Ventoy/SQUASHFS/squashfs-tools-4.4/squashfs-tools/read_fs.c
2020-04-05 00:08:01 +08:00

997 lines
29 KiB
C

/*
* Read a squashfs filesystem. This is a highly compressed read only
* filesystem.
*
* Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010,
* 2012, 2013, 2014, 2019
* Phillip Lougher <phillip@squashfs.org.uk>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2,
* or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*
* read_fs.c
*/
#define TRUE 1
#define FALSE 0
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <string.h>
#include <sys/mman.h>
#include <limits.h>
#include <dirent.h>
#ifndef linux
#define __BYTE_ORDER BYTE_ORDER
#define __BIG_ENDIAN BIG_ENDIAN
#define __LITTLE_ENDIAN LITTLE_ENDIAN
#else
#include <endian.h>
#endif
#include <stdlib.h>
#include "squashfs_fs.h"
#include "squashfs_swap.h"
#include "compressor.h"
#include "xattr.h"
#include "error.h"
#include "mksquashfs.h"
int read_block(int fd, long long start, long long *next, int expected,
void *block)
{
unsigned short c_byte;
int res, compressed;
int outlen = expected ? expected : SQUASHFS_METADATA_SIZE;
/* Read block size */
res = read_fs_bytes(fd, start, 2, &c_byte);
if(res == 0)
return 0;
SQUASHFS_INSWAP_SHORTS(&c_byte, 1);
compressed = SQUASHFS_COMPRESSED(c_byte);
c_byte = SQUASHFS_COMPRESSED_SIZE(c_byte);
/*
* The block size should not be larger than
* the uncompressed size (or max uncompressed size if
* expected is 0)
*/
if (c_byte > outlen)
return 0;
if(compressed) {
char buffer[c_byte];
int error;
res = read_fs_bytes(fd, start + 2, c_byte, buffer);
if(res == 0)
return 0;
res = compressor_uncompress(comp, block, buffer, c_byte,
outlen, &error);
if(res == -1) {
ERROR("%s uncompress failed with error code %d\n",
comp->name, error);
return 0;
}
} else {
res = read_fs_bytes(fd, start + 2, c_byte, block);
if(res == 0)
return 0;
res = c_byte;
}
if(next)
*next = start + 2 + c_byte;
/*
* if expected, then check the (uncompressed) return data
* is of the expected size
*/
if(expected && expected != res)
return 0;
else
return res;
}
#define NO_BYTES(SIZE) \
(bytes - (cur_ptr - inode_table) < (SIZE))
#define NO_INODE_BYTES(INODE) NO_BYTES(sizeof(struct INODE))
unsigned char *scan_inode_table(int fd, long long start, long long end,
long long root_inode_start, int root_inode_offset,
struct squashfs_super_block *sBlk, union squashfs_inode_header
*dir_inode, unsigned int *root_inode_block, unsigned int
*root_inode_size, long long *uncompressed_file, unsigned int
*uncompressed_directory, int *file_count, int *sym_count, int
*dev_count, int *dir_count, int *fifo_count, int *sock_count,
unsigned int *id_table)
{
unsigned char *cur_ptr;
unsigned char *inode_table = NULL;
int byte, files = 0;
unsigned int directory_start_block, bytes = 0, size = 0;
struct squashfs_base_inode_header base;
TRACE("scan_inode_table: start 0x%llx, end 0x%llx, root_inode_start "
"0x%llx\n", start, end, root_inode_start);
*root_inode_block = UINT_MAX;
while(start < end) {
if(start == root_inode_start) {
TRACE("scan_inode_table: read compressed block 0x%llx "
"containing root inode\n", start);
*root_inode_block = bytes;
}
if(size - bytes < SQUASHFS_METADATA_SIZE) {
inode_table = realloc(inode_table, size
+= SQUASHFS_METADATA_SIZE);
if(inode_table == NULL)
MEM_ERROR();
}
TRACE("scan_inode_table: reading block 0x%llx\n", start);
byte = read_block(fd, start, &start, 0, inode_table + bytes);
if(byte == 0)
goto corrupted;
bytes += byte;
/* If this is not the last metadata block in the inode table
* then it should be SQUASHFS_METADATA_SIZE in size.
* Note, we can't use expected in read_block() above for this
* because we don't know if this is the last block until
* after reading.
*/
if(start != end && byte != SQUASHFS_METADATA_SIZE)
goto corrupted;
}
/*
* We expect to have found the metadata block containing the
* root inode in the above inode_table metadata block scan. If it
* hasn't been found then the filesystem is corrupted
*/
if(*root_inode_block == UINT_MAX)
goto corrupted;
/*
* The number of bytes available after the root inode medata block
* should be at least the root inode offset + the size of a
* regular directory inode, if not the filesystem is corrupted
*
* +-----------------------+-----------------------+
* | | directory |
* | | inode |
* +-----------------------+-----------------------+
* ^ ^ ^
* *root_inode_block root_inode_offset bytes
*/
if((bytes - *root_inode_block) < (root_inode_offset +
sizeof(struct squashfs_dir_inode_header)))
goto corrupted;
/*
* Read last inode entry which is the root directory inode, and obtain
* the last directory start block index. This is used when calculating
* the total uncompressed directory size. The directory bytes in the
* last * block will be counted as normal.
*
* Note, the previous check ensures the following calculation won't
* underflow, and we won't access beyond the buffer
*/
*root_inode_size = bytes - (*root_inode_block + root_inode_offset);
bytes = *root_inode_block + root_inode_offset;
SQUASHFS_SWAP_DIR_INODE_HEADER(inode_table + bytes, &dir_inode->dir);
if(dir_inode->base.inode_type == SQUASHFS_DIR_TYPE)
directory_start_block = dir_inode->dir.start_block;
else if(dir_inode->base.inode_type == SQUASHFS_LDIR_TYPE) {
if(*root_inode_size < sizeof(struct squashfs_ldir_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_LDIR_INODE_HEADER(inode_table + bytes,
&dir_inode->ldir);
directory_start_block = dir_inode->ldir.start_block;
} else
/* bad type, corrupted filesystem */
goto corrupted;
get_uid(id_table[dir_inode->base.uid]);
get_guid(id_table[dir_inode->base.guid]);
/* allocate fragment to file mapping table */
file_mapping = calloc(sBlk->fragments, sizeof(struct append_file *));
if(file_mapping == NULL)
MEM_ERROR();
for(cur_ptr = inode_table; cur_ptr < inode_table + bytes; files ++) {
if(NO_INODE_BYTES(squashfs_base_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_BASE_INODE_HEADER(cur_ptr, &base);
TRACE("scan_inode_table: processing inode @ byte position "
"0x%x, type 0x%x\n",
(unsigned int) (cur_ptr - inode_table),
base.inode_type);
get_uid(id_table[base.uid]);
get_guid(id_table[base.guid]);
switch(base.inode_type) {
case SQUASHFS_FILE_TYPE: {
struct squashfs_reg_inode_header inode;
int frag_bytes, blocks, i;
long long start, file_bytes = 0;
unsigned int *block_list;
if(NO_INODE_BYTES(squashfs_reg_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_REG_INODE_HEADER(cur_ptr, &inode);
frag_bytes = inode.fragment == SQUASHFS_INVALID_FRAG ?
0 : inode.file_size % sBlk->block_size;
blocks = inode.fragment == SQUASHFS_INVALID_FRAG ?
(inode.file_size + sBlk->block_size - 1) >>
sBlk->block_log : inode.file_size >>
sBlk->block_log;
start = inode.start_block;
TRACE("scan_inode_table: regular file, file_size %d, "
"blocks %d\n", inode.file_size, blocks);
if(NO_BYTES(blocks * sizeof(unsigned int)))
/* corrupted filesystem */
goto corrupted;
block_list = malloc(blocks * sizeof(unsigned int));
if(block_list == NULL)
MEM_ERROR();
cur_ptr += sizeof(inode);
SQUASHFS_SWAP_INTS(cur_ptr, block_list, blocks);
*uncompressed_file += inode.file_size;
(*file_count) ++;
for(i = 0; i < blocks; i++)
file_bytes +=
SQUASHFS_COMPRESSED_SIZE_BLOCK
(block_list[i]);
if(inode.fragment != SQUASHFS_INVALID_FRAG &&
inode.fragment >= sBlk->fragments) {
free(block_list);
goto corrupted;
}
add_file(start, inode.file_size, file_bytes,
block_list, blocks, inode.fragment,
inode.offset, frag_bytes);
cur_ptr += blocks * sizeof(unsigned int);
break;
}
case SQUASHFS_LREG_TYPE: {
struct squashfs_lreg_inode_header inode;
int frag_bytes, blocks, i;
long long start, file_bytes = 0;
unsigned int *block_list;
if(NO_INODE_BYTES(squashfs_lreg_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_LREG_INODE_HEADER(cur_ptr, &inode);
frag_bytes = inode.fragment == SQUASHFS_INVALID_FRAG ?
0 : inode.file_size % sBlk->block_size;
blocks = inode.fragment == SQUASHFS_INVALID_FRAG ?
(inode.file_size + sBlk->block_size - 1) >>
sBlk->block_log : inode.file_size >>
sBlk->block_log;
start = inode.start_block;
TRACE("scan_inode_table: extended regular "
"file, file_size %lld, blocks %d\n",
inode.file_size, blocks);
if(NO_BYTES(blocks * sizeof(unsigned int)))
/* corrupted filesystem */
goto corrupted;
block_list = malloc(blocks * sizeof(unsigned int));
if(block_list == NULL)
MEM_ERROR();
cur_ptr += sizeof(inode);
SQUASHFS_SWAP_INTS(cur_ptr, block_list, blocks);
*uncompressed_file += inode.file_size;
(*file_count) ++;
for(i = 0; i < blocks; i++)
file_bytes +=
SQUASHFS_COMPRESSED_SIZE_BLOCK
(block_list[i]);
if(inode.fragment != SQUASHFS_INVALID_FRAG &&
inode.fragment >= sBlk->fragments) {
free(block_list);
goto corrupted;
}
add_file(start, inode.file_size, file_bytes,
block_list, blocks, inode.fragment,
inode.offset, frag_bytes);
cur_ptr += blocks * sizeof(unsigned int);
break;
}
case SQUASHFS_SYMLINK_TYPE:
case SQUASHFS_LSYMLINK_TYPE: {
struct squashfs_symlink_inode_header inode;
if(NO_INODE_BYTES(squashfs_symlink_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_SYMLINK_INODE_HEADER(cur_ptr, &inode);
(*sym_count) ++;
if (inode.inode_type == SQUASHFS_LSYMLINK_TYPE) {
if(NO_BYTES(inode.symlink_size +
sizeof(unsigned int)))
/* corrupted filesystem */
goto corrupted;
cur_ptr += sizeof(inode) + inode.symlink_size +
sizeof(unsigned int);
} else {
if(NO_BYTES(inode.symlink_size))
/* corrupted filesystem */
goto corrupted;
cur_ptr += sizeof(inode) + inode.symlink_size;
}
break;
}
case SQUASHFS_DIR_TYPE: {
struct squashfs_dir_inode_header dir_inode;
if(NO_INODE_BYTES(squashfs_dir_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_DIR_INODE_HEADER(cur_ptr, &dir_inode);
if(dir_inode.start_block < directory_start_block)
*uncompressed_directory += dir_inode.file_size;
(*dir_count) ++;
cur_ptr += sizeof(struct squashfs_dir_inode_header);
break;
}
case SQUASHFS_LDIR_TYPE: {
struct squashfs_ldir_inode_header dir_inode;
int i;
if(NO_INODE_BYTES(squashfs_ldir_inode_header))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_LDIR_INODE_HEADER(cur_ptr, &dir_inode);
if(dir_inode.start_block < directory_start_block)
*uncompressed_directory += dir_inode.file_size;
(*dir_count) ++;
cur_ptr += sizeof(struct squashfs_ldir_inode_header);
for(i = 0; i < dir_inode.i_count; i++) {
struct squashfs_dir_index index;
if(NO_BYTES(sizeof(index)))
/* corrupted filesystem */
goto corrupted;
SQUASHFS_SWAP_DIR_INDEX(cur_ptr, &index);
if(NO_BYTES(index.size + 1))
/* corrupted filesystem */
goto corrupted;
cur_ptr += sizeof(index) + index.size + 1;
}
break;
}
case SQUASHFS_BLKDEV_TYPE:
case SQUASHFS_CHRDEV_TYPE:
if(NO_INODE_BYTES(squashfs_dev_inode_header))
/* corrupted filesystem */
goto corrupted;
(*dev_count) ++;
cur_ptr += sizeof(struct squashfs_dev_inode_header);
break;
case SQUASHFS_LBLKDEV_TYPE:
case SQUASHFS_LCHRDEV_TYPE:
if(NO_INODE_BYTES(squashfs_ldev_inode_header))
/* corrupted filesystem */
goto corrupted;
(*dev_count) ++;
cur_ptr += sizeof(struct squashfs_ldev_inode_header);
break;
case SQUASHFS_FIFO_TYPE:
if(NO_INODE_BYTES(squashfs_ipc_inode_header))
/* corrupted filesystem */
goto corrupted;
(*fifo_count) ++;
cur_ptr += sizeof(struct squashfs_ipc_inode_header);
break;
case SQUASHFS_LFIFO_TYPE:
if(NO_INODE_BYTES(squashfs_lipc_inode_header))
/* corrupted filesystem */
goto corrupted;
(*fifo_count) ++;
cur_ptr += sizeof(struct squashfs_lipc_inode_header);
break;
case SQUASHFS_SOCKET_TYPE:
if(NO_INODE_BYTES(squashfs_ipc_inode_header))
/* corrupted filesystem */
goto corrupted;
(*sock_count) ++;
cur_ptr += sizeof(struct squashfs_ipc_inode_header);
break;
case SQUASHFS_LSOCKET_TYPE:
if(NO_INODE_BYTES(squashfs_lipc_inode_header))
/* corrupted filesystem */
goto corrupted;
(*sock_count) ++;
cur_ptr += sizeof(struct squashfs_lipc_inode_header);
break;
default:
ERROR("Unknown inode type %d in scan_inode_table!\n",
base.inode_type);
goto corrupted;
}
}
printf("Read existing filesystem, %d inodes scanned\n", files);
return inode_table;
corrupted:
ERROR("scan_inode_table: filesystem corruption detected in "
"scanning metadata\n");
free(inode_table);
return NULL;
}
struct compressor *read_super(int fd, struct squashfs_super_block *sBlk, char *source)
{
int res, bytes = 0;
char buffer[SQUASHFS_METADATA_SIZE] __attribute__ ((aligned));
res = read_fs_bytes(fd, SQUASHFS_START, sizeof(struct squashfs_super_block),
sBlk);
if(res == 0) {
ERROR("Can't find a SQUASHFS superblock on %s\n",
source);
ERROR("Wrong filesystem or filesystem is corrupted!\n");
goto failed_mount;
}
SQUASHFS_INSWAP_SUPER_BLOCK(sBlk);
if(sBlk->s_magic != SQUASHFS_MAGIC) {
if(sBlk->s_magic == SQUASHFS_MAGIC_SWAP)
ERROR("Pre 4.0 big-endian filesystem on %s, appending"
" to this is unsupported\n", source);
else {
ERROR("Can't find a SQUASHFS superblock on %s\n",
source);
ERROR("Wrong filesystem or filesystem is corrupted!\n");
}
goto failed_mount;
}
/* Check the MAJOR & MINOR versions */
if(sBlk->s_major != SQUASHFS_MAJOR || sBlk->s_minor > SQUASHFS_MINOR) {
if(sBlk->s_major < 4)
ERROR("Filesystem on %s is a SQUASHFS %d.%d filesystem."
" Appending\nto SQUASHFS %d.%d filesystems is "
"not supported. Please convert it to a "
"SQUASHFS 4 filesystem\n", source,
sBlk->s_major,
sBlk->s_minor, sBlk->s_major, sBlk->s_minor);
else
ERROR("Filesystem on %s is %d.%d, which is a later "
"filesystem version than I support\n",
source, sBlk->s_major, sBlk->s_minor);
goto failed_mount;
}
/* Check the compression type */
comp = lookup_compressor_id(sBlk->compression);
if(!comp->supported) {
ERROR("Filesystem on %s uses %s compression, this is "
"unsupported by this version\n", source, comp->name);
ERROR("Compressors available:\n");
display_compressors("", "");
goto failed_mount;
}
/*
* Read extended superblock information from disk.
*
* Read compressor specific options from disk if present, and pass
* to compressor to set compressor options.
*
* Note, if there's no compressor options present, the compressor
* is still called to set the default options (the defaults may have
* been changed by the user specifying options on the command
* line which need to be over-ridden).
*
* Compressor_extract_options is also used to ensure that
* we know how decompress a filesystem compressed with these
* compression options.
*/
if(SQUASHFS_COMP_OPTS(sBlk->flags)) {
bytes = read_block(fd, sizeof(*sBlk), NULL, 0, buffer);
if(bytes == 0) {
ERROR("Failed to read compressor options from append "
"filesystem\n");
ERROR("Filesystem corrupted?\n");
goto failed_mount;
}
}
res = compressor_extract_options(comp, sBlk->block_size, buffer, bytes);
if(res == -1) {
ERROR("Compressor failed to set compressor options\n");
goto failed_mount;
}
printf("Found a valid %sSQUASHFS superblock on %s.\n",
SQUASHFS_EXPORTABLE(sBlk->flags) ? "exportable " : "", source);
printf("\tCompression used %s\n", comp->name);
printf("\tInodes are %scompressed\n",
SQUASHFS_UNCOMPRESSED_INODES(sBlk->flags) ? "un" : "");
printf("\tData is %scompressed\n",
SQUASHFS_UNCOMPRESSED_DATA(sBlk->flags) ? "un" : "");
printf("\tFragments are %scompressed\n",
SQUASHFS_UNCOMPRESSED_FRAGMENTS(sBlk->flags) ? "un" : "");
printf("\tXattrs are %scompressed\n",
SQUASHFS_UNCOMPRESSED_XATTRS(sBlk->flags) ? "un" : "");
printf("\tFragments are %spresent in the filesystem\n",
SQUASHFS_NO_FRAGMENTS(sBlk->flags) ? "not " : "");
printf("\tAlways-use-fragments option is %sspecified\n",
SQUASHFS_ALWAYS_FRAGMENTS(sBlk->flags) ? "" : "not ");
printf("\tDuplicates are %sremoved\n",
SQUASHFS_DUPLICATES(sBlk->flags) ? "" : "not ");
printf("\tXattrs are %sstored\n",
SQUASHFS_NO_XATTRS(sBlk->flags) ? "not " : "");
printf("\tFilesystem size %.2f Kbytes (%.2f Mbytes)\n",
sBlk->bytes_used / 1024.0, sBlk->bytes_used
/ (1024.0 * 1024.0));
printf("\tBlock size %d\n", sBlk->block_size);
printf("\tNumber of fragments %d\n", sBlk->fragments);
printf("\tNumber of inodes %d\n", sBlk->inodes);
printf("\tNumber of ids %d\n", sBlk->no_ids);
TRACE("sBlk->inode_table_start %llx\n", sBlk->inode_table_start);
TRACE("sBlk->directory_table_start %llx\n",
sBlk->directory_table_start);
TRACE("sBlk->id_table_start %llx\n", sBlk->id_table_start);
TRACE("sBlk->fragment_table_start %llx\n", sBlk->fragment_table_start);
TRACE("sBlk->lookup_table_start %llx\n", sBlk->lookup_table_start);
TRACE("sBlk->xattr_id_table_start %llx\n", sBlk->xattr_id_table_start);
printf("\n");
return comp;
failed_mount:
return NULL;
}
unsigned char *squashfs_readdir(int fd, int root_entries,
unsigned int directory_start_block, int offset, int size,
unsigned int *last_directory_block, struct squashfs_super_block *sBlk,
void (push_directory_entry)(char *, squashfs_inode, int, int))
{
struct squashfs_dir_header dirh;
char buffer[sizeof(struct squashfs_dir_entry) + SQUASHFS_NAME_LEN + 1]
__attribute__ ((aligned));
struct squashfs_dir_entry *dire = (struct squashfs_dir_entry *) buffer;
unsigned char *directory_table = NULL;
int byte, bytes = 0, dir_count;
long long start = sBlk->directory_table_start + directory_start_block,
last_start_block = start;
size += offset;
directory_table = malloc((size + SQUASHFS_METADATA_SIZE * 2 - 1) &
~(SQUASHFS_METADATA_SIZE - 1));
if(directory_table == NULL)
MEM_ERROR();
while(bytes < size) {
int expected = (size - bytes) >= SQUASHFS_METADATA_SIZE ?
SQUASHFS_METADATA_SIZE : 0;
TRACE("squashfs_readdir: reading block 0x%llx, bytes read so "
"far %d\n", start, bytes);
last_start_block = start;
byte = read_block(fd, start, &start, expected, directory_table + bytes);
if(byte == 0) {
ERROR("Failed to read directory\n");
ERROR("Filesystem corrupted?\n");
free(directory_table);
return NULL;
}
bytes += byte;
}
if(!root_entries)
goto all_done;
bytes = offset;
while(bytes < size) {
SQUASHFS_SWAP_DIR_HEADER(directory_table + bytes, &dirh);
dir_count = dirh.count + 1;
/* dir_count should never be larger than SQUASHFS_DIR_COUNT */
if(dir_count > SQUASHFS_DIR_COUNT) {
ERROR("File system corrupted: too many entries in directory\n");
free(directory_table);
return NULL;
}
TRACE("squashfs_readdir: Read directory header @ byte position "
"0x%x, 0x%x directory entries\n", bytes, dir_count);
bytes += sizeof(dirh);
while(dir_count--) {
SQUASHFS_SWAP_DIR_ENTRY(directory_table + bytes, dire);
bytes += sizeof(*dire);
/* size should never be SQUASHFS_NAME_LEN or larger */
if(dire->size >= SQUASHFS_NAME_LEN) {
ERROR("File system corrupted: filename too long\n");
free(directory_table);
return NULL;
}
memcpy(dire->name, directory_table + bytes,
dire->size + 1);
dire->name[dire->size + 1] = '\0';
TRACE("squashfs_readdir: pushing directory entry %s, "
"inode %x:%x, type 0x%x\n", dire->name,
dirh.start_block, dire->offset, dire->type);
push_directory_entry(dire->name,
SQUASHFS_MKINODE(dirh.start_block,
dire->offset), dirh.inode_number +
dire->inode_number, dire->type);
bytes += dire->size + 1;
}
}
all_done:
*last_directory_block = (unsigned int) last_start_block -
sBlk->directory_table_start;
return directory_table;
}
unsigned int *read_id_table(int fd, struct squashfs_super_block *sBlk)
{
int indexes = SQUASHFS_ID_BLOCKS(sBlk->no_ids);
long long index[indexes];
int bytes = SQUASHFS_ID_BYTES(sBlk->no_ids);
unsigned int *id_table;
int res, i;
id_table = malloc(bytes);
if(id_table == NULL)
MEM_ERROR();
res = read_fs_bytes(fd, sBlk->id_table_start,
SQUASHFS_ID_BLOCK_BYTES(sBlk->no_ids), index);
if(res == 0) {
ERROR("Failed to read id table index\n");
ERROR("Filesystem corrupted?\n");
free(id_table);
return NULL;
}
SQUASHFS_INSWAP_ID_BLOCKS(index, indexes);
for(i = 0; i < indexes; i++) {
int expected = (i + 1) != indexes ? SQUASHFS_METADATA_SIZE :
bytes & (SQUASHFS_METADATA_SIZE - 1);
int length = read_block(fd, index[i], NULL, expected,
((unsigned char *) id_table) +
(i * SQUASHFS_METADATA_SIZE));
TRACE("Read id table block %d, from 0x%llx, length %d\n", i,
index[i], length);
if(length == 0) {
ERROR("Failed to read id table block %d, from 0x%llx, "
"length %d\n", i, index[i], length);
ERROR("Filesystem corrupted?\n");
free(id_table);
return NULL;
}
}
SQUASHFS_INSWAP_INTS(id_table, sBlk->no_ids);
for(i = 0; i < sBlk->no_ids; i++) {
TRACE("Adding id %d to id tables\n", id_table[i]);
create_id(id_table[i]);
}
return id_table;
}
struct squashfs_fragment_entry *read_fragment_table(int fd, struct squashfs_super_block *sBlk)
{
int res, i;
int bytes = SQUASHFS_FRAGMENT_BYTES(sBlk->fragments);
int indexes = SQUASHFS_FRAGMENT_INDEXES(sBlk->fragments);
long long fragment_table_index[indexes];
struct squashfs_fragment_entry *fragment_table;
TRACE("read_fragment_table: %d fragments, reading %d fragment indexes "
"from 0x%llx\n", sBlk->fragments, indexes,
sBlk->fragment_table_start);
fragment_table = malloc(bytes);
if(fragment_table == NULL)
MEM_ERROR();
res = read_fs_bytes(fd, sBlk->fragment_table_start,
SQUASHFS_FRAGMENT_INDEX_BYTES(sBlk->fragments),
fragment_table_index);
if(res == 0) {
ERROR("Failed to read fragment table index\n");
ERROR("Filesystem corrupted?\n");
free(fragment_table);
return NULL;
}
SQUASHFS_INSWAP_FRAGMENT_INDEXES(fragment_table_index, indexes);
for(i = 0; i < indexes; i++) {
int expected = (i + 1) != indexes ? SQUASHFS_METADATA_SIZE :
bytes & (SQUASHFS_METADATA_SIZE - 1);
int length = read_block(fd, fragment_table_index[i], NULL,
expected, ((unsigned char *) fragment_table) +
(i * SQUASHFS_METADATA_SIZE));
TRACE("Read fragment table block %d, from 0x%llx, length %d\n",
i, fragment_table_index[i], length);
if(length == 0) {
ERROR("Failed to read fragment table block %d, from "
"0x%llx, length %d\n", i,
fragment_table_index[i], length);
ERROR("Filesystem corrupted?\n");
free(fragment_table);
return NULL;
}
}
for(i = 0; i < sBlk->fragments; i++)
SQUASHFS_INSWAP_FRAGMENT_ENTRY(&fragment_table[i]);
return fragment_table;
}
squashfs_inode *read_inode_lookup_table(int fd, struct squashfs_super_block *sBlk)
{
int lookup_bytes = SQUASHFS_LOOKUP_BYTES(sBlk->inodes);
int indexes = SQUASHFS_LOOKUP_BLOCKS(sBlk->inodes);
long long index[indexes];
int res, i;
squashfs_inode *inode_lookup_table;
inode_lookup_table = malloc(lookup_bytes);
if(inode_lookup_table == NULL)
MEM_ERROR();
res = read_fs_bytes(fd, sBlk->lookup_table_start,
SQUASHFS_LOOKUP_BLOCK_BYTES(sBlk->inodes), index);
if(res == 0) {
ERROR("Failed to read inode lookup table index\n");
ERROR("Filesystem corrupted?\n");
free(inode_lookup_table);
return NULL;
}
SQUASHFS_INSWAP_LONG_LONGS(index, indexes);
for(i = 0; i < indexes; i++) {
int expected = (i + 1) != indexes ? SQUASHFS_METADATA_SIZE :
lookup_bytes & (SQUASHFS_METADATA_SIZE - 1);
int length = read_block(fd, index[i], NULL, expected,
((unsigned char *) inode_lookup_table) +
(i * SQUASHFS_METADATA_SIZE));
TRACE("Read inode lookup table block %d, from 0x%llx, length "
"%d\n", i, index[i], length);
if(length == 0) {
ERROR("Failed to read inode lookup table block %d, "
"from 0x%llx, length %d\n", i, index[i],
length);
ERROR("Filesystem corrupted?\n");
free(inode_lookup_table);
return NULL;
}
}
SQUASHFS_INSWAP_LONG_LONGS(inode_lookup_table, sBlk->inodes);
return inode_lookup_table;
}
long long read_filesystem(char *root_name, int fd, struct squashfs_super_block *sBlk,
char **cinode_table, char **data_cache, char **cdirectory_table,
char **directory_data_cache, unsigned int *last_directory_block,
unsigned int *inode_dir_offset, unsigned int *inode_dir_file_size,
unsigned int *root_inode_size, unsigned int *inode_dir_start_block,
int *file_count, int *sym_count, int *dev_count, int *dir_count,
int *fifo_count, int *sock_count, long long *uncompressed_file,
unsigned int *uncompressed_inode, unsigned int *uncompressed_directory,
unsigned int *inode_dir_inode_number,
unsigned int *inode_dir_parent_inode,
void (push_directory_entry)(char *, squashfs_inode, int, int),
struct squashfs_fragment_entry **fragment_table,
squashfs_inode **inode_lookup_table)
{
unsigned char *inode_table = NULL, *directory_table = NULL;
long long start = sBlk->inode_table_start;
long long end = sBlk->directory_table_start;
long long root_inode_start = start +
SQUASHFS_INODE_BLK(sBlk->root_inode);
unsigned int root_inode_offset =
SQUASHFS_INODE_OFFSET(sBlk->root_inode);
unsigned int root_inode_block;
union squashfs_inode_header inode;
unsigned int *id_table = NULL;
int res;
printf("Scanning existing filesystem...\n");
if(get_xattrs(fd, sBlk) == 0)
goto error;
if(sBlk->fragments > 0) {
*fragment_table = read_fragment_table(fd, sBlk);
if(*fragment_table == NULL)
goto error;
}
if(sBlk->lookup_table_start != SQUASHFS_INVALID_BLK) {
*inode_lookup_table = read_inode_lookup_table(fd, sBlk);
if(*inode_lookup_table == NULL)
goto error;
}
id_table = read_id_table(fd, sBlk);
if(id_table == NULL)
goto error;
inode_table = scan_inode_table(fd, start, end, root_inode_start,
root_inode_offset, sBlk, &inode, &root_inode_block,
root_inode_size, uncompressed_file, uncompressed_directory,
file_count, sym_count, dev_count, dir_count, fifo_count,
sock_count, id_table);
if(inode_table == NULL)
goto error;
*uncompressed_inode = root_inode_block;
if(inode.base.inode_type == SQUASHFS_DIR_TYPE ||
inode.base.inode_type == SQUASHFS_LDIR_TYPE) {
if(inode.base.inode_type == SQUASHFS_DIR_TYPE) {
*inode_dir_start_block = inode.dir.start_block;
*inode_dir_offset = inode.dir.offset;
*inode_dir_file_size = inode.dir.file_size - 3;
*inode_dir_inode_number = inode.dir.inode_number;
*inode_dir_parent_inode = inode.dir.parent_inode;
} else {
*inode_dir_start_block = inode.ldir.start_block;
*inode_dir_offset = inode.ldir.offset;
*inode_dir_file_size = inode.ldir.file_size - 3;
*inode_dir_inode_number = inode.ldir.inode_number;
*inode_dir_parent_inode = inode.ldir.parent_inode;
}
directory_table = squashfs_readdir(fd, !root_name,
*inode_dir_start_block, *inode_dir_offset,
*inode_dir_file_size, last_directory_block, sBlk,
push_directory_entry);
if(directory_table == NULL)
goto error;
root_inode_start -= start;
*cinode_table = malloc(root_inode_start);
if(*cinode_table == NULL)
MEM_ERROR();
res = read_fs_bytes(fd, start, root_inode_start, *cinode_table);
if(res == 0) {
ERROR("Failed to read inode table\n");
ERROR("Filesystem corrupted?\n");
goto error;
}
*cdirectory_table = malloc(*last_directory_block);
if(*cdirectory_table == NULL)
MEM_ERROR();
res = read_fs_bytes(fd, sBlk->directory_table_start,
*last_directory_block, *cdirectory_table);
if(res == 0) {
ERROR("Failed to read directory table\n");
ERROR("Filesystem corrupted?\n");
goto error;
}
*data_cache = malloc(root_inode_offset + *root_inode_size);
if(*data_cache == NULL)
MEM_ERROR();
memcpy(*data_cache, inode_table + root_inode_block,
root_inode_offset + *root_inode_size);
*directory_data_cache = malloc(*inode_dir_offset +
*inode_dir_file_size);
if(*directory_data_cache == NULL)
MEM_ERROR();
memcpy(*directory_data_cache, directory_table,
*inode_dir_offset + *inode_dir_file_size);
free(id_table);
free(inode_table);
free(directory_table);
return sBlk->inode_table_start;
}
error:
free(id_table);
free(inode_table);
free(directory_table);
return 0;
}