|
| 1 | +// Copyright 2017 The Rust Project Developers. See the COPYRIGHT |
| 2 | +// file at the top-level directory of this distribution and at |
| 3 | +// http://rust-lang.org/COPYRIGHT. |
| 4 | +// |
| 5 | +// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
| 6 | +// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
| 7 | +// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your |
| 8 | +// option. This file may not be copied, modified, or distributed |
| 9 | +// except according to those terms. |
| 10 | + |
| 11 | +//! Management of the encoding of LLVM bytecode into rlibs |
| 12 | +//! |
| 13 | +//! This module contains the management of encoding LLVM bytecode into rlibs, |
| 14 | +//! primarily for the usage in LTO situations. Currently the compiler will |
| 15 | +//! unconditionally encode LLVM-IR into rlibs regardless of what's happening |
| 16 | +//! elsewhere, so we currently compress the bytecode via deflate to avoid taking |
| 17 | +//! up too much space on disk. |
| 18 | +//! |
| 19 | +//! After compressing the bytecode we then have the rest of the format to |
| 20 | +//! basically deal with various bugs in various archive implementations. The |
| 21 | +//! format currently is: |
| 22 | +//! |
| 23 | +//! RLIB LLVM-BYTECODE OBJECT LAYOUT |
| 24 | +//! Version 2 |
| 25 | +//! Bytes Data |
| 26 | +//! 0..10 "RUST_OBJECT" encoded in ASCII |
| 27 | +//! 11..14 format version as little-endian u32 |
| 28 | +//! 15..19 the length of the module identifier string |
| 29 | +//! 20..n the module identifier string |
| 30 | +//! n..n+8 size in bytes of deflate compressed LLVM bitcode as |
| 31 | +//! little-endian u64 |
| 32 | +//! n+9.. compressed LLVM bitcode |
| 33 | +//! ? maybe a byte to make this whole thing even length |
| 34 | +
|
| 35 | +use std::io::{Read, Write}; |
| 36 | +use std::ptr; |
| 37 | +use std::str; |
| 38 | + |
| 39 | +use flate2::Compression; |
| 40 | +use flate2::read::DeflateDecoder; |
| 41 | +use flate2::write::DeflateEncoder; |
| 42 | + |
| 43 | +// This is the "magic number" expected at the beginning of a LLVM bytecode |
| 44 | +// object in an rlib. |
| 45 | +pub const RLIB_BYTECODE_OBJECT_MAGIC: &'static [u8] = b"RUST_OBJECT"; |
| 46 | + |
| 47 | +// The version number this compiler will write to bytecode objects in rlibs |
| 48 | +pub const RLIB_BYTECODE_OBJECT_VERSION: u8 = 2; |
| 49 | + |
| 50 | +pub const RLIB_BYTECODE_EXTENSION: &str = "bytecode.encoded"; |
| 51 | + |
| 52 | +pub fn encode(identifier: &str, bytecode: &[u8]) -> Vec<u8> { |
| 53 | + let mut encoded = Vec::new(); |
| 54 | + |
| 55 | + // Start off with the magic string |
| 56 | + encoded.extend_from_slice(RLIB_BYTECODE_OBJECT_MAGIC); |
| 57 | + |
| 58 | + // Next up is the version |
| 59 | + encoded.extend_from_slice(&[RLIB_BYTECODE_OBJECT_VERSION, 0, 0, 0]); |
| 60 | + |
| 61 | + // Next is the LLVM module identifier length + contents |
| 62 | + let identifier_len = identifier.len(); |
| 63 | + encoded.extend_from_slice(&[ |
| 64 | + (identifier_len >> 0) as u8, |
| 65 | + (identifier_len >> 8) as u8, |
| 66 | + (identifier_len >> 16) as u8, |
| 67 | + (identifier_len >> 24) as u8, |
| 68 | + ]); |
| 69 | + encoded.extend_from_slice(identifier.as_bytes()); |
| 70 | + |
| 71 | + // Next is the LLVM module deflate compressed, prefixed with its length. We |
| 72 | + // don't know its length yet, so fill in 0s |
| 73 | + let deflated_size_pos = encoded.len(); |
| 74 | + encoded.extend_from_slice(&[0, 0, 0, 0, 0, 0, 0, 0]); |
| 75 | + |
| 76 | + let before = encoded.len(); |
| 77 | + DeflateEncoder::new(&mut encoded, Compression::Fast) |
| 78 | + .write_all(bytecode) |
| 79 | + .unwrap(); |
| 80 | + let after = encoded.len(); |
| 81 | + |
| 82 | + // Fill in the length we reserved space for before |
| 83 | + let bytecode_len = (after - before) as u64; |
| 84 | + encoded[deflated_size_pos + 0] = (bytecode_len >> 0) as u8; |
| 85 | + encoded[deflated_size_pos + 1] = (bytecode_len >> 8) as u8; |
| 86 | + encoded[deflated_size_pos + 2] = (bytecode_len >> 16) as u8; |
| 87 | + encoded[deflated_size_pos + 3] = (bytecode_len >> 24) as u8; |
| 88 | + encoded[deflated_size_pos + 4] = (bytecode_len >> 32) as u8; |
| 89 | + encoded[deflated_size_pos + 5] = (bytecode_len >> 40) as u8; |
| 90 | + encoded[deflated_size_pos + 6] = (bytecode_len >> 48) as u8; |
| 91 | + encoded[deflated_size_pos + 7] = (bytecode_len >> 56) as u8; |
| 92 | + |
| 93 | + // If the number of bytes written to the object so far is odd, add a |
| 94 | + // padding byte to make it even. This works around a crash bug in LLDB |
| 95 | + // (see issue #15950) |
| 96 | + if encoded.len() % 2 == 1 { |
| 97 | + encoded.push(0); |
| 98 | + } |
| 99 | + |
| 100 | + return encoded |
| 101 | +} |
| 102 | + |
| 103 | +pub struct DecodedBytecode<'a> { |
| 104 | + identifier: &'a str, |
| 105 | + encoded_bytecode: &'a [u8], |
| 106 | +} |
| 107 | + |
| 108 | +impl<'a> DecodedBytecode<'a> { |
| 109 | + pub fn new(data: &'a [u8]) -> Result<DecodedBytecode<'a>, String> { |
| 110 | + if !data.starts_with(RLIB_BYTECODE_OBJECT_MAGIC) { |
| 111 | + return Err(format!("magic bytecode prefix not found")) |
| 112 | + } |
| 113 | + let data = &data[RLIB_BYTECODE_OBJECT_MAGIC.len()..]; |
| 114 | + if !data.starts_with(&[RLIB_BYTECODE_OBJECT_VERSION, 0, 0, 0]) { |
| 115 | + return Err(format!("wrong version prefix found in bytecode")) |
| 116 | + } |
| 117 | + let data = &data[4..]; |
| 118 | + if data.len() < 4 { |
| 119 | + return Err(format!("bytecode corrupted")) |
| 120 | + } |
| 121 | + let identifier_len = unsafe { |
| 122 | + u32::from_le(ptr::read_unaligned(data.as_ptr() as *const u32)) as usize |
| 123 | + }; |
| 124 | + let data = &data[4..]; |
| 125 | + if data.len() < identifier_len { |
| 126 | + return Err(format!("bytecode corrupted")) |
| 127 | + } |
| 128 | + let identifier = match str::from_utf8(&data[..identifier_len]) { |
| 129 | + Ok(s) => s, |
| 130 | + Err(_) => return Err(format!("bytecode corrupted")) |
| 131 | + }; |
| 132 | + let data = &data[identifier_len..]; |
| 133 | + if data.len() < 8 { |
| 134 | + return Err(format!("bytecode corrupted")) |
| 135 | + } |
| 136 | + let bytecode_len = unsafe { |
| 137 | + u64::from_le(ptr::read_unaligned(data.as_ptr() as *const u64)) as usize |
| 138 | + }; |
| 139 | + let data = &data[8..]; |
| 140 | + if data.len() < bytecode_len { |
| 141 | + return Err(format!("bytecode corrupted")) |
| 142 | + } |
| 143 | + let encoded_bytecode = &data[..bytecode_len]; |
| 144 | + |
| 145 | + Ok(DecodedBytecode { |
| 146 | + identifier, |
| 147 | + encoded_bytecode, |
| 148 | + }) |
| 149 | + } |
| 150 | + |
| 151 | + pub fn bytecode(&self) -> Vec<u8> { |
| 152 | + let mut data = Vec::new(); |
| 153 | + DeflateDecoder::new(self.encoded_bytecode).read_to_end(&mut data).unwrap(); |
| 154 | + return data |
| 155 | + } |
| 156 | + |
| 157 | + pub fn identifier(&self) -> &'a str { |
| 158 | + self.identifier |
| 159 | + } |
| 160 | +} |
0 commit comments