Refactor weak symbols in std::sys::unix

This makes a few changes to the weak symbol macros in `sys::unix`:

- `dlsym!` is added to keep the functionality for runtime `dlsym`
  lookups, like for `__pthread_get_minstack@GLIBC_PRIVATE` that we don't
  want to show up in ELF symbol tables.
- `weak!` now uses `#[linkage = "extern_weak"]` symbols, so its runtime
  behavior is just a simple null check. This is also used by `syscall!`.
  - On non-ELF targets (macos/ios) where that linkage is not known to
    behave, `weak!` is just an alias to `dlsym!` for the old behavior.
- `raw_syscall!` is added to always call `libc::syscall` on linux and
  android, for cases like `clone3` that have no known libc wrapper.

The new `weak!` linkage does mean that you'll get versioned symbols if
you build with a newer glibc, like `WEAK DEFAULT UND statx@GLIBC_2.28`.
This might seem problematic, but old non-weak symbols can tie the build
to new versions too, like `dlsym@GLIBC_2.34` from their recent library
unification. If you build with an old glibc like `dist-x86_64-linux`
does, you'll still get unversioned `WEAK DEFAULT UND statx`, which may
be resolved based on the runtime glibc.

I also found a few functions that don't need to be weak anymore:

- Android can directly use `ftruncate64`, `pread64`, and `pwrite64`, as
  these were added in API 12, and our baseline is API 14.
- Linux can directly use `splice`, added way back in glibc 2.5 and
  similarly old musl. Android only added it in API 21 though.

1 files changed, 81 insertions, 21 deletions

diff --git a/library/std/src/sys/unix/weak.rs b/library/std/src/sys/unix/weak.rs
index ba432ec5494..32072affe8a 100644
--- a/library/std/src/sys/unix/weak.rs
+++ b/library/std/src/sys/unix/weak.rs
@@ -6,7 +6,7 @@
 //! detection.
 //!
 //! One option to use here is weak linkage, but that is unfortunately only
-//! really workable on Linux. Hence, use dlsym to get the symbol value at
+//! really workable with ELF. Otherwise, use dlsym to get the symbol value at
 //! runtime. This is also done for compatibility with older versions of glibc,
 //! and to avoid creating dependencies on GLIBC_PRIVATE symbols. It assumes that
 //! we've been dynamically linked to the library the symbol comes from, but that
@@ -14,7 +14,8 @@
 //!
 //! A long time ago this used weak linkage for the __pthread_get_minstack
 //! symbol, but that caused Debian to detect an unnecessarily strict versioned
-//! dependency on libc6 (#23628).
+//! dependency on libc6 (#23628) because it is GLIBC_PRIVATE. We now use `dlsym`
+//! for a runtime lookup of that symbol to avoid the ELF versioned dependency.
 
 // There are a variety of `#[cfg]`s controlling which targets are involved in
 // each instance of `weak!` and `syscall!`. Rather than trying to unify all of
@@ -22,31 +23,75 @@
 #![allow(dead_code, unused_macros)]
 
 use crate::ffi::CStr;
-use crate::marker;
+use crate::marker::PhantomData;
 use crate::mem;
 use crate::sync::atomic::{self, AtomicUsize, Ordering};
 
+// We can use true weak linkage on ELF targets.
+#[cfg(not(any(target_os = "macos", target_os = "ios")))]
 pub(crate) macro weak {
     (fn $name:ident($($t:ty),*) -> $ret:ty) => (
-        #[allow(non_upper_case_globals)]
-        static $name: crate::sys::weak::Weak<unsafe extern "C" fn($($t),*) -> $ret> =
-            crate::sys::weak::Weak::new(concat!(stringify!($name), '\0'));
+        let ref $name: ExternWeak<unsafe extern "C" fn($($t),*) -> $ret> = {
+            extern "C" {
+                #[linkage = "extern_weak"]
+                static $name: *const libc::c_void;
+            }
+            #[allow(unused_unsafe)]
+            ExternWeak::new(unsafe { $name })
+        };
     )
 }
 
-pub struct Weak<F> {
+// On non-ELF targets, use the dlsym approximation of weak linkage.
+#[cfg(any(target_os = "macos", target_os = "ios"))]
+pub(crate) use self::dlsym as weak;
+
+pub(crate) struct ExternWeak<F> {
+    weak_ptr: *const libc::c_void,
+    _marker: PhantomData<F>,
+}
+
+impl<F> ExternWeak<F> {
+    #[inline]
+    pub(crate) fn new(weak_ptr: *const libc::c_void) -> Self {
+        ExternWeak { weak_ptr, _marker: PhantomData }
+    }
+}
+
+impl<F> ExternWeak<F> {
+    #[inline]
+    pub(crate) fn get(&self) -> Option<F> {
+        unsafe {
+            if self.weak_ptr.is_null() {
+                None
+            } else {
+                Some(mem::transmute_copy::<*const libc::c_void, F>(&self.weak_ptr))
+            }
+        }
+    }
+}
+
+pub(crate) macro dlsym {
+    (fn $name:ident($($t:ty),*) -> $ret:ty) => (
+        static DLSYM: DlsymWeak<unsafe extern "C" fn($($t),*) -> $ret> =
+            DlsymWeak::new(concat!(stringify!($name), '\0'));
+        let $name = &DLSYM;
+    )
+}
+
+pub(crate) struct DlsymWeak<F> {
     name: &'static str,
     addr: AtomicUsize,
-    _marker: marker::PhantomData<F>,
+    _marker: PhantomData<F>,
 }
 
-impl<F> Weak<F> {
-    pub const fn new(name: &'static str) -> Weak<F> {
-        Weak { name, addr: AtomicUsize::new(1), _marker: marker::PhantomData }
+impl<F> DlsymWeak<F> {
+    pub(crate) const fn new(name: &'static str) -> Self {
+        DlsymWeak { name, addr: AtomicUsize::new(1), _marker: PhantomData }
     }
 
-    pub fn get(&self) -> Option<F> {
-        assert_eq!(mem::size_of::<F>(), mem::size_of::<usize>());
+    #[inline]
+    pub(crate) fn get(&self) -> Option<F> {
         unsafe {
             // Relaxed is fine here because we fence before reading through the
             // pointer (see the comment below).
@@ -82,6 +127,8 @@ impl<F> Weak<F> {
     // Cold because it should only happen during first-time initalization.
     #[cold]
     unsafe fn initialize(&self) -> Option<F> {
+        assert_eq!(mem::size_of::<F>(), mem::size_of::<usize>());
+
         let val = fetch(self.name);
         // This synchronizes with the acquire fence in `get`.
         self.addr.store(val, Ordering::Release);
@@ -105,14 +152,12 @@ unsafe fn fetch(name: &str) -> usize {
 pub(crate) macro syscall {
     (fn $name:ident($($arg_name:ident: $t:ty),*) -> $ret:ty) => (
         unsafe fn $name($($arg_name: $t),*) -> $ret {
-            use super::os;
-
             weak! { fn $name($($t),*) -> $ret }
 
             if let Some(fun) = $name.get() {
                 fun($($arg_name),*)
             } else {
-                os::set_errno(libc::ENOSYS);
+                super::os::set_errno(libc::ENOSYS);
                 -1
             }
         }
@@ -123,11 +168,6 @@ pub(crate) macro syscall {
 pub(crate) macro syscall {
     (fn $name:ident($($arg_name:ident: $t:ty),*) -> $ret:ty) => (
         unsafe fn $name($($arg_name:$t),*) -> $ret {
-            use weak;
-            // This looks like a hack, but concat_idents only accepts idents
-            // (not paths).
-            use libc::*;
-
             weak! { fn $name($($t),*) -> $ret }
 
             // Use a weak symbol from libc when possible, allowing `LD_PRELOAD`
@@ -135,6 +175,10 @@ pub(crate) macro syscall {
             if let Some(fun) = $name.get() {
                 fun($($arg_name),*)
             } else {
+                // This looks like a hack, but concat_idents only accepts idents
+                // (not paths).
+                use libc::*;
+
                 syscall(
                     concat_idents!(SYS_, $name),
                     $($arg_name),*
@@ -143,3 +187,19 @@ pub(crate) macro syscall {
         }
     )
 }
+
+#[cfg(any(target_os = "linux", target_os = "android"))]
+pub(crate) macro raw_syscall {
+    (fn $name:ident($($arg_name:ident: $t:ty),*) -> $ret:ty) => (
+        unsafe fn $name($($arg_name:$t),*) -> $ret {
+            // This looks like a hack, but concat_idents only accepts idents
+            // (not paths).
+            use libc::*;
+
+            syscall(
+                concat_idents!(SYS_, $name),
+                $($arg_name),*
+            ) as $ret
+        }
+    )
+}