freebsd-dev/lib/libarchive/archive_read_support_format_iso9660.c
Tim Kientzle cdad0e17a1 Merge r558,567,569,571,581,582,583,598 from libarchive.googlecode.com:
Support Joliet extensions.  This currently ignores Rockridge extensions
if both exist on the same disk unless the '!joliet' option is provided.
e.g.: tar -xvf example.iso --options '!joliet'
Thanks to: Andreas Henriksson
2009-03-07 02:24:32 +00:00

1500 lines
44 KiB
C

/*-
* Copyright (c) 2003-2007 Tim Kientzle
* Copyright (c) 2009 Andreas Henriksson <andreas@fatal.se>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "archive_platform.h"
__FBSDID("$FreeBSD$");
#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif
/* #include <stdint.h> */ /* See archive_platform.h */
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include <time.h>
#include "archive.h"
#include "archive_entry.h"
#include "archive_private.h"
#include "archive_read_private.h"
#include "archive_string.h"
/*
* An overview of ISO 9660 format:
*
* Each disk is laid out as follows:
* * 32k reserved for private use
* * Volume descriptor table. Each volume descriptor
* is 2k and specifies basic format information.
* The "Primary Volume Descriptor" (PVD) is defined by the
* standard and should always be present; other volume
* descriptors include various vendor-specific extensions.
* * Files and directories. Each file/dir is specified by
* an "extent" (starting sector and length in bytes).
* Dirs are just files with directory records packed one
* after another. The PVD contains a single dir entry
* specifying the location of the root directory. Everything
* else follows from there.
*
* This module works by first reading the volume descriptors, then
* building a list of directory entries, sorted by starting
* sector. At each step, I look for the earliest dir entry that
* hasn't yet been read, seek forward to that location and read
* that entry. If it's a dir, I slurp in the new dir entries and
* add them to the heap; if it's a regular file, I return the
* corresponding archive_entry and wait for the client to request
* the file body. This strategy allows us to read most compliant
* CDs with a single pass through the data, as required by libarchive.
*/
/* Structure of on-disk primary volume descriptor. */
#define PVD_type_offset 0
#define PVD_type_size 1
#define PVD_id_offset (PVD_type_offset + PVD_type_size)
#define PVD_id_size 5
#define PVD_version_offset (PVD_id_offset + PVD_id_size)
#define PVD_version_size 1
#define PVD_reserved1_offset (PVD_version_offset + PVD_version_size)
#define PVD_reserved1_size 1
#define PVD_system_id_offset (PVD_reserved1_offset + PVD_reserved1_size)
#define PVD_system_id_size 32
#define PVD_volume_id_offset (PVD_system_id_offset + PVD_system_id_size)
#define PVD_volume_id_size 32
#define PVD_reserved2_offset (PVD_volume_id_offset + PVD_volume_id_size)
#define PVD_reserved2_size 8
#define PVD_volume_space_size_offset (PVD_reserved2_offset + PVD_reserved2_size)
#define PVD_volume_space_size_size 8
#define PVD_reserved3_offset (PVD_volume_space_size_offset + PVD_volume_space_size_size)
#define PVD_reserved3_size 32
#define PVD_volume_set_size_offset (PVD_reserved3_offset + PVD_reserved3_size)
#define PVD_volume_set_size_size 4
#define PVD_volume_sequence_number_offset (PVD_volume_set_size_offset + PVD_volume_set_size_size)
#define PVD_volume_sequence_number_size 4
#define PVD_logical_block_size_offset (PVD_volume_sequence_number_offset + PVD_volume_sequence_number_size)
#define PVD_logical_block_size_size 4
#define PVD_path_table_size_offset (PVD_logical_block_size_offset + PVD_logical_block_size_size)
#define PVD_path_table_size_size 8
#define PVD_type_1_path_table_offset (PVD_path_table_size_offset + PVD_path_table_size_size)
#define PVD_type_1_path_table_size 4
#define PVD_opt_type_1_path_table_offset (PVD_type_1_path_table_offset + PVD_type_1_path_table_size)
#define PVD_opt_type_1_path_table_size 4
#define PVD_type_m_path_table_offset (PVD_opt_type_1_path_table_offset + PVD_opt_type_1_path_table_size)
#define PVD_type_m_path_table_size 4
#define PVD_opt_type_m_path_table_offset (PVD_type_m_path_table_offset + PVD_type_m_path_table_size)
#define PVD_opt_type_m_path_table_size 4
#define PVD_root_directory_record_offset (PVD_opt_type_m_path_table_offset + PVD_opt_type_m_path_table_size)
#define PVD_root_directory_record_size 34
#define PVD_volume_set_id_offset (PVD_root_directory_record_offset + PVD_root_directory_record_size)
#define PVD_volume_set_id_size 128
#define PVD_publisher_id_offset (PVD_volume_set_id_offset + PVD_volume_set_id_size)
#define PVD_publisher_id_size 128
#define PVD_preparer_id_offset (PVD_publisher_id_offset + PVD_publisher_id_size)
#define PVD_preparer_id_size 128
#define PVD_application_id_offset (PVD_preparer_id_offset + PVD_preparer_id_size)
#define PVD_application_id_size 128
#define PVD_copyright_file_id_offset (PVD_application_id_offset + PVD_application_id_size)
#define PVD_copyright_file_id_size 37
#define PVD_abstract_file_id_offset (PVD_copyright_file_id_offset + PVD_copyright_file_id_size)
#define PVD_abstract_file_id_size 37
#define PVD_bibliographic_file_id_offset (PVD_abstract_file_id_offset + PVD_abstract_file_id_size)
#define PVD_bibliographic_file_id_size 37
#define PVD_creation_date_offset (PVD_bibliographic_file_id_offset + PVD_bibliographic_file_id_size)
#define PVD_creation_date_size 17
#define PVD_modification_date_offset (PVD_creation_date_offset + PVD_creation_date_size)
#define PVD_modification_date_size 17
#define PVD_expiration_date_offset (PVD_modification_date_offset + PVD_modification_date_size)
#define PVD_expiration_date_size 17
#define PVD_effective_date_offset (PVD_expiration_date_offset + PVD_expiration_date_size)
#define PVD_effective_date_size 17
#define PVD_file_structure_version_offset (PVD_effective_date_offset + PVD_effective_date_size)
#define PVD_file_structure_version_size 1
#define PVD_reserved4_offset (PVD_file_structure_version_offset + PVD_file_structure_version_size)
#define PVD_reserved4_size 1
#define PVD_application_data_offset (PVD_reserved4_offset + PVD_reserved4_size)
#define PVD_application_data_size 512
#define PVD_reserved5_offset (PVD_application_data_offset + PVD_application_data_size)
#define PVD_reserved5_size (2048 - PVD_reserved5_offset)
/* TODO: It would make future maintenance easier to just hardcode the
* above values. In particular, ECMA119 states the offsets as part of
* the standard. That would eliminate the need for the following check.*/
#if PVD_reserved5_offset != 1395
#error PVD offset and size definitions are wrong.
#endif
/* Structure of optional on-disk supplementary volume descriptor. */
#define SVD_type_offset 0
#define SVD_type_size 1
#define SVD_id_offset (SVD_type_offset + SVD_type_size)
#define SVD_id_size 5
#define SVD_version_offset (SVD_id_offset + SVD_id_size)
#define SVD_version_size 1
/* ... */
#define SVD_volume_space_size_offset 80
#define SVD_volume_space_size_size 8
#define SVD_escape_sequences_offset (SVD_volume_space_size_offset + SVD_volume_space_size_size)
#define SVD_escape_sequences_size 32
/* ... */
#define SVD_logical_block_size_offset 128
#define SVD_logical_block_size_size 4
/* ... */
#define SVD_root_directory_record_offset 156
#define SVD_root_directory_record_size 34
/* ... */
/* FIXME: validate correctness of last SVD entry offset. */
/* Structure of an on-disk directory record. */
/* Note: ISO9660 stores each multi-byte integer twice, once in
* each byte order. The sizes here are the size of just one
* of the two integers. (This is why the offset of a field isn't
* the same as the offset+size of the previous field.) */
#define DR_length_offset 0
#define DR_length_size 1
#define DR_ext_attr_length_offset 1
#define DR_ext_attr_length_size 1
#define DR_extent_offset 2
#define DR_extent_size 4
#define DR_size_offset 10
#define DR_size_size 4
#define DR_date_offset 18
#define DR_date_size 7
#define DR_flags_offset 25
#define DR_flags_size 1
#define DR_file_unit_size_offset 26
#define DR_file_unit_size_size 1
#define DR_interleave_offset 27
#define DR_interleave_size 1
#define DR_volume_sequence_number_offset 28
#define DR_volume_sequence_number_size 2
#define DR_name_len_offset 32
#define DR_name_len_size 1
#define DR_name_offset 33
/* In-memory storage for a directory record. */
struct file_info {
struct file_info *parent;
int refcount;
uint64_t offset; /* Offset on disk. */
uint64_t size; /* File size in bytes. */
uint64_t ce_offset; /* Offset of CE */
uint64_t ce_size; /* Size of CE */
time_t birthtime; /* File created time. */
time_t mtime; /* File last modified time. */
time_t atime; /* File last accessed time. */
time_t ctime; /* File attribute change time. */
uint64_t rdev; /* Device number */
mode_t mode;
uid_t uid;
gid_t gid;
ino_t inode;
int nlinks;
struct archive_string name; /* Pathname */
char name_continues; /* Non-zero if name continues */
struct archive_string symlink;
char symlink_continues; /* Non-zero if link continues */
};
struct iso9660 {
int magic;
#define ISO9660_MAGIC 0x96609660
int option_ignore_joliet;
struct archive_string pathname;
char seenRockridge; /* Set true if RR extensions are used. */
unsigned char suspOffset;
char seenJoliet;
uint64_t previous_offset;
uint64_t previous_size;
struct archive_string previous_pathname;
/* TODO: Make this a heap for fast inserts and deletions. */
struct file_info **pending_files;
int pending_files_allocated;
int pending_files_used;
uint64_t current_position;
ssize_t logical_block_size;
uint64_t volume_size; /* Total size of volume in bytes. */
off_t entry_sparse_offset;
int64_t entry_bytes_remaining;
};
static void add_entry(struct iso9660 *iso9660, struct file_info *file);
static int archive_read_format_iso9660_bid(struct archive_read *);
static int archive_read_format_iso9660_options(struct archive_read *,
const char *, const char *);
static int archive_read_format_iso9660_cleanup(struct archive_read *);
static int archive_read_format_iso9660_read_data(struct archive_read *,
const void **, size_t *, off_t *);
static int archive_read_format_iso9660_read_data_skip(struct archive_read *);
static int archive_read_format_iso9660_read_header(struct archive_read *,
struct archive_entry *);
static const char *build_pathname(struct archive_string *, struct file_info *);
#if DEBUG
static void dump_isodirrec(FILE *, const unsigned char *isodirrec);
#endif
static time_t time_from_tm(struct tm *);
static time_t isodate17(const unsigned char *);
static time_t isodate7(const unsigned char *);
static int isJolietSVD(struct iso9660 *, const unsigned char *);
static int isPVD(struct iso9660 *, const unsigned char *);
static struct file_info *next_entry(struct iso9660 *);
static int next_entry_seek(struct archive_read *a, struct iso9660 *iso9660,
struct file_info **pfile);
static struct file_info *
parse_file_info(struct iso9660 *iso9660,
struct file_info *parent, const unsigned char *isodirrec);
static void parse_rockridge(struct iso9660 *iso9660,
struct file_info *file, const unsigned char *start,
const unsigned char *end);
static void parse_rockridge_NM1(struct file_info *,
const unsigned char *, int);
static void parse_rockridge_SL1(struct file_info *,
const unsigned char *, int);
static void parse_rockridge_TF1(struct file_info *,
const unsigned char *, int);
static void release_file(struct iso9660 *, struct file_info *);
static unsigned toi(const void *p, int n);
int
archive_read_support_format_iso9660(struct archive *_a)
{
struct archive_read *a = (struct archive_read *)_a;
struct iso9660 *iso9660;
int r;
iso9660 = (struct iso9660 *)malloc(sizeof(*iso9660));
if (iso9660 == NULL) {
archive_set_error(&a->archive, ENOMEM, "Can't allocate iso9660 data");
return (ARCHIVE_FATAL);
}
memset(iso9660, 0, sizeof(*iso9660));
iso9660->magic = ISO9660_MAGIC;
r = __archive_read_register_format(a,
iso9660,
"iso9660",
archive_read_format_iso9660_bid,
archive_read_format_iso9660_options,
archive_read_format_iso9660_read_header,
archive_read_format_iso9660_read_data,
archive_read_format_iso9660_read_data_skip,
archive_read_format_iso9660_cleanup);
if (r != ARCHIVE_OK) {
free(iso9660);
return (r);
}
return (ARCHIVE_OK);
}
static int
archive_read_format_iso9660_bid(struct archive_read *a)
{
struct iso9660 *iso9660;
ssize_t bytes_read, brsvd;
const void *h;
const unsigned char *p, *psvd;
int bid;
iso9660 = (struct iso9660 *)(a->format->data);
/*
* Skip the first 32k (reserved area) and get the first
* 8 sectors of the volume descriptor table. Of course,
* if the I/O layer gives us more, we'll take it.
*/
h = __archive_read_ahead(a, 32768 + 8*2048, &bytes_read);
if (h == NULL)
return (-1);
p = (const unsigned char *)h;
/* Skip the reserved area. */
bytes_read -= 32768;
p += 32768;
/* Check each volume descriptor to locate possible SVD with Joliet. */
for (brsvd = bytes_read, psvd = p;
!iso9660->option_ignore_joliet && brsvd > 2048;
brsvd -= 2048, psvd += 2048) {
bid = isJolietSVD(iso9660, psvd);
if (bid > 0)
return (bid);
if (*p == '\177') /* End-of-volume-descriptor marker. */
break;
}
/* Check each volume descriptor to locate the PVD. */
for (; bytes_read > 2048; bytes_read -= 2048, p += 2048) {
bid = isPVD(iso9660, p);
if (bid > 0)
return (bid);
if (*p == '\177') /* End-of-volume-descriptor marker. */
break;
}
/* We didn't find a valid PVD; return a bid of zero. */
return (0);
}
static int
archive_read_format_iso9660_options(struct archive_read *a,
const char *key, const char *val)
{
struct iso9660 *iso9660;
iso9660 = (struct iso9660 *)(a->format->data);
if (strcmp(key, "joliet") == 0) {
if (val == NULL || strcmp(val, "off") == 0 ||
strcmp(val, "ignore") == 0 ||
strcmp(val, "disable") == 0 ||
strcmp(val, "0") == 0)
iso9660->option_ignore_joliet = 1;
else
iso9660->option_ignore_joliet = 0;
return (ARCHIVE_OK);
}
/* Note: The "warn" return is just to inform the options
* supervisor that we didn't handle it. It will generate
* a suitable error if noone used this option. */
return (ARCHIVE_WARN);
}
static int
isJolietSVD(struct iso9660 *iso9660, const unsigned char *h)
{
struct file_info *file;
const unsigned char *p;
/* Type 2 means it's a SVD. */
if (h[SVD_type_offset] != 2)
return (0);
/* ID must be "CD001" */
if (memcmp(h + SVD_id_offset, "CD001", 5) != 0)
return (0);
/* FIXME: do more validations according to joliet spec. */
/* check if this SVD contains joliet extension! */
p = h + SVD_escape_sequences_offset;
/* N.B. Joliet spec says p[1] == '\\', but.... */
if (p[0] == '%' && p[1] == '/') {
int level = 0;
if (p[2] == '@')
level = 1;
else if (p[2] == 'C')
level = 2;
else if (p[2] == 'E')
level = 3;
else /* not joliet */
return (0);
iso9660->seenJoliet = level;
} else /* not joliet */
return (0);
iso9660->logical_block_size = toi(h + SVD_logical_block_size_offset, 2);
if (iso9660->logical_block_size <= 0)
return (0);
iso9660->volume_size = iso9660->logical_block_size
* (uint64_t)toi(h + SVD_volume_space_size_offset, 4);
#if DEBUG
fprintf(stderr, "Joliet UCS-2 level %d with "
"logical block size:%d, volume size:%d\n",
iso9660->seenJoliet,
iso9660->logical_block_size, iso9660->volume_size);
#endif
/* Store the root directory in the pending list. */
file = parse_file_info(iso9660, NULL, h + SVD_root_directory_record_offset);
add_entry(iso9660, file);
return (48);
}
static int
isPVD(struct iso9660 *iso9660, const unsigned char *h)
{
struct file_info *file;
int i;
/* Type of the Primary Volume Descriptor must be 1. */
if (h[PVD_type_offset] != 1)
return (0);
/* ID must be "CD001" */
if (memcmp(h + PVD_id_offset, "CD001", 5) != 0)
return (0);
/* PVD version must be 1. */
if (h[PVD_version_offset] != 1)
return (0);
/* Reserved field must be 0. */
if (h[PVD_reserved1_offset] != 0)
return (0);
/* Reserved field must be 0. */
for (i = 0; i < PVD_reserved2_size; ++i)
if (h[PVD_reserved2_offset + i] != 0)
return (0);
/* Reserved field must be 0. */
for (i = 0; i < PVD_reserved3_size; ++i)
if (h[PVD_reserved3_offset + i] != 0)
return (0);
/* Logical block size must be > 0. */
/* I've looked at Ecma 119 and can't find any stronger
* restriction on this field. */
iso9660->logical_block_size = toi(h + PVD_logical_block_size_offset, 2);
if (iso9660->logical_block_size <= 0)
return (0);
iso9660->volume_size = iso9660->logical_block_size
* (uint64_t)toi(h + PVD_volume_space_size_offset, 4);
/* File structure version must be 1 for ISO9660/ECMA119. */
if (h[PVD_file_structure_version_offset] != 1)
return (0);
/* Reserved field must be 0. */
for (i = 0; i < PVD_reserved4_size; ++i)
if (h[PVD_reserved4_offset + i] != 0)
return (0);
/* Reserved field must be 0. */
for (i = 0; i < PVD_reserved5_size; ++i)
if (h[PVD_reserved5_offset + i] != 0)
return (0);
/* XXX TODO: Check other values for sanity; reject more
* malformed PVDs. XXX */
/* Store the root directory in the pending list. */
file = parse_file_info(iso9660, NULL, h + PVD_root_directory_record_offset);
add_entry(iso9660, file);
return (48);
}
static int
archive_read_format_iso9660_read_header(struct archive_read *a,
struct archive_entry *entry)
{
struct iso9660 *iso9660;
struct file_info *file;
int r;
iso9660 = (struct iso9660 *)(a->format->data);
if (!a->archive.archive_format) {
a->archive.archive_format = ARCHIVE_FORMAT_ISO9660;
a->archive.archive_format_name = "ISO9660";
}
/* Get the next entry that appears after the current offset. */
r = next_entry_seek(a, iso9660, &file);
if (r != ARCHIVE_OK) {
release_file(iso9660, file);
return (r);
}
iso9660->entry_bytes_remaining = file->size;
iso9660->entry_sparse_offset = 0; /* Offset for sparse-file-aware clients. */
if (file->offset + file->size > iso9660->volume_size) {
archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
"File is beyond end-of-media: %s", file->name);
iso9660->entry_bytes_remaining = 0;
iso9660->entry_sparse_offset = 0;
release_file(iso9660, file);
return (ARCHIVE_WARN);
}
/* Set up the entry structure with information about this entry. */
archive_entry_set_mode(entry, file->mode);
archive_entry_set_uid(entry, file->uid);
archive_entry_set_gid(entry, file->gid);
archive_entry_set_nlink(entry, file->nlinks);
archive_entry_set_ino(entry, file->inode);
archive_entry_set_birthtime(entry, file->birthtime, 0);
archive_entry_set_mtime(entry, file->mtime, 0);
archive_entry_set_ctime(entry, file->ctime, 0);
archive_entry_set_atime(entry, file->atime, 0);
/* N.B.: Rock Ridge supports 64-bit device numbers. */
archive_entry_set_rdev(entry, (dev_t)file->rdev);
archive_entry_set_size(entry, iso9660->entry_bytes_remaining);
archive_string_empty(&iso9660->pathname);
archive_entry_set_pathname(entry,
build_pathname(&iso9660->pathname, file));
if (file->symlink.s != NULL)
archive_entry_copy_symlink(entry, file->symlink.s);
/* If this entry points to the same data as the previous
* entry, convert this into a hardlink to that entry.
* But don't bother for zero-length files. */
if (file->offset == iso9660->previous_offset
&& file->size == iso9660->previous_size
&& file->size > 0) {
archive_entry_set_hardlink(entry,
iso9660->previous_pathname.s);
iso9660->entry_bytes_remaining = 0;
iso9660->entry_sparse_offset = 0;
release_file(iso9660, file);
return (ARCHIVE_OK);
}
/* If the offset is before our current position, we can't
* seek backwards to extract it, so issue a warning. */
if (file->offset < iso9660->current_position) {
archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
"Ignoring out-of-order file @%x (%s) %jd < %jd",
file,
iso9660->pathname.s,
file->offset, iso9660->current_position);
iso9660->entry_bytes_remaining = 0;
iso9660->entry_sparse_offset = 0;
release_file(iso9660, file);
return (ARCHIVE_WARN);
}
iso9660->previous_size = file->size;
iso9660->previous_offset = file->offset;
archive_strcpy(&iso9660->previous_pathname, iso9660->pathname.s);
/* If this is a directory, read in all of the entries right now. */
if (archive_entry_filetype(entry) == AE_IFDIR) {
while (iso9660->entry_bytes_remaining > 0) {
const void *block;
const unsigned char *p;
ssize_t step = iso9660->logical_block_size;
if (step > iso9660->entry_bytes_remaining)
step = iso9660->entry_bytes_remaining;
block = __archive_read_ahead(a, step, NULL);
if (block == NULL) {
archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
"Failed to read full block when scanning ISO9660 directory list");
release_file(iso9660, file);
return (ARCHIVE_FATAL);
}
__archive_read_consume(a, step);
iso9660->current_position += step;
iso9660->entry_bytes_remaining -= step;
for (p = (const unsigned char *)block;
*p != 0 && p < (const unsigned char *)block + step;
p += *p) {
struct file_info *child;
/* N.B.: these special directory identifiers
* are 8 bit "values" even on a
* Joliet CD with UCS-2 (16bit) encoding.
*/
/* Skip '.' entry. */
if (*(p + DR_name_len_offset) == 1
&& *(p + DR_name_offset) == '\0')
continue;
/* Skip '..' entry. */
if (*(p + DR_name_len_offset) == 1
&& *(p + DR_name_offset) == '\001')
continue;
child = parse_file_info(iso9660, file, p);
add_entry(iso9660, child);
if (iso9660->seenRockridge) {
a->archive.archive_format =
ARCHIVE_FORMAT_ISO9660_ROCKRIDGE;
a->archive.archive_format_name =
"ISO9660 with Rockridge extensions";
}
}
}
}
release_file(iso9660, file);
return (ARCHIVE_OK);
}
static int
archive_read_format_iso9660_read_data_skip(struct archive_read *a)
{
/* Because read_next_header always does an explicit skip
* to the next entry, we don't need to do anything here. */
(void)a; /* UNUSED */
return (ARCHIVE_OK);
}
static int
archive_read_format_iso9660_read_data(struct archive_read *a,
const void **buff, size_t *size, off_t *offset)
{
ssize_t bytes_read;
struct iso9660 *iso9660;
iso9660 = (struct iso9660 *)(a->format->data);
if (iso9660->entry_bytes_remaining <= 0) {
*buff = NULL;
*size = 0;
*offset = iso9660->entry_sparse_offset;
return (ARCHIVE_EOF);
}
*buff = __archive_read_ahead(a, 1, &bytes_read);
if (bytes_read == 0)
archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
"Truncated input file");
if (buff == NULL)
return (ARCHIVE_FATAL);
if (bytes_read > iso9660->entry_bytes_remaining)
bytes_read = iso9660->entry_bytes_remaining;
*size = bytes_read;
*offset = iso9660->entry_sparse_offset;
iso9660->entry_sparse_offset += bytes_read;
iso9660->entry_bytes_remaining -= bytes_read;
iso9660->current_position += bytes_read;
__archive_read_consume(a, bytes_read);
return (ARCHIVE_OK);
}
static int
archive_read_format_iso9660_cleanup(struct archive_read *a)
{
struct iso9660 *iso9660;
struct file_info *file;
iso9660 = (struct iso9660 *)(a->format->data);
while ((file = next_entry(iso9660)) != NULL)
release_file(iso9660, file);
archive_string_free(&iso9660->pathname);
archive_string_free(&iso9660->previous_pathname);
if (iso9660->pending_files)
free(iso9660->pending_files);
free(iso9660);
(a->format->data) = NULL;
return (ARCHIVE_OK);
}
/*
* This routine parses a single ISO directory record, makes sense
* of any extensions, and stores the result in memory.
*/
static struct file_info *
parse_file_info(struct iso9660 *iso9660, struct file_info *parent,
const unsigned char *isodirrec)
{
struct file_info *file;
size_t name_len;
const unsigned char *rr_start, *rr_end;
const unsigned char *p;
int flags;
/* TODO: Sanity check that name_len doesn't exceed length, etc. */
/* Create a new file entry and copy data from the ISO dir record. */
file = (struct file_info *)malloc(sizeof(*file));
if (file == NULL)
return (NULL);
memset(file, 0, sizeof(*file));
file->parent = parent;
if (parent != NULL)
parent->refcount++;
file->offset = (uint64_t)toi(isodirrec + DR_extent_offset, DR_extent_size)
* iso9660->logical_block_size;
file->size = toi(isodirrec + DR_size_offset, DR_size_size);
file->mtime = isodate7(isodirrec + DR_date_offset);
file->ctime = file->atime = file->mtime;
name_len = (size_t)isodirrec[DR_name_len_offset];
p = isodirrec + DR_name_offset;
/* Rockridge extensions (if any) follow name. Compute this
* before fidgeting the name_len below. */
rr_start = p + name_len + (name_len & 1 ? 0 : 1) + iso9660->suspOffset;
rr_end = isodirrec + isodirrec[DR_length_offset];
if (iso9660->seenJoliet) {
/* Joliet names are max 64 chars (128 bytes) according to spec,
* but genisoimage (and others?) will allow you to have more.
*/
wchar_t wbuff[64+1], *wp;
const unsigned char *c;
/* TODO: warn when name_len > 128 ? */
/* convert BE UTF-16 to wchar_t */
for (c = p, wp = wbuff;
c < (p + name_len) &&
wp < (wbuff + sizeof(wbuff)/sizeof(*wbuff) - 1);
c += 2) {
*wp++ = (((255 & (int)c[0]) << 8) | (255 & (int)c[1]));
}
*wp = L'\0';
#if 0 /* untested code, is it at all useful on Joliet? */
/* trim trailing first version and dot from filename.
*
* Remember we where in UTF-16BE land!
* SEPARATOR 1 (.) and SEPARATOR 2 (;) are both
* 16 bits big endian characters on Joliet.
*
* TODO: sanitize filename?
* Joliet allows any UCS-2 char except:
* *, /, :, ;, ? and \.
*/
/* Chop off trailing ';1' from files. */
if (*(wp-2) == ';' && *(wp-1) == '1') {
wp-=2;
*wp = L'\0';
}
/* Chop off trailing '.' from filenames. */
if (*(wp-1) == '.')
*(--wp) = L'\0';
#endif
/* store the result in the file name field. */
archive_strappend_w_utf8(&file->name, wbuff);
} else {
/* Chop off trailing ';1' from files. */
if (name_len > 2 && p[name_len - 2] == ';' &&
p[name_len - 1] == '1')
name_len -= 2;
/* Chop off trailing '.' from filenames. */
if (name_len > 1 && p[name_len - 1] == '.')
--name_len;
archive_strncpy(&file->name, (const char *)p, name_len);
}
flags = isodirrec[DR_flags_offset];
if (flags & 0x02)
file->mode = AE_IFDIR | 0700;
else
file->mode = AE_IFREG | 0400;
/* Rockridge extensions overwrite information from above. */
parse_rockridge(iso9660, file, rr_start, rr_end);
#if DEBUG
/* DEBUGGING: Warn about attributes I don't yet fully support. */
if ((flags & ~0x02) != 0) {
fprintf(stderr, "\n ** Unrecognized flag: ");
dump_isodirrec(stderr, isodirrec);
fprintf(stderr, "\n");
} else if (toi(isodirrec + DR_volume_sequence_number_offset, 2) != 1) {
fprintf(stderr, "\n ** Unrecognized sequence number: ");
dump_isodirrec(stderr, isodirrec);
fprintf(stderr, "\n");
} else if (*(isodirrec + DR_file_unit_size_offset) != 0) {
fprintf(stderr, "\n ** Unexpected file unit size: ");
dump_isodirrec(stderr, isodirrec);
fprintf(stderr, "\n");
} else if (*(isodirrec + DR_interleave_offset) != 0) {
fprintf(stderr, "\n ** Unexpected interleave: ");
dump_isodirrec(stderr, isodirrec);
fprintf(stderr, "\n");
} else if (*(isodirrec + DR_ext_attr_length_offset) != 0) {
fprintf(stderr, "\n ** Unexpected extended attribute length: ");
dump_isodirrec(stderr, isodirrec);
fprintf(stderr, "\n");
}
#endif
return (file);
}
static void
add_entry(struct iso9660 *iso9660, struct file_info *file)
{
/* Expand our pending files list as necessary. */
if (iso9660->pending_files_used >= iso9660->pending_files_allocated) {
struct file_info **new_pending_files;
int new_size = iso9660->pending_files_allocated * 2;
if (iso9660->pending_files_allocated < 1024)
new_size = 1024;
/* Overflow might keep us from growing the list. */
if (new_size <= iso9660->pending_files_allocated)
__archive_errx(1, "Out of memory");
new_pending_files = (struct file_info **)malloc(new_size * sizeof(new_pending_files[0]));
if (new_pending_files == NULL)
__archive_errx(1, "Out of memory");
memcpy(new_pending_files, iso9660->pending_files,
iso9660->pending_files_allocated * sizeof(new_pending_files[0]));
if (iso9660->pending_files != NULL)
free(iso9660->pending_files);
iso9660->pending_files = new_pending_files;
iso9660->pending_files_allocated = new_size;
}
iso9660->pending_files[iso9660->pending_files_used++] = file;
}
static void
parse_rockridge(struct iso9660 *iso9660, struct file_info *file,
const unsigned char *p, const unsigned char *end)
{
(void)iso9660; /* UNUSED */
file->name_continues = 0;
file->symlink_continues = 0;
while (p + 4 < end /* Enough space for another entry. */
&& p[0] >= 'A' && p[0] <= 'Z' /* Sanity-check 1st char of name. */
&& p[1] >= 'A' && p[1] <= 'Z' /* Sanity-check 2nd char of name. */
&& p[2] >= 4 /* Sanity-check length. */
&& p + p[2] <= end) { /* Sanity-check length. */
const unsigned char *data = p + 4;
int data_length = p[2] - 4;
int version = p[3];
/*
* Yes, each 'if' here does test p[0] again.
* Otherwise, the fall-through handling to catch
* unsupported extensions doesn't work.
*/
switch(p[0]) {
case 'C':
if (p[0] == 'C' && p[1] == 'E') {
if (version == 1 && data_length == 24) {
/*
* CE extension comprises:
* 8 byte sector containing extension
* 8 byte offset w/in above sector
* 8 byte length of continuation
*/
file->ce_offset = (uint64_t)toi(data, 4)
* iso9660->logical_block_size
+ toi(data + 8, 4);
file->ce_size = toi(data + 16, 4);
/* If the result is rediculous,
* ignore it. */
if (file->ce_offset + file->ce_size
> iso9660->volume_size) {
file->ce_offset = 0;
file->ce_size = 0;
}
}
break;
}
/* FALLTHROUGH */
case 'N':
if (p[0] == 'N' && p[1] == 'M') {
if (version == 1)
parse_rockridge_NM1(file,
data, data_length);
break;
}
/* FALLTHROUGH */
case 'P':
if (p[0] == 'P' && p[1] == 'D') {
/*
* PD extension is padding;
* contents are always ignored.
*/
break;
}
if (p[0] == 'P' && p[1] == 'N') {
if (version == 1 && data_length == 16) {
file->rdev = toi(data,4);
file->rdev <<= 32;
file->rdev |= toi(data + 8, 4);
}
break;
}
if (p[0] == 'P' && p[1] == 'X') {
/*
* PX extension comprises:
* 8 bytes for mode,
* 8 bytes for nlinks,
* 8 bytes for uid,
* 8 bytes for gid,
* 8 bytes for inode.
*/
if (version == 1) {
if (data_length >= 8)
file->mode
= toi(data, 4);
if (data_length >= 16)
file->nlinks
= toi(data + 8, 4);
if (data_length >= 24)
file->uid
= toi(data + 16, 4);
if (data_length >= 32)
file->gid
= toi(data + 24, 4);
if (data_length >= 40)
file->inode
= toi(data + 32, 4);
}
break;
}
/* FALLTHROUGH */
case 'R':
if (p[0] == 'R' && p[1] == 'R' && version == 1) {
iso9660->seenRockridge = 1;
/*
* RR extension comprises:
* one byte flag value
*/
/* TODO: Handle RR extension. */
break;
}
/* FALLTHROUGH */
case 'S':
if (p[0] == 'S' && p[1] == 'L') {
if (version == 1)
parse_rockridge_SL1(file,
data, data_length);
break;
}
if (p[0] == 'S' && p[1] == 'P'
&& version == 1 && data_length == 3
&& data[0] == (unsigned char)'\xbe'
&& data[1] == (unsigned char)'\xef') {
/*
* SP extension stores the suspOffset
* (Number of bytes to skip between
* filename and SUSP records.)
* It is mandatory by the SUSP standard
* (IEEE 1281).
*
* It allows SUSP to coexist with
* non-SUSP uses of the System
* Use Area by placing non-SUSP data
* before SUSP data.
*
* TODO: Add a check for 'SP' in
* first directory entry, disable all SUSP
* processing if not found.
*/
iso9660->suspOffset = data[2];
break;
}
if (p[0] == 'S' && p[1] == 'T'
&& data_length == 0 && version == 1) {
/*
* ST extension marks end of this
* block of SUSP entries.
*
* It allows SUSP to coexist with
* non-SUSP uses of the System
* Use Area by placing non-SUSP data
* after SUSP data.
*/
return;
}
case 'T':
if (p[0] == 'T' && p[1] == 'F') {
if (version == 1)
parse_rockridge_TF1(file,
data, data_length);
break;
}
/* FALLTHROUGH */
default:
/* The FALLTHROUGHs above leave us here for
* any unsupported extension. */
#if DEBUG
{
const unsigned char *t;
fprintf(stderr, "\nUnsupported RRIP extension for %s\n", file->name.s);
fprintf(stderr, " %c%c(%d):", p[0], p[1], data_length);
for (t = data; t < data + data_length && t < data + 16; t++)
fprintf(stderr, " %02x", *t);
fprintf(stderr, "\n");
}
#endif
break;
}
p += p[2];
}
}
static void
parse_rockridge_NM1(struct file_info *file,
const unsigned char *data, int data_length)
{
if (!file->name_continues)
archive_string_empty(&file->name);
file->name_continues = 0;
if (data_length < 1)
return;
/*
* NM version 1 extension comprises:
* 1 byte flag, value is one of:
* = 0: remainder is name
* = 1: remainder is name, next NM entry continues name
* = 2: "."
* = 4: ".."
* = 32: Implementation specific
* All other values are reserved.
*/
switch(data[0]) {
case 0:
if (data_length < 2)
return;
archive_strncat(&file->name, (const char *)data + 1, data_length - 1);
break;
case 1:
if (data_length < 2)
return;
archive_strncat(&file->name, (const char *)data + 1, data_length - 1);
file->name_continues = 1;
break;
case 2:
archive_strcat(&file->name, ".");
break;
case 4:
archive_strcat(&file->name, "..");
break;
default:
return;
}
}
static void
parse_rockridge_TF1(struct file_info *file, const unsigned char *data,
int data_length)
{
char flag;
/*
* TF extension comprises:
* one byte flag
* create time (optional)
* modify time (optional)
* access time (optional)
* attribute time (optional)
* Time format and presence of fields
* is controlled by flag bits.
*/
if (data_length < 1)
return;
flag = data[0];
++data;
--data_length;
if (flag & 0x80) {
/* Use 17-byte time format. */
if ((flag & 1) && data_length >= 17) {
/* Create time. */
file->birthtime = isodate17(data);
data += 17;
data_length -= 17;
}
if ((flag & 2) && data_length >= 17) {
/* Modify time. */
file->mtime = isodate17(data);
data += 17;
data_length -= 17;
}
if ((flag & 4) && data_length >= 17) {
/* Access time. */
file->atime = isodate17(data);
data += 17;
data_length -= 17;
}
if ((flag & 8) && data_length >= 17) {
/* Attribute change time. */
file->ctime = isodate17(data);
data += 17;
data_length -= 17;
}
} else {
/* Use 7-byte time format. */
if ((flag & 1) && data_length >= 7) {
/* Create time. */
file->birthtime = isodate17(data);
data += 7;
data_length -= 7;
}
if ((flag & 2) && data_length >= 7) {
/* Modify time. */
file->mtime = isodate7(data);
data += 7;
data_length -= 7;
}
if ((flag & 4) && data_length >= 7) {
/* Access time. */
file->atime = isodate7(data);
data += 7;
data_length -= 7;
}
if ((flag & 8) && data_length >= 7) {
/* Attribute change time. */
file->ctime = isodate7(data);
data += 7;
data_length -= 7;
}
}
}
static void
parse_rockridge_SL1(struct file_info *file, const unsigned char *data,
int data_length)
{
int component_continues = 1;
if (!file->symlink_continues)
archive_string_empty(&file->symlink);
else
archive_strcat(&file->symlink, "/");
file->symlink_continues = 0;
/*
* Defined flag values:
* 0: This is the last SL record for this symbolic link
* 1: this symbolic link field continues in next SL entry
* All other values are reserved.
*/
if (data_length < 1)
return;
switch(*data) {
case 0:
break;
case 1:
file->symlink_continues = 1;
break;
default:
return;
}
++data; /* Skip flag byte. */
--data_length;
/*
* SL extension body stores "components".
* Basically, this is a complicated way of storing
* a POSIX path. It also interferes with using
* symlinks for storing non-path data. <sigh>
*
* Each component is 2 bytes (flag and length)
* possibly followed by name data.
*/
while (data_length >= 2) {
unsigned char flag = *data++;
unsigned char nlen = *data++;
data_length -= 2;
if (!component_continues)
archive_strcat(&file->symlink, "/");
component_continues = 0;
switch(flag) {
case 0: /* Usual case, this is text. */
if (data_length < nlen)
return;
archive_strncat(&file->symlink,
(const char *)data, nlen);
break;
case 0x01: /* Text continues in next component. */
if (data_length < nlen)
return;
archive_strncat(&file->symlink,
(const char *)data, nlen);
component_continues = 1;
break;
case 0x02: /* Current dir. */
archive_strcat(&file->symlink, ".");
break;
case 0x04: /* Parent dir. */
archive_strcat(&file->symlink, "..");
break;
case 0x08: /* Root of filesystem. */
archive_string_empty(&file->symlink);
archive_strcat(&file->symlink, "/");
break;
case 0x10: /* Undefined (historically "volume root" */
archive_string_empty(&file->symlink);
archive_strcat(&file->symlink, "ROOT");
break;
case 0x20: /* Undefined (historically "hostname") */
archive_strcat(&file->symlink, "hostname");
break;
default:
/* TODO: issue a warning ? */
return;
}
data += nlen;
data_length -= nlen;
}
}
static void
release_file(struct iso9660 *iso9660, struct file_info *file)
{
struct file_info *parent;
if (file == NULL)
return;
if (file->refcount == 0) {
parent = file->parent;
archive_string_free(&file->name);
archive_string_free(&file->symlink);
free(file);
if (parent != NULL) {
parent->refcount--;
release_file(iso9660, parent);
}
}
}
static int
next_entry_seek(struct archive_read *a, struct iso9660 *iso9660,
struct file_info **pfile)
{
struct file_info *file;
uint64_t offset;
*pfile = NULL;
for (;;) {
*pfile = file = next_entry(iso9660);
if (file == NULL)
return (ARCHIVE_EOF);
/* CE area precedes actual file data? Ignore it. */
if (file->ce_offset > file->offset) {
/* fprintf(stderr, " *** Discarding CE data.\n"); */
file->ce_offset = 0;
file->ce_size = 0;
}
/* Don't waste time seeking for zero-length bodies. */
if (file->size == 0) {
file->offset = iso9660->current_position;
}
/* If CE exists, find and read it now. */
if (file->ce_offset > 0)
offset = file->ce_offset;
else
offset = file->offset;
/* Seek forward to the start of the entry. */
if (iso9660->current_position < offset) {
off_t step = offset - iso9660->current_position;
off_t bytes_read;
bytes_read = __archive_read_skip(a, step);
if (bytes_read < 0)
return (bytes_read);
iso9660->current_position = offset;
}
/* We found body of file; handle it now. */
if (offset == file->offset)
return (ARCHIVE_OK);
/* Found CE? Process it and push the file back onto list. */
if (offset == file->ce_offset) {
const void *p;
ssize_t size = file->ce_size;
const unsigned char *rr_start;
file->ce_offset = 0;
file->ce_size = 0;
p = __archive_read_ahead(a, size, NULL);
if (p == NULL)
return (ARCHIVE_FATAL);
rr_start = (const unsigned char *)p;
parse_rockridge(iso9660, file, rr_start,
rr_start + size);
__archive_read_consume(a, size);
iso9660->current_position += size;
add_entry(iso9660, file);
}
}
}
static struct file_info *
next_entry(struct iso9660 *iso9660)
{
int least_index;
uint64_t least_end_offset;
int i;
struct file_info *r;
if (iso9660->pending_files_used < 1)
return (NULL);
/* Assume the first file in the list is the earliest on disk. */
least_index = 0;
least_end_offset = iso9660->pending_files[0]->offset
+ iso9660->pending_files[0]->size;
/* Now, try to find an earlier one. */
for (i = 0; i < iso9660->pending_files_used; i++) {
/* Use the position of the file *end* as our comparison. */
uint64_t end_offset = iso9660->pending_files[i]->offset
+ iso9660->pending_files[i]->size;
if (iso9660->pending_files[i]->ce_offset > 0
&& iso9660->pending_files[i]->ce_offset < iso9660->pending_files[i]->offset)
end_offset = iso9660->pending_files[i]->ce_offset
+ iso9660->pending_files[i]->ce_size;
if (least_end_offset > end_offset) {
least_index = i;
least_end_offset = end_offset;
}
}
r = iso9660->pending_files[least_index];
iso9660->pending_files[least_index]
= iso9660->pending_files[--iso9660->pending_files_used];
return (r);
}
static unsigned int
toi(const void *p, int n)
{
const unsigned char *v = (const unsigned char *)p;
if (n > 1)
return v[0] + 256 * toi(v + 1, n - 1);
if (n == 1)
return v[0];
return (0);
}
static time_t
isodate7(const unsigned char *v)
{
struct tm tm;
int offset;
memset(&tm, 0, sizeof(tm));
tm.tm_year = v[0];
tm.tm_mon = v[1] - 1;
tm.tm_mday = v[2];
tm.tm_hour = v[3];
tm.tm_min = v[4];
tm.tm_sec = v[5];
/* v[6] is the signed timezone offset, in 1/4-hour increments. */
offset = ((const signed char *)v)[6];
if (offset > -48 && offset < 52) {
tm.tm_hour -= offset / 4;
tm.tm_min -= (offset % 4) * 15;
}
return (time_from_tm(&tm));
}
static time_t
isodate17(const unsigned char *v)
{
struct tm tm;
int offset;
memset(&tm, 0, sizeof(tm));
tm.tm_year = (v[0] - '0') * 1000 + (v[1] - '0') * 100
+ (v[2] - '0') * 10 + (v[3] - '0')
- 1900;
tm.tm_mon = (v[4] - '0') * 10 + (v[5] - '0');
tm.tm_mday = (v[6] - '0') * 10 + (v[7] - '0');
tm.tm_hour = (v[8] - '0') * 10 + (v[9] - '0');
tm.tm_min = (v[10] - '0') * 10 + (v[11] - '0');
tm.tm_sec = (v[12] - '0') * 10 + (v[13] - '0');
/* v[16] is the signed timezone offset, in 1/4-hour increments. */
offset = ((const signed char *)v)[16];
if (offset > -48 && offset < 52) {
tm.tm_hour -= offset / 4;
tm.tm_min -= (offset % 4) * 15;
}
return (time_from_tm(&tm));
}
static time_t
time_from_tm(struct tm *t)
{
#if HAVE_TIMEGM
/* Use platform timegm() if available. */
return (timegm(t));
#else
/* Else use direct calculation using POSIX assumptions. */
/* First, fix up tm_yday based on the year/month/day. */
mktime(t);
/* Then we can compute timegm() from first principles. */
return (t->tm_sec + t->tm_min * 60 + t->tm_hour * 3600
+ t->tm_yday * 86400 + (t->tm_year - 70) * 31536000
+ ((t->tm_year - 69) / 4) * 86400 -
((t->tm_year - 1) / 100) * 86400
+ ((t->tm_year + 299) / 400) * 86400);
#endif
}
static const char *
build_pathname(struct archive_string *as, struct file_info *file)
{
if (file->parent != NULL && archive_strlen(&file->parent->name) > 0) {
build_pathname(as, file->parent);
archive_strcat(as, "/");
}
if (archive_strlen(&file->name) == 0)
archive_strcat(as, ".");
else
archive_string_concat(as, &file->name);
return (as->s);
}
#if DEBUG
static void
dump_isodirrec(FILE *out, const unsigned char *isodirrec)
{
fprintf(out, " l %d,",
toi(isodirrec + DR_length_offset, DR_length_size));
fprintf(out, " a %d,",
toi(isodirrec + DR_ext_attr_length_offset, DR_ext_attr_length_size));
fprintf(out, " ext 0x%x,",
toi(isodirrec + DR_extent_offset, DR_extent_size));
fprintf(out, " s %d,",
toi(isodirrec + DR_size_offset, DR_extent_size));
fprintf(out, " f 0x%02x,",
toi(isodirrec + DR_flags_offset, DR_flags_size));
fprintf(out, " u %d,",
toi(isodirrec + DR_file_unit_size_offset, DR_file_unit_size_size));
fprintf(out, " ilv %d,",
toi(isodirrec + DR_interleave_offset, DR_interleave_size));
fprintf(out, " seq %d,",
toi(isodirrec + DR_volume_sequence_number_offset, DR_volume_sequence_number_size));
fprintf(out, " nl %d:",
toi(isodirrec + DR_name_len_offset, DR_name_len_size));
fprintf(out, " `%.*s'",
toi(isodirrec + DR_name_len_offset, DR_name_len_size), isodirrec + DR_name_offset);
}
#endif