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