13 KiB
BOOTBOOT Bootable Disk Image Creator
See BOOTBOOT Protocol for common details.
This is an all-in-one, multiplatform, dependency-free disk image creator tool. You pass a disk configuration to it in a very
flexible JSON, and it generates ESP FAT boot partition with the required loader files, GPT partitioning table, PMBR, etc. It
also creates an initrd from a directory (currently cpio
, tar
, jamesm
(James Molloy's initrd), echfs
and FS/Z
supported, but the code is written in a way that it is easily expandable). In contrast, partitions can be generated from
directories for fat
(with long file names), minix
(Minix V3 with POSIX permissions and device files), tar
, echfs
,
FS/Z
filesystems.
The generated image was tested with fdisk, and with the verify function of gdisk. The FAT partition was tested with fsck.vfat and with TianoCore UEFI firmware and on Raspberry Pi. The ISO9660 part tested with iat (ISO9660 Analyzer Tool) and Linux mount.
Operating Modes
$ ./mkbootimg
BOOTBOOT mkbootimg utility - bztsrc@gitlab
BOOTBOOT Copyright (c) bzt MIT https://gitlab.com/bztsrc/bootboot
deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler
Raspbery Pi Firmware Copyright (c) Broadcom Corp, Raspberry Pi (Trading) Ltd
Validates ELF or PE executables for being BOOTBOOT compatible, otherwise
creates a bootable hybrid image or Option ROM image for your hobby OS.
Usage:
./mkbootimg check <kernel elf / pe>
./mkbootimg <configuration json> initrd.rom
./mkbootimg <configuration json> bootpart.bin
./mkbootimg <configuration json> <output disk image name>
Examples:
./mkbootimg check mykernel/mykernel.x86_64.elf
./mkbootimg myos.json initrd.rom
./mkbootimg myos.json bootpart.bin
./mkbootimg myos.json myos.img
If the first argument is check
, then it's followed by a kernel filename. The utility will check the executable for
BOOTBOOT compliance, and it will report all errors and if passed, which BOOTBOOT Protocol level it conforms to.
Otherwise the first argument is the configuration JSON file. If the second argument is initrd.rom
, then it will generate
a BIOS Option ROM image from the initrd directory. If that is bootpart.bin
, then it saves the boot partition image
(and only the partition image). Every other filename will make it generate a whole disk image with GPT.
The tool is multilingual. It will detect your operating system's language and if it has a dictionary for it, it will use that.
You can override the autodetection from the command line by using the -l <lang>
flag as the first argument (available for
all operating modes). Language is given in two characters long code and fallbacks to en
.
Configuration
The JSON is simple and flexible, accepts many variations. At the top level, you can define the output disk parameters.
Top Level
Field | Type | Description |
---|---|---|
diskguid | GUID | optional, the disk GUID. If not given, or full zeros, it will be generated |
disksize | integer | optional, the size of the disk image in Megabytes. If not given, it is calculated |
align | integer | optional, the partition alignment in Kilobytes. Zero gives sector alignment |
iso9660 | boolean | optional, wether to generate ISO9660 Boot Catalog into the image. Defaults to false |
config | filename | BOOTBOOT configuration file. It is parsed for the kernel filename |
initrd | struct | the initial ramdisk's definition, see below |
partitions | array | partition definitions, see below |
Example:
{
"diskguid": "00000000-0000-0000-0000-000000000000",
"disksize": 128,
"align": 1024,
"iso9660": true,
"config": "boot/sys/config",
"initrd": { "type": "tar", "gzip": true, "directory": "boot" },
"partitions": [
{ "type": "boot", "size": 16 },
{ "type": "ext4", "size": 128, "name": "Linux Exchange" },
{ "type": "ntfs", "size": 128, "name": "Windows Exchange" },
{ "type": "Microsoft basic data", "size": 32, "name": "MyOS usr", "file": "usrpart.bin" },
{ "type": "00000000-0000-0000-0000-000000000000", "size": 32, "name": "MyOS var", "file": "varpart.bin" }
]
}
Initrd
Field | Type | Description |
---|---|---|
gzip | boolean | optional, wether to compress the initrd image, defaults to true |
type | string | format of the initrd image. When invalid value given, it lists the options |
file | filename | the filename of the image file to be used |
directory | folder | path to a folder, its contents will be used to generate the initrd |
file | array | for multiarch images |
directory | array | for multiarch images |
The fields file
and directory
are mutually exclusive. They can be both strings (if there's only one architecture),
or arrays (one array element for each architecture). Currently three architecture supported, which means there can be
three strings in the arrays. Which architecture is used depends on the kernel's architecture in that folder or image
file. Type is only mandatory for directory
.
Examples:
"initrd": { "file": "initrd.bin" },
"initrd": { "type": "tar", "gzip": 0, "directory": "boot" },
"initrd": { "gzip": true, "file": [ "initrd-x86.bin", "initrd-arm.bin", "initrd-rv64.bin" ] },
"initrd": { "type": "cpio", "gzip": true, "directory": [ "boot/arm", "boot/x86", "boot/riscv64" ] },
Partitions
It is somewhat unusual, as the first array element is different than the rest. It specifies the boot partition,
therefore it has different types, and file
/ directory
and name
are not interpreted because that partition image is
always dinamically generated with the implicit name of "EFI System Partition". For the same reason, size
is mandatory
for the first (boot) partition.
Field | Type | Description |
---|---|---|
size | integer | optional, the size of the partition in Megabytes. If not given, it is calculated |
file | filename | optional, path to a partition image to be used |
directory | folder | optional, path to a folder, its contents will be used to generate the partition |
driver | string | optional, in case type can't specify the format without a doubt |
type | string | format of the partition. When invalid value given, it lists the options |
name | string | UTF-8 partition names, limited to UNICODE code points 32 to 65535 (BMP) |
For the first entry, valid values for type
are: boot
(or explicit fat16
and fat32
). Generates only 8+3 file names.
The utility handles this comfortably, it tries to use FAT16 if possible to save storage space. There's a minimal size
for the boot partition, 16 Megabytes. Although both the image creator and BOOTBOOT is capable of handling smaller sizes,
some UEFI firmware incorrectly assumes FAT12 when there are too few clusters on the file system. If the partition size is
bigger than 128 Megabytes, then it automatically switches to FAT32. If you don't use iso9660
, then you can also set FAT32
for smaller images, but at least 33 Megabytes (that's a hard lower limit for FAT32). With iso9660
, each cluster must
be 2048 bytes aligned, which is achieved by 4 sectors per cluster. The same problem applies here, both the image
creator and the BOOTBOOT loader capable of handling FAT32 with smaller cluster numbers, but some UEFI firmware don't,
and falsely assumes FAT16. To guarantee the minimum number of clusters, with ISO9660 and FAT32 the boot partition's
minimum size is 128 Megabytes (128*1024*1024/512/4 = 65536, just one larger than what fits in 16 bits).
For the other entries (starting from the second), type
is either a GUID or one of a pre-defined file system aliases.
Here fat
will decide between FAT16 and FAT32 based on the number of clusters, and it can generate long file names.
With an invalid string, the utility will list all possible values.
Example:
mkbootimg: partition #2 doesn't have a valid type. Accepted values:
"65706154-4120-6372-6968-766520465320" / "tar"
"5A2F534F-0000-5346-2F5A-000000000000" / "FS/Z"
"6A898CC3-1DD2-11B2-99A6-080020736631" / "ZFS"
"EBD0A0A2-B9E5-4433-87C0-68B6B72699C7" / "ntfs"
"0FC63DAF-8483-4772-8E79-3D69D8477DE4" / "ext4"
"516E7CB6-6ECF-11D6-8FF8-00022D09712B" / "ufs"
"C91818F9-8025-47AF-89D2-F030D7000C2C" / "p9"
"D3BFE2DE-3DAF-11DF-BA40-E3A556D89593" / "Intel Fast Flash"
"21686148-6449-6E6F-744E-656564454649" / "BIOS boot"
...
"77719A0C-A4A0-11E3-A47E-000C29745A24" / "VMware Virsto"
"9198EFFC-31C0-11DB-8F78-000C2911D1B8" / "VMware Reserved"
"824CC7A0-36A8-11E3-890A-952519AD3F61" / "OpenBSD data"
"CEF5A9AD-73BC-4601-89F3-CDEEEEE321A1" / "QNX6 file system"
"C91818F9-8025-47AF-89D2-F030D7000C2C" / "Plan 9 partition"
"5B193300-FC78-40CD-8002-E86C45580B47" / "HiFive Unleashed FSBL"
"2E54B353-1271-4842-806F-E436D6AF6985" / "HiFive Unleashed BBL"
...or any non-zero GUID in the form "%08X-%04X-%04X-%04X-%12X"
If file
given, then the partition is filled with data from that file. If size
is not given or smaller than
the file's size, then the file's size will be the partition's size. If both given, and size
is larger, then the
difference is filled up with zeros. Partition sizes will always be multiple of align
Kilobytes. Using 1024
as alignment gives you 1 Megabyte aligned partitions. For the first entry, only size
is valid, file
isn't.
Alternatively to file
, you might also be able to use directory
to generate the partition image from the contents
of a directory. This option is only available if the file system driver is implemented for type
. Because there might
be no one-to-one relation between partition types and file system types, you can use driver
to explicily set the
latter. This is only relevant when the directory
directive is used. For example:
{ "type": "5A2F534F-8664-5346-2F5A-000075737200", "driver": "FS/Z", "size": 32, "name": "usr", "directory": "myusr" },
{ "type": "Linux home", "driver": "minix", "size": 32, "name": "home", "directory": "myhome" },
{ "type": "Microsoft basic data", "driver": "fat", "size": 32, "name": "data", "directory": "mydata" },
Finally, name
is just an UTF-8 string, name of the partition. Maximum length is 35 characters. Not valid for the first entry.
Adding More File Systems
Types are listed in the fs registry, in the file fs.h
. You can freely add new file system types. For file systems that you
want to use for generating partition images or initrd as well, you must implement three functions, like:
void somefs_open(gpt_t *gpt_entry);
void somefs_add(struct stat *st, char *name, unsigned char *content, int size);
void somefs_close();
The first, the "open" is called whenever a new file system is to be created. The gpt_entry
is NULL when called for initrd
creation. As the given directory is recursively parsed, for each directory entry an "add" call is made. This should add the
file or directory to the file system image. Here st
is the stat struct for the file, name
is the filename with full path,
content
and size
are the file's content, or in case of a symbolic link, the pointed path. Finally when the parsing is
done, the "close" function is called to finalize the image. Only the "add" function is mandatory, the other two are optional.
These functions can use two global variables, fs_base
and fs_len
which holds the buffer for the filesystem image
in memory (this implies that partitions are limited to few gigabytes, depending how much RAM you have). In case they want
to report error, fs_no
is the number of the partition the driver is generating for.
In lack of these functions, the file system still can be used in the partition's type
field, but then only the GPT entry
will be created, not the content of the partition. The driver
field only accepts file system types which have these functions.
Keeping the built-in binaries up-to-date
To avoid dependencies, the image creator includes all the necessary binaries. If these are updated, then delete data.c
and run make
which will regenerate it. If there are missing files, then in the aarch64-rpi
directory run make getfw
,
that will download the latest Raspberry Pi firmware files. Then make
in this directory should run without problems.