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 (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 return -EFSCORRUPTED;
96
97 ext4_lock_group(sb, block_group);
98 if (buffer_verified(bh))
99 goto verified;
100 blk = ext4_inode_bitmap(sb, desc);
101 if (!ext4_inode_bitmap_csum_verify(sb, 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 (!grp || unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 fatal = -EFSCORRUPTED;
298 goto error_return;
299 }
300 }
301
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
303 fatal = ext4_journal_get_write_access(handle, 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 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*flex_size / 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
539 fallback:
540 ngroups = real_ngroups;
541 avefreei = freei / ngroups;
542 fallback_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
find_group_other(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode)568 static 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
recently_deleted(struct super_block * sb,ext4_group_t group,int ino)673 static 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;
711 out:
712 brelse(bh);
713 return ret;
714 }
715
find_inode_bit(struct super_block * sb,ext4_group_t group,struct buffer_head * bitmap,unsigned long * ino)716 static 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
722 next:
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;
737 not_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
ext4_mark_inode_used(struct super_block * sb,int ino)750 int 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);
862 out:
863 return err;
864 }
865
ext4_xattr_credits_for_new_inode(struct inode * dir,mode_t mode,bool encrypt)866 static 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 */
__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)923 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
924 umode_t mode, const struct qstr *qstr,
925 __u32 goal, uid_t *owner, __u32 i_flags,
926 int handle_type, unsigned int line_no,
927 int nblocks)
928 {
929 struct super_block *sb;
930 struct buffer_head *inode_bitmap_bh = NULL;
931 struct buffer_head *group_desc_bh;
932 ext4_group_t ngroups, group = 0;
933 unsigned long ino = 0;
934 struct inode *inode;
935 struct ext4_group_desc *gdp = NULL;
936 struct ext4_inode_info *ei;
937 struct ext4_sb_info *sbi;
938 int ret2, err;
939 struct inode *ret;
940 ext4_group_t i;
941 ext4_group_t flex_group;
942 struct ext4_group_info *grp = NULL;
943 bool encrypt = false;
944
945 /* Cannot create files in a deleted directory */
946 if (!dir || !dir->i_nlink)
947 return ERR_PTR(-EPERM);
948
949 sb = dir->i_sb;
950 sbi = EXT4_SB(sb);
951
952 if (unlikely(ext4_forced_shutdown(sbi)))
953 return ERR_PTR(-EIO);
954
955 ngroups = ext4_get_groups_count(sb);
956 trace_ext4_request_inode(dir, mode);
957 inode = new_inode(sb);
958 if (!inode)
959 return ERR_PTR(-ENOMEM);
960 ei = EXT4_I(inode);
961
962 /*
963 * Initialize owners and quota early so that we don't have to account
964 * for quota initialization worst case in standard inode creating
965 * transaction
966 */
967 if (owner) {
968 inode->i_mode = mode;
969 i_uid_write(inode, owner[0]);
970 i_gid_write(inode, owner[1]);
971 } else if (test_opt(sb, GRPID)) {
972 inode->i_mode = mode;
973 inode->i_uid = current_fsuid();
974 inode->i_gid = dir->i_gid;
975 } else
976 inode_init_owner(inode, dir, mode);
977
978 if (ext4_has_feature_project(sb) &&
979 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
980 ei->i_projid = EXT4_I(dir)->i_projid;
981 else
982 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
983
984 if (!(i_flags & EXT4_EA_INODE_FL)) {
985 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
986 if (err)
987 goto out;
988 }
989
990 err = dquot_initialize(inode);
991 if (err)
992 goto out;
993
994 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
995 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
996 if (ret2 < 0) {
997 err = ret2;
998 goto out;
999 }
1000 nblocks += ret2;
1001 }
1002
1003 if (!goal)
1004 goal = sbi->s_inode_goal;
1005
1006 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1007 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1008 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1009 ret2 = 0;
1010 goto got_group;
1011 }
1012
1013 if (S_ISDIR(mode))
1014 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1015 else
1016 ret2 = find_group_other(sb, dir, &group, mode);
1017
1018 got_group:
1019 EXT4_I(dir)->i_last_alloc_group = group;
1020 err = -ENOSPC;
1021 if (ret2 == -1)
1022 goto out;
1023
1024 /*
1025 * Normally we will only go through one pass of this loop,
1026 * unless we get unlucky and it turns out the group we selected
1027 * had its last inode grabbed by someone else.
1028 */
1029 for (i = 0; i < ngroups; i++, ino = 0) {
1030 err = -EIO;
1031
1032 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1033 if (!gdp)
1034 goto out;
1035
1036 /*
1037 * Check free inodes count before loading bitmap.
1038 */
1039 if (ext4_free_inodes_count(sb, gdp) == 0)
1040 goto next_group;
1041
1042 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1043 grp = ext4_get_group_info(sb, group);
1044 /*
1045 * Skip groups with already-known suspicious inode
1046 * tables
1047 */
1048 if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1049 goto next_group;
1050 }
1051
1052 brelse(inode_bitmap_bh);
1053 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1054 /* Skip groups with suspicious inode tables */
1055 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1056 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1057 IS_ERR(inode_bitmap_bh)) {
1058 inode_bitmap_bh = NULL;
1059 goto next_group;
1060 }
1061
1062 repeat_in_this_group:
1063 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1064 if (!ret2)
1065 goto next_group;
1066
1067 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1068 ext4_error(sb, "reserved inode found cleared - "
1069 "inode=%lu", ino + 1);
1070 ext4_mark_group_bitmap_corrupted(sb, group,
1071 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1072 goto next_group;
1073 }
1074
1075 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1076 BUG_ON(nblocks <= 0);
1077 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1078 handle_type, nblocks, 0,
1079 ext4_trans_default_revoke_credits(sb));
1080 if (IS_ERR(handle)) {
1081 err = PTR_ERR(handle);
1082 ext4_std_error(sb, err);
1083 goto out;
1084 }
1085 }
1086 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1087 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1088 if (err) {
1089 ext4_std_error(sb, err);
1090 goto out;
1091 }
1092 ext4_lock_group(sb, group);
1093 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1094 if (ret2) {
1095 /* Someone already took the bit. Repeat the search
1096 * with lock held.
1097 */
1098 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1099 if (ret2) {
1100 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1101 ret2 = 0;
1102 } else {
1103 ret2 = 1; /* we didn't grab the inode */
1104 }
1105 }
1106 ext4_unlock_group(sb, group);
1107 ino++; /* the inode bitmap is zero-based */
1108 if (!ret2)
1109 goto got; /* we grabbed the inode! */
1110
1111 if (ino < EXT4_INODES_PER_GROUP(sb))
1112 goto repeat_in_this_group;
1113 next_group:
1114 if (++group == ngroups)
1115 group = 0;
1116 }
1117 err = -ENOSPC;
1118 goto out;
1119
1120 got:
1121 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1122 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1123 if (err) {
1124 ext4_std_error(sb, err);
1125 goto out;
1126 }
1127
1128 BUFFER_TRACE(group_desc_bh, "get_write_access");
1129 err = ext4_journal_get_write_access(handle, group_desc_bh);
1130 if (err) {
1131 ext4_std_error(sb, err);
1132 goto out;
1133 }
1134
1135 /* We may have to initialize the block bitmap if it isn't already */
1136 if (ext4_has_group_desc_csum(sb) &&
1137 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1138 struct buffer_head *block_bitmap_bh;
1139
1140 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1141 if (IS_ERR(block_bitmap_bh)) {
1142 err = PTR_ERR(block_bitmap_bh);
1143 goto out;
1144 }
1145 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1146 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1147 if (err) {
1148 brelse(block_bitmap_bh);
1149 ext4_std_error(sb, err);
1150 goto out;
1151 }
1152
1153 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1154 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1155
1156 /* recheck and clear flag under lock if we still need to */
1157 ext4_lock_group(sb, group);
1158 if (ext4_has_group_desc_csum(sb) &&
1159 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1160 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1161 ext4_free_group_clusters_set(sb, gdp,
1162 ext4_free_clusters_after_init(sb, group, gdp));
1163 ext4_block_bitmap_csum_set(sb, group, gdp,
1164 block_bitmap_bh);
1165 ext4_group_desc_csum_set(sb, group, gdp);
1166 }
1167 ext4_unlock_group(sb, group);
1168 brelse(block_bitmap_bh);
1169
1170 if (err) {
1171 ext4_std_error(sb, err);
1172 goto out;
1173 }
1174 }
1175
1176 /* Update the relevant bg descriptor fields */
1177 if (ext4_has_group_desc_csum(sb)) {
1178 int free;
1179 struct ext4_group_info *grp = NULL;
1180
1181 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1182 grp = ext4_get_group_info(sb, group);
1183 if (!grp) {
1184 err = -EFSCORRUPTED;
1185 goto out;
1186 }
1187 down_read(&grp->alloc_sem); /*
1188 * protect vs itable
1189 * lazyinit
1190 */
1191 }
1192 ext4_lock_group(sb, group); /* while we modify the bg desc */
1193 free = EXT4_INODES_PER_GROUP(sb) -
1194 ext4_itable_unused_count(sb, gdp);
1195 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1196 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1197 free = 0;
1198 }
1199 /*
1200 * Check the relative inode number against the last used
1201 * relative inode number in this group. if it is greater
1202 * we need to update the bg_itable_unused count
1203 */
1204 if (ino > free)
1205 ext4_itable_unused_set(sb, gdp,
1206 (EXT4_INODES_PER_GROUP(sb) - ino));
1207 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1208 up_read(&grp->alloc_sem);
1209 } else {
1210 ext4_lock_group(sb, group);
1211 }
1212
1213 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1214 if (S_ISDIR(mode)) {
1215 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1216 if (sbi->s_log_groups_per_flex) {
1217 ext4_group_t f = ext4_flex_group(sbi, group);
1218
1219 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1220 f)->used_dirs);
1221 }
1222 }
1223 if (ext4_has_group_desc_csum(sb)) {
1224 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1225 EXT4_INODES_PER_GROUP(sb) / 8);
1226 ext4_group_desc_csum_set(sb, group, gdp);
1227 }
1228 ext4_unlock_group(sb, group);
1229
1230 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1231 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1232 if (err) {
1233 ext4_std_error(sb, err);
1234 goto out;
1235 }
1236
1237 percpu_counter_dec(&sbi->s_freeinodes_counter);
1238 if (S_ISDIR(mode))
1239 percpu_counter_inc(&sbi->s_dirs_counter);
1240
1241 if (sbi->s_log_groups_per_flex) {
1242 flex_group = ext4_flex_group(sbi, group);
1243 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1244 flex_group)->free_inodes);
1245 }
1246
1247 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1248 /* This is the optimal IO size (for stat), not the fs block size */
1249 inode->i_blocks = 0;
1250 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1251 ei->i_crtime = inode->i_mtime;
1252
1253 memset(ei->i_data, 0, sizeof(ei->i_data));
1254 ei->i_dir_start_lookup = 0;
1255 ei->i_disksize = 0;
1256
1257 /* Don't inherit extent flag from directory, amongst others. */
1258 ei->i_flags =
1259 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1260 ei->i_flags |= i_flags;
1261 ei->i_file_acl = 0;
1262 ei->i_dtime = 0;
1263 ei->i_block_group = group;
1264 ei->i_last_alloc_group = ~0;
1265
1266 ext4_set_inode_flags(inode, true);
1267 if (IS_DIRSYNC(inode))
1268 ext4_handle_sync(handle);
1269 if (insert_inode_locked(inode) < 0) {
1270 /*
1271 * Likely a bitmap corruption causing inode to be allocated
1272 * twice.
1273 */
1274 err = -EIO;
1275 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1276 inode->i_ino);
1277 ext4_mark_group_bitmap_corrupted(sb, group,
1278 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1279 goto out;
1280 }
1281 inode->i_generation = prandom_u32();
1282
1283 /* Precompute checksum seed for inode metadata */
1284 if (ext4_has_metadata_csum(sb)) {
1285 __u32 csum;
1286 __le32 inum = cpu_to_le32(inode->i_ino);
1287 __le32 gen = cpu_to_le32(inode->i_generation);
1288 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1289 sizeof(inum));
1290 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1291 sizeof(gen));
1292 }
1293
1294 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1295 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1296
1297 ei->i_extra_isize = sbi->s_want_extra_isize;
1298 ei->i_inline_off = 0;
1299 if (ext4_has_feature_inline_data(sb) &&
1300 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1301 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1302 ret = inode;
1303 err = dquot_alloc_inode(inode);
1304 if (err)
1305 goto fail_drop;
1306
1307 /*
1308 * Since the encryption xattr will always be unique, create it first so
1309 * that it's less likely to end up in an external xattr block and
1310 * prevent its deduplication.
1311 */
1312 if (encrypt) {
1313 err = fscrypt_set_context(inode, handle);
1314 if (err)
1315 goto fail_free_drop;
1316 }
1317
1318 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1319 err = ext4_init_acl(handle, inode, dir);
1320 if (err)
1321 goto fail_free_drop;
1322
1323 err = ext4_init_security(handle, inode, dir, qstr);
1324 if (err)
1325 goto fail_free_drop;
1326 }
1327
1328 if (ext4_has_feature_extents(sb)) {
1329 /* set extent flag only for directory, file and normal symlink*/
1330 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1331 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1332 ext4_ext_tree_init(handle, inode);
1333 }
1334 }
1335
1336 if (ext4_handle_valid(handle)) {
1337 ei->i_sync_tid = handle->h_transaction->t_tid;
1338 ei->i_datasync_tid = handle->h_transaction->t_tid;
1339 }
1340
1341 err = ext4_mark_inode_dirty(handle, inode);
1342 if (err) {
1343 ext4_std_error(sb, err);
1344 goto fail_free_drop;
1345 }
1346
1347 ext4_debug("allocating inode %lu\n", inode->i_ino);
1348 trace_ext4_allocate_inode(inode, dir, mode);
1349 brelse(inode_bitmap_bh);
1350 return ret;
1351
1352 fail_free_drop:
1353 dquot_free_inode(inode);
1354 fail_drop:
1355 clear_nlink(inode);
1356 unlock_new_inode(inode);
1357 out:
1358 dquot_drop(inode);
1359 inode->i_flags |= S_NOQUOTA;
1360 iput(inode);
1361 brelse(inode_bitmap_bh);
1362 return ERR_PTR(err);
1363 }
1364
1365 /* Verify that we are loading a valid orphan from disk */
ext4_orphan_get(struct super_block * sb,unsigned long ino)1366 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1367 {
1368 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1369 ext4_group_t block_group;
1370 int bit;
1371 struct buffer_head *bitmap_bh = NULL;
1372 struct inode *inode = NULL;
1373 int err = -EFSCORRUPTED;
1374
1375 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1376 goto bad_orphan;
1377
1378 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1379 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1380 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1381 if (IS_ERR(bitmap_bh))
1382 return ERR_CAST(bitmap_bh);
1383
1384 /* Having the inode bit set should be a 100% indicator that this
1385 * is a valid orphan (no e2fsck run on fs). Orphans also include
1386 * inodes that were being truncated, so we can't check i_nlink==0.
1387 */
1388 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1389 goto bad_orphan;
1390
1391 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1392 if (IS_ERR(inode)) {
1393 err = PTR_ERR(inode);
1394 ext4_error_err(sb, -err,
1395 "couldn't read orphan inode %lu (err %d)",
1396 ino, err);
1397 brelse(bitmap_bh);
1398 return inode;
1399 }
1400
1401 /*
1402 * If the orphans has i_nlinks > 0 then it should be able to
1403 * be truncated, otherwise it won't be removed from the orphan
1404 * list during processing and an infinite loop will result.
1405 * Similarly, it must not be a bad inode.
1406 */
1407 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1408 is_bad_inode(inode))
1409 goto bad_orphan;
1410
1411 if (NEXT_ORPHAN(inode) > max_ino)
1412 goto bad_orphan;
1413 brelse(bitmap_bh);
1414 return inode;
1415
1416 bad_orphan:
1417 ext4_error(sb, "bad orphan inode %lu", ino);
1418 if (bitmap_bh)
1419 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1420 bit, (unsigned long long)bitmap_bh->b_blocknr,
1421 ext4_test_bit(bit, bitmap_bh->b_data));
1422 if (inode) {
1423 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1424 is_bad_inode(inode));
1425 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1426 NEXT_ORPHAN(inode));
1427 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1428 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1429 /* Avoid freeing blocks if we got a bad deleted inode */
1430 if (inode->i_nlink == 0)
1431 inode->i_blocks = 0;
1432 iput(inode);
1433 }
1434 brelse(bitmap_bh);
1435 return ERR_PTR(err);
1436 }
1437
ext4_count_free_inodes(struct super_block * sb)1438 unsigned long ext4_count_free_inodes(struct super_block *sb)
1439 {
1440 unsigned long desc_count;
1441 struct ext4_group_desc *gdp;
1442 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1443 #ifdef EXT4FS_DEBUG
1444 struct ext4_super_block *es;
1445 unsigned long bitmap_count, x;
1446 struct buffer_head *bitmap_bh = NULL;
1447
1448 es = EXT4_SB(sb)->s_es;
1449 desc_count = 0;
1450 bitmap_count = 0;
1451 gdp = NULL;
1452 for (i = 0; i < ngroups; i++) {
1453 gdp = ext4_get_group_desc(sb, i, NULL);
1454 if (!gdp)
1455 continue;
1456 desc_count += ext4_free_inodes_count(sb, gdp);
1457 brelse(bitmap_bh);
1458 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1459 if (IS_ERR(bitmap_bh)) {
1460 bitmap_bh = NULL;
1461 continue;
1462 }
1463
1464 x = ext4_count_free(bitmap_bh->b_data,
1465 EXT4_INODES_PER_GROUP(sb) / 8);
1466 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1467 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1468 bitmap_count += x;
1469 }
1470 brelse(bitmap_bh);
1471 printk(KERN_DEBUG "ext4_count_free_inodes: "
1472 "stored = %u, computed = %lu, %lu\n",
1473 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1474 return desc_count;
1475 #else
1476 desc_count = 0;
1477 for (i = 0; i < ngroups; i++) {
1478 gdp = ext4_get_group_desc(sb, i, NULL);
1479 if (!gdp)
1480 continue;
1481 desc_count += ext4_free_inodes_count(sb, gdp);
1482 cond_resched();
1483 }
1484 return desc_count;
1485 #endif
1486 }
1487
1488 /* Called at mount-time, super-block is locked */
ext4_count_dirs(struct super_block * sb)1489 unsigned long ext4_count_dirs(struct super_block * sb)
1490 {
1491 unsigned long count = 0;
1492 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1493
1494 for (i = 0; i < ngroups; i++) {
1495 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1496 if (!gdp)
1497 continue;
1498 count += ext4_used_dirs_count(sb, gdp);
1499 }
1500 return count;
1501 }
1502
1503 /*
1504 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1505 * inode table. Must be called without any spinlock held. The only place
1506 * where it is called from on active part of filesystem is ext4lazyinit
1507 * thread, so we do not need any special locks, however we have to prevent
1508 * inode allocation from the current group, so we take alloc_sem lock, to
1509 * block ext4_new_inode() until we are finished.
1510 */
ext4_init_inode_table(struct super_block * sb,ext4_group_t group,int barrier)1511 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1512 int barrier)
1513 {
1514 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1515 struct ext4_sb_info *sbi = EXT4_SB(sb);
1516 struct ext4_group_desc *gdp = NULL;
1517 struct buffer_head *group_desc_bh;
1518 handle_t *handle;
1519 ext4_fsblk_t blk;
1520 int num, ret = 0, used_blks = 0;
1521 unsigned long used_inos = 0;
1522
1523 /* This should not happen, but just to be sure check this */
1524 if (sb_rdonly(sb)) {
1525 ret = 1;
1526 goto out;
1527 }
1528
1529 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1530 if (!gdp || !grp)
1531 goto out;
1532
1533 /*
1534 * We do not need to lock this, because we are the only one
1535 * handling this flag.
1536 */
1537 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1538 goto out;
1539
1540 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1541 if (IS_ERR(handle)) {
1542 ret = PTR_ERR(handle);
1543 goto out;
1544 }
1545
1546 down_write(&grp->alloc_sem);
1547 /*
1548 * If inode bitmap was already initialized there may be some
1549 * used inodes so we need to skip blocks with used inodes in
1550 * inode table.
1551 */
1552 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1553 used_inos = EXT4_INODES_PER_GROUP(sb) -
1554 ext4_itable_unused_count(sb, gdp);
1555 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1556
1557 /* Bogus inode unused count? */
1558 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1559 ext4_error(sb, "Something is wrong with group %u: "
1560 "used itable blocks: %d; "
1561 "itable unused count: %u",
1562 group, used_blks,
1563 ext4_itable_unused_count(sb, gdp));
1564 ret = 1;
1565 goto err_out;
1566 }
1567
1568 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1569 /*
1570 * Are there some uninitialized inodes in the inode table
1571 * before the first normal inode?
1572 */
1573 if ((used_blks != sbi->s_itb_per_group) &&
1574 (used_inos < EXT4_FIRST_INO(sb))) {
1575 ext4_error(sb, "Something is wrong with group %u: "
1576 "itable unused count: %u; "
1577 "itables initialized count: %ld",
1578 group, ext4_itable_unused_count(sb, gdp),
1579 used_inos);
1580 ret = 1;
1581 goto err_out;
1582 }
1583 }
1584
1585 blk = ext4_inode_table(sb, gdp) + used_blks;
1586 num = sbi->s_itb_per_group - used_blks;
1587
1588 BUFFER_TRACE(group_desc_bh, "get_write_access");
1589 ret = ext4_journal_get_write_access(handle,
1590 group_desc_bh);
1591 if (ret)
1592 goto err_out;
1593
1594 /*
1595 * Skip zeroout if the inode table is full. But we set the ZEROED
1596 * flag anyway, because obviously, when it is full it does not need
1597 * further zeroing.
1598 */
1599 if (unlikely(num == 0))
1600 goto skip_zeroout;
1601
1602 ext4_debug("going to zero out inode table in group %d\n",
1603 group);
1604 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1605 if (ret < 0)
1606 goto err_out;
1607 if (barrier)
1608 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1609
1610 skip_zeroout:
1611 ext4_lock_group(sb, group);
1612 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1613 ext4_group_desc_csum_set(sb, group, gdp);
1614 ext4_unlock_group(sb, group);
1615
1616 BUFFER_TRACE(group_desc_bh,
1617 "call ext4_handle_dirty_metadata");
1618 ret = ext4_handle_dirty_metadata(handle, NULL,
1619 group_desc_bh);
1620
1621 err_out:
1622 up_write(&grp->alloc_sem);
1623 ext4_journal_stop(handle);
1624 out:
1625 return ret;
1626 }
1627