1 # -*- coding: utf-8 -*-
2 # kate: space-indent on; indent-width 2; mixedindent off; indent-mode python;
4 # Copyright (C) 2009 Amand 'alrj' Tihon <amand.tihon@alrj.org>
6 # This file is part of bold, the Byte Optimized Linker.
8 # You can redistribute this file and/or modify it under the terms of the
9 # GNU General Public License as published by the Free Software Foundation,
10 # either version 3 of the License or (at your option) any later version.
13 Main entry point for the bold linker.
16 from constants import *
17 from BinArray import BinArray
18 from elf import Elf64, Elf64_Phdr, Elf64_Shdr, TextSegment, DataSegment
19 from elf import SStrtab, SSymtab, SProgBits, SNobits, Dynamic, Interpreter
21 from ctypes.util import find_library
26 """Caculate the hash of the function name.
27 @param name: the string to hash
28 @return: 32 bits hash value.
32 h = ((h * 0x21) ^ ord(c)) & 0xffffffff
36 class BoldLinker(object):
37 """A Linker object takes one or more objects files, optional shared libs,
38 and arranges all this in an executable.
46 self.entry_point = "_start"
48 self.global_symbols = {}
49 self.undefined_symbols = set()
50 self.common_symbols = set()
53 def add_object(self, filename):
54 """Add a relocatable file as input.
55 @param filename: path to relocatable object file to add
63 def build_symbols_tables(self):
64 """Find out the globally available symbols, as well as the globally
65 undefined ones (which should be found in external libraries."""
67 # Gather the "extern" and common symbols from each input files.
69 self.undefined_symbols.update(i.undefined_symbols)
70 self.common_symbols.update(i.common_symbols)
72 # Make a dict with all the symbols declared globally.
73 # Key is the symbol name, value will later be set to the final
74 # virtual address. Currently, we're only interrested in the declaration.
75 # The virtual addresses are set to None, they'll be resolved later.
77 for s in i.global_symbols:
78 if s in self.global_symbols:
79 raise RedefinedSymbol(s)
80 self.global_symbols[s] = None
82 # Add a few useful symbols. They'll be resolved ater as well.
83 self.global_symbols["_dt_debug"] = None
84 self.global_symbols["_DYNAMIC"] = None
86 # Find out which symbols aren't really defined anywhere
87 self.undefined_symbols.difference_update(self.global_symbols)
89 # A symbol declared as COMMON in one object may very well have been
90 # defined in another. In this case, it will be present in the
92 # Take a copy because we can't change the set's size inside the loop
93 for i in self.common_symbols.copy():
94 if i[0] in self.global_symbols:
95 self.common_symbols.remove(i)
98 def build_external(self, with_jump=False, align_jump=False):
100 Generate a fake relocatable object, for dynamic linking.
101 This object is then automatically added in the list of ebjects to link.
102 TODO: This part is extremely non-portable.
105 # Find out all the undefined symbols. They're the one we'll need to resolve
107 symbols = sorted(list(self.undefined_symbols))
109 # Those three will soon be known...
110 symbols.remove('_bold__functions_count')
111 symbols.remove('_bold__functions_hash')
112 symbols.remove('_bold__functions_pointers')
114 # Create the fake ELF object.
115 fo = Elf64() # Don't care about most parts of ELF header (?)
116 fo.filename = "Internal dynamic linker"
118 # We need a .data section, a .bss section and a possibly a .text section
119 data_shdr = Elf64_Shdr()
120 data_shdr.sh_type = SHT_PROGBITS
121 data_shdr.sh_flags = (SHF_WRITE | SHF_ALLOC)
122 data_shdr.sh_size = len(symbols) * 4
123 fmt = "<" + "I" * len(symbols)
124 data_shdr.content = BinArray(struct.pack(fmt, *[hash_name(s) for s in symbols]))
125 fo.shdrs.append(data_shdr)
126 fo.sections['.data'] = data_shdr
128 # .bss will contain pointers to resolved external functions, as well as
129 # the COMMON symbols (from C tentative declaration).
130 bss_size = len(symbols) * 8
131 for s_name, s_size, s_alignment in self.common_symbols:
132 padding = (s_alignment - (bss_size % s_alignment))
133 bss_size += padding + s_size
135 bss_shdr = Elf64_Shdr()
136 bss_shdr.sh_type = SHT_NOBITS
137 bss_shdr.sh_flags = (SHF_WRITE | SHF_ALLOC)
138 bss_shdr.sh_size = bss_size
139 bss_shdr.content = BinArray("")
140 fo.shdrs.append(bss_shdr)
141 fo.sections['.bss'] = bss_shdr
144 text_shdr = Elf64_Shdr()
145 text_shdr.sh_type = SHT_PROGBITS
146 text_shdr.sh_flags = (SHF_ALLOC | SHF_EXECINSTR)
147 text_shdr.sh_size = len(symbols) * 8
149 fmt = '\xff\x25\x00\x00\x00\x00\x00\x00' # ff 25 = jmp [rel label]
152 fmt = '\xff\x25\x00\x00\x00\x00'
154 text_shdr.content = BinArray(fmt * len(symbols))
155 fo.shdrs.append(text_shdr)
156 fo.sections['.text'] = text_shdr
158 # Cheating here. All symbols declared as global so we don't need to create
159 # a symtab from scratch.
160 fo.global_symbols = {}
161 fo.global_symbols['_bold__functions_count'] = (SHN_ABS, len(symbols))
162 fo.global_symbols['_bold__functions_hash'] = (data_shdr, 0)
163 fo.global_symbols['_bold__functions_pointers'] = (bss_shdr, 0)
165 # The COMMON symbols. Assign an offset in .bss, declare as global.
166 bss_common_offset = len(symbols) * 8
167 for s_name, s_size, s_alignment in self.common_symbols:
168 padding = (s_alignment - (bss_common_offset % s_alignment))
169 bss_common_offset += padding
170 fo.global_symbols[s_name] = (bss_shdr, bss_common_offset)
171 bss_common_offset += s_size
174 for n, i in enumerate(symbols):
175 # The hash is always in .data
176 h = "_bold__hash_%s" % i
177 fo.global_symbols[h] = (data_shdr, n * 4) # Section, offset
180 # the symbol is in .text, can be called directly
181 fo.global_symbols[i] = (text_shdr, n * jmp_size)
182 # another symbol can be used to reference the pointer, just in case.
184 fo.global_symbols[p] = (bss_shdr, n * 8)
187 # The symbol is in .bss, must be called indirectly
188 fo.global_symbols[i] = (bss_shdr, n * 8)
191 # Add relocation entries for the jumps
192 # Relocation will be done for the .text, for every jmp instruction.
194 rela_shdr = Elf64_Shdr()
195 rela_shdr.sh_type = SHT_RELA
196 rela_shdr.target = text_shdr
197 rela_shdr.sh_flags = 0
198 rela_shdr._content = dummy() # We only need a container for relatab...
199 relatab = [] # Prepare a relatab
200 rela_shdr.content.relatab = relatab
202 for n, i in enumerate(symbols):
203 # Create a relocation entry for each symbol
205 reloc.r_offset = (n * jmp_size) + 2 # Beginning of the cell to update
207 reloc.r_type = R_X86_64_PC32
208 reloc.symbol = dummy()
209 reloc.symbol.st_shndx = SHN_UNDEF
210 reloc.symbol.name = "_bold__%s" % i
211 relatab.append(reloc)
212 fo.shdrs.append(rela_shdr)
213 fo.sections['.rela.text'] = rela_shdr
215 # Ok, let's add this fake object
219 def add_shlib(self, libname):
220 """Add a shared library to link against."""
221 # Note : we use ctypes' find_library to find the real name
222 fullname = find_library(libname)
224 raise LibNotFound(libname)
225 self.shlibs.append(fullname)
229 """Do the actual linking."""
230 # Prepare two segments. One for .text, the other for .data + .bss
231 self.text_segment = TextSegment()
232 # .data will be mapped 0x100000 bytes further
233 self.data_segment = DataSegment(align=0x100000)
234 self.output.add_segment(self.text_segment)
235 self.output.add_segment(self.data_segment)
237 # Adjust the ELF header
238 self.output.header.e_ident.make_default_amd64()
239 self.output.header.e_phoff = self.output.header.size
240 self.output.header.e_type = ET_EXEC
241 # Elf header lies inside .text
242 self.text_segment.add_content(self.output.header)
244 # Create the four Program Headers. They'll be inside .text
245 # The first Program Header defines .text
246 ph_text = Elf64_Phdr()
247 ph_text.p_type = PT_LOAD
248 ph_text.p_align = 0x100000
249 self.output.add_phdr(ph_text)
250 self.text_segment.add_content(ph_text)
252 # Second one defines .data + .bss
253 ph_data = Elf64_Phdr()
254 ph_data.p_type = PT_LOAD
255 ph_data.p_align = 0x100000
256 self.output.add_phdr(ph_data)
257 self.text_segment.add_content(ph_data)
259 # Third one is only there to define the DYNAMIC section
260 ph_dynamic = Elf64_Phdr()
261 ph_dynamic.p_type = PT_DYNAMIC
262 self.output.add_phdr(ph_dynamic)
263 self.text_segment.add_content(ph_dynamic)
265 # Fourth one is for interp
266 ph_interp = Elf64_Phdr()
267 ph_interp.p_type = PT_INTERP
268 self.output.add_phdr(ph_interp)
269 self.text_segment.add_content(ph_interp)
271 # We have all the needed program headers, update ELF header
272 self.output.header.ph_num = len(self.output.phdrs)
274 # Create the actual content for the interpreter section
275 interp = Interpreter()
276 self.text_segment.add_content(interp)
278 # Then the Dynamic section
280 # for all the requested libs, add a reference in the Dynamic table
281 for lib in self.shlibs:
282 dynamic.add_shlib(lib)
283 # Add an empty symtab, symbol resolution is not done.
284 dynamic.add_symtab(0)
285 # And we need a DT_DEBUG
288 # This belongs to .data
289 self.data_segment.add_content(dynamic)
290 # The dynamic table links to a string table for the libs' names.
291 self.text_segment.add_content(dynamic.strtab)
293 # We can now add the interesting sections to the corresponding segments
296 # Only ALLOC sections are worth it.
297 # This might require change in the future
298 if not (sh.sh_flags & SHF_ALLOC):
301 if (sh.sh_flags & SHF_EXECINSTR):
302 self.text_segment.add_content(sh.content)
303 else: # No exec, it's for .data or .bss
304 if (sh.sh_type == SHT_NOBITS):
305 self.data_segment.add_nobits(sh.content)
307 self.data_segment.add_content(sh.content)
309 # Now, everything is at its place.
310 # Knowing the base address, we can determine where everyone will fall
311 self.output.layout(base_vaddr=0x400000)
313 # Knowing the addresses of all the parts, Program Headers can be filled
314 # This will put the correct p_offset, p_vaddr, p_filesz and p_memsz
315 ph_text.update_from_content(self.text_segment)
316 ph_data.update_from_content(self.data_segment)
317 ph_interp.update_from_content(interp)
318 ph_dynamic.update_from_content(dynamic)
320 # All parts are at their final address, find out the symbols' addresses
322 for s in i.global_symbols:
323 # Final address is the section's base address + the symbol's offset
324 if i.global_symbols[s][0] == SHN_ABS:
325 addr = i.global_symbols[s][1]
327 addr = i.global_symbols[s][0].content.virt_addr
328 addr += i.global_symbols[s][1]
330 self.global_symbols[s] = addr
332 # Resolve the few useful symbols
333 self.global_symbols["_dt_debug"] = dynamic.dt_debug_address
334 self.global_symbols["_DYNAMIC"] = dynamic.virt_addr
336 # We can now do the actual relocation
338 i.apply_relocation(self.global_symbols)
340 # And update the ELF header with the entry point
341 if not self.entry_point in self.global_symbols:
342 raise UndefinedSymbol(self.entry_point)
343 self.output.header.e_entry = self.global_symbols[self.entry_point]
348 def toBinArray(self):
349 return self.output.toBinArray()
352 def tofile(self, file_object):
353 return self.output.toBinArray().tofile(file_object)