1=======================
2Early userspace support
3=======================
4
5Last update: 2004-12-20 tlh
6
7
8"Early userspace" is a set of libraries and programs that provide
9various pieces of functionality that are important enough to be
10available while a Linux kernel is coming up, but that don't need to be
11run inside the kernel itself.
12
13It consists of several major infrastructure components:
14
15- gen_init_cpio, a program that builds a cpio-format archive
16  containing a root filesystem image.  This archive is compressed, and
17  the compressed image is linked into the kernel image.
18- initramfs, a chunk of code that unpacks the compressed cpio image
19  midway through the kernel boot process.
20- klibc, a userspace C library, currently packaged separately, that is
21  optimized for correctness and small size.
22
23The cpio file format used by initramfs is the "newc" (aka "cpio -H newc")
24format, and is documented in the file "buffer-format.txt".  There are
25two ways to add an early userspace image: specify an existing cpio
26archive to be used as the image or have the kernel build process build
27the image from specifications.
28
29CPIO ARCHIVE method
30-------------------
31
32You can create a cpio archive that contains the early userspace image.
33Your cpio archive should be specified in CONFIG_INITRAMFS_SOURCE and it
34will be used directly.  Only a single cpio file may be specified in
35CONFIG_INITRAMFS_SOURCE and directory and file names are not allowed in
36combination with a cpio archive.
37
38IMAGE BUILDING method
39---------------------
40
41The kernel build process can also build an early userspace image from
42source parts rather than supplying a cpio archive.  This method provides
43a way to create images with root-owned files even though the image was
44built by an unprivileged user.
45
46The image is specified as one or more sources in
47CONFIG_INITRAMFS_SOURCE.  Sources can be either directories or files -
48cpio archives are *not* allowed when building from sources.
49
50A source directory will have it and all of its contents packaged.  The
51specified directory name will be mapped to '/'.  When packaging a
52directory, limited user and group ID translation can be performed.
53INITRAMFS_ROOT_UID can be set to a user ID that needs to be mapped to
54user root (0).  INITRAMFS_ROOT_GID can be set to a group ID that needs
55to be mapped to group root (0).
56
57A source file must be directives in the format required by the
58usr/gen_init_cpio utility (run 'usr/gen_init_cpio -h' to get the
59file format).  The directives in the file will be passed directly to
60usr/gen_init_cpio.
61
62When a combination of directories and files are specified then the
63initramfs image will be an aggregate of all of them.  In this way a user
64can create a 'root-image' directory and install all files into it.
65Because device-special files cannot be created by a unprivileged user,
66special files can be listed in a 'root-files' file.  Both 'root-image'
67and 'root-files' can be listed in CONFIG_INITRAMFS_SOURCE and a complete
68early userspace image can be built by an unprivileged user.
69
70As a technical note, when directories and files are specified, the
71entire CONFIG_INITRAMFS_SOURCE is passed to
72usr/gen_initramfs_list.sh.  This means that CONFIG_INITRAMFS_SOURCE
73can really be interpreted as any legal argument to
74gen_initramfs_list.sh.  If a directory is specified as an argument then
75the contents are scanned, uid/gid translation is performed, and
76usr/gen_init_cpio file directives are output.  If a directory is
77specified as an argument to usr/gen_initramfs_list.sh then the
78contents of the file are simply copied to the output.  All of the output
79directives from directory scanning and file contents copying are
80processed by usr/gen_init_cpio.
81
82See also 'usr/gen_initramfs_list.sh -h'.
83
84Where's this all leading?
85=========================
86
87The klibc distribution contains some of the necessary software to make
88early userspace useful.  The klibc distribution is currently
89maintained separately from the kernel.
90
91You can obtain somewhat infrequent snapshots of klibc from
92https://www.kernel.org/pub/linux/libs/klibc/
93
94For active users, you are better off using the klibc git
95repository, at https://git.kernel.org/?p=libs/klibc/klibc.git
96
97The standalone klibc distribution currently provides three components,
98in addition to the klibc library:
99
100- ipconfig, a program that configures network interfaces.  It can
101  configure them statically, or use DHCP to obtain information
102  dynamically (aka "IP autoconfiguration").
103- nfsmount, a program that can mount an NFS filesystem.
104- kinit, the "glue" that uses ipconfig and nfsmount to replace the old
105  support for IP autoconfig, mount a filesystem over NFS, and continue
106  system boot using that filesystem as root.
107
108kinit is built as a single statically linked binary to save space.
109
110Eventually, several more chunks of kernel functionality will hopefully
111move to early userspace:
112
113- Almost all of init/do_mounts* (the beginning of this is already in
114  place)
115- ACPI table parsing
116- Insert unwieldy subsystem that doesn't really need to be in kernel
117  space here
118
119If kinit doesn't meet your current needs and you've got bytes to burn,
120the klibc distribution includes a small Bourne-compatible shell (ash)
121and a number of other utilities, so you can replace kinit and build
122custom initramfs images that meet your needs exactly.
123
124For questions and help, you can sign up for the early userspace
125mailing list at https://www.zytor.com/mailman/listinfo/klibc
126
127How does it work?
128=================
129
130The kernel has currently 3 ways to mount the root filesystem:
131
132a) all required device and filesystem drivers compiled into the kernel, no
133   initrd.  init/main.c:init() will call prepare_namespace() to mount the
134   final root filesystem, based on the root= option and optional init= to run
135   some other init binary than listed at the end of init/main.c:init().
136
137b) some device and filesystem drivers built as modules and stored in an
138   initrd.  The initrd must contain a binary '/linuxrc' which is supposed to
139   load these driver modules.  It is also possible to mount the final root
140   filesystem via linuxrc and use the pivot_root syscall.  The initrd is
141   mounted and executed via prepare_namespace().
142
143c) using initramfs.  The call to prepare_namespace() must be skipped.
144   This means that a binary must do all the work.  Said binary can be stored
145   into initramfs either via modifying usr/gen_init_cpio.c or via the new
146   initrd format, an cpio archive.  It must be called "/init".  This binary
147   is responsible to do all the things prepare_namespace() would do.
148
149   To maintain backwards compatibility, the /init binary will only run if it
150   comes via an initramfs cpio archive.  If this is not the case,
151   init/main.c:init() will run prepare_namespace() to mount the final root
152   and exec one of the predefined init binaries.
153
154Bryan O'Sullivan <bos@serpentine.com>
155