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 /// Rate of spurious failures for compare_exchange_weak atomic operations,
106 /// between 0.0 and 1.0, defaulting to 0.8 (80% chance of failure).
107 pub cmpxchg_weak_failure_rate: f64,
108 /// If `Some`, enable the `measureme` profiler, writing results to a file
109 /// with the specified prefix.
110 pub measureme_out: Option<String>,
111 /// Panic when unsupported functionality is encountered.
112 pub panic_on_unsupported: bool,
113 /// Which style to use for printing backtraces.
114 pub backtrace_style: BacktraceStyle,
115 /// Which provenance to use for int2ptr casts
116 pub provenance_mode: ProvenanceMode,
117 /// Whether to ignore any output by the program. This is helpful when debugging miri
118 /// as its messages don't get intermingled with the program messages.
119 pub mute_stdout_stderr: bool,
120 /// The probability of the active thread being preempted at the end of each basic block.
121 pub preemption_rate: f64,
122 /// Report the current instruction being executed every N basic blocks.
123 pub report_progress: Option<u32>,
124 /// Whether Stacked Borrows retagging should recurse into fields of datatypes.
125 pub retag_fields: bool,
128 impl Default for MiriConfig {
129 fn default() -> MiriConfig {
132 stacked_borrows: true,
133 check_alignment: AlignmentCheck::Int,
135 isolated_op: IsolatedOp::Reject(RejectOpWith::Abort),
137 excluded_env_vars: vec![],
138 forwarded_env_vars: vec![],
141 tracked_pointer_tags: HashSet::default(),
142 tracked_call_ids: HashSet::default(),
143 tracked_alloc_ids: HashSet::default(),
144 data_race_detector: true,
145 weak_memory_emulation: true,
146 cmpxchg_weak_failure_rate: 0.8, // 80%
148 panic_on_unsupported: false,
149 backtrace_style: BacktraceStyle::Short,
150 provenance_mode: ProvenanceMode::Default,
151 mute_stdout_stderr: false,
152 preemption_rate: 0.01, // 1%
153 report_progress: None,
159 /// Returns a freshly created `InterpCx`, along with an `MPlaceTy` representing
160 /// the location where the return value of the `start` function will be
162 /// Public because this is also used by `priroda`.
163 pub fn create_ecx<'mir, 'tcx: 'mir>(
166 entry_type: EntryFnType,
168 ) -> InterpResult<'tcx, (InterpCx<'mir, 'tcx, Evaluator<'mir, 'tcx>>, MPlaceTy<'tcx, Tag>)> {
169 let param_env = ty::ParamEnv::reveal_all();
170 let layout_cx = LayoutCx { tcx, param_env };
171 let mut ecx = InterpCx::new(
173 rustc_span::source_map::DUMMY_SP,
175 Evaluator::new(config, layout_cx),
178 // Capture the current interpreter stack state (which should be empty) so that we can emit
179 // allocation-tracking and tag-tracking diagnostics for allocations which are part of the
180 // early runtime setup.
181 let info = ecx.preprocess_diagnostics();
183 // Some parts of initialization require a full `InterpCx`.
184 Evaluator::late_init(&mut ecx, config)?;
186 // Make sure we have MIR. We check MIR for some stable monomorphic function in libcore.
187 let sentinel = ecx.try_resolve_path(&["core", "ascii", "escape_default"]);
188 if !matches!(sentinel, Some(s) if tcx.is_mir_available(s.def.def_id())) {
190 "the current sysroot was built without `-Zalways-encode-mir`, or libcore seems missing. \
191 Use `cargo miri setup` to prepare a sysroot that is suitable for Miri."
195 // Setup first stack frame.
196 let entry_instance = ty::Instance::mono(tcx, entry_id);
198 // First argument is constructed later, because it's skipped if the entry function uses #[start].
200 // Second argument (argc): length of `config.args`.
201 let argc = Scalar::from_machine_usize(u64::try_from(config.args.len()).unwrap(), &ecx);
202 // Third argument (`argv`): created from `config.args`.
204 // Put each argument in memory, collect pointers.
205 let mut argvs = Vec::<Immediate<Tag>>::new();
206 for arg in config.args.iter() {
207 // Make space for `0` terminator.
208 let size = u64::try_from(arg.len()).unwrap().checked_add(1).unwrap();
209 let arg_type = tcx.mk_array(tcx.types.u8, size);
211 ecx.allocate(ecx.layout_of(arg_type)?, MiriMemoryKind::Machine.into())?;
212 ecx.write_os_str_to_c_str(OsStr::new(arg), arg_place.ptr, size)?;
213 ecx.mark_immutable(&arg_place);
214 argvs.push(arg_place.to_ref(&ecx));
216 // Make an array with all these pointers, in the Miri memory.
217 let argvs_layout = ecx.layout_of(
218 tcx.mk_array(tcx.mk_imm_ptr(tcx.types.u8), u64::try_from(argvs.len()).unwrap()),
220 let argvs_place = ecx.allocate(argvs_layout, MiriMemoryKind::Machine.into())?;
221 for (idx, arg) in argvs.into_iter().enumerate() {
222 let place = ecx.mplace_field(&argvs_place, idx)?;
223 ecx.write_immediate(arg, &place.into())?;
225 ecx.mark_immutable(&argvs_place);
226 // A pointer to that place is the 3rd argument for main.
227 let argv = argvs_place.to_ref(&ecx);
228 // Store `argc` and `argv` for macOS `_NSGetArg{c,v}`.
231 ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
232 ecx.write_scalar(argc, &argc_place.into())?;
233 ecx.mark_immutable(&argc_place);
234 ecx.machine.argc = Some(*argc_place);
236 let argv_place = ecx.allocate(
237 ecx.layout_of(tcx.mk_imm_ptr(tcx.types.unit))?,
238 MiriMemoryKind::Machine.into(),
240 ecx.write_immediate(argv, &argv_place.into())?;
241 ecx.mark_immutable(&argv_place);
242 ecx.machine.argv = Some(*argv_place);
244 // Store command line as UTF-16 for Windows `GetCommandLineW`.
246 // Construct a command string with all the aguments.
247 let cmd_utf16: Vec<u16> = args_to_utf16_command_string(config.args.iter());
249 let cmd_type = tcx.mk_array(tcx.types.u16, u64::try_from(cmd_utf16.len()).unwrap());
251 ecx.allocate(ecx.layout_of(cmd_type)?, MiriMemoryKind::Machine.into())?;
252 ecx.machine.cmd_line = Some(*cmd_place);
253 // Store the UTF-16 string. We just allocated so we know the bounds are fine.
254 for (idx, &c) in cmd_utf16.iter().enumerate() {
255 let place = ecx.mplace_field(&cmd_place, idx)?;
256 ecx.write_scalar(Scalar::from_u16(c), &place.into())?;
258 ecx.mark_immutable(&cmd_place);
263 // Return place (in static memory so that it does not count as leak).
264 let ret_place = ecx.allocate(ecx.machine.layouts.isize, MiriMemoryKind::Machine.into())?;
265 // Call start function.
268 EntryFnType::Main => {
269 let start_id = tcx.lang_items().start_fn().unwrap();
270 let main_ret_ty = tcx.fn_sig(entry_id).output();
271 let main_ret_ty = main_ret_ty.no_bound_vars().unwrap();
272 let start_instance = ty::Instance::resolve(
274 ty::ParamEnv::reveal_all(),
276 tcx.mk_substs(::std::iter::once(ty::subst::GenericArg::from(main_ret_ty))),
281 let main_ptr = ecx.create_fn_alloc_ptr(FnVal::Instance(entry_instance));
286 &[Scalar::from_pointer(main_ptr, &ecx).into(), argc.into(), argv],
287 Some(&ret_place.into()),
288 StackPopCleanup::Root { cleanup: true },
291 EntryFnType::Start => {
295 &[argc.into(), argv],
296 Some(&ret_place.into()),
297 StackPopCleanup::Root { cleanup: true },
302 // Emit any diagnostics related to the setup process for the runtime, so that when the
303 // interpreter loop starts there are no unprocessed diagnostics.
304 ecx.process_diagnostics(info);
309 /// Evaluates the entry function specified by `entry_id`.
310 /// Returns `Some(return_code)` if program executed completed.
311 /// Returns `None` if an evaluation error occured.
312 pub fn eval_entry<'tcx>(
315 entry_type: EntryFnType,
318 // Copy setting before we move `config`.
319 let ignore_leaks = config.ignore_leaks;
321 let (mut ecx, ret_place) = match create_ecx(tcx, entry_id, entry_type, &config) {
324 err.print_backtrace();
325 panic!("Miri initialization error: {}", err.kind())
329 // Perform the main execution.
330 let res: thread::Result<InterpResult<'_, i64>> = panic::catch_unwind(AssertUnwindSafe(|| {
333 let info = ecx.preprocess_diagnostics();
334 match ecx.schedule()? {
335 SchedulingAction::ExecuteStep => {
336 assert!(ecx.step()?, "a terminated thread was scheduled for execution");
338 SchedulingAction::ExecuteTimeoutCallback => {
340 ecx.machine.communicate(),
341 "scheduler callbacks require disabled isolation, but the code \
342 that created the callback did not check it"
344 ecx.run_timeout_callback()?;
346 SchedulingAction::ExecuteDtors => {
347 // This will either enable the thread again (so we go back
348 // to `ExecuteStep`), or determine that this thread is done
350 ecx.schedule_next_tls_dtor_for_active_thread()?;
352 SchedulingAction::Stop => {
356 ecx.process_diagnostics(info);
358 let return_code = ecx.read_scalar(&ret_place.into())?.to_machine_isize(&ecx)?;
361 let res = res.unwrap_or_else(|panic_payload| {
363 panic::resume_unwind(panic_payload)
366 // Machine cleanup. Only do this if all threads have terminated; threads that are still running
367 // might cause data races (https://github.com/rust-lang/miri/issues/2020) or Stacked Borrows
368 // errors (https://github.com/rust-lang/miri/issues/2396) if we deallocate here.
369 if ecx.have_all_terminated() {
370 EnvVars::cleanup(&mut ecx).unwrap();
373 // Process the result.
377 // Check for thread leaks.
378 if !ecx.have_all_terminated() {
380 "the main thread terminated without waiting for all remaining threads",
382 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
385 // Check for memory leaks.
386 info!("Additonal static roots: {:?}", ecx.machine.static_roots);
387 let leaks = ecx.leak_report(&ecx.machine.static_roots);
389 tcx.sess.err("the evaluated program leaked memory");
390 tcx.sess.note_without_error("pass `-Zmiri-ignore-leaks` to disable this check");
391 // Ignore the provided return code - let the reported error
392 // determine the return code.
398 Err(e) => report_error(&ecx, e),
402 /// Turns an array of arguments into a Windows command line string.
404 /// The string will be UTF-16 encoded and NUL terminated.
406 /// Panics if the zeroth argument contains the `"` character because doublequotes
407 /// in `argv[0]` cannot be encoded using the standard command line parsing rules.
410 /// * [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)
411 /// * [The C/C++ Parameter Parsing Rules](https://daviddeley.com/autohotkey/parameters/parameters.htm#WINCRULES)
412 fn args_to_utf16_command_string<I, T>(mut args: I) -> Vec<u16>
414 I: Iterator<Item = T>,
417 // Parse argv[0]. Slashes aren't escaped. Literal double quotes are not allowed.
419 let arg0 = if let Some(arg0) = args.next() {
424 let arg0 = arg0.as_ref();
425 if arg0.contains('"') {
426 panic!("argv[0] cannot contain a doublequote (\") character");
428 // Always surround argv[0] with quotes.
429 let mut s = String::new();
437 // Build the other arguments.
439 let arg = arg.as_ref();
442 cmd.push_str("\"\"");
443 } else if !arg.bytes().any(|c| matches!(c, b'"' | b'\t' | b' ')) {
444 // No quote, tab, or space -- no escaping required.
447 // Spaces and tabs are escaped by surrounding them in quotes.
448 // Quotes are themselves escaped by using backslashes when in a
450 // Backslashes only need to be escaped when one or more are directly
451 // followed by a quote. Otherwise they are taken literally.
454 let mut chars = arg.chars().peekable();
456 let mut nslashes = 0;
457 while let Some(&'\\') = chars.peek() {
464 cmd.extend(iter::repeat('\\').take(nslashes * 2 + 1));
468 cmd.extend(iter::repeat('\\').take(nslashes));
472 cmd.extend(iter::repeat('\\').take(nslashes * 2));
481 if cmd.contains('\0') {
482 panic!("interior null in command line arguments");
484 cmd.encode_utf16().chain(iter::once(0)).collect()
491 #[should_panic(expected = "argv[0] cannot contain a doublequote (\") character")]
492 fn windows_argv0_panic_on_quote() {
493 args_to_utf16_command_string(["\""].iter());
496 fn windows_argv0_no_escape() {
497 // Ensure that a trailing backslash in argv[0] is not escaped.
498 let cmd = String::from_utf16_lossy(&args_to_utf16_command_string(
499 [r"C:\Program Files\", "arg1", "arg 2", "arg \" 3"].iter(),
501 assert_eq!(cmd.trim_end_matches('\0'), r#""C:\Program Files\" arg1 "arg 2" "arg \" 3""#);