p_vmlinx.cpp

UPX 源代码

/* p_vmlinx.cpp -- pack vmlinux ET_EXEC file (before bootsect or setup)

   This file is part of the UPX executable compressor.

   Copyright (C) 2004-2007 John Reiser
   Copyright (C) 1996-2007 Markus Franz Xaver Johannes Oberhumer
   Copyright (C) 1996-2007 Laszlo Molnar
   All Rights Reserved.

   UPX and the UCL library are free software; you can redistribute them
   and/or modify them under the terms of the GNU General Public License as
   published by the Free Software Foundation; either version 2 of
   the License, or (at your option) any later version.

   This program 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 for more details.

   You should have received a copy of the GNU General Public License
   along with this program; see the file COPYING.
   If not, write to the Free Software Foundation, Inc.,
   59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

   Markus F.X.J. Oberhumer   Laszlo Molnar
   [e-mail]      [e-mail]

   John Reiser
   [e-mail]
 */


#include "conf.h"

#include "file.h"
#include "filter.h"
#include "packer.h"
#include "p_vmlinx.h"
#include "linker.h"

static const
#include "stub/i386-linux.kernel.vmlinux.h"
static const
#include "stub/amd64-linux.kernel.vmlinux.h"
static const
#include "stub/arm-linux.kernel.vmlinux.h"
static const
#include "stub/armeb-linux.kernel.vmlinux.h"
static const
#include "stub/powerpc-linux.kernel.vmlinux.h"


/*************************************************************************
//
**************************************************************************/

template <class T>
PackVmlinuxBase<T>::PackVmlinuxBase(InputFile *f,
                                    unsigned e_machine, unsigned elfclass, unsigned elfdata,
                                    char const *const boot_label) :
    super(f),
    my_e_machine(e_machine), my_elfclass(elfclass), my_elfdata(elfdata),
    my_boot_label(boot_label),
    n_ptload(0), phdri(NULL), shdri(NULL), shstrtab(NULL)
{
    ElfClass::compileTimeAssertions();
    bele = N_BELE_CTP::getRTP<BeLePolicy>();
}

template <class T>
PackVmlinuxBase<T>::~PackVmlinuxBase()
{
    delete [] phdri;
    delete [] shdri;
    delete [] shstrtab;
}

template <class T>
int PackVmlinuxBase<T>::getStrategy(Filter &/*ft*/)
{
    // Called just before reading and compressing each block.
    // Might want to adjust blocksize, etc.

    // If user specified the filter, then use it (-2==strategy).
    // Else try the first two filters, and pick the better (2==strategy).
    return (opt->no_filter ? -3 : ((opt->filter > 0) ? -2 : 2));
}

template <class T>
int __acc_cdecl_qsort
PackVmlinuxBase<T>::compare_Phdr(void const *aa, void const *bb)
{
    Phdr const *const a = (Phdr const *)aa;
    Phdr const *const b = (Phdr const *)bb;
    unsigned const xa = a->p_type - Phdr::PT_LOAD;
    unsigned const xb = b->p_type - Phdr::PT_LOAD;
            if (xa < xb)         return -1;  // PT_LOAD first
            if (xa > xb)         return  1;
    if (a->p_paddr < b->p_paddr) return -1;  // ascending by .p_paddr
    if (a->p_paddr > b->p_paddr) return  1;
                                 return  0;
}

template <class T>
typename T::Shdr const *PackVmlinuxBase<T>::getElfSections()
{
    Shdr const *p, *shstrsec=0;
    shdri = new Shdr[(unsigned) ehdri.e_shnum];
    fi->seek(ehdri.e_shoff, SEEK_SET);
    fi->readx(shdri, ehdri.e_shnum * sizeof(*shdri));
    int j;
    for (p = shdri, j= ehdri.e_shnum; --j>=0; ++p) {
        if (Shdr::SHT_STRTAB==p->sh_type
        &&  (p->sh_size + p->sh_offset) <= (unsigned) file_size
        &&  (10+ p->sh_name) <= p->sh_size  // 1+ strlen(".shstrtab")
        ) {
            delete [] shstrtab;
            shstrtab = new char[1+ p->sh_size];
            fi->seek(p->sh_offset, SEEK_SET);
            fi->readx(shstrtab, p->sh_size);
            shstrtab[p->sh_size] = '\0';
            if (0==strcmp(".shstrtab", shstrtab + p->sh_name)) {
                shstrsec = p;
                break;
            }
        }
    }
    return shstrsec;
}

template <class T>
bool PackVmlinuxBase<T>::canPack()
{
    fi->seek(0, SEEK_SET);
    fi->readx(&ehdri, sizeof(ehdri));

    // now check the ELF header
    if (memcmp(&ehdri, "\x7f\x45\x4c\x46", 4)
    ||  ehdri.e_ident[Ehdr::EI_CLASS] != my_elfclass
    ||  ehdri.e_ident[Ehdr::EI_DATA] != my_elfdata
    ||  ehdri.e_ident[Ehdr::EI_VERSION] != Ehdr::EV_CURRENT
    ||  !memcmp(&ehdri.e_ident[8], "FreeBSD", 7)  // branded
    ||  ehdri.e_machine != my_e_machine
    ||  ehdri.e_version != 1  // version
    ||  ehdri.e_ehsize != sizeof(ehdri)  // different <elf.h> ?
    ) {
        return false;
    }

    // additional requirements for vmlinux
    if (ehdri.e_type != Ehdr::ET_EXEC
    ||  ehdri.e_phoff != sizeof(ehdri)  // Phdr not contiguous with Ehdr
    ||  ehdri.e_phentsize!=sizeof(Phdr)
    ||  !is_valid_e_entry(ehdri.e_entry)
    ) {
        return false;
    }

    // A Linux kernel must have a __ksymtab section. [??]
    Shdr const *p, *const shstrsec = getElfSections();
    if (0==shstrsec) {
        return false;
    }
    {
        int j;
        for (p = shdri, j= ehdri.e_shnum; --j>=0; ++p) {
            if (Shdr::SHT_PROGBITS==p->sh_type
            && 0==strcmp("__ksymtab", p->sh_name + shstrtab)) {
                break;
            }
        }
        if (j < 0) {
            return false;
        }
    }

    phdri = new Phdr[(unsigned) ehdri.e_phnum];
    fi->seek(ehdri.e_phoff, SEEK_SET);
    fi->readx(phdri, ehdri.e_phnum * sizeof(*phdri));

    // Put PT_LOAD together at the beginning, ascending by .p_paddr.
    qsort(phdri, ehdri.e_phnum, sizeof(*phdri), compare_Phdr);

    // Find convex hull of physical addresses, and count the PT_LOAD.
    // Ignore ".bss": .p_filesz < .p_memsz
    unsigned phys_lo= ~0u, phys_hi= 0u;
    for (unsigned j = 0; j < ehdri.e_phnum; ++j) {
        if (Phdr::PT_LOAD==phdri[j].p_type) {
            // Check for general sanity (not necessarily required.)
            if (0xfff & (phdri[j].p_offset | phdri[j].p_paddr
                       | phdri[j].p_align  | phdri[j].p_vaddr) ) {
                return false;
            }
            if (phys_lo > phdri[j].p_paddr) {
                phys_lo = phdri[j].p_paddr;
            }
            if (phys_hi < (phdri[j].p_filesz + phdri[j].p_paddr)) {
                phys_hi = (phdri[j].p_filesz + phdri[j].p_paddr);
            }
            ++n_ptload;
        }
    }
    paddr_min = phys_lo;
    sz_ptload = phys_hi - phys_lo;
    return 0 < n_ptload;
}

#include "p_elf.h"

template <class T>
void PackVmlinuxBase<T>::pack(OutputFile *fo)
{
    unsigned fo_off = 0;
    Ehdr ehdro;
    U32 tmp_u32;

    // NULL
    // .text(PT_LOADs) .note(1st page) .note(rest)
    // .shstrtab .symtab .strtab
    Shdr shdro[1+3+3];
    memset(shdro, 0, sizeof(shdro));

    ibuf.alloc(file_size);
    obuf.allocForCompression(file_size);

    // .e_ident, .e_machine, .e_version, .e_flags
    memcpy(&ehdro, &ehdri, sizeof(ehdro));
    ehdro.e_type = Ehdr::ET_REL;
    ehdro.e_entry = 0;
    ehdro.e_phoff = 0;
    ehdro.e_shoff = sizeof(ehdro);
    ehdro.e_phentsize = 0;
    ehdro.e_phnum = 0;
    ehdro.e_shnum = 1+3+3;
    ehdro.e_shstrndx = 4;
    fo->write(&ehdro, sizeof(ehdro)); fo_off+= sizeof(ehdro);
    fo->write(shdro, sizeof(shdro)); fo_off+= sizeof(shdro);

// Notice overlap [containment] of physical PT_LOAD[2] into PTLOAD[1]
// in this vmlinux for x86_64 from Fedora Core 6 on 2007-01-07:
//Program Headers:
//  Type           Offset             VirtAddr           PhysAddr
//                 FileSiz            MemSiz              Flags  Align
//  LOAD           0x0000000000200000 0xffffffff80200000 0x0000000000200000
//                 0x000000000034bce8 0x000000000034bce8  R E    200000
//  LOAD           0x000000000054c000 0xffffffff8054c000 0x000000000054c000
//                 0x00000000000ed004 0x00000000001702a4  RWE    200000
//  LOAD           0x0000000000800000 0xffffffffff600000 0x00000000005f5000
//                 0x0000000000000c08 0x0000000000000c08  RWE    200000
//  NOTE           0x0000000000000000 0x0000000000000000 0x0000000000000000
//                 0x0000000000000000 0x0000000000000000  R      8
// Therefore we must "compose" the convex hull to be loaded.

    ph.u_len = sz_ptload;
    memset(ibuf, 0, sz_ptload);
    for (unsigned j = 0; j < ehdri.e_phnum; ++j) {
        if (Phdr::PT_LOAD==phdri[j].p_type) {
            fi->seek(phdri[j].p_offset, SEEK_SET);
            fi->readx(ibuf + ((unsigned) phdri[j].p_paddr - paddr_min), phdri[j].p_filesz);
        }
    }
    checkAlreadyPacked(ibuf + ph.u_len - 1024, 1024);

    // prepare filter
    ph.filter = 0;
    Filter ft(ph.level);
    ft.buf_len = ph.u_len;
    ft.addvalue = 0;  // we are independent of actual runtime address; see ckt32

    upx_compress_config_t cconf; cconf.reset();
    // limit stack size needed for runtime decompression
    cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28KB stack

    unsigned ppc32_extra = 0;
    if (Ehdr::EM_PPC==my_e_machine) {
        // output layout:
        //      .long UPX_MAGIC_LE32
        //      .long L20 - L10
        // L10:
        //      b_info for Ehdr; compressed Ehdr; .balign 4  // one block only
        //      b_info for LOAD; compressed LOAD; .balign 4  // possibly many blocks
        //          // This allows per-block filters!
        // L20:
        //      b f_decompress
        // +4:  f_unfilter(char *buf, unsigned len, unsigned cto8, unsigned ftid)
        //          // Code for multiple filters can "daisy chain" on ftid.
        //      f_decompress(char const *src, unsigned  src_len,
        //                   char       *dst, unsigned *dst_len, int method)
        unsigned tmp;
        tmp = UPX_MAGIC_LE32; fo->write(&tmp, sizeof(tmp)); fo_off += sizeof(tmp);
        tmp = 0;              fo->write(&tmp, sizeof(tmp)); fo_off += sizeof(tmp);
        ppc32_extra += 2*sizeof(tmp);
        unsigned const len_unc = sizeof(ehdri) + sizeof(Phdr) * ehdri.e_phnum;
        MemBuffer unc_hdr(len_unc);
        MemBuffer cpr_hdr; cpr_hdr.allocForCompression(len_unc);
        memcpy(&unc_hdr[0],             &ehdri, sizeof(ehdri));
        memcpy(&unc_hdr[sizeof(ehdri)],  phdri, sizeof(Phdr) * ehdri.e_phnum);
        unsigned len_cpr = 0;
        int const r = upx_compress(unc_hdr, len_unc, cpr_hdr, &len_cpr,
            NULL, ph.method, 10, NULL, NULL );
        if (UPX_E_OK!=r || len_unc<=len_cpr)  // FIXME: allow no compression
            throwInternalError("Ehdr compression failed");

        struct b_info { // 12-byte header before each compressed block
            unsigned sz_unc;  // uncompressed_size
            unsigned sz_cpr;  //   compressed_size
            unsigned char b_method;  // compression algorithm
            unsigned char b_ftid;  // filter id
            unsigned char b_cto8;  // filter parameter
            unsigned char b_unused;  // FIXME: !=0  for partial-block unfilter
            // unsigned f_offset, f_len;  // only if    partial-block unfilter
        }
        __attribute_packed;

        struct b_info hdr_info;
        set_be32(&hdr_info.sz_unc, len_unc);
        set_be32(&hdr_info.sz_cpr, len_cpr);
        hdr_info.b_method = ph.method;
        hdr_info.b_ftid = 0;
        hdr_info.b_cto8 = 0;
        hdr_info.b_unused = 0;
        fo->write(&hdr_info, sizeof(hdr_info)); fo_off += sizeof(hdr_info);
        unsigned const frag = (3& (0u-len_cpr));
        ppc32_extra += sizeof(hdr_info) + len_cpr + frag;
        fo_off += len_cpr + frag;
        fo->write(cpr_hdr, len_cpr + frag);

        // Partial filter: .text and following contiguous SHF_EXECINSTR
        upx_bytep f_ptr = ibuf;
        unsigned  f_len = 0;
        Shdr const *shdr = 1+ shdri;  // skip empty shdr[0]
        if (ft.buf_len==0  // not specified yet  FIXME: was set near construction
        && (Shdr::SHF_ALLOC     & shdr->sh_flags)
        && (Shdr::SHF_EXECINSTR & shdr->sh_flags)) {
            // shdr[1] is instructions (probably .text)
            f_ptr = ibuf + (unsigned) (shdr->sh_offset - phdri[0].p_offset);
            f_len = shdr->sh_size;
            ++shdr;
            for (int j= ehdri.e_shnum - 2; --j>=0; ++shdr) {
                unsigned prev_end = shdr[-1].sh_size + shdr[-1].sh_offset;
                prev_end += ~(0u-shdr[0].sh_addralign) & (0u-prev_end);  // align_up
                if ((Shdr::SHF_ALLOC     & shdr->sh_flags)
                &&  (Shdr::SHF_EXECINSTR & shdr->sh_flags)
                &&  shdr[0].sh_offset==prev_end) {
                    f_len += shdr->sh_size;
                }
                else {
                    break;
                }
            }
        }
        else { // ft.buf_len already specified, or Shdr[1] not instructions
            f_ptr = ibuf;
            f_len = ph.u_len;
        }
        compressWithFilters(ibuf, ph.u_len, obuf,
            f_ptr, f_len,  // filter range
            NULL, 0,  // hdr_ptr, hdr_len
            &ft, 512, &cconf, getStrategy(ft));

        set_be32(&hdr_info.sz_unc, ph.u_len);
        set_be32(&hdr_info.sz_cpr, ph.c_len);
        hdr_info.b_ftid = ft.id;
        hdr_info.b_cto8 = ft.cto;
        if (ph.u_len!=f_len) {
            hdr_info.b_unused = 1;  // flag for partial filter