Re: $ORIGIN

[christos%astron.com@localhost (Christos Zoulas)]

In article <20140125080410.GA11641%netbsd.org@localhost>,
David Holland  <dholland-tech%netbsd.org@localhost> wrote:
>So, a large part of the reason $ORIGIN still doesn't work is that we
>have believed that it isn't possible to identify binaries that use
>$ORIGIN from within the kernel, so supporting it fully would add
>overhead to all execs.
>
>However, while trawling through elf.h tonight looking for something
>else, I noticed this:
>
>   /* Flag values for DT_FLAGS */
>   #define DF_ORIGIN       0x00000001      /* uses $ORIGIN */
>
>and it appears with some quick experimentation that with at least the
>amd64 toolchain, this flag appears (in a flags entry in the dynamic
>section) if and only if "-z origin" is given when linking:
>
>   % gcc foo.o -o foo
>   % readelf -d foo | grep FLAG
>   Exit 1
>   % gcc -Wl,-z,origin foo.o -o foo
>   % readelf -d foo | grep FLAG
>    0x000000006ffffffb (FLAGS_1)            Flags: ORIGIN
>   % 
>
>Which seems to me to mean that we can (with only a bit of fuss) find
>this flag during exec and do the (somewhat) expensive extra lookups
>for supporting $ORIGIN only if it's found.
>
>Am I missing something?

The problem is that we don't have a reliable name -> inode mapping. It mostly
does not have to do with performance.

christos