elf_object.cc revision 4111
112SN/A/* 21762SN/A * Copyright (c) 2003-2005 The Regents of The University of Michigan 312SN/A * All rights reserved. 412SN/A * 512SN/A * Redistribution and use in source and binary forms, with or without 612SN/A * modification, are permitted provided that the following conditions are 712SN/A * met: redistributions of source code must retain the above copyright 812SN/A * notice, this list of conditions and the following disclaimer; 912SN/A * redistributions in binary form must reproduce the above copyright 1012SN/A * notice, this list of conditions and the following disclaimer in the 1112SN/A * documentation and/or other materials provided with the distribution; 1212SN/A * neither the name of the copyright holders nor the names of its 1312SN/A * contributors may be used to endorse or promote products derived from 1412SN/A * this software without specific prior written permission. 1512SN/A * 1612SN/A * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 1712SN/A * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 1812SN/A * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 1912SN/A * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 2012SN/A * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 2112SN/A * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 2212SN/A * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 2312SN/A * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 2412SN/A * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 2512SN/A * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 2612SN/A * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 272665Ssaidi@eecs.umich.edu * 282665Ssaidi@eecs.umich.edu * Authors: Steve Reinhardt 292665Ssaidi@eecs.umich.edu * Ali Saidi 3012SN/A */ 3112SN/A 3212SN/A#include <string> 3312SN/A 34468SN/A// Because of the -Wundef flag we have to do this 35468SN/A#define __LIBELF_INTERNAL__ 0 36468SN/A#define __LIBELF_NEED_LINK_H 0 37661SN/A#define __LIBELF_SYMBOL_VERSIONS 0 38468SN/A 392634Sstever@eecs.umich.edu#include "gelf.h" 40468SN/A 4156SN/A#include "base/loader/elf_object.hh" 422439SN/A#include "base/misc.hh" 4312SN/A 4456SN/A#include "base/loader/symtab.hh" 4512SN/A 4656SN/A#include "base/trace.hh" // for DPRINTF 4712SN/A 482423SN/A#include "sim/byteswap.hh" 492423SN/A 5012SN/A 5112SN/Ausing namespace std; 5212SN/A 5312SN/AObjectFile * 5412SN/AElfObject::tryFile(const string &fname, int fd, size_t len, uint8_t *data) 5512SN/A{ 56443SN/A Elf *elf; 57443SN/A GElf_Ehdr ehdr; 582207SN/A Arch arch = UnknownArch; 592207SN/A OpSys opSys = UnknownOpSys; 60443SN/A 61468SN/A // check that header matches library version 621708SN/A if (elf_version(EV_CURRENT) == EV_NONE) 631708SN/A panic("wrong elf version number!"); 64443SN/A 65468SN/A // get a pointer to elf structure 66443SN/A elf = elf_memory((char*)data,len); 67468SN/A // will only fail if fd is invalid 68443SN/A assert(elf != NULL); 69443SN/A 70468SN/A // Check that we actually have a elf file 71468SN/A if (gelf_getehdr(elf, &ehdr) ==0) { 72443SN/A DPRINTFR(Loader, "Not ELF\n"); 73443SN/A elf_end(elf); 74443SN/A return NULL; 752476SN/A } else { 762207SN/A //Detect the architecture 772207SN/A //Since we don't know how to check for alpha right now, we'll 782207SN/A //just assume if it wasn't something else and it's 64 bit, that's 792207SN/A //what it must be. 802207SN/A if (ehdr.e_machine == EM_SPARC64 || 814111Sgblack@eecs.umich.edu (ehdr.e_machine == EM_SPARC && 824111Sgblack@eecs.umich.edu ehdr.e_ident[EI_CLASS] == ELFCLASS64)|| 832620SN/A ehdr.e_machine == EM_SPARCV9) { 844111Sgblack@eecs.umich.edu arch = ObjectFile::SPARC64; 854111Sgblack@eecs.umich.edu } else if (ehdr.e_machine == EM_SPARC32PLUS || 864111Sgblack@eecs.umich.edu (ehdr.e_machine == EM_SPARC && 874111Sgblack@eecs.umich.edu ehdr.e_ident[EI_CLASS] == ELFCLASS32)) { 884111Sgblack@eecs.umich.edu arch = ObjectFile::SPARC32; 892207SN/A } else if (ehdr.e_machine == EM_MIPS 902207SN/A && ehdr.e_ident[EI_CLASS] == ELFCLASS32) { 912472SN/A arch = ObjectFile::Mips; 922207SN/A } else if (ehdr.e_ident[EI_CLASS] == ELFCLASS64) { 932207SN/A arch = ObjectFile::Alpha; 942207SN/A } else { 952600SN/A warn("Unknown architecture: %d\n", ehdr.e_machine); 962207SN/A arch = ObjectFile::UnknownArch; 972207SN/A } 982207SN/A 992207SN/A //Detect the operating system 1002207SN/A switch (ehdr.e_ident[EI_OSABI]) 1012207SN/A { 1022238SN/A 1032207SN/A case ELFOSABI_LINUX: 1042207SN/A opSys = ObjectFile::Linux; 1052207SN/A break; 1062207SN/A case ELFOSABI_SOLARIS: 1072207SN/A opSys = ObjectFile::Solaris; 1082238SN/A break; 1092207SN/A case ELFOSABI_TRU64: 1102207SN/A opSys = ObjectFile::Tru64; 1112238SN/A break; 1122207SN/A default: 1132207SN/A opSys = ObjectFile::UnknownOpSys; 1142207SN/A } 1152207SN/A 1162238SN/A //take a look at the .note.ABI section 1172238SN/A //It can let us know what's what. 1182600SN/A if (opSys == ObjectFile::UnknownOpSys) { 1192238SN/A Elf_Scn *section; 1202238SN/A GElf_Shdr shdr; 1212238SN/A Elf_Data *data; 1222238SN/A uint32_t osAbi;; 1232238SN/A int secIdx = 1; 1242238SN/A 1252238SN/A // Get the first section 1262238SN/A section = elf_getscn(elf, secIdx); 1272238SN/A 1282238SN/A // While there are no more sections 1292600SN/A while (section != NULL && opSys == ObjectFile::UnknownOpSys) { 1302238SN/A gelf_getshdr(section, &shdr); 1312238SN/A if (shdr.sh_type == SHT_NOTE && !strcmp(".note.ABI-tag", 1322238SN/A elf_strptr(elf, ehdr.e_shstrndx, shdr.sh_name))) { 1332238SN/A // we have found a ABI note section 1342238SN/A // Check the 5th 32bit word for OS 0 == linux, 1 == hurd, 1352238SN/A // 2 == solaris, 3 == freebsd 1362238SN/A data = elf_rawdata(section, NULL); 1372238SN/A assert(data->d_buf); 1382238SN/A if(ehdr.e_ident[EI_DATA] == ELFDATA2LSB) 1392238SN/A osAbi = htole(((uint32_t*)data->d_buf)[4]); 1402238SN/A else 1412238SN/A osAbi = htobe(((uint32_t*)data->d_buf)[4]); 1422238SN/A 1432238SN/A switch(osAbi) { 1442238SN/A case 0: 1452238SN/A opSys = ObjectFile::Linux; 1462238SN/A break; 1472238SN/A case 2: 1482238SN/A opSys = ObjectFile::Solaris; 1492238SN/A break; 1502238SN/A } 1512238SN/A } // if section found 1522600SN/A if (!strcmp(".SUNW_version", elf_strptr(elf, ehdr.e_shstrndx, shdr.sh_name))) 1532600SN/A opSys = ObjectFile::Solaris; 1542600SN/A if (!strcmp(".stab.index", elf_strptr(elf, ehdr.e_shstrndx, shdr.sh_name))) 1552600SN/A opSys = ObjectFile::Solaris; 1562600SN/A 1572238SN/A section = elf_getscn(elf, ++secIdx); 1582238SN/A } // while sections 1592238SN/A } 1602472SN/A 1612976Sgblack@eecs.umich.edu ElfObject * result = new ElfObject(fname, fd, len, data, arch, opSys); 1622976Sgblack@eecs.umich.edu 1632976Sgblack@eecs.umich.edu //The number of headers in the file 1642976Sgblack@eecs.umich.edu result->_programHeaderCount = ehdr.e_phnum; 1652976Sgblack@eecs.umich.edu //Record the size of each entry 1662976Sgblack@eecs.umich.edu result->_programHeaderSize = ehdr.e_phentsize; 1672976Sgblack@eecs.umich.edu if(result->_programHeaderCount) //If there is a program header table 1682976Sgblack@eecs.umich.edu { 1692976Sgblack@eecs.umich.edu //Figure out the virtual address of the header table in the 1702976Sgblack@eecs.umich.edu //final memory image. We use the program headers themselves 1712976Sgblack@eecs.umich.edu //to translate from a file offset to the address in the image. 1722976Sgblack@eecs.umich.edu GElf_Phdr phdr; 1732976Sgblack@eecs.umich.edu uint64_t e_phoff = ehdr.e_phoff; 1742976Sgblack@eecs.umich.edu result->_programHeaderTable = 0; 1752976Sgblack@eecs.umich.edu for(int hdrnum = 0; hdrnum < result->_programHeaderCount; hdrnum++) 1762976Sgblack@eecs.umich.edu { 1772976Sgblack@eecs.umich.edu gelf_getphdr(elf, hdrnum, &phdr); 1782976Sgblack@eecs.umich.edu //Check if we've found the segment with the headers in it 1792976Sgblack@eecs.umich.edu if(phdr.p_offset <= e_phoff && 1802976Sgblack@eecs.umich.edu phdr.p_offset + phdr.p_filesz > e_phoff) 1812976Sgblack@eecs.umich.edu { 1822976Sgblack@eecs.umich.edu result->_programHeaderTable = phdr.p_vaddr + e_phoff; 1832976Sgblack@eecs.umich.edu break; 1842976Sgblack@eecs.umich.edu } 1852976Sgblack@eecs.umich.edu } 1862976Sgblack@eecs.umich.edu } 1872976Sgblack@eecs.umich.edu else 1882976Sgblack@eecs.umich.edu result->_programHeaderTable = 0; 1892976Sgblack@eecs.umich.edu 1902976Sgblack@eecs.umich.edu 1912238SN/A elf_end(elf); 1922976Sgblack@eecs.umich.edu return result; 19312SN/A } 19412SN/A} 19512SN/A 19612SN/A 19712SN/AElfObject::ElfObject(const string &_filename, int _fd, 198360SN/A size_t _len, uint8_t *_data, 199360SN/A Arch _arch, OpSys _opSys) 200360SN/A : ObjectFile(_filename, _fd, _len, _data, _arch, _opSys) 201443SN/A 20212SN/A{ 203443SN/A Elf *elf; 204443SN/A GElf_Ehdr ehdr; 20512SN/A 206468SN/A // check that header matches library version 2071708SN/A if (elf_version(EV_CURRENT) == EV_NONE) 2081708SN/A panic("wrong elf version number!"); 20912SN/A 210468SN/A // get a pointer to elf structure 211443SN/A elf = elf_memory((char*)fileData,len); 212468SN/A // will only fail if fd is invalid 213443SN/A assert(elf != NULL); 21412SN/A 215468SN/A // Check that we actually have a elf file 216468SN/A if (gelf_getehdr(elf, &ehdr) ==0) { 217443SN/A panic("Not ELF, shouldn't be here"); 21812SN/A } 21912SN/A 220468SN/A entry = ehdr.e_entry; 22112SN/A 2222472SN/A 223468SN/A // initialize segment sizes to 0 in case they're not present 224468SN/A text.size = data.size = bss.size = 0; 225468SN/A 226468SN/A for (int i = 0; i < ehdr.e_phnum; ++i) { 227468SN/A GElf_Phdr phdr; 228468SN/A if (gelf_getphdr(elf, i, &phdr) == 0) { 229468SN/A panic("gelf_getphdr failed for section %d", i); 230468SN/A } 231468SN/A 232468SN/A // for now we don't care about non-loadable segments 233468SN/A if (!(phdr.p_type & PT_LOAD)) 234468SN/A continue; 235468SN/A 236468SN/A // the headers don't explicitly distinguish text from data, 237468SN/A // but empirically the text segment comes first. 238468SN/A if (text.size == 0) { // haven't seen text segment yet 239468SN/A text.baseAddr = phdr.p_vaddr; 240468SN/A text.size = phdr.p_filesz; 2412420SN/A text.fileImage = fileData + phdr.p_offset; 242468SN/A // if there's any padding at the end that's not in the 243468SN/A // file, call it the bss. This happens in the "text" 244468SN/A // segment if there's only one loadable segment (as for 245468SN/A // kernel images). 246468SN/A bss.size = phdr.p_memsz - phdr.p_filesz; 247468SN/A bss.baseAddr = phdr.p_vaddr + phdr.p_filesz; 2482420SN/A bss.fileImage = NULL; 2492476SN/A } else if (data.size == 0) { // have text, this must be data 250468SN/A data.baseAddr = phdr.p_vaddr; 251468SN/A data.size = phdr.p_filesz; 2522420SN/A data.fileImage = fileData + phdr.p_offset; 253468SN/A // if there's any padding at the end that's not in the 254468SN/A // file, call it the bss. Warn if this happens for both 255468SN/A // the text & data segments (should only have one bss). 256468SN/A if (phdr.p_memsz - phdr.p_filesz > 0 && bss.size != 0) { 257468SN/A warn("Two implied bss segments in file!\n"); 258468SN/A } 259468SN/A bss.size = phdr.p_memsz - phdr.p_filesz; 260468SN/A bss.baseAddr = phdr.p_vaddr + phdr.p_filesz; 2612420SN/A bss.fileImage = NULL; 2622476SN/A } else { 2632476SN/A warn("More than two loadable segments in ELF object."); 2642476SN/A warn("Ignoring segment @ 0x%x length 0x%x.", 2652476SN/A phdr.p_vaddr, phdr.p_filesz); 266468SN/A } 267468SN/A } 268468SN/A 269468SN/A // should have found at least one loadable segment 270468SN/A assert(text.size != 0); 271468SN/A 272468SN/A DPRINTFR(Loader, "text: 0x%x %d\ndata: 0x%x %d\nbss: 0x%x %d\n", 273468SN/A text.baseAddr, text.size, data.baseAddr, data.size, 274468SN/A bss.baseAddr, bss.size); 275468SN/A 276443SN/A elf_end(elf); 277443SN/A 278468SN/A // We will actually read the sections when we need to load them 27912SN/A} 28012SN/A 28112SN/A 28212SN/Abool 283468SN/AElfObject::loadSomeSymbols(SymbolTable *symtab, int binding) 28412SN/A{ 285443SN/A Elf *elf; 286766SN/A int sec_idx = 1; // there is a 0 but it is nothing, go figure 287443SN/A Elf_Scn *section; 288443SN/A GElf_Shdr shdr; 289443SN/A Elf_Data *data; 290443SN/A int count, ii; 291443SN/A bool found = false; 292443SN/A GElf_Sym sym; 293443SN/A 294443SN/A if (!symtab) 295443SN/A return false; 296443SN/A 297468SN/A // check that header matches library version 2981708SN/A if (elf_version(EV_CURRENT) == EV_NONE) 2991708SN/A panic("wrong elf version number!"); 300443SN/A 301468SN/A // get a pointer to elf structure 302443SN/A elf = elf_memory((char*)fileData,len); 303443SN/A 304443SN/A assert(elf != NULL); 305443SN/A 306468SN/A // Get the first section 307454SN/A section = elf_getscn(elf, sec_idx); 308443SN/A 309468SN/A // While there are no more sections 310468SN/A while (section != NULL) { 311443SN/A gelf_getshdr(section, &shdr); 312443SN/A 313468SN/A if (shdr.sh_type == SHT_SYMTAB) { 314443SN/A found = true; 315443SN/A data = elf_getdata(section, NULL); 316443SN/A count = shdr.sh_size / shdr.sh_entsize; 317443SN/A DPRINTF(Loader, "Found Symbol Table, %d symbols present\n", count); 318443SN/A 319468SN/A // loop through all the symbols, only loading global ones 320468SN/A for (ii = 0; ii < count; ++ii) { 321443SN/A gelf_getsym(data, ii, &sym); 322836SN/A if (GELF_ST_BIND(sym.st_info) == binding) { 323468SN/A symtab->insert(sym.st_value, 324468SN/A elf_strptr(elf, shdr.sh_link, sym.st_name)); 325443SN/A } 326443SN/A } 327443SN/A } 328454SN/A ++sec_idx; 329454SN/A section = elf_getscn(elf, sec_idx); 330443SN/A } 331443SN/A 332443SN/A elf_end(elf); 333443SN/A 334443SN/A return found; 33512SN/A} 33612SN/A 33712SN/Abool 3383812Ssaidi@eecs.umich.eduElfObject::loadGlobalSymbols(SymbolTable *symtab, Addr addrMask) 339468SN/A{ 340468SN/A return loadSomeSymbols(symtab, STB_GLOBAL); 341468SN/A} 342468SN/A 343468SN/Abool 3443812Ssaidi@eecs.umich.eduElfObject::loadLocalSymbols(SymbolTable *symtab, Addr addrMask) 34512SN/A{ 346468SN/A return loadSomeSymbols(symtab, STB_LOCAL); 34712SN/A} 3483917Ssaidi@eecs.umich.edu 3493917Ssaidi@eecs.umich.edubool 3503917Ssaidi@eecs.umich.eduElfObject::isDynamic() 3513917Ssaidi@eecs.umich.edu{ 3523917Ssaidi@eecs.umich.edu Elf *elf; 3533917Ssaidi@eecs.umich.edu int sec_idx = 1; // there is a 0 but it is nothing, go figure 3543917Ssaidi@eecs.umich.edu Elf_Scn *section; 3553917Ssaidi@eecs.umich.edu GElf_Shdr shdr; 3563917Ssaidi@eecs.umich.edu 3573917Ssaidi@eecs.umich.edu GElf_Ehdr ehdr; 3583917Ssaidi@eecs.umich.edu 3593917Ssaidi@eecs.umich.edu // check that header matches library version 3603917Ssaidi@eecs.umich.edu if (elf_version(EV_CURRENT) == EV_NONE) 3613917Ssaidi@eecs.umich.edu panic("wrong elf version number!"); 3623917Ssaidi@eecs.umich.edu 3633917Ssaidi@eecs.umich.edu // get a pointer to elf structure 3643917Ssaidi@eecs.umich.edu elf = elf_memory((char*)fileData,len); 3653917Ssaidi@eecs.umich.edu assert(elf != NULL); 3663917Ssaidi@eecs.umich.edu 3673917Ssaidi@eecs.umich.edu // Check that we actually have a elf file 3683917Ssaidi@eecs.umich.edu if (gelf_getehdr(elf, &ehdr) ==0) { 3693917Ssaidi@eecs.umich.edu panic("Not ELF, shouldn't be here"); 3703917Ssaidi@eecs.umich.edu } 3713917Ssaidi@eecs.umich.edu 3723917Ssaidi@eecs.umich.edu // Get the first section 3733917Ssaidi@eecs.umich.edu section = elf_getscn(elf, sec_idx); 3743917Ssaidi@eecs.umich.edu 3753917Ssaidi@eecs.umich.edu // While there are no more sections 3763917Ssaidi@eecs.umich.edu while (section != NULL) { 3773917Ssaidi@eecs.umich.edu gelf_getshdr(section, &shdr); 3783925Ssaidi@eecs.umich.edu if (!strcmp(".interp", elf_strptr(elf, ehdr.e_shstrndx, shdr.sh_name))) 3793917Ssaidi@eecs.umich.edu return true; 3803917Ssaidi@eecs.umich.edu section = elf_getscn(elf, ++sec_idx); 3813917Ssaidi@eecs.umich.edu } // while sections 3823917Ssaidi@eecs.umich.edu return false; 3833917Ssaidi@eecs.umich.edu} 3843917Ssaidi@eecs.umich.edu 3853917Ssaidi@eecs.umich.edu 386