507bfb3344d41d1cc857022e61eb7c09ab184231
[sfrench/cifs-2.6.git] / 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/stat.h>
18 #include <linux/string.h>
19 #include <linux/quotaops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/random.h>
22 #include <linux/bitops.h>
23 #include <linux/blkdev.h>
24 #include <linux/cred.h>
25
26 #include <asm/byteorder.h>
27
28 #include "ext4.h"
29 #include "ext4_jbd2.h"
30 #include "xattr.h"
31 #include "acl.h"
32
33 #include <trace/events/ext4.h>
34
35 /*
36  * ialloc.c contains the inodes allocation and deallocation routines
37  */
38
39 /*
40  * The free inodes are managed by bitmaps.  A file system contains several
41  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
42  * block for inodes, N blocks for the inode table and data blocks.
43  *
44  * The file system contains group descriptors which are located after the
45  * super block.  Each descriptor contains the number of the bitmap block and
46  * the free blocks count in the block.
47  */
48
49 /*
50  * To avoid calling the atomic setbit hundreds or thousands of times, we only
51  * need to use it within a single byte (to ensure we get endianness right).
52  * We can use memset for the rest of the bitmap as there are no other users.
53  */
54 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
55 {
56         int i;
57
58         if (start_bit >= end_bit)
59                 return;
60
61         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
62         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
63                 ext4_set_bit(i, bitmap);
64         if (i < end_bit)
65                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
66 }
67
68 /* Initializes an uninitialized inode bitmap */
69 static int ext4_init_inode_bitmap(struct super_block *sb,
70                                        struct buffer_head *bh,
71                                        ext4_group_t block_group,
72                                        struct ext4_group_desc *gdp)
73 {
74         struct ext4_group_info *grp;
75         struct ext4_sb_info *sbi = EXT4_SB(sb);
76         J_ASSERT_BH(bh, buffer_locked(bh));
77
78         /* If checksum is bad mark all blocks and inodes use to prevent
79          * allocation, essentially implementing a per-group read-only flag. */
80         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
81                 grp = ext4_get_group_info(sb, block_group);
82                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
83                         percpu_counter_sub(&sbi->s_freeclusters_counter,
84                                            grp->bb_free);
85                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
86                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
87                         int count;
88                         count = ext4_free_inodes_count(sb, gdp);
89                         percpu_counter_sub(&sbi->s_freeinodes_counter,
90                                            count);
91                 }
92                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
93                 return -EFSBADCRC;
94         }
95
96         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
97         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
98                         bh->b_data);
99         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
100                                    EXT4_INODES_PER_GROUP(sb) / 8);
101         ext4_group_desc_csum_set(sb, block_group, gdp);
102
103         return 0;
104 }
105
106 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
107 {
108         if (uptodate) {
109                 set_buffer_uptodate(bh);
110                 set_bitmap_uptodate(bh);
111         }
112         unlock_buffer(bh);
113         put_bh(bh);
114 }
115
116 static int ext4_validate_inode_bitmap(struct super_block *sb,
117                                       struct ext4_group_desc *desc,
118                                       ext4_group_t block_group,
119                                       struct buffer_head *bh)
120 {
121         ext4_fsblk_t    blk;
122         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
123         struct ext4_sb_info *sbi = EXT4_SB(sb);
124
125         if (buffer_verified(bh))
126                 return 0;
127         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
128                 return -EFSCORRUPTED;
129
130         ext4_lock_group(sb, block_group);
131         blk = ext4_inode_bitmap(sb, desc);
132         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
133                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
134                 ext4_unlock_group(sb, block_group);
135                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
136                            "inode_bitmap = %llu", block_group, blk);
137                 grp = ext4_get_group_info(sb, block_group);
138                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
139                         int count;
140                         count = ext4_free_inodes_count(sb, desc);
141                         percpu_counter_sub(&sbi->s_freeinodes_counter,
142                                            count);
143                 }
144                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
145                 return -EFSBADCRC;
146         }
147         set_buffer_verified(bh);
148         ext4_unlock_group(sb, block_group);
149         return 0;
150 }
151
152 /*
153  * Read the inode allocation bitmap for a given block_group, reading
154  * into the specified slot in the superblock's bitmap cache.
155  *
156  * Return buffer_head of bitmap on success or NULL.
157  */
158 static struct buffer_head *
159 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
160 {
161         struct ext4_group_desc *desc;
162         struct buffer_head *bh = NULL;
163         ext4_fsblk_t bitmap_blk;
164         int err;
165
166         desc = ext4_get_group_desc(sb, block_group, NULL);
167         if (!desc)
168                 return ERR_PTR(-EFSCORRUPTED);
169
170         bitmap_blk = ext4_inode_bitmap(sb, desc);
171         bh = sb_getblk(sb, bitmap_blk);
172         if (unlikely(!bh)) {
173                 ext4_error(sb, "Cannot read inode bitmap - "
174                             "block_group = %u, inode_bitmap = %llu",
175                             block_group, bitmap_blk);
176                 return ERR_PTR(-EIO);
177         }
178         if (bitmap_uptodate(bh))
179                 goto verify;
180
181         lock_buffer(bh);
182         if (bitmap_uptodate(bh)) {
183                 unlock_buffer(bh);
184                 goto verify;
185         }
186
187         ext4_lock_group(sb, block_group);
188         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
189                 err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
190                 set_bitmap_uptodate(bh);
191                 set_buffer_uptodate(bh);
192                 set_buffer_verified(bh);
193                 ext4_unlock_group(sb, block_group);
194                 unlock_buffer(bh);
195                 if (err) {
196                         ext4_error(sb, "Failed to init inode bitmap for group "
197                                    "%u: %d", block_group, err);
198                         goto out;
199                 }
200                 return bh;
201         }
202         ext4_unlock_group(sb, block_group);
203
204         if (buffer_uptodate(bh)) {
205                 /*
206                  * if not uninit if bh is uptodate,
207                  * bitmap is also uptodate
208                  */
209                 set_bitmap_uptodate(bh);
210                 unlock_buffer(bh);
211                 goto verify;
212         }
213         /*
214          * submit the buffer_head for reading
215          */
216         trace_ext4_load_inode_bitmap(sb, block_group);
217         bh->b_end_io = ext4_end_bitmap_read;
218         get_bh(bh);
219         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
220         wait_on_buffer(bh);
221         if (!buffer_uptodate(bh)) {
222                 put_bh(bh);
223                 ext4_error(sb, "Cannot read inode bitmap - "
224                            "block_group = %u, inode_bitmap = %llu",
225                            block_group, bitmap_blk);
226                 return ERR_PTR(-EIO);
227         }
228
229 verify:
230         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
231         if (err)
232                 goto out;
233         return bh;
234 out:
235         put_bh(bh);
236         return ERR_PTR(err);
237 }
238
239 /*
240  * NOTE! When we get the inode, we're the only people
241  * that have access to it, and as such there are no
242  * race conditions we have to worry about. The inode
243  * is not on the hash-lists, and it cannot be reached
244  * through the filesystem because the directory entry
245  * has been deleted earlier.
246  *
247  * HOWEVER: we must make sure that we get no aliases,
248  * which means that we have to call "clear_inode()"
249  * _before_ we mark the inode not in use in the inode
250  * bitmaps. Otherwise a newly created file might use
251  * the same inode number (not actually the same pointer
252  * though), and then we'd have two inodes sharing the
253  * same inode number and space on the harddisk.
254  */
255 void ext4_free_inode(handle_t *handle, struct inode *inode)
256 {
257         struct super_block *sb = inode->i_sb;
258         int is_directory;
259         unsigned long ino;
260         struct buffer_head *bitmap_bh = NULL;
261         struct buffer_head *bh2;
262         ext4_group_t block_group;
263         unsigned long bit;
264         struct ext4_group_desc *gdp;
265         struct ext4_super_block *es;
266         struct ext4_sb_info *sbi;
267         int fatal = 0, err, count, cleared;
268         struct ext4_group_info *grp;
269
270         if (!sb) {
271                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
272                        "nonexistent device\n", __func__, __LINE__);
273                 return;
274         }
275         if (atomic_read(&inode->i_count) > 1) {
276                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
277                          __func__, __LINE__, inode->i_ino,
278                          atomic_read(&inode->i_count));
279                 return;
280         }
281         if (inode->i_nlink) {
282                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
283                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
284                 return;
285         }
286         sbi = EXT4_SB(sb);
287
288         ino = inode->i_ino;
289         ext4_debug("freeing inode %lu\n", ino);
290         trace_ext4_free_inode(inode);
291
292         /*
293          * Note: we must free any quota before locking the superblock,
294          * as writing the quota to disk may need the lock as well.
295          */
296         dquot_initialize(inode);
297         dquot_free_inode(inode);
298         dquot_drop(inode);
299
300         is_directory = S_ISDIR(inode->i_mode);
301
302         /* Do this BEFORE marking the inode not in use or returning an error */
303         ext4_clear_inode(inode);
304
305         es = EXT4_SB(sb)->s_es;
306         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
307                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
308                 goto error_return;
309         }
310         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
311         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
312         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
313         /* Don't bother if the inode bitmap is corrupt. */
314         grp = ext4_get_group_info(sb, block_group);
315         if (IS_ERR(bitmap_bh)) {
316                 fatal = PTR_ERR(bitmap_bh);
317                 bitmap_bh = NULL;
318                 goto error_return;
319         }
320         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
321                 fatal = -EFSCORRUPTED;
322                 goto error_return;
323         }
324
325         BUFFER_TRACE(bitmap_bh, "get_write_access");
326         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
327         if (fatal)
328                 goto error_return;
329
330         fatal = -ESRCH;
331         gdp = ext4_get_group_desc(sb, block_group, &bh2);
332         if (gdp) {
333                 BUFFER_TRACE(bh2, "get_write_access");
334                 fatal = ext4_journal_get_write_access(handle, bh2);
335         }
336         ext4_lock_group(sb, block_group);
337         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
338         if (fatal || !cleared) {
339                 ext4_unlock_group(sb, block_group);
340                 goto out;
341         }
342
343         count = ext4_free_inodes_count(sb, gdp) + 1;
344         ext4_free_inodes_set(sb, gdp, count);
345         if (is_directory) {
346                 count = ext4_used_dirs_count(sb, gdp) - 1;
347                 ext4_used_dirs_set(sb, gdp, count);
348                 percpu_counter_dec(&sbi->s_dirs_counter);
349         }
350         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
351                                    EXT4_INODES_PER_GROUP(sb) / 8);
352         ext4_group_desc_csum_set(sb, block_group, gdp);
353         ext4_unlock_group(sb, block_group);
354
355         percpu_counter_inc(&sbi->s_freeinodes_counter);
356         if (sbi->s_log_groups_per_flex) {
357                 ext4_group_t f = ext4_flex_group(sbi, block_group);
358
359                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
360                 if (is_directory)
361                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
362         }
363         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
364         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
365 out:
366         if (cleared) {
367                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
368                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
369                 if (!fatal)
370                         fatal = err;
371         } else {
372                 ext4_error(sb, "bit already cleared for inode %lu", ino);
373                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
374                         int count;
375                         count = ext4_free_inodes_count(sb, gdp);
376                         percpu_counter_sub(&sbi->s_freeinodes_counter,
377                                            count);
378                 }
379                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
380         }
381
382 error_return:
383         brelse(bitmap_bh);
384         ext4_std_error(sb, fatal);
385 }
386
387 struct orlov_stats {
388         __u64 free_clusters;
389         __u32 free_inodes;
390         __u32 used_dirs;
391 };
392
393 /*
394  * Helper function for Orlov's allocator; returns critical information
395  * for a particular block group or flex_bg.  If flex_size is 1, then g
396  * is a block group number; otherwise it is flex_bg number.
397  */
398 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
399                             int flex_size, struct orlov_stats *stats)
400 {
401         struct ext4_group_desc *desc;
402         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
403
404         if (flex_size > 1) {
405                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
406                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
407                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
408                 return;
409         }
410
411         desc = ext4_get_group_desc(sb, g, NULL);
412         if (desc) {
413                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
414                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
415                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
416         } else {
417                 stats->free_inodes = 0;
418                 stats->free_clusters = 0;
419                 stats->used_dirs = 0;
420         }
421 }
422
423 /*
424  * Orlov's allocator for directories.
425  *
426  * We always try to spread first-level directories.
427  *
428  * If there are blockgroups with both free inodes and free blocks counts
429  * not worse than average we return one with smallest directory count.
430  * Otherwise we simply return a random group.
431  *
432  * For the rest rules look so:
433  *
434  * It's OK to put directory into a group unless
435  * it has too many directories already (max_dirs) or
436  * it has too few free inodes left (min_inodes) or
437  * it has too few free blocks left (min_blocks) or
438  * Parent's group is preferred, if it doesn't satisfy these
439  * conditions we search cyclically through the rest. If none
440  * of the groups look good we just look for a group with more
441  * free inodes than average (starting at parent's group).
442  */
443
444 static int find_group_orlov(struct super_block *sb, struct inode *parent,
445                             ext4_group_t *group, umode_t mode,
446                             const struct qstr *qstr)
447 {
448         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
449         struct ext4_sb_info *sbi = EXT4_SB(sb);
450         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
451         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
452         unsigned int freei, avefreei, grp_free;
453         ext4_fsblk_t freeb, avefreec;
454         unsigned int ndirs;
455         int max_dirs, min_inodes;
456         ext4_grpblk_t min_clusters;
457         ext4_group_t i, grp, g, ngroups;
458         struct ext4_group_desc *desc;
459         struct orlov_stats stats;
460         int flex_size = ext4_flex_bg_size(sbi);
461         struct dx_hash_info hinfo;
462
463         ngroups = real_ngroups;
464         if (flex_size > 1) {
465                 ngroups = (real_ngroups + flex_size - 1) >>
466                         sbi->s_log_groups_per_flex;
467                 parent_group >>= sbi->s_log_groups_per_flex;
468         }
469
470         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
471         avefreei = freei / ngroups;
472         freeb = EXT4_C2B(sbi,
473                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
474         avefreec = freeb;
475         do_div(avefreec, ngroups);
476         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
477
478         if (S_ISDIR(mode) &&
479             ((parent == d_inode(sb->s_root)) ||
480              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
481                 int best_ndir = inodes_per_group;
482                 int ret = -1;
483
484                 if (qstr) {
485                         hinfo.hash_version = DX_HASH_HALF_MD4;
486                         hinfo.seed = sbi->s_hash_seed;
487                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
488                         grp = hinfo.hash;
489                 } else
490                         grp = prandom_u32();
491                 parent_group = (unsigned)grp % ngroups;
492                 for (i = 0; i < ngroups; i++) {
493                         g = (parent_group + i) % ngroups;
494                         get_orlov_stats(sb, g, flex_size, &stats);
495                         if (!stats.free_inodes)
496                                 continue;
497                         if (stats.used_dirs >= best_ndir)
498                                 continue;
499                         if (stats.free_inodes < avefreei)
500                                 continue;
501                         if (stats.free_clusters < avefreec)
502                                 continue;
503                         grp = g;
504                         ret = 0;
505                         best_ndir = stats.used_dirs;
506                 }
507                 if (ret)
508                         goto fallback;
509         found_flex_bg:
510                 if (flex_size == 1) {
511                         *group = grp;
512                         return 0;
513                 }
514
515                 /*
516                  * We pack inodes at the beginning of the flexgroup's
517                  * inode tables.  Block allocation decisions will do
518                  * something similar, although regular files will
519                  * start at 2nd block group of the flexgroup.  See
520                  * ext4_ext_find_goal() and ext4_find_near().
521                  */
522                 grp *= flex_size;
523                 for (i = 0; i < flex_size; i++) {
524                         if (grp+i >= real_ngroups)
525                                 break;
526                         desc = ext4_get_group_desc(sb, grp+i, NULL);
527                         if (desc && ext4_free_inodes_count(sb, desc)) {
528                                 *group = grp+i;
529                                 return 0;
530                         }
531                 }
532                 goto fallback;
533         }
534
535         max_dirs = ndirs / ngroups + inodes_per_group / 16;
536         min_inodes = avefreei - inodes_per_group*flex_size / 4;
537         if (min_inodes < 1)
538                 min_inodes = 1;
539         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
540
541         /*
542          * Start looking in the flex group where we last allocated an
543          * inode for this parent directory
544          */
545         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
546                 parent_group = EXT4_I(parent)->i_last_alloc_group;
547                 if (flex_size > 1)
548                         parent_group >>= sbi->s_log_groups_per_flex;
549         }
550
551         for (i = 0; i < ngroups; i++) {
552                 grp = (parent_group + i) % ngroups;
553                 get_orlov_stats(sb, grp, flex_size, &stats);
554                 if (stats.used_dirs >= max_dirs)
555                         continue;
556                 if (stats.free_inodes < min_inodes)
557                         continue;
558                 if (stats.free_clusters < min_clusters)
559                         continue;
560                 goto found_flex_bg;
561         }
562
563 fallback:
564         ngroups = real_ngroups;
565         avefreei = freei / ngroups;
566 fallback_retry:
567         parent_group = EXT4_I(parent)->i_block_group;
568         for (i = 0; i < ngroups; i++) {
569                 grp = (parent_group + i) % ngroups;
570                 desc = ext4_get_group_desc(sb, grp, NULL);
571                 if (desc) {
572                         grp_free = ext4_free_inodes_count(sb, desc);
573                         if (grp_free && grp_free >= avefreei) {
574                                 *group = grp;
575                                 return 0;
576                         }
577                 }
578         }
579
580         if (avefreei) {
581                 /*
582                  * The free-inodes counter is approximate, and for really small
583                  * filesystems the above test can fail to find any blockgroups
584                  */
585                 avefreei = 0;
586                 goto fallback_retry;
587         }
588
589         return -1;
590 }
591
592 static int find_group_other(struct super_block *sb, struct inode *parent,
593                             ext4_group_t *group, umode_t mode)
594 {
595         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
596         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
597         struct ext4_group_desc *desc;
598         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
599
600         /*
601          * Try to place the inode is the same flex group as its
602          * parent.  If we can't find space, use the Orlov algorithm to
603          * find another flex group, and store that information in the
604          * parent directory's inode information so that use that flex
605          * group for future allocations.
606          */
607         if (flex_size > 1) {
608                 int retry = 0;
609
610         try_again:
611                 parent_group &= ~(flex_size-1);
612                 last = parent_group + flex_size;
613                 if (last > ngroups)
614                         last = ngroups;
615                 for  (i = parent_group; i < last; i++) {
616                         desc = ext4_get_group_desc(sb, i, NULL);
617                         if (desc && ext4_free_inodes_count(sb, desc)) {
618                                 *group = i;
619                                 return 0;
620                         }
621                 }
622                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
623                         retry = 1;
624                         parent_group = EXT4_I(parent)->i_last_alloc_group;
625                         goto try_again;
626                 }
627                 /*
628                  * If this didn't work, use the Orlov search algorithm
629                  * to find a new flex group; we pass in the mode to
630                  * avoid the topdir algorithms.
631                  */
632                 *group = parent_group + flex_size;
633                 if (*group > ngroups)
634                         *group = 0;
635                 return find_group_orlov(sb, parent, group, mode, NULL);
636         }
637
638         /*
639          * Try to place the inode in its parent directory
640          */
641         *group = parent_group;
642         desc = ext4_get_group_desc(sb, *group, NULL);
643         if (desc && ext4_free_inodes_count(sb, desc) &&
644             ext4_free_group_clusters(sb, desc))
645                 return 0;
646
647         /*
648          * We're going to place this inode in a different blockgroup from its
649          * parent.  We want to cause files in a common directory to all land in
650          * the same blockgroup.  But we want files which are in a different
651          * directory which shares a blockgroup with our parent to land in a
652          * different blockgroup.
653          *
654          * So add our directory's i_ino into the starting point for the hash.
655          */
656         *group = (*group + parent->i_ino) % ngroups;
657
658         /*
659          * Use a quadratic hash to find a group with a free inode and some free
660          * blocks.
661          */
662         for (i = 1; i < ngroups; i <<= 1) {
663                 *group += i;
664                 if (*group >= ngroups)
665                         *group -= ngroups;
666                 desc = ext4_get_group_desc(sb, *group, NULL);
667                 if (desc && ext4_free_inodes_count(sb, desc) &&
668                     ext4_free_group_clusters(sb, desc))
669                         return 0;
670         }
671
672         /*
673          * That failed: try linear search for a free inode, even if that group
674          * has no free blocks.
675          */
676         *group = parent_group;
677         for (i = 0; i < ngroups; i++) {
678                 if (++*group >= ngroups)
679                         *group = 0;
680                 desc = ext4_get_group_desc(sb, *group, NULL);
681                 if (desc && ext4_free_inodes_count(sb, desc))
682                         return 0;
683         }
684
685         return -1;
686 }
687
688 /*
689  * In no journal mode, if an inode has recently been deleted, we want
690  * to avoid reusing it until we're reasonably sure the inode table
691  * block has been written back to disk.  (Yes, these values are
692  * somewhat arbitrary...)
693  */
694 #define RECENTCY_MIN    5
695 #define RECENTCY_DIRTY  30
696
697 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
698 {
699         struct ext4_group_desc  *gdp;
700         struct ext4_inode       *raw_inode;
701         struct buffer_head      *bh;
702         unsigned long           dtime, now;
703         int     inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
704         int     offset, ret = 0, recentcy = RECENTCY_MIN;
705
706         gdp = ext4_get_group_desc(sb, group, NULL);
707         if (unlikely(!gdp))
708                 return 0;
709
710         bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
711                        (ino / inodes_per_block));
712         if (unlikely(!bh) || !buffer_uptodate(bh))
713                 /*
714                  * If the block is not in the buffer cache, then it
715                  * must have been written out.
716                  */
717                 goto out;
718
719         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
720         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
721         dtime = le32_to_cpu(raw_inode->i_dtime);
722         now = get_seconds();
723         if (buffer_dirty(bh))
724                 recentcy += RECENTCY_DIRTY;
725
726         if (dtime && (dtime < now) && (now < dtime + recentcy))
727                 ret = 1;
728 out:
729         brelse(bh);
730         return ret;
731 }
732
733 /*
734  * There are two policies for allocating an inode.  If the new inode is
735  * a directory, then a forward search is made for a block group with both
736  * free space and a low directory-to-inode ratio; if that fails, then of
737  * the groups with above-average free space, that group with the fewest
738  * directories already is chosen.
739  *
740  * For other inodes, search forward from the parent directory's block
741  * group to find a free inode.
742  */
743 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
744                                umode_t mode, const struct qstr *qstr,
745                                __u32 goal, uid_t *owner, __u32 i_flags,
746                                int handle_type, unsigned int line_no,
747                                int nblocks)
748 {
749         struct super_block *sb;
750         struct buffer_head *inode_bitmap_bh = NULL;
751         struct buffer_head *group_desc_bh;
752         ext4_group_t ngroups, group = 0;
753         unsigned long ino = 0;
754         struct inode *inode;
755         struct ext4_group_desc *gdp = NULL;
756         struct ext4_inode_info *ei;
757         struct ext4_sb_info *sbi;
758         int ret2, err;
759         struct inode *ret;
760         ext4_group_t i;
761         ext4_group_t flex_group;
762         struct ext4_group_info *grp;
763         int encrypt = 0;
764
765         /* Cannot create files in a deleted directory */
766         if (!dir || !dir->i_nlink)
767                 return ERR_PTR(-EPERM);
768
769         sb = dir->i_sb;
770         sbi = EXT4_SB(sb);
771
772         if (unlikely(ext4_forced_shutdown(sbi)))
773                 return ERR_PTR(-EIO);
774
775         if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
776             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
777             !(i_flags & EXT4_EA_INODE_FL)) {
778                 err = fscrypt_get_encryption_info(dir);
779                 if (err)
780                         return ERR_PTR(err);
781                 if (!fscrypt_has_encryption_key(dir))
782                         return ERR_PTR(-ENOKEY);
783                 encrypt = 1;
784         }
785
786         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
787 #ifdef CONFIG_EXT4_FS_POSIX_ACL
788                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
789
790                 if (p) {
791                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
792
793                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
794                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
795                                         NULL /* block_bh */, acl_size,
796                                         true /* is_create */);
797                         posix_acl_release(p);
798                 }
799 #endif
800
801 #ifdef CONFIG_SECURITY
802                 {
803                         int num_security_xattrs = 1;
804
805 #ifdef CONFIG_INTEGRITY
806                         num_security_xattrs++;
807 #endif
808                         /*
809                          * We assume that security xattrs are never
810                          * more than 1k.  In practice they are under
811                          * 128 bytes.
812                          */
813                         nblocks += num_security_xattrs *
814                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
815                                         NULL /* block_bh */, 1024,
816                                         true /* is_create */);
817                 }
818 #endif
819                 if (encrypt)
820                         nblocks += __ext4_xattr_set_credits(sb,
821                                         NULL /* inode */, NULL /* block_bh */,
822                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
823                                         true /* is_create */);
824         }
825
826         ngroups = ext4_get_groups_count(sb);
827         trace_ext4_request_inode(dir, mode);
828         inode = new_inode(sb);
829         if (!inode)
830                 return ERR_PTR(-ENOMEM);
831         ei = EXT4_I(inode);
832
833         /*
834          * Initialize owners and quota early so that we don't have to account
835          * for quota initialization worst case in standard inode creating
836          * transaction
837          */
838         if (owner) {
839                 inode->i_mode = mode;
840                 i_uid_write(inode, owner[0]);
841                 i_gid_write(inode, owner[1]);
842         } else if (test_opt(sb, GRPID)) {
843                 inode->i_mode = mode;
844                 inode->i_uid = current_fsuid();
845                 inode->i_gid = dir->i_gid;
846         } else
847                 inode_init_owner(inode, dir, mode);
848
849         if (ext4_has_feature_project(sb) &&
850             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
851                 ei->i_projid = EXT4_I(dir)->i_projid;
852         else
853                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
854
855         err = dquot_initialize(inode);
856         if (err)
857                 goto out;
858
859         if (!goal)
860                 goal = sbi->s_inode_goal;
861
862         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
863                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
864                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
865                 ret2 = 0;
866                 goto got_group;
867         }
868
869         if (S_ISDIR(mode))
870                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
871         else
872                 ret2 = find_group_other(sb, dir, &group, mode);
873
874 got_group:
875         EXT4_I(dir)->i_last_alloc_group = group;
876         err = -ENOSPC;
877         if (ret2 == -1)
878                 goto out;
879
880         /*
881          * Normally we will only go through one pass of this loop,
882          * unless we get unlucky and it turns out the group we selected
883          * had its last inode grabbed by someone else.
884          */
885         for (i = 0; i < ngroups; i++, ino = 0) {
886                 err = -EIO;
887
888                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
889                 if (!gdp)
890                         goto out;
891
892                 /*
893                  * Check free inodes count before loading bitmap.
894                  */
895                 if (ext4_free_inodes_count(sb, gdp) == 0) {
896                         if (++group == ngroups)
897                                 group = 0;
898                         continue;
899                 }
900
901                 grp = ext4_get_group_info(sb, group);
902                 /* Skip groups with already-known suspicious inode tables */
903                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
904                         if (++group == ngroups)
905                                 group = 0;
906                         continue;
907                 }
908
909                 brelse(inode_bitmap_bh);
910                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
911                 /* Skip groups with suspicious inode tables */
912                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
913                     IS_ERR(inode_bitmap_bh)) {
914                         inode_bitmap_bh = NULL;
915                         if (++group == ngroups)
916                                 group = 0;
917                         continue;
918                 }
919
920 repeat_in_this_group:
921                 ino = ext4_find_next_zero_bit((unsigned long *)
922                                               inode_bitmap_bh->b_data,
923                                               EXT4_INODES_PER_GROUP(sb), ino);
924                 if (ino >= EXT4_INODES_PER_GROUP(sb))
925                         goto next_group;
926                 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
927                         ext4_error(sb, "reserved inode found cleared - "
928                                    "inode=%lu", ino + 1);
929                         continue;
930                 }
931                 if ((EXT4_SB(sb)->s_journal == NULL) &&
932                     recently_deleted(sb, group, ino)) {
933                         ino++;
934                         goto next_inode;
935                 }
936                 if (!handle) {
937                         BUG_ON(nblocks <= 0);
938                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
939                                                          handle_type, nblocks,
940                                                          0);
941                         if (IS_ERR(handle)) {
942                                 err = PTR_ERR(handle);
943                                 ext4_std_error(sb, err);
944                                 goto out;
945                         }
946                 }
947                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
948                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
949                 if (err) {
950                         ext4_std_error(sb, err);
951                         goto out;
952                 }
953                 ext4_lock_group(sb, group);
954                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
955                 ext4_unlock_group(sb, group);
956                 ino++;          /* the inode bitmap is zero-based */
957                 if (!ret2)
958                         goto got; /* we grabbed the inode! */
959 next_inode:
960                 if (ino < EXT4_INODES_PER_GROUP(sb))
961                         goto repeat_in_this_group;
962 next_group:
963                 if (++group == ngroups)
964                         group = 0;
965         }
966         err = -ENOSPC;
967         goto out;
968
969 got:
970         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
971         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
972         if (err) {
973                 ext4_std_error(sb, err);
974                 goto out;
975         }
976
977         BUFFER_TRACE(group_desc_bh, "get_write_access");
978         err = ext4_journal_get_write_access(handle, group_desc_bh);
979         if (err) {
980                 ext4_std_error(sb, err);
981                 goto out;
982         }
983
984         /* We may have to initialize the block bitmap if it isn't already */
985         if (ext4_has_group_desc_csum(sb) &&
986             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
987                 struct buffer_head *block_bitmap_bh;
988
989                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
990                 if (IS_ERR(block_bitmap_bh)) {
991                         err = PTR_ERR(block_bitmap_bh);
992                         goto out;
993                 }
994                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
995                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
996                 if (err) {
997                         brelse(block_bitmap_bh);
998                         ext4_std_error(sb, err);
999                         goto out;
1000                 }
1001
1002                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1003                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1004
1005                 /* recheck and clear flag under lock if we still need to */
1006                 ext4_lock_group(sb, group);
1007                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1008                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1009                         ext4_free_group_clusters_set(sb, gdp,
1010                                 ext4_free_clusters_after_init(sb, group, gdp));
1011                         ext4_block_bitmap_csum_set(sb, group, gdp,
1012                                                    block_bitmap_bh);
1013                         ext4_group_desc_csum_set(sb, group, gdp);
1014                 }
1015                 ext4_unlock_group(sb, group);
1016                 brelse(block_bitmap_bh);
1017
1018                 if (err) {
1019                         ext4_std_error(sb, err);
1020                         goto out;
1021                 }
1022         }
1023
1024         /* Update the relevant bg descriptor fields */
1025         if (ext4_has_group_desc_csum(sb)) {
1026                 int free;
1027                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1028
1029                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1030                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1031                 free = EXT4_INODES_PER_GROUP(sb) -
1032                         ext4_itable_unused_count(sb, gdp);
1033                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1034                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1035                         free = 0;
1036                 }
1037                 /*
1038                  * Check the relative inode number against the last used
1039                  * relative inode number in this group. if it is greater
1040                  * we need to update the bg_itable_unused count
1041                  */
1042                 if (ino > free)
1043                         ext4_itable_unused_set(sb, gdp,
1044                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1045                 up_read(&grp->alloc_sem);
1046         } else {
1047                 ext4_lock_group(sb, group);
1048         }
1049
1050         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1051         if (S_ISDIR(mode)) {
1052                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1053                 if (sbi->s_log_groups_per_flex) {
1054                         ext4_group_t f = ext4_flex_group(sbi, group);
1055
1056                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1057                 }
1058         }
1059         if (ext4_has_group_desc_csum(sb)) {
1060                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1061                                            EXT4_INODES_PER_GROUP(sb) / 8);
1062                 ext4_group_desc_csum_set(sb, group, gdp);
1063         }
1064         ext4_unlock_group(sb, group);
1065
1066         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1067         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1068         if (err) {
1069                 ext4_std_error(sb, err);
1070                 goto out;
1071         }
1072
1073         percpu_counter_dec(&sbi->s_freeinodes_counter);
1074         if (S_ISDIR(mode))
1075                 percpu_counter_inc(&sbi->s_dirs_counter);
1076
1077         if (sbi->s_log_groups_per_flex) {
1078                 flex_group = ext4_flex_group(sbi, group);
1079                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1080         }
1081
1082         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1083         /* This is the optimal IO size (for stat), not the fs block size */
1084         inode->i_blocks = 0;
1085         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1086                                                        current_time(inode);
1087
1088         memset(ei->i_data, 0, sizeof(ei->i_data));
1089         ei->i_dir_start_lookup = 0;
1090         ei->i_disksize = 0;
1091
1092         /* Don't inherit extent flag from directory, amongst others. */
1093         ei->i_flags =
1094                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1095         ei->i_flags |= i_flags;
1096         ei->i_file_acl = 0;
1097         ei->i_dtime = 0;
1098         ei->i_block_group = group;
1099         ei->i_last_alloc_group = ~0;
1100
1101         ext4_set_inode_flags(inode);
1102         if (IS_DIRSYNC(inode))
1103                 ext4_handle_sync(handle);
1104         if (insert_inode_locked(inode) < 0) {
1105                 /*
1106                  * Likely a bitmap corruption causing inode to be allocated
1107                  * twice.
1108                  */
1109                 err = -EIO;
1110                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1111                            inode->i_ino);
1112                 goto out;
1113         }
1114         spin_lock(&sbi->s_next_gen_lock);
1115         inode->i_generation = sbi->s_next_generation++;
1116         spin_unlock(&sbi->s_next_gen_lock);
1117
1118         /* Precompute checksum seed for inode metadata */
1119         if (ext4_has_metadata_csum(sb)) {
1120                 __u32 csum;
1121                 __le32 inum = cpu_to_le32(inode->i_ino);
1122                 __le32 gen = cpu_to_le32(inode->i_generation);
1123                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1124                                    sizeof(inum));
1125                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1126                                               sizeof(gen));
1127         }
1128
1129         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1130         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1131
1132         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1133         ei->i_inline_off = 0;
1134         if (ext4_has_feature_inline_data(sb))
1135                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1136         ret = inode;
1137         err = dquot_alloc_inode(inode);
1138         if (err)
1139                 goto fail_drop;
1140
1141         /*
1142          * Since the encryption xattr will always be unique, create it first so
1143          * that it's less likely to end up in an external xattr block and
1144          * prevent its deduplication.
1145          */
1146         if (encrypt) {
1147                 err = fscrypt_inherit_context(dir, inode, handle, true);
1148                 if (err)
1149                         goto fail_free_drop;
1150         }
1151
1152         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1153                 err = ext4_init_acl(handle, inode, dir);
1154                 if (err)
1155                         goto fail_free_drop;
1156
1157                 err = ext4_init_security(handle, inode, dir, qstr);
1158                 if (err)
1159                         goto fail_free_drop;
1160         }
1161
1162         if (ext4_has_feature_extents(sb)) {
1163                 /* set extent flag only for directory, file and normal symlink*/
1164                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1165                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1166                         ext4_ext_tree_init(handle, inode);
1167                 }
1168         }
1169
1170         if (ext4_handle_valid(handle)) {
1171                 ei->i_sync_tid = handle->h_transaction->t_tid;
1172                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1173         }
1174
1175         err = ext4_mark_inode_dirty(handle, inode);
1176         if (err) {
1177                 ext4_std_error(sb, err);
1178                 goto fail_free_drop;
1179         }
1180
1181         ext4_debug("allocating inode %lu\n", inode->i_ino);
1182         trace_ext4_allocate_inode(inode, dir, mode);
1183         brelse(inode_bitmap_bh);
1184         return ret;
1185
1186 fail_free_drop:
1187         dquot_free_inode(inode);
1188 fail_drop:
1189         clear_nlink(inode);
1190         unlock_new_inode(inode);
1191 out:
1192         dquot_drop(inode);
1193         inode->i_flags |= S_NOQUOTA;
1194         iput(inode);
1195         brelse(inode_bitmap_bh);
1196         return ERR_PTR(err);
1197 }
1198
1199 /* Verify that we are loading a valid orphan from disk */
1200 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1201 {
1202         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1203         ext4_group_t block_group;
1204         int bit;
1205         struct buffer_head *bitmap_bh = NULL;
1206         struct inode *inode = NULL;
1207         int err = -EFSCORRUPTED;
1208
1209         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1210                 goto bad_orphan;
1211
1212         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1213         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1214         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1215         if (IS_ERR(bitmap_bh)) {
1216                 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1217                            ino, PTR_ERR(bitmap_bh));
1218                 return (struct inode *) bitmap_bh;
1219         }
1220
1221         /* Having the inode bit set should be a 100% indicator that this
1222          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1223          * inodes that were being truncated, so we can't check i_nlink==0.
1224          */
1225         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1226                 goto bad_orphan;
1227
1228         inode = ext4_iget(sb, ino);
1229         if (IS_ERR(inode)) {
1230                 err = PTR_ERR(inode);
1231                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1232                            ino, err);
1233                 return inode;
1234         }
1235
1236         /*
1237          * If the orphans has i_nlinks > 0 then it should be able to
1238          * be truncated, otherwise it won't be removed from the orphan
1239          * list during processing and an infinite loop will result.
1240          * Similarly, it must not be a bad inode.
1241          */
1242         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1243             is_bad_inode(inode))
1244                 goto bad_orphan;
1245
1246         if (NEXT_ORPHAN(inode) > max_ino)
1247                 goto bad_orphan;
1248         brelse(bitmap_bh);
1249         return inode;
1250
1251 bad_orphan:
1252         ext4_error(sb, "bad orphan inode %lu", ino);
1253         if (bitmap_bh)
1254                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1255                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1256                        ext4_test_bit(bit, bitmap_bh->b_data));
1257         if (inode) {
1258                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1259                        is_bad_inode(inode));
1260                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1261                        NEXT_ORPHAN(inode));
1262                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1263                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1264                 /* Avoid freeing blocks if we got a bad deleted inode */
1265                 if (inode->i_nlink == 0)
1266                         inode->i_blocks = 0;
1267                 iput(inode);
1268         }
1269         brelse(bitmap_bh);
1270         return ERR_PTR(err);
1271 }
1272
1273 unsigned long ext4_count_free_inodes(struct super_block *sb)
1274 {
1275         unsigned long desc_count;
1276         struct ext4_group_desc *gdp;
1277         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1278 #ifdef EXT4FS_DEBUG
1279         struct ext4_super_block *es;
1280         unsigned long bitmap_count, x;
1281         struct buffer_head *bitmap_bh = NULL;
1282
1283         es = EXT4_SB(sb)->s_es;
1284         desc_count = 0;
1285         bitmap_count = 0;
1286         gdp = NULL;
1287         for (i = 0; i < ngroups; i++) {
1288                 gdp = ext4_get_group_desc(sb, i, NULL);
1289                 if (!gdp)
1290                         continue;
1291                 desc_count += ext4_free_inodes_count(sb, gdp);
1292                 brelse(bitmap_bh);
1293                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1294                 if (IS_ERR(bitmap_bh)) {
1295                         bitmap_bh = NULL;
1296                         continue;
1297                 }
1298
1299                 x = ext4_count_free(bitmap_bh->b_data,
1300                                     EXT4_INODES_PER_GROUP(sb) / 8);
1301                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1302                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1303                 bitmap_count += x;
1304         }
1305         brelse(bitmap_bh);
1306         printk(KERN_DEBUG "ext4_count_free_inodes: "
1307                "stored = %u, computed = %lu, %lu\n",
1308                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1309         return desc_count;
1310 #else
1311         desc_count = 0;
1312         for (i = 0; i < ngroups; i++) {
1313                 gdp = ext4_get_group_desc(sb, i, NULL);
1314                 if (!gdp)
1315                         continue;
1316                 desc_count += ext4_free_inodes_count(sb, gdp);
1317                 cond_resched();
1318         }
1319         return desc_count;
1320 #endif
1321 }
1322
1323 /* Called at mount-time, super-block is locked */
1324 unsigned long ext4_count_dirs(struct super_block * sb)
1325 {
1326         unsigned long count = 0;
1327         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1328
1329         for (i = 0; i < ngroups; i++) {
1330                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1331                 if (!gdp)
1332                         continue;
1333                 count += ext4_used_dirs_count(sb, gdp);
1334         }
1335         return count;
1336 }
1337
1338 /*
1339  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1340  * inode table. Must be called without any spinlock held. The only place
1341  * where it is called from on active part of filesystem is ext4lazyinit
1342  * thread, so we do not need any special locks, however we have to prevent
1343  * inode allocation from the current group, so we take alloc_sem lock, to
1344  * block ext4_new_inode() until we are finished.
1345  */
1346 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1347                                  int barrier)
1348 {
1349         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1350         struct ext4_sb_info *sbi = EXT4_SB(sb);
1351         struct ext4_group_desc *gdp = NULL;
1352         struct buffer_head *group_desc_bh;
1353         handle_t *handle;
1354         ext4_fsblk_t blk;
1355         int num, ret = 0, used_blks = 0;
1356
1357         /* This should not happen, but just to be sure check this */
1358         if (sb->s_flags & MS_RDONLY) {
1359                 ret = 1;
1360                 goto out;
1361         }
1362
1363         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1364         if (!gdp)
1365                 goto out;
1366
1367         /*
1368          * We do not need to lock this, because we are the only one
1369          * handling this flag.
1370          */
1371         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1372                 goto out;
1373
1374         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1375         if (IS_ERR(handle)) {
1376                 ret = PTR_ERR(handle);
1377                 goto out;
1378         }
1379
1380         down_write(&grp->alloc_sem);
1381         /*
1382          * If inode bitmap was already initialized there may be some
1383          * used inodes so we need to skip blocks with used inodes in
1384          * inode table.
1385          */
1386         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1387                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1388                             ext4_itable_unused_count(sb, gdp)),
1389                             sbi->s_inodes_per_block);
1390
1391         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1392                 ext4_error(sb, "Something is wrong with group %u: "
1393                            "used itable blocks: %d; "
1394                            "itable unused count: %u",
1395                            group, used_blks,
1396                            ext4_itable_unused_count(sb, gdp));
1397                 ret = 1;
1398                 goto err_out;
1399         }
1400
1401         blk = ext4_inode_table(sb, gdp) + used_blks;
1402         num = sbi->s_itb_per_group - used_blks;
1403
1404         BUFFER_TRACE(group_desc_bh, "get_write_access");
1405         ret = ext4_journal_get_write_access(handle,
1406                                             group_desc_bh);
1407         if (ret)
1408                 goto err_out;
1409
1410         /*
1411          * Skip zeroout if the inode table is full. But we set the ZEROED
1412          * flag anyway, because obviously, when it is full it does not need
1413          * further zeroing.
1414          */
1415         if (unlikely(num == 0))
1416                 goto skip_zeroout;
1417
1418         ext4_debug("going to zero out inode table in group %d\n",
1419                    group);
1420         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1421         if (ret < 0)
1422                 goto err_out;
1423         if (barrier)
1424                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1425
1426 skip_zeroout:
1427         ext4_lock_group(sb, group);
1428         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1429         ext4_group_desc_csum_set(sb, group, gdp);
1430         ext4_unlock_group(sb, group);
1431
1432         BUFFER_TRACE(group_desc_bh,
1433                      "call ext4_handle_dirty_metadata");
1434         ret = ext4_handle_dirty_metadata(handle, NULL,
1435                                          group_desc_bh);
1436
1437 err_out:
1438         up_write(&grp->alloc_sem);
1439         ext4_journal_stop(handle);
1440 out:
1441         return ret;
1442 }