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 * 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
48static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
49 ext4_group_t group) {
50 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
51 EXT4_DESC_PER_BLOCK_BITS(sb);
52}
53
54static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
55 ext4_group_t group) {
56 group = ext4_meta_bg_first_group(sb, group);
57 return ext4_group_first_block_no(sb, group);
58}
59
60static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
61 ext4_group_t group) {
62 ext4_grpblk_t overhead;
63 overhead = ext4_bg_num_gdb(sb, group);
64 if (ext4_bg_has_super(sb, group))
65 overhead += 1 +
66 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
67 return overhead;
68}
69
70#define outside(b, first, last) ((b) < (first) || (b) >= (last))
71#define inside(b, first, last) ((b) >= (first) && (b) < (last))
72
73static int verify_group_input(struct super_block *sb,
74 struct ext4_new_group_data *input)
75{
76 struct ext4_sb_info *sbi = EXT4_SB(sb);
77 struct ext4_super_block *es = sbi->s_es;
78 ext4_fsblk_t start = ext4_blocks_count(es);
79 ext4_fsblk_t end = start + input->blocks_count;
80 ext4_group_t group = input->group;
81 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
82 unsigned overhead;
83 ext4_fsblk_t metaend;
84 struct buffer_head *bh = NULL;
85 ext4_grpblk_t free_blocks_count, offset;
86 int err = -EINVAL;
87
88 if (group != sbi->s_groups_count) {
89 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
90 input->group, sbi->s_groups_count);
91 return -EINVAL;
92 }
93
94 overhead = ext4_group_overhead_blocks(sb, group);
95 metaend = start + overhead;
96 input->free_blocks_count = free_blocks_count =
97 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
98
99 if (test_opt(sb, DEBUG))
100 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
101 "(%d free, %u reserved)\n",
102 ext4_bg_has_super(sb, input->group) ? "normal" :
103 "no-super", input->group, input->blocks_count,
104 free_blocks_count, input->reserved_blocks);
105
106 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
107 if (offset != 0)
108 ext4_warning(sb, "Last group not full");
109 else if (input->reserved_blocks > input->blocks_count / 5)
110 ext4_warning(sb, "Reserved blocks too high (%u)",
111 input->reserved_blocks);
112 else if (free_blocks_count < 0)
113 ext4_warning(sb, "Bad blocks count %u",
114 input->blocks_count);
115 else if (!(bh = sb_bread(sb, end - 1)))
116 ext4_warning(sb, "Cannot read last block (%llu)",
117 end - 1);
118 else if (outside(input->block_bitmap, start, end))
119 ext4_warning(sb, "Block bitmap not in group (block %llu)",
120 (unsigned long long)input->block_bitmap);
121 else if (outside(input->inode_bitmap, start, end))
122 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
123 (unsigned long long)input->inode_bitmap);
124 else if (outside(input->inode_table, start, end) ||
125 outside(itend - 1, start, end))
126 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
127 (unsigned long long)input->inode_table, itend - 1);
128 else if (input->inode_bitmap == input->block_bitmap)
129 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
130 (unsigned long long)input->block_bitmap);
131 else if (inside(input->block_bitmap, input->inode_table, itend))
132 ext4_warning(sb, "Block bitmap (%llu) in inode table "
133 "(%llu-%llu)",
134 (unsigned long long)input->block_bitmap,
135 (unsigned long long)input->inode_table, itend - 1);
136 else if (inside(input->inode_bitmap, input->inode_table, itend))
137 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
138 "(%llu-%llu)",
139 (unsigned long long)input->inode_bitmap,
140 (unsigned long long)input->inode_table, itend - 1);
141 else if (inside(input->block_bitmap, start, metaend))
142 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
143 (unsigned long long)input->block_bitmap,
144 start, metaend - 1);
145 else if (inside(input->inode_bitmap, start, metaend))
146 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
147 (unsigned long long)input->inode_bitmap,
148 start, metaend - 1);
149 else if (inside(input->inode_table, start, metaend) ||
150 inside(itend - 1, start, metaend))
151 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
152 "(%llu-%llu)",
153 (unsigned long long)input->inode_table,
154 itend - 1, start, metaend - 1);
155 else
156 err = 0;
157 brelse(bh);
158
159 return err;
160}
161
162/*
163 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
164 * group each time.
165 */
166struct ext4_new_flex_group_data {
167 struct ext4_new_group_data *groups; /* new_group_data for groups
168 in the flex group */
169 __u16 *bg_flags; /* block group flags of groups
170 in @groups */
171 ext4_group_t count; /* number of groups in @groups
172 */
173};
174
175/*
176 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
177 * @flexbg_size.
178 *
179 * Returns NULL on failure otherwise address of the allocated structure.
180 */
181static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
182{
183 struct ext4_new_flex_group_data *flex_gd;
184
185 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
186 if (flex_gd == NULL)
187 goto out3;
188
189 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
190 goto out2;
191 flex_gd->count = flexbg_size;
192
193 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
194 flexbg_size, GFP_NOFS);
195 if (flex_gd->groups == NULL)
196 goto out2;
197
198 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
199 if (flex_gd->bg_flags == NULL)
200 goto out1;
201
202 return flex_gd;
203
204out1:
205 kfree(flex_gd->groups);
206out2:
207 kfree(flex_gd);
208out3:
209 return NULL;
210}
211
212static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
213{
214 kfree(flex_gd->bg_flags);
215 kfree(flex_gd->groups);
216 kfree(flex_gd);
217}
218
219/*
220 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
221 * and inode tables for a flex group.
222 *
223 * This function is used by 64bit-resize. Note that this function allocates
224 * group tables from the 1st group of groups contained by @flexgd, which may
225 * be a partial of a flex group.
226 *
227 * @sb: super block of fs to which the groups belongs
228 *
229 * Returns 0 on a successful allocation of the metadata blocks in the
230 * block group.
231 */
232static int ext4_alloc_group_tables(struct super_block *sb,
233 struct ext4_new_flex_group_data *flex_gd,
234 int flexbg_size)
235{
236 struct ext4_new_group_data *group_data = flex_gd->groups;
237 ext4_fsblk_t start_blk;
238 ext4_fsblk_t last_blk;
239 ext4_group_t src_group;
240 ext4_group_t bb_index = 0;
241 ext4_group_t ib_index = 0;
242 ext4_group_t it_index = 0;
243 ext4_group_t group;
244 ext4_group_t last_group;
245 unsigned overhead;
246 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
247
248 BUG_ON(flex_gd->count == 0 || group_data == NULL);
249
250 src_group = group_data[0].group;
251 last_group = src_group + flex_gd->count - 1;
252
253 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
254 (last_group & ~(flexbg_size - 1))));
255next_group:
256 group = group_data[0].group;
257 if (src_group >= group_data[0].group + flex_gd->count)
258 return -ENOSPC;
259 start_blk = ext4_group_first_block_no(sb, src_group);
260 last_blk = start_blk + group_data[src_group - group].blocks_count;
261
262 overhead = ext4_group_overhead_blocks(sb, src_group);
263
264 start_blk += overhead;
265
266 /* We collect contiguous blocks as much as possible. */
267 src_group++;
268 for (; src_group <= last_group; src_group++) {
269 overhead = ext4_group_overhead_blocks(sb, src_group);
270 if (overhead == 0)
271 last_blk += group_data[src_group - group].blocks_count;
272 else
273 break;
274 }
275
276 /* Allocate block bitmaps */
277 for (; bb_index < flex_gd->count; bb_index++) {
278 if (start_blk >= last_blk)
279 goto next_group;
280 group_data[bb_index].block_bitmap = start_blk++;
281 group = ext4_get_group_number(sb, start_blk - 1);
282 group -= group_data[0].group;
283 group_data[group].free_blocks_count--;
284 flex_gd->bg_flags[group] &= uninit_mask;
285 }
286
287 /* Allocate inode bitmaps */
288 for (; ib_index < flex_gd->count; ib_index++) {
289 if (start_blk >= last_blk)
290 goto next_group;
291 group_data[ib_index].inode_bitmap = start_blk++;
292 group = ext4_get_group_number(sb, start_blk - 1);
293 group -= group_data[0].group;
294 group_data[group].free_blocks_count--;
295 flex_gd->bg_flags[group] &= uninit_mask;
296 }
297
298 /* Allocate inode tables */
299 for (; it_index < flex_gd->count; it_index++) {
300 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
301 ext4_fsblk_t next_group_start;
302
303 if (start_blk + itb > last_blk)
304 goto next_group;
305 group_data[it_index].inode_table = start_blk;
306 group = ext4_get_group_number(sb, start_blk);
307 next_group_start = ext4_group_first_block_no(sb, group + 1);
308 group -= group_data[0].group;
309
310 if (start_blk + itb > next_group_start) {
311 flex_gd->bg_flags[group + 1] &= uninit_mask;
312 overhead = start_blk + itb - next_group_start;
313 group_data[group + 1].free_blocks_count -= overhead;
314 itb -= overhead;
315 }
316
317 group_data[group].free_blocks_count -= itb;
318 flex_gd->bg_flags[group] &= uninit_mask;
319 start_blk += EXT4_SB(sb)->s_itb_per_group;
320 }
321
322 if (test_opt(sb, DEBUG)) {
323 int i;
324 group = group_data[0].group;
325
326 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
327 "%d groups, flexbg size is %d:\n", flex_gd->count,
328 flexbg_size);
329
330 for (i = 0; i < flex_gd->count; i++) {
331 printk(KERN_DEBUG "adding %s group %u: %u "
332 "blocks (%d free)\n",
333 ext4_bg_has_super(sb, group + i) ? "normal" :
334 "no-super", group + i,
335 group_data[i].blocks_count,
336 group_data[i].free_blocks_count);
337 }
338 }
339 return 0;
340}
341
342static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
343 ext4_fsblk_t blk)
344{
345 struct buffer_head *bh;
346 int err;
347
348 bh = sb_getblk(sb, blk);
349 if (unlikely(!bh))
350 return ERR_PTR(-ENOMEM);
351 if ((err = ext4_journal_get_write_access(handle, bh))) {
352 brelse(bh);
353 bh = ERR_PTR(err);
354 } else {
355 memset(bh->b_data, 0, sb->s_blocksize);
356 set_buffer_uptodate(bh);
357 }
358
359 return bh;
360}
361
362/*
363 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
364 * If that fails, restart the transaction & regain write access for the
365 * buffer head which is used for block_bitmap modifications.
366 */
367static int extend_or_restart_transaction(handle_t *handle, int thresh)
368{
369 int err;
370
371 if (ext4_handle_has_enough_credits(handle, thresh))
372 return 0;
373
374 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
375 if (err < 0)
376 return err;
377 if (err) {
378 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
379 if (err)
380 return err;
381 }
382
383 return 0;
384}
385
386/*
387 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
388 *
389 * Helper function for ext4_setup_new_group_blocks() which set .
390 *
391 * @sb: super block
392 * @handle: journal handle
393 * @flex_gd: flex group data
394 */
395static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
396 struct ext4_new_flex_group_data *flex_gd,
397 ext4_fsblk_t block, ext4_group_t count)
398{
399 ext4_group_t count2;
400
401 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
402 for (count2 = count; count > 0; count -= count2, block += count2) {
403 ext4_fsblk_t start;
404 struct buffer_head *bh;
405 ext4_group_t group;
406 int err;
407
408 group = ext4_get_group_number(sb, block);
409 start = ext4_group_first_block_no(sb, group);
410 group -= flex_gd->groups[0].group;
411
412 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
413 if (count2 > count)
414 count2 = count;
415
416 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
417 BUG_ON(flex_gd->count > 1);
418 continue;
419 }
420
421 err = extend_or_restart_transaction(handle, 1);
422 if (err)
423 return err;
424
425 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
426 if (unlikely(!bh))
427 return -ENOMEM;
428
429 err = ext4_journal_get_write_access(handle, bh);
430 if (err)
431 return err;
432 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
433 block - start, count2);
434 ext4_set_bits(bh->b_data, block - start, count2);
435
436 err = ext4_handle_dirty_metadata(handle, NULL, bh);
437 if (unlikely(err))
438 return err;
439 brelse(bh);
440 }
441
442 return 0;
443}
444
445/*
446 * Set up the block and inode bitmaps, and the inode table for the new groups.
447 * This doesn't need to be part of the main transaction, since we are only
448 * changing blocks outside the actual filesystem. We still do journaling to
449 * ensure the recovery is correct in case of a failure just after resize.
450 * If any part of this fails, we simply abort the resize.
451 *
452 * setup_new_flex_group_blocks handles a flex group as follow:
453 * 1. copy super block and GDT, and initialize group tables if necessary.
454 * In this step, we only set bits in blocks bitmaps for blocks taken by
455 * super block and GDT.
456 * 2. allocate group tables in block bitmaps, that is, set bits in block
457 * bitmap for blocks taken by group tables.
458 */
459static int setup_new_flex_group_blocks(struct super_block *sb,
460 struct ext4_new_flex_group_data *flex_gd)
461{
462 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
463 ext4_fsblk_t start;
464 ext4_fsblk_t block;
465 struct ext4_sb_info *sbi = EXT4_SB(sb);
466 struct ext4_super_block *es = sbi->s_es;
467 struct ext4_new_group_data *group_data = flex_gd->groups;
468 __u16 *bg_flags = flex_gd->bg_flags;
469 handle_t *handle;
470 ext4_group_t group, count;
471 struct buffer_head *bh = NULL;
472 int reserved_gdb, i, j, err = 0, err2;
473 int meta_bg;
474
475 BUG_ON(!flex_gd->count || !group_data ||
476 group_data[0].group != sbi->s_groups_count);
477
478 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
479 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
480
481 /* This transaction may be extended/restarted along the way */
482 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
483 if (IS_ERR(handle))
484 return PTR_ERR(handle);
485
486 group = group_data[0].group;
487 for (i = 0; i < flex_gd->count; i++, group++) {
488 unsigned long gdblocks;
489 ext4_grpblk_t overhead;
490
491 gdblocks = ext4_bg_num_gdb(sb, group);
492 start = ext4_group_first_block_no(sb, group);
493
494 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
495 goto handle_itb;
496
497 if (meta_bg == 1) {
498 ext4_group_t first_group;
499 first_group = ext4_meta_bg_first_group(sb, group);
500 if (first_group != group + 1 &&
501 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
502 goto handle_itb;
503 }
504
505 block = start + ext4_bg_has_super(sb, group);
506 /* Copy all of the GDT blocks into the backup in this group */
507 for (j = 0; j < gdblocks; j++, block++) {
508 struct buffer_head *gdb;
509
510 ext4_debug("update backup group %#04llx\n", block);
511 err = extend_or_restart_transaction(handle, 1);
512 if (err)
513 goto out;
514
515 gdb = sb_getblk(sb, block);
516 if (unlikely(!gdb)) {
517 err = -ENOMEM;
518 goto out;
519 }
520
521 err = ext4_journal_get_write_access(handle, gdb);
522 if (err) {
523 brelse(gdb);
524 goto out;
525 }
526 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
527 gdb->b_size);
528 set_buffer_uptodate(gdb);
529
530 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
531 if (unlikely(err)) {
532 brelse(gdb);
533 goto out;
534 }
535 brelse(gdb);
536 }
537
538 /* Zero out all of the reserved backup group descriptor
539 * table blocks
540 */
541 if (ext4_bg_has_super(sb, group)) {
542 err = sb_issue_zeroout(sb, gdblocks + start + 1,
543 reserved_gdb, GFP_NOFS);
544 if (err)
545 goto out;
546 }
547
548handle_itb:
549 /* Initialize group tables of the grop @group */
550 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
551 goto handle_bb;
552
553 /* Zero out all of the inode table blocks */
554 block = group_data[i].inode_table;
555 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
556 block, sbi->s_itb_per_group);
557 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
558 GFP_NOFS);
559 if (err)
560 goto out;
561
562handle_bb:
563 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
564 goto handle_ib;
565
566 /* Initialize block bitmap of the @group */
567 block = group_data[i].block_bitmap;
568 err = extend_or_restart_transaction(handle, 1);
569 if (err)
570 goto out;
571
572 bh = bclean(handle, sb, block);
573 if (IS_ERR(bh)) {
574 err = PTR_ERR(bh);
575 goto out;
576 }
577 overhead = ext4_group_overhead_blocks(sb, group);
578 if (overhead != 0) {
579 ext4_debug("mark backup superblock %#04llx (+0)\n",
580 start);
581 ext4_set_bits(bh->b_data, 0, overhead);
582 }
583 ext4_mark_bitmap_end(group_data[i].blocks_count,
584 sb->s_blocksize * 8, bh->b_data);
585 err = ext4_handle_dirty_metadata(handle, NULL, bh);
586 if (err)
587 goto out;
588 brelse(bh);
589
590handle_ib:
591 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
592 continue;
593
594 /* Initialize inode bitmap of the @group */
595 block = group_data[i].inode_bitmap;
596 err = extend_or_restart_transaction(handle, 1);
597 if (err)
598 goto out;
599 /* Mark unused entries in inode bitmap used */
600 bh = bclean(handle, sb, block);
601 if (IS_ERR(bh)) {
602 err = PTR_ERR(bh);
603 goto out;
604 }
605
606 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
607 sb->s_blocksize * 8, bh->b_data);
608 err = ext4_handle_dirty_metadata(handle, NULL, bh);
609 if (err)
610 goto out;
611 brelse(bh);
612 }
613 bh = NULL;
614
615 /* Mark group tables in block bitmap */
616 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
617 count = group_table_count[j];
618 start = (&group_data[0].block_bitmap)[j];
619 block = start;
620 for (i = 1; i < flex_gd->count; i++) {
621 block += group_table_count[j];
622 if (block == (&group_data[i].block_bitmap)[j]) {
623 count += group_table_count[j];
624 continue;
625 }
626 err = set_flexbg_block_bitmap(sb, handle,
627 flex_gd, start, count);
628 if (err)
629 goto out;
630 count = group_table_count[j];
631 start = (&group_data[i].block_bitmap)[j];
632 block = start;
633 }
634
635 if (count) {
636 err = set_flexbg_block_bitmap(sb, handle,
637 flex_gd, start, count);
638 if (err)
639 goto out;
640 }
641 }
642
643out:
644 brelse(bh);
645 err2 = ext4_journal_stop(handle);
646 if (err2 && !err)
647 err = err2;
648
649 return err;
650}
651
652/*
653 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
654 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
655 * calling this for the first time. In a sparse filesystem it will be the
656 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
657 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
658 */
659static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
660 unsigned *five, unsigned *seven)
661{
662 unsigned *min = three;
663 int mult = 3;
664 unsigned ret;
665
666 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
667 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
668 ret = *min;
669 *min += 1;
670 return ret;
671 }
672
673 if (*five < *min) {
674 min = five;
675 mult = 5;
676 }
677 if (*seven < *min) {
678 min = seven;
679 mult = 7;
680 }
681
682 ret = *min;
683 *min *= mult;
684
685 return ret;
686}
687
688/*
689 * Check that all of the backup GDT blocks are held in the primary GDT block.
690 * It is assumed that they are stored in group order. Returns the number of
691 * groups in current filesystem that have BACKUPS, or -ve error code.
692 */
693static int verify_reserved_gdb(struct super_block *sb,
694 ext4_group_t end,
695 struct buffer_head *primary)
696{
697 const ext4_fsblk_t blk = primary->b_blocknr;
698 unsigned three = 1;
699 unsigned five = 5;
700 unsigned seven = 7;
701 unsigned grp;
702 __le32 *p = (__le32 *)primary->b_data;
703 int gdbackups = 0;
704
705 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
706 if (le32_to_cpu(*p++) !=
707 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
708 ext4_warning(sb, "reserved GDT %llu"
709 " missing grp %d (%llu)",
710 blk, grp,
711 grp *
712 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
713 blk);
714 return -EINVAL;
715 }
716 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
717 return -EFBIG;
718 }
719
720 return gdbackups;
721}
722
723/*
724 * Called when we need to bring a reserved group descriptor table block into
725 * use from the resize inode. The primary copy of the new GDT block currently
726 * is an indirect block (under the double indirect block in the resize inode).
727 * The new backup GDT blocks will be stored as leaf blocks in this indirect
728 * block, in group order. Even though we know all the block numbers we need,
729 * we check to ensure that the resize inode has actually reserved these blocks.
730 *
731 * Don't need to update the block bitmaps because the blocks are still in use.
732 *
733 * We get all of the error cases out of the way, so that we are sure to not
734 * fail once we start modifying the data on disk, because JBD has no rollback.
735 */
736static int add_new_gdb(handle_t *handle, struct inode *inode,
737 ext4_group_t group)
738{
739 struct super_block *sb = inode->i_sb;
740 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
741 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
742 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
743 struct buffer_head **o_group_desc, **n_group_desc;
744 struct buffer_head *dind;
745 struct buffer_head *gdb_bh;
746 int gdbackups;
747 struct ext4_iloc iloc;
748 __le32 *data;
749 int err;
750
751 if (test_opt(sb, DEBUG))
752 printk(KERN_DEBUG
753 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
754 gdb_num);
755
756 /*
757 * If we are not using the primary superblock/GDT copy don't resize,
758 * because the user tools have no way of handling this. Probably a
759 * bad time to do it anyways.
760 */
761 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
762 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
763 ext4_warning(sb, "won't resize using backup superblock at %llu",
764 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
765 return -EPERM;
766 }
767
768 gdb_bh = sb_bread(sb, gdblock);
769 if (!gdb_bh)
770 return -EIO;
771
772 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
773 if (gdbackups < 0) {
774 err = gdbackups;
775 goto exit_bh;
776 }
777
778 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
779 dind = sb_bread(sb, le32_to_cpu(*data));
780 if (!dind) {
781 err = -EIO;
782 goto exit_bh;
783 }
784
785 data = (__le32 *)dind->b_data;
786 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
787 ext4_warning(sb, "new group %u GDT block %llu not reserved",
788 group, gdblock);
789 err = -EINVAL;
790 goto exit_dind;
791 }
792
793 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
794 if (unlikely(err))
795 goto exit_dind;
796
797 err = ext4_journal_get_write_access(handle, gdb_bh);
798 if (unlikely(err))
799 goto exit_dind;
800
801 err = ext4_journal_get_write_access(handle, dind);
802 if (unlikely(err))
803 ext4_std_error(sb, err);
804
805 /* ext4_reserve_inode_write() gets a reference on the iloc */
806 err = ext4_reserve_inode_write(handle, inode, &iloc);
807 if (unlikely(err))
808 goto exit_dind;
809
810 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
811 sizeof(struct buffer_head *),
812 GFP_NOFS);
813 if (!n_group_desc) {
814 err = -ENOMEM;
815 ext4_warning(sb, "not enough memory for %lu groups",
816 gdb_num + 1);
817 goto exit_inode;
818 }
819
820 /*
821 * Finally, we have all of the possible failures behind us...
822 *
823 * Remove new GDT block from inode double-indirect block and clear out
824 * the new GDT block for use (which also "frees" the backup GDT blocks
825 * from the reserved inode). We don't need to change the bitmaps for
826 * these blocks, because they are marked as in-use from being in the
827 * reserved inode, and will become GDT blocks (primary and backup).
828 */
829 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
830 err = ext4_handle_dirty_metadata(handle, NULL, dind);
831 if (unlikely(err)) {
832 ext4_std_error(sb, err);
833 goto exit_inode;
834 }
835 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
836 ext4_mark_iloc_dirty(handle, inode, &iloc);
837 memset(gdb_bh->b_data, 0, sb->s_blocksize);
838 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
839 if (unlikely(err)) {
840 ext4_std_error(sb, err);
841 goto exit_inode;
842 }
843 brelse(dind);
844
845 o_group_desc = EXT4_SB(sb)->s_group_desc;
846 memcpy(n_group_desc, o_group_desc,
847 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
848 n_group_desc[gdb_num] = gdb_bh;
849 EXT4_SB(sb)->s_group_desc = n_group_desc;
850 EXT4_SB(sb)->s_gdb_count++;
851 ext4_kvfree(o_group_desc);
852
853 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
854 err = ext4_handle_dirty_super(handle, sb);
855 if (err)
856 ext4_std_error(sb, err);
857
858 return err;
859
860exit_inode:
861 ext4_kvfree(n_group_desc);
862 brelse(iloc.bh);
863exit_dind:
864 brelse(dind);
865exit_bh:
866 brelse(gdb_bh);
867
868 ext4_debug("leaving with error %d\n", err);
869 return err;
870}
871
872/*
873 * add_new_gdb_meta_bg is the sister of add_new_gdb.
874 */
875static int add_new_gdb_meta_bg(struct super_block *sb,
876 handle_t *handle, ext4_group_t group) {
877 ext4_fsblk_t gdblock;
878 struct buffer_head *gdb_bh;
879 struct buffer_head **o_group_desc, **n_group_desc;
880 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
881 int err;
882
883 gdblock = ext4_meta_bg_first_block_no(sb, group) +
884 ext4_bg_has_super(sb, group);
885 gdb_bh = sb_bread(sb, gdblock);
886 if (!gdb_bh)
887 return -EIO;
888 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
889 sizeof(struct buffer_head *),
890 GFP_NOFS);
891 if (!n_group_desc) {
892 err = -ENOMEM;
893 ext4_warning(sb, "not enough memory for %lu groups",
894 gdb_num + 1);
895 return err;
896 }
897
898 o_group_desc = EXT4_SB(sb)->s_group_desc;
899 memcpy(n_group_desc, o_group_desc,
900 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
901 n_group_desc[gdb_num] = gdb_bh;
902 EXT4_SB(sb)->s_group_desc = n_group_desc;
903 EXT4_SB(sb)->s_gdb_count++;
904 ext4_kvfree(o_group_desc);
905 err = ext4_journal_get_write_access(handle, gdb_bh);
906 if (unlikely(err))
907 brelse(gdb_bh);
908 return err;
909}
910
911/*
912 * Called when we are adding a new group which has a backup copy of each of
913 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
914 * We need to add these reserved backup GDT blocks to the resize inode, so
915 * that they are kept for future resizing and not allocated to files.
916 *
917 * Each reserved backup GDT block will go into a different indirect block.
918 * The indirect blocks are actually the primary reserved GDT blocks,
919 * so we know in advance what their block numbers are. We only get the
920 * double-indirect block to verify it is pointing to the primary reserved
921 * GDT blocks so we don't overwrite a data block by accident. The reserved
922 * backup GDT blocks are stored in their reserved primary GDT block.
923 */
924static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
925 ext4_group_t group)
926{
927 struct super_block *sb = inode->i_sb;
928 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
929 struct buffer_head **primary;
930 struct buffer_head *dind;
931 struct ext4_iloc iloc;
932 ext4_fsblk_t blk;
933 __le32 *data, *end;
934 int gdbackups = 0;
935 int res, i;
936 int err;
937
938 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
939 if (!primary)
940 return -ENOMEM;
941
942 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
943 dind = sb_bread(sb, le32_to_cpu(*data));
944 if (!dind) {
945 err = -EIO;
946 goto exit_free;
947 }
948
949 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
950 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
951 EXT4_ADDR_PER_BLOCK(sb));
952 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
953
954 /* Get each reserved primary GDT block and verify it holds backups */
955 for (res = 0; res < reserved_gdb; res++, blk++) {
956 if (le32_to_cpu(*data) != blk) {
957 ext4_warning(sb, "reserved block %llu"
958 " not at offset %ld",
959 blk,
960 (long)(data - (__le32 *)dind->b_data));
961 err = -EINVAL;
962 goto exit_bh;
963 }
964 primary[res] = sb_bread(sb, blk);
965 if (!primary[res]) {
966 err = -EIO;
967 goto exit_bh;
968 }
969 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
970 if (gdbackups < 0) {
971 brelse(primary[res]);
972 err = gdbackups;
973 goto exit_bh;
974 }
975 if (++data >= end)
976 data = (__le32 *)dind->b_data;
977 }
978
979 for (i = 0; i < reserved_gdb; i++) {
980 if ((err = ext4_journal_get_write_access(handle, primary[i])))
981 goto exit_bh;
982 }
983
984 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
985 goto exit_bh;
986
987 /*
988 * Finally we can add each of the reserved backup GDT blocks from
989 * the new group to its reserved primary GDT block.
990 */
991 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
992 for (i = 0; i < reserved_gdb; i++) {
993 int err2;
994 data = (__le32 *)primary[i]->b_data;
995 /* printk("reserving backup %lu[%u] = %lu\n",
996 primary[i]->b_blocknr, gdbackups,
997 blk + primary[i]->b_blocknr); */
998 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
999 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1000 if (!err)
1001 err = err2;
1002 }
1003 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1004 ext4_mark_iloc_dirty(handle, inode, &iloc);
1005
1006exit_bh:
1007 while (--res >= 0)
1008 brelse(primary[res]);
1009 brelse(dind);
1010
1011exit_free:
1012 kfree(primary);
1013
1014 return err;
1015}
1016
1017/*
1018 * Update the backup copies of the ext4 metadata. These don't need to be part
1019 * of the main resize transaction, because e2fsck will re-write them if there
1020 * is a problem (basically only OOM will cause a problem). However, we
1021 * _should_ update the backups if possible, in case the primary gets trashed
1022 * for some reason and we need to run e2fsck from a backup superblock. The
1023 * important part is that the new block and inode counts are in the backup
1024 * superblocks, and the location of the new group metadata in the GDT backups.
1025 *
1026 * We do not need take the s_resize_lock for this, because these
1027 * blocks are not otherwise touched by the filesystem code when it is
1028 * mounted. We don't need to worry about last changing from
1029 * sbi->s_groups_count, because the worst that can happen is that we
1030 * do not copy the full number of backups at this time. The resize
1031 * which changed s_groups_count will backup again.
1032 */
1033static void update_backups(struct super_block *sb, int blk_off, char *data,
1034 int size, int meta_bg)
1035{
1036 struct ext4_sb_info *sbi = EXT4_SB(sb);
1037 ext4_group_t last;
1038 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1039 unsigned three = 1;
1040 unsigned five = 5;
1041 unsigned seven = 7;
1042 ext4_group_t group = 0;
1043 int rest = sb->s_blocksize - size;
1044 handle_t *handle;
1045 int err = 0, err2;
1046
1047 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1048 if (IS_ERR(handle)) {
1049 group = 1;
1050 err = PTR_ERR(handle);
1051 goto exit_err;
1052 }
1053
1054 if (meta_bg == 0) {
1055 group = ext4_list_backups(sb, &three, &five, &seven);
1056 last = sbi->s_groups_count;
1057 } else {
1058 group = ext4_meta_bg_first_group(sb, group) + 1;
1059 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1060 }
1061
1062 while (group < sbi->s_groups_count) {
1063 struct buffer_head *bh;
1064 ext4_fsblk_t backup_block;
1065
1066 /* Out of journal space, and can't get more - abort - so sad */
1067 if (ext4_handle_valid(handle) &&
1068 handle->h_buffer_credits == 0 &&
1069 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1070 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1071 break;
1072
1073 if (meta_bg == 0)
1074 backup_block = group * bpg + blk_off;
1075 else
1076 backup_block = (ext4_group_first_block_no(sb, group) +
1077 ext4_bg_has_super(sb, group));
1078
1079 bh = sb_getblk(sb, backup_block);
1080 if (unlikely(!bh)) {
1081 err = -ENOMEM;
1082 break;
1083 }
1084 ext4_debug("update metadata backup %llu(+%llu)\n",
1085 backup_block, backup_block -
1086 ext4_group_first_block_no(sb, group));
1087 if ((err = ext4_journal_get_write_access(handle, bh)))
1088 break;
1089 lock_buffer(bh);
1090 memcpy(bh->b_data, data, size);
1091 if (rest)
1092 memset(bh->b_data + size, 0, rest);
1093 set_buffer_uptodate(bh);
1094 unlock_buffer(bh);
1095 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1096 if (unlikely(err))
1097 ext4_std_error(sb, err);
1098 brelse(bh);
1099
1100 if (meta_bg == 0)
1101 group = ext4_list_backups(sb, &three, &five, &seven);
1102 else if (group == last)
1103 break;
1104 else
1105 group = last;
1106 }
1107 if ((err2 = ext4_journal_stop(handle)) && !err)
1108 err = err2;
1109
1110 /*
1111 * Ugh! Need to have e2fsck write the backup copies. It is too
1112 * late to revert the resize, we shouldn't fail just because of
1113 * the backup copies (they are only needed in case of corruption).
1114 *
1115 * However, if we got here we have a journal problem too, so we
1116 * can't really start a transaction to mark the superblock.
1117 * Chicken out and just set the flag on the hope it will be written
1118 * to disk, and if not - we will simply wait until next fsck.
1119 */
1120exit_err:
1121 if (err) {
1122 ext4_warning(sb, "can't update backup for group %u (err %d), "
1123 "forcing fsck on next reboot", group, err);
1124 sbi->s_mount_state &= ~EXT4_VALID_FS;
1125 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1126 mark_buffer_dirty(sbi->s_sbh);
1127 }
1128}
1129
1130/*
1131 * ext4_add_new_descs() adds @count group descriptor of groups
1132 * starting at @group
1133 *
1134 * @handle: journal handle
1135 * @sb: super block
1136 * @group: the group no. of the first group desc to be added
1137 * @resize_inode: the resize inode
1138 * @count: number of group descriptors to be added
1139 */
1140static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1141 ext4_group_t group, struct inode *resize_inode,
1142 ext4_group_t count)
1143{
1144 struct ext4_sb_info *sbi = EXT4_SB(sb);
1145 struct ext4_super_block *es = sbi->s_es;
1146 struct buffer_head *gdb_bh;
1147 int i, gdb_off, gdb_num, err = 0;
1148 int meta_bg;
1149
1150 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1151 for (i = 0; i < count; i++, group++) {
1152 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1153 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1154
1155 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1156 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1157
1158 /*
1159 * We will only either add reserved group blocks to a backup group
1160 * or remove reserved blocks for the first group in a new group block.
1161 * Doing both would be mean more complex code, and sane people don't
1162 * use non-sparse filesystems anymore. This is already checked above.
1163 */
1164 if (gdb_off) {
1165 gdb_bh = sbi->s_group_desc[gdb_num];
1166 err = ext4_journal_get_write_access(handle, gdb_bh);
1167
1168 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1169 err = reserve_backup_gdb(handle, resize_inode, group);
1170 } else if (meta_bg != 0) {
1171 err = add_new_gdb_meta_bg(sb, handle, group);
1172 } else {
1173 err = add_new_gdb(handle, resize_inode, group);
1174 }
1175 if (err)
1176 break;
1177 }
1178 return err;
1179}
1180
1181static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1182{
1183 struct buffer_head *bh = sb_getblk(sb, block);
1184 if (unlikely(!bh))
1185 return NULL;
1186 if (!bh_uptodate_or_lock(bh)) {
1187 if (bh_submit_read(bh) < 0) {
1188 brelse(bh);
1189 return NULL;
1190 }
1191 }
1192
1193 return bh;
1194}
1195
1196static int ext4_set_bitmap_checksums(struct super_block *sb,
1197 ext4_group_t group,
1198 struct ext4_group_desc *gdp,
1199 struct ext4_new_group_data *group_data)
1200{
1201 struct buffer_head *bh;
1202
1203 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1204 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1205 return 0;
1206
1207 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1208 if (!bh)
1209 return -EIO;
1210 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1211 EXT4_INODES_PER_GROUP(sb) / 8);
1212 brelse(bh);
1213
1214 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1215 if (!bh)
1216 return -EIO;
1217 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1218 brelse(bh);
1219
1220 return 0;
1221}
1222
1223/*
1224 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1225 */
1226static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1227 struct ext4_new_flex_group_data *flex_gd)
1228{
1229 struct ext4_new_group_data *group_data = flex_gd->groups;
1230 struct ext4_group_desc *gdp;
1231 struct ext4_sb_info *sbi = EXT4_SB(sb);
1232 struct buffer_head *gdb_bh;
1233 ext4_group_t group;
1234 __u16 *bg_flags = flex_gd->bg_flags;
1235 int i, gdb_off, gdb_num, err = 0;
1236
1237
1238 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1239 group = group_data->group;
1240
1241 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1242 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1243
1244 /*
1245 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1246 */
1247 gdb_bh = sbi->s_group_desc[gdb_num];
1248 /* Update group descriptor block for new group */
1249 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1250 gdb_off * EXT4_DESC_SIZE(sb));
1251
1252 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1253 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1254 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1255 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1256 if (err) {
1257 ext4_std_error(sb, err);
1258 break;
1259 }
1260
1261 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1262 ext4_free_group_clusters_set(sb, gdp,
1263 EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1264 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1265 if (ext4_has_group_desc_csum(sb))
1266 ext4_itable_unused_set(sb, gdp,
1267 EXT4_INODES_PER_GROUP(sb));
1268 gdp->bg_flags = cpu_to_le16(*bg_flags);
1269 ext4_group_desc_csum_set(sb, group, gdp);
1270
1271 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1272 if (unlikely(err)) {
1273 ext4_std_error(sb, err);
1274 break;
1275 }
1276
1277 /*
1278 * We can allocate memory for mb_alloc based on the new group
1279 * descriptor
1280 */
1281 err = ext4_mb_add_groupinfo(sb, group, gdp);
1282 if (err)
1283 break;
1284 }
1285 return err;
1286}
1287
1288/*
1289 * ext4_update_super() updates the super block so that the newly added
1290 * groups can be seen by the filesystem.
1291 *
1292 * @sb: super block
1293 * @flex_gd: new added groups
1294 */
1295static void ext4_update_super(struct super_block *sb,
1296 struct ext4_new_flex_group_data *flex_gd)
1297{
1298 ext4_fsblk_t blocks_count = 0;
1299 ext4_fsblk_t free_blocks = 0;
1300 ext4_fsblk_t reserved_blocks = 0;
1301 struct ext4_new_group_data *group_data = flex_gd->groups;
1302 struct ext4_sb_info *sbi = EXT4_SB(sb);
1303 struct ext4_super_block *es = sbi->s_es;
1304 int i;
1305
1306 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1307 /*
1308 * Make the new blocks and inodes valid next. We do this before
1309 * increasing the group count so that once the group is enabled,
1310 * all of its blocks and inodes are already valid.
1311 *
1312 * We always allocate group-by-group, then block-by-block or
1313 * inode-by-inode within a group, so enabling these
1314 * blocks/inodes before the group is live won't actually let us
1315 * allocate the new space yet.
1316 */
1317 for (i = 0; i < flex_gd->count; i++) {
1318 blocks_count += group_data[i].blocks_count;
1319 free_blocks += group_data[i].free_blocks_count;
1320 }
1321
1322 reserved_blocks = ext4_r_blocks_count(es) * 100;
1323 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1324 reserved_blocks *= blocks_count;
1325 do_div(reserved_blocks, 100);
1326
1327 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1328 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1329 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1330 flex_gd->count);
1331 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1332 flex_gd->count);
1333
1334 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1335 /*
1336 * We need to protect s_groups_count against other CPUs seeing
1337 * inconsistent state in the superblock.
1338 *
1339 * The precise rules we use are:
1340 *
1341 * * Writers must perform a smp_wmb() after updating all
1342 * dependent data and before modifying the groups count
1343 *
1344 * * Readers must perform an smp_rmb() after reading the groups
1345 * count and before reading any dependent data.
1346 *
1347 * NB. These rules can be relaxed when checking the group count
1348 * while freeing data, as we can only allocate from a block
1349 * group after serialising against the group count, and we can
1350 * only then free after serialising in turn against that
1351 * allocation.
1352 */
1353 smp_wmb();
1354
1355 /* Update the global fs size fields */
1356 sbi->s_groups_count += flex_gd->count;
1357 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1358 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1359
1360 /* Update the reserved block counts only once the new group is
1361 * active. */
1362 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1363 reserved_blocks);
1364
1365 /* Update the free space counts */
1366 percpu_counter_add(&sbi->s_freeclusters_counter,
1367 EXT4_NUM_B2C(sbi, free_blocks));
1368 percpu_counter_add(&sbi->s_freeinodes_counter,
1369 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1370
1371 ext4_debug("free blocks count %llu",
1372 percpu_counter_read(&sbi->s_freeclusters_counter));
1373 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1374 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1375 sbi->s_log_groups_per_flex) {
1376 ext4_group_t flex_group;
1377 flex_group = ext4_flex_group(sbi, group_data[0].group);
1378 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1379 &sbi->s_flex_groups[flex_group].free_clusters);
1380 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1381 &sbi->s_flex_groups[flex_group].free_inodes);
1382 }
1383
1384 /*
1385 * Update the fs overhead information
1386 */
1387 ext4_calculate_overhead(sb);
1388
1389 if (test_opt(sb, DEBUG))
1390 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1391 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1392 blocks_count, free_blocks, reserved_blocks);
1393}
1394
1395/* Add a flex group to an fs. Ensure we handle all possible error conditions
1396 * _before_ we start modifying the filesystem, because we cannot abort the
1397 * transaction and not have it write the data to disk.
1398 */
1399static int ext4_flex_group_add(struct super_block *sb,
1400 struct inode *resize_inode,
1401 struct ext4_new_flex_group_data *flex_gd)
1402{
1403 struct ext4_sb_info *sbi = EXT4_SB(sb);
1404 struct ext4_super_block *es = sbi->s_es;
1405 ext4_fsblk_t o_blocks_count;
1406 ext4_grpblk_t last;
1407 ext4_group_t group;
1408 handle_t *handle;
1409 unsigned reserved_gdb;
1410 int err = 0, err2 = 0, credit;
1411
1412 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1413
1414 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1415 o_blocks_count = ext4_blocks_count(es);
1416 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1417 BUG_ON(last);
1418
1419 err = setup_new_flex_group_blocks(sb, flex_gd);
1420 if (err)
1421 goto exit;
1422 /*
1423 * We will always be modifying at least the superblock and GDT
1424 * block. If we are adding a group past the last current GDT block,
1425 * we will also modify the inode and the dindirect block. If we
1426 * are adding a group with superblock/GDT backups we will also
1427 * modify each of the reserved GDT dindirect blocks.
1428 */
1429 credit = flex_gd->count * 4 + reserved_gdb;
1430 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1431 if (IS_ERR(handle)) {
1432 err = PTR_ERR(handle);
1433 goto exit;
1434 }
1435
1436 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1437 if (err)
1438 goto exit_journal;
1439
1440 group = flex_gd->groups[0].group;
1441 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1442 err = ext4_add_new_descs(handle, sb, group,
1443 resize_inode, flex_gd->count);
1444 if (err)
1445 goto exit_journal;
1446
1447 err = ext4_setup_new_descs(handle, sb, flex_gd);
1448 if (err)
1449 goto exit_journal;
1450
1451 ext4_update_super(sb, flex_gd);
1452
1453 err = ext4_handle_dirty_super(handle, sb);
1454
1455exit_journal:
1456 err2 = ext4_journal_stop(handle);
1457 if (!err)
1458 err = err2;
1459
1460 if (!err) {
1461 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1462 int gdb_num_end = ((group + flex_gd->count - 1) /
1463 EXT4_DESC_PER_BLOCK(sb));
1464 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1465 EXT4_FEATURE_INCOMPAT_META_BG);
1466 sector_t old_gdb = 0;
1467
1468 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1469 sizeof(struct ext4_super_block), 0);
1470 for (; gdb_num <= gdb_num_end; gdb_num++) {
1471 struct buffer_head *gdb_bh;
1472
1473 gdb_bh = sbi->s_group_desc[gdb_num];
1474 if (old_gdb == gdb_bh->b_blocknr)
1475 continue;
1476 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1477 gdb_bh->b_size, meta_bg);
1478 old_gdb = gdb_bh->b_blocknr;
1479 }
1480 }
1481exit:
1482 return err;
1483}
1484
1485static int ext4_setup_next_flex_gd(struct super_block *sb,
1486 struct ext4_new_flex_group_data *flex_gd,
1487 ext4_fsblk_t n_blocks_count,
1488 unsigned long flexbg_size)
1489{
1490 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1491 struct ext4_new_group_data *group_data = flex_gd->groups;
1492 ext4_fsblk_t o_blocks_count;
1493 ext4_group_t n_group;
1494 ext4_group_t group;
1495 ext4_group_t last_group;
1496 ext4_grpblk_t last;
1497 ext4_grpblk_t blocks_per_group;
1498 unsigned long i;
1499
1500 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1501
1502 o_blocks_count = ext4_blocks_count(es);
1503
1504 if (o_blocks_count == n_blocks_count)
1505 return 0;
1506
1507 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1508 BUG_ON(last);
1509 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1510
1511 last_group = group | (flexbg_size - 1);
1512 if (last_group > n_group)
1513 last_group = n_group;
1514
1515 flex_gd->count = last_group - group + 1;
1516
1517 for (i = 0; i < flex_gd->count; i++) {
1518 int overhead;
1519
1520 group_data[i].group = group + i;
1521 group_data[i].blocks_count = blocks_per_group;
1522 overhead = ext4_group_overhead_blocks(sb, group + i);
1523 group_data[i].free_blocks_count = blocks_per_group - overhead;
1524 if (ext4_has_group_desc_csum(sb)) {
1525 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1526 EXT4_BG_INODE_UNINIT;
1527 if (!test_opt(sb, INIT_INODE_TABLE))
1528 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1529 } else
1530 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1531 }
1532
1533 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1534 /* We need to initialize block bitmap of last group. */
1535 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1536
1537 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1538 group_data[i - 1].blocks_count = last + 1;
1539 group_data[i - 1].free_blocks_count -= blocks_per_group-
1540 last - 1;
1541 }
1542
1543 return 1;
1544}
1545
1546/* Add group descriptor data to an existing or new group descriptor block.
1547 * Ensure we handle all possible error conditions _before_ we start modifying
1548 * the filesystem, because we cannot abort the transaction and not have it
1549 * write the data to disk.
1550 *
1551 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1552 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1553 *
1554 * We only need to hold the superblock lock while we are actually adding
1555 * in the new group's counts to the superblock. Prior to that we have
1556 * not really "added" the group at all. We re-check that we are still
1557 * adding in the last group in case things have changed since verifying.
1558 */
1559int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1560{
1561 struct ext4_new_flex_group_data flex_gd;
1562 struct ext4_sb_info *sbi = EXT4_SB(sb);
1563 struct ext4_super_block *es = sbi->s_es;
1564 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1565 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1566 struct inode *inode = NULL;
1567 int gdb_off;
1568 int err;
1569 __u16 bg_flags = 0;
1570
1571 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1572
1573 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1574 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1575 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1576 return -EPERM;
1577 }
1578
1579 if (ext4_blocks_count(es) + input->blocks_count <
1580 ext4_blocks_count(es)) {
1581 ext4_warning(sb, "blocks_count overflow");
1582 return -EINVAL;
1583 }
1584
1585 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1586 le32_to_cpu(es->s_inodes_count)) {
1587 ext4_warning(sb, "inodes_count overflow");
1588 return -EINVAL;
1589 }
1590
1591 if (reserved_gdb || gdb_off == 0) {
1592 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1593 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1594 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1595 ext4_warning(sb,
1596 "No reserved GDT blocks, can't resize");
1597 return -EPERM;
1598 }
1599 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1600 if (IS_ERR(inode)) {
1601 ext4_warning(sb, "Error opening resize inode");
1602 return PTR_ERR(inode);
1603 }
1604 }
1605
1606
1607 err = verify_group_input(sb, input);
1608 if (err)
1609 goto out;
1610
1611 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1612 if (err)
1613 goto out;
1614
1615 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1616 if (err)
1617 goto out;
1618
1619 flex_gd.count = 1;
1620 flex_gd.groups = input;
1621 flex_gd.bg_flags = &bg_flags;
1622 err = ext4_flex_group_add(sb, inode, &flex_gd);
1623out:
1624 iput(inode);
1625 return err;
1626} /* ext4_group_add */
1627
1628/*
1629 * extend a group without checking assuming that checking has been done.
1630 */
1631static int ext4_group_extend_no_check(struct super_block *sb,
1632 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1633{
1634 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1635 handle_t *handle;
1636 int err = 0, err2;
1637
1638 /* We will update the superblock, one block bitmap, and
1639 * one group descriptor via ext4_group_add_blocks().
1640 */
1641 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1642 if (IS_ERR(handle)) {
1643 err = PTR_ERR(handle);
1644 ext4_warning(sb, "error %d on journal start", err);
1645 return err;
1646 }
1647
1648 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1649 if (err) {
1650 ext4_warning(sb, "error %d on journal write access", err);
1651 goto errout;
1652 }
1653
1654 ext4_blocks_count_set(es, o_blocks_count + add);
1655 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1656 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1657 o_blocks_count + add);
1658 /* We add the blocks to the bitmap and set the group need init bit */
1659 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1660 if (err)
1661 goto errout;
1662 ext4_handle_dirty_super(handle, sb);
1663 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1664 o_blocks_count + add);
1665errout:
1666 err2 = ext4_journal_stop(handle);
1667 if (err2 && !err)
1668 err = err2;
1669
1670 if (!err) {
1671 if (test_opt(sb, DEBUG))
1672 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1673 "blocks\n", ext4_blocks_count(es));
1674 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1675 (char *)es, sizeof(struct ext4_super_block), 0);
1676 }
1677 return err;
1678}
1679
1680/*
1681 * Extend the filesystem to the new number of blocks specified. This entry
1682 * point is only used to extend the current filesystem to the end of the last
1683 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1684 * for emergencies (because it has no dependencies on reserved blocks).
1685 *
1686 * If we _really_ wanted, we could use default values to call ext4_group_add()
1687 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1688 * GDT blocks are reserved to grow to the desired size.
1689 */
1690int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1691 ext4_fsblk_t n_blocks_count)
1692{
1693 ext4_fsblk_t o_blocks_count;
1694 ext4_grpblk_t last;
1695 ext4_grpblk_t add;
1696 struct buffer_head *bh;
1697 int err;
1698 ext4_group_t group;
1699
1700 o_blocks_count = ext4_blocks_count(es);
1701
1702 if (test_opt(sb, DEBUG))
1703 ext4_msg(sb, KERN_DEBUG,
1704 "extending last group from %llu to %llu blocks",
1705 o_blocks_count, n_blocks_count);
1706
1707 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1708 return 0;
1709
1710 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1711 ext4_msg(sb, KERN_ERR,
1712 "filesystem too large to resize to %llu blocks safely",
1713 n_blocks_count);
1714 if (sizeof(sector_t) < 8)
1715 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1716 return -EINVAL;
1717 }
1718
1719 if (n_blocks_count < o_blocks_count) {
1720 ext4_warning(sb, "can't shrink FS - resize aborted");
1721 return -EINVAL;
1722 }
1723
1724 /* Handle the remaining blocks in the last group only. */
1725 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1726
1727 if (last == 0) {
1728 ext4_warning(sb, "need to use ext2online to resize further");
1729 return -EPERM;
1730 }
1731
1732 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1733
1734 if (o_blocks_count + add < o_blocks_count) {
1735 ext4_warning(sb, "blocks_count overflow");
1736 return -EINVAL;
1737 }
1738
1739 if (o_blocks_count + add > n_blocks_count)
1740 add = n_blocks_count - o_blocks_count;
1741
1742 if (o_blocks_count + add < n_blocks_count)
1743 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1744 o_blocks_count + add, add);
1745
1746 /* See if the device is actually as big as what was requested */
1747 bh = sb_bread(sb, o_blocks_count + add - 1);
1748 if (!bh) {
1749 ext4_warning(sb, "can't read last block, resize aborted");
1750 return -ENOSPC;
1751 }
1752 brelse(bh);
1753
1754 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1755 return err;
1756} /* ext4_group_extend */
1757
1758
1759static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1760{
1761 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1762}
1763
1764/*
1765 * Release the resize inode and drop the resize_inode feature if there
1766 * are no more reserved gdt blocks, and then convert the file system
1767 * to enable meta_bg
1768 */
1769static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1770{
1771 handle_t *handle;
1772 struct ext4_sb_info *sbi = EXT4_SB(sb);
1773 struct ext4_super_block *es = sbi->s_es;
1774 struct ext4_inode_info *ei = EXT4_I(inode);
1775 ext4_fsblk_t nr;
1776 int i, ret, err = 0;
1777 int credits = 1;
1778
1779 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1780 if (inode) {
1781 if (es->s_reserved_gdt_blocks) {
1782 ext4_error(sb, "Unexpected non-zero "
1783 "s_reserved_gdt_blocks");
1784 return -EPERM;
1785 }
1786
1787 /* Do a quick sanity check of the resize inode */
1788 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1789 goto invalid_resize_inode;
1790 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1791 if (i == EXT4_DIND_BLOCK) {
1792 if (ei->i_data[i])
1793 continue;
1794 else
1795 goto invalid_resize_inode;
1796 }
1797 if (ei->i_data[i])
1798 goto invalid_resize_inode;
1799 }
1800 credits += 3; /* block bitmap, bg descriptor, resize inode */
1801 }
1802
1803 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1804 if (IS_ERR(handle))
1805 return PTR_ERR(handle);
1806
1807 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1808 if (err)
1809 goto errout;
1810
1811 EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1812 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1813 sbi->s_es->s_first_meta_bg =
1814 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1815
1816 err = ext4_handle_dirty_super(handle, sb);
1817 if (err) {
1818 ext4_std_error(sb, err);
1819 goto errout;
1820 }
1821
1822 if (inode) {
1823 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1824 ext4_free_blocks(handle, inode, NULL, nr, 1,
1825 EXT4_FREE_BLOCKS_METADATA |
1826 EXT4_FREE_BLOCKS_FORGET);
1827 ei->i_data[EXT4_DIND_BLOCK] = 0;
1828 inode->i_blocks = 0;
1829
1830 err = ext4_mark_inode_dirty(handle, inode);
1831 if (err)
1832 ext4_std_error(sb, err);
1833 }
1834
1835errout:
1836 ret = ext4_journal_stop(handle);
1837 if (!err)
1838 err = ret;
1839 return ret;
1840
1841invalid_resize_inode:
1842 ext4_error(sb, "corrupted/inconsistent resize inode");
1843 return -EINVAL;
1844}
1845
1846/*
1847 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1848 *
1849 * @sb: super block of the fs to be resized
1850 * @n_blocks_count: the number of blocks resides in the resized fs
1851 */
1852int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1853{
1854 struct ext4_new_flex_group_data *flex_gd = NULL;
1855 struct ext4_sb_info *sbi = EXT4_SB(sb);
1856 struct ext4_super_block *es = sbi->s_es;
1857 struct buffer_head *bh;
1858 struct inode *resize_inode = NULL;
1859 ext4_grpblk_t add, offset;
1860 unsigned long n_desc_blocks;
1861 unsigned long o_desc_blocks;
1862 ext4_group_t o_group;
1863 ext4_group_t n_group;
1864 ext4_fsblk_t o_blocks_count;
1865 ext4_fsblk_t n_blocks_count_retry = 0;
1866 unsigned long last_update_time = 0;
1867 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1868 int meta_bg;
1869
1870 /* See if the device is actually as big as what was requested */
1871 bh = sb_bread(sb, n_blocks_count - 1);
1872 if (!bh) {
1873 ext4_warning(sb, "can't read last block, resize aborted");
1874 return -ENOSPC;
1875 }
1876 brelse(bh);
1877
1878retry:
1879 o_blocks_count = ext4_blocks_count(es);
1880
1881 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1882 "to %llu blocks", o_blocks_count, n_blocks_count);
1883
1884 if (n_blocks_count < o_blocks_count) {
1885 /* On-line shrinking not supported */
1886 ext4_warning(sb, "can't shrink FS - resize aborted");
1887 return -EINVAL;
1888 }
1889
1890 if (n_blocks_count == o_blocks_count)
1891 /* Nothing need to do */
1892 return 0;
1893
1894 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1895 if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1896 ext4_warning(sb, "resize would cause inodes_count overflow");
1897 return -EINVAL;
1898 }
1899 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1900
1901 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1902 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1903
1904 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1905
1906 if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1907 if (meta_bg) {
1908 ext4_error(sb, "resize_inode and meta_bg enabled "
1909 "simultaneously");
1910 return -EINVAL;
1911 }
1912 if (n_desc_blocks > o_desc_blocks +
1913 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1914 n_blocks_count_retry = n_blocks_count;
1915 n_desc_blocks = o_desc_blocks +
1916 le16_to_cpu(es->s_reserved_gdt_blocks);
1917 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1918 n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1919 n_group--; /* set to last group number */
1920 }
1921
1922 if (!resize_inode)
1923 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1924 if (IS_ERR(resize_inode)) {
1925 ext4_warning(sb, "Error opening resize inode");
1926 return PTR_ERR(resize_inode);
1927 }
1928 }
1929
1930 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1931 err = ext4_convert_meta_bg(sb, resize_inode);
1932 if (err)
1933 goto out;
1934 if (resize_inode) {
1935 iput(resize_inode);
1936 resize_inode = NULL;
1937 }
1938 if (n_blocks_count_retry) {
1939 n_blocks_count = n_blocks_count_retry;
1940 n_blocks_count_retry = 0;
1941 goto retry;
1942 }
1943 }
1944
1945 /* extend the last group */
1946 if (n_group == o_group)
1947 add = n_blocks_count - o_blocks_count;
1948 else
1949 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1950 if (add > 0) {
1951 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1952 if (err)
1953 goto out;
1954 }
1955
1956 if (ext4_blocks_count(es) == n_blocks_count)
1957 goto out;
1958
1959 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1960 if (err)
1961 return err;
1962
1963 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1964 if (err)
1965 goto out;
1966
1967 flex_gd = alloc_flex_gd(flexbg_size);
1968 if (flex_gd == NULL) {
1969 err = -ENOMEM;
1970 goto out;
1971 }
1972
1973 /* Add flex groups. Note that a regular group is a
1974 * flex group with 1 group.
1975 */
1976 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1977 flexbg_size)) {
1978 if (jiffies - last_update_time > HZ * 10) {
1979 if (last_update_time)
1980 ext4_msg(sb, KERN_INFO,
1981 "resized to %llu blocks",
1982 ext4_blocks_count(es));
1983 last_update_time = jiffies;
1984 }
1985 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1986 break;
1987 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1988 if (unlikely(err))
1989 break;
1990 }
1991
1992 if (!err && n_blocks_count_retry) {
1993 n_blocks_count = n_blocks_count_retry;
1994 n_blocks_count_retry = 0;
1995 free_flex_gd(flex_gd);
1996 flex_gd = NULL;
1997 goto retry;
1998 }
1999
2000out:
2001 if (flex_gd)
2002 free_flex_gd(flex_gd);
2003 if (resize_inode != NULL)
2004 iput(resize_inode);
2005 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2006 return err;
2007}