xref: /openbmc/linux/fs/ext4/resize.c (revision 61a9c11e)
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 (!(bh = sb_bread(sb, end - 1)))
131 		ext4_warning(sb, "Cannot read last block (%llu)",
132 			     end - 1);
133 	else if (outside(input->block_bitmap, start, end))
134 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
135 			     (unsigned long long)input->block_bitmap);
136 	else if (outside(input->inode_bitmap, start, end))
137 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
138 			     (unsigned long long)input->inode_bitmap);
139 	else if (outside(input->inode_table, start, end) ||
140 		 outside(itend - 1, start, end))
141 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
142 			     (unsigned long long)input->inode_table, itend - 1);
143 	else if (input->inode_bitmap == input->block_bitmap)
144 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
145 			     (unsigned long long)input->block_bitmap);
146 	else if (inside(input->block_bitmap, input->inode_table, itend))
147 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
148 			     "(%llu-%llu)",
149 			     (unsigned long long)input->block_bitmap,
150 			     (unsigned long long)input->inode_table, itend - 1);
151 	else if (inside(input->inode_bitmap, input->inode_table, itend))
152 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
153 			     "(%llu-%llu)",
154 			     (unsigned long long)input->inode_bitmap,
155 			     (unsigned long long)input->inode_table, itend - 1);
156 	else if (inside(input->block_bitmap, start, metaend))
157 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
158 			     (unsigned long long)input->block_bitmap,
159 			     start, metaend - 1);
160 	else if (inside(input->inode_bitmap, start, metaend))
161 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
162 			     (unsigned long long)input->inode_bitmap,
163 			     start, metaend - 1);
164 	else if (inside(input->inode_table, start, metaend) ||
165 		 inside(itend - 1, start, metaend))
166 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
167 			     "(%llu-%llu)",
168 			     (unsigned long long)input->inode_table,
169 			     itend - 1, start, metaend - 1);
170 	else
171 		err = 0;
172 	brelse(bh);
173 
174 	return err;
175 }
176 
177 /*
178  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
179  * group each time.
180  */
181 struct ext4_new_flex_group_data {
182 	struct ext4_new_group_data *groups;	/* new_group_data for groups
183 						   in the flex group */
184 	__u16 *bg_flags;			/* block group flags of groups
185 						   in @groups */
186 	ext4_group_t count;			/* number of groups in @groups
187 						 */
188 };
189 
190 /*
191  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
192  * @flexbg_size.
193  *
194  * Returns NULL on failure otherwise address of the allocated structure.
195  */
196 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
197 {
198 	struct ext4_new_flex_group_data *flex_gd;
199 
200 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
201 	if (flex_gd == NULL)
202 		goto out3;
203 
204 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
205 		goto out2;
206 	flex_gd->count = flexbg_size;
207 
208 	flex_gd->groups = kmalloc_array(flexbg_size,
209 					sizeof(struct ext4_new_group_data),
210 					GFP_NOFS);
211 	if (flex_gd->groups == NULL)
212 		goto out2;
213 
214 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
215 					  GFP_NOFS);
216 	if (flex_gd->bg_flags == NULL)
217 		goto out1;
218 
219 	return flex_gd;
220 
221 out1:
222 	kfree(flex_gd->groups);
223 out2:
224 	kfree(flex_gd);
225 out3:
226 	return NULL;
227 }
228 
229 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
230 {
231 	kfree(flex_gd->bg_flags);
232 	kfree(flex_gd->groups);
233 	kfree(flex_gd);
234 }
235 
236 /*
237  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
238  * and inode tables for a flex group.
239  *
240  * This function is used by 64bit-resize.  Note that this function allocates
241  * group tables from the 1st group of groups contained by @flexgd, which may
242  * be a partial of a flex group.
243  *
244  * @sb: super block of fs to which the groups belongs
245  *
246  * Returns 0 on a successful allocation of the metadata blocks in the
247  * block group.
248  */
249 static int ext4_alloc_group_tables(struct super_block *sb,
250 				struct ext4_new_flex_group_data *flex_gd,
251 				int flexbg_size)
252 {
253 	struct ext4_new_group_data *group_data = flex_gd->groups;
254 	ext4_fsblk_t start_blk;
255 	ext4_fsblk_t last_blk;
256 	ext4_group_t src_group;
257 	ext4_group_t bb_index = 0;
258 	ext4_group_t ib_index = 0;
259 	ext4_group_t it_index = 0;
260 	ext4_group_t group;
261 	ext4_group_t last_group;
262 	unsigned overhead;
263 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
264 	int i;
265 
266 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
267 
268 	src_group = group_data[0].group;
269 	last_group  = src_group + flex_gd->count - 1;
270 
271 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
272 	       (last_group & ~(flexbg_size - 1))));
273 next_group:
274 	group = group_data[0].group;
275 	if (src_group >= group_data[0].group + flex_gd->count)
276 		return -ENOSPC;
277 	start_blk = ext4_group_first_block_no(sb, src_group);
278 	last_blk = start_blk + group_data[src_group - group].blocks_count;
279 
280 	overhead = ext4_group_overhead_blocks(sb, src_group);
281 
282 	start_blk += overhead;
283 
284 	/* We collect contiguous blocks as much as possible. */
285 	src_group++;
286 	for (; src_group <= last_group; src_group++) {
287 		overhead = ext4_group_overhead_blocks(sb, src_group);
288 		if (overhead == 0)
289 			last_blk += group_data[src_group - group].blocks_count;
290 		else
291 			break;
292 	}
293 
294 	/* Allocate block bitmaps */
295 	for (; bb_index < flex_gd->count; bb_index++) {
296 		if (start_blk >= last_blk)
297 			goto next_group;
298 		group_data[bb_index].block_bitmap = start_blk++;
299 		group = ext4_get_group_number(sb, start_blk - 1);
300 		group -= group_data[0].group;
301 		group_data[group].mdata_blocks++;
302 		flex_gd->bg_flags[group] &= uninit_mask;
303 	}
304 
305 	/* Allocate inode bitmaps */
306 	for (; ib_index < flex_gd->count; ib_index++) {
307 		if (start_blk >= last_blk)
308 			goto next_group;
309 		group_data[ib_index].inode_bitmap = start_blk++;
310 		group = ext4_get_group_number(sb, start_blk - 1);
311 		group -= group_data[0].group;
312 		group_data[group].mdata_blocks++;
313 		flex_gd->bg_flags[group] &= uninit_mask;
314 	}
315 
316 	/* Allocate inode tables */
317 	for (; it_index < flex_gd->count; it_index++) {
318 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
319 		ext4_fsblk_t next_group_start;
320 
321 		if (start_blk + itb > last_blk)
322 			goto next_group;
323 		group_data[it_index].inode_table = start_blk;
324 		group = ext4_get_group_number(sb, start_blk);
325 		next_group_start = ext4_group_first_block_no(sb, group + 1);
326 		group -= group_data[0].group;
327 
328 		if (start_blk + itb > next_group_start) {
329 			flex_gd->bg_flags[group + 1] &= uninit_mask;
330 			overhead = start_blk + itb - next_group_start;
331 			group_data[group + 1].mdata_blocks += overhead;
332 			itb -= overhead;
333 		}
334 
335 		group_data[group].mdata_blocks += itb;
336 		flex_gd->bg_flags[group] &= uninit_mask;
337 		start_blk += EXT4_SB(sb)->s_itb_per_group;
338 	}
339 
340 	/* Update free clusters count to exclude metadata blocks */
341 	for (i = 0; i < flex_gd->count; i++) {
342 		group_data[i].free_clusters_count -=
343 				EXT4_NUM_B2C(EXT4_SB(sb),
344 					     group_data[i].mdata_blocks);
345 	}
346 
347 	if (test_opt(sb, DEBUG)) {
348 		int i;
349 		group = group_data[0].group;
350 
351 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
352 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
353 		       flexbg_size);
354 
355 		for (i = 0; i < flex_gd->count; i++) {
356 			ext4_debug(
357 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
358 			       ext4_bg_has_super(sb, group + i) ? "normal" :
359 			       "no-super", group + i,
360 			       group_data[i].blocks_count,
361 			       group_data[i].free_clusters_count,
362 			       group_data[i].mdata_blocks);
363 		}
364 	}
365 	return 0;
366 }
367 
368 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
369 				  ext4_fsblk_t blk)
370 {
371 	struct buffer_head *bh;
372 	int err;
373 
374 	bh = sb_getblk(sb, blk);
375 	if (unlikely(!bh))
376 		return ERR_PTR(-ENOMEM);
377 	BUFFER_TRACE(bh, "get_write_access");
378 	if ((err = ext4_journal_get_write_access(handle, bh))) {
379 		brelse(bh);
380 		bh = ERR_PTR(err);
381 	} else {
382 		memset(bh->b_data, 0, sb->s_blocksize);
383 		set_buffer_uptodate(bh);
384 	}
385 
386 	return bh;
387 }
388 
389 /*
390  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
391  * If that fails, restart the transaction & regain write access for the
392  * buffer head which is used for block_bitmap modifications.
393  */
394 static int extend_or_restart_transaction(handle_t *handle, int thresh)
395 {
396 	int err;
397 
398 	if (ext4_handle_has_enough_credits(handle, thresh))
399 		return 0;
400 
401 	err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
402 	if (err < 0)
403 		return err;
404 	if (err) {
405 		err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
406 		if (err)
407 			return err;
408 	}
409 
410 	return 0;
411 }
412 
413 /*
414  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
415  *
416  * Helper function for ext4_setup_new_group_blocks() which set .
417  *
418  * @sb: super block
419  * @handle: journal handle
420  * @flex_gd: flex group data
421  */
422 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
423 			struct ext4_new_flex_group_data *flex_gd,
424 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
425 {
426 	struct ext4_sb_info *sbi = EXT4_SB(sb);
427 	ext4_group_t count = last_cluster - first_cluster + 1;
428 	ext4_group_t count2;
429 
430 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
431 		   last_cluster);
432 	for (count2 = count; count > 0;
433 	     count -= count2, first_cluster += count2) {
434 		ext4_fsblk_t start;
435 		struct buffer_head *bh;
436 		ext4_group_t group;
437 		int err;
438 
439 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
440 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
441 		group -= flex_gd->groups[0].group;
442 
443 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
444 		if (count2 > count)
445 			count2 = count;
446 
447 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
448 			BUG_ON(flex_gd->count > 1);
449 			continue;
450 		}
451 
452 		err = extend_or_restart_transaction(handle, 1);
453 		if (err)
454 			return err;
455 
456 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
457 		if (unlikely(!bh))
458 			return -ENOMEM;
459 
460 		BUFFER_TRACE(bh, "get_write_access");
461 		err = ext4_journal_get_write_access(handle, bh);
462 		if (err) {
463 			brelse(bh);
464 			return err;
465 		}
466 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
467 			   first_cluster, first_cluster - start, count2);
468 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
469 
470 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
471 		brelse(bh);
472 		if (unlikely(err))
473 			return err;
474 	}
475 
476 	return 0;
477 }
478 
479 /*
480  * Set up the block and inode bitmaps, and the inode table for the new groups.
481  * This doesn't need to be part of the main transaction, since we are only
482  * changing blocks outside the actual filesystem.  We still do journaling to
483  * ensure the recovery is correct in case of a failure just after resize.
484  * If any part of this fails, we simply abort the resize.
485  *
486  * setup_new_flex_group_blocks handles a flex group as follow:
487  *  1. copy super block and GDT, and initialize group tables if necessary.
488  *     In this step, we only set bits in blocks bitmaps for blocks taken by
489  *     super block and GDT.
490  *  2. allocate group tables in block bitmaps, that is, set bits in block
491  *     bitmap for blocks taken by group tables.
492  */
493 static int setup_new_flex_group_blocks(struct super_block *sb,
494 				struct ext4_new_flex_group_data *flex_gd)
495 {
496 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
497 	ext4_fsblk_t start;
498 	ext4_fsblk_t block;
499 	struct ext4_sb_info *sbi = EXT4_SB(sb);
500 	struct ext4_super_block *es = sbi->s_es;
501 	struct ext4_new_group_data *group_data = flex_gd->groups;
502 	__u16 *bg_flags = flex_gd->bg_flags;
503 	handle_t *handle;
504 	ext4_group_t group, count;
505 	struct buffer_head *bh = NULL;
506 	int reserved_gdb, i, j, err = 0, err2;
507 	int meta_bg;
508 
509 	BUG_ON(!flex_gd->count || !group_data ||
510 	       group_data[0].group != sbi->s_groups_count);
511 
512 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
513 	meta_bg = ext4_has_feature_meta_bg(sb);
514 
515 	/* This transaction may be extended/restarted along the way */
516 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
517 	if (IS_ERR(handle))
518 		return PTR_ERR(handle);
519 
520 	group = group_data[0].group;
521 	for (i = 0; i < flex_gd->count; i++, group++) {
522 		unsigned long gdblocks;
523 		ext4_grpblk_t overhead;
524 
525 		gdblocks = ext4_bg_num_gdb(sb, group);
526 		start = ext4_group_first_block_no(sb, group);
527 
528 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
529 			goto handle_itb;
530 
531 		if (meta_bg == 1) {
532 			ext4_group_t first_group;
533 			first_group = ext4_meta_bg_first_group(sb, group);
534 			if (first_group != group + 1 &&
535 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
536 				goto handle_itb;
537 		}
538 
539 		block = start + ext4_bg_has_super(sb, group);
540 		/* Copy all of the GDT blocks into the backup in this group */
541 		for (j = 0; j < gdblocks; j++, block++) {
542 			struct buffer_head *gdb;
543 
544 			ext4_debug("update backup group %#04llx\n", block);
545 			err = extend_or_restart_transaction(handle, 1);
546 			if (err)
547 				goto out;
548 
549 			gdb = sb_getblk(sb, block);
550 			if (unlikely(!gdb)) {
551 				err = -ENOMEM;
552 				goto out;
553 			}
554 
555 			BUFFER_TRACE(gdb, "get_write_access");
556 			err = ext4_journal_get_write_access(handle, gdb);
557 			if (err) {
558 				brelse(gdb);
559 				goto out;
560 			}
561 			memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
562 			       gdb->b_size);
563 			set_buffer_uptodate(gdb);
564 
565 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
566 			if (unlikely(err)) {
567 				brelse(gdb);
568 				goto out;
569 			}
570 			brelse(gdb);
571 		}
572 
573 		/* Zero out all of the reserved backup group descriptor
574 		 * table blocks
575 		 */
576 		if (ext4_bg_has_super(sb, group)) {
577 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
578 					reserved_gdb, GFP_NOFS);
579 			if (err)
580 				goto out;
581 		}
582 
583 handle_itb:
584 		/* Initialize group tables of the grop @group */
585 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
586 			goto handle_bb;
587 
588 		/* Zero out all of the inode table blocks */
589 		block = group_data[i].inode_table;
590 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
591 			   block, sbi->s_itb_per_group);
592 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
593 				       GFP_NOFS);
594 		if (err)
595 			goto out;
596 
597 handle_bb:
598 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
599 			goto handle_ib;
600 
601 		/* Initialize block bitmap of the @group */
602 		block = group_data[i].block_bitmap;
603 		err = extend_or_restart_transaction(handle, 1);
604 		if (err)
605 			goto out;
606 
607 		bh = bclean(handle, sb, block);
608 		if (IS_ERR(bh)) {
609 			err = PTR_ERR(bh);
610 			goto out;
611 		}
612 		overhead = ext4_group_overhead_blocks(sb, group);
613 		if (overhead != 0) {
614 			ext4_debug("mark backup superblock %#04llx (+0)\n",
615 				   start);
616 			ext4_set_bits(bh->b_data, 0,
617 				      EXT4_NUM_B2C(sbi, overhead));
618 		}
619 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
620 				     sb->s_blocksize * 8, bh->b_data);
621 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
622 		brelse(bh);
623 		if (err)
624 			goto out;
625 
626 handle_ib:
627 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
628 			continue;
629 
630 		/* Initialize inode bitmap of the @group */
631 		block = group_data[i].inode_bitmap;
632 		err = extend_or_restart_transaction(handle, 1);
633 		if (err)
634 			goto out;
635 		/* Mark unused entries in inode bitmap used */
636 		bh = bclean(handle, sb, block);
637 		if (IS_ERR(bh)) {
638 			err = PTR_ERR(bh);
639 			goto out;
640 		}
641 
642 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
643 				     sb->s_blocksize * 8, bh->b_data);
644 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
645 		brelse(bh);
646 		if (err)
647 			goto out;
648 	}
649 
650 	/* Mark group tables in block bitmap */
651 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
652 		count = group_table_count[j];
653 		start = (&group_data[0].block_bitmap)[j];
654 		block = start;
655 		for (i = 1; i < flex_gd->count; i++) {
656 			block += group_table_count[j];
657 			if (block == (&group_data[i].block_bitmap)[j]) {
658 				count += group_table_count[j];
659 				continue;
660 			}
661 			err = set_flexbg_block_bitmap(sb, handle,
662 						      flex_gd,
663 						      EXT4_B2C(sbi, start),
664 						      EXT4_B2C(sbi,
665 							       start + count
666 							       - 1));
667 			if (err)
668 				goto out;
669 			count = group_table_count[j];
670 			start = (&group_data[i].block_bitmap)[j];
671 			block = start;
672 		}
673 
674 		if (count) {
675 			err = set_flexbg_block_bitmap(sb, handle,
676 						      flex_gd,
677 						      EXT4_B2C(sbi, start),
678 						      EXT4_B2C(sbi,
679 							       start + count
680 							       - 1));
681 			if (err)
682 				goto out;
683 		}
684 	}
685 
686 out:
687 	err2 = ext4_journal_stop(handle);
688 	if (err2 && !err)
689 		err = err2;
690 
691 	return err;
692 }
693 
694 /*
695  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
696  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
697  * calling this for the first time.  In a sparse filesystem it will be the
698  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
699  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
700  */
701 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
702 				  unsigned *five, unsigned *seven)
703 {
704 	unsigned *min = three;
705 	int mult = 3;
706 	unsigned ret;
707 
708 	if (!ext4_has_feature_sparse_super(sb)) {
709 		ret = *min;
710 		*min += 1;
711 		return ret;
712 	}
713 
714 	if (*five < *min) {
715 		min = five;
716 		mult = 5;
717 	}
718 	if (*seven < *min) {
719 		min = seven;
720 		mult = 7;
721 	}
722 
723 	ret = *min;
724 	*min *= mult;
725 
726 	return ret;
727 }
728 
729 /*
730  * Check that all of the backup GDT blocks are held in the primary GDT block.
731  * It is assumed that they are stored in group order.  Returns the number of
732  * groups in current filesystem that have BACKUPS, or -ve error code.
733  */
734 static int verify_reserved_gdb(struct super_block *sb,
735 			       ext4_group_t end,
736 			       struct buffer_head *primary)
737 {
738 	const ext4_fsblk_t blk = primary->b_blocknr;
739 	unsigned three = 1;
740 	unsigned five = 5;
741 	unsigned seven = 7;
742 	unsigned grp;
743 	__le32 *p = (__le32 *)primary->b_data;
744 	int gdbackups = 0;
745 
746 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
747 		if (le32_to_cpu(*p++) !=
748 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
749 			ext4_warning(sb, "reserved GDT %llu"
750 				     " missing grp %d (%llu)",
751 				     blk, grp,
752 				     grp *
753 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
754 				     blk);
755 			return -EINVAL;
756 		}
757 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
758 			return -EFBIG;
759 	}
760 
761 	return gdbackups;
762 }
763 
764 /*
765  * Called when we need to bring a reserved group descriptor table block into
766  * use from the resize inode.  The primary copy of the new GDT block currently
767  * is an indirect block (under the double indirect block in the resize inode).
768  * The new backup GDT blocks will be stored as leaf blocks in this indirect
769  * block, in group order.  Even though we know all the block numbers we need,
770  * we check to ensure that the resize inode has actually reserved these blocks.
771  *
772  * Don't need to update the block bitmaps because the blocks are still in use.
773  *
774  * We get all of the error cases out of the way, so that we are sure to not
775  * fail once we start modifying the data on disk, because JBD has no rollback.
776  */
777 static int add_new_gdb(handle_t *handle, struct inode *inode,
778 		       ext4_group_t group)
779 {
780 	struct super_block *sb = inode->i_sb;
781 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
782 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
783 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
784 	struct buffer_head **o_group_desc, **n_group_desc;
785 	struct buffer_head *dind;
786 	struct buffer_head *gdb_bh;
787 	int gdbackups;
788 	struct ext4_iloc iloc;
789 	__le32 *data;
790 	int err;
791 
792 	if (test_opt(sb, DEBUG))
793 		printk(KERN_DEBUG
794 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
795 		       gdb_num);
796 
797 	gdb_bh = sb_bread(sb, gdblock);
798 	if (!gdb_bh)
799 		return -EIO;
800 
801 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
802 	if (gdbackups < 0) {
803 		err = gdbackups;
804 		goto exit_bh;
805 	}
806 
807 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
808 	dind = sb_bread(sb, le32_to_cpu(*data));
809 	if (!dind) {
810 		err = -EIO;
811 		goto exit_bh;
812 	}
813 
814 	data = (__le32 *)dind->b_data;
815 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
816 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
817 			     group, gdblock);
818 		err = -EINVAL;
819 		goto exit_dind;
820 	}
821 
822 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
823 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
824 	if (unlikely(err))
825 		goto exit_dind;
826 
827 	BUFFER_TRACE(gdb_bh, "get_write_access");
828 	err = ext4_journal_get_write_access(handle, gdb_bh);
829 	if (unlikely(err))
830 		goto exit_dind;
831 
832 	BUFFER_TRACE(dind, "get_write_access");
833 	err = ext4_journal_get_write_access(handle, dind);
834 	if (unlikely(err))
835 		ext4_std_error(sb, err);
836 
837 	/* ext4_reserve_inode_write() gets a reference on the iloc */
838 	err = ext4_reserve_inode_write(handle, inode, &iloc);
839 	if (unlikely(err))
840 		goto exit_dind;
841 
842 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
843 				     sizeof(struct buffer_head *),
844 				     GFP_NOFS);
845 	if (!n_group_desc) {
846 		err = -ENOMEM;
847 		ext4_warning(sb, "not enough memory for %lu groups",
848 			     gdb_num + 1);
849 		goto exit_inode;
850 	}
851 
852 	/*
853 	 * Finally, we have all of the possible failures behind us...
854 	 *
855 	 * Remove new GDT block from inode double-indirect block and clear out
856 	 * the new GDT block for use (which also "frees" the backup GDT blocks
857 	 * from the reserved inode).  We don't need to change the bitmaps for
858 	 * these blocks, because they are marked as in-use from being in the
859 	 * reserved inode, and will become GDT blocks (primary and backup).
860 	 */
861 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
862 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
863 	if (unlikely(err)) {
864 		ext4_std_error(sb, err);
865 		goto exit_inode;
866 	}
867 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
868 			   (9 - EXT4_SB(sb)->s_cluster_bits);
869 	ext4_mark_iloc_dirty(handle, inode, &iloc);
870 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
871 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
872 	if (unlikely(err)) {
873 		ext4_std_error(sb, err);
874 		goto exit_inode;
875 	}
876 	brelse(dind);
877 
878 	o_group_desc = EXT4_SB(sb)->s_group_desc;
879 	memcpy(n_group_desc, o_group_desc,
880 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
881 	n_group_desc[gdb_num] = gdb_bh;
882 	EXT4_SB(sb)->s_group_desc = n_group_desc;
883 	EXT4_SB(sb)->s_gdb_count++;
884 	kvfree(o_group_desc);
885 
886 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
887 	err = ext4_handle_dirty_super(handle, sb);
888 	if (err)
889 		ext4_std_error(sb, err);
890 
891 	return err;
892 
893 exit_inode:
894 	kvfree(n_group_desc);
895 	brelse(iloc.bh);
896 exit_dind:
897 	brelse(dind);
898 exit_bh:
899 	brelse(gdb_bh);
900 
901 	ext4_debug("leaving with error %d\n", err);
902 	return err;
903 }
904 
905 /*
906  * add_new_gdb_meta_bg is the sister of add_new_gdb.
907  */
908 static int add_new_gdb_meta_bg(struct super_block *sb,
909 			       handle_t *handle, ext4_group_t group) {
910 	ext4_fsblk_t gdblock;
911 	struct buffer_head *gdb_bh;
912 	struct buffer_head **o_group_desc, **n_group_desc;
913 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
914 	int err;
915 
916 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
917 		   ext4_bg_has_super(sb, group);
918 	gdb_bh = sb_bread(sb, gdblock);
919 	if (!gdb_bh)
920 		return -EIO;
921 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
922 				     sizeof(struct buffer_head *),
923 				     GFP_NOFS);
924 	if (!n_group_desc) {
925 		brelse(gdb_bh);
926 		err = -ENOMEM;
927 		ext4_warning(sb, "not enough memory for %lu groups",
928 			     gdb_num + 1);
929 		return err;
930 	}
931 
932 	o_group_desc = EXT4_SB(sb)->s_group_desc;
933 	memcpy(n_group_desc, o_group_desc,
934 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
935 	n_group_desc[gdb_num] = gdb_bh;
936 	EXT4_SB(sb)->s_group_desc = n_group_desc;
937 	EXT4_SB(sb)->s_gdb_count++;
938 	kvfree(o_group_desc);
939 	BUFFER_TRACE(gdb_bh, "get_write_access");
940 	err = ext4_journal_get_write_access(handle, gdb_bh);
941 	return err;
942 }
943 
944 /*
945  * Called when we are adding a new group which has a backup copy of each of
946  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
947  * We need to add these reserved backup GDT blocks to the resize inode, so
948  * that they are kept for future resizing and not allocated to files.
949  *
950  * Each reserved backup GDT block will go into a different indirect block.
951  * The indirect blocks are actually the primary reserved GDT blocks,
952  * so we know in advance what their block numbers are.  We only get the
953  * double-indirect block to verify it is pointing to the primary reserved
954  * GDT blocks so we don't overwrite a data block by accident.  The reserved
955  * backup GDT blocks are stored in their reserved primary GDT block.
956  */
957 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
958 			      ext4_group_t group)
959 {
960 	struct super_block *sb = inode->i_sb;
961 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
962 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
963 	struct buffer_head **primary;
964 	struct buffer_head *dind;
965 	struct ext4_iloc iloc;
966 	ext4_fsblk_t blk;
967 	__le32 *data, *end;
968 	int gdbackups = 0;
969 	int res, i;
970 	int err;
971 
972 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
973 	if (!primary)
974 		return -ENOMEM;
975 
976 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
977 	dind = sb_bread(sb, le32_to_cpu(*data));
978 	if (!dind) {
979 		err = -EIO;
980 		goto exit_free;
981 	}
982 
983 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
984 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
985 					 EXT4_ADDR_PER_BLOCK(sb));
986 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
987 
988 	/* Get each reserved primary GDT block and verify it holds backups */
989 	for (res = 0; res < reserved_gdb; res++, blk++) {
990 		if (le32_to_cpu(*data) != blk) {
991 			ext4_warning(sb, "reserved block %llu"
992 				     " not at offset %ld",
993 				     blk,
994 				     (long)(data - (__le32 *)dind->b_data));
995 			err = -EINVAL;
996 			goto exit_bh;
997 		}
998 		primary[res] = sb_bread(sb, blk);
999 		if (!primary[res]) {
1000 			err = -EIO;
1001 			goto exit_bh;
1002 		}
1003 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1004 		if (gdbackups < 0) {
1005 			brelse(primary[res]);
1006 			err = gdbackups;
1007 			goto exit_bh;
1008 		}
1009 		if (++data >= end)
1010 			data = (__le32 *)dind->b_data;
1011 	}
1012 
1013 	for (i = 0; i < reserved_gdb; i++) {
1014 		BUFFER_TRACE(primary[i], "get_write_access");
1015 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1016 			goto exit_bh;
1017 	}
1018 
1019 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1020 		goto exit_bh;
1021 
1022 	/*
1023 	 * Finally we can add each of the reserved backup GDT blocks from
1024 	 * the new group to its reserved primary GDT block.
1025 	 */
1026 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1027 	for (i = 0; i < reserved_gdb; i++) {
1028 		int err2;
1029 		data = (__le32 *)primary[i]->b_data;
1030 		/* printk("reserving backup %lu[%u] = %lu\n",
1031 		       primary[i]->b_blocknr, gdbackups,
1032 		       blk + primary[i]->b_blocknr); */
1033 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1034 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1035 		if (!err)
1036 			err = err2;
1037 	}
1038 
1039 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1040 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1041 
1042 exit_bh:
1043 	while (--res >= 0)
1044 		brelse(primary[res]);
1045 	brelse(dind);
1046 
1047 exit_free:
1048 	kfree(primary);
1049 
1050 	return err;
1051 }
1052 
1053 /*
1054  * Update the backup copies of the ext4 metadata.  These don't need to be part
1055  * of the main resize transaction, because e2fsck will re-write them if there
1056  * is a problem (basically only OOM will cause a problem).  However, we
1057  * _should_ update the backups if possible, in case the primary gets trashed
1058  * for some reason and we need to run e2fsck from a backup superblock.  The
1059  * important part is that the new block and inode counts are in the backup
1060  * superblocks, and the location of the new group metadata in the GDT backups.
1061  *
1062  * We do not need take the s_resize_lock for this, because these
1063  * blocks are not otherwise touched by the filesystem code when it is
1064  * mounted.  We don't need to worry about last changing from
1065  * sbi->s_groups_count, because the worst that can happen is that we
1066  * do not copy the full number of backups at this time.  The resize
1067  * which changed s_groups_count will backup again.
1068  */
1069 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1070 			   int size, int meta_bg)
1071 {
1072 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1073 	ext4_group_t last;
1074 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1075 	unsigned three = 1;
1076 	unsigned five = 5;
1077 	unsigned seven = 7;
1078 	ext4_group_t group = 0;
1079 	int rest = sb->s_blocksize - size;
1080 	handle_t *handle;
1081 	int err = 0, err2;
1082 
1083 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1084 	if (IS_ERR(handle)) {
1085 		group = 1;
1086 		err = PTR_ERR(handle);
1087 		goto exit_err;
1088 	}
1089 
1090 	if (meta_bg == 0) {
1091 		group = ext4_list_backups(sb, &three, &five, &seven);
1092 		last = sbi->s_groups_count;
1093 	} else {
1094 		group = ext4_get_group_number(sb, blk_off) + 1;
1095 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1096 	}
1097 
1098 	while (group < sbi->s_groups_count) {
1099 		struct buffer_head *bh;
1100 		ext4_fsblk_t backup_block;
1101 
1102 		/* Out of journal space, and can't get more - abort - so sad */
1103 		if (ext4_handle_valid(handle) &&
1104 		    handle->h_buffer_credits == 0 &&
1105 		    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1106 		    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1107 			break;
1108 
1109 		if (meta_bg == 0)
1110 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1111 		else
1112 			backup_block = (ext4_group_first_block_no(sb, group) +
1113 					ext4_bg_has_super(sb, group));
1114 
1115 		bh = sb_getblk(sb, backup_block);
1116 		if (unlikely(!bh)) {
1117 			err = -ENOMEM;
1118 			break;
1119 		}
1120 		ext4_debug("update metadata backup %llu(+%llu)\n",
1121 			   backup_block, backup_block -
1122 			   ext4_group_first_block_no(sb, group));
1123 		BUFFER_TRACE(bh, "get_write_access");
1124 		if ((err = ext4_journal_get_write_access(handle, bh)))
1125 			break;
1126 		lock_buffer(bh);
1127 		memcpy(bh->b_data, data, size);
1128 		if (rest)
1129 			memset(bh->b_data + size, 0, rest);
1130 		set_buffer_uptodate(bh);
1131 		unlock_buffer(bh);
1132 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1133 		if (unlikely(err))
1134 			ext4_std_error(sb, err);
1135 		brelse(bh);
1136 
1137 		if (meta_bg == 0)
1138 			group = ext4_list_backups(sb, &three, &five, &seven);
1139 		else if (group == last)
1140 			break;
1141 		else
1142 			group = last;
1143 	}
1144 	if ((err2 = ext4_journal_stop(handle)) && !err)
1145 		err = err2;
1146 
1147 	/*
1148 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1149 	 * late to revert the resize, we shouldn't fail just because of
1150 	 * the backup copies (they are only needed in case of corruption).
1151 	 *
1152 	 * However, if we got here we have a journal problem too, so we
1153 	 * can't really start a transaction to mark the superblock.
1154 	 * Chicken out and just set the flag on the hope it will be written
1155 	 * to disk, and if not - we will simply wait until next fsck.
1156 	 */
1157 exit_err:
1158 	if (err) {
1159 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1160 			     "forcing fsck on next reboot", group, err);
1161 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1162 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1163 		mark_buffer_dirty(sbi->s_sbh);
1164 	}
1165 }
1166 
1167 /*
1168  * ext4_add_new_descs() adds @count group descriptor of groups
1169  * starting at @group
1170  *
1171  * @handle: journal handle
1172  * @sb: super block
1173  * @group: the group no. of the first group desc to be added
1174  * @resize_inode: the resize inode
1175  * @count: number of group descriptors to be added
1176  */
1177 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1178 			      ext4_group_t group, struct inode *resize_inode,
1179 			      ext4_group_t count)
1180 {
1181 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1182 	struct ext4_super_block *es = sbi->s_es;
1183 	struct buffer_head *gdb_bh;
1184 	int i, gdb_off, gdb_num, err = 0;
1185 	int meta_bg;
1186 
1187 	meta_bg = ext4_has_feature_meta_bg(sb);
1188 	for (i = 0; i < count; i++, group++) {
1189 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1190 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1191 
1192 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1193 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1194 
1195 		/*
1196 		 * We will only either add reserved group blocks to a backup group
1197 		 * or remove reserved blocks for the first group in a new group block.
1198 		 * Doing both would be mean more complex code, and sane people don't
1199 		 * use non-sparse filesystems anymore.  This is already checked above.
1200 		 */
1201 		if (gdb_off) {
1202 			gdb_bh = sbi->s_group_desc[gdb_num];
1203 			BUFFER_TRACE(gdb_bh, "get_write_access");
1204 			err = ext4_journal_get_write_access(handle, gdb_bh);
1205 
1206 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1207 				err = reserve_backup_gdb(handle, resize_inode, group);
1208 		} else if (meta_bg != 0) {
1209 			err = add_new_gdb_meta_bg(sb, handle, group);
1210 		} else {
1211 			err = add_new_gdb(handle, resize_inode, group);
1212 		}
1213 		if (err)
1214 			break;
1215 	}
1216 	return err;
1217 }
1218 
1219 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1220 {
1221 	struct buffer_head *bh = sb_getblk(sb, block);
1222 	if (unlikely(!bh))
1223 		return NULL;
1224 	if (!bh_uptodate_or_lock(bh)) {
1225 		if (bh_submit_read(bh) < 0) {
1226 			brelse(bh);
1227 			return NULL;
1228 		}
1229 	}
1230 
1231 	return bh;
1232 }
1233 
1234 static int ext4_set_bitmap_checksums(struct super_block *sb,
1235 				     ext4_group_t group,
1236 				     struct ext4_group_desc *gdp,
1237 				     struct ext4_new_group_data *group_data)
1238 {
1239 	struct buffer_head *bh;
1240 
1241 	if (!ext4_has_metadata_csum(sb))
1242 		return 0;
1243 
1244 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1245 	if (!bh)
1246 		return -EIO;
1247 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1248 				   EXT4_INODES_PER_GROUP(sb) / 8);
1249 	brelse(bh);
1250 
1251 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1252 	if (!bh)
1253 		return -EIO;
1254 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1255 	brelse(bh);
1256 
1257 	return 0;
1258 }
1259 
1260 /*
1261  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1262  */
1263 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1264 				struct ext4_new_flex_group_data *flex_gd)
1265 {
1266 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1267 	struct ext4_group_desc		*gdp;
1268 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1269 	struct buffer_head		*gdb_bh;
1270 	ext4_group_t			group;
1271 	__u16				*bg_flags = flex_gd->bg_flags;
1272 	int				i, gdb_off, gdb_num, err = 0;
1273 
1274 
1275 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1276 		group = group_data->group;
1277 
1278 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1279 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1280 
1281 		/*
1282 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1283 		 */
1284 		gdb_bh = sbi->s_group_desc[gdb_num];
1285 		/* Update group descriptor block for new group */
1286 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1287 						 gdb_off * EXT4_DESC_SIZE(sb));
1288 
1289 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1290 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1291 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1292 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1293 		if (err) {
1294 			ext4_std_error(sb, err);
1295 			break;
1296 		}
1297 
1298 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1299 		ext4_free_group_clusters_set(sb, gdp,
1300 					     group_data->free_clusters_count);
1301 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1302 		if (ext4_has_group_desc_csum(sb))
1303 			ext4_itable_unused_set(sb, gdp,
1304 					       EXT4_INODES_PER_GROUP(sb));
1305 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1306 		ext4_group_desc_csum_set(sb, group, gdp);
1307 
1308 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1309 		if (unlikely(err)) {
1310 			ext4_std_error(sb, err);
1311 			break;
1312 		}
1313 
1314 		/*
1315 		 * We can allocate memory for mb_alloc based on the new group
1316 		 * descriptor
1317 		 */
1318 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1319 		if (err)
1320 			break;
1321 	}
1322 	return err;
1323 }
1324 
1325 /*
1326  * ext4_update_super() updates the super block so that the newly added
1327  * groups can be seen by the filesystem.
1328  *
1329  * @sb: super block
1330  * @flex_gd: new added groups
1331  */
1332 static void ext4_update_super(struct super_block *sb,
1333 			     struct ext4_new_flex_group_data *flex_gd)
1334 {
1335 	ext4_fsblk_t blocks_count = 0;
1336 	ext4_fsblk_t free_blocks = 0;
1337 	ext4_fsblk_t reserved_blocks = 0;
1338 	struct ext4_new_group_data *group_data = flex_gd->groups;
1339 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1340 	struct ext4_super_block *es = sbi->s_es;
1341 	int i;
1342 
1343 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1344 	/*
1345 	 * Make the new blocks and inodes valid next.  We do this before
1346 	 * increasing the group count so that once the group is enabled,
1347 	 * all of its blocks and inodes are already valid.
1348 	 *
1349 	 * We always allocate group-by-group, then block-by-block or
1350 	 * inode-by-inode within a group, so enabling these
1351 	 * blocks/inodes before the group is live won't actually let us
1352 	 * allocate the new space yet.
1353 	 */
1354 	for (i = 0; i < flex_gd->count; i++) {
1355 		blocks_count += group_data[i].blocks_count;
1356 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1357 	}
1358 
1359 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1360 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1361 	reserved_blocks *= blocks_count;
1362 	do_div(reserved_blocks, 100);
1363 
1364 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1365 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1366 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1367 		     flex_gd->count);
1368 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1369 		     flex_gd->count);
1370 
1371 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1372 	/*
1373 	 * We need to protect s_groups_count against other CPUs seeing
1374 	 * inconsistent state in the superblock.
1375 	 *
1376 	 * The precise rules we use are:
1377 	 *
1378 	 * * Writers must perform a smp_wmb() after updating all
1379 	 *   dependent data and before modifying the groups count
1380 	 *
1381 	 * * Readers must perform an smp_rmb() after reading the groups
1382 	 *   count and before reading any dependent data.
1383 	 *
1384 	 * NB. These rules can be relaxed when checking the group count
1385 	 * while freeing data, as we can only allocate from a block
1386 	 * group after serialising against the group count, and we can
1387 	 * only then free after serialising in turn against that
1388 	 * allocation.
1389 	 */
1390 	smp_wmb();
1391 
1392 	/* Update the global fs size fields */
1393 	sbi->s_groups_count += flex_gd->count;
1394 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1395 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1396 
1397 	/* Update the reserved block counts only once the new group is
1398 	 * active. */
1399 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1400 				reserved_blocks);
1401 
1402 	/* Update the free space counts */
1403 	percpu_counter_add(&sbi->s_freeclusters_counter,
1404 			   EXT4_NUM_B2C(sbi, free_blocks));
1405 	percpu_counter_add(&sbi->s_freeinodes_counter,
1406 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1407 
1408 	ext4_debug("free blocks count %llu",
1409 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1410 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1411 		ext4_group_t flex_group;
1412 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1413 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1414 			     &sbi->s_flex_groups[flex_group].free_clusters);
1415 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1416 			   &sbi->s_flex_groups[flex_group].free_inodes);
1417 	}
1418 
1419 	/*
1420 	 * Update the fs overhead information
1421 	 */
1422 	ext4_calculate_overhead(sb);
1423 
1424 	if (test_opt(sb, DEBUG))
1425 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1426 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1427 		       blocks_count, free_blocks, reserved_blocks);
1428 }
1429 
1430 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1431  * _before_ we start modifying the filesystem, because we cannot abort the
1432  * transaction and not have it write the data to disk.
1433  */
1434 static int ext4_flex_group_add(struct super_block *sb,
1435 			       struct inode *resize_inode,
1436 			       struct ext4_new_flex_group_data *flex_gd)
1437 {
1438 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1439 	struct ext4_super_block *es = sbi->s_es;
1440 	ext4_fsblk_t o_blocks_count;
1441 	ext4_grpblk_t last;
1442 	ext4_group_t group;
1443 	handle_t *handle;
1444 	unsigned reserved_gdb;
1445 	int err = 0, err2 = 0, credit;
1446 
1447 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1448 
1449 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1450 	o_blocks_count = ext4_blocks_count(es);
1451 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1452 	BUG_ON(last);
1453 
1454 	err = setup_new_flex_group_blocks(sb, flex_gd);
1455 	if (err)
1456 		goto exit;
1457 	/*
1458 	 * We will always be modifying at least the superblock and  GDT
1459 	 * blocks.  If we are adding a group past the last current GDT block,
1460 	 * we will also modify the inode and the dindirect block.  If we
1461 	 * are adding a group with superblock/GDT backups  we will also
1462 	 * modify each of the reserved GDT dindirect blocks.
1463 	 */
1464 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1465 	/* GDT blocks */
1466 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1467 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1468 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1469 	if (IS_ERR(handle)) {
1470 		err = PTR_ERR(handle);
1471 		goto exit;
1472 	}
1473 
1474 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1475 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1476 	if (err)
1477 		goto exit_journal;
1478 
1479 	group = flex_gd->groups[0].group;
1480 	BUG_ON(group != sbi->s_groups_count);
1481 	err = ext4_add_new_descs(handle, sb, group,
1482 				resize_inode, flex_gd->count);
1483 	if (err)
1484 		goto exit_journal;
1485 
1486 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1487 	if (err)
1488 		goto exit_journal;
1489 
1490 	ext4_update_super(sb, flex_gd);
1491 
1492 	err = ext4_handle_dirty_super(handle, sb);
1493 
1494 exit_journal:
1495 	err2 = ext4_journal_stop(handle);
1496 	if (!err)
1497 		err = err2;
1498 
1499 	if (!err) {
1500 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1501 		int gdb_num_end = ((group + flex_gd->count - 1) /
1502 				   EXT4_DESC_PER_BLOCK(sb));
1503 		int meta_bg = ext4_has_feature_meta_bg(sb);
1504 		sector_t old_gdb = 0;
1505 
1506 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1507 			       sizeof(struct ext4_super_block), 0);
1508 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1509 			struct buffer_head *gdb_bh;
1510 
1511 			gdb_bh = sbi->s_group_desc[gdb_num];
1512 			if (old_gdb == gdb_bh->b_blocknr)
1513 				continue;
1514 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1515 				       gdb_bh->b_size, meta_bg);
1516 			old_gdb = gdb_bh->b_blocknr;
1517 		}
1518 	}
1519 exit:
1520 	return err;
1521 }
1522 
1523 static int ext4_setup_next_flex_gd(struct super_block *sb,
1524 				    struct ext4_new_flex_group_data *flex_gd,
1525 				    ext4_fsblk_t n_blocks_count,
1526 				    unsigned long flexbg_size)
1527 {
1528 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1529 	struct ext4_super_block *es = sbi->s_es;
1530 	struct ext4_new_group_data *group_data = flex_gd->groups;
1531 	ext4_fsblk_t o_blocks_count;
1532 	ext4_group_t n_group;
1533 	ext4_group_t group;
1534 	ext4_group_t last_group;
1535 	ext4_grpblk_t last;
1536 	ext4_grpblk_t clusters_per_group;
1537 	unsigned long i;
1538 
1539 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1540 
1541 	o_blocks_count = ext4_blocks_count(es);
1542 
1543 	if (o_blocks_count == n_blocks_count)
1544 		return 0;
1545 
1546 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1547 	BUG_ON(last);
1548 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1549 
1550 	last_group = group | (flexbg_size - 1);
1551 	if (last_group > n_group)
1552 		last_group = n_group;
1553 
1554 	flex_gd->count = last_group - group + 1;
1555 
1556 	for (i = 0; i < flex_gd->count; i++) {
1557 		int overhead;
1558 
1559 		group_data[i].group = group + i;
1560 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1561 		overhead = ext4_group_overhead_blocks(sb, group + i);
1562 		group_data[i].mdata_blocks = overhead;
1563 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1564 		if (ext4_has_group_desc_csum(sb)) {
1565 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1566 					       EXT4_BG_INODE_UNINIT;
1567 			if (!test_opt(sb, INIT_INODE_TABLE))
1568 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1569 		} else
1570 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1571 	}
1572 
1573 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1574 		/* We need to initialize block bitmap of last group. */
1575 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1576 
1577 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1578 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1579 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1580 						       last - 1;
1581 	}
1582 
1583 	return 1;
1584 }
1585 
1586 /* Add group descriptor data to an existing or new group descriptor block.
1587  * Ensure we handle all possible error conditions _before_ we start modifying
1588  * the filesystem, because we cannot abort the transaction and not have it
1589  * write the data to disk.
1590  *
1591  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1592  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1593  *
1594  * We only need to hold the superblock lock while we are actually adding
1595  * in the new group's counts to the superblock.  Prior to that we have
1596  * not really "added" the group at all.  We re-check that we are still
1597  * adding in the last group in case things have changed since verifying.
1598  */
1599 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1600 {
1601 	struct ext4_new_flex_group_data flex_gd;
1602 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1603 	struct ext4_super_block *es = sbi->s_es;
1604 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1605 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1606 	struct inode *inode = NULL;
1607 	int gdb_off;
1608 	int err;
1609 	__u16 bg_flags = 0;
1610 
1611 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1612 
1613 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1614 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1615 		return -EPERM;
1616 	}
1617 
1618 	if (ext4_blocks_count(es) + input->blocks_count <
1619 	    ext4_blocks_count(es)) {
1620 		ext4_warning(sb, "blocks_count overflow");
1621 		return -EINVAL;
1622 	}
1623 
1624 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1625 	    le32_to_cpu(es->s_inodes_count)) {
1626 		ext4_warning(sb, "inodes_count overflow");
1627 		return -EINVAL;
1628 	}
1629 
1630 	if (reserved_gdb || gdb_off == 0) {
1631 		if (ext4_has_feature_resize_inode(sb) ||
1632 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1633 			ext4_warning(sb,
1634 				     "No reserved GDT blocks, can't resize");
1635 			return -EPERM;
1636 		}
1637 		inode = ext4_iget(sb, EXT4_RESIZE_INO);
1638 		if (IS_ERR(inode)) {
1639 			ext4_warning(sb, "Error opening resize inode");
1640 			return PTR_ERR(inode);
1641 		}
1642 	}
1643 
1644 
1645 	err = verify_group_input(sb, input);
1646 	if (err)
1647 		goto out;
1648 
1649 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1650 	if (err)
1651 		goto out;
1652 
1653 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1654 	if (err)
1655 		goto out;
1656 
1657 	flex_gd.count = 1;
1658 	flex_gd.groups = input;
1659 	flex_gd.bg_flags = &bg_flags;
1660 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1661 out:
1662 	iput(inode);
1663 	return err;
1664 } /* ext4_group_add */
1665 
1666 /*
1667  * extend a group without checking assuming that checking has been done.
1668  */
1669 static int ext4_group_extend_no_check(struct super_block *sb,
1670 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1671 {
1672 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1673 	handle_t *handle;
1674 	int err = 0, err2;
1675 
1676 	/* We will update the superblock, one block bitmap, and
1677 	 * one group descriptor via ext4_group_add_blocks().
1678 	 */
1679 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1680 	if (IS_ERR(handle)) {
1681 		err = PTR_ERR(handle);
1682 		ext4_warning(sb, "error %d on journal start", err);
1683 		return err;
1684 	}
1685 
1686 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1687 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1688 	if (err) {
1689 		ext4_warning(sb, "error %d on journal write access", err);
1690 		goto errout;
1691 	}
1692 
1693 	ext4_blocks_count_set(es, o_blocks_count + add);
1694 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1695 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1696 		   o_blocks_count + add);
1697 	/* We add the blocks to the bitmap and set the group need init bit */
1698 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1699 	if (err)
1700 		goto errout;
1701 	ext4_handle_dirty_super(handle, sb);
1702 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1703 		   o_blocks_count + add);
1704 errout:
1705 	err2 = ext4_journal_stop(handle);
1706 	if (err2 && !err)
1707 		err = err2;
1708 
1709 	if (!err) {
1710 		if (test_opt(sb, DEBUG))
1711 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1712 			       "blocks\n", ext4_blocks_count(es));
1713 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1714 			       (char *)es, sizeof(struct ext4_super_block), 0);
1715 	}
1716 	return err;
1717 }
1718 
1719 /*
1720  * Extend the filesystem to the new number of blocks specified.  This entry
1721  * point is only used to extend the current filesystem to the end of the last
1722  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1723  * for emergencies (because it has no dependencies on reserved blocks).
1724  *
1725  * If we _really_ wanted, we could use default values to call ext4_group_add()
1726  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1727  * GDT blocks are reserved to grow to the desired size.
1728  */
1729 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1730 		      ext4_fsblk_t n_blocks_count)
1731 {
1732 	ext4_fsblk_t o_blocks_count;
1733 	ext4_grpblk_t last;
1734 	ext4_grpblk_t add;
1735 	struct buffer_head *bh;
1736 	int err;
1737 	ext4_group_t group;
1738 
1739 	o_blocks_count = ext4_blocks_count(es);
1740 
1741 	if (test_opt(sb, DEBUG))
1742 		ext4_msg(sb, KERN_DEBUG,
1743 			 "extending last group from %llu to %llu blocks",
1744 			 o_blocks_count, n_blocks_count);
1745 
1746 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1747 		return 0;
1748 
1749 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1750 		ext4_msg(sb, KERN_ERR,
1751 			 "filesystem too large to resize to %llu blocks safely",
1752 			 n_blocks_count);
1753 		if (sizeof(sector_t) < 8)
1754 			ext4_warning(sb, "CONFIG_LBDAF not enabled");
1755 		return -EINVAL;
1756 	}
1757 
1758 	if (n_blocks_count < o_blocks_count) {
1759 		ext4_warning(sb, "can't shrink FS - resize aborted");
1760 		return -EINVAL;
1761 	}
1762 
1763 	/* Handle the remaining blocks in the last group only. */
1764 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1765 
1766 	if (last == 0) {
1767 		ext4_warning(sb, "need to use ext2online to resize further");
1768 		return -EPERM;
1769 	}
1770 
1771 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1772 
1773 	if (o_blocks_count + add < o_blocks_count) {
1774 		ext4_warning(sb, "blocks_count overflow");
1775 		return -EINVAL;
1776 	}
1777 
1778 	if (o_blocks_count + add > n_blocks_count)
1779 		add = n_blocks_count - o_blocks_count;
1780 
1781 	if (o_blocks_count + add < n_blocks_count)
1782 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1783 			     o_blocks_count + add, add);
1784 
1785 	/* See if the device is actually as big as what was requested */
1786 	bh = sb_bread(sb, o_blocks_count + add - 1);
1787 	if (!bh) {
1788 		ext4_warning(sb, "can't read last block, resize aborted");
1789 		return -ENOSPC;
1790 	}
1791 	brelse(bh);
1792 
1793 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1794 	return err;
1795 } /* ext4_group_extend */
1796 
1797 
1798 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1799 {
1800 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1801 }
1802 
1803 /*
1804  * Release the resize inode and drop the resize_inode feature if there
1805  * are no more reserved gdt blocks, and then convert the file system
1806  * to enable meta_bg
1807  */
1808 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1809 {
1810 	handle_t *handle;
1811 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1812 	struct ext4_super_block *es = sbi->s_es;
1813 	struct ext4_inode_info *ei = EXT4_I(inode);
1814 	ext4_fsblk_t nr;
1815 	int i, ret, err = 0;
1816 	int credits = 1;
1817 
1818 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1819 	if (inode) {
1820 		if (es->s_reserved_gdt_blocks) {
1821 			ext4_error(sb, "Unexpected non-zero "
1822 				   "s_reserved_gdt_blocks");
1823 			return -EPERM;
1824 		}
1825 
1826 		/* Do a quick sanity check of the resize inode */
1827 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1828 					     (9 - sbi->s_cluster_bits)))
1829 			goto invalid_resize_inode;
1830 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1831 			if (i == EXT4_DIND_BLOCK) {
1832 				if (ei->i_data[i])
1833 					continue;
1834 				else
1835 					goto invalid_resize_inode;
1836 			}
1837 			if (ei->i_data[i])
1838 				goto invalid_resize_inode;
1839 		}
1840 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1841 	}
1842 
1843 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1844 	if (IS_ERR(handle))
1845 		return PTR_ERR(handle);
1846 
1847 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1848 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1849 	if (err)
1850 		goto errout;
1851 
1852 	ext4_clear_feature_resize_inode(sb);
1853 	ext4_set_feature_meta_bg(sb);
1854 	sbi->s_es->s_first_meta_bg =
1855 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1856 
1857 	err = ext4_handle_dirty_super(handle, sb);
1858 	if (err) {
1859 		ext4_std_error(sb, err);
1860 		goto errout;
1861 	}
1862 
1863 	if (inode) {
1864 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1865 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1866 				 EXT4_FREE_BLOCKS_METADATA |
1867 				 EXT4_FREE_BLOCKS_FORGET);
1868 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1869 		inode->i_blocks = 0;
1870 
1871 		err = ext4_mark_inode_dirty(handle, inode);
1872 		if (err)
1873 			ext4_std_error(sb, err);
1874 	}
1875 
1876 errout:
1877 	ret = ext4_journal_stop(handle);
1878 	if (!err)
1879 		err = ret;
1880 	return ret;
1881 
1882 invalid_resize_inode:
1883 	ext4_error(sb, "corrupted/inconsistent resize inode");
1884 	return -EINVAL;
1885 }
1886 
1887 /*
1888  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1889  *
1890  * @sb: super block of the fs to be resized
1891  * @n_blocks_count: the number of blocks resides in the resized fs
1892  */
1893 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1894 {
1895 	struct ext4_new_flex_group_data *flex_gd = NULL;
1896 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1897 	struct ext4_super_block *es = sbi->s_es;
1898 	struct buffer_head *bh;
1899 	struct inode *resize_inode = NULL;
1900 	ext4_grpblk_t add, offset;
1901 	unsigned long n_desc_blocks;
1902 	unsigned long o_desc_blocks;
1903 	ext4_group_t o_group;
1904 	ext4_group_t n_group;
1905 	ext4_fsblk_t o_blocks_count;
1906 	ext4_fsblk_t n_blocks_count_retry = 0;
1907 	unsigned long last_update_time = 0;
1908 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1909 	int meta_bg;
1910 
1911 	/* See if the device is actually as big as what was requested */
1912 	bh = sb_bread(sb, n_blocks_count - 1);
1913 	if (!bh) {
1914 		ext4_warning(sb, "can't read last block, resize aborted");
1915 		return -ENOSPC;
1916 	}
1917 	brelse(bh);
1918 
1919 retry:
1920 	o_blocks_count = ext4_blocks_count(es);
1921 
1922 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1923 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1924 
1925 	if (n_blocks_count < o_blocks_count) {
1926 		/* On-line shrinking not supported */
1927 		ext4_warning(sb, "can't shrink FS - resize aborted");
1928 		return -EINVAL;
1929 	}
1930 
1931 	if (n_blocks_count == o_blocks_count)
1932 		/* Nothing need to do */
1933 		return 0;
1934 
1935 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1936 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1937 		ext4_warning(sb, "resize would cause inodes_count overflow");
1938 		return -EINVAL;
1939 	}
1940 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1941 
1942 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1943 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1944 
1945 	meta_bg = ext4_has_feature_meta_bg(sb);
1946 
1947 	if (ext4_has_feature_resize_inode(sb)) {
1948 		if (meta_bg) {
1949 			ext4_error(sb, "resize_inode and meta_bg enabled "
1950 				   "simultaneously");
1951 			return -EINVAL;
1952 		}
1953 		if (n_desc_blocks > o_desc_blocks +
1954 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1955 			n_blocks_count_retry = n_blocks_count;
1956 			n_desc_blocks = o_desc_blocks +
1957 				le16_to_cpu(es->s_reserved_gdt_blocks);
1958 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1959 			n_blocks_count = (ext4_fsblk_t)n_group *
1960 				EXT4_BLOCKS_PER_GROUP(sb);
1961 			n_group--; /* set to last group number */
1962 		}
1963 
1964 		if (!resize_inode)
1965 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1966 		if (IS_ERR(resize_inode)) {
1967 			ext4_warning(sb, "Error opening resize inode");
1968 			return PTR_ERR(resize_inode);
1969 		}
1970 	}
1971 
1972 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1973 		err = ext4_convert_meta_bg(sb, resize_inode);
1974 		if (err)
1975 			goto out;
1976 		if (resize_inode) {
1977 			iput(resize_inode);
1978 			resize_inode = NULL;
1979 		}
1980 		if (n_blocks_count_retry) {
1981 			n_blocks_count = n_blocks_count_retry;
1982 			n_blocks_count_retry = 0;
1983 			goto retry;
1984 		}
1985 	}
1986 
1987 	/*
1988 	 * Make sure the last group has enough space so that it's
1989 	 * guaranteed to have enough space for all metadata blocks
1990 	 * that it might need to hold.  (We might not need to store
1991 	 * the inode table blocks in the last block group, but there
1992 	 * will be cases where this might be needed.)
1993 	 */
1994 	if ((ext4_group_first_block_no(sb, n_group) +
1995 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
1996 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
1997 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
1998 		n_group--;
1999 		n_blocks_count_retry = 0;
2000 		if (resize_inode) {
2001 			iput(resize_inode);
2002 			resize_inode = NULL;
2003 		}
2004 		goto retry;
2005 	}
2006 
2007 	/* extend the last group */
2008 	if (n_group == o_group)
2009 		add = n_blocks_count - o_blocks_count;
2010 	else
2011 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2012 	if (add > 0) {
2013 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2014 		if (err)
2015 			goto out;
2016 	}
2017 
2018 	if (ext4_blocks_count(es) == n_blocks_count)
2019 		goto out;
2020 
2021 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2022 	if (err)
2023 		return err;
2024 
2025 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2026 	if (err)
2027 		goto out;
2028 
2029 	flex_gd = alloc_flex_gd(flexbg_size);
2030 	if (flex_gd == NULL) {
2031 		err = -ENOMEM;
2032 		goto out;
2033 	}
2034 
2035 	/* Add flex groups. Note that a regular group is a
2036 	 * flex group with 1 group.
2037 	 */
2038 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2039 					      flexbg_size)) {
2040 		if (jiffies - last_update_time > HZ * 10) {
2041 			if (last_update_time)
2042 				ext4_msg(sb, KERN_INFO,
2043 					 "resized to %llu blocks",
2044 					 ext4_blocks_count(es));
2045 			last_update_time = jiffies;
2046 		}
2047 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2048 			break;
2049 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2050 		if (unlikely(err))
2051 			break;
2052 	}
2053 
2054 	if (!err && n_blocks_count_retry) {
2055 		n_blocks_count = n_blocks_count_retry;
2056 		n_blocks_count_retry = 0;
2057 		free_flex_gd(flex_gd);
2058 		flex_gd = NULL;
2059 		goto retry;
2060 	}
2061 
2062 out:
2063 	if (flex_gd)
2064 		free_flex_gd(flex_gd);
2065 	if (resize_inode != NULL)
2066 		iput(resize_inode);
2067 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2068 	return err;
2069 }
2070