use std::collections::{HashMap, HashSet};
use elf::{endian::LittleEndian, ElfBytes};
use crate::error::LinkerError;
const ET_REL: u16 = 1;
const EM_BPF: u16 = 247;
const SHT_PROGBITS: u32 = 1;
const SHT_REL: u32 = 9;
const SHF_ALLOC: u64 = 1 << 1;
const SHF_EXECINSTR: u64 = 1 << 2;
const R_BPF_64_64: u32 = 1;
const R_BPF_64_32: u32 = 10;
const EBPF_OP_CALL: u8 = 0x05u8 | 0x80u8;
const EBPF_OP_LDDW: u8 = 0x18;
#[derive(Copy, Clone, Debug)]
struct EbpfInsn {
opcode: u8,
dst: u8,
src: u8,
offset: i16,
imm: i32,
}
impl EbpfInsn {
fn from_u64(insn: u64) -> Self {
Self {
opcode: (insn & 0xFF) as u8,
dst: ((insn >> 8) & 0xF) as u8,
src: ((insn >> 12) & 0xF) as u8,
offset: ((insn >> 16) & 0xFFFF) as i16,
imm: (insn >> 32) as i32,
}
}
fn to_u64(&self) -> u64 {
(self.opcode as u64)
| ((self.dst as u64) << 8)
| ((self.src as u64) << 12)
| ((self.offset as u16 as u64) << 16)
| ((self.imm as u32 as u64) << 32)
}
}
pub fn link_elf(
input: &mut [u8],
vbase: usize,
ext_func_table: &HashMap<&str, i32>,
) -> Result<HashMap<String, (usize, usize)>, LinkerError> {
let elf = ElfBytes::<LittleEndian>::minimal_parse(input)?;
if elf.ehdr.class != elf::file::Class::ELF64
|| elf.ehdr.version != 1
|| elf.ehdr.osabi != 0
|| elf.ehdr.e_type != ET_REL
|| elf.ehdr.e_machine != EM_BPF
{
return Err(LinkerError::InvalidElf("invalid ELF header"));
}
let (Some(sht), Some(sht_strtab)) = elf.section_headers_with_strtab()? else {
return Err(LinkerError::InvalidElf("missing section headers"));
};
let Some(symtab) = elf.symbol_table()? else {
return Err(LinkerError::InvalidElf("missing symbol table"));
};
let mut code_sections: HashMap<String, (usize, usize)> = HashMap::new();
let mut code_section_indexes: HashSet<usize> = HashSet::new();
for (cs_index, cs) in sht.iter().enumerate() {
if cs.sh_type != SHT_PROGBITS || cs.sh_flags != SHF_ALLOC | SHF_EXECINSTR {
continue;
}
let Ok(cs_name) = sht_strtab.get(cs.sh_name as usize) else {
continue;
};
elf.section_data(&cs)?;
code_sections.insert(
cs_name.to_string(),
(vbase + cs.sh_offset as usize, cs.sh_size as usize),
);
code_section_indexes.insert(cs_index);
}
let mut insn_rewrites: Vec<(usize, u64)> = vec![];
for sec in sht.iter() {
if sec.sh_type != SHT_REL {
continue;
}
let target_section_index = sec.sh_info as usize;
if !code_section_indexes.contains(&target_section_index) {
continue;
}
let target_section = sht.get(target_section_index)?;
let target_section_name = sht_strtab
.get(target_section.sh_name as usize)
.unwrap_or_default();
let (target_section_data, _) = elf.section_data(&target_section)?;
let relocs = elf.section_data_as_rels(&sec)?;
for reloc in relocs {
let end = (reloc.r_offset as usize).saturating_add(8);
if reloc.r_offset % 8 != 0 || end > target_section_data.len() {
return Err(LinkerError::InvalidElf("relocation: invalid offset"));
}
let insn = u64::from_le_bytes(
target_section_data[reloc.r_offset as usize..end]
.try_into()
.unwrap(),
);
let mut insn = EbpfInsn::from_u64(insn);
let sym = symtab.0.get(reloc.r_sym as usize)?;
let sym_name = symtab.1.get(sym.st_name as usize)?;
if reloc.r_type == R_BPF_64_32 {
if insn.opcode != EBPF_OP_CALL {
return Err(LinkerError::Reloc(
format!("R_BPF_64_32: not a call instruction: {:?}", insn),
reloc,
));
}
let Some(&func_index) = ext_func_table.get(sym_name) else {
return Err(LinkerError::Reloc(
format!(
"R_BPF_64_32: unknown symbol {} in section {}",
sym_name, target_section_name
),
reloc,
));
};
insn.imm = func_index;
insn.src = 0;
insn_rewrites.push((
target_section.sh_offset as usize + reloc.r_offset as usize,
insn.to_u64(),
));
} else if reloc.r_type == R_BPF_64_64 {
if insn.opcode != EBPF_OP_LDDW {
return Err(LinkerError::Reloc(
"R_BPF_64_64: not a lddw instruction".to_string(),
reloc,
));
}
if end.saturating_add(8) > target_section_data.len() {
return Err(LinkerError::Reloc(
"R_BPF_64_64: out of bounds".to_string(),
reloc,
));
}
let data_section = sht.get(sym.st_shndx as usize)?;
if data_section.sh_type != SHT_PROGBITS || (data_section.sh_flags & SHF_ALLOC) == 0 {
return Err(LinkerError::Reloc(
"R_BPF_64_64: data section not SHT_PROGBITS or does not have SHF_ALLOC".to_string(),
reloc,
));
}
if sym.st_size.saturating_add(sym.st_value) > data_section.sh_size {
return Err(LinkerError::Reloc(
"R_BPF_64_64: data section out of bounds".to_string(),
reloc,
));
}
let oldimm = insn.imm as u32 as u64
+ ((u32::from_le_bytes(
<[u8; 4]>::try_from(&target_section_data[end + 4..end + 8]).unwrap(),
) as u64)
<< 32);
let imm = (vbase as u64)
.wrapping_add(data_section.sh_offset)
.wrapping_add(sym.st_value)
.wrapping_add(oldimm);
insn.imm = imm as u32 as i32;
insn_rewrites.push((
target_section.sh_offset as usize + reloc.r_offset as usize,
insn.to_u64(),
));
insn_rewrites.push((
target_section.sh_offset as usize + reloc.r_offset as usize + 8,
(imm >> 32) << 32,
));
} else {
return Err(LinkerError::Reloc(
"unsupported relocation type".to_string(),
reloc,
));
}
}
}
for (offset, value) in insn_rewrites {
input[offset..offset + 8].copy_from_slice(&value.to_le_bytes());
}
for (cs_name, &(vaddr, len)) in code_sections.iter() {
if len % 8 != 0 {
return Err(LinkerError::InvalidElf(
"code section size is not multiple of 8",
));
}
for i in (0..len).step_by(8) {
let offset = vaddr - vbase + i;
let insn = u64::from_le_bytes(input[offset..offset + 8].try_into().unwrap());
let insn = EbpfInsn::from_u64(insn);
if insn.opcode == EBPF_OP_CALL && insn.src != 0 {
return Err(LinkerError::Rejected(format!(
"call op rejected (src != 0) in section {} at offset {}",
cs_name, offset
)));
}
}
}
Ok(code_sections)
}