1 files changed, 73 insertions, 8 deletions

diff --git a/third_party/rust/minidump-writer/src/linux/ptrace_dumper.rs b/third_party/rust/minidump-writer/src/linux/ptrace_dumper.rs
index f75499bcdd..0dd0fa2719 100644
--- a/third_party/rust/minidump-writer/src/linux/ptrace_dumper.rs
+++ b/third_party/rust/minidump-writer/src/linux/ptrace_dumper.rs
@@ -15,10 +15,20 @@ use crate::{
 use goblin::elf;
 use nix::{
     errno::Errno,
-    sys::{ptrace, wait},
+    sys::{ptrace, signal, wait},
+};
+use procfs_core::{
+    process::{MMPermissions, ProcState, Stat},
+    FromRead, ProcError,
+};
+use std::{
+    collections::HashMap,
+    ffi::c_void,
+    io::BufReader,
+    path,
+    result::Result,
+    time::{Duration, Instant},
 };
-use procfs_core::process::MMPermissions;
-use std::{collections::HashMap, ffi::c_void, io::BufReader, path, result::Result};
 
 #[derive(Debug, Clone)]
 pub struct Thread {
@@ -45,9 +55,27 @@ impl Drop for PtraceDumper {
     fn drop(&mut self) {
         // Always try to resume all threads (e.g. in case of error)
         let _ = self.resume_threads();
+        // Always allow the process to continue.
+        let _ = self.continue_process();
     }
 }
 
+#[derive(Debug, thiserror::Error)]
+enum StopProcessError {
+    #[error("Failed to stop the process")]
+    Stop(#[from] Errno),
+    #[error("Failed to get the process state")]
+    State(#[from] ProcError),
+    #[error("Timeout waiting for process to stop")]
+    Timeout,
+}
+
+#[derive(Debug, thiserror::Error)]
+enum ContinueProcessError {
+    #[error("Failed to continue the process")]
+    Continue(#[from] Errno),
+}
+
 /// PTRACE_DETACH the given pid.
 ///
 /// This handles special errno cases (ESRCH) which we won't consider errors.
@@ -67,7 +95,7 @@ fn ptrace_detach(child: Pid) -> Result<(), DumperError> {
 impl PtraceDumper {
     /// Constructs a dumper for extracting information of a given process
     /// with a process ID of |pid|.
-    pub fn new(pid: Pid) -> Result<Self, InitError> {
+    pub fn new(pid: Pid, stop_timeout: Duration) -> Result<Self, InitError> {
         let mut dumper = PtraceDumper {
             pid,
             threads_suspended: false,
@@ -76,12 +104,16 @@ impl PtraceDumper {
             mappings: Vec::new(),
             page_size: 0,
         };
-        dumper.init()?;
+        dumper.init(stop_timeout)?;
         Ok(dumper)
     }
 
     // TODO: late_init for chromeos and android
-    pub fn init(&mut self) -> Result<(), InitError> {
+    pub fn init(&mut self, stop_timeout: Duration) -> Result<(), InitError> {
+        // Stopping the process is best-effort.
+        if let Err(e) = self.stop_process(stop_timeout) {
+            log::warn!("failed to stop process {}: {e}", self.pid);
+        }
         self.read_auxv()?;
         self.enumerate_threads()?;
         self.enumerate_mappings()?;
@@ -207,6 +239,38 @@ impl PtraceDumper {
         result
     }
 
+    /// Send SIGSTOP to the process so that we can get a consistent state.
+    ///
+    /// This will block waiting for the process to stop until `timeout` has passed.
+    fn stop_process(&mut self, timeout: Duration) -> Result<(), StopProcessError> {
+        signal::kill(nix::unistd::Pid::from_raw(self.pid), Some(signal::SIGSTOP))?;
+
+        // Something like waitpid for non-child processes would be better, but we have no such
+        // tool, so we poll the status.
+        const POLL_INTERVAL: Duration = Duration::from_millis(1);
+        let proc_file = format!("/proc/{}/stat", self.pid);
+        let end = Instant::now() + timeout;
+
+        loop {
+            if let Ok(ProcState::Stopped) = Stat::from_file(&proc_file)?.state() {
+                return Ok(());
+            }
+
+            std::thread::sleep(POLL_INTERVAL);
+            if Instant::now() > end {
+                return Err(StopProcessError::Timeout);
+            }
+        }
+    }
+
+    /// Send SIGCONT to the process to continue.
+    ///
+    /// Unlike `stop_process`, this function does not wait for the process to continue.
+    fn continue_process(&mut self) -> Result<(), ContinueProcessError> {
+        signal::kill(nix::unistd::Pid::from_raw(self.pid), Some(signal::SIGCONT))?;
+        Ok(())
+    }
+
     /// Parse /proc/$pid/task to list all the threads of the process identified by
     /// pid.
     fn enumerate_threads(&mut self) -> Result<(), InitError> {
@@ -334,8 +398,9 @@ impl PtraceDumper {
         let mut mapping = self.find_mapping(stack_pointer);
 
         // The guard page has been 1 MiB in size since kernel 4.12, older
-        // kernels used a 4 KiB one instead.
-        let guard_page_max_addr = stack_pointer + (1024 * 1024);
+        // kernels used a 4 KiB one instead. Note the saturating add, as 32-bit
+        // processes can have a stack pointer within 1MiB of usize::MAX
+        let guard_page_max_addr = stack_pointer.saturating_add(1024 * 1024);
 
         // If we found no mapping, or the mapping we found has no permissions
         // then we might have hit a guard page, try looking for a mapping in