xref: /freebsd/lib/libproc/proc_sym.c (revision e72055b7feba695a760d45f01f0f8268b1cb4a74)
1 /*-
2  * Copyright (c) 2010 The FreeBSD Foundation
3  * Copyright (c) 2008 John Birrell (jb@freebsd.org)
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Rui Paulo under sponsorship
7  * from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  * $FreeBSD$
31  */
32 
33 #include <sys/types.h>
34 #include <sys/user.h>
35 
36 #include <assert.h>
37 #include <err.h>
38 #include <stdio.h>
39 #include <libgen.h>
40 #include <string.h>
41 #include <stdlib.h>
42 #include <fcntl.h>
43 #include <string.h>
44 #include <unistd.h>
45 #include <libutil.h>
46 
47 #include "_libproc.h"
48 
49 #ifndef NO_CXA_DEMANGLE
50 extern char *__cxa_demangle(const char *, char *, size_t *, int *);
51 #endif /* NO_CXA_DEMANGLE */
52 
53 static void	proc_rdl2prmap(rd_loadobj_t *, prmap_t *);
54 
55 static void
56 demangle(const char *symbol, char *buf, size_t len)
57 {
58 #ifndef NO_CXA_DEMANGLE
59 	char *dembuf;
60 
61 	if (symbol[0] == '_' && symbol[1] == 'Z' && symbol[2]) {
62 		dembuf = __cxa_demangle(symbol, NULL, NULL, NULL);
63 		if (!dembuf)
64 			goto fail;
65 		strlcpy(buf, dembuf, len);
66 		free(dembuf);
67 		return;
68 	}
69 fail:
70 #endif /* NO_CXA_DEMANGLE */
71 	strlcpy(buf, symbol, len);
72 }
73 
74 static void
75 proc_rdl2prmap(rd_loadobj_t *rdl, prmap_t *map)
76 {
77 	map->pr_vaddr = rdl->rdl_saddr;
78 	map->pr_size = rdl->rdl_eaddr - rdl->rdl_saddr;
79 	map->pr_offset = rdl->rdl_offset;
80 	map->pr_mflags = 0;
81 	if (rdl->rdl_prot & RD_RDL_R)
82 		map->pr_mflags |= MA_READ;
83 	if (rdl->rdl_prot & RD_RDL_W)
84 		map->pr_mflags |= MA_WRITE;
85 	if (rdl->rdl_prot & RD_RDL_X)
86 		map->pr_mflags |= MA_EXEC;
87 	strlcpy(map->pr_mapname, rdl->rdl_path,
88 	    sizeof(map->pr_mapname));
89 }
90 
91 char *
92 proc_objname(struct proc_handle *p, uintptr_t addr, char *objname,
93     size_t objnamesz)
94 {
95 	size_t i;
96 	rd_loadobj_t *rdl;
97 
98 	for (i = 0; i < p->nobjs; i++) {
99 		rdl = &p->rdobjs[i];
100 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
101 			strlcpy(objname, rdl->rdl_path, objnamesz);
102 			return (objname);
103 		}
104 	}
105 	return (NULL);
106 }
107 
108 prmap_t *
109 proc_obj2map(struct proc_handle *p, const char *objname)
110 {
111 	size_t i;
112 	prmap_t *map;
113 	rd_loadobj_t *rdl;
114 	char path[MAXPATHLEN];
115 
116 	rdl = NULL;
117 	for (i = 0; i < p->nobjs; i++) {
118 		basename_r(p->rdobjs[i].rdl_path, path);
119 		if (strcmp(path, objname) == 0) {
120 			rdl = &p->rdobjs[i];
121 			break;
122 		}
123 	}
124 	if (rdl == NULL) {
125 		if (strcmp(objname, "a.out") == 0 && p->rdexec != NULL)
126 			rdl = p->rdexec;
127 		else
128 			return (NULL);
129 	}
130 
131 	if ((map = malloc(sizeof(*map))) == NULL)
132 		return (NULL);
133 	proc_rdl2prmap(rdl, map);
134 	return (map);
135 }
136 
137 int
138 proc_iter_objs(struct proc_handle *p, proc_map_f *func, void *cd)
139 {
140 	size_t i;
141 	rd_loadobj_t *rdl;
142 	prmap_t map;
143 	char path[MAXPATHLEN];
144 	char last[MAXPATHLEN];
145 
146 	if (p->nobjs == 0)
147 		return (-1);
148 	memset(last, 0, sizeof(last));
149 	for (i = 0; i < p->nobjs; i++) {
150 		rdl = &p->rdobjs[i];
151 		proc_rdl2prmap(rdl, &map);
152 		basename_r(rdl->rdl_path, path);
153 		/*
154 		 * We shouldn't call the callback twice with the same object.
155 		 * To do that we are assuming the fact that if there are
156 		 * repeated object names (i.e. different mappings for the
157 		 * same object) they occur next to each other.
158 		 */
159 		if (strcmp(path, last) == 0)
160 			continue;
161 		(*func)(cd, &map, path);
162 		strlcpy(last, path, sizeof(last));
163 	}
164 
165 	return (0);
166 }
167 
168 prmap_t *
169 proc_addr2map(struct proc_handle *p, uintptr_t addr)
170 {
171 	size_t i;
172 	int cnt, lastvn = 0;
173 	prmap_t *map;
174 	rd_loadobj_t *rdl;
175 	struct kinfo_vmentry *kves, *kve;
176 
177 	/*
178 	 * If we don't have a cache of listed objects, we need to query
179 	 * it ourselves.
180 	 */
181 	if (p->nobjs == 0) {
182 		if ((kves = kinfo_getvmmap(p->pid, &cnt)) == NULL)
183 			return (NULL);
184 		for (i = 0; i < (size_t)cnt; i++) {
185 			kve = kves + i;
186 			if (kve->kve_type == KVME_TYPE_VNODE)
187 				lastvn = i;
188 			if (addr >= kve->kve_start && addr < kve->kve_end) {
189 				if ((map = malloc(sizeof(*map))) == NULL) {
190 					free(kves);
191 					return (NULL);
192 				}
193 				map->pr_vaddr = kve->kve_start;
194 				map->pr_size = kve->kve_end - kve->kve_start;
195 				map->pr_offset = kve->kve_offset;
196 				map->pr_mflags = 0;
197 				if (kve->kve_protection & KVME_PROT_READ)
198 					map->pr_mflags |= MA_READ;
199 				if (kve->kve_protection & KVME_PROT_WRITE)
200 					map->pr_mflags |= MA_WRITE;
201 				if (kve->kve_protection & KVME_PROT_EXEC)
202 					map->pr_mflags |= MA_EXEC;
203 				if (kve->kve_flags & KVME_FLAG_COW)
204 					map->pr_mflags |= MA_COW;
205 				if (kve->kve_flags & KVME_FLAG_NEEDS_COPY)
206 					map->pr_mflags |= MA_NEEDS_COPY;
207 				if (kve->kve_flags & KVME_FLAG_NOCOREDUMP)
208 					map->pr_mflags |= MA_NOCOREDUMP;
209 				strlcpy(map->pr_mapname, kves[lastvn].kve_path,
210 				    sizeof(map->pr_mapname));
211 				free(kves);
212 				return (map);
213 			}
214 		}
215 		free(kves);
216 		return (NULL);
217 	}
218 
219 	for (i = 0; i < p->nobjs; i++) {
220 		rdl = &p->rdobjs[i];
221 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
222 			if ((map = malloc(sizeof(*map))) == NULL)
223 				return (NULL);
224 			proc_rdl2prmap(rdl, map);
225 			return (map);
226 		}
227 	}
228 	return (NULL);
229 }
230 
231 int
232 proc_addr2sym(struct proc_handle *p, uintptr_t addr, char *name,
233     size_t namesz, GElf_Sym *symcopy)
234 {
235 	Elf *e;
236 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
237 	Elf_Data *data;
238 	GElf_Shdr shdr;
239 	GElf_Sym sym;
240 	GElf_Ehdr ehdr;
241 	int fd, error = -1;
242 	size_t i;
243 	uint64_t rsym;
244 	prmap_t *map;
245 	char *s;
246 	unsigned long symtabstridx = 0, dynsymstridx = 0;
247 
248 	if ((map = proc_addr2map(p, addr)) == NULL)
249 		return (-1);
250 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
251 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
252 		goto err0;
253 	}
254 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
255 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
256 		goto err1;
257 	}
258 	if (gelf_getehdr(e, &ehdr) == NULL) {
259 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
260 		goto err2;
261 	}
262 	/*
263 	 * Find the index of the STRTAB and SYMTAB sections to locate
264 	 * symbol names.
265 	 */
266 	scn = NULL;
267 	while ((scn = elf_nextscn(e, scn)) != NULL) {
268 		gelf_getshdr(scn, &shdr);
269 		switch (shdr.sh_type) {
270 		case SHT_SYMTAB:
271 			symtabscn = scn;
272 			symtabstridx = shdr.sh_link;
273 			break;
274 		case SHT_DYNSYM:
275 			dynsymscn = scn;
276 			dynsymstridx = shdr.sh_link;
277 			break;
278 		default:
279 			break;
280 		}
281 	}
282 	/*
283 	 * Iterate over the Dynamic Symbols table to find the symbol.
284 	 * Then look up the string name in STRTAB (.dynstr)
285 	 */
286 	if ((data = elf_getdata(dynsymscn, NULL)) == NULL) {
287 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
288 		goto symtab;
289 	}
290 	i = 0;
291 	while (gelf_getsym(data, i++, &sym) != NULL) {
292 		/*
293 		 * Calculate the address mapped to the virtual memory
294 		 * by rtld.
295 		 */
296 		if (ehdr.e_type != ET_EXEC)
297 			rsym = map->pr_vaddr + sym.st_value;
298 		else
299 			rsym = sym.st_value;
300 		if (addr >= rsym && addr < rsym + sym.st_size) {
301 			s = elf_strptr(e, dynsymstridx, sym.st_name);
302 			if (s) {
303 				demangle(s, name, namesz);
304 				memcpy(symcopy, &sym, sizeof(sym));
305 				/*
306 				 * DTrace expects the st_value to contain
307 				 * only the address relative to the start of
308 				 * the function.
309 				 */
310 				symcopy->st_value = rsym;
311 				error = 0;
312 				goto out;
313 			}
314 		}
315 	}
316 symtab:
317 	/*
318 	 * Iterate over the Symbols Table to find the symbol.
319 	 * Then look up the string name in STRTAB (.dynstr)
320 	 */
321 	if ((data = elf_getdata(symtabscn, NULL)) == NULL) {
322 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
323 		goto err2;
324 	}
325 	i = 0;
326 	while (gelf_getsym(data, i++, &sym) != NULL) {
327 		/*
328 		 * Calculate the address mapped to the virtual memory
329 		 * by rtld.
330 		 */
331 		if (ehdr.e_type != ET_EXEC)
332 			rsym = map->pr_vaddr + sym.st_value;
333 		else
334 			rsym = sym.st_value;
335 		if (addr >= rsym && addr < rsym + sym.st_size) {
336 			s = elf_strptr(e, symtabstridx, sym.st_name);
337 			if (s) {
338 				demangle(s, name, namesz);
339 				memcpy(symcopy, &sym, sizeof(sym));
340 				/*
341 				 * DTrace expects the st_value to contain
342 				 * only the address relative to the start of
343 				 * the function.
344 				 */
345 				symcopy->st_value = rsym;
346 				error = 0;
347 				goto out;
348 			}
349 		}
350 	}
351 out:
352 err2:
353 	elf_end(e);
354 err1:
355 	close(fd);
356 err0:
357 	free(map);
358 	return (error);
359 }
360 
361 prmap_t *
362 proc_name2map(struct proc_handle *p, const char *name)
363 {
364 	size_t i;
365 	int cnt;
366 	prmap_t *map;
367 	char tmppath[MAXPATHLEN];
368 	struct kinfo_vmentry *kves, *kve;
369 	rd_loadobj_t *rdl;
370 
371 	/*
372 	 * If we haven't iterated over the list of loaded objects,
373 	 * librtld_db isn't yet initialized and it's very likely
374 	 * that librtld_db called us. We need to do the heavy
375 	 * lifting here to find the symbol librtld_db is looking for.
376 	 */
377 	if (p->nobjs == 0) {
378 		if ((kves = kinfo_getvmmap(proc_getpid(p), &cnt)) == NULL)
379 			return (NULL);
380 		for (i = 0; i < (size_t)cnt; i++) {
381 			kve = kves + i;
382 			basename_r(kve->kve_path, tmppath);
383 			if (strcmp(tmppath, name) == 0) {
384 				map = proc_addr2map(p, kve->kve_start);
385 				free(kves);
386 				return (map);
387 			}
388 		}
389 		free(kves);
390 		return (NULL);
391 	}
392 	if ((name == NULL || strcmp(name, "a.out") == 0) &&
393 	    p->rdexec != NULL) {
394 		map = proc_addr2map(p, p->rdexec->rdl_saddr);
395 		return (map);
396 	}
397 	for (i = 0; i < p->nobjs; i++) {
398 		rdl = &p->rdobjs[i];
399 		basename_r(rdl->rdl_path, tmppath);
400 		if (strcmp(tmppath, name) == 0) {
401 			if ((map = malloc(sizeof(*map))) == NULL)
402 				return (NULL);
403 			proc_rdl2prmap(rdl, map);
404 			return (map);
405 		}
406 	}
407 
408 	return (NULL);
409 }
410 
411 int
412 proc_name2sym(struct proc_handle *p, const char *object, const char *symbol,
413     GElf_Sym *symcopy)
414 {
415 	Elf *e;
416 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
417 	Elf_Data *data;
418 	GElf_Shdr shdr;
419 	GElf_Sym sym;
420 	GElf_Ehdr ehdr;
421 	int fd, error = -1;
422 	size_t i;
423 	prmap_t *map;
424 	char *s;
425 	unsigned long symtabstridx = 0, dynsymstridx = 0;
426 
427 	if ((map = proc_name2map(p, object)) == NULL) {
428 		DPRINTFX("ERROR: couldn't find object %s", object);
429 		goto err0;
430 	}
431 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
432 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
433 		goto err0;
434 	}
435 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
436 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
437 		goto err1;
438 	}
439 	if (gelf_getehdr(e, &ehdr) == NULL) {
440 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
441 		goto err2;
442 	}
443 	/*
444 	 * Find the index of the STRTAB and SYMTAB sections to locate
445 	 * symbol names.
446 	 */
447 	scn = NULL;
448 	while ((scn = elf_nextscn(e, scn)) != NULL) {
449 		gelf_getshdr(scn, &shdr);
450 		switch (shdr.sh_type) {
451 		case SHT_SYMTAB:
452 			symtabscn = scn;
453 			symtabstridx = shdr.sh_link;
454 			break;
455 		case SHT_DYNSYM:
456 			dynsymscn = scn;
457 			dynsymstridx = shdr.sh_link;
458 			break;
459 		default:
460 			break;
461 		}
462 	}
463 	/*
464 	 * Iterate over the Dynamic Symbols table to find the symbol.
465 	 * Then look up the string name in STRTAB (.dynstr)
466 	 */
467 	if ((data = elf_getdata(dynsymscn, NULL))) {
468 		i = 0;
469 		while (gelf_getsym(data, i++, &sym) != NULL) {
470 			s = elf_strptr(e, dynsymstridx, sym.st_name);
471 			if (s && strcmp(s, symbol) == 0) {
472 				memcpy(symcopy, &sym, sizeof(sym));
473 				if (ehdr.e_type != ET_EXEC)
474 					symcopy->st_value += map->pr_vaddr;
475 				error = 0;
476 				goto out;
477 			}
478 		}
479 	}
480 	/*
481 	 * Iterate over the Symbols Table to find the symbol.
482 	 * Then look up the string name in STRTAB (.dynstr)
483 	 */
484 	if ((data = elf_getdata(symtabscn, NULL))) {
485 		i = 0;
486 		while (gelf_getsym(data, i++, &sym) != NULL) {
487 			s = elf_strptr(e, symtabstridx, sym.st_name);
488 			if (s && strcmp(s, symbol) == 0) {
489 				memcpy(symcopy, &sym, sizeof(sym));
490 				if (ehdr.e_type != ET_EXEC)
491 					symcopy->st_value += map->pr_vaddr;
492 				error = 0;
493 				goto out;
494 			}
495 		}
496 	}
497 out:
498 	DPRINTFX("found addr 0x%lx for %s", symcopy->st_value, symbol);
499 err2:
500 	elf_end(e);
501 err1:
502 	close(fd);
503 err0:
504 	free(map);
505 
506 	return (error);
507 }
508 
509 
510 int
511 proc_iter_symbyaddr(struct proc_handle *p, const char *object, int which,
512     int mask, proc_sym_f *func, void *cd)
513 {
514 	Elf *e;
515 	int i, fd;
516 	prmap_t *map;
517 	Elf_Scn *scn, *foundscn = NULL;
518 	Elf_Data *data;
519 	GElf_Ehdr ehdr;
520 	GElf_Shdr shdr;
521 	GElf_Sym sym;
522 	unsigned long stridx = -1;
523 	char *s;
524 	int error = -1;
525 
526 	if ((map = proc_name2map(p, object)) == NULL)
527 		return (-1);
528 	if ((fd = open(map->pr_mapname, O_RDONLY)) < 0) {
529 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
530 		goto err0;
531 	}
532 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
533 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
534 		goto err1;
535 	}
536 	if (gelf_getehdr(e, &ehdr) == NULL) {
537 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
538 		goto err2;
539 	}
540 	/*
541 	 * Find the section we are looking for.
542 	 */
543 	scn = NULL;
544 	while ((scn = elf_nextscn(e, scn)) != NULL) {
545 		gelf_getshdr(scn, &shdr);
546 		if (which == PR_SYMTAB &&
547 		    shdr.sh_type == SHT_SYMTAB) {
548 			foundscn = scn;
549 			break;
550 		} else if (which == PR_DYNSYM &&
551 		    shdr.sh_type == SHT_DYNSYM) {
552 			foundscn = scn;
553 			break;
554 		}
555 	}
556 	if (!foundscn)
557 		return (-1);
558 	stridx = shdr.sh_link;
559 	if ((data = elf_getdata(foundscn, NULL)) == NULL) {
560 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
561 		goto err2;
562 	}
563 	i = 0;
564 	while (gelf_getsym(data, i++, &sym) != NULL) {
565 		if (GELF_ST_BIND(sym.st_info) == STB_LOCAL &&
566 		    (mask & BIND_LOCAL) == 0)
567 			continue;
568 		if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL &&
569 		    (mask & BIND_GLOBAL) == 0)
570 			continue;
571 		if (GELF_ST_BIND(sym.st_info) == STB_WEAK &&
572 		    (mask & BIND_WEAK) == 0)
573 			continue;
574 		if (GELF_ST_TYPE(sym.st_info) == STT_NOTYPE &&
575 		    (mask & TYPE_NOTYPE) == 0)
576 			continue;
577 		if (GELF_ST_TYPE(sym.st_info) == STT_OBJECT &&
578 		    (mask & TYPE_OBJECT) == 0)
579 			continue;
580 		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC &&
581 		    (mask & TYPE_FUNC) == 0)
582 			continue;
583 		if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
584 		    (mask & TYPE_SECTION) == 0)
585 			continue;
586 		if (GELF_ST_TYPE(sym.st_info) == STT_FILE &&
587 		    (mask & TYPE_FILE) == 0)
588 			continue;
589 		s = elf_strptr(e, stridx, sym.st_name);
590 		if (ehdr.e_type != ET_EXEC)
591 			sym.st_value += map->pr_vaddr;
592 		(*func)(cd, &sym, s);
593 	}
594 	error = 0;
595 err2:
596 	elf_end(e);
597 err1:
598 	close(fd);
599 err0:
600 	free(map);
601 	return (error);
602 }
603