/xDL

:fire: xDL is an enhanced implementation of the Android DL series functions.

Primary LanguageCMIT LicenseMIT

xDL

xDL is an enhanced implementation of the Android DL series functions.

简体中文

Features

  • Enhanced dlopen() + dlsym() + dladdr().
    • Bypass the restrictions of Android 7.0+ linker namespace.
    • Lookup dynamic link symbols in .dynsym.
    • Lookup debuging symbols in .symtab and ".symtab in .gnu_debugdata".
  • Enhanced dl_iterate_phdr().
    • Compatible with Android 4.x on ARM32.
    • Including linker / linker64 (for Android <= 8.x).
    • Return full pathname instead of basename (for Android 5.x).
    • Return app_process32 / app_process64 instead of package name.
  • Support Android 4.1 - 15 (API level 16 - 35).
  • Support armeabi-v7a, arm64-v8a, x86 and x86_64.
  • MIT licensed.

Artifacts Size

If xDL is compiled into an independent dynamic library:

ABI Compressed (KB) Uncompressed (KB)
armeabi-v7a 7.9 15
arm64-v8a 9.1 20
x86 9.0 18
x86_64 9.1 20

Usage

1. Add dependency in build.gradle

xDL is published on Maven Central, and uses Prefab package format for native dependencies, which is supported by Android Gradle Plugin 4.0+.

android {
    buildFeatures {
        prefab true
    }
}

dependencies {
    implementation 'io.github.hexhacking:xdl:2.2.0'
}

NOTE:

  1. Starting from version 2.0.0 of xDL, group ID changed from io.hexhacking to io.github.hexhacking.
version range group ID artifact ID Repository URL
[1.0.3, 1.2.1] io.hexhacking xdl repo
[2.0.0, ) io.github.hexhacking xdl repo
  1. xDL uses the prefab package schema v2, which is configured by default since Android Gradle Plugin 7.1.0. If you are using Android Gradle Plugin earlier than 7.1.0, please add the following configuration to gradle.properties:
android.prefabVersion=2.0.0

2. Add dependency in CMakeLists.txt or Android.mk

CMakeLists.txt

find_package(xdl REQUIRED CONFIG)

add_library(mylib SHARED mylib.c)
target_link_libraries(mylib xdl::xdl)

Android.mk

include $(CLEAR_VARS)
LOCAL_MODULE           := mylib
LOCAL_SRC_FILES        := mylib.c
LOCAL_SHARED_LIBRARIES += xdl
include $(BUILD_SHARED_LIBRARY)

$(call import-module,prefab/xdl)

3. Specify one or more ABI(s) you need

android {
    defaultConfig {
        ndk {
            abiFilters 'armeabi-v7a', 'arm64-v8a', 'x86', 'x86_64'
        }
    }
}

4. Add packaging options

If you are using xDL in an SDK project, you may need to avoid packaging libxdl.so into your AAR, so as not to encounter duplicate libxdl.so file when packaging the app project.

android {
    packagingOptions {
        exclude '**/libxdl.so'
    }
}

On the other hand, if you are using xDL in an APP project, you may need to add some options to deal with conflicts caused by duplicate libxdl.so file.

android {
    packagingOptions {
        pickFirst '**/libxdl.so'
    }
}

There is a sample app in the xdl-sample folder you can refer to.

API

#include "xdl.h"

1. xdl_open() and xdl_close()

#define XDL_DEFAULT           0x00
#define XDL_TRY_FORCE_LOAD    0x01
#define XDL_ALWAYS_FORCE_LOAD 0x02

void *xdl_open(const char *filename, int flags);
void *xdl_open2(struct dl_phdr_info *info);
void *xdl_close(void *handle);

They are very similar to dlopen() and dlclose(). But xdl_open() can bypass the restrictions of Android 7.0+ linker namespace.

Depending on the value of the flags parameter, the behavior of xdl_open() will have some differences:

  • XDL_DEFAULT: If the library has been loaded into memory, xdl_open() will not dlopen() it again. (But it will still return a valid handle)
  • XDL_TRY_FORCE_LOAD: If the library has not been loaded into memory, xdl_open() will try to dlopen() it.
  • XDL_ALWAYS_FORCE_LOAD: xdl_open() will always dlopen() the library.

If xdl_open() really uses dlopen() to load the library, xdl_close() will return the handle from linker (the return value of dlopen()), and then you can decide whether and when to close it with standard dlclose(). Otherwise, NULL will be returned.

filename can be basename or full pathname. However, Android linker has used the namespace mechanism since 7.0. If you pass basename, you need to make sure that no duplicate ELF is loaded into the current process. xdl_open() will only return the first matching ELF. Please consider this fragment of /proc/self/maps on Android 10:

756fc2c000-756fc7c000 r--p 00000000 fd:03 2985  /system/lib64/vndk-sp-29/libc++.so
756fc7c000-756fcee000 --xp 00050000 fd:03 2985  /system/lib64/vndk-sp-29/libc++.so
756fcee000-756fcef000 rw-p 000c2000 fd:03 2985  /system/lib64/vndk-sp-29/libc++.so
756fcef000-756fcf7000 r--p 000c3000 fd:03 2985  /system/lib64/vndk-sp-29/libc++.so
7571fdd000-757202d000 r--p 00000000 07:38 20    /apex/com.android.conscrypt/lib64/libc++.so
757202d000-757209f000 --xp 00050000 07:38 20    /apex/com.android.conscrypt/lib64/libc++.so
757209f000-75720a0000 rw-p 000c2000 07:38 20    /apex/com.android.conscrypt/lib64/libc++.so
75720a0000-75720a8000 r--p 000c3000 07:38 20    /apex/com.android.conscrypt/lib64/libc++.so
760b9df000-760ba2f000 r--p 00000000 fd:03 2441  /system/lib64/libc++.so
760ba2f000-760baa1000 --xp 00050000 fd:03 2441  /system/lib64/libc++.so
760baa1000-760baa2000 rw-p 000c2000 fd:03 2441  /system/lib64/libc++.so
760baa2000-760baaa000 r--p 000c3000 fd:03 2441  /system/lib64/libc++.so

xdl_open2() creates a handle from struct dl_phdr_info. xdl_open2() is always XDL_DEFAULT semantics, i.e. it will not try to load ELF with dlopen().

2. xdl_sym() and xdl_dsym()

void *xdl_sym(void *handle, const char *symbol, size_t *symbol_size);
void *xdl_dsym(void *handle, const char *symbol, size_t *symbol_size);

They are very similar to dlsym(). They all takes a "handle" of an ELF returned by xdl_open() and the null-terminated symbol name, returning the address where that symbol is loaded into memory.

If the symbol_size parameter is not NULL, it will be assigned as "the bytes occupied by the content corresponding to the symbol in the ELF". If you don't need this information, just pass NULL.

xdl_sym() lookup "dynamic link symbols" in .dynsym as dlsym() does.

xdl_dsym() lookup "debuging symbols" in .symtab and ".symtab in .gnu_debugdata".

Notice:

  • The symbol sets in .dynsym and .symtab do not contain each other. Some symbols only exist in .dynsym, and some only exist in .symtab. You may need to use tools such as readelf to determine which ELF section the symbol you are looking for is in.
  • xdl_dsym() needs to load debuging symbols from disk file, and xdl_sym() only lookup dynamic link symbols from memory. So xdl_dsym() runs slower than xdl_sym().
  • The dynamic linker only uses symbols in .dynsym. The debugger actually uses the symbols in both .dynsym and .symtab.

3. xdl_addr()

typedef struct
{
    const char       *dli_fname;
    void             *dli_fbase;
    const char       *dli_sname;
    void             *dli_saddr;
    size_t            dli_ssize;
    const ElfW(Phdr) *dlpi_phdr;
    size_t            dlpi_phnum;
} xdl_info_t;

#define XDL_DEFAULT 0x00
#define XDL_NON_SYM 0x01

int xdl_addr(void *addr, xdl_info_t *info, void **cache);
int xdl_addr4(void *addr, xdl_info_t *info, void **cache, int flags);
void xdl_addr_clean(void **cache);

xdl_addr() is similar to dladdr(). But there are a few differences:

  • xdl_addr() can lookup not only dynamic link symbols, but also debugging symbols.
  • xdl_addr() uses the xdl_info_t structure instead of the Dl_info structure, which contains more extended information: dli_ssize is the number of bytes occupied by the current symbol; dlpi_phdr points to the program headers array of the ELF where the current symbol is located; dlpi_phnum is the number of elements in the dlpi_phdr array.
  • xdl_addr() needs to pass an additional parameter (cache), which will cache the ELF handle opened during the execution of xdl_addr(). The purpose of caching is to make subsequent executions of xdl_addr() of the same ELF faster. When you do not need to execute xdl_addr(), please use xdl_addr_clean() to clear the cache. For example:
void *cache = NULL;
xdl_info_t info;
xdl_addr(addr_1, &info, &cache);
xdl_addr(addr_2, &info, &cache);
xdl_addr(addr_3, &info, &cache);
xdl_addr_clean(&cache);
  • xdl_addr4() is similar to xdl_addr(), except that it adds the flags parameter. When the flags value is XDL_DEFAULT, the behavior of xdl_addr4() is the same as xdl_addr(). When the flags value is XDL_NON_SYM, xdl_addr4() will not obtain symbol-related information (the values of dli_sname, dli_saddr, and dli_ssize in xdl_info_t are all 0).

4. xdl_iterate_phdr()

#define XDL_DEFAULT       0x00
#define XDL_FULL_PATHNAME 0x01

int xdl_iterate_phdr(int (*callback)(struct dl_phdr_info *, size_t, void *), void *data, int flags);

xdl_iterate_phdr() is similar to dl_iterate_phdr(). But xdl_iterate_phdr() is compatible with android 4.x on ARM32, and always including linker / linker64.

xdl_iterate_phdr() has an additional "flags" parameter, one or more flags can be bitwise-or'd in it:

  • XDL_DEFAULT: Default behavior.
  • XDL_FULL_PATHNAME: Always return full pathname instead of basename.

These flags are needed because these capabilities require additional execution time, and you don't always need them.

5. xdl_info()

#define XDL_DI_DLINFO 1  // type of info: xdl_info_t

int xdl_info(void *handle, int request, void *info);

xdl_info() is similar to dlinfo(). xdl_info() obtains information about the dynamically loaded object referred to by handle (obtained by an earlier call to xdl_open).

The only request parameter currently supported is XDL_DI_DLINFO, which means to return data of type xdl_info_t through the info parameter (note that the values of dli_sname, dli_saddr, dli_ssize in the returned xdl_info_t at this time both are 0).

On success, xdl_info() returns 0. On failure, it returns -1.

Support

Contributing

License

xDL is MIT licensed, as found in the LICENSE file.

History

xCrash 2.x contains a very rudimentary module xc_dl for searching system library symbols, which has many problems in performance and compatibility. xCrash 2.x uses it to search a few symbols from libart, libc and libc++.

Later, some other projects began to use the xc_dl module alone, including in some performance-sensitive usage scenarios. At this time, we began to realize that we need to rewrite this module, and we need a better implementation.