static dr_emit_flags_t event_basic_block(void *drcontext, void *tag, instrlist_t *bb, bool for_trace, bool translating) { instr_t *instr, *next_instr; #ifdef VERBOSE dr_printf("in dr_basic_block(tag="PFX")\n", tag); # if VERBOSE_VERBOSE instrlist_disassemble(drcontext, tag, bb, STDOUT); # endif #endif for (instr = instrlist_first_app(bb); instr != NULL; instr = next_instr) { next_instr = instr_get_next_app(instr); if (!instr_opcode_valid(instr)) continue; /* instrument calls and returns -- ignore far calls/rets */ if (instr_is_call_direct(instr)) { dr_insert_call_instrumentation(drcontext, bb, instr, (app_pc)at_call); } else if (instr_is_call_indirect(instr)) { dr_insert_mbr_instrumentation(drcontext, bb, instr, (app_pc)at_call_ind, SPILL_SLOT_1); } else if (instr_is_return(instr)) { dr_insert_mbr_instrumentation(drcontext, bb, instr, (app_pc)at_return, SPILL_SLOT_1); } } return DR_EMIT_DEFAULT; }
/* return the branch type of the (branch) inst */ uint instr_branch_type(instr_t *cti_instr) { instr_get_opcode(cti_instr); /* ensure opcode is valid */ if (instr_get_opcode(cti_instr) == OP_blx) { /* To handle the mode switch we go through the ibl. * FIXME i#1551: once we have far linking through stubs we should * remove this and have a faster link through the stub. */ return LINK_INDIRECT|LINK_CALL; } /* We treate a predicated call as a cbr, not a call */ else if (instr_is_cbr_arch(cti_instr) || instr_is_ubr_arch(cti_instr)) return LINK_DIRECT|LINK_JMP; else if (instr_is_call_direct(cti_instr)) return LINK_DIRECT|LINK_CALL; else if (instr_is_call_indirect(cti_instr)) return LINK_INDIRECT|LINK_CALL; else if (instr_is_return(cti_instr)) return LINK_INDIRECT|LINK_RETURN; else if (instr_is_mbr_arch(cti_instr)) return LINK_INDIRECT|LINK_JMP; else CLIENT_ASSERT(false, "instr_branch_type: unknown opcode"); return LINK_INDIRECT; }
/* returns whether found a syscall * - found_eax: whether the caller has seen "mov imm => %eax" * - found_edx: whether the caller has seen "mov $0x7ffe0300 => %edx", * xref the comment below about "mov $0x7ffe0300 => %edx". */ static bool process_syscall_instr(void *dcontext, instr_t *instr, bool found_eax, bool found_edx) { /* 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) || /* win10 has imm in edx and a near call */ found_edx))) return true; return false; }
static dr_emit_flags_t event_instruction(void *drcontext, void *tag, instrlist_t *bb, instr_t *instr, bool for_trace, bool translating, void *user_data) { /* ignore tool-inserted instrumentation */ if (!instr_is_app(instr)) return DR_EMIT_DEFAULT; /* instrument calls and returns -- ignore far calls/rets */ if (instr_is_call_direct(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_direct_calls)); } else if (instr_is_call_indirect(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_indirect_calls)); } else if (instr_is_return(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_returns)); } return DR_EMIT_DEFAULT; }
static dr_emit_flags_t event_basic_block(void *drcontext, void *tag, instrlist_t *bb, bool for_trace, bool translating) { instr_t *instr, *next_instr; #ifdef VERBOSE dr_printf("in dynamorio_basic_block(tag="PFX")\n", tag); # ifdef VERBOSE_VERBOSE instrlist_disassemble(drcontext, tag, bb, STDOUT); # endif #endif for (instr = instrlist_first(bb); instr != NULL; instr = next_instr) { /* grab next now so we don't go over instructions we insert */ next_instr = instr_get_next(instr); /* instrument calls and returns -- ignore far calls/rets */ if (instr_is_call_direct(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_direct_calls)); } else if (instr_is_call_indirect(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_indirect_calls)); } else if (instr_is_return(instr)) { insert_counter_update(drcontext, bb, instr, offsetof(per_thread_t, num_returns)); } } #if defined(VERBOSE) && defined(VERBOSE_VERBOSE) dr_printf("Finished instrumenting dynamorio_basic_block(tag="PFX")\n", tag); instrlist_disassemble(drcontext, tag, bb, STDOUT); #endif return DR_EMIT_DEFAULT; }
/* 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 whether found a syscall * - found_eax: whether the caller has seen "mov imm => %eax" * - found_edx: whether the caller has seen "mov $0x7ffe0300 => %edx", * xref the comment in process_syscall_instr. */ static bool process_syscall_call(void *dcontext, byte *next_pc, instr_t *call, bool found_eax, bool found_edx) { int num_instr; byte *pc; instr_t instr; bool found_syscall = false; assert(instr_get_opcode(call) == OP_call && opnd_is_pc(instr_get_target(call))); pc = opnd_get_pc(instr_get_target(call)); if (pc > next_pc + MAX_SYSENTER_CALLEE_OFFSET || pc <= next_pc /* assuming the call won't go backward */) return false; /* 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 * * or DrMem-i#1366-c#2 * USER32!NtUserCreateWindowStation: * 75caea7a b841110000 mov eax,0x1141 * 75caea7f e838000000 call user32!...+0xd (75caeabc) * 75caea84 c22000 ret 0x20 * ... * USER32!GetWindowStationName: * 75caea8c 8bff mov edi,edi * 75caea8e 55 push ebp * ... * 75caeabc 8bd4 mov edx,esp * 75caeabe 0f34 sysenter * 75caeac0 c3 ret */ /* We expect the win8 x86 sysenter adjacent "inlined" callee to be as simple as * 75caeabc 8bd4 mov edx,esp * 75caeabe 0f34 sysenter * 75caeac0 c3 ret */ instr_init(dcontext, &instr); num_instr = 0; do { instr_reset(dcontext, &instr); pc = decode(dcontext, pc, &instr); if (verbose) dr_print_instr(dcontext, STDOUT, &instr, ""); if (pc == NULL || !instr_valid(&instr)) break; if (instr_is_syscall(&instr) || instr_is_call_indirect(&instr)) { found_syscall = process_syscall_instr(dcontext, &instr, found_eax, found_edx); break; } else if (instr_is_cti(&instr)) { break; } num_instr++; } while (num_instr <= MAX_INSTRS_SYSENTER_CALLEE); instr_free(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; }