1 The design of crazy_linker: 2 =========================== 3 4 Introduction: 5 ------------- 6 7 A system linker (e.g. ld.so on Linux, or /system/bin/linker on Android), is a 8 particularly sophisticated piece of code because it is used to load and start 9 _executables_ on the system. This requires dealing with really low-level 10 details like: 11 12 - The way the kernel loads and initializes binaries into a new process. 13 14 - The way it passes initialization data (e.g. command-line arguments) to 15 the process being launched. 16 17 - Setting up the C runtime library, thread-local storage, and others properly 18 before calling main(). 19 20 - Be very careful in the way it operates, due to the fact that it will be used 21 to load set-uid programs. 22 23 - Need to support a flurry of exotic flags and environment variables that 24 affect runtime behaviour in "interesting" but mostly unpredictable ways 25 (see the manpages for dlopen, dlsym and ld.so for details). 26 27 Add to this that most of this must be done without the C library being loaded or 28 initialized yet. No wonder this code is really complex. 29 30 By contrast, crazy_linker is a static library whose only purpose is to load 31 ELF shared libraries, inside an _existing_ executable process. This makes it 32 considerably simpler: 33 34 - The runtime environment (C library, libstdc++) is available and properly 35 initialized. 36 37 - No need to care about kernel interfaces. Everything uses mmap() and simple 38 file accesses. 39 40 - The API is simple, and straightforward (no hidden behaviour changes due to 41 environment variables). 42 43 This document explains how the crazy_linker works. A good understanding of the 44 ELF file format is recommended, though not necessary. 45 46 47 I. ELF Loading Basics: 48 ---------------------- 49 50 When it comes to loading shared libraries, an ELF file mainly consists in the 51 following parts: 52 53 - A fixed-size header that identifies the file as an ELF file and gives 54 offsets/sizes to other tables. 55 56 - A table (called the "program header table"), containing entries describing 57 'segments' of interest in the ELF file. 58 59 - A table (called the "dynamic table"), containing entries describing 60 properties of the ELF library. The most interesting ones are the list 61 of libraries the current one depends on. 62 63 - A table describing the symbols (function or global names) that the library 64 references or exports. 65 66 - One or more tables containing 'relocations'. Because libraries can be loaded 67 at any page-aligned address in memory, numerical pointers they contain must 68 be adjusted after load. That's what the relocation entries do. They can 69 also reference symbols to be found in other libraries. 70 71 The process of loading a given ELF shared library can be decomposed into 4 steps: 72 73 1) Map loadable segments into memory. 74 75 This step parses the program header table to identify 'loadable' segments, 76 reserve the corresponding address space, then map them directly into 77 memory with mmap(). 78 79 Related: src/crazy_linker_elf_loader.cpp 80 81 82 2) Load library dependencies. 83 84 This step parses the dynamic table to identify all the other shared 85 libraries the current one depends on, then will _recursively_ load them. 86 87 Related: src/crazy_linker_library_list.cpp 88 (crazy::LibraryList::LoadLibrary()) 89 90 3) Apply all relocations. 91 92 This steps adjusts all pointers within the library for the actual load 93 address. This can also reference symbols that appear in other libraries 94 loaded in step 2). 95 96 Related: src/crazy_linker_elf_relocator.cpp 97 98 4) Run constructors. 99 100 Libraries include a list of functions to be run at load time, typically 101 to perform static C++ initialization. 102 103 Related: src/crazy_linker_shared_library.cpp 104 (SharedLibrary::RunConstructors()) 105 106 Unloading a library is similar, but in reverse order: 107 108 1) Run destructors. 109 2) Unload dependencies recursively. 110 3) Unmap loadable segments. 111 112 113 II. Managing the list of libraries: 114 ----------------------------------- 115 116 It is crucial to avoid loading the same library twice in the same process, 117 otherwise some really bad undefined behaviour may happen. 118 119 This implies that, inside an Android application process, all system libraries 120 should be loaded by the system linker (because otherwise, the Dalvik-based 121 framework might load the same library on demand, at an unpredictable time). 122 123 To handle this, the crazy_linker uses a custom class (crazy::LibraryList) where 124 each entry (crazy::LibraryView) is reference-counted, and either references: 125 126 - An application shared libraries, loaded by the crazy_linker itself. 127 - A system shared libraries, loaded through the system dlopen(). 128 129 Libraries loaded by the crazy_linker are modelled by a crazy::SharedLibrary 130 object. The source code comments often refer to these objects as 131 "crazy libraries", as opposed to "system libraries". 132 133 As an example, here's a diagram that shows the list after loading a library 134 'libfoo.so' that depends on the system libraries 'libc.so', 'libm.so' and 135 'libOpenSLES.so'. 136 137 +-------------+ 138 | LibraryList | 139 +-------------+ 140 | 141 | +-------------+ 142 +----| LibraryView | ----> libc.so 143 | +-------------+ 144 | 145 | +-------------+ 146 +----| LibraryView | ----> libm.so 147 | +-------------+ 148 | 149 | +-------------+ 150 +----| LibraryView | ----> libOpenSLES.so 151 | +-------------+ 152 | 153 | +-------------+ +-------------+ 154 +----| LibraryView |----->|SharedLibrary| ---> libfoo.so 155 | +-------------+ +-------------+ 156 | 157 ___ 158 _ 159 160 System libraries are identified by name. Only the official NDK-official system 161 libraries are listed. It is likely that using crazy_linker to load non-NDK 162 system libraries will not work correctly, so don't do it. 163 164 165 III. Wrapping of linker symbols within crazy ones: 166 -------------------------------------------------- 167 168 Libraries loaded by the crazy linker are not visible to the system linker. 169 170 This means that directly calling the system dlopen() or dlsym() from a library 171 code loaded by the crazy_linker will not work properly. 172 173 To work-around this, crazy_linker redirects all linker symbols to its own 174 wrapper implementation. This redirection happens transparently. 175 176 Related: src/crazy_linker_wrappers.cpp 177 178 This also includes a few "hidden" dynamic linker symbols which are used for 179 stack-unwinding. This guarantees that C++ exception propagation works. 180 181 182 IV. GDB support: 183 ---------------- 184 185 The crazy_linker contains support code to ensure that libraries loaded with it 186 are visible through GDB at runtime. For more details, see the extensive comments 187 in src/crazy_linker_rdebug.h 188 189 190 V. Other Implementation details: 191 -------------------------------- 192 193 The crazy_linker is written in C++, but its API is completely C-based. 194 195 The implementation doesn't require any C++ STL feature (except for new 196 and delete). 197 198 Very little of the code is actually Android-specific. The target system's 199 bitness is abstracted through a C++ traits class (see src/elf_traits.h). 200 201 Written originally for Chrome, so follows the Chromium coding style. Which can 202 be enforced by using the 'clang-format' tool with: 203 204 cd /path/to/crazy_linker/ 205 find . -name "*.h" -o -name "*.cpp" | xargs clang-format -style Chromium -i 206