Do not allow to load ET_DYN object with DF_1_PIE flag set.

[kib]

Summary

Linkers are supposed to mark PIE binaries with DF_1_PIE, such binary cannot be correctly and usefully loaded neither by dlopen(3) nor as a dependency of other object.

Requested by: emaste

Test Plan

I cannot test, because it seems -fPIE -Wl,-pie is broken for all current toolchains, both base and ports.

[emaste]

lld 10 refuses to link with -fPIE -Wl,--pie with errors like

ld: error: cannot preempt symbol: _DYNAMIC
>>> defined in 
>>> referenced by crt1.c:62 (/usr/home/emaste/src/freebsd-svn/head/lib/csu/amd64/crt1.c:62)
>>>               /usr/lib/crt1.o:(_start)
...

lld 8 proceeds:

$ cc -fuse-ld=lld80 -fPIE -Wl,--pie -Wl,-z,notext ~/hello.c
$

but the resulting binary segfaults on startup

[kib]

In D25086#552699, @emaste wrote:

lld 10 refuses to link with -fPIE -Wl,--pie with errors like

ld: error: cannot preempt symbol: _DYNAMIC
>>> defined in 
>>> referenced by crt1.c:62 (/usr/home/emaste/src/freebsd-svn/head/lib/csu/amd64/crt1.c:62)
>>>               /usr/lib/crt1.o:(_start)
...

lld 8 proceeds:

$ cc -fuse-ld=lld80 -fPIE -Wl,--pie -Wl,-z,notext ~/hello.c
$

but the resulting binary segfaults on startup

Problem is that all drivers link with crt1.o instead of Scrt1.o. See r200038.

[emaste]

Problem is that all drivers link with crt1.o instead of Scrt1.o. See r200038.

Hrm,

$ cc -g -pie -Wl,--pie -Wl,-z,notext ~/hello.c
$ ./a.out
Hello, world
$ file a.out
a.out: ELF 64-bit LSB shared object, x86-64, version 1 (FreeBSD), dynamically linked, interpreter /libexec/ld-elf.so.1, for FreeBSD 13.0 (1300087), FreeBSD-style, with debug_info, not stripped

[emaste]

Following up after discussions on IRC, the canonical way do do this is:
CFLAGS=-fPIC
LDFLAGS=-pic
or if compiling a test app on the command line,
cc -fPIC -pic ...

[emaste]

This looks reasonable to me. We'll need to bring in a couple of LLVM commits to be able to test.

[kib]

In D25086#552814, @emaste wrote:

This looks reasonable to me. We'll need to bring in a couple of LLVM commits to be able to test.

I tested it with gcc 10.1 / binutils 2.34.

[arichardson]

head/libexec/rtld-elf/rtld.c

2586

Since this could be a user-visible message, maybe spell out dynamic shared object or use uppercase DSO to indicate that it's an abbreviation?