Reimplement r266925 after r300793.

[dchagin]

Summary

Use execpath if possible. To allow to run the interpreter itself add a new ELF branding type. Allow Linux ABI to run ELF interpreter.

[kib]

Apart from the minor notes which I put inline, I have global and strong objection against the proposed addition. We do not base any decisions about binary ABI based on the binary name, at least we did not up to this point.

And I do not see why would we start claiming that some random binary found to have a name ld-linux.so.2 is neccessarly the Linux glibc interpreter.

IMO something more precise must be developed if the need to properly brand interpreter alone is important.

sys/kern/imgact_elf.c

366

fname might be NULL there.

367

Why '<' and not '!=' ?

369

Why strncmp ?

[dchagin]

In D6968#146101, @kib wrote:

Apart from the minor notes which I put inline, I have global and strong objection against the proposed addition. We do not base any decisions about binary ABI based on the binary name, at least we did not up to this point.

And I do not see why would we start claiming that some random binary found to have a name ld-linux.so.2 is neccessarly the Linux glibc interpreter.

np

IMO something more precise must be developed if the need to properly brand interpreter alone is important.

unfortunately, there is no other way except to compare the file name or use fallback_brand, which is not always suitable