xref: /linux/fs/ext4/resize.c (revision c5d3cdad688ed75fb311a3a671eb30ba7106d7d3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/resize.c
4  *
5  * Support for resizing an ext4 filesystem while it is mounted.
6  *
7  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8  *
9  * This could probably be made into a module, because it is not often in use.
10  */
11 
12 
13 #define EXT4FS_DEBUG
14 
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17 
18 #include "ext4_jbd2.h"
19 
20 struct ext4_rcu_ptr {
21 	struct rcu_head rcu;
22 	void *ptr;
23 };
24 
25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
26 {
27 	struct ext4_rcu_ptr *ptr;
28 
29 	ptr = container_of(head, struct ext4_rcu_ptr, rcu);
30 	kvfree(ptr->ptr);
31 	kfree(ptr);
32 }
33 
34 void ext4_kvfree_array_rcu(void *to_free)
35 {
36 	struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
37 
38 	if (ptr) {
39 		ptr->ptr = to_free;
40 		call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
41 		return;
42 	}
43 	synchronize_rcu();
44 	kvfree(to_free);
45 }
46 
47 int ext4_resize_begin(struct super_block *sb)
48 {
49 	struct ext4_sb_info *sbi = EXT4_SB(sb);
50 	int ret = 0;
51 
52 	if (!capable(CAP_SYS_RESOURCE))
53 		return -EPERM;
54 
55 	/*
56 	 * If we are not using the primary superblock/GDT copy don't resize,
57          * because the user tools have no way of handling this.  Probably a
58          * bad time to do it anyways.
59          */
60 	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
61 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
62 		ext4_warning(sb, "won't resize using backup superblock at %llu",
63 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
64 		return -EPERM;
65 	}
66 
67 	/*
68 	 * We are not allowed to do online-resizing on a filesystem mounted
69 	 * with error, because it can destroy the filesystem easily.
70 	 */
71 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
72 		ext4_warning(sb, "There are errors in the filesystem, "
73 			     "so online resizing is not allowed");
74 		return -EPERM;
75 	}
76 
77 	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
78 				  &EXT4_SB(sb)->s_ext4_flags))
79 		ret = -EBUSY;
80 
81 	return ret;
82 }
83 
84 void ext4_resize_end(struct super_block *sb)
85 {
86 	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
87 	smp_mb__after_atomic();
88 }
89 
90 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
91 					     ext4_group_t group) {
92 	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
93 	       EXT4_DESC_PER_BLOCK_BITS(sb);
94 }
95 
96 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
97 					     ext4_group_t group) {
98 	group = ext4_meta_bg_first_group(sb, group);
99 	return ext4_group_first_block_no(sb, group);
100 }
101 
102 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
103 						ext4_group_t group) {
104 	ext4_grpblk_t overhead;
105 	overhead = ext4_bg_num_gdb(sb, group);
106 	if (ext4_bg_has_super(sb, group))
107 		overhead += 1 +
108 			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
109 	return overhead;
110 }
111 
112 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
113 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
114 
115 static int verify_group_input(struct super_block *sb,
116 			      struct ext4_new_group_data *input)
117 {
118 	struct ext4_sb_info *sbi = EXT4_SB(sb);
119 	struct ext4_super_block *es = sbi->s_es;
120 	ext4_fsblk_t start = ext4_blocks_count(es);
121 	ext4_fsblk_t end = start + input->blocks_count;
122 	ext4_group_t group = input->group;
123 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
124 	unsigned overhead;
125 	ext4_fsblk_t metaend;
126 	struct buffer_head *bh = NULL;
127 	ext4_grpblk_t free_blocks_count, offset;
128 	int err = -EINVAL;
129 
130 	if (group != sbi->s_groups_count) {
131 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
132 			     input->group, sbi->s_groups_count);
133 		return -EINVAL;
134 	}
135 
136 	overhead = ext4_group_overhead_blocks(sb, group);
137 	metaend = start + overhead;
138 	input->free_clusters_count = free_blocks_count =
139 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
140 
141 	if (test_opt(sb, DEBUG))
142 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
143 		       "(%d free, %u reserved)\n",
144 		       ext4_bg_has_super(sb, input->group) ? "normal" :
145 		       "no-super", input->group, input->blocks_count,
146 		       free_blocks_count, input->reserved_blocks);
147 
148 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
149 	if (offset != 0)
150 			ext4_warning(sb, "Last group not full");
151 	else if (input->reserved_blocks > input->blocks_count / 5)
152 		ext4_warning(sb, "Reserved blocks too high (%u)",
153 			     input->reserved_blocks);
154 	else if (free_blocks_count < 0)
155 		ext4_warning(sb, "Bad blocks count %u",
156 			     input->blocks_count);
157 	else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
158 		err = PTR_ERR(bh);
159 		bh = NULL;
160 		ext4_warning(sb, "Cannot read last block (%llu)",
161 			     end - 1);
162 	} else if (outside(input->block_bitmap, start, end))
163 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
164 			     (unsigned long long)input->block_bitmap);
165 	else if (outside(input->inode_bitmap, start, end))
166 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
167 			     (unsigned long long)input->inode_bitmap);
168 	else if (outside(input->inode_table, start, end) ||
169 		 outside(itend - 1, start, end))
170 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
171 			     (unsigned long long)input->inode_table, itend - 1);
172 	else if (input->inode_bitmap == input->block_bitmap)
173 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
174 			     (unsigned long long)input->block_bitmap);
175 	else if (inside(input->block_bitmap, input->inode_table, itend))
176 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
177 			     "(%llu-%llu)",
178 			     (unsigned long long)input->block_bitmap,
179 			     (unsigned long long)input->inode_table, itend - 1);
180 	else if (inside(input->inode_bitmap, input->inode_table, itend))
181 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
182 			     "(%llu-%llu)",
183 			     (unsigned long long)input->inode_bitmap,
184 			     (unsigned long long)input->inode_table, itend - 1);
185 	else if (inside(input->block_bitmap, start, metaend))
186 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
187 			     (unsigned long long)input->block_bitmap,
188 			     start, metaend - 1);
189 	else if (inside(input->inode_bitmap, start, metaend))
190 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
191 			     (unsigned long long)input->inode_bitmap,
192 			     start, metaend - 1);
193 	else if (inside(input->inode_table, start, metaend) ||
194 		 inside(itend - 1, start, metaend))
195 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
196 			     "(%llu-%llu)",
197 			     (unsigned long long)input->inode_table,
198 			     itend - 1, start, metaend - 1);
199 	else
200 		err = 0;
201 	brelse(bh);
202 
203 	return err;
204 }
205 
206 /*
207  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
208  * group each time.
209  */
210 struct ext4_new_flex_group_data {
211 	struct ext4_new_group_data *groups;	/* new_group_data for groups
212 						   in the flex group */
213 	__u16 *bg_flags;			/* block group flags of groups
214 						   in @groups */
215 	ext4_group_t count;			/* number of groups in @groups
216 						 */
217 };
218 
219 /*
220  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
221  * @flexbg_size.
222  *
223  * Returns NULL on failure otherwise address of the allocated structure.
224  */
225 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
226 {
227 	struct ext4_new_flex_group_data *flex_gd;
228 
229 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
230 	if (flex_gd == NULL)
231 		goto out3;
232 
233 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
234 		goto out2;
235 	flex_gd->count = flexbg_size;
236 
237 	flex_gd->groups = kmalloc_array(flexbg_size,
238 					sizeof(struct ext4_new_group_data),
239 					GFP_NOFS);
240 	if (flex_gd->groups == NULL)
241 		goto out2;
242 
243 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
244 					  GFP_NOFS);
245 	if (flex_gd->bg_flags == NULL)
246 		goto out1;
247 
248 	return flex_gd;
249 
250 out1:
251 	kfree(flex_gd->groups);
252 out2:
253 	kfree(flex_gd);
254 out3:
255 	return NULL;
256 }
257 
258 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
259 {
260 	kfree(flex_gd->bg_flags);
261 	kfree(flex_gd->groups);
262 	kfree(flex_gd);
263 }
264 
265 /*
266  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
267  * and inode tables for a flex group.
268  *
269  * This function is used by 64bit-resize.  Note that this function allocates
270  * group tables from the 1st group of groups contained by @flexgd, which may
271  * be a partial of a flex group.
272  *
273  * @sb: super block of fs to which the groups belongs
274  *
275  * Returns 0 on a successful allocation of the metadata blocks in the
276  * block group.
277  */
278 static int ext4_alloc_group_tables(struct super_block *sb,
279 				struct ext4_new_flex_group_data *flex_gd,
280 				int flexbg_size)
281 {
282 	struct ext4_new_group_data *group_data = flex_gd->groups;
283 	ext4_fsblk_t start_blk;
284 	ext4_fsblk_t last_blk;
285 	ext4_group_t src_group;
286 	ext4_group_t bb_index = 0;
287 	ext4_group_t ib_index = 0;
288 	ext4_group_t it_index = 0;
289 	ext4_group_t group;
290 	ext4_group_t last_group;
291 	unsigned overhead;
292 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
293 	int i;
294 
295 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
296 
297 	src_group = group_data[0].group;
298 	last_group  = src_group + flex_gd->count - 1;
299 
300 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
301 	       (last_group & ~(flexbg_size - 1))));
302 next_group:
303 	group = group_data[0].group;
304 	if (src_group >= group_data[0].group + flex_gd->count)
305 		return -ENOSPC;
306 	start_blk = ext4_group_first_block_no(sb, src_group);
307 	last_blk = start_blk + group_data[src_group - group].blocks_count;
308 
309 	overhead = ext4_group_overhead_blocks(sb, src_group);
310 
311 	start_blk += overhead;
312 
313 	/* We collect contiguous blocks as much as possible. */
314 	src_group++;
315 	for (; src_group <= last_group; src_group++) {
316 		overhead = ext4_group_overhead_blocks(sb, src_group);
317 		if (overhead == 0)
318 			last_blk += group_data[src_group - group].blocks_count;
319 		else
320 			break;
321 	}
322 
323 	/* Allocate block bitmaps */
324 	for (; bb_index < flex_gd->count; bb_index++) {
325 		if (start_blk >= last_blk)
326 			goto next_group;
327 		group_data[bb_index].block_bitmap = start_blk++;
328 		group = ext4_get_group_number(sb, start_blk - 1);
329 		group -= group_data[0].group;
330 		group_data[group].mdata_blocks++;
331 		flex_gd->bg_flags[group] &= uninit_mask;
332 	}
333 
334 	/* Allocate inode bitmaps */
335 	for (; ib_index < flex_gd->count; ib_index++) {
336 		if (start_blk >= last_blk)
337 			goto next_group;
338 		group_data[ib_index].inode_bitmap = start_blk++;
339 		group = ext4_get_group_number(sb, start_blk - 1);
340 		group -= group_data[0].group;
341 		group_data[group].mdata_blocks++;
342 		flex_gd->bg_flags[group] &= uninit_mask;
343 	}
344 
345 	/* Allocate inode tables */
346 	for (; it_index < flex_gd->count; it_index++) {
347 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
348 		ext4_fsblk_t next_group_start;
349 
350 		if (start_blk + itb > last_blk)
351 			goto next_group;
352 		group_data[it_index].inode_table = start_blk;
353 		group = ext4_get_group_number(sb, start_blk);
354 		next_group_start = ext4_group_first_block_no(sb, group + 1);
355 		group -= group_data[0].group;
356 
357 		if (start_blk + itb > next_group_start) {
358 			flex_gd->bg_flags[group + 1] &= uninit_mask;
359 			overhead = start_blk + itb - next_group_start;
360 			group_data[group + 1].mdata_blocks += overhead;
361 			itb -= overhead;
362 		}
363 
364 		group_data[group].mdata_blocks += itb;
365 		flex_gd->bg_flags[group] &= uninit_mask;
366 		start_blk += EXT4_SB(sb)->s_itb_per_group;
367 	}
368 
369 	/* Update free clusters count to exclude metadata blocks */
370 	for (i = 0; i < flex_gd->count; i++) {
371 		group_data[i].free_clusters_count -=
372 				EXT4_NUM_B2C(EXT4_SB(sb),
373 					     group_data[i].mdata_blocks);
374 	}
375 
376 	if (test_opt(sb, DEBUG)) {
377 		int i;
378 		group = group_data[0].group;
379 
380 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
381 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
382 		       flexbg_size);
383 
384 		for (i = 0; i < flex_gd->count; i++) {
385 			ext4_debug(
386 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
387 			       ext4_bg_has_super(sb, group + i) ? "normal" :
388 			       "no-super", group + i,
389 			       group_data[i].blocks_count,
390 			       group_data[i].free_clusters_count,
391 			       group_data[i].mdata_blocks);
392 		}
393 	}
394 	return 0;
395 }
396 
397 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
398 				  ext4_fsblk_t blk)
399 {
400 	struct buffer_head *bh;
401 	int err;
402 
403 	bh = sb_getblk(sb, blk);
404 	if (unlikely(!bh))
405 		return ERR_PTR(-ENOMEM);
406 	BUFFER_TRACE(bh, "get_write_access");
407 	if ((err = ext4_journal_get_write_access(handle, bh))) {
408 		brelse(bh);
409 		bh = ERR_PTR(err);
410 	} else {
411 		memset(bh->b_data, 0, sb->s_blocksize);
412 		set_buffer_uptodate(bh);
413 	}
414 
415 	return bh;
416 }
417 
418 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
419 {
420 	return ext4_journal_ensure_credits_fn(handle, credits,
421 		EXT4_MAX_TRANS_DATA, 0, 0);
422 }
423 
424 /*
425  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
426  *
427  * Helper function for ext4_setup_new_group_blocks() which set .
428  *
429  * @sb: super block
430  * @handle: journal handle
431  * @flex_gd: flex group data
432  */
433 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
434 			struct ext4_new_flex_group_data *flex_gd,
435 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
436 {
437 	struct ext4_sb_info *sbi = EXT4_SB(sb);
438 	ext4_group_t count = last_cluster - first_cluster + 1;
439 	ext4_group_t count2;
440 
441 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
442 		   last_cluster);
443 	for (count2 = count; count > 0;
444 	     count -= count2, first_cluster += count2) {
445 		ext4_fsblk_t start;
446 		struct buffer_head *bh;
447 		ext4_group_t group;
448 		int err;
449 
450 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
451 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
452 		group -= flex_gd->groups[0].group;
453 
454 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
455 		if (count2 > count)
456 			count2 = count;
457 
458 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
459 			BUG_ON(flex_gd->count > 1);
460 			continue;
461 		}
462 
463 		err = ext4_resize_ensure_credits_batch(handle, 1);
464 		if (err < 0)
465 			return err;
466 
467 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
468 		if (unlikely(!bh))
469 			return -ENOMEM;
470 
471 		BUFFER_TRACE(bh, "get_write_access");
472 		err = ext4_journal_get_write_access(handle, bh);
473 		if (err) {
474 			brelse(bh);
475 			return err;
476 		}
477 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
478 			   first_cluster, first_cluster - start, count2);
479 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
480 
481 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
482 		brelse(bh);
483 		if (unlikely(err))
484 			return err;
485 	}
486 
487 	return 0;
488 }
489 
490 /*
491  * Set up the block and inode bitmaps, and the inode table for the new groups.
492  * This doesn't need to be part of the main transaction, since we are only
493  * changing blocks outside the actual filesystem.  We still do journaling to
494  * ensure the recovery is correct in case of a failure just after resize.
495  * If any part of this fails, we simply abort the resize.
496  *
497  * setup_new_flex_group_blocks handles a flex group as follow:
498  *  1. copy super block and GDT, and initialize group tables if necessary.
499  *     In this step, we only set bits in blocks bitmaps for blocks taken by
500  *     super block and GDT.
501  *  2. allocate group tables in block bitmaps, that is, set bits in block
502  *     bitmap for blocks taken by group tables.
503  */
504 static int setup_new_flex_group_blocks(struct super_block *sb,
505 				struct ext4_new_flex_group_data *flex_gd)
506 {
507 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
508 	ext4_fsblk_t start;
509 	ext4_fsblk_t block;
510 	struct ext4_sb_info *sbi = EXT4_SB(sb);
511 	struct ext4_super_block *es = sbi->s_es;
512 	struct ext4_new_group_data *group_data = flex_gd->groups;
513 	__u16 *bg_flags = flex_gd->bg_flags;
514 	handle_t *handle;
515 	ext4_group_t group, count;
516 	struct buffer_head *bh = NULL;
517 	int reserved_gdb, i, j, err = 0, err2;
518 	int meta_bg;
519 
520 	BUG_ON(!flex_gd->count || !group_data ||
521 	       group_data[0].group != sbi->s_groups_count);
522 
523 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
524 	meta_bg = ext4_has_feature_meta_bg(sb);
525 
526 	/* This transaction may be extended/restarted along the way */
527 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
528 	if (IS_ERR(handle))
529 		return PTR_ERR(handle);
530 
531 	group = group_data[0].group;
532 	for (i = 0; i < flex_gd->count; i++, group++) {
533 		unsigned long gdblocks;
534 		ext4_grpblk_t overhead;
535 
536 		gdblocks = ext4_bg_num_gdb(sb, group);
537 		start = ext4_group_first_block_no(sb, group);
538 
539 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
540 			goto handle_itb;
541 
542 		if (meta_bg == 1) {
543 			ext4_group_t first_group;
544 			first_group = ext4_meta_bg_first_group(sb, group);
545 			if (first_group != group + 1 &&
546 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
547 				goto handle_itb;
548 		}
549 
550 		block = start + ext4_bg_has_super(sb, group);
551 		/* Copy all of the GDT blocks into the backup in this group */
552 		for (j = 0; j < gdblocks; j++, block++) {
553 			struct buffer_head *gdb;
554 
555 			ext4_debug("update backup group %#04llx\n", block);
556 			err = ext4_resize_ensure_credits_batch(handle, 1);
557 			if (err < 0)
558 				goto out;
559 
560 			gdb = sb_getblk(sb, block);
561 			if (unlikely(!gdb)) {
562 				err = -ENOMEM;
563 				goto out;
564 			}
565 
566 			BUFFER_TRACE(gdb, "get_write_access");
567 			err = ext4_journal_get_write_access(handle, gdb);
568 			if (err) {
569 				brelse(gdb);
570 				goto out;
571 			}
572 			memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
573 				s_group_desc, j)->b_data, gdb->b_size);
574 			set_buffer_uptodate(gdb);
575 
576 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
577 			if (unlikely(err)) {
578 				brelse(gdb);
579 				goto out;
580 			}
581 			brelse(gdb);
582 		}
583 
584 		/* Zero out all of the reserved backup group descriptor
585 		 * table blocks
586 		 */
587 		if (ext4_bg_has_super(sb, group)) {
588 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
589 					reserved_gdb, GFP_NOFS);
590 			if (err)
591 				goto out;
592 		}
593 
594 handle_itb:
595 		/* Initialize group tables of the grop @group */
596 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
597 			goto handle_bb;
598 
599 		/* Zero out all of the inode table blocks */
600 		block = group_data[i].inode_table;
601 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
602 			   block, sbi->s_itb_per_group);
603 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
604 				       GFP_NOFS);
605 		if (err)
606 			goto out;
607 
608 handle_bb:
609 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
610 			goto handle_ib;
611 
612 		/* Initialize block bitmap of the @group */
613 		block = group_data[i].block_bitmap;
614 		err = ext4_resize_ensure_credits_batch(handle, 1);
615 		if (err < 0)
616 			goto out;
617 
618 		bh = bclean(handle, sb, block);
619 		if (IS_ERR(bh)) {
620 			err = PTR_ERR(bh);
621 			goto out;
622 		}
623 		overhead = ext4_group_overhead_blocks(sb, group);
624 		if (overhead != 0) {
625 			ext4_debug("mark backup superblock %#04llx (+0)\n",
626 				   start);
627 			ext4_set_bits(bh->b_data, 0,
628 				      EXT4_NUM_B2C(sbi, overhead));
629 		}
630 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
631 				     sb->s_blocksize * 8, bh->b_data);
632 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
633 		brelse(bh);
634 		if (err)
635 			goto out;
636 
637 handle_ib:
638 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
639 			continue;
640 
641 		/* Initialize inode bitmap of the @group */
642 		block = group_data[i].inode_bitmap;
643 		err = ext4_resize_ensure_credits_batch(handle, 1);
644 		if (err < 0)
645 			goto out;
646 		/* Mark unused entries in inode bitmap used */
647 		bh = bclean(handle, sb, block);
648 		if (IS_ERR(bh)) {
649 			err = PTR_ERR(bh);
650 			goto out;
651 		}
652 
653 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
654 				     sb->s_blocksize * 8, bh->b_data);
655 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
656 		brelse(bh);
657 		if (err)
658 			goto out;
659 	}
660 
661 	/* Mark group tables in block bitmap */
662 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
663 		count = group_table_count[j];
664 		start = (&group_data[0].block_bitmap)[j];
665 		block = start;
666 		for (i = 1; i < flex_gd->count; i++) {
667 			block += group_table_count[j];
668 			if (block == (&group_data[i].block_bitmap)[j]) {
669 				count += group_table_count[j];
670 				continue;
671 			}
672 			err = set_flexbg_block_bitmap(sb, handle,
673 						      flex_gd,
674 						      EXT4_B2C(sbi, start),
675 						      EXT4_B2C(sbi,
676 							       start + count
677 							       - 1));
678 			if (err)
679 				goto out;
680 			count = group_table_count[j];
681 			start = (&group_data[i].block_bitmap)[j];
682 			block = start;
683 		}
684 
685 		if (count) {
686 			err = set_flexbg_block_bitmap(sb, handle,
687 						      flex_gd,
688 						      EXT4_B2C(sbi, start),
689 						      EXT4_B2C(sbi,
690 							       start + count
691 							       - 1));
692 			if (err)
693 				goto out;
694 		}
695 	}
696 
697 out:
698 	err2 = ext4_journal_stop(handle);
699 	if (err2 && !err)
700 		err = err2;
701 
702 	return err;
703 }
704 
705 /*
706  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
707  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
708  * calling this for the first time.  In a sparse filesystem it will be the
709  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
710  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
711  */
712 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
713 				  unsigned *five, unsigned *seven)
714 {
715 	unsigned *min = three;
716 	int mult = 3;
717 	unsigned ret;
718 
719 	if (!ext4_has_feature_sparse_super(sb)) {
720 		ret = *min;
721 		*min += 1;
722 		return ret;
723 	}
724 
725 	if (*five < *min) {
726 		min = five;
727 		mult = 5;
728 	}
729 	if (*seven < *min) {
730 		min = seven;
731 		mult = 7;
732 	}
733 
734 	ret = *min;
735 	*min *= mult;
736 
737 	return ret;
738 }
739 
740 /*
741  * Check that all of the backup GDT blocks are held in the primary GDT block.
742  * It is assumed that they are stored in group order.  Returns the number of
743  * groups in current filesystem that have BACKUPS, or -ve error code.
744  */
745 static int verify_reserved_gdb(struct super_block *sb,
746 			       ext4_group_t end,
747 			       struct buffer_head *primary)
748 {
749 	const ext4_fsblk_t blk = primary->b_blocknr;
750 	unsigned three = 1;
751 	unsigned five = 5;
752 	unsigned seven = 7;
753 	unsigned grp;
754 	__le32 *p = (__le32 *)primary->b_data;
755 	int gdbackups = 0;
756 
757 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
758 		if (le32_to_cpu(*p++) !=
759 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
760 			ext4_warning(sb, "reserved GDT %llu"
761 				     " missing grp %d (%llu)",
762 				     blk, grp,
763 				     grp *
764 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
765 				     blk);
766 			return -EINVAL;
767 		}
768 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
769 			return -EFBIG;
770 	}
771 
772 	return gdbackups;
773 }
774 
775 /*
776  * Called when we need to bring a reserved group descriptor table block into
777  * use from the resize inode.  The primary copy of the new GDT block currently
778  * is an indirect block (under the double indirect block in the resize inode).
779  * The new backup GDT blocks will be stored as leaf blocks in this indirect
780  * block, in group order.  Even though we know all the block numbers we need,
781  * we check to ensure that the resize inode has actually reserved these blocks.
782  *
783  * Don't need to update the block bitmaps because the blocks are still in use.
784  *
785  * We get all of the error cases out of the way, so that we are sure to not
786  * fail once we start modifying the data on disk, because JBD has no rollback.
787  */
788 static int add_new_gdb(handle_t *handle, struct inode *inode,
789 		       ext4_group_t group)
790 {
791 	struct super_block *sb = inode->i_sb;
792 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
793 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
794 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
795 	struct buffer_head **o_group_desc, **n_group_desc = NULL;
796 	struct buffer_head *dind = NULL;
797 	struct buffer_head *gdb_bh = NULL;
798 	int gdbackups;
799 	struct ext4_iloc iloc = { .bh = NULL };
800 	__le32 *data;
801 	int err;
802 
803 	if (test_opt(sb, DEBUG))
804 		printk(KERN_DEBUG
805 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
806 		       gdb_num);
807 
808 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
809 	if (IS_ERR(gdb_bh))
810 		return PTR_ERR(gdb_bh);
811 
812 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
813 	if (gdbackups < 0) {
814 		err = gdbackups;
815 		goto errout;
816 	}
817 
818 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
819 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
820 	if (IS_ERR(dind)) {
821 		err = PTR_ERR(dind);
822 		dind = NULL;
823 		goto errout;
824 	}
825 
826 	data = (__le32 *)dind->b_data;
827 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
828 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
829 			     group, gdblock);
830 		err = -EINVAL;
831 		goto errout;
832 	}
833 
834 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
835 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
836 	if (unlikely(err))
837 		goto errout;
838 
839 	BUFFER_TRACE(gdb_bh, "get_write_access");
840 	err = ext4_journal_get_write_access(handle, gdb_bh);
841 	if (unlikely(err))
842 		goto errout;
843 
844 	BUFFER_TRACE(dind, "get_write_access");
845 	err = ext4_journal_get_write_access(handle, dind);
846 	if (unlikely(err))
847 		ext4_std_error(sb, err);
848 
849 	/* ext4_reserve_inode_write() gets a reference on the iloc */
850 	err = ext4_reserve_inode_write(handle, inode, &iloc);
851 	if (unlikely(err))
852 		goto errout;
853 
854 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
855 				GFP_KERNEL);
856 	if (!n_group_desc) {
857 		err = -ENOMEM;
858 		ext4_warning(sb, "not enough memory for %lu groups",
859 			     gdb_num + 1);
860 		goto errout;
861 	}
862 
863 	/*
864 	 * Finally, we have all of the possible failures behind us...
865 	 *
866 	 * Remove new GDT block from inode double-indirect block and clear out
867 	 * the new GDT block for use (which also "frees" the backup GDT blocks
868 	 * from the reserved inode).  We don't need to change the bitmaps for
869 	 * these blocks, because they are marked as in-use from being in the
870 	 * reserved inode, and will become GDT blocks (primary and backup).
871 	 */
872 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
873 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
874 	if (unlikely(err)) {
875 		ext4_std_error(sb, err);
876 		goto errout;
877 	}
878 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
879 			   (9 - EXT4_SB(sb)->s_cluster_bits);
880 	ext4_mark_iloc_dirty(handle, inode, &iloc);
881 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
882 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
883 	if (unlikely(err)) {
884 		ext4_std_error(sb, err);
885 		iloc.bh = NULL;
886 		goto errout;
887 	}
888 	brelse(dind);
889 
890 	rcu_read_lock();
891 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
892 	memcpy(n_group_desc, o_group_desc,
893 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
894 	rcu_read_unlock();
895 	n_group_desc[gdb_num] = gdb_bh;
896 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
897 	EXT4_SB(sb)->s_gdb_count++;
898 	ext4_kvfree_array_rcu(o_group_desc);
899 
900 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
901 	err = ext4_handle_dirty_super(handle, sb);
902 	if (err)
903 		ext4_std_error(sb, err);
904 	return err;
905 errout:
906 	kvfree(n_group_desc);
907 	brelse(iloc.bh);
908 	brelse(dind);
909 	brelse(gdb_bh);
910 
911 	ext4_debug("leaving with error %d\n", err);
912 	return err;
913 }
914 
915 /*
916  * add_new_gdb_meta_bg is the sister of add_new_gdb.
917  */
918 static int add_new_gdb_meta_bg(struct super_block *sb,
919 			       handle_t *handle, ext4_group_t group) {
920 	ext4_fsblk_t gdblock;
921 	struct buffer_head *gdb_bh;
922 	struct buffer_head **o_group_desc, **n_group_desc;
923 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
924 	int err;
925 
926 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
927 		   ext4_bg_has_super(sb, group);
928 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
929 	if (IS_ERR(gdb_bh))
930 		return PTR_ERR(gdb_bh);
931 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
932 				GFP_KERNEL);
933 	if (!n_group_desc) {
934 		brelse(gdb_bh);
935 		err = -ENOMEM;
936 		ext4_warning(sb, "not enough memory for %lu groups",
937 			     gdb_num + 1);
938 		return err;
939 	}
940 
941 	rcu_read_lock();
942 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
943 	memcpy(n_group_desc, o_group_desc,
944 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
945 	rcu_read_unlock();
946 	n_group_desc[gdb_num] = gdb_bh;
947 
948 	BUFFER_TRACE(gdb_bh, "get_write_access");
949 	err = ext4_journal_get_write_access(handle, gdb_bh);
950 	if (err) {
951 		kvfree(n_group_desc);
952 		brelse(gdb_bh);
953 		return err;
954 	}
955 
956 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
957 	EXT4_SB(sb)->s_gdb_count++;
958 	ext4_kvfree_array_rcu(o_group_desc);
959 	return err;
960 }
961 
962 /*
963  * Called when we are adding a new group which has a backup copy of each of
964  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
965  * We need to add these reserved backup GDT blocks to the resize inode, so
966  * that they are kept for future resizing and not allocated to files.
967  *
968  * Each reserved backup GDT block will go into a different indirect block.
969  * The indirect blocks are actually the primary reserved GDT blocks,
970  * so we know in advance what their block numbers are.  We only get the
971  * double-indirect block to verify it is pointing to the primary reserved
972  * GDT blocks so we don't overwrite a data block by accident.  The reserved
973  * backup GDT blocks are stored in their reserved primary GDT block.
974  */
975 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
976 			      ext4_group_t group)
977 {
978 	struct super_block *sb = inode->i_sb;
979 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
980 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
981 	struct buffer_head **primary;
982 	struct buffer_head *dind;
983 	struct ext4_iloc iloc;
984 	ext4_fsblk_t blk;
985 	__le32 *data, *end;
986 	int gdbackups = 0;
987 	int res, i;
988 	int err;
989 
990 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
991 	if (!primary)
992 		return -ENOMEM;
993 
994 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
995 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
996 	if (IS_ERR(dind)) {
997 		err = PTR_ERR(dind);
998 		dind = NULL;
999 		goto exit_free;
1000 	}
1001 
1002 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1003 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1004 					 EXT4_ADDR_PER_BLOCK(sb));
1005 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1006 
1007 	/* Get each reserved primary GDT block and verify it holds backups */
1008 	for (res = 0; res < reserved_gdb; res++, blk++) {
1009 		if (le32_to_cpu(*data) != blk) {
1010 			ext4_warning(sb, "reserved block %llu"
1011 				     " not at offset %ld",
1012 				     blk,
1013 				     (long)(data - (__le32 *)dind->b_data));
1014 			err = -EINVAL;
1015 			goto exit_bh;
1016 		}
1017 		primary[res] = ext4_sb_bread(sb, blk, 0);
1018 		if (IS_ERR(primary[res])) {
1019 			err = PTR_ERR(primary[res]);
1020 			primary[res] = NULL;
1021 			goto exit_bh;
1022 		}
1023 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1024 		if (gdbackups < 0) {
1025 			brelse(primary[res]);
1026 			err = gdbackups;
1027 			goto exit_bh;
1028 		}
1029 		if (++data >= end)
1030 			data = (__le32 *)dind->b_data;
1031 	}
1032 
1033 	for (i = 0; i < reserved_gdb; i++) {
1034 		BUFFER_TRACE(primary[i], "get_write_access");
1035 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1036 			goto exit_bh;
1037 	}
1038 
1039 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1040 		goto exit_bh;
1041 
1042 	/*
1043 	 * Finally we can add each of the reserved backup GDT blocks from
1044 	 * the new group to its reserved primary GDT block.
1045 	 */
1046 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1047 	for (i = 0; i < reserved_gdb; i++) {
1048 		int err2;
1049 		data = (__le32 *)primary[i]->b_data;
1050 		/* printk("reserving backup %lu[%u] = %lu\n",
1051 		       primary[i]->b_blocknr, gdbackups,
1052 		       blk + primary[i]->b_blocknr); */
1053 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1054 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1055 		if (!err)
1056 			err = err2;
1057 	}
1058 
1059 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1060 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1061 
1062 exit_bh:
1063 	while (--res >= 0)
1064 		brelse(primary[res]);
1065 	brelse(dind);
1066 
1067 exit_free:
1068 	kfree(primary);
1069 
1070 	return err;
1071 }
1072 
1073 /*
1074  * Update the backup copies of the ext4 metadata.  These don't need to be part
1075  * of the main resize transaction, because e2fsck will re-write them if there
1076  * is a problem (basically only OOM will cause a problem).  However, we
1077  * _should_ update the backups if possible, in case the primary gets trashed
1078  * for some reason and we need to run e2fsck from a backup superblock.  The
1079  * important part is that the new block and inode counts are in the backup
1080  * superblocks, and the location of the new group metadata in the GDT backups.
1081  *
1082  * We do not need take the s_resize_lock for this, because these
1083  * blocks are not otherwise touched by the filesystem code when it is
1084  * mounted.  We don't need to worry about last changing from
1085  * sbi->s_groups_count, because the worst that can happen is that we
1086  * do not copy the full number of backups at this time.  The resize
1087  * which changed s_groups_count will backup again.
1088  */
1089 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1090 			   int size, int meta_bg)
1091 {
1092 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1093 	ext4_group_t last;
1094 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1095 	unsigned three = 1;
1096 	unsigned five = 5;
1097 	unsigned seven = 7;
1098 	ext4_group_t group = 0;
1099 	int rest = sb->s_blocksize - size;
1100 	handle_t *handle;
1101 	int err = 0, err2;
1102 
1103 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1104 	if (IS_ERR(handle)) {
1105 		group = 1;
1106 		err = PTR_ERR(handle);
1107 		goto exit_err;
1108 	}
1109 
1110 	if (meta_bg == 0) {
1111 		group = ext4_list_backups(sb, &three, &five, &seven);
1112 		last = sbi->s_groups_count;
1113 	} else {
1114 		group = ext4_get_group_number(sb, blk_off) + 1;
1115 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1116 	}
1117 
1118 	while (group < sbi->s_groups_count) {
1119 		struct buffer_head *bh;
1120 		ext4_fsblk_t backup_block;
1121 
1122 		/* Out of journal space, and can't get more - abort - so sad */
1123 		err = ext4_resize_ensure_credits_batch(handle, 1);
1124 		if (err < 0)
1125 			break;
1126 
1127 		if (meta_bg == 0)
1128 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1129 		else
1130 			backup_block = (ext4_group_first_block_no(sb, group) +
1131 					ext4_bg_has_super(sb, group));
1132 
1133 		bh = sb_getblk(sb, backup_block);
1134 		if (unlikely(!bh)) {
1135 			err = -ENOMEM;
1136 			break;
1137 		}
1138 		ext4_debug("update metadata backup %llu(+%llu)\n",
1139 			   backup_block, backup_block -
1140 			   ext4_group_first_block_no(sb, group));
1141 		BUFFER_TRACE(bh, "get_write_access");
1142 		if ((err = ext4_journal_get_write_access(handle, bh))) {
1143 			brelse(bh);
1144 			break;
1145 		}
1146 		lock_buffer(bh);
1147 		memcpy(bh->b_data, data, size);
1148 		if (rest)
1149 			memset(bh->b_data + size, 0, rest);
1150 		set_buffer_uptodate(bh);
1151 		unlock_buffer(bh);
1152 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1153 		if (unlikely(err))
1154 			ext4_std_error(sb, err);
1155 		brelse(bh);
1156 
1157 		if (meta_bg == 0)
1158 			group = ext4_list_backups(sb, &three, &five, &seven);
1159 		else if (group == last)
1160 			break;
1161 		else
1162 			group = last;
1163 	}
1164 	if ((err2 = ext4_journal_stop(handle)) && !err)
1165 		err = err2;
1166 
1167 	/*
1168 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1169 	 * late to revert the resize, we shouldn't fail just because of
1170 	 * the backup copies (they are only needed in case of corruption).
1171 	 *
1172 	 * However, if we got here we have a journal problem too, so we
1173 	 * can't really start a transaction to mark the superblock.
1174 	 * Chicken out and just set the flag on the hope it will be written
1175 	 * to disk, and if not - we will simply wait until next fsck.
1176 	 */
1177 exit_err:
1178 	if (err) {
1179 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1180 			     "forcing fsck on next reboot", group, err);
1181 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1182 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1183 		mark_buffer_dirty(sbi->s_sbh);
1184 	}
1185 }
1186 
1187 /*
1188  * ext4_add_new_descs() adds @count group descriptor of groups
1189  * starting at @group
1190  *
1191  * @handle: journal handle
1192  * @sb: super block
1193  * @group: the group no. of the first group desc to be added
1194  * @resize_inode: the resize inode
1195  * @count: number of group descriptors to be added
1196  */
1197 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1198 			      ext4_group_t group, struct inode *resize_inode,
1199 			      ext4_group_t count)
1200 {
1201 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1202 	struct ext4_super_block *es = sbi->s_es;
1203 	struct buffer_head *gdb_bh;
1204 	int i, gdb_off, gdb_num, err = 0;
1205 	int meta_bg;
1206 
1207 	meta_bg = ext4_has_feature_meta_bg(sb);
1208 	for (i = 0; i < count; i++, group++) {
1209 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1210 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1211 
1212 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1213 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1214 
1215 		/*
1216 		 * We will only either add reserved group blocks to a backup group
1217 		 * or remove reserved blocks for the first group in a new group block.
1218 		 * Doing both would be mean more complex code, and sane people don't
1219 		 * use non-sparse filesystems anymore.  This is already checked above.
1220 		 */
1221 		if (gdb_off) {
1222 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1223 						     gdb_num);
1224 			BUFFER_TRACE(gdb_bh, "get_write_access");
1225 			err = ext4_journal_get_write_access(handle, gdb_bh);
1226 
1227 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1228 				err = reserve_backup_gdb(handle, resize_inode, group);
1229 		} else if (meta_bg != 0) {
1230 			err = add_new_gdb_meta_bg(sb, handle, group);
1231 		} else {
1232 			err = add_new_gdb(handle, resize_inode, group);
1233 		}
1234 		if (err)
1235 			break;
1236 	}
1237 	return err;
1238 }
1239 
1240 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1241 {
1242 	struct buffer_head *bh = sb_getblk(sb, block);
1243 	if (unlikely(!bh))
1244 		return NULL;
1245 	if (!bh_uptodate_or_lock(bh)) {
1246 		if (bh_submit_read(bh) < 0) {
1247 			brelse(bh);
1248 			return NULL;
1249 		}
1250 	}
1251 
1252 	return bh;
1253 }
1254 
1255 static int ext4_set_bitmap_checksums(struct super_block *sb,
1256 				     ext4_group_t group,
1257 				     struct ext4_group_desc *gdp,
1258 				     struct ext4_new_group_data *group_data)
1259 {
1260 	struct buffer_head *bh;
1261 
1262 	if (!ext4_has_metadata_csum(sb))
1263 		return 0;
1264 
1265 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1266 	if (!bh)
1267 		return -EIO;
1268 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1269 				   EXT4_INODES_PER_GROUP(sb) / 8);
1270 	brelse(bh);
1271 
1272 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1273 	if (!bh)
1274 		return -EIO;
1275 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1276 	brelse(bh);
1277 
1278 	return 0;
1279 }
1280 
1281 /*
1282  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1283  */
1284 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1285 				struct ext4_new_flex_group_data *flex_gd)
1286 {
1287 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1288 	struct ext4_group_desc		*gdp;
1289 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1290 	struct buffer_head		*gdb_bh;
1291 	ext4_group_t			group;
1292 	__u16				*bg_flags = flex_gd->bg_flags;
1293 	int				i, gdb_off, gdb_num, err = 0;
1294 
1295 
1296 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1297 		group = group_data->group;
1298 
1299 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1300 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1301 
1302 		/*
1303 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1304 		 */
1305 		gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1306 		/* Update group descriptor block for new group */
1307 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1308 						 gdb_off * EXT4_DESC_SIZE(sb));
1309 
1310 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1311 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1312 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1313 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1314 		if (err) {
1315 			ext4_std_error(sb, err);
1316 			break;
1317 		}
1318 
1319 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1320 		ext4_free_group_clusters_set(sb, gdp,
1321 					     group_data->free_clusters_count);
1322 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1323 		if (ext4_has_group_desc_csum(sb))
1324 			ext4_itable_unused_set(sb, gdp,
1325 					       EXT4_INODES_PER_GROUP(sb));
1326 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1327 		ext4_group_desc_csum_set(sb, group, gdp);
1328 
1329 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1330 		if (unlikely(err)) {
1331 			ext4_std_error(sb, err);
1332 			break;
1333 		}
1334 
1335 		/*
1336 		 * We can allocate memory for mb_alloc based on the new group
1337 		 * descriptor
1338 		 */
1339 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1340 		if (err)
1341 			break;
1342 	}
1343 	return err;
1344 }
1345 
1346 /*
1347  * ext4_update_super() updates the super block so that the newly added
1348  * groups can be seen by the filesystem.
1349  *
1350  * @sb: super block
1351  * @flex_gd: new added groups
1352  */
1353 static void ext4_update_super(struct super_block *sb,
1354 			     struct ext4_new_flex_group_data *flex_gd)
1355 {
1356 	ext4_fsblk_t blocks_count = 0;
1357 	ext4_fsblk_t free_blocks = 0;
1358 	ext4_fsblk_t reserved_blocks = 0;
1359 	struct ext4_new_group_data *group_data = flex_gd->groups;
1360 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1361 	struct ext4_super_block *es = sbi->s_es;
1362 	int i;
1363 
1364 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1365 	/*
1366 	 * Make the new blocks and inodes valid next.  We do this before
1367 	 * increasing the group count so that once the group is enabled,
1368 	 * all of its blocks and inodes are already valid.
1369 	 *
1370 	 * We always allocate group-by-group, then block-by-block or
1371 	 * inode-by-inode within a group, so enabling these
1372 	 * blocks/inodes before the group is live won't actually let us
1373 	 * allocate the new space yet.
1374 	 */
1375 	for (i = 0; i < flex_gd->count; i++) {
1376 		blocks_count += group_data[i].blocks_count;
1377 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1378 	}
1379 
1380 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1381 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1382 	reserved_blocks *= blocks_count;
1383 	do_div(reserved_blocks, 100);
1384 
1385 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1386 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1387 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1388 		     flex_gd->count);
1389 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1390 		     flex_gd->count);
1391 
1392 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1393 	/*
1394 	 * We need to protect s_groups_count against other CPUs seeing
1395 	 * inconsistent state in the superblock.
1396 	 *
1397 	 * The precise rules we use are:
1398 	 *
1399 	 * * Writers must perform a smp_wmb() after updating all
1400 	 *   dependent data and before modifying the groups count
1401 	 *
1402 	 * * Readers must perform an smp_rmb() after reading the groups
1403 	 *   count and before reading any dependent data.
1404 	 *
1405 	 * NB. These rules can be relaxed when checking the group count
1406 	 * while freeing data, as we can only allocate from a block
1407 	 * group after serialising against the group count, and we can
1408 	 * only then free after serialising in turn against that
1409 	 * allocation.
1410 	 */
1411 	smp_wmb();
1412 
1413 	/* Update the global fs size fields */
1414 	sbi->s_groups_count += flex_gd->count;
1415 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1416 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1417 
1418 	/* Update the reserved block counts only once the new group is
1419 	 * active. */
1420 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1421 				reserved_blocks);
1422 
1423 	/* Update the free space counts */
1424 	percpu_counter_add(&sbi->s_freeclusters_counter,
1425 			   EXT4_NUM_B2C(sbi, free_blocks));
1426 	percpu_counter_add(&sbi->s_freeinodes_counter,
1427 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1428 
1429 	ext4_debug("free blocks count %llu",
1430 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1431 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1432 		ext4_group_t flex_group;
1433 		struct flex_groups *fg;
1434 
1435 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1436 		fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1437 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1438 			     &fg->free_clusters);
1439 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1440 			   &fg->free_inodes);
1441 	}
1442 
1443 	/*
1444 	 * Update the fs overhead information
1445 	 */
1446 	ext4_calculate_overhead(sb);
1447 
1448 	if (test_opt(sb, DEBUG))
1449 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1450 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1451 		       blocks_count, free_blocks, reserved_blocks);
1452 }
1453 
1454 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1455  * _before_ we start modifying the filesystem, because we cannot abort the
1456  * transaction and not have it write the data to disk.
1457  */
1458 static int ext4_flex_group_add(struct super_block *sb,
1459 			       struct inode *resize_inode,
1460 			       struct ext4_new_flex_group_data *flex_gd)
1461 {
1462 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1463 	struct ext4_super_block *es = sbi->s_es;
1464 	ext4_fsblk_t o_blocks_count;
1465 	ext4_grpblk_t last;
1466 	ext4_group_t group;
1467 	handle_t *handle;
1468 	unsigned reserved_gdb;
1469 	int err = 0, err2 = 0, credit;
1470 
1471 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1472 
1473 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1474 	o_blocks_count = ext4_blocks_count(es);
1475 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1476 	BUG_ON(last);
1477 
1478 	err = setup_new_flex_group_blocks(sb, flex_gd);
1479 	if (err)
1480 		goto exit;
1481 	/*
1482 	 * We will always be modifying at least the superblock and  GDT
1483 	 * blocks.  If we are adding a group past the last current GDT block,
1484 	 * we will also modify the inode and the dindirect block.  If we
1485 	 * are adding a group with superblock/GDT backups  we will also
1486 	 * modify each of the reserved GDT dindirect blocks.
1487 	 */
1488 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1489 	/* GDT blocks */
1490 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1491 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1492 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1493 	if (IS_ERR(handle)) {
1494 		err = PTR_ERR(handle);
1495 		goto exit;
1496 	}
1497 
1498 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1499 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1500 	if (err)
1501 		goto exit_journal;
1502 
1503 	group = flex_gd->groups[0].group;
1504 	BUG_ON(group != sbi->s_groups_count);
1505 	err = ext4_add_new_descs(handle, sb, group,
1506 				resize_inode, flex_gd->count);
1507 	if (err)
1508 		goto exit_journal;
1509 
1510 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1511 	if (err)
1512 		goto exit_journal;
1513 
1514 	ext4_update_super(sb, flex_gd);
1515 
1516 	err = ext4_handle_dirty_super(handle, sb);
1517 
1518 exit_journal:
1519 	err2 = ext4_journal_stop(handle);
1520 	if (!err)
1521 		err = err2;
1522 
1523 	if (!err) {
1524 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1525 		int gdb_num_end = ((group + flex_gd->count - 1) /
1526 				   EXT4_DESC_PER_BLOCK(sb));
1527 		int meta_bg = ext4_has_feature_meta_bg(sb);
1528 		sector_t old_gdb = 0;
1529 
1530 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1531 			       sizeof(struct ext4_super_block), 0);
1532 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1533 			struct buffer_head *gdb_bh;
1534 
1535 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1536 						     gdb_num);
1537 			if (old_gdb == gdb_bh->b_blocknr)
1538 				continue;
1539 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1540 				       gdb_bh->b_size, meta_bg);
1541 			old_gdb = gdb_bh->b_blocknr;
1542 		}
1543 	}
1544 exit:
1545 	return err;
1546 }
1547 
1548 static int ext4_setup_next_flex_gd(struct super_block *sb,
1549 				    struct ext4_new_flex_group_data *flex_gd,
1550 				    ext4_fsblk_t n_blocks_count,
1551 				    unsigned long flexbg_size)
1552 {
1553 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1554 	struct ext4_super_block *es = sbi->s_es;
1555 	struct ext4_new_group_data *group_data = flex_gd->groups;
1556 	ext4_fsblk_t o_blocks_count;
1557 	ext4_group_t n_group;
1558 	ext4_group_t group;
1559 	ext4_group_t last_group;
1560 	ext4_grpblk_t last;
1561 	ext4_grpblk_t clusters_per_group;
1562 	unsigned long i;
1563 
1564 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1565 
1566 	o_blocks_count = ext4_blocks_count(es);
1567 
1568 	if (o_blocks_count == n_blocks_count)
1569 		return 0;
1570 
1571 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1572 	BUG_ON(last);
1573 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1574 
1575 	last_group = group | (flexbg_size - 1);
1576 	if (last_group > n_group)
1577 		last_group = n_group;
1578 
1579 	flex_gd->count = last_group - group + 1;
1580 
1581 	for (i = 0; i < flex_gd->count; i++) {
1582 		int overhead;
1583 
1584 		group_data[i].group = group + i;
1585 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1586 		overhead = ext4_group_overhead_blocks(sb, group + i);
1587 		group_data[i].mdata_blocks = overhead;
1588 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1589 		if (ext4_has_group_desc_csum(sb)) {
1590 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1591 					       EXT4_BG_INODE_UNINIT;
1592 			if (!test_opt(sb, INIT_INODE_TABLE))
1593 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1594 		} else
1595 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1596 	}
1597 
1598 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1599 		/* We need to initialize block bitmap of last group. */
1600 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1601 
1602 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1603 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1604 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1605 						       last - 1;
1606 	}
1607 
1608 	return 1;
1609 }
1610 
1611 /* Add group descriptor data to an existing or new group descriptor block.
1612  * Ensure we handle all possible error conditions _before_ we start modifying
1613  * the filesystem, because we cannot abort the transaction and not have it
1614  * write the data to disk.
1615  *
1616  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1617  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1618  *
1619  * We only need to hold the superblock lock while we are actually adding
1620  * in the new group's counts to the superblock.  Prior to that we have
1621  * not really "added" the group at all.  We re-check that we are still
1622  * adding in the last group in case things have changed since verifying.
1623  */
1624 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1625 {
1626 	struct ext4_new_flex_group_data flex_gd;
1627 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1628 	struct ext4_super_block *es = sbi->s_es;
1629 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1630 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1631 	struct inode *inode = NULL;
1632 	int gdb_off;
1633 	int err;
1634 	__u16 bg_flags = 0;
1635 
1636 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1637 
1638 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1639 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1640 		return -EPERM;
1641 	}
1642 
1643 	if (ext4_blocks_count(es) + input->blocks_count <
1644 	    ext4_blocks_count(es)) {
1645 		ext4_warning(sb, "blocks_count overflow");
1646 		return -EINVAL;
1647 	}
1648 
1649 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1650 	    le32_to_cpu(es->s_inodes_count)) {
1651 		ext4_warning(sb, "inodes_count overflow");
1652 		return -EINVAL;
1653 	}
1654 
1655 	if (reserved_gdb || gdb_off == 0) {
1656 		if (!ext4_has_feature_resize_inode(sb) ||
1657 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1658 			ext4_warning(sb,
1659 				     "No reserved GDT blocks, can't resize");
1660 			return -EPERM;
1661 		}
1662 		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1663 		if (IS_ERR(inode)) {
1664 			ext4_warning(sb, "Error opening resize inode");
1665 			return PTR_ERR(inode);
1666 		}
1667 	}
1668 
1669 
1670 	err = verify_group_input(sb, input);
1671 	if (err)
1672 		goto out;
1673 
1674 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1675 	if (err)
1676 		goto out;
1677 
1678 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1679 	if (err)
1680 		goto out;
1681 
1682 	flex_gd.count = 1;
1683 	flex_gd.groups = input;
1684 	flex_gd.bg_flags = &bg_flags;
1685 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1686 out:
1687 	iput(inode);
1688 	return err;
1689 } /* ext4_group_add */
1690 
1691 /*
1692  * extend a group without checking assuming that checking has been done.
1693  */
1694 static int ext4_group_extend_no_check(struct super_block *sb,
1695 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1696 {
1697 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1698 	handle_t *handle;
1699 	int err = 0, err2;
1700 
1701 	/* We will update the superblock, one block bitmap, and
1702 	 * one group descriptor via ext4_group_add_blocks().
1703 	 */
1704 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1705 	if (IS_ERR(handle)) {
1706 		err = PTR_ERR(handle);
1707 		ext4_warning(sb, "error %d on journal start", err);
1708 		return err;
1709 	}
1710 
1711 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1712 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1713 	if (err) {
1714 		ext4_warning(sb, "error %d on journal write access", err);
1715 		goto errout;
1716 	}
1717 
1718 	ext4_blocks_count_set(es, o_blocks_count + add);
1719 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1720 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1721 		   o_blocks_count + add);
1722 	/* We add the blocks to the bitmap and set the group need init bit */
1723 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1724 	if (err)
1725 		goto errout;
1726 	ext4_handle_dirty_super(handle, sb);
1727 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1728 		   o_blocks_count + add);
1729 errout:
1730 	err2 = ext4_journal_stop(handle);
1731 	if (err2 && !err)
1732 		err = err2;
1733 
1734 	if (!err) {
1735 		if (test_opt(sb, DEBUG))
1736 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1737 			       "blocks\n", ext4_blocks_count(es));
1738 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1739 			       (char *)es, sizeof(struct ext4_super_block), 0);
1740 	}
1741 	return err;
1742 }
1743 
1744 /*
1745  * Extend the filesystem to the new number of blocks specified.  This entry
1746  * point is only used to extend the current filesystem to the end of the last
1747  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1748  * for emergencies (because it has no dependencies on reserved blocks).
1749  *
1750  * If we _really_ wanted, we could use default values to call ext4_group_add()
1751  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1752  * GDT blocks are reserved to grow to the desired size.
1753  */
1754 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1755 		      ext4_fsblk_t n_blocks_count)
1756 {
1757 	ext4_fsblk_t o_blocks_count;
1758 	ext4_grpblk_t last;
1759 	ext4_grpblk_t add;
1760 	struct buffer_head *bh;
1761 	int err;
1762 	ext4_group_t group;
1763 
1764 	o_blocks_count = ext4_blocks_count(es);
1765 
1766 	if (test_opt(sb, DEBUG))
1767 		ext4_msg(sb, KERN_DEBUG,
1768 			 "extending last group from %llu to %llu blocks",
1769 			 o_blocks_count, n_blocks_count);
1770 
1771 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1772 		return 0;
1773 
1774 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1775 		ext4_msg(sb, KERN_ERR,
1776 			 "filesystem too large to resize to %llu blocks safely",
1777 			 n_blocks_count);
1778 		return -EINVAL;
1779 	}
1780 
1781 	if (n_blocks_count < o_blocks_count) {
1782 		ext4_warning(sb, "can't shrink FS - resize aborted");
1783 		return -EINVAL;
1784 	}
1785 
1786 	/* Handle the remaining blocks in the last group only. */
1787 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1788 
1789 	if (last == 0) {
1790 		ext4_warning(sb, "need to use ext2online to resize further");
1791 		return -EPERM;
1792 	}
1793 
1794 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1795 
1796 	if (o_blocks_count + add < o_blocks_count) {
1797 		ext4_warning(sb, "blocks_count overflow");
1798 		return -EINVAL;
1799 	}
1800 
1801 	if (o_blocks_count + add > n_blocks_count)
1802 		add = n_blocks_count - o_blocks_count;
1803 
1804 	if (o_blocks_count + add < n_blocks_count)
1805 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1806 			     o_blocks_count + add, add);
1807 
1808 	/* See if the device is actually as big as what was requested */
1809 	bh = sb_bread(sb, o_blocks_count + add - 1);
1810 	if (!bh) {
1811 		ext4_warning(sb, "can't read last block, resize aborted");
1812 		return -ENOSPC;
1813 	}
1814 	brelse(bh);
1815 
1816 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1817 	return err;
1818 } /* ext4_group_extend */
1819 
1820 
1821 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1822 {
1823 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1824 }
1825 
1826 /*
1827  * Release the resize inode and drop the resize_inode feature if there
1828  * are no more reserved gdt blocks, and then convert the file system
1829  * to enable meta_bg
1830  */
1831 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1832 {
1833 	handle_t *handle;
1834 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1835 	struct ext4_super_block *es = sbi->s_es;
1836 	struct ext4_inode_info *ei = EXT4_I(inode);
1837 	ext4_fsblk_t nr;
1838 	int i, ret, err = 0;
1839 	int credits = 1;
1840 
1841 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1842 	if (inode) {
1843 		if (es->s_reserved_gdt_blocks) {
1844 			ext4_error(sb, "Unexpected non-zero "
1845 				   "s_reserved_gdt_blocks");
1846 			return -EPERM;
1847 		}
1848 
1849 		/* Do a quick sanity check of the resize inode */
1850 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1851 					     (9 - sbi->s_cluster_bits)))
1852 			goto invalid_resize_inode;
1853 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1854 			if (i == EXT4_DIND_BLOCK) {
1855 				if (ei->i_data[i])
1856 					continue;
1857 				else
1858 					goto invalid_resize_inode;
1859 			}
1860 			if (ei->i_data[i])
1861 				goto invalid_resize_inode;
1862 		}
1863 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1864 	}
1865 
1866 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1867 	if (IS_ERR(handle))
1868 		return PTR_ERR(handle);
1869 
1870 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1871 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1872 	if (err)
1873 		goto errout;
1874 
1875 	ext4_clear_feature_resize_inode(sb);
1876 	ext4_set_feature_meta_bg(sb);
1877 	sbi->s_es->s_first_meta_bg =
1878 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1879 
1880 	err = ext4_handle_dirty_super(handle, sb);
1881 	if (err) {
1882 		ext4_std_error(sb, err);
1883 		goto errout;
1884 	}
1885 
1886 	if (inode) {
1887 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1888 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1889 				 EXT4_FREE_BLOCKS_METADATA |
1890 				 EXT4_FREE_BLOCKS_FORGET);
1891 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1892 		inode->i_blocks = 0;
1893 
1894 		err = ext4_mark_inode_dirty(handle, inode);
1895 		if (err)
1896 			ext4_std_error(sb, err);
1897 	}
1898 
1899 errout:
1900 	ret = ext4_journal_stop(handle);
1901 	if (!err)
1902 		err = ret;
1903 	return ret;
1904 
1905 invalid_resize_inode:
1906 	ext4_error(sb, "corrupted/inconsistent resize inode");
1907 	return -EINVAL;
1908 }
1909 
1910 /*
1911  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1912  *
1913  * @sb: super block of the fs to be resized
1914  * @n_blocks_count: the number of blocks resides in the resized fs
1915  */
1916 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1917 {
1918 	struct ext4_new_flex_group_data *flex_gd = NULL;
1919 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1920 	struct ext4_super_block *es = sbi->s_es;
1921 	struct buffer_head *bh;
1922 	struct inode *resize_inode = NULL;
1923 	ext4_grpblk_t add, offset;
1924 	unsigned long n_desc_blocks;
1925 	unsigned long o_desc_blocks;
1926 	ext4_group_t o_group;
1927 	ext4_group_t n_group;
1928 	ext4_fsblk_t o_blocks_count;
1929 	ext4_fsblk_t n_blocks_count_retry = 0;
1930 	unsigned long last_update_time = 0;
1931 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1932 	int meta_bg;
1933 
1934 	/* See if the device is actually as big as what was requested */
1935 	bh = sb_bread(sb, n_blocks_count - 1);
1936 	if (!bh) {
1937 		ext4_warning(sb, "can't read last block, resize aborted");
1938 		return -ENOSPC;
1939 	}
1940 	brelse(bh);
1941 
1942 retry:
1943 	o_blocks_count = ext4_blocks_count(es);
1944 
1945 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1946 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1947 
1948 	if (n_blocks_count < o_blocks_count) {
1949 		/* On-line shrinking not supported */
1950 		ext4_warning(sb, "can't shrink FS - resize aborted");
1951 		return -EINVAL;
1952 	}
1953 
1954 	if (n_blocks_count == o_blocks_count)
1955 		/* Nothing need to do */
1956 		return 0;
1957 
1958 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1959 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1960 		ext4_warning(sb, "resize would cause inodes_count overflow");
1961 		return -EINVAL;
1962 	}
1963 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1964 
1965 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1966 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1967 
1968 	meta_bg = ext4_has_feature_meta_bg(sb);
1969 
1970 	if (ext4_has_feature_resize_inode(sb)) {
1971 		if (meta_bg) {
1972 			ext4_error(sb, "resize_inode and meta_bg enabled "
1973 				   "simultaneously");
1974 			return -EINVAL;
1975 		}
1976 		if (n_desc_blocks > o_desc_blocks +
1977 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1978 			n_blocks_count_retry = n_blocks_count;
1979 			n_desc_blocks = o_desc_blocks +
1980 				le16_to_cpu(es->s_reserved_gdt_blocks);
1981 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1982 			n_blocks_count = (ext4_fsblk_t)n_group *
1983 				EXT4_BLOCKS_PER_GROUP(sb) +
1984 				le32_to_cpu(es->s_first_data_block);
1985 			n_group--; /* set to last group number */
1986 		}
1987 
1988 		if (!resize_inode)
1989 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1990 						 EXT4_IGET_SPECIAL);
1991 		if (IS_ERR(resize_inode)) {
1992 			ext4_warning(sb, "Error opening resize inode");
1993 			return PTR_ERR(resize_inode);
1994 		}
1995 	}
1996 
1997 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1998 		err = ext4_convert_meta_bg(sb, resize_inode);
1999 		if (err)
2000 			goto out;
2001 		if (resize_inode) {
2002 			iput(resize_inode);
2003 			resize_inode = NULL;
2004 		}
2005 		if (n_blocks_count_retry) {
2006 			n_blocks_count = n_blocks_count_retry;
2007 			n_blocks_count_retry = 0;
2008 			goto retry;
2009 		}
2010 	}
2011 
2012 	/*
2013 	 * Make sure the last group has enough space so that it's
2014 	 * guaranteed to have enough space for all metadata blocks
2015 	 * that it might need to hold.  (We might not need to store
2016 	 * the inode table blocks in the last block group, but there
2017 	 * will be cases where this might be needed.)
2018 	 */
2019 	if ((ext4_group_first_block_no(sb, n_group) +
2020 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
2021 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2022 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2023 		n_group--;
2024 		n_blocks_count_retry = 0;
2025 		if (resize_inode) {
2026 			iput(resize_inode);
2027 			resize_inode = NULL;
2028 		}
2029 		goto retry;
2030 	}
2031 
2032 	/* extend the last group */
2033 	if (n_group == o_group)
2034 		add = n_blocks_count - o_blocks_count;
2035 	else
2036 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2037 	if (add > 0) {
2038 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2039 		if (err)
2040 			goto out;
2041 	}
2042 
2043 	if (ext4_blocks_count(es) == n_blocks_count)
2044 		goto out;
2045 
2046 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2047 	if (err)
2048 		goto out;
2049 
2050 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2051 	if (err)
2052 		goto out;
2053 
2054 	flex_gd = alloc_flex_gd(flexbg_size);
2055 	if (flex_gd == NULL) {
2056 		err = -ENOMEM;
2057 		goto out;
2058 	}
2059 
2060 	/* Add flex groups. Note that a regular group is a
2061 	 * flex group with 1 group.
2062 	 */
2063 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2064 					      flexbg_size)) {
2065 		if (jiffies - last_update_time > HZ * 10) {
2066 			if (last_update_time)
2067 				ext4_msg(sb, KERN_INFO,
2068 					 "resized to %llu blocks",
2069 					 ext4_blocks_count(es));
2070 			last_update_time = jiffies;
2071 		}
2072 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2073 			break;
2074 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2075 		if (unlikely(err))
2076 			break;
2077 	}
2078 
2079 	if (!err && n_blocks_count_retry) {
2080 		n_blocks_count = n_blocks_count_retry;
2081 		n_blocks_count_retry = 0;
2082 		free_flex_gd(flex_gd);
2083 		flex_gd = NULL;
2084 		if (resize_inode) {
2085 			iput(resize_inode);
2086 			resize_inode = NULL;
2087 		}
2088 		goto retry;
2089 	}
2090 
2091 out:
2092 	if (flex_gd)
2093 		free_flex_gd(flex_gd);
2094 	if (resize_inode != NULL)
2095 		iput(resize_inode);
2096 	if (err)
2097 		ext4_warning(sb, "error (%d) occurred during "
2098 			     "file system resize", err);
2099 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2100 		 ext4_blocks_count(es));
2101 	return err;
2102 }
2103