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 
ext4_resize_begin(struct super_block * sb)20 int ext4_resize_begin(struct super_block *sb)
21 {
22 	struct ext4_sb_info *sbi = EXT4_SB(sb);
23 	int ret = 0;
24 
25 	if (!capable(CAP_SYS_RESOURCE))
26 		return -EPERM;
27 
28 	/*
29 	 * If we are not using the primary superblock/GDT copy don't resize,
30          * because the user tools have no way of handling this.  Probably a
31          * bad time to do it anyways.
32          */
33 	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
34 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
35 		ext4_warning(sb, "won't resize using backup superblock at %llu",
36 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
37 		return -EPERM;
38 	}
39 
40 	/*
41 	 * We are not allowed to do online-resizing on a filesystem mounted
42 	 * with error, because it can destroy the filesystem easily.
43 	 */
44 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
45 		ext4_warning(sb, "There are errors in the filesystem, "
46 			     "so online resizing is not allowed");
47 		return -EPERM;
48 	}
49 
50 	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
51 				  &EXT4_SB(sb)->s_ext4_flags))
52 		ret = -EBUSY;
53 
54 	return ret;
55 }
56 
ext4_resize_end(struct super_block * sb)57 void ext4_resize_end(struct super_block *sb)
58 {
59 	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
60 	smp_mb__after_atomic();
61 }
62 
ext4_meta_bg_first_group(struct super_block * sb,ext4_group_t group)63 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
64 					     ext4_group_t group) {
65 	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
66 	       EXT4_DESC_PER_BLOCK_BITS(sb);
67 }
68 
ext4_meta_bg_first_block_no(struct super_block * sb,ext4_group_t group)69 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
70 					     ext4_group_t group) {
71 	group = ext4_meta_bg_first_group(sb, group);
72 	return ext4_group_first_block_no(sb, group);
73 }
74 
ext4_group_overhead_blocks(struct super_block * sb,ext4_group_t group)75 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
76 						ext4_group_t group) {
77 	ext4_grpblk_t overhead;
78 	overhead = ext4_bg_num_gdb(sb, group);
79 	if (ext4_bg_has_super(sb, group))
80 		overhead += 1 +
81 			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
82 	return overhead;
83 }
84 
85 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
86 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
87 
verify_group_input(struct super_block * sb,struct ext4_new_group_data * input)88 static int verify_group_input(struct super_block *sb,
89 			      struct ext4_new_group_data *input)
90 {
91 	struct ext4_sb_info *sbi = EXT4_SB(sb);
92 	struct ext4_super_block *es = sbi->s_es;
93 	ext4_fsblk_t start = ext4_blocks_count(es);
94 	ext4_fsblk_t end = start + input->blocks_count;
95 	ext4_group_t group = input->group;
96 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
97 	unsigned overhead;
98 	ext4_fsblk_t metaend;
99 	struct buffer_head *bh = NULL;
100 	ext4_grpblk_t free_blocks_count, offset;
101 	int err = -EINVAL;
102 
103 	if (group != sbi->s_groups_count) {
104 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
105 			     input->group, sbi->s_groups_count);
106 		return -EINVAL;
107 	}
108 
109 	overhead = ext4_group_overhead_blocks(sb, group);
110 	metaend = start + overhead;
111 	input->free_clusters_count = free_blocks_count =
112 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
113 
114 	if (test_opt(sb, DEBUG))
115 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
116 		       "(%d free, %u reserved)\n",
117 		       ext4_bg_has_super(sb, input->group) ? "normal" :
118 		       "no-super", input->group, input->blocks_count,
119 		       free_blocks_count, input->reserved_blocks);
120 
121 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
122 	if (offset != 0)
123 			ext4_warning(sb, "Last group not full");
124 	else if (input->reserved_blocks > input->blocks_count / 5)
125 		ext4_warning(sb, "Reserved blocks too high (%u)",
126 			     input->reserved_blocks);
127 	else if (free_blocks_count < 0)
128 		ext4_warning(sb, "Bad blocks count %u",
129 			     input->blocks_count);
130 	else if (!(bh = sb_bread(sb, end - 1)))
131 		ext4_warning(sb, "Cannot read last block (%llu)",
132 			     end - 1);
133 	else if (outside(input->block_bitmap, start, end))
134 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
135 			     (unsigned long long)input->block_bitmap);
136 	else if (outside(input->inode_bitmap, start, end))
137 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
138 			     (unsigned long long)input->inode_bitmap);
139 	else if (outside(input->inode_table, start, end) ||
140 		 outside(itend - 1, start, end))
141 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
142 			     (unsigned long long)input->inode_table, itend - 1);
143 	else if (input->inode_bitmap == input->block_bitmap)
144 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
145 			     (unsigned long long)input->block_bitmap);
146 	else if (inside(input->block_bitmap, input->inode_table, itend))
147 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
148 			     "(%llu-%llu)",
149 			     (unsigned long long)input->block_bitmap,
150 			     (unsigned long long)input->inode_table, itend - 1);
151 	else if (inside(input->inode_bitmap, input->inode_table, itend))
152 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
153 			     "(%llu-%llu)",
154 			     (unsigned long long)input->inode_bitmap,
155 			     (unsigned long long)input->inode_table, itend - 1);
156 	else if (inside(input->block_bitmap, start, metaend))
157 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
158 			     (unsigned long long)input->block_bitmap,
159 			     start, metaend - 1);
160 	else if (inside(input->inode_bitmap, start, metaend))
161 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
162 			     (unsigned long long)input->inode_bitmap,
163 			     start, metaend - 1);
164 	else if (inside(input->inode_table, start, metaend) ||
165 		 inside(itend - 1, start, metaend))
166 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
167 			     "(%llu-%llu)",
168 			     (unsigned long long)input->inode_table,
169 			     itend - 1, start, metaend - 1);
170 	else
171 		err = 0;
172 	brelse(bh);
173 
174 	return err;
175 }
176 
177 /*
178  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
179  * group each time.
180  */
181 struct ext4_new_flex_group_data {
182 	struct ext4_new_group_data *groups;	/* new_group_data for groups
183 						   in the flex group */
184 	__u16 *bg_flags;			/* block group flags of groups
185 						   in @groups */
186 	ext4_group_t count;			/* number of groups in @groups
187 						 */
188 };
189 
190 /*
191  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
192  * @flexbg_size.
193  *
194  * Returns NULL on failure otherwise address of the allocated structure.
195  */
alloc_flex_gd(unsigned long flexbg_size)196 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
197 {
198 	struct ext4_new_flex_group_data *flex_gd;
199 
200 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
201 	if (flex_gd == NULL)
202 		goto out3;
203 
204 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
205 		goto out2;
206 	flex_gd->count = flexbg_size;
207 
208 	flex_gd->groups = kmalloc_array(flexbg_size,
209 					sizeof(struct ext4_new_group_data),
210 					GFP_NOFS);
211 	if (flex_gd->groups == NULL)
212 		goto out2;
213 
214 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
215 					  GFP_NOFS);
216 	if (flex_gd->bg_flags == NULL)
217 		goto out1;
218 
219 	return flex_gd;
220 
221 out1:
222 	kfree(flex_gd->groups);
223 out2:
224 	kfree(flex_gd);
225 out3:
226 	return NULL;
227 }
228 
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)229 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
230 {
231 	kfree(flex_gd->bg_flags);
232 	kfree(flex_gd->groups);
233 	kfree(flex_gd);
234 }
235 
236 /*
237  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
238  * and inode tables for a flex group.
239  *
240  * This function is used by 64bit-resize.  Note that this function allocates
241  * group tables from the 1st group of groups contained by @flexgd, which may
242  * be a partial of a flex group.
243  *
244  * @sb: super block of fs to which the groups belongs
245  *
246  * Returns 0 on a successful allocation of the metadata blocks in the
247  * block group.
248  */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,int flexbg_size)249 static int ext4_alloc_group_tables(struct super_block *sb,
250 				struct ext4_new_flex_group_data *flex_gd,
251 				int flexbg_size)
252 {
253 	struct ext4_new_group_data *group_data = flex_gd->groups;
254 	ext4_fsblk_t start_blk;
255 	ext4_fsblk_t last_blk;
256 	ext4_group_t src_group;
257 	ext4_group_t bb_index = 0;
258 	ext4_group_t ib_index = 0;
259 	ext4_group_t it_index = 0;
260 	ext4_group_t group;
261 	ext4_group_t last_group;
262 	unsigned overhead;
263 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
264 	int i;
265 
266 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
267 
268 	src_group = group_data[0].group;
269 	last_group  = src_group + flex_gd->count - 1;
270 
271 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
272 	       (last_group & ~(flexbg_size - 1))));
273 next_group:
274 	group = group_data[0].group;
275 	if (src_group >= group_data[0].group + flex_gd->count)
276 		return -ENOSPC;
277 	start_blk = ext4_group_first_block_no(sb, src_group);
278 	last_blk = start_blk + group_data[src_group - group].blocks_count;
279 
280 	overhead = ext4_group_overhead_blocks(sb, src_group);
281 
282 	start_blk += overhead;
283 
284 	/* We collect contiguous blocks as much as possible. */
285 	src_group++;
286 	for (; src_group <= last_group; src_group++) {
287 		overhead = ext4_group_overhead_blocks(sb, src_group);
288 		if (overhead == 0)
289 			last_blk += group_data[src_group - group].blocks_count;
290 		else
291 			break;
292 	}
293 
294 	/* Allocate block bitmaps */
295 	for (; bb_index < flex_gd->count; bb_index++) {
296 		if (start_blk >= last_blk)
297 			goto next_group;
298 		group_data[bb_index].block_bitmap = start_blk++;
299 		group = ext4_get_group_number(sb, start_blk - 1);
300 		group -= group_data[0].group;
301 		group_data[group].mdata_blocks++;
302 		flex_gd->bg_flags[group] &= uninit_mask;
303 	}
304 
305 	/* Allocate inode bitmaps */
306 	for (; ib_index < flex_gd->count; ib_index++) {
307 		if (start_blk >= last_blk)
308 			goto next_group;
309 		group_data[ib_index].inode_bitmap = start_blk++;
310 		group = ext4_get_group_number(sb, start_blk - 1);
311 		group -= group_data[0].group;
312 		group_data[group].mdata_blocks++;
313 		flex_gd->bg_flags[group] &= uninit_mask;
314 	}
315 
316 	/* Allocate inode tables */
317 	for (; it_index < flex_gd->count; it_index++) {
318 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
319 		ext4_fsblk_t next_group_start;
320 
321 		if (start_blk + itb > last_blk)
322 			goto next_group;
323 		group_data[it_index].inode_table = start_blk;
324 		group = ext4_get_group_number(sb, start_blk);
325 		next_group_start = ext4_group_first_block_no(sb, group + 1);
326 		group -= group_data[0].group;
327 
328 		if (start_blk + itb > next_group_start) {
329 			flex_gd->bg_flags[group + 1] &= uninit_mask;
330 			overhead = start_blk + itb - next_group_start;
331 			group_data[group + 1].mdata_blocks += overhead;
332 			itb -= overhead;
333 		}
334 
335 		group_data[group].mdata_blocks += itb;
336 		flex_gd->bg_flags[group] &= uninit_mask;
337 		start_blk += EXT4_SB(sb)->s_itb_per_group;
338 	}
339 
340 	/* Update free clusters count to exclude metadata blocks */
341 	for (i = 0; i < flex_gd->count; i++) {
342 		group_data[i].free_clusters_count -=
343 				EXT4_NUM_B2C(EXT4_SB(sb),
344 					     group_data[i].mdata_blocks);
345 	}
346 
347 	if (test_opt(sb, DEBUG)) {
348 		int i;
349 		group = group_data[0].group;
350 
351 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
352 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
353 		       flexbg_size);
354 
355 		for (i = 0; i < flex_gd->count; i++) {
356 			ext4_debug(
357 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
358 			       ext4_bg_has_super(sb, group + i) ? "normal" :
359 			       "no-super", group + i,
360 			       group_data[i].blocks_count,
361 			       group_data[i].free_clusters_count,
362 			       group_data[i].mdata_blocks);
363 		}
364 	}
365 	return 0;
366 }
367 
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)368 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
369 				  ext4_fsblk_t blk)
370 {
371 	struct buffer_head *bh;
372 	int err;
373 
374 	bh = sb_getblk(sb, blk);
375 	if (unlikely(!bh))
376 		return ERR_PTR(-ENOMEM);
377 	BUFFER_TRACE(bh, "get_write_access");
378 	if ((err = ext4_journal_get_write_access(handle, bh))) {
379 		brelse(bh);
380 		bh = ERR_PTR(err);
381 	} else {
382 		memset(bh->b_data, 0, sb->s_blocksize);
383 		set_buffer_uptodate(bh);
384 	}
385 
386 	return bh;
387 }
388 
389 /*
390  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
391  * If that fails, restart the transaction & regain write access for the
392  * buffer head which is used for block_bitmap modifications.
393  */
extend_or_restart_transaction(handle_t * handle,int thresh)394 static int extend_or_restart_transaction(handle_t *handle, int thresh)
395 {
396 	int err;
397 
398 	if (ext4_handle_has_enough_credits(handle, thresh))
399 		return 0;
400 
401 	err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
402 	if (err < 0)
403 		return err;
404 	if (err) {
405 		err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
406 		if (err)
407 			return err;
408 	}
409 
410 	return 0;
411 }
412 
413 /*
414  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
415  *
416  * Helper function for ext4_setup_new_group_blocks() which set .
417  *
418  * @sb: super block
419  * @handle: journal handle
420  * @flex_gd: flex group data
421  */
set_flexbg_block_bitmap(struct super_block * sb,handle_t * handle,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t first_cluster,ext4_fsblk_t last_cluster)422 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
423 			struct ext4_new_flex_group_data *flex_gd,
424 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
425 {
426 	struct ext4_sb_info *sbi = EXT4_SB(sb);
427 	ext4_group_t count = last_cluster - first_cluster + 1;
428 	ext4_group_t count2;
429 
430 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
431 		   last_cluster);
432 	for (count2 = count; count > 0;
433 	     count -= count2, first_cluster += count2) {
434 		ext4_fsblk_t start;
435 		struct buffer_head *bh;
436 		ext4_group_t group;
437 		int err;
438 
439 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
440 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
441 		group -= flex_gd->groups[0].group;
442 
443 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
444 		if (count2 > count)
445 			count2 = count;
446 
447 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
448 			BUG_ON(flex_gd->count > 1);
449 			continue;
450 		}
451 
452 		err = extend_or_restart_transaction(handle, 1);
453 		if (err)
454 			return err;
455 
456 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
457 		if (unlikely(!bh))
458 			return -ENOMEM;
459 
460 		BUFFER_TRACE(bh, "get_write_access");
461 		err = ext4_journal_get_write_access(handle, bh);
462 		if (err)
463 			return err;
464 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
465 			   first_cluster, first_cluster - start, count2);
466 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
467 
468 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
469 		if (unlikely(err))
470 			return err;
471 		brelse(bh);
472 	}
473 
474 	return 0;
475 }
476 
477 /*
478  * Set up the block and inode bitmaps, and the inode table for the new groups.
479  * This doesn't need to be part of the main transaction, since we are only
480  * changing blocks outside the actual filesystem.  We still do journaling to
481  * ensure the recovery is correct in case of a failure just after resize.
482  * If any part of this fails, we simply abort the resize.
483  *
484  * setup_new_flex_group_blocks handles a flex group as follow:
485  *  1. copy super block and GDT, and initialize group tables if necessary.
486  *     In this step, we only set bits in blocks bitmaps for blocks taken by
487  *     super block and GDT.
488  *  2. allocate group tables in block bitmaps, that is, set bits in block
489  *     bitmap for blocks taken by group tables.
490  */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)491 static int setup_new_flex_group_blocks(struct super_block *sb,
492 				struct ext4_new_flex_group_data *flex_gd)
493 {
494 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
495 	ext4_fsblk_t start;
496 	ext4_fsblk_t block;
497 	struct ext4_sb_info *sbi = EXT4_SB(sb);
498 	struct ext4_super_block *es = sbi->s_es;
499 	struct ext4_new_group_data *group_data = flex_gd->groups;
500 	__u16 *bg_flags = flex_gd->bg_flags;
501 	handle_t *handle;
502 	ext4_group_t group, count;
503 	struct buffer_head *bh = NULL;
504 	int reserved_gdb, i, j, err = 0, err2;
505 	int meta_bg;
506 
507 	BUG_ON(!flex_gd->count || !group_data ||
508 	       group_data[0].group != sbi->s_groups_count);
509 
510 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
511 	meta_bg = ext4_has_feature_meta_bg(sb);
512 
513 	/* This transaction may be extended/restarted along the way */
514 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
515 	if (IS_ERR(handle))
516 		return PTR_ERR(handle);
517 
518 	group = group_data[0].group;
519 	for (i = 0; i < flex_gd->count; i++, group++) {
520 		unsigned long gdblocks;
521 		ext4_grpblk_t overhead;
522 
523 		gdblocks = ext4_bg_num_gdb(sb, group);
524 		start = ext4_group_first_block_no(sb, group);
525 
526 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
527 			goto handle_itb;
528 
529 		if (meta_bg == 1) {
530 			ext4_group_t first_group;
531 			first_group = ext4_meta_bg_first_group(sb, group);
532 			if (first_group != group + 1 &&
533 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
534 				goto handle_itb;
535 		}
536 
537 		block = start + ext4_bg_has_super(sb, group);
538 		/* Copy all of the GDT blocks into the backup in this group */
539 		for (j = 0; j < gdblocks; j++, block++) {
540 			struct buffer_head *gdb;
541 
542 			ext4_debug("update backup group %#04llx\n", block);
543 			err = extend_or_restart_transaction(handle, 1);
544 			if (err)
545 				goto out;
546 
547 			gdb = sb_getblk(sb, block);
548 			if (unlikely(!gdb)) {
549 				err = -ENOMEM;
550 				goto out;
551 			}
552 
553 			BUFFER_TRACE(gdb, "get_write_access");
554 			err = ext4_journal_get_write_access(handle, gdb);
555 			if (err) {
556 				brelse(gdb);
557 				goto out;
558 			}
559 			memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
560 			       gdb->b_size);
561 			set_buffer_uptodate(gdb);
562 
563 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
564 			if (unlikely(err)) {
565 				brelse(gdb);
566 				goto out;
567 			}
568 			brelse(gdb);
569 		}
570 
571 		/* Zero out all of the reserved backup group descriptor
572 		 * table blocks
573 		 */
574 		if (ext4_bg_has_super(sb, group)) {
575 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
576 					reserved_gdb, GFP_NOFS);
577 			if (err)
578 				goto out;
579 		}
580 
581 handle_itb:
582 		/* Initialize group tables of the grop @group */
583 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
584 			goto handle_bb;
585 
586 		/* Zero out all of the inode table blocks */
587 		block = group_data[i].inode_table;
588 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
589 			   block, sbi->s_itb_per_group);
590 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
591 				       GFP_NOFS);
592 		if (err)
593 			goto out;
594 
595 handle_bb:
596 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
597 			goto handle_ib;
598 
599 		/* Initialize block bitmap of the @group */
600 		block = group_data[i].block_bitmap;
601 		err = extend_or_restart_transaction(handle, 1);
602 		if (err)
603 			goto out;
604 
605 		bh = bclean(handle, sb, block);
606 		if (IS_ERR(bh)) {
607 			err = PTR_ERR(bh);
608 			bh = NULL;
609 			goto out;
610 		}
611 		overhead = ext4_group_overhead_blocks(sb, group);
612 		if (overhead != 0) {
613 			ext4_debug("mark backup superblock %#04llx (+0)\n",
614 				   start);
615 			ext4_set_bits(bh->b_data, 0,
616 				      EXT4_NUM_B2C(sbi, overhead));
617 		}
618 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
619 				     sb->s_blocksize * 8, bh->b_data);
620 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
621 		if (err)
622 			goto out;
623 		brelse(bh);
624 
625 handle_ib:
626 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
627 			continue;
628 
629 		/* Initialize inode bitmap of the @group */
630 		block = group_data[i].inode_bitmap;
631 		err = extend_or_restart_transaction(handle, 1);
632 		if (err)
633 			goto out;
634 		/* Mark unused entries in inode bitmap used */
635 		bh = bclean(handle, sb, block);
636 		if (IS_ERR(bh)) {
637 			err = PTR_ERR(bh);
638 			bh = NULL;
639 			goto out;
640 		}
641 
642 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
643 				     sb->s_blocksize * 8, bh->b_data);
644 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
645 		if (err)
646 			goto out;
647 		brelse(bh);
648 	}
649 	bh = NULL;
650 
651 	/* Mark group tables in block bitmap */
652 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
653 		count = group_table_count[j];
654 		start = (&group_data[0].block_bitmap)[j];
655 		block = start;
656 		for (i = 1; i < flex_gd->count; i++) {
657 			block += group_table_count[j];
658 			if (block == (&group_data[i].block_bitmap)[j]) {
659 				count += group_table_count[j];
660 				continue;
661 			}
662 			err = set_flexbg_block_bitmap(sb, handle,
663 						      flex_gd,
664 						      EXT4_B2C(sbi, start),
665 						      EXT4_B2C(sbi,
666 							       start + count
667 							       - 1));
668 			if (err)
669 				goto out;
670 			count = group_table_count[j];
671 			start = (&group_data[i].block_bitmap)[j];
672 			block = start;
673 		}
674 
675 		if (count) {
676 			err = set_flexbg_block_bitmap(sb, handle,
677 						      flex_gd,
678 						      EXT4_B2C(sbi, start),
679 						      EXT4_B2C(sbi,
680 							       start + count
681 							       - 1));
682 			if (err)
683 				goto out;
684 		}
685 	}
686 
687 out:
688 	brelse(bh);
689 	err2 = ext4_journal_stop(handle);
690 	if (err2 && !err)
691 		err = err2;
692 
693 	return err;
694 }
695 
696 /*
697  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
698  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
699  * calling this for the first time.  In a sparse filesystem it will be the
700  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
701  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
702  */
ext4_list_backups(struct super_block * sb,unsigned * three,unsigned * five,unsigned * seven)703 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
704 				  unsigned *five, unsigned *seven)
705 {
706 	unsigned *min = three;
707 	int mult = 3;
708 	unsigned ret;
709 
710 	if (!ext4_has_feature_sparse_super(sb)) {
711 		ret = *min;
712 		*min += 1;
713 		return ret;
714 	}
715 
716 	if (*five < *min) {
717 		min = five;
718 		mult = 5;
719 	}
720 	if (*seven < *min) {
721 		min = seven;
722 		mult = 7;
723 	}
724 
725 	ret = *min;
726 	*min *= mult;
727 
728 	return ret;
729 }
730 
731 /*
732  * Check that all of the backup GDT blocks are held in the primary GDT block.
733  * It is assumed that they are stored in group order.  Returns the number of
734  * groups in current filesystem that have BACKUPS, or -ve error code.
735  */
verify_reserved_gdb(struct super_block * sb,ext4_group_t end,struct buffer_head * primary)736 static int verify_reserved_gdb(struct super_block *sb,
737 			       ext4_group_t end,
738 			       struct buffer_head *primary)
739 {
740 	const ext4_fsblk_t blk = primary->b_blocknr;
741 	unsigned three = 1;
742 	unsigned five = 5;
743 	unsigned seven = 7;
744 	unsigned grp;
745 	__le32 *p = (__le32 *)primary->b_data;
746 	int gdbackups = 0;
747 
748 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
749 		if (le32_to_cpu(*p++) !=
750 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
751 			ext4_warning(sb, "reserved GDT %llu"
752 				     " missing grp %d (%llu)",
753 				     blk, grp,
754 				     grp *
755 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
756 				     blk);
757 			return -EINVAL;
758 		}
759 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
760 			return -EFBIG;
761 	}
762 
763 	return gdbackups;
764 }
765 
766 /*
767  * Called when we need to bring a reserved group descriptor table block into
768  * use from the resize inode.  The primary copy of the new GDT block currently
769  * is an indirect block (under the double indirect block in the resize inode).
770  * The new backup GDT blocks will be stored as leaf blocks in this indirect
771  * block, in group order.  Even though we know all the block numbers we need,
772  * we check to ensure that the resize inode has actually reserved these blocks.
773  *
774  * Don't need to update the block bitmaps because the blocks are still in use.
775  *
776  * We get all of the error cases out of the way, so that we are sure to not
777  * fail once we start modifying the data on disk, because JBD has no rollback.
778  */
add_new_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)779 static int add_new_gdb(handle_t *handle, struct inode *inode,
780 		       ext4_group_t group)
781 {
782 	struct super_block *sb = inode->i_sb;
783 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
784 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
785 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
786 	struct buffer_head **o_group_desc, **n_group_desc;
787 	struct buffer_head *dind;
788 	struct buffer_head *gdb_bh;
789 	int gdbackups;
790 	struct ext4_iloc iloc;
791 	__le32 *data;
792 	int err;
793 
794 	if (test_opt(sb, DEBUG))
795 		printk(KERN_DEBUG
796 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
797 		       gdb_num);
798 
799 	gdb_bh = sb_bread(sb, gdblock);
800 	if (!gdb_bh)
801 		return -EIO;
802 
803 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
804 	if (gdbackups < 0) {
805 		err = gdbackups;
806 		goto exit_bh;
807 	}
808 
809 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
810 	dind = sb_bread(sb, le32_to_cpu(*data));
811 	if (!dind) {
812 		err = -EIO;
813 		goto exit_bh;
814 	}
815 
816 	data = (__le32 *)dind->b_data;
817 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
818 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
819 			     group, gdblock);
820 		err = -EINVAL;
821 		goto exit_dind;
822 	}
823 
824 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
825 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
826 	if (unlikely(err))
827 		goto exit_dind;
828 
829 	BUFFER_TRACE(gdb_bh, "get_write_access");
830 	err = ext4_journal_get_write_access(handle, gdb_bh);
831 	if (unlikely(err))
832 		goto exit_dind;
833 
834 	BUFFER_TRACE(dind, "get_write_access");
835 	err = ext4_journal_get_write_access(handle, dind);
836 	if (unlikely(err))
837 		ext4_std_error(sb, err);
838 
839 	/* ext4_reserve_inode_write() gets a reference on the iloc */
840 	err = ext4_reserve_inode_write(handle, inode, &iloc);
841 	if (unlikely(err))
842 		goto exit_dind;
843 
844 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
845 				     sizeof(struct buffer_head *),
846 				     GFP_NOFS);
847 	if (!n_group_desc) {
848 		err = -ENOMEM;
849 		ext4_warning(sb, "not enough memory for %lu groups",
850 			     gdb_num + 1);
851 		goto exit_inode;
852 	}
853 
854 	/*
855 	 * Finally, we have all of the possible failures behind us...
856 	 *
857 	 * Remove new GDT block from inode double-indirect block and clear out
858 	 * the new GDT block for use (which also "frees" the backup GDT blocks
859 	 * from the reserved inode).  We don't need to change the bitmaps for
860 	 * these blocks, because they are marked as in-use from being in the
861 	 * reserved inode, and will become GDT blocks (primary and backup).
862 	 */
863 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
864 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
865 	if (unlikely(err)) {
866 		ext4_std_error(sb, err);
867 		goto exit_inode;
868 	}
869 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
870 			   (9 - EXT4_SB(sb)->s_cluster_bits);
871 	ext4_mark_iloc_dirty(handle, inode, &iloc);
872 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
873 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
874 	if (unlikely(err)) {
875 		ext4_std_error(sb, err);
876 		goto exit_inode;
877 	}
878 	brelse(dind);
879 
880 	o_group_desc = EXT4_SB(sb)->s_group_desc;
881 	memcpy(n_group_desc, o_group_desc,
882 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
883 	n_group_desc[gdb_num] = gdb_bh;
884 	EXT4_SB(sb)->s_group_desc = n_group_desc;
885 	EXT4_SB(sb)->s_gdb_count++;
886 	kvfree(o_group_desc);
887 
888 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
889 	err = ext4_handle_dirty_super(handle, sb);
890 	if (err)
891 		ext4_std_error(sb, err);
892 
893 	return err;
894 
895 exit_inode:
896 	kvfree(n_group_desc);
897 	brelse(iloc.bh);
898 exit_dind:
899 	brelse(dind);
900 exit_bh:
901 	brelse(gdb_bh);
902 
903 	ext4_debug("leaving with error %d\n", err);
904 	return err;
905 }
906 
907 /*
908  * add_new_gdb_meta_bg is the sister of add_new_gdb.
909  */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)910 static int add_new_gdb_meta_bg(struct super_block *sb,
911 			       handle_t *handle, ext4_group_t group) {
912 	ext4_fsblk_t gdblock;
913 	struct buffer_head *gdb_bh;
914 	struct buffer_head **o_group_desc, **n_group_desc;
915 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
916 	int err;
917 
918 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
919 		   ext4_bg_has_super(sb, group);
920 	gdb_bh = sb_bread(sb, gdblock);
921 	if (!gdb_bh)
922 		return -EIO;
923 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
924 				     sizeof(struct buffer_head *),
925 				     GFP_NOFS);
926 	if (!n_group_desc) {
927 		err = -ENOMEM;
928 		ext4_warning(sb, "not enough memory for %lu groups",
929 			     gdb_num + 1);
930 		return err;
931 	}
932 
933 	o_group_desc = EXT4_SB(sb)->s_group_desc;
934 	memcpy(n_group_desc, o_group_desc,
935 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
936 	n_group_desc[gdb_num] = gdb_bh;
937 	EXT4_SB(sb)->s_group_desc = n_group_desc;
938 	EXT4_SB(sb)->s_gdb_count++;
939 	kvfree(o_group_desc);
940 	BUFFER_TRACE(gdb_bh, "get_write_access");
941 	err = ext4_journal_get_write_access(handle, gdb_bh);
942 	if (unlikely(err))
943 		brelse(gdb_bh);
944 	return err;
945 }
946 
947 /*
948  * Called when we are adding a new group which has a backup copy of each of
949  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
950  * We need to add these reserved backup GDT blocks to the resize inode, so
951  * that they are kept for future resizing and not allocated to files.
952  *
953  * Each reserved backup GDT block will go into a different indirect block.
954  * The indirect blocks are actually the primary reserved GDT blocks,
955  * so we know in advance what their block numbers are.  We only get the
956  * double-indirect block to verify it is pointing to the primary reserved
957  * GDT blocks so we don't overwrite a data block by accident.  The reserved
958  * backup GDT blocks are stored in their reserved primary GDT block.
959  */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)960 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
961 			      ext4_group_t group)
962 {
963 	struct super_block *sb = inode->i_sb;
964 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
965 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
966 	struct buffer_head **primary;
967 	struct buffer_head *dind;
968 	struct ext4_iloc iloc;
969 	ext4_fsblk_t blk;
970 	__le32 *data, *end;
971 	int gdbackups = 0;
972 	int res, i;
973 	int err;
974 
975 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
976 	if (!primary)
977 		return -ENOMEM;
978 
979 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
980 	dind = sb_bread(sb, le32_to_cpu(*data));
981 	if (!dind) {
982 		err = -EIO;
983 		goto exit_free;
984 	}
985 
986 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
987 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
988 					 EXT4_ADDR_PER_BLOCK(sb));
989 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
990 
991 	/* Get each reserved primary GDT block and verify it holds backups */
992 	for (res = 0; res < reserved_gdb; res++, blk++) {
993 		if (le32_to_cpu(*data) != blk) {
994 			ext4_warning(sb, "reserved block %llu"
995 				     " not at offset %ld",
996 				     blk,
997 				     (long)(data - (__le32 *)dind->b_data));
998 			err = -EINVAL;
999 			goto exit_bh;
1000 		}
1001 		primary[res] = sb_bread(sb, blk);
1002 		if (!primary[res]) {
1003 			err = -EIO;
1004 			goto exit_bh;
1005 		}
1006 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1007 		if (gdbackups < 0) {
1008 			brelse(primary[res]);
1009 			err = gdbackups;
1010 			goto exit_bh;
1011 		}
1012 		if (++data >= end)
1013 			data = (__le32 *)dind->b_data;
1014 	}
1015 
1016 	for (i = 0; i < reserved_gdb; i++) {
1017 		BUFFER_TRACE(primary[i], "get_write_access");
1018 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1019 			goto exit_bh;
1020 	}
1021 
1022 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1023 		goto exit_bh;
1024 
1025 	/*
1026 	 * Finally we can add each of the reserved backup GDT blocks from
1027 	 * the new group to its reserved primary GDT block.
1028 	 */
1029 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1030 	for (i = 0; i < reserved_gdb; i++) {
1031 		int err2;
1032 		data = (__le32 *)primary[i]->b_data;
1033 		/* printk("reserving backup %lu[%u] = %lu\n",
1034 		       primary[i]->b_blocknr, gdbackups,
1035 		       blk + primary[i]->b_blocknr); */
1036 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1037 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1038 		if (!err)
1039 			err = err2;
1040 	}
1041 
1042 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1043 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1044 
1045 exit_bh:
1046 	while (--res >= 0)
1047 		brelse(primary[res]);
1048 	brelse(dind);
1049 
1050 exit_free:
1051 	kfree(primary);
1052 
1053 	return err;
1054 }
1055 
1056 /*
1057  * Update the backup copies of the ext4 metadata.  These don't need to be part
1058  * of the main resize transaction, because e2fsck will re-write them if there
1059  * is a problem (basically only OOM will cause a problem).  However, we
1060  * _should_ update the backups if possible, in case the primary gets trashed
1061  * for some reason and we need to run e2fsck from a backup superblock.  The
1062  * important part is that the new block and inode counts are in the backup
1063  * superblocks, and the location of the new group metadata in the GDT backups.
1064  *
1065  * We do not need take the s_resize_lock for this, because these
1066  * blocks are not otherwise touched by the filesystem code when it is
1067  * mounted.  We don't need to worry about last changing from
1068  * sbi->s_groups_count, because the worst that can happen is that we
1069  * do not copy the full number of backups at this time.  The resize
1070  * which changed s_groups_count will backup again.
1071  */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1072 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1073 			   int size, int meta_bg)
1074 {
1075 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1076 	ext4_group_t last;
1077 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1078 	unsigned three = 1;
1079 	unsigned five = 5;
1080 	unsigned seven = 7;
1081 	ext4_group_t group = 0;
1082 	int rest = sb->s_blocksize - size;
1083 	handle_t *handle;
1084 	int err = 0, err2;
1085 
1086 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1087 	if (IS_ERR(handle)) {
1088 		group = 1;
1089 		err = PTR_ERR(handle);
1090 		goto exit_err;
1091 	}
1092 
1093 	if (meta_bg == 0) {
1094 		group = ext4_list_backups(sb, &three, &five, &seven);
1095 		last = sbi->s_groups_count;
1096 	} else {
1097 		group = ext4_get_group_number(sb, blk_off) + 1;
1098 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1099 	}
1100 
1101 	while (group < sbi->s_groups_count) {
1102 		struct buffer_head *bh;
1103 		ext4_fsblk_t backup_block;
1104 
1105 		/* Out of journal space, and can't get more - abort - so sad */
1106 		if (ext4_handle_valid(handle) &&
1107 		    handle->h_buffer_credits == 0 &&
1108 		    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1109 		    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1110 			break;
1111 
1112 		if (meta_bg == 0)
1113 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1114 		else
1115 			backup_block = (ext4_group_first_block_no(sb, group) +
1116 					ext4_bg_has_super(sb, group));
1117 
1118 		bh = sb_getblk(sb, backup_block);
1119 		if (unlikely(!bh)) {
1120 			err = -ENOMEM;
1121 			break;
1122 		}
1123 		ext4_debug("update metadata backup %llu(+%llu)\n",
1124 			   backup_block, backup_block -
1125 			   ext4_group_first_block_no(sb, group));
1126 		BUFFER_TRACE(bh, "get_write_access");
1127 		if ((err = ext4_journal_get_write_access(handle, bh)))
1128 			break;
1129 		lock_buffer(bh);
1130 		memcpy(bh->b_data, data, size);
1131 		if (rest)
1132 			memset(bh->b_data + size, 0, rest);
1133 		set_buffer_uptodate(bh);
1134 		unlock_buffer(bh);
1135 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1136 		if (unlikely(err))
1137 			ext4_std_error(sb, err);
1138 		brelse(bh);
1139 
1140 		if (meta_bg == 0)
1141 			group = ext4_list_backups(sb, &three, &five, &seven);
1142 		else if (group == last)
1143 			break;
1144 		else
1145 			group = last;
1146 	}
1147 	if ((err2 = ext4_journal_stop(handle)) && !err)
1148 		err = err2;
1149 
1150 	/*
1151 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1152 	 * late to revert the resize, we shouldn't fail just because of
1153 	 * the backup copies (they are only needed in case of corruption).
1154 	 *
1155 	 * However, if we got here we have a journal problem too, so we
1156 	 * can't really start a transaction to mark the superblock.
1157 	 * Chicken out and just set the flag on the hope it will be written
1158 	 * to disk, and if not - we will simply wait until next fsck.
1159 	 */
1160 exit_err:
1161 	if (err) {
1162 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1163 			     "forcing fsck on next reboot", group, err);
1164 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1165 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1166 		mark_buffer_dirty(sbi->s_sbh);
1167 	}
1168 }
1169 
1170 /*
1171  * ext4_add_new_descs() adds @count group descriptor of groups
1172  * starting at @group
1173  *
1174  * @handle: journal handle
1175  * @sb: super block
1176  * @group: the group no. of the first group desc to be added
1177  * @resize_inode: the resize inode
1178  * @count: number of group descriptors to be added
1179  */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1180 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1181 			      ext4_group_t group, struct inode *resize_inode,
1182 			      ext4_group_t count)
1183 {
1184 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1185 	struct ext4_super_block *es = sbi->s_es;
1186 	struct buffer_head *gdb_bh;
1187 	int i, gdb_off, gdb_num, err = 0;
1188 	int meta_bg;
1189 
1190 	meta_bg = ext4_has_feature_meta_bg(sb);
1191 	for (i = 0; i < count; i++, group++) {
1192 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1193 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1194 
1195 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1196 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1197 
1198 		/*
1199 		 * We will only either add reserved group blocks to a backup group
1200 		 * or remove reserved blocks for the first group in a new group block.
1201 		 * Doing both would be mean more complex code, and sane people don't
1202 		 * use non-sparse filesystems anymore.  This is already checked above.
1203 		 */
1204 		if (gdb_off) {
1205 			gdb_bh = sbi->s_group_desc[gdb_num];
1206 			BUFFER_TRACE(gdb_bh, "get_write_access");
1207 			err = ext4_journal_get_write_access(handle, gdb_bh);
1208 
1209 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1210 				err = reserve_backup_gdb(handle, resize_inode, group);
1211 		} else if (meta_bg != 0) {
1212 			err = add_new_gdb_meta_bg(sb, handle, group);
1213 		} else {
1214 			err = add_new_gdb(handle, resize_inode, group);
1215 		}
1216 		if (err)
1217 			break;
1218 	}
1219 	return err;
1220 }
1221 
ext4_get_bitmap(struct super_block * sb,__u64 block)1222 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1223 {
1224 	struct buffer_head *bh = sb_getblk(sb, block);
1225 	if (unlikely(!bh))
1226 		return NULL;
1227 	if (!bh_uptodate_or_lock(bh)) {
1228 		if (bh_submit_read(bh) < 0) {
1229 			brelse(bh);
1230 			return NULL;
1231 		}
1232 	}
1233 
1234 	return bh;
1235 }
1236 
ext4_set_bitmap_checksums(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1237 static int ext4_set_bitmap_checksums(struct super_block *sb,
1238 				     ext4_group_t group,
1239 				     struct ext4_group_desc *gdp,
1240 				     struct ext4_new_group_data *group_data)
1241 {
1242 	struct buffer_head *bh;
1243 
1244 	if (!ext4_has_metadata_csum(sb))
1245 		return 0;
1246 
1247 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1248 	if (!bh)
1249 		return -EIO;
1250 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1251 				   EXT4_INODES_PER_GROUP(sb) / 8);
1252 	brelse(bh);
1253 
1254 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1255 	if (!bh)
1256 		return -EIO;
1257 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1258 	brelse(bh);
1259 
1260 	return 0;
1261 }
1262 
1263 /*
1264  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1265  */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1266 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1267 				struct ext4_new_flex_group_data *flex_gd)
1268 {
1269 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1270 	struct ext4_group_desc		*gdp;
1271 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1272 	struct buffer_head		*gdb_bh;
1273 	ext4_group_t			group;
1274 	__u16				*bg_flags = flex_gd->bg_flags;
1275 	int				i, gdb_off, gdb_num, err = 0;
1276 
1277 
1278 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1279 		group = group_data->group;
1280 
1281 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1282 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1283 
1284 		/*
1285 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1286 		 */
1287 		gdb_bh = sbi->s_group_desc[gdb_num];
1288 		/* Update group descriptor block for new group */
1289 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1290 						 gdb_off * EXT4_DESC_SIZE(sb));
1291 
1292 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1293 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1294 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1295 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1296 		if (err) {
1297 			ext4_std_error(sb, err);
1298 			break;
1299 		}
1300 
1301 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1302 		ext4_free_group_clusters_set(sb, gdp,
1303 					     group_data->free_clusters_count);
1304 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1305 		if (ext4_has_group_desc_csum(sb))
1306 			ext4_itable_unused_set(sb, gdp,
1307 					       EXT4_INODES_PER_GROUP(sb));
1308 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1309 		ext4_group_desc_csum_set(sb, group, gdp);
1310 
1311 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1312 		if (unlikely(err)) {
1313 			ext4_std_error(sb, err);
1314 			break;
1315 		}
1316 
1317 		/*
1318 		 * We can allocate memory for mb_alloc based on the new group
1319 		 * descriptor
1320 		 */
1321 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1322 		if (err)
1323 			break;
1324 	}
1325 	return err;
1326 }
1327 
1328 /*
1329  * ext4_update_super() updates the super block so that the newly added
1330  * groups can be seen by the filesystem.
1331  *
1332  * @sb: super block
1333  * @flex_gd: new added groups
1334  */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1335 static void ext4_update_super(struct super_block *sb,
1336 			     struct ext4_new_flex_group_data *flex_gd)
1337 {
1338 	ext4_fsblk_t blocks_count = 0;
1339 	ext4_fsblk_t free_blocks = 0;
1340 	ext4_fsblk_t reserved_blocks = 0;
1341 	struct ext4_new_group_data *group_data = flex_gd->groups;
1342 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1343 	struct ext4_super_block *es = sbi->s_es;
1344 	int i;
1345 
1346 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1347 	/*
1348 	 * Make the new blocks and inodes valid next.  We do this before
1349 	 * increasing the group count so that once the group is enabled,
1350 	 * all of its blocks and inodes are already valid.
1351 	 *
1352 	 * We always allocate group-by-group, then block-by-block or
1353 	 * inode-by-inode within a group, so enabling these
1354 	 * blocks/inodes before the group is live won't actually let us
1355 	 * allocate the new space yet.
1356 	 */
1357 	for (i = 0; i < flex_gd->count; i++) {
1358 		blocks_count += group_data[i].blocks_count;
1359 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1360 	}
1361 
1362 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1363 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1364 	reserved_blocks *= blocks_count;
1365 	do_div(reserved_blocks, 100);
1366 
1367 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1368 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1369 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1370 		     flex_gd->count);
1371 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1372 		     flex_gd->count);
1373 
1374 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1375 	/*
1376 	 * We need to protect s_groups_count against other CPUs seeing
1377 	 * inconsistent state in the superblock.
1378 	 *
1379 	 * The precise rules we use are:
1380 	 *
1381 	 * * Writers must perform a smp_wmb() after updating all
1382 	 *   dependent data and before modifying the groups count
1383 	 *
1384 	 * * Readers must perform an smp_rmb() after reading the groups
1385 	 *   count and before reading any dependent data.
1386 	 *
1387 	 * NB. These rules can be relaxed when checking the group count
1388 	 * while freeing data, as we can only allocate from a block
1389 	 * group after serialising against the group count, and we can
1390 	 * only then free after serialising in turn against that
1391 	 * allocation.
1392 	 */
1393 	smp_wmb();
1394 
1395 	/* Update the global fs size fields */
1396 	sbi->s_groups_count += flex_gd->count;
1397 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1398 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1399 
1400 	/* Update the reserved block counts only once the new group is
1401 	 * active. */
1402 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1403 				reserved_blocks);
1404 
1405 	/* Update the free space counts */
1406 	percpu_counter_add(&sbi->s_freeclusters_counter,
1407 			   EXT4_NUM_B2C(sbi, free_blocks));
1408 	percpu_counter_add(&sbi->s_freeinodes_counter,
1409 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1410 
1411 	ext4_debug("free blocks count %llu",
1412 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1413 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1414 		ext4_group_t flex_group;
1415 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1416 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1417 			     &sbi->s_flex_groups[flex_group].free_clusters);
1418 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1419 			   &sbi->s_flex_groups[flex_group].free_inodes);
1420 	}
1421 
1422 	/*
1423 	 * Update the fs overhead information
1424 	 */
1425 	ext4_calculate_overhead(sb);
1426 
1427 	if (test_opt(sb, DEBUG))
1428 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1429 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1430 		       blocks_count, free_blocks, reserved_blocks);
1431 }
1432 
1433 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1434  * _before_ we start modifying the filesystem, because we cannot abort the
1435  * transaction and not have it write the data to disk.
1436  */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1437 static int ext4_flex_group_add(struct super_block *sb,
1438 			       struct inode *resize_inode,
1439 			       struct ext4_new_flex_group_data *flex_gd)
1440 {
1441 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1442 	struct ext4_super_block *es = sbi->s_es;
1443 	ext4_fsblk_t o_blocks_count;
1444 	ext4_grpblk_t last;
1445 	ext4_group_t group;
1446 	handle_t *handle;
1447 	unsigned reserved_gdb;
1448 	int err = 0, err2 = 0, credit;
1449 
1450 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1451 
1452 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1453 	o_blocks_count = ext4_blocks_count(es);
1454 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1455 	BUG_ON(last);
1456 
1457 	err = setup_new_flex_group_blocks(sb, flex_gd);
1458 	if (err)
1459 		goto exit;
1460 	/*
1461 	 * We will always be modifying at least the superblock and  GDT
1462 	 * blocks.  If we are adding a group past the last current GDT block,
1463 	 * we will also modify the inode and the dindirect block.  If we
1464 	 * are adding a group with superblock/GDT backups  we will also
1465 	 * modify each of the reserved GDT dindirect blocks.
1466 	 */
1467 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1468 	/* GDT blocks */
1469 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1470 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1471 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1472 	if (IS_ERR(handle)) {
1473 		err = PTR_ERR(handle);
1474 		goto exit;
1475 	}
1476 
1477 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1478 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1479 	if (err)
1480 		goto exit_journal;
1481 
1482 	group = flex_gd->groups[0].group;
1483 	BUG_ON(group != sbi->s_groups_count);
1484 	err = ext4_add_new_descs(handle, sb, group,
1485 				resize_inode, flex_gd->count);
1486 	if (err)
1487 		goto exit_journal;
1488 
1489 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1490 	if (err)
1491 		goto exit_journal;
1492 
1493 	ext4_update_super(sb, flex_gd);
1494 
1495 	err = ext4_handle_dirty_super(handle, sb);
1496 
1497 exit_journal:
1498 	err2 = ext4_journal_stop(handle);
1499 	if (!err)
1500 		err = err2;
1501 
1502 	if (!err) {
1503 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1504 		int gdb_num_end = ((group + flex_gd->count - 1) /
1505 				   EXT4_DESC_PER_BLOCK(sb));
1506 		int meta_bg = ext4_has_feature_meta_bg(sb);
1507 		sector_t old_gdb = 0;
1508 
1509 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1510 			       sizeof(struct ext4_super_block), 0);
1511 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1512 			struct buffer_head *gdb_bh;
1513 
1514 			gdb_bh = sbi->s_group_desc[gdb_num];
1515 			if (old_gdb == gdb_bh->b_blocknr)
1516 				continue;
1517 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1518 				       gdb_bh->b_size, meta_bg);
1519 			old_gdb = gdb_bh->b_blocknr;
1520 		}
1521 	}
1522 exit:
1523 	return err;
1524 }
1525 
ext4_setup_next_flex_gd(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t n_blocks_count,unsigned long flexbg_size)1526 static int ext4_setup_next_flex_gd(struct super_block *sb,
1527 				    struct ext4_new_flex_group_data *flex_gd,
1528 				    ext4_fsblk_t n_blocks_count,
1529 				    unsigned long flexbg_size)
1530 {
1531 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1532 	struct ext4_super_block *es = sbi->s_es;
1533 	struct ext4_new_group_data *group_data = flex_gd->groups;
1534 	ext4_fsblk_t o_blocks_count;
1535 	ext4_group_t n_group;
1536 	ext4_group_t group;
1537 	ext4_group_t last_group;
1538 	ext4_grpblk_t last;
1539 	ext4_grpblk_t clusters_per_group;
1540 	unsigned long i;
1541 
1542 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1543 
1544 	o_blocks_count = ext4_blocks_count(es);
1545 
1546 	if (o_blocks_count == n_blocks_count)
1547 		return 0;
1548 
1549 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1550 	BUG_ON(last);
1551 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1552 
1553 	last_group = group | (flexbg_size - 1);
1554 	if (last_group > n_group)
1555 		last_group = n_group;
1556 
1557 	flex_gd->count = last_group - group + 1;
1558 
1559 	for (i = 0; i < flex_gd->count; i++) {
1560 		int overhead;
1561 
1562 		group_data[i].group = group + i;
1563 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1564 		overhead = ext4_group_overhead_blocks(sb, group + i);
1565 		group_data[i].mdata_blocks = overhead;
1566 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1567 		if (ext4_has_group_desc_csum(sb)) {
1568 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1569 					       EXT4_BG_INODE_UNINIT;
1570 			if (!test_opt(sb, INIT_INODE_TABLE))
1571 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1572 		} else
1573 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1574 	}
1575 
1576 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1577 		/* We need to initialize block bitmap of last group. */
1578 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1579 
1580 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1581 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1582 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1583 						       last - 1;
1584 	}
1585 
1586 	return 1;
1587 }
1588 
1589 /* Add group descriptor data to an existing or new group descriptor block.
1590  * Ensure we handle all possible error conditions _before_ we start modifying
1591  * the filesystem, because we cannot abort the transaction and not have it
1592  * write the data to disk.
1593  *
1594  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1595  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1596  *
1597  * We only need to hold the superblock lock while we are actually adding
1598  * in the new group's counts to the superblock.  Prior to that we have
1599  * not really "added" the group at all.  We re-check that we are still
1600  * adding in the last group in case things have changed since verifying.
1601  */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1602 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1603 {
1604 	struct ext4_new_flex_group_data flex_gd;
1605 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1606 	struct ext4_super_block *es = sbi->s_es;
1607 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1608 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1609 	struct inode *inode = NULL;
1610 	int gdb_off;
1611 	int err;
1612 	__u16 bg_flags = 0;
1613 
1614 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1615 
1616 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1617 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1618 		return -EPERM;
1619 	}
1620 
1621 	if (ext4_blocks_count(es) + input->blocks_count <
1622 	    ext4_blocks_count(es)) {
1623 		ext4_warning(sb, "blocks_count overflow");
1624 		return -EINVAL;
1625 	}
1626 
1627 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1628 	    le32_to_cpu(es->s_inodes_count)) {
1629 		ext4_warning(sb, "inodes_count overflow");
1630 		return -EINVAL;
1631 	}
1632 
1633 	if (reserved_gdb || gdb_off == 0) {
1634 		if (ext4_has_feature_resize_inode(sb) ||
1635 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1636 			ext4_warning(sb,
1637 				     "No reserved GDT blocks, can't resize");
1638 			return -EPERM;
1639 		}
1640 		inode = ext4_iget(sb, EXT4_RESIZE_INO);
1641 		if (IS_ERR(inode)) {
1642 			ext4_warning(sb, "Error opening resize inode");
1643 			return PTR_ERR(inode);
1644 		}
1645 	}
1646 
1647 
1648 	err = verify_group_input(sb, input);
1649 	if (err)
1650 		goto out;
1651 
1652 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1653 	if (err)
1654 		goto out;
1655 
1656 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1657 	if (err)
1658 		goto out;
1659 
1660 	flex_gd.count = 1;
1661 	flex_gd.groups = input;
1662 	flex_gd.bg_flags = &bg_flags;
1663 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1664 out:
1665 	iput(inode);
1666 	return err;
1667 } /* ext4_group_add */
1668 
1669 /*
1670  * extend a group without checking assuming that checking has been done.
1671  */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1672 static int ext4_group_extend_no_check(struct super_block *sb,
1673 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1674 {
1675 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1676 	handle_t *handle;
1677 	int err = 0, err2;
1678 
1679 	/* We will update the superblock, one block bitmap, and
1680 	 * one group descriptor via ext4_group_add_blocks().
1681 	 */
1682 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1683 	if (IS_ERR(handle)) {
1684 		err = PTR_ERR(handle);
1685 		ext4_warning(sb, "error %d on journal start", err);
1686 		return err;
1687 	}
1688 
1689 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1690 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1691 	if (err) {
1692 		ext4_warning(sb, "error %d on journal write access", err);
1693 		goto errout;
1694 	}
1695 
1696 	ext4_blocks_count_set(es, o_blocks_count + add);
1697 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1698 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1699 		   o_blocks_count + add);
1700 	/* We add the blocks to the bitmap and set the group need init bit */
1701 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1702 	if (err)
1703 		goto errout;
1704 	ext4_handle_dirty_super(handle, sb);
1705 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1706 		   o_blocks_count + add);
1707 errout:
1708 	err2 = ext4_journal_stop(handle);
1709 	if (err2 && !err)
1710 		err = err2;
1711 
1712 	if (!err) {
1713 		if (test_opt(sb, DEBUG))
1714 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1715 			       "blocks\n", ext4_blocks_count(es));
1716 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1717 			       (char *)es, sizeof(struct ext4_super_block), 0);
1718 	}
1719 	return err;
1720 }
1721 
1722 /*
1723  * Extend the filesystem to the new number of blocks specified.  This entry
1724  * point is only used to extend the current filesystem to the end of the last
1725  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1726  * for emergencies (because it has no dependencies on reserved blocks).
1727  *
1728  * If we _really_ wanted, we could use default values to call ext4_group_add()
1729  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1730  * GDT blocks are reserved to grow to the desired size.
1731  */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1732 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1733 		      ext4_fsblk_t n_blocks_count)
1734 {
1735 	ext4_fsblk_t o_blocks_count;
1736 	ext4_grpblk_t last;
1737 	ext4_grpblk_t add;
1738 	struct buffer_head *bh;
1739 	int err;
1740 	ext4_group_t group;
1741 
1742 	o_blocks_count = ext4_blocks_count(es);
1743 
1744 	if (test_opt(sb, DEBUG))
1745 		ext4_msg(sb, KERN_DEBUG,
1746 			 "extending last group from %llu to %llu blocks",
1747 			 o_blocks_count, n_blocks_count);
1748 
1749 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1750 		return 0;
1751 
1752 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1753 		ext4_msg(sb, KERN_ERR,
1754 			 "filesystem too large to resize to %llu blocks safely",
1755 			 n_blocks_count);
1756 		if (sizeof(sector_t) < 8)
1757 			ext4_warning(sb, "CONFIG_LBDAF not enabled");
1758 		return -EINVAL;
1759 	}
1760 
1761 	if (n_blocks_count < o_blocks_count) {
1762 		ext4_warning(sb, "can't shrink FS - resize aborted");
1763 		return -EINVAL;
1764 	}
1765 
1766 	/* Handle the remaining blocks in the last group only. */
1767 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1768 
1769 	if (last == 0) {
1770 		ext4_warning(sb, "need to use ext2online to resize further");
1771 		return -EPERM;
1772 	}
1773 
1774 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1775 
1776 	if (o_blocks_count + add < o_blocks_count) {
1777 		ext4_warning(sb, "blocks_count overflow");
1778 		return -EINVAL;
1779 	}
1780 
1781 	if (o_blocks_count + add > n_blocks_count)
1782 		add = n_blocks_count - o_blocks_count;
1783 
1784 	if (o_blocks_count + add < n_blocks_count)
1785 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1786 			     o_blocks_count + add, add);
1787 
1788 	/* See if the device is actually as big as what was requested */
1789 	bh = sb_bread(sb, o_blocks_count + add - 1);
1790 	if (!bh) {
1791 		ext4_warning(sb, "can't read last block, resize aborted");
1792 		return -ENOSPC;
1793 	}
1794 	brelse(bh);
1795 
1796 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1797 	return err;
1798 } /* ext4_group_extend */
1799 
1800 
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1801 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1802 {
1803 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1804 }
1805 
1806 /*
1807  * Release the resize inode and drop the resize_inode feature if there
1808  * are no more reserved gdt blocks, and then convert the file system
1809  * to enable meta_bg
1810  */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1811 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1812 {
1813 	handle_t *handle;
1814 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1815 	struct ext4_super_block *es = sbi->s_es;
1816 	struct ext4_inode_info *ei = EXT4_I(inode);
1817 	ext4_fsblk_t nr;
1818 	int i, ret, err = 0;
1819 	int credits = 1;
1820 
1821 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1822 	if (inode) {
1823 		if (es->s_reserved_gdt_blocks) {
1824 			ext4_error(sb, "Unexpected non-zero "
1825 				   "s_reserved_gdt_blocks");
1826 			return -EPERM;
1827 		}
1828 
1829 		/* Do a quick sanity check of the resize inode */
1830 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1831 					     (9 - sbi->s_cluster_bits)))
1832 			goto invalid_resize_inode;
1833 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1834 			if (i == EXT4_DIND_BLOCK) {
1835 				if (ei->i_data[i])
1836 					continue;
1837 				else
1838 					goto invalid_resize_inode;
1839 			}
1840 			if (ei->i_data[i])
1841 				goto invalid_resize_inode;
1842 		}
1843 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1844 	}
1845 
1846 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1847 	if (IS_ERR(handle))
1848 		return PTR_ERR(handle);
1849 
1850 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1851 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1852 	if (err)
1853 		goto errout;
1854 
1855 	ext4_clear_feature_resize_inode(sb);
1856 	ext4_set_feature_meta_bg(sb);
1857 	sbi->s_es->s_first_meta_bg =
1858 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1859 
1860 	err = ext4_handle_dirty_super(handle, sb);
1861 	if (err) {
1862 		ext4_std_error(sb, err);
1863 		goto errout;
1864 	}
1865 
1866 	if (inode) {
1867 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1868 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1869 				 EXT4_FREE_BLOCKS_METADATA |
1870 				 EXT4_FREE_BLOCKS_FORGET);
1871 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1872 		inode->i_blocks = 0;
1873 
1874 		err = ext4_mark_inode_dirty(handle, inode);
1875 		if (err)
1876 			ext4_std_error(sb, err);
1877 	}
1878 
1879 errout:
1880 	ret = ext4_journal_stop(handle);
1881 	if (!err)
1882 		err = ret;
1883 	return ret;
1884 
1885 invalid_resize_inode:
1886 	ext4_error(sb, "corrupted/inconsistent resize inode");
1887 	return -EINVAL;
1888 }
1889 
1890 /*
1891  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1892  *
1893  * @sb: super block of the fs to be resized
1894  * @n_blocks_count: the number of blocks resides in the resized fs
1895  */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)1896 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1897 {
1898 	struct ext4_new_flex_group_data *flex_gd = NULL;
1899 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1900 	struct ext4_super_block *es = sbi->s_es;
1901 	struct buffer_head *bh;
1902 	struct inode *resize_inode = NULL;
1903 	ext4_grpblk_t add, offset;
1904 	unsigned long n_desc_blocks;
1905 	unsigned long o_desc_blocks;
1906 	ext4_group_t o_group;
1907 	ext4_group_t n_group;
1908 	ext4_fsblk_t o_blocks_count;
1909 	ext4_fsblk_t n_blocks_count_retry = 0;
1910 	unsigned long last_update_time = 0;
1911 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1912 	int meta_bg;
1913 
1914 	/* See if the device is actually as big as what was requested */
1915 	bh = sb_bread(sb, n_blocks_count - 1);
1916 	if (!bh) {
1917 		ext4_warning(sb, "can't read last block, resize aborted");
1918 		return -ENOSPC;
1919 	}
1920 	brelse(bh);
1921 
1922 retry:
1923 	o_blocks_count = ext4_blocks_count(es);
1924 
1925 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1926 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1927 
1928 	if (n_blocks_count < o_blocks_count) {
1929 		/* On-line shrinking not supported */
1930 		ext4_warning(sb, "can't shrink FS - resize aborted");
1931 		return -EINVAL;
1932 	}
1933 
1934 	if (n_blocks_count == o_blocks_count)
1935 		/* Nothing need to do */
1936 		return 0;
1937 
1938 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1939 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1940 		ext4_warning(sb, "resize would cause inodes_count overflow");
1941 		return -EINVAL;
1942 	}
1943 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1944 
1945 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1946 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1947 
1948 	meta_bg = ext4_has_feature_meta_bg(sb);
1949 
1950 	if (ext4_has_feature_resize_inode(sb)) {
1951 		if (meta_bg) {
1952 			ext4_error(sb, "resize_inode and meta_bg enabled "
1953 				   "simultaneously");
1954 			return -EINVAL;
1955 		}
1956 		if (n_desc_blocks > o_desc_blocks +
1957 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1958 			n_blocks_count_retry = n_blocks_count;
1959 			n_desc_blocks = o_desc_blocks +
1960 				le16_to_cpu(es->s_reserved_gdt_blocks);
1961 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1962 			n_blocks_count = (ext4_fsblk_t)n_group *
1963 				EXT4_BLOCKS_PER_GROUP(sb);
1964 			n_group--; /* set to last group number */
1965 		}
1966 
1967 		if (!resize_inode)
1968 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1969 		if (IS_ERR(resize_inode)) {
1970 			ext4_warning(sb, "Error opening resize inode");
1971 			return PTR_ERR(resize_inode);
1972 		}
1973 	}
1974 
1975 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1976 		err = ext4_convert_meta_bg(sb, resize_inode);
1977 		if (err)
1978 			goto out;
1979 		if (resize_inode) {
1980 			iput(resize_inode);
1981 			resize_inode = NULL;
1982 		}
1983 		if (n_blocks_count_retry) {
1984 			n_blocks_count = n_blocks_count_retry;
1985 			n_blocks_count_retry = 0;
1986 			goto retry;
1987 		}
1988 	}
1989 
1990 	/*
1991 	 * Make sure the last group has enough space so that it's
1992 	 * guaranteed to have enough space for all metadata blocks
1993 	 * that it might need to hold.  (We might not need to store
1994 	 * the inode table blocks in the last block group, but there
1995 	 * will be cases where this might be needed.)
1996 	 */
1997 	if ((ext4_group_first_block_no(sb, n_group) +
1998 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
1999 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2000 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2001 		n_group--;
2002 		n_blocks_count_retry = 0;
2003 		if (resize_inode) {
2004 			iput(resize_inode);
2005 			resize_inode = NULL;
2006 		}
2007 		goto retry;
2008 	}
2009 
2010 	/* extend the last group */
2011 	if (n_group == o_group)
2012 		add = n_blocks_count - o_blocks_count;
2013 	else
2014 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2015 	if (add > 0) {
2016 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2017 		if (err)
2018 			goto out;
2019 	}
2020 
2021 	if (ext4_blocks_count(es) == n_blocks_count)
2022 		goto out;
2023 
2024 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2025 	if (err)
2026 		return err;
2027 
2028 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2029 	if (err)
2030 		goto out;
2031 
2032 	flex_gd = alloc_flex_gd(flexbg_size);
2033 	if (flex_gd == NULL) {
2034 		err = -ENOMEM;
2035 		goto out;
2036 	}
2037 
2038 	/* Add flex groups. Note that a regular group is a
2039 	 * flex group with 1 group.
2040 	 */
2041 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2042 					      flexbg_size)) {
2043 		if (jiffies - last_update_time > HZ * 10) {
2044 			if (last_update_time)
2045 				ext4_msg(sb, KERN_INFO,
2046 					 "resized to %llu blocks",
2047 					 ext4_blocks_count(es));
2048 			last_update_time = jiffies;
2049 		}
2050 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2051 			break;
2052 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2053 		if (unlikely(err))
2054 			break;
2055 	}
2056 
2057 	if (!err && n_blocks_count_retry) {
2058 		n_blocks_count = n_blocks_count_retry;
2059 		n_blocks_count_retry = 0;
2060 		free_flex_gd(flex_gd);
2061 		flex_gd = NULL;
2062 		goto retry;
2063 	}
2064 
2065 out:
2066 	if (flex_gd)
2067 		free_flex_gd(flex_gd);
2068 	if (resize_inode != NULL)
2069 		iput(resize_inode);
2070 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2071 	return err;
2072 }
2073