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
87 if (buffer_verified(bh))
88 return 0;
89 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
90 return -EFSCORRUPTED;
91
92 ext4_lock_group(sb, block_group);
93 if (buffer_verified(bh))
94 goto verified;
95 blk = ext4_inode_bitmap(sb, desc);
96 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
97 EXT4_INODES_PER_GROUP(sb) / 8) ||
98 ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
99 ext4_unlock_group(sb, block_group);
100 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
101 "inode_bitmap = %llu", block_group, blk);
102 ext4_mark_group_bitmap_corrupted(sb, block_group,
103 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
104 return -EFSBADCRC;
105 }
106 set_buffer_verified(bh);
107verified:
108 ext4_unlock_group(sb, block_group);
109 return 0;
110}
111
112/*
113 * Read the inode allocation bitmap for a given block_group, reading
114 * into the specified slot in the superblock's bitmap cache.
115 *
116 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
117 */
118static struct buffer_head *
119ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
120{
121 struct ext4_group_desc *desc;
122 struct ext4_sb_info *sbi = EXT4_SB(sb);
123 struct buffer_head *bh = NULL;
124 ext4_fsblk_t bitmap_blk;
125 int err;
126
127 desc = ext4_get_group_desc(sb, block_group, NULL);
128 if (!desc)
129 return ERR_PTR(-EFSCORRUPTED);
130
131 bitmap_blk = ext4_inode_bitmap(sb, desc);
132 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
133 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
134 ext4_error(sb, "Invalid inode bitmap blk %llu in "
135 "block_group %u", bitmap_blk, block_group);
136 ext4_mark_group_bitmap_corrupted(sb, block_group,
137 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140 bh = sb_getblk(sb, bitmap_blk);
141 if (unlikely(!bh)) {
142 ext4_warning(sb, "Cannot read inode bitmap - "
143 "block_group = %u, inode_bitmap = %llu",
144 block_group, bitmap_blk);
145 return ERR_PTR(-ENOMEM);
146 }
147 if (bitmap_uptodate(bh))
148 goto verify;
149
150 lock_buffer(bh);
151 if (bitmap_uptodate(bh)) {
152 unlock_buffer(bh);
153 goto verify;
154 }
155
156 ext4_lock_group(sb, block_group);
157 if (ext4_has_group_desc_csum(sb) &&
158 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
159 if (block_group == 0) {
160 ext4_unlock_group(sb, block_group);
161 unlock_buffer(bh);
162 ext4_error(sb, "Inode bitmap for bg 0 marked "
163 "uninitialized");
164 err = -EFSCORRUPTED;
165 goto out;
166 }
167 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
168 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
169 sb->s_blocksize * 8, bh->b_data);
170 set_bitmap_uptodate(bh);
171 set_buffer_uptodate(bh);
172 set_buffer_verified(bh);
173 ext4_unlock_group(sb, block_group);
174 unlock_buffer(bh);
175 return bh;
176 }
177 ext4_unlock_group(sb, block_group);
178
179 if (buffer_uptodate(bh)) {
180 /*
181 * if not uninit if bh is uptodate,
182 * bitmap is also uptodate
183 */
184 set_bitmap_uptodate(bh);
185 unlock_buffer(bh);
186 goto verify;
187 }
188 /*
189 * submit the buffer_head for reading
190 */
191 trace_ext4_load_inode_bitmap(sb, block_group);
192 bh->b_end_io = ext4_end_bitmap_read;
193 get_bh(bh);
194 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
195 wait_on_buffer(bh);
196 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
197 if (!buffer_uptodate(bh)) {
198 put_bh(bh);
199 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
200 "block_group = %u, inode_bitmap = %llu",
201 block_group, bitmap_blk);
202 ext4_mark_group_bitmap_corrupted(sb, block_group,
203 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
204 return ERR_PTR(-EIO);
205 }
206
207verify:
208 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
209 if (err)
210 goto out;
211 return bh;
212out:
213 put_bh(bh);
214 return ERR_PTR(err);
215}
216
217/*
218 * NOTE! When we get the inode, we're the only people
219 * that have access to it, and as such there are no
220 * race conditions we have to worry about. The inode
221 * is not on the hash-lists, and it cannot be reached
222 * through the filesystem because the directory entry
223 * has been deleted earlier.
224 *
225 * HOWEVER: we must make sure that we get no aliases,
226 * which means that we have to call "clear_inode()"
227 * _before_ we mark the inode not in use in the inode
228 * bitmaps. Otherwise a newly created file might use
229 * the same inode number (not actually the same pointer
230 * though), and then we'd have two inodes sharing the
231 * same inode number and space on the harddisk.
232 */
233void ext4_free_inode(handle_t *handle, struct inode *inode)
234{
235 struct super_block *sb = inode->i_sb;
236 int is_directory;
237 unsigned long ino;
238 struct buffer_head *bitmap_bh = NULL;
239 struct buffer_head *bh2;
240 ext4_group_t block_group;
241 unsigned long bit;
242 struct ext4_group_desc *gdp;
243 struct ext4_super_block *es;
244 struct ext4_sb_info *sbi;
245 int fatal = 0, err, count, cleared;
246 struct ext4_group_info *grp;
247
248 if (!sb) {
249 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
250 "nonexistent device\n", __func__, __LINE__);
251 return;
252 }
253 if (atomic_read(&inode->i_count) > 1) {
254 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
255 __func__, __LINE__, inode->i_ino,
256 atomic_read(&inode->i_count));
257 return;
258 }
259 if (inode->i_nlink) {
260 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
261 __func__, __LINE__, inode->i_ino, inode->i_nlink);
262 return;
263 }
264 sbi = EXT4_SB(sb);
265
266 ino = inode->i_ino;
267 ext4_debug("freeing inode %lu\n", ino);
268 trace_ext4_free_inode(inode);
269
270 dquot_initialize(inode);
271 dquot_free_inode(inode);
272
273 is_directory = S_ISDIR(inode->i_mode);
274
275 /* Do this BEFORE marking the inode not in use or returning an error */
276 ext4_clear_inode(inode);
277
278 es = sbi->s_es;
279 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
280 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
281 goto error_return;
282 }
283 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
284 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
285 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
286 /* Don't bother if the inode bitmap is corrupt. */
287 grp = ext4_get_group_info(sb, block_group);
288 if (IS_ERR(bitmap_bh)) {
289 fatal = PTR_ERR(bitmap_bh);
290 bitmap_bh = NULL;
291 goto error_return;
292 }
293 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
294 fatal = -EFSCORRUPTED;
295 goto error_return;
296 }
297
298 BUFFER_TRACE(bitmap_bh, "get_write_access");
299 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
300 if (fatal)
301 goto error_return;
302
303 fatal = -ESRCH;
304 gdp = ext4_get_group_desc(sb, block_group, &bh2);
305 if (gdp) {
306 BUFFER_TRACE(bh2, "get_write_access");
307 fatal = ext4_journal_get_write_access(handle, bh2);
308 }
309 ext4_lock_group(sb, block_group);
310 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
311 if (fatal || !cleared) {
312 ext4_unlock_group(sb, block_group);
313 goto out;
314 }
315
316 count = ext4_free_inodes_count(sb, gdp) + 1;
317 ext4_free_inodes_set(sb, gdp, count);
318 if (is_directory) {
319 count = ext4_used_dirs_count(sb, gdp) - 1;
320 ext4_used_dirs_set(sb, gdp, count);
321 percpu_counter_dec(&sbi->s_dirs_counter);
322 }
323 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
324 EXT4_INODES_PER_GROUP(sb) / 8);
325 ext4_group_desc_csum_set(sb, block_group, gdp);
326 ext4_unlock_group(sb, block_group);
327
328 percpu_counter_inc(&sbi->s_freeinodes_counter);
329 if (sbi->s_log_groups_per_flex) {
330 struct flex_groups *fg;
331
332 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
333 ext4_flex_group(sbi, block_group));
334 atomic_inc(&fg->free_inodes);
335 if (is_directory)
336 atomic_dec(&fg->used_dirs);
337 }
338 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
339 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
340out:
341 if (cleared) {
342 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
343 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
344 if (!fatal)
345 fatal = err;
346 } else {
347 ext4_error(sb, "bit already cleared for inode %lu", ino);
348 ext4_mark_group_bitmap_corrupted(sb, block_group,
349 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
350 }
351
352error_return:
353 brelse(bitmap_bh);
354 ext4_std_error(sb, fatal);
355}
356
357struct orlov_stats {
358 __u64 free_clusters;
359 __u32 free_inodes;
360 __u32 used_dirs;
361};
362
363/*
364 * Helper function for Orlov's allocator; returns critical information
365 * for a particular block group or flex_bg. If flex_size is 1, then g
366 * is a block group number; otherwise it is flex_bg number.
367 */
368static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
369 int flex_size, struct orlov_stats *stats)
370{
371 struct ext4_group_desc *desc;
372
373 if (flex_size > 1) {
374 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
375 s_flex_groups, g);
376 stats->free_inodes = atomic_read(&fg->free_inodes);
377 stats->free_clusters = atomic64_read(&fg->free_clusters);
378 stats->used_dirs = atomic_read(&fg->used_dirs);
379 return;
380 }
381
382 desc = ext4_get_group_desc(sb, g, NULL);
383 if (desc) {
384 stats->free_inodes = ext4_free_inodes_count(sb, desc);
385 stats->free_clusters = ext4_free_group_clusters(sb, desc);
386 stats->used_dirs = ext4_used_dirs_count(sb, desc);
387 } else {
388 stats->free_inodes = 0;
389 stats->free_clusters = 0;
390 stats->used_dirs = 0;
391 }
392}
393
394/*
395 * Orlov's allocator for directories.
396 *
397 * We always try to spread first-level directories.
398 *
399 * If there are blockgroups with both free inodes and free blocks counts
400 * not worse than average we return one with smallest directory count.
401 * Otherwise we simply return a random group.
402 *
403 * For the rest rules look so:
404 *
405 * It's OK to put directory into a group unless
406 * it has too many directories already (max_dirs) or
407 * it has too few free inodes left (min_inodes) or
408 * it has too few free blocks left (min_blocks) or
409 * Parent's group is preferred, if it doesn't satisfy these
410 * conditions we search cyclically through the rest. If none
411 * of the groups look good we just look for a group with more
412 * free inodes than average (starting at parent's group).
413 */
414
415static int find_group_orlov(struct super_block *sb, struct inode *parent,
416 ext4_group_t *group, umode_t mode,
417 const struct qstr *qstr)
418{
419 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
420 struct ext4_sb_info *sbi = EXT4_SB(sb);
421 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
422 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
423 unsigned int freei, avefreei, grp_free;
424 ext4_fsblk_t freeb, avefreec;
425 unsigned int ndirs;
426 int max_dirs, min_inodes;
427 ext4_grpblk_t min_clusters;
428 ext4_group_t i, grp, g, ngroups;
429 struct ext4_group_desc *desc;
430 struct orlov_stats stats;
431 int flex_size = ext4_flex_bg_size(sbi);
432 struct dx_hash_info hinfo;
433
434 ngroups = real_ngroups;
435 if (flex_size > 1) {
436 ngroups = (real_ngroups + flex_size - 1) >>
437 sbi->s_log_groups_per_flex;
438 parent_group >>= sbi->s_log_groups_per_flex;
439 }
440
441 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
442 avefreei = freei / ngroups;
443 freeb = EXT4_C2B(sbi,
444 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
445 avefreec = freeb;
446 do_div(avefreec, ngroups);
447 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
448
449 if (S_ISDIR(mode) &&
450 ((parent == d_inode(sb->s_root)) ||
451 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
452 int best_ndir = inodes_per_group;
453 int ret = -1;
454
455 if (qstr) {
456 hinfo.hash_version = DX_HASH_HALF_MD4;
457 hinfo.seed = sbi->s_hash_seed;
458 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
459 grp = hinfo.hash;
460 } else
461 grp = prandom_u32();
462 parent_group = (unsigned)grp % ngroups;
463 for (i = 0; i < ngroups; i++) {
464 g = (parent_group + i) % ngroups;
465 get_orlov_stats(sb, g, flex_size, &stats);
466 if (!stats.free_inodes)
467 continue;
468 if (stats.used_dirs >= best_ndir)
469 continue;
470 if (stats.free_inodes < avefreei)
471 continue;
472 if (stats.free_clusters < avefreec)
473 continue;
474 grp = g;
475 ret = 0;
476 best_ndir = stats.used_dirs;
477 }
478 if (ret)
479 goto fallback;
480 found_flex_bg:
481 if (flex_size == 1) {
482 *group = grp;
483 return 0;
484 }
485
486 /*
487 * We pack inodes at the beginning of the flexgroup's
488 * inode tables. Block allocation decisions will do
489 * something similar, although regular files will
490 * start at 2nd block group of the flexgroup. See
491 * ext4_ext_find_goal() and ext4_find_near().
492 */
493 grp *= flex_size;
494 for (i = 0; i < flex_size; i++) {
495 if (grp+i >= real_ngroups)
496 break;
497 desc = ext4_get_group_desc(sb, grp+i, NULL);
498 if (desc && ext4_free_inodes_count(sb, desc)) {
499 *group = grp+i;
500 return 0;
501 }
502 }
503 goto fallback;
504 }
505
506 max_dirs = ndirs / ngroups + inodes_per_group / 16;
507 min_inodes = avefreei - inodes_per_group*flex_size / 4;
508 if (min_inodes < 1)
509 min_inodes = 1;
510 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
511
512 /*
513 * Start looking in the flex group where we last allocated an
514 * inode for this parent directory
515 */
516 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
517 parent_group = EXT4_I(parent)->i_last_alloc_group;
518 if (flex_size > 1)
519 parent_group >>= sbi->s_log_groups_per_flex;
520 }
521
522 for (i = 0; i < ngroups; i++) {
523 grp = (parent_group + i) % ngroups;
524 get_orlov_stats(sb, grp, flex_size, &stats);
525 if (stats.used_dirs >= max_dirs)
526 continue;
527 if (stats.free_inodes < min_inodes)
528 continue;
529 if (stats.free_clusters < min_clusters)
530 continue;
531 goto found_flex_bg;
532 }
533
534fallback:
535 ngroups = real_ngroups;
536 avefreei = freei / ngroups;
537fallback_retry:
538 parent_group = EXT4_I(parent)->i_block_group;
539 for (i = 0; i < ngroups; i++) {
540 grp = (parent_group + i) % ngroups;
541 desc = ext4_get_group_desc(sb, grp, NULL);
542 if (desc) {
543 grp_free = ext4_free_inodes_count(sb, desc);
544 if (grp_free && grp_free >= avefreei) {
545 *group = grp;
546 return 0;
547 }
548 }
549 }
550
551 if (avefreei) {
552 /*
553 * The free-inodes counter is approximate, and for really small
554 * filesystems the above test can fail to find any blockgroups
555 */
556 avefreei = 0;
557 goto fallback_retry;
558 }
559
560 return -1;
561}
562
563static int find_group_other(struct super_block *sb, struct inode *parent,
564 ext4_group_t *group, umode_t mode)
565{
566 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
567 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
568 struct ext4_group_desc *desc;
569 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
570
571 /*
572 * Try to place the inode is the same flex group as its
573 * parent. If we can't find space, use the Orlov algorithm to
574 * find another flex group, and store that information in the
575 * parent directory's inode information so that use that flex
576 * group for future allocations.
577 */
578 if (flex_size > 1) {
579 int retry = 0;
580
581 try_again:
582 parent_group &= ~(flex_size-1);
583 last = parent_group + flex_size;
584 if (last > ngroups)
585 last = ngroups;
586 for (i = parent_group; i < last; i++) {
587 desc = ext4_get_group_desc(sb, i, NULL);
588 if (desc && ext4_free_inodes_count(sb, desc)) {
589 *group = i;
590 return 0;
591 }
592 }
593 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
594 retry = 1;
595 parent_group = EXT4_I(parent)->i_last_alloc_group;
596 goto try_again;
597 }
598 /*
599 * If this didn't work, use the Orlov search algorithm
600 * to find a new flex group; we pass in the mode to
601 * avoid the topdir algorithms.
602 */
603 *group = parent_group + flex_size;
604 if (*group > ngroups)
605 *group = 0;
606 return find_group_orlov(sb, parent, group, mode, NULL);
607 }
608
609 /*
610 * Try to place the inode in its parent directory
611 */
612 *group = parent_group;
613 desc = ext4_get_group_desc(sb, *group, NULL);
614 if (desc && ext4_free_inodes_count(sb, desc) &&
615 ext4_free_group_clusters(sb, desc))
616 return 0;
617
618 /*
619 * We're going to place this inode in a different blockgroup from its
620 * parent. We want to cause files in a common directory to all land in
621 * the same blockgroup. But we want files which are in a different
622 * directory which shares a blockgroup with our parent to land in a
623 * different blockgroup.
624 *
625 * So add our directory's i_ino into the starting point for the hash.
626 */
627 *group = (*group + parent->i_ino) % ngroups;
628
629 /*
630 * Use a quadratic hash to find a group with a free inode and some free
631 * blocks.
632 */
633 for (i = 1; i < ngroups; i <<= 1) {
634 *group += i;
635 if (*group >= ngroups)
636 *group -= ngroups;
637 desc = ext4_get_group_desc(sb, *group, NULL);
638 if (desc && ext4_free_inodes_count(sb, desc) &&
639 ext4_free_group_clusters(sb, desc))
640 return 0;
641 }
642
643 /*
644 * That failed: try linear search for a free inode, even if that group
645 * has no free blocks.
646 */
647 *group = parent_group;
648 for (i = 0; i < ngroups; i++) {
649 if (++*group >= ngroups)
650 *group = 0;
651 desc = ext4_get_group_desc(sb, *group, NULL);
652 if (desc && ext4_free_inodes_count(sb, desc))
653 return 0;
654 }
655
656 return -1;
657}
658
659/*
660 * In no journal mode, if an inode has recently been deleted, we want
661 * to avoid reusing it until we're reasonably sure the inode table
662 * block has been written back to disk. (Yes, these values are
663 * somewhat arbitrary...)
664 */
665#define RECENTCY_MIN 60
666#define RECENTCY_DIRTY 300
667
668static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
669{
670 struct ext4_group_desc *gdp;
671 struct ext4_inode *raw_inode;
672 struct buffer_head *bh;
673 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
674 int offset, ret = 0;
675 int recentcy = RECENTCY_MIN;
676 u32 dtime, now;
677
678 gdp = ext4_get_group_desc(sb, group, NULL);
679 if (unlikely(!gdp))
680 return 0;
681
682 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
683 (ino / inodes_per_block));
684 if (!bh || !buffer_uptodate(bh))
685 /*
686 * If the block is not in the buffer cache, then it
687 * must have been written out.
688 */
689 goto out;
690
691 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
692 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
693
694 /* i_dtime is only 32 bits on disk, but we only care about relative
695 * times in the range of a few minutes (i.e. long enough to sync a
696 * recently-deleted inode to disk), so using the low 32 bits of the
697 * clock (a 68 year range) is enough, see time_before32() */
698 dtime = le32_to_cpu(raw_inode->i_dtime);
699 now = ktime_get_real_seconds();
700 if (buffer_dirty(bh))
701 recentcy += RECENTCY_DIRTY;
702
703 if (dtime && time_before32(dtime, now) &&
704 time_before32(now, dtime + recentcy))
705 ret = 1;
706out:
707 brelse(bh);
708 return ret;
709}
710
711static int find_inode_bit(struct super_block *sb, ext4_group_t group,
712 struct buffer_head *bitmap, unsigned long *ino)
713{
714 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
715 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
716
717next:
718 *ino = ext4_find_next_zero_bit((unsigned long *)
719 bitmap->b_data,
720 EXT4_INODES_PER_GROUP(sb), *ino);
721 if (*ino >= EXT4_INODES_PER_GROUP(sb))
722 goto not_found;
723
724 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
725 recently_deleted_ino = *ino;
726 *ino = *ino + 1;
727 if (*ino < EXT4_INODES_PER_GROUP(sb))
728 goto next;
729 goto not_found;
730 }
731 return 1;
732not_found:
733 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
734 return 0;
735 /*
736 * Not reusing recently deleted inodes is mostly a preference. We don't
737 * want to report ENOSPC or skew allocation patterns because of that.
738 * So return even recently deleted inode if we could find better in the
739 * given range.
740 */
741 *ino = recently_deleted_ino;
742 return 1;
743}
744
745/*
746 * There are two policies for allocating an inode. If the new inode is
747 * a directory, then a forward search is made for a block group with both
748 * free space and a low directory-to-inode ratio; if that fails, then of
749 * the groups with above-average free space, that group with the fewest
750 * directories already is chosen.
751 *
752 * For other inodes, search forward from the parent directory's block
753 * group to find a free inode.
754 */
755struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
756 umode_t mode, const struct qstr *qstr,
757 __u32 goal, uid_t *owner, __u32 i_flags,
758 int handle_type, unsigned int line_no,
759 int nblocks)
760{
761 struct super_block *sb;
762 struct buffer_head *inode_bitmap_bh = NULL;
763 struct buffer_head *group_desc_bh;
764 ext4_group_t ngroups, group = 0;
765 unsigned long ino = 0;
766 struct inode *inode;
767 struct ext4_group_desc *gdp = NULL;
768 struct ext4_inode_info *ei;
769 struct ext4_sb_info *sbi;
770 int ret2, err;
771 struct inode *ret;
772 ext4_group_t i;
773 ext4_group_t flex_group;
774 struct ext4_group_info *grp;
775 int encrypt = 0;
776
777 /* Cannot create files in a deleted directory */
778 if (!dir || !dir->i_nlink)
779 return ERR_PTR(-EPERM);
780
781 sb = dir->i_sb;
782 sbi = EXT4_SB(sb);
783
784 if (unlikely(ext4_forced_shutdown(sbi)))
785 return ERR_PTR(-EIO);
786
787 if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
788 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
789 !(i_flags & EXT4_EA_INODE_FL)) {
790 err = fscrypt_get_encryption_info(dir);
791 if (err)
792 return ERR_PTR(err);
793 if (!fscrypt_has_encryption_key(dir))
794 return ERR_PTR(-ENOKEY);
795 encrypt = 1;
796 }
797
798 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
799#ifdef CONFIG_EXT4_FS_POSIX_ACL
800 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
801
802 if (IS_ERR(p))
803 return ERR_CAST(p);
804 if (p) {
805 int acl_size = p->a_count * sizeof(ext4_acl_entry);
806
807 nblocks += (S_ISDIR(mode) ? 2 : 1) *
808 __ext4_xattr_set_credits(sb, NULL /* inode */,
809 NULL /* block_bh */, acl_size,
810 true /* is_create */);
811 posix_acl_release(p);
812 }
813#endif
814
815#ifdef CONFIG_SECURITY
816 {
817 int num_security_xattrs = 1;
818
819#ifdef CONFIG_INTEGRITY
820 num_security_xattrs++;
821#endif
822 /*
823 * We assume that security xattrs are never
824 * more than 1k. In practice they are under
825 * 128 bytes.
826 */
827 nblocks += num_security_xattrs *
828 __ext4_xattr_set_credits(sb, NULL /* inode */,
829 NULL /* block_bh */, 1024,
830 true /* is_create */);
831 }
832#endif
833 if (encrypt)
834 nblocks += __ext4_xattr_set_credits(sb,
835 NULL /* inode */, NULL /* block_bh */,
836 FSCRYPT_SET_CONTEXT_MAX_SIZE,
837 true /* is_create */);
838 }
839
840 ngroups = ext4_get_groups_count(sb);
841 trace_ext4_request_inode(dir, mode);
842 inode = new_inode(sb);
843 if (!inode)
844 return ERR_PTR(-ENOMEM);
845 ei = EXT4_I(inode);
846
847 /*
848 * Initialize owners and quota early so that we don't have to account
849 * for quota initialization worst case in standard inode creating
850 * transaction
851 */
852 if (owner) {
853 inode->i_mode = mode;
854 i_uid_write(inode, owner[0]);
855 i_gid_write(inode, owner[1]);
856 } else if (test_opt(sb, GRPID)) {
857 inode->i_mode = mode;
858 inode->i_uid = current_fsuid();
859 inode->i_gid = dir->i_gid;
860 } else
861 inode_init_owner(inode, dir, mode);
862
863 if (ext4_has_feature_project(sb) &&
864 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
865 ei->i_projid = EXT4_I(dir)->i_projid;
866 else
867 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
868
869 err = dquot_initialize(inode);
870 if (err)
871 goto out;
872
873 if (!goal)
874 goal = sbi->s_inode_goal;
875
876 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
877 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
878 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
879 ret2 = 0;
880 goto got_group;
881 }
882
883 if (S_ISDIR(mode))
884 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
885 else
886 ret2 = find_group_other(sb, dir, &group, mode);
887
888got_group:
889 EXT4_I(dir)->i_last_alloc_group = group;
890 err = -ENOSPC;
891 if (ret2 == -1)
892 goto out;
893
894 /*
895 * Normally we will only go through one pass of this loop,
896 * unless we get unlucky and it turns out the group we selected
897 * had its last inode grabbed by someone else.
898 */
899 for (i = 0; i < ngroups; i++, ino = 0) {
900 err = -EIO;
901
902 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
903 if (!gdp)
904 goto out;
905
906 /*
907 * Check free inodes count before loading bitmap.
908 */
909 if (ext4_free_inodes_count(sb, gdp) == 0)
910 goto next_group;
911
912 grp = ext4_get_group_info(sb, group);
913 /* Skip groups with already-known suspicious inode tables */
914 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
915 goto next_group;
916
917 brelse(inode_bitmap_bh);
918 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
919 /* Skip groups with suspicious inode tables */
920 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
921 IS_ERR(inode_bitmap_bh)) {
922 inode_bitmap_bh = NULL;
923 goto next_group;
924 }
925
926repeat_in_this_group:
927 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
928 if (!ret2)
929 goto next_group;
930
931 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
932 ext4_error(sb, "reserved inode found cleared - "
933 "inode=%lu", ino + 1);
934 ext4_mark_group_bitmap_corrupted(sb, group,
935 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
936 goto next_group;
937 }
938
939 if (!handle) {
940 BUG_ON(nblocks <= 0);
941 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
942 handle_type, nblocks, 0,
943 ext4_trans_default_revoke_credits(sb));
944 if (IS_ERR(handle)) {
945 err = PTR_ERR(handle);
946 ext4_std_error(sb, err);
947 goto out;
948 }
949 }
950 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
951 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
952 if (err) {
953 ext4_std_error(sb, err);
954 goto out;
955 }
956 ext4_lock_group(sb, group);
957 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
958 if (ret2) {
959 /* Someone already took the bit. Repeat the search
960 * with lock held.
961 */
962 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
963 if (ret2) {
964 ext4_set_bit(ino, inode_bitmap_bh->b_data);
965 ret2 = 0;
966 } else {
967 ret2 = 1; /* we didn't grab the inode */
968 }
969 }
970 ext4_unlock_group(sb, group);
971 ino++; /* the inode bitmap is zero-based */
972 if (!ret2)
973 goto got; /* we grabbed the inode! */
974
975 if (ino < EXT4_INODES_PER_GROUP(sb))
976 goto repeat_in_this_group;
977next_group:
978 if (++group == ngroups)
979 group = 0;
980 }
981 err = -ENOSPC;
982 goto out;
983
984got:
985 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
986 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
987 if (err) {
988 ext4_std_error(sb, err);
989 goto out;
990 }
991
992 BUFFER_TRACE(group_desc_bh, "get_write_access");
993 err = ext4_journal_get_write_access(handle, group_desc_bh);
994 if (err) {
995 ext4_std_error(sb, err);
996 goto out;
997 }
998
999 /* We may have to initialize the block bitmap if it isn't already */
1000 if (ext4_has_group_desc_csum(sb) &&
1001 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1002 struct buffer_head *block_bitmap_bh;
1003
1004 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1005 if (IS_ERR(block_bitmap_bh)) {
1006 err = PTR_ERR(block_bitmap_bh);
1007 goto out;
1008 }
1009 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1010 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1011 if (err) {
1012 brelse(block_bitmap_bh);
1013 ext4_std_error(sb, err);
1014 goto out;
1015 }
1016
1017 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1018 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1019
1020 /* recheck and clear flag under lock if we still need to */
1021 ext4_lock_group(sb, group);
1022 if (ext4_has_group_desc_csum(sb) &&
1023 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1024 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1025 ext4_free_group_clusters_set(sb, gdp,
1026 ext4_free_clusters_after_init(sb, group, gdp));
1027 ext4_block_bitmap_csum_set(sb, group, gdp,
1028 block_bitmap_bh);
1029 ext4_group_desc_csum_set(sb, group, gdp);
1030 }
1031 ext4_unlock_group(sb, group);
1032 brelse(block_bitmap_bh);
1033
1034 if (err) {
1035 ext4_std_error(sb, err);
1036 goto out;
1037 }
1038 }
1039
1040 /* Update the relevant bg descriptor fields */
1041 if (ext4_has_group_desc_csum(sb)) {
1042 int free;
1043 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1044
1045 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1046 ext4_lock_group(sb, group); /* while we modify the bg desc */
1047 free = EXT4_INODES_PER_GROUP(sb) -
1048 ext4_itable_unused_count(sb, gdp);
1049 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1050 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1051 free = 0;
1052 }
1053 /*
1054 * Check the relative inode number against the last used
1055 * relative inode number in this group. if it is greater
1056 * we need to update the bg_itable_unused count
1057 */
1058 if (ino > free)
1059 ext4_itable_unused_set(sb, gdp,
1060 (EXT4_INODES_PER_GROUP(sb) - ino));
1061 up_read(&grp->alloc_sem);
1062 } else {
1063 ext4_lock_group(sb, group);
1064 }
1065
1066 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1067 if (S_ISDIR(mode)) {
1068 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1069 if (sbi->s_log_groups_per_flex) {
1070 ext4_group_t f = ext4_flex_group(sbi, group);
1071
1072 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1073 f)->used_dirs);
1074 }
1075 }
1076 if (ext4_has_group_desc_csum(sb)) {
1077 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1078 EXT4_INODES_PER_GROUP(sb) / 8);
1079 ext4_group_desc_csum_set(sb, group, gdp);
1080 }
1081 ext4_unlock_group(sb, group);
1082
1083 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1084 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1085 if (err) {
1086 ext4_std_error(sb, err);
1087 goto out;
1088 }
1089
1090 percpu_counter_dec(&sbi->s_freeinodes_counter);
1091 if (S_ISDIR(mode))
1092 percpu_counter_inc(&sbi->s_dirs_counter);
1093
1094 if (sbi->s_log_groups_per_flex) {
1095 flex_group = ext4_flex_group(sbi, group);
1096 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1097 flex_group)->free_inodes);
1098 }
1099
1100 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1101 /* This is the optimal IO size (for stat), not the fs block size */
1102 inode->i_blocks = 0;
1103 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1104 ei->i_crtime = inode->i_mtime;
1105
1106 memset(ei->i_data, 0, sizeof(ei->i_data));
1107 ei->i_dir_start_lookup = 0;
1108 ei->i_disksize = 0;
1109
1110 /* Don't inherit extent flag from directory, amongst others. */
1111 ei->i_flags =
1112 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1113 ei->i_flags |= i_flags;
1114 ei->i_file_acl = 0;
1115 ei->i_dtime = 0;
1116 ei->i_block_group = group;
1117 ei->i_last_alloc_group = ~0;
1118
1119 ext4_set_inode_flags(inode, true);
1120 if (IS_DIRSYNC(inode))
1121 ext4_handle_sync(handle);
1122 if (insert_inode_locked(inode) < 0) {
1123 /*
1124 * Likely a bitmap corruption causing inode to be allocated
1125 * twice.
1126 */
1127 err = -EIO;
1128 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1129 inode->i_ino);
1130 ext4_mark_group_bitmap_corrupted(sb, group,
1131 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1132 goto out;
1133 }
1134 inode->i_generation = prandom_u32();
1135
1136 /* Precompute checksum seed for inode metadata */
1137 if (ext4_has_metadata_csum(sb)) {
1138 __u32 csum;
1139 __le32 inum = cpu_to_le32(inode->i_ino);
1140 __le32 gen = cpu_to_le32(inode->i_generation);
1141 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1142 sizeof(inum));
1143 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1144 sizeof(gen));
1145 }
1146
1147 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1148 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1149
1150 ei->i_extra_isize = sbi->s_want_extra_isize;
1151 ei->i_inline_off = 0;
1152 if (ext4_has_feature_inline_data(sb))
1153 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1154 ret = inode;
1155 err = dquot_alloc_inode(inode);
1156 if (err)
1157 goto fail_drop;
1158
1159 /*
1160 * Since the encryption xattr will always be unique, create it first so
1161 * that it's less likely to end up in an external xattr block and
1162 * prevent its deduplication.
1163 */
1164 if (encrypt) {
1165 err = fscrypt_inherit_context(dir, inode, handle, true);
1166 if (err)
1167 goto fail_free_drop;
1168 }
1169
1170 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1171 err = ext4_init_acl(handle, inode, dir);
1172 if (err)
1173 goto fail_free_drop;
1174
1175 err = ext4_init_security(handle, inode, dir, qstr);
1176 if (err)
1177 goto fail_free_drop;
1178 }
1179
1180 if (ext4_has_feature_extents(sb)) {
1181 /* set extent flag only for directory, file and normal symlink*/
1182 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1183 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1184 ext4_ext_tree_init(handle, inode);
1185 }
1186 }
1187
1188 if (ext4_handle_valid(handle)) {
1189 ei->i_sync_tid = handle->h_transaction->t_tid;
1190 ei->i_datasync_tid = handle->h_transaction->t_tid;
1191 }
1192
1193 err = ext4_mark_inode_dirty(handle, inode);
1194 if (err) {
1195 ext4_std_error(sb, err);
1196 goto fail_free_drop;
1197 }
1198
1199 ext4_debug("allocating inode %lu\n", inode->i_ino);
1200 trace_ext4_allocate_inode(inode, dir, mode);
1201 brelse(inode_bitmap_bh);
1202 return ret;
1203
1204fail_free_drop:
1205 dquot_free_inode(inode);
1206fail_drop:
1207 clear_nlink(inode);
1208 unlock_new_inode(inode);
1209out:
1210 dquot_drop(inode);
1211 inode->i_flags |= S_NOQUOTA;
1212 iput(inode);
1213 brelse(inode_bitmap_bh);
1214 return ERR_PTR(err);
1215}
1216
1217/* Verify that we are loading a valid orphan from disk */
1218struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1219{
1220 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1221 ext4_group_t block_group;
1222 int bit;
1223 struct buffer_head *bitmap_bh = NULL;
1224 struct inode *inode = NULL;
1225 int err = -EFSCORRUPTED;
1226
1227 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1228 goto bad_orphan;
1229
1230 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1231 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1232 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1233 if (IS_ERR(bitmap_bh))
1234 return ERR_CAST(bitmap_bh);
1235
1236 /* Having the inode bit set should be a 100% indicator that this
1237 * is a valid orphan (no e2fsck run on fs). Orphans also include
1238 * inodes that were being truncated, so we can't check i_nlink==0.
1239 */
1240 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1241 goto bad_orphan;
1242
1243 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1244 if (IS_ERR(inode)) {
1245 err = PTR_ERR(inode);
1246 ext4_error_err(sb, -err,
1247 "couldn't read orphan inode %lu (err %d)",
1248 ino, err);
1249 brelse(bitmap_bh);
1250 return inode;
1251 }
1252
1253 /*
1254 * If the orphans has i_nlinks > 0 then it should be able to
1255 * be truncated, otherwise it won't be removed from the orphan
1256 * list during processing and an infinite loop will result.
1257 * Similarly, it must not be a bad inode.
1258 */
1259 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1260 is_bad_inode(inode))
1261 goto bad_orphan;
1262
1263 if (NEXT_ORPHAN(inode) > max_ino)
1264 goto bad_orphan;
1265 brelse(bitmap_bh);
1266 return inode;
1267
1268bad_orphan:
1269 ext4_error(sb, "bad orphan inode %lu", ino);
1270 if (bitmap_bh)
1271 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1272 bit, (unsigned long long)bitmap_bh->b_blocknr,
1273 ext4_test_bit(bit, bitmap_bh->b_data));
1274 if (inode) {
1275 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1276 is_bad_inode(inode));
1277 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1278 NEXT_ORPHAN(inode));
1279 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1280 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1281 /* Avoid freeing blocks if we got a bad deleted inode */
1282 if (inode->i_nlink == 0)
1283 inode->i_blocks = 0;
1284 iput(inode);
1285 }
1286 brelse(bitmap_bh);
1287 return ERR_PTR(err);
1288}
1289
1290unsigned long ext4_count_free_inodes(struct super_block *sb)
1291{
1292 unsigned long desc_count;
1293 struct ext4_group_desc *gdp;
1294 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1295#ifdef EXT4FS_DEBUG
1296 struct ext4_super_block *es;
1297 unsigned long bitmap_count, x;
1298 struct buffer_head *bitmap_bh = NULL;
1299
1300 es = EXT4_SB(sb)->s_es;
1301 desc_count = 0;
1302 bitmap_count = 0;
1303 gdp = NULL;
1304 for (i = 0; i < ngroups; i++) {
1305 gdp = ext4_get_group_desc(sb, i, NULL);
1306 if (!gdp)
1307 continue;
1308 desc_count += ext4_free_inodes_count(sb, gdp);
1309 brelse(bitmap_bh);
1310 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1311 if (IS_ERR(bitmap_bh)) {
1312 bitmap_bh = NULL;
1313 continue;
1314 }
1315
1316 x = ext4_count_free(bitmap_bh->b_data,
1317 EXT4_INODES_PER_GROUP(sb) / 8);
1318 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1319 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1320 bitmap_count += x;
1321 }
1322 brelse(bitmap_bh);
1323 printk(KERN_DEBUG "ext4_count_free_inodes: "
1324 "stored = %u, computed = %lu, %lu\n",
1325 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1326 return desc_count;
1327#else
1328 desc_count = 0;
1329 for (i = 0; i < ngroups; i++) {
1330 gdp = ext4_get_group_desc(sb, i, NULL);
1331 if (!gdp)
1332 continue;
1333 desc_count += ext4_free_inodes_count(sb, gdp);
1334 cond_resched();
1335 }
1336 return desc_count;
1337#endif
1338}
1339
1340/* Called at mount-time, super-block is locked */
1341unsigned long ext4_count_dirs(struct super_block * sb)
1342{
1343 unsigned long count = 0;
1344 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1345
1346 for (i = 0; i < ngroups; i++) {
1347 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1348 if (!gdp)
1349 continue;
1350 count += ext4_used_dirs_count(sb, gdp);
1351 }
1352 return count;
1353}
1354
1355/*
1356 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1357 * inode table. Must be called without any spinlock held. The only place
1358 * where it is called from on active part of filesystem is ext4lazyinit
1359 * thread, so we do not need any special locks, however we have to prevent
1360 * inode allocation from the current group, so we take alloc_sem lock, to
1361 * block ext4_new_inode() until we are finished.
1362 */
1363int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1364 int barrier)
1365{
1366 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1367 struct ext4_sb_info *sbi = EXT4_SB(sb);
1368 struct ext4_group_desc *gdp = NULL;
1369 struct buffer_head *group_desc_bh;
1370 handle_t *handle;
1371 ext4_fsblk_t blk;
1372 int num, ret = 0, used_blks = 0;
1373
1374 /* This should not happen, but just to be sure check this */
1375 if (sb_rdonly(sb)) {
1376 ret = 1;
1377 goto out;
1378 }
1379
1380 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1381 if (!gdp)
1382 goto out;
1383
1384 /*
1385 * We do not need to lock this, because we are the only one
1386 * handling this flag.
1387 */
1388 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1389 goto out;
1390
1391 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1392 if (IS_ERR(handle)) {
1393 ret = PTR_ERR(handle);
1394 goto out;
1395 }
1396
1397 down_write(&grp->alloc_sem);
1398 /*
1399 * If inode bitmap was already initialized there may be some
1400 * used inodes so we need to skip blocks with used inodes in
1401 * inode table.
1402 */
1403 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1404 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1405 ext4_itable_unused_count(sb, gdp)),
1406 sbi->s_inodes_per_block);
1407
1408 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1409 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1410 ext4_itable_unused_count(sb, gdp)) <
1411 EXT4_FIRST_INO(sb)))) {
1412 ext4_error(sb, "Something is wrong with group %u: "
1413 "used itable blocks: %d; "
1414 "itable unused count: %u",
1415 group, used_blks,
1416 ext4_itable_unused_count(sb, gdp));
1417 ret = 1;
1418 goto err_out;
1419 }
1420
1421 blk = ext4_inode_table(sb, gdp) + used_blks;
1422 num = sbi->s_itb_per_group - used_blks;
1423
1424 BUFFER_TRACE(group_desc_bh, "get_write_access");
1425 ret = ext4_journal_get_write_access(handle,
1426 group_desc_bh);
1427 if (ret)
1428 goto err_out;
1429
1430 /*
1431 * Skip zeroout if the inode table is full. But we set the ZEROED
1432 * flag anyway, because obviously, when it is full it does not need
1433 * further zeroing.
1434 */
1435 if (unlikely(num == 0))
1436 goto skip_zeroout;
1437
1438 ext4_debug("going to zero out inode table in group %d\n",
1439 group);
1440 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1441 if (ret < 0)
1442 goto err_out;
1443 if (barrier)
1444 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1445
1446skip_zeroout:
1447 ext4_lock_group(sb, group);
1448 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1449 ext4_group_desc_csum_set(sb, group, gdp);
1450 ext4_unlock_group(sb, group);
1451
1452 BUFFER_TRACE(group_desc_bh,
1453 "call ext4_handle_dirty_metadata");
1454 ret = ext4_handle_dirty_metadata(handle, NULL,
1455 group_desc_bh);
1456
1457err_out:
1458 up_write(&grp->alloc_sem);
1459 ext4_journal_stop(handle);
1460out:
1461 return ret;
1462}