Skip to contentSkip to conversation

Keep initrd tasks running?

DISCUSSION John Richard Moser 6 participants 11 messages Oct 18, 2005 – Oct 23, 2005

lore
Share & tools
lore
#0 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have no idea who's the best to ask for this.

I want to start a task in an initrd and have it stay running after init is started. Pretty much:

- kernel boot
- initrd loaded
- linuxrc executes
- /bin/mydaemon runs
- mount rootfs
- pivot_root
- exec /sbin/init (PID=1; linuxrc and sh is replaced)
- mydaemon keeps running, reparented under init, uninterrupted

What's the feasibility of this without the system balking and vomiting chunks everywhere? I'm pretty sure 'exec /sbin/init' from linuxrc (PID=1) will replace the process image of sh (linuxrc) with init, keeping PID=1; but I'm worried this may terminate children too. Haven't tried.

(this actually has a useful application)
- --
All content of all messages exchanged herein are left in the Public Domain, unless otherwise explicitly stated.

Creative brains are a valuable, limited resource. They shouldn't be wasted on re-inventing the wheel when there are so many fascinating new problems waiting out there.
-- Eric Steven Raymond
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDVUlKhDd4aOud5P8RAuTeAJ9z7F+aeLZgnHzuzyviSXhJG/d5egCcCoxs GpD7wLP+ZngCsYFLGsmEXb0=
=w6xA
-----END PGP SIGNATURE-----

#1 lore raw
Quoted 10 lines · John Richard Moser

I have no idea who's the best to ask for this.

I want to start a task in an initrd and have it stay running after init is started. Pretty much:

What's the feasibility of this without the system balking and vomiting chunks everywhere? I'm pretty sure 'exec /sbin/init' from linuxrc (PID=1) will replace the process image of sh (linuxrc) with init, keeping PID=1; but I'm worried this may terminate children too. Haven't tried.

This is much more easily supported in Breezy. usplash is started at the top of the initramfs (from the init-top hook) and lives until we start gdm.

The biggest constraint is that you don't have write access to the target root filesystem (since it's mounted readonly). However, /dev is a tmpfs that is move mounted to the new root system. If you need to have sockets open or store data, you can use that. usplash does this for its socket.

Note that the initramfs startup sequence isn't at all similar to the old initrd startups. It should be easy for you to cleanly add what you want under /etc/mkinitramfs/scripts and not have to modify the initramfs-tools package. /usr/share/doc/initramfs-tools/HACKING contains some starter information.

Hope this helps!

Tks,
Jeff Bailey

#2 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 15 lines · Jeff Bailey
Quoted 11 lines · John Richard Moser

I have no idea who's the best to ask for this.

I want to start a task in an initrd and have it stay running after init is started. Pretty much:

What's the feasibility of this without the system balking and vomiting chunks everywhere? I'm pretty sure 'exec /sbin/init' from linuxrc (PID=1) will replace the process image of sh (linuxrc) with init, keeping PID=1; but I'm worried this may terminate children too. Haven't tried.

Quoted 4 lines

This is much more easily supported in Breezy. usplash is started at the top of the initramfs (from the init-top hook) and lives until we start gdm.

So in short it's possible?

Quoted 6 lines

The biggest constraint is that you don't have write access to the target root filesystem (since it's mounted readonly). However, /dev is a tmpfs that is move mounted to the new root system. If you need to have sockets open or store data, you can use that. usplash does this for its socket.

That's not much of a problem for me. What I'm contemplating is a FUSE file system driver that gets started in the initrd, and a kernel that has a file system driver built-in for something stupid like cramfs or MINIX.

The idea is that (as proof of concept) it should be possible to supply something like ext3 as a FUSE driver, and boot off it as the rootfs without building ext3 into the kernel or ever modprobing it. Besides just being damn cool, and a show of quasi-hybrid microkernelism at work (uh oh now Linus is going to remove FUSE from mainline :), it'd allow some real visible macro-benchmarking of FUSE.

Quoted 6 lines

Note that the initramfs startup sequence isn't at all similar to the old initrd startups. It should be easy for you to cleanly add what you want under /etc/mkinitramfs/scripts and not have to modify the initramfs-tools package. /usr/share/doc/initramfs-tools/HACKING contains some starter information.

Quoted 8 lines

Hope this helps!

Tks,
Jeff Bailey

- --
All content of all messages exchanged herein are left in the Public Domain, unless otherwise explicitly stated.

Creative brains are a valuable, limited resource. They shouldn't be wasted on re-inventing the wheel when there are so many fascinating new problems waiting out there.
-- Eric Steven Raymond
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDVU8bhDd4aOud5P8RAoFSAJwImO0TNH5tJKyrhZ7mSs8zUWDMUwCfYCT6 0IaiHMCZxglMNnlVTvLHm8o=
=+0g0
-----END PGP SIGNATURE-----

#3 lore raw
Quoted 5 lines · John Richard Moser
Quoted 3 lines

This is much more easily supported in Breezy. usplash is started at the top of the initramfs (from the init-top hook) and lives until we start gdm.

So in short it's possible?

I make no promises until I"m holding the code in my hand and have tested it myself. =) I don't see any reason why it wouldn't work.

Quoted 3 lines

That's not much of a problem for me. What I'm contemplating is a FUSE file system driver that gets started in the initrd, and a kernel that has a file system driver built-in for something stupid like cramfs or MINIX.

Ah, okay. Jerry Haltom was looking at how to support booting from a loopback filesystem (for embedded systems shiping a .bin blob) and he was able to get it running. This is hopefully just another twist on the same thing. I haven't played with FUSE, though, so I don't know what's needed.

Lemme know if you run into specific problems. If you have access to IRC, I'm on freenode as jbailey in #ubuntu-devel.

Tks,
Jeff Bailey

#4 lore raw
In reply to Jeff Bailey

I am confused. I thought that once the initramfs init execs the real init, the initramfs is freed. It can't be freed if there are processes that still have open files there, so that would seem to prevent any processes being started in the initramfs and continuing after the real system is booted.

Quoted 27 lines · Jeff Bailey

This is much more easily supported in Breezy. usplash is started at the top of the initramfs (from the init-top hook) and lives until we start gdm.

The biggest constraint is that you don't have write access to the target root filesystem (since it's mounted readonly). However, /dev is a tmpfs that is move mounted to the new root system. If you need to have sockets open or store data, you can use that. usplash does this for its socket.

Note that the initramfs startup sequence isn't at all similar to the old initrd startups. It should be easy for you to cleanly add what you want under /etc/mkinitramfs/scripts and not have to modify the initramfs-tools package. /usr/share/doc/initramfs-tools/HACKING contains some starter information.

Hope this helps!

Tks,
Jeff Bailey

#5 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 6 lines · Phillip Susi

I am confused. I thought that once the initramfs init execs the real init, the initramfs is freed. It can't be freed if there are processes that still have open files there, so that would seem to prevent any processes being started in the initramfs and continuing after the real system is booted.

AFAIK it's pivoted and then umounted, which frees it. This doesn't mean it has to be freed.. . .

Quoted 20 lines
Quoted 18 lines · Jeff Bailey

This is much more easily supported in Breezy. usplash is started at the top of the initramfs (from the init-top hook) and lives until we start gdm.

The biggest constraint is that you don't have write access to the target root filesystem (since it's mounted readonly). However, /dev is a tmpfs that is move mounted to the new root system. If you need to have sockets open or store data, you can use that. usplash does this for its socket.

Note that the initramfs startup sequence isn't at all similar to the old initrd startups. It should be easy for you to cleanly add what you want under /etc/mkinitramfs/scripts and not have to modify the initramfs-tools package. /usr/share/doc/initramfs-tools/HACKING contains some starter information.

Hope this helps!

Quoted 11 lines
Quoted 9 lines

Tks,
Jeff Bailey

- --
All content of all messages exchanged herein are left in the Public Domain, unless otherwise explicitly stated.

Creative brains are a valuable, limited resource. They shouldn't be wasted on re-inventing the wheel when there are so many fascinating new problems waiting out there.
-- Eric Steven Raymond
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDVcnzhDd4aOud5P8RAmk/AJ48RK9V5CrJeNXu/ORxuNa3I/+jrACeOkUl z2l1tNTWm6dtjysddaZ81co=
=7GLq
-----END PGP SIGNATURE-----

#6 lore raw

I thought that the key difference between initrd and initramfs is that initrd uses pivot_root, initramfs does not. From what I have read, some consider pivit_root to be an ugly hack, so initramfs was made in such a way that you don't pivot. Is this incorrect?

The whole reason that pivot was made was to allow the initrd filesystem to be moved and the real root mounted, all while processes ( at least init, because it can't exit ) continued to execute, maintaining their open files on the initrd filesystem. Just instead of those files being in / they got moved to /initrd.

Under that system it is possible to leave things running, but I thought that with initramfs, init actually exited and the kernel unmounted the initramfs, and mounted the real root and ran the real init.

Quoted 5 lines · John Richard Moser

AFAIK it's pivoted and then umounted, which frees it. This doesn't mean it has to be freed.. . .

#7 lore raw
Quoted 2 lines · Phillip Susi

thought that with initramfs, init actually exited and the kernel unmounted the initramfs, and mounted the real root and ran the real init.

This is not what happens; look at run-init in klibc if you want to understand how it does work.

Signature · 1 line
 - mdz
#8 lore raw
In reply to John Richard Moser
Quoted 10 lines · John Richard Moser
Quoted 6 lines · Phillip Susi

I am confused. I thought that once the initramfs init execs the real init, the initramfs is freed. It can't be freed if there are processes that still have open files there, so that would seem to prevent any processes being started in the initramfs and continuing after the real system is booted.

AFAIK it's pivoted and then umounted, which frees it. This doesn't mean it has to be freed.. . .

It is neither pivoted nor unmounted.

Signature · 1 line
 - mdz
#9 lore raw
In reply to John Richard Moser
Quoted 19 lines · John Richard Moser

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have no idea who's the best to ask for this.

I want to start a task in an initrd and have it stay running after init is started. Pretty much:

- kernel boot
- initrd loaded
- linuxrc executes
- /bin/mydaemon runs
- mount rootfs
- pivot_root
- exec /sbin/init (PID=1; linuxrc and sh is replaced)
- mydaemon keeps running, reparented under init, uninterrupted

Quoted 7 lines

What's the feasibility of this without the system balking and vomiting chunks everywhere? I'm pretty sure 'exec /sbin/init' from linuxrc (PID=1) will replace the process image of sh (linuxrc) with init, keeping PID=1; but I'm worried this may terminate children too. Haven't tried.

(this actually has a useful application)

Note that the 'init' of linuxrc (/sbin/nash or /sbin/sash) also has a PID of 1. It gets replaced with /sbin/init or whatever you use for an emergency shell.

You can load a kernel thread from a driver that gets installed by initrd. This would allow you to have processes running with PIDs greater than 1 before the final init gets started. The present kernel starts quite a few threads up to PID=8 right now.

Once 'init' gets started, you are going to have either new 'inits' exec(ed) with PID 1 or children with PIDs greater than 1.

Cheers,
Dick Johnson
Penguin : Linux version 2.6.13.4 on an i686 machine (5589.55 BogoMips). Warning : 98.36% of all statistics are fiction.
.

****************************************************************
The information transmitted in this message is confidential and may be privileged. Any review, retransmission, dissemination, or other use of this information by persons or entities other than the intended recipient is prohibited. If you are not the intended recipient, please notify Analogic Corporation immediately - by replying to this message or by sending an email to [email protected] - and destroy all copies of this information, including any attachments, without reading or disclosing them.

Thank you.

#10 lore raw
In reply to John Richard Moser
Quoted 19 lines · John Richard Moser

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I have no idea who's the best to ask for this.

I want to start a task in an initrd and have it stay running after init is started. Pretty much:

- kernel boot
- initrd loaded
- linuxrc executes
- /bin/mydaemon runs
- mount rootfs
- pivot_root
- exec /sbin/init (PID=1; linuxrc and sh is replaced)
- mydaemon keeps running, reparented under init, uninterrupted

Quoted 5 lines

What's the feasibility of this without the system balking and vomiting chunks everywhere? I'm pretty sure 'exec /sbin/init' from linuxrc (PID=1) will replace the process image of sh (linuxrc) with init, keeping PID=1; but I'm worried this may terminate children too. Haven't tried.

It won't terminate children. Try it manually by booting with init=/bin/sh. --
vda