1 //! Main evaluator loop and setting up the initial stack frame.
3 use std::collections::HashSet;
6 use std::panic::{self, AssertUnwindSafe};
11 use rustc_hir::def_id::DefId;
12 use rustc_middle::ty::{
14 layout::{LayoutCx, LayoutOf},
17 use rustc_target::spec::abi::Abi;
19 use rustc_session::config::EntryFnType;
23 #[derive(Copy, Clone, Debug, PartialEq)]
24 pub enum AlignmentCheck {
25 /// Do not check alignment.
27 /// Check alignment "symbolically", i.e., using only the requested alignment for an allocation and not its real base address.
29 /// Check alignment on the actual physical integer address.
33 #[derive(Copy, Clone, Debug, PartialEq)]
34 pub enum RejectOpWith {
35 /// Isolated op is rejected with an abort of the machine.
38 /// If not Abort, miri returns an error for an isolated op.
39 /// Following options determine if user should be warned about such error.
40 /// Do not print warning about rejected isolated op.
43 /// Print a warning about rejected isolated op, with backtrace.
46 /// Print a warning about rejected isolated op, without backtrace.
47 WarningWithoutBacktrace,
50 #[derive(Copy, Clone, Debug, PartialEq)]
52 /// Reject an op requiring communication with the host. By
53 /// default, miri rejects the op with an abort. If not, it returns
54 /// an error code, and prints a warning about it. Warning levels
55 /// are controlled by `RejectOpWith` enum.
58 /// Execute op requiring communication with the host, i.e. disable isolation.
62 #[derive(Copy, Clone, PartialEq, Eq)]
63 pub enum BacktraceStyle {
64 /// Prints a terser backtrace which ideally only contains relevant information.
66 /// Prints a backtrace with all possible information.
68 /// Prints only the frame that the error occurs in.
72 /// Configuration needed to spawn a Miri instance.
74 pub struct MiriConfig {
75 /// Determine if validity checking is enabled.
77 /// Determines if Stacked Borrows is enabled.
78 pub stacked_borrows: bool,
79 /// Controls alignment checking.
80 pub check_alignment: AlignmentCheck,
81 /// Controls function [ABI](Abi) checking.
83 /// Action for an op requiring communication with the host.
84 pub isolated_op: IsolatedOp,
85 /// Determines if memory leaks should be ignored.
86 pub ignore_leaks: bool,
87 /// Environment variables that should always be isolated from the host.
88 pub excluded_env_vars: Vec<String>,
89 /// Environment variables that should always be forwarded from the host.
90 pub forwarded_env_vars: Vec<String>,
91 /// Command-line arguments passed to the interpreted program.
92 pub args: Vec<String>,
93 /// The seed to use when non-determinism or randomness are required (e.g. ptr-to-int cast, `getrandom()`).
94 pub seed: Option<u64>,
95 /// The stacked borrows pointer ids to report about
96 pub tracked_pointer_tags: HashSet<SbTag>,
97 /// The stacked borrows call IDs to report about
98 pub tracked_call_ids: HashSet<CallId>,
99 /// The allocation ids to report about.
100 pub tracked_alloc_ids: HashSet<AllocId>,
101 /// Determine if data race detection should be enabled
102 pub data_race_detector: bool,
103 /// Determine if weak memory emulation should be enabled. Requires data race detection to be enabled
104 pub weak_memory_emulation: bool,
105 /// Track when an outdated (weak memory) load happens.
106 pub track_outdated_loads: bool,
107 /// Rate of spurious failures for compare_exchange_weak atomic operations,
108 /// between 0.0 and 1.0, defaulting to 0.8 (80% chance of failure).
109 pub cmpxchg_weak_failure_rate: f64,
110 /// If `Some`, enable the `measureme` profiler, writing results to a file
111 /// with the specified prefix.
112 pub measureme_out: Option<String>,
113 /// Panic when unsupported functionality is encountered.
114 pub panic_on_unsupported: bool,
115 /// Which style to use for printing backtraces.
116 pub backtrace_style: BacktraceStyle,
117 /// Which provenance to use for int2ptr casts
118 pub provenance_mode: ProvenanceMode,
119 /// Whether to ignore any output by the program. This is helpful when debugging miri
120 /// as its messages don't get intermingled with the program messages.
121 pub mute_stdout_stderr: bool,
122 /// The probability of the active thread being preempted at the end of each basic block.
123 pub preemption_rate: f64,
124 /// Report the current instruction being executed every N basic blocks.
125 pub report_progress: Option<u32>,
126 /// Whether Stacked Borrows retagging should recurse into fields of datatypes.
127 pub retag_fields: bool,
130 impl Default for MiriConfig {
131 fn default() -> MiriConfig {
134 stacked_borrows: true,
135 check_alignment: AlignmentCheck::Int,
137 isolated_op: IsolatedOp::Reject(RejectOpWith::Abort),
139 excluded_env_vars: vec![],
140 forwarded_env_vars: vec![],
143 tracked_pointer_tags: HashSet::default(),
144 tracked_call_ids: HashSet::default(),
145 tracked_alloc_ids: HashSet::default(),
146 data_race_detector: true,
147 weak_memory_emulation: true,
148 track_outdated_loads: false,
149 cmpxchg_weak_failure_rate: 0.8, // 80%
151 panic_on_unsupported: false,
152 backtrace_style: BacktraceStyle::Short,
153 provenance_mode: ProvenanceMode::Default,
154 mute_stdout_stderr: false,
155 preemption_rate: 0.01, // 1%
156 report_progress: None,
162 /// Returns a freshly created `InterpCx`, along with an `MPlaceTy` representing
163 /// the location where the return value of the `start` function will be
165 /// Public because this is also used by `priroda`.
166 pub fn create_ecx<'mir, 'tcx: 'mir>(
169 entry_type: EntryFnType,
171 ) -> InterpResult<'tcx, (InterpCx<'mir, 'tcx, Evaluator<'mir, 'tcx>>, MPlaceTy<'tcx, Provenance>)> {
172 let param_env = ty::ParamEnv::reveal_all();
173 let layout_cx = LayoutCx { tcx, param_env };
174 let mut ecx = InterpCx::new(
176 rustc_span::source_map::DUMMY_SP,
178 Evaluator::new(config, layout_cx),
181 // Capture the current interpreter stack state (which should be empty) so that we can emit
182 // allocation-tracking and tag-tracking diagnostics for allocations which are part of the
183 // early runtime setup.
184 let info = ecx.preprocess_diagnostics();
186 // Some parts of initialization require a full `InterpCx`.
187 Evaluator::late_init(&mut ecx, config)?;
189 // Make sure we have MIR. We check MIR for some stable monomorphic function in libcore.
190 let sentinel = ecx.try_resolve_path(&["core", "ascii", "escape_default"]);
191 if !matches!(sentinel, Some(s) if tcx.is_mir_available(s.def.def_id())) {
193 "the current sysroot was built without `-Zalways-encode-mir`, or libcore seems missing. \
194 Use `cargo miri setup` to prepare a sysroot that is suitable for Miri."
198 // Setup first stack frame.
199 let entry_instance = ty::Instance::mono(tcx, entry_id);
201 // First argument is constructed later, because it's skipped if the entry function uses #[start].
203 // Second argument (argc): length of `config.args`.
204 let argc = Scalar::from_machine_usize(u64::try_from(config.args.len()).unwrap(), &ecx);
205 // Third argument (`argv`): created from `config.args`.
207 // Put each argument in memory, collect pointers.
208 let mut argvs = Vec::<Immediate<Provenance>>::new();
209 for arg in config.args.iter() {
210 // Make space for `0` terminator.
211 let size = u64::try_from(arg.len()).unwrap().checked_add(1).unwrap();
212 let arg_type = tcx.mk_array(tcx.types.u8, size);
214 ecx.allocate(ecx.layout_of(arg_type)?, MiriMemoryKind::Machine.into())?;
215 ecx.write_os_str_to_c_str(OsStr::new(arg), arg_place.ptr, size)?;
216 ecx.mark_immutable(&arg_place);
217 argvs.push(arg_place.to_ref(&ecx));
219 // Make an array with all these pointers, in the Miri memory.
220 let argvs_layout = ecx.layout_of(
221 tcx.mk_array(tcx.mk_imm_ptr(tcx.types.u8), u64::try_from(argvs.len()).unwrap()),
223 let argvs_place = ecx.allocate(argvs_layout, MiriMemoryKind::Machine.into())?;
224 for (idx, arg) in argvs.into_iter().enumerate() {
225 let place = ecx.mplace_field(&argvs_place, idx)?;
226 ecx.write_immediate(arg, &place.into())?;
228 ecx.mark_immutable(&argvs_place);
229 // A pointer to that place is the 3rd argument for main.
230 let argv = argvs_place.to_ref(&ecx);
231 // Store `argc` and `argv` for macOS `_NSGetArg{c,v}`.
234 ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
235 ecx.write_scalar(argc, &argc_place.into())?;
236 ecx.mark_immutable(&argc_place);
237 ecx.machine.argc = Some(*argc_place);
239 let argv_place = ecx.allocate(
240 ecx.layout_of(tcx.mk_imm_ptr(tcx.types.unit))?,
241 MiriMemoryKind::Machine.into(),
243 ecx.write_immediate(argv, &argv_place.into())?;
244 ecx.mark_immutable(&argv_place);
245 ecx.machine.argv = Some(*argv_place);
247 // Store command line as UTF-16 for Windows `GetCommandLineW`.
249 // Construct a command string with all the aguments.
250 let cmd_utf16: Vec<u16> = args_to_utf16_command_string(config.args.iter());
252 let cmd_type = tcx.mk_array(tcx.types.u16, u64::try_from(cmd_utf16.len()).unwrap());
254 ecx.allocate(ecx.layout_of(cmd_type)?, MiriMemoryKind::Machine.into())?;
255 ecx.machine.cmd_line = Some(*cmd_place);
256 // Store the UTF-16 string. We just allocated so we know the bounds are fine.
257 for (idx, &c) in cmd_utf16.iter().enumerate() {
258 let place = ecx.mplace_field(&cmd_place, idx)?;
259 ecx.write_scalar(Scalar::from_u16(c), &place.into())?;
261 ecx.mark_immutable(&cmd_place);
266 // Return place (in static memory so that it does not count as leak).
267 let ret_place = ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
268 // Call start function.
271 EntryFnType::Main => {
272 let start_id = tcx.lang_items().start_fn().unwrap();
273 let main_ret_ty = tcx.fn_sig(entry_id).output();
274 let main_ret_ty = main_ret_ty.no_bound_vars().unwrap();
275 let start_instance = ty::Instance::resolve(
277 ty::ParamEnv::reveal_all(),
279 tcx.mk_substs(::std::iter::once(ty::subst::GenericArg::from(main_ret_ty))),
284 let main_ptr = ecx.create_fn_alloc_ptr(FnVal::Instance(entry_instance));
289 &[Scalar::from_pointer(main_ptr, &ecx).into(), argc.into(), argv],
290 Some(&ret_place.into()),
291 StackPopCleanup::Root { cleanup: true },
294 EntryFnType::Start => {
298 &[argc.into(), argv],
299 Some(&ret_place.into()),
300 StackPopCleanup::Root { cleanup: true },
305 // Emit any diagnostics related to the setup process for the runtime, so that when the
306 // interpreter loop starts there are no unprocessed diagnostics.
307 ecx.process_diagnostics(info);
312 /// Evaluates the entry function specified by `entry_id`.
313 /// Returns `Some(return_code)` if program executed completed.
314 /// Returns `None` if an evaluation error occured.
315 pub fn eval_entry<'tcx>(
318 entry_type: EntryFnType,
321 // Copy setting before we move `config`.
322 let ignore_leaks = config.ignore_leaks;
324 let (mut ecx, ret_place) = match create_ecx(tcx, entry_id, entry_type, &config) {
327 err.print_backtrace();
328 panic!("Miri initialization error: {}", err.kind())
332 // Perform the main execution.
333 let res: thread::Result<InterpResult<'_, i64>> = panic::catch_unwind(AssertUnwindSafe(|| {
336 let info = ecx.preprocess_diagnostics();
337 match ecx.schedule()? {
338 SchedulingAction::ExecuteStep => {
339 assert!(ecx.step()?, "a terminated thread was scheduled for execution");
341 SchedulingAction::ExecuteTimeoutCallback => {
343 ecx.machine.communicate(),
344 "scheduler callbacks require disabled isolation, but the code \
345 that created the callback did not check it"
347 ecx.run_timeout_callback()?;
349 SchedulingAction::ExecuteDtors => {
350 // This will either enable the thread again (so we go back
351 // to `ExecuteStep`), or determine that this thread is done
353 ecx.schedule_next_tls_dtor_for_active_thread()?;
355 SchedulingAction::Stop => {
359 ecx.process_diagnostics(info);
361 let return_code = ecx.read_scalar(&ret_place.into())?.to_machine_isize(&ecx)?;
364 let res = res.unwrap_or_else(|panic_payload| {
366 panic::resume_unwind(panic_payload)
369 // Machine cleanup. Only do this if all threads have terminated; threads that are still running
370 // might cause data races (https://github.com/rust-lang/miri/issues/2020) or Stacked Borrows
371 // errors (https://github.com/rust-lang/miri/issues/2396) if we deallocate here.
372 if ecx.have_all_terminated() {
373 EnvVars::cleanup(&mut ecx).unwrap();
376 // Process the result.
380 // Check for thread leaks.
381 if !ecx.have_all_terminated() {
383 "the main thread terminated without waiting for all remaining threads",
385 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
388 // Check for memory leaks.
389 info!("Additonal static roots: {:?}", ecx.machine.static_roots);
390 let leaks = ecx.leak_report(&ecx.machine.static_roots);
392 tcx.sess.err("the evaluated program leaked memory");
393 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
394 // Ignore the provided return code - let the reported error
395 // determine the return code.
401 Err(e) => report_error(&ecx, e),
405 /// Turns an array of arguments into a Windows command line string.
407 /// The string will be UTF-16 encoded and NUL terminated.
409 /// Panics if the zeroth argument contains the `"` character because doublequotes
410 /// in `argv[0]` cannot be encoded using the standard command line parsing rules.
413 /// * [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)
414 /// * [The C/C++ Parameter Parsing Rules](https://daviddeley.com/autohotkey/parameters/parameters.htm#WINCRULES)
415 fn args_to_utf16_command_string<I, T>(mut args: I) -> Vec<u16>
417 I: Iterator<Item = T>,
420 // Parse argv[0]. Slashes aren't escaped. Literal double quotes are not allowed.
422 let arg0 = if let Some(arg0) = args.next() {
427 let arg0 = arg0.as_ref();
428 if arg0.contains('"') {
429 panic!("argv[0] cannot contain a doublequote (\") character");
431 // Always surround argv[0] with quotes.
432 let mut s = String::new();
440 // Build the other arguments.
442 let arg = arg.as_ref();
445 cmd.push_str("\"\"");
446 } else if !arg.bytes().any(|c| matches!(c, b'"' | b'\t' | b' ')) {
447 // No quote, tab, or space -- no escaping required.
450 // Spaces and tabs are escaped by surrounding them in quotes.
451 // Quotes are themselves escaped by using backslashes when in a
453 // Backslashes only need to be escaped when one or more are directly
454 // followed by a quote. Otherwise they are taken literally.
457 let mut chars = arg.chars().peekable();
459 let mut nslashes = 0;
460 while let Some(&'\\') = chars.peek() {
467 cmd.extend(iter::repeat('\\').take(nslashes * 2 + 1));
471 cmd.extend(iter::repeat('\\').take(nslashes));
475 cmd.extend(iter::repeat('\\').take(nslashes * 2));
484 if cmd.contains('\0') {
485 panic!("interior null in command line arguments");
487 cmd.encode_utf16().chain(iter::once(0)).collect()
494 #[should_panic(expected = "argv[0] cannot contain a doublequote (\") character")]
495 fn windows_argv0_panic_on_quote() {
496 args_to_utf16_command_string(["\""].iter());
499 fn windows_argv0_no_escape() {
500 // Ensure that a trailing backslash in argv[0] is not escaped.
501 let cmd = String::from_utf16_lossy(&args_to_utf16_command_string(
502 [r"C:\Program Files\", "arg1", "arg 2", "arg \" 3"].iter(),
504 assert_eq!(cmd.trim_end_matches('\0'), r#""C:\Program Files\" arg1 "arg 2" "arg \" 3""#);