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