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 #define EXT4FS_DEBUG
14
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17
18 #include "ext4_jbd2.h"
19
20 struct ext4_rcu_ptr {
21 struct rcu_head rcu;
22 void *ptr;
23 };
24
ext4_rcu_ptr_callback(struct rcu_head * head)25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
26 {
27 struct ext4_rcu_ptr *ptr;
28
29 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
30 kvfree(ptr->ptr);
31 kfree(ptr);
32 }
33
ext4_kvfree_array_rcu(void * to_free)34 void ext4_kvfree_array_rcu(void *to_free)
35 {
36 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
37
38 if (ptr) {
39 ptr->ptr = to_free;
40 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
41 return;
42 }
43 synchronize_rcu();
44 kvfree(to_free);
45 }
46
ext4_resize_begin(struct super_block * sb)47 int ext4_resize_begin(struct super_block *sb)
48 {
49 struct ext4_sb_info *sbi = EXT4_SB(sb);
50 int ret = 0;
51
52 if (!capable(CAP_SYS_RESOURCE))
53 return -EPERM;
54
55 /*
56 * If the reserved GDT blocks is non-zero, the resize_inode feature
57 * should always be set.
58 */
59 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
60 !ext4_has_feature_resize_inode(sb)) {
61 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
62 return -EFSCORRUPTED;
63 }
64
65 /*
66 * If we are not using the primary superblock/GDT copy don't resize,
67 * because the user tools have no way of handling this. Probably a
68 * bad time to do it anyways.
69 */
70 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
71 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
72 ext4_warning(sb, "won't resize using backup superblock at %llu",
73 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
74 return -EPERM;
75 }
76
77 /*
78 * We are not allowed to do online-resizing on a filesystem mounted
79 * with error, because it can destroy the filesystem easily.
80 */
81 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
82 ext4_warning(sb, "There are errors in the filesystem, "
83 "so online resizing is not allowed");
84 return -EPERM;
85 }
86
87 if (ext4_has_feature_sparse_super2(sb)) {
88 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
89 return -EOPNOTSUPP;
90 }
91
92 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
93 &EXT4_SB(sb)->s_ext4_flags))
94 ret = -EBUSY;
95
96 return ret;
97 }
98
ext4_resize_end(struct super_block * sb)99 void ext4_resize_end(struct super_block *sb)
100 {
101 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
102 smp_mb__after_atomic();
103 }
104
ext4_meta_bg_first_group(struct super_block * sb,ext4_group_t group)105 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
106 ext4_group_t group) {
107 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
108 EXT4_DESC_PER_BLOCK_BITS(sb);
109 }
110
ext4_meta_bg_first_block_no(struct super_block * sb,ext4_group_t group)111 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
112 ext4_group_t group) {
113 group = ext4_meta_bg_first_group(sb, group);
114 return ext4_group_first_block_no(sb, group);
115 }
116
ext4_group_overhead_blocks(struct super_block * sb,ext4_group_t group)117 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
118 ext4_group_t group) {
119 ext4_grpblk_t overhead;
120 overhead = ext4_bg_num_gdb(sb, group);
121 if (ext4_bg_has_super(sb, group))
122 overhead += 1 +
123 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
124 return overhead;
125 }
126
127 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
128 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
129
verify_group_input(struct super_block * sb,struct ext4_new_group_data * input)130 static int verify_group_input(struct super_block *sb,
131 struct ext4_new_group_data *input)
132 {
133 struct ext4_sb_info *sbi = EXT4_SB(sb);
134 struct ext4_super_block *es = sbi->s_es;
135 ext4_fsblk_t start = ext4_blocks_count(es);
136 ext4_fsblk_t end = start + input->blocks_count;
137 ext4_group_t group = input->group;
138 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
139 unsigned overhead;
140 ext4_fsblk_t metaend;
141 struct buffer_head *bh = NULL;
142 ext4_grpblk_t free_blocks_count, offset;
143 int err = -EINVAL;
144
145 if (group != sbi->s_groups_count) {
146 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
147 input->group, sbi->s_groups_count);
148 return -EINVAL;
149 }
150
151 overhead = ext4_group_overhead_blocks(sb, group);
152 metaend = start + overhead;
153 input->free_clusters_count = free_blocks_count =
154 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
155
156 if (test_opt(sb, DEBUG))
157 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
158 "(%d free, %u reserved)\n",
159 ext4_bg_has_super(sb, input->group) ? "normal" :
160 "no-super", input->group, input->blocks_count,
161 free_blocks_count, input->reserved_blocks);
162
163 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
164 if (offset != 0)
165 ext4_warning(sb, "Last group not full");
166 else if (input->reserved_blocks > input->blocks_count / 5)
167 ext4_warning(sb, "Reserved blocks too high (%u)",
168 input->reserved_blocks);
169 else if (free_blocks_count < 0)
170 ext4_warning(sb, "Bad blocks count %u",
171 input->blocks_count);
172 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
173 err = PTR_ERR(bh);
174 bh = NULL;
175 ext4_warning(sb, "Cannot read last block (%llu)",
176 end - 1);
177 } else if (outside(input->block_bitmap, start, end))
178 ext4_warning(sb, "Block bitmap not in group (block %llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (outside(input->inode_bitmap, start, end))
181 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
182 (unsigned long long)input->inode_bitmap);
183 else if (outside(input->inode_table, start, end) ||
184 outside(itend - 1, start, end))
185 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
186 (unsigned long long)input->inode_table, itend - 1);
187 else if (input->inode_bitmap == input->block_bitmap)
188 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
189 (unsigned long long)input->block_bitmap);
190 else if (inside(input->block_bitmap, input->inode_table, itend))
191 ext4_warning(sb, "Block bitmap (%llu) in inode table "
192 "(%llu-%llu)",
193 (unsigned long long)input->block_bitmap,
194 (unsigned long long)input->inode_table, itend - 1);
195 else if (inside(input->inode_bitmap, input->inode_table, itend))
196 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
197 "(%llu-%llu)",
198 (unsigned long long)input->inode_bitmap,
199 (unsigned long long)input->inode_table, itend - 1);
200 else if (inside(input->block_bitmap, start, metaend))
201 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
202 (unsigned long long)input->block_bitmap,
203 start, metaend - 1);
204 else if (inside(input->inode_bitmap, start, metaend))
205 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
206 (unsigned long long)input->inode_bitmap,
207 start, metaend - 1);
208 else if (inside(input->inode_table, start, metaend) ||
209 inside(itend - 1, start, metaend))
210 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
211 "(%llu-%llu)",
212 (unsigned long long)input->inode_table,
213 itend - 1, start, metaend - 1);
214 else
215 err = 0;
216 brelse(bh);
217
218 return err;
219 }
220
221 /*
222 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
223 * group each time.
224 */
225 struct ext4_new_flex_group_data {
226 struct ext4_new_group_data *groups; /* new_group_data for groups
227 in the flex group */
228 __u16 *bg_flags; /* block group flags of groups
229 in @groups */
230 ext4_group_t resize_bg; /* number of allocated
231 new_group_data */
232 ext4_group_t count; /* number of groups in @groups
233 */
234 };
235
236 /*
237 * Avoiding memory allocation failures due to too many groups added each time.
238 */
239 #define MAX_RESIZE_BG 16384
240
241 /*
242 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
243 * @flexbg_size.
244 *
245 * Returns NULL on failure otherwise address of the allocated structure.
246 */
alloc_flex_gd(unsigned int flexbg_size)247 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size)
248 {
249 struct ext4_new_flex_group_data *flex_gd;
250
251 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
252 if (flex_gd == NULL)
253 goto out3;
254
255 if (unlikely(flexbg_size > MAX_RESIZE_BG))
256 flex_gd->resize_bg = MAX_RESIZE_BG;
257 else
258 flex_gd->resize_bg = flexbg_size;
259
260 flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
261 sizeof(struct ext4_new_group_data),
262 GFP_NOFS);
263 if (flex_gd->groups == NULL)
264 goto out2;
265
266 flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
267 GFP_NOFS);
268 if (flex_gd->bg_flags == NULL)
269 goto out1;
270
271 return flex_gd;
272
273 out1:
274 kfree(flex_gd->groups);
275 out2:
276 kfree(flex_gd);
277 out3:
278 return NULL;
279 }
280
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)281 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
282 {
283 kfree(flex_gd->bg_flags);
284 kfree(flex_gd->groups);
285 kfree(flex_gd);
286 }
287
288 /*
289 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
290 * and inode tables for a flex group.
291 *
292 * This function is used by 64bit-resize. Note that this function allocates
293 * group tables from the 1st group of groups contained by @flexgd, which may
294 * be a partial of a flex group.
295 *
296 * @sb: super block of fs to which the groups belongs
297 *
298 * Returns 0 on a successful allocation of the metadata blocks in the
299 * block group.
300 */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,unsigned int flexbg_size)301 static int ext4_alloc_group_tables(struct super_block *sb,
302 struct ext4_new_flex_group_data *flex_gd,
303 unsigned int flexbg_size)
304 {
305 struct ext4_new_group_data *group_data = flex_gd->groups;
306 ext4_fsblk_t start_blk;
307 ext4_fsblk_t last_blk;
308 ext4_group_t src_group;
309 ext4_group_t bb_index = 0;
310 ext4_group_t ib_index = 0;
311 ext4_group_t it_index = 0;
312 ext4_group_t group;
313 ext4_group_t last_group;
314 unsigned overhead;
315 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
316 int i;
317
318 BUG_ON(flex_gd->count == 0 || group_data == NULL);
319
320 src_group = group_data[0].group;
321 last_group = src_group + flex_gd->count - 1;
322
323 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
324 (last_group & ~(flexbg_size - 1))));
325 next_group:
326 group = group_data[0].group;
327 if (src_group >= group_data[0].group + flex_gd->count)
328 return -ENOSPC;
329 start_blk = ext4_group_first_block_no(sb, src_group);
330 last_blk = start_blk + group_data[src_group - group].blocks_count;
331
332 overhead = ext4_group_overhead_blocks(sb, src_group);
333
334 start_blk += overhead;
335
336 /* We collect contiguous blocks as much as possible. */
337 src_group++;
338 for (; src_group <= last_group; src_group++) {
339 overhead = ext4_group_overhead_blocks(sb, src_group);
340 if (overhead == 0)
341 last_blk += group_data[src_group - group].blocks_count;
342 else
343 break;
344 }
345
346 /* Allocate block bitmaps */
347 for (; bb_index < flex_gd->count; bb_index++) {
348 if (start_blk >= last_blk)
349 goto next_group;
350 group_data[bb_index].block_bitmap = start_blk++;
351 group = ext4_get_group_number(sb, start_blk - 1);
352 group -= group_data[0].group;
353 group_data[group].mdata_blocks++;
354 flex_gd->bg_flags[group] &= uninit_mask;
355 }
356
357 /* Allocate inode bitmaps */
358 for (; ib_index < flex_gd->count; ib_index++) {
359 if (start_blk >= last_blk)
360 goto next_group;
361 group_data[ib_index].inode_bitmap = start_blk++;
362 group = ext4_get_group_number(sb, start_blk - 1);
363 group -= group_data[0].group;
364 group_data[group].mdata_blocks++;
365 flex_gd->bg_flags[group] &= uninit_mask;
366 }
367
368 /* Allocate inode tables */
369 for (; it_index < flex_gd->count; it_index++) {
370 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
371 ext4_fsblk_t next_group_start;
372
373 if (start_blk + itb > last_blk)
374 goto next_group;
375 group_data[it_index].inode_table = start_blk;
376 group = ext4_get_group_number(sb, start_blk);
377 next_group_start = ext4_group_first_block_no(sb, group + 1);
378 group -= group_data[0].group;
379
380 if (start_blk + itb > next_group_start) {
381 flex_gd->bg_flags[group + 1] &= uninit_mask;
382 overhead = start_blk + itb - next_group_start;
383 group_data[group + 1].mdata_blocks += overhead;
384 itb -= overhead;
385 }
386
387 group_data[group].mdata_blocks += itb;
388 flex_gd->bg_flags[group] &= uninit_mask;
389 start_blk += EXT4_SB(sb)->s_itb_per_group;
390 }
391
392 /* Update free clusters count to exclude metadata blocks */
393 for (i = 0; i < flex_gd->count; i++) {
394 group_data[i].free_clusters_count -=
395 EXT4_NUM_B2C(EXT4_SB(sb),
396 group_data[i].mdata_blocks);
397 }
398
399 if (test_opt(sb, DEBUG)) {
400 int i;
401 group = group_data[0].group;
402
403 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
404 "%u groups, flexbg size is %u:\n", flex_gd->count,
405 flexbg_size);
406
407 for (i = 0; i < flex_gd->count; i++) {
408 ext4_debug(
409 "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
410 ext4_bg_has_super(sb, group + i) ? "normal" :
411 "no-super", group + i,
412 group_data[i].blocks_count,
413 group_data[i].free_clusters_count,
414 group_data[i].mdata_blocks);
415 }
416 }
417 return 0;
418 }
419
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)420 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
421 ext4_fsblk_t blk)
422 {
423 struct buffer_head *bh;
424 int err;
425
426 bh = sb_getblk(sb, blk);
427 if (unlikely(!bh))
428 return ERR_PTR(-ENOMEM);
429 BUFFER_TRACE(bh, "get_write_access");
430 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
431 if (err) {
432 brelse(bh);
433 bh = ERR_PTR(err);
434 } else {
435 memset(bh->b_data, 0, sb->s_blocksize);
436 set_buffer_uptodate(bh);
437 }
438
439 return bh;
440 }
441
ext4_resize_ensure_credits_batch(handle_t * handle,int credits)442 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
443 {
444 return ext4_journal_ensure_credits_fn(handle, credits,
445 EXT4_MAX_TRANS_DATA, 0, 0);
446 }
447
448 /*
449 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
450 *
451 * Helper function for ext4_setup_new_group_blocks() which set .
452 *
453 * @sb: super block
454 * @handle: journal handle
455 * @flex_gd: flex group data
456 */
set_flexbg_block_bitmap(struct super_block * sb,handle_t * handle,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t first_cluster,ext4_fsblk_t last_cluster)457 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
458 struct ext4_new_flex_group_data *flex_gd,
459 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
460 {
461 struct ext4_sb_info *sbi = EXT4_SB(sb);
462 ext4_group_t count = last_cluster - first_cluster + 1;
463 ext4_group_t count2;
464
465 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
466 last_cluster);
467 for (count2 = count; count > 0;
468 count -= count2, first_cluster += count2) {
469 ext4_fsblk_t start;
470 struct buffer_head *bh;
471 ext4_group_t group;
472 int err;
473
474 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
475 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
476 group -= flex_gd->groups[0].group;
477
478 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
479 if (count2 > count)
480 count2 = count;
481
482 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
483 BUG_ON(flex_gd->count > 1);
484 continue;
485 }
486
487 err = ext4_resize_ensure_credits_batch(handle, 1);
488 if (err < 0)
489 return err;
490
491 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
492 if (unlikely(!bh))
493 return -ENOMEM;
494
495 BUFFER_TRACE(bh, "get_write_access");
496 err = ext4_journal_get_write_access(handle, sb, bh,
497 EXT4_JTR_NONE);
498 if (err) {
499 brelse(bh);
500 return err;
501 }
502 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
503 first_cluster, first_cluster - start, count2);
504 ext4_set_bits(bh->b_data, first_cluster - start, count2);
505
506 err = ext4_handle_dirty_metadata(handle, NULL, bh);
507 brelse(bh);
508 if (unlikely(err))
509 return err;
510 }
511
512 return 0;
513 }
514
515 /*
516 * Set up the block and inode bitmaps, and the inode table for the new groups.
517 * This doesn't need to be part of the main transaction, since we are only
518 * changing blocks outside the actual filesystem. We still do journaling to
519 * ensure the recovery is correct in case of a failure just after resize.
520 * If any part of this fails, we simply abort the resize.
521 *
522 * setup_new_flex_group_blocks handles a flex group as follow:
523 * 1. copy super block and GDT, and initialize group tables if necessary.
524 * In this step, we only set bits in blocks bitmaps for blocks taken by
525 * super block and GDT.
526 * 2. allocate group tables in block bitmaps, that is, set bits in block
527 * bitmap for blocks taken by group tables.
528 */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)529 static int setup_new_flex_group_blocks(struct super_block *sb,
530 struct ext4_new_flex_group_data *flex_gd)
531 {
532 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
533 ext4_fsblk_t start;
534 ext4_fsblk_t block;
535 struct ext4_sb_info *sbi = EXT4_SB(sb);
536 struct ext4_super_block *es = sbi->s_es;
537 struct ext4_new_group_data *group_data = flex_gd->groups;
538 __u16 *bg_flags = flex_gd->bg_flags;
539 handle_t *handle;
540 ext4_group_t group, count;
541 struct buffer_head *bh = NULL;
542 int reserved_gdb, i, j, err = 0, err2;
543 int meta_bg;
544
545 BUG_ON(!flex_gd->count || !group_data ||
546 group_data[0].group != sbi->s_groups_count);
547
548 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
549 meta_bg = ext4_has_feature_meta_bg(sb);
550
551 /* This transaction may be extended/restarted along the way */
552 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
553 if (IS_ERR(handle))
554 return PTR_ERR(handle);
555
556 group = group_data[0].group;
557 for (i = 0; i < flex_gd->count; i++, group++) {
558 unsigned long gdblocks;
559 ext4_grpblk_t overhead;
560
561 gdblocks = ext4_bg_num_gdb(sb, group);
562 start = ext4_group_first_block_no(sb, group);
563
564 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
565 goto handle_itb;
566
567 if (meta_bg == 1)
568 goto handle_itb;
569
570 block = start + ext4_bg_has_super(sb, group);
571 /* Copy all of the GDT blocks into the backup in this group */
572 for (j = 0; j < gdblocks; j++, block++) {
573 struct buffer_head *gdb;
574
575 ext4_debug("update backup group %#04llx\n", block);
576 err = ext4_resize_ensure_credits_batch(handle, 1);
577 if (err < 0)
578 goto out;
579
580 gdb = sb_getblk(sb, block);
581 if (unlikely(!gdb)) {
582 err = -ENOMEM;
583 goto out;
584 }
585
586 BUFFER_TRACE(gdb, "get_write_access");
587 err = ext4_journal_get_write_access(handle, sb, gdb,
588 EXT4_JTR_NONE);
589 if (err) {
590 brelse(gdb);
591 goto out;
592 }
593 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
594 s_group_desc, j)->b_data, gdb->b_size);
595 set_buffer_uptodate(gdb);
596
597 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
598 if (unlikely(err)) {
599 brelse(gdb);
600 goto out;
601 }
602 brelse(gdb);
603 }
604
605 /* Zero out all of the reserved backup group descriptor
606 * table blocks
607 */
608 if (ext4_bg_has_super(sb, group)) {
609 err = sb_issue_zeroout(sb, gdblocks + start + 1,
610 reserved_gdb, GFP_NOFS);
611 if (err)
612 goto out;
613 }
614
615 handle_itb:
616 /* Initialize group tables of the grop @group */
617 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
618 goto handle_bb;
619
620 /* Zero out all of the inode table blocks */
621 block = group_data[i].inode_table;
622 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
623 block, sbi->s_itb_per_group);
624 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
625 GFP_NOFS);
626 if (err)
627 goto out;
628
629 handle_bb:
630 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
631 goto handle_ib;
632
633 /* Initialize block bitmap of the @group */
634 block = group_data[i].block_bitmap;
635 err = ext4_resize_ensure_credits_batch(handle, 1);
636 if (err < 0)
637 goto out;
638
639 bh = bclean(handle, sb, block);
640 if (IS_ERR(bh)) {
641 err = PTR_ERR(bh);
642 goto out;
643 }
644 overhead = ext4_group_overhead_blocks(sb, group);
645 if (overhead != 0) {
646 ext4_debug("mark backup superblock %#04llx (+0)\n",
647 start);
648 ext4_set_bits(bh->b_data, 0,
649 EXT4_NUM_B2C(sbi, overhead));
650 }
651 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
652 sb->s_blocksize * 8, bh->b_data);
653 err = ext4_handle_dirty_metadata(handle, NULL, bh);
654 brelse(bh);
655 if (err)
656 goto out;
657
658 handle_ib:
659 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
660 continue;
661
662 /* Initialize inode bitmap of the @group */
663 block = group_data[i].inode_bitmap;
664 err = ext4_resize_ensure_credits_batch(handle, 1);
665 if (err < 0)
666 goto out;
667 /* Mark unused entries in inode bitmap used */
668 bh = bclean(handle, sb, block);
669 if (IS_ERR(bh)) {
670 err = PTR_ERR(bh);
671 goto out;
672 }
673
674 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
675 sb->s_blocksize * 8, bh->b_data);
676 err = ext4_handle_dirty_metadata(handle, NULL, bh);
677 brelse(bh);
678 if (err)
679 goto out;
680 }
681
682 /* Mark group tables in block bitmap */
683 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
684 count = group_table_count[j];
685 start = (&group_data[0].block_bitmap)[j];
686 block = start;
687 for (i = 1; i < flex_gd->count; i++) {
688 block += group_table_count[j];
689 if (block == (&group_data[i].block_bitmap)[j]) {
690 count += group_table_count[j];
691 continue;
692 }
693 err = set_flexbg_block_bitmap(sb, handle,
694 flex_gd,
695 EXT4_B2C(sbi, start),
696 EXT4_B2C(sbi,
697 start + count
698 - 1));
699 if (err)
700 goto out;
701 count = group_table_count[j];
702 start = (&group_data[i].block_bitmap)[j];
703 block = start;
704 }
705
706 if (count) {
707 err = set_flexbg_block_bitmap(sb, handle,
708 flex_gd,
709 EXT4_B2C(sbi, start),
710 EXT4_B2C(sbi,
711 start + count
712 - 1));
713 if (err)
714 goto out;
715 }
716 }
717
718 out:
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 */
ext4_list_backups(struct super_block * sb,unsigned * three,unsigned * five,unsigned * seven)733 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
734 unsigned *five, unsigned *seven)
735 {
736 unsigned *min = three;
737 int mult = 3;
738 unsigned ret;
739
740 if (!ext4_has_feature_sparse_super(sb)) {
741 ret = *min;
742 *min += 1;
743 return ret;
744 }
745
746 if (*five < *min) {
747 min = five;
748 mult = 5;
749 }
750 if (*seven < *min) {
751 min = seven;
752 mult = 7;
753 }
754
755 ret = *min;
756 *min *= mult;
757
758 return ret;
759 }
760
761 /*
762 * Check that all of the backup GDT blocks are held in the primary GDT block.
763 * It is assumed that they are stored in group order. Returns the number of
764 * groups in current filesystem that have BACKUPS, or -ve error code.
765 */
verify_reserved_gdb(struct super_block * sb,ext4_group_t end,struct buffer_head * primary)766 static int verify_reserved_gdb(struct super_block *sb,
767 ext4_group_t end,
768 struct buffer_head *primary)
769 {
770 const ext4_fsblk_t blk = primary->b_blocknr;
771 unsigned three = 1;
772 unsigned five = 5;
773 unsigned seven = 7;
774 unsigned grp;
775 __le32 *p = (__le32 *)primary->b_data;
776 int gdbackups = 0;
777
778 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
779 if (le32_to_cpu(*p++) !=
780 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
781 ext4_warning(sb, "reserved GDT %llu"
782 " missing grp %d (%llu)",
783 blk, grp,
784 grp *
785 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
786 blk);
787 return -EINVAL;
788 }
789 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
790 return -EFBIG;
791 }
792
793 return gdbackups;
794 }
795
796 /*
797 * Called when we need to bring a reserved group descriptor table block into
798 * use from the resize inode. The primary copy of the new GDT block currently
799 * is an indirect block (under the double indirect block in the resize inode).
800 * The new backup GDT blocks will be stored as leaf blocks in this indirect
801 * block, in group order. Even though we know all the block numbers we need,
802 * we check to ensure that the resize inode has actually reserved these blocks.
803 *
804 * Don't need to update the block bitmaps because the blocks are still in use.
805 *
806 * We get all of the error cases out of the way, so that we are sure to not
807 * fail once we start modifying the data on disk, because JBD has no rollback.
808 */
add_new_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)809 static int add_new_gdb(handle_t *handle, struct inode *inode,
810 ext4_group_t group)
811 {
812 struct super_block *sb = inode->i_sb;
813 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
814 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
815 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
816 struct buffer_head **o_group_desc, **n_group_desc = NULL;
817 struct buffer_head *dind = NULL;
818 struct buffer_head *gdb_bh = NULL;
819 int gdbackups;
820 struct ext4_iloc iloc = { .bh = NULL };
821 __le32 *data;
822 int err;
823
824 if (test_opt(sb, DEBUG))
825 printk(KERN_DEBUG
826 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
827 gdb_num);
828
829 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
830 if (IS_ERR(gdb_bh))
831 return PTR_ERR(gdb_bh);
832
833 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
834 if (gdbackups < 0) {
835 err = gdbackups;
836 goto errout;
837 }
838
839 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
840 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
841 if (IS_ERR(dind)) {
842 err = PTR_ERR(dind);
843 dind = NULL;
844 goto errout;
845 }
846
847 data = (__le32 *)dind->b_data;
848 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
849 ext4_warning(sb, "new group %u GDT block %llu not reserved",
850 group, gdblock);
851 err = -EINVAL;
852 goto errout;
853 }
854
855 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
856 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
857 EXT4_JTR_NONE);
858 if (unlikely(err))
859 goto errout;
860
861 BUFFER_TRACE(gdb_bh, "get_write_access");
862 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
863 if (unlikely(err))
864 goto errout;
865
866 BUFFER_TRACE(dind, "get_write_access");
867 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
868 if (unlikely(err)) {
869 ext4_std_error(sb, err);
870 goto errout;
871 }
872
873 /* ext4_reserve_inode_write() gets a reference on the iloc */
874 err = ext4_reserve_inode_write(handle, inode, &iloc);
875 if (unlikely(err))
876 goto errout;
877
878 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
879 GFP_KERNEL);
880 if (!n_group_desc) {
881 err = -ENOMEM;
882 ext4_warning(sb, "not enough memory for %lu groups",
883 gdb_num + 1);
884 goto errout;
885 }
886
887 /*
888 * Finally, we have all of the possible failures behind us...
889 *
890 * Remove new GDT block from inode double-indirect block and clear out
891 * the new GDT block for use (which also "frees" the backup GDT blocks
892 * from the reserved inode). We don't need to change the bitmaps for
893 * these blocks, because they are marked as in-use from being in the
894 * reserved inode, and will become GDT blocks (primary and backup).
895 */
896 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
897 err = ext4_handle_dirty_metadata(handle, NULL, dind);
898 if (unlikely(err)) {
899 ext4_std_error(sb, err);
900 goto errout;
901 }
902 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
903 (9 - EXT4_SB(sb)->s_cluster_bits);
904 ext4_mark_iloc_dirty(handle, inode, &iloc);
905 memset(gdb_bh->b_data, 0, sb->s_blocksize);
906 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
907 if (unlikely(err)) {
908 ext4_std_error(sb, err);
909 iloc.bh = NULL;
910 goto errout;
911 }
912 brelse(dind);
913
914 rcu_read_lock();
915 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
916 memcpy(n_group_desc, o_group_desc,
917 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
918 rcu_read_unlock();
919 n_group_desc[gdb_num] = gdb_bh;
920 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
921 EXT4_SB(sb)->s_gdb_count++;
922 ext4_kvfree_array_rcu(o_group_desc);
923
924 lock_buffer(EXT4_SB(sb)->s_sbh);
925 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
926 ext4_superblock_csum_set(sb);
927 unlock_buffer(EXT4_SB(sb)->s_sbh);
928 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
929 if (err)
930 ext4_std_error(sb, err);
931 return err;
932 errout:
933 kvfree(n_group_desc);
934 brelse(iloc.bh);
935 brelse(dind);
936 brelse(gdb_bh);
937
938 ext4_debug("leaving with error %d\n", err);
939 return err;
940 }
941
942 /*
943 * add_new_gdb_meta_bg is the sister of add_new_gdb.
944 */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)945 static int add_new_gdb_meta_bg(struct super_block *sb,
946 handle_t *handle, ext4_group_t group) {
947 ext4_fsblk_t gdblock;
948 struct buffer_head *gdb_bh;
949 struct buffer_head **o_group_desc, **n_group_desc;
950 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
951 int err;
952
953 gdblock = ext4_meta_bg_first_block_no(sb, group) +
954 ext4_bg_has_super(sb, group);
955 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
956 if (IS_ERR(gdb_bh))
957 return PTR_ERR(gdb_bh);
958 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
959 GFP_KERNEL);
960 if (!n_group_desc) {
961 brelse(gdb_bh);
962 err = -ENOMEM;
963 ext4_warning(sb, "not enough memory for %lu groups",
964 gdb_num + 1);
965 return err;
966 }
967
968 rcu_read_lock();
969 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
970 memcpy(n_group_desc, o_group_desc,
971 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
972 rcu_read_unlock();
973 n_group_desc[gdb_num] = gdb_bh;
974
975 BUFFER_TRACE(gdb_bh, "get_write_access");
976 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
977 if (err) {
978 kvfree(n_group_desc);
979 brelse(gdb_bh);
980 return err;
981 }
982
983 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
984 EXT4_SB(sb)->s_gdb_count++;
985 ext4_kvfree_array_rcu(o_group_desc);
986 return err;
987 }
988
989 /*
990 * Called when we are adding a new group which has a backup copy of each of
991 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
992 * We need to add these reserved backup GDT blocks to the resize inode, so
993 * that they are kept for future resizing and not allocated to files.
994 *
995 * Each reserved backup GDT block will go into a different indirect block.
996 * The indirect blocks are actually the primary reserved GDT blocks,
997 * so we know in advance what their block numbers are. We only get the
998 * double-indirect block to verify it is pointing to the primary reserved
999 * GDT blocks so we don't overwrite a data block by accident. The reserved
1000 * backup GDT blocks are stored in their reserved primary GDT block.
1001 */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)1002 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1003 ext4_group_t group)
1004 {
1005 struct super_block *sb = inode->i_sb;
1006 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1007 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1008 struct buffer_head **primary;
1009 struct buffer_head *dind;
1010 struct ext4_iloc iloc;
1011 ext4_fsblk_t blk;
1012 __le32 *data, *end;
1013 int gdbackups = 0;
1014 int res, i;
1015 int err;
1016
1017 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1018 if (!primary)
1019 return -ENOMEM;
1020
1021 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1022 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1023 if (IS_ERR(dind)) {
1024 err = PTR_ERR(dind);
1025 dind = NULL;
1026 goto exit_free;
1027 }
1028
1029 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1030 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1031 EXT4_ADDR_PER_BLOCK(sb));
1032 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1033
1034 /* Get each reserved primary GDT block and verify it holds backups */
1035 for (res = 0; res < reserved_gdb; res++, blk++) {
1036 if (le32_to_cpu(*data) != blk) {
1037 ext4_warning(sb, "reserved block %llu"
1038 " not at offset %ld",
1039 blk,
1040 (long)(data - (__le32 *)dind->b_data));
1041 err = -EINVAL;
1042 goto exit_bh;
1043 }
1044 primary[res] = ext4_sb_bread(sb, blk, 0);
1045 if (IS_ERR(primary[res])) {
1046 err = PTR_ERR(primary[res]);
1047 primary[res] = NULL;
1048 goto exit_bh;
1049 }
1050 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1051 if (gdbackups < 0) {
1052 brelse(primary[res]);
1053 err = gdbackups;
1054 goto exit_bh;
1055 }
1056 if (++data >= end)
1057 data = (__le32 *)dind->b_data;
1058 }
1059
1060 for (i = 0; i < reserved_gdb; i++) {
1061 BUFFER_TRACE(primary[i], "get_write_access");
1062 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1063 EXT4_JTR_NONE)))
1064 goto exit_bh;
1065 }
1066
1067 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1068 goto exit_bh;
1069
1070 /*
1071 * Finally we can add each of the reserved backup GDT blocks from
1072 * the new group to its reserved primary GDT block.
1073 */
1074 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1075 for (i = 0; i < reserved_gdb; i++) {
1076 int err2;
1077 data = (__le32 *)primary[i]->b_data;
1078 /* printk("reserving backup %lu[%u] = %lu\n",
1079 primary[i]->b_blocknr, gdbackups,
1080 blk + primary[i]->b_blocknr); */
1081 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1082 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1083 if (!err)
1084 err = err2;
1085 }
1086
1087 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1088 ext4_mark_iloc_dirty(handle, inode, &iloc);
1089
1090 exit_bh:
1091 while (--res >= 0)
1092 brelse(primary[res]);
1093 brelse(dind);
1094
1095 exit_free:
1096 kfree(primary);
1097
1098 return err;
1099 }
1100
1101 /*
1102 * Update the backup copies of the ext4 metadata. These don't need to be part
1103 * of the main resize transaction, because e2fsck will re-write them if there
1104 * is a problem (basically only OOM will cause a problem). However, we
1105 * _should_ update the backups if possible, in case the primary gets trashed
1106 * for some reason and we need to run e2fsck from a backup superblock. The
1107 * important part is that the new block and inode counts are in the backup
1108 * superblocks, and the location of the new group metadata in the GDT backups.
1109 *
1110 * We do not need take the s_resize_lock for this, because these
1111 * blocks are not otherwise touched by the filesystem code when it is
1112 * mounted. We don't need to worry about last changing from
1113 * sbi->s_groups_count, because the worst that can happen is that we
1114 * do not copy the full number of backups at this time. The resize
1115 * which changed s_groups_count will backup again.
1116 */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1117 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1118 int size, int meta_bg)
1119 {
1120 struct ext4_sb_info *sbi = EXT4_SB(sb);
1121 ext4_group_t last;
1122 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1123 unsigned three = 1;
1124 unsigned five = 5;
1125 unsigned seven = 7;
1126 ext4_group_t group = 0;
1127 int rest = sb->s_blocksize - size;
1128 handle_t *handle;
1129 int err = 0, err2;
1130
1131 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1132 if (IS_ERR(handle)) {
1133 group = 1;
1134 err = PTR_ERR(handle);
1135 goto exit_err;
1136 }
1137
1138 if (meta_bg == 0) {
1139 group = ext4_list_backups(sb, &three, &five, &seven);
1140 last = sbi->s_groups_count;
1141 } else {
1142 group = ext4_get_group_number(sb, blk_off) + 1;
1143 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1144 }
1145
1146 while (group < sbi->s_groups_count) {
1147 struct buffer_head *bh;
1148 ext4_fsblk_t backup_block;
1149
1150 /* Out of journal space, and can't get more - abort - so sad */
1151 err = ext4_resize_ensure_credits_batch(handle, 1);
1152 if (err < 0)
1153 break;
1154
1155 if (meta_bg == 0)
1156 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1157 else
1158 backup_block = (ext4_group_first_block_no(sb, group) +
1159 ext4_bg_has_super(sb, group));
1160
1161 bh = sb_getblk(sb, backup_block);
1162 if (unlikely(!bh)) {
1163 err = -ENOMEM;
1164 break;
1165 }
1166 ext4_debug("update metadata backup %llu(+%llu)\n",
1167 backup_block, backup_block -
1168 ext4_group_first_block_no(sb, group));
1169 BUFFER_TRACE(bh, "get_write_access");
1170 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1171 EXT4_JTR_NONE))) {
1172 brelse(bh);
1173 break;
1174 }
1175 lock_buffer(bh);
1176 memcpy(bh->b_data, data, size);
1177 if (rest)
1178 memset(bh->b_data + size, 0, rest);
1179 set_buffer_uptodate(bh);
1180 unlock_buffer(bh);
1181 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1182 if (unlikely(err))
1183 ext4_std_error(sb, err);
1184 brelse(bh);
1185
1186 if (meta_bg == 0)
1187 group = ext4_list_backups(sb, &three, &five, &seven);
1188 else if (group == last)
1189 break;
1190 else
1191 group = last;
1192 }
1193 if ((err2 = ext4_journal_stop(handle)) && !err)
1194 err = err2;
1195
1196 /*
1197 * Ugh! Need to have e2fsck write the backup copies. It is too
1198 * late to revert the resize, we shouldn't fail just because of
1199 * the backup copies (they are only needed in case of corruption).
1200 *
1201 * However, if we got here we have a journal problem too, so we
1202 * can't really start a transaction to mark the superblock.
1203 * Chicken out and just set the flag on the hope it will be written
1204 * to disk, and if not - we will simply wait until next fsck.
1205 */
1206 exit_err:
1207 if (err) {
1208 ext4_warning(sb, "can't update backup for group %u (err %d), "
1209 "forcing fsck on next reboot", group, err);
1210 sbi->s_mount_state &= ~EXT4_VALID_FS;
1211 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1212 mark_buffer_dirty(sbi->s_sbh);
1213 }
1214 }
1215
1216 /*
1217 * ext4_add_new_descs() adds @count group descriptor of groups
1218 * starting at @group
1219 *
1220 * @handle: journal handle
1221 * @sb: super block
1222 * @group: the group no. of the first group desc to be added
1223 * @resize_inode: the resize inode
1224 * @count: number of group descriptors to be added
1225 */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1226 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1227 ext4_group_t group, struct inode *resize_inode,
1228 ext4_group_t count)
1229 {
1230 struct ext4_sb_info *sbi = EXT4_SB(sb);
1231 struct ext4_super_block *es = sbi->s_es;
1232 struct buffer_head *gdb_bh;
1233 int i, gdb_off, gdb_num, err = 0;
1234 int meta_bg;
1235
1236 meta_bg = ext4_has_feature_meta_bg(sb);
1237 for (i = 0; i < count; i++, group++) {
1238 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1239 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1240
1241 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1242 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1243
1244 /*
1245 * We will only either add reserved group blocks to a backup group
1246 * or remove reserved blocks for the first group in a new group block.
1247 * Doing both would be mean more complex code, and sane people don't
1248 * use non-sparse filesystems anymore. This is already checked above.
1249 */
1250 if (gdb_off) {
1251 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1252 gdb_num);
1253 BUFFER_TRACE(gdb_bh, "get_write_access");
1254 err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1255 EXT4_JTR_NONE);
1256
1257 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1258 err = reserve_backup_gdb(handle, resize_inode, group);
1259 } else if (meta_bg != 0) {
1260 err = add_new_gdb_meta_bg(sb, handle, group);
1261 } else {
1262 err = add_new_gdb(handle, resize_inode, group);
1263 }
1264 if (err)
1265 break;
1266 }
1267 return err;
1268 }
1269
ext4_get_bitmap(struct super_block * sb,__u64 block)1270 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1271 {
1272 struct buffer_head *bh = sb_getblk(sb, block);
1273 if (unlikely(!bh))
1274 return NULL;
1275 if (!bh_uptodate_or_lock(bh)) {
1276 if (ext4_read_bh(bh, 0, NULL) < 0) {
1277 brelse(bh);
1278 return NULL;
1279 }
1280 }
1281
1282 return bh;
1283 }
1284
ext4_set_bitmap_checksums(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1285 static int ext4_set_bitmap_checksums(struct super_block *sb,
1286 ext4_group_t group,
1287 struct ext4_group_desc *gdp,
1288 struct ext4_new_group_data *group_data)
1289 {
1290 struct buffer_head *bh;
1291
1292 if (!ext4_has_metadata_csum(sb))
1293 return 0;
1294
1295 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1296 if (!bh)
1297 return -EIO;
1298 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1299 EXT4_INODES_PER_GROUP(sb) / 8);
1300 brelse(bh);
1301
1302 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1303 if (!bh)
1304 return -EIO;
1305 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1306 brelse(bh);
1307
1308 return 0;
1309 }
1310
1311 /*
1312 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1313 */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1314 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1315 struct ext4_new_flex_group_data *flex_gd)
1316 {
1317 struct ext4_new_group_data *group_data = flex_gd->groups;
1318 struct ext4_group_desc *gdp;
1319 struct ext4_sb_info *sbi = EXT4_SB(sb);
1320 struct buffer_head *gdb_bh;
1321 ext4_group_t group;
1322 __u16 *bg_flags = flex_gd->bg_flags;
1323 int i, gdb_off, gdb_num, err = 0;
1324
1325
1326 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1327 group = group_data->group;
1328
1329 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1330 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1331
1332 /*
1333 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1334 */
1335 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1336 /* Update group descriptor block for new group */
1337 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1338 gdb_off * EXT4_DESC_SIZE(sb));
1339
1340 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1341 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1342 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1343 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1344 if (err) {
1345 ext4_std_error(sb, err);
1346 break;
1347 }
1348
1349 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1350 ext4_free_group_clusters_set(sb, gdp,
1351 group_data->free_clusters_count);
1352 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1353 if (ext4_has_group_desc_csum(sb))
1354 ext4_itable_unused_set(sb, gdp,
1355 EXT4_INODES_PER_GROUP(sb));
1356 gdp->bg_flags = cpu_to_le16(*bg_flags);
1357 ext4_group_desc_csum_set(sb, group, gdp);
1358
1359 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1360 if (unlikely(err)) {
1361 ext4_std_error(sb, err);
1362 break;
1363 }
1364
1365 /*
1366 * We can allocate memory for mb_alloc based on the new group
1367 * descriptor
1368 */
1369 err = ext4_mb_add_groupinfo(sb, group, gdp);
1370 if (err)
1371 break;
1372 }
1373 return err;
1374 }
1375
1376 /*
1377 * ext4_update_super() updates the super block so that the newly added
1378 * groups can be seen by the filesystem.
1379 *
1380 * @sb: super block
1381 * @flex_gd: new added groups
1382 */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1383 static void ext4_update_super(struct super_block *sb,
1384 struct ext4_new_flex_group_data *flex_gd)
1385 {
1386 ext4_fsblk_t blocks_count = 0;
1387 ext4_fsblk_t free_blocks = 0;
1388 ext4_fsblk_t reserved_blocks = 0;
1389 struct ext4_new_group_data *group_data = flex_gd->groups;
1390 struct ext4_sb_info *sbi = EXT4_SB(sb);
1391 struct ext4_super_block *es = sbi->s_es;
1392 int i;
1393
1394 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1395 /*
1396 * Make the new blocks and inodes valid next. We do this before
1397 * increasing the group count so that once the group is enabled,
1398 * all of its blocks and inodes are already valid.
1399 *
1400 * We always allocate group-by-group, then block-by-block or
1401 * inode-by-inode within a group, so enabling these
1402 * blocks/inodes before the group is live won't actually let us
1403 * allocate the new space yet.
1404 */
1405 for (i = 0; i < flex_gd->count; i++) {
1406 blocks_count += group_data[i].blocks_count;
1407 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1408 }
1409
1410 reserved_blocks = ext4_r_blocks_count(es) * 100;
1411 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1412 reserved_blocks *= blocks_count;
1413 do_div(reserved_blocks, 100);
1414
1415 lock_buffer(sbi->s_sbh);
1416 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1417 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1418 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1419 flex_gd->count);
1420 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1421 flex_gd->count);
1422
1423 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1424 /*
1425 * We need to protect s_groups_count against other CPUs seeing
1426 * inconsistent state in the superblock.
1427 *
1428 * The precise rules we use are:
1429 *
1430 * * Writers must perform a smp_wmb() after updating all
1431 * dependent data and before modifying the groups count
1432 *
1433 * * Readers must perform an smp_rmb() after reading the groups
1434 * count and before reading any dependent data.
1435 *
1436 * NB. These rules can be relaxed when checking the group count
1437 * while freeing data, as we can only allocate from a block
1438 * group after serialising against the group count, and we can
1439 * only then free after serialising in turn against that
1440 * allocation.
1441 */
1442 smp_wmb();
1443
1444 /* Update the global fs size fields */
1445 sbi->s_groups_count += flex_gd->count;
1446 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1447 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1448
1449 /* Update the reserved block counts only once the new group is
1450 * active. */
1451 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1452 reserved_blocks);
1453
1454 /* Update the free space counts */
1455 percpu_counter_add(&sbi->s_freeclusters_counter,
1456 EXT4_NUM_B2C(sbi, free_blocks));
1457 percpu_counter_add(&sbi->s_freeinodes_counter,
1458 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1459
1460 ext4_debug("free blocks count %llu",
1461 percpu_counter_read(&sbi->s_freeclusters_counter));
1462 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1463 ext4_group_t flex_group;
1464 struct flex_groups *fg;
1465
1466 flex_group = ext4_flex_group(sbi, group_data[0].group);
1467 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1468 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1469 &fg->free_clusters);
1470 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1471 &fg->free_inodes);
1472 }
1473
1474 /*
1475 * Update the fs overhead information
1476 */
1477 ext4_calculate_overhead(sb);
1478 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1479
1480 ext4_superblock_csum_set(sb);
1481 unlock_buffer(sbi->s_sbh);
1482 if (test_opt(sb, DEBUG))
1483 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1484 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1485 blocks_count, free_blocks, reserved_blocks);
1486 }
1487
1488 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1489 * _before_ we start modifying the filesystem, because we cannot abort the
1490 * transaction and not have it write the data to disk.
1491 */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1492 static int ext4_flex_group_add(struct super_block *sb,
1493 struct inode *resize_inode,
1494 struct ext4_new_flex_group_data *flex_gd)
1495 {
1496 struct ext4_sb_info *sbi = EXT4_SB(sb);
1497 struct ext4_super_block *es = sbi->s_es;
1498 ext4_fsblk_t o_blocks_count;
1499 ext4_grpblk_t last;
1500 ext4_group_t group;
1501 handle_t *handle;
1502 unsigned reserved_gdb;
1503 int err = 0, err2 = 0, credit;
1504
1505 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1506
1507 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1508 o_blocks_count = ext4_blocks_count(es);
1509 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1510 BUG_ON(last);
1511
1512 err = setup_new_flex_group_blocks(sb, flex_gd);
1513 if (err)
1514 goto exit;
1515 /*
1516 * We will always be modifying at least the superblock and GDT
1517 * blocks. If we are adding a group past the last current GDT block,
1518 * we will also modify the inode and the dindirect block. If we
1519 * are adding a group with superblock/GDT backups we will also
1520 * modify each of the reserved GDT dindirect blocks.
1521 */
1522 credit = 3; /* sb, resize inode, resize inode dindirect */
1523 /* GDT blocks */
1524 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1525 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1526 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1527 if (IS_ERR(handle)) {
1528 err = PTR_ERR(handle);
1529 goto exit;
1530 }
1531
1532 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1533 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1534 EXT4_JTR_NONE);
1535 if (err)
1536 goto exit_journal;
1537
1538 group = flex_gd->groups[0].group;
1539 BUG_ON(group != sbi->s_groups_count);
1540 err = ext4_add_new_descs(handle, sb, group,
1541 resize_inode, flex_gd->count);
1542 if (err)
1543 goto exit_journal;
1544
1545 err = ext4_setup_new_descs(handle, sb, flex_gd);
1546 if (err)
1547 goto exit_journal;
1548
1549 ext4_update_super(sb, flex_gd);
1550
1551 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1552
1553 exit_journal:
1554 err2 = ext4_journal_stop(handle);
1555 if (!err)
1556 err = err2;
1557
1558 if (!err) {
1559 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1560 int gdb_num_end = ((group + flex_gd->count - 1) /
1561 EXT4_DESC_PER_BLOCK(sb));
1562 int meta_bg = ext4_has_feature_meta_bg(sb);
1563 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1564 ext4_group_first_block_no(sb, 0);
1565 sector_t old_gdb = 0;
1566
1567 update_backups(sb, ext4_group_first_block_no(sb, 0),
1568 (char *)es, sizeof(struct ext4_super_block), 0);
1569 for (; gdb_num <= gdb_num_end; gdb_num++) {
1570 struct buffer_head *gdb_bh;
1571
1572 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1573 gdb_num);
1574 if (old_gdb == gdb_bh->b_blocknr)
1575 continue;
1576 update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1577 gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1578 old_gdb = gdb_bh->b_blocknr;
1579 }
1580 }
1581 exit:
1582 return err;
1583 }
1584
ext4_setup_next_flex_gd(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t n_blocks_count)1585 static int ext4_setup_next_flex_gd(struct super_block *sb,
1586 struct ext4_new_flex_group_data *flex_gd,
1587 ext4_fsblk_t n_blocks_count)
1588 {
1589 struct ext4_sb_info *sbi = EXT4_SB(sb);
1590 struct ext4_super_block *es = sbi->s_es;
1591 struct ext4_new_group_data *group_data = flex_gd->groups;
1592 ext4_fsblk_t o_blocks_count;
1593 ext4_group_t n_group;
1594 ext4_group_t group;
1595 ext4_group_t last_group;
1596 ext4_grpblk_t last;
1597 ext4_grpblk_t clusters_per_group;
1598 unsigned long i;
1599
1600 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1601
1602 o_blocks_count = ext4_blocks_count(es);
1603
1604 if (o_blocks_count == n_blocks_count)
1605 return 0;
1606
1607 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1608 BUG_ON(last);
1609 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1610
1611 last_group = group | (flex_gd->resize_bg - 1);
1612 if (last_group > n_group)
1613 last_group = n_group;
1614
1615 flex_gd->count = last_group - group + 1;
1616
1617 for (i = 0; i < flex_gd->count; i++) {
1618 int overhead;
1619
1620 group_data[i].group = group + i;
1621 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1622 overhead = ext4_group_overhead_blocks(sb, group + i);
1623 group_data[i].mdata_blocks = overhead;
1624 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1625 if (ext4_has_group_desc_csum(sb)) {
1626 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1627 EXT4_BG_INODE_UNINIT;
1628 if (!test_opt(sb, INIT_INODE_TABLE))
1629 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1630 } else
1631 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1632 }
1633
1634 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1635 /* We need to initialize block bitmap of last group. */
1636 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1637
1638 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1639 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1640 group_data[i - 1].free_clusters_count -= clusters_per_group -
1641 last - 1;
1642 }
1643
1644 return 1;
1645 }
1646
1647 /* Add group descriptor data to an existing or new group descriptor block.
1648 * Ensure we handle all possible error conditions _before_ we start modifying
1649 * the filesystem, because we cannot abort the transaction and not have it
1650 * write the data to disk.
1651 *
1652 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1653 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1654 *
1655 * We only need to hold the superblock lock while we are actually adding
1656 * in the new group's counts to the superblock. Prior to that we have
1657 * not really "added" the group at all. We re-check that we are still
1658 * adding in the last group in case things have changed since verifying.
1659 */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1660 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1661 {
1662 struct ext4_new_flex_group_data flex_gd;
1663 struct ext4_sb_info *sbi = EXT4_SB(sb);
1664 struct ext4_super_block *es = sbi->s_es;
1665 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1666 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1667 struct inode *inode = NULL;
1668 int gdb_off;
1669 int err;
1670 __u16 bg_flags = 0;
1671
1672 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1673
1674 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1675 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1676 return -EPERM;
1677 }
1678
1679 if (ext4_blocks_count(es) + input->blocks_count <
1680 ext4_blocks_count(es)) {
1681 ext4_warning(sb, "blocks_count overflow");
1682 return -EINVAL;
1683 }
1684
1685 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1686 le32_to_cpu(es->s_inodes_count)) {
1687 ext4_warning(sb, "inodes_count overflow");
1688 return -EINVAL;
1689 }
1690
1691 if (reserved_gdb || gdb_off == 0) {
1692 if (!ext4_has_feature_resize_inode(sb) ||
1693 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1694 ext4_warning(sb,
1695 "No reserved GDT blocks, can't resize");
1696 return -EPERM;
1697 }
1698 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1699 if (IS_ERR(inode)) {
1700 ext4_warning(sb, "Error opening resize inode");
1701 return PTR_ERR(inode);
1702 }
1703 }
1704
1705
1706 err = verify_group_input(sb, input);
1707 if (err)
1708 goto out;
1709
1710 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1711 if (err)
1712 goto out;
1713
1714 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1715 if (err)
1716 goto out;
1717
1718 flex_gd.count = 1;
1719 flex_gd.groups = input;
1720 flex_gd.bg_flags = &bg_flags;
1721 err = ext4_flex_group_add(sb, inode, &flex_gd);
1722 out:
1723 iput(inode);
1724 return err;
1725 } /* ext4_group_add */
1726
1727 /*
1728 * extend a group without checking assuming that checking has been done.
1729 */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1730 static int ext4_group_extend_no_check(struct super_block *sb,
1731 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1732 {
1733 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1734 handle_t *handle;
1735 int err = 0, err2;
1736
1737 /* We will update the superblock, one block bitmap, and
1738 * one group descriptor via ext4_group_add_blocks().
1739 */
1740 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1741 if (IS_ERR(handle)) {
1742 err = PTR_ERR(handle);
1743 ext4_warning(sb, "error %d on journal start", err);
1744 return err;
1745 }
1746
1747 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1748 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1749 EXT4_JTR_NONE);
1750 if (err) {
1751 ext4_warning(sb, "error %d on journal write access", err);
1752 goto errout;
1753 }
1754
1755 lock_buffer(EXT4_SB(sb)->s_sbh);
1756 ext4_blocks_count_set(es, o_blocks_count + add);
1757 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1758 ext4_superblock_csum_set(sb);
1759 unlock_buffer(EXT4_SB(sb)->s_sbh);
1760 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1761 o_blocks_count + add);
1762 /* We add the blocks to the bitmap and set the group need init bit */
1763 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1764 if (err)
1765 goto errout;
1766 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1767 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1768 o_blocks_count + add);
1769 errout:
1770 err2 = ext4_journal_stop(handle);
1771 if (err2 && !err)
1772 err = err2;
1773
1774 if (!err) {
1775 if (test_opt(sb, DEBUG))
1776 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1777 "blocks\n", ext4_blocks_count(es));
1778 update_backups(sb, ext4_group_first_block_no(sb, 0),
1779 (char *)es, sizeof(struct ext4_super_block), 0);
1780 }
1781 return err;
1782 }
1783
1784 /*
1785 * Extend the filesystem to the new number of blocks specified. This entry
1786 * point is only used to extend the current filesystem to the end of the last
1787 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1788 * for emergencies (because it has no dependencies on reserved blocks).
1789 *
1790 * If we _really_ wanted, we could use default values to call ext4_group_add()
1791 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1792 * GDT blocks are reserved to grow to the desired size.
1793 */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1794 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1795 ext4_fsblk_t n_blocks_count)
1796 {
1797 ext4_fsblk_t o_blocks_count;
1798 ext4_grpblk_t last;
1799 ext4_grpblk_t add;
1800 struct buffer_head *bh;
1801 int err;
1802 ext4_group_t group;
1803
1804 o_blocks_count = ext4_blocks_count(es);
1805
1806 if (test_opt(sb, DEBUG))
1807 ext4_msg(sb, KERN_DEBUG,
1808 "extending last group from %llu to %llu blocks",
1809 o_blocks_count, n_blocks_count);
1810
1811 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1812 return 0;
1813
1814 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1815 ext4_msg(sb, KERN_ERR,
1816 "filesystem too large to resize to %llu blocks safely",
1817 n_blocks_count);
1818 return -EINVAL;
1819 }
1820
1821 if (n_blocks_count < o_blocks_count) {
1822 ext4_warning(sb, "can't shrink FS - resize aborted");
1823 return -EINVAL;
1824 }
1825
1826 /* Handle the remaining blocks in the last group only. */
1827 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1828
1829 if (last == 0) {
1830 ext4_warning(sb, "need to use ext2online to resize further");
1831 return -EPERM;
1832 }
1833
1834 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1835
1836 if (o_blocks_count + add < o_blocks_count) {
1837 ext4_warning(sb, "blocks_count overflow");
1838 return -EINVAL;
1839 }
1840
1841 if (o_blocks_count + add > n_blocks_count)
1842 add = n_blocks_count - o_blocks_count;
1843
1844 if (o_blocks_count + add < n_blocks_count)
1845 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1846 o_blocks_count + add, add);
1847
1848 /* See if the device is actually as big as what was requested */
1849 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1850 if (IS_ERR(bh)) {
1851 ext4_warning(sb, "can't read last block, resize aborted");
1852 return -ENOSPC;
1853 }
1854 brelse(bh);
1855
1856 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1857 return err;
1858 } /* ext4_group_extend */
1859
1860
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1861 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1862 {
1863 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1864 }
1865
1866 /*
1867 * Release the resize inode and drop the resize_inode feature if there
1868 * are no more reserved gdt blocks, and then convert the file system
1869 * to enable meta_bg
1870 */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1871 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1872 {
1873 handle_t *handle;
1874 struct ext4_sb_info *sbi = EXT4_SB(sb);
1875 struct ext4_super_block *es = sbi->s_es;
1876 struct ext4_inode_info *ei = EXT4_I(inode);
1877 ext4_fsblk_t nr;
1878 int i, ret, err = 0;
1879 int credits = 1;
1880
1881 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1882 if (inode) {
1883 if (es->s_reserved_gdt_blocks) {
1884 ext4_error(sb, "Unexpected non-zero "
1885 "s_reserved_gdt_blocks");
1886 return -EPERM;
1887 }
1888
1889 /* Do a quick sanity check of the resize inode */
1890 if (inode->i_blocks != 1 << (inode->i_blkbits -
1891 (9 - sbi->s_cluster_bits)))
1892 goto invalid_resize_inode;
1893 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1894 if (i == EXT4_DIND_BLOCK) {
1895 if (ei->i_data[i])
1896 continue;
1897 else
1898 goto invalid_resize_inode;
1899 }
1900 if (ei->i_data[i])
1901 goto invalid_resize_inode;
1902 }
1903 credits += 3; /* block bitmap, bg descriptor, resize inode */
1904 }
1905
1906 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1907 if (IS_ERR(handle))
1908 return PTR_ERR(handle);
1909
1910 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1911 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1912 EXT4_JTR_NONE);
1913 if (err)
1914 goto errout;
1915
1916 lock_buffer(sbi->s_sbh);
1917 ext4_clear_feature_resize_inode(sb);
1918 ext4_set_feature_meta_bg(sb);
1919 sbi->s_es->s_first_meta_bg =
1920 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1921 ext4_superblock_csum_set(sb);
1922 unlock_buffer(sbi->s_sbh);
1923
1924 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1925 if (err) {
1926 ext4_std_error(sb, err);
1927 goto errout;
1928 }
1929
1930 if (inode) {
1931 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1932 ext4_free_blocks(handle, inode, NULL, nr, 1,
1933 EXT4_FREE_BLOCKS_METADATA |
1934 EXT4_FREE_BLOCKS_FORGET);
1935 ei->i_data[EXT4_DIND_BLOCK] = 0;
1936 inode->i_blocks = 0;
1937
1938 err = ext4_mark_inode_dirty(handle, inode);
1939 if (err)
1940 ext4_std_error(sb, err);
1941 }
1942
1943 errout:
1944 ret = ext4_journal_stop(handle);
1945 return err ? err : ret;
1946
1947 invalid_resize_inode:
1948 ext4_error(sb, "corrupted/inconsistent resize inode");
1949 return -EINVAL;
1950 }
1951
1952 /*
1953 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1954 *
1955 * @sb: super block of the fs to be resized
1956 * @n_blocks_count: the number of blocks resides in the resized fs
1957 */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)1958 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1959 {
1960 struct ext4_new_flex_group_data *flex_gd = NULL;
1961 struct ext4_sb_info *sbi = EXT4_SB(sb);
1962 struct ext4_super_block *es = sbi->s_es;
1963 struct buffer_head *bh;
1964 struct inode *resize_inode = NULL;
1965 ext4_grpblk_t add, offset;
1966 unsigned long n_desc_blocks;
1967 unsigned long o_desc_blocks;
1968 ext4_group_t o_group;
1969 ext4_group_t n_group;
1970 ext4_fsblk_t o_blocks_count;
1971 ext4_fsblk_t n_blocks_count_retry = 0;
1972 unsigned long last_update_time = 0;
1973 int err = 0;
1974 int meta_bg;
1975 unsigned int flexbg_size = ext4_flex_bg_size(sbi);
1976
1977 /* See if the device is actually as big as what was requested */
1978 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1979 if (IS_ERR(bh)) {
1980 ext4_warning(sb, "can't read last block, resize aborted");
1981 return -ENOSPC;
1982 }
1983 brelse(bh);
1984
1985 /*
1986 * For bigalloc, trim the requested size to the nearest cluster
1987 * boundary to avoid creating an unusable filesystem. We do this
1988 * silently, instead of returning an error, to avoid breaking
1989 * callers that blindly resize the filesystem to the full size of
1990 * the underlying block device.
1991 */
1992 if (ext4_has_feature_bigalloc(sb))
1993 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1994
1995 retry:
1996 o_blocks_count = ext4_blocks_count(es);
1997
1998 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1999 "to %llu blocks", o_blocks_count, n_blocks_count);
2000
2001 if (n_blocks_count < o_blocks_count) {
2002 /* On-line shrinking not supported */
2003 ext4_warning(sb, "can't shrink FS - resize aborted");
2004 return -EINVAL;
2005 }
2006
2007 if (n_blocks_count == o_blocks_count)
2008 /* Nothing need to do */
2009 return 0;
2010
2011 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2012 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2013 ext4_warning(sb, "resize would cause inodes_count overflow");
2014 return -EINVAL;
2015 }
2016 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2017
2018 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2019 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2020
2021 meta_bg = ext4_has_feature_meta_bg(sb);
2022
2023 if (ext4_has_feature_resize_inode(sb)) {
2024 if (meta_bg) {
2025 ext4_error(sb, "resize_inode and meta_bg enabled "
2026 "simultaneously");
2027 return -EINVAL;
2028 }
2029 if (n_desc_blocks > o_desc_blocks +
2030 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2031 n_blocks_count_retry = n_blocks_count;
2032 n_desc_blocks = o_desc_blocks +
2033 le16_to_cpu(es->s_reserved_gdt_blocks);
2034 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2035 n_blocks_count = (ext4_fsblk_t)n_group *
2036 EXT4_BLOCKS_PER_GROUP(sb) +
2037 le32_to_cpu(es->s_first_data_block);
2038 n_group--; /* set to last group number */
2039 }
2040
2041 if (!resize_inode)
2042 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2043 EXT4_IGET_SPECIAL);
2044 if (IS_ERR(resize_inode)) {
2045 ext4_warning(sb, "Error opening resize inode");
2046 return PTR_ERR(resize_inode);
2047 }
2048 }
2049
2050 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2051 err = ext4_convert_meta_bg(sb, resize_inode);
2052 if (err)
2053 goto out;
2054 if (resize_inode) {
2055 iput(resize_inode);
2056 resize_inode = NULL;
2057 }
2058 if (n_blocks_count_retry) {
2059 n_blocks_count = n_blocks_count_retry;
2060 n_blocks_count_retry = 0;
2061 goto retry;
2062 }
2063 }
2064
2065 /*
2066 * Make sure the last group has enough space so that it's
2067 * guaranteed to have enough space for all metadata blocks
2068 * that it might need to hold. (We might not need to store
2069 * the inode table blocks in the last block group, but there
2070 * will be cases where this might be needed.)
2071 */
2072 if ((ext4_group_first_block_no(sb, n_group) +
2073 ext4_group_overhead_blocks(sb, n_group) + 2 +
2074 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2075 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2076 n_group--;
2077 n_blocks_count_retry = 0;
2078 if (resize_inode) {
2079 iput(resize_inode);
2080 resize_inode = NULL;
2081 }
2082 goto retry;
2083 }
2084
2085 /* extend the last group */
2086 if (n_group == o_group)
2087 add = n_blocks_count - o_blocks_count;
2088 else
2089 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2090 if (add > 0) {
2091 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2092 if (err)
2093 goto out;
2094 }
2095
2096 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2097 goto out;
2098
2099 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2100 if (err)
2101 goto out;
2102
2103 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2104 if (err)
2105 goto out;
2106
2107 flex_gd = alloc_flex_gd(flexbg_size);
2108 if (flex_gd == NULL) {
2109 err = -ENOMEM;
2110 goto out;
2111 }
2112
2113 /* Add flex groups. Note that a regular group is a
2114 * flex group with 1 group.
2115 */
2116 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2117 if (jiffies - last_update_time > HZ * 10) {
2118 if (last_update_time)
2119 ext4_msg(sb, KERN_INFO,
2120 "resized to %llu blocks",
2121 ext4_blocks_count(es));
2122 last_update_time = jiffies;
2123 }
2124 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2125 break;
2126 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2127 if (unlikely(err))
2128 break;
2129 }
2130
2131 if (!err && n_blocks_count_retry) {
2132 n_blocks_count = n_blocks_count_retry;
2133 n_blocks_count_retry = 0;
2134 free_flex_gd(flex_gd);
2135 flex_gd = NULL;
2136 if (resize_inode) {
2137 iput(resize_inode);
2138 resize_inode = NULL;
2139 }
2140 goto retry;
2141 }
2142
2143 out:
2144 if (flex_gd)
2145 free_flex_gd(flex_gd);
2146 if (resize_inode != NULL)
2147 iput(resize_inode);
2148 if (err)
2149 ext4_warning(sb, "error (%d) occurred during "
2150 "file system resize", err);
2151 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2152 ext4_blocks_count(es));
2153 return err;
2154 }
2155