xref: /freebsd/lib/libproc/proc_sym.c (revision 7f9dff23d3092aa33ad45b2b63e52469b3c13a6e)
1 /*-
2  * Copyright (c) 2016 Mark Johnston <markj@FreeBSD.org>
3  * Copyright (c) 2010 The FreeBSD Foundation
4  * Copyright (c) 2008 John Birrell (jb@freebsd.org)
5  * All rights reserved.
6  *
7  * Portions of this software were developed by Rui Paulo under sponsorship
8  * from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/types.h>
36 #ifndef NO_CTF
37 #include <sys/ctf.h>
38 #include <sys/ctf_api.h>
39 #endif
40 #include <sys/user.h>
41 
42 #include <assert.h>
43 #include <err.h>
44 #include <fcntl.h>
45 #include <libgen.h>
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <string.h>
49 #include <unistd.h>
50 #ifndef NO_CTF
51 #include <libctf.h>
52 #endif
53 #include <libutil.h>
54 
55 #include "crc32.h"
56 #include "_libproc.h"
57 
58 #define	PATH_DEBUG_DIR	"/usr/lib/debug"
59 
60 #ifdef NO_CTF
61 typedef struct ctf_file ctf_file_t;
62 #endif
63 
64 #ifndef NO_CXA_DEMANGLE
65 extern char *__cxa_demangle(const char *, char *, size_t *, int *);
66 #endif /* NO_CXA_DEMANGLE */
67 
68 static int
69 crc32_file(int fd, uint32_t *crc)
70 {
71 	uint8_t buf[PAGE_SIZE], *p;
72 	size_t n;
73 
74 	*crc = ~0;
75 	while ((n = read(fd, buf, sizeof(buf))) > 0) {
76 		p = &buf[0];
77 		while (n-- > 0)
78 			*crc = crc32_tab[(*crc ^ *p++) & 0xff] ^ (*crc >> 8);
79 	}
80 	*crc = ~*crc;
81 	return (n);
82 }
83 
84 static void
85 demangle(const char *symbol, char *buf, size_t len)
86 {
87 #ifndef NO_CXA_DEMANGLE
88 	char *dembuf;
89 
90 	if (symbol[0] == '_' && symbol[1] == 'Z' && symbol[2]) {
91 		dembuf = __cxa_demangle(symbol, NULL, NULL, NULL);
92 		if (!dembuf)
93 			goto fail;
94 		strlcpy(buf, dembuf, len);
95 		free(dembuf);
96 		return;
97 	}
98 fail:
99 #endif /* NO_CXA_DEMANGLE */
100 	strlcpy(buf, symbol, len);
101 }
102 
103 static int
104 symvalcomp(void *thunk, const void *a1, const void *a2)
105 {
106 	struct symtab *symtab;
107 	GElf_Sym sym1, sym2;
108 	u_int i1, i2;
109 	int ret;
110 
111 	i1 = *(const u_int *)a1;
112 	i2 = *(const u_int *)a2;
113 	symtab = thunk;
114 
115 	(void)gelf_getsym(symtab->data, i1, &sym1);
116 	(void)gelf_getsym(symtab->data, i2, &sym2);
117 	if (sym1.st_value < sym2.st_value)
118 		ret = -1;
119 	else if (sym1.st_value == sym2.st_value)
120 		ret = 0;
121 	else
122 		ret = 1;
123 	return (ret);
124 }
125 
126 static int
127 load_symtab(Elf *e, struct symtab *symtab, u_long sh_type)
128 {
129 	GElf_Ehdr ehdr;
130 	GElf_Shdr shdr;
131 	Elf_Scn *scn;
132 	u_int nsyms;
133 
134 	if (gelf_getehdr(e, &ehdr) == NULL)
135 		return (-1);
136 
137 	scn = NULL;
138 	while ((scn = elf_nextscn(e, scn)) != NULL) {
139 		(void)gelf_getshdr(scn, &shdr);
140 		if (shdr.sh_type == sh_type)
141 			break;
142 	}
143 	if (scn == NULL)
144 		return (-1);
145 
146 	if ((symtab->data = elf_getdata(scn, NULL)) == NULL)
147 		return (-1);
148 
149 	nsyms = shdr.sh_size / shdr.sh_entsize;
150 
151 	symtab->index = calloc(nsyms, sizeof(u_int));
152 	if (symtab->index == NULL)
153 		return (-1);
154 	for (u_int i = 0; i < nsyms; i++)
155 		symtab->index[i] = i;
156 	qsort_r(symtab->index, nsyms, sizeof(u_int), symtab, symvalcomp);
157 	symtab->nsyms = nsyms;
158 	symtab->stridx = shdr.sh_link;
159 	return (0);
160 }
161 
162 static void
163 load_symtabs(struct file_info *file)
164 {
165 
166 	file->symtab.nsyms = file->dynsymtab.nsyms = 0;
167 	(void)load_symtab(file->elf, &file->symtab, SHT_SYMTAB);
168 	(void)load_symtab(file->elf, &file->dynsymtab, SHT_DYNSYM);
169 }
170 
171 static int
172 open_debug_file(char *path, const char *debugfile, uint32_t crc)
173 {
174 	size_t n;
175 	uint32_t compcrc;
176 	int fd;
177 
178 	fd = -1;
179 	if ((n = strlcat(path, "/", PATH_MAX)) >= PATH_MAX)
180 		return (fd);
181 	if (strlcat(path, debugfile, PATH_MAX) >= PATH_MAX)
182 		goto out;
183 	if ((fd = open(path, O_RDONLY | O_CLOEXEC)) < 0)
184 		goto out;
185 	if (crc32_file(fd, &compcrc) != 0 || crc != compcrc) {
186 		DPRINTFX("ERROR: CRC32 mismatch for %s", path);
187 		(void)close(fd);
188 		fd = -1;
189 	}
190 out:
191 	path[n] = '\0';
192 	return (fd);
193 }
194 
195 /*
196  * Obtain an ELF descriptor for the specified mapped object. If a GNU debuglink
197  * section is present, a descriptor for the corresponding debug file is
198  * returned.
199  */
200 static int
201 open_object(struct map_info *mapping)
202 {
203 	char path[PATH_MAX];
204 	GElf_Shdr shdr;
205 	Elf *e, *e2;
206 	Elf_Data *data;
207 	Elf_Scn *scn;
208 	struct file_info *file;
209 	prmap_t *map;
210 	const char *debugfile, *scnname;
211 	size_t ndx;
212 	uint32_t crc;
213 	int fd, fd2;
214 
215 	if (mapping->map.pr_mapname[0] == '\0')
216 		return (-1); /* anonymous object */
217 	if (mapping->file->elf != NULL)
218 		return (0); /* already loaded */
219 
220 	file = mapping->file;
221 	map = &mapping->map;
222 	if ((fd = open(map->pr_mapname, O_RDONLY | O_CLOEXEC)) < 0) {
223 		DPRINTF("ERROR: open %s failed", map->pr_mapname);
224 		return (-1);
225 	}
226 	if ((e = elf_begin(fd, ELF_C_READ, NULL)) == NULL) {
227 		DPRINTFX("ERROR: elf_begin() failed: %s", elf_errmsg(-1));
228 		goto err;
229 	}
230 	if (gelf_getehdr(e, &file->ehdr) != &file->ehdr) {
231 		DPRINTFX("ERROR: elf_getehdr() failed: %s", elf_errmsg(-1));
232 		goto err;
233 	}
234 
235 	scn = NULL;
236 	while ((scn = elf_nextscn(e, scn)) != NULL) {
237 		if (gelf_getshdr(scn, &shdr) != &shdr) {
238 			DPRINTFX("ERROR: gelf_getshdr failed: %s",
239 			    elf_errmsg(-1));
240 			goto err;
241 		}
242 		if (shdr.sh_type != SHT_PROGBITS)
243 			continue;
244 		if (elf_getshdrstrndx(e, &ndx) != 0) {
245 			DPRINTFX("ERROR: elf_getshdrstrndx failed: %s",
246 			    elf_errmsg(-1));
247 			goto err;
248 		}
249 		if ((scnname = elf_strptr(e, ndx, shdr.sh_name)) == NULL)
250 			continue;
251 
252 		if (strcmp(scnname, ".gnu_debuglink") == 0)
253 			break;
254 	}
255 	if (scn == NULL)
256 		goto internal;
257 
258 	if ((data = elf_getdata(scn, NULL)) == NULL) {
259 		DPRINTFX("ERROR: elf_getdata failed: %s", elf_errmsg(-1));
260 		goto err;
261 	}
262 
263 	/*
264 	 * The data contains a null-terminated file name followed by a 4-byte
265 	 * CRC.
266 	 */
267 	if (data->d_size < sizeof(crc) + 1) {
268 		DPRINTFX("ERROR: debuglink section is too small (%zd bytes)",
269 		    data->d_size);
270 		goto internal;
271 	}
272 	if (strnlen(data->d_buf, data->d_size) >= data->d_size - sizeof(crc)) {
273 		DPRINTFX("ERROR: no null-terminator in gnu_debuglink section");
274 		goto internal;
275 	}
276 
277 	debugfile = data->d_buf;
278 	memcpy(&crc, (char *)data->d_buf + data->d_size - sizeof(crc),
279 	    sizeof(crc));
280 
281 	/*
282 	 * Search for the debug file using the algorithm described in the gdb
283 	 * documentation:
284 	 * - look in the directory containing the object,
285 	 * - look in the subdirectory ".debug" of the directory containing the
286 	 *   object,
287 	 * - look in the global debug directories (currently /usr/lib/debug).
288 	 */
289 	(void)strlcpy(path, map->pr_mapname, sizeof(path));
290 	(void)dirname(path);
291 
292 	if ((fd2 = open_debug_file(path, debugfile, crc)) >= 0)
293 		goto external;
294 
295 	if (strlcat(path, "/.debug", sizeof(path)) < sizeof(path) &&
296 	    (fd2 = open_debug_file(path, debugfile, crc)) >= 0)
297 		goto external;
298 
299 	(void)snprintf(path, sizeof(path), PATH_DEBUG_DIR);
300 	if (strlcat(path, map->pr_mapname, sizeof(path)) < sizeof(path)) {
301 		(void)dirname(path);
302 		if ((fd2 = open_debug_file(path, debugfile, crc)) >= 0)
303 			goto external;
304 	}
305 
306 internal:
307 	/* We didn't find a debug file, just return the object's descriptor. */
308 	file->elf = e;
309 	file->fd = fd;
310 	load_symtabs(file);
311 	return (0);
312 
313 external:
314 	if ((e2 = elf_begin(fd2, ELF_C_READ, NULL)) == NULL) {
315 		DPRINTFX("ERROR: elf_begin failed: %s", elf_errmsg(-1));
316 		(void)close(fd2);
317 		goto err;
318 	}
319 	(void)elf_end(e);
320 	(void)close(fd);
321 	file->elf = e2;
322 	file->fd = fd2;
323 	load_symtabs(file);
324 	return (0);
325 
326 err:
327 	if (e != NULL)
328 		(void)elf_end(e);
329 	(void)close(fd);
330 	return (-1);
331 }
332 
333 char *
334 proc_objname(struct proc_handle *p, uintptr_t addr, char *objname,
335     size_t objnamesz)
336 {
337 	prmap_t *map;
338 	size_t i;
339 
340 	for (i = 0; i < p->nmappings; i++) {
341 		map = &p->mappings[i].map;
342 		if (addr >= map->pr_vaddr &&
343 		    addr < map->pr_vaddr + map->pr_size) {
344 			strlcpy(objname, map->pr_mapname, objnamesz);
345 			return (objname);
346 		}
347 	}
348 	return (NULL);
349 }
350 
351 int
352 proc_iter_objs(struct proc_handle *p, proc_map_f *func, void *cd)
353 {
354 	char last[MAXPATHLEN], path[MAXPATHLEN], *base;
355 	prmap_t *map;
356 	size_t i;
357 	int error;
358 
359 	if (p->nmappings == 0)
360 		if (proc_rdagent(p) == NULL)
361 			return (-1);
362 
363 	error = 0;
364 	memset(last, 0, sizeof(last));
365 	for (i = 0; i < p->nmappings; i++) {
366 		map = &p->mappings[i].map;
367 		strlcpy(path, map->pr_mapname, sizeof(path));
368 		base = basename(path);
369 		/*
370 		 * We shouldn't call the callback twice with the same object.
371 		 * To do that we are assuming the fact that if there are
372 		 * repeated object names (i.e. different mappings for the
373 		 * same object) they occur next to each other.
374 		 */
375 		if (strcmp(base, last) == 0)
376 			continue;
377 		if ((error = (*func)(cd, map, base)) != 0)
378 			break;
379 		strlcpy(last, path, sizeof(last));
380 	}
381 	return (error);
382 }
383 
384 static struct map_info *
385 _proc_addr2map(struct proc_handle *p, uintptr_t addr)
386 {
387 	struct map_info *mapping;
388 	size_t i;
389 
390 	if (p->nmappings == 0)
391 		if (proc_rdagent(p) == NULL)
392 			return (NULL);
393 	for (i = 0; i < p->nmappings; i++) {
394 		mapping = &p->mappings[i];
395 		if (addr >= mapping->map.pr_vaddr &&
396 		    addr < mapping->map.pr_vaddr + mapping->map.pr_size)
397 			return (mapping);
398 	}
399 	return (NULL);
400 }
401 
402 prmap_t *
403 proc_addr2map(struct proc_handle *p, uintptr_t addr)
404 {
405 
406 	return (&_proc_addr2map(p, addr)->map);
407 }
408 
409 /*
410  * Look up the symbol at addr using a binary search, returning a copy of the
411  * symbol and its name.
412  */
413 static int
414 lookup_symbol_by_addr(Elf *elf, struct symtab *symtab, uintptr_t addr,
415     const char **namep, GElf_Sym *sym)
416 {
417 	Elf_Data *data;
418 	const char *s;
419 	int min, max, mid;
420 
421 	data = symtab->data;
422 	min = 0;
423 	max = symtab->nsyms - 1;
424 
425 	while (min <= max) {
426 		mid = (max + min) / 2;
427 		(void)gelf_getsym(data, symtab->index[mid], sym);
428 		if (addr >= sym->st_value &&
429 		    addr < sym->st_value + sym->st_size) {
430 			s = elf_strptr(elf, symtab->stridx, sym->st_name);
431 			if (s != NULL && namep != NULL)
432 				*namep = s;
433 			return (0);
434 		}
435 
436 		if (addr < sym->st_value)
437 			max = mid - 1;
438 		else
439 			min = mid + 1;
440 	}
441 	return (ENOENT);
442 }
443 
444 int
445 proc_addr2sym(struct proc_handle *p, uintptr_t addr, char *name,
446     size_t namesz, GElf_Sym *symcopy)
447 {
448 	struct file_info *file;
449 	struct map_info *mapping;
450 	const char *s;
451 	uintptr_t off;
452 	int error;
453 
454 	if ((mapping = _proc_addr2map(p, addr)) == NULL) {
455 		DPRINTFX("ERROR: proc_addr2map failed to resolve 0x%jx", addr);
456 		return (-1);
457 	}
458 	if (open_object(mapping) != 0) {
459 		DPRINTFX("ERROR: failed to open object %s",
460 		    mapping->map.pr_mapname);
461 		return (-1);
462 	}
463 
464 	file = mapping->file;
465 	off = file->ehdr.e_type == ET_DYN ? mapping->map.pr_vaddr : 0;
466 	if (addr < off)
467 		return (ENOENT);
468 	addr -= off;
469 
470 	error = lookup_symbol_by_addr(file->elf, &file->dynsymtab, addr, &s,
471 	    symcopy);
472 	if (error == ENOENT)
473 		error = lookup_symbol_by_addr(file->elf, &file->symtab, addr,
474 		    &s, symcopy);
475 	if (error == 0) {
476 		symcopy->st_value += off;
477 		demangle(s, name, namesz);
478 	}
479 	return (error);
480 }
481 
482 static struct map_info *
483 _proc_name2map(struct proc_handle *p, const char *name)
484 {
485 	char path[MAXPATHLEN], *base;
486 	struct map_info *mapping;
487 	size_t i, len;
488 
489 	if ((len = strlen(name)) == 0)
490 		return (NULL);
491 	if (p->nmappings == 0)
492 		if (proc_rdagent(p) == NULL)
493 			return (NULL);
494 	for (i = 0; i < p->nmappings; i++) {
495 		mapping = &p->mappings[i];
496 		(void)strlcpy(path, mapping->map.pr_mapname, sizeof(path));
497 		base = basename(path);
498 		if (strcmp(base, name) == 0)
499 			return (mapping);
500 	}
501 	/* If we didn't find a match, try matching prefixes of the basename. */
502 	for (i = 0; i < p->nmappings; i++) {
503 		strlcpy(path, p->mappings[i].map.pr_mapname, sizeof(path));
504 		base = basename(path);
505 		if (strncmp(base, name, len) == 0)
506 			return (&p->mappings[i]);
507 	}
508 	if (strcmp(name, "a.out") == 0)
509 		return (_proc_addr2map(p, p->exec_map->pr_vaddr));
510 	return (NULL);
511 }
512 
513 prmap_t *
514 proc_name2map(struct proc_handle *p, const char *name)
515 {
516 
517 	return (&_proc_name2map(p, name)->map);
518 }
519 
520 /*
521  * Look up the symbol with the given name and return a copy of it.
522  */
523 static int
524 lookup_symbol_by_name(Elf *elf, struct symtab *symtab, const char *symbol,
525     GElf_Sym *symcopy, prsyminfo_t *si)
526 {
527 	GElf_Sym sym;
528 	Elf_Data *data;
529 	char *s;
530 	int i;
531 
532 	if (symtab->nsyms == 0)
533 		return (ENOENT);
534 	data = symtab->data;
535 	for (i = 0; gelf_getsym(data, i, &sym) != NULL; i++) {
536 		s = elf_strptr(elf, symtab->stridx, sym.st_name);
537 		if (s != NULL && strcmp(s, symbol) == 0) {
538 			memcpy(symcopy, &sym, sizeof(*symcopy));
539 			if (si != NULL)
540 				si->prs_id = i;
541 			return (0);
542 		}
543 	}
544 	return (ENOENT);
545 }
546 
547 int
548 proc_name2sym(struct proc_handle *p, const char *object, const char *symbol,
549     GElf_Sym *symcopy, prsyminfo_t *si)
550 {
551 	struct file_info *file;
552 	struct map_info *mapping;
553 	uintptr_t off;
554 	int error;
555 
556 	if ((mapping = _proc_name2map(p, object)) == NULL) {
557 		DPRINTFX("ERROR: proc_name2map failed to resolve %s", object);
558 		return (-1);
559 	}
560 	if (open_object(mapping) != 0) {
561 		DPRINTFX("ERROR: failed to open object %s",
562 		    mapping->map.pr_mapname);
563 		return (-1);
564 	}
565 
566 	file = mapping->file;
567 	off = file->ehdr.e_type == ET_DYN ? mapping->map.pr_vaddr : 0;
568 
569 	error = lookup_symbol_by_name(file->elf, &file->dynsymtab, symbol,
570 	    symcopy, si);
571 	if (error == ENOENT)
572 		error = lookup_symbol_by_name(file->elf, &file->symtab, symbol,
573 		    symcopy, si);
574 	if (error == 0)
575 		symcopy->st_value += off;
576 	return (error);
577 }
578 
579 ctf_file_t *
580 proc_name2ctf(struct proc_handle *p, const char *name)
581 {
582 #ifndef NO_CTF
583 	ctf_file_t *ctf;
584 	prmap_t *map;
585 	int error;
586 
587 	if ((map = proc_name2map(p, name)) == NULL)
588 		return (NULL);
589 
590 	ctf = ctf_open(map->pr_mapname, &error);
591 	return (ctf);
592 #else
593 	(void)p;
594 	(void)name;
595 	return (NULL);
596 #endif
597 }
598 
599 int
600 proc_iter_symbyaddr(struct proc_handle *p, const char *object, int which,
601     int mask, proc_sym_f *func, void *cd)
602 {
603 	GElf_Sym sym;
604 	struct file_info *file;
605 	struct map_info *mapping;
606 	struct symtab *symtab;
607 	const char *s;
608 	int error, i;
609 
610 	if ((mapping = _proc_name2map(p, object)) == NULL) {
611 		DPRINTFX("ERROR: proc_name2map failed to resolve %s", object);
612 		return (-1);
613 	}
614 	if (open_object(mapping) != 0) {
615 		DPRINTFX("ERROR: failed to open object %s",
616 		    mapping->map.pr_mapname);
617 		return (-1);
618 	}
619 
620 	file = mapping->file;
621 	symtab = which == PR_SYMTAB ? &file->symtab : &file->dynsymtab;
622 	if (symtab->nsyms == 0)
623 		return (-1);
624 
625 	error = 0;
626 	for (i = 0; gelf_getsym(symtab->data, i, &sym) != NULL; i++) {
627 		if (GELF_ST_BIND(sym.st_info) == STB_LOCAL &&
628 		    (mask & BIND_LOCAL) == 0)
629 			continue;
630 		if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL &&
631 		    (mask & BIND_GLOBAL) == 0)
632 			continue;
633 		if (GELF_ST_BIND(sym.st_info) == STB_WEAK &&
634 		    (mask & BIND_WEAK) == 0)
635 			continue;
636 		if (GELF_ST_TYPE(sym.st_info) == STT_NOTYPE &&
637 		    (mask & TYPE_NOTYPE) == 0)
638 			continue;
639 		if (GELF_ST_TYPE(sym.st_info) == STT_OBJECT &&
640 		    (mask & TYPE_OBJECT) == 0)
641 			continue;
642 		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC &&
643 		    (mask & TYPE_FUNC) == 0)
644 			continue;
645 		if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
646 		    (mask & TYPE_SECTION) == 0)
647 			continue;
648 		if (GELF_ST_TYPE(sym.st_info) == STT_FILE &&
649 		    (mask & TYPE_FILE) == 0)
650 			continue;
651 		s = elf_strptr(file->elf, symtab->stridx, sym.st_name);
652 		if (file->ehdr.e_type == ET_DYN)
653 			sym.st_value += mapping->map.pr_vaddr;
654 		if ((error = (*func)(cd, &sym, s)) != 0)
655 			break;
656 	}
657 	return (error);
658 }
659