Explicitly pass -no-pie

[sfackler]

Some linkers (e.g. Debian and Arch's) that are configured to make PIEs
by default will make dynamically linked executables that don't actually
dynamically link to anything. Fix that by explicitly passing -no-pie in
those cases.

Closes #43647

r? @alexcrichton

[alexcrichton]

@bors: r+

[bors]

📌 Commit 1d838a2 has been approved by alexcrichton

[aidanhs]

I'm 80% sure that #43647 is already fixed - comparing the link args , the one missing link flag before and after #40113 (aside from -l unwind cc #44069) is -static. Running rustc -C link-args="-static" --target x86_64-unknown-linux-musl x.rs on stable on an arch linux that reproduces the issue, produces a full static binary (as -static appears to default to -no-pie).

So I don't see what we're fixing by merging this PR. Even if -static defaulted to -pie, that's valid and we should respect the system default - fully static PIE binaries work fine.

All of this aside, I think we should postpone this PR so we can fix #44069 first.

[alexcrichton]

@aidanhs do you think, though, that we shouldn't land this?

[aidanhs]

@alexcrichton that is correct. pie for shared executables is pretty widely accepted as A Good Thing, but the case is a bit murkier for static executables (until recently-ish I think it required some arcane incantations to get a pie static executable from gcc).

I think we should defer to the system linker and leave pie alone for static executables - most systems will end up with non-pie since that's what gcc seems to default to for -static, but I see no advantage to enforcing it (and some disadvantages, like not being able to opt-in to ASLR on a system-level).

[alexcrichton]

@bors: r-

Ok @sfackler do you agree as well?

Should we close #43647

[aidanhs]

I'm going to test #43647 with the nightly tonight, I'll close if it seems fixed.

[sfackler]

I'm happy as long as binaries that don't link to glibc or anything else dynamically don't depend on a dynamic loader.

[alexcrichton]

Ok, if @aidanhs's tests come back green sounds like we can close!

[aidanhs]

Closing per #43647 (comment)