1 //! Main evaluator loop and setting up the initial stack frame.
3 use std::ffi::{OsStr, OsString};
5 use std::panic::{self, AssertUnwindSafe};
6 use std::path::PathBuf;
11 use rustc_data_structures::fx::FxHashSet;
12 use rustc_hir::def_id::DefId;
13 use rustc_middle::ty::{
15 layout::{LayoutCx, LayoutOf},
18 use rustc_target::spec::abi::Abi;
20 use rustc_session::config::EntryFnType;
24 #[derive(Copy, Clone, Debug, PartialEq)]
25 pub enum AlignmentCheck {
26 /// Do not check alignment.
28 /// Check alignment "symbolically", i.e., using only the requested alignment for an allocation and not its real base address.
30 /// Check alignment on the actual physical integer address.
34 #[derive(Copy, Clone, Debug, PartialEq)]
35 pub enum RejectOpWith {
36 /// Isolated op is rejected with an abort of the machine.
39 /// If not Abort, miri returns an error for an isolated op.
40 /// Following options determine if user should be warned about such error.
41 /// Do not print warning about rejected isolated op.
44 /// Print a warning about rejected isolated op, with backtrace.
47 /// Print a warning about rejected isolated op, without backtrace.
48 WarningWithoutBacktrace,
51 #[derive(Copy, Clone, Debug, PartialEq)]
53 /// Reject an op requiring communication with the host. By
54 /// default, miri rejects the op with an abort. If not, it returns
55 /// an error code, and prints a warning about it. Warning levels
56 /// are controlled by `RejectOpWith` enum.
59 /// Execute op requiring communication with the host, i.e. disable isolation.
63 #[derive(Copy, Clone, PartialEq, Eq)]
64 pub enum BacktraceStyle {
65 /// Prints a terser backtrace which ideally only contains relevant information.
67 /// Prints a backtrace with all possible information.
69 /// Prints only the frame that the error occurs in.
73 /// Configuration needed to spawn a Miri instance.
75 pub struct MiriConfig {
76 /// The host environment snapshot to use as basis for what is provided to the interpreted program.
77 /// (This is still subject to isolation as well as `forwarded_env_vars`.)
78 pub env: Vec<(OsString, OsString)>,
79 /// Determine if validity checking is enabled.
81 /// Determines if Stacked Borrows is enabled.
82 pub stacked_borrows: bool,
83 /// Controls alignment checking.
84 pub check_alignment: AlignmentCheck,
85 /// Controls function [ABI](Abi) checking.
87 /// Action for an op requiring communication with the host.
88 pub isolated_op: IsolatedOp,
89 /// Determines if memory leaks should be ignored.
90 pub ignore_leaks: bool,
91 /// Environment variables that should always be forwarded from the host.
92 pub forwarded_env_vars: Vec<String>,
93 /// Command-line arguments passed to the interpreted program.
94 pub args: Vec<String>,
95 /// The seed to use when non-determinism or randomness are required (e.g. ptr-to-int cast, `getrandom()`).
96 pub seed: Option<u64>,
97 /// The stacked borrows pointer ids to report about
98 pub tracked_pointer_tags: FxHashSet<SbTag>,
99 /// The stacked borrows call IDs to report about
100 pub tracked_call_ids: FxHashSet<CallId>,
101 /// The allocation ids to report about.
102 pub tracked_alloc_ids: FxHashSet<AllocId>,
103 /// Determine if data race detection should be enabled
104 pub data_race_detector: bool,
105 /// Determine if weak memory emulation should be enabled. Requires data race detection to be enabled
106 pub weak_memory_emulation: bool,
107 /// Track when an outdated (weak memory) load happens.
108 pub track_outdated_loads: bool,
109 /// Rate of spurious failures for compare_exchange_weak atomic operations,
110 /// between 0.0 and 1.0, defaulting to 0.8 (80% chance of failure).
111 pub cmpxchg_weak_failure_rate: f64,
112 /// If `Some`, enable the `measureme` profiler, writing results to a file
113 /// with the specified prefix.
114 pub measureme_out: Option<String>,
115 /// Panic when unsupported functionality is encountered.
116 pub panic_on_unsupported: bool,
117 /// Which style to use for printing backtraces.
118 pub backtrace_style: BacktraceStyle,
119 /// Which provenance to use for int2ptr casts
120 pub provenance_mode: ProvenanceMode,
121 /// Whether to ignore any output by the program. This is helpful when debugging miri
122 /// as its messages don't get intermingled with the program messages.
123 pub mute_stdout_stderr: bool,
124 /// The probability of the active thread being preempted at the end of each basic block.
125 pub preemption_rate: f64,
126 /// Report the current instruction being executed every N basic blocks.
127 pub report_progress: Option<u32>,
128 /// Whether Stacked Borrows retagging should recurse into fields of datatypes.
129 pub retag_fields: bool,
130 /// The location of a shared object file to load when calling external functions
131 /// FIXME! consider allowing users to specify paths to multiple SO files, or to a directory
132 pub external_so_file: Option<PathBuf>,
133 /// Run a garbage collector for SbTags every N basic blocks.
134 pub gc_interval: u32,
137 impl Default for MiriConfig {
138 fn default() -> MiriConfig {
142 stacked_borrows: true,
143 check_alignment: AlignmentCheck::Int,
145 isolated_op: IsolatedOp::Reject(RejectOpWith::Abort),
147 forwarded_env_vars: vec![],
150 tracked_pointer_tags: FxHashSet::default(),
151 tracked_call_ids: FxHashSet::default(),
152 tracked_alloc_ids: FxHashSet::default(),
153 data_race_detector: true,
154 weak_memory_emulation: true,
155 track_outdated_loads: false,
156 cmpxchg_weak_failure_rate: 0.8, // 80%
158 panic_on_unsupported: false,
159 backtrace_style: BacktraceStyle::Short,
160 provenance_mode: ProvenanceMode::Default,
161 mute_stdout_stderr: false,
162 preemption_rate: 0.01, // 1%
163 report_progress: None,
165 external_so_file: None,
171 /// Returns a freshly created `InterpCx`, along with an `MPlaceTy` representing
172 /// the location where the return value of the `start` function will be
174 /// Public because this is also used by `priroda`.
175 pub fn create_ecx<'mir, 'tcx: 'mir>(
178 entry_type: EntryFnType,
180 ) -> InterpResult<'tcx, (InterpCx<'mir, 'tcx, MiriMachine<'mir, 'tcx>>, MPlaceTy<'tcx, Provenance>)>
182 let param_env = ty::ParamEnv::reveal_all();
183 let layout_cx = LayoutCx { tcx, param_env };
184 let mut ecx = InterpCx::new(
186 rustc_span::source_map::DUMMY_SP,
188 MiriMachine::new(config, layout_cx),
191 // Some parts of initialization require a full `InterpCx`.
192 MiriMachine::late_init(&mut ecx, config)?;
194 // Make sure we have MIR. We check MIR for some stable monomorphic function in libcore.
195 let sentinel = ecx.try_resolve_path(&["core", "ascii", "escape_default"]);
196 if !matches!(sentinel, Some(s) if tcx.is_mir_available(s.def.def_id())) {
198 "the current sysroot was built without `-Zalways-encode-mir`, or libcore seems missing. \
199 Use `cargo miri setup` to prepare a sysroot that is suitable for Miri."
203 // Setup first stack frame.
204 let entry_instance = ty::Instance::mono(tcx, entry_id);
206 // First argument is constructed later, because it's skipped if the entry function uses #[start].
208 // Second argument (argc): length of `config.args`.
209 let argc = Scalar::from_machine_usize(u64::try_from(config.args.len()).unwrap(), &ecx);
210 // Third argument (`argv`): created from `config.args`.
212 // Put each argument in memory, collect pointers.
213 let mut argvs = Vec::<Immediate<Provenance>>::new();
214 for arg in config.args.iter() {
215 // Make space for `0` terminator.
216 let size = u64::try_from(arg.len()).unwrap().checked_add(1).unwrap();
217 let arg_type = tcx.mk_array(tcx.types.u8, size);
219 ecx.allocate(ecx.layout_of(arg_type)?, MiriMemoryKind::Machine.into())?;
220 ecx.write_os_str_to_c_str(OsStr::new(arg), arg_place.ptr, size)?;
221 ecx.mark_immutable(&arg_place);
222 argvs.push(arg_place.to_ref(&ecx));
224 // Make an array with all these pointers, in the Miri memory.
225 let argvs_layout = ecx.layout_of(
226 tcx.mk_array(tcx.mk_imm_ptr(tcx.types.u8), u64::try_from(argvs.len()).unwrap()),
228 let argvs_place = ecx.allocate(argvs_layout, MiriMemoryKind::Machine.into())?;
229 for (idx, arg) in argvs.into_iter().enumerate() {
230 let place = ecx.mplace_field(&argvs_place, idx)?;
231 ecx.write_immediate(arg, &place.into())?;
233 ecx.mark_immutable(&argvs_place);
234 // A pointer to that place is the 3rd argument for main.
235 let argv = argvs_place.to_ref(&ecx);
236 // Store `argc` and `argv` for macOS `_NSGetArg{c,v}`.
239 ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
240 ecx.write_scalar(argc, &argc_place.into())?;
241 ecx.mark_immutable(&argc_place);
242 ecx.machine.argc = Some(*argc_place);
244 let argv_place = ecx.allocate(
245 ecx.layout_of(tcx.mk_imm_ptr(tcx.types.unit))?,
246 MiriMemoryKind::Machine.into(),
248 ecx.write_immediate(argv, &argv_place.into())?;
249 ecx.mark_immutable(&argv_place);
250 ecx.machine.argv = Some(*argv_place);
252 // Store command line as UTF-16 for Windows `GetCommandLineW`.
254 // Construct a command string with all the arguments.
255 let cmd_utf16: Vec<u16> = args_to_utf16_command_string(config.args.iter());
257 let cmd_type = tcx.mk_array(tcx.types.u16, u64::try_from(cmd_utf16.len()).unwrap());
259 ecx.allocate(ecx.layout_of(cmd_type)?, MiriMemoryKind::Machine.into())?;
260 ecx.machine.cmd_line = Some(*cmd_place);
261 // Store the UTF-16 string. We just allocated so we know the bounds are fine.
262 for (idx, &c) in cmd_utf16.iter().enumerate() {
263 let place = ecx.mplace_field(&cmd_place, idx)?;
264 ecx.write_scalar(Scalar::from_u16(c), &place.into())?;
266 ecx.mark_immutable(&cmd_place);
271 // Return place (in static memory so that it does not count as leak).
272 let ret_place = ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
273 // Call start function.
276 EntryFnType::Main { .. } => {
277 let start_id = tcx.lang_items().start_fn().unwrap();
278 let main_ret_ty = tcx.fn_sig(entry_id).output();
279 let main_ret_ty = main_ret_ty.no_bound_vars().unwrap();
280 let start_instance = ty::Instance::resolve(
282 ty::ParamEnv::reveal_all(),
284 tcx.mk_substs(::std::iter::once(ty::subst::GenericArg::from(main_ret_ty))),
289 let main_ptr = ecx.create_fn_alloc_ptr(FnVal::Instance(entry_instance));
291 // Inlining of `DEFAULT` from
292 // https://github.com/rust-lang/rust/blob/master/compiler/rustc_session/src/config/sigpipe.rs.
293 // Alaways using DEFAULT is okay since we don't support signals in Miri anyway.
300 Scalar::from_pointer(main_ptr, &ecx).into(),
303 Scalar::from_u8(sigpipe).into(),
305 Some(&ret_place.into()),
306 StackPopCleanup::Root { cleanup: true },
309 EntryFnType::Start => {
313 &[argc.into(), argv],
314 Some(&ret_place.into()),
315 StackPopCleanup::Root { cleanup: true },
323 /// Evaluates the entry function specified by `entry_id`.
324 /// Returns `Some(return_code)` if program executed completed.
325 /// Returns `None` if an evaluation error occurred.
326 #[allow(clippy::needless_lifetimes)]
327 pub fn eval_entry<'tcx>(
330 entry_type: EntryFnType,
333 // Copy setting before we move `config`.
334 let ignore_leaks = config.ignore_leaks;
336 let (mut ecx, ret_place) = match create_ecx(tcx, entry_id, entry_type, &config) {
339 err.print_backtrace();
340 panic!("Miri initialization error: {}", err.kind())
344 // Perform the main execution.
345 let res: thread::Result<InterpResult<'_, i64>> = panic::catch_unwind(AssertUnwindSafe(|| {
348 match ecx.schedule()? {
349 SchedulingAction::ExecuteStep => {
350 assert!(ecx.step()?, "a terminated thread was scheduled for execution");
352 SchedulingAction::ExecuteTimeoutCallback => {
353 ecx.run_timeout_callback()?;
355 SchedulingAction::ExecuteDtors => {
356 // This will either enable the thread again (so we go back
357 // to `ExecuteStep`), or determine that this thread is done
359 ecx.schedule_next_tls_dtor_for_active_thread()?;
361 SchedulingAction::Stop => {
366 let return_code = ecx.read_scalar(&ret_place.into())?.to_machine_isize(&ecx)?;
369 let res = res.unwrap_or_else(|panic_payload| {
371 panic::resume_unwind(panic_payload)
374 // Machine cleanup. Only do this if all threads have terminated; threads that are still running
375 // might cause Stacked Borrows errors (https://github.com/rust-lang/miri/issues/2396).
376 if ecx.have_all_terminated() {
377 // Even if all threads have terminated, we have to beware of data races since some threads
378 // might not have joined the main thread (https://github.com/rust-lang/miri/issues/2020,
379 // https://github.com/rust-lang/miri/issues/2508).
380 ecx.allow_data_races_all_threads_done();
381 EnvVars::cleanup(&mut ecx).expect("error during env var cleanup");
384 // Process the result.
388 // Check for thread leaks.
389 if !ecx.have_all_terminated() {
391 "the main thread terminated without waiting for all remaining threads",
393 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
396 // Check for memory leaks.
397 info!("Additonal static roots: {:?}", ecx.machine.static_roots);
398 let leaks = ecx.leak_report(&ecx.machine.static_roots);
400 tcx.sess.err("the evaluated program leaked memory");
401 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
402 // Ignore the provided return code - let the reported error
403 // determine the return code.
409 Err(e) => report_error(&ecx, e),
413 /// Turns an array of arguments into a Windows command line string.
415 /// The string will be UTF-16 encoded and NUL terminated.
417 /// Panics if the zeroth argument contains the `"` character because doublequotes
418 /// in `argv[0]` cannot be encoded using the standard command line parsing rules.
421 /// * [Parsing C++ command-line arguments](https://docs.microsoft.com/en-us/cpp/cpp/main-function-command-line-args?view=msvc-160#parsing-c-command-line-arguments)
422 /// * [The C/C++ Parameter Parsing Rules](https://daviddeley.com/autohotkey/parameters/parameters.htm#WINCRULES)
423 fn args_to_utf16_command_string<I, T>(mut args: I) -> Vec<u16>
425 I: Iterator<Item = T>,
428 // Parse argv[0]. Slashes aren't escaped. Literal double quotes are not allowed.
430 let arg0 = if let Some(arg0) = args.next() {
435 let arg0 = arg0.as_ref();
436 if arg0.contains('"') {
437 panic!("argv[0] cannot contain a doublequote (\") character");
439 // Always surround argv[0] with quotes.
440 let mut s = String::new();
448 // Build the other arguments.
450 let arg = arg.as_ref();
453 cmd.push_str("\"\"");
454 } else if !arg.bytes().any(|c| matches!(c, b'"' | b'\t' | b' ')) {
455 // No quote, tab, or space -- no escaping required.
458 // Spaces and tabs are escaped by surrounding them in quotes.
459 // Quotes are themselves escaped by using backslashes when in a
461 // Backslashes only need to be escaped when one or more are directly
462 // followed by a quote. Otherwise they are taken literally.
465 let mut chars = arg.chars().peekable();
467 let mut nslashes = 0;
468 while let Some(&'\\') = chars.peek() {
475 cmd.extend(iter::repeat('\\').take(nslashes * 2 + 1));
479 cmd.extend(iter::repeat('\\').take(nslashes));
483 cmd.extend(iter::repeat('\\').take(nslashes * 2));
492 if cmd.contains('\0') {
493 panic!("interior null in command line arguments");
495 cmd.encode_utf16().chain(iter::once(0)).collect()
502 #[should_panic(expected = "argv[0] cannot contain a doublequote (\") character")]
503 fn windows_argv0_panic_on_quote() {
504 args_to_utf16_command_string(["\""].iter());
507 fn windows_argv0_no_escape() {
508 // Ensure that a trailing backslash in argv[0] is not escaped.
509 let cmd = String::from_utf16_lossy(&args_to_utf16_command_string(
510 [r"C:\Program Files\", "arg1", "arg 2", "arg \" 3"].iter(),
512 assert_eq!(cmd.trim_end_matches('\0'), r#""C:\Program Files\" arg1 "arg 2" "arg \" 3""#);