Factor out retrieving the interpreter path from the main ELF loader routine.

[trasz]

Summary

Factor out retrieving the interpreter path from the main ELF
loader routine.

One small difference is that it no longer errors out if there
are multiple PT_INTERP sections; it just uses the first one.

[kib]

If you only move some parts of the case PT_INTERP into the helper, you can 1. avoid one more pass over phdrs 2. easily keep the logic to error on more than one interpreter.

Also I do not completely like the removal of the interp_name_len argument from the get_brandinfo args.

sys/kern/imgact_elf.c

285 ↗

(On Diff #55458)

If you insist on doing this arg removal, at least write this code as

interp_name_len = interp != NULL ? strlen(interp) : 0;

[trasz]

Kib's suggestions.

Updating D19715: Factor out retrieving the interpreter path from the main ELF

loader routine.

[kib]

sys/kern/imgact_elf.c

901 ↗

(On Diff #55461)

Add assertions: phdr type is PT_INTERP, vp is exclusively locked.

902 ↗

(On Diff #55461)

curthread is used only once, it seems. I do not see a need for td.

930 ↗

(On Diff #55461)

If you restore the 'else' clause, then *interpp = interp; return (0); can be left once in the end of the function.

[trasz]

Improve.

Updating D19715: Factor out retrieving the interpreter path from the main ELF

loader routine.

[trasz]

sys/kern/imgact_elf.c

930 ↗

(On Diff #55461)

I find the current version easier to read, to be honest.

[kib]

sys/kern/imgact_elf.c

930 ↗

(On Diff #55461)

For me (the reader) it is reverse. I need to understand that all successful exit paths satisfy the function interface, and single success-exit with non-duplicated code makes it easier to see.

I would not stop you from committing this due to the rearrangement of three lines, but I prefer the described version.