xref: /freebsd/lib/libproc/proc_sym.c (revision ee7b0571c2c18bdec848ed2044223cc88db29bd8)
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 	size_t demlen;
61 
62 	if (symbol[0] == '_' && symbol[1] == 'Z' && symbol[2]) {
63 		dembuf = malloc(len);
64 		if (!dembuf)
65 			goto fail;
66 		demlen = len;
67 		dembuf = __cxa_demangle(symbol, dembuf, &demlen, NULL);
68 		if (!dembuf)
69 			goto fail;
70 		strlcpy(buf, dembuf, len);
71 		free(dembuf);
72 	}
73 
74 	return;
75 fail:
76 #endif /* NO_CXA_DEMANGLE */
77 	strlcpy(buf, symbol, len);
78 }
79 
80 static void
81 proc_rdl2prmap(rd_loadobj_t *rdl, prmap_t *map)
82 {
83 	map->pr_vaddr = rdl->rdl_saddr;
84 	map->pr_size = rdl->rdl_eaddr - rdl->rdl_saddr;
85 	map->pr_offset = rdl->rdl_offset;
86 	map->pr_mflags = 0;
87 	if (rdl->rdl_prot & RD_RDL_R)
88 		map->pr_mflags |= MA_READ;
89 	if (rdl->rdl_prot & RD_RDL_W)
90 		map->pr_mflags |= MA_WRITE;
91 	if (rdl->rdl_prot & RD_RDL_X)
92 		map->pr_mflags |= MA_EXEC;
93 	strlcpy(map->pr_mapname, rdl->rdl_path,
94 	    sizeof(map->pr_mapname));
95 }
96 
97 char *
98 proc_objname(struct proc_handle *p, uintptr_t addr, char *objname,
99     size_t objnamesz)
100 {
101 	size_t i;
102 	rd_loadobj_t *rdl;
103 
104 	for (i = 0; i < p->nobjs; i++) {
105 		rdl = &p->rdobjs[i];
106 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
107 			strlcpy(objname, rdl->rdl_path, objnamesz);
108 			return (objname);
109 		}
110 	}
111 	return (NULL);
112 }
113 
114 prmap_t *
115 proc_obj2map(struct proc_handle *p, const char *objname)
116 {
117 	size_t i;
118 	prmap_t *map;
119 	rd_loadobj_t *rdl;
120 	char path[MAXPATHLEN];
121 
122 	rdl = NULL;
123 	for (i = 0; i < p->nobjs; i++) {
124 		basename_r(p->rdobjs[i].rdl_path, path);
125 		if (strcmp(path, objname) == 0) {
126 			rdl = &p->rdobjs[i];
127 			break;
128 		}
129 	}
130 	if (rdl == NULL && strcmp(objname, "a.out") == 0 && p->rdexec != NULL)
131 		rdl = p->rdexec;
132 	else
133 		return (NULL);
134 
135 	if ((map = malloc(sizeof(*map))) == NULL)
136 		return (NULL);
137 	proc_rdl2prmap(rdl, map);
138 	return (map);
139 }
140 
141 int
142 proc_iter_objs(struct proc_handle *p, proc_map_f *func, void *cd)
143 {
144 	size_t i;
145 	rd_loadobj_t *rdl;
146 	prmap_t map;
147 	char path[MAXPATHLEN];
148 	char last[MAXPATHLEN];
149 
150 	if (p->nobjs == 0)
151 		return (-1);
152 	memset(last, 0, sizeof(last));
153 	for (i = 0; i < p->nobjs; i++) {
154 		rdl = &p->rdobjs[i];
155 		proc_rdl2prmap(rdl, &map);
156 		basename_r(rdl->rdl_path, path);
157 		/*
158 		 * We shouldn't call the callback twice with the same object.
159 		 * To do that we are assuming the fact that if there are
160 		 * repeated object names (i.e. different mappings for the
161 		 * same object) they occur next to each other.
162 		 */
163 		if (strcmp(path, last) == 0)
164 			continue;
165 		(*func)(cd, &map, path);
166 		strlcpy(last, path, sizeof(last));
167 	}
168 
169 	return (0);
170 }
171 
172 prmap_t *
173 proc_addr2map(struct proc_handle *p, uintptr_t addr)
174 {
175 	size_t i;
176 	int cnt, lastvn = 0;
177 	prmap_t *map;
178 	rd_loadobj_t *rdl;
179 	struct kinfo_vmentry *kves, *kve;
180 
181 	/*
182 	 * If we don't have a cache of listed objects, we need to query
183 	 * it ourselves.
184 	 */
185 	if (p->nobjs == 0) {
186 		if ((kves = kinfo_getvmmap(p->pid, &cnt)) == NULL)
187 			return (NULL);
188 		for (i = 0; i < (size_t)cnt; i++) {
189 			kve = kves + i;
190 			if (kve->kve_type == KVME_TYPE_VNODE)
191 				lastvn = i;
192 			if (addr >= kve->kve_start && addr < kve->kve_end) {
193 				if ((map = malloc(sizeof(*map))) == NULL) {
194 					free(kves);
195 					return (NULL);
196 				}
197 				map->pr_vaddr = kve->kve_start;
198 				map->pr_size = kve->kve_end - kve->kve_start;
199 				map->pr_offset = kve->kve_offset;
200 				map->pr_mflags = 0;
201 				if (kve->kve_protection & KVME_PROT_READ)
202 					map->pr_mflags |= MA_READ;
203 				if (kve->kve_protection & KVME_PROT_WRITE)
204 					map->pr_mflags |= MA_WRITE;
205 				if (kve->kve_protection & KVME_PROT_EXEC)
206 					map->pr_mflags |= MA_EXEC;
207 				if (kve->kve_flags & KVME_FLAG_COW)
208 					map->pr_mflags |= MA_COW;
209 				if (kve->kve_flags & KVME_FLAG_NEEDS_COPY)
210 					map->pr_mflags |= MA_NEEDS_COPY;
211 				if (kve->kve_flags & KVME_FLAG_NOCOREDUMP)
212 					map->pr_mflags |= MA_NOCOREDUMP;
213 				strlcpy(map->pr_mapname, kves[lastvn].kve_path,
214 				    sizeof(map->pr_mapname));
215 				free(kves);
216 				return (map);
217 			}
218 		}
219 		free(kves);
220 		return (NULL);
221 	}
222 
223 	for (i = 0; i < p->nobjs; i++) {
224 		rdl = &p->rdobjs[i];
225 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
226 			if ((map = malloc(sizeof(*map))) == NULL)
227 				return (NULL);
228 			proc_rdl2prmap(rdl, map);
229 			return (map);
230 		}
231 	}
232 	return (NULL);
233 }
234 
235 int
236 proc_addr2sym(struct proc_handle *p, uintptr_t addr, char *name,
237     size_t namesz, GElf_Sym *symcopy)
238 {
239 	Elf *e;
240 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
241 	Elf_Data *data;
242 	GElf_Shdr shdr;
243 	GElf_Sym sym;
244 	GElf_Ehdr ehdr;
245 	int fd, error = -1;
246 	size_t i;
247 	uint64_t rsym;
248 	prmap_t *map;
249 	char *s;
250 	unsigned long symtabstridx = 0, dynsymstridx = 0;
251 
252 	if ((map = proc_addr2map(p, addr)) == NULL)
253 		return (-1);
254 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
255 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
256 		goto err0;
257 	}
258 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
259 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
260 		goto err1;
261 	}
262 	if (gelf_getehdr(e, &ehdr) == NULL) {
263 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
264 		goto err2;
265 	}
266 	/*
267 	 * Find the index of the STRTAB and SYMTAB sections to locate
268 	 * symbol names.
269 	 */
270 	scn = NULL;
271 	while ((scn = elf_nextscn(e, scn)) != NULL) {
272 		gelf_getshdr(scn, &shdr);
273 		switch (shdr.sh_type) {
274 		case SHT_SYMTAB:
275 			symtabscn = scn;
276 			symtabstridx = shdr.sh_link;
277 			break;
278 		case SHT_DYNSYM:
279 			dynsymscn = scn;
280 			dynsymstridx = shdr.sh_link;
281 			break;
282 		default:
283 			break;
284 		}
285 	}
286 	/*
287 	 * Iterate over the Dynamic Symbols table to find the symbol.
288 	 * Then look up the string name in STRTAB (.dynstr)
289 	 */
290 	if ((data = elf_getdata(dynsymscn, NULL)) == NULL) {
291 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
292 		goto symtab;
293 	}
294 	i = 0;
295 	while (gelf_getsym(data, i++, &sym) != NULL) {
296 		/*
297 		 * Calculate the address mapped to the virtual memory
298 		 * by rtld.
299 		 */
300 		if (ehdr.e_type != ET_EXEC)
301 			rsym = map->pr_vaddr + sym.st_value;
302 		else
303 			rsym = sym.st_value;
304 		if (addr >= rsym && addr < rsym + sym.st_size) {
305 			s = elf_strptr(e, dynsymstridx, sym.st_name);
306 			if (s) {
307 				demangle(s, name, namesz);
308 				memcpy(symcopy, &sym, sizeof(sym));
309 				/*
310 				 * DTrace expects the st_value to contain
311 				 * only the address relative to the start of
312 				 * the function.
313 				 */
314 				symcopy->st_value = rsym;
315 				error = 0;
316 				goto out;
317 			}
318 		}
319 	}
320 symtab:
321 	/*
322 	 * Iterate over the Symbols Table to find the symbol.
323 	 * Then look up the string name in STRTAB (.dynstr)
324 	 */
325 	if ((data = elf_getdata(symtabscn, NULL)) == NULL) {
326 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
327 		goto err2;
328 	}
329 	i = 0;
330 	while (gelf_getsym(data, i++, &sym) != NULL) {
331 		/*
332 		 * Calculate the address mapped to the virtual memory
333 		 * by rtld.
334 		 */
335 		if (ehdr.e_type != ET_EXEC)
336 			rsym = map->pr_vaddr + sym.st_value;
337 		else
338 			rsym = sym.st_value;
339 		if (addr >= rsym && addr < rsym + sym.st_size) {
340 			s = elf_strptr(e, symtabstridx, sym.st_name);
341 			if (s) {
342 				demangle(s, name, namesz);
343 				memcpy(symcopy, &sym, sizeof(sym));
344 				/*
345 				 * DTrace expects the st_value to contain
346 				 * only the address relative to the start of
347 				 * the function.
348 				 */
349 				symcopy->st_value = rsym;
350 				error = 0;
351 				goto out;
352 			}
353 		}
354 	}
355 out:
356 err2:
357 	elf_end(e);
358 err1:
359 	close(fd);
360 err0:
361 	free(map);
362 	return (error);
363 }
364 
365 prmap_t *
366 proc_name2map(struct proc_handle *p, const char *name)
367 {
368 	size_t i;
369 	int cnt;
370 	prmap_t *map;
371 	char tmppath[MAXPATHLEN];
372 	struct kinfo_vmentry *kves, *kve;
373 	rd_loadobj_t *rdl;
374 
375 	/*
376 	 * If we haven't iterated over the list of loaded objects,
377 	 * librtld_db isn't yet initialized and it's very likely
378 	 * that librtld_db called us. We need to do the heavy
379 	 * lifting here to find the symbol librtld_db is looking for.
380 	 */
381 	if (p->nobjs == 0) {
382 		if ((kves = kinfo_getvmmap(proc_getpid(p), &cnt)) == NULL)
383 			return (NULL);
384 		for (i = 0; i < (size_t)cnt; i++) {
385 			kve = kves + i;
386 			basename_r(kve->kve_path, tmppath);
387 			if (strcmp(tmppath, name) == 0) {
388 				map = proc_addr2map(p, kve->kve_start);
389 				free(kves);
390 				return (map);
391 			}
392 		}
393 		free(kves);
394 		return (NULL);
395 	}
396 	if ((name == NULL || strcmp(name, "a.out") == 0) &&
397 	    p->rdexec != NULL) {
398 		map = proc_addr2map(p, p->rdexec->rdl_saddr);
399 		return (map);
400 	}
401 	for (i = 0; i < p->nobjs; i++) {
402 		rdl = &p->rdobjs[i];
403 		basename_r(rdl->rdl_path, tmppath);
404 		if (strcmp(tmppath, name) == 0) {
405 			if ((map = malloc(sizeof(*map))) == NULL)
406 				return (NULL);
407 			proc_rdl2prmap(rdl, map);
408 			return (map);
409 		}
410 	}
411 
412 	return (NULL);
413 }
414 
415 int
416 proc_name2sym(struct proc_handle *p, const char *object, const char *symbol,
417     GElf_Sym *symcopy)
418 {
419 	Elf *e;
420 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
421 	Elf_Data *data;
422 	GElf_Shdr shdr;
423 	GElf_Sym sym;
424 	GElf_Ehdr ehdr;
425 	int fd, error = -1;
426 	size_t i;
427 	prmap_t *map;
428 	char *s;
429 	unsigned long symtabstridx = 0, dynsymstridx = 0;
430 
431 	if ((map = proc_name2map(p, object)) == NULL) {
432 		DPRINTFX("ERROR: couldn't find object %s", object);
433 		goto err0;
434 	}
435 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
436 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
437 		goto err0;
438 	}
439 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
440 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
441 		goto err1;
442 	}
443 	if (gelf_getehdr(e, &ehdr) == NULL) {
444 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
445 		goto err2;
446 	}
447 	/*
448 	 * Find the index of the STRTAB and SYMTAB sections to locate
449 	 * symbol names.
450 	 */
451 	scn = NULL;
452 	while ((scn = elf_nextscn(e, scn)) != NULL) {
453 		gelf_getshdr(scn, &shdr);
454 		switch (shdr.sh_type) {
455 		case SHT_SYMTAB:
456 			symtabscn = scn;
457 			symtabstridx = shdr.sh_link;
458 			break;
459 		case SHT_DYNSYM:
460 			dynsymscn = scn;
461 			dynsymstridx = shdr.sh_link;
462 			break;
463 		default:
464 			break;
465 		}
466 	}
467 	/*
468 	 * Iterate over the Dynamic Symbols table to find the symbol.
469 	 * Then look up the string name in STRTAB (.dynstr)
470 	 */
471 	if ((data = elf_getdata(dynsymscn, NULL))) {
472 		i = 0;
473 		while (gelf_getsym(data, i++, &sym) != NULL) {
474 			s = elf_strptr(e, dynsymstridx, sym.st_name);
475 			if (s && strcmp(s, symbol) == 0) {
476 				memcpy(symcopy, &sym, sizeof(sym));
477 				if (ehdr.e_type != ET_EXEC)
478 					symcopy->st_value += map->pr_vaddr;
479 				error = 0;
480 				goto out;
481 			}
482 		}
483 	}
484 	/*
485 	 * Iterate over the Symbols Table to find the symbol.
486 	 * Then look up the string name in STRTAB (.dynstr)
487 	 */
488 	if ((data = elf_getdata(symtabscn, NULL))) {
489 		i = 0;
490 		while (gelf_getsym(data, i++, &sym) != NULL) {
491 			s = elf_strptr(e, symtabstridx, sym.st_name);
492 			if (s && strcmp(s, symbol) == 0) {
493 				memcpy(symcopy, &sym, sizeof(sym));
494 				if (ehdr.e_type != ET_EXEC)
495 					symcopy->st_value += map->pr_vaddr;
496 				error = 0;
497 				goto out;
498 			}
499 		}
500 	}
501 out:
502 	DPRINTFX("found addr 0x%lx for %s", symcopy->st_value, symbol);
503 err2:
504 	elf_end(e);
505 err1:
506 	close(fd);
507 err0:
508 	free(map);
509 
510 	return (error);
511 }
512 
513 
514 int
515 proc_iter_symbyaddr(struct proc_handle *p, const char *object, int which,
516     int mask, proc_sym_f *func, void *cd)
517 {
518 	Elf *e;
519 	int i, fd;
520 	prmap_t *map;
521 	Elf_Scn *scn, *foundscn = NULL;
522 	Elf_Data *data;
523 	GElf_Ehdr ehdr;
524 	GElf_Shdr shdr;
525 	GElf_Sym sym;
526 	unsigned long stridx = -1;
527 	char *s;
528 	int error = -1;
529 
530 	if ((map = proc_name2map(p, object)) == NULL)
531 		return (-1);
532 	if ((fd = open(map->pr_mapname, O_RDONLY)) < 0) {
533 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
534 		goto err0;
535 	}
536 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
537 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
538 		goto err1;
539 	}
540 	if (gelf_getehdr(e, &ehdr) == NULL) {
541 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
542 		goto err2;
543 	}
544 	/*
545 	 * Find the section we are looking for.
546 	 */
547 	scn = NULL;
548 	while ((scn = elf_nextscn(e, scn)) != NULL) {
549 		gelf_getshdr(scn, &shdr);
550 		if (which == PR_SYMTAB &&
551 		    shdr.sh_type == SHT_SYMTAB) {
552 			foundscn = scn;
553 			break;
554 		} else if (which == PR_DYNSYM &&
555 		    shdr.sh_type == SHT_DYNSYM) {
556 			foundscn = scn;
557 			break;
558 		}
559 	}
560 	if (!foundscn)
561 		return (-1);
562 	stridx = shdr.sh_link;
563 	if ((data = elf_getdata(foundscn, NULL)) == NULL) {
564 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
565 		goto err2;
566 	}
567 	i = 0;
568 	while (gelf_getsym(data, i++, &sym) != NULL) {
569 		if (GELF_ST_BIND(sym.st_info) == STB_LOCAL &&
570 		    (mask & BIND_LOCAL) == 0)
571 			continue;
572 		if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL &&
573 		    (mask & BIND_GLOBAL) == 0)
574 			continue;
575 		if (GELF_ST_BIND(sym.st_info) == STB_WEAK &&
576 		    (mask & BIND_WEAK) == 0)
577 			continue;
578 		if (GELF_ST_TYPE(sym.st_info) == STT_NOTYPE &&
579 		    (mask & TYPE_NOTYPE) == 0)
580 			continue;
581 		if (GELF_ST_TYPE(sym.st_info) == STT_OBJECT &&
582 		    (mask & TYPE_OBJECT) == 0)
583 			continue;
584 		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC &&
585 		    (mask & TYPE_FUNC) == 0)
586 			continue;
587 		if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
588 		    (mask & TYPE_SECTION) == 0)
589 			continue;
590 		if (GELF_ST_TYPE(sym.st_info) == STT_FILE &&
591 		    (mask & TYPE_FILE) == 0)
592 			continue;
593 		s = elf_strptr(e, stridx, sym.st_name);
594 		if (ehdr.e_type != ET_EXEC)
595 			sym.st_value += map->pr_vaddr;
596 		(*func)(cd, &sym, s);
597 	}
598 	error = 0;
599 err2:
600 	elf_end(e);
601 err1:
602 	close(fd);
603 err0:
604 	free(map);
605 	return (error);
606 }
607