ext4fixup.c revision 97fc910ce0e05862888fd1d9e1938feba40f7539
1/*
2 * Copyright (C) 2010 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16#include "ext4_utils.h"
17#include "make_ext4fs.h"
18#include "ext4_extents.h"
19#include "output_file.h"
20#include "backed_block.h"
21#include "allocate.h"
22#include "ext4fixup.h"
23
24#include <sys/types.h>
25#include <sys/stat.h>
26#include <sys/types.h>
27#include <sys/mman.h>
28#include <fcntl.h>
29#include <unistd.h>
30
31#if defined(__APPLE__) && defined(__MACH__)
32#define lseek64 lseek
33#define off64_t off_t
34#endif
35
36/* The inode block count for a file/directory is in units of 512 byte blocks,
37 * _NOT_ the filesystem block size!
38 */
39#define INODE_BLOCK_SIZE 512
40
41#define MAX_EXT4_BLOCK_SIZE 4096
42
43/* The two modes the recurse_dir() can be in */
44#define SANITY_CHECK_PASS 1
45#define MARK_INODE_NUMS   2
46#define UPDATE_INODE_NUMS 3
47
48/* Magic numbers to indicate what state the update process is in */
49#define MAGIC_STATE_MARKING_INUMS  0x7000151515565512ll
50#define MAGIC_STATE_UPDATING_INUMS 0x6121131211735123ll
51#define MAGIC_STATE_UPDATING_SB    0x15e1715151558477ll
52
53/* Internal state variables corresponding to the magic numbers */
54#define STATE_UNSET          0
55#define STATE_MARKING_INUMS  1
56#define STATE_UPDATING_INUMS 2
57#define STATE_UPDATING_SB    3
58
59/* global flags */
60static int verbose = 0;
61static int no_write = 0;
62
63static int new_inodes_per_group = 0;
64
65static int no_write_fixup_state = 0;
66
67static int compute_new_inum(unsigned int old_inum)
68{
69    unsigned int group, offset;
70
71    group = (old_inum - 1) / info.inodes_per_group;
72    offset = (old_inum -1) % info.inodes_per_group;
73
74    return (group * new_inodes_per_group) + offset + 1;
75}
76
77/* Function to read the primary superblock */
78static void read_sb(int fd, struct ext4_super_block *sb)
79{
80    off64_t ret;
81
82    ret = lseek64(fd, 1024, SEEK_SET);
83    if (ret < 0)
84        critical_error_errno("failed to seek to superblock");
85
86    ret = read(fd, sb, sizeof(*sb));
87    if (ret < 0)
88        critical_error_errno("failed to read superblock");
89    if (ret != sizeof(*sb))
90        critical_error("failed to read all of superblock");
91}
92
93/* Function to write a primary or backup superblock at a given offset */
94static void write_sb(int fd, unsigned long long offset, struct ext4_super_block *sb)
95{
96    off64_t ret;
97
98    if (no_write) {
99        return;
100    }
101
102    ret = lseek64(fd, offset, SEEK_SET);
103    if (ret < 0)
104        critical_error_errno("failed to seek to superblock");
105
106    ret = write(fd, sb, sizeof(*sb));
107    if (ret < 0)
108        critical_error_errno("failed to write superblock");
109    if (ret != sizeof(*sb))
110        critical_error("failed to write all of superblock");
111}
112
113static int get_fs_fixup_state(int fd)
114{
115    unsigned long long magic;
116    int ret, len;
117
118    if (no_write) {
119        return no_write_fixup_state;
120    }
121
122    lseek64(fd, 0, SEEK_SET);
123    len = read(fd, &magic, sizeof(magic));
124    if (len != sizeof(magic)) {
125        critical_error("cannot read fixup_state\n");
126    }
127
128    switch (magic) {
129        case MAGIC_STATE_MARKING_INUMS:
130            ret = STATE_MARKING_INUMS;
131            break;
132        case MAGIC_STATE_UPDATING_INUMS:
133            ret = STATE_UPDATING_INUMS;
134            break;
135        case MAGIC_STATE_UPDATING_SB:
136            ret = STATE_UPDATING_SB;
137            break;
138        default:
139            ret = STATE_UNSET;
140    }
141    return ret;
142}
143
144static int set_fs_fixup_state(int fd, int state)
145{
146    unsigned long long magic;
147    struct ext4_super_block sb;
148    int len;
149
150    if (no_write) {
151        no_write_fixup_state = state;
152        return 0;
153    }
154
155    switch (state) {
156        case STATE_MARKING_INUMS:
157            magic = MAGIC_STATE_MARKING_INUMS;
158            break;
159        case STATE_UPDATING_INUMS:
160            magic = MAGIC_STATE_UPDATING_INUMS;
161            break;
162        case STATE_UPDATING_SB:
163            magic = MAGIC_STATE_UPDATING_SB;
164            break;
165        case STATE_UNSET:
166        default:
167            magic = 0ll;
168            break;
169    }
170
171    lseek64(fd, 0, SEEK_SET);
172    len = write(fd, &magic, sizeof(magic));
173    if (len != sizeof(magic)) {
174        critical_error("cannot write fixup_state\n");
175    }
176
177    read_sb(fd, &sb);
178    if (magic) {
179        /* If we are in the process of updating the filesystem, make it unmountable */
180        sb.s_desc_size |= 1;
181    } else {
182        /* we are done, so make the filesystem mountable again */
183        sb.s_desc_size &= ~1;
184    }
185    write_sb(fd, 1024, &sb);
186
187    return 0;
188}
189
190static int read_ext(int fd)
191{
192    off64_t ret;
193    struct ext4_super_block sb;
194    unsigned int i;
195
196    read_sb(fd, &sb);
197
198    ext4_parse_sb(&sb);
199
200    if (info.feat_incompat & EXT4_FEATURE_INCOMPAT_RECOVER) {
201        critical_error("Filesystem needs recovery first, mount and unmount to do that\n");
202    }
203
204    /* Clear the low bit which is set while this tool is in progress.
205     * If the tool crashes, it will still be set when we restart.
206     * The low bit is set to make the filesystem unmountable while
207     * it is being fixed up.  Also allow 0, which means the old ext2
208     * size is in use.
209     */
210    if (((sb.s_desc_size & ~1) != sizeof(struct ext2_group_desc)) &&
211        ((sb.s_desc_size & ~1) != 0))
212        critical_error("error: bg_desc_size != sizeof(struct ext2_group_desc)\n");
213
214    ret = lseek64(fd, info.len, SEEK_SET);
215    if (ret < 0)
216        critical_error_errno("failed to seek to end of input image");
217
218    ret = lseek64(fd, info.block_size * (aux_info.first_data_block + 1), SEEK_SET);
219    if (ret < 0)
220        critical_error_errno("failed to seek to block group descriptors");
221
222    ret = read(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
223    if (ret < 0)
224        critical_error_errno("failed to read block group descriptors");
225    if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
226        critical_error("failed to read all of block group descriptors");
227
228    if (verbose) {
229        printf("Found filesystem with parameters:\n");
230        printf("    Size: %llu\n", info.len);
231        printf("    Block size: %d\n", info.block_size);
232        printf("    Blocks per group: %d\n", info.blocks_per_group);
233        printf("    Inodes per group: %d\n", info.inodes_per_group);
234        printf("    Inode size: %d\n", info.inode_size);
235        printf("    Label: %s\n", info.label);
236        printf("    Blocks: %llu\n", aux_info.len_blocks);
237        printf("    Block groups: %d\n", aux_info.groups);
238        printf("    Reserved block group size: %d\n", info.bg_desc_reserve_blocks);
239        printf("    Used %d/%d inodes and %d/%d blocks\n",
240                aux_info.sb->s_inodes_count - aux_info.sb->s_free_inodes_count,
241                aux_info.sb->s_inodes_count,
242                aux_info.sb->s_blocks_count_lo - aux_info.sb->s_free_blocks_count_lo,
243                aux_info.sb->s_blocks_count_lo);
244    }
245
246    return 0;
247}
248
249static int read_inode(int fd, unsigned int inum, struct ext4_inode *inode)
250{
251    unsigned int bg_num, bg_offset;
252    off64_t inode_offset;
253    int len;
254
255    bg_num = (inum-1) / info.inodes_per_group;
256    bg_offset = (inum-1) % info.inodes_per_group;
257
258    inode_offset = ((unsigned long long)aux_info.bg_desc[bg_num].bg_inode_table * info.block_size) +
259                    (bg_offset * info.inode_size);
260
261    if (lseek64(fd, inode_offset, SEEK_SET) < 0) {
262        critical_error_errno("failed to seek to inode %d\n", inum);
263    }
264
265    len=read(fd, inode, sizeof(*inode));
266    if (len != sizeof(*inode)) {
267        critical_error_errno("failed to read inode %d\n", inum);
268    }
269
270    return 0;
271}
272
273static int read_block(int fd, unsigned long long block_num, void *block)
274{
275    off64_t off;
276    unsigned int len;
277
278    off = block_num * info.block_size;
279
280    if (lseek64(fd, off, SEEK_SET) , 0) {
281        critical_error_errno("failed to seek to block %lld\n", block_num);
282    }
283
284    len=read(fd, block, info.block_size);
285    if (len != info.block_size) {
286        critical_error_errno("failed to read block %lld\n", block_num);
287    }
288
289    return 0;
290}
291
292static int write_block(int fd, unsigned long long block_num, void *block)
293{
294    off64_t off;
295    unsigned int len;
296
297    if (no_write) {
298        return 0;
299    }
300
301    off = block_num * info.block_size;
302
303    if (lseek64(fd, off, SEEK_SET) < 0) {
304        critical_error_errno("failed to seek to block %lld\n", block_num);
305    }
306
307    len=write(fd, block, info.block_size);
308    if (len != info.block_size) {
309        critical_error_errno("failed to write block %lld\n", block_num);
310    }
311
312    return 0;
313}
314
315static int bitmap_get_bit(u8 *bitmap, u32 bit)
316{
317        if (bitmap[bit / 8] & (1 << (bit % 8)))
318                return 1;
319
320        return 0;
321}
322
323static void bitmap_clear_bit(u8 *bitmap, u32 bit)
324{
325        bitmap[bit / 8] &= ~(1 << (bit % 8));
326
327        return;
328}
329
330static void check_inode_bitmap(int fd, unsigned int bg_num)
331{
332    unsigned int inode_bitmap_block_num;
333    unsigned char block[MAX_EXT4_BLOCK_SIZE];
334    int i, bitmap_updated = 0;
335
336    /* Using the bg_num, aux_info.bg_desc[], info.inodes_per_group and
337     * new_inodes_per_group, retrieve the inode bitmap, and make sure
338     * the bits between the old and new size are clear
339     */
340    inode_bitmap_block_num = aux_info.bg_desc[bg_num].bg_inode_bitmap;
341
342    read_block(fd, inode_bitmap_block_num, block);
343
344    for (i = info.inodes_per_group; i < new_inodes_per_group; i++) {
345        if (bitmap_get_bit(block, i)) {
346            bitmap_clear_bit(block, i);
347            bitmap_updated = 1;
348        }
349    }
350
351    if (bitmap_updated) {
352        if (verbose) {
353            printf("Warning: updated inode bitmap for block group %d\n", bg_num);
354        }
355        write_block(fd, inode_bitmap_block_num, block);
356    }
357
358    return;
359}
360
361/* Update the superblock and bgdesc of the specified block group */
362static int update_superblocks_and_bg_desc(int fd, int state)
363{
364    off64_t ret;
365    struct ext4_super_block sb;
366    unsigned int num_block_groups, total_new_inodes;
367    unsigned int i;
368
369
370    read_sb(fd, &sb);
371
372    /* Compute how many more inodes are now available */
373    num_block_groups = DIV_ROUND_UP(aux_info.len_blocks, info.blocks_per_group);
374    total_new_inodes = num_block_groups * (new_inodes_per_group - sb.s_inodes_per_group);
375
376    if (verbose) {
377        printf("created %d additional inodes\n", total_new_inodes);
378    }
379
380    /* Update the free inodes count in each block group descriptor */
381    for (i = 0; i < num_block_groups; i++) {
382       if (state == STATE_UPDATING_SB) {
383           aux_info.bg_desc[i].bg_free_inodes_count += (new_inodes_per_group - sb.s_inodes_per_group);
384       }
385       check_inode_bitmap(fd, i);
386    }
387
388    /* First some sanity checks */
389    if ((sb.s_inodes_count + total_new_inodes) != (new_inodes_per_group * num_block_groups)) {
390        critical_error("Failed sanity check on new inode count\n");
391    }
392    if (new_inodes_per_group % (info.block_size/info.inode_size)) {
393        critical_error("Failed sanity check on new inode per group alignment\n");
394    }
395
396    /* Update the free inodes count in the superblock */
397    sb.s_inodes_count += total_new_inodes;
398    sb.s_free_inodes_count += total_new_inodes;
399    sb.s_inodes_per_group = new_inodes_per_group;
400
401    for (i = 0; i < aux_info.groups; i++) {
402        if (ext4_bg_has_super_block(i)) {
403            unsigned int sb_offset;
404
405            if (i == 0) {
406              /* The first superblock is offset by 1K to leave room for boot sectors */
407              sb_offset = 1024;
408            } else {
409              sb_offset = 0;
410            }
411
412            sb.s_block_group_nr = i;
413            /* Don't write out the backup superblocks with the bit set in the s_desc_size
414             * which prevents the filesystem from mounting.  The bit for the primary
415             * superblock will be cleared on the final call to set_fs_fixup_state() */
416            if (i != 0) {
417                sb.s_desc_size &= ~1;
418            }
419
420            write_sb(fd, (unsigned long long)i * info.blocks_per_group * info.block_size + sb_offset, &sb);
421
422            ret = lseek64(fd, ((unsigned long long)i * info.blocks_per_group * info.block_size) +
423                              (info.block_size * (aux_info.first_data_block + 1)), SEEK_SET);
424            if (ret < 0)
425                critical_error_errno("failed to seek to block group descriptors");
426
427            if (!no_write) {
428                ret = write(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
429                if (ret < 0)
430                    critical_error_errno("failed to write block group descriptors");
431                if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
432                    critical_error("failed to write all of block group descriptors");
433            }
434        }
435    }
436
437    return 0;
438}
439
440
441static int get_direct_blocks(struct ext4_inode *inode, unsigned long long *block_list,
442                                                       unsigned int *count)
443{
444    unsigned int i = 0;
445    unsigned int ret = 0;
446    unsigned int sectors_per_block;
447
448    sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
449    while ((i < (inode->i_blocks_lo / sectors_per_block)) && (i < EXT4_NDIR_BLOCKS)) {
450        block_list[i] = inode->i_block[i];
451        i++;
452    }
453
454    *count += i;
455
456    if ((inode->i_blocks_lo / sectors_per_block) > EXT4_NDIR_BLOCKS) {
457        ret = 1;
458    }
459
460    return ret;
461}
462
463static int get_indirect_blocks(int fd, struct ext4_inode *inode,
464                               unsigned long long *block_list, unsigned int *count)
465{
466    unsigned int i;
467    unsigned int *indirect_block;
468    unsigned int sectors_per_block;
469
470    sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
471
472    indirect_block = (unsigned int *)malloc(info.block_size);
473    if (indirect_block == 0) {
474        critical_error("failed to allocate memory for indirect_block\n");
475    }
476
477    read_block(fd, inode->i_block[EXT4_NDIR_BLOCKS], indirect_block);
478
479    for(i = 0; i < (inode->i_blocks_lo / sectors_per_block - EXT4_NDIR_BLOCKS); i++) {
480       block_list[EXT4_NDIR_BLOCKS+i] = indirect_block[i];
481    }
482
483    *count += i;
484
485    free(indirect_block);
486
487    return 0;
488}
489
490static int get_block_list_indirect(int fd, struct ext4_inode *inode, unsigned long long *block_list)
491{
492    unsigned int count=0;
493
494    if (get_direct_blocks(inode, block_list, &count)) {
495        get_indirect_blocks(fd, inode, block_list, &count);
496    }
497
498    return count;
499}
500
501static int get_extent_ents(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
502{
503    int i, j;
504    struct ext4_extent *extent;
505    off64_t fs_block_num;
506
507    if (ext_hdr->eh_depth != 0) {
508        critical_error("get_extent_ents called with eh_depth != 0\n");
509    }
510
511    /* The extent entries immediately follow the header, so add 1 to the pointer
512     * and cast it to an extent pointer.
513     */
514    extent = (struct ext4_extent *)(ext_hdr + 1);
515
516    for (i = 0; i < ext_hdr->eh_entries; i++) {
517         fs_block_num = ((off64_t)extent->ee_start_hi << 32) | extent->ee_start_lo;
518         for (j = 0; j < extent->ee_len; j++) {
519             block_list[extent->ee_block+j] = fs_block_num+j;
520         }
521         extent++;
522    }
523
524    return 0;
525}
526
527static int get_extent_idx(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
528{
529    int i;
530    struct ext4_extent_idx *extent_idx;
531    struct ext4_extent_header *tmp_ext_hdr;
532    off64_t fs_block_num;
533    unsigned char block[MAX_EXT4_BLOCK_SIZE];
534
535    /* Sanity check */
536    if (ext_hdr->eh_depth == 0) {
537        critical_error("get_extent_idx called with eh_depth == 0\n");
538    }
539
540    /* The extent entries immediately follow the header, so add 1 to the pointer
541     * and cast it to an extent pointer.
542     */
543    extent_idx = (struct ext4_extent_idx *)(ext_hdr + 1);
544
545    for (i = 0; i < ext_hdr->eh_entries; i++) {
546         fs_block_num = ((off64_t)extent_idx->ei_leaf_hi << 32) | extent_idx->ei_leaf_lo;
547         read_block(fd, fs_block_num, block);
548         tmp_ext_hdr = (struct ext4_extent_header *)block;
549
550         if (tmp_ext_hdr->eh_depth == 0) {
551             get_extent_ents(fd, tmp_ext_hdr, block_list); /* leaf node, fill in block_list */
552         } else {
553             get_extent_idx(fd, tmp_ext_hdr, block_list); /* recurse down the tree */
554         }
555    }
556
557    return 0;
558}
559
560static int get_block_list_extents(int fd, struct ext4_inode *inode, unsigned long long *block_list)
561{
562    struct ext4_extent_header *extent_hdr;
563
564    extent_hdr = (struct ext4_extent_header *)inode->i_block;
565
566    if (extent_hdr->eh_magic != EXT4_EXT_MAGIC) {
567        critical_error("extent header has unexpected magic value 0x%4.4x\n",
568                       extent_hdr->eh_magic);
569    }
570
571    if (extent_hdr->eh_depth == 0) {
572         get_extent_ents(fd, (struct ext4_extent_header *)inode->i_block, block_list);
573         return 0;
574    }
575
576    get_extent_idx(fd, (struct ext4_extent_header *)inode->i_block, block_list);
577
578    return 0;
579}
580
581static int is_entry_dir(int fd, struct ext4_dir_entry_2 *dirp, int pass)
582{
583    struct ext4_inode inode;
584    int ret = 0;
585
586    if (dirp->file_type == EXT4_FT_DIR) {
587        ret = 1;
588    } else if (dirp->file_type == EXT4_FT_UNKNOWN) {
589        /* Somebody was too lazy to fill in the dir entry,
590         * so we have to go fetch it from the inode. Grrr.
591         */
592        /* if UPDATE_INODE_NUMS pass and the inode high bit is not
593         * set return false so we don't recurse down the tree that is
594         * already updated.  Otherwise, fetch inode, and return answer.
595         */
596        if ((pass == UPDATE_INODE_NUMS) && !(dirp->inode & 0x80000000)) {
597            ret = 0;
598        } else {
599            read_inode(fd, (dirp->inode & 0x7fffffff), &inode);
600            if (S_ISDIR(inode.i_mode)) {
601                ret = 1;
602            }
603        }
604    }
605
606    return ret;
607}
608
609static int recurse_dir(int fd, struct ext4_inode *inode, char *dirbuf, int dirsize, int mode)
610{
611    unsigned long long *block_list;
612    unsigned int num_blocks;
613    struct ext4_dir_entry_2 *dirp, *prev_dirp = 0;
614    char name[256];
615    unsigned int i, leftover_space, is_dir;
616    struct ext4_inode tmp_inode;
617    int tmp_dirsize;
618    char *tmp_dirbuf;
619
620    switch (mode) {
621        case SANITY_CHECK_PASS:
622        case MARK_INODE_NUMS:
623        case UPDATE_INODE_NUMS:
624            break;
625        default:
626            critical_error("recurse_dir() called witn unknown mode!\n");
627    }
628
629    if (dirsize % info.block_size) {
630        critical_error("dirsize %d not a multiple of block_size %d.  This is unexpected!\n",
631                dirsize, info.block_size);
632    }
633
634    num_blocks = dirsize / info.block_size;
635
636    block_list = malloc((num_blocks + 1) * sizeof(*block_list));
637    if (block_list == 0) {
638        critical_error("failed to allocate memory for block_list\n");
639    }
640
641    if (inode->i_flags & EXT4_EXTENTS_FL) {
642        get_block_list_extents(fd, inode, block_list);
643    } else {
644        /* A directory that requires doubly or triply indirect blocks in huge indeed,
645         * and will almost certainly not exist, especially since make_ext4fs only creates
646         * directories with extents, and the kernel will too, but check to make sure the
647         * directory is not that big and give an error if so.  Our limit is 12 direct blocks,
648         * plus block_size/4 singly indirect blocks, which for a filesystem with 4K blocks
649         * is a directory 1036 blocks long, or 4,243,456 bytes long!  Assuming an average
650         * filename length of 20 (which I think is generous) thats 20 + 8 bytes overhead
651         * per entry, or 151,552 entries in the directory!
652         */
653        if (num_blocks > (info.block_size / 4 + EXT4_NDIR_BLOCKS)) {
654            critical_error("Non-extent based directory is too big!\n");
655        }
656        get_block_list_indirect(fd, inode, block_list);
657    }
658
659    /* Read in all the blocks for this directory */
660    for (i = 0; i < num_blocks; i++) {
661        read_block(fd, block_list[i], dirbuf + (i * info.block_size));
662    }
663
664    dirp = (struct ext4_dir_entry_2 *)dirbuf;
665    while (dirp < (struct ext4_dir_entry_2 *)(dirbuf + dirsize)) {
666        leftover_space = (char *)(dirbuf + dirsize) - (char *)dirp;
667        if (((mode == SANITY_CHECK_PASS) || (mode == UPDATE_INODE_NUMS)) &&
668            (leftover_space <= 8) && prev_dirp) {
669            /* This is a bug in an older version of make_ext4fs, where it
670             * didn't properly include the rest of the block in rec_len.
671             * Update rec_len on the previous entry to include the rest of
672             * the block and exit the loop.
673             */
674            if (verbose) {
675                printf("fixing up short rec_len for diretory entry for %s\n", name);
676            }
677            prev_dirp->rec_len += leftover_space;
678            break;
679        }
680
681        if (dirp->inode == 0) {
682            /* This is the last entry in the directory */
683            break;
684        }
685
686        strncpy(name, dirp->name, dirp->name_len);
687        name[dirp->name_len]='\0';
688
689        /* Only recurse on pass UPDATE_INODE_NUMS if the high bit is set.
690         * Otherwise, this inode entry has already been updated
691         * and we'll do the wrong thing.  Also don't recurse on . or ..,
692         * and certainly not on non-directories!
693         */
694        /* Hrm, looks like filesystems made by fastboot on stingray set the file_type
695         * flag, but the lost+found directory has the type set to Unknown, which
696         * seems to imply I need to read the inode and get it.
697         */
698        is_dir = is_entry_dir(fd, dirp, mode);
699        if ( is_dir && (strcmp(name, ".") && strcmp(name, "..")) &&
700            ((mode == SANITY_CHECK_PASS) || (mode == MARK_INODE_NUMS) ||
701              ((mode == UPDATE_INODE_NUMS) && (dirp->inode & 0x80000000))) ) {
702            /* A directory!  Recurse! */
703            read_inode(fd, dirp->inode & 0x7fffffff, &tmp_inode);
704
705            if (!S_ISDIR(tmp_inode.i_mode)) {
706                critical_error("inode %d for name %s does not point to a directory\n",
707                        dirp->inode & 0x7fffffff, name);
708            }
709            if (verbose) {
710                printf("inode %d %s use extents\n", dirp->inode & 0x7fffffff,
711                       (tmp_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
712            }
713
714            tmp_dirsize = tmp_inode.i_blocks_lo * INODE_BLOCK_SIZE;
715            if (verbose) {
716                printf("dir size = %d bytes\n", tmp_dirsize);
717            }
718
719            tmp_dirbuf = malloc(tmp_dirsize);
720            if (tmp_dirbuf == 0) {
721                critical_error("failed to allocate memory for tmp_dirbuf\n");
722            }
723
724            recurse_dir(fd, &tmp_inode, tmp_dirbuf, tmp_dirsize, mode);
725
726            free(tmp_dirbuf);
727        }
728
729        if (verbose) {
730            if (is_dir) {
731                printf("Directory %s\n", name);
732            } else {
733                printf("Non-directory %s\n", name);
734            }
735        }
736
737        /* Process entry based on current mode.  Either set high bit or change inode number */
738        if (mode == MARK_INODE_NUMS) {
739            dirp->inode |= 0x80000000;
740        } else if (mode == UPDATE_INODE_NUMS) {
741            if (dirp->inode & 0x80000000) {
742                dirp->inode = compute_new_inum(dirp->inode & 0x7fffffff);
743            }
744        }
745
746        /* Point dirp at the next entry */
747        prev_dirp = dirp;
748        dirp = (struct ext4_dir_entry_2*)((char *)dirp + dirp->rec_len);
749    }
750
751    /* Write out all the blocks for this directory */
752    for (i = 0; i < num_blocks; i++) {
753        write_block(fd, block_list[i], dirbuf + (i * info.block_size));
754    }
755
756    free(block_list);
757
758    return 0;
759}
760
761int ext4fixup(char *fsdev)
762{
763    return ext4fixup_internal(fsdev, 0, 0);
764}
765
766int ext4fixup_internal(char *fsdev, int v_flag, int n_flag)
767{
768    int fd;
769    struct ext4_inode root_inode;
770    unsigned int dirsize;
771    char *dirbuf;
772
773    if (setjmp(setjmp_env))
774        return EXIT_FAILURE; /* Handle a call to longjmp() */
775
776    verbose = v_flag;
777    no_write = n_flag;
778
779    fd = open(fsdev, O_RDWR);
780
781    if (fd < 0)
782        critical_error_errno("failed to open filesystem image");
783
784    read_ext(fd);
785
786    if ((info.feat_incompat & EXT4_FEATURE_INCOMPAT_FILETYPE) == 0) {
787        critical_error("Expected filesystem to have filetype flag set\n");
788    }
789
790#if 0 // If we have to fix the directory rec_len issue, we can't use this check
791    /* Check to see if the inodes/group is copacetic */
792    if (info.inodes_per_blockgroup % (info.block_size/info.inode_size) == 0) {
793             /* This filesystem has either already been updated, or was
794              * made correctly.
795              */
796             if (verbose) {
797                 printf("%s: filesystem correct, no work to do\n", me);
798             }
799             exit(0);
800    }
801#endif
802
803    /* Compute what the new value of inodes_per_blockgroup will be when we're done */
804    new_inodes_per_group=ALIGN(info.inodes_per_group,(info.block_size/info.inode_size));
805
806    read_inode(fd, EXT4_ROOT_INO, &root_inode);
807
808    if (!S_ISDIR(root_inode.i_mode)) {
809        critical_error("root inode %d does not point to a directory\n", EXT4_ROOT_INO);
810    }
811    if (verbose) {
812        printf("inode %d %s use extents\n", EXT4_ROOT_INO,
813               (root_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
814    }
815
816    dirsize = root_inode.i_blocks_lo * INODE_BLOCK_SIZE;
817    if (verbose) {
818        printf("root dir size = %d bytes\n", dirsize);
819    }
820
821    dirbuf = malloc(dirsize);
822    if (dirbuf == 0) {
823        critical_error("failed to allocate memory for dirbuf\n");
824    }
825
826    /* Perform a sanity check pass first, try to catch any errors that will occur
827     * before we actually change anything, so we don't leave a filesystem in a
828     * corrupted, unrecoverable state.  Set no_write, make it quiet, and do a recurse
829     * pass and a update_superblock pass.  Set flags back to requested state when done.
830     * Only perform sanity check if the state is unset.  If the state is _NOT_ unset,
831     * then the tool has already been run and interrupted, and it presumably ran and
832     * passed sanity checked before it got interrupted.  It is _NOT_ safe to run sanity
833     * check if state is unset because it assumes inodes are to be computed using the
834     * old inodes/group, but some inode numbers may be updated to the new number.
835     */
836    if (get_fs_fixup_state(fd) == STATE_UNSET) {
837        verbose = 0;
838        no_write = 1;
839        recurse_dir(fd, &root_inode, dirbuf, dirsize, SANITY_CHECK_PASS);
840        update_superblocks_and_bg_desc(fd, STATE_UNSET);
841        verbose = v_flag;
842        no_write = n_flag;
843
844        set_fs_fixup_state(fd, STATE_MARKING_INUMS);
845    }
846
847    if (get_fs_fixup_state(fd) == STATE_MARKING_INUMS) {
848        if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, MARK_INODE_NUMS)) {
849            set_fs_fixup_state(fd, STATE_UPDATING_INUMS);
850        }
851    }
852
853    if (get_fs_fixup_state(fd) == STATE_UPDATING_INUMS) {
854        if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, UPDATE_INODE_NUMS)) {
855            set_fs_fixup_state(fd, STATE_UPDATING_SB);
856        }
857    }
858
859    if (get_fs_fixup_state(fd) == STATE_UPDATING_SB) {
860        /* set the new inodes/blockgroup number,
861         * and sets the state back to 0.
862         */
863        if (!update_superblocks_and_bg_desc(fd, STATE_UPDATING_SB)) {
864            set_fs_fixup_state(fd, STATE_UNSET);
865        }
866    }
867
868    close(fd);
869
870    return 0;
871}
872