指令级工具Dobby源码阅读
2022-7-29 18:8:51 Author: 看雪学苑(查看原文) 阅读量:18 收藏


本文为看雪论坛优秀文章

看雪论坛作者ID:KerryS

Dobby一共两个功能,其一是inlinehook,其二是指令插桩,两者原理差不多,主要介绍指令插桩。

所谓指令插桩,就是在任意一条指令(函数头或者函数内部都行),进行插桩,执行到这条指令的时候,会去执行我们定义的回调函数,然后再回来执行原来的指令流。使用方法:
int res_instument = DobbyInstrument((void *) addr, offset_name_handler);//handler即我们自定义的回调 //RegisterContext为寄存器上下文,HookEntrtInfo为hook一些必要信息,比如hook地址等void offset_name_handler(RegisterContext *ctx, const HookEntryInfo *info)typedef struct _RegisterContext {  uint32_t dummy_0;  uint32_t dummy_1;   uint32_t dummy_2;  uint32_t sp;   union {    uint32_t r[13];    struct {      uint32_t r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12;    } regs;  } general;   uint32_t lr;} RegisterContext //HookEntryInfo为hook地址及idtypedef struct _HookEntryInfo {  int hook_id;  union {    void *target_address;    void *function_address;    void *instruction_address;  };} HookEntryInfo;


工作原理

所谓听君一席话,胜读十年书;看君一张图,胜过十席话。用图来说明最好不过了,我在阅读的过程中也是一边梳理一边画图。
被插桩指令处被替换为:
------------------------------------------------------------------------------process 60890x9d639d32 nop0x9d639d34 ldr.w pc, [pc, #-0x0]0x9d639d38  //地址 0xcea0a0ac0x9d639d38            0  1  2  3  4  5  6  7  8  9  A  B  C  D  E  F  0123456789ABCDEF00000000  ac a0 a0 ce                                      ....------------------------------------------------------------------------------
arm处理器采用指令流水技术,即取指译码执行三阶段同步进行,pc寄存器指向的是正在取指的指令,arm模式中为当前执行的指令地址+8,thhumb模式中为当前位置+4,故而上面的ldr执行的时候,pc寄存器值为ldr指令位置+4,所以ldr,pc,[pc,-0x0]刚好是把下一条内容放入pc中,即跳转了。
这种跳转方式支持的范围是一个寄存器的宽度,也就是32位,4g内存,linux进程的虚拟地址空间好像也是4g,这样就可以进程全地址跳转了。那会跳转到哪里呢,跳转到
prologue_dispatch_bridge
0xcea0a0ac。
0xcea0a0ac ldr ip, [pc]0xcea0a0b0 ldr pc, [pc]0xcea0a0b4  //地址 0xa2305b800xcea0a0b8  //地址 0xcea0a0000xcea0a0b4            0  1  2  3  4  5  6  7  8  9  A  B  C  D  E  F  0123456789ABCDEF00000000  80 5b 30 a2 00 a0 a0 ce
主要做了两件事,第一,把0xa2305b80放到ip寄存器,第二,跳转到0xcea0a000。注意,这里的是arm模式的指令,pc偏移是8。
其中,0xcea0a000就是closure bridge上半场。
0xcea0a000 sub sp, sp, #0x380xcea0a004 str lr, [sp, #0x34]0xcea0a008 str ip, [sp, #0x30]0xcea0a00c str fp, [sp, #0x2c]0xcea0a010 str sl, [sp, #0x28]0xcea0a014 str sb, [sp, #0x24]0xcea0a018 str r8, [sp, #0x20]0xcea0a01c str r7, [sp, #0x1c]0xcea0a020 str r6, [sp, #0x18]0xcea0a024 str r5, [sp, #0x14]0xcea0a028 str r4, [sp, #0x10]0xcea0a02c str r3, [sp, #0xc]0xcea0a030 str r2, [sp, #8]0xcea0a034 str r1, [sp, #4]0xcea0a038 str r0, [sp]0xcea0a03c add r0, sp, #0x380xcea0a040 sub sp, sp, #80xcea0a044 str r0, [sp, #4]0xcea0a048 sub sp, sp, #80xcea0a04c mov r0, sp0xcea0a050 mov r1, ip0xcea0a054 bl #0xcea0a05c0xcea0a058 b #0xcea0a0640xcea0a05c ldr pc, [pc, #-4]0xcea0a060 //地址 0x9d2b43e10xcea0a060            0  1  2  3  4  5  6  7  8  9  A  B  C  D  E  F  0123456789ABCDEF00000000  e1 43 2b 9d
这里0xcea0a060处的0x9d2b43e1是高层handler,高层handler会调用我们自定义的handler,就是它。

instrument_call_forward_handler
void instrument_call_forward_handler(RegisterContext *ctx, HookEntry *entry) {  DynamicBinaryInstrumentRouting *route = (DynamicBinaryInstrumentRouting *)entry->route;  if (route->handler) {    DBICallTy handler;    HookEntryInfo entry_info;    entry_info.hook_id = entry->id;    entry_info.instruction_address = entry->instruction_address;    handler = (DBICallTy)route->handler;    (*handler)(ctx, (const HookEntryInfo *)&entry_info);  }   // set prologue bridge next hop address with origin instructions that have been relocated(patched)  set_routing_bridge_next_hop(ctx, entry->relocated_origin_instructions);}
这个handler除了调用我们的handler,还做了一件苟且的事情,后面会说到。

梳理一下这个closure bridge,首先保存寄存器环境,然后到地址0xcea0a054时,用bl指令跳到 0xcea0a05c,0xcea0a05c通过ldr方式找到高层handler地址并且调用,注意,bl指令会把下一条指令地址,即0xcea0a058放入lr寄存器,当bl跳到指定函数并且执行之后,函数会返回到lr寄存器保存的地址,即0xcea0a058 b #0xcea0a064,看看0xcea0a064内容。

closure bridge下半场
0xcea0a064 add sp, sp, #80xcea0a068 add sp, sp, #80xcea0a06c pop {r0}0xcea0a070 pop {r1}0xcea0a074 pop {r2}0xcea0a078 pop {r3}0xcea0a07c pop {r4}0xcea0a080 pop {r5}0xcea0a084 pop {r6}0xcea0a088 pop {r7}0xcea0a08c pop {r8}0xcea0a090 pop {sb}0xcea0a094 pop {sl}0xcea0a098 pop {fp}0xcea0a09c pop {ip}0xcea0a0a0 pop {lr}0xcea0a0a4 mov pc, ip
做的事情很平常,就是把之前上半场保存的寄存器出栈,同时恢复栈平衡;只有一点不平常,就是最后一条mov,pc,ip,跳到ip寄存存保存的地址,那么ip寄存起保存的地址是啥呢,还记得上文说的苟且之事吗?
instrument_call_forward_handler函数的最后一句
// set prologue bridge next hop address with origin instructions that have been relocated(patched)  set_routing_bridge_next_hop(ctx, entry->relocated_origin_instructions); void set_routing_bridge_next_hop(RegisterContext *ctx, void *address) {  *reinterpret_cast<void **>(&ctx->general.regs.r12) = address;}
就是把entry->relocated_origin_instructions的内容赋给r12寄存器,这个entry->relocated_origin_instructions就是原始指令的重定位之后的位置。
因为原始指令被我们patch成了ldr pc,[pc,-4]以及一条地址,这些被patch的指令会被修复好,放在entry->relocated_origin_instructions(指令修复问题后文继续说),执行完修复好的原始指令之后,会跳回到被patch的原始指令之后的那些指令,继续执行,这个过程大致如下:
原指令
因为是patch需要至少8字节,而这里原始指令是thumb,所以patch了四条,修复好的。

重定位后的指令
------------------------------------------------------------------------------process 60890xcea0a0c0 nop0xcea0a0c2 nop0xcea0a0c4 push {r0, r1, r2, lr}0xcea0a0c6 nop0xcea0a0c8 cbz r0, #0xcea0a0cc0xcea0a0ca nop0xcea0a0cc b.w #0xcea0a0d00xcea0a0d0 ldr.w pc, [pc, #0x14]  0xcea0a0d0 + 0x14+thumb_pc_offset(4)=0xcea0a0e8,即 0x9d639d450xcea0a0d4 nop0xcea0a0d6 nop0xcea0a0d8 add r2, sp, #80xcea0a0da nop0xcea0a0dc str r1, [r2, #-0x4]!0xcea0a0e0 ldr.w pc, [pc, #-0x0]   同理,0x9d639d3d0xcea0a0e4 //地址 0x9d639d3d0xcea0a0e8 //地址 0x9d639d450xcea0a0e4            0  1  2  3  4  5  6  7  8  9  A  B  C  D  E  F  0123456789ABCDEF00000000  3d 9d 63 9d 45 9d 63 9d                          =.c.E.c.------------------------------------------------------------------------------
指令修复的逻辑是,pc相关的指令,采用ldr给它跳回正确的位置;pc无关的指令,直接复制过来,这里push指令,add指令以及str.w指令都被直接复制过来,插入的一些nop指令是为了4字节对齐,至于为啥要对齐我就不知道了,印象中thumb指令似乎是两字节对齐就可以了。
原指令中,只有cbz是pc相关的,这条指令的语义是,看r0寄存器是否为零,为零则跳转到到给定位置,这个例子中是跳转到当前位置+0x10,即偏移25ED44处;可以看到,Dobby的修复手段是,修改cbz指令,如果r0为0,则通过ldr(0xcea0a0d0 ldr.w pc, [pc, #0x14])指令,给它跳回偏移25ED44处(0x9d639d45)。
若不为0,则通过ldr(0xcea0a0e0 ldr.w pc, [pc, #-0x0])跳转到被patch指令之后的指令继续去执行,在这里是偏移25ED3C(0x9d639d3d),这两条地址都加了个1,是因为原指令是thumb指令,arm处理器通过末尾地址是否为1来确定采用arm模式还是thumb模式,为1采用thum模式。至此,整个Dobby指令插桩的逻辑就完了。

代码详解

先遍历一个HookEntry链表,这个链表保存了每一次插桩的信息;每插桩一条指令,都会生成一个HookEntry结构体并且添加到这个链表。遍历这个链表可以判断当前要插桩的指令是否被插过。
route->DispatchRouting();为重点方法,这个方法完成几乎所有的插桩工作,route->DispatchRouting()调用了两个方法BuildDynamicBinaryInstrumentRouting()和GenerateRelocatedCode(trampolinebuffer->getSize())。
void DynamicBinaryInstrumentRouting::DispatchRouting() {  BuildDynamicBinaryInstrumentRouting();   // generate relocated code which size == trampoline size  GenerateRelocatedCode(trampoline_buffer_->getSize());}

BuildDynamicBinaryInstrumentRouting()

void DynamicBinaryInstrumentRouting::BuildDynamicBinaryInstrumentRouting() {  // create closure trampoline jump to prologue_routing_dispath with the `entry_` data   ClosureTrampolineEntry *closure_trampoline;   void *handler = (void *)instrument_routing_dispatch;#if __APPLE__#if __has_feature(ptrauth_calls)  handler = __builtin_ptrauth_strip(handler, ptrauth_key_asia);#endif#endif    closure_trampoline = ClosureTrampoline::CreateClosureTrampoline(entry_, handler);  this->SetTrampolineTarget(closure_trampoline->address);    DLOG(0, "[closure bridge] Carry data %p ", entry_);    DLOG(0, "[closure bridge] Create prologue_dispatch_bridge %p", closure_trampoline->address);   // generate trampoline buffer, run before `GenerateRelocatedCode`  GenerateTrampolineBuffer(entry_->target_address, GetTrampolineTarget());}
其中,closuretrampoline = ClosureTrampoline::CreateClosureTrampoline(entry, handler);会生成
prologue_dispatch_bridge的那些汇编指令,其中__ EmitAddress((uint32_t)get_closure_bridge())是一个重点,closure_bridge的指令在这里生成。
ClosureTrampolineEntry *ClosureTrampoline::CreateClosureTrampoline(void *carry_data, void *carry_handler) {    ClosureTrampolineEntry *entry = nullptr;  entry = new ClosureTrampolineEntry; #ifdef ENABLE_CLOSURE_TRAMPOLINE_TEMPLATE#define CLOSURE_TRAMPOLINE_SIZE (7 * 4)  // use closure trampoline template code, find the executable memory and patch it.  Code *code = Code::FinalizeCodeFromAddress(closure_trampoline_template, CLOSURE_TRAMPOLINE_SIZE);#else // use assembler and codegen modules instead of template_code#include "TrampolineBridge/ClosureTrampolineBridge/AssemblyClosureTrampoline.h"#define _ turbo_assembler_.  TurboAssembler turbo_assembler_(0);   PseudoLabel entry_label;  PseudoLabel forward_bridge_label;     _ Ldr(r12, &entry_label);  _ Ldr(pc, &forward_bridge_label);  _ PseudoBind(&entry_label);  _ EmitAddress((uint32_t)entry);  _ PseudoBind(&forward_bridge_label);  _ EmitAddress((uint32_t)get_closure_bridge());   AssemblyCodeChunk *code = nullptr;  code = AssemblyCodeBuilder::FinalizeFromTurboAssembler(&turbo_assembler_);   entry->address = (void *)code->raw_instruction_start();  entry->size = code->raw_instruction_size();  entry->carry_data = carry_data;  entry->carry_handler = carry_handler;   delete code;  return entry;#endif}
void *get_closure_bridge() {   // if already initialized, just return.  if (closure_bridge)    return closure_bridge; // check if enable the inline-assembly closure_bridge_template#if ENABLE_CLOSURE_BRIDGE_TEMPLATE  extern void closure_bridge_tempate();  closure_bridge = closure_bridge_template;// otherwise, use the Assembler build the closure_bridge#else#define _ turbo_assembler_.  TurboAssembler turbo_assembler_(0);   _ sub(sp, sp, Operand(14 * 4));  _ str(lr, MemOperand(sp, 13 * 4));  _ str(r12, MemOperand(sp, 12 * 4));  _ str(r11, MemOperand(sp, 11 * 4));  _ str(r10, MemOperand(sp, 10 * 4));  _ str(r9, MemOperand(sp, 9 * 4));  _ str(r8, MemOperand(sp, 8 * 4));  _ str(r7, MemOperand(sp, 7 * 4));  _ str(r6, MemOperand(sp, 6 * 4));  _ str(r5, MemOperand(sp, 5 * 4));  _ str(r4, MemOperand(sp, 4 * 4));  _ str(r3, MemOperand(sp, 3 * 4));  _ str(r2, MemOperand(sp, 2 * 4));  _ str(r1, MemOperand(sp, 1 * 4));  _ str(r0, MemOperand(sp, 0 * 4));   // store sp  _ add(r0, sp, Operand(14 * 4));  _ sub(sp, sp, Operand(8));  _ str(r0, MemOperand(sp, 4));   // stack align  _ sub(sp, sp, Operand(8));   _ mov(r0, Operand(sp));  _ mov(r1, Operand(r12));   _ CallFunction(ExternalReference((void *)intercept_routing_common_bridge_handler));   // stack align  _ add(sp, sp, Operand(8));   // restore sp placeholder stack  _ add(sp, sp, Operand(8));   _ ldr(r0, MemOperand(sp, 4, PostIndex));  _ ldr(r1, MemOperand(sp, 4, PostIndex));  _ ldr(r2, MemOperand(sp, 4, PostIndex));  _ ldr(r3, MemOperand(sp, 4, PostIndex));  _ ldr(r4, MemOperand(sp, 4, PostIndex));  _ ldr(r5, MemOperand(sp, 4, PostIndex));  _ ldr(r6, MemOperand(sp, 4, PostIndex));  _ ldr(r7, MemOperand(sp, 4, PostIndex));  _ ldr(r8, MemOperand(sp, 4, PostIndex));  _ ldr(r9, MemOperand(sp, 4, PostIndex));  _ ldr(r10, MemOperand(sp, 4, PostIndex));  _ ldr(r11, MemOperand(sp, 4, PostIndex));  _ ldr(r12, MemOperand(sp, 4, PostIndex));  _ ldr(lr, MemOperand(sp, 4, PostIndex));   // auto switch A32 & T32 with `least significant bit`, refer `docs/A32_T32_states_switch.md`  _ mov(pc, Operand(r12));   AssemblyCodeChunk *code = AssemblyCodeBuilder::FinalizeFromTurboAssembler(&turbo_assembler_);  closure_bridge = (void *)code->raw_instruction_start();   DLOG(0, "[closure bridge] Build the closure bridge at %p", closure_bridge);#endif  return (void *)closure_bridge;}
BuildDynamicBinaryInstrumentRouting()还调用了这个GenerateTrampolineBuffer(entry_->target_address, GetTrampolineTarget()); 这个方法生成了TrampolineBuffer,也就是用于patch原始指令的那些指令,流程图的第二个小方块。
bool InterceptRouting::GenerateTrampolineBuffer(void *src, void *dst) {  CodeBufferBase *trampoline_buffer = NULL;  // if near branch trampoline plugin enabled  if (RoutingPluginManager::near_branch_trampoline) {    RoutingPluginInterface *plugin = NULL;    plugin = reinterpret_cast<RoutingPluginInterface *>(RoutingPluginManager::near_branch_trampoline);    if (plugin->GenerateTrampolineBuffer(this, src, dst) == false) {      DLOG(0, "Failed enable near branch trampoline plugin");    }  }   if (this->GetTrampolineBuffer() == NULL) {    trampoline_buffer = GenerateNormalTrampolineBuffer((addr_t)src, (addr_t)dst);    this->SetTrampolineBuffer(trampoline_buffer);     DLOG(0, "[trampoline] Generate trampoline buffer %p -> %p", src, dst);  }  return true;}

GenerateRelocatedCode(trampolinebuffer->getSize())

bool InterceptRouting::GenerateRelocatedCode(int tramp_size) {  // generate original code  AssemblyCodeChunk *origin = NULL;  origin = AssemblyCodeBuilder::FinalizeFromAddress((addr_t)entry_->target_address, tramp_size);  origin_ = origin;   // generate the relocated code  AssemblyCodeChunk *relocated = NULL;  relocated = AssemblyCodeBuilder::FinalizeFromAddress(0, 0);  relocated_ = relocated;   void *relocate_buffer = NULL;  relocate_buffer = entry_->target_address;   GenRelocateCodeAndBranch(relocate_buffer, origin, relocated);  if (relocated->raw_instruction_start() == 0)    return false;   // set the relocated instruction address  entry_->relocated_origin_instructions = (void *)relocated->raw_instruction_start();   DLOG(0, "[insn relocate] origin %p - %d", origin->raw_instruction_start(), origin->raw_instruction_size());    DLOG(0, "[insn relocate] relocated %p - %d", relocated->raw_instruction_start(), relocated->raw_instruction_size());    // save original prologue  memcpy((void *)entry_->origin_chunk_.chunk_buffer, (void *)origin_->raw_instruction_start(),         origin_->raw_instruction_size());  entry_->origin_chunk_.chunk.re_init_region_range(origin_);  return true;}
其中GenRelocateCodeAndBranch(relocate_buffer, origin, relocated);是重点,它会生成重定位代码,放在relocated指针指向的地址空间中。
void GenRelocateCodeAndBranch(void *buffer, AssemblyCodeChunk *origin, AssemblyCodeChunk *relocated) {  CodeBuffer *code_buffer = new CodeBuffer(64);   ThumbTurboAssembler thumb_turbo_assembler_(0, code_buffer);#define thumb_ thumb_turbo_assembler_.  TurboAssembler arm_turbo_assembler_(0, code_buffer);#define arm_ arm_turbo_assembler_.   Assembler *curr_assembler_ = NULL;   AssemblyCodeChunk origin_chunk;  origin_chunk.init_region_range(origin->raw_instruction_start(), origin->raw_instruction_size());   bool entry_is_thumb = origin->raw_instruction_start() % 2;  if (entry_is_thumb) {    origin->re_init_region_range(origin->raw_instruction_start() - THUMB_ADDRESS_FLAG, origin->raw_instruction_size());  }   LiteMutableArray relo_map(8); relocate_remain:  addr32_t execute_state_changed_pc = 0;   bool is_thumb = origin_chunk.raw_instruction_start() % 2;  if (is_thumb) {    curr_assembler_ = &thumb_turbo_assembler_;     buffer = (void *)((addr_t)buffer - THUMB_ADDRESS_FLAG);     addr32_t origin_code_start_aligned = origin_chunk.raw_instruction_start() - THUMB_ADDRESS_FLAG;    // remove thumb address flag    origin_chunk.re_init_region_range(origin_code_start_aligned, origin_chunk.raw_instruction_size());     gen_thumb_relocate_code(&relo_map, &thumb_turbo_assembler_, buffer, &origin_chunk, relocated,                            &execute_state_changed_pc);    if (thumb_turbo_assembler_.GetExecuteState() == ARMExecuteState) {      // relocate interrupt as execute state changed      if (execute_state_changed_pc < origin_chunk.raw_instruction_start() + origin_chunk.raw_instruction_size()) {        // re-init the origin        int relocate_remain_size =            origin_chunk.raw_instruction_start() + origin_chunk.raw_instruction_size() - execute_state_changed_pc;        // current execute state is ARMExecuteState, so not need `+ THUMB_ADDRESS_FLAG`        origin_chunk.re_init_region_range(execute_state_changed_pc, relocate_remain_size);         // update buffer        buffer = (void *)((addr_t)buffer + (execute_state_changed_pc - origin_code_start_aligned));         // add nop to align ARM        if (thumb_turbo_assembler_.pc_offset() % 4)          thumb_turbo_assembler_.t1_nop();        goto relocate_remain;      }    }  } else {    curr_assembler_ = &arm_turbo_assembler_;     gen_arm_relocate_code(&relo_map, &arm_turbo_assembler_, buffer, &origin_chunk, relocated,                          &execute_state_changed_pc);    if (arm_turbo_assembler_.GetExecuteState() == ThumbExecuteState) {      // relocate interrupt as execute state changed      if (execute_state_changed_pc < origin_chunk.raw_instruction_start() + origin_chunk.raw_instruction_size()) {        // re-init the origin        int relocate_remain_size =            origin_chunk.raw_instruction_start() + origin_chunk.raw_instruction_size() - execute_state_changed_pc;        // current execute state is ThumbExecuteState, add THUMB_ADDRESS_FLAG        origin_chunk.re_init_region_range(execute_state_changed_pc + THUMB_ADDRESS_FLAG, relocate_remain_size);         // update buffer        buffer = (void *)((addr_t)buffer + (execute_state_changed_pc - origin_chunk.raw_instruction_start()));        goto relocate_remain;      }    }  }   // TODO:  // if last instr is unlink branch, skip  //dkl 调回插桩点之后继续执行  addr32_t rest_instr_addr = origin_chunk.raw_instruction_start() + origin_chunk.raw_instruction_size();  if (curr_assembler_ == &thumb_turbo_assembler_) {    // Branch to the rest of instructions    thumb_ AlignThumbNop();    thumb_ t2_ldr(pc, MemOperand(pc, 0));    // Get the real branch address    thumb_ EmitAddress(rest_instr_addr + THUMB_ADDRESS_FLAG);  } else {    // Branch to the rest of instructions    CodeGen codegen(&arm_turbo_assembler_);    // Get the real branch address    codegen.LiteralLdrBranch(rest_instr_addr);  }   // Realize all the Pseudo-Label-Data  thumb_turbo_assembler_.RelocBind();   // Realize all the Pseudo-Label-Data  //dkl 在这里会修正之前lable link的ldr指令,  arm_turbo_assembler_.RelocBind();   // Generate executable code  {    // assembler without specific memory address    AssemblyCodeChunk *cchunk;    cchunk = MemoryArena::AllocateCodeChunk(code_buffer->getSize());    if (cchunk == nullptr)      return;     thumb_turbo_assembler_.SetRealizedAddress(cchunk->address);    arm_turbo_assembler_.SetRealizedAddress(cchunk->address);     // fixup the instr branch into trampoline(has been modified)     reloc_label_fixup(origin, &relo_map, &thumb_turbo_assembler_, &arm_turbo_assembler_);     AssemblyCodeChunk *code = NULL;    code = AssemblyCodeBuilder::FinalizeFromTurboAssembler(curr_assembler_);    relocated->re_init_region_range(code->raw_instruction_start(), code->raw_instruction_size());    delete code;  }   // thumb  if (entry_is_thumb) {    // add thumb address flag    relocated->re_init_region_range(relocated->raw_instruction_start() + THUMB_ADDRESS_FLAG,                                    relocated->raw_instruction_size());  }   // clean  {    thumb_turbo_assembler_.ClearCodeBuffer();    arm_turbo_assembler_.ClearCodeBuffer();     delete code_buffer;  }}
感觉有点啰嗦了,重点说一下指令修复那块吧,我们的例子中,要修复的指令是thumb1指令,最终会调到这里;我省略了其他指令的修复,只看cbz的,细节就不说了。
大概思路就是,用ldr pc,[pc,xxx]去跳转,但是第一次生成ldr指令的时候,xxx是没用的,等到全部重定位指令都生成之后,这些ldr都会被修正,因为ldr跳转的地址,都是储存在所有指令之后的,从流程图以及上面说的各个块的汇编指令也可以看出,地址都是存在指令末尾。
static void Thumb1RelocateSingleInstr(ThumbTurboAssembler *turbo_assembler, LiteMutableArray *thumb_labels,                                      int16_t instr, addr32_t from_pc, addr32_t to_pc,                                      addr32_t *execute_state_changed_pc_ptr) {  bool is_instr_relocated = false;   _ AlignThumbNop();   uint32_t val = 0, op = 0, rt = 0, rm = 0, rn = 0, rd = 0, shift = 0, cond = 0;  int32_t offset = 0;   int32_t op0 = 0, op1 = 0;  op0 = bits(instr, 10, 15);  // [F3.2.3 Special data instructions and branch and exchange]  if (op0 == 0b010001) {    op0 = bits(instr, 8, 9);    // [Add, subtract, compare, move (two high registers)]    if (op0 != 0b11) {      int rs = bits(instr, 3, 6);      // rs is PC register      if (rs == 15) {        val = from_pc;         uint16_t rewrite_inst = 0;        rewrite_inst = (instr & 0xff87) | LeftShift((VOLATILE_REGISTER.code()), 4, 3);         ThumbRelocLabelEntry *label = new ThumbRelocLabelEntry(val, false);        _ AppendRelocLabelEntry(label);         _ T2_Ldr(VOLATILE_REGISTER, label);        _ EmitInt16(rewrite_inst);         is_instr_relocated = true;      }    }     // compare branch (cbz, cbnz)  if ((instr & 0xf500) == 0xb100) {    uint16_t imm5 = bits(instr, 3, 7);    uint16_t i = bit(instr, 9);    uint32_t offset = (i << 6) | (imm5 << 1);    val = from_pc + offset;    rn = bits(instr, 0, 2);     //ThumbTurboAssembler 的data_labels_记录所有的ThumbRelocLabelEntry,保存着要跳转的地址,同时绑定了跳转指令,等待后续把要跳转的地址找到合适的内存储存后,一起修复好//    即,修复前 ldr pc,xxx  修复后 ldr pc, [pc,offset],pc+offset就是存储要跳转地址的内存    ThumbRelocLabelEntry *label = new ThumbRelocLabelEntry(val + 1, true);    _ AppendRelocLabelEntry(label); //    imm5 = bits(0x4 >> 1, 1, 5);    //dkl 修复      imm5 = bits(0, 1, 5);    i = bit(0x4 >> 1, 6);     _ EmitInt16((instr & 0xfd07) | imm5 << 3 | i << 9);    _ t1_nop(); // manual align    _ t2_b(0);    //这个label持有要跳转过去的地址,跳转采用ldr pc 的方式,这个label同时又采用PseudoLabelInstruction结构体绑定到指令上,所以,已经具备了跳转的全部信息了,    // 只差把跳转地址存到合适的位置,然后修复ldr即可,修复工作好像是后面统一处理, thumb_turbo_assembler_.RelocBind();在这里修正    _ T2_Ldr(pc, label);     is_instr_relocated = true;  }     // if the instr do not needed relocate, just rewrite the origin  if (!is_instr_relocated) {#if 0        if (from_pc % Thumb2_INST_LEN)            _ t1_nop();#endif    _ EmitInt16(instr);  }}
至此,代码详解也结束了,其实代码修复主要是解析指令,这一块稍微繁琐一点。


收获

最主要的是有了一次完整的源码阅读经验,同时学到了一些工程技巧,比如c++的链表技巧。

先定义一个通用链表头:

具体数据节点:
这样写的好处是,遍历链表时,直接采用NodHead指针去遍历,然后需要读取数据的时候,把NodHead转为 EntryNod即可,因为结构体指针就是结构体首项地址,这NodHead和EntryNod值都是一样的。这样就可以写出一个通用的链表模板,以后设么链表都可以用这套模板,把EntryNod改改就行。
 
第二个收获是,一些经典宏,比如##可以连接字符串,比如这个宏,可以通过类类型,类成员名称,类成员地址取到类的this指针,参考 container_of宏(https://blog.csdn.net/lezardfu/article/details/44916167)。
#define offsetof(t, d) __builtin_offsetof(t, d) #define container_of(ptr, type, member)                                                                                \  ({                                                                                                                   \    const __typeof(((type *)0)->member) *__mptr = (ptr);                                                               \    (type *)((char *)__mptr - offsetof(type, member));                                                                 \  })
同时,Dobby有自己的内存分配模块,他会把每次分配的相同属性的内存记录下来,等到需要申请内存的时候,先查看已经分配的内存是否有可用的,这样就避免了频繁的内存分配。


使用Dobby过程中遇到的问题

我总共遇到了三个问题,第一个问题是插桩的时候,正好那条指令正在执行,这样就会出错。修复办法有两个,第一个是在so加载的第一时间就完成插桩;第二个办法是,通过异常使进程中断,自定义信号处理函数,在异常处理过程中完成插桩。
我采用的是第一个办法,在so加载的第一时间就完成插桩.android的so加载最终都是通过linker的do_dlopen加载so,而do_dlopen会调用
soinfo* si = find_library(ns, translated_name, flags, extinfo, caller);在这里可以拿到soinfo指针,有了soinfo就有了一切。
所以只需要hook这个函数即可。实际上,在aosp10,这个函数是内联的,所以我hook了find_library中的si->increment_ref_count();这个函数拿到的soinfo指针。

第二个问题是,mproterct问题,因为需要patch 的原指令,但是原指令一般内存属性是只读的,需要使用mprotect去把属性改成可写,mprotect是按页整数倍进行修改的,Dobby会把需要插桩的那条指令所在页面权限修改,大多数情况下没有问题。
但是偶尔,被插桩的指令位于页面底部,而patch又需要至少8字节,这就导致了会横跨两个页面,而Dobby只是修改了一个页面,需要注意一下。

第三个问题,sigll,这个问题主要是指令修复的时候,没有生成正确的汇编,跳错地方了,这个需要针对性的根据源码来修复了,这也是我去看Dobby源码的原因。


总结

目前逆向工具中,ida是静态分析的王者,frida(估计)是动态分析的王者,但是frida是函数级的工作,粒度不够,需要Dobby配合使用,即可达到指令级的动态分析。

调试器虽然也可以达到目的,但是调试器容易引入很多其他的问题,我开始就是使用gdb的,但是遇到了很多问题,比如gdb把进程暂停了,android一些广播超时,就把我的进程杀了,或者不小心摸了一下屏幕,屏幕响应超时,又把我杀了,有时候gdb识别不出thumb指令,还得给它手动设置模式,体验不好;不过gdb有个内存断点,估计有时候不得不用一下。

看雪ID:KerryS

https://bbs.pediy.com/user-home-844633.htm

*本文由看雪论坛 KerryS 原创,转载请注明来自看雪社区

# 往期推荐

1.AFL速通——流程及afl-fuzz.c源码简析

2.sql注入学习笔记

3.文件上传和文件包含的各种姿势

4.React Native Hermes 逆向实践

5.2022CISCN初赛 ez_usb WriteUp

6.APT Turla样本分析

球分享

球点赞

球在看

点击“阅读原文”,了解更多!


文章来源: http://mp.weixin.qq.com/s?__biz=MjM5NTc2MDYxMw==&mid=2458459349&idx=1&sn=c7befdac063330a9ada2e3d1b0e396ef&chksm=b18e2c5f86f9a5492113d4584d85a484eedb3384f8e4ad14235273dd830e4cd57615f08ec926#rd
如有侵权请联系:admin#unsafe.sh