mirrors/reverie
1
0
Fork 0
mirror of https://github.com/facebookexperimental/reverie.git synced 2025-01-22 21:04:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make sure that VDSO patch code is properly aligned (#22)

Browse source 
Summary:
The code that is injected into the tracee processes' VDSO needs to be 8 byte aligned in order to satisfy conditions imposed by the ptrace interface on 64 bit architectures. There have been reproducible reports of that not always being the case:
https://github.com/facebookexperimental/hermit/issues/41. Use an explicitly aligned data structure to enforce this.

Pull Request resolved: https://github.com/facebookexperimental/reverie/pull/22

Reviewed By: VladimirMakaev

Differential Revision: D49195126

Pulled By: jasonwhite

fbshipit-source-id: 8e21cd4306ee2bf75ec321b21e40137be5292f9f
This commit is contained in:
Bjoern Doebel 2023-09-12 12:48:51 -07:00 committed by Facebook GitHub Bot
parent e3c97826ef
commit 9b5480352e
1 changed files with 43 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

61
reverie-ptrace/src/vdso.rs
View file

@ -21,41 +21,56 @@ use reverie::Guest;
use reverie::Tool;
use tracing::debug;
#[repr(align(64))]
struct BufferAligned<const N: usize>([u8; N]);
// Byte code for the new pseudo vdso functions which do the actual syscalls.
// Note: the byte code must be 8 bytes aligned
#[cfg(target_arch = "x86_64")]
mod vdso_syms {
#![allow(non_upper_case_globals)]
pub const time: &[u8; 8] = &[
use crate::vdso::BufferAligned;
const time_code: BufferAligned<8> = BufferAligned::<8>([
0xb8, 0xc9, 0x00, 0x00, 0x00, // mov %SYS_time, %eax
0x0f, 0x05, // syscall
0xc3, // retq
];
]);
pub const clock_gettime: &[u8; 8] = &[
pub const time: &[u8; 8] = &time_code.0;
const clock_gettime_code: BufferAligned<8> = BufferAligned::<8>([
0xb8, 0xe4, 0x00, 0x00, 0x00, // mov SYS_clock_gettime, %eax
0x0f, 0x05, // syscall
0xc3, // retq
];
]);
pub const getcpu: &[u8; 8] = &[
pub const clock_gettime: &[u8; 8] = &clock_gettime_code.0;
const getcpu_code: BufferAligned<8> = BufferAligned::<8>([
0xb8, 0x35, 0x01, 0x00, 0x00, // mov SYS_getcpu, %eax
0x0f, 0x05, // syscall
0xc3, // retq
];
]);
pub const gettimeofday: &[u8; 8] = &[
pub const getcpu: &[u8; 8] = &getcpu_code.0;
const gettimeofday_code: BufferAligned<8> = BufferAligned::<8>([
0xb8, 0x60, 0x00, 0x00, 0x00, // mov SYS_gettimeofday, %eax
0x0f, 0x05, // syscall
0xc3, // retq
];
]);
pub const clock_getres: &[u8; 8] = &[
pub const gettimeofday: &[u8; 8] = &gettimeofday_code.0;
const clock_getres_code: BufferAligned<8> = BufferAligned::<8>([
0xb8, 0xe5, 0x00, 0x00, 0x00, // mov SYS_clock_getres, %eax
0x0f, 0x05, // syscall
0xc3, // retq
];
]);
pub const clock_getres: &[u8; 8] = &clock_getres_code.0;
}
#[cfg(target_arch = "aarch64")]
@ -87,26 +102,34 @@ mod vdso_syms {
// See also
// https://developer.arm.com/documentation/ddi0596/2021-06/Base-Instructions/BTI--Branch-Target-Identification-
pub const clock_getres: &[u8; 16] = &[
use crate::vdso::BufferAligned;
const clock_getres_code: BufferAligned<16> = BufferAligned::<16>([
0x5f, 0x24, 0x03, 0xd5, // bti c
0x48, 0x0e, 0x80, 0xd2, // mov x8, 114 (#__NR_clock_getres)
0x01, 0x00, 0x00, 0xd4, // svc 0
0xc0, 0x03, 0x5f, 0xd6, // ret
];
]);
pub const clock_gettime: &[u8; 16] = &[
pub const clock_getres: &[u8; 16] = &clock_getres_code.0;
const clock_gettime_code: BufferAligned<16> = BufferAligned::<16>([
0x5f, 0x24, 0x03, 0xd5, // bti c
0x28, 0x0e, 0x80, 0xd2, // mov x8, 113 (#__NR_clock_gettime)
0x01, 0x00, 0x00, 0xd4, // svc 0
0xc0, 0x03, 0x5f, 0xd6, // ret
];
]);
pub const gettimeofday: &[u8; 16] = &[
pub const clock_gettime: &[u8; 16] = &clock_gettime_code.0;
const gettimeofday_code: BufferAligned<16> = BufferAligned::<16>([
0x5f, 0x24, 0x03, 0xd5, // bti c
0x28, 0x15, 0x80, 0xd2, // mov x8, 169 (#__NR_gettimeofday)
0x01, 0x00, 0x00, 0xd4, // svc 0
0xc0, 0x03, 0x5f, 0xd6, // ret
];
]);
pub const gettimeofday: &[u8; 16] = &gettimeofday_code.0;
// On aarch64, the vdso version of rt_sigreturn is only 8 bytes, so our
// patch can't exceed that size. However, since this syscall doesn't return,
@ -115,10 +138,12 @@ mod vdso_syms {
// NOTE: This is currently *exactly* how the kernel implements the
// rt_sigreturn vdso, so we could probably get away with not even patching
// it. See also `linux/arch/arm64/kernel/vdso/sigreturn.S`.
pub const rt_sigreturn: &[u8; 8] = &[
const rt_sigreturn_code: BufferAligned<8> = BufferAligned::<8>([
0x68, 0x11, 0x80, 0xd2, // mov x8, 139 (#__NR_rt_sigreturn)
0x01, 0x00, 0x00, 0xd4, // svc 0
];
]);
pub const rt_sigreturn: &[u8; 8] = &rt_sigreturn_code.0;
}
#[cfg(target_arch = "x86_64")]