Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/ialloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
14 */
15
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
24#include <linux/blkdev.h>
25#include <linux/cred.h>
26
27#include <asm/byteorder.h>
28
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34#include <trace/events/ext4.h>
35
36/*
37 * ialloc.c contains the inodes allocation and deallocation routines
38 */
39
40/*
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
44 *
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
48 */
49
50/*
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
54 */
55void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56{
57 int i;
58
59 if (start_bit >= end_bit)
60 return;
61
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
65 if (i < end_bit)
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67}
68
69void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70{
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
74 }
75 unlock_buffer(bh);
76 put_bh(bh);
77}
78
79static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
83{
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
86
87 if (buffer_verified(bh))
88 return 0;
89 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
90 return -EFSCORRUPTED;
91
92 ext4_lock_group(sb, block_group);
93 if (buffer_verified(bh))
94 goto verified;
95 blk = ext4_inode_bitmap(sb, desc);
96 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
97 EXT4_INODES_PER_GROUP(sb) / 8) ||
98 ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
99 ext4_unlock_group(sb, block_group);
100 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
101 "inode_bitmap = %llu", block_group, blk);
102 ext4_mark_group_bitmap_corrupted(sb, block_group,
103 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
104 return -EFSBADCRC;
105 }
106 set_buffer_verified(bh);
107verified:
108 ext4_unlock_group(sb, block_group);
109 return 0;
110}
111
112/*
113 * Read the inode allocation bitmap for a given block_group, reading
114 * into the specified slot in the superblock's bitmap cache.
115 *
116 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
117 */
118static struct buffer_head *
119ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
120{
121 struct ext4_group_desc *desc;
122 struct ext4_sb_info *sbi = EXT4_SB(sb);
123 struct buffer_head *bh = NULL;
124 ext4_fsblk_t bitmap_blk;
125 int err;
126
127 desc = ext4_get_group_desc(sb, block_group, NULL);
128 if (!desc)
129 return ERR_PTR(-EFSCORRUPTED);
130
131 bitmap_blk = ext4_inode_bitmap(sb, desc);
132 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
133 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
134 ext4_error(sb, "Invalid inode bitmap blk %llu in "
135 "block_group %u", bitmap_blk, block_group);
136 ext4_mark_group_bitmap_corrupted(sb, block_group,
137 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140 bh = sb_getblk(sb, bitmap_blk);
141 if (unlikely(!bh)) {
142 ext4_warning(sb, "Cannot read inode bitmap - "
143 "block_group = %u, inode_bitmap = %llu",
144 block_group, bitmap_blk);
145 return ERR_PTR(-ENOMEM);
146 }
147 if (bitmap_uptodate(bh))
148 goto verify;
149
150 lock_buffer(bh);
151 if (bitmap_uptodate(bh)) {
152 unlock_buffer(bh);
153 goto verify;
154 }
155
156 ext4_lock_group(sb, block_group);
157 if (ext4_has_group_desc_csum(sb) &&
158 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
159 if (block_group == 0) {
160 ext4_unlock_group(sb, block_group);
161 unlock_buffer(bh);
162 ext4_error(sb, "Inode bitmap for bg 0 marked "
163 "uninitialized");
164 err = -EFSCORRUPTED;
165 goto out;
166 }
167 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
168 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
169 sb->s_blocksize * 8, bh->b_data);
170 set_bitmap_uptodate(bh);
171 set_buffer_uptodate(bh);
172 set_buffer_verified(bh);
173 ext4_unlock_group(sb, block_group);
174 unlock_buffer(bh);
175 return bh;
176 }
177 ext4_unlock_group(sb, block_group);
178
179 if (buffer_uptodate(bh)) {
180 /*
181 * if not uninit if bh is uptodate,
182 * bitmap is also uptodate
183 */
184 set_bitmap_uptodate(bh);
185 unlock_buffer(bh);
186 goto verify;
187 }
188 /*
189 * submit the buffer_head for reading
190 */
191 trace_ext4_load_inode_bitmap(sb, block_group);
192 bh->b_end_io = ext4_end_bitmap_read;
193 get_bh(bh);
194 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
195 wait_on_buffer(bh);
196 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
197 if (!buffer_uptodate(bh)) {
198 put_bh(bh);
199 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
200 "block_group = %u, inode_bitmap = %llu",
201 block_group, bitmap_blk);
202 ext4_mark_group_bitmap_corrupted(sb, block_group,
203 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
204 return ERR_PTR(-EIO);
205 }
206
207verify:
208 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
209 if (err)
210 goto out;
211 return bh;
212out:
213 put_bh(bh);
214 return ERR_PTR(err);
215}
216
217/*
218 * NOTE! When we get the inode, we're the only people
219 * that have access to it, and as such there are no
220 * race conditions we have to worry about. The inode
221 * is not on the hash-lists, and it cannot be reached
222 * through the filesystem because the directory entry
223 * has been deleted earlier.
224 *
225 * HOWEVER: we must make sure that we get no aliases,
226 * which means that we have to call "clear_inode()"
227 * _before_ we mark the inode not in use in the inode
228 * bitmaps. Otherwise a newly created file might use
229 * the same inode number (not actually the same pointer
230 * though), and then we'd have two inodes sharing the
231 * same inode number and space on the harddisk.
232 */
233void ext4_free_inode(handle_t *handle, struct inode *inode)
234{
235 struct super_block *sb = inode->i_sb;
236 int is_directory;
237 unsigned long ino;
238 struct buffer_head *bitmap_bh = NULL;
239 struct buffer_head *bh2;
240 ext4_group_t block_group;
241 unsigned long bit;
242 struct ext4_group_desc *gdp;
243 struct ext4_super_block *es;
244 struct ext4_sb_info *sbi;
245 int fatal = 0, err, count, cleared;
246 struct ext4_group_info *grp;
247
248 if (!sb) {
249 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
250 "nonexistent device\n", __func__, __LINE__);
251 return;
252 }
253 if (atomic_read(&inode->i_count) > 1) {
254 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
255 __func__, __LINE__, inode->i_ino,
256 atomic_read(&inode->i_count));
257 return;
258 }
259 if (inode->i_nlink) {
260 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
261 __func__, __LINE__, inode->i_ino, inode->i_nlink);
262 return;
263 }
264 sbi = EXT4_SB(sb);
265
266 ino = inode->i_ino;
267 ext4_debug("freeing inode %lu\n", ino);
268 trace_ext4_free_inode(inode);
269
270 dquot_initialize(inode);
271 dquot_free_inode(inode);
272
273 is_directory = S_ISDIR(inode->i_mode);
274
275 /* Do this BEFORE marking the inode not in use or returning an error */
276 ext4_clear_inode(inode);
277
278 es = sbi->s_es;
279 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
280 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
281 goto error_return;
282 }
283 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
284 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
285 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
286 /* Don't bother if the inode bitmap is corrupt. */
287 grp = ext4_get_group_info(sb, block_group);
288 if (IS_ERR(bitmap_bh)) {
289 fatal = PTR_ERR(bitmap_bh);
290 bitmap_bh = NULL;
291 goto error_return;
292 }
293 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
294 fatal = -EFSCORRUPTED;
295 goto error_return;
296 }
297
298 BUFFER_TRACE(bitmap_bh, "get_write_access");
299 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
300 if (fatal)
301 goto error_return;
302
303 fatal = -ESRCH;
304 gdp = ext4_get_group_desc(sb, block_group, &bh2);
305 if (gdp) {
306 BUFFER_TRACE(bh2, "get_write_access");
307 fatal = ext4_journal_get_write_access(handle, bh2);
308 }
309 ext4_lock_group(sb, block_group);
310 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
311 if (fatal || !cleared) {
312 ext4_unlock_group(sb, block_group);
313 goto out;
314 }
315
316 count = ext4_free_inodes_count(sb, gdp) + 1;
317 ext4_free_inodes_set(sb, gdp, count);
318 if (is_directory) {
319 count = ext4_used_dirs_count(sb, gdp) - 1;
320 ext4_used_dirs_set(sb, gdp, count);
321 percpu_counter_dec(&sbi->s_dirs_counter);
322 }
323 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
324 EXT4_INODES_PER_GROUP(sb) / 8);
325 ext4_group_desc_csum_set(sb, block_group, gdp);
326 ext4_unlock_group(sb, block_group);
327
328 percpu_counter_inc(&sbi->s_freeinodes_counter);
329 if (sbi->s_log_groups_per_flex) {
330 struct flex_groups *fg;
331
332 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
333 ext4_flex_group(sbi, block_group));
334 atomic_inc(&fg->free_inodes);
335 if (is_directory)
336 atomic_dec(&fg->used_dirs);
337 }
338 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
339 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
340out:
341 if (cleared) {
342 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
343 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
344 if (!fatal)
345 fatal = err;
346 } else {
347 ext4_error(sb, "bit already cleared for inode %lu", ino);
348 ext4_mark_group_bitmap_corrupted(sb, block_group,
349 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
350 }
351
352error_return:
353 brelse(bitmap_bh);
354 ext4_std_error(sb, fatal);
355}
356
357struct orlov_stats {
358 __u64 free_clusters;
359 __u32 free_inodes;
360 __u32 used_dirs;
361};
362
363/*
364 * Helper function for Orlov's allocator; returns critical information
365 * for a particular block group or flex_bg. If flex_size is 1, then g
366 * is a block group number; otherwise it is flex_bg number.
367 */
368static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
369 int flex_size, struct orlov_stats *stats)
370{
371 struct ext4_group_desc *desc;
372
373 if (flex_size > 1) {
374 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
375 s_flex_groups, g);
376 stats->free_inodes = atomic_read(&fg->free_inodes);
377 stats->free_clusters = atomic64_read(&fg->free_clusters);
378 stats->used_dirs = atomic_read(&fg->used_dirs);
379 return;
380 }
381
382 desc = ext4_get_group_desc(sb, g, NULL);
383 if (desc) {
384 stats->free_inodes = ext4_free_inodes_count(sb, desc);
385 stats->free_clusters = ext4_free_group_clusters(sb, desc);
386 stats->used_dirs = ext4_used_dirs_count(sb, desc);
387 } else {
388 stats->free_inodes = 0;
389 stats->free_clusters = 0;
390 stats->used_dirs = 0;
391 }
392}
393
394/*
395 * Orlov's allocator for directories.
396 *
397 * We always try to spread first-level directories.
398 *
399 * If there are blockgroups with both free inodes and free blocks counts
400 * not worse than average we return one with smallest directory count.
401 * Otherwise we simply return a random group.
402 *
403 * For the rest rules look so:
404 *
405 * It's OK to put directory into a group unless
406 * it has too many directories already (max_dirs) or
407 * it has too few free inodes left (min_inodes) or
408 * it has too few free blocks left (min_blocks) or
409 * Parent's group is preferred, if it doesn't satisfy these
410 * conditions we search cyclically through the rest. If none
411 * of the groups look good we just look for a group with more
412 * free inodes than average (starting at parent's group).
413 */
414
415static int find_group_orlov(struct super_block *sb, struct inode *parent,
416 ext4_group_t *group, umode_t mode,
417 const struct qstr *qstr)
418{
419 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
420 struct ext4_sb_info *sbi = EXT4_SB(sb);
421 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
422 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
423 unsigned int freei, avefreei, grp_free;
424 ext4_fsblk_t freeb, avefreec;
425 unsigned int ndirs;
426 int max_dirs, min_inodes;
427 ext4_grpblk_t min_clusters;
428 ext4_group_t i, grp, g, ngroups;
429 struct ext4_group_desc *desc;
430 struct orlov_stats stats;
431 int flex_size = ext4_flex_bg_size(sbi);
432 struct dx_hash_info hinfo;
433
434 ngroups = real_ngroups;
435 if (flex_size > 1) {
436 ngroups = (real_ngroups + flex_size - 1) >>
437 sbi->s_log_groups_per_flex;
438 parent_group >>= sbi->s_log_groups_per_flex;
439 }
440
441 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
442 avefreei = freei / ngroups;
443 freeb = EXT4_C2B(sbi,
444 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
445 avefreec = freeb;
446 do_div(avefreec, ngroups);
447 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
448
449 if (S_ISDIR(mode) &&
450 ((parent == d_inode(sb->s_root)) ||
451 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
452 int best_ndir = inodes_per_group;
453 int ret = -1;
454
455 if (qstr) {
456 hinfo.hash_version = DX_HASH_HALF_MD4;
457 hinfo.seed = sbi->s_hash_seed;
458 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
459 grp = hinfo.hash;
460 } else
461 grp = prandom_u32();
462 parent_group = (unsigned)grp % ngroups;
463 for (i = 0; i < ngroups; i++) {
464 g = (parent_group + i) % ngroups;
465 get_orlov_stats(sb, g, flex_size, &stats);
466 if (!stats.free_inodes)
467 continue;
468 if (stats.used_dirs >= best_ndir)
469 continue;
470 if (stats.free_inodes < avefreei)
471 continue;
472 if (stats.free_clusters < avefreec)
473 continue;
474 grp = g;
475 ret = 0;
476 best_ndir = stats.used_dirs;
477 }
478 if (ret)
479 goto fallback;
480 found_flex_bg:
481 if (flex_size == 1) {
482 *group = grp;
483 return 0;
484 }
485
486 /*
487 * We pack inodes at the beginning of the flexgroup's
488 * inode tables. Block allocation decisions will do
489 * something similar, although regular files will
490 * start at 2nd block group of the flexgroup. See
491 * ext4_ext_find_goal() and ext4_find_near().
492 */
493 grp *= flex_size;
494 for (i = 0; i < flex_size; i++) {
495 if (grp+i >= real_ngroups)
496 break;
497 desc = ext4_get_group_desc(sb, grp+i, NULL);
498 if (desc && ext4_free_inodes_count(sb, desc)) {
499 *group = grp+i;
500 return 0;
501 }
502 }
503 goto fallback;
504 }
505
506 max_dirs = ndirs / ngroups + inodes_per_group / 16;
507 min_inodes = avefreei - inodes_per_group*flex_size / 4;
508 if (min_inodes < 1)
509 min_inodes = 1;
510 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
511
512 /*
513 * Start looking in the flex group where we last allocated an
514 * inode for this parent directory
515 */
516 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
517 parent_group = EXT4_I(parent)->i_last_alloc_group;
518 if (flex_size > 1)
519 parent_group >>= sbi->s_log_groups_per_flex;
520 }
521
522 for (i = 0; i < ngroups; i++) {
523 grp = (parent_group + i) % ngroups;
524 get_orlov_stats(sb, grp, flex_size, &stats);
525 if (stats.used_dirs >= max_dirs)
526 continue;
527 if (stats.free_inodes < min_inodes)
528 continue;
529 if (stats.free_clusters < min_clusters)
530 continue;
531 goto found_flex_bg;
532 }
533
534fallback:
535 ngroups = real_ngroups;
536 avefreei = freei / ngroups;
537fallback_retry:
538 parent_group = EXT4_I(parent)->i_block_group;
539 for (i = 0; i < ngroups; i++) {
540 grp = (parent_group + i) % ngroups;
541 desc = ext4_get_group_desc(sb, grp, NULL);
542 if (desc) {
543 grp_free = ext4_free_inodes_count(sb, desc);
544 if (grp_free && grp_free >= avefreei) {
545 *group = grp;
546 return 0;
547 }
548 }
549 }
550
551 if (avefreei) {
552 /*
553 * The free-inodes counter is approximate, and for really small
554 * filesystems the above test can fail to find any blockgroups
555 */
556 avefreei = 0;
557 goto fallback_retry;
558 }
559
560 return -1;
561}
562
563static int find_group_other(struct super_block *sb, struct inode *parent,
564 ext4_group_t *group, umode_t mode)
565{
566 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
567 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
568 struct ext4_group_desc *desc;
569 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
570
571 /*
572 * Try to place the inode is the same flex group as its
573 * parent. If we can't find space, use the Orlov algorithm to
574 * find another flex group, and store that information in the
575 * parent directory's inode information so that use that flex
576 * group for future allocations.
577 */
578 if (flex_size > 1) {
579 int retry = 0;
580
581 try_again:
582 parent_group &= ~(flex_size-1);
583 last = parent_group + flex_size;
584 if (last > ngroups)
585 last = ngroups;
586 for (i = parent_group; i < last; i++) {
587 desc = ext4_get_group_desc(sb, i, NULL);
588 if (desc && ext4_free_inodes_count(sb, desc)) {
589 *group = i;
590 return 0;
591 }
592 }
593 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
594 retry = 1;
595 parent_group = EXT4_I(parent)->i_last_alloc_group;
596 goto try_again;
597 }
598 /*
599 * If this didn't work, use the Orlov search algorithm
600 * to find a new flex group; we pass in the mode to
601 * avoid the topdir algorithms.
602 */
603 *group = parent_group + flex_size;
604 if (*group > ngroups)
605 *group = 0;
606 return find_group_orlov(sb, parent, group, mode, NULL);
607 }
608
609 /*
610 * Try to place the inode in its parent directory
611 */
612 *group = parent_group;
613 desc = ext4_get_group_desc(sb, *group, NULL);
614 if (desc && ext4_free_inodes_count(sb, desc) &&
615 ext4_free_group_clusters(sb, desc))
616 return 0;
617
618 /*
619 * We're going to place this inode in a different blockgroup from its
620 * parent. We want to cause files in a common directory to all land in
621 * the same blockgroup. But we want files which are in a different
622 * directory which shares a blockgroup with our parent to land in a
623 * different blockgroup.
624 *
625 * So add our directory's i_ino into the starting point for the hash.
626 */
627 *group = (*group + parent->i_ino) % ngroups;
628
629 /*
630 * Use a quadratic hash to find a group with a free inode and some free
631 * blocks.
632 */
633 for (i = 1; i < ngroups; i <<= 1) {
634 *group += i;
635 if (*group >= ngroups)
636 *group -= ngroups;
637 desc = ext4_get_group_desc(sb, *group, NULL);
638 if (desc && ext4_free_inodes_count(sb, desc) &&
639 ext4_free_group_clusters(sb, desc))
640 return 0;
641 }
642
643 /*
644 * That failed: try linear search for a free inode, even if that group
645 * has no free blocks.
646 */
647 *group = parent_group;
648 for (i = 0; i < ngroups; i++) {
649 if (++*group >= ngroups)
650 *group = 0;
651 desc = ext4_get_group_desc(sb, *group, NULL);
652 if (desc && ext4_free_inodes_count(sb, desc))
653 return 0;
654 }
655
656 return -1;
657}
658
659/*
660 * In no journal mode, if an inode has recently been deleted, we want
661 * to avoid reusing it until we're reasonably sure the inode table
662 * block has been written back to disk. (Yes, these values are
663 * somewhat arbitrary...)
664 */
665#define RECENTCY_MIN 60
666#define RECENTCY_DIRTY 300
667
668static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
669{
670 struct ext4_group_desc *gdp;
671 struct ext4_inode *raw_inode;
672 struct buffer_head *bh;
673 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
674 int offset, ret = 0;
675 int recentcy = RECENTCY_MIN;
676 u32 dtime, now;
677
678 gdp = ext4_get_group_desc(sb, group, NULL);
679 if (unlikely(!gdp))
680 return 0;
681
682 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
683 (ino / inodes_per_block));
684 if (!bh || !buffer_uptodate(bh))
685 /*
686 * If the block is not in the buffer cache, then it
687 * must have been written out.
688 */
689 goto out;
690
691 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
692 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
693
694 /* i_dtime is only 32 bits on disk, but we only care about relative
695 * times in the range of a few minutes (i.e. long enough to sync a
696 * recently-deleted inode to disk), so using the low 32 bits of the
697 * clock (a 68 year range) is enough, see time_before32() */
698 dtime = le32_to_cpu(raw_inode->i_dtime);
699 now = ktime_get_real_seconds();
700 if (buffer_dirty(bh))
701 recentcy += RECENTCY_DIRTY;
702
703 if (dtime && time_before32(dtime, now) &&
704 time_before32(now, dtime + recentcy))
705 ret = 1;
706out:
707 brelse(bh);
708 return ret;
709}
710
711static int find_inode_bit(struct super_block *sb, ext4_group_t group,
712 struct buffer_head *bitmap, unsigned long *ino)
713{
714 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
715 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
716
717next:
718 *ino = ext4_find_next_zero_bit((unsigned long *)
719 bitmap->b_data,
720 EXT4_INODES_PER_GROUP(sb), *ino);
721 if (*ino >= EXT4_INODES_PER_GROUP(sb))
722 goto not_found;
723
724 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
725 recently_deleted_ino = *ino;
726 *ino = *ino + 1;
727 if (*ino < EXT4_INODES_PER_GROUP(sb))
728 goto next;
729 goto not_found;
730 }
731 return 1;
732not_found:
733 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
734 return 0;
735 /*
736 * Not reusing recently deleted inodes is mostly a preference. We don't
737 * want to report ENOSPC or skew allocation patterns because of that.
738 * So return even recently deleted inode if we could find better in the
739 * given range.
740 */
741 *ino = recently_deleted_ino;
742 return 1;
743}
744
745/*
746 * There are two policies for allocating an inode. If the new inode is
747 * a directory, then a forward search is made for a block group with both
748 * free space and a low directory-to-inode ratio; if that fails, then of
749 * the groups with above-average free space, that group with the fewest
750 * directories already is chosen.
751 *
752 * For other inodes, search forward from the parent directory's block
753 * group to find a free inode.
754 */
755struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
756 umode_t mode, const struct qstr *qstr,
757 __u32 goal, uid_t *owner, __u32 i_flags,
758 int handle_type, unsigned int line_no,
759 int nblocks)
760{
761 struct super_block *sb;
762 struct buffer_head *inode_bitmap_bh = NULL;
763 struct buffer_head *group_desc_bh;
764 ext4_group_t ngroups, group = 0;
765 unsigned long ino = 0;
766 struct inode *inode;
767 struct ext4_group_desc *gdp = NULL;
768 struct ext4_inode_info *ei;
769 struct ext4_sb_info *sbi;
770 int ret2, err;
771 struct inode *ret;
772 ext4_group_t i;
773 ext4_group_t flex_group;
774 struct ext4_group_info *grp;
775 int encrypt = 0;
776
777 /* Cannot create files in a deleted directory */
778 if (!dir || !dir->i_nlink)
779 return ERR_PTR(-EPERM);
780
781 sb = dir->i_sb;
782 sbi = EXT4_SB(sb);
783
784 if (unlikely(ext4_forced_shutdown(sbi)))
785 return ERR_PTR(-EIO);
786
787 if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
788 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
789 !(i_flags & EXT4_EA_INODE_FL)) {
790 err = fscrypt_get_encryption_info(dir);
791 if (err)
792 return ERR_PTR(err);
793 if (!fscrypt_has_encryption_key(dir))
794 return ERR_PTR(-ENOKEY);
795 encrypt = 1;
796 }
797
798 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
799#ifdef CONFIG_EXT4_FS_POSIX_ACL
800 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
801
802 if (IS_ERR(p))
803 return ERR_CAST(p);
804 if (p) {
805 int acl_size = p->a_count * sizeof(ext4_acl_entry);
806
807 nblocks += (S_ISDIR(mode) ? 2 : 1) *
808 __ext4_xattr_set_credits(sb, NULL /* inode */,
809 NULL /* block_bh */, acl_size,
810 true /* is_create */);
811 posix_acl_release(p);
812 }
813#endif
814
815#ifdef CONFIG_SECURITY
816 {
817 int num_security_xattrs = 1;
818
819#ifdef CONFIG_INTEGRITY
820 num_security_xattrs++;
821#endif
822 /*
823 * We assume that security xattrs are never
824 * more than 1k. In practice they are under
825 * 128 bytes.
826 */
827 nblocks += num_security_xattrs *
828 __ext4_xattr_set_credits(sb, NULL /* inode */,
829 NULL /* block_bh */, 1024,
830 true /* is_create */);
831 }
832#endif
833 if (encrypt)
834 nblocks += __ext4_xattr_set_credits(sb,
835 NULL /* inode */, NULL /* block_bh */,
836 FSCRYPT_SET_CONTEXT_MAX_SIZE,
837 true /* is_create */);
838 }
839
840 ngroups = ext4_get_groups_count(sb);
841 trace_ext4_request_inode(dir, mode);
842 inode = new_inode(sb);
843 if (!inode)
844 return ERR_PTR(-ENOMEM);
845 ei = EXT4_I(inode);
846
847 /*
848 * Initialize owners and quota early so that we don't have to account
849 * for quota initialization worst case in standard inode creating
850 * transaction
851 */
852 if (owner) {
853 inode->i_mode = mode;
854 i_uid_write(inode, owner[0]);
855 i_gid_write(inode, owner[1]);
856 } else if (test_opt(sb, GRPID)) {
857 inode->i_mode = mode;
858 inode->i_uid = current_fsuid();
859 inode->i_gid = dir->i_gid;
860 } else
861 inode_init_owner(inode, dir, mode);
862
863 if (ext4_has_feature_project(sb) &&
864 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
865 ei->i_projid = EXT4_I(dir)->i_projid;
866 else
867 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
868
869 err = dquot_initialize(inode);
870 if (err)
871 goto out;
872
873 if (!goal)
874 goal = sbi->s_inode_goal;
875
876 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
877 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
878 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
879 ret2 = 0;
880 goto got_group;
881 }
882
883 if (S_ISDIR(mode))
884 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
885 else
886 ret2 = find_group_other(sb, dir, &group, mode);
887
888got_group:
889 EXT4_I(dir)->i_last_alloc_group = group;
890 err = -ENOSPC;
891 if (ret2 == -1)
892 goto out;
893
894 /*
895 * Normally we will only go through one pass of this loop,
896 * unless we get unlucky and it turns out the group we selected
897 * had its last inode grabbed by someone else.
898 */
899 for (i = 0; i < ngroups; i++, ino = 0) {
900 err = -EIO;
901
902 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
903 if (!gdp)
904 goto out;
905
906 /*
907 * Check free inodes count before loading bitmap.
908 */
909 if (ext4_free_inodes_count(sb, gdp) == 0)
910 goto next_group;
911
912 grp = ext4_get_group_info(sb, group);
913 /* Skip groups with already-known suspicious inode tables */
914 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
915 goto next_group;
916
917 brelse(inode_bitmap_bh);
918 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
919 /* Skip groups with suspicious inode tables */
920 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
921 IS_ERR(inode_bitmap_bh)) {
922 inode_bitmap_bh = NULL;
923 goto next_group;
924 }
925
926repeat_in_this_group:
927 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
928 if (!ret2)
929 goto next_group;
930
931 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
932 ext4_error(sb, "reserved inode found cleared - "
933 "inode=%lu", ino + 1);
934 ext4_mark_group_bitmap_corrupted(sb, group,
935 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
936 goto next_group;
937 }
938
939 if (!handle) {
940 BUG_ON(nblocks <= 0);
941 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
942 handle_type, nblocks, 0,
943 ext4_trans_default_revoke_credits(sb));
944 if (IS_ERR(handle)) {
945 err = PTR_ERR(handle);
946 ext4_std_error(sb, err);
947 goto out;
948 }
949 }
950 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
951 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
952 if (err) {
953 ext4_std_error(sb, err);
954 goto out;
955 }
956 ext4_lock_group(sb, group);
957 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
958 if (ret2) {
959 /* Someone already took the bit. Repeat the search
960 * with lock held.
961 */
962 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
963 if (ret2) {
964 ext4_set_bit(ino, inode_bitmap_bh->b_data);
965 ret2 = 0;
966 } else {
967 ret2 = 1; /* we didn't grab the inode */
968 }
969 }
970 ext4_unlock_group(sb, group);
971 ino++; /* the inode bitmap is zero-based */
972 if (!ret2)
973 goto got; /* we grabbed the inode! */
974
975 if (ino < EXT4_INODES_PER_GROUP(sb))
976 goto repeat_in_this_group;
977next_group:
978 if (++group == ngroups)
979 group = 0;
980 }
981 err = -ENOSPC;
982 goto out;
983
984got:
985 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
986 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
987 if (err) {
988 ext4_std_error(sb, err);
989 goto out;
990 }
991
992 BUFFER_TRACE(group_desc_bh, "get_write_access");
993 err = ext4_journal_get_write_access(handle, group_desc_bh);
994 if (err) {
995 ext4_std_error(sb, err);
996 goto out;
997 }
998
999 /* We may have to initialize the block bitmap if it isn't already */
1000 if (ext4_has_group_desc_csum(sb) &&
1001 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1002 struct buffer_head *block_bitmap_bh;
1003
1004 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1005 if (IS_ERR(block_bitmap_bh)) {
1006 err = PTR_ERR(block_bitmap_bh);
1007 goto out;
1008 }
1009 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1010 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1011 if (err) {
1012 brelse(block_bitmap_bh);
1013 ext4_std_error(sb, err);
1014 goto out;
1015 }
1016
1017 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1018 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1019
1020 /* recheck and clear flag under lock if we still need to */
1021 ext4_lock_group(sb, group);
1022 if (ext4_has_group_desc_csum(sb) &&
1023 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1024 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1025 ext4_free_group_clusters_set(sb, gdp,
1026 ext4_free_clusters_after_init(sb, group, gdp));
1027 ext4_block_bitmap_csum_set(sb, group, gdp,
1028 block_bitmap_bh);
1029 ext4_group_desc_csum_set(sb, group, gdp);
1030 }
1031 ext4_unlock_group(sb, group);
1032 brelse(block_bitmap_bh);
1033
1034 if (err) {
1035 ext4_std_error(sb, err);
1036 goto out;
1037 }
1038 }
1039
1040 /* Update the relevant bg descriptor fields */
1041 if (ext4_has_group_desc_csum(sb)) {
1042 int free;
1043 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1044
1045 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1046 ext4_lock_group(sb, group); /* while we modify the bg desc */
1047 free = EXT4_INODES_PER_GROUP(sb) -
1048 ext4_itable_unused_count(sb, gdp);
1049 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1050 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1051 free = 0;
1052 }
1053 /*
1054 * Check the relative inode number against the last used
1055 * relative inode number in this group. if it is greater
1056 * we need to update the bg_itable_unused count
1057 */
1058 if (ino > free)
1059 ext4_itable_unused_set(sb, gdp,
1060 (EXT4_INODES_PER_GROUP(sb) - ino));
1061 up_read(&grp->alloc_sem);
1062 } else {
1063 ext4_lock_group(sb, group);
1064 }
1065
1066 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1067 if (S_ISDIR(mode)) {
1068 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1069 if (sbi->s_log_groups_per_flex) {
1070 ext4_group_t f = ext4_flex_group(sbi, group);
1071
1072 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1073 f)->used_dirs);
1074 }
1075 }
1076 if (ext4_has_group_desc_csum(sb)) {
1077 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1078 EXT4_INODES_PER_GROUP(sb) / 8);
1079 ext4_group_desc_csum_set(sb, group, gdp);
1080 }
1081 ext4_unlock_group(sb, group);
1082
1083 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1084 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1085 if (err) {
1086 ext4_std_error(sb, err);
1087 goto out;
1088 }
1089
1090 percpu_counter_dec(&sbi->s_freeinodes_counter);
1091 if (S_ISDIR(mode))
1092 percpu_counter_inc(&sbi->s_dirs_counter);
1093
1094 if (sbi->s_log_groups_per_flex) {
1095 flex_group = ext4_flex_group(sbi, group);
1096 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1097 flex_group)->free_inodes);
1098 }
1099
1100 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1101 /* This is the optimal IO size (for stat), not the fs block size */
1102 inode->i_blocks = 0;
1103 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1104 ei->i_crtime = inode->i_mtime;
1105
1106 memset(ei->i_data, 0, sizeof(ei->i_data));
1107 ei->i_dir_start_lookup = 0;
1108 ei->i_disksize = 0;
1109
1110 /* Don't inherit extent flag from directory, amongst others. */
1111 ei->i_flags =
1112 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1113 ei->i_flags |= i_flags;
1114 ei->i_file_acl = 0;
1115 ei->i_dtime = 0;
1116 ei->i_block_group = group;
1117 ei->i_last_alloc_group = ~0;
1118
1119 ext4_set_inode_flags(inode, true);
1120 if (IS_DIRSYNC(inode))
1121 ext4_handle_sync(handle);
1122 if (insert_inode_locked(inode) < 0) {
1123 /*
1124 * Likely a bitmap corruption causing inode to be allocated
1125 * twice.
1126 */
1127 err = -EIO;
1128 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1129 inode->i_ino);
1130 ext4_mark_group_bitmap_corrupted(sb, group,
1131 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1132 goto out;
1133 }
1134 inode->i_generation = prandom_u32();
1135
1136 /* Precompute checksum seed for inode metadata */
1137 if (ext4_has_metadata_csum(sb)) {
1138 __u32 csum;
1139 __le32 inum = cpu_to_le32(inode->i_ino);
1140 __le32 gen = cpu_to_le32(inode->i_generation);
1141 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1142 sizeof(inum));
1143 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1144 sizeof(gen));
1145 }
1146
1147 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1148 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1149
1150 ei->i_extra_isize = sbi->s_want_extra_isize;
1151 ei->i_inline_off = 0;
1152 if (ext4_has_feature_inline_data(sb))
1153 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1154 ret = inode;
1155 err = dquot_alloc_inode(inode);
1156 if (err)
1157 goto fail_drop;
1158
1159 /*
1160 * Since the encryption xattr will always be unique, create it first so
1161 * that it's less likely to end up in an external xattr block and
1162 * prevent its deduplication.
1163 */
1164 if (encrypt) {
1165 err = fscrypt_inherit_context(dir, inode, handle, true);
1166 if (err)
1167 goto fail_free_drop;
1168 }
1169
1170 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1171 err = ext4_init_acl(handle, inode, dir);
1172 if (err)
1173 goto fail_free_drop;
1174
1175 err = ext4_init_security(handle, inode, dir, qstr);
1176 if (err)
1177 goto fail_free_drop;
1178 }
1179
1180 if (ext4_has_feature_extents(sb)) {
1181 /* set extent flag only for directory, file and normal symlink*/
1182 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1183 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1184 ext4_ext_tree_init(handle, inode);
1185 }
1186 }
1187
1188 if (ext4_handle_valid(handle)) {
1189 ei->i_sync_tid = handle->h_transaction->t_tid;
1190 ei->i_datasync_tid = handle->h_transaction->t_tid;
1191 }
1192
1193 err = ext4_mark_inode_dirty(handle, inode);
1194 if (err) {
1195 ext4_std_error(sb, err);
1196 goto fail_free_drop;
1197 }
1198
1199 ext4_debug("allocating inode %lu\n", inode->i_ino);
1200 trace_ext4_allocate_inode(inode, dir, mode);
1201 brelse(inode_bitmap_bh);
1202 return ret;
1203
1204fail_free_drop:
1205 dquot_free_inode(inode);
1206fail_drop:
1207 clear_nlink(inode);
1208 unlock_new_inode(inode);
1209out:
1210 dquot_drop(inode);
1211 inode->i_flags |= S_NOQUOTA;
1212 iput(inode);
1213 brelse(inode_bitmap_bh);
1214 return ERR_PTR(err);
1215}
1216
1217/* Verify that we are loading a valid orphan from disk */
1218struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1219{
1220 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1221 ext4_group_t block_group;
1222 int bit;
1223 struct buffer_head *bitmap_bh = NULL;
1224 struct inode *inode = NULL;
1225 int err = -EFSCORRUPTED;
1226
1227 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1228 goto bad_orphan;
1229
1230 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1231 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1232 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1233 if (IS_ERR(bitmap_bh))
1234 return ERR_CAST(bitmap_bh);
1235
1236 /* Having the inode bit set should be a 100% indicator that this
1237 * is a valid orphan (no e2fsck run on fs). Orphans also include
1238 * inodes that were being truncated, so we can't check i_nlink==0.
1239 */
1240 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1241 goto bad_orphan;
1242
1243 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1244 if (IS_ERR(inode)) {
1245 err = PTR_ERR(inode);
1246 ext4_error_err(sb, -err,
1247 "couldn't read orphan inode %lu (err %d)",
1248 ino, err);
1249 brelse(bitmap_bh);
1250 return inode;
1251 }
1252
1253 /*
1254 * If the orphans has i_nlinks > 0 then it should be able to
1255 * be truncated, otherwise it won't be removed from the orphan
1256 * list during processing and an infinite loop will result.
1257 * Similarly, it must not be a bad inode.
1258 */
1259 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1260 is_bad_inode(inode))
1261 goto bad_orphan;
1262
1263 if (NEXT_ORPHAN(inode) > max_ino)
1264 goto bad_orphan;
1265 brelse(bitmap_bh);
1266 return inode;
1267
1268bad_orphan:
1269 ext4_error(sb, "bad orphan inode %lu", ino);
1270 if (bitmap_bh)
1271 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1272 bit, (unsigned long long)bitmap_bh->b_blocknr,
1273 ext4_test_bit(bit, bitmap_bh->b_data));
1274 if (inode) {
1275 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1276 is_bad_inode(inode));
1277 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1278 NEXT_ORPHAN(inode));
1279 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1280 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1281 /* Avoid freeing blocks if we got a bad deleted inode */
1282 if (inode->i_nlink == 0)
1283 inode->i_blocks = 0;
1284 iput(inode);
1285 }
1286 brelse(bitmap_bh);
1287 return ERR_PTR(err);
1288}
1289
1290unsigned long ext4_count_free_inodes(struct super_block *sb)
1291{
1292 unsigned long desc_count;
1293 struct ext4_group_desc *gdp;
1294 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1295#ifdef EXT4FS_DEBUG
1296 struct ext4_super_block *es;
1297 unsigned long bitmap_count, x;
1298 struct buffer_head *bitmap_bh = NULL;
1299
1300 es = EXT4_SB(sb)->s_es;
1301 desc_count = 0;
1302 bitmap_count = 0;
1303 gdp = NULL;
1304 for (i = 0; i < ngroups; i++) {
1305 gdp = ext4_get_group_desc(sb, i, NULL);
1306 if (!gdp)
1307 continue;
1308 desc_count += ext4_free_inodes_count(sb, gdp);
1309 brelse(bitmap_bh);
1310 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1311 if (IS_ERR(bitmap_bh)) {
1312 bitmap_bh = NULL;
1313 continue;
1314 }
1315
1316 x = ext4_count_free(bitmap_bh->b_data,
1317 EXT4_INODES_PER_GROUP(sb) / 8);
1318 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1319 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1320 bitmap_count += x;
1321 }
1322 brelse(bitmap_bh);
1323 printk(KERN_DEBUG "ext4_count_free_inodes: "
1324 "stored = %u, computed = %lu, %lu\n",
1325 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1326 return desc_count;
1327#else
1328 desc_count = 0;
1329 for (i = 0; i < ngroups; i++) {
1330 gdp = ext4_get_group_desc(sb, i, NULL);
1331 if (!gdp)
1332 continue;
1333 desc_count += ext4_free_inodes_count(sb, gdp);
1334 cond_resched();
1335 }
1336 return desc_count;
1337#endif
1338}
1339
1340/* Called at mount-time, super-block is locked */
1341unsigned long ext4_count_dirs(struct super_block * sb)
1342{
1343 unsigned long count = 0;
1344 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1345
1346 for (i = 0; i < ngroups; i++) {
1347 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1348 if (!gdp)
1349 continue;
1350 count += ext4_used_dirs_count(sb, gdp);
1351 }
1352 return count;
1353}
1354
1355/*
1356 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1357 * inode table. Must be called without any spinlock held. The only place
1358 * where it is called from on active part of filesystem is ext4lazyinit
1359 * thread, so we do not need any special locks, however we have to prevent
1360 * inode allocation from the current group, so we take alloc_sem lock, to
1361 * block ext4_new_inode() until we are finished.
1362 */
1363int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1364 int barrier)
1365{
1366 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1367 struct ext4_sb_info *sbi = EXT4_SB(sb);
1368 struct ext4_group_desc *gdp = NULL;
1369 struct buffer_head *group_desc_bh;
1370 handle_t *handle;
1371 ext4_fsblk_t blk;
1372 int num, ret = 0, used_blks = 0;
1373
1374 /* This should not happen, but just to be sure check this */
1375 if (sb_rdonly(sb)) {
1376 ret = 1;
1377 goto out;
1378 }
1379
1380 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1381 if (!gdp)
1382 goto out;
1383
1384 /*
1385 * We do not need to lock this, because we are the only one
1386 * handling this flag.
1387 */
1388 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1389 goto out;
1390
1391 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1392 if (IS_ERR(handle)) {
1393 ret = PTR_ERR(handle);
1394 goto out;
1395 }
1396
1397 down_write(&grp->alloc_sem);
1398 /*
1399 * If inode bitmap was already initialized there may be some
1400 * used inodes so we need to skip blocks with used inodes in
1401 * inode table.
1402 */
1403 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1404 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1405 ext4_itable_unused_count(sb, gdp)),
1406 sbi->s_inodes_per_block);
1407
1408 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1409 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1410 ext4_itable_unused_count(sb, gdp)) <
1411 EXT4_FIRST_INO(sb)))) {
1412 ext4_error(sb, "Something is wrong with group %u: "
1413 "used itable blocks: %d; "
1414 "itable unused count: %u",
1415 group, used_blks,
1416 ext4_itable_unused_count(sb, gdp));
1417 ret = 1;
1418 goto err_out;
1419 }
1420
1421 blk = ext4_inode_table(sb, gdp) + used_blks;
1422 num = sbi->s_itb_per_group - used_blks;
1423
1424 BUFFER_TRACE(group_desc_bh, "get_write_access");
1425 ret = ext4_journal_get_write_access(handle,
1426 group_desc_bh);
1427 if (ret)
1428 goto err_out;
1429
1430 /*
1431 * Skip zeroout if the inode table is full. But we set the ZEROED
1432 * flag anyway, because obviously, when it is full it does not need
1433 * further zeroing.
1434 */
1435 if (unlikely(num == 0))
1436 goto skip_zeroout;
1437
1438 ext4_debug("going to zero out inode table in group %d\n",
1439 group);
1440 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1441 if (ret < 0)
1442 goto err_out;
1443 if (barrier)
1444 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1445
1446skip_zeroout:
1447 ext4_lock_group(sb, group);
1448 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1449 ext4_group_desc_csum_set(sb, group, gdp);
1450 ext4_unlock_group(sb, group);
1451
1452 BUFFER_TRACE(group_desc_bh,
1453 "call ext4_handle_dirty_metadata");
1454 ret = ext4_handle_dirty_metadata(handle, NULL,
1455 group_desc_bh);
1456
1457err_out:
1458 up_write(&grp->alloc_sem);
1459 ext4_journal_stop(handle);
1460out:
1461 return ret;
1462}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/ialloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
14 */
15
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
24#include <linux/blkdev.h>
25#include <linux/cred.h>
26
27#include <asm/byteorder.h>
28
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34#include <trace/events/ext4.h>
35
36/*
37 * ialloc.c contains the inodes allocation and deallocation routines
38 */
39
40/*
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
44 *
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
48 */
49
50/*
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
54 */
55void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56{
57 int i;
58
59 if (start_bit >= end_bit)
60 return;
61
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
65 if (i < end_bit)
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67}
68
69void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70{
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
74 }
75 unlock_buffer(bh);
76 put_bh(bh);
77}
78
79static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
83{
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp;
86
87 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 return 0;
89
90 grp = ext4_get_group_info(sb, block_group);
91
92 if (buffer_verified(bh))
93 return 0;
94 if (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);
112verified:
113 ext4_unlock_group(sb, block_group);
114 return 0;
115}
116
117/*
118 * Read the inode allocation bitmap for a given block_group, reading
119 * into the specified slot in the superblock's bitmap cache.
120 *
121 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122 */
123static struct buffer_head *
124ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125{
126 struct ext4_group_desc *desc;
127 struct ext4_sb_info *sbi = EXT4_SB(sb);
128 struct buffer_head *bh = NULL;
129 ext4_fsblk_t bitmap_blk;
130 int err;
131
132 desc = ext4_get_group_desc(sb, block_group, NULL);
133 if (!desc)
134 return ERR_PTR(-EFSCORRUPTED);
135
136 bitmap_blk = ext4_inode_bitmap(sb, desc);
137 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 "block_group %u", bitmap_blk, block_group);
141 ext4_mark_group_bitmap_corrupted(sb, block_group,
142 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 return ERR_PTR(-EFSCORRUPTED);
144 }
145 bh = sb_getblk(sb, bitmap_blk);
146 if (unlikely(!bh)) {
147 ext4_warning(sb, "Cannot read inode bitmap - "
148 "block_group = %u, inode_bitmap = %llu",
149 block_group, bitmap_blk);
150 return ERR_PTR(-ENOMEM);
151 }
152 if (bitmap_uptodate(bh))
153 goto verify;
154
155 lock_buffer(bh);
156 if (bitmap_uptodate(bh)) {
157 unlock_buffer(bh);
158 goto verify;
159 }
160
161 ext4_lock_group(sb, block_group);
162 if (ext4_has_group_desc_csum(sb) &&
163 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 if (block_group == 0) {
165 ext4_unlock_group(sb, block_group);
166 unlock_buffer(bh);
167 ext4_error(sb, "Inode bitmap for bg 0 marked "
168 "uninitialized");
169 err = -EFSCORRUPTED;
170 goto out;
171 }
172 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 sb->s_blocksize * 8, bh->b_data);
175 set_bitmap_uptodate(bh);
176 set_buffer_uptodate(bh);
177 set_buffer_verified(bh);
178 ext4_unlock_group(sb, block_group);
179 unlock_buffer(bh);
180 return bh;
181 }
182 ext4_unlock_group(sb, block_group);
183
184 if (buffer_uptodate(bh)) {
185 /*
186 * if not uninit if bh is uptodate,
187 * bitmap is also uptodate
188 */
189 set_bitmap_uptodate(bh);
190 unlock_buffer(bh);
191 goto verify;
192 }
193 /*
194 * submit the buffer_head for reading
195 */
196 trace_ext4_load_inode_bitmap(sb, block_group);
197 ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 if (!buffer_uptodate(bh)) {
200 put_bh(bh);
201 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 "block_group = %u, inode_bitmap = %llu",
203 block_group, bitmap_blk);
204 ext4_mark_group_bitmap_corrupted(sb, block_group,
205 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 return ERR_PTR(-EIO);
207 }
208
209verify:
210 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 if (err)
212 goto out;
213 return bh;
214out:
215 put_bh(bh);
216 return ERR_PTR(err);
217}
218
219/*
220 * NOTE! When we get the inode, we're the only people
221 * that have access to it, and as such there are no
222 * race conditions we have to worry about. The inode
223 * is not on the hash-lists, and it cannot be reached
224 * through the filesystem because the directory entry
225 * has been deleted earlier.
226 *
227 * HOWEVER: we must make sure that we get no aliases,
228 * which means that we have to call "clear_inode()"
229 * _before_ we mark the inode not in use in the inode
230 * bitmaps. Otherwise a newly created file might use
231 * the same inode number (not actually the same pointer
232 * though), and then we'd have two inodes sharing the
233 * same inode number and space on the harddisk.
234 */
235void ext4_free_inode(handle_t *handle, struct inode *inode)
236{
237 struct super_block *sb = inode->i_sb;
238 int is_directory;
239 unsigned long ino;
240 struct buffer_head *bitmap_bh = NULL;
241 struct buffer_head *bh2;
242 ext4_group_t block_group;
243 unsigned long bit;
244 struct ext4_group_desc *gdp;
245 struct ext4_super_block *es;
246 struct ext4_sb_info *sbi;
247 int fatal = 0, err, count, cleared;
248 struct ext4_group_info *grp;
249
250 if (!sb) {
251 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 "nonexistent device\n", __func__, __LINE__);
253 return;
254 }
255 if (atomic_read(&inode->i_count) > 1) {
256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 __func__, __LINE__, inode->i_ino,
258 atomic_read(&inode->i_count));
259 return;
260 }
261 if (inode->i_nlink) {
262 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 return;
265 }
266 sbi = EXT4_SB(sb);
267
268 ino = inode->i_ino;
269 ext4_debug("freeing inode %lu\n", ino);
270 trace_ext4_free_inode(inode);
271
272 dquot_initialize(inode);
273 dquot_free_inode(inode);
274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
278 ext4_clear_inode(inode);
279
280 es = sbi->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 goto error_return;
284 }
285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 /* Don't bother if the inode bitmap is corrupt. */
289 if (IS_ERR(bitmap_bh)) {
290 fatal = PTR_ERR(bitmap_bh);
291 bitmap_bh = NULL;
292 goto error_return;
293 }
294 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 grp = ext4_get_group_info(sb, block_group);
296 if (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 if (percpu_counter_initialized(&sbi->s_dirs_counter))
326 percpu_counter_dec(&sbi->s_dirs_counter);
327 }
328 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
329 EXT4_INODES_PER_GROUP(sb) / 8);
330 ext4_group_desc_csum_set(sb, block_group, gdp);
331 ext4_unlock_group(sb, block_group);
332
333 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
334 percpu_counter_inc(&sbi->s_freeinodes_counter);
335 if (sbi->s_log_groups_per_flex) {
336 struct flex_groups *fg;
337
338 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
339 ext4_flex_group(sbi, block_group));
340 atomic_inc(&fg->free_inodes);
341 if (is_directory)
342 atomic_dec(&fg->used_dirs);
343 }
344 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
345 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
346out:
347 if (cleared) {
348 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
349 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
350 if (!fatal)
351 fatal = err;
352 } else {
353 ext4_error(sb, "bit already cleared for inode %lu", ino);
354 ext4_mark_group_bitmap_corrupted(sb, block_group,
355 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
356 }
357
358error_return:
359 brelse(bitmap_bh);
360 ext4_std_error(sb, fatal);
361}
362
363struct orlov_stats {
364 __u64 free_clusters;
365 __u32 free_inodes;
366 __u32 used_dirs;
367};
368
369/*
370 * Helper function for Orlov's allocator; returns critical information
371 * for a particular block group or flex_bg. If flex_size is 1, then g
372 * is a block group number; otherwise it is flex_bg number.
373 */
374static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
375 int flex_size, struct orlov_stats *stats)
376{
377 struct ext4_group_desc *desc;
378
379 if (flex_size > 1) {
380 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
381 s_flex_groups, g);
382 stats->free_inodes = atomic_read(&fg->free_inodes);
383 stats->free_clusters = atomic64_read(&fg->free_clusters);
384 stats->used_dirs = atomic_read(&fg->used_dirs);
385 return;
386 }
387
388 desc = ext4_get_group_desc(sb, g, NULL);
389 if (desc) {
390 stats->free_inodes = ext4_free_inodes_count(sb, desc);
391 stats->free_clusters = ext4_free_group_clusters(sb, desc);
392 stats->used_dirs = ext4_used_dirs_count(sb, desc);
393 } else {
394 stats->free_inodes = 0;
395 stats->free_clusters = 0;
396 stats->used_dirs = 0;
397 }
398}
399
400/*
401 * Orlov's allocator for directories.
402 *
403 * We always try to spread first-level directories.
404 *
405 * If there are blockgroups with both free inodes and free clusters counts
406 * not worse than average we return one with smallest directory count.
407 * Otherwise we simply return a random group.
408 *
409 * For the rest rules look so:
410 *
411 * It's OK to put directory into a group unless
412 * it has too many directories already (max_dirs) or
413 * it has too few free inodes left (min_inodes) or
414 * it has too few free clusters left (min_clusters) or
415 * Parent's group is preferred, if it doesn't satisfy these
416 * conditions we search cyclically through the rest. If none
417 * of the groups look good we just look for a group with more
418 * free inodes than average (starting at parent's group).
419 */
420
421static int find_group_orlov(struct super_block *sb, struct inode *parent,
422 ext4_group_t *group, umode_t mode,
423 const struct qstr *qstr)
424{
425 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
426 struct ext4_sb_info *sbi = EXT4_SB(sb);
427 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
428 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
429 unsigned int freei, avefreei, grp_free;
430 ext4_fsblk_t freec, avefreec;
431 unsigned int ndirs;
432 int max_dirs, min_inodes;
433 ext4_grpblk_t min_clusters;
434 ext4_group_t i, grp, g, ngroups;
435 struct ext4_group_desc *desc;
436 struct orlov_stats stats;
437 int flex_size = ext4_flex_bg_size(sbi);
438 struct dx_hash_info hinfo;
439
440 ngroups = real_ngroups;
441 if (flex_size > 1) {
442 ngroups = (real_ngroups + flex_size - 1) >>
443 sbi->s_log_groups_per_flex;
444 parent_group >>= sbi->s_log_groups_per_flex;
445 }
446
447 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
448 avefreei = freei / ngroups;
449 freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
450 avefreec = freec;
451 do_div(avefreec, ngroups);
452 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
453
454 if (S_ISDIR(mode) &&
455 ((parent == d_inode(sb->s_root)) ||
456 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
457 int best_ndir = inodes_per_group;
458 int ret = -1;
459
460 if (qstr) {
461 hinfo.hash_version = DX_HASH_HALF_MD4;
462 hinfo.seed = sbi->s_hash_seed;
463 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
464 grp = hinfo.hash;
465 } else
466 grp = prandom_u32();
467 parent_group = (unsigned)grp % ngroups;
468 for (i = 0; i < ngroups; i++) {
469 g = (parent_group + i) % ngroups;
470 get_orlov_stats(sb, g, flex_size, &stats);
471 if (!stats.free_inodes)
472 continue;
473 if (stats.used_dirs >= best_ndir)
474 continue;
475 if (stats.free_inodes < avefreei)
476 continue;
477 if (stats.free_clusters < avefreec)
478 continue;
479 grp = g;
480 ret = 0;
481 best_ndir = stats.used_dirs;
482 }
483 if (ret)
484 goto fallback;
485 found_flex_bg:
486 if (flex_size == 1) {
487 *group = grp;
488 return 0;
489 }
490
491 /*
492 * We pack inodes at the beginning of the flexgroup's
493 * inode tables. Block allocation decisions will do
494 * something similar, although regular files will
495 * start at 2nd block group of the flexgroup. See
496 * ext4_ext_find_goal() and ext4_find_near().
497 */
498 grp *= flex_size;
499 for (i = 0; i < flex_size; i++) {
500 if (grp+i >= real_ngroups)
501 break;
502 desc = ext4_get_group_desc(sb, grp+i, NULL);
503 if (desc && ext4_free_inodes_count(sb, desc)) {
504 *group = grp+i;
505 return 0;
506 }
507 }
508 goto fallback;
509 }
510
511 max_dirs = ndirs / ngroups + inodes_per_group / 16;
512 min_inodes = avefreei - inodes_per_group*flex_size / 4;
513 if (min_inodes < 1)
514 min_inodes = 1;
515 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
516
517 /*
518 * Start looking in the flex group where we last allocated an
519 * inode for this parent directory
520 */
521 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
522 parent_group = EXT4_I(parent)->i_last_alloc_group;
523 if (flex_size > 1)
524 parent_group >>= sbi->s_log_groups_per_flex;
525 }
526
527 for (i = 0; i < ngroups; i++) {
528 grp = (parent_group + i) % ngroups;
529 get_orlov_stats(sb, grp, flex_size, &stats);
530 if (stats.used_dirs >= max_dirs)
531 continue;
532 if (stats.free_inodes < min_inodes)
533 continue;
534 if (stats.free_clusters < min_clusters)
535 continue;
536 goto found_flex_bg;
537 }
538
539fallback:
540 ngroups = real_ngroups;
541 avefreei = freei / ngroups;
542fallback_retry:
543 parent_group = EXT4_I(parent)->i_block_group;
544 for (i = 0; i < ngroups; i++) {
545 grp = (parent_group + i) % ngroups;
546 desc = ext4_get_group_desc(sb, grp, NULL);
547 if (desc) {
548 grp_free = ext4_free_inodes_count(sb, desc);
549 if (grp_free && grp_free >= avefreei) {
550 *group = grp;
551 return 0;
552 }
553 }
554 }
555
556 if (avefreei) {
557 /*
558 * The free-inodes counter is approximate, and for really small
559 * filesystems the above test can fail to find any blockgroups
560 */
561 avefreei = 0;
562 goto fallback_retry;
563 }
564
565 return -1;
566}
567
568static int find_group_other(struct super_block *sb, struct inode *parent,
569 ext4_group_t *group, umode_t mode)
570{
571 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
572 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
573 struct ext4_group_desc *desc;
574 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
575
576 /*
577 * Try to place the inode is the same flex group as its
578 * parent. If we can't find space, use the Orlov algorithm to
579 * find another flex group, and store that information in the
580 * parent directory's inode information so that use that flex
581 * group for future allocations.
582 */
583 if (flex_size > 1) {
584 int retry = 0;
585
586 try_again:
587 parent_group &= ~(flex_size-1);
588 last = parent_group + flex_size;
589 if (last > ngroups)
590 last = ngroups;
591 for (i = parent_group; i < last; i++) {
592 desc = ext4_get_group_desc(sb, i, NULL);
593 if (desc && ext4_free_inodes_count(sb, desc)) {
594 *group = i;
595 return 0;
596 }
597 }
598 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
599 retry = 1;
600 parent_group = EXT4_I(parent)->i_last_alloc_group;
601 goto try_again;
602 }
603 /*
604 * If this didn't work, use the Orlov search algorithm
605 * to find a new flex group; we pass in the mode to
606 * avoid the topdir algorithms.
607 */
608 *group = parent_group + flex_size;
609 if (*group > ngroups)
610 *group = 0;
611 return find_group_orlov(sb, parent, group, mode, NULL);
612 }
613
614 /*
615 * Try to place the inode in its parent directory
616 */
617 *group = parent_group;
618 desc = ext4_get_group_desc(sb, *group, NULL);
619 if (desc && ext4_free_inodes_count(sb, desc) &&
620 ext4_free_group_clusters(sb, desc))
621 return 0;
622
623 /*
624 * We're going to place this inode in a different blockgroup from its
625 * parent. We want to cause files in a common directory to all land in
626 * the same blockgroup. But we want files which are in a different
627 * directory which shares a blockgroup with our parent to land in a
628 * different blockgroup.
629 *
630 * So add our directory's i_ino into the starting point for the hash.
631 */
632 *group = (*group + parent->i_ino) % ngroups;
633
634 /*
635 * Use a quadratic hash to find a group with a free inode and some free
636 * blocks.
637 */
638 for (i = 1; i < ngroups; i <<= 1) {
639 *group += i;
640 if (*group >= ngroups)
641 *group -= ngroups;
642 desc = ext4_get_group_desc(sb, *group, NULL);
643 if (desc && ext4_free_inodes_count(sb, desc) &&
644 ext4_free_group_clusters(sb, desc))
645 return 0;
646 }
647
648 /*
649 * That failed: try linear search for a free inode, even if that group
650 * has no free blocks.
651 */
652 *group = parent_group;
653 for (i = 0; i < ngroups; i++) {
654 if (++*group >= ngroups)
655 *group = 0;
656 desc = ext4_get_group_desc(sb, *group, NULL);
657 if (desc && ext4_free_inodes_count(sb, desc))
658 return 0;
659 }
660
661 return -1;
662}
663
664/*
665 * In no journal mode, if an inode has recently been deleted, we want
666 * to avoid reusing it until we're reasonably sure the inode table
667 * block has been written back to disk. (Yes, these values are
668 * somewhat arbitrary...)
669 */
670#define RECENTCY_MIN 60
671#define RECENTCY_DIRTY 300
672
673static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
674{
675 struct ext4_group_desc *gdp;
676 struct ext4_inode *raw_inode;
677 struct buffer_head *bh;
678 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
679 int offset, ret = 0;
680 int recentcy = RECENTCY_MIN;
681 u32 dtime, now;
682
683 gdp = ext4_get_group_desc(sb, group, NULL);
684 if (unlikely(!gdp))
685 return 0;
686
687 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
688 (ino / inodes_per_block));
689 if (!bh || !buffer_uptodate(bh))
690 /*
691 * If the block is not in the buffer cache, then it
692 * must have been written out.
693 */
694 goto out;
695
696 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
697 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
698
699 /* i_dtime is only 32 bits on disk, but we only care about relative
700 * times in the range of a few minutes (i.e. long enough to sync a
701 * recently-deleted inode to disk), so using the low 32 bits of the
702 * clock (a 68 year range) is enough, see time_before32() */
703 dtime = le32_to_cpu(raw_inode->i_dtime);
704 now = ktime_get_real_seconds();
705 if (buffer_dirty(bh))
706 recentcy += RECENTCY_DIRTY;
707
708 if (dtime && time_before32(dtime, now) &&
709 time_before32(now, dtime + recentcy))
710 ret = 1;
711out:
712 brelse(bh);
713 return ret;
714}
715
716static int find_inode_bit(struct super_block *sb, ext4_group_t group,
717 struct buffer_head *bitmap, unsigned long *ino)
718{
719 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
720 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
721
722next:
723 *ino = ext4_find_next_zero_bit((unsigned long *)
724 bitmap->b_data,
725 EXT4_INODES_PER_GROUP(sb), *ino);
726 if (*ino >= EXT4_INODES_PER_GROUP(sb))
727 goto not_found;
728
729 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
730 recently_deleted_ino = *ino;
731 *ino = *ino + 1;
732 if (*ino < EXT4_INODES_PER_GROUP(sb))
733 goto next;
734 goto not_found;
735 }
736 return 1;
737not_found:
738 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
739 return 0;
740 /*
741 * Not reusing recently deleted inodes is mostly a preference. We don't
742 * want to report ENOSPC or skew allocation patterns because of that.
743 * So return even recently deleted inode if we could find better in the
744 * given range.
745 */
746 *ino = recently_deleted_ino;
747 return 1;
748}
749
750int ext4_mark_inode_used(struct super_block *sb, int ino)
751{
752 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
753 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
754 struct ext4_group_desc *gdp;
755 ext4_group_t group;
756 int bit;
757 int err = -EFSCORRUPTED;
758
759 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
760 goto out;
761
762 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
763 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
764 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
765 if (IS_ERR(inode_bitmap_bh))
766 return PTR_ERR(inode_bitmap_bh);
767
768 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
769 err = 0;
770 goto out;
771 }
772
773 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
774 if (!gdp || !group_desc_bh) {
775 err = -EINVAL;
776 goto out;
777 }
778
779 ext4_set_bit(bit, inode_bitmap_bh->b_data);
780
781 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
782 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
783 if (err) {
784 ext4_std_error(sb, err);
785 goto out;
786 }
787 err = sync_dirty_buffer(inode_bitmap_bh);
788 if (err) {
789 ext4_std_error(sb, err);
790 goto out;
791 }
792
793 /* We may have to initialize the block bitmap if it isn't already */
794 if (ext4_has_group_desc_csum(sb) &&
795 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
796 struct buffer_head *block_bitmap_bh;
797
798 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
799 if (IS_ERR(block_bitmap_bh)) {
800 err = PTR_ERR(block_bitmap_bh);
801 goto out;
802 }
803
804 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
805 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
806 sync_dirty_buffer(block_bitmap_bh);
807
808 /* recheck and clear flag under lock if we still need to */
809 ext4_lock_group(sb, group);
810 if (ext4_has_group_desc_csum(sb) &&
811 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
812 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
813 ext4_free_group_clusters_set(sb, gdp,
814 ext4_free_clusters_after_init(sb, group, gdp));
815 ext4_block_bitmap_csum_set(sb, group, gdp,
816 block_bitmap_bh);
817 ext4_group_desc_csum_set(sb, group, gdp);
818 }
819 ext4_unlock_group(sb, group);
820 brelse(block_bitmap_bh);
821
822 if (err) {
823 ext4_std_error(sb, err);
824 goto out;
825 }
826 }
827
828 /* Update the relevant bg descriptor fields */
829 if (ext4_has_group_desc_csum(sb)) {
830 int free;
831
832 ext4_lock_group(sb, group); /* while we modify the bg desc */
833 free = EXT4_INODES_PER_GROUP(sb) -
834 ext4_itable_unused_count(sb, gdp);
835 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
836 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
837 free = 0;
838 }
839
840 /*
841 * Check the relative inode number against the last used
842 * relative inode number in this group. if it is greater
843 * we need to update the bg_itable_unused count
844 */
845 if (bit >= free)
846 ext4_itable_unused_set(sb, gdp,
847 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
848 } else {
849 ext4_lock_group(sb, group);
850 }
851
852 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
853 if (ext4_has_group_desc_csum(sb)) {
854 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
855 EXT4_INODES_PER_GROUP(sb) / 8);
856 ext4_group_desc_csum_set(sb, group, gdp);
857 }
858
859 ext4_unlock_group(sb, group);
860 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
861 sync_dirty_buffer(group_desc_bh);
862out:
863 return err;
864}
865
866static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
867 bool encrypt)
868{
869 struct super_block *sb = dir->i_sb;
870 int nblocks = 0;
871#ifdef CONFIG_EXT4_FS_POSIX_ACL
872 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
873
874 if (IS_ERR(p))
875 return PTR_ERR(p);
876 if (p) {
877 int acl_size = p->a_count * sizeof(ext4_acl_entry);
878
879 nblocks += (S_ISDIR(mode) ? 2 : 1) *
880 __ext4_xattr_set_credits(sb, NULL /* inode */,
881 NULL /* block_bh */, acl_size,
882 true /* is_create */);
883 posix_acl_release(p);
884 }
885#endif
886
887#ifdef CONFIG_SECURITY
888 {
889 int num_security_xattrs = 1;
890
891#ifdef CONFIG_INTEGRITY
892 num_security_xattrs++;
893#endif
894 /*
895 * We assume that security xattrs are never more than 1k.
896 * In practice they are under 128 bytes.
897 */
898 nblocks += num_security_xattrs *
899 __ext4_xattr_set_credits(sb, NULL /* inode */,
900 NULL /* block_bh */, 1024,
901 true /* is_create */);
902 }
903#endif
904 if (encrypt)
905 nblocks += __ext4_xattr_set_credits(sb,
906 NULL /* inode */,
907 NULL /* block_bh */,
908 FSCRYPT_SET_CONTEXT_MAX_SIZE,
909 true /* is_create */);
910 return nblocks;
911}
912
913/*
914 * There are two policies for allocating an inode. If the new inode is
915 * a directory, then a forward search is made for a block group with both
916 * free space and a low directory-to-inode ratio; if that fails, then of
917 * the groups with above-average free space, that group with the fewest
918 * directories already is chosen.
919 *
920 * For other inodes, search forward from the parent directory's block
921 * group to find a free inode.
922 */
923struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
924 handle_t *handle, struct inode *dir,
925 umode_t mode, const struct qstr *qstr,
926 __u32 goal, uid_t *owner, __u32 i_flags,
927 int handle_type, unsigned int line_no,
928 int nblocks)
929{
930 struct super_block *sb;
931 struct buffer_head *inode_bitmap_bh = NULL;
932 struct buffer_head *group_desc_bh;
933 ext4_group_t ngroups, group = 0;
934 unsigned long ino = 0;
935 struct inode *inode;
936 struct ext4_group_desc *gdp = NULL;
937 struct ext4_inode_info *ei;
938 struct ext4_sb_info *sbi;
939 int ret2, err;
940 struct inode *ret;
941 ext4_group_t i;
942 ext4_group_t flex_group;
943 struct ext4_group_info *grp = NULL;
944 bool encrypt = false;
945
946 /* Cannot create files in a deleted directory */
947 if (!dir || !dir->i_nlink)
948 return ERR_PTR(-EPERM);
949
950 sb = dir->i_sb;
951 sbi = EXT4_SB(sb);
952
953 if (unlikely(ext4_forced_shutdown(sbi)))
954 return ERR_PTR(-EIO);
955
956 ngroups = ext4_get_groups_count(sb);
957 trace_ext4_request_inode(dir, mode);
958 inode = new_inode(sb);
959 if (!inode)
960 return ERR_PTR(-ENOMEM);
961 ei = EXT4_I(inode);
962
963 /*
964 * Initialize owners and quota early so that we don't have to account
965 * for quota initialization worst case in standard inode creating
966 * transaction
967 */
968 if (owner) {
969 inode->i_mode = mode;
970 i_uid_write(inode, owner[0]);
971 i_gid_write(inode, owner[1]);
972 } else if (test_opt(sb, GRPID)) {
973 inode->i_mode = mode;
974 inode_fsuid_set(inode, mnt_userns);
975 inode->i_gid = dir->i_gid;
976 } else
977 inode_init_owner(mnt_userns, inode, dir, mode);
978
979 if (ext4_has_feature_project(sb) &&
980 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
981 ei->i_projid = EXT4_I(dir)->i_projid;
982 else
983 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
984
985 if (!(i_flags & EXT4_EA_INODE_FL)) {
986 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
987 if (err)
988 goto out;
989 }
990
991 err = dquot_initialize(inode);
992 if (err)
993 goto out;
994
995 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
996 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
997 if (ret2 < 0) {
998 err = ret2;
999 goto out;
1000 }
1001 nblocks += ret2;
1002 }
1003
1004 if (!goal)
1005 goal = sbi->s_inode_goal;
1006
1007 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1008 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1009 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1010 ret2 = 0;
1011 goto got_group;
1012 }
1013
1014 if (S_ISDIR(mode))
1015 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1016 else
1017 ret2 = find_group_other(sb, dir, &group, mode);
1018
1019got_group:
1020 EXT4_I(dir)->i_last_alloc_group = group;
1021 err = -ENOSPC;
1022 if (ret2 == -1)
1023 goto out;
1024
1025 /*
1026 * Normally we will only go through one pass of this loop,
1027 * unless we get unlucky and it turns out the group we selected
1028 * had its last inode grabbed by someone else.
1029 */
1030 for (i = 0; i < ngroups; i++, ino = 0) {
1031 err = -EIO;
1032
1033 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1034 if (!gdp)
1035 goto out;
1036
1037 /*
1038 * Check free inodes count before loading bitmap.
1039 */
1040 if (ext4_free_inodes_count(sb, gdp) == 0)
1041 goto next_group;
1042
1043 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1044 grp = ext4_get_group_info(sb, group);
1045 /*
1046 * Skip groups with already-known suspicious inode
1047 * tables
1048 */
1049 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1050 goto next_group;
1051 }
1052
1053 brelse(inode_bitmap_bh);
1054 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1055 /* Skip groups with suspicious inode tables */
1056 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1057 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1058 IS_ERR(inode_bitmap_bh)) {
1059 inode_bitmap_bh = NULL;
1060 goto next_group;
1061 }
1062
1063repeat_in_this_group:
1064 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1065 if (!ret2)
1066 goto next_group;
1067
1068 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1069 ext4_error(sb, "reserved inode found cleared - "
1070 "inode=%lu", ino + 1);
1071 ext4_mark_group_bitmap_corrupted(sb, group,
1072 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1073 goto next_group;
1074 }
1075
1076 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1077 BUG_ON(nblocks <= 0);
1078 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1079 handle_type, nblocks, 0,
1080 ext4_trans_default_revoke_credits(sb));
1081 if (IS_ERR(handle)) {
1082 err = PTR_ERR(handle);
1083 ext4_std_error(sb, err);
1084 goto out;
1085 }
1086 }
1087 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1088 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1089 if (err) {
1090 ext4_std_error(sb, err);
1091 goto out;
1092 }
1093 ext4_lock_group(sb, group);
1094 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1095 if (ret2) {
1096 /* Someone already took the bit. Repeat the search
1097 * with lock held.
1098 */
1099 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1100 if (ret2) {
1101 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1102 ret2 = 0;
1103 } else {
1104 ret2 = 1; /* we didn't grab the inode */
1105 }
1106 }
1107 ext4_unlock_group(sb, group);
1108 ino++; /* the inode bitmap is zero-based */
1109 if (!ret2)
1110 goto got; /* we grabbed the inode! */
1111
1112 if (ino < EXT4_INODES_PER_GROUP(sb))
1113 goto repeat_in_this_group;
1114next_group:
1115 if (++group == ngroups)
1116 group = 0;
1117 }
1118 err = -ENOSPC;
1119 goto out;
1120
1121got:
1122 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1123 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1124 if (err) {
1125 ext4_std_error(sb, err);
1126 goto out;
1127 }
1128
1129 BUFFER_TRACE(group_desc_bh, "get_write_access");
1130 err = ext4_journal_get_write_access(handle, group_desc_bh);
1131 if (err) {
1132 ext4_std_error(sb, err);
1133 goto out;
1134 }
1135
1136 /* We may have to initialize the block bitmap if it isn't already */
1137 if (ext4_has_group_desc_csum(sb) &&
1138 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1139 struct buffer_head *block_bitmap_bh;
1140
1141 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1142 if (IS_ERR(block_bitmap_bh)) {
1143 err = PTR_ERR(block_bitmap_bh);
1144 goto out;
1145 }
1146 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1147 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1148 if (err) {
1149 brelse(block_bitmap_bh);
1150 ext4_std_error(sb, err);
1151 goto out;
1152 }
1153
1154 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1155 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1156
1157 /* recheck and clear flag under lock if we still need to */
1158 ext4_lock_group(sb, group);
1159 if (ext4_has_group_desc_csum(sb) &&
1160 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1161 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1162 ext4_free_group_clusters_set(sb, gdp,
1163 ext4_free_clusters_after_init(sb, group, gdp));
1164 ext4_block_bitmap_csum_set(sb, group, gdp,
1165 block_bitmap_bh);
1166 ext4_group_desc_csum_set(sb, group, gdp);
1167 }
1168 ext4_unlock_group(sb, group);
1169 brelse(block_bitmap_bh);
1170
1171 if (err) {
1172 ext4_std_error(sb, err);
1173 goto out;
1174 }
1175 }
1176
1177 /* Update the relevant bg descriptor fields */
1178 if (ext4_has_group_desc_csum(sb)) {
1179 int free;
1180 struct ext4_group_info *grp = NULL;
1181
1182 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1183 grp = ext4_get_group_info(sb, group);
1184 down_read(&grp->alloc_sem); /*
1185 * protect vs itable
1186 * lazyinit
1187 */
1188 }
1189 ext4_lock_group(sb, group); /* while we modify the bg desc */
1190 free = EXT4_INODES_PER_GROUP(sb) -
1191 ext4_itable_unused_count(sb, gdp);
1192 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1193 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1194 free = 0;
1195 }
1196 /*
1197 * Check the relative inode number against the last used
1198 * relative inode number in this group. if it is greater
1199 * we need to update the bg_itable_unused count
1200 */
1201 if (ino > free)
1202 ext4_itable_unused_set(sb, gdp,
1203 (EXT4_INODES_PER_GROUP(sb) - ino));
1204 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1205 up_read(&grp->alloc_sem);
1206 } else {
1207 ext4_lock_group(sb, group);
1208 }
1209
1210 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1211 if (S_ISDIR(mode)) {
1212 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1213 if (sbi->s_log_groups_per_flex) {
1214 ext4_group_t f = ext4_flex_group(sbi, group);
1215
1216 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1217 f)->used_dirs);
1218 }
1219 }
1220 if (ext4_has_group_desc_csum(sb)) {
1221 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1222 EXT4_INODES_PER_GROUP(sb) / 8);
1223 ext4_group_desc_csum_set(sb, group, gdp);
1224 }
1225 ext4_unlock_group(sb, group);
1226
1227 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1228 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1229 if (err) {
1230 ext4_std_error(sb, err);
1231 goto out;
1232 }
1233
1234 percpu_counter_dec(&sbi->s_freeinodes_counter);
1235 if (S_ISDIR(mode))
1236 percpu_counter_inc(&sbi->s_dirs_counter);
1237
1238 if (sbi->s_log_groups_per_flex) {
1239 flex_group = ext4_flex_group(sbi, group);
1240 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1241 flex_group)->free_inodes);
1242 }
1243
1244 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1245 /* This is the optimal IO size (for stat), not the fs block size */
1246 inode->i_blocks = 0;
1247 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1248 ei->i_crtime = inode->i_mtime;
1249
1250 memset(ei->i_data, 0, sizeof(ei->i_data));
1251 ei->i_dir_start_lookup = 0;
1252 ei->i_disksize = 0;
1253
1254 /* Don't inherit extent flag from directory, amongst others. */
1255 ei->i_flags =
1256 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1257 ei->i_flags |= i_flags;
1258 ei->i_file_acl = 0;
1259 ei->i_dtime = 0;
1260 ei->i_block_group = group;
1261 ei->i_last_alloc_group = ~0;
1262
1263 ext4_set_inode_flags(inode, true);
1264 if (IS_DIRSYNC(inode))
1265 ext4_handle_sync(handle);
1266 if (insert_inode_locked(inode) < 0) {
1267 /*
1268 * Likely a bitmap corruption causing inode to be allocated
1269 * twice.
1270 */
1271 err = -EIO;
1272 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1273 inode->i_ino);
1274 ext4_mark_group_bitmap_corrupted(sb, group,
1275 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1276 goto out;
1277 }
1278 inode->i_generation = prandom_u32();
1279
1280 /* Precompute checksum seed for inode metadata */
1281 if (ext4_has_metadata_csum(sb)) {
1282 __u32 csum;
1283 __le32 inum = cpu_to_le32(inode->i_ino);
1284 __le32 gen = cpu_to_le32(inode->i_generation);
1285 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1286 sizeof(inum));
1287 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1288 sizeof(gen));
1289 }
1290
1291 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1292 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1293
1294 ei->i_extra_isize = sbi->s_want_extra_isize;
1295 ei->i_inline_off = 0;
1296 if (ext4_has_feature_inline_data(sb) &&
1297 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1298 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1299 ret = inode;
1300 err = dquot_alloc_inode(inode);
1301 if (err)
1302 goto fail_drop;
1303
1304 /*
1305 * Since the encryption xattr will always be unique, create it first so
1306 * that it's less likely to end up in an external xattr block and
1307 * prevent its deduplication.
1308 */
1309 if (encrypt) {
1310 err = fscrypt_set_context(inode, handle);
1311 if (err)
1312 goto fail_free_drop;
1313 }
1314
1315 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1316 err = ext4_init_acl(handle, inode, dir);
1317 if (err)
1318 goto fail_free_drop;
1319
1320 err = ext4_init_security(handle, inode, dir, qstr);
1321 if (err)
1322 goto fail_free_drop;
1323 }
1324
1325 if (ext4_has_feature_extents(sb)) {
1326 /* set extent flag only for directory, file and normal symlink*/
1327 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1328 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1329 ext4_ext_tree_init(handle, inode);
1330 }
1331 }
1332
1333 if (ext4_handle_valid(handle)) {
1334 ei->i_sync_tid = handle->h_transaction->t_tid;
1335 ei->i_datasync_tid = handle->h_transaction->t_tid;
1336 }
1337
1338 err = ext4_mark_inode_dirty(handle, inode);
1339 if (err) {
1340 ext4_std_error(sb, err);
1341 goto fail_free_drop;
1342 }
1343
1344 ext4_debug("allocating inode %lu\n", inode->i_ino);
1345 trace_ext4_allocate_inode(inode, dir, mode);
1346 brelse(inode_bitmap_bh);
1347 return ret;
1348
1349fail_free_drop:
1350 dquot_free_inode(inode);
1351fail_drop:
1352 clear_nlink(inode);
1353 unlock_new_inode(inode);
1354out:
1355 dquot_drop(inode);
1356 inode->i_flags |= S_NOQUOTA;
1357 iput(inode);
1358 brelse(inode_bitmap_bh);
1359 return ERR_PTR(err);
1360}
1361
1362/* Verify that we are loading a valid orphan from disk */
1363struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1364{
1365 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1366 ext4_group_t block_group;
1367 int bit;
1368 struct buffer_head *bitmap_bh = NULL;
1369 struct inode *inode = NULL;
1370 int err = -EFSCORRUPTED;
1371
1372 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1373 goto bad_orphan;
1374
1375 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1376 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1377 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1378 if (IS_ERR(bitmap_bh))
1379 return ERR_CAST(bitmap_bh);
1380
1381 /* Having the inode bit set should be a 100% indicator that this
1382 * is a valid orphan (no e2fsck run on fs). Orphans also include
1383 * inodes that were being truncated, so we can't check i_nlink==0.
1384 */
1385 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1386 goto bad_orphan;
1387
1388 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1389 if (IS_ERR(inode)) {
1390 err = PTR_ERR(inode);
1391 ext4_error_err(sb, -err,
1392 "couldn't read orphan inode %lu (err %d)",
1393 ino, err);
1394 brelse(bitmap_bh);
1395 return inode;
1396 }
1397
1398 /*
1399 * If the orphans has i_nlinks > 0 then it should be able to
1400 * be truncated, otherwise it won't be removed from the orphan
1401 * list during processing and an infinite loop will result.
1402 * Similarly, it must not be a bad inode.
1403 */
1404 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1405 is_bad_inode(inode))
1406 goto bad_orphan;
1407
1408 if (NEXT_ORPHAN(inode) > max_ino)
1409 goto bad_orphan;
1410 brelse(bitmap_bh);
1411 return inode;
1412
1413bad_orphan:
1414 ext4_error(sb, "bad orphan inode %lu", ino);
1415 if (bitmap_bh)
1416 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1417 bit, (unsigned long long)bitmap_bh->b_blocknr,
1418 ext4_test_bit(bit, bitmap_bh->b_data));
1419 if (inode) {
1420 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1421 is_bad_inode(inode));
1422 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1423 NEXT_ORPHAN(inode));
1424 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1425 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1426 /* Avoid freeing blocks if we got a bad deleted inode */
1427 if (inode->i_nlink == 0)
1428 inode->i_blocks = 0;
1429 iput(inode);
1430 }
1431 brelse(bitmap_bh);
1432 return ERR_PTR(err);
1433}
1434
1435unsigned long ext4_count_free_inodes(struct super_block *sb)
1436{
1437 unsigned long desc_count;
1438 struct ext4_group_desc *gdp;
1439 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1440#ifdef EXT4FS_DEBUG
1441 struct ext4_super_block *es;
1442 unsigned long bitmap_count, x;
1443 struct buffer_head *bitmap_bh = NULL;
1444
1445 es = EXT4_SB(sb)->s_es;
1446 desc_count = 0;
1447 bitmap_count = 0;
1448 gdp = NULL;
1449 for (i = 0; i < ngroups; i++) {
1450 gdp = ext4_get_group_desc(sb, i, NULL);
1451 if (!gdp)
1452 continue;
1453 desc_count += ext4_free_inodes_count(sb, gdp);
1454 brelse(bitmap_bh);
1455 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1456 if (IS_ERR(bitmap_bh)) {
1457 bitmap_bh = NULL;
1458 continue;
1459 }
1460
1461 x = ext4_count_free(bitmap_bh->b_data,
1462 EXT4_INODES_PER_GROUP(sb) / 8);
1463 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1464 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1465 bitmap_count += x;
1466 }
1467 brelse(bitmap_bh);
1468 printk(KERN_DEBUG "ext4_count_free_inodes: "
1469 "stored = %u, computed = %lu, %lu\n",
1470 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1471 return desc_count;
1472#else
1473 desc_count = 0;
1474 for (i = 0; i < ngroups; i++) {
1475 gdp = ext4_get_group_desc(sb, i, NULL);
1476 if (!gdp)
1477 continue;
1478 desc_count += ext4_free_inodes_count(sb, gdp);
1479 cond_resched();
1480 }
1481 return desc_count;
1482#endif
1483}
1484
1485/* Called at mount-time, super-block is locked */
1486unsigned long ext4_count_dirs(struct super_block * sb)
1487{
1488 unsigned long count = 0;
1489 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1490
1491 for (i = 0; i < ngroups; i++) {
1492 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1493 if (!gdp)
1494 continue;
1495 count += ext4_used_dirs_count(sb, gdp);
1496 }
1497 return count;
1498}
1499
1500/*
1501 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1502 * inode table. Must be called without any spinlock held. The only place
1503 * where it is called from on active part of filesystem is ext4lazyinit
1504 * thread, so we do not need any special locks, however we have to prevent
1505 * inode allocation from the current group, so we take alloc_sem lock, to
1506 * block ext4_new_inode() until we are finished.
1507 */
1508int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1509 int barrier)
1510{
1511 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1512 struct ext4_sb_info *sbi = EXT4_SB(sb);
1513 struct ext4_group_desc *gdp = NULL;
1514 struct buffer_head *group_desc_bh;
1515 handle_t *handle;
1516 ext4_fsblk_t blk;
1517 int num, ret = 0, used_blks = 0;
1518 unsigned long used_inos = 0;
1519
1520 /* This should not happen, but just to be sure check this */
1521 if (sb_rdonly(sb)) {
1522 ret = 1;
1523 goto out;
1524 }
1525
1526 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1527 if (!gdp)
1528 goto out;
1529
1530 /*
1531 * We do not need to lock this, because we are the only one
1532 * handling this flag.
1533 */
1534 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1535 goto out;
1536
1537 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1538 if (IS_ERR(handle)) {
1539 ret = PTR_ERR(handle);
1540 goto out;
1541 }
1542
1543 down_write(&grp->alloc_sem);
1544 /*
1545 * If inode bitmap was already initialized there may be some
1546 * used inodes so we need to skip blocks with used inodes in
1547 * inode table.
1548 */
1549 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1550 used_inos = EXT4_INODES_PER_GROUP(sb) -
1551 ext4_itable_unused_count(sb, gdp);
1552 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1553
1554 /* Bogus inode unused count? */
1555 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1556 ext4_error(sb, "Something is wrong with group %u: "
1557 "used itable blocks: %d; "
1558 "itable unused count: %u",
1559 group, used_blks,
1560 ext4_itable_unused_count(sb, gdp));
1561 ret = 1;
1562 goto err_out;
1563 }
1564
1565 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1566 /*
1567 * Are there some uninitialized inodes in the inode table
1568 * before the first normal inode?
1569 */
1570 if ((used_blks != sbi->s_itb_per_group) &&
1571 (used_inos < EXT4_FIRST_INO(sb))) {
1572 ext4_error(sb, "Something is wrong with group %u: "
1573 "itable unused count: %u; "
1574 "itables initialized count: %ld",
1575 group, ext4_itable_unused_count(sb, gdp),
1576 used_inos);
1577 ret = 1;
1578 goto err_out;
1579 }
1580 }
1581
1582 blk = ext4_inode_table(sb, gdp) + used_blks;
1583 num = sbi->s_itb_per_group - used_blks;
1584
1585 BUFFER_TRACE(group_desc_bh, "get_write_access");
1586 ret = ext4_journal_get_write_access(handle,
1587 group_desc_bh);
1588 if (ret)
1589 goto err_out;
1590
1591 /*
1592 * Skip zeroout if the inode table is full. But we set the ZEROED
1593 * flag anyway, because obviously, when it is full it does not need
1594 * further zeroing.
1595 */
1596 if (unlikely(num == 0))
1597 goto skip_zeroout;
1598
1599 ext4_debug("going to zero out inode table in group %d\n",
1600 group);
1601 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1602 if (ret < 0)
1603 goto err_out;
1604 if (barrier)
1605 blkdev_issue_flush(sb->s_bdev);
1606
1607skip_zeroout:
1608 ext4_lock_group(sb, group);
1609 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1610 ext4_group_desc_csum_set(sb, group, gdp);
1611 ext4_unlock_group(sb, group);
1612
1613 BUFFER_TRACE(group_desc_bh,
1614 "call ext4_handle_dirty_metadata");
1615 ret = ext4_handle_dirty_metadata(handle, NULL,
1616 group_desc_bh);
1617
1618err_out:
1619 up_write(&grp->alloc_sem);
1620 ext4_journal_stop(handle);
1621out:
1622 return ret;
1623}