d4337d869f
loads and stores (resp.) The ldq_u and stq_u instruction mask off the lower 3 bits of the final address before loading from or storing to the address, so as to avoid unaligned loads and stores. They do not themselves allow loads from or stores to unaligned addresses. Replace the macro definitions by a packed struct dereference. Submitted by: Richard Henderson (rth at twiddle dot net)
530 lines
15 KiB
C
530 lines
15 KiB
C
/*-
|
|
* Copyright 1996, 1997, 1998, 1999 John D. Polstra.
|
|
* All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
|
|
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
|
|
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
|
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
|
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
*
|
|
* $FreeBSD$
|
|
*/
|
|
|
|
/*
|
|
* Dynamic linker for ELF.
|
|
*
|
|
* John Polstra <jdp@polstra.com>.
|
|
*/
|
|
|
|
#include <sys/param.h>
|
|
#include <sys/mman.h>
|
|
|
|
#include <dlfcn.h>
|
|
#include <err.h>
|
|
#include <errno.h>
|
|
#include <fcntl.h>
|
|
#include <stdarg.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <unistd.h>
|
|
|
|
#include "debug.h"
|
|
#include "rtld.h"
|
|
|
|
extern Elf_Dyn _GOT_END_;
|
|
|
|
/*
|
|
* Macros for loading/storing unaligned 64-bit values. These are
|
|
* needed because relocations can point to unaligned data. This
|
|
* occurs in the DWARF2 exception frame tables generated by the
|
|
* compiler, for instance.
|
|
*
|
|
* We don't use these when relocating jump slots and GOT entries,
|
|
* since they are guaranteed to be aligned.
|
|
*/
|
|
|
|
struct ualong {
|
|
Elf_Addr x __attribute__((packed));
|
|
};
|
|
|
|
#define load64(p) (((struct ualong *)(p))->x)
|
|
#define store64(p,v) (((struct ualong *)(p))->x = (v))
|
|
|
|
/* Relocate a non-PLT object with addend. */
|
|
static int
|
|
reloc_non_plt_obj(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
|
|
SymCache *cache)
|
|
{
|
|
Elf_Addr *where = (Elf_Addr *) (obj->relocbase + rela->r_offset);
|
|
|
|
switch (ELF_R_TYPE(rela->r_info)) {
|
|
|
|
case R_ALPHA_NONE:
|
|
break;
|
|
|
|
case R_ALPHA_REFQUAD: {
|
|
const Elf_Sym *def;
|
|
const Obj_Entry *defobj;
|
|
|
|
def = find_symdef(ELF_R_SYM(rela->r_info), obj,
|
|
&defobj, false, cache);
|
|
if (def == NULL)
|
|
return -1;
|
|
store64(where,
|
|
(Elf_Addr) (defobj->relocbase + def->st_value) +
|
|
load64(where) + rela->r_addend);
|
|
}
|
|
break;
|
|
|
|
case R_ALPHA_GLOB_DAT: {
|
|
const Elf_Sym *def;
|
|
const Obj_Entry *defobj;
|
|
Elf_Addr val;
|
|
|
|
def = find_symdef(ELF_R_SYM(rela->r_info), obj,
|
|
&defobj, false, cache);
|
|
if (def == NULL)
|
|
return -1;
|
|
val = (Elf_Addr) (defobj->relocbase + def->st_value +
|
|
rela->r_addend);
|
|
if (load64(where) != val)
|
|
store64(where, val);
|
|
}
|
|
break;
|
|
|
|
case R_ALPHA_RELATIVE: {
|
|
if (obj != obj_rtld ||
|
|
(caddr_t)where < (caddr_t)_GLOBAL_OFFSET_TABLE_ ||
|
|
(caddr_t)where >= (caddr_t)&_GOT_END_)
|
|
store64(where,
|
|
load64(where) + (Elf_Addr) obj->relocbase);
|
|
}
|
|
break;
|
|
|
|
case R_ALPHA_COPY: {
|
|
/*
|
|
* These are deferred until all other relocations
|
|
* have been done. All we do here is make sure
|
|
* that the COPY relocation is not in a shared
|
|
* library. They are allowed only in executable
|
|
* files.
|
|
*/
|
|
if (!obj->mainprog) {
|
|
_rtld_error("%s: Unexpected R_COPY "
|
|
" relocation in shared library",
|
|
obj->path);
|
|
return -1;
|
|
}
|
|
}
|
|
break;
|
|
|
|
default:
|
|
_rtld_error("%s: Unsupported relocation type %d"
|
|
" in non-PLT relocations\n", obj->path,
|
|
ELF_R_TYPE(rela->r_info));
|
|
return -1;
|
|
}
|
|
return(0);
|
|
}
|
|
|
|
/* Process the non-PLT relocations. */
|
|
int
|
|
reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld)
|
|
{
|
|
const Elf_Rel *rellim;
|
|
const Elf_Rel *rel;
|
|
const Elf_Rela *relalim;
|
|
const Elf_Rela *rela;
|
|
SymCache *cache;
|
|
int bytes = obj->nchains * sizeof(SymCache);
|
|
int r = -1;
|
|
|
|
/*
|
|
* The dynamic loader may be called from a thread, we have
|
|
* limited amounts of stack available so we cannot use alloca().
|
|
*/
|
|
cache = mmap(NULL, bytes, PROT_READ|PROT_WRITE, MAP_ANON, -1, 0);
|
|
if (cache == MAP_FAILED)
|
|
cache = NULL;
|
|
|
|
/* Perform relocations without addend if there are any: */
|
|
rellim = (const Elf_Rel *) ((caddr_t) obj->rel + obj->relsize);
|
|
for (rel = obj->rel; obj->rel != NULL && rel < rellim; rel++) {
|
|
Elf_Rela locrela;
|
|
|
|
locrela.r_info = rel->r_info;
|
|
locrela.r_offset = rel->r_offset;
|
|
locrela.r_addend = 0;
|
|
if (reloc_non_plt_obj(obj_rtld, obj, &locrela, cache))
|
|
goto done;
|
|
}
|
|
|
|
/* Perform relocations with addend if there are any: */
|
|
relalim = (const Elf_Rela *) ((caddr_t) obj->rela + obj->relasize);
|
|
for (rela = obj->rela; obj->rela != NULL && rela < relalim; rela++) {
|
|
if (reloc_non_plt_obj(obj_rtld, obj, rela, cache))
|
|
goto done;
|
|
}
|
|
r = 0;
|
|
done:
|
|
if (cache)
|
|
munmap(cache, bytes);
|
|
return(r);
|
|
}
|
|
|
|
/* Process the PLT relocations. */
|
|
int
|
|
reloc_plt(Obj_Entry *obj)
|
|
{
|
|
/* All PLT relocations are the same kind: either Elf_Rel or Elf_Rela. */
|
|
if (obj->pltrelsize != 0) {
|
|
const Elf_Rel *rellim;
|
|
const Elf_Rel *rel;
|
|
|
|
rellim = (const Elf_Rel *)((char *)obj->pltrel + obj->pltrelsize);
|
|
for (rel = obj->pltrel; rel < rellim; rel++) {
|
|
Elf_Addr *where;
|
|
|
|
assert(ELF_R_TYPE(rel->r_info) == R_ALPHA_JMP_SLOT);
|
|
|
|
/* Relocate the GOT slot pointing into the PLT. */
|
|
where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
|
|
*where += (Elf_Addr)obj->relocbase;
|
|
}
|
|
} else {
|
|
const Elf_Rela *relalim;
|
|
const Elf_Rela *rela;
|
|
|
|
relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize);
|
|
for (rela = obj->pltrela; rela < relalim; rela++) {
|
|
Elf_Addr *where;
|
|
|
|
assert(ELF_R_TYPE(rela->r_info) == R_ALPHA_JMP_SLOT);
|
|
|
|
/* Relocate the GOT slot pointing into the PLT. */
|
|
where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
|
|
*where += (Elf_Addr)obj->relocbase;
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
/* Relocate the jump slots in an object. */
|
|
int
|
|
reloc_jmpslots(Obj_Entry *obj)
|
|
{
|
|
if (obj->jmpslots_done)
|
|
return 0;
|
|
/* All PLT relocations are the same kind: either Elf_Rel or Elf_Rela. */
|
|
if (obj->pltrelsize != 0) {
|
|
const Elf_Rel *rellim;
|
|
const Elf_Rel *rel;
|
|
|
|
rellim = (const Elf_Rel *)((char *)obj->pltrel + obj->pltrelsize);
|
|
for (rel = obj->pltrel; rel < rellim; rel++) {
|
|
Elf_Addr *where;
|
|
const Elf_Sym *def;
|
|
const Obj_Entry *defobj;
|
|
|
|
assert(ELF_R_TYPE(rel->r_info) == R_ALPHA_JMP_SLOT);
|
|
where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
|
|
def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, true,
|
|
NULL);
|
|
if (def == NULL)
|
|
return -1;
|
|
reloc_jmpslot(where,
|
|
(Elf_Addr)(defobj->relocbase + def->st_value),
|
|
defobj, obj, rel);
|
|
}
|
|
} else {
|
|
const Elf_Rela *relalim;
|
|
const Elf_Rela *rela;
|
|
|
|
relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize);
|
|
for (rela = obj->pltrela; rela < relalim; rela++) {
|
|
Elf_Addr *where;
|
|
const Elf_Sym *def;
|
|
const Obj_Entry *defobj;
|
|
|
|
assert(ELF_R_TYPE(rela->r_info) == R_ALPHA_JMP_SLOT);
|
|
where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
|
|
def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, true,
|
|
NULL);
|
|
if (def == NULL)
|
|
return -1;
|
|
reloc_jmpslot(where,
|
|
(Elf_Addr)(defobj->relocbase + def->st_value),
|
|
defobj, obj, (Elf_Rel *)rela);
|
|
}
|
|
}
|
|
obj->jmpslots_done = true;
|
|
return 0;
|
|
}
|
|
|
|
/* Fixup the jump slot at "where" to transfer control to "target". */
|
|
Elf_Addr
|
|
reloc_jmpslot(Elf_Addr *where, Elf_Addr target, const Obj_Entry *obj,
|
|
const Obj_Entry *refobj, const Elf_Rel *rel)
|
|
{
|
|
Elf_Addr stubaddr;
|
|
|
|
dbg(" reloc_jmpslot: where=%p, target=%p", (void *)where, (void *)target);
|
|
stubaddr = *where;
|
|
if (stubaddr != target) {
|
|
int64_t delta;
|
|
u_int32_t inst[3];
|
|
int instct;
|
|
Elf_Addr pc;
|
|
int64_t idisp;
|
|
u_int32_t *stubptr;
|
|
|
|
/* Point this GOT entry directly at the target. */
|
|
*where = target;
|
|
|
|
/*
|
|
* There may be multiple GOT tables, each with an entry
|
|
* pointing to the stub in the PLT. But we can only find and
|
|
* fix up the first GOT entry. So we must rewrite the stub as
|
|
* well, to perform a call to the target if it is executed.
|
|
*
|
|
* When the stub gets control, register pv ($27) contains its
|
|
* address. We adjust its value so that it points to the
|
|
* target, and then jump indirect through it.
|
|
*
|
|
* Each PLT entry has room for 3 instructions. If the
|
|
* adjustment amount fits in a signed 32-bit integer, we can
|
|
* simply add it to register pv. Otherwise we must load the
|
|
* GOT entry itself into the pv register.
|
|
*/
|
|
delta = target - stubaddr;
|
|
dbg(" stubaddr=%p, where-stubaddr=%ld, delta=%ld", (void *)stubaddr,
|
|
(long)where - (long)stubaddr, (long)delta);
|
|
instct = 0;
|
|
if ((int32_t)delta == delta) {
|
|
/*
|
|
* We can adjust pv with a LDA, LDAH sequence.
|
|
*
|
|
* First build an LDA instruction to adjust the low 16 bits.
|
|
*/
|
|
inst[instct++] = 0x08 << 26 | 27 << 21 | 27 << 16 |
|
|
(delta & 0xffff);
|
|
dbg(" LDA $27,%d($27)", (int16_t)delta);
|
|
/*
|
|
* Adjust the delta to account for the effects of the LDA,
|
|
* including sign-extension.
|
|
*/
|
|
delta -= (int16_t)delta;
|
|
if (delta != 0) {
|
|
/* Build an LDAH instruction to adjust the high 16 bits. */
|
|
inst[instct++] = 0x09 << 26 | 27 << 21 | 27 << 16 |
|
|
(delta >> 16 & 0xffff);
|
|
dbg(" LDAH $27,%d($27)", (int16_t)(delta >> 16));
|
|
}
|
|
} else {
|
|
int64_t dhigh;
|
|
|
|
/* We must load the GOT entry from memory. */
|
|
delta = (Elf_Addr)where - stubaddr;
|
|
/*
|
|
* If the GOT entry is too far away from the PLT entry,
|
|
* then punt. This PLT entry will have to be looked up
|
|
* manually for all GOT entries except the first one.
|
|
* The program will still run, albeit very slowly. It's
|
|
* extremely unlikely that this case could ever arise in
|
|
* practice, but we might as well handle it correctly if
|
|
* it does.
|
|
*/
|
|
if ((int32_t)delta != delta) {
|
|
dbg(" PLT stub too far from GOT to relocate");
|
|
return target;
|
|
}
|
|
dhigh = delta - (int16_t)delta;
|
|
if (dhigh != 0) {
|
|
/* Build an LDAH instruction to adjust the high 16 bits. */
|
|
inst[instct++] = 0x09 << 26 | 27 << 21 | 27 << 16 |
|
|
(dhigh >> 16 & 0xffff);
|
|
dbg(" LDAH $27,%d($27)", (int16_t)(dhigh >> 16));
|
|
}
|
|
/* Build an LDQ to load the GOT entry. */
|
|
inst[instct++] = 0x29 << 26 | 27 << 21 | 27 << 16 |
|
|
(delta & 0xffff);
|
|
dbg(" LDQ $27,%d($27)", (int16_t)delta);
|
|
}
|
|
|
|
/*
|
|
* Build a JMP or BR instruction to jump to the target. If
|
|
* the instruction displacement fits in a sign-extended 21-bit
|
|
* field, we can use the more efficient BR instruction.
|
|
* Otherwise we have to jump indirect through the pv register.
|
|
*/
|
|
pc = stubaddr + 4 * (instct + 1);
|
|
idisp = (int64_t)(target - pc) >> 2;
|
|
if (-0x100000 <= idisp && idisp < 0x100000) {
|
|
inst[instct++] = 0x30 << 26 | 31 << 21 | (idisp & 0x1fffff);
|
|
dbg(" BR $31,%p", (void *)target);
|
|
} else {
|
|
inst[instct++] = 0x1a << 26 | 31 << 21 | 27 << 16 |
|
|
(idisp & 0x3fff);
|
|
dbg(" JMP $31,($27),%d", (int)(idisp & 0x3fff));
|
|
}
|
|
|
|
/*
|
|
* Fill in the tail of the PLT entry first for reentrancy.
|
|
* Until we have overwritten the first instruction (an
|
|
* unconditional branch), the remaining instructions have no
|
|
* effect.
|
|
*/
|
|
stubptr = (u_int32_t *)stubaddr;
|
|
while (instct > 1) {
|
|
instct--;
|
|
stubptr[instct] = inst[instct];
|
|
}
|
|
/*
|
|
* Commit the tail of the instruction sequence to memory
|
|
* before overwriting the first instruction.
|
|
*/
|
|
__asm__ __volatile__("wmb" : : : "memory");
|
|
stubptr[0] = inst[0];
|
|
}
|
|
|
|
return target;
|
|
}
|
|
|
|
/* Process an R_ALPHA_COPY relocation. */
|
|
static int
|
|
do_copy_relocation(Obj_Entry *dstobj, const Elf_Rela *rela)
|
|
{
|
|
void *dstaddr;
|
|
const Elf_Sym *dstsym;
|
|
const char *name;
|
|
unsigned long hash;
|
|
size_t size;
|
|
const void *srcaddr;
|
|
const Elf_Sym *srcsym;
|
|
Obj_Entry *srcobj;
|
|
|
|
dstaddr = (void *) (dstobj->relocbase + rela->r_offset);
|
|
dstsym = dstobj->symtab + ELF_R_SYM(rela->r_info);
|
|
name = dstobj->strtab + dstsym->st_name;
|
|
hash = elf_hash(name);
|
|
size = dstsym->st_size;
|
|
|
|
for (srcobj = dstobj->next; srcobj != NULL; srcobj = srcobj->next)
|
|
if ((srcsym = symlook_obj(name, hash, srcobj, false)) != NULL)
|
|
break;
|
|
|
|
if (srcobj == NULL) {
|
|
_rtld_error("Undefined symbol \"%s\" referenced from COPY"
|
|
" relocation in %s", name, dstobj->path);
|
|
return -1;
|
|
}
|
|
|
|
srcaddr = (const void *) (srcobj->relocbase + srcsym->st_value);
|
|
memcpy(dstaddr, srcaddr, size);
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Process the special R_ALPHA_COPY relocations in the main program. These
|
|
* copy data from a shared object into a region in the main program's BSS
|
|
* segment.
|
|
*
|
|
* Returns 0 on success, -1 on failure.
|
|
*/
|
|
int
|
|
do_copy_relocations(Obj_Entry *dstobj)
|
|
{
|
|
const Elf_Rel *rellim;
|
|
const Elf_Rel *rel;
|
|
const Elf_Rela *relalim;
|
|
const Elf_Rela *rela;
|
|
|
|
assert(dstobj->mainprog); /* COPY relocations are invalid elsewhere */
|
|
|
|
rellim = (const Elf_Rel *) ((caddr_t) dstobj->rel + dstobj->relsize);
|
|
for (rel = dstobj->rel; dstobj->rel != NULL && rel < rellim; rel++) {
|
|
if (ELF_R_TYPE(rel->r_info) == R_ALPHA_COPY) {
|
|
Elf_Rela locrela;
|
|
|
|
locrela.r_info = rel->r_info;
|
|
locrela.r_offset = rel->r_offset;
|
|
locrela.r_addend = 0;
|
|
if (do_copy_relocation(dstobj, &locrela))
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
relalim = (const Elf_Rela *) ((caddr_t) dstobj->rela +
|
|
dstobj->relasize);
|
|
for (rela = dstobj->rela; dstobj->rela != NULL && rela < relalim;
|
|
rela++) {
|
|
if (ELF_R_TYPE(rela->r_info) == R_ALPHA_COPY) {
|
|
if (do_copy_relocation(dstobj, rela))
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
/* Initialize the special PLT entries. */
|
|
void
|
|
init_pltgot(Obj_Entry *obj)
|
|
{
|
|
u_int32_t *pltgot;
|
|
|
|
if (obj->pltgot != NULL &&
|
|
(obj->pltrelsize != 0 || obj->pltrelasize != 0)) {
|
|
/*
|
|
* This function will be called to perform the relocation.
|
|
* Look for the ldah instruction from the old PLT format since
|
|
* that will tell us what format we are trying to relocate.
|
|
*/
|
|
pltgot = (u_int32_t *) obj->pltgot;
|
|
if ((pltgot[8] & 0xffff0000) == 0x279f0000)
|
|
obj->pltgot[2] = (Elf_Addr) &_rtld_bind_start_old;
|
|
else
|
|
obj->pltgot[2] = (Elf_Addr) &_rtld_bind_start;
|
|
/* Identify this shared object */
|
|
obj->pltgot[3] = (Elf_Addr) obj;
|
|
}
|
|
}
|
|
|
|
void
|
|
allocate_initial_tls(Obj_Entry *list)
|
|
{
|
|
void *tls;
|
|
|
|
/*
|
|
* Fix the size of the static TLS block by using the maximum
|
|
* offset allocated so far and adding a bit for dynamic modules to
|
|
* use.
|
|
*/
|
|
tls_static_space = tls_last_offset + tls_last_size + RTLD_STATIC_TLS_EXTRA;
|
|
tls = allocate_tls(list, 0, 16, 16);
|
|
alpha_pal_wrunique((u_int64_t) tls);
|
|
}
|
|
|
|
void *__tls_get_addr(tls_index* ti)
|
|
{
|
|
Elf_Addr** tp = (Elf_Addr**) alpha_pal_rdunique();
|
|
|
|
return tls_get_addr_common(tp, ti->ti_module, ti->ti_offset);
|
|
}
|