various: (ab)use the new syscall wrappers and symbols; drop libc entirely

Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: I19ecd9801cf6e04adcedd3003d9fc59d6a6a6964

.cargo/config.toml

@@ -1,3 +1,5 @@
-# https://github.com/rui314/mold?tab=readme-ov-file#how-to-use
+# Link with Mold, and without libc! We use nostartfiles to avoid the C runtime
+# See:
+#  <https://github.com/rui314/mold?tab=readme-ov-file#how-to-use>
 [target.'cfg(target_os = "linux")']
-rustflags = [ "-C", "link-arg=-fuse-ld=mold", "-C", "link-arg=-lc", "-C", "link-arg=-lgcc_s" ]
+rustflags = [ "-C", "link-arg=-fuse-ld=mold", "-C", "link-arg=-nostartfiles" ]

crates/lib/src/lib.rs

@@ -142,22 +142,34 @@ impl<T> Drop for OnceLock<T> {
   }
 }
 
-// Access to the environ pointer (provided by libc startup code)
-unsafe extern "C" {
-  static environ: *const *const u8;
-}
+// Store the environment pointer internally,initialized from `main()`. This
+// helps avoid the libc dependency *completely*.
+static ENVP: AtomicPtr<*const u8> = AtomicPtr::new(core::ptr::null_mut());
 
-/// Gets an environment variable by name (without using std).
+/// Initialize the environment pointer. Must be called before any `getenv()`
+/// calls. This is called from `main()` with the calculated `envp`.
 ///
 /// # Safety
 ///
-/// This function reads from the environ global which is initialized
-/// by the C runtime before `main()` is called.
+/// envp must be a valid null-terminated array of C strings, or null if
+/// no environment is available.
+#[inline]
+pub unsafe fn init_env(envp: *const *const u8) {
+  ENVP.store(envp.cast_mut(), Ordering::Release);
+}
+
+/// Gets the current environment pointer.
+#[inline]
+#[must_use]
+fn get_envp() -> *const *const u8 {
+  ENVP.load(Ordering::Acquire)
+}
+
+/// Gets an environment variable by name without using std or libc by reading
+/// from the environment pointer set by [`init_env`].
 #[must_use]
 pub fn getenv(name: &str) -> Option<&'static [u8]> {
-  // SAFETY: environ is set up by the C runtime before main() runs
-  // and remains valid for the lifetime of the program
-  let envp = unsafe { environ };
+  let envp = get_envp();
   if envp.is_null() {
     return None;
   }
@@ -237,6 +249,7 @@ impl UtsName {
     if unsafe { sys_uname(uts.as_mut_ptr()) } != 0 {
       return Err(Error::last_os_error());
     }
+
     Ok(Self(unsafe { uts.assume_init() }))
   }
 

microfetch/src/main.rs

@@ -3,40 +3,49 @@
 
 extern crate alloc;
 
-use microfetch_alloc::BumpAlloc;
-use microfetch_asm::sys_write;
-#[cfg(not(test))]
-use {core::panic::PanicInfo, microfetch_asm::sys_exit};
+use core::panic::PanicInfo;
 
+use microfetch_alloc::BumpAllocator;
+// Re-export libc replacement functions from asm crate
+pub use microfetch_asm::{memcpy, memset, strlen};
+use microfetch_asm::{sys_exit, sys_write};
+
+// Global allocator
 #[global_allocator]
-static ALLOCATOR: BumpAlloc = BumpAlloc::new();
+static ALLOCATOR: BumpAllocator = BumpAllocator::new();
 
-/// Receives argc and argv directly. The C runtime will call this after
-/// initializing the environment. Cool right?
+/// Main application entry point. Called by the asm crate's entry point
+/// after setting up argc, argv, and envp.
 ///
 /// # Safety
 ///
 /// argv must be a valid pointer to an array of argc C strings.
 #[unsafe(no_mangle)]
 pub unsafe extern "C" fn main(argc: i32, argv: *const *const u8) -> i32 {
-  // SAFETY: argc and argv are provided by the C runtime and are valid
+  // Calculate envp from argv. On Linux, envp is right after argv on the stack
+  // but I bet 12 cents that there will be at least one exception.
+  let argc_usize = usize::try_from(argc).unwrap_or(0);
+  let envp = unsafe { argv.add(argc_usize + 1) };
+
+  // Initialize the environment pointer
   unsafe {
-    match microfetch_lib::run(argc, argv) {
-      Ok(()) => 0,
-      Err(e) => {
-        // Print error message to stderr (fd 2)
-        let msg = alloc::format!("Error: {e}\n");
-        let _ = sys_write(2, msg.as_ptr(), msg.len());
-        1
-      },
-    }
+    microfetch_lib::init_env(envp);
+  }
+
+  // Run the main application logic
+  match unsafe { microfetch_lib::run(argc, argv) } {
+    Ok(()) => 0,
+    Err(e) => {
+      let msg = alloc::format!("Error: {e}\n");
+      let _ = unsafe { sys_write(2, msg.as_ptr(), msg.len()) };
+      1
+    },
   }
 }
 
 #[cfg(not(test))]
 #[panic_handler]
 fn panic(_info: &PanicInfo) -> ! {
-  // Write "panic" to stderr and exit
   const PANIC_MSG: &[u8] = b"panic\n";
   unsafe {
     let _ = sys_write(2, PANIC_MSG.as_ptr(), PANIC_MSG.len());
@@ -44,11 +53,7 @@ fn panic(_info: &PanicInfo) -> ! {
   }
 }
 
-// FIXME: Stubs for Rust exception handling symbols needed when using alloc with
-// panic=abort These are normally provided by the unwinding runtime, but we're
-// using panic=abort. I don't actually think this is the correct approach, but I
-// cannot think of anything better.
-
+// Stubs for Rust exception handling
 #[cfg(not(test))]
 #[unsafe(no_mangle)]
 const extern "C" fn rust_eh_personality() {}