lib/mscat: fix logging in wscript
[Samba.git] / buildtools / wafsamba / symbols.py
blob502407c60c60b15376baa986d9599067494727f9
1 # a waf tool to extract symbols from object files or libraries
2 # using nm, producing a set of exposed defined/undefined symbols
4 import os, re, subprocess
5 from waflib import Utils, Build, Options, Logs, Errors
6 from waflib.Logs import debug
7 from samba_utils import TO_LIST, LOCAL_CACHE, get_tgt_list, os_path_relpath
9 # these are the data structures used in symbols.py:
11 # bld.env.symbol_map : dictionary mapping public symbol names to list of
12 # subsystem names where that symbol exists
14 # t.in_library : list of libraries that t is in
16 # bld.env.public_symbols: set of public symbols for each subsystem
17 # bld.env.used_symbols : set of used symbols for each subsystem
19 # bld.env.syslib_symbols: dictionary mapping system library name to set of symbols
20 # for that library
21 # bld.env.library_dict : dictionary mapping built library paths to subsystem names
23 # LOCAL_CACHE(bld, 'TARGET_TYPE') : dictionary mapping subsystem name to target type
26 def symbols_extract(bld, objfiles, dynamic=False):
27 '''extract symbols from objfile, returning a dictionary containing
28 the set of undefined and public symbols for each file'''
30 ret = {}
32 # see if we can get some results from the nm cache
33 if not bld.env.nm_cache:
34 bld.env.nm_cache = {}
36 objfiles = set(objfiles).copy()
38 remaining = set()
39 for obj in objfiles:
40 if obj in bld.env.nm_cache:
41 ret[obj] = bld.env.nm_cache[obj].copy()
42 else:
43 remaining.add(obj)
44 objfiles = remaining
46 if len(objfiles) == 0:
47 return ret
49 cmd = ["nm"]
50 if dynamic:
51 # needed for some .so files
52 cmd.append("-D")
53 cmd.extend(list(objfiles))
55 nmpipe = subprocess.Popen(cmd, stdout=subprocess.PIPE).stdout
56 if len(objfiles) == 1:
57 filename = list(objfiles)[0]
58 ret[filename] = { "PUBLIC": set(), "UNDEFINED" : set()}
60 for line in nmpipe:
61 line = line.strip()
62 if line.endswith(':'):
63 filename = line[:-1]
64 ret[filename] = { "PUBLIC": set(), "UNDEFINED" : set() }
65 continue
66 cols = line.split(" ")
67 if cols == ['']:
68 continue
69 # see if the line starts with an address
70 if len(cols) == 3:
71 symbol_type = cols[1]
72 symbol = cols[2]
73 else:
74 symbol_type = cols[0]
75 symbol = cols[1]
76 if symbol_type in "BDGTRVWSi":
77 # its a public symbol
78 ret[filename]["PUBLIC"].add(symbol)
79 elif symbol_type in "U":
80 ret[filename]["UNDEFINED"].add(symbol)
82 # add to the cache
83 for obj in objfiles:
84 if obj in ret:
85 bld.env.nm_cache[obj] = ret[obj].copy()
86 else:
87 bld.env.nm_cache[obj] = { "PUBLIC": set(), "UNDEFINED" : set() }
89 return ret
92 def real_name(name):
93 if name.find(".objlist") != -1:
94 name = name[:-8]
95 return name
98 def find_ldd_path(bld, libname, binary):
99 '''find the path to the syslib we will link against'''
100 ret = None
101 if not bld.env.syslib_paths:
102 bld.env.syslib_paths = {}
103 if libname in bld.env.syslib_paths:
104 return bld.env.syslib_paths[libname]
106 lddpipe = subprocess.Popen(['ldd', binary], stdout=subprocess.PIPE).stdout
107 for line in lddpipe:
108 line = line.strip()
109 cols = line.split(" ")
110 if len(cols) < 3 or cols[1] != "=>":
111 continue
112 if cols[0].startswith("libc."):
113 # save this one too
114 bld.env.libc_path = cols[2]
115 if cols[0].startswith(libname):
116 ret = cols[2]
117 bld.env.syslib_paths[libname] = ret
118 return ret
121 # some regular expressions for parsing readelf output
122 re_sharedlib = re.compile('Shared library: \[(.*)\]')
123 re_rpath = re.compile('Library rpath: \[(.*)\]')
125 def get_libs(bld, binname):
126 '''find the list of linked libraries for any binary or library
127 binname is the path to the binary/library on disk
129 We do this using readelf instead of ldd as we need to avoid recursing
130 into system libraries
133 # see if we can get the result from the ldd cache
134 if not bld.env.lib_cache:
135 bld.env.lib_cache = {}
136 if binname in bld.env.lib_cache:
137 return bld.env.lib_cache[binname].copy()
139 rpath = []
140 libs = set()
142 elfpipe = subprocess.Popen(['readelf', '--dynamic', binname], stdout=subprocess.PIPE).stdout
143 for line in elfpipe:
144 m = re_sharedlib.search(line)
145 if m:
146 libs.add(m.group(1))
147 m = re_rpath.search(line)
148 if m:
149 rpath.extend(m.group(1).split(":"))
151 ret = set()
152 for lib in libs:
153 found = False
154 for r in rpath:
155 path = os.path.join(r, lib)
156 if os.path.exists(path):
157 ret.add(os.path.realpath(path))
158 found = True
159 break
160 if not found:
161 # we didn't find this lib using rpath. It is probably a system
162 # library, so to find the path to it we either need to use ldd
163 # or we need to start parsing /etc/ld.so.conf* ourselves. We'll
164 # use ldd for now, even though it is slow
165 path = find_ldd_path(bld, lib, binname)
166 if path:
167 ret.add(os.path.realpath(path))
169 bld.env.lib_cache[binname] = ret.copy()
171 return ret
174 def get_libs_recursive(bld, binname, seen):
175 '''find the recursive list of linked libraries for any binary or library
176 binname is the path to the binary/library on disk. seen is a set used
177 to prevent loops
179 if binname in seen:
180 return set()
181 ret = get_libs(bld, binname)
182 seen.add(binname)
183 for lib in ret:
184 # we don't want to recurse into system libraries. If a system
185 # library that we use (eg. libcups) happens to use another library
186 # (such as libkrb5) which contains common symbols with our own
187 # libraries, then that is not an error
188 if lib in bld.env.library_dict:
189 ret = ret.union(get_libs_recursive(bld, lib, seen))
190 return ret
194 def find_syslib_path(bld, libname, deps):
195 '''find the path to the syslib we will link against'''
196 # the strategy is to use the targets that depend on the library, and run ldd
197 # on it to find the real location of the library that is used
199 linkpath = deps[0].link_task.outputs[0].abspath(bld.env)
201 if libname == "python":
202 libname += bld.env.PYTHON_VERSION
204 return find_ldd_path(bld, "lib%s" % libname.lower(), linkpath)
207 def build_symbol_sets(bld, tgt_list):
208 '''build the public_symbols and undefined_symbols attributes for each target'''
210 if bld.env.public_symbols:
211 return
213 objlist = [] # list of object file
214 objmap = {} # map from object filename to target (subsystem) name
216 for t in tgt_list:
217 t.public_symbols = set()
218 t.undefined_symbols = set()
219 t.used_symbols = set()
220 for tsk in getattr(t, 'compiled_tasks', []):
221 for output in tsk.outputs:
222 objpath = output.abspath(bld.env)
223 objlist.append(objpath)
224 objmap[objpath] = t
226 symbols = symbols_extract(bld, objlist)
227 for obj in objlist:
228 t = objmap[obj]
229 t.public_symbols = t.public_symbols.union(symbols[obj]["PUBLIC"])
230 t.undefined_symbols = t.undefined_symbols.union(symbols[obj]["UNDEFINED"])
231 t.used_symbols = t.used_symbols.union(symbols[obj]["UNDEFINED"])
233 t.undefined_symbols = t.undefined_symbols.difference(t.public_symbols)
235 # and the reverse map of public symbols to subsystem name
236 bld.env.symbol_map = {}
238 for t in tgt_list:
239 for s in t.public_symbols:
240 if not s in bld.env.symbol_map:
241 bld.env.symbol_map[s] = []
242 bld.env.symbol_map[s].append(real_name(t.sname))
244 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
246 bld.env.public_symbols = {}
247 for t in tgt_list:
248 name = real_name(t.sname)
249 if name in bld.env.public_symbols:
250 bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t.public_symbols)
251 else:
252 bld.env.public_symbols[name] = t.public_symbols
253 if t.samba_type == 'LIBRARY':
254 for dep in t.add_objects:
255 t2 = bld.get_tgen_by_name(dep)
256 bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
257 bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t2.public_symbols)
259 bld.env.used_symbols = {}
260 for t in tgt_list:
261 name = real_name(t.sname)
262 if name in bld.env.used_symbols:
263 bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t.used_symbols)
264 else:
265 bld.env.used_symbols[name] = t.used_symbols
266 if t.samba_type == 'LIBRARY':
267 for dep in t.add_objects:
268 t2 = bld.get_tgen_by_name(dep)
269 bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
270 bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t2.used_symbols)
273 def build_library_dict(bld, tgt_list):
274 '''build the library_dict dictionary'''
276 if bld.env.library_dict:
277 return
279 bld.env.library_dict = {}
281 for t in tgt_list:
282 if t.samba_type in [ 'LIBRARY', 'PYTHON' ]:
283 linkpath = os.path.realpath(t.link_task.outputs[0].abspath(bld.env))
284 bld.env.library_dict[linkpath] = t.sname
287 def build_syslib_sets(bld, tgt_list):
288 '''build the public_symbols for all syslibs'''
290 if bld.env.syslib_symbols:
291 return
293 # work out what syslibs we depend on, and what targets those are used in
294 syslibs = {}
295 objmap = {}
296 for t in tgt_list:
297 if getattr(t, 'uselib', []) and t.samba_type in [ 'LIBRARY', 'BINARY', 'PYTHON' ]:
298 for lib in t.uselib:
299 if lib in ['PYEMBED', 'PYEXT']:
300 lib = "python"
301 if not lib in syslibs:
302 syslibs[lib] = []
303 syslibs[lib].append(t)
305 # work out the paths to each syslib
306 syslib_paths = []
307 for lib in syslibs:
308 path = find_syslib_path(bld, lib, syslibs[lib])
309 if path is None:
310 Logs.warn("Unable to find syslib path for %s" % lib)
311 if path is not None:
312 syslib_paths.append(path)
313 objmap[path] = lib.lower()
315 # add in libc
316 syslib_paths.append(bld.env.libc_path)
317 objmap[bld.env.libc_path] = 'c'
319 symbols = symbols_extract(bld, syslib_paths, dynamic=True)
321 # keep a map of syslib names to public symbols
322 bld.env.syslib_symbols = {}
323 for lib in symbols:
324 bld.env.syslib_symbols[lib] = symbols[lib]["PUBLIC"]
326 # add to the map of symbols to dependencies
327 for lib in symbols:
328 for sym in symbols[lib]["PUBLIC"]:
329 if not sym in bld.env.symbol_map:
330 bld.env.symbol_map[sym] = []
331 bld.env.symbol_map[sym].append(objmap[lib])
333 # keep the libc symbols as well, as these are useful for some of the
334 # sanity checks
335 bld.env.libc_symbols = symbols[bld.env.libc_path]["PUBLIC"]
337 # add to the combined map of dependency name to public_symbols
338 for lib in bld.env.syslib_symbols:
339 bld.env.public_symbols[objmap[lib]] = bld.env.syslib_symbols[lib]
342 def build_autodeps(bld, t):
343 '''build the set of dependencies for a target'''
344 deps = set()
345 name = real_name(t.sname)
347 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
349 for sym in t.undefined_symbols:
350 if sym in t.public_symbols:
351 continue
352 if sym in bld.env.symbol_map:
353 depname = bld.env.symbol_map[sym]
354 if depname == [ name ]:
355 # self dependencies aren't interesting
356 continue
357 if t.in_library == depname:
358 # no need to depend on the library we are part of
359 continue
360 if depname[0] in ['c', 'python']:
361 # these don't go into autodeps
362 continue
363 if targets[depname[0]] in [ 'SYSLIB' ]:
364 deps.add(depname[0])
365 continue
366 t2 = bld.get_tgen_by_name(depname[0])
367 if len(t2.in_library) != 1:
368 deps.add(depname[0])
369 continue
370 if t2.in_library == t.in_library:
371 # if we're part of the same library, we don't need to autodep
372 continue
373 deps.add(t2.in_library[0])
374 t.autodeps = deps
377 def build_library_names(bld, tgt_list):
378 '''add a in_library attribute to all targets that are part of a library'''
380 if bld.env.done_build_library_names:
381 return
383 for t in tgt_list:
384 t.in_library = []
386 for t in tgt_list:
387 if t.samba_type in [ 'LIBRARY' ]:
388 for obj in t.samba_deps_extended:
389 t2 = bld.get_tgen_by_name(obj)
390 if t2 and t2.samba_type in [ 'SUBSYSTEM', 'ASN1' ]:
391 if not t.sname in t2.in_library:
392 t2.in_library.append(t.sname)
393 bld.env.done_build_library_names = True
396 def check_library_deps(bld, t):
397 '''check that all the autodeps that have mutual dependency of this
398 target are in the same library as the target'''
400 name = real_name(t.sname)
402 if len(t.in_library) > 1:
403 Logs.warn("WARNING: Target '%s' in multiple libraries: %s" % (t.sname, t.in_library))
405 for dep in t.autodeps:
406 t2 = bld.get_tgen_by_name(dep)
407 if t2 is None:
408 continue
409 for dep2 in t2.autodeps:
410 if dep2 == name and t.in_library != t2.in_library:
411 Logs.warn("WARNING: mutual dependency %s <=> %s" % (name, real_name(t2.sname)))
412 Logs.warn("Libraries should match. %s != %s" % (t.in_library, t2.in_library))
413 # raise Errors.WafError("illegal mutual dependency")
416 def check_syslib_collisions(bld, tgt_list):
417 '''check if a target has any symbol collisions with a syslib
419 We do not want any code in Samba to use a symbol name from a
420 system library. The chance of that causing problems is just too
421 high. Note that libreplace uses a rep_XX approach of renaming
422 symbols via macros
425 has_error = False
426 for t in tgt_list:
427 for lib in bld.env.syslib_symbols:
428 common = t.public_symbols.intersection(bld.env.syslib_symbols[lib])
429 if common:
430 Logs.error("ERROR: Target '%s' has symbols '%s' which is also in syslib '%s'" % (t.sname, common, lib))
431 has_error = True
432 if has_error:
433 raise Errors.WafError("symbols in common with system libraries")
436 def check_dependencies(bld, t):
437 '''check for depenencies that should be changed'''
439 if bld.get_tgen_by_name(t.sname + ".objlist"):
440 return
442 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
444 remaining = t.undefined_symbols.copy()
445 remaining = remaining.difference(t.public_symbols)
447 sname = real_name(t.sname)
449 deps = set(t.samba_deps)
450 for d in t.samba_deps:
451 if targets[d] in [ 'EMPTY', 'DISABLED', 'SYSLIB', 'GENERATOR' ]:
452 continue
453 bld.ASSERT(d in bld.env.public_symbols, "Failed to find symbol list for dependency '%s'" % d)
454 diff = remaining.intersection(bld.env.public_symbols[d])
455 if not diff and targets[sname] != 'LIBRARY':
456 Logs.info("Target '%s' has no dependency on %s" % (sname, d))
457 else:
458 remaining = remaining.difference(diff)
460 t.unsatisfied_symbols = set()
461 needed = {}
462 for sym in remaining:
463 if sym in bld.env.symbol_map:
464 dep = bld.env.symbol_map[sym]
465 if not dep[0] in needed:
466 needed[dep[0]] = set()
467 needed[dep[0]].add(sym)
468 else:
469 t.unsatisfied_symbols.add(sym)
471 for dep in needed:
472 Logs.info("Target '%s' should add dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
476 def check_syslib_dependencies(bld, t):
477 '''check for syslib depenencies'''
479 if bld.get_tgen_by_name(t.sname + ".objlist"):
480 return
482 sname = real_name(t.sname)
484 remaining = set()
486 features = TO_LIST(t.features)
487 if 'pyembed' in features or 'pyext' in features:
488 if 'python' in bld.env.public_symbols:
489 t.unsatisfied_symbols = t.unsatisfied_symbols.difference(bld.env.public_symbols['python'])
491 needed = {}
492 for sym in t.unsatisfied_symbols:
493 if sym in bld.env.symbol_map:
494 dep = bld.env.symbol_map[sym][0]
495 if dep == 'c':
496 continue
497 if not dep in needed:
498 needed[dep] = set()
499 needed[dep].add(sym)
500 else:
501 remaining.add(sym)
503 for dep in needed:
504 Logs.info("Target '%s' should add syslib dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
506 if remaining:
507 debug("deps: Target '%s' has unsatisfied symbols: %s" % (sname, " ".join(remaining)))
511 def symbols_symbolcheck(task):
512 '''check the internal dependency lists'''
513 bld = task.env.bld
514 tgt_list = get_tgt_list(bld)
516 build_symbol_sets(bld, tgt_list)
517 build_library_names(bld, tgt_list)
519 for t in tgt_list:
520 t.autodeps = set()
521 if getattr(t, 'source', ''):
522 build_autodeps(bld, t)
524 for t in tgt_list:
525 check_dependencies(bld, t)
527 for t in tgt_list:
528 check_library_deps(bld, t)
530 def symbols_syslibcheck(task):
531 '''check the syslib dependencies'''
532 bld = task.env.bld
533 tgt_list = get_tgt_list(bld)
535 build_syslib_sets(bld, tgt_list)
536 check_syslib_collisions(bld, tgt_list)
538 for t in tgt_list:
539 check_syslib_dependencies(bld, t)
542 def symbols_whyneeded(task):
543 """check why 'target' needs to link to 'subsystem'"""
544 bld = task.env.bld
545 tgt_list = get_tgt_list(bld)
547 why = Options.options.WHYNEEDED.split(":")
548 if len(why) != 2:
549 raise Errors.WafError("usage: WHYNEEDED=TARGET:DEPENDENCY")
550 target = why[0]
551 subsystem = why[1]
553 build_symbol_sets(bld, tgt_list)
554 build_library_names(bld, tgt_list)
555 build_syslib_sets(bld, tgt_list)
557 Logs.info("Checking why %s needs to link to %s" % (target, subsystem))
558 if not target in bld.env.used_symbols:
559 Logs.warn("unable to find target '%s' in used_symbols dict" % target)
560 return
561 if not subsystem in bld.env.public_symbols:
562 Logs.warn("unable to find subsystem '%s' in public_symbols dict" % subsystem)
563 return
564 overlap = bld.env.used_symbols[target].intersection(bld.env.public_symbols[subsystem])
565 if not overlap:
566 Logs.info("target '%s' doesn't use any public symbols from '%s'" % (target, subsystem))
567 else:
568 Logs.info("target '%s' uses symbols %s from '%s'" % (target, overlap, subsystem))
571 def report_duplicate(bld, binname, sym, libs, fail_on_error):
572 '''report duplicated symbols'''
573 if sym in ['_init', '_fini', '_edata', '_end', '__bss_start']:
574 return
575 libnames = []
576 for lib in libs:
577 if lib in bld.env.library_dict:
578 libnames.append(bld.env.library_dict[lib])
579 else:
580 libnames.append(lib)
581 if fail_on_error:
582 raise Errors.WafError("%s: Symbol %s linked in multiple libraries %s" % (binname, sym, libnames))
583 else:
584 print("%s: Symbol %s linked in multiple libraries %s" % (binname, sym, libnames))
587 def symbols_dupcheck_binary(bld, binname, fail_on_error):
588 '''check for duplicated symbols in one binary'''
590 libs = get_libs_recursive(bld, binname, set())
591 symlist = symbols_extract(bld, libs, dynamic=True)
593 symmap = {}
594 for libpath in symlist:
595 for sym in symlist[libpath]['PUBLIC']:
596 if sym == '_GLOBAL_OFFSET_TABLE_':
597 continue
598 if not sym in symmap:
599 symmap[sym] = set()
600 symmap[sym].add(libpath)
601 for sym in symmap:
602 if len(symmap[sym]) > 1:
603 for libpath in symmap[sym]:
604 if libpath in bld.env.library_dict:
605 report_duplicate(bld, binname, sym, symmap[sym], fail_on_error)
606 break
608 def symbols_dupcheck(task, fail_on_error=False):
609 '''check for symbols defined in two different subsystems'''
610 bld = task.env.bld
611 tgt_list = get_tgt_list(bld)
613 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
615 build_library_dict(bld, tgt_list)
616 for t in tgt_list:
617 if t.samba_type == 'BINARY':
618 binname = os_path_relpath(t.link_task.outputs[0].abspath(bld.env), os.getcwd())
619 symbols_dupcheck_binary(bld, binname, fail_on_error)
622 def symbols_dupcheck_fatal(task):
623 '''check for symbols defined in two different subsystems (and fail if duplicates are found)'''
624 symbols_dupcheck(task, fail_on_error=True)
627 def SYMBOL_CHECK(bld):
628 '''check our dependency lists'''
629 if Options.options.SYMBOLCHECK:
630 bld.SET_BUILD_GROUP('symbolcheck')
631 task = bld(rule=symbols_symbolcheck, always=True, name='symbol checking')
632 task.env.bld = bld
634 bld.SET_BUILD_GROUP('syslibcheck')
635 task = bld(rule=symbols_syslibcheck, always=True, name='syslib checking')
636 task.env.bld = bld
638 bld.SET_BUILD_GROUP('syslibcheck')
639 task = bld(rule=symbols_dupcheck, always=True, name='symbol duplicate checking')
640 task.env.bld = bld
642 if Options.options.WHYNEEDED:
643 bld.SET_BUILD_GROUP('syslibcheck')
644 task = bld(rule=symbols_whyneeded, always=True, name='check why a dependency is needed')
645 task.env.bld = bld
648 Build.BuildContext.SYMBOL_CHECK = SYMBOL_CHECK
650 def DUP_SYMBOL_CHECK(bld):
651 if Options.options.DUP_SYMBOLCHECK and bld.env.DEVELOPER:
652 '''check for duplicate symbols'''
653 bld.SET_BUILD_GROUP('syslibcheck')
654 task = bld(rule=symbols_dupcheck_fatal, always=True, name='symbol duplicate checking')
655 task.env.bld = bld
657 Build.BuildContext.DUP_SYMBOL_CHECK = DUP_SYMBOL_CHECK