Add shim for realpath on unix

LegNeato · LegNeato · commit 086c8ee00f0c · 2022-07-02T00:54:59.000-07:00
diff --git a/src/helpers.rs b/src/helpers.rs
@@ -37,6 +37,7 @@ const UNIX_IO_ERROR_TABLE: &[(std::io::ErrorKind, &str)] = {
         (NotFound, "ENOENT"),
         (Interrupted, "EINTR"),
         (InvalidInput, "EINVAL"),
+        (InvalidFilename, "ENAMETOOLONG"),
         (TimedOut, "ETIMEDOUT"),
         (AlreadyExists, "EEXIST"),
         (WouldBlock, "EWOULDBLOCK"),
diff --git a/src/shims/unix/foreign_items.rs b/src/shims/unix/foreign_items.rs
@@ -138,6 +138,11 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                 let result = this.readlink(pathname, buf, bufsize)?;
                 this.write_scalar(Scalar::from_machine_isize(result, this), dest)?;
             }
+            "realpath" => {
+                let [path, resolved_path] = this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+                let result = this.realpath(path, resolved_path)?;
+                this.write_pointer(result, dest)?;
+            }
 
             // Time related shims
             "gettimeofday" => {
diff --git a/src/shims/unix/fs.rs b/src/shims/unix/fs.rs
@@ -1,5 +1,6 @@
 use std::borrow::Cow;
 use std::collections::BTreeMap;
+use std::convert::TryInto;
 use std::fs::{
     read_dir, remove_dir, remove_file, rename, DirBuilder, File, FileType, OpenOptions, ReadDir,
 };
@@ -1659,6 +1660,67 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
             }
         }
     }
+
+    fn realpath(
+        &mut self,
+        path_op: &OpTy<'tcx, Tag>,
+        processed_path_op: &OpTy<'tcx, Tag>,
+    ) -> InterpResult<'tcx, Pointer<Option<Tag>>> {
+        let this = self.eval_context_mut();
+        this.assert_target_os_is_unix("realpath");
+
+        let pathname = this.read_path_from_c_str(this.read_pointer(path_op)?)?;
+        let processed_ptr = this.read_pointer(processed_path_op)?;
+
+        // Reject if isolation is enabled.
+        if let IsolatedOp::Reject(reject_with) = this.machine.isolated_op {
+            this.reject_in_isolation("`realpath`", reject_with)?;
+            let eacc = this.eval_libc("EACCES")?;
+            this.set_last_error(eacc)?;
+            return Ok(Pointer::null());
+        }
+
+        let result = std::fs::canonicalize(pathname);
+        match result {
+            Ok(resolved) => {
+                let path_max = this
+                    .eval_libc_i32("PATH_MAX")?
+                    .try_into()
+                    .expect("PATH_MAX does not fit in u64");
+                let dest = if this.ptr_is_null(processed_ptr)? {
+                    // POSIX says behavior when passing a null pointer is implementation-defined,
+                    // but GNU/linux, freebsd, netbsd, bionic/android, and macos all treat a null pointer
+                    // similarly to:
+                    //
+                    // "If resolved_path is specified as NULL, then realpath() uses
+                    // malloc(3) to allocate a buffer of up to PATH_MAX bytes to hold
+                    // the resolved pathname, and returns a pointer to this buffer.  The
+                    // caller should deallocate this buffer using free(3)."
+                    // <https://man7.org/linux/man-pages/man3/realpath.3.html>
+                    this.alloc_os_str_as_c_str(resolved.as_os_str(), MiriMemoryKind::C.into())?
+                } else {
+                    let (wrote_path, _) =
+                        this.write_path_to_c_str(&resolved, processed_ptr, path_max)?;
+
+                    if !wrote_path {
+                        // Note that we do not explicitly handle `FILENAME_MAX`
+                        // (different from `PATH_MAX` above) as it is Linux-specific and
+                        // seems like a bit of a mess anyway: <https://eklitzke.org/path-max-is-tricky>.
+                        let enametoolong = this.eval_libc("ENAMETOOLONG")?;
+                        this.set_last_error(enametoolong)?;
+                        return Ok(Pointer::null());
+                    }
+                    processed_ptr
+                };
+
+                Ok(dest)
+            }
+            Err(e) => {
+                this.set_last_error_from_io_error(e.kind())?;
+                Ok(Pointer::null())
+            }
+        }
+    }
 }
 
 /// Extracts the number of seconds and nanoseconds elapsed between `time` and the unix epoch when
diff --git a/src/shims/unix/macos/foreign_items.rs b/src/shims/unix/macos/foreign_items.rs
@@ -67,6 +67,12 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                 let result = this.ftruncate64(fd, length)?;
                 this.write_scalar(Scalar::from_i32(result), dest)?;
             }
+            "realpath" | "realpath$DARWIN_EXTSN" => {
+                let [path, resolved_path] =
+                    this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+                let result = this.realpath(path, resolved_path)?;
+                this.write_pointer(result, dest)?;
+            }
 
             // Environment related shims
             "_NSGetEnviron" => {
diff --git a/tests/pass/fs.rs b/tests/pass/fs.rs
@@ -25,6 +25,7 @@ fn main() {
     test_errors();
     test_rename();
     test_directory();
+    test_canonicalize();
     test_dup_stdout_stderr();
 
     // These all require unix, if the test is changed to no longer `ignore-windows`, move these to a unix test
@@ -364,6 +365,24 @@ fn test_rename() {
     remove_file(&path2).unwrap();
 }
 
+fn test_canonicalize() {
+    use std::fs::canonicalize;
+    let dir_path = prepare_dir("miri_test_fs_dir");
+    create_dir(&dir_path).unwrap();
+    let path = dir_path.join("test_file");
+    drop(File::create(&path).unwrap());
+
+    let p = canonicalize(format!("{}/./test_file", dir_path.to_string_lossy())).unwrap();
+    assert_eq!(p.to_string_lossy().find('.'), None);
+
+    remove_dir_all(&dir_path).unwrap();
+
+    // Make sure we get an error for long paths.
+    use std::convert::TryInto;
+    let too_long = "x/".repeat(libc::PATH_MAX.try_into().unwrap());
+    assert!(canonicalize(too_long).is_err());
+}
+
 fn test_directory() {
     let dir_path = prepare_dir("miri_test_fs_dir");
     // Creating a directory should succeed.
diff --git a/tests/pass/libc.rs b/tests/pass/libc.rs
@@ -1,17 +1,142 @@
 // ignore-windows: No libc on Windows
 // compile-flags: -Zmiri-disable-isolation
 
+#![feature(io_error_more)]
 #![feature(rustc_private)]
 
 extern crate libc;
 
-#[cfg(any(target_os = "linux", target_os = "freebsd"))]
 fn tmp() -> std::path::PathBuf {
     std::env::var("MIRI_TEMP")
         .map(std::path::PathBuf::from)
         .unwrap_or_else(|_| std::env::temp_dir())
 }
 
+/// Test allocating variant of `realpath`.
+fn test_posix_realpath_alloc() {
+    use std::ffi::OsString;
+    use std::ffi::{CStr, CString};
+    use std::fs::{remove_file, File};
+    use std::os::unix::ffi::OsStrExt;
+    use std::os::unix::ffi::OsStringExt;
+    use std::path::PathBuf;
+
+    let buf;
+    let path = tmp().join("miri_test_libc_posix_realpath_alloc");
+    let c_path = CString::new(path.as_os_str().as_bytes()).expect("CString::new failed");
+
+    // Cleanup before test.
+    remove_file(&path).ok();
+    // Create file.
+    drop(File::create(&path).unwrap());
+    unsafe {
+        let r = libc::realpath(c_path.as_ptr(), std::ptr::null_mut());
+        assert!(!r.is_null());
+        buf = CStr::from_ptr(r).to_bytes().to_vec();
+        libc::free(r as *mut _);
+    }
+    let canonical = PathBuf::from(OsString::from_vec(buf));
+    // Cleanup after test.
+    remove_file(&path).unwrap();
+
+    assert_eq!(path.file_name(), canonical.file_name());
+}
+
+/// Test non-allocating variant of `realpath`.
+fn test_posix_realpath_noalloc() {
+    use std::ffi::{CStr, CString};
+    use std::fs::{remove_file, File};
+    use std::os::unix::ffi::OsStrExt;
+    use std::path::PathBuf;
+
+    let path = tmp().join("miri_test_libc_posix_realpath_noalloc");
+    let c_path = CString::new(path.as_os_str().as_bytes()).expect("CString::new failed");
+
+    let mut v = vec![0; libc::PATH_MAX as usize];
+
+    // Cleanup before test.
+    remove_file(&path).ok();
+    // Create file.
+    drop(File::create(&path).unwrap());
+    unsafe {
+        let r = libc::realpath(c_path.as_ptr(), v.as_mut_ptr());
+        assert!(!r.is_null());
+    }
+    let c = unsafe { CStr::from_ptr(v.as_ptr()) };
+    let canonical = PathBuf::from(c.to_str().expect("CStr to str"));
+    // Cleanup after test.
+    remove_file(&path).unwrap();
+
+    assert_eq!(path.file_name(), canonical.file_name());
+}
+
+/// Test failure cases for `realpath`.
+fn test_posix_realpath_errors() {
+    use std::convert::TryInto;
+    use std::ffi::CString;
+    use std::fs::{create_dir_all, remove_dir, remove_file};
+    use std::io::ErrorKind;
+    use std::os::unix::ffi::OsStrExt;
+    use std::os::unix::fs::symlink;
+
+    // Test non-existent path returns an error.
+    let c_path = CString::new("./nothing_to_see_here").expect("CString::new failed");
+    let r = unsafe { libc::realpath(c_path.as_ptr(), std::ptr::null_mut()) };
+    assert!(r.is_null());
+    let e = std::io::Error::last_os_error();
+    assert_eq!(e.raw_os_error(), Some(libc::ENOENT));
+    assert_eq!(e.kind(), ErrorKind::NotFound);
+
+    // Test that a long path returns an error.
+    //
+    // Linux first checks if the path to exists and macos does not.
+    // Using an existing path ensures all platforms return `ENAMETOOLONG` given a long path.
+    //
+    // Rather than creating a bunch of directories, we create two directories containing symlinks.
+    // Sadly we can't avoid creating directories via a path like "./././././" or"/../../" as linux
+    // appears to collapse "." and ".." before checking path length.
+    let path = tmp();
+
+    // The directories we will put symlinks in.
+    let x = path.join("x/");
+    let y = path.join("y/");
+
+    // The symlinks in each directory pointing to each other.
+    let yx_sym = y.join("x");
+    let xy_sym = x.join("y");
+
+    // Cleanup before test.
+    remove_file(&yx_sym).ok();
+    remove_file(&xy_sym).ok();
+    remove_dir(&x).ok();
+    remove_dir(&y).ok();
+
+    // Create directories.
+    create_dir_all(&x).expect("dir x");
+    create_dir_all(&y).expect("dir y");
+
+    // Create symlinks between directories.
+    symlink(&x, &yx_sym).expect("symlink x");
+    symlink(&y, &xy_sym).expect("symlink y ");
+
+    // This pass exists due to the symlinks created above but is too long.
+    let too_long = path.join("x/y/".repeat(libc::PATH_MAX.try_into().unwrap()));
+
+    let c_path = CString::new(too_long.into_os_string().as_bytes()).expect("CString::new failed");
+    let r = unsafe { libc::realpath(c_path.as_ptr(), std::ptr::null_mut()) };
+    let e = std::io::Error::last_os_error();
+
+    // Cleanup after test.
+    remove_file(&yx_sym).ok();
+    remove_file(&xy_sym).ok();
+    remove_dir(&x).ok();
+    remove_dir(&y).ok();
+
+    assert!(r.is_null());
+    assert_eq!(e.raw_os_error(), Some(libc::ENAMETOOLONG));
+    assert_eq!(e.kind(), ErrorKind::InvalidFilename);
+}
+
 #[cfg(any(target_os = "linux", target_os = "freebsd"))]
 fn test_posix_fadvise() {
     use std::convert::TryInto;
@@ -317,6 +442,10 @@ fn main() {
 
     test_posix_gettimeofday();
 
+    test_posix_realpath_alloc();
+    test_posix_realpath_noalloc();
+    test_posix_realpath_errors();
+
     #[cfg(any(target_os = "linux", target_os = "freebsd"))]
     test_sync_file_range();
 
