Loading...
1/*
2 * linux/fs/ext4/ialloc.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * BSD ufs-inspired inode and directory allocation by
10 * Stephen Tweedie (sct@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
13 */
14
15#include <linux/time.h>
16#include <linux/fs.h>
17#include <linux/stat.h>
18#include <linux/string.h>
19#include <linux/quotaops.h>
20#include <linux/buffer_head.h>
21#include <linux/random.h>
22#include <linux/bitops.h>
23#include <linux/blkdev.h>
24#include <asm/byteorder.h>
25
26#include "ext4.h"
27#include "ext4_jbd2.h"
28#include "xattr.h"
29#include "acl.h"
30
31#include <trace/events/ext4.h>
32
33/*
34 * ialloc.c contains the inodes allocation and deallocation routines
35 */
36
37/*
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
41 *
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
45 */
46
47/*
48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
49 * need to use it within a single byte (to ensure we get endianness right).
50 * We can use memset for the rest of the bitmap as there are no other users.
51 */
52void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53{
54 int i;
55
56 if (start_bit >= end_bit)
57 return;
58
59 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 ext4_set_bit(i, bitmap);
62 if (i < end_bit)
63 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64}
65
66/* Initializes an uninitialized inode bitmap */
67static int ext4_init_inode_bitmap(struct super_block *sb,
68 struct buffer_head *bh,
69 ext4_group_t block_group,
70 struct ext4_group_desc *gdp)
71{
72 struct ext4_group_info *grp;
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74 J_ASSERT_BH(bh, buffer_locked(bh));
75
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
78 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
79 grp = ext4_get_group_info(sb, block_group);
80 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
81 percpu_counter_sub(&sbi->s_freeclusters_counter,
82 grp->bb_free);
83 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
84 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
85 int count;
86 count = ext4_free_inodes_count(sb, gdp);
87 percpu_counter_sub(&sbi->s_freeinodes_counter,
88 count);
89 }
90 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
91 return -EFSBADCRC;
92 }
93
94 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
95 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
96 bh->b_data);
97 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
98 EXT4_INODES_PER_GROUP(sb) / 8);
99 ext4_group_desc_csum_set(sb, block_group, gdp);
100
101 return 0;
102}
103
104void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
105{
106 if (uptodate) {
107 set_buffer_uptodate(bh);
108 set_bitmap_uptodate(bh);
109 }
110 unlock_buffer(bh);
111 put_bh(bh);
112}
113
114static int ext4_validate_inode_bitmap(struct super_block *sb,
115 struct ext4_group_desc *desc,
116 ext4_group_t block_group,
117 struct buffer_head *bh)
118{
119 ext4_fsblk_t blk;
120 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
121 struct ext4_sb_info *sbi = EXT4_SB(sb);
122
123 if (buffer_verified(bh))
124 return 0;
125 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
126 return -EFSCORRUPTED;
127
128 ext4_lock_group(sb, block_group);
129 blk = ext4_inode_bitmap(sb, desc);
130 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
131 EXT4_INODES_PER_GROUP(sb) / 8)) {
132 ext4_unlock_group(sb, block_group);
133 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
134 "inode_bitmap = %llu", block_group, blk);
135 grp = ext4_get_group_info(sb, block_group);
136 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
137 int count;
138 count = ext4_free_inodes_count(sb, desc);
139 percpu_counter_sub(&sbi->s_freeinodes_counter,
140 count);
141 }
142 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
143 return -EFSBADCRC;
144 }
145 set_buffer_verified(bh);
146 ext4_unlock_group(sb, block_group);
147 return 0;
148}
149
150/*
151 * Read the inode allocation bitmap for a given block_group, reading
152 * into the specified slot in the superblock's bitmap cache.
153 *
154 * Return buffer_head of bitmap on success or NULL.
155 */
156static struct buffer_head *
157ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
158{
159 struct ext4_group_desc *desc;
160 struct buffer_head *bh = NULL;
161 ext4_fsblk_t bitmap_blk;
162 int err;
163
164 desc = ext4_get_group_desc(sb, block_group, NULL);
165 if (!desc)
166 return ERR_PTR(-EFSCORRUPTED);
167
168 bitmap_blk = ext4_inode_bitmap(sb, desc);
169 bh = sb_getblk(sb, bitmap_blk);
170 if (unlikely(!bh)) {
171 ext4_error(sb, "Cannot read inode bitmap - "
172 "block_group = %u, inode_bitmap = %llu",
173 block_group, bitmap_blk);
174 return ERR_PTR(-EIO);
175 }
176 if (bitmap_uptodate(bh))
177 goto verify;
178
179 lock_buffer(bh);
180 if (bitmap_uptodate(bh)) {
181 unlock_buffer(bh);
182 goto verify;
183 }
184
185 ext4_lock_group(sb, block_group);
186 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
187 err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
188 set_bitmap_uptodate(bh);
189 set_buffer_uptodate(bh);
190 set_buffer_verified(bh);
191 ext4_unlock_group(sb, block_group);
192 unlock_buffer(bh);
193 if (err) {
194 ext4_error(sb, "Failed to init inode bitmap for group "
195 "%u: %d", block_group, err);
196 goto out;
197 }
198 return bh;
199 }
200 ext4_unlock_group(sb, block_group);
201
202 if (buffer_uptodate(bh)) {
203 /*
204 * if not uninit if bh is uptodate,
205 * bitmap is also uptodate
206 */
207 set_bitmap_uptodate(bh);
208 unlock_buffer(bh);
209 goto verify;
210 }
211 /*
212 * submit the buffer_head for reading
213 */
214 trace_ext4_load_inode_bitmap(sb, block_group);
215 bh->b_end_io = ext4_end_bitmap_read;
216 get_bh(bh);
217 submit_bh(READ | REQ_META | REQ_PRIO, bh);
218 wait_on_buffer(bh);
219 if (!buffer_uptodate(bh)) {
220 put_bh(bh);
221 ext4_error(sb, "Cannot read inode bitmap - "
222 "block_group = %u, inode_bitmap = %llu",
223 block_group, bitmap_blk);
224 return ERR_PTR(-EIO);
225 }
226
227verify:
228 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
229 if (err)
230 goto out;
231 return bh;
232out:
233 put_bh(bh);
234 return ERR_PTR(err);
235}
236
237/*
238 * NOTE! When we get the inode, we're the only people
239 * that have access to it, and as such there are no
240 * race conditions we have to worry about. The inode
241 * is not on the hash-lists, and it cannot be reached
242 * through the filesystem because the directory entry
243 * has been deleted earlier.
244 *
245 * HOWEVER: we must make sure that we get no aliases,
246 * which means that we have to call "clear_inode()"
247 * _before_ we mark the inode not in use in the inode
248 * bitmaps. Otherwise a newly created file might use
249 * the same inode number (not actually the same pointer
250 * though), and then we'd have two inodes sharing the
251 * same inode number and space on the harddisk.
252 */
253void ext4_free_inode(handle_t *handle, struct inode *inode)
254{
255 struct super_block *sb = inode->i_sb;
256 int is_directory;
257 unsigned long ino;
258 struct buffer_head *bitmap_bh = NULL;
259 struct buffer_head *bh2;
260 ext4_group_t block_group;
261 unsigned long bit;
262 struct ext4_group_desc *gdp;
263 struct ext4_super_block *es;
264 struct ext4_sb_info *sbi;
265 int fatal = 0, err, count, cleared;
266 struct ext4_group_info *grp;
267
268 if (!sb) {
269 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
270 "nonexistent device\n", __func__, __LINE__);
271 return;
272 }
273 if (atomic_read(&inode->i_count) > 1) {
274 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
275 __func__, __LINE__, inode->i_ino,
276 atomic_read(&inode->i_count));
277 return;
278 }
279 if (inode->i_nlink) {
280 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
281 __func__, __LINE__, inode->i_ino, inode->i_nlink);
282 return;
283 }
284 sbi = EXT4_SB(sb);
285
286 ino = inode->i_ino;
287 ext4_debug("freeing inode %lu\n", ino);
288 trace_ext4_free_inode(inode);
289
290 /*
291 * Note: we must free any quota before locking the superblock,
292 * as writing the quota to disk may need the lock as well.
293 */
294 dquot_initialize(inode);
295 ext4_xattr_delete_inode(handle, inode);
296 dquot_free_inode(inode);
297 dquot_drop(inode);
298
299 is_directory = S_ISDIR(inode->i_mode);
300
301 /* Do this BEFORE marking the inode not in use or returning an error */
302 ext4_clear_inode(inode);
303
304 es = EXT4_SB(sb)->s_es;
305 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
306 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
307 goto error_return;
308 }
309 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
310 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
311 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
312 /* Don't bother if the inode bitmap is corrupt. */
313 grp = ext4_get_group_info(sb, block_group);
314 if (IS_ERR(bitmap_bh)) {
315 fatal = PTR_ERR(bitmap_bh);
316 bitmap_bh = NULL;
317 goto error_return;
318 }
319 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
320 fatal = -EFSCORRUPTED;
321 goto error_return;
322 }
323
324 BUFFER_TRACE(bitmap_bh, "get_write_access");
325 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
326 if (fatal)
327 goto error_return;
328
329 fatal = -ESRCH;
330 gdp = ext4_get_group_desc(sb, block_group, &bh2);
331 if (gdp) {
332 BUFFER_TRACE(bh2, "get_write_access");
333 fatal = ext4_journal_get_write_access(handle, bh2);
334 }
335 ext4_lock_group(sb, block_group);
336 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
337 if (fatal || !cleared) {
338 ext4_unlock_group(sb, block_group);
339 goto out;
340 }
341
342 count = ext4_free_inodes_count(sb, gdp) + 1;
343 ext4_free_inodes_set(sb, gdp, count);
344 if (is_directory) {
345 count = ext4_used_dirs_count(sb, gdp) - 1;
346 ext4_used_dirs_set(sb, gdp, count);
347 percpu_counter_dec(&sbi->s_dirs_counter);
348 }
349 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
350 EXT4_INODES_PER_GROUP(sb) / 8);
351 ext4_group_desc_csum_set(sb, block_group, gdp);
352 ext4_unlock_group(sb, block_group);
353
354 percpu_counter_inc(&sbi->s_freeinodes_counter);
355 if (sbi->s_log_groups_per_flex) {
356 ext4_group_t f = ext4_flex_group(sbi, block_group);
357
358 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
359 if (is_directory)
360 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
361 }
362 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
363 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
364out:
365 if (cleared) {
366 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
367 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
368 if (!fatal)
369 fatal = err;
370 } else {
371 ext4_error(sb, "bit already cleared for inode %lu", ino);
372 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
373 int count;
374 count = ext4_free_inodes_count(sb, gdp);
375 percpu_counter_sub(&sbi->s_freeinodes_counter,
376 count);
377 }
378 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
379 }
380
381error_return:
382 brelse(bitmap_bh);
383 ext4_std_error(sb, fatal);
384}
385
386struct orlov_stats {
387 __u64 free_clusters;
388 __u32 free_inodes;
389 __u32 used_dirs;
390};
391
392/*
393 * Helper function for Orlov's allocator; returns critical information
394 * for a particular block group or flex_bg. If flex_size is 1, then g
395 * is a block group number; otherwise it is flex_bg number.
396 */
397static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
398 int flex_size, struct orlov_stats *stats)
399{
400 struct ext4_group_desc *desc;
401 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
402
403 if (flex_size > 1) {
404 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
405 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
406 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
407 return;
408 }
409
410 desc = ext4_get_group_desc(sb, g, NULL);
411 if (desc) {
412 stats->free_inodes = ext4_free_inodes_count(sb, desc);
413 stats->free_clusters = ext4_free_group_clusters(sb, desc);
414 stats->used_dirs = ext4_used_dirs_count(sb, desc);
415 } else {
416 stats->free_inodes = 0;
417 stats->free_clusters = 0;
418 stats->used_dirs = 0;
419 }
420}
421
422/*
423 * Orlov's allocator for directories.
424 *
425 * We always try to spread first-level directories.
426 *
427 * If there are blockgroups with both free inodes and free blocks counts
428 * not worse than average we return one with smallest directory count.
429 * Otherwise we simply return a random group.
430 *
431 * For the rest rules look so:
432 *
433 * It's OK to put directory into a group unless
434 * it has too many directories already (max_dirs) or
435 * it has too few free inodes left (min_inodes) or
436 * it has too few free blocks left (min_blocks) or
437 * Parent's group is preferred, if it doesn't satisfy these
438 * conditions we search cyclically through the rest. If none
439 * of the groups look good we just look for a group with more
440 * free inodes than average (starting at parent's group).
441 */
442
443static int find_group_orlov(struct super_block *sb, struct inode *parent,
444 ext4_group_t *group, umode_t mode,
445 const struct qstr *qstr)
446{
447 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
448 struct ext4_sb_info *sbi = EXT4_SB(sb);
449 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
450 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
451 unsigned int freei, avefreei, grp_free;
452 ext4_fsblk_t freeb, avefreec;
453 unsigned int ndirs;
454 int max_dirs, min_inodes;
455 ext4_grpblk_t min_clusters;
456 ext4_group_t i, grp, g, ngroups;
457 struct ext4_group_desc *desc;
458 struct orlov_stats stats;
459 int flex_size = ext4_flex_bg_size(sbi);
460 struct dx_hash_info hinfo;
461
462 ngroups = real_ngroups;
463 if (flex_size > 1) {
464 ngroups = (real_ngroups + flex_size - 1) >>
465 sbi->s_log_groups_per_flex;
466 parent_group >>= sbi->s_log_groups_per_flex;
467 }
468
469 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
470 avefreei = freei / ngroups;
471 freeb = EXT4_C2B(sbi,
472 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
473 avefreec = freeb;
474 do_div(avefreec, ngroups);
475 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
476
477 if (S_ISDIR(mode) &&
478 ((parent == d_inode(sb->s_root)) ||
479 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
480 int best_ndir = inodes_per_group;
481 int ret = -1;
482
483 if (qstr) {
484 hinfo.hash_version = DX_HASH_HALF_MD4;
485 hinfo.seed = sbi->s_hash_seed;
486 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
487 grp = hinfo.hash;
488 } else
489 grp = prandom_u32();
490 parent_group = (unsigned)grp % ngroups;
491 for (i = 0; i < ngroups; i++) {
492 g = (parent_group + i) % ngroups;
493 get_orlov_stats(sb, g, flex_size, &stats);
494 if (!stats.free_inodes)
495 continue;
496 if (stats.used_dirs >= best_ndir)
497 continue;
498 if (stats.free_inodes < avefreei)
499 continue;
500 if (stats.free_clusters < avefreec)
501 continue;
502 grp = g;
503 ret = 0;
504 best_ndir = stats.used_dirs;
505 }
506 if (ret)
507 goto fallback;
508 found_flex_bg:
509 if (flex_size == 1) {
510 *group = grp;
511 return 0;
512 }
513
514 /*
515 * We pack inodes at the beginning of the flexgroup's
516 * inode tables. Block allocation decisions will do
517 * something similar, although regular files will
518 * start at 2nd block group of the flexgroup. See
519 * ext4_ext_find_goal() and ext4_find_near().
520 */
521 grp *= flex_size;
522 for (i = 0; i < flex_size; i++) {
523 if (grp+i >= real_ngroups)
524 break;
525 desc = ext4_get_group_desc(sb, grp+i, NULL);
526 if (desc && ext4_free_inodes_count(sb, desc)) {
527 *group = grp+i;
528 return 0;
529 }
530 }
531 goto fallback;
532 }
533
534 max_dirs = ndirs / ngroups + inodes_per_group / 16;
535 min_inodes = avefreei - inodes_per_group*flex_size / 4;
536 if (min_inodes < 1)
537 min_inodes = 1;
538 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
539
540 /*
541 * Start looking in the flex group where we last allocated an
542 * inode for this parent directory
543 */
544 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
545 parent_group = EXT4_I(parent)->i_last_alloc_group;
546 if (flex_size > 1)
547 parent_group >>= sbi->s_log_groups_per_flex;
548 }
549
550 for (i = 0; i < ngroups; i++) {
551 grp = (parent_group + i) % ngroups;
552 get_orlov_stats(sb, grp, flex_size, &stats);
553 if (stats.used_dirs >= max_dirs)
554 continue;
555 if (stats.free_inodes < min_inodes)
556 continue;
557 if (stats.free_clusters < min_clusters)
558 continue;
559 goto found_flex_bg;
560 }
561
562fallback:
563 ngroups = real_ngroups;
564 avefreei = freei / ngroups;
565fallback_retry:
566 parent_group = EXT4_I(parent)->i_block_group;
567 for (i = 0; i < ngroups; i++) {
568 grp = (parent_group + i) % ngroups;
569 desc = ext4_get_group_desc(sb, grp, NULL);
570 if (desc) {
571 grp_free = ext4_free_inodes_count(sb, desc);
572 if (grp_free && grp_free >= avefreei) {
573 *group = grp;
574 return 0;
575 }
576 }
577 }
578
579 if (avefreei) {
580 /*
581 * The free-inodes counter is approximate, and for really small
582 * filesystems the above test can fail to find any blockgroups
583 */
584 avefreei = 0;
585 goto fallback_retry;
586 }
587
588 return -1;
589}
590
591static int find_group_other(struct super_block *sb, struct inode *parent,
592 ext4_group_t *group, umode_t mode)
593{
594 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
595 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
596 struct ext4_group_desc *desc;
597 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
598
599 /*
600 * Try to place the inode is the same flex group as its
601 * parent. If we can't find space, use the Orlov algorithm to
602 * find another flex group, and store that information in the
603 * parent directory's inode information so that use that flex
604 * group for future allocations.
605 */
606 if (flex_size > 1) {
607 int retry = 0;
608
609 try_again:
610 parent_group &= ~(flex_size-1);
611 last = parent_group + flex_size;
612 if (last > ngroups)
613 last = ngroups;
614 for (i = parent_group; i < last; i++) {
615 desc = ext4_get_group_desc(sb, i, NULL);
616 if (desc && ext4_free_inodes_count(sb, desc)) {
617 *group = i;
618 return 0;
619 }
620 }
621 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
622 retry = 1;
623 parent_group = EXT4_I(parent)->i_last_alloc_group;
624 goto try_again;
625 }
626 /*
627 * If this didn't work, use the Orlov search algorithm
628 * to find a new flex group; we pass in the mode to
629 * avoid the topdir algorithms.
630 */
631 *group = parent_group + flex_size;
632 if (*group > ngroups)
633 *group = 0;
634 return find_group_orlov(sb, parent, group, mode, NULL);
635 }
636
637 /*
638 * Try to place the inode in its parent directory
639 */
640 *group = parent_group;
641 desc = ext4_get_group_desc(sb, *group, NULL);
642 if (desc && ext4_free_inodes_count(sb, desc) &&
643 ext4_free_group_clusters(sb, desc))
644 return 0;
645
646 /*
647 * We're going to place this inode in a different blockgroup from its
648 * parent. We want to cause files in a common directory to all land in
649 * the same blockgroup. But we want files which are in a different
650 * directory which shares a blockgroup with our parent to land in a
651 * different blockgroup.
652 *
653 * So add our directory's i_ino into the starting point for the hash.
654 */
655 *group = (*group + parent->i_ino) % ngroups;
656
657 /*
658 * Use a quadratic hash to find a group with a free inode and some free
659 * blocks.
660 */
661 for (i = 1; i < ngroups; i <<= 1) {
662 *group += i;
663 if (*group >= ngroups)
664 *group -= ngroups;
665 desc = ext4_get_group_desc(sb, *group, NULL);
666 if (desc && ext4_free_inodes_count(sb, desc) &&
667 ext4_free_group_clusters(sb, desc))
668 return 0;
669 }
670
671 /*
672 * That failed: try linear search for a free inode, even if that group
673 * has no free blocks.
674 */
675 *group = parent_group;
676 for (i = 0; i < ngroups; i++) {
677 if (++*group >= ngroups)
678 *group = 0;
679 desc = ext4_get_group_desc(sb, *group, NULL);
680 if (desc && ext4_free_inodes_count(sb, desc))
681 return 0;
682 }
683
684 return -1;
685}
686
687/*
688 * In no journal mode, if an inode has recently been deleted, we want
689 * to avoid reusing it until we're reasonably sure the inode table
690 * block has been written back to disk. (Yes, these values are
691 * somewhat arbitrary...)
692 */
693#define RECENTCY_MIN 5
694#define RECENTCY_DIRTY 30
695
696static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
697{
698 struct ext4_group_desc *gdp;
699 struct ext4_inode *raw_inode;
700 struct buffer_head *bh;
701 unsigned long dtime, now;
702 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
703 int offset, ret = 0, recentcy = RECENTCY_MIN;
704
705 gdp = ext4_get_group_desc(sb, group, NULL);
706 if (unlikely(!gdp))
707 return 0;
708
709 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
710 (ino / inodes_per_block));
711 if (unlikely(!bh) || !buffer_uptodate(bh))
712 /*
713 * If the block is not in the buffer cache, then it
714 * must have been written out.
715 */
716 goto out;
717
718 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
719 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
720 dtime = le32_to_cpu(raw_inode->i_dtime);
721 now = get_seconds();
722 if (buffer_dirty(bh))
723 recentcy += RECENTCY_DIRTY;
724
725 if (dtime && (dtime < now) && (now < dtime + recentcy))
726 ret = 1;
727out:
728 brelse(bh);
729 return ret;
730}
731
732/*
733 * There are two policies for allocating an inode. If the new inode is
734 * a directory, then a forward search is made for a block group with both
735 * free space and a low directory-to-inode ratio; if that fails, then of
736 * the groups with above-average free space, that group with the fewest
737 * directories already is chosen.
738 *
739 * For other inodes, search forward from the parent directory's block
740 * group to find a free inode.
741 */
742struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
743 umode_t mode, const struct qstr *qstr,
744 __u32 goal, uid_t *owner, int handle_type,
745 unsigned int line_no, int nblocks)
746{
747 struct super_block *sb;
748 struct buffer_head *inode_bitmap_bh = NULL;
749 struct buffer_head *group_desc_bh;
750 ext4_group_t ngroups, group = 0;
751 unsigned long ino = 0;
752 struct inode *inode;
753 struct ext4_group_desc *gdp = NULL;
754 struct ext4_inode_info *ei;
755 struct ext4_sb_info *sbi;
756 int ret2, err;
757 struct inode *ret;
758 ext4_group_t i;
759 ext4_group_t flex_group;
760 struct ext4_group_info *grp;
761 int encrypt = 0;
762
763 /* Cannot create files in a deleted directory */
764 if (!dir || !dir->i_nlink)
765 return ERR_PTR(-EPERM);
766
767 if ((ext4_encrypted_inode(dir) ||
768 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
769 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
770 err = ext4_get_encryption_info(dir);
771 if (err)
772 return ERR_PTR(err);
773 if (ext4_encryption_info(dir) == NULL)
774 return ERR_PTR(-EPERM);
775 if (!handle)
776 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
777 encrypt = 1;
778 }
779
780 sb = dir->i_sb;
781 ngroups = ext4_get_groups_count(sb);
782 trace_ext4_request_inode(dir, mode);
783 inode = new_inode(sb);
784 if (!inode)
785 return ERR_PTR(-ENOMEM);
786 ei = EXT4_I(inode);
787 sbi = EXT4_SB(sb);
788
789 /*
790 * Initialize owners and quota early so that we don't have to account
791 * for quota initialization worst case in standard inode creating
792 * transaction
793 */
794 if (owner) {
795 inode->i_mode = mode;
796 i_uid_write(inode, owner[0]);
797 i_gid_write(inode, owner[1]);
798 } else if (test_opt(sb, GRPID)) {
799 inode->i_mode = mode;
800 inode->i_uid = current_fsuid();
801 inode->i_gid = dir->i_gid;
802 } else
803 inode_init_owner(inode, dir, mode);
804
805 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
806 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
807 ei->i_projid = EXT4_I(dir)->i_projid;
808 else
809 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
810
811 err = dquot_initialize(inode);
812 if (err)
813 goto out;
814
815 if (!goal)
816 goal = sbi->s_inode_goal;
817
818 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
819 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
820 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
821 ret2 = 0;
822 goto got_group;
823 }
824
825 if (S_ISDIR(mode))
826 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
827 else
828 ret2 = find_group_other(sb, dir, &group, mode);
829
830got_group:
831 EXT4_I(dir)->i_last_alloc_group = group;
832 err = -ENOSPC;
833 if (ret2 == -1)
834 goto out;
835
836 /*
837 * Normally we will only go through one pass of this loop,
838 * unless we get unlucky and it turns out the group we selected
839 * had its last inode grabbed by someone else.
840 */
841 for (i = 0; i < ngroups; i++, ino = 0) {
842 err = -EIO;
843
844 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
845 if (!gdp)
846 goto out;
847
848 /*
849 * Check free inodes count before loading bitmap.
850 */
851 if (ext4_free_inodes_count(sb, gdp) == 0) {
852 if (++group == ngroups)
853 group = 0;
854 continue;
855 }
856
857 grp = ext4_get_group_info(sb, group);
858 /* Skip groups with already-known suspicious inode tables */
859 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
860 if (++group == ngroups)
861 group = 0;
862 continue;
863 }
864
865 brelse(inode_bitmap_bh);
866 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
867 /* Skip groups with suspicious inode tables */
868 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
869 IS_ERR(inode_bitmap_bh)) {
870 inode_bitmap_bh = NULL;
871 if (++group == ngroups)
872 group = 0;
873 continue;
874 }
875
876repeat_in_this_group:
877 ino = ext4_find_next_zero_bit((unsigned long *)
878 inode_bitmap_bh->b_data,
879 EXT4_INODES_PER_GROUP(sb), ino);
880 if (ino >= EXT4_INODES_PER_GROUP(sb))
881 goto next_group;
882 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
883 ext4_error(sb, "reserved inode found cleared - "
884 "inode=%lu", ino + 1);
885 continue;
886 }
887 if ((EXT4_SB(sb)->s_journal == NULL) &&
888 recently_deleted(sb, group, ino)) {
889 ino++;
890 goto next_inode;
891 }
892 if (!handle) {
893 BUG_ON(nblocks <= 0);
894 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
895 handle_type, nblocks,
896 0);
897 if (IS_ERR(handle)) {
898 err = PTR_ERR(handle);
899 ext4_std_error(sb, err);
900 goto out;
901 }
902 }
903 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
904 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
905 if (err) {
906 ext4_std_error(sb, err);
907 goto out;
908 }
909 ext4_lock_group(sb, group);
910 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
911 ext4_unlock_group(sb, group);
912 ino++; /* the inode bitmap is zero-based */
913 if (!ret2)
914 goto got; /* we grabbed the inode! */
915next_inode:
916 if (ino < EXT4_INODES_PER_GROUP(sb))
917 goto repeat_in_this_group;
918next_group:
919 if (++group == ngroups)
920 group = 0;
921 }
922 err = -ENOSPC;
923 goto out;
924
925got:
926 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
927 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
928 if (err) {
929 ext4_std_error(sb, err);
930 goto out;
931 }
932
933 BUFFER_TRACE(group_desc_bh, "get_write_access");
934 err = ext4_journal_get_write_access(handle, group_desc_bh);
935 if (err) {
936 ext4_std_error(sb, err);
937 goto out;
938 }
939
940 /* We may have to initialize the block bitmap if it isn't already */
941 if (ext4_has_group_desc_csum(sb) &&
942 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
943 struct buffer_head *block_bitmap_bh;
944
945 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
946 if (IS_ERR(block_bitmap_bh)) {
947 err = PTR_ERR(block_bitmap_bh);
948 goto out;
949 }
950 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
951 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
952 if (err) {
953 brelse(block_bitmap_bh);
954 ext4_std_error(sb, err);
955 goto out;
956 }
957
958 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
959 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
960
961 /* recheck and clear flag under lock if we still need to */
962 ext4_lock_group(sb, group);
963 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
964 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
965 ext4_free_group_clusters_set(sb, gdp,
966 ext4_free_clusters_after_init(sb, group, gdp));
967 ext4_block_bitmap_csum_set(sb, group, gdp,
968 block_bitmap_bh);
969 ext4_group_desc_csum_set(sb, group, gdp);
970 }
971 ext4_unlock_group(sb, group);
972 brelse(block_bitmap_bh);
973
974 if (err) {
975 ext4_std_error(sb, err);
976 goto out;
977 }
978 }
979
980 /* Update the relevant bg descriptor fields */
981 if (ext4_has_group_desc_csum(sb)) {
982 int free;
983 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
984
985 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
986 ext4_lock_group(sb, group); /* while we modify the bg desc */
987 free = EXT4_INODES_PER_GROUP(sb) -
988 ext4_itable_unused_count(sb, gdp);
989 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
990 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
991 free = 0;
992 }
993 /*
994 * Check the relative inode number against the last used
995 * relative inode number in this group. if it is greater
996 * we need to update the bg_itable_unused count
997 */
998 if (ino > free)
999 ext4_itable_unused_set(sb, gdp,
1000 (EXT4_INODES_PER_GROUP(sb) - ino));
1001 up_read(&grp->alloc_sem);
1002 } else {
1003 ext4_lock_group(sb, group);
1004 }
1005
1006 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1007 if (S_ISDIR(mode)) {
1008 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1009 if (sbi->s_log_groups_per_flex) {
1010 ext4_group_t f = ext4_flex_group(sbi, group);
1011
1012 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1013 }
1014 }
1015 if (ext4_has_group_desc_csum(sb)) {
1016 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1017 EXT4_INODES_PER_GROUP(sb) / 8);
1018 ext4_group_desc_csum_set(sb, group, gdp);
1019 }
1020 ext4_unlock_group(sb, group);
1021
1022 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1023 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1024 if (err) {
1025 ext4_std_error(sb, err);
1026 goto out;
1027 }
1028
1029 percpu_counter_dec(&sbi->s_freeinodes_counter);
1030 if (S_ISDIR(mode))
1031 percpu_counter_inc(&sbi->s_dirs_counter);
1032
1033 if (sbi->s_log_groups_per_flex) {
1034 flex_group = ext4_flex_group(sbi, group);
1035 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1036 }
1037
1038 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1039 /* This is the optimal IO size (for stat), not the fs block size */
1040 inode->i_blocks = 0;
1041 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1042 ext4_current_time(inode);
1043
1044 memset(ei->i_data, 0, sizeof(ei->i_data));
1045 ei->i_dir_start_lookup = 0;
1046 ei->i_disksize = 0;
1047
1048 /* Don't inherit extent flag from directory, amongst others. */
1049 ei->i_flags =
1050 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1051 ei->i_file_acl = 0;
1052 ei->i_dtime = 0;
1053 ei->i_block_group = group;
1054 ei->i_last_alloc_group = ~0;
1055
1056 ext4_set_inode_flags(inode);
1057 if (IS_DIRSYNC(inode))
1058 ext4_handle_sync(handle);
1059 if (insert_inode_locked(inode) < 0) {
1060 /*
1061 * Likely a bitmap corruption causing inode to be allocated
1062 * twice.
1063 */
1064 err = -EIO;
1065 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1066 inode->i_ino);
1067 goto out;
1068 }
1069 spin_lock(&sbi->s_next_gen_lock);
1070 inode->i_generation = sbi->s_next_generation++;
1071 spin_unlock(&sbi->s_next_gen_lock);
1072
1073 /* Precompute checksum seed for inode metadata */
1074 if (ext4_has_metadata_csum(sb)) {
1075 __u32 csum;
1076 __le32 inum = cpu_to_le32(inode->i_ino);
1077 __le32 gen = cpu_to_le32(inode->i_generation);
1078 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1079 sizeof(inum));
1080 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1081 sizeof(gen));
1082 }
1083
1084 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1085 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1086
1087 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1088 ei->i_inline_off = 0;
1089 if (ext4_has_feature_inline_data(sb))
1090 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1091 ret = inode;
1092 err = dquot_alloc_inode(inode);
1093 if (err)
1094 goto fail_drop;
1095
1096 err = ext4_init_acl(handle, inode, dir);
1097 if (err)
1098 goto fail_free_drop;
1099
1100 err = ext4_init_security(handle, inode, dir, qstr);
1101 if (err)
1102 goto fail_free_drop;
1103
1104 if (ext4_has_feature_extents(sb)) {
1105 /* set extent flag only for directory, file and normal symlink*/
1106 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1107 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1108 ext4_ext_tree_init(handle, inode);
1109 }
1110 }
1111
1112 if (ext4_handle_valid(handle)) {
1113 ei->i_sync_tid = handle->h_transaction->t_tid;
1114 ei->i_datasync_tid = handle->h_transaction->t_tid;
1115 }
1116
1117 if (encrypt) {
1118 err = ext4_inherit_context(dir, inode);
1119 if (err)
1120 goto fail_free_drop;
1121 }
1122
1123 err = ext4_mark_inode_dirty(handle, inode);
1124 if (err) {
1125 ext4_std_error(sb, err);
1126 goto fail_free_drop;
1127 }
1128
1129 ext4_debug("allocating inode %lu\n", inode->i_ino);
1130 trace_ext4_allocate_inode(inode, dir, mode);
1131 brelse(inode_bitmap_bh);
1132 return ret;
1133
1134fail_free_drop:
1135 dquot_free_inode(inode);
1136fail_drop:
1137 clear_nlink(inode);
1138 unlock_new_inode(inode);
1139out:
1140 dquot_drop(inode);
1141 inode->i_flags |= S_NOQUOTA;
1142 iput(inode);
1143 brelse(inode_bitmap_bh);
1144 return ERR_PTR(err);
1145}
1146
1147/* Verify that we are loading a valid orphan from disk */
1148struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1149{
1150 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1151 ext4_group_t block_group;
1152 int bit;
1153 struct buffer_head *bitmap_bh;
1154 struct inode *inode = NULL;
1155 long err = -EIO;
1156
1157 /* Error cases - e2fsck has already cleaned up for us */
1158 if (ino > max_ino) {
1159 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1160 err = -EFSCORRUPTED;
1161 goto error;
1162 }
1163
1164 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1165 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1166 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1167 if (IS_ERR(bitmap_bh)) {
1168 err = PTR_ERR(bitmap_bh);
1169 ext4_warning(sb, "inode bitmap error %ld for orphan %lu",
1170 ino, err);
1171 goto error;
1172 }
1173
1174 /* Having the inode bit set should be a 100% indicator that this
1175 * is a valid orphan (no e2fsck run on fs). Orphans also include
1176 * inodes that were being truncated, so we can't check i_nlink==0.
1177 */
1178 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1179 goto bad_orphan;
1180
1181 inode = ext4_iget(sb, ino);
1182 if (IS_ERR(inode))
1183 goto iget_failed;
1184
1185 /*
1186 * If the orphans has i_nlinks > 0 then it should be able to be
1187 * truncated, otherwise it won't be removed from the orphan list
1188 * during processing and an infinite loop will result.
1189 */
1190 if (inode->i_nlink && !ext4_can_truncate(inode))
1191 goto bad_orphan;
1192
1193 if (NEXT_ORPHAN(inode) > max_ino)
1194 goto bad_orphan;
1195 brelse(bitmap_bh);
1196 return inode;
1197
1198iget_failed:
1199 err = PTR_ERR(inode);
1200 inode = NULL;
1201bad_orphan:
1202 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1203 printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1204 bit, (unsigned long long)bitmap_bh->b_blocknr,
1205 ext4_test_bit(bit, bitmap_bh->b_data));
1206 printk(KERN_WARNING "inode=%p\n", inode);
1207 if (inode) {
1208 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
1209 is_bad_inode(inode));
1210 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
1211 NEXT_ORPHAN(inode));
1212 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1213 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
1214 /* Avoid freeing blocks if we got a bad deleted inode */
1215 if (inode->i_nlink == 0)
1216 inode->i_blocks = 0;
1217 iput(inode);
1218 }
1219 brelse(bitmap_bh);
1220error:
1221 return ERR_PTR(err);
1222}
1223
1224unsigned long ext4_count_free_inodes(struct super_block *sb)
1225{
1226 unsigned long desc_count;
1227 struct ext4_group_desc *gdp;
1228 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1229#ifdef EXT4FS_DEBUG
1230 struct ext4_super_block *es;
1231 unsigned long bitmap_count, x;
1232 struct buffer_head *bitmap_bh = NULL;
1233
1234 es = EXT4_SB(sb)->s_es;
1235 desc_count = 0;
1236 bitmap_count = 0;
1237 gdp = NULL;
1238 for (i = 0; i < ngroups; i++) {
1239 gdp = ext4_get_group_desc(sb, i, NULL);
1240 if (!gdp)
1241 continue;
1242 desc_count += ext4_free_inodes_count(sb, gdp);
1243 brelse(bitmap_bh);
1244 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1245 if (IS_ERR(bitmap_bh)) {
1246 bitmap_bh = NULL;
1247 continue;
1248 }
1249
1250 x = ext4_count_free(bitmap_bh->b_data,
1251 EXT4_INODES_PER_GROUP(sb) / 8);
1252 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1253 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1254 bitmap_count += x;
1255 }
1256 brelse(bitmap_bh);
1257 printk(KERN_DEBUG "ext4_count_free_inodes: "
1258 "stored = %u, computed = %lu, %lu\n",
1259 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1260 return desc_count;
1261#else
1262 desc_count = 0;
1263 for (i = 0; i < ngroups; i++) {
1264 gdp = ext4_get_group_desc(sb, i, NULL);
1265 if (!gdp)
1266 continue;
1267 desc_count += ext4_free_inodes_count(sb, gdp);
1268 cond_resched();
1269 }
1270 return desc_count;
1271#endif
1272}
1273
1274/* Called at mount-time, super-block is locked */
1275unsigned long ext4_count_dirs(struct super_block * sb)
1276{
1277 unsigned long count = 0;
1278 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1279
1280 for (i = 0; i < ngroups; i++) {
1281 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1282 if (!gdp)
1283 continue;
1284 count += ext4_used_dirs_count(sb, gdp);
1285 }
1286 return count;
1287}
1288
1289/*
1290 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1291 * inode table. Must be called without any spinlock held. The only place
1292 * where it is called from on active part of filesystem is ext4lazyinit
1293 * thread, so we do not need any special locks, however we have to prevent
1294 * inode allocation from the current group, so we take alloc_sem lock, to
1295 * block ext4_new_inode() until we are finished.
1296 */
1297int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1298 int barrier)
1299{
1300 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1301 struct ext4_sb_info *sbi = EXT4_SB(sb);
1302 struct ext4_group_desc *gdp = NULL;
1303 struct buffer_head *group_desc_bh;
1304 handle_t *handle;
1305 ext4_fsblk_t blk;
1306 int num, ret = 0, used_blks = 0;
1307
1308 /* This should not happen, but just to be sure check this */
1309 if (sb->s_flags & MS_RDONLY) {
1310 ret = 1;
1311 goto out;
1312 }
1313
1314 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1315 if (!gdp)
1316 goto out;
1317
1318 /*
1319 * We do not need to lock this, because we are the only one
1320 * handling this flag.
1321 */
1322 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1323 goto out;
1324
1325 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1326 if (IS_ERR(handle)) {
1327 ret = PTR_ERR(handle);
1328 goto out;
1329 }
1330
1331 down_write(&grp->alloc_sem);
1332 /*
1333 * If inode bitmap was already initialized there may be some
1334 * used inodes so we need to skip blocks with used inodes in
1335 * inode table.
1336 */
1337 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1338 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1339 ext4_itable_unused_count(sb, gdp)),
1340 sbi->s_inodes_per_block);
1341
1342 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1343 ext4_error(sb, "Something is wrong with group %u: "
1344 "used itable blocks: %d; "
1345 "itable unused count: %u",
1346 group, used_blks,
1347 ext4_itable_unused_count(sb, gdp));
1348 ret = 1;
1349 goto err_out;
1350 }
1351
1352 blk = ext4_inode_table(sb, gdp) + used_blks;
1353 num = sbi->s_itb_per_group - used_blks;
1354
1355 BUFFER_TRACE(group_desc_bh, "get_write_access");
1356 ret = ext4_journal_get_write_access(handle,
1357 group_desc_bh);
1358 if (ret)
1359 goto err_out;
1360
1361 /*
1362 * Skip zeroout if the inode table is full. But we set the ZEROED
1363 * flag anyway, because obviously, when it is full it does not need
1364 * further zeroing.
1365 */
1366 if (unlikely(num == 0))
1367 goto skip_zeroout;
1368
1369 ext4_debug("going to zero out inode table in group %d\n",
1370 group);
1371 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1372 if (ret < 0)
1373 goto err_out;
1374 if (barrier)
1375 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1376
1377skip_zeroout:
1378 ext4_lock_group(sb, group);
1379 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1380 ext4_group_desc_csum_set(sb, group, gdp);
1381 ext4_unlock_group(sb, group);
1382
1383 BUFFER_TRACE(group_desc_bh,
1384 "call ext4_handle_dirty_metadata");
1385 ret = ext4_handle_dirty_metadata(handle, NULL,
1386 group_desc_bh);
1387
1388err_out:
1389 up_write(&grp->alloc_sem);
1390 ext4_journal_stop(handle);
1391out:
1392 return ret;
1393}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/ialloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
14 */
15
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/stat.h>
19#include <linux/string.h>
20#include <linux/quotaops.h>
21#include <linux/buffer_head.h>
22#include <linux/random.h>
23#include <linux/bitops.h>
24#include <linux/blkdev.h>
25#include <linux/cred.h>
26
27#include <asm/byteorder.h>
28
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34#include <trace/events/ext4.h>
35
36/*
37 * ialloc.c contains the inodes allocation and deallocation routines
38 */
39
40/*
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
44 *
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
48 */
49
50/*
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
54 */
55void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56{
57 int i;
58
59 if (start_bit >= end_bit)
60 return;
61
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
65 if (i < end_bit)
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67}
68
69void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70{
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
74 }
75 unlock_buffer(bh);
76 put_bh(bh);
77}
78
79static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
83{
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp;
86
87 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 return 0;
89
90 grp = ext4_get_group_info(sb, block_group);
91
92 if (buffer_verified(bh))
93 return 0;
94 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 return -EFSCORRUPTED;
96
97 ext4_lock_group(sb, block_group);
98 if (buffer_verified(bh))
99 goto verified;
100 blk = ext4_inode_bitmap(sb, desc);
101 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
102 EXT4_INODES_PER_GROUP(sb) / 8) ||
103 ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
104 ext4_unlock_group(sb, block_group);
105 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
106 "inode_bitmap = %llu", block_group, blk);
107 ext4_mark_group_bitmap_corrupted(sb, block_group,
108 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
109 return -EFSBADCRC;
110 }
111 set_buffer_verified(bh);
112verified:
113 ext4_unlock_group(sb, block_group);
114 return 0;
115}
116
117/*
118 * Read the inode allocation bitmap for a given block_group, reading
119 * into the specified slot in the superblock's bitmap cache.
120 *
121 * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122 */
123static struct buffer_head *
124ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125{
126 struct ext4_group_desc *desc;
127 struct ext4_sb_info *sbi = EXT4_SB(sb);
128 struct buffer_head *bh = NULL;
129 ext4_fsblk_t bitmap_blk;
130 int err;
131
132 desc = ext4_get_group_desc(sb, block_group, NULL);
133 if (!desc)
134 return ERR_PTR(-EFSCORRUPTED);
135
136 bitmap_blk = ext4_inode_bitmap(sb, desc);
137 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 "block_group %u", bitmap_blk, block_group);
141 ext4_mark_group_bitmap_corrupted(sb, block_group,
142 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 return ERR_PTR(-EFSCORRUPTED);
144 }
145 bh = sb_getblk(sb, bitmap_blk);
146 if (unlikely(!bh)) {
147 ext4_warning(sb, "Cannot read inode bitmap - "
148 "block_group = %u, inode_bitmap = %llu",
149 block_group, bitmap_blk);
150 return ERR_PTR(-ENOMEM);
151 }
152 if (bitmap_uptodate(bh))
153 goto verify;
154
155 lock_buffer(bh);
156 if (bitmap_uptodate(bh)) {
157 unlock_buffer(bh);
158 goto verify;
159 }
160
161 ext4_lock_group(sb, block_group);
162 if (ext4_has_group_desc_csum(sb) &&
163 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 if (block_group == 0) {
165 ext4_unlock_group(sb, block_group);
166 unlock_buffer(bh);
167 ext4_error(sb, "Inode bitmap for bg 0 marked "
168 "uninitialized");
169 err = -EFSCORRUPTED;
170 goto out;
171 }
172 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 sb->s_blocksize * 8, bh->b_data);
175 set_bitmap_uptodate(bh);
176 set_buffer_uptodate(bh);
177 set_buffer_verified(bh);
178 ext4_unlock_group(sb, block_group);
179 unlock_buffer(bh);
180 return bh;
181 }
182 ext4_unlock_group(sb, block_group);
183
184 if (buffer_uptodate(bh)) {
185 /*
186 * if not uninit if bh is uptodate,
187 * bitmap is also uptodate
188 */
189 set_bitmap_uptodate(bh);
190 unlock_buffer(bh);
191 goto verify;
192 }
193 /*
194 * submit the buffer_head for reading
195 */
196 trace_ext4_load_inode_bitmap(sb, block_group);
197 ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 if (!buffer_uptodate(bh)) {
200 put_bh(bh);
201 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 "block_group = %u, inode_bitmap = %llu",
203 block_group, bitmap_blk);
204 ext4_mark_group_bitmap_corrupted(sb, block_group,
205 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 return ERR_PTR(-EIO);
207 }
208
209verify:
210 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 if (err)
212 goto out;
213 return bh;
214out:
215 put_bh(bh);
216 return ERR_PTR(err);
217}
218
219/*
220 * NOTE! When we get the inode, we're the only people
221 * that have access to it, and as such there are no
222 * race conditions we have to worry about. The inode
223 * is not on the hash-lists, and it cannot be reached
224 * through the filesystem because the directory entry
225 * has been deleted earlier.
226 *
227 * HOWEVER: we must make sure that we get no aliases,
228 * which means that we have to call "clear_inode()"
229 * _before_ we mark the inode not in use in the inode
230 * bitmaps. Otherwise a newly created file might use
231 * the same inode number (not actually the same pointer
232 * though), and then we'd have two inodes sharing the
233 * same inode number and space on the harddisk.
234 */
235void ext4_free_inode(handle_t *handle, struct inode *inode)
236{
237 struct super_block *sb = inode->i_sb;
238 int is_directory;
239 unsigned long ino;
240 struct buffer_head *bitmap_bh = NULL;
241 struct buffer_head *bh2;
242 ext4_group_t block_group;
243 unsigned long bit;
244 struct ext4_group_desc *gdp;
245 struct ext4_super_block *es;
246 struct ext4_sb_info *sbi;
247 int fatal = 0, err, count, cleared;
248 struct ext4_group_info *grp;
249
250 if (!sb) {
251 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 "nonexistent device\n", __func__, __LINE__);
253 return;
254 }
255 if (atomic_read(&inode->i_count) > 1) {
256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 __func__, __LINE__, inode->i_ino,
258 atomic_read(&inode->i_count));
259 return;
260 }
261 if (inode->i_nlink) {
262 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 return;
265 }
266 sbi = EXT4_SB(sb);
267
268 ino = inode->i_ino;
269 ext4_debug("freeing inode %lu\n", ino);
270 trace_ext4_free_inode(inode);
271
272 dquot_initialize(inode);
273 dquot_free_inode(inode);
274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
278 ext4_clear_inode(inode);
279
280 es = sbi->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 goto error_return;
284 }
285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 /* Don't bother if the inode bitmap is corrupt. */
289 if (IS_ERR(bitmap_bh)) {
290 fatal = PTR_ERR(bitmap_bh);
291 bitmap_bh = NULL;
292 goto error_return;
293 }
294 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 grp = ext4_get_group_info(sb, block_group);
296 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 fatal = -EFSCORRUPTED;
298 goto error_return;
299 }
300 }
301
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
303 fatal = ext4_journal_get_write_access(handle, sb, bitmap_bh,
304 EXT4_JTR_NONE);
305 if (fatal)
306 goto error_return;
307
308 fatal = -ESRCH;
309 gdp = ext4_get_group_desc(sb, block_group, &bh2);
310 if (gdp) {
311 BUFFER_TRACE(bh2, "get_write_access");
312 fatal = ext4_journal_get_write_access(handle, sb, bh2,
313 EXT4_JTR_NONE);
314 }
315 ext4_lock_group(sb, block_group);
316 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
317 if (fatal || !cleared) {
318 ext4_unlock_group(sb, block_group);
319 goto out;
320 }
321
322 count = ext4_free_inodes_count(sb, gdp) + 1;
323 ext4_free_inodes_set(sb, gdp, count);
324 if (is_directory) {
325 count = ext4_used_dirs_count(sb, gdp) - 1;
326 ext4_used_dirs_set(sb, gdp, count);
327 if (percpu_counter_initialized(&sbi->s_dirs_counter))
328 percpu_counter_dec(&sbi->s_dirs_counter);
329 }
330 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
331 EXT4_INODES_PER_GROUP(sb) / 8);
332 ext4_group_desc_csum_set(sb, block_group, gdp);
333 ext4_unlock_group(sb, block_group);
334
335 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
336 percpu_counter_inc(&sbi->s_freeinodes_counter);
337 if (sbi->s_log_groups_per_flex) {
338 struct flex_groups *fg;
339
340 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
341 ext4_flex_group(sbi, block_group));
342 atomic_inc(&fg->free_inodes);
343 if (is_directory)
344 atomic_dec(&fg->used_dirs);
345 }
346 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
347 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
348out:
349 if (cleared) {
350 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
351 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
352 if (!fatal)
353 fatal = err;
354 } else {
355 ext4_error(sb, "bit already cleared for inode %lu", ino);
356 ext4_mark_group_bitmap_corrupted(sb, block_group,
357 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
358 }
359
360error_return:
361 brelse(bitmap_bh);
362 ext4_std_error(sb, fatal);
363}
364
365struct orlov_stats {
366 __u64 free_clusters;
367 __u32 free_inodes;
368 __u32 used_dirs;
369};
370
371/*
372 * Helper function for Orlov's allocator; returns critical information
373 * for a particular block group or flex_bg. If flex_size is 1, then g
374 * is a block group number; otherwise it is flex_bg number.
375 */
376static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
377 int flex_size, struct orlov_stats *stats)
378{
379 struct ext4_group_desc *desc;
380
381 if (flex_size > 1) {
382 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
383 s_flex_groups, g);
384 stats->free_inodes = atomic_read(&fg->free_inodes);
385 stats->free_clusters = atomic64_read(&fg->free_clusters);
386 stats->used_dirs = atomic_read(&fg->used_dirs);
387 return;
388 }
389
390 desc = ext4_get_group_desc(sb, g, NULL);
391 if (desc) {
392 stats->free_inodes = ext4_free_inodes_count(sb, desc);
393 stats->free_clusters = ext4_free_group_clusters(sb, desc);
394 stats->used_dirs = ext4_used_dirs_count(sb, desc);
395 } else {
396 stats->free_inodes = 0;
397 stats->free_clusters = 0;
398 stats->used_dirs = 0;
399 }
400}
401
402/*
403 * Orlov's allocator for directories.
404 *
405 * We always try to spread first-level directories.
406 *
407 * If there are blockgroups with both free inodes and free clusters counts
408 * not worse than average we return one with smallest directory count.
409 * Otherwise we simply return a random group.
410 *
411 * For the rest rules look so:
412 *
413 * It's OK to put directory into a group unless
414 * it has too many directories already (max_dirs) or
415 * it has too few free inodes left (min_inodes) or
416 * it has too few free clusters left (min_clusters) or
417 * Parent's group is preferred, if it doesn't satisfy these
418 * conditions we search cyclically through the rest. If none
419 * of the groups look good we just look for a group with more
420 * free inodes than average (starting at parent's group).
421 */
422
423static int find_group_orlov(struct super_block *sb, struct inode *parent,
424 ext4_group_t *group, umode_t mode,
425 const struct qstr *qstr)
426{
427 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
428 struct ext4_sb_info *sbi = EXT4_SB(sb);
429 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
430 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
431 unsigned int freei, avefreei, grp_free;
432 ext4_fsblk_t freec, avefreec;
433 unsigned int ndirs;
434 int max_dirs, min_inodes;
435 ext4_grpblk_t min_clusters;
436 ext4_group_t i, grp, g, ngroups;
437 struct ext4_group_desc *desc;
438 struct orlov_stats stats;
439 int flex_size = ext4_flex_bg_size(sbi);
440 struct dx_hash_info hinfo;
441
442 ngroups = real_ngroups;
443 if (flex_size > 1) {
444 ngroups = (real_ngroups + flex_size - 1) >>
445 sbi->s_log_groups_per_flex;
446 parent_group >>= sbi->s_log_groups_per_flex;
447 }
448
449 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
450 avefreei = freei / ngroups;
451 freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
452 avefreec = freec;
453 do_div(avefreec, ngroups);
454 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
455
456 if (S_ISDIR(mode) &&
457 ((parent == d_inode(sb->s_root)) ||
458 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
459 int best_ndir = inodes_per_group;
460 int ret = -1;
461
462 if (qstr) {
463 hinfo.hash_version = DX_HASH_HALF_MD4;
464 hinfo.seed = sbi->s_hash_seed;
465 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
466 parent_group = hinfo.hash % ngroups;
467 } else
468 parent_group = get_random_u32_below(ngroups);
469 for (i = 0; i < ngroups; i++) {
470 g = (parent_group + i) % ngroups;
471 get_orlov_stats(sb, g, flex_size, &stats);
472 if (!stats.free_inodes)
473 continue;
474 if (stats.used_dirs >= best_ndir)
475 continue;
476 if (stats.free_inodes < avefreei)
477 continue;
478 if (stats.free_clusters < avefreec)
479 continue;
480 grp = g;
481 ret = 0;
482 best_ndir = stats.used_dirs;
483 }
484 if (ret)
485 goto fallback;
486 found_flex_bg:
487 if (flex_size == 1) {
488 *group = grp;
489 return 0;
490 }
491
492 /*
493 * We pack inodes at the beginning of the flexgroup's
494 * inode tables. Block allocation decisions will do
495 * something similar, although regular files will
496 * start at 2nd block group of the flexgroup. See
497 * ext4_ext_find_goal() and ext4_find_near().
498 */
499 grp *= flex_size;
500 for (i = 0; i < flex_size; i++) {
501 if (grp+i >= real_ngroups)
502 break;
503 desc = ext4_get_group_desc(sb, grp+i, NULL);
504 if (desc && ext4_free_inodes_count(sb, desc)) {
505 *group = grp+i;
506 return 0;
507 }
508 }
509 goto fallback;
510 }
511
512 max_dirs = ndirs / ngroups + inodes_per_group*flex_size / 16;
513 min_inodes = avefreei - inodes_per_group*flex_size / 4;
514 if (min_inodes < 1)
515 min_inodes = 1;
516 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
517
518 /*
519 * Start looking in the flex group where we last allocated an
520 * inode for this parent directory
521 */
522 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
523 parent_group = EXT4_I(parent)->i_last_alloc_group;
524 if (flex_size > 1)
525 parent_group >>= sbi->s_log_groups_per_flex;
526 }
527
528 for (i = 0; i < ngroups; i++) {
529 grp = (parent_group + i) % ngroups;
530 get_orlov_stats(sb, grp, flex_size, &stats);
531 if (stats.used_dirs >= max_dirs)
532 continue;
533 if (stats.free_inodes < min_inodes)
534 continue;
535 if (stats.free_clusters < min_clusters)
536 continue;
537 goto found_flex_bg;
538 }
539
540fallback:
541 ngroups = real_ngroups;
542 avefreei = freei / ngroups;
543fallback_retry:
544 parent_group = EXT4_I(parent)->i_block_group;
545 for (i = 0; i < ngroups; i++) {
546 grp = (parent_group + i) % ngroups;
547 desc = ext4_get_group_desc(sb, grp, NULL);
548 if (desc) {
549 grp_free = ext4_free_inodes_count(sb, desc);
550 if (grp_free && grp_free >= avefreei) {
551 *group = grp;
552 return 0;
553 }
554 }
555 }
556
557 if (avefreei) {
558 /*
559 * The free-inodes counter is approximate, and for really small
560 * filesystems the above test can fail to find any blockgroups
561 */
562 avefreei = 0;
563 goto fallback_retry;
564 }
565
566 return -1;
567}
568
569static int find_group_other(struct super_block *sb, struct inode *parent,
570 ext4_group_t *group, umode_t mode)
571{
572 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
573 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
574 struct ext4_group_desc *desc;
575 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
576
577 /*
578 * Try to place the inode is the same flex group as its
579 * parent. If we can't find space, use the Orlov algorithm to
580 * find another flex group, and store that information in the
581 * parent directory's inode information so that use that flex
582 * group for future allocations.
583 */
584 if (flex_size > 1) {
585 int retry = 0;
586
587 try_again:
588 parent_group &= ~(flex_size-1);
589 last = parent_group + flex_size;
590 if (last > ngroups)
591 last = ngroups;
592 for (i = parent_group; i < last; i++) {
593 desc = ext4_get_group_desc(sb, i, NULL);
594 if (desc && ext4_free_inodes_count(sb, desc)) {
595 *group = i;
596 return 0;
597 }
598 }
599 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
600 retry = 1;
601 parent_group = EXT4_I(parent)->i_last_alloc_group;
602 goto try_again;
603 }
604 /*
605 * If this didn't work, use the Orlov search algorithm
606 * to find a new flex group; we pass in the mode to
607 * avoid the topdir algorithms.
608 */
609 *group = parent_group + flex_size;
610 if (*group > ngroups)
611 *group = 0;
612 return find_group_orlov(sb, parent, group, mode, NULL);
613 }
614
615 /*
616 * Try to place the inode in its parent directory
617 */
618 *group = parent_group;
619 desc = ext4_get_group_desc(sb, *group, NULL);
620 if (desc && ext4_free_inodes_count(sb, desc) &&
621 ext4_free_group_clusters(sb, desc))
622 return 0;
623
624 /*
625 * We're going to place this inode in a different blockgroup from its
626 * parent. We want to cause files in a common directory to all land in
627 * the same blockgroup. But we want files which are in a different
628 * directory which shares a blockgroup with our parent to land in a
629 * different blockgroup.
630 *
631 * So add our directory's i_ino into the starting point for the hash.
632 */
633 *group = (*group + parent->i_ino) % ngroups;
634
635 /*
636 * Use a quadratic hash to find a group with a free inode and some free
637 * blocks.
638 */
639 for (i = 1; i < ngroups; i <<= 1) {
640 *group += i;
641 if (*group >= ngroups)
642 *group -= ngroups;
643 desc = ext4_get_group_desc(sb, *group, NULL);
644 if (desc && ext4_free_inodes_count(sb, desc) &&
645 ext4_free_group_clusters(sb, desc))
646 return 0;
647 }
648
649 /*
650 * That failed: try linear search for a free inode, even if that group
651 * has no free blocks.
652 */
653 *group = parent_group;
654 for (i = 0; i < ngroups; i++) {
655 if (++*group >= ngroups)
656 *group = 0;
657 desc = ext4_get_group_desc(sb, *group, NULL);
658 if (desc && ext4_free_inodes_count(sb, desc))
659 return 0;
660 }
661
662 return -1;
663}
664
665/*
666 * In no journal mode, if an inode has recently been deleted, we want
667 * to avoid reusing it until we're reasonably sure the inode table
668 * block has been written back to disk. (Yes, these values are
669 * somewhat arbitrary...)
670 */
671#define RECENTCY_MIN 60
672#define RECENTCY_DIRTY 300
673
674static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
675{
676 struct ext4_group_desc *gdp;
677 struct ext4_inode *raw_inode;
678 struct buffer_head *bh;
679 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
680 int offset, ret = 0;
681 int recentcy = RECENTCY_MIN;
682 u32 dtime, now;
683
684 gdp = ext4_get_group_desc(sb, group, NULL);
685 if (unlikely(!gdp))
686 return 0;
687
688 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
689 (ino / inodes_per_block));
690 if (!bh || !buffer_uptodate(bh))
691 /*
692 * If the block is not in the buffer cache, then it
693 * must have been written out.
694 */
695 goto out;
696
697 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
698 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
699
700 /* i_dtime is only 32 bits on disk, but we only care about relative
701 * times in the range of a few minutes (i.e. long enough to sync a
702 * recently-deleted inode to disk), so using the low 32 bits of the
703 * clock (a 68 year range) is enough, see time_before32() */
704 dtime = le32_to_cpu(raw_inode->i_dtime);
705 now = ktime_get_real_seconds();
706 if (buffer_dirty(bh))
707 recentcy += RECENTCY_DIRTY;
708
709 if (dtime && time_before32(dtime, now) &&
710 time_before32(now, dtime + recentcy))
711 ret = 1;
712out:
713 brelse(bh);
714 return ret;
715}
716
717static int find_inode_bit(struct super_block *sb, ext4_group_t group,
718 struct buffer_head *bitmap, unsigned long *ino)
719{
720 bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
721 unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
722
723next:
724 *ino = ext4_find_next_zero_bit((unsigned long *)
725 bitmap->b_data,
726 EXT4_INODES_PER_GROUP(sb), *ino);
727 if (*ino >= EXT4_INODES_PER_GROUP(sb))
728 goto not_found;
729
730 if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
731 recently_deleted_ino = *ino;
732 *ino = *ino + 1;
733 if (*ino < EXT4_INODES_PER_GROUP(sb))
734 goto next;
735 goto not_found;
736 }
737 return 1;
738not_found:
739 if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
740 return 0;
741 /*
742 * Not reusing recently deleted inodes is mostly a preference. We don't
743 * want to report ENOSPC or skew allocation patterns because of that.
744 * So return even recently deleted inode if we could find better in the
745 * given range.
746 */
747 *ino = recently_deleted_ino;
748 return 1;
749}
750
751int ext4_mark_inode_used(struct super_block *sb, int ino)
752{
753 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
754 struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
755 struct ext4_group_desc *gdp;
756 ext4_group_t group;
757 int bit;
758 int err = -EFSCORRUPTED;
759
760 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
761 goto out;
762
763 group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
764 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
765 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
766 if (IS_ERR(inode_bitmap_bh))
767 return PTR_ERR(inode_bitmap_bh);
768
769 if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
770 err = 0;
771 goto out;
772 }
773
774 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
775 if (!gdp || !group_desc_bh) {
776 err = -EINVAL;
777 goto out;
778 }
779
780 ext4_set_bit(bit, inode_bitmap_bh->b_data);
781
782 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
783 err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
784 if (err) {
785 ext4_std_error(sb, err);
786 goto out;
787 }
788 err = sync_dirty_buffer(inode_bitmap_bh);
789 if (err) {
790 ext4_std_error(sb, err);
791 goto out;
792 }
793
794 /* We may have to initialize the block bitmap if it isn't already */
795 if (ext4_has_group_desc_csum(sb) &&
796 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
797 struct buffer_head *block_bitmap_bh;
798
799 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
800 if (IS_ERR(block_bitmap_bh)) {
801 err = PTR_ERR(block_bitmap_bh);
802 goto out;
803 }
804
805 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
806 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
807 sync_dirty_buffer(block_bitmap_bh);
808
809 /* recheck and clear flag under lock if we still need to */
810 ext4_lock_group(sb, group);
811 if (ext4_has_group_desc_csum(sb) &&
812 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
813 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
814 ext4_free_group_clusters_set(sb, gdp,
815 ext4_free_clusters_after_init(sb, group, gdp));
816 ext4_block_bitmap_csum_set(sb, group, gdp,
817 block_bitmap_bh);
818 ext4_group_desc_csum_set(sb, group, gdp);
819 }
820 ext4_unlock_group(sb, group);
821 brelse(block_bitmap_bh);
822
823 if (err) {
824 ext4_std_error(sb, err);
825 goto out;
826 }
827 }
828
829 /* Update the relevant bg descriptor fields */
830 if (ext4_has_group_desc_csum(sb)) {
831 int free;
832
833 ext4_lock_group(sb, group); /* while we modify the bg desc */
834 free = EXT4_INODES_PER_GROUP(sb) -
835 ext4_itable_unused_count(sb, gdp);
836 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
837 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
838 free = 0;
839 }
840
841 /*
842 * Check the relative inode number against the last used
843 * relative inode number in this group. if it is greater
844 * we need to update the bg_itable_unused count
845 */
846 if (bit >= free)
847 ext4_itable_unused_set(sb, gdp,
848 (EXT4_INODES_PER_GROUP(sb) - bit - 1));
849 } else {
850 ext4_lock_group(sb, group);
851 }
852
853 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
854 if (ext4_has_group_desc_csum(sb)) {
855 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
856 EXT4_INODES_PER_GROUP(sb) / 8);
857 ext4_group_desc_csum_set(sb, group, gdp);
858 }
859
860 ext4_unlock_group(sb, group);
861 err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
862 sync_dirty_buffer(group_desc_bh);
863out:
864 return err;
865}
866
867static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
868 bool encrypt)
869{
870 struct super_block *sb = dir->i_sb;
871 int nblocks = 0;
872#ifdef CONFIG_EXT4_FS_POSIX_ACL
873 struct posix_acl *p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
874
875 if (IS_ERR(p))
876 return PTR_ERR(p);
877 if (p) {
878 int acl_size = p->a_count * sizeof(ext4_acl_entry);
879
880 nblocks += (S_ISDIR(mode) ? 2 : 1) *
881 __ext4_xattr_set_credits(sb, NULL /* inode */,
882 NULL /* block_bh */, acl_size,
883 true /* is_create */);
884 posix_acl_release(p);
885 }
886#endif
887
888#ifdef CONFIG_SECURITY
889 {
890 int num_security_xattrs = 1;
891
892#ifdef CONFIG_INTEGRITY
893 num_security_xattrs++;
894#endif
895 /*
896 * We assume that security xattrs are never more than 1k.
897 * In practice they are under 128 bytes.
898 */
899 nblocks += num_security_xattrs *
900 __ext4_xattr_set_credits(sb, NULL /* inode */,
901 NULL /* block_bh */, 1024,
902 true /* is_create */);
903 }
904#endif
905 if (encrypt)
906 nblocks += __ext4_xattr_set_credits(sb,
907 NULL /* inode */,
908 NULL /* block_bh */,
909 FSCRYPT_SET_CONTEXT_MAX_SIZE,
910 true /* is_create */);
911 return nblocks;
912}
913
914/*
915 * There are two policies for allocating an inode. If the new inode is
916 * a directory, then a forward search is made for a block group with both
917 * free space and a low directory-to-inode ratio; if that fails, then of
918 * the groups with above-average free space, that group with the fewest
919 * directories already is chosen.
920 *
921 * For other inodes, search forward from the parent directory's block
922 * group to find a free inode.
923 */
924struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
925 handle_t *handle, struct inode *dir,
926 umode_t mode, const struct qstr *qstr,
927 __u32 goal, uid_t *owner, __u32 i_flags,
928 int handle_type, unsigned int line_no,
929 int nblocks)
930{
931 struct super_block *sb;
932 struct buffer_head *inode_bitmap_bh = NULL;
933 struct buffer_head *group_desc_bh;
934 ext4_group_t ngroups, group = 0;
935 unsigned long ino = 0;
936 struct inode *inode;
937 struct ext4_group_desc *gdp = NULL;
938 struct ext4_inode_info *ei;
939 struct ext4_sb_info *sbi;
940 int ret2, err;
941 struct inode *ret;
942 ext4_group_t i;
943 ext4_group_t flex_group;
944 struct ext4_group_info *grp = NULL;
945 bool encrypt = false;
946
947 /* Cannot create files in a deleted directory */
948 if (!dir || !dir->i_nlink)
949 return ERR_PTR(-EPERM);
950
951 sb = dir->i_sb;
952 sbi = EXT4_SB(sb);
953
954 if (unlikely(ext4_forced_shutdown(sbi)))
955 return ERR_PTR(-EIO);
956
957 ngroups = ext4_get_groups_count(sb);
958 trace_ext4_request_inode(dir, mode);
959 inode = new_inode(sb);
960 if (!inode)
961 return ERR_PTR(-ENOMEM);
962 ei = EXT4_I(inode);
963
964 /*
965 * Initialize owners and quota early so that we don't have to account
966 * for quota initialization worst case in standard inode creating
967 * transaction
968 */
969 if (owner) {
970 inode->i_mode = mode;
971 i_uid_write(inode, owner[0]);
972 i_gid_write(inode, owner[1]);
973 } else if (test_opt(sb, GRPID)) {
974 inode->i_mode = mode;
975 inode_fsuid_set(inode, mnt_userns);
976 inode->i_gid = dir->i_gid;
977 } else
978 inode_init_owner(mnt_userns, inode, dir, mode);
979
980 if (ext4_has_feature_project(sb) &&
981 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
982 ei->i_projid = EXT4_I(dir)->i_projid;
983 else
984 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
985
986 if (!(i_flags & EXT4_EA_INODE_FL)) {
987 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
988 if (err)
989 goto out;
990 }
991
992 err = dquot_initialize(inode);
993 if (err)
994 goto out;
995
996 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
997 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
998 if (ret2 < 0) {
999 err = ret2;
1000 goto out;
1001 }
1002 nblocks += ret2;
1003 }
1004
1005 if (!goal)
1006 goal = sbi->s_inode_goal;
1007
1008 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1009 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1010 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1011 ret2 = 0;
1012 goto got_group;
1013 }
1014
1015 if (S_ISDIR(mode))
1016 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1017 else
1018 ret2 = find_group_other(sb, dir, &group, mode);
1019
1020got_group:
1021 EXT4_I(dir)->i_last_alloc_group = group;
1022 err = -ENOSPC;
1023 if (ret2 == -1)
1024 goto out;
1025
1026 /*
1027 * Normally we will only go through one pass of this loop,
1028 * unless we get unlucky and it turns out the group we selected
1029 * had its last inode grabbed by someone else.
1030 */
1031 for (i = 0; i < ngroups; i++, ino = 0) {
1032 err = -EIO;
1033
1034 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1035 if (!gdp)
1036 goto out;
1037
1038 /*
1039 * Check free inodes count before loading bitmap.
1040 */
1041 if (ext4_free_inodes_count(sb, gdp) == 0)
1042 goto next_group;
1043
1044 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1045 grp = ext4_get_group_info(sb, group);
1046 /*
1047 * Skip groups with already-known suspicious inode
1048 * tables
1049 */
1050 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1051 goto next_group;
1052 }
1053
1054 brelse(inode_bitmap_bh);
1055 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1056 /* Skip groups with suspicious inode tables */
1057 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1058 && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1059 IS_ERR(inode_bitmap_bh)) {
1060 inode_bitmap_bh = NULL;
1061 goto next_group;
1062 }
1063
1064repeat_in_this_group:
1065 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1066 if (!ret2)
1067 goto next_group;
1068
1069 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1070 ext4_error(sb, "reserved inode found cleared - "
1071 "inode=%lu", ino + 1);
1072 ext4_mark_group_bitmap_corrupted(sb, group,
1073 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1074 goto next_group;
1075 }
1076
1077 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1078 BUG_ON(nblocks <= 0);
1079 handle = __ext4_journal_start_sb(NULL, dir->i_sb,
1080 line_no, handle_type, nblocks, 0,
1081 ext4_trans_default_revoke_credits(sb));
1082 if (IS_ERR(handle)) {
1083 err = PTR_ERR(handle);
1084 ext4_std_error(sb, err);
1085 goto out;
1086 }
1087 }
1088 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1089 err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,
1090 EXT4_JTR_NONE);
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;
1116next_group:
1117 if (++group == ngroups)
1118 group = 0;
1119 }
1120 err = -ENOSPC;
1121 goto out;
1122
1123got:
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, sb, group_desc_bh,
1133 EXT4_JTR_NONE);
1134 if (err) {
1135 ext4_std_error(sb, err);
1136 goto out;
1137 }
1138
1139 /* We may have to initialize the block bitmap if it isn't already */
1140 if (ext4_has_group_desc_csum(sb) &&
1141 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1142 struct buffer_head *block_bitmap_bh;
1143
1144 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1145 if (IS_ERR(block_bitmap_bh)) {
1146 err = PTR_ERR(block_bitmap_bh);
1147 goto out;
1148 }
1149 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1150 err = ext4_journal_get_write_access(handle, sb, block_bitmap_bh,
1151 EXT4_JTR_NONE);
1152 if (err) {
1153 brelse(block_bitmap_bh);
1154 ext4_std_error(sb, err);
1155 goto out;
1156 }
1157
1158 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1159 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1160
1161 /* recheck and clear flag under lock if we still need to */
1162 ext4_lock_group(sb, group);
1163 if (ext4_has_group_desc_csum(sb) &&
1164 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1165 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1166 ext4_free_group_clusters_set(sb, gdp,
1167 ext4_free_clusters_after_init(sb, group, gdp));
1168 ext4_block_bitmap_csum_set(sb, group, gdp,
1169 block_bitmap_bh);
1170 ext4_group_desc_csum_set(sb, group, gdp);
1171 }
1172 ext4_unlock_group(sb, group);
1173 brelse(block_bitmap_bh);
1174
1175 if (err) {
1176 ext4_std_error(sb, err);
1177 goto out;
1178 }
1179 }
1180
1181 /* Update the relevant bg descriptor fields */
1182 if (ext4_has_group_desc_csum(sb)) {
1183 int free;
1184 struct ext4_group_info *grp = NULL;
1185
1186 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1187 grp = ext4_get_group_info(sb, group);
1188 down_read(&grp->alloc_sem); /*
1189 * protect vs itable
1190 * lazyinit
1191 */
1192 }
1193 ext4_lock_group(sb, group); /* while we modify the bg desc */
1194 free = EXT4_INODES_PER_GROUP(sb) -
1195 ext4_itable_unused_count(sb, gdp);
1196 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1197 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1198 free = 0;
1199 }
1200 /*
1201 * Check the relative inode number against the last used
1202 * relative inode number in this group. if it is greater
1203 * we need to update the bg_itable_unused count
1204 */
1205 if (ino > free)
1206 ext4_itable_unused_set(sb, gdp,
1207 (EXT4_INODES_PER_GROUP(sb) - ino));
1208 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1209 up_read(&grp->alloc_sem);
1210 } else {
1211 ext4_lock_group(sb, group);
1212 }
1213
1214 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1215 if (S_ISDIR(mode)) {
1216 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1217 if (sbi->s_log_groups_per_flex) {
1218 ext4_group_t f = ext4_flex_group(sbi, group);
1219
1220 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1221 f)->used_dirs);
1222 }
1223 }
1224 if (ext4_has_group_desc_csum(sb)) {
1225 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1226 EXT4_INODES_PER_GROUP(sb) / 8);
1227 ext4_group_desc_csum_set(sb, group, gdp);
1228 }
1229 ext4_unlock_group(sb, group);
1230
1231 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1232 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1233 if (err) {
1234 ext4_std_error(sb, err);
1235 goto out;
1236 }
1237
1238 percpu_counter_dec(&sbi->s_freeinodes_counter);
1239 if (S_ISDIR(mode))
1240 percpu_counter_inc(&sbi->s_dirs_counter);
1241
1242 if (sbi->s_log_groups_per_flex) {
1243 flex_group = ext4_flex_group(sbi, group);
1244 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1245 flex_group)->free_inodes);
1246 }
1247
1248 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1249 /* This is the optimal IO size (for stat), not the fs block size */
1250 inode->i_blocks = 0;
1251 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1252 ei->i_crtime = inode->i_mtime;
1253
1254 memset(ei->i_data, 0, sizeof(ei->i_data));
1255 ei->i_dir_start_lookup = 0;
1256 ei->i_disksize = 0;
1257
1258 /* Don't inherit extent flag from directory, amongst others. */
1259 ei->i_flags =
1260 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1261 ei->i_flags |= i_flags;
1262 ei->i_file_acl = 0;
1263 ei->i_dtime = 0;
1264 ei->i_block_group = group;
1265 ei->i_last_alloc_group = ~0;
1266
1267 ext4_set_inode_flags(inode, true);
1268 if (IS_DIRSYNC(inode))
1269 ext4_handle_sync(handle);
1270 if (insert_inode_locked(inode) < 0) {
1271 /*
1272 * Likely a bitmap corruption causing inode to be allocated
1273 * twice.
1274 */
1275 err = -EIO;
1276 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1277 inode->i_ino);
1278 ext4_mark_group_bitmap_corrupted(sb, group,
1279 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1280 goto out;
1281 }
1282 inode->i_generation = get_random_u32();
1283
1284 /* Precompute checksum seed for inode metadata */
1285 if (ext4_has_metadata_csum(sb)) {
1286 __u32 csum;
1287 __le32 inum = cpu_to_le32(inode->i_ino);
1288 __le32 gen = cpu_to_le32(inode->i_generation);
1289 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1290 sizeof(inum));
1291 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1292 sizeof(gen));
1293 }
1294
1295 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1296 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1297
1298 ei->i_extra_isize = sbi->s_want_extra_isize;
1299 ei->i_inline_off = 0;
1300 if (ext4_has_feature_inline_data(sb) &&
1301 (!(ei->i_flags & EXT4_DAX_FL) || S_ISDIR(mode)))
1302 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1303 ret = inode;
1304 err = dquot_alloc_inode(inode);
1305 if (err)
1306 goto fail_drop;
1307
1308 /*
1309 * Since the encryption xattr will always be unique, create it first so
1310 * that it's less likely to end up in an external xattr block and
1311 * prevent its deduplication.
1312 */
1313 if (encrypt) {
1314 err = fscrypt_set_context(inode, handle);
1315 if (err)
1316 goto fail_free_drop;
1317 }
1318
1319 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1320 err = ext4_init_acl(handle, inode, dir);
1321 if (err)
1322 goto fail_free_drop;
1323
1324 err = ext4_init_security(handle, inode, dir, qstr);
1325 if (err)
1326 goto fail_free_drop;
1327 }
1328
1329 if (ext4_has_feature_extents(sb)) {
1330 /* set extent flag only for directory, file and normal symlink*/
1331 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1332 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1333 ext4_ext_tree_init(handle, inode);
1334 }
1335 }
1336
1337 if (ext4_handle_valid(handle)) {
1338 ei->i_sync_tid = handle->h_transaction->t_tid;
1339 ei->i_datasync_tid = handle->h_transaction->t_tid;
1340 }
1341
1342 err = ext4_mark_inode_dirty(handle, inode);
1343 if (err) {
1344 ext4_std_error(sb, err);
1345 goto fail_free_drop;
1346 }
1347
1348 ext4_debug("allocating inode %lu\n", inode->i_ino);
1349 trace_ext4_allocate_inode(inode, dir, mode);
1350 brelse(inode_bitmap_bh);
1351 return ret;
1352
1353fail_free_drop:
1354 dquot_free_inode(inode);
1355fail_drop:
1356 clear_nlink(inode);
1357 unlock_new_inode(inode);
1358out:
1359 dquot_drop(inode);
1360 inode->i_flags |= S_NOQUOTA;
1361 iput(inode);
1362 brelse(inode_bitmap_bh);
1363 return ERR_PTR(err);
1364}
1365
1366/* Verify that we are loading a valid orphan from disk */
1367struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1368{
1369 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1370 ext4_group_t block_group;
1371 int bit;
1372 struct buffer_head *bitmap_bh = NULL;
1373 struct inode *inode = NULL;
1374 int err = -EFSCORRUPTED;
1375
1376 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1377 goto bad_orphan;
1378
1379 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1380 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1381 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1382 if (IS_ERR(bitmap_bh))
1383 return ERR_CAST(bitmap_bh);
1384
1385 /* Having the inode bit set should be a 100% indicator that this
1386 * is a valid orphan (no e2fsck run on fs). Orphans also include
1387 * inodes that were being truncated, so we can't check i_nlink==0.
1388 */
1389 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1390 goto bad_orphan;
1391
1392 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1393 if (IS_ERR(inode)) {
1394 err = PTR_ERR(inode);
1395 ext4_error_err(sb, -err,
1396 "couldn't read orphan inode %lu (err %d)",
1397 ino, err);
1398 brelse(bitmap_bh);
1399 return inode;
1400 }
1401
1402 /*
1403 * If the orphans has i_nlinks > 0 then it should be able to
1404 * be truncated, otherwise it won't be removed from the orphan
1405 * list during processing and an infinite loop will result.
1406 * Similarly, it must not be a bad inode.
1407 */
1408 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1409 is_bad_inode(inode))
1410 goto bad_orphan;
1411
1412 if (NEXT_ORPHAN(inode) > max_ino)
1413 goto bad_orphan;
1414 brelse(bitmap_bh);
1415 return inode;
1416
1417bad_orphan:
1418 ext4_error(sb, "bad orphan inode %lu", ino);
1419 if (bitmap_bh)
1420 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1421 bit, (unsigned long long)bitmap_bh->b_blocknr,
1422 ext4_test_bit(bit, bitmap_bh->b_data));
1423 if (inode) {
1424 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1425 is_bad_inode(inode));
1426 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1427 NEXT_ORPHAN(inode));
1428 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1429 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1430 /* Avoid freeing blocks if we got a bad deleted inode */
1431 if (inode->i_nlink == 0)
1432 inode->i_blocks = 0;
1433 iput(inode);
1434 }
1435 brelse(bitmap_bh);
1436 return ERR_PTR(err);
1437}
1438
1439unsigned long ext4_count_free_inodes(struct super_block *sb)
1440{
1441 unsigned long desc_count;
1442 struct ext4_group_desc *gdp;
1443 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1444#ifdef EXT4FS_DEBUG
1445 struct ext4_super_block *es;
1446 unsigned long bitmap_count, x;
1447 struct buffer_head *bitmap_bh = NULL;
1448
1449 es = EXT4_SB(sb)->s_es;
1450 desc_count = 0;
1451 bitmap_count = 0;
1452 gdp = NULL;
1453 for (i = 0; i < ngroups; i++) {
1454 gdp = ext4_get_group_desc(sb, i, NULL);
1455 if (!gdp)
1456 continue;
1457 desc_count += ext4_free_inodes_count(sb, gdp);
1458 brelse(bitmap_bh);
1459 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1460 if (IS_ERR(bitmap_bh)) {
1461 bitmap_bh = NULL;
1462 continue;
1463 }
1464
1465 x = ext4_count_free(bitmap_bh->b_data,
1466 EXT4_INODES_PER_GROUP(sb) / 8);
1467 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1468 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1469 bitmap_count += x;
1470 }
1471 brelse(bitmap_bh);
1472 printk(KERN_DEBUG "ext4_count_free_inodes: "
1473 "stored = %u, computed = %lu, %lu\n",
1474 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1475 return desc_count;
1476#else
1477 desc_count = 0;
1478 for (i = 0; i < ngroups; i++) {
1479 gdp = ext4_get_group_desc(sb, i, NULL);
1480 if (!gdp)
1481 continue;
1482 desc_count += ext4_free_inodes_count(sb, gdp);
1483 cond_resched();
1484 }
1485 return desc_count;
1486#endif
1487}
1488
1489/* Called at mount-time, super-block is locked */
1490unsigned long ext4_count_dirs(struct super_block * sb)
1491{
1492 unsigned long count = 0;
1493 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1494
1495 for (i = 0; i < ngroups; i++) {
1496 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1497 if (!gdp)
1498 continue;
1499 count += ext4_used_dirs_count(sb, gdp);
1500 }
1501 return count;
1502}
1503
1504/*
1505 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1506 * inode table. Must be called without any spinlock held. The only place
1507 * where it is called from on active part of filesystem is ext4lazyinit
1508 * thread, so we do not need any special locks, however we have to prevent
1509 * inode allocation from the current group, so we take alloc_sem lock, to
1510 * block ext4_new_inode() until we are finished.
1511 */
1512int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1513 int barrier)
1514{
1515 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1516 struct ext4_sb_info *sbi = EXT4_SB(sb);
1517 struct ext4_group_desc *gdp = NULL;
1518 struct buffer_head *group_desc_bh;
1519 handle_t *handle;
1520 ext4_fsblk_t blk;
1521 int num, ret = 0, used_blks = 0;
1522 unsigned long used_inos = 0;
1523
1524 /* This should not happen, but just to be sure check this */
1525 if (sb_rdonly(sb)) {
1526 ret = 1;
1527 goto out;
1528 }
1529
1530 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1531 if (!gdp)
1532 goto out;
1533
1534 /*
1535 * We do not need to lock this, because we are the only one
1536 * handling this flag.
1537 */
1538 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1539 goto out;
1540
1541 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1542 if (IS_ERR(handle)) {
1543 ret = PTR_ERR(handle);
1544 goto out;
1545 }
1546
1547 down_write(&grp->alloc_sem);
1548 /*
1549 * If inode bitmap was already initialized there may be some
1550 * used inodes so we need to skip blocks with used inodes in
1551 * inode table.
1552 */
1553 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1554 used_inos = EXT4_INODES_PER_GROUP(sb) -
1555 ext4_itable_unused_count(sb, gdp);
1556 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1557
1558 /* Bogus inode unused count? */
1559 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1560 ext4_error(sb, "Something is wrong with group %u: "
1561 "used itable blocks: %d; "
1562 "itable unused count: %u",
1563 group, used_blks,
1564 ext4_itable_unused_count(sb, gdp));
1565 ret = 1;
1566 goto err_out;
1567 }
1568
1569 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1570 /*
1571 * Are there some uninitialized inodes in the inode table
1572 * before the first normal inode?
1573 */
1574 if ((used_blks != sbi->s_itb_per_group) &&
1575 (used_inos < EXT4_FIRST_INO(sb))) {
1576 ext4_error(sb, "Something is wrong with group %u: "
1577 "itable unused count: %u; "
1578 "itables initialized count: %ld",
1579 group, ext4_itable_unused_count(sb, gdp),
1580 used_inos);
1581 ret = 1;
1582 goto err_out;
1583 }
1584 }
1585
1586 blk = ext4_inode_table(sb, gdp) + used_blks;
1587 num = sbi->s_itb_per_group - used_blks;
1588
1589 BUFFER_TRACE(group_desc_bh, "get_write_access");
1590 ret = ext4_journal_get_write_access(handle, sb, group_desc_bh,
1591 EXT4_JTR_NONE);
1592 if (ret)
1593 goto err_out;
1594
1595 /*
1596 * Skip zeroout if the inode table is full. But we set the ZEROED
1597 * flag anyway, because obviously, when it is full it does not need
1598 * further zeroing.
1599 */
1600 if (unlikely(num == 0))
1601 goto skip_zeroout;
1602
1603 ext4_debug("going to zero out inode table in group %d\n",
1604 group);
1605 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1606 if (ret < 0)
1607 goto err_out;
1608 if (barrier)
1609 blkdev_issue_flush(sb->s_bdev);
1610
1611skip_zeroout:
1612 ext4_lock_group(sb, group);
1613 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1614 ext4_group_desc_csum_set(sb, group, gdp);
1615 ext4_unlock_group(sb, group);
1616
1617 BUFFER_TRACE(group_desc_bh,
1618 "call ext4_handle_dirty_metadata");
1619 ret = ext4_handle_dirty_metadata(handle, NULL,
1620 group_desc_bh);
1621
1622err_out:
1623 up_write(&grp->alloc_sem);
1624 ext4_journal_stop(handle);
1625out:
1626 return ret;
1627}