read/elf/hash.rs

use core::mem;

use crate::elf;
use crate::read::{ReadError, ReadRef, Result};
use crate::{U32, U64};

use super::{FileHeader, Sym, SymbolTable};

/// A SysV symbol hash table in an ELF file.
#[derive(Debug)]
pub struct HashTable<'data, Elf: FileHeader> {
    buckets: &'data [U32<Elf::Endian>],
    chains: &'data [U32<Elf::Endian>],
}

impl<'data, Elf: FileHeader> HashTable<'data, Elf> {
    /// Parse a SysV hash table.
    ///
    /// `data` should be from a `SHT_HASH` section, or from a
    /// segment pointed to via the `DT_HASH` entry.
    ///
    /// The header is read at offset 0 in the given `data`.
    pub fn parse(endian: Elf::Endian, data: &'data [u8]) -> Result<Self> {
        let mut offset = 0;
        let header = data
            .read::<elf::HashHeader<Elf::Endian>>(&mut offset)
            .read_error("Invalid hash header")?;
        let buckets = data
            .read_slice(&mut offset, header.bucket_count.get(endian) as usize)
            .read_error("Invalid hash buckets")?;
        let chains = data
            .read_slice(&mut offset, header.chain_count.get(endian) as usize)
            .read_error("Invalid hash chains")?;
        Ok(HashTable { buckets, chains })
    }

    /// Return the symbol table length.
    pub fn symbol_table_length(&self) -> u32 {
        self.chains.len() as u32
    }

    /// Use the hash table to find the symbol table entry with the given name and hash.
    pub fn find<R: ReadRef<'data>>(
        &self,
        endian: Elf::Endian,
        name: &[u8],
        hash: u32,
        symbols: &SymbolTable<'data, Elf, R>,
    ) -> Option<&'data Elf::Sym> {
        // Get the chain start from the bucket for this hash.
        let mut index = self.buckets[(hash as usize) % self.buckets.len()].get(endian) as usize;
        // Avoid infinite loop.
        let mut i = 0;
        let strings = symbols.strings();
        while index != 0 && i < self.chains.len() {
            if let Ok(symbol) = symbols.symbol(index) {
                if symbol.name(endian, strings) == Ok(name) {
                    return Some(symbol);
                }
            }
            index = self.chains.get(index)?.get(endian) as usize;
            i += 1;
        }
        None
    }
}

/// A GNU symbol hash table in an ELF file.
#[derive(Debug)]
pub struct GnuHashTable<'data, Elf: FileHeader> {
    symbol_base: u32,
    bloom_shift: u32,
    bloom_filters: &'data [u8],
    buckets: &'data [U32<Elf::Endian>],
    values: &'data [U32<Elf::Endian>],
}

impl<'data, Elf: FileHeader> GnuHashTable<'data, Elf> {
    /// Parse a GNU hash table.
    ///
    /// `data` should be from a `SHT_GNU_HASH` section, or from a
    /// segment pointed to via the `DT_GNU_HASH` entry.
    ///
    /// The header is read at offset 0 in the given `data`.
    ///
    /// The header does not contain a length field, and so all of `data`
    /// will be used as the hash table values. It does not matter if this
    /// is longer than needed, and this will often the case when accessing
    /// the hash table via the `DT_GNU_HASH` entry.
    pub fn parse(endian: Elf::Endian, data: &'data [u8]) -> Result<Self> {
        let mut offset = 0;
        let header = data
            .read::<elf::GnuHashHeader<Elf::Endian>>(&mut offset)
            .read_error("Invalid GNU hash header")?;
        let bloom_len =
            u64::from(header.bloom_count.get(endian)) * mem::size_of::<Elf::Word>() as u64;
        let bloom_filters = data
            .read_bytes(&mut offset, bloom_len)
            .read_error("Invalid GNU hash bloom filters")?;
        let buckets = data
            .read_slice(&mut offset, header.bucket_count.get(endian) as usize)
            .read_error("Invalid GNU hash buckets")?;
        let chain_count = (data.len() - offset as usize) / 4;
        let values = data
            .read_slice(&mut offset, chain_count)
            .read_error("Invalid GNU hash values")?;
        Ok(GnuHashTable {
            symbol_base: header.symbol_base.get(endian),
            bloom_shift: header.bloom_shift.get(endian),
            bloom_filters,
            buckets,
            values,
        })
    }

    /// Return the symbol table index of the first symbol in the hash table.
    pub fn symbol_base(&self) -> u32 {
        self.symbol_base
    }

    /// Determine the symbol table length by finding the last entry in the hash table.
    ///
    /// Returns `None` if the hash table is empty or invalid.
    pub fn symbol_table_length(&self, endian: Elf::Endian) -> Option<u32> {
        // Ensure we find a non-empty bucket.
        if self.symbol_base == 0 {
            return None;
        }

        // Find the highest chain index in a bucket.
        let mut max_symbol = 0;
        for bucket in self.buckets {
            let bucket = bucket.get(endian);
            if max_symbol < bucket {
                max_symbol = bucket;
            }
        }

        // Find the end of the chain.
        for value in self
            .values
            .get(max_symbol.checked_sub(self.symbol_base)? as usize..)?
        {
            max_symbol += 1;
            if value.get(endian) & 1 != 0 {
                return Some(max_symbol);
            }
        }

        None
    }

    /// Use the hash table to find the symbol table entry with the given name and hash.
    pub fn find<R: ReadRef<'data>>(
        &self,
        endian: Elf::Endian,
        name: &[u8],
        hash: u32,
        symbols: &SymbolTable<'data, Elf, R>,
    ) -> Option<&'data Elf::Sym> {
        let word_bits = mem::size_of::<Elf::Word>() as u32 * 8;

        // Test against bloom filter.
        let bloom_count = self.bloom_filters.len() / mem::size_of::<Elf::Word>();
        let offset =
            ((hash / word_bits) & (bloom_count as u32 - 1)) * mem::size_of::<Elf::Word>() as u32;
        let filter = if word_bits == 64 {
            self.bloom_filters
                .read_at::<U64<Elf::Endian>>(offset.into())
                .ok()?
                .get(endian)
        } else {
            self.bloom_filters
                .read_at::<U32<Elf::Endian>>(offset.into())
                .ok()?
                .get(endian)
                .into()
        };
        if filter & (1 << (hash % word_bits)) == 0 {
            return None;
        }
        if filter & (1 << ((hash >> self.bloom_shift) % word_bits)) == 0 {
            return None;
        }

        // Get the chain start from the bucket for this hash.
        let index = self.buckets[(hash as usize) % self.buckets.len()].get(endian) as usize;
        if index == 0 {
            return None;
        }

        // Test symbols in the chain.
        let strings = symbols.strings();
        let symbols = symbols.symbols().get(index..)?;
        let values = self
            .values
            .get(index.checked_sub(self.symbol_base as usize)?..)?;
        for (symbol, value) in symbols.iter().zip(values.iter()) {
            let value = value.get(endian);
            if value | 1 == hash | 1 {
                if symbol.name(endian, strings) == Ok(name) {
                    return Some(symbol);
                }
            }
            if value & 1 != 0 {
                break;
            }
        }
        None
    }
}