1 /*
2  *  linux/fs/ext4/ialloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  BSD ufs-inspired inode and directory allocation by
10  *  Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14 
15 #include <linux/time.h>
16 #include <linux/fs.h>
17 #include <linux/stat.h>
18 #include <linux/string.h>
19 #include <linux/quotaops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/random.h>
22 #include <linux/bitops.h>
23 #include <linux/blkdev.h>
24 #include <asm/byteorder.h>
25 
26 #include "ext4.h"
27 #include "ext4_jbd2.h"
28 #include "xattr.h"
29 #include "acl.h"
30 
31 #include <trace/events/ext4.h>
32 
33 /*
34  * ialloc.c contains the inodes allocation and deallocation routines
35  */
36 
37 /*
38  * The free inodes are managed by bitmaps.  A file system contains several
39  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
40  * block for inodes, N blocks for the inode table and data blocks.
41  *
42  * The file system contains group descriptors which are located after the
43  * super block.  Each descriptor contains the number of the bitmap block and
44  * the free blocks count in the block.
45  */
46 
47 /*
48  * To avoid calling the atomic setbit hundreds or thousands of times, we only
49  * need to use it within a single byte (to ensure we get endianness right).
50  * We can use memset for the rest of the bitmap as there are no other users.
51  */
ext4_mark_bitmap_end(int start_bit,int end_bit,char * bitmap)52 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53 {
54 	int i;
55 
56 	if (start_bit >= end_bit)
57 		return;
58 
59 	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 		ext4_set_bit(i, bitmap);
62 	if (i < end_bit)
63 		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64 }
65 
66 /* Initializes an uninitialized inode bitmap */
ext4_init_inode_bitmap(struct super_block * sb,struct buffer_head * bh,ext4_group_t block_group,struct ext4_group_desc * gdp)67 static int ext4_init_inode_bitmap(struct super_block *sb,
68 				       struct buffer_head *bh,
69 				       ext4_group_t block_group,
70 				       struct ext4_group_desc *gdp)
71 {
72 	struct ext4_group_info *grp;
73 	struct ext4_sb_info *sbi = EXT4_SB(sb);
74 	J_ASSERT_BH(bh, buffer_locked(bh));
75 
76 	/* If checksum is bad mark all blocks and inodes use to prevent
77 	 * allocation, essentially implementing a per-group read-only flag. */
78 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
79 		grp = ext4_get_group_info(sb, block_group);
80 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
81 			percpu_counter_sub(&sbi->s_freeclusters_counter,
82 					   grp->bb_free);
83 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
84 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
85 			int count;
86 			count = ext4_free_inodes_count(sb, gdp);
87 			percpu_counter_sub(&sbi->s_freeinodes_counter,
88 					   count);
89 		}
90 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
91 		return -EFSBADCRC;
92 	}
93 
94 	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
95 	ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
96 			bh->b_data);
97 	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
98 				   EXT4_INODES_PER_GROUP(sb) / 8);
99 	ext4_group_desc_csum_set(sb, block_group, gdp);
100 
101 	return 0;
102 }
103 
ext4_end_bitmap_read(struct buffer_head * bh,int uptodate)104 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
105 {
106 	if (uptodate) {
107 		set_buffer_uptodate(bh);
108 		set_bitmap_uptodate(bh);
109 	}
110 	unlock_buffer(bh);
111 	put_bh(bh);
112 }
113 
ext4_validate_inode_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)114 static int ext4_validate_inode_bitmap(struct super_block *sb,
115 				      struct ext4_group_desc *desc,
116 				      ext4_group_t block_group,
117 				      struct buffer_head *bh)
118 {
119 	ext4_fsblk_t	blk;
120 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
121 	struct ext4_sb_info *sbi = EXT4_SB(sb);
122 
123 	if (buffer_verified(bh))
124 		return 0;
125 	if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
126 		return -EFSCORRUPTED;
127 
128 	ext4_lock_group(sb, block_group);
129 	blk = ext4_inode_bitmap(sb, desc);
130 	if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
131 					   EXT4_INODES_PER_GROUP(sb) / 8)) {
132 		ext4_unlock_group(sb, block_group);
133 		ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
134 			   "inode_bitmap = %llu", block_group, blk);
135 		grp = ext4_get_group_info(sb, block_group);
136 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
137 			int count;
138 			count = ext4_free_inodes_count(sb, desc);
139 			percpu_counter_sub(&sbi->s_freeinodes_counter,
140 					   count);
141 		}
142 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
143 		return -EFSBADCRC;
144 	}
145 	set_buffer_verified(bh);
146 	ext4_unlock_group(sb, block_group);
147 	return 0;
148 }
149 
150 /*
151  * Read the inode allocation bitmap for a given block_group, reading
152  * into the specified slot in the superblock's bitmap cache.
153  *
154  * Return buffer_head of bitmap on success or NULL.
155  */
156 static struct buffer_head *
ext4_read_inode_bitmap(struct super_block * sb,ext4_group_t block_group)157 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
158 {
159 	struct ext4_group_desc *desc;
160 	struct buffer_head *bh = NULL;
161 	ext4_fsblk_t bitmap_blk;
162 	int err;
163 
164 	desc = ext4_get_group_desc(sb, block_group, NULL);
165 	if (!desc)
166 		return ERR_PTR(-EFSCORRUPTED);
167 
168 	bitmap_blk = ext4_inode_bitmap(sb, desc);
169 	bh = sb_getblk(sb, bitmap_blk);
170 	if (unlikely(!bh)) {
171 		ext4_error(sb, "Cannot read inode bitmap - "
172 			    "block_group = %u, inode_bitmap = %llu",
173 			    block_group, bitmap_blk);
174 		return ERR_PTR(-EIO);
175 	}
176 	if (bitmap_uptodate(bh))
177 		goto verify;
178 
179 	lock_buffer(bh);
180 	if (bitmap_uptodate(bh)) {
181 		unlock_buffer(bh);
182 		goto verify;
183 	}
184 
185 	ext4_lock_group(sb, block_group);
186 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
187 		err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
188 		set_bitmap_uptodate(bh);
189 		set_buffer_uptodate(bh);
190 		set_buffer_verified(bh);
191 		ext4_unlock_group(sb, block_group);
192 		unlock_buffer(bh);
193 		if (err) {
194 			ext4_error(sb, "Failed to init inode bitmap for group "
195 				   "%u: %d", block_group, err);
196 			goto out;
197 		}
198 		return bh;
199 	}
200 	ext4_unlock_group(sb, block_group);
201 
202 	if (buffer_uptodate(bh)) {
203 		/*
204 		 * if not uninit if bh is uptodate,
205 		 * bitmap is also uptodate
206 		 */
207 		set_bitmap_uptodate(bh);
208 		unlock_buffer(bh);
209 		goto verify;
210 	}
211 	/*
212 	 * submit the buffer_head for reading
213 	 */
214 	trace_ext4_load_inode_bitmap(sb, block_group);
215 	bh->b_end_io = ext4_end_bitmap_read;
216 	get_bh(bh);
217 	submit_bh(READ | REQ_META | REQ_PRIO, bh);
218 	wait_on_buffer(bh);
219 	if (!buffer_uptodate(bh)) {
220 		put_bh(bh);
221 		ext4_error(sb, "Cannot read inode bitmap - "
222 			   "block_group = %u, inode_bitmap = %llu",
223 			   block_group, bitmap_blk);
224 		return ERR_PTR(-EIO);
225 	}
226 
227 verify:
228 	err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
229 	if (err)
230 		goto out;
231 	return bh;
232 out:
233 	put_bh(bh);
234 	return ERR_PTR(err);
235 }
236 
237 /*
238  * NOTE! When we get the inode, we're the only people
239  * that have access to it, and as such there are no
240  * race conditions we have to worry about. The inode
241  * is not on the hash-lists, and it cannot be reached
242  * through the filesystem because the directory entry
243  * has been deleted earlier.
244  *
245  * HOWEVER: we must make sure that we get no aliases,
246  * which means that we have to call "clear_inode()"
247  * _before_ we mark the inode not in use in the inode
248  * bitmaps. Otherwise a newly created file might use
249  * the same inode number (not actually the same pointer
250  * though), and then we'd have two inodes sharing the
251  * same inode number and space on the harddisk.
252  */
ext4_free_inode(handle_t * handle,struct inode * inode)253 void ext4_free_inode(handle_t *handle, struct inode *inode)
254 {
255 	struct super_block *sb = inode->i_sb;
256 	int is_directory;
257 	unsigned long ino;
258 	struct buffer_head *bitmap_bh = NULL;
259 	struct buffer_head *bh2;
260 	ext4_group_t block_group;
261 	unsigned long bit;
262 	struct ext4_group_desc *gdp;
263 	struct ext4_super_block *es;
264 	struct ext4_sb_info *sbi;
265 	int fatal = 0, err, count, cleared;
266 	struct ext4_group_info *grp;
267 
268 	if (!sb) {
269 		printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
270 		       "nonexistent device\n", __func__, __LINE__);
271 		return;
272 	}
273 	if (atomic_read(&inode->i_count) > 1) {
274 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
275 			 __func__, __LINE__, inode->i_ino,
276 			 atomic_read(&inode->i_count));
277 		return;
278 	}
279 	if (inode->i_nlink) {
280 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
281 			 __func__, __LINE__, inode->i_ino, inode->i_nlink);
282 		return;
283 	}
284 	sbi = EXT4_SB(sb);
285 
286 	ino = inode->i_ino;
287 	ext4_debug("freeing inode %lu\n", ino);
288 	trace_ext4_free_inode(inode);
289 
290 	/*
291 	 * Note: we must free any quota before locking the superblock,
292 	 * as writing the quota to disk may need the lock as well.
293 	 */
294 	dquot_initialize(inode);
295 	ext4_xattr_delete_inode(handle, inode);
296 	dquot_free_inode(inode);
297 	dquot_drop(inode);
298 
299 	is_directory = S_ISDIR(inode->i_mode);
300 
301 	/* Do this BEFORE marking the inode not in use or returning an error */
302 	ext4_clear_inode(inode);
303 
304 	es = EXT4_SB(sb)->s_es;
305 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
306 		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
307 		goto error_return;
308 	}
309 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
310 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
311 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
312 	/* Don't bother if the inode bitmap is corrupt. */
313 	grp = ext4_get_group_info(sb, block_group);
314 	if (IS_ERR(bitmap_bh)) {
315 		fatal = PTR_ERR(bitmap_bh);
316 		bitmap_bh = NULL;
317 		goto error_return;
318 	}
319 	if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
320 		fatal = -EFSCORRUPTED;
321 		goto error_return;
322 	}
323 
324 	BUFFER_TRACE(bitmap_bh, "get_write_access");
325 	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
326 	if (fatal)
327 		goto error_return;
328 
329 	fatal = -ESRCH;
330 	gdp = ext4_get_group_desc(sb, block_group, &bh2);
331 	if (gdp) {
332 		BUFFER_TRACE(bh2, "get_write_access");
333 		fatal = ext4_journal_get_write_access(handle, bh2);
334 	}
335 	ext4_lock_group(sb, block_group);
336 	cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
337 	if (fatal || !cleared) {
338 		ext4_unlock_group(sb, block_group);
339 		goto out;
340 	}
341 
342 	count = ext4_free_inodes_count(sb, gdp) + 1;
343 	ext4_free_inodes_set(sb, gdp, count);
344 	if (is_directory) {
345 		count = ext4_used_dirs_count(sb, gdp) - 1;
346 		ext4_used_dirs_set(sb, gdp, count);
347 		percpu_counter_dec(&sbi->s_dirs_counter);
348 	}
349 	ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
350 				   EXT4_INODES_PER_GROUP(sb) / 8);
351 	ext4_group_desc_csum_set(sb, block_group, gdp);
352 	ext4_unlock_group(sb, block_group);
353 
354 	percpu_counter_inc(&sbi->s_freeinodes_counter);
355 	if (sbi->s_log_groups_per_flex) {
356 		ext4_group_t f = ext4_flex_group(sbi, block_group);
357 
358 		atomic_inc(&sbi->s_flex_groups[f].free_inodes);
359 		if (is_directory)
360 			atomic_dec(&sbi->s_flex_groups[f].used_dirs);
361 	}
362 	BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
363 	fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
364 out:
365 	if (cleared) {
366 		BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
367 		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
368 		if (!fatal)
369 			fatal = err;
370 	} else {
371 		ext4_error(sb, "bit already cleared for inode %lu", ino);
372 		if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
373 			int count;
374 			count = ext4_free_inodes_count(sb, gdp);
375 			percpu_counter_sub(&sbi->s_freeinodes_counter,
376 					   count);
377 		}
378 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
379 	}
380 
381 error_return:
382 	brelse(bitmap_bh);
383 	ext4_std_error(sb, fatal);
384 }
385 
386 struct orlov_stats {
387 	__u64 free_clusters;
388 	__u32 free_inodes;
389 	__u32 used_dirs;
390 };
391 
392 /*
393  * Helper function for Orlov's allocator; returns critical information
394  * for a particular block group or flex_bg.  If flex_size is 1, then g
395  * is a block group number; otherwise it is flex_bg number.
396  */
get_orlov_stats(struct super_block * sb,ext4_group_t g,int flex_size,struct orlov_stats * stats)397 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
398 			    int flex_size, struct orlov_stats *stats)
399 {
400 	struct ext4_group_desc *desc;
401 	struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
402 
403 	if (flex_size > 1) {
404 		stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
405 		stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
406 		stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
407 		return;
408 	}
409 
410 	desc = ext4_get_group_desc(sb, g, NULL);
411 	if (desc) {
412 		stats->free_inodes = ext4_free_inodes_count(sb, desc);
413 		stats->free_clusters = ext4_free_group_clusters(sb, desc);
414 		stats->used_dirs = ext4_used_dirs_count(sb, desc);
415 	} else {
416 		stats->free_inodes = 0;
417 		stats->free_clusters = 0;
418 		stats->used_dirs = 0;
419 	}
420 }
421 
422 /*
423  * Orlov's allocator for directories.
424  *
425  * We always try to spread first-level directories.
426  *
427  * If there are blockgroups with both free inodes and free blocks counts
428  * not worse than average we return one with smallest directory count.
429  * Otherwise we simply return a random group.
430  *
431  * For the rest rules look so:
432  *
433  * It's OK to put directory into a group unless
434  * it has too many directories already (max_dirs) or
435  * it has too few free inodes left (min_inodes) or
436  * it has too few free blocks left (min_blocks) or
437  * Parent's group is preferred, if it doesn't satisfy these
438  * conditions we search cyclically through the rest. If none
439  * of the groups look good we just look for a group with more
440  * free inodes than average (starting at parent's group).
441  */
442 
find_group_orlov(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode,const struct qstr * qstr)443 static int find_group_orlov(struct super_block *sb, struct inode *parent,
444 			    ext4_group_t *group, umode_t mode,
445 			    const struct qstr *qstr)
446 {
447 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
448 	struct ext4_sb_info *sbi = EXT4_SB(sb);
449 	ext4_group_t real_ngroups = ext4_get_groups_count(sb);
450 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
451 	unsigned int freei, avefreei, grp_free;
452 	ext4_fsblk_t freeb, avefreec;
453 	unsigned int ndirs;
454 	int max_dirs, min_inodes;
455 	ext4_grpblk_t min_clusters;
456 	ext4_group_t i, grp, g, ngroups;
457 	struct ext4_group_desc *desc;
458 	struct orlov_stats stats;
459 	int flex_size = ext4_flex_bg_size(sbi);
460 	struct dx_hash_info hinfo;
461 
462 	ngroups = real_ngroups;
463 	if (flex_size > 1) {
464 		ngroups = (real_ngroups + flex_size - 1) >>
465 			sbi->s_log_groups_per_flex;
466 		parent_group >>= sbi->s_log_groups_per_flex;
467 	}
468 
469 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
470 	avefreei = freei / ngroups;
471 	freeb = EXT4_C2B(sbi,
472 		percpu_counter_read_positive(&sbi->s_freeclusters_counter));
473 	avefreec = freeb;
474 	do_div(avefreec, ngroups);
475 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
476 
477 	if (S_ISDIR(mode) &&
478 	    ((parent == d_inode(sb->s_root)) ||
479 	     (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
480 		int best_ndir = inodes_per_group;
481 		int ret = -1;
482 
483 		if (qstr) {
484 			hinfo.hash_version = DX_HASH_HALF_MD4;
485 			hinfo.seed = sbi->s_hash_seed;
486 			ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
487 			grp = hinfo.hash;
488 		} else
489 			grp = prandom_u32();
490 		parent_group = (unsigned)grp % ngroups;
491 		for (i = 0; i < ngroups; i++) {
492 			g = (parent_group + i) % ngroups;
493 			get_orlov_stats(sb, g, flex_size, &stats);
494 			if (!stats.free_inodes)
495 				continue;
496 			if (stats.used_dirs >= best_ndir)
497 				continue;
498 			if (stats.free_inodes < avefreei)
499 				continue;
500 			if (stats.free_clusters < avefreec)
501 				continue;
502 			grp = g;
503 			ret = 0;
504 			best_ndir = stats.used_dirs;
505 		}
506 		if (ret)
507 			goto fallback;
508 	found_flex_bg:
509 		if (flex_size == 1) {
510 			*group = grp;
511 			return 0;
512 		}
513 
514 		/*
515 		 * We pack inodes at the beginning of the flexgroup's
516 		 * inode tables.  Block allocation decisions will do
517 		 * something similar, although regular files will
518 		 * start at 2nd block group of the flexgroup.  See
519 		 * ext4_ext_find_goal() and ext4_find_near().
520 		 */
521 		grp *= flex_size;
522 		for (i = 0; i < flex_size; i++) {
523 			if (grp+i >= real_ngroups)
524 				break;
525 			desc = ext4_get_group_desc(sb, grp+i, NULL);
526 			if (desc && ext4_free_inodes_count(sb, desc)) {
527 				*group = grp+i;
528 				return 0;
529 			}
530 		}
531 		goto fallback;
532 	}
533 
534 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
535 	min_inodes = avefreei - inodes_per_group*flex_size / 4;
536 	if (min_inodes < 1)
537 		min_inodes = 1;
538 	min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
539 
540 	/*
541 	 * Start looking in the flex group where we last allocated an
542 	 * inode for this parent directory
543 	 */
544 	if (EXT4_I(parent)->i_last_alloc_group != ~0) {
545 		parent_group = EXT4_I(parent)->i_last_alloc_group;
546 		if (flex_size > 1)
547 			parent_group >>= sbi->s_log_groups_per_flex;
548 	}
549 
550 	for (i = 0; i < ngroups; i++) {
551 		grp = (parent_group + i) % ngroups;
552 		get_orlov_stats(sb, grp, flex_size, &stats);
553 		if (stats.used_dirs >= max_dirs)
554 			continue;
555 		if (stats.free_inodes < min_inodes)
556 			continue;
557 		if (stats.free_clusters < min_clusters)
558 			continue;
559 		goto found_flex_bg;
560 	}
561 
562 fallback:
563 	ngroups = real_ngroups;
564 	avefreei = freei / ngroups;
565 fallback_retry:
566 	parent_group = EXT4_I(parent)->i_block_group;
567 	for (i = 0; i < ngroups; i++) {
568 		grp = (parent_group + i) % ngroups;
569 		desc = ext4_get_group_desc(sb, grp, NULL);
570 		if (desc) {
571 			grp_free = ext4_free_inodes_count(sb, desc);
572 			if (grp_free && grp_free >= avefreei) {
573 				*group = grp;
574 				return 0;
575 			}
576 		}
577 	}
578 
579 	if (avefreei) {
580 		/*
581 		 * The free-inodes counter is approximate, and for really small
582 		 * filesystems the above test can fail to find any blockgroups
583 		 */
584 		avefreei = 0;
585 		goto fallback_retry;
586 	}
587 
588 	return -1;
589 }
590 
find_group_other(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode)591 static int find_group_other(struct super_block *sb, struct inode *parent,
592 			    ext4_group_t *group, umode_t mode)
593 {
594 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
595 	ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
596 	struct ext4_group_desc *desc;
597 	int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
598 
599 	/*
600 	 * Try to place the inode is the same flex group as its
601 	 * parent.  If we can't find space, use the Orlov algorithm to
602 	 * find another flex group, and store that information in the
603 	 * parent directory's inode information so that use that flex
604 	 * group for future allocations.
605 	 */
606 	if (flex_size > 1) {
607 		int retry = 0;
608 
609 	try_again:
610 		parent_group &= ~(flex_size-1);
611 		last = parent_group + flex_size;
612 		if (last > ngroups)
613 			last = ngroups;
614 		for  (i = parent_group; i < last; i++) {
615 			desc = ext4_get_group_desc(sb, i, NULL);
616 			if (desc && ext4_free_inodes_count(sb, desc)) {
617 				*group = i;
618 				return 0;
619 			}
620 		}
621 		if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
622 			retry = 1;
623 			parent_group = EXT4_I(parent)->i_last_alloc_group;
624 			goto try_again;
625 		}
626 		/*
627 		 * If this didn't work, use the Orlov search algorithm
628 		 * to find a new flex group; we pass in the mode to
629 		 * avoid the topdir algorithms.
630 		 */
631 		*group = parent_group + flex_size;
632 		if (*group > ngroups)
633 			*group = 0;
634 		return find_group_orlov(sb, parent, group, mode, NULL);
635 	}
636 
637 	/*
638 	 * Try to place the inode in its parent directory
639 	 */
640 	*group = parent_group;
641 	desc = ext4_get_group_desc(sb, *group, NULL);
642 	if (desc && ext4_free_inodes_count(sb, desc) &&
643 	    ext4_free_group_clusters(sb, desc))
644 		return 0;
645 
646 	/*
647 	 * We're going to place this inode in a different blockgroup from its
648 	 * parent.  We want to cause files in a common directory to all land in
649 	 * the same blockgroup.  But we want files which are in a different
650 	 * directory which shares a blockgroup with our parent to land in a
651 	 * different blockgroup.
652 	 *
653 	 * So add our directory's i_ino into the starting point for the hash.
654 	 */
655 	*group = (*group + parent->i_ino) % ngroups;
656 
657 	/*
658 	 * Use a quadratic hash to find a group with a free inode and some free
659 	 * blocks.
660 	 */
661 	for (i = 1; i < ngroups; i <<= 1) {
662 		*group += i;
663 		if (*group >= ngroups)
664 			*group -= ngroups;
665 		desc = ext4_get_group_desc(sb, *group, NULL);
666 		if (desc && ext4_free_inodes_count(sb, desc) &&
667 		    ext4_free_group_clusters(sb, desc))
668 			return 0;
669 	}
670 
671 	/*
672 	 * That failed: try linear search for a free inode, even if that group
673 	 * has no free blocks.
674 	 */
675 	*group = parent_group;
676 	for (i = 0; i < ngroups; i++) {
677 		if (++*group >= ngroups)
678 			*group = 0;
679 		desc = ext4_get_group_desc(sb, *group, NULL);
680 		if (desc && ext4_free_inodes_count(sb, desc))
681 			return 0;
682 	}
683 
684 	return -1;
685 }
686 
687 /*
688  * In no journal mode, if an inode has recently been deleted, we want
689  * to avoid reusing it until we're reasonably sure the inode table
690  * block has been written back to disk.  (Yes, these values are
691  * somewhat arbitrary...)
692  */
693 #define RECENTCY_MIN	5
694 #define RECENTCY_DIRTY	30
695 
recently_deleted(struct super_block * sb,ext4_group_t group,int ino)696 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
697 {
698 	struct ext4_group_desc	*gdp;
699 	struct ext4_inode	*raw_inode;
700 	struct buffer_head	*bh;
701 	unsigned long		dtime, now;
702 	int	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
703 	int	offset, ret = 0, recentcy = RECENTCY_MIN;
704 
705 	gdp = ext4_get_group_desc(sb, group, NULL);
706 	if (unlikely(!gdp))
707 		return 0;
708 
709 	bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
710 		       (ino / inodes_per_block));
711 	if (unlikely(!bh) || !buffer_uptodate(bh))
712 		/*
713 		 * If the block is not in the buffer cache, then it
714 		 * must have been written out.
715 		 */
716 		goto out;
717 
718 	offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
719 	raw_inode = (struct ext4_inode *) (bh->b_data + offset);
720 	dtime = le32_to_cpu(raw_inode->i_dtime);
721 	now = get_seconds();
722 	if (buffer_dirty(bh))
723 		recentcy += RECENTCY_DIRTY;
724 
725 	if (dtime && (dtime < now) && (now < dtime + recentcy))
726 		ret = 1;
727 out:
728 	brelse(bh);
729 	return ret;
730 }
731 
732 /*
733  * There are two policies for allocating an inode.  If the new inode is
734  * a directory, then a forward search is made for a block group with both
735  * free space and a low directory-to-inode ratio; if that fails, then of
736  * the groups with above-average free space, that group with the fewest
737  * directories already is chosen.
738  *
739  * For other inodes, search forward from the parent directory's block
740  * group to find a free inode.
741  */
__ext4_new_inode(handle_t * handle,struct inode * dir,umode_t mode,const struct qstr * qstr,__u32 goal,uid_t * owner,int handle_type,unsigned int line_no,int nblocks)742 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
743 			       umode_t mode, const struct qstr *qstr,
744 			       __u32 goal, uid_t *owner, int handle_type,
745 			       unsigned int line_no, int nblocks)
746 {
747 	struct super_block *sb;
748 	struct buffer_head *inode_bitmap_bh = NULL;
749 	struct buffer_head *group_desc_bh;
750 	ext4_group_t ngroups, group = 0;
751 	unsigned long ino = 0;
752 	struct inode *inode;
753 	struct ext4_group_desc *gdp = NULL;
754 	struct ext4_inode_info *ei;
755 	struct ext4_sb_info *sbi;
756 	int ret2, err;
757 	struct inode *ret;
758 	ext4_group_t i;
759 	ext4_group_t flex_group;
760 	struct ext4_group_info *grp;
761 	int encrypt = 0;
762 
763 	/* Cannot create files in a deleted directory */
764 	if (!dir || !dir->i_nlink)
765 		return ERR_PTR(-EPERM);
766 
767 	if ((ext4_encrypted_inode(dir) ||
768 	     DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
769 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
770 		err = ext4_get_encryption_info(dir);
771 		if (err)
772 			return ERR_PTR(err);
773 		if (ext4_encryption_info(dir) == NULL)
774 			return ERR_PTR(-EPERM);
775 		if (!handle)
776 			nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
777 		encrypt = 1;
778 	}
779 
780 	sb = dir->i_sb;
781 	ngroups = ext4_get_groups_count(sb);
782 	trace_ext4_request_inode(dir, mode);
783 	inode = new_inode(sb);
784 	if (!inode)
785 		return ERR_PTR(-ENOMEM);
786 	ei = EXT4_I(inode);
787 	sbi = EXT4_SB(sb);
788 
789 	/*
790 	 * Initalize owners and quota early so that we don't have to account
791 	 * for quota initialization worst case in standard inode creating
792 	 * transaction
793 	 */
794 	if (owner) {
795 		inode->i_mode = mode;
796 		i_uid_write(inode, owner[0]);
797 		i_gid_write(inode, owner[1]);
798 	} else if (test_opt(sb, GRPID)) {
799 		inode->i_mode = mode;
800 		inode->i_uid = current_fsuid();
801 		inode->i_gid = dir->i_gid;
802 	} else
803 		inode_init_owner(inode, dir, mode);
804 	err = dquot_initialize(inode);
805 	if (err)
806 		goto out;
807 
808 	if (!goal)
809 		goal = sbi->s_inode_goal;
810 
811 	if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
812 		group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
813 		ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
814 		ret2 = 0;
815 		goto got_group;
816 	}
817 
818 	if (S_ISDIR(mode))
819 		ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
820 	else
821 		ret2 = find_group_other(sb, dir, &group, mode);
822 
823 got_group:
824 	EXT4_I(dir)->i_last_alloc_group = group;
825 	err = -ENOSPC;
826 	if (ret2 == -1)
827 		goto out;
828 
829 	/*
830 	 * Normally we will only go through one pass of this loop,
831 	 * unless we get unlucky and it turns out the group we selected
832 	 * had its last inode grabbed by someone else.
833 	 */
834 	for (i = 0; i < ngroups; i++, ino = 0) {
835 		err = -EIO;
836 
837 		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
838 		if (!gdp)
839 			goto out;
840 
841 		/*
842 		 * Check free inodes count before loading bitmap.
843 		 */
844 		if (ext4_free_inodes_count(sb, gdp) == 0) {
845 			if (++group == ngroups)
846 				group = 0;
847 			continue;
848 		}
849 
850 		grp = ext4_get_group_info(sb, group);
851 		/* Skip groups with already-known suspicious inode tables */
852 		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
853 			if (++group == ngroups)
854 				group = 0;
855 			continue;
856 		}
857 
858 		brelse(inode_bitmap_bh);
859 		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
860 		/* Skip groups with suspicious inode tables */
861 		if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
862 		    IS_ERR(inode_bitmap_bh)) {
863 			inode_bitmap_bh = NULL;
864 			if (++group == ngroups)
865 				group = 0;
866 			continue;
867 		}
868 
869 repeat_in_this_group:
870 		ino = ext4_find_next_zero_bit((unsigned long *)
871 					      inode_bitmap_bh->b_data,
872 					      EXT4_INODES_PER_GROUP(sb), ino);
873 		if (ino >= EXT4_INODES_PER_GROUP(sb))
874 			goto next_group;
875 		if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
876 			ext4_error(sb, "reserved inode found cleared - "
877 				   "inode=%lu", ino + 1);
878 			continue;
879 		}
880 		if ((EXT4_SB(sb)->s_journal == NULL) &&
881 		    recently_deleted(sb, group, ino)) {
882 			ino++;
883 			goto next_inode;
884 		}
885 		if (!handle) {
886 			BUG_ON(nblocks <= 0);
887 			handle = __ext4_journal_start_sb(dir->i_sb, line_no,
888 							 handle_type, nblocks,
889 							 0);
890 			if (IS_ERR(handle)) {
891 				err = PTR_ERR(handle);
892 				ext4_std_error(sb, err);
893 				goto out;
894 			}
895 		}
896 		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
897 		err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
898 		if (err) {
899 			ext4_std_error(sb, err);
900 			goto out;
901 		}
902 		ext4_lock_group(sb, group);
903 		ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
904 		ext4_unlock_group(sb, group);
905 		ino++;		/* the inode bitmap is zero-based */
906 		if (!ret2)
907 			goto got; /* we grabbed the inode! */
908 next_inode:
909 		if (ino < EXT4_INODES_PER_GROUP(sb))
910 			goto repeat_in_this_group;
911 next_group:
912 		if (++group == ngroups)
913 			group = 0;
914 	}
915 	err = -ENOSPC;
916 	goto out;
917 
918 got:
919 	BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
920 	err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
921 	if (err) {
922 		ext4_std_error(sb, err);
923 		goto out;
924 	}
925 
926 	BUFFER_TRACE(group_desc_bh, "get_write_access");
927 	err = ext4_journal_get_write_access(handle, group_desc_bh);
928 	if (err) {
929 		ext4_std_error(sb, err);
930 		goto out;
931 	}
932 
933 	/* We may have to initialize the block bitmap if it isn't already */
934 	if (ext4_has_group_desc_csum(sb) &&
935 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
936 		struct buffer_head *block_bitmap_bh;
937 
938 		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
939 		if (IS_ERR(block_bitmap_bh)) {
940 			err = PTR_ERR(block_bitmap_bh);
941 			goto out;
942 		}
943 		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
944 		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
945 		if (err) {
946 			brelse(block_bitmap_bh);
947 			ext4_std_error(sb, err);
948 			goto out;
949 		}
950 
951 		BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
952 		err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
953 
954 		/* recheck and clear flag under lock if we still need to */
955 		ext4_lock_group(sb, group);
956 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
957 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
958 			ext4_free_group_clusters_set(sb, gdp,
959 				ext4_free_clusters_after_init(sb, group, gdp));
960 			ext4_block_bitmap_csum_set(sb, group, gdp,
961 						   block_bitmap_bh);
962 			ext4_group_desc_csum_set(sb, group, gdp);
963 		}
964 		ext4_unlock_group(sb, group);
965 		brelse(block_bitmap_bh);
966 
967 		if (err) {
968 			ext4_std_error(sb, err);
969 			goto out;
970 		}
971 	}
972 
973 	/* Update the relevant bg descriptor fields */
974 	if (ext4_has_group_desc_csum(sb)) {
975 		int free;
976 		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
977 
978 		down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
979 		ext4_lock_group(sb, group); /* while we modify the bg desc */
980 		free = EXT4_INODES_PER_GROUP(sb) -
981 			ext4_itable_unused_count(sb, gdp);
982 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
983 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
984 			free = 0;
985 		}
986 		/*
987 		 * Check the relative inode number against the last used
988 		 * relative inode number in this group. if it is greater
989 		 * we need to update the bg_itable_unused count
990 		 */
991 		if (ino > free)
992 			ext4_itable_unused_set(sb, gdp,
993 					(EXT4_INODES_PER_GROUP(sb) - ino));
994 		up_read(&grp->alloc_sem);
995 	} else {
996 		ext4_lock_group(sb, group);
997 	}
998 
999 	ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1000 	if (S_ISDIR(mode)) {
1001 		ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1002 		if (sbi->s_log_groups_per_flex) {
1003 			ext4_group_t f = ext4_flex_group(sbi, group);
1004 
1005 			atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1006 		}
1007 	}
1008 	if (ext4_has_group_desc_csum(sb)) {
1009 		ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1010 					   EXT4_INODES_PER_GROUP(sb) / 8);
1011 		ext4_group_desc_csum_set(sb, group, gdp);
1012 	}
1013 	ext4_unlock_group(sb, group);
1014 
1015 	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1016 	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1017 	if (err) {
1018 		ext4_std_error(sb, err);
1019 		goto out;
1020 	}
1021 
1022 	percpu_counter_dec(&sbi->s_freeinodes_counter);
1023 	if (S_ISDIR(mode))
1024 		percpu_counter_inc(&sbi->s_dirs_counter);
1025 
1026 	if (sbi->s_log_groups_per_flex) {
1027 		flex_group = ext4_flex_group(sbi, group);
1028 		atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1029 	}
1030 
1031 	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1032 	/* This is the optimal IO size (for stat), not the fs block size */
1033 	inode->i_blocks = 0;
1034 	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1035 						       ext4_current_time(inode);
1036 
1037 	memset(ei->i_data, 0, sizeof(ei->i_data));
1038 	ei->i_dir_start_lookup = 0;
1039 	ei->i_disksize = 0;
1040 
1041 	/* Don't inherit extent flag from directory, amongst others. */
1042 	ei->i_flags =
1043 		ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1044 	ei->i_file_acl = 0;
1045 	ei->i_dtime = 0;
1046 	ei->i_block_group = group;
1047 	ei->i_last_alloc_group = ~0;
1048 
1049 	ext4_set_inode_flags(inode);
1050 	if (IS_DIRSYNC(inode))
1051 		ext4_handle_sync(handle);
1052 	if (insert_inode_locked(inode) < 0) {
1053 		/*
1054 		 * Likely a bitmap corruption causing inode to be allocated
1055 		 * twice.
1056 		 */
1057 		err = -EIO;
1058 		ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1059 			   inode->i_ino);
1060 		goto out;
1061 	}
1062 	spin_lock(&sbi->s_next_gen_lock);
1063 	inode->i_generation = sbi->s_next_generation++;
1064 	spin_unlock(&sbi->s_next_gen_lock);
1065 
1066 	/* Precompute checksum seed for inode metadata */
1067 	if (ext4_has_metadata_csum(sb)) {
1068 		__u32 csum;
1069 		__le32 inum = cpu_to_le32(inode->i_ino);
1070 		__le32 gen = cpu_to_le32(inode->i_generation);
1071 		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1072 				   sizeof(inum));
1073 		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1074 					      sizeof(gen));
1075 	}
1076 
1077 	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1078 	ext4_set_inode_state(inode, EXT4_STATE_NEW);
1079 
1080 	ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1081 	ei->i_inline_off = 0;
1082 	if (ext4_has_feature_inline_data(sb))
1083 		ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1084 	ret = inode;
1085 	err = dquot_alloc_inode(inode);
1086 	if (err)
1087 		goto fail_drop;
1088 
1089 	err = ext4_init_acl(handle, inode, dir);
1090 	if (err)
1091 		goto fail_free_drop;
1092 
1093 	err = ext4_init_security(handle, inode, dir, qstr);
1094 	if (err)
1095 		goto fail_free_drop;
1096 
1097 	if (ext4_has_feature_extents(sb)) {
1098 		/* set extent flag only for directory, file and normal symlink*/
1099 		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1100 			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1101 			ext4_ext_tree_init(handle, inode);
1102 		}
1103 	}
1104 
1105 	if (ext4_handle_valid(handle)) {
1106 		ei->i_sync_tid = handle->h_transaction->t_tid;
1107 		ei->i_datasync_tid = handle->h_transaction->t_tid;
1108 	}
1109 
1110 	if (encrypt) {
1111 		err = ext4_inherit_context(dir, inode);
1112 		if (err)
1113 			goto fail_free_drop;
1114 	}
1115 
1116 	err = ext4_mark_inode_dirty(handle, inode);
1117 	if (err) {
1118 		ext4_std_error(sb, err);
1119 		goto fail_free_drop;
1120 	}
1121 
1122 	ext4_debug("allocating inode %lu\n", inode->i_ino);
1123 	trace_ext4_allocate_inode(inode, dir, mode);
1124 	brelse(inode_bitmap_bh);
1125 	return ret;
1126 
1127 fail_free_drop:
1128 	dquot_free_inode(inode);
1129 fail_drop:
1130 	clear_nlink(inode);
1131 	unlock_new_inode(inode);
1132 out:
1133 	dquot_drop(inode);
1134 	inode->i_flags |= S_NOQUOTA;
1135 	iput(inode);
1136 	brelse(inode_bitmap_bh);
1137 	return ERR_PTR(err);
1138 }
1139 
1140 /* Verify that we are loading a valid orphan from disk */
ext4_orphan_get(struct super_block * sb,unsigned long ino)1141 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1142 {
1143 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1144 	ext4_group_t block_group;
1145 	int bit;
1146 	struct buffer_head *bitmap_bh = NULL;
1147 	struct inode *inode = NULL;
1148 	int err = -EFSCORRUPTED;
1149 
1150 	if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1151 		goto bad_orphan;
1152 
1153 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1154 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1155 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1156 	if (IS_ERR(bitmap_bh)) {
1157 		ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1158 			   ino, PTR_ERR(bitmap_bh));
1159 		return (struct inode *) bitmap_bh;
1160 	}
1161 
1162 	/* Having the inode bit set should be a 100% indicator that this
1163 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
1164 	 * inodes that were being truncated, so we can't check i_nlink==0.
1165 	 */
1166 	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1167 		goto bad_orphan;
1168 
1169 	inode = ext4_iget(sb, ino);
1170 	if (IS_ERR(inode)) {
1171 		err = PTR_ERR(inode);
1172 		ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1173 			   ino, err);
1174 		return inode;
1175 	}
1176 
1177 	/*
1178 	 * If the orphans has i_nlinks > 0 then it should be able to
1179 	 * be truncated, otherwise it won't be removed from the orphan
1180 	 * list during processing and an infinite loop will result.
1181 	 * Similarly, it must not be a bad inode.
1182 	 */
1183 	if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1184 	    is_bad_inode(inode))
1185 		goto bad_orphan;
1186 
1187 	if (NEXT_ORPHAN(inode) > max_ino)
1188 		goto bad_orphan;
1189 	brelse(bitmap_bh);
1190 	return inode;
1191 
1192 bad_orphan:
1193 	ext4_error(sb, "bad orphan inode %lu", ino);
1194 	if (bitmap_bh)
1195 		printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1196 		       bit, (unsigned long long)bitmap_bh->b_blocknr,
1197 		       ext4_test_bit(bit, bitmap_bh->b_data));
1198 	if (inode) {
1199 		printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1200 		       is_bad_inode(inode));
1201 		printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1202 		       NEXT_ORPHAN(inode));
1203 		printk(KERN_ERR "max_ino=%lu\n", max_ino);
1204 		printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1205 		/* Avoid freeing blocks if we got a bad deleted inode */
1206 		if (inode->i_nlink == 0)
1207 			inode->i_blocks = 0;
1208 		iput(inode);
1209 	}
1210 	brelse(bitmap_bh);
1211 	return ERR_PTR(err);
1212 }
1213 
ext4_count_free_inodes(struct super_block * sb)1214 unsigned long ext4_count_free_inodes(struct super_block *sb)
1215 {
1216 	unsigned long desc_count;
1217 	struct ext4_group_desc *gdp;
1218 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1219 #ifdef EXT4FS_DEBUG
1220 	struct ext4_super_block *es;
1221 	unsigned long bitmap_count, x;
1222 	struct buffer_head *bitmap_bh = NULL;
1223 
1224 	es = EXT4_SB(sb)->s_es;
1225 	desc_count = 0;
1226 	bitmap_count = 0;
1227 	gdp = NULL;
1228 	for (i = 0; i < ngroups; i++) {
1229 		gdp = ext4_get_group_desc(sb, i, NULL);
1230 		if (!gdp)
1231 			continue;
1232 		desc_count += ext4_free_inodes_count(sb, gdp);
1233 		brelse(bitmap_bh);
1234 		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1235 		if (IS_ERR(bitmap_bh)) {
1236 			bitmap_bh = NULL;
1237 			continue;
1238 		}
1239 
1240 		x = ext4_count_free(bitmap_bh->b_data,
1241 				    EXT4_INODES_PER_GROUP(sb) / 8);
1242 		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1243 			(unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1244 		bitmap_count += x;
1245 	}
1246 	brelse(bitmap_bh);
1247 	printk(KERN_DEBUG "ext4_count_free_inodes: "
1248 	       "stored = %u, computed = %lu, %lu\n",
1249 	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1250 	return desc_count;
1251 #else
1252 	desc_count = 0;
1253 	for (i = 0; i < ngroups; i++) {
1254 		gdp = ext4_get_group_desc(sb, i, NULL);
1255 		if (!gdp)
1256 			continue;
1257 		desc_count += ext4_free_inodes_count(sb, gdp);
1258 		cond_resched();
1259 	}
1260 	return desc_count;
1261 #endif
1262 }
1263 
1264 /* Called at mount-time, super-block is locked */
ext4_count_dirs(struct super_block * sb)1265 unsigned long ext4_count_dirs(struct super_block * sb)
1266 {
1267 	unsigned long count = 0;
1268 	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1269 
1270 	for (i = 0; i < ngroups; i++) {
1271 		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1272 		if (!gdp)
1273 			continue;
1274 		count += ext4_used_dirs_count(sb, gdp);
1275 	}
1276 	return count;
1277 }
1278 
1279 /*
1280  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1281  * inode table. Must be called without any spinlock held. The only place
1282  * where it is called from on active part of filesystem is ext4lazyinit
1283  * thread, so we do not need any special locks, however we have to prevent
1284  * inode allocation from the current group, so we take alloc_sem lock, to
1285  * block ext4_new_inode() until we are finished.
1286  */
ext4_init_inode_table(struct super_block * sb,ext4_group_t group,int barrier)1287 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1288 				 int barrier)
1289 {
1290 	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1291 	struct ext4_sb_info *sbi = EXT4_SB(sb);
1292 	struct ext4_group_desc *gdp = NULL;
1293 	struct buffer_head *group_desc_bh;
1294 	handle_t *handle;
1295 	ext4_fsblk_t blk;
1296 	int num, ret = 0, used_blks = 0;
1297 
1298 	/* This should not happen, but just to be sure check this */
1299 	if (sb->s_flags & MS_RDONLY) {
1300 		ret = 1;
1301 		goto out;
1302 	}
1303 
1304 	gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1305 	if (!gdp)
1306 		goto out;
1307 
1308 	/*
1309 	 * We do not need to lock this, because we are the only one
1310 	 * handling this flag.
1311 	 */
1312 	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1313 		goto out;
1314 
1315 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1316 	if (IS_ERR(handle)) {
1317 		ret = PTR_ERR(handle);
1318 		goto out;
1319 	}
1320 
1321 	down_write(&grp->alloc_sem);
1322 	/*
1323 	 * If inode bitmap was already initialized there may be some
1324 	 * used inodes so we need to skip blocks with used inodes in
1325 	 * inode table.
1326 	 */
1327 	if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1328 		used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1329 			    ext4_itable_unused_count(sb, gdp)),
1330 			    sbi->s_inodes_per_block);
1331 
1332 	if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1333 		ext4_error(sb, "Something is wrong with group %u: "
1334 			   "used itable blocks: %d; "
1335 			   "itable unused count: %u",
1336 			   group, used_blks,
1337 			   ext4_itable_unused_count(sb, gdp));
1338 		ret = 1;
1339 		goto err_out;
1340 	}
1341 
1342 	blk = ext4_inode_table(sb, gdp) + used_blks;
1343 	num = sbi->s_itb_per_group - used_blks;
1344 
1345 	BUFFER_TRACE(group_desc_bh, "get_write_access");
1346 	ret = ext4_journal_get_write_access(handle,
1347 					    group_desc_bh);
1348 	if (ret)
1349 		goto err_out;
1350 
1351 	/*
1352 	 * Skip zeroout if the inode table is full. But we set the ZEROED
1353 	 * flag anyway, because obviously, when it is full it does not need
1354 	 * further zeroing.
1355 	 */
1356 	if (unlikely(num == 0))
1357 		goto skip_zeroout;
1358 
1359 	ext4_debug("going to zero out inode table in group %d\n",
1360 		   group);
1361 	ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1362 	if (ret < 0)
1363 		goto err_out;
1364 	if (barrier)
1365 		blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1366 
1367 skip_zeroout:
1368 	ext4_lock_group(sb, group);
1369 	gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1370 	ext4_group_desc_csum_set(sb, group, gdp);
1371 	ext4_unlock_group(sb, group);
1372 
1373 	BUFFER_TRACE(group_desc_bh,
1374 		     "call ext4_handle_dirty_metadata");
1375 	ret = ext4_handle_dirty_metadata(handle, NULL,
1376 					 group_desc_bh);
1377 
1378 err_out:
1379 	up_write(&grp->alloc_sem);
1380 	ext4_journal_stop(handle);
1381 out:
1382 	return ret;
1383 }
1384