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 */
ext4_mark_bitmap_end(int start_bit,int end_bit,char * bitmap)55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56 {
57 int i;
58
59 if (start_bit >= end_bit)
60 return;
61
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
65 if (i < end_bit)
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67 }
68
ext4_end_bitmap_read(struct buffer_head * bh,int uptodate)69 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70 {
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
74 }
75 unlock_buffer(bh);
76 put_bh(bh);
77 }
78
ext4_validate_inode_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)79 static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
83 {
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp;
86
87 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 return 0;
89
90 grp = ext4_get_group_info(sb, block_group);
91
92 if (buffer_verified(bh))
93 return 0;
94 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 return -EFSCORRUPTED;
96
97 ext4_lock_group(sb, block_group);
98 if (buffer_verified(bh))
99 goto verified;
100 blk = ext4_inode_bitmap(sb, desc);
101 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
102 EXT4_INODES_PER_GROUP(sb) / 8) ||
103 ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
104 ext4_unlock_group(sb, block_group);
105 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
106 "inode_bitmap = %llu", block_group, blk);
107 ext4_mark_group_bitmap_corrupted(sb, block_group,
108 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
109 return -EFSBADCRC;
110 }
111 set_buffer_verified(bh);
112 verified:
113 ext4_unlock_group(sb, block_group);
114 return 0;
115 }
116
117 /*
118 * Read the inode allocation bitmap for a given block_group, reading
119 * into the specified slot in the superblock's bitmap cache.
120 *
121 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122 */
123 static struct buffer_head *
ext4_read_inode_bitmap(struct super_block * sb,ext4_group_t block_group)124 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125 {
126 struct ext4_group_desc *desc;
127 struct ext4_sb_info *sbi = EXT4_SB(sb);
128 struct buffer_head *bh = NULL;
129 ext4_fsblk_t bitmap_blk;
130 int err;
131
132 desc = ext4_get_group_desc(sb, block_group, NULL);
133 if (!desc)
134 return ERR_PTR(-EFSCORRUPTED);
135
136 bitmap_blk = ext4_inode_bitmap(sb, desc);
137 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 "block_group %u", bitmap_blk, block_group);
141 ext4_mark_group_bitmap_corrupted(sb, block_group,
142 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 return ERR_PTR(-EFSCORRUPTED);
144 }
145 bh = sb_getblk(sb, bitmap_blk);
146 if (unlikely(!bh)) {
147 ext4_warning(sb, "Cannot read inode bitmap - "
148 "block_group = %u, inode_bitmap = %llu",
149 block_group, bitmap_blk);
150 return ERR_PTR(-ENOMEM);
151 }
152 if (bitmap_uptodate(bh))
153 goto verify;
154
155 lock_buffer(bh);
156 if (bitmap_uptodate(bh)) {
157 unlock_buffer(bh);
158 goto verify;
159 }
160
161 ext4_lock_group(sb, block_group);
162 if (ext4_has_group_desc_csum(sb) &&
163 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 if (block_group == 0) {
165 ext4_unlock_group(sb, block_group);
166 unlock_buffer(bh);
167 ext4_error(sb, "Inode bitmap for bg 0 marked "
168 "uninitialized");
169 err = -EFSCORRUPTED;
170 goto out;
171 }
172 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 sb->s_blocksize * 8, bh->b_data);
175 set_bitmap_uptodate(bh);
176 set_buffer_uptodate(bh);
177 set_buffer_verified(bh);
178 ext4_unlock_group(sb, block_group);
179 unlock_buffer(bh);
180 return bh;
181 }
182 ext4_unlock_group(sb, block_group);
183
184 if (buffer_uptodate(bh)) {
185 /*
186 * if not uninit if bh is uptodate,
187 * bitmap is also uptodate
188 */
189 set_bitmap_uptodate(bh);
190 unlock_buffer(bh);
191 goto verify;
192 }
193 /*
194 * submit the buffer_head for reading
195 */
196 trace_ext4_load_inode_bitmap(sb, block_group);
197 ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 if (!buffer_uptodate(bh)) {
200 put_bh(bh);
201 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 "block_group = %u, inode_bitmap = %llu",
203 block_group, bitmap_blk);
204 ext4_mark_group_bitmap_corrupted(sb, block_group,
205 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 return ERR_PTR(-EIO);
207 }
208
209 verify:
210 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 if (err)
212 goto out;
213 return bh;
214 out:
215 put_bh(bh);
216 return ERR_PTR(err);
217 }
218
219 /*
220 * NOTE! When we get the inode, we're the only people
221 * that have access to it, and as such there are no
222 * race conditions we have to worry about. The inode
223 * is not on the hash-lists, and it cannot be reached
224 * through the filesystem because the directory entry
225 * has been deleted earlier.
226 *
227 * HOWEVER: we must make sure that we get no aliases,
228 * which means that we have to call "clear_inode()"
229 * _before_ we mark the inode not in use in the inode
230 * bitmaps. Otherwise a newly created file might use
231 * the same inode number (not actually the same pointer
232 * though), and then we'd have two inodes sharing the
233 * same inode number and space on the harddisk.
234 */
ext4_free_inode(handle_t * handle,struct inode * inode)235 void ext4_free_inode(handle_t *handle, struct inode *inode)
236 {
237 struct super_block *sb = inode->i_sb;
238 int is_directory;
239 unsigned long ino;
240 struct buffer_head *bitmap_bh = NULL;
241 struct buffer_head *bh2;
242 ext4_group_t block_group;
243 unsigned long bit;
244 struct ext4_group_desc *gdp;
245 struct ext4_super_block *es;
246 struct ext4_sb_info *sbi;
247 int fatal = 0, err, count, cleared;
248 struct ext4_group_info *grp;
249
250 if (!sb) {
251 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 "nonexistent device\n", __func__, __LINE__);
253 return;
254 }
255 if (atomic_read(&inode->i_count) > 1) {
256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 __func__, __LINE__, inode->i_ino,
258 atomic_read(&inode->i_count));
259 return;
260 }
261 if (inode->i_nlink) {
262 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 return;
265 }
266 sbi = EXT4_SB(sb);
267
268 ino = inode->i_ino;
269 ext4_debug("freeing inode %lu\n", ino);
270 trace_ext4_free_inode(inode);
271
272 dquot_initialize(inode);
273 dquot_free_inode(inode);
274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
278 ext4_clear_inode(inode);
279
280 es = sbi->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 goto error_return;
284 }
285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 /* Don't bother if the inode bitmap is corrupt. */
289 if (IS_ERR(bitmap_bh)) {
290 fatal = PTR_ERR(bitmap_bh);
291 bitmap_bh = NULL;
292 goto error_return;
293 }
294 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 grp = ext4_get_group_info(sb, block_group);
296 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 fatal = -EFSCORRUPTED;
298 goto error_return;
299 }
300 }
301
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
303 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
304 if (fatal)
305 goto error_return;
306
307 fatal = -ESRCH;
308 gdp = ext4_get_group_desc(sb, block_group, &bh2);
309 if (gdp) {
310 BUFFER_TRACE(bh2, "get_write_access");
311 fatal = ext4_journal_get_write_access(handle, bh2);
312 }
313 ext4_lock_group(sb, block_group);
314 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
315 if (fatal || !cleared) {
316 ext4_unlock_group(sb, block_group);
317 goto out;
318 }
319
320 count = ext4_free_inodes_count(sb, gdp) + 1;
321 ext4_free_inodes_set(sb, gdp, count);
322 if (is_directory) {
323 count = ext4_used_dirs_count(sb, gdp) - 1;
324 ext4_used_dirs_set(sb, gdp, count);
325 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);
346 out:
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
358 error_return:
359 brelse(bitmap_bh);
360 ext4_std_error(sb, fatal);
361 }
362
363 struct 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 */
get_orlov_stats(struct super_block * sb,ext4_group_t g,int flex_size,struct orlov_stats * stats)374 static 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
find_group_orlov(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode,const struct qstr * qstr)421 static 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 if (ext4_hash_in_dirent(parent))
462 hinfo.hash_version = DX_HASH_SIPHASH;
463 else
464 hinfo.hash_version = DX_HASH_HALF_MD4;
465 hinfo.seed = sbi->s_hash_seed;
466 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
467 grp = hinfo.hash;
468 } else
469 grp = prandom_u32();
470 parent_group = (unsigned)grp % ngroups;
471 for (i = 0; i < ngroups; i++) {
472 g = (parent_group + i) % ngroups;
473 get_orlov_stats(sb, g, flex_size, &stats);
474 if (!stats.free_inodes)
475 continue;
476 if (stats.used_dirs >= best_ndir)
477 continue;
478 if (stats.free_inodes < avefreei)
479 continue;
480 if (stats.free_clusters < avefreec)
481 continue;
482 grp = g;
483 ret = 0;
484 best_ndir = stats.used_dirs;
485 }
486 if (ret)
487 goto fallback;
488 found_flex_bg:
489 if (flex_size == 1) {
490 *group = grp;
491 return 0;
492 }
493
494 /*
495 * We pack inodes at the beginning of the flexgroup's
496 * inode tables. Block allocation decisions will do
497 * something similar, although regular files will
498 * start at 2nd block group of the flexgroup. See
499 * ext4_ext_find_goal() and ext4_find_near().
500 */
501 grp *= flex_size;
502 for (i = 0; i < flex_size; i++) {
503 if (grp+i >= real_ngroups)
504 break;
505 desc = ext4_get_group_desc(sb, grp+i, NULL);
506 if (desc && ext4_free_inodes_count(sb, desc)) {
507 *group = grp+i;
508 return 0;
509 }
510 }
511 goto fallback;
512 }
513
514 max_dirs = ndirs / ngroups + inodes_per_group*flex_size / 16;
515 min_inodes = avefreei - inodes_per_group*flex_size / 4;
516 if (min_inodes < 1)
517 min_inodes = 1;
518 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
519
520 /*
521 * Start looking in the flex group where we last allocated an
522 * inode for this parent directory
523 */
524 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
525 parent_group = EXT4_I(parent)->i_last_alloc_group;
526 if (flex_size > 1)
527 parent_group >>= sbi->s_log_groups_per_flex;
528 }
529
530 for (i = 0; i < ngroups; i++) {
531 grp = (parent_group + i) % ngroups;
532 get_orlov_stats(sb, grp, flex_size, &stats);
533 if (stats.used_dirs >= max_dirs)
534 continue;
535 if (stats.free_inodes < min_inodes)
536 continue;
537 if (stats.free_clusters < min_clusters)
538 continue;
539 goto found_flex_bg;
540 }
541
542 fallback:
543 ngroups = real_ngroups;
544 avefreei = freei / ngroups;
545 fallback_retry:
546 parent_group = EXT4_I(parent)->i_block_group;
547 for (i = 0; i < ngroups; i++) {
548 grp = (parent_group + i) % ngroups;
549 desc = ext4_get_group_desc(sb, grp, NULL);
550 if (desc) {
551 grp_free = ext4_free_inodes_count(sb, desc);
552 if (grp_free && grp_free >= avefreei) {
553 *group = grp;
554 return 0;
555 }
556 }
557 }
558
559 if (avefreei) {
560 /*
561 * The free-inodes counter is approximate, and for really small
562 * filesystems the above test can fail to find any blockgroups
563 */
564 avefreei = 0;
565 goto fallback_retry;
566 }
567
568 return -1;
569 }
570
find_group_other(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode)571 static int find_group_other(struct super_block *sb, struct inode *parent,
572 ext4_group_t *group, umode_t mode)
573 {
574 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
575 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
576 struct ext4_group_desc *desc;
577 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
578
579 /*
580 * Try to place the inode is the same flex group as its
581 * parent. If we can't find space, use the Orlov algorithm to
582 * find another flex group, and store that information in the
583 * parent directory's inode information so that use that flex
584 * group for future allocations.
585 */
586 if (flex_size > 1) {
587 int retry = 0;
588
589 try_again:
590 parent_group &= ~(flex_size-1);
591 last = parent_group + flex_size;
592 if (last > ngroups)
593 last = ngroups;
594 for (i = parent_group; i < last; i++) {
595 desc = ext4_get_group_desc(sb, i, NULL);
596 if (desc && ext4_free_inodes_count(sb, desc)) {
597 *group = i;
598 return 0;
599 }
600 }
601 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
602 retry = 1;
603 parent_group = EXT4_I(parent)->i_last_alloc_group;
604 goto try_again;
605 }
606 /*
607 * If this didn't work, use the Orlov search algorithm
608 * to find a new flex group; we pass in the mode to
609 * avoid the topdir algorithms.
610 */
611 *group = parent_group + flex_size;
612 if (*group > ngroups)
613 *group = 0;
614 return find_group_orlov(sb, parent, group, mode, NULL);
615 }
616
617 /*
618 * Try to place the inode in its parent directory
619 */
620 *group = parent_group;
621 desc = ext4_get_group_desc(sb, *group, NULL);
622 if (desc && ext4_free_inodes_count(sb, desc) &&
623 ext4_free_group_clusters(sb, desc))
624 return 0;
625
626 /*
627 * We're going to place this inode in a different blockgroup from its
628 * parent. We want to cause files in a common directory to all land in
629 * the same blockgroup. But we want files which are in a different
630 * directory which shares a blockgroup with our parent to land in a
631 * different blockgroup.
632 *
633 * So add our directory's i_ino into the starting point for the hash.
634 */
635 *group = (*group + parent->i_ino) % ngroups;
636
637 /*
638 * Use a quadratic hash to find a group with a free inode and some free
639 * blocks.
640 */
641 for (i = 1; i < ngroups; i <<= 1) {
642 *group += i;
643 if (*group >= ngroups)
644 *group -= ngroups;
645 desc = ext4_get_group_desc(sb, *group, NULL);
646 if (desc && ext4_free_inodes_count(sb, desc) &&
647 ext4_free_group_clusters(sb, desc))
648 return 0;
649 }
650
651 /*
652 * That failed: try linear search for a free inode, even if that group
653 * has no free blocks.
654 */
655 *group = parent_group;
656 for (i = 0; i < ngroups; i++) {
657 if (++*group >= ngroups)
658 *group = 0;
659 desc = ext4_get_group_desc(sb, *group, NULL);
660 if (desc && ext4_free_inodes_count(sb, desc))
661 return 0;
662 }
663
664 return -1;
665 }
666
667 /*
668 * In no journal mode, if an inode has recently been deleted, we want
669 * to avoid reusing it until we're reasonably sure the inode table
670 * block has been written back to disk. (Yes, these values are
671 * somewhat arbitrary...)
672 */
673 #define RECENTCY_MIN 60
674 #define RECENTCY_DIRTY 300
675
recently_deleted(struct super_block * sb,ext4_group_t group,int ino)676 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
677 {
678 struct ext4_group_desc *gdp;
679 struct ext4_inode *raw_inode;
680 struct buffer_head *bh;
681 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
682 int offset, ret = 0;
683 int recentcy = RECENTCY_MIN;
684 u32 dtime, now;
685
686 gdp = ext4_get_group_desc(sb, group, NULL);
687 if (unlikely(!gdp))
688 return 0;
689
690 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
691 (ino / inodes_per_block));
692 if (!bh || !buffer_uptodate(bh))
693 /*
694 * If the block is not in the buffer cache, then it
695 * must have been written out.
696 */
697 goto out;
698
699 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
700 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
701
702 /* i_dtime is only 32 bits on disk, but we only care about relative
703 * times in the range of a few minutes (i.e. long enough to sync a
704 * recently-deleted inode to disk), so using the low 32 bits of the
705 * clock (a 68 year range) is enough, see time_before32() */
706 dtime = le32_to_cpu(raw_inode->i_dtime);
707 now = ktime_get_real_seconds();
708 if (buffer_dirty(bh))
709 recentcy += RECENTCY_DIRTY;
710
711 if (dtime && time_before32(dtime, now) &&
712 time_before32(now, dtime + recentcy))
713 ret = 1;
714 out:
715 brelse(bh);
716 return ret;
717 }
718
find_inode_bit(struct super_block * sb,ext4_group_t group,struct buffer_head * bitmap,unsigned long * ino)719 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
720 struct buffer_head *bitmap, unsigned long *ino)
721 {
722 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
723 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
724
725 next:
726 *ino = ext4_find_next_zero_bit((unsigned long *)
727 bitmap->b_data,
728 EXT4_INODES_PER_GROUP(sb), *ino);
729 if (*ino >= EXT4_INODES_PER_GROUP(sb))
730 goto not_found;
731
732 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
733 recently_deleted_ino = *ino;
734 *ino = *ino + 1;
735 if (*ino < EXT4_INODES_PER_GROUP(sb))
736 goto next;
737 goto not_found;
738 }
739 return 1;
740 not_found:
741 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
742 return 0;
743 /*
744 * Not reusing recently deleted inodes is mostly a preference. We don't
745 * want to report ENOSPC or skew allocation patterns because of that.
746 * So return even recently deleted inode if we could find better in the
747 * given range.
748 */
749 *ino = recently_deleted_ino;
750 return 1;
751 }
752
ext4_mark_inode_used(struct super_block * sb,int ino)753 int ext4_mark_inode_used(struct super_block *sb, int ino)
754 {
755 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
756 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
757 struct ext4_group_desc *gdp;
758 ext4_group_t group;
759 int bit;
760 int err = -EFSCORRUPTED;
761
762 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
763 goto out;
764
765 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
766 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
767 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
768 if (IS_ERR(inode_bitmap_bh))
769 return PTR_ERR(inode_bitmap_bh);
770
771 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
772 err = 0;
773 goto out;
774 }
775
776 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
777 if (!gdp || !group_desc_bh) {
778 err = -EINVAL;
779 goto out;
780 }
781
782 ext4_set_bit(bit, inode_bitmap_bh->b_data);
783
784 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
785 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
786 if (err) {
787 ext4_std_error(sb, err);
788 goto out;
789 }
790 err = sync_dirty_buffer(inode_bitmap_bh);
791 if (err) {
792 ext4_std_error(sb, err);
793 goto out;
794 }
795
796 /* We may have to initialize the block bitmap if it isn't already */
797 if (ext4_has_group_desc_csum(sb) &&
798 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
799 struct buffer_head *block_bitmap_bh;
800
801 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
802 if (IS_ERR(block_bitmap_bh)) {
803 err = PTR_ERR(block_bitmap_bh);
804 goto out;
805 }
806
807 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
808 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
809 sync_dirty_buffer(block_bitmap_bh);
810
811 /* recheck and clear flag under lock if we still need to */
812 ext4_lock_group(sb, group);
813 if (ext4_has_group_desc_csum(sb) &&
814 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
815 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
816 ext4_free_group_clusters_set(sb, gdp,
817 ext4_free_clusters_after_init(sb, group, gdp));
818 ext4_block_bitmap_csum_set(sb, group, gdp,
819 block_bitmap_bh);
820 ext4_group_desc_csum_set(sb, group, gdp);
821 }
822 ext4_unlock_group(sb, group);
823 brelse(block_bitmap_bh);
824
825 if (err) {
826 ext4_std_error(sb, err);
827 goto out;
828 }
829 }
830
831 /* Update the relevant bg descriptor fields */
832 if (ext4_has_group_desc_csum(sb)) {
833 int free;
834
835 ext4_lock_group(sb, group); /* while we modify the bg desc */
836 free = EXT4_INODES_PER_GROUP(sb) -
837 ext4_itable_unused_count(sb, gdp);
838 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
839 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
840 free = 0;
841 }
842
843 /*
844 * Check the relative inode number against the last used
845 * relative inode number in this group. if it is greater
846 * we need to update the bg_itable_unused count
847 */
848 if (bit >= free)
849 ext4_itable_unused_set(sb, gdp,
850 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
851 } else {
852 ext4_lock_group(sb, group);
853 }
854
855 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
856 if (ext4_has_group_desc_csum(sb)) {
857 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
858 EXT4_INODES_PER_GROUP(sb) / 8);
859 ext4_group_desc_csum_set(sb, group, gdp);
860 }
861
862 ext4_unlock_group(sb, group);
863 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
864 sync_dirty_buffer(group_desc_bh);
865 out:
866 return err;
867 }
868
ext4_xattr_credits_for_new_inode(struct inode * dir,mode_t mode,bool encrypt)869 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
870 bool encrypt)
871 {
872 struct super_block *sb = dir->i_sb;
873 int nblocks = 0;
874 #ifdef CONFIG_EXT4_FS_POSIX_ACL
875 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
876
877 if (IS_ERR(p))
878 return PTR_ERR(p);
879 if (p) {
880 int acl_size = p->a_count * sizeof(ext4_acl_entry);
881
882 nblocks += (S_ISDIR(mode) ? 2 : 1) *
883 __ext4_xattr_set_credits(sb, NULL /* inode */,
884 NULL /* block_bh */, acl_size,
885 true /* is_create */);
886 posix_acl_release(p);
887 }
888 #endif
889
890 #ifdef CONFIG_SECURITY
891 {
892 int num_security_xattrs = 1;
893
894 #ifdef CONFIG_INTEGRITY
895 num_security_xattrs++;
896 #endif
897 /*
898 * We assume that security xattrs are never more than 1k.
899 * In practice they are under 128 bytes.
900 */
901 nblocks += num_security_xattrs *
902 __ext4_xattr_set_credits(sb, NULL /* inode */,
903 NULL /* block_bh */, 1024,
904 true /* is_create */);
905 }
906 #endif
907 if (encrypt)
908 nblocks += __ext4_xattr_set_credits(sb,
909 NULL /* inode */,
910 NULL /* block_bh */,
911 FSCRYPT_SET_CONTEXT_MAX_SIZE,
912 true /* is_create */);
913 return nblocks;
914 }
915
916 /*
917 * There are two policies for allocating an inode. If the new inode is
918 * a directory, then a forward search is made for a block group with both
919 * free space and a low directory-to-inode ratio; if that fails, then of
920 * the groups with above-average free space, that group with the fewest
921 * directories already is chosen.
922 *
923 * For other inodes, search forward from the parent directory's block
924 * group to find a free inode.
925 */
__ext4_new_inode(handle_t * handle,struct inode * dir,umode_t mode,const struct qstr * qstr,__u32 goal,uid_t * owner,__u32 i_flags,int handle_type,unsigned int line_no,int nblocks)926 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
927 umode_t mode, const struct qstr *qstr,
928 __u32 goal, uid_t *owner, __u32 i_flags,
929 int handle_type, unsigned int line_no,
930 int nblocks)
931 {
932 struct super_block *sb;
933 struct buffer_head *inode_bitmap_bh = NULL;
934 struct buffer_head *group_desc_bh;
935 ext4_group_t ngroups, group = 0;
936 unsigned long ino = 0;
937 struct inode *inode;
938 struct ext4_group_desc *gdp = NULL;
939 struct ext4_inode_info *ei;
940 struct ext4_sb_info *sbi;
941 int ret2, err;
942 struct inode *ret;
943 ext4_group_t i;
944 ext4_group_t flex_group;
945 struct ext4_group_info *grp = NULL;
946 bool encrypt = false;
947
948 /* Cannot create files in a deleted directory */
949 if (!dir || !dir->i_nlink)
950 return ERR_PTR(-EPERM);
951
952 sb = dir->i_sb;
953 sbi = EXT4_SB(sb);
954
955 if (unlikely(ext4_forced_shutdown(sbi)))
956 return ERR_PTR(-EIO);
957
958 ngroups = ext4_get_groups_count(sb);
959 trace_ext4_request_inode(dir, mode);
960 inode = new_inode(sb);
961 if (!inode)
962 return ERR_PTR(-ENOMEM);
963 ei = EXT4_I(inode);
964
965 /*
966 * Initialize owners and quota early so that we don't have to account
967 * for quota initialization worst case in standard inode creating
968 * transaction
969 */
970 if (owner) {
971 inode->i_mode = mode;
972 i_uid_write(inode, owner[0]);
973 i_gid_write(inode, owner[1]);
974 } else if (test_opt(sb, GRPID)) {
975 inode->i_mode = mode;
976 inode->i_uid = current_fsuid();
977 inode->i_gid = dir->i_gid;
978 } else
979 inode_init_owner(inode, dir, mode);
980
981 if (ext4_has_feature_project(sb) &&
982 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
983 ei->i_projid = EXT4_I(dir)->i_projid;
984 else
985 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
986
987 if (!(i_flags & EXT4_EA_INODE_FL)) {
988 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
989 if (err)
990 goto out;
991 }
992
993 err = dquot_initialize(inode);
994 if (err)
995 goto out;
996
997 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
998 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
999 if (ret2 < 0) {
1000 err = ret2;
1001 goto out;
1002 }
1003 nblocks += ret2;
1004 }
1005
1006 if (!goal)
1007 goal = sbi->s_inode_goal;
1008
1009 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1010 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1011 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1012 ret2 = 0;
1013 goto got_group;
1014 }
1015
1016 if (S_ISDIR(mode))
1017 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1018 else
1019 ret2 = find_group_other(sb, dir, &group, mode);
1020
1021 got_group:
1022 EXT4_I(dir)->i_last_alloc_group = group;
1023 err = -ENOSPC;
1024 if (ret2 == -1)
1025 goto out;
1026
1027 /*
1028 * Normally we will only go through one pass of this loop,
1029 * unless we get unlucky and it turns out the group we selected
1030 * had its last inode grabbed by someone else.
1031 */
1032 for (i = 0; i < ngroups; i++, ino = 0) {
1033 err = -EIO;
1034
1035 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1036 if (!gdp)
1037 goto out;
1038
1039 /*
1040 * Check free inodes count before loading bitmap.
1041 */
1042 if (ext4_free_inodes_count(sb, gdp) == 0)
1043 goto next_group;
1044
1045 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1046 grp = ext4_get_group_info(sb, group);
1047 /*
1048 * Skip groups with already-known suspicious inode
1049 * tables
1050 */
1051 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1052 goto next_group;
1053 }
1054
1055 brelse(inode_bitmap_bh);
1056 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1057 /* Skip groups with suspicious inode tables */
1058 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1059 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1060 IS_ERR(inode_bitmap_bh)) {
1061 inode_bitmap_bh = NULL;
1062 goto next_group;
1063 }
1064
1065 repeat_in_this_group:
1066 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1067 if (!ret2)
1068 goto next_group;
1069
1070 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1071 ext4_error(sb, "reserved inode found cleared - "
1072 "inode=%lu", ino + 1);
1073 ext4_mark_group_bitmap_corrupted(sb, group,
1074 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1075 goto next_group;
1076 }
1077
1078 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1079 BUG_ON(nblocks <= 0);
1080 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1081 handle_type, nblocks, 0,
1082 ext4_trans_default_revoke_credits(sb));
1083 if (IS_ERR(handle)) {
1084 err = PTR_ERR(handle);
1085 ext4_std_error(sb, err);
1086 goto out;
1087 }
1088 }
1089 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1090 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1091 if (err) {
1092 ext4_std_error(sb, err);
1093 goto out;
1094 }
1095 ext4_lock_group(sb, group);
1096 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1097 if (ret2) {
1098 /* Someone already took the bit. Repeat the search
1099 * with lock held.
1100 */
1101 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1102 if (ret2) {
1103 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1104 ret2 = 0;
1105 } else {
1106 ret2 = 1; /* we didn't grab the inode */
1107 }
1108 }
1109 ext4_unlock_group(sb, group);
1110 ino++; /* the inode bitmap is zero-based */
1111 if (!ret2)
1112 goto got; /* we grabbed the inode! */
1113
1114 if (ino < EXT4_INODES_PER_GROUP(sb))
1115 goto repeat_in_this_group;
1116 next_group:
1117 if (++group == ngroups)
1118 group = 0;
1119 }
1120 err = -ENOSPC;
1121 goto out;
1122
1123 got:
1124 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1125 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1126 if (err) {
1127 ext4_std_error(sb, err);
1128 goto out;
1129 }
1130
1131 BUFFER_TRACE(group_desc_bh, "get_write_access");
1132 err = ext4_journal_get_write_access(handle, group_desc_bh);
1133 if (err) {
1134 ext4_std_error(sb, err);
1135 goto out;
1136 }
1137
1138 /* We may have to initialize the block bitmap if it isn't already */
1139 if (ext4_has_group_desc_csum(sb) &&
1140 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1141 struct buffer_head *block_bitmap_bh;
1142
1143 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1144 if (IS_ERR(block_bitmap_bh)) {
1145 err = PTR_ERR(block_bitmap_bh);
1146 goto out;
1147 }
1148 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1149 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1150 if (err) {
1151 brelse(block_bitmap_bh);
1152 ext4_std_error(sb, err);
1153 goto out;
1154 }
1155
1156 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1157 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1158
1159 /* recheck and clear flag under lock if we still need to */
1160 ext4_lock_group(sb, group);
1161 if (ext4_has_group_desc_csum(sb) &&
1162 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1163 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1164 ext4_free_group_clusters_set(sb, gdp,
1165 ext4_free_clusters_after_init(sb, group, gdp));
1166 ext4_block_bitmap_csum_set(sb, group, gdp,
1167 block_bitmap_bh);
1168 ext4_group_desc_csum_set(sb, group, gdp);
1169 }
1170 ext4_unlock_group(sb, group);
1171 brelse(block_bitmap_bh);
1172
1173 if (err) {
1174 ext4_std_error(sb, err);
1175 goto out;
1176 }
1177 }
1178
1179 /* Update the relevant bg descriptor fields */
1180 if (ext4_has_group_desc_csum(sb)) {
1181 int free;
1182 struct ext4_group_info *grp = NULL;
1183
1184 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1185 grp = ext4_get_group_info(sb, group);
1186 down_read(&grp->alloc_sem); /*
1187 * protect vs itable
1188 * lazyinit
1189 */
1190 }
1191 ext4_lock_group(sb, group); /* while we modify the bg desc */
1192 free = EXT4_INODES_PER_GROUP(sb) -
1193 ext4_itable_unused_count(sb, gdp);
1194 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1195 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1196 free = 0;
1197 }
1198 /*
1199 * Check the relative inode number against the last used
1200 * relative inode number in this group. if it is greater
1201 * we need to update the bg_itable_unused count
1202 */
1203 if (ino > free)
1204 ext4_itable_unused_set(sb, gdp,
1205 (EXT4_INODES_PER_GROUP(sb) - ino));
1206 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1207 up_read(&grp->alloc_sem);
1208 } else {
1209 ext4_lock_group(sb, group);
1210 }
1211
1212 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1213 if (S_ISDIR(mode)) {
1214 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1215 if (sbi->s_log_groups_per_flex) {
1216 ext4_group_t f = ext4_flex_group(sbi, group);
1217
1218 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1219 f)->used_dirs);
1220 }
1221 }
1222 if (ext4_has_group_desc_csum(sb)) {
1223 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1224 EXT4_INODES_PER_GROUP(sb) / 8);
1225 ext4_group_desc_csum_set(sb, group, gdp);
1226 }
1227 ext4_unlock_group(sb, group);
1228
1229 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1230 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1231 if (err) {
1232 ext4_std_error(sb, err);
1233 goto out;
1234 }
1235
1236 percpu_counter_dec(&sbi->s_freeinodes_counter);
1237 if (S_ISDIR(mode))
1238 percpu_counter_inc(&sbi->s_dirs_counter);
1239
1240 if (sbi->s_log_groups_per_flex) {
1241 flex_group = ext4_flex_group(sbi, group);
1242 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1243 flex_group)->free_inodes);
1244 }
1245
1246 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1247 /* This is the optimal IO size (for stat), not the fs block size */
1248 inode->i_blocks = 0;
1249 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1250 ei->i_crtime = inode->i_mtime;
1251
1252 memset(ei->i_data, 0, sizeof(ei->i_data));
1253 ei->i_dir_start_lookup = 0;
1254 ei->i_disksize = 0;
1255
1256 /* Don't inherit extent flag from directory, amongst others. */
1257 ei->i_flags =
1258 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1259 ei->i_flags |= i_flags;
1260 ei->i_file_acl = 0;
1261 ei->i_dtime = 0;
1262 ei->i_block_group = group;
1263 ei->i_last_alloc_group = ~0;
1264
1265 ext4_set_inode_flags(inode, true);
1266 if (IS_DIRSYNC(inode))
1267 ext4_handle_sync(handle);
1268 if (insert_inode_locked(inode) < 0) {
1269 /*
1270 * Likely a bitmap corruption causing inode to be allocated
1271 * twice.
1272 */
1273 err = -EIO;
1274 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1275 inode->i_ino);
1276 ext4_mark_group_bitmap_corrupted(sb, group,
1277 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1278 goto out;
1279 }
1280 inode->i_generation = prandom_u32();
1281
1282 /* Precompute checksum seed for inode metadata */
1283 if (ext4_has_metadata_csum(sb)) {
1284 __u32 csum;
1285 __le32 inum = cpu_to_le32(inode->i_ino);
1286 __le32 gen = cpu_to_le32(inode->i_generation);
1287 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1288 sizeof(inum));
1289 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1290 sizeof(gen));
1291 }
1292
1293 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1294 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1295
1296 ei->i_extra_isize = sbi->s_want_extra_isize;
1297 ei->i_inline_off = 0;
1298 if (ext4_has_feature_inline_data(sb) &&
1299 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1300 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1301 ret = inode;
1302 err = dquot_alloc_inode(inode);
1303 if (err)
1304 goto fail_drop;
1305
1306 /*
1307 * Since the encryption xattr will always be unique, create it first so
1308 * that it's less likely to end up in an external xattr block and
1309 * prevent its deduplication.
1310 */
1311 if (encrypt) {
1312 err = fscrypt_set_context(inode, handle);
1313 if (err)
1314 goto fail_free_drop;
1315 }
1316
1317 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1318 err = ext4_init_acl(handle, inode, dir);
1319 if (err)
1320 goto fail_free_drop;
1321
1322 err = ext4_init_security(handle, inode, dir, qstr);
1323 if (err)
1324 goto fail_free_drop;
1325 }
1326
1327 if (ext4_has_feature_extents(sb)) {
1328 /* set extent flag only for directory, file and normal symlink*/
1329 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1330 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1331 ext4_ext_tree_init(handle, inode);
1332 }
1333 }
1334
1335 if (ext4_handle_valid(handle)) {
1336 ei->i_sync_tid = handle->h_transaction->t_tid;
1337 ei->i_datasync_tid = handle->h_transaction->t_tid;
1338 }
1339
1340 err = ext4_mark_inode_dirty(handle, inode);
1341 if (err) {
1342 ext4_std_error(sb, err);
1343 goto fail_free_drop;
1344 }
1345
1346 ext4_debug("allocating inode %lu\n", inode->i_ino);
1347 trace_ext4_allocate_inode(inode, dir, mode);
1348 brelse(inode_bitmap_bh);
1349 return ret;
1350
1351 fail_free_drop:
1352 dquot_free_inode(inode);
1353 fail_drop:
1354 clear_nlink(inode);
1355 unlock_new_inode(inode);
1356 out:
1357 dquot_drop(inode);
1358 inode->i_flags |= S_NOQUOTA;
1359 iput(inode);
1360 brelse(inode_bitmap_bh);
1361 return ERR_PTR(err);
1362 }
1363
1364 /* Verify that we are loading a valid orphan from disk */
ext4_orphan_get(struct super_block * sb,unsigned long ino)1365 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1366 {
1367 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1368 ext4_group_t block_group;
1369 int bit;
1370 struct buffer_head *bitmap_bh = NULL;
1371 struct inode *inode = NULL;
1372 int err = -EFSCORRUPTED;
1373
1374 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1375 goto bad_orphan;
1376
1377 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1378 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1379 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1380 if (IS_ERR(bitmap_bh))
1381 return ERR_CAST(bitmap_bh);
1382
1383 /* Having the inode bit set should be a 100% indicator that this
1384 * is a valid orphan (no e2fsck run on fs). Orphans also include
1385 * inodes that were being truncated, so we can't check i_nlink==0.
1386 */
1387 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1388 goto bad_orphan;
1389
1390 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1391 if (IS_ERR(inode)) {
1392 err = PTR_ERR(inode);
1393 ext4_error_err(sb, -err,
1394 "couldn't read orphan inode %lu (err %d)",
1395 ino, err);
1396 brelse(bitmap_bh);
1397 return inode;
1398 }
1399
1400 /*
1401 * If the orphans has i_nlinks > 0 then it should be able to
1402 * be truncated, otherwise it won't be removed from the orphan
1403 * list during processing and an infinite loop will result.
1404 * Similarly, it must not be a bad inode.
1405 */
1406 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1407 is_bad_inode(inode))
1408 goto bad_orphan;
1409
1410 if (NEXT_ORPHAN(inode) > max_ino)
1411 goto bad_orphan;
1412 brelse(bitmap_bh);
1413 return inode;
1414
1415 bad_orphan:
1416 ext4_error(sb, "bad orphan inode %lu", ino);
1417 if (bitmap_bh)
1418 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1419 bit, (unsigned long long)bitmap_bh->b_blocknr,
1420 ext4_test_bit(bit, bitmap_bh->b_data));
1421 if (inode) {
1422 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1423 is_bad_inode(inode));
1424 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1425 NEXT_ORPHAN(inode));
1426 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1427 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1428 /* Avoid freeing blocks if we got a bad deleted inode */
1429 if (inode->i_nlink == 0)
1430 inode->i_blocks = 0;
1431 iput(inode);
1432 }
1433 brelse(bitmap_bh);
1434 return ERR_PTR(err);
1435 }
1436
ext4_count_free_inodes(struct super_block * sb)1437 unsigned long ext4_count_free_inodes(struct super_block *sb)
1438 {
1439 unsigned long desc_count;
1440 struct ext4_group_desc *gdp;
1441 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1442 #ifdef EXT4FS_DEBUG
1443 struct ext4_super_block *es;
1444 unsigned long bitmap_count, x;
1445 struct buffer_head *bitmap_bh = NULL;
1446
1447 es = EXT4_SB(sb)->s_es;
1448 desc_count = 0;
1449 bitmap_count = 0;
1450 gdp = NULL;
1451 for (i = 0; i < ngroups; i++) {
1452 gdp = ext4_get_group_desc(sb, i, NULL);
1453 if (!gdp)
1454 continue;
1455 desc_count += ext4_free_inodes_count(sb, gdp);
1456 brelse(bitmap_bh);
1457 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1458 if (IS_ERR(bitmap_bh)) {
1459 bitmap_bh = NULL;
1460 continue;
1461 }
1462
1463 x = ext4_count_free(bitmap_bh->b_data,
1464 EXT4_INODES_PER_GROUP(sb) / 8);
1465 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1466 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1467 bitmap_count += x;
1468 }
1469 brelse(bitmap_bh);
1470 printk(KERN_DEBUG "ext4_count_free_inodes: "
1471 "stored = %u, computed = %lu, %lu\n",
1472 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1473 return desc_count;
1474 #else
1475 desc_count = 0;
1476 for (i = 0; i < ngroups; i++) {
1477 gdp = ext4_get_group_desc(sb, i, NULL);
1478 if (!gdp)
1479 continue;
1480 desc_count += ext4_free_inodes_count(sb, gdp);
1481 cond_resched();
1482 }
1483 return desc_count;
1484 #endif
1485 }
1486
1487 /* Called at mount-time, super-block is locked */
ext4_count_dirs(struct super_block * sb)1488 unsigned long ext4_count_dirs(struct super_block * sb)
1489 {
1490 unsigned long count = 0;
1491 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1492
1493 for (i = 0; i < ngroups; i++) {
1494 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1495 if (!gdp)
1496 continue;
1497 count += ext4_used_dirs_count(sb, gdp);
1498 }
1499 return count;
1500 }
1501
1502 /*
1503 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1504 * inode table. Must be called without any spinlock held. The only place
1505 * where it is called from on active part of filesystem is ext4lazyinit
1506 * thread, so we do not need any special locks, however we have to prevent
1507 * inode allocation from the current group, so we take alloc_sem lock, to
1508 * block ext4_new_inode() until we are finished.
1509 */
ext4_init_inode_table(struct super_block * sb,ext4_group_t group,int barrier)1510 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1511 int barrier)
1512 {
1513 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1514 struct ext4_sb_info *sbi = EXT4_SB(sb);
1515 struct ext4_group_desc *gdp = NULL;
1516 struct buffer_head *group_desc_bh;
1517 handle_t *handle;
1518 ext4_fsblk_t blk;
1519 int num, ret = 0, used_blks = 0;
1520 unsigned long used_inos = 0;
1521
1522 /* This should not happen, but just to be sure check this */
1523 if (sb_rdonly(sb)) {
1524 ret = 1;
1525 goto out;
1526 }
1527
1528 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1529 if (!gdp)
1530 goto out;
1531
1532 /*
1533 * We do not need to lock this, because we are the only one
1534 * handling this flag.
1535 */
1536 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1537 goto out;
1538
1539 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1540 if (IS_ERR(handle)) {
1541 ret = PTR_ERR(handle);
1542 goto out;
1543 }
1544
1545 down_write(&grp->alloc_sem);
1546 /*
1547 * If inode bitmap was already initialized there may be some
1548 * used inodes so we need to skip blocks with used inodes in
1549 * inode table.
1550 */
1551 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1552 used_inos = EXT4_INODES_PER_GROUP(sb) -
1553 ext4_itable_unused_count(sb, gdp);
1554 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1555
1556 /* Bogus inode unused count? */
1557 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1558 ext4_error(sb, "Something is wrong with group %u: "
1559 "used itable blocks: %d; "
1560 "itable unused count: %u",
1561 group, used_blks,
1562 ext4_itable_unused_count(sb, gdp));
1563 ret = 1;
1564 goto err_out;
1565 }
1566
1567 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1568 /*
1569 * Are there some uninitialized inodes in the inode table
1570 * before the first normal inode?
1571 */
1572 if ((used_blks != sbi->s_itb_per_group) &&
1573 (used_inos < EXT4_FIRST_INO(sb))) {
1574 ext4_error(sb, "Something is wrong with group %u: "
1575 "itable unused count: %u; "
1576 "itables initialized count: %ld",
1577 group, ext4_itable_unused_count(sb, gdp),
1578 used_inos);
1579 ret = 1;
1580 goto err_out;
1581 }
1582 }
1583
1584 blk = ext4_inode_table(sb, gdp) + used_blks;
1585 num = sbi->s_itb_per_group - used_blks;
1586
1587 BUFFER_TRACE(group_desc_bh, "get_write_access");
1588 ret = ext4_journal_get_write_access(handle,
1589 group_desc_bh);
1590 if (ret)
1591 goto err_out;
1592
1593 /*
1594 * Skip zeroout if the inode table is full. But we set the ZEROED
1595 * flag anyway, because obviously, when it is full it does not need
1596 * further zeroing.
1597 */
1598 if (unlikely(num == 0))
1599 goto skip_zeroout;
1600
1601 ext4_debug("going to zero out inode table in group %d\n",
1602 group);
1603 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1604 if (ret < 0)
1605 goto err_out;
1606 if (barrier)
1607 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1608
1609 skip_zeroout:
1610 ext4_lock_group(sb, group);
1611 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1612 ext4_group_desc_csum_set(sb, group, gdp);
1613 ext4_unlock_group(sb, group);
1614
1615 BUFFER_TRACE(group_desc_bh,
1616 "call ext4_handle_dirty_metadata");
1617 ret = ext4_handle_dirty_metadata(handle, NULL,
1618 group_desc_bh);
1619
1620 err_out:
1621 up_write(&grp->alloc_sem);
1622 ext4_journal_stop(handle);
1623 out:
1624 return ret;
1625 }
1626