pass-lang/src/format.c

/// This file handles concerns the executable file format.
/// This includes keeping track of the current virtual address,
/// performing relocations, and creating the executable file header.
///
/// The set of features we actually use is very small, so hopefully
/// this will turn out to be able to port this across executable formats
/// and architectures with relatively few modifications.
#include "format.h"
#include "io.h"

#include <assert.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

struct symbol {
    uint64_t vaddr;
};

#define MAX_SYMBOLS 65535
static uint32_t symbol_count = 0;
static struct symbol symbols[MAX_SYMBOLS];

enum relocation_type {
    REL_PC32 = 2,
    REL_GOTPCREL = 9,
    REL_SIZE32 = 32,
    REL_SIZE64 = 33,
};

struct relocation {
    enum relocation_type type;
    symbol symbol;
    uint64_t offset;
};

#define MAX_RELOCATIONS 65535
static uint32_t relocation_count = 0;
static struct relocation relocations[MAX_RELOCATIONS];

static uint64_t file_here = 0;

/// Not the size of the ELF header per se, but rather th ELF header
/// plus the program headers and section headers we include.
/// We reserve this much space at the beginning of every file
/// to fill in once the executable is finished.
#define ELF_HEADER_SIZE 0xb0

void elf_executable(void) {
    reserve(ELF_HEADER_SIZE);
    file_here += ELF_HEADER_SIZE;
}

void finish_executable(symbol entry_point) {
    uint64_t file_len = file_here;

    // Hardcoded ELF header for statically-linked position-independent executable.
    // Since we only support Linux amd64 static PIE, there's no need to abstract over this for now.
    uint8_t elf_header[ELF_HEADER_SIZE] = {
        // ELF header
        0x7F, 'E', 'L', 'F',          // ELF magic
        2, 1, 1, 3, 0,                // 64-bit little-endian Linux, ELF version 1
        0, 0, 0, 0, 0, 0, 0,          // padding
        3, 0, 0x3E, 0, 1, 0, 0, 0,    // dynamic executable, amd64, ELF version 1 again
        0, 0, 0, 0, 0, 0, 0, 0,       // PATCHME: entry point address
        0x40, 0, 0, 0, 0, 0, 0, 0,    // program header table offset (immediately after ELF)
        0, 0, 0, 0, 0, 0, 0, 0,       // section eader table offset (none)
        0, 0, 0, 0, 0x40, 0, 0x38, 0, // flags (none), header sizes
        2, 0, 0, 0, 0, 0, 0, 0,       // 2 segments, no sections

        // program header segment
        6, 0, 0, 0, 4, 0, 0, 0,       // program header segment, readable
        0x40, 0, 0, 0, 0, 0, 0, 0,    // immediately after ELF header
        0x40, 0, 0, 0, 0, 0, 0, 0,    // virtual address
        0, 0, 0, 0, 0, 0, 0, 0,       // physical address
        0x70, 0, 0, 0, 0, 0, 0, 0,    // size in file (2 * size of program header)
        0x70, 0, 0, 0, 0, 0, 0, 0,    // size in memory
        8, 0, 0, 0, 0, 0, 0, 0,       // alignment

        // executable segment
        1, 0, 0, 0, 5, 0, 0, 0,       // loadable segment, readable and executable
        0, 0, 0, 0, 0, 0, 0, 0,       // whole file
        0, 0, 0, 0, 0, 0, 0, 0,       // virtual address
        0, 0, 0, 0, 0, 0, 0, 0,       // physical address
        0, 0, 0, 0, 0, 0, 0, 0,       // PATCHME: size in file
        0, 0, 0, 0, 0, 0, 0, 0,       // PATCHME: size in memory
        0, 0x10, 0, 0, 0, 0, 0, 0,    // alignment (4K)
    };
    uint64_t ep = (uint64_t) symbols[entry_point].vaddr;
    uint64_t fl = (uint64_t) file_len;
    memcpy(&elf_header[0x18], &ep, sizeof(uint64_t));
    memcpy(&elf_header[0x98], &fl, sizeof(uint64_t));
    memcpy(&elf_header[0x98 + sizeof(uint64_t)], &fl, sizeof(uint64_t));

    patch(0, elf_header, ELF_HEADER_SIZE);

    for (uint32_t i = 0; i < relocation_count; i++) {
        struct relocation rel = relocations[i];
        assert(rel.type == REL_PC32);
        uint64_t vaddr = symbols[rel.symbol].vaddr;
        assert(vaddr != (uint64_t) -1);
        // PC-relative is from the *end* of the instruction,
        // and the displacement is 4 bytes (32 bits).
        int64_t disp = (int64_t) vaddr - ((int64_t) rel.offset + 4);
        assert(disp >= INT32_MIN && disp <= INT32_MAX);
        patch_u32(rel.offset, (int32_t) disp);
    }
}

symbol new_symbol(void) {
    struct symbol* sym = &symbols[symbol_count];
    sym->vaddr = (uint64_t) -1;
    if (symbol_count == MAX_SYMBOLS) {
        fprintf(stderr, "error: exceeded maximum number of symbols\n");
        exit(1);
    }
    return symbol_count++;
}

void define_executable_symbol(symbol s) {
    struct symbol* sym = &symbols[s];
    sym->vaddr = file_here;
}

void define_readonly_symbol(symbol sym) {
    // TODO:
    assert(0);
}

void append_data(size_t size, const void* buf) {
    file_here += size;
    emit(buf, size);
}

void append_u8(uint8_t x) {
    file_here += 1;
    emit_u8(x);
}

void append_u32(uint32_t x) {
    file_here += 4;
    emit_u32(x);
}

void append_u64(uint64_t x) {
    file_here += 8;
    emit_u64(x);
}

static struct relocation* new_relocation(void) {
    if (relocation_count == MAX_RELOCATIONS) {
        fprintf(stderr, "error: exceeded maximum number of relocations\n");
        exit(1);
    }
    struct relocation* rel = &relocations[relocation_count];
    relocation_count++;
    return rel;
}

void relocate_pc32(symbol sym) {
    int32_t offset = symbol_offset(sym, 4);
    if (offset != INT32_MAX) {
        append_u32((uint32_t) offset);
        return;
    }
    struct relocation* rel = new_relocation();
    rel->type = REL_PC32;
    rel->offset = file_here;
    rel->symbol = sym;
    append_u32(0);
}

int32_t symbol_offset(symbol sym, int8_t off) {
    uint64_t vaddr = symbols[sym].vaddr;
    if (vaddr == (uint64_t) -1) {
        return INT32_MAX;
    }
    int64_t disp = (int64_t) vaddr - ((int64_t) file_here + off);
    if (disp >= INT32_MAX || disp <= INT32_MIN) {
        return INT32_MAX;
    }
    return disp;
}

void relocate_gotpcrel(symbol sym) {
    // TODO
    assert(0);
}

void relocate_size32(symbol sym) {
    // TODO
    assert(0);
}

void relocate_size64(symbol sym) {
    // TODO
    assert(0);
}