PTR_ERR(bh);
619 bh = NULL;
620 goto out;
621 }
622
623 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
624 sb->s_blocksize * 8, bh->b_data);
625 err = ext4_handle_dirty_metadata(handle, NULL, bh);
626 if (err)
627 goto out;
628 brelse(bh);
629 }
630 bh = NULL;
631
632 /* Mark group tables in block bitmap */
633 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
634 count = group_table_count[j];
635 start = (&group_data[0].block_bitmap)[j];
636 block = start;
637 for (i = 1; i < flex_gd->count; i++) {
638 block += group_table_count[j];
639 if (block == (&group_data[i].block_bitmap)[j]) {
640 count += group_table_count[j];
641 continue;
642 }
643 err = set_flexbg_block_bitmap(sb, handle,
644 flex_gd, start, count);
645 if (err)
646 goto out;
647 count = group_table_count[j];
648 start = (&group_data[i].block_bitmap)[j];
649 block = start;
650 }
651
652 if (count) {
653 err = set_flexbg_block_bitmap(sb, handle,
654 flex_gd, start, count);
655 if (err)
656 goto out;
657 }
658 }
659
660out:
661 brelse(bh);
662 err2 = ext4_journal_stop(handle);
663 if (err2 && !err)
664 err = err2;
665
666 return err;
667}
668
669/*
670 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
671 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
672 * calling this for the first time. In a sparse filesystem it will be the
673 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
674 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
675 */
676static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
677 unsigned *five, unsigned *seven)
678{
679 unsigned *min = three;
680 int mult = 3;
681 unsigned ret;
682
683 if (!ext4_has_feature_sparse_super(sb)) {
684 ret = *min;
685 *min += 1;
686 return ret;
687 }
688
689 if (*five < *min) {
690 min = five;
691 mult = 5;
692 }
693 if (*seven < *min) {
694 min = seven;
695 mult = 7;
696 }
697
698 ret = *min;
699 *min *= mult;
700
701 return ret;
702}
703
704/*
705 * Check that all of the backup GDT blocks are held in the primary GDT block.
706 * It is assumed that they are stored in group order. Returns the number of
707 * groups in current filesystem that have BACKUPS, or -ve error code.
708 */
709static int verify_reserved_gdb(struct super_block *sb,
710 ext4_group_t end,
711 struct buffer_head *primary)
712{
713 const ext4_fsblk_t blk = primary->b_blocknr;
714 unsigned three = 1;
715 unsigned five = 5;
716 unsigned seven = 7;
717 unsigned grp;
718 __le32 *p = (__le32 *)primary->b_data;
719 int gdbackups = 0;
720
721 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
722 if (le32_to_cpu(*p++) !=
723 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
724 ext4_warning(sb, "reserved GDT %llu"
725 " missing grp %d (%llu)",
726 blk, grp,
727 grp *
728 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
729 blk);
730 return -EINVAL;
731 }
732 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
733 return -EFBIG;
734 }
735
736 return gdbackups;
737}
738
739/*
740 * Called when we need to bring a reserved group descriptor table block into
741 * use from the resize inode. The primary copy of the new GDT block currently
742 * is an indirect block (under the double indirect block in the resize inode).
743 * The new backup GDT blocks will be stored as leaf blocks in this indirect
744 * block, in group order. Even though we know all the block numbers we need,
745 * we check to ensure that the resize inode has actually reserved these blocks.
746 *
747 * Don't need to update the block bitmaps because the blocks are still in use.
748 *
749 * We get all of the error cases out of the way, so that we are sure to not
750 * fail once we start modifying the data on disk, because JBD has no rollback.
751 */
752static int add_new_gdb(handle_t *handle, struct inode *inode,
753 ext4_group_t group)
754{
755 struct super_block *sb = inode->i_sb;
756 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
757 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
758 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
759 struct buffer_head **o_group_desc, **n_group_desc;
760 struct buffer_head *dind;
761 struct buffer_head *gdb_bh;
762 int gdbackups;
763 struct ext4_iloc iloc;
764 __le32 *data;
765 int err;
766
767 if (test_opt(sb, DEBUG))
768 printk(KERN_DEBUG
769 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
770 gdb_num);
771
772 gdb_bh = sb_bread(sb, gdblock);
773 if (!gdb_bh)
774 return -EIO;
775
776 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
777 if (gdbackups < 0) {
778 err = gdbackups;
779 goto exit_bh;
780 }
781
782 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
783 dind = sb_bread(sb, le32_to_cpu(*data));
784 if (!dind) {
785 err = -EIO;
786 goto exit_bh;
787 }
788
789 data = (__le32 *)dind->b_data;
790 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
791 ext4_warning(sb, "new group %u GDT block %llu not reserved",
792 group, gdblock);
793 err = -EINVAL;
794 goto exit_dind;
795 }
796
797 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
798 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
799 if (unlikely(err))
800 goto exit_dind;
801
802 BUFFER_TRACE(gdb_bh, "get_write_access");
803 err = ext4_journal_get_write_access(handle, gdb_bh);
804 if (unlikely(err))
805 goto exit_dind;
806
807 BUFFER_TRACE(dind, "get_write_access");
808 err = ext4_journal_get_write_access(handle, dind);
809 if (unlikely(err))
810 ext4_std_error(sb, err);
811
812 /* ext4_reserve_inode_write() gets a reference on the iloc */
813 err = ext4_reserve_inode_write(handle, inode, &iloc);
814 if (unlikely(err))
815 goto exit_dind;
816
817 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
818 sizeof(struct buffer_head *),
819 GFP_NOFS);
820 if (!n_group_desc) {
821 err = -ENOMEM;
822 ext4_warning(sb, "not enough memory for %lu groups",
823 gdb_num + 1);
824 goto exit_inode;
825 }
826
827 /*
828 * Finally, we have all of the possible failures behind us...
829 *
830 * Remove new GDT block from inode double-indirect block and clear out
831 * the new GDT block for use (which also "frees" the backup GDT blocks
832 * from the reserved inode). We don't need to change the bitmaps for
833 * these blocks, because they are marked as in-use from being in the
834 * reserved inode, and will become GDT blocks (primary and backup).
835 */
836 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
837 err = ext4_handle_dirty_metadata(handle, NULL, dind);
838 if (unlikely(err)) {
839 ext4_std_error(sb, err);
840 goto exit_inode;
841 }
842 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
843 ext4_mark_iloc_dirty(handle, inode, &iloc);
844 memset(gdb_bh->b_data, 0, sb->s_blocksize);
845 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
846 if (unlikely(err)) {
847 ext4_std_error(sb, err);
848 goto exit_inode;
849 }
850 brelse(dind);
851
852 o_group_desc = EXT4_SB(sb)->s_group_desc;
853 memcpy(n_group_desc, o_group_desc,
854 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
855 n_group_desc[gdb_num] = gdb_bh;
856 EXT4_SB(sb)->s_group_desc = n_group_desc;
857 EXT4_SB(sb)->s_gdb_count++;
858 kvfree(o_group_desc);
859
860 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
861 err = ext4_handle_dirty_super(handle, sb);
862 if (err)
863 ext4_std_error(sb, err);
864
865 return err;
866
867exit_inode:
868 kvfree(n_group_desc);
869 brelse(iloc.bh);
870exit_dind:
871 brelse(dind);
872exit_bh:
873 brelse(gdb_bh);
874
875 ext4_debug("leaving with error %d\n", err);
876 return err;
877}
878
879/*
880 * add_new_gdb_meta_bg is the sister of add_new_gdb.
881 */
882static int add_new_gdb_meta_bg(struct super_block *sb,
883 handle_t *handle, ext4_group_t group) {
884 ext4_fsblk_t gdblock;
885 struct buffer_head *gdb_bh;
886 struct buffer_head **o_group_desc, **n_group_desc;
887 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
888 int err;
889
890 gdblock = ext4_meta_bg_first_block_no(sb, group) +
891 ext4_bg_has_super(sb, group);
892 gdb_bh = sb_bread(sb, gdblock);
893 if (!gdb_bh)
894 return -EIO;
895 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
896 sizeof(struct buffer_head *),
897 GFP_NOFS);
898 if (!n_group_desc) {
899 err = -ENOMEM;
900 ext4_warning(sb, "not enough memory for %lu groups",
901 gdb_num + 1);
902 return err;
903 }
904
905 o_group_desc = EXT4_SB(sb)->s_group_desc;
906 memcpy(n_group_desc, o_group_desc,
907 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
908 n_group_desc[gdb_num] = gdb_bh;
909 EXT4_SB(sb)->s_group_desc = n_group_desc;
910 EXT4_SB(sb)->s_gdb_count++;
911 kvfree(o_group_desc);
912 BUFFER_TRACE(gdb_bh, "get_write_access");
913 err = ext4_journal_get_write_access(handle, gdb_bh);
914 if (unlikely(err))
915 brelse(gdb_bh);
916 return err;
917}
918
919/*
920 * Called when we are adding a new group which has a backup copy of each of
921 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
922 * We need to add these reserved backup GDT blocks to the resize inode, so
923 * that they are kept for future resizing and not allocated to files.
924 *
925 * Each reserved backup GDT block will go into a different indirect block.
926 * The indirect blocks are actually the primary reserved GDT blocks,
927 * so we know in advance what their block numbers are. We only get the
928 * double-indirect block to verify it is pointing to the primary reserved
929 * GDT blocks so we don't overwrite a data block by accident. The reserved
930 * backup GDT blocks are stored in their reserved primary GDT block.
931 */
932static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
933 ext4_group_t group)
934{
935 struct super_block *sb = inode->i_sb;
936 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
937 struct buffer_head **primary;
938 struct buffer_head *dind;
939 struct ext4_iloc iloc;
940 ext4_fsblk_t blk;
941 __le32 *data, *end;
942 int gdbackups = 0;
943 int res, i;
944 int err;
945
946 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
947 if (!primary)
948 return -ENOMEM;
949
950 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
951 dind = sb_bread(sb, le32_to_cpu(*data));
952 if (!dind) {
953 err = -EIO;
954 goto exit_free;
955 }
956
957 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
958 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
959 EXT4_ADDR_PER_BLOCK(sb));
960 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
961
962 /* Get each reserved primary GDT block and verify it holds backups */
963 for (res = 0; res < reserved_gdb; res++, blk++) {
964 if (le32_to_cpu(*data) != blk) {
965 ext4_warning(sb, "reserved block %llu"
966 " not at offset %ld",
967 blk,
968 (long)(data - (__le32 *)dind->b_data));
969 err = -EINVAL;
970 goto exit_bh;
971 }
972 primary[res] = sb_bread(sb, blk);
973 if (!primary[res]) {
974 err = -EIO;
975 goto exit_bh;
976 }
977 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
978 if (gdbackups < 0) {
979 brelse(primary[res]);
980 err = gdbackups;
981 goto exit_bh;
982 }
983 if (++data >= end)
984 data = (__le32 *)dind->b_data;
985 }
986
987 for (i = 0; i < reserved_gdb; i++) {
988 BUFFER_TRACE(primary[i], "get_write_access");
989 if ((err = ext4_journal_get_write_access(handle, primary[i])))
990 goto exit_bh;
991 }
992
993 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
994 goto exit_bh;
995
996 /*
997 * Finally we can add each of the reserved backup GDT blocks from
998 * the new group to its reserved primary GDT block.
999 */
1000 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1001 for (i = 0; i < reserved_gdb; i++) {
1002 int err2;
1003 data = (__le32 *)primary[i]->b_data;
1004 /* printk("reserving backup %lu[%u] = %lu\n",
1005 primary[i]->b_blocknr, gdbackups,
1006 blk + primary[i]->b_blocknr); */
1007 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1008 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1009 if (!err)
1010 err = err2;
1011 }
1012 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1013 ext4_mark_iloc_dirty(handle, inode, &iloc);
1014
1015exit_bh:
1016 while (--res >= 0)
1017 brelse(primary[res]);
1018 brelse(dind);
1019
1020exit_free:
1021 kfree(primary);
1022
1023 return err;
1024}
1025
1026/*
1027 * Update the backup copies of the ext4 metadata. These don't need to be part
1028 * of the main resize transaction, because e2fsck will re-write them if there
1029 * is a problem (basically only OOM will cause a problem). However, we
1030 * _should_ update the backups if possible, in case the primary gets trashed
1031 * for some reason and we need to run e2fsck from a backup superblock. The
1032 * important part is that the new block and inode counts are in the backup
1033 * superblocks, and the location of the new group metadata in the GDT backups.
1034 *
1035 * We do not need take the s_resize_lock for this, because these
1036 * blocks are not otherwise touched by the filesystem code when it is
1037 * mounted. We don't need to worry about last changing from
1038 * sbi->s_groups_count, because the worst that can happen is that we
1039 * do not copy the full number of backups at this time. The resize
1040 * which changed s_groups_count will backup again.
1041 */
1042static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1043 int size, int meta_bg)
1044{
1045 struct ext4_sb_info *sbi = EXT4_SB(sb);
1046 ext4_group_t last;
1047 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1048 unsigned three = 1;
1049 unsigned five = 5;
1050 unsigned seven = 7;
1051 ext4_group_t group = 0;
1052 int rest = sb->s_blocksize - size;
1053 handle_t *handle;
1054 int err = 0, err2;
1055
1056 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1057 if (IS_ERR(handle)) {
1058 group = 1;
1059 err = PTR_ERR(handle);
1060 goto exit_err;
1061 }
1062
1063 if (meta_bg == 0) {
1064 group = ext4_list_backups(sb, &three, &five, &seven);
1065 last = sbi->s_groups_count;
1066 } else {
1067 group = ext4_get_group_number(sb, blk_off) + 1;
1068 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1069 }
1070
1071 while (group < sbi->s_groups_count) {
1072 struct buffer_head *bh;
1073 ext4_fsblk_t backup_block;
1074
1075 /* Out of journal space, and can't get more - abort - so sad */
1076 if (ext4_handle_valid(handle) &&
1077 handle->h_buffer_credits == 0 &&
1078 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1079 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1080 break;
1081
1082 if (meta_bg == 0)
1083 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1084 else
1085 backup_block = (ext4_group_first_block_no(sb, group) +
1086 ext4_bg_has_super(sb, group));
1087
1088 bh = sb_getblk(sb, backup_block);
1089 if (unlikely(!bh)) {
1090 err = -ENOMEM;
1091 break;
1092 }
1093 ext4_debug("update metadata backup %llu(+%llu)\n",
1094 backup_block, backup_block -
1095 ext4_group_first_block_no(sb, group));
1096 BUFFER_TRACE(bh, "get_write_access");
1097 if ((err = ext4_journal_get_write_access(handle, bh)))
1098 break;
1099 lock_buffer(bh);
1100 memcpy(bh->b_data, data, size);
1101 if (rest)
1102 memset(bh->b_data + size, 0, rest);
1103 set_buffer_uptodate(bh);
1104 unlock_buffer(bh);
1105 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1106 if (unlikely(err))
1107 ext4_std_error(sb, err);
1108 brelse(bh);
1109
1110 if (meta_bg == 0)
1111 group = ext4_list_backups(sb, &three, &five, &seven);
1112 else if (group == last)
1113 break;
1114 else
1115 group = last;
1116 }
1117 if ((err2 = ext4_journal_stop(handle)) && !err)
1118 err = err2;
1119
1120 /*
1121 * Ugh! Need to have e2fsck write the backup copies. It is too
1122 * late to revert the resize, we shouldn't fail just because of
1123 * the backup copies (they are only needed in case of corruption).
1124 *
1125 * However, if we got here we have a journal problem too, so we
1126 * can't really start a transaction to mark the superblock.
1127 * Chicken out and just set the flag on the hope it will be written
1128 * to disk, and if not - we will simply wait until next fsck.
1129 */
1130exit_err:
1131 if (err) {
1132 ext4_warning(sb, "can't update backup for group %u (err %d), "
1133 "forcing fsck on next reboot", group, err);
1134 sbi->s_mount_state &= ~EXT4_VALID_FS;
1135 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1136 mark_buffer_dirty(sbi->s_sbh);
1137 }
1138}
1139
1140/*
1141 * ext4_add_new_descs() adds @count group descriptor of groups
1142 * starting at @group
1143 *
1144 * @handle: journal handle
1145 * @sb: super block
1146 * @group: the group no. of the first group desc to be added
1147 * @resize_inode: the resize inode
1148 * @count: number of group descriptors to be added
1149 */
1150static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1151 ext4_group_t group, struct inode *resize_inode,
1152 ext4_group_t count)
1153{
1154 struct ext4_sb_info *sbi = EXT4_SB(sb);
1155 struct ext4_super_block *es = sbi->s_es;
1156 struct buffer_head *gdb_bh;
1157 int i, gdb_off, gdb_num, err = 0;
1158 int meta_bg;
1159
1160 meta_bg = ext4_has_feature_meta_bg(sb);
1161 for (i = 0; i < count; i++, group++) {
1162 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1163 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1164
1165 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1166 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1167
1168 /*
1169 * We will only either add reserved group blocks to a backup group
1170 * or remove reserved blocks for the first group in a new group block.
1171 * Doing both would be mean more complex code, and sane people don't
1172 * use non-sparse filesystems anymore. This is already checked above.
1173 */
1174 if (gdb_off) {
1175 gdb_bh = sbi->s_group_desc[gdb_num];
1176 BUFFER_TRACE(gdb_bh, "get_write_access");
1177 err = ext4_journal_get_write_access(handle, gdb_bh);
1178
1179 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1180 err = reserve_backup_gdb(handle, resize_inode, group);
1181 } else if (meta_bg != 0) {
1182 err = add_new_gdb_meta_bg(sb, handle, group);
1183 } else {
1184 err = add_new_gdb(handle, resize_inode, group);
1185 }
1186 if (err)
1187 break;
1188 }
1189 return err;
1190}
1191
1192static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1193{
1194 struct buffer_head *bh = sb_getblk(sb, block);
1195 if (unlikely(!bh))
1196 return NULL;
1197 if (!bh_uptodate_or_lock(bh)) {
1198 if (bh_submit_read(bh) < 0) {
1199 brelse(bh);
1200 return NULL;
1201 }
1202 }
1203
1204 return bh;
1205}
1206
1207static int ext4_set_bitmap_checksums(struct super_block *sb,
1208 ext4_group_t group,
1209 struct ext4_group_desc *gdp,
1210 struct ext4_new_group_data *group_data)
1211{
1212 struct buffer_head *bh;
1213
1214 if (!ext4_has_metadata_csum(sb))
1215 return 0;
1216
1217 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1218 if (!bh)
1219 return -EIO;
1220 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1221 EXT4_INODES_PER_GROUP(sb) / 8);
1222 brelse(bh);
1223
1224 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1225 if (!bh)
1226 return -EIO;
1227 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1228 brelse(bh);
1229
1230 return 0;
1231}
1232
1233/*
1234 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1235 */
1236static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1237 struct ext4_new_flex_group_data *flex_gd)
1238{
1239 struct ext4_new_group_data *group_data = flex_gd->groups;
1240 struct ext4_group_desc *gdp;
1241 struct ext4_sb_info *sbi = EXT4_SB(sb);
1242 struct buffer_head *gdb_bh;
1243 ext4_group_t group;
1244 __u16 *bg_flags = flex_gd->bg_flags;
1245 int i, gdb_off, gdb_num, err = 0;
1246
1247
1248 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1249 group = group_data->group;
1250
1251 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1252 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1253
1254 /*
1255 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1256 */
1257 gdb_bh = sbi->s_group_desc[gdb_num];
1258 /* Update group descriptor block for new group */
1259 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1260 gdb_off * EXT4_DESC_SIZE(sb));
1261
1262 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1263 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1264 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1265 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1266 if (err) {
1267 ext4_std_error(sb, err);
1268 break;
1269 }
1270
1271 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1272 ext4_free_group_clusters_set(sb, gdp,
1273 EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1274 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1275 if (ext4_has_group_desc_csum(sb))
1276 ext4_itable_unused_set(sb, gdp,
1277 EXT4_INODES_PER_GROUP(sb));
1278 gdp->bg_flags = cpu_to_le16(*bg_flags);
1279 ext4_group_desc_csum_set(sb, group, gdp);
1280
1281 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1282 if (unlikely(err)) {
1283 ext4_std_error(sb, err);
1284 break;
1285 }
1286
1287 /*
1288 * We can allocate memory for mb_alloc based on the new group
1289 * descriptor
1290 */
1291 err = ext4_mb_add_groupinfo(sb, group, gdp);
1292 if (err)
1293 break;
1294 }
1295 return err;
1296}
1297
1298/*
1299 * ext4_update_super() updates the super block so that the newly added
1300 * groups can be seen by the filesystem.
1301 *
1302 * @sb: super block
1303 * @flex_gd: new added groups
1304 */
1305static void ext4_update_super(struct super_block *sb,
1306 struct ext4_new_flex_group_data *flex_gd)
1307{
1308 ext4_fsblk_t blocks_count = 0;
1309 ext4_fsblk_t free_blocks = 0;
1310 ext4_fsblk_t reserved_blocks = 0;
1311 struct ext4_new_group_data *group_data = flex_gd->groups;
1312 struct ext4_sb_info *sbi = EXT4_SB(sb);
1313 struct ext4_super_block *es = sbi->s_es;
1314 int i;
1315
1316 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1317 /*
1318 * Make the new blocks and inodes valid next. We do this before
1319 * increasing the group count so that once the group is enabled,
1320 * all of its blocks and inodes are already valid.
1321 *
1322 * We always allocate group-by-group, then block-by-block or
1323 * inode-by-inode within a group, so enabling these
1324 * blocks/inodes before the group is live won't actually let us
1325 * allocate the new space yet.
1326 */
1327 for (i = 0; i < flex_gd->count; i++) {
1328 blocks_count += group_data[i].blocks_count;
1329 free_blocks += group_data[i].free_blocks_count;
1330 }
1331
1332 reserved_blocks = ext4_r_blocks_count(es) * 100;
1333 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1334 reserved_blocks *= blocks_count;
1335 do_div(reserved_blocks, 100);
1336
1337 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1338 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1339 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1340 flex_gd->count);
1341 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1342 flex_gd->count);
1343
1344 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1345 /*
1346 * We need to protect s_groups_count against other CPUs seeing
1347 * inconsistent state in the superblock.
1348 *
1349 * The precise rules we use are:
1350 *
1351 * * Writers must perform a smp_wmb() after updating all
1352 * dependent data and before modifying the groups count
1353 *
1354 * * Readers must perform an smp_rmb() after reading the groups
1355 * count and before reading any dependent data.
1356 *
1357 * NB. These rules can be relaxed when checking the group count
1358 * while freeing data, as we can only allocate from a block
1359 * group after serialising against the group count, and we can
1360 * only then free after serialising in turn against that
1361 * allocation.
1362 */
1363 smp_wmb();
1364
1365 /* Update the global fs size fields */
1366 sbi->s_groups_count += flex_gd->count;
1367 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1368 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1369
1370 /* Update the reserved block counts only once the new group is
1371 * active. */
1372 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1373 reserved_blocks);
1374
1375 /* Update the free space counts */
1376 percpu_counter_add(&sbi->s_freeclusters_counter,
1377 EXT4_NUM_B2C(sbi, free_blocks));
1378 percpu_counter_add(&sbi->s_freeinodes_counter,
1379 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1380
1381 ext4_debug("free blocks count %llu",
1382 percpu_counter_read(&sbi->s_freeclusters_counter));
1383 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1384 ext4_group_t flex_group;
1385 flex_group = ext4_flex_group(sbi, group_data[0].group);
1386 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1387 &sbi->s_flex_groups[flex_group].free_clusters);
1388 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1389 &sbi->s_flex_groups[flex_group].free_inodes);
1390 }
1391
1392 /*
1393 * Update the fs overhead information
1394 */
1395 ext4_calculate_overhead(sb);
1396
1397 if (test_opt(sb, DEBUG))
1398 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1399 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1400 blocks_count, free_blocks, reserved_blocks);
1401}
1402
1403/* Add a flex group to an fs. Ensure we handle all possible error conditions
1404 * _before_ we start modifying the filesystem, because we cannot abort the
1405 * transaction and not have it write the data to disk.
1406 */
1407static int ext4_flex_group_add(struct super_block *sb,
1408 struct inode *resize_inode,
1409 struct ext4_new_flex_group_data *flex_gd)
1410{
1411 struct ext4_sb_info *sbi = EXT4_SB(sb);
1412 struct ext4_super_block *es = sbi->s_es;
1413 ext4_fsblk_t o_blocks_count;
1414 ext4_grpblk_t last;
1415 ext4_group_t group;
1416 handle_t *handle;
1417 unsigned reserved_gdb;
1418 int err = 0, err2 = 0, credit;
1419
1420 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1421
1422 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1423 o_blocks_count = ext4_blocks_count(es);
1424 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1425 BUG_ON(last);
1426
1427 err = setup_new_flex_group_blocks(sb, flex_gd);
1428 if (err)
1429 goto exit;
1430 /*
1431 * We will always be modifying at least the superblock and GDT
1432 * blocks. If we are adding a group past the last current GDT block,
1433 * we will also modify the inode and the dindirect block. If we
1434 * are adding a group with superblock/GDT backups we will also
1435 * modify each of the reserved GDT dindirect blocks.
1436 */
1437 credit = 3; /* sb, resize inode, resize inode dindirect */
1438 /* GDT blocks */
1439 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1440 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1441 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1442 if (IS_ERR(handle)) {
1443 err = PTR_ERR(handle);
1444 goto exit;
1445 }
1446
1447 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1448 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1449 if (err)
1450 goto exit_journal;
1451
1452 group = flex_gd->groups[0].group;
1453 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1454 err = ext4_add_new_descs(handle, sb, group,
1455 resize_inode, flex_gd->count);
1456 if (err)
1457 goto exit_journal;
1458
1459 err = ext4_setup_new_descs(handle, sb, flex_gd);
1460 if (err)
1461 goto exit_journal;
1462
1463 ext4_update_super(sb, flex_gd);
1464
1465 err = ext4_handle_dirty_super(handle, sb);
1466
1467exit_journal:
1468 err2 = ext4_journal_stop(handle);
1469 if (!err)
1470 err = err2;
1471
1472 if (!err) {
1473 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1474 int gdb_num_end = ((group + flex_gd->count - 1) /
1475 EXT4_DESC_PER_BLOCK(sb));
1476 int meta_bg = ext4_has_feature_meta_bg(sb);
1477 sector_t old_gdb = 0;
1478
1479 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1480 sizeof(struct ext4_super_block), 0);
1481 for (; gdb_num <= gdb_num_end; gdb_num++) {
1482 struct buffer_head *gdb_bh;
1483
1484 gdb_bh = sbi->s_group_desc[gdb_num];
1485 if (old_gdb == gdb_bh->b_blocknr)
1486 continue;
1487 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1488 gdb_bh->b_size, meta_bg);
1489 old_gdb = gdb_bh->b_blocknr;
1490 }
1491 }
1492exit:
1493 return err;
1494}
1495
1496static int ext4_setup_next_flex_gd(struct super_block *sb,
1497 struct ext4_new_flex_group_data *flex_gd,
1498 ext4_fsblk_t n_blocks_count,
1499 unsigned long flexbg_size)
1500{
1501 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1502 struct ext4_new_group_data *group_data = flex_gd->groups;
1503 ext4_fsblk_t o_blocks_count;
1504 ext4_group_t n_group;
1505 ext4_group_t group;
1506 ext4_group_t last_group;
1507 ext4_grpblk_t last;
1508 ext4_grpblk_t blocks_per_group;
1509 unsigned long i;
1510
1511 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1512
1513 o_blocks_count = ext4_blocks_count(es);
1514
1515 if (o_blocks_count == n_blocks_count)
1516 return 0;
1517
1518 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1519 BUG_ON(last);
1520 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1521
1522 last_group = group | (flexbg_size - 1);
1523 if (last_group > n_group)
1524 last_group = n_group;
1525
1526 flex_gd->count = last_group - group + 1;
1527
1528 for (i = 0; i < flex_gd->count; i++) {
1529 int overhead;
1530
1531 group_data[i].group = group + i;
1532 group_data[i].blocks_count = blocks_per_group;
1533 overhead = ext4_group_overhead_blocks(sb, group + i);
1534 group_data[i].free_blocks_count = blocks_per_group - overhead;
1535 if (ext4_has_group_desc_csum(sb)) {
1536 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1537 EXT4_BG_INODE_UNINIT;
1538 if (!test_opt(sb, INIT_INODE_TABLE))
1539 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1540 } else
1541 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1542 }
1543
1544 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1545 /* We need to initialize block bitmap of last group. */
1546 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1547
1548 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1549 group_data[i - 1].blocks_count = last + 1;
1550 group_data[i - 1].free_blocks_count -= blocks_per_group-
1551 last - 1;
1552 }
1553
1554 return 1;
1555}
1556
1557/* Add group descriptor data to an existing or new group descriptor block.
1558 * Ensure we handle all possible error conditions _before_ we start modifying
1559 * the filesystem, because we cannot abort the transaction and not have it
1560 * write the data to disk.
1561 *
1562 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1563 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1564 *
1565 * We only need to hold the superblock lock while we are actually adding
1566 * in the new group's counts to the superblock. Prior to that we have
1567 * not really "added" the group at all. We re-check that we are still
1568 * adding in the last group in case things have changed since verifying.
1569 */
1570int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1571{
1572 struct ext4_new_flex_group_data flex_gd;
1573 struct ext4_sb_info *sbi = EXT4_SB(sb);
1574 struct ext4_super_block *es = sbi->s_es;
1575 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1576 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1577 struct inode *inode = NULL;
1578 int gdb_off;
1579 int err;
1580 __u16 bg_flags = 0;
1581
1582 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1583
1584 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1585 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1586 return -EPERM;
1587 }
1588
1589 if (ext4_blocks_count(es) + input->blocks_count <
1590 ext4_blocks_count(es)) {
1591 ext4_warning(sb, "blocks_count overflow");
1592 return -EINVAL;
1593 }
1594
1595 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1596 le32_to_cpu(es->s_inodes_count)) {
1597 ext4_warning(sb, "inodes_count overflow");
1598 return -EINVAL;
1599 }
1600
1601 if (reserved_gdb || gdb_off == 0) {
1602 if (ext4_has_feature_resize_inode(sb) ||
1603 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1604 ext4_warning(sb,
1605 "No reserved GDT blocks, can't resize");
1606 return -EPERM;
1607 }
1608 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1609 if (IS_ERR(inode)) {
1610 ext4_warning(sb, "Error opening resize inode");
1611 return PTR_ERR(inode);
1612 }
1613 }
1614
1615
1616 err = verify_group_input(sb, input);
1617 if (err)
1618 goto out;
1619
1620 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1621 if (err)
1622 goto out;
1623
1624 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1625 if (err)
1626 goto out;
1627
1628 flex_gd.count = 1;
1629 flex_gd.groups = input;
1630 flex_gd.bg_flags = &bg_flags;
1631 err = ext4_flex_group_add(sb, inode, &flex_gd);
1632out:
1633 iput(inode);
1634 return err;
1635} /* ext4_group_add */
1636
1637/*
1638 * extend a group without checking assuming that checking has been done.
1639 */
1640static int ext4_group_extend_no_check(struct super_block *sb,
1641 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1642{
1643 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1644 handle_t *handle;
1645 int err = 0, err2;
1646
1647 /* We will update the superblock, one block bitmap, and
1648 * one group descriptor via ext4_group_add_blocks().
1649 */
1650 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1651 if (IS_ERR(handle)) {
1652 err = PTR_ERR(handle);
1653 ext4_warning(sb, "error %d on journal start", err);
1654 return err;
1655 }
1656
1657 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1658 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1659 if (err) {
1660 ext4_warning(sb, "error %d on journal write access", err);
1661 goto errout;
1662 }
1663
1664 ext4_blocks_count_set(es, o_blocks_count + add);
1665 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1666 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1667 o_blocks_count + add);
1668 /* We add the blocks to the bitmap and set the group need init bit */
1669 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1670 if (err)
1671 goto errout;
1672 ext4_handle_dirty_super(handle, sb);
1673 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1674 o_blocks_count + add);
1675errout:
1676 err2 = ext4_journal_stop(handle);
1677 if (err2 && !err)
1678 err = err2;
1679
1680 if (!err) {
1681 if (test_opt(sb, DEBUG))
1682 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1683 "blocks\n", ext4_blocks_count(es));
1684 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1685 (char *)es, sizeof(struct ext4_super_block), 0);
1686 }
1687 return err;
1688}
1689
1690/*
1691 * Extend the filesystem to the new number of blocks specified. This entry
1692 * point is only used to extend the current filesystem to the end of the last
1693 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1694 * for emergencies (because it has no dependencies on reserved blocks).
1695 *
1696 * If we _really_ wanted, we could use default values to call ext4_group_add()
1697 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1698 * GDT blocks are reserved to grow to the desired size.
1699 */
1700int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1701 ext4_fsblk_t n_blocks_count)
1702{
1703 ext4_fsblk_t o_blocks_count;
1704 ext4_grpblk_t last;
1705 ext4_grpblk_t add;
1706 struct buffer_head *bh;
1707 int err;
1708 ext4_group_t group;
1709
1710 o_blocks_count = ext4_blocks_count(es);
1711
1712 if (test_opt(sb, DEBUG))
1713 ext4_msg(sb, KERN_DEBUG,
1714 "extending last group from %llu to %llu blocks",
1715 o_blocks_count, n_blocks_count);
1716
1717 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1718 return 0;
1719
1720 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1721 ext4_msg(sb, KERN_ERR,
1722 "filesystem too large to resize to %llu blocks safely",
1723 n_blocks_count);
1724 if (sizeof(sector_t) < 8)
1725 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1726 return -EINVAL;
1727 }
1728
1729 if (n_blocks_count < o_blocks_count) {
1730 ext4_warning(sb, "can't shrink FS - resize aborted");
1731 return -EINVAL;
1732 }
1733
1734 /* Handle the remaining blocks in the last group only. */
1735 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1736
1737 if (last == 0) {
1738 ext4_warning(sb, "need to use ext2online to resize further");
1739 return -EPERM;
1740 }
1741
1742 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1743
1744 if (o_blocks_count + add < o_blocks_count) {
1745 ext4_warning(sb, "blocks_count overflow");
1746 return -EINVAL;
1747 }
1748
1749 if (o_blocks_count + add > n_blocks_count)
1750 add = n_blocks_count - o_blocks_count;
1751
1752 if (o_blocks_count + add < n_blocks_count)
1753 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1754 o_blocks_count + add, add);
1755
1756 /* See if the device is actually as big as what was requested */
1757 bh = sb_bread(sb, o_blocks_count + add - 1);
1758 if (!bh) {
1759 ext4_warning(sb, "can't read last block, resize aborted");
1760 return -ENOSPC;
1761 }
1762 brelse(bh);
1763
1764 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1765 return err;
1766} /* ext4_group_extend */
1767
1768
1769static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1770{
1771 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1772}
1773
1774/*
1775 * Release the resize inode and drop the resize_inode feature if there
1776 * are no more reserved gdt blocks, and then convert the file system
1777 * to enable meta_bg
1778 */
1779static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1780{
1781 handle_t *handle;
1782 struct ext4_sb_info *sbi = EXT4_SB(sb);
1783 struct ext4_super_block *es = sbi->s_es;
1784 struct ext4_inode_info *ei = EXT4_I(inode);
1785 ext4_fsblk_t nr;
1786 int i, ret, err = 0;
1787 int credits = 1;
1788
1789 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1790 if (inode) {
1791 if (es->s_reserved_gdt_blocks) {
1792 ext4_error(sb, "Unexpected non-zero "
1793 "s_reserved_gdt_blocks");
1794 return -EPERM;
1795 }
1796
1797 /* Do a quick sanity check of the resize inode */
1798 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1799 goto invalid_resize_inode;
1800 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1801 if (i == EXT4_DIND_BLOCK) {
1802 if (ei->i_data[i])
1803 continue;
1804 else
1805 goto invalid_resize_inode;
1806 }
1807 if (ei->i_data[i])
1808 goto invalid_resize_inode;
1809 }
1810 credits += 3; /* block bitmap, bg descriptor, resize inode */
1811 }
1812
1813 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1814 if (IS_ERR(handle))
1815 return PTR_ERR(handle);
1816
1817 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1818 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1819 if (err)
1820 goto errout;
1821
1822 ext4_clear_feature_resize_inode(sb);
1823 ext4_set_feature_meta_bg(sb);
1824 sbi->s_es->s_first_meta_bg =
1825 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1826
1827 err = ext4_handle_dirty_super(handle, sb);
1828 if (err) {
1829 ext4_std_error(sb, err);
1830 goto errout;
1831 }
1832
1833 if (inode) {
1834 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1835 ext4_free_blocks(handle, inode, NULL, nr, 1,
1836 EXT4_FREE_BLOCKS_METADATA |
1837 EXT4_FREE_BLOCKS_FORGET);
1838 ei->i_data[EXT4_DIND_BLOCK] = 0;
1839 inode->i_blocks = 0;
1840
1841 err = ext4_mark_inode_dirty(handle, inode);
1842 if (err)
1843 ext4_std_error(sb, err);
1844 }
1845
1846errout:
1847 ret = ext4_journal_stop(handle);
1848 if (!err)
1849 err = ret;
1850 return ret;
1851
1852invalid_resize_inode:
1853 ext4_error(sb, "corrupted/inconsistent resize inode");
1854 return -EINVAL;
1855}
1856
1857/*
1858 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1859 *
1860 * @sb: super block of the fs to be resized
1861 * @n_blocks_count: the number of blocks resides in the resized fs
1862 */
1863int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1864{
1865 struct ext4_new_flex_group_data *flex_gd = NULL;
1866 struct ext4_sb_info *sbi = EXT4_SB(sb);
1867 struct ext4_super_block *es = sbi->s_es;
1868 struct buffer_head *bh;
1869 struct inode *resize_inode = NULL;
1870 ext4_grpblk_t add, offset;
1871 unsigned long n_desc_blocks;
1872 unsigned long o_desc_blocks;
1873 ext4_group_t o_group;
1874 ext4_group_t n_group;
1875 ext4_fsblk_t o_blocks_count;
1876 ext4_fsblk_t n_blocks_count_retry = 0;
1877 unsigned long last_update_time = 0;
1878 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1879 int meta_bg;
1880
1881 /* See if the device is actually as big as what was requested */
1882 bh = sb_bread(sb, n_blocks_count - 1);
1883 if (!bh) {
1884 ext4_warning(sb, "can't read last block, resize aborted");
1885 return -ENOSPC;
1886 }
1887 brelse(bh);
1888
1889retry:
1890 o_blocks_count = ext4_blocks_count(es);
1891
1892 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1893 "to %llu blocks", o_blocks_count, n_blocks_count);
1894
1895 if (n_blocks_count < o_blocks_count) {
1896 /* On-line shrinking not supported */
1897 ext4_warning(sb, "can't shrink FS - resize aborted");
1898 return -EINVAL;
1899 }
1900
1901 if (n_blocks_count == o_blocks_count)
1902 /* Nothing need to do */
1903 return 0;
1904
1905 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1906 if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1907 ext4_warning(sb, "resize would cause inodes_count overflow");
1908 return -EINVAL;
1909 }
1910 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1911
1912 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1913 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1914
1915 meta_bg = ext4_has_feature_meta_bg(sb);
1916
1917 if (ext4_has_feature_resize_inode(sb)) {
1918 if (meta_bg) {
1919 ext4_error(sb, "resize_inode and meta_bg enabled "
1920 "simultaneously");
1921 return -EINVAL;
1922 }
1923 if (n_desc_blocks > o_desc_blocks +
1924 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1925 n_blocks_count_retry = n_blocks_count;
1926 n_desc_blocks = o_desc_blocks +
1927 le16_to_cpu(es->s_reserved_gdt_blocks);
1928 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1929 n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1930 n_group--; /* set to last group number */
1931 }
1932
1933 if (!resize_inode)
1934 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1935 if (IS_ERR(resize_inode)) {
1936 ext4_warning(sb, "Error opening resize inode");
1937 return PTR_ERR(resize_inode);
1938 }
1939 }
1940
1941 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1942 err = ext4_convert_meta_bg(sb, resize_inode);
1943 if (err)
1944 goto out;
1945 if (resize_inode) {
1946 iput(resize_inode);
1947 resize_inode = NULL;
1948 }
1949 if (n_blocks_count_retry) {
1950 n_blocks_count = n_blocks_count_retry;
1951 n_blocks_count_retry = 0;
1952 goto retry;
1953 }
1954 }
1955
1956 /* extend the last group */
1957 if (n_group == o_group)
1958 add = n_blocks_count - o_blocks_count;
1959 else
1960 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1961 if (add > 0) {
1962 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1963 if (err)
1964 goto out;
1965 }
1966
1967 if (ext4_blocks_count(es) == n_blocks_count)
1968 goto out;
1969
1970 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1971 if (err)
1972 return err;
1973
1974 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1975 if (err)
1976 goto out;
1977
1978 flex_gd = alloc_flex_gd(flexbg_size);
1979 if (flex_gd == NULL) {
1980 err = -ENOMEM;
1981 goto out;
1982 }
1983
1984 /* Add flex groups. Note that a regular group is a
1985 * flex group with 1 group.
1986 */
1987 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1988 flexbg_size)) {
1989 if (jiffies - last_update_time > HZ * 10) {
1990 if (last_update_time)
1991 ext4_msg(sb, KERN_INFO,
1992 "resized to %llu blocks",
1993 ext4_blocks_count(es));
1994 last_update_time = jiffies;
1995 }
1996 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1997 break;
1998 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1999 if (unlikely(err))
2000 break;
2001 }
2002
2003 if (!err && n_blocks_count_retry) {
2004 n_blocks_count = n_blocks_count_retry;
2005 n_blocks_count_retry = 0;
2006 free_flex_gd(flex_gd);
2007 flex_gd = NULL;
2008 goto retry;
2009 }
2010
2011out:
2012 if (flex_gd)
2013 free_flex_gd(flex_gd);
2014 if (resize_inode != NULL)
2015 iput(resize_inode);
2016 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2017 return err;
2018}