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// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/resize.c
4 *
5 * Support for resizing an ext4 filesystem while it is mounted.
6 *
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 *
9 * This could probably be made into a module, because it is not often in use.
10 */
11
12
13#include <linux/errno.h>
14#include <linux/slab.h>
15#include <linux/jiffies.h>
16
17#include "ext4_jbd2.h"
18
19struct ext4_rcu_ptr {
20 struct rcu_head rcu;
21 void *ptr;
22};
23
24static void ext4_rcu_ptr_callback(struct rcu_head *head)
25{
26 struct ext4_rcu_ptr *ptr;
27
28 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
29 kvfree(ptr->ptr);
30 kfree(ptr);
31}
32
33void ext4_kvfree_array_rcu(void *to_free)
34{
35 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
36
37 if (ptr) {
38 ptr->ptr = to_free;
39 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
40 return;
41 }
42 synchronize_rcu();
43 kvfree(to_free);
44}
45
46int ext4_resize_begin(struct super_block *sb)
47{
48 struct ext4_sb_info *sbi = EXT4_SB(sb);
49 int ret = 0;
50
51 if (!capable(CAP_SYS_RESOURCE))
52 return -EPERM;
53
54 /*
55 * If the reserved GDT blocks is non-zero, the resize_inode feature
56 * should always be set.
57 */
58 if (sbi->s_es->s_reserved_gdt_blocks &&
59 !ext4_has_feature_resize_inode(sb)) {
60 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
61 return -EFSCORRUPTED;
62 }
63
64 /*
65 * If we are not using the primary superblock/GDT copy don't resize,
66 * because the user tools have no way of handling this. Probably a
67 * bad time to do it anyways.
68 */
69 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
70 le32_to_cpu(sbi->s_es->s_first_data_block)) {
71 ext4_warning(sb, "won't resize using backup superblock at %llu",
72 (unsigned long long)sbi->s_sbh->b_blocknr);
73 return -EPERM;
74 }
75
76 /*
77 * We are not allowed to do online-resizing on a filesystem mounted
78 * with error, because it can destroy the filesystem easily.
79 */
80 if (sbi->s_mount_state & EXT4_ERROR_FS) {
81 ext4_warning(sb, "There are errors in the filesystem, "
82 "so online resizing is not allowed");
83 return -EPERM;
84 }
85
86 if (ext4_has_feature_sparse_super2(sb)) {
87 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
88 return -EOPNOTSUPP;
89 }
90
91 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
92 &sbi->s_ext4_flags))
93 ret = -EBUSY;
94
95 return ret;
96}
97
98int ext4_resize_end(struct super_block *sb, bool update_backups)
99{
100 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
101 smp_mb__after_atomic();
102 if (update_backups)
103 return ext4_update_overhead(sb, true);
104 return 0;
105}
106
107static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
108 ext4_group_t group) {
109 ext4_grpblk_t overhead;
110 overhead = ext4_bg_num_gdb(sb, group);
111 if (ext4_bg_has_super(sb, group))
112 overhead += 1 +
113 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
114 return overhead;
115}
116
117#define outside(b, first, last) ((b) < (first) || (b) >= (last))
118#define inside(b, first, last) ((b) >= (first) && (b) < (last))
119
120static int verify_group_input(struct super_block *sb,
121 struct ext4_new_group_data *input)
122{
123 struct ext4_sb_info *sbi = EXT4_SB(sb);
124 struct ext4_super_block *es = sbi->s_es;
125 ext4_fsblk_t start = ext4_blocks_count(es);
126 ext4_fsblk_t end = start + input->blocks_count;
127 ext4_group_t group = input->group;
128 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
129 unsigned overhead;
130 ext4_fsblk_t metaend;
131 struct buffer_head *bh = NULL;
132 ext4_grpblk_t free_blocks_count, offset;
133 int err = -EINVAL;
134
135 if (group != sbi->s_groups_count) {
136 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
137 input->group, sbi->s_groups_count);
138 return -EINVAL;
139 }
140
141 overhead = ext4_group_overhead_blocks(sb, group);
142 metaend = start + overhead;
143 free_blocks_count = input->blocks_count - 2 - overhead -
144 sbi->s_itb_per_group;
145 input->free_clusters_count = EXT4_B2C(sbi, free_blocks_count);
146
147 if (test_opt(sb, DEBUG))
148 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
149 "(%d free, %u reserved)\n",
150 ext4_bg_has_super(sb, input->group) ? "normal" :
151 "no-super", input->group, input->blocks_count,
152 free_blocks_count, input->reserved_blocks);
153
154 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
155 if (offset != 0)
156 ext4_warning(sb, "Last group not full");
157 else if (input->reserved_blocks > input->blocks_count / 5)
158 ext4_warning(sb, "Reserved blocks too high (%u)",
159 input->reserved_blocks);
160 else if (free_blocks_count < 0)
161 ext4_warning(sb, "Bad blocks count %u",
162 input->blocks_count);
163 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
164 err = PTR_ERR(bh);
165 bh = NULL;
166 ext4_warning(sb, "Cannot read last block (%llu)",
167 end - 1);
168 } else if (outside(input->block_bitmap, start, end))
169 ext4_warning(sb, "Block bitmap not in group (block %llu)",
170 (unsigned long long)input->block_bitmap);
171 else if (outside(input->inode_bitmap, start, end))
172 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
173 (unsigned long long)input->inode_bitmap);
174 else if (outside(input->inode_table, start, end) ||
175 outside(itend - 1, start, end))
176 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
177 (unsigned long long)input->inode_table, itend - 1);
178 else if (input->inode_bitmap == input->block_bitmap)
179 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
180 (unsigned long long)input->block_bitmap);
181 else if (inside(input->block_bitmap, input->inode_table, itend))
182 ext4_warning(sb, "Block bitmap (%llu) in inode table "
183 "(%llu-%llu)",
184 (unsigned long long)input->block_bitmap,
185 (unsigned long long)input->inode_table, itend - 1);
186 else if (inside(input->inode_bitmap, input->inode_table, itend))
187 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
188 "(%llu-%llu)",
189 (unsigned long long)input->inode_bitmap,
190 (unsigned long long)input->inode_table, itend - 1);
191 else if (inside(input->block_bitmap, start, metaend))
192 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
193 (unsigned long long)input->block_bitmap,
194 start, metaend - 1);
195 else if (inside(input->inode_bitmap, start, metaend))
196 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
197 (unsigned long long)input->inode_bitmap,
198 start, metaend - 1);
199 else if (inside(input->inode_table, start, metaend) ||
200 inside(itend - 1, start, metaend))
201 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
202 "(%llu-%llu)",
203 (unsigned long long)input->inode_table,
204 itend - 1, start, metaend - 1);
205 else
206 err = 0;
207 brelse(bh);
208
209 return err;
210}
211
212/*
213 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
214 * group each time.
215 */
216struct ext4_new_flex_group_data {
217 struct ext4_new_group_data *groups; /* new_group_data for groups
218 in the flex group */
219 __u16 *bg_flags; /* block group flags of groups
220 in @groups */
221 ext4_group_t resize_bg; /* number of allocated
222 new_group_data */
223 ext4_group_t count; /* number of groups in @groups
224 */
225};
226
227/*
228 * Avoiding memory allocation failures due to too many groups added each time.
229 */
230#define MAX_RESIZE_BG 16384
231
232/*
233 * alloc_flex_gd() allocates an ext4_new_flex_group_data that satisfies the
234 * resizing from @o_group to @n_group, its size is typically @flexbg_size.
235 *
236 * Returns NULL on failure otherwise address of the allocated structure.
237 */
238static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size,
239 ext4_group_t o_group, ext4_group_t n_group)
240{
241 ext4_group_t last_group;
242 unsigned int max_resize_bg;
243 struct ext4_new_flex_group_data *flex_gd;
244
245 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
246 if (flex_gd == NULL)
247 goto out3;
248
249 max_resize_bg = umin(flexbg_size, MAX_RESIZE_BG);
250 flex_gd->resize_bg = max_resize_bg;
251
252 /* Avoid allocating large 'groups' array if not needed */
253 last_group = o_group | (flex_gd->resize_bg - 1);
254 if (n_group <= last_group)
255 flex_gd->resize_bg = 1 << fls(n_group - o_group);
256 else if (n_group - last_group < flex_gd->resize_bg)
257 flex_gd->resize_bg = 1 << max(fls(last_group - o_group),
258 fls(n_group - last_group));
259
260 if (WARN_ON_ONCE(flex_gd->resize_bg > max_resize_bg))
261 flex_gd->resize_bg = max_resize_bg;
262
263 flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
264 sizeof(struct ext4_new_group_data),
265 GFP_NOFS);
266 if (flex_gd->groups == NULL)
267 goto out2;
268
269 flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
270 GFP_NOFS);
271 if (flex_gd->bg_flags == NULL)
272 goto out1;
273
274 return flex_gd;
275
276out1:
277 kfree(flex_gd->groups);
278out2:
279 kfree(flex_gd);
280out3:
281 return NULL;
282}
283
284static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
285{
286 kfree(flex_gd->bg_flags);
287 kfree(flex_gd->groups);
288 kfree(flex_gd);
289}
290
291/*
292 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
293 * and inode tables for a flex group.
294 *
295 * This function is used by 64bit-resize. Note that this function allocates
296 * group tables from the 1st group of groups contained by @flexgd, which may
297 * be a partial of a flex group.
298 *
299 * @sb: super block of fs to which the groups belongs
300 *
301 * Returns 0 on a successful allocation of the metadata blocks in the
302 * block group.
303 */
304static int ext4_alloc_group_tables(struct super_block *sb,
305 struct ext4_new_flex_group_data *flex_gd,
306 unsigned int flexbg_size)
307{
308 struct ext4_new_group_data *group_data = flex_gd->groups;
309 ext4_fsblk_t start_blk;
310 ext4_fsblk_t last_blk;
311 ext4_group_t src_group;
312 ext4_group_t bb_index = 0;
313 ext4_group_t ib_index = 0;
314 ext4_group_t it_index = 0;
315 ext4_group_t group;
316 ext4_group_t last_group;
317 unsigned overhead;
318 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
319 int i;
320
321 BUG_ON(flex_gd->count == 0 || group_data == NULL);
322
323 src_group = group_data[0].group;
324 last_group = src_group + flex_gd->count - 1;
325
326 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
327 (last_group & ~(flexbg_size - 1))));
328next_group:
329 group = group_data[0].group;
330 if (src_group >= group_data[0].group + flex_gd->count)
331 return -ENOSPC;
332 start_blk = ext4_group_first_block_no(sb, src_group);
333 last_blk = start_blk + group_data[src_group - group].blocks_count;
334
335 overhead = ext4_group_overhead_blocks(sb, src_group);
336
337 start_blk += overhead;
338
339 /* We collect contiguous blocks as much as possible. */
340 src_group++;
341 for (; src_group <= last_group; src_group++) {
342 overhead = ext4_group_overhead_blocks(sb, src_group);
343 if (overhead == 0)
344 last_blk += group_data[src_group - group].blocks_count;
345 else
346 break;
347 }
348
349 /* Allocate block bitmaps */
350 for (; bb_index < flex_gd->count; bb_index++) {
351 if (start_blk >= last_blk)
352 goto next_group;
353 group_data[bb_index].block_bitmap = start_blk++;
354 group = ext4_get_group_number(sb, start_blk - 1);
355 group -= group_data[0].group;
356 group_data[group].mdata_blocks++;
357 flex_gd->bg_flags[group] &= uninit_mask;
358 }
359
360 /* Allocate inode bitmaps */
361 for (; ib_index < flex_gd->count; ib_index++) {
362 if (start_blk >= last_blk)
363 goto next_group;
364 group_data[ib_index].inode_bitmap = start_blk++;
365 group = ext4_get_group_number(sb, start_blk - 1);
366 group -= group_data[0].group;
367 group_data[group].mdata_blocks++;
368 flex_gd->bg_flags[group] &= uninit_mask;
369 }
370
371 /* Allocate inode tables */
372 for (; it_index < flex_gd->count; it_index++) {
373 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
374 ext4_fsblk_t next_group_start;
375
376 if (start_blk + itb > last_blk)
377 goto next_group;
378 group_data[it_index].inode_table = start_blk;
379 group = ext4_get_group_number(sb, start_blk);
380 next_group_start = ext4_group_first_block_no(sb, group + 1);
381 group -= group_data[0].group;
382
383 if (start_blk + itb > next_group_start) {
384 flex_gd->bg_flags[group + 1] &= uninit_mask;
385 overhead = start_blk + itb - next_group_start;
386 group_data[group + 1].mdata_blocks += overhead;
387 itb -= overhead;
388 }
389
390 group_data[group].mdata_blocks += itb;
391 flex_gd->bg_flags[group] &= uninit_mask;
392 start_blk += EXT4_SB(sb)->s_itb_per_group;
393 }
394
395 /* Update free clusters count to exclude metadata blocks */
396 for (i = 0; i < flex_gd->count; i++) {
397 group_data[i].free_clusters_count -=
398 EXT4_NUM_B2C(EXT4_SB(sb),
399 group_data[i].mdata_blocks);
400 }
401
402 if (test_opt(sb, DEBUG)) {
403 int i;
404 group = group_data[0].group;
405
406 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
407 "%u groups, flexbg size is %u:\n", flex_gd->count,
408 flexbg_size);
409
410 for (i = 0; i < flex_gd->count; i++) {
411 ext4_debug(
412 "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
413 ext4_bg_has_super(sb, group + i) ? "normal" :
414 "no-super", group + i,
415 group_data[i].blocks_count,
416 group_data[i].free_clusters_count,
417 group_data[i].mdata_blocks);
418 }
419 }
420 return 0;
421}
422
423static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
424 ext4_fsblk_t blk)
425{
426 struct buffer_head *bh;
427 int err;
428
429 bh = sb_getblk(sb, blk);
430 if (unlikely(!bh))
431 return ERR_PTR(-ENOMEM);
432 BUFFER_TRACE(bh, "get_write_access");
433 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
434 if (err) {
435 brelse(bh);
436 bh = ERR_PTR(err);
437 } else {
438 memset(bh->b_data, 0, sb->s_blocksize);
439 set_buffer_uptodate(bh);
440 }
441
442 return bh;
443}
444
445static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
446{
447 return ext4_journal_ensure_credits_fn(handle, credits,
448 EXT4_MAX_TRANS_DATA, 0, 0);
449}
450
451/*
452 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
453 *
454 * Helper function for ext4_setup_new_group_blocks() which set .
455 *
456 * @sb: super block
457 * @handle: journal handle
458 * @flex_gd: flex group data
459 */
460static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
461 struct ext4_new_flex_group_data *flex_gd,
462 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
463{
464 struct ext4_sb_info *sbi = EXT4_SB(sb);
465 ext4_group_t count = last_cluster - first_cluster + 1;
466 ext4_group_t count2;
467
468 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
469 last_cluster);
470 for (; count > 0; count -= count2, first_cluster += count2) {
471 ext4_fsblk_t start;
472 struct buffer_head *bh;
473 ext4_group_t group;
474 int err;
475
476 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
477 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
478 group -= flex_gd->groups[0].group;
479
480 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
481 if (count2 > count)
482 count2 = count;
483
484 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
485 BUG_ON(flex_gd->count > 1);
486 continue;
487 }
488
489 err = ext4_resize_ensure_credits_batch(handle, 1);
490 if (err < 0)
491 return err;
492
493 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
494 if (unlikely(!bh))
495 return -ENOMEM;
496
497 BUFFER_TRACE(bh, "get_write_access");
498 err = ext4_journal_get_write_access(handle, sb, bh,
499 EXT4_JTR_NONE);
500 if (err) {
501 brelse(bh);
502 return err;
503 }
504 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
505 first_cluster, first_cluster - start, count2);
506 mb_set_bits(bh->b_data, first_cluster - start, count2);
507
508 err = ext4_handle_dirty_metadata(handle, NULL, bh);
509 brelse(bh);
510 if (unlikely(err))
511 return err;
512 }
513
514 return 0;
515}
516
517/*
518 * Set up the block and inode bitmaps, and the inode table for the new groups.
519 * This doesn't need to be part of the main transaction, since we are only
520 * changing blocks outside the actual filesystem. We still do journaling to
521 * ensure the recovery is correct in case of a failure just after resize.
522 * If any part of this fails, we simply abort the resize.
523 *
524 * setup_new_flex_group_blocks handles a flex group as follow:
525 * 1. copy super block and GDT, and initialize group tables if necessary.
526 * In this step, we only set bits in blocks bitmaps for blocks taken by
527 * super block and GDT.
528 * 2. allocate group tables in block bitmaps, that is, set bits in block
529 * bitmap for blocks taken by group tables.
530 */
531static int setup_new_flex_group_blocks(struct super_block *sb,
532 struct ext4_new_flex_group_data *flex_gd)
533{
534 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
535 ext4_fsblk_t start;
536 ext4_fsblk_t block;
537 struct ext4_sb_info *sbi = EXT4_SB(sb);
538 struct ext4_super_block *es = sbi->s_es;
539 struct ext4_new_group_data *group_data = flex_gd->groups;
540 __u16 *bg_flags = flex_gd->bg_flags;
541 handle_t *handle;
542 ext4_group_t group, count;
543 struct buffer_head *bh = NULL;
544 int reserved_gdb, i, j, err = 0, err2;
545 int meta_bg;
546
547 BUG_ON(!flex_gd->count || !group_data ||
548 group_data[0].group != sbi->s_groups_count);
549
550 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
551 meta_bg = ext4_has_feature_meta_bg(sb);
552
553 /* This transaction may be extended/restarted along the way */
554 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
555 if (IS_ERR(handle))
556 return PTR_ERR(handle);
557
558 group = group_data[0].group;
559 for (i = 0; i < flex_gd->count; i++, group++) {
560 unsigned long gdblocks;
561 ext4_grpblk_t overhead;
562
563 gdblocks = ext4_bg_num_gdb(sb, group);
564 start = ext4_group_first_block_no(sb, group);
565
566 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
567 goto handle_itb;
568
569 if (meta_bg == 1)
570 goto handle_itb;
571
572 block = start + ext4_bg_has_super(sb, group);
573 /* Copy all of the GDT blocks into the backup in this group */
574 for (j = 0; j < gdblocks; j++, block++) {
575 struct buffer_head *gdb;
576
577 ext4_debug("update backup group %#04llx\n", block);
578 err = ext4_resize_ensure_credits_batch(handle, 1);
579 if (err < 0)
580 goto out;
581
582 gdb = sb_getblk(sb, block);
583 if (unlikely(!gdb)) {
584 err = -ENOMEM;
585 goto out;
586 }
587
588 BUFFER_TRACE(gdb, "get_write_access");
589 err = ext4_journal_get_write_access(handle, sb, gdb,
590 EXT4_JTR_NONE);
591 if (err) {
592 brelse(gdb);
593 goto out;
594 }
595 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
596 s_group_desc, j)->b_data, gdb->b_size);
597 set_buffer_uptodate(gdb);
598
599 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
600 if (unlikely(err)) {
601 brelse(gdb);
602 goto out;
603 }
604 brelse(gdb);
605 }
606
607 /* Zero out all of the reserved backup group descriptor
608 * table blocks
609 */
610 if (ext4_bg_has_super(sb, group)) {
611 err = sb_issue_zeroout(sb, gdblocks + start + 1,
612 reserved_gdb, GFP_NOFS);
613 if (err)
614 goto out;
615 }
616
617handle_itb:
618 /* Initialize group tables of the group @group */
619 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
620 goto handle_bb;
621
622 /* Zero out all of the inode table blocks */
623 block = group_data[i].inode_table;
624 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
625 block, sbi->s_itb_per_group);
626 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
627 GFP_NOFS);
628 if (err)
629 goto out;
630
631handle_bb:
632 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
633 goto handle_ib;
634
635 /* Initialize block bitmap of the @group */
636 block = group_data[i].block_bitmap;
637 err = ext4_resize_ensure_credits_batch(handle, 1);
638 if (err < 0)
639 goto out;
640
641 bh = bclean(handle, sb, block);
642 if (IS_ERR(bh)) {
643 err = PTR_ERR(bh);
644 goto out;
645 }
646 overhead = ext4_group_overhead_blocks(sb, group);
647 if (overhead != 0) {
648 ext4_debug("mark backup superblock %#04llx (+0)\n",
649 start);
650 mb_set_bits(bh->b_data, 0,
651 EXT4_NUM_B2C(sbi, overhead));
652 }
653 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
654 sb->s_blocksize * 8, bh->b_data);
655 err = ext4_handle_dirty_metadata(handle, NULL, bh);
656 brelse(bh);
657 if (err)
658 goto out;
659
660handle_ib:
661 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
662 continue;
663
664 /* Initialize inode bitmap of the @group */
665 block = group_data[i].inode_bitmap;
666 err = ext4_resize_ensure_credits_batch(handle, 1);
667 if (err < 0)
668 goto out;
669 /* Mark unused entries in inode bitmap used */
670 bh = bclean(handle, sb, block);
671 if (IS_ERR(bh)) {
672 err = PTR_ERR(bh);
673 goto out;
674 }
675
676 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
677 sb->s_blocksize * 8, bh->b_data);
678 err = ext4_handle_dirty_metadata(handle, NULL, bh);
679 brelse(bh);
680 if (err)
681 goto out;
682 }
683
684 /* Mark group tables in block bitmap */
685 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
686 count = group_table_count[j];
687 start = (&group_data[0].block_bitmap)[j];
688 block = start;
689 for (i = 1; i < flex_gd->count; i++) {
690 block += group_table_count[j];
691 if (block == (&group_data[i].block_bitmap)[j]) {
692 count += group_table_count[j];
693 continue;
694 }
695 err = set_flexbg_block_bitmap(sb, handle,
696 flex_gd,
697 EXT4_B2C(sbi, start),
698 EXT4_B2C(sbi,
699 start + count
700 - 1));
701 if (err)
702 goto out;
703 count = group_table_count[j];
704 start = (&group_data[i].block_bitmap)[j];
705 block = start;
706 }
707
708 err = set_flexbg_block_bitmap(sb, handle,
709 flex_gd,
710 EXT4_B2C(sbi, start),
711 EXT4_B2C(sbi,
712 start + count
713 - 1));
714 if (err)
715 goto out;
716 }
717
718out:
719 err2 = ext4_journal_stop(handle);
720 if (err2 && !err)
721 err = err2;
722
723 return err;
724}
725
726/*
727 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
728 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
729 * calling this for the first time. In a sparse filesystem it will be the
730 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
731 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
732 */
733unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
734 unsigned int *five, unsigned int *seven)
735{
736 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
737 unsigned int *min = three;
738 int mult = 3;
739 unsigned int ret;
740
741 if (ext4_has_feature_sparse_super2(sb)) {
742 do {
743 if (*min > 2)
744 return UINT_MAX;
745 ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
746 *min += 1;
747 } while (!ret);
748 return ret;
749 }
750
751 if (!ext4_has_feature_sparse_super(sb)) {
752 ret = *min;
753 *min += 1;
754 return ret;
755 }
756
757 if (*five < *min) {
758 min = five;
759 mult = 5;
760 }
761 if (*seven < *min) {
762 min = seven;
763 mult = 7;
764 }
765
766 ret = *min;
767 *min *= mult;
768
769 return ret;
770}
771
772/*
773 * Check that all of the backup GDT blocks are held in the primary GDT block.
774 * It is assumed that they are stored in group order. Returns the number of
775 * groups in current filesystem that have BACKUPS, or -ve error code.
776 */
777static int verify_reserved_gdb(struct super_block *sb,
778 ext4_group_t end,
779 struct buffer_head *primary)
780{
781 const ext4_fsblk_t blk = primary->b_blocknr;
782 unsigned three = 1;
783 unsigned five = 5;
784 unsigned seven = 7;
785 unsigned grp;
786 __le32 *p = (__le32 *)primary->b_data;
787 int gdbackups = 0;
788
789 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
790 if (le32_to_cpu(*p++) !=
791 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
792 ext4_warning(sb, "reserved GDT %llu"
793 " missing grp %d (%llu)",
794 blk, grp,
795 grp *
796 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
797 blk);
798 return -EINVAL;
799 }
800 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
801 return -EFBIG;
802 }
803
804 return gdbackups;
805}
806
807/*
808 * Called when we need to bring a reserved group descriptor table block into
809 * use from the resize inode. The primary copy of the new GDT block currently
810 * is an indirect block (under the double indirect block in the resize inode).
811 * The new backup GDT blocks will be stored as leaf blocks in this indirect
812 * block, in group order. Even though we know all the block numbers we need,
813 * we check to ensure that the resize inode has actually reserved these blocks.
814 *
815 * Don't need to update the block bitmaps because the blocks are still in use.
816 *
817 * We get all of the error cases out of the way, so that we are sure to not
818 * fail once we start modifying the data on disk, because JBD has no rollback.
819 */
820static int add_new_gdb(handle_t *handle, struct inode *inode,
821 ext4_group_t group)
822{
823 struct super_block *sb = inode->i_sb;
824 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
825 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
826 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
827 struct buffer_head **o_group_desc, **n_group_desc = NULL;
828 struct buffer_head *dind = NULL;
829 struct buffer_head *gdb_bh = NULL;
830 int gdbackups;
831 struct ext4_iloc iloc = { .bh = NULL };
832 __le32 *data;
833 int err;
834
835 if (test_opt(sb, DEBUG))
836 printk(KERN_DEBUG
837 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
838 gdb_num);
839
840 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
841 if (IS_ERR(gdb_bh))
842 return PTR_ERR(gdb_bh);
843
844 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
845 if (gdbackups < 0) {
846 err = gdbackups;
847 goto errout;
848 }
849
850 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
851 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
852 if (IS_ERR(dind)) {
853 err = PTR_ERR(dind);
854 dind = NULL;
855 goto errout;
856 }
857
858 data = (__le32 *)dind->b_data;
859 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
860 ext4_warning(sb, "new group %u GDT block %llu not reserved",
861 group, gdblock);
862 err = -EINVAL;
863 goto errout;
864 }
865
866 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
867 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
868 EXT4_JTR_NONE);
869 if (unlikely(err))
870 goto errout;
871
872 BUFFER_TRACE(gdb_bh, "get_write_access");
873 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
874 if (unlikely(err))
875 goto errout;
876
877 BUFFER_TRACE(dind, "get_write_access");
878 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
879 if (unlikely(err)) {
880 ext4_std_error(sb, err);
881 goto errout;
882 }
883
884 /* ext4_reserve_inode_write() gets a reference on the iloc */
885 err = ext4_reserve_inode_write(handle, inode, &iloc);
886 if (unlikely(err))
887 goto errout;
888
889 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
890 GFP_KERNEL);
891 if (!n_group_desc) {
892 err = -ENOMEM;
893 ext4_warning(sb, "not enough memory for %lu groups",
894 gdb_num + 1);
895 goto errout;
896 }
897
898 /*
899 * Finally, we have all of the possible failures behind us...
900 *
901 * Remove new GDT block from inode double-indirect block and clear out
902 * the new GDT block for use (which also "frees" the backup GDT blocks
903 * from the reserved inode). We don't need to change the bitmaps for
904 * these blocks, because they are marked as in-use from being in the
905 * reserved inode, and will become GDT blocks (primary and backup).
906 */
907 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
908 err = ext4_handle_dirty_metadata(handle, NULL, dind);
909 if (unlikely(err)) {
910 ext4_std_error(sb, err);
911 goto errout;
912 }
913 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
914 (9 - EXT4_SB(sb)->s_cluster_bits);
915 ext4_mark_iloc_dirty(handle, inode, &iloc);
916 memset(gdb_bh->b_data, 0, sb->s_blocksize);
917 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
918 if (unlikely(err)) {
919 ext4_std_error(sb, err);
920 iloc.bh = NULL;
921 goto errout;
922 }
923 brelse(dind);
924
925 rcu_read_lock();
926 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
927 memcpy(n_group_desc, o_group_desc,
928 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
929 rcu_read_unlock();
930 n_group_desc[gdb_num] = gdb_bh;
931 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
932 EXT4_SB(sb)->s_gdb_count++;
933 ext4_kvfree_array_rcu(o_group_desc);
934
935 lock_buffer(EXT4_SB(sb)->s_sbh);
936 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
937 ext4_superblock_csum_set(sb);
938 unlock_buffer(EXT4_SB(sb)->s_sbh);
939 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
940 if (err)
941 ext4_std_error(sb, err);
942 return err;
943errout:
944 kvfree(n_group_desc);
945 brelse(iloc.bh);
946 brelse(dind);
947 brelse(gdb_bh);
948
949 ext4_debug("leaving with error %d\n", err);
950 return err;
951}
952
953/*
954 * If there is no available space in the existing block group descriptors for
955 * the new block group and there are no reserved block group descriptors, then
956 * the meta_bg feature will get enabled, and es->s_first_meta_bg will get set
957 * to the first block group that is managed using meta_bg and s_first_meta_bg
958 * must be a multiple of EXT4_DESC_PER_BLOCK(sb).
959 * This function will be called when first group of meta_bg is added to bring
960 * new group descriptors block of new added meta_bg.
961 */
962static int add_new_gdb_meta_bg(struct super_block *sb,
963 handle_t *handle, ext4_group_t group) {
964 ext4_fsblk_t gdblock;
965 struct buffer_head *gdb_bh;
966 struct buffer_head **o_group_desc, **n_group_desc;
967 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
968 int err;
969
970 gdblock = ext4_group_first_block_no(sb, group) +
971 ext4_bg_has_super(sb, group);
972 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
973 if (IS_ERR(gdb_bh))
974 return PTR_ERR(gdb_bh);
975 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
976 GFP_KERNEL);
977 if (!n_group_desc) {
978 brelse(gdb_bh);
979 err = -ENOMEM;
980 ext4_warning(sb, "not enough memory for %lu groups",
981 gdb_num + 1);
982 return err;
983 }
984
985 rcu_read_lock();
986 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
987 memcpy(n_group_desc, o_group_desc,
988 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
989 rcu_read_unlock();
990 n_group_desc[gdb_num] = gdb_bh;
991
992 BUFFER_TRACE(gdb_bh, "get_write_access");
993 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
994 if (err) {
995 kvfree(n_group_desc);
996 brelse(gdb_bh);
997 return err;
998 }
999
1000 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
1001 EXT4_SB(sb)->s_gdb_count++;
1002 ext4_kvfree_array_rcu(o_group_desc);
1003 return err;
1004}
1005
1006/*
1007 * Called when we are adding a new group which has a backup copy of each of
1008 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
1009 * We need to add these reserved backup GDT blocks to the resize inode, so
1010 * that they are kept for future resizing and not allocated to files.
1011 *
1012 * Each reserved backup GDT block will go into a different indirect block.
1013 * The indirect blocks are actually the primary reserved GDT blocks,
1014 * so we know in advance what their block numbers are. We only get the
1015 * double-indirect block to verify it is pointing to the primary reserved
1016 * GDT blocks so we don't overwrite a data block by accident. The reserved
1017 * backup GDT blocks are stored in their reserved primary GDT block.
1018 */
1019static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1020 ext4_group_t group)
1021{
1022 struct super_block *sb = inode->i_sb;
1023 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1024 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1025 struct buffer_head **primary;
1026 struct buffer_head *dind;
1027 struct ext4_iloc iloc;
1028 ext4_fsblk_t blk;
1029 __le32 *data, *end;
1030 int gdbackups = 0;
1031 int res, i;
1032 int err;
1033
1034 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1035 if (!primary)
1036 return -ENOMEM;
1037
1038 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1039 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1040 if (IS_ERR(dind)) {
1041 err = PTR_ERR(dind);
1042 dind = NULL;
1043 goto exit_free;
1044 }
1045
1046 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1047 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1048 EXT4_ADDR_PER_BLOCK(sb));
1049 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1050
1051 /* Get each reserved primary GDT block and verify it holds backups */
1052 for (res = 0; res < reserved_gdb; res++, blk++) {
1053 if (le32_to_cpu(*data) != blk) {
1054 ext4_warning(sb, "reserved block %llu"
1055 " not at offset %ld",
1056 blk,
1057 (long)(data - (__le32 *)dind->b_data));
1058 err = -EINVAL;
1059 goto exit_bh;
1060 }
1061 primary[res] = ext4_sb_bread(sb, blk, 0);
1062 if (IS_ERR(primary[res])) {
1063 err = PTR_ERR(primary[res]);
1064 primary[res] = NULL;
1065 goto exit_bh;
1066 }
1067 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1068 if (gdbackups < 0) {
1069 brelse(primary[res]);
1070 err = gdbackups;
1071 goto exit_bh;
1072 }
1073 if (++data >= end)
1074 data = (__le32 *)dind->b_data;
1075 }
1076
1077 for (i = 0; i < reserved_gdb; i++) {
1078 BUFFER_TRACE(primary[i], "get_write_access");
1079 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1080 EXT4_JTR_NONE)))
1081 goto exit_bh;
1082 }
1083
1084 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1085 goto exit_bh;
1086
1087 /*
1088 * Finally we can add each of the reserved backup GDT blocks from
1089 * the new group to its reserved primary GDT block.
1090 */
1091 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1092 for (i = 0; i < reserved_gdb; i++) {
1093 int err2;
1094 data = (__le32 *)primary[i]->b_data;
1095 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1096 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1097 if (!err)
1098 err = err2;
1099 }
1100
1101 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1102 ext4_mark_iloc_dirty(handle, inode, &iloc);
1103
1104exit_bh:
1105 while (--res >= 0)
1106 brelse(primary[res]);
1107 brelse(dind);
1108
1109exit_free:
1110 kfree(primary);
1111
1112 return err;
1113}
1114
1115static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1116 ext4_group_t group)
1117{
1118 struct ext4_super_block *es = (struct ext4_super_block *) data;
1119
1120 es->s_block_group_nr = cpu_to_le16(group);
1121 if (ext4_has_metadata_csum(sb))
1122 es->s_checksum = ext4_superblock_csum(sb, es);
1123}
1124
1125/*
1126 * Update the backup copies of the ext4 metadata. These don't need to be part
1127 * of the main resize transaction, because e2fsck will re-write them if there
1128 * is a problem (basically only OOM will cause a problem). However, we
1129 * _should_ update the backups if possible, in case the primary gets trashed
1130 * for some reason and we need to run e2fsck from a backup superblock. The
1131 * important part is that the new block and inode counts are in the backup
1132 * superblocks, and the location of the new group metadata in the GDT backups.
1133 *
1134 * We do not need take the s_resize_lock for this, because these
1135 * blocks are not otherwise touched by the filesystem code when it is
1136 * mounted. We don't need to worry about last changing from
1137 * sbi->s_groups_count, because the worst that can happen is that we
1138 * do not copy the full number of backups at this time. The resize
1139 * which changed s_groups_count will backup again.
1140 */
1141static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1142 int size, int meta_bg)
1143{
1144 struct ext4_sb_info *sbi = EXT4_SB(sb);
1145 ext4_group_t last;
1146 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1147 unsigned three = 1;
1148 unsigned five = 5;
1149 unsigned seven = 7;
1150 ext4_group_t group = 0;
1151 int rest = sb->s_blocksize - size;
1152 handle_t *handle;
1153 int err = 0, err2;
1154
1155 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1156 if (IS_ERR(handle)) {
1157 group = 1;
1158 err = PTR_ERR(handle);
1159 goto exit_err;
1160 }
1161
1162 if (meta_bg == 0) {
1163 group = ext4_list_backups(sb, &three, &five, &seven);
1164 last = sbi->s_groups_count;
1165 } else {
1166 group = ext4_get_group_number(sb, blk_off) + 1;
1167 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1168 }
1169
1170 while (group < sbi->s_groups_count) {
1171 struct buffer_head *bh;
1172 ext4_fsblk_t backup_block;
1173 int has_super = ext4_bg_has_super(sb, group);
1174 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1175
1176 /* Out of journal space, and can't get more - abort - so sad */
1177 err = ext4_resize_ensure_credits_batch(handle, 1);
1178 if (err < 0)
1179 break;
1180
1181 if (meta_bg == 0)
1182 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1183 else
1184 backup_block = first_block + has_super;
1185
1186 bh = sb_getblk(sb, backup_block);
1187 if (unlikely(!bh)) {
1188 err = -ENOMEM;
1189 break;
1190 }
1191 ext4_debug("update metadata backup %llu(+%llu)\n",
1192 backup_block, backup_block -
1193 ext4_group_first_block_no(sb, group));
1194 BUFFER_TRACE(bh, "get_write_access");
1195 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1196 EXT4_JTR_NONE))) {
1197 brelse(bh);
1198 break;
1199 }
1200 lock_buffer(bh);
1201 memcpy(bh->b_data, data, size);
1202 if (rest)
1203 memset(bh->b_data + size, 0, rest);
1204 if (has_super && (backup_block == first_block))
1205 ext4_set_block_group_nr(sb, bh->b_data, group);
1206 set_buffer_uptodate(bh);
1207 unlock_buffer(bh);
1208 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1209 if (unlikely(err))
1210 ext4_std_error(sb, err);
1211 brelse(bh);
1212
1213 if (meta_bg == 0)
1214 group = ext4_list_backups(sb, &three, &five, &seven);
1215 else if (group == last)
1216 break;
1217 else
1218 group = last;
1219 }
1220 if ((err2 = ext4_journal_stop(handle)) && !err)
1221 err = err2;
1222
1223 /*
1224 * Ugh! Need to have e2fsck write the backup copies. It is too
1225 * late to revert the resize, we shouldn't fail just because of
1226 * the backup copies (they are only needed in case of corruption).
1227 *
1228 * However, if we got here we have a journal problem too, so we
1229 * can't really start a transaction to mark the superblock.
1230 * Chicken out and just set the flag on the hope it will be written
1231 * to disk, and if not - we will simply wait until next fsck.
1232 */
1233exit_err:
1234 if (err) {
1235 ext4_warning(sb, "can't update backup for group %u (err %d), "
1236 "forcing fsck on next reboot", group, err);
1237 sbi->s_mount_state &= ~EXT4_VALID_FS;
1238 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1239 mark_buffer_dirty(sbi->s_sbh);
1240 }
1241}
1242
1243/*
1244 * ext4_add_new_descs() adds @count group descriptor of groups
1245 * starting at @group
1246 *
1247 * @handle: journal handle
1248 * @sb: super block
1249 * @group: the group no. of the first group desc to be added
1250 * @resize_inode: the resize inode
1251 * @count: number of group descriptors to be added
1252 */
1253static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1254 ext4_group_t group, struct inode *resize_inode,
1255 ext4_group_t count)
1256{
1257 struct ext4_sb_info *sbi = EXT4_SB(sb);
1258 struct ext4_super_block *es = sbi->s_es;
1259 struct buffer_head *gdb_bh;
1260 int i, gdb_off, gdb_num, err = 0;
1261 int meta_bg;
1262
1263 meta_bg = ext4_has_feature_meta_bg(sb);
1264 for (i = 0; i < count; i++, group++) {
1265 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1266 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1267
1268 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1269 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1270
1271 /*
1272 * We will only either add reserved group blocks to a backup group
1273 * or remove reserved blocks for the first group in a new group block.
1274 * Doing both would be mean more complex code, and sane people don't
1275 * use non-sparse filesystems anymore. This is already checked above.
1276 */
1277 if (gdb_off) {
1278 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1279 gdb_num);
1280 BUFFER_TRACE(gdb_bh, "get_write_access");
1281 err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1282 EXT4_JTR_NONE);
1283
1284 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1285 err = reserve_backup_gdb(handle, resize_inode, group);
1286 } else if (meta_bg != 0) {
1287 err = add_new_gdb_meta_bg(sb, handle, group);
1288 } else {
1289 err = add_new_gdb(handle, resize_inode, group);
1290 }
1291 if (err)
1292 break;
1293 }
1294 return err;
1295}
1296
1297static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1298{
1299 struct buffer_head *bh = sb_getblk(sb, block);
1300 if (unlikely(!bh))
1301 return NULL;
1302 if (!bh_uptodate_or_lock(bh)) {
1303 if (ext4_read_bh(bh, 0, NULL, false) < 0) {
1304 brelse(bh);
1305 return NULL;
1306 }
1307 }
1308
1309 return bh;
1310}
1311
1312static int ext4_set_bitmap_checksums(struct super_block *sb,
1313 struct ext4_group_desc *gdp,
1314 struct ext4_new_group_data *group_data)
1315{
1316 struct buffer_head *bh;
1317
1318 if (!ext4_has_metadata_csum(sb))
1319 return 0;
1320
1321 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1322 if (!bh)
1323 return -EIO;
1324 ext4_inode_bitmap_csum_set(sb, gdp, bh);
1325 brelse(bh);
1326
1327 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1328 if (!bh)
1329 return -EIO;
1330 ext4_block_bitmap_csum_set(sb, gdp, bh);
1331 brelse(bh);
1332
1333 return 0;
1334}
1335
1336/*
1337 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1338 */
1339static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1340 struct ext4_new_flex_group_data *flex_gd)
1341{
1342 struct ext4_new_group_data *group_data = flex_gd->groups;
1343 struct ext4_group_desc *gdp;
1344 struct ext4_sb_info *sbi = EXT4_SB(sb);
1345 struct buffer_head *gdb_bh;
1346 ext4_group_t group;
1347 __u16 *bg_flags = flex_gd->bg_flags;
1348 int i, gdb_off, gdb_num, err = 0;
1349
1350
1351 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1352 group = group_data->group;
1353
1354 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1355 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1356
1357 /*
1358 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1359 */
1360 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1361 /* Update group descriptor block for new group */
1362 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1363 gdb_off * EXT4_DESC_SIZE(sb));
1364
1365 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1366 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1367 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1368 err = ext4_set_bitmap_checksums(sb, gdp, group_data);
1369 if (err) {
1370 ext4_std_error(sb, err);
1371 break;
1372 }
1373
1374 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1375 ext4_free_group_clusters_set(sb, gdp,
1376 group_data->free_clusters_count);
1377 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1378 if (ext4_has_group_desc_csum(sb))
1379 ext4_itable_unused_set(sb, gdp,
1380 EXT4_INODES_PER_GROUP(sb));
1381 gdp->bg_flags = cpu_to_le16(*bg_flags);
1382 ext4_group_desc_csum_set(sb, group, gdp);
1383
1384 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1385 if (unlikely(err)) {
1386 ext4_std_error(sb, err);
1387 break;
1388 }
1389
1390 /*
1391 * We can allocate memory for mb_alloc based on the new group
1392 * descriptor
1393 */
1394 err = ext4_mb_add_groupinfo(sb, group, gdp);
1395 if (err)
1396 break;
1397 }
1398 return err;
1399}
1400
1401static void ext4_add_overhead(struct super_block *sb,
1402 const ext4_fsblk_t overhead)
1403{
1404 struct ext4_sb_info *sbi = EXT4_SB(sb);
1405 struct ext4_super_block *es = sbi->s_es;
1406
1407 sbi->s_overhead += overhead;
1408 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1409 smp_wmb();
1410}
1411
1412/*
1413 * ext4_update_super() updates the super block so that the newly added
1414 * groups can be seen by the filesystem.
1415 *
1416 * @sb: super block
1417 * @flex_gd: new added groups
1418 */
1419static void ext4_update_super(struct super_block *sb,
1420 struct ext4_new_flex_group_data *flex_gd)
1421{
1422 ext4_fsblk_t blocks_count = 0;
1423 ext4_fsblk_t free_blocks = 0;
1424 ext4_fsblk_t reserved_blocks = 0;
1425 struct ext4_new_group_data *group_data = flex_gd->groups;
1426 struct ext4_sb_info *sbi = EXT4_SB(sb);
1427 struct ext4_super_block *es = sbi->s_es;
1428 int i;
1429
1430 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1431 /*
1432 * Make the new blocks and inodes valid next. We do this before
1433 * increasing the group count so that once the group is enabled,
1434 * all of its blocks and inodes are already valid.
1435 *
1436 * We always allocate group-by-group, then block-by-block or
1437 * inode-by-inode within a group, so enabling these
1438 * blocks/inodes before the group is live won't actually let us
1439 * allocate the new space yet.
1440 */
1441 for (i = 0; i < flex_gd->count; i++) {
1442 blocks_count += group_data[i].blocks_count;
1443 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1444 }
1445
1446 reserved_blocks = ext4_r_blocks_count(es) * 100;
1447 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1448 reserved_blocks *= blocks_count;
1449 do_div(reserved_blocks, 100);
1450
1451 lock_buffer(sbi->s_sbh);
1452 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1453 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1454 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1455 flex_gd->count);
1456 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1457 flex_gd->count);
1458
1459 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1460 /*
1461 * We need to protect s_groups_count against other CPUs seeing
1462 * inconsistent state in the superblock.
1463 *
1464 * The precise rules we use are:
1465 *
1466 * * Writers must perform a smp_wmb() after updating all
1467 * dependent data and before modifying the groups count
1468 *
1469 * * Readers must perform an smp_rmb() after reading the groups
1470 * count and before reading any dependent data.
1471 *
1472 * NB. These rules can be relaxed when checking the group count
1473 * while freeing data, as we can only allocate from a block
1474 * group after serialising against the group count, and we can
1475 * only then free after serialising in turn against that
1476 * allocation.
1477 */
1478 smp_wmb();
1479
1480 /* Update the global fs size fields */
1481 sbi->s_groups_count += flex_gd->count;
1482 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1483 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1484
1485 /* Update the reserved block counts only once the new group is
1486 * active. */
1487 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1488 reserved_blocks);
1489
1490 /* Update the free space counts */
1491 percpu_counter_add(&sbi->s_freeclusters_counter,
1492 EXT4_NUM_B2C(sbi, free_blocks));
1493 percpu_counter_add(&sbi->s_freeinodes_counter,
1494 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1495
1496 ext4_debug("free blocks count %llu",
1497 percpu_counter_read(&sbi->s_freeclusters_counter));
1498 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1499 ext4_group_t flex_group;
1500 struct flex_groups *fg;
1501
1502 flex_group = ext4_flex_group(sbi, group_data[0].group);
1503 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1504 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1505 &fg->free_clusters);
1506 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1507 &fg->free_inodes);
1508 }
1509
1510 /*
1511 * Update the fs overhead information.
1512 *
1513 * For bigalloc, if the superblock already has a properly calculated
1514 * overhead, update it with a value based on numbers already computed
1515 * above for the newly allocated capacity.
1516 */
1517 if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1518 ext4_add_overhead(sb,
1519 EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1520 else
1521 ext4_calculate_overhead(sb);
1522 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1523
1524 ext4_superblock_csum_set(sb);
1525 unlock_buffer(sbi->s_sbh);
1526 if (test_opt(sb, DEBUG))
1527 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1528 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1529 blocks_count, free_blocks, reserved_blocks);
1530}
1531
1532/* Add a flex group to an fs. Ensure we handle all possible error conditions
1533 * _before_ we start modifying the filesystem, because we cannot abort the
1534 * transaction and not have it write the data to disk.
1535 */
1536static int ext4_flex_group_add(struct super_block *sb,
1537 struct inode *resize_inode,
1538 struct ext4_new_flex_group_data *flex_gd)
1539{
1540 struct ext4_sb_info *sbi = EXT4_SB(sb);
1541 struct ext4_super_block *es = sbi->s_es;
1542 ext4_fsblk_t o_blocks_count;
1543 ext4_grpblk_t last;
1544 ext4_group_t group;
1545 handle_t *handle;
1546 unsigned reserved_gdb;
1547 int err = 0, err2 = 0, credit;
1548
1549 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1550
1551 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1552 o_blocks_count = ext4_blocks_count(es);
1553 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1554 BUG_ON(last);
1555
1556 err = setup_new_flex_group_blocks(sb, flex_gd);
1557 if (err)
1558 goto exit;
1559 /*
1560 * We will always be modifying at least the superblock and GDT
1561 * blocks. If we are adding a group past the last current GDT block,
1562 * we will also modify the inode and the dindirect block. If we
1563 * are adding a group with superblock/GDT backups we will also
1564 * modify each of the reserved GDT dindirect blocks.
1565 */
1566 credit = 3; /* sb, resize inode, resize inode dindirect */
1567 /* GDT blocks */
1568 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1569 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1570 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1571 if (IS_ERR(handle)) {
1572 err = PTR_ERR(handle);
1573 goto exit;
1574 }
1575
1576 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1577 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1578 EXT4_JTR_NONE);
1579 if (err)
1580 goto exit_journal;
1581
1582 group = flex_gd->groups[0].group;
1583 BUG_ON(group != sbi->s_groups_count);
1584 err = ext4_add_new_descs(handle, sb, group,
1585 resize_inode, flex_gd->count);
1586 if (err)
1587 goto exit_journal;
1588
1589 err = ext4_setup_new_descs(handle, sb, flex_gd);
1590 if (err)
1591 goto exit_journal;
1592
1593 ext4_update_super(sb, flex_gd);
1594
1595 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1596
1597exit_journal:
1598 err2 = ext4_journal_stop(handle);
1599 if (!err)
1600 err = err2;
1601
1602 if (!err) {
1603 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1604 int gdb_num_end = ((group + flex_gd->count - 1) /
1605 EXT4_DESC_PER_BLOCK(sb));
1606 int meta_bg = ext4_has_feature_meta_bg(sb) &&
1607 gdb_num >= le32_to_cpu(es->s_first_meta_bg);
1608 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1609 ext4_group_first_block_no(sb, 0);
1610
1611 update_backups(sb, ext4_group_first_block_no(sb, 0),
1612 (char *)es, sizeof(struct ext4_super_block), 0);
1613 for (; gdb_num <= gdb_num_end; gdb_num++) {
1614 struct buffer_head *gdb_bh;
1615
1616 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1617 gdb_num);
1618 update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1619 gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1620 }
1621 }
1622exit:
1623 return err;
1624}
1625
1626static int ext4_setup_next_flex_gd(struct super_block *sb,
1627 struct ext4_new_flex_group_data *flex_gd,
1628 ext4_fsblk_t n_blocks_count)
1629{
1630 struct ext4_sb_info *sbi = EXT4_SB(sb);
1631 struct ext4_super_block *es = sbi->s_es;
1632 struct ext4_new_group_data *group_data = flex_gd->groups;
1633 ext4_fsblk_t o_blocks_count;
1634 ext4_group_t n_group;
1635 ext4_group_t group;
1636 ext4_group_t last_group;
1637 ext4_grpblk_t last;
1638 ext4_grpblk_t clusters_per_group;
1639 unsigned long i;
1640
1641 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1642
1643 o_blocks_count = ext4_blocks_count(es);
1644
1645 if (o_blocks_count == n_blocks_count)
1646 return 0;
1647
1648 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1649 BUG_ON(last);
1650 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1651
1652 last_group = group | (flex_gd->resize_bg - 1);
1653 if (last_group > n_group)
1654 last_group = n_group;
1655
1656 flex_gd->count = last_group - group + 1;
1657
1658 for (i = 0; i < flex_gd->count; i++) {
1659 int overhead;
1660
1661 group_data[i].group = group + i;
1662 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1663 overhead = ext4_group_overhead_blocks(sb, group + i);
1664 group_data[i].mdata_blocks = overhead;
1665 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1666 if (ext4_has_group_desc_csum(sb)) {
1667 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1668 EXT4_BG_INODE_UNINIT;
1669 if (!test_opt(sb, INIT_INODE_TABLE))
1670 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1671 } else
1672 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1673 }
1674
1675 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1676 /* We need to initialize block bitmap of last group. */
1677 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1678
1679 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1680 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1681 group_data[i - 1].free_clusters_count -= clusters_per_group -
1682 last - 1;
1683 }
1684
1685 return 1;
1686}
1687
1688/* Add group descriptor data to an existing or new group descriptor block.
1689 * Ensure we handle all possible error conditions _before_ we start modifying
1690 * the filesystem, because we cannot abort the transaction and not have it
1691 * write the data to disk.
1692 *
1693 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1694 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1695 *
1696 * We only need to hold the superblock lock while we are actually adding
1697 * in the new group's counts to the superblock. Prior to that we have
1698 * not really "added" the group at all. We re-check that we are still
1699 * adding in the last group in case things have changed since verifying.
1700 */
1701int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1702{
1703 struct ext4_new_flex_group_data flex_gd;
1704 struct ext4_sb_info *sbi = EXT4_SB(sb);
1705 struct ext4_super_block *es = sbi->s_es;
1706 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1707 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1708 struct inode *inode = NULL;
1709 int gdb_off;
1710 int err;
1711 __u16 bg_flags = 0;
1712
1713 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1714
1715 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1716 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1717 return -EPERM;
1718 }
1719
1720 if (ext4_blocks_count(es) + input->blocks_count <
1721 ext4_blocks_count(es)) {
1722 ext4_warning(sb, "blocks_count overflow");
1723 return -EINVAL;
1724 }
1725
1726 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1727 le32_to_cpu(es->s_inodes_count)) {
1728 ext4_warning(sb, "inodes_count overflow");
1729 return -EINVAL;
1730 }
1731
1732 if (reserved_gdb || gdb_off == 0) {
1733 if (!ext4_has_feature_resize_inode(sb) ||
1734 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1735 ext4_warning(sb,
1736 "No reserved GDT blocks, can't resize");
1737 return -EPERM;
1738 }
1739 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1740 if (IS_ERR(inode)) {
1741 ext4_warning(sb, "Error opening resize inode");
1742 return PTR_ERR(inode);
1743 }
1744 }
1745
1746
1747 err = verify_group_input(sb, input);
1748 if (err)
1749 goto out;
1750
1751 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1752 if (err)
1753 goto out;
1754
1755 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1756 if (err)
1757 goto out;
1758
1759 flex_gd.count = 1;
1760 flex_gd.groups = input;
1761 flex_gd.bg_flags = &bg_flags;
1762 err = ext4_flex_group_add(sb, inode, &flex_gd);
1763out:
1764 iput(inode);
1765 return err;
1766} /* ext4_group_add */
1767
1768/*
1769 * extend a group without checking assuming that checking has been done.
1770 */
1771static int ext4_group_extend_no_check(struct super_block *sb,
1772 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1773{
1774 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1775 handle_t *handle;
1776 int err = 0, err2;
1777
1778 /* We will update the superblock, one block bitmap, and
1779 * one group descriptor via ext4_group_add_blocks().
1780 */
1781 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1782 if (IS_ERR(handle)) {
1783 err = PTR_ERR(handle);
1784 ext4_warning(sb, "error %d on journal start", err);
1785 return err;
1786 }
1787
1788 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1789 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1790 EXT4_JTR_NONE);
1791 if (err) {
1792 ext4_warning(sb, "error %d on journal write access", err);
1793 goto errout;
1794 }
1795
1796 lock_buffer(EXT4_SB(sb)->s_sbh);
1797 ext4_blocks_count_set(es, o_blocks_count + add);
1798 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1799 ext4_superblock_csum_set(sb);
1800 unlock_buffer(EXT4_SB(sb)->s_sbh);
1801 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1802 o_blocks_count + add);
1803 /* We add the blocks to the bitmap and set the group need init bit */
1804 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1805 if (err)
1806 goto errout;
1807 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1808 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1809 o_blocks_count + add);
1810errout:
1811 err2 = ext4_journal_stop(handle);
1812 if (err2 && !err)
1813 err = err2;
1814
1815 if (!err) {
1816 if (test_opt(sb, DEBUG))
1817 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1818 "blocks\n", ext4_blocks_count(es));
1819 update_backups(sb, ext4_group_first_block_no(sb, 0),
1820 (char *)es, sizeof(struct ext4_super_block), 0);
1821 }
1822 return err;
1823}
1824
1825/*
1826 * Extend the filesystem to the new number of blocks specified. This entry
1827 * point is only used to extend the current filesystem to the end of the last
1828 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1829 * for emergencies (because it has no dependencies on reserved blocks).
1830 *
1831 * If we _really_ wanted, we could use default values to call ext4_group_add()
1832 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1833 * GDT blocks are reserved to grow to the desired size.
1834 */
1835int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1836 ext4_fsblk_t n_blocks_count)
1837{
1838 ext4_fsblk_t o_blocks_count;
1839 ext4_grpblk_t last;
1840 ext4_grpblk_t add;
1841 struct buffer_head *bh;
1842 ext4_group_t group;
1843
1844 o_blocks_count = ext4_blocks_count(es);
1845
1846 if (test_opt(sb, DEBUG))
1847 ext4_msg(sb, KERN_DEBUG,
1848 "extending last group from %llu to %llu blocks",
1849 o_blocks_count, n_blocks_count);
1850
1851 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1852 return 0;
1853
1854 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1855 ext4_msg(sb, KERN_ERR,
1856 "filesystem too large to resize to %llu blocks safely",
1857 n_blocks_count);
1858 return -EINVAL;
1859 }
1860
1861 if (n_blocks_count < o_blocks_count) {
1862 ext4_warning(sb, "can't shrink FS - resize aborted");
1863 return -EINVAL;
1864 }
1865
1866 /* Handle the remaining blocks in the last group only. */
1867 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1868
1869 if (last == 0) {
1870 ext4_warning(sb, "need to use ext2online to resize further");
1871 return -EPERM;
1872 }
1873
1874 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1875
1876 if (o_blocks_count + add < o_blocks_count) {
1877 ext4_warning(sb, "blocks_count overflow");
1878 return -EINVAL;
1879 }
1880
1881 if (o_blocks_count + add > n_blocks_count)
1882 add = n_blocks_count - o_blocks_count;
1883
1884 if (o_blocks_count + add < n_blocks_count)
1885 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1886 o_blocks_count + add, add);
1887
1888 /* See if the device is actually as big as what was requested */
1889 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1890 if (IS_ERR(bh)) {
1891 ext4_warning(sb, "can't read last block, resize aborted");
1892 return -ENOSPC;
1893 }
1894 brelse(bh);
1895
1896 return ext4_group_extend_no_check(sb, o_blocks_count, add);
1897} /* ext4_group_extend */
1898
1899
1900static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1901{
1902 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1903}
1904
1905/*
1906 * Release the resize inode and drop the resize_inode feature if there
1907 * are no more reserved gdt blocks, and then convert the file system
1908 * to enable meta_bg
1909 */
1910static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1911{
1912 handle_t *handle;
1913 struct ext4_sb_info *sbi = EXT4_SB(sb);
1914 struct ext4_super_block *es = sbi->s_es;
1915 struct ext4_inode_info *ei = EXT4_I(inode);
1916 ext4_fsblk_t nr;
1917 int i, ret, err = 0;
1918 int credits = 1;
1919
1920 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1921 if (inode) {
1922 if (es->s_reserved_gdt_blocks) {
1923 ext4_error(sb, "Unexpected non-zero "
1924 "s_reserved_gdt_blocks");
1925 return -EPERM;
1926 }
1927
1928 /* Do a quick sanity check of the resize inode */
1929 if (inode->i_blocks != 1 << (inode->i_blkbits -
1930 (9 - sbi->s_cluster_bits)))
1931 goto invalid_resize_inode;
1932 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1933 if (i == EXT4_DIND_BLOCK) {
1934 if (ei->i_data[i])
1935 continue;
1936 else
1937 goto invalid_resize_inode;
1938 }
1939 if (ei->i_data[i])
1940 goto invalid_resize_inode;
1941 }
1942 credits += 3; /* block bitmap, bg descriptor, resize inode */
1943 }
1944
1945 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1946 if (IS_ERR(handle))
1947 return PTR_ERR(handle);
1948
1949 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1950 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1951 EXT4_JTR_NONE);
1952 if (err)
1953 goto errout;
1954
1955 lock_buffer(sbi->s_sbh);
1956 ext4_clear_feature_resize_inode(sb);
1957 ext4_set_feature_meta_bg(sb);
1958 sbi->s_es->s_first_meta_bg =
1959 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1960 ext4_superblock_csum_set(sb);
1961 unlock_buffer(sbi->s_sbh);
1962
1963 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1964 if (err) {
1965 ext4_std_error(sb, err);
1966 goto errout;
1967 }
1968
1969 if (inode) {
1970 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1971 ext4_free_blocks(handle, inode, NULL, nr, 1,
1972 EXT4_FREE_BLOCKS_METADATA |
1973 EXT4_FREE_BLOCKS_FORGET);
1974 ei->i_data[EXT4_DIND_BLOCK] = 0;
1975 inode->i_blocks = 0;
1976
1977 err = ext4_mark_inode_dirty(handle, inode);
1978 if (err)
1979 ext4_std_error(sb, err);
1980 }
1981
1982errout:
1983 ret = ext4_journal_stop(handle);
1984 return err ? err : ret;
1985
1986invalid_resize_inode:
1987 ext4_error(sb, "corrupted/inconsistent resize inode");
1988 return -EINVAL;
1989}
1990
1991/*
1992 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1993 *
1994 * @sb: super block of the fs to be resized
1995 * @n_blocks_count: the number of blocks resides in the resized fs
1996 */
1997int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1998{
1999 struct ext4_new_flex_group_data *flex_gd = NULL;
2000 struct ext4_sb_info *sbi = EXT4_SB(sb);
2001 struct ext4_super_block *es = sbi->s_es;
2002 struct buffer_head *bh;
2003 struct inode *resize_inode = NULL;
2004 ext4_grpblk_t add, offset;
2005 unsigned long n_desc_blocks;
2006 unsigned long o_desc_blocks;
2007 ext4_group_t o_group;
2008 ext4_group_t n_group;
2009 ext4_fsblk_t o_blocks_count;
2010 ext4_fsblk_t n_blocks_count_retry = 0;
2011 unsigned long last_update_time = 0;
2012 int err = 0;
2013 int meta_bg;
2014 unsigned int flexbg_size = ext4_flex_bg_size(sbi);
2015
2016 /* See if the device is actually as big as what was requested */
2017 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2018 if (IS_ERR(bh)) {
2019 ext4_warning(sb, "can't read last block, resize aborted");
2020 return -ENOSPC;
2021 }
2022 brelse(bh);
2023
2024 /*
2025 * For bigalloc, trim the requested size to the nearest cluster
2026 * boundary to avoid creating an unusable filesystem. We do this
2027 * silently, instead of returning an error, to avoid breaking
2028 * callers that blindly resize the filesystem to the full size of
2029 * the underlying block device.
2030 */
2031 if (ext4_has_feature_bigalloc(sb))
2032 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2033
2034retry:
2035 o_blocks_count = ext4_blocks_count(es);
2036
2037 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2038 "to %llu blocks", o_blocks_count, n_blocks_count);
2039
2040 if (n_blocks_count < o_blocks_count) {
2041 /* On-line shrinking not supported */
2042 ext4_warning(sb, "can't shrink FS - resize aborted");
2043 return -EINVAL;
2044 }
2045
2046 if (n_blocks_count == o_blocks_count)
2047 /* Nothing need to do */
2048 return 0;
2049
2050 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2051 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2052 ext4_warning(sb, "resize would cause inodes_count overflow");
2053 return -EINVAL;
2054 }
2055 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2056
2057 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2058 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2059
2060 meta_bg = ext4_has_feature_meta_bg(sb);
2061
2062 if (ext4_has_feature_resize_inode(sb)) {
2063 if (meta_bg) {
2064 ext4_error(sb, "resize_inode and meta_bg enabled "
2065 "simultaneously");
2066 return -EINVAL;
2067 }
2068 if (n_desc_blocks > o_desc_blocks +
2069 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2070 n_blocks_count_retry = n_blocks_count;
2071 n_desc_blocks = o_desc_blocks +
2072 le16_to_cpu(es->s_reserved_gdt_blocks);
2073 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2074 n_blocks_count = (ext4_fsblk_t)n_group *
2075 EXT4_BLOCKS_PER_GROUP(sb) +
2076 le32_to_cpu(es->s_first_data_block);
2077 n_group--; /* set to last group number */
2078 }
2079
2080 if (!resize_inode)
2081 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2082 EXT4_IGET_SPECIAL);
2083 if (IS_ERR(resize_inode)) {
2084 ext4_warning(sb, "Error opening resize inode");
2085 return PTR_ERR(resize_inode);
2086 }
2087 }
2088
2089 if ((!resize_inode && !meta_bg && n_desc_blocks > o_desc_blocks) || n_blocks_count == o_blocks_count) {
2090 err = ext4_convert_meta_bg(sb, resize_inode);
2091 if (err)
2092 goto out;
2093 if (resize_inode) {
2094 iput(resize_inode);
2095 resize_inode = NULL;
2096 }
2097 if (n_blocks_count_retry) {
2098 n_blocks_count = n_blocks_count_retry;
2099 n_blocks_count_retry = 0;
2100 goto retry;
2101 }
2102 }
2103
2104 /*
2105 * Make sure the last group has enough space so that it's
2106 * guaranteed to have enough space for all metadata blocks
2107 * that it might need to hold. (We might not need to store
2108 * the inode table blocks in the last block group, but there
2109 * will be cases where this might be needed.)
2110 */
2111 if ((ext4_group_first_block_no(sb, n_group) +
2112 ext4_group_overhead_blocks(sb, n_group) + 2 +
2113 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2114 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2115 n_group--;
2116 n_blocks_count_retry = 0;
2117 if (resize_inode) {
2118 iput(resize_inode);
2119 resize_inode = NULL;
2120 }
2121 goto retry;
2122 }
2123
2124 /* extend the last group */
2125 if (n_group == o_group)
2126 add = n_blocks_count - o_blocks_count;
2127 else
2128 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2129 if (add > 0) {
2130 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2131 if (err)
2132 goto out;
2133 }
2134
2135 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2136 goto out;
2137
2138 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2139 if (err)
2140 goto out;
2141
2142 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2143 if (err)
2144 goto out;
2145
2146 flex_gd = alloc_flex_gd(flexbg_size, o_group, n_group);
2147 if (flex_gd == NULL) {
2148 err = -ENOMEM;
2149 goto out;
2150 }
2151
2152 /* Add flex groups. Note that a regular group is a
2153 * flex group with 1 group.
2154 */
2155 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2156 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2157 if (last_update_time)
2158 ext4_msg(sb, KERN_INFO,
2159 "resized to %llu blocks",
2160 ext4_blocks_count(es));
2161 last_update_time = jiffies;
2162 }
2163 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2164 break;
2165 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2166 if (unlikely(err))
2167 break;
2168 }
2169
2170 if (!err && n_blocks_count_retry) {
2171 n_blocks_count = n_blocks_count_retry;
2172 n_blocks_count_retry = 0;
2173 free_flex_gd(flex_gd);
2174 flex_gd = NULL;
2175 if (resize_inode) {
2176 iput(resize_inode);
2177 resize_inode = NULL;
2178 }
2179 goto retry;
2180 }
2181
2182out:
2183 if (flex_gd)
2184 free_flex_gd(flex_gd);
2185 if (resize_inode != NULL)
2186 iput(resize_inode);
2187 if (err)
2188 ext4_warning(sb, "error (%d) occurred during "
2189 "file system resize", err);
2190 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2191 ext4_blocks_count(es));
2192 return err;
2193}