__tty_open(): use the correct type for saved_flags
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext4 / ialloc.c
blob47b84e8df56859ca1e741cfb6543b5966ee45657
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>
26 #include "ext4.h"
27 #include "ext4_jbd2.h"
28 #include "xattr.h"
29 #include "acl.h"
30 #include "group.h"
33 * ialloc.c contains the inodes allocation and deallocation routines
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.
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.
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.
51 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53 int i;
55 if (start_bit >= end_bit)
56 return;
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);
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)
70 struct ext4_sb_info *sbi = EXT4_SB(sb);
72 J_ASSERT_BH(bh, buffer_locked(bh));
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;
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);
90 return EXT4_INODES_PER_GROUP(sb);
94 * Read the inode allocation bitmap for a given block_group, reading
95 * into the specified slot in the superblock's bitmap cache.
97 * Return buffer_head of bitmap on success or NULL.
99 static struct buffer_head *
100 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
102 struct ext4_group_desc *desc;
103 struct buffer_head *bh = NULL;
104 ext4_fsblk_t bitmap_blk;
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;
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 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;
135 spin_unlock(sb_bgl_lock(EXT4_SB(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, __func__,
155 "Cannot read inode bitmap - "
156 "block_group = %u, inode_bitmap = %llu",
157 block_group, bitmap_blk);
158 return NULL;
160 return bh;
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.
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.
179 void ext4_free_inode(handle_t *handle, struct inode *inode)
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;
193 if (atomic_read(&inode->i_count) > 1) {
194 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
195 atomic_read(&inode->i_count));
196 return;
198 if (inode->i_nlink) {
199 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
200 inode->i_nlink);
201 return;
203 if (!sb) {
204 printk(KERN_ERR "ext4_free_inode: inode on "
205 "nonexistent device\n");
206 return;
208 sbi = EXT4_SB(sb);
210 ino = inode->i_ino;
211 ext4_debug("freeing inode %lu\n", ino);
212 trace_mark(ext4_free_inode,
213 "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu",
214 sb->s_id, inode->i_ino, inode->i_mode,
215 (unsigned long) inode->i_uid, (unsigned long) inode->i_gid,
216 (unsigned long long) inode->i_blocks);
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
222 vfs_dq_init(inode);
223 ext4_xattr_delete_inode(handle, inode);
224 vfs_dq_free_inode(inode);
225 vfs_dq_drop(inode);
227 is_directory = S_ISDIR(inode->i_mode);
229 /* Do this BEFORE marking the inode not in use or returning an error */
230 clear_inode(inode);
232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
234 ext4_error(sb, "ext4_free_inode",
235 "reserved or nonexistent inode %lu", ino);
236 goto error_return;
238 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
239 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
240 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
241 if (!bitmap_bh)
242 goto error_return;
244 BUFFER_TRACE(bitmap_bh, "get_write_access");
245 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
246 if (fatal)
247 goto error_return;
249 /* Ok, now we can actually update the inode bitmaps.. */
250 spin_lock(sb_bgl_lock(sbi, block_group));
251 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
252 spin_unlock(sb_bgl_lock(sbi, block_group));
253 if (!cleared)
254 ext4_error(sb, "ext4_free_inode",
255 "bit already cleared for inode %lu", ino);
256 else {
257 gdp = ext4_get_group_desc(sb, block_group, &bh2);
259 BUFFER_TRACE(bh2, "get_write_access");
260 fatal = ext4_journal_get_write_access(handle, bh2);
261 if (fatal) goto error_return;
263 if (gdp) {
264 spin_lock(sb_bgl_lock(sbi, block_group));
265 count = ext4_free_inodes_count(sb, gdp) + 1;
266 ext4_free_inodes_set(sb, gdp, count);
267 if (is_directory) {
268 count = ext4_used_dirs_count(sb, gdp) - 1;
269 ext4_used_dirs_set(sb, gdp, count);
270 if (sbi->s_log_groups_per_flex) {
271 ext4_group_t f;
273 f = ext4_flex_group(sbi, block_group);
274 atomic_dec(&sbi->s_flex_groups[f].free_inodes);
278 gdp->bg_checksum = ext4_group_desc_csum(sbi,
279 block_group, gdp);
280 spin_unlock(sb_bgl_lock(sbi, block_group));
281 percpu_counter_inc(&sbi->s_freeinodes_counter);
282 if (is_directory)
283 percpu_counter_dec(&sbi->s_dirs_counter);
285 if (sbi->s_log_groups_per_flex) {
286 ext4_group_t f;
288 f = ext4_flex_group(sbi, block_group);
289 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
292 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
293 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
294 if (!fatal) fatal = err;
296 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
297 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
298 if (!fatal)
299 fatal = err;
300 sb->s_dirt = 1;
301 error_return:
302 brelse(bitmap_bh);
303 ext4_std_error(sb, fatal);
307 * There are two policies for allocating an inode. If the new inode is
308 * a directory, then a forward search is made for a block group with both
309 * free space and a low directory-to-inode ratio; if that fails, then of
310 * the groups with above-average free space, that group with the fewest
311 * directories already is chosen.
313 * For other inodes, search forward from the parent directory\'s block
314 * group to find a free inode.
316 static int find_group_dir(struct super_block *sb, struct inode *parent,
317 ext4_group_t *best_group)
319 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
320 unsigned int freei, avefreei;
321 struct ext4_group_desc *desc, *best_desc = NULL;
322 ext4_group_t group;
323 int ret = -1;
325 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
326 avefreei = freei / ngroups;
328 for (group = 0; group < ngroups; group++) {
329 desc = ext4_get_group_desc(sb, group, NULL);
330 if (!desc || !ext4_free_inodes_count(sb, desc))
331 continue;
332 if (ext4_free_inodes_count(sb, desc) < avefreei)
333 continue;
334 if (!best_desc ||
335 (ext4_free_blks_count(sb, desc) >
336 ext4_free_blks_count(sb, best_desc))) {
337 *best_group = group;
338 best_desc = desc;
339 ret = 0;
342 return ret;
345 #define free_block_ratio 10
347 static int find_group_flex(struct super_block *sb, struct inode *parent,
348 ext4_group_t *best_group)
350 struct ext4_sb_info *sbi = EXT4_SB(sb);
351 struct ext4_group_desc *desc;
352 struct buffer_head *bh;
353 struct flex_groups *flex_group = sbi->s_flex_groups;
354 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
355 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
356 ext4_group_t ngroups = sbi->s_groups_count;
357 int flex_size = ext4_flex_bg_size(sbi);
358 ext4_group_t best_flex = parent_fbg_group;
359 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
360 int flexbg_free_blocks;
361 int flex_freeb_ratio;
362 ext4_group_t n_fbg_groups;
363 ext4_group_t i;
365 n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >>
366 sbi->s_log_groups_per_flex;
368 find_close_to_parent:
369 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
370 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
371 if (atomic_read(&flex_group[best_flex].free_inodes) &&
372 flex_freeb_ratio > free_block_ratio)
373 goto found_flexbg;
375 if (best_flex && best_flex == parent_fbg_group) {
376 best_flex--;
377 goto find_close_to_parent;
380 for (i = 0; i < n_fbg_groups; i++) {
381 if (i == parent_fbg_group || i == parent_fbg_group - 1)
382 continue;
384 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
385 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
387 if (flex_freeb_ratio > free_block_ratio &&
388 (atomic_read(&flex_group[i].free_inodes))) {
389 best_flex = i;
390 goto found_flexbg;
393 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
394 ((atomic_read(&flex_group[i].free_blocks) >
395 atomic_read(&flex_group[best_flex].free_blocks)) &&
396 atomic_read(&flex_group[i].free_inodes)))
397 best_flex = i;
400 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
401 !atomic_read(&flex_group[best_flex].free_blocks))
402 return -1;
404 found_flexbg:
405 for (i = best_flex * flex_size; i < ngroups &&
406 i < (best_flex + 1) * flex_size; i++) {
407 desc = ext4_get_group_desc(sb, i, &bh);
408 if (ext4_free_inodes_count(sb, desc)) {
409 *best_group = i;
410 goto out;
414 return -1;
415 out:
416 return 0;
419 struct orlov_stats {
420 __u32 free_inodes;
421 __u32 free_blocks;
422 __u32 used_dirs;
426 * Helper function for Orlov's allocator; returns critical information
427 * for a particular block group or flex_bg. If flex_size is 1, then g
428 * is a block group number; otherwise it is flex_bg number.
430 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
431 int flex_size, struct orlov_stats *stats)
433 struct ext4_group_desc *desc;
434 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
436 if (flex_size > 1) {
437 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
438 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
439 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
440 return;
443 desc = ext4_get_group_desc(sb, g, NULL);
444 if (desc) {
445 stats->free_inodes = ext4_free_inodes_count(sb, desc);
446 stats->free_blocks = ext4_free_blks_count(sb, desc);
447 stats->used_dirs = ext4_used_dirs_count(sb, desc);
448 } else {
449 stats->free_inodes = 0;
450 stats->free_blocks = 0;
451 stats->used_dirs = 0;
456 * Orlov's allocator for directories.
458 * We always try to spread first-level directories.
460 * If there are blockgroups with both free inodes and free blocks counts
461 * not worse than average we return one with smallest directory count.
462 * Otherwise we simply return a random group.
464 * For the rest rules look so:
466 * It's OK to put directory into a group unless
467 * it has too many directories already (max_dirs) or
468 * it has too few free inodes left (min_inodes) or
469 * it has too few free blocks left (min_blocks) or
470 * Parent's group is preferred, if it doesn't satisfy these
471 * conditions we search cyclically through the rest. If none
472 * of the groups look good we just look for a group with more
473 * free inodes than average (starting at parent's group).
476 static int find_group_orlov(struct super_block *sb, struct inode *parent,
477 ext4_group_t *group, int mode)
479 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
480 struct ext4_sb_info *sbi = EXT4_SB(sb);
481 ext4_group_t ngroups = sbi->s_groups_count;
482 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
483 unsigned int freei, avefreei;
484 ext4_fsblk_t freeb, avefreeb;
485 unsigned int ndirs;
486 int max_dirs, min_inodes;
487 ext4_grpblk_t min_blocks;
488 ext4_group_t i, grp, g;
489 struct ext4_group_desc *desc;
490 struct orlov_stats stats;
491 int flex_size = ext4_flex_bg_size(sbi);
493 if (flex_size > 1) {
494 ngroups = (ngroups + flex_size - 1) >>
495 sbi->s_log_groups_per_flex;
496 parent_group >>= sbi->s_log_groups_per_flex;
499 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
500 avefreei = freei / ngroups;
501 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
502 avefreeb = freeb;
503 do_div(avefreeb, ngroups);
504 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
506 if (S_ISDIR(mode) &&
507 ((parent == sb->s_root->d_inode) ||
508 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
509 int best_ndir = inodes_per_group;
510 int ret = -1;
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 >= sbi->s_groups_count)
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 = sbi->s_groups_count;
587 avefreei = freei / ngroups;
588 parent_group = EXT4_I(parent)->i_block_group;
589 for (i = 0; i < ngroups; i++) {
590 grp = (parent_group + i) % ngroups;
591 desc = ext4_get_group_desc(sb, grp, NULL);
592 if (desc && ext4_free_inodes_count(sb, desc) &&
593 ext4_free_inodes_count(sb, desc) >= avefreei) {
594 *group = grp;
595 return 0;
599 if (avefreei) {
601 * The free-inodes counter is approximate, and for really small
602 * filesystems the above test can fail to find any blockgroups
604 avefreei = 0;
605 goto fallback;
608 return -1;
611 static int find_group_other(struct super_block *sb, struct inode *parent,
612 ext4_group_t *group, int mode)
614 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
615 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
616 struct ext4_group_desc *desc;
617 ext4_group_t i, last;
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);
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 sb_bgl_lock
711 * and clear the uninit flag. The inode bitmap update
712 * and group desc uninit flag clear should be done
713 * after holding sb_bgl_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 spin_lock(sb_bgl_lock(sbi, 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 spin_unlock(sb_bgl_lock(sbi, group));
734 ext4_error(sb, __func__,
735 "reserved inode or inode > inodes count - "
736 "block_group = %u, inode=%lu", group,
737 ino + group * EXT4_INODES_PER_GROUP(sb));
738 return 1;
740 /* If we didn't allocate from within the initialized part of the inode
741 * table then we need to initialize up to this inode. */
742 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
744 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
745 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
746 /* When marking the block group with
747 * ~EXT4_BG_INODE_UNINIT we don't want to depend
748 * on the value of bg_itable_unused even though
749 * mke2fs could have initialized the same for us.
750 * Instead we calculated the value below
753 free = 0;
754 } else {
755 free = EXT4_INODES_PER_GROUP(sb) -
756 ext4_itable_unused_count(sb, gdp);
760 * Check the relative inode number against the last used
761 * relative inode number in this group. if it is greater
762 * we need to update the bg_itable_unused count
765 if (ino > free)
766 ext4_itable_unused_set(sb, gdp,
767 (EXT4_INODES_PER_GROUP(sb) - ino));
769 count = ext4_free_inodes_count(sb, gdp) - 1;
770 ext4_free_inodes_set(sb, gdp, count);
771 if (S_ISDIR(mode)) {
772 count = ext4_used_dirs_count(sb, gdp) + 1;
773 ext4_used_dirs_set(sb, gdp, count);
774 if (sbi->s_log_groups_per_flex) {
775 ext4_group_t f = ext4_flex_group(sbi, group);
777 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
780 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
781 err_ret:
782 spin_unlock(sb_bgl_lock(sbi, group));
783 return retval;
787 * There are two policies for allocating an inode. If the new inode is
788 * a directory, then a forward search is made for a block group with both
789 * free space and a low directory-to-inode ratio; if that fails, then of
790 * the groups with above-average free space, that group with the fewest
791 * directories already is chosen.
793 * For other inodes, search forward from the parent directory's block
794 * group to find a free inode.
796 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
798 struct super_block *sb;
799 struct buffer_head *inode_bitmap_bh = NULL;
800 struct buffer_head *group_desc_bh;
801 ext4_group_t group = 0;
802 unsigned long ino = 0;
803 struct inode *inode;
804 struct ext4_group_desc *gdp = NULL;
805 struct ext4_super_block *es;
806 struct ext4_inode_info *ei;
807 struct ext4_sb_info *sbi;
808 int ret2, err = 0;
809 struct inode *ret;
810 ext4_group_t i;
811 int free = 0;
812 static int once = 1;
813 ext4_group_t flex_group;
815 /* Cannot create files in a deleted directory */
816 if (!dir || !dir->i_nlink)
817 return ERR_PTR(-EPERM);
819 sb = dir->i_sb;
820 trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id,
821 dir->i_ino, mode);
822 inode = new_inode(sb);
823 if (!inode)
824 return ERR_PTR(-ENOMEM);
825 ei = EXT4_I(inode);
827 sbi = EXT4_SB(sb);
828 es = sbi->s_es;
830 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
831 ret2 = find_group_flex(sb, dir, &group);
832 if (ret2 == -1) {
833 ret2 = find_group_other(sb, dir, &group, mode);
834 if (ret2 == 0 && once)
835 once = 0;
836 printk(KERN_NOTICE "ext4: find_group_flex "
837 "failed, fallback succeeded dir %lu\n",
838 dir->i_ino);
840 goto got_group;
843 if (S_ISDIR(mode)) {
844 if (test_opt(sb, OLDALLOC))
845 ret2 = find_group_dir(sb, dir, &group);
846 else
847 ret2 = find_group_orlov(sb, dir, &group, mode);
848 } else
849 ret2 = find_group_other(sb, dir, &group, mode);
851 got_group:
852 EXT4_I(dir)->i_last_alloc_group = group;
853 err = -ENOSPC;
854 if (ret2 == -1)
855 goto out;
857 for (i = 0; i < sbi->s_groups_count; i++) {
858 err = -EIO;
860 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
861 if (!gdp)
862 goto fail;
864 brelse(inode_bitmap_bh);
865 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
866 if (!inode_bitmap_bh)
867 goto fail;
869 ino = 0;
871 repeat_in_this_group:
872 ino = ext4_find_next_zero_bit((unsigned long *)
873 inode_bitmap_bh->b_data,
874 EXT4_INODES_PER_GROUP(sb), ino);
876 if (ino < EXT4_INODES_PER_GROUP(sb)) {
878 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
879 err = ext4_journal_get_write_access(handle,
880 inode_bitmap_bh);
881 if (err)
882 goto fail;
884 BUFFER_TRACE(group_desc_bh, "get_write_access");
885 err = ext4_journal_get_write_access(handle,
886 group_desc_bh);
887 if (err)
888 goto fail;
889 if (!ext4_claim_inode(sb, inode_bitmap_bh,
890 ino, group, mode)) {
891 /* we won it */
892 BUFFER_TRACE(inode_bitmap_bh,
893 "call ext4_handle_dirty_metadata");
894 err = ext4_handle_dirty_metadata(handle,
895 inode,
896 inode_bitmap_bh);
897 if (err)
898 goto fail;
899 /* zero bit is inode number 1*/
900 ino++;
901 goto got;
903 /* we lost it */
904 ext4_handle_release_buffer(handle, inode_bitmap_bh);
905 ext4_handle_release_buffer(handle, group_desc_bh);
907 if (++ino < EXT4_INODES_PER_GROUP(sb))
908 goto repeat_in_this_group;
912 * This case is possible in concurrent environment. It is very
913 * rare. We cannot repeat the find_group_xxx() call because
914 * that will simply return the same blockgroup, because the
915 * group descriptor metadata has not yet been updated.
916 * So we just go onto the next blockgroup.
918 if (++group == sbi->s_groups_count)
919 group = 0;
921 err = -ENOSPC;
922 goto out;
924 got:
925 /* We may have to initialize the block bitmap if it isn't already */
926 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
927 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
928 struct buffer_head *block_bitmap_bh;
930 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
931 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
932 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
933 if (err) {
934 brelse(block_bitmap_bh);
935 goto fail;
938 free = 0;
939 spin_lock(sb_bgl_lock(sbi, group));
940 /* recheck and clear flag under lock if we still need to */
941 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
942 free = ext4_free_blocks_after_init(sb, group, gdp);
943 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
944 ext4_free_blks_set(sb, gdp, free);
945 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
946 gdp);
948 spin_unlock(sb_bgl_lock(sbi, group));
950 /* Don't need to dirty bitmap block if we didn't change it */
951 if (free) {
952 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
953 err = ext4_handle_dirty_metadata(handle,
954 NULL, block_bitmap_bh);
957 brelse(block_bitmap_bh);
958 if (err)
959 goto fail;
961 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
962 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
963 if (err)
964 goto fail;
966 percpu_counter_dec(&sbi->s_freeinodes_counter);
967 if (S_ISDIR(mode))
968 percpu_counter_inc(&sbi->s_dirs_counter);
969 sb->s_dirt = 1;
971 if (sbi->s_log_groups_per_flex) {
972 flex_group = ext4_flex_group(sbi, group);
973 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
976 inode->i_uid = current_fsuid();
977 if (test_opt(sb, GRPID))
978 inode->i_gid = dir->i_gid;
979 else if (dir->i_mode & S_ISGID) {
980 inode->i_gid = dir->i_gid;
981 if (S_ISDIR(mode))
982 mode |= S_ISGID;
983 } else
984 inode->i_gid = current_fsgid();
985 inode->i_mode = mode;
987 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
988 /* This is the optimal IO size (for stat), not the fs block size */
989 inode->i_blocks = 0;
990 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
991 ext4_current_time(inode);
993 memset(ei->i_data, 0, sizeof(ei->i_data));
994 ei->i_dir_start_lookup = 0;
995 ei->i_disksize = 0;
998 * Don't inherit extent flag from directory, amongst others. We set
999 * extent flag on newly created directory and file only if -o extent
1000 * mount option is specified
1002 ei->i_flags =
1003 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1004 ei->i_file_acl = 0;
1005 ei->i_dtime = 0;
1006 ei->i_block_group = group;
1007 ei->i_last_alloc_group = ~0;
1009 ext4_set_inode_flags(inode);
1010 if (IS_DIRSYNC(inode))
1011 ext4_handle_sync(handle);
1012 if (insert_inode_locked(inode) < 0) {
1013 err = -EINVAL;
1014 goto fail_drop;
1016 spin_lock(&sbi->s_next_gen_lock);
1017 inode->i_generation = sbi->s_next_generation++;
1018 spin_unlock(&sbi->s_next_gen_lock);
1020 ei->i_state = EXT4_STATE_NEW;
1022 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1024 ret = inode;
1025 if (vfs_dq_alloc_inode(inode)) {
1026 err = -EDQUOT;
1027 goto fail_drop;
1030 err = ext4_init_acl(handle, inode, dir);
1031 if (err)
1032 goto fail_free_drop;
1034 err = ext4_init_security(handle, inode, dir);
1035 if (err)
1036 goto fail_free_drop;
1038 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1039 /* set extent flag only for directory, file and normal symlink*/
1040 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1041 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1042 ext4_ext_tree_init(handle, inode);
1046 err = ext4_mark_inode_dirty(handle, inode);
1047 if (err) {
1048 ext4_std_error(sb, err);
1049 goto fail_free_drop;
1052 ext4_debug("allocating inode %lu\n", inode->i_ino);
1053 trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d",
1054 sb->s_id, inode->i_ino, dir->i_ino, mode);
1055 goto really_out;
1056 fail:
1057 ext4_std_error(sb, err);
1058 out:
1059 iput(inode);
1060 ret = ERR_PTR(err);
1061 really_out:
1062 brelse(inode_bitmap_bh);
1063 return ret;
1065 fail_free_drop:
1066 vfs_dq_free_inode(inode);
1068 fail_drop:
1069 vfs_dq_drop(inode);
1070 inode->i_flags |= S_NOQUOTA;
1071 inode->i_nlink = 0;
1072 unlock_new_inode(inode);
1073 iput(inode);
1074 brelse(inode_bitmap_bh);
1075 return ERR_PTR(err);
1078 /* Verify that we are loading a valid orphan from disk */
1079 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1081 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1082 ext4_group_t block_group;
1083 int bit;
1084 struct buffer_head *bitmap_bh;
1085 struct inode *inode = NULL;
1086 long err = -EIO;
1088 /* Error cases - e2fsck has already cleaned up for us */
1089 if (ino > max_ino) {
1090 ext4_warning(sb, __func__,
1091 "bad orphan ino %lu! e2fsck was run?", ino);
1092 goto error;
1095 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1096 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1097 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1098 if (!bitmap_bh) {
1099 ext4_warning(sb, __func__,
1100 "inode bitmap error for orphan %lu", ino);
1101 goto error;
1104 /* Having the inode bit set should be a 100% indicator that this
1105 * is a valid orphan (no e2fsck run on fs). Orphans also include
1106 * inodes that were being truncated, so we can't check i_nlink==0.
1108 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1109 goto bad_orphan;
1111 inode = ext4_iget(sb, ino);
1112 if (IS_ERR(inode))
1113 goto iget_failed;
1116 * If the orphans has i_nlinks > 0 then it should be able to be
1117 * truncated, otherwise it won't be removed from the orphan list
1118 * during processing and an infinite loop will result.
1120 if (inode->i_nlink && !ext4_can_truncate(inode))
1121 goto bad_orphan;
1123 if (NEXT_ORPHAN(inode) > max_ino)
1124 goto bad_orphan;
1125 brelse(bitmap_bh);
1126 return inode;
1128 iget_failed:
1129 err = PTR_ERR(inode);
1130 inode = NULL;
1131 bad_orphan:
1132 ext4_warning(sb, __func__,
1133 "bad orphan inode %lu! e2fsck was run?", ino);
1134 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1135 bit, (unsigned long long)bitmap_bh->b_blocknr,
1136 ext4_test_bit(bit, bitmap_bh->b_data));
1137 printk(KERN_NOTICE "inode=%p\n", inode);
1138 if (inode) {
1139 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1140 is_bad_inode(inode));
1141 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1142 NEXT_ORPHAN(inode));
1143 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1144 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1145 /* Avoid freeing blocks if we got a bad deleted inode */
1146 if (inode->i_nlink == 0)
1147 inode->i_blocks = 0;
1148 iput(inode);
1150 brelse(bitmap_bh);
1151 error:
1152 return ERR_PTR(err);
1155 unsigned long ext4_count_free_inodes(struct super_block *sb)
1157 unsigned long desc_count;
1158 struct ext4_group_desc *gdp;
1159 ext4_group_t i;
1160 #ifdef EXT4FS_DEBUG
1161 struct ext4_super_block *es;
1162 unsigned long bitmap_count, x;
1163 struct buffer_head *bitmap_bh = NULL;
1165 es = EXT4_SB(sb)->s_es;
1166 desc_count = 0;
1167 bitmap_count = 0;
1168 gdp = NULL;
1169 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1170 gdp = ext4_get_group_desc(sb, i, NULL);
1171 if (!gdp)
1172 continue;
1173 desc_count += ext4_free_inodes_count(sb, gdp);
1174 brelse(bitmap_bh);
1175 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1176 if (!bitmap_bh)
1177 continue;
1179 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1180 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1181 i, ext4_free_inodes_count(sb, gdp), x);
1182 bitmap_count += x;
1184 brelse(bitmap_bh);
1185 printk(KERN_DEBUG "ext4_count_free_inodes: "
1186 "stored = %u, computed = %lu, %lu\n",
1187 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1188 return desc_count;
1189 #else
1190 desc_count = 0;
1191 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1192 gdp = ext4_get_group_desc(sb, i, NULL);
1193 if (!gdp)
1194 continue;
1195 desc_count += ext4_free_inodes_count(sb, gdp);
1196 cond_resched();
1198 return desc_count;
1199 #endif
1202 /* Called at mount-time, super-block is locked */
1203 unsigned long ext4_count_dirs(struct super_block * sb)
1205 unsigned long count = 0;
1206 ext4_group_t i;
1208 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1209 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1210 if (!gdp)
1211 continue;
1212 count += ext4_used_dirs_count(sb, gdp);
1214 return count;