mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / ext4 / ialloc.c
blob85c2a7ea5ea2d58f8479e0d67331c382fe354074
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/ialloc.c
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)
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
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>
27 #include <asm/byteorder.h>
29 #include "ext4.h"
30 #include "ext4_jbd2.h"
31 #include "xattr.h"
32 #include "acl.h"
34 #include <trace/events/ext4.h>
37 * ialloc.c contains the inodes allocation and deallocation routines
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.
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.
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.
55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
57 int i;
59 if (start_bit >= end_bit)
60 return;
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);
69 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
71 if (uptodate) {
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
75 unlock_buffer(bh);
76 put_bh(bh);
79 static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
84 ext4_fsblk_t blk;
85 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
86 struct ext4_sb_info *sbi = EXT4_SB(sb);
88 if (buffer_verified(bh))
89 return 0;
90 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
91 return -EFSCORRUPTED;
93 ext4_lock_group(sb, block_group);
94 if (buffer_verified(bh))
95 goto verified;
96 blk = ext4_inode_bitmap(sb, desc);
97 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
98 EXT4_INODES_PER_GROUP(sb) / 8)) {
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 grp = ext4_get_group_info(sb, block_group);
103 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
104 int count;
105 count = ext4_free_inodes_count(sb, desc);
106 percpu_counter_sub(&sbi->s_freeinodes_counter,
107 count);
109 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
110 return -EFSBADCRC;
112 set_buffer_verified(bh);
113 verified:
114 ext4_unlock_group(sb, block_group);
115 return 0;
119 * Read the inode allocation bitmap for a given block_group, reading
120 * into the specified slot in the superblock's bitmap cache.
122 * Return buffer_head of bitmap on success or NULL.
124 static struct buffer_head *
125 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
127 struct ext4_group_desc *desc;
128 struct ext4_sb_info *sbi = EXT4_SB(sb);
129 struct buffer_head *bh = NULL;
130 ext4_fsblk_t bitmap_blk;
131 int err;
133 desc = ext4_get_group_desc(sb, block_group, NULL);
134 if (!desc)
135 return ERR_PTR(-EFSCORRUPTED);
137 bitmap_blk = ext4_inode_bitmap(sb, desc);
138 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
139 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
140 ext4_error(sb, "Invalid inode bitmap blk %llu in "
141 "block_group %u", bitmap_blk, block_group);
142 return ERR_PTR(-EFSCORRUPTED);
144 bh = sb_getblk(sb, bitmap_blk);
145 if (unlikely(!bh)) {
146 ext4_error(sb, "Cannot read inode bitmap - "
147 "block_group = %u, inode_bitmap = %llu",
148 block_group, bitmap_blk);
149 return ERR_PTR(-EIO);
151 if (bitmap_uptodate(bh))
152 goto verify;
154 lock_buffer(bh);
155 if (bitmap_uptodate(bh)) {
156 unlock_buffer(bh);
157 goto verify;
160 ext4_lock_group(sb, block_group);
161 if (ext4_has_group_desc_csum(sb) &&
162 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
163 if (block_group == 0) {
164 ext4_unlock_group(sb, block_group);
165 unlock_buffer(bh);
166 ext4_error(sb, "Inode bitmap for bg 0 marked "
167 "uninitialized");
168 err = -EFSCORRUPTED;
169 goto out;
171 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
172 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
173 sb->s_blocksize * 8, bh->b_data);
174 set_bitmap_uptodate(bh);
175 set_buffer_uptodate(bh);
176 set_buffer_verified(bh);
177 ext4_unlock_group(sb, block_group);
178 unlock_buffer(bh);
179 return bh;
181 ext4_unlock_group(sb, block_group);
183 if (buffer_uptodate(bh)) {
185 * if not uninit if bh is uptodate,
186 * bitmap is also uptodate
188 set_bitmap_uptodate(bh);
189 unlock_buffer(bh);
190 goto verify;
193 * submit the buffer_head for reading
195 trace_ext4_load_inode_bitmap(sb, block_group);
196 bh->b_end_io = ext4_end_bitmap_read;
197 get_bh(bh);
198 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
199 wait_on_buffer(bh);
200 if (!buffer_uptodate(bh)) {
201 put_bh(bh);
202 ext4_error(sb, "Cannot read inode bitmap - "
203 "block_group = %u, inode_bitmap = %llu",
204 block_group, bitmap_blk);
205 return ERR_PTR(-EIO);
208 verify:
209 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
210 if (err)
211 goto out;
212 return bh;
213 out:
214 put_bh(bh);
215 return ERR_PTR(err);
219 * NOTE! When we get the inode, we're the only people
220 * that have access to it, and as such there are no
221 * race conditions we have to worry about. The inode
222 * is not on the hash-lists, and it cannot be reached
223 * through the filesystem because the directory entry
224 * has been deleted earlier.
226 * HOWEVER: we must make sure that we get no aliases,
227 * which means that we have to call "clear_inode()"
228 * _before_ we mark the inode not in use in the inode
229 * bitmaps. Otherwise a newly created file might use
230 * the same inode number (not actually the same pointer
231 * though), and then we'd have two inodes sharing the
232 * same inode number and space on the harddisk.
234 void ext4_free_inode(handle_t *handle, struct inode *inode)
236 struct super_block *sb = inode->i_sb;
237 int is_directory;
238 unsigned long ino;
239 struct buffer_head *bitmap_bh = NULL;
240 struct buffer_head *bh2;
241 ext4_group_t block_group;
242 unsigned long bit;
243 struct ext4_group_desc *gdp;
244 struct ext4_super_block *es;
245 struct ext4_sb_info *sbi;
246 int fatal = 0, err, count, cleared;
247 struct ext4_group_info *grp;
249 if (!sb) {
250 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
251 "nonexistent device\n", __func__, __LINE__);
252 return;
254 if (atomic_read(&inode->i_count) > 1) {
255 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
256 __func__, __LINE__, inode->i_ino,
257 atomic_read(&inode->i_count));
258 return;
260 if (inode->i_nlink) {
261 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
262 __func__, __LINE__, inode->i_ino, inode->i_nlink);
263 return;
265 sbi = EXT4_SB(sb);
267 ino = inode->i_ino;
268 ext4_debug("freeing inode %lu\n", ino);
269 trace_ext4_free_inode(inode);
272 * Note: we must free any quota before locking the superblock,
273 * as writing the quota to disk may need the lock as well.
275 dquot_initialize(inode);
276 dquot_free_inode(inode);
277 dquot_drop(inode);
279 is_directory = S_ISDIR(inode->i_mode);
281 /* Do this BEFORE marking the inode not in use or returning an error */
282 ext4_clear_inode(inode);
284 es = EXT4_SB(sb)->s_es;
285 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
286 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
287 goto error_return;
289 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
290 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
291 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
292 /* Don't bother if the inode bitmap is corrupt. */
293 grp = ext4_get_group_info(sb, block_group);
294 if (IS_ERR(bitmap_bh)) {
295 fatal = PTR_ERR(bitmap_bh);
296 bitmap_bh = NULL;
297 goto error_return;
299 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
300 fatal = -EFSCORRUPTED;
301 goto error_return;
304 BUFFER_TRACE(bitmap_bh, "get_write_access");
305 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
306 if (fatal)
307 goto error_return;
309 fatal = -ESRCH;
310 gdp = ext4_get_group_desc(sb, block_group, &bh2);
311 if (gdp) {
312 BUFFER_TRACE(bh2, "get_write_access");
313 fatal = ext4_journal_get_write_access(handle, bh2);
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;
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 percpu_counter_dec(&sbi->s_dirs_counter);
329 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
330 EXT4_INODES_PER_GROUP(sb) / 8);
331 ext4_group_desc_csum_set(sb, block_group, gdp);
332 ext4_unlock_group(sb, block_group);
334 percpu_counter_inc(&sbi->s_freeinodes_counter);
335 if (sbi->s_log_groups_per_flex) {
336 struct flex_groups *fg;
338 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
339 ext4_flex_group(sbi, block_group));
340 atomic_inc(&fg->free_inodes);
341 if (is_directory)
342 atomic_dec(&fg->used_dirs);
344 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
345 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
346 out:
347 if (cleared) {
348 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
349 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
350 if (!fatal)
351 fatal = err;
352 } else {
353 ext4_error(sb, "bit already cleared for inode %lu", ino);
354 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
355 int count;
356 count = ext4_free_inodes_count(sb, gdp);
357 percpu_counter_sub(&sbi->s_freeinodes_counter,
358 count);
360 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
363 error_return:
364 brelse(bitmap_bh);
365 ext4_std_error(sb, fatal);
368 struct orlov_stats {
369 __u64 free_clusters;
370 __u32 free_inodes;
371 __u32 used_dirs;
375 * Helper function for Orlov's allocator; returns critical information
376 * for a particular block group or flex_bg. If flex_size is 1, then g
377 * is a block group number; otherwise it is flex_bg number.
379 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
380 int flex_size, struct orlov_stats *stats)
382 struct ext4_group_desc *desc;
384 if (flex_size > 1) {
385 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
386 s_flex_groups, g);
387 stats->free_inodes = atomic_read(&fg->free_inodes);
388 stats->free_clusters = atomic64_read(&fg->free_clusters);
389 stats->used_dirs = atomic_read(&fg->used_dirs);
390 return;
393 desc = ext4_get_group_desc(sb, g, NULL);
394 if (desc) {
395 stats->free_inodes = ext4_free_inodes_count(sb, desc);
396 stats->free_clusters = ext4_free_group_clusters(sb, desc);
397 stats->used_dirs = ext4_used_dirs_count(sb, desc);
398 } else {
399 stats->free_inodes = 0;
400 stats->free_clusters = 0;
401 stats->used_dirs = 0;
406 * Orlov's allocator for directories.
408 * We always try to spread first-level directories.
410 * If there are blockgroups with both free inodes and free blocks counts
411 * not worse than average we return one with smallest directory count.
412 * Otherwise we simply return a random group.
414 * For the rest rules look so:
416 * It's OK to put directory into a group unless
417 * it has too many directories already (max_dirs) or
418 * it has too few free inodes left (min_inodes) or
419 * it has too few free blocks left (min_blocks) or
420 * Parent's group is preferred, if it doesn't satisfy these
421 * conditions we search cyclically through the rest. If none
422 * of the groups look good we just look for a group with more
423 * free inodes than average (starting at parent's group).
426 static int find_group_orlov(struct super_block *sb, struct inode *parent,
427 ext4_group_t *group, umode_t mode,
428 const struct qstr *qstr)
430 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
431 struct ext4_sb_info *sbi = EXT4_SB(sb);
432 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
433 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
434 unsigned int freei, avefreei, grp_free;
435 ext4_fsblk_t freeb, avefreec;
436 unsigned int ndirs;
437 int max_dirs, min_inodes;
438 ext4_grpblk_t min_clusters;
439 ext4_group_t i, grp, g, ngroups;
440 struct ext4_group_desc *desc;
441 struct orlov_stats stats;
442 int flex_size = ext4_flex_bg_size(sbi);
443 struct dx_hash_info hinfo;
445 ngroups = real_ngroups;
446 if (flex_size > 1) {
447 ngroups = (real_ngroups + flex_size - 1) >>
448 sbi->s_log_groups_per_flex;
449 parent_group >>= sbi->s_log_groups_per_flex;
452 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
453 avefreei = freei / ngroups;
454 freeb = EXT4_C2B(sbi,
455 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
456 avefreec = freeb;
457 do_div(avefreec, ngroups);
458 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
460 if (S_ISDIR(mode) &&
461 ((parent == d_inode(sb->s_root)) ||
462 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
463 int best_ndir = inodes_per_group;
464 int ret = -1;
466 if (qstr) {
467 hinfo.hash_version = DX_HASH_HALF_MD4;
468 hinfo.seed = sbi->s_hash_seed;
469 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
470 grp = hinfo.hash;
471 } else
472 grp = prandom_u32();
473 parent_group = (unsigned)grp % ngroups;
474 for (i = 0; i < ngroups; i++) {
475 g = (parent_group + i) % ngroups;
476 get_orlov_stats(sb, g, flex_size, &stats);
477 if (!stats.free_inodes)
478 continue;
479 if (stats.used_dirs >= best_ndir)
480 continue;
481 if (stats.free_inodes < avefreei)
482 continue;
483 if (stats.free_clusters < avefreec)
484 continue;
485 grp = g;
486 ret = 0;
487 best_ndir = stats.used_dirs;
489 if (ret)
490 goto fallback;
491 found_flex_bg:
492 if (flex_size == 1) {
493 *group = grp;
494 return 0;
498 * We pack inodes at the beginning of the flexgroup's
499 * inode tables. Block allocation decisions will do
500 * something similar, although regular files will
501 * start at 2nd block group of the flexgroup. See
502 * ext4_ext_find_goal() and ext4_find_near().
504 grp *= flex_size;
505 for (i = 0; i < flex_size; i++) {
506 if (grp+i >= real_ngroups)
507 break;
508 desc = ext4_get_group_desc(sb, grp+i, NULL);
509 if (desc && ext4_free_inodes_count(sb, desc)) {
510 *group = grp+i;
511 return 0;
514 goto fallback;
517 max_dirs = ndirs / ngroups + inodes_per_group / 16;
518 min_inodes = avefreei - inodes_per_group*flex_size / 4;
519 if (min_inodes < 1)
520 min_inodes = 1;
521 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
524 * Start looking in the flex group where we last allocated an
525 * inode for this parent directory
527 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
528 parent_group = EXT4_I(parent)->i_last_alloc_group;
529 if (flex_size > 1)
530 parent_group >>= sbi->s_log_groups_per_flex;
533 for (i = 0; i < ngroups; i++) {
534 grp = (parent_group + i) % ngroups;
535 get_orlov_stats(sb, grp, flex_size, &stats);
536 if (stats.used_dirs >= max_dirs)
537 continue;
538 if (stats.free_inodes < min_inodes)
539 continue;
540 if (stats.free_clusters < min_clusters)
541 continue;
542 goto found_flex_bg;
545 fallback:
546 ngroups = real_ngroups;
547 avefreei = freei / ngroups;
548 fallback_retry:
549 parent_group = EXT4_I(parent)->i_block_group;
550 for (i = 0; i < ngroups; i++) {
551 grp = (parent_group + i) % ngroups;
552 desc = ext4_get_group_desc(sb, grp, NULL);
553 if (desc) {
554 grp_free = ext4_free_inodes_count(sb, desc);
555 if (grp_free && grp_free >= avefreei) {
556 *group = grp;
557 return 0;
562 if (avefreei) {
564 * The free-inodes counter is approximate, and for really small
565 * filesystems the above test can fail to find any blockgroups
567 avefreei = 0;
568 goto fallback_retry;
571 return -1;
574 static int find_group_other(struct super_block *sb, struct inode *parent,
575 ext4_group_t *group, umode_t mode)
577 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
578 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
579 struct ext4_group_desc *desc;
580 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
583 * Try to place the inode is the same flex group as its
584 * parent. If we can't find space, use the Orlov algorithm to
585 * find another flex group, and store that information in the
586 * parent directory's inode information so that use that flex
587 * group for future allocations.
589 if (flex_size > 1) {
590 int retry = 0;
592 try_again:
593 parent_group &= ~(flex_size-1);
594 last = parent_group + flex_size;
595 if (last > ngroups)
596 last = ngroups;
597 for (i = parent_group; i < last; i++) {
598 desc = ext4_get_group_desc(sb, i, NULL);
599 if (desc && ext4_free_inodes_count(sb, desc)) {
600 *group = i;
601 return 0;
604 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
605 retry = 1;
606 parent_group = EXT4_I(parent)->i_last_alloc_group;
607 goto try_again;
610 * If this didn't work, use the Orlov search algorithm
611 * to find a new flex group; we pass in the mode to
612 * avoid the topdir algorithms.
614 *group = parent_group + flex_size;
615 if (*group > ngroups)
616 *group = 0;
617 return find_group_orlov(sb, parent, group, mode, NULL);
621 * Try to place the inode in its parent directory
623 *group = parent_group;
624 desc = ext4_get_group_desc(sb, *group, NULL);
625 if (desc && ext4_free_inodes_count(sb, desc) &&
626 ext4_free_group_clusters(sb, desc))
627 return 0;
630 * We're going to place this inode in a different blockgroup from its
631 * parent. We want to cause files in a common directory to all land in
632 * the same blockgroup. But we want files which are in a different
633 * directory which shares a blockgroup with our parent to land in a
634 * different blockgroup.
636 * So add our directory's i_ino into the starting point for the hash.
638 *group = (*group + parent->i_ino) % ngroups;
641 * Use a quadratic hash to find a group with a free inode and some free
642 * blocks.
644 for (i = 1; i < ngroups; i <<= 1) {
645 *group += i;
646 if (*group >= ngroups)
647 *group -= ngroups;
648 desc = ext4_get_group_desc(sb, *group, NULL);
649 if (desc && ext4_free_inodes_count(sb, desc) &&
650 ext4_free_group_clusters(sb, desc))
651 return 0;
655 * That failed: try linear search for a free inode, even if that group
656 * has no free blocks.
658 *group = parent_group;
659 for (i = 0; i < ngroups; i++) {
660 if (++*group >= ngroups)
661 *group = 0;
662 desc = ext4_get_group_desc(sb, *group, NULL);
663 if (desc && ext4_free_inodes_count(sb, desc))
664 return 0;
667 return -1;
671 * In no journal mode, if an inode has recently been deleted, we want
672 * to avoid reusing it until we're reasonably sure the inode table
673 * block has been written back to disk. (Yes, these values are
674 * somewhat arbitrary...)
676 #define RECENTCY_MIN 60
677 #define RECENTCY_DIRTY 300
679 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
681 struct ext4_group_desc *gdp;
682 struct ext4_inode *raw_inode;
683 struct buffer_head *bh;
684 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
685 int offset, ret = 0;
686 int recentcy = RECENTCY_MIN;
687 u32 dtime, now;
689 gdp = ext4_get_group_desc(sb, group, NULL);
690 if (unlikely(!gdp))
691 return 0;
693 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
694 (ino / inodes_per_block));
695 if (!bh || !buffer_uptodate(bh))
697 * If the block is not in the buffer cache, then it
698 * must have been written out.
700 goto out;
702 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
703 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
705 /* i_dtime is only 32 bits on disk, but we only care about relative
706 * times in the range of a few minutes (i.e. long enough to sync a
707 * recently-deleted inode to disk), so using the low 32 bits of the
708 * clock (a 68 year range) is enough, see time_before32() */
709 dtime = le32_to_cpu(raw_inode->i_dtime);
710 now = ktime_get_real_seconds();
711 if (buffer_dirty(bh))
712 recentcy += RECENTCY_DIRTY;
714 if (dtime && time_before32(dtime, now) &&
715 time_before32(now, dtime + recentcy))
716 ret = 1;
717 out:
718 brelse(bh);
719 return ret;
722 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
723 struct buffer_head *bitmap, unsigned long *ino)
725 next:
726 *ino = ext4_find_next_zero_bit((unsigned long *)
727 bitmap->b_data,
728 EXT4_INODES_PER_GROUP(sb), *ino);
729 if (*ino >= EXT4_INODES_PER_GROUP(sb))
730 return 0;
732 if ((EXT4_SB(sb)->s_journal == NULL) &&
733 recently_deleted(sb, group, *ino)) {
734 *ino = *ino + 1;
735 if (*ino < EXT4_INODES_PER_GROUP(sb))
736 goto next;
737 return 0;
740 return 1;
744 * There are two policies for allocating an inode. If the new inode is
745 * a directory, then a forward search is made for a block group with both
746 * free space and a low directory-to-inode ratio; if that fails, then of
747 * the groups with above-average free space, that group with the fewest
748 * directories already is chosen.
750 * For other inodes, search forward from the parent directory's block
751 * group to find a free inode.
753 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
754 umode_t mode, const struct qstr *qstr,
755 __u32 goal, uid_t *owner, __u32 i_flags,
756 int handle_type, unsigned int line_no,
757 int nblocks)
759 struct super_block *sb;
760 struct buffer_head *inode_bitmap_bh = NULL;
761 struct buffer_head *group_desc_bh;
762 ext4_group_t ngroups, group = 0;
763 unsigned long ino = 0;
764 struct inode *inode;
765 struct ext4_group_desc *gdp = NULL;
766 struct ext4_inode_info *ei;
767 struct ext4_sb_info *sbi;
768 int ret2, err;
769 struct inode *ret;
770 ext4_group_t i;
771 ext4_group_t flex_group;
772 struct ext4_group_info *grp;
773 int encrypt = 0;
775 /* Cannot create files in a deleted directory */
776 if (!dir || !dir->i_nlink)
777 return ERR_PTR(-EPERM);
779 sb = dir->i_sb;
780 sbi = EXT4_SB(sb);
782 if (unlikely(ext4_forced_shutdown(sbi)))
783 return ERR_PTR(-EIO);
785 if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
786 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
787 !(i_flags & EXT4_EA_INODE_FL)) {
788 err = fscrypt_get_encryption_info(dir);
789 if (err)
790 return ERR_PTR(err);
791 if (!fscrypt_has_encryption_key(dir))
792 return ERR_PTR(-ENOKEY);
793 encrypt = 1;
796 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
797 #ifdef CONFIG_EXT4_FS_POSIX_ACL
798 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
800 if (IS_ERR(p))
801 return ERR_CAST(p);
802 if (p) {
803 int acl_size = p->a_count * sizeof(ext4_acl_entry);
805 nblocks += (S_ISDIR(mode) ? 2 : 1) *
806 __ext4_xattr_set_credits(sb, NULL /* inode */,
807 NULL /* block_bh */, acl_size,
808 true /* is_create */);
809 posix_acl_release(p);
811 #endif
813 #ifdef CONFIG_SECURITY
815 int num_security_xattrs = 1;
817 #ifdef CONFIG_INTEGRITY
818 num_security_xattrs++;
819 #endif
821 * We assume that security xattrs are never
822 * more than 1k. In practice they are under
823 * 128 bytes.
825 nblocks += num_security_xattrs *
826 __ext4_xattr_set_credits(sb, NULL /* inode */,
827 NULL /* block_bh */, 1024,
828 true /* is_create */);
830 #endif
831 if (encrypt)
832 nblocks += __ext4_xattr_set_credits(sb,
833 NULL /* inode */, NULL /* block_bh */,
834 FSCRYPT_SET_CONTEXT_MAX_SIZE,
835 true /* is_create */);
838 ngroups = ext4_get_groups_count(sb);
839 trace_ext4_request_inode(dir, mode);
840 inode = new_inode(sb);
841 if (!inode)
842 return ERR_PTR(-ENOMEM);
843 ei = EXT4_I(inode);
846 * Initialize owners and quota early so that we don't have to account
847 * for quota initialization worst case in standard inode creating
848 * transaction
850 if (owner) {
851 inode->i_mode = mode;
852 i_uid_write(inode, owner[0]);
853 i_gid_write(inode, owner[1]);
854 } else if (test_opt(sb, GRPID)) {
855 inode->i_mode = mode;
856 inode->i_uid = current_fsuid();
857 inode->i_gid = dir->i_gid;
858 } else
859 inode_init_owner(inode, dir, mode);
861 if (ext4_has_feature_project(sb) &&
862 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
863 ei->i_projid = EXT4_I(dir)->i_projid;
864 else
865 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
867 err = dquot_initialize(inode);
868 if (err)
869 goto out;
871 if (!goal)
872 goal = sbi->s_inode_goal;
874 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
875 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
876 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
877 ret2 = 0;
878 goto got_group;
881 if (S_ISDIR(mode))
882 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
883 else
884 ret2 = find_group_other(sb, dir, &group, mode);
886 got_group:
887 EXT4_I(dir)->i_last_alloc_group = group;
888 err = -ENOSPC;
889 if (ret2 == -1)
890 goto out;
893 * Normally we will only go through one pass of this loop,
894 * unless we get unlucky and it turns out the group we selected
895 * had its last inode grabbed by someone else.
897 for (i = 0; i < ngroups; i++, ino = 0) {
898 err = -EIO;
900 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
901 if (!gdp)
902 goto out;
905 * Check free inodes count before loading bitmap.
907 if (ext4_free_inodes_count(sb, gdp) == 0)
908 goto next_group;
910 grp = ext4_get_group_info(sb, group);
911 /* Skip groups with already-known suspicious inode tables */
912 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
913 goto next_group;
915 brelse(inode_bitmap_bh);
916 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
917 /* Skip groups with suspicious inode tables */
918 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
919 IS_ERR(inode_bitmap_bh)) {
920 inode_bitmap_bh = NULL;
921 goto next_group;
924 repeat_in_this_group:
925 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
926 if (!ret2)
927 goto next_group;
929 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
930 ext4_error(sb, "reserved inode found cleared - "
931 "inode=%lu", ino + 1);
932 goto next_group;
935 if (!handle) {
936 BUG_ON(nblocks <= 0);
937 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
938 handle_type, nblocks,
940 if (IS_ERR(handle)) {
941 err = PTR_ERR(handle);
942 ext4_std_error(sb, err);
943 goto out;
946 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
947 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
948 if (err) {
949 ext4_std_error(sb, err);
950 goto out;
952 ext4_lock_group(sb, group);
953 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
954 if (ret2) {
955 /* Someone already took the bit. Repeat the search
956 * with lock held.
958 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
959 if (ret2) {
960 ext4_set_bit(ino, inode_bitmap_bh->b_data);
961 ret2 = 0;
962 } else {
963 ret2 = 1; /* we didn't grab the inode */
966 ext4_unlock_group(sb, group);
967 ino++; /* the inode bitmap is zero-based */
968 if (!ret2)
969 goto got; /* we grabbed the inode! */
971 if (ino < EXT4_INODES_PER_GROUP(sb))
972 goto repeat_in_this_group;
973 next_group:
974 if (++group == ngroups)
975 group = 0;
977 err = -ENOSPC;
978 goto out;
980 got:
981 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
982 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
983 if (err) {
984 ext4_std_error(sb, err);
985 goto out;
988 BUFFER_TRACE(group_desc_bh, "get_write_access");
989 err = ext4_journal_get_write_access(handle, group_desc_bh);
990 if (err) {
991 ext4_std_error(sb, err);
992 goto out;
995 /* We may have to initialize the block bitmap if it isn't already */
996 if (ext4_has_group_desc_csum(sb) &&
997 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
998 struct buffer_head *block_bitmap_bh;
1000 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1001 if (IS_ERR(block_bitmap_bh)) {
1002 err = PTR_ERR(block_bitmap_bh);
1003 goto out;
1005 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1006 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1007 if (err) {
1008 brelse(block_bitmap_bh);
1009 ext4_std_error(sb, err);
1010 goto out;
1013 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1014 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1016 /* recheck and clear flag under lock if we still need to */
1017 ext4_lock_group(sb, group);
1018 if (ext4_has_group_desc_csum(sb) &&
1019 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1020 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1021 ext4_free_group_clusters_set(sb, gdp,
1022 ext4_free_clusters_after_init(sb, group, gdp));
1023 ext4_block_bitmap_csum_set(sb, group, gdp,
1024 block_bitmap_bh);
1025 ext4_group_desc_csum_set(sb, group, gdp);
1027 ext4_unlock_group(sb, group);
1028 brelse(block_bitmap_bh);
1030 if (err) {
1031 ext4_std_error(sb, err);
1032 goto out;
1036 /* Update the relevant bg descriptor fields */
1037 if (ext4_has_group_desc_csum(sb)) {
1038 int free;
1039 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1041 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1042 ext4_lock_group(sb, group); /* while we modify the bg desc */
1043 free = EXT4_INODES_PER_GROUP(sb) -
1044 ext4_itable_unused_count(sb, gdp);
1045 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1046 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1047 free = 0;
1050 * Check the relative inode number against the last used
1051 * relative inode number in this group. if it is greater
1052 * we need to update the bg_itable_unused count
1054 if (ino > free)
1055 ext4_itable_unused_set(sb, gdp,
1056 (EXT4_INODES_PER_GROUP(sb) - ino));
1057 up_read(&grp->alloc_sem);
1058 } else {
1059 ext4_lock_group(sb, group);
1062 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1063 if (S_ISDIR(mode)) {
1064 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1065 if (sbi->s_log_groups_per_flex) {
1066 ext4_group_t f = ext4_flex_group(sbi, group);
1068 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1069 f)->used_dirs);
1072 if (ext4_has_group_desc_csum(sb)) {
1073 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1074 EXT4_INODES_PER_GROUP(sb) / 8);
1075 ext4_group_desc_csum_set(sb, group, gdp);
1077 ext4_unlock_group(sb, group);
1079 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1080 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1081 if (err) {
1082 ext4_std_error(sb, err);
1083 goto out;
1086 percpu_counter_dec(&sbi->s_freeinodes_counter);
1087 if (S_ISDIR(mode))
1088 percpu_counter_inc(&sbi->s_dirs_counter);
1090 if (sbi->s_log_groups_per_flex) {
1091 flex_group = ext4_flex_group(sbi, group);
1092 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1093 flex_group)->free_inodes);
1096 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1097 /* This is the optimal IO size (for stat), not the fs block size */
1098 inode->i_blocks = 0;
1099 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1100 current_time(inode);
1102 memset(ei->i_data, 0, sizeof(ei->i_data));
1103 ei->i_dir_start_lookup = 0;
1104 ei->i_disksize = 0;
1106 /* Don't inherit extent flag from directory, amongst others. */
1107 ei->i_flags =
1108 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1109 ei->i_flags |= i_flags;
1110 ei->i_file_acl = 0;
1111 ei->i_dtime = 0;
1112 ei->i_block_group = group;
1113 ei->i_last_alloc_group = ~0;
1115 ext4_set_inode_flags(inode);
1116 if (IS_DIRSYNC(inode))
1117 ext4_handle_sync(handle);
1118 if (insert_inode_locked(inode) < 0) {
1120 * Likely a bitmap corruption causing inode to be allocated
1121 * twice.
1123 err = -EIO;
1124 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1125 inode->i_ino);
1126 goto out;
1128 spin_lock(&sbi->s_next_gen_lock);
1129 inode->i_generation = sbi->s_next_generation++;
1130 spin_unlock(&sbi->s_next_gen_lock);
1132 /* Precompute checksum seed for inode metadata */
1133 if (ext4_has_metadata_csum(sb)) {
1134 __u32 csum;
1135 __le32 inum = cpu_to_le32(inode->i_ino);
1136 __le32 gen = cpu_to_le32(inode->i_generation);
1137 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1138 sizeof(inum));
1139 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1140 sizeof(gen));
1143 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1144 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1146 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1147 ei->i_inline_off = 0;
1148 if (ext4_has_feature_inline_data(sb))
1149 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1150 ret = inode;
1151 err = dquot_alloc_inode(inode);
1152 if (err)
1153 goto fail_drop;
1156 * Since the encryption xattr will always be unique, create it first so
1157 * that it's less likely to end up in an external xattr block and
1158 * prevent its deduplication.
1160 if (encrypt) {
1161 err = fscrypt_inherit_context(dir, inode, handle, true);
1162 if (err)
1163 goto fail_free_drop;
1166 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1167 err = ext4_init_acl(handle, inode, dir);
1168 if (err)
1169 goto fail_free_drop;
1171 err = ext4_init_security(handle, inode, dir, qstr);
1172 if (err)
1173 goto fail_free_drop;
1176 if (ext4_has_feature_extents(sb)) {
1177 /* set extent flag only for directory, file and normal symlink*/
1178 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1179 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1180 ext4_ext_tree_init(handle, inode);
1184 if (ext4_handle_valid(handle)) {
1185 ei->i_sync_tid = handle->h_transaction->t_tid;
1186 ei->i_datasync_tid = handle->h_transaction->t_tid;
1189 err = ext4_mark_inode_dirty(handle, inode);
1190 if (err) {
1191 ext4_std_error(sb, err);
1192 goto fail_free_drop;
1195 ext4_debug("allocating inode %lu\n", inode->i_ino);
1196 trace_ext4_allocate_inode(inode, dir, mode);
1197 brelse(inode_bitmap_bh);
1198 return ret;
1200 fail_free_drop:
1201 dquot_free_inode(inode);
1202 fail_drop:
1203 clear_nlink(inode);
1204 unlock_new_inode(inode);
1205 out:
1206 dquot_drop(inode);
1207 inode->i_flags |= S_NOQUOTA;
1208 iput(inode);
1209 brelse(inode_bitmap_bh);
1210 return ERR_PTR(err);
1213 /* Verify that we are loading a valid orphan from disk */
1214 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1216 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1217 ext4_group_t block_group;
1218 int bit;
1219 struct buffer_head *bitmap_bh = NULL;
1220 struct inode *inode = NULL;
1221 int err = -EFSCORRUPTED;
1223 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1224 goto bad_orphan;
1226 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1227 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1228 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1229 if (IS_ERR(bitmap_bh)) {
1230 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1231 ino, PTR_ERR(bitmap_bh));
1232 return (struct inode *) bitmap_bh;
1235 /* Having the inode bit set should be a 100% indicator that this
1236 * is a valid orphan (no e2fsck run on fs). Orphans also include
1237 * inodes that were being truncated, so we can't check i_nlink==0.
1239 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1240 goto bad_orphan;
1242 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1243 if (IS_ERR(inode)) {
1244 err = PTR_ERR(inode);
1245 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1246 ino, err);
1247 return inode;
1251 * If the orphans has i_nlinks > 0 then it should be able to
1252 * be truncated, otherwise it won't be removed from the orphan
1253 * list during processing and an infinite loop will result.
1254 * Similarly, it must not be a bad inode.
1256 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1257 is_bad_inode(inode))
1258 goto bad_orphan;
1260 if (NEXT_ORPHAN(inode) > max_ino)
1261 goto bad_orphan;
1262 brelse(bitmap_bh);
1263 return inode;
1265 bad_orphan:
1266 ext4_error(sb, "bad orphan inode %lu", ino);
1267 if (bitmap_bh)
1268 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1269 bit, (unsigned long long)bitmap_bh->b_blocknr,
1270 ext4_test_bit(bit, bitmap_bh->b_data));
1271 if (inode) {
1272 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1273 is_bad_inode(inode));
1274 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1275 NEXT_ORPHAN(inode));
1276 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1277 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1278 /* Avoid freeing blocks if we got a bad deleted inode */
1279 if (inode->i_nlink == 0)
1280 inode->i_blocks = 0;
1281 iput(inode);
1283 brelse(bitmap_bh);
1284 return ERR_PTR(err);
1287 unsigned long ext4_count_free_inodes(struct super_block *sb)
1289 unsigned long desc_count;
1290 struct ext4_group_desc *gdp;
1291 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1292 #ifdef EXT4FS_DEBUG
1293 struct ext4_super_block *es;
1294 unsigned long bitmap_count, x;
1295 struct buffer_head *bitmap_bh = NULL;
1297 es = EXT4_SB(sb)->s_es;
1298 desc_count = 0;
1299 bitmap_count = 0;
1300 gdp = NULL;
1301 for (i = 0; i < ngroups; i++) {
1302 gdp = ext4_get_group_desc(sb, i, NULL);
1303 if (!gdp)
1304 continue;
1305 desc_count += ext4_free_inodes_count(sb, gdp);
1306 brelse(bitmap_bh);
1307 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1308 if (IS_ERR(bitmap_bh)) {
1309 bitmap_bh = NULL;
1310 continue;
1313 x = ext4_count_free(bitmap_bh->b_data,
1314 EXT4_INODES_PER_GROUP(sb) / 8);
1315 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1316 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1317 bitmap_count += x;
1319 brelse(bitmap_bh);
1320 printk(KERN_DEBUG "ext4_count_free_inodes: "
1321 "stored = %u, computed = %lu, %lu\n",
1322 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1323 return desc_count;
1324 #else
1325 desc_count = 0;
1326 for (i = 0; i < ngroups; i++) {
1327 gdp = ext4_get_group_desc(sb, i, NULL);
1328 if (!gdp)
1329 continue;
1330 desc_count += ext4_free_inodes_count(sb, gdp);
1331 cond_resched();
1333 return desc_count;
1334 #endif
1337 /* Called at mount-time, super-block is locked */
1338 unsigned long ext4_count_dirs(struct super_block * sb)
1340 unsigned long count = 0;
1341 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1343 for (i = 0; i < ngroups; i++) {
1344 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1345 if (!gdp)
1346 continue;
1347 count += ext4_used_dirs_count(sb, gdp);
1349 return count;
1353 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1354 * inode table. Must be called without any spinlock held. The only place
1355 * where it is called from on active part of filesystem is ext4lazyinit
1356 * thread, so we do not need any special locks, however we have to prevent
1357 * inode allocation from the current group, so we take alloc_sem lock, to
1358 * block ext4_new_inode() until we are finished.
1360 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1361 int barrier)
1363 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1364 struct ext4_sb_info *sbi = EXT4_SB(sb);
1365 struct ext4_group_desc *gdp = NULL;
1366 struct buffer_head *group_desc_bh;
1367 handle_t *handle;
1368 ext4_fsblk_t blk;
1369 int num, ret = 0, used_blks = 0;
1371 /* This should not happen, but just to be sure check this */
1372 if (sb_rdonly(sb)) {
1373 ret = 1;
1374 goto out;
1377 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1378 if (!gdp)
1379 goto out;
1382 * We do not need to lock this, because we are the only one
1383 * handling this flag.
1385 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1386 goto out;
1388 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1389 if (IS_ERR(handle)) {
1390 ret = PTR_ERR(handle);
1391 goto out;
1394 down_write(&grp->alloc_sem);
1396 * If inode bitmap was already initialized there may be some
1397 * used inodes so we need to skip blocks with used inodes in
1398 * inode table.
1400 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1401 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1402 ext4_itable_unused_count(sb, gdp)),
1403 sbi->s_inodes_per_block);
1405 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1406 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1407 ext4_itable_unused_count(sb, gdp)) <
1408 EXT4_FIRST_INO(sb)))) {
1409 ext4_error(sb, "Something is wrong with group %u: "
1410 "used itable blocks: %d; "
1411 "itable unused count: %u",
1412 group, used_blks,
1413 ext4_itable_unused_count(sb, gdp));
1414 ret = 1;
1415 goto err_out;
1418 blk = ext4_inode_table(sb, gdp) + used_blks;
1419 num = sbi->s_itb_per_group - used_blks;
1421 BUFFER_TRACE(group_desc_bh, "get_write_access");
1422 ret = ext4_journal_get_write_access(handle,
1423 group_desc_bh);
1424 if (ret)
1425 goto err_out;
1428 * Skip zeroout if the inode table is full. But we set the ZEROED
1429 * flag anyway, because obviously, when it is full it does not need
1430 * further zeroing.
1432 if (unlikely(num == 0))
1433 goto skip_zeroout;
1435 ext4_debug("going to zero out inode table in group %d\n",
1436 group);
1437 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1438 if (ret < 0)
1439 goto err_out;
1440 if (barrier)
1441 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1443 skip_zeroout:
1444 ext4_lock_group(sb, group);
1445 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1446 ext4_group_desc_csum_set(sb, group, gdp);
1447 ext4_unlock_group(sb, group);
1449 BUFFER_TRACE(group_desc_bh,
1450 "call ext4_handle_dirty_metadata");
1451 ret = ext4_handle_dirty_metadata(handle, NULL,
1452 group_desc_bh);
1454 err_out:
1455 up_write(&grp->alloc_sem);
1456 ext4_journal_stop(handle);
1457 out:
1458 return ret;