xref: /openbmc/linux/fs/ext4/resize.c (revision 63dc02bd)
1 /*
2  *  linux/fs/ext4/resize.c
3  *
4  * Support for resizing an ext4 filesystem while it is mounted.
5  *
6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7  *
8  * This could probably be made into a module, because it is not often in use.
9  */
10 
11 
12 #define EXT4FS_DEBUG
13 
14 #include <linux/errno.h>
15 #include <linux/slab.h>
16 
17 #include "ext4_jbd2.h"
18 
19 int ext4_resize_begin(struct super_block *sb)
20 {
21 	int ret = 0;
22 
23 	if (!capable(CAP_SYS_RESOURCE))
24 		return -EPERM;
25 
26 	/*
27 	 * We are not allowed to do online-resizing on a filesystem mounted
28 	 * with error, because it can destroy the filesystem easily.
29 	 */
30 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 		ext4_warning(sb, "There are errors in the filesystem, "
32 			     "so online resizing is not allowed\n");
33 		return -EPERM;
34 	}
35 
36 	if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 		ret = -EBUSY;
38 
39 	return ret;
40 }
41 
42 void ext4_resize_end(struct super_block *sb)
43 {
44 	clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 	smp_mb__after_clear_bit();
46 }
47 
48 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
49 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
50 
51 static int verify_group_input(struct super_block *sb,
52 			      struct ext4_new_group_data *input)
53 {
54 	struct ext4_sb_info *sbi = EXT4_SB(sb);
55 	struct ext4_super_block *es = sbi->s_es;
56 	ext4_fsblk_t start = ext4_blocks_count(es);
57 	ext4_fsblk_t end = start + input->blocks_count;
58 	ext4_group_t group = input->group;
59 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 	unsigned overhead = ext4_bg_has_super(sb, group) ?
61 		(1 + ext4_bg_num_gdb(sb, group) +
62 		 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
63 	ext4_fsblk_t metaend = start + overhead;
64 	struct buffer_head *bh = NULL;
65 	ext4_grpblk_t free_blocks_count, offset;
66 	int err = -EINVAL;
67 
68 	input->free_blocks_count = free_blocks_count =
69 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
70 
71 	if (test_opt(sb, DEBUG))
72 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
73 		       "(%d free, %u reserved)\n",
74 		       ext4_bg_has_super(sb, input->group) ? "normal" :
75 		       "no-super", input->group, input->blocks_count,
76 		       free_blocks_count, input->reserved_blocks);
77 
78 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
79 	if (group != sbi->s_groups_count)
80 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
81 			     input->group, sbi->s_groups_count);
82 	else if (offset != 0)
83 			ext4_warning(sb, "Last group not full");
84 	else if (input->reserved_blocks > input->blocks_count / 5)
85 		ext4_warning(sb, "Reserved blocks too high (%u)",
86 			     input->reserved_blocks);
87 	else if (free_blocks_count < 0)
88 		ext4_warning(sb, "Bad blocks count %u",
89 			     input->blocks_count);
90 	else if (!(bh = sb_bread(sb, end - 1)))
91 		ext4_warning(sb, "Cannot read last block (%llu)",
92 			     end - 1);
93 	else if (outside(input->block_bitmap, start, end))
94 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
95 			     (unsigned long long)input->block_bitmap);
96 	else if (outside(input->inode_bitmap, start, end))
97 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
98 			     (unsigned long long)input->inode_bitmap);
99 	else if (outside(input->inode_table, start, end) ||
100 		 outside(itend - 1, start, end))
101 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
102 			     (unsigned long long)input->inode_table, itend - 1);
103 	else if (input->inode_bitmap == input->block_bitmap)
104 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
105 			     (unsigned long long)input->block_bitmap);
106 	else if (inside(input->block_bitmap, input->inode_table, itend))
107 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
108 			     "(%llu-%llu)",
109 			     (unsigned long long)input->block_bitmap,
110 			     (unsigned long long)input->inode_table, itend - 1);
111 	else if (inside(input->inode_bitmap, input->inode_table, itend))
112 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
113 			     "(%llu-%llu)",
114 			     (unsigned long long)input->inode_bitmap,
115 			     (unsigned long long)input->inode_table, itend - 1);
116 	else if (inside(input->block_bitmap, start, metaend))
117 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 			     (unsigned long long)input->block_bitmap,
119 			     start, metaend - 1);
120 	else if (inside(input->inode_bitmap, start, metaend))
121 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 			     (unsigned long long)input->inode_bitmap,
123 			     start, metaend - 1);
124 	else if (inside(input->inode_table, start, metaend) ||
125 		 inside(itend - 1, start, metaend))
126 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
127 			     "(%llu-%llu)",
128 			     (unsigned long long)input->inode_table,
129 			     itend - 1, start, metaend - 1);
130 	else
131 		err = 0;
132 	brelse(bh);
133 
134 	return err;
135 }
136 
137 /*
138  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
139  * group each time.
140  */
141 struct ext4_new_flex_group_data {
142 	struct ext4_new_group_data *groups;	/* new_group_data for groups
143 						   in the flex group */
144 	__u16 *bg_flags;			/* block group flags of groups
145 						   in @groups */
146 	ext4_group_t count;			/* number of groups in @groups
147 						 */
148 };
149 
150 /*
151  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
152  * @flexbg_size.
153  *
154  * Returns NULL on failure otherwise address of the allocated structure.
155  */
156 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
157 {
158 	struct ext4_new_flex_group_data *flex_gd;
159 
160 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
161 	if (flex_gd == NULL)
162 		goto out3;
163 
164 	flex_gd->count = flexbg_size;
165 
166 	flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
167 				  flexbg_size, GFP_NOFS);
168 	if (flex_gd->groups == NULL)
169 		goto out2;
170 
171 	flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
172 	if (flex_gd->bg_flags == NULL)
173 		goto out1;
174 
175 	return flex_gd;
176 
177 out1:
178 	kfree(flex_gd->groups);
179 out2:
180 	kfree(flex_gd);
181 out3:
182 	return NULL;
183 }
184 
185 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
186 {
187 	kfree(flex_gd->bg_flags);
188 	kfree(flex_gd->groups);
189 	kfree(flex_gd);
190 }
191 
192 /*
193  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
194  * and inode tables for a flex group.
195  *
196  * This function is used by 64bit-resize.  Note that this function allocates
197  * group tables from the 1st group of groups contained by @flexgd, which may
198  * be a partial of a flex group.
199  *
200  * @sb: super block of fs to which the groups belongs
201  */
202 static void ext4_alloc_group_tables(struct super_block *sb,
203 				struct ext4_new_flex_group_data *flex_gd,
204 				int flexbg_size)
205 {
206 	struct ext4_new_group_data *group_data = flex_gd->groups;
207 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
208 	ext4_fsblk_t start_blk;
209 	ext4_fsblk_t last_blk;
210 	ext4_group_t src_group;
211 	ext4_group_t bb_index = 0;
212 	ext4_group_t ib_index = 0;
213 	ext4_group_t it_index = 0;
214 	ext4_group_t group;
215 	ext4_group_t last_group;
216 	unsigned overhead;
217 
218 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
219 
220 	src_group = group_data[0].group;
221 	last_group  = src_group + flex_gd->count - 1;
222 
223 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
224 	       (last_group & ~(flexbg_size - 1))));
225 next_group:
226 	group = group_data[0].group;
227 	start_blk = ext4_group_first_block_no(sb, src_group);
228 	last_blk = start_blk + group_data[src_group - group].blocks_count;
229 
230 	overhead = ext4_bg_has_super(sb, src_group) ?
231 		   (1 + ext4_bg_num_gdb(sb, src_group) +
232 		    le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
233 
234 	start_blk += overhead;
235 
236 	BUG_ON(src_group >= group_data[0].group + flex_gd->count);
237 	/* We collect contiguous blocks as much as possible. */
238 	src_group++;
239 	for (; src_group <= last_group; src_group++)
240 		if (!ext4_bg_has_super(sb, src_group))
241 			last_blk += group_data[src_group - group].blocks_count;
242 		else
243 			break;
244 
245 	/* Allocate block bitmaps */
246 	for (; bb_index < flex_gd->count; bb_index++) {
247 		if (start_blk >= last_blk)
248 			goto next_group;
249 		group_data[bb_index].block_bitmap = start_blk++;
250 		ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
251 		group -= group_data[0].group;
252 		group_data[group].free_blocks_count--;
253 		if (flexbg_size > 1)
254 			flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
255 	}
256 
257 	/* Allocate inode bitmaps */
258 	for (; ib_index < flex_gd->count; ib_index++) {
259 		if (start_blk >= last_blk)
260 			goto next_group;
261 		group_data[ib_index].inode_bitmap = start_blk++;
262 		ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
263 		group -= group_data[0].group;
264 		group_data[group].free_blocks_count--;
265 		if (flexbg_size > 1)
266 			flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
267 	}
268 
269 	/* Allocate inode tables */
270 	for (; it_index < flex_gd->count; it_index++) {
271 		if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
272 			goto next_group;
273 		group_data[it_index].inode_table = start_blk;
274 		ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
275 		group -= group_data[0].group;
276 		group_data[group].free_blocks_count -=
277 					EXT4_SB(sb)->s_itb_per_group;
278 		if (flexbg_size > 1)
279 			flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
280 
281 		start_blk += EXT4_SB(sb)->s_itb_per_group;
282 	}
283 
284 	if (test_opt(sb, DEBUG)) {
285 		int i;
286 		group = group_data[0].group;
287 
288 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
289 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
290 		       flexbg_size);
291 
292 		for (i = 0; i < flex_gd->count; i++) {
293 			printk(KERN_DEBUG "adding %s group %u: %u "
294 			       "blocks (%d free)\n",
295 			       ext4_bg_has_super(sb, group + i) ? "normal" :
296 			       "no-super", group + i,
297 			       group_data[i].blocks_count,
298 			       group_data[i].free_blocks_count);
299 		}
300 	}
301 }
302 
303 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
304 				  ext4_fsblk_t blk)
305 {
306 	struct buffer_head *bh;
307 	int err;
308 
309 	bh = sb_getblk(sb, blk);
310 	if (!bh)
311 		return ERR_PTR(-EIO);
312 	if ((err = ext4_journal_get_write_access(handle, bh))) {
313 		brelse(bh);
314 		bh = ERR_PTR(err);
315 	} else {
316 		memset(bh->b_data, 0, sb->s_blocksize);
317 		set_buffer_uptodate(bh);
318 	}
319 
320 	return bh;
321 }
322 
323 /*
324  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
325  * If that fails, restart the transaction & regain write access for the
326  * buffer head which is used for block_bitmap modifications.
327  */
328 static int extend_or_restart_transaction(handle_t *handle, int thresh)
329 {
330 	int err;
331 
332 	if (ext4_handle_has_enough_credits(handle, thresh))
333 		return 0;
334 
335 	err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
336 	if (err < 0)
337 		return err;
338 	if (err) {
339 		err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
340 		if (err)
341 			return err;
342 	}
343 
344 	return 0;
345 }
346 
347 /*
348  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
349  *
350  * Helper function for ext4_setup_new_group_blocks() which set .
351  *
352  * @sb: super block
353  * @handle: journal handle
354  * @flex_gd: flex group data
355  */
356 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
357 			struct ext4_new_flex_group_data *flex_gd,
358 			ext4_fsblk_t block, ext4_group_t count)
359 {
360 	ext4_group_t count2;
361 
362 	ext4_debug("mark blocks [%llu/%u] used\n", block, count);
363 	for (count2 = count; count > 0; count -= count2, block += count2) {
364 		ext4_fsblk_t start;
365 		struct buffer_head *bh;
366 		ext4_group_t group;
367 		int err;
368 
369 		ext4_get_group_no_and_offset(sb, block, &group, NULL);
370 		start = ext4_group_first_block_no(sb, group);
371 		group -= flex_gd->groups[0].group;
372 
373 		count2 = sb->s_blocksize * 8 - (block - start);
374 		if (count2 > count)
375 			count2 = count;
376 
377 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
378 			BUG_ON(flex_gd->count > 1);
379 			continue;
380 		}
381 
382 		err = extend_or_restart_transaction(handle, 1);
383 		if (err)
384 			return err;
385 
386 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
387 		if (!bh)
388 			return -EIO;
389 
390 		err = ext4_journal_get_write_access(handle, bh);
391 		if (err)
392 			return err;
393 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
394 			   block - start, count2);
395 		ext4_set_bits(bh->b_data, block - start, count2);
396 
397 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
398 		if (unlikely(err))
399 			return err;
400 		brelse(bh);
401 	}
402 
403 	return 0;
404 }
405 
406 /*
407  * Set up the block and inode bitmaps, and the inode table for the new groups.
408  * This doesn't need to be part of the main transaction, since we are only
409  * changing blocks outside the actual filesystem.  We still do journaling to
410  * ensure the recovery is correct in case of a failure just after resize.
411  * If any part of this fails, we simply abort the resize.
412  *
413  * setup_new_flex_group_blocks handles a flex group as follow:
414  *  1. copy super block and GDT, and initialize group tables if necessary.
415  *     In this step, we only set bits in blocks bitmaps for blocks taken by
416  *     super block and GDT.
417  *  2. allocate group tables in block bitmaps, that is, set bits in block
418  *     bitmap for blocks taken by group tables.
419  */
420 static int setup_new_flex_group_blocks(struct super_block *sb,
421 				struct ext4_new_flex_group_data *flex_gd)
422 {
423 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
424 	ext4_fsblk_t start;
425 	ext4_fsblk_t block;
426 	struct ext4_sb_info *sbi = EXT4_SB(sb);
427 	struct ext4_super_block *es = sbi->s_es;
428 	struct ext4_new_group_data *group_data = flex_gd->groups;
429 	__u16 *bg_flags = flex_gd->bg_flags;
430 	handle_t *handle;
431 	ext4_group_t group, count;
432 	struct buffer_head *bh = NULL;
433 	int reserved_gdb, i, j, err = 0, err2;
434 
435 	BUG_ON(!flex_gd->count || !group_data ||
436 	       group_data[0].group != sbi->s_groups_count);
437 
438 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
439 
440 	/* This transaction may be extended/restarted along the way */
441 	handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
442 	if (IS_ERR(handle))
443 		return PTR_ERR(handle);
444 
445 	group = group_data[0].group;
446 	for (i = 0; i < flex_gd->count; i++, group++) {
447 		unsigned long gdblocks;
448 
449 		gdblocks = ext4_bg_num_gdb(sb, group);
450 		start = ext4_group_first_block_no(sb, group);
451 
452 		/* Copy all of the GDT blocks into the backup in this group */
453 		for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
454 			struct buffer_head *gdb;
455 
456 			ext4_debug("update backup group %#04llx\n", block);
457 			err = extend_or_restart_transaction(handle, 1);
458 			if (err)
459 				goto out;
460 
461 			gdb = sb_getblk(sb, block);
462 			if (!gdb) {
463 				err = -EIO;
464 				goto out;
465 			}
466 
467 			err = ext4_journal_get_write_access(handle, gdb);
468 			if (err) {
469 				brelse(gdb);
470 				goto out;
471 			}
472 			memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
473 			       gdb->b_size);
474 			set_buffer_uptodate(gdb);
475 
476 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
477 			if (unlikely(err)) {
478 				brelse(gdb);
479 				goto out;
480 			}
481 			brelse(gdb);
482 		}
483 
484 		/* Zero out all of the reserved backup group descriptor
485 		 * table blocks
486 		 */
487 		if (ext4_bg_has_super(sb, group)) {
488 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
489 					reserved_gdb, GFP_NOFS);
490 			if (err)
491 				goto out;
492 		}
493 
494 		/* Initialize group tables of the grop @group */
495 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
496 			goto handle_bb;
497 
498 		/* Zero out all of the inode table blocks */
499 		block = group_data[i].inode_table;
500 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
501 			   block, sbi->s_itb_per_group);
502 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
503 				       GFP_NOFS);
504 		if (err)
505 			goto out;
506 
507 handle_bb:
508 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
509 			goto handle_ib;
510 
511 		/* Initialize block bitmap of the @group */
512 		block = group_data[i].block_bitmap;
513 		err = extend_or_restart_transaction(handle, 1);
514 		if (err)
515 			goto out;
516 
517 		bh = bclean(handle, sb, block);
518 		if (IS_ERR(bh)) {
519 			err = PTR_ERR(bh);
520 			goto out;
521 		}
522 		if (ext4_bg_has_super(sb, group)) {
523 			ext4_debug("mark backup superblock %#04llx (+0)\n",
524 				   start);
525 			ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
526 						     1);
527 		}
528 		ext4_mark_bitmap_end(group_data[i].blocks_count,
529 				     sb->s_blocksize * 8, bh->b_data);
530 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
531 		if (err)
532 			goto out;
533 		brelse(bh);
534 
535 handle_ib:
536 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
537 			continue;
538 
539 		/* Initialize inode bitmap of the @group */
540 		block = group_data[i].inode_bitmap;
541 		err = extend_or_restart_transaction(handle, 1);
542 		if (err)
543 			goto out;
544 		/* Mark unused entries in inode bitmap used */
545 		bh = bclean(handle, sb, block);
546 		if (IS_ERR(bh)) {
547 			err = PTR_ERR(bh);
548 			goto out;
549 		}
550 
551 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
552 				     sb->s_blocksize * 8, bh->b_data);
553 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
554 		if (err)
555 			goto out;
556 		brelse(bh);
557 	}
558 	bh = NULL;
559 
560 	/* Mark group tables in block bitmap */
561 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
562 		count = group_table_count[j];
563 		start = (&group_data[0].block_bitmap)[j];
564 		block = start;
565 		for (i = 1; i < flex_gd->count; i++) {
566 			block += group_table_count[j];
567 			if (block == (&group_data[i].block_bitmap)[j]) {
568 				count += group_table_count[j];
569 				continue;
570 			}
571 			err = set_flexbg_block_bitmap(sb, handle,
572 						flex_gd, start, count);
573 			if (err)
574 				goto out;
575 			count = group_table_count[j];
576 			start = group_data[i].block_bitmap;
577 			block = start;
578 		}
579 
580 		if (count) {
581 			err = set_flexbg_block_bitmap(sb, handle,
582 						flex_gd, start, count);
583 			if (err)
584 				goto out;
585 		}
586 	}
587 
588 out:
589 	brelse(bh);
590 	err2 = ext4_journal_stop(handle);
591 	if (err2 && !err)
592 		err = err2;
593 
594 	return err;
595 }
596 
597 /*
598  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
599  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
600  * calling this for the first time.  In a sparse filesystem it will be the
601  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
602  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
603  */
604 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
605 				  unsigned *five, unsigned *seven)
606 {
607 	unsigned *min = three;
608 	int mult = 3;
609 	unsigned ret;
610 
611 	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
612 					EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
613 		ret = *min;
614 		*min += 1;
615 		return ret;
616 	}
617 
618 	if (*five < *min) {
619 		min = five;
620 		mult = 5;
621 	}
622 	if (*seven < *min) {
623 		min = seven;
624 		mult = 7;
625 	}
626 
627 	ret = *min;
628 	*min *= mult;
629 
630 	return ret;
631 }
632 
633 /*
634  * Check that all of the backup GDT blocks are held in the primary GDT block.
635  * It is assumed that they are stored in group order.  Returns the number of
636  * groups in current filesystem that have BACKUPS, or -ve error code.
637  */
638 static int verify_reserved_gdb(struct super_block *sb,
639 			       ext4_group_t end,
640 			       struct buffer_head *primary)
641 {
642 	const ext4_fsblk_t blk = primary->b_blocknr;
643 	unsigned three = 1;
644 	unsigned five = 5;
645 	unsigned seven = 7;
646 	unsigned grp;
647 	__le32 *p = (__le32 *)primary->b_data;
648 	int gdbackups = 0;
649 
650 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
651 		if (le32_to_cpu(*p++) !=
652 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
653 			ext4_warning(sb, "reserved GDT %llu"
654 				     " missing grp %d (%llu)",
655 				     blk, grp,
656 				     grp *
657 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
658 				     blk);
659 			return -EINVAL;
660 		}
661 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
662 			return -EFBIG;
663 	}
664 
665 	return gdbackups;
666 }
667 
668 /*
669  * Called when we need to bring a reserved group descriptor table block into
670  * use from the resize inode.  The primary copy of the new GDT block currently
671  * is an indirect block (under the double indirect block in the resize inode).
672  * The new backup GDT blocks will be stored as leaf blocks in this indirect
673  * block, in group order.  Even though we know all the block numbers we need,
674  * we check to ensure that the resize inode has actually reserved these blocks.
675  *
676  * Don't need to update the block bitmaps because the blocks are still in use.
677  *
678  * We get all of the error cases out of the way, so that we are sure to not
679  * fail once we start modifying the data on disk, because JBD has no rollback.
680  */
681 static int add_new_gdb(handle_t *handle, struct inode *inode,
682 		       ext4_group_t group)
683 {
684 	struct super_block *sb = inode->i_sb;
685 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
686 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
687 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
688 	struct buffer_head **o_group_desc, **n_group_desc;
689 	struct buffer_head *dind;
690 	struct buffer_head *gdb_bh;
691 	int gdbackups;
692 	struct ext4_iloc iloc;
693 	__le32 *data;
694 	int err;
695 
696 	if (test_opt(sb, DEBUG))
697 		printk(KERN_DEBUG
698 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
699 		       gdb_num);
700 
701 	/*
702 	 * If we are not using the primary superblock/GDT copy don't resize,
703          * because the user tools have no way of handling this.  Probably a
704          * bad time to do it anyways.
705          */
706 	if (EXT4_SB(sb)->s_sbh->b_blocknr !=
707 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
708 		ext4_warning(sb, "won't resize using backup superblock at %llu",
709 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
710 		return -EPERM;
711 	}
712 
713 	gdb_bh = sb_bread(sb, gdblock);
714 	if (!gdb_bh)
715 		return -EIO;
716 
717 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
718 	if (gdbackups < 0) {
719 		err = gdbackups;
720 		goto exit_bh;
721 	}
722 
723 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
724 	dind = sb_bread(sb, le32_to_cpu(*data));
725 	if (!dind) {
726 		err = -EIO;
727 		goto exit_bh;
728 	}
729 
730 	data = (__le32 *)dind->b_data;
731 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
732 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
733 			     group, gdblock);
734 		err = -EINVAL;
735 		goto exit_dind;
736 	}
737 
738 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
739 	if (unlikely(err))
740 		goto exit_dind;
741 
742 	err = ext4_journal_get_write_access(handle, gdb_bh);
743 	if (unlikely(err))
744 		goto exit_sbh;
745 
746 	err = ext4_journal_get_write_access(handle, dind);
747 	if (unlikely(err))
748 		ext4_std_error(sb, err);
749 
750 	/* ext4_reserve_inode_write() gets a reference on the iloc */
751 	err = ext4_reserve_inode_write(handle, inode, &iloc);
752 	if (unlikely(err))
753 		goto exit_dindj;
754 
755 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
756 				     sizeof(struct buffer_head *),
757 				     GFP_NOFS);
758 	if (!n_group_desc) {
759 		err = -ENOMEM;
760 		ext4_warning(sb, "not enough memory for %lu groups",
761 			     gdb_num + 1);
762 		goto exit_inode;
763 	}
764 
765 	/*
766 	 * Finally, we have all of the possible failures behind us...
767 	 *
768 	 * Remove new GDT block from inode double-indirect block and clear out
769 	 * the new GDT block for use (which also "frees" the backup GDT blocks
770 	 * from the reserved inode).  We don't need to change the bitmaps for
771 	 * these blocks, because they are marked as in-use from being in the
772 	 * reserved inode, and will become GDT blocks (primary and backup).
773 	 */
774 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
775 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
776 	if (unlikely(err)) {
777 		ext4_std_error(sb, err);
778 		goto exit_inode;
779 	}
780 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
781 	ext4_mark_iloc_dirty(handle, inode, &iloc);
782 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
783 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
784 	if (unlikely(err)) {
785 		ext4_std_error(sb, err);
786 		goto exit_inode;
787 	}
788 	brelse(dind);
789 
790 	o_group_desc = EXT4_SB(sb)->s_group_desc;
791 	memcpy(n_group_desc, o_group_desc,
792 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
793 	n_group_desc[gdb_num] = gdb_bh;
794 	EXT4_SB(sb)->s_group_desc = n_group_desc;
795 	EXT4_SB(sb)->s_gdb_count++;
796 	ext4_kvfree(o_group_desc);
797 
798 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
799 	err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
800 	if (err)
801 		ext4_std_error(sb, err);
802 
803 	return err;
804 
805 exit_inode:
806 	ext4_kvfree(n_group_desc);
807 	/* ext4_handle_release_buffer(handle, iloc.bh); */
808 	brelse(iloc.bh);
809 exit_dindj:
810 	/* ext4_handle_release_buffer(handle, dind); */
811 exit_sbh:
812 	/* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
813 exit_dind:
814 	brelse(dind);
815 exit_bh:
816 	brelse(gdb_bh);
817 
818 	ext4_debug("leaving with error %d\n", err);
819 	return err;
820 }
821 
822 /*
823  * Called when we are adding a new group which has a backup copy of each of
824  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
825  * We need to add these reserved backup GDT blocks to the resize inode, so
826  * that they are kept for future resizing and not allocated to files.
827  *
828  * Each reserved backup GDT block will go into a different indirect block.
829  * The indirect blocks are actually the primary reserved GDT blocks,
830  * so we know in advance what their block numbers are.  We only get the
831  * double-indirect block to verify it is pointing to the primary reserved
832  * GDT blocks so we don't overwrite a data block by accident.  The reserved
833  * backup GDT blocks are stored in their reserved primary GDT block.
834  */
835 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
836 			      ext4_group_t group)
837 {
838 	struct super_block *sb = inode->i_sb;
839 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
840 	struct buffer_head **primary;
841 	struct buffer_head *dind;
842 	struct ext4_iloc iloc;
843 	ext4_fsblk_t blk;
844 	__le32 *data, *end;
845 	int gdbackups = 0;
846 	int res, i;
847 	int err;
848 
849 	primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
850 	if (!primary)
851 		return -ENOMEM;
852 
853 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
854 	dind = sb_bread(sb, le32_to_cpu(*data));
855 	if (!dind) {
856 		err = -EIO;
857 		goto exit_free;
858 	}
859 
860 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
861 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
862 					 EXT4_ADDR_PER_BLOCK(sb));
863 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
864 
865 	/* Get each reserved primary GDT block and verify it holds backups */
866 	for (res = 0; res < reserved_gdb; res++, blk++) {
867 		if (le32_to_cpu(*data) != blk) {
868 			ext4_warning(sb, "reserved block %llu"
869 				     " not at offset %ld",
870 				     blk,
871 				     (long)(data - (__le32 *)dind->b_data));
872 			err = -EINVAL;
873 			goto exit_bh;
874 		}
875 		primary[res] = sb_bread(sb, blk);
876 		if (!primary[res]) {
877 			err = -EIO;
878 			goto exit_bh;
879 		}
880 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
881 		if (gdbackups < 0) {
882 			brelse(primary[res]);
883 			err = gdbackups;
884 			goto exit_bh;
885 		}
886 		if (++data >= end)
887 			data = (__le32 *)dind->b_data;
888 	}
889 
890 	for (i = 0; i < reserved_gdb; i++) {
891 		if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
892 			/*
893 			int j;
894 			for (j = 0; j < i; j++)
895 				ext4_handle_release_buffer(handle, primary[j]);
896 			 */
897 			goto exit_bh;
898 		}
899 	}
900 
901 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
902 		goto exit_bh;
903 
904 	/*
905 	 * Finally we can add each of the reserved backup GDT blocks from
906 	 * the new group to its reserved primary GDT block.
907 	 */
908 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
909 	for (i = 0; i < reserved_gdb; i++) {
910 		int err2;
911 		data = (__le32 *)primary[i]->b_data;
912 		/* printk("reserving backup %lu[%u] = %lu\n",
913 		       primary[i]->b_blocknr, gdbackups,
914 		       blk + primary[i]->b_blocknr); */
915 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
916 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
917 		if (!err)
918 			err = err2;
919 	}
920 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
921 	ext4_mark_iloc_dirty(handle, inode, &iloc);
922 
923 exit_bh:
924 	while (--res >= 0)
925 		brelse(primary[res]);
926 	brelse(dind);
927 
928 exit_free:
929 	kfree(primary);
930 
931 	return err;
932 }
933 
934 /*
935  * Update the backup copies of the ext4 metadata.  These don't need to be part
936  * of the main resize transaction, because e2fsck will re-write them if there
937  * is a problem (basically only OOM will cause a problem).  However, we
938  * _should_ update the backups if possible, in case the primary gets trashed
939  * for some reason and we need to run e2fsck from a backup superblock.  The
940  * important part is that the new block and inode counts are in the backup
941  * superblocks, and the location of the new group metadata in the GDT backups.
942  *
943  * We do not need take the s_resize_lock for this, because these
944  * blocks are not otherwise touched by the filesystem code when it is
945  * mounted.  We don't need to worry about last changing from
946  * sbi->s_groups_count, because the worst that can happen is that we
947  * do not copy the full number of backups at this time.  The resize
948  * which changed s_groups_count will backup again.
949  */
950 static void update_backups(struct super_block *sb,
951 			   int blk_off, char *data, int size)
952 {
953 	struct ext4_sb_info *sbi = EXT4_SB(sb);
954 	const ext4_group_t last = sbi->s_groups_count;
955 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
956 	unsigned three = 1;
957 	unsigned five = 5;
958 	unsigned seven = 7;
959 	ext4_group_t group;
960 	int rest = sb->s_blocksize - size;
961 	handle_t *handle;
962 	int err = 0, err2;
963 
964 	handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
965 	if (IS_ERR(handle)) {
966 		group = 1;
967 		err = PTR_ERR(handle);
968 		goto exit_err;
969 	}
970 
971 	while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
972 		struct buffer_head *bh;
973 
974 		/* Out of journal space, and can't get more - abort - so sad */
975 		if (ext4_handle_valid(handle) &&
976 		    handle->h_buffer_credits == 0 &&
977 		    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
978 		    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
979 			break;
980 
981 		bh = sb_getblk(sb, group * bpg + blk_off);
982 		if (!bh) {
983 			err = -EIO;
984 			break;
985 		}
986 		ext4_debug("update metadata backup %#04lx\n",
987 			  (unsigned long)bh->b_blocknr);
988 		if ((err = ext4_journal_get_write_access(handle, bh)))
989 			break;
990 		lock_buffer(bh);
991 		memcpy(bh->b_data, data, size);
992 		if (rest)
993 			memset(bh->b_data + size, 0, rest);
994 		set_buffer_uptodate(bh);
995 		unlock_buffer(bh);
996 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
997 		if (unlikely(err))
998 			ext4_std_error(sb, err);
999 		brelse(bh);
1000 	}
1001 	if ((err2 = ext4_journal_stop(handle)) && !err)
1002 		err = err2;
1003 
1004 	/*
1005 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1006 	 * late to revert the resize, we shouldn't fail just because of
1007 	 * the backup copies (they are only needed in case of corruption).
1008 	 *
1009 	 * However, if we got here we have a journal problem too, so we
1010 	 * can't really start a transaction to mark the superblock.
1011 	 * Chicken out and just set the flag on the hope it will be written
1012 	 * to disk, and if not - we will simply wait until next fsck.
1013 	 */
1014 exit_err:
1015 	if (err) {
1016 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1017 			     "forcing fsck on next reboot", group, err);
1018 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1019 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1020 		mark_buffer_dirty(sbi->s_sbh);
1021 	}
1022 }
1023 
1024 /*
1025  * ext4_add_new_descs() adds @count group descriptor of groups
1026  * starting at @group
1027  *
1028  * @handle: journal handle
1029  * @sb: super block
1030  * @group: the group no. of the first group desc to be added
1031  * @resize_inode: the resize inode
1032  * @count: number of group descriptors to be added
1033  */
1034 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1035 			      ext4_group_t group, struct inode *resize_inode,
1036 			      ext4_group_t count)
1037 {
1038 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1039 	struct ext4_super_block *es = sbi->s_es;
1040 	struct buffer_head *gdb_bh;
1041 	int i, gdb_off, gdb_num, err = 0;
1042 
1043 	for (i = 0; i < count; i++, group++) {
1044 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1045 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1046 
1047 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1048 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1049 
1050 		/*
1051 		 * We will only either add reserved group blocks to a backup group
1052 		 * or remove reserved blocks for the first group in a new group block.
1053 		 * Doing both would be mean more complex code, and sane people don't
1054 		 * use non-sparse filesystems anymore.  This is already checked above.
1055 		 */
1056 		if (gdb_off) {
1057 			gdb_bh = sbi->s_group_desc[gdb_num];
1058 			err = ext4_journal_get_write_access(handle, gdb_bh);
1059 
1060 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1061 				err = reserve_backup_gdb(handle, resize_inode, group);
1062 		} else
1063 			err = add_new_gdb(handle, resize_inode, group);
1064 		if (err)
1065 			break;
1066 	}
1067 	return err;
1068 }
1069 
1070 /*
1071  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1072  */
1073 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1074 				struct ext4_new_flex_group_data *flex_gd)
1075 {
1076 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1077 	struct ext4_group_desc		*gdp;
1078 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1079 	struct buffer_head		*gdb_bh;
1080 	ext4_group_t			group;
1081 	__u16				*bg_flags = flex_gd->bg_flags;
1082 	int				i, gdb_off, gdb_num, err = 0;
1083 
1084 
1085 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1086 		group = group_data->group;
1087 
1088 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1089 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1090 
1091 		/*
1092 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1093 		 */
1094 		gdb_bh = sbi->s_group_desc[gdb_num];
1095 		/* Update group descriptor block for new group */
1096 		gdp = (struct ext4_group_desc *)((char *)gdb_bh->b_data +
1097 						 gdb_off * EXT4_DESC_SIZE(sb));
1098 
1099 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1100 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1101 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1102 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1103 		ext4_free_group_clusters_set(sb, gdp,
1104 					     EXT4_B2C(sbi, group_data->free_blocks_count));
1105 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1106 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1107 		gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1108 
1109 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1110 		if (unlikely(err)) {
1111 			ext4_std_error(sb, err);
1112 			break;
1113 		}
1114 
1115 		/*
1116 		 * We can allocate memory for mb_alloc based on the new group
1117 		 * descriptor
1118 		 */
1119 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1120 		if (err)
1121 			break;
1122 	}
1123 	return err;
1124 }
1125 
1126 /*
1127  * ext4_update_super() updates the super block so that the newly added
1128  * groups can be seen by the filesystem.
1129  *
1130  * @sb: super block
1131  * @flex_gd: new added groups
1132  */
1133 static void ext4_update_super(struct super_block *sb,
1134 			     struct ext4_new_flex_group_data *flex_gd)
1135 {
1136 	ext4_fsblk_t blocks_count = 0;
1137 	ext4_fsblk_t free_blocks = 0;
1138 	ext4_fsblk_t reserved_blocks = 0;
1139 	struct ext4_new_group_data *group_data = flex_gd->groups;
1140 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1141 	struct ext4_super_block *es = sbi->s_es;
1142 	int i;
1143 
1144 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1145 	/*
1146 	 * Make the new blocks and inodes valid next.  We do this before
1147 	 * increasing the group count so that once the group is enabled,
1148 	 * all of its blocks and inodes are already valid.
1149 	 *
1150 	 * We always allocate group-by-group, then block-by-block or
1151 	 * inode-by-inode within a group, so enabling these
1152 	 * blocks/inodes before the group is live won't actually let us
1153 	 * allocate the new space yet.
1154 	 */
1155 	for (i = 0; i < flex_gd->count; i++) {
1156 		blocks_count += group_data[i].blocks_count;
1157 		free_blocks += group_data[i].free_blocks_count;
1158 	}
1159 
1160 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1161 	do_div(reserved_blocks, ext4_blocks_count(es));
1162 	reserved_blocks *= blocks_count;
1163 	do_div(reserved_blocks, 100);
1164 
1165 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1166 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1167 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1168 		     flex_gd->count);
1169 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1170 		     flex_gd->count);
1171 
1172 	/*
1173 	 * We need to protect s_groups_count against other CPUs seeing
1174 	 * inconsistent state in the superblock.
1175 	 *
1176 	 * The precise rules we use are:
1177 	 *
1178 	 * * Writers must perform a smp_wmb() after updating all
1179 	 *   dependent data and before modifying the groups count
1180 	 *
1181 	 * * Readers must perform an smp_rmb() after reading the groups
1182 	 *   count and before reading any dependent data.
1183 	 *
1184 	 * NB. These rules can be relaxed when checking the group count
1185 	 * while freeing data, as we can only allocate from a block
1186 	 * group after serialising against the group count, and we can
1187 	 * only then free after serialising in turn against that
1188 	 * allocation.
1189 	 */
1190 	smp_wmb();
1191 
1192 	/* Update the global fs size fields */
1193 	sbi->s_groups_count += flex_gd->count;
1194 
1195 	/* Update the reserved block counts only once the new group is
1196 	 * active. */
1197 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1198 				reserved_blocks);
1199 
1200 	/* Update the free space counts */
1201 	percpu_counter_add(&sbi->s_freeclusters_counter,
1202 			   EXT4_B2C(sbi, free_blocks));
1203 	percpu_counter_add(&sbi->s_freeinodes_counter,
1204 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1205 
1206 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1207 				      EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1208 	    sbi->s_log_groups_per_flex) {
1209 		ext4_group_t flex_group;
1210 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1211 		atomic_add(EXT4_B2C(sbi, free_blocks),
1212 			   &sbi->s_flex_groups[flex_group].free_clusters);
1213 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1214 			   &sbi->s_flex_groups[flex_group].free_inodes);
1215 	}
1216 
1217 	if (test_opt(sb, DEBUG))
1218 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1219 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1220 		       blocks_count, free_blocks, reserved_blocks);
1221 }
1222 
1223 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1224  * _before_ we start modifying the filesystem, because we cannot abort the
1225  * transaction and not have it write the data to disk.
1226  */
1227 static int ext4_flex_group_add(struct super_block *sb,
1228 			       struct inode *resize_inode,
1229 			       struct ext4_new_flex_group_data *flex_gd)
1230 {
1231 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1232 	struct ext4_super_block *es = sbi->s_es;
1233 	ext4_fsblk_t o_blocks_count;
1234 	ext4_grpblk_t last;
1235 	ext4_group_t group;
1236 	handle_t *handle;
1237 	unsigned reserved_gdb;
1238 	int err = 0, err2 = 0, credit;
1239 
1240 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1241 
1242 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1243 	o_blocks_count = ext4_blocks_count(es);
1244 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1245 	BUG_ON(last);
1246 
1247 	err = setup_new_flex_group_blocks(sb, flex_gd);
1248 	if (err)
1249 		goto exit;
1250 	/*
1251 	 * We will always be modifying at least the superblock and  GDT
1252 	 * block.  If we are adding a group past the last current GDT block,
1253 	 * we will also modify the inode and the dindirect block.  If we
1254 	 * are adding a group with superblock/GDT backups  we will also
1255 	 * modify each of the reserved GDT dindirect blocks.
1256 	 */
1257 	credit = flex_gd->count * 4 + reserved_gdb;
1258 	handle = ext4_journal_start_sb(sb, credit);
1259 	if (IS_ERR(handle)) {
1260 		err = PTR_ERR(handle);
1261 		goto exit;
1262 	}
1263 
1264 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1265 	if (err)
1266 		goto exit_journal;
1267 
1268 	group = flex_gd->groups[0].group;
1269 	BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1270 	err = ext4_add_new_descs(handle, sb, group,
1271 				resize_inode, flex_gd->count);
1272 	if (err)
1273 		goto exit_journal;
1274 
1275 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1276 	if (err)
1277 		goto exit_journal;
1278 
1279 	ext4_update_super(sb, flex_gd);
1280 
1281 	err = ext4_handle_dirty_super(handle, sb);
1282 
1283 exit_journal:
1284 	err2 = ext4_journal_stop(handle);
1285 	if (!err)
1286 		err = err2;
1287 
1288 	if (!err) {
1289 		int i;
1290 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1291 			       sizeof(struct ext4_super_block));
1292 		for (i = 0; i < flex_gd->count; i++, group++) {
1293 			struct buffer_head *gdb_bh;
1294 			int gdb_num;
1295 			gdb_num = group / EXT4_BLOCKS_PER_GROUP(sb);
1296 			gdb_bh = sbi->s_group_desc[gdb_num];
1297 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1298 				       gdb_bh->b_size);
1299 		}
1300 	}
1301 exit:
1302 	return err;
1303 }
1304 
1305 static int ext4_setup_next_flex_gd(struct super_block *sb,
1306 				    struct ext4_new_flex_group_data *flex_gd,
1307 				    ext4_fsblk_t n_blocks_count,
1308 				    unsigned long flexbg_size)
1309 {
1310 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1311 	struct ext4_new_group_data *group_data = flex_gd->groups;
1312 	ext4_fsblk_t o_blocks_count;
1313 	ext4_group_t n_group;
1314 	ext4_group_t group;
1315 	ext4_group_t last_group;
1316 	ext4_grpblk_t last;
1317 	ext4_grpblk_t blocks_per_group;
1318 	unsigned long i;
1319 
1320 	blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1321 
1322 	o_blocks_count = ext4_blocks_count(es);
1323 
1324 	if (o_blocks_count == n_blocks_count)
1325 		return 0;
1326 
1327 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1328 	BUG_ON(last);
1329 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1330 
1331 	last_group = group | (flexbg_size - 1);
1332 	if (last_group > n_group)
1333 		last_group = n_group;
1334 
1335 	flex_gd->count = last_group - group + 1;
1336 
1337 	for (i = 0; i < flex_gd->count; i++) {
1338 		int overhead;
1339 
1340 		group_data[i].group = group + i;
1341 		group_data[i].blocks_count = blocks_per_group;
1342 		overhead = ext4_bg_has_super(sb, group + i) ?
1343 			   (1 + ext4_bg_num_gdb(sb, group + i) +
1344 			    le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
1345 		group_data[i].free_blocks_count = blocks_per_group - overhead;
1346 		if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1347 					       EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
1348 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1349 					       EXT4_BG_INODE_UNINIT;
1350 		else
1351 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1352 	}
1353 
1354 	if (last_group == n_group &&
1355 	    EXT4_HAS_RO_COMPAT_FEATURE(sb,
1356 				       EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
1357 		/* We need to initialize block bitmap of last group. */
1358 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1359 
1360 	if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1361 		group_data[i - 1].blocks_count = last + 1;
1362 		group_data[i - 1].free_blocks_count -= blocks_per_group-
1363 					last - 1;
1364 	}
1365 
1366 	return 1;
1367 }
1368 
1369 /* Add group descriptor data to an existing or new group descriptor block.
1370  * Ensure we handle all possible error conditions _before_ we start modifying
1371  * the filesystem, because we cannot abort the transaction and not have it
1372  * write the data to disk.
1373  *
1374  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1375  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1376  *
1377  * We only need to hold the superblock lock while we are actually adding
1378  * in the new group's counts to the superblock.  Prior to that we have
1379  * not really "added" the group at all.  We re-check that we are still
1380  * adding in the last group in case things have changed since verifying.
1381  */
1382 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1383 {
1384 	struct ext4_new_flex_group_data flex_gd;
1385 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1386 	struct ext4_super_block *es = sbi->s_es;
1387 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1388 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1389 	struct inode *inode = NULL;
1390 	int gdb_off, gdb_num;
1391 	int err;
1392 	__u16 bg_flags = 0;
1393 
1394 	gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1395 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1396 
1397 	if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1398 					EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1399 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1400 		return -EPERM;
1401 	}
1402 
1403 	if (ext4_blocks_count(es) + input->blocks_count <
1404 	    ext4_blocks_count(es)) {
1405 		ext4_warning(sb, "blocks_count overflow");
1406 		return -EINVAL;
1407 	}
1408 
1409 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1410 	    le32_to_cpu(es->s_inodes_count)) {
1411 		ext4_warning(sb, "inodes_count overflow");
1412 		return -EINVAL;
1413 	}
1414 
1415 	if (reserved_gdb || gdb_off == 0) {
1416 		if (!EXT4_HAS_COMPAT_FEATURE(sb,
1417 					     EXT4_FEATURE_COMPAT_RESIZE_INODE)
1418 		    || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1419 			ext4_warning(sb,
1420 				     "No reserved GDT blocks, can't resize");
1421 			return -EPERM;
1422 		}
1423 		inode = ext4_iget(sb, EXT4_RESIZE_INO);
1424 		if (IS_ERR(inode)) {
1425 			ext4_warning(sb, "Error opening resize inode");
1426 			return PTR_ERR(inode);
1427 		}
1428 	}
1429 
1430 
1431 	err = verify_group_input(sb, input);
1432 	if (err)
1433 		goto out;
1434 
1435 	flex_gd.count = 1;
1436 	flex_gd.groups = input;
1437 	flex_gd.bg_flags = &bg_flags;
1438 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1439 out:
1440 	iput(inode);
1441 	return err;
1442 } /* ext4_group_add */
1443 
1444 /*
1445  * extend a group without checking assuming that checking has been done.
1446  */
1447 static int ext4_group_extend_no_check(struct super_block *sb,
1448 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1449 {
1450 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1451 	handle_t *handle;
1452 	int err = 0, err2;
1453 
1454 	/* We will update the superblock, one block bitmap, and
1455 	 * one group descriptor via ext4_group_add_blocks().
1456 	 */
1457 	handle = ext4_journal_start_sb(sb, 3);
1458 	if (IS_ERR(handle)) {
1459 		err = PTR_ERR(handle);
1460 		ext4_warning(sb, "error %d on journal start", err);
1461 		return err;
1462 	}
1463 
1464 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1465 	if (err) {
1466 		ext4_warning(sb, "error %d on journal write access", err);
1467 		goto errout;
1468 	}
1469 
1470 	ext4_blocks_count_set(es, o_blocks_count + add);
1471 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1472 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1473 		   o_blocks_count + add);
1474 	/* We add the blocks to the bitmap and set the group need init bit */
1475 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1476 	if (err)
1477 		goto errout;
1478 	ext4_handle_dirty_super(handle, sb);
1479 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1480 		   o_blocks_count + add);
1481 errout:
1482 	err2 = ext4_journal_stop(handle);
1483 	if (err2 && !err)
1484 		err = err2;
1485 
1486 	if (!err) {
1487 		if (test_opt(sb, DEBUG))
1488 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1489 			       "blocks\n", ext4_blocks_count(es));
1490 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1491 			       sizeof(struct ext4_super_block));
1492 	}
1493 	return err;
1494 }
1495 
1496 /*
1497  * Extend the filesystem to the new number of blocks specified.  This entry
1498  * point is only used to extend the current filesystem to the end of the last
1499  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1500  * for emergencies (because it has no dependencies on reserved blocks).
1501  *
1502  * If we _really_ wanted, we could use default values to call ext4_group_add()
1503  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1504  * GDT blocks are reserved to grow to the desired size.
1505  */
1506 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1507 		      ext4_fsblk_t n_blocks_count)
1508 {
1509 	ext4_fsblk_t o_blocks_count;
1510 	ext4_grpblk_t last;
1511 	ext4_grpblk_t add;
1512 	struct buffer_head *bh;
1513 	int err;
1514 	ext4_group_t group;
1515 
1516 	o_blocks_count = ext4_blocks_count(es);
1517 
1518 	if (test_opt(sb, DEBUG))
1519 		ext4_msg(sb, KERN_DEBUG,
1520 			 "extending last group from %llu to %llu blocks",
1521 			 o_blocks_count, n_blocks_count);
1522 
1523 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1524 		return 0;
1525 
1526 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1527 		ext4_msg(sb, KERN_ERR,
1528 			 "filesystem too large to resize to %llu blocks safely",
1529 			 n_blocks_count);
1530 		if (sizeof(sector_t) < 8)
1531 			ext4_warning(sb, "CONFIG_LBDAF not enabled");
1532 		return -EINVAL;
1533 	}
1534 
1535 	if (n_blocks_count < o_blocks_count) {
1536 		ext4_warning(sb, "can't shrink FS - resize aborted");
1537 		return -EINVAL;
1538 	}
1539 
1540 	/* Handle the remaining blocks in the last group only. */
1541 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1542 
1543 	if (last == 0) {
1544 		ext4_warning(sb, "need to use ext2online to resize further");
1545 		return -EPERM;
1546 	}
1547 
1548 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1549 
1550 	if (o_blocks_count + add < o_blocks_count) {
1551 		ext4_warning(sb, "blocks_count overflow");
1552 		return -EINVAL;
1553 	}
1554 
1555 	if (o_blocks_count + add > n_blocks_count)
1556 		add = n_blocks_count - o_blocks_count;
1557 
1558 	if (o_blocks_count + add < n_blocks_count)
1559 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1560 			     o_blocks_count + add, add);
1561 
1562 	/* See if the device is actually as big as what was requested */
1563 	bh = sb_bread(sb, o_blocks_count + add - 1);
1564 	if (!bh) {
1565 		ext4_warning(sb, "can't read last block, resize aborted");
1566 		return -ENOSPC;
1567 	}
1568 	brelse(bh);
1569 
1570 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1571 	return err;
1572 } /* ext4_group_extend */
1573 
1574 /*
1575  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1576  *
1577  * @sb: super block of the fs to be resized
1578  * @n_blocks_count: the number of blocks resides in the resized fs
1579  */
1580 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1581 {
1582 	struct ext4_new_flex_group_data *flex_gd = NULL;
1583 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1584 	struct ext4_super_block *es = sbi->s_es;
1585 	struct buffer_head *bh;
1586 	struct inode *resize_inode;
1587 	ext4_fsblk_t o_blocks_count;
1588 	ext4_group_t o_group;
1589 	ext4_group_t n_group;
1590 	ext4_grpblk_t offset, add;
1591 	unsigned long n_desc_blocks;
1592 	unsigned long o_desc_blocks;
1593 	unsigned long desc_blocks;
1594 	int err = 0, flexbg_size = 1;
1595 
1596 	o_blocks_count = ext4_blocks_count(es);
1597 
1598 	if (test_opt(sb, DEBUG))
1599 		ext4_msg(sb, KERN_DEBUG, "resizing filesystem from %llu "
1600 		       "to %llu blocks", o_blocks_count, n_blocks_count);
1601 
1602 	if (n_blocks_count < o_blocks_count) {
1603 		/* On-line shrinking not supported */
1604 		ext4_warning(sb, "can't shrink FS - resize aborted");
1605 		return -EINVAL;
1606 	}
1607 
1608 	if (n_blocks_count == o_blocks_count)
1609 		/* Nothing need to do */
1610 		return 0;
1611 
1612 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
1613 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1614 
1615 	n_desc_blocks = (n_group + EXT4_DESC_PER_BLOCK(sb)) /
1616 			EXT4_DESC_PER_BLOCK(sb);
1617 	o_desc_blocks = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1618 			EXT4_DESC_PER_BLOCK(sb);
1619 	desc_blocks = n_desc_blocks - o_desc_blocks;
1620 
1621 	if (desc_blocks &&
1622 	    (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE) ||
1623 	     le16_to_cpu(es->s_reserved_gdt_blocks) < desc_blocks)) {
1624 		ext4_warning(sb, "No reserved GDT blocks, can't resize");
1625 		return -EPERM;
1626 	}
1627 
1628 	resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1629 	if (IS_ERR(resize_inode)) {
1630 		ext4_warning(sb, "Error opening resize inode");
1631 		return PTR_ERR(resize_inode);
1632 	}
1633 
1634 	/* See if the device is actually as big as what was requested */
1635 	bh = sb_bread(sb, n_blocks_count - 1);
1636 	if (!bh) {
1637 		ext4_warning(sb, "can't read last block, resize aborted");
1638 		return -ENOSPC;
1639 	}
1640 	brelse(bh);
1641 
1642 	/* extend the last group */
1643 	if (n_group == o_group)
1644 		add = n_blocks_count - o_blocks_count;
1645 	else
1646 		add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1647 	if (add > 0) {
1648 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1649 		if (err)
1650 			goto out;
1651 	}
1652 
1653 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1654 	    es->s_log_groups_per_flex)
1655 		flexbg_size = 1 << es->s_log_groups_per_flex;
1656 
1657 	o_blocks_count = ext4_blocks_count(es);
1658 	if (o_blocks_count == n_blocks_count)
1659 		goto out;
1660 
1661 	flex_gd = alloc_flex_gd(flexbg_size);
1662 	if (flex_gd == NULL) {
1663 		err = -ENOMEM;
1664 		goto out;
1665 	}
1666 
1667 	/* Add flex groups. Note that a regular group is a
1668 	 * flex group with 1 group.
1669 	 */
1670 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1671 					      flexbg_size)) {
1672 		ext4_alloc_group_tables(sb, flex_gd, flexbg_size);
1673 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1674 		if (unlikely(err))
1675 			break;
1676 	}
1677 
1678 out:
1679 	if (flex_gd)
1680 		free_flex_gd(flex_gd);
1681 
1682 	iput(resize_inode);
1683 	if (test_opt(sb, DEBUG))
1684 		ext4_msg(sb, KERN_DEBUG, "resized filesystem from %llu "
1685 		       "upto %llu blocks", o_blocks_count, n_blocks_count);
1686 	return err;
1687 }
1688