initial work on adding 'read/write/alloc wide_str' functions

Author: JOE1994

src/helpers.rs

@@ -1,4 +1,4 @@
-use std::ffi::OsStr;
+use std::ffi::{OsStr, OsString};
 use std::{iter, mem};
 
 use rustc::mir;
@@ -447,29 +447,77 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         }
     }
 
+    fn read_os_str_from_target_str<'a>(&'a self, scalar: Scalar<Tag>) -> InterpResult<'tcx, OsString> {
+        if cfg!(target_os = "unix") {
+            self.read_os_str_from_c_str(scalar)
+        } else if cfg!(target_os = "windows") {
+            self.read_os_str_from_wide_str(scalar)
+        } else {
+            throw_unsup_format!("support for target OS not yet available")
+        }
+    }
+
     /// Helper function to read an OsString from a null-terminated sequence of bytes, which is what
     /// the Unix APIs usually handle.
-    fn read_os_str_from_c_str<'a>(&'a self, scalar: Scalar<Tag>) -> InterpResult<'tcx, &'a OsStr>
+    fn read_os_str_from_c_str<'a>(&'a self, scalar: Scalar<Tag>) -> InterpResult<'tcx, OsString>
     where
         'tcx: 'a,
         'mir: 'a,
     {
         #[cfg(target_os = "unix")]
-        fn bytes_to_os_str<'tcx, 'a>(bytes: &'a [u8]) -> InterpResult<'tcx, &'a OsStr> {
-            Ok(std::os::unix::ffi::OsStringExt::from_bytes(bytes))
+        fn bytes_to_os_str<'tcx>(bytes: &[u8]) -> InterpResult<'tcx, OsString> {
+            Ok(std::os::unix::ffi::OsStringExt::from_bytes(bytes).to_os_string())
         }
         #[cfg(not(target_os = "unix"))]
-        fn bytes_to_os_str<'tcx, 'a>(bytes: &'a [u8]) -> InterpResult<'tcx, &'a OsStr> {
+        fn bytes_to_os_str<'tcx>(bytes: &[u8]) -> InterpResult<'tcx, OsString> {
             let s = std::str::from_utf8(bytes)
                 .map_err(|_| err_unsup_format!("{:?} is not a valid utf-8 string", bytes))?;
-            Ok(&OsStr::new(s))
+            Ok(OsStr::new(s).to_os_string())
         }
 
         let this = self.eval_context_ref();
         let bytes = this.memory.read_c_str(scalar)?;
         bytes_to_os_str(bytes)
     }
 
+    /// Helper function to read an OsString from a 0x0000-terminated sequence of u16,
+    /// which is what the Windows APIs usually handle.
+    fn read_os_str_from_wide_str<'a>(&'a self, scalar: Scalar<Tag>) -> InterpResult<'tcx, OsString>
+    where
+        'tcx: 'a,
+        'mir: 'a,
+    {
+        #[cfg(target_os = "windows")]
+        fn u16vec_to_osstring<'tcx>(u16_vec: Vec<u16>) -> InterpResult<'tcx, OsString> {
+            Ok(std::os::windows::ffi::OsStringExt::from_wide(&u16_vec[..]))
+        }
+        #[cfg(not(target_os = "windows"))]
+        fn u16vec_to_osstring<'tcx>(u16_vec: Vec<u16>) -> InterpResult<'tcx, OsString> {
+            let s = String::from_utf16(&u16_vec[..])
+                .map_err(|_| err_unsup_format!("{:?} is not a valid utf-16 string", u16_vec))?;
+            Ok(s.into())
+        }
+
+        let this = self.eval_context_ref();
+        let u16_vec = this.memory.read_wide_str(scalar)?;
+        u16vec_to_osstring(u16_vec)
+    }
+
+    fn write_os_str_to_target_str(
+        &mut self,
+        os_str: &OsStr,
+        scalar: Scalar<Tag>,
+        size: u64,
+    ) -> InterpResult<'tcx, (bool, u64)> {
+        if cfg!(target_os = "unix") {
+            self.write_os_str_to_c_str(os_str, scalar, size)
+        } else if cfg!(target_os = "windows") {
+            self.write_os_str_to_wide_str(os_str, scalar, size)
+        } else {
+            panic!("support for target OS not yet available")
+        }
+    }
+
     /// Helper function to write an OsStr as a null-terminated sequence of bytes, which is what
     /// the Unix APIs usually handle. This function returns `Ok((false, length))` without trying
     /// to write if `size` is not large enough to fit the contents of `os_string` plus a null
@@ -509,6 +557,62 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         Ok((true, string_length))
     }
 
+    /// Helper function to write an OsStr as a 0x0000-terminated u16-sequence, which is what
+    /// the Windows APIs usually handle. This function returns `Ok((false, length))` without trying
+    /// to write if `size` is not large enough to fit the contents of `os_string` plus a null
+    /// terminator. It returns `Ok((true, length))` if the writing process was successful. The
+    /// string length returned does not include the null terminator.
+    fn write_os_str_to_wide_str(
+        &mut self,
+        os_str: &OsStr,
+        scalar: Scalar<Tag>,
+        size: u64,
+    ) -> InterpResult<'tcx, (bool, u64)> {
+        #[cfg(target_os = "windows")]
+        fn os_str_to_u16vec(os_str: &OsStr) -> Vec<u16> {
+            std::os::windows::ffi::OsStrExt::encode_wide(os_str).collect()
+        }
+        #[cfg(not(target_os = "windows"))]
+        fn os_str_to_u16vec(os_str: &OsStr) -> Vec<u16> {
+            os_str.to_str().encode_utf16().collect()
+        }
+
+        let u16_vec = os_str_to_u16vec(os_str);
+        // If `size` is smaller or equal than `bytes.len()`, writing `bytes` plus the required null
+        // terminator to memory using the `ptr` pointer would cause an out-of-bounds access.
+        let string_length = u16_vec.len() as u64;
+        if size <= string_length {
+            return Ok((false, string_length));
+        }
+
+        let this = self.eval_context_mut();
+
+        // Store the UTF-16 string. We just allocated so we know the bounds are fine.
+        let char_size = Size::from_bytes(2);
+        let place_ptr = scalar.assert_ptr();
+        /*
+        for (idx, &c) in u16_vec.iter().enumerate() {
+            let place = this.mplace_field(place, idx as u64)?; 
+            this.write_scalar(Scalar::from_uint(c, char_size), place.into())?;
+        }
+        */
+        Ok((true, string_length))
+    }
+
+    fn alloc_os_str_as_target_str(
+        &mut self,
+        os_str: &OsStr,
+        memkind: MemoryKind<MiriMemoryKind>,
+    ) -> Pointer<Tag> {
+        if cfg!(target_os = "unix") {
+            self.alloc_os_str_as_c_str(os_str, memkind)
+        } else if cfg!(target_os = "windows") {
+            self.alloc_os_str_as_wide_str(os_str, memkind)
+        } else {
+            panic!("support for target OS not yet available")
+        }
+    }
+
     fn alloc_os_str_as_c_str(
         &mut self,
         os_str: &OsStr,
@@ -522,6 +626,20 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         self.write_os_str_to_c_str(os_str, arg_place.ptr, size).unwrap();
         arg_place.ptr.assert_ptr()
     }
+
+    fn alloc_os_str_as_wide_str(
+        &mut self,
+        os_str: &OsStr,
+        memkind: MemoryKind<MiriMemoryKind>,
+    ) -> Pointer<Tag> {
+        let size = os_str.len() as u64 + 1; // Make space for `0x0000` terminator.
+        let this = self.eval_context_mut();
+
+        let arg_type = this.tcx.mk_array(this.tcx.types.u16, size);
+        let arg_place = this.allocate(this.layout_of(arg_type).unwrap(), memkind);
+        self.write_os_str_to_wide_str(os_str, arg_place.ptr, size).unwrap();
+        arg_place.ptr.assert_ptr()
+    }
 }
 
 pub fn immty_from_int_checked<'tcx>(

src/shims/env.rs

@@ -28,7 +28,7 @@ impl<'tcx> EnvVars<'tcx> {
             for (name, value) in env::vars() {
                 if !excluded_env_vars.contains(&name) {
                     let var_ptr =
-                        alloc_env_var_as_c_str(name.as_ref(), value.as_ref(), ecx);
+                        alloc_env_var_as_target_str(name.as_ref(), value.as_ref(), ecx);
                     ecx.machine.env_vars.map.insert(OsString::from(name), var_ptr);
                 }
             }
@@ -37,15 +37,15 @@ impl<'tcx> EnvVars<'tcx> {
     }
 }
 
-fn alloc_env_var_as_c_str<'mir, 'tcx>(
+fn alloc_env_var_as_target_str<'mir, 'tcx>(
     name: &OsStr,
     value: &OsStr,
     ecx: &mut InterpCx<'mir, 'tcx, Evaluator<'tcx>>,
 ) -> Pointer<Tag> {
     let mut name_osstring = name.to_os_string();
     name_osstring.push("=");
     name_osstring.push(value);
-    ecx.alloc_os_str_as_c_str(name_osstring.as_os_str(), MiriMemoryKind::Machine.into())
+    ecx.alloc_os_str_as_target_str(name_osstring.as_os_str(), MiriMemoryKind::Machine.into())
 }
 
 impl<'mir, 'tcx> EvalContextExt<'mir, 'tcx> for crate::MiriEvalContext<'mir, 'tcx> {}
@@ -54,8 +54,8 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         let this = self.eval_context_mut();
 
         let name_ptr = this.read_scalar(name_op)?.not_undef()?;
-        let name = this.read_os_str_from_c_str(name_ptr)?;
-        Ok(match this.machine.env_vars.map.get(name) {
+        let name = this.read_os_str_from_target_str(name_ptr)?;
+        Ok(match this.machine.env_vars.map.get(&name) {
             // The offset is used to strip the "{name}=" part of the string.
             Some(var_ptr) => {
                 Scalar::from(var_ptr.offset(Size::from_bytes(name.len() as u64 + 1), this)?)
@@ -64,6 +64,11 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         })
     }
 
+
+    fn getenvironmentvariablew() {
+
+    }
+
     fn setenv(
         &mut self,
         name_op: OpTy<'tcx, Tag>,
@@ -73,16 +78,16 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
 
         let name_ptr = this.read_scalar(name_op)?.not_undef()?;
         let value_ptr = this.read_scalar(value_op)?.not_undef()?;
-        let value = this.read_os_str_from_c_str(value_ptr)?;
+        let value = this.read_os_str_from_target_str(value_ptr)?;
         let mut new = None;
         if !this.is_null(name_ptr)? {
-            let name = this.read_os_str_from_c_str(name_ptr)?;
+            let name = this.read_os_str_from_target_str(name_ptr)?;
             if !name.is_empty() && !name.to_string_lossy().contains('=') {
                 new = Some((name.to_owned(), value.to_owned()));
             }
         }
         if let Some((name, value)) = new {
-            let var_ptr = alloc_env_var_as_c_str(&name, &value, &mut this);
+            let var_ptr = alloc_env_var_as_target_str(&name, &value, &mut this);
             if let Some(var) = this.machine.env_vars.map.insert(name.to_owned(), var_ptr) {
                 this.memory
                     .deallocate(var, None, MiriMemoryKind::Machine.into())?;
@@ -94,13 +99,17 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         }
     }
 
+    fn setenvironmentvariablew() {
+
+    }
+
     fn unsetenv(&mut self, name_op: OpTy<'tcx, Tag>) -> InterpResult<'tcx, i32> {
         let this = self.eval_context_mut();
 
         let name_ptr = this.read_scalar(name_op)?.not_undef()?;
         let mut success = None;
         if !this.is_null(name_ptr)? {
-            let name = this.read_os_str_from_c_str(name_ptr)?.to_owned();
+            let name = this.read_os_str_from_target_str(name_ptr)?.to_owned();
             if !name.is_empty() && !name.to_string_lossy().contains('=') {
                 success = Some(this.machine.env_vars.map.remove(&name));
             }