• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
131 		err = PTR_ERR(bh);
132 		bh = NULL;
133 		ext4_warning(sb, "Cannot read last block (%llu)",
134 			     end - 1);
135 	} else if (outside(input->block_bitmap, start, end))
136 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
137 			     (unsigned long long)input->block_bitmap);
138 	else if (outside(input->inode_bitmap, start, end))
139 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
140 			     (unsigned long long)input->inode_bitmap);
141 	else if (outside(input->inode_table, start, end) ||
142 		 outside(itend - 1, start, end))
143 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
144 			     (unsigned long long)input->inode_table, itend - 1);
145 	else if (input->inode_bitmap == input->block_bitmap)
146 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
147 			     (unsigned long long)input->block_bitmap);
148 	else if (inside(input->block_bitmap, input->inode_table, itend))
149 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
150 			     "(%llu-%llu)",
151 			     (unsigned long long)input->block_bitmap,
152 			     (unsigned long long)input->inode_table, itend - 1);
153 	else if (inside(input->inode_bitmap, input->inode_table, itend))
154 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
155 			     "(%llu-%llu)",
156 			     (unsigned long long)input->inode_bitmap,
157 			     (unsigned long long)input->inode_table, itend - 1);
158 	else if (inside(input->block_bitmap, start, metaend))
159 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
160 			     (unsigned long long)input->block_bitmap,
161 			     start, metaend - 1);
162 	else if (inside(input->inode_bitmap, start, metaend))
163 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
164 			     (unsigned long long)input->inode_bitmap,
165 			     start, metaend - 1);
166 	else if (inside(input->inode_table, start, metaend) ||
167 		 inside(itend - 1, start, metaend))
168 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
169 			     "(%llu-%llu)",
170 			     (unsigned long long)input->inode_table,
171 			     itend - 1, start, metaend - 1);
172 	else
173 		err = 0;
174 	brelse(bh);
175 
176 	return err;
177 }
178 
179 /*
180  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
181  * group each time.
182  */
183 struct ext4_new_flex_group_data {
184 	struct ext4_new_group_data *groups;	/* new_group_data for groups
185 						   in the flex group */
186 	__u16 *bg_flags;			/* block group flags of groups
187 						   in @groups */
188 	ext4_group_t count;			/* number of groups in @groups
189 						 */
190 };
191 
192 /*
193  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
194  * @flexbg_size.
195  *
196  * Returns NULL on failure otherwise address of the allocated structure.
197  */
alloc_flex_gd(unsigned long flexbg_size)198 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
199 {
200 	struct ext4_new_flex_group_data *flex_gd;
201 
202 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
203 	if (flex_gd == NULL)
204 		goto out3;
205 
206 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
207 		goto out2;
208 	flex_gd->count = flexbg_size;
209 
210 	flex_gd->groups = kmalloc_array(flexbg_size,
211 					sizeof(struct ext4_new_group_data),
212 					GFP_NOFS);
213 	if (flex_gd->groups == NULL)
214 		goto out2;
215 
216 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
217 					  GFP_NOFS);
218 	if (flex_gd->bg_flags == NULL)
219 		goto out1;
220 
221 	return flex_gd;
222 
223 out1:
224 	kfree(flex_gd->groups);
225 out2:
226 	kfree(flex_gd);
227 out3:
228 	return NULL;
229 }
230 
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)231 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
232 {
233 	kfree(flex_gd->bg_flags);
234 	kfree(flex_gd->groups);
235 	kfree(flex_gd);
236 }
237 
238 /*
239  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
240  * and inode tables for a flex group.
241  *
242  * This function is used by 64bit-resize.  Note that this function allocates
243  * group tables from the 1st group of groups contained by @flexgd, which may
244  * be a partial of a flex group.
245  *
246  * @sb: super block of fs to which the groups belongs
247  *
248  * Returns 0 on a successful allocation of the metadata blocks in the
249  * block group.
250  */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,int flexbg_size)251 static int ext4_alloc_group_tables(struct super_block *sb,
252 				struct ext4_new_flex_group_data *flex_gd,
253 				int flexbg_size)
254 {
255 	struct ext4_new_group_data *group_data = flex_gd->groups;
256 	ext4_fsblk_t start_blk;
257 	ext4_fsblk_t last_blk;
258 	ext4_group_t src_group;
259 	ext4_group_t bb_index = 0;
260 	ext4_group_t ib_index = 0;
261 	ext4_group_t it_index = 0;
262 	ext4_group_t group;
263 	ext4_group_t last_group;
264 	unsigned overhead;
265 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
266 	int i;
267 
268 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
269 
270 	src_group = group_data[0].group;
271 	last_group  = src_group + flex_gd->count - 1;
272 
273 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
274 	       (last_group & ~(flexbg_size - 1))));
275 next_group:
276 	group = group_data[0].group;
277 	if (src_group >= group_data[0].group + flex_gd->count)
278 		return -ENOSPC;
279 	start_blk = ext4_group_first_block_no(sb, src_group);
280 	last_blk = start_blk + group_data[src_group - group].blocks_count;
281 
282 	overhead = ext4_group_overhead_blocks(sb, src_group);
283 
284 	start_blk += overhead;
285 
286 	/* We collect contiguous blocks as much as possible. */
287 	src_group++;
288 	for (; src_group <= last_group; src_group++) {
289 		overhead = ext4_group_overhead_blocks(sb, src_group);
290 		if (overhead == 0)
291 			last_blk += group_data[src_group - group].blocks_count;
292 		else
293 			break;
294 	}
295 
296 	/* Allocate block bitmaps */
297 	for (; bb_index < flex_gd->count; bb_index++) {
298 		if (start_blk >= last_blk)
299 			goto next_group;
300 		group_data[bb_index].block_bitmap = start_blk++;
301 		group = ext4_get_group_number(sb, start_blk - 1);
302 		group -= group_data[0].group;
303 		group_data[group].mdata_blocks++;
304 		flex_gd->bg_flags[group] &= uninit_mask;
305 	}
306 
307 	/* Allocate inode bitmaps */
308 	for (; ib_index < flex_gd->count; ib_index++) {
309 		if (start_blk >= last_blk)
310 			goto next_group;
311 		group_data[ib_index].inode_bitmap = start_blk++;
312 		group = ext4_get_group_number(sb, start_blk - 1);
313 		group -= group_data[0].group;
314 		group_data[group].mdata_blocks++;
315 		flex_gd->bg_flags[group] &= uninit_mask;
316 	}
317 
318 	/* Allocate inode tables */
319 	for (; it_index < flex_gd->count; it_index++) {
320 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
321 		ext4_fsblk_t next_group_start;
322 
323 		if (start_blk + itb > last_blk)
324 			goto next_group;
325 		group_data[it_index].inode_table = start_blk;
326 		group = ext4_get_group_number(sb, start_blk);
327 		next_group_start = ext4_group_first_block_no(sb, group + 1);
328 		group -= group_data[0].group;
329 
330 		if (start_blk + itb > next_group_start) {
331 			flex_gd->bg_flags[group + 1] &= uninit_mask;
332 			overhead = start_blk + itb - next_group_start;
333 			group_data[group + 1].mdata_blocks += overhead;
334 			itb -= overhead;
335 		}
336 
337 		group_data[group].mdata_blocks += itb;
338 		flex_gd->bg_flags[group] &= uninit_mask;
339 		start_blk += EXT4_SB(sb)->s_itb_per_group;
340 	}
341 
342 	/* Update free clusters count to exclude metadata blocks */
343 	for (i = 0; i < flex_gd->count; i++) {
344 		group_data[i].free_clusters_count -=
345 				EXT4_NUM_B2C(EXT4_SB(sb),
346 					     group_data[i].mdata_blocks);
347 	}
348 
349 	if (test_opt(sb, DEBUG)) {
350 		int i;
351 		group = group_data[0].group;
352 
353 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
354 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
355 		       flexbg_size);
356 
357 		for (i = 0; i < flex_gd->count; i++) {
358 			ext4_debug(
359 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
360 			       ext4_bg_has_super(sb, group + i) ? "normal" :
361 			       "no-super", group + i,
362 			       group_data[i].blocks_count,
363 			       group_data[i].free_clusters_count,
364 			       group_data[i].mdata_blocks);
365 		}
366 	}
367 	return 0;
368 }
369 
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)370 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
371 				  ext4_fsblk_t blk)
372 {
373 	struct buffer_head *bh;
374 	int err;
375 
376 	bh = sb_getblk(sb, blk);
377 	if (unlikely(!bh))
378 		return ERR_PTR(-ENOMEM);
379 	BUFFER_TRACE(bh, "get_write_access");
380 	if ((err = ext4_journal_get_write_access(handle, bh))) {
381 		brelse(bh);
382 		bh = ERR_PTR(err);
383 	} else {
384 		memset(bh->b_data, 0, sb->s_blocksize);
385 		set_buffer_uptodate(bh);
386 	}
387 
388 	return bh;
389 }
390 
391 /*
392  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
393  * If that fails, restart the transaction & regain write access for the
394  * buffer head which is used for block_bitmap modifications.
395  */
extend_or_restart_transaction(handle_t * handle,int thresh)396 static int extend_or_restart_transaction(handle_t *handle, int thresh)
397 {
398 	int err;
399 
400 	if (ext4_handle_has_enough_credits(handle, thresh))
401 		return 0;
402 
403 	err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
404 	if (err < 0)
405 		return err;
406 	if (err) {
407 		err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
408 		if (err)
409 			return err;
410 	}
411 
412 	return 0;
413 }
414 
415 /*
416  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
417  *
418  * Helper function for ext4_setup_new_group_blocks() which set .
419  *
420  * @sb: super block
421  * @handle: journal handle
422  * @flex_gd: flex group data
423  */
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)424 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
425 			struct ext4_new_flex_group_data *flex_gd,
426 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
427 {
428 	struct ext4_sb_info *sbi = EXT4_SB(sb);
429 	ext4_group_t count = last_cluster - first_cluster + 1;
430 	ext4_group_t count2;
431 
432 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
433 		   last_cluster);
434 	for (count2 = count; count > 0;
435 	     count -= count2, first_cluster += count2) {
436 		ext4_fsblk_t start;
437 		struct buffer_head *bh;
438 		ext4_group_t group;
439 		int err;
440 
441 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
442 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
443 		group -= flex_gd->groups[0].group;
444 
445 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
446 		if (count2 > count)
447 			count2 = count;
448 
449 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
450 			BUG_ON(flex_gd->count > 1);
451 			continue;
452 		}
453 
454 		err = extend_or_restart_transaction(handle, 1);
455 		if (err)
456 			return err;
457 
458 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
459 		if (unlikely(!bh))
460 			return -ENOMEM;
461 
462 		BUFFER_TRACE(bh, "get_write_access");
463 		err = ext4_journal_get_write_access(handle, bh);
464 		if (err) {
465 			brelse(bh);
466 			return err;
467 		}
468 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
469 			   first_cluster, first_cluster - start, count2);
470 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
471 
472 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
473 		brelse(bh);
474 		if (unlikely(err))
475 			return err;
476 	}
477 
478 	return 0;
479 }
480 
481 /*
482  * Set up the block and inode bitmaps, and the inode table for the new groups.
483  * This doesn't need to be part of the main transaction, since we are only
484  * changing blocks outside the actual filesystem.  We still do journaling to
485  * ensure the recovery is correct in case of a failure just after resize.
486  * If any part of this fails, we simply abort the resize.
487  *
488  * setup_new_flex_group_blocks handles a flex group as follow:
489  *  1. copy super block and GDT, and initialize group tables if necessary.
490  *     In this step, we only set bits in blocks bitmaps for blocks taken by
491  *     super block and GDT.
492  *  2. allocate group tables in block bitmaps, that is, set bits in block
493  *     bitmap for blocks taken by group tables.
494  */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)495 static int setup_new_flex_group_blocks(struct super_block *sb,
496 				struct ext4_new_flex_group_data *flex_gd)
497 {
498 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
499 	ext4_fsblk_t start;
500 	ext4_fsblk_t block;
501 	struct ext4_sb_info *sbi = EXT4_SB(sb);
502 	struct ext4_super_block *es = sbi->s_es;
503 	struct ext4_new_group_data *group_data = flex_gd->groups;
504 	__u16 *bg_flags = flex_gd->bg_flags;
505 	handle_t *handle;
506 	ext4_group_t group, count;
507 	struct buffer_head *bh = NULL;
508 	int reserved_gdb, i, j, err = 0, err2;
509 	int meta_bg;
510 
511 	BUG_ON(!flex_gd->count || !group_data ||
512 	       group_data[0].group != sbi->s_groups_count);
513 
514 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
515 	meta_bg = ext4_has_feature_meta_bg(sb);
516 
517 	/* This transaction may be extended/restarted along the way */
518 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
519 	if (IS_ERR(handle))
520 		return PTR_ERR(handle);
521 
522 	group = group_data[0].group;
523 	for (i = 0; i < flex_gd->count; i++, group++) {
524 		unsigned long gdblocks;
525 		ext4_grpblk_t overhead;
526 
527 		gdblocks = ext4_bg_num_gdb(sb, group);
528 		start = ext4_group_first_block_no(sb, group);
529 
530 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
531 			goto handle_itb;
532 
533 		if (meta_bg == 1) {
534 			ext4_group_t first_group;
535 			first_group = ext4_meta_bg_first_group(sb, group);
536 			if (first_group != group + 1 &&
537 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
538 				goto handle_itb;
539 		}
540 
541 		block = start + ext4_bg_has_super(sb, group);
542 		/* Copy all of the GDT blocks into the backup in this group */
543 		for (j = 0; j < gdblocks; j++, block++) {
544 			struct buffer_head *gdb;
545 
546 			ext4_debug("update backup group %#04llx\n", block);
547 			err = extend_or_restart_transaction(handle, 1);
548 			if (err)
549 				goto out;
550 
551 			gdb = sb_getblk(sb, block);
552 			if (unlikely(!gdb)) {
553 				err = -ENOMEM;
554 				goto out;
555 			}
556 
557 			BUFFER_TRACE(gdb, "get_write_access");
558 			err = ext4_journal_get_write_access(handle, gdb);
559 			if (err) {
560 				brelse(gdb);
561 				goto out;
562 			}
563 			memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
564 			       gdb->b_size);
565 			set_buffer_uptodate(gdb);
566 
567 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
568 			if (unlikely(err)) {
569 				brelse(gdb);
570 				goto out;
571 			}
572 			brelse(gdb);
573 		}
574 
575 		/* Zero out all of the reserved backup group descriptor
576 		 * table blocks
577 		 */
578 		if (ext4_bg_has_super(sb, group)) {
579 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
580 					reserved_gdb, GFP_NOFS);
581 			if (err)
582 				goto out;
583 		}
584 
585 handle_itb:
586 		/* Initialize group tables of the grop @group */
587 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
588 			goto handle_bb;
589 
590 		/* Zero out all of the inode table blocks */
591 		block = group_data[i].inode_table;
592 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
593 			   block, sbi->s_itb_per_group);
594 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
595 				       GFP_NOFS);
596 		if (err)
597 			goto out;
598 
599 handle_bb:
600 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
601 			goto handle_ib;
602 
603 		/* Initialize block bitmap of the @group */
604 		block = group_data[i].block_bitmap;
605 		err = extend_or_restart_transaction(handle, 1);
606 		if (err)
607 			goto out;
608 
609 		bh = bclean(handle, sb, block);
610 		if (IS_ERR(bh)) {
611 			err = PTR_ERR(bh);
612 			goto out;
613 		}
614 		overhead = ext4_group_overhead_blocks(sb, group);
615 		if (overhead != 0) {
616 			ext4_debug("mark backup superblock %#04llx (+0)\n",
617 				   start);
618 			ext4_set_bits(bh->b_data, 0,
619 				      EXT4_NUM_B2C(sbi, overhead));
620 		}
621 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
622 				     sb->s_blocksize * 8, bh->b_data);
623 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
624 		brelse(bh);
625 		if (err)
626 			goto out;
627 
628 handle_ib:
629 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
630 			continue;
631 
632 		/* Initialize inode bitmap of the @group */
633 		block = group_data[i].inode_bitmap;
634 		err = extend_or_restart_transaction(handle, 1);
635 		if (err)
636 			goto out;
637 		/* Mark unused entries in inode bitmap used */
638 		bh = bclean(handle, sb, block);
639 		if (IS_ERR(bh)) {
640 			err = PTR_ERR(bh);
641 			goto out;
642 		}
643 
644 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
645 				     sb->s_blocksize * 8, bh->b_data);
646 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
647 		brelse(bh);
648 		if (err)
649 			goto out;
650 	}
651 
652 	/* Mark group tables in block bitmap */
653 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
654 		count = group_table_count[j];
655 		start = (&group_data[0].block_bitmap)[j];
656 		block = start;
657 		for (i = 1; i < flex_gd->count; i++) {
658 			block += group_table_count[j];
659 			if (block == (&group_data[i].block_bitmap)[j]) {
660 				count += group_table_count[j];
661 				continue;
662 			}
663 			err = set_flexbg_block_bitmap(sb, handle,
664 						      flex_gd,
665 						      EXT4_B2C(sbi, start),
666 						      EXT4_B2C(sbi,
667 							       start + count
668 							       - 1));
669 			if (err)
670 				goto out;
671 			count = group_table_count[j];
672 			start = (&group_data[i].block_bitmap)[j];
673 			block = start;
674 		}
675 
676 		if (count) {
677 			err = set_flexbg_block_bitmap(sb, handle,
678 						      flex_gd,
679 						      EXT4_B2C(sbi, start),
680 						      EXT4_B2C(sbi,
681 							       start + count
682 							       - 1));
683 			if (err)
684 				goto out;
685 		}
686 	}
687 
688 out:
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 = NULL;
787 	struct buffer_head *dind = NULL;
788 	struct buffer_head *gdb_bh = NULL;
789 	int gdbackups;
790 	struct ext4_iloc iloc = { .bh = NULL };
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 = ext4_sb_bread(sb, gdblock, 0);
800 	if (IS_ERR(gdb_bh))
801 		return PTR_ERR(gdb_bh);
802 
803 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
804 	if (gdbackups < 0) {
805 		err = gdbackups;
806 		goto errout;
807 	}
808 
809 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
810 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
811 	if (IS_ERR(dind)) {
812 		err = PTR_ERR(dind);
813 		dind = NULL;
814 		goto errout;
815 	}
816 
817 	data = (__le32 *)dind->b_data;
818 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
819 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
820 			     group, gdblock);
821 		err = -EINVAL;
822 		goto errout;
823 	}
824 
825 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
826 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
827 	if (unlikely(err))
828 		goto errout;
829 
830 	BUFFER_TRACE(gdb_bh, "get_write_access");
831 	err = ext4_journal_get_write_access(handle, gdb_bh);
832 	if (unlikely(err))
833 		goto errout;
834 
835 	BUFFER_TRACE(dind, "get_write_access");
836 	err = ext4_journal_get_write_access(handle, dind);
837 	if (unlikely(err))
838 		ext4_std_error(sb, err);
839 
840 	/* ext4_reserve_inode_write() gets a reference on the iloc */
841 	err = ext4_reserve_inode_write(handle, inode, &iloc);
842 	if (unlikely(err))
843 		goto errout;
844 
845 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
846 				     sizeof(struct buffer_head *),
847 				     GFP_NOFS);
848 	if (!n_group_desc) {
849 		err = -ENOMEM;
850 		ext4_warning(sb, "not enough memory for %lu groups",
851 			     gdb_num + 1);
852 		goto errout;
853 	}
854 
855 	/*
856 	 * Finally, we have all of the possible failures behind us...
857 	 *
858 	 * Remove new GDT block from inode double-indirect block and clear out
859 	 * the new GDT block for use (which also "frees" the backup GDT blocks
860 	 * from the reserved inode).  We don't need to change the bitmaps for
861 	 * these blocks, because they are marked as in-use from being in the
862 	 * reserved inode, and will become GDT blocks (primary and backup).
863 	 */
864 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
865 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
866 	if (unlikely(err)) {
867 		ext4_std_error(sb, err);
868 		goto errout;
869 	}
870 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
871 			   (9 - EXT4_SB(sb)->s_cluster_bits);
872 	ext4_mark_iloc_dirty(handle, inode, &iloc);
873 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
874 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
875 	if (unlikely(err)) {
876 		ext4_std_error(sb, err);
877 		iloc.bh = NULL;
878 		goto errout;
879 	}
880 	brelse(dind);
881 
882 	o_group_desc = EXT4_SB(sb)->s_group_desc;
883 	memcpy(n_group_desc, o_group_desc,
884 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
885 	n_group_desc[gdb_num] = gdb_bh;
886 	EXT4_SB(sb)->s_group_desc = n_group_desc;
887 	EXT4_SB(sb)->s_gdb_count++;
888 	kvfree(o_group_desc);
889 
890 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
891 	err = ext4_handle_dirty_super(handle, sb);
892 	if (err)
893 		ext4_std_error(sb, err);
894 	return err;
895 errout:
896 	kvfree(n_group_desc);
897 	brelse(iloc.bh);
898 	brelse(dind);
899 	brelse(gdb_bh);
900 
901 	ext4_debug("leaving with error %d\n", err);
902 	return err;
903 }
904 
905 /*
906  * add_new_gdb_meta_bg is the sister of add_new_gdb.
907  */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)908 static int add_new_gdb_meta_bg(struct super_block *sb,
909 			       handle_t *handle, ext4_group_t group) {
910 	ext4_fsblk_t gdblock;
911 	struct buffer_head *gdb_bh;
912 	struct buffer_head **o_group_desc, **n_group_desc;
913 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
914 	int err;
915 
916 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
917 		   ext4_bg_has_super(sb, group);
918 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
919 	if (IS_ERR(gdb_bh))
920 		return PTR_ERR(gdb_bh);
921 	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
922 				     sizeof(struct buffer_head *),
923 				     GFP_NOFS);
924 	if (!n_group_desc) {
925 		brelse(gdb_bh);
926 		err = -ENOMEM;
927 		ext4_warning(sb, "not enough memory for %lu groups",
928 			     gdb_num + 1);
929 		return err;
930 	}
931 
932 	o_group_desc = EXT4_SB(sb)->s_group_desc;
933 	memcpy(n_group_desc, o_group_desc,
934 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
935 	n_group_desc[gdb_num] = gdb_bh;
936 
937 	BUFFER_TRACE(gdb_bh, "get_write_access");
938 	err = ext4_journal_get_write_access(handle, gdb_bh);
939 	if (err) {
940 		kvfree(n_group_desc);
941 		brelse(gdb_bh);
942 		return err;
943 	}
944 
945 	EXT4_SB(sb)->s_group_desc = n_group_desc;
946 	EXT4_SB(sb)->s_gdb_count++;
947 	kvfree(o_group_desc);
948 	return err;
949 }
950 
951 /*
952  * Called when we are adding a new group which has a backup copy of each of
953  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
954  * We need to add these reserved backup GDT blocks to the resize inode, so
955  * that they are kept for future resizing and not allocated to files.
956  *
957  * Each reserved backup GDT block will go into a different indirect block.
958  * The indirect blocks are actually the primary reserved GDT blocks,
959  * so we know in advance what their block numbers are.  We only get the
960  * double-indirect block to verify it is pointing to the primary reserved
961  * GDT blocks so we don't overwrite a data block by accident.  The reserved
962  * backup GDT blocks are stored in their reserved primary GDT block.
963  */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)964 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
965 			      ext4_group_t group)
966 {
967 	struct super_block *sb = inode->i_sb;
968 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
969 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
970 	struct buffer_head **primary;
971 	struct buffer_head *dind;
972 	struct ext4_iloc iloc;
973 	ext4_fsblk_t blk;
974 	__le32 *data, *end;
975 	int gdbackups = 0;
976 	int res, i;
977 	int err;
978 
979 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
980 	if (!primary)
981 		return -ENOMEM;
982 
983 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
984 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
985 	if (IS_ERR(dind)) {
986 		err = PTR_ERR(dind);
987 		dind = NULL;
988 		goto exit_free;
989 	}
990 
991 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
992 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
993 					 EXT4_ADDR_PER_BLOCK(sb));
994 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
995 
996 	/* Get each reserved primary GDT block and verify it holds backups */
997 	for (res = 0; res < reserved_gdb; res++, blk++) {
998 		if (le32_to_cpu(*data) != blk) {
999 			ext4_warning(sb, "reserved block %llu"
1000 				     " not at offset %ld",
1001 				     blk,
1002 				     (long)(data - (__le32 *)dind->b_data));
1003 			err = -EINVAL;
1004 			goto exit_bh;
1005 		}
1006 		primary[res] = ext4_sb_bread(sb, blk, 0);
1007 		if (IS_ERR(primary[res])) {
1008 			err = PTR_ERR(primary[res]);
1009 			primary[res] = NULL;
1010 			goto exit_bh;
1011 		}
1012 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1013 		if (gdbackups < 0) {
1014 			brelse(primary[res]);
1015 			err = gdbackups;
1016 			goto exit_bh;
1017 		}
1018 		if (++data >= end)
1019 			data = (__le32 *)dind->b_data;
1020 	}
1021 
1022 	for (i = 0; i < reserved_gdb; i++) {
1023 		BUFFER_TRACE(primary[i], "get_write_access");
1024 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1025 			goto exit_bh;
1026 	}
1027 
1028 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1029 		goto exit_bh;
1030 
1031 	/*
1032 	 * Finally we can add each of the reserved backup GDT blocks from
1033 	 * the new group to its reserved primary GDT block.
1034 	 */
1035 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1036 	for (i = 0; i < reserved_gdb; i++) {
1037 		int err2;
1038 		data = (__le32 *)primary[i]->b_data;
1039 		/* printk("reserving backup %lu[%u] = %lu\n",
1040 		       primary[i]->b_blocknr, gdbackups,
1041 		       blk + primary[i]->b_blocknr); */
1042 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1043 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1044 		if (!err)
1045 			err = err2;
1046 	}
1047 
1048 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1049 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1050 
1051 exit_bh:
1052 	while (--res >= 0)
1053 		brelse(primary[res]);
1054 	brelse(dind);
1055 
1056 exit_free:
1057 	kfree(primary);
1058 
1059 	return err;
1060 }
1061 
1062 /*
1063  * Update the backup copies of the ext4 metadata.  These don't need to be part
1064  * of the main resize transaction, because e2fsck will re-write them if there
1065  * is a problem (basically only OOM will cause a problem).  However, we
1066  * _should_ update the backups if possible, in case the primary gets trashed
1067  * for some reason and we need to run e2fsck from a backup superblock.  The
1068  * important part is that the new block and inode counts are in the backup
1069  * superblocks, and the location of the new group metadata in the GDT backups.
1070  *
1071  * We do not need take the s_resize_lock for this, because these
1072  * blocks are not otherwise touched by the filesystem code when it is
1073  * mounted.  We don't need to worry about last changing from
1074  * sbi->s_groups_count, because the worst that can happen is that we
1075  * do not copy the full number of backups at this time.  The resize
1076  * which changed s_groups_count will backup again.
1077  */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1078 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1079 			   int size, int meta_bg)
1080 {
1081 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1082 	ext4_group_t last;
1083 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1084 	unsigned three = 1;
1085 	unsigned five = 5;
1086 	unsigned seven = 7;
1087 	ext4_group_t group = 0;
1088 	int rest = sb->s_blocksize - size;
1089 	handle_t *handle;
1090 	int err = 0, err2;
1091 
1092 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1093 	if (IS_ERR(handle)) {
1094 		group = 1;
1095 		err = PTR_ERR(handle);
1096 		goto exit_err;
1097 	}
1098 
1099 	if (meta_bg == 0) {
1100 		group = ext4_list_backups(sb, &three, &five, &seven);
1101 		last = sbi->s_groups_count;
1102 	} else {
1103 		group = ext4_get_group_number(sb, blk_off) + 1;
1104 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1105 	}
1106 
1107 	while (group < sbi->s_groups_count) {
1108 		struct buffer_head *bh;
1109 		ext4_fsblk_t backup_block;
1110 
1111 		/* Out of journal space, and can't get more - abort - so sad */
1112 		if (ext4_handle_valid(handle) &&
1113 		    handle->h_buffer_credits == 0 &&
1114 		    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1115 		    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1116 			break;
1117 
1118 		if (meta_bg == 0)
1119 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1120 		else
1121 			backup_block = (ext4_group_first_block_no(sb, group) +
1122 					ext4_bg_has_super(sb, group));
1123 
1124 		bh = sb_getblk(sb, backup_block);
1125 		if (unlikely(!bh)) {
1126 			err = -ENOMEM;
1127 			break;
1128 		}
1129 		ext4_debug("update metadata backup %llu(+%llu)\n",
1130 			   backup_block, backup_block -
1131 			   ext4_group_first_block_no(sb, group));
1132 		BUFFER_TRACE(bh, "get_write_access");
1133 		if ((err = ext4_journal_get_write_access(handle, bh))) {
1134 			brelse(bh);
1135 			break;
1136 		}
1137 		lock_buffer(bh);
1138 		memcpy(bh->b_data, data, size);
1139 		if (rest)
1140 			memset(bh->b_data + size, 0, rest);
1141 		set_buffer_uptodate(bh);
1142 		unlock_buffer(bh);
1143 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1144 		if (unlikely(err))
1145 			ext4_std_error(sb, err);
1146 		brelse(bh);
1147 
1148 		if (meta_bg == 0)
1149 			group = ext4_list_backups(sb, &three, &five, &seven);
1150 		else if (group == last)
1151 			break;
1152 		else
1153 			group = last;
1154 	}
1155 	if ((err2 = ext4_journal_stop(handle)) && !err)
1156 		err = err2;
1157 
1158 	/*
1159 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1160 	 * late to revert the resize, we shouldn't fail just because of
1161 	 * the backup copies (they are only needed in case of corruption).
1162 	 *
1163 	 * However, if we got here we have a journal problem too, so we
1164 	 * can't really start a transaction to mark the superblock.
1165 	 * Chicken out and just set the flag on the hope it will be written
1166 	 * to disk, and if not - we will simply wait until next fsck.
1167 	 */
1168 exit_err:
1169 	if (err) {
1170 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1171 			     "forcing fsck on next reboot", group, err);
1172 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1173 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1174 		mark_buffer_dirty(sbi->s_sbh);
1175 	}
1176 }
1177 
1178 /*
1179  * ext4_add_new_descs() adds @count group descriptor of groups
1180  * starting at @group
1181  *
1182  * @handle: journal handle
1183  * @sb: super block
1184  * @group: the group no. of the first group desc to be added
1185  * @resize_inode: the resize inode
1186  * @count: number of group descriptors to be added
1187  */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1188 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1189 			      ext4_group_t group, struct inode *resize_inode,
1190 			      ext4_group_t count)
1191 {
1192 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1193 	struct ext4_super_block *es = sbi->s_es;
1194 	struct buffer_head *gdb_bh;
1195 	int i, gdb_off, gdb_num, err = 0;
1196 	int meta_bg;
1197 
1198 	meta_bg = ext4_has_feature_meta_bg(sb);
1199 	for (i = 0; i < count; i++, group++) {
1200 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1201 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1202 
1203 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1204 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1205 
1206 		/*
1207 		 * We will only either add reserved group blocks to a backup group
1208 		 * or remove reserved blocks for the first group in a new group block.
1209 		 * Doing both would be mean more complex code, and sane people don't
1210 		 * use non-sparse filesystems anymore.  This is already checked above.
1211 		 */
1212 		if (gdb_off) {
1213 			gdb_bh = sbi->s_group_desc[gdb_num];
1214 			BUFFER_TRACE(gdb_bh, "get_write_access");
1215 			err = ext4_journal_get_write_access(handle, gdb_bh);
1216 
1217 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1218 				err = reserve_backup_gdb(handle, resize_inode, group);
1219 		} else if (meta_bg != 0) {
1220 			err = add_new_gdb_meta_bg(sb, handle, group);
1221 		} else {
1222 			err = add_new_gdb(handle, resize_inode, group);
1223 		}
1224 		if (err)
1225 			break;
1226 	}
1227 	return err;
1228 }
1229 
ext4_get_bitmap(struct super_block * sb,__u64 block)1230 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1231 {
1232 	struct buffer_head *bh = sb_getblk(sb, block);
1233 	if (unlikely(!bh))
1234 		return NULL;
1235 	if (!bh_uptodate_or_lock(bh)) {
1236 		if (bh_submit_read(bh) < 0) {
1237 			brelse(bh);
1238 			return NULL;
1239 		}
1240 	}
1241 
1242 	return bh;
1243 }
1244 
ext4_set_bitmap_checksums(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1245 static int ext4_set_bitmap_checksums(struct super_block *sb,
1246 				     ext4_group_t group,
1247 				     struct ext4_group_desc *gdp,
1248 				     struct ext4_new_group_data *group_data)
1249 {
1250 	struct buffer_head *bh;
1251 
1252 	if (!ext4_has_metadata_csum(sb))
1253 		return 0;
1254 
1255 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1256 	if (!bh)
1257 		return -EIO;
1258 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1259 				   EXT4_INODES_PER_GROUP(sb) / 8);
1260 	brelse(bh);
1261 
1262 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1263 	if (!bh)
1264 		return -EIO;
1265 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1266 	brelse(bh);
1267 
1268 	return 0;
1269 }
1270 
1271 /*
1272  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1273  */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1274 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1275 				struct ext4_new_flex_group_data *flex_gd)
1276 {
1277 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1278 	struct ext4_group_desc		*gdp;
1279 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1280 	struct buffer_head		*gdb_bh;
1281 	ext4_group_t			group;
1282 	__u16				*bg_flags = flex_gd->bg_flags;
1283 	int				i, gdb_off, gdb_num, err = 0;
1284 
1285 
1286 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1287 		group = group_data->group;
1288 
1289 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1290 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1291 
1292 		/*
1293 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1294 		 */
1295 		gdb_bh = sbi->s_group_desc[gdb_num];
1296 		/* Update group descriptor block for new group */
1297 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1298 						 gdb_off * EXT4_DESC_SIZE(sb));
1299 
1300 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1301 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1302 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1303 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1304 		if (err) {
1305 			ext4_std_error(sb, err);
1306 			break;
1307 		}
1308 
1309 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1310 		ext4_free_group_clusters_set(sb, gdp,
1311 					     group_data->free_clusters_count);
1312 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1313 		if (ext4_has_group_desc_csum(sb))
1314 			ext4_itable_unused_set(sb, gdp,
1315 					       EXT4_INODES_PER_GROUP(sb));
1316 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1317 		ext4_group_desc_csum_set(sb, group, gdp);
1318 
1319 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1320 		if (unlikely(err)) {
1321 			ext4_std_error(sb, err);
1322 			break;
1323 		}
1324 
1325 		/*
1326 		 * We can allocate memory for mb_alloc based on the new group
1327 		 * descriptor
1328 		 */
1329 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1330 		if (err)
1331 			break;
1332 	}
1333 	return err;
1334 }
1335 
1336 /*
1337  * ext4_update_super() updates the super block so that the newly added
1338  * groups can be seen by the filesystem.
1339  *
1340  * @sb: super block
1341  * @flex_gd: new added groups
1342  */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1343 static void ext4_update_super(struct super_block *sb,
1344 			     struct ext4_new_flex_group_data *flex_gd)
1345 {
1346 	ext4_fsblk_t blocks_count = 0;
1347 	ext4_fsblk_t free_blocks = 0;
1348 	ext4_fsblk_t reserved_blocks = 0;
1349 	struct ext4_new_group_data *group_data = flex_gd->groups;
1350 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1351 	struct ext4_super_block *es = sbi->s_es;
1352 	int i;
1353 
1354 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1355 	/*
1356 	 * Make the new blocks and inodes valid next.  We do this before
1357 	 * increasing the group count so that once the group is enabled,
1358 	 * all of its blocks and inodes are already valid.
1359 	 *
1360 	 * We always allocate group-by-group, then block-by-block or
1361 	 * inode-by-inode within a group, so enabling these
1362 	 * blocks/inodes before the group is live won't actually let us
1363 	 * allocate the new space yet.
1364 	 */
1365 	for (i = 0; i < flex_gd->count; i++) {
1366 		blocks_count += group_data[i].blocks_count;
1367 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1368 	}
1369 
1370 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1371 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1372 	reserved_blocks *= blocks_count;
1373 	do_div(reserved_blocks, 100);
1374 
1375 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1376 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1377 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1378 		     flex_gd->count);
1379 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1380 		     flex_gd->count);
1381 
1382 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1383 	/*
1384 	 * We need to protect s_groups_count against other CPUs seeing
1385 	 * inconsistent state in the superblock.
1386 	 *
1387 	 * The precise rules we use are:
1388 	 *
1389 	 * * Writers must perform a smp_wmb() after updating all
1390 	 *   dependent data and before modifying the groups count
1391 	 *
1392 	 * * Readers must perform an smp_rmb() after reading the groups
1393 	 *   count and before reading any dependent data.
1394 	 *
1395 	 * NB. These rules can be relaxed when checking the group count
1396 	 * while freeing data, as we can only allocate from a block
1397 	 * group after serialising against the group count, and we can
1398 	 * only then free after serialising in turn against that
1399 	 * allocation.
1400 	 */
1401 	smp_wmb();
1402 
1403 	/* Update the global fs size fields */
1404 	sbi->s_groups_count += flex_gd->count;
1405 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1406 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1407 
1408 	/* Update the reserved block counts only once the new group is
1409 	 * active. */
1410 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1411 				reserved_blocks);
1412 
1413 	/* Update the free space counts */
1414 	percpu_counter_add(&sbi->s_freeclusters_counter,
1415 			   EXT4_NUM_B2C(sbi, free_blocks));
1416 	percpu_counter_add(&sbi->s_freeinodes_counter,
1417 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1418 
1419 	ext4_debug("free blocks count %llu",
1420 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1421 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1422 		ext4_group_t flex_group;
1423 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1424 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1425 			     &sbi->s_flex_groups[flex_group].free_clusters);
1426 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1427 			   &sbi->s_flex_groups[flex_group].free_inodes);
1428 	}
1429 
1430 	/*
1431 	 * Update the fs overhead information
1432 	 */
1433 	ext4_calculate_overhead(sb);
1434 
1435 	if (test_opt(sb, DEBUG))
1436 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1437 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1438 		       blocks_count, free_blocks, reserved_blocks);
1439 }
1440 
1441 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1442  * _before_ we start modifying the filesystem, because we cannot abort the
1443  * transaction and not have it write the data to disk.
1444  */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1445 static int ext4_flex_group_add(struct super_block *sb,
1446 			       struct inode *resize_inode,
1447 			       struct ext4_new_flex_group_data *flex_gd)
1448 {
1449 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1450 	struct ext4_super_block *es = sbi->s_es;
1451 	ext4_fsblk_t o_blocks_count;
1452 	ext4_grpblk_t last;
1453 	ext4_group_t group;
1454 	handle_t *handle;
1455 	unsigned reserved_gdb;
1456 	int err = 0, err2 = 0, credit;
1457 
1458 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1459 
1460 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1461 	o_blocks_count = ext4_blocks_count(es);
1462 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1463 	BUG_ON(last);
1464 
1465 	err = setup_new_flex_group_blocks(sb, flex_gd);
1466 	if (err)
1467 		goto exit;
1468 	/*
1469 	 * We will always be modifying at least the superblock and  GDT
1470 	 * blocks.  If we are adding a group past the last current GDT block,
1471 	 * we will also modify the inode and the dindirect block.  If we
1472 	 * are adding a group with superblock/GDT backups  we will also
1473 	 * modify each of the reserved GDT dindirect blocks.
1474 	 */
1475 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1476 	/* GDT blocks */
1477 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1478 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1479 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1480 	if (IS_ERR(handle)) {
1481 		err = PTR_ERR(handle);
1482 		goto exit;
1483 	}
1484 
1485 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1486 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1487 	if (err)
1488 		goto exit_journal;
1489 
1490 	group = flex_gd->groups[0].group;
1491 	BUG_ON(group != sbi->s_groups_count);
1492 	err = ext4_add_new_descs(handle, sb, group,
1493 				resize_inode, flex_gd->count);
1494 	if (err)
1495 		goto exit_journal;
1496 
1497 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1498 	if (err)
1499 		goto exit_journal;
1500 
1501 	ext4_update_super(sb, flex_gd);
1502 
1503 	err = ext4_handle_dirty_super(handle, sb);
1504 
1505 exit_journal:
1506 	err2 = ext4_journal_stop(handle);
1507 	if (!err)
1508 		err = err2;
1509 
1510 	if (!err) {
1511 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1512 		int gdb_num_end = ((group + flex_gd->count - 1) /
1513 				   EXT4_DESC_PER_BLOCK(sb));
1514 		int meta_bg = ext4_has_feature_meta_bg(sb);
1515 		sector_t old_gdb = 0;
1516 
1517 		update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1518 			       sizeof(struct ext4_super_block), 0);
1519 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1520 			struct buffer_head *gdb_bh;
1521 
1522 			gdb_bh = sbi->s_group_desc[gdb_num];
1523 			if (old_gdb == gdb_bh->b_blocknr)
1524 				continue;
1525 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1526 				       gdb_bh->b_size, meta_bg);
1527 			old_gdb = gdb_bh->b_blocknr;
1528 		}
1529 	}
1530 exit:
1531 	return err;
1532 }
1533 
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)1534 static int ext4_setup_next_flex_gd(struct super_block *sb,
1535 				    struct ext4_new_flex_group_data *flex_gd,
1536 				    ext4_fsblk_t n_blocks_count,
1537 				    unsigned long flexbg_size)
1538 {
1539 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1540 	struct ext4_super_block *es = sbi->s_es;
1541 	struct ext4_new_group_data *group_data = flex_gd->groups;
1542 	ext4_fsblk_t o_blocks_count;
1543 	ext4_group_t n_group;
1544 	ext4_group_t group;
1545 	ext4_group_t last_group;
1546 	ext4_grpblk_t last;
1547 	ext4_grpblk_t clusters_per_group;
1548 	unsigned long i;
1549 
1550 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1551 
1552 	o_blocks_count = ext4_blocks_count(es);
1553 
1554 	if (o_blocks_count == n_blocks_count)
1555 		return 0;
1556 
1557 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1558 	BUG_ON(last);
1559 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1560 
1561 	last_group = group | (flexbg_size - 1);
1562 	if (last_group > n_group)
1563 		last_group = n_group;
1564 
1565 	flex_gd->count = last_group - group + 1;
1566 
1567 	for (i = 0; i < flex_gd->count; i++) {
1568 		int overhead;
1569 
1570 		group_data[i].group = group + i;
1571 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1572 		overhead = ext4_group_overhead_blocks(sb, group + i);
1573 		group_data[i].mdata_blocks = overhead;
1574 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1575 		if (ext4_has_group_desc_csum(sb)) {
1576 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1577 					       EXT4_BG_INODE_UNINIT;
1578 			if (!test_opt(sb, INIT_INODE_TABLE))
1579 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1580 		} else
1581 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1582 	}
1583 
1584 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1585 		/* We need to initialize block bitmap of last group. */
1586 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1587 
1588 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1589 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1590 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1591 						       last - 1;
1592 	}
1593 
1594 	return 1;
1595 }
1596 
1597 /* Add group descriptor data to an existing or new group descriptor block.
1598  * Ensure we handle all possible error conditions _before_ we start modifying
1599  * the filesystem, because we cannot abort the transaction and not have it
1600  * write the data to disk.
1601  *
1602  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1603  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1604  *
1605  * We only need to hold the superblock lock while we are actually adding
1606  * in the new group's counts to the superblock.  Prior to that we have
1607  * not really "added" the group at all.  We re-check that we are still
1608  * adding in the last group in case things have changed since verifying.
1609  */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1610 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1611 {
1612 	struct ext4_new_flex_group_data flex_gd;
1613 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1614 	struct ext4_super_block *es = sbi->s_es;
1615 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1616 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1617 	struct inode *inode = NULL;
1618 	int gdb_off;
1619 	int err;
1620 	__u16 bg_flags = 0;
1621 
1622 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1623 
1624 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1625 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1626 		return -EPERM;
1627 	}
1628 
1629 	if (ext4_blocks_count(es) + input->blocks_count <
1630 	    ext4_blocks_count(es)) {
1631 		ext4_warning(sb, "blocks_count overflow");
1632 		return -EINVAL;
1633 	}
1634 
1635 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1636 	    le32_to_cpu(es->s_inodes_count)) {
1637 		ext4_warning(sb, "inodes_count overflow");
1638 		return -EINVAL;
1639 	}
1640 
1641 	if (reserved_gdb || gdb_off == 0) {
1642 		if (!ext4_has_feature_resize_inode(sb) ||
1643 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1644 			ext4_warning(sb,
1645 				     "No reserved GDT blocks, can't resize");
1646 			return -EPERM;
1647 		}
1648 		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1649 		if (IS_ERR(inode)) {
1650 			ext4_warning(sb, "Error opening resize inode");
1651 			return PTR_ERR(inode);
1652 		}
1653 	}
1654 
1655 
1656 	err = verify_group_input(sb, input);
1657 	if (err)
1658 		goto out;
1659 
1660 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1661 	if (err)
1662 		goto out;
1663 
1664 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1665 	if (err)
1666 		goto out;
1667 
1668 	flex_gd.count = 1;
1669 	flex_gd.groups = input;
1670 	flex_gd.bg_flags = &bg_flags;
1671 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1672 out:
1673 	iput(inode);
1674 	return err;
1675 } /* ext4_group_add */
1676 
1677 /*
1678  * extend a group without checking assuming that checking has been done.
1679  */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1680 static int ext4_group_extend_no_check(struct super_block *sb,
1681 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1682 {
1683 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1684 	handle_t *handle;
1685 	int err = 0, err2;
1686 
1687 	/* We will update the superblock, one block bitmap, and
1688 	 * one group descriptor via ext4_group_add_blocks().
1689 	 */
1690 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1691 	if (IS_ERR(handle)) {
1692 		err = PTR_ERR(handle);
1693 		ext4_warning(sb, "error %d on journal start", err);
1694 		return err;
1695 	}
1696 
1697 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1698 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1699 	if (err) {
1700 		ext4_warning(sb, "error %d on journal write access", err);
1701 		goto errout;
1702 	}
1703 
1704 	ext4_blocks_count_set(es, o_blocks_count + add);
1705 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1706 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1707 		   o_blocks_count + add);
1708 	/* We add the blocks to the bitmap and set the group need init bit */
1709 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1710 	if (err)
1711 		goto errout;
1712 	ext4_handle_dirty_super(handle, sb);
1713 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1714 		   o_blocks_count + add);
1715 errout:
1716 	err2 = ext4_journal_stop(handle);
1717 	if (err2 && !err)
1718 		err = err2;
1719 
1720 	if (!err) {
1721 		if (test_opt(sb, DEBUG))
1722 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1723 			       "blocks\n", ext4_blocks_count(es));
1724 		update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1725 			       (char *)es, sizeof(struct ext4_super_block), 0);
1726 	}
1727 	return err;
1728 }
1729 
1730 /*
1731  * Extend the filesystem to the new number of blocks specified.  This entry
1732  * point is only used to extend the current filesystem to the end of the last
1733  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1734  * for emergencies (because it has no dependencies on reserved blocks).
1735  *
1736  * If we _really_ wanted, we could use default values to call ext4_group_add()
1737  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1738  * GDT blocks are reserved to grow to the desired size.
1739  */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1740 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1741 		      ext4_fsblk_t n_blocks_count)
1742 {
1743 	ext4_fsblk_t o_blocks_count;
1744 	ext4_grpblk_t last;
1745 	ext4_grpblk_t add;
1746 	struct buffer_head *bh;
1747 	int err;
1748 	ext4_group_t group;
1749 
1750 	o_blocks_count = ext4_blocks_count(es);
1751 
1752 	if (test_opt(sb, DEBUG))
1753 		ext4_msg(sb, KERN_DEBUG,
1754 			 "extending last group from %llu to %llu blocks",
1755 			 o_blocks_count, n_blocks_count);
1756 
1757 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1758 		return 0;
1759 
1760 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1761 		ext4_msg(sb, KERN_ERR,
1762 			 "filesystem too large to resize to %llu blocks safely",
1763 			 n_blocks_count);
1764 		return -EINVAL;
1765 	}
1766 
1767 	if (n_blocks_count < o_blocks_count) {
1768 		ext4_warning(sb, "can't shrink FS - resize aborted");
1769 		return -EINVAL;
1770 	}
1771 
1772 	/* Handle the remaining blocks in the last group only. */
1773 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1774 
1775 	if (last == 0) {
1776 		ext4_warning(sb, "need to use ext2online to resize further");
1777 		return -EPERM;
1778 	}
1779 
1780 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1781 
1782 	if (o_blocks_count + add < o_blocks_count) {
1783 		ext4_warning(sb, "blocks_count overflow");
1784 		return -EINVAL;
1785 	}
1786 
1787 	if (o_blocks_count + add > n_blocks_count)
1788 		add = n_blocks_count - o_blocks_count;
1789 
1790 	if (o_blocks_count + add < n_blocks_count)
1791 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1792 			     o_blocks_count + add, add);
1793 
1794 	/* See if the device is actually as big as what was requested */
1795 	bh = sb_bread(sb, o_blocks_count + add - 1);
1796 	if (!bh) {
1797 		ext4_warning(sb, "can't read last block, resize aborted");
1798 		return -ENOSPC;
1799 	}
1800 	brelse(bh);
1801 
1802 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1803 	return err;
1804 } /* ext4_group_extend */
1805 
1806 
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1807 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1808 {
1809 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1810 }
1811 
1812 /*
1813  * Release the resize inode and drop the resize_inode feature if there
1814  * are no more reserved gdt blocks, and then convert the file system
1815  * to enable meta_bg
1816  */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1817 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1818 {
1819 	handle_t *handle;
1820 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1821 	struct ext4_super_block *es = sbi->s_es;
1822 	struct ext4_inode_info *ei = EXT4_I(inode);
1823 	ext4_fsblk_t nr;
1824 	int i, ret, err = 0;
1825 	int credits = 1;
1826 
1827 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1828 	if (inode) {
1829 		if (es->s_reserved_gdt_blocks) {
1830 			ext4_error(sb, "Unexpected non-zero "
1831 				   "s_reserved_gdt_blocks");
1832 			return -EPERM;
1833 		}
1834 
1835 		/* Do a quick sanity check of the resize inode */
1836 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1837 					     (9 - sbi->s_cluster_bits)))
1838 			goto invalid_resize_inode;
1839 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1840 			if (i == EXT4_DIND_BLOCK) {
1841 				if (ei->i_data[i])
1842 					continue;
1843 				else
1844 					goto invalid_resize_inode;
1845 			}
1846 			if (ei->i_data[i])
1847 				goto invalid_resize_inode;
1848 		}
1849 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1850 	}
1851 
1852 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1853 	if (IS_ERR(handle))
1854 		return PTR_ERR(handle);
1855 
1856 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1857 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1858 	if (err)
1859 		goto errout;
1860 
1861 	ext4_clear_feature_resize_inode(sb);
1862 	ext4_set_feature_meta_bg(sb);
1863 	sbi->s_es->s_first_meta_bg =
1864 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1865 
1866 	err = ext4_handle_dirty_super(handle, sb);
1867 	if (err) {
1868 		ext4_std_error(sb, err);
1869 		goto errout;
1870 	}
1871 
1872 	if (inode) {
1873 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1874 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1875 				 EXT4_FREE_BLOCKS_METADATA |
1876 				 EXT4_FREE_BLOCKS_FORGET);
1877 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1878 		inode->i_blocks = 0;
1879 
1880 		err = ext4_mark_inode_dirty(handle, inode);
1881 		if (err)
1882 			ext4_std_error(sb, err);
1883 	}
1884 
1885 errout:
1886 	ret = ext4_journal_stop(handle);
1887 	if (!err)
1888 		err = ret;
1889 	return ret;
1890 
1891 invalid_resize_inode:
1892 	ext4_error(sb, "corrupted/inconsistent resize inode");
1893 	return -EINVAL;
1894 }
1895 
1896 /*
1897  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1898  *
1899  * @sb: super block of the fs to be resized
1900  * @n_blocks_count: the number of blocks resides in the resized fs
1901  */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)1902 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1903 {
1904 	struct ext4_new_flex_group_data *flex_gd = NULL;
1905 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1906 	struct ext4_super_block *es = sbi->s_es;
1907 	struct buffer_head *bh;
1908 	struct inode *resize_inode = NULL;
1909 	ext4_grpblk_t add, offset;
1910 	unsigned long n_desc_blocks;
1911 	unsigned long o_desc_blocks;
1912 	ext4_group_t o_group;
1913 	ext4_group_t n_group;
1914 	ext4_fsblk_t o_blocks_count;
1915 	ext4_fsblk_t n_blocks_count_retry = 0;
1916 	unsigned long last_update_time = 0;
1917 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1918 	int meta_bg;
1919 
1920 	/* See if the device is actually as big as what was requested */
1921 	bh = sb_bread(sb, n_blocks_count - 1);
1922 	if (!bh) {
1923 		ext4_warning(sb, "can't read last block, resize aborted");
1924 		return -ENOSPC;
1925 	}
1926 	brelse(bh);
1927 
1928 retry:
1929 	o_blocks_count = ext4_blocks_count(es);
1930 
1931 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1932 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1933 
1934 	if (n_blocks_count < o_blocks_count) {
1935 		/* On-line shrinking not supported */
1936 		ext4_warning(sb, "can't shrink FS - resize aborted");
1937 		return -EINVAL;
1938 	}
1939 
1940 	if (n_blocks_count == o_blocks_count)
1941 		/* Nothing need to do */
1942 		return 0;
1943 
1944 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1945 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1946 		ext4_warning(sb, "resize would cause inodes_count overflow");
1947 		return -EINVAL;
1948 	}
1949 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1950 
1951 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1952 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1953 
1954 	meta_bg = ext4_has_feature_meta_bg(sb);
1955 
1956 	if (ext4_has_feature_resize_inode(sb)) {
1957 		if (meta_bg) {
1958 			ext4_error(sb, "resize_inode and meta_bg enabled "
1959 				   "simultaneously");
1960 			return -EINVAL;
1961 		}
1962 		if (n_desc_blocks > o_desc_blocks +
1963 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1964 			n_blocks_count_retry = n_blocks_count;
1965 			n_desc_blocks = o_desc_blocks +
1966 				le16_to_cpu(es->s_reserved_gdt_blocks);
1967 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1968 			n_blocks_count = (ext4_fsblk_t)n_group *
1969 				EXT4_BLOCKS_PER_GROUP(sb) +
1970 				le32_to_cpu(es->s_first_data_block);
1971 			n_group--; /* set to last group number */
1972 		}
1973 
1974 		if (!resize_inode)
1975 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1976 						 EXT4_IGET_SPECIAL);
1977 		if (IS_ERR(resize_inode)) {
1978 			ext4_warning(sb, "Error opening resize inode");
1979 			return PTR_ERR(resize_inode);
1980 		}
1981 	}
1982 
1983 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1984 		err = ext4_convert_meta_bg(sb, resize_inode);
1985 		if (err)
1986 			goto out;
1987 		if (resize_inode) {
1988 			iput(resize_inode);
1989 			resize_inode = NULL;
1990 		}
1991 		if (n_blocks_count_retry) {
1992 			n_blocks_count = n_blocks_count_retry;
1993 			n_blocks_count_retry = 0;
1994 			goto retry;
1995 		}
1996 	}
1997 
1998 	/*
1999 	 * Make sure the last group has enough space so that it's
2000 	 * guaranteed to have enough space for all metadata blocks
2001 	 * that it might need to hold.  (We might not need to store
2002 	 * the inode table blocks in the last block group, but there
2003 	 * will be cases where this might be needed.)
2004 	 */
2005 	if ((ext4_group_first_block_no(sb, n_group) +
2006 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
2007 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2008 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2009 		n_group--;
2010 		n_blocks_count_retry = 0;
2011 		if (resize_inode) {
2012 			iput(resize_inode);
2013 			resize_inode = NULL;
2014 		}
2015 		goto retry;
2016 	}
2017 
2018 	/* extend the last group */
2019 	if (n_group == o_group)
2020 		add = n_blocks_count - o_blocks_count;
2021 	else
2022 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2023 	if (add > 0) {
2024 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2025 		if (err)
2026 			goto out;
2027 	}
2028 
2029 	if (ext4_blocks_count(es) == n_blocks_count)
2030 		goto out;
2031 
2032 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2033 	if (err)
2034 		goto out;
2035 
2036 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2037 	if (err)
2038 		goto out;
2039 
2040 	flex_gd = alloc_flex_gd(flexbg_size);
2041 	if (flex_gd == NULL) {
2042 		err = -ENOMEM;
2043 		goto out;
2044 	}
2045 
2046 	/* Add flex groups. Note that a regular group is a
2047 	 * flex group with 1 group.
2048 	 */
2049 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2050 					      flexbg_size)) {
2051 		if (jiffies - last_update_time > HZ * 10) {
2052 			if (last_update_time)
2053 				ext4_msg(sb, KERN_INFO,
2054 					 "resized to %llu blocks",
2055 					 ext4_blocks_count(es));
2056 			last_update_time = jiffies;
2057 		}
2058 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2059 			break;
2060 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2061 		if (unlikely(err))
2062 			break;
2063 	}
2064 
2065 	if (!err && n_blocks_count_retry) {
2066 		n_blocks_count = n_blocks_count_retry;
2067 		n_blocks_count_retry = 0;
2068 		free_flex_gd(flex_gd);
2069 		flex_gd = NULL;
2070 		if (resize_inode) {
2071 			iput(resize_inode);
2072 			resize_inode = NULL;
2073 		}
2074 		goto retry;
2075 	}
2076 
2077 out:
2078 	if (flex_gd)
2079 		free_flex_gd(flex_gd);
2080 	if (resize_inode != NULL)
2081 		iput(resize_inode);
2082 	if (err)
2083 		ext4_warning(sb, "error (%d) occurred during "
2084 			     "file system resize", err);
2085 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2086 		 ext4_blocks_count(es));
2087 	return err;
2088 }
2089