Re: [PATCH 3/7] inflate pt1: clean up input logic

[John Reiser <jreiser@BitWagon.com>]

Matt Mackall wrote:
> On Sat, Feb 25, 2006 at 09:22:48PM +0000, Russell King wrote:
> 
>>init/do_mounts_rd.c:#include "../lib/inflate.c"
>>init/initramfs.c:#include "../lib/inflate.c"
>>
>>for these your arguments that halting is fine is _NOT_ correct nor is it
>>desirable.
> 
> 
> If you have an argument for why we shouldn't halt on failed
> init{rd,ramfs} decompression, I look forward to hearing it.

It depends on the nature of the error, and which parts were decompressed
successfully.  Gzip has optional "re-sync" capability, and ideally a
decompression failure might invoke some kind of bad-block tagging
for the output instead of halting the machine.  Some init{rd,ramfs}
have all the network drivers, all the SCSI drivers, all the sound
drivers, etc., but the user may care only about those for the current
machine.  Other init{rd,ramfs} contain only "essential" pieces.
Even then, the pieces that are deemed more important can be at the
beginning.  It might be possible to work without a sound driver,
but perhaps not without a SCSI driver.

> In my mind, being unable to decompress init* is every bit as fatal as
> being unable to mount root.

It may be possible to recover, at least partially, from one error
more than from another.  It would be nice if the decompressor itself
was a minor influence instead of a major one.

--