xref: /openbmc/linux/fs/ext4/resize.c (revision 37b6a3ba)
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 #include <linux/jiffies.h>
18 
19 #include "ext4_jbd2.h"
20 
21 struct ext4_rcu_ptr {
22 	struct rcu_head rcu;
23 	void *ptr;
24 };
25 
ext4_rcu_ptr_callback(struct rcu_head * head)26 static void ext4_rcu_ptr_callback(struct rcu_head *head)
27 {
28 	struct ext4_rcu_ptr *ptr;
29 
30 	ptr = container_of(head, struct ext4_rcu_ptr, rcu);
31 	kvfree(ptr->ptr);
32 	kfree(ptr);
33 }
34 
ext4_kvfree_array_rcu(void * to_free)35 void ext4_kvfree_array_rcu(void *to_free)
36 {
37 	struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
38 
39 	if (ptr) {
40 		ptr->ptr = to_free;
41 		call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
42 		return;
43 	}
44 	synchronize_rcu();
45 	kvfree(to_free);
46 }
47 
ext4_resize_begin(struct super_block * sb)48 int ext4_resize_begin(struct super_block *sb)
49 {
50 	struct ext4_sb_info *sbi = EXT4_SB(sb);
51 	int ret = 0;
52 
53 	if (!capable(CAP_SYS_RESOURCE))
54 		return -EPERM;
55 
56 	/*
57 	 * If the reserved GDT blocks is non-zero, the resize_inode feature
58 	 * should always be set.
59 	 */
60 	if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
61 	    !ext4_has_feature_resize_inode(sb)) {
62 		ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
63 		return -EFSCORRUPTED;
64 	}
65 
66 	/*
67 	 * If we are not using the primary superblock/GDT copy don't resize,
68          * because the user tools have no way of handling this.  Probably a
69          * bad time to do it anyways.
70          */
71 	if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
72 	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
73 		ext4_warning(sb, "won't resize using backup superblock at %llu",
74 			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
75 		return -EPERM;
76 	}
77 
78 	/*
79 	 * We are not allowed to do online-resizing on a filesystem mounted
80 	 * with error, because it can destroy the filesystem easily.
81 	 */
82 	if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
83 		ext4_warning(sb, "There are errors in the filesystem, "
84 			     "so online resizing is not allowed");
85 		return -EPERM;
86 	}
87 
88 	if (ext4_has_feature_sparse_super2(sb)) {
89 		ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
90 		return -EOPNOTSUPP;
91 	}
92 
93 	if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
94 				  &EXT4_SB(sb)->s_ext4_flags))
95 		ret = -EBUSY;
96 
97 	return ret;
98 }
99 
ext4_resize_end(struct super_block * sb,bool update_backups)100 int ext4_resize_end(struct super_block *sb, bool update_backups)
101 {
102 	clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
103 	smp_mb__after_atomic();
104 	if (update_backups)
105 		return ext4_update_overhead(sb, true);
106 	return 0;
107 }
108 
ext4_meta_bg_first_group(struct super_block * sb,ext4_group_t group)109 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
110 					     ext4_group_t group) {
111 	return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
112 	       EXT4_DESC_PER_BLOCK_BITS(sb);
113 }
114 
ext4_meta_bg_first_block_no(struct super_block * sb,ext4_group_t group)115 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
116 					     ext4_group_t group) {
117 	group = ext4_meta_bg_first_group(sb, group);
118 	return ext4_group_first_block_no(sb, group);
119 }
120 
ext4_group_overhead_blocks(struct super_block * sb,ext4_group_t group)121 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
122 						ext4_group_t group) {
123 	ext4_grpblk_t overhead;
124 	overhead = ext4_bg_num_gdb(sb, group);
125 	if (ext4_bg_has_super(sb, group))
126 		overhead += 1 +
127 			  le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
128 	return overhead;
129 }
130 
131 #define outside(b, first, last)	((b) < (first) || (b) >= (last))
132 #define inside(b, first, last)	((b) >= (first) && (b) < (last))
133 
verify_group_input(struct super_block * sb,struct ext4_new_group_data * input)134 static int verify_group_input(struct super_block *sb,
135 			      struct ext4_new_group_data *input)
136 {
137 	struct ext4_sb_info *sbi = EXT4_SB(sb);
138 	struct ext4_super_block *es = sbi->s_es;
139 	ext4_fsblk_t start = ext4_blocks_count(es);
140 	ext4_fsblk_t end = start + input->blocks_count;
141 	ext4_group_t group = input->group;
142 	ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
143 	unsigned overhead;
144 	ext4_fsblk_t metaend;
145 	struct buffer_head *bh = NULL;
146 	ext4_grpblk_t free_blocks_count, offset;
147 	int err = -EINVAL;
148 
149 	if (group != sbi->s_groups_count) {
150 		ext4_warning(sb, "Cannot add at group %u (only %u groups)",
151 			     input->group, sbi->s_groups_count);
152 		return -EINVAL;
153 	}
154 
155 	overhead = ext4_group_overhead_blocks(sb, group);
156 	metaend = start + overhead;
157 	input->free_clusters_count = free_blocks_count =
158 		input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
159 
160 	if (test_opt(sb, DEBUG))
161 		printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
162 		       "(%d free, %u reserved)\n",
163 		       ext4_bg_has_super(sb, input->group) ? "normal" :
164 		       "no-super", input->group, input->blocks_count,
165 		       free_blocks_count, input->reserved_blocks);
166 
167 	ext4_get_group_no_and_offset(sb, start, NULL, &offset);
168 	if (offset != 0)
169 			ext4_warning(sb, "Last group not full");
170 	else if (input->reserved_blocks > input->blocks_count / 5)
171 		ext4_warning(sb, "Reserved blocks too high (%u)",
172 			     input->reserved_blocks);
173 	else if (free_blocks_count < 0)
174 		ext4_warning(sb, "Bad blocks count %u",
175 			     input->blocks_count);
176 	else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
177 		err = PTR_ERR(bh);
178 		bh = NULL;
179 		ext4_warning(sb, "Cannot read last block (%llu)",
180 			     end - 1);
181 	} else if (outside(input->block_bitmap, start, end))
182 		ext4_warning(sb, "Block bitmap not in group (block %llu)",
183 			     (unsigned long long)input->block_bitmap);
184 	else if (outside(input->inode_bitmap, start, end))
185 		ext4_warning(sb, "Inode bitmap not in group (block %llu)",
186 			     (unsigned long long)input->inode_bitmap);
187 	else if (outside(input->inode_table, start, end) ||
188 		 outside(itend - 1, start, end))
189 		ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
190 			     (unsigned long long)input->inode_table, itend - 1);
191 	else if (input->inode_bitmap == input->block_bitmap)
192 		ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
193 			     (unsigned long long)input->block_bitmap);
194 	else if (inside(input->block_bitmap, input->inode_table, itend))
195 		ext4_warning(sb, "Block bitmap (%llu) in inode table "
196 			     "(%llu-%llu)",
197 			     (unsigned long long)input->block_bitmap,
198 			     (unsigned long long)input->inode_table, itend - 1);
199 	else if (inside(input->inode_bitmap, input->inode_table, itend))
200 		ext4_warning(sb, "Inode bitmap (%llu) in inode table "
201 			     "(%llu-%llu)",
202 			     (unsigned long long)input->inode_bitmap,
203 			     (unsigned long long)input->inode_table, itend - 1);
204 	else if (inside(input->block_bitmap, start, metaend))
205 		ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
206 			     (unsigned long long)input->block_bitmap,
207 			     start, metaend - 1);
208 	else if (inside(input->inode_bitmap, start, metaend))
209 		ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
210 			     (unsigned long long)input->inode_bitmap,
211 			     start, metaend - 1);
212 	else if (inside(input->inode_table, start, metaend) ||
213 		 inside(itend - 1, start, metaend))
214 		ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
215 			     "(%llu-%llu)",
216 			     (unsigned long long)input->inode_table,
217 			     itend - 1, start, metaend - 1);
218 	else
219 		err = 0;
220 	brelse(bh);
221 
222 	return err;
223 }
224 
225 /*
226  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
227  * group each time.
228  */
229 struct ext4_new_flex_group_data {
230 	struct ext4_new_group_data *groups;	/* new_group_data for groups
231 						   in the flex group */
232 	__u16 *bg_flags;			/* block group flags of groups
233 						   in @groups */
234 	ext4_group_t resize_bg;			/* number of allocated
235 						   new_group_data */
236 	ext4_group_t count;			/* number of groups in @groups
237 						 */
238 };
239 
240 /*
241  * Avoiding memory allocation failures due to too many groups added each time.
242  */
243 #define MAX_RESIZE_BG				16384
244 
245 /*
246  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
247  * @flexbg_size.
248  *
249  * Returns NULL on failure otherwise address of the allocated structure.
250  */
alloc_flex_gd(unsigned int flexbg_size)251 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size)
252 {
253 	struct ext4_new_flex_group_data *flex_gd;
254 
255 	flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
256 	if (flex_gd == NULL)
257 		goto out3;
258 
259 	if (unlikely(flexbg_size > MAX_RESIZE_BG))
260 		flex_gd->resize_bg = MAX_RESIZE_BG;
261 	else
262 		flex_gd->resize_bg = flexbg_size;
263 
264 	flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
265 					sizeof(struct ext4_new_group_data),
266 					GFP_NOFS);
267 	if (flex_gd->groups == NULL)
268 		goto out2;
269 
270 	flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
271 					  GFP_NOFS);
272 	if (flex_gd->bg_flags == NULL)
273 		goto out1;
274 
275 	return flex_gd;
276 
277 out1:
278 	kfree(flex_gd->groups);
279 out2:
280 	kfree(flex_gd);
281 out3:
282 	return NULL;
283 }
284 
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)285 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
286 {
287 	kfree(flex_gd->bg_flags);
288 	kfree(flex_gd->groups);
289 	kfree(flex_gd);
290 }
291 
292 /*
293  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
294  * and inode tables for a flex group.
295  *
296  * This function is used by 64bit-resize.  Note that this function allocates
297  * group tables from the 1st group of groups contained by @flexgd, which may
298  * be a partial of a flex group.
299  *
300  * @sb: super block of fs to which the groups belongs
301  *
302  * Returns 0 on a successful allocation of the metadata blocks in the
303  * block group.
304  */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,unsigned int flexbg_size)305 static int ext4_alloc_group_tables(struct super_block *sb,
306 				struct ext4_new_flex_group_data *flex_gd,
307 				unsigned int flexbg_size)
308 {
309 	struct ext4_new_group_data *group_data = flex_gd->groups;
310 	ext4_fsblk_t start_blk;
311 	ext4_fsblk_t last_blk;
312 	ext4_group_t src_group;
313 	ext4_group_t bb_index = 0;
314 	ext4_group_t ib_index = 0;
315 	ext4_group_t it_index = 0;
316 	ext4_group_t group;
317 	ext4_group_t last_group;
318 	unsigned overhead;
319 	__u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
320 	int i;
321 
322 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
323 
324 	src_group = group_data[0].group;
325 	last_group  = src_group + flex_gd->count - 1;
326 
327 	BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
328 	       (last_group & ~(flexbg_size - 1))));
329 next_group:
330 	group = group_data[0].group;
331 	if (src_group >= group_data[0].group + flex_gd->count)
332 		return -ENOSPC;
333 	start_blk = ext4_group_first_block_no(sb, src_group);
334 	last_blk = start_blk + group_data[src_group - group].blocks_count;
335 
336 	overhead = ext4_group_overhead_blocks(sb, src_group);
337 
338 	start_blk += overhead;
339 
340 	/* We collect contiguous blocks as much as possible. */
341 	src_group++;
342 	for (; src_group <= last_group; src_group++) {
343 		overhead = ext4_group_overhead_blocks(sb, src_group);
344 		if (overhead == 0)
345 			last_blk += group_data[src_group - group].blocks_count;
346 		else
347 			break;
348 	}
349 
350 	/* Allocate block bitmaps */
351 	for (; bb_index < flex_gd->count; bb_index++) {
352 		if (start_blk >= last_blk)
353 			goto next_group;
354 		group_data[bb_index].block_bitmap = start_blk++;
355 		group = ext4_get_group_number(sb, start_blk - 1);
356 		group -= group_data[0].group;
357 		group_data[group].mdata_blocks++;
358 		flex_gd->bg_flags[group] &= uninit_mask;
359 	}
360 
361 	/* Allocate inode bitmaps */
362 	for (; ib_index < flex_gd->count; ib_index++) {
363 		if (start_blk >= last_blk)
364 			goto next_group;
365 		group_data[ib_index].inode_bitmap = start_blk++;
366 		group = ext4_get_group_number(sb, start_blk - 1);
367 		group -= group_data[0].group;
368 		group_data[group].mdata_blocks++;
369 		flex_gd->bg_flags[group] &= uninit_mask;
370 	}
371 
372 	/* Allocate inode tables */
373 	for (; it_index < flex_gd->count; it_index++) {
374 		unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
375 		ext4_fsblk_t next_group_start;
376 
377 		if (start_blk + itb > last_blk)
378 			goto next_group;
379 		group_data[it_index].inode_table = start_blk;
380 		group = ext4_get_group_number(sb, start_blk);
381 		next_group_start = ext4_group_first_block_no(sb, group + 1);
382 		group -= group_data[0].group;
383 
384 		if (start_blk + itb > next_group_start) {
385 			flex_gd->bg_flags[group + 1] &= uninit_mask;
386 			overhead = start_blk + itb - next_group_start;
387 			group_data[group + 1].mdata_blocks += overhead;
388 			itb -= overhead;
389 		}
390 
391 		group_data[group].mdata_blocks += itb;
392 		flex_gd->bg_flags[group] &= uninit_mask;
393 		start_blk += EXT4_SB(sb)->s_itb_per_group;
394 	}
395 
396 	/* Update free clusters count to exclude metadata blocks */
397 	for (i = 0; i < flex_gd->count; i++) {
398 		group_data[i].free_clusters_count -=
399 				EXT4_NUM_B2C(EXT4_SB(sb),
400 					     group_data[i].mdata_blocks);
401 	}
402 
403 	if (test_opt(sb, DEBUG)) {
404 		int i;
405 		group = group_data[0].group;
406 
407 		printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
408 		       "%u groups, flexbg size is %u:\n", flex_gd->count,
409 		       flexbg_size);
410 
411 		for (i = 0; i < flex_gd->count; i++) {
412 			ext4_debug(
413 			       "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
414 			       ext4_bg_has_super(sb, group + i) ? "normal" :
415 			       "no-super", group + i,
416 			       group_data[i].blocks_count,
417 			       group_data[i].free_clusters_count,
418 			       group_data[i].mdata_blocks);
419 		}
420 	}
421 	return 0;
422 }
423 
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)424 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
425 				  ext4_fsblk_t blk)
426 {
427 	struct buffer_head *bh;
428 	int err;
429 
430 	bh = sb_getblk(sb, blk);
431 	if (unlikely(!bh))
432 		return ERR_PTR(-ENOMEM);
433 	BUFFER_TRACE(bh, "get_write_access");
434 	err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
435 	if (err) {
436 		brelse(bh);
437 		bh = ERR_PTR(err);
438 	} else {
439 		memset(bh->b_data, 0, sb->s_blocksize);
440 		set_buffer_uptodate(bh);
441 	}
442 
443 	return bh;
444 }
445 
ext4_resize_ensure_credits_batch(handle_t * handle,int credits)446 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
447 {
448 	return ext4_journal_ensure_credits_fn(handle, credits,
449 		EXT4_MAX_TRANS_DATA, 0, 0);
450 }
451 
452 /*
453  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
454  *
455  * Helper function for ext4_setup_new_group_blocks() which set .
456  *
457  * @sb: super block
458  * @handle: journal handle
459  * @flex_gd: flex group data
460  */
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)461 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
462 			struct ext4_new_flex_group_data *flex_gd,
463 			ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
464 {
465 	struct ext4_sb_info *sbi = EXT4_SB(sb);
466 	ext4_group_t count = last_cluster - first_cluster + 1;
467 	ext4_group_t count2;
468 
469 	ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
470 		   last_cluster);
471 	for (count2 = count; count > 0;
472 	     count -= count2, first_cluster += count2) {
473 		ext4_fsblk_t start;
474 		struct buffer_head *bh;
475 		ext4_group_t group;
476 		int err;
477 
478 		group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
479 		start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
480 		group -= flex_gd->groups[0].group;
481 
482 		count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
483 		if (count2 > count)
484 			count2 = count;
485 
486 		if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
487 			BUG_ON(flex_gd->count > 1);
488 			continue;
489 		}
490 
491 		err = ext4_resize_ensure_credits_batch(handle, 1);
492 		if (err < 0)
493 			return err;
494 
495 		bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
496 		if (unlikely(!bh))
497 			return -ENOMEM;
498 
499 		BUFFER_TRACE(bh, "get_write_access");
500 		err = ext4_journal_get_write_access(handle, sb, bh,
501 						    EXT4_JTR_NONE);
502 		if (err) {
503 			brelse(bh);
504 			return err;
505 		}
506 		ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
507 			   first_cluster, first_cluster - start, count2);
508 		mb_set_bits(bh->b_data, first_cluster - start, count2);
509 
510 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
511 		brelse(bh);
512 		if (unlikely(err))
513 			return err;
514 	}
515 
516 	return 0;
517 }
518 
519 /*
520  * Set up the block and inode bitmaps, and the inode table for the new groups.
521  * This doesn't need to be part of the main transaction, since we are only
522  * changing blocks outside the actual filesystem.  We still do journaling to
523  * ensure the recovery is correct in case of a failure just after resize.
524  * If any part of this fails, we simply abort the resize.
525  *
526  * setup_new_flex_group_blocks handles a flex group as follow:
527  *  1. copy super block and GDT, and initialize group tables if necessary.
528  *     In this step, we only set bits in blocks bitmaps for blocks taken by
529  *     super block and GDT.
530  *  2. allocate group tables in block bitmaps, that is, set bits in block
531  *     bitmap for blocks taken by group tables.
532  */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)533 static int setup_new_flex_group_blocks(struct super_block *sb,
534 				struct ext4_new_flex_group_data *flex_gd)
535 {
536 	int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
537 	ext4_fsblk_t start;
538 	ext4_fsblk_t block;
539 	struct ext4_sb_info *sbi = EXT4_SB(sb);
540 	struct ext4_super_block *es = sbi->s_es;
541 	struct ext4_new_group_data *group_data = flex_gd->groups;
542 	__u16 *bg_flags = flex_gd->bg_flags;
543 	handle_t *handle;
544 	ext4_group_t group, count;
545 	struct buffer_head *bh = NULL;
546 	int reserved_gdb, i, j, err = 0, err2;
547 	int meta_bg;
548 
549 	BUG_ON(!flex_gd->count || !group_data ||
550 	       group_data[0].group != sbi->s_groups_count);
551 
552 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
553 	meta_bg = ext4_has_feature_meta_bg(sb);
554 
555 	/* This transaction may be extended/restarted along the way */
556 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
557 	if (IS_ERR(handle))
558 		return PTR_ERR(handle);
559 
560 	group = group_data[0].group;
561 	for (i = 0; i < flex_gd->count; i++, group++) {
562 		unsigned long gdblocks;
563 		ext4_grpblk_t overhead;
564 
565 		gdblocks = ext4_bg_num_gdb(sb, group);
566 		start = ext4_group_first_block_no(sb, group);
567 
568 		if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
569 			goto handle_itb;
570 
571 		if (meta_bg == 1)
572 			goto handle_itb;
573 
574 		block = start + ext4_bg_has_super(sb, group);
575 		/* Copy all of the GDT blocks into the backup in this group */
576 		for (j = 0; j < gdblocks; j++, block++) {
577 			struct buffer_head *gdb;
578 
579 			ext4_debug("update backup group %#04llx\n", block);
580 			err = ext4_resize_ensure_credits_batch(handle, 1);
581 			if (err < 0)
582 				goto out;
583 
584 			gdb = sb_getblk(sb, block);
585 			if (unlikely(!gdb)) {
586 				err = -ENOMEM;
587 				goto out;
588 			}
589 
590 			BUFFER_TRACE(gdb, "get_write_access");
591 			err = ext4_journal_get_write_access(handle, sb, gdb,
592 							    EXT4_JTR_NONE);
593 			if (err) {
594 				brelse(gdb);
595 				goto out;
596 			}
597 			memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
598 				s_group_desc, j)->b_data, gdb->b_size);
599 			set_buffer_uptodate(gdb);
600 
601 			err = ext4_handle_dirty_metadata(handle, NULL, gdb);
602 			if (unlikely(err)) {
603 				brelse(gdb);
604 				goto out;
605 			}
606 			brelse(gdb);
607 		}
608 
609 		/* Zero out all of the reserved backup group descriptor
610 		 * table blocks
611 		 */
612 		if (ext4_bg_has_super(sb, group)) {
613 			err = sb_issue_zeroout(sb, gdblocks + start + 1,
614 					reserved_gdb, GFP_NOFS);
615 			if (err)
616 				goto out;
617 		}
618 
619 handle_itb:
620 		/* Initialize group tables of the grop @group */
621 		if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
622 			goto handle_bb;
623 
624 		/* Zero out all of the inode table blocks */
625 		block = group_data[i].inode_table;
626 		ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
627 			   block, sbi->s_itb_per_group);
628 		err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
629 				       GFP_NOFS);
630 		if (err)
631 			goto out;
632 
633 handle_bb:
634 		if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
635 			goto handle_ib;
636 
637 		/* Initialize block bitmap of the @group */
638 		block = group_data[i].block_bitmap;
639 		err = ext4_resize_ensure_credits_batch(handle, 1);
640 		if (err < 0)
641 			goto out;
642 
643 		bh = bclean(handle, sb, block);
644 		if (IS_ERR(bh)) {
645 			err = PTR_ERR(bh);
646 			goto out;
647 		}
648 		overhead = ext4_group_overhead_blocks(sb, group);
649 		if (overhead != 0) {
650 			ext4_debug("mark backup superblock %#04llx (+0)\n",
651 				   start);
652 			mb_set_bits(bh->b_data, 0,
653 				      EXT4_NUM_B2C(sbi, overhead));
654 		}
655 		ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
656 				     sb->s_blocksize * 8, bh->b_data);
657 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
658 		brelse(bh);
659 		if (err)
660 			goto out;
661 
662 handle_ib:
663 		if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
664 			continue;
665 
666 		/* Initialize inode bitmap of the @group */
667 		block = group_data[i].inode_bitmap;
668 		err = ext4_resize_ensure_credits_batch(handle, 1);
669 		if (err < 0)
670 			goto out;
671 		/* Mark unused entries in inode bitmap used */
672 		bh = bclean(handle, sb, block);
673 		if (IS_ERR(bh)) {
674 			err = PTR_ERR(bh);
675 			goto out;
676 		}
677 
678 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
679 				     sb->s_blocksize * 8, bh->b_data);
680 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
681 		brelse(bh);
682 		if (err)
683 			goto out;
684 	}
685 
686 	/* Mark group tables in block bitmap */
687 	for (j = 0; j < GROUP_TABLE_COUNT; j++) {
688 		count = group_table_count[j];
689 		start = (&group_data[0].block_bitmap)[j];
690 		block = start;
691 		for (i = 1; i < flex_gd->count; i++) {
692 			block += group_table_count[j];
693 			if (block == (&group_data[i].block_bitmap)[j]) {
694 				count += group_table_count[j];
695 				continue;
696 			}
697 			err = set_flexbg_block_bitmap(sb, handle,
698 						      flex_gd,
699 						      EXT4_B2C(sbi, start),
700 						      EXT4_B2C(sbi,
701 							       start + count
702 							       - 1));
703 			if (err)
704 				goto out;
705 			count = group_table_count[j];
706 			start = (&group_data[i].block_bitmap)[j];
707 			block = start;
708 		}
709 
710 		if (count) {
711 			err = set_flexbg_block_bitmap(sb, handle,
712 						      flex_gd,
713 						      EXT4_B2C(sbi, start),
714 						      EXT4_B2C(sbi,
715 							       start + count
716 							       - 1));
717 			if (err)
718 				goto out;
719 		}
720 	}
721 
722 out:
723 	err2 = ext4_journal_stop(handle);
724 	if (err2 && !err)
725 		err = err2;
726 
727 	return err;
728 }
729 
730 /*
731  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
732  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
733  * calling this for the first time.  In a sparse filesystem it will be the
734  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
735  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
736  */
ext4_list_backups(struct super_block * sb,unsigned int * three,unsigned int * five,unsigned int * seven)737 unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
738 			       unsigned int *five, unsigned int *seven)
739 {
740 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
741 	unsigned int *min = three;
742 	int mult = 3;
743 	unsigned int ret;
744 
745 	if (ext4_has_feature_sparse_super2(sb)) {
746 		do {
747 			if (*min > 2)
748 				return UINT_MAX;
749 			ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
750 			*min += 1;
751 		} while (!ret);
752 		return ret;
753 	}
754 
755 	if (!ext4_has_feature_sparse_super(sb)) {
756 		ret = *min;
757 		*min += 1;
758 		return ret;
759 	}
760 
761 	if (*five < *min) {
762 		min = five;
763 		mult = 5;
764 	}
765 	if (*seven < *min) {
766 		min = seven;
767 		mult = 7;
768 	}
769 
770 	ret = *min;
771 	*min *= mult;
772 
773 	return ret;
774 }
775 
776 /*
777  * Check that all of the backup GDT blocks are held in the primary GDT block.
778  * It is assumed that they are stored in group order.  Returns the number of
779  * groups in current filesystem that have BACKUPS, or -ve error code.
780  */
verify_reserved_gdb(struct super_block * sb,ext4_group_t end,struct buffer_head * primary)781 static int verify_reserved_gdb(struct super_block *sb,
782 			       ext4_group_t end,
783 			       struct buffer_head *primary)
784 {
785 	const ext4_fsblk_t blk = primary->b_blocknr;
786 	unsigned three = 1;
787 	unsigned five = 5;
788 	unsigned seven = 7;
789 	unsigned grp;
790 	__le32 *p = (__le32 *)primary->b_data;
791 	int gdbackups = 0;
792 
793 	while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
794 		if (le32_to_cpu(*p++) !=
795 		    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
796 			ext4_warning(sb, "reserved GDT %llu"
797 				     " missing grp %d (%llu)",
798 				     blk, grp,
799 				     grp *
800 				     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
801 				     blk);
802 			return -EINVAL;
803 		}
804 		if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
805 			return -EFBIG;
806 	}
807 
808 	return gdbackups;
809 }
810 
811 /*
812  * Called when we need to bring a reserved group descriptor table block into
813  * use from the resize inode.  The primary copy of the new GDT block currently
814  * is an indirect block (under the double indirect block in the resize inode).
815  * The new backup GDT blocks will be stored as leaf blocks in this indirect
816  * block, in group order.  Even though we know all the block numbers we need,
817  * we check to ensure that the resize inode has actually reserved these blocks.
818  *
819  * Don't need to update the block bitmaps because the blocks are still in use.
820  *
821  * We get all of the error cases out of the way, so that we are sure to not
822  * fail once we start modifying the data on disk, because JBD has no rollback.
823  */
add_new_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)824 static int add_new_gdb(handle_t *handle, struct inode *inode,
825 		       ext4_group_t group)
826 {
827 	struct super_block *sb = inode->i_sb;
828 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
829 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
830 	ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
831 	struct buffer_head **o_group_desc, **n_group_desc = NULL;
832 	struct buffer_head *dind = NULL;
833 	struct buffer_head *gdb_bh = NULL;
834 	int gdbackups;
835 	struct ext4_iloc iloc = { .bh = NULL };
836 	__le32 *data;
837 	int err;
838 
839 	if (test_opt(sb, DEBUG))
840 		printk(KERN_DEBUG
841 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
842 		       gdb_num);
843 
844 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
845 	if (IS_ERR(gdb_bh))
846 		return PTR_ERR(gdb_bh);
847 
848 	gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
849 	if (gdbackups < 0) {
850 		err = gdbackups;
851 		goto errout;
852 	}
853 
854 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
855 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
856 	if (IS_ERR(dind)) {
857 		err = PTR_ERR(dind);
858 		dind = NULL;
859 		goto errout;
860 	}
861 
862 	data = (__le32 *)dind->b_data;
863 	if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
864 		ext4_warning(sb, "new group %u GDT block %llu not reserved",
865 			     group, gdblock);
866 		err = -EINVAL;
867 		goto errout;
868 	}
869 
870 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
871 	err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
872 					    EXT4_JTR_NONE);
873 	if (unlikely(err))
874 		goto errout;
875 
876 	BUFFER_TRACE(gdb_bh, "get_write_access");
877 	err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
878 	if (unlikely(err))
879 		goto errout;
880 
881 	BUFFER_TRACE(dind, "get_write_access");
882 	err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
883 	if (unlikely(err)) {
884 		ext4_std_error(sb, err);
885 		goto errout;
886 	}
887 
888 	/* ext4_reserve_inode_write() gets a reference on the iloc */
889 	err = ext4_reserve_inode_write(handle, inode, &iloc);
890 	if (unlikely(err))
891 		goto errout;
892 
893 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
894 				GFP_KERNEL);
895 	if (!n_group_desc) {
896 		err = -ENOMEM;
897 		ext4_warning(sb, "not enough memory for %lu groups",
898 			     gdb_num + 1);
899 		goto errout;
900 	}
901 
902 	/*
903 	 * Finally, we have all of the possible failures behind us...
904 	 *
905 	 * Remove new GDT block from inode double-indirect block and clear out
906 	 * the new GDT block for use (which also "frees" the backup GDT blocks
907 	 * from the reserved inode).  We don't need to change the bitmaps for
908 	 * these blocks, because they are marked as in-use from being in the
909 	 * reserved inode, and will become GDT blocks (primary and backup).
910 	 */
911 	data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
912 	err = ext4_handle_dirty_metadata(handle, NULL, dind);
913 	if (unlikely(err)) {
914 		ext4_std_error(sb, err);
915 		goto errout;
916 	}
917 	inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
918 			   (9 - EXT4_SB(sb)->s_cluster_bits);
919 	ext4_mark_iloc_dirty(handle, inode, &iloc);
920 	memset(gdb_bh->b_data, 0, sb->s_blocksize);
921 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
922 	if (unlikely(err)) {
923 		ext4_std_error(sb, err);
924 		iloc.bh = NULL;
925 		goto errout;
926 	}
927 	brelse(dind);
928 
929 	rcu_read_lock();
930 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
931 	memcpy(n_group_desc, o_group_desc,
932 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
933 	rcu_read_unlock();
934 	n_group_desc[gdb_num] = gdb_bh;
935 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
936 	EXT4_SB(sb)->s_gdb_count++;
937 	ext4_kvfree_array_rcu(o_group_desc);
938 
939 	lock_buffer(EXT4_SB(sb)->s_sbh);
940 	le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
941 	ext4_superblock_csum_set(sb);
942 	unlock_buffer(EXT4_SB(sb)->s_sbh);
943 	err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
944 	if (err)
945 		ext4_std_error(sb, err);
946 	return err;
947 errout:
948 	kvfree(n_group_desc);
949 	brelse(iloc.bh);
950 	brelse(dind);
951 	brelse(gdb_bh);
952 
953 	ext4_debug("leaving with error %d\n", err);
954 	return err;
955 }
956 
957 /*
958  * add_new_gdb_meta_bg is the sister of add_new_gdb.
959  */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)960 static int add_new_gdb_meta_bg(struct super_block *sb,
961 			       handle_t *handle, ext4_group_t group) {
962 	ext4_fsblk_t gdblock;
963 	struct buffer_head *gdb_bh;
964 	struct buffer_head **o_group_desc, **n_group_desc;
965 	unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
966 	int err;
967 
968 	gdblock = ext4_meta_bg_first_block_no(sb, group) +
969 		   ext4_bg_has_super(sb, group);
970 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
971 	if (IS_ERR(gdb_bh))
972 		return PTR_ERR(gdb_bh);
973 	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
974 				GFP_KERNEL);
975 	if (!n_group_desc) {
976 		brelse(gdb_bh);
977 		err = -ENOMEM;
978 		ext4_warning(sb, "not enough memory for %lu groups",
979 			     gdb_num + 1);
980 		return err;
981 	}
982 
983 	rcu_read_lock();
984 	o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
985 	memcpy(n_group_desc, o_group_desc,
986 	       EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
987 	rcu_read_unlock();
988 	n_group_desc[gdb_num] = gdb_bh;
989 
990 	BUFFER_TRACE(gdb_bh, "get_write_access");
991 	err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
992 	if (err) {
993 		kvfree(n_group_desc);
994 		brelse(gdb_bh);
995 		return err;
996 	}
997 
998 	rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
999 	EXT4_SB(sb)->s_gdb_count++;
1000 	ext4_kvfree_array_rcu(o_group_desc);
1001 	return err;
1002 }
1003 
1004 /*
1005  * Called when we are adding a new group which has a backup copy of each of
1006  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
1007  * We need to add these reserved backup GDT blocks to the resize inode, so
1008  * that they are kept for future resizing and not allocated to files.
1009  *
1010  * Each reserved backup GDT block will go into a different indirect block.
1011  * The indirect blocks are actually the primary reserved GDT blocks,
1012  * so we know in advance what their block numbers are.  We only get the
1013  * double-indirect block to verify it is pointing to the primary reserved
1014  * GDT blocks so we don't overwrite a data block by accident.  The reserved
1015  * backup GDT blocks are stored in their reserved primary GDT block.
1016  */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)1017 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1018 			      ext4_group_t group)
1019 {
1020 	struct super_block *sb = inode->i_sb;
1021 	int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1022 	int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1023 	struct buffer_head **primary;
1024 	struct buffer_head *dind;
1025 	struct ext4_iloc iloc;
1026 	ext4_fsblk_t blk;
1027 	__le32 *data, *end;
1028 	int gdbackups = 0;
1029 	int res, i;
1030 	int err;
1031 
1032 	primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1033 	if (!primary)
1034 		return -ENOMEM;
1035 
1036 	data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1037 	dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1038 	if (IS_ERR(dind)) {
1039 		err = PTR_ERR(dind);
1040 		dind = NULL;
1041 		goto exit_free;
1042 	}
1043 
1044 	blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1045 	data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1046 					 EXT4_ADDR_PER_BLOCK(sb));
1047 	end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1048 
1049 	/* Get each reserved primary GDT block and verify it holds backups */
1050 	for (res = 0; res < reserved_gdb; res++, blk++) {
1051 		if (le32_to_cpu(*data) != blk) {
1052 			ext4_warning(sb, "reserved block %llu"
1053 				     " not at offset %ld",
1054 				     blk,
1055 				     (long)(data - (__le32 *)dind->b_data));
1056 			err = -EINVAL;
1057 			goto exit_bh;
1058 		}
1059 		primary[res] = ext4_sb_bread(sb, blk, 0);
1060 		if (IS_ERR(primary[res])) {
1061 			err = PTR_ERR(primary[res]);
1062 			primary[res] = NULL;
1063 			goto exit_bh;
1064 		}
1065 		gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1066 		if (gdbackups < 0) {
1067 			brelse(primary[res]);
1068 			err = gdbackups;
1069 			goto exit_bh;
1070 		}
1071 		if (++data >= end)
1072 			data = (__le32 *)dind->b_data;
1073 	}
1074 
1075 	for (i = 0; i < reserved_gdb; i++) {
1076 		BUFFER_TRACE(primary[i], "get_write_access");
1077 		if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1078 							 EXT4_JTR_NONE)))
1079 			goto exit_bh;
1080 	}
1081 
1082 	if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1083 		goto exit_bh;
1084 
1085 	/*
1086 	 * Finally we can add each of the reserved backup GDT blocks from
1087 	 * the new group to its reserved primary GDT block.
1088 	 */
1089 	blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1090 	for (i = 0; i < reserved_gdb; i++) {
1091 		int err2;
1092 		data = (__le32 *)primary[i]->b_data;
1093 		/* printk("reserving backup %lu[%u] = %lu\n",
1094 		       primary[i]->b_blocknr, gdbackups,
1095 		       blk + primary[i]->b_blocknr); */
1096 		data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1097 		err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1098 		if (!err)
1099 			err = err2;
1100 	}
1101 
1102 	inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1103 	ext4_mark_iloc_dirty(handle, inode, &iloc);
1104 
1105 exit_bh:
1106 	while (--res >= 0)
1107 		brelse(primary[res]);
1108 	brelse(dind);
1109 
1110 exit_free:
1111 	kfree(primary);
1112 
1113 	return err;
1114 }
1115 
ext4_set_block_group_nr(struct super_block * sb,char * data,ext4_group_t group)1116 static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1117 					   ext4_group_t group)
1118 {
1119 	struct ext4_super_block *es = (struct ext4_super_block *) data;
1120 
1121 	es->s_block_group_nr = cpu_to_le16(group);
1122 	if (ext4_has_metadata_csum(sb))
1123 		es->s_checksum = ext4_superblock_csum(sb, es);
1124 }
1125 
1126 /*
1127  * Update the backup copies of the ext4 metadata.  These don't need to be part
1128  * of the main resize transaction, because e2fsck will re-write them if there
1129  * is a problem (basically only OOM will cause a problem).  However, we
1130  * _should_ update the backups if possible, in case the primary gets trashed
1131  * for some reason and we need to run e2fsck from a backup superblock.  The
1132  * important part is that the new block and inode counts are in the backup
1133  * superblocks, and the location of the new group metadata in the GDT backups.
1134  *
1135  * We do not need take the s_resize_lock for this, because these
1136  * blocks are not otherwise touched by the filesystem code when it is
1137  * mounted.  We don't need to worry about last changing from
1138  * sbi->s_groups_count, because the worst that can happen is that we
1139  * do not copy the full number of backups at this time.  The resize
1140  * which changed s_groups_count will backup again.
1141  */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1142 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1143 			   int size, int meta_bg)
1144 {
1145 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1146 	ext4_group_t last;
1147 	const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1148 	unsigned three = 1;
1149 	unsigned five = 5;
1150 	unsigned seven = 7;
1151 	ext4_group_t group = 0;
1152 	int rest = sb->s_blocksize - size;
1153 	handle_t *handle;
1154 	int err = 0, err2;
1155 
1156 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1157 	if (IS_ERR(handle)) {
1158 		group = 1;
1159 		err = PTR_ERR(handle);
1160 		goto exit_err;
1161 	}
1162 
1163 	if (meta_bg == 0) {
1164 		group = ext4_list_backups(sb, &three, &five, &seven);
1165 		last = sbi->s_groups_count;
1166 	} else {
1167 		group = ext4_get_group_number(sb, blk_off) + 1;
1168 		last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1169 	}
1170 
1171 	while (group < sbi->s_groups_count) {
1172 		struct buffer_head *bh;
1173 		ext4_fsblk_t backup_block;
1174 		int has_super = ext4_bg_has_super(sb, group);
1175 		ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1176 
1177 		/* Out of journal space, and can't get more - abort - so sad */
1178 		err = ext4_resize_ensure_credits_batch(handle, 1);
1179 		if (err < 0)
1180 			break;
1181 
1182 		if (meta_bg == 0)
1183 			backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1184 		else
1185 			backup_block = first_block + has_super;
1186 
1187 		bh = sb_getblk(sb, backup_block);
1188 		if (unlikely(!bh)) {
1189 			err = -ENOMEM;
1190 			break;
1191 		}
1192 		ext4_debug("update metadata backup %llu(+%llu)\n",
1193 			   backup_block, backup_block -
1194 			   ext4_group_first_block_no(sb, group));
1195 		BUFFER_TRACE(bh, "get_write_access");
1196 		if ((err = ext4_journal_get_write_access(handle, sb, bh,
1197 							 EXT4_JTR_NONE))) {
1198 			brelse(bh);
1199 			break;
1200 		}
1201 		lock_buffer(bh);
1202 		memcpy(bh->b_data, data, size);
1203 		if (rest)
1204 			memset(bh->b_data + size, 0, rest);
1205 		if (has_super && (backup_block == first_block))
1206 			ext4_set_block_group_nr(sb, bh->b_data, group);
1207 		set_buffer_uptodate(bh);
1208 		unlock_buffer(bh);
1209 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1210 		if (unlikely(err))
1211 			ext4_std_error(sb, err);
1212 		brelse(bh);
1213 
1214 		if (meta_bg == 0)
1215 			group = ext4_list_backups(sb, &three, &five, &seven);
1216 		else if (group == last)
1217 			break;
1218 		else
1219 			group = last;
1220 	}
1221 	if ((err2 = ext4_journal_stop(handle)) && !err)
1222 		err = err2;
1223 
1224 	/*
1225 	 * Ugh! Need to have e2fsck write the backup copies.  It is too
1226 	 * late to revert the resize, we shouldn't fail just because of
1227 	 * the backup copies (they are only needed in case of corruption).
1228 	 *
1229 	 * However, if we got here we have a journal problem too, so we
1230 	 * can't really start a transaction to mark the superblock.
1231 	 * Chicken out and just set the flag on the hope it will be written
1232 	 * to disk, and if not - we will simply wait until next fsck.
1233 	 */
1234 exit_err:
1235 	if (err) {
1236 		ext4_warning(sb, "can't update backup for group %u (err %d), "
1237 			     "forcing fsck on next reboot", group, err);
1238 		sbi->s_mount_state &= ~EXT4_VALID_FS;
1239 		sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1240 		mark_buffer_dirty(sbi->s_sbh);
1241 	}
1242 }
1243 
1244 /*
1245  * ext4_add_new_descs() adds @count group descriptor of groups
1246  * starting at @group
1247  *
1248  * @handle: journal handle
1249  * @sb: super block
1250  * @group: the group no. of the first group desc to be added
1251  * @resize_inode: the resize inode
1252  * @count: number of group descriptors to be added
1253  */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1254 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1255 			      ext4_group_t group, struct inode *resize_inode,
1256 			      ext4_group_t count)
1257 {
1258 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1259 	struct ext4_super_block *es = sbi->s_es;
1260 	struct buffer_head *gdb_bh;
1261 	int i, gdb_off, gdb_num, err = 0;
1262 	int meta_bg;
1263 
1264 	meta_bg = ext4_has_feature_meta_bg(sb);
1265 	for (i = 0; i < count; i++, group++) {
1266 		int reserved_gdb = ext4_bg_has_super(sb, group) ?
1267 			le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1268 
1269 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1270 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1271 
1272 		/*
1273 		 * We will only either add reserved group blocks to a backup group
1274 		 * or remove reserved blocks for the first group in a new group block.
1275 		 * Doing both would be mean more complex code, and sane people don't
1276 		 * use non-sparse filesystems anymore.  This is already checked above.
1277 		 */
1278 		if (gdb_off) {
1279 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1280 						     gdb_num);
1281 			BUFFER_TRACE(gdb_bh, "get_write_access");
1282 			err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1283 							    EXT4_JTR_NONE);
1284 
1285 			if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1286 				err = reserve_backup_gdb(handle, resize_inode, group);
1287 		} else if (meta_bg != 0) {
1288 			err = add_new_gdb_meta_bg(sb, handle, group);
1289 		} else {
1290 			err = add_new_gdb(handle, resize_inode, group);
1291 		}
1292 		if (err)
1293 			break;
1294 	}
1295 	return err;
1296 }
1297 
ext4_get_bitmap(struct super_block * sb,__u64 block)1298 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1299 {
1300 	struct buffer_head *bh = sb_getblk(sb, block);
1301 	if (unlikely(!bh))
1302 		return NULL;
1303 	if (!bh_uptodate_or_lock(bh)) {
1304 		if (ext4_read_bh(bh, 0, NULL) < 0) {
1305 			brelse(bh);
1306 			return NULL;
1307 		}
1308 	}
1309 
1310 	return bh;
1311 }
1312 
ext4_set_bitmap_checksums(struct super_block * sb,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1313 static int ext4_set_bitmap_checksums(struct super_block *sb,
1314 				     struct ext4_group_desc *gdp,
1315 				     struct ext4_new_group_data *group_data)
1316 {
1317 	struct buffer_head *bh;
1318 
1319 	if (!ext4_has_metadata_csum(sb))
1320 		return 0;
1321 
1322 	bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1323 	if (!bh)
1324 		return -EIO;
1325 	ext4_inode_bitmap_csum_set(sb, gdp, bh,
1326 				   EXT4_INODES_PER_GROUP(sb) / 8);
1327 	brelse(bh);
1328 
1329 	bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1330 	if (!bh)
1331 		return -EIO;
1332 	ext4_block_bitmap_csum_set(sb, gdp, bh);
1333 	brelse(bh);
1334 
1335 	return 0;
1336 }
1337 
1338 /*
1339  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1340  */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1341 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1342 				struct ext4_new_flex_group_data *flex_gd)
1343 {
1344 	struct ext4_new_group_data	*group_data = flex_gd->groups;
1345 	struct ext4_group_desc		*gdp;
1346 	struct ext4_sb_info		*sbi = EXT4_SB(sb);
1347 	struct buffer_head		*gdb_bh;
1348 	ext4_group_t			group;
1349 	__u16				*bg_flags = flex_gd->bg_flags;
1350 	int				i, gdb_off, gdb_num, err = 0;
1351 
1352 
1353 	for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1354 		group = group_data->group;
1355 
1356 		gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1357 		gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1358 
1359 		/*
1360 		 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1361 		 */
1362 		gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1363 		/* Update group descriptor block for new group */
1364 		gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1365 						 gdb_off * EXT4_DESC_SIZE(sb));
1366 
1367 		memset(gdp, 0, EXT4_DESC_SIZE(sb));
1368 		ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1369 		ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1370 		err = ext4_set_bitmap_checksums(sb, gdp, group_data);
1371 		if (err) {
1372 			ext4_std_error(sb, err);
1373 			break;
1374 		}
1375 
1376 		ext4_inode_table_set(sb, gdp, group_data->inode_table);
1377 		ext4_free_group_clusters_set(sb, gdp,
1378 					     group_data->free_clusters_count);
1379 		ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1380 		if (ext4_has_group_desc_csum(sb))
1381 			ext4_itable_unused_set(sb, gdp,
1382 					       EXT4_INODES_PER_GROUP(sb));
1383 		gdp->bg_flags = cpu_to_le16(*bg_flags);
1384 		ext4_group_desc_csum_set(sb, group, gdp);
1385 
1386 		err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1387 		if (unlikely(err)) {
1388 			ext4_std_error(sb, err);
1389 			break;
1390 		}
1391 
1392 		/*
1393 		 * We can allocate memory for mb_alloc based on the new group
1394 		 * descriptor
1395 		 */
1396 		err = ext4_mb_add_groupinfo(sb, group, gdp);
1397 		if (err)
1398 			break;
1399 	}
1400 	return err;
1401 }
1402 
ext4_add_overhead(struct super_block * sb,const ext4_fsblk_t overhead)1403 static void ext4_add_overhead(struct super_block *sb,
1404                               const ext4_fsblk_t overhead)
1405 {
1406        struct ext4_sb_info *sbi = EXT4_SB(sb);
1407        struct ext4_super_block *es = sbi->s_es;
1408 
1409        sbi->s_overhead += overhead;
1410        es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1411        smp_wmb();
1412 }
1413 
1414 /*
1415  * ext4_update_super() updates the super block so that the newly added
1416  * groups can be seen by the filesystem.
1417  *
1418  * @sb: super block
1419  * @flex_gd: new added groups
1420  */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1421 static void ext4_update_super(struct super_block *sb,
1422 			     struct ext4_new_flex_group_data *flex_gd)
1423 {
1424 	ext4_fsblk_t blocks_count = 0;
1425 	ext4_fsblk_t free_blocks = 0;
1426 	ext4_fsblk_t reserved_blocks = 0;
1427 	struct ext4_new_group_data *group_data = flex_gd->groups;
1428 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1429 	struct ext4_super_block *es = sbi->s_es;
1430 	int i;
1431 
1432 	BUG_ON(flex_gd->count == 0 || group_data == NULL);
1433 	/*
1434 	 * Make the new blocks and inodes valid next.  We do this before
1435 	 * increasing the group count so that once the group is enabled,
1436 	 * all of its blocks and inodes are already valid.
1437 	 *
1438 	 * We always allocate group-by-group, then block-by-block or
1439 	 * inode-by-inode within a group, so enabling these
1440 	 * blocks/inodes before the group is live won't actually let us
1441 	 * allocate the new space yet.
1442 	 */
1443 	for (i = 0; i < flex_gd->count; i++) {
1444 		blocks_count += group_data[i].blocks_count;
1445 		free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1446 	}
1447 
1448 	reserved_blocks = ext4_r_blocks_count(es) * 100;
1449 	reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1450 	reserved_blocks *= blocks_count;
1451 	do_div(reserved_blocks, 100);
1452 
1453 	lock_buffer(sbi->s_sbh);
1454 	ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1455 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1456 	le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1457 		     flex_gd->count);
1458 	le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1459 		     flex_gd->count);
1460 
1461 	ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1462 	/*
1463 	 * We need to protect s_groups_count against other CPUs seeing
1464 	 * inconsistent state in the superblock.
1465 	 *
1466 	 * The precise rules we use are:
1467 	 *
1468 	 * * Writers must perform a smp_wmb() after updating all
1469 	 *   dependent data and before modifying the groups count
1470 	 *
1471 	 * * Readers must perform an smp_rmb() after reading the groups
1472 	 *   count and before reading any dependent data.
1473 	 *
1474 	 * NB. These rules can be relaxed when checking the group count
1475 	 * while freeing data, as we can only allocate from a block
1476 	 * group after serialising against the group count, and we can
1477 	 * only then free after serialising in turn against that
1478 	 * allocation.
1479 	 */
1480 	smp_wmb();
1481 
1482 	/* Update the global fs size fields */
1483 	sbi->s_groups_count += flex_gd->count;
1484 	sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1485 			(EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1486 
1487 	/* Update the reserved block counts only once the new group is
1488 	 * active. */
1489 	ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1490 				reserved_blocks);
1491 
1492 	/* Update the free space counts */
1493 	percpu_counter_add(&sbi->s_freeclusters_counter,
1494 			   EXT4_NUM_B2C(sbi, free_blocks));
1495 	percpu_counter_add(&sbi->s_freeinodes_counter,
1496 			   EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1497 
1498 	ext4_debug("free blocks count %llu",
1499 		   percpu_counter_read(&sbi->s_freeclusters_counter));
1500 	if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1501 		ext4_group_t flex_group;
1502 		struct flex_groups *fg;
1503 
1504 		flex_group = ext4_flex_group(sbi, group_data[0].group);
1505 		fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1506 		atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1507 			     &fg->free_clusters);
1508 		atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1509 			   &fg->free_inodes);
1510 	}
1511 
1512 	/*
1513 	 * Update the fs overhead information.
1514 	 *
1515 	 * For bigalloc, if the superblock already has a properly calculated
1516 	 * overhead, update it with a value based on numbers already computed
1517 	 * above for the newly allocated capacity.
1518 	 */
1519 	if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1520 		ext4_add_overhead(sb,
1521 			EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1522 	else
1523 		ext4_calculate_overhead(sb);
1524 	es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1525 
1526 	ext4_superblock_csum_set(sb);
1527 	unlock_buffer(sbi->s_sbh);
1528 	if (test_opt(sb, DEBUG))
1529 		printk(KERN_DEBUG "EXT4-fs: added group %u:"
1530 		       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1531 		       blocks_count, free_blocks, reserved_blocks);
1532 }
1533 
1534 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1535  * _before_ we start modifying the filesystem, because we cannot abort the
1536  * transaction and not have it write the data to disk.
1537  */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1538 static int ext4_flex_group_add(struct super_block *sb,
1539 			       struct inode *resize_inode,
1540 			       struct ext4_new_flex_group_data *flex_gd)
1541 {
1542 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1543 	struct ext4_super_block *es = sbi->s_es;
1544 	ext4_fsblk_t o_blocks_count;
1545 	ext4_grpblk_t last;
1546 	ext4_group_t group;
1547 	handle_t *handle;
1548 	unsigned reserved_gdb;
1549 	int err = 0, err2 = 0, credit;
1550 
1551 	BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1552 
1553 	reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1554 	o_blocks_count = ext4_blocks_count(es);
1555 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1556 	BUG_ON(last);
1557 
1558 	err = setup_new_flex_group_blocks(sb, flex_gd);
1559 	if (err)
1560 		goto exit;
1561 	/*
1562 	 * We will always be modifying at least the superblock and  GDT
1563 	 * blocks.  If we are adding a group past the last current GDT block,
1564 	 * we will also modify the inode and the dindirect block.  If we
1565 	 * are adding a group with superblock/GDT backups  we will also
1566 	 * modify each of the reserved GDT dindirect blocks.
1567 	 */
1568 	credit = 3;	/* sb, resize inode, resize inode dindirect */
1569 	/* GDT blocks */
1570 	credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1571 	credit += reserved_gdb;	/* Reserved GDT dindirect blocks */
1572 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1573 	if (IS_ERR(handle)) {
1574 		err = PTR_ERR(handle);
1575 		goto exit;
1576 	}
1577 
1578 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1579 	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1580 					    EXT4_JTR_NONE);
1581 	if (err)
1582 		goto exit_journal;
1583 
1584 	group = flex_gd->groups[0].group;
1585 	BUG_ON(group != sbi->s_groups_count);
1586 	err = ext4_add_new_descs(handle, sb, group,
1587 				resize_inode, flex_gd->count);
1588 	if (err)
1589 		goto exit_journal;
1590 
1591 	err = ext4_setup_new_descs(handle, sb, flex_gd);
1592 	if (err)
1593 		goto exit_journal;
1594 
1595 	ext4_update_super(sb, flex_gd);
1596 
1597 	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1598 
1599 exit_journal:
1600 	err2 = ext4_journal_stop(handle);
1601 	if (!err)
1602 		err = err2;
1603 
1604 	if (!err) {
1605 		int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1606 		int gdb_num_end = ((group + flex_gd->count - 1) /
1607 				   EXT4_DESC_PER_BLOCK(sb));
1608 		int meta_bg = ext4_has_feature_meta_bg(sb) &&
1609 			      gdb_num >= le32_to_cpu(es->s_first_meta_bg);
1610 		sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1611 					 ext4_group_first_block_no(sb, 0);
1612 		sector_t old_gdb = 0;
1613 
1614 		update_backups(sb, ext4_group_first_block_no(sb, 0),
1615 			       (char *)es, sizeof(struct ext4_super_block), 0);
1616 		for (; gdb_num <= gdb_num_end; gdb_num++) {
1617 			struct buffer_head *gdb_bh;
1618 
1619 			gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1620 						     gdb_num);
1621 			if (old_gdb == gdb_bh->b_blocknr)
1622 				continue;
1623 			update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1624 				       gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1625 			old_gdb = gdb_bh->b_blocknr;
1626 		}
1627 	}
1628 exit:
1629 	return err;
1630 }
1631 
ext4_setup_next_flex_gd(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t n_blocks_count)1632 static int ext4_setup_next_flex_gd(struct super_block *sb,
1633 				    struct ext4_new_flex_group_data *flex_gd,
1634 				    ext4_fsblk_t n_blocks_count)
1635 {
1636 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1637 	struct ext4_super_block *es = sbi->s_es;
1638 	struct ext4_new_group_data *group_data = flex_gd->groups;
1639 	ext4_fsblk_t o_blocks_count;
1640 	ext4_group_t n_group;
1641 	ext4_group_t group;
1642 	ext4_group_t last_group;
1643 	ext4_grpblk_t last;
1644 	ext4_grpblk_t clusters_per_group;
1645 	unsigned long i;
1646 
1647 	clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1648 
1649 	o_blocks_count = ext4_blocks_count(es);
1650 
1651 	if (o_blocks_count == n_blocks_count)
1652 		return 0;
1653 
1654 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1655 	BUG_ON(last);
1656 	ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1657 
1658 	last_group = group | (flex_gd->resize_bg - 1);
1659 	if (last_group > n_group)
1660 		last_group = n_group;
1661 
1662 	flex_gd->count = last_group - group + 1;
1663 
1664 	for (i = 0; i < flex_gd->count; i++) {
1665 		int overhead;
1666 
1667 		group_data[i].group = group + i;
1668 		group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1669 		overhead = ext4_group_overhead_blocks(sb, group + i);
1670 		group_data[i].mdata_blocks = overhead;
1671 		group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1672 		if (ext4_has_group_desc_csum(sb)) {
1673 			flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1674 					       EXT4_BG_INODE_UNINIT;
1675 			if (!test_opt(sb, INIT_INODE_TABLE))
1676 				flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1677 		} else
1678 			flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1679 	}
1680 
1681 	if (last_group == n_group && ext4_has_group_desc_csum(sb))
1682 		/* We need to initialize block bitmap of last group. */
1683 		flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1684 
1685 	if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1686 		group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1687 		group_data[i - 1].free_clusters_count -= clusters_per_group -
1688 						       last - 1;
1689 	}
1690 
1691 	return 1;
1692 }
1693 
1694 /* Add group descriptor data to an existing or new group descriptor block.
1695  * Ensure we handle all possible error conditions _before_ we start modifying
1696  * the filesystem, because we cannot abort the transaction and not have it
1697  * write the data to disk.
1698  *
1699  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1700  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1701  *
1702  * We only need to hold the superblock lock while we are actually adding
1703  * in the new group's counts to the superblock.  Prior to that we have
1704  * not really "added" the group at all.  We re-check that we are still
1705  * adding in the last group in case things have changed since verifying.
1706  */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1707 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1708 {
1709 	struct ext4_new_flex_group_data flex_gd;
1710 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1711 	struct ext4_super_block *es = sbi->s_es;
1712 	int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1713 		le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1714 	struct inode *inode = NULL;
1715 	int gdb_off;
1716 	int err;
1717 	__u16 bg_flags = 0;
1718 
1719 	gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1720 
1721 	if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1722 		ext4_warning(sb, "Can't resize non-sparse filesystem further");
1723 		return -EPERM;
1724 	}
1725 
1726 	if (ext4_blocks_count(es) + input->blocks_count <
1727 	    ext4_blocks_count(es)) {
1728 		ext4_warning(sb, "blocks_count overflow");
1729 		return -EINVAL;
1730 	}
1731 
1732 	if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1733 	    le32_to_cpu(es->s_inodes_count)) {
1734 		ext4_warning(sb, "inodes_count overflow");
1735 		return -EINVAL;
1736 	}
1737 
1738 	if (reserved_gdb || gdb_off == 0) {
1739 		if (!ext4_has_feature_resize_inode(sb) ||
1740 		    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1741 			ext4_warning(sb,
1742 				     "No reserved GDT blocks, can't resize");
1743 			return -EPERM;
1744 		}
1745 		inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1746 		if (IS_ERR(inode)) {
1747 			ext4_warning(sb, "Error opening resize inode");
1748 			return PTR_ERR(inode);
1749 		}
1750 	}
1751 
1752 
1753 	err = verify_group_input(sb, input);
1754 	if (err)
1755 		goto out;
1756 
1757 	err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1758 	if (err)
1759 		goto out;
1760 
1761 	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1762 	if (err)
1763 		goto out;
1764 
1765 	flex_gd.count = 1;
1766 	flex_gd.groups = input;
1767 	flex_gd.bg_flags = &bg_flags;
1768 	err = ext4_flex_group_add(sb, inode, &flex_gd);
1769 out:
1770 	iput(inode);
1771 	return err;
1772 } /* ext4_group_add */
1773 
1774 /*
1775  * extend a group without checking assuming that checking has been done.
1776  */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1777 static int ext4_group_extend_no_check(struct super_block *sb,
1778 				      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1779 {
1780 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1781 	handle_t *handle;
1782 	int err = 0, err2;
1783 
1784 	/* We will update the superblock, one block bitmap, and
1785 	 * one group descriptor via ext4_group_add_blocks().
1786 	 */
1787 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1788 	if (IS_ERR(handle)) {
1789 		err = PTR_ERR(handle);
1790 		ext4_warning(sb, "error %d on journal start", err);
1791 		return err;
1792 	}
1793 
1794 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1795 	err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1796 					    EXT4_JTR_NONE);
1797 	if (err) {
1798 		ext4_warning(sb, "error %d on journal write access", err);
1799 		goto errout;
1800 	}
1801 
1802 	lock_buffer(EXT4_SB(sb)->s_sbh);
1803 	ext4_blocks_count_set(es, o_blocks_count + add);
1804 	ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1805 	ext4_superblock_csum_set(sb);
1806 	unlock_buffer(EXT4_SB(sb)->s_sbh);
1807 	ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1808 		   o_blocks_count + add);
1809 	/* We add the blocks to the bitmap and set the group need init bit */
1810 	err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1811 	if (err)
1812 		goto errout;
1813 	ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1814 	ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1815 		   o_blocks_count + add);
1816 errout:
1817 	err2 = ext4_journal_stop(handle);
1818 	if (err2 && !err)
1819 		err = err2;
1820 
1821 	if (!err) {
1822 		if (test_opt(sb, DEBUG))
1823 			printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1824 			       "blocks\n", ext4_blocks_count(es));
1825 		update_backups(sb, ext4_group_first_block_no(sb, 0),
1826 			       (char *)es, sizeof(struct ext4_super_block), 0);
1827 	}
1828 	return err;
1829 }
1830 
1831 /*
1832  * Extend the filesystem to the new number of blocks specified.  This entry
1833  * point is only used to extend the current filesystem to the end of the last
1834  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1835  * for emergencies (because it has no dependencies on reserved blocks).
1836  *
1837  * If we _really_ wanted, we could use default values to call ext4_group_add()
1838  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1839  * GDT blocks are reserved to grow to the desired size.
1840  */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1841 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1842 		      ext4_fsblk_t n_blocks_count)
1843 {
1844 	ext4_fsblk_t o_blocks_count;
1845 	ext4_grpblk_t last;
1846 	ext4_grpblk_t add;
1847 	struct buffer_head *bh;
1848 	ext4_group_t group;
1849 
1850 	o_blocks_count = ext4_blocks_count(es);
1851 
1852 	if (test_opt(sb, DEBUG))
1853 		ext4_msg(sb, KERN_DEBUG,
1854 			 "extending last group from %llu to %llu blocks",
1855 			 o_blocks_count, n_blocks_count);
1856 
1857 	if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1858 		return 0;
1859 
1860 	if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1861 		ext4_msg(sb, KERN_ERR,
1862 			 "filesystem too large to resize to %llu blocks safely",
1863 			 n_blocks_count);
1864 		return -EINVAL;
1865 	}
1866 
1867 	if (n_blocks_count < o_blocks_count) {
1868 		ext4_warning(sb, "can't shrink FS - resize aborted");
1869 		return -EINVAL;
1870 	}
1871 
1872 	/* Handle the remaining blocks in the last group only. */
1873 	ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1874 
1875 	if (last == 0) {
1876 		ext4_warning(sb, "need to use ext2online to resize further");
1877 		return -EPERM;
1878 	}
1879 
1880 	add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1881 
1882 	if (o_blocks_count + add < o_blocks_count) {
1883 		ext4_warning(sb, "blocks_count overflow");
1884 		return -EINVAL;
1885 	}
1886 
1887 	if (o_blocks_count + add > n_blocks_count)
1888 		add = n_blocks_count - o_blocks_count;
1889 
1890 	if (o_blocks_count + add < n_blocks_count)
1891 		ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1892 			     o_blocks_count + add, add);
1893 
1894 	/* See if the device is actually as big as what was requested */
1895 	bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1896 	if (IS_ERR(bh)) {
1897 		ext4_warning(sb, "can't read last block, resize aborted");
1898 		return -ENOSPC;
1899 	}
1900 	brelse(bh);
1901 
1902 	return ext4_group_extend_no_check(sb, o_blocks_count, add);
1903 } /* ext4_group_extend */
1904 
1905 
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1906 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1907 {
1908 	return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1909 }
1910 
1911 /*
1912  * Release the resize inode and drop the resize_inode feature if there
1913  * are no more reserved gdt blocks, and then convert the file system
1914  * to enable meta_bg
1915  */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1916 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1917 {
1918 	handle_t *handle;
1919 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1920 	struct ext4_super_block *es = sbi->s_es;
1921 	struct ext4_inode_info *ei = EXT4_I(inode);
1922 	ext4_fsblk_t nr;
1923 	int i, ret, err = 0;
1924 	int credits = 1;
1925 
1926 	ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1927 	if (inode) {
1928 		if (es->s_reserved_gdt_blocks) {
1929 			ext4_error(sb, "Unexpected non-zero "
1930 				   "s_reserved_gdt_blocks");
1931 			return -EPERM;
1932 		}
1933 
1934 		/* Do a quick sanity check of the resize inode */
1935 		if (inode->i_blocks != 1 << (inode->i_blkbits -
1936 					     (9 - sbi->s_cluster_bits)))
1937 			goto invalid_resize_inode;
1938 		for (i = 0; i < EXT4_N_BLOCKS; i++) {
1939 			if (i == EXT4_DIND_BLOCK) {
1940 				if (ei->i_data[i])
1941 					continue;
1942 				else
1943 					goto invalid_resize_inode;
1944 			}
1945 			if (ei->i_data[i])
1946 				goto invalid_resize_inode;
1947 		}
1948 		credits += 3;	/* block bitmap, bg descriptor, resize inode */
1949 	}
1950 
1951 	handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1952 	if (IS_ERR(handle))
1953 		return PTR_ERR(handle);
1954 
1955 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1956 	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1957 					    EXT4_JTR_NONE);
1958 	if (err)
1959 		goto errout;
1960 
1961 	lock_buffer(sbi->s_sbh);
1962 	ext4_clear_feature_resize_inode(sb);
1963 	ext4_set_feature_meta_bg(sb);
1964 	sbi->s_es->s_first_meta_bg =
1965 		cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1966 	ext4_superblock_csum_set(sb);
1967 	unlock_buffer(sbi->s_sbh);
1968 
1969 	err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1970 	if (err) {
1971 		ext4_std_error(sb, err);
1972 		goto errout;
1973 	}
1974 
1975 	if (inode) {
1976 		nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1977 		ext4_free_blocks(handle, inode, NULL, nr, 1,
1978 				 EXT4_FREE_BLOCKS_METADATA |
1979 				 EXT4_FREE_BLOCKS_FORGET);
1980 		ei->i_data[EXT4_DIND_BLOCK] = 0;
1981 		inode->i_blocks = 0;
1982 
1983 		err = ext4_mark_inode_dirty(handle, inode);
1984 		if (err)
1985 			ext4_std_error(sb, err);
1986 	}
1987 
1988 errout:
1989 	ret = ext4_journal_stop(handle);
1990 	return err ? err : ret;
1991 
1992 invalid_resize_inode:
1993 	ext4_error(sb, "corrupted/inconsistent resize inode");
1994 	return -EINVAL;
1995 }
1996 
1997 /*
1998  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1999  *
2000  * @sb: super block of the fs to be resized
2001  * @n_blocks_count: the number of blocks resides in the resized fs
2002  */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)2003 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
2004 {
2005 	struct ext4_new_flex_group_data *flex_gd = NULL;
2006 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2007 	struct ext4_super_block *es = sbi->s_es;
2008 	struct buffer_head *bh;
2009 	struct inode *resize_inode = NULL;
2010 	ext4_grpblk_t add, offset;
2011 	unsigned long n_desc_blocks;
2012 	unsigned long o_desc_blocks;
2013 	ext4_group_t o_group;
2014 	ext4_group_t n_group;
2015 	ext4_fsblk_t o_blocks_count;
2016 	ext4_fsblk_t n_blocks_count_retry = 0;
2017 	unsigned long last_update_time = 0;
2018 	int err = 0;
2019 	int meta_bg;
2020 	unsigned int flexbg_size = ext4_flex_bg_size(sbi);
2021 
2022 	/* See if the device is actually as big as what was requested */
2023 	bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2024 	if (IS_ERR(bh)) {
2025 		ext4_warning(sb, "can't read last block, resize aborted");
2026 		return -ENOSPC;
2027 	}
2028 	brelse(bh);
2029 
2030 	/*
2031 	 * For bigalloc, trim the requested size to the nearest cluster
2032 	 * boundary to avoid creating an unusable filesystem. We do this
2033 	 * silently, instead of returning an error, to avoid breaking
2034 	 * callers that blindly resize the filesystem to the full size of
2035 	 * the underlying block device.
2036 	 */
2037 	if (ext4_has_feature_bigalloc(sb))
2038 		n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2039 
2040 retry:
2041 	o_blocks_count = ext4_blocks_count(es);
2042 
2043 	ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2044 		 "to %llu blocks", o_blocks_count, n_blocks_count);
2045 
2046 	if (n_blocks_count < o_blocks_count) {
2047 		/* On-line shrinking not supported */
2048 		ext4_warning(sb, "can't shrink FS - resize aborted");
2049 		return -EINVAL;
2050 	}
2051 
2052 	if (n_blocks_count == o_blocks_count)
2053 		/* Nothing need to do */
2054 		return 0;
2055 
2056 	n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2057 	if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2058 		ext4_warning(sb, "resize would cause inodes_count overflow");
2059 		return -EINVAL;
2060 	}
2061 	ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2062 
2063 	n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2064 	o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2065 
2066 	meta_bg = ext4_has_feature_meta_bg(sb);
2067 
2068 	if (ext4_has_feature_resize_inode(sb)) {
2069 		if (meta_bg) {
2070 			ext4_error(sb, "resize_inode and meta_bg enabled "
2071 				   "simultaneously");
2072 			return -EINVAL;
2073 		}
2074 		if (n_desc_blocks > o_desc_blocks +
2075 		    le16_to_cpu(es->s_reserved_gdt_blocks)) {
2076 			n_blocks_count_retry = n_blocks_count;
2077 			n_desc_blocks = o_desc_blocks +
2078 				le16_to_cpu(es->s_reserved_gdt_blocks);
2079 			n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2080 			n_blocks_count = (ext4_fsblk_t)n_group *
2081 				EXT4_BLOCKS_PER_GROUP(sb) +
2082 				le32_to_cpu(es->s_first_data_block);
2083 			n_group--; /* set to last group number */
2084 		}
2085 
2086 		if (!resize_inode)
2087 			resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2088 						 EXT4_IGET_SPECIAL);
2089 		if (IS_ERR(resize_inode)) {
2090 			ext4_warning(sb, "Error opening resize inode");
2091 			return PTR_ERR(resize_inode);
2092 		}
2093 	}
2094 
2095 	if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2096 		err = ext4_convert_meta_bg(sb, resize_inode);
2097 		if (err)
2098 			goto out;
2099 		if (resize_inode) {
2100 			iput(resize_inode);
2101 			resize_inode = NULL;
2102 		}
2103 		if (n_blocks_count_retry) {
2104 			n_blocks_count = n_blocks_count_retry;
2105 			n_blocks_count_retry = 0;
2106 			goto retry;
2107 		}
2108 	}
2109 
2110 	/*
2111 	 * Make sure the last group has enough space so that it's
2112 	 * guaranteed to have enough space for all metadata blocks
2113 	 * that it might need to hold.  (We might not need to store
2114 	 * the inode table blocks in the last block group, but there
2115 	 * will be cases where this might be needed.)
2116 	 */
2117 	if ((ext4_group_first_block_no(sb, n_group) +
2118 	     ext4_group_overhead_blocks(sb, n_group) + 2 +
2119 	     sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2120 		n_blocks_count = ext4_group_first_block_no(sb, n_group);
2121 		n_group--;
2122 		n_blocks_count_retry = 0;
2123 		if (resize_inode) {
2124 			iput(resize_inode);
2125 			resize_inode = NULL;
2126 		}
2127 		goto retry;
2128 	}
2129 
2130 	/* extend the last group */
2131 	if (n_group == o_group)
2132 		add = n_blocks_count - o_blocks_count;
2133 	else
2134 		add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2135 	if (add > 0) {
2136 		err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2137 		if (err)
2138 			goto out;
2139 	}
2140 
2141 	if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2142 		goto out;
2143 
2144 	err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2145 	if (err)
2146 		goto out;
2147 
2148 	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2149 	if (err)
2150 		goto out;
2151 
2152 	flex_gd = alloc_flex_gd(flexbg_size);
2153 	if (flex_gd == NULL) {
2154 		err = -ENOMEM;
2155 		goto out;
2156 	}
2157 
2158 	/* Add flex groups. Note that a regular group is a
2159 	 * flex group with 1 group.
2160 	 */
2161 	while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2162 		if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2163 			if (last_update_time)
2164 				ext4_msg(sb, KERN_INFO,
2165 					 "resized to %llu blocks",
2166 					 ext4_blocks_count(es));
2167 			last_update_time = jiffies;
2168 		}
2169 		if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2170 			break;
2171 		err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2172 		if (unlikely(err))
2173 			break;
2174 	}
2175 
2176 	if (!err && n_blocks_count_retry) {
2177 		n_blocks_count = n_blocks_count_retry;
2178 		n_blocks_count_retry = 0;
2179 		free_flex_gd(flex_gd);
2180 		flex_gd = NULL;
2181 		if (resize_inode) {
2182 			iput(resize_inode);
2183 			resize_inode = NULL;
2184 		}
2185 		goto retry;
2186 	}
2187 
2188 out:
2189 	if (flex_gd)
2190 		free_flex_gd(flex_gd);
2191 	if (resize_inode != NULL)
2192 		iput(resize_inode);
2193 	if (err)
2194 		ext4_warning(sb, "error (%d) occurred during "
2195 			     "file system resize", err);
2196 	ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2197 		 ext4_blocks_count(es));
2198 	return err;
2199 }
2200