Android Dynamic Framework :
Native Hook Mechanism in Bionic Linker
Mai-Hsuan Chia
Shih-Wei Liao
Department of Computer Science and Information Engineering
National Taiwan University

Outline
● Background
● Motivation
● Native Hook Mechanism
● Experiment
● Applications
● Future works
● Conclusion

Background
● JNI
● Android Dynamic Framework
● Bionic

JNI
● Enable Java code can call or can be called by native applications

JNI
Java method
JNI
Native functionC/C++
Java

Java calls native
class HelloWorld {
private native void print(); // print() is native function
public static void main(String[] args) {
new HelloWorld().print();
}
static {
System.loadLibrary("hello"); // This loads libhello.so
}
}

● A framework which is able to dynamically replace Java
methods in ART Runtime without modifying APKs.
Android Dynamic Framework

Android Dynamic Framework
Class A Class B
HookTable
...
class
linker
Method A1
Method A2
Method B1
Method B2

Android Dynamic Framework
Class A Class B
HookTable
...
class
linker
Method A1
Method A2
Method B1
Method B2
0. Do linking

Android Dynamic Framework
Class A Class B
HookTable
...
class
linker
Method A1
Method A2
Method B1
Method B2
1. Query HookTable

Android Dynamic Framework
Class A Class B
HookTable
...
class
linker
Method A1
Method A2
Method B1
Method B2
Replace
ClassA::A1 with
ClassB::B1
1. Query HookTable

Android Dynamic Framework
Class A Class B
HookTable
...
class
linker
Method A2
Method B1
Method B2
Method B1
2. Do method hooking

● C library in Android
● Forked from BSDs rather than from GNU/Linux
○ To avoid license problems
● Smaller
● Faster
Bionic

● Components
○ libc
○ libm
○ libdl (written from scratch)
○ dynamic linker
■ /system/bin/linker (written from scratch)
Bionic

Motivation
● Only Java methods can be replaced in Android Dynamic
Framework

Class A
Method A2
Method B1
Class B
Method B1
Method B2
JNI
libd.so
Func D1
Func D2
libe.so
Func E1
Func E2
(1) method hook
Method A3
libc.so
Func C1
Func C2
native call
hooking path

Class A
Method A2
Method B1
Class B
Method B1
Method B2
JNI
libd.so
Func D1
Func D2
libe.so
Func E1
Func E2
(1) method hook
Method A3
libc.so
Func C1
Func C2
native call
hooking path

Class A
Method A2
Method B1
Class B
Method B1
Method B2
JNI
libd.so
Func D1
Func D2
libe.so
Func E1
Func E2
(1) method hook
Method A3
libc.so
Func D1
Func C2
native call
hooking path
(2) dlopen native hook
(1) method hook

Class A
Method A2
Method B1
Class B
Method B1
Method B2
JNI
libd.so
Func D1
Func D2
libe.so
Func E1
Func E2
(1) method hook
Method A3
libc.so
Func D1
Func C2
native call
hooking path
(2) dlopen native hook
(1) method hook
(2) dlopen native hook

Class A
Method A2
Method B1
Class B
Method B1
Method B2
JNI
libd.so
Func D1
Func E2
libe.so
Func E1
Func E2
(1) method hook
Method A3
libc.so
Func D1
Func C2
native call
hooking path
(2) dlopen native hook
(1) method hook
(2) dlopen native hook
(3) native to native hook

Motivation
● (1) method hook can be done in the existing Android Dynamic
Framework
● However, (2) dlopen native hook and (3) native to native hook
cannot not be done.

Motivation
● Native hook mechanism can do both (2) dlopen native hook and
(3) native to native hook

Motivation
With Native hook mechanism integrated,
Android Dynamic Framework can be more complete and powerful

Native hook mechanism
● Implemented in Bionic Linker

Review
● How Bionic Linker loads an executable
● Dynamic linking flow
● Dynamic loading flow

How Bionic Linker loads an executable

OS creates a process image
● Based on the interpreter’s segments. high
low
Memory space
/system/bin/linker
linker

Linker links itself
● __linker_init() high
low
Memory space
/system/bin/linker

Load the executable
● __linker_init_post_relocation()
/system/bin/linker
high
low
Memory space
exe
executable

Get needed libraries names
● __linker_init_post_relocation()
/system/bin/linker
high
low
Memory space
executable
exe
.dynamic
DT_NEEDED
ptr_to_liba1.so_name
DT_NEEDED
ptr_to_liba2.so_name
…
DT_NULL

Get needed libraries names
● __linker_init_post_relocation()
/system/bin/linker
high
low
Memory space
executable
exe
DT_NEEDED
ptr_to_liba1.so_name
DT_NEEDED
ptr_to_liba2.so_name
…
DT_NULL
.dynamic
char needed_libraries_names[] = {
“liba1.so”,
“liba2.so”
}

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree
exe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree
exe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree
exe
liba1.so
liba2.so
libb1.so
libb2.so
libb3.so
libb4.so
...
...
Loaded
Not
Loaded
p.s.

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree
exe
liba1.so
liba2.so
libb2.so
libb3.so
libb4.so
...
...
Loaded
Not
Loaded
p.s.
libb1.so

Load needed libraries
● find_libaries(exe, needed_libraries_names)
○ step 1 : load libraries and build dependencies tree
exe
liba1.so
liba2.so
libb2.so
libb3.so
libb4.so
...
...
...
...
...
...
...
Loaded
Not
Loaded
p.s.
libb1.so

Load needed libraries
liba1.soexe liba2.so libb1.so libb2.so ...
● find_libaries(exe, needed_libraries_names)
○ step 2 : turn dependencies tree into libraries_list in
Breadth First Search(BFS) order
libraries_list
dependencies tree

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
foreach lib in libraries_list {
foreach rel in lib.dynamic_relocation_table {
symbol = rel.sym;
soinfo_do_lookup(symbol, lib, libraries_list);
}
}

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;
NOT FOUND

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;
NOT FOUND

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;
FOUND
ok

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
if sym_k is defined in lib:
sym_k = lib.find(sym)
else:
lib = lib->next;
ok

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
exe
sym_1
sym_n
...
ok
ok

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
liba1.so
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
...
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;

Link the application and all libraries
● find_libaries(exe, needed_libraries_names)
○ step 3 : relocate all to-be-relocated symbols in the
application and libraries
liba1.soexe liba2.so libb1.so libb2.so ...
libraries_list
...
sym_1
sym_n
...
if sym_1 is defined in lib:
sym_1 = lib.find(sym)
else:
lib = lib->next;
It is DONE until all libraries are linked

Jump to the application’s entry
/system/bin/linker
high
low
Memory space
executable
liba1.so
liba2.so
libb1.so
...
● jump to executable’s _start.

The executable is loaded successfully
● And start to execute
/system/bin/linker
high
low
Memory space
liba1.so
liba2.so
libb1.so
...
.text section
_start:
….
….
executable

Bionic linker linking & loading flow
● Dynamic linking flow
● Dynamic loading flow

__linker_init_post_relocation
Dynamic linking
dlopen_ext
do_dlopen
find_library
find_libraries
find_library_internal
load_library
Dynamic loading
...
load all libraries
…
relocate all symbols

Native hook mechanism
Modified codes are mainly in two parts
● Load hooking libraries in find_libraries()
○ Init native_hook_table
○ Look up native_hook_table
○ Load hooking_library
● Replace hooked_symbol with hooking_symbol in soinfo_do_lookup()
○ Look up native_hook_table
○ Replace every hooked_symbol in hooked_library with hooking_symbol in
hooking_library

Native hook file format
in /system/nh_file.txt
< hooked_lib_name:hooked_symbol:hooking_lib_name:hooking_symbol >

System flow
hooking
lib
nh_file
ROM
/system/bin/linker
__linker_init_post_relocation
find_libraries
init native_hook_table
look up native hook table
soinfo_do_lookup
look up native hook table
replace hooked symbol
with hooking symbol
New
Process
load hooking library

Load hooking libraries
linkerexe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.
liba1.so:hi:libhooking.so:ha
...
Native Hook Table

Load hooking libraries
linkerexe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
0. load liba1.so

Load hooking libraries
linkerexe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
1. look up the native
hook table
HOOKED LIB “liba1.so” FOUND

Load hooking libraries
linkerexe
liba1.so
liba2.so Loaded
Not
Loaded
p.s.
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
2. load libhooking.so
libhooking.
so

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi
hi ha
linker
0. relocate symbol

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
NOT FOUND
hi

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
FOUND
hi

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
FOUND
hi
1. look up native
hook table
liba1.so:hi is to be hooked

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
hi
1. look up native
hook table
2. find libhooking.so:ha

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
ha
3. relocate hooked_symbol “hi”
with the hooking_symbol “ha”

// in libnativehook.so
#include “native_hook.h”
void* find_lib_symbol(char* lib_name, char* symbol)
{
// Using dl_iterate_phdr() to get the symbol’s address
// in the loaded library whose name is lib_name.
…
return ptr_to_symbol;
}
Before/After hook SDK

How find_lib_symbol() works ?
With the following facts, we can get the hooked_symbol in hooked_library with
dl_iterate_phdr(callback, void* data)
● hooked_lib is loaded in the memory
● dl_iterate_phdr()iterates all loaded libraries in the process, and get each
library’s program header and base address.
● With library’s program header, we can get .dynamic segment, and therefore we get
.dynstr and .dynsym section
● With .dynsym and .dynstr, we can find the offset of hooked_symbol in hooked_lib.
● hooked_symbol_addr = base address + offset

// in libmine.so
#include “native_hook.h”
double my_sin(double x)
{
char hooked_lib[] = "/system/lib/libm.so";
char hooked_symbol[] = "sin";
double (*hooked_sin)(double) = find_lib_symbol(hooked_lib, hooked_symbol);
/*
before hook : you can do something before calling hooked_func
*/
double result = hooked_sin(x);
/*
after hook : you can do something after calling hooked_func
*/
result += 5566;
return result;
}
After hook example

After hook example
// in main.c
#include <math.h>
#include <stdio.h>
#define PI 3.14159265
int main(void)
{
double angle = 30.0;
double result = sin((angle * PI) /
180);
printf(“sin(%lf) = %lfn”, angle,
result);
return 0;
}
libm.so:sin:libmine.so:my_sin
...
Native Hook Table
$ ./main
sin(30.000000) = 5566.500000

double my_sin(double x)
{
char hooked_lib[] = "/system/lib/libm.so";
char hooked_symbol[] = "sin";
static void* cache_ptr = NULL;
double (*hooked_sin)(double) = NULL;
if (cache_ptr) {
hooked_sin = cache_ptr;
} else {
hooked_sin = find_lib_symbol(hooked_lib, hooked_symbol);
}
if (hooked_sin) {
cache_ptr = (void*)hooked_sin;
}
double result = hooked_sin(x);
result += 5566;
return result;
}
Before/After hook with cache

Experiment
1,000 100,000 1,000,000 10,000,000
Baseline 0.10 0.14 0.52 4.07
Normal hook 0.20 0.23 0.60 4.15
Before/After hook without cache 0.25 1.9 17.12 169.03
Before/After hook with cache 0.22 0.24 0.69 4.77
iterations

Experiment
169.03

Applications
● Profiling
● Boosting apps performance
● Security sandbox

Profiling
Target function
Before hook
After hook
● Input Distribution Analysis
● Function call Analysis
● Output Analysis

● Hook functions that affect the performance of applications in
Android
● Scenario
○ Functions in libm.so are not good enough for some special
purpose, we can hook the function with the optimized one.
Boosting apps performance

libm_opt.so
optimized_sin:
...
libbenchmark.so
getScore:
…
call <sin>
App
libm.so
sin:
...
JNI

libm_opt.so
optimized_sin:
...
libbenchmark.so
getScore:
…
call <sin>
App
libm.so
sin:
...
JNI
Replace ‘sin’ with
‘optimized_sin’

Security sandbox
● Use “before hook” to hook the open()in libc
● Examine the filename and other parameters in advance
○ If the to-be-written file is a critical file, we let the app open another file to write
without consciousness.

Security sandbox
f = open(“/data/critical.txt”, ‘w’);
...
modifying critical.txt ...
...
App
Sandbox

Security sandbox
f = open(“/data/critical.txt”, ‘w’);
...
modifying critical.txt ...
...
App
Sandbox
/data/critical.txt
should not be
modified.

Security sandbox
f = open(“/data/critical.txt”, ‘w’);
...
modifying critical.txt ...
...
App
Sandbox
f = open(“/data/another.txt”, ‘w’);
In the sandbox, app is deceived to write to
“/data/another.txt” instead of
“/data/critical.txt”.

Security sandbox
App
Sandbox
f = open(“/data/another.txt”, ‘w’);
f = open(“/data/another.txt”, ‘w’);
...
modifying another.txt ...
...

● Provide more easy-to-use API for Native Hook in Android
○ Native Hook SDK
Future works

● Completely integrate Native Hook into Android Dynamic
Framework
○ Provide hooking between Java method and native functions.
Future works
Integrated Hook Table
liba.so:funca:libb.so:funcb # hook native to native
classA:methoda:classB:methodb # hook java to java
classA:methoda:libb.so:funcb # hook java to native
libb.so:funcb:classA:methoda # hook native to java
...

Conclusion
● Native Hook mechanism is a strong and useful framework in
Android allowing developers to replace native functions at
runtime without modifying the existing functions.
● Native Hook is more powerful than Java method hook
mechanisms because it is implemented in Bionic Linker.
● With Before/After hook mechanism, you can do whatever you
want before/after any existing function.
● With Native Hook enabled, it suffers only little overhead to
load nh_file and hooking libraries.

Q & A

Thank you for your
listening

Backup slides

void* find_lib_symbol(char* lib_name, char* symbol)
{
// Using dl_iterate_phdr() to get the symbol’s address
// in the loaded library whose name is lib_name.
static void* unordered_map<std::string, void*> cache = nullptr;
std::string lib_symbol = std::string(lib_name) + symbol;
if (cache) {
unordered_map<std::string, void*>::iterator it = cache.find(lib_symbol);
if (it != cache.end()) {
return it->second;
}
}
…
// find ptr_to_symbol
if (ptr_to_symbol) {
cache[lib_symbol] = ptr_to_symbol;
}
return ptr_to_symbol;
}
Before/After hook with cache in find_lib_symbol

Replace hooked_symbol with hooking_symbol
liba1.soexe liba2.so libhooking.s
o
libraries_list
liba1.so:hi:libhooking.so:ha
...
Native Hook Table
exe
hi ha
linker
FOUND
hi
1. look up native
hook table
liba1.so:hi is to be hooked
2. find libhooking.so:ha