[Toybox] [PATCH] fix dmesg default size/level filtering/performance

[Rob Landley]

On 11/26/14 15:10, enh wrote:
> On Tue, Nov 25, 2014 at 7:42 PM, Rob Landley <rob at landley.net> wrote:
>>> +    if (size < 0) size = 16384;
>>
>> Hmmm, klogctl type 10 added in the 2.6.6 kernel... I think I'm ok with
>> forcing earlier kernels to specify -s for now, and waiting for somebody
>> to complain. (I want to perror_exit() when it gives an error, because it
>> shouldn't do that...)
> 
> yeah, sorry, i meant to mention that. (it slipped my mind because for
> the Android platform the minimum supported kernel version is always
> pretty high. i think it's 3.4 right now, for example.)

People still use older kernels in stuff like:
http://www.bigmessowires.com/2014/11/17/68-katy-68000-linux-on-a-solderless-breadboard/

Which is cool... but doesn't support even half the modern architectures,
and one of the design assumptions I made was a posix-2008 build
enviornment. (2.6.0 was 2003.) I'm using all sorts of openat() variants
that just weren't there in kernels much older than 2008.

So yeah, yay klogctl type 10.

>> The giant comment at the start of lib/args.c explains the syntax, and
>> there's a somewhat more verbose introduction at:
>>
>> http://landley.net/toybox/code.html#lib_args
>>
>> The FLAG_x macros are generated automatically, you mask them against
>> toys.optargs to see if the relevant bit is set. (Each option letter gets
>> a bit, in the same order as binary digits. So for "abcd", d=1, c=2, b=4,
>> a=8. Sometimes we use the value of a flag to simpify the code, but I try
>> to comment that when it happens.)
> 
> i think the one piece i'm missing to start working on the SELinux
> flags is how to probe for the library. i've seen the recent examples
> of probing for header files, but i'm guessing that for the -lselinux
> side of things i need to edit scripts/make.sh?
> 
> -for i in util crypt m resolv
> +for i in util crypt m resolv selinux

Should work. If the --as-needed doesn't fully drop out selinux.so when
it's configed out (hopefully it doesn't pull the same sort of weak
symbol shenanigans -lpthread does) I can always add a config check here
so it's only in the list when it's enabled in kconfig.

I normally try to avoid external library dependencies (to the point I'm
implementing my own inflate/deflate code rather than sucking in zlib)
for similar reasons busybox does, as described at the end of the
features part of http://landley.net/toybox/design.html .

(Basically I want a minimal self-bootstrapping system to require as few
packages as possible to get up and running on a new target, so you don't
have to cross compile anything more than absolutely necessary before you
can start natively building under the new target. It should also
statically link to a completely self-contained binary you can drop on a
system when doing rescue sort of things. And I want everything it does
to be auditable, which is harder if it includes code from other
packages; the interactions between different implementations of that
package may not be something we ever tested.)

That said, as long as selinux is an optional thing you can completely
switch off, I don't have to personally be a fan of it, or a domain expert.

A bigger concern is you might want to ping the tizen guys who've already
implemented -Z based on libattr and libsmack. Their repository with
their additions is at:

https://git.tizen.org/cgit/platform/upstream/toybox.git

No idea what their upstreaming plans are. I should ping Xavier and see
if he's caught up from his broken leg yet...

Rob

 1417072207.0