Root/fs/ext4/ialloc.c

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 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 */
68unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
69                ext4_group_t block_group,
70                struct ext4_group_desc *gdp)
71{
72    struct ext4_sb_info *sbi = EXT4_SB(sb);
73
74    J_ASSERT_BH(bh, buffer_locked(bh));
75
76    /* If checksum is bad mark all blocks and inodes use to prevent
77     * allocation, essentially implementing a per-group read-only flag. */
78    if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
79        ext4_error(sb, __func__, "Checksum bad for group %u",
80               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    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    bitmap_blk = ext4_inode_bitmap(sb, desc);
112    bh = sb_getblk(sb, bitmap_blk);
113    if (unlikely(!bh)) {
114        ext4_error(sb, __func__,
115                "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    ext4_lock_group(sb, block_group);
129    if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
130        ext4_init_inode_bitmap(sb, bh, block_group, desc);
131        set_bitmap_uptodate(bh);
132        set_buffer_uptodate(bh);
133        ext4_unlock_group(sb, block_group);
134        unlock_buffer(bh);
135        return bh;
136    }
137    ext4_unlock_group(sb, block_group);
138    if (buffer_uptodate(bh)) {
139        /*
140         * if not uninit if bh is uptodate,
141         * bitmap is also uptodate
142         */
143        set_bitmap_uptodate(bh);
144        unlock_buffer(bh);
145        return bh;
146    }
147    /*
148     * submit the buffer_head for read. We can
149     * safely mark the bitmap as uptodate now.
150     * We do it here so the bitmap uptodate bit
151     * get set with buffer lock held.
152     */
153    set_bitmap_uptodate(bh);
154    if (bh_submit_read(bh) < 0) {
155        put_bh(bh);
156        ext4_error(sb, __func__,
157                "Cannot read inode bitmap - "
158                "block_group = %u, inode_bitmap = %llu",
159                block_group, bitmap_blk);
160        return NULL;
161    }
162    return bh;
163}
164
165/*
166 * NOTE! When we get the inode, we're the only people
167 * that have access to it, and as such there are no
168 * race conditions we have to worry about. The inode
169 * is not on the hash-lists, and it cannot be reached
170 * through the filesystem because the directory entry
171 * has been deleted earlier.
172 *
173 * HOWEVER: we must make sure that we get no aliases,
174 * which means that we have to call "clear_inode()"
175 * _before_ we mark the inode not in use in the inode
176 * bitmaps. Otherwise a newly created file might use
177 * the same inode number (not actually the same pointer
178 * though), and then we'd have two inodes sharing the
179 * same inode number and space on the harddisk.
180 */
181void ext4_free_inode(handle_t *handle, struct inode *inode)
182{
183    struct super_block *sb = inode->i_sb;
184    int is_directory;
185    unsigned long ino;
186    struct buffer_head *bitmap_bh = NULL;
187    struct buffer_head *bh2;
188    ext4_group_t block_group;
189    unsigned long bit;
190    struct ext4_group_desc *gdp;
191    struct ext4_super_block *es;
192    struct ext4_sb_info *sbi;
193    int fatal = 0, err, count, cleared;
194
195    if (atomic_read(&inode->i_count) > 1) {
196        printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
197               atomic_read(&inode->i_count));
198        return;
199    }
200    if (inode->i_nlink) {
201        printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
202               inode->i_nlink);
203        return;
204    }
205    if (!sb) {
206        printk(KERN_ERR "ext4_free_inode: inode on "
207               "nonexistent device\n");
208        return;
209    }
210    sbi = EXT4_SB(sb);
211
212    ino = inode->i_ino;
213    ext4_debug("freeing inode %lu\n", ino);
214    trace_ext4_free_inode(inode);
215
216    /*
217     * Note: we must free any quota before locking the superblock,
218     * as writing the quota to disk may need the lock as well.
219     */
220    vfs_dq_init(inode);
221    ext4_xattr_delete_inode(handle, inode);
222    vfs_dq_free_inode(inode);
223    vfs_dq_drop(inode);
224
225    is_directory = S_ISDIR(inode->i_mode);
226
227    /* Do this BEFORE marking the inode not in use or returning an error */
228    clear_inode(inode);
229
230    es = EXT4_SB(sb)->s_es;
231    if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
232        ext4_error(sb, "ext4_free_inode",
233               "reserved or nonexistent inode %lu", ino);
234        goto error_return;
235    }
236    block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
237    bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
238    bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
239    if (!bitmap_bh)
240        goto error_return;
241
242    BUFFER_TRACE(bitmap_bh, "get_write_access");
243    fatal = ext4_journal_get_write_access(handle, bitmap_bh);
244    if (fatal)
245        goto error_return;
246
247    /* Ok, now we can actually update the inode bitmaps.. */
248    cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
249                    bit, bitmap_bh->b_data);
250    if (!cleared)
251        ext4_error(sb, "ext4_free_inode",
252               "bit already cleared for inode %lu", ino);
253    else {
254        gdp = ext4_get_group_desc(sb, block_group, &bh2);
255
256        BUFFER_TRACE(bh2, "get_write_access");
257        fatal = ext4_journal_get_write_access(handle, bh2);
258        if (fatal) goto error_return;
259
260        if (gdp) {
261            ext4_lock_group(sb, block_group);
262            count = ext4_free_inodes_count(sb, gdp) + 1;
263            ext4_free_inodes_set(sb, gdp, count);
264            if (is_directory) {
265                count = ext4_used_dirs_count(sb, gdp) - 1;
266                ext4_used_dirs_set(sb, gdp, count);
267                if (sbi->s_log_groups_per_flex) {
268                    ext4_group_t f;
269
270                    f = ext4_flex_group(sbi, block_group);
271                    atomic_dec(&sbi->s_flex_groups[f].free_inodes);
272                }
273
274            }
275            gdp->bg_checksum = ext4_group_desc_csum(sbi,
276                            block_group, gdp);
277            ext4_unlock_group(sb, block_group);
278            percpu_counter_inc(&sbi->s_freeinodes_counter);
279            if (is_directory)
280                percpu_counter_dec(&sbi->s_dirs_counter);
281
282            if (sbi->s_log_groups_per_flex) {
283                ext4_group_t f;
284
285                f = ext4_flex_group(sbi, block_group);
286                atomic_inc(&sbi->s_flex_groups[f].free_inodes);
287            }
288        }
289        BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
290        err = ext4_handle_dirty_metadata(handle, NULL, bh2);
291        if (!fatal) fatal = err;
292    }
293    BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
294    err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
295    if (!fatal)
296        fatal = err;
297    sb->s_dirt = 1;
298error_return:
299    brelse(bitmap_bh);
300    ext4_std_error(sb, fatal);
301}
302
303/*
304 * There are two policies for allocating an inode. If the new inode is
305 * a directory, then a forward search is made for a block group with both
306 * free space and a low directory-to-inode ratio; if that fails, then of
307 * the groups with above-average free space, that group with the fewest
308 * directories already is chosen.
309 *
310 * For other inodes, search forward from the parent directory\'s block
311 * group to find a free inode.
312 */
313static int find_group_dir(struct super_block *sb, struct inode *parent,
314                ext4_group_t *best_group)
315{
316    ext4_group_t ngroups = ext4_get_groups_count(sb);
317    unsigned int freei, avefreei;
318    struct ext4_group_desc *desc, *best_desc = NULL;
319    ext4_group_t group;
320    int ret = -1;
321
322    freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
323    avefreei = freei / ngroups;
324
325    for (group = 0; group < ngroups; group++) {
326        desc = ext4_get_group_desc(sb, group, NULL);
327        if (!desc || !ext4_free_inodes_count(sb, desc))
328            continue;
329        if (ext4_free_inodes_count(sb, desc) < avefreei)
330            continue;
331        if (!best_desc ||
332            (ext4_free_blks_count(sb, desc) >
333             ext4_free_blks_count(sb, best_desc))) {
334            *best_group = group;
335            best_desc = desc;
336            ret = 0;
337        }
338    }
339    return ret;
340}
341
342#define free_block_ratio 10
343
344static int find_group_flex(struct super_block *sb, struct inode *parent,
345               ext4_group_t *best_group)
346{
347    struct ext4_sb_info *sbi = EXT4_SB(sb);
348    struct ext4_group_desc *desc;
349    struct flex_groups *flex_group = sbi->s_flex_groups;
350    ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
351    ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
352    ext4_group_t ngroups = ext4_get_groups_count(sb);
353    int flex_size = ext4_flex_bg_size(sbi);
354    ext4_group_t best_flex = parent_fbg_group;
355    int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
356    int flexbg_free_blocks;
357    int flex_freeb_ratio;
358    ext4_group_t n_fbg_groups;
359    ext4_group_t i;
360
361    n_fbg_groups = (ngroups + flex_size - 1) >>
362        sbi->s_log_groups_per_flex;
363
364find_close_to_parent:
365    flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
366    flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
367    if (atomic_read(&flex_group[best_flex].free_inodes) &&
368        flex_freeb_ratio > free_block_ratio)
369        goto found_flexbg;
370
371    if (best_flex && best_flex == parent_fbg_group) {
372        best_flex--;
373        goto find_close_to_parent;
374    }
375
376    for (i = 0; i < n_fbg_groups; i++) {
377        if (i == parent_fbg_group || i == parent_fbg_group - 1)
378            continue;
379
380        flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
381        flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
382
383        if (flex_freeb_ratio > free_block_ratio &&
384            (atomic_read(&flex_group[i].free_inodes))) {
385            best_flex = i;
386            goto found_flexbg;
387        }
388
389        if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
390            ((atomic_read(&flex_group[i].free_blocks) >
391              atomic_read(&flex_group[best_flex].free_blocks)) &&
392             atomic_read(&flex_group[i].free_inodes)))
393            best_flex = i;
394    }
395
396    if (!atomic_read(&flex_group[best_flex].free_inodes) ||
397        !atomic_read(&flex_group[best_flex].free_blocks))
398        return -1;
399
400found_flexbg:
401    for (i = best_flex * flex_size; i < ngroups &&
402             i < (best_flex + 1) * flex_size; i++) {
403        desc = ext4_get_group_desc(sb, i, NULL);
404        if (ext4_free_inodes_count(sb, desc)) {
405            *best_group = i;
406            goto out;
407        }
408    }
409
410    return -1;
411out:
412    return 0;
413}
414
415struct orlov_stats {
416    __u32 free_inodes;
417    __u32 free_blocks;
418    __u32 used_dirs;
419};
420
421/*
422 * Helper function for Orlov's allocator; returns critical information
423 * for a particular block group or flex_bg. If flex_size is 1, then g
424 * is a block group number; otherwise it is flex_bg number.
425 */
426void get_orlov_stats(struct super_block *sb, ext4_group_t g,
427               int flex_size, struct orlov_stats *stats)
428{
429    struct ext4_group_desc *desc;
430    struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
431
432    if (flex_size > 1) {
433        stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
434        stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
435        stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
436        return;
437    }
438
439    desc = ext4_get_group_desc(sb, g, NULL);
440    if (desc) {
441        stats->free_inodes = ext4_free_inodes_count(sb, desc);
442        stats->free_blocks = ext4_free_blks_count(sb, desc);
443        stats->used_dirs = ext4_used_dirs_count(sb, desc);
444    } else {
445        stats->free_inodes = 0;
446        stats->free_blocks = 0;
447        stats->used_dirs = 0;
448    }
449}
450
451/*
452 * Orlov's allocator for directories.
453 *
454 * We always try to spread first-level directories.
455 *
456 * If there are blockgroups with both free inodes and free blocks counts
457 * not worse than average we return one with smallest directory count.
458 * Otherwise we simply return a random group.
459 *
460 * For the rest rules look so:
461 *
462 * It's OK to put directory into a group unless
463 * it has too many directories already (max_dirs) or
464 * it has too few free inodes left (min_inodes) or
465 * it has too few free blocks left (min_blocks) or
466 * Parent's group is preferred, if it doesn't satisfy these
467 * conditions we search cyclically through the rest. If none
468 * of the groups look good we just look for a group with more
469 * free inodes than average (starting at parent's group).
470 */
471
472static int find_group_orlov(struct super_block *sb, struct inode *parent,
473                ext4_group_t *group, int mode,
474                const struct qstr *qstr)
475{
476    ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
477    struct ext4_sb_info *sbi = EXT4_SB(sb);
478    ext4_group_t real_ngroups = ext4_get_groups_count(sb);
479    int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
480    unsigned int freei, avefreei;
481    ext4_fsblk_t freeb, avefreeb;
482    unsigned int ndirs;
483    int max_dirs, min_inodes;
484    ext4_grpblk_t min_blocks;
485    ext4_group_t i, grp, g, ngroups;
486    struct ext4_group_desc *desc;
487    struct orlov_stats stats;
488    int flex_size = ext4_flex_bg_size(sbi);
489    struct dx_hash_info hinfo;
490
491    ngroups = real_ngroups;
492    if (flex_size > 1) {
493        ngroups = (real_ngroups + flex_size - 1) >>
494            sbi->s_log_groups_per_flex;
495        parent_group >>= sbi->s_log_groups_per_flex;
496    }
497
498    freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
499    avefreei = freei / ngroups;
500    freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
501    avefreeb = freeb;
502    do_div(avefreeb, ngroups);
503    ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
504
505    if (S_ISDIR(mode) &&
506        ((parent == sb->s_root->d_inode) ||
507         (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
508        int best_ndir = inodes_per_group;
509        int ret = -1;
510
511        if (qstr) {
512            hinfo.hash_version = DX_HASH_HALF_MD4;
513            hinfo.seed = sbi->s_hash_seed;
514            ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
515            grp = hinfo.hash;
516        } else
517            get_random_bytes(&grp, sizeof(grp));
518        parent_group = (unsigned)grp % ngroups;
519        for (i = 0; i < ngroups; i++) {
520            g = (parent_group + i) % ngroups;
521            get_orlov_stats(sb, g, flex_size, &stats);
522            if (!stats.free_inodes)
523                continue;
524            if (stats.used_dirs >= best_ndir)
525                continue;
526            if (stats.free_inodes < avefreei)
527                continue;
528            if (stats.free_blocks < avefreeb)
529                continue;
530            grp = g;
531            ret = 0;
532            best_ndir = stats.used_dirs;
533        }
534        if (ret)
535            goto fallback;
536    found_flex_bg:
537        if (flex_size == 1) {
538            *group = grp;
539            return 0;
540        }
541
542        /*
543         * We pack inodes at the beginning of the flexgroup's
544         * inode tables. Block allocation decisions will do
545         * something similar, although regular files will
546         * start at 2nd block group of the flexgroup. See
547         * ext4_ext_find_goal() and ext4_find_near().
548         */
549        grp *= flex_size;
550        for (i = 0; i < flex_size; i++) {
551            if (grp+i >= real_ngroups)
552                break;
553            desc = ext4_get_group_desc(sb, grp+i, NULL);
554            if (desc && ext4_free_inodes_count(sb, desc)) {
555                *group = grp+i;
556                return 0;
557            }
558        }
559        goto fallback;
560    }
561
562    max_dirs = ndirs / ngroups + inodes_per_group / 16;
563    min_inodes = avefreei - inodes_per_group*flex_size / 4;
564    if (min_inodes < 1)
565        min_inodes = 1;
566    min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
567
568    /*
569     * Start looking in the flex group where we last allocated an
570     * inode for this parent directory
571     */
572    if (EXT4_I(parent)->i_last_alloc_group != ~0) {
573        parent_group = EXT4_I(parent)->i_last_alloc_group;
574        if (flex_size > 1)
575            parent_group >>= sbi->s_log_groups_per_flex;
576    }
577
578    for (i = 0; i < ngroups; i++) {
579        grp = (parent_group + i) % ngroups;
580        get_orlov_stats(sb, grp, flex_size, &stats);
581        if (stats.used_dirs >= max_dirs)
582            continue;
583        if (stats.free_inodes < min_inodes)
584            continue;
585        if (stats.free_blocks < min_blocks)
586            continue;
587        goto found_flex_bg;
588    }
589
590fallback:
591    ngroups = real_ngroups;
592    avefreei = freei / ngroups;
593fallback_retry:
594    parent_group = EXT4_I(parent)->i_block_group;
595    for (i = 0; i < ngroups; i++) {
596        grp = (parent_group + i) % ngroups;
597        desc = ext4_get_group_desc(sb, grp, NULL);
598        if (desc && ext4_free_inodes_count(sb, desc) &&
599            ext4_free_inodes_count(sb, desc) >= avefreei) {
600            *group = grp;
601            return 0;
602        }
603    }
604
605    if (avefreei) {
606        /*
607         * The free-inodes counter is approximate, and for really small
608         * filesystems the above test can fail to find any blockgroups
609         */
610        avefreei = 0;
611        goto fallback_retry;
612    }
613
614    return -1;
615}
616
617static int find_group_other(struct super_block *sb, struct inode *parent,
618                ext4_group_t *group, int mode)
619{
620    ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
621    ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
622    struct ext4_group_desc *desc;
623    int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
624
625    /*
626     * Try to place the inode is the same flex group as its
627     * parent. If we can't find space, use the Orlov algorithm to
628     * find another flex group, and store that information in the
629     * parent directory's inode information so that use that flex
630     * group for future allocations.
631     */
632    if (flex_size > 1) {
633        int retry = 0;
634
635    try_again:
636        parent_group &= ~(flex_size-1);
637        last = parent_group + flex_size;
638        if (last > ngroups)
639            last = ngroups;
640        for (i = parent_group; i < last; i++) {
641            desc = ext4_get_group_desc(sb, i, NULL);
642            if (desc && ext4_free_inodes_count(sb, desc)) {
643                *group = i;
644                return 0;
645            }
646        }
647        if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
648            retry = 1;
649            parent_group = EXT4_I(parent)->i_last_alloc_group;
650            goto try_again;
651        }
652        /*
653         * If this didn't work, use the Orlov search algorithm
654         * to find a new flex group; we pass in the mode to
655         * avoid the topdir algorithms.
656         */
657        *group = parent_group + flex_size;
658        if (*group > ngroups)
659            *group = 0;
660        return find_group_orlov(sb, parent, group, mode, 0);
661    }
662
663    /*
664     * Try to place the inode in its parent directory
665     */
666    *group = parent_group;
667    desc = ext4_get_group_desc(sb, *group, NULL);
668    if (desc && ext4_free_inodes_count(sb, desc) &&
669            ext4_free_blks_count(sb, desc))
670        return 0;
671
672    /*
673     * We're going to place this inode in a different blockgroup from its
674     * parent. We want to cause files in a common directory to all land in
675     * the same blockgroup. But we want files which are in a different
676     * directory which shares a blockgroup with our parent to land in a
677     * different blockgroup.
678     *
679     * So add our directory's i_ino into the starting point for the hash.
680     */
681    *group = (*group + parent->i_ino) % ngroups;
682
683    /*
684     * Use a quadratic hash to find a group with a free inode and some free
685     * blocks.
686     */
687    for (i = 1; i < ngroups; i <<= 1) {
688        *group += i;
689        if (*group >= ngroups)
690            *group -= ngroups;
691        desc = ext4_get_group_desc(sb, *group, NULL);
692        if (desc && ext4_free_inodes_count(sb, desc) &&
693                ext4_free_blks_count(sb, desc))
694            return 0;
695    }
696
697    /*
698     * That failed: try linear search for a free inode, even if that group
699     * has no free blocks.
700     */
701    *group = parent_group;
702    for (i = 0; i < ngroups; i++) {
703        if (++*group >= ngroups)
704            *group = 0;
705        desc = ext4_get_group_desc(sb, *group, NULL);
706        if (desc && ext4_free_inodes_count(sb, desc))
707            return 0;
708    }
709
710    return -1;
711}
712
713/*
714 * claim the inode from the inode bitmap. If the group
715 * is uninit we need to take the groups's ext4_group_lock
716 * and clear the uninit flag. The inode bitmap update
717 * and group desc uninit flag clear should be done
718 * after holding ext4_group_lock so that ext4_read_inode_bitmap
719 * doesn't race with the ext4_claim_inode
720 */
721static int ext4_claim_inode(struct super_block *sb,
722            struct buffer_head *inode_bitmap_bh,
723            unsigned long ino, ext4_group_t group, int mode)
724{
725    int free = 0, retval = 0, count;
726    struct ext4_sb_info *sbi = EXT4_SB(sb);
727    struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
728
729    ext4_lock_group(sb, group);
730    if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
731        /* not a free inode */
732        retval = 1;
733        goto err_ret;
734    }
735    ino++;
736    if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
737            ino > EXT4_INODES_PER_GROUP(sb)) {
738        ext4_unlock_group(sb, group);
739        ext4_error(sb, __func__,
740               "reserved inode or inode > inodes count - "
741               "block_group = %u, inode=%lu", group,
742               ino + group * EXT4_INODES_PER_GROUP(sb));
743        return 1;
744    }
745    /* If we didn't allocate from within the initialized part of the inode
746     * table then we need to initialize up to this inode. */
747    if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
748
749        if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
750            gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
751            /* When marking the block group with
752             * ~EXT4_BG_INODE_UNINIT we don't want to depend
753             * on the value of bg_itable_unused even though
754             * mke2fs could have initialized the same for us.
755             * Instead we calculated the value below
756             */
757
758            free = 0;
759        } else {
760            free = EXT4_INODES_PER_GROUP(sb) -
761                ext4_itable_unused_count(sb, gdp);
762        }
763
764        /*
765         * Check the relative inode number against the last used
766         * relative inode number in this group. if it is greater
767         * we need to update the bg_itable_unused count
768         *
769         */
770        if (ino > free)
771            ext4_itable_unused_set(sb, gdp,
772                    (EXT4_INODES_PER_GROUP(sb) - ino));
773    }
774    count = ext4_free_inodes_count(sb, gdp) - 1;
775    ext4_free_inodes_set(sb, gdp, count);
776    if (S_ISDIR(mode)) {
777        count = ext4_used_dirs_count(sb, gdp) + 1;
778        ext4_used_dirs_set(sb, gdp, count);
779        if (sbi->s_log_groups_per_flex) {
780            ext4_group_t f = ext4_flex_group(sbi, group);
781
782            atomic_inc(&sbi->s_flex_groups[f].free_inodes);
783        }
784    }
785    gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
786err_ret:
787    ext4_unlock_group(sb, group);
788    return retval;
789}
790
791/*
792 * There are two policies for allocating an inode. If the new inode is
793 * a directory, then a forward search is made for a block group with both
794 * free space and a low directory-to-inode ratio; if that fails, then of
795 * the groups with above-average free space, that group with the fewest
796 * directories already is chosen.
797 *
798 * For other inodes, search forward from the parent directory's block
799 * group to find a free inode.
800 */
801struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
802                 const struct qstr *qstr, __u32 goal)
803{
804    struct super_block *sb;
805    struct buffer_head *inode_bitmap_bh = NULL;
806    struct buffer_head *group_desc_bh;
807    ext4_group_t ngroups, group = 0;
808    unsigned long ino = 0;
809    struct inode *inode;
810    struct ext4_group_desc *gdp = NULL;
811    struct ext4_inode_info *ei;
812    struct ext4_sb_info *sbi;
813    int ret2, err = 0;
814    struct inode *ret;
815    ext4_group_t i;
816    int free = 0;
817    static int once = 1;
818    ext4_group_t flex_group;
819
820    /* Cannot create files in a deleted directory */
821    if (!dir || !dir->i_nlink)
822        return ERR_PTR(-EPERM);
823
824    sb = dir->i_sb;
825    ngroups = ext4_get_groups_count(sb);
826    trace_ext4_request_inode(dir, mode);
827    inode = new_inode(sb);
828    if (!inode)
829        return ERR_PTR(-ENOMEM);
830    ei = EXT4_I(inode);
831    sbi = EXT4_SB(sb);
832
833    if (!goal)
834        goal = sbi->s_inode_goal;
835
836    if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
837        group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
838        ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
839        ret2 = 0;
840        goto got_group;
841    }
842
843    if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
844        ret2 = find_group_flex(sb, dir, &group);
845        if (ret2 == -1) {
846            ret2 = find_group_other(sb, dir, &group, mode);
847            if (ret2 == 0 && once) {
848                once = 0;
849                printk(KERN_NOTICE "ext4: find_group_flex "
850                       "failed, fallback succeeded dir %lu\n",
851                       dir->i_ino);
852            }
853        }
854        goto got_group;
855    }
856
857    if (S_ISDIR(mode)) {
858        if (test_opt(sb, OLDALLOC))
859            ret2 = find_group_dir(sb, dir, &group);
860        else
861            ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
862    } else
863        ret2 = find_group_other(sb, dir, &group, mode);
864
865got_group:
866    EXT4_I(dir)->i_last_alloc_group = group;
867    err = -ENOSPC;
868    if (ret2 == -1)
869        goto out;
870
871    for (i = 0; i < ngroups; i++, ino = 0) {
872        err = -EIO;
873
874        gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
875        if (!gdp)
876            goto fail;
877
878        brelse(inode_bitmap_bh);
879        inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
880        if (!inode_bitmap_bh)
881            goto fail;
882
883repeat_in_this_group:
884        ino = ext4_find_next_zero_bit((unsigned long *)
885                          inode_bitmap_bh->b_data,
886                          EXT4_INODES_PER_GROUP(sb), ino);
887
888        if (ino < EXT4_INODES_PER_GROUP(sb)) {
889
890            BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
891            err = ext4_journal_get_write_access(handle,
892                                inode_bitmap_bh);
893            if (err)
894                goto fail;
895
896            BUFFER_TRACE(group_desc_bh, "get_write_access");
897            err = ext4_journal_get_write_access(handle,
898                                group_desc_bh);
899            if (err)
900                goto fail;
901            if (!ext4_claim_inode(sb, inode_bitmap_bh,
902                        ino, group, mode)) {
903                /* we won it */
904                BUFFER_TRACE(inode_bitmap_bh,
905                    "call ext4_handle_dirty_metadata");
906                err = ext4_handle_dirty_metadata(handle,
907                                 inode,
908                            inode_bitmap_bh);
909                if (err)
910                    goto fail;
911                /* zero bit is inode number 1*/
912                ino++;
913                goto got;
914            }
915            /* we lost it */
916            ext4_handle_release_buffer(handle, inode_bitmap_bh);
917            ext4_handle_release_buffer(handle, group_desc_bh);
918
919            if (++ino < EXT4_INODES_PER_GROUP(sb))
920                goto repeat_in_this_group;
921        }
922
923        /*
924         * This case is possible in concurrent environment. It is very
925         * rare. We cannot repeat the find_group_xxx() call because
926         * that will simply return the same blockgroup, because the
927         * group descriptor metadata has not yet been updated.
928         * So we just go onto the next blockgroup.
929         */
930        if (++group == ngroups)
931            group = 0;
932    }
933    err = -ENOSPC;
934    goto out;
935
936got:
937    /* We may have to initialize the block bitmap if it isn't already */
938    if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
939        gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
940        struct buffer_head *block_bitmap_bh;
941
942        block_bitmap_bh = ext4_read_block_bitmap(sb, group);
943        BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
944        err = ext4_journal_get_write_access(handle, block_bitmap_bh);
945        if (err) {
946            brelse(block_bitmap_bh);
947            goto fail;
948        }
949
950        free = 0;
951        ext4_lock_group(sb, group);
952        /* recheck and clear flag under lock if we still need to */
953        if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
954            free = ext4_free_blocks_after_init(sb, group, gdp);
955            gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
956            ext4_free_blks_set(sb, gdp, free);
957            gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
958                                gdp);
959        }
960        ext4_unlock_group(sb, group);
961
962        /* Don't need to dirty bitmap block if we didn't change it */
963        if (free) {
964            BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
965            err = ext4_handle_dirty_metadata(handle,
966                            NULL, block_bitmap_bh);
967        }
968
969        brelse(block_bitmap_bh);
970        if (err)
971            goto fail;
972    }
973    BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
974    err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
975    if (err)
976        goto fail;
977
978    percpu_counter_dec(&sbi->s_freeinodes_counter);
979    if (S_ISDIR(mode))
980        percpu_counter_inc(&sbi->s_dirs_counter);
981    sb->s_dirt = 1;
982
983    if (sbi->s_log_groups_per_flex) {
984        flex_group = ext4_flex_group(sbi, group);
985        atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
986    }
987
988    inode->i_uid = current_fsuid();
989    if (test_opt(sb, GRPID))
990        inode->i_gid = dir->i_gid;
991    else if (dir->i_mode & S_ISGID) {
992        inode->i_gid = dir->i_gid;
993        if (S_ISDIR(mode))
994            mode |= S_ISGID;
995    } else
996        inode->i_gid = current_fsgid();
997    inode->i_mode = mode;
998
999    inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1000    /* This is the optimal IO size (for stat), not the fs block size */
1001    inode->i_blocks = 0;
1002    inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1003                               ext4_current_time(inode);
1004
1005    memset(ei->i_data, 0, sizeof(ei->i_data));
1006    ei->i_dir_start_lookup = 0;
1007    ei->i_disksize = 0;
1008
1009    /*
1010     * Don't inherit extent flag from directory, amongst others. We set
1011     * extent flag on newly created directory and file only if -o extent
1012     * mount option is specified
1013     */
1014    ei->i_flags =
1015        ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1016    ei->i_file_acl = 0;
1017    ei->i_dtime = 0;
1018    ei->i_block_group = group;
1019    ei->i_last_alloc_group = ~0;
1020
1021    ext4_set_inode_flags(inode);
1022    if (IS_DIRSYNC(inode))
1023        ext4_handle_sync(handle);
1024    if (insert_inode_locked(inode) < 0) {
1025        err = -EINVAL;
1026        goto fail_drop;
1027    }
1028    spin_lock(&sbi->s_next_gen_lock);
1029    inode->i_generation = sbi->s_next_generation++;
1030    spin_unlock(&sbi->s_next_gen_lock);
1031
1032    ei->i_state = EXT4_STATE_NEW;
1033
1034    ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1035
1036    ret = inode;
1037    if (vfs_dq_alloc_inode(inode)) {
1038        err = -EDQUOT;
1039        goto fail_drop;
1040    }
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);
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_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1054            ext4_ext_tree_init(handle, inode);
1055        }
1056    }
1057
1058    err = ext4_mark_inode_dirty(handle, inode);
1059    if (err) {
1060        ext4_std_error(sb, err);
1061        goto fail_free_drop;
1062    }
1063
1064    ext4_debug("allocating inode %lu\n", inode->i_ino);
1065    trace_ext4_allocate_inode(inode, dir, mode);
1066    goto really_out;
1067fail:
1068    ext4_std_error(sb, err);
1069out:
1070    iput(inode);
1071    ret = ERR_PTR(err);
1072really_out:
1073    brelse(inode_bitmap_bh);
1074    return ret;
1075
1076fail_free_drop:
1077    vfs_dq_free_inode(inode);
1078
1079fail_drop:
1080    vfs_dq_drop(inode);
1081    inode->i_flags |= S_NOQUOTA;
1082    inode->i_nlink = 0;
1083    unlock_new_inode(inode);
1084    iput(inode);
1085    brelse(inode_bitmap_bh);
1086    return ERR_PTR(err);
1087}
1088
1089/* Verify that we are loading a valid orphan from disk */
1090struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1091{
1092    unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1093    ext4_group_t block_group;
1094    int bit;
1095    struct buffer_head *bitmap_bh;
1096    struct inode *inode = NULL;
1097    long err = -EIO;
1098
1099    /* Error cases - e2fsck has already cleaned up for us */
1100    if (ino > max_ino) {
1101        ext4_warning(sb, __func__,
1102                 "bad orphan ino %lu! e2fsck was run?", ino);
1103        goto error;
1104    }
1105
1106    block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1107    bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1108    bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1109    if (!bitmap_bh) {
1110        ext4_warning(sb, __func__,
1111                 "inode bitmap error for orphan %lu", ino);
1112        goto error;
1113    }
1114
1115    /* Having the inode bit set should be a 100% indicator that this
1116     * is a valid orphan (no e2fsck run on fs). Orphans also include
1117     * inodes that were being truncated, so we can't check i_nlink==0.
1118     */
1119    if (!ext4_test_bit(bit, bitmap_bh->b_data))
1120        goto bad_orphan;
1121
1122    inode = ext4_iget(sb, ino);
1123    if (IS_ERR(inode))
1124        goto iget_failed;
1125
1126    /*
1127     * If the orphans has i_nlinks > 0 then it should be able to be
1128     * truncated, otherwise it won't be removed from the orphan list
1129     * during processing and an infinite loop will result.
1130     */
1131    if (inode->i_nlink && !ext4_can_truncate(inode))
1132        goto bad_orphan;
1133
1134    if (NEXT_ORPHAN(inode) > max_ino)
1135        goto bad_orphan;
1136    brelse(bitmap_bh);
1137    return inode;
1138
1139iget_failed:
1140    err = PTR_ERR(inode);
1141    inode = NULL;
1142bad_orphan:
1143    ext4_warning(sb, __func__,
1144             "bad orphan inode %lu! e2fsck was run?", ino);
1145    printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1146           bit, (unsigned long long)bitmap_bh->b_blocknr,
1147           ext4_test_bit(bit, bitmap_bh->b_data));
1148    printk(KERN_NOTICE "inode=%p\n", inode);
1149    if (inode) {
1150        printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1151               is_bad_inode(inode));
1152        printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1153               NEXT_ORPHAN(inode));
1154        printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1155        printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1156        /* Avoid freeing blocks if we got a bad deleted inode */
1157        if (inode->i_nlink == 0)
1158            inode->i_blocks = 0;
1159        iput(inode);
1160    }
1161    brelse(bitmap_bh);
1162error:
1163    return ERR_PTR(err);
1164}
1165
1166unsigned long ext4_count_free_inodes(struct super_block *sb)
1167{
1168    unsigned long desc_count;
1169    struct ext4_group_desc *gdp;
1170    ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1171#ifdef EXT4FS_DEBUG
1172    struct ext4_super_block *es;
1173    unsigned long bitmap_count, x;
1174    struct buffer_head *bitmap_bh = NULL;
1175
1176    es = EXT4_SB(sb)->s_es;
1177    desc_count = 0;
1178    bitmap_count = 0;
1179    gdp = NULL;
1180    for (i = 0; i < ngroups; i++) {
1181        gdp = ext4_get_group_desc(sb, i, NULL);
1182        if (!gdp)
1183            continue;
1184        desc_count += ext4_free_inodes_count(sb, gdp);
1185        brelse(bitmap_bh);
1186        bitmap_bh = ext4_read_inode_bitmap(sb, i);
1187        if (!bitmap_bh)
1188            continue;
1189
1190        x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1191        printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1192            i, ext4_free_inodes_count(sb, gdp), x);
1193        bitmap_count += x;
1194    }
1195    brelse(bitmap_bh);
1196    printk(KERN_DEBUG "ext4_count_free_inodes: "
1197           "stored = %u, computed = %lu, %lu\n",
1198           le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1199    return desc_count;
1200#else
1201    desc_count = 0;
1202    for (i = 0; i < ngroups; i++) {
1203        gdp = ext4_get_group_desc(sb, i, NULL);
1204        if (!gdp)
1205            continue;
1206        desc_count += ext4_free_inodes_count(sb, gdp);
1207        cond_resched();
1208    }
1209    return desc_count;
1210#endif
1211}
1212
1213/* Called at mount-time, super-block is locked */
1214unsigned long ext4_count_dirs(struct super_block * sb)
1215{
1216    unsigned long count = 0;
1217    ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1218
1219    for (i = 0; i < ngroups; i++) {
1220        struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1221        if (!gdp)
1222            continue;
1223        count += ext4_used_dirs_count(sb, gdp);
1224    }
1225    return count;
1226}
1227

Archive Download this file



interactive