Uses a `spin::Mutex` for the global static `_r_debug` instance to remove
warning about a global mutable static.
Guarantees exclusive mutable access to the `_r_debug` variable.
Fixes warning about not using compiler intrinsics.
Calls into the `panic` macro instead of directly calling the `abort`
intrinsic.
Slightly changes the behavior of the function by printing the
panic message, unwinding, and aborting.
Closes: #212
The fix is simply to not follow links when opening a file to be renamed.
O_NOFOLLOW, a non-POSIX extension, does exactly that while not needing
renameat or openat.
`pthread_atfork` should be exported in pthread.h according to the
standard. We only exported our function in unistd.h. `glibc` exports it
in both pthread.h and unistd.h whereas musl only exports it in pthread.h
(which is standards compliant).
I exported it in both headers. Cbindgen doesn't seem to reexport `pub
use` so I declared the function twice. We might have to reexamine our
`pub use` to check what's exported and what isn't.
Linux's variant uses the syscall as intended. Redox's variant uses fpath
to build a path to pass to fstat from the file descriptor plus the file
name. Unlike the syscall, this isn't atomic so the liminal space between
fpath/getcwd and fstat is subject to TOCTOU.
Beyond fstatat, I moved the stat test to its correct location and added
an assert since the output of the test is unchecked.
I also added AT_FDCWD which seems to be -100 across Unixes. The other
AT_* constants are unimplemented for now.
The implementation for confstr is straightforward. Most of the constants
return 1 on both musl and glibc. The only constant that doesn't is
required for Fish.
I also switched an #[unsafe(no_mangle)] from my last patch back
to a #[no_mangle] because we need to bump cbindgen. #[unsafe(no_mangle)]
is required for Rust 2024.
I copied the constants and definitions from musl which we can presume is
correct. Relibc is missing some features which affects our sysconf as
well, such as constants in limits.h or a working resources.h.
Most platforms use a c_int though glibc uses a u32. GNU defines a type,
__rlimit_resource_t, for the RLIMIT enum that is a u32.
Using a c_int is nicer for a few reasons. The first is that our
(unimplemented) functions receive a c_int for the enum - using a u32 is
technically wrong but doesn't affect anything since the constants are
turned into macros by cbindgen.
The second reason is that a c_int is nicer for libc and nix (the crate)
too since we don't need to pollute the crates with guards for Redox.
elle