Home | History | Annotate | Download | only in ext4_utils
      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 "allocate.h"
     20 #include "ext4fixup.h"
     21 
     22 #include <sparse/sparse.h>
     23 
     24 #include <sys/types.h>
     25 #include <sys/stat.h>
     26 #include <sys/types.h>
     27 #include <fcntl.h>
     28 #include <unistd.h>
     29 
     30 #ifndef USE_MINGW
     31 #include <sys/mman.h>
     32 #endif
     33 
     34 #if defined(__APPLE__) && defined(__MACH__)
     35 #define lseek64 lseek
     36 #define off64_t off_t
     37 #endif
     38 
     39 /* The inode block count for a file/directory is in units of 512 byte blocks,
     40  * _NOT_ the filesystem block size!
     41  */
     42 #define INODE_BLOCK_SIZE 512
     43 
     44 #define MAX_EXT4_BLOCK_SIZE 4096
     45 
     46 /* The two modes the recurse_dir() can be in */
     47 #define SANITY_CHECK_PASS 1
     48 #define MARK_INODE_NUMS   2
     49 #define UPDATE_INODE_NUMS 3
     50 
     51 /* Magic numbers to indicate what state the update process is in */
     52 #define MAGIC_STATE_MARKING_INUMS  0x7000151515565512ll
     53 #define MAGIC_STATE_UPDATING_INUMS 0x6121131211735123ll
     54 #define MAGIC_STATE_UPDATING_SB    0x15e1715151558477ll
     55 
     56 /* Internal state variables corresponding to the magic numbers */
     57 #define STATE_UNSET          0
     58 #define STATE_MARKING_INUMS  1
     59 #define STATE_UPDATING_INUMS 2
     60 #define STATE_UPDATING_SB    3
     61 
     62 /* Used for automated testing of this programs ability to stop and be restarted wthout error */
     63 static int bail_phase = 0;
     64 static int bail_loc = 0;
     65 static int bail_count = 0;
     66 static int count = 0;
     67 
     68 /* global flags */
     69 static int verbose = 0;
     70 static int no_write = 0;
     71 
     72 static int new_inodes_per_group = 0;
     73 
     74 static int no_write_fixup_state = 0;
     75 
     76 static int compute_new_inum(unsigned int old_inum)
     77 {
     78     unsigned int group, offset;
     79 
     80     group = (old_inum - 1) / info.inodes_per_group;
     81     offset = (old_inum -1) % info.inodes_per_group;
     82 
     83     return (group * new_inodes_per_group) + offset + 1;
     84 }
     85 
     86 /* Function to read the primary superblock */
     87 static void read_sb(int fd, struct ext4_super_block *sb)
     88 {
     89     off64_t ret;
     90 
     91     ret = lseek64(fd, 1024, SEEK_SET);
     92     if (ret < 0)
     93         critical_error_errno("failed to seek to superblock");
     94 
     95     ret = read(fd, sb, sizeof(*sb));
     96     if (ret < 0)
     97         critical_error_errno("failed to read superblock");
     98     if (ret != sizeof(*sb))
     99         critical_error("failed to read all of superblock");
    100 }
    101 
    102 /* Function to write a primary or backup superblock at a given offset */
    103 static void write_sb(int fd, unsigned long long offset, struct ext4_super_block *sb)
    104 {
    105     off64_t ret;
    106 
    107     if (no_write) {
    108         return;
    109     }
    110 
    111     ret = lseek64(fd, offset, SEEK_SET);
    112     if (ret < 0)
    113         critical_error_errno("failed to seek to superblock");
    114 
    115     ret = write(fd, sb, sizeof(*sb));
    116     if (ret < 0)
    117         critical_error_errno("failed to write superblock");
    118     if (ret != sizeof(*sb))
    119         critical_error("failed to write all of superblock");
    120 }
    121 
    122 static int get_fs_fixup_state(int fd)
    123 {
    124     unsigned long long magic;
    125     int ret, len;
    126 
    127     if (no_write) {
    128         return no_write_fixup_state;
    129     }
    130 
    131     lseek64(fd, 0, SEEK_SET);
    132     len = read(fd, &magic, sizeof(magic));
    133     if (len != sizeof(magic)) {
    134         critical_error("cannot read fixup_state\n");
    135     }
    136 
    137     switch (magic) {
    138         case MAGIC_STATE_MARKING_INUMS:
    139             ret = STATE_MARKING_INUMS;
    140             break;
    141         case MAGIC_STATE_UPDATING_INUMS:
    142             ret = STATE_UPDATING_INUMS;
    143             break;
    144         case MAGIC_STATE_UPDATING_SB:
    145             ret = STATE_UPDATING_SB;
    146             break;
    147         default:
    148             ret = STATE_UNSET;
    149     }
    150     return ret;
    151 }
    152 
    153 static int set_fs_fixup_state(int fd, int state)
    154 {
    155     unsigned long long magic;
    156     struct ext4_super_block sb;
    157     int len;
    158 
    159     if (no_write) {
    160         no_write_fixup_state = state;
    161         return 0;
    162     }
    163 
    164     switch (state) {
    165         case STATE_MARKING_INUMS:
    166             magic = MAGIC_STATE_MARKING_INUMS;
    167             break;
    168         case STATE_UPDATING_INUMS:
    169             magic = MAGIC_STATE_UPDATING_INUMS;
    170             break;
    171         case STATE_UPDATING_SB:
    172             magic = MAGIC_STATE_UPDATING_SB;
    173             break;
    174         case STATE_UNSET:
    175         default:
    176             magic = 0ll;
    177             break;
    178     }
    179 
    180     lseek64(fd, 0, SEEK_SET);
    181     len = write(fd, &magic, sizeof(magic));
    182     if (len != sizeof(magic)) {
    183         critical_error("cannot write fixup_state\n");
    184     }
    185 
    186     read_sb(fd, &sb);
    187     if (magic) {
    188         /* If we are in the process of updating the filesystem, make it unmountable */
    189         sb.s_desc_size |= 1;
    190     } else {
    191         /* we are done, so make the filesystem mountable again */
    192         sb.s_desc_size &= ~1;
    193     }
    194     write_sb(fd, 1024, &sb);
    195 
    196     return 0;
    197 }
    198 
    199 static int read_ext(int fd)
    200 {
    201     off64_t ret;
    202     struct ext4_super_block sb;
    203 
    204     read_sb(fd, &sb);
    205 
    206     ext4_parse_sb(&sb);
    207 
    208     if (info.feat_incompat & EXT4_FEATURE_INCOMPAT_RECOVER) {
    209         critical_error("Filesystem needs recovery first, mount and unmount to do that\n");
    210     }
    211 
    212     /* Clear the low bit which is set while this tool is in progress.
    213      * If the tool crashes, it will still be set when we restart.
    214      * The low bit is set to make the filesystem unmountable while
    215      * it is being fixed up.  Also allow 0, which means the old ext2
    216      * size is in use.
    217      */
    218     if (((sb.s_desc_size & ~1) != sizeof(struct ext2_group_desc)) &&
    219         ((sb.s_desc_size & ~1) != 0))
    220         critical_error("error: bg_desc_size != sizeof(struct ext2_group_desc)\n");
    221 
    222     ret = lseek64(fd, info.len, SEEK_SET);
    223     if (ret < 0)
    224         critical_error_errno("failed to seek to end of input image");
    225 
    226     ret = lseek64(fd, info.block_size * (aux_info.first_data_block + 1), SEEK_SET);
    227     if (ret < 0)
    228         critical_error_errno("failed to seek to block group descriptors");
    229 
    230     ret = read(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
    231     if (ret < 0)
    232         critical_error_errno("failed to read block group descriptors");
    233     if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
    234         critical_error("failed to read all of block group descriptors");
    235 
    236     if (verbose) {
    237         printf("Found filesystem with parameters:\n");
    238         printf("    Size: %llu\n", info.len);
    239         printf("    Block size: %d\n", info.block_size);
    240         printf("    Blocks per group: %d\n", info.blocks_per_group);
    241         printf("    Inodes per group: %d\n", info.inodes_per_group);
    242         printf("    Inode size: %d\n", info.inode_size);
    243         printf("    Label: %s\n", info.label);
    244         printf("    Blocks: %llu\n", aux_info.len_blocks);
    245         printf("    Block groups: %d\n", aux_info.groups);
    246         printf("    Reserved block group size: %d\n", info.bg_desc_reserve_blocks);
    247         printf("    Used %d/%d inodes and %d/%d blocks\n",
    248                 aux_info.sb->s_inodes_count - aux_info.sb->s_free_inodes_count,
    249                 aux_info.sb->s_inodes_count,
    250                 aux_info.sb->s_blocks_count_lo - aux_info.sb->s_free_blocks_count_lo,
    251                 aux_info.sb->s_blocks_count_lo);
    252     }
    253 
    254     return 0;
    255 }
    256 
    257 static int read_inode(int fd, unsigned int inum, struct ext4_inode *inode)
    258 {
    259     unsigned int bg_num, bg_offset;
    260     off64_t inode_offset;
    261     int len;
    262 
    263     bg_num = (inum-1) / info.inodes_per_group;
    264     bg_offset = (inum-1) % info.inodes_per_group;
    265 
    266     inode_offset = ((unsigned long long)aux_info.bg_desc[bg_num].bg_inode_table * info.block_size) +
    267                     (bg_offset * info.inode_size);
    268 
    269     if (lseek64(fd, inode_offset, SEEK_SET) < 0) {
    270         critical_error_errno("failed to seek to inode %d\n", inum);
    271     }
    272 
    273     len=read(fd, inode, sizeof(*inode));
    274     if (len != sizeof(*inode)) {
    275         critical_error_errno("failed to read inode %d\n", inum);
    276     }
    277 
    278     return 0;
    279 }
    280 
    281 static int read_block(int fd, unsigned long long block_num, void *block)
    282 {
    283     off64_t off;
    284     unsigned int len;
    285 
    286     off = block_num * info.block_size;
    287 
    288     if (lseek64(fd, off, SEEK_SET) , 0) {
    289         critical_error_errno("failed to seek to block %lld\n", block_num);
    290     }
    291 
    292     len=read(fd, block, info.block_size);
    293     if (len != info.block_size) {
    294         critical_error_errno("failed to read block %lld\n", block_num);
    295     }
    296 
    297     return 0;
    298 }
    299 
    300 static int write_block(int fd, unsigned long long block_num, void *block)
    301 {
    302     off64_t off;
    303     unsigned int len;
    304 
    305     if (no_write) {
    306         return 0;
    307     }
    308 
    309     off = block_num * info.block_size;
    310 
    311     if (lseek64(fd, off, SEEK_SET) < 0) {
    312         critical_error_errno("failed to seek to block %lld\n", block_num);
    313     }
    314 
    315     len=write(fd, block, info.block_size);
    316     if (len != info.block_size) {
    317         critical_error_errno("failed to write block %lld\n", block_num);
    318     }
    319 
    320     return 0;
    321 }
    322 
    323 static int bitmap_get_bit(u8 *bitmap, u32 bit)
    324 {
    325         if (bitmap[bit / 8] & (1 << (bit % 8)))
    326                 return 1;
    327 
    328         return 0;
    329 }
    330 
    331 static void bitmap_clear_bit(u8 *bitmap, u32 bit)
    332 {
    333         bitmap[bit / 8] &= ~(1 << (bit % 8));
    334 
    335         return;
    336 }
    337 
    338 static void check_inode_bitmap(int fd, unsigned int bg_num)
    339 {
    340     unsigned int inode_bitmap_block_num;
    341     unsigned char block[MAX_EXT4_BLOCK_SIZE];
    342     int i, bitmap_updated = 0;
    343 
    344     /* Using the bg_num, aux_info.bg_desc[], info.inodes_per_group and
    345      * new_inodes_per_group, retrieve the inode bitmap, and make sure
    346      * the bits between the old and new size are clear
    347      */
    348     inode_bitmap_block_num = aux_info.bg_desc[bg_num].bg_inode_bitmap;
    349 
    350     read_block(fd, inode_bitmap_block_num, block);
    351 
    352     for (i = info.inodes_per_group; i < new_inodes_per_group; i++) {
    353         if (bitmap_get_bit(block, i)) {
    354             bitmap_clear_bit(block, i);
    355             bitmap_updated = 1;
    356         }
    357     }
    358 
    359     if (bitmap_updated) {
    360         if (verbose) {
    361             printf("Warning: updated inode bitmap for block group %d\n", bg_num);
    362         }
    363         write_block(fd, inode_bitmap_block_num, block);
    364     }
    365 
    366     return;
    367 }
    368 
    369 /* Update the superblock and bgdesc of the specified block group */
    370 static int update_superblocks_and_bg_desc(int fd, int state)
    371 {
    372     off64_t ret;
    373     struct ext4_super_block sb;
    374     unsigned int num_block_groups, total_new_inodes;
    375     unsigned int i;
    376 
    377 
    378     read_sb(fd, &sb);
    379 
    380     /* Compute how many more inodes are now available */
    381     num_block_groups = DIV_ROUND_UP(aux_info.len_blocks, info.blocks_per_group);
    382     total_new_inodes = num_block_groups * (new_inodes_per_group - sb.s_inodes_per_group);
    383 
    384     if (verbose) {
    385         printf("created %d additional inodes\n", total_new_inodes);
    386     }
    387 
    388     /* Update the free inodes count in each block group descriptor */
    389     for (i = 0; i < num_block_groups; i++) {
    390        if (state == STATE_UPDATING_SB) {
    391            aux_info.bg_desc[i].bg_free_inodes_count += (new_inodes_per_group - sb.s_inodes_per_group);
    392        }
    393        check_inode_bitmap(fd, i);
    394     }
    395 
    396     /* First some sanity checks */
    397     if ((sb.s_inodes_count + total_new_inodes) != (new_inodes_per_group * num_block_groups)) {
    398         critical_error("Failed sanity check on new inode count\n");
    399     }
    400     if (new_inodes_per_group % (info.block_size/info.inode_size)) {
    401         critical_error("Failed sanity check on new inode per group alignment\n");
    402     }
    403 
    404     /* Update the free inodes count in the superblock */
    405     sb.s_inodes_count += total_new_inodes;
    406     sb.s_free_inodes_count += total_new_inodes;
    407     sb.s_inodes_per_group = new_inodes_per_group;
    408 
    409     for (i = 0; i < aux_info.groups; i++) {
    410         if (ext4_bg_has_super_block(i)) {
    411             unsigned int sb_offset;
    412 
    413             if (i == 0) {
    414               /* The first superblock is offset by 1K to leave room for boot sectors */
    415               sb_offset = 1024;
    416             } else {
    417               sb_offset = 0;
    418             }
    419 
    420             sb.s_block_group_nr = i;
    421             /* Don't write out the backup superblocks with the bit set in the s_desc_size
    422              * which prevents the filesystem from mounting.  The bit for the primary
    423              * superblock will be cleared on the final call to set_fs_fixup_state() */
    424             if (i != 0) {
    425                 sb.s_desc_size &= ~1;
    426             }
    427 
    428             write_sb(fd, (unsigned long long)i * info.blocks_per_group * info.block_size + sb_offset, &sb);
    429 
    430             ret = lseek64(fd, ((unsigned long long)i * info.blocks_per_group * info.block_size) +
    431                               (info.block_size * (aux_info.first_data_block + 1)), SEEK_SET);
    432             if (ret < 0)
    433                 critical_error_errno("failed to seek to block group descriptors");
    434 
    435             if (!no_write) {
    436                 ret = write(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
    437                 if (ret < 0)
    438                     critical_error_errno("failed to write block group descriptors");
    439                 if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
    440                     critical_error("failed to write all of block group descriptors");
    441             }
    442         }
    443         if ((bail_phase == 4) && ((unsigned int)bail_count == i)) {
    444             critical_error("bailing at phase 4\n");
    445         }
    446     }
    447 
    448     return 0;
    449 }
    450 
    451 
    452 static int get_direct_blocks(struct ext4_inode *inode, unsigned long long *block_list,
    453                                                        unsigned int *count)
    454 {
    455     unsigned int i = 0;
    456     unsigned int ret = 0;
    457     unsigned int sectors_per_block;
    458 
    459     sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
    460     while ((i < (inode->i_blocks_lo / sectors_per_block)) && (i < EXT4_NDIR_BLOCKS)) {
    461         block_list[i] = inode->i_block[i];
    462         i++;
    463     }
    464 
    465     *count += i;
    466 
    467     if ((inode->i_blocks_lo / sectors_per_block) > EXT4_NDIR_BLOCKS) {
    468         ret = 1;
    469     }
    470 
    471     return ret;
    472 }
    473 
    474 static int get_indirect_blocks(int fd, struct ext4_inode *inode,
    475                                unsigned long long *block_list, unsigned int *count)
    476 {
    477     unsigned int i;
    478     unsigned int *indirect_block;
    479     unsigned int sectors_per_block;
    480 
    481     sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
    482 
    483     indirect_block = (unsigned int *)malloc(info.block_size);
    484     if (indirect_block == 0) {
    485         critical_error("failed to allocate memory for indirect_block\n");
    486     }
    487 
    488     read_block(fd, inode->i_block[EXT4_NDIR_BLOCKS], indirect_block);
    489 
    490     for(i = 0; i < (inode->i_blocks_lo / sectors_per_block - EXT4_NDIR_BLOCKS); i++) {
    491        block_list[EXT4_NDIR_BLOCKS+i] = indirect_block[i];
    492     }
    493 
    494     *count += i;
    495 
    496     free(indirect_block);
    497 
    498     return 0;
    499 }
    500 
    501 static int get_block_list_indirect(int fd, struct ext4_inode *inode, unsigned long long *block_list)
    502 {
    503     unsigned int count=0;
    504 
    505     if (get_direct_blocks(inode, block_list, &count)) {
    506         get_indirect_blocks(fd, inode, block_list, &count);
    507     }
    508 
    509     return count;
    510 }
    511 
    512 static int get_extent_ents(struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
    513 {
    514     int i, j;
    515     struct ext4_extent *extent;
    516     off64_t fs_block_num;
    517 
    518     if (ext_hdr->eh_depth != 0) {
    519         critical_error("get_extent_ents called with eh_depth != 0\n");
    520     }
    521 
    522     /* The extent entries immediately follow the header, so add 1 to the pointer
    523      * and cast it to an extent pointer.
    524      */
    525     extent = (struct ext4_extent *)(ext_hdr + 1);
    526 
    527     for (i = 0; i < ext_hdr->eh_entries; i++) {
    528          fs_block_num = ((off64_t)extent->ee_start_hi << 32) | extent->ee_start_lo;
    529          for (j = 0; j < extent->ee_len; j++) {
    530              block_list[extent->ee_block+j] = fs_block_num+j;
    531          }
    532          extent++;
    533     }
    534 
    535     return 0;
    536 }
    537 
    538 static int get_extent_idx(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
    539 {
    540     int i;
    541     struct ext4_extent_idx *extent_idx;
    542     struct ext4_extent_header *tmp_ext_hdr;
    543     off64_t fs_block_num;
    544     unsigned char block[MAX_EXT4_BLOCK_SIZE];
    545 
    546     /* Sanity check */
    547     if (ext_hdr->eh_depth == 0) {
    548         critical_error("get_extent_idx called with eh_depth == 0\n");
    549     }
    550 
    551     /* The extent entries immediately follow the header, so add 1 to the pointer
    552      * and cast it to an extent pointer.
    553      */
    554     extent_idx = (struct ext4_extent_idx *)(ext_hdr + 1);
    555 
    556     for (i = 0; i < ext_hdr->eh_entries; i++) {
    557          fs_block_num = ((off64_t)extent_idx->ei_leaf_hi << 32) | extent_idx->ei_leaf_lo;
    558          read_block(fd, fs_block_num, block);
    559          tmp_ext_hdr = (struct ext4_extent_header *)block;
    560 
    561          if (tmp_ext_hdr->eh_depth == 0) {
    562              get_extent_ents(tmp_ext_hdr, block_list); /* leaf node, fill in block_list */
    563          } else {
    564              get_extent_idx(fd, tmp_ext_hdr, block_list); /* recurse down the tree */
    565          }
    566     }
    567 
    568     return 0;
    569 }
    570 
    571 static int get_block_list_extents(int fd, struct ext4_inode *inode, unsigned long long *block_list)
    572 {
    573     struct ext4_extent_header *extent_hdr;
    574 
    575     extent_hdr = (struct ext4_extent_header *)inode->i_block;
    576 
    577     if (extent_hdr->eh_magic != EXT4_EXT_MAGIC) {
    578         critical_error("extent header has unexpected magic value 0x%4.4x\n",
    579                        extent_hdr->eh_magic);
    580     }
    581 
    582     if (extent_hdr->eh_depth == 0) {
    583          get_extent_ents((struct ext4_extent_header *)inode->i_block, block_list);
    584          return 0;
    585     }
    586 
    587     get_extent_idx(fd, (struct ext4_extent_header *)inode->i_block, block_list);
    588 
    589     return 0;
    590 }
    591 
    592 static int is_entry_dir(int fd, struct ext4_dir_entry_2 *dirp, int pass)
    593 {
    594     struct ext4_inode inode;
    595     int ret = 0;
    596 
    597     if (dirp->file_type == EXT4_FT_DIR) {
    598         ret = 1;
    599     } else if (dirp->file_type == EXT4_FT_UNKNOWN) {
    600         /* Somebody was too lazy to fill in the dir entry,
    601          * so we have to go fetch it from the inode. Grrr.
    602          */
    603         /* if UPDATE_INODE_NUMS pass and the inode high bit is not
    604          * set return false so we don't recurse down the tree that is
    605          * already updated.  Otherwise, fetch inode, and return answer.
    606          */
    607         if ((pass == UPDATE_INODE_NUMS) && !(dirp->inode & 0x80000000)) {
    608             ret = 0;
    609         } else {
    610             read_inode(fd, (dirp->inode & 0x7fffffff), &inode);
    611             if (S_ISDIR(inode.i_mode)) {
    612                 ret = 1;
    613             }
    614         }
    615     }
    616 
    617     return ret;
    618 }
    619 
    620 static int recurse_dir(int fd, struct ext4_inode *inode, char *dirbuf, int dirsize, int mode)
    621 {
    622     unsigned long long *block_list;
    623     unsigned int num_blocks;
    624     struct ext4_dir_entry_2 *dirp, *prev_dirp = 0;
    625     char name[256];
    626     unsigned int i, leftover_space, is_dir;
    627     struct ext4_inode tmp_inode;
    628     int tmp_dirsize;
    629     char *tmp_dirbuf;
    630 
    631     switch (mode) {
    632         case SANITY_CHECK_PASS:
    633         case MARK_INODE_NUMS:
    634         case UPDATE_INODE_NUMS:
    635             break;
    636         default:
    637             critical_error("recurse_dir() called witn unknown mode!\n");
    638     }
    639 
    640     if (dirsize % info.block_size) {
    641         critical_error("dirsize %d not a multiple of block_size %d.  This is unexpected!\n",
    642                 dirsize, info.block_size);
    643     }
    644 
    645     num_blocks = dirsize / info.block_size;
    646 
    647     block_list = malloc((num_blocks + 1) * sizeof(*block_list));
    648     if (block_list == 0) {
    649         critical_error("failed to allocate memory for block_list\n");
    650     }
    651 
    652     if (inode->i_flags & EXT4_EXTENTS_FL) {
    653         get_block_list_extents(fd, inode, block_list);
    654     } else {
    655         /* A directory that requires doubly or triply indirect blocks in huge indeed,
    656          * and will almost certainly not exist, especially since make_ext4fs only creates
    657          * directories with extents, and the kernel will too, but check to make sure the
    658          * directory is not that big and give an error if so.  Our limit is 12 direct blocks,
    659          * plus block_size/4 singly indirect blocks, which for a filesystem with 4K blocks
    660          * is a directory 1036 blocks long, or 4,243,456 bytes long!  Assuming an average
    661          * filename length of 20 (which I think is generous) thats 20 + 8 bytes overhead
    662          * per entry, or 151,552 entries in the directory!
    663          */
    664         if (num_blocks > (info.block_size / 4 + EXT4_NDIR_BLOCKS)) {
    665             critical_error("Non-extent based directory is too big!\n");
    666         }
    667         get_block_list_indirect(fd, inode, block_list);
    668     }
    669 
    670     /* Read in all the blocks for this directory */
    671     for (i = 0; i < num_blocks; i++) {
    672         read_block(fd, block_list[i], dirbuf + (i * info.block_size));
    673     }
    674 
    675     dirp = (struct ext4_dir_entry_2 *)dirbuf;
    676     while (dirp < (struct ext4_dir_entry_2 *)(dirbuf + dirsize)) {
    677         count++;
    678         leftover_space = (char *)(dirbuf + dirsize) - (char *)dirp;
    679         if (((mode == SANITY_CHECK_PASS) || (mode == UPDATE_INODE_NUMS)) &&
    680             (leftover_space <= 8) && prev_dirp) {
    681             /* This is a bug in an older version of make_ext4fs, where it
    682              * didn't properly include the rest of the block in rec_len.
    683              * Update rec_len on the previous entry to include the rest of
    684              * the block and exit the loop.
    685              */
    686             if (verbose) {
    687                 printf("fixing up short rec_len for diretory entry for %s\n", name);
    688             }
    689             prev_dirp->rec_len += leftover_space;
    690             break;
    691         }
    692 
    693         if (dirp->inode == 0) {
    694             /* This is the last entry in the directory */
    695             break;
    696         }
    697 
    698         strncpy(name, dirp->name, dirp->name_len);
    699         name[dirp->name_len]='\0';
    700 
    701         /* Only recurse on pass UPDATE_INODE_NUMS if the high bit is set.
    702          * Otherwise, this inode entry has already been updated
    703          * and we'll do the wrong thing.  Also don't recurse on . or ..,
    704          * and certainly not on non-directories!
    705          */
    706         /* Hrm, looks like filesystems made by fastboot on stingray set the file_type
    707          * flag, but the lost+found directory has the type set to Unknown, which
    708          * seems to imply I need to read the inode and get it.
    709          */
    710         is_dir = is_entry_dir(fd, dirp, mode);
    711         if ( is_dir && (strcmp(name, ".") && strcmp(name, "..")) &&
    712             ((mode == SANITY_CHECK_PASS) || (mode == MARK_INODE_NUMS) ||
    713               ((mode == UPDATE_INODE_NUMS) && (dirp->inode & 0x80000000))) ) {
    714             /* A directory!  Recurse! */
    715             read_inode(fd, dirp->inode & 0x7fffffff, &tmp_inode);
    716 
    717             if (!S_ISDIR(tmp_inode.i_mode)) {
    718                 critical_error("inode %d for name %s does not point to a directory\n",
    719                         dirp->inode & 0x7fffffff, name);
    720             }
    721             if (verbose) {
    722                 printf("inode %d %s use extents\n", dirp->inode & 0x7fffffff,
    723                        (tmp_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
    724             }
    725 
    726             tmp_dirsize = tmp_inode.i_blocks_lo * INODE_BLOCK_SIZE;
    727             if (verbose) {
    728                 printf("dir size = %d bytes\n", tmp_dirsize);
    729             }
    730 
    731             tmp_dirbuf = malloc(tmp_dirsize);
    732             if (tmp_dirbuf == 0) {
    733                 critical_error("failed to allocate memory for tmp_dirbuf\n");
    734             }
    735 
    736             recurse_dir(fd, &tmp_inode, tmp_dirbuf, tmp_dirsize, mode);
    737 
    738             free(tmp_dirbuf);
    739         }
    740 
    741         if (verbose) {
    742             if (is_dir) {
    743                 printf("Directory %s\n", name);
    744             } else {
    745                 printf("Non-directory %s\n", name);
    746             }
    747         }
    748 
    749         /* Process entry based on current mode.  Either set high bit or change inode number */
    750         if (mode == MARK_INODE_NUMS) {
    751             dirp->inode |= 0x80000000;
    752         } else if (mode == UPDATE_INODE_NUMS) {
    753             if (dirp->inode & 0x80000000) {
    754                 dirp->inode = compute_new_inum(dirp->inode & 0x7fffffff);
    755             }
    756         }
    757 
    758         if ((bail_phase == mode) && (bail_loc == 1) && (bail_count == count)) {
    759             critical_error("Bailing at phase %d, loc 1 and count %d\n", mode, count);
    760         }
    761 
    762         /* Point dirp at the next entry */
    763         prev_dirp = dirp;
    764         dirp = (struct ext4_dir_entry_2*)((char *)dirp + dirp->rec_len);
    765     }
    766 
    767     /* Write out all the blocks for this directory */
    768     for (i = 0; i < num_blocks; i++) {
    769         write_block(fd, block_list[i], dirbuf + (i * info.block_size));
    770         if ((bail_phase == mode) && (bail_loc == 2) && (bail_count <= count)) {
    771             critical_error("Bailing at phase %d, loc 2 and count %d\n", mode, count);
    772         }
    773     }
    774 
    775     free(block_list);
    776 
    777     return 0;
    778 }
    779 
    780 int ext4fixup(char *fsdev)
    781 {
    782     return ext4fixup_internal(fsdev, 0, 0, 0, 0, 0);
    783 }
    784 
    785 int ext4fixup_internal(char *fsdev, int v_flag, int n_flag,
    786                        int stop_phase, int stop_loc, int stop_count)
    787 {
    788     int fd;
    789     struct ext4_inode root_inode;
    790     unsigned int dirsize;
    791     char *dirbuf;
    792 
    793     if (setjmp(setjmp_env))
    794         return EXIT_FAILURE; /* Handle a call to longjmp() */
    795 
    796     verbose = v_flag;
    797     no_write = n_flag;
    798 
    799     bail_phase = stop_phase;
    800     bail_loc = stop_loc;
    801     bail_count = stop_count;
    802 
    803     fd = open(fsdev, O_RDWR);
    804 
    805     if (fd < 0)
    806         critical_error_errno("failed to open filesystem image");
    807 
    808     read_ext(fd);
    809 
    810     if ((info.feat_incompat & EXT4_FEATURE_INCOMPAT_FILETYPE) == 0) {
    811         critical_error("Expected filesystem to have filetype flag set\n");
    812     }
    813 
    814 #if 0 // If we have to fix the directory rec_len issue, we can't use this check
    815     /* Check to see if the inodes/group is copacetic */
    816     if (info.inodes_per_blockgroup % (info.block_size/info.inode_size) == 0) {
    817              /* This filesystem has either already been updated, or was
    818               * made correctly.
    819               */
    820              if (verbose) {
    821                  printf("%s: filesystem correct, no work to do\n", me);
    822              }
    823              exit(0);
    824     }
    825 #endif
    826 
    827     /* Compute what the new value of inodes_per_blockgroup will be when we're done */
    828     new_inodes_per_group=ALIGN(info.inodes_per_group,(info.block_size/info.inode_size));
    829 
    830     read_inode(fd, EXT4_ROOT_INO, &root_inode);
    831 
    832     if (!S_ISDIR(root_inode.i_mode)) {
    833         critical_error("root inode %d does not point to a directory\n", EXT4_ROOT_INO);
    834     }
    835     if (verbose) {
    836         printf("inode %d %s use extents\n", EXT4_ROOT_INO,
    837                (root_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
    838     }
    839 
    840     dirsize = root_inode.i_blocks_lo * INODE_BLOCK_SIZE;
    841     if (verbose) {
    842         printf("root dir size = %d bytes\n", dirsize);
    843     }
    844 
    845     dirbuf = malloc(dirsize);
    846     if (dirbuf == 0) {
    847         critical_error("failed to allocate memory for dirbuf\n");
    848     }
    849 
    850     /* Perform a sanity check pass first, try to catch any errors that will occur
    851      * before we actually change anything, so we don't leave a filesystem in a
    852      * corrupted, unrecoverable state.  Set no_write, make it quiet, and do a recurse
    853      * pass and a update_superblock pass.  Set flags back to requested state when done.
    854      * Only perform sanity check if the state is unset.  If the state is _NOT_ unset,
    855      * then the tool has already been run and interrupted, and it presumably ran and
    856      * passed sanity checked before it got interrupted.  It is _NOT_ safe to run sanity
    857      * check if state is unset because it assumes inodes are to be computed using the
    858      * old inodes/group, but some inode numbers may be updated to the new number.
    859      */
    860     if (get_fs_fixup_state(fd) == STATE_UNSET) {
    861         verbose = 0;
    862         no_write = 1;
    863         recurse_dir(fd, &root_inode, dirbuf, dirsize, SANITY_CHECK_PASS);
    864         update_superblocks_and_bg_desc(fd, STATE_UNSET);
    865         verbose = v_flag;
    866         no_write = n_flag;
    867 
    868         set_fs_fixup_state(fd, STATE_MARKING_INUMS);
    869     }
    870 
    871     if (get_fs_fixup_state(fd) == STATE_MARKING_INUMS) {
    872         count = 0; /* Reset debugging counter */
    873         if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, MARK_INODE_NUMS)) {
    874             set_fs_fixup_state(fd, STATE_UPDATING_INUMS);
    875         }
    876     }
    877 
    878     if (get_fs_fixup_state(fd) == STATE_UPDATING_INUMS) {
    879         count = 0; /* Reset debugging counter */
    880         if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, UPDATE_INODE_NUMS)) {
    881             set_fs_fixup_state(fd, STATE_UPDATING_SB);
    882         }
    883     }
    884 
    885     if (get_fs_fixup_state(fd) == STATE_UPDATING_SB) {
    886         /* set the new inodes/blockgroup number,
    887          * and sets the state back to 0.
    888          */
    889         if (!update_superblocks_and_bg_desc(fd, STATE_UPDATING_SB)) {
    890             set_fs_fixup_state(fd, STATE_UNSET);
    891         }
    892     }
    893 
    894     close(fd);
    895 
    896     return 0;
    897 }
    898