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