xref: /linux/fs/ext4/ialloc.c (revision 0c93ea4064a209cdc36de8a9a3003d43d08f46f7)
1 /*
2  *  linux/fs/ext4/ialloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  BSD ufs-inspired inode and directory allocation by
10  *  Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14 
15 #include <linux/time.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/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 <asm/byteorder.h>
26 #include "ext4.h"
27 #include "ext4_jbd2.h"
28 #include "xattr.h"
29 #include "acl.h"
30 #include "group.h"
31 
32 /*
33  * ialloc.c contains the inodes allocation and deallocation routines
34  */
35 
36 /*
37  * The free inodes are managed by bitmaps.  A file system contains several
38  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
39  * block for inodes, N blocks for the inode table and data blocks.
40  *
41  * The file system contains group descriptors which are located after the
42  * super block.  Each descriptor contains the number of the bitmap block and
43  * the free blocks count in the block.
44  */
45 
46 /*
47  * To avoid calling the atomic setbit hundreds or thousands of times, we only
48  * need to use it within a single byte (to ensure we get endianness right).
49  * We can use memset for the rest of the bitmap as there are no other users.
50  */
51 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
52 {
53 	int i;
54 
55 	if (start_bit >= end_bit)
56 		return;
57 
58 	ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
59 	for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
60 		ext4_set_bit(i, bitmap);
61 	if (i < end_bit)
62 		memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
63 }
64 
65 /* Initializes an uninitialized inode bitmap */
66 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
67 				ext4_group_t block_group,
68 				struct ext4_group_desc *gdp)
69 {
70 	struct ext4_sb_info *sbi = EXT4_SB(sb);
71 
72 	J_ASSERT_BH(bh, buffer_locked(bh));
73 
74 	/* If checksum is bad mark all blocks and inodes use to prevent
75 	 * allocation, essentially implementing a per-group read-only flag. */
76 	if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
77 		ext4_error(sb, __func__, "Checksum bad for group %u",
78 			   block_group);
79 		ext4_free_blks_set(sb, gdp, 0);
80 		ext4_free_inodes_set(sb, gdp, 0);
81 		ext4_itable_unused_set(sb, gdp, 0);
82 		memset(bh->b_data, 0xff, sb->s_blocksize);
83 		return 0;
84 	}
85 
86 	memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
87 	mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
88 			bh->b_data);
89 
90 	return EXT4_INODES_PER_GROUP(sb);
91 }
92 
93 /*
94  * Read the inode allocation bitmap for a given block_group, reading
95  * into the specified slot in the superblock's bitmap cache.
96  *
97  * Return buffer_head of bitmap on success or NULL.
98  */
99 static struct buffer_head *
100 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
101 {
102 	struct ext4_group_desc *desc;
103 	struct buffer_head *bh = NULL;
104 	ext4_fsblk_t bitmap_blk;
105 
106 	desc = ext4_get_group_desc(sb, block_group, NULL);
107 	if (!desc)
108 		return NULL;
109 	bitmap_blk = ext4_inode_bitmap(sb, desc);
110 	bh = sb_getblk(sb, bitmap_blk);
111 	if (unlikely(!bh)) {
112 		ext4_error(sb, __func__,
113 			    "Cannot read inode bitmap - "
114 			    "block_group = %u, inode_bitmap = %llu",
115 			    block_group, bitmap_blk);
116 		return NULL;
117 	}
118 	if (bitmap_uptodate(bh))
119 		return bh;
120 
121 	lock_buffer(bh);
122 	if (bitmap_uptodate(bh)) {
123 		unlock_buffer(bh);
124 		return bh;
125 	}
126 	spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
127 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
128 		ext4_init_inode_bitmap(sb, bh, block_group, desc);
129 		set_bitmap_uptodate(bh);
130 		set_buffer_uptodate(bh);
131 		spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
132 		unlock_buffer(bh);
133 		return bh;
134 	}
135 	spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
136 	if (buffer_uptodate(bh)) {
137 		/*
138 		 * if not uninit if bh is uptodate,
139 		 * bitmap is also uptodate
140 		 */
141 		set_bitmap_uptodate(bh);
142 		unlock_buffer(bh);
143 		return bh;
144 	}
145 	/*
146 	 * submit the buffer_head for read. We can
147 	 * safely mark the bitmap as uptodate now.
148 	 * We do it here so the bitmap uptodate bit
149 	 * get set with buffer lock held.
150 	 */
151 	set_bitmap_uptodate(bh);
152 	if (bh_submit_read(bh) < 0) {
153 		put_bh(bh);
154 		ext4_error(sb, __func__,
155 			    "Cannot read inode bitmap - "
156 			    "block_group = %u, inode_bitmap = %llu",
157 			    block_group, bitmap_blk);
158 		return NULL;
159 	}
160 	return bh;
161 }
162 
163 /*
164  * NOTE! When we get the inode, we're the only people
165  * that have access to it, and as such there are no
166  * race conditions we have to worry about. The inode
167  * is not on the hash-lists, and it cannot be reached
168  * through the filesystem because the directory entry
169  * has been deleted earlier.
170  *
171  * HOWEVER: we must make sure that we get no aliases,
172  * which means that we have to call "clear_inode()"
173  * _before_ we mark the inode not in use in the inode
174  * bitmaps. Otherwise a newly created file might use
175  * the same inode number (not actually the same pointer
176  * though), and then we'd have two inodes sharing the
177  * same inode number and space on the harddisk.
178  */
179 void ext4_free_inode(handle_t *handle, struct inode *inode)
180 {
181 	struct super_block *sb = inode->i_sb;
182 	int is_directory;
183 	unsigned long ino;
184 	struct buffer_head *bitmap_bh = NULL;
185 	struct buffer_head *bh2;
186 	ext4_group_t block_group;
187 	unsigned long bit;
188 	struct ext4_group_desc *gdp;
189 	struct ext4_super_block *es;
190 	struct ext4_sb_info *sbi;
191 	int fatal = 0, err, count, cleared;
192 	ext4_group_t flex_group;
193 
194 	if (atomic_read(&inode->i_count) > 1) {
195 		printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
196 		       atomic_read(&inode->i_count));
197 		return;
198 	}
199 	if (inode->i_nlink) {
200 		printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
201 		       inode->i_nlink);
202 		return;
203 	}
204 	if (!sb) {
205 		printk(KERN_ERR "ext4_free_inode: inode on "
206 		       "nonexistent device\n");
207 		return;
208 	}
209 	sbi = EXT4_SB(sb);
210 
211 	ino = inode->i_ino;
212 	ext4_debug("freeing inode %lu\n", ino);
213 	trace_mark(ext4_free_inode,
214 		   "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu",
215 		   sb->s_id, inode->i_ino, inode->i_mode,
216 		   (unsigned long) inode->i_uid, (unsigned long) inode->i_gid,
217 		   (unsigned long long) inode->i_blocks);
218 
219 	/*
220 	 * Note: we must free any quota before locking the superblock,
221 	 * as writing the quota to disk may need the lock as well.
222 	 */
223 	DQUOT_INIT(inode);
224 	ext4_xattr_delete_inode(handle, inode);
225 	DQUOT_FREE_INODE(inode);
226 	DQUOT_DROP(inode);
227 
228 	is_directory = S_ISDIR(inode->i_mode);
229 
230 	/* Do this BEFORE marking the inode not in use or returning an error */
231 	clear_inode(inode);
232 
233 	es = EXT4_SB(sb)->s_es;
234 	if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
235 		ext4_error(sb, "ext4_free_inode",
236 			   "reserved or nonexistent inode %lu", ino);
237 		goto error_return;
238 	}
239 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
240 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
241 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
242 	if (!bitmap_bh)
243 		goto error_return;
244 
245 	BUFFER_TRACE(bitmap_bh, "get_write_access");
246 	fatal = ext4_journal_get_write_access(handle, bitmap_bh);
247 	if (fatal)
248 		goto error_return;
249 
250 	/* Ok, now we can actually update the inode bitmaps.. */
251 	spin_lock(sb_bgl_lock(sbi, block_group));
252 	cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
253 	spin_unlock(sb_bgl_lock(sbi, block_group));
254 	if (!cleared)
255 		ext4_error(sb, "ext4_free_inode",
256 			   "bit already cleared for inode %lu", ino);
257 	else {
258 		gdp = ext4_get_group_desc(sb, block_group, &bh2);
259 
260 		BUFFER_TRACE(bh2, "get_write_access");
261 		fatal = ext4_journal_get_write_access(handle, bh2);
262 		if (fatal) goto error_return;
263 
264 		if (gdp) {
265 			spin_lock(sb_bgl_lock(sbi, block_group));
266 			count = ext4_free_inodes_count(sb, gdp) + 1;
267 			ext4_free_inodes_set(sb, gdp, count);
268 			if (is_directory) {
269 				count = ext4_used_dirs_count(sb, gdp) - 1;
270 				ext4_used_dirs_set(sb, gdp, count);
271 			}
272 			gdp->bg_checksum = ext4_group_desc_csum(sbi,
273 							block_group, gdp);
274 			spin_unlock(sb_bgl_lock(sbi, block_group));
275 			percpu_counter_inc(&sbi->s_freeinodes_counter);
276 			if (is_directory)
277 				percpu_counter_dec(&sbi->s_dirs_counter);
278 
279 			if (sbi->s_log_groups_per_flex) {
280 				flex_group = ext4_flex_group(sbi, block_group);
281 				spin_lock(sb_bgl_lock(sbi, flex_group));
282 				sbi->s_flex_groups[flex_group].free_inodes++;
283 				spin_unlock(sb_bgl_lock(sbi, flex_group));
284 			}
285 		}
286 		BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
287 		err = ext4_handle_dirty_metadata(handle, NULL, bh2);
288 		if (!fatal) fatal = err;
289 	}
290 	BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
291 	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
292 	if (!fatal)
293 		fatal = err;
294 	sb->s_dirt = 1;
295 error_return:
296 	brelse(bitmap_bh);
297 	ext4_std_error(sb, fatal);
298 }
299 
300 /*
301  * There are two policies for allocating an inode.  If the new inode is
302  * a directory, then a forward search is made for a block group with both
303  * free space and a low directory-to-inode ratio; if that fails, then of
304  * the groups with above-average free space, that group with the fewest
305  * directories already is chosen.
306  *
307  * For other inodes, search forward from the parent directory\'s block
308  * group to find a free inode.
309  */
310 static int find_group_dir(struct super_block *sb, struct inode *parent,
311 				ext4_group_t *best_group)
312 {
313 	ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
314 	unsigned int freei, avefreei;
315 	struct ext4_group_desc *desc, *best_desc = NULL;
316 	ext4_group_t group;
317 	int ret = -1;
318 
319 	freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
320 	avefreei = freei / ngroups;
321 
322 	for (group = 0; group < ngroups; group++) {
323 		desc = ext4_get_group_desc(sb, group, NULL);
324 		if (!desc || !ext4_free_inodes_count(sb, desc))
325 			continue;
326 		if (ext4_free_inodes_count(sb, desc) < avefreei)
327 			continue;
328 		if (!best_desc ||
329 		    (ext4_free_blks_count(sb, desc) >
330 		     ext4_free_blks_count(sb, best_desc))) {
331 			*best_group = group;
332 			best_desc = desc;
333 			ret = 0;
334 		}
335 	}
336 	return ret;
337 }
338 
339 #define free_block_ratio 10
340 
341 static int find_group_flex(struct super_block *sb, struct inode *parent,
342 			   ext4_group_t *best_group)
343 {
344 	struct ext4_sb_info *sbi = EXT4_SB(sb);
345 	struct ext4_group_desc *desc;
346 	struct buffer_head *bh;
347 	struct flex_groups *flex_group = sbi->s_flex_groups;
348 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
349 	ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
350 	ext4_group_t ngroups = sbi->s_groups_count;
351 	int flex_size = ext4_flex_bg_size(sbi);
352 	ext4_group_t best_flex = parent_fbg_group;
353 	int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
354 	int flexbg_free_blocks;
355 	int flex_freeb_ratio;
356 	ext4_group_t n_fbg_groups;
357 	ext4_group_t i;
358 
359 	n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >>
360 		sbi->s_log_groups_per_flex;
361 
362 find_close_to_parent:
363 	flexbg_free_blocks = flex_group[best_flex].free_blocks;
364 	flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
365 	if (flex_group[best_flex].free_inodes &&
366 	    flex_freeb_ratio > free_block_ratio)
367 		goto found_flexbg;
368 
369 	if (best_flex && best_flex == parent_fbg_group) {
370 		best_flex--;
371 		goto find_close_to_parent;
372 	}
373 
374 	for (i = 0; i < n_fbg_groups; i++) {
375 		if (i == parent_fbg_group || i == parent_fbg_group - 1)
376 			continue;
377 
378 		flexbg_free_blocks = flex_group[i].free_blocks;
379 		flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
380 
381 		if (flex_freeb_ratio > free_block_ratio &&
382 		    flex_group[i].free_inodes) {
383 			best_flex = i;
384 			goto found_flexbg;
385 		}
386 
387 		if (flex_group[best_flex].free_inodes == 0 ||
388 		    (flex_group[i].free_blocks >
389 		     flex_group[best_flex].free_blocks &&
390 		     flex_group[i].free_inodes))
391 			best_flex = i;
392 	}
393 
394 	if (!flex_group[best_flex].free_inodes ||
395 	    !flex_group[best_flex].free_blocks)
396 		return -1;
397 
398 found_flexbg:
399 	for (i = best_flex * flex_size; i < ngroups &&
400 		     i < (best_flex + 1) * flex_size; i++) {
401 		desc = ext4_get_group_desc(sb, i, &bh);
402 		if (ext4_free_inodes_count(sb, desc)) {
403 			*best_group = i;
404 			goto out;
405 		}
406 	}
407 
408 	return -1;
409 out:
410 	return 0;
411 }
412 
413 /*
414  * Orlov's allocator for directories.
415  *
416  * We always try to spread first-level directories.
417  *
418  * If there are blockgroups with both free inodes and free blocks counts
419  * not worse than average we return one with smallest directory count.
420  * Otherwise we simply return a random group.
421  *
422  * For the rest rules look so:
423  *
424  * It's OK to put directory into a group unless
425  * it has too many directories already (max_dirs) or
426  * it has too few free inodes left (min_inodes) or
427  * it has too few free blocks left (min_blocks) or
428  * it's already running too large debt (max_debt).
429  * Parent's group is preferred, if it doesn't satisfy these
430  * conditions we search cyclically through the rest. If none
431  * of the groups look good we just look for a group with more
432  * free inodes than average (starting at parent's group).
433  *
434  * Debt is incremented each time we allocate a directory and decremented
435  * when we allocate an inode, within 0--255.
436  */
437 
438 #define INODE_COST 64
439 #define BLOCK_COST 256
440 
441 static int find_group_orlov(struct super_block *sb, struct inode *parent,
442 				ext4_group_t *group)
443 {
444 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
445 	struct ext4_sb_info *sbi = EXT4_SB(sb);
446 	struct ext4_super_block *es = sbi->s_es;
447 	ext4_group_t ngroups = sbi->s_groups_count;
448 	int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
449 	unsigned int freei, avefreei;
450 	ext4_fsblk_t freeb, avefreeb;
451 	ext4_fsblk_t blocks_per_dir;
452 	unsigned int ndirs;
453 	int max_debt, max_dirs, min_inodes;
454 	ext4_grpblk_t min_blocks;
455 	ext4_group_t i;
456 	struct ext4_group_desc *desc;
457 
458 	freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
459 	avefreei = freei / ngroups;
460 	freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
461 	avefreeb = freeb;
462 	do_div(avefreeb, ngroups);
463 	ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
464 
465 	if ((parent == sb->s_root->d_inode) ||
466 	    (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
467 		int best_ndir = inodes_per_group;
468 		ext4_group_t grp;
469 		int ret = -1;
470 
471 		get_random_bytes(&grp, sizeof(grp));
472 		parent_group = (unsigned)grp % ngroups;
473 		for (i = 0; i < ngroups; i++) {
474 			grp = (parent_group + i) % ngroups;
475 			desc = ext4_get_group_desc(sb, grp, NULL);
476 			if (!desc || !ext4_free_inodes_count(sb, desc))
477 				continue;
478 			if (ext4_used_dirs_count(sb, desc) >= best_ndir)
479 				continue;
480 			if (ext4_free_inodes_count(sb, desc) < avefreei)
481 				continue;
482 			if (ext4_free_blks_count(sb, desc) < avefreeb)
483 				continue;
484 			*group = grp;
485 			ret = 0;
486 			best_ndir = ext4_used_dirs_count(sb, desc);
487 		}
488 		if (ret == 0)
489 			return ret;
490 		goto fallback;
491 	}
492 
493 	blocks_per_dir = ext4_blocks_count(es) - freeb;
494 	do_div(blocks_per_dir, ndirs);
495 
496 	max_dirs = ndirs / ngroups + inodes_per_group / 16;
497 	min_inodes = avefreei - inodes_per_group / 4;
498 	min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
499 
500 	max_debt = EXT4_BLOCKS_PER_GROUP(sb);
501 	max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
502 	if (max_debt * INODE_COST > inodes_per_group)
503 		max_debt = inodes_per_group / INODE_COST;
504 	if (max_debt > 255)
505 		max_debt = 255;
506 	if (max_debt == 0)
507 		max_debt = 1;
508 
509 	for (i = 0; i < ngroups; i++) {
510 		*group = (parent_group + i) % ngroups;
511 		desc = ext4_get_group_desc(sb, *group, NULL);
512 		if (!desc || !ext4_free_inodes_count(sb, desc))
513 			continue;
514 		if (ext4_used_dirs_count(sb, desc) >= max_dirs)
515 			continue;
516 		if (ext4_free_inodes_count(sb, desc) < min_inodes)
517 			continue;
518 		if (ext4_free_blks_count(sb, desc) < min_blocks)
519 			continue;
520 		return 0;
521 	}
522 
523 fallback:
524 	for (i = 0; i < ngroups; i++) {
525 		*group = (parent_group + i) % ngroups;
526 		desc = ext4_get_group_desc(sb, *group, NULL);
527 		if (desc && ext4_free_inodes_count(sb, desc) &&
528 			ext4_free_inodes_count(sb, desc) >= avefreei)
529 			return 0;
530 	}
531 
532 	if (avefreei) {
533 		/*
534 		 * The free-inodes counter is approximate, and for really small
535 		 * filesystems the above test can fail to find any blockgroups
536 		 */
537 		avefreei = 0;
538 		goto fallback;
539 	}
540 
541 	return -1;
542 }
543 
544 static int find_group_other(struct super_block *sb, struct inode *parent,
545 				ext4_group_t *group)
546 {
547 	ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
548 	ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
549 	struct ext4_group_desc *desc;
550 	ext4_group_t i;
551 
552 	/*
553 	 * Try to place the inode in its parent directory
554 	 */
555 	*group = parent_group;
556 	desc = ext4_get_group_desc(sb, *group, NULL);
557 	if (desc && ext4_free_inodes_count(sb, desc) &&
558 			ext4_free_blks_count(sb, desc))
559 		return 0;
560 
561 	/*
562 	 * We're going to place this inode in a different blockgroup from its
563 	 * parent.  We want to cause files in a common directory to all land in
564 	 * the same blockgroup.  But we want files which are in a different
565 	 * directory which shares a blockgroup with our parent to land in a
566 	 * different blockgroup.
567 	 *
568 	 * So add our directory's i_ino into the starting point for the hash.
569 	 */
570 	*group = (*group + parent->i_ino) % ngroups;
571 
572 	/*
573 	 * Use a quadratic hash to find a group with a free inode and some free
574 	 * blocks.
575 	 */
576 	for (i = 1; i < ngroups; i <<= 1) {
577 		*group += i;
578 		if (*group >= ngroups)
579 			*group -= ngroups;
580 		desc = ext4_get_group_desc(sb, *group, NULL);
581 		if (desc && ext4_free_inodes_count(sb, desc) &&
582 				ext4_free_blks_count(sb, desc))
583 			return 0;
584 	}
585 
586 	/*
587 	 * That failed: try linear search for a free inode, even if that group
588 	 * has no free blocks.
589 	 */
590 	*group = parent_group;
591 	for (i = 0; i < ngroups; i++) {
592 		if (++*group >= ngroups)
593 			*group = 0;
594 		desc = ext4_get_group_desc(sb, *group, NULL);
595 		if (desc && ext4_free_inodes_count(sb, desc))
596 			return 0;
597 	}
598 
599 	return -1;
600 }
601 
602 /*
603  * claim the inode from the inode bitmap. If the group
604  * is uninit we need to take the groups's sb_bgl_lock
605  * and clear the uninit flag. The inode bitmap update
606  * and group desc uninit flag clear should be done
607  * after holding sb_bgl_lock so that ext4_read_inode_bitmap
608  * doesn't race with the ext4_claim_inode
609  */
610 static int ext4_claim_inode(struct super_block *sb,
611 			struct buffer_head *inode_bitmap_bh,
612 			unsigned long ino, ext4_group_t group, int mode)
613 {
614 	int free = 0, retval = 0, count;
615 	struct ext4_sb_info *sbi = EXT4_SB(sb);
616 	struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
617 
618 	spin_lock(sb_bgl_lock(sbi, group));
619 	if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
620 		/* not a free inode */
621 		retval = 1;
622 		goto err_ret;
623 	}
624 	ino++;
625 	if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
626 			ino > EXT4_INODES_PER_GROUP(sb)) {
627 		spin_unlock(sb_bgl_lock(sbi, group));
628 		ext4_error(sb, __func__,
629 			   "reserved inode or inode > inodes count - "
630 			   "block_group = %u, inode=%lu", group,
631 			   ino + group * EXT4_INODES_PER_GROUP(sb));
632 		return 1;
633 	}
634 	/* If we didn't allocate from within the initialized part of the inode
635 	 * table then we need to initialize up to this inode. */
636 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
637 
638 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
639 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
640 			/* When marking the block group with
641 			 * ~EXT4_BG_INODE_UNINIT we don't want to depend
642 			 * on the value of bg_itable_unused even though
643 			 * mke2fs could have initialized the same for us.
644 			 * Instead we calculated the value below
645 			 */
646 
647 			free = 0;
648 		} else {
649 			free = EXT4_INODES_PER_GROUP(sb) -
650 				ext4_itable_unused_count(sb, gdp);
651 		}
652 
653 		/*
654 		 * Check the relative inode number against the last used
655 		 * relative inode number in this group. if it is greater
656 		 * we need to  update the bg_itable_unused count
657 		 *
658 		 */
659 		if (ino > free)
660 			ext4_itable_unused_set(sb, gdp,
661 					(EXT4_INODES_PER_GROUP(sb) - ino));
662 	}
663 	count = ext4_free_inodes_count(sb, gdp) - 1;
664 	ext4_free_inodes_set(sb, gdp, count);
665 	if (S_ISDIR(mode)) {
666 		count = ext4_used_dirs_count(sb, gdp) + 1;
667 		ext4_used_dirs_set(sb, gdp, count);
668 	}
669 	gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
670 err_ret:
671 	spin_unlock(sb_bgl_lock(sbi, group));
672 	return retval;
673 }
674 
675 /*
676  * There are two policies for allocating an inode.  If the new inode is
677  * a directory, then a forward search is made for a block group with both
678  * free space and a low directory-to-inode ratio; if that fails, then of
679  * the groups with above-average free space, that group with the fewest
680  * directories already is chosen.
681  *
682  * For other inodes, search forward from the parent directory's block
683  * group to find a free inode.
684  */
685 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
686 {
687 	struct super_block *sb;
688 	struct buffer_head *inode_bitmap_bh = NULL;
689 	struct buffer_head *group_desc_bh;
690 	ext4_group_t group = 0;
691 	unsigned long ino = 0;
692 	struct inode *inode;
693 	struct ext4_group_desc *gdp = NULL;
694 	struct ext4_super_block *es;
695 	struct ext4_inode_info *ei;
696 	struct ext4_sb_info *sbi;
697 	int ret2, err = 0;
698 	struct inode *ret;
699 	ext4_group_t i;
700 	int free = 0;
701 	static int once = 1;
702 	ext4_group_t flex_group;
703 
704 	/* Cannot create files in a deleted directory */
705 	if (!dir || !dir->i_nlink)
706 		return ERR_PTR(-EPERM);
707 
708 	sb = dir->i_sb;
709 	trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id,
710 		   dir->i_ino, mode);
711 	inode = new_inode(sb);
712 	if (!inode)
713 		return ERR_PTR(-ENOMEM);
714 	ei = EXT4_I(inode);
715 
716 	sbi = EXT4_SB(sb);
717 	es = sbi->s_es;
718 
719 	if (sbi->s_log_groups_per_flex) {
720 		ret2 = find_group_flex(sb, dir, &group);
721 		if (ret2 == -1) {
722 			ret2 = find_group_other(sb, dir, &group);
723 			if (ret2 == 0 && once)
724 				once = 0;
725 				printk(KERN_NOTICE "ext4: find_group_flex "
726 				       "failed, fallback succeeded dir %lu\n",
727 				       dir->i_ino);
728 		}
729 		goto got_group;
730 	}
731 
732 	if (S_ISDIR(mode)) {
733 		if (test_opt(sb, OLDALLOC))
734 			ret2 = find_group_dir(sb, dir, &group);
735 		else
736 			ret2 = find_group_orlov(sb, dir, &group);
737 	} else
738 		ret2 = find_group_other(sb, dir, &group);
739 
740 got_group:
741 	err = -ENOSPC;
742 	if (ret2 == -1)
743 		goto out;
744 
745 	for (i = 0; i < sbi->s_groups_count; i++) {
746 		err = -EIO;
747 
748 		gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
749 		if (!gdp)
750 			goto fail;
751 
752 		brelse(inode_bitmap_bh);
753 		inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
754 		if (!inode_bitmap_bh)
755 			goto fail;
756 
757 		ino = 0;
758 
759 repeat_in_this_group:
760 		ino = ext4_find_next_zero_bit((unsigned long *)
761 					      inode_bitmap_bh->b_data,
762 					      EXT4_INODES_PER_GROUP(sb), ino);
763 
764 		if (ino < EXT4_INODES_PER_GROUP(sb)) {
765 
766 			BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
767 			err = ext4_journal_get_write_access(handle,
768 							    inode_bitmap_bh);
769 			if (err)
770 				goto fail;
771 
772 			BUFFER_TRACE(group_desc_bh, "get_write_access");
773 			err = ext4_journal_get_write_access(handle,
774 								group_desc_bh);
775 			if (err)
776 				goto fail;
777 			if (!ext4_claim_inode(sb, inode_bitmap_bh,
778 						ino, group, mode)) {
779 				/* we won it */
780 				BUFFER_TRACE(inode_bitmap_bh,
781 					"call ext4_handle_dirty_metadata");
782 				err = ext4_handle_dirty_metadata(handle,
783 								 inode,
784 							inode_bitmap_bh);
785 				if (err)
786 					goto fail;
787 				/* zero bit is inode number 1*/
788 				ino++;
789 				goto got;
790 			}
791 			/* we lost it */
792 			ext4_handle_release_buffer(handle, inode_bitmap_bh);
793 			ext4_handle_release_buffer(handle, group_desc_bh);
794 
795 			if (++ino < EXT4_INODES_PER_GROUP(sb))
796 				goto repeat_in_this_group;
797 		}
798 
799 		/*
800 		 * This case is possible in concurrent environment.  It is very
801 		 * rare.  We cannot repeat the find_group_xxx() call because
802 		 * that will simply return the same blockgroup, because the
803 		 * group descriptor metadata has not yet been updated.
804 		 * So we just go onto the next blockgroup.
805 		 */
806 		if (++group == sbi->s_groups_count)
807 			group = 0;
808 	}
809 	err = -ENOSPC;
810 	goto out;
811 
812 got:
813 	/* We may have to initialize the block bitmap if it isn't already */
814 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
815 	    gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
816 		struct buffer_head *block_bitmap_bh;
817 
818 		block_bitmap_bh = ext4_read_block_bitmap(sb, group);
819 		BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
820 		err = ext4_journal_get_write_access(handle, block_bitmap_bh);
821 		if (err) {
822 			brelse(block_bitmap_bh);
823 			goto fail;
824 		}
825 
826 		free = 0;
827 		spin_lock(sb_bgl_lock(sbi, group));
828 		/* recheck and clear flag under lock if we still need to */
829 		if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
830 			free = ext4_free_blocks_after_init(sb, group, gdp);
831 			gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
832 			ext4_free_blks_set(sb, gdp, free);
833 			gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
834 								gdp);
835 		}
836 		spin_unlock(sb_bgl_lock(sbi, group));
837 
838 		/* Don't need to dirty bitmap block if we didn't change it */
839 		if (free) {
840 			BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
841 			err = ext4_handle_dirty_metadata(handle,
842 							NULL, block_bitmap_bh);
843 		}
844 
845 		brelse(block_bitmap_bh);
846 		if (err)
847 			goto fail;
848 	}
849 	BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
850 	err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
851 	if (err)
852 		goto fail;
853 
854 	percpu_counter_dec(&sbi->s_freeinodes_counter);
855 	if (S_ISDIR(mode))
856 		percpu_counter_inc(&sbi->s_dirs_counter);
857 	sb->s_dirt = 1;
858 
859 	if (sbi->s_log_groups_per_flex) {
860 		flex_group = ext4_flex_group(sbi, group);
861 		spin_lock(sb_bgl_lock(sbi, flex_group));
862 		sbi->s_flex_groups[flex_group].free_inodes--;
863 		spin_unlock(sb_bgl_lock(sbi, flex_group));
864 	}
865 
866 	inode->i_uid = current_fsuid();
867 	if (test_opt(sb, GRPID))
868 		inode->i_gid = dir->i_gid;
869 	else if (dir->i_mode & S_ISGID) {
870 		inode->i_gid = dir->i_gid;
871 		if (S_ISDIR(mode))
872 			mode |= S_ISGID;
873 	} else
874 		inode->i_gid = current_fsgid();
875 	inode->i_mode = mode;
876 
877 	inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
878 	/* This is the optimal IO size (for stat), not the fs block size */
879 	inode->i_blocks = 0;
880 	inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
881 						       ext4_current_time(inode);
882 
883 	memset(ei->i_data, 0, sizeof(ei->i_data));
884 	ei->i_dir_start_lookup = 0;
885 	ei->i_disksize = 0;
886 
887 	/*
888 	 * Don't inherit extent flag from directory. We set extent flag on
889 	 * newly created directory and file only if -o extent mount option is
890 	 * specified
891 	 */
892 	ei->i_flags = EXT4_I(dir)->i_flags & ~(EXT4_INDEX_FL|EXT4_EXTENTS_FL);
893 	if (S_ISLNK(mode))
894 		ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
895 	/* dirsync only applies to directories */
896 	if (!S_ISDIR(mode))
897 		ei->i_flags &= ~EXT4_DIRSYNC_FL;
898 	ei->i_file_acl = 0;
899 	ei->i_dtime = 0;
900 	ei->i_block_group = group;
901 
902 	ext4_set_inode_flags(inode);
903 	if (IS_DIRSYNC(inode))
904 		ext4_handle_sync(handle);
905 	if (insert_inode_locked(inode) < 0) {
906 		err = -EINVAL;
907 		goto fail_drop;
908 	}
909 	spin_lock(&sbi->s_next_gen_lock);
910 	inode->i_generation = sbi->s_next_generation++;
911 	spin_unlock(&sbi->s_next_gen_lock);
912 
913 	ei->i_state = EXT4_STATE_NEW;
914 
915 	ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
916 
917 	ret = inode;
918 	if (DQUOT_ALLOC_INODE(inode)) {
919 		err = -EDQUOT;
920 		goto fail_drop;
921 	}
922 
923 	err = ext4_init_acl(handle, inode, dir);
924 	if (err)
925 		goto fail_free_drop;
926 
927 	err = ext4_init_security(handle, inode, dir);
928 	if (err)
929 		goto fail_free_drop;
930 
931 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
932 		/* set extent flag only for directory, file and normal symlink*/
933 		if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
934 			EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
935 			ext4_ext_tree_init(handle, inode);
936 		}
937 	}
938 
939 	err = ext4_mark_inode_dirty(handle, inode);
940 	if (err) {
941 		ext4_std_error(sb, err);
942 		goto fail_free_drop;
943 	}
944 
945 	ext4_debug("allocating inode %lu\n", inode->i_ino);
946 	trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d",
947 		   sb->s_id, inode->i_ino, dir->i_ino, mode);
948 	goto really_out;
949 fail:
950 	ext4_std_error(sb, err);
951 out:
952 	iput(inode);
953 	ret = ERR_PTR(err);
954 really_out:
955 	brelse(inode_bitmap_bh);
956 	return ret;
957 
958 fail_free_drop:
959 	DQUOT_FREE_INODE(inode);
960 
961 fail_drop:
962 	DQUOT_DROP(inode);
963 	inode->i_flags |= S_NOQUOTA;
964 	inode->i_nlink = 0;
965 	unlock_new_inode(inode);
966 	iput(inode);
967 	brelse(inode_bitmap_bh);
968 	return ERR_PTR(err);
969 }
970 
971 /* Verify that we are loading a valid orphan from disk */
972 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
973 {
974 	unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
975 	ext4_group_t block_group;
976 	int bit;
977 	struct buffer_head *bitmap_bh;
978 	struct inode *inode = NULL;
979 	long err = -EIO;
980 
981 	/* Error cases - e2fsck has already cleaned up for us */
982 	if (ino > max_ino) {
983 		ext4_warning(sb, __func__,
984 			     "bad orphan ino %lu!  e2fsck was run?", ino);
985 		goto error;
986 	}
987 
988 	block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
989 	bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
990 	bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
991 	if (!bitmap_bh) {
992 		ext4_warning(sb, __func__,
993 			     "inode bitmap error for orphan %lu", ino);
994 		goto error;
995 	}
996 
997 	/* Having the inode bit set should be a 100% indicator that this
998 	 * is a valid orphan (no e2fsck run on fs).  Orphans also include
999 	 * inodes that were being truncated, so we can't check i_nlink==0.
1000 	 */
1001 	if (!ext4_test_bit(bit, bitmap_bh->b_data))
1002 		goto bad_orphan;
1003 
1004 	inode = ext4_iget(sb, ino);
1005 	if (IS_ERR(inode))
1006 		goto iget_failed;
1007 
1008 	/*
1009 	 * If the orphans has i_nlinks > 0 then it should be able to be
1010 	 * truncated, otherwise it won't be removed from the orphan list
1011 	 * during processing and an infinite loop will result.
1012 	 */
1013 	if (inode->i_nlink && !ext4_can_truncate(inode))
1014 		goto bad_orphan;
1015 
1016 	if (NEXT_ORPHAN(inode) > max_ino)
1017 		goto bad_orphan;
1018 	brelse(bitmap_bh);
1019 	return inode;
1020 
1021 iget_failed:
1022 	err = PTR_ERR(inode);
1023 	inode = NULL;
1024 bad_orphan:
1025 	ext4_warning(sb, __func__,
1026 		     "bad orphan inode %lu!  e2fsck was run?", ino);
1027 	printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1028 	       bit, (unsigned long long)bitmap_bh->b_blocknr,
1029 	       ext4_test_bit(bit, bitmap_bh->b_data));
1030 	printk(KERN_NOTICE "inode=%p\n", inode);
1031 	if (inode) {
1032 		printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1033 		       is_bad_inode(inode));
1034 		printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1035 		       NEXT_ORPHAN(inode));
1036 		printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1037 		printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1038 		/* Avoid freeing blocks if we got a bad deleted inode */
1039 		if (inode->i_nlink == 0)
1040 			inode->i_blocks = 0;
1041 		iput(inode);
1042 	}
1043 	brelse(bitmap_bh);
1044 error:
1045 	return ERR_PTR(err);
1046 }
1047 
1048 unsigned long ext4_count_free_inodes(struct super_block *sb)
1049 {
1050 	unsigned long desc_count;
1051 	struct ext4_group_desc *gdp;
1052 	ext4_group_t i;
1053 #ifdef EXT4FS_DEBUG
1054 	struct ext4_super_block *es;
1055 	unsigned long bitmap_count, x;
1056 	struct buffer_head *bitmap_bh = NULL;
1057 
1058 	es = EXT4_SB(sb)->s_es;
1059 	desc_count = 0;
1060 	bitmap_count = 0;
1061 	gdp = NULL;
1062 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1063 		gdp = ext4_get_group_desc(sb, i, NULL);
1064 		if (!gdp)
1065 			continue;
1066 		desc_count += ext4_free_inodes_count(sb, gdp);
1067 		brelse(bitmap_bh);
1068 		bitmap_bh = ext4_read_inode_bitmap(sb, i);
1069 		if (!bitmap_bh)
1070 			continue;
1071 
1072 		x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1073 		printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1074 			i, ext4_free_inodes_count(sb, gdp), x);
1075 		bitmap_count += x;
1076 	}
1077 	brelse(bitmap_bh);
1078 	printk(KERN_DEBUG "ext4_count_free_inodes: "
1079 	       "stored = %u, computed = %lu, %lu\n",
1080 	       le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1081 	return desc_count;
1082 #else
1083 	desc_count = 0;
1084 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1085 		gdp = ext4_get_group_desc(sb, i, NULL);
1086 		if (!gdp)
1087 			continue;
1088 		desc_count += ext4_free_inodes_count(sb, gdp);
1089 		cond_resched();
1090 	}
1091 	return desc_count;
1092 #endif
1093 }
1094 
1095 /* Called at mount-time, super-block is locked */
1096 unsigned long ext4_count_dirs(struct super_block * sb)
1097 {
1098 	unsigned long count = 0;
1099 	ext4_group_t i;
1100 
1101 	for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1102 		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1103 		if (!gdp)
1104 			continue;
1105 		count += ext4_used_dirs_count(sb, gdp);
1106 	}
1107 	return count;
1108 }
1109