Merge pull request #4418 from RalfJung/native-lib-tracing

various minor native-lib-tracing tweaks

Author: RalfJung

src/tools/miri/README.md

@@ -419,11 +419,9 @@ to Miri failing to detect cases of undefined behavior in a program.
   Finally, the flag is **unsound** in the sense that Miri stops tracking details such as
   initialization and provenance on memory shared with native code, so it is easily possible to write
   code that has UB which is missed by Miri.
-* `-Zmiri-force-old-native-lib-mode` disables the WIP improved native code access tracking. If for
-  whatever reason enabling native calls leads to odd behaviours or causes Miri to panic, disabling
-  the tracer *might* fix this. This will likely be removed once the tracer has been adequately
-  battle-tested. Note that this flag is only meaningful on Linux systems; other Unixes (currently)
-  exclusively use the old native-lib code.
+* `-Zmiri-native-lib-enable-tracing` enables the WIP detailed tracing mode for invoking native code.
+  Note that this flag is only meaningful on Linux systems; other Unixes (currently) do not support
+  tracing mode.
 * `-Zmiri-measureme=<name>` enables `measureme` profiling for the interpreted program.
    This can be used to find which parts of your program are executing slowly under Miri.
    The profile is written out to a file inside a directory called `<name>`, and can be processed

src/tools/miri/src/alloc/isolated_alloc.rs

@@ -305,12 +305,12 @@ impl IsolatedAlloc {
     /// Returns a vector of page addresses managed by the allocator.
     pub fn pages(&self) -> Vec<usize> {
         let mut pages: Vec<usize> =
-            self.page_ptrs.clone().into_iter().map(|p| p.expose_provenance().get()).collect();
-        self.huge_ptrs.iter().for_each(|(ptr, size)| {
+            self.page_ptrs.iter().map(|p| p.expose_provenance().get()).collect();
+        for (ptr, size) in self.huge_ptrs.iter() {
             for i in 0..size / self.page_size {
                 pages.push(ptr.expose_provenance().get().strict_add(i * self.page_size));
             }
-        });
+        }
         pages
     }
 

src/tools/miri/src/bin/miri.rs

@@ -228,7 +228,7 @@ impl rustc_driver::Callbacks for MiriCompilerCalls {
             let return_code = miri::eval_entry(tcx, entry_def_id, entry_type, &config, None)
                 .unwrap_or_else(|| {
                     #[cfg(target_os = "linux")]
-                    miri::register_retcode_sv(rustc_driver::EXIT_FAILURE);
+                    miri::native_lib::register_retcode_sv(rustc_driver::EXIT_FAILURE);
                     tcx.dcx().abort_if_errors();
                     rustc_driver::EXIT_FAILURE
                 });
@@ -724,8 +724,8 @@ fn main() {
             } else {
                 show_error!("-Zmiri-native-lib `{}` does not exist", filename);
             }
-        } else if arg == "-Zmiri-force-old-native-lib-mode" {
-            miri_config.force_old_native_lib = true;
+        } else if arg == "-Zmiri-native-lib-enable-tracing" {
+            miri_config.native_lib_enable_tracing = true;
         } else if let Some(param) = arg.strip_prefix("-Zmiri-num-cpus=") {
             let num_cpus = param
                 .parse::<u32>()
@@ -797,14 +797,14 @@ fn main() {
     debug!("rustc arguments: {:?}", rustc_args);
     debug!("crate arguments: {:?}", miri_config.args);
     #[cfg(target_os = "linux")]
-    if !miri_config.native_lib.is_empty() && !miri_config.force_old_native_lib {
+    if !miri_config.native_lib.is_empty() && miri_config.native_lib_enable_tracing {
         // FIXME: This should display a diagnostic / warning on error
         // SAFETY: If any other threads exist at this point (namely for the ctrlc
         // handler), they will not interact with anything on the main rustc/Miri
         // thread in an async-signal-unsafe way such as by accessing shared
         // semaphores, etc.; the handler only calls `sleep()` and `exit()`, which
         // are async-signal-safe, as is accessing atomics
-        let _ = unsafe { miri::init_sv() };
+        let _ = unsafe { miri::native_lib::init_sv() };
     }
     run_compiler_and_exit(
         &rustc_args,

src/tools/miri/src/diagnostics.rs

@@ -132,8 +132,9 @@ pub enum NonHaltingDiagnostic {
     Int2Ptr {
         details: bool,
     },
-    NativeCallSharedMem,
-    NativeCallNoTrace,
+    NativeCallSharedMem {
+        tracing: bool,
+    },
     WeakMemoryOutdatedLoad {
         ptr: Pointer,
     },
@@ -628,10 +629,8 @@ impl<'tcx> MiriMachine<'tcx> {
             RejectedIsolatedOp(_) =>
                 ("operation rejected by isolation".to_string(), DiagLevel::Warning),
             Int2Ptr { .. } => ("integer-to-pointer cast".to_string(), DiagLevel::Warning),
-            NativeCallSharedMem =>
+            NativeCallSharedMem { .. } =>
                 ("sharing memory with a native function".to_string(), DiagLevel::Warning),
-            NativeCallNoTrace =>
-                ("unable to trace native code memory accesses".to_string(), DiagLevel::Warning),
             ExternTypeReborrow =>
                 ("reborrow of reference to `extern type`".to_string(), DiagLevel::Warning),
             CreatedPointerTag(..)
@@ -666,11 +665,8 @@ impl<'tcx> MiriMachine<'tcx> {
             ProgressReport { .. } =>
                 format!("progress report: current operation being executed is here"),
             Int2Ptr { .. } => format!("integer-to-pointer cast"),
-            NativeCallSharedMem => format!("sharing memory with a native function called via FFI"),
-            NativeCallNoTrace =>
-                format!(
-                    "sharing memory with a native function called via FFI, and unable to use ptrace"
-                ),
+            NativeCallSharedMem { .. } =>
+                format!("sharing memory with a native function called via FFI"),
             WeakMemoryOutdatedLoad { ptr } =>
                 format!("weak memory emulation: outdated value returned from load at {ptr}"),
             ExternTypeReborrow =>
@@ -716,42 +712,41 @@ impl<'tcx> MiriMachine<'tcx> {
                 }
                 v
             }
-            NativeCallSharedMem => {
-                vec![
-                    note!(
-                        "when memory is shared with a native function call, Miri can only track initialisation and provenance on a best-effort basis"
-                    ),
-                    note!(
-                        "in particular, Miri assumes that the native call initializes all memory it has written to"
-                    ),
-                    note!(
-                        "Miri also assumes that any part of this memory may be a pointer that is permitted to point to arbitrary exposed memory"
-                    ),
-                    note!(
-                        "what this means is that Miri will easily miss Undefined Behavior related to incorrect usage of this shared memory, so you should not take a clean Miri run as a signal that your FFI code is UB-free"
-                    ),
-                ]
-            }
-            NativeCallNoTrace => {
-                vec![
-                    note!(
-                        "when memory is shared with a native function call, Miri stops tracking initialization and provenance for that memory"
-                    ),
-                    note!(
-                        "in particular, Miri assumes that the native call initializes all memory it has access to"
-                    ),
-                    note!(
-                        "Miri also assumes that any part of this memory may be a pointer that is permitted to point to arbitrary exposed memory"
-                    ),
-                    note!(
-                        "what this means is that Miri will easily miss Undefined Behavior related to incorrect usage of this shared memory, so you should not take a clean Miri run as a signal that your FFI code is UB-free"
-                    ),
-                    #[cfg(target_os = "linux")]
-                    note!(
-                        "this is normally partially mitigated, but either -Zmiri-force-old-native-lib-mode was passed or ptrace is disabled on your system"
-                    ),
-                ]
-            }
+            NativeCallSharedMem { tracing } =>
+                if *tracing {
+                    vec![
+                        note!(
+                            "when memory is shared with a native function call, Miri can only track initialisation and provenance on a best-effort basis"
+                        ),
+                        note!(
+                            "in particular, Miri assumes that the native call initializes all memory it has written to"
+                        ),
+                        note!(
+                            "Miri also assumes that any part of this memory may be a pointer that is permitted to point to arbitrary exposed memory"
+                        ),
+                        note!(
+                            "what this means is that Miri will easily miss Undefined Behavior related to incorrect usage of this shared memory, so you should not take a clean Miri run as a signal that your FFI code is UB-free"
+                        ),
+                        note!(
+                            "tracing memory accesses in native code is not yet fully implemented, so there can be further imprecisions beyond what is documented here"
+                        ),
+                    ]
+                } else {
+                    vec![
+                        note!(
+                            "when memory is shared with a native function call, Miri stops tracking initialization and provenance for that memory"
+                        ),
+                        note!(
+                            "in particular, Miri assumes that the native call initializes all memory it has access to"
+                        ),
+                        note!(
+                            "Miri also assumes that any part of this memory may be a pointer that is permitted to point to arbitrary exposed memory"
+                        ),
+                        note!(
+                            "what this means is that Miri will easily miss Undefined Behavior related to incorrect usage of this shared memory, so you should not take a clean Miri run as a signal that your FFI code is UB-free"
+                        ),
+                    ]
+                },
             ExternTypeReborrow => {
                 assert!(self.borrow_tracker.as_ref().is_some_and(|b| {
                     matches!(

src/tools/miri/src/eval.rs

@@ -150,8 +150,8 @@ pub struct MiriConfig {
     pub retag_fields: RetagFields,
     /// The location of the shared object files to load when calling external functions
     pub native_lib: Vec<PathBuf>,
-    /// Whether to force using the old native lib behaviour even if ptrace might be supported.
-    pub force_old_native_lib: bool,
+    /// Whether to enable the new native lib tracing system.
+    pub native_lib_enable_tracing: bool,
     /// Run a garbage collector for BorTags every N basic blocks.
     pub gc_interval: u32,
     /// The number of CPUs to be reported by miri.
@@ -201,7 +201,7 @@ impl Default for MiriConfig {
             report_progress: None,
             retag_fields: RetagFields::Yes,
             native_lib: vec![],
-            force_old_native_lib: false,
+            native_lib_enable_tracing: false,
             gc_interval: 10_000,
             num_cpus: 1,
             page_size: None,

src/tools/miri/src/lib.rs

@@ -100,7 +100,9 @@ use rustc_middle::{bug, span_bug};
 use tracing::{info, trace};
 
 #[cfg(target_os = "linux")]
-pub use crate::shims::trace::{init_sv, register_retcode_sv};
+pub mod native_lib {
+    pub use crate::shims::{init_sv, register_retcode_sv};
+}
 
 // Type aliases that set the provenance parameter.
 pub type Pointer = interpret::Pointer<Option<machine::Provenance>>;

src/tools/miri/src/shims/mod.rs

@@ -19,10 +19,10 @@ pub mod os_str;
 pub mod panic;
 pub mod time;
 pub mod tls;
-#[cfg(target_os = "linux")]
-pub mod trace;
 
 pub use self::files::FdTable;
+#[cfg(target_os = "linux")]
+pub use self::native_lib::trace::{init_sv, register_retcode_sv};
 pub use self::unix::{DirTable, EpollInterestTable};
 
 /// What needs to be done after emulating an item (a shim or an intrinsic) is done.