xref: /linux/fs/ext4/resize.c (revision 9a6b55ac)
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 int ext4_resize_begin(struct super_block *sb)
21 {
22 	struct ext4_sb_info *sbi = EXT4_SB(sb);
23 	int ret = 0;
24 
25 	if (!capable(CAP_SYS_RESOURCE))
26 		return -EPERM;
27 
28 	/*
29 	 * If we are not using the primary superblock/GDT copy don't resize,
30          * because the user tools have no way of handling this.  Probably a
31          * bad time to do it anyways.
32          */
33 	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
34 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
35 		ext4_warning(sb, "won't resize using backup superblock at %llu",
36 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
37 		return -EPERM;
38 	}
39 
40 	/*
41 	 * We are not allowed to do online-resizing on a filesystem mounted
42 	 * with error, because it can destroy the filesystem easily.
43 	 */
44 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
45 		ext4_warning(sb, "There are errors in the filesystem, "
46 			     "so online resizing is not allowed");
47 		return -EPERM;
48 	}
49 
50 	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
51 				  &EXT4_SB(sb)->s_ext4_flags))
52 		ret = -EBUSY;
53 
54 	return ret;
55 }
56 
57 void ext4_resize_end(struct super_block *sb)
58 {
59 	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
60 	smp_mb__after_atomic();
61 }
62 
63 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
64 					     ext4_group_t group) {
65 	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
66 	       EXT4_DESC_PER_BLOCK_BITS(sb);
67 }
68 
69 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
70 					     ext4_group_t group) {
71 	group = ext4_meta_bg_first_group(sb, group);
72 	return ext4_group_first_block_no(sb, group);
73 }
74 
75 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
76 						ext4_group_t group) {
77 	ext4_grpblk_t overhead;
78 	overhead = ext4_bg_num_gdb(sb, group);
79 	if (ext4_bg_has_super(sb, group))
80 		overhead += 1 +
81 			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
82 	return overhead;
83 }
84 
85 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
86 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
87 
88 static int verify_group_input(struct super_block *sb,
89 			      struct ext4_new_group_data *input)
90 {
91 	struct ext4_sb_info *sbi = EXT4_SB(sb);
92 	struct ext4_super_block *es = sbi->s_es;
93 	ext4_fsblk_t start = ext4_blocks_count(es);
94 	ext4_fsblk_t end = start + input->blocks_count;
95 	ext4_group_t group = input->group;
96 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
97 	unsigned overhead;
98 	ext4_fsblk_t metaend;
99 	struct buffer_head *bh = NULL;
100 	ext4_grpblk_t free_blocks_count, offset;
101 	int err = -EINVAL;
102 
103 	if (group != sbi->s_groups_count) {
104 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
105 			     input->group, sbi->s_groups_count);
106 		return -EINVAL;
107 	}
108 
109 	overhead = ext4_group_overhead_blocks(sb, group);
110 	metaend = start + overhead;
111 	input->free_clusters_count = free_blocks_count =
112 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
113 
114 	if (test_opt(sb, DEBUG))
115 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
116 		       "(%d free, %u reserved)\n",
117 		       ext4_bg_has_super(sb, input->group) ? "normal" :
118 		       "no-super", input->group, input->blocks_count,
119 		       free_blocks_count, input->reserved_blocks);
120 
121 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
122 	if (offset != 0)
123 			ext4_warning(sb, "Last group not full");
124 	else if (input->reserved_blocks > input->blocks_count / 5)
125 		ext4_warning(sb, "Reserved blocks too high (%u)",
126 			     input->reserved_blocks);
127 	else if (free_blocks_count < 0)
128 		ext4_warning(sb, "Bad blocks count %u",
129 			     input->blocks_count);
130 	else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
131 		err = PTR_ERR(bh);
132 		bh = NULL;
133 		ext4_warning(sb, "Cannot read last block (%llu)",
134 			     end - 1);
135 	} else if (outside(input->block_bitmap, start, end))
136 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
137 			     (unsigned long long)input->block_bitmap);
138 	else if (outside(input->inode_bitmap, start, end))
139 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
140 			     (unsigned long long)input->inode_bitmap);
141 	else if (outside(input->inode_table, start, end) ||
142 		 outside(itend - 1, start, end))
143 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
144 			     (unsigned long long)input->inode_table, itend - 1);
145 	else if (input->inode_bitmap == input->block_bitmap)
146 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
147 			     (unsigned long long)input->block_bitmap);
148 	else if (inside(input->block_bitmap, input->inode_table, itend))
149 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
150 			     "(%llu-%llu)",
151 			     (unsigned long long)input->block_bitmap,
152 			     (unsigned long long)input->inode_table, itend - 1);
153 	else if (inside(input->inode_bitmap, input->inode_table, itend))
154 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
155 			     "(%llu-%llu)",
156 			     (unsigned long long)input->inode_bitmap,
157 			     (unsigned long long)input->inode_table, itend - 1);
158 	else if (inside(input->block_bitmap, start, metaend))
159 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
160 			     (unsigned long long)input->block_bitmap,
161 			     start, metaend - 1);
162 	else if (inside(input->inode_bitmap, start, metaend))
163 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
164 			     (unsigned long long)input->inode_bitmap,
165 			     start, metaend - 1);
166 	else if (inside(input->inode_table, start, metaend) ||
167 		 inside(itend - 1, start, metaend))
168 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
169 			     "(%llu-%llu)",
170 			     (unsigned long long)input->inode_table,
171 			     itend - 1, start, metaend - 1);
172 	else
173 		err = 0;
174 	brelse(bh);
175 
176 	return err;
177 }
178 
179 /*
180  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
181  * group each time.
182  */
183 struct ext4_new_flex_group_data {
184 	struct ext4_new_group_data *groups;	/* new_group_data for groups
185 						   in the flex group */
186 	__u16 *bg_flags;			/* block group flags of groups
187 						   in @groups */
188 	ext4_group_t count;			/* number of groups in @groups
189 						 */
190 };
191 
192 /*
193  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
194  * @flexbg_size.
195  *
196  * Returns NULL on failure otherwise address of the allocated structure.
197  */
198 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
199 {
200 	struct ext4_new_flex_group_data *flex_gd;
201 
202 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
203 	if (flex_gd == NULL)
204 		goto out3;
205 
206 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
207 		goto out2;
208 	flex_gd->count = flexbg_size;
209 
210 	flex_gd->groups = kmalloc_array(flexbg_size,
211 					sizeof(struct ext4_new_group_data),
212 					GFP_NOFS);
213 	if (flex_gd->groups == NULL)
214 		goto out2;
215 
216 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
217 					  GFP_NOFS);
218 	if (flex_gd->bg_flags == NULL)
219 		goto out1;
220 
221 	return flex_gd;
222 
223 out1:
224 	kfree(flex_gd->groups);
225 out2:
226 	kfree(flex_gd);
227 out3:
228 	return NULL;
229 }
230 
231 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
232 {
233 	kfree(flex_gd->bg_flags);
234 	kfree(flex_gd->groups);
235 	kfree(flex_gd);
236 }
237 
238 /*
239  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
240  * and inode tables for a flex group.
241  *
242  * This function is used by 64bit-resize.  Note that this function allocates
243  * group tables from the 1st group of groups contained by @flexgd, which may
244  * be a partial of a flex group.
245  *
246  * @sb: super block of fs to which the groups belongs
247  *
248  * Returns 0 on a successful allocation of the metadata blocks in the
249  * block group.
250  */
251 static int ext4_alloc_group_tables(struct super_block *sb,
252 				struct ext4_new_flex_group_data *flex_gd,
253 				int flexbg_size)
254 {
255 	struct ext4_new_group_data *group_data = flex_gd->groups;
256 	ext4_fsblk_t start_blk;
257 	ext4_fsblk_t last_blk;
258 	ext4_group_t src_group;
259 	ext4_group_t bb_index = 0;
260 	ext4_group_t ib_index = 0;
261 	ext4_group_t it_index = 0;
262 	ext4_group_t group;
263 	ext4_group_t last_group;
264 	unsigned overhead;
265 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
266 	int i;
267 
268 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
269 
270 	src_group = group_data[0].group;
271 	last_group  = src_group + flex_gd->count - 1;
272 
273 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
274 	       (last_group & ~(flexbg_size - 1))));
275 next_group:
276 	group = group_data[0].group;
277 	if (src_group >= group_data[0].group + flex_gd->count)
278 		return -ENOSPC;
279 	start_blk = ext4_group_first_block_no(sb, src_group);
280 	last_blk = start_blk + group_data[src_group - group].blocks_count;
281 
282 	overhead = ext4_group_overhead_blocks(sb, src_group);
283 
284 	start_blk += overhead;
285 
286 	/* We collect contiguous blocks as much as possible. */
287 	src_group++;
288 	for (; src_group <= last_group; src_group++) {
289 		overhead = ext4_group_overhead_blocks(sb, src_group);
290 		if (overhead == 0)
291 			last_blk += group_data[src_group - group].blocks_count;
292 		else
293 			break;
294 	}
295 
296 	/* Allocate block bitmaps */
297 	for (; bb_index < flex_gd->count; bb_index++) {
298 		if (start_blk >= last_blk)
299 			goto next_group;
300 		group_data[bb_index].block_bitmap = start_blk++;
301 		group = ext4_get_group_number(sb, start_blk - 1);
302 		group -= group_data[0].group;
303 		group_data[group].mdata_blocks++;
304 		flex_gd->bg_flags[group] &= uninit_mask;
305 	}
306 
307 	/* Allocate inode bitmaps */
308 	for (; ib_index < flex_gd->count; ib_index++) {
309 		if (start_blk >= last_blk)
310 			goto next_group;
311 		group_data[ib_index].inode_bitmap = start_blk++;
312 		group = ext4_get_group_number(sb, start_blk - 1);
313 		group -= group_data[0].group;
314 		group_data[group].mdata_blocks++;
315 		flex_gd->bg_flags[group] &= uninit_mask;
316 	}
317 
318 	/* Allocate inode tables */
319 	for (; it_index < flex_gd->count; it_index++) {
320 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
321 		ext4_fsblk_t next_group_start;
322 
323 		if (start_blk + itb > last_blk)
324 			goto next_group;
325 		group_data[it_index].inode_table = start_blk;
326 		group = ext4_get_group_number(sb, start_blk);
327 		next_group_start = ext4_group_first_block_no(sb, group + 1);
328 		group -= group_data[0].group;
329 
330 		if (start_blk + itb > next_group_start) {
331 			flex_gd->bg_flags[group + 1] &= uninit_mask;
332 			overhead = start_blk + itb - next_group_start;
333 			group_data[group + 1].mdata_blocks += overhead;
334 			itb -= overhead;
335 		}
336 
337 		group_data[group].mdata_blocks += itb;
338 		flex_gd->bg_flags[group] &= uninit_mask;
339 		start_blk += EXT4_SB(sb)->s_itb_per_group;
340 	}
341 
342 	/* Update free clusters count to exclude metadata blocks */
343 	for (i = 0; i < flex_gd->count; i++) {
344 		group_data[i].free_clusters_count -=
345 				EXT4_NUM_B2C(EXT4_SB(sb),
346 					     group_data[i].mdata_blocks);
347 	}
348 
349 	if (test_opt(sb, DEBUG)) {
350 		int i;
351 		group = group_data[0].group;
352 
353 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
354 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
355 		       flexbg_size);
356 
357 		for (i = 0; i < flex_gd->count; i++) {
358 			ext4_debug(
359 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
360 			       ext4_bg_has_super(sb, group + i) ? "normal" :
361 			       "no-super", group + i,
362 			       group_data[i].blocks_count,
363 			       group_data[i].free_clusters_count,
364 			       group_data[i].mdata_blocks);
365 		}
366 	}
367 	return 0;
368 }
369 
370 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
371 				  ext4_fsblk_t blk)
372 {
373 	struct buffer_head *bh;
374 	int err;
375 
376 	bh = sb_getblk(sb, blk);
377 	if (unlikely(!bh))
378 		return ERR_PTR(-ENOMEM);
379 	BUFFER_TRACE(bh, "get_write_access");
380 	if ((err = ext4_journal_get_write_access(handle, bh))) {
381 		brelse(bh);
382 		bh = ERR_PTR(err);
383 	} else {
384 		memset(bh->b_data, 0, sb->s_blocksize);
385 		set_buffer_uptodate(bh);
386 	}
387 
388 	return bh;
389 }
390 
391 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
392 {
393 	return ext4_journal_ensure_credits_fn(handle, credits,
394 		EXT4_MAX_TRANS_DATA, 0, 0);
395 }
396 
397 /*
398  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
399  *
400  * Helper function for ext4_setup_new_group_blocks() which set .
401  *
402  * @sb: super block
403  * @handle: journal handle
404  * @flex_gd: flex group data
405  */
406 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
407 			struct ext4_new_flex_group_data *flex_gd,
408 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
409 {
410 	struct ext4_sb_info *sbi = EXT4_SB(sb);
411 	ext4_group_t count = last_cluster - first_cluster + 1;
412 	ext4_group_t count2;
413 
414 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
415 		   last_cluster);
416 	for (count2 = count; count > 0;
417 	     count -= count2, first_cluster += count2) {
418 		ext4_fsblk_t start;
419 		struct buffer_head *bh;
420 		ext4_group_t group;
421 		int err;
422 
423 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
424 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
425 		group -= flex_gd->groups[0].group;
426 
427 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
428 		if (count2 > count)
429 			count2 = count;
430 
431 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
432 			BUG_ON(flex_gd->count > 1);
433 			continue;
434 		}
435 
436 		err = ext4_resize_ensure_credits_batch(handle, 1);
437 		if (err < 0)
438 			return err;
439 
440 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
441 		if (unlikely(!bh))
442 			return -ENOMEM;
443 
444 		BUFFER_TRACE(bh, "get_write_access");
445 		err = ext4_journal_get_write_access(handle, bh);
446 		if (err) {
447 			brelse(bh);
448 			return err;
449 		}
450 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
451 			   first_cluster, first_cluster - start, count2);
452 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
453 
454 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
455 		brelse(bh);
456 		if (unlikely(err))
457 			return err;
458 	}
459 
460 	return 0;
461 }
462 
463 /*
464  * Set up the block and inode bitmaps, and the inode table for the new groups.
465  * This doesn't need to be part of the main transaction, since we are only
466  * changing blocks outside the actual filesystem.  We still do journaling to
467  * ensure the recovery is correct in case of a failure just after resize.
468  * If any part of this fails, we simply abort the resize.
469  *
470  * setup_new_flex_group_blocks handles a flex group as follow:
471  *  1. copy super block and GDT, and initialize group tables if necessary.
472  *     In this step, we only set bits in blocks bitmaps for blocks taken by
473  *     super block and GDT.
474  *  2. allocate group tables in block bitmaps, that is, set bits in block
475  *     bitmap for blocks taken by group tables.
476  */
477 static int setup_new_flex_group_blocks(struct super_block *sb,
478 				struct ext4_new_flex_group_data *flex_gd)
479 {
480 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
481 	ext4_fsblk_t start;
482 	ext4_fsblk_t block;
483 	struct ext4_sb_info *sbi = EXT4_SB(sb);
484 	struct ext4_super_block *es = sbi->s_es;
485 	struct ext4_new_group_data *group_data = flex_gd->groups;
486 	__u16 *bg_flags = flex_gd->bg_flags;
487 	handle_t *handle;
488 	ext4_group_t group, count;
489 	struct buffer_head *bh = NULL;
490 	int reserved_gdb, i, j, err = 0, err2;
491 	int meta_bg;
492 
493 	BUG_ON(!flex_gd->count || !group_data ||
494 	       group_data[0].group != sbi->s_groups_count);
495 
496 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
497 	meta_bg = ext4_has_feature_meta_bg(sb);
498 
499 	/* This transaction may be extended/restarted along the way */
500 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
501 	if (IS_ERR(handle))
502 		return PTR_ERR(handle);
503 
504 	group = group_data[0].group;
505 	for (i = 0; i < flex_gd->count; i++, group++) {
506 		unsigned long gdblocks;
507 		ext4_grpblk_t overhead;
508 
509 		gdblocks = ext4_bg_num_gdb(sb, group);
510 		start = ext4_group_first_block_no(sb, group);
511 
512 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
513 			goto handle_itb;
514 
515 		if (meta_bg == 1) {
516 			ext4_group_t first_group;
517 			first_group = ext4_meta_bg_first_group(sb, group);
518 			if (first_group != group + 1 &&
519 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
520 				goto handle_itb;
521 		}
522 
523 		block = start + ext4_bg_has_super(sb, group);
524 		/* Copy all of the GDT blocks into the backup in this group */
525 		for (j = 0; j < gdblocks; j++, block++) {
526 			struct buffer_head *gdb;
527 
528 			ext4_debug("update backup group %#04llx\n", block);
529 			err = ext4_resize_ensure_credits_batch(handle, 1);
530 			if (err < 0)
531 				goto out;
532 
533 			gdb = sb_getblk(sb, block);
534 			if (unlikely(!gdb)) {
535 				err = -ENOMEM;
536 				goto out;
537 			}
538 
539 			BUFFER_TRACE(gdb, "get_write_access");
540 			err = ext4_journal_get_write_access(handle, gdb);
541 			if (err) {
542 				brelse(gdb);
543 				goto out;
544 			}
545 			memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
546 			       gdb->b_size);
547 			set_buffer_uptodate(gdb);
548 
549 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
550 			if (unlikely(err)) {
551 				brelse(gdb);
552 				goto out;
553 			}
554 			brelse(gdb);
555 		}
556 
557 		/* Zero out all of the reserved backup group descriptor
558 		 * table blocks
559 		 */
560 		if (ext4_bg_has_super(sb, group)) {
561 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
562 					reserved_gdb, GFP_NOFS);
563 			if (err)
564 				goto out;
565 		}
566 
567 handle_itb:
568 		/* Initialize group tables of the grop @group */
569 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
570 			goto handle_bb;
571 
572 		/* Zero out all of the inode table blocks */
573 		block = group_data[i].inode_table;
574 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
575 			   block, sbi->s_itb_per_group);
576 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
577 				       GFP_NOFS);
578 		if (err)
579 			goto out;
580 
581 handle_bb:
582 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
583 			goto handle_ib;
584 
585 		/* Initialize block bitmap of the @group */
586 		block = group_data[i].block_bitmap;
587 		err = ext4_resize_ensure_credits_batch(handle, 1);
588 		if (err < 0)
589 			goto out;
590 
591 		bh = bclean(handle, sb, block);
592 		if (IS_ERR(bh)) {
593 			err = PTR_ERR(bh);
594 			goto out;
595 		}
596 		overhead = ext4_group_overhead_blocks(sb, group);
597 		if (overhead != 0) {
598 			ext4_debug("mark backup superblock %#04llx (+0)\n",
599 				   start);
600 			ext4_set_bits(bh->b_data, 0,
601 				      EXT4_NUM_B2C(sbi, overhead));
602 		}
603 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
604 				     sb->s_blocksize * 8, bh->b_data);
605 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
606 		brelse(bh);
607 		if (err)
608 			goto out;
609 
610 handle_ib:
611 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
612 			continue;
613 
614 		/* Initialize inode bitmap of the @group */
615 		block = group_data[i].inode_bitmap;
616 		err = ext4_resize_ensure_credits_batch(handle, 1);
617 		if (err < 0)
618 			goto out;
619 		/* Mark unused entries in inode bitmap used */
620 		bh = bclean(handle, sb, block);
621 		if (IS_ERR(bh)) {
622 			err = PTR_ERR(bh);
623 			goto out;
624 		}
625 
626 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
627 				     sb->s_blocksize * 8, bh->b_data);
628 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
629 		brelse(bh);
630 		if (err)
631 			goto out;
632 	}
633 
634 	/* Mark group tables in block bitmap */
635 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
636 		count = group_table_count[j];
637 		start = (&group_data[0].block_bitmap)[j];
638 		block = start;
639 		for (i = 1; i < flex_gd->count; i++) {
640 			block += group_table_count[j];
641 			if (block == (&group_data[i].block_bitmap)[j]) {
642 				count += group_table_count[j];
643 				continue;
644 			}
645 			err = set_flexbg_block_bitmap(sb, handle,
646 						      flex_gd,
647 						      EXT4_B2C(sbi, start),
648 						      EXT4_B2C(sbi,
649 							       start + count
650 							       - 1));
651 			if (err)
652 				goto out;
653 			count = group_table_count[j];
654 			start = (&group_data[i].block_bitmap)[j];
655 			block = start;
656 		}
657 
658 		if (count) {
659 			err = set_flexbg_block_bitmap(sb, handle,
660 						      flex_gd,
661 						      EXT4_B2C(sbi, start),
662 						      EXT4_B2C(sbi,
663 							       start + count
664 							       - 1));
665 			if (err)
666 				goto out;
667 		}
668 	}
669 
670 out:
671 	err2 = ext4_journal_stop(handle);
672 	if (err2 && !err)
673 		err = err2;
674 
675 	return err;
676 }
677 
678 /*
679  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
680  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
681  * calling this for the first time.  In a sparse filesystem it will be the
682  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
683  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
684  */
685 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
686 				  unsigned *five, unsigned *seven)
687 {
688 	unsigned *min = three;
689 	int mult = 3;
690 	unsigned ret;
691 
692 	if (!ext4_has_feature_sparse_super(sb)) {
693 		ret = *min;
694 		*min += 1;
695 		return ret;
696 	}
697 
698 	if (*five < *min) {
699 		min = five;
700 		mult = 5;
701 	}
702 	if (*seven < *min) {
703 		min = seven;
704 		mult = 7;
705 	}
706 
707 	ret = *min;
708 	*min *= mult;
709 
710 	return ret;
711 }
712 
713 /*
714  * Check that all of the backup GDT blocks are held in the primary GDT block.
715  * It is assumed that they are stored in group order.  Returns the number of
716  * groups in current filesystem that have BACKUPS, or -ve error code.
717  */
718 static int verify_reserved_gdb(struct super_block *sb,
719 			       ext4_group_t end,
720 			       struct buffer_head *primary)
721 {
722 	const ext4_fsblk_t blk = primary->b_blocknr;
723 	unsigned three = 1;
724 	unsigned five = 5;
725 	unsigned seven = 7;
726 	unsigned grp;
727 	__le32 *p = (__le32 *)primary->b_data;
728 	int gdbackups = 0;
729 
730 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
731 		if (le32_to_cpu(*p++) !=
732 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
733 			ext4_warning(sb, "reserved GDT %llu"
734 				     " missing grp %d (%llu)",
735 				     blk, grp,
736 				     grp *
737 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
738 				     blk);
739 			return -EINVAL;
740 		}
741 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
742 			return -EFBIG;
743 	}
744 
745 	return gdbackups;
746 }
747 
748 /*
749  * Called when we need to bring a reserved group descriptor table block into
750  * use from the resize inode.  The primary copy of the new GDT block currently
751  * is an indirect block (under the double indirect block in the resize inode).
752  * The new backup GDT blocks will be stored as leaf blocks in this indirect
753  * block, in group order.  Even though we know all the block numbers we need,
754  * we check to ensure that the resize inode has actually reserved these blocks.
755  *
756  * Don't need to update the block bitmaps because the blocks are still in use.
757  *
758  * We get all of the error cases out of the way, so that we are sure to not
759  * fail once we start modifying the data on disk, because JBD has no rollback.
760  */
761 static int add_new_gdb(handle_t *handle, struct inode *inode,
762 		       ext4_group_t group)
763 {
764 	struct super_block *sb = inode->i_sb;
765 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
766 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
767 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
768 	struct buffer_head **o_group_desc, **n_group_desc = NULL;
769 	struct buffer_head *dind = NULL;
770 	struct buffer_head *gdb_bh = NULL;
771 	int gdbackups;
772 	struct ext4_iloc iloc = { .bh = NULL };
773 	__le32 *data;
774 	int err;
775 
776 	if (test_opt(sb, DEBUG))
777 		printk(KERN_DEBUG
778 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
779 		       gdb_num);
780 
781 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
782 	if (IS_ERR(gdb_bh))
783 		return PTR_ERR(gdb_bh);
784 
785 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
786 	if (gdbackups < 0) {
787 		err = gdbackups;
788 		goto errout;
789 	}
790 
791 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
792 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
793 	if (IS_ERR(dind)) {
794 		err = PTR_ERR(dind);
795 		dind = NULL;
796 		goto errout;
797 	}
798 
799 	data = (__le32 *)dind->b_data;
800 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
801 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
802 			     group, gdblock);
803 		err = -EINVAL;
804 		goto errout;
805 	}
806 
807 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
808 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
809 	if (unlikely(err))
810 		goto errout;
811 
812 	BUFFER_TRACE(gdb_bh, "get_write_access");
813 	err = ext4_journal_get_write_access(handle, gdb_bh);
814 	if (unlikely(err))
815 		goto errout;
816 
817 	BUFFER_TRACE(dind, "get_write_access");
818 	err = ext4_journal_get_write_access(handle, dind);
819 	if (unlikely(err))
820 		ext4_std_error(sb, err);
821 
822 	/* ext4_reserve_inode_write() gets a reference on the iloc */
823 	err = ext4_reserve_inode_write(handle, inode, &iloc);
824 	if (unlikely(err))
825 		goto errout;
826 
827 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
828 				     sizeof(struct buffer_head *),
829 				     GFP_NOFS);
830 	if (!n_group_desc) {
831 		err = -ENOMEM;
832 		ext4_warning(sb, "not enough memory for %lu groups",
833 			     gdb_num + 1);
834 		goto errout;
835 	}
836 
837 	/*
838 	 * Finally, we have all of the possible failures behind us...
839 	 *
840 	 * Remove new GDT block from inode double-indirect block and clear out
841 	 * the new GDT block for use (which also "frees" the backup GDT blocks
842 	 * from the reserved inode).  We don't need to change the bitmaps for
843 	 * these blocks, because they are marked as in-use from being in the
844 	 * reserved inode, and will become GDT blocks (primary and backup).
845 	 */
846 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
847 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
848 	if (unlikely(err)) {
849 		ext4_std_error(sb, err);
850 		goto errout;
851 	}
852 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
853 			   (9 - EXT4_SB(sb)->s_cluster_bits);
854 	ext4_mark_iloc_dirty(handle, inode, &iloc);
855 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
856 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
857 	if (unlikely(err)) {
858 		ext4_std_error(sb, err);
859 		iloc.bh = NULL;
860 		goto errout;
861 	}
862 	brelse(dind);
863 
864 	o_group_desc = EXT4_SB(sb)->s_group_desc;
865 	memcpy(n_group_desc, o_group_desc,
866 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
867 	n_group_desc[gdb_num] = gdb_bh;
868 	EXT4_SB(sb)->s_group_desc = n_group_desc;
869 	EXT4_SB(sb)->s_gdb_count++;
870 	kvfree(o_group_desc);
871 
872 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
873 	err = ext4_handle_dirty_super(handle, sb);
874 	if (err)
875 		ext4_std_error(sb, err);
876 	return err;
877 errout:
878 	kvfree(n_group_desc);
879 	brelse(iloc.bh);
880 	brelse(dind);
881 	brelse(gdb_bh);
882 
883 	ext4_debug("leaving with error %d\n", err);
884 	return err;
885 }
886 
887 /*
888  * add_new_gdb_meta_bg is the sister of add_new_gdb.
889  */
890 static int add_new_gdb_meta_bg(struct super_block *sb,
891 			       handle_t *handle, ext4_group_t group) {
892 	ext4_fsblk_t gdblock;
893 	struct buffer_head *gdb_bh;
894 	struct buffer_head **o_group_desc, **n_group_desc;
895 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
896 	int err;
897 
898 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
899 		   ext4_bg_has_super(sb, group);
900 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
901 	if (IS_ERR(gdb_bh))
902 		return PTR_ERR(gdb_bh);
903 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
904 				     sizeof(struct buffer_head *),
905 				     GFP_NOFS);
906 	if (!n_group_desc) {
907 		brelse(gdb_bh);
908 		err = -ENOMEM;
909 		ext4_warning(sb, "not enough memory for %lu groups",
910 			     gdb_num + 1);
911 		return err;
912 	}
913 
914 	o_group_desc = EXT4_SB(sb)->s_group_desc;
915 	memcpy(n_group_desc, o_group_desc,
916 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
917 	n_group_desc[gdb_num] = gdb_bh;
918 
919 	BUFFER_TRACE(gdb_bh, "get_write_access");
920 	err = ext4_journal_get_write_access(handle, gdb_bh);
921 	if (err) {
922 		kvfree(n_group_desc);
923 		brelse(gdb_bh);
924 		return err;
925 	}
926 
927 	EXT4_SB(sb)->s_group_desc = n_group_desc;
928 	EXT4_SB(sb)->s_gdb_count++;
929 	kvfree(o_group_desc);
930 	return err;
931 }
932 
933 /*
934  * Called when we are adding a new group which has a backup copy of each of
935  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
936  * We need to add these reserved backup GDT blocks to the resize inode, so
937  * that they are kept for future resizing and not allocated to files.
938  *
939  * Each reserved backup GDT block will go into a different indirect block.
940  * The indirect blocks are actually the primary reserved GDT blocks,
941  * so we know in advance what their block numbers are.  We only get the
942  * double-indirect block to verify it is pointing to the primary reserved
943  * GDT blocks so we don't overwrite a data block by accident.  The reserved
944  * backup GDT blocks are stored in their reserved primary GDT block.
945  */
946 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
947 			      ext4_group_t group)
948 {
949 	struct super_block *sb = inode->i_sb;
950 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
951 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
952 	struct buffer_head **primary;
953 	struct buffer_head *dind;
954 	struct ext4_iloc iloc;
955 	ext4_fsblk_t blk;
956 	__le32 *data, *end;
957 	int gdbackups = 0;
958 	int res, i;
959 	int err;
960 
961 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
962 	if (!primary)
963 		return -ENOMEM;
964 
965 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
966 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
967 	if (IS_ERR(dind)) {
968 		err = PTR_ERR(dind);
969 		dind = NULL;
970 		goto exit_free;
971 	}
972 
973 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
974 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
975 					 EXT4_ADDR_PER_BLOCK(sb));
976 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
977 
978 	/* Get each reserved primary GDT block and verify it holds backups */
979 	for (res = 0; res < reserved_gdb; res++, blk++) {
980 		if (le32_to_cpu(*data) != blk) {
981 			ext4_warning(sb, "reserved block %llu"
982 				     " not at offset %ld",
983 				     blk,
984 				     (long)(data - (__le32 *)dind->b_data));
985 			err = -EINVAL;
986 			goto exit_bh;
987 		}
988 		primary[res] = ext4_sb_bread(sb, blk, 0);
989 		if (IS_ERR(primary[res])) {
990 			err = PTR_ERR(primary[res]);
991 			primary[res] = NULL;
992 			goto exit_bh;
993 		}
994 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
995 		if (gdbackups < 0) {
996 			brelse(primary[res]);
997 			err = gdbackups;
998 			goto exit_bh;
999 		}
1000 		if (++data >= end)
1001 			data = (__le32 *)dind->b_data;
1002 	}
1003 
1004 	for (i = 0; i < reserved_gdb; i++) {
1005 		BUFFER_TRACE(primary[i], "get_write_access");
1006 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1007 			goto exit_bh;
1008 	}
1009 
1010 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1011 		goto exit_bh;
1012 
1013 	/*
1014 	 * Finally we can add each of the reserved backup GDT blocks from
1015 	 * the new group to its reserved primary GDT block.
1016 	 */
1017 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1018 	for (i = 0; i < reserved_gdb; i++) {
1019 		int err2;
1020 		data = (__le32 *)primary[i]->b_data;
1021 		/* printk("reserving backup %lu[%u] = %lu\n",
1022 		       primary[i]->b_blocknr, gdbackups,
1023 		       blk + primary[i]->b_blocknr); */
1024 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1025 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1026 		if (!err)
1027 			err = err2;
1028 	}
1029 
1030 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1031 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1032 
1033 exit_bh:
1034 	while (--res >= 0)
1035 		brelse(primary[res]);
1036 	brelse(dind);
1037 
1038 exit_free:
1039 	kfree(primary);
1040 
1041 	return err;
1042 }
1043 
1044 /*
1045  * Update the backup copies of the ext4 metadata.  These don't need to be part
1046  * of the main resize transaction, because e2fsck will re-write them if there
1047  * is a problem (basically only OOM will cause a problem).  However, we
1048  * _should_ update the backups if possible, in case the primary gets trashed
1049  * for some reason and we need to run e2fsck from a backup superblock.  The
1050  * important part is that the new block and inode counts are in the backup
1051  * superblocks, and the location of the new group metadata in the GDT backups.
1052  *
1053  * We do not need take the s_resize_lock for this, because these
1054  * blocks are not otherwise touched by the filesystem code when it is
1055  * mounted.  We don't need to worry about last changing from
1056  * sbi->s_groups_count, because the worst that can happen is that we
1057  * do not copy the full number of backups at this time.  The resize
1058  * which changed s_groups_count will backup again.
1059  */
1060 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1061 			   int size, int meta_bg)
1062 {
1063 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1064 	ext4_group_t last;
1065 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1066 	unsigned three = 1;
1067 	unsigned five = 5;
1068 	unsigned seven = 7;
1069 	ext4_group_t group = 0;
1070 	int rest = sb->s_blocksize - size;
1071 	handle_t *handle;
1072 	int err = 0, err2;
1073 
1074 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1075 	if (IS_ERR(handle)) {
1076 		group = 1;
1077 		err = PTR_ERR(handle);
1078 		goto exit_err;
1079 	}
1080 
1081 	if (meta_bg == 0) {
1082 		group = ext4_list_backups(sb, &three, &five, &seven);
1083 		last = sbi->s_groups_count;
1084 	} else {
1085 		group = ext4_get_group_number(sb, blk_off) + 1;
1086 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1087 	}
1088 
1089 	while (group < sbi->s_groups_count) {
1090 		struct buffer_head *bh;
1091 		ext4_fsblk_t backup_block;
1092 
1093 		/* Out of journal space, and can't get more - abort - so sad */
1094 		err = ext4_resize_ensure_credits_batch(handle, 1);
1095 		if (err < 0)
1096 			break;
1097 
1098 		if (meta_bg == 0)
1099 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1100 		else
1101 			backup_block = (ext4_group_first_block_no(sb, group) +
1102 					ext4_bg_has_super(sb, group));
1103 
1104 		bh = sb_getblk(sb, backup_block);
1105 		if (unlikely(!bh)) {
1106 			err = -ENOMEM;
1107 			break;
1108 		}
1109 		ext4_debug("update metadata backup %llu(+%llu)\n",
1110 			   backup_block, backup_block -
1111 			   ext4_group_first_block_no(sb, group));
1112 		BUFFER_TRACE(bh, "get_write_access");
1113 		if ((err = ext4_journal_get_write_access(handle, bh))) {
1114 			brelse(bh);
1115 			break;
1116 		}
1117 		lock_buffer(bh);
1118 		memcpy(bh->b_data, data, size);
1119 		if (rest)
1120 			memset(bh->b_data + size, 0, rest);
1121 		set_buffer_uptodate(bh);
1122 		unlock_buffer(bh);
1123 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1124 		if (unlikely(err))
1125 			ext4_std_error(sb, err);
1126 		brelse(bh);
1127 
1128 		if (meta_bg == 0)
1129 			group = ext4_list_backups(sb, &three, &five, &seven);
1130 		else if (group == last)
1131 			break;
1132 		else
1133 			group = last;
1134 	}
1135 	if ((err2 = ext4_journal_stop(handle)) && !err)
1136 		err = err2;
1137 
1138 	/*
1139 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1140 	 * late to revert the resize, we shouldn't fail just because of
1141 	 * the backup copies (they are only needed in case of corruption).
1142 	 *
1143 	 * However, if we got here we have a journal problem too, so we
1144 	 * can't really start a transaction to mark the superblock.
1145 	 * Chicken out and just set the flag on the hope it will be written
1146 	 * to disk, and if not - we will simply wait until next fsck.
1147 	 */
1148 exit_err:
1149 	if (err) {
1150 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1151 			     "forcing fsck on next reboot", group, err);
1152 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1153 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1154 		mark_buffer_dirty(sbi->s_sbh);
1155 	}
1156 }
1157 
1158 /*
1159  * ext4_add_new_descs() adds @count group descriptor of groups
1160  * starting at @group
1161  *
1162  * @handle: journal handle
1163  * @sb: super block
1164  * @group: the group no. of the first group desc to be added
1165  * @resize_inode: the resize inode
1166  * @count: number of group descriptors to be added
1167  */
1168 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1169 			      ext4_group_t group, struct inode *resize_inode,
1170 			      ext4_group_t count)
1171 {
1172 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1173 	struct ext4_super_block *es = sbi->s_es;
1174 	struct buffer_head *gdb_bh;
1175 	int i, gdb_off, gdb_num, err = 0;
1176 	int meta_bg;
1177 
1178 	meta_bg = ext4_has_feature_meta_bg(sb);
1179 	for (i = 0; i < count; i++, group++) {
1180 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1181 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1182 
1183 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1184 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1185 
1186 		/*
1187 		 * We will only either add reserved group blocks to a backup group
1188 		 * or remove reserved blocks for the first group in a new group block.
1189 		 * Doing both would be mean more complex code, and sane people don't
1190 		 * use non-sparse filesystems anymore.  This is already checked above.
1191 		 */
1192 		if (gdb_off) {
1193 			gdb_bh = sbi->s_group_desc[gdb_num];
1194 			BUFFER_TRACE(gdb_bh, "get_write_access");
1195 			err = ext4_journal_get_write_access(handle, gdb_bh);
1196 
1197 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1198 				err = reserve_backup_gdb(handle, resize_inode, group);
1199 		} else if (meta_bg != 0) {
1200 			err = add_new_gdb_meta_bg(sb, handle, group);
1201 		} else {
1202 			err = add_new_gdb(handle, resize_inode, group);
1203 		}
1204 		if (err)
1205 			break;
1206 	}
1207 	return err;
1208 }
1209 
1210 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1211 {
1212 	struct buffer_head *bh = sb_getblk(sb, block);
1213 	if (unlikely(!bh))
1214 		return NULL;
1215 	if (!bh_uptodate_or_lock(bh)) {
1216 		if (bh_submit_read(bh) < 0) {
1217 			brelse(bh);
1218 			return NULL;
1219 		}
1220 	}
1221 
1222 	return bh;
1223 }
1224 
1225 static int ext4_set_bitmap_checksums(struct super_block *sb,
1226 				     ext4_group_t group,
1227 				     struct ext4_group_desc *gdp,
1228 				     struct ext4_new_group_data *group_data)
1229 {
1230 	struct buffer_head *bh;
1231 
1232 	if (!ext4_has_metadata_csum(sb))
1233 		return 0;
1234 
1235 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1236 	if (!bh)
1237 		return -EIO;
1238 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1239 				   EXT4_INODES_PER_GROUP(sb) / 8);
1240 	brelse(bh);
1241 
1242 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1243 	if (!bh)
1244 		return -EIO;
1245 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1246 	brelse(bh);
1247 
1248 	return 0;
1249 }
1250 
1251 /*
1252  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1253  */
1254 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1255 				struct ext4_new_flex_group_data *flex_gd)
1256 {
1257 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1258 	struct ext4_group_desc		*gdp;
1259 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1260 	struct buffer_head		*gdb_bh;
1261 	ext4_group_t			group;
1262 	__u16				*bg_flags = flex_gd->bg_flags;
1263 	int				i, gdb_off, gdb_num, err = 0;
1264 
1265 
1266 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1267 		group = group_data->group;
1268 
1269 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1270 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1271 
1272 		/*
1273 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1274 		 */
1275 		gdb_bh = sbi->s_group_desc[gdb_num];
1276 		/* Update group descriptor block for new group */
1277 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1278 						 gdb_off * EXT4_DESC_SIZE(sb));
1279 
1280 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1281 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1282 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1283 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1284 		if (err) {
1285 			ext4_std_error(sb, err);
1286 			break;
1287 		}
1288 
1289 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1290 		ext4_free_group_clusters_set(sb, gdp,
1291 					     group_data->free_clusters_count);
1292 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1293 		if (ext4_has_group_desc_csum(sb))
1294 			ext4_itable_unused_set(sb, gdp,
1295 					       EXT4_INODES_PER_GROUP(sb));
1296 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1297 		ext4_group_desc_csum_set(sb, group, gdp);
1298 
1299 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1300 		if (unlikely(err)) {
1301 			ext4_std_error(sb, err);
1302 			break;
1303 		}
1304 
1305 		/*
1306 		 * We can allocate memory for mb_alloc based on the new group
1307 		 * descriptor
1308 		 */
1309 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1310 		if (err)
1311 			break;
1312 	}
1313 	return err;
1314 }
1315 
1316 /*
1317  * ext4_update_super() updates the super block so that the newly added
1318  * groups can be seen by the filesystem.
1319  *
1320  * @sb: super block
1321  * @flex_gd: new added groups
1322  */
1323 static void ext4_update_super(struct super_block *sb,
1324 			     struct ext4_new_flex_group_data *flex_gd)
1325 {
1326 	ext4_fsblk_t blocks_count = 0;
1327 	ext4_fsblk_t free_blocks = 0;
1328 	ext4_fsblk_t reserved_blocks = 0;
1329 	struct ext4_new_group_data *group_data = flex_gd->groups;
1330 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1331 	struct ext4_super_block *es = sbi->s_es;
1332 	int i;
1333 
1334 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1335 	/*
1336 	 * Make the new blocks and inodes valid next.  We do this before
1337 	 * increasing the group count so that once the group is enabled,
1338 	 * all of its blocks and inodes are already valid.
1339 	 *
1340 	 * We always allocate group-by-group, then block-by-block or
1341 	 * inode-by-inode within a group, so enabling these
1342 	 * blocks/inodes before the group is live won't actually let us
1343 	 * allocate the new space yet.
1344 	 */
1345 	for (i = 0; i < flex_gd->count; i++) {
1346 		blocks_count += group_data[i].blocks_count;
1347 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1348 	}
1349 
1350 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1351 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1352 	reserved_blocks *= blocks_count;
1353 	do_div(reserved_blocks, 100);
1354 
1355 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1356 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1357 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1358 		     flex_gd->count);
1359 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1360 		     flex_gd->count);
1361 
1362 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1363 	/*
1364 	 * We need to protect s_groups_count against other CPUs seeing
1365 	 * inconsistent state in the superblock.
1366 	 *
1367 	 * The precise rules we use are:
1368 	 *
1369 	 * * Writers must perform a smp_wmb() after updating all
1370 	 *   dependent data and before modifying the groups count
1371 	 *
1372 	 * * Readers must perform an smp_rmb() after reading the groups
1373 	 *   count and before reading any dependent data.
1374 	 *
1375 	 * NB. These rules can be relaxed when checking the group count
1376 	 * while freeing data, as we can only allocate from a block
1377 	 * group after serialising against the group count, and we can
1378 	 * only then free after serialising in turn against that
1379 	 * allocation.
1380 	 */
1381 	smp_wmb();
1382 
1383 	/* Update the global fs size fields */
1384 	sbi->s_groups_count += flex_gd->count;
1385 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1386 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1387 
1388 	/* Update the reserved block counts only once the new group is
1389 	 * active. */
1390 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1391 				reserved_blocks);
1392 
1393 	/* Update the free space counts */
1394 	percpu_counter_add(&sbi->s_freeclusters_counter,
1395 			   EXT4_NUM_B2C(sbi, free_blocks));
1396 	percpu_counter_add(&sbi->s_freeinodes_counter,
1397 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1398 
1399 	ext4_debug("free blocks count %llu",
1400 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1401 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1402 		ext4_group_t flex_group;
1403 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1404 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1405 			     &sbi->s_flex_groups[flex_group].free_clusters);
1406 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1407 			   &sbi->s_flex_groups[flex_group].free_inodes);
1408 	}
1409 
1410 	/*
1411 	 * Update the fs overhead information
1412 	 */
1413 	ext4_calculate_overhead(sb);
1414 
1415 	if (test_opt(sb, DEBUG))
1416 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1417 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1418 		       blocks_count, free_blocks, reserved_blocks);
1419 }
1420 
1421 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1422  * _before_ we start modifying the filesystem, because we cannot abort the
1423  * transaction and not have it write the data to disk.
1424  */
1425 static int ext4_flex_group_add(struct super_block *sb,
1426 			       struct inode *resize_inode,
1427 			       struct ext4_new_flex_group_data *flex_gd)
1428 {
1429 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1430 	struct ext4_super_block *es = sbi->s_es;
1431 	ext4_fsblk_t o_blocks_count;
1432 	ext4_grpblk_t last;
1433 	ext4_group_t group;
1434 	handle_t *handle;
1435 	unsigned reserved_gdb;
1436 	int err = 0, err2 = 0, credit;
1437 
1438 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1439 
1440 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1441 	o_blocks_count = ext4_blocks_count(es);
1442 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1443 	BUG_ON(last);
1444 
1445 	err = setup_new_flex_group_blocks(sb, flex_gd);
1446 	if (err)
1447 		goto exit;
1448 	/*
1449 	 * We will always be modifying at least the superblock and  GDT
1450 	 * blocks.  If we are adding a group past the last current GDT block,
1451 	 * we will also modify the inode and the dindirect block.  If we
1452 	 * are adding a group with superblock/GDT backups  we will also
1453 	 * modify each of the reserved GDT dindirect blocks.
1454 	 */
1455 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1456 	/* GDT blocks */
1457 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1458 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1459 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1460 	if (IS_ERR(handle)) {
1461 		err = PTR_ERR(handle);
1462 		goto exit;
1463 	}
1464 
1465 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1466 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1467 	if (err)
1468 		goto exit_journal;
1469 
1470 	group = flex_gd->groups[0].group;
1471 	BUG_ON(group != sbi->s_groups_count);
1472 	err = ext4_add_new_descs(handle, sb, group,
1473 				resize_inode, flex_gd->count);
1474 	if (err)
1475 		goto exit_journal;
1476 
1477 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1478 	if (err)
1479 		goto exit_journal;
1480 
1481 	ext4_update_super(sb, flex_gd);
1482 
1483 	err = ext4_handle_dirty_super(handle, sb);
1484 
1485 exit_journal:
1486 	err2 = ext4_journal_stop(handle);
1487 	if (!err)
1488 		err = err2;
1489 
1490 	if (!err) {
1491 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1492 		int gdb_num_end = ((group + flex_gd->count - 1) /
1493 				   EXT4_DESC_PER_BLOCK(sb));
1494 		int meta_bg = ext4_has_feature_meta_bg(sb);
1495 		sector_t old_gdb = 0;
1496 
1497 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1498 			       sizeof(struct ext4_super_block), 0);
1499 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1500 			struct buffer_head *gdb_bh;
1501 
1502 			gdb_bh = sbi->s_group_desc[gdb_num];
1503 			if (old_gdb == gdb_bh->b_blocknr)
1504 				continue;
1505 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1506 				       gdb_bh->b_size, meta_bg);
1507 			old_gdb = gdb_bh->b_blocknr;
1508 		}
1509 	}
1510 exit:
1511 	return err;
1512 }
1513 
1514 static int ext4_setup_next_flex_gd(struct super_block *sb,
1515 				    struct ext4_new_flex_group_data *flex_gd,
1516 				    ext4_fsblk_t n_blocks_count,
1517 				    unsigned long flexbg_size)
1518 {
1519 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1520 	struct ext4_super_block *es = sbi->s_es;
1521 	struct ext4_new_group_data *group_data = flex_gd->groups;
1522 	ext4_fsblk_t o_blocks_count;
1523 	ext4_group_t n_group;
1524 	ext4_group_t group;
1525 	ext4_group_t last_group;
1526 	ext4_grpblk_t last;
1527 	ext4_grpblk_t clusters_per_group;
1528 	unsigned long i;
1529 
1530 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1531 
1532 	o_blocks_count = ext4_blocks_count(es);
1533 
1534 	if (o_blocks_count == n_blocks_count)
1535 		return 0;
1536 
1537 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1538 	BUG_ON(last);
1539 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1540 
1541 	last_group = group | (flexbg_size - 1);
1542 	if (last_group > n_group)
1543 		last_group = n_group;
1544 
1545 	flex_gd->count = last_group - group + 1;
1546 
1547 	for (i = 0; i < flex_gd->count; i++) {
1548 		int overhead;
1549 
1550 		group_data[i].group = group + i;
1551 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1552 		overhead = ext4_group_overhead_blocks(sb, group + i);
1553 		group_data[i].mdata_blocks = overhead;
1554 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1555 		if (ext4_has_group_desc_csum(sb)) {
1556 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1557 					       EXT4_BG_INODE_UNINIT;
1558 			if (!test_opt(sb, INIT_INODE_TABLE))
1559 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1560 		} else
1561 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1562 	}
1563 
1564 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1565 		/* We need to initialize block bitmap of last group. */
1566 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1567 
1568 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1569 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1570 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1571 						       last - 1;
1572 	}
1573 
1574 	return 1;
1575 }
1576 
1577 /* Add group descriptor data to an existing or new group descriptor block.
1578  * Ensure we handle all possible error conditions _before_ we start modifying
1579  * the filesystem, because we cannot abort the transaction and not have it
1580  * write the data to disk.
1581  *
1582  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1583  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1584  *
1585  * We only need to hold the superblock lock while we are actually adding
1586  * in the new group's counts to the superblock.  Prior to that we have
1587  * not really "added" the group at all.  We re-check that we are still
1588  * adding in the last group in case things have changed since verifying.
1589  */
1590 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1591 {
1592 	struct ext4_new_flex_group_data flex_gd;
1593 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1594 	struct ext4_super_block *es = sbi->s_es;
1595 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1596 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1597 	struct inode *inode = NULL;
1598 	int gdb_off;
1599 	int err;
1600 	__u16 bg_flags = 0;
1601 
1602 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1603 
1604 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1605 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1606 		return -EPERM;
1607 	}
1608 
1609 	if (ext4_blocks_count(es) + input->blocks_count <
1610 	    ext4_blocks_count(es)) {
1611 		ext4_warning(sb, "blocks_count overflow");
1612 		return -EINVAL;
1613 	}
1614 
1615 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1616 	    le32_to_cpu(es->s_inodes_count)) {
1617 		ext4_warning(sb, "inodes_count overflow");
1618 		return -EINVAL;
1619 	}
1620 
1621 	if (reserved_gdb || gdb_off == 0) {
1622 		if (!ext4_has_feature_resize_inode(sb) ||
1623 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1624 			ext4_warning(sb,
1625 				     "No reserved GDT blocks, can't resize");
1626 			return -EPERM;
1627 		}
1628 		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1629 		if (IS_ERR(inode)) {
1630 			ext4_warning(sb, "Error opening resize inode");
1631 			return PTR_ERR(inode);
1632 		}
1633 	}
1634 
1635 
1636 	err = verify_group_input(sb, input);
1637 	if (err)
1638 		goto out;
1639 
1640 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1641 	if (err)
1642 		goto out;
1643 
1644 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1645 	if (err)
1646 		goto out;
1647 
1648 	flex_gd.count = 1;
1649 	flex_gd.groups = input;
1650 	flex_gd.bg_flags = &bg_flags;
1651 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1652 out:
1653 	iput(inode);
1654 	return err;
1655 } /* ext4_group_add */
1656 
1657 /*
1658  * extend a group without checking assuming that checking has been done.
1659  */
1660 static int ext4_group_extend_no_check(struct super_block *sb,
1661 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1662 {
1663 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1664 	handle_t *handle;
1665 	int err = 0, err2;
1666 
1667 	/* We will update the superblock, one block bitmap, and
1668 	 * one group descriptor via ext4_group_add_blocks().
1669 	 */
1670 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1671 	if (IS_ERR(handle)) {
1672 		err = PTR_ERR(handle);
1673 		ext4_warning(sb, "error %d on journal start", err);
1674 		return err;
1675 	}
1676 
1677 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1678 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1679 	if (err) {
1680 		ext4_warning(sb, "error %d on journal write access", err);
1681 		goto errout;
1682 	}
1683 
1684 	ext4_blocks_count_set(es, o_blocks_count + add);
1685 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1686 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1687 		   o_blocks_count + add);
1688 	/* We add the blocks to the bitmap and set the group need init bit */
1689 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1690 	if (err)
1691 		goto errout;
1692 	ext4_handle_dirty_super(handle, sb);
1693 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1694 		   o_blocks_count + add);
1695 errout:
1696 	err2 = ext4_journal_stop(handle);
1697 	if (err2 && !err)
1698 		err = err2;
1699 
1700 	if (!err) {
1701 		if (test_opt(sb, DEBUG))
1702 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1703 			       "blocks\n", ext4_blocks_count(es));
1704 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1705 			       (char *)es, sizeof(struct ext4_super_block), 0);
1706 	}
1707 	return err;
1708 }
1709 
1710 /*
1711  * Extend the filesystem to the new number of blocks specified.  This entry
1712  * point is only used to extend the current filesystem to the end of the last
1713  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1714  * for emergencies (because it has no dependencies on reserved blocks).
1715  *
1716  * If we _really_ wanted, we could use default values to call ext4_group_add()
1717  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1718  * GDT blocks are reserved to grow to the desired size.
1719  */
1720 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1721 		      ext4_fsblk_t n_blocks_count)
1722 {
1723 	ext4_fsblk_t o_blocks_count;
1724 	ext4_grpblk_t last;
1725 	ext4_grpblk_t add;
1726 	struct buffer_head *bh;
1727 	int err;
1728 	ext4_group_t group;
1729 
1730 	o_blocks_count = ext4_blocks_count(es);
1731 
1732 	if (test_opt(sb, DEBUG))
1733 		ext4_msg(sb, KERN_DEBUG,
1734 			 "extending last group from %llu to %llu blocks",
1735 			 o_blocks_count, n_blocks_count);
1736 
1737 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1738 		return 0;
1739 
1740 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1741 		ext4_msg(sb, KERN_ERR,
1742 			 "filesystem too large to resize to %llu blocks safely",
1743 			 n_blocks_count);
1744 		return -EINVAL;
1745 	}
1746 
1747 	if (n_blocks_count < o_blocks_count) {
1748 		ext4_warning(sb, "can't shrink FS - resize aborted");
1749 		return -EINVAL;
1750 	}
1751 
1752 	/* Handle the remaining blocks in the last group only. */
1753 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1754 
1755 	if (last == 0) {
1756 		ext4_warning(sb, "need to use ext2online to resize further");
1757 		return -EPERM;
1758 	}
1759 
1760 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1761 
1762 	if (o_blocks_count + add < o_blocks_count) {
1763 		ext4_warning(sb, "blocks_count overflow");
1764 		return -EINVAL;
1765 	}
1766 
1767 	if (o_blocks_count + add > n_blocks_count)
1768 		add = n_blocks_count - o_blocks_count;
1769 
1770 	if (o_blocks_count + add < n_blocks_count)
1771 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1772 			     o_blocks_count + add, add);
1773 
1774 	/* See if the device is actually as big as what was requested */
1775 	bh = sb_bread(sb, o_blocks_count + add - 1);
1776 	if (!bh) {
1777 		ext4_warning(sb, "can't read last block, resize aborted");
1778 		return -ENOSPC;
1779 	}
1780 	brelse(bh);
1781 
1782 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1783 	return err;
1784 } /* ext4_group_extend */
1785 
1786 
1787 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1788 {
1789 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1790 }
1791 
1792 /*
1793  * Release the resize inode and drop the resize_inode feature if there
1794  * are no more reserved gdt blocks, and then convert the file system
1795  * to enable meta_bg
1796  */
1797 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1798 {
1799 	handle_t *handle;
1800 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1801 	struct ext4_super_block *es = sbi->s_es;
1802 	struct ext4_inode_info *ei = EXT4_I(inode);
1803 	ext4_fsblk_t nr;
1804 	int i, ret, err = 0;
1805 	int credits = 1;
1806 
1807 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1808 	if (inode) {
1809 		if (es->s_reserved_gdt_blocks) {
1810 			ext4_error(sb, "Unexpected non-zero "
1811 				   "s_reserved_gdt_blocks");
1812 			return -EPERM;
1813 		}
1814 
1815 		/* Do a quick sanity check of the resize inode */
1816 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1817 					     (9 - sbi->s_cluster_bits)))
1818 			goto invalid_resize_inode;
1819 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1820 			if (i == EXT4_DIND_BLOCK) {
1821 				if (ei->i_data[i])
1822 					continue;
1823 				else
1824 					goto invalid_resize_inode;
1825 			}
1826 			if (ei->i_data[i])
1827 				goto invalid_resize_inode;
1828 		}
1829 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1830 	}
1831 
1832 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1833 	if (IS_ERR(handle))
1834 		return PTR_ERR(handle);
1835 
1836 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1837 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1838 	if (err)
1839 		goto errout;
1840 
1841 	ext4_clear_feature_resize_inode(sb);
1842 	ext4_set_feature_meta_bg(sb);
1843 	sbi->s_es->s_first_meta_bg =
1844 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1845 
1846 	err = ext4_handle_dirty_super(handle, sb);
1847 	if (err) {
1848 		ext4_std_error(sb, err);
1849 		goto errout;
1850 	}
1851 
1852 	if (inode) {
1853 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1854 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1855 				 EXT4_FREE_BLOCKS_METADATA |
1856 				 EXT4_FREE_BLOCKS_FORGET);
1857 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1858 		inode->i_blocks = 0;
1859 
1860 		err = ext4_mark_inode_dirty(handle, inode);
1861 		if (err)
1862 			ext4_std_error(sb, err);
1863 	}
1864 
1865 errout:
1866 	ret = ext4_journal_stop(handle);
1867 	if (!err)
1868 		err = ret;
1869 	return ret;
1870 
1871 invalid_resize_inode:
1872 	ext4_error(sb, "corrupted/inconsistent resize inode");
1873 	return -EINVAL;
1874 }
1875 
1876 /*
1877  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1878  *
1879  * @sb: super block of the fs to be resized
1880  * @n_blocks_count: the number of blocks resides in the resized fs
1881  */
1882 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1883 {
1884 	struct ext4_new_flex_group_data *flex_gd = NULL;
1885 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1886 	struct ext4_super_block *es = sbi->s_es;
1887 	struct buffer_head *bh;
1888 	struct inode *resize_inode = NULL;
1889 	ext4_grpblk_t add, offset;
1890 	unsigned long n_desc_blocks;
1891 	unsigned long o_desc_blocks;
1892 	ext4_group_t o_group;
1893 	ext4_group_t n_group;
1894 	ext4_fsblk_t o_blocks_count;
1895 	ext4_fsblk_t n_blocks_count_retry = 0;
1896 	unsigned long last_update_time = 0;
1897 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1898 	int meta_bg;
1899 
1900 	/* See if the device is actually as big as what was requested */
1901 	bh = sb_bread(sb, n_blocks_count - 1);
1902 	if (!bh) {
1903 		ext4_warning(sb, "can't read last block, resize aborted");
1904 		return -ENOSPC;
1905 	}
1906 	brelse(bh);
1907 
1908 retry:
1909 	o_blocks_count = ext4_blocks_count(es);
1910 
1911 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1912 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1913 
1914 	if (n_blocks_count < o_blocks_count) {
1915 		/* On-line shrinking not supported */
1916 		ext4_warning(sb, "can't shrink FS - resize aborted");
1917 		return -EINVAL;
1918 	}
1919 
1920 	if (n_blocks_count == o_blocks_count)
1921 		/* Nothing need to do */
1922 		return 0;
1923 
1924 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1925 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1926 		ext4_warning(sb, "resize would cause inodes_count overflow");
1927 		return -EINVAL;
1928 	}
1929 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1930 
1931 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1932 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1933 
1934 	meta_bg = ext4_has_feature_meta_bg(sb);
1935 
1936 	if (ext4_has_feature_resize_inode(sb)) {
1937 		if (meta_bg) {
1938 			ext4_error(sb, "resize_inode and meta_bg enabled "
1939 				   "simultaneously");
1940 			return -EINVAL;
1941 		}
1942 		if (n_desc_blocks > o_desc_blocks +
1943 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1944 			n_blocks_count_retry = n_blocks_count;
1945 			n_desc_blocks = o_desc_blocks +
1946 				le16_to_cpu(es->s_reserved_gdt_blocks);
1947 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1948 			n_blocks_count = (ext4_fsblk_t)n_group *
1949 				EXT4_BLOCKS_PER_GROUP(sb) +
1950 				le32_to_cpu(es->s_first_data_block);
1951 			n_group--; /* set to last group number */
1952 		}
1953 
1954 		if (!resize_inode)
1955 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1956 						 EXT4_IGET_SPECIAL);
1957 		if (IS_ERR(resize_inode)) {
1958 			ext4_warning(sb, "Error opening resize inode");
1959 			return PTR_ERR(resize_inode);
1960 		}
1961 	}
1962 
1963 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1964 		err = ext4_convert_meta_bg(sb, resize_inode);
1965 		if (err)
1966 			goto out;
1967 		if (resize_inode) {
1968 			iput(resize_inode);
1969 			resize_inode = NULL;
1970 		}
1971 		if (n_blocks_count_retry) {
1972 			n_blocks_count = n_blocks_count_retry;
1973 			n_blocks_count_retry = 0;
1974 			goto retry;
1975 		}
1976 	}
1977 
1978 	/*
1979 	 * Make sure the last group has enough space so that it's
1980 	 * guaranteed to have enough space for all metadata blocks
1981 	 * that it might need to hold.  (We might not need to store
1982 	 * the inode table blocks in the last block group, but there
1983 	 * will be cases where this might be needed.)
1984 	 */
1985 	if ((ext4_group_first_block_no(sb, n_group) +
1986 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
1987 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
1988 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
1989 		n_group--;
1990 		n_blocks_count_retry = 0;
1991 		if (resize_inode) {
1992 			iput(resize_inode);
1993 			resize_inode = NULL;
1994 		}
1995 		goto retry;
1996 	}
1997 
1998 	/* extend the last group */
1999 	if (n_group == o_group)
2000 		add = n_blocks_count - o_blocks_count;
2001 	else
2002 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2003 	if (add > 0) {
2004 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2005 		if (err)
2006 			goto out;
2007 	}
2008 
2009 	if (ext4_blocks_count(es) == n_blocks_count)
2010 		goto out;
2011 
2012 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2013 	if (err)
2014 		goto out;
2015 
2016 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2017 	if (err)
2018 		goto out;
2019 
2020 	flex_gd = alloc_flex_gd(flexbg_size);
2021 	if (flex_gd == NULL) {
2022 		err = -ENOMEM;
2023 		goto out;
2024 	}
2025 
2026 	/* Add flex groups. Note that a regular group is a
2027 	 * flex group with 1 group.
2028 	 */
2029 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2030 					      flexbg_size)) {
2031 		if (jiffies - last_update_time > HZ * 10) {
2032 			if (last_update_time)
2033 				ext4_msg(sb, KERN_INFO,
2034 					 "resized to %llu blocks",
2035 					 ext4_blocks_count(es));
2036 			last_update_time = jiffies;
2037 		}
2038 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2039 			break;
2040 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2041 		if (unlikely(err))
2042 			break;
2043 	}
2044 
2045 	if (!err && n_blocks_count_retry) {
2046 		n_blocks_count = n_blocks_count_retry;
2047 		n_blocks_count_retry = 0;
2048 		free_flex_gd(flex_gd);
2049 		flex_gd = NULL;
2050 		if (resize_inode) {
2051 			iput(resize_inode);
2052 			resize_inode = NULL;
2053 		}
2054 		goto retry;
2055 	}
2056 
2057 out:
2058 	if (flex_gd)
2059 		free_flex_gd(flex_gd);
2060 	if (resize_inode != NULL)
2061 		iput(resize_inode);
2062 	if (err)
2063 		ext4_warning(sb, "error (%d) occurred during "
2064 			     "file system resize", err);
2065 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2066 		 ext4_blocks_count(es));
2067 	return err;
2068 }
2069