Loading...
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/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
24#include <linux/blkdev.h>
25#include <asm/byteorder.h>
26
27#include "ext4.h"
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31
32#include <trace/events/ext4.h>
33
34/*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38/*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
48/*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
53void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54{
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65}
66
67/* Initializes an uninitialized inode bitmap */
68static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
72{
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74
75 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
80 ext4_error(sb, "Checksum bad for group %u", block_group);
81 ext4_free_blks_set(sb, gdp, 0);
82 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
84 memset(bh->b_data, 0xff, sb->s_blocksize);
85 return 0;
86 }
87
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
90 bh->b_data);
91
92 return EXT4_INODES_PER_GROUP(sb);
93}
94
95/*
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
98 *
99 * Return buffer_head of bitmap on success or NULL.
100 */
101static struct buffer_head *
102ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
103{
104 struct ext4_group_desc *desc;
105 struct buffer_head *bh = NULL;
106 ext4_fsblk_t bitmap_blk;
107
108 desc = ext4_get_group_desc(sb, block_group, NULL);
109 if (!desc)
110 return NULL;
111
112 bitmap_blk = ext4_inode_bitmap(sb, desc);
113 bh = sb_getblk(sb, bitmap_blk);
114 if (unlikely(!bh)) {
115 ext4_error(sb, "Cannot read inode bitmap - "
116 "block_group = %u, inode_bitmap = %llu",
117 block_group, bitmap_blk);
118 return NULL;
119 }
120 if (bitmap_uptodate(bh))
121 return bh;
122
123 lock_buffer(bh);
124 if (bitmap_uptodate(bh)) {
125 unlock_buffer(bh);
126 return bh;
127 }
128
129 ext4_lock_group(sb, block_group);
130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
132 set_bitmap_uptodate(bh);
133 set_buffer_uptodate(bh);
134 ext4_unlock_group(sb, block_group);
135 unlock_buffer(bh);
136 return bh;
137 }
138 ext4_unlock_group(sb, block_group);
139
140 if (buffer_uptodate(bh)) {
141 /*
142 * if not uninit if bh is uptodate,
143 * bitmap is also uptodate
144 */
145 set_bitmap_uptodate(bh);
146 unlock_buffer(bh);
147 return bh;
148 }
149 /*
150 * submit the buffer_head for read. We can
151 * safely mark the bitmap as uptodate now.
152 * We do it here so the bitmap uptodate bit
153 * get set with buffer lock held.
154 */
155 trace_ext4_load_inode_bitmap(sb, block_group);
156 set_bitmap_uptodate(bh);
157 if (bh_submit_read(bh) < 0) {
158 put_bh(bh);
159 ext4_error(sb, "Cannot read inode bitmap - "
160 "block_group = %u, inode_bitmap = %llu",
161 block_group, bitmap_blk);
162 return NULL;
163 }
164 return bh;
165}
166
167/*
168 * NOTE! When we get the inode, we're the only people
169 * that have access to it, and as such there are no
170 * race conditions we have to worry about. The inode
171 * is not on the hash-lists, and it cannot be reached
172 * through the filesystem because the directory entry
173 * has been deleted earlier.
174 *
175 * HOWEVER: we must make sure that we get no aliases,
176 * which means that we have to call "clear_inode()"
177 * _before_ we mark the inode not in use in the inode
178 * bitmaps. Otherwise a newly created file might use
179 * the same inode number (not actually the same pointer
180 * though), and then we'd have two inodes sharing the
181 * same inode number and space on the harddisk.
182 */
183void ext4_free_inode(handle_t *handle, struct inode *inode)
184{
185 struct super_block *sb = inode->i_sb;
186 int is_directory;
187 unsigned long ino;
188 struct buffer_head *bitmap_bh = NULL;
189 struct buffer_head *bh2;
190 ext4_group_t block_group;
191 unsigned long bit;
192 struct ext4_group_desc *gdp;
193 struct ext4_super_block *es;
194 struct ext4_sb_info *sbi;
195 int fatal = 0, err, count, cleared;
196
197 if (atomic_read(&inode->i_count) > 1) {
198 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
199 atomic_read(&inode->i_count));
200 return;
201 }
202 if (inode->i_nlink) {
203 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
204 inode->i_nlink);
205 return;
206 }
207 if (!sb) {
208 printk(KERN_ERR "ext4_free_inode: inode on "
209 "nonexistent device\n");
210 return;
211 }
212 sbi = EXT4_SB(sb);
213
214 ino = inode->i_ino;
215 ext4_debug("freeing inode %lu\n", ino);
216 trace_ext4_free_inode(inode);
217
218 /*
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
221 */
222 dquot_initialize(inode);
223 ext4_xattr_delete_inode(handle, inode);
224 dquot_free_inode(inode);
225 dquot_drop(inode);
226
227 is_directory = S_ISDIR(inode->i_mode);
228
229 /* Do this BEFORE marking the inode not in use or returning an error */
230 ext4_clear_inode(inode);
231
232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
234 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
235 goto error_return;
236 }
237 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
238 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
239 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
240 if (!bitmap_bh)
241 goto error_return;
242
243 BUFFER_TRACE(bitmap_bh, "get_write_access");
244 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
245 if (fatal)
246 goto error_return;
247
248 fatal = -ESRCH;
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
250 if (gdp) {
251 BUFFER_TRACE(bh2, "get_write_access");
252 fatal = ext4_journal_get_write_access(handle, bh2);
253 }
254 ext4_lock_group(sb, block_group);
255 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
256 if (fatal || !cleared) {
257 ext4_unlock_group(sb, block_group);
258 goto out;
259 }
260
261 count = ext4_free_inodes_count(sb, gdp) + 1;
262 ext4_free_inodes_set(sb, gdp, count);
263 if (is_directory) {
264 count = ext4_used_dirs_count(sb, gdp) - 1;
265 ext4_used_dirs_set(sb, gdp, count);
266 percpu_counter_dec(&sbi->s_dirs_counter);
267 }
268 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
269 ext4_unlock_group(sb, block_group);
270
271 percpu_counter_inc(&sbi->s_freeinodes_counter);
272 if (sbi->s_log_groups_per_flex) {
273 ext4_group_t f = ext4_flex_group(sbi, block_group);
274
275 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
276 if (is_directory)
277 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
278 }
279 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
280 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
281out:
282 if (cleared) {
283 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
284 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
285 if (!fatal)
286 fatal = err;
287 ext4_mark_super_dirty(sb);
288 } else
289 ext4_error(sb, "bit already cleared for inode %lu", ino);
290
291error_return:
292 brelse(bitmap_bh);
293 ext4_std_error(sb, fatal);
294}
295
296/*
297 * There are two policies for allocating an inode. If the new inode is
298 * a directory, then a forward search is made for a block group with both
299 * free space and a low directory-to-inode ratio; if that fails, then of
300 * the groups with above-average free space, that group with the fewest
301 * directories already is chosen.
302 *
303 * For other inodes, search forward from the parent directory\'s block
304 * group to find a free inode.
305 */
306static int find_group_dir(struct super_block *sb, struct inode *parent,
307 ext4_group_t *best_group)
308{
309 ext4_group_t ngroups = ext4_get_groups_count(sb);
310 unsigned int freei, avefreei;
311 struct ext4_group_desc *desc, *best_desc = NULL;
312 ext4_group_t group;
313 int ret = -1;
314
315 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
316 avefreei = freei / ngroups;
317
318 for (group = 0; group < ngroups; group++) {
319 desc = ext4_get_group_desc(sb, group, NULL);
320 if (!desc || !ext4_free_inodes_count(sb, desc))
321 continue;
322 if (ext4_free_inodes_count(sb, desc) < avefreei)
323 continue;
324 if (!best_desc ||
325 (ext4_free_blks_count(sb, desc) >
326 ext4_free_blks_count(sb, best_desc))) {
327 *best_group = group;
328 best_desc = desc;
329 ret = 0;
330 }
331 }
332 return ret;
333}
334
335#define free_block_ratio 10
336
337static int find_group_flex(struct super_block *sb, struct inode *parent,
338 ext4_group_t *best_group)
339{
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
341 struct ext4_group_desc *desc;
342 struct flex_groups *flex_group = sbi->s_flex_groups;
343 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
344 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
345 ext4_group_t ngroups = ext4_get_groups_count(sb);
346 int flex_size = ext4_flex_bg_size(sbi);
347 ext4_group_t best_flex = parent_fbg_group;
348 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
349 int flexbg_free_blocks;
350 int flex_freeb_ratio;
351 ext4_group_t n_fbg_groups;
352 ext4_group_t i;
353
354 n_fbg_groups = (ngroups + flex_size - 1) >>
355 sbi->s_log_groups_per_flex;
356
357find_close_to_parent:
358 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
359 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
360 if (atomic_read(&flex_group[best_flex].free_inodes) &&
361 flex_freeb_ratio > free_block_ratio)
362 goto found_flexbg;
363
364 if (best_flex && best_flex == parent_fbg_group) {
365 best_flex--;
366 goto find_close_to_parent;
367 }
368
369 for (i = 0; i < n_fbg_groups; i++) {
370 if (i == parent_fbg_group || i == parent_fbg_group - 1)
371 continue;
372
373 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
374 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
375
376 if (flex_freeb_ratio > free_block_ratio &&
377 (atomic_read(&flex_group[i].free_inodes))) {
378 best_flex = i;
379 goto found_flexbg;
380 }
381
382 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
383 ((atomic_read(&flex_group[i].free_blocks) >
384 atomic_read(&flex_group[best_flex].free_blocks)) &&
385 atomic_read(&flex_group[i].free_inodes)))
386 best_flex = i;
387 }
388
389 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
390 !atomic_read(&flex_group[best_flex].free_blocks))
391 return -1;
392
393found_flexbg:
394 for (i = best_flex * flex_size; i < ngroups &&
395 i < (best_flex + 1) * flex_size; i++) {
396 desc = ext4_get_group_desc(sb, i, NULL);
397 if (ext4_free_inodes_count(sb, desc)) {
398 *best_group = i;
399 goto out;
400 }
401 }
402
403 return -1;
404out:
405 return 0;
406}
407
408struct orlov_stats {
409 __u32 free_inodes;
410 __u32 free_blocks;
411 __u32 used_dirs;
412};
413
414/*
415 * Helper function for Orlov's allocator; returns critical information
416 * for a particular block group or flex_bg. If flex_size is 1, then g
417 * is a block group number; otherwise it is flex_bg number.
418 */
419static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
420 int flex_size, struct orlov_stats *stats)
421{
422 struct ext4_group_desc *desc;
423 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
424
425 if (flex_size > 1) {
426 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
427 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
428 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
429 return;
430 }
431
432 desc = ext4_get_group_desc(sb, g, NULL);
433 if (desc) {
434 stats->free_inodes = ext4_free_inodes_count(sb, desc);
435 stats->free_blocks = ext4_free_blks_count(sb, desc);
436 stats->used_dirs = ext4_used_dirs_count(sb, desc);
437 } else {
438 stats->free_inodes = 0;
439 stats->free_blocks = 0;
440 stats->used_dirs = 0;
441 }
442}
443
444/*
445 * Orlov's allocator for directories.
446 *
447 * We always try to spread first-level directories.
448 *
449 * If there are blockgroups with both free inodes and free blocks counts
450 * not worse than average we return one with smallest directory count.
451 * Otherwise we simply return a random group.
452 *
453 * For the rest rules look so:
454 *
455 * It's OK to put directory into a group unless
456 * it has too many directories already (max_dirs) or
457 * it has too few free inodes left (min_inodes) or
458 * it has too few free blocks left (min_blocks) or
459 * Parent's group is preferred, if it doesn't satisfy these
460 * conditions we search cyclically through the rest. If none
461 * of the groups look good we just look for a group with more
462 * free inodes than average (starting at parent's group).
463 */
464
465static int find_group_orlov(struct super_block *sb, struct inode *parent,
466 ext4_group_t *group, int mode,
467 const struct qstr *qstr)
468{
469 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
470 struct ext4_sb_info *sbi = EXT4_SB(sb);
471 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
472 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
473 unsigned int freei, avefreei;
474 ext4_fsblk_t freeb, avefreeb;
475 unsigned int ndirs;
476 int max_dirs, min_inodes;
477 ext4_grpblk_t min_blocks;
478 ext4_group_t i, grp, g, ngroups;
479 struct ext4_group_desc *desc;
480 struct orlov_stats stats;
481 int flex_size = ext4_flex_bg_size(sbi);
482 struct dx_hash_info hinfo;
483
484 ngroups = real_ngroups;
485 if (flex_size > 1) {
486 ngroups = (real_ngroups + flex_size - 1) >>
487 sbi->s_log_groups_per_flex;
488 parent_group >>= sbi->s_log_groups_per_flex;
489 }
490
491 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
492 avefreei = freei / ngroups;
493 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
494 avefreeb = freeb;
495 do_div(avefreeb, ngroups);
496 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
497
498 if (S_ISDIR(mode) &&
499 ((parent == sb->s_root->d_inode) ||
500 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
501 int best_ndir = inodes_per_group;
502 int ret = -1;
503
504 if (qstr) {
505 hinfo.hash_version = DX_HASH_HALF_MD4;
506 hinfo.seed = sbi->s_hash_seed;
507 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
508 grp = hinfo.hash;
509 } else
510 get_random_bytes(&grp, sizeof(grp));
511 parent_group = (unsigned)grp % ngroups;
512 for (i = 0; i < ngroups; i++) {
513 g = (parent_group + i) % ngroups;
514 get_orlov_stats(sb, g, flex_size, &stats);
515 if (!stats.free_inodes)
516 continue;
517 if (stats.used_dirs >= best_ndir)
518 continue;
519 if (stats.free_inodes < avefreei)
520 continue;
521 if (stats.free_blocks < avefreeb)
522 continue;
523 grp = g;
524 ret = 0;
525 best_ndir = stats.used_dirs;
526 }
527 if (ret)
528 goto fallback;
529 found_flex_bg:
530 if (flex_size == 1) {
531 *group = grp;
532 return 0;
533 }
534
535 /*
536 * We pack inodes at the beginning of the flexgroup's
537 * inode tables. Block allocation decisions will do
538 * something similar, although regular files will
539 * start at 2nd block group of the flexgroup. See
540 * ext4_ext_find_goal() and ext4_find_near().
541 */
542 grp *= flex_size;
543 for (i = 0; i < flex_size; i++) {
544 if (grp+i >= real_ngroups)
545 break;
546 desc = ext4_get_group_desc(sb, grp+i, NULL);
547 if (desc && ext4_free_inodes_count(sb, desc)) {
548 *group = grp+i;
549 return 0;
550 }
551 }
552 goto fallback;
553 }
554
555 max_dirs = ndirs / ngroups + inodes_per_group / 16;
556 min_inodes = avefreei - inodes_per_group*flex_size / 4;
557 if (min_inodes < 1)
558 min_inodes = 1;
559 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
560
561 /*
562 * Start looking in the flex group where we last allocated an
563 * inode for this parent directory
564 */
565 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
566 parent_group = EXT4_I(parent)->i_last_alloc_group;
567 if (flex_size > 1)
568 parent_group >>= sbi->s_log_groups_per_flex;
569 }
570
571 for (i = 0; i < ngroups; i++) {
572 grp = (parent_group + i) % ngroups;
573 get_orlov_stats(sb, grp, flex_size, &stats);
574 if (stats.used_dirs >= max_dirs)
575 continue;
576 if (stats.free_inodes < min_inodes)
577 continue;
578 if (stats.free_blocks < min_blocks)
579 continue;
580 goto found_flex_bg;
581 }
582
583fallback:
584 ngroups = real_ngroups;
585 avefreei = freei / ngroups;
586fallback_retry:
587 parent_group = EXT4_I(parent)->i_block_group;
588 for (i = 0; i < ngroups; i++) {
589 grp = (parent_group + i) % ngroups;
590 desc = ext4_get_group_desc(sb, grp, NULL);
591 if (desc && ext4_free_inodes_count(sb, desc) &&
592 ext4_free_inodes_count(sb, desc) >= avefreei) {
593 *group = grp;
594 return 0;
595 }
596 }
597
598 if (avefreei) {
599 /*
600 * The free-inodes counter is approximate, and for really small
601 * filesystems the above test can fail to find any blockgroups
602 */
603 avefreei = 0;
604 goto fallback_retry;
605 }
606
607 return -1;
608}
609
610static int find_group_other(struct super_block *sb, struct inode *parent,
611 ext4_group_t *group, int mode)
612{
613 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
614 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
615 struct ext4_group_desc *desc;
616 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
617
618 /*
619 * Try to place the inode is the same flex group as its
620 * parent. If we can't find space, use the Orlov algorithm to
621 * find another flex group, and store that information in the
622 * parent directory's inode information so that use that flex
623 * group for future allocations.
624 */
625 if (flex_size > 1) {
626 int retry = 0;
627
628 try_again:
629 parent_group &= ~(flex_size-1);
630 last = parent_group + flex_size;
631 if (last > ngroups)
632 last = ngroups;
633 for (i = parent_group; i < last; i++) {
634 desc = ext4_get_group_desc(sb, i, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc)) {
636 *group = i;
637 return 0;
638 }
639 }
640 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
641 retry = 1;
642 parent_group = EXT4_I(parent)->i_last_alloc_group;
643 goto try_again;
644 }
645 /*
646 * If this didn't work, use the Orlov search algorithm
647 * to find a new flex group; we pass in the mode to
648 * avoid the topdir algorithms.
649 */
650 *group = parent_group + flex_size;
651 if (*group > ngroups)
652 *group = 0;
653 return find_group_orlov(sb, parent, group, mode, NULL);
654 }
655
656 /*
657 * Try to place the inode in its parent directory
658 */
659 *group = parent_group;
660 desc = ext4_get_group_desc(sb, *group, NULL);
661 if (desc && ext4_free_inodes_count(sb, desc) &&
662 ext4_free_blks_count(sb, desc))
663 return 0;
664
665 /*
666 * We're going to place this inode in a different blockgroup from its
667 * parent. We want to cause files in a common directory to all land in
668 * the same blockgroup. But we want files which are in a different
669 * directory which shares a blockgroup with our parent to land in a
670 * different blockgroup.
671 *
672 * So add our directory's i_ino into the starting point for the hash.
673 */
674 *group = (*group + parent->i_ino) % ngroups;
675
676 /*
677 * Use a quadratic hash to find a group with a free inode and some free
678 * blocks.
679 */
680 for (i = 1; i < ngroups; i <<= 1) {
681 *group += i;
682 if (*group >= ngroups)
683 *group -= ngroups;
684 desc = ext4_get_group_desc(sb, *group, NULL);
685 if (desc && ext4_free_inodes_count(sb, desc) &&
686 ext4_free_blks_count(sb, desc))
687 return 0;
688 }
689
690 /*
691 * That failed: try linear search for a free inode, even if that group
692 * has no free blocks.
693 */
694 *group = parent_group;
695 for (i = 0; i < ngroups; i++) {
696 if (++*group >= ngroups)
697 *group = 0;
698 desc = ext4_get_group_desc(sb, *group, NULL);
699 if (desc && ext4_free_inodes_count(sb, desc))
700 return 0;
701 }
702
703 return -1;
704}
705
706/*
707 * claim the inode from the inode bitmap. If the group
708 * is uninit we need to take the groups's ext4_group_lock
709 * and clear the uninit flag. The inode bitmap update
710 * and group desc uninit flag clear should be done
711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
712 * doesn't race with the ext4_claim_inode
713 */
714static int ext4_claim_inode(struct super_block *sb,
715 struct buffer_head *inode_bitmap_bh,
716 unsigned long ino, ext4_group_t group, int mode)
717{
718 int free = 0, retval = 0, count;
719 struct ext4_sb_info *sbi = EXT4_SB(sb);
720 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
721 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
722
723 /*
724 * We have to be sure that new inode allocation does not race with
725 * inode table initialization, because otherwise we may end up
726 * allocating and writing new inode right before sb_issue_zeroout
727 * takes place and overwriting our new inode with zeroes. So we
728 * take alloc_sem to prevent it.
729 */
730 down_read(&grp->alloc_sem);
731 ext4_lock_group(sb, group);
732 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
733 /* not a free inode */
734 retval = 1;
735 goto err_ret;
736 }
737 ino++;
738 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
739 ino > EXT4_INODES_PER_GROUP(sb)) {
740 ext4_unlock_group(sb, group);
741 up_read(&grp->alloc_sem);
742 ext4_error(sb, "reserved inode or inode > inodes count - "
743 "block_group = %u, inode=%lu", group,
744 ino + group * EXT4_INODES_PER_GROUP(sb));
745 return 1;
746 }
747 /* If we didn't allocate from within the initialized part of the inode
748 * table then we need to initialize up to this inode. */
749 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
750
751 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
752 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
753 /* When marking the block group with
754 * ~EXT4_BG_INODE_UNINIT we don't want to depend
755 * on the value of bg_itable_unused even though
756 * mke2fs could have initialized the same for us.
757 * Instead we calculated the value below
758 */
759
760 free = 0;
761 } else {
762 free = EXT4_INODES_PER_GROUP(sb) -
763 ext4_itable_unused_count(sb, gdp);
764 }
765
766 /*
767 * Check the relative inode number against the last used
768 * relative inode number in this group. if it is greater
769 * we need to update the bg_itable_unused count
770 *
771 */
772 if (ino > free)
773 ext4_itable_unused_set(sb, gdp,
774 (EXT4_INODES_PER_GROUP(sb) - ino));
775 }
776 count = ext4_free_inodes_count(sb, gdp) - 1;
777 ext4_free_inodes_set(sb, gdp, count);
778 if (S_ISDIR(mode)) {
779 count = ext4_used_dirs_count(sb, gdp) + 1;
780 ext4_used_dirs_set(sb, gdp, count);
781 if (sbi->s_log_groups_per_flex) {
782 ext4_group_t f = ext4_flex_group(sbi, group);
783
784 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
785 }
786 }
787 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
788err_ret:
789 ext4_unlock_group(sb, group);
790 up_read(&grp->alloc_sem);
791 return retval;
792}
793
794/*
795 * There are two policies for allocating an inode. If the new inode is
796 * a directory, then a forward search is made for a block group with both
797 * free space and a low directory-to-inode ratio; if that fails, then of
798 * the groups with above-average free space, that group with the fewest
799 * directories already is chosen.
800 *
801 * For other inodes, search forward from the parent directory's block
802 * group to find a free inode.
803 */
804struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
805 const struct qstr *qstr, __u32 goal)
806{
807 struct super_block *sb;
808 struct buffer_head *inode_bitmap_bh = NULL;
809 struct buffer_head *group_desc_bh;
810 ext4_group_t ngroups, group = 0;
811 unsigned long ino = 0;
812 struct inode *inode;
813 struct ext4_group_desc *gdp = NULL;
814 struct ext4_inode_info *ei;
815 struct ext4_sb_info *sbi;
816 int ret2, err = 0;
817 struct inode *ret;
818 ext4_group_t i;
819 int free = 0;
820 static int once = 1;
821 ext4_group_t flex_group;
822
823 /* Cannot create files in a deleted directory */
824 if (!dir || !dir->i_nlink)
825 return ERR_PTR(-EPERM);
826
827 sb = dir->i_sb;
828 ngroups = ext4_get_groups_count(sb);
829 trace_ext4_request_inode(dir, mode);
830 inode = new_inode(sb);
831 if (!inode)
832 return ERR_PTR(-ENOMEM);
833 ei = EXT4_I(inode);
834 sbi = EXT4_SB(sb);
835
836 if (!goal)
837 goal = sbi->s_inode_goal;
838
839 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
840 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
841 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
842 ret2 = 0;
843 goto got_group;
844 }
845
846 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
847 ret2 = find_group_flex(sb, dir, &group);
848 if (ret2 == -1) {
849 ret2 = find_group_other(sb, dir, &group, mode);
850 if (ret2 == 0 && once) {
851 once = 0;
852 printk(KERN_NOTICE "ext4: find_group_flex "
853 "failed, fallback succeeded dir %lu\n",
854 dir->i_ino);
855 }
856 }
857 goto got_group;
858 }
859
860 if (S_ISDIR(mode)) {
861 if (test_opt(sb, OLDALLOC))
862 ret2 = find_group_dir(sb, dir, &group);
863 else
864 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
865 } else
866 ret2 = find_group_other(sb, dir, &group, mode);
867
868got_group:
869 EXT4_I(dir)->i_last_alloc_group = group;
870 err = -ENOSPC;
871 if (ret2 == -1)
872 goto out;
873
874 for (i = 0; i < ngroups; i++, ino = 0) {
875 err = -EIO;
876
877 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
878 if (!gdp)
879 goto fail;
880
881 brelse(inode_bitmap_bh);
882 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
883 if (!inode_bitmap_bh)
884 goto fail;
885
886repeat_in_this_group:
887 ino = ext4_find_next_zero_bit((unsigned long *)
888 inode_bitmap_bh->b_data,
889 EXT4_INODES_PER_GROUP(sb), ino);
890
891 if (ino < EXT4_INODES_PER_GROUP(sb)) {
892
893 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
894 err = ext4_journal_get_write_access(handle,
895 inode_bitmap_bh);
896 if (err)
897 goto fail;
898
899 BUFFER_TRACE(group_desc_bh, "get_write_access");
900 err = ext4_journal_get_write_access(handle,
901 group_desc_bh);
902 if (err)
903 goto fail;
904 if (!ext4_claim_inode(sb, inode_bitmap_bh,
905 ino, group, mode)) {
906 /* we won it */
907 BUFFER_TRACE(inode_bitmap_bh,
908 "call ext4_handle_dirty_metadata");
909 err = ext4_handle_dirty_metadata(handle,
910 NULL,
911 inode_bitmap_bh);
912 if (err)
913 goto fail;
914 /* zero bit is inode number 1*/
915 ino++;
916 goto got;
917 }
918 /* we lost it */
919 ext4_handle_release_buffer(handle, inode_bitmap_bh);
920 ext4_handle_release_buffer(handle, group_desc_bh);
921
922 if (++ino < EXT4_INODES_PER_GROUP(sb))
923 goto repeat_in_this_group;
924 }
925
926 /*
927 * This case is possible in concurrent environment. It is very
928 * rare. We cannot repeat the find_group_xxx() call because
929 * that will simply return the same blockgroup, because the
930 * group descriptor metadata has not yet been updated.
931 * So we just go onto the next blockgroup.
932 */
933 if (++group == ngroups)
934 group = 0;
935 }
936 err = -ENOSPC;
937 goto out;
938
939got:
940 /* We may have to initialize the block bitmap if it isn't already */
941 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
942 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
943 struct buffer_head *block_bitmap_bh;
944
945 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
946 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
947 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
948 if (err) {
949 brelse(block_bitmap_bh);
950 goto fail;
951 }
952
953 free = 0;
954 ext4_lock_group(sb, group);
955 /* recheck and clear flag under lock if we still need to */
956 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
957 free = ext4_free_blocks_after_init(sb, group, gdp);
958 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
959 ext4_free_blks_set(sb, gdp, free);
960 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
961 gdp);
962 }
963 ext4_unlock_group(sb, group);
964
965 /* Don't need to dirty bitmap block if we didn't change it */
966 if (free) {
967 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
968 err = ext4_handle_dirty_metadata(handle,
969 NULL, block_bitmap_bh);
970 }
971
972 brelse(block_bitmap_bh);
973 if (err)
974 goto fail;
975 }
976 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
977 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
978 if (err)
979 goto fail;
980
981 percpu_counter_dec(&sbi->s_freeinodes_counter);
982 if (S_ISDIR(mode))
983 percpu_counter_inc(&sbi->s_dirs_counter);
984 ext4_mark_super_dirty(sb);
985
986 if (sbi->s_log_groups_per_flex) {
987 flex_group = ext4_flex_group(sbi, group);
988 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
989 }
990
991 if (test_opt(sb, GRPID)) {
992 inode->i_mode = mode;
993 inode->i_uid = current_fsuid();
994 inode->i_gid = dir->i_gid;
995 } else
996 inode_init_owner(inode, dir, mode);
997
998 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
999 /* This is the optimal IO size (for stat), not the fs block size */
1000 inode->i_blocks = 0;
1001 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1002 ext4_current_time(inode);
1003
1004 memset(ei->i_data, 0, sizeof(ei->i_data));
1005 ei->i_dir_start_lookup = 0;
1006 ei->i_disksize = 0;
1007
1008 /*
1009 * Don't inherit extent flag from directory, amongst others. We set
1010 * extent flag on newly created directory and file only if -o extent
1011 * mount option is specified
1012 */
1013 ei->i_flags =
1014 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1015 ei->i_file_acl = 0;
1016 ei->i_dtime = 0;
1017 ei->i_block_group = group;
1018 ei->i_last_alloc_group = ~0;
1019
1020 ext4_set_inode_flags(inode);
1021 if (IS_DIRSYNC(inode))
1022 ext4_handle_sync(handle);
1023 if (insert_inode_locked(inode) < 0) {
1024 err = -EINVAL;
1025 goto fail_drop;
1026 }
1027 spin_lock(&sbi->s_next_gen_lock);
1028 inode->i_generation = sbi->s_next_generation++;
1029 spin_unlock(&sbi->s_next_gen_lock);
1030
1031 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1032 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1033
1034 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1035
1036 ret = inode;
1037 dquot_initialize(inode);
1038 err = dquot_alloc_inode(inode);
1039 if (err)
1040 goto fail_drop;
1041
1042 err = ext4_init_acl(handle, inode, dir);
1043 if (err)
1044 goto fail_free_drop;
1045
1046 err = ext4_init_security(handle, inode, dir, qstr);
1047 if (err)
1048 goto fail_free_drop;
1049
1050 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1051 /* set extent flag only for directory, file and normal symlink*/
1052 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1053 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1054 ext4_ext_tree_init(handle, inode);
1055 }
1056 }
1057
1058 if (ext4_handle_valid(handle)) {
1059 ei->i_sync_tid = handle->h_transaction->t_tid;
1060 ei->i_datasync_tid = handle->h_transaction->t_tid;
1061 }
1062
1063 err = ext4_mark_inode_dirty(handle, inode);
1064 if (err) {
1065 ext4_std_error(sb, err);
1066 goto fail_free_drop;
1067 }
1068
1069 ext4_debug("allocating inode %lu\n", inode->i_ino);
1070 trace_ext4_allocate_inode(inode, dir, mode);
1071 goto really_out;
1072fail:
1073 ext4_std_error(sb, err);
1074out:
1075 iput(inode);
1076 ret = ERR_PTR(err);
1077really_out:
1078 brelse(inode_bitmap_bh);
1079 return ret;
1080
1081fail_free_drop:
1082 dquot_free_inode(inode);
1083
1084fail_drop:
1085 dquot_drop(inode);
1086 inode->i_flags |= S_NOQUOTA;
1087 inode->i_nlink = 0;
1088 unlock_new_inode(inode);
1089 iput(inode);
1090 brelse(inode_bitmap_bh);
1091 return ERR_PTR(err);
1092}
1093
1094/* Verify that we are loading a valid orphan from disk */
1095struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1096{
1097 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1098 ext4_group_t block_group;
1099 int bit;
1100 struct buffer_head *bitmap_bh;
1101 struct inode *inode = NULL;
1102 long err = -EIO;
1103
1104 /* Error cases - e2fsck has already cleaned up for us */
1105 if (ino > max_ino) {
1106 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1107 goto error;
1108 }
1109
1110 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1111 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1112 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1113 if (!bitmap_bh) {
1114 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1115 goto error;
1116 }
1117
1118 /* Having the inode bit set should be a 100% indicator that this
1119 * is a valid orphan (no e2fsck run on fs). Orphans also include
1120 * inodes that were being truncated, so we can't check i_nlink==0.
1121 */
1122 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1123 goto bad_orphan;
1124
1125 inode = ext4_iget(sb, ino);
1126 if (IS_ERR(inode))
1127 goto iget_failed;
1128
1129 /*
1130 * If the orphans has i_nlinks > 0 then it should be able to be
1131 * truncated, otherwise it won't be removed from the orphan list
1132 * during processing and an infinite loop will result.
1133 */
1134 if (inode->i_nlink && !ext4_can_truncate(inode))
1135 goto bad_orphan;
1136
1137 if (NEXT_ORPHAN(inode) > max_ino)
1138 goto bad_orphan;
1139 brelse(bitmap_bh);
1140 return inode;
1141
1142iget_failed:
1143 err = PTR_ERR(inode);
1144 inode = NULL;
1145bad_orphan:
1146 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1147 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1148 bit, (unsigned long long)bitmap_bh->b_blocknr,
1149 ext4_test_bit(bit, bitmap_bh->b_data));
1150 printk(KERN_NOTICE "inode=%p\n", inode);
1151 if (inode) {
1152 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1153 is_bad_inode(inode));
1154 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1155 NEXT_ORPHAN(inode));
1156 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1157 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1158 /* Avoid freeing blocks if we got a bad deleted inode */
1159 if (inode->i_nlink == 0)
1160 inode->i_blocks = 0;
1161 iput(inode);
1162 }
1163 brelse(bitmap_bh);
1164error:
1165 return ERR_PTR(err);
1166}
1167
1168unsigned long ext4_count_free_inodes(struct super_block *sb)
1169{
1170 unsigned long desc_count;
1171 struct ext4_group_desc *gdp;
1172 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1173#ifdef EXT4FS_DEBUG
1174 struct ext4_super_block *es;
1175 unsigned long bitmap_count, x;
1176 struct buffer_head *bitmap_bh = NULL;
1177
1178 es = EXT4_SB(sb)->s_es;
1179 desc_count = 0;
1180 bitmap_count = 0;
1181 gdp = NULL;
1182 for (i = 0; i < ngroups; i++) {
1183 gdp = ext4_get_group_desc(sb, i, NULL);
1184 if (!gdp)
1185 continue;
1186 desc_count += ext4_free_inodes_count(sb, gdp);
1187 brelse(bitmap_bh);
1188 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1189 if (!bitmap_bh)
1190 continue;
1191
1192 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1193 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1194 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1195 bitmap_count += x;
1196 }
1197 brelse(bitmap_bh);
1198 printk(KERN_DEBUG "ext4_count_free_inodes: "
1199 "stored = %u, computed = %lu, %lu\n",
1200 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1201 return desc_count;
1202#else
1203 desc_count = 0;
1204 for (i = 0; i < ngroups; i++) {
1205 gdp = ext4_get_group_desc(sb, i, NULL);
1206 if (!gdp)
1207 continue;
1208 desc_count += ext4_free_inodes_count(sb, gdp);
1209 cond_resched();
1210 }
1211 return desc_count;
1212#endif
1213}
1214
1215/* Called at mount-time, super-block is locked */
1216unsigned long ext4_count_dirs(struct super_block * sb)
1217{
1218 unsigned long count = 0;
1219 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1220
1221 for (i = 0; i < ngroups; i++) {
1222 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1223 if (!gdp)
1224 continue;
1225 count += ext4_used_dirs_count(sb, gdp);
1226 }
1227 return count;
1228}
1229
1230/*
1231 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1232 * inode table. Must be called without any spinlock held. The only place
1233 * where it is called from on active part of filesystem is ext4lazyinit
1234 * thread, so we do not need any special locks, however we have to prevent
1235 * inode allocation from the current group, so we take alloc_sem lock, to
1236 * block ext4_claim_inode until we are finished.
1237 */
1238extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1239 int barrier)
1240{
1241 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1242 struct ext4_sb_info *sbi = EXT4_SB(sb);
1243 struct ext4_group_desc *gdp = NULL;
1244 struct buffer_head *group_desc_bh;
1245 handle_t *handle;
1246 ext4_fsblk_t blk;
1247 int num, ret = 0, used_blks = 0;
1248
1249 /* This should not happen, but just to be sure check this */
1250 if (sb->s_flags & MS_RDONLY) {
1251 ret = 1;
1252 goto out;
1253 }
1254
1255 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1256 if (!gdp)
1257 goto out;
1258
1259 /*
1260 * We do not need to lock this, because we are the only one
1261 * handling this flag.
1262 */
1263 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1264 goto out;
1265
1266 handle = ext4_journal_start_sb(sb, 1);
1267 if (IS_ERR(handle)) {
1268 ret = PTR_ERR(handle);
1269 goto out;
1270 }
1271
1272 down_write(&grp->alloc_sem);
1273 /*
1274 * If inode bitmap was already initialized there may be some
1275 * used inodes so we need to skip blocks with used inodes in
1276 * inode table.
1277 */
1278 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1279 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1280 ext4_itable_unused_count(sb, gdp)),
1281 sbi->s_inodes_per_block);
1282
1283 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1284 ext4_error(sb, "Something is wrong with group %u\n"
1285 "Used itable blocks: %d"
1286 "itable unused count: %u\n",
1287 group, used_blks,
1288 ext4_itable_unused_count(sb, gdp));
1289 ret = 1;
1290 goto err_out;
1291 }
1292
1293 blk = ext4_inode_table(sb, gdp) + used_blks;
1294 num = sbi->s_itb_per_group - used_blks;
1295
1296 BUFFER_TRACE(group_desc_bh, "get_write_access");
1297 ret = ext4_journal_get_write_access(handle,
1298 group_desc_bh);
1299 if (ret)
1300 goto err_out;
1301
1302 /*
1303 * Skip zeroout if the inode table is full. But we set the ZEROED
1304 * flag anyway, because obviously, when it is full it does not need
1305 * further zeroing.
1306 */
1307 if (unlikely(num == 0))
1308 goto skip_zeroout;
1309
1310 ext4_debug("going to zero out inode table in group %d\n",
1311 group);
1312 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1313 if (ret < 0)
1314 goto err_out;
1315 if (barrier)
1316 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1317
1318skip_zeroout:
1319 ext4_lock_group(sb, group);
1320 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1321 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1322 ext4_unlock_group(sb, group);
1323
1324 BUFFER_TRACE(group_desc_bh,
1325 "call ext4_handle_dirty_metadata");
1326 ret = ext4_handle_dirty_metadata(handle, NULL,
1327 group_desc_bh);
1328
1329err_out:
1330 up_write(&grp->alloc_sem);
1331 ext4_journal_stop(handle);
1332out:
1333 return ret;
1334}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/ialloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
14 */
15
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
24#include <linux/blkdev.h>
25#include <linux/cred.h>
26
27#include <asm/byteorder.h>
28
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34#include <trace/events/ext4.h>
35
36/*
37 * ialloc.c contains the inodes allocation and deallocation routines
38 */
39
40/*
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
44 *
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
48 */
49
50/*
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
54 */
55void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56{
57 int i;
58
59 if (start_bit >= end_bit)
60 return;
61
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
65 if (i < end_bit)
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67}
68
69void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70{
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
74 }
75 unlock_buffer(bh);
76 put_bh(bh);
77}
78
79static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
83{
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
86 struct ext4_sb_info *sbi = EXT4_SB(sb);
87
88 if (buffer_verified(bh))
89 return 0;
90 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
91 return -EFSCORRUPTED;
92
93 ext4_lock_group(sb, block_group);
94 blk = ext4_inode_bitmap(sb, desc);
95 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
96 EXT4_INODES_PER_GROUP(sb) / 8)) {
97 ext4_unlock_group(sb, block_group);
98 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
99 "inode_bitmap = %llu", block_group, blk);
100 grp = ext4_get_group_info(sb, block_group);
101 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
102 int count;
103 count = ext4_free_inodes_count(sb, desc);
104 percpu_counter_sub(&sbi->s_freeinodes_counter,
105 count);
106 }
107 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
108 return -EFSBADCRC;
109 }
110 set_buffer_verified(bh);
111 ext4_unlock_group(sb, block_group);
112 return 0;
113}
114
115/*
116 * Read the inode allocation bitmap for a given block_group, reading
117 * into the specified slot in the superblock's bitmap cache.
118 *
119 * Return buffer_head of bitmap on success or NULL.
120 */
121static struct buffer_head *
122ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
123{
124 struct ext4_group_desc *desc;
125 struct ext4_sb_info *sbi = EXT4_SB(sb);
126 struct buffer_head *bh = NULL;
127 ext4_fsblk_t bitmap_blk;
128 int err;
129
130 desc = ext4_get_group_desc(sb, block_group, NULL);
131 if (!desc)
132 return ERR_PTR(-EFSCORRUPTED);
133
134 bitmap_blk = ext4_inode_bitmap(sb, desc);
135 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
136 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
137 ext4_error(sb, "Invalid inode bitmap blk %llu in "
138 "block_group %u", bitmap_blk, block_group);
139 return ERR_PTR(-EFSCORRUPTED);
140 }
141 bh = sb_getblk(sb, bitmap_blk);
142 if (unlikely(!bh)) {
143 ext4_error(sb, "Cannot read inode bitmap - "
144 "block_group = %u, inode_bitmap = %llu",
145 block_group, bitmap_blk);
146 return ERR_PTR(-EIO);
147 }
148 if (bitmap_uptodate(bh))
149 goto verify;
150
151 lock_buffer(bh);
152 if (bitmap_uptodate(bh)) {
153 unlock_buffer(bh);
154 goto verify;
155 }
156
157 ext4_lock_group(sb, block_group);
158 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
159 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
160 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
161 sb->s_blocksize * 8, bh->b_data);
162 set_bitmap_uptodate(bh);
163 set_buffer_uptodate(bh);
164 set_buffer_verified(bh);
165 ext4_unlock_group(sb, block_group);
166 unlock_buffer(bh);
167 return bh;
168 }
169 ext4_unlock_group(sb, block_group);
170
171 if (buffer_uptodate(bh)) {
172 /*
173 * if not uninit if bh is uptodate,
174 * bitmap is also uptodate
175 */
176 set_bitmap_uptodate(bh);
177 unlock_buffer(bh);
178 goto verify;
179 }
180 /*
181 * submit the buffer_head for reading
182 */
183 trace_ext4_load_inode_bitmap(sb, block_group);
184 bh->b_end_io = ext4_end_bitmap_read;
185 get_bh(bh);
186 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
187 wait_on_buffer(bh);
188 if (!buffer_uptodate(bh)) {
189 put_bh(bh);
190 ext4_error(sb, "Cannot read inode bitmap - "
191 "block_group = %u, inode_bitmap = %llu",
192 block_group, bitmap_blk);
193 return ERR_PTR(-EIO);
194 }
195
196verify:
197 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
198 if (err)
199 goto out;
200 return bh;
201out:
202 put_bh(bh);
203 return ERR_PTR(err);
204}
205
206/*
207 * NOTE! When we get the inode, we're the only people
208 * that have access to it, and as such there are no
209 * race conditions we have to worry about. The inode
210 * is not on the hash-lists, and it cannot be reached
211 * through the filesystem because the directory entry
212 * has been deleted earlier.
213 *
214 * HOWEVER: we must make sure that we get no aliases,
215 * which means that we have to call "clear_inode()"
216 * _before_ we mark the inode not in use in the inode
217 * bitmaps. Otherwise a newly created file might use
218 * the same inode number (not actually the same pointer
219 * though), and then we'd have two inodes sharing the
220 * same inode number and space on the harddisk.
221 */
222void ext4_free_inode(handle_t *handle, struct inode *inode)
223{
224 struct super_block *sb = inode->i_sb;
225 int is_directory;
226 unsigned long ino;
227 struct buffer_head *bitmap_bh = NULL;
228 struct buffer_head *bh2;
229 ext4_group_t block_group;
230 unsigned long bit;
231 struct ext4_group_desc *gdp;
232 struct ext4_super_block *es;
233 struct ext4_sb_info *sbi;
234 int fatal = 0, err, count, cleared;
235 struct ext4_group_info *grp;
236
237 if (!sb) {
238 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
239 "nonexistent device\n", __func__, __LINE__);
240 return;
241 }
242 if (atomic_read(&inode->i_count) > 1) {
243 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
244 __func__, __LINE__, inode->i_ino,
245 atomic_read(&inode->i_count));
246 return;
247 }
248 if (inode->i_nlink) {
249 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
250 __func__, __LINE__, inode->i_ino, inode->i_nlink);
251 return;
252 }
253 sbi = EXT4_SB(sb);
254
255 ino = inode->i_ino;
256 ext4_debug("freeing inode %lu\n", ino);
257 trace_ext4_free_inode(inode);
258
259 /*
260 * Note: we must free any quota before locking the superblock,
261 * as writing the quota to disk may need the lock as well.
262 */
263 dquot_initialize(inode);
264 dquot_free_inode(inode);
265 dquot_drop(inode);
266
267 is_directory = S_ISDIR(inode->i_mode);
268
269 /* Do this BEFORE marking the inode not in use or returning an error */
270 ext4_clear_inode(inode);
271
272 es = sbi->s_es;
273 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
274 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
275 goto error_return;
276 }
277 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
278 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
279 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
280 /* Don't bother if the inode bitmap is corrupt. */
281 grp = ext4_get_group_info(sb, block_group);
282 if (IS_ERR(bitmap_bh)) {
283 fatal = PTR_ERR(bitmap_bh);
284 bitmap_bh = NULL;
285 goto error_return;
286 }
287 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
288 fatal = -EFSCORRUPTED;
289 goto error_return;
290 }
291
292 BUFFER_TRACE(bitmap_bh, "get_write_access");
293 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
294 if (fatal)
295 goto error_return;
296
297 fatal = -ESRCH;
298 gdp = ext4_get_group_desc(sb, block_group, &bh2);
299 if (gdp) {
300 BUFFER_TRACE(bh2, "get_write_access");
301 fatal = ext4_journal_get_write_access(handle, bh2);
302 }
303 ext4_lock_group(sb, block_group);
304 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
305 if (fatal || !cleared) {
306 ext4_unlock_group(sb, block_group);
307 goto out;
308 }
309
310 count = ext4_free_inodes_count(sb, gdp) + 1;
311 ext4_free_inodes_set(sb, gdp, count);
312 if (is_directory) {
313 count = ext4_used_dirs_count(sb, gdp) - 1;
314 ext4_used_dirs_set(sb, gdp, count);
315 percpu_counter_dec(&sbi->s_dirs_counter);
316 }
317 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
318 EXT4_INODES_PER_GROUP(sb) / 8);
319 ext4_group_desc_csum_set(sb, block_group, gdp);
320 ext4_unlock_group(sb, block_group);
321
322 percpu_counter_inc(&sbi->s_freeinodes_counter);
323 if (sbi->s_log_groups_per_flex) {
324 ext4_group_t f = ext4_flex_group(sbi, block_group);
325
326 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
327 if (is_directory)
328 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
329 }
330 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
331 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
332out:
333 if (cleared) {
334 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
335 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
336 if (!fatal)
337 fatal = err;
338 } else {
339 ext4_error(sb, "bit already cleared for inode %lu", ino);
340 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
341 int count;
342 count = ext4_free_inodes_count(sb, gdp);
343 percpu_counter_sub(&sbi->s_freeinodes_counter,
344 count);
345 }
346 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
347 }
348
349error_return:
350 brelse(bitmap_bh);
351 ext4_std_error(sb, fatal);
352}
353
354struct orlov_stats {
355 __u64 free_clusters;
356 __u32 free_inodes;
357 __u32 used_dirs;
358};
359
360/*
361 * Helper function for Orlov's allocator; returns critical information
362 * for a particular block group or flex_bg. If flex_size is 1, then g
363 * is a block group number; otherwise it is flex_bg number.
364 */
365static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
366 int flex_size, struct orlov_stats *stats)
367{
368 struct ext4_group_desc *desc;
369 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
370
371 if (flex_size > 1) {
372 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
373 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
374 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
375 return;
376 }
377
378 desc = ext4_get_group_desc(sb, g, NULL);
379 if (desc) {
380 stats->free_inodes = ext4_free_inodes_count(sb, desc);
381 stats->free_clusters = ext4_free_group_clusters(sb, desc);
382 stats->used_dirs = ext4_used_dirs_count(sb, desc);
383 } else {
384 stats->free_inodes = 0;
385 stats->free_clusters = 0;
386 stats->used_dirs = 0;
387 }
388}
389
390/*
391 * Orlov's allocator for directories.
392 *
393 * We always try to spread first-level directories.
394 *
395 * If there are blockgroups with both free inodes and free blocks counts
396 * not worse than average we return one with smallest directory count.
397 * Otherwise we simply return a random group.
398 *
399 * For the rest rules look so:
400 *
401 * It's OK to put directory into a group unless
402 * it has too many directories already (max_dirs) or
403 * it has too few free inodes left (min_inodes) or
404 * it has too few free blocks left (min_blocks) or
405 * Parent's group is preferred, if it doesn't satisfy these
406 * conditions we search cyclically through the rest. If none
407 * of the groups look good we just look for a group with more
408 * free inodes than average (starting at parent's group).
409 */
410
411static int find_group_orlov(struct super_block *sb, struct inode *parent,
412 ext4_group_t *group, umode_t mode,
413 const struct qstr *qstr)
414{
415 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
416 struct ext4_sb_info *sbi = EXT4_SB(sb);
417 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
418 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
419 unsigned int freei, avefreei, grp_free;
420 ext4_fsblk_t freeb, avefreec;
421 unsigned int ndirs;
422 int max_dirs, min_inodes;
423 ext4_grpblk_t min_clusters;
424 ext4_group_t i, grp, g, ngroups;
425 struct ext4_group_desc *desc;
426 struct orlov_stats stats;
427 int flex_size = ext4_flex_bg_size(sbi);
428 struct dx_hash_info hinfo;
429
430 ngroups = real_ngroups;
431 if (flex_size > 1) {
432 ngroups = (real_ngroups + flex_size - 1) >>
433 sbi->s_log_groups_per_flex;
434 parent_group >>= sbi->s_log_groups_per_flex;
435 }
436
437 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
438 avefreei = freei / ngroups;
439 freeb = EXT4_C2B(sbi,
440 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
441 avefreec = freeb;
442 do_div(avefreec, ngroups);
443 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
444
445 if (S_ISDIR(mode) &&
446 ((parent == d_inode(sb->s_root)) ||
447 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
448 int best_ndir = inodes_per_group;
449 int ret = -1;
450
451 if (qstr) {
452 hinfo.hash_version = DX_HASH_HALF_MD4;
453 hinfo.seed = sbi->s_hash_seed;
454 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
455 grp = hinfo.hash;
456 } else
457 grp = prandom_u32();
458 parent_group = (unsigned)grp % ngroups;
459 for (i = 0; i < ngroups; i++) {
460 g = (parent_group + i) % ngroups;
461 get_orlov_stats(sb, g, flex_size, &stats);
462 if (!stats.free_inodes)
463 continue;
464 if (stats.used_dirs >= best_ndir)
465 continue;
466 if (stats.free_inodes < avefreei)
467 continue;
468 if (stats.free_clusters < avefreec)
469 continue;
470 grp = g;
471 ret = 0;
472 best_ndir = stats.used_dirs;
473 }
474 if (ret)
475 goto fallback;
476 found_flex_bg:
477 if (flex_size == 1) {
478 *group = grp;
479 return 0;
480 }
481
482 /*
483 * We pack inodes at the beginning of the flexgroup's
484 * inode tables. Block allocation decisions will do
485 * something similar, although regular files will
486 * start at 2nd block group of the flexgroup. See
487 * ext4_ext_find_goal() and ext4_find_near().
488 */
489 grp *= flex_size;
490 for (i = 0; i < flex_size; i++) {
491 if (grp+i >= real_ngroups)
492 break;
493 desc = ext4_get_group_desc(sb, grp+i, NULL);
494 if (desc && ext4_free_inodes_count(sb, desc)) {
495 *group = grp+i;
496 return 0;
497 }
498 }
499 goto fallback;
500 }
501
502 max_dirs = ndirs / ngroups + inodes_per_group / 16;
503 min_inodes = avefreei - inodes_per_group*flex_size / 4;
504 if (min_inodes < 1)
505 min_inodes = 1;
506 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
507
508 /*
509 * Start looking in the flex group where we last allocated an
510 * inode for this parent directory
511 */
512 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
513 parent_group = EXT4_I(parent)->i_last_alloc_group;
514 if (flex_size > 1)
515 parent_group >>= sbi->s_log_groups_per_flex;
516 }
517
518 for (i = 0; i < ngroups; i++) {
519 grp = (parent_group + i) % ngroups;
520 get_orlov_stats(sb, grp, flex_size, &stats);
521 if (stats.used_dirs >= max_dirs)
522 continue;
523 if (stats.free_inodes < min_inodes)
524 continue;
525 if (stats.free_clusters < min_clusters)
526 continue;
527 goto found_flex_bg;
528 }
529
530fallback:
531 ngroups = real_ngroups;
532 avefreei = freei / ngroups;
533fallback_retry:
534 parent_group = EXT4_I(parent)->i_block_group;
535 for (i = 0; i < ngroups; i++) {
536 grp = (parent_group + i) % ngroups;
537 desc = ext4_get_group_desc(sb, grp, NULL);
538 if (desc) {
539 grp_free = ext4_free_inodes_count(sb, desc);
540 if (grp_free && grp_free >= avefreei) {
541 *group = grp;
542 return 0;
543 }
544 }
545 }
546
547 if (avefreei) {
548 /*
549 * The free-inodes counter is approximate, and for really small
550 * filesystems the above test can fail to find any blockgroups
551 */
552 avefreei = 0;
553 goto fallback_retry;
554 }
555
556 return -1;
557}
558
559static int find_group_other(struct super_block *sb, struct inode *parent,
560 ext4_group_t *group, umode_t mode)
561{
562 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
563 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
564 struct ext4_group_desc *desc;
565 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
566
567 /*
568 * Try to place the inode is the same flex group as its
569 * parent. If we can't find space, use the Orlov algorithm to
570 * find another flex group, and store that information in the
571 * parent directory's inode information so that use that flex
572 * group for future allocations.
573 */
574 if (flex_size > 1) {
575 int retry = 0;
576
577 try_again:
578 parent_group &= ~(flex_size-1);
579 last = parent_group + flex_size;
580 if (last > ngroups)
581 last = ngroups;
582 for (i = parent_group; i < last; i++) {
583 desc = ext4_get_group_desc(sb, i, NULL);
584 if (desc && ext4_free_inodes_count(sb, desc)) {
585 *group = i;
586 return 0;
587 }
588 }
589 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
590 retry = 1;
591 parent_group = EXT4_I(parent)->i_last_alloc_group;
592 goto try_again;
593 }
594 /*
595 * If this didn't work, use the Orlov search algorithm
596 * to find a new flex group; we pass in the mode to
597 * avoid the topdir algorithms.
598 */
599 *group = parent_group + flex_size;
600 if (*group > ngroups)
601 *group = 0;
602 return find_group_orlov(sb, parent, group, mode, NULL);
603 }
604
605 /*
606 * Try to place the inode in its parent directory
607 */
608 *group = parent_group;
609 desc = ext4_get_group_desc(sb, *group, NULL);
610 if (desc && ext4_free_inodes_count(sb, desc) &&
611 ext4_free_group_clusters(sb, desc))
612 return 0;
613
614 /*
615 * We're going to place this inode in a different blockgroup from its
616 * parent. We want to cause files in a common directory to all land in
617 * the same blockgroup. But we want files which are in a different
618 * directory which shares a blockgroup with our parent to land in a
619 * different blockgroup.
620 *
621 * So add our directory's i_ino into the starting point for the hash.
622 */
623 *group = (*group + parent->i_ino) % ngroups;
624
625 /*
626 * Use a quadratic hash to find a group with a free inode and some free
627 * blocks.
628 */
629 for (i = 1; i < ngroups; i <<= 1) {
630 *group += i;
631 if (*group >= ngroups)
632 *group -= ngroups;
633 desc = ext4_get_group_desc(sb, *group, NULL);
634 if (desc && ext4_free_inodes_count(sb, desc) &&
635 ext4_free_group_clusters(sb, desc))
636 return 0;
637 }
638
639 /*
640 * That failed: try linear search for a free inode, even if that group
641 * has no free blocks.
642 */
643 *group = parent_group;
644 for (i = 0; i < ngroups; i++) {
645 if (++*group >= ngroups)
646 *group = 0;
647 desc = ext4_get_group_desc(sb, *group, NULL);
648 if (desc && ext4_free_inodes_count(sb, desc))
649 return 0;
650 }
651
652 return -1;
653}
654
655/*
656 * In no journal mode, if an inode has recently been deleted, we want
657 * to avoid reusing it until we're reasonably sure the inode table
658 * block has been written back to disk. (Yes, these values are
659 * somewhat arbitrary...)
660 */
661#define RECENTCY_MIN 5
662#define RECENTCY_DIRTY 300
663
664static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
665{
666 struct ext4_group_desc *gdp;
667 struct ext4_inode *raw_inode;
668 struct buffer_head *bh;
669 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
670 int offset, ret = 0;
671 int recentcy = RECENTCY_MIN;
672 u32 dtime, now;
673
674 gdp = ext4_get_group_desc(sb, group, NULL);
675 if (unlikely(!gdp))
676 return 0;
677
678 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
679 (ino / inodes_per_block));
680 if (!bh || !buffer_uptodate(bh))
681 /*
682 * If the block is not in the buffer cache, then it
683 * must have been written out.
684 */
685 goto out;
686
687 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
688 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
689
690 /* i_dtime is only 32 bits on disk, but we only care about relative
691 * times in the range of a few minutes (i.e. long enough to sync a
692 * recently-deleted inode to disk), so using the low 32 bits of the
693 * clock (a 68 year range) is enough, see time_before32() */
694 dtime = le32_to_cpu(raw_inode->i_dtime);
695 now = ktime_get_real_seconds();
696 if (buffer_dirty(bh))
697 recentcy += RECENTCY_DIRTY;
698
699 if (dtime && time_before32(dtime, now) &&
700 time_before32(now, dtime + recentcy))
701 ret = 1;
702out:
703 brelse(bh);
704 return ret;
705}
706
707static int find_inode_bit(struct super_block *sb, ext4_group_t group,
708 struct buffer_head *bitmap, unsigned long *ino)
709{
710next:
711 *ino = ext4_find_next_zero_bit((unsigned long *)
712 bitmap->b_data,
713 EXT4_INODES_PER_GROUP(sb), *ino);
714 if (*ino >= EXT4_INODES_PER_GROUP(sb))
715 return 0;
716
717 if ((EXT4_SB(sb)->s_journal == NULL) &&
718 recently_deleted(sb, group, *ino)) {
719 *ino = *ino + 1;
720 if (*ino < EXT4_INODES_PER_GROUP(sb))
721 goto next;
722 return 0;
723 }
724
725 return 1;
726}
727
728/*
729 * There are two policies for allocating an inode. If the new inode is
730 * a directory, then a forward search is made for a block group with both
731 * free space and a low directory-to-inode ratio; if that fails, then of
732 * the groups with above-average free space, that group with the fewest
733 * directories already is chosen.
734 *
735 * For other inodes, search forward from the parent directory's block
736 * group to find a free inode.
737 */
738struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
739 umode_t mode, const struct qstr *qstr,
740 __u32 goal, uid_t *owner, __u32 i_flags,
741 int handle_type, unsigned int line_no,
742 int nblocks)
743{
744 struct super_block *sb;
745 struct buffer_head *inode_bitmap_bh = NULL;
746 struct buffer_head *group_desc_bh;
747 ext4_group_t ngroups, group = 0;
748 unsigned long ino = 0;
749 struct inode *inode;
750 struct ext4_group_desc *gdp = NULL;
751 struct ext4_inode_info *ei;
752 struct ext4_sb_info *sbi;
753 int ret2, err;
754 struct inode *ret;
755 ext4_group_t i;
756 ext4_group_t flex_group;
757 struct ext4_group_info *grp;
758 int encrypt = 0;
759
760 /* Cannot create files in a deleted directory */
761 if (!dir || !dir->i_nlink)
762 return ERR_PTR(-EPERM);
763
764 sb = dir->i_sb;
765 sbi = EXT4_SB(sb);
766
767 if (unlikely(ext4_forced_shutdown(sbi)))
768 return ERR_PTR(-EIO);
769
770 if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
771 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
772 !(i_flags & EXT4_EA_INODE_FL)) {
773 err = fscrypt_get_encryption_info(dir);
774 if (err)
775 return ERR_PTR(err);
776 if (!fscrypt_has_encryption_key(dir))
777 return ERR_PTR(-ENOKEY);
778 encrypt = 1;
779 }
780
781 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
782#ifdef CONFIG_EXT4_FS_POSIX_ACL
783 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
784
785 if (IS_ERR(p))
786 return ERR_CAST(p);
787 if (p) {
788 int acl_size = p->a_count * sizeof(ext4_acl_entry);
789
790 nblocks += (S_ISDIR(mode) ? 2 : 1) *
791 __ext4_xattr_set_credits(sb, NULL /* inode */,
792 NULL /* block_bh */, acl_size,
793 true /* is_create */);
794 posix_acl_release(p);
795 }
796#endif
797
798#ifdef CONFIG_SECURITY
799 {
800 int num_security_xattrs = 1;
801
802#ifdef CONFIG_INTEGRITY
803 num_security_xattrs++;
804#endif
805 /*
806 * We assume that security xattrs are never
807 * more than 1k. In practice they are under
808 * 128 bytes.
809 */
810 nblocks += num_security_xattrs *
811 __ext4_xattr_set_credits(sb, NULL /* inode */,
812 NULL /* block_bh */, 1024,
813 true /* is_create */);
814 }
815#endif
816 if (encrypt)
817 nblocks += __ext4_xattr_set_credits(sb,
818 NULL /* inode */, NULL /* block_bh */,
819 FSCRYPT_SET_CONTEXT_MAX_SIZE,
820 true /* is_create */);
821 }
822
823 ngroups = ext4_get_groups_count(sb);
824 trace_ext4_request_inode(dir, mode);
825 inode = new_inode(sb);
826 if (!inode)
827 return ERR_PTR(-ENOMEM);
828 ei = EXT4_I(inode);
829
830 /*
831 * Initialize owners and quota early so that we don't have to account
832 * for quota initialization worst case in standard inode creating
833 * transaction
834 */
835 if (owner) {
836 inode->i_mode = mode;
837 i_uid_write(inode, owner[0]);
838 i_gid_write(inode, owner[1]);
839 } else if (test_opt(sb, GRPID)) {
840 inode->i_mode = mode;
841 inode->i_uid = current_fsuid();
842 inode->i_gid = dir->i_gid;
843 } else
844 inode_init_owner(inode, dir, mode);
845
846 if (ext4_has_feature_project(sb) &&
847 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
848 ei->i_projid = EXT4_I(dir)->i_projid;
849 else
850 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
851
852 err = dquot_initialize(inode);
853 if (err)
854 goto out;
855
856 if (!goal)
857 goal = sbi->s_inode_goal;
858
859 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
860 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
861 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
862 ret2 = 0;
863 goto got_group;
864 }
865
866 if (S_ISDIR(mode))
867 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
868 else
869 ret2 = find_group_other(sb, dir, &group, mode);
870
871got_group:
872 EXT4_I(dir)->i_last_alloc_group = group;
873 err = -ENOSPC;
874 if (ret2 == -1)
875 goto out;
876
877 /*
878 * Normally we will only go through one pass of this loop,
879 * unless we get unlucky and it turns out the group we selected
880 * had its last inode grabbed by someone else.
881 */
882 for (i = 0; i < ngroups; i++, ino = 0) {
883 err = -EIO;
884
885 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
886 if (!gdp)
887 goto out;
888
889 /*
890 * Check free inodes count before loading bitmap.
891 */
892 if (ext4_free_inodes_count(sb, gdp) == 0)
893 goto next_group;
894
895 grp = ext4_get_group_info(sb, group);
896 /* Skip groups with already-known suspicious inode tables */
897 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
898 goto next_group;
899
900 brelse(inode_bitmap_bh);
901 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
902 /* Skip groups with suspicious inode tables */
903 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
904 IS_ERR(inode_bitmap_bh)) {
905 inode_bitmap_bh = NULL;
906 goto next_group;
907 }
908
909repeat_in_this_group:
910 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
911 if (!ret2)
912 goto next_group;
913
914 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
915 ext4_error(sb, "reserved inode found cleared - "
916 "inode=%lu", ino + 1);
917 goto next_group;
918 }
919
920 if (!handle) {
921 BUG_ON(nblocks <= 0);
922 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
923 handle_type, nblocks,
924 0);
925 if (IS_ERR(handle)) {
926 err = PTR_ERR(handle);
927 ext4_std_error(sb, err);
928 goto out;
929 }
930 }
931 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
932 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
933 if (err) {
934 ext4_std_error(sb, err);
935 goto out;
936 }
937 ext4_lock_group(sb, group);
938 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
939 if (ret2) {
940 /* Someone already took the bit. Repeat the search
941 * with lock held.
942 */
943 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
944 if (ret2) {
945 ext4_set_bit(ino, inode_bitmap_bh->b_data);
946 ret2 = 0;
947 } else {
948 ret2 = 1; /* we didn't grab the inode */
949 }
950 }
951 ext4_unlock_group(sb, group);
952 ino++; /* the inode bitmap is zero-based */
953 if (!ret2)
954 goto got; /* we grabbed the inode! */
955
956 if (ino < EXT4_INODES_PER_GROUP(sb))
957 goto repeat_in_this_group;
958next_group:
959 if (++group == ngroups)
960 group = 0;
961 }
962 err = -ENOSPC;
963 goto out;
964
965got:
966 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
967 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
968 if (err) {
969 ext4_std_error(sb, err);
970 goto out;
971 }
972
973 BUFFER_TRACE(group_desc_bh, "get_write_access");
974 err = ext4_journal_get_write_access(handle, group_desc_bh);
975 if (err) {
976 ext4_std_error(sb, err);
977 goto out;
978 }
979
980 /* We may have to initialize the block bitmap if it isn't already */
981 if (ext4_has_group_desc_csum(sb) &&
982 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
983 struct buffer_head *block_bitmap_bh;
984
985 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
986 if (IS_ERR(block_bitmap_bh)) {
987 err = PTR_ERR(block_bitmap_bh);
988 goto out;
989 }
990 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
991 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
992 if (err) {
993 brelse(block_bitmap_bh);
994 ext4_std_error(sb, err);
995 goto out;
996 }
997
998 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
999 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1000
1001 /* recheck and clear flag under lock if we still need to */
1002 ext4_lock_group(sb, group);
1003 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1004 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1005 ext4_free_group_clusters_set(sb, gdp,
1006 ext4_free_clusters_after_init(sb, group, gdp));
1007 ext4_block_bitmap_csum_set(sb, group, gdp,
1008 block_bitmap_bh);
1009 ext4_group_desc_csum_set(sb, group, gdp);
1010 }
1011 ext4_unlock_group(sb, group);
1012 brelse(block_bitmap_bh);
1013
1014 if (err) {
1015 ext4_std_error(sb, err);
1016 goto out;
1017 }
1018 }
1019
1020 /* Update the relevant bg descriptor fields */
1021 if (ext4_has_group_desc_csum(sb)) {
1022 int free;
1023 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1024
1025 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1026 ext4_lock_group(sb, group); /* while we modify the bg desc */
1027 free = EXT4_INODES_PER_GROUP(sb) -
1028 ext4_itable_unused_count(sb, gdp);
1029 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1030 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1031 free = 0;
1032 }
1033 /*
1034 * Check the relative inode number against the last used
1035 * relative inode number in this group. if it is greater
1036 * we need to update the bg_itable_unused count
1037 */
1038 if (ino > free)
1039 ext4_itable_unused_set(sb, gdp,
1040 (EXT4_INODES_PER_GROUP(sb) - ino));
1041 up_read(&grp->alloc_sem);
1042 } else {
1043 ext4_lock_group(sb, group);
1044 }
1045
1046 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1047 if (S_ISDIR(mode)) {
1048 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1049 if (sbi->s_log_groups_per_flex) {
1050 ext4_group_t f = ext4_flex_group(sbi, group);
1051
1052 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1053 }
1054 }
1055 if (ext4_has_group_desc_csum(sb)) {
1056 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1057 EXT4_INODES_PER_GROUP(sb) / 8);
1058 ext4_group_desc_csum_set(sb, group, gdp);
1059 }
1060 ext4_unlock_group(sb, group);
1061
1062 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1063 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1064 if (err) {
1065 ext4_std_error(sb, err);
1066 goto out;
1067 }
1068
1069 percpu_counter_dec(&sbi->s_freeinodes_counter);
1070 if (S_ISDIR(mode))
1071 percpu_counter_inc(&sbi->s_dirs_counter);
1072
1073 if (sbi->s_log_groups_per_flex) {
1074 flex_group = ext4_flex_group(sbi, group);
1075 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1076 }
1077
1078 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1079 /* This is the optimal IO size (for stat), not the fs block size */
1080 inode->i_blocks = 0;
1081 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1082 current_time(inode);
1083
1084 memset(ei->i_data, 0, sizeof(ei->i_data));
1085 ei->i_dir_start_lookup = 0;
1086 ei->i_disksize = 0;
1087
1088 /* Don't inherit extent flag from directory, amongst others. */
1089 ei->i_flags =
1090 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1091 ei->i_flags |= i_flags;
1092 ei->i_file_acl = 0;
1093 ei->i_dtime = 0;
1094 ei->i_block_group = group;
1095 ei->i_last_alloc_group = ~0;
1096
1097 ext4_set_inode_flags(inode);
1098 if (IS_DIRSYNC(inode))
1099 ext4_handle_sync(handle);
1100 if (insert_inode_locked(inode) < 0) {
1101 /*
1102 * Likely a bitmap corruption causing inode to be allocated
1103 * twice.
1104 */
1105 err = -EIO;
1106 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1107 inode->i_ino);
1108 goto out;
1109 }
1110 inode->i_generation = prandom_u32();
1111
1112 /* Precompute checksum seed for inode metadata */
1113 if (ext4_has_metadata_csum(sb)) {
1114 __u32 csum;
1115 __le32 inum = cpu_to_le32(inode->i_ino);
1116 __le32 gen = cpu_to_le32(inode->i_generation);
1117 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1118 sizeof(inum));
1119 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1120 sizeof(gen));
1121 }
1122
1123 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1124 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1125
1126 ei->i_extra_isize = sbi->s_want_extra_isize;
1127 ei->i_inline_off = 0;
1128 if (ext4_has_feature_inline_data(sb))
1129 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1130 ret = inode;
1131 err = dquot_alloc_inode(inode);
1132 if (err)
1133 goto fail_drop;
1134
1135 /*
1136 * Since the encryption xattr will always be unique, create it first so
1137 * that it's less likely to end up in an external xattr block and
1138 * prevent its deduplication.
1139 */
1140 if (encrypt) {
1141 err = fscrypt_inherit_context(dir, inode, handle, true);
1142 if (err)
1143 goto fail_free_drop;
1144 }
1145
1146 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1147 err = ext4_init_acl(handle, inode, dir);
1148 if (err)
1149 goto fail_free_drop;
1150
1151 err = ext4_init_security(handle, inode, dir, qstr);
1152 if (err)
1153 goto fail_free_drop;
1154 }
1155
1156 if (ext4_has_feature_extents(sb)) {
1157 /* set extent flag only for directory, file and normal symlink*/
1158 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1159 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1160 ext4_ext_tree_init(handle, inode);
1161 }
1162 }
1163
1164 if (ext4_handle_valid(handle)) {
1165 ei->i_sync_tid = handle->h_transaction->t_tid;
1166 ei->i_datasync_tid = handle->h_transaction->t_tid;
1167 }
1168
1169 err = ext4_mark_inode_dirty(handle, inode);
1170 if (err) {
1171 ext4_std_error(sb, err);
1172 goto fail_free_drop;
1173 }
1174
1175 ext4_debug("allocating inode %lu\n", inode->i_ino);
1176 trace_ext4_allocate_inode(inode, dir, mode);
1177 brelse(inode_bitmap_bh);
1178 return ret;
1179
1180fail_free_drop:
1181 dquot_free_inode(inode);
1182fail_drop:
1183 clear_nlink(inode);
1184 unlock_new_inode(inode);
1185out:
1186 dquot_drop(inode);
1187 inode->i_flags |= S_NOQUOTA;
1188 iput(inode);
1189 brelse(inode_bitmap_bh);
1190 return ERR_PTR(err);
1191}
1192
1193/* Verify that we are loading a valid orphan from disk */
1194struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1195{
1196 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1197 ext4_group_t block_group;
1198 int bit;
1199 struct buffer_head *bitmap_bh = NULL;
1200 struct inode *inode = NULL;
1201 int err = -EFSCORRUPTED;
1202
1203 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1204 goto bad_orphan;
1205
1206 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1207 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1208 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1209 if (IS_ERR(bitmap_bh)) {
1210 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1211 ino, PTR_ERR(bitmap_bh));
1212 return (struct inode *) bitmap_bh;
1213 }
1214
1215 /* Having the inode bit set should be a 100% indicator that this
1216 * is a valid orphan (no e2fsck run on fs). Orphans also include
1217 * inodes that were being truncated, so we can't check i_nlink==0.
1218 */
1219 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1220 goto bad_orphan;
1221
1222 inode = ext4_iget(sb, ino);
1223 if (IS_ERR(inode)) {
1224 err = PTR_ERR(inode);
1225 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1226 ino, err);
1227 return inode;
1228 }
1229
1230 /*
1231 * If the orphans has i_nlinks > 0 then it should be able to
1232 * be truncated, otherwise it won't be removed from the orphan
1233 * list during processing and an infinite loop will result.
1234 * Similarly, it must not be a bad inode.
1235 */
1236 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1237 is_bad_inode(inode))
1238 goto bad_orphan;
1239
1240 if (NEXT_ORPHAN(inode) > max_ino)
1241 goto bad_orphan;
1242 brelse(bitmap_bh);
1243 return inode;
1244
1245bad_orphan:
1246 ext4_error(sb, "bad orphan inode %lu", ino);
1247 if (bitmap_bh)
1248 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1249 bit, (unsigned long long)bitmap_bh->b_blocknr,
1250 ext4_test_bit(bit, bitmap_bh->b_data));
1251 if (inode) {
1252 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1253 is_bad_inode(inode));
1254 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1255 NEXT_ORPHAN(inode));
1256 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1257 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1258 /* Avoid freeing blocks if we got a bad deleted inode */
1259 if (inode->i_nlink == 0)
1260 inode->i_blocks = 0;
1261 iput(inode);
1262 }
1263 brelse(bitmap_bh);
1264 return ERR_PTR(err);
1265}
1266
1267unsigned long ext4_count_free_inodes(struct super_block *sb)
1268{
1269 unsigned long desc_count;
1270 struct ext4_group_desc *gdp;
1271 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1272#ifdef EXT4FS_DEBUG
1273 struct ext4_super_block *es;
1274 unsigned long bitmap_count, x;
1275 struct buffer_head *bitmap_bh = NULL;
1276
1277 es = EXT4_SB(sb)->s_es;
1278 desc_count = 0;
1279 bitmap_count = 0;
1280 gdp = NULL;
1281 for (i = 0; i < ngroups; i++) {
1282 gdp = ext4_get_group_desc(sb, i, NULL);
1283 if (!gdp)
1284 continue;
1285 desc_count += ext4_free_inodes_count(sb, gdp);
1286 brelse(bitmap_bh);
1287 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1288 if (IS_ERR(bitmap_bh)) {
1289 bitmap_bh = NULL;
1290 continue;
1291 }
1292
1293 x = ext4_count_free(bitmap_bh->b_data,
1294 EXT4_INODES_PER_GROUP(sb) / 8);
1295 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1296 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1297 bitmap_count += x;
1298 }
1299 brelse(bitmap_bh);
1300 printk(KERN_DEBUG "ext4_count_free_inodes: "
1301 "stored = %u, computed = %lu, %lu\n",
1302 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1303 return desc_count;
1304#else
1305 desc_count = 0;
1306 for (i = 0; i < ngroups; i++) {
1307 gdp = ext4_get_group_desc(sb, i, NULL);
1308 if (!gdp)
1309 continue;
1310 desc_count += ext4_free_inodes_count(sb, gdp);
1311 cond_resched();
1312 }
1313 return desc_count;
1314#endif
1315}
1316
1317/* Called at mount-time, super-block is locked */
1318unsigned long ext4_count_dirs(struct super_block * sb)
1319{
1320 unsigned long count = 0;
1321 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1322
1323 for (i = 0; i < ngroups; i++) {
1324 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1325 if (!gdp)
1326 continue;
1327 count += ext4_used_dirs_count(sb, gdp);
1328 }
1329 return count;
1330}
1331
1332/*
1333 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1334 * inode table. Must be called without any spinlock held. The only place
1335 * where it is called from on active part of filesystem is ext4lazyinit
1336 * thread, so we do not need any special locks, however we have to prevent
1337 * inode allocation from the current group, so we take alloc_sem lock, to
1338 * block ext4_new_inode() until we are finished.
1339 */
1340int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1341 int barrier)
1342{
1343 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1344 struct ext4_sb_info *sbi = EXT4_SB(sb);
1345 struct ext4_group_desc *gdp = NULL;
1346 struct buffer_head *group_desc_bh;
1347 handle_t *handle;
1348 ext4_fsblk_t blk;
1349 int num, ret = 0, used_blks = 0;
1350
1351 /* This should not happen, but just to be sure check this */
1352 if (sb_rdonly(sb)) {
1353 ret = 1;
1354 goto out;
1355 }
1356
1357 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1358 if (!gdp)
1359 goto out;
1360
1361 /*
1362 * We do not need to lock this, because we are the only one
1363 * handling this flag.
1364 */
1365 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1366 goto out;
1367
1368 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1369 if (IS_ERR(handle)) {
1370 ret = PTR_ERR(handle);
1371 goto out;
1372 }
1373
1374 down_write(&grp->alloc_sem);
1375 /*
1376 * If inode bitmap was already initialized there may be some
1377 * used inodes so we need to skip blocks with used inodes in
1378 * inode table.
1379 */
1380 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1381 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1382 ext4_itable_unused_count(sb, gdp)),
1383 sbi->s_inodes_per_block);
1384
1385 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1386 ext4_error(sb, "Something is wrong with group %u: "
1387 "used itable blocks: %d; "
1388 "itable unused count: %u",
1389 group, used_blks,
1390 ext4_itable_unused_count(sb, gdp));
1391 ret = 1;
1392 goto err_out;
1393 }
1394
1395 blk = ext4_inode_table(sb, gdp) + used_blks;
1396 num = sbi->s_itb_per_group - used_blks;
1397
1398 BUFFER_TRACE(group_desc_bh, "get_write_access");
1399 ret = ext4_journal_get_write_access(handle,
1400 group_desc_bh);
1401 if (ret)
1402 goto err_out;
1403
1404 /*
1405 * Skip zeroout if the inode table is full. But we set the ZEROED
1406 * flag anyway, because obviously, when it is full it does not need
1407 * further zeroing.
1408 */
1409 if (unlikely(num == 0))
1410 goto skip_zeroout;
1411
1412 ext4_debug("going to zero out inode table in group %d\n",
1413 group);
1414 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1415 if (ret < 0)
1416 goto err_out;
1417 if (barrier)
1418 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1419
1420skip_zeroout:
1421 ext4_lock_group(sb, group);
1422 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1423 ext4_group_desc_csum_set(sb, group, gdp);
1424 ext4_unlock_group(sb, group);
1425
1426 BUFFER_TRACE(group_desc_bh,
1427 "call ext4_handle_dirty_metadata");
1428 ret = ext4_handle_dirty_metadata(handle, NULL,
1429 group_desc_bh);
1430
1431err_out:
1432 up_write(&grp->alloc_sem);
1433 ext4_journal_stop(handle);
1434out:
1435 return ret;
1436}