]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/ext4/ialloc.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'ovl-update-5.13' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs
[mirror_ubuntu-jammy-kernel.git] / fs / ext4 / ialloc.c
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 (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, block_group, 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 (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, bitmap_bh);
304 if (fatal)
305 goto error_return;
306
307 fatal = -ESRCH;
308 gdp = ext4_get_group_desc(sb, block_group, &bh2);
309 if (gdp) {
310 BUFFER_TRACE(bh2, "get_write_access");
311 fatal = ext4_journal_get_write_access(handle, bh2);
312 }
313 ext4_lock_group(sb, block_group);
314 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
315 if (fatal || !cleared) {
316 ext4_unlock_group(sb, block_group);
317 goto out;
318 }
319
320 count = ext4_free_inodes_count(sb, gdp) + 1;
321 ext4_free_inodes_set(sb, gdp, count);
322 if (is_directory) {
323 count = ext4_used_dirs_count(sb, gdp) - 1;
324 ext4_used_dirs_set(sb, gdp, count);
325 percpu_counter_dec(&sbi->s_dirs_counter);
326 }
327 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
328 EXT4_INODES_PER_GROUP(sb) / 8);
329 ext4_group_desc_csum_set(sb, block_group, gdp);
330 ext4_unlock_group(sb, block_group);
331
332 percpu_counter_inc(&sbi->s_freeinodes_counter);
333 if (sbi->s_log_groups_per_flex) {
334 struct flex_groups *fg;
335
336 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
337 ext4_flex_group(sbi, block_group));
338 atomic_inc(&fg->free_inodes);
339 if (is_directory)
340 atomic_dec(&fg->used_dirs);
341 }
342 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
343 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
344 out:
345 if (cleared) {
346 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
347 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
348 if (!fatal)
349 fatal = err;
350 } else {
351 ext4_error(sb, "bit already cleared for inode %lu", ino);
352 ext4_mark_group_bitmap_corrupted(sb, block_group,
353 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
354 }
355
356 error_return:
357 brelse(bitmap_bh);
358 ext4_std_error(sb, fatal);
359 }
360
361 struct orlov_stats {
362 __u64 free_clusters;
363 __u32 free_inodes;
364 __u32 used_dirs;
365 };
366
367 /*
368 * Helper function for Orlov's allocator; returns critical information
369 * for a particular block group or flex_bg. If flex_size is 1, then g
370 * is a block group number; otherwise it is flex_bg number.
371 */
372 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
373 int flex_size, struct orlov_stats *stats)
374 {
375 struct ext4_group_desc *desc;
376
377 if (flex_size > 1) {
378 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
379 s_flex_groups, g);
380 stats->free_inodes = atomic_read(&fg->free_inodes);
381 stats->free_clusters = atomic64_read(&fg->free_clusters);
382 stats->used_dirs = atomic_read(&fg->used_dirs);
383 return;
384 }
385
386 desc = ext4_get_group_desc(sb, g, NULL);
387 if (desc) {
388 stats->free_inodes = ext4_free_inodes_count(sb, desc);
389 stats->free_clusters = ext4_free_group_clusters(sb, desc);
390 stats->used_dirs = ext4_used_dirs_count(sb, desc);
391 } else {
392 stats->free_inodes = 0;
393 stats->free_clusters = 0;
394 stats->used_dirs = 0;
395 }
396 }
397
398 /*
399 * Orlov's allocator for directories.
400 *
401 * We always try to spread first-level directories.
402 *
403 * If there are blockgroups with both free inodes and free blocks counts
404 * not worse than average we return one with smallest directory count.
405 * Otherwise we simply return a random group.
406 *
407 * For the rest rules look so:
408 *
409 * It's OK to put directory into a group unless
410 * it has too many directories already (max_dirs) or
411 * it has too few free inodes left (min_inodes) or
412 * it has too few free blocks left (min_blocks) or
413 * Parent's group is preferred, if it doesn't satisfy these
414 * conditions we search cyclically through the rest. If none
415 * of the groups look good we just look for a group with more
416 * free inodes than average (starting at parent's group).
417 */
418
419 static int find_group_orlov(struct super_block *sb, struct inode *parent,
420 ext4_group_t *group, umode_t mode,
421 const struct qstr *qstr)
422 {
423 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
424 struct ext4_sb_info *sbi = EXT4_SB(sb);
425 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
426 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
427 unsigned int freei, avefreei, grp_free;
428 ext4_fsblk_t freeb, avefreec;
429 unsigned int ndirs;
430 int max_dirs, min_inodes;
431 ext4_grpblk_t min_clusters;
432 ext4_group_t i, grp, g, ngroups;
433 struct ext4_group_desc *desc;
434 struct orlov_stats stats;
435 int flex_size = ext4_flex_bg_size(sbi);
436 struct dx_hash_info hinfo;
437
438 ngroups = real_ngroups;
439 if (flex_size > 1) {
440 ngroups = (real_ngroups + flex_size - 1) >>
441 sbi->s_log_groups_per_flex;
442 parent_group >>= sbi->s_log_groups_per_flex;
443 }
444
445 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
446 avefreei = freei / ngroups;
447 freeb = EXT4_C2B(sbi,
448 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
449 avefreec = freeb;
450 do_div(avefreec, ngroups);
451 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
452
453 if (S_ISDIR(mode) &&
454 ((parent == d_inode(sb->s_root)) ||
455 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
456 int best_ndir = inodes_per_group;
457 int ret = -1;
458
459 if (qstr) {
460 hinfo.hash_version = DX_HASH_HALF_MD4;
461 hinfo.seed = sbi->s_hash_seed;
462 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
463 grp = hinfo.hash;
464 } else
465 grp = prandom_u32();
466 parent_group = (unsigned)grp % ngroups;
467 for (i = 0; i < ngroups; i++) {
468 g = (parent_group + i) % ngroups;
469 get_orlov_stats(sb, g, flex_size, &stats);
470 if (!stats.free_inodes)
471 continue;
472 if (stats.used_dirs >= best_ndir)
473 continue;
474 if (stats.free_inodes < avefreei)
475 continue;
476 if (stats.free_clusters < avefreec)
477 continue;
478 grp = g;
479 ret = 0;
480 best_ndir = stats.used_dirs;
481 }
482 if (ret)
483 goto fallback;
484 found_flex_bg:
485 if (flex_size == 1) {
486 *group = grp;
487 return 0;
488 }
489
490 /*
491 * We pack inodes at the beginning of the flexgroup's
492 * inode tables. Block allocation decisions will do
493 * something similar, although regular files will
494 * start at 2nd block group of the flexgroup. See
495 * ext4_ext_find_goal() and ext4_find_near().
496 */
497 grp *= flex_size;
498 for (i = 0; i < flex_size; i++) {
499 if (grp+i >= real_ngroups)
500 break;
501 desc = ext4_get_group_desc(sb, grp+i, NULL);
502 if (desc && ext4_free_inodes_count(sb, desc)) {
503 *group = grp+i;
504 return 0;
505 }
506 }
507 goto fallback;
508 }
509
510 max_dirs = ndirs / ngroups + inodes_per_group / 16;
511 min_inodes = avefreei - inodes_per_group*flex_size / 4;
512 if (min_inodes < 1)
513 min_inodes = 1;
514 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
515
516 /*
517 * Start looking in the flex group where we last allocated an
518 * inode for this parent directory
519 */
520 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
521 parent_group = EXT4_I(parent)->i_last_alloc_group;
522 if (flex_size > 1)
523 parent_group >>= sbi->s_log_groups_per_flex;
524 }
525
526 for (i = 0; i < ngroups; i++) {
527 grp = (parent_group + i) % ngroups;
528 get_orlov_stats(sb, grp, flex_size, &stats);
529 if (stats.used_dirs >= max_dirs)
530 continue;
531 if (stats.free_inodes < min_inodes)
532 continue;
533 if (stats.free_clusters < min_clusters)
534 continue;
535 goto found_flex_bg;
536 }
537
538 fallback:
539 ngroups = real_ngroups;
540 avefreei = freei / ngroups;
541 fallback_retry:
542 parent_group = EXT4_I(parent)->i_block_group;
543 for (i = 0; i < ngroups; i++) {
544 grp = (parent_group + i) % ngroups;
545 desc = ext4_get_group_desc(sb, grp, NULL);
546 if (desc) {
547 grp_free = ext4_free_inodes_count(sb, desc);
548 if (grp_free && grp_free >= avefreei) {
549 *group = grp;
550 return 0;
551 }
552 }
553 }
554
555 if (avefreei) {
556 /*
557 * The free-inodes counter is approximate, and for really small
558 * filesystems the above test can fail to find any blockgroups
559 */
560 avefreei = 0;
561 goto fallback_retry;
562 }
563
564 return -1;
565 }
566
567 static int find_group_other(struct super_block *sb, struct inode *parent,
568 ext4_group_t *group, umode_t mode)
569 {
570 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
571 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
572 struct ext4_group_desc *desc;
573 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
574
575 /*
576 * Try to place the inode is the same flex group as its
577 * parent. If we can't find space, use the Orlov algorithm to
578 * find another flex group, and store that information in the
579 * parent directory's inode information so that use that flex
580 * group for future allocations.
581 */
582 if (flex_size > 1) {
583 int retry = 0;
584
585 try_again:
586 parent_group &= ~(flex_size-1);
587 last = parent_group + flex_size;
588 if (last > ngroups)
589 last = ngroups;
590 for (i = parent_group; i < last; i++) {
591 desc = ext4_get_group_desc(sb, i, NULL);
592 if (desc && ext4_free_inodes_count(sb, desc)) {
593 *group = i;
594 return 0;
595 }
596 }
597 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
598 retry = 1;
599 parent_group = EXT4_I(parent)->i_last_alloc_group;
600 goto try_again;
601 }
602 /*
603 * If this didn't work, use the Orlov search algorithm
604 * to find a new flex group; we pass in the mode to
605 * avoid the topdir algorithms.
606 */
607 *group = parent_group + flex_size;
608 if (*group > ngroups)
609 *group = 0;
610 return find_group_orlov(sb, parent, group, mode, NULL);
611 }
612
613 /*
614 * Try to place the inode in its parent directory
615 */
616 *group = parent_group;
617 desc = ext4_get_group_desc(sb, *group, NULL);
618 if (desc && ext4_free_inodes_count(sb, desc) &&
619 ext4_free_group_clusters(sb, desc))
620 return 0;
621
622 /*
623 * We're going to place this inode in a different blockgroup from its
624 * parent. We want to cause files in a common directory to all land in
625 * the same blockgroup. But we want files which are in a different
626 * directory which shares a blockgroup with our parent to land in a
627 * different blockgroup.
628 *
629 * So add our directory's i_ino into the starting point for the hash.
630 */
631 *group = (*group + parent->i_ino) % ngroups;
632
633 /*
634 * Use a quadratic hash to find a group with a free inode and some free
635 * blocks.
636 */
637 for (i = 1; i < ngroups; i <<= 1) {
638 *group += i;
639 if (*group >= ngroups)
640 *group -= ngroups;
641 desc = ext4_get_group_desc(sb, *group, NULL);
642 if (desc && ext4_free_inodes_count(sb, desc) &&
643 ext4_free_group_clusters(sb, desc))
644 return 0;
645 }
646
647 /*
648 * That failed: try linear search for a free inode, even if that group
649 * has no free blocks.
650 */
651 *group = parent_group;
652 for (i = 0; i < ngroups; i++) {
653 if (++*group >= ngroups)
654 *group = 0;
655 desc = ext4_get_group_desc(sb, *group, NULL);
656 if (desc && ext4_free_inodes_count(sb, desc))
657 return 0;
658 }
659
660 return -1;
661 }
662
663 /*
664 * In no journal mode, if an inode has recently been deleted, we want
665 * to avoid reusing it until we're reasonably sure the inode table
666 * block has been written back to disk. (Yes, these values are
667 * somewhat arbitrary...)
668 */
669 #define RECENTCY_MIN 60
670 #define RECENTCY_DIRTY 300
671
672 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
673 {
674 struct ext4_group_desc *gdp;
675 struct ext4_inode *raw_inode;
676 struct buffer_head *bh;
677 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
678 int offset, ret = 0;
679 int recentcy = RECENTCY_MIN;
680 u32 dtime, now;
681
682 gdp = ext4_get_group_desc(sb, group, NULL);
683 if (unlikely(!gdp))
684 return 0;
685
686 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
687 (ino / inodes_per_block));
688 if (!bh || !buffer_uptodate(bh))
689 /*
690 * If the block is not in the buffer cache, then it
691 * must have been written out.
692 */
693 goto out;
694
695 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
696 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
697
698 /* i_dtime is only 32 bits on disk, but we only care about relative
699 * times in the range of a few minutes (i.e. long enough to sync a
700 * recently-deleted inode to disk), so using the low 32 bits of the
701 * clock (a 68 year range) is enough, see time_before32() */
702 dtime = le32_to_cpu(raw_inode->i_dtime);
703 now = ktime_get_real_seconds();
704 if (buffer_dirty(bh))
705 recentcy += RECENTCY_DIRTY;
706
707 if (dtime && time_before32(dtime, now) &&
708 time_before32(now, dtime + recentcy))
709 ret = 1;
710 out:
711 brelse(bh);
712 return ret;
713 }
714
715 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
716 struct buffer_head *bitmap, unsigned long *ino)
717 {
718 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
719 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
720
721 next:
722 *ino = ext4_find_next_zero_bit((unsigned long *)
723 bitmap->b_data,
724 EXT4_INODES_PER_GROUP(sb), *ino);
725 if (*ino >= EXT4_INODES_PER_GROUP(sb))
726 goto not_found;
727
728 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
729 recently_deleted_ino = *ino;
730 *ino = *ino + 1;
731 if (*ino < EXT4_INODES_PER_GROUP(sb))
732 goto next;
733 goto not_found;
734 }
735 return 1;
736 not_found:
737 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
738 return 0;
739 /*
740 * Not reusing recently deleted inodes is mostly a preference. We don't
741 * want to report ENOSPC or skew allocation patterns because of that.
742 * So return even recently deleted inode if we could find better in the
743 * given range.
744 */
745 *ino = recently_deleted_ino;
746 return 1;
747 }
748
749 int ext4_mark_inode_used(struct super_block *sb, int ino)
750 {
751 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
752 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
753 struct ext4_group_desc *gdp;
754 ext4_group_t group;
755 int bit;
756 int err = -EFSCORRUPTED;
757
758 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
759 goto out;
760
761 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
762 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
763 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
764 if (IS_ERR(inode_bitmap_bh))
765 return PTR_ERR(inode_bitmap_bh);
766
767 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
768 err = 0;
769 goto out;
770 }
771
772 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
773 if (!gdp || !group_desc_bh) {
774 err = -EINVAL;
775 goto out;
776 }
777
778 ext4_set_bit(bit, inode_bitmap_bh->b_data);
779
780 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
781 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
782 if (err) {
783 ext4_std_error(sb, err);
784 goto out;
785 }
786 err = sync_dirty_buffer(inode_bitmap_bh);
787 if (err) {
788 ext4_std_error(sb, err);
789 goto out;
790 }
791
792 /* We may have to initialize the block bitmap if it isn't already */
793 if (ext4_has_group_desc_csum(sb) &&
794 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
795 struct buffer_head *block_bitmap_bh;
796
797 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
798 if (IS_ERR(block_bitmap_bh)) {
799 err = PTR_ERR(block_bitmap_bh);
800 goto out;
801 }
802
803 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
804 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
805 sync_dirty_buffer(block_bitmap_bh);
806
807 /* recheck and clear flag under lock if we still need to */
808 ext4_lock_group(sb, group);
809 if (ext4_has_group_desc_csum(sb) &&
810 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
811 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
812 ext4_free_group_clusters_set(sb, gdp,
813 ext4_free_clusters_after_init(sb, group, gdp));
814 ext4_block_bitmap_csum_set(sb, group, gdp,
815 block_bitmap_bh);
816 ext4_group_desc_csum_set(sb, group, gdp);
817 }
818 ext4_unlock_group(sb, group);
819 brelse(block_bitmap_bh);
820
821 if (err) {
822 ext4_std_error(sb, err);
823 goto out;
824 }
825 }
826
827 /* Update the relevant bg descriptor fields */
828 if (ext4_has_group_desc_csum(sb)) {
829 int free;
830
831 ext4_lock_group(sb, group); /* while we modify the bg desc */
832 free = EXT4_INODES_PER_GROUP(sb) -
833 ext4_itable_unused_count(sb, gdp);
834 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
835 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
836 free = 0;
837 }
838
839 /*
840 * Check the relative inode number against the last used
841 * relative inode number in this group. if it is greater
842 * we need to update the bg_itable_unused count
843 */
844 if (bit >= free)
845 ext4_itable_unused_set(sb, gdp,
846 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
847 } else {
848 ext4_lock_group(sb, group);
849 }
850
851 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
852 if (ext4_has_group_desc_csum(sb)) {
853 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
854 EXT4_INODES_PER_GROUP(sb) / 8);
855 ext4_group_desc_csum_set(sb, group, gdp);
856 }
857
858 ext4_unlock_group(sb, group);
859 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
860 sync_dirty_buffer(group_desc_bh);
861 out:
862 return err;
863 }
864
865 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
866 bool encrypt)
867 {
868 struct super_block *sb = dir->i_sb;
869 int nblocks = 0;
870 #ifdef CONFIG_EXT4_FS_POSIX_ACL
871 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
872
873 if (IS_ERR(p))
874 return PTR_ERR(p);
875 if (p) {
876 int acl_size = p->a_count * sizeof(ext4_acl_entry);
877
878 nblocks += (S_ISDIR(mode) ? 2 : 1) *
879 __ext4_xattr_set_credits(sb, NULL /* inode */,
880 NULL /* block_bh */, acl_size,
881 true /* is_create */);
882 posix_acl_release(p);
883 }
884 #endif
885
886 #ifdef CONFIG_SECURITY
887 {
888 int num_security_xattrs = 1;
889
890 #ifdef CONFIG_INTEGRITY
891 num_security_xattrs++;
892 #endif
893 /*
894 * We assume that security xattrs are never more than 1k.
895 * In practice they are under 128 bytes.
896 */
897 nblocks += num_security_xattrs *
898 __ext4_xattr_set_credits(sb, NULL /* inode */,
899 NULL /* block_bh */, 1024,
900 true /* is_create */);
901 }
902 #endif
903 if (encrypt)
904 nblocks += __ext4_xattr_set_credits(sb,
905 NULL /* inode */,
906 NULL /* block_bh */,
907 FSCRYPT_SET_CONTEXT_MAX_SIZE,
908 true /* is_create */);
909 return nblocks;
910 }
911
912 /*
913 * There are two policies for allocating an inode. If the new inode is
914 * a directory, then a forward search is made for a block group with both
915 * free space and a low directory-to-inode ratio; if that fails, then of
916 * the groups with above-average free space, that group with the fewest
917 * directories already is chosen.
918 *
919 * For other inodes, search forward from the parent directory's block
920 * group to find a free inode.
921 */
922 struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
923 handle_t *handle, struct inode *dir,
924 umode_t mode, const struct qstr *qstr,
925 __u32 goal, uid_t *owner, __u32 i_flags,
926 int handle_type, unsigned int line_no,
927 int nblocks)
928 {
929 struct super_block *sb;
930 struct buffer_head *inode_bitmap_bh = NULL;
931 struct buffer_head *group_desc_bh;
932 ext4_group_t ngroups, group = 0;
933 unsigned long ino = 0;
934 struct inode *inode;
935 struct ext4_group_desc *gdp = NULL;
936 struct ext4_inode_info *ei;
937 struct ext4_sb_info *sbi;
938 int ret2, err;
939 struct inode *ret;
940 ext4_group_t i;
941 ext4_group_t flex_group;
942 struct ext4_group_info *grp = NULL;
943 bool encrypt = false;
944
945 /* Cannot create files in a deleted directory */
946 if (!dir || !dir->i_nlink)
947 return ERR_PTR(-EPERM);
948
949 sb = dir->i_sb;
950 sbi = EXT4_SB(sb);
951
952 if (unlikely(ext4_forced_shutdown(sbi)))
953 return ERR_PTR(-EIO);
954
955 ngroups = ext4_get_groups_count(sb);
956 trace_ext4_request_inode(dir, mode);
957 inode = new_inode(sb);
958 if (!inode)
959 return ERR_PTR(-ENOMEM);
960 ei = EXT4_I(inode);
961
962 /*
963 * Initialize owners and quota early so that we don't have to account
964 * for quota initialization worst case in standard inode creating
965 * transaction
966 */
967 if (owner) {
968 inode->i_mode = mode;
969 i_uid_write(inode, owner[0]);
970 i_gid_write(inode, owner[1]);
971 } else if (test_opt(sb, GRPID)) {
972 inode->i_mode = mode;
973 inode_fsuid_set(inode, mnt_userns);
974 inode->i_gid = dir->i_gid;
975 } else
976 inode_init_owner(mnt_userns, inode, dir, mode);
977
978 if (ext4_has_feature_project(sb) &&
979 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
980 ei->i_projid = EXT4_I(dir)->i_projid;
981 else
982 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
983
984 if (!(i_flags & EXT4_EA_INODE_FL)) {
985 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
986 if (err)
987 goto out;
988 }
989
990 err = dquot_initialize(inode);
991 if (err)
992 goto out;
993
994 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
995 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
996 if (ret2 < 0) {
997 err = ret2;
998 goto out;
999 }
1000 nblocks += ret2;
1001 }
1002
1003 if (!goal)
1004 goal = sbi->s_inode_goal;
1005
1006 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1007 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1008 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1009 ret2 = 0;
1010 goto got_group;
1011 }
1012
1013 if (S_ISDIR(mode))
1014 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1015 else
1016 ret2 = find_group_other(sb, dir, &group, mode);
1017
1018 got_group:
1019 EXT4_I(dir)->i_last_alloc_group = group;
1020 err = -ENOSPC;
1021 if (ret2 == -1)
1022 goto out;
1023
1024 /*
1025 * Normally we will only go through one pass of this loop,
1026 * unless we get unlucky and it turns out the group we selected
1027 * had its last inode grabbed by someone else.
1028 */
1029 for (i = 0; i < ngroups; i++, ino = 0) {
1030 err = -EIO;
1031
1032 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1033 if (!gdp)
1034 goto out;
1035
1036 /*
1037 * Check free inodes count before loading bitmap.
1038 */
1039 if (ext4_free_inodes_count(sb, gdp) == 0)
1040 goto next_group;
1041
1042 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1043 grp = ext4_get_group_info(sb, group);
1044 /*
1045 * Skip groups with already-known suspicious inode
1046 * tables
1047 */
1048 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1049 goto next_group;
1050 }
1051
1052 brelse(inode_bitmap_bh);
1053 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1054 /* Skip groups with suspicious inode tables */
1055 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1056 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1057 IS_ERR(inode_bitmap_bh)) {
1058 inode_bitmap_bh = NULL;
1059 goto next_group;
1060 }
1061
1062 repeat_in_this_group:
1063 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1064 if (!ret2)
1065 goto next_group;
1066
1067 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1068 ext4_error(sb, "reserved inode found cleared - "
1069 "inode=%lu", ino + 1);
1070 ext4_mark_group_bitmap_corrupted(sb, group,
1071 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1072 goto next_group;
1073 }
1074
1075 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1076 BUG_ON(nblocks <= 0);
1077 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1078 handle_type, nblocks, 0,
1079 ext4_trans_default_revoke_credits(sb));
1080 if (IS_ERR(handle)) {
1081 err = PTR_ERR(handle);
1082 ext4_std_error(sb, err);
1083 goto out;
1084 }
1085 }
1086 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1087 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1088 if (err) {
1089 ext4_std_error(sb, err);
1090 goto out;
1091 }
1092 ext4_lock_group(sb, group);
1093 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1094 if (ret2) {
1095 /* Someone already took the bit. Repeat the search
1096 * with lock held.
1097 */
1098 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1099 if (ret2) {
1100 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1101 ret2 = 0;
1102 } else {
1103 ret2 = 1; /* we didn't grab the inode */
1104 }
1105 }
1106 ext4_unlock_group(sb, group);
1107 ino++; /* the inode bitmap is zero-based */
1108 if (!ret2)
1109 goto got; /* we grabbed the inode! */
1110
1111 if (ino < EXT4_INODES_PER_GROUP(sb))
1112 goto repeat_in_this_group;
1113 next_group:
1114 if (++group == ngroups)
1115 group = 0;
1116 }
1117 err = -ENOSPC;
1118 goto out;
1119
1120 got:
1121 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1122 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1123 if (err) {
1124 ext4_std_error(sb, err);
1125 goto out;
1126 }
1127
1128 BUFFER_TRACE(group_desc_bh, "get_write_access");
1129 err = ext4_journal_get_write_access(handle, group_desc_bh);
1130 if (err) {
1131 ext4_std_error(sb, err);
1132 goto out;
1133 }
1134
1135 /* We may have to initialize the block bitmap if it isn't already */
1136 if (ext4_has_group_desc_csum(sb) &&
1137 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1138 struct buffer_head *block_bitmap_bh;
1139
1140 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1141 if (IS_ERR(block_bitmap_bh)) {
1142 err = PTR_ERR(block_bitmap_bh);
1143 goto out;
1144 }
1145 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1146 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1147 if (err) {
1148 brelse(block_bitmap_bh);
1149 ext4_std_error(sb, err);
1150 goto out;
1151 }
1152
1153 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1154 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1155
1156 /* recheck and clear flag under lock if we still need to */
1157 ext4_lock_group(sb, group);
1158 if (ext4_has_group_desc_csum(sb) &&
1159 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1160 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1161 ext4_free_group_clusters_set(sb, gdp,
1162 ext4_free_clusters_after_init(sb, group, gdp));
1163 ext4_block_bitmap_csum_set(sb, group, gdp,
1164 block_bitmap_bh);
1165 ext4_group_desc_csum_set(sb, group, gdp);
1166 }
1167 ext4_unlock_group(sb, group);
1168 brelse(block_bitmap_bh);
1169
1170 if (err) {
1171 ext4_std_error(sb, err);
1172 goto out;
1173 }
1174 }
1175
1176 /* Update the relevant bg descriptor fields */
1177 if (ext4_has_group_desc_csum(sb)) {
1178 int free;
1179 struct ext4_group_info *grp = NULL;
1180
1181 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1182 grp = ext4_get_group_info(sb, group);
1183 down_read(&grp->alloc_sem); /*
1184 * protect vs itable
1185 * lazyinit
1186 */
1187 }
1188 ext4_lock_group(sb, group); /* while we modify the bg desc */
1189 free = EXT4_INODES_PER_GROUP(sb) -
1190 ext4_itable_unused_count(sb, gdp);
1191 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1192 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1193 free = 0;
1194 }
1195 /*
1196 * Check the relative inode number against the last used
1197 * relative inode number in this group. if it is greater
1198 * we need to update the bg_itable_unused count
1199 */
1200 if (ino > free)
1201 ext4_itable_unused_set(sb, gdp,
1202 (EXT4_INODES_PER_GROUP(sb) - ino));
1203 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1204 up_read(&grp->alloc_sem);
1205 } else {
1206 ext4_lock_group(sb, group);
1207 }
1208
1209 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1210 if (S_ISDIR(mode)) {
1211 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1212 if (sbi->s_log_groups_per_flex) {
1213 ext4_group_t f = ext4_flex_group(sbi, group);
1214
1215 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1216 f)->used_dirs);
1217 }
1218 }
1219 if (ext4_has_group_desc_csum(sb)) {
1220 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1221 EXT4_INODES_PER_GROUP(sb) / 8);
1222 ext4_group_desc_csum_set(sb, group, gdp);
1223 }
1224 ext4_unlock_group(sb, group);
1225
1226 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1227 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1228 if (err) {
1229 ext4_std_error(sb, err);
1230 goto out;
1231 }
1232
1233 percpu_counter_dec(&sbi->s_freeinodes_counter);
1234 if (S_ISDIR(mode))
1235 percpu_counter_inc(&sbi->s_dirs_counter);
1236
1237 if (sbi->s_log_groups_per_flex) {
1238 flex_group = ext4_flex_group(sbi, group);
1239 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1240 flex_group)->free_inodes);
1241 }
1242
1243 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1244 /* This is the optimal IO size (for stat), not the fs block size */
1245 inode->i_blocks = 0;
1246 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1247 ei->i_crtime = inode->i_mtime;
1248
1249 memset(ei->i_data, 0, sizeof(ei->i_data));
1250 ei->i_dir_start_lookup = 0;
1251 ei->i_disksize = 0;
1252
1253 /* Don't inherit extent flag from directory, amongst others. */
1254 ei->i_flags =
1255 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1256 ei->i_flags |= i_flags;
1257 ei->i_file_acl = 0;
1258 ei->i_dtime = 0;
1259 ei->i_block_group = group;
1260 ei->i_last_alloc_group = ~0;
1261
1262 ext4_set_inode_flags(inode, true);
1263 if (IS_DIRSYNC(inode))
1264 ext4_handle_sync(handle);
1265 if (insert_inode_locked(inode) < 0) {
1266 /*
1267 * Likely a bitmap corruption causing inode to be allocated
1268 * twice.
1269 */
1270 err = -EIO;
1271 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1272 inode->i_ino);
1273 ext4_mark_group_bitmap_corrupted(sb, group,
1274 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1275 goto out;
1276 }
1277 inode->i_generation = prandom_u32();
1278
1279 /* Precompute checksum seed for inode metadata */
1280 if (ext4_has_metadata_csum(sb)) {
1281 __u32 csum;
1282 __le32 inum = cpu_to_le32(inode->i_ino);
1283 __le32 gen = cpu_to_le32(inode->i_generation);
1284 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1285 sizeof(inum));
1286 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1287 sizeof(gen));
1288 }
1289
1290 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1291 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1292
1293 ei->i_extra_isize = sbi->s_want_extra_isize;
1294 ei->i_inline_off = 0;
1295 if (ext4_has_feature_inline_data(sb))
1296 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1297 ret = inode;
1298 err = dquot_alloc_inode(inode);
1299 if (err)
1300 goto fail_drop;
1301
1302 /*
1303 * Since the encryption xattr will always be unique, create it first so
1304 * that it's less likely to end up in an external xattr block and
1305 * prevent its deduplication.
1306 */
1307 if (encrypt) {
1308 err = fscrypt_set_context(inode, handle);
1309 if (err)
1310 goto fail_free_drop;
1311 }
1312
1313 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1314 err = ext4_init_acl(handle, inode, dir);
1315 if (err)
1316 goto fail_free_drop;
1317
1318 err = ext4_init_security(handle, inode, dir, qstr);
1319 if (err)
1320 goto fail_free_drop;
1321 }
1322
1323 if (ext4_has_feature_extents(sb)) {
1324 /* set extent flag only for directory, file and normal symlink*/
1325 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1326 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1327 ext4_ext_tree_init(handle, inode);
1328 }
1329 }
1330
1331 if (ext4_handle_valid(handle)) {
1332 ei->i_sync_tid = handle->h_transaction->t_tid;
1333 ei->i_datasync_tid = handle->h_transaction->t_tid;
1334 }
1335
1336 err = ext4_mark_inode_dirty(handle, inode);
1337 if (err) {
1338 ext4_std_error(sb, err);
1339 goto fail_free_drop;
1340 }
1341
1342 ext4_debug("allocating inode %lu\n", inode->i_ino);
1343 trace_ext4_allocate_inode(inode, dir, mode);
1344 brelse(inode_bitmap_bh);
1345 return ret;
1346
1347 fail_free_drop:
1348 dquot_free_inode(inode);
1349 fail_drop:
1350 clear_nlink(inode);
1351 unlock_new_inode(inode);
1352 out:
1353 dquot_drop(inode);
1354 inode->i_flags |= S_NOQUOTA;
1355 iput(inode);
1356 brelse(inode_bitmap_bh);
1357 return ERR_PTR(err);
1358 }
1359
1360 /* Verify that we are loading a valid orphan from disk */
1361 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1362 {
1363 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1364 ext4_group_t block_group;
1365 int bit;
1366 struct buffer_head *bitmap_bh = NULL;
1367 struct inode *inode = NULL;
1368 int err = -EFSCORRUPTED;
1369
1370 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1371 goto bad_orphan;
1372
1373 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1374 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1375 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1376 if (IS_ERR(bitmap_bh))
1377 return ERR_CAST(bitmap_bh);
1378
1379 /* Having the inode bit set should be a 100% indicator that this
1380 * is a valid orphan (no e2fsck run on fs). Orphans also include
1381 * inodes that were being truncated, so we can't check i_nlink==0.
1382 */
1383 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1384 goto bad_orphan;
1385
1386 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1387 if (IS_ERR(inode)) {
1388 err = PTR_ERR(inode);
1389 ext4_error_err(sb, -err,
1390 "couldn't read orphan inode %lu (err %d)",
1391 ino, err);
1392 brelse(bitmap_bh);
1393 return inode;
1394 }
1395
1396 /*
1397 * If the orphans has i_nlinks > 0 then it should be able to
1398 * be truncated, otherwise it won't be removed from the orphan
1399 * list during processing and an infinite loop will result.
1400 * Similarly, it must not be a bad inode.
1401 */
1402 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1403 is_bad_inode(inode))
1404 goto bad_orphan;
1405
1406 if (NEXT_ORPHAN(inode) > max_ino)
1407 goto bad_orphan;
1408 brelse(bitmap_bh);
1409 return inode;
1410
1411 bad_orphan:
1412 ext4_error(sb, "bad orphan inode %lu", ino);
1413 if (bitmap_bh)
1414 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1415 bit, (unsigned long long)bitmap_bh->b_blocknr,
1416 ext4_test_bit(bit, bitmap_bh->b_data));
1417 if (inode) {
1418 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1419 is_bad_inode(inode));
1420 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1421 NEXT_ORPHAN(inode));
1422 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1423 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1424 /* Avoid freeing blocks if we got a bad deleted inode */
1425 if (inode->i_nlink == 0)
1426 inode->i_blocks = 0;
1427 iput(inode);
1428 }
1429 brelse(bitmap_bh);
1430 return ERR_PTR(err);
1431 }
1432
1433 unsigned long ext4_count_free_inodes(struct super_block *sb)
1434 {
1435 unsigned long desc_count;
1436 struct ext4_group_desc *gdp;
1437 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1438 #ifdef EXT4FS_DEBUG
1439 struct ext4_super_block *es;
1440 unsigned long bitmap_count, x;
1441 struct buffer_head *bitmap_bh = NULL;
1442
1443 es = EXT4_SB(sb)->s_es;
1444 desc_count = 0;
1445 bitmap_count = 0;
1446 gdp = NULL;
1447 for (i = 0; i < ngroups; i++) {
1448 gdp = ext4_get_group_desc(sb, i, NULL);
1449 if (!gdp)
1450 continue;
1451 desc_count += ext4_free_inodes_count(sb, gdp);
1452 brelse(bitmap_bh);
1453 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1454 if (IS_ERR(bitmap_bh)) {
1455 bitmap_bh = NULL;
1456 continue;
1457 }
1458
1459 x = ext4_count_free(bitmap_bh->b_data,
1460 EXT4_INODES_PER_GROUP(sb) / 8);
1461 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1462 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1463 bitmap_count += x;
1464 }
1465 brelse(bitmap_bh);
1466 printk(KERN_DEBUG "ext4_count_free_inodes: "
1467 "stored = %u, computed = %lu, %lu\n",
1468 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1469 return desc_count;
1470 #else
1471 desc_count = 0;
1472 for (i = 0; i < ngroups; i++) {
1473 gdp = ext4_get_group_desc(sb, i, NULL);
1474 if (!gdp)
1475 continue;
1476 desc_count += ext4_free_inodes_count(sb, gdp);
1477 cond_resched();
1478 }
1479 return desc_count;
1480 #endif
1481 }
1482
1483 /* Called at mount-time, super-block is locked */
1484 unsigned long ext4_count_dirs(struct super_block * sb)
1485 {
1486 unsigned long count = 0;
1487 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1488
1489 for (i = 0; i < ngroups; i++) {
1490 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1491 if (!gdp)
1492 continue;
1493 count += ext4_used_dirs_count(sb, gdp);
1494 }
1495 return count;
1496 }
1497
1498 /*
1499 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1500 * inode table. Must be called without any spinlock held. The only place
1501 * where it is called from on active part of filesystem is ext4lazyinit
1502 * thread, so we do not need any special locks, however we have to prevent
1503 * inode allocation from the current group, so we take alloc_sem lock, to
1504 * block ext4_new_inode() until we are finished.
1505 */
1506 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1507 int barrier)
1508 {
1509 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1510 struct ext4_sb_info *sbi = EXT4_SB(sb);
1511 struct ext4_group_desc *gdp = NULL;
1512 struct buffer_head *group_desc_bh;
1513 handle_t *handle;
1514 ext4_fsblk_t blk;
1515 int num, ret = 0, used_blks = 0;
1516
1517 /* This should not happen, but just to be sure check this */
1518 if (sb_rdonly(sb)) {
1519 ret = 1;
1520 goto out;
1521 }
1522
1523 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1524 if (!gdp)
1525 goto out;
1526
1527 /*
1528 * We do not need to lock this, because we are the only one
1529 * handling this flag.
1530 */
1531 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1532 goto out;
1533
1534 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1535 if (IS_ERR(handle)) {
1536 ret = PTR_ERR(handle);
1537 goto out;
1538 }
1539
1540 down_write(&grp->alloc_sem);
1541 /*
1542 * If inode bitmap was already initialized there may be some
1543 * used inodes so we need to skip blocks with used inodes in
1544 * inode table.
1545 */
1546 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1547 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1548 ext4_itable_unused_count(sb, gdp)),
1549 sbi->s_inodes_per_block);
1550
1551 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1552 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1553 ext4_itable_unused_count(sb, gdp)) <
1554 EXT4_FIRST_INO(sb)))) {
1555 ext4_error(sb, "Something is wrong with group %u: "
1556 "used itable blocks: %d; "
1557 "itable unused count: %u",
1558 group, used_blks,
1559 ext4_itable_unused_count(sb, gdp));
1560 ret = 1;
1561 goto err_out;
1562 }
1563
1564 blk = ext4_inode_table(sb, gdp) + used_blks;
1565 num = sbi->s_itb_per_group - used_blks;
1566
1567 BUFFER_TRACE(group_desc_bh, "get_write_access");
1568 ret = ext4_journal_get_write_access(handle,
1569 group_desc_bh);
1570 if (ret)
1571 goto err_out;
1572
1573 /*
1574 * Skip zeroout if the inode table is full. But we set the ZEROED
1575 * flag anyway, because obviously, when it is full it does not need
1576 * further zeroing.
1577 */
1578 if (unlikely(num == 0))
1579 goto skip_zeroout;
1580
1581 ext4_debug("going to zero out inode table in group %d\n",
1582 group);
1583 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1584 if (ret < 0)
1585 goto err_out;
1586 if (barrier)
1587 blkdev_issue_flush(sb->s_bdev);
1588
1589 skip_zeroout:
1590 ext4_lock_group(sb, group);
1591 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1592 ext4_group_desc_csum_set(sb, group, gdp);
1593 ext4_unlock_group(sb, group);
1594
1595 BUFFER_TRACE(group_desc_bh,
1596 "call ext4_handle_dirty_metadata");
1597 ret = ext4_handle_dirty_metadata(handle, NULL,
1598 group_desc_bh);
1599
1600 err_out:
1601 up_write(&grp->alloc_sem);
1602 ext4_journal_stop(handle);
1603 out:
1604 return ret;
1605 }
Ubuntu Jammy Linux kernel mirror