xref: /openbmc/linux/fs/ext4/ialloc.c (revision fd2d42912f9f09e5250cb3b024ee0625704e9cb7)
1 /*
2  *  linux/fs/ext4/ialloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  BSD ufs-inspired inode and directory allocation by
10  *  Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14 
15 #include <linux/time.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.h>
20 #include <linux/stat.h>
21 #include <linux/string.h>
22 #include <linux/quotaops.h>
23 #include <linux/buffer_head.h>
24 #include <linux/random.h>
25 #include <linux/bitops.h>
26 #include <linux/blkdev.h>
27 #include <asm/byteorder.h>
28 
29 #include "xattr.h"
30 #include "acl.h"
31 #include "group.h"
32 
33 /*
34  * ialloc.c contains the inodes allocation and deallocation routines
35  */
36 
37 /*
38  * The free inodes are managed by bitmaps.  A file system contains several
39  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
40  * block for inodes, N blocks for the inode table and data blocks.
41  *
42  * The file system contains group descriptors which are located after the
43  * super block.  Each descriptor contains the number of the bitmap block and
44  * the free blocks count in the block.
45  */
46 
47 /*
48  * To avoid calling the atomic setbit hundreds or thousands of times, we only
49  * need to use it within a single byte (to ensure we get endianness right).
50  * We can use memset for the rest of the bitmap as there are no other users.
51  */
52 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53 {
54 	int i;
55 
56 	if (start_bit >= end_bit)
57 		return;
58 
59 	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 		ext4_set_bit(i, bitmap);
62 	if (i < end_bit)
63 		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64 }
65 
66 /* Initializes an uninitialized inode bitmap */
67 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
68 				ext4_group_t block_group,
69 				struct ext4_group_desc *gdp)
70 {
71 	struct ext4_sb_info *sbi = EXT4_SB(sb);
72 
73 	J_ASSERT_BH(bh, buffer_locked(bh));
74 
75 	/* If checksum is bad mark all blocks and inodes use to prevent
76 	 * allocation, essentially implementing a per-group read-only flag. */
77 	if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
78 		ext4_error(sb, __FUNCTION__, "Checksum bad for group %lu\n",
79 			   block_group);
80 		gdp->bg_free_blocks_count = 0;
81 		gdp->bg_free_inodes_count = 0;
82 		gdp->bg_itable_unused = 0;
83 		memset(bh->b_data, 0xff, sb->s_blocksize);
84 		return 0;
85 	}
86 
87 	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
88 	mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
89 			bh->b_data);
90 
91 	return EXT4_INODES_PER_GROUP(sb);
92 }
93 
94 /*
95  * Read the inode allocation bitmap for a given block_group, reading
96  * into the specified slot in the superblock's bitmap cache.
97  *
98  * Return buffer_head of bitmap on success or NULL.
99  */
100 static struct buffer_head *
101 read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
102 {
103 	struct ext4_group_desc *desc;
104 	struct buffer_head *bh = NULL;
105 
106 	desc = ext4_get_group_desc(sb, block_group, NULL);
107 	if (!desc)
108 		goto error_out;
109 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
110 		bh = sb_getblk(sb, ext4_inode_bitmap(sb, desc));
111 		if (!buffer_uptodate(bh)) {
112 			lock_buffer(bh);
113 			if (!buffer_uptodate(bh)) {
114 				ext4_init_inode_bitmap(sb, bh, block_group,
115 						       desc);
116 				set_buffer_uptodate(bh);
117 			}
118 			unlock_buffer(bh);
119 		}
120 	} else {
121 		bh = sb_bread(sb, ext4_inode_bitmap(sb, desc));
122 	}
123 	if (!bh)
124 		ext4_error(sb, "read_inode_bitmap",
125 			    "Cannot read inode bitmap - "
126 			    "block_group = %lu, inode_bitmap = %llu",
127 			    block_group, ext4_inode_bitmap(sb, desc));
128 error_out:
129 	return bh;
130 }
131 
132 /*
133  * NOTE! When we get the inode, we're the only people
134  * that have access to it, and as such there are no
135  * race conditions we have to worry about. The inode
136  * is not on the hash-lists, and it cannot be reached
137  * through the filesystem because the directory entry
138  * has been deleted earlier.
139  *
140  * HOWEVER: we must make sure that we get no aliases,
141  * which means that we have to call "clear_inode()"
142  * _before_ we mark the inode not in use in the inode
143  * bitmaps. Otherwise a newly created file might use
144  * the same inode number (not actually the same pointer
145  * though), and then we'd have two inodes sharing the
146  * same inode number and space on the harddisk.
147  */
148 void ext4_free_inode (handle_t *handle, struct inode * inode)
149 {
150 	struct super_block * sb = inode->i_sb;
151 	int is_directory;
152 	unsigned long ino;
153 	struct buffer_head *bitmap_bh = NULL;
154 	struct buffer_head *bh2;
155 	ext4_group_t block_group;
156 	unsigned long bit;
157 	struct ext4_group_desc * gdp;
158 	struct ext4_super_block * es;
159 	struct ext4_sb_info *sbi;
160 	int fatal = 0, err;
161 
162 	if (atomic_read(&inode->i_count) > 1) {
163 		printk ("ext4_free_inode: inode has count=%d\n",
164 					atomic_read(&inode->i_count));
165 		return;
166 	}
167 	if (inode->i_nlink) {
168 		printk ("ext4_free_inode: inode has nlink=%d\n",
169 			inode->i_nlink);
170 		return;
171 	}
172 	if (!sb) {
173 		printk("ext4_free_inode: inode on nonexistent device\n");
174 		return;
175 	}
176 	sbi = EXT4_SB(sb);
177 
178 	ino = inode->i_ino;
179 	ext4_debug ("freeing inode %lu\n", ino);
180 
181 	/*
182 	 * Note: we must free any quota before locking the superblock,
183 	 * as writing the quota to disk may need the lock as well.
184 	 */
185 	DQUOT_INIT(inode);
186 	ext4_xattr_delete_inode(handle, inode);
187 	DQUOT_FREE_INODE(inode);
188 	DQUOT_DROP(inode);
189 
190 	is_directory = S_ISDIR(inode->i_mode);
191 
192 	/* Do this BEFORE marking the inode not in use or returning an error */
193 	clear_inode (inode);
194 
195 	es = EXT4_SB(sb)->s_es;
196 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
197 		ext4_error (sb, "ext4_free_inode",
198 			    "reserved or nonexistent inode %lu", ino);
199 		goto error_return;
200 	}
201 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
202 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
203 	bitmap_bh = read_inode_bitmap(sb, block_group);
204 	if (!bitmap_bh)
205 		goto error_return;
206 
207 	BUFFER_TRACE(bitmap_bh, "get_write_access");
208 	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
209 	if (fatal)
210 		goto error_return;
211 
212 	/* Ok, now we can actually update the inode bitmaps.. */
213 	if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
214 					bit, bitmap_bh->b_data))
215 		ext4_error (sb, "ext4_free_inode",
216 			      "bit already cleared for inode %lu", ino);
217 	else {
218 		gdp = ext4_get_group_desc (sb, block_group, &bh2);
219 
220 		BUFFER_TRACE(bh2, "get_write_access");
221 		fatal = ext4_journal_get_write_access(handle, bh2);
222 		if (fatal) goto error_return;
223 
224 		if (gdp) {
225 			spin_lock(sb_bgl_lock(sbi, block_group));
226 			gdp->bg_free_inodes_count = cpu_to_le16(
227 				le16_to_cpu(gdp->bg_free_inodes_count) + 1);
228 			if (is_directory)
229 				gdp->bg_used_dirs_count = cpu_to_le16(
230 				  le16_to_cpu(gdp->bg_used_dirs_count) - 1);
231 			gdp->bg_checksum = ext4_group_desc_csum(sbi,
232 							block_group, gdp);
233 			spin_unlock(sb_bgl_lock(sbi, block_group));
234 			percpu_counter_inc(&sbi->s_freeinodes_counter);
235 			if (is_directory)
236 				percpu_counter_dec(&sbi->s_dirs_counter);
237 
238 		}
239 		BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
240 		err = ext4_journal_dirty_metadata(handle, bh2);
241 		if (!fatal) fatal = err;
242 	}
243 	BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata");
244 	err = ext4_journal_dirty_metadata(handle, bitmap_bh);
245 	if (!fatal)
246 		fatal = err;
247 	sb->s_dirt = 1;
248 error_return:
249 	brelse(bitmap_bh);
250 	ext4_std_error(sb, fatal);
251 }
252 
253 /*
254  * There are two policies for allocating an inode.  If the new inode is
255  * a directory, then a forward search is made for a block group with both
256  * free space and a low directory-to-inode ratio; if that fails, then of
257  * the groups with above-average free space, that group with the fewest
258  * directories already is chosen.
259  *
260  * For other inodes, search forward from the parent directory\'s block
261  * group to find a free inode.
262  */
263 static ext4_group_t find_group_dir(struct super_block *sb, struct inode *parent)
264 {
265 	ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
266 	unsigned int freei, avefreei;
267 	struct ext4_group_desc *desc, *best_desc = NULL;
268 	ext4_group_t group, best_group = -1;
269 
270 	freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
271 	avefreei = freei / ngroups;
272 
273 	for (group = 0; group < ngroups; group++) {
274 		desc = ext4_get_group_desc (sb, group, NULL);
275 		if (!desc || !desc->bg_free_inodes_count)
276 			continue;
277 		if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
278 			continue;
279 		if (!best_desc ||
280 		    (le16_to_cpu(desc->bg_free_blocks_count) >
281 		     le16_to_cpu(best_desc->bg_free_blocks_count))) {
282 			best_group = group;
283 			best_desc = desc;
284 		}
285 	}
286 	return best_group;
287 }
288 
289 /*
290  * Orlov's allocator for directories.
291  *
292  * We always try to spread first-level directories.
293  *
294  * If there are blockgroups with both free inodes and free blocks counts
295  * not worse than average we return one with smallest directory count.
296  * Otherwise we simply return a random group.
297  *
298  * For the rest rules look so:
299  *
300  * It's OK to put directory into a group unless
301  * it has too many directories already (max_dirs) or
302  * it has too few free inodes left (min_inodes) or
303  * it has too few free blocks left (min_blocks) or
304  * it's already running too large debt (max_debt).
305  * Parent's group is prefered, if it doesn't satisfy these
306  * conditions we search cyclically through the rest. If none
307  * of the groups look good we just look for a group with more
308  * free inodes than average (starting at parent's group).
309  *
310  * Debt is incremented each time we allocate a directory and decremented
311  * when we allocate an inode, within 0--255.
312  */
313 
314 #define INODE_COST 64
315 #define BLOCK_COST 256
316 
317 static ext4_group_t find_group_orlov(struct super_block *sb,
318 				      struct inode *parent)
319 {
320 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
321 	struct ext4_sb_info *sbi = EXT4_SB(sb);
322 	struct ext4_super_block *es = sbi->s_es;
323 	ext4_group_t ngroups = sbi->s_groups_count;
324 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
325 	unsigned int freei, avefreei;
326 	ext4_fsblk_t freeb, avefreeb;
327 	ext4_fsblk_t blocks_per_dir;
328 	unsigned int ndirs;
329 	int max_debt, max_dirs, min_inodes;
330 	ext4_grpblk_t min_blocks;
331 	ext4_group_t group = -1, i;
332 	struct ext4_group_desc *desc;
333 
334 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
335 	avefreei = freei / ngroups;
336 	freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
337 	avefreeb = freeb;
338 	do_div(avefreeb, ngroups);
339 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
340 
341 	if ((parent == sb->s_root->d_inode) ||
342 	    (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
343 		int best_ndir = inodes_per_group;
344 		ext4_group_t best_group = -1;
345 
346 		get_random_bytes(&group, sizeof(group));
347 		parent_group = (unsigned)group % ngroups;
348 		for (i = 0; i < ngroups; i++) {
349 			group = (parent_group + i) % ngroups;
350 			desc = ext4_get_group_desc (sb, group, NULL);
351 			if (!desc || !desc->bg_free_inodes_count)
352 				continue;
353 			if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
354 				continue;
355 			if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
356 				continue;
357 			if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
358 				continue;
359 			best_group = group;
360 			best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
361 		}
362 		if (best_group >= 0)
363 			return best_group;
364 		goto fallback;
365 	}
366 
367 	blocks_per_dir = ext4_blocks_count(es) - freeb;
368 	do_div(blocks_per_dir, ndirs);
369 
370 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
371 	min_inodes = avefreei - inodes_per_group / 4;
372 	min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
373 
374 	max_debt = EXT4_BLOCKS_PER_GROUP(sb);
375 	max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
376 	if (max_debt * INODE_COST > inodes_per_group)
377 		max_debt = inodes_per_group / INODE_COST;
378 	if (max_debt > 255)
379 		max_debt = 255;
380 	if (max_debt == 0)
381 		max_debt = 1;
382 
383 	for (i = 0; i < ngroups; i++) {
384 		group = (parent_group + i) % ngroups;
385 		desc = ext4_get_group_desc (sb, group, NULL);
386 		if (!desc || !desc->bg_free_inodes_count)
387 			continue;
388 		if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
389 			continue;
390 		if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
391 			continue;
392 		if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
393 			continue;
394 		return group;
395 	}
396 
397 fallback:
398 	for (i = 0; i < ngroups; i++) {
399 		group = (parent_group + i) % ngroups;
400 		desc = ext4_get_group_desc (sb, group, NULL);
401 		if (!desc || !desc->bg_free_inodes_count)
402 			continue;
403 		if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
404 			return group;
405 	}
406 
407 	if (avefreei) {
408 		/*
409 		 * The free-inodes counter is approximate, and for really small
410 		 * filesystems the above test can fail to find any blockgroups
411 		 */
412 		avefreei = 0;
413 		goto fallback;
414 	}
415 
416 	return -1;
417 }
418 
419 static ext4_group_t find_group_other(struct super_block *sb,
420 					struct inode *parent)
421 {
422 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
423 	ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
424 	struct ext4_group_desc *desc;
425 	ext4_group_t group, i;
426 
427 	/*
428 	 * Try to place the inode in its parent directory
429 	 */
430 	group = parent_group;
431 	desc = ext4_get_group_desc (sb, group, NULL);
432 	if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
433 			le16_to_cpu(desc->bg_free_blocks_count))
434 		return group;
435 
436 	/*
437 	 * We're going to place this inode in a different blockgroup from its
438 	 * parent.  We want to cause files in a common directory to all land in
439 	 * the same blockgroup.  But we want files which are in a different
440 	 * directory which shares a blockgroup with our parent to land in a
441 	 * different blockgroup.
442 	 *
443 	 * So add our directory's i_ino into the starting point for the hash.
444 	 */
445 	group = (group + parent->i_ino) % ngroups;
446 
447 	/*
448 	 * Use a quadratic hash to find a group with a free inode and some free
449 	 * blocks.
450 	 */
451 	for (i = 1; i < ngroups; i <<= 1) {
452 		group += i;
453 		if (group >= ngroups)
454 			group -= ngroups;
455 		desc = ext4_get_group_desc (sb, group, NULL);
456 		if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
457 				le16_to_cpu(desc->bg_free_blocks_count))
458 			return group;
459 	}
460 
461 	/*
462 	 * That failed: try linear search for a free inode, even if that group
463 	 * has no free blocks.
464 	 */
465 	group = parent_group;
466 	for (i = 0; i < ngroups; i++) {
467 		if (++group >= ngroups)
468 			group = 0;
469 		desc = ext4_get_group_desc (sb, group, NULL);
470 		if (desc && le16_to_cpu(desc->bg_free_inodes_count))
471 			return group;
472 	}
473 
474 	return -1;
475 }
476 
477 /*
478  * There are two policies for allocating an inode.  If the new inode is
479  * a directory, then a forward search is made for a block group with both
480  * free space and a low directory-to-inode ratio; if that fails, then of
481  * the groups with above-average free space, that group with the fewest
482  * directories already is chosen.
483  *
484  * For other inodes, search forward from the parent directory's block
485  * group to find a free inode.
486  */
487 struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
488 {
489 	struct super_block *sb;
490 	struct buffer_head *bitmap_bh = NULL;
491 	struct buffer_head *bh2;
492 	ext4_group_t group;
493 	unsigned long ino = 0;
494 	struct inode * inode;
495 	struct ext4_group_desc * gdp = NULL;
496 	struct ext4_super_block * es;
497 	struct ext4_inode_info *ei;
498 	struct ext4_sb_info *sbi;
499 	int err = 0;
500 	struct inode *ret;
501 	int i, free = 0;
502 
503 	/* Cannot create files in a deleted directory */
504 	if (!dir || !dir->i_nlink)
505 		return ERR_PTR(-EPERM);
506 
507 	sb = dir->i_sb;
508 	inode = new_inode(sb);
509 	if (!inode)
510 		return ERR_PTR(-ENOMEM);
511 	ei = EXT4_I(inode);
512 
513 	sbi = EXT4_SB(sb);
514 	es = sbi->s_es;
515 	if (S_ISDIR(mode)) {
516 		if (test_opt (sb, OLDALLOC))
517 			group = find_group_dir(sb, dir);
518 		else
519 			group = find_group_orlov(sb, dir);
520 	} else
521 		group = find_group_other(sb, dir);
522 
523 	err = -ENOSPC;
524 	if (group == -1)
525 		goto out;
526 
527 	for (i = 0; i < sbi->s_groups_count; i++) {
528 		err = -EIO;
529 
530 		gdp = ext4_get_group_desc(sb, group, &bh2);
531 		if (!gdp)
532 			goto fail;
533 
534 		brelse(bitmap_bh);
535 		bitmap_bh = read_inode_bitmap(sb, group);
536 		if (!bitmap_bh)
537 			goto fail;
538 
539 		ino = 0;
540 
541 repeat_in_this_group:
542 		ino = ext4_find_next_zero_bit((unsigned long *)
543 				bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
544 		if (ino < EXT4_INODES_PER_GROUP(sb)) {
545 
546 			BUFFER_TRACE(bitmap_bh, "get_write_access");
547 			err = ext4_journal_get_write_access(handle, bitmap_bh);
548 			if (err)
549 				goto fail;
550 
551 			if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
552 						ino, bitmap_bh->b_data)) {
553 				/* we won it */
554 				BUFFER_TRACE(bitmap_bh,
555 					"call ext4_journal_dirty_metadata");
556 				err = ext4_journal_dirty_metadata(handle,
557 								bitmap_bh);
558 				if (err)
559 					goto fail;
560 				goto got;
561 			}
562 			/* we lost it */
563 			jbd2_journal_release_buffer(handle, bitmap_bh);
564 
565 			if (++ino < EXT4_INODES_PER_GROUP(sb))
566 				goto repeat_in_this_group;
567 		}
568 
569 		/*
570 		 * This case is possible in concurrent environment.  It is very
571 		 * rare.  We cannot repeat the find_group_xxx() call because
572 		 * that will simply return the same blockgroup, because the
573 		 * group descriptor metadata has not yet been updated.
574 		 * So we just go onto the next blockgroup.
575 		 */
576 		if (++group == sbi->s_groups_count)
577 			group = 0;
578 	}
579 	err = -ENOSPC;
580 	goto out;
581 
582 got:
583 	ino++;
584 	if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
585 	    ino > EXT4_INODES_PER_GROUP(sb)) {
586 		ext4_error(sb, __FUNCTION__,
587 			   "reserved inode or inode > inodes count - "
588 			   "block_group = %lu, inode=%lu", group,
589 			   ino + group * EXT4_INODES_PER_GROUP(sb));
590 		err = -EIO;
591 		goto fail;
592 	}
593 
594 	BUFFER_TRACE(bh2, "get_write_access");
595 	err = ext4_journal_get_write_access(handle, bh2);
596 	if (err) goto fail;
597 
598 	/* We may have to initialize the block bitmap if it isn't already */
599 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
600 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
601 		struct buffer_head *block_bh = read_block_bitmap(sb, group);
602 
603 		BUFFER_TRACE(block_bh, "get block bitmap access");
604 		err = ext4_journal_get_write_access(handle, block_bh);
605 		if (err) {
606 			brelse(block_bh);
607 			goto fail;
608 		}
609 
610 		free = 0;
611 		spin_lock(sb_bgl_lock(sbi, group));
612 		/* recheck and clear flag under lock if we still need to */
613 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
614 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
615 			free = ext4_free_blocks_after_init(sb, group, gdp);
616 			gdp->bg_free_blocks_count = cpu_to_le16(free);
617 		}
618 		spin_unlock(sb_bgl_lock(sbi, group));
619 
620 		/* Don't need to dirty bitmap block if we didn't change it */
621 		if (free) {
622 			BUFFER_TRACE(block_bh, "dirty block bitmap");
623 			err = ext4_journal_dirty_metadata(handle, block_bh);
624 		}
625 
626 		brelse(block_bh);
627 		if (err)
628 			goto fail;
629 	}
630 
631 	spin_lock(sb_bgl_lock(sbi, group));
632 	/* If we didn't allocate from within the initialized part of the inode
633 	 * table then we need to initialize up to this inode. */
634 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
635 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
636 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
637 
638 			/* When marking the block group with
639 			 * ~EXT4_BG_INODE_UNINIT we don't want to depend
640 			 * on the value of bg_itable_unsed even though
641 			 * mke2fs could have initialized the same for us.
642 			 * Instead we calculated the value below
643 			 */
644 
645 			free = 0;
646 		} else {
647 			free = EXT4_INODES_PER_GROUP(sb) -
648 				le16_to_cpu(gdp->bg_itable_unused);
649 		}
650 
651 		/*
652 		 * Check the relative inode number against the last used
653 		 * relative inode number in this group. if it is greater
654 		 * we need to  update the bg_itable_unused count
655 		 *
656 		 */
657 		if (ino > free)
658 			gdp->bg_itable_unused =
659 				cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino);
660 	}
661 
662 	gdp->bg_free_inodes_count =
663 		cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
664 	if (S_ISDIR(mode)) {
665 		gdp->bg_used_dirs_count =
666 			cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
667 	}
668 	gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
669 	spin_unlock(sb_bgl_lock(sbi, group));
670 	BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
671 	err = ext4_journal_dirty_metadata(handle, bh2);
672 	if (err) goto fail;
673 
674 	percpu_counter_dec(&sbi->s_freeinodes_counter);
675 	if (S_ISDIR(mode))
676 		percpu_counter_inc(&sbi->s_dirs_counter);
677 	sb->s_dirt = 1;
678 
679 	inode->i_uid = current->fsuid;
680 	if (test_opt (sb, GRPID))
681 		inode->i_gid = dir->i_gid;
682 	else if (dir->i_mode & S_ISGID) {
683 		inode->i_gid = dir->i_gid;
684 		if (S_ISDIR(mode))
685 			mode |= S_ISGID;
686 	} else
687 		inode->i_gid = current->fsgid;
688 	inode->i_mode = mode;
689 
690 	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
691 	/* This is the optimal IO size (for stat), not the fs block size */
692 	inode->i_blocks = 0;
693 	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
694 						       ext4_current_time(inode);
695 
696 	memset(ei->i_data, 0, sizeof(ei->i_data));
697 	ei->i_dir_start_lookup = 0;
698 	ei->i_disksize = 0;
699 
700 	ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL;
701 	if (S_ISLNK(mode))
702 		ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
703 	/* dirsync only applies to directories */
704 	if (!S_ISDIR(mode))
705 		ei->i_flags &= ~EXT4_DIRSYNC_FL;
706 	ei->i_file_acl = 0;
707 	ei->i_dir_acl = 0;
708 	ei->i_dtime = 0;
709 	ei->i_block_alloc_info = NULL;
710 	ei->i_block_group = group;
711 
712 	ext4_set_inode_flags(inode);
713 	if (IS_DIRSYNC(inode))
714 		handle->h_sync = 1;
715 	insert_inode_hash(inode);
716 	spin_lock(&sbi->s_next_gen_lock);
717 	inode->i_generation = sbi->s_next_generation++;
718 	spin_unlock(&sbi->s_next_gen_lock);
719 
720 	ei->i_state = EXT4_STATE_NEW;
721 
722 	ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
723 
724 	ret = inode;
725 	if(DQUOT_ALLOC_INODE(inode)) {
726 		err = -EDQUOT;
727 		goto fail_drop;
728 	}
729 
730 	err = ext4_init_acl(handle, inode, dir);
731 	if (err)
732 		goto fail_free_drop;
733 
734 	err = ext4_init_security(handle,inode, dir);
735 	if (err)
736 		goto fail_free_drop;
737 
738 	err = ext4_mark_inode_dirty(handle, inode);
739 	if (err) {
740 		ext4_std_error(sb, err);
741 		goto fail_free_drop;
742 	}
743 	if (test_opt(sb, EXTENTS)) {
744 		EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
745 		ext4_ext_tree_init(handle, inode);
746 		if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
747 			err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
748 			if (err) goto fail;
749 			EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS);
750 			BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata");
751 			err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
752 		}
753 	}
754 
755 	ext4_debug("allocating inode %lu\n", inode->i_ino);
756 	goto really_out;
757 fail:
758 	ext4_std_error(sb, err);
759 out:
760 	iput(inode);
761 	ret = ERR_PTR(err);
762 really_out:
763 	brelse(bitmap_bh);
764 	return ret;
765 
766 fail_free_drop:
767 	DQUOT_FREE_INODE(inode);
768 
769 fail_drop:
770 	DQUOT_DROP(inode);
771 	inode->i_flags |= S_NOQUOTA;
772 	inode->i_nlink = 0;
773 	iput(inode);
774 	brelse(bitmap_bh);
775 	return ERR_PTR(err);
776 }
777 
778 /* Verify that we are loading a valid orphan from disk */
779 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
780 {
781 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
782 	ext4_group_t block_group;
783 	int bit;
784 	struct buffer_head *bitmap_bh = NULL;
785 	struct inode *inode = NULL;
786 
787 	/* Error cases - e2fsck has already cleaned up for us */
788 	if (ino > max_ino) {
789 		ext4_warning(sb, __FUNCTION__,
790 			     "bad orphan ino %lu!  e2fsck was run?", ino);
791 		goto out;
792 	}
793 
794 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
795 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
796 	bitmap_bh = read_inode_bitmap(sb, block_group);
797 	if (!bitmap_bh) {
798 		ext4_warning(sb, __FUNCTION__,
799 			     "inode bitmap error for orphan %lu", ino);
800 		goto out;
801 	}
802 
803 	/* Having the inode bit set should be a 100% indicator that this
804 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
805 	 * inodes that were being truncated, so we can't check i_nlink==0.
806 	 */
807 	if (!ext4_test_bit(bit, bitmap_bh->b_data) ||
808 			!(inode = iget(sb, ino)) || is_bad_inode(inode) ||
809 			NEXT_ORPHAN(inode) > max_ino) {
810 		ext4_warning(sb, __FUNCTION__,
811 			     "bad orphan inode %lu!  e2fsck was run?", ino);
812 		printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
813 		       bit, (unsigned long long)bitmap_bh->b_blocknr,
814 		       ext4_test_bit(bit, bitmap_bh->b_data));
815 		printk(KERN_NOTICE "inode=%p\n", inode);
816 		if (inode) {
817 			printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
818 			       is_bad_inode(inode));
819 			printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
820 			       NEXT_ORPHAN(inode));
821 			printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
822 		}
823 		/* Avoid freeing blocks if we got a bad deleted inode */
824 		if (inode && inode->i_nlink == 0)
825 			inode->i_blocks = 0;
826 		iput(inode);
827 		inode = NULL;
828 	}
829 out:
830 	brelse(bitmap_bh);
831 	return inode;
832 }
833 
834 unsigned long ext4_count_free_inodes (struct super_block * sb)
835 {
836 	unsigned long desc_count;
837 	struct ext4_group_desc *gdp;
838 	ext4_group_t i;
839 #ifdef EXT4FS_DEBUG
840 	struct ext4_super_block *es;
841 	unsigned long bitmap_count, x;
842 	struct buffer_head *bitmap_bh = NULL;
843 
844 	es = EXT4_SB(sb)->s_es;
845 	desc_count = 0;
846 	bitmap_count = 0;
847 	gdp = NULL;
848 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
849 		gdp = ext4_get_group_desc (sb, i, NULL);
850 		if (!gdp)
851 			continue;
852 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
853 		brelse(bitmap_bh);
854 		bitmap_bh = read_inode_bitmap(sb, i);
855 		if (!bitmap_bh)
856 			continue;
857 
858 		x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
859 		printk("group %d: stored = %d, counted = %lu\n",
860 			i, le16_to_cpu(gdp->bg_free_inodes_count), x);
861 		bitmap_count += x;
862 	}
863 	brelse(bitmap_bh);
864 	printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
865 		le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
866 	return desc_count;
867 #else
868 	desc_count = 0;
869 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
870 		gdp = ext4_get_group_desc (sb, i, NULL);
871 		if (!gdp)
872 			continue;
873 		desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
874 		cond_resched();
875 	}
876 	return desc_count;
877 #endif
878 }
879 
880 /* Called at mount-time, super-block is locked */
881 unsigned long ext4_count_dirs (struct super_block * sb)
882 {
883 	unsigned long count = 0;
884 	ext4_group_t i;
885 
886 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
887 		struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
888 		if (!gdp)
889 			continue;
890 		count += le16_to_cpu(gdp->bg_used_dirs_count);
891 	}
892 	return count;
893 }
894 
895