xref: /titanic_44/usr/src/cmd/dis/dis_target.c (revision 9729663dccf3af95363464bd23939c153a918fd9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24  *
25  * Copyright 2011 Jason King.  All rights reserved.
26  */
27 
28 #include <assert.h>
29 #include <errno.h>
30 #include <fcntl.h>
31 #include <gelf.h>
32 #include <libelf.h>
33 #include <stdlib.h>
34 #include <string.h>
35 #include <unistd.h>
36 
37 #include <sys/fcntl.h>
38 #include <sys/stat.h>
39 #include <sys/sysmacros.h>
40 #include <sys/types.h>
41 
42 #include "dis_target.h"
43 #include "dis_util.h"
44 
45 /*
46  * Standard ELF disassembler target.
47  *
48  * We only support disassembly of ELF files, though this target interface could
49  * be extended in the future.  Each basic type (target, func, section) contains
50  * enough information to uniquely identify the location within the file.  The
51  * interfaces use libelf(3LIB) to do the actual processing of the file.
52  */
53 
54 /*
55  * Symbol table entry type.  We maintain our own symbol table sorted by address,
56  * with the symbol name already resolved against the ELF symbol table.
57  */
58 typedef struct sym_entry {
59 	GElf_Sym	se_sym;		/* value of symbol */
60 	char		*se_name;	/* name of symbol */
61 	int		se_shndx;	/* section where symbol is located */
62 } sym_entry_t;
63 
64 /*
65  * Create a map of the virtual address ranges of every section.  This will
66  * allow us to create dummpy mappings for unassigned addresses.  Otherwise
67  * multiple sections with unassigned addresses will appear to overlap and
68  * mess up symbol resolution (which uses the virtual address).
69  */
70 typedef struct dis_shnmap {
71 	const char 	*dm_name;	/* name of section */
72 	uint64_t	dm_start;	/* virtual address of section */
73 	size_t		dm_length;	/* address length */
74 	boolean_t	dm_mapped;	/* did we assign the mapping */
75 } dis_shnmap_t;
76 
77 /*
78  * Target data structure.  This structure keeps track of the ELF file
79  * information, a few bits of pre-processed section index information, and
80  * sorted versions of the symbol table.  We also keep track of the last symbol
81  * looked up, as the majority of lookups remain within the same symbol.
82  */
83 struct dis_tgt {
84 	Elf		*dt_elf;	/* libelf handle */
85 	Elf		*dt_elf_root;	/* main libelf handle (for archives) */
86 	const char	*dt_filename;	/* name of file */
87 	int		dt_fd;		/* underlying file descriptor */
88 	size_t		dt_shstrndx;	/* section index of .shstrtab */
89 	size_t		dt_symidx;	/* section index of symbol table */
90 	sym_entry_t	*dt_symcache;	/* last symbol looked up */
91 	sym_entry_t	*dt_symtab;	/* sorted symbol table */
92 	int		dt_symcount;	/* # of symbol table entries */
93 	struct dis_tgt	*dt_next;	/* next target (for archives) */
94 	Elf_Arhdr	*dt_arhdr;	/* archive header (for archives) */
95 	dis_shnmap_t	*dt_shnmap;	/* section address map */
96 	size_t		dt_shncount;	/* # of sections in target */
97 };
98 
99 /*
100  * Function data structure.  We resolve the symbol and lookup the associated ELF
101  * data when building this structure.  The offset is calculated based on the
102  * section's starting address.
103  */
104 struct dis_func {
105 	sym_entry_t	*df_sym;	/* symbol table reference */
106 	Elf_Data	*df_data;	/* associated ELF data */
107 	size_t		df_offset;	/* offset within data */
108 };
109 
110 /*
111  * Section data structure.  We store the entire section header so that we can
112  * determine some properties (such as whether or not it contains text) after
113  * building the structure.
114  */
115 struct dis_scn {
116 	GElf_Shdr	ds_shdr;
117 	const char	*ds_name;
118 	Elf_Data	*ds_data;
119 };
120 
121 /* Lifted from Psymtab.c, omitting STT_TLS */
122 #define	DATA_TYPES      \
123 	((1 << STT_OBJECT) | (1 << STT_FUNC) | (1 << STT_COMMON))
124 #define	IS_DATA_TYPE(tp)	(((1 << (tp)) & DATA_TYPES) != 0)
125 
126 /*
127  * Save the virtual address range for this section and select the
128  * best section to use as the symbol table.  We prefer SHT_SYMTAB
129  * over SHT_DYNSYM.
130  */
131 /* ARGSUSED */
132 static void
133 tgt_scn_init(dis_tgt_t *tgt, dis_scn_t *scn, void *data)
134 {
135 	int *index = data;
136 
137 	*index += 1;
138 
139 	tgt->dt_shnmap[*index].dm_name = scn->ds_name;
140 	tgt->dt_shnmap[*index].dm_start = scn->ds_shdr.sh_addr;
141 	tgt->dt_shnmap[*index].dm_length = scn->ds_shdr.sh_size;
142 	tgt->dt_shnmap[*index].dm_mapped = B_FALSE;
143 
144 	/*
145 	 * Prefer SHT_SYMTAB over SHT_DYNSYM
146 	 */
147 	if (scn->ds_shdr.sh_type == SHT_DYNSYM && tgt->dt_symidx == 0)
148 		tgt->dt_symidx = *index;
149 	else if (scn->ds_shdr.sh_type == SHT_SYMTAB)
150 		tgt->dt_symidx = *index;
151 }
152 
153 static int
154 sym_compare(const void *a, const void *b)
155 {
156 	const sym_entry_t *syma = a;
157 	const sym_entry_t *symb = b;
158 	const char *aname = syma->se_name;
159 	const char *bname = symb->se_name;
160 
161 	if (syma->se_sym.st_value < symb->se_sym.st_value)
162 		return (-1);
163 
164 	if (syma->se_sym.st_value > symb->se_sym.st_value)
165 		return (1);
166 
167 	/*
168 	 * Prefer functions over non-functions
169 	 */
170 	if (GELF_ST_TYPE(syma->se_sym.st_info) !=
171 	    GELF_ST_TYPE(symb->se_sym.st_info)) {
172 		if (GELF_ST_TYPE(syma->se_sym.st_info) == STT_FUNC)
173 			return (-1);
174 		if (GELF_ST_TYPE(symb->se_sym.st_info) == STT_FUNC)
175 			return (1);
176 	}
177 
178 	/*
179 	 * For symbols with the same address and type, we sort them according to
180 	 * a hierarchy:
181 	 *
182 	 * 	1. weak symbols (common name)
183 	 * 	2. global symbols (external name)
184 	 * 	3. local symbols
185 	 */
186 	if (GELF_ST_BIND(syma->se_sym.st_info) !=
187 	    GELF_ST_BIND(symb->se_sym.st_info)) {
188 		if (GELF_ST_BIND(syma->se_sym.st_info) == STB_WEAK)
189 			return (-1);
190 		if (GELF_ST_BIND(symb->se_sym.st_info) == STB_WEAK)
191 			return (1);
192 
193 		if (GELF_ST_BIND(syma->se_sym.st_info) == STB_GLOBAL)
194 			return (-1);
195 		if (GELF_ST_BIND(symb->se_sym.st_info) == STB_GLOBAL)
196 			return (1);
197 	}
198 
199 	/*
200 	 * As a last resort, if we have multiple symbols of the same type at the
201 	 * same address, prefer the version with the fewest leading underscores.
202 	 */
203 	if (aname == NULL)
204 		return (-1);
205 	if (bname == NULL)
206 		return (1);
207 
208 	while (*aname == '_' && *bname == '_') {
209 		aname++;
210 		bname++;
211 	}
212 
213 	if (*bname == '_')
214 		return (-1);
215 	if (*aname == '_')
216 		return (1);
217 
218 	/*
219 	 * Prefer the symbol with the smaller size.
220 	 */
221 	if (syma->se_sym.st_size < symb->se_sym.st_size)
222 		return (-1);
223 	if (syma->se_sym.st_size > symb->se_sym.st_size)
224 		return (1);
225 
226 	/*
227 	 * We really do have two identical symbols for some reason.  Just report
228 	 * them as equal, and to the lucky one go the spoils.
229 	 */
230 	return (0);
231 }
232 
233 /*
234  * Construct an optimized symbol table sorted by starting address.
235  */
236 static void
237 construct_symtab(dis_tgt_t *tgt)
238 {
239 	Elf_Scn *scn;
240 	GElf_Shdr shdr;
241 	Elf_Data *symdata;
242 	int i;
243 	GElf_Word *symshndx = NULL;
244 	int symshndx_size;
245 	sym_entry_t *sym;
246 	sym_entry_t *p_symtab = NULL;
247 	int nsym = 0; /* count of symbols we're not interested in */
248 
249 	/*
250 	 * Find the symshndx section, if any
251 	 */
252 	for (scn = elf_nextscn(tgt->dt_elf, NULL); scn != NULL;
253 	    scn = elf_nextscn(tgt->dt_elf, scn)) {
254 		if (gelf_getshdr(scn, &shdr) == NULL)
255 			break;
256 		if (shdr.sh_type == SHT_SYMTAB_SHNDX &&
257 		    shdr.sh_link == tgt->dt_symidx) {
258 			Elf_Data	*data;
259 
260 			if ((data = elf_getdata(scn, NULL)) != NULL) {
261 				symshndx = (GElf_Word *)data->d_buf;
262 				symshndx_size = data->d_size /
263 				    sizeof (GElf_Word);
264 				break;
265 			}
266 		}
267 	}
268 
269 	if ((scn = elf_getscn(tgt->dt_elf, tgt->dt_symidx)) == NULL)
270 		die("%s: failed to get section information", tgt->dt_filename);
271 	if (gelf_getshdr(scn, &shdr) == NULL)
272 		die("%s: failed to get section header", tgt->dt_filename);
273 	if (shdr.sh_entsize == 0)
274 		die("%s: symbol table has zero size", tgt->dt_filename);
275 
276 	if ((symdata = elf_getdata(scn, NULL)) == NULL)
277 		die("%s: failed to get symbol table", tgt->dt_filename);
278 
279 	tgt->dt_symcount = symdata->d_size / gelf_fsize(tgt->dt_elf, ELF_T_SYM,
280 	    1, EV_CURRENT);
281 
282 	p_symtab = safe_malloc(tgt->dt_symcount * sizeof (sym_entry_t));
283 
284 	for (i = 0, sym = p_symtab; i < tgt->dt_symcount; i++) {
285 		if (gelf_getsym(symdata, i, &(sym->se_sym)) == NULL) {
286 			warn("%s: gelf_getsym returned NULL for %d",
287 			    tgt->dt_filename, i);
288 			nsym++;
289 			continue;
290 		}
291 
292 		/*
293 		 * We're only interested in data symbols.
294 		 */
295 		if (!IS_DATA_TYPE(GELF_ST_TYPE(sym->se_sym.st_info))) {
296 			nsym++;
297 			continue;
298 		}
299 
300 		if (sym->se_sym.st_shndx == SHN_XINDEX && symshndx != NULL) {
301 			if (i > symshndx_size) {
302 				warn("%s: bad SHNX_XINDEX %d",
303 				    tgt->dt_filename, i);
304 				sym->se_shndx = -1;
305 			} else {
306 				sym->se_shndx = symshndx[i];
307 			}
308 		} else {
309 			sym->se_shndx = sym->se_sym.st_shndx;
310 		}
311 
312 		if ((sym->se_name = elf_strptr(tgt->dt_elf, shdr.sh_link,
313 		    (size_t)sym->se_sym.st_name)) == NULL) {
314 			warn("%s: failed to lookup symbol %d name",
315 			    tgt->dt_filename, i);
316 			nsym++;
317 			continue;
318 		}
319 
320 		/*
321 		 * If we had to map this section, its symbol value
322 		 * also needs to be mapped.
323 		 */
324 		if (tgt->dt_shnmap[sym->se_shndx].dm_mapped)
325 			sym->se_sym.st_value +=
326 			    tgt->dt_shnmap[sym->se_shndx].dm_start;
327 
328 		sym++;
329 	}
330 
331 	tgt->dt_symcount -= nsym;
332 	tgt->dt_symtab = realloc(p_symtab, tgt->dt_symcount *
333 	    sizeof (sym_entry_t));
334 
335 	qsort(tgt->dt_symtab, tgt->dt_symcount, sizeof (sym_entry_t),
336 	    sym_compare);
337 }
338 
339 /*
340  * Assign virtual address ranges for sections that need it
341  */
342 static void
343 create_addrmap(dis_tgt_t *tgt)
344 {
345 	uint64_t addr;
346 	int i;
347 
348 	if (tgt->dt_shnmap == NULL)
349 		return;
350 
351 	/* find the greatest used address */
352 	for (addr = 0, i = 1; i < tgt->dt_shncount; i++)
353 		if (tgt->dt_shnmap[i].dm_start > addr)
354 			addr = tgt->dt_shnmap[i].dm_start +
355 			    tgt->dt_shnmap[i].dm_length;
356 
357 	addr = P2ROUNDUP(addr, 0x1000);
358 
359 	/*
360 	 * Assign section a starting address beyond the largest mapped section
361 	 * if no address was given.
362 	 */
363 	for (i = 1; i < tgt->dt_shncount; i++) {
364 		if (tgt->dt_shnmap[i].dm_start != 0)
365 			continue;
366 
367 		tgt->dt_shnmap[i].dm_start = addr;
368 		tgt->dt_shnmap[i].dm_mapped = B_TRUE;
369 		addr = P2ROUNDUP(addr + tgt->dt_shnmap[i].dm_length, 0x1000);
370 	}
371 }
372 
373 /*
374  * Create a target backed by an ELF file.
375  */
376 dis_tgt_t *
377 dis_tgt_create(const char *file)
378 {
379 	dis_tgt_t *tgt, *current;
380 	int idx;
381 	Elf *elf;
382 	GElf_Ehdr ehdr;
383 	Elf_Arhdr *arhdr = NULL;
384 	int cmd;
385 
386 	if (elf_version(EV_CURRENT) == EV_NONE)
387 		die("libelf(3ELF) out of date");
388 
389 	tgt = safe_malloc(sizeof (dis_tgt_t));
390 
391 	if ((tgt->dt_fd = open(file, O_RDONLY)) < 0) {
392 		warn("%s: failed opening file, reason: %s", file,
393 		    strerror(errno));
394 		free(tgt);
395 		return (NULL);
396 	}
397 
398 	if ((tgt->dt_elf_root =
399 	    elf_begin(tgt->dt_fd, ELF_C_READ, NULL)) == NULL) {
400 		warn("%s: invalid or corrupt ELF file", file);
401 		dis_tgt_destroy(tgt);
402 		return (NULL);
403 	}
404 
405 	current = tgt;
406 	cmd = ELF_C_READ;
407 	while ((elf = elf_begin(tgt->dt_fd, cmd, tgt->dt_elf_root)) != NULL) {
408 
409 		if (elf_kind(tgt->dt_elf_root) == ELF_K_AR &&
410 		    (arhdr = elf_getarhdr(elf)) == NULL) {
411 			warn("%s: malformed archive", file);
412 			dis_tgt_destroy(tgt);
413 			return (NULL);
414 		}
415 
416 		/*
417 		 * Make sure that this Elf file is sane
418 		 */
419 		if (gelf_getehdr(elf, &ehdr) == NULL) {
420 			if (arhdr != NULL) {
421 				/*
422 				 * For archives, we drive on in the face of bad
423 				 * members.  The "/" and "//" members are
424 				 * special, and should be silently ignored.
425 				 */
426 				if (strcmp(arhdr->ar_name, "/") != 0 &&
427 				    strcmp(arhdr->ar_name, "//") != 0)
428 					warn("%s[%s]: invalid file type",
429 					    file, arhdr->ar_name);
430 				cmd = elf_next(elf);
431 				(void) elf_end(elf);
432 				continue;
433 			}
434 
435 			warn("%s: invalid file type", file);
436 			dis_tgt_destroy(tgt);
437 			return (NULL);
438 		}
439 
440 		/*
441 		 * If we're seeing a new Elf object, then we have an
442 		 * archive. In this case, we create a new target, and chain it
443 		 * off the master target.  We can later iterate over these
444 		 * targets using dis_tgt_next().
445 		 */
446 		if (current->dt_elf != NULL) {
447 			dis_tgt_t *next = safe_malloc(sizeof (dis_tgt_t));
448 			next->dt_elf_root = tgt->dt_elf_root;
449 			next->dt_fd = -1;
450 			current->dt_next = next;
451 			current = next;
452 		}
453 		current->dt_elf = elf;
454 		current->dt_arhdr = arhdr;
455 
456 		if (elf_getshdrstrndx(elf, &current->dt_shstrndx) == -1) {
457 			warn("%s: failed to get section string table for "
458 			    "file", file);
459 			dis_tgt_destroy(tgt);
460 			return (NULL);
461 		}
462 
463 		current->dt_shnmap = safe_malloc(sizeof (dis_shnmap_t) *
464 		    ehdr.e_shnum);
465 		current->dt_shncount = ehdr.e_shnum;
466 
467 		idx = 0;
468 		dis_tgt_section_iter(current, tgt_scn_init, &idx);
469 
470 		create_addrmap(current);
471 		if (current->dt_symidx != 0)
472 			construct_symtab(current);
473 
474 		current->dt_filename = file;
475 
476 		cmd = elf_next(elf);
477 	}
478 
479 	/*
480 	 * Final sanity check.  If we had an archive with no members, then bail
481 	 * out with a nice message.
482 	 */
483 	if (tgt->dt_elf == NULL) {
484 		warn("%s: empty archive\n", file);
485 		dis_tgt_destroy(tgt);
486 		return (NULL);
487 	}
488 
489 	return (tgt);
490 }
491 
492 /*
493  * Return the filename associated with the target.
494  */
495 const char *
496 dis_tgt_name(dis_tgt_t *tgt)
497 {
498 	return (tgt->dt_filename);
499 }
500 
501 /*
502  * Return the archive member name, if any.
503  */
504 const char *
505 dis_tgt_member(dis_tgt_t *tgt)
506 {
507 	if (tgt->dt_arhdr)
508 		return (tgt->dt_arhdr->ar_name);
509 	else
510 		return (NULL);
511 }
512 
513 /*
514  * Return the Elf_Ehdr associated with this target.  Needed to determine which
515  * disassembler to use.
516  */
517 void
518 dis_tgt_ehdr(dis_tgt_t *tgt, GElf_Ehdr *ehdr)
519 {
520 	(void) gelf_getehdr(tgt->dt_elf, ehdr);
521 }
522 
523 /*
524  * Return the next target in the list, if this is an archive.
525  */
526 dis_tgt_t *
527 dis_tgt_next(dis_tgt_t *tgt)
528 {
529 	return (tgt->dt_next);
530 }
531 
532 /*
533  * Destroy a target and free up any associated memory.
534  */
535 void
536 dis_tgt_destroy(dis_tgt_t *tgt)
537 {
538 	dis_tgt_t *current, *next;
539 
540 	current = tgt->dt_next;
541 	while (current != NULL) {
542 		next = current->dt_next;
543 		if (current->dt_elf)
544 			(void) elf_end(current->dt_elf);
545 		if (current->dt_symtab)
546 			free(current->dt_symtab);
547 		free(current);
548 		current = next;
549 	}
550 
551 	if (tgt->dt_elf)
552 		(void) elf_end(tgt->dt_elf);
553 	if (tgt->dt_elf_root)
554 		(void) elf_end(tgt->dt_elf_root);
555 
556 	if (tgt->dt_symtab)
557 		free(tgt->dt_symtab);
558 
559 	free(tgt);
560 }
561 
562 /*
563  * Given an address, return the section it is in and set the offset within
564  * the section.
565  */
566 const char *
567 dis_find_section(dis_tgt_t *tgt, uint64_t addr, off_t *offset)
568 {
569 	int i;
570 
571 	for (i = 1; i < tgt->dt_shncount; i++) {
572 		if ((addr >= tgt->dt_shnmap[i].dm_start) &&
573 		    (addr < tgt->dt_shnmap[i].dm_start +
574 		    tgt->dt_shnmap[i].dm_length)) {
575 			*offset = addr - tgt->dt_shnmap[i].dm_start;
576 			return (tgt->dt_shnmap[i].dm_name);
577 		}
578 	}
579 
580 	*offset = 0;
581 	return (NULL);
582 }
583 
584 /*
585  * Given an address, returns the name of the corresponding symbol, as well as
586  * the offset within that symbol.  If no matching symbol is found, then NULL is
587  * returned.
588  *
589  * If 'cache_result' is specified, then we keep track of the resulting symbol.
590  * This cached result is consulted first on subsequent lookups in order to avoid
591  * unecessary lookups.  This flag should be used for resolving the current PC,
592  * as the majority of addresses stay within the current function.
593  */
594 const char *
595 dis_tgt_lookup(dis_tgt_t *tgt, uint64_t addr, off_t *offset, int cache_result,
596     size_t *size, int *isfunc)
597 {
598 	int lo, hi, mid;
599 	sym_entry_t *sym, *osym, *match;
600 	int found;
601 
602 	if (tgt->dt_symcache != NULL &&
603 	    addr >= tgt->dt_symcache->se_sym.st_value &&
604 	    addr < tgt->dt_symcache->se_sym.st_value +
605 	    tgt->dt_symcache->se_sym.st_size) {
606 		*offset = addr - tgt->dt_symcache->se_sym.st_value;
607 		*size = tgt->dt_symcache->se_sym.st_size;
608 		return (tgt->dt_symcache->se_name);
609 	}
610 
611 	lo = 0;
612 	hi = (tgt->dt_symcount - 1);
613 	found = 0;
614 	match = osym = NULL;
615 	while (lo <= hi) {
616 		mid = (lo + hi) / 2;
617 
618 		sym = &tgt->dt_symtab[mid];
619 
620 		if (addr >= sym->se_sym.st_value &&
621 		    addr < sym->se_sym.st_value + sym->se_sym.st_size &&
622 		    (!found || sym->se_sym.st_value > osym->se_sym.st_value)) {
623 			osym = sym;
624 			found = 1;
625 		} else if (addr == sym->se_sym.st_value) {
626 			/*
627 			 * Particularly for .plt objects, it's possible to have
628 			 * a zero sized object.  We want to return this, but we
629 			 * want it to be a last resort.
630 			 */
631 			match = sym;
632 		}
633 
634 		if (addr < sym->se_sym.st_value)
635 			hi = mid - 1;
636 		else
637 			lo = mid + 1;
638 	}
639 
640 	if (!found) {
641 		if (match)
642 			osym = match;
643 		else
644 			return (NULL);
645 	}
646 
647 	/*
648 	 * Walk backwards to find the best match.
649 	 */
650 	do {
651 		sym = osym;
652 
653 		if (osym == tgt->dt_symtab)
654 			break;
655 
656 		osym = osym - 1;
657 	} while ((sym->se_sym.st_value == osym->se_sym.st_value) &&
658 	    (addr >= osym->se_sym.st_value) &&
659 	    (addr < osym->se_sym.st_value + osym->se_sym.st_size));
660 
661 	if (cache_result)
662 		tgt->dt_symcache = sym;
663 
664 	*offset = addr - sym->se_sym.st_value;
665 	*size = sym->se_sym.st_size;
666 	if (isfunc)
667 		*isfunc = (GELF_ST_TYPE(sym->se_sym.st_info) == STT_FUNC);
668 
669 	return (sym->se_name);
670 }
671 
672 /*
673  * Given an address, return the starting offset of the next symbol in the file.
674  */
675 off_t
676 dis_tgt_next_symbol(dis_tgt_t *tgt, uint64_t addr)
677 {
678 	sym_entry_t *sym;
679 
680 	for (sym = tgt->dt_symcache;
681 	    sym != tgt->dt_symtab + tgt->dt_symcount;
682 	    sym++) {
683 		if (sym->se_sym.st_value >= addr)
684 			return (sym->se_sym.st_value - addr);
685 	}
686 
687 	return (0);
688 }
689 
690 /*
691  * Iterate over all sections in the target, executing the given callback for
692  * each.
693  */
694 void
695 dis_tgt_section_iter(dis_tgt_t *tgt, section_iter_f func, void *data)
696 {
697 	dis_scn_t sdata;
698 	Elf_Scn *scn;
699 	int idx;
700 
701 	for (scn = elf_nextscn(tgt->dt_elf, NULL), idx = 1; scn != NULL;
702 	    scn = elf_nextscn(tgt->dt_elf, scn), idx++) {
703 
704 		if (gelf_getshdr(scn, &sdata.ds_shdr) == NULL) {
705 			warn("%s: failed to get section %d header",
706 			    tgt->dt_filename, idx);
707 			continue;
708 		}
709 
710 		if ((sdata.ds_name = elf_strptr(tgt->dt_elf, tgt->dt_shstrndx,
711 		    sdata.ds_shdr.sh_name)) == NULL) {
712 			warn("%s: failed to get section %d name",
713 			    tgt->dt_filename, idx);
714 			continue;
715 		}
716 
717 		if ((sdata.ds_data = elf_getdata(scn, NULL)) == NULL) {
718 			warn("%s: failed to get data for section '%s'",
719 			    tgt->dt_filename, sdata.ds_name);
720 			continue;
721 		}
722 
723 		/*
724 		 * dis_tgt_section_iter is also used before the section map
725 		 * is initialized, so only check when we need to.  If the
726 		 * section map is uninitialized, it will return 0 and have
727 		 * no net effect.
728 		 */
729 		if (sdata.ds_shdr.sh_addr == 0)
730 			sdata.ds_shdr.sh_addr = tgt->dt_shnmap[idx].dm_start;
731 
732 		func(tgt, &sdata, data);
733 	}
734 }
735 
736 /*
737  * Return 1 if the given section contains text, 0 otherwise.
738  */
739 int
740 dis_section_istext(dis_scn_t *scn)
741 {
742 	return ((scn->ds_shdr.sh_type == SHT_PROGBITS) &&
743 	    (scn->ds_shdr.sh_flags == (SHF_ALLOC | SHF_EXECINSTR)));
744 }
745 
746 /*
747  * Return a pointer to the section data.
748  */
749 void *
750 dis_section_data(dis_scn_t *scn)
751 {
752 	return (scn->ds_data->d_buf);
753 }
754 
755 /*
756  * Return the size of the section data.
757  */
758 size_t
759 dis_section_size(dis_scn_t *scn)
760 {
761 	return (scn->ds_data->d_size);
762 }
763 
764 /*
765  * Return the address for the given section.
766  */
767 uint64_t
768 dis_section_addr(dis_scn_t *scn)
769 {
770 	return (scn->ds_shdr.sh_addr);
771 }
772 
773 /*
774  * Return the name of the current section.
775  */
776 const char *
777 dis_section_name(dis_scn_t *scn)
778 {
779 	return (scn->ds_name);
780 }
781 
782 /*
783  * Create an allocated copy of the given section
784  */
785 dis_scn_t *
786 dis_section_copy(dis_scn_t *scn)
787 {
788 	dis_scn_t *new;
789 
790 	new = safe_malloc(sizeof (dis_scn_t));
791 	(void) memcpy(new, scn, sizeof (dis_scn_t));
792 
793 	return (new);
794 }
795 
796 /*
797  * Free section memory
798  */
799 void
800 dis_section_free(dis_scn_t *scn)
801 {
802 	free(scn);
803 }
804 
805 /*
806  * Iterate over all functions in the target, executing the given callback for
807  * each one.
808  */
809 void
810 dis_tgt_function_iter(dis_tgt_t *tgt, function_iter_f func, void *data)
811 {
812 	int i;
813 	sym_entry_t *sym;
814 	dis_func_t df;
815 	Elf_Scn *scn;
816 	GElf_Shdr	shdr;
817 
818 	for (i = 0, sym = tgt->dt_symtab; i < tgt->dt_symcount; i++, sym++) {
819 
820 		/* ignore non-functions */
821 		if ((GELF_ST_TYPE(sym->se_sym.st_info) != STT_FUNC) ||
822 		    (sym->se_name == NULL) ||
823 		    (sym->se_sym.st_size == 0) ||
824 		    (sym->se_shndx >= SHN_LORESERVE))
825 			continue;
826 
827 		/* get the ELF data associated with this function */
828 		if ((scn = elf_getscn(tgt->dt_elf, sym->se_shndx)) == NULL ||
829 		    gelf_getshdr(scn, &shdr) == NULL ||
830 		    (df.df_data = elf_getdata(scn, NULL)) == NULL ||
831 		    df.df_data->d_size == 0) {
832 			warn("%s: failed to read section %d",
833 			    tgt->dt_filename, sym->se_shndx);
834 			continue;
835 		}
836 
837 		if (tgt->dt_shnmap[sym->se_shndx].dm_mapped)
838 			shdr.sh_addr = tgt->dt_shnmap[sym->se_shndx].dm_start;
839 
840 		/*
841 		 * Verify that the address lies within the section that we think
842 		 * it does.
843 		 */
844 		if (sym->se_sym.st_value < shdr.sh_addr ||
845 		    (sym->se_sym.st_value + sym->se_sym.st_size) >
846 		    (shdr.sh_addr + shdr.sh_size)) {
847 			warn("%s: bad section %d for address %p",
848 			    tgt->dt_filename, sym->se_sym.st_shndx,
849 			    sym->se_sym.st_value);
850 			continue;
851 		}
852 
853 		df.df_sym = sym;
854 		df.df_offset = sym->se_sym.st_value - shdr.sh_addr;
855 
856 		func(tgt, &df, data);
857 	}
858 }
859 
860 /*
861  * Return the data associated with a given function.
862  */
863 void *
864 dis_function_data(dis_func_t *func)
865 {
866 	return ((char *)func->df_data->d_buf + func->df_offset);
867 }
868 
869 /*
870  * Return the size of a function.
871  */
872 size_t
873 dis_function_size(dis_func_t *func)
874 {
875 	return (func->df_sym->se_sym.st_size);
876 }
877 
878 /*
879  * Return the address of a function.
880  */
881 uint64_t
882 dis_function_addr(dis_func_t *func)
883 {
884 	return (func->df_sym->se_sym.st_value);
885 }
886 
887 /*
888  * Return the name of the function
889  */
890 const char *
891 dis_function_name(dis_func_t *func)
892 {
893 	return (func->df_sym->se_name);
894 }
895 
896 /*
897  * Return a copy of a function.
898  */
899 dis_func_t *
900 dis_function_copy(dis_func_t *func)
901 {
902 	dis_func_t *new;
903 
904 	new = safe_malloc(sizeof (dis_func_t));
905 	(void) memcpy(new, func, sizeof (dis_func_t));
906 
907 	return (new);
908 }
909 
910 /*
911  * Free function memory
912  */
913 void
914 dis_function_free(dis_func_t *func)
915 {
916 	free(func);
917 }
918