Explaining the partition offset feature

Thomas Schmitt 12 years ago
parent b86abe8c65
commit d02d8642b9
  1. 1
  2. 159
  3. 4
  4. 2

@ -228,6 +228,7 @@ EXTRA_DIST = \
version.h.in \
doc/comments \
doc/doxygen.conf.in \
doc/partition_offset.wiki \

@ -0,0 +1,159 @@
The partition offset feature of libisofs can produce ISO 9660 images which bear
a quite conventional partition table if copied onto a USB stick. The first
partition marks the size of the ISO image but starts at a non-zero address.
Thus it marks a small part of the device as unclaimed by partitions and
available for storing boot loader code.
Nevertheless the USB stick is mountable via its overall device file as well as
via the partition device file. E.g. on GNU/Linux: /dev/sdb and /dev/sdb1.
This is achieved by two distinct sets of meta-data which refer to the same
file content.
The dual-mount feature supports Rock Ridge and eventually Joliet too.
It is capable of multi-session.
Currently only offset 32 kB seems to make sense. Smaller offsets are prohibited
by fundamental assumptions of libisofs and libisoburn. Larger offsets would
extend the unclaimed area into vital blocks of the ISO image.
Testing mountability and ISOLINUX bootability from USB stick and CD.
The test image was derived from one year old RIPLinux-9.3-non-X.iso which
has an isohybrid MBR. Syslinux version seems to be 3.82. That MBR and the file
tree from the mounted RIPLinux image was used to build a new ISO image
with 16 * 2kB partition offset. Isohybrid MBR patching was done by xorriso.
The first 32 kB of an ISO 9660 image are called System Area and may host any
byte pattern. In the case of RIPLinux-9.3-non-X.iso only the first 512 bytes
are non-zero. But to avoid any assumptions, all 32 kB get copied here.
dd if=RIPLinux-9.3-non-X.iso bs=1K count=32 of=RIPLinux-9.3-non-X.sysarea
The files of the image are made accessible for reading
mount -o loop RIPLinux-9.3-non-X.iso /mnt
A new ISO image gets composed. The first three lines of arguments are taken
from the prescriptions of ISOLINUX wiki and adapted to the names used in
Option -isohybrid-mbr imports the copied System Area and patches the MBR
according to rules published by hpa on Syslinux mailing list.
Option -partition_offset 16 causes the first partition to start at 2 kB block
number 16. It also prepares the image to be mountable by this partition, too.
xorriso -as mkisofs \
-o new_image.iso \
-b boot/isolinux/isolinux.bin -c boot/boot.cat \
-no-emul-boot -boot-load-size 4 -boot-info-table \
-isohybrid-mbr RIPLinux-9.3-non-X.sysarea \
-partition_offset 16 \
The image was copied onto a USB stick
dd if=new_image.iso of=/dev/sdc
and plugged into a Debian system.
fdisk -lu /dev/sdb
Device Boot Start End Blocks Id System
/dev/sdb1 * 64 119215 59576 17 Hidden HPFS/NTFS
I can mount /dev/sdb and /dev/sdb1 alike:
mount /dev/sdb1 /mnt1
mount -o loop /dev/sdb /mnt
-o loop avoids failure with "mount: /dev/sdb already mounted or /mnt busy".
A comparison by
diff -r /mnt /mnt1
reports no difference.
Human readable files look ok.
Test-reading all content by
tar cf - /mnt | wc
yields a reasonable byte count of 60743680 and no errors.
The machine boots RIPLinux from this USB stick with no visible problems.
It can then mount /dev/sdb as well as /dev/sdb1.
The ISO image boots from CD too.
Mounting the partition can be simulated with an image file on hard disk by
cutting off the first partition_offset blocks of 2 KB:
dd if=new_image.iso of=partition_image.iso bs=2048 skip=16
mount -o loop partition_image.iso /mnt1
Open questions:
- Shall the partition of an isohybrid image be marked bootable ?
Currently xorriso keeps the 0x80 mark of an eventually imported MBR
resp. the 0x80 mark which xorriso eventually sets by its own MBR
- If not to be marked bootable:
What equipment would the partition need to justify having the mark ?
- I am still puzzled by the cylinder-head-sector ambiguity. How does a reader
of the partition table determine the numbers for heads/cylinder and
sectors/head ?
fdisk tells about my /dev/sdb
105 heads, 17 sectors/track, 2193 cylinders, total 3915776 sectors
My partiton table rather was computed with 255 heads/cylinder, 63 sectors/head
Debian does not mind the difference. (It uses the 32 bit LBAs, i assume.)
The partition offset feature can be controlled by libisofs API call
int iso_write_opts_set_part_offset(IsoWriteOpts *opts,
uint32_t block_offset_2k,
int secs_512_per_head,
int heads_per_cyl);
resp. by libisoburn calls
int isoburn_igopt_set_part_offset(struct isoburn_imgen_opts *opts,
uint32_t block_offset_2k,
int secs_512_per_head, int heads_per_cyl);
int isoburn_igopt_get_part_offset(struct isoburn_imgen_opts *opts,
uint32_t *block_offset_2k,
int *secs_512_per_head, int *heads_per_cyl);
resp. by xorriso options
-boot_image any partition_offset=(2kb_block_adr)
-boot_image any partition_sec_hd=(number)
-boot_image any partition_hd_cyl=(number)
-as mkisofs ... -partition_offset (2kb_block_adr) ...
As stated above, an offset larger than 16 would expose vital parts of the
ISO image as unclaimed space. Values smaller than 16 are not accepted.
So use either an offset of 16 blocks or keep the feature disabled by
offset 0.

@ -204,6 +204,9 @@ copy_files xorriso/xorriso_buildstamp_none.h \
copy_files xorriso/xorriso_buildstamp_none.h \
create_dir "$lone_dir"/doc
copy_files doc/partition_offset.wiki "$lone_dir"/doc
create_dir "$lone_dir"/test
copy_files \
test/compare_file.c \
@ -214,7 +217,6 @@ copy_files \
create_dir "$lone_dir"/libisofs
create_dir "$lone_dir"/libisofs/filters
create_dir "$lone_dir"/doc
goto_dir "$current_dir"/nglibisofs-develop
copy_files libisofs/*.[ch] "$lone_dir"/libisofs
copy_files libisofs/filters/*.[ch] "$lone_dir"/libisofs/filters

@ -1 +1 @@
#define Xorriso_timestamP "2010.09.21.150325"
#define Xorriso_timestamP "2010.09.23.135551"