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
17#include "ext4fixup.h"
18
19#ifndef _LARGEFILE64_SOURCE
20#define _LARGEFILE64_SOURCE 1
21#endif
22
23#include <fcntl.h>
24#include <sys/stat.h>
25#include <sys/types.h>
26#include <unistd.h>
27
28#include <sparse/sparse.h>
29
30#include "allocate.h"
31#include "ext4_utils/ext4_extents.h"
32#include "ext4_utils/ext4_utils.h"
33#include "ext4_utils/make_ext4fs.h"
34
35#ifndef _WIN32
36#include <sys/mman.h>
37#endif
38
39#if defined(__APPLE__) && defined(__MACH__)
40#define lseek64 lseek
41#define off64_t off_t
42#endif
43
44/* The inode block count for a file/directory is in units of 512 byte blocks,
45 * _NOT_ the filesystem block size!
46 */
47#define INODE_BLOCK_SIZE 512
48
49#define MAX_EXT4_BLOCK_SIZE 4096
50
51/* The two modes the recurse_dir() can be in */
52#define SANITY_CHECK_PASS 1
53#define MARK_INODE_NUMS   2
54#define UPDATE_INODE_NUMS 3
55
56/* Magic numbers to indicate what state the update process is in */
57#define MAGIC_STATE_MARKING_INUMS  0x7000151515565512ll
58#define MAGIC_STATE_UPDATING_INUMS 0x6121131211735123ll
59#define MAGIC_STATE_UPDATING_SB    0x15e1715151558477ll
60
61/* Internal state variables corresponding to the magic numbers */
62#define STATE_UNSET          0
63#define STATE_MARKING_INUMS  1
64#define STATE_UPDATING_INUMS 2
65#define STATE_UPDATING_SB    3
66
67/* Used for automated testing of this programs ability to stop and be restarted wthout error */
68static int bail_phase = 0;
69static int bail_loc = 0;
70static int bail_count = 0;
71static int count = 0;
72
73/* global flags */
74static int verbose = 0;
75static int no_write = 0;
76
77static int new_inodes_per_group = 0;
78
79static int no_write_fixup_state = 0;
80
81static int compute_new_inum(unsigned int old_inum)
82{
83    unsigned int group, offset;
84
85    group = (old_inum - 1) / info.inodes_per_group;
86    offset = (old_inum -1) % info.inodes_per_group;
87
88    return (group * new_inodes_per_group) + offset + 1;
89}
90
91static int get_fs_fixup_state(int fd)
92{
93    unsigned long long magic;
94    int ret, len;
95
96    if (no_write) {
97        return no_write_fixup_state;
98    }
99
100    lseek64(fd, 0, SEEK_SET);
101    len = read(fd, &magic, sizeof(magic));
102    if (len != sizeof(magic)) {
103        critical_error("cannot read fixup_state\n");
104    }
105
106    switch (magic) {
107        case MAGIC_STATE_MARKING_INUMS:
108            ret = STATE_MARKING_INUMS;
109            break;
110        case MAGIC_STATE_UPDATING_INUMS:
111            ret = STATE_UPDATING_INUMS;
112            break;
113        case MAGIC_STATE_UPDATING_SB:
114            ret = STATE_UPDATING_SB;
115            break;
116        default:
117            ret = STATE_UNSET;
118    }
119    return ret;
120}
121
122static int set_fs_fixup_state(int fd, int state)
123{
124    unsigned long long magic;
125    struct ext4_super_block sb;
126    int len;
127
128    if (no_write) {
129        no_write_fixup_state = state;
130        return 0;
131    }
132
133    switch (state) {
134        case STATE_MARKING_INUMS:
135            magic = MAGIC_STATE_MARKING_INUMS;
136            break;
137        case STATE_UPDATING_INUMS:
138            magic = MAGIC_STATE_UPDATING_INUMS;
139            break;
140        case STATE_UPDATING_SB:
141            magic = MAGIC_STATE_UPDATING_SB;
142            break;
143        case STATE_UNSET:
144        default:
145            magic = 0ll;
146            break;
147    }
148
149    lseek64(fd, 0, SEEK_SET);
150    len = write(fd, &magic, sizeof(magic));
151    if (len != sizeof(magic)) {
152        critical_error("cannot write fixup_state\n");
153    }
154
155    read_sb(fd, &sb);
156    if (magic) {
157        /* If we are in the process of updating the filesystem, make it unmountable */
158        sb.s_desc_size |= 1;
159    } else {
160        /* we are done, so make the filesystem mountable again */
161        sb.s_desc_size &= ~1;
162    }
163
164    if (!no_write) {
165        write_sb(fd, 1024, &sb);
166    }
167
168    return 0;
169}
170
171static int read_inode(int fd, unsigned int inum, struct ext4_inode *inode)
172{
173    unsigned int bg_num, bg_offset;
174    off64_t inode_offset;
175    int len;
176
177    bg_num = (inum-1) / info.inodes_per_group;
178    bg_offset = (inum-1) % info.inodes_per_group;
179
180    inode_offset = ((unsigned long long)aux_info.bg_desc[bg_num].bg_inode_table * info.block_size) +
181                    (bg_offset * info.inode_size);
182
183    if (lseek64(fd, inode_offset, SEEK_SET) < 0) {
184        critical_error_errno("failed to seek to inode %d\n", inum);
185    }
186
187    len=read(fd, inode, sizeof(*inode));
188    if (len != sizeof(*inode)) {
189        critical_error_errno("failed to read inode %d\n", inum);
190    }
191
192    return 0;
193}
194
195static int read_block(int fd, unsigned long long block_num, void *block)
196{
197    off64_t off;
198    unsigned int len;
199
200    off = block_num * info.block_size;
201
202    if (lseek64(fd, off, SEEK_SET) , 0) {
203        critical_error_errno("failed to seek to block %lld\n", block_num);
204    }
205
206    len=read(fd, block, info.block_size);
207    if (len != info.block_size) {
208        critical_error_errno("failed to read block %lld\n", block_num);
209    }
210
211    return 0;
212}
213
214static int write_block(int fd, unsigned long long block_num, void *block)
215{
216    off64_t off;
217    unsigned int len;
218
219    if (no_write) {
220        return 0;
221    }
222
223    off = block_num * info.block_size;
224
225    if (lseek64(fd, off, SEEK_SET) < 0) {
226        critical_error_errno("failed to seek to block %lld\n", block_num);
227    }
228
229    len=write(fd, block, info.block_size);
230    if (len != info.block_size) {
231        critical_error_errno("failed to write block %lld\n", block_num);
232    }
233
234    return 0;
235}
236
237static void check_inode_bitmap(int fd, unsigned int bg_num)
238{
239    unsigned int inode_bitmap_block_num;
240    unsigned char block[MAX_EXT4_BLOCK_SIZE];
241    int i, bitmap_updated = 0;
242
243    /* Using the bg_num, aux_info.bg_desc[], info.inodes_per_group and
244     * new_inodes_per_group, retrieve the inode bitmap, and make sure
245     * the bits between the old and new size are clear
246     */
247    inode_bitmap_block_num = aux_info.bg_desc[bg_num].bg_inode_bitmap;
248
249    read_block(fd, inode_bitmap_block_num, block);
250
251    for (i = info.inodes_per_group; i < new_inodes_per_group; i++) {
252        if (bitmap_get_bit(block, i)) {
253            bitmap_clear_bit(block, i);
254            bitmap_updated = 1;
255        }
256    }
257
258    if (bitmap_updated) {
259        if (verbose) {
260            printf("Warning: updated inode bitmap for block group %d\n", bg_num);
261        }
262        write_block(fd, inode_bitmap_block_num, block);
263    }
264
265    return;
266}
267
268/* Update the superblock and bgdesc of the specified block group */
269static int update_superblocks_and_bg_desc(int fd, int state)
270{
271    off64_t ret;
272    struct ext4_super_block sb;
273    unsigned int num_block_groups, total_new_inodes;
274    unsigned int i;
275
276
277    read_sb(fd, &sb);
278
279    /* Compute how many more inodes are now available */
280    num_block_groups = DIV_ROUND_UP(aux_info.len_blocks, info.blocks_per_group);
281    total_new_inodes = num_block_groups * (new_inodes_per_group - sb.s_inodes_per_group);
282
283    if (verbose) {
284        printf("created %d additional inodes\n", total_new_inodes);
285    }
286
287    /* Update the free inodes count in each block group descriptor */
288    for (i = 0; i < num_block_groups; i++) {
289       if (state == STATE_UPDATING_SB) {
290           aux_info.bg_desc[i].bg_free_inodes_count += (new_inodes_per_group - sb.s_inodes_per_group);
291       }
292       check_inode_bitmap(fd, i);
293    }
294
295    /* First some sanity checks */
296    if ((sb.s_inodes_count + total_new_inodes) != (new_inodes_per_group * num_block_groups)) {
297        critical_error("Failed sanity check on new inode count\n");
298    }
299    if (new_inodes_per_group % (info.block_size/info.inode_size)) {
300        critical_error("Failed sanity check on new inode per group alignment\n");
301    }
302
303    /* Update the free inodes count in the superblock */
304    sb.s_inodes_count += total_new_inodes;
305    sb.s_free_inodes_count += total_new_inodes;
306    sb.s_inodes_per_group = new_inodes_per_group;
307
308    for (i = 0; i < aux_info.groups; i++) {
309        if (ext4_bg_has_super_block(i)) {
310            unsigned int sb_offset;
311
312            if (i == 0) {
313              /* The first superblock is offset by 1K to leave room for boot sectors */
314              sb_offset = 1024;
315            } else {
316              sb_offset = 0;
317            }
318
319            sb.s_block_group_nr = i;
320            /* Don't write out the backup superblocks with the bit set in the s_desc_size
321             * which prevents the filesystem from mounting.  The bit for the primary
322             * superblock will be cleared on the final call to set_fs_fixup_state() */
323            if (i != 0) {
324                sb.s_desc_size &= ~1;
325            }
326
327            if (!no_write) {
328                write_sb(fd,
329                         (unsigned long long)i
330                         * info.blocks_per_group * info.block_size
331                         + sb_offset,
332                         &sb);
333            }
334
335            ret = lseek64(fd, ((unsigned long long)i * info.blocks_per_group * info.block_size) +
336                              (info.block_size * (aux_info.first_data_block + 1)), SEEK_SET);
337            if (ret < 0)
338                critical_error_errno("failed to seek to block group descriptors");
339
340            if (!no_write) {
341                ret = write(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
342                if (ret < 0)
343                    critical_error_errno("failed to write block group descriptors");
344                if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
345                    critical_error("failed to write all of block group descriptors");
346            }
347        }
348        if ((bail_phase == 4) && ((unsigned int)bail_count == i)) {
349            critical_error("bailing at phase 4\n");
350        }
351    }
352
353    return 0;
354}
355
356
357static int get_direct_blocks(struct ext4_inode *inode, unsigned long long *block_list,
358                                                       unsigned int *count)
359{
360    unsigned int i = 0;
361    unsigned int ret = 0;
362    unsigned int sectors_per_block;
363
364    sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
365    while ((i < (inode->i_blocks_lo / sectors_per_block)) && (i < EXT4_NDIR_BLOCKS)) {
366        block_list[i] = inode->i_block[i];
367        i++;
368    }
369
370    *count += i;
371
372    if ((inode->i_blocks_lo / sectors_per_block) > EXT4_NDIR_BLOCKS) {
373        ret = 1;
374    }
375
376    return ret;
377}
378
379static int get_indirect_blocks(int fd, struct ext4_inode *inode,
380                               unsigned long long *block_list, unsigned int *count)
381{
382    unsigned int i;
383    unsigned int *indirect_block;
384    unsigned int sectors_per_block;
385
386    sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
387
388    indirect_block = (unsigned int *)malloc(info.block_size);
389    if (indirect_block == 0) {
390        critical_error("failed to allocate memory for indirect_block\n");
391    }
392
393    read_block(fd, inode->i_block[EXT4_NDIR_BLOCKS], indirect_block);
394
395    for(i = 0; i < (inode->i_blocks_lo / sectors_per_block - EXT4_NDIR_BLOCKS); i++) {
396       block_list[EXT4_NDIR_BLOCKS+i] = indirect_block[i];
397    }
398
399    *count += i;
400
401    free(indirect_block);
402
403    return 0;
404}
405
406static int get_block_list_indirect(int fd, struct ext4_inode *inode, unsigned long long *block_list)
407{
408    unsigned int count=0;
409
410    if (get_direct_blocks(inode, block_list, &count)) {
411        get_indirect_blocks(fd, inode, block_list, &count);
412    }
413
414    return count;
415}
416
417static int get_extent_ents(struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
418{
419    int i, j;
420    struct ext4_extent *extent;
421    off64_t fs_block_num;
422
423    if (ext_hdr->eh_depth != 0) {
424        critical_error("get_extent_ents called with eh_depth != 0\n");
425    }
426
427    /* The extent entries immediately follow the header, so add 1 to the pointer
428     * and cast it to an extent pointer.
429     */
430    extent = (struct ext4_extent *)(ext_hdr + 1);
431
432    for (i = 0; i < ext_hdr->eh_entries; i++) {
433         fs_block_num = ((off64_t)extent->ee_start_hi << 32) | extent->ee_start_lo;
434         for (j = 0; j < extent->ee_len; j++) {
435             block_list[extent->ee_block+j] = fs_block_num+j;
436         }
437         extent++;
438    }
439
440    return 0;
441}
442
443static int get_extent_idx(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
444{
445    int i;
446    struct ext4_extent_idx *extent_idx;
447    struct ext4_extent_header *tmp_ext_hdr;
448    off64_t fs_block_num;
449    unsigned char block[MAX_EXT4_BLOCK_SIZE];
450
451    /* Sanity check */
452    if (ext_hdr->eh_depth == 0) {
453        critical_error("get_extent_idx called with eh_depth == 0\n");
454    }
455
456    /* The extent entries immediately follow the header, so add 1 to the pointer
457     * and cast it to an extent pointer.
458     */
459    extent_idx = (struct ext4_extent_idx *)(ext_hdr + 1);
460
461    for (i = 0; i < ext_hdr->eh_entries; i++) {
462         fs_block_num = ((off64_t)extent_idx->ei_leaf_hi << 32) | extent_idx->ei_leaf_lo;
463         read_block(fd, fs_block_num, block);
464         tmp_ext_hdr = (struct ext4_extent_header *)block;
465
466         if (tmp_ext_hdr->eh_depth == 0) {
467             get_extent_ents(tmp_ext_hdr, block_list); /* leaf node, fill in block_list */
468         } else {
469             get_extent_idx(fd, tmp_ext_hdr, block_list); /* recurse down the tree */
470         }
471    }
472
473    return 0;
474}
475
476static int get_block_list_extents(int fd, struct ext4_inode *inode, unsigned long long *block_list)
477{
478    struct ext4_extent_header *extent_hdr;
479
480    extent_hdr = (struct ext4_extent_header *)inode->i_block;
481
482    if (extent_hdr->eh_magic != EXT4_EXT_MAGIC) {
483        critical_error("extent header has unexpected magic value 0x%4.4x\n",
484                       extent_hdr->eh_magic);
485    }
486
487    if (extent_hdr->eh_depth == 0) {
488         get_extent_ents((struct ext4_extent_header *)inode->i_block, block_list);
489         return 0;
490    }
491
492    get_extent_idx(fd, (struct ext4_extent_header *)inode->i_block, block_list);
493
494    return 0;
495}
496
497static int is_entry_dir(int fd, struct ext4_dir_entry_2 *dirp, int pass)
498{
499    struct ext4_inode inode;
500    int ret = 0;
501
502    if (dirp->file_type == EXT4_FT_DIR) {
503        ret = 1;
504    } else if (dirp->file_type == EXT4_FT_UNKNOWN) {
505        /* Somebody was too lazy to fill in the dir entry,
506         * so we have to go fetch it from the inode. Grrr.
507         */
508        /* if UPDATE_INODE_NUMS pass and the inode high bit is not
509         * set return false so we don't recurse down the tree that is
510         * already updated.  Otherwise, fetch inode, and return answer.
511         */
512        if ((pass == UPDATE_INODE_NUMS) && !(dirp->inode & 0x80000000)) {
513            ret = 0;
514        } else {
515            read_inode(fd, (dirp->inode & 0x7fffffff), &inode);
516            if (S_ISDIR(inode.i_mode)) {
517                ret = 1;
518            }
519        }
520    }
521
522    return ret;
523}
524
525static int recurse_dir(int fd, struct ext4_inode *inode, char *dirbuf, int dirsize, int mode)
526{
527    unsigned long long *block_list;
528    unsigned int num_blocks;
529    struct ext4_dir_entry_2 *dirp, *prev_dirp = 0;
530    char name[256];
531    unsigned int i, leftover_space, is_dir;
532    struct ext4_inode tmp_inode;
533    int tmp_dirsize;
534    char *tmp_dirbuf;
535
536    switch (mode) {
537        case SANITY_CHECK_PASS:
538        case MARK_INODE_NUMS:
539        case UPDATE_INODE_NUMS:
540            break;
541        default:
542            critical_error("recurse_dir() called witn unknown mode!\n");
543    }
544
545    if (dirsize % info.block_size) {
546        critical_error("dirsize %d not a multiple of block_size %d.  This is unexpected!\n",
547                dirsize, info.block_size);
548    }
549
550    num_blocks = dirsize / info.block_size;
551
552    block_list = malloc((num_blocks + 1) * sizeof(*block_list));
553    if (block_list == 0) {
554        critical_error("failed to allocate memory for block_list\n");
555    }
556
557    if (inode->i_flags & EXT4_EXTENTS_FL) {
558        get_block_list_extents(fd, inode, block_list);
559    } else {
560        /* A directory that requires doubly or triply indirect blocks in huge indeed,
561         * and will almost certainly not exist, especially since make_ext4fs only creates
562         * directories with extents, and the kernel will too, but check to make sure the
563         * directory is not that big and give an error if so.  Our limit is 12 direct blocks,
564         * plus block_size/4 singly indirect blocks, which for a filesystem with 4K blocks
565         * is a directory 1036 blocks long, or 4,243,456 bytes long!  Assuming an average
566         * filename length of 20 (which I think is generous) thats 20 + 8 bytes overhead
567         * per entry, or 151,552 entries in the directory!
568         */
569        if (num_blocks > (info.block_size / 4 + EXT4_NDIR_BLOCKS)) {
570            critical_error("Non-extent based directory is too big!\n");
571        }
572        get_block_list_indirect(fd, inode, block_list);
573    }
574
575    /* Read in all the blocks for this directory */
576    for (i = 0; i < num_blocks; i++) {
577        read_block(fd, block_list[i], dirbuf + (i * info.block_size));
578    }
579
580    dirp = (struct ext4_dir_entry_2 *)dirbuf;
581    while (dirp < (struct ext4_dir_entry_2 *)(dirbuf + dirsize)) {
582        count++;
583        leftover_space = (char *)(dirbuf + dirsize) - (char *)dirp;
584        if (((mode == SANITY_CHECK_PASS) || (mode == UPDATE_INODE_NUMS)) &&
585            (leftover_space <= 8) && prev_dirp) {
586            /* This is a bug in an older version of make_ext4fs, where it
587             * didn't properly include the rest of the block in rec_len.
588             * Update rec_len on the previous entry to include the rest of
589             * the block and exit the loop.
590             */
591            if (verbose) {
592                printf("fixing up short rec_len for diretory entry for %s\n", name);
593            }
594            prev_dirp->rec_len += leftover_space;
595            break;
596        }
597
598        if (dirp->inode == 0) {
599            /* This is the last entry in the directory */
600            break;
601        }
602
603        strncpy(name, dirp->name, dirp->name_len);
604        name[dirp->name_len]='\0';
605
606        /* Only recurse on pass UPDATE_INODE_NUMS if the high bit is set.
607         * Otherwise, this inode entry has already been updated
608         * and we'll do the wrong thing.  Also don't recurse on . or ..,
609         * and certainly not on non-directories!
610         */
611        /* Hrm, looks like filesystems made by fastboot on stingray set the file_type
612         * flag, but the lost+found directory has the type set to Unknown, which
613         * seems to imply I need to read the inode and get it.
614         */
615        is_dir = is_entry_dir(fd, dirp, mode);
616        if ( is_dir && (strcmp(name, ".") && strcmp(name, "..")) &&
617            ((mode == SANITY_CHECK_PASS) || (mode == MARK_INODE_NUMS) ||
618              ((mode == UPDATE_INODE_NUMS) && (dirp->inode & 0x80000000))) ) {
619            /* A directory!  Recurse! */
620            read_inode(fd, dirp->inode & 0x7fffffff, &tmp_inode);
621
622            if (!S_ISDIR(tmp_inode.i_mode)) {
623                critical_error("inode %d for name %s does not point to a directory\n",
624                        dirp->inode & 0x7fffffff, name);
625            }
626            if (verbose) {
627                printf("inode %d %s use extents\n", dirp->inode & 0x7fffffff,
628                       (tmp_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
629            }
630
631            tmp_dirsize = tmp_inode.i_blocks_lo * INODE_BLOCK_SIZE;
632            if (verbose) {
633                printf("dir size = %d bytes\n", tmp_dirsize);
634            }
635
636            tmp_dirbuf = malloc(tmp_dirsize);
637            if (tmp_dirbuf == 0) {
638                critical_error("failed to allocate memory for tmp_dirbuf\n");
639            }
640
641            recurse_dir(fd, &tmp_inode, tmp_dirbuf, tmp_dirsize, mode);
642
643            free(tmp_dirbuf);
644        }
645
646        if (verbose) {
647            if (is_dir) {
648                printf("Directory %s\n", name);
649            } else {
650                printf("Non-directory %s\n", name);
651            }
652        }
653
654        /* Process entry based on current mode.  Either set high bit or change inode number */
655        if (mode == MARK_INODE_NUMS) {
656            dirp->inode |= 0x80000000;
657        } else if (mode == UPDATE_INODE_NUMS) {
658            if (dirp->inode & 0x80000000) {
659                dirp->inode = compute_new_inum(dirp->inode & 0x7fffffff);
660            }
661        }
662
663        if ((bail_phase == mode) && (bail_loc == 1) && (bail_count == count)) {
664            critical_error("Bailing at phase %d, loc 1 and count %d\n", mode, count);
665        }
666
667        /* Point dirp at the next entry */
668        prev_dirp = dirp;
669        dirp = (struct ext4_dir_entry_2*)((char *)dirp + dirp->rec_len);
670    }
671
672    /* Write out all the blocks for this directory */
673    for (i = 0; i < num_blocks; i++) {
674        write_block(fd, block_list[i], dirbuf + (i * info.block_size));
675        if ((bail_phase == mode) && (bail_loc == 2) && (bail_count <= count)) {
676            critical_error("Bailing at phase %d, loc 2 and count %d\n", mode, count);
677        }
678    }
679
680    free(block_list);
681
682    return 0;
683}
684
685int ext4fixup(char *fsdev)
686{
687    return ext4fixup_internal(fsdev, 0, 0, 0, 0, 0);
688}
689
690int ext4fixup_internal(char *fsdev, int v_flag, int n_flag,
691                       int stop_phase, int stop_loc, int stop_count)
692{
693    int fd;
694    struct ext4_inode root_inode;
695    unsigned int dirsize;
696    char *dirbuf;
697
698    if (setjmp(setjmp_env))
699        return EXIT_FAILURE; /* Handle a call to longjmp() */
700
701    verbose = v_flag;
702    no_write = n_flag;
703
704    bail_phase = stop_phase;
705    bail_loc = stop_loc;
706    bail_count = stop_count;
707
708    fd = open(fsdev, O_RDWR);
709
710    if (fd < 0)
711        critical_error_errno("failed to open filesystem image");
712
713    read_ext(fd, verbose);
714
715    if (info.feat_incompat & EXT4_FEATURE_INCOMPAT_RECOVER) {
716        critical_error("Filesystem needs recovery first, mount and unmount to do that\n");
717    }
718
719    /* Clear the low bit which is set while this tool is in progress.
720     * If the tool crashes, it will still be set when we restart.
721     * The low bit is set to make the filesystem unmountable while
722     * it is being fixed up.  Also allow 0, which means the old ext2
723     * size is in use.
724     */
725    if (((aux_info.sb->s_desc_size & ~1) != sizeof(struct ext2_group_desc)) &&
726        ((aux_info.sb->s_desc_size & ~1) != 0))
727        critical_error("error: bg_desc_size != sizeof(struct ext2_group_desc)\n");
728
729    if ((info.feat_incompat & EXT4_FEATURE_INCOMPAT_FILETYPE) == 0) {
730        critical_error("Expected filesystem to have filetype flag set\n");
731    }
732
733#if 0 // If we have to fix the directory rec_len issue, we can't use this check
734    /* Check to see if the inodes/group is copacetic */
735    if (info.inodes_per_blockgroup % (info.block_size/info.inode_size) == 0) {
736             /* This filesystem has either already been updated, or was
737              * made correctly.
738              */
739             if (verbose) {
740                 printf("%s: filesystem correct, no work to do\n", me);
741             }
742             exit(0);
743    }
744#endif
745
746    /* Compute what the new value of inodes_per_blockgroup will be when we're done */
747    new_inodes_per_group=EXT4_ALIGN(info.inodes_per_group,(info.block_size/info.inode_size));
748
749    read_inode(fd, EXT4_ROOT_INO, &root_inode);
750
751    if (!S_ISDIR(root_inode.i_mode)) {
752        critical_error("root inode %d does not point to a directory\n", EXT4_ROOT_INO);
753    }
754    if (verbose) {
755        printf("inode %d %s use extents\n", EXT4_ROOT_INO,
756               (root_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
757    }
758
759    dirsize = root_inode.i_blocks_lo * INODE_BLOCK_SIZE;
760    if (verbose) {
761        printf("root dir size = %d bytes\n", dirsize);
762    }
763
764    dirbuf = malloc(dirsize);
765    if (dirbuf == 0) {
766        critical_error("failed to allocate memory for dirbuf\n");
767    }
768
769    /* Perform a sanity check pass first, try to catch any errors that will occur
770     * before we actually change anything, so we don't leave a filesystem in a
771     * corrupted, unrecoverable state.  Set no_write, make it quiet, and do a recurse
772     * pass and a update_superblock pass.  Set flags back to requested state when done.
773     * Only perform sanity check if the state is unset.  If the state is _NOT_ unset,
774     * then the tool has already been run and interrupted, and it presumably ran and
775     * passed sanity checked before it got interrupted.  It is _NOT_ safe to run sanity
776     * check if state is unset because it assumes inodes are to be computed using the
777     * old inodes/group, but some inode numbers may be updated to the new number.
778     */
779    if (get_fs_fixup_state(fd) == STATE_UNSET) {
780        verbose = 0;
781        no_write = 1;
782        recurse_dir(fd, &root_inode, dirbuf, dirsize, SANITY_CHECK_PASS);
783        update_superblocks_and_bg_desc(fd, STATE_UNSET);
784        verbose = v_flag;
785        no_write = n_flag;
786
787        set_fs_fixup_state(fd, STATE_MARKING_INUMS);
788    }
789
790    if (get_fs_fixup_state(fd) == STATE_MARKING_INUMS) {
791        count = 0; /* Reset debugging counter */
792        if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, MARK_INODE_NUMS)) {
793            set_fs_fixup_state(fd, STATE_UPDATING_INUMS);
794        }
795    }
796
797    if (get_fs_fixup_state(fd) == STATE_UPDATING_INUMS) {
798        count = 0; /* Reset debugging counter */
799        if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, UPDATE_INODE_NUMS)) {
800            set_fs_fixup_state(fd, STATE_UPDATING_SB);
801        }
802    }
803
804    if (get_fs_fixup_state(fd) == STATE_UPDATING_SB) {
805        /* set the new inodes/blockgroup number,
806         * and sets the state back to 0.
807         */
808        if (!update_superblocks_and_bg_desc(fd, STATE_UPDATING_SB)) {
809            set_fs_fixup_state(fd, STATE_UNSET);
810        }
811    }
812
813    close(fd);
814    free(dirbuf);
815
816    return 0;
817}
818