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 */
55 void 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
69 void 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
79 static 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);
112 verified:
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 */
123 static struct buffer_head *
124 ext4_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
209 verify:
210 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 if (err)
212 goto out;
213 return bh;
214 out:
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 */
235 void 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);
348 out:
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
360 error_return:
361 brelse(bitmap_bh);
362 ext4_std_error(sb, fatal);
363 }
364
365 struct 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 */
376 static 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
423 static 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 if (min_clusters < 0)
518 min_clusters = 0;
519
520 /*
521 * Start looking in the flex group where we last allocated an
522 * inode for this parent directory
523 */
524 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
525 parent_group = EXT4_I(parent)->i_last_alloc_group;
526 if (flex_size > 1)
527 parent_group >>= sbi->s_log_groups_per_flex;
528 }
529
530 for (i = 0; i < ngroups; i++) {
531 grp = (parent_group + i) % ngroups;
532 get_orlov_stats(sb, grp, flex_size, &stats);
533 if (stats.used_dirs >= max_dirs)
534 continue;
535 if (stats.free_inodes < min_inodes)
536 continue;
537 if (stats.free_clusters < min_clusters)
538 continue;
539 goto found_flex_bg;
540 }
541
542 fallback:
543 ngroups = real_ngroups;
544 avefreei = freei / ngroups;
545 fallback_retry:
546 parent_group = EXT4_I(parent)->i_block_group;
547 for (i = 0; i < ngroups; i++) {
548 grp = (parent_group + i) % ngroups;
549 desc = ext4_get_group_desc(sb, grp, NULL);
550 if (desc) {
551 grp_free = ext4_free_inodes_count(sb, desc);
552 if (grp_free && grp_free >= avefreei) {
553 *group = grp;
554 return 0;
555 }
556 }
557 }
558
559 if (avefreei) {
560 /*
561 * The free-inodes counter is approximate, and for really small
562 * filesystems the above test can fail to find any blockgroups
563 */
564 avefreei = 0;
565 goto fallback_retry;
566 }
567
568 return -1;
569 }
570
571 static int find_group_other(struct super_block *sb, struct inode *parent,
572 ext4_group_t *group, umode_t mode)
573 {
574 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
575 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
576 struct ext4_group_desc *desc;
577 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
578
579 /*
580 * Try to place the inode is the same flex group as its
581 * parent. If we can't find space, use the Orlov algorithm to
582 * find another flex group, and store that information in the
583 * parent directory's inode information so that use that flex
584 * group for future allocations.
585 */
586 if (flex_size > 1) {
587 int retry = 0;
588
589 try_again:
590 parent_group &= ~(flex_size-1);
591 last = parent_group + flex_size;
592 if (last > ngroups)
593 last = ngroups;
594 for (i = parent_group; i < last; i++) {
595 desc = ext4_get_group_desc(sb, i, NULL);
596 if (desc && ext4_free_inodes_count(sb, desc)) {
597 *group = i;
598 return 0;
599 }
600 }
601 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
602 retry = 1;
603 parent_group = EXT4_I(parent)->i_last_alloc_group;
604 goto try_again;
605 }
606 /*
607 * If this didn't work, use the Orlov search algorithm
608 * to find a new flex group; we pass in the mode to
609 * avoid the topdir algorithms.
610 */
611 *group = parent_group + flex_size;
612 if (*group > ngroups)
613 *group = 0;
614 return find_group_orlov(sb, parent, group, mode, NULL);
615 }
616
617 /*
618 * Try to place the inode in its parent directory
619 */
620 *group = parent_group;
621 desc = ext4_get_group_desc(sb, *group, NULL);
622 if (desc && ext4_free_inodes_count(sb, desc) &&
623 ext4_free_group_clusters(sb, desc))
624 return 0;
625
626 /*
627 * We're going to place this inode in a different blockgroup from its
628 * parent. We want to cause files in a common directory to all land in
629 * the same blockgroup. But we want files which are in a different
630 * directory which shares a blockgroup with our parent to land in a
631 * different blockgroup.
632 *
633 * So add our directory's i_ino into the starting point for the hash.
634 */
635 *group = (*group + parent->i_ino) % ngroups;
636
637 /*
638 * Use a quadratic hash to find a group with a free inode and some free
639 * blocks.
640 */
641 for (i = 1; i < ngroups; i <<= 1) {
642 *group += i;
643 if (*group >= ngroups)
644 *group -= ngroups;
645 desc = ext4_get_group_desc(sb, *group, NULL);
646 if (desc && ext4_free_inodes_count(sb, desc) &&
647 ext4_free_group_clusters(sb, desc))
648 return 0;
649 }
650
651 /*
652 * That failed: try linear search for a free inode, even if that group
653 * has no free blocks.
654 */
655 *group = parent_group;
656 for (i = 0; i < ngroups; i++) {
657 if (++*group >= ngroups)
658 *group = 0;
659 desc = ext4_get_group_desc(sb, *group, NULL);
660 if (desc && ext4_free_inodes_count(sb, desc))
661 return 0;
662 }
663
664 return -1;
665 }
666
667 /*
668 * In no journal mode, if an inode has recently been deleted, we want
669 * to avoid reusing it until we're reasonably sure the inode table
670 * block has been written back to disk. (Yes, these values are
671 * somewhat arbitrary...)
672 */
673 #define RECENTCY_MIN 60
674 #define RECENTCY_DIRTY 300
675
676 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
677 {
678 struct ext4_group_desc *gdp;
679 struct ext4_inode *raw_inode;
680 struct buffer_head *bh;
681 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
682 int offset, ret = 0;
683 int recentcy = RECENTCY_MIN;
684 u32 dtime, now;
685
686 gdp = ext4_get_group_desc(sb, group, NULL);
687 if (unlikely(!gdp))
688 return 0;
689
690 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
691 (ino / inodes_per_block));
692 if (!bh || !buffer_uptodate(bh))
693 /*
694 * If the block is not in the buffer cache, then it
695 * must have been written out.
696 */
697 goto out;
698
699 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
700 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
701
702 /* i_dtime is only 32 bits on disk, but we only care about relative
703 * times in the range of a few minutes (i.e. long enough to sync a
704 * recently-deleted inode to disk), so using the low 32 bits of the
705 * clock (a 68 year range) is enough, see time_before32() */
706 dtime = le32_to_cpu(raw_inode->i_dtime);
707 now = ktime_get_real_seconds();
708 if (buffer_dirty(bh))
709 recentcy += RECENTCY_DIRTY;
710
711 if (dtime && time_before32(dtime, now) &&
712 time_before32(now, dtime + recentcy))
713 ret = 1;
714 out:
715 brelse(bh);
716 return ret;
717 }
718
719 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
720 struct buffer_head *bitmap, unsigned long *ino)
721 {
722 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
723 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
724
725 next:
726 *ino = ext4_find_next_zero_bit((unsigned long *)
727 bitmap->b_data,
728 EXT4_INODES_PER_GROUP(sb), *ino);
729 if (*ino >= EXT4_INODES_PER_GROUP(sb))
730 goto not_found;
731
732 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
733 recently_deleted_ino = *ino;
734 *ino = *ino + 1;
735 if (*ino < EXT4_INODES_PER_GROUP(sb))
736 goto next;
737 goto not_found;
738 }
739 return 1;
740 not_found:
741 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
742 return 0;
743 /*
744 * Not reusing recently deleted inodes is mostly a preference. We don't
745 * want to report ENOSPC or skew allocation patterns because of that.
746 * So return even recently deleted inode if we could find better in the
747 * given range.
748 */
749 *ino = recently_deleted_ino;
750 return 1;
751 }
752
753 int ext4_mark_inode_used(struct super_block *sb, int ino)
754 {
755 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
756 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
757 struct ext4_group_desc *gdp;
758 ext4_group_t group;
759 int bit;
760 int err;
761
762 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
763 return -EFSCORRUPTED;
764
765 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
766 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
767 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
768 if (IS_ERR(inode_bitmap_bh))
769 return PTR_ERR(inode_bitmap_bh);
770
771 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
772 err = 0;
773 goto out;
774 }
775
776 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
777 if (!gdp || !group_desc_bh) {
778 err = -EINVAL;
779 goto out;
780 }
781
782 ext4_set_bit(bit, inode_bitmap_bh->b_data);
783
784 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
785 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
786 if (err) {
787 ext4_std_error(sb, err);
788 goto out;
789 }
790 err = sync_dirty_buffer(inode_bitmap_bh);
791 if (err) {
792 ext4_std_error(sb, err);
793 goto out;
794 }
795
796 /* We may have to initialize the block bitmap if it isn't already */
797 if (ext4_has_group_desc_csum(sb) &&
798 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
799 struct buffer_head *block_bitmap_bh;
800
801 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
802 if (IS_ERR(block_bitmap_bh)) {
803 err = PTR_ERR(block_bitmap_bh);
804 goto out;
805 }
806
807 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
808 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
809 sync_dirty_buffer(block_bitmap_bh);
810
811 /* recheck and clear flag under lock if we still need to */
812 ext4_lock_group(sb, group);
813 if (ext4_has_group_desc_csum(sb) &&
814 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
815 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
816 ext4_free_group_clusters_set(sb, gdp,
817 ext4_free_clusters_after_init(sb, group, gdp));
818 ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
819 ext4_group_desc_csum_set(sb, group, gdp);
820 }
821 ext4_unlock_group(sb, group);
822 brelse(block_bitmap_bh);
823
824 if (err) {
825 ext4_std_error(sb, err);
826 goto out;
827 }
828 }
829
830 /* Update the relevant bg descriptor fields */
831 if (ext4_has_group_desc_csum(sb)) {
832 int free;
833
834 ext4_lock_group(sb, group); /* while we modify the bg desc */
835 free = EXT4_INODES_PER_GROUP(sb) -
836 ext4_itable_unused_count(sb, gdp);
837 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
838 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
839 free = 0;
840 }
841
842 /*
843 * Check the relative inode number against the last used
844 * relative inode number in this group. if it is greater
845 * we need to update the bg_itable_unused count
846 */
847 if (bit >= free)
848 ext4_itable_unused_set(sb, gdp,
849 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
850 } else {
851 ext4_lock_group(sb, group);
852 }
853
854 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
855 if (ext4_has_group_desc_csum(sb)) {
856 ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh,
857 EXT4_INODES_PER_GROUP(sb) / 8);
858 ext4_group_desc_csum_set(sb, group, gdp);
859 }
860
861 ext4_unlock_group(sb, group);
862 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
863 sync_dirty_buffer(group_desc_bh);
864 out:
865 brelse(inode_bitmap_bh);
866 return err;
867 }
868
869 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
870 bool encrypt)
871 {
872 struct super_block *sb = dir->i_sb;
873 int nblocks = 0;
874 #ifdef CONFIG_EXT4_FS_POSIX_ACL
875 struct posix_acl *p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
876
877 if (IS_ERR(p))
878 return PTR_ERR(p);
879 if (p) {
880 int acl_size = p->a_count * sizeof(ext4_acl_entry);
881
882 nblocks += (S_ISDIR(mode) ? 2 : 1) *
883 __ext4_xattr_set_credits(sb, NULL /* inode */,
884 NULL /* block_bh */, acl_size,
885 true /* is_create */);
886 posix_acl_release(p);
887 }
888 #endif
889
890 #ifdef CONFIG_SECURITY
891 {
892 int num_security_xattrs = 1;
893
894 #ifdef CONFIG_INTEGRITY
895 num_security_xattrs++;
896 #endif
897 /*
898 * We assume that security xattrs are never more than 1k.
899 * In practice they are under 128 bytes.
900 */
901 nblocks += num_security_xattrs *
902 __ext4_xattr_set_credits(sb, NULL /* inode */,
903 NULL /* block_bh */, 1024,
904 true /* is_create */);
905 }
906 #endif
907 if (encrypt)
908 nblocks += __ext4_xattr_set_credits(sb,
909 NULL /* inode */,
910 NULL /* block_bh */,
911 FSCRYPT_SET_CONTEXT_MAX_SIZE,
912 true /* is_create */);
913 return nblocks;
914 }
915
916 /*
917 * There are two policies for allocating an inode. If the new inode is
918 * a directory, then a forward search is made for a block group with both
919 * free space and a low directory-to-inode ratio; if that fails, then of
920 * the groups with above-average free space, that group with the fewest
921 * directories already is chosen.
922 *
923 * For other inodes, search forward from the parent directory's block
924 * group to find a free inode.
925 */
926 struct inode *__ext4_new_inode(struct mnt_idmap *idmap,
927 handle_t *handle, struct inode *dir,
928 umode_t mode, const struct qstr *qstr,
929 __u32 goal, uid_t *owner, __u32 i_flags,
930 int handle_type, unsigned int line_no,
931 int nblocks)
932 {
933 struct super_block *sb;
934 struct buffer_head *inode_bitmap_bh = NULL;
935 struct buffer_head *group_desc_bh;
936 ext4_group_t ngroups, group = 0;
937 unsigned long ino = 0;
938 struct inode *inode;
939 struct ext4_group_desc *gdp = NULL;
940 struct ext4_inode_info *ei;
941 struct ext4_sb_info *sbi;
942 int ret2, err;
943 struct inode *ret;
944 ext4_group_t i;
945 ext4_group_t flex_group;
946 struct ext4_group_info *grp = NULL;
947 bool encrypt = false;
948
949 /* Cannot create files in a deleted directory */
950 if (!dir || !dir->i_nlink)
951 return ERR_PTR(-EPERM);
952
953 sb = dir->i_sb;
954 sbi = EXT4_SB(sb);
955
956 if (unlikely(ext4_forced_shutdown(sb)))
957 return ERR_PTR(-EIO);
958
959 ngroups = ext4_get_groups_count(sb);
960 trace_ext4_request_inode(dir, mode);
961 inode = new_inode(sb);
962 if (!inode)
963 return ERR_PTR(-ENOMEM);
964 ei = EXT4_I(inode);
965
966 /*
967 * Initialize owners and quota early so that we don't have to account
968 * for quota initialization worst case in standard inode creating
969 * transaction
970 */
971 if (owner) {
972 inode->i_mode = mode;
973 i_uid_write(inode, owner[0]);
974 i_gid_write(inode, owner[1]);
975 } else if (test_opt(sb, GRPID)) {
976 inode->i_mode = mode;
977 inode_fsuid_set(inode, idmap);
978 inode->i_gid = dir->i_gid;
979 } else
980 inode_init_owner(idmap, inode, dir, mode);
981
982 if (ext4_has_feature_project(sb) &&
983 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
984 ei->i_projid = EXT4_I(dir)->i_projid;
985 else
986 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
987
988 if (!(i_flags & EXT4_EA_INODE_FL)) {
989 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
990 if (err)
991 goto out;
992 }
993
994 err = dquot_initialize(inode);
995 if (err)
996 goto out;
997
998 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
999 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
1000 if (ret2 < 0) {
1001 err = ret2;
1002 goto out;
1003 }
1004 nblocks += ret2;
1005 }
1006
1007 if (!goal)
1008 goal = sbi->s_inode_goal;
1009
1010 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1011 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1012 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1013 ret2 = 0;
1014 goto got_group;
1015 }
1016
1017 if (S_ISDIR(mode))
1018 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1019 else
1020 ret2 = find_group_other(sb, dir, &group, mode);
1021
1022 got_group:
1023 EXT4_I(dir)->i_last_alloc_group = group;
1024 err = -ENOSPC;
1025 if (ret2 == -1)
1026 goto out;
1027
1028 /*
1029 * Normally we will only go through one pass of this loop,
1030 * unless we get unlucky and it turns out the group we selected
1031 * had its last inode grabbed by someone else.
1032 */
1033 for (i = 0; i < ngroups; i++, ino = 0) {
1034 err = -EIO;
1035
1036 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1037 if (!gdp)
1038 goto out;
1039
1040 /*
1041 * Check free inodes count before loading bitmap.
1042 */
1043 if (ext4_free_inodes_count(sb, gdp) == 0)
1044 goto next_group;
1045
1046 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1047 grp = ext4_get_group_info(sb, group);
1048 /*
1049 * Skip groups with already-known suspicious inode
1050 * tables
1051 */
1052 if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1053 goto next_group;
1054 }
1055
1056 brelse(inode_bitmap_bh);
1057 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1058 /* Skip groups with suspicious inode tables */
1059 if (IS_ERR(inode_bitmap_bh)) {
1060 inode_bitmap_bh = NULL;
1061 goto next_group;
1062 }
1063 if (!(sbi->s_mount_state & EXT4_FC_REPLAY) &&
1064 EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1065 goto next_group;
1066
1067 repeat_in_this_group:
1068 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1069 if (!ret2)
1070 goto next_group;
1071
1072 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1073 ext4_error(sb, "reserved inode found cleared - "
1074 "inode=%lu", ino + 1);
1075 ext4_mark_group_bitmap_corrupted(sb, group,
1076 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1077 goto next_group;
1078 }
1079
1080 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1081 BUG_ON(nblocks <= 0);
1082 handle = __ext4_journal_start_sb(NULL, dir->i_sb,
1083 line_no, handle_type, nblocks, 0,
1084 ext4_trans_default_revoke_credits(sb));
1085 if (IS_ERR(handle)) {
1086 err = PTR_ERR(handle);
1087 ext4_std_error(sb, err);
1088 goto out;
1089 }
1090 }
1091 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1092 err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,
1093 EXT4_JTR_NONE);
1094 if (err) {
1095 ext4_std_error(sb, err);
1096 goto out;
1097 }
1098 ext4_lock_group(sb, group);
1099 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1100 if (ret2) {
1101 /* Someone already took the bit. Repeat the search
1102 * with lock held.
1103 */
1104 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1105 if (ret2) {
1106 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1107 ret2 = 0;
1108 } else {
1109 ret2 = 1; /* we didn't grab the inode */
1110 }
1111 }
1112 ext4_unlock_group(sb, group);
1113 ino++; /* the inode bitmap is zero-based */
1114 if (!ret2)
1115 goto got; /* we grabbed the inode! */
1116
1117 if (ino < EXT4_INODES_PER_GROUP(sb))
1118 goto repeat_in_this_group;
1119 next_group:
1120 if (++group == ngroups)
1121 group = 0;
1122 }
1123 err = -ENOSPC;
1124 goto out;
1125
1126 got:
1127 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1128 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1129 if (err) {
1130 ext4_std_error(sb, err);
1131 goto out;
1132 }
1133
1134 BUFFER_TRACE(group_desc_bh, "get_write_access");
1135 err = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1136 EXT4_JTR_NONE);
1137 if (err) {
1138 ext4_std_error(sb, err);
1139 goto out;
1140 }
1141
1142 /* We may have to initialize the block bitmap if it isn't already */
1143 if (ext4_has_group_desc_csum(sb) &&
1144 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1145 struct buffer_head *block_bitmap_bh;
1146
1147 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1148 if (IS_ERR(block_bitmap_bh)) {
1149 err = PTR_ERR(block_bitmap_bh);
1150 goto out;
1151 }
1152 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1153 err = ext4_journal_get_write_access(handle, sb, block_bitmap_bh,
1154 EXT4_JTR_NONE);
1155 if (err) {
1156 brelse(block_bitmap_bh);
1157 ext4_std_error(sb, err);
1158 goto out;
1159 }
1160
1161 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1162 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1163
1164 /* recheck and clear flag under lock if we still need to */
1165 ext4_lock_group(sb, group);
1166 if (ext4_has_group_desc_csum(sb) &&
1167 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1168 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1169 ext4_free_group_clusters_set(sb, gdp,
1170 ext4_free_clusters_after_init(sb, group, gdp));
1171 ext4_block_bitmap_csum_set(sb, gdp, block_bitmap_bh);
1172 ext4_group_desc_csum_set(sb, group, gdp);
1173 }
1174 ext4_unlock_group(sb, group);
1175 brelse(block_bitmap_bh);
1176
1177 if (err) {
1178 ext4_std_error(sb, err);
1179 goto out;
1180 }
1181 }
1182
1183 /* Update the relevant bg descriptor fields */
1184 if (ext4_has_group_desc_csum(sb)) {
1185 int free;
1186 struct ext4_group_info *grp = NULL;
1187
1188 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1189 grp = ext4_get_group_info(sb, group);
1190 if (!grp) {
1191 err = -EFSCORRUPTED;
1192 goto out;
1193 }
1194 down_read(&grp->alloc_sem); /*
1195 * protect vs itable
1196 * lazyinit
1197 */
1198 }
1199 ext4_lock_group(sb, group); /* while we modify the bg desc */
1200 free = EXT4_INODES_PER_GROUP(sb) -
1201 ext4_itable_unused_count(sb, gdp);
1202 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1203 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1204 free = 0;
1205 }
1206 /*
1207 * Check the relative inode number against the last used
1208 * relative inode number in this group. if it is greater
1209 * we need to update the bg_itable_unused count
1210 */
1211 if (ino > free)
1212 ext4_itable_unused_set(sb, gdp,
1213 (EXT4_INODES_PER_GROUP(sb) - ino));
1214 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1215 up_read(&grp->alloc_sem);
1216 } else {
1217 ext4_lock_group(sb, group);
1218 }
1219
1220 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1221 if (S_ISDIR(mode)) {
1222 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1223 if (sbi->s_log_groups_per_flex) {
1224 ext4_group_t f = ext4_flex_group(sbi, group);
1225
1226 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1227 f)->used_dirs);
1228 }
1229 }
1230 if (ext4_has_group_desc_csum(sb)) {
1231 ext4_inode_bitmap_csum_set(sb, gdp, inode_bitmap_bh,
1232 EXT4_INODES_PER_GROUP(sb) / 8);
1233 ext4_group_desc_csum_set(sb, group, gdp);
1234 }
1235 ext4_unlock_group(sb, group);
1236
1237 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1238 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1239 if (err) {
1240 ext4_std_error(sb, err);
1241 goto out;
1242 }
1243
1244 percpu_counter_dec(&sbi->s_freeinodes_counter);
1245 if (S_ISDIR(mode))
1246 percpu_counter_inc(&sbi->s_dirs_counter);
1247
1248 if (sbi->s_log_groups_per_flex) {
1249 flex_group = ext4_flex_group(sbi, group);
1250 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1251 flex_group)->free_inodes);
1252 }
1253
1254 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1255 /* This is the optimal IO size (for stat), not the fs block size */
1256 inode->i_blocks = 0;
1257 simple_inode_init_ts(inode);
1258 ei->i_crtime = inode_get_mtime(inode);
1259
1260 memset(ei->i_data, 0, sizeof(ei->i_data));
1261 ei->i_dir_start_lookup = 0;
1262 ei->i_disksize = 0;
1263
1264 /* Don't inherit extent flag from directory, amongst others. */
1265 ei->i_flags =
1266 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1267 ei->i_flags |= i_flags;
1268 ei->i_file_acl = 0;
1269 ei->i_dtime = 0;
1270 ei->i_block_group = group;
1271 ei->i_last_alloc_group = ~0;
1272
1273 ext4_set_inode_flags(inode, true);
1274 if (IS_DIRSYNC(inode))
1275 ext4_handle_sync(handle);
1276 if (insert_inode_locked(inode) < 0) {
1277 /*
1278 * Likely a bitmap corruption causing inode to be allocated
1279 * twice.
1280 */
1281 err = -EIO;
1282 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1283 inode->i_ino);
1284 ext4_mark_group_bitmap_corrupted(sb, group,
1285 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1286 goto out;
1287 }
1288 inode->i_generation = get_random_u32();
1289
1290 /* Precompute checksum seed for inode metadata */
1291 if (ext4_has_metadata_csum(sb)) {
1292 __u32 csum;
1293 __le32 inum = cpu_to_le32(inode->i_ino);
1294 __le32 gen = cpu_to_le32(inode->i_generation);
1295 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1296 sizeof(inum));
1297 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1298 sizeof(gen));
1299 }
1300
1301 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1302 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1303
1304 ei->i_extra_isize = sbi->s_want_extra_isize;
1305 ei->i_inline_off = 0;
1306 if (ext4_has_feature_inline_data(sb) &&
1307 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1308 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1309 ret = inode;
1310 err = dquot_alloc_inode(inode);
1311 if (err)
1312 goto fail_drop;
1313
1314 /*
1315 * Since the encryption xattr will always be unique, create it first so
1316 * that it's less likely to end up in an external xattr block and
1317 * prevent its deduplication.
1318 */
1319 if (encrypt) {
1320 err = fscrypt_set_context(inode, handle);
1321 if (err)
1322 goto fail_free_drop;
1323 }
1324
1325 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1326 err = ext4_init_acl(handle, inode, dir);
1327 if (err)
1328 goto fail_free_drop;
1329
1330 err = ext4_init_security(handle, inode, dir, qstr);
1331 if (err)
1332 goto fail_free_drop;
1333 }
1334
1335 if (ext4_has_feature_extents(sb)) {
1336 /* set extent flag only for directory, file and normal symlink*/
1337 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1338 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1339 ext4_ext_tree_init(handle, inode);
1340 }
1341 }
1342
1343 ext4_update_inode_fsync_trans(handle, inode, 1);
1344
1345 err = ext4_mark_inode_dirty(handle, inode);
1346 if (err) {
1347 ext4_std_error(sb, err);
1348 goto fail_free_drop;
1349 }
1350
1351 ext4_debug("allocating inode %lu\n", inode->i_ino);
1352 trace_ext4_allocate_inode(inode, dir, mode);
1353 brelse(inode_bitmap_bh);
1354 return ret;
1355
1356 fail_free_drop:
1357 dquot_free_inode(inode);
1358 fail_drop:
1359 clear_nlink(inode);
1360 unlock_new_inode(inode);
1361 out:
1362 dquot_drop(inode);
1363 inode->i_flags |= S_NOQUOTA;
1364 iput(inode);
1365 brelse(inode_bitmap_bh);
1366 return ERR_PTR(err);
1367 }
1368
1369 /* Verify that we are loading a valid orphan from disk */
1370 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1371 {
1372 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1373 ext4_group_t block_group;
1374 int bit;
1375 struct buffer_head *bitmap_bh = NULL;
1376 struct inode *inode = NULL;
1377 int err = -EFSCORRUPTED;
1378
1379 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1380 goto bad_orphan;
1381
1382 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1383 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1384 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1385 if (IS_ERR(bitmap_bh))
1386 return ERR_CAST(bitmap_bh);
1387
1388 /* Having the inode bit set should be a 100% indicator that this
1389 * is a valid orphan (no e2fsck run on fs). Orphans also include
1390 * inodes that were being truncated, so we can't check i_nlink==0.
1391 */
1392 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1393 goto bad_orphan;
1394
1395 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1396 if (IS_ERR(inode)) {
1397 err = PTR_ERR(inode);
1398 ext4_error_err(sb, -err,
1399 "couldn't read orphan inode %lu (err %d)",
1400 ino, err);
1401 brelse(bitmap_bh);
1402 return inode;
1403 }
1404
1405 /*
1406 * If the orphans has i_nlinks > 0 then it should be able to
1407 * be truncated, otherwise it won't be removed from the orphan
1408 * list during processing and an infinite loop will result.
1409 * Similarly, it must not be a bad inode.
1410 */
1411 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1412 is_bad_inode(inode))
1413 goto bad_orphan;
1414
1415 if (NEXT_ORPHAN(inode) > max_ino)
1416 goto bad_orphan;
1417 brelse(bitmap_bh);
1418 return inode;
1419
1420 bad_orphan:
1421 ext4_error(sb, "bad orphan inode %lu", ino);
1422 if (bitmap_bh)
1423 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1424 bit, (unsigned long long)bitmap_bh->b_blocknr,
1425 ext4_test_bit(bit, bitmap_bh->b_data));
1426 if (inode) {
1427 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1428 is_bad_inode(inode));
1429 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1430 NEXT_ORPHAN(inode));
1431 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1432 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1433 /* Avoid freeing blocks if we got a bad deleted inode */
1434 if (inode->i_nlink == 0)
1435 inode->i_blocks = 0;
1436 iput(inode);
1437 }
1438 brelse(bitmap_bh);
1439 return ERR_PTR(err);
1440 }
1441
1442 unsigned long ext4_count_free_inodes(struct super_block *sb)
1443 {
1444 unsigned long desc_count;
1445 struct ext4_group_desc *gdp;
1446 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1447 #ifdef EXT4FS_DEBUG
1448 struct ext4_super_block *es;
1449 unsigned long bitmap_count, x;
1450 struct buffer_head *bitmap_bh = NULL;
1451
1452 es = EXT4_SB(sb)->s_es;
1453 desc_count = 0;
1454 bitmap_count = 0;
1455 gdp = NULL;
1456 for (i = 0; i < ngroups; i++) {
1457 gdp = ext4_get_group_desc(sb, i, NULL);
1458 if (!gdp)
1459 continue;
1460 desc_count += ext4_free_inodes_count(sb, gdp);
1461 brelse(bitmap_bh);
1462 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1463 if (IS_ERR(bitmap_bh)) {
1464 bitmap_bh = NULL;
1465 continue;
1466 }
1467
1468 x = ext4_count_free(bitmap_bh->b_data,
1469 EXT4_INODES_PER_GROUP(sb) / 8);
1470 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1471 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1472 bitmap_count += x;
1473 }
1474 brelse(bitmap_bh);
1475 printk(KERN_DEBUG "ext4_count_free_inodes: "
1476 "stored = %u, computed = %lu, %lu\n",
1477 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1478 return desc_count;
1479 #else
1480 desc_count = 0;
1481 for (i = 0; i < ngroups; i++) {
1482 gdp = ext4_get_group_desc(sb, i, NULL);
1483 if (!gdp)
1484 continue;
1485 desc_count += ext4_free_inodes_count(sb, gdp);
1486 cond_resched();
1487 }
1488 return desc_count;
1489 #endif
1490 }
1491
1492 /* Called at mount-time, super-block is locked */
1493 unsigned long ext4_count_dirs(struct super_block * sb)
1494 {
1495 unsigned long count = 0;
1496 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1497
1498 for (i = 0; i < ngroups; i++) {
1499 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1500 if (!gdp)
1501 continue;
1502 count += ext4_used_dirs_count(sb, gdp);
1503 }
1504 return count;
1505 }
1506
1507 /*
1508 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1509 * inode table. Must be called without any spinlock held. The only place
1510 * where it is called from on active part of filesystem is ext4lazyinit
1511 * thread, so we do not need any special locks, however we have to prevent
1512 * inode allocation from the current group, so we take alloc_sem lock, to
1513 * block ext4_new_inode() until we are finished.
1514 */
1515 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1516 int barrier)
1517 {
1518 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1519 struct ext4_sb_info *sbi = EXT4_SB(sb);
1520 struct ext4_group_desc *gdp = NULL;
1521 struct buffer_head *group_desc_bh;
1522 handle_t *handle;
1523 ext4_fsblk_t blk;
1524 int num, ret = 0, used_blks = 0;
1525 unsigned long used_inos = 0;
1526
1527 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1528 if (!gdp || !grp)
1529 goto out;
1530
1531 /*
1532 * We do not need to lock this, because we are the only one
1533 * handling this flag.
1534 */
1535 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1536 goto out;
1537
1538 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1539 if (IS_ERR(handle)) {
1540 ret = PTR_ERR(handle);
1541 goto out;
1542 }
1543
1544 down_write(&grp->alloc_sem);
1545 /*
1546 * If inode bitmap was already initialized there may be some
1547 * used inodes so we need to skip blocks with used inodes in
1548 * inode table.
1549 */
1550 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1551 used_inos = EXT4_INODES_PER_GROUP(sb) -
1552 ext4_itable_unused_count(sb, gdp);
1553 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1554
1555 /* Bogus inode unused count? */
1556 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1557 ext4_error(sb, "Something is wrong with group %u: "
1558 "used itable blocks: %d; "
1559 "itable unused count: %u",
1560 group, used_blks,
1561 ext4_itable_unused_count(sb, gdp));
1562 ret = 1;
1563 goto err_out;
1564 }
1565
1566 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1567 /*
1568 * Are there some uninitialized inodes in the inode table
1569 * before the first normal inode?
1570 */
1571 if ((used_blks != sbi->s_itb_per_group) &&
1572 (used_inos < EXT4_FIRST_INO(sb))) {
1573 ext4_error(sb, "Something is wrong with group %u: "
1574 "itable unused count: %u; "
1575 "itables initialized count: %ld",
1576 group, ext4_itable_unused_count(sb, gdp),
1577 used_inos);
1578 ret = 1;
1579 goto err_out;
1580 }
1581 }
1582
1583 blk = ext4_inode_table(sb, gdp) + used_blks;
1584 num = sbi->s_itb_per_group - used_blks;
1585
1586 BUFFER_TRACE(group_desc_bh, "get_write_access");
1587 ret = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1588 EXT4_JTR_NONE);
1589 if (ret)
1590 goto err_out;
1591
1592 /*
1593 * Skip zeroout if the inode table is full. But we set the ZEROED
1594 * flag anyway, because obviously, when it is full it does not need
1595 * further zeroing.
1596 */
1597 if (unlikely(num == 0))
1598 goto skip_zeroout;
1599
1600 ext4_debug("going to zero out inode table in group %d\n",
1601 group);
1602 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1603 if (ret < 0)
1604 goto err_out;
1605 if (barrier)
1606 blkdev_issue_flush(sb->s_bdev);
1607
1608 skip_zeroout:
1609 ext4_lock_group(sb, group);
1610 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1611 ext4_group_desc_csum_set(sb, group, gdp);
1612 ext4_unlock_group(sb, group);
1613
1614 BUFFER_TRACE(group_desc_bh,
1615 "call ext4_handle_dirty_metadata");
1616 ret = ext4_handle_dirty_metadata(handle, NULL,
1617 group_desc_bh);
1618
1619 err_out:
1620 up_write(&grp->alloc_sem);
1621 ext4_journal_stop(handle);
1622 out:
1623 return ret;
1624 }
1625
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.