Merge git://git.infradead.org/mtd-2.6
[linux-2.6] / fs / ext4 / ialloc.c
1 /*
2  *  linux/fs/ext4/ialloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  BSD ufs-inspired inode and directory allocation by
10  *  Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14
15 #include <linux/time.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.h>
20 #include <linux/stat.h>
21 #include <linux/string.h>
22 #include <linux/quotaops.h>
23 #include <linux/buffer_head.h>
24 #include <linux/random.h>
25 #include <linux/bitops.h>
26 #include <linux/blkdev.h>
27 #include <asm/byteorder.h>
28
29 #include "xattr.h"
30 #include "acl.h"
31 #include "group.h"
32
33 /*
34  * ialloc.c contains the inodes allocation and deallocation routines
35  */
36
37 /*
38  * The free inodes are managed by bitmaps.  A file system contains several
39  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
40  * block for inodes, N blocks for the inode table and data blocks.
41  *
42  * The file system contains group descriptors which are located after the
43  * super block.  Each descriptor contains the number of the bitmap block and
44  * the free blocks count in the block.
45  */
46
47 /*
48  * To avoid calling the atomic setbit hundreds or thousands of times, we only
49  * need to use it within a single byte (to ensure we get endianness right).
50  * We can use memset for the rest of the bitmap as there are no other users.
51  */
52 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53 {
54         int i;
55
56         if (start_bit >= end_bit)
57                 return;
58
59         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61                 ext4_set_bit(i, bitmap);
62         if (i < end_bit)
63                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64 }
65
66 /* Initializes an uninitialized inode bitmap */
67 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
68                                 ext4_group_t block_group,
69                                 struct ext4_group_desc *gdp)
70 {
71         struct ext4_sb_info *sbi = EXT4_SB(sb);
72
73         J_ASSERT_BH(bh, buffer_locked(bh));
74
75         /* If checksum is bad mark all blocks and inodes use to prevent
76          * allocation, essentially implementing a per-group read-only flag. */
77         if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
78                 ext4_error(sb, __FUNCTION__, "Checksum bad for group %lu\n",
79                            block_group);
80                 gdp->bg_free_blocks_count = 0;
81                 gdp->bg_free_inodes_count = 0;
82                 gdp->bg_itable_unused = 0;
83                 memset(bh->b_data, 0xff, sb->s_blocksize);
84                 return 0;
85         }
86
87         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
88         mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
89                         bh->b_data);
90
91         return EXT4_INODES_PER_GROUP(sb);
92 }
93
94 /*
95  * Read the inode allocation bitmap for a given block_group, reading
96  * into the specified slot in the superblock's bitmap cache.
97  *
98  * Return buffer_head of bitmap on success or NULL.
99  */
100 static struct buffer_head *
101 read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
102 {
103         struct ext4_group_desc *desc;
104         struct buffer_head *bh = NULL;
105
106         desc = ext4_get_group_desc(sb, block_group, NULL);
107         if (!desc)
108                 goto error_out;
109         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
110                 bh = sb_getblk(sb, ext4_inode_bitmap(sb, desc));
111                 if (!buffer_uptodate(bh)) {
112                         lock_buffer(bh);
113                         if (!buffer_uptodate(bh)) {
114                                 ext4_init_inode_bitmap(sb, bh, block_group,
115                                                        desc);
116                                 set_buffer_uptodate(bh);
117                         }
118                         unlock_buffer(bh);
119                 }
120         } else {
121                 bh = sb_bread(sb, ext4_inode_bitmap(sb, desc));
122         }
123         if (!bh)
124                 ext4_error(sb, "read_inode_bitmap",
125                             "Cannot read inode bitmap - "
126                             "block_group = %lu, inode_bitmap = %llu",
127                             block_group, ext4_inode_bitmap(sb, desc));
128 error_out:
129         return bh;
130 }
131
132 /*
133  * NOTE! When we get the inode, we're the only people
134  * that have access to it, and as such there are no
135  * race conditions we have to worry about. The inode
136  * is not on the hash-lists, and it cannot be reached
137  * through the filesystem because the directory entry
138  * has been deleted earlier.
139  *
140  * HOWEVER: we must make sure that we get no aliases,
141  * which means that we have to call "clear_inode()"
142  * _before_ we mark the inode not in use in the inode
143  * bitmaps. Otherwise a newly created file might use
144  * the same inode number (not actually the same pointer
145  * though), and then we'd have two inodes sharing the
146  * same inode number and space on the harddisk.
147  */
148 void ext4_free_inode (handle_t *handle, struct inode * inode)
149 {
150         struct super_block * sb = inode->i_sb;
151         int is_directory;
152         unsigned long ino;
153         struct buffer_head *bitmap_bh = NULL;
154         struct buffer_head *bh2;
155         ext4_group_t block_group;
156         unsigned long bit;
157         struct ext4_group_desc * gdp;
158         struct ext4_super_block * es;
159         struct ext4_sb_info *sbi;
160         int fatal = 0, err;
161
162         if (atomic_read(&inode->i_count) > 1) {
163                 printk ("ext4_free_inode: inode has count=%d\n",
164                                         atomic_read(&inode->i_count));
165                 return;
166         }
167         if (inode->i_nlink) {
168                 printk ("ext4_free_inode: inode has nlink=%d\n",
169                         inode->i_nlink);
170                 return;
171         }
172         if (!sb) {
173                 printk("ext4_free_inode: inode on nonexistent device\n");
174                 return;
175         }
176         sbi = EXT4_SB(sb);
177
178         ino = inode->i_ino;
179         ext4_debug ("freeing inode %lu\n", ino);
180
181         /*
182          * Note: we must free any quota before locking the superblock,
183          * as writing the quota to disk may need the lock as well.
184          */
185         DQUOT_INIT(inode);
186         ext4_xattr_delete_inode(handle, inode);
187         DQUOT_FREE_INODE(inode);
188         DQUOT_DROP(inode);
189
190         is_directory = S_ISDIR(inode->i_mode);
191
192         /* Do this BEFORE marking the inode not in use or returning an error */
193         clear_inode (inode);
194
195         es = EXT4_SB(sb)->s_es;
196         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
197                 ext4_error (sb, "ext4_free_inode",
198                             "reserved or nonexistent inode %lu", ino);
199                 goto error_return;
200         }
201         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
202         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
203         bitmap_bh = read_inode_bitmap(sb, block_group);
204         if (!bitmap_bh)
205                 goto error_return;
206
207         BUFFER_TRACE(bitmap_bh, "get_write_access");
208         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
209         if (fatal)
210                 goto error_return;
211
212         /* Ok, now we can actually update the inode bitmaps.. */
213         if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
214                                         bit, bitmap_bh->b_data))
215                 ext4_error (sb, "ext4_free_inode",
216                               "bit already cleared for inode %lu", ino);
217         else {
218                 gdp = ext4_get_group_desc (sb, block_group, &bh2);
219
220                 BUFFER_TRACE(bh2, "get_write_access");
221                 fatal = ext4_journal_get_write_access(handle, bh2);
222                 if (fatal) goto error_return;
223
224                 if (gdp) {
225                         spin_lock(sb_bgl_lock(sbi, block_group));
226                         gdp->bg_free_inodes_count = cpu_to_le16(
227                                 le16_to_cpu(gdp->bg_free_inodes_count) + 1);
228                         if (is_directory)
229                                 gdp->bg_used_dirs_count = cpu_to_le16(
230                                   le16_to_cpu(gdp->bg_used_dirs_count) - 1);
231                         gdp->bg_checksum = ext4_group_desc_csum(sbi,
232                                                         block_group, gdp);
233                         spin_unlock(sb_bgl_lock(sbi, block_group));
234                         percpu_counter_inc(&sbi->s_freeinodes_counter);
235                         if (is_directory)
236                                 percpu_counter_dec(&sbi->s_dirs_counter);
237
238                 }
239                 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
240                 err = ext4_journal_dirty_metadata(handle, bh2);
241                 if (!fatal) fatal = err;
242         }
243         BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata");
244         err = ext4_journal_dirty_metadata(handle, bitmap_bh);
245         if (!fatal)
246                 fatal = err;
247         sb->s_dirt = 1;
248 error_return:
249         brelse(bitmap_bh);
250         ext4_std_error(sb, fatal);
251 }
252
253 /*
254  * There are two policies for allocating an inode.  If the new inode is
255  * a directory, then a forward search is made for a block group with both
256  * free space and a low directory-to-inode ratio; if that fails, then of
257  * the groups with above-average free space, that group with the fewest
258  * directories already is chosen.
259  *
260  * For other inodes, search forward from the parent directory\'s block
261  * group to find a free inode.
262  */
263 static int find_group_dir(struct super_block *sb, struct inode *parent,
264                                 ext4_group_t *best_group)
265 {
266         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
267         unsigned int freei, avefreei;
268         struct ext4_group_desc *desc, *best_desc = NULL;
269         ext4_group_t group;
270         int ret = -1;
271
272         freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
273         avefreei = freei / ngroups;
274
275         for (group = 0; group < ngroups; group++) {
276                 desc = ext4_get_group_desc (sb, group, NULL);
277                 if (!desc || !desc->bg_free_inodes_count)
278                         continue;
279                 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
280                         continue;
281                 if (!best_desc ||
282                     (le16_to_cpu(desc->bg_free_blocks_count) >
283                      le16_to_cpu(best_desc->bg_free_blocks_count))) {
284                         *best_group = group;
285                         best_desc = desc;
286                         ret = 0;
287                 }
288         }
289         return ret;
290 }
291
292 /*
293  * Orlov's allocator for directories.
294  *
295  * We always try to spread first-level directories.
296  *
297  * If there are blockgroups with both free inodes and free blocks counts
298  * not worse than average we return one with smallest directory count.
299  * Otherwise we simply return a random group.
300  *
301  * For the rest rules look so:
302  *
303  * It's OK to put directory into a group unless
304  * it has too many directories already (max_dirs) or
305  * it has too few free inodes left (min_inodes) or
306  * it has too few free blocks left (min_blocks) or
307  * it's already running too large debt (max_debt).
308  * Parent's group is prefered, if it doesn't satisfy these
309  * conditions we search cyclically through the rest. If none
310  * of the groups look good we just look for a group with more
311  * free inodes than average (starting at parent's group).
312  *
313  * Debt is incremented each time we allocate a directory and decremented
314  * when we allocate an inode, within 0--255.
315  */
316
317 #define INODE_COST 64
318 #define BLOCK_COST 256
319
320 static int find_group_orlov(struct super_block *sb, struct inode *parent,
321                                 ext4_group_t *group)
322 {
323         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
324         struct ext4_sb_info *sbi = EXT4_SB(sb);
325         struct ext4_super_block *es = sbi->s_es;
326         ext4_group_t ngroups = sbi->s_groups_count;
327         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
328         unsigned int freei, avefreei;
329         ext4_fsblk_t freeb, avefreeb;
330         ext4_fsblk_t blocks_per_dir;
331         unsigned int ndirs;
332         int max_debt, max_dirs, min_inodes;
333         ext4_grpblk_t min_blocks;
334         ext4_group_t i;
335         struct ext4_group_desc *desc;
336
337         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
338         avefreei = freei / ngroups;
339         freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
340         avefreeb = freeb;
341         do_div(avefreeb, ngroups);
342         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
343
344         if ((parent == sb->s_root->d_inode) ||
345             (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
346                 int best_ndir = inodes_per_group;
347                 ext4_group_t grp;
348                 int ret = -1;
349
350                 get_random_bytes(&grp, sizeof(grp));
351                 parent_group = (unsigned)grp % ngroups;
352                 for (i = 0; i < ngroups; i++) {
353                         grp = (parent_group + i) % ngroups;
354                         desc = ext4_get_group_desc(sb, grp, NULL);
355                         if (!desc || !desc->bg_free_inodes_count)
356                                 continue;
357                         if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
358                                 continue;
359                         if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
360                                 continue;
361                         if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
362                                 continue;
363                         *group = grp;
364                         ret = 0;
365                         best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
366                 }
367                 if (ret == 0)
368                         return ret;
369                 goto fallback;
370         }
371
372         blocks_per_dir = ext4_blocks_count(es) - freeb;
373         do_div(blocks_per_dir, ndirs);
374
375         max_dirs = ndirs / ngroups + inodes_per_group / 16;
376         min_inodes = avefreei - inodes_per_group / 4;
377         min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
378
379         max_debt = EXT4_BLOCKS_PER_GROUP(sb);
380         max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
381         if (max_debt * INODE_COST > inodes_per_group)
382                 max_debt = inodes_per_group / INODE_COST;
383         if (max_debt > 255)
384                 max_debt = 255;
385         if (max_debt == 0)
386                 max_debt = 1;
387
388         for (i = 0; i < ngroups; i++) {
389                 *group = (parent_group + i) % ngroups;
390                 desc = ext4_get_group_desc(sb, *group, NULL);
391                 if (!desc || !desc->bg_free_inodes_count)
392                         continue;
393                 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
394                         continue;
395                 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
396                         continue;
397                 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
398                         continue;
399                 return 0;
400         }
401
402 fallback:
403         for (i = 0; i < ngroups; i++) {
404                 *group = (parent_group + i) % ngroups;
405                 desc = ext4_get_group_desc(sb, *group, NULL);
406                 if (desc && desc->bg_free_inodes_count &&
407                         le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
408                         return 0;
409         }
410
411         if (avefreei) {
412                 /*
413                  * The free-inodes counter is approximate, and for really small
414                  * filesystems the above test can fail to find any blockgroups
415                  */
416                 avefreei = 0;
417                 goto fallback;
418         }
419
420         return -1;
421 }
422
423 static int find_group_other(struct super_block *sb, struct inode *parent,
424                                 ext4_group_t *group)
425 {
426         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
427         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
428         struct ext4_group_desc *desc;
429         ext4_group_t i;
430
431         /*
432          * Try to place the inode in its parent directory
433          */
434         *group = parent_group;
435         desc = ext4_get_group_desc(sb, *group, NULL);
436         if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
437                         le16_to_cpu(desc->bg_free_blocks_count))
438                 return 0;
439
440         /*
441          * We're going to place this inode in a different blockgroup from its
442          * parent.  We want to cause files in a common directory to all land in
443          * the same blockgroup.  But we want files which are in a different
444          * directory which shares a blockgroup with our parent to land in a
445          * different blockgroup.
446          *
447          * So add our directory's i_ino into the starting point for the hash.
448          */
449         *group = (*group + parent->i_ino) % ngroups;
450
451         /*
452          * Use a quadratic hash to find a group with a free inode and some free
453          * blocks.
454          */
455         for (i = 1; i < ngroups; i <<= 1) {
456                 *group += i;
457                 if (*group >= ngroups)
458                         *group -= ngroups;
459                 desc = ext4_get_group_desc(sb, *group, NULL);
460                 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
461                                 le16_to_cpu(desc->bg_free_blocks_count))
462                         return 0;
463         }
464
465         /*
466          * That failed: try linear search for a free inode, even if that group
467          * has no free blocks.
468          */
469         *group = parent_group;
470         for (i = 0; i < ngroups; i++) {
471                 if (++*group >= ngroups)
472                         *group = 0;
473                 desc = ext4_get_group_desc(sb, *group, NULL);
474                 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
475                         return 0;
476         }
477
478         return -1;
479 }
480
481 /*
482  * There are two policies for allocating an inode.  If the new inode is
483  * a directory, then a forward search is made for a block group with both
484  * free space and a low directory-to-inode ratio; if that fails, then of
485  * the groups with above-average free space, that group with the fewest
486  * directories already is chosen.
487  *
488  * For other inodes, search forward from the parent directory's block
489  * group to find a free inode.
490  */
491 struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
492 {
493         struct super_block *sb;
494         struct buffer_head *bitmap_bh = NULL;
495         struct buffer_head *bh2;
496         ext4_group_t group = 0;
497         unsigned long ino = 0;
498         struct inode * inode;
499         struct ext4_group_desc * gdp = NULL;
500         struct ext4_super_block * es;
501         struct ext4_inode_info *ei;
502         struct ext4_sb_info *sbi;
503         int ret2, err = 0;
504         struct inode *ret;
505         ext4_group_t i;
506         int free = 0;
507
508         /* Cannot create files in a deleted directory */
509         if (!dir || !dir->i_nlink)
510                 return ERR_PTR(-EPERM);
511
512         sb = dir->i_sb;
513         inode = new_inode(sb);
514         if (!inode)
515                 return ERR_PTR(-ENOMEM);
516         ei = EXT4_I(inode);
517
518         sbi = EXT4_SB(sb);
519         es = sbi->s_es;
520         if (S_ISDIR(mode)) {
521                 if (test_opt (sb, OLDALLOC))
522                         ret2 = find_group_dir(sb, dir, &group);
523                 else
524                         ret2 = find_group_orlov(sb, dir, &group);
525         } else
526                 ret2 = find_group_other(sb, dir, &group);
527
528         err = -ENOSPC;
529         if (ret2 == -1)
530                 goto out;
531
532         for (i = 0; i < sbi->s_groups_count; i++) {
533                 err = -EIO;
534
535                 gdp = ext4_get_group_desc(sb, group, &bh2);
536                 if (!gdp)
537                         goto fail;
538
539                 brelse(bitmap_bh);
540                 bitmap_bh = read_inode_bitmap(sb, group);
541                 if (!bitmap_bh)
542                         goto fail;
543
544                 ino = 0;
545
546 repeat_in_this_group:
547                 ino = ext4_find_next_zero_bit((unsigned long *)
548                                 bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
549                 if (ino < EXT4_INODES_PER_GROUP(sb)) {
550
551                         BUFFER_TRACE(bitmap_bh, "get_write_access");
552                         err = ext4_journal_get_write_access(handle, bitmap_bh);
553                         if (err)
554                                 goto fail;
555
556                         if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group),
557                                                 ino, bitmap_bh->b_data)) {
558                                 /* we won it */
559                                 BUFFER_TRACE(bitmap_bh,
560                                         "call ext4_journal_dirty_metadata");
561                                 err = ext4_journal_dirty_metadata(handle,
562                                                                 bitmap_bh);
563                                 if (err)
564                                         goto fail;
565                                 goto got;
566                         }
567                         /* we lost it */
568                         jbd2_journal_release_buffer(handle, bitmap_bh);
569
570                         if (++ino < EXT4_INODES_PER_GROUP(sb))
571                                 goto repeat_in_this_group;
572                 }
573
574                 /*
575                  * This case is possible in concurrent environment.  It is very
576                  * rare.  We cannot repeat the find_group_xxx() call because
577                  * that will simply return the same blockgroup, because the
578                  * group descriptor metadata has not yet been updated.
579                  * So we just go onto the next blockgroup.
580                  */
581                 if (++group == sbi->s_groups_count)
582                         group = 0;
583         }
584         err = -ENOSPC;
585         goto out;
586
587 got:
588         ino++;
589         if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
590             ino > EXT4_INODES_PER_GROUP(sb)) {
591                 ext4_error(sb, __FUNCTION__,
592                            "reserved inode or inode > inodes count - "
593                            "block_group = %lu, inode=%lu", group,
594                            ino + group * EXT4_INODES_PER_GROUP(sb));
595                 err = -EIO;
596                 goto fail;
597         }
598
599         BUFFER_TRACE(bh2, "get_write_access");
600         err = ext4_journal_get_write_access(handle, bh2);
601         if (err) goto fail;
602
603         /* We may have to initialize the block bitmap if it isn't already */
604         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
605             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
606                 struct buffer_head *block_bh = read_block_bitmap(sb, group);
607
608                 BUFFER_TRACE(block_bh, "get block bitmap access");
609                 err = ext4_journal_get_write_access(handle, block_bh);
610                 if (err) {
611                         brelse(block_bh);
612                         goto fail;
613                 }
614
615                 free = 0;
616                 spin_lock(sb_bgl_lock(sbi, group));
617                 /* recheck and clear flag under lock if we still need to */
618                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
619                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
620                         free = ext4_free_blocks_after_init(sb, group, gdp);
621                         gdp->bg_free_blocks_count = cpu_to_le16(free);
622                 }
623                 spin_unlock(sb_bgl_lock(sbi, group));
624
625                 /* Don't need to dirty bitmap block if we didn't change it */
626                 if (free) {
627                         BUFFER_TRACE(block_bh, "dirty block bitmap");
628                         err = ext4_journal_dirty_metadata(handle, block_bh);
629                 }
630
631                 brelse(block_bh);
632                 if (err)
633                         goto fail;
634         }
635
636         spin_lock(sb_bgl_lock(sbi, group));
637         /* If we didn't allocate from within the initialized part of the inode
638          * table then we need to initialize up to this inode. */
639         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
640                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
641                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
642
643                         /* When marking the block group with
644                          * ~EXT4_BG_INODE_UNINIT we don't want to depend
645                          * on the value of bg_itable_unsed even though
646                          * mke2fs could have initialized the same for us.
647                          * Instead we calculated the value below
648                          */
649
650                         free = 0;
651                 } else {
652                         free = EXT4_INODES_PER_GROUP(sb) -
653                                 le16_to_cpu(gdp->bg_itable_unused);
654                 }
655
656                 /*
657                  * Check the relative inode number against the last used
658                  * relative inode number in this group. if it is greater
659                  * we need to  update the bg_itable_unused count
660                  *
661                  */
662                 if (ino > free)
663                         gdp->bg_itable_unused =
664                                 cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino);
665         }
666
667         gdp->bg_free_inodes_count =
668                 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
669         if (S_ISDIR(mode)) {
670                 gdp->bg_used_dirs_count =
671                         cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
672         }
673         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
674         spin_unlock(sb_bgl_lock(sbi, group));
675         BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
676         err = ext4_journal_dirty_metadata(handle, bh2);
677         if (err) goto fail;
678
679         percpu_counter_dec(&sbi->s_freeinodes_counter);
680         if (S_ISDIR(mode))
681                 percpu_counter_inc(&sbi->s_dirs_counter);
682         sb->s_dirt = 1;
683
684         inode->i_uid = current->fsuid;
685         if (test_opt (sb, GRPID))
686                 inode->i_gid = dir->i_gid;
687         else if (dir->i_mode & S_ISGID) {
688                 inode->i_gid = dir->i_gid;
689                 if (S_ISDIR(mode))
690                         mode |= S_ISGID;
691         } else
692                 inode->i_gid = current->fsgid;
693         inode->i_mode = mode;
694
695         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
696         /* This is the optimal IO size (for stat), not the fs block size */
697         inode->i_blocks = 0;
698         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
699                                                        ext4_current_time(inode);
700
701         memset(ei->i_data, 0, sizeof(ei->i_data));
702         ei->i_dir_start_lookup = 0;
703         ei->i_disksize = 0;
704
705         ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL;
706         if (S_ISLNK(mode))
707                 ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL);
708         /* dirsync only applies to directories */
709         if (!S_ISDIR(mode))
710                 ei->i_flags &= ~EXT4_DIRSYNC_FL;
711         ei->i_file_acl = 0;
712         ei->i_dtime = 0;
713         ei->i_block_alloc_info = NULL;
714         ei->i_block_group = group;
715
716         ext4_set_inode_flags(inode);
717         if (IS_DIRSYNC(inode))
718                 handle->h_sync = 1;
719         insert_inode_hash(inode);
720         spin_lock(&sbi->s_next_gen_lock);
721         inode->i_generation = sbi->s_next_generation++;
722         spin_unlock(&sbi->s_next_gen_lock);
723
724         ei->i_state = EXT4_STATE_NEW;
725
726         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
727
728         ret = inode;
729         if(DQUOT_ALLOC_INODE(inode)) {
730                 err = -EDQUOT;
731                 goto fail_drop;
732         }
733
734         err = ext4_init_acl(handle, inode, dir);
735         if (err)
736                 goto fail_free_drop;
737
738         err = ext4_init_security(handle,inode, dir);
739         if (err)
740                 goto fail_free_drop;
741
742         err = ext4_mark_inode_dirty(handle, inode);
743         if (err) {
744                 ext4_std_error(sb, err);
745                 goto fail_free_drop;
746         }
747         if (test_opt(sb, EXTENTS)) {
748                 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
749                 ext4_ext_tree_init(handle, inode);
750                 err = ext4_update_incompat_feature(handle, sb,
751                                                 EXT4_FEATURE_INCOMPAT_EXTENTS);
752                 if (err)
753                         goto fail;
754         }
755
756         ext4_debug("allocating inode %lu\n", inode->i_ino);
757         goto really_out;
758 fail:
759         ext4_std_error(sb, err);
760 out:
761         iput(inode);
762         ret = ERR_PTR(err);
763 really_out:
764         brelse(bitmap_bh);
765         return ret;
766
767 fail_free_drop:
768         DQUOT_FREE_INODE(inode);
769
770 fail_drop:
771         DQUOT_DROP(inode);
772         inode->i_flags |= S_NOQUOTA;
773         inode->i_nlink = 0;
774         iput(inode);
775         brelse(bitmap_bh);
776         return ERR_PTR(err);
777 }
778
779 /* Verify that we are loading a valid orphan from disk */
780 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
781 {
782         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
783         ext4_group_t block_group;
784         int bit;
785         struct buffer_head *bitmap_bh;
786         struct inode *inode = NULL;
787         long err = -EIO;
788
789         /* Error cases - e2fsck has already cleaned up for us */
790         if (ino > max_ino) {
791                 ext4_warning(sb, __FUNCTION__,
792                              "bad orphan ino %lu!  e2fsck was run?", ino);
793                 goto error;
794         }
795
796         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
797         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
798         bitmap_bh = read_inode_bitmap(sb, block_group);
799         if (!bitmap_bh) {
800                 ext4_warning(sb, __FUNCTION__,
801                              "inode bitmap error for orphan %lu", ino);
802                 goto error;
803         }
804
805         /* Having the inode bit set should be a 100% indicator that this
806          * is a valid orphan (no e2fsck run on fs).  Orphans also include
807          * inodes that were being truncated, so we can't check i_nlink==0.
808          */
809         if (!ext4_test_bit(bit, bitmap_bh->b_data))
810                 goto bad_orphan;
811
812         inode = ext4_iget(sb, ino);
813         if (IS_ERR(inode))
814                 goto iget_failed;
815
816         if (NEXT_ORPHAN(inode) > max_ino)
817                 goto bad_orphan;
818         brelse(bitmap_bh);
819         return inode;
820
821 iget_failed:
822         err = PTR_ERR(inode);
823         inode = NULL;
824 bad_orphan:
825         ext4_warning(sb, __FUNCTION__,
826                      "bad orphan inode %lu!  e2fsck was run?", ino);
827         printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
828                bit, (unsigned long long)bitmap_bh->b_blocknr,
829                ext4_test_bit(bit, bitmap_bh->b_data));
830         printk(KERN_NOTICE "inode=%p\n", inode);
831         if (inode) {
832                 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
833                        is_bad_inode(inode));
834                 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
835                        NEXT_ORPHAN(inode));
836                 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
837                 /* Avoid freeing blocks if we got a bad deleted inode */
838                 if (inode->i_nlink == 0)
839                         inode->i_blocks = 0;
840                 iput(inode);
841         }
842         brelse(bitmap_bh);
843 error:
844         return ERR_PTR(err);
845 }
846
847 unsigned long ext4_count_free_inodes (struct super_block * sb)
848 {
849         unsigned long desc_count;
850         struct ext4_group_desc *gdp;
851         ext4_group_t i;
852 #ifdef EXT4FS_DEBUG
853         struct ext4_super_block *es;
854         unsigned long bitmap_count, x;
855         struct buffer_head *bitmap_bh = NULL;
856
857         es = EXT4_SB(sb)->s_es;
858         desc_count = 0;
859         bitmap_count = 0;
860         gdp = NULL;
861         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
862                 gdp = ext4_get_group_desc (sb, i, NULL);
863                 if (!gdp)
864                         continue;
865                 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
866                 brelse(bitmap_bh);
867                 bitmap_bh = read_inode_bitmap(sb, i);
868                 if (!bitmap_bh)
869                         continue;
870
871                 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
872                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
873                         i, le16_to_cpu(gdp->bg_free_inodes_count), x);
874                 bitmap_count += x;
875         }
876         brelse(bitmap_bh);
877         printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
878                 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
879         return desc_count;
880 #else
881         desc_count = 0;
882         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
883                 gdp = ext4_get_group_desc (sb, i, NULL);
884                 if (!gdp)
885                         continue;
886                 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
887                 cond_resched();
888         }
889         return desc_count;
890 #endif
891 }
892
893 /* Called at mount-time, super-block is locked */
894 unsigned long ext4_count_dirs (struct super_block * sb)
895 {
896         unsigned long count = 0;
897         ext4_group_t i;
898
899         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
900                 struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
901                 if (!gdp)
902                         continue;
903                 count += le16_to_cpu(gdp->bg_used_dirs_count);
904         }
905         return count;
906 }
907