Fixed the godawful branch execution loop and added more prints

master
xtremegamer1 2 years ago
parent e2e578bf55
commit 64d2adebc7

@ -26,6 +26,7 @@ class emu_t {
~emu_t();
bool init();
bool emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn);
static const bool s_log_instructions = true;
private:
uc_engine* uc;

@ -1,5 +1,8 @@
#include <string>
#include <vmemu_t.hpp>
#include <unordered_set>
int g_new_delete_tracker;
namespace vm {
emu_t::emu_t(vm::vmctx_t* vm_ctx) : m_vm(vm_ctx) {};
@ -58,6 +61,8 @@ bool emu_t::init() {
}
bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
std::printf("\nSTARTING FROM NEW VMENTER AT %p (%p)\n\n", vmenter_rva + m_vm->m_module_base,
vmenter_rva + m_vm->m_image_base);
uc_err err;
vrtn.m_rva = vmenter_rva;
@ -83,6 +88,8 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
}
std::printf("> beginning execution at = %p (%p)\n", rip, rip - m_vm->m_module_base + m_vm->m_image_base);
std::printf("vsp: %s, vip: %s\n", ZydisRegisterGetString(cc_blk->m_vm.vsp),
ZydisRegisterGetString(cc_blk->m_vm.vip));
if ((err = uc_emu_start(uc, rip, 0ull, 0ull, 0ull))) {
std::printf("> error starting emu... reason = %d\n", err);
return false;
@ -92,8 +99,8 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
std::printf("> emulated blk_%p\n\n", cc_blk->m_vip.img_based);
// keep track of the emulated blocks... by their addresses...
std::vector<std::uintptr_t> blk_addrs;
blk_addrs.push_back(blk.m_vip.rva + m_vm->m_module_base);
std::unordered_set<std::uintptr_t> blk_addrs;
blk_addrs.insert(blk.m_vip.rva + m_vm->m_module_base);
// the vector containing the vblk's grows inside of this for loop
// thus we cannot use an advanced for loop (which uses itr's)...
@ -103,8 +110,7 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
// force the emulation of all branches...
for (const auto br : blk.branches) {
// only emulate blocks that havent been emulated before...
if (std::find(blk_addrs.begin(), blk_addrs.end(), br) !=
blk_addrs.end())
if (blk_addrs.find(br) != blk_addrs.end())
continue;
std::uintptr_t vsp = 0ull;
@ -119,9 +125,12 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
cc_blk = &new_blk;
// emulate the branch...
// The stack is now set up with vsp pointing to the next branch to execute, therefore the jmp will go to that branch.
uc_mem_write(uc, vsp, &br, sizeof br);
std::printf("> beginning execution at = %p (%p)\n", blk.m_jmp.rip,
blk.m_jmp.rip - m_vm->m_module_base + m_vm->m_image_base);
std::printf("vsp: %s, vip: %s\n", ZydisRegisterGetString(cc_blk->m_vm.vsp),
ZydisRegisterGetString(cc_blk->m_vm.vip));
if ((err = uc_emu_start(uc, blk.m_jmp.rip, 0ull, 0ull, 0ull))) {
std::printf("> error starting emu... reason = %d\n", err);
return false;
@ -129,6 +138,8 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
extract_branch_data();
std::printf("> emulated blk_%p\n", cc_blk->m_vip.img_based);
// We needn't execute a block more than once for a single vrtn.
blk_addrs.insert(br);
}
}
}
@ -138,7 +149,7 @@ bool emu_t::emulate(std::uint32_t vmenter_rva, vm::instrs::vrtn_t& vrtn) {
[&](vm::instrs::vblk_t& blk) {
if (blk.m_jmp.ctx) uct_context_free(blk.m_jmp.ctx);
if (blk.m_jmp.stack) delete[] blk.m_jmp.stack;
if (blk.m_jmp.stack) delete[] blk.m_jmp.stack; if (blk.m_jmp.stack) g_new_delete_tracker--;
});
return true;
@ -200,6 +211,7 @@ void emu_t::extract_branch_data() {
vm::utils::scn::executable(m_vm->m_module_base, imm_mod_based)) {
cc_blk->branches.push_back(imm_mod_based);
cc_blk->branch_type = vm::instrs::vbranch_type::absolute;
std::printf("> absolute virtual jmp uncovered... branch = %p\n", imm_mod_based);
}
} else {
std::printf("> jump table detected... review instruction stream...\n");
@ -264,7 +276,7 @@ bool emu_t::branch_pred_spec_exec(uc_engine* uc, uint64_t address,
if (!obj->cc_trace.m_instrs.size()) {
obj->cc_trace.m_vip = obj->cc_blk->m_vm.vip;
obj->cc_trace.m_vsp = obj->cc_blk->m_vm.vsp;
obj->cc_trace.m_stack = new std::uint8_t[STACK_SIZE];
obj->cc_trace.m_stack = new std::uint8_t[STACK_SIZE]; g_new_delete_tracker++;
uc_mem_read(uc, STACK_BASE, obj->cc_trace.m_stack, STACK_SIZE);
}
@ -309,7 +321,7 @@ bool emu_t::branch_pred_spec_exec(uc_engine* uc, uint64_t address,
uct_context_free(instr.m_cpu);
});
delete[] obj->cc_trace.m_stack;
delete[] obj->cc_trace.m_stack; if (obj->cc_trace.m_stack) g_new_delete_tracker--;
obj->cc_trace.m_instrs.clear();
if (vinstr.mnemonic != vm::instrs::mnemonic_t::jmp) {
@ -352,7 +364,7 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
if (!obj->cc_trace.m_instrs.size()) {
obj->cc_trace.m_vip = obj->cc_blk->m_vm.vip;
obj->cc_trace.m_vsp = obj->cc_blk->m_vm.vsp;
obj->cc_trace.m_stack = new std::uint8_t[STACK_SIZE];
obj->cc_trace.m_stack = new std::uint8_t[STACK_SIZE]; g_new_delete_tracker++;
obj->cc_trace.m_begin = address;
uc_mem_read(uc, STACK_BASE, obj->cc_trace.m_stack, STACK_SIZE);
}
@ -369,7 +381,9 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
// find the last MOV REG, DWORD PTR [VIP] in the instruction stream, then
// remove any instructions from this instruction to the JMP/RET...
const auto rva_fetch = std::find_if(
// This may be more performant and elegant, but it fucks my new jmp profiler.
// I will keep it in as a comment for now in case I can make a jmp profiler compatible with this.
/* const auto rva_fetch = std::find_if(
obj->cc_trace.m_instrs.rbegin(), obj->cc_trace.m_instrs.rend(),
[& vip = obj->cc_trace.m_vip](
const vm::instrs::emu_instr_t& instr) -> bool {
@ -388,7 +402,7 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
}
obj->cc_trace.m_instrs.erase((rva_fetch + 1).base(),
--obj->cc_trace.m_instrs.end());
}
} */
// set the virtual code block vip address information...
if (!obj->cc_blk->m_vip.rva || !obj->cc_blk->m_vip.img_based) {
@ -418,20 +432,22 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
} else {
const auto vinstr = vm::instrs::determine(obj->cc_trace);
if (vinstr.mnemonic != vm::instrs::mnemonic_t::unknown) {
std::printf("%p: ", obj->cc_trace.m_begin + obj->m_vm->m_image_base - obj->m_vm->m_module_base);
if (vinstr.imm.has_imm)
if (vinstr.mnemonic == instrs::mnemonic_t::lreg || vinstr.mnemonic == instrs::mnemonic_t::sreg)
std::printf("> %s %s_%s\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str(),
vm::reg_names::prefixes[(vinstr.imm.val / 8) % 8].c_str(),
vm::reg_names::suffixes[(vinstr.imm.val / 8) / 8].c_str());
if (s_log_instructions) {
std::printf("%p: ", obj->cc_trace.m_begin + obj->m_vm->m_image_base - obj->m_vm->m_module_base);
if (vinstr.imm.has_imm)
if (vinstr.mnemonic == instrs::mnemonic_t::lreg || vinstr.mnemonic == instrs::mnemonic_t::sreg)
std::printf("> %s %s_%s\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str(),
vm::reg_names::prefixes[(vinstr.imm.val / 8) % 8].c_str(),
vm::reg_names::suffixes[(vinstr.imm.val / 8) / 8].c_str());
else
std::printf("> %s %p\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str(),
vinstr.imm.val);
else
std::printf("> %s %p\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str(),
vinstr.imm.val);
else
std::printf("> %s\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str());
std::printf("> %s\n",
vm::instrs::get_profile(vinstr.mnemonic)->name.c_str());
}
} else {
zydis_rtn_t inst_stream;
std::for_each(obj->cc_trace.m_instrs.begin(),
@ -444,8 +460,6 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
"> err: please define the following vm handler (at = %p):\n",
(obj->cc_trace.m_begin - obj->m_vm->m_module_base) +
obj->m_vm->m_image_base);
std::printf("vsp: %s, vip: %s\n", ZydisRegisterGetString(obj->cc_trace.m_vsp),
ZydisRegisterGetString(obj->cc_trace.m_vip));
vm::utils::print(inst_stream);
uc_emu_stop(uc);
return false;
@ -471,7 +485,7 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
// set current code block virtual jmp instruction information...
obj->cc_blk->m_jmp.ctx = copy;
obj->cc_blk->m_jmp.rip = obj->cc_trace.m_begin;
obj->cc_blk->m_jmp.stack = new std::uint8_t[STACK_SIZE];
obj->cc_blk->m_jmp.stack = new std::uint8_t[STACK_SIZE]; g_new_delete_tracker++;
obj->cc_blk->m_jmp.m_vm = {obj->cc_trace.m_vip, obj->cc_trace.m_vsp};
std::memcpy(obj->cc_blk->m_jmp.stack, obj->cc_trace.m_stack,
STACK_SIZE);
@ -491,7 +505,7 @@ bool emu_t::code_exec_callback(uc_engine* uc, uint64_t address, uint32_t size,
uct_context_free(instr.m_cpu);
});
delete[] obj->cc_trace.m_stack;
delete[] obj->cc_trace.m_stack; if (obj->cc_trace.m_stack) g_new_delete_tracker--;
obj->cc_trace.m_instrs.clear();
}
return true;
@ -537,7 +551,7 @@ bool emu_t::legit_branch(vm::instrs::vblk_t& vblk, std::uintptr_t branch_addr) {
uc_context* backup;
uct_context_alloc(uc, &backup);
uc_context_save(uc, backup);
std::uint8_t* stack = new std::uint8_t[STACK_SIZE];
std::uint8_t* stack = new std::uint8_t[STACK_SIZE]; g_new_delete_tracker++;
uc_mem_read(uc, STACK_BASE, stack, STACK_SIZE);
// restore cpu and stack back to the virtual jump handler...
@ -557,7 +571,7 @@ bool emu_t::legit_branch(vm::instrs::vblk_t& vblk, std::uintptr_t branch_addr) {
uc_mem_write(uc, STACK_BASE, stack, STACK_SIZE);
uc_context_restore(uc, backup);
uct_context_free(backup);
delete[] stack;
delete[] stack; if (stack) g_new_delete_tracker--;
// add normal execution callback back...
uc_hook_del(uc, branch_pred_hook);

Loading…
Cancel
Save