libata: fix NULL sdev dereference race in atapi_qc_complete()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext4 / ialloc.c
blob55a93f5bb0031a9c3836601fbe51374033310bd5
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, __func__, "Checksum bad for group %u",
80 block_group);
81 ext4_free_blks_set(sb, gdp, 0);
82 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
84 memset(bh->b_data, 0xff, sb->s_blocksize);
85 return 0;
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
89 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
90 bh->b_data);
92 return EXT4_INODES_PER_GROUP(sb);
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
99 * Return buffer_head of bitmap on success or NULL.
101 static struct buffer_head *
102 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
104 struct ext4_group_desc *desc;
105 struct buffer_head *bh = NULL;
106 ext4_fsblk_t bitmap_blk;
108 desc = ext4_get_group_desc(sb, block_group, NULL);
109 if (!desc)
110 return NULL;
111 bitmap_blk = ext4_inode_bitmap(sb, desc);
112 bh = sb_getblk(sb, bitmap_blk);
113 if (unlikely(!bh)) {
114 ext4_error(sb, __func__,
115 "Cannot read inode bitmap - "
116 "block_group = %u, inode_bitmap = %llu",
117 block_group, bitmap_blk);
118 return NULL;
120 if (bitmap_uptodate(bh))
121 return bh;
123 lock_buffer(bh);
124 if (bitmap_uptodate(bh)) {
125 unlock_buffer(bh);
126 return bh;
128 ext4_lock_group(sb, block_group);
129 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
130 ext4_init_inode_bitmap(sb, bh, block_group, desc);
131 set_bitmap_uptodate(bh);
132 set_buffer_uptodate(bh);
133 ext4_unlock_group(sb, block_group);
134 unlock_buffer(bh);
135 return bh;
137 ext4_unlock_group(sb, block_group);
138 if (buffer_uptodate(bh)) {
140 * if not uninit if bh is uptodate,
141 * bitmap is also uptodate
143 set_bitmap_uptodate(bh);
144 unlock_buffer(bh);
145 return bh;
148 * submit the buffer_head for read. We can
149 * safely mark the bitmap as uptodate now.
150 * We do it here so the bitmap uptodate bit
151 * get set with buffer lock held.
153 set_bitmap_uptodate(bh);
154 if (bh_submit_read(bh) < 0) {
155 put_bh(bh);
156 ext4_error(sb, __func__,
157 "Cannot read inode bitmap - "
158 "block_group = %u, inode_bitmap = %llu",
159 block_group, bitmap_blk);
160 return NULL;
162 return bh;
166 * NOTE! When we get the inode, we're the only people
167 * that have access to it, and as such there are no
168 * race conditions we have to worry about. The inode
169 * is not on the hash-lists, and it cannot be reached
170 * through the filesystem because the directory entry
171 * has been deleted earlier.
173 * HOWEVER: we must make sure that we get no aliases,
174 * which means that we have to call "clear_inode()"
175 * _before_ we mark the inode not in use in the inode
176 * bitmaps. Otherwise a newly created file might use
177 * the same inode number (not actually the same pointer
178 * though), and then we'd have two inodes sharing the
179 * same inode number and space on the harddisk.
181 void ext4_free_inode(handle_t *handle, struct inode *inode)
183 struct super_block *sb = inode->i_sb;
184 int is_directory;
185 unsigned long ino;
186 struct buffer_head *bitmap_bh = NULL;
187 struct buffer_head *bh2;
188 ext4_group_t block_group;
189 unsigned long bit;
190 struct ext4_group_desc *gdp;
191 struct ext4_super_block *es;
192 struct ext4_sb_info *sbi;
193 int fatal = 0, err, count, cleared;
195 if (atomic_read(&inode->i_count) > 1) {
196 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
197 atomic_read(&inode->i_count));
198 return;
200 if (inode->i_nlink) {
201 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
202 inode->i_nlink);
203 return;
205 if (!sb) {
206 printk(KERN_ERR "ext4_free_inode: inode on "
207 "nonexistent device\n");
208 return;
210 sbi = EXT4_SB(sb);
212 ino = inode->i_ino;
213 ext4_debug("freeing inode %lu\n", ino);
214 trace_ext4_free_inode(inode);
217 * Note: we must free any quota before locking the superblock,
218 * as writing the quota to disk may need the lock as well.
220 vfs_dq_init(inode);
221 ext4_xattr_delete_inode(handle, inode);
222 vfs_dq_free_inode(inode);
223 vfs_dq_drop(inode);
225 is_directory = S_ISDIR(inode->i_mode);
227 /* Do this BEFORE marking the inode not in use or returning an error */
228 clear_inode(inode);
230 es = EXT4_SB(sb)->s_es;
231 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
232 ext4_error(sb, "ext4_free_inode",
233 "reserved or nonexistent inode %lu", ino);
234 goto error_return;
236 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
237 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
238 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
239 if (!bitmap_bh)
240 goto error_return;
242 BUFFER_TRACE(bitmap_bh, "get_write_access");
243 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
244 if (fatal)
245 goto error_return;
247 fatal = -ESRCH;
248 gdp = ext4_get_group_desc(sb, block_group, &bh2);
249 if (gdp) {
250 BUFFER_TRACE(bh2, "get_write_access");
251 fatal = ext4_journal_get_write_access(handle, bh2);
253 ext4_lock_group(sb, block_group);
254 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
255 if (fatal || !cleared) {
256 ext4_unlock_group(sb, block_group);
257 goto out;
260 count = ext4_free_inodes_count(sb, gdp) + 1;
261 ext4_free_inodes_set(sb, gdp, count);
262 if (is_directory) {
263 count = ext4_used_dirs_count(sb, gdp) - 1;
264 ext4_used_dirs_set(sb, gdp, count);
265 percpu_counter_dec(&sbi->s_dirs_counter);
267 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
268 ext4_unlock_group(sb, block_group);
270 percpu_counter_inc(&sbi->s_freeinodes_counter);
271 if (sbi->s_log_groups_per_flex) {
272 ext4_group_t f = ext4_flex_group(sbi, block_group);
274 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
275 if (is_directory)
276 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
278 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
279 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
280 out:
281 if (cleared) {
282 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
283 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
284 if (!fatal)
285 fatal = err;
286 sb->s_dirt = 1;
287 } else
288 ext4_error(sb, "ext4_free_inode",
289 "bit already cleared for inode %lu", ino);
291 error_return:
292 brelse(bitmap_bh);
293 ext4_std_error(sb, fatal);
297 * There are two policies for allocating an inode. If the new inode is
298 * a directory, then a forward search is made for a block group with both
299 * free space and a low directory-to-inode ratio; if that fails, then of
300 * the groups with above-average free space, that group with the fewest
301 * directories already is chosen.
303 * For other inodes, search forward from the parent directory\'s block
304 * group to find a free inode.
306 static int find_group_dir(struct super_block *sb, struct inode *parent,
307 ext4_group_t *best_group)
309 ext4_group_t ngroups = ext4_get_groups_count(sb);
310 unsigned int freei, avefreei;
311 struct ext4_group_desc *desc, *best_desc = NULL;
312 ext4_group_t group;
313 int ret = -1;
315 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
316 avefreei = freei / ngroups;
318 for (group = 0; group < ngroups; group++) {
319 desc = ext4_get_group_desc(sb, group, NULL);
320 if (!desc || !ext4_free_inodes_count(sb, desc))
321 continue;
322 if (ext4_free_inodes_count(sb, desc) < avefreei)
323 continue;
324 if (!best_desc ||
325 (ext4_free_blks_count(sb, desc) >
326 ext4_free_blks_count(sb, best_desc))) {
327 *best_group = group;
328 best_desc = desc;
329 ret = 0;
332 return ret;
335 #define free_block_ratio 10
337 static int find_group_flex(struct super_block *sb, struct inode *parent,
338 ext4_group_t *best_group)
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
341 struct ext4_group_desc *desc;
342 struct flex_groups *flex_group = sbi->s_flex_groups;
343 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
344 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
345 ext4_group_t ngroups = ext4_get_groups_count(sb);
346 int flex_size = ext4_flex_bg_size(sbi);
347 ext4_group_t best_flex = parent_fbg_group;
348 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
349 int flexbg_free_blocks;
350 int flex_freeb_ratio;
351 ext4_group_t n_fbg_groups;
352 ext4_group_t i;
354 n_fbg_groups = (ngroups + flex_size - 1) >>
355 sbi->s_log_groups_per_flex;
357 find_close_to_parent:
358 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
359 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
360 if (atomic_read(&flex_group[best_flex].free_inodes) &&
361 flex_freeb_ratio > free_block_ratio)
362 goto found_flexbg;
364 if (best_flex && best_flex == parent_fbg_group) {
365 best_flex--;
366 goto find_close_to_parent;
369 for (i = 0; i < n_fbg_groups; i++) {
370 if (i == parent_fbg_group || i == parent_fbg_group - 1)
371 continue;
373 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
374 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
376 if (flex_freeb_ratio > free_block_ratio &&
377 (atomic_read(&flex_group[i].free_inodes))) {
378 best_flex = i;
379 goto found_flexbg;
382 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
383 ((atomic_read(&flex_group[i].free_blocks) >
384 atomic_read(&flex_group[best_flex].free_blocks)) &&
385 atomic_read(&flex_group[i].free_inodes)))
386 best_flex = i;
389 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
390 !atomic_read(&flex_group[best_flex].free_blocks))
391 return -1;
393 found_flexbg:
394 for (i = best_flex * flex_size; i < ngroups &&
395 i < (best_flex + 1) * flex_size; i++) {
396 desc = ext4_get_group_desc(sb, i, NULL);
397 if (ext4_free_inodes_count(sb, desc)) {
398 *best_group = i;
399 goto out;
403 return -1;
404 out:
405 return 0;
408 struct orlov_stats {
409 __u32 free_inodes;
410 __u32 free_blocks;
411 __u32 used_dirs;
415 * Helper function for Orlov's allocator; returns critical information
416 * for a particular block group or flex_bg. If flex_size is 1, then g
417 * is a block group number; otherwise it is flex_bg number.
419 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
420 int flex_size, struct orlov_stats *stats)
422 struct ext4_group_desc *desc;
423 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
425 if (flex_size > 1) {
426 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
427 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
428 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
429 return;
432 desc = ext4_get_group_desc(sb, g, NULL);
433 if (desc) {
434 stats->free_inodes = ext4_free_inodes_count(sb, desc);
435 stats->free_blocks = ext4_free_blks_count(sb, desc);
436 stats->used_dirs = ext4_used_dirs_count(sb, desc);
437 } else {
438 stats->free_inodes = 0;
439 stats->free_blocks = 0;
440 stats->used_dirs = 0;
445 * Orlov's allocator for directories.
447 * We always try to spread first-level directories.
449 * If there are blockgroups with both free inodes and free blocks counts
450 * not worse than average we return one with smallest directory count.
451 * Otherwise we simply return a random group.
453 * For the rest rules look so:
455 * It's OK to put directory into a group unless
456 * it has too many directories already (max_dirs) or
457 * it has too few free inodes left (min_inodes) or
458 * it has too few free blocks left (min_blocks) or
459 * Parent's group is preferred, if it doesn't satisfy these
460 * conditions we search cyclically through the rest. If none
461 * of the groups look good we just look for a group with more
462 * free inodes than average (starting at parent's group).
465 static int find_group_orlov(struct super_block *sb, struct inode *parent,
466 ext4_group_t *group, int mode,
467 const struct qstr *qstr)
469 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
470 struct ext4_sb_info *sbi = EXT4_SB(sb);
471 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
472 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
473 unsigned int freei, avefreei;
474 ext4_fsblk_t freeb, avefreeb;
475 unsigned int ndirs;
476 int max_dirs, min_inodes;
477 ext4_grpblk_t min_blocks;
478 ext4_group_t i, grp, g, ngroups;
479 struct ext4_group_desc *desc;
480 struct orlov_stats stats;
481 int flex_size = ext4_flex_bg_size(sbi);
482 struct dx_hash_info hinfo;
484 ngroups = real_ngroups;
485 if (flex_size > 1) {
486 ngroups = (real_ngroups + flex_size - 1) >>
487 sbi->s_log_groups_per_flex;
488 parent_group >>= sbi->s_log_groups_per_flex;
491 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
492 avefreei = freei / ngroups;
493 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
494 avefreeb = freeb;
495 do_div(avefreeb, ngroups);
496 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
498 if (S_ISDIR(mode) &&
499 ((parent == sb->s_root->d_inode) ||
500 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
501 int best_ndir = inodes_per_group;
502 int ret = -1;
504 if (qstr) {
505 hinfo.hash_version = DX_HASH_HALF_MD4;
506 hinfo.seed = sbi->s_hash_seed;
507 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
508 grp = hinfo.hash;
509 } else
510 get_random_bytes(&grp, sizeof(grp));
511 parent_group = (unsigned)grp % ngroups;
512 for (i = 0; i < ngroups; i++) {
513 g = (parent_group + i) % ngroups;
514 get_orlov_stats(sb, g, flex_size, &stats);
515 if (!stats.free_inodes)
516 continue;
517 if (stats.used_dirs >= best_ndir)
518 continue;
519 if (stats.free_inodes < avefreei)
520 continue;
521 if (stats.free_blocks < avefreeb)
522 continue;
523 grp = g;
524 ret = 0;
525 best_ndir = stats.used_dirs;
527 if (ret)
528 goto fallback;
529 found_flex_bg:
530 if (flex_size == 1) {
531 *group = grp;
532 return 0;
536 * We pack inodes at the beginning of the flexgroup's
537 * inode tables. Block allocation decisions will do
538 * something similar, although regular files will
539 * start at 2nd block group of the flexgroup. See
540 * ext4_ext_find_goal() and ext4_find_near().
542 grp *= flex_size;
543 for (i = 0; i < flex_size; i++) {
544 if (grp+i >= real_ngroups)
545 break;
546 desc = ext4_get_group_desc(sb, grp+i, NULL);
547 if (desc && ext4_free_inodes_count(sb, desc)) {
548 *group = grp+i;
549 return 0;
552 goto fallback;
555 max_dirs = ndirs / ngroups + inodes_per_group / 16;
556 min_inodes = avefreei - inodes_per_group*flex_size / 4;
557 if (min_inodes < 1)
558 min_inodes = 1;
559 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
562 * Start looking in the flex group where we last allocated an
563 * inode for this parent directory
565 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
566 parent_group = EXT4_I(parent)->i_last_alloc_group;
567 if (flex_size > 1)
568 parent_group >>= sbi->s_log_groups_per_flex;
571 for (i = 0; i < ngroups; i++) {
572 grp = (parent_group + i) % ngroups;
573 get_orlov_stats(sb, grp, flex_size, &stats);
574 if (stats.used_dirs >= max_dirs)
575 continue;
576 if (stats.free_inodes < min_inodes)
577 continue;
578 if (stats.free_blocks < min_blocks)
579 continue;
580 goto found_flex_bg;
583 fallback:
584 ngroups = real_ngroups;
585 avefreei = freei / ngroups;
586 fallback_retry:
587 parent_group = EXT4_I(parent)->i_block_group;
588 for (i = 0; i < ngroups; i++) {
589 grp = (parent_group + i) % ngroups;
590 desc = ext4_get_group_desc(sb, grp, NULL);
591 if (desc && ext4_free_inodes_count(sb, desc) &&
592 ext4_free_inodes_count(sb, desc) >= avefreei) {
593 *group = grp;
594 return 0;
598 if (avefreei) {
600 * The free-inodes counter is approximate, and for really small
601 * filesystems the above test can fail to find any blockgroups
603 avefreei = 0;
604 goto fallback_retry;
607 return -1;
610 static int find_group_other(struct super_block *sb, struct inode *parent,
611 ext4_group_t *group, int mode)
613 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
614 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
615 struct ext4_group_desc *desc;
616 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
619 * Try to place the inode is the same flex group as its
620 * parent. If we can't find space, use the Orlov algorithm to
621 * find another flex group, and store that information in the
622 * parent directory's inode information so that use that flex
623 * group for future allocations.
625 if (flex_size > 1) {
626 int retry = 0;
628 try_again:
629 parent_group &= ~(flex_size-1);
630 last = parent_group + flex_size;
631 if (last > ngroups)
632 last = ngroups;
633 for (i = parent_group; i < last; i++) {
634 desc = ext4_get_group_desc(sb, i, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc)) {
636 *group = i;
637 return 0;
640 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
641 retry = 1;
642 parent_group = EXT4_I(parent)->i_last_alloc_group;
643 goto try_again;
646 * If this didn't work, use the Orlov search algorithm
647 * to find a new flex group; we pass in the mode to
648 * avoid the topdir algorithms.
650 *group = parent_group + flex_size;
651 if (*group > ngroups)
652 *group = 0;
653 return find_group_orlov(sb, parent, group, mode, 0);
657 * Try to place the inode in its parent directory
659 *group = parent_group;
660 desc = ext4_get_group_desc(sb, *group, NULL);
661 if (desc && ext4_free_inodes_count(sb, desc) &&
662 ext4_free_blks_count(sb, desc))
663 return 0;
666 * We're going to place this inode in a different blockgroup from its
667 * parent. We want to cause files in a common directory to all land in
668 * the same blockgroup. But we want files which are in a different
669 * directory which shares a blockgroup with our parent to land in a
670 * different blockgroup.
672 * So add our directory's i_ino into the starting point for the hash.
674 *group = (*group + parent->i_ino) % ngroups;
677 * Use a quadratic hash to find a group with a free inode and some free
678 * blocks.
680 for (i = 1; i < ngroups; i <<= 1) {
681 *group += i;
682 if (*group >= ngroups)
683 *group -= ngroups;
684 desc = ext4_get_group_desc(sb, *group, NULL);
685 if (desc && ext4_free_inodes_count(sb, desc) &&
686 ext4_free_blks_count(sb, desc))
687 return 0;
691 * That failed: try linear search for a free inode, even if that group
692 * has no free blocks.
694 *group = parent_group;
695 for (i = 0; i < ngroups; i++) {
696 if (++*group >= ngroups)
697 *group = 0;
698 desc = ext4_get_group_desc(sb, *group, NULL);
699 if (desc && ext4_free_inodes_count(sb, desc))
700 return 0;
703 return -1;
707 * claim the inode from the inode bitmap. If the group
708 * is uninit we need to take the groups's ext4_group_lock
709 * and clear the uninit flag. The inode bitmap update
710 * and group desc uninit flag clear should be done
711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
712 * doesn't race with the ext4_claim_inode
714 static int ext4_claim_inode(struct super_block *sb,
715 struct buffer_head *inode_bitmap_bh,
716 unsigned long ino, ext4_group_t group, int mode)
718 int free = 0, retval = 0, count;
719 struct ext4_sb_info *sbi = EXT4_SB(sb);
720 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
722 ext4_lock_group(sb, group);
723 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
724 /* not a free inode */
725 retval = 1;
726 goto err_ret;
728 ino++;
729 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
730 ino > EXT4_INODES_PER_GROUP(sb)) {
731 ext4_unlock_group(sb, group);
732 ext4_error(sb, __func__,
733 "reserved inode or inode > inodes count - "
734 "block_group = %u, inode=%lu", group,
735 ino + group * EXT4_INODES_PER_GROUP(sb));
736 return 1;
738 /* If we didn't allocate from within the initialized part of the inode
739 * table then we need to initialize up to this inode. */
740 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
742 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
743 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
744 /* When marking the block group with
745 * ~EXT4_BG_INODE_UNINIT we don't want to depend
746 * on the value of bg_itable_unused even though
747 * mke2fs could have initialized the same for us.
748 * Instead we calculated the value below
751 free = 0;
752 } else {
753 free = EXT4_INODES_PER_GROUP(sb) -
754 ext4_itable_unused_count(sb, gdp);
758 * Check the relative inode number against the last used
759 * relative inode number in this group. if it is greater
760 * we need to update the bg_itable_unused count
763 if (ino > free)
764 ext4_itable_unused_set(sb, gdp,
765 (EXT4_INODES_PER_GROUP(sb) - ino));
767 count = ext4_free_inodes_count(sb, gdp) - 1;
768 ext4_free_inodes_set(sb, gdp, count);
769 if (S_ISDIR(mode)) {
770 count = ext4_used_dirs_count(sb, gdp) + 1;
771 ext4_used_dirs_set(sb, gdp, count);
772 if (sbi->s_log_groups_per_flex) {
773 ext4_group_t f = ext4_flex_group(sbi, group);
775 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
778 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
779 err_ret:
780 ext4_unlock_group(sb, group);
781 return retval;
785 * There are two policies for allocating an inode. If the new inode is
786 * a directory, then a forward search is made for a block group with both
787 * free space and a low directory-to-inode ratio; if that fails, then of
788 * the groups with above-average free space, that group with the fewest
789 * directories already is chosen.
791 * For other inodes, search forward from the parent directory's block
792 * group to find a free inode.
794 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
795 const struct qstr *qstr, __u32 goal)
797 struct super_block *sb;
798 struct buffer_head *inode_bitmap_bh = NULL;
799 struct buffer_head *group_desc_bh;
800 ext4_group_t ngroups, group = 0;
801 unsigned long ino = 0;
802 struct inode *inode;
803 struct ext4_group_desc *gdp = NULL;
804 struct ext4_inode_info *ei;
805 struct ext4_sb_info *sbi;
806 int ret2, err = 0;
807 struct inode *ret;
808 ext4_group_t i;
809 int free = 0;
810 static int once = 1;
811 ext4_group_t flex_group;
813 /* Cannot create files in a deleted directory */
814 if (!dir || !dir->i_nlink)
815 return ERR_PTR(-EPERM);
817 sb = dir->i_sb;
818 ngroups = ext4_get_groups_count(sb);
819 trace_ext4_request_inode(dir, mode);
820 inode = new_inode(sb);
821 if (!inode)
822 return ERR_PTR(-ENOMEM);
823 ei = EXT4_I(inode);
824 sbi = EXT4_SB(sb);
826 if (!goal)
827 goal = sbi->s_inode_goal;
829 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
830 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
831 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
832 ret2 = 0;
833 goto got_group;
836 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
837 ret2 = find_group_flex(sb, dir, &group);
838 if (ret2 == -1) {
839 ret2 = find_group_other(sb, dir, &group, mode);
840 if (ret2 == 0 && once) {
841 once = 0;
842 printk(KERN_NOTICE "ext4: find_group_flex "
843 "failed, fallback succeeded dir %lu\n",
844 dir->i_ino);
847 goto got_group;
850 if (S_ISDIR(mode)) {
851 if (test_opt(sb, OLDALLOC))
852 ret2 = find_group_dir(sb, dir, &group);
853 else
854 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
855 } else
856 ret2 = find_group_other(sb, dir, &group, mode);
858 got_group:
859 EXT4_I(dir)->i_last_alloc_group = group;
860 err = -ENOSPC;
861 if (ret2 == -1)
862 goto out;
864 for (i = 0; i < ngroups; i++, ino = 0) {
865 err = -EIO;
867 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
868 if (!gdp)
869 goto fail;
871 brelse(inode_bitmap_bh);
872 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
873 if (!inode_bitmap_bh)
874 goto fail;
876 repeat_in_this_group:
877 ino = ext4_find_next_zero_bit((unsigned long *)
878 inode_bitmap_bh->b_data,
879 EXT4_INODES_PER_GROUP(sb), ino);
881 if (ino < EXT4_INODES_PER_GROUP(sb)) {
883 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
884 err = ext4_journal_get_write_access(handle,
885 inode_bitmap_bh);
886 if (err)
887 goto fail;
889 BUFFER_TRACE(group_desc_bh, "get_write_access");
890 err = ext4_journal_get_write_access(handle,
891 group_desc_bh);
892 if (err)
893 goto fail;
894 if (!ext4_claim_inode(sb, inode_bitmap_bh,
895 ino, group, mode)) {
896 /* we won it */
897 BUFFER_TRACE(inode_bitmap_bh,
898 "call ext4_handle_dirty_metadata");
899 err = ext4_handle_dirty_metadata(handle,
900 NULL,
901 inode_bitmap_bh);
902 if (err)
903 goto fail;
904 /* zero bit is inode number 1*/
905 ino++;
906 goto got;
908 /* we lost it */
909 ext4_handle_release_buffer(handle, inode_bitmap_bh);
910 ext4_handle_release_buffer(handle, group_desc_bh);
912 if (++ino < EXT4_INODES_PER_GROUP(sb))
913 goto repeat_in_this_group;
917 * This case is possible in concurrent environment. It is very
918 * rare. We cannot repeat the find_group_xxx() call because
919 * that will simply return the same blockgroup, because the
920 * group descriptor metadata has not yet been updated.
921 * So we just go onto the next blockgroup.
923 if (++group == ngroups)
924 group = 0;
926 err = -ENOSPC;
927 goto out;
929 got:
930 /* We may have to initialize the block bitmap if it isn't already */
931 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
932 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
933 struct buffer_head *block_bitmap_bh;
935 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
936 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
937 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
938 if (err) {
939 brelse(block_bitmap_bh);
940 goto fail;
943 free = 0;
944 ext4_lock_group(sb, group);
945 /* recheck and clear flag under lock if we still need to */
946 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
947 free = ext4_free_blocks_after_init(sb, group, gdp);
948 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
949 ext4_free_blks_set(sb, gdp, free);
950 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
951 gdp);
953 ext4_unlock_group(sb, group);
955 /* Don't need to dirty bitmap block if we didn't change it */
956 if (free) {
957 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
958 err = ext4_handle_dirty_metadata(handle,
959 NULL, block_bitmap_bh);
962 brelse(block_bitmap_bh);
963 if (err)
964 goto fail;
966 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
967 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
968 if (err)
969 goto fail;
971 percpu_counter_dec(&sbi->s_freeinodes_counter);
972 if (S_ISDIR(mode))
973 percpu_counter_inc(&sbi->s_dirs_counter);
974 sb->s_dirt = 1;
976 if (sbi->s_log_groups_per_flex) {
977 flex_group = ext4_flex_group(sbi, group);
978 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
981 inode->i_uid = current_fsuid();
982 if (test_opt(sb, GRPID))
983 inode->i_gid = dir->i_gid;
984 else if (dir->i_mode & S_ISGID) {
985 inode->i_gid = dir->i_gid;
986 if (S_ISDIR(mode))
987 mode |= S_ISGID;
988 } else
989 inode->i_gid = current_fsgid();
990 inode->i_mode = mode;
992 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
993 /* This is the optimal IO size (for stat), not the fs block size */
994 inode->i_blocks = 0;
995 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
996 ext4_current_time(inode);
998 memset(ei->i_data, 0, sizeof(ei->i_data));
999 ei->i_dir_start_lookup = 0;
1000 ei->i_disksize = 0;
1003 * Don't inherit extent flag from directory, amongst others. We set
1004 * extent flag on newly created directory and file only if -o extent
1005 * mount option is specified
1007 ei->i_flags =
1008 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1009 ei->i_file_acl = 0;
1010 ei->i_dtime = 0;
1011 ei->i_block_group = group;
1012 ei->i_last_alloc_group = ~0;
1014 ext4_set_inode_flags(inode);
1015 if (IS_DIRSYNC(inode))
1016 ext4_handle_sync(handle);
1017 if (insert_inode_locked(inode) < 0) {
1018 err = -EINVAL;
1019 goto fail_drop;
1021 spin_lock(&sbi->s_next_gen_lock);
1022 inode->i_generation = sbi->s_next_generation++;
1023 spin_unlock(&sbi->s_next_gen_lock);
1025 ei->i_state_flags = 0;
1026 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1028 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1030 ret = inode;
1031 if (vfs_dq_alloc_inode(inode)) {
1032 err = -EDQUOT;
1033 goto fail_drop;
1036 err = ext4_init_acl(handle, inode, dir);
1037 if (err)
1038 goto fail_free_drop;
1040 err = ext4_init_security(handle, inode, dir);
1041 if (err)
1042 goto fail_free_drop;
1044 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1045 /* set extent flag only for directory, file and normal symlink*/
1046 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1047 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1048 ext4_ext_tree_init(handle, inode);
1052 err = ext4_mark_inode_dirty(handle, inode);
1053 if (err) {
1054 ext4_std_error(sb, err);
1055 goto fail_free_drop;
1058 ext4_debug("allocating inode %lu\n", inode->i_ino);
1059 trace_ext4_allocate_inode(inode, dir, mode);
1060 goto really_out;
1061 fail:
1062 ext4_std_error(sb, err);
1063 out:
1064 iput(inode);
1065 ret = ERR_PTR(err);
1066 really_out:
1067 brelse(inode_bitmap_bh);
1068 return ret;
1070 fail_free_drop:
1071 vfs_dq_free_inode(inode);
1073 fail_drop:
1074 vfs_dq_drop(inode);
1075 inode->i_flags |= S_NOQUOTA;
1076 inode->i_nlink = 0;
1077 unlock_new_inode(inode);
1078 iput(inode);
1079 brelse(inode_bitmap_bh);
1080 return ERR_PTR(err);
1083 /* Verify that we are loading a valid orphan from disk */
1084 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1086 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1087 ext4_group_t block_group;
1088 int bit;
1089 struct buffer_head *bitmap_bh;
1090 struct inode *inode = NULL;
1091 long err = -EIO;
1093 /* Error cases - e2fsck has already cleaned up for us */
1094 if (ino > max_ino) {
1095 ext4_warning(sb, __func__,
1096 "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, __func__,
1105 "inode bitmap error for orphan %lu", ino);
1106 goto error;
1109 /* Having the inode bit set should be a 100% indicator that this
1110 * is a valid orphan (no e2fsck run on fs). Orphans also include
1111 * inodes that were being truncated, so we can't check i_nlink==0.
1113 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1114 goto bad_orphan;
1116 inode = ext4_iget(sb, ino);
1117 if (IS_ERR(inode))
1118 goto iget_failed;
1121 * If the orphans has i_nlinks > 0 then it should be able to be
1122 * truncated, otherwise it won't be removed from the orphan list
1123 * during processing and an infinite loop will result.
1125 if (inode->i_nlink && !ext4_can_truncate(inode))
1126 goto bad_orphan;
1128 if (NEXT_ORPHAN(inode) > max_ino)
1129 goto bad_orphan;
1130 brelse(bitmap_bh);
1131 return inode;
1133 iget_failed:
1134 err = PTR_ERR(inode);
1135 inode = NULL;
1136 bad_orphan:
1137 ext4_warning(sb, __func__,
1138 "bad orphan inode %lu! e2fsck was run?", ino);
1139 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1140 bit, (unsigned long long)bitmap_bh->b_blocknr,
1141 ext4_test_bit(bit, bitmap_bh->b_data));
1142 printk(KERN_NOTICE "inode=%p\n", inode);
1143 if (inode) {
1144 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1145 is_bad_inode(inode));
1146 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1147 NEXT_ORPHAN(inode));
1148 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1149 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1150 /* Avoid freeing blocks if we got a bad deleted inode */
1151 if (inode->i_nlink == 0)
1152 inode->i_blocks = 0;
1153 iput(inode);
1155 brelse(bitmap_bh);
1156 error:
1157 return ERR_PTR(err);
1160 unsigned long ext4_count_free_inodes(struct super_block *sb)
1162 unsigned long desc_count;
1163 struct ext4_group_desc *gdp;
1164 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1165 #ifdef EXT4FS_DEBUG
1166 struct ext4_super_block *es;
1167 unsigned long bitmap_count, x;
1168 struct buffer_head *bitmap_bh = NULL;
1170 es = EXT4_SB(sb)->s_es;
1171 desc_count = 0;
1172 bitmap_count = 0;
1173 gdp = NULL;
1174 for (i = 0; i < ngroups; i++) {
1175 gdp = ext4_get_group_desc(sb, i, NULL);
1176 if (!gdp)
1177 continue;
1178 desc_count += ext4_free_inodes_count(sb, gdp);
1179 brelse(bitmap_bh);
1180 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1181 if (!bitmap_bh)
1182 continue;
1184 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1185 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1186 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1187 bitmap_count += x;
1189 brelse(bitmap_bh);
1190 printk(KERN_DEBUG "ext4_count_free_inodes: "
1191 "stored = %u, computed = %lu, %lu\n",
1192 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1193 return desc_count;
1194 #else
1195 desc_count = 0;
1196 for (i = 0; i < ngroups; i++) {
1197 gdp = ext4_get_group_desc(sb, i, NULL);
1198 if (!gdp)
1199 continue;
1200 desc_count += ext4_free_inodes_count(sb, gdp);
1201 cond_resched();
1203 return desc_count;
1204 #endif
1207 /* Called at mount-time, super-block is locked */
1208 unsigned long ext4_count_dirs(struct super_block * sb)
1210 unsigned long count = 0;
1211 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1213 for (i = 0; i < ngroups; i++) {
1214 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1215 if (!gdp)
1216 continue;
1217 count += ext4_used_dirs_count(sb, gdp);
1219 return count;