Why initramfs stops although root device is mountable and ready to use

[Shachar Shemesh]

Ilya A. Volynets-Evenbakh wrote:
>
> I think it's the other way around - if you exit from your linuxrc when
> running from initramfs,
> it'll continue on with standard boot sequence, while ending linuxrc in
> initrd will panic the kernel
> with "Trying to kill PID 1" error.
>
>   
An initramfs exists for any 2.6 kernel. Sometimes it's just a very small 
one. The way the kernel decides whether to execute it is to check for a 
*/init* file in it. If it finds one, it runs it. If not, it does not.

As such, if you have a /linuxrc file in your initramfs, and it simply 
exists at the end, all that will happen is that the kernel will not find 
a /init and will skip running anything from the initramfs at all. In 
that sense you might be said to be right that if your linuxrc exists, 
the normal boot sequence will continue.

The history goes something like this:
Originally, the drivers for the root file system had to be compiled into 
the kernel. This was very difficult, if not impossible, for distribution 
kernels, as the list of potential drivers included, well, everything, 
and the kernels were just too big. Then the idea of initrd came - we 
compile the drivers only for the ram disk and the initrd file system 
(typically, cramfs) into the kernel. The initrd loads the relevant 
drivers into the kernel and quits, restoring the original boot sequence. 
Then someone pointed out that, sometimes, an initrd is all you need. The 
convention then arose to pass "boot=/dev/ram0" to the kernel, which 
tells it that the initrd stage is the last one. It then became common to 
put the real "boot=" option on the kernel command line, but then, from 
initrd, to load the values for "/dev/ram0" into the proc entry for the 
boot device from linuxrc, manually mount the real root file system, make 
it the root file system using pivot_root, and exec init from there.

Then, for 2.6, people figured that if this is what you normally do 
anyways, there is no point in carrying around the drivers for the ram 
disk and cramfs (compiled into the kernel, which means it cannot be 
unloaded). Instead, use the much lighter tmpfs file system, and use a 
cpio archive for the actual files. The tmpfs file system was made 
mandatory, whether initramfs was used or not, and so you couldn't move 
it from the root (only mount on top of it). This means you cannot use 
pivot_root, and switch_root was invented. While at it, the "legacy" boot 
sequence was removed. Assuming, as Gilad seems to, that we leave initrd 
is out of the discussion, your options are either run an initramfs and 
perform the entire root mounting from the /init script there, or compile 
the root file system drivers into the kernel and let the kernel mount 
them using the boot= kernel option.

Shachar

-- 
Shachar Shemesh
Lingnu Open Source Consulting Ltd.
http://www.lingnu.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.huji.ac.il/pipermail/linux-il/attachments/20090707/c615b8d5/attachment.html>