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/jbd2.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 <asm/byteorder.h>
26
27#include "ext4.h"
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31
32#include <trace/events/ext4.h>
33
34/*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38/*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
48/*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
53void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54{
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65}
66
67/* Initializes an uninitialized inode bitmap */
68static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
72{
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74
75 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
80 ext4_error(sb, "Checksum bad for group %u", block_group);
81 ext4_free_blks_set(sb, gdp, 0);
82 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
84 memset(bh->b_data, 0xff, sb->s_blocksize);
85 return 0;
86 }
87
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
90 bh->b_data);
91
92 return EXT4_INODES_PER_GROUP(sb);
93}
94
95/*
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
98 *
99 * Return buffer_head of bitmap on success or NULL.
100 */
101static struct buffer_head *
102ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
103{
104 struct ext4_group_desc *desc;
105 struct buffer_head *bh = NULL;
106 ext4_fsblk_t bitmap_blk;
107
108 desc = ext4_get_group_desc(sb, block_group, NULL);
109 if (!desc)
110 return NULL;
111
112 bitmap_blk = ext4_inode_bitmap(sb, desc);
113 bh = sb_getblk(sb, bitmap_blk);
114 if (unlikely(!bh)) {
115 ext4_error(sb, "Cannot read inode bitmap - "
116 "block_group = %u, inode_bitmap = %llu",
117 block_group, bitmap_blk);
118 return NULL;
119 }
120 if (bitmap_uptodate(bh))
121 return bh;
122
123 lock_buffer(bh);
124 if (bitmap_uptodate(bh)) {
125 unlock_buffer(bh);
126 return bh;
127 }
128
129 ext4_lock_group(sb, block_group);
130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
132 set_bitmap_uptodate(bh);
133 set_buffer_uptodate(bh);
134 ext4_unlock_group(sb, block_group);
135 unlock_buffer(bh);
136 return bh;
137 }
138 ext4_unlock_group(sb, block_group);
139
140 if (buffer_uptodate(bh)) {
141 /*
142 * if not uninit if bh is uptodate,
143 * bitmap is also uptodate
144 */
145 set_bitmap_uptodate(bh);
146 unlock_buffer(bh);
147 return bh;
148 }
149 /*
150 * submit the buffer_head for read. We can
151 * safely mark the bitmap as uptodate now.
152 * We do it here so the bitmap uptodate bit
153 * get set with buffer lock held.
154 */
155 trace_ext4_load_inode_bitmap(sb, block_group);
156 set_bitmap_uptodate(bh);
157 if (bh_submit_read(bh) < 0) {
158 put_bh(bh);
159 ext4_error(sb, "Cannot read inode bitmap - "
160 "block_group = %u, inode_bitmap = %llu",
161 block_group, bitmap_blk);
162 return NULL;
163 }
164 return bh;
165}
166
167/*
168 * NOTE! When we get the inode, we're the only people
169 * that have access to it, and as such there are no
170 * race conditions we have to worry about. The inode
171 * is not on the hash-lists, and it cannot be reached
172 * through the filesystem because the directory entry
173 * has been deleted earlier.
174 *
175 * HOWEVER: we must make sure that we get no aliases,
176 * which means that we have to call "clear_inode()"
177 * _before_ we mark the inode not in use in the inode
178 * bitmaps. Otherwise a newly created file might use
179 * the same inode number (not actually the same pointer
180 * though), and then we'd have two inodes sharing the
181 * same inode number and space on the harddisk.
182 */
183void ext4_free_inode(handle_t *handle, struct inode *inode)
184{
185 struct super_block *sb = inode->i_sb;
186 int is_directory;
187 unsigned long ino;
188 struct buffer_head *bitmap_bh = NULL;
189 struct buffer_head *bh2;
190 ext4_group_t block_group;
191 unsigned long bit;
192 struct ext4_group_desc *gdp;
193 struct ext4_super_block *es;
194 struct ext4_sb_info *sbi;
195 int fatal = 0, err, count, cleared;
196
197 if (atomic_read(&inode->i_count) > 1) {
198 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
199 atomic_read(&inode->i_count));
200 return;
201 }
202 if (inode->i_nlink) {
203 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
204 inode->i_nlink);
205 return;
206 }
207 if (!sb) {
208 printk(KERN_ERR "ext4_free_inode: inode on "
209 "nonexistent device\n");
210 return;
211 }
212 sbi = EXT4_SB(sb);
213
214 ino = inode->i_ino;
215 ext4_debug("freeing inode %lu\n", ino);
216 trace_ext4_free_inode(inode);
217
218 /*
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
221 */
222 dquot_initialize(inode);
223 ext4_xattr_delete_inode(handle, inode);
224 dquot_free_inode(inode);
225 dquot_drop(inode);
226
227 is_directory = S_ISDIR(inode->i_mode);
228
229 /* Do this BEFORE marking the inode not in use or returning an error */
230 ext4_clear_inode(inode);
231
232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
234 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
235 goto error_return;
236 }
237 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
238 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
239 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
240 if (!bitmap_bh)
241 goto error_return;
242
243 BUFFER_TRACE(bitmap_bh, "get_write_access");
244 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
245 if (fatal)
246 goto error_return;
247
248 fatal = -ESRCH;
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
250 if (gdp) {
251 BUFFER_TRACE(bh2, "get_write_access");
252 fatal = ext4_journal_get_write_access(handle, bh2);
253 }
254 ext4_lock_group(sb, block_group);
255 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
256 if (fatal || !cleared) {
257 ext4_unlock_group(sb, block_group);
258 goto out;
259 }
260
261 count = ext4_free_inodes_count(sb, gdp) + 1;
262 ext4_free_inodes_set(sb, gdp, count);
263 if (is_directory) {
264 count = ext4_used_dirs_count(sb, gdp) - 1;
265 ext4_used_dirs_set(sb, gdp, count);
266 percpu_counter_dec(&sbi->s_dirs_counter);
267 }
268 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
269 ext4_unlock_group(sb, block_group);
270
271 percpu_counter_inc(&sbi->s_freeinodes_counter);
272 if (sbi->s_log_groups_per_flex) {
273 ext4_group_t f = ext4_flex_group(sbi, block_group);
274
275 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
276 if (is_directory)
277 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
278 }
279 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
280 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
281out:
282 if (cleared) {
283 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
284 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
285 if (!fatal)
286 fatal = err;
287 ext4_mark_super_dirty(sb);
288 } else
289 ext4_error(sb, "bit already cleared for inode %lu", ino);
290
291error_return:
292 brelse(bitmap_bh);
293 ext4_std_error(sb, fatal);
294}
295
296/*
297 * There are two policies for allocating an inode. If the new inode is
298 * a directory, then a forward search is made for a block group with both
299 * free space and a low directory-to-inode ratio; if that fails, then of
300 * the groups with above-average free space, that group with the fewest
301 * directories already is chosen.
302 *
303 * For other inodes, search forward from the parent directory\'s block
304 * group to find a free inode.
305 */
306static int find_group_dir(struct super_block *sb, struct inode *parent,
307 ext4_group_t *best_group)
308{
309 ext4_group_t ngroups = ext4_get_groups_count(sb);
310 unsigned int freei, avefreei;
311 struct ext4_group_desc *desc, *best_desc = NULL;
312 ext4_group_t group;
313 int ret = -1;
314
315 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
316 avefreei = freei / ngroups;
317
318 for (group = 0; group < ngroups; group++) {
319 desc = ext4_get_group_desc(sb, group, NULL);
320 if (!desc || !ext4_free_inodes_count(sb, desc))
321 continue;
322 if (ext4_free_inodes_count(sb, desc) < avefreei)
323 continue;
324 if (!best_desc ||
325 (ext4_free_blks_count(sb, desc) >
326 ext4_free_blks_count(sb, best_desc))) {
327 *best_group = group;
328 best_desc = desc;
329 ret = 0;
330 }
331 }
332 return ret;
333}
334
335#define free_block_ratio 10
336
337static int find_group_flex(struct super_block *sb, struct inode *parent,
338 ext4_group_t *best_group)
339{
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
341 struct ext4_group_desc *desc;
342 struct flex_groups *flex_group = sbi->s_flex_groups;
343 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
344 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
345 ext4_group_t ngroups = ext4_get_groups_count(sb);
346 int flex_size = ext4_flex_bg_size(sbi);
347 ext4_group_t best_flex = parent_fbg_group;
348 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
349 int flexbg_free_blocks;
350 int flex_freeb_ratio;
351 ext4_group_t n_fbg_groups;
352 ext4_group_t i;
353
354 n_fbg_groups = (ngroups + flex_size - 1) >>
355 sbi->s_log_groups_per_flex;
356
357find_close_to_parent:
358 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
359 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
360 if (atomic_read(&flex_group[best_flex].free_inodes) &&
361 flex_freeb_ratio > free_block_ratio)
362 goto found_flexbg;
363
364 if (best_flex && best_flex == parent_fbg_group) {
365 best_flex--;
366 goto find_close_to_parent;
367 }
368
369 for (i = 0; i < n_fbg_groups; i++) {
370 if (i == parent_fbg_group || i == parent_fbg_group - 1)
371 continue;
372
373 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
374 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
375
376 if (flex_freeb_ratio > free_block_ratio &&
377 (atomic_read(&flex_group[i].free_inodes))) {
378 best_flex = i;
379 goto found_flexbg;
380 }
381
382 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
383 ((atomic_read(&flex_group[i].free_blocks) >
384 atomic_read(&flex_group[best_flex].free_blocks)) &&
385 atomic_read(&flex_group[i].free_inodes)))
386 best_flex = i;
387 }
388
389 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
390 !atomic_read(&flex_group[best_flex].free_blocks))
391 return -1;
392
393found_flexbg:
394 for (i = best_flex * flex_size; i < ngroups &&
395 i < (best_flex + 1) * flex_size; i++) {
396 desc = ext4_get_group_desc(sb, i, NULL);
397 if (ext4_free_inodes_count(sb, desc)) {
398 *best_group = i;
399 goto out;
400 }
401 }
402
403 return -1;
404out:
405 return 0;
406}
407
408struct orlov_stats {
409 __u32 free_inodes;
410 __u32 free_blocks;
411 __u32 used_dirs;
412};
413
414/*
415 * Helper function for Orlov's allocator; returns critical information
416 * for a particular block group or flex_bg. If flex_size is 1, then g
417 * is a block group number; otherwise it is flex_bg number.
418 */
419static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
420 int flex_size, struct orlov_stats *stats)
421{
422 struct ext4_group_desc *desc;
423 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
424
425 if (flex_size > 1) {
426 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
427 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
428 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
429 return;
430 }
431
432 desc = ext4_get_group_desc(sb, g, NULL);
433 if (desc) {
434 stats->free_inodes = ext4_free_inodes_count(sb, desc);
435 stats->free_blocks = ext4_free_blks_count(sb, desc);
436 stats->used_dirs = ext4_used_dirs_count(sb, desc);
437 } else {
438 stats->free_inodes = 0;
439 stats->free_blocks = 0;
440 stats->used_dirs = 0;
441 }
442}
443
444/*
445 * Orlov's allocator for directories.
446 *
447 * We always try to spread first-level directories.
448 *
449 * If there are blockgroups with both free inodes and free blocks counts
450 * not worse than average we return one with smallest directory count.
451 * Otherwise we simply return a random group.
452 *
453 * For the rest rules look so:
454 *
455 * It's OK to put directory into a group unless
456 * it has too many directories already (max_dirs) or
457 * it has too few free inodes left (min_inodes) or
458 * it has too few free blocks left (min_blocks) or
459 * Parent's group is preferred, if it doesn't satisfy these
460 * conditions we search cyclically through the rest. If none
461 * of the groups look good we just look for a group with more
462 * free inodes than average (starting at parent's group).
463 */
464
465static int find_group_orlov(struct super_block *sb, struct inode *parent,
466 ext4_group_t *group, int mode,
467 const struct qstr *qstr)
468{
469 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
470 struct ext4_sb_info *sbi = EXT4_SB(sb);
471 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
472 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
473 unsigned int freei, avefreei;
474 ext4_fsblk_t freeb, avefreeb;
475 unsigned int ndirs;
476 int max_dirs, min_inodes;
477 ext4_grpblk_t min_blocks;
478 ext4_group_t i, grp, g, ngroups;
479 struct ext4_group_desc *desc;
480 struct orlov_stats stats;
481 int flex_size = ext4_flex_bg_size(sbi);
482 struct dx_hash_info hinfo;
483
484 ngroups = real_ngroups;
485 if (flex_size > 1) {
486 ngroups = (real_ngroups + flex_size - 1) >>
487 sbi->s_log_groups_per_flex;
488 parent_group >>= sbi->s_log_groups_per_flex;
489 }
490
491 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
492 avefreei = freei / ngroups;
493 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
494 avefreeb = freeb;
495 do_div(avefreeb, ngroups);
496 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
497
498 if (S_ISDIR(mode) &&
499 ((parent == sb->s_root->d_inode) ||
500 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
501 int best_ndir = inodes_per_group;
502 int ret = -1;
503
504 if (qstr) {
505 hinfo.hash_version = DX_HASH_HALF_MD4;
506 hinfo.seed = sbi->s_hash_seed;
507 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
508 grp = hinfo.hash;
509 } else
510 get_random_bytes(&grp, sizeof(grp));
511 parent_group = (unsigned)grp % ngroups;
512 for (i = 0; i < ngroups; i++) {
513 g = (parent_group + i) % ngroups;
514 get_orlov_stats(sb, g, flex_size, &stats);
515 if (!stats.free_inodes)
516 continue;
517 if (stats.used_dirs >= best_ndir)
518 continue;
519 if (stats.free_inodes < avefreei)
520 continue;
521 if (stats.free_blocks < avefreeb)
522 continue;
523 grp = g;
524 ret = 0;
525 best_ndir = stats.used_dirs;
526 }
527 if (ret)
528 goto fallback;
529 found_flex_bg:
530 if (flex_size == 1) {
531 *group = grp;
532 return 0;
533 }
534
535 /*
536 * We pack inodes at the beginning of the flexgroup's
537 * inode tables. Block allocation decisions will do
538 * something similar, although regular files will
539 * start at 2nd block group of the flexgroup. See
540 * ext4_ext_find_goal() and ext4_find_near().
541 */
542 grp *= flex_size;
543 for (i = 0; i < flex_size; i++) {
544 if (grp+i >= real_ngroups)
545 break;
546 desc = ext4_get_group_desc(sb, grp+i, NULL);
547 if (desc && ext4_free_inodes_count(sb, desc)) {
548 *group = grp+i;
549 return 0;
550 }
551 }
552 goto fallback;
553 }
554
555 max_dirs = ndirs / ngroups + inodes_per_group / 16;
556 min_inodes = avefreei - inodes_per_group*flex_size / 4;
557 if (min_inodes < 1)
558 min_inodes = 1;
559 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
560
561 /*
562 * Start looking in the flex group where we last allocated an
563 * inode for this parent directory
564 */
565 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
566 parent_group = EXT4_I(parent)->i_last_alloc_group;
567 if (flex_size > 1)
568 parent_group >>= sbi->s_log_groups_per_flex;
569 }
570
571 for (i = 0; i < ngroups; i++) {
572 grp = (parent_group + i) % ngroups;
573 get_orlov_stats(sb, grp, flex_size, &stats);
574 if (stats.used_dirs >= max_dirs)
575 continue;
576 if (stats.free_inodes < min_inodes)
577 continue;
578 if (stats.free_blocks < min_blocks)
579 continue;
580 goto found_flex_bg;
581 }
582
583fallback:
584 ngroups = real_ngroups;
585 avefreei = freei / ngroups;
586fallback_retry:
587 parent_group = EXT4_I(parent)->i_block_group;
588 for (i = 0; i < ngroups; i++) {
589 grp = (parent_group + i) % ngroups;
590 desc = ext4_get_group_desc(sb, grp, NULL);
591 if (desc && ext4_free_inodes_count(sb, desc) &&
592 ext4_free_inodes_count(sb, desc) >= avefreei) {
593 *group = grp;
594 return 0;
595 }
596 }
597
598 if (avefreei) {
599 /*
600 * The free-inodes counter is approximate, and for really small
601 * filesystems the above test can fail to find any blockgroups
602 */
603 avefreei = 0;
604 goto fallback_retry;
605 }
606
607 return -1;
608}
609
610static int find_group_other(struct super_block *sb, struct inode *parent,
611 ext4_group_t *group, int mode)
612{
613 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
614 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
615 struct ext4_group_desc *desc;
616 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
617
618 /*
619 * Try to place the inode is the same flex group as its
620 * parent. If we can't find space, use the Orlov algorithm to
621 * find another flex group, and store that information in the
622 * parent directory's inode information so that use that flex
623 * group for future allocations.
624 */
625 if (flex_size > 1) {
626 int retry = 0;
627
628 try_again:
629 parent_group &= ~(flex_size-1);
630 last = parent_group + flex_size;
631 if (last > ngroups)
632 last = ngroups;
633 for (i = parent_group; i < last; i++) {
634 desc = ext4_get_group_desc(sb, i, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc)) {
636 *group = i;
637 return 0;
638 }
639 }
640 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
641 retry = 1;
642 parent_group = EXT4_I(parent)->i_last_alloc_group;
643 goto try_again;
644 }
645 /*
646 * If this didn't work, use the Orlov search algorithm
647 * to find a new flex group; we pass in the mode to
648 * avoid the topdir algorithms.
649 */
650 *group = parent_group + flex_size;
651 if (*group > ngroups)
652 *group = 0;
653 return find_group_orlov(sb, parent, group, mode, NULL);
654 }
655
656 /*
657 * Try to place the inode in its parent directory
658 */
659 *group = parent_group;
660 desc = ext4_get_group_desc(sb, *group, NULL);
661 if (desc && ext4_free_inodes_count(sb, desc) &&
662 ext4_free_blks_count(sb, desc))
663 return 0;
664
665 /*
666 * We're going to place this inode in a different blockgroup from its
667 * parent. We want to cause files in a common directory to all land in
668 * the same blockgroup. But we want files which are in a different
669 * directory which shares a blockgroup with our parent to land in a
670 * different blockgroup.
671 *
672 * So add our directory's i_ino into the starting point for the hash.
673 */
674 *group = (*group + parent->i_ino) % ngroups;
675
676 /*
677 * Use a quadratic hash to find a group with a free inode and some free
678 * blocks.
679 */
680 for (i = 1; i < ngroups; i <<= 1) {
681 *group += i;
682 if (*group >= ngroups)
683 *group -= ngroups;
684 desc = ext4_get_group_desc(sb, *group, NULL);
685 if (desc && ext4_free_inodes_count(sb, desc) &&
686 ext4_free_blks_count(sb, desc))
687 return 0;
688 }
689
690 /*
691 * That failed: try linear search for a free inode, even if that group
692 * has no free blocks.
693 */
694 *group = parent_group;
695 for (i = 0; i < ngroups; i++) {
696 if (++*group >= ngroups)
697 *group = 0;
698 desc = ext4_get_group_desc(sb, *group, NULL);
699 if (desc && ext4_free_inodes_count(sb, desc))
700 return 0;
701 }
702
703 return -1;
704}
705
706/*
707 * claim the inode from the inode bitmap. If the group
708 * is uninit we need to take the groups's ext4_group_lock
709 * and clear the uninit flag. The inode bitmap update
710 * and group desc uninit flag clear should be done
711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
712 * doesn't race with the ext4_claim_inode
713 */
714static int ext4_claim_inode(struct super_block *sb,
715 struct buffer_head *inode_bitmap_bh,
716 unsigned long ino, ext4_group_t group, int mode)
717{
718 int free = 0, retval = 0, count;
719 struct ext4_sb_info *sbi = EXT4_SB(sb);
720 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
721 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
722
723 /*
724 * We have to be sure that new inode allocation does not race with
725 * inode table initialization, because otherwise we may end up
726 * allocating and writing new inode right before sb_issue_zeroout
727 * takes place and overwriting our new inode with zeroes. So we
728 * take alloc_sem to prevent it.
729 */
730 down_read(&grp->alloc_sem);
731 ext4_lock_group(sb, group);
732 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
733 /* not a free inode */
734 retval = 1;
735 goto err_ret;
736 }
737 ino++;
738 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
739 ino > EXT4_INODES_PER_GROUP(sb)) {
740 ext4_unlock_group(sb, group);
741 up_read(&grp->alloc_sem);
742 ext4_error(sb, "reserved inode or inode > inodes count - "
743 "block_group = %u, inode=%lu", group,
744 ino + group * EXT4_INODES_PER_GROUP(sb));
745 return 1;
746 }
747 /* If we didn't allocate from within the initialized part of the inode
748 * table then we need to initialize up to this inode. */
749 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
750
751 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
752 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
753 /* When marking the block group with
754 * ~EXT4_BG_INODE_UNINIT we don't want to depend
755 * on the value of bg_itable_unused even though
756 * mke2fs could have initialized the same for us.
757 * Instead we calculated the value below
758 */
759
760 free = 0;
761 } else {
762 free = EXT4_INODES_PER_GROUP(sb) -
763 ext4_itable_unused_count(sb, gdp);
764 }
765
766 /*
767 * Check the relative inode number against the last used
768 * relative inode number in this group. if it is greater
769 * we need to update the bg_itable_unused count
770 *
771 */
772 if (ino > free)
773 ext4_itable_unused_set(sb, gdp,
774 (EXT4_INODES_PER_GROUP(sb) - ino));
775 }
776 count = ext4_free_inodes_count(sb, gdp) - 1;
777 ext4_free_inodes_set(sb, gdp, count);
778 if (S_ISDIR(mode)) {
779 count = ext4_used_dirs_count(sb, gdp) + 1;
780 ext4_used_dirs_set(sb, gdp, count);
781 if (sbi->s_log_groups_per_flex) {
782 ext4_group_t f = ext4_flex_group(sbi, group);
783
784 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
785 }
786 }
787 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
788err_ret:
789 ext4_unlock_group(sb, group);
790 up_read(&grp->alloc_sem);
791 return retval;
792}
793
794/*
795 * There are two policies for allocating an inode. If the new inode is
796 * a directory, then a forward search is made for a block group with both
797 * free space and a low directory-to-inode ratio; if that fails, then of
798 * the groups with above-average free space, that group with the fewest
799 * directories already is chosen.
800 *
801 * For other inodes, search forward from the parent directory's block
802 * group to find a free inode.
803 */
804struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
805 const struct qstr *qstr, __u32 goal)
806{
807 struct super_block *sb;
808 struct buffer_head *inode_bitmap_bh = NULL;
809 struct buffer_head *group_desc_bh;
810 ext4_group_t ngroups, group = 0;
811 unsigned long ino = 0;
812 struct inode *inode;
813 struct ext4_group_desc *gdp = NULL;
814 struct ext4_inode_info *ei;
815 struct ext4_sb_info *sbi;
816 int ret2, err = 0;
817 struct inode *ret;
818 ext4_group_t i;
819 int free = 0;
820 static int once = 1;
821 ext4_group_t flex_group;
822
823 /* Cannot create files in a deleted directory */
824 if (!dir || !dir->i_nlink)
825 return ERR_PTR(-EPERM);
826
827 sb = dir->i_sb;
828 ngroups = ext4_get_groups_count(sb);
829 trace_ext4_request_inode(dir, mode);
830 inode = new_inode(sb);
831 if (!inode)
832 return ERR_PTR(-ENOMEM);
833 ei = EXT4_I(inode);
834 sbi = EXT4_SB(sb);
835
836 if (!goal)
837 goal = sbi->s_inode_goal;
838
839 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
840 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
841 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
842 ret2 = 0;
843 goto got_group;
844 }
845
846 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
847 ret2 = find_group_flex(sb, dir, &group);
848 if (ret2 == -1) {
849 ret2 = find_group_other(sb, dir, &group, mode);
850 if (ret2 == 0 && once) {
851 once = 0;
852 printk(KERN_NOTICE "ext4: find_group_flex "
853 "failed, fallback succeeded dir %lu\n",
854 dir->i_ino);
855 }
856 }
857 goto got_group;
858 }
859
860 if (S_ISDIR(mode)) {
861 if (test_opt(sb, OLDALLOC))
862 ret2 = find_group_dir(sb, dir, &group);
863 else
864 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
865 } else
866 ret2 = find_group_other(sb, dir, &group, mode);
867
868got_group:
869 EXT4_I(dir)->i_last_alloc_group = group;
870 err = -ENOSPC;
871 if (ret2 == -1)
872 goto out;
873
874 for (i = 0; i < ngroups; i++, ino = 0) {
875 err = -EIO;
876
877 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
878 if (!gdp)
879 goto fail;
880
881 brelse(inode_bitmap_bh);
882 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
883 if (!inode_bitmap_bh)
884 goto fail;
885
886repeat_in_this_group:
887 ino = ext4_find_next_zero_bit((unsigned long *)
888 inode_bitmap_bh->b_data,
889 EXT4_INODES_PER_GROUP(sb), ino);
890
891 if (ino < EXT4_INODES_PER_GROUP(sb)) {
892
893 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
894 err = ext4_journal_get_write_access(handle,
895 inode_bitmap_bh);
896 if (err)
897 goto fail;
898
899 BUFFER_TRACE(group_desc_bh, "get_write_access");
900 err = ext4_journal_get_write_access(handle,
901 group_desc_bh);
902 if (err)
903 goto fail;
904 if (!ext4_claim_inode(sb, inode_bitmap_bh,
905 ino, group, mode)) {
906 /* we won it */
907 BUFFER_TRACE(inode_bitmap_bh,
908 "call ext4_handle_dirty_metadata");
909 err = ext4_handle_dirty_metadata(handle,
910 NULL,
911 inode_bitmap_bh);
912 if (err)
913 goto fail;
914 /* zero bit is inode number 1*/
915 ino++;
916 goto got;
917 }
918 /* we lost it */
919 ext4_handle_release_buffer(handle, inode_bitmap_bh);
920 ext4_handle_release_buffer(handle, group_desc_bh);
921
922 if (++ino < EXT4_INODES_PER_GROUP(sb))
923 goto repeat_in_this_group;
924 }
925
926 /*
927 * This case is possible in concurrent environment. It is very
928 * rare. We cannot repeat the find_group_xxx() call because
929 * that will simply return the same blockgroup, because the
930 * group descriptor metadata has not yet been updated.
931 * So we just go onto the next blockgroup.
932 */
933 if (++group == ngroups)
934 group = 0;
935 }
936 err = -ENOSPC;
937 goto out;
938
939got:
940 /* We may have to initialize the block bitmap if it isn't already */
941 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
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 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
947 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
948 if (err) {
949 brelse(block_bitmap_bh);
950 goto fail;
951 }
952
953 free = 0;
954 ext4_lock_group(sb, group);
955 /* recheck and clear flag under lock if we still need to */
956 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
957 free = ext4_free_blocks_after_init(sb, group, gdp);
958 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
959 ext4_free_blks_set(sb, gdp, free);
960 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
961 gdp);
962 }
963 ext4_unlock_group(sb, group);
964
965 /* Don't need to dirty bitmap block if we didn't change it */
966 if (free) {
967 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
968 err = ext4_handle_dirty_metadata(handle,
969 NULL, block_bitmap_bh);
970 }
971
972 brelse(block_bitmap_bh);
973 if (err)
974 goto fail;
975 }
976 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
977 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
978 if (err)
979 goto fail;
980
981 percpu_counter_dec(&sbi->s_freeinodes_counter);
982 if (S_ISDIR(mode))
983 percpu_counter_inc(&sbi->s_dirs_counter);
984 ext4_mark_super_dirty(sb);
985
986 if (sbi->s_log_groups_per_flex) {
987 flex_group = ext4_flex_group(sbi, group);
988 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
989 }
990
991 if (test_opt(sb, GRPID)) {
992 inode->i_mode = mode;
993 inode->i_uid = current_fsuid();
994 inode->i_gid = dir->i_gid;
995 } else
996 inode_init_owner(inode, dir, mode);
997
998 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
999 /* This is the optimal IO size (for stat), not the fs block size */
1000 inode->i_blocks = 0;
1001 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1002 ext4_current_time(inode);
1003
1004 memset(ei->i_data, 0, sizeof(ei->i_data));
1005 ei->i_dir_start_lookup = 0;
1006 ei->i_disksize = 0;
1007
1008 /*
1009 * Don't inherit extent flag from directory, amongst others. We set
1010 * extent flag on newly created directory and file only if -o extent
1011 * mount option is specified
1012 */
1013 ei->i_flags =
1014 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1015 ei->i_file_acl = 0;
1016 ei->i_dtime = 0;
1017 ei->i_block_group = group;
1018 ei->i_last_alloc_group = ~0;
1019
1020 ext4_set_inode_flags(inode);
1021 if (IS_DIRSYNC(inode))
1022 ext4_handle_sync(handle);
1023 if (insert_inode_locked(inode) < 0) {
1024 err = -EINVAL;
1025 goto fail_drop;
1026 }
1027 spin_lock(&sbi->s_next_gen_lock);
1028 inode->i_generation = sbi->s_next_generation++;
1029 spin_unlock(&sbi->s_next_gen_lock);
1030
1031 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1032 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1033
1034 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1035
1036 ret = inode;
1037 dquot_initialize(inode);
1038 err = dquot_alloc_inode(inode);
1039 if (err)
1040 goto fail_drop;
1041
1042 err = ext4_init_acl(handle, inode, dir);
1043 if (err)
1044 goto fail_free_drop;
1045
1046 err = ext4_init_security(handle, inode, dir, qstr);
1047 if (err)
1048 goto fail_free_drop;
1049
1050 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1051 /* set extent flag only for directory, file and normal symlink*/
1052 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1053 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1054 ext4_ext_tree_init(handle, inode);
1055 }
1056 }
1057
1058 if (ext4_handle_valid(handle)) {
1059 ei->i_sync_tid = handle->h_transaction->t_tid;
1060 ei->i_datasync_tid = handle->h_transaction->t_tid;
1061 }
1062
1063 err = ext4_mark_inode_dirty(handle, inode);
1064 if (err) {
1065 ext4_std_error(sb, err);
1066 goto fail_free_drop;
1067 }
1068
1069 ext4_debug("allocating inode %lu\n", inode->i_ino);
1070 trace_ext4_allocate_inode(inode, dir, mode);
1071 goto really_out;
1072fail:
1073 ext4_std_error(sb, err);
1074out:
1075 iput(inode);
1076 ret = ERR_PTR(err);
1077really_out:
1078 brelse(inode_bitmap_bh);
1079 return ret;
1080
1081fail_free_drop:
1082 dquot_free_inode(inode);
1083
1084fail_drop:
1085 dquot_drop(inode);
1086 inode->i_flags |= S_NOQUOTA;
1087 inode->i_nlink = 0;
1088 unlock_new_inode(inode);
1089 iput(inode);
1090 brelse(inode_bitmap_bh);
1091 return ERR_PTR(err);
1092}
1093
1094/* Verify that we are loading a valid orphan from disk */
1095struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1096{
1097 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1098 ext4_group_t block_group;
1099 int bit;
1100 struct buffer_head *bitmap_bh;
1101 struct inode *inode = NULL;
1102 long err = -EIO;
1103
1104 /* Error cases - e2fsck has already cleaned up for us */
1105 if (ino > max_ino) {
1106 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1107 goto error;
1108 }
1109
1110 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1111 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1112 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1113 if (!bitmap_bh) {
1114 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1115 goto error;
1116 }
1117
1118 /* Having the inode bit set should be a 100% indicator that this
1119 * is a valid orphan (no e2fsck run on fs). Orphans also include
1120 * inodes that were being truncated, so we can't check i_nlink==0.
1121 */
1122 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1123 goto bad_orphan;
1124
1125 inode = ext4_iget(sb, ino);
1126 if (IS_ERR(inode))
1127 goto iget_failed;
1128
1129 /*
1130 * If the orphans has i_nlinks > 0 then it should be able to be
1131 * truncated, otherwise it won't be removed from the orphan list
1132 * during processing and an infinite loop will result.
1133 */
1134 if (inode->i_nlink && !ext4_can_truncate(inode))
1135 goto bad_orphan;
1136
1137 if (NEXT_ORPHAN(inode) > max_ino)
1138 goto bad_orphan;
1139 brelse(bitmap_bh);
1140 return inode;
1141
1142iget_failed:
1143 err = PTR_ERR(inode);
1144 inode = NULL;
1145bad_orphan:
1146 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1147 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1148 bit, (unsigned long long)bitmap_bh->b_blocknr,
1149 ext4_test_bit(bit, bitmap_bh->b_data));
1150 printk(KERN_NOTICE "inode=%p\n", inode);
1151 if (inode) {
1152 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1153 is_bad_inode(inode));
1154 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1155 NEXT_ORPHAN(inode));
1156 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1157 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1158 /* Avoid freeing blocks if we got a bad deleted inode */
1159 if (inode->i_nlink == 0)
1160 inode->i_blocks = 0;
1161 iput(inode);
1162 }
1163 brelse(bitmap_bh);
1164error:
1165 return ERR_PTR(err);
1166}
1167
1168unsigned long ext4_count_free_inodes(struct super_block *sb)
1169{
1170 unsigned long desc_count;
1171 struct ext4_group_desc *gdp;
1172 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1173#ifdef EXT4FS_DEBUG
1174 struct ext4_super_block *es;
1175 unsigned long bitmap_count, x;
1176 struct buffer_head *bitmap_bh = NULL;
1177
1178 es = EXT4_SB(sb)->s_es;
1179 desc_count = 0;
1180 bitmap_count = 0;
1181 gdp = NULL;
1182 for (i = 0; i < ngroups; i++) {
1183 gdp = ext4_get_group_desc(sb, i, NULL);
1184 if (!gdp)
1185 continue;
1186 desc_count += ext4_free_inodes_count(sb, gdp);
1187 brelse(bitmap_bh);
1188 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1189 if (!bitmap_bh)
1190 continue;
1191
1192 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1193 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1194 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1195 bitmap_count += x;
1196 }
1197 brelse(bitmap_bh);
1198 printk(KERN_DEBUG "ext4_count_free_inodes: "
1199 "stored = %u, computed = %lu, %lu\n",
1200 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1201 return desc_count;
1202#else
1203 desc_count = 0;
1204 for (i = 0; i < ngroups; i++) {
1205 gdp = ext4_get_group_desc(sb, i, NULL);
1206 if (!gdp)
1207 continue;
1208 desc_count += ext4_free_inodes_count(sb, gdp);
1209 cond_resched();
1210 }
1211 return desc_count;
1212#endif
1213}
1214
1215/* Called at mount-time, super-block is locked */
1216unsigned long ext4_count_dirs(struct super_block * sb)
1217{
1218 unsigned long count = 0;
1219 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1220
1221 for (i = 0; i < ngroups; i++) {
1222 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1223 if (!gdp)
1224 continue;
1225 count += ext4_used_dirs_count(sb, gdp);
1226 }
1227 return count;
1228}
1229
1230/*
1231 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1232 * inode table. Must be called without any spinlock held. The only place
1233 * where it is called from on active part of filesystem is ext4lazyinit
1234 * thread, so we do not need any special locks, however we have to prevent
1235 * inode allocation from the current group, so we take alloc_sem lock, to
1236 * block ext4_claim_inode until we are finished.
1237 */
1238extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1239 int barrier)
1240{
1241 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1242 struct ext4_sb_info *sbi = EXT4_SB(sb);
1243 struct ext4_group_desc *gdp = NULL;
1244 struct buffer_head *group_desc_bh;
1245 handle_t *handle;
1246 ext4_fsblk_t blk;
1247 int num, ret = 0, used_blks = 0;
1248
1249 /* This should not happen, but just to be sure check this */
1250 if (sb->s_flags & MS_RDONLY) {
1251 ret = 1;
1252 goto out;
1253 }
1254
1255 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1256 if (!gdp)
1257 goto out;
1258
1259 /*
1260 * We do not need to lock this, because we are the only one
1261 * handling this flag.
1262 */
1263 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1264 goto out;
1265
1266 handle = ext4_journal_start_sb(sb, 1);
1267 if (IS_ERR(handle)) {
1268 ret = PTR_ERR(handle);
1269 goto out;
1270 }
1271
1272 down_write(&grp->alloc_sem);
1273 /*
1274 * If inode bitmap was already initialized there may be some
1275 * used inodes so we need to skip blocks with used inodes in
1276 * inode table.
1277 */
1278 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1279 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1280 ext4_itable_unused_count(sb, gdp)),
1281 sbi->s_inodes_per_block);
1282
1283 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1284 ext4_error(sb, "Something is wrong with group %u\n"
1285 "Used itable blocks: %d"
1286 "itable unused count: %u\n",
1287 group, used_blks,
1288 ext4_itable_unused_count(sb, gdp));
1289 ret = 1;
1290 goto err_out;
1291 }
1292
1293 blk = ext4_inode_table(sb, gdp) + used_blks;
1294 num = sbi->s_itb_per_group - used_blks;
1295
1296 BUFFER_TRACE(group_desc_bh, "get_write_access");
1297 ret = ext4_journal_get_write_access(handle,
1298 group_desc_bh);
1299 if (ret)
1300 goto err_out;
1301
1302 /*
1303 * Skip zeroout if the inode table is full. But we set the ZEROED
1304 * flag anyway, because obviously, when it is full it does not need
1305 * further zeroing.
1306 */
1307 if (unlikely(num == 0))
1308 goto skip_zeroout;
1309
1310 ext4_debug("going to zero out inode table in group %d\n",
1311 group);
1312 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1313 if (ret < 0)
1314 goto err_out;
1315 if (barrier)
1316 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1317
1318skip_zeroout:
1319 ext4_lock_group(sb, group);
1320 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1321 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1322 ext4_unlock_group(sb, group);
1323
1324 BUFFER_TRACE(group_desc_bh,
1325 "call ext4_handle_dirty_metadata");
1326 ret = ext4_handle_dirty_metadata(handle, NULL,
1327 group_desc_bh);
1328
1329err_out:
1330 up_write(&grp->alloc_sem);
1331 ext4_journal_stop(handle);
1332out:
1333 return ret;
1334}
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/jbd2.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 <asm/byteorder.h>
26
27#include "ext4.h"
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31
32#include <trace/events/ext4.h>
33
34/*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38/*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
48/*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
53void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54{
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65}
66
67/* Initializes an uninitialized inode bitmap */
68static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
72{
73 J_ASSERT_BH(bh, buffer_locked(bh));
74
75 /* If checksum is bad mark all blocks and inodes use to prevent
76 * allocation, essentially implementing a per-group read-only flag. */
77 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
78 ext4_error(sb, "Checksum bad for group %u", block_group);
79 ext4_free_group_clusters_set(sb, gdp, 0);
80 ext4_free_inodes_set(sb, gdp, 0);
81 ext4_itable_unused_set(sb, gdp, 0);
82 memset(bh->b_data, 0xff, sb->s_blocksize);
83 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
84 EXT4_INODES_PER_GROUP(sb) / 8);
85 return 0;
86 }
87
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
90 bh->b_data);
91 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
92 EXT4_INODES_PER_GROUP(sb) / 8);
93 ext4_group_desc_csum_set(sb, block_group, gdp);
94
95 return EXT4_INODES_PER_GROUP(sb);
96}
97
98void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
99{
100 if (uptodate) {
101 set_buffer_uptodate(bh);
102 set_bitmap_uptodate(bh);
103 }
104 unlock_buffer(bh);
105 put_bh(bh);
106}
107
108/*
109 * Read the inode allocation bitmap for a given block_group, reading
110 * into the specified slot in the superblock's bitmap cache.
111 *
112 * Return buffer_head of bitmap on success or NULL.
113 */
114static struct buffer_head *
115ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
116{
117 struct ext4_group_desc *desc;
118 struct buffer_head *bh = NULL;
119 ext4_fsblk_t bitmap_blk;
120
121 desc = ext4_get_group_desc(sb, block_group, NULL);
122 if (!desc)
123 return NULL;
124
125 bitmap_blk = ext4_inode_bitmap(sb, desc);
126 bh = sb_getblk(sb, bitmap_blk);
127 if (unlikely(!bh)) {
128 ext4_error(sb, "Cannot read inode bitmap - "
129 "block_group = %u, inode_bitmap = %llu",
130 block_group, bitmap_blk);
131 return NULL;
132 }
133 if (bitmap_uptodate(bh))
134 goto verify;
135
136 lock_buffer(bh);
137 if (bitmap_uptodate(bh)) {
138 unlock_buffer(bh);
139 goto verify;
140 }
141
142 ext4_lock_group(sb, block_group);
143 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
144 ext4_init_inode_bitmap(sb, bh, block_group, desc);
145 set_bitmap_uptodate(bh);
146 set_buffer_uptodate(bh);
147 set_buffer_verified(bh);
148 ext4_unlock_group(sb, block_group);
149 unlock_buffer(bh);
150 return bh;
151 }
152 ext4_unlock_group(sb, block_group);
153
154 if (buffer_uptodate(bh)) {
155 /*
156 * if not uninit if bh is uptodate,
157 * bitmap is also uptodate
158 */
159 set_bitmap_uptodate(bh);
160 unlock_buffer(bh);
161 goto verify;
162 }
163 /*
164 * submit the buffer_head for reading
165 */
166 trace_ext4_load_inode_bitmap(sb, block_group);
167 bh->b_end_io = ext4_end_bitmap_read;
168 get_bh(bh);
169 submit_bh(READ, bh);
170 wait_on_buffer(bh);
171 if (!buffer_uptodate(bh)) {
172 put_bh(bh);
173 ext4_error(sb, "Cannot read inode bitmap - "
174 "block_group = %u, inode_bitmap = %llu",
175 block_group, bitmap_blk);
176 return NULL;
177 }
178
179verify:
180 ext4_lock_group(sb, block_group);
181 if (!buffer_verified(bh) &&
182 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
183 EXT4_INODES_PER_GROUP(sb) / 8)) {
184 ext4_unlock_group(sb, block_group);
185 put_bh(bh);
186 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
187 "inode_bitmap = %llu", block_group, bitmap_blk);
188 return NULL;
189 }
190 ext4_unlock_group(sb, block_group);
191 set_buffer_verified(bh);
192 return bh;
193}
194
195/*
196 * NOTE! When we get the inode, we're the only people
197 * that have access to it, and as such there are no
198 * race conditions we have to worry about. The inode
199 * is not on the hash-lists, and it cannot be reached
200 * through the filesystem because the directory entry
201 * has been deleted earlier.
202 *
203 * HOWEVER: we must make sure that we get no aliases,
204 * which means that we have to call "clear_inode()"
205 * _before_ we mark the inode not in use in the inode
206 * bitmaps. Otherwise a newly created file might use
207 * the same inode number (not actually the same pointer
208 * though), and then we'd have two inodes sharing the
209 * same inode number and space on the harddisk.
210 */
211void ext4_free_inode(handle_t *handle, struct inode *inode)
212{
213 struct super_block *sb = inode->i_sb;
214 int is_directory;
215 unsigned long ino;
216 struct buffer_head *bitmap_bh = NULL;
217 struct buffer_head *bh2;
218 ext4_group_t block_group;
219 unsigned long bit;
220 struct ext4_group_desc *gdp;
221 struct ext4_super_block *es;
222 struct ext4_sb_info *sbi;
223 int fatal = 0, err, count, cleared;
224
225 if (!sb) {
226 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
227 "nonexistent device\n", __func__, __LINE__);
228 return;
229 }
230 if (atomic_read(&inode->i_count) > 1) {
231 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
232 __func__, __LINE__, inode->i_ino,
233 atomic_read(&inode->i_count));
234 return;
235 }
236 if (inode->i_nlink) {
237 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
238 __func__, __LINE__, inode->i_ino, inode->i_nlink);
239 return;
240 }
241 sbi = EXT4_SB(sb);
242
243 ino = inode->i_ino;
244 ext4_debug("freeing inode %lu\n", ino);
245 trace_ext4_free_inode(inode);
246
247 /*
248 * Note: we must free any quota before locking the superblock,
249 * as writing the quota to disk may need the lock as well.
250 */
251 dquot_initialize(inode);
252 ext4_xattr_delete_inode(handle, inode);
253 dquot_free_inode(inode);
254 dquot_drop(inode);
255
256 is_directory = S_ISDIR(inode->i_mode);
257
258 /* Do this BEFORE marking the inode not in use or returning an error */
259 ext4_clear_inode(inode);
260
261 es = EXT4_SB(sb)->s_es;
262 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
263 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
264 goto error_return;
265 }
266 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
267 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
268 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
269 if (!bitmap_bh)
270 goto error_return;
271
272 BUFFER_TRACE(bitmap_bh, "get_write_access");
273 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
274 if (fatal)
275 goto error_return;
276
277 fatal = -ESRCH;
278 gdp = ext4_get_group_desc(sb, block_group, &bh2);
279 if (gdp) {
280 BUFFER_TRACE(bh2, "get_write_access");
281 fatal = ext4_journal_get_write_access(handle, bh2);
282 }
283 ext4_lock_group(sb, block_group);
284 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
285 if (fatal || !cleared) {
286 ext4_unlock_group(sb, block_group);
287 goto out;
288 }
289
290 count = ext4_free_inodes_count(sb, gdp) + 1;
291 ext4_free_inodes_set(sb, gdp, count);
292 if (is_directory) {
293 count = ext4_used_dirs_count(sb, gdp) - 1;
294 ext4_used_dirs_set(sb, gdp, count);
295 percpu_counter_dec(&sbi->s_dirs_counter);
296 }
297 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
298 EXT4_INODES_PER_GROUP(sb) / 8);
299 ext4_group_desc_csum_set(sb, block_group, gdp);
300 ext4_unlock_group(sb, block_group);
301
302 percpu_counter_inc(&sbi->s_freeinodes_counter);
303 if (sbi->s_log_groups_per_flex) {
304 ext4_group_t f = ext4_flex_group(sbi, block_group);
305
306 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
307 if (is_directory)
308 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
309 }
310 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
311 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
312out:
313 if (cleared) {
314 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
315 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
316 if (!fatal)
317 fatal = err;
318 ext4_mark_super_dirty(sb);
319 } else
320 ext4_error(sb, "bit already cleared for inode %lu", ino);
321
322error_return:
323 brelse(bitmap_bh);
324 ext4_std_error(sb, fatal);
325}
326
327struct orlov_stats {
328 __u32 free_inodes;
329 __u32 free_clusters;
330 __u32 used_dirs;
331};
332
333/*
334 * Helper function for Orlov's allocator; returns critical information
335 * for a particular block group or flex_bg. If flex_size is 1, then g
336 * is a block group number; otherwise it is flex_bg number.
337 */
338static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
339 int flex_size, struct orlov_stats *stats)
340{
341 struct ext4_group_desc *desc;
342 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
343
344 if (flex_size > 1) {
345 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
346 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
347 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
348 return;
349 }
350
351 desc = ext4_get_group_desc(sb, g, NULL);
352 if (desc) {
353 stats->free_inodes = ext4_free_inodes_count(sb, desc);
354 stats->free_clusters = ext4_free_group_clusters(sb, desc);
355 stats->used_dirs = ext4_used_dirs_count(sb, desc);
356 } else {
357 stats->free_inodes = 0;
358 stats->free_clusters = 0;
359 stats->used_dirs = 0;
360 }
361}
362
363/*
364 * Orlov's allocator for directories.
365 *
366 * We always try to spread first-level directories.
367 *
368 * If there are blockgroups with both free inodes and free blocks counts
369 * not worse than average we return one with smallest directory count.
370 * Otherwise we simply return a random group.
371 *
372 * For the rest rules look so:
373 *
374 * It's OK to put directory into a group unless
375 * it has too many directories already (max_dirs) or
376 * it has too few free inodes left (min_inodes) or
377 * it has too few free blocks left (min_blocks) or
378 * Parent's group is preferred, if it doesn't satisfy these
379 * conditions we search cyclically through the rest. If none
380 * of the groups look good we just look for a group with more
381 * free inodes than average (starting at parent's group).
382 */
383
384static int find_group_orlov(struct super_block *sb, struct inode *parent,
385 ext4_group_t *group, umode_t mode,
386 const struct qstr *qstr)
387{
388 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
389 struct ext4_sb_info *sbi = EXT4_SB(sb);
390 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
391 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
392 unsigned int freei, avefreei, grp_free;
393 ext4_fsblk_t freeb, avefreec;
394 unsigned int ndirs;
395 int max_dirs, min_inodes;
396 ext4_grpblk_t min_clusters;
397 ext4_group_t i, grp, g, ngroups;
398 struct ext4_group_desc *desc;
399 struct orlov_stats stats;
400 int flex_size = ext4_flex_bg_size(sbi);
401 struct dx_hash_info hinfo;
402
403 ngroups = real_ngroups;
404 if (flex_size > 1) {
405 ngroups = (real_ngroups + flex_size - 1) >>
406 sbi->s_log_groups_per_flex;
407 parent_group >>= sbi->s_log_groups_per_flex;
408 }
409
410 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
411 avefreei = freei / ngroups;
412 freeb = EXT4_C2B(sbi,
413 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
414 avefreec = freeb;
415 do_div(avefreec, ngroups);
416 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
417
418 if (S_ISDIR(mode) &&
419 ((parent == sb->s_root->d_inode) ||
420 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
421 int best_ndir = inodes_per_group;
422 int ret = -1;
423
424 if (qstr) {
425 hinfo.hash_version = DX_HASH_HALF_MD4;
426 hinfo.seed = sbi->s_hash_seed;
427 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
428 grp = hinfo.hash;
429 } else
430 get_random_bytes(&grp, sizeof(grp));
431 parent_group = (unsigned)grp % ngroups;
432 for (i = 0; i < ngroups; i++) {
433 g = (parent_group + i) % ngroups;
434 get_orlov_stats(sb, g, flex_size, &stats);
435 if (!stats.free_inodes)
436 continue;
437 if (stats.used_dirs >= best_ndir)
438 continue;
439 if (stats.free_inodes < avefreei)
440 continue;
441 if (stats.free_clusters < avefreec)
442 continue;
443 grp = g;
444 ret = 0;
445 best_ndir = stats.used_dirs;
446 }
447 if (ret)
448 goto fallback;
449 found_flex_bg:
450 if (flex_size == 1) {
451 *group = grp;
452 return 0;
453 }
454
455 /*
456 * We pack inodes at the beginning of the flexgroup's
457 * inode tables. Block allocation decisions will do
458 * something similar, although regular files will
459 * start at 2nd block group of the flexgroup. See
460 * ext4_ext_find_goal() and ext4_find_near().
461 */
462 grp *= flex_size;
463 for (i = 0; i < flex_size; i++) {
464 if (grp+i >= real_ngroups)
465 break;
466 desc = ext4_get_group_desc(sb, grp+i, NULL);
467 if (desc && ext4_free_inodes_count(sb, desc)) {
468 *group = grp+i;
469 return 0;
470 }
471 }
472 goto fallback;
473 }
474
475 max_dirs = ndirs / ngroups + inodes_per_group / 16;
476 min_inodes = avefreei - inodes_per_group*flex_size / 4;
477 if (min_inodes < 1)
478 min_inodes = 1;
479 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
480
481 /*
482 * Start looking in the flex group where we last allocated an
483 * inode for this parent directory
484 */
485 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
486 parent_group = EXT4_I(parent)->i_last_alloc_group;
487 if (flex_size > 1)
488 parent_group >>= sbi->s_log_groups_per_flex;
489 }
490
491 for (i = 0; i < ngroups; i++) {
492 grp = (parent_group + i) % ngroups;
493 get_orlov_stats(sb, grp, flex_size, &stats);
494 if (stats.used_dirs >= max_dirs)
495 continue;
496 if (stats.free_inodes < min_inodes)
497 continue;
498 if (stats.free_clusters < min_clusters)
499 continue;
500 goto found_flex_bg;
501 }
502
503fallback:
504 ngroups = real_ngroups;
505 avefreei = freei / ngroups;
506fallback_retry:
507 parent_group = EXT4_I(parent)->i_block_group;
508 for (i = 0; i < ngroups; i++) {
509 grp = (parent_group + i) % ngroups;
510 desc = ext4_get_group_desc(sb, grp, NULL);
511 if (desc) {
512 grp_free = ext4_free_inodes_count(sb, desc);
513 if (grp_free && grp_free >= avefreei) {
514 *group = grp;
515 return 0;
516 }
517 }
518 }
519
520 if (avefreei) {
521 /*
522 * The free-inodes counter is approximate, and for really small
523 * filesystems the above test can fail to find any blockgroups
524 */
525 avefreei = 0;
526 goto fallback_retry;
527 }
528
529 return -1;
530}
531
532static int find_group_other(struct super_block *sb, struct inode *parent,
533 ext4_group_t *group, umode_t mode)
534{
535 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
536 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
537 struct ext4_group_desc *desc;
538 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
539
540 /*
541 * Try to place the inode is the same flex group as its
542 * parent. If we can't find space, use the Orlov algorithm to
543 * find another flex group, and store that information in the
544 * parent directory's inode information so that use that flex
545 * group for future allocations.
546 */
547 if (flex_size > 1) {
548 int retry = 0;
549
550 try_again:
551 parent_group &= ~(flex_size-1);
552 last = parent_group + flex_size;
553 if (last > ngroups)
554 last = ngroups;
555 for (i = parent_group; i < last; i++) {
556 desc = ext4_get_group_desc(sb, i, NULL);
557 if (desc && ext4_free_inodes_count(sb, desc)) {
558 *group = i;
559 return 0;
560 }
561 }
562 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
563 retry = 1;
564 parent_group = EXT4_I(parent)->i_last_alloc_group;
565 goto try_again;
566 }
567 /*
568 * If this didn't work, use the Orlov search algorithm
569 * to find a new flex group; we pass in the mode to
570 * avoid the topdir algorithms.
571 */
572 *group = parent_group + flex_size;
573 if (*group > ngroups)
574 *group = 0;
575 return find_group_orlov(sb, parent, group, mode, NULL);
576 }
577
578 /*
579 * Try to place the inode in its parent directory
580 */
581 *group = parent_group;
582 desc = ext4_get_group_desc(sb, *group, NULL);
583 if (desc && ext4_free_inodes_count(sb, desc) &&
584 ext4_free_group_clusters(sb, desc))
585 return 0;
586
587 /*
588 * We're going to place this inode in a different blockgroup from its
589 * parent. We want to cause files in a common directory to all land in
590 * the same blockgroup. But we want files which are in a different
591 * directory which shares a blockgroup with our parent to land in a
592 * different blockgroup.
593 *
594 * So add our directory's i_ino into the starting point for the hash.
595 */
596 *group = (*group + parent->i_ino) % ngroups;
597
598 /*
599 * Use a quadratic hash to find a group with a free inode and some free
600 * blocks.
601 */
602 for (i = 1; i < ngroups; i <<= 1) {
603 *group += i;
604 if (*group >= ngroups)
605 *group -= ngroups;
606 desc = ext4_get_group_desc(sb, *group, NULL);
607 if (desc && ext4_free_inodes_count(sb, desc) &&
608 ext4_free_group_clusters(sb, desc))
609 return 0;
610 }
611
612 /*
613 * That failed: try linear search for a free inode, even if that group
614 * has no free blocks.
615 */
616 *group = parent_group;
617 for (i = 0; i < ngroups; i++) {
618 if (++*group >= ngroups)
619 *group = 0;
620 desc = ext4_get_group_desc(sb, *group, NULL);
621 if (desc && ext4_free_inodes_count(sb, desc))
622 return 0;
623 }
624
625 return -1;
626}
627
628/*
629 * There are two policies for allocating an inode. If the new inode is
630 * a directory, then a forward search is made for a block group with both
631 * free space and a low directory-to-inode ratio; if that fails, then of
632 * the groups with above-average free space, that group with the fewest
633 * directories already is chosen.
634 *
635 * For other inodes, search forward from the parent directory's block
636 * group to find a free inode.
637 */
638struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, umode_t mode,
639 const struct qstr *qstr, __u32 goal, uid_t *owner)
640{
641 struct super_block *sb;
642 struct buffer_head *inode_bitmap_bh = NULL;
643 struct buffer_head *group_desc_bh;
644 ext4_group_t ngroups, group = 0;
645 unsigned long ino = 0;
646 struct inode *inode;
647 struct ext4_group_desc *gdp = NULL;
648 struct ext4_inode_info *ei;
649 struct ext4_sb_info *sbi;
650 int ret2, err = 0;
651 struct inode *ret;
652 ext4_group_t i;
653 ext4_group_t flex_group;
654
655 /* Cannot create files in a deleted directory */
656 if (!dir || !dir->i_nlink)
657 return ERR_PTR(-EPERM);
658
659 sb = dir->i_sb;
660 ngroups = ext4_get_groups_count(sb);
661 trace_ext4_request_inode(dir, mode);
662 inode = new_inode(sb);
663 if (!inode)
664 return ERR_PTR(-ENOMEM);
665 ei = EXT4_I(inode);
666 sbi = EXT4_SB(sb);
667
668 if (!goal)
669 goal = sbi->s_inode_goal;
670
671 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
672 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
673 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
674 ret2 = 0;
675 goto got_group;
676 }
677
678 if (S_ISDIR(mode))
679 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
680 else
681 ret2 = find_group_other(sb, dir, &group, mode);
682
683got_group:
684 EXT4_I(dir)->i_last_alloc_group = group;
685 err = -ENOSPC;
686 if (ret2 == -1)
687 goto out;
688
689 /*
690 * Normally we will only go through one pass of this loop,
691 * unless we get unlucky and it turns out the group we selected
692 * had its last inode grabbed by someone else.
693 */
694 for (i = 0; i < ngroups; i++, ino = 0) {
695 err = -EIO;
696
697 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
698 if (!gdp)
699 goto fail;
700
701 brelse(inode_bitmap_bh);
702 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
703 if (!inode_bitmap_bh)
704 goto fail;
705
706repeat_in_this_group:
707 ino = ext4_find_next_zero_bit((unsigned long *)
708 inode_bitmap_bh->b_data,
709 EXT4_INODES_PER_GROUP(sb), ino);
710 if (ino >= EXT4_INODES_PER_GROUP(sb)) {
711 if (++group == ngroups)
712 group = 0;
713 continue;
714 }
715 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
716 ext4_error(sb, "reserved inode found cleared - "
717 "inode=%lu", ino + 1);
718 continue;
719 }
720 ext4_lock_group(sb, group);
721 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
722 ext4_unlock_group(sb, group);
723 ino++; /* the inode bitmap is zero-based */
724 if (!ret2)
725 goto got; /* we grabbed the inode! */
726 if (ino < EXT4_INODES_PER_GROUP(sb))
727 goto repeat_in_this_group;
728 }
729 err = -ENOSPC;
730 goto out;
731
732got:
733 /* We may have to initialize the block bitmap if it isn't already */
734 if (ext4_has_group_desc_csum(sb) &&
735 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
736 struct buffer_head *block_bitmap_bh;
737
738 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
739 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
740 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
741 if (err) {
742 brelse(block_bitmap_bh);
743 goto fail;
744 }
745
746 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
747 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
748 brelse(block_bitmap_bh);
749
750 /* recheck and clear flag under lock if we still need to */
751 ext4_lock_group(sb, group);
752 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
753 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
754 ext4_free_group_clusters_set(sb, gdp,
755 ext4_free_clusters_after_init(sb, group, gdp));
756 ext4_block_bitmap_csum_set(sb, group, gdp,
757 block_bitmap_bh,
758 EXT4_BLOCKS_PER_GROUP(sb) /
759 8);
760 ext4_group_desc_csum_set(sb, group, gdp);
761 }
762 ext4_unlock_group(sb, group);
763
764 if (err)
765 goto fail;
766 }
767
768 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
769 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
770 if (err)
771 goto fail;
772
773 BUFFER_TRACE(group_desc_bh, "get_write_access");
774 err = ext4_journal_get_write_access(handle, group_desc_bh);
775 if (err)
776 goto fail;
777
778 /* Update the relevant bg descriptor fields */
779 if (ext4_has_group_desc_csum(sb)) {
780 int free;
781 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
782
783 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
784 ext4_lock_group(sb, group); /* while we modify the bg desc */
785 free = EXT4_INODES_PER_GROUP(sb) -
786 ext4_itable_unused_count(sb, gdp);
787 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
788 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
789 free = 0;
790 }
791 /*
792 * Check the relative inode number against the last used
793 * relative inode number in this group. if it is greater
794 * we need to update the bg_itable_unused count
795 */
796 if (ino > free)
797 ext4_itable_unused_set(sb, gdp,
798 (EXT4_INODES_PER_GROUP(sb) - ino));
799 up_read(&grp->alloc_sem);
800 } else {
801 ext4_lock_group(sb, group);
802 }
803
804 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
805 if (S_ISDIR(mode)) {
806 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
807 if (sbi->s_log_groups_per_flex) {
808 ext4_group_t f = ext4_flex_group(sbi, group);
809
810 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
811 }
812 }
813 if (ext4_has_group_desc_csum(sb)) {
814 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
815 EXT4_INODES_PER_GROUP(sb) / 8);
816 ext4_group_desc_csum_set(sb, group, gdp);
817 }
818 ext4_unlock_group(sb, group);
819
820 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
821 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
822 if (err)
823 goto fail;
824
825 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
826 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
827 if (err)
828 goto fail;
829
830 percpu_counter_dec(&sbi->s_freeinodes_counter);
831 if (S_ISDIR(mode))
832 percpu_counter_inc(&sbi->s_dirs_counter);
833 ext4_mark_super_dirty(sb);
834
835 if (sbi->s_log_groups_per_flex) {
836 flex_group = ext4_flex_group(sbi, group);
837 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
838 }
839 if (owner) {
840 inode->i_mode = mode;
841 i_uid_write(inode, owner[0]);
842 i_gid_write(inode, owner[1]);
843 } else if (test_opt(sb, GRPID)) {
844 inode->i_mode = mode;
845 inode->i_uid = current_fsuid();
846 inode->i_gid = dir->i_gid;
847 } else
848 inode_init_owner(inode, dir, mode);
849
850 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
851 /* This is the optimal IO size (for stat), not the fs block size */
852 inode->i_blocks = 0;
853 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
854 ext4_current_time(inode);
855
856 memset(ei->i_data, 0, sizeof(ei->i_data));
857 ei->i_dir_start_lookup = 0;
858 ei->i_disksize = 0;
859
860 /* Don't inherit extent flag from directory, amongst others. */
861 ei->i_flags =
862 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
863 ei->i_file_acl = 0;
864 ei->i_dtime = 0;
865 ei->i_block_group = group;
866 ei->i_last_alloc_group = ~0;
867
868 ext4_set_inode_flags(inode);
869 if (IS_DIRSYNC(inode))
870 ext4_handle_sync(handle);
871 if (insert_inode_locked(inode) < 0) {
872 /*
873 * Likely a bitmap corruption causing inode to be allocated
874 * twice.
875 */
876 err = -EIO;
877 goto fail;
878 }
879 spin_lock(&sbi->s_next_gen_lock);
880 inode->i_generation = sbi->s_next_generation++;
881 spin_unlock(&sbi->s_next_gen_lock);
882
883 /* Precompute checksum seed for inode metadata */
884 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
885 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
886 __u32 csum;
887 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
888 __le32 inum = cpu_to_le32(inode->i_ino);
889 __le32 gen = cpu_to_le32(inode->i_generation);
890 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
891 sizeof(inum));
892 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
893 sizeof(gen));
894 }
895
896 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
897 ext4_set_inode_state(inode, EXT4_STATE_NEW);
898
899 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
900
901 ret = inode;
902 dquot_initialize(inode);
903 err = dquot_alloc_inode(inode);
904 if (err)
905 goto fail_drop;
906
907 err = ext4_init_acl(handle, inode, dir);
908 if (err)
909 goto fail_free_drop;
910
911 err = ext4_init_security(handle, inode, dir, qstr);
912 if (err)
913 goto fail_free_drop;
914
915 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
916 /* set extent flag only for directory, file and normal symlink*/
917 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
918 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
919 ext4_ext_tree_init(handle, inode);
920 }
921 }
922
923 if (ext4_handle_valid(handle)) {
924 ei->i_sync_tid = handle->h_transaction->t_tid;
925 ei->i_datasync_tid = handle->h_transaction->t_tid;
926 }
927
928 err = ext4_mark_inode_dirty(handle, inode);
929 if (err) {
930 ext4_std_error(sb, err);
931 goto fail_free_drop;
932 }
933
934 ext4_debug("allocating inode %lu\n", inode->i_ino);
935 trace_ext4_allocate_inode(inode, dir, mode);
936 goto really_out;
937fail:
938 ext4_std_error(sb, err);
939out:
940 iput(inode);
941 ret = ERR_PTR(err);
942really_out:
943 brelse(inode_bitmap_bh);
944 return ret;
945
946fail_free_drop:
947 dquot_free_inode(inode);
948
949fail_drop:
950 dquot_drop(inode);
951 inode->i_flags |= S_NOQUOTA;
952 clear_nlink(inode);
953 unlock_new_inode(inode);
954 iput(inode);
955 brelse(inode_bitmap_bh);
956 return ERR_PTR(err);
957}
958
959/* Verify that we are loading a valid orphan from disk */
960struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
961{
962 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
963 ext4_group_t block_group;
964 int bit;
965 struct buffer_head *bitmap_bh;
966 struct inode *inode = NULL;
967 long err = -EIO;
968
969 /* Error cases - e2fsck has already cleaned up for us */
970 if (ino > max_ino) {
971 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
972 goto error;
973 }
974
975 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
976 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
977 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
978 if (!bitmap_bh) {
979 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
980 goto error;
981 }
982
983 /* Having the inode bit set should be a 100% indicator that this
984 * is a valid orphan (no e2fsck run on fs). Orphans also include
985 * inodes that were being truncated, so we can't check i_nlink==0.
986 */
987 if (!ext4_test_bit(bit, bitmap_bh->b_data))
988 goto bad_orphan;
989
990 inode = ext4_iget(sb, ino);
991 if (IS_ERR(inode))
992 goto iget_failed;
993
994 /*
995 * If the orphans has i_nlinks > 0 then it should be able to be
996 * truncated, otherwise it won't be removed from the orphan list
997 * during processing and an infinite loop will result.
998 */
999 if (inode->i_nlink && !ext4_can_truncate(inode))
1000 goto bad_orphan;
1001
1002 if (NEXT_ORPHAN(inode) > max_ino)
1003 goto bad_orphan;
1004 brelse(bitmap_bh);
1005 return inode;
1006
1007iget_failed:
1008 err = PTR_ERR(inode);
1009 inode = NULL;
1010bad_orphan:
1011 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1012 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1013 bit, (unsigned long long)bitmap_bh->b_blocknr,
1014 ext4_test_bit(bit, bitmap_bh->b_data));
1015 printk(KERN_NOTICE "inode=%p\n", inode);
1016 if (inode) {
1017 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1018 is_bad_inode(inode));
1019 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1020 NEXT_ORPHAN(inode));
1021 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1022 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1023 /* Avoid freeing blocks if we got a bad deleted inode */
1024 if (inode->i_nlink == 0)
1025 inode->i_blocks = 0;
1026 iput(inode);
1027 }
1028 brelse(bitmap_bh);
1029error:
1030 return ERR_PTR(err);
1031}
1032
1033unsigned long ext4_count_free_inodes(struct super_block *sb)
1034{
1035 unsigned long desc_count;
1036 struct ext4_group_desc *gdp;
1037 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1038#ifdef EXT4FS_DEBUG
1039 struct ext4_super_block *es;
1040 unsigned long bitmap_count, x;
1041 struct buffer_head *bitmap_bh = NULL;
1042
1043 es = EXT4_SB(sb)->s_es;
1044 desc_count = 0;
1045 bitmap_count = 0;
1046 gdp = NULL;
1047 for (i = 0; i < ngroups; i++) {
1048 gdp = ext4_get_group_desc(sb, i, NULL);
1049 if (!gdp)
1050 continue;
1051 desc_count += ext4_free_inodes_count(sb, gdp);
1052 brelse(bitmap_bh);
1053 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1054 if (!bitmap_bh)
1055 continue;
1056
1057 x = ext4_count_free(bitmap_bh->b_data,
1058 EXT4_INODES_PER_GROUP(sb) / 8);
1059 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1060 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1061 bitmap_count += x;
1062 }
1063 brelse(bitmap_bh);
1064 printk(KERN_DEBUG "ext4_count_free_inodes: "
1065 "stored = %u, computed = %lu, %lu\n",
1066 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1067 return desc_count;
1068#else
1069 desc_count = 0;
1070 for (i = 0; i < ngroups; i++) {
1071 gdp = ext4_get_group_desc(sb, i, NULL);
1072 if (!gdp)
1073 continue;
1074 desc_count += ext4_free_inodes_count(sb, gdp);
1075 cond_resched();
1076 }
1077 return desc_count;
1078#endif
1079}
1080
1081/* Called at mount-time, super-block is locked */
1082unsigned long ext4_count_dirs(struct super_block * sb)
1083{
1084 unsigned long count = 0;
1085 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1086
1087 for (i = 0; i < ngroups; i++) {
1088 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1089 if (!gdp)
1090 continue;
1091 count += ext4_used_dirs_count(sb, gdp);
1092 }
1093 return count;
1094}
1095
1096/*
1097 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1098 * inode table. Must be called without any spinlock held. The only place
1099 * where it is called from on active part of filesystem is ext4lazyinit
1100 * thread, so we do not need any special locks, however we have to prevent
1101 * inode allocation from the current group, so we take alloc_sem lock, to
1102 * block ext4_new_inode() until we are finished.
1103 */
1104int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1105 int barrier)
1106{
1107 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1108 struct ext4_sb_info *sbi = EXT4_SB(sb);
1109 struct ext4_group_desc *gdp = NULL;
1110 struct buffer_head *group_desc_bh;
1111 handle_t *handle;
1112 ext4_fsblk_t blk;
1113 int num, ret = 0, used_blks = 0;
1114
1115 /* This should not happen, but just to be sure check this */
1116 if (sb->s_flags & MS_RDONLY) {
1117 ret = 1;
1118 goto out;
1119 }
1120
1121 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1122 if (!gdp)
1123 goto out;
1124
1125 /*
1126 * We do not need to lock this, because we are the only one
1127 * handling this flag.
1128 */
1129 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1130 goto out;
1131
1132 handle = ext4_journal_start_sb(sb, 1);
1133 if (IS_ERR(handle)) {
1134 ret = PTR_ERR(handle);
1135 goto out;
1136 }
1137
1138 down_write(&grp->alloc_sem);
1139 /*
1140 * If inode bitmap was already initialized there may be some
1141 * used inodes so we need to skip blocks with used inodes in
1142 * inode table.
1143 */
1144 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1145 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1146 ext4_itable_unused_count(sb, gdp)),
1147 sbi->s_inodes_per_block);
1148
1149 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1150 ext4_error(sb, "Something is wrong with group %u: "
1151 "used itable blocks: %d; "
1152 "itable unused count: %u",
1153 group, used_blks,
1154 ext4_itable_unused_count(sb, gdp));
1155 ret = 1;
1156 goto err_out;
1157 }
1158
1159 blk = ext4_inode_table(sb, gdp) + used_blks;
1160 num = sbi->s_itb_per_group - used_blks;
1161
1162 BUFFER_TRACE(group_desc_bh, "get_write_access");
1163 ret = ext4_journal_get_write_access(handle,
1164 group_desc_bh);
1165 if (ret)
1166 goto err_out;
1167
1168 /*
1169 * Skip zeroout if the inode table is full. But we set the ZEROED
1170 * flag anyway, because obviously, when it is full it does not need
1171 * further zeroing.
1172 */
1173 if (unlikely(num == 0))
1174 goto skip_zeroout;
1175
1176 ext4_debug("going to zero out inode table in group %d\n",
1177 group);
1178 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1179 if (ret < 0)
1180 goto err_out;
1181 if (barrier)
1182 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1183
1184skip_zeroout:
1185 ext4_lock_group(sb, group);
1186 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1187 ext4_group_desc_csum_set(sb, group, gdp);
1188 ext4_unlock_group(sb, group);
1189
1190 BUFFER_TRACE(group_desc_bh,
1191 "call ext4_handle_dirty_metadata");
1192 ret = ext4_handle_dirty_metadata(handle, NULL,
1193 group_desc_bh);
1194
1195err_out:
1196 up_write(&grp->alloc_sem);
1197 ext4_journal_stop(handle);
1198out:
1199 return ret;
1200}