/* First tried something like this, but we hit too many issues in decode and encode */
bool
compare_pages(void *drcontext, byte *start1, byte *start2)
{
byte *p1 = start1, *p2 = start2;
int skipped_bytes = 0, identical_skipped_bytes = 0;
while (p1 < start1 + PAGE_SIZE) {
int instr_size = decode_sizeof(drcontext, p1, NULL _IF_X64(NULL));
if (p1 + instr_size > start1 + PAGE_SIZE) {
/* We're overlapping the end of the page, skip these. */
int end_skip = start1 + PAGE_SIZE - p1;
VVERBOSE_PRINT("Passing PAGE_END %d bytes", end_skip);
skipped_bytes += end_skip;
if (memcmp(p1, p2, end_skip) == 0)
identical_skipped_bytes += end_skip;
break;
}
if (decode_sizeof(drcontext, p2, NULL _IF_X64(NULL)) != instr_size) {
VVERBOSE_PRINT("Instruction alignment mismatch\n");
return false;
}
/* assumption - instructions <= 4 bytes in size won't have relocations */
if (instr_size < 5) {
if (memcmp(p1, p2, instr_size) != 0) {
VVERBOSE_PRINT("Difference found in small instr\n");
return false;
}
p1 += size;
p2 += size;
} else {
/* guess if there could be a relocation */
instr_t *instr1 = instr_create(drcontext);
instr_t *instr2 = instr_create(drcontext);
p1 = decode(drcontext, p1, instr1);
p2 = decode(drcontext, p2, instr2);
if (p1 - start1 != p2 - start2) {
VVERBOSE_PRINT("Instruction alignment mismatch on full decode\n");
/* Fixme - free instr, don't expect this to happen */
return false;
}
if (instr_get_num_srcs(instr1) != instr_get_num_srcs(instr2) ||
instr_get_num_dsts(instr1) != instr_get_num_dsts(instr2)) {
VVERBOSE_PRINT("Full decode operand mismatch");
return false;
}
for (i = instr_get_num_srcs(instr1); i > 0; i--) {
opnd_t opnd = instr_get_src(instr1, i);
if (opnd_is_immed_int(opnd) && opnd_get_immed_int(opnd) > 0x10000) {
instr_set_src(instr1, i, opnd_create_immed_int(opnd_get_immed_int(opnd), opnd_get_size(opnd)));
}
}
}
}
}
/* PR 332254: test xchg vs nop */
static void
test_nop_xchg(void *dc)
{
/* 0x0000000000671460 87 c0 xchg %eax %eax -> %eax %eax
* 0x0000000000671460 48 87 c0 xchg %rax %rax -> %rax %rax
* 0x0000000000671460 41 87 c0 xchg %r8d %eax -> %r8d %eax
* 0x0000000000671460 46 90 nop
* 0x0000000000671460 4e 90 nop
* 0x0000000000671460 41 90 xchg %r8d %eax -> %r8d %eax
*/
instr_t *instr;
instr = INSTR_CREATE_xchg(dc, opnd_create_reg(REG_EAX), opnd_create_reg(REG_EAX));
test_instr_encode(dc, instr, 2);
#ifdef X64
/* we don't do the optimal "48 90" instead of "48 87 c0" */
instr = INSTR_CREATE_xchg(dc, opnd_create_reg(REG_RAX), opnd_create_reg(REG_RAX));
test_instr_encode(dc, instr, 3);
/* we don't do the optimal "41 90" instead of "41 87 c0" */
instr = INSTR_CREATE_xchg(dc, opnd_create_reg(REG_R8D), opnd_create_reg(REG_EAX));
test_instr_encode(dc, instr, 3);
/* ensure we treat as nop and NOT xchg if doesn't have rex.b */
buf[0] = 0x46;
buf[1] = 0x90;
instr = instr_create(dc);
# if VERBOSE
disassemble_with_info(dc, buf, STDOUT, true, true);
# endif
decode(dc, buf, instr);
ASSERT(instr_get_opcode(instr) == OP_nop);
instr_destroy(dc, instr);
buf[0] = 0x4e;
buf[1] = 0x90;
instr = instr_create(dc);
# if VERBOSE
disassemble_with_info(dc, buf, STDOUT, true, true);
# endif
decode(dc, buf, instr);
ASSERT(instr_get_opcode(instr) == OP_nop);
instr_destroy(dc, instr);
buf[0] = 0x41;
buf[1] = 0x90;
instr = instr_create(dc);
# if VERBOSE
disassemble_with_info(dc, buf, STDOUT, true, true);
# endif
decode(dc, buf, instr);
ASSERT(instr_get_opcode(instr) == OP_xchg);
instr_destroy(dc, instr);
#endif
}
void memory_write(void *pc, void *prev_pc)
{
void *drcontext = dr_get_current_drcontext();
instr_t *instr = instr_create(drcontext);
dr_mcontext_t mctx;
opnd_t dst;
ctx_t ctx;
pc = dr_app_pc_for_decoding(pc);
mctx.flags = DR_MC_CONTROL|DR_MC_INTEGER;
mctx.size = sizeof(mctx);
dr_get_mcontext(drcontext, &mctx);
instr_init(drcontext, instr);
if (!decode(drcontext, pc, instr))
{
dr_printf("Decode of instruction at %p failed\n", pc);
return;
}
ctx.addr = prev_pc;
ctx.dr_addr = prev_pc;
for (int i = 0; i < instr_num_dsts(instr); i++)
{
dst = instr_get_dst(instr, i);
check_opnd(dst, pc, 0, drcontext, &mctx, &ctx);
}
instr_destroy(drcontext, instr);
}
/* emits the instruction to buf (for tests that wish to do additional checks on
* the output) */
static void
test_instr_encode_and_decode(void *dc, instr_t *instr, uint len_expect,
/* also checks one operand's size */
bool src, uint opnum, opnd_size_t sz, uint bytes)
{
opnd_t op;
opnd_size_t opsz;
instr_t *decin;
uint len;
byte *pc = instr_encode(dc, instr, buf);
len = (int) (pc - (byte *)buf);
#if VERBOSE
disassemble_with_info(dc, buf, STDOUT, true, true);
#endif
ASSERT(len == len_expect);
decin = instr_create(dc);
decode(dc, buf, decin);
ASSERT(instr_same(instr, decin));
/* PR 245805: variable sizes should be resolved on decode */
if (src)
op = instr_get_src(decin, opnum);
else
op = instr_get_dst(decin, opnum);
opsz = opnd_get_size(op);
ASSERT(opsz == sz && opnd_size_in_bytes(opsz) == bytes);
instr_destroy(dc, instr);
instr_destroy(dc, decin);
}
/* returns false on failure */
static bool
decode_function(void *dcontext, byte *entry)
{
byte *pc, *pre_pc;
int num_instr = 0;
bool found_ret = false;
instr_t *instr;
if (entry == NULL)
return false;
instr = instr_create(dcontext);
pc = entry;
while (true) {
instr_reset(dcontext, instr);
pre_pc = pc;
pc = decode(dcontext, pc, instr);
instr_set_translation(instr, pre_pc);
dr_print_instr(dcontext, STDOUT, instr, "");
if (instr_is_return(instr)) {
found_ret = true;
break;
}
num_instr++;
if (num_instr > MAX_INSTRS_IN_FUNCTION) {
print("ERROR: hit max instr limit %d\n", MAX_INSTRS_IN_FUNCTION);
break;
}
}
instr_destroy(dcontext, instr);
return found_ret;
}
/* Decode instruction from callee and return the next_pc to be decoded. */
static app_pc
decode_callee_instr(dcontext_t *dcontext, callee_info_t *ci, app_pc instr_pc)
{
instrlist_t *ilist = ci->ilist;
instr_t *instr;
app_pc next_pc = NULL;
instr = instr_create(GLOBAL_DCONTEXT);
instrlist_append(ilist, instr);
ci->num_instrs++;
TRY_EXCEPT(dcontext, { next_pc = decode(GLOBAL_DCONTEXT, instr_pc, instr); },
{ /* EXCEPT */
/* emits the instruction to buf (for tests that wish to do additional checks on
* the output) */
static void
test_instr_encode(void *dc, instr_t *instr, uint len_expect)
{
instr_t *decin;
uint len;
byte *pc = instr_encode(dc, instr, buf);
len = (int) (pc - (byte *)buf);
#if VERBOSE
disassemble_with_info(dc, buf, STDOUT, true, true);
#endif
ASSERT(len == len_expect);
decin = instr_create(dc);
decode(dc, buf, decin);
ASSERT(instr_same(instr, decin));
instr_destroy(dc, instr);
instr_destroy(dc, decin);
}
static void
look_for_usercall(void *dcontext, byte *entry, const char *sym, LOADED_IMAGE *img,
const char *modpath)
{
bool found_push_imm = false;
int imm = 0;
app_pc pc, pre_pc;
instr_t *instr;
if (entry == NULL)
return;
instr = instr_create(dcontext);
pc = entry;
while (true) {
instr_reset(dcontext, instr);
pre_pc = pc;
pc = decode(dcontext, pc, instr);
if (verbose) {
instr_set_translation(instr, pre_pc);
dr_print_instr(dcontext, STDOUT, instr, "");
}
if (pc == NULL || !instr_valid(instr))
break;
if (instr_get_opcode(instr) == OP_push_imm) {
found_push_imm = true;
imm = (int) opnd_get_immed_int(instr_get_src(instr, 0));
} else if (instr_is_call_direct(instr) && found_push_imm) {
app_pc tgt = opnd_get_pc(instr_get_target(instr));
bool found = false;
int i;
for (i = 0; i < NUM_USERCALL; i++) {
if (tgt == usercall_addr[i]) {
dr_printf("Call #0x%02x to %s at %s+0x%x\n", imm, usercall_names[i],
sym, pre_pc - entry);
found = true;
break;
}
}
if (found)
break;
} else if (instr_is_return(instr))
break;
if (pc - entry > MAX_BYTES_BEFORE_USERCALL)
break;
}
instr_destroy(dcontext, instr);
}
// return the addr of the previous instructions
// a bad addr can be return in some specific case but generally work good
void *get_prev_instr_pc(void *pc, void *drc)
{
instr_t *instr = instr_create(drc);
void *tmp_pc;
for (int ct = 1; ; ct++)
{
tmp_pc = dr_app_pc_for_decoding(pc - ct);
if (decode(drc, tmp_pc, instr))
{
if (instr_is_call(instr) && decode_next_pc(drc, tmp_pc) == pc)
break;
}
instr_reuse(drc, instr);
}
instr_destroy(drc, instr);
return tmp_pc;
}
static void
test_instr_opnds(void *dc)
{
/* Verbose disasm looks like this:
* 32-bit:
* 0x080f1ae0 ff 25 e7 1a 0f 08 jmp 0x080f1ae7
* 0x080f1ae6 b8 ef be ad de mov $0xdeadbeef -> %eax
* 0x080f1ae0 a0 e6 1a 0f 08 mov 0x080f1ae6 -> %al
* 0x080f1ae5 b8 ef be ad de mov $0xdeadbeef -> %eax
* 64-bit:
* 0x00000000006b8de0 ff 25 02 00 00 00 jmp <rel> 0x00000000006b8de8
* 0x00000000006b8de6 48 b8 ef be ad de 00 mov $0x00000000deadbeef -> %rax
* 00 00 00
* 0x00000000006b8de0 8a 05 02 00 00 00 mov <rel> 0x00000000006b8de8 -> %al
* 0x00000000006b8de6 48 b8 ef be ad de 00 mov $0x00000000deadbeef -> %rax
* 00 00 00
*/
instrlist_t *ilist;
instr_t *tgt, *instr;
byte *pc;
short disp;
ilist = instrlist_create(dc);
/* test mem instr as ind jmp target */
tgt = INSTR_CREATE_mov_imm(dc, opnd_create_reg(DR_REG_XAX),
opnd_create_immed_int(0xdeadbeef, OPSZ_PTR));
/* skip rex+opcode */
disp = IF_X64_ELSE(2,1);
instrlist_append(ilist, INSTR_CREATE_jmp_ind
(dc, opnd_create_mem_instr(tgt, disp, OPSZ_PTR)));
instrlist_append(ilist, tgt);
pc = instrlist_encode(dc, ilist, buf, true/*instr targets*/);
ASSERT(pc != NULL);
instrlist_clear(dc, ilist);
#if VERBOSE
pc = disassemble_with_info(dc, buf, STDOUT, true, true);
pc = disassemble_with_info(dc, pc, STDOUT, true, true);
#endif
pc = buf;
instr = instr_create(dc);
pc = decode(dc, pc, instr);
ASSERT(pc != NULL);
ASSERT(instr_get_opcode(instr) == OP_jmp_ind);
#ifdef X64
ASSERT(opnd_is_rel_addr(instr_get_src(instr, 0)));
ASSERT(opnd_get_addr(instr_get_src(instr, 0)) == pc + disp);
#else
ASSERT(opnd_is_base_disp(instr_get_src(instr, 0)));
ASSERT(opnd_get_base(instr_get_src(instr, 0)) == REG_NULL);
ASSERT(opnd_get_index(instr_get_src(instr, 0)) == REG_NULL);
ASSERT(opnd_get_disp(instr_get_src(instr, 0)) == (ptr_int_t)pc + disp);
#endif
/* test mem instr as TYPE_O */
tgt = INSTR_CREATE_mov_imm(dc, opnd_create_reg(DR_REG_XAX),
opnd_create_immed_int(0xdeadbeef, OPSZ_PTR));
/* skip rex+opcode */
disp = IF_X64_ELSE(2,1);
instrlist_append(ilist, INSTR_CREATE_mov_ld
(dc, opnd_create_reg(DR_REG_AL),
opnd_create_mem_instr(tgt, disp, OPSZ_1)));
instrlist_append(ilist, tgt);
pc = instrlist_encode(dc, ilist, buf, true/*instr targets*/);
ASSERT(pc != NULL);
instrlist_clear(dc, ilist);
#if VERBOSE
pc = disassemble_with_info(dc, buf, STDOUT, true, true);
pc = disassemble_with_info(dc, pc, STDOUT, true, true);
#endif
pc = buf;
instr_reset(dc, instr);
pc = decode(dc, pc, instr);
ASSERT(pc != NULL);
ASSERT(instr_get_opcode(instr) == OP_mov_ld);
#ifdef X64
ASSERT(opnd_is_rel_addr(instr_get_src(instr, 0)));
ASSERT(opnd_get_addr(instr_get_src(instr, 0)) == pc + disp);
#else
ASSERT(opnd_is_base_disp(instr_get_src(instr, 0)));
ASSERT(opnd_get_base(instr_get_src(instr, 0)) == REG_NULL);
ASSERT(opnd_get_index(instr_get_src(instr, 0)) == REG_NULL);
ASSERT(opnd_get_disp(instr_get_src(instr, 0)) == (ptr_int_t)pc + disp);
#endif
instr_free(dc, instr);
instrlist_destroy(dc, ilist);
}
/* make sure the following are consistent (though they could still all be wrong :))
* with respect to instr length and opcode:
* - decode_fast
* - decode
* - INSTR_CREATE_
* - encode
*/
static void
test_all_opcodes(void *dc)
{
byte *pc, *next_pc;
byte *end;
instrlist_t *ilist = instrlist_create(dc);
instr_t *instr;
/* we cannot pass on variadic args as separate args to another
* macro, so we must split ours by # args (xref PR 208603)
*/
# define MEMARG(sz) (opnd_create_base_disp(REG_XCX, REG_NULL, 0, 0x37, sz))
# define IMMARG(sz) opnd_create_immed_int(37, sz)
# define TGTARG opnd_create_instr(instrlist_last(ilist))
# define REGARG(reg) opnd_create_reg(REG_##reg)
# define X86_ONLY 1
# define X64_ONLY 2
# define OPCODE(opc, icnm, ...) \
int len_##icnm;
# include "ir_0args.h"
# include "ir_1args.h"
# include "ir_2args.h"
# include "ir_3args.h"
# include "ir_4args.h"
# undef OPCODE
/* we can encode+fast-decode some instrs cross-platform but we
* leave that testing to the regression run on that platform */
# define OPCODE(opc, icnm, flags) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0) { \
instrlist_append(ilist, INSTR_CREATE_##icnm(dc)); \
len_##icnm = instr_length(dc, instrlist_last(ilist)); \
} } while (0);
# include "ir_0args.h"
# undef OPCODE
# define OPCODE(opc, icnm, flags, arg1) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0) { \
instrlist_append(ilist, INSTR_CREATE_##icnm(dc, arg1)); \
len_##icnm = instr_length(dc, instrlist_last(ilist)); \
} } while (0);
# include "ir_1args.h"
# undef OPCODE
# define OPCODE(opc, icnm, flags, arg1, arg2) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0) { \
instrlist_append(ilist, INSTR_CREATE_##icnm(dc, arg1, arg2)); \
len_##icnm = instr_length(dc, instrlist_last(ilist)); \
} } while (0);
# include "ir_2args.h"
# undef OPCODE
# define OPCODE(opc, icnm, flags, arg1, arg2, arg3) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0) { \
instrlist_append(ilist, INSTR_CREATE_##icnm(dc, arg1, arg2, arg3)); \
len_##icnm = instr_length(dc, instrlist_last(ilist)); \
} } while (0);
# include "ir_3args.h"
# undef OPCODE
# define OPCODE(opc, icnm, flags, arg1, arg2, arg3, arg4) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0) { \
instrlist_append(ilist, INSTR_CREATE_##icnm(dc, arg1, arg2, arg3, arg4)); \
len_##icnm = instr_length(dc, instrlist_last(ilist)); \
} } while (0);
# include "ir_4args.h"
# undef OPCODE
end = instrlist_encode(dc, ilist, buf, false);
instr = instr_create(dc);
pc = buf;
# define OPCODE(opc, icnm, flags, ...) do { \
if ((flags & IF_X64_ELSE(X86_ONLY, X64_ONLY)) == 0 && len_##icnm != 0) { \
instr_reset(dc, instr); \
next_pc = decode(dc, pc, instr); \
ASSERT((next_pc - pc) == decode_sizeof(dc, pc, NULL _IF_X64(NULL))); \
ASSERT((next_pc - pc) == len_##icnm); \
ASSERT(instr_get_opcode(instr) == OP_##opc); \
pc = next_pc; \
} } while (0);
# include "ir_0args.h"
# include "ir_1args.h"
# include "ir_2args.h"
# include "ir_3args.h"
# include "ir_4args.h"
# undef OPCODE
#if VERBOSE
for (pc = buf; pc < end; )
pc = disassemble_with_info(dc, pc, STDOUT, true, true);
#endif
instr_destroy(dc, instr);
instrlist_clear_and_destroy(dc, ilist);
}
void Shade::detour(void *address, void *target, void *&trampoline)
{
const size_t instr_max = 17;
auto list = instrlist_create(dr);
byte instr_data[instr_max];
byte *current = (byte *)address;
byte *min_pos = (byte *)address + 5;
size_t size = 0;
while(current < min_pos)
{
read(current, instr_data, instr_max);
auto instr = instr_create(dr);
byte *decoded = decode_from_copy(dr, instr_data, current, instr);
if(!decoded)
error("Unknown instruction");
instrlist_append(list, instr);
instr_make_persistent(dr, instr);
current += (size_t)(decoded - instr_data);
size += instr_length(dr, instr);
}
auto instr = INSTR_CREATE_jmp(dr, opnd_create_pc(current));
size += instr_length(dr, instr);
instrlist_append(list, instr);
auto local_trampoline = alloca(size);
if(!local_trampoline)
error("Out of memory");
void *remote = code_section.allocate(size, 4);
if(!instrlist_encode_to_copy(dr, list, (byte *)local_trampoline, (byte *)remote, 0, true))
error("Unable to encode instructions");
instrlist_clear_and_destroy(dr, list);
write(remote, local_trampoline, size);
trampoline = remote;
char code[5];
DWORD offset = (size_t)target - (size_t)address - 5;
code[0] = 0xE9;
*(DWORD *)(code + 1) = offset;
access(address, 5, [&] {
write(address, code, 5);
});
}
/* returns false on failure */
static bool
decode_syscall_num(void *dcontext, byte *entry, syscall_info_t *info, LOADED_IMAGE *img)
{
/* FIXME: would like to fail gracefully rather than have a DR assertion
* on non-code! => use DEBUG=0 INTERNAL=1 DR build!
*/
bool found_syscall = false, found_eax = false, found_edx = false, found_ecx = false;
bool found_ret = false;
byte *pc, *pre_pc;
int num_instr = 0;
instr_t *instr;
byte *preferred = get_preferred_base(img);
if (entry == NULL)
return false;
info->num_args = -1; /* if find sysnum but not args */
info->sysnum = -1;
info->fixup_index = -1;
instr = instr_create(dcontext);
pc = entry;
/* FIXME - we don't support decoding 64bit instructions in 32bit mode, but I want
* this to work on 32bit machines. Hack fix based on the wrapper pattern, we skip
* the first instruction (mov r10, rcx) here, the rest should decode ok.
* Xref PR 236203. */
if (expect_x64 && *pc == 0x4c && *(pc+1) == 0x8b && *(pc+2) == 0xd1)
pc += 3;
while (true) {
instr_reset(dcontext, instr);
pre_pc = pc;
pc = decode(dcontext, pc, instr);
if (verbose) {
instr_set_translation(instr, pre_pc);
dr_print_instr(dcontext, STDOUT, instr, "");
}
if (pc == NULL || !instr_valid(instr))
break;
if (instr_is_syscall(instr) || instr_is_call_indirect(instr)) {
/* If we see a syscall instr or an indirect call which is not syscall,
* we assume this is not a syscall wrapper.
*/
found_syscall = process_syscall_instr(dcontext, instr, found_eax, found_edx);
if (!found_syscall)
break; /* assume not a syscall wrapper, give up gracefully */
} else if (instr_is_return(instr)) {
/* we must break on return to avoid case like win8 x86
* which has sysenter callee adjacent-"inlined"
* ntdll!NtYieldExecution:
* 77d7422c b801000000 mov eax,1
* 77d74231 e801000000 call ntdll!NtYieldExecution+0xb (77d74237)
* 77d74236 c3 ret
* 77d74237 8bd4 mov edx,esp
* 77d74239 0f34 sysenter
* 77d7423b c3 ret
*/
if (!found_ret) {
process_ret(instr, info);
found_ret = true;
}
break;
} else if (instr_get_opcode(instr) == OP_call) {
found_syscall = process_syscall_call(dcontext, pc, instr,
found_eax, found_edx);
/* If we see a call and it is not a sysenter callee,
* we assume this is not a syscall wrapper.
*/
if (!found_syscall)
break; /* assume not a syscall wrapper, give up gracefully */
} else if (instr_is_cti(instr)) {
/* We expect only ctis like ret or ret imm, syscall, and call, which are
* handled above. Give up gracefully if we hit any other cti.
* XXX: what about jmp to shared ret (seen in the past on some syscalls)?
*/
/* Update: win10 TH2 1511 x64 has a cti:
* ntdll!NtContinue:
* 00007ff9`13185630 4c8bd1 mov r10,rcx
* 00007ff9`13185633 b843000000 mov eax,43h
* 00007ff9`13185638 f604250803fe7f01 test byte ptr [SharedUserData+0x308 (00000000`7ffe0308)],1
* 00007ff9`13185640 7503 jne ntdll!NtContinue+0x15 (00007ff9`13185645)
* 00007ff9`13185642 0f05 syscall
* 00007ff9`13185644 c3 ret
* 00007ff9`13185645 cd2e int 2Eh
* 00007ff9`13185647 c3 ret
*/
if (expect_x64 && instr_is_cbr(instr) &&
opnd_get_pc(instr_get_target(instr)) == pc + 3/*syscall;ret*/) {
/* keep going */
} else
break;
} else if ((!found_eax || !found_edx || !found_ecx) &&
instr_get_opcode(instr) == OP_mov_imm &&
opnd_is_reg(instr_get_dst(instr, 0))) {
if (!found_eax && opnd_get_reg(instr_get_dst(instr, 0)) == REG_EAX) {
info->sysnum = (int) opnd_get_immed_int(instr_get_src(instr, 0));
found_eax = true;
} else if (!found_edx && opnd_get_reg(instr_get_dst(instr, 0)) == REG_EDX) {
uint imm = (uint) opnd_get_immed_int(instr_get_src(instr, 0));
if (imm == 0x7ffe0300 ||
/* On Win10 the immed is ntdll!Wow64SystemServiceCall */
(expect_wow && imm > (ptr_uint_t)preferred &&
imm < (ptr_uint_t)preferred + img->SizeOfImage))
found_edx = true;
} else if (!found_ecx && opnd_get_reg(instr_get_dst(instr, 0)) == REG_ECX) {
found_ecx = true;
info->fixup_index = (int) opnd_get_immed_int(instr_get_src(instr, 0));
}
} else if (instr_get_opcode(instr) == OP_xor &&
opnd_is_reg(instr_get_src(instr, 0)) &&
opnd_get_reg(instr_get_src(instr, 0)) == REG_ECX &&
opnd_is_reg(instr_get_dst(instr, 0)) &&
opnd_get_reg(instr_get_dst(instr, 0)) == REG_ECX) {
/* xor to 0 */
found_ecx = true;
info->fixup_index = 0;
}
num_instr++;
if (num_instr > MAX_INSTRS_BEFORE_SYSCALL) /* wrappers should be short! */
break; /* avoid weird cases like NPXEMULATORTABLE */
}
instr_destroy(dcontext, instr);
return found_syscall;
}
/* returns false on failure */
static bool
decode_syscall_num(void *dcontext, byte *entry, syscall_info_t *info)
{
/* FIXME: would like to fail gracefully rather than have a DR assertion
* on non-code! => use DEBUG=0 INTERNAL=1 DR build!
*/
bool found_syscall = false, found_eax = false, found_edx = false, found_ecx = false;
bool found_ret = false;
byte *pc;
int num_instr = 0;
instr_t *instr;
if (entry == NULL)
return false;
info->num_args = -1; /* if find sysnum but not args */
info->sysnum = -1;
info->fixup_index = -1;
instr = instr_create(dcontext);
pc = entry;
/* FIXME - we don't support decoding 64bit instructions in 32bit mode, but I want
* this to work on 32bit machines. Hack fix based on the wrapper pattern, we skip
* the first instruction (mov r10, rcx) here, the rest should decode ok.
* Xref PR 236203. */
if (expect_x64 && *pc == 0x4c && *(pc+1) == 0x8b && *(pc+2) == 0xd1)
pc += 3;
while (true) {
instr_reset(dcontext, instr);
pc = decode(dcontext, pc, instr);
if (verbose)
dr_print_instr(dcontext, STDOUT, instr, "");
if (pc == NULL || !instr_valid(instr))
break;
/* ASSUMPTION: a mov imm of 0x7ffe0300 into edx followed by an
* indirect call via edx is a system call on XP and later
* On XP SP1 it's call *edx, while on XP SP2 it's call *(edx)
* For wow it's a call through fs.
* FIXME - core exports various is_*_syscall routines (such as
* instr_is_wow64_syscall()) which we could use here instead of
* duplicating if they were more flexible about when they could
* be called (instr_is_wow64_syscall() for ex. asserts if not
* in a wow process).
*/
if (/* int 2e or x64 or win8 sysenter */
(instr_is_syscall(instr) && found_eax && (expect_int2e || expect_x64 || expect_sysenter)) ||
/* sysenter case */
(expect_sysenter && found_edx && found_eax &&
instr_is_call_indirect(instr) &&
/* XP SP{0,1}, 2003 SP0: call *edx */
((opnd_is_reg(instr_get_target(instr)) &&
opnd_get_reg(instr_get_target(instr)) == REG_EDX) ||
/* XP SP2, 2003 SP1: call *(edx) */
(opnd_is_base_disp(instr_get_target(instr)) &&
opnd_get_base(instr_get_target(instr)) == REG_EDX &&
opnd_get_index(instr_get_target(instr)) == REG_NULL &&
opnd_get_disp(instr_get_target(instr)) == 0))) ||
/* wow case
* we don't require found_ecx b/c win8 does not use ecx
*/
(expect_wow && found_eax &&
instr_is_call_indirect(instr) &&
opnd_is_far_base_disp(instr_get_target(instr)) &&
opnd_get_base(instr_get_target(instr)) == REG_NULL &&
opnd_get_index(instr_get_target(instr)) == REG_NULL &&
opnd_get_segment(instr_get_target(instr)) == SEG_FS)) {
found_syscall = true;
} else if (instr_is_return(instr)) {
if (!found_ret) {
process_ret(instr, info);
found_ret = true;
}
break;
} else if (instr_is_cti(instr)) {
if (instr_get_opcode(instr) == OP_call) {
/* handle win8 x86 which has sysenter callee adjacent-"inlined"
* ntdll!NtYieldExecution:
* 77d7422c b801000000 mov eax,1
* 77d74231 e801000000 call ntdll!NtYieldExecution+0xb (77d74237)
* 77d74236 c3 ret
* 77d74237 8bd4 mov edx,esp
* 77d74239 0f34 sysenter
* 77d7423b c3 ret
*/
byte *tgt;
assert(opnd_is_pc(instr_get_target(instr)));
tgt = opnd_get_pc(instr_get_target(instr));
/* we expect only ret or ret imm, and possibly some nops (in gdi32).
* XXX: what about jmp to shared ret (seen in the past on some syscalls)?
*/
if (tgt > pc && tgt <= pc + 16) {
bool ok = false;
do {
if (pc == tgt) {
ok = true;
break;
}
instr_reset(dcontext, instr);
pc = decode(dcontext, pc, instr);
if (verbose)
dr_print_instr(dcontext, STDOUT, instr, "");
if (instr_is_return(instr)) {
process_ret(instr, info);
found_ret = true;
} else if (!instr_is_nop(instr))
break;
num_instr++;
} while (num_instr <= MAX_INSTRS_BEFORE_SYSCALL);
if (ok)
continue;
}
}
/* assume not a syscall wrapper if we hit a cti */
break; /* give up gracefully */
} else if ((!found_eax || !found_edx || !found_ecx) &&
instr_get_opcode(instr) == OP_mov_imm &&
opnd_is_reg(instr_get_dst(instr, 0))) {
if (!found_eax && opnd_get_reg(instr_get_dst(instr, 0)) == REG_EAX) {
info->sysnum = (int) opnd_get_immed_int(instr_get_src(instr, 0));
found_eax = true;
} else if (!found_edx && opnd_get_reg(instr_get_dst(instr, 0)) == REG_EDX) {
int imm = (int) opnd_get_immed_int(instr_get_src(instr, 0));
if (imm == 0x7ffe0300)
found_edx = true;
} else if (!found_ecx && opnd_get_reg(instr_get_dst(instr, 0)) == REG_ECX) {
found_ecx = true;
info->fixup_index = (int) opnd_get_immed_int(instr_get_src(instr, 0));
}
} else if (instr_get_opcode(instr) == OP_xor &&
opnd_is_reg(instr_get_src(instr, 0)) &&
opnd_get_reg(instr_get_src(instr, 0)) == REG_ECX &&
opnd_is_reg(instr_get_dst(instr, 0)) &&
opnd_get_reg(instr_get_dst(instr, 0)) == REG_ECX) {
/* xor to 0 */
found_ecx = true;
info->fixup_index = 0;
}
num_instr++;
if (num_instr > MAX_INSTRS_BEFORE_SYSCALL) /* wrappers should be short! */
break; /* avoid weird cases like NPXEMULATORTABLE */
}
instr_destroy(dcontext, instr);
return found_syscall;
}
