blob: 6249527c956010fa12d78528d0d172fa218238d9 [file] [log] [blame]
/*
* Copyright (c) 2003, 2022, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#include <unistd.h>
#include <search.h>
#include <stddef.h>
#include <stdlib.h>
#include <string.h>
#include <db.h>
#include <fcntl.h>
#include "libproc_impl.h"
#include "symtab.h"
#ifndef __APPLE__
#include "salibelf.h"
#endif // __APPLE__
// ----------------------------------------------------
// functions for symbol lookups
// ----------------------------------------------------
typedef struct symtab_symbol {
char *name; // name like __ZThread_...
uintptr_t offset; // to loaded address
uintptr_t size; // size strlen
} symtab_symbol;
typedef struct symtab {
char *strs; // all symbols "__symbol1__'\0'__symbol2__...."
size_t num_symbols;
DB* hash_table;
symtab_symbol* symbols;
} symtab_t;
#ifdef __APPLE__
void build_search_table(symtab_t *symtab) {
int i;
print_debug("build_search_table\n");
for (i = 0; i < symtab->num_symbols; i++) {
DBT key, value;
key.data = symtab->symbols[i].name;
key.size = strlen(key.data) + 1;
value.data = &(symtab->symbols[i]);
value.size = sizeof(symtab_symbol);
//print_debug("build_search_table: %d 0x%x %s\n", i, symtab->symbols[i].offset, symtab->symbols[i].name);
(*symtab->hash_table->put)(symtab->hash_table, &key, &value, 0);
// check result
if (is_debug()) {
DBT rkey, rvalue;
char* tmp = (char *)malloc(strlen(symtab->symbols[i].name) + 1);
if (tmp == NULL) {
print_debug("error allocating array in build_search_table\n");
} else {
strcpy(tmp, symtab->symbols[i].name);
rkey.data = tmp;
rkey.size = strlen(tmp) + 1;
(*symtab->hash_table->get)(symtab->hash_table, &rkey, &rvalue, 0);
// we may get a copy back so compare contents
symtab_symbol *res = (symtab_symbol *)rvalue.data;
if (strcmp(res->name, symtab->symbols[i].name) ||
res->offset != symtab->symbols[i].offset ||
res->size != symtab->symbols[i].size) {
print_debug("error to get hash_table value!\n");
}
free(tmp);
}
}
}
}
// read symbol table from given fd.
struct symtab* build_symtab(int fd, size_t *p_max_offset) {
symtab_t* symtab = NULL;
int i, j;
mach_header_64 header;
off_t image_start;
size_t max_offset = 0;
print_debug("build_symtab\n");
if (!get_arch_off(fd, CPU_TYPE_X86_64, &image_start)) {
print_debug("failed in get fat header\n");
return NULL;
}
lseek(fd, image_start, SEEK_SET);
if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
print_debug("reading header failed!\n");
return NULL;
}
// header
if (header.magic != MH_MAGIC_64) {
print_debug("not a valid .dylib file\n");
return NULL;
}
load_command lcmd;
symtab_command symtabcmd;
nlist_64 lentry;
bool lcsymtab_exist = false;
long filepos = ltell(fd);
for (i = 0; i < header.ncmds; i++) {
lseek(fd, filepos, SEEK_SET);
if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
print_debug("read load_command failed for file\n");
return NULL;
}
filepos += lcmd.cmdsize; // next command position
if (lcmd.cmd == LC_SYMTAB) {
lseek(fd, -sizeof(load_command), SEEK_CUR);
lcsymtab_exist = true;
break;
}
}
if (!lcsymtab_exist) {
print_debug("No symtab command found!\n");
return NULL;
}
if (read(fd, (void *)&symtabcmd, sizeof(symtab_command)) != sizeof(symtab_command)) {
print_debug("read symtab_command failed for file");
return NULL;
}
symtab = (symtab_t *)malloc(sizeof(symtab_t));
if (symtab == NULL) {
print_debug("out of memory: allocating symtab\n");
return NULL;
}
// create hash table, we use berkeley db to
// manipulate the hash table.
symtab->hash_table = dbopen(NULL, O_CREAT | O_RDWR, 0600, DB_HASH, NULL);
if (symtab->hash_table == NULL)
goto quit;
// allocate the symtab
symtab->num_symbols = symtabcmd.nsyms;
symtab->symbols = (symtab_symbol *)malloc(sizeof(symtab_symbol) * symtab->num_symbols);
symtab->strs = (char *)malloc(sizeof(char) * symtabcmd.strsize);
if (symtab->symbols == NULL || symtab->strs == NULL) {
print_debug("out of memory: allocating symtab.symbol or symtab.strs\n");
goto quit;
}
// read in the string table
lseek(fd, image_start + symtabcmd.stroff, SEEK_SET);
int size = read(fd, (void *)(symtab->strs), symtabcmd.strsize * sizeof(char));
if (size != symtabcmd.strsize * sizeof(char)) {
print_debug("reading string table failed\n");
goto quit;
}
// read in each nlist_64 from the symbol table and use to fill in symtab->symbols
lseek(fd, image_start + symtabcmd.symoff, SEEK_SET);
i = 0;
for (j = 0; j < symtab->num_symbols; j++) {
if (read(fd, (void *)&lentry, sizeof(nlist_64)) != sizeof(nlist_64)) {
print_debug("read nlist_64 failed at %j\n", j);
goto quit;
}
uintptr_t offset = lentry.n_value; // offset of the symbol code/data in the file
uintptr_t stridx = lentry.n_un.n_strx; // offset of symbol string in the symtabcmd.symoff section
if (stridx == 0 || offset == 0) {
continue; // Skip this entry. It's not a reference to code or data
}
if (lentry.n_type == N_OSO) {
// This is an object file name/path. These entries have something other than
// an offset in lentry.n_value, so we need to ignore them.
continue;
}
symtab->symbols[i].offset = offset;
symtab->symbols[i].name = symtab->strs + stridx;
symtab->symbols[i].size = strlen(symtab->symbols[i].name);
if (symtab->symbols[i].size == 0) {
continue; // Skip this entry. It points to an empty string.
}
// Track the maximum offset we've seen. This is used to determine the address range
// that the library covers.
if (offset > max_offset) {
max_offset = (offset + 4096) & ~0xfff; // Round up to next page boundary
}
print_debug("symbol read: %d %d n_type=0x%x n_sect=0x%x n_desc=0x%x n_strx=0x%lx offset=0x%lx %s\n",
j, i, lentry.n_type, lentry.n_sect, lentry.n_desc, stridx, offset, symtab->symbols[i].name);
i++;
}
// Update symtab->num_symbols to be the actual number of symbols we added. Since the symbols
// array was allocated larger, reallocate it to the proper size.
print_debug("build_symtab: included %d of %d entries.\n", i, symtab->num_symbols);
symtab->num_symbols = i;
symtab->symbols = (symtab_symbol *)realloc(symtab->symbols, sizeof(symtab_symbol) * symtab->num_symbols);
if (symtab->symbols == NULL) {
print_debug("out of memory: reallocating symtab.symbol\n");
goto quit;
}
// build a hashtable for fast query
build_search_table(symtab);
*p_max_offset = max_offset;
return symtab;
quit:
if (symtab) destroy_symtab(symtab);
return NULL;
}
#else // __APPLE__
struct elf_section {
ELF_SHDR *c_shdr;
void *c_data;
};
// read symbol table from given fd.
struct symtab* build_symtab(int fd) {
ELF_EHDR ehdr;
struct symtab* symtab = NULL;
// Reading of elf header
struct elf_section *scn_cache = NULL;
int cnt = 0;
ELF_SHDR* shbuf = NULL;
ELF_SHDR* cursct = NULL;
int symtab_found = 0;
int dynsym_found = 0;
uint32_t symsection = SHT_SYMTAB;
uintptr_t baseaddr = (uintptr_t)-1;
lseek(fd, (off_t)0L, SEEK_SET);
if (! read_elf_header(fd, &ehdr)) {
// not an elf
return NULL;
}
// read ELF header
if ((shbuf = read_section_header_table(fd, &ehdr)) == NULL) {
goto quit;
}
baseaddr = find_base_address(fd, &ehdr);
scn_cache = calloc(ehdr.e_shnum, sizeof(*scn_cache));
if (scn_cache == NULL) {
goto quit;
}
for (cursct = shbuf, cnt = 0; cnt < ehdr.e_shnum; cnt++) {
scn_cache[cnt].c_shdr = cursct;
if (cursct->sh_type == SHT_SYMTAB ||
cursct->sh_type == SHT_STRTAB ||
cursct->sh_type == SHT_DYNSYM) {
if ( (scn_cache[cnt].c_data = read_section_data(fd, &ehdr, cursct)) == NULL) {
goto quit;
}
}
if (cursct->sh_type == SHT_SYMTAB)
symtab_found++;
if (cursct->sh_type == SHT_DYNSYM)
dynsym_found++;
cursct++;
}
if (!symtab_found && dynsym_found)
symsection = SHT_DYNSYM;
for (cnt = 1; cnt < ehdr.e_shnum; cnt++) {
ELF_SHDR *shdr = scn_cache[cnt].c_shdr;
if (shdr->sh_type == symsection) {
ELF_SYM *syms;
int j, n;
size_t size;
// FIXME: there could be multiple data buffers associated with the
// same ELF section. Here we can handle only one buffer. See man page
// for elf_getdata on Solaris.
// guarantee(symtab == NULL, "multiple symtab");
symtab = calloc(1, sizeof(*symtab));
if (symtab == NULL) {
goto quit;
}
// the symbol table
syms = (ELF_SYM *)scn_cache[cnt].c_data;
// number of symbols
n = shdr->sh_size / shdr->sh_entsize;
// create hash table, we use berkeley db to
// manipulate the hash table.
symtab->hash_table = dbopen(NULL, O_CREAT | O_RDWR, 0600, DB_HASH, NULL);
// guarantee(symtab->hash_table, "unexpected failure: dbopen");
if (symtab->hash_table == NULL)
goto bad;
// shdr->sh_link points to the section that contains the actual strings
// for symbol names. the st_name field in ELF_SYM is just the
// string table index. we make a copy of the string table so the
// strings will not be destroyed by elf_end.
size = scn_cache[shdr->sh_link].c_shdr->sh_size;
symtab->strs = malloc(size);
if (symtab->strs == NULL)
goto bad;
memcpy(symtab->strs, scn_cache[shdr->sh_link].c_data, size);
// allocate memory for storing symbol offset and size;
symtab->num_symbols = n;
symtab->symbols = calloc(n , sizeof(*symtab->symbols));
if (symtab->symbols == NULL)
goto bad;
// copy symbols info our symtab and enter them info the hash table
for (j = 0; j < n; j++, syms++) {
DBT key, value;
char *sym_name = symtab->strs + syms->st_name;
// skip non-object and non-function symbols
int st_type = ELF_ST_TYPE(syms->st_info);
if ( st_type != STT_FUNC && st_type != STT_OBJECT)
continue;
// skip empty strings and undefined symbols
if (*sym_name == '\0' || syms->st_shndx == SHN_UNDEF) continue;
symtab->symbols[j].name = sym_name;
symtab->symbols[j].offset = syms->st_value - baseaddr;
symtab->symbols[j].size = syms->st_size;
key.data = sym_name;
key.size = strlen(sym_name) + 1;
value.data = &(symtab->symbols[j]);
value.size = sizeof(symtab_symbol);
(*symtab->hash_table->put)(symtab->hash_table, &key, &value, 0);
}
}
}
goto quit;
bad:
destroy_symtab(symtab);
symtab = NULL;
quit:
if (shbuf) free(shbuf);
if (scn_cache) {
for (cnt = 0; cnt < ehdr.e_shnum; cnt++) {
if (scn_cache[cnt].c_data != NULL) {
free(scn_cache[cnt].c_data);
}
}
free(scn_cache);
}
return symtab;
}
#endif // __APPLE__
void destroy_symtab(symtab_t* symtab) {
if (!symtab) return;
free(symtab->strs);
free(symtab->symbols);
free(symtab);
}
uintptr_t search_symbol(struct symtab* symtab, uintptr_t base, const char *sym_name, int *sym_size) {
DBT key, value;
int ret;
// library does not have symbol table
if (!symtab || !symtab->hash_table) {
return 0;
}
key.data = (char*)(uintptr_t)sym_name;
key.size = strlen(sym_name) + 1;
ret = (*symtab->hash_table->get)(symtab->hash_table, &key, &value, 0);
if (ret == 0) {
symtab_symbol *sym = value.data;
uintptr_t rslt = (uintptr_t) ((char*)base + sym->offset);
if (sym_size) *sym_size = sym->size;
return rslt;
}
return 0;
}
const char* nearest_symbol(struct symtab* symtab, uintptr_t offset,
uintptr_t* poffset) {
int n = 0;
char* result = NULL;
ptrdiff_t lowest_offset_from_sym = -1;
if (!symtab) return NULL;
// Search the symbol table for the symbol that is closest to the specified offset, but is not under.
//
// Note we can't just use the first symbol that is >= the offset because the symbols may not be
// sorted by offset.
//
// Note this is a rather slow search that is O(n/2), and libjvm has as many as 250k symbols.
// Probably would be good to sort the array and do a binary search, or use a hash table like
// we do for name -> address lookups. However, this functionality is not used often and
// generally just involves one lookup, such as with the clhsdb "findpc" command.
for (; n < symtab->num_symbols; n++) {
symtab_symbol* sym = &(symtab->symbols[n]);
if (sym->size != 0 && offset >= sym->offset) {
ptrdiff_t offset_from_sym = offset - sym->offset;
if (offset_from_sym >= 0) { // ignore symbols that come after "offset"
if (lowest_offset_from_sym == -1 || offset_from_sym < lowest_offset_from_sym) {
lowest_offset_from_sym = offset_from_sym;
result = sym->name;
//print_debug("nearest_symbol: found %d %s 0x%x 0x%x 0x%x\n",
// n, sym->name, offset, sym->offset, lowest_offset_from_sym);
}
}
}
}
print_debug("nearest_symbol: found symbol %d file_offset=0x%lx sym_offset=0x%lx %s\n",
n, offset, lowest_offset_from_sym, result);
// Save the offset from the symbol if requested.
if (result != NULL && poffset) {
*poffset = lowest_offset_from_sym;
}
return result;
}