Add RCU read bindings (#143) and each_process() iterator

Add an RcuReadGuard object for calling functions that expect to be
RCU-protected, and add a Rust equivalent of for_each_process and a very
light task_struct bindings sufficient to write a simple test/demo.

Author: geofft

build.rs

@@ -45,6 +45,8 @@ const INCLUDED_VARS: &[&str] = &[
     "SEEK_CUR",
     "SEEK_END",
     "O_NONBLOCK",
+    "init_task",
+    "TASK_COMM_LEN",
 ];
 const OPAQUE_TYPES: &[&str] = &[
     // These need to be opaque because they're both packed and aligned, which rustc

src/bindings_helper.h

@@ -2,6 +2,7 @@
 #include <linux/fs.h>
 #include <linux/module.h>
 #include <linux/random.h>
+#include <linux/sched/task.h>
 #include <linux/slab.h>
 #include <linux/uaccess.h>
 #include <linux/version.h>

src/helpers.c

@@ -1,7 +1,10 @@
 #include <linux/bug.h>
 #include <linux/printk.h>
+#include <linux/rcupdate.h>
 #include <linux/uaccess.h>
 #include <linux/version.h>
+#include <linux/sched/signal.h>
+#include <linux/sched/task.h>
 
 
 int printk_helper(const unsigned char *s, int len)
@@ -23,6 +26,26 @@ int access_ok_helper(const void __user *addr, unsigned long n)
 #endif
 }
 
+void rcu_read_lock_helper(void) {
+    rcu_read_lock();
+}
+
+void rcu_read_unlock_helper(void) {
+    rcu_read_unlock();
+}
+
+struct task_struct *next_task_helper(struct task_struct *p) {
+    return next_task(p);
+}
+
+void task_lock_helper(struct task_struct *p) {
+    return task_lock(p);
+}
+
+void task_unlock_helper(struct task_struct *p) {
+    return task_unlock(p);
+}
+
 /* see https://github.com/rust-lang/rust-bindgen/issues/1671 */
 _Static_assert(__builtin_types_compatible_p(size_t, uintptr_t),
                "size_t must match uintptr_t, what architecture is this??");

src/lib.rs

@@ -15,6 +15,8 @@ pub mod filesystem;
 pub mod printk;
 #[cfg(kernel_4_13_0_or_greater)]
 pub mod random;
+pub mod rcu;
+pub mod sched;
 pub mod sysctl;
 mod types;
 pub mod user_ptr;

src/rcu.rs

@@ -0,0 +1,42 @@
+//! Bindings to RCU (read-copy-update), a high-performance lockless
+//! synchronization system used by many kernel data structures. At the
+//! moment, only calling functions that perform RCU reads is supported.
+
+extern "C" {
+    fn rcu_read_lock_helper();
+    fn rcu_read_unlock_helper();
+}
+
+/// A guard representing an RCU read-side critical section. Its
+/// constructor calls `rcu_read_lock()` and its destructor calls
+/// `rcu_read_unlock()`.
+///
+/// Within a read-side critical section (i.e., while at least one
+/// RcuReadGuard object is instantiated), objects behind RCU-protected
+/// pointers are guaranteed not to change, and so reading from them
+/// (after gaining a pointer with `rcu_dereference()`) is safe.
+///
+/// It is an error (risk of deadlock, but not memory unsafety) to block
+/// or schedule while holding an RcuReadGuard. It is also an error
+/// (guaranteed deadlock) to call `synchronize_rcu()` while holding an
+/// RcuReadGuard. Holding multiple guards (i.e., nesting read-side
+/// critical sections) is safe.
+pub struct RcuReadGuard(());
+
+#[allow(clippy::new_without_default)]
+impl RcuReadGuard {
+    pub fn new() -> Self {
+        unsafe {
+            rcu_read_lock_helper();
+        }
+        RcuReadGuard(())
+    }
+}
+
+impl Drop for RcuReadGuard {
+    fn drop(&mut self) {
+        unsafe {
+            rcu_read_unlock_helper();
+        }
+    }
+}

src/sched.rs

@@ -0,0 +1,97 @@
+//! APIs for interacting with the scheduler and with processes,
+//! corresponding to <linux/sched.h> and related header files.
+#![allow(improper_ctypes)]
+
+use core::ptr;
+
+use crate::bindings;
+use crate::rcu;
+
+extern "C" {
+    fn task_lock_helper(p: *mut bindings::task_struct);
+    fn task_unlock_helper(p: *mut bindings::task_struct);
+    fn next_task_helper(p: *mut bindings::task_struct) -> *mut bindings::task_struct;
+}
+
+/// Represents a `struct task_struct *`.
+pub struct TaskStruct<'a>(&'a mut bindings::task_struct);
+
+impl TaskStruct<'_> {
+    /// Returns the threadgroup ID (what userspace calls the process ID).
+    pub fn tgid(&self) -> i32 {
+        self.0.tgid
+    }
+
+    /// Returns the command name / process title. This is a short name,
+    /// typically the base name of the command, and does not have the
+    /// full path or arguments. It's a fixed-sized set of bytes, but by
+    /// convention it's interpreted as NUL-terminated.
+    pub fn comm(&mut self) -> [u8; bindings::TASK_COMM_LEN as usize] {
+        let mut result = [0u8; bindings::TASK_COMM_LEN as usize];
+        unsafe {
+            task_lock_helper(self.0);
+        }
+        // if only char were unsigned char
+        let mut it = result.iter_mut();
+        for byte in &self.0.comm {
+            if *byte == 0 {
+                break;
+            }
+            match it.next() {
+                Some(lvalue) => {
+                    *lvalue = *byte as _;
+                }
+                None => {
+                    break;
+                }
+            }
+        }
+        unsafe {
+            task_unlock_helper(self.0);
+        }
+        result
+    }
+}
+
+/// Iterate over every process on the system. Returns only processes,
+/// i.e., thread group leaders.
+///
+/// ```
+/// let g = rcu::RcuReadGuard::new();
+/// for p in each_process(&g) {
+///     println!("{:?}", p.comm());
+/// }
+/// ```
+pub struct EachProcess<'g> {
+    p: *mut bindings::task_struct,
+    _g: &'g rcu::RcuReadGuard,
+}
+
+pub fn each_process(g: &rcu::RcuReadGuard) -> EachProcess {
+    // unsafe is bogus here because we don't read it
+    // https://github.com/rust-lang/rust/issues/74843
+    EachProcess {
+        p: unsafe { &mut bindings::init_task },
+        _g: g,
+    }
+}
+
+impl<'g> Iterator for EachProcess<'g> {
+    type Item = TaskStruct<'g>;
+
+    fn next(&mut self) -> Option<TaskStruct<'g>> {
+        // Safety:
+        // - oldp is valid if not null, because it is either &init_task
+        //   (a static location) or updated by this function.
+        // - next_task calls rcu_dereference internally, which is safe
+        //   because we hold self._g.
+        // - The returned reference has lifetime 'g, which is valid
+        //   because self._g lives at least that long.
+        let oldp = unsafe { self.p.as_mut()? };
+        self.p = unsafe { next_task_helper(self.p) };
+        if self.p == unsafe { &mut bindings::init_task } {
+            self.p = ptr::null_mut();
+        }
+        Some(TaskStruct(oldp))
+    }
+}

tests/for-each-process/Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "for-each-process-tests"
+version = "0.1.0"
+authors = ["Alex Gaynor <[email protected]>", "Geoffrey Thomas <[email protected]>"]
+edition = "2018"
+
+[lib]
+crate-type = ["staticlib"]
+test = false
+
+[features]
+default = ["linux-kernel-module"]
+
+[dependencies]
+linux-kernel-module = { path = "../..", optional = true }
+
+[dev-dependencies]
+kernel-module-testlib = { path = "../../testlib" }
+libc = "0.2.58"
+tempfile = "3"

tests/for-each-process/src/lib.rs

@@ -0,0 +1,28 @@
+#![no_std]
+
+use linux_kernel_module::{self, println, rcu, sched};
+
+struct ForEachProcessTestModule;
+
+impl linux_kernel_module::KernelModule for ForEachProcessTestModule {
+    fn init() -> linux_kernel_module::KernelResult<Self> {
+        let g = rcu::RcuReadGuard::new();
+        for mut p in sched::each_process(&g) {
+            let comm = p.comm();
+            let comm_until_nul = comm.split(|c| *c == 0).next().unwrap();
+            println!(
+                "for-each-process: {:8} {}",
+                p.tgid(),
+                core::str::from_utf8(comm_until_nul).unwrap_or("[invalid UTF-8]")
+            );
+        }
+        Ok(ForEachProcessTestModule)
+    }
+}
+
+linux_kernel_module::kernel_module!(
+    ForEachProcessTestModule,
+    author: b"Fish in a Barrel Contributors",
+    description: b"A module for testing EachProcess",
+    license: b"GPL"
+);

tests/for-each-process/tests/tests.rs

@@ -0,0 +1,12 @@
+use std::fs::File;
+use std::io::Write;
+
+use kernel_module_testlib::{assert_dmesg_contains, with_kernel_module};
+
+#[test]
+fn test_for_each_process() {
+    File::create("/proc/self/comm").unwrap().write_all(b"areyouthere").unwrap();
+    with_kernel_module(|| {
+        assert_dmesg_contains(&[b"areyouthere"]);
+    });
+}