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;
86
87 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 return 0;
89
90 grp = ext4_get_group_info(sb, block_group);
91
92 if (buffer_verified(bh))
93 return 0;
94 if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 return -EFSCORRUPTED;
96
97 ext4_lock_group(sb, block_group);
98 if (buffer_verified(bh))
99 goto verified;
100 blk = ext4_inode_bitmap(sb, desc);
101 if (!ext4_inode_bitmap_csum_verify(sb, desc, bh,
102 EXT4_INODES_PER_GROUP(sb) / 8) ||
103 ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
104 ext4_unlock_group(sb, block_group);
105 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
106 "inode_bitmap = %llu", block_group, blk);
107 ext4_mark_group_bitmap_corrupted(sb, block_group,
108 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
109 return -EFSBADCRC;
110 }
111 set_buffer_verified(bh);
112verified:
113 ext4_unlock_group(sb, block_group);
114 return 0;
115}
116
117/*
118 * Read the inode allocation bitmap for a given block_group, reading
119 * into the specified slot in the superblock's bitmap cache.
120 *
121 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122 */
123static struct buffer_head *
124ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125{
126 struct ext4_group_desc *desc;
127 struct ext4_sb_info *sbi = EXT4_SB(sb);
128 struct buffer_head *bh = NULL;
129 ext4_fsblk_t bitmap_blk;
130 int err;
131
132 desc = ext4_get_group_desc(sb, block_group, NULL);
133 if (!desc)
134 return ERR_PTR(-EFSCORRUPTED);
135
136 bitmap_blk = ext4_inode_bitmap(sb, desc);
137 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 "block_group %u", bitmap_blk, block_group);
141 ext4_mark_group_bitmap_corrupted(sb, block_group,
142 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 return ERR_PTR(-EFSCORRUPTED);
144 }
145 bh = sb_getblk(sb, bitmap_blk);
146 if (unlikely(!bh)) {
147 ext4_warning(sb, "Cannot read inode bitmap - "
148 "block_group = %u, inode_bitmap = %llu",
149 block_group, bitmap_blk);
150 return ERR_PTR(-ENOMEM);
151 }
152 if (bitmap_uptodate(bh))
153 goto verify;
154
155 lock_buffer(bh);
156 if (bitmap_uptodate(bh)) {
157 unlock_buffer(bh);
158 goto verify;
159 }
160
161 ext4_lock_group(sb, block_group);
162 if (ext4_has_group_desc_csum(sb) &&
163 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 if (block_group == 0) {
165 ext4_unlock_group(sb, block_group);
166 unlock_buffer(bh);
167 ext4_error(sb, "Inode bitmap for bg 0 marked "
168 "uninitialized");
169 err = -EFSCORRUPTED;
170 goto out;
171 }
172 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 sb->s_blocksize * 8, bh->b_data);
175 set_bitmap_uptodate(bh);
176 set_buffer_uptodate(bh);
177 set_buffer_verified(bh);
178 ext4_unlock_group(sb, block_group);
179 unlock_buffer(bh);
180 return bh;
181 }
182 ext4_unlock_group(sb, block_group);
183
184 if (buffer_uptodate(bh)) {
185 /*
186 * if not uninit if bh is uptodate,
187 * bitmap is also uptodate
188 */
189 set_bitmap_uptodate(bh);
190 unlock_buffer(bh);
191 goto verify;
192 }
193 /*
194 * submit the buffer_head for reading
195 */
196 trace_ext4_load_inode_bitmap(sb, block_group);
197 ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 if (!buffer_uptodate(bh)) {
200 put_bh(bh);
201 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 "block_group = %u, inode_bitmap = %llu",
203 block_group, bitmap_blk);
204 ext4_mark_group_bitmap_corrupted(sb, block_group,
205 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 return ERR_PTR(-EIO);
207 }
208
209verify:
210 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 if (err)
212 goto out;
213 return bh;
214out:
215 put_bh(bh);
216 return ERR_PTR(err);
217}
218
219/*
220 * NOTE! When we get the inode, we're the only people
221 * that have access to it, and as such there are no
222 * race conditions we have to worry about. The inode
223 * is not on the hash-lists, and it cannot be reached
224 * through the filesystem because the directory entry
225 * has been deleted earlier.
226 *
227 * HOWEVER: we must make sure that we get no aliases,
228 * which means that we have to call "clear_inode()"
229 * _before_ we mark the inode not in use in the inode
230 * bitmaps. Otherwise a newly created file might use
231 * the same inode number (not actually the same pointer
232 * though), and then we'd have two inodes sharing the
233 * same inode number and space on the harddisk.
234 */
235void ext4_free_inode(handle_t *handle, struct inode *inode)
236{
237 struct super_block *sb = inode->i_sb;
238 int is_directory;
239 unsigned long ino;
240 struct buffer_head *bitmap_bh = NULL;
241 struct buffer_head *bh2;
242 ext4_group_t block_group;
243 unsigned long bit;
244 struct ext4_group_desc *gdp;
245 struct ext4_super_block *es;
246 struct ext4_sb_info *sbi;
247 int fatal = 0, err, count, cleared;
248 struct ext4_group_info *grp;
249
250 if (!sb) {
251 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 "nonexistent device\n", __func__, __LINE__);
253 return;
254 }
255 if (atomic_read(&inode->i_count) > 1) {
256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 __func__, __LINE__, inode->i_ino,
258 atomic_read(&inode->i_count));
259 return;
260 }
261 if (inode->i_nlink) {
262 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 return;
265 }
266 sbi = EXT4_SB(sb);
267
268 ino = inode->i_ino;
269 ext4_debug("freeing inode %lu\n", ino);
270 trace_ext4_free_inode(inode);
271
272 dquot_initialize(inode);
273 dquot_free_inode(inode);
274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
278 ext4_clear_inode(inode);
279
280 es = sbi->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 goto error_return;
284 }
285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 /* Don't bother if the inode bitmap is corrupt. */
289 if (IS_ERR(bitmap_bh)) {
290 fatal = PTR_ERR(bitmap_bh);
291 bitmap_bh = NULL;
292 goto error_return;
293 }
294 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 grp = ext4_get_group_info(sb, block_group);
296 if (!grp || unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 fatal = -EFSCORRUPTED;
298 goto error_return;
299 }
300 }
301
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
303 fatal = ext4_journal_get_write_access(handle, sb, bitmap_bh,
304 EXT4_JTR_NONE);
305 if (fatal)
306 goto error_return;
307
308 fatal = -ESRCH;
309 gdp = ext4_get_group_desc(sb, block_group, &bh2);
310 if (gdp) {
311 BUFFER_TRACE(bh2, "get_write_access");
312 fatal = ext4_journal_get_write_access(handle, sb, bh2,
313 EXT4_JTR_NONE);
314 }
315 ext4_lock_group(sb, block_group);
316 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
317 if (fatal || !cleared) {
318 ext4_unlock_group(sb, block_group);
319 goto out;
320 }
321
322 count = ext4_free_inodes_count(sb, gdp) + 1;
323 ext4_free_inodes_set(sb, gdp, count);
324 if (is_directory) {
325 count = ext4_used_dirs_count(sb, gdp) - 1;
326 ext4_used_dirs_set(sb, gdp, count);
327 if (percpu_counter_initialized(&sbi->s_dirs_counter))
328 percpu_counter_dec(&sbi->s_dirs_counter);
329 }
330 ext4_inode_bitmap_csum_set(sb, gdp, bitmap_bh,
331 EXT4_INODES_PER_GROUP(sb) / 8);
332 ext4_group_desc_csum_set(sb, block_group, gdp);
333 ext4_unlock_group(sb, block_group);
334
335 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
336 percpu_counter_inc(&sbi->s_freeinodes_counter);
337 if (sbi->s_log_groups_per_flex) {
338 struct flex_groups *fg;
339
340 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
341 ext4_flex_group(sbi, block_group));
342 atomic_inc(&fg->free_inodes);
343 if (is_directory)
344 atomic_dec(&fg->used_dirs);
345 }
346 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
347 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
348out:
349 if (cleared) {
350 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
351 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
352 if (!fatal)
353 fatal = err;
354 } else {
355 ext4_error(sb, "bit already cleared for inode %lu", ino);
356 ext4_mark_group_bitmap_corrupted(sb, block_group,
357 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
358 }
359
360error_return:
361 brelse(bitmap_bh);
362 ext4_std_error(sb, fatal);
363}
364
365struct orlov_stats {
366 __u64 free_clusters;
367 __u32 free_inodes;
368 __u32 used_dirs;
369};
370
371/*
372 * Helper function for Orlov's allocator; returns critical information
373 * for a particular block group or flex_bg. If flex_size is 1, then g
374 * is a block group number; otherwise it is flex_bg number.
375 */
376static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
377 int flex_size, struct orlov_stats *stats)
378{
379 struct ext4_group_desc *desc;
380
381 if (flex_size > 1) {
382 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
383 s_flex_groups, g);
384 stats->free_inodes = atomic_read(&fg->free_inodes);
385 stats->free_clusters = atomic64_read(&fg->free_clusters);
386 stats->used_dirs = atomic_read(&fg->used_dirs);
387 return;
388 }
389
390 desc = ext4_get_group_desc(sb, g, NULL);
391 if (desc) {
392 stats->free_inodes = ext4_free_inodes_count(sb, desc);
393 stats->free_clusters = ext4_free_group_clusters(sb, desc);
394 stats->used_dirs = ext4_used_dirs_count(sb, desc);
395 } else {
396 stats->free_inodes = 0;
397 stats->free_clusters = 0;
398 stats->used_dirs = 0;
399 }
400}
401
402/*
403 * Orlov's allocator for directories.
404 *
405 * We always try to spread first-level directories.
406 *
407 * If there are blockgroups with both free inodes and free clusters counts
408 * not worse than average we return one with smallest directory count.
409 * Otherwise we simply return a random group.
410 *
411 * For the rest rules look so:
412 *
413 * It's OK to put directory into a group unless
414 * it has too many directories already (max_dirs) or
415 * it has too few free inodes left (min_inodes) or
416 * it has too few free clusters left (min_clusters) or
417 * Parent's group is preferred, if it doesn't satisfy these
418 * conditions we search cyclically through the rest. If none
419 * of the groups look good we just look for a group with more
420 * free inodes than average (starting at parent's group).
421 */
422
423static int find_group_orlov(struct super_block *sb, struct inode *parent,
424 ext4_group_t *group, umode_t mode,
425 const struct qstr *qstr)
426{
427 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
428 struct ext4_sb_info *sbi = EXT4_SB(sb);
429 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
430 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
431 unsigned int freei, avefreei, grp_free;
432 ext4_fsblk_t freec, avefreec;
433 unsigned int ndirs;
434 int max_dirs, min_inodes;
435 ext4_grpblk_t min_clusters;
436 ext4_group_t i, grp, g, ngroups;
437 struct ext4_group_desc *desc;
438 struct orlov_stats stats;
439 int flex_size = ext4_flex_bg_size(sbi);
440 struct dx_hash_info hinfo;
441
442 ngroups = real_ngroups;
443 if (flex_size > 1) {
444 ngroups = (real_ngroups + flex_size - 1) >>
445 sbi->s_log_groups_per_flex;
446 parent_group >>= sbi->s_log_groups_per_flex;
447 }
448
449 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
450 avefreei = freei / ngroups;
451 freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
452 avefreec = freec;
453 do_div(avefreec, ngroups);
454 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
455
456 if (S_ISDIR(mode) &&
457 ((parent == d_inode(sb->s_root)) ||
458 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
459 int best_ndir = inodes_per_group;
460 int ret = -1;
461
462 if (qstr) {
463 hinfo.hash_version = DX_HASH_HALF_MD4;
464 hinfo.seed = sbi->s_hash_seed;
465 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
466 parent_group = hinfo.hash % ngroups;
467 } else
468 parent_group = get_random_u32_below(ngroups);
469 for (i = 0; i < ngroups; i++) {
470 g = (parent_group + i) % ngroups;
471 get_orlov_stats(sb, g, flex_size, &stats);
472 if (!stats.free_inodes)
473 continue;
474 if (stats.used_dirs >= best_ndir)
475 continue;
476 if (stats.free_inodes < avefreei)
477 continue;
478 if (stats.free_clusters < avefreec)
479 continue;
480 grp = g;
481 ret = 0;
482 best_ndir = stats.used_dirs;
483 }
484 if (ret)
485 goto fallback;
486 found_flex_bg:
487 if (flex_size == 1) {
488 *group = grp;
489 return 0;
490 }
491
492 /*
493 * We pack inodes at the beginning of the flexgroup's
494 * inode tables. Block allocation decisions will do
495 * something similar, although regular files will
496 * start at 2nd block group of the flexgroup. See
497 * ext4_ext_find_goal() and ext4_find_near().
498 */
499 grp *= flex_size;
500 for (i = 0; i < flex_size; i++) {
501 if (grp+i >= real_ngroups)
502 break;
503 desc = ext4_get_group_desc(sb, grp+i, NULL);
504 if (desc && ext4_free_inodes_count(sb, desc)) {
505 *group = grp+i;
506 return 0;
507 }
508 }
509 goto fallback;
510 }
511
512 max_dirs = ndirs / ngroups + inodes_per_group*flex_size / 16;
513 min_inodes = avefreei - inodes_per_group*flex_size / 4;
514 if (min_inodes < 1)
515 min_inodes = 1;
516 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
517
518 /*
519 * Start looking in the flex group where we last allocated an
520 * inode for this parent directory
521 */
522 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
523 parent_group = EXT4_I(parent)->i_last_alloc_group;
524 if (flex_size > 1)
525 parent_group >>= sbi->s_log_groups_per_flex;
526 }
527
528 for (i = 0; i < ngroups; i++) {
529 grp = (parent_group + i) % ngroups;
530 get_orlov_stats(sb, grp, flex_size, &stats);
531 if (stats.used_dirs >= max_dirs)
532 continue;
533 if (stats.free_inodes < min_inodes)
534 continue;
535 if (stats.free_clusters < min_clusters)
536 continue;
537 goto found_flex_bg;
538 }
539
540fallback:
541 ngroups = real_ngroups;
542 avefreei = freei / ngroups;
543fallback_retry:
544 parent_group = EXT4_I(parent)->i_block_group;
545 for (i = 0; i < ngroups; i++) {
546 grp = (parent_group + i) % ngroups;
547 desc = ext4_get_group_desc(sb, grp, NULL);
548 if (desc) {
549 grp_free = ext4_free_inodes_count(sb, desc);
550 if (grp_free && grp_free >= avefreei) {
551 *group = grp;
552 return 0;
553 }
554 }
555 }
556
557 if (avefreei) {
558 /*
559 * The free-inodes counter is approximate, and for really small
560 * filesystems the above test can fail to find any blockgroups
561 */
562 avefreei = 0;
563 goto fallback_retry;
564 }
565
566 return -1;
567}
568
569static int find_group_other(struct super_block *sb, struct inode *parent,
570 ext4_group_t *group, umode_t mode)
571{
572 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
573 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
574 struct ext4_group_desc *desc;
575 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
576
577 /*
578 * Try to place the inode is the same flex group as its
579 * parent. If we can't find space, use the Orlov algorithm to
580 * find another flex group, and store that information in the
581 * parent directory's inode information so that use that flex
582 * group for future allocations.
583 */
584 if (flex_size > 1) {
585 int retry = 0;
586
587 try_again:
588 parent_group &= ~(flex_size-1);
589 last = parent_group + flex_size;
590 if (last > ngroups)
591 last = ngroups;
592 for (i = parent_group; i < last; i++) {
593 desc = ext4_get_group_desc(sb, i, NULL);
594 if (desc && ext4_free_inodes_count(sb, desc)) {
595 *group = i;
596 return 0;
597 }
598 }
599 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
600 retry = 1;
601 parent_group = EXT4_I(parent)->i_last_alloc_group;
602 goto try_again;
603 }
604 /*
605 * If this didn't work, use the Orlov search algorithm
606 * to find a new flex group; we pass in the mode to
607 * avoid the topdir algorithms.
608 */
609 *group = parent_group + flex_size;
610 if (*group > ngroups)
611 *group = 0;
612 return find_group_orlov(sb, parent, group, mode, NULL);
613 }
614
615 /*
616 * Try to place the inode in its parent directory
617 */
618 *group = parent_group;
619 desc = ext4_get_group_desc(sb, *group, NULL);
620 if (desc && ext4_free_inodes_count(sb, desc) &&
621 ext4_free_group_clusters(sb, desc))
622 return 0;
623
624 /*
625 * We're going to place this inode in a different blockgroup from its
626 * parent. We want to cause files in a common directory to all land in
627 * the same blockgroup. But we want files which are in a different
628 * directory which shares a blockgroup with our parent to land in a
629 * different blockgroup.
630 *
631 * So add our directory's i_ino into the starting point for the hash.
632 */
633 *group = (*group + parent->i_ino) % ngroups;
634
635 /*
636 * Use a quadratic hash to find a group with a free inode and some free
637 * blocks.
638 */
639 for (i = 1; i < ngroups; i <<= 1) {
640 *group += i;
641 if (*group >= ngroups)
642 *group -= ngroups;
643 desc = ext4_get_group_desc(sb, *group, NULL);
644 if (desc && ext4_free_inodes_count(sb, desc) &&
645 ext4_free_group_clusters(sb, desc))
646 return 0;
647 }
648
649 /*
650 * That failed: try linear search for a free inode, even if that group
651 * has no free blocks.
652 */
653 *group = parent_group;
654 for (i = 0; i < ngroups; i++) {
655 if (++*group >= ngroups)
656 *group = 0;
657 desc = ext4_get_group_desc(sb, *group, NULL);
658 if (desc && ext4_free_inodes_count(sb, desc))
659 return 0;
660 }
661
662 return -1;
663}
664
665/*
666 * In no journal mode, if an inode has recently been deleted, we want
667 * to avoid reusing it until we're reasonably sure the inode table
668 * block has been written back to disk. (Yes, these values are
669 * somewhat arbitrary...)
670 */
671#define RECENTCY_MIN 60
672#define RECENTCY_DIRTY 300
673
674static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
675{
676 struct ext4_group_desc *gdp;
677 struct ext4_inode *raw_inode;
678 struct buffer_head *bh;
679 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
680 int offset, ret = 0;
681 int recentcy = RECENTCY_MIN;
682 u32 dtime, now;
683
684 gdp = ext4_get_group_desc(sb, group, NULL);
685 if (unlikely(!gdp))
686 return 0;
687
688 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
689 (ino / inodes_per_block));
690 if (!bh || !buffer_uptodate(bh))
691 /*
692 * If the block is not in the buffer cache, then it
693 * must have been written out.
694 */
695 goto out;
696
697 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
698 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
699
700 /* i_dtime is only 32 bits on disk, but we only care about relative
701 * times in the range of a few minutes (i.e. long enough to sync a
702 * recently-deleted inode to disk), so using the low 32 bits of the
703 * clock (a 68 year range) is enough, see time_before32() */
704 dtime = le32_to_cpu(raw_inode->i_dtime);
705 now = ktime_get_real_seconds();
706 if (buffer_dirty(bh))
707 recentcy += RECENTCY_DIRTY;
708
709 if (dtime && time_before32(dtime, now) &&
710 time_before32(now, dtime + recentcy))
711 ret = 1;
712out:
713 brelse(bh);
714 return ret;
715}
716
717static int find_inode_bit(struct super_block *sb, ext4_group_t group,
718 struct buffer_head *bitmap, unsigned long *ino)
719{
720 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
721 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
722
723next:
724 *ino = ext4_find_next_zero_bit((unsigned long *)
725 bitmap->b_data,
726 EXT4_INODES_PER_GROUP(sb), *ino);
727 if (*ino >= EXT4_INODES_PER_GROUP(sb))
728 goto not_found;
729
730 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
731 recently_deleted_ino = *ino;
732 *ino = *ino + 1;
733 if (*ino < EXT4_INODES_PER_GROUP(sb))
734 goto next;
735 goto not_found;
736 }
737 return 1;
738not_found:
739 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
740 return 0;
741 /*
742 * Not reusing recently deleted inodes is mostly a preference. We don't
743 * want to report ENOSPC or skew allocation patterns because of that.
744 * So return even recently deleted inode if we could find better in the
745 * given range.
746 */
747 *ino = recently_deleted_ino;
748 return 1;
749}
750
751int ext4_mark_inode_used(struct super_block *sb, int ino)
752{
753 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
754 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
755 struct ext4_group_desc *gdp;
756 ext4_group_t group;
757 int bit;
758 int err = -EFSCORRUPTED;
759
760 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
761 goto out;
762
763 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
764 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
765 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
766 if (IS_ERR(inode_bitmap_bh))
767 return PTR_ERR(inode_bitmap_bh);
768
769 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
770 err = 0;
771 goto out;
772 }
773
774 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
775 if (!gdp || !group_desc_bh) {
776 err = -EINVAL;
777 goto out;
778 }
779
780 ext4_set_bit(bit, inode_bitmap_bh->b_data);
781
782 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
783 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
784 if (err) {
785 ext4_std_error(sb, err);
786 goto out;
787 }
788 err = sync_dirty_buffer(inode_bitmap_bh);
789 if (err) {
790 ext4_std_error(sb, err);
791 goto out;
792 }
793
794 /* We may have to initialize the block bitmap if it isn't already */
795 if (ext4_has_group_desc_csum(sb) &&
796 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
797 struct buffer_head *block_bitmap_bh;
798
799 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
800 if (IS_ERR(block_bitmap_bh)) {
801 err = PTR_ERR(block_bitmap_bh);
802 goto out;
803 }
804
805 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
806 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
807 sync_dirty_buffer(block_bitmap_bh);
808
809 /* recheck and clear flag under lock if we still need to */
810 ext4_lock_group(sb, group);
811 if (ext4_has_group_desc_csum(sb) &&
812 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
813 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
814 ext4_free_group_clusters_set(sb, gdp,
815 ext4_free_clusters_after_init(sb, group, gdp));
816 ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
817 ext4_group_desc_csum_set(sb, group, gdp);
818 }
819 ext4_unlock_group(sb, group);
820 brelse(block_bitmap_bh);
821
822 if (err) {
823 ext4_std_error(sb, err);
824 goto out;
825 }
826 }
827
828 /* Update the relevant bg descriptor fields */
829 if (ext4_has_group_desc_csum(sb)) {
830 int free;
831
832 ext4_lock_group(sb, group); /* while we modify the bg desc */
833 free = EXT4_INODES_PER_GROUP(sb) -
834 ext4_itable_unused_count(sb, gdp);
835 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
836 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
837 free = 0;
838 }
839
840 /*
841 * Check the relative inode number against the last used
842 * relative inode number in this group. if it is greater
843 * we need to update the bg_itable_unused count
844 */
845 if (bit >= free)
846 ext4_itable_unused_set(sb, gdp,
847 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
848 } else {
849 ext4_lock_group(sb, group);
850 }
851
852 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
853 if (ext4_has_group_desc_csum(sb)) {
854 ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh,
855 EXT4_INODES_PER_GROUP(sb) / 8);
856 ext4_group_desc_csum_set(sb, group, gdp);
857 }
858
859 ext4_unlock_group(sb, group);
860 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
861 sync_dirty_buffer(group_desc_bh);
862out:
863 return err;
864}
865
866static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
867 bool encrypt)
868{
869 struct super_block *sb = dir->i_sb;
870 int nblocks = 0;
871#ifdef CONFIG_EXT4_FS_POSIX_ACL
872 struct posix_acl *p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
873
874 if (IS_ERR(p))
875 return PTR_ERR(p);
876 if (p) {
877 int acl_size = p->a_count * sizeof(ext4_acl_entry);
878
879 nblocks += (S_ISDIR(mode) ? 2 : 1) *
880 __ext4_xattr_set_credits(sb, NULL /* inode */,
881 NULL /* block_bh */, acl_size,
882 true /* is_create */);
883 posix_acl_release(p);
884 }
885#endif
886
887#ifdef CONFIG_SECURITY
888 {
889 int num_security_xattrs = 1;
890
891#ifdef CONFIG_INTEGRITY
892 num_security_xattrs++;
893#endif
894 /*
895 * We assume that security xattrs are never more than 1k.
896 * In practice they are under 128 bytes.
897 */
898 nblocks += num_security_xattrs *
899 __ext4_xattr_set_credits(sb, NULL /* inode */,
900 NULL /* block_bh */, 1024,
901 true /* is_create */);
902 }
903#endif
904 if (encrypt)
905 nblocks += __ext4_xattr_set_credits(sb,
906 NULL /* inode */,
907 NULL /* block_bh */,
908 FSCRYPT_SET_CONTEXT_MAX_SIZE,
909 true /* is_create */);
910 return nblocks;
911}
912
913/*
914 * There are two policies for allocating an inode. If the new inode is
915 * a directory, then a forward search is made for a block group with both
916 * free space and a low directory-to-inode ratio; if that fails, then of
917 * the groups with above-average free space, that group with the fewest
918 * directories already is chosen.
919 *
920 * For other inodes, search forward from the parent directory's block
921 * group to find a free inode.
922 */
923struct inode *__ext4_new_inode(struct mnt_idmap *idmap,
924 handle_t *handle, struct inode *dir,
925 umode_t mode, const struct qstr *qstr,
926 __u32 goal, uid_t *owner, __u32 i_flags,
927 int handle_type, unsigned int line_no,
928 int nblocks)
929{
930 struct super_block *sb;
931 struct buffer_head *inode_bitmap_bh = NULL;
932 struct buffer_head *group_desc_bh;
933 ext4_group_t ngroups, group = 0;
934 unsigned long ino = 0;
935 struct inode *inode;
936 struct ext4_group_desc *gdp = NULL;
937 struct ext4_inode_info *ei;
938 struct ext4_sb_info *sbi;
939 int ret2, err;
940 struct inode *ret;
941 ext4_group_t i;
942 ext4_group_t flex_group;
943 struct ext4_group_info *grp = NULL;
944 bool encrypt = false;
945
946 /* Cannot create files in a deleted directory */
947 if (!dir || !dir->i_nlink)
948 return ERR_PTR(-EPERM);
949
950 sb = dir->i_sb;
951 sbi = EXT4_SB(sb);
952
953 if (unlikely(ext4_forced_shutdown(sb)))
954 return ERR_PTR(-EIO);
955
956 ngroups = ext4_get_groups_count(sb);
957 trace_ext4_request_inode(dir, mode);
958 inode = new_inode(sb);
959 if (!inode)
960 return ERR_PTR(-ENOMEM);
961 ei = EXT4_I(inode);
962
963 /*
964 * Initialize owners and quota early so that we don't have to account
965 * for quota initialization worst case in standard inode creating
966 * transaction
967 */
968 if (owner) {
969 inode->i_mode = mode;
970 i_uid_write(inode, owner[0]);
971 i_gid_write(inode, owner[1]);
972 } else if (test_opt(sb, GRPID)) {
973 inode->i_mode = mode;
974 inode_fsuid_set(inode, idmap);
975 inode->i_gid = dir->i_gid;
976 } else
977 inode_init_owner(idmap, inode, dir, mode);
978
979 if (ext4_has_feature_project(sb) &&
980 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
981 ei->i_projid = EXT4_I(dir)->i_projid;
982 else
983 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
984
985 if (!(i_flags & EXT4_EA_INODE_FL)) {
986 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
987 if (err)
988 goto out;
989 }
990
991 err = dquot_initialize(inode);
992 if (err)
993 goto out;
994
995 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
996 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
997 if (ret2 < 0) {
998 err = ret2;
999 goto out;
1000 }
1001 nblocks += ret2;
1002 }
1003
1004 if (!goal)
1005 goal = sbi->s_inode_goal;
1006
1007 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1008 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1009 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1010 ret2 = 0;
1011 goto got_group;
1012 }
1013
1014 if (S_ISDIR(mode))
1015 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1016 else
1017 ret2 = find_group_other(sb, dir, &group, mode);
1018
1019got_group:
1020 EXT4_I(dir)->i_last_alloc_group = group;
1021 err = -ENOSPC;
1022 if (ret2 == -1)
1023 goto out;
1024
1025 /*
1026 * Normally we will only go through one pass of this loop,
1027 * unless we get unlucky and it turns out the group we selected
1028 * had its last inode grabbed by someone else.
1029 */
1030 for (i = 0; i < ngroups; i++, ino = 0) {
1031 err = -EIO;
1032
1033 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1034 if (!gdp)
1035 goto out;
1036
1037 /*
1038 * Check free inodes count before loading bitmap.
1039 */
1040 if (ext4_free_inodes_count(sb, gdp) == 0)
1041 goto next_group;
1042
1043 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1044 grp = ext4_get_group_info(sb, group);
1045 /*
1046 * Skip groups with already-known suspicious inode
1047 * tables
1048 */
1049 if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1050 goto next_group;
1051 }
1052
1053 brelse(inode_bitmap_bh);
1054 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1055 /* Skip groups with suspicious inode tables */
1056 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1057 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1058 IS_ERR(inode_bitmap_bh)) {
1059 inode_bitmap_bh = NULL;
1060 goto next_group;
1061 }
1062
1063repeat_in_this_group:
1064 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1065 if (!ret2)
1066 goto next_group;
1067
1068 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1069 ext4_error(sb, "reserved inode found cleared - "
1070 "inode=%lu", ino + 1);
1071 ext4_mark_group_bitmap_corrupted(sb, group,
1072 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1073 goto next_group;
1074 }
1075
1076 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1077 BUG_ON(nblocks <= 0);
1078 handle = __ext4_journal_start_sb(NULL, dir->i_sb,
1079 line_no, handle_type, nblocks, 0,
1080 ext4_trans_default_revoke_credits(sb));
1081 if (IS_ERR(handle)) {
1082 err = PTR_ERR(handle);
1083 ext4_std_error(sb, err);
1084 goto out;
1085 }
1086 }
1087 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1088 err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,
1089 EXT4_JTR_NONE);
1090 if (err) {
1091 ext4_std_error(sb, err);
1092 goto out;
1093 }
1094 ext4_lock_group(sb, group);
1095 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1096 if (ret2) {
1097 /* Someone already took the bit. Repeat the search
1098 * with lock held.
1099 */
1100 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1101 if (ret2) {
1102 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1103 ret2 = 0;
1104 } else {
1105 ret2 = 1; /* we didn't grab the inode */
1106 }
1107 }
1108 ext4_unlock_group(sb, group);
1109 ino++; /* the inode bitmap is zero-based */
1110 if (!ret2)
1111 goto got; /* we grabbed the inode! */
1112
1113 if (ino < EXT4_INODES_PER_GROUP(sb))
1114 goto repeat_in_this_group;
1115next_group:
1116 if (++group == ngroups)
1117 group = 0;
1118 }
1119 err = -ENOSPC;
1120 goto out;
1121
1122got:
1123 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1124 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1125 if (err) {
1126 ext4_std_error(sb, err);
1127 goto out;
1128 }
1129
1130 BUFFER_TRACE(group_desc_bh, "get_write_access");
1131 err = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1132 EXT4_JTR_NONE);
1133 if (err) {
1134 ext4_std_error(sb, err);
1135 goto out;
1136 }
1137
1138 /* We may have to initialize the block bitmap if it isn't already */
1139 if (ext4_has_group_desc_csum(sb) &&
1140 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1141 struct buffer_head *block_bitmap_bh;
1142
1143 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1144 if (IS_ERR(block_bitmap_bh)) {
1145 err = PTR_ERR(block_bitmap_bh);
1146 goto out;
1147 }
1148 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1149 err = ext4_journal_get_write_access(handle, sb, block_bitmap_bh,
1150 EXT4_JTR_NONE);
1151 if (err) {
1152 brelse(block_bitmap_bh);
1153 ext4_std_error(sb, err);
1154 goto out;
1155 }
1156
1157 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1158 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1159
1160 /* recheck and clear flag under lock if we still need to */
1161 ext4_lock_group(sb, group);
1162 if (ext4_has_group_desc_csum(sb) &&
1163 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1164 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1165 ext4_free_group_clusters_set(sb, gdp,
1166 ext4_free_clusters_after_init(sb, group, gdp));
1167 ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
1168 ext4_group_desc_csum_set(sb, group, gdp);
1169 }
1170 ext4_unlock_group(sb, group);
1171 brelse(block_bitmap_bh);
1172
1173 if (err) {
1174 ext4_std_error(sb, err);
1175 goto out;
1176 }
1177 }
1178
1179 /* Update the relevant bg descriptor fields */
1180 if (ext4_has_group_desc_csum(sb)) {
1181 int free;
1182 struct ext4_group_info *grp = NULL;
1183
1184 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1185 grp = ext4_get_group_info(sb, group);
1186 if (!grp) {
1187 err = -EFSCORRUPTED;
1188 goto out;
1189 }
1190 down_read(&grp->alloc_sem); /*
1191 * protect vs itable
1192 * lazyinit
1193 */
1194 }
1195 ext4_lock_group(sb, group); /* while we modify the bg desc */
1196 free = EXT4_INODES_PER_GROUP(sb) -
1197 ext4_itable_unused_count(sb, gdp);
1198 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1199 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1200 free = 0;
1201 }
1202 /*
1203 * Check the relative inode number against the last used
1204 * relative inode number in this group. if it is greater
1205 * we need to update the bg_itable_unused count
1206 */
1207 if (ino > free)
1208 ext4_itable_unused_set(sb, gdp,
1209 (EXT4_INODES_PER_GROUP(sb) - ino));
1210 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1211 up_read(&grp->alloc_sem);
1212 } else {
1213 ext4_lock_group(sb, group);
1214 }
1215
1216 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1217 if (S_ISDIR(mode)) {
1218 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1219 if (sbi->s_log_groups_per_flex) {
1220 ext4_group_t f = ext4_flex_group(sbi, group);
1221
1222 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1223 f)->used_dirs);
1224 }
1225 }
1226 if (ext4_has_group_desc_csum(sb)) {
1227 ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh,
1228 EXT4_INODES_PER_GROUP(sb) / 8);
1229 ext4_group_desc_csum_set(sb, group, gdp);
1230 }
1231 ext4_unlock_group(sb, group);
1232
1233 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1234 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1235 if (err) {
1236 ext4_std_error(sb, err);
1237 goto out;
1238 }
1239
1240 percpu_counter_dec(&sbi->s_freeinodes_counter);
1241 if (S_ISDIR(mode))
1242 percpu_counter_inc(&sbi->s_dirs_counter);
1243
1244 if (sbi->s_log_groups_per_flex) {
1245 flex_group = ext4_flex_group(sbi, group);
1246 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1247 flex_group)->free_inodes);
1248 }
1249
1250 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1251 /* This is the optimal IO size (for stat), not the fs block size */
1252 inode->i_blocks = 0;
1253 simple_inode_init_ts(inode);
1254 ei->i_crtime = inode_get_mtime(inode);
1255
1256 memset(ei->i_data, 0, sizeof(ei->i_data));
1257 ei->i_dir_start_lookup = 0;
1258 ei->i_disksize = 0;
1259
1260 /* Don't inherit extent flag from directory, amongst others. */
1261 ei->i_flags =
1262 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1263 ei->i_flags |= i_flags;
1264 ei->i_file_acl = 0;
1265 ei->i_dtime = 0;
1266 ei->i_block_group = group;
1267 ei->i_last_alloc_group = ~0;
1268
1269 ext4_set_inode_flags(inode, true);
1270 if (IS_DIRSYNC(inode))
1271 ext4_handle_sync(handle);
1272 if (insert_inode_locked(inode) < 0) {
1273 /*
1274 * Likely a bitmap corruption causing inode to be allocated
1275 * twice.
1276 */
1277 err = -EIO;
1278 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1279 inode->i_ino);
1280 ext4_mark_group_bitmap_corrupted(sb, group,
1281 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1282 goto out;
1283 }
1284 inode->i_generation = get_random_u32();
1285
1286 /* Precompute checksum seed for inode metadata */
1287 if (ext4_has_metadata_csum(sb)) {
1288 __u32 csum;
1289 __le32 inum = cpu_to_le32(inode->i_ino);
1290 __le32 gen = cpu_to_le32(inode->i_generation);
1291 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1292 sizeof(inum));
1293 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1294 sizeof(gen));
1295 }
1296
1297 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1298 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1299
1300 ei->i_extra_isize = sbi->s_want_extra_isize;
1301 ei->i_inline_off = 0;
1302 if (ext4_has_feature_inline_data(sb) &&
1303 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1304 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1305 ret = inode;
1306 err = dquot_alloc_inode(inode);
1307 if (err)
1308 goto fail_drop;
1309
1310 /*
1311 * Since the encryption xattr will always be unique, create it first so
1312 * that it's less likely to end up in an external xattr block and
1313 * prevent its deduplication.
1314 */
1315 if (encrypt) {
1316 err = fscrypt_set_context(inode, handle);
1317 if (err)
1318 goto fail_free_drop;
1319 }
1320
1321 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1322 err = ext4_init_acl(handle, inode, dir);
1323 if (err)
1324 goto fail_free_drop;
1325
1326 err = ext4_init_security(handle, inode, dir, qstr);
1327 if (err)
1328 goto fail_free_drop;
1329 }
1330
1331 if (ext4_has_feature_extents(sb)) {
1332 /* set extent flag only for directory, file and normal symlink*/
1333 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1334 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1335 ext4_ext_tree_init(handle, inode);
1336 }
1337 }
1338
1339 if (ext4_handle_valid(handle)) {
1340 ei->i_sync_tid = handle->h_transaction->t_tid;
1341 ei->i_datasync_tid = handle->h_transaction->t_tid;
1342 }
1343
1344 err = ext4_mark_inode_dirty(handle, inode);
1345 if (err) {
1346 ext4_std_error(sb, err);
1347 goto fail_free_drop;
1348 }
1349
1350 ext4_debug("allocating inode %lu\n", inode->i_ino);
1351 trace_ext4_allocate_inode(inode, dir, mode);
1352 brelse(inode_bitmap_bh);
1353 return ret;
1354
1355fail_free_drop:
1356 dquot_free_inode(inode);
1357fail_drop:
1358 clear_nlink(inode);
1359 unlock_new_inode(inode);
1360out:
1361 dquot_drop(inode);
1362 inode->i_flags |= S_NOQUOTA;
1363 iput(inode);
1364 brelse(inode_bitmap_bh);
1365 return ERR_PTR(err);
1366}
1367
1368/* Verify that we are loading a valid orphan from disk */
1369struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1370{
1371 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1372 ext4_group_t block_group;
1373 int bit;
1374 struct buffer_head *bitmap_bh = NULL;
1375 struct inode *inode = NULL;
1376 int err = -EFSCORRUPTED;
1377
1378 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1379 goto bad_orphan;
1380
1381 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1382 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1383 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1384 if (IS_ERR(bitmap_bh))
1385 return ERR_CAST(bitmap_bh);
1386
1387 /* Having the inode bit set should be a 100% indicator that this
1388 * is a valid orphan (no e2fsck run on fs). Orphans also include
1389 * inodes that were being truncated, so we can't check i_nlink==0.
1390 */
1391 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1392 goto bad_orphan;
1393
1394 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1395 if (IS_ERR(inode)) {
1396 err = PTR_ERR(inode);
1397 ext4_error_err(sb, -err,
1398 "couldn't read orphan inode %lu (err %d)",
1399 ino, err);
1400 brelse(bitmap_bh);
1401 return inode;
1402 }
1403
1404 /*
1405 * If the orphans has i_nlinks > 0 then it should be able to
1406 * be truncated, otherwise it won't be removed from the orphan
1407 * list during processing and an infinite loop will result.
1408 * Similarly, it must not be a bad inode.
1409 */
1410 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1411 is_bad_inode(inode))
1412 goto bad_orphan;
1413
1414 if (NEXT_ORPHAN(inode) > max_ino)
1415 goto bad_orphan;
1416 brelse(bitmap_bh);
1417 return inode;
1418
1419bad_orphan:
1420 ext4_error(sb, "bad orphan inode %lu", ino);
1421 if (bitmap_bh)
1422 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1423 bit, (unsigned long long)bitmap_bh->b_blocknr,
1424 ext4_test_bit(bit, bitmap_bh->b_data));
1425 if (inode) {
1426 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1427 is_bad_inode(inode));
1428 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1429 NEXT_ORPHAN(inode));
1430 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1431 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1432 /* Avoid freeing blocks if we got a bad deleted inode */
1433 if (inode->i_nlink == 0)
1434 inode->i_blocks = 0;
1435 iput(inode);
1436 }
1437 brelse(bitmap_bh);
1438 return ERR_PTR(err);
1439}
1440
1441unsigned long ext4_count_free_inodes(struct super_block *sb)
1442{
1443 unsigned long desc_count;
1444 struct ext4_group_desc *gdp;
1445 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1446#ifdef EXT4FS_DEBUG
1447 struct ext4_super_block *es;
1448 unsigned long bitmap_count, x;
1449 struct buffer_head *bitmap_bh = NULL;
1450
1451 es = EXT4_SB(sb)->s_es;
1452 desc_count = 0;
1453 bitmap_count = 0;
1454 gdp = NULL;
1455 for (i = 0; i < ngroups; i++) {
1456 gdp = ext4_get_group_desc(sb, i, NULL);
1457 if (!gdp)
1458 continue;
1459 desc_count += ext4_free_inodes_count(sb, gdp);
1460 brelse(bitmap_bh);
1461 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1462 if (IS_ERR(bitmap_bh)) {
1463 bitmap_bh = NULL;
1464 continue;
1465 }
1466
1467 x = ext4_count_free(bitmap_bh->b_data,
1468 EXT4_INODES_PER_GROUP(sb) / 8);
1469 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1470 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1471 bitmap_count += x;
1472 }
1473 brelse(bitmap_bh);
1474 printk(KERN_DEBUG "ext4_count_free_inodes: "
1475 "stored = %u, computed = %lu, %lu\n",
1476 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1477 return desc_count;
1478#else
1479 desc_count = 0;
1480 for (i = 0; i < ngroups; i++) {
1481 gdp = ext4_get_group_desc(sb, i, NULL);
1482 if (!gdp)
1483 continue;
1484 desc_count += ext4_free_inodes_count(sb, gdp);
1485 cond_resched();
1486 }
1487 return desc_count;
1488#endif
1489}
1490
1491/* Called at mount-time, super-block is locked */
1492unsigned long ext4_count_dirs(struct super_block * sb)
1493{
1494 unsigned long count = 0;
1495 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1496
1497 for (i = 0; i < ngroups; i++) {
1498 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1499 if (!gdp)
1500 continue;
1501 count += ext4_used_dirs_count(sb, gdp);
1502 }
1503 return count;
1504}
1505
1506/*
1507 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1508 * inode table. Must be called without any spinlock held. The only place
1509 * where it is called from on active part of filesystem is ext4lazyinit
1510 * thread, so we do not need any special locks, however we have to prevent
1511 * inode allocation from the current group, so we take alloc_sem lock, to
1512 * block ext4_new_inode() until we are finished.
1513 */
1514int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1515 int barrier)
1516{
1517 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1518 struct ext4_sb_info *sbi = EXT4_SB(sb);
1519 struct ext4_group_desc *gdp = NULL;
1520 struct buffer_head *group_desc_bh;
1521 handle_t *handle;
1522 ext4_fsblk_t blk;
1523 int num, ret = 0, used_blks = 0;
1524 unsigned long used_inos = 0;
1525
1526 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1527 if (!gdp || !grp)
1528 goto out;
1529
1530 /*
1531 * We do not need to lock this, because we are the only one
1532 * handling this flag.
1533 */
1534 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1535 goto out;
1536
1537 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1538 if (IS_ERR(handle)) {
1539 ret = PTR_ERR(handle);
1540 goto out;
1541 }
1542
1543 down_write(&grp->alloc_sem);
1544 /*
1545 * If inode bitmap was already initialized there may be some
1546 * used inodes so we need to skip blocks with used inodes in
1547 * inode table.
1548 */
1549 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1550 used_inos = EXT4_INODES_PER_GROUP(sb) -
1551 ext4_itable_unused_count(sb, gdp);
1552 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1553
1554 /* Bogus inode unused count? */
1555 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1556 ext4_error(sb, "Something is wrong with group %u: "
1557 "used itable blocks: %d; "
1558 "itable unused count: %u",
1559 group, used_blks,
1560 ext4_itable_unused_count(sb, gdp));
1561 ret = 1;
1562 goto err_out;
1563 }
1564
1565 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1566 /*
1567 * Are there some uninitialized inodes in the inode table
1568 * before the first normal inode?
1569 */
1570 if ((used_blks != sbi->s_itb_per_group) &&
1571 (used_inos < EXT4_FIRST_INO(sb))) {
1572 ext4_error(sb, "Something is wrong with group %u: "
1573 "itable unused count: %u; "
1574 "itables initialized count: %ld",
1575 group, ext4_itable_unused_count(sb, gdp),
1576 used_inos);
1577 ret = 1;
1578 goto err_out;
1579 }
1580 }
1581
1582 blk = ext4_inode_table(sb, gdp) + used_blks;
1583 num = sbi->s_itb_per_group - used_blks;
1584
1585 BUFFER_TRACE(group_desc_bh, "get_write_access");
1586 ret = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1587 EXT4_JTR_NONE);
1588 if (ret)
1589 goto err_out;
1590
1591 /*
1592 * Skip zeroout if the inode table is full. But we set the ZEROED
1593 * flag anyway, because obviously, when it is full it does not need
1594 * further zeroing.
1595 */
1596 if (unlikely(num == 0))
1597 goto skip_zeroout;
1598
1599 ext4_debug("going to zero out inode table in group %d\n",
1600 group);
1601 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1602 if (ret < 0)
1603 goto err_out;
1604 if (barrier)
1605 blkdev_issue_flush(sb->s_bdev);
1606
1607skip_zeroout:
1608 ext4_lock_group(sb, group);
1609 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1610 ext4_group_desc_csum_set(sb, group, gdp);
1611 ext4_unlock_group(sb, group);
1612
1613 BUFFER_TRACE(group_desc_bh,
1614 "call ext4_handle_dirty_metadata");
1615 ret = ext4_handle_dirty_metadata(handle, NULL,
1616 group_desc_bh);
1617
1618err_out:
1619 up_write(&grp->alloc_sem);
1620 ext4_journal_stop(handle);
1621out:
1622 return ret;
1623}