xref: /openbmc/linux/fs/ext4/ialloc.c (revision b7ec62d7)
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 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70 {
71 	if (uptodate) {
72 		set_buffer_uptodate(bh);
73 		set_bitmap_uptodate(bh);
74 	}
75 	unlock_buffer(bh);
76 	put_bh(bh);
77 }
78 
79 static int ext4_validate_inode_bitmap(struct super_block *sb,
80 				      struct ext4_group_desc *desc,
81 				      ext4_group_t block_group,
82 				      struct buffer_head *bh)
83 {
84 	ext4_fsblk_t	blk;
85 	struct ext4_group_info *grp;
86 
87 	if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 		return 0;
89 
90 	grp = ext4_get_group_info(sb, block_group);
91 
92 	if (buffer_verified(bh))
93 		return 0;
94 	if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 		return -EFSCORRUPTED;
96 
97 	ext4_lock_group(sb, block_group);
98 	if (buffer_verified(bh))
99 		goto verified;
100 	blk = ext4_inode_bitmap(sb, desc);
101 	if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
102 					   EXT4_INODES_PER_GROUP(sb) / 8) ||
103 	    ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
104 		ext4_unlock_group(sb, block_group);
105 		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
106 			   "inode_bitmap = %llu", block_group, blk);
107 		ext4_mark_group_bitmap_corrupted(sb, block_group,
108 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
109 		return -EFSBADCRC;
110 	}
111 	set_buffer_verified(bh);
112 verified:
113 	ext4_unlock_group(sb, block_group);
114 	return 0;
115 }
116 
117 /*
118  * Read the inode allocation bitmap for a given block_group, reading
119  * into the specified slot in the superblock's bitmap cache.
120  *
121  * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122  */
123 static struct buffer_head *
124 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125 {
126 	struct ext4_group_desc *desc;
127 	struct ext4_sb_info *sbi = EXT4_SB(sb);
128 	struct buffer_head *bh = NULL;
129 	ext4_fsblk_t bitmap_blk;
130 	int err;
131 
132 	desc = ext4_get_group_desc(sb, block_group, NULL);
133 	if (!desc)
134 		return ERR_PTR(-EFSCORRUPTED);
135 
136 	bitmap_blk = ext4_inode_bitmap(sb, desc);
137 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 		ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 			   "block_group %u", bitmap_blk, block_group);
141 		ext4_mark_group_bitmap_corrupted(sb, block_group,
142 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 		return ERR_PTR(-EFSCORRUPTED);
144 	}
145 	bh = sb_getblk(sb, bitmap_blk);
146 	if (unlikely(!bh)) {
147 		ext4_warning(sb, "Cannot read inode bitmap - "
148 			     "block_group = %u, inode_bitmap = %llu",
149 			     block_group, bitmap_blk);
150 		return ERR_PTR(-ENOMEM);
151 	}
152 	if (bitmap_uptodate(bh))
153 		goto verify;
154 
155 	lock_buffer(bh);
156 	if (bitmap_uptodate(bh)) {
157 		unlock_buffer(bh);
158 		goto verify;
159 	}
160 
161 	ext4_lock_group(sb, block_group);
162 	if (ext4_has_group_desc_csum(sb) &&
163 	    (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 		if (block_group == 0) {
165 			ext4_unlock_group(sb, block_group);
166 			unlock_buffer(bh);
167 			ext4_error(sb, "Inode bitmap for bg 0 marked "
168 				   "uninitialized");
169 			err = -EFSCORRUPTED;
170 			goto out;
171 		}
172 		memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 		ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 				     sb->s_blocksize * 8, bh->b_data);
175 		set_bitmap_uptodate(bh);
176 		set_buffer_uptodate(bh);
177 		set_buffer_verified(bh);
178 		ext4_unlock_group(sb, block_group);
179 		unlock_buffer(bh);
180 		return bh;
181 	}
182 	ext4_unlock_group(sb, block_group);
183 
184 	if (buffer_uptodate(bh)) {
185 		/*
186 		 * if not uninit if bh is uptodate,
187 		 * bitmap is also uptodate
188 		 */
189 		set_bitmap_uptodate(bh);
190 		unlock_buffer(bh);
191 		goto verify;
192 	}
193 	/*
194 	 * submit the buffer_head for reading
195 	 */
196 	trace_ext4_load_inode_bitmap(sb, block_group);
197 	ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 	if (!buffer_uptodate(bh)) {
200 		put_bh(bh);
201 		ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 			       "block_group = %u, inode_bitmap = %llu",
203 			       block_group, bitmap_blk);
204 		ext4_mark_group_bitmap_corrupted(sb, block_group,
205 				EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 		return ERR_PTR(-EIO);
207 	}
208 
209 verify:
210 	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 	if (err)
212 		goto out;
213 	return bh;
214 out:
215 	put_bh(bh);
216 	return ERR_PTR(err);
217 }
218 
219 /*
220  * NOTE! When we get the inode, we're the only people
221  * that have access to it, and as such there are no
222  * race conditions we have to worry about. The inode
223  * is not on the hash-lists, and it cannot be reached
224  * through the filesystem because the directory entry
225  * has been deleted earlier.
226  *
227  * HOWEVER: we must make sure that we get no aliases,
228  * which means that we have to call "clear_inode()"
229  * _before_ we mark the inode not in use in the inode
230  * bitmaps. Otherwise a newly created file might use
231  * the same inode number (not actually the same pointer
232  * though), and then we'd have two inodes sharing the
233  * same inode number and space on the harddisk.
234  */
235 void ext4_free_inode(handle_t *handle, struct inode *inode)
236 {
237 	struct super_block *sb = inode->i_sb;
238 	int is_directory;
239 	unsigned long ino;
240 	struct buffer_head *bitmap_bh = NULL;
241 	struct buffer_head *bh2;
242 	ext4_group_t block_group;
243 	unsigned long bit;
244 	struct ext4_group_desc *gdp;
245 	struct ext4_super_block *es;
246 	struct ext4_sb_info *sbi;
247 	int fatal = 0, err, count, cleared;
248 	struct ext4_group_info *grp;
249 
250 	if (!sb) {
251 		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 		       "nonexistent device\n", __func__, __LINE__);
253 		return;
254 	}
255 	if (atomic_read(&inode->i_count) > 1) {
256 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 			 __func__, __LINE__, inode->i_ino,
258 			 atomic_read(&inode->i_count));
259 		return;
260 	}
261 	if (inode->i_nlink) {
262 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 		return;
265 	}
266 	sbi = EXT4_SB(sb);
267 
268 	ino = inode->i_ino;
269 	ext4_debug("freeing inode %lu\n", ino);
270 	trace_ext4_free_inode(inode);
271 
272 	dquot_initialize(inode);
273 	dquot_free_inode(inode);
274 
275 	is_directory = S_ISDIR(inode->i_mode);
276 
277 	/* Do this BEFORE marking the inode not in use or returning an error */
278 	ext4_clear_inode(inode);
279 
280 	es = sbi->s_es;
281 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 		goto error_return;
284 	}
285 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 	/* Don't bother if the inode bitmap is corrupt. */
289 	if (IS_ERR(bitmap_bh)) {
290 		fatal = PTR_ERR(bitmap_bh);
291 		bitmap_bh = NULL;
292 		goto error_return;
293 	}
294 	if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 		grp = ext4_get_group_info(sb, block_group);
296 		if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 			fatal = -EFSCORRUPTED;
298 			goto error_return;
299 		}
300 	}
301 
302 	BUFFER_TRACE(bitmap_bh, "get_write_access");
303 	fatal = ext4_journal_get_write_access(handle, sb, bitmap_bh,
304 					      EXT4_JTR_NONE);
305 	if (fatal)
306 		goto error_return;
307 
308 	fatal = -ESRCH;
309 	gdp = ext4_get_group_desc(sb, block_group, &bh2);
310 	if (gdp) {
311 		BUFFER_TRACE(bh2, "get_write_access");
312 		fatal = ext4_journal_get_write_access(handle, sb, bh2,
313 						      EXT4_JTR_NONE);
314 	}
315 	ext4_lock_group(sb, block_group);
316 	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
317 	if (fatal || !cleared) {
318 		ext4_unlock_group(sb, block_group);
319 		goto out;
320 	}
321 
322 	count = ext4_free_inodes_count(sb, gdp) + 1;
323 	ext4_free_inodes_set(sb, gdp, count);
324 	if (is_directory) {
325 		count = ext4_used_dirs_count(sb, gdp) - 1;
326 		ext4_used_dirs_set(sb, gdp, count);
327 		if (percpu_counter_initialized(&sbi->s_dirs_counter))
328 			percpu_counter_dec(&sbi->s_dirs_counter);
329 	}
330 	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
331 				   EXT4_INODES_PER_GROUP(sb) / 8);
332 	ext4_group_desc_csum_set(sb, block_group, gdp);
333 	ext4_unlock_group(sb, block_group);
334 
335 	if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
336 		percpu_counter_inc(&sbi->s_freeinodes_counter);
337 	if (sbi->s_log_groups_per_flex) {
338 		struct flex_groups *fg;
339 
340 		fg = sbi_array_rcu_deref(sbi, s_flex_groups,
341 					 ext4_flex_group(sbi, block_group));
342 		atomic_inc(&fg->free_inodes);
343 		if (is_directory)
344 			atomic_dec(&fg->used_dirs);
345 	}
346 	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
347 	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
348 out:
349 	if (cleared) {
350 		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
351 		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
352 		if (!fatal)
353 			fatal = err;
354 	} else {
355 		ext4_error(sb, "bit already cleared for inode %lu", ino);
356 		ext4_mark_group_bitmap_corrupted(sb, block_group,
357 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
358 	}
359 
360 error_return:
361 	brelse(bitmap_bh);
362 	ext4_std_error(sb, fatal);
363 }
364 
365 struct orlov_stats {
366 	__u64 free_clusters;
367 	__u32 free_inodes;
368 	__u32 used_dirs;
369 };
370 
371 /*
372  * Helper function for Orlov's allocator; returns critical information
373  * for a particular block group or flex_bg.  If flex_size is 1, then g
374  * is a block group number; otherwise it is flex_bg number.
375  */
376 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
377 			    int flex_size, struct orlov_stats *stats)
378 {
379 	struct ext4_group_desc *desc;
380 
381 	if (flex_size > 1) {
382 		struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
383 							     s_flex_groups, g);
384 		stats->free_inodes = atomic_read(&fg->free_inodes);
385 		stats->free_clusters = atomic64_read(&fg->free_clusters);
386 		stats->used_dirs = atomic_read(&fg->used_dirs);
387 		return;
388 	}
389 
390 	desc = ext4_get_group_desc(sb, g, NULL);
391 	if (desc) {
392 		stats->free_inodes = ext4_free_inodes_count(sb, desc);
393 		stats->free_clusters = ext4_free_group_clusters(sb, desc);
394 		stats->used_dirs = ext4_used_dirs_count(sb, desc);
395 	} else {
396 		stats->free_inodes = 0;
397 		stats->free_clusters = 0;
398 		stats->used_dirs = 0;
399 	}
400 }
401 
402 /*
403  * Orlov's allocator for directories.
404  *
405  * We always try to spread first-level directories.
406  *
407  * If there are blockgroups with both free inodes and free clusters counts
408  * not worse than average we return one with smallest directory count.
409  * Otherwise we simply return a random group.
410  *
411  * For the rest rules look so:
412  *
413  * It's OK to put directory into a group unless
414  * it has too many directories already (max_dirs) or
415  * it has too few free inodes left (min_inodes) or
416  * it has too few free clusters left (min_clusters) or
417  * Parent's group is preferred, if it doesn't satisfy these
418  * conditions we search cyclically through the rest. If none
419  * of the groups look good we just look for a group with more
420  * free inodes than average (starting at parent's group).
421  */
422 
423 static int find_group_orlov(struct super_block *sb, struct inode *parent,
424 			    ext4_group_t *group, umode_t mode,
425 			    const struct qstr *qstr)
426 {
427 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
428 	struct ext4_sb_info *sbi = EXT4_SB(sb);
429 	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
430 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
431 	unsigned int freei, avefreei, grp_free;
432 	ext4_fsblk_t freec, avefreec;
433 	unsigned int ndirs;
434 	int max_dirs, min_inodes;
435 	ext4_grpblk_t min_clusters;
436 	ext4_group_t i, grp, g, ngroups;
437 	struct ext4_group_desc *desc;
438 	struct orlov_stats stats;
439 	int flex_size = ext4_flex_bg_size(sbi);
440 	struct dx_hash_info hinfo;
441 
442 	ngroups = real_ngroups;
443 	if (flex_size > 1) {
444 		ngroups = (real_ngroups + flex_size - 1) >>
445 			sbi->s_log_groups_per_flex;
446 		parent_group >>= sbi->s_log_groups_per_flex;
447 	}
448 
449 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
450 	avefreei = freei / ngroups;
451 	freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
452 	avefreec = freec;
453 	do_div(avefreec, ngroups);
454 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
455 
456 	if (S_ISDIR(mode) &&
457 	    ((parent == d_inode(sb->s_root)) ||
458 	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
459 		int best_ndir = inodes_per_group;
460 		int ret = -1;
461 
462 		if (qstr) {
463 			hinfo.hash_version = DX_HASH_HALF_MD4;
464 			hinfo.seed = sbi->s_hash_seed;
465 			ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
466 			grp = hinfo.hash;
467 		} else
468 			grp = prandom_u32();
469 		parent_group = (unsigned)grp % ngroups;
470 		for (i = 0; i < ngroups; i++) {
471 			g = (parent_group + i) % ngroups;
472 			get_orlov_stats(sb, g, flex_size, &stats);
473 			if (!stats.free_inodes)
474 				continue;
475 			if (stats.used_dirs >= best_ndir)
476 				continue;
477 			if (stats.free_inodes < avefreei)
478 				continue;
479 			if (stats.free_clusters < avefreec)
480 				continue;
481 			grp = g;
482 			ret = 0;
483 			best_ndir = stats.used_dirs;
484 		}
485 		if (ret)
486 			goto fallback;
487 	found_flex_bg:
488 		if (flex_size == 1) {
489 			*group = grp;
490 			return 0;
491 		}
492 
493 		/*
494 		 * We pack inodes at the beginning of the flexgroup's
495 		 * inode tables.  Block allocation decisions will do
496 		 * something similar, although regular files will
497 		 * start at 2nd block group of the flexgroup.  See
498 		 * ext4_ext_find_goal() and ext4_find_near().
499 		 */
500 		grp *= flex_size;
501 		for (i = 0; i < flex_size; i++) {
502 			if (grp+i >= real_ngroups)
503 				break;
504 			desc = ext4_get_group_desc(sb, grp+i, NULL);
505 			if (desc && ext4_free_inodes_count(sb, desc)) {
506 				*group = grp+i;
507 				return 0;
508 			}
509 		}
510 		goto fallback;
511 	}
512 
513 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
514 	min_inodes = avefreei - inodes_per_group*flex_size / 4;
515 	if (min_inodes < 1)
516 		min_inodes = 1;
517 	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
518 
519 	/*
520 	 * Start looking in the flex group where we last allocated an
521 	 * inode for this parent directory
522 	 */
523 	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
524 		parent_group = EXT4_I(parent)->i_last_alloc_group;
525 		if (flex_size > 1)
526 			parent_group >>= sbi->s_log_groups_per_flex;
527 	}
528 
529 	for (i = 0; i < ngroups; i++) {
530 		grp = (parent_group + i) % ngroups;
531 		get_orlov_stats(sb, grp, flex_size, &stats);
532 		if (stats.used_dirs >= max_dirs)
533 			continue;
534 		if (stats.free_inodes < min_inodes)
535 			continue;
536 		if (stats.free_clusters < min_clusters)
537 			continue;
538 		goto found_flex_bg;
539 	}
540 
541 fallback:
542 	ngroups = real_ngroups;
543 	avefreei = freei / ngroups;
544 fallback_retry:
545 	parent_group = EXT4_I(parent)->i_block_group;
546 	for (i = 0; i < ngroups; i++) {
547 		grp = (parent_group + i) % ngroups;
548 		desc = ext4_get_group_desc(sb, grp, NULL);
549 		if (desc) {
550 			grp_free = ext4_free_inodes_count(sb, desc);
551 			if (grp_free && grp_free >= avefreei) {
552 				*group = grp;
553 				return 0;
554 			}
555 		}
556 	}
557 
558 	if (avefreei) {
559 		/*
560 		 * The free-inodes counter is approximate, and for really small
561 		 * filesystems the above test can fail to find any blockgroups
562 		 */
563 		avefreei = 0;
564 		goto fallback_retry;
565 	}
566 
567 	return -1;
568 }
569 
570 static int find_group_other(struct super_block *sb, struct inode *parent,
571 			    ext4_group_t *group, umode_t mode)
572 {
573 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
574 	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
575 	struct ext4_group_desc *desc;
576 	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
577 
578 	/*
579 	 * Try to place the inode is the same flex group as its
580 	 * parent.  If we can't find space, use the Orlov algorithm to
581 	 * find another flex group, and store that information in the
582 	 * parent directory's inode information so that use that flex
583 	 * group for future allocations.
584 	 */
585 	if (flex_size > 1) {
586 		int retry = 0;
587 
588 	try_again:
589 		parent_group &= ~(flex_size-1);
590 		last = parent_group + flex_size;
591 		if (last > ngroups)
592 			last = ngroups;
593 		for  (i = parent_group; i < last; i++) {
594 			desc = ext4_get_group_desc(sb, i, NULL);
595 			if (desc && ext4_free_inodes_count(sb, desc)) {
596 				*group = i;
597 				return 0;
598 			}
599 		}
600 		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
601 			retry = 1;
602 			parent_group = EXT4_I(parent)->i_last_alloc_group;
603 			goto try_again;
604 		}
605 		/*
606 		 * If this didn't work, use the Orlov search algorithm
607 		 * to find a new flex group; we pass in the mode to
608 		 * avoid the topdir algorithms.
609 		 */
610 		*group = parent_group + flex_size;
611 		if (*group > ngroups)
612 			*group = 0;
613 		return find_group_orlov(sb, parent, group, mode, NULL);
614 	}
615 
616 	/*
617 	 * Try to place the inode in its parent directory
618 	 */
619 	*group = parent_group;
620 	desc = ext4_get_group_desc(sb, *group, NULL);
621 	if (desc && ext4_free_inodes_count(sb, desc) &&
622 	    ext4_free_group_clusters(sb, desc))
623 		return 0;
624 
625 	/*
626 	 * We're going to place this inode in a different blockgroup from its
627 	 * parent.  We want to cause files in a common directory to all land in
628 	 * the same blockgroup.  But we want files which are in a different
629 	 * directory which shares a blockgroup with our parent to land in a
630 	 * different blockgroup.
631 	 *
632 	 * So add our directory's i_ino into the starting point for the hash.
633 	 */
634 	*group = (*group + parent->i_ino) % ngroups;
635 
636 	/*
637 	 * Use a quadratic hash to find a group with a free inode and some free
638 	 * blocks.
639 	 */
640 	for (i = 1; i < ngroups; i <<= 1) {
641 		*group += i;
642 		if (*group >= ngroups)
643 			*group -= ngroups;
644 		desc = ext4_get_group_desc(sb, *group, NULL);
645 		if (desc && ext4_free_inodes_count(sb, desc) &&
646 		    ext4_free_group_clusters(sb, desc))
647 			return 0;
648 	}
649 
650 	/*
651 	 * That failed: try linear search for a free inode, even if that group
652 	 * has no free blocks.
653 	 */
654 	*group = parent_group;
655 	for (i = 0; i < ngroups; i++) {
656 		if (++*group >= ngroups)
657 			*group = 0;
658 		desc = ext4_get_group_desc(sb, *group, NULL);
659 		if (desc && ext4_free_inodes_count(sb, desc))
660 			return 0;
661 	}
662 
663 	return -1;
664 }
665 
666 /*
667  * In no journal mode, if an inode has recently been deleted, we want
668  * to avoid reusing it until we're reasonably sure the inode table
669  * block has been written back to disk.  (Yes, these values are
670  * somewhat arbitrary...)
671  */
672 #define RECENTCY_MIN	60
673 #define RECENTCY_DIRTY	300
674 
675 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
676 {
677 	struct ext4_group_desc	*gdp;
678 	struct ext4_inode	*raw_inode;
679 	struct buffer_head	*bh;
680 	int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
681 	int offset, ret = 0;
682 	int recentcy = RECENTCY_MIN;
683 	u32 dtime, now;
684 
685 	gdp = ext4_get_group_desc(sb, group, NULL);
686 	if (unlikely(!gdp))
687 		return 0;
688 
689 	bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
690 		       (ino / inodes_per_block));
691 	if (!bh || !buffer_uptodate(bh))
692 		/*
693 		 * If the block is not in the buffer cache, then it
694 		 * must have been written out.
695 		 */
696 		goto out;
697 
698 	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
699 	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
700 
701 	/* i_dtime is only 32 bits on disk, but we only care about relative
702 	 * times in the range of a few minutes (i.e. long enough to sync a
703 	 * recently-deleted inode to disk), so using the low 32 bits of the
704 	 * clock (a 68 year range) is enough, see time_before32() */
705 	dtime = le32_to_cpu(raw_inode->i_dtime);
706 	now = ktime_get_real_seconds();
707 	if (buffer_dirty(bh))
708 		recentcy += RECENTCY_DIRTY;
709 
710 	if (dtime && time_before32(dtime, now) &&
711 	    time_before32(now, dtime + recentcy))
712 		ret = 1;
713 out:
714 	brelse(bh);
715 	return ret;
716 }
717 
718 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
719 			  struct buffer_head *bitmap, unsigned long *ino)
720 {
721 	bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
722 	unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
723 
724 next:
725 	*ino = ext4_find_next_zero_bit((unsigned long *)
726 				       bitmap->b_data,
727 				       EXT4_INODES_PER_GROUP(sb), *ino);
728 	if (*ino >= EXT4_INODES_PER_GROUP(sb))
729 		goto not_found;
730 
731 	if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
732 		recently_deleted_ino = *ino;
733 		*ino = *ino + 1;
734 		if (*ino < EXT4_INODES_PER_GROUP(sb))
735 			goto next;
736 		goto not_found;
737 	}
738 	return 1;
739 not_found:
740 	if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
741 		return 0;
742 	/*
743 	 * Not reusing recently deleted inodes is mostly a preference. We don't
744 	 * want to report ENOSPC or skew allocation patterns because of that.
745 	 * So return even recently deleted inode if we could find better in the
746 	 * given range.
747 	 */
748 	*ino = recently_deleted_ino;
749 	return 1;
750 }
751 
752 int ext4_mark_inode_used(struct super_block *sb, int ino)
753 {
754 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
755 	struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
756 	struct ext4_group_desc *gdp;
757 	ext4_group_t group;
758 	int bit;
759 	int err = -EFSCORRUPTED;
760 
761 	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
762 		goto out;
763 
764 	group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
765 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
766 	inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
767 	if (IS_ERR(inode_bitmap_bh))
768 		return PTR_ERR(inode_bitmap_bh);
769 
770 	if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
771 		err = 0;
772 		goto out;
773 	}
774 
775 	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
776 	if (!gdp || !group_desc_bh) {
777 		err = -EINVAL;
778 		goto out;
779 	}
780 
781 	ext4_set_bit(bit, inode_bitmap_bh->b_data);
782 
783 	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
784 	err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
785 	if (err) {
786 		ext4_std_error(sb, err);
787 		goto out;
788 	}
789 	err = sync_dirty_buffer(inode_bitmap_bh);
790 	if (err) {
791 		ext4_std_error(sb, err);
792 		goto out;
793 	}
794 
795 	/* We may have to initialize the block bitmap if it isn't already */
796 	if (ext4_has_group_desc_csum(sb) &&
797 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
798 		struct buffer_head *block_bitmap_bh;
799 
800 		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
801 		if (IS_ERR(block_bitmap_bh)) {
802 			err = PTR_ERR(block_bitmap_bh);
803 			goto out;
804 		}
805 
806 		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
807 		err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
808 		sync_dirty_buffer(block_bitmap_bh);
809 
810 		/* recheck and clear flag under lock if we still need to */
811 		ext4_lock_group(sb, group);
812 		if (ext4_has_group_desc_csum(sb) &&
813 		    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
814 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
815 			ext4_free_group_clusters_set(sb, gdp,
816 				ext4_free_clusters_after_init(sb, group, gdp));
817 			ext4_block_bitmap_csum_set(sb, group, gdp,
818 						   block_bitmap_bh);
819 			ext4_group_desc_csum_set(sb, group, gdp);
820 		}
821 		ext4_unlock_group(sb, group);
822 		brelse(block_bitmap_bh);
823 
824 		if (err) {
825 			ext4_std_error(sb, err);
826 			goto out;
827 		}
828 	}
829 
830 	/* Update the relevant bg descriptor fields */
831 	if (ext4_has_group_desc_csum(sb)) {
832 		int free;
833 
834 		ext4_lock_group(sb, group); /* while we modify the bg desc */
835 		free = EXT4_INODES_PER_GROUP(sb) -
836 			ext4_itable_unused_count(sb, gdp);
837 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
838 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
839 			free = 0;
840 		}
841 
842 		/*
843 		 * Check the relative inode number against the last used
844 		 * relative inode number in this group. if it is greater
845 		 * we need to update the bg_itable_unused count
846 		 */
847 		if (bit >= free)
848 			ext4_itable_unused_set(sb, gdp,
849 					(EXT4_INODES_PER_GROUP(sb) - bit - 1));
850 	} else {
851 		ext4_lock_group(sb, group);
852 	}
853 
854 	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
855 	if (ext4_has_group_desc_csum(sb)) {
856 		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
857 					   EXT4_INODES_PER_GROUP(sb) / 8);
858 		ext4_group_desc_csum_set(sb, group, gdp);
859 	}
860 
861 	ext4_unlock_group(sb, group);
862 	err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
863 	sync_dirty_buffer(group_desc_bh);
864 out:
865 	return err;
866 }
867 
868 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
869 					    bool encrypt)
870 {
871 	struct super_block *sb = dir->i_sb;
872 	int nblocks = 0;
873 #ifdef CONFIG_EXT4_FS_POSIX_ACL
874 	struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
875 
876 	if (IS_ERR(p))
877 		return PTR_ERR(p);
878 	if (p) {
879 		int acl_size = p->a_count * sizeof(ext4_acl_entry);
880 
881 		nblocks += (S_ISDIR(mode) ? 2 : 1) *
882 			__ext4_xattr_set_credits(sb, NULL /* inode */,
883 						 NULL /* block_bh */, acl_size,
884 						 true /* is_create */);
885 		posix_acl_release(p);
886 	}
887 #endif
888 
889 #ifdef CONFIG_SECURITY
890 	{
891 		int num_security_xattrs = 1;
892 
893 #ifdef CONFIG_INTEGRITY
894 		num_security_xattrs++;
895 #endif
896 		/*
897 		 * We assume that security xattrs are never more than 1k.
898 		 * In practice they are under 128 bytes.
899 		 */
900 		nblocks += num_security_xattrs *
901 			__ext4_xattr_set_credits(sb, NULL /* inode */,
902 						 NULL /* block_bh */, 1024,
903 						 true /* is_create */);
904 	}
905 #endif
906 	if (encrypt)
907 		nblocks += __ext4_xattr_set_credits(sb,
908 						    NULL /* inode */,
909 						    NULL /* block_bh */,
910 						    FSCRYPT_SET_CONTEXT_MAX_SIZE,
911 						    true /* is_create */);
912 	return nblocks;
913 }
914 
915 /*
916  * There are two policies for allocating an inode.  If the new inode is
917  * a directory, then a forward search is made for a block group with both
918  * free space and a low directory-to-inode ratio; if that fails, then of
919  * the groups with above-average free space, that group with the fewest
920  * directories already is chosen.
921  *
922  * For other inodes, search forward from the parent directory's block
923  * group to find a free inode.
924  */
925 struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
926 			       handle_t *handle, struct inode *dir,
927 			       umode_t mode, const struct qstr *qstr,
928 			       __u32 goal, uid_t *owner, __u32 i_flags,
929 			       int handle_type, unsigned int line_no,
930 			       int nblocks)
931 {
932 	struct super_block *sb;
933 	struct buffer_head *inode_bitmap_bh = NULL;
934 	struct buffer_head *group_desc_bh;
935 	ext4_group_t ngroups, group = 0;
936 	unsigned long ino = 0;
937 	struct inode *inode;
938 	struct ext4_group_desc *gdp = NULL;
939 	struct ext4_inode_info *ei;
940 	struct ext4_sb_info *sbi;
941 	int ret2, err;
942 	struct inode *ret;
943 	ext4_group_t i;
944 	ext4_group_t flex_group;
945 	struct ext4_group_info *grp = NULL;
946 	bool encrypt = false;
947 
948 	/* Cannot create files in a deleted directory */
949 	if (!dir || !dir->i_nlink)
950 		return ERR_PTR(-EPERM);
951 
952 	sb = dir->i_sb;
953 	sbi = EXT4_SB(sb);
954 
955 	if (unlikely(ext4_forced_shutdown(sbi)))
956 		return ERR_PTR(-EIO);
957 
958 	ngroups = ext4_get_groups_count(sb);
959 	trace_ext4_request_inode(dir, mode);
960 	inode = new_inode(sb);
961 	if (!inode)
962 		return ERR_PTR(-ENOMEM);
963 	ei = EXT4_I(inode);
964 
965 	/*
966 	 * Initialize owners and quota early so that we don't have to account
967 	 * for quota initialization worst case in standard inode creating
968 	 * transaction
969 	 */
970 	if (owner) {
971 		inode->i_mode = mode;
972 		i_uid_write(inode, owner[0]);
973 		i_gid_write(inode, owner[1]);
974 	} else if (test_opt(sb, GRPID)) {
975 		inode->i_mode = mode;
976 		inode_fsuid_set(inode, mnt_userns);
977 		inode->i_gid = dir->i_gid;
978 	} else
979 		inode_init_owner(mnt_userns, inode, dir, mode);
980 
981 	if (ext4_has_feature_project(sb) &&
982 	    ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
983 		ei->i_projid = EXT4_I(dir)->i_projid;
984 	else
985 		ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
986 
987 	if (!(i_flags & EXT4_EA_INODE_FL)) {
988 		err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
989 		if (err)
990 			goto out;
991 	}
992 
993 	err = dquot_initialize(inode);
994 	if (err)
995 		goto out;
996 
997 	if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
998 		ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
999 		if (ret2 < 0) {
1000 			err = ret2;
1001 			goto out;
1002 		}
1003 		nblocks += ret2;
1004 	}
1005 
1006 	if (!goal)
1007 		goal = sbi->s_inode_goal;
1008 
1009 	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1010 		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1011 		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1012 		ret2 = 0;
1013 		goto got_group;
1014 	}
1015 
1016 	if (S_ISDIR(mode))
1017 		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1018 	else
1019 		ret2 = find_group_other(sb, dir, &group, mode);
1020 
1021 got_group:
1022 	EXT4_I(dir)->i_last_alloc_group = group;
1023 	err = -ENOSPC;
1024 	if (ret2 == -1)
1025 		goto out;
1026 
1027 	/*
1028 	 * Normally we will only go through one pass of this loop,
1029 	 * unless we get unlucky and it turns out the group we selected
1030 	 * had its last inode grabbed by someone else.
1031 	 */
1032 	for (i = 0; i < ngroups; i++, ino = 0) {
1033 		err = -EIO;
1034 
1035 		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1036 		if (!gdp)
1037 			goto out;
1038 
1039 		/*
1040 		 * Check free inodes count before loading bitmap.
1041 		 */
1042 		if (ext4_free_inodes_count(sb, gdp) == 0)
1043 			goto next_group;
1044 
1045 		if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1046 			grp = ext4_get_group_info(sb, group);
1047 			/*
1048 			 * Skip groups with already-known suspicious inode
1049 			 * tables
1050 			 */
1051 			if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1052 				goto next_group;
1053 		}
1054 
1055 		brelse(inode_bitmap_bh);
1056 		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1057 		/* Skip groups with suspicious inode tables */
1058 		if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1059 		     && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1060 		    IS_ERR(inode_bitmap_bh)) {
1061 			inode_bitmap_bh = NULL;
1062 			goto next_group;
1063 		}
1064 
1065 repeat_in_this_group:
1066 		ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1067 		if (!ret2)
1068 			goto next_group;
1069 
1070 		if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1071 			ext4_error(sb, "reserved inode found cleared - "
1072 				   "inode=%lu", ino + 1);
1073 			ext4_mark_group_bitmap_corrupted(sb, group,
1074 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1075 			goto next_group;
1076 		}
1077 
1078 		if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1079 			BUG_ON(nblocks <= 0);
1080 			handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1081 				 handle_type, nblocks, 0,
1082 				 ext4_trans_default_revoke_credits(sb));
1083 			if (IS_ERR(handle)) {
1084 				err = PTR_ERR(handle);
1085 				ext4_std_error(sb, err);
1086 				goto out;
1087 			}
1088 		}
1089 		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1090 		err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,
1091 						    EXT4_JTR_NONE);
1092 		if (err) {
1093 			ext4_std_error(sb, err);
1094 			goto out;
1095 		}
1096 		ext4_lock_group(sb, group);
1097 		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1098 		if (ret2) {
1099 			/* Someone already took the bit. Repeat the search
1100 			 * with lock held.
1101 			 */
1102 			ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1103 			if (ret2) {
1104 				ext4_set_bit(ino, inode_bitmap_bh->b_data);
1105 				ret2 = 0;
1106 			} else {
1107 				ret2 = 1; /* we didn't grab the inode */
1108 			}
1109 		}
1110 		ext4_unlock_group(sb, group);
1111 		ino++;		/* the inode bitmap is zero-based */
1112 		if (!ret2)
1113 			goto got; /* we grabbed the inode! */
1114 
1115 		if (ino < EXT4_INODES_PER_GROUP(sb))
1116 			goto repeat_in_this_group;
1117 next_group:
1118 		if (++group == ngroups)
1119 			group = 0;
1120 	}
1121 	err = -ENOSPC;
1122 	goto out;
1123 
1124 got:
1125 	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1126 	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1127 	if (err) {
1128 		ext4_std_error(sb, err);
1129 		goto out;
1130 	}
1131 
1132 	BUFFER_TRACE(group_desc_bh, "get_write_access");
1133 	err = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1134 					    EXT4_JTR_NONE);
1135 	if (err) {
1136 		ext4_std_error(sb, err);
1137 		goto out;
1138 	}
1139 
1140 	/* We may have to initialize the block bitmap if it isn't already */
1141 	if (ext4_has_group_desc_csum(sb) &&
1142 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1143 		struct buffer_head *block_bitmap_bh;
1144 
1145 		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1146 		if (IS_ERR(block_bitmap_bh)) {
1147 			err = PTR_ERR(block_bitmap_bh);
1148 			goto out;
1149 		}
1150 		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1151 		err = ext4_journal_get_write_access(handle, sb, block_bitmap_bh,
1152 						    EXT4_JTR_NONE);
1153 		if (err) {
1154 			brelse(block_bitmap_bh);
1155 			ext4_std_error(sb, err);
1156 			goto out;
1157 		}
1158 
1159 		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1160 		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1161 
1162 		/* recheck and clear flag under lock if we still need to */
1163 		ext4_lock_group(sb, group);
1164 		if (ext4_has_group_desc_csum(sb) &&
1165 		    (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1166 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1167 			ext4_free_group_clusters_set(sb, gdp,
1168 				ext4_free_clusters_after_init(sb, group, gdp));
1169 			ext4_block_bitmap_csum_set(sb, group, gdp,
1170 						   block_bitmap_bh);
1171 			ext4_group_desc_csum_set(sb, group, gdp);
1172 		}
1173 		ext4_unlock_group(sb, group);
1174 		brelse(block_bitmap_bh);
1175 
1176 		if (err) {
1177 			ext4_std_error(sb, err);
1178 			goto out;
1179 		}
1180 	}
1181 
1182 	/* Update the relevant bg descriptor fields */
1183 	if (ext4_has_group_desc_csum(sb)) {
1184 		int free;
1185 		struct ext4_group_info *grp = NULL;
1186 
1187 		if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1188 			grp = ext4_get_group_info(sb, group);
1189 			down_read(&grp->alloc_sem); /*
1190 						     * protect vs itable
1191 						     * lazyinit
1192 						     */
1193 		}
1194 		ext4_lock_group(sb, group); /* while we modify the bg desc */
1195 		free = EXT4_INODES_PER_GROUP(sb) -
1196 			ext4_itable_unused_count(sb, gdp);
1197 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1198 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1199 			free = 0;
1200 		}
1201 		/*
1202 		 * Check the relative inode number against the last used
1203 		 * relative inode number in this group. if it is greater
1204 		 * we need to update the bg_itable_unused count
1205 		 */
1206 		if (ino > free)
1207 			ext4_itable_unused_set(sb, gdp,
1208 					(EXT4_INODES_PER_GROUP(sb) - ino));
1209 		if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1210 			up_read(&grp->alloc_sem);
1211 	} else {
1212 		ext4_lock_group(sb, group);
1213 	}
1214 
1215 	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1216 	if (S_ISDIR(mode)) {
1217 		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1218 		if (sbi->s_log_groups_per_flex) {
1219 			ext4_group_t f = ext4_flex_group(sbi, group);
1220 
1221 			atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1222 							f)->used_dirs);
1223 		}
1224 	}
1225 	if (ext4_has_group_desc_csum(sb)) {
1226 		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1227 					   EXT4_INODES_PER_GROUP(sb) / 8);
1228 		ext4_group_desc_csum_set(sb, group, gdp);
1229 	}
1230 	ext4_unlock_group(sb, group);
1231 
1232 	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1233 	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1234 	if (err) {
1235 		ext4_std_error(sb, err);
1236 		goto out;
1237 	}
1238 
1239 	percpu_counter_dec(&sbi->s_freeinodes_counter);
1240 	if (S_ISDIR(mode))
1241 		percpu_counter_inc(&sbi->s_dirs_counter);
1242 
1243 	if (sbi->s_log_groups_per_flex) {
1244 		flex_group = ext4_flex_group(sbi, group);
1245 		atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1246 						flex_group)->free_inodes);
1247 	}
1248 
1249 	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1250 	/* This is the optimal IO size (for stat), not the fs block size */
1251 	inode->i_blocks = 0;
1252 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1253 	ei->i_crtime = inode->i_mtime;
1254 
1255 	memset(ei->i_data, 0, sizeof(ei->i_data));
1256 	ei->i_dir_start_lookup = 0;
1257 	ei->i_disksize = 0;
1258 
1259 	/* Don't inherit extent flag from directory, amongst others. */
1260 	ei->i_flags =
1261 		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1262 	ei->i_flags |= i_flags;
1263 	ei->i_file_acl = 0;
1264 	ei->i_dtime = 0;
1265 	ei->i_block_group = group;
1266 	ei->i_last_alloc_group = ~0;
1267 
1268 	ext4_set_inode_flags(inode, true);
1269 	if (IS_DIRSYNC(inode))
1270 		ext4_handle_sync(handle);
1271 	if (insert_inode_locked(inode) < 0) {
1272 		/*
1273 		 * Likely a bitmap corruption causing inode to be allocated
1274 		 * twice.
1275 		 */
1276 		err = -EIO;
1277 		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1278 			   inode->i_ino);
1279 		ext4_mark_group_bitmap_corrupted(sb, group,
1280 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1281 		goto out;
1282 	}
1283 	inode->i_generation = prandom_u32();
1284 
1285 	/* Precompute checksum seed for inode metadata */
1286 	if (ext4_has_metadata_csum(sb)) {
1287 		__u32 csum;
1288 		__le32 inum = cpu_to_le32(inode->i_ino);
1289 		__le32 gen = cpu_to_le32(inode->i_generation);
1290 		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1291 				   sizeof(inum));
1292 		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1293 					      sizeof(gen));
1294 	}
1295 
1296 	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1297 	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1298 
1299 	ei->i_extra_isize = sbi->s_want_extra_isize;
1300 	ei->i_inline_off = 0;
1301 	if (ext4_has_feature_inline_data(sb) &&
1302 	    (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1303 		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1304 	ret = inode;
1305 	err = dquot_alloc_inode(inode);
1306 	if (err)
1307 		goto fail_drop;
1308 
1309 	/*
1310 	 * Since the encryption xattr will always be unique, create it first so
1311 	 * that it's less likely to end up in an external xattr block and
1312 	 * prevent its deduplication.
1313 	 */
1314 	if (encrypt) {
1315 		err = fscrypt_set_context(inode, handle);
1316 		if (err)
1317 			goto fail_free_drop;
1318 	}
1319 
1320 	if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1321 		err = ext4_init_acl(handle, inode, dir);
1322 		if (err)
1323 			goto fail_free_drop;
1324 
1325 		err = ext4_init_security(handle, inode, dir, qstr);
1326 		if (err)
1327 			goto fail_free_drop;
1328 	}
1329 
1330 	if (ext4_has_feature_extents(sb)) {
1331 		/* set extent flag only for directory, file and normal symlink*/
1332 		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1333 			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1334 			ext4_ext_tree_init(handle, inode);
1335 		}
1336 	}
1337 
1338 	if (ext4_handle_valid(handle)) {
1339 		ei->i_sync_tid = handle->h_transaction->t_tid;
1340 		ei->i_datasync_tid = handle->h_transaction->t_tid;
1341 	}
1342 
1343 	err = ext4_mark_inode_dirty(handle, inode);
1344 	if (err) {
1345 		ext4_std_error(sb, err);
1346 		goto fail_free_drop;
1347 	}
1348 
1349 	ext4_debug("allocating inode %lu\n", inode->i_ino);
1350 	trace_ext4_allocate_inode(inode, dir, mode);
1351 	brelse(inode_bitmap_bh);
1352 	return ret;
1353 
1354 fail_free_drop:
1355 	dquot_free_inode(inode);
1356 fail_drop:
1357 	clear_nlink(inode);
1358 	unlock_new_inode(inode);
1359 out:
1360 	dquot_drop(inode);
1361 	inode->i_flags |= S_NOQUOTA;
1362 	iput(inode);
1363 	brelse(inode_bitmap_bh);
1364 	return ERR_PTR(err);
1365 }
1366 
1367 /* Verify that we are loading a valid orphan from disk */
1368 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1369 {
1370 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1371 	ext4_group_t block_group;
1372 	int bit;
1373 	struct buffer_head *bitmap_bh = NULL;
1374 	struct inode *inode = NULL;
1375 	int err = -EFSCORRUPTED;
1376 
1377 	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1378 		goto bad_orphan;
1379 
1380 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1381 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1382 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1383 	if (IS_ERR(bitmap_bh))
1384 		return ERR_CAST(bitmap_bh);
1385 
1386 	/* Having the inode bit set should be a 100% indicator that this
1387 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1388 	 * inodes that were being truncated, so we can't check i_nlink==0.
1389 	 */
1390 	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1391 		goto bad_orphan;
1392 
1393 	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1394 	if (IS_ERR(inode)) {
1395 		err = PTR_ERR(inode);
1396 		ext4_error_err(sb, -err,
1397 			       "couldn't read orphan inode %lu (err %d)",
1398 			       ino, err);
1399 		brelse(bitmap_bh);
1400 		return inode;
1401 	}
1402 
1403 	/*
1404 	 * If the orphans has i_nlinks > 0 then it should be able to
1405 	 * be truncated, otherwise it won't be removed from the orphan
1406 	 * list during processing and an infinite loop will result.
1407 	 * Similarly, it must not be a bad inode.
1408 	 */
1409 	if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1410 	    is_bad_inode(inode))
1411 		goto bad_orphan;
1412 
1413 	if (NEXT_ORPHAN(inode) > max_ino)
1414 		goto bad_orphan;
1415 	brelse(bitmap_bh);
1416 	return inode;
1417 
1418 bad_orphan:
1419 	ext4_error(sb, "bad orphan inode %lu", ino);
1420 	if (bitmap_bh)
1421 		printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1422 		       bit, (unsigned long long)bitmap_bh->b_blocknr,
1423 		       ext4_test_bit(bit, bitmap_bh->b_data));
1424 	if (inode) {
1425 		printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1426 		       is_bad_inode(inode));
1427 		printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1428 		       NEXT_ORPHAN(inode));
1429 		printk(KERN_ERR "max_ino=%lu\n", max_ino);
1430 		printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1431 		/* Avoid freeing blocks if we got a bad deleted inode */
1432 		if (inode->i_nlink == 0)
1433 			inode->i_blocks = 0;
1434 		iput(inode);
1435 	}
1436 	brelse(bitmap_bh);
1437 	return ERR_PTR(err);
1438 }
1439 
1440 unsigned long ext4_count_free_inodes(struct super_block *sb)
1441 {
1442 	unsigned long desc_count;
1443 	struct ext4_group_desc *gdp;
1444 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1445 #ifdef EXT4FS_DEBUG
1446 	struct ext4_super_block *es;
1447 	unsigned long bitmap_count, x;
1448 	struct buffer_head *bitmap_bh = NULL;
1449 
1450 	es = EXT4_SB(sb)->s_es;
1451 	desc_count = 0;
1452 	bitmap_count = 0;
1453 	gdp = NULL;
1454 	for (i = 0; i < ngroups; i++) {
1455 		gdp = ext4_get_group_desc(sb, i, NULL);
1456 		if (!gdp)
1457 			continue;
1458 		desc_count += ext4_free_inodes_count(sb, gdp);
1459 		brelse(bitmap_bh);
1460 		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1461 		if (IS_ERR(bitmap_bh)) {
1462 			bitmap_bh = NULL;
1463 			continue;
1464 		}
1465 
1466 		x = ext4_count_free(bitmap_bh->b_data,
1467 				    EXT4_INODES_PER_GROUP(sb) / 8);
1468 		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1469 			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1470 		bitmap_count += x;
1471 	}
1472 	brelse(bitmap_bh);
1473 	printk(KERN_DEBUG "ext4_count_free_inodes: "
1474 	       "stored = %u, computed = %lu, %lu\n",
1475 	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1476 	return desc_count;
1477 #else
1478 	desc_count = 0;
1479 	for (i = 0; i < ngroups; i++) {
1480 		gdp = ext4_get_group_desc(sb, i, NULL);
1481 		if (!gdp)
1482 			continue;
1483 		desc_count += ext4_free_inodes_count(sb, gdp);
1484 		cond_resched();
1485 	}
1486 	return desc_count;
1487 #endif
1488 }
1489 
1490 /* Called at mount-time, super-block is locked */
1491 unsigned long ext4_count_dirs(struct super_block * sb)
1492 {
1493 	unsigned long count = 0;
1494 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1495 
1496 	for (i = 0; i < ngroups; i++) {
1497 		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1498 		if (!gdp)
1499 			continue;
1500 		count += ext4_used_dirs_count(sb, gdp);
1501 	}
1502 	return count;
1503 }
1504 
1505 /*
1506  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1507  * inode table. Must be called without any spinlock held. The only place
1508  * where it is called from on active part of filesystem is ext4lazyinit
1509  * thread, so we do not need any special locks, however we have to prevent
1510  * inode allocation from the current group, so we take alloc_sem lock, to
1511  * block ext4_new_inode() until we are finished.
1512  */
1513 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1514 				 int barrier)
1515 {
1516 	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1517 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1518 	struct ext4_group_desc *gdp = NULL;
1519 	struct buffer_head *group_desc_bh;
1520 	handle_t *handle;
1521 	ext4_fsblk_t blk;
1522 	int num, ret = 0, used_blks = 0;
1523 	unsigned long used_inos = 0;
1524 
1525 	/* This should not happen, but just to be sure check this */
1526 	if (sb_rdonly(sb)) {
1527 		ret = 1;
1528 		goto out;
1529 	}
1530 
1531 	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1532 	if (!gdp)
1533 		goto out;
1534 
1535 	/*
1536 	 * We do not need to lock this, because we are the only one
1537 	 * handling this flag.
1538 	 */
1539 	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1540 		goto out;
1541 
1542 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1543 	if (IS_ERR(handle)) {
1544 		ret = PTR_ERR(handle);
1545 		goto out;
1546 	}
1547 
1548 	down_write(&grp->alloc_sem);
1549 	/*
1550 	 * If inode bitmap was already initialized there may be some
1551 	 * used inodes so we need to skip blocks with used inodes in
1552 	 * inode table.
1553 	 */
1554 	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1555 		used_inos = EXT4_INODES_PER_GROUP(sb) -
1556 			    ext4_itable_unused_count(sb, gdp);
1557 		used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1558 
1559 		/* Bogus inode unused count? */
1560 		if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1561 			ext4_error(sb, "Something is wrong with group %u: "
1562 				   "used itable blocks: %d; "
1563 				   "itable unused count: %u",
1564 				   group, used_blks,
1565 				   ext4_itable_unused_count(sb, gdp));
1566 			ret = 1;
1567 			goto err_out;
1568 		}
1569 
1570 		used_inos += group * EXT4_INODES_PER_GROUP(sb);
1571 		/*
1572 		 * Are there some uninitialized inodes in the inode table
1573 		 * before the first normal inode?
1574 		 */
1575 		if ((used_blks != sbi->s_itb_per_group) &&
1576 		     (used_inos < EXT4_FIRST_INO(sb))) {
1577 			ext4_error(sb, "Something is wrong with group %u: "
1578 				   "itable unused count: %u; "
1579 				   "itables initialized count: %ld",
1580 				   group, ext4_itable_unused_count(sb, gdp),
1581 				   used_inos);
1582 			ret = 1;
1583 			goto err_out;
1584 		}
1585 	}
1586 
1587 	blk = ext4_inode_table(sb, gdp) + used_blks;
1588 	num = sbi->s_itb_per_group - used_blks;
1589 
1590 	BUFFER_TRACE(group_desc_bh, "get_write_access");
1591 	ret = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1592 					    EXT4_JTR_NONE);
1593 	if (ret)
1594 		goto err_out;
1595 
1596 	/*
1597 	 * Skip zeroout if the inode table is full. But we set the ZEROED
1598 	 * flag anyway, because obviously, when it is full it does not need
1599 	 * further zeroing.
1600 	 */
1601 	if (unlikely(num == 0))
1602 		goto skip_zeroout;
1603 
1604 	ext4_debug("going to zero out inode table in group %d\n",
1605 		   group);
1606 	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1607 	if (ret < 0)
1608 		goto err_out;
1609 	if (barrier)
1610 		blkdev_issue_flush(sb->s_bdev);
1611 
1612 skip_zeroout:
1613 	ext4_lock_group(sb, group);
1614 	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1615 	ext4_group_desc_csum_set(sb, group, gdp);
1616 	ext4_unlock_group(sb, group);
1617 
1618 	BUFFER_TRACE(group_desc_bh,
1619 		     "call ext4_handle_dirty_metadata");
1620 	ret = ext4_handle_dirty_metadata(handle, NULL,
1621 					 group_desc_bh);
1622 
1623 err_out:
1624 	up_write(&grp->alloc_sem);
1625 	ext4_journal_stop(handle);
1626 out:
1627 	return ret;
1628 }
1629