xref: /freebsd/lib/libproc/proc_sym.c (revision 9ecd54f24fe9fa373e07c9fd7c052deb2188f545)
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 extern char *__cxa_demangle(const char *, char *, size_t *, int *);
50 
51 static void	proc_rdl2prmap(rd_loadobj_t *, prmap_t *);
52 
53 static void
54 demangle(const char *symbol, char *buf, size_t len)
55 {
56 	char *dembuf;
57 	size_t demlen = len;
58 
59 	dembuf = malloc(len);
60 	if (!dembuf)
61 		goto fail;
62 	dembuf = __cxa_demangle(symbol, dembuf, &demlen, NULL);
63 	if (!dembuf)
64 		goto fail;
65 	strlcpy(buf, dembuf, len);
66 	free(dembuf);
67 
68 	return;
69 fail:
70 	strlcpy(buf, symbol, len);
71 }
72 
73 static void
74 proc_rdl2prmap(rd_loadobj_t *rdl, prmap_t *map)
75 {
76 	map->pr_vaddr = rdl->rdl_saddr;
77 	map->pr_size = rdl->rdl_eaddr - rdl->rdl_saddr;
78 	map->pr_offset = rdl->rdl_offset;
79 	map->pr_mflags = 0;
80 	if (rdl->rdl_prot & RD_RDL_R)
81 		map->pr_mflags |= MA_READ;
82 	if (rdl->rdl_prot & RD_RDL_W)
83 		map->pr_mflags |= MA_WRITE;
84 	if (rdl->rdl_prot & RD_RDL_X)
85 		map->pr_mflags |= MA_EXEC;
86 	strlcpy(map->pr_mapname, rdl->rdl_path,
87 	    sizeof(map->pr_mapname));
88 }
89 
90 char *
91 proc_objname(struct proc_handle *p, uintptr_t addr, char *objname,
92     size_t objnamesz)
93 {
94 	size_t i;
95 	rd_loadobj_t *rdl;
96 
97 	for (i = 0; i < p->nobjs; i++) {
98 		rdl = &p->rdobjs[i];
99 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
100 			strlcpy(objname, rdl->rdl_path, objnamesz);
101 			return (objname);
102 		}
103 	}
104 	return (NULL);
105 }
106 
107 prmap_t *
108 proc_obj2map(struct proc_handle *p, const char *objname)
109 {
110 	size_t i;
111 	prmap_t *map;
112 	rd_loadobj_t *rdl;
113 	char path[MAXPATHLEN];
114 
115 	for (i = 0; i < p->nobjs; i++) {
116 		rdl = &p->rdobjs[i];
117 		basename_r(rdl->rdl_path, path);
118 		if (strcmp(path, objname) == 0) {
119 			if ((map = malloc(sizeof(*map))) == NULL)
120 				return (NULL);
121 			proc_rdl2prmap(rdl, map);
122 			return (map);
123 		}
124 	}
125 	return (NULL);
126 }
127 
128 int
129 proc_iter_objs(struct proc_handle *p, proc_map_f *func, void *cd)
130 {
131 	size_t i;
132 	rd_loadobj_t *rdl;
133 	prmap_t map;
134 	char path[MAXPATHLEN];
135 	char last[MAXPATHLEN];
136 
137 	if (p->nobjs == 0)
138 		return (-1);
139 	memset(last, 0, sizeof(last));
140 	for (i = 0; i < p->nobjs; i++) {
141 		rdl = &p->rdobjs[i];
142 		proc_rdl2prmap(rdl, &map);
143 		basename_r(rdl->rdl_path, path);
144 		/*
145 		 * We shouldn't call the callback twice with the same object.
146 		 * To do that we are assuming the fact that if there are
147 		 * repeated object names (i.e. different mappings for the
148 		 * same object) they occur next to each other.
149 		 */
150 		if (strcmp(path, last) == 0)
151 			continue;
152 		(*func)(cd, &map, path);
153 		strlcpy(last, path, sizeof(last));
154 	}
155 
156 	return (0);
157 }
158 
159 prmap_t *
160 proc_addr2map(struct proc_handle *p, uintptr_t addr)
161 {
162 	size_t i;
163 	int cnt, lastvn = 0;
164 	prmap_t *map;
165 	rd_loadobj_t *rdl;
166 	struct kinfo_vmentry *kves, *kve;
167 
168 	/*
169 	 * If we don't have a cache of listed objects, we need to query
170 	 * it ourselves.
171 	 */
172 	if (p->nobjs == 0) {
173 		if ((kves = kinfo_getvmmap(p->pid, &cnt)) == NULL)
174 			return (NULL);
175 		for (i = 0; i < (size_t)cnt; i++) {
176 			kve = kves + i;
177 			if (kve->kve_type == KVME_TYPE_VNODE)
178 				lastvn = i;
179 			if (addr >= kve->kve_start && addr < kve->kve_end) {
180 				if ((map = malloc(sizeof(*map))) == NULL) {
181 					free(kves);
182 					return (NULL);
183 				}
184 				map->pr_vaddr = kve->kve_start;
185 				map->pr_size = kve->kve_end - kve->kve_start;
186 				map->pr_offset = kve->kve_offset;
187 				map->pr_mflags = 0;
188 				if (kve->kve_protection & KVME_PROT_READ)
189 					map->pr_mflags |= MA_READ;
190 				if (kve->kve_protection & KVME_PROT_WRITE)
191 					map->pr_mflags |= MA_WRITE;
192 				if (kve->kve_protection & KVME_PROT_EXEC)
193 					map->pr_mflags |= MA_EXEC;
194 				if (kve->kve_flags & KVME_FLAG_COW)
195 					map->pr_mflags |= MA_COW;
196 				if (kve->kve_flags & KVME_FLAG_NEEDS_COPY)
197 					map->pr_mflags |= MA_NEEDS_COPY;
198 				if (kve->kve_flags & KVME_FLAG_NOCOREDUMP)
199 					map->pr_mflags |= MA_NOCOREDUMP;
200 				strlcpy(map->pr_mapname, kves[lastvn].kve_path,
201 				    sizeof(map->pr_mapname));
202 				free(kves);
203 				return (map);
204 			}
205 		}
206 		free(kves);
207 		return (NULL);
208 	}
209 
210 	for (i = 0; i < p->nobjs; i++) {
211 		rdl = &p->rdobjs[i];
212 		if (addr >= rdl->rdl_saddr && addr < rdl->rdl_eaddr) {
213 			if ((map = malloc(sizeof(*map))) == NULL)
214 				return (NULL);
215 			proc_rdl2prmap(rdl, map);
216 			return (map);
217 		}
218 	}
219 	return (NULL);
220 }
221 
222 int
223 proc_addr2sym(struct proc_handle *p, uintptr_t addr, char *name,
224     size_t namesz, GElf_Sym *symcopy)
225 {
226 	Elf *e;
227 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
228 	Elf_Data *data;
229 	GElf_Shdr shdr;
230 	GElf_Sym sym;
231 	GElf_Ehdr ehdr;
232 	int fd, error = -1;
233 	size_t i;
234 	uint64_t rsym;
235 	prmap_t *map;
236 	char *s;
237 	unsigned long symtabstridx = 0, dynsymstridx = 0;
238 
239 	if ((map = proc_addr2map(p, addr)) == NULL)
240 		return (-1);
241 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
242 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
243 		goto err0;
244 	}
245 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
246 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
247 		goto err1;
248 	}
249 	if (gelf_getehdr(e, &ehdr) == NULL) {
250 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
251 		goto err2;
252 	}
253 	/*
254 	 * Find the index of the STRTAB and SYMTAB sections to locate
255 	 * symbol names.
256 	 */
257 	scn = NULL;
258 	while ((scn = elf_nextscn(e, scn)) != NULL) {
259 		gelf_getshdr(scn, &shdr);
260 		switch (shdr.sh_type) {
261 		case SHT_SYMTAB:
262 			symtabscn = scn;
263 			symtabstridx = shdr.sh_link;
264 			break;
265 		case SHT_DYNSYM:
266 			dynsymscn = scn;
267 			dynsymstridx = shdr.sh_link;
268 			break;
269 		default:
270 			break;
271 		}
272 	}
273 	/*
274 	 * Iterate over the Dynamic Symbols table to find the symbol.
275 	 * Then look up the string name in STRTAB (.dynstr)
276 	 */
277 	if ((data = elf_getdata(dynsymscn, NULL)) == NULL) {
278 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
279 		goto symtab;
280 	}
281 	i = 0;
282 	while (gelf_getsym(data, i++, &sym) != NULL) {
283 		/*
284 		 * Calculate the address mapped to the virtual memory
285 		 * by rtld.
286 		 */
287 		if (ehdr.e_type != ET_EXEC)
288 			rsym = map->pr_vaddr + sym.st_value;
289 		else
290 			rsym = sym.st_value;
291 		if (addr >= rsym && addr < rsym + sym.st_size) {
292 			s = elf_strptr(e, dynsymstridx, sym.st_name);
293 			if (s) {
294 				if (s[0] == '_' && s[1] == 'Z' && s[2])
295 					demangle(s, name, namesz);
296 				else
297 					strlcpy(name, s, namesz);
298 				memcpy(symcopy, &sym, sizeof(sym));
299 				/*
300 				 * DTrace expects the st_value to contain
301 				 * only the address relative to the start of
302 				 * the function.
303 				 */
304 				symcopy->st_value = rsym;
305 				error = 0;
306 				goto out;
307 			}
308 		}
309 	}
310 symtab:
311 	/*
312 	 * Iterate over the Symbols Table to find the symbol.
313 	 * Then look up the string name in STRTAB (.dynstr)
314 	 */
315 	if ((data = elf_getdata(symtabscn, NULL)) == NULL) {
316 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
317 		goto err2;
318 	}
319 	i = 0;
320 	while (gelf_getsym(data, i++, &sym) != NULL) {
321 		/*
322 		 * Calculate the address mapped to the virtual memory
323 		 * by rtld.
324 		 */
325 		if (ehdr.e_type != ET_EXEC)
326 			rsym = map->pr_vaddr + sym.st_value;
327 		else
328 			rsym = sym.st_value;
329 		if (addr >= rsym && addr < rsym + sym.st_size) {
330 			s = elf_strptr(e, symtabstridx, sym.st_name);
331 			if (s) {
332 				if (s[0] == '_' && s[1] == 'Z' && s[2])
333 					demangle(s, name, namesz);
334 				else
335 					strlcpy(name, s, namesz);
336 				memcpy(symcopy, &sym, sizeof(sym));
337 				/*
338 				 * DTrace expects the st_value to contain
339 				 * only the address relative to the start of
340 				 * the function.
341 				 */
342 				symcopy->st_value = rsym;
343 				error = 0;
344 				goto out;
345 			}
346 		}
347 	}
348 out:
349 err2:
350 	elf_end(e);
351 err1:
352 	close(fd);
353 err0:
354 	free(map);
355 	return (error);
356 }
357 
358 prmap_t *
359 proc_name2map(struct proc_handle *p, const char *name)
360 {
361 	size_t i;
362 	int cnt;
363 	prmap_t *map;
364 	char tmppath[MAXPATHLEN];
365 	struct kinfo_vmentry *kves, *kve;
366 	rd_loadobj_t *rdl;
367 
368 	/*
369 	 * If we haven't iterated over the list of loaded objects,
370 	 * librtld_db isn't yet initialized and it's very likely
371 	 * that librtld_db called us. We need to do the heavy
372 	 * lifting here to find the symbol librtld_db is looking for.
373 	 */
374 	if (p->nobjs == 0) {
375 		if ((kves = kinfo_getvmmap(proc_getpid(p), &cnt)) == NULL)
376 			return (NULL);
377 		for (i = 0; i < (size_t)cnt; i++) {
378 			kve = kves + i;
379 			basename_r(kve->kve_path, tmppath);
380 			if (strcmp(tmppath, name) == 0) {
381 				map = proc_addr2map(p, kve->kve_start);
382 				free(kves);
383 				return (map);
384 			}
385 		}
386 		free(kves);
387 		return (NULL);
388 	}
389 	if (name == NULL || strcmp(name, "a.out") == 0) {
390 		map = proc_addr2map(p, p->rdobjs[0].rdl_saddr);
391 		return (map);
392 	}
393 	for (i = 0; i < p->nobjs; i++) {
394 		rdl = &p->rdobjs[i];
395 		basename_r(rdl->rdl_path, tmppath);
396 		if (strcmp(tmppath, name) == 0) {
397 			if ((map = malloc(sizeof(*map))) == NULL)
398 				return (NULL);
399 			proc_rdl2prmap(rdl, map);
400 			return (map);
401 		}
402 	}
403 
404 	return (NULL);
405 }
406 
407 int
408 proc_name2sym(struct proc_handle *p, const char *object, const char *symbol,
409     GElf_Sym *symcopy)
410 {
411 	Elf *e;
412 	Elf_Scn *scn, *dynsymscn = NULL, *symtabscn = NULL;
413 	Elf_Data *data;
414 	GElf_Shdr shdr;
415 	GElf_Sym sym;
416 	GElf_Ehdr ehdr;
417 	int fd, error = -1;
418 	size_t i;
419 	prmap_t *map;
420 	char *s;
421 	unsigned long symtabstridx = 0, dynsymstridx = 0;
422 
423 	if ((map = proc_name2map(p, object)) == NULL) {
424 		DPRINTFX("ERROR: couldn't find object %s", object);
425 		goto err0;
426 	}
427 	if ((fd = open(map->pr_mapname, O_RDONLY, 0)) < 0) {
428 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
429 		goto err0;
430 	}
431 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
432 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
433 		goto err1;
434 	}
435 	if (gelf_getehdr(e, &ehdr) == NULL) {
436 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
437 		goto err2;
438 	}
439 	/*
440 	 * Find the index of the STRTAB and SYMTAB sections to locate
441 	 * symbol names.
442 	 */
443 	scn = NULL;
444 	while ((scn = elf_nextscn(e, scn)) != NULL) {
445 		gelf_getshdr(scn, &shdr);
446 		switch (shdr.sh_type) {
447 		case SHT_SYMTAB:
448 			symtabscn = scn;
449 			symtabstridx = shdr.sh_link;
450 			break;
451 		case SHT_DYNSYM:
452 			dynsymscn = scn;
453 			dynsymstridx = shdr.sh_link;
454 			break;
455 		default:
456 			break;
457 		}
458 	}
459 	/*
460 	 * Iterate over the Dynamic Symbols table to find the symbol.
461 	 * Then look up the string name in STRTAB (.dynstr)
462 	 */
463 	if ((data = elf_getdata(dynsymscn, NULL))) {
464 		i = 0;
465 		while (gelf_getsym(data, i++, &sym) != NULL) {
466 			s = elf_strptr(e, dynsymstridx, sym.st_name);
467 			if (s && strcmp(s, symbol) == 0) {
468 				memcpy(symcopy, &sym, sizeof(sym));
469 				if (ehdr.e_type != ET_EXEC)
470 					symcopy->st_value += map->pr_vaddr;
471 				error = 0;
472 				goto out;
473 			}
474 		}
475 	}
476 	/*
477 	 * Iterate over the Symbols Table to find the symbol.
478 	 * Then look up the string name in STRTAB (.dynstr)
479 	 */
480 	if ((data = elf_getdata(symtabscn, NULL))) {
481 		i = 0;
482 		while (gelf_getsym(data, i++, &sym) != NULL) {
483 			s = elf_strptr(e, symtabstridx, sym.st_name);
484 			if (s && strcmp(s, symbol) == 0) {
485 				memcpy(symcopy, &sym, sizeof(sym));
486 				if (ehdr.e_type != ET_EXEC)
487 					symcopy->st_value += map->pr_vaddr;
488 				error = 0;
489 				goto out;
490 			}
491 		}
492 	}
493 out:
494 	DPRINTFX("found addr 0x%lx for %s", symcopy->st_value, symbol);
495 err2:
496 	elf_end(e);
497 err1:
498 	close(fd);
499 err0:
500 	free(map);
501 
502 	return (error);
503 }
504 
505 
506 int
507 proc_iter_symbyaddr(struct proc_handle *p, const char *object, int which,
508     int mask, proc_sym_f *func, void *cd)
509 {
510 	Elf *e;
511 	int i, fd;
512 	prmap_t *map;
513 	Elf_Scn *scn, *foundscn = NULL;
514 	Elf_Data *data;
515 	GElf_Ehdr ehdr;
516 	GElf_Shdr shdr;
517 	GElf_Sym sym;
518 	unsigned long stridx = -1;
519 	char *s;
520 	int error = -1;
521 
522 	if ((map = proc_name2map(p, object)) == NULL)
523 		return (-1);
524 	if ((fd = open(map->pr_mapname, O_RDONLY)) < 0) {
525 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
526 		goto err0;
527 	}
528 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
529 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
530 		goto err1;
531 	}
532 	if (gelf_getehdr(e, &ehdr) == NULL) {
533 		DPRINTFX("ERROR: gelf_getehdr() failed: %s", elf_errmsg(-1));
534 		goto err2;
535 	}
536 	/*
537 	 * Find the section we are looking for.
538 	 */
539 	scn = NULL;
540 	while ((scn = elf_nextscn(e, scn)) != NULL) {
541 		gelf_getshdr(scn, &shdr);
542 		if (which == PR_SYMTAB &&
543 		    shdr.sh_type == SHT_SYMTAB) {
544 			foundscn = scn;
545 			break;
546 		} else if (which == PR_DYNSYM &&
547 		    shdr.sh_type == SHT_DYNSYM) {
548 			foundscn = scn;
549 			break;
550 		}
551 	}
552 	if (!foundscn)
553 		return (-1);
554 	stridx = shdr.sh_link;
555 	if ((data = elf_getdata(foundscn, NULL)) == NULL) {
556 		DPRINTFX("ERROR: elf_getdata() failed: %s", elf_errmsg(-1));
557 		goto err2;
558 	}
559 	i = 0;
560 	while (gelf_getsym(data, i++, &sym) != NULL) {
561 		if (GELF_ST_BIND(sym.st_info) == STB_LOCAL &&
562 		    (mask & BIND_LOCAL) == 0)
563 			continue;
564 		if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL &&
565 		    (mask & BIND_GLOBAL) == 0)
566 			continue;
567 		if (GELF_ST_BIND(sym.st_info) == STB_WEAK &&
568 		    (mask & BIND_WEAK) == 0)
569 			continue;
570 		if (GELF_ST_TYPE(sym.st_info) == STT_NOTYPE &&
571 		    (mask & TYPE_NOTYPE) == 0)
572 			continue;
573 		if (GELF_ST_TYPE(sym.st_info) == STT_OBJECT &&
574 		    (mask & TYPE_OBJECT) == 0)
575 			continue;
576 		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC &&
577 		    (mask & TYPE_FUNC) == 0)
578 			continue;
579 		if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
580 		    (mask & TYPE_SECTION) == 0)
581 			continue;
582 		if (GELF_ST_TYPE(sym.st_info) == STT_FILE &&
583 		    (mask & TYPE_FILE) == 0)
584 			continue;
585 		s = elf_strptr(e, stridx, sym.st_name);
586 		if (ehdr.e_type != ET_EXEC)
587 			sym.st_value += map->pr_vaddr;
588 		(*func)(cd, &sym, s);
589 	}
590 	error = 0;
591 err2:
592 	elf_end(e);
593 err1:
594 	close(fd);
595 err0:
596 	free(map);
597 	return (error);
598 }
599