scx_rustland_core: Forbid mmap() syscall

.github/include/flake.nix

@@ -95,6 +95,7 @@
                   gnused
                   isort
                   jq
+                  libseccomp.lib
                   llvmPackages.libclang
                   llvmPackages.libllvm
                   pkg-config
@@ -137,6 +138,7 @@
                       elfutils.out
                       zlib
                       zstd.out
+                      libseccomp.lib
                     ]))) + "'")
                   ]
                 ];

rust/scx_rustland_core/Cargo.toml

@@ -12,6 +12,7 @@ anyhow = "1.0.65"
 plain = "0.2.3"
 libbpf-rs = "=0.25.0-beta.1"
 libc = "0.2.137"
+seccomp = "0.1"
 scx_utils = { path = "../scx_utils", version = "1.0.14" }
 
 [build-dependencies]

rust/scx_rustland_core/assets/bpf.rs

@@ -205,10 +205,6 @@ impl<'cb> BpfScheduler<'cb> {
         skel_builder.obj_builder.debug(debug);
         let mut skel = scx_ops_open!(skel_builder, open_object, rustland)?;
 
-        // Lock all the memory to prevent page faults that could trigger potential deadlocks during
-        // scheduling.
-        ALLOCATOR.lock_memory();
-
         // Copy one item from the ring buffer.
         //
         // # Safety
@@ -282,6 +278,11 @@ impl<'cb> BpfScheduler<'cb> {
         let dispatched = libbpf_rs::UserRingBuffer::new(&maps.dispatched)
             .expect("failed to create user ringbuf");
 
+        // Lock all the memory to prevent page faults that could trigger potential deadlocks during
+        // scheduling.
+        ALLOCATOR.lock_memory();
+        ALLOCATOR.disable_mmap().expect("Failed to disable mmap");
+
         // Make sure to use the SCHED_EXT class at least for the scheduler itself.
         match Self::use_sched_ext() {
             0 => Ok(Self {

rust/scx_rustland_core/src/alloc.rs

@@ -26,6 +26,7 @@
 // This software may be used and distributed according to the terms of the
 // GNU General Public License version 2.
 
+use seccomp::*;
 use std::alloc::{GlobalAlloc, Layout};
 use std::fs::File;
 use std::io::{BufRead, BufReader, Write};
@@ -456,6 +457,24 @@ impl UserAllocator {
         }
     }
 
+    // Enable a seccomp filter that sends a SIGSYS when mmap() is called.
+    #[allow(static_mut_refs)]
+    pub fn disable_mmap(&self) -> Result<(), Box<dyn std::error::Error>> {
+        let mut ctx = seccomp::Context::default(Action::Allow)?;
+        let syscall_nr = libc::SYS_mmap as usize;
+        let cmp = Compare::arg(0)
+            .using(Op::MaskedEq)
+            .with(0)
+            .build()
+            .ok_or("Failed to build seccomp filter")?;
+
+        let rule = Rule::new(syscall_nr, cmp, Action::Errno(libc::EPERM));
+        ctx.add_rule(rule)?;
+        ctx.load()?;
+
+        Ok(())
+    }
+
     #[allow(static_mut_refs)]
     pub fn lock_memory(&self) {
         unsafe {

scheds/rust/scx_rustland/src/main.rs

@@ -219,6 +219,8 @@ struct Scheduler<'a> {
 
 impl<'a> Scheduler<'a> {
     fn init(opts: &Opts, open_object: &'a mut MaybeUninit<OpenObject>) -> Result<Self> {
+        let stats_server = StatsServer::new(stats::server_data()).launch()?;
+
         // Low-level BPF connector.
         let bpf = BpfScheduler::init(
             open_object,
@@ -227,7 +229,6 @@ impl<'a> Scheduler<'a> {
             opts.verbose,
             true, // Enable built-in idle CPU selection policy
         )?;
-        let stats_server = StatsServer::new(stats::server_data()).launch()?;
 
         info!("{} scheduler attached", SCHEDULER_NAME);