xref: /openbmc/linux/fs/ext4/ialloc.c (revision 82e6fdd6)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/ialloc.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  BSD ufs-inspired inode and directory allocation by
11  *  Stephen Tweedie (sct@redhat.com), 1993
12  *  Big-endian to little-endian byte-swapping/bitmaps by
13  *        David S. Miller (davem@caip.rutgers.edu), 1995
14  */
15 
16 #include <linux/time.h>
17 #include <linux/fs.h>
18 #include <linux/stat.h>
19 #include <linux/string.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
22 #include <linux/random.h>
23 #include <linux/bitops.h>
24 #include <linux/blkdev.h>
25 #include <linux/cred.h>
26 
27 #include <asm/byteorder.h>
28 
29 #include "ext4.h"
30 #include "ext4_jbd2.h"
31 #include "xattr.h"
32 #include "acl.h"
33 
34 #include <trace/events/ext4.h>
35 
36 /*
37  * ialloc.c contains the inodes allocation and deallocation routines
38  */
39 
40 /*
41  * The free inodes are managed by bitmaps.  A file system contains several
42  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
43  * block for inodes, N blocks for the inode table and data blocks.
44  *
45  * The file system contains group descriptors which are located after the
46  * super block.  Each descriptor contains the number of the bitmap block and
47  * the free blocks count in the block.
48  */
49 
50 /*
51  * To avoid calling the atomic setbit hundreds or thousands of times, we only
52  * need to use it within a single byte (to ensure we get endianness right).
53  * We can use memset for the rest of the bitmap as there are no other users.
54  */
55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56 {
57 	int i;
58 
59 	if (start_bit >= end_bit)
60 		return;
61 
62 	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 		ext4_set_bit(i, bitmap);
65 	if (i < end_bit)
66 		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67 }
68 
69 /* Initializes an uninitialized inode bitmap */
70 static int ext4_init_inode_bitmap(struct super_block *sb,
71 				       struct buffer_head *bh,
72 				       ext4_group_t block_group,
73 				       struct ext4_group_desc *gdp)
74 {
75 	struct ext4_group_info *grp;
76 	struct ext4_sb_info *sbi = EXT4_SB(sb);
77 	J_ASSERT_BH(bh, buffer_locked(bh));
78 
79 	/* If checksum is bad mark all blocks and inodes use to prevent
80 	 * allocation, essentially implementing a per-group read-only flag. */
81 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
82 		grp = ext4_get_group_info(sb, block_group);
83 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
84 			percpu_counter_sub(&sbi->s_freeclusters_counter,
85 					   grp->bb_free);
86 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
87 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
88 			int count;
89 			count = ext4_free_inodes_count(sb, gdp);
90 			percpu_counter_sub(&sbi->s_freeinodes_counter,
91 					   count);
92 		}
93 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
94 		return -EFSBADCRC;
95 	}
96 
97 	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
98 	ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
99 			bh->b_data);
100 	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
101 				   EXT4_INODES_PER_GROUP(sb) / 8);
102 	ext4_group_desc_csum_set(sb, block_group, gdp);
103 
104 	return 0;
105 }
106 
107 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
108 {
109 	if (uptodate) {
110 		set_buffer_uptodate(bh);
111 		set_bitmap_uptodate(bh);
112 	}
113 	unlock_buffer(bh);
114 	put_bh(bh);
115 }
116 
117 static int ext4_validate_inode_bitmap(struct super_block *sb,
118 				      struct ext4_group_desc *desc,
119 				      ext4_group_t block_group,
120 				      struct buffer_head *bh)
121 {
122 	ext4_fsblk_t	blk;
123 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
124 	struct ext4_sb_info *sbi = EXT4_SB(sb);
125 
126 	if (buffer_verified(bh))
127 		return 0;
128 	if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
129 		return -EFSCORRUPTED;
130 
131 	ext4_lock_group(sb, block_group);
132 	blk = ext4_inode_bitmap(sb, desc);
133 	if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
134 					   EXT4_INODES_PER_GROUP(sb) / 8)) {
135 		ext4_unlock_group(sb, block_group);
136 		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
137 			   "inode_bitmap = %llu", block_group, blk);
138 		grp = ext4_get_group_info(sb, block_group);
139 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
140 			int count;
141 			count = ext4_free_inodes_count(sb, desc);
142 			percpu_counter_sub(&sbi->s_freeinodes_counter,
143 					   count);
144 		}
145 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
146 		return -EFSBADCRC;
147 	}
148 	set_buffer_verified(bh);
149 	ext4_unlock_group(sb, block_group);
150 	return 0;
151 }
152 
153 /*
154  * Read the inode allocation bitmap for a given block_group, reading
155  * into the specified slot in the superblock's bitmap cache.
156  *
157  * Return buffer_head of bitmap on success or NULL.
158  */
159 static struct buffer_head *
160 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
161 {
162 	struct ext4_group_desc *desc;
163 	struct buffer_head *bh = NULL;
164 	ext4_fsblk_t bitmap_blk;
165 	int err;
166 
167 	desc = ext4_get_group_desc(sb, block_group, NULL);
168 	if (!desc)
169 		return ERR_PTR(-EFSCORRUPTED);
170 
171 	bitmap_blk = ext4_inode_bitmap(sb, desc);
172 	bh = sb_getblk(sb, bitmap_blk);
173 	if (unlikely(!bh)) {
174 		ext4_error(sb, "Cannot read inode bitmap - "
175 			    "block_group = %u, inode_bitmap = %llu",
176 			    block_group, bitmap_blk);
177 		return ERR_PTR(-EIO);
178 	}
179 	if (bitmap_uptodate(bh))
180 		goto verify;
181 
182 	lock_buffer(bh);
183 	if (bitmap_uptodate(bh)) {
184 		unlock_buffer(bh);
185 		goto verify;
186 	}
187 
188 	ext4_lock_group(sb, block_group);
189 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
190 		err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
191 		set_bitmap_uptodate(bh);
192 		set_buffer_uptodate(bh);
193 		set_buffer_verified(bh);
194 		ext4_unlock_group(sb, block_group);
195 		unlock_buffer(bh);
196 		if (err) {
197 			ext4_error(sb, "Failed to init inode bitmap for group "
198 				   "%u: %d", block_group, err);
199 			goto out;
200 		}
201 		return bh;
202 	}
203 	ext4_unlock_group(sb, block_group);
204 
205 	if (buffer_uptodate(bh)) {
206 		/*
207 		 * if not uninit if bh is uptodate,
208 		 * bitmap is also uptodate
209 		 */
210 		set_bitmap_uptodate(bh);
211 		unlock_buffer(bh);
212 		goto verify;
213 	}
214 	/*
215 	 * submit the buffer_head for reading
216 	 */
217 	trace_ext4_load_inode_bitmap(sb, block_group);
218 	bh->b_end_io = ext4_end_bitmap_read;
219 	get_bh(bh);
220 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
221 	wait_on_buffer(bh);
222 	if (!buffer_uptodate(bh)) {
223 		put_bh(bh);
224 		ext4_error(sb, "Cannot read inode bitmap - "
225 			   "block_group = %u, inode_bitmap = %llu",
226 			   block_group, bitmap_blk);
227 		return ERR_PTR(-EIO);
228 	}
229 
230 verify:
231 	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
232 	if (err)
233 		goto out;
234 	return bh;
235 out:
236 	put_bh(bh);
237 	return ERR_PTR(err);
238 }
239 
240 /*
241  * NOTE! When we get the inode, we're the only people
242  * that have access to it, and as such there are no
243  * race conditions we have to worry about. The inode
244  * is not on the hash-lists, and it cannot be reached
245  * through the filesystem because the directory entry
246  * has been deleted earlier.
247  *
248  * HOWEVER: we must make sure that we get no aliases,
249  * which means that we have to call "clear_inode()"
250  * _before_ we mark the inode not in use in the inode
251  * bitmaps. Otherwise a newly created file might use
252  * the same inode number (not actually the same pointer
253  * though), and then we'd have two inodes sharing the
254  * same inode number and space on the harddisk.
255  */
256 void ext4_free_inode(handle_t *handle, struct inode *inode)
257 {
258 	struct super_block *sb = inode->i_sb;
259 	int is_directory;
260 	unsigned long ino;
261 	struct buffer_head *bitmap_bh = NULL;
262 	struct buffer_head *bh2;
263 	ext4_group_t block_group;
264 	unsigned long bit;
265 	struct ext4_group_desc *gdp;
266 	struct ext4_super_block *es;
267 	struct ext4_sb_info *sbi;
268 	int fatal = 0, err, count, cleared;
269 	struct ext4_group_info *grp;
270 
271 	if (!sb) {
272 		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
273 		       "nonexistent device\n", __func__, __LINE__);
274 		return;
275 	}
276 	if (atomic_read(&inode->i_count) > 1) {
277 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
278 			 __func__, __LINE__, inode->i_ino,
279 			 atomic_read(&inode->i_count));
280 		return;
281 	}
282 	if (inode->i_nlink) {
283 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
284 			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
285 		return;
286 	}
287 	sbi = EXT4_SB(sb);
288 
289 	ino = inode->i_ino;
290 	ext4_debug("freeing inode %lu\n", ino);
291 	trace_ext4_free_inode(inode);
292 
293 	/*
294 	 * Note: we must free any quota before locking the superblock,
295 	 * as writing the quota to disk may need the lock as well.
296 	 */
297 	dquot_initialize(inode);
298 	dquot_free_inode(inode);
299 	dquot_drop(inode);
300 
301 	is_directory = S_ISDIR(inode->i_mode);
302 
303 	/* Do this BEFORE marking the inode not in use or returning an error */
304 	ext4_clear_inode(inode);
305 
306 	es = sbi->s_es;
307 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
308 		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
309 		goto error_return;
310 	}
311 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
312 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
313 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
314 	/* Don't bother if the inode bitmap is corrupt. */
315 	grp = ext4_get_group_info(sb, block_group);
316 	if (IS_ERR(bitmap_bh)) {
317 		fatal = PTR_ERR(bitmap_bh);
318 		bitmap_bh = NULL;
319 		goto error_return;
320 	}
321 	if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
322 		fatal = -EFSCORRUPTED;
323 		goto error_return;
324 	}
325 
326 	BUFFER_TRACE(bitmap_bh, "get_write_access");
327 	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
328 	if (fatal)
329 		goto error_return;
330 
331 	fatal = -ESRCH;
332 	gdp = ext4_get_group_desc(sb, block_group, &bh2);
333 	if (gdp) {
334 		BUFFER_TRACE(bh2, "get_write_access");
335 		fatal = ext4_journal_get_write_access(handle, bh2);
336 	}
337 	ext4_lock_group(sb, block_group);
338 	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
339 	if (fatal || !cleared) {
340 		ext4_unlock_group(sb, block_group);
341 		goto out;
342 	}
343 
344 	count = ext4_free_inodes_count(sb, gdp) + 1;
345 	ext4_free_inodes_set(sb, gdp, count);
346 	if (is_directory) {
347 		count = ext4_used_dirs_count(sb, gdp) - 1;
348 		ext4_used_dirs_set(sb, gdp, count);
349 		percpu_counter_dec(&sbi->s_dirs_counter);
350 	}
351 	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
352 				   EXT4_INODES_PER_GROUP(sb) / 8);
353 	ext4_group_desc_csum_set(sb, block_group, gdp);
354 	ext4_unlock_group(sb, block_group);
355 
356 	percpu_counter_inc(&sbi->s_freeinodes_counter);
357 	if (sbi->s_log_groups_per_flex) {
358 		ext4_group_t f = ext4_flex_group(sbi, block_group);
359 
360 		atomic_inc(&sbi->s_flex_groups[f].free_inodes);
361 		if (is_directory)
362 			atomic_dec(&sbi->s_flex_groups[f].used_dirs);
363 	}
364 	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
365 	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
366 out:
367 	if (cleared) {
368 		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
369 		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
370 		if (!fatal)
371 			fatal = err;
372 	} else {
373 		ext4_error(sb, "bit already cleared for inode %lu", ino);
374 		if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
375 			int count;
376 			count = ext4_free_inodes_count(sb, gdp);
377 			percpu_counter_sub(&sbi->s_freeinodes_counter,
378 					   count);
379 		}
380 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
381 	}
382 
383 error_return:
384 	brelse(bitmap_bh);
385 	ext4_std_error(sb, fatal);
386 }
387 
388 struct orlov_stats {
389 	__u64 free_clusters;
390 	__u32 free_inodes;
391 	__u32 used_dirs;
392 };
393 
394 /*
395  * Helper function for Orlov's allocator; returns critical information
396  * for a particular block group or flex_bg.  If flex_size is 1, then g
397  * is a block group number; otherwise it is flex_bg number.
398  */
399 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
400 			    int flex_size, struct orlov_stats *stats)
401 {
402 	struct ext4_group_desc *desc;
403 	struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
404 
405 	if (flex_size > 1) {
406 		stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
407 		stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
408 		stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
409 		return;
410 	}
411 
412 	desc = ext4_get_group_desc(sb, g, NULL);
413 	if (desc) {
414 		stats->free_inodes = ext4_free_inodes_count(sb, desc);
415 		stats->free_clusters = ext4_free_group_clusters(sb, desc);
416 		stats->used_dirs = ext4_used_dirs_count(sb, desc);
417 	} else {
418 		stats->free_inodes = 0;
419 		stats->free_clusters = 0;
420 		stats->used_dirs = 0;
421 	}
422 }
423 
424 /*
425  * Orlov's allocator for directories.
426  *
427  * We always try to spread first-level directories.
428  *
429  * If there are blockgroups with both free inodes and free blocks counts
430  * not worse than average we return one with smallest directory count.
431  * Otherwise we simply return a random group.
432  *
433  * For the rest rules look so:
434  *
435  * It's OK to put directory into a group unless
436  * it has too many directories already (max_dirs) or
437  * it has too few free inodes left (min_inodes) or
438  * it has too few free blocks left (min_blocks) or
439  * Parent's group is preferred, if it doesn't satisfy these
440  * conditions we search cyclically through the rest. If none
441  * of the groups look good we just look for a group with more
442  * free inodes than average (starting at parent's group).
443  */
444 
445 static int find_group_orlov(struct super_block *sb, struct inode *parent,
446 			    ext4_group_t *group, umode_t mode,
447 			    const struct qstr *qstr)
448 {
449 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
450 	struct ext4_sb_info *sbi = EXT4_SB(sb);
451 	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
452 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
453 	unsigned int freei, avefreei, grp_free;
454 	ext4_fsblk_t freeb, avefreec;
455 	unsigned int ndirs;
456 	int max_dirs, min_inodes;
457 	ext4_grpblk_t min_clusters;
458 	ext4_group_t i, grp, g, ngroups;
459 	struct ext4_group_desc *desc;
460 	struct orlov_stats stats;
461 	int flex_size = ext4_flex_bg_size(sbi);
462 	struct dx_hash_info hinfo;
463 
464 	ngroups = real_ngroups;
465 	if (flex_size > 1) {
466 		ngroups = (real_ngroups + flex_size - 1) >>
467 			sbi->s_log_groups_per_flex;
468 		parent_group >>= sbi->s_log_groups_per_flex;
469 	}
470 
471 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
472 	avefreei = freei / ngroups;
473 	freeb = EXT4_C2B(sbi,
474 		percpu_counter_read_positive(&sbi->s_freeclusters_counter));
475 	avefreec = freeb;
476 	do_div(avefreec, ngroups);
477 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
478 
479 	if (S_ISDIR(mode) &&
480 	    ((parent == d_inode(sb->s_root)) ||
481 	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
482 		int best_ndir = inodes_per_group;
483 		int ret = -1;
484 
485 		if (qstr) {
486 			hinfo.hash_version = DX_HASH_HALF_MD4;
487 			hinfo.seed = sbi->s_hash_seed;
488 			ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
489 			grp = hinfo.hash;
490 		} else
491 			grp = prandom_u32();
492 		parent_group = (unsigned)grp % ngroups;
493 		for (i = 0; i < ngroups; i++) {
494 			g = (parent_group + i) % ngroups;
495 			get_orlov_stats(sb, g, flex_size, &stats);
496 			if (!stats.free_inodes)
497 				continue;
498 			if (stats.used_dirs >= best_ndir)
499 				continue;
500 			if (stats.free_inodes < avefreei)
501 				continue;
502 			if (stats.free_clusters < avefreec)
503 				continue;
504 			grp = g;
505 			ret = 0;
506 			best_ndir = stats.used_dirs;
507 		}
508 		if (ret)
509 			goto fallback;
510 	found_flex_bg:
511 		if (flex_size == 1) {
512 			*group = grp;
513 			return 0;
514 		}
515 
516 		/*
517 		 * We pack inodes at the beginning of the flexgroup's
518 		 * inode tables.  Block allocation decisions will do
519 		 * something similar, although regular files will
520 		 * start at 2nd block group of the flexgroup.  See
521 		 * ext4_ext_find_goal() and ext4_find_near().
522 		 */
523 		grp *= flex_size;
524 		for (i = 0; i < flex_size; i++) {
525 			if (grp+i >= real_ngroups)
526 				break;
527 			desc = ext4_get_group_desc(sb, grp+i, NULL);
528 			if (desc && ext4_free_inodes_count(sb, desc)) {
529 				*group = grp+i;
530 				return 0;
531 			}
532 		}
533 		goto fallback;
534 	}
535 
536 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
537 	min_inodes = avefreei - inodes_per_group*flex_size / 4;
538 	if (min_inodes < 1)
539 		min_inodes = 1;
540 	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
541 
542 	/*
543 	 * Start looking in the flex group where we last allocated an
544 	 * inode for this parent directory
545 	 */
546 	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
547 		parent_group = EXT4_I(parent)->i_last_alloc_group;
548 		if (flex_size > 1)
549 			parent_group >>= sbi->s_log_groups_per_flex;
550 	}
551 
552 	for (i = 0; i < ngroups; i++) {
553 		grp = (parent_group + i) % ngroups;
554 		get_orlov_stats(sb, grp, flex_size, &stats);
555 		if (stats.used_dirs >= max_dirs)
556 			continue;
557 		if (stats.free_inodes < min_inodes)
558 			continue;
559 		if (stats.free_clusters < min_clusters)
560 			continue;
561 		goto found_flex_bg;
562 	}
563 
564 fallback:
565 	ngroups = real_ngroups;
566 	avefreei = freei / ngroups;
567 fallback_retry:
568 	parent_group = EXT4_I(parent)->i_block_group;
569 	for (i = 0; i < ngroups; i++) {
570 		grp = (parent_group + i) % ngroups;
571 		desc = ext4_get_group_desc(sb, grp, NULL);
572 		if (desc) {
573 			grp_free = ext4_free_inodes_count(sb, desc);
574 			if (grp_free && grp_free >= avefreei) {
575 				*group = grp;
576 				return 0;
577 			}
578 		}
579 	}
580 
581 	if (avefreei) {
582 		/*
583 		 * The free-inodes counter is approximate, and for really small
584 		 * filesystems the above test can fail to find any blockgroups
585 		 */
586 		avefreei = 0;
587 		goto fallback_retry;
588 	}
589 
590 	return -1;
591 }
592 
593 static int find_group_other(struct super_block *sb, struct inode *parent,
594 			    ext4_group_t *group, umode_t mode)
595 {
596 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
597 	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
598 	struct ext4_group_desc *desc;
599 	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
600 
601 	/*
602 	 * Try to place the inode is the same flex group as its
603 	 * parent.  If we can't find space, use the Orlov algorithm to
604 	 * find another flex group, and store that information in the
605 	 * parent directory's inode information so that use that flex
606 	 * group for future allocations.
607 	 */
608 	if (flex_size > 1) {
609 		int retry = 0;
610 
611 	try_again:
612 		parent_group &= ~(flex_size-1);
613 		last = parent_group + flex_size;
614 		if (last > ngroups)
615 			last = ngroups;
616 		for  (i = parent_group; i < last; i++) {
617 			desc = ext4_get_group_desc(sb, i, NULL);
618 			if (desc && ext4_free_inodes_count(sb, desc)) {
619 				*group = i;
620 				return 0;
621 			}
622 		}
623 		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
624 			retry = 1;
625 			parent_group = EXT4_I(parent)->i_last_alloc_group;
626 			goto try_again;
627 		}
628 		/*
629 		 * If this didn't work, use the Orlov search algorithm
630 		 * to find a new flex group; we pass in the mode to
631 		 * avoid the topdir algorithms.
632 		 */
633 		*group = parent_group + flex_size;
634 		if (*group > ngroups)
635 			*group = 0;
636 		return find_group_orlov(sb, parent, group, mode, NULL);
637 	}
638 
639 	/*
640 	 * Try to place the inode in its parent directory
641 	 */
642 	*group = parent_group;
643 	desc = ext4_get_group_desc(sb, *group, NULL);
644 	if (desc && ext4_free_inodes_count(sb, desc) &&
645 	    ext4_free_group_clusters(sb, desc))
646 		return 0;
647 
648 	/*
649 	 * We're going to place this inode in a different blockgroup from its
650 	 * parent.  We want to cause files in a common directory to all land in
651 	 * the same blockgroup.  But we want files which are in a different
652 	 * directory which shares a blockgroup with our parent to land in a
653 	 * different blockgroup.
654 	 *
655 	 * So add our directory's i_ino into the starting point for the hash.
656 	 */
657 	*group = (*group + parent->i_ino) % ngroups;
658 
659 	/*
660 	 * Use a quadratic hash to find a group with a free inode and some free
661 	 * blocks.
662 	 */
663 	for (i = 1; i < ngroups; i <<= 1) {
664 		*group += i;
665 		if (*group >= ngroups)
666 			*group -= ngroups;
667 		desc = ext4_get_group_desc(sb, *group, NULL);
668 		if (desc && ext4_free_inodes_count(sb, desc) &&
669 		    ext4_free_group_clusters(sb, desc))
670 			return 0;
671 	}
672 
673 	/*
674 	 * That failed: try linear search for a free inode, even if that group
675 	 * has no free blocks.
676 	 */
677 	*group = parent_group;
678 	for (i = 0; i < ngroups; i++) {
679 		if (++*group >= ngroups)
680 			*group = 0;
681 		desc = ext4_get_group_desc(sb, *group, NULL);
682 		if (desc && ext4_free_inodes_count(sb, desc))
683 			return 0;
684 	}
685 
686 	return -1;
687 }
688 
689 /*
690  * In no journal mode, if an inode has recently been deleted, we want
691  * to avoid reusing it until we're reasonably sure the inode table
692  * block has been written back to disk.  (Yes, these values are
693  * somewhat arbitrary...)
694  */
695 #define RECENTCY_MIN	5
696 #define RECENTCY_DIRTY	300
697 
698 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
699 {
700 	struct ext4_group_desc	*gdp;
701 	struct ext4_inode	*raw_inode;
702 	struct buffer_head	*bh;
703 	int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
704 	int offset, ret = 0;
705 	int recentcy = RECENTCY_MIN;
706 	u32 dtime, now;
707 
708 	gdp = ext4_get_group_desc(sb, group, NULL);
709 	if (unlikely(!gdp))
710 		return 0;
711 
712 	bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
713 		       (ino / inodes_per_block));
714 	if (!bh || !buffer_uptodate(bh))
715 		/*
716 		 * If the block is not in the buffer cache, then it
717 		 * must have been written out.
718 		 */
719 		goto out;
720 
721 	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
722 	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
723 
724 	/* i_dtime is only 32 bits on disk, but we only care about relative
725 	 * times in the range of a few minutes (i.e. long enough to sync a
726 	 * recently-deleted inode to disk), so using the low 32 bits of the
727 	 * clock (a 68 year range) is enough, see time_before32() */
728 	dtime = le32_to_cpu(raw_inode->i_dtime);
729 	now = ktime_get_real_seconds();
730 	if (buffer_dirty(bh))
731 		recentcy += RECENTCY_DIRTY;
732 
733 	if (dtime && time_before32(dtime, now) &&
734 	    time_before32(now, dtime + recentcy))
735 		ret = 1;
736 out:
737 	brelse(bh);
738 	return ret;
739 }
740 
741 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
742 			  struct buffer_head *bitmap, unsigned long *ino)
743 {
744 next:
745 	*ino = ext4_find_next_zero_bit((unsigned long *)
746 				       bitmap->b_data,
747 				       EXT4_INODES_PER_GROUP(sb), *ino);
748 	if (*ino >= EXT4_INODES_PER_GROUP(sb))
749 		return 0;
750 
751 	if ((EXT4_SB(sb)->s_journal == NULL) &&
752 	    recently_deleted(sb, group, *ino)) {
753 		*ino = *ino + 1;
754 		if (*ino < EXT4_INODES_PER_GROUP(sb))
755 			goto next;
756 		return 0;
757 	}
758 
759 	return 1;
760 }
761 
762 /*
763  * There are two policies for allocating an inode.  If the new inode is
764  * a directory, then a forward search is made for a block group with both
765  * free space and a low directory-to-inode ratio; if that fails, then of
766  * the groups with above-average free space, that group with the fewest
767  * directories already is chosen.
768  *
769  * For other inodes, search forward from the parent directory's block
770  * group to find a free inode.
771  */
772 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
773 			       umode_t mode, const struct qstr *qstr,
774 			       __u32 goal, uid_t *owner, __u32 i_flags,
775 			       int handle_type, unsigned int line_no,
776 			       int nblocks)
777 {
778 	struct super_block *sb;
779 	struct buffer_head *inode_bitmap_bh = NULL;
780 	struct buffer_head *group_desc_bh;
781 	ext4_group_t ngroups, group = 0;
782 	unsigned long ino = 0;
783 	struct inode *inode;
784 	struct ext4_group_desc *gdp = NULL;
785 	struct ext4_inode_info *ei;
786 	struct ext4_sb_info *sbi;
787 	int ret2, err;
788 	struct inode *ret;
789 	ext4_group_t i;
790 	ext4_group_t flex_group;
791 	struct ext4_group_info *grp;
792 	int encrypt = 0;
793 
794 	/* Cannot create files in a deleted directory */
795 	if (!dir || !dir->i_nlink)
796 		return ERR_PTR(-EPERM);
797 
798 	sb = dir->i_sb;
799 	sbi = EXT4_SB(sb);
800 
801 	if (unlikely(ext4_forced_shutdown(sbi)))
802 		return ERR_PTR(-EIO);
803 
804 	if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
805 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
806 	    !(i_flags & EXT4_EA_INODE_FL)) {
807 		err = fscrypt_get_encryption_info(dir);
808 		if (err)
809 			return ERR_PTR(err);
810 		if (!fscrypt_has_encryption_key(dir))
811 			return ERR_PTR(-ENOKEY);
812 		encrypt = 1;
813 	}
814 
815 	if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
816 #ifdef CONFIG_EXT4_FS_POSIX_ACL
817 		struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
818 
819 		if (IS_ERR(p))
820 			return ERR_CAST(p);
821 		if (p) {
822 			int acl_size = p->a_count * sizeof(ext4_acl_entry);
823 
824 			nblocks += (S_ISDIR(mode) ? 2 : 1) *
825 				__ext4_xattr_set_credits(sb, NULL /* inode */,
826 					NULL /* block_bh */, acl_size,
827 					true /* is_create */);
828 			posix_acl_release(p);
829 		}
830 #endif
831 
832 #ifdef CONFIG_SECURITY
833 		{
834 			int num_security_xattrs = 1;
835 
836 #ifdef CONFIG_INTEGRITY
837 			num_security_xattrs++;
838 #endif
839 			/*
840 			 * We assume that security xattrs are never
841 			 * more than 1k.  In practice they are under
842 			 * 128 bytes.
843 			 */
844 			nblocks += num_security_xattrs *
845 				__ext4_xattr_set_credits(sb, NULL /* inode */,
846 					NULL /* block_bh */, 1024,
847 					true /* is_create */);
848 		}
849 #endif
850 		if (encrypt)
851 			nblocks += __ext4_xattr_set_credits(sb,
852 					NULL /* inode */, NULL /* block_bh */,
853 					FSCRYPT_SET_CONTEXT_MAX_SIZE,
854 					true /* is_create */);
855 	}
856 
857 	ngroups = ext4_get_groups_count(sb);
858 	trace_ext4_request_inode(dir, mode);
859 	inode = new_inode(sb);
860 	if (!inode)
861 		return ERR_PTR(-ENOMEM);
862 	ei = EXT4_I(inode);
863 
864 	/*
865 	 * Initialize owners and quota early so that we don't have to account
866 	 * for quota initialization worst case in standard inode creating
867 	 * transaction
868 	 */
869 	if (owner) {
870 		inode->i_mode = mode;
871 		i_uid_write(inode, owner[0]);
872 		i_gid_write(inode, owner[1]);
873 	} else if (test_opt(sb, GRPID)) {
874 		inode->i_mode = mode;
875 		inode->i_uid = current_fsuid();
876 		inode->i_gid = dir->i_gid;
877 	} else
878 		inode_init_owner(inode, dir, mode);
879 
880 	if (ext4_has_feature_project(sb) &&
881 	    ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
882 		ei->i_projid = EXT4_I(dir)->i_projid;
883 	else
884 		ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
885 
886 	err = dquot_initialize(inode);
887 	if (err)
888 		goto out;
889 
890 	if (!goal)
891 		goal = sbi->s_inode_goal;
892 
893 	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
894 		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
895 		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
896 		ret2 = 0;
897 		goto got_group;
898 	}
899 
900 	if (S_ISDIR(mode))
901 		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
902 	else
903 		ret2 = find_group_other(sb, dir, &group, mode);
904 
905 got_group:
906 	EXT4_I(dir)->i_last_alloc_group = group;
907 	err = -ENOSPC;
908 	if (ret2 == -1)
909 		goto out;
910 
911 	/*
912 	 * Normally we will only go through one pass of this loop,
913 	 * unless we get unlucky and it turns out the group we selected
914 	 * had its last inode grabbed by someone else.
915 	 */
916 	for (i = 0; i < ngroups; i++, ino = 0) {
917 		err = -EIO;
918 
919 		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
920 		if (!gdp)
921 			goto out;
922 
923 		/*
924 		 * Check free inodes count before loading bitmap.
925 		 */
926 		if (ext4_free_inodes_count(sb, gdp) == 0)
927 			goto next_group;
928 
929 		grp = ext4_get_group_info(sb, group);
930 		/* Skip groups with already-known suspicious inode tables */
931 		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
932 			goto next_group;
933 
934 		brelse(inode_bitmap_bh);
935 		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
936 		/* Skip groups with suspicious inode tables */
937 		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
938 		    IS_ERR(inode_bitmap_bh)) {
939 			inode_bitmap_bh = NULL;
940 			goto next_group;
941 		}
942 
943 repeat_in_this_group:
944 		ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
945 		if (!ret2)
946 			goto next_group;
947 
948 		if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
949 			ext4_error(sb, "reserved inode found cleared - "
950 				   "inode=%lu", ino + 1);
951 			goto next_group;
952 		}
953 
954 		if (!handle) {
955 			BUG_ON(nblocks <= 0);
956 			handle = __ext4_journal_start_sb(dir->i_sb, line_no,
957 							 handle_type, nblocks,
958 							 0);
959 			if (IS_ERR(handle)) {
960 				err = PTR_ERR(handle);
961 				ext4_std_error(sb, err);
962 				goto out;
963 			}
964 		}
965 		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
966 		err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
967 		if (err) {
968 			ext4_std_error(sb, err);
969 			goto out;
970 		}
971 		ext4_lock_group(sb, group);
972 		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
973 		if (ret2) {
974 			/* Someone already took the bit. Repeat the search
975 			 * with lock held.
976 			 */
977 			ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
978 			if (ret2) {
979 				ext4_set_bit(ino, inode_bitmap_bh->b_data);
980 				ret2 = 0;
981 			} else {
982 				ret2 = 1; /* we didn't grab the inode */
983 			}
984 		}
985 		ext4_unlock_group(sb, group);
986 		ino++;		/* the inode bitmap is zero-based */
987 		if (!ret2)
988 			goto got; /* we grabbed the inode! */
989 
990 		if (ino < EXT4_INODES_PER_GROUP(sb))
991 			goto repeat_in_this_group;
992 next_group:
993 		if (++group == ngroups)
994 			group = 0;
995 	}
996 	err = -ENOSPC;
997 	goto out;
998 
999 got:
1000 	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1001 	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1002 	if (err) {
1003 		ext4_std_error(sb, err);
1004 		goto out;
1005 	}
1006 
1007 	BUFFER_TRACE(group_desc_bh, "get_write_access");
1008 	err = ext4_journal_get_write_access(handle, group_desc_bh);
1009 	if (err) {
1010 		ext4_std_error(sb, err);
1011 		goto out;
1012 	}
1013 
1014 	/* We may have to initialize the block bitmap if it isn't already */
1015 	if (ext4_has_group_desc_csum(sb) &&
1016 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1017 		struct buffer_head *block_bitmap_bh;
1018 
1019 		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1020 		if (IS_ERR(block_bitmap_bh)) {
1021 			err = PTR_ERR(block_bitmap_bh);
1022 			goto out;
1023 		}
1024 		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1025 		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1026 		if (err) {
1027 			brelse(block_bitmap_bh);
1028 			ext4_std_error(sb, err);
1029 			goto out;
1030 		}
1031 
1032 		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1033 		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1034 
1035 		/* recheck and clear flag under lock if we still need to */
1036 		ext4_lock_group(sb, group);
1037 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1038 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1039 			ext4_free_group_clusters_set(sb, gdp,
1040 				ext4_free_clusters_after_init(sb, group, gdp));
1041 			ext4_block_bitmap_csum_set(sb, group, gdp,
1042 						   block_bitmap_bh);
1043 			ext4_group_desc_csum_set(sb, group, gdp);
1044 		}
1045 		ext4_unlock_group(sb, group);
1046 		brelse(block_bitmap_bh);
1047 
1048 		if (err) {
1049 			ext4_std_error(sb, err);
1050 			goto out;
1051 		}
1052 	}
1053 
1054 	/* Update the relevant bg descriptor fields */
1055 	if (ext4_has_group_desc_csum(sb)) {
1056 		int free;
1057 		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1058 
1059 		down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1060 		ext4_lock_group(sb, group); /* while we modify the bg desc */
1061 		free = EXT4_INODES_PER_GROUP(sb) -
1062 			ext4_itable_unused_count(sb, gdp);
1063 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1064 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1065 			free = 0;
1066 		}
1067 		/*
1068 		 * Check the relative inode number against the last used
1069 		 * relative inode number in this group. if it is greater
1070 		 * we need to update the bg_itable_unused count
1071 		 */
1072 		if (ino > free)
1073 			ext4_itable_unused_set(sb, gdp,
1074 					(EXT4_INODES_PER_GROUP(sb) - ino));
1075 		up_read(&grp->alloc_sem);
1076 	} else {
1077 		ext4_lock_group(sb, group);
1078 	}
1079 
1080 	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1081 	if (S_ISDIR(mode)) {
1082 		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1083 		if (sbi->s_log_groups_per_flex) {
1084 			ext4_group_t f = ext4_flex_group(sbi, group);
1085 
1086 			atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1087 		}
1088 	}
1089 	if (ext4_has_group_desc_csum(sb)) {
1090 		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1091 					   EXT4_INODES_PER_GROUP(sb) / 8);
1092 		ext4_group_desc_csum_set(sb, group, gdp);
1093 	}
1094 	ext4_unlock_group(sb, group);
1095 
1096 	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1097 	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1098 	if (err) {
1099 		ext4_std_error(sb, err);
1100 		goto out;
1101 	}
1102 
1103 	percpu_counter_dec(&sbi->s_freeinodes_counter);
1104 	if (S_ISDIR(mode))
1105 		percpu_counter_inc(&sbi->s_dirs_counter);
1106 
1107 	if (sbi->s_log_groups_per_flex) {
1108 		flex_group = ext4_flex_group(sbi, group);
1109 		atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1110 	}
1111 
1112 	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1113 	/* This is the optimal IO size (for stat), not the fs block size */
1114 	inode->i_blocks = 0;
1115 	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1116 						       current_time(inode);
1117 
1118 	memset(ei->i_data, 0, sizeof(ei->i_data));
1119 	ei->i_dir_start_lookup = 0;
1120 	ei->i_disksize = 0;
1121 
1122 	/* Don't inherit extent flag from directory, amongst others. */
1123 	ei->i_flags =
1124 		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1125 	ei->i_flags |= i_flags;
1126 	ei->i_file_acl = 0;
1127 	ei->i_dtime = 0;
1128 	ei->i_block_group = group;
1129 	ei->i_last_alloc_group = ~0;
1130 
1131 	ext4_set_inode_flags(inode);
1132 	if (IS_DIRSYNC(inode))
1133 		ext4_handle_sync(handle);
1134 	if (insert_inode_locked(inode) < 0) {
1135 		/*
1136 		 * Likely a bitmap corruption causing inode to be allocated
1137 		 * twice.
1138 		 */
1139 		err = -EIO;
1140 		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1141 			   inode->i_ino);
1142 		goto out;
1143 	}
1144 	inode->i_generation = prandom_u32();
1145 
1146 	/* Precompute checksum seed for inode metadata */
1147 	if (ext4_has_metadata_csum(sb)) {
1148 		__u32 csum;
1149 		__le32 inum = cpu_to_le32(inode->i_ino);
1150 		__le32 gen = cpu_to_le32(inode->i_generation);
1151 		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1152 				   sizeof(inum));
1153 		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1154 					      sizeof(gen));
1155 	}
1156 
1157 	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1158 	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1159 
1160 	ei->i_extra_isize = sbi->s_want_extra_isize;
1161 	ei->i_inline_off = 0;
1162 	if (ext4_has_feature_inline_data(sb))
1163 		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1164 	ret = inode;
1165 	err = dquot_alloc_inode(inode);
1166 	if (err)
1167 		goto fail_drop;
1168 
1169 	/*
1170 	 * Since the encryption xattr will always be unique, create it first so
1171 	 * that it's less likely to end up in an external xattr block and
1172 	 * prevent its deduplication.
1173 	 */
1174 	if (encrypt) {
1175 		err = fscrypt_inherit_context(dir, inode, handle, true);
1176 		if (err)
1177 			goto fail_free_drop;
1178 	}
1179 
1180 	if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1181 		err = ext4_init_acl(handle, inode, dir);
1182 		if (err)
1183 			goto fail_free_drop;
1184 
1185 		err = ext4_init_security(handle, inode, dir, qstr);
1186 		if (err)
1187 			goto fail_free_drop;
1188 	}
1189 
1190 	if (ext4_has_feature_extents(sb)) {
1191 		/* set extent flag only for directory, file and normal symlink*/
1192 		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1193 			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1194 			ext4_ext_tree_init(handle, inode);
1195 		}
1196 	}
1197 
1198 	if (ext4_handle_valid(handle)) {
1199 		ei->i_sync_tid = handle->h_transaction->t_tid;
1200 		ei->i_datasync_tid = handle->h_transaction->t_tid;
1201 	}
1202 
1203 	err = ext4_mark_inode_dirty(handle, inode);
1204 	if (err) {
1205 		ext4_std_error(sb, err);
1206 		goto fail_free_drop;
1207 	}
1208 
1209 	ext4_debug("allocating inode %lu\n", inode->i_ino);
1210 	trace_ext4_allocate_inode(inode, dir, mode);
1211 	brelse(inode_bitmap_bh);
1212 	return ret;
1213 
1214 fail_free_drop:
1215 	dquot_free_inode(inode);
1216 fail_drop:
1217 	clear_nlink(inode);
1218 	unlock_new_inode(inode);
1219 out:
1220 	dquot_drop(inode);
1221 	inode->i_flags |= S_NOQUOTA;
1222 	iput(inode);
1223 	brelse(inode_bitmap_bh);
1224 	return ERR_PTR(err);
1225 }
1226 
1227 /* Verify that we are loading a valid orphan from disk */
1228 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1229 {
1230 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1231 	ext4_group_t block_group;
1232 	int bit;
1233 	struct buffer_head *bitmap_bh = NULL;
1234 	struct inode *inode = NULL;
1235 	int err = -EFSCORRUPTED;
1236 
1237 	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1238 		goto bad_orphan;
1239 
1240 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1241 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1242 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1243 	if (IS_ERR(bitmap_bh)) {
1244 		ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1245 			   ino, PTR_ERR(bitmap_bh));
1246 		return (struct inode *) bitmap_bh;
1247 	}
1248 
1249 	/* Having the inode bit set should be a 100% indicator that this
1250 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1251 	 * inodes that were being truncated, so we can't check i_nlink==0.
1252 	 */
1253 	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1254 		goto bad_orphan;
1255 
1256 	inode = ext4_iget(sb, ino);
1257 	if (IS_ERR(inode)) {
1258 		err = PTR_ERR(inode);
1259 		ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1260 			   ino, err);
1261 		return inode;
1262 	}
1263 
1264 	/*
1265 	 * If the orphans has i_nlinks > 0 then it should be able to
1266 	 * be truncated, otherwise it won't be removed from the orphan
1267 	 * list during processing and an infinite loop will result.
1268 	 * Similarly, it must not be a bad inode.
1269 	 */
1270 	if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1271 	    is_bad_inode(inode))
1272 		goto bad_orphan;
1273 
1274 	if (NEXT_ORPHAN(inode) > max_ino)
1275 		goto bad_orphan;
1276 	brelse(bitmap_bh);
1277 	return inode;
1278 
1279 bad_orphan:
1280 	ext4_error(sb, "bad orphan inode %lu", ino);
1281 	if (bitmap_bh)
1282 		printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1283 		       bit, (unsigned long long)bitmap_bh->b_blocknr,
1284 		       ext4_test_bit(bit, bitmap_bh->b_data));
1285 	if (inode) {
1286 		printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1287 		       is_bad_inode(inode));
1288 		printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1289 		       NEXT_ORPHAN(inode));
1290 		printk(KERN_ERR "max_ino=%lu\n", max_ino);
1291 		printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1292 		/* Avoid freeing blocks if we got a bad deleted inode */
1293 		if (inode->i_nlink == 0)
1294 			inode->i_blocks = 0;
1295 		iput(inode);
1296 	}
1297 	brelse(bitmap_bh);
1298 	return ERR_PTR(err);
1299 }
1300 
1301 unsigned long ext4_count_free_inodes(struct super_block *sb)
1302 {
1303 	unsigned long desc_count;
1304 	struct ext4_group_desc *gdp;
1305 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1306 #ifdef EXT4FS_DEBUG
1307 	struct ext4_super_block *es;
1308 	unsigned long bitmap_count, x;
1309 	struct buffer_head *bitmap_bh = NULL;
1310 
1311 	es = EXT4_SB(sb)->s_es;
1312 	desc_count = 0;
1313 	bitmap_count = 0;
1314 	gdp = NULL;
1315 	for (i = 0; i < ngroups; i++) {
1316 		gdp = ext4_get_group_desc(sb, i, NULL);
1317 		if (!gdp)
1318 			continue;
1319 		desc_count += ext4_free_inodes_count(sb, gdp);
1320 		brelse(bitmap_bh);
1321 		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1322 		if (IS_ERR(bitmap_bh)) {
1323 			bitmap_bh = NULL;
1324 			continue;
1325 		}
1326 
1327 		x = ext4_count_free(bitmap_bh->b_data,
1328 				    EXT4_INODES_PER_GROUP(sb) / 8);
1329 		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1330 			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1331 		bitmap_count += x;
1332 	}
1333 	brelse(bitmap_bh);
1334 	printk(KERN_DEBUG "ext4_count_free_inodes: "
1335 	       "stored = %u, computed = %lu, %lu\n",
1336 	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1337 	return desc_count;
1338 #else
1339 	desc_count = 0;
1340 	for (i = 0; i < ngroups; i++) {
1341 		gdp = ext4_get_group_desc(sb, i, NULL);
1342 		if (!gdp)
1343 			continue;
1344 		desc_count += ext4_free_inodes_count(sb, gdp);
1345 		cond_resched();
1346 	}
1347 	return desc_count;
1348 #endif
1349 }
1350 
1351 /* Called at mount-time, super-block is locked */
1352 unsigned long ext4_count_dirs(struct super_block * sb)
1353 {
1354 	unsigned long count = 0;
1355 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1356 
1357 	for (i = 0; i < ngroups; i++) {
1358 		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1359 		if (!gdp)
1360 			continue;
1361 		count += ext4_used_dirs_count(sb, gdp);
1362 	}
1363 	return count;
1364 }
1365 
1366 /*
1367  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1368  * inode table. Must be called without any spinlock held. The only place
1369  * where it is called from on active part of filesystem is ext4lazyinit
1370  * thread, so we do not need any special locks, however we have to prevent
1371  * inode allocation from the current group, so we take alloc_sem lock, to
1372  * block ext4_new_inode() until we are finished.
1373  */
1374 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1375 				 int barrier)
1376 {
1377 	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1378 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1379 	struct ext4_group_desc *gdp = NULL;
1380 	struct buffer_head *group_desc_bh;
1381 	handle_t *handle;
1382 	ext4_fsblk_t blk;
1383 	int num, ret = 0, used_blks = 0;
1384 
1385 	/* This should not happen, but just to be sure check this */
1386 	if (sb_rdonly(sb)) {
1387 		ret = 1;
1388 		goto out;
1389 	}
1390 
1391 	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1392 	if (!gdp)
1393 		goto out;
1394 
1395 	/*
1396 	 * We do not need to lock this, because we are the only one
1397 	 * handling this flag.
1398 	 */
1399 	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1400 		goto out;
1401 
1402 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1403 	if (IS_ERR(handle)) {
1404 		ret = PTR_ERR(handle);
1405 		goto out;
1406 	}
1407 
1408 	down_write(&grp->alloc_sem);
1409 	/*
1410 	 * If inode bitmap was already initialized there may be some
1411 	 * used inodes so we need to skip blocks with used inodes in
1412 	 * inode table.
1413 	 */
1414 	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1415 		used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1416 			    ext4_itable_unused_count(sb, gdp)),
1417 			    sbi->s_inodes_per_block);
1418 
1419 	if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1420 		ext4_error(sb, "Something is wrong with group %u: "
1421 			   "used itable blocks: %d; "
1422 			   "itable unused count: %u",
1423 			   group, used_blks,
1424 			   ext4_itable_unused_count(sb, gdp));
1425 		ret = 1;
1426 		goto err_out;
1427 	}
1428 
1429 	blk = ext4_inode_table(sb, gdp) + used_blks;
1430 	num = sbi->s_itb_per_group - used_blks;
1431 
1432 	BUFFER_TRACE(group_desc_bh, "get_write_access");
1433 	ret = ext4_journal_get_write_access(handle,
1434 					    group_desc_bh);
1435 	if (ret)
1436 		goto err_out;
1437 
1438 	/*
1439 	 * Skip zeroout if the inode table is full. But we set the ZEROED
1440 	 * flag anyway, because obviously, when it is full it does not need
1441 	 * further zeroing.
1442 	 */
1443 	if (unlikely(num == 0))
1444 		goto skip_zeroout;
1445 
1446 	ext4_debug("going to zero out inode table in group %d\n",
1447 		   group);
1448 	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1449 	if (ret < 0)
1450 		goto err_out;
1451 	if (barrier)
1452 		blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1453 
1454 skip_zeroout:
1455 	ext4_lock_group(sb, group);
1456 	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1457 	ext4_group_desc_csum_set(sb, group, gdp);
1458 	ext4_unlock_group(sb, group);
1459 
1460 	BUFFER_TRACE(group_desc_bh,
1461 		     "call ext4_handle_dirty_metadata");
1462 	ret = ext4_handle_dirty_metadata(handle, NULL,
1463 					 group_desc_bh);
1464 
1465 err_out:
1466 	up_write(&grp->alloc_sem);
1467 	ext4_journal_stop(handle);
1468 out:
1469 	return ret;
1470 }
1471