Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext4 / ialloc.c
blob57f6eef6ccd6c03afbd4c35d0fa364f4e350bbf8
1 /*
2 * linux/fs/ext4/ialloc.c
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)
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
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>
27 #include "ext4.h"
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
32 #include <trace/events/ext4.h>
35 * ialloc.c contains the inodes allocation and deallocation routines
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
53 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
55 int i;
57 if (start_bit >= end_bit)
58 return;
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67 /* Initializes an uninitialized inode bitmap */
68 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
69 ext4_group_t block_group,
70 struct ext4_group_desc *gdp)
72 struct ext4_sb_info *sbi = EXT4_SB(sb);
74 J_ASSERT_BH(bh, buffer_locked(bh));
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(sbi, block_group, gdp)) {
79 ext4_error(sb, "Checksum bad for group %u", block_group);
80 ext4_free_blks_set(sb, gdp, 0);
81 ext4_free_inodes_set(sb, gdp, 0);
82 ext4_itable_unused_set(sb, gdp, 0);
83 memset(bh->b_data, 0xff, sb->s_blocksize);
84 return 0;
87 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
88 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
89 bh->b_data);
91 return EXT4_INODES_PER_GROUP(sb);
95 * Read the inode allocation bitmap for a given block_group, reading
96 * into the specified slot in the superblock's bitmap cache.
98 * Return buffer_head of bitmap on success or NULL.
100 static struct buffer_head *
101 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
103 struct ext4_group_desc *desc;
104 struct buffer_head *bh = NULL;
105 ext4_fsblk_t bitmap_blk;
107 desc = ext4_get_group_desc(sb, block_group, NULL);
108 if (!desc)
109 return NULL;
110 bitmap_blk = ext4_inode_bitmap(sb, desc);
111 bh = sb_getblk(sb, bitmap_blk);
112 if (unlikely(!bh)) {
113 ext4_error(sb, "Cannot read inode bitmap - "
114 "block_group = %u, inode_bitmap = %llu",
115 block_group, bitmap_blk);
116 return NULL;
118 if (bitmap_uptodate(bh))
119 return bh;
121 lock_buffer(bh);
122 if (bitmap_uptodate(bh)) {
123 unlock_buffer(bh);
124 return bh;
126 ext4_lock_group(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 ext4_unlock_group(sb, block_group);
132 unlock_buffer(bh);
133 return bh;
135 ext4_unlock_group(sb, block_group);
136 if (buffer_uptodate(bh)) {
138 * if not uninit if bh is uptodate,
139 * bitmap is also uptodate
141 set_bitmap_uptodate(bh);
142 unlock_buffer(bh);
143 return bh;
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.
151 set_bitmap_uptodate(bh);
152 if (bh_submit_read(bh) < 0) {
153 put_bh(bh);
154 ext4_error(sb, "Cannot read inode bitmap - "
155 "block_group = %u, inode_bitmap = %llu",
156 block_group, bitmap_blk);
157 return NULL;
159 return bh;
163 * NOTE! When we get the inode, we're the only people
164 * that have access to it, and as such there are no
165 * race conditions we have to worry about. The inode
166 * is not on the hash-lists, and it cannot be reached
167 * through the filesystem because the directory entry
168 * has been deleted earlier.
170 * HOWEVER: we must make sure that we get no aliases,
171 * which means that we have to call "clear_inode()"
172 * _before_ we mark the inode not in use in the inode
173 * bitmaps. Otherwise a newly created file might use
174 * the same inode number (not actually the same pointer
175 * though), and then we'd have two inodes sharing the
176 * same inode number and space on the harddisk.
178 void ext4_free_inode(handle_t *handle, struct inode *inode)
180 struct super_block *sb = inode->i_sb;
181 int is_directory;
182 unsigned long ino;
183 struct buffer_head *bitmap_bh = NULL;
184 struct buffer_head *bh2;
185 ext4_group_t block_group;
186 unsigned long bit;
187 struct ext4_group_desc *gdp;
188 struct ext4_super_block *es;
189 struct ext4_sb_info *sbi;
190 int fatal = 0, err, count, cleared;
192 if (atomic_read(&inode->i_count) > 1) {
193 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
194 atomic_read(&inode->i_count));
195 return;
197 if (inode->i_nlink) {
198 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
199 inode->i_nlink);
200 return;
202 if (!sb) {
203 printk(KERN_ERR "ext4_free_inode: inode on "
204 "nonexistent device\n");
205 return;
207 sbi = EXT4_SB(sb);
209 ino = inode->i_ino;
210 ext4_debug("freeing inode %lu\n", ino);
211 trace_ext4_free_inode(inode);
214 * Note: we must free any quota before locking the superblock,
215 * as writing the quota to disk may need the lock as well.
217 dquot_initialize(inode);
218 ext4_xattr_delete_inode(handle, inode);
219 dquot_free_inode(inode);
220 dquot_drop(inode);
222 is_directory = S_ISDIR(inode->i_mode);
224 /* Do this BEFORE marking the inode not in use or returning an error */
225 clear_inode(inode);
227 es = EXT4_SB(sb)->s_es;
228 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
229 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
230 goto error_return;
232 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
233 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
234 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
235 if (!bitmap_bh)
236 goto error_return;
238 BUFFER_TRACE(bitmap_bh, "get_write_access");
239 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
240 if (fatal)
241 goto error_return;
243 /* Ok, now we can actually update the inode bitmaps.. */
244 cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
245 bit, bitmap_bh->b_data);
246 if (!cleared)
247 ext4_error(sb, "bit already cleared for inode %lu", ino);
248 else {
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
251 BUFFER_TRACE(bh2, "get_write_access");
252 fatal = ext4_journal_get_write_access(handle, bh2);
253 if (fatal) goto error_return;
255 if (gdp) {
256 ext4_lock_group(sb, block_group);
257 count = ext4_free_inodes_count(sb, gdp) + 1;
258 ext4_free_inodes_set(sb, gdp, count);
259 if (is_directory) {
260 count = ext4_used_dirs_count(sb, gdp) - 1;
261 ext4_used_dirs_set(sb, gdp, count);
262 if (sbi->s_log_groups_per_flex) {
263 ext4_group_t f;
265 f = ext4_flex_group(sbi, block_group);
266 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
270 gdp->bg_checksum = ext4_group_desc_csum(sbi,
271 block_group, gdp);
272 ext4_unlock_group(sb, block_group);
273 percpu_counter_inc(&sbi->s_freeinodes_counter);
274 if (is_directory)
275 percpu_counter_dec(&sbi->s_dirs_counter);
277 if (sbi->s_log_groups_per_flex) {
278 ext4_group_t f;
280 f = ext4_flex_group(sbi, block_group);
281 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
284 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
285 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
286 if (!fatal) fatal = err;
288 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
289 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
290 if (!fatal)
291 fatal = err;
292 sb->s_dirt = 1;
293 error_return:
294 brelse(bitmap_bh);
295 ext4_std_error(sb, fatal);
299 * There are two policies for allocating an inode. If the new inode is
300 * a directory, then a forward search is made for a block group with both
301 * free space and a low directory-to-inode ratio; if that fails, then of
302 * the groups with above-average free space, that group with the fewest
303 * directories already is chosen.
305 * For other inodes, search forward from the parent directory\'s block
306 * group to find a free inode.
308 static int find_group_dir(struct super_block *sb, struct inode *parent,
309 ext4_group_t *best_group)
311 ext4_group_t ngroups = ext4_get_groups_count(sb);
312 unsigned int freei, avefreei;
313 struct ext4_group_desc *desc, *best_desc = NULL;
314 ext4_group_t group;
315 int ret = -1;
317 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
318 avefreei = freei / ngroups;
320 for (group = 0; group < ngroups; group++) {
321 desc = ext4_get_group_desc(sb, group, NULL);
322 if (!desc || !ext4_free_inodes_count(sb, desc))
323 continue;
324 if (ext4_free_inodes_count(sb, desc) < avefreei)
325 continue;
326 if (!best_desc ||
327 (ext4_free_blks_count(sb, desc) >
328 ext4_free_blks_count(sb, best_desc))) {
329 *best_group = group;
330 best_desc = desc;
331 ret = 0;
334 return ret;
337 #define free_block_ratio 10
339 static int find_group_flex(struct super_block *sb, struct inode *parent,
340 ext4_group_t *best_group)
342 struct ext4_sb_info *sbi = EXT4_SB(sb);
343 struct ext4_group_desc *desc;
344 struct flex_groups *flex_group = sbi->s_flex_groups;
345 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
346 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
347 ext4_group_t ngroups = ext4_get_groups_count(sb);
348 int flex_size = ext4_flex_bg_size(sbi);
349 ext4_group_t best_flex = parent_fbg_group;
350 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
351 int flexbg_free_blocks;
352 int flex_freeb_ratio;
353 ext4_group_t n_fbg_groups;
354 ext4_group_t i;
356 n_fbg_groups = (ngroups + flex_size - 1) >>
357 sbi->s_log_groups_per_flex;
359 find_close_to_parent:
360 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
361 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
362 if (atomic_read(&flex_group[best_flex].free_inodes) &&
363 flex_freeb_ratio > free_block_ratio)
364 goto found_flexbg;
366 if (best_flex && best_flex == parent_fbg_group) {
367 best_flex--;
368 goto find_close_to_parent;
371 for (i = 0; i < n_fbg_groups; i++) {
372 if (i == parent_fbg_group || i == parent_fbg_group - 1)
373 continue;
375 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
376 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
378 if (flex_freeb_ratio > free_block_ratio &&
379 (atomic_read(&flex_group[i].free_inodes))) {
380 best_flex = i;
381 goto found_flexbg;
384 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
385 ((atomic_read(&flex_group[i].free_blocks) >
386 atomic_read(&flex_group[best_flex].free_blocks)) &&
387 atomic_read(&flex_group[i].free_inodes)))
388 best_flex = i;
391 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
392 !atomic_read(&flex_group[best_flex].free_blocks))
393 return -1;
395 found_flexbg:
396 for (i = best_flex * flex_size; i < ngroups &&
397 i < (best_flex + 1) * flex_size; i++) {
398 desc = ext4_get_group_desc(sb, i, NULL);
399 if (ext4_free_inodes_count(sb, desc)) {
400 *best_group = i;
401 goto out;
405 return -1;
406 out:
407 return 0;
410 struct orlov_stats {
411 __u32 free_inodes;
412 __u32 free_blocks;
413 __u32 used_dirs;
417 * Helper function for Orlov's allocator; returns critical information
418 * for a particular block group or flex_bg. If flex_size is 1, then g
419 * is a block group number; otherwise it is flex_bg number.
421 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
422 int flex_size, struct orlov_stats *stats)
424 struct ext4_group_desc *desc;
425 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
427 if (flex_size > 1) {
428 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
429 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
430 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
431 return;
434 desc = ext4_get_group_desc(sb, g, NULL);
435 if (desc) {
436 stats->free_inodes = ext4_free_inodes_count(sb, desc);
437 stats->free_blocks = ext4_free_blks_count(sb, desc);
438 stats->used_dirs = ext4_used_dirs_count(sb, desc);
439 } else {
440 stats->free_inodes = 0;
441 stats->free_blocks = 0;
442 stats->used_dirs = 0;
447 * Orlov's allocator for directories.
449 * We always try to spread first-level directories.
451 * If there are blockgroups with both free inodes and free blocks counts
452 * not worse than average we return one with smallest directory count.
453 * Otherwise we simply return a random group.
455 * For the rest rules look so:
457 * It's OK to put directory into a group unless
458 * it has too many directories already (max_dirs) or
459 * it has too few free inodes left (min_inodes) or
460 * it has too few free blocks left (min_blocks) or
461 * Parent's group is preferred, if it doesn't satisfy these
462 * conditions we search cyclically through the rest. If none
463 * of the groups look good we just look for a group with more
464 * free inodes than average (starting at parent's group).
467 static int find_group_orlov(struct super_block *sb, struct inode *parent,
468 ext4_group_t *group, int mode,
469 const struct qstr *qstr)
471 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
472 struct ext4_sb_info *sbi = EXT4_SB(sb);
473 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
474 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
475 unsigned int freei, avefreei;
476 ext4_fsblk_t freeb, avefreeb;
477 unsigned int ndirs;
478 int max_dirs, min_inodes;
479 ext4_grpblk_t min_blocks;
480 ext4_group_t i, grp, g, ngroups;
481 struct ext4_group_desc *desc;
482 struct orlov_stats stats;
483 int flex_size = ext4_flex_bg_size(sbi);
484 struct dx_hash_info hinfo;
486 ngroups = real_ngroups;
487 if (flex_size > 1) {
488 ngroups = (real_ngroups + flex_size - 1) >>
489 sbi->s_log_groups_per_flex;
490 parent_group >>= sbi->s_log_groups_per_flex;
493 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
494 avefreei = freei / ngroups;
495 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
496 avefreeb = freeb;
497 do_div(avefreeb, ngroups);
498 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
500 if (S_ISDIR(mode) &&
501 ((parent == sb->s_root->d_inode) ||
502 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
503 int best_ndir = inodes_per_group;
504 int ret = -1;
506 if (qstr) {
507 hinfo.hash_version = DX_HASH_HALF_MD4;
508 hinfo.seed = sbi->s_hash_seed;
509 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
510 grp = hinfo.hash;
511 } else
512 get_random_bytes(&grp, sizeof(grp));
513 parent_group = (unsigned)grp % ngroups;
514 for (i = 0; i < ngroups; i++) {
515 g = (parent_group + i) % ngroups;
516 get_orlov_stats(sb, g, flex_size, &stats);
517 if (!stats.free_inodes)
518 continue;
519 if (stats.used_dirs >= best_ndir)
520 continue;
521 if (stats.free_inodes < avefreei)
522 continue;
523 if (stats.free_blocks < avefreeb)
524 continue;
525 grp = g;
526 ret = 0;
527 best_ndir = stats.used_dirs;
529 if (ret)
530 goto fallback;
531 found_flex_bg:
532 if (flex_size == 1) {
533 *group = grp;
534 return 0;
538 * We pack inodes at the beginning of the flexgroup's
539 * inode tables. Block allocation decisions will do
540 * something similar, although regular files will
541 * start at 2nd block group of the flexgroup. See
542 * ext4_ext_find_goal() and ext4_find_near().
544 grp *= flex_size;
545 for (i = 0; i < flex_size; i++) {
546 if (grp+i >= real_ngroups)
547 break;
548 desc = ext4_get_group_desc(sb, grp+i, NULL);
549 if (desc && ext4_free_inodes_count(sb, desc)) {
550 *group = grp+i;
551 return 0;
554 goto fallback;
557 max_dirs = ndirs / ngroups + inodes_per_group / 16;
558 min_inodes = avefreei - inodes_per_group*flex_size / 4;
559 if (min_inodes < 1)
560 min_inodes = 1;
561 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
564 * Start looking in the flex group where we last allocated an
565 * inode for this parent directory
567 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
568 parent_group = EXT4_I(parent)->i_last_alloc_group;
569 if (flex_size > 1)
570 parent_group >>= sbi->s_log_groups_per_flex;
573 for (i = 0; i < ngroups; i++) {
574 grp = (parent_group + i) % ngroups;
575 get_orlov_stats(sb, grp, flex_size, &stats);
576 if (stats.used_dirs >= max_dirs)
577 continue;
578 if (stats.free_inodes < min_inodes)
579 continue;
580 if (stats.free_blocks < min_blocks)
581 continue;
582 goto found_flex_bg;
585 fallback:
586 ngroups = real_ngroups;
587 avefreei = freei / ngroups;
588 fallback_retry:
589 parent_group = EXT4_I(parent)->i_block_group;
590 for (i = 0; i < ngroups; i++) {
591 grp = (parent_group + i) % ngroups;
592 desc = ext4_get_group_desc(sb, grp, NULL);
593 if (desc && ext4_free_inodes_count(sb, desc) &&
594 ext4_free_inodes_count(sb, desc) >= avefreei) {
595 *group = grp;
596 return 0;
600 if (avefreei) {
602 * The free-inodes counter is approximate, and for really small
603 * filesystems the above test can fail to find any blockgroups
605 avefreei = 0;
606 goto fallback_retry;
609 return -1;
612 static int find_group_other(struct super_block *sb, struct inode *parent,
613 ext4_group_t *group, int mode)
615 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
616 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
617 struct ext4_group_desc *desc;
618 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
621 * Try to place the inode is the same flex group as its
622 * parent. If we can't find space, use the Orlov algorithm to
623 * find another flex group, and store that information in the
624 * parent directory's inode information so that use that flex
625 * group for future allocations.
627 if (flex_size > 1) {
628 int retry = 0;
630 try_again:
631 parent_group &= ~(flex_size-1);
632 last = parent_group + flex_size;
633 if (last > ngroups)
634 last = ngroups;
635 for (i = parent_group; i < last; i++) {
636 desc = ext4_get_group_desc(sb, i, NULL);
637 if (desc && ext4_free_inodes_count(sb, desc)) {
638 *group = i;
639 return 0;
642 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
643 retry = 1;
644 parent_group = EXT4_I(parent)->i_last_alloc_group;
645 goto try_again;
648 * If this didn't work, use the Orlov search algorithm
649 * to find a new flex group; we pass in the mode to
650 * avoid the topdir algorithms.
652 *group = parent_group + flex_size;
653 if (*group > ngroups)
654 *group = 0;
655 return find_group_orlov(sb, parent, group, mode, 0);
659 * Try to place the inode in its parent directory
661 *group = parent_group;
662 desc = ext4_get_group_desc(sb, *group, NULL);
663 if (desc && ext4_free_inodes_count(sb, desc) &&
664 ext4_free_blks_count(sb, desc))
665 return 0;
668 * We're going to place this inode in a different blockgroup from its
669 * parent. We want to cause files in a common directory to all land in
670 * the same blockgroup. But we want files which are in a different
671 * directory which shares a blockgroup with our parent to land in a
672 * different blockgroup.
674 * So add our directory's i_ino into the starting point for the hash.
676 *group = (*group + parent->i_ino) % ngroups;
679 * Use a quadratic hash to find a group with a free inode and some free
680 * blocks.
682 for (i = 1; i < ngroups; i <<= 1) {
683 *group += i;
684 if (*group >= ngroups)
685 *group -= ngroups;
686 desc = ext4_get_group_desc(sb, *group, NULL);
687 if (desc && ext4_free_inodes_count(sb, desc) &&
688 ext4_free_blks_count(sb, desc))
689 return 0;
693 * That failed: try linear search for a free inode, even if that group
694 * has no free blocks.
696 *group = parent_group;
697 for (i = 0; i < ngroups; i++) {
698 if (++*group >= ngroups)
699 *group = 0;
700 desc = ext4_get_group_desc(sb, *group, NULL);
701 if (desc && ext4_free_inodes_count(sb, desc))
702 return 0;
705 return -1;
709 * claim the inode from the inode bitmap. If the group
710 * is uninit we need to take the groups's ext4_group_lock
711 * and clear the uninit flag. The inode bitmap update
712 * and group desc uninit flag clear should be done
713 * after holding ext4_group_lock so that ext4_read_inode_bitmap
714 * doesn't race with the ext4_claim_inode
716 static int ext4_claim_inode(struct super_block *sb,
717 struct buffer_head *inode_bitmap_bh,
718 unsigned long ino, ext4_group_t group, int mode)
720 int free = 0, retval = 0, count;
721 struct ext4_sb_info *sbi = EXT4_SB(sb);
722 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
724 ext4_lock_group(sb, group);
725 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
726 /* not a free inode */
727 retval = 1;
728 goto err_ret;
730 ino++;
731 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
732 ino > EXT4_INODES_PER_GROUP(sb)) {
733 ext4_unlock_group(sb, group);
734 ext4_error(sb, "reserved inode or inode > inodes count - "
735 "block_group = %u, inode=%lu", group,
736 ino + group * EXT4_INODES_PER_GROUP(sb));
737 return 1;
739 /* If we didn't allocate from within the initialized part of the inode
740 * table then we need to initialize up to this inode. */
741 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
743 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
744 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
745 /* When marking the block group with
746 * ~EXT4_BG_INODE_UNINIT we don't want to depend
747 * on the value of bg_itable_unused even though
748 * mke2fs could have initialized the same for us.
749 * Instead we calculated the value below
752 free = 0;
753 } else {
754 free = EXT4_INODES_PER_GROUP(sb) -
755 ext4_itable_unused_count(sb, gdp);
759 * Check the relative inode number against the last used
760 * relative inode number in this group. if it is greater
761 * we need to update the bg_itable_unused count
764 if (ino > free)
765 ext4_itable_unused_set(sb, gdp,
766 (EXT4_INODES_PER_GROUP(sb) - ino));
768 count = ext4_free_inodes_count(sb, gdp) - 1;
769 ext4_free_inodes_set(sb, gdp, count);
770 if (S_ISDIR(mode)) {
771 count = ext4_used_dirs_count(sb, gdp) + 1;
772 ext4_used_dirs_set(sb, gdp, count);
773 if (sbi->s_log_groups_per_flex) {
774 ext4_group_t f = ext4_flex_group(sbi, group);
776 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
779 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
780 err_ret:
781 ext4_unlock_group(sb, group);
782 return retval;
786 * There are two policies for allocating an inode. If the new inode is
787 * a directory, then a forward search is made for a block group with both
788 * free space and a low directory-to-inode ratio; if that fails, then of
789 * the groups with above-average free space, that group with the fewest
790 * directories already is chosen.
792 * For other inodes, search forward from the parent directory's block
793 * group to find a free inode.
795 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
796 const struct qstr *qstr, __u32 goal)
798 struct super_block *sb;
799 struct buffer_head *inode_bitmap_bh = NULL;
800 struct buffer_head *group_desc_bh;
801 ext4_group_t ngroups, group = 0;
802 unsigned long ino = 0;
803 struct inode *inode;
804 struct ext4_group_desc *gdp = NULL;
805 struct ext4_inode_info *ei;
806 struct ext4_sb_info *sbi;
807 int ret2, err = 0;
808 struct inode *ret;
809 ext4_group_t i;
810 int free = 0;
811 static int once = 1;
812 ext4_group_t flex_group;
814 /* Cannot create files in a deleted directory */
815 if (!dir || !dir->i_nlink)
816 return ERR_PTR(-EPERM);
818 sb = dir->i_sb;
819 ngroups = ext4_get_groups_count(sb);
820 trace_ext4_request_inode(dir, mode);
821 inode = new_inode(sb);
822 if (!inode)
823 return ERR_PTR(-ENOMEM);
824 ei = EXT4_I(inode);
825 sbi = EXT4_SB(sb);
827 if (!goal)
828 goal = sbi->s_inode_goal;
830 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
831 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
832 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
833 ret2 = 0;
834 goto got_group;
837 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
838 ret2 = find_group_flex(sb, dir, &group);
839 if (ret2 == -1) {
840 ret2 = find_group_other(sb, dir, &group, mode);
841 if (ret2 == 0 && once) {
842 once = 0;
843 printk(KERN_NOTICE "ext4: find_group_flex "
844 "failed, fallback succeeded dir %lu\n",
845 dir->i_ino);
848 goto got_group;
851 if (S_ISDIR(mode)) {
852 if (test_opt(sb, OLDALLOC))
853 ret2 = find_group_dir(sb, dir, &group);
854 else
855 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
856 } else
857 ret2 = find_group_other(sb, dir, &group, mode);
859 got_group:
860 EXT4_I(dir)->i_last_alloc_group = group;
861 err = -ENOSPC;
862 if (ret2 == -1)
863 goto out;
865 for (i = 0; i < ngroups; i++, ino = 0) {
866 err = -EIO;
868 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
869 if (!gdp)
870 goto fail;
872 brelse(inode_bitmap_bh);
873 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
874 if (!inode_bitmap_bh)
875 goto fail;
877 repeat_in_this_group:
878 ino = ext4_find_next_zero_bit((unsigned long *)
879 inode_bitmap_bh->b_data,
880 EXT4_INODES_PER_GROUP(sb), ino);
882 if (ino < EXT4_INODES_PER_GROUP(sb)) {
884 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
885 err = ext4_journal_get_write_access(handle,
886 inode_bitmap_bh);
887 if (err)
888 goto fail;
890 BUFFER_TRACE(group_desc_bh, "get_write_access");
891 err = ext4_journal_get_write_access(handle,
892 group_desc_bh);
893 if (err)
894 goto fail;
895 if (!ext4_claim_inode(sb, inode_bitmap_bh,
896 ino, group, mode)) {
897 /* we won it */
898 BUFFER_TRACE(inode_bitmap_bh,
899 "call ext4_handle_dirty_metadata");
900 err = ext4_handle_dirty_metadata(handle,
901 NULL,
902 inode_bitmap_bh);
903 if (err)
904 goto fail;
905 /* zero bit is inode number 1*/
906 ino++;
907 goto got;
909 /* we lost it */
910 ext4_handle_release_buffer(handle, inode_bitmap_bh);
911 ext4_handle_release_buffer(handle, group_desc_bh);
913 if (++ino < EXT4_INODES_PER_GROUP(sb))
914 goto repeat_in_this_group;
918 * This case is possible in concurrent environment. It is very
919 * rare. We cannot repeat the find_group_xxx() call because
920 * that will simply return the same blockgroup, because the
921 * group descriptor metadata has not yet been updated.
922 * So we just go onto the next blockgroup.
924 if (++group == ngroups)
925 group = 0;
927 err = -ENOSPC;
928 goto out;
930 got:
931 /* We may have to initialize the block bitmap if it isn't already */
932 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
933 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
934 struct buffer_head *block_bitmap_bh;
936 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
937 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
938 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
939 if (err) {
940 brelse(block_bitmap_bh);
941 goto fail;
944 free = 0;
945 ext4_lock_group(sb, group);
946 /* recheck and clear flag under lock if we still need to */
947 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
948 free = ext4_free_blocks_after_init(sb, group, gdp);
949 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
950 ext4_free_blks_set(sb, gdp, free);
951 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
952 gdp);
954 ext4_unlock_group(sb, group);
956 /* Don't need to dirty bitmap block if we didn't change it */
957 if (free) {
958 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
959 err = ext4_handle_dirty_metadata(handle,
960 NULL, block_bitmap_bh);
963 brelse(block_bitmap_bh);
964 if (err)
965 goto fail;
967 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
968 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
969 if (err)
970 goto fail;
972 percpu_counter_dec(&sbi->s_freeinodes_counter);
973 if (S_ISDIR(mode))
974 percpu_counter_inc(&sbi->s_dirs_counter);
975 sb->s_dirt = 1;
977 if (sbi->s_log_groups_per_flex) {
978 flex_group = ext4_flex_group(sbi, group);
979 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
982 inode->i_uid = current_fsuid();
983 if (test_opt(sb, GRPID))
984 inode->i_gid = dir->i_gid;
985 else if (dir->i_mode & S_ISGID) {
986 inode->i_gid = dir->i_gid;
987 if (S_ISDIR(mode))
988 mode |= S_ISGID;
989 } else
990 inode->i_gid = current_fsgid();
991 inode->i_mode = mode;
993 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
994 /* This is the optimal IO size (for stat), not the fs block size */
995 inode->i_blocks = 0;
996 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
997 ext4_current_time(inode);
999 memset(ei->i_data, 0, sizeof(ei->i_data));
1000 ei->i_dir_start_lookup = 0;
1001 ei->i_disksize = 0;
1004 * Don't inherit extent flag from directory, amongst others. We set
1005 * extent flag on newly created directory and file only if -o extent
1006 * mount option is specified
1008 ei->i_flags =
1009 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1010 ei->i_file_acl = 0;
1011 ei->i_dtime = 0;
1012 ei->i_block_group = group;
1013 ei->i_last_alloc_group = ~0;
1015 ext4_set_inode_flags(inode);
1016 if (IS_DIRSYNC(inode))
1017 ext4_handle_sync(handle);
1018 if (insert_inode_locked(inode) < 0) {
1019 err = -EINVAL;
1020 goto fail_drop;
1022 spin_lock(&sbi->s_next_gen_lock);
1023 inode->i_generation = sbi->s_next_generation++;
1024 spin_unlock(&sbi->s_next_gen_lock);
1026 ei->i_state_flags = 0;
1027 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1029 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1031 ret = inode;
1032 dquot_initialize(inode);
1033 err = dquot_alloc_inode(inode);
1034 if (err)
1035 goto fail_drop;
1037 err = ext4_init_acl(handle, inode, dir);
1038 if (err)
1039 goto fail_free_drop;
1041 err = ext4_init_security(handle, inode, dir);
1042 if (err)
1043 goto fail_free_drop;
1045 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1046 /* set extent flag only for directory, file and normal symlink*/
1047 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1048 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1049 ext4_ext_tree_init(handle, inode);
1053 err = ext4_mark_inode_dirty(handle, inode);
1054 if (err) {
1055 ext4_std_error(sb, err);
1056 goto fail_free_drop;
1059 ext4_debug("allocating inode %lu\n", inode->i_ino);
1060 trace_ext4_allocate_inode(inode, dir, mode);
1061 goto really_out;
1062 fail:
1063 ext4_std_error(sb, err);
1064 out:
1065 iput(inode);
1066 ret = ERR_PTR(err);
1067 really_out:
1068 brelse(inode_bitmap_bh);
1069 return ret;
1071 fail_free_drop:
1072 dquot_free_inode(inode);
1074 fail_drop:
1075 dquot_drop(inode);
1076 inode->i_flags |= S_NOQUOTA;
1077 inode->i_nlink = 0;
1078 unlock_new_inode(inode);
1079 iput(inode);
1080 brelse(inode_bitmap_bh);
1081 return ERR_PTR(err);
1084 /* Verify that we are loading a valid orphan from disk */
1085 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1087 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1088 ext4_group_t block_group;
1089 int bit;
1090 struct buffer_head *bitmap_bh;
1091 struct inode *inode = NULL;
1092 long err = -EIO;
1094 /* Error cases - e2fsck has already cleaned up for us */
1095 if (ino > max_ino) {
1096 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1097 goto error;
1100 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1101 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1102 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1103 if (!bitmap_bh) {
1104 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1105 goto error;
1108 /* Having the inode bit set should be a 100% indicator that this
1109 * is a valid orphan (no e2fsck run on fs). Orphans also include
1110 * inodes that were being truncated, so we can't check i_nlink==0.
1112 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1113 goto bad_orphan;
1115 inode = ext4_iget(sb, ino);
1116 if (IS_ERR(inode))
1117 goto iget_failed;
1120 * If the orphans has i_nlinks > 0 then it should be able to be
1121 * truncated, otherwise it won't be removed from the orphan list
1122 * during processing and an infinite loop will result.
1124 if (inode->i_nlink && !ext4_can_truncate(inode))
1125 goto bad_orphan;
1127 if (NEXT_ORPHAN(inode) > max_ino)
1128 goto bad_orphan;
1129 brelse(bitmap_bh);
1130 return inode;
1132 iget_failed:
1133 err = PTR_ERR(inode);
1134 inode = NULL;
1135 bad_orphan:
1136 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1137 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1138 bit, (unsigned long long)bitmap_bh->b_blocknr,
1139 ext4_test_bit(bit, bitmap_bh->b_data));
1140 printk(KERN_NOTICE "inode=%p\n", inode);
1141 if (inode) {
1142 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1143 is_bad_inode(inode));
1144 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1145 NEXT_ORPHAN(inode));
1146 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1147 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1148 /* Avoid freeing blocks if we got a bad deleted inode */
1149 if (inode->i_nlink == 0)
1150 inode->i_blocks = 0;
1151 iput(inode);
1153 brelse(bitmap_bh);
1154 error:
1155 return ERR_PTR(err);
1158 unsigned long ext4_count_free_inodes(struct super_block *sb)
1160 unsigned long desc_count;
1161 struct ext4_group_desc *gdp;
1162 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1163 #ifdef EXT4FS_DEBUG
1164 struct ext4_super_block *es;
1165 unsigned long bitmap_count, x;
1166 struct buffer_head *bitmap_bh = NULL;
1168 es = EXT4_SB(sb)->s_es;
1169 desc_count = 0;
1170 bitmap_count = 0;
1171 gdp = NULL;
1172 for (i = 0; i < ngroups; i++) {
1173 gdp = ext4_get_group_desc(sb, i, NULL);
1174 if (!gdp)
1175 continue;
1176 desc_count += ext4_free_inodes_count(sb, gdp);
1177 brelse(bitmap_bh);
1178 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1179 if (!bitmap_bh)
1180 continue;
1182 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1183 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1184 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1185 bitmap_count += x;
1187 brelse(bitmap_bh);
1188 printk(KERN_DEBUG "ext4_count_free_inodes: "
1189 "stored = %u, computed = %lu, %lu\n",
1190 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1191 return desc_count;
1192 #else
1193 desc_count = 0;
1194 for (i = 0; i < ngroups; i++) {
1195 gdp = ext4_get_group_desc(sb, i, NULL);
1196 if (!gdp)
1197 continue;
1198 desc_count += ext4_free_inodes_count(sb, gdp);
1199 cond_resched();
1201 return desc_count;
1202 #endif
1205 /* Called at mount-time, super-block is locked */
1206 unsigned long ext4_count_dirs(struct super_block * sb)
1208 unsigned long count = 0;
1209 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1211 for (i = 0; i < ngroups; i++) {
1212 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1213 if (!gdp)
1214 continue;
1215 count += ext4_used_dirs_count(sb, gdp);
1217 return count;