• 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 
20 struct ext4_rcu_ptr {
21 	struct rcu_head rcu;
22 	void *ptr;
23 };
24 
ext4_rcu_ptr_callback(struct rcu_head * head)25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
26 {
27 	struct ext4_rcu_ptr *ptr;
28 
29 	ptr = container_of(head, struct ext4_rcu_ptr, rcu);
30 	kvfree(ptr->ptr);
31 	kfree(ptr);
32 }
33 
ext4_kvfree_array_rcu(void * to_free)34 void ext4_kvfree_array_rcu(void *to_free)
35 {
36 	struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
37 
38 	if (ptr) {
39 		ptr->ptr = to_free;
40 		call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
41 		return;
42 	}
43 	synchronize_rcu();
44 	kvfree(to_free);
45 }
46 
ext4_resize_begin(struct super_block * sb)47 int ext4_resize_begin(struct super_block *sb)
48 {
49 	struct ext4_sb_info *sbi = EXT4_SB(sb);
50 	int ret = 0;
51 
52 	if (!capable(CAP_SYS_RESOURCE))
53 		return -EPERM;
54 
55 	/*
56 	 * If the reserved GDT blocks is non-zero, the resize_inode feature
57 	 * should always be set.
58 	 */
59 	if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
60 	    !ext4_has_feature_resize_inode(sb)) {
61 		ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
62 		return -EFSCORRUPTED;
63 	}
64 
65 	/*
66 	 * If we are not using the primary superblock/GDT copy don't resize,
67          * because the user tools have no way of handling this.  Probably a
68          * bad time to do it anyways.
69          */
70 	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
71 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
72 		ext4_warning(sb, "won't resize using backup superblock at %llu",
73 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
74 		return -EPERM;
75 	}
76 
77 	/*
78 	 * We are not allowed to do online-resizing on a filesystem mounted
79 	 * with error, because it can destroy the filesystem easily.
80 	 */
81 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
82 		ext4_warning(sb, "There are errors in the filesystem, "
83 			     "so online resizing is not allowed");
84 		return -EPERM;
85 	}
86 
87 	if (ext4_has_feature_sparse_super2(sb)) {
88 		ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
89 		return -EOPNOTSUPP;
90 	}
91 
92 	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
93 				  &EXT4_SB(sb)->s_ext4_flags))
94 		ret = -EBUSY;
95 
96 	return ret;
97 }
98 
ext4_resize_end(struct super_block * sb)99 void ext4_resize_end(struct super_block *sb)
100 {
101 	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
102 	smp_mb__after_atomic();
103 }
104 
ext4_meta_bg_first_group(struct super_block * sb,ext4_group_t group)105 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
106 					     ext4_group_t group) {
107 	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
108 	       EXT4_DESC_PER_BLOCK_BITS(sb);
109 }
110 
ext4_meta_bg_first_block_no(struct super_block * sb,ext4_group_t group)111 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
112 					     ext4_group_t group) {
113 	group = ext4_meta_bg_first_group(sb, group);
114 	return ext4_group_first_block_no(sb, group);
115 }
116 
ext4_group_overhead_blocks(struct super_block * sb,ext4_group_t group)117 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
118 						ext4_group_t group) {
119 	ext4_grpblk_t overhead;
120 	overhead = ext4_bg_num_gdb(sb, group);
121 	if (ext4_bg_has_super(sb, group))
122 		overhead += 1 +
123 			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
124 	return overhead;
125 }
126 
127 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
128 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
129 
verify_group_input(struct super_block * sb,struct ext4_new_group_data * input)130 static int verify_group_input(struct super_block *sb,
131 			      struct ext4_new_group_data *input)
132 {
133 	struct ext4_sb_info *sbi = EXT4_SB(sb);
134 	struct ext4_super_block *es = sbi->s_es;
135 	ext4_fsblk_t start = ext4_blocks_count(es);
136 	ext4_fsblk_t end = start + input->blocks_count;
137 	ext4_group_t group = input->group;
138 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
139 	unsigned overhead;
140 	ext4_fsblk_t metaend;
141 	struct buffer_head *bh = NULL;
142 	ext4_grpblk_t free_blocks_count, offset;
143 	int err = -EINVAL;
144 
145 	if (group != sbi->s_groups_count) {
146 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
147 			     input->group, sbi->s_groups_count);
148 		return -EINVAL;
149 	}
150 
151 	overhead = ext4_group_overhead_blocks(sb, group);
152 	metaend = start + overhead;
153 	input->free_clusters_count = free_blocks_count =
154 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
155 
156 	if (test_opt(sb, DEBUG))
157 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
158 		       "(%d free, %u reserved)\n",
159 		       ext4_bg_has_super(sb, input->group) ? "normal" :
160 		       "no-super", input->group, input->blocks_count,
161 		       free_blocks_count, input->reserved_blocks);
162 
163 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
164 	if (offset != 0)
165 			ext4_warning(sb, "Last group not full");
166 	else if (input->reserved_blocks > input->blocks_count / 5)
167 		ext4_warning(sb, "Reserved blocks too high (%u)",
168 			     input->reserved_blocks);
169 	else if (free_blocks_count < 0)
170 		ext4_warning(sb, "Bad blocks count %u",
171 			     input->blocks_count);
172 	else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
173 		err = PTR_ERR(bh);
174 		bh = NULL;
175 		ext4_warning(sb, "Cannot read last block (%llu)",
176 			     end - 1);
177 	} else if (outside(input->block_bitmap, start, end))
178 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
179 			     (unsigned long long)input->block_bitmap);
180 	else if (outside(input->inode_bitmap, start, end))
181 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
182 			     (unsigned long long)input->inode_bitmap);
183 	else if (outside(input->inode_table, start, end) ||
184 		 outside(itend - 1, start, end))
185 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
186 			     (unsigned long long)input->inode_table, itend - 1);
187 	else if (input->inode_bitmap == input->block_bitmap)
188 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
189 			     (unsigned long long)input->block_bitmap);
190 	else if (inside(input->block_bitmap, input->inode_table, itend))
191 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
192 			     "(%llu-%llu)",
193 			     (unsigned long long)input->block_bitmap,
194 			     (unsigned long long)input->inode_table, itend - 1);
195 	else if (inside(input->inode_bitmap, input->inode_table, itend))
196 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
197 			     "(%llu-%llu)",
198 			     (unsigned long long)input->inode_bitmap,
199 			     (unsigned long long)input->inode_table, itend - 1);
200 	else if (inside(input->block_bitmap, start, metaend))
201 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
202 			     (unsigned long long)input->block_bitmap,
203 			     start, metaend - 1);
204 	else if (inside(input->inode_bitmap, start, metaend))
205 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
206 			     (unsigned long long)input->inode_bitmap,
207 			     start, metaend - 1);
208 	else if (inside(input->inode_table, start, metaend) ||
209 		 inside(itend - 1, start, metaend))
210 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
211 			     "(%llu-%llu)",
212 			     (unsigned long long)input->inode_table,
213 			     itend - 1, start, metaend - 1);
214 	else
215 		err = 0;
216 	brelse(bh);
217 
218 	return err;
219 }
220 
221 /*
222  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
223  * group each time.
224  */
225 struct ext4_new_flex_group_data {
226 	struct ext4_new_group_data *groups;	/* new_group_data for groups
227 						   in the flex group */
228 	__u16 *bg_flags;			/* block group flags of groups
229 						   in @groups */
230 	ext4_group_t count;			/* number of groups in @groups
231 						 */
232 };
233 
234 /*
235  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
236  * @flexbg_size.
237  *
238  * Returns NULL on failure otherwise address of the allocated structure.
239  */
alloc_flex_gd(unsigned long flexbg_size)240 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
241 {
242 	struct ext4_new_flex_group_data *flex_gd;
243 
244 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
245 	if (flex_gd == NULL)
246 		goto out3;
247 
248 	if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
249 		goto out2;
250 	flex_gd->count = flexbg_size;
251 
252 	flex_gd->groups = kmalloc_array(flexbg_size,
253 					sizeof(struct ext4_new_group_data),
254 					GFP_NOFS);
255 	if (flex_gd->groups == NULL)
256 		goto out2;
257 
258 	flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
259 					  GFP_NOFS);
260 	if (flex_gd->bg_flags == NULL)
261 		goto out1;
262 
263 	return flex_gd;
264 
265 out1:
266 	kfree(flex_gd->groups);
267 out2:
268 	kfree(flex_gd);
269 out3:
270 	return NULL;
271 }
272 
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)273 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
274 {
275 	kfree(flex_gd->bg_flags);
276 	kfree(flex_gd->groups);
277 	kfree(flex_gd);
278 }
279 
280 /*
281  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
282  * and inode tables for a flex group.
283  *
284  * This function is used by 64bit-resize.  Note that this function allocates
285  * group tables from the 1st group of groups contained by @flexgd, which may
286  * be a partial of a flex group.
287  *
288  * @sb: super block of fs to which the groups belongs
289  *
290  * Returns 0 on a successful allocation of the metadata blocks in the
291  * block group.
292  */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,int flexbg_size)293 static int ext4_alloc_group_tables(struct super_block *sb,
294 				struct ext4_new_flex_group_data *flex_gd,
295 				int flexbg_size)
296 {
297 	struct ext4_new_group_data *group_data = flex_gd->groups;
298 	ext4_fsblk_t start_blk;
299 	ext4_fsblk_t last_blk;
300 	ext4_group_t src_group;
301 	ext4_group_t bb_index = 0;
302 	ext4_group_t ib_index = 0;
303 	ext4_group_t it_index = 0;
304 	ext4_group_t group;
305 	ext4_group_t last_group;
306 	unsigned overhead;
307 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
308 	int i;
309 
310 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
311 
312 	src_group = group_data[0].group;
313 	last_group  = src_group + flex_gd->count - 1;
314 
315 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
316 	       (last_group & ~(flexbg_size - 1))));
317 next_group:
318 	group = group_data[0].group;
319 	if (src_group >= group_data[0].group + flex_gd->count)
320 		return -ENOSPC;
321 	start_blk = ext4_group_first_block_no(sb, src_group);
322 	last_blk = start_blk + group_data[src_group - group].blocks_count;
323 
324 	overhead = ext4_group_overhead_blocks(sb, src_group);
325 
326 	start_blk += overhead;
327 
328 	/* We collect contiguous blocks as much as possible. */
329 	src_group++;
330 	for (; src_group <= last_group; src_group++) {
331 		overhead = ext4_group_overhead_blocks(sb, src_group);
332 		if (overhead == 0)
333 			last_blk += group_data[src_group - group].blocks_count;
334 		else
335 			break;
336 	}
337 
338 	/* Allocate block bitmaps */
339 	for (; bb_index < flex_gd->count; bb_index++) {
340 		if (start_blk >= last_blk)
341 			goto next_group;
342 		group_data[bb_index].block_bitmap = start_blk++;
343 		group = ext4_get_group_number(sb, start_blk - 1);
344 		group -= group_data[0].group;
345 		group_data[group].mdata_blocks++;
346 		flex_gd->bg_flags[group] &= uninit_mask;
347 	}
348 
349 	/* Allocate inode bitmaps */
350 	for (; ib_index < flex_gd->count; ib_index++) {
351 		if (start_blk >= last_blk)
352 			goto next_group;
353 		group_data[ib_index].inode_bitmap = start_blk++;
354 		group = ext4_get_group_number(sb, start_blk - 1);
355 		group -= group_data[0].group;
356 		group_data[group].mdata_blocks++;
357 		flex_gd->bg_flags[group] &= uninit_mask;
358 	}
359 
360 	/* Allocate inode tables */
361 	for (; it_index < flex_gd->count; it_index++) {
362 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
363 		ext4_fsblk_t next_group_start;
364 
365 		if (start_blk + itb > last_blk)
366 			goto next_group;
367 		group_data[it_index].inode_table = start_blk;
368 		group = ext4_get_group_number(sb, start_blk);
369 		next_group_start = ext4_group_first_block_no(sb, group + 1);
370 		group -= group_data[0].group;
371 
372 		if (start_blk + itb > next_group_start) {
373 			flex_gd->bg_flags[group + 1] &= uninit_mask;
374 			overhead = start_blk + itb - next_group_start;
375 			group_data[group + 1].mdata_blocks += overhead;
376 			itb -= overhead;
377 		}
378 
379 		group_data[group].mdata_blocks += itb;
380 		flex_gd->bg_flags[group] &= uninit_mask;
381 		start_blk += EXT4_SB(sb)->s_itb_per_group;
382 	}
383 
384 	/* Update free clusters count to exclude metadata blocks */
385 	for (i = 0; i < flex_gd->count; i++) {
386 		group_data[i].free_clusters_count -=
387 				EXT4_NUM_B2C(EXT4_SB(sb),
388 					     group_data[i].mdata_blocks);
389 	}
390 
391 	if (test_opt(sb, DEBUG)) {
392 		int i;
393 		group = group_data[0].group;
394 
395 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
396 		       "%d groups, flexbg size is %d:\n", flex_gd->count,
397 		       flexbg_size);
398 
399 		for (i = 0; i < flex_gd->count; i++) {
400 			ext4_debug(
401 			       "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
402 			       ext4_bg_has_super(sb, group + i) ? "normal" :
403 			       "no-super", group + i,
404 			       group_data[i].blocks_count,
405 			       group_data[i].free_clusters_count,
406 			       group_data[i].mdata_blocks);
407 		}
408 	}
409 	return 0;
410 }
411 
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)412 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
413 				  ext4_fsblk_t blk)
414 {
415 	struct buffer_head *bh;
416 	int err;
417 
418 	bh = sb_getblk(sb, blk);
419 	if (unlikely(!bh))
420 		return ERR_PTR(-ENOMEM);
421 	BUFFER_TRACE(bh, "get_write_access");
422 	if ((err = ext4_journal_get_write_access(handle, bh))) {
423 		brelse(bh);
424 		bh = ERR_PTR(err);
425 	} else {
426 		memset(bh->b_data, 0, sb->s_blocksize);
427 		set_buffer_uptodate(bh);
428 	}
429 
430 	return bh;
431 }
432 
ext4_resize_ensure_credits_batch(handle_t * handle,int credits)433 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
434 {
435 	return ext4_journal_ensure_credits_fn(handle, credits,
436 		EXT4_MAX_TRANS_DATA, 0, 0);
437 }
438 
439 /*
440  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
441  *
442  * Helper function for ext4_setup_new_group_blocks() which set .
443  *
444  * @sb: super block
445  * @handle: journal handle
446  * @flex_gd: flex group data
447  */
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)448 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
449 			struct ext4_new_flex_group_data *flex_gd,
450 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
451 {
452 	struct ext4_sb_info *sbi = EXT4_SB(sb);
453 	ext4_group_t count = last_cluster - first_cluster + 1;
454 	ext4_group_t count2;
455 
456 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
457 		   last_cluster);
458 	for (count2 = count; count > 0;
459 	     count -= count2, first_cluster += count2) {
460 		ext4_fsblk_t start;
461 		struct buffer_head *bh;
462 		ext4_group_t group;
463 		int err;
464 
465 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
466 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
467 		group -= flex_gd->groups[0].group;
468 
469 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
470 		if (count2 > count)
471 			count2 = count;
472 
473 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
474 			BUG_ON(flex_gd->count > 1);
475 			continue;
476 		}
477 
478 		err = ext4_resize_ensure_credits_batch(handle, 1);
479 		if (err < 0)
480 			return err;
481 
482 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
483 		if (unlikely(!bh))
484 			return -ENOMEM;
485 
486 		BUFFER_TRACE(bh, "get_write_access");
487 		err = ext4_journal_get_write_access(handle, bh);
488 		if (err) {
489 			brelse(bh);
490 			return err;
491 		}
492 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
493 			   first_cluster, first_cluster - start, count2);
494 		ext4_set_bits(bh->b_data, first_cluster - start, count2);
495 
496 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
497 		brelse(bh);
498 		if (unlikely(err))
499 			return err;
500 	}
501 
502 	return 0;
503 }
504 
505 /*
506  * Set up the block and inode bitmaps, and the inode table for the new groups.
507  * This doesn't need to be part of the main transaction, since we are only
508  * changing blocks outside the actual filesystem.  We still do journaling to
509  * ensure the recovery is correct in case of a failure just after resize.
510  * If any part of this fails, we simply abort the resize.
511  *
512  * setup_new_flex_group_blocks handles a flex group as follow:
513  *  1. copy super block and GDT, and initialize group tables if necessary.
514  *     In this step, we only set bits in blocks bitmaps for blocks taken by
515  *     super block and GDT.
516  *  2. allocate group tables in block bitmaps, that is, set bits in block
517  *     bitmap for blocks taken by group tables.
518  */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)519 static int setup_new_flex_group_blocks(struct super_block *sb,
520 				struct ext4_new_flex_group_data *flex_gd)
521 {
522 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
523 	ext4_fsblk_t start;
524 	ext4_fsblk_t block;
525 	struct ext4_sb_info *sbi = EXT4_SB(sb);
526 	struct ext4_super_block *es = sbi->s_es;
527 	struct ext4_new_group_data *group_data = flex_gd->groups;
528 	__u16 *bg_flags = flex_gd->bg_flags;
529 	handle_t *handle;
530 	ext4_group_t group, count;
531 	struct buffer_head *bh = NULL;
532 	int reserved_gdb, i, j, err = 0, err2;
533 	int meta_bg;
534 
535 	BUG_ON(!flex_gd->count || !group_data ||
536 	       group_data[0].group != sbi->s_groups_count);
537 
538 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
539 	meta_bg = ext4_has_feature_meta_bg(sb);
540 
541 	/* This transaction may be extended/restarted along the way */
542 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
543 	if (IS_ERR(handle))
544 		return PTR_ERR(handle);
545 
546 	group = group_data[0].group;
547 	for (i = 0; i < flex_gd->count; i++, group++) {
548 		unsigned long gdblocks;
549 		ext4_grpblk_t overhead;
550 
551 		gdblocks = ext4_bg_num_gdb(sb, group);
552 		start = ext4_group_first_block_no(sb, group);
553 
554 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
555 			goto handle_itb;
556 
557 		if (meta_bg == 1) {
558 			ext4_group_t first_group;
559 			first_group = ext4_meta_bg_first_group(sb, group);
560 			if (first_group != group + 1 &&
561 			    first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
562 				goto handle_itb;
563 		}
564 
565 		block = start + ext4_bg_has_super(sb, group);
566 		/* Copy all of the GDT blocks into the backup in this group */
567 		for (j = 0; j < gdblocks; j++, block++) {
568 			struct buffer_head *gdb;
569 
570 			ext4_debug("update backup group %#04llx\n", block);
571 			err = ext4_resize_ensure_credits_batch(handle, 1);
572 			if (err < 0)
573 				goto out;
574 
575 			gdb = sb_getblk(sb, block);
576 			if (unlikely(!gdb)) {
577 				err = -ENOMEM;
578 				goto out;
579 			}
580 
581 			BUFFER_TRACE(gdb, "get_write_access");
582 			err = ext4_journal_get_write_access(handle, gdb);
583 			if (err) {
584 				brelse(gdb);
585 				goto out;
586 			}
587 			memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
588 				s_group_desc, j)->b_data, gdb->b_size);
589 			set_buffer_uptodate(gdb);
590 
591 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
592 			if (unlikely(err)) {
593 				brelse(gdb);
594 				goto out;
595 			}
596 			brelse(gdb);
597 		}
598 
599 		/* Zero out all of the reserved backup group descriptor
600 		 * table blocks
601 		 */
602 		if (ext4_bg_has_super(sb, group)) {
603 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
604 					reserved_gdb, GFP_NOFS);
605 			if (err)
606 				goto out;
607 		}
608 
609 handle_itb:
610 		/* Initialize group tables of the grop @group */
611 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
612 			goto handle_bb;
613 
614 		/* Zero out all of the inode table blocks */
615 		block = group_data[i].inode_table;
616 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
617 			   block, sbi->s_itb_per_group);
618 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
619 				       GFP_NOFS);
620 		if (err)
621 			goto out;
622 
623 handle_bb:
624 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
625 			goto handle_ib;
626 
627 		/* Initialize block bitmap of the @group */
628 		block = group_data[i].block_bitmap;
629 		err = ext4_resize_ensure_credits_batch(handle, 1);
630 		if (err < 0)
631 			goto out;
632 
633 		bh = bclean(handle, sb, block);
634 		if (IS_ERR(bh)) {
635 			err = PTR_ERR(bh);
636 			goto out;
637 		}
638 		overhead = ext4_group_overhead_blocks(sb, group);
639 		if (overhead != 0) {
640 			ext4_debug("mark backup superblock %#04llx (+0)\n",
641 				   start);
642 			ext4_set_bits(bh->b_data, 0,
643 				      EXT4_NUM_B2C(sbi, overhead));
644 		}
645 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
646 				     sb->s_blocksize * 8, bh->b_data);
647 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
648 		brelse(bh);
649 		if (err)
650 			goto out;
651 
652 handle_ib:
653 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
654 			continue;
655 
656 		/* Initialize inode bitmap of the @group */
657 		block = group_data[i].inode_bitmap;
658 		err = ext4_resize_ensure_credits_batch(handle, 1);
659 		if (err < 0)
660 			goto out;
661 		/* Mark unused entries in inode bitmap used */
662 		bh = bclean(handle, sb, block);
663 		if (IS_ERR(bh)) {
664 			err = PTR_ERR(bh);
665 			goto out;
666 		}
667 
668 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
669 				     sb->s_blocksize * 8, bh->b_data);
670 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
671 		brelse(bh);
672 		if (err)
673 			goto out;
674 	}
675 
676 	/* Mark group tables in block bitmap */
677 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
678 		count = group_table_count[j];
679 		start = (&group_data[0].block_bitmap)[j];
680 		block = start;
681 		for (i = 1; i < flex_gd->count; i++) {
682 			block += group_table_count[j];
683 			if (block == (&group_data[i].block_bitmap)[j]) {
684 				count += group_table_count[j];
685 				continue;
686 			}
687 			err = set_flexbg_block_bitmap(sb, handle,
688 						      flex_gd,
689 						      EXT4_B2C(sbi, start),
690 						      EXT4_B2C(sbi,
691 							       start + count
692 							       - 1));
693 			if (err)
694 				goto out;
695 			count = group_table_count[j];
696 			start = (&group_data[i].block_bitmap)[j];
697 			block = start;
698 		}
699 
700 		if (count) {
701 			err = set_flexbg_block_bitmap(sb, handle,
702 						      flex_gd,
703 						      EXT4_B2C(sbi, start),
704 						      EXT4_B2C(sbi,
705 							       start + count
706 							       - 1));
707 			if (err)
708 				goto out;
709 		}
710 	}
711 
712 out:
713 	err2 = ext4_journal_stop(handle);
714 	if (err2 && !err)
715 		err = err2;
716 
717 	return err;
718 }
719 
720 /*
721  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
722  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
723  * calling this for the first time.  In a sparse filesystem it will be the
724  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
725  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
726  */
ext4_list_backups(struct super_block * sb,unsigned * three,unsigned * five,unsigned * seven)727 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
728 				  unsigned *five, unsigned *seven)
729 {
730 	unsigned *min = three;
731 	int mult = 3;
732 	unsigned ret;
733 
734 	if (!ext4_has_feature_sparse_super(sb)) {
735 		ret = *min;
736 		*min += 1;
737 		return ret;
738 	}
739 
740 	if (*five < *min) {
741 		min = five;
742 		mult = 5;
743 	}
744 	if (*seven < *min) {
745 		min = seven;
746 		mult = 7;
747 	}
748 
749 	ret = *min;
750 	*min *= mult;
751 
752 	return ret;
753 }
754 
755 /*
756  * Check that all of the backup GDT blocks are held in the primary GDT block.
757  * It is assumed that they are stored in group order.  Returns the number of
758  * groups in current filesystem that have BACKUPS, or -ve error code.
759  */
verify_reserved_gdb(struct super_block * sb,ext4_group_t end,struct buffer_head * primary)760 static int verify_reserved_gdb(struct super_block *sb,
761 			       ext4_group_t end,
762 			       struct buffer_head *primary)
763 {
764 	const ext4_fsblk_t blk = primary->b_blocknr;
765 	unsigned three = 1;
766 	unsigned five = 5;
767 	unsigned seven = 7;
768 	unsigned grp;
769 	__le32 *p = (__le32 *)primary->b_data;
770 	int gdbackups = 0;
771 
772 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
773 		if (le32_to_cpu(*p++) !=
774 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
775 			ext4_warning(sb, "reserved GDT %llu"
776 				     " missing grp %d (%llu)",
777 				     blk, grp,
778 				     grp *
779 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
780 				     blk);
781 			return -EINVAL;
782 		}
783 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
784 			return -EFBIG;
785 	}
786 
787 	return gdbackups;
788 }
789 
790 /*
791  * Called when we need to bring a reserved group descriptor table block into
792  * use from the resize inode.  The primary copy of the new GDT block currently
793  * is an indirect block (under the double indirect block in the resize inode).
794  * The new backup GDT blocks will be stored as leaf blocks in this indirect
795  * block, in group order.  Even though we know all the block numbers we need,
796  * we check to ensure that the resize inode has actually reserved these blocks.
797  *
798  * Don't need to update the block bitmaps because the blocks are still in use.
799  *
800  * We get all of the error cases out of the way, so that we are sure to not
801  * fail once we start modifying the data on disk, because JBD has no rollback.
802  */
add_new_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)803 static int add_new_gdb(handle_t *handle, struct inode *inode,
804 		       ext4_group_t group)
805 {
806 	struct super_block *sb = inode->i_sb;
807 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
808 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
809 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
810 	struct buffer_head **o_group_desc, **n_group_desc = NULL;
811 	struct buffer_head *dind = NULL;
812 	struct buffer_head *gdb_bh = NULL;
813 	int gdbackups;
814 	struct ext4_iloc iloc = { .bh = NULL };
815 	__le32 *data;
816 	int err;
817 
818 	if (test_opt(sb, DEBUG))
819 		printk(KERN_DEBUG
820 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
821 		       gdb_num);
822 
823 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
824 	if (IS_ERR(gdb_bh))
825 		return PTR_ERR(gdb_bh);
826 
827 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
828 	if (gdbackups < 0) {
829 		err = gdbackups;
830 		goto errout;
831 	}
832 
833 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
834 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
835 	if (IS_ERR(dind)) {
836 		err = PTR_ERR(dind);
837 		dind = NULL;
838 		goto errout;
839 	}
840 
841 	data = (__le32 *)dind->b_data;
842 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
843 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
844 			     group, gdblock);
845 		err = -EINVAL;
846 		goto errout;
847 	}
848 
849 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
850 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
851 	if (unlikely(err))
852 		goto errout;
853 
854 	BUFFER_TRACE(gdb_bh, "get_write_access");
855 	err = ext4_journal_get_write_access(handle, gdb_bh);
856 	if (unlikely(err))
857 		goto errout;
858 
859 	BUFFER_TRACE(dind, "get_write_access");
860 	err = ext4_journal_get_write_access(handle, dind);
861 	if (unlikely(err)) {
862 		ext4_std_error(sb, err);
863 		goto errout;
864 	}
865 
866 	/* ext4_reserve_inode_write() gets a reference on the iloc */
867 	err = ext4_reserve_inode_write(handle, inode, &iloc);
868 	if (unlikely(err))
869 		goto errout;
870 
871 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
872 				GFP_KERNEL);
873 	if (!n_group_desc) {
874 		err = -ENOMEM;
875 		ext4_warning(sb, "not enough memory for %lu groups",
876 			     gdb_num + 1);
877 		goto errout;
878 	}
879 
880 	/*
881 	 * Finally, we have all of the possible failures behind us...
882 	 *
883 	 * Remove new GDT block from inode double-indirect block and clear out
884 	 * the new GDT block for use (which also "frees" the backup GDT blocks
885 	 * from the reserved inode).  We don't need to change the bitmaps for
886 	 * these blocks, because they are marked as in-use from being in the
887 	 * reserved inode, and will become GDT blocks (primary and backup).
888 	 */
889 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
890 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
891 	if (unlikely(err)) {
892 		ext4_std_error(sb, err);
893 		goto errout;
894 	}
895 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
896 			   (9 - EXT4_SB(sb)->s_cluster_bits);
897 	ext4_mark_iloc_dirty(handle, inode, &iloc);
898 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
899 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
900 	if (unlikely(err)) {
901 		ext4_std_error(sb, err);
902 		iloc.bh = NULL;
903 		goto errout;
904 	}
905 	brelse(dind);
906 
907 	rcu_read_lock();
908 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
909 	memcpy(n_group_desc, o_group_desc,
910 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
911 	rcu_read_unlock();
912 	n_group_desc[gdb_num] = gdb_bh;
913 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
914 	EXT4_SB(sb)->s_gdb_count++;
915 	ext4_kvfree_array_rcu(o_group_desc);
916 
917 	lock_buffer(EXT4_SB(sb)->s_sbh);
918 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
919 	ext4_superblock_csum_set(sb);
920 	unlock_buffer(EXT4_SB(sb)->s_sbh);
921 	err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
922 	if (err)
923 		ext4_std_error(sb, err);
924 	return err;
925 errout:
926 	kvfree(n_group_desc);
927 	brelse(iloc.bh);
928 	brelse(dind);
929 	brelse(gdb_bh);
930 
931 	ext4_debug("leaving with error %d\n", err);
932 	return err;
933 }
934 
935 /*
936  * add_new_gdb_meta_bg is the sister of add_new_gdb.
937  */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)938 static int add_new_gdb_meta_bg(struct super_block *sb,
939 			       handle_t *handle, ext4_group_t group) {
940 	ext4_fsblk_t gdblock;
941 	struct buffer_head *gdb_bh;
942 	struct buffer_head **o_group_desc, **n_group_desc;
943 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
944 	int err;
945 
946 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
947 		   ext4_bg_has_super(sb, group);
948 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
949 	if (IS_ERR(gdb_bh))
950 		return PTR_ERR(gdb_bh);
951 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
952 				GFP_KERNEL);
953 	if (!n_group_desc) {
954 		brelse(gdb_bh);
955 		err = -ENOMEM;
956 		ext4_warning(sb, "not enough memory for %lu groups",
957 			     gdb_num + 1);
958 		return err;
959 	}
960 
961 	rcu_read_lock();
962 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
963 	memcpy(n_group_desc, o_group_desc,
964 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
965 	rcu_read_unlock();
966 	n_group_desc[gdb_num] = gdb_bh;
967 
968 	BUFFER_TRACE(gdb_bh, "get_write_access");
969 	err = ext4_journal_get_write_access(handle, gdb_bh);
970 	if (err) {
971 		kvfree(n_group_desc);
972 		brelse(gdb_bh);
973 		return err;
974 	}
975 
976 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
977 	EXT4_SB(sb)->s_gdb_count++;
978 	ext4_kvfree_array_rcu(o_group_desc);
979 	return err;
980 }
981 
982 /*
983  * Called when we are adding a new group which has a backup copy of each of
984  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
985  * We need to add these reserved backup GDT blocks to the resize inode, so
986  * that they are kept for future resizing and not allocated to files.
987  *
988  * Each reserved backup GDT block will go into a different indirect block.
989  * The indirect blocks are actually the primary reserved GDT blocks,
990  * so we know in advance what their block numbers are.  We only get the
991  * double-indirect block to verify it is pointing to the primary reserved
992  * GDT blocks so we don't overwrite a data block by accident.  The reserved
993  * backup GDT blocks are stored in their reserved primary GDT block.
994  */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)995 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
996 			      ext4_group_t group)
997 {
998 	struct super_block *sb = inode->i_sb;
999 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1000 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1001 	struct buffer_head **primary;
1002 	struct buffer_head *dind;
1003 	struct ext4_iloc iloc;
1004 	ext4_fsblk_t blk;
1005 	__le32 *data, *end;
1006 	int gdbackups = 0;
1007 	int res, i;
1008 	int err;
1009 
1010 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1011 	if (!primary)
1012 		return -ENOMEM;
1013 
1014 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1015 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1016 	if (IS_ERR(dind)) {
1017 		err = PTR_ERR(dind);
1018 		dind = NULL;
1019 		goto exit_free;
1020 	}
1021 
1022 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1023 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1024 					 EXT4_ADDR_PER_BLOCK(sb));
1025 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1026 
1027 	/* Get each reserved primary GDT block and verify it holds backups */
1028 	for (res = 0; res < reserved_gdb; res++, blk++) {
1029 		if (le32_to_cpu(*data) != blk) {
1030 			ext4_warning(sb, "reserved block %llu"
1031 				     " not at offset %ld",
1032 				     blk,
1033 				     (long)(data - (__le32 *)dind->b_data));
1034 			err = -EINVAL;
1035 			goto exit_bh;
1036 		}
1037 		primary[res] = ext4_sb_bread(sb, blk, 0);
1038 		if (IS_ERR(primary[res])) {
1039 			err = PTR_ERR(primary[res]);
1040 			primary[res] = NULL;
1041 			goto exit_bh;
1042 		}
1043 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1044 		if (gdbackups < 0) {
1045 			brelse(primary[res]);
1046 			err = gdbackups;
1047 			goto exit_bh;
1048 		}
1049 		if (++data >= end)
1050 			data = (__le32 *)dind->b_data;
1051 	}
1052 
1053 	for (i = 0; i < reserved_gdb; i++) {
1054 		BUFFER_TRACE(primary[i], "get_write_access");
1055 		if ((err = ext4_journal_get_write_access(handle, primary[i])))
1056 			goto exit_bh;
1057 	}
1058 
1059 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1060 		goto exit_bh;
1061 
1062 	/*
1063 	 * Finally we can add each of the reserved backup GDT blocks from
1064 	 * the new group to its reserved primary GDT block.
1065 	 */
1066 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1067 	for (i = 0; i < reserved_gdb; i++) {
1068 		int err2;
1069 		data = (__le32 *)primary[i]->b_data;
1070 		/* printk("reserving backup %lu[%u] = %lu\n",
1071 		       primary[i]->b_blocknr, gdbackups,
1072 		       blk + primary[i]->b_blocknr); */
1073 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1074 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1075 		if (!err)
1076 			err = err2;
1077 	}
1078 
1079 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1080 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1081 
1082 exit_bh:
1083 	while (--res >= 0)
1084 		brelse(primary[res]);
1085 	brelse(dind);
1086 
1087 exit_free:
1088 	kfree(primary);
1089 
1090 	return err;
1091 }
1092 
1093 /*
1094  * Update the backup copies of the ext4 metadata.  These don't need to be part
1095  * of the main resize transaction, because e2fsck will re-write them if there
1096  * is a problem (basically only OOM will cause a problem).  However, we
1097  * _should_ update the backups if possible, in case the primary gets trashed
1098  * for some reason and we need to run e2fsck from a backup superblock.  The
1099  * important part is that the new block and inode counts are in the backup
1100  * superblocks, and the location of the new group metadata in the GDT backups.
1101  *
1102  * We do not need take the s_resize_lock for this, because these
1103  * blocks are not otherwise touched by the filesystem code when it is
1104  * mounted.  We don't need to worry about last changing from
1105  * sbi->s_groups_count, because the worst that can happen is that we
1106  * do not copy the full number of backups at this time.  The resize
1107  * which changed s_groups_count will backup again.
1108  */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1109 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1110 			   int size, int meta_bg)
1111 {
1112 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1113 	ext4_group_t last;
1114 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1115 	unsigned three = 1;
1116 	unsigned five = 5;
1117 	unsigned seven = 7;
1118 	ext4_group_t group = 0;
1119 	int rest = sb->s_blocksize - size;
1120 	handle_t *handle;
1121 	int err = 0, err2;
1122 
1123 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1124 	if (IS_ERR(handle)) {
1125 		group = 1;
1126 		err = PTR_ERR(handle);
1127 		goto exit_err;
1128 	}
1129 
1130 	if (meta_bg == 0) {
1131 		group = ext4_list_backups(sb, &three, &five, &seven);
1132 		last = sbi->s_groups_count;
1133 	} else {
1134 		group = ext4_get_group_number(sb, blk_off) + 1;
1135 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1136 	}
1137 
1138 	while (group < sbi->s_groups_count) {
1139 		struct buffer_head *bh;
1140 		ext4_fsblk_t backup_block;
1141 
1142 		/* Out of journal space, and can't get more - abort - so sad */
1143 		err = ext4_resize_ensure_credits_batch(handle, 1);
1144 		if (err < 0)
1145 			break;
1146 
1147 		if (meta_bg == 0)
1148 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1149 		else
1150 			backup_block = (ext4_group_first_block_no(sb, group) +
1151 					ext4_bg_has_super(sb, group));
1152 
1153 		bh = sb_getblk(sb, backup_block);
1154 		if (unlikely(!bh)) {
1155 			err = -ENOMEM;
1156 			break;
1157 		}
1158 		ext4_debug("update metadata backup %llu(+%llu)\n",
1159 			   backup_block, backup_block -
1160 			   ext4_group_first_block_no(sb, group));
1161 		BUFFER_TRACE(bh, "get_write_access");
1162 		if ((err = ext4_journal_get_write_access(handle, bh))) {
1163 			brelse(bh);
1164 			break;
1165 		}
1166 		lock_buffer(bh);
1167 		memcpy(bh->b_data, data, size);
1168 		if (rest)
1169 			memset(bh->b_data + size, 0, rest);
1170 		set_buffer_uptodate(bh);
1171 		unlock_buffer(bh);
1172 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1173 		if (unlikely(err))
1174 			ext4_std_error(sb, err);
1175 		brelse(bh);
1176 
1177 		if (meta_bg == 0)
1178 			group = ext4_list_backups(sb, &three, &five, &seven);
1179 		else if (group == last)
1180 			break;
1181 		else
1182 			group = last;
1183 	}
1184 	if ((err2 = ext4_journal_stop(handle)) && !err)
1185 		err = err2;
1186 
1187 	/*
1188 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1189 	 * late to revert the resize, we shouldn't fail just because of
1190 	 * the backup copies (they are only needed in case of corruption).
1191 	 *
1192 	 * However, if we got here we have a journal problem too, so we
1193 	 * can't really start a transaction to mark the superblock.
1194 	 * Chicken out and just set the flag on the hope it will be written
1195 	 * to disk, and if not - we will simply wait until next fsck.
1196 	 */
1197 exit_err:
1198 	if (err) {
1199 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1200 			     "forcing fsck on next reboot", group, err);
1201 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1202 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1203 		mark_buffer_dirty(sbi->s_sbh);
1204 	}
1205 }
1206 
1207 /*
1208  * ext4_add_new_descs() adds @count group descriptor of groups
1209  * starting at @group
1210  *
1211  * @handle: journal handle
1212  * @sb: super block
1213  * @group: the group no. of the first group desc to be added
1214  * @resize_inode: the resize inode
1215  * @count: number of group descriptors to be added
1216  */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1217 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1218 			      ext4_group_t group, struct inode *resize_inode,
1219 			      ext4_group_t count)
1220 {
1221 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1222 	struct ext4_super_block *es = sbi->s_es;
1223 	struct buffer_head *gdb_bh;
1224 	int i, gdb_off, gdb_num, err = 0;
1225 	int meta_bg;
1226 
1227 	meta_bg = ext4_has_feature_meta_bg(sb);
1228 	for (i = 0; i < count; i++, group++) {
1229 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1230 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1231 
1232 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1233 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1234 
1235 		/*
1236 		 * We will only either add reserved group blocks to a backup group
1237 		 * or remove reserved blocks for the first group in a new group block.
1238 		 * Doing both would be mean more complex code, and sane people don't
1239 		 * use non-sparse filesystems anymore.  This is already checked above.
1240 		 */
1241 		if (gdb_off) {
1242 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1243 						     gdb_num);
1244 			BUFFER_TRACE(gdb_bh, "get_write_access");
1245 			err = ext4_journal_get_write_access(handle, gdb_bh);
1246 
1247 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1248 				err = reserve_backup_gdb(handle, resize_inode, group);
1249 		} else if (meta_bg != 0) {
1250 			err = add_new_gdb_meta_bg(sb, handle, group);
1251 		} else {
1252 			err = add_new_gdb(handle, resize_inode, group);
1253 		}
1254 		if (err)
1255 			break;
1256 	}
1257 	return err;
1258 }
1259 
ext4_get_bitmap(struct super_block * sb,__u64 block)1260 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1261 {
1262 	struct buffer_head *bh = sb_getblk(sb, block);
1263 	if (unlikely(!bh))
1264 		return NULL;
1265 	if (!bh_uptodate_or_lock(bh)) {
1266 		if (ext4_read_bh(bh, 0, NULL) < 0) {
1267 			brelse(bh);
1268 			return NULL;
1269 		}
1270 	}
1271 
1272 	return bh;
1273 }
1274 
ext4_set_bitmap_checksums(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1275 static int ext4_set_bitmap_checksums(struct super_block *sb,
1276 				     ext4_group_t group,
1277 				     struct ext4_group_desc *gdp,
1278 				     struct ext4_new_group_data *group_data)
1279 {
1280 	struct buffer_head *bh;
1281 
1282 	if (!ext4_has_metadata_csum(sb))
1283 		return 0;
1284 
1285 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1286 	if (!bh)
1287 		return -EIO;
1288 	ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1289 				   EXT4_INODES_PER_GROUP(sb) / 8);
1290 	brelse(bh);
1291 
1292 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1293 	if (!bh)
1294 		return -EIO;
1295 	ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1296 	brelse(bh);
1297 
1298 	return 0;
1299 }
1300 
1301 /*
1302  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1303  */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1304 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1305 				struct ext4_new_flex_group_data *flex_gd)
1306 {
1307 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1308 	struct ext4_group_desc		*gdp;
1309 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1310 	struct buffer_head		*gdb_bh;
1311 	ext4_group_t			group;
1312 	__u16				*bg_flags = flex_gd->bg_flags;
1313 	int				i, gdb_off, gdb_num, err = 0;
1314 
1315 
1316 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1317 		group = group_data->group;
1318 
1319 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1320 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1321 
1322 		/*
1323 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1324 		 */
1325 		gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1326 		/* Update group descriptor block for new group */
1327 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1328 						 gdb_off * EXT4_DESC_SIZE(sb));
1329 
1330 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1331 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1332 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1333 		err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1334 		if (err) {
1335 			ext4_std_error(sb, err);
1336 			break;
1337 		}
1338 
1339 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1340 		ext4_free_group_clusters_set(sb, gdp,
1341 					     group_data->free_clusters_count);
1342 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1343 		if (ext4_has_group_desc_csum(sb))
1344 			ext4_itable_unused_set(sb, gdp,
1345 					       EXT4_INODES_PER_GROUP(sb));
1346 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1347 		ext4_group_desc_csum_set(sb, group, gdp);
1348 
1349 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1350 		if (unlikely(err)) {
1351 			ext4_std_error(sb, err);
1352 			break;
1353 		}
1354 
1355 		/*
1356 		 * We can allocate memory for mb_alloc based on the new group
1357 		 * descriptor
1358 		 */
1359 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1360 		if (err)
1361 			break;
1362 	}
1363 	return err;
1364 }
1365 
1366 /*
1367  * ext4_update_super() updates the super block so that the newly added
1368  * groups can be seen by the filesystem.
1369  *
1370  * @sb: super block
1371  * @flex_gd: new added groups
1372  */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1373 static void ext4_update_super(struct super_block *sb,
1374 			     struct ext4_new_flex_group_data *flex_gd)
1375 {
1376 	ext4_fsblk_t blocks_count = 0;
1377 	ext4_fsblk_t free_blocks = 0;
1378 	ext4_fsblk_t reserved_blocks = 0;
1379 	struct ext4_new_group_data *group_data = flex_gd->groups;
1380 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1381 	struct ext4_super_block *es = sbi->s_es;
1382 	int i;
1383 
1384 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1385 	/*
1386 	 * Make the new blocks and inodes valid next.  We do this before
1387 	 * increasing the group count so that once the group is enabled,
1388 	 * all of its blocks and inodes are already valid.
1389 	 *
1390 	 * We always allocate group-by-group, then block-by-block or
1391 	 * inode-by-inode within a group, so enabling these
1392 	 * blocks/inodes before the group is live won't actually let us
1393 	 * allocate the new space yet.
1394 	 */
1395 	for (i = 0; i < flex_gd->count; i++) {
1396 		blocks_count += group_data[i].blocks_count;
1397 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1398 	}
1399 
1400 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1401 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1402 	reserved_blocks *= blocks_count;
1403 	do_div(reserved_blocks, 100);
1404 
1405 	lock_buffer(sbi->s_sbh);
1406 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1407 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1408 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1409 		     flex_gd->count);
1410 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1411 		     flex_gd->count);
1412 
1413 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1414 	/*
1415 	 * We need to protect s_groups_count against other CPUs seeing
1416 	 * inconsistent state in the superblock.
1417 	 *
1418 	 * The precise rules we use are:
1419 	 *
1420 	 * * Writers must perform a smp_wmb() after updating all
1421 	 *   dependent data and before modifying the groups count
1422 	 *
1423 	 * * Readers must perform an smp_rmb() after reading the groups
1424 	 *   count and before reading any dependent data.
1425 	 *
1426 	 * NB. These rules can be relaxed when checking the group count
1427 	 * while freeing data, as we can only allocate from a block
1428 	 * group after serialising against the group count, and we can
1429 	 * only then free after serialising in turn against that
1430 	 * allocation.
1431 	 */
1432 	smp_wmb();
1433 
1434 	/* Update the global fs size fields */
1435 	sbi->s_groups_count += flex_gd->count;
1436 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1437 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1438 
1439 	/* Update the reserved block counts only once the new group is
1440 	 * active. */
1441 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1442 				reserved_blocks);
1443 	ext4_superblock_csum_set(sb);
1444 	unlock_buffer(sbi->s_sbh);
1445 
1446 	/* Update the free space counts */
1447 	percpu_counter_add(&sbi->s_freeclusters_counter,
1448 			   EXT4_NUM_B2C(sbi, free_blocks));
1449 	percpu_counter_add(&sbi->s_freeinodes_counter,
1450 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1451 
1452 	ext4_debug("free blocks count %llu",
1453 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1454 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1455 		ext4_group_t flex_group;
1456 		struct flex_groups *fg;
1457 
1458 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1459 		fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1460 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1461 			     &fg->free_clusters);
1462 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1463 			   &fg->free_inodes);
1464 	}
1465 
1466 	/*
1467 	 * Update the fs overhead information
1468 	 */
1469 	ext4_calculate_overhead(sb);
1470 	es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1471 
1472 	if (test_opt(sb, DEBUG))
1473 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1474 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1475 		       blocks_count, free_blocks, reserved_blocks);
1476 }
1477 
1478 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1479  * _before_ we start modifying the filesystem, because we cannot abort the
1480  * transaction and not have it write the data to disk.
1481  */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1482 static int ext4_flex_group_add(struct super_block *sb,
1483 			       struct inode *resize_inode,
1484 			       struct ext4_new_flex_group_data *flex_gd)
1485 {
1486 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1487 	struct ext4_super_block *es = sbi->s_es;
1488 	ext4_fsblk_t o_blocks_count;
1489 	ext4_grpblk_t last;
1490 	ext4_group_t group;
1491 	handle_t *handle;
1492 	unsigned reserved_gdb;
1493 	int err = 0, err2 = 0, credit;
1494 
1495 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1496 
1497 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1498 	o_blocks_count = ext4_blocks_count(es);
1499 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1500 	BUG_ON(last);
1501 
1502 	err = setup_new_flex_group_blocks(sb, flex_gd);
1503 	if (err)
1504 		goto exit;
1505 	/*
1506 	 * We will always be modifying at least the superblock and  GDT
1507 	 * blocks.  If we are adding a group past the last current GDT block,
1508 	 * we will also modify the inode and the dindirect block.  If we
1509 	 * are adding a group with superblock/GDT backups  we will also
1510 	 * modify each of the reserved GDT dindirect blocks.
1511 	 */
1512 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1513 	/* GDT blocks */
1514 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1515 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1516 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1517 	if (IS_ERR(handle)) {
1518 		err = PTR_ERR(handle);
1519 		goto exit;
1520 	}
1521 
1522 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1523 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1524 	if (err)
1525 		goto exit_journal;
1526 
1527 	group = flex_gd->groups[0].group;
1528 	BUG_ON(group != sbi->s_groups_count);
1529 	err = ext4_add_new_descs(handle, sb, group,
1530 				resize_inode, flex_gd->count);
1531 	if (err)
1532 		goto exit_journal;
1533 
1534 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1535 	if (err)
1536 		goto exit_journal;
1537 
1538 	ext4_update_super(sb, flex_gd);
1539 
1540 	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1541 
1542 exit_journal:
1543 	err2 = ext4_journal_stop(handle);
1544 	if (!err)
1545 		err = err2;
1546 
1547 	if (!err) {
1548 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1549 		int gdb_num_end = ((group + flex_gd->count - 1) /
1550 				   EXT4_DESC_PER_BLOCK(sb));
1551 		int meta_bg = ext4_has_feature_meta_bg(sb);
1552 		sector_t old_gdb = 0;
1553 
1554 		update_backups(sb, ext4_group_first_block_no(sb, 0),
1555 			       (char *)es, sizeof(struct ext4_super_block), 0);
1556 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1557 			struct buffer_head *gdb_bh;
1558 
1559 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1560 						     gdb_num);
1561 			if (old_gdb == gdb_bh->b_blocknr)
1562 				continue;
1563 			update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1564 				       gdb_bh->b_size, meta_bg);
1565 			old_gdb = gdb_bh->b_blocknr;
1566 		}
1567 	}
1568 exit:
1569 	return err;
1570 }
1571 
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)1572 static int ext4_setup_next_flex_gd(struct super_block *sb,
1573 				    struct ext4_new_flex_group_data *flex_gd,
1574 				    ext4_fsblk_t n_blocks_count,
1575 				    unsigned long flexbg_size)
1576 {
1577 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1578 	struct ext4_super_block *es = sbi->s_es;
1579 	struct ext4_new_group_data *group_data = flex_gd->groups;
1580 	ext4_fsblk_t o_blocks_count;
1581 	ext4_group_t n_group;
1582 	ext4_group_t group;
1583 	ext4_group_t last_group;
1584 	ext4_grpblk_t last;
1585 	ext4_grpblk_t clusters_per_group;
1586 	unsigned long i;
1587 
1588 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1589 
1590 	o_blocks_count = ext4_blocks_count(es);
1591 
1592 	if (o_blocks_count == n_blocks_count)
1593 		return 0;
1594 
1595 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1596 	BUG_ON(last);
1597 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1598 
1599 	last_group = group | (flexbg_size - 1);
1600 	if (last_group > n_group)
1601 		last_group = n_group;
1602 
1603 	flex_gd->count = last_group - group + 1;
1604 
1605 	for (i = 0; i < flex_gd->count; i++) {
1606 		int overhead;
1607 
1608 		group_data[i].group = group + i;
1609 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1610 		overhead = ext4_group_overhead_blocks(sb, group + i);
1611 		group_data[i].mdata_blocks = overhead;
1612 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1613 		if (ext4_has_group_desc_csum(sb)) {
1614 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1615 					       EXT4_BG_INODE_UNINIT;
1616 			if (!test_opt(sb, INIT_INODE_TABLE))
1617 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1618 		} else
1619 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1620 	}
1621 
1622 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1623 		/* We need to initialize block bitmap of last group. */
1624 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1625 
1626 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1627 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1628 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1629 						       last - 1;
1630 	}
1631 
1632 	return 1;
1633 }
1634 
1635 /* Add group descriptor data to an existing or new group descriptor block.
1636  * Ensure we handle all possible error conditions _before_ we start modifying
1637  * the filesystem, because we cannot abort the transaction and not have it
1638  * write the data to disk.
1639  *
1640  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1641  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1642  *
1643  * We only need to hold the superblock lock while we are actually adding
1644  * in the new group's counts to the superblock.  Prior to that we have
1645  * not really "added" the group at all.  We re-check that we are still
1646  * adding in the last group in case things have changed since verifying.
1647  */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1648 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1649 {
1650 	struct ext4_new_flex_group_data flex_gd;
1651 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1652 	struct ext4_super_block *es = sbi->s_es;
1653 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1654 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1655 	struct inode *inode = NULL;
1656 	int gdb_off;
1657 	int err;
1658 	__u16 bg_flags = 0;
1659 
1660 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1661 
1662 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1663 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1664 		return -EPERM;
1665 	}
1666 
1667 	if (ext4_blocks_count(es) + input->blocks_count <
1668 	    ext4_blocks_count(es)) {
1669 		ext4_warning(sb, "blocks_count overflow");
1670 		return -EINVAL;
1671 	}
1672 
1673 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1674 	    le32_to_cpu(es->s_inodes_count)) {
1675 		ext4_warning(sb, "inodes_count overflow");
1676 		return -EINVAL;
1677 	}
1678 
1679 	if (reserved_gdb || gdb_off == 0) {
1680 		if (!ext4_has_feature_resize_inode(sb) ||
1681 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1682 			ext4_warning(sb,
1683 				     "No reserved GDT blocks, can't resize");
1684 			return -EPERM;
1685 		}
1686 		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1687 		if (IS_ERR(inode)) {
1688 			ext4_warning(sb, "Error opening resize inode");
1689 			return PTR_ERR(inode);
1690 		}
1691 	}
1692 
1693 
1694 	err = verify_group_input(sb, input);
1695 	if (err)
1696 		goto out;
1697 
1698 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1699 	if (err)
1700 		goto out;
1701 
1702 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1703 	if (err)
1704 		goto out;
1705 
1706 	flex_gd.count = 1;
1707 	flex_gd.groups = input;
1708 	flex_gd.bg_flags = &bg_flags;
1709 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1710 out:
1711 	iput(inode);
1712 	return err;
1713 } /* ext4_group_add */
1714 
1715 /*
1716  * extend a group without checking assuming that checking has been done.
1717  */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1718 static int ext4_group_extend_no_check(struct super_block *sb,
1719 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1720 {
1721 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1722 	handle_t *handle;
1723 	int err = 0, err2;
1724 
1725 	/* We will update the superblock, one block bitmap, and
1726 	 * one group descriptor via ext4_group_add_blocks().
1727 	 */
1728 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1729 	if (IS_ERR(handle)) {
1730 		err = PTR_ERR(handle);
1731 		ext4_warning(sb, "error %d on journal start", err);
1732 		return err;
1733 	}
1734 
1735 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1736 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1737 	if (err) {
1738 		ext4_warning(sb, "error %d on journal write access", err);
1739 		goto errout;
1740 	}
1741 
1742 	lock_buffer(EXT4_SB(sb)->s_sbh);
1743 	ext4_blocks_count_set(es, o_blocks_count + add);
1744 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1745 	ext4_superblock_csum_set(sb);
1746 	unlock_buffer(EXT4_SB(sb)->s_sbh);
1747 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1748 		   o_blocks_count + add);
1749 	/* We add the blocks to the bitmap and set the group need init bit */
1750 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1751 	if (err)
1752 		goto errout;
1753 	ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1754 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1755 		   o_blocks_count + add);
1756 errout:
1757 	err2 = ext4_journal_stop(handle);
1758 	if (err2 && !err)
1759 		err = err2;
1760 
1761 	if (!err) {
1762 		if (test_opt(sb, DEBUG))
1763 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1764 			       "blocks\n", ext4_blocks_count(es));
1765 		update_backups(sb, ext4_group_first_block_no(sb, 0),
1766 			       (char *)es, sizeof(struct ext4_super_block), 0);
1767 	}
1768 	return err;
1769 }
1770 
1771 /*
1772  * Extend the filesystem to the new number of blocks specified.  This entry
1773  * point is only used to extend the current filesystem to the end of the last
1774  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1775  * for emergencies (because it has no dependencies on reserved blocks).
1776  *
1777  * If we _really_ wanted, we could use default values to call ext4_group_add()
1778  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1779  * GDT blocks are reserved to grow to the desired size.
1780  */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1781 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1782 		      ext4_fsblk_t n_blocks_count)
1783 {
1784 	ext4_fsblk_t o_blocks_count;
1785 	ext4_grpblk_t last;
1786 	ext4_grpblk_t add;
1787 	struct buffer_head *bh;
1788 	int err;
1789 	ext4_group_t group;
1790 
1791 	o_blocks_count = ext4_blocks_count(es);
1792 
1793 	if (test_opt(sb, DEBUG))
1794 		ext4_msg(sb, KERN_DEBUG,
1795 			 "extending last group from %llu to %llu blocks",
1796 			 o_blocks_count, n_blocks_count);
1797 
1798 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1799 		return 0;
1800 
1801 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1802 		ext4_msg(sb, KERN_ERR,
1803 			 "filesystem too large to resize to %llu blocks safely",
1804 			 n_blocks_count);
1805 		return -EINVAL;
1806 	}
1807 
1808 	if (n_blocks_count < o_blocks_count) {
1809 		ext4_warning(sb, "can't shrink FS - resize aborted");
1810 		return -EINVAL;
1811 	}
1812 
1813 	/* Handle the remaining blocks in the last group only. */
1814 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1815 
1816 	if (last == 0) {
1817 		ext4_warning(sb, "need to use ext2online to resize further");
1818 		return -EPERM;
1819 	}
1820 
1821 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1822 
1823 	if (o_blocks_count + add < o_blocks_count) {
1824 		ext4_warning(sb, "blocks_count overflow");
1825 		return -EINVAL;
1826 	}
1827 
1828 	if (o_blocks_count + add > n_blocks_count)
1829 		add = n_blocks_count - o_blocks_count;
1830 
1831 	if (o_blocks_count + add < n_blocks_count)
1832 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1833 			     o_blocks_count + add, add);
1834 
1835 	/* See if the device is actually as big as what was requested */
1836 	bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1837 	if (IS_ERR(bh)) {
1838 		ext4_warning(sb, "can't read last block, resize aborted");
1839 		return -ENOSPC;
1840 	}
1841 	brelse(bh);
1842 
1843 	err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1844 	return err;
1845 } /* ext4_group_extend */
1846 
1847 
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1848 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1849 {
1850 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1851 }
1852 
1853 /*
1854  * Release the resize inode and drop the resize_inode feature if there
1855  * are no more reserved gdt blocks, and then convert the file system
1856  * to enable meta_bg
1857  */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1858 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1859 {
1860 	handle_t *handle;
1861 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1862 	struct ext4_super_block *es = sbi->s_es;
1863 	struct ext4_inode_info *ei = EXT4_I(inode);
1864 	ext4_fsblk_t nr;
1865 	int i, ret, err = 0;
1866 	int credits = 1;
1867 
1868 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1869 	if (inode) {
1870 		if (es->s_reserved_gdt_blocks) {
1871 			ext4_error(sb, "Unexpected non-zero "
1872 				   "s_reserved_gdt_blocks");
1873 			return -EPERM;
1874 		}
1875 
1876 		/* Do a quick sanity check of the resize inode */
1877 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1878 					     (9 - sbi->s_cluster_bits)))
1879 			goto invalid_resize_inode;
1880 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1881 			if (i == EXT4_DIND_BLOCK) {
1882 				if (ei->i_data[i])
1883 					continue;
1884 				else
1885 					goto invalid_resize_inode;
1886 			}
1887 			if (ei->i_data[i])
1888 				goto invalid_resize_inode;
1889 		}
1890 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1891 	}
1892 
1893 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1894 	if (IS_ERR(handle))
1895 		return PTR_ERR(handle);
1896 
1897 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1898 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1899 	if (err)
1900 		goto errout;
1901 
1902 	lock_buffer(sbi->s_sbh);
1903 	ext4_clear_feature_resize_inode(sb);
1904 	ext4_set_feature_meta_bg(sb);
1905 	sbi->s_es->s_first_meta_bg =
1906 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1907 	ext4_superblock_csum_set(sb);
1908 	unlock_buffer(sbi->s_sbh);
1909 
1910 	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1911 	if (err) {
1912 		ext4_std_error(sb, err);
1913 		goto errout;
1914 	}
1915 
1916 	if (inode) {
1917 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1918 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1919 				 EXT4_FREE_BLOCKS_METADATA |
1920 				 EXT4_FREE_BLOCKS_FORGET);
1921 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1922 		inode->i_blocks = 0;
1923 
1924 		err = ext4_mark_inode_dirty(handle, inode);
1925 		if (err)
1926 			ext4_std_error(sb, err);
1927 	}
1928 
1929 errout:
1930 	ret = ext4_journal_stop(handle);
1931 	if (!err)
1932 		err = ret;
1933 	return ret;
1934 
1935 invalid_resize_inode:
1936 	ext4_error(sb, "corrupted/inconsistent resize inode");
1937 	return -EINVAL;
1938 }
1939 
1940 /*
1941  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1942  *
1943  * @sb: super block of the fs to be resized
1944  * @n_blocks_count: the number of blocks resides in the resized fs
1945  */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)1946 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1947 {
1948 	struct ext4_new_flex_group_data *flex_gd = NULL;
1949 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1950 	struct ext4_super_block *es = sbi->s_es;
1951 	struct buffer_head *bh;
1952 	struct inode *resize_inode = NULL;
1953 	ext4_grpblk_t add, offset;
1954 	unsigned long n_desc_blocks;
1955 	unsigned long o_desc_blocks;
1956 	ext4_group_t o_group;
1957 	ext4_group_t n_group;
1958 	ext4_fsblk_t o_blocks_count;
1959 	ext4_fsblk_t n_blocks_count_retry = 0;
1960 	unsigned long last_update_time = 0;
1961 	int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1962 	int meta_bg;
1963 
1964 	/* See if the device is actually as big as what was requested */
1965 	bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1966 	if (IS_ERR(bh)) {
1967 		ext4_warning(sb, "can't read last block, resize aborted");
1968 		return -ENOSPC;
1969 	}
1970 	brelse(bh);
1971 
1972 	/*
1973 	 * For bigalloc, trim the requested size to the nearest cluster
1974 	 * boundary to avoid creating an unusable filesystem. We do this
1975 	 * silently, instead of returning an error, to avoid breaking
1976 	 * callers that blindly resize the filesystem to the full size of
1977 	 * the underlying block device.
1978 	 */
1979 	if (ext4_has_feature_bigalloc(sb))
1980 		n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1981 
1982 retry:
1983 	o_blocks_count = ext4_blocks_count(es);
1984 
1985 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1986 		 "to %llu blocks", o_blocks_count, n_blocks_count);
1987 
1988 	if (n_blocks_count < o_blocks_count) {
1989 		/* On-line shrinking not supported */
1990 		ext4_warning(sb, "can't shrink FS - resize aborted");
1991 		return -EINVAL;
1992 	}
1993 
1994 	if (n_blocks_count == o_blocks_count)
1995 		/* Nothing need to do */
1996 		return 0;
1997 
1998 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1999 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2000 		ext4_warning(sb, "resize would cause inodes_count overflow");
2001 		return -EINVAL;
2002 	}
2003 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2004 
2005 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2006 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2007 
2008 	meta_bg = ext4_has_feature_meta_bg(sb);
2009 
2010 	if (ext4_has_feature_resize_inode(sb)) {
2011 		if (meta_bg) {
2012 			ext4_error(sb, "resize_inode and meta_bg enabled "
2013 				   "simultaneously");
2014 			return -EINVAL;
2015 		}
2016 		if (n_desc_blocks > o_desc_blocks +
2017 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
2018 			n_blocks_count_retry = n_blocks_count;
2019 			n_desc_blocks = o_desc_blocks +
2020 				le16_to_cpu(es->s_reserved_gdt_blocks);
2021 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2022 			n_blocks_count = (ext4_fsblk_t)n_group *
2023 				EXT4_BLOCKS_PER_GROUP(sb) +
2024 				le32_to_cpu(es->s_first_data_block);
2025 			n_group--; /* set to last group number */
2026 		}
2027 
2028 		if (!resize_inode)
2029 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2030 						 EXT4_IGET_SPECIAL);
2031 		if (IS_ERR(resize_inode)) {
2032 			ext4_warning(sb, "Error opening resize inode");
2033 			return PTR_ERR(resize_inode);
2034 		}
2035 	}
2036 
2037 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2038 		err = ext4_convert_meta_bg(sb, resize_inode);
2039 		if (err)
2040 			goto out;
2041 		if (resize_inode) {
2042 			iput(resize_inode);
2043 			resize_inode = NULL;
2044 		}
2045 		if (n_blocks_count_retry) {
2046 			n_blocks_count = n_blocks_count_retry;
2047 			n_blocks_count_retry = 0;
2048 			goto retry;
2049 		}
2050 	}
2051 
2052 	/*
2053 	 * Make sure the last group has enough space so that it's
2054 	 * guaranteed to have enough space for all metadata blocks
2055 	 * that it might need to hold.  (We might not need to store
2056 	 * the inode table blocks in the last block group, but there
2057 	 * will be cases where this might be needed.)
2058 	 */
2059 	if ((ext4_group_first_block_no(sb, n_group) +
2060 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
2061 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2062 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2063 		n_group--;
2064 		n_blocks_count_retry = 0;
2065 		if (resize_inode) {
2066 			iput(resize_inode);
2067 			resize_inode = NULL;
2068 		}
2069 		goto retry;
2070 	}
2071 
2072 	/* extend the last group */
2073 	if (n_group == o_group)
2074 		add = n_blocks_count - o_blocks_count;
2075 	else
2076 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2077 	if (add > 0) {
2078 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2079 		if (err)
2080 			goto out;
2081 	}
2082 
2083 	if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2084 		goto out;
2085 
2086 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2087 	if (err)
2088 		goto out;
2089 
2090 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2091 	if (err)
2092 		goto out;
2093 
2094 	flex_gd = alloc_flex_gd(flexbg_size);
2095 	if (flex_gd == NULL) {
2096 		err = -ENOMEM;
2097 		goto out;
2098 	}
2099 
2100 	/* Add flex groups. Note that a regular group is a
2101 	 * flex group with 1 group.
2102 	 */
2103 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2104 					      flexbg_size)) {
2105 		if (jiffies - last_update_time > HZ * 10) {
2106 			if (last_update_time)
2107 				ext4_msg(sb, KERN_INFO,
2108 					 "resized to %llu blocks",
2109 					 ext4_blocks_count(es));
2110 			last_update_time = jiffies;
2111 		}
2112 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2113 			break;
2114 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2115 		if (unlikely(err))
2116 			break;
2117 	}
2118 
2119 	if (!err && n_blocks_count_retry) {
2120 		n_blocks_count = n_blocks_count_retry;
2121 		n_blocks_count_retry = 0;
2122 		free_flex_gd(flex_gd);
2123 		flex_gd = NULL;
2124 		if (resize_inode) {
2125 			iput(resize_inode);
2126 			resize_inode = NULL;
2127 		}
2128 		goto retry;
2129 	}
2130 
2131 out:
2132 	if (flex_gd)
2133 		free_flex_gd(flex_gd);
2134 	if (resize_inode != NULL)
2135 		iput(resize_inode);
2136 	if (err)
2137 		ext4_warning(sb, "error (%d) occurred during "
2138 			     "file system resize", err);
2139 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2140 		 ext4_blocks_count(es));
2141 	return err;
2142 }
2143