Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Refactor - JIT noop insertion #584

Merged
merged 3 commits into from
Sep 5, 2024
Merged

Refactor - JIT noop insertion #584

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7064f2f
Uses the Uniform distribution type from rand for noop insertion inste…
Lichtso Sep 4, 2024
322e9cd
Makes the ExhaustedTextSegment check more conservative.
Lichtso Sep 4, 2024
ef1fe8d
Makes cargo fmt happy.
Lichtso Sep 4, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 10 additions & 4 deletions src/jit.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,11 @@ use rand::{thread_rng, Rng};
#[cfg(feature = "shuttle-test")]
use shuttle::rand::{thread_rng, Rng};

use rand::{rngs::SmallRng, SeedableRng};
use rand::{
distributions::{Distribution, Uniform},
rngs::SmallRng,
SeedableRng,
};
use std::{fmt::Debug, mem, ptr};

use crate::{
Expand Down Expand Up @@ -321,6 +325,7 @@ pub struct JitCompiler<'a, C: ContextObject> {
pc: usize,
last_instruction_meter_validation_pc: usize,
next_noop_insertion: u32,
noop_range: Uniform<u32>,
runtime_environment_key: i32,
diversification_rng: SmallRng,
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not related to this PR, but I found out that SmallRng isn't good when security is a concern:

Furthermore, SmallRng is not a good choice when:
Security against prediction is important. Use StdRng instead.

https://docs.rs/rand/latest/rand/rngs/struct.SmallRng.html

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The PRNG here needs to be fast, not high quality. The actual security comes from diversity of seeds in validators across the cluster.

stopwatch_is_active: bool,
Expand Down Expand Up @@ -372,6 +377,7 @@ impl<'a, C: ContextObject> JitCompiler<'a, C> {
pc: 0,
last_instruction_meter_validation_pc: 0,
next_noop_insertion: if config.noop_instruction_rate == 0 { u32::MAX } else { diversification_rng.gen_range(0..config.noop_instruction_rate * 2) },
noop_range: Uniform::new_inclusive(0, config.noop_instruction_rate * 2),
runtime_environment_key,
diversification_rng,
stopwatch_is_active: false,
Expand All @@ -393,7 +399,7 @@ impl<'a, C: ContextObject> JitCompiler<'a, C> {
self.emit_subroutines();

while self.pc * ebpf::INSN_SIZE < self.program.len() {
if self.offset_in_text_section + MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION > self.result.text_section.len() {
if self.offset_in_text_section + MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION * 2 >= self.result.text_section.len() {
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this overestimation is too high. We could decrease MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION to 90. Also, a nop instruction is a single byte. If we emit a maximum of (e.g.) 8 x86 instructions per SBPF instruction, we could do MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION + 8. In the worst case scenario, there will be one nop per x86 instruction.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is high yes, but it can only run into the end one sBPF instruction earlier. And that should be covered by MAX_EMPTY_PROGRAM_MACHINE_CODE_LENGTH. Generally I would be rather conservative and overestimate here as this is the only bounds-check for the emitted code size.

return Err(EbpfError::ExhaustedTextSegment(self.pc));
}
let mut insn = ebpf::get_insn_unchecked(self.program, self.pc);
Expand Down Expand Up @@ -723,7 +729,7 @@ impl<'a, C: ContextObject> JitCompiler<'a, C> {
}

// Bumper in case there was no final exit
if self.offset_in_text_section + MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION > self.result.text_section.len() {
if self.offset_in_text_section + MAX_MACHINE_CODE_LENGTH_PER_INSTRUCTION * 2 >= self.result.text_section.len() {
return Err(EbpfError::ExhaustedTextSegment(self.pc));
}
self.emit_validate_and_profile_instruction_count(true, Some(self.pc + 2));
Expand Down Expand Up @@ -786,7 +792,7 @@ impl<'a, C: ContextObject> JitCompiler<'a, C> {
pub fn emit_ins(&mut self, instruction: X86Instruction) {
instruction.emit(self);
if self.next_noop_insertion == 0 {
self.next_noop_insertion = self.diversification_rng.gen_range(0..self.config.noop_instruction_rate * 2);
self.next_noop_insertion = self.noop_range.sample(&mut self.diversification_rng);
// X86Instruction::noop().emit(self)?;
self.emit::<u8>(0x90);
} else {
Expand Down
6 changes: 4 additions & 2 deletions src/syscalls.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,8 +79,10 @@ declare_builtin_function!(
);

declare_builtin_function!(
/// Same as `void *memfrob(void *s, size_t n);` in `string.h` in C. See the GNU manual page (in
/// section 3) for `memfrob`. The memory is directly modified, and the syscall returns 0 in all
/// Same as `void *memfrob(void *s, size_t n);` in `string.h` in C.
///
/// See the GNU manual page (in section 3) for `memfrob`.
/// The memory is directly modified, and the syscall returns 0 in all
/// cases. Arguments 3 to 5 are unused.
SyscallMemFrob,
fn rust(
Expand Down
13 changes: 1 addition & 12 deletions src/verifier.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,18 +10,7 @@
// the MIT license <http://opensource.org/licenses/MIT>, at your option. This file may not be
// copied, modified, or distributed except according to those terms.

//! This “verifier” performs simple checks when the eBPF program is loaded into the VM (before it is
//! interpreted or JIT-compiled). It has nothing to do with the much more elaborated verifier inside
//! Linux kernel. There is no verification regarding the program flow control (should be a Direct
//! Acyclic Graph) or the consistency for registers usage (the verifier of the kernel assigns types
//! to the registers and is much stricter).
//!
//! On the other hand, rbpf is not expected to run in kernel space.
//!
//! Improving the verifier would be nice, but this is not trivial (and Linux kernel is under GPL
//! license, so we cannot copy it).
//!
//! Contrary to the verifier of the Linux kernel, this one does not modify the bytecode at all.
//! Verifies that the bytecode is valid for the given config.

use crate::{
ebpf,
Expand Down