内容简介:So IIt included this snippet of Rust code:
Tags: rust
So I tweeted :
This happened in an instant. I am not even sorry.
It included this snippet of Rust code:
use std::os::raw::{c_int, c_void}; use std::sync::atomic::{AtomicU64, Ordering::SeqCst}; use std::time::Instant; #[repr(C)] struct timespec { tv_sec: u64, tv_nsec: u64, } static TIME_COUNTER: AtomicU64 = AtomicU64::new(0); #[no_mangle] extern "C" fn clock_gettime(clk_id: *mut c_void, tp: *mut timespec) -> c_int { unsafe { let next_tick = TIME_COUNTER.fetch_add(1, SeqCst); (*tp).tv_sec = next_tick; (*tp).tv_nsec = 0; 0 } } fn main() { let a = Instant::now(); let b = Instant::now(); let diff = b.duration_since(a); println!("{:?}", diff); }
( Playground link )
And I was then asked what that piece of code does:
@varjmes: what am i seeing here? started reading the rust book yesterday :)
( Tweet )
So let me explain.
First let's take this from a tweet-worthy piece of code to be more production ready with cross-platform support. I'm on macOS most of the day. Let's add support for it:
#[cfg(all(unix, not(target_os = "macos")))] use std::os::raw::{c_int, c_void}; use std::sync::atomic::{AtomicU64, Ordering::SeqCst}; use std::time::Instant; #[cfg(all(unix, not(target_os = "macos")))] #[repr(C)] struct timespec { tv_sec: u64, tv_nsec: u64, } static TIME_COUNTER: AtomicU64 = AtomicU64::new(0); #[cfg(all(unix, not(target_os = "macos")))] #[no_mangle] extern "C" fn clock_gettime(_clk_id: *mut c_void, tp: *mut timespec) -> c_int { unsafe { let next_tick = TIME_COUNTER.fetch_add(1, SeqCst); (*tp).tv_sec = next_tick; (*tp).tv_nsec = 0; 0 } } #[cfg(target_os = "macos")] #[no_mangle] extern "C" fn mach_absolute_time() -> u64 { const NANOSECONDS_IN_A_SECOND: u64 = 1_000_000_000; TIME_COUNTER.fetch_add(1, SeqCst) * NANOSECONDS_IN_A_SECOND } fn main() { let a = Instant::now(); let b = Instant::now(); let diff = b.duration_since(a); println!("{:?}", diff); }
We can compile that:
rustc instant.rs
And then run it:
> ./instant 1s
What is happening here?
Time is an illusion
Computers provide clock sources that allow a program to make assumptions about time passed.
If all you need is measure the time passed between some instant a
and instant b
the Rust libstd provides you with Instant
, a measurement of a monotonically nondecreasing clock
(I'm sorry to inform you that guarantees about monotonically increasing clocks are lacking
). Instant
s are an opaque thing and only good for getting you the difference between two of them, resulting in a Duration
.
Good enough if you can rely on the operating system to not lie to you.
How is Instant
implemented?
That depends on the operating system your running your code on.
Luckily the documentation lists the (currently) used system calls to get time information.
System Call | |
---|---|
UNIX |
clock_gettime
(Monotonic Clock) |
Darwin |
(: Trimmed down to the platforms I work on.)
So that's where Rust gets its time from.
The operating system is an illusion
But how do we replace operating system functionality? Turns out functions are merely a name and the linker tries to figure out where the code behind them is and then points to that.
If we look at a simpler version of our code without any #[no_mangle] extern "C"
functions we can see where these come from:
(macOS) ❯ nm -a ./instant | rg mach_absolute_time U _mach_absolute_time (linux) ❯ nm -a instant | rg clock_gettime U clock_gettime@@GLIBC_2.17
nm
lists symbols from object files.
That U
right there stands for undefined
: The binary doesn't have knowledge where it comes from
(though in the case of Linux it gave it a slightly different name letting us guess what to expect).
Details about this symbol are later filled in, once your program is loaded and symbols to dynamic libaries are resolved
(and I'm sure the "Making our own executable packer" series
by @fasterthanlime
will have lots more details on it, I should go read it).
What if we do that work before even running the program?
Strings are strings, no matter what order
So now let's try to expose a function under the same name from our code.
Note: I'm only doing the macOS part here, it works the same for Linux.
If we start with the plain function like this:
fn mach_absolute_time() -> u64 { 1 }
it will not turn up in the final binary at all and rustc
right fully complains:
warning: function is never used: `mach_absolute_time`
Even making it pub
won't work.
By default symbols such as function names get mangled
; some additional information is encoded into the name as well and at the same time this ensures uniqueness of names
(see also Name Mangling
).
One can disable that in Rust using the #[no_mangle]
attribute. So we apply that.
We're trying to override a function that is defined in terms of C.
C has slightly different calling conventions than Rust.
It's part of their respective ABIs: how data is passed in and out of functions.
In Rust one can define the used ABI with the extern "name"
tag.
So we add that.
And thus we end up with
#[no_mangle] extern "C" fn mach_absolute_time() -> u64 { 1 }
If we compile our code again and look at the symbols we get this:
(macOS) ❯ nm -a ./instant | rg mach_absolute_time 0000000100000c20 T _mach_absolute_time (linux) ❯ nm -a instant | rg clock_gettime 0000000000005250 T clock_gettime
Now we know that the symbol is defined somewhere in the binary ("T - The symbol is in the text (code) section.") and also the location ( 0000000100000c20 / 0000000000005250 ).
If this program is run it will still need to resolve undefined symbols ... but this time our functions are not undefined anymore. As the stdlib just calls whatever is behind that respective name it now calls our code instead!
And yes, this works for pretty much all the functions defined by other libraries or libc:
use std::fs::File; use std::os::raw::{c_char, c_int}; #[no_mangle] extern "C" fn open(_path: *mut c_char, _flag: c_int) -> c_int { -1 } fn main() { File::open("/etc/passwd").unwrap(); }
❯ rustc file.rs ❯ ./file thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Os { code: 0, kind: Other, message: "Undefined error: 0" }', file.rs:10:5 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Please don't do this.
If you liked this you might also like STDSHOUT!!!1! , I even gave a talk about it .
以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持 码农网
猜你喜欢:本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们。
Java程序员修炼之道
[英] Benjamin J. Evans、[荷兰] Martijn Verburg / 吴海星 / 人民邮电出版社 / 2013-7 / 89.00元
本书分为四部分,第一部分全面介绍Java 7 的新特性,第二部分探讨Java 关键编程知识和技术,第三部分讨论JVM 上的新语言和多语言编程,第四部分将平台和多语言编程知识付诸实践。从介绍Java 7 的新特性入手,本书涵盖了Java 开发中最重要的技术,比如依赖注入、测试驱动的开发和持续集成,探索了JVM 上的非Java 语言,并详细讲解了多语言项目, 特别是涉及Groovy、Scala 和Cl......一起来看看 《Java程序员修炼之道》 这本书的介绍吧!
Markdown 在线编辑器
Markdown 在线编辑器
RGB CMYK 转换工具
RGB CMYK 互转工具