Psymtab.c revision 9900:1b86d65a4f9e
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 2009 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#include <assert.h>
28#include <stdio.h>
29#include <stdlib.h>
30#include <stddef.h>
31#include <unistd.h>
32#include <ctype.h>
33#include <fcntl.h>
34#include <string.h>
35#include <strings.h>
36#include <memory.h>
37#include <errno.h>
38#include <dirent.h>
39#include <signal.h>
40#include <limits.h>
41#include <libgen.h>
42#include <sys/types.h>
43#include <sys/stat.h>
44#include <sys/systeminfo.h>
45#include <sys/sysmacros.h>
46
47#include "libproc.h"
48#include "Pcontrol.h"
49#include "Putil.h"
50#include "Psymtab_machelf.h"
51
52static file_info_t *build_map_symtab(struct ps_prochandle *, map_info_t *);
53static map_info_t *exec_map(struct ps_prochandle *);
54static map_info_t *object_to_map(struct ps_prochandle *, Lmid_t, const char *);
55static map_info_t *object_name_to_map(struct ps_prochandle *,
56	Lmid_t, const char *);
57static GElf_Sym *sym_by_name(sym_tbl_t *, const char *, GElf_Sym *, uint_t *);
58static int read_ehdr32(struct ps_prochandle *, Elf32_Ehdr *, uint_t *,
59    uintptr_t);
60#ifdef _LP64
61static int read_ehdr64(struct ps_prochandle *, Elf64_Ehdr *, uint_t *,
62    uintptr_t);
63#endif
64
65#define	DATA_TYPES	\
66	((1 << STT_OBJECT) | (1 << STT_FUNC) | \
67	(1 << STT_COMMON) | (1 << STT_TLS))
68#define	IS_DATA_TYPE(tp)	(((1 << (tp)) & DATA_TYPES) != 0)
69
70#define	MA_RWX	(MA_READ | MA_WRITE | MA_EXEC)
71
72typedef enum {
73	PRO_NATURAL,
74	PRO_BYADDR,
75	PRO_BYNAME
76} pr_order_t;
77
78static int
79addr_cmp(const void *aa, const void *bb)
80{
81	uintptr_t a = *((uintptr_t *)aa);
82	uintptr_t b = *((uintptr_t *)bb);
83
84	if (a > b)
85		return (1);
86	if (a < b)
87		return (-1);
88	return (0);
89}
90
91/*
92 * This function creates a list of addresses for a load object's sections.
93 * The list is in ascending address order and alternates start address
94 * then end address for each section we're interested in. The function
95 * returns a pointer to the list, which must be freed by the caller.
96 */
97static uintptr_t *
98get_saddrs(struct ps_prochandle *P, uintptr_t ehdr_start, uint_t *n)
99{
100	uintptr_t a, addr, *addrs, last = 0;
101	uint_t i, naddrs = 0, unordered = 0;
102
103	if (P->status.pr_dmodel == PR_MODEL_ILP32) {
104		Elf32_Ehdr ehdr;
105		Elf32_Phdr phdr;
106		uint_t phnum;
107
108		if (read_ehdr32(P, &ehdr, &phnum, ehdr_start) != 0)
109			return (NULL);
110
111		addrs = malloc(sizeof (uintptr_t) * phnum * 2);
112		a = ehdr_start + ehdr.e_phoff;
113		for (i = 0; i < phnum; i++, a += ehdr.e_phentsize) {
114			if (Pread(P, &phdr, sizeof (phdr), a) !=
115			    sizeof (phdr)) {
116				free(addrs);
117				return (NULL);
118			}
119			if (phdr.p_type != PT_LOAD || phdr.p_memsz == 0)
120				continue;
121
122			addr = phdr.p_vaddr;
123			if (ehdr.e_type == ET_DYN)
124				addr += ehdr_start;
125			if (last > addr)
126				unordered = 1;
127			addrs[naddrs++] = addr;
128			addrs[naddrs++] = last = addr + phdr.p_memsz - 1;
129		}
130#ifdef _LP64
131	} else {
132		Elf64_Ehdr ehdr;
133		Elf64_Phdr phdr;
134		uint_t phnum;
135
136		if (read_ehdr64(P, &ehdr, &phnum, ehdr_start) != 0)
137			return (NULL);
138
139		addrs = malloc(sizeof (uintptr_t) * phnum * 2);
140		a = ehdr_start + ehdr.e_phoff;
141		for (i = 0; i < phnum; i++, a += ehdr.e_phentsize) {
142			if (Pread(P, &phdr, sizeof (phdr), a) !=
143			    sizeof (phdr)) {
144				free(addrs);
145				return (NULL);
146			}
147			if (phdr.p_type != PT_LOAD || phdr.p_memsz == 0)
148				continue;
149
150			addr = phdr.p_vaddr;
151			if (ehdr.e_type == ET_DYN)
152				addr += ehdr_start;
153			if (last > addr)
154				unordered = 1;
155			addrs[naddrs++] = addr;
156			addrs[naddrs++] = last = addr + phdr.p_memsz - 1;
157		}
158#endif
159	}
160
161	if (unordered)
162		qsort(addrs, naddrs, sizeof (uintptr_t), addr_cmp);
163
164	*n = naddrs;
165	return (addrs);
166}
167
168/*
169 * Allocation function for a new file_info_t
170 */
171file_info_t *
172file_info_new(struct ps_prochandle *P, map_info_t *mptr)
173{
174	file_info_t *fptr;
175	map_info_t *mp;
176	uintptr_t mstart, mend, sstart, send;
177	uint_t i;
178
179	if ((fptr = calloc(1, sizeof (file_info_t))) == NULL)
180		return (NULL);
181
182	list_link(fptr, &P->file_head);
183	(void) strcpy(fptr->file_pname, mptr->map_pmap.pr_mapname);
184	mptr->map_file = fptr;
185	fptr->file_ref = 1;
186	fptr->file_fd = -1;
187	P->num_files++;
188
189	/*
190	 * To figure out which map_info_t instances correspond to the mappings
191	 * for this load object we try to obtain the start and end address
192	 * for each section of our in-memory ELF image. If successful, we
193	 * walk down the list of addresses and the list of map_info_t
194	 * instances in lock step to correctly find the mappings that
195	 * correspond to this load object.
196	 */
197	if ((fptr->file_saddrs = get_saddrs(P, mptr->map_pmap.pr_vaddr,
198	    &fptr->file_nsaddrs)) == NULL)
199		return (fptr);
200
201	mp = P->mappings;
202	i = 0;
203	while (mp < P->mappings + P->map_count && i < fptr->file_nsaddrs) {
204
205		/* Calculate the start and end of the mapping and section */
206		mstart = mp->map_pmap.pr_vaddr;
207		mend = mp->map_pmap.pr_vaddr + mp->map_pmap.pr_size;
208		sstart = fptr->file_saddrs[i];
209		send = fptr->file_saddrs[i + 1];
210
211		if (mend <= sstart) {
212			/* This mapping is below the current section */
213			mp++;
214		} else if (mstart >= send) {
215			/* This mapping is above the current section */
216			i += 2;
217		} else {
218			/* This mapping overlaps the current section */
219			if (mp->map_file == NULL) {
220				dprintf("file_info_new: associating "
221				    "segment at %p\n",
222				    (void *)mp->map_pmap.pr_vaddr);
223				mp->map_file = fptr;
224				fptr->file_ref++;
225			} else {
226				dprintf("file_info_new: segment at %p "
227				    "already associated with %s\n",
228				    (void *)mp->map_pmap.pr_vaddr,
229				    (mp == mptr ? "this file" :
230				    mp->map_file->file_pname));
231			}
232			mp++;
233		}
234	}
235
236	return (fptr);
237}
238
239/*
240 * Deallocation function for a file_info_t
241 */
242static void
243file_info_free(struct ps_prochandle *P, file_info_t *fptr)
244{
245	if (--fptr->file_ref == 0) {
246		list_unlink(fptr);
247		if (fptr->file_symtab.sym_elf) {
248			(void) elf_end(fptr->file_symtab.sym_elf);
249			free(fptr->file_symtab.sym_elfmem);
250		}
251		if (fptr->file_symtab.sym_byname)
252			free(fptr->file_symtab.sym_byname);
253		if (fptr->file_symtab.sym_byaddr)
254			free(fptr->file_symtab.sym_byaddr);
255
256		if (fptr->file_dynsym.sym_elf) {
257			(void) elf_end(fptr->file_dynsym.sym_elf);
258			free(fptr->file_dynsym.sym_elfmem);
259		}
260		if (fptr->file_dynsym.sym_byname)
261			free(fptr->file_dynsym.sym_byname);
262		if (fptr->file_dynsym.sym_byaddr)
263			free(fptr->file_dynsym.sym_byaddr);
264
265		if (fptr->file_lo)
266			free(fptr->file_lo);
267		if (fptr->file_lname)
268			free(fptr->file_lname);
269		if (fptr->file_rname)
270			free(fptr->file_rname);
271		if (fptr->file_elf)
272			(void) elf_end(fptr->file_elf);
273		if (fptr->file_elfmem != NULL)
274			free(fptr->file_elfmem);
275		if (fptr->file_fd >= 0)
276			(void) close(fptr->file_fd);
277		if (fptr->file_ctfp) {
278			ctf_close(fptr->file_ctfp);
279			free(fptr->file_ctf_buf);
280		}
281		if (fptr->file_saddrs)
282			free(fptr->file_saddrs);
283		free(fptr);
284		P->num_files--;
285	}
286}
287
288/*
289 * Deallocation function for a map_info_t
290 */
291static void
292map_info_free(struct ps_prochandle *P, map_info_t *mptr)
293{
294	file_info_t *fptr;
295
296	if ((fptr = mptr->map_file) != NULL) {
297		if (fptr->file_map == mptr)
298			fptr->file_map = NULL;
299		file_info_free(P, fptr);
300	}
301	if (P->execname && mptr == P->map_exec) {
302		free(P->execname);
303		P->execname = NULL;
304	}
305	if (P->auxv && (mptr == P->map_exec || mptr == P->map_ldso)) {
306		free(P->auxv);
307		P->auxv = NULL;
308		P->nauxv = 0;
309	}
310	if (mptr == P->map_exec)
311		P->map_exec = NULL;
312	if (mptr == P->map_ldso)
313		P->map_ldso = NULL;
314}
315
316/*
317 * Call-back function for librtld_db to iterate through all of its shared
318 * libraries.  We use this to get the load object names for the mappings.
319 */
320static int
321map_iter(const rd_loadobj_t *lop, void *cd)
322{
323	char buf[PATH_MAX];
324	struct ps_prochandle *P = cd;
325	map_info_t *mptr;
326	file_info_t *fptr;
327
328	dprintf("encountered rd object at %p\n", (void *)lop->rl_base);
329
330	if ((mptr = Paddr2mptr(P, lop->rl_base)) == NULL) {
331		dprintf("map_iter: base address doesn't match any mapping\n");
332		return (1); /* Base address does not match any mapping */
333	}
334
335	if ((fptr = mptr->map_file) == NULL &&
336	    (fptr = file_info_new(P, mptr)) == NULL) {
337		dprintf("map_iter: failed to allocate a new file_info_t\n");
338		return (1); /* Failed to allocate a new file_info_t */
339	}
340
341	if ((fptr->file_lo == NULL) &&
342	    (fptr->file_lo = malloc(sizeof (rd_loadobj_t))) == NULL) {
343		dprintf("map_iter: failed to allocate rd_loadobj_t\n");
344		file_info_free(P, fptr);
345		return (1); /* Failed to allocate rd_loadobj_t */
346	}
347
348	fptr->file_map = mptr;
349	*fptr->file_lo = *lop;
350
351	fptr->file_lo->rl_plt_base = fptr->file_plt_base;
352	fptr->file_lo->rl_plt_size = fptr->file_plt_size;
353
354	if (fptr->file_lname) {
355		free(fptr->file_lname);
356		fptr->file_lname = NULL;
357		fptr->file_lbase = NULL;
358	}
359	if (fptr->file_rname) {
360		free(fptr->file_rname);
361		fptr->file_rname = NULL;
362		fptr->file_rbase = NULL;
363	}
364
365	if (Pread_string(P, buf, sizeof (buf), lop->rl_nameaddr) > 0) {
366		if ((fptr->file_lname = strdup(buf)) != NULL)
367			fptr->file_lbase = basename(fptr->file_lname);
368	} else {
369		dprintf("map_iter: failed to read string at %p\n",
370		    (void *)lop->rl_nameaddr);
371	}
372
373	if ((Pfindmap(P, mptr, buf, sizeof (buf)) != NULL) &&
374	    ((fptr->file_rname = strdup(buf)) != NULL))
375		fptr->file_rbase = basename(fptr->file_rname);
376
377	dprintf("loaded rd object %s lmid %lx\n",
378	    fptr->file_lname ? buf : "<NULL>", lop->rl_lmident);
379	return (1);
380}
381
382static void
383map_set(struct ps_prochandle *P, map_info_t *mptr, const char *lname)
384{
385	file_info_t *fptr;
386	char buf[PATH_MAX];
387
388	if ((fptr = mptr->map_file) == NULL &&
389	    (fptr = file_info_new(P, mptr)) == NULL)
390		return; /* Failed to allocate a new file_info_t */
391
392	fptr->file_map = mptr;
393
394	if ((fptr->file_lo == NULL) &&
395	    (fptr->file_lo = malloc(sizeof (rd_loadobj_t))) == NULL) {
396		file_info_free(P, fptr);
397		return; /* Failed to allocate rd_loadobj_t */
398	}
399
400	(void) memset(fptr->file_lo, 0, sizeof (rd_loadobj_t));
401	fptr->file_lo->rl_base = mptr->map_pmap.pr_vaddr;
402	fptr->file_lo->rl_bend =
403	    mptr->map_pmap.pr_vaddr + mptr->map_pmap.pr_size;
404
405	fptr->file_lo->rl_plt_base = fptr->file_plt_base;
406	fptr->file_lo->rl_plt_size = fptr->file_plt_size;
407
408	if ((fptr->file_lname == NULL) &&
409	    (fptr->file_lname = strdup(lname)) != NULL)
410		fptr->file_lbase = basename(fptr->file_lname);
411
412	if ((Pfindmap(P, mptr, buf, sizeof (buf)) != NULL) &&
413	    ((fptr->file_rname = strdup(buf)) != NULL))
414		fptr->file_rbase = basename(fptr->file_rname);
415}
416
417static void
418load_static_maps(struct ps_prochandle *P)
419{
420	map_info_t *mptr;
421
422	/*
423	 * Construct the map for the a.out.
424	 */
425	if ((mptr = object_name_to_map(P, PR_LMID_EVERY, PR_OBJ_EXEC)) != NULL)
426		map_set(P, mptr, "a.out");
427
428	/*
429	 * If the dynamic linker exists for this process,
430	 * construct the map for it.
431	 */
432	if (Pgetauxval(P, AT_BASE) != -1L &&
433	    (mptr = object_name_to_map(P, PR_LMID_EVERY, PR_OBJ_LDSO)) != NULL)
434		map_set(P, mptr, "ld.so.1");
435}
436
437/*
438 * Go through all the address space mappings, validating or updating
439 * the information already gathered, or gathering new information.
440 *
441 * This function is only called when we suspect that the mappings have changed
442 * because this is the first time we're calling it or because of rtld activity.
443 */
444void
445Pupdate_maps(struct ps_prochandle *P)
446{
447	char mapfile[PATH_MAX];
448	int mapfd;
449	struct stat statb;
450	prmap_t *Pmap = NULL;
451	prmap_t *pmap;
452	ssize_t nmap;
453	int i;
454	uint_t oldmapcount;
455	map_info_t *newmap, *newp;
456	map_info_t *mptr;
457
458	if (P->info_valid || P->state == PS_UNDEAD)
459		return;
460
461	Preadauxvec(P);
462
463	(void) snprintf(mapfile, sizeof (mapfile), "%s/%d/map",
464	    procfs_path, (int)P->pid);
465	if ((mapfd = open(mapfile, O_RDONLY)) < 0 ||
466	    fstat(mapfd, &statb) != 0 ||
467	    statb.st_size < sizeof (prmap_t) ||
468	    (Pmap = malloc(statb.st_size)) == NULL ||
469	    (nmap = pread(mapfd, Pmap, statb.st_size, 0L)) <= 0 ||
470	    (nmap /= sizeof (prmap_t)) == 0) {
471		if (Pmap != NULL)
472			free(Pmap);
473		if (mapfd >= 0)
474			(void) close(mapfd);
475		Preset_maps(P);	/* utter failure; destroy tables */
476		return;
477	}
478	(void) close(mapfd);
479
480	if ((newmap = calloc(1, nmap * sizeof (map_info_t))) == NULL)
481		return;
482
483	/*
484	 * We try to merge any file information we may have for existing
485	 * mappings, to avoid having to rebuild the file info.
486	 */
487	mptr = P->mappings;
488	pmap = Pmap;
489	newp = newmap;
490	oldmapcount = P->map_count;
491	for (i = 0; i < nmap; i++, pmap++, newp++) {
492
493		if (oldmapcount == 0) {
494			/*
495			 * We've exhausted all the old mappings.  Every new
496			 * mapping should be added.
497			 */
498			newp->map_pmap = *pmap;
499
500		} else if (pmap->pr_vaddr == mptr->map_pmap.pr_vaddr &&
501		    pmap->pr_size == mptr->map_pmap.pr_size &&
502		    pmap->pr_offset == mptr->map_pmap.pr_offset &&
503		    (pmap->pr_mflags & ~(MA_BREAK | MA_STACK)) ==
504		    (mptr->map_pmap.pr_mflags & ~(MA_BREAK | MA_STACK)) &&
505		    pmap->pr_pagesize == mptr->map_pmap.pr_pagesize &&
506		    pmap->pr_shmid == mptr->map_pmap.pr_shmid &&
507		    strcmp(pmap->pr_mapname, mptr->map_pmap.pr_mapname) == 0) {
508
509			/*
510			 * This mapping matches exactly.  Copy over the old
511			 * mapping, taking care to get the latest flags.
512			 * Make sure the associated file_info_t is updated
513			 * appropriately.
514			 */
515			*newp = *mptr;
516			if (P->map_exec == mptr)
517				P->map_exec = newp;
518			if (P->map_ldso == mptr)
519				P->map_ldso = newp;
520			newp->map_pmap.pr_mflags = pmap->pr_mflags;
521			if (mptr->map_file != NULL &&
522			    mptr->map_file->file_map == mptr)
523				mptr->map_file->file_map = newp;
524			oldmapcount--;
525			mptr++;
526
527		} else if (pmap->pr_vaddr + pmap->pr_size >
528		    mptr->map_pmap.pr_vaddr) {
529
530			/*
531			 * The old mapping doesn't exist any more, remove it
532			 * from the list.
533			 */
534			map_info_free(P, mptr);
535			oldmapcount--;
536			i--;
537			newp--;
538			pmap--;
539			mptr++;
540
541		} else {
542
543			/*
544			 * This is a new mapping, add it directly.
545			 */
546			newp->map_pmap = *pmap;
547		}
548	}
549
550	/*
551	 * Free any old maps
552	 */
553	while (oldmapcount) {
554		map_info_free(P, mptr);
555		oldmapcount--;
556		mptr++;
557	}
558
559	free(Pmap);
560	if (P->mappings != NULL)
561		free(P->mappings);
562	P->mappings = newmap;
563	P->map_count = P->map_alloc = nmap;
564	P->info_valid = 1;
565
566	/*
567	 * Consult librtld_db to get the load object
568	 * names for all of the shared libraries.
569	 */
570	if (P->rap != NULL)
571		(void) rd_loadobj_iter(P->rap, map_iter, P);
572}
573
574/*
575 * Update all of the mappings and rtld_db as if by Pupdate_maps(), and then
576 * forcibly cache all of the symbol tables associated with all object files.
577 */
578void
579Pupdate_syms(struct ps_prochandle *P)
580{
581	file_info_t *fptr;
582	int i;
583
584	Pupdate_maps(P);
585
586	for (i = 0, fptr = list_next(&P->file_head); i < P->num_files;
587	    i++, fptr = list_next(fptr)) {
588		Pbuild_file_symtab(P, fptr);
589		(void) Pbuild_file_ctf(P, fptr);
590	}
591}
592
593/*
594 * Return the librtld_db agent handle for the victim process.
595 * The handle will become invalid at the next successful exec() and the
596 * client (caller of proc_rd_agent()) must not use it beyond that point.
597 * If the process is already dead, we've already tried our best to
598 * create the agent during core file initialization.
599 */
600rd_agent_t *
601Prd_agent(struct ps_prochandle *P)
602{
603	if (P->rap == NULL && P->state != PS_DEAD && P->state != PS_IDLE) {
604		Pupdate_maps(P);
605		if (P->num_files == 0)
606			load_static_maps(P);
607		rd_log(_libproc_debug);
608		if ((P->rap = rd_new(P)) != NULL)
609			(void) rd_loadobj_iter(P->rap, map_iter, P);
610	}
611	return (P->rap);
612}
613
614/*
615 * Return the prmap_t structure containing 'addr', but only if it
616 * is in the dynamic linker's link map and is the text section.
617 */
618const prmap_t *
619Paddr_to_text_map(struct ps_prochandle *P, uintptr_t addr)
620{
621	map_info_t *mptr;
622
623	if (!P->info_valid)
624		Pupdate_maps(P);
625
626	if ((mptr = Paddr2mptr(P, addr)) != NULL) {
627		file_info_t *fptr = build_map_symtab(P, mptr);
628		const prmap_t *pmp = &mptr->map_pmap;
629
630		/*
631		 * Assume that if rl_data_base is NULL, it means that no
632		 * data section was found for this load object, and that
633		 * a section must be text. Otherwise, a section will be
634		 * text unless it ends above the start of the data
635		 * section.
636		 */
637		if (fptr != NULL && fptr->file_lo != NULL &&
638		    (fptr->file_lo->rl_data_base == NULL ||
639		    pmp->pr_vaddr + pmp->pr_size <
640		    fptr->file_lo->rl_data_base))
641			return (pmp);
642	}
643
644	return (NULL);
645}
646
647/*
648 * Return the prmap_t structure containing 'addr' (no restrictions on
649 * the type of mapping).
650 */
651const prmap_t *
652Paddr_to_map(struct ps_prochandle *P, uintptr_t addr)
653{
654	map_info_t *mptr;
655
656	if (!P->info_valid)
657		Pupdate_maps(P);
658
659	if ((mptr = Paddr2mptr(P, addr)) != NULL)
660		return (&mptr->map_pmap);
661
662	return (NULL);
663}
664
665/*
666 * Convert a full or partial load object name to the prmap_t for its
667 * corresponding primary text mapping.
668 */
669const prmap_t *
670Plmid_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *name)
671{
672	map_info_t *mptr;
673
674	if (name == PR_OBJ_EVERY)
675		return (NULL); /* A reasonable mistake */
676
677	if ((mptr = object_name_to_map(P, lmid, name)) != NULL)
678		return (&mptr->map_pmap);
679
680	return (NULL);
681}
682
683const prmap_t *
684Pname_to_map(struct ps_prochandle *P, const char *name)
685{
686	return (Plmid_to_map(P, PR_LMID_EVERY, name));
687}
688
689const rd_loadobj_t *
690Paddr_to_loadobj(struct ps_prochandle *P, uintptr_t addr)
691{
692	map_info_t *mptr;
693
694	if (!P->info_valid)
695		Pupdate_maps(P);
696
697	if ((mptr = Paddr2mptr(P, addr)) == NULL)
698		return (NULL);
699
700	/*
701	 * By building the symbol table, we implicitly bring the PLT
702	 * information up to date in the load object.
703	 */
704	(void) build_map_symtab(P, mptr);
705
706	return (mptr->map_file->file_lo);
707}
708
709const rd_loadobj_t *
710Plmid_to_loadobj(struct ps_prochandle *P, Lmid_t lmid, const char *name)
711{
712	map_info_t *mptr;
713
714	if (name == PR_OBJ_EVERY)
715		return (NULL);
716
717	if ((mptr = object_name_to_map(P, lmid, name)) == NULL)
718		return (NULL);
719
720	/*
721	 * By building the symbol table, we implicitly bring the PLT
722	 * information up to date in the load object.
723	 */
724	(void) build_map_symtab(P, mptr);
725
726	return (mptr->map_file->file_lo);
727}
728
729const rd_loadobj_t *
730Pname_to_loadobj(struct ps_prochandle *P, const char *name)
731{
732	return (Plmid_to_loadobj(P, PR_LMID_EVERY, name));
733}
734
735ctf_file_t *
736Pbuild_file_ctf(struct ps_prochandle *P, file_info_t *fptr)
737{
738	ctf_sect_t ctdata, symtab, strtab;
739	sym_tbl_t *symp;
740	int err;
741
742	if (fptr->file_ctfp != NULL)
743		return (fptr->file_ctfp);
744
745	Pbuild_file_symtab(P, fptr);
746
747	if (fptr->file_ctf_size == 0)
748		return (NULL);
749
750	symp = fptr->file_ctf_dyn ? &fptr->file_dynsym : &fptr->file_symtab;
751	if (symp->sym_data_pri == NULL)
752		return (NULL);
753
754	/*
755	 * The buffer may alread be allocated if this is a core file that
756	 * contained CTF data for this file.
757	 */
758	if (fptr->file_ctf_buf == NULL) {
759		fptr->file_ctf_buf = malloc(fptr->file_ctf_size);
760		if (fptr->file_ctf_buf == NULL) {
761			dprintf("failed to allocate ctf buffer\n");
762			return (NULL);
763		}
764
765		if (pread(fptr->file_fd, fptr->file_ctf_buf,
766		    fptr->file_ctf_size, fptr->file_ctf_off) !=
767		    fptr->file_ctf_size) {
768			free(fptr->file_ctf_buf);
769			fptr->file_ctf_buf = NULL;
770			dprintf("failed to read ctf data\n");
771			return (NULL);
772		}
773	}
774
775	ctdata.cts_name = ".SUNW_ctf";
776	ctdata.cts_type = SHT_PROGBITS;
777	ctdata.cts_flags = 0;
778	ctdata.cts_data = fptr->file_ctf_buf;
779	ctdata.cts_size = fptr->file_ctf_size;
780	ctdata.cts_entsize = 1;
781	ctdata.cts_offset = 0;
782
783	symtab.cts_name = fptr->file_ctf_dyn ? ".dynsym" : ".symtab";
784	symtab.cts_type = symp->sym_hdr_pri.sh_type;
785	symtab.cts_flags = symp->sym_hdr_pri.sh_flags;
786	symtab.cts_data = symp->sym_data_pri->d_buf;
787	symtab.cts_size = symp->sym_hdr_pri.sh_size;
788	symtab.cts_entsize = symp->sym_hdr_pri.sh_entsize;
789	symtab.cts_offset = symp->sym_hdr_pri.sh_offset;
790
791	strtab.cts_name = fptr->file_ctf_dyn ? ".dynstr" : ".strtab";
792	strtab.cts_type = symp->sym_strhdr.sh_type;
793	strtab.cts_flags = symp->sym_strhdr.sh_flags;
794	strtab.cts_data = symp->sym_strs;
795	strtab.cts_size = symp->sym_strhdr.sh_size;
796	strtab.cts_entsize = symp->sym_strhdr.sh_entsize;
797	strtab.cts_offset = symp->sym_strhdr.sh_offset;
798
799	fptr->file_ctfp = ctf_bufopen(&ctdata, &symtab, &strtab, &err);
800	if (fptr->file_ctfp == NULL) {
801		dprintf("ctf_bufopen() failed, error code %d\n", err);
802		free(fptr->file_ctf_buf);
803		fptr->file_ctf_buf = NULL;
804		return (NULL);
805	}
806
807	dprintf("loaded %lu bytes of CTF data for %s\n",
808	    (ulong_t)fptr->file_ctf_size, fptr->file_pname);
809
810	return (fptr->file_ctfp);
811}
812
813ctf_file_t *
814Paddr_to_ctf(struct ps_prochandle *P, uintptr_t addr)
815{
816	map_info_t *mptr;
817	file_info_t *fptr;
818
819	if (!P->info_valid)
820		Pupdate_maps(P);
821
822	if ((mptr = Paddr2mptr(P, addr)) == NULL ||
823	    (fptr = mptr->map_file) == NULL)
824		return (NULL);
825
826	return (Pbuild_file_ctf(P, fptr));
827}
828
829ctf_file_t *
830Plmid_to_ctf(struct ps_prochandle *P, Lmid_t lmid, const char *name)
831{
832	map_info_t *mptr;
833	file_info_t *fptr;
834
835	if (name == PR_OBJ_EVERY)
836		return (NULL);
837
838	if ((mptr = object_name_to_map(P, lmid, name)) == NULL ||
839	    (fptr = mptr->map_file) == NULL)
840		return (NULL);
841
842	return (Pbuild_file_ctf(P, fptr));
843}
844
845ctf_file_t *
846Pname_to_ctf(struct ps_prochandle *P, const char *name)
847{
848	return (Plmid_to_ctf(P, PR_LMID_EVERY, name));
849}
850
851/*
852 * If we're not a core file, re-read the /proc/<pid>/auxv file and store
853 * its contents in P->auxv.  In the case of a core file, we either
854 * initialized P->auxv in Pcore() from the NT_AUXV, or we don't have an
855 * auxv because the note was missing.
856 */
857void
858Preadauxvec(struct ps_prochandle *P)
859{
860	char auxfile[64];
861	struct stat statb;
862	ssize_t naux;
863	int fd;
864
865	if (P->state == PS_DEAD)
866		return; /* Already read during Pgrab_core() */
867	if (P->state == PS_IDLE)
868		return; /* No aux vec for Pgrab_file() */
869
870	if (P->auxv != NULL) {
871		free(P->auxv);
872		P->auxv = NULL;
873		P->nauxv = 0;
874	}
875
876	(void) snprintf(auxfile, sizeof (auxfile), "%s/%d/auxv",
877	    procfs_path, (int)P->pid);
878	if ((fd = open(auxfile, O_RDONLY)) < 0)
879		return;
880
881	if (fstat(fd, &statb) == 0 &&
882	    statb.st_size >= sizeof (auxv_t) &&
883	    (P->auxv = malloc(statb.st_size + sizeof (auxv_t))) != NULL) {
884		if ((naux = read(fd, P->auxv, statb.st_size)) < 0 ||
885		    (naux /= sizeof (auxv_t)) < 1) {
886			free(P->auxv);
887			P->auxv = NULL;
888		} else {
889			P->auxv[naux].a_type = AT_NULL;
890			P->auxv[naux].a_un.a_val = 0L;
891			P->nauxv = (int)naux;
892		}
893	}
894
895	(void) close(fd);
896}
897
898/*
899 * Return a requested element from the process's aux vector.
900 * Return -1 on failure (this is adequate for our purposes).
901 */
902long
903Pgetauxval(struct ps_prochandle *P, int type)
904{
905	auxv_t *auxv;
906
907	if (P->auxv == NULL)
908		Preadauxvec(P);
909
910	if (P->auxv == NULL)
911		return (-1);
912
913	for (auxv = P->auxv; auxv->a_type != AT_NULL; auxv++) {
914		if (auxv->a_type == type)
915			return (auxv->a_un.a_val);
916	}
917
918	return (-1);
919}
920
921/*
922 * Return a pointer to our internal copy of the process's aux vector.
923 * The caller should not hold on to this pointer across any libproc calls.
924 */
925const auxv_t *
926Pgetauxvec(struct ps_prochandle *P)
927{
928	static const auxv_t empty = { AT_NULL, 0L };
929
930	if (P->auxv == NULL)
931		Preadauxvec(P);
932
933	if (P->auxv == NULL)
934		return (&empty);
935
936	return (P->auxv);
937}
938
939/*
940 * Return 1 if the given mapping corresponds to the given file_info_t's
941 * load object; return 0 otherwise.
942 */
943static int
944is_mapping_in_file(struct ps_prochandle *P, map_info_t *mptr, file_info_t *fptr)
945{
946	prmap_t *pmap = &mptr->map_pmap;
947	rd_loadobj_t *lop = fptr->file_lo;
948	uint_t i;
949	uintptr_t mstart, mend, sstart, send;
950
951	/*
952	 * We can get for free the start address of the text and data
953	 * sections of the load object. Start by seeing if the mapping
954	 * encloses either of these.
955	 */
956	if ((pmap->pr_vaddr <= lop->rl_base &&
957	    lop->rl_base < pmap->pr_vaddr + pmap->pr_size) ||
958	    (pmap->pr_vaddr <= lop->rl_data_base &&
959	    lop->rl_data_base < pmap->pr_vaddr + pmap->pr_size))
960		return (1);
961
962	/*
963	 * It's still possible that this mapping correponds to the load
964	 * object. Consider the example of a mapping whose start and end
965	 * addresses correspond to those of the load object's text section.
966	 * If the mapping splits, e.g. as a result of a segment demotion,
967	 * then although both mappings are still backed by the same section,
968	 * only one will be seen to enclose that section's start address.
969	 * Thus, to be rigorous, we ask not whether this mapping encloses
970	 * the start of a section, but whether there exists a section that
971	 * overlaps this mapping.
972	 *
973	 * If we don't already have the section addresses, and we successfully
974	 * get them, then we cache them in case we come here again.
975	 */
976	if (fptr->file_saddrs == NULL &&
977	    (fptr->file_saddrs = get_saddrs(P,
978	    fptr->file_map->map_pmap.pr_vaddr, &fptr->file_nsaddrs)) == NULL)
979		return (0);
980
981	mstart = mptr->map_pmap.pr_vaddr;
982	mend = mptr->map_pmap.pr_vaddr + mptr->map_pmap.pr_size;
983	for (i = 0; i < fptr->file_nsaddrs; i += 2) {
984		/* Does this section overlap the mapping? */
985		sstart = fptr->file_saddrs[i];
986		send = fptr->file_saddrs[i + 1];
987		if (!(mend <= sstart || mstart >= send))
988			return (1);
989	}
990
991	return (0);
992}
993
994/*
995 * Find or build the symbol table for the given mapping.
996 */
997static file_info_t *
998build_map_symtab(struct ps_prochandle *P, map_info_t *mptr)
999{
1000	prmap_t *pmap = &mptr->map_pmap;
1001	file_info_t *fptr;
1002	uint_t i;
1003
1004	if ((fptr = mptr->map_file) != NULL) {
1005		Pbuild_file_symtab(P, fptr);
1006		return (fptr);
1007	}
1008
1009	if (pmap->pr_mapname[0] == '\0')
1010		return (NULL);
1011
1012	/*
1013	 * Attempt to find a matching file.
1014	 * (A file can be mapped at several different addresses.)
1015	 */
1016	for (i = 0, fptr = list_next(&P->file_head); i < P->num_files;
1017	    i++, fptr = list_next(fptr)) {
1018		if (strcmp(fptr->file_pname, pmap->pr_mapname) == 0 &&
1019		    fptr->file_lo && is_mapping_in_file(P, mptr, fptr)) {
1020			mptr->map_file = fptr;
1021			fptr->file_ref++;
1022			Pbuild_file_symtab(P, fptr);
1023			return (fptr);
1024		}
1025	}
1026
1027	/*
1028	 * If we need to create a new file_info structure, iterate
1029	 * through the load objects in order to attempt to connect
1030	 * this new file with its primary text mapping.  We again
1031	 * need to handle ld.so as a special case because we need
1032	 * to be able to bootstrap librtld_db.
1033	 */
1034	if ((fptr = file_info_new(P, mptr)) == NULL)
1035		return (NULL);
1036
1037	if (P->map_ldso != mptr) {
1038		if (P->rap != NULL)
1039			(void) rd_loadobj_iter(P->rap, map_iter, P);
1040		else
1041			(void) Prd_agent(P);
1042	} else {
1043		fptr->file_map = mptr;
1044	}
1045
1046	/*
1047	 * If librtld_db wasn't able to help us connect the file to a primary
1048	 * text mapping, set file_map to the current mapping because we require
1049	 * fptr->file_map to be set in Pbuild_file_symtab.  librtld_db may be
1050	 * unaware of what's going on in the rare case that a legitimate ELF
1051	 * file has been mmap(2)ed into the process address space *without*
1052	 * the use of dlopen(3x).
1053	 */
1054	if (fptr->file_map == NULL)
1055		fptr->file_map = mptr;
1056
1057	Pbuild_file_symtab(P, fptr);
1058
1059	return (fptr);
1060}
1061
1062static int
1063read_ehdr32(struct ps_prochandle *P, Elf32_Ehdr *ehdr, uint_t *phnum,
1064    uintptr_t addr)
1065{
1066	if (Pread(P, ehdr, sizeof (*ehdr), addr) != sizeof (*ehdr))
1067		return (-1);
1068
1069	if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1070	    ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1071	    ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1072	    ehdr->e_ident[EI_MAG3] != ELFMAG3 ||
1073	    ehdr->e_ident[EI_CLASS] != ELFCLASS32 ||
1074#ifdef _BIG_ENDIAN
1075	    ehdr->e_ident[EI_DATA] != ELFDATA2MSB ||
1076#else
1077	    ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
1078#endif
1079	    ehdr->e_ident[EI_VERSION] != EV_CURRENT)
1080		return (-1);
1081
1082	if ((*phnum = ehdr->e_phnum) == PN_XNUM) {
1083		Elf32_Shdr shdr0;
1084
1085		if (ehdr->e_shoff == 0 || ehdr->e_shentsize < sizeof (shdr0) ||
1086		    Pread(P, &shdr0, sizeof (shdr0), addr + ehdr->e_shoff) !=
1087		    sizeof (shdr0))
1088			return (-1);
1089
1090		if (shdr0.sh_info != 0)
1091			*phnum = shdr0.sh_info;
1092	}
1093
1094	return (0);
1095}
1096
1097static int
1098read_dynamic_phdr32(struct ps_prochandle *P, const Elf32_Ehdr *ehdr,
1099    uint_t phnum, Elf32_Phdr *phdr, uintptr_t addr)
1100{
1101	uint_t i;
1102
1103	for (i = 0; i < phnum; i++) {
1104		uintptr_t a = addr + ehdr->e_phoff + i * ehdr->e_phentsize;
1105		if (Pread(P, phdr, sizeof (*phdr), a) != sizeof (*phdr))
1106			return (-1);
1107
1108		if (phdr->p_type == PT_DYNAMIC)
1109			return (0);
1110	}
1111
1112	return (-1);
1113}
1114
1115#ifdef _LP64
1116static int
1117read_ehdr64(struct ps_prochandle *P, Elf64_Ehdr *ehdr, uint_t *phnum,
1118    uintptr_t addr)
1119{
1120	if (Pread(P, ehdr, sizeof (Elf64_Ehdr), addr) != sizeof (Elf64_Ehdr))
1121		return (-1);
1122
1123	if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1124	    ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1125	    ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1126	    ehdr->e_ident[EI_MAG3] != ELFMAG3 ||
1127	    ehdr->e_ident[EI_CLASS] != ELFCLASS64 ||
1128#ifdef _BIG_ENDIAN
1129	    ehdr->e_ident[EI_DATA] != ELFDATA2MSB ||
1130#else
1131	    ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
1132#endif
1133	    ehdr->e_ident[EI_VERSION] != EV_CURRENT)
1134		return (-1);
1135
1136	if ((*phnum = ehdr->e_phnum) == PN_XNUM) {
1137		Elf64_Shdr shdr0;
1138
1139		if (ehdr->e_shoff == 0 || ehdr->e_shentsize < sizeof (shdr0) ||
1140		    Pread(P, &shdr0, sizeof (shdr0), addr + ehdr->e_shoff) !=
1141		    sizeof (shdr0))
1142			return (-1);
1143
1144		if (shdr0.sh_info != 0)
1145			*phnum = shdr0.sh_info;
1146	}
1147
1148	return (0);
1149}
1150
1151static int
1152read_dynamic_phdr64(struct ps_prochandle *P, const Elf64_Ehdr *ehdr,
1153    uint_t phnum, Elf64_Phdr *phdr, uintptr_t addr)
1154{
1155	uint_t i;
1156
1157	for (i = 0; i < phnum; i++) {
1158		uintptr_t a = addr + ehdr->e_phoff + i * ehdr->e_phentsize;
1159		if (Pread(P, phdr, sizeof (*phdr), a) != sizeof (*phdr))
1160			return (-1);
1161
1162		if (phdr->p_type == PT_DYNAMIC)
1163			return (0);
1164	}
1165
1166	return (-1);
1167}
1168#endif	/* _LP64 */
1169
1170/*
1171 * The text segment for each load object contains the elf header and
1172 * program headers. We can use this information to determine if the
1173 * file that corresponds to the load object is the same file that
1174 * was loaded into the process's address space. There can be a discrepency
1175 * if a file is recompiled after the process is started or if the target
1176 * represents a core file from a differently configured system -- two
1177 * common examples. The DT_CHECKSUM entry in the dynamic section
1178 * provides an easy method of comparison. It is important to note that
1179 * the dynamic section usually lives in the data segment, but the meta
1180 * data we use to find the dynamic section lives in the text segment so
1181 * if either of those segments is absent we can't proceed.
1182 *
1183 * We're looking through the elf file for several items: the symbol tables
1184 * (both dynsym and symtab), the procedure linkage table (PLT) base,
1185 * size, and relocation base, and the CTF information. Most of this can
1186 * be recovered from the loaded image of the file itself, the exceptions
1187 * being the symtab and CTF data.
1188 *
1189 * First we try to open the file that we think corresponds to the load
1190 * object, if the DT_CHECKSUM values match, we're all set, and can simply
1191 * recover all the information we need from the file. If the values of
1192 * DT_CHECKSUM don't match, or if we can't access the file for whatever
1193 * reasaon, we fake up a elf file to use in its stead. If we can't read
1194 * the elf data in the process's address space, we fall back to using
1195 * the file even though it may give inaccurate information.
1196 *
1197 * The elf file that we fake up has to consist of sections for the
1198 * dynsym, the PLT and the dynamic section. Note that in the case of a
1199 * core file, we'll get the CTF data in the file_info_t later on from
1200 * a section embedded the core file (if it's present).
1201 *
1202 * file_differs() conservatively looks for mismatched files, identifying
1203 * a match when there is any ambiguity (since that's the legacy behavior).
1204 */
1205static int
1206file_differs(struct ps_prochandle *P, Elf *elf, file_info_t *fptr)
1207{
1208	Elf_Scn *scn;
1209	GElf_Shdr shdr;
1210	GElf_Dyn dyn;
1211	Elf_Data *data;
1212	uint_t i, ndyn;
1213	GElf_Xword cksum;
1214	uintptr_t addr;
1215
1216	if (fptr->file_map == NULL)
1217		return (0);
1218
1219	if ((Pcontent(P) & (CC_CONTENT_TEXT | CC_CONTENT_DATA)) !=
1220	    (CC_CONTENT_TEXT | CC_CONTENT_DATA))
1221		return (0);
1222
1223	/*
1224	 * First, we find the checksum value in the elf file.
1225	 */
1226	scn = NULL;
1227	while ((scn = elf_nextscn(elf, scn)) != NULL) {
1228		if (gelf_getshdr(scn, &shdr) != NULL &&
1229		    shdr.sh_type == SHT_DYNAMIC)
1230			goto found_shdr;
1231	}
1232	return (0);
1233
1234found_shdr:
1235	if ((data = elf_getdata(scn, NULL)) == NULL)
1236		return (0);
1237
1238	if (P->status.pr_dmodel == PR_MODEL_ILP32)
1239		ndyn = shdr.sh_size / sizeof (Elf32_Dyn);
1240#ifdef _LP64
1241	else if (P->status.pr_dmodel == PR_MODEL_LP64)
1242		ndyn = shdr.sh_size / sizeof (Elf64_Dyn);
1243#endif
1244	else
1245		return (0);
1246
1247	for (i = 0; i < ndyn; i++) {
1248		if (gelf_getdyn(data, i, &dyn) != NULL &&
1249		    dyn.d_tag == DT_CHECKSUM)
1250			goto found_cksum;
1251	}
1252
1253	/*
1254	 * The in-memory ELF has no DT_CHECKSUM section, but we will report it
1255	 * as matching the file anyhow.
1256	 */
1257	return (0);
1258
1259found_cksum:
1260	cksum = dyn.d_un.d_val;
1261	dprintf("elf cksum value is %llx\n", (u_longlong_t)cksum);
1262
1263	/*
1264	 * Get the base of the text mapping that corresponds to this file.
1265	 */
1266	addr = fptr->file_map->map_pmap.pr_vaddr;
1267
1268	if (P->status.pr_dmodel == PR_MODEL_ILP32) {
1269		Elf32_Ehdr ehdr;
1270		Elf32_Phdr phdr;
1271		Elf32_Dyn dync, *dynp;
1272		uint_t phnum, i;
1273
1274		if (read_ehdr32(P, &ehdr, &phnum, addr) != 0 ||
1275		    read_dynamic_phdr32(P, &ehdr, phnum, &phdr, addr) != 0)
1276			return (0);
1277
1278		if (ehdr.e_type == ET_DYN)
1279			phdr.p_vaddr += addr;
1280		if ((dynp = malloc(phdr.p_filesz)) == NULL)
1281			return (0);
1282		dync.d_tag = DT_NULL;
1283		if (Pread(P, dynp, phdr.p_filesz, phdr.p_vaddr) !=
1284		    phdr.p_filesz) {
1285			free(dynp);
1286			return (0);
1287		}
1288
1289		for (i = 0; i < phdr.p_filesz / sizeof (Elf32_Dyn); i++) {
1290			if (dynp[i].d_tag == DT_CHECKSUM)
1291				dync = dynp[i];
1292		}
1293
1294		free(dynp);
1295
1296		if (dync.d_tag != DT_CHECKSUM)
1297			return (0);
1298
1299		dprintf("image cksum value is %llx\n",
1300		    (u_longlong_t)dync.d_un.d_val);
1301		return (dync.d_un.d_val != cksum);
1302#ifdef _LP64
1303	} else if (P->status.pr_dmodel == PR_MODEL_LP64) {
1304		Elf64_Ehdr ehdr;
1305		Elf64_Phdr phdr;
1306		Elf64_Dyn dync, *dynp;
1307		uint_t phnum, i;
1308
1309		if (read_ehdr64(P, &ehdr, &phnum, addr) != 0 ||
1310		    read_dynamic_phdr64(P, &ehdr, phnum, &phdr, addr) != 0)
1311			return (0);
1312
1313		if (ehdr.e_type == ET_DYN)
1314			phdr.p_vaddr += addr;
1315		if ((dynp = malloc(phdr.p_filesz)) == NULL)
1316			return (0);
1317		dync.d_tag = DT_NULL;
1318		if (Pread(P, dynp, phdr.p_filesz, phdr.p_vaddr) !=
1319		    phdr.p_filesz) {
1320			free(dynp);
1321			return (0);
1322		}
1323
1324		for (i = 0; i < phdr.p_filesz / sizeof (Elf64_Dyn); i++) {
1325			if (dynp[i].d_tag == DT_CHECKSUM)
1326				dync = dynp[i];
1327		}
1328
1329		free(dynp);
1330
1331		if (dync.d_tag != DT_CHECKSUM)
1332			return (0);
1333
1334		dprintf("image cksum value is %llx\n",
1335		    (u_longlong_t)dync.d_un.d_val);
1336		return (dync.d_un.d_val != cksum);
1337#endif	/* _LP64 */
1338	}
1339
1340	return (0);
1341}
1342
1343/*
1344 * Read data from the specified process and construct an in memory
1345 * image of an ELF file that represents it well enough to let
1346 * us probe it for information.
1347 */
1348static Elf *
1349fake_elf(struct ps_prochandle *P, file_info_t *fptr)
1350{
1351	Elf *elf;
1352	uintptr_t addr;
1353	uint_t phnum;
1354
1355	if (fptr->file_map == NULL)
1356		return (NULL);
1357
1358	if ((Pcontent(P) & (CC_CONTENT_TEXT | CC_CONTENT_DATA)) !=
1359	    (CC_CONTENT_TEXT | CC_CONTENT_DATA))
1360		return (NULL);
1361
1362	addr = fptr->file_map->map_pmap.pr_vaddr;
1363
1364	if (P->status.pr_dmodel == PR_MODEL_ILP32) {
1365		Elf32_Ehdr ehdr;
1366		Elf32_Phdr phdr;
1367
1368		if ((read_ehdr32(P, &ehdr, &phnum, addr) != 0) ||
1369		    read_dynamic_phdr32(P, &ehdr, phnum, &phdr, addr) != 0)
1370			return (NULL);
1371
1372		elf = fake_elf32(P, fptr, addr, &ehdr, phnum, &phdr);
1373#ifdef _LP64
1374	} else {
1375		Elf64_Ehdr ehdr;
1376		Elf64_Phdr phdr;
1377
1378		if (read_ehdr64(P, &ehdr, &phnum, addr) != 0 ||
1379		    read_dynamic_phdr64(P, &ehdr, phnum, &phdr, addr) != 0)
1380			return (NULL);
1381
1382		elf = fake_elf64(P, fptr, addr, &ehdr, phnum, &phdr);
1383#endif
1384	}
1385
1386	return (elf);
1387}
1388
1389/*
1390 * We wouldn't need these if qsort(3C) took an argument for the callback...
1391 */
1392static mutex_t sort_mtx = DEFAULTMUTEX;
1393static char *sort_strs;
1394static GElf_Sym *sort_syms;
1395
1396int
1397byaddr_cmp_common(GElf_Sym *a, char *aname, GElf_Sym *b, char *bname)
1398{
1399	if (a->st_value < b->st_value)
1400		return (-1);
1401	if (a->st_value > b->st_value)
1402		return (1);
1403
1404	/*
1405	 * Prefer the function to the non-function.
1406	 */
1407	if (GELF_ST_TYPE(a->st_info) != GELF_ST_TYPE(b->st_info)) {
1408		if (GELF_ST_TYPE(a->st_info) == STT_FUNC)
1409			return (-1);
1410		if (GELF_ST_TYPE(b->st_info) == STT_FUNC)
1411			return (1);
1412	}
1413
1414	/*
1415	 * Prefer the weak or strong global symbol to the local symbol.
1416	 */
1417	if (GELF_ST_BIND(a->st_info) != GELF_ST_BIND(b->st_info)) {
1418		if (GELF_ST_BIND(b->st_info) == STB_LOCAL)
1419			return (-1);
1420		if (GELF_ST_BIND(a->st_info) == STB_LOCAL)
1421			return (1);
1422	}
1423
1424	/*
1425	 * Prefer the symbol that doesn't begin with a '$' since compilers and
1426	 * other symbol generators often use it as a prefix.
1427	 */
1428	if (*bname == '$')
1429		return (-1);
1430	if (*aname == '$')
1431		return (1);
1432
1433	/*
1434	 * Prefer the name with fewer leading underscores in the name.
1435	 */
1436	while (*aname == '_' && *bname == '_') {
1437		aname++;
1438		bname++;
1439	}
1440
1441	if (*bname == '_')
1442		return (-1);
1443	if (*aname == '_')
1444		return (1);
1445
1446	/*
1447	 * Prefer the symbol with the smaller size.
1448	 */
1449	if (a->st_size < b->st_size)
1450		return (-1);
1451	if (a->st_size > b->st_size)
1452		return (1);
1453
1454	/*
1455	 * All other factors being equal, fall back to lexicographic order.
1456	 */
1457	return (strcmp(aname, bname));
1458}
1459
1460static int
1461byaddr_cmp(const void *aa, const void *bb)
1462{
1463	GElf_Sym *a = &sort_syms[*(uint_t *)aa];
1464	GElf_Sym *b = &sort_syms[*(uint_t *)bb];
1465	char *aname = sort_strs + a->st_name;
1466	char *bname = sort_strs + b->st_name;
1467
1468	return (byaddr_cmp_common(a, aname, b, bname));
1469}
1470
1471static int
1472byname_cmp(const void *aa, const void *bb)
1473{
1474	GElf_Sym *a = &sort_syms[*(uint_t *)aa];
1475	GElf_Sym *b = &sort_syms[*(uint_t *)bb];
1476	char *aname = sort_strs + a->st_name;
1477	char *bname = sort_strs + b->st_name;
1478
1479	return (strcmp(aname, bname));
1480}
1481
1482/*
1483 * Given a symbol index, look up the corresponding symbol from the
1484 * given symbol table.
1485 *
1486 * This function allows the caller to treat the symbol table as a single
1487 * logical entity even though there may be 2 actual ELF symbol tables
1488 * involved. See the comments in Pcontrol.h for details.
1489 */
1490static GElf_Sym *
1491symtab_getsym(sym_tbl_t *symtab, int ndx, GElf_Sym *dst)
1492{
1493	/* If index is in range of primary symtab, look it up there */
1494	if (ndx >= symtab->sym_symn_aux) {
1495		return (gelf_getsym(symtab->sym_data_pri,
1496		    ndx - symtab->sym_symn_aux, dst));
1497	}
1498
1499	/* Not in primary: Look it up in the auxiliary symtab */
1500	return (gelf_getsym(symtab->sym_data_aux, ndx, dst));
1501}
1502
1503void
1504optimize_symtab(sym_tbl_t *symtab)
1505{
1506	GElf_Sym *symp, *syms;
1507	uint_t i, *indexa, *indexb;
1508	size_t symn, strsz, count;
1509
1510	if (symtab == NULL || symtab->sym_data_pri == NULL ||
1511	    symtab->sym_byaddr != NULL)
1512		return;
1513
1514	symn = symtab->sym_symn;
1515	strsz = symtab->sym_strsz;
1516
1517	symp = syms = malloc(sizeof (GElf_Sym) * symn);
1518	if (symp == NULL) {
1519		dprintf("optimize_symtab: failed to malloc symbol array");
1520		return;
1521	}
1522
1523	/*
1524	 * First record all the symbols into a table and count up the ones
1525	 * that we're interested in. We mark symbols as invalid by setting
1526	 * the st_name to an illegal value.
1527	 */
1528	for (i = 0, count = 0; i < symn; i++, symp++) {
1529		if (symtab_getsym(symtab, i, symp) != NULL &&
1530		    symp->st_name < strsz &&
1531		    IS_DATA_TYPE(GELF_ST_TYPE(symp->st_info)))
1532			count++;
1533		else
1534			symp->st_name = strsz;
1535	}
1536
1537	/*
1538	 * Allocate sufficient space for both tables and populate them
1539	 * with the same symbols we just counted.
1540	 */
1541	symtab->sym_count = count;
1542	indexa = symtab->sym_byaddr = calloc(sizeof (uint_t), count);
1543	indexb = symtab->sym_byname = calloc(sizeof (uint_t), count);
1544	if (indexa == NULL || indexb == NULL) {
1545		dprintf(
1546		    "optimize_symtab: failed to malloc symbol index arrays");
1547		symtab->sym_count = 0;
1548		if (indexa != NULL) {	/* First alloc succeeded. Free it */
1549			free(indexa);
1550			symtab->sym_byaddr = NULL;
1551		}
1552		free(syms);
1553		return;
1554	}
1555	for (i = 0, symp = syms; i < symn; i++, symp++) {
1556		if (symp->st_name < strsz)
1557			*indexa++ = *indexb++ = i;
1558	}
1559
1560	/*
1561	 * Sort the two tables according to the appropriate criteria,
1562	 * unless the user has overridden this behaviour.
1563	 *
1564	 * An example where we might not sort the tables is the relatively
1565	 * unusual case of a process with very large symbol tables in which
1566	 * we perform few lookups. In such a case the total time would be
1567	 * dominated by the sort. It is difficult to determine a priori
1568	 * how many lookups an arbitrary client will perform, and
1569	 * hence whether the symbol tables should be sorted. We therefore
1570	 * sort the tables by default, but provide the user with a
1571	 * "chicken switch" in the form of the LIBPROC_NO_QSORT
1572	 * environment variable.
1573	 */
1574	if (!_libproc_no_qsort) {
1575		(void) mutex_lock(&sort_mtx);
1576		sort_strs = symtab->sym_strs;
1577		sort_syms = syms;
1578
1579		qsort(symtab->sym_byaddr, count, sizeof (uint_t), byaddr_cmp);
1580		qsort(symtab->sym_byname, count, sizeof (uint_t), byname_cmp);
1581
1582		sort_strs = NULL;
1583		sort_syms = NULL;
1584		(void) mutex_unlock(&sort_mtx);
1585	}
1586
1587	free(syms);
1588}
1589
1590/*
1591 * Build the symbol table for the given mapped file.
1592 */
1593void
1594Pbuild_file_symtab(struct ps_prochandle *P, file_info_t *fptr)
1595{
1596	char objectfile[PATH_MAX];
1597	uint_t i;
1598
1599	GElf_Ehdr ehdr;
1600	GElf_Sym s;
1601
1602	Elf_Data *shdata;
1603	Elf_Scn *scn;
1604	Elf *elf;
1605	size_t nshdrs, shstrndx;
1606
1607	struct {
1608		GElf_Shdr c_shdr;
1609		Elf_Data *c_data;
1610		const char *c_name;
1611	} *cp, *cache = NULL, *dyn = NULL, *plt = NULL, *ctf = NULL;
1612
1613	if (fptr->file_init)
1614		return;	/* We've already processed this file */
1615
1616	/*
1617	 * Mark the file_info struct as having the symbol table initialized
1618	 * even if we fail below.  We tried once; we don't try again.
1619	 */
1620	fptr->file_init = 1;
1621
1622	if (elf_version(EV_CURRENT) == EV_NONE) {
1623		dprintf("libproc ELF version is more recent than libelf\n");
1624		return;
1625	}
1626
1627	if (P->state == PS_DEAD || P->state == PS_IDLE) {
1628		char *name;
1629		/*
1630		 * If we're a not live, we can't open files from the /proc
1631		 * object directory; we have only the mapping and file names
1632		 * to guide us.  We prefer the file_lname, but need to handle
1633		 * the case of it being NULL in order to bootstrap: we first
1634		 * come here during rd_new() when the only information we have
1635		 * is interpreter name associated with the AT_BASE mapping.
1636		 *
1637		 * Also, if the zone associated with the core file seems
1638		 * to exists on this machine we'll try to open the object
1639		 * file within the zone.
1640		 */
1641		if (fptr->file_rname != NULL)
1642			name = fptr->file_rname;
1643		else if (fptr->file_lname != NULL)
1644			name = fptr->file_lname;
1645		else
1646			name = fptr->file_pname;
1647		(void) strlcpy(objectfile, name, sizeof (objectfile));
1648	} else {
1649		(void) snprintf(objectfile, sizeof (objectfile),
1650		    "%s/%d/object/%s",
1651		    procfs_path, (int)P->pid, fptr->file_pname);
1652	}
1653
1654	/*
1655	 * Open the object file, create the elf file, and then get the elf
1656	 * header and .shstrtab data buffer so we can process sections by
1657	 * name. If anything goes wrong try to fake up an elf file from
1658	 * the in-core elf image.
1659	 */
1660	if ((fptr->file_fd = open(objectfile, O_RDONLY)) < 0) {
1661		dprintf("Pbuild_file_symtab: failed to open %s: %s\n",
1662		    objectfile, strerror(errno));
1663
1664		if ((elf = fake_elf(P, fptr)) == NULL ||
1665		    elf_kind(elf) != ELF_K_ELF ||
1666		    gelf_getehdr(elf, &ehdr) == NULL ||
1667		    elf_getshdrnum(elf, &nshdrs) == -1 ||
1668		    elf_getshdrstrndx(elf, &shstrndx) == -1 ||
1669		    (scn = elf_getscn(elf, shstrndx)) == NULL ||
1670		    (shdata = elf_getdata(scn, NULL)) == NULL) {
1671			dprintf("failed to fake up ELF file\n");
1672			return;
1673		}
1674
1675	} else if ((elf = elf_begin(fptr->file_fd, ELF_C_READ, NULL)) == NULL ||
1676	    elf_kind(elf) != ELF_K_ELF ||
1677	    gelf_getehdr(elf, &ehdr) == NULL ||
1678	    elf_getshdrnum(elf, &nshdrs) == -1 ||
1679	    elf_getshdrstrndx(elf, &shstrndx) == -1 ||
1680	    (scn = elf_getscn(elf, shstrndx)) == NULL ||
1681	    (shdata = elf_getdata(scn, NULL)) == NULL) {
1682		int err = elf_errno();
1683
1684		dprintf("failed to process ELF file %s: %s\n",
1685		    objectfile, (err == 0) ? "<null>" : elf_errmsg(err));
1686
1687		if ((elf = fake_elf(P, fptr)) == NULL ||
1688		    elf_kind(elf) != ELF_K_ELF ||
1689		    gelf_getehdr(elf, &ehdr) == NULL ||
1690		    elf_getshdrnum(elf, &nshdrs) == -1 ||
1691		    elf_getshdrstrndx(elf, &shstrndx) == -1 ||
1692		    (scn = elf_getscn(elf, shstrndx)) == NULL ||
1693		    (shdata = elf_getdata(scn, NULL)) == NULL) {
1694			dprintf("failed to fake up ELF file\n");
1695			goto bad;
1696		}
1697
1698	} else if (file_differs(P, elf, fptr)) {
1699		Elf *newelf;
1700
1701		/*
1702		 * Before we get too excited about this elf file, we'll check
1703		 * its checksum value against the value we have in memory. If
1704		 * they don't agree, we try to fake up a new elf file and
1705		 * proceed with that instead.
1706		 */
1707
1708		dprintf("ELF file %s (%lx) doesn't match in-core image\n",
1709		    fptr->file_pname,
1710		    (ulong_t)fptr->file_map->map_pmap.pr_vaddr);
1711
1712		if ((newelf = fake_elf(P, fptr)) == NULL ||
1713		    elf_kind(newelf) != ELF_K_ELF ||
1714		    gelf_getehdr(newelf, &ehdr) == NULL ||
1715		    elf_getshdrnum(newelf, &nshdrs) == -1 ||
1716		    elf_getshdrstrndx(newelf, &shstrndx) == -1 ||
1717		    (scn = elf_getscn(newelf, shstrndx)) == NULL ||
1718		    (shdata = elf_getdata(scn, NULL)) == NULL) {
1719			dprintf("failed to fake up ELF file\n");
1720		} else {
1721			(void) elf_end(elf);
1722			elf = newelf;
1723
1724			dprintf("switched to faked up ELF file\n");
1725		}
1726	}
1727
1728	if ((cache = malloc(nshdrs * sizeof (*cache))) == NULL) {
1729		dprintf("failed to malloc section cache for %s\n", objectfile);
1730		goto bad;
1731	}
1732
1733	dprintf("processing ELF file %s\n", objectfile);
1734	fptr->file_class = ehdr.e_ident[EI_CLASS];
1735	fptr->file_etype = ehdr.e_type;
1736	fptr->file_elf = elf;
1737	fptr->file_shstrs = shdata->d_buf;
1738	fptr->file_shstrsz = shdata->d_size;
1739
1740	/*
1741	 * Iterate through each section, caching its section header, data
1742	 * pointer, and name.  We use this for handling sh_link values below.
1743	 */
1744	for (cp = cache + 1, scn = NULL; scn = elf_nextscn(elf, scn); cp++) {
1745		if (gelf_getshdr(scn, &cp->c_shdr) == NULL) {
1746			dprintf("Pbuild_file_symtab: Failed to get section "
1747			    "header\n");
1748			goto bad; /* Failed to get section header */
1749		}
1750
1751		if ((cp->c_data = elf_getdata(scn, NULL)) == NULL) {
1752			dprintf("Pbuild_file_symtab: Failed to get section "
1753			    "data\n");
1754			goto bad; /* Failed to get section data */
1755		}
1756
1757		if (cp->c_shdr.sh_name >= shdata->d_size) {
1758			dprintf("Pbuild_file_symtab: corrupt section name");
1759			goto bad; /* Corrupt section name */
1760		}
1761
1762		cp->c_name = (const char *)shdata->d_buf + cp->c_shdr.sh_name;
1763	}
1764
1765	/*
1766	 * Now iterate through the section cache in order to locate info
1767	 * for the .symtab, .dynsym, .SUNW_ldynsym, .dynamic, .plt,
1768	 * and .SUNW_ctf sections:
1769	 */
1770	for (i = 1, cp = cache + 1; i < nshdrs; i++, cp++) {
1771		GElf_Shdr *shp = &cp->c_shdr;
1772
1773		if (shp->sh_type == SHT_SYMTAB || shp->sh_type == SHT_DYNSYM) {
1774			sym_tbl_t *symp = shp->sh_type == SHT_SYMTAB ?
1775			    &fptr->file_symtab : &fptr->file_dynsym;
1776			/*
1777			 * It's possible that the we already got the symbol
1778			 * table from the core file itself. Either the file
1779			 * differs in which case our faked up elf file will
1780			 * only contain the dynsym (not the symtab) or the
1781			 * file matches in which case we'll just be replacing
1782			 * the symbol table we pulled out of the core file
1783			 * with an equivalent one. In either case, this
1784			 * check isn't essential, but it's a good idea.
1785			 */
1786			if (symp->sym_data_pri == NULL) {
1787				dprintf("Symbol table found for %s\n",
1788				    objectfile);
1789				symp->sym_data_pri = cp->c_data;
1790				symp->sym_symn +=
1791				    shp->sh_size / shp->sh_entsize;
1792				symp->sym_strs =
1793				    cache[shp->sh_link].c_data->d_buf;
1794				symp->sym_strsz =
1795				    cache[shp->sh_link].c_data->d_size;
1796				symp->sym_hdr_pri = cp->c_shdr;
1797				symp->sym_strhdr = cache[shp->sh_link].c_shdr;
1798			} else {
1799				dprintf("Symbol table already there for %s\n",
1800				    objectfile);
1801			}
1802		} else if (shp->sh_type == SHT_SUNW_LDYNSYM) {
1803			/* .SUNW_ldynsym section is auxiliary to .dynsym */
1804			if (fptr->file_dynsym.sym_data_aux == NULL) {
1805				dprintf(".SUNW_ldynsym symbol table"
1806				    " found for %s\n", objectfile);
1807				fptr->file_dynsym.sym_data_aux = cp->c_data;
1808				fptr->file_dynsym.sym_symn_aux =
1809				    shp->sh_size / shp->sh_entsize;
1810				fptr->file_dynsym.sym_symn +=
1811				    fptr->file_dynsym.sym_symn_aux;
1812				fptr->file_dynsym.sym_hdr_aux = cp->c_shdr;
1813			} else {
1814				dprintf(".SUNW_ldynsym symbol table already"
1815				    " there for %s\n", objectfile);
1816			}
1817		} else if (shp->sh_type == SHT_DYNAMIC) {
1818			dyn = cp;
1819		} else if (strcmp(cp->c_name, ".plt") == 0) {
1820			plt = cp;
1821		} else if (strcmp(cp->c_name, ".SUNW_ctf") == 0) {
1822			/*
1823			 * Skip over bogus CTF sections so they don't come back
1824			 * to haunt us later.
1825			 */
1826			if (shp->sh_link == 0 ||
1827			    shp->sh_link >= nshdrs ||
1828			    (cache[shp->sh_link].c_shdr.sh_type != SHT_DYNSYM &&
1829			    cache[shp->sh_link].c_shdr.sh_type != SHT_SYMTAB)) {
1830				dprintf("Bad sh_link %d for "
1831				    "CTF\n", shp->sh_link);
1832				continue;
1833			}
1834			ctf = cp;
1835		}
1836	}
1837
1838	/*
1839	 * At this point, we've found all the symbol tables we're ever going
1840	 * to find: the ones in the loop above and possibly the symtab that
1841	 * was included in the core file. Before we perform any lookups, we
1842	 * create sorted versions to optimize for lookups.
1843	 */
1844	optimize_symtab(&fptr->file_symtab);
1845	optimize_symtab(&fptr->file_dynsym);
1846
1847	/*
1848	 * Fill in the base address of the text mapping for shared libraries.
1849	 * This allows us to translate symbols before librtld_db is ready.
1850	 */
1851	if (fptr->file_etype == ET_DYN) {
1852		fptr->file_dyn_base = fptr->file_map->map_pmap.pr_vaddr -
1853		    fptr->file_map->map_pmap.pr_offset;
1854		dprintf("setting file_dyn_base for %s to %lx\n",
1855		    objectfile, (long)fptr->file_dyn_base);
1856	}
1857
1858	/*
1859	 * Record the CTF section information in the file info structure.
1860	 */
1861	if (ctf != NULL) {
1862		fptr->file_ctf_off = ctf->c_shdr.sh_offset;
1863		fptr->file_ctf_size = ctf->c_shdr.sh_size;
1864		if (ctf->c_shdr.sh_link != 0 &&
1865		    cache[ctf->c_shdr.sh_link].c_shdr.sh_type == SHT_DYNSYM)
1866			fptr->file_ctf_dyn = 1;
1867	}
1868
1869	if (fptr->file_lo == NULL)
1870		goto done; /* Nothing else to do if no load object info */
1871
1872	/*
1873	 * If the object is a shared library and we have a different rl_base
1874	 * value, reset file_dyn_base according to librtld_db's information.
1875	 */
1876	if (fptr->file_etype == ET_DYN &&
1877	    fptr->file_lo->rl_base != fptr->file_dyn_base) {
1878		dprintf("resetting file_dyn_base for %s to %lx\n",
1879		    objectfile, (long)fptr->file_lo->rl_base);
1880		fptr->file_dyn_base = fptr->file_lo->rl_base;
1881	}
1882
1883	/*
1884	 * Fill in the PLT information for this file if a PLT symbol is found.
1885	 */
1886	if (sym_by_name(&fptr->file_dynsym, "_PROCEDURE_LINKAGE_TABLE_", &s,
1887	    NULL) != NULL) {
1888		fptr->file_plt_base = s.st_value + fptr->file_dyn_base;
1889		fptr->file_plt_size = (plt != NULL) ? plt->c_shdr.sh_size : 0;
1890
1891		/*
1892		 * Bring the load object up to date; it is the only way the
1893		 * user has to access the PLT data. The PLT information in the
1894		 * rd_loadobj_t is not set in the call to map_iter() (the
1895		 * callback for rd_loadobj_iter) where we set file_lo.
1896		 */
1897		fptr->file_lo->rl_plt_base = fptr->file_plt_base;
1898		fptr->file_lo->rl_plt_size = fptr->file_plt_size;
1899
1900		dprintf("PLT found at %p, size = %lu\n",
1901		    (void *)fptr->file_plt_base, (ulong_t)fptr->file_plt_size);
1902	}
1903
1904	/*
1905	 * Fill in the PLT information.
1906	 */
1907	if (dyn != NULL) {
1908		uintptr_t dynaddr = dyn->c_shdr.sh_addr + fptr->file_dyn_base;
1909		size_t ndyn = dyn->c_shdr.sh_size / dyn->c_shdr.sh_entsize;
1910		GElf_Dyn d;
1911
1912		for (i = 0; i < ndyn; i++) {
1913			if (gelf_getdyn(dyn->c_data, i, &d) != NULL &&
1914			    d.d_tag == DT_JMPREL) {
1915				dprintf("DT_JMPREL is %p\n",
1916				    (void *)(uintptr_t)d.d_un.d_ptr);
1917				fptr->file_jmp_rel =
1918				    d.d_un.d_ptr + fptr->file_dyn_base;
1919				break;
1920			}
1921		}
1922
1923		dprintf("_DYNAMIC found at %p, %lu entries, DT_JMPREL = %p\n",
1924		    (void *)dynaddr, (ulong_t)ndyn, (void *)fptr->file_jmp_rel);
1925	}
1926
1927done:
1928	free(cache);
1929	return;
1930
1931bad:
1932	if (cache != NULL)
1933		free(cache);
1934
1935	(void) elf_end(elf);
1936	fptr->file_elf = NULL;
1937	if (fptr->file_elfmem != NULL) {
1938		free(fptr->file_elfmem);
1939		fptr->file_elfmem = NULL;
1940	}
1941	(void) close(fptr->file_fd);
1942	fptr->file_fd = -1;
1943}
1944
1945/*
1946 * Given a process virtual address, return the map_info_t containing it.
1947 * If none found, return NULL.
1948 */
1949map_info_t *
1950Paddr2mptr(struct ps_prochandle *P, uintptr_t addr)
1951{
1952	int lo = 0;
1953	int hi = P->map_count - 1;
1954	int mid;
1955	map_info_t *mp;
1956
1957	while (lo <= hi) {
1958
1959		mid = (lo + hi) / 2;
1960		mp = &P->mappings[mid];
1961
1962		/* check that addr is in [vaddr, vaddr + size) */
1963		if ((addr - mp->map_pmap.pr_vaddr) < mp->map_pmap.pr_size)
1964			return (mp);
1965
1966		if (addr < mp->map_pmap.pr_vaddr)
1967			hi = mid - 1;
1968		else
1969			lo = mid + 1;
1970	}
1971
1972	return (NULL);
1973}
1974
1975/*
1976 * Return the map_info_t for the executable file.
1977 * If not found, return NULL.
1978 */
1979static map_info_t *
1980exec_map(struct ps_prochandle *P)
1981{
1982	uint_t i;
1983	map_info_t *mptr;
1984	map_info_t *mold = NULL;
1985	file_info_t *fptr;
1986	uintptr_t base;
1987
1988	for (i = 0, mptr = P->mappings; i < P->map_count; i++, mptr++) {
1989		if (mptr->map_pmap.pr_mapname[0] == '\0')
1990			continue;
1991		if (strcmp(mptr->map_pmap.pr_mapname, "a.out") == 0) {
1992			if ((fptr = mptr->map_file) != NULL &&
1993			    fptr->file_lo != NULL) {
1994				base = fptr->file_lo->rl_base;
1995				if (base >= mptr->map_pmap.pr_vaddr &&
1996				    base < mptr->map_pmap.pr_vaddr +
1997				    mptr->map_pmap.pr_size)	/* text space */
1998					return (mptr);
1999				mold = mptr;	/* must be the data */
2000				continue;
2001			}
2002			/* This is a poor way to test for text space */
2003			if (!(mptr->map_pmap.pr_mflags & MA_EXEC) ||
2004			    (mptr->map_pmap.pr_mflags & MA_WRITE)) {
2005				mold = mptr;
2006				continue;
2007			}
2008			return (mptr);
2009		}
2010	}
2011
2012	return (mold);
2013}
2014
2015/*
2016 * Given a shared object name, return the map_info_t for it.  If no matching
2017 * object is found, return NULL.  Normally, the link maps contain the full
2018 * object pathname, e.g. /usr/lib/libc.so.1.  We allow the object name to
2019 * take one of the following forms:
2020 *
2021 * 1. An exact match (i.e. a full pathname): "/usr/lib/libc.so.1"
2022 * 2. An exact basename match: "libc.so.1"
2023 * 3. An initial basename match up to a '.' suffix: "libc.so" or "libc"
2024 * 4. The literal string "a.out" is an alias for the executable mapping
2025 *
2026 * The third case is a convenience for callers and may not be necessary.
2027 *
2028 * As the exact same object name may be loaded on different link maps (see
2029 * dlmopen(3DL)), we also allow the caller to resolve the object name by
2030 * specifying a particular link map id.  If lmid is PR_LMID_EVERY, the
2031 * first matching name will be returned, regardless of the link map id.
2032 */
2033static map_info_t *
2034object_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *objname)
2035{
2036	map_info_t *mp;
2037	file_info_t *fp;
2038	size_t objlen;
2039	uint_t i;
2040
2041	/*
2042	 * If we have no rtld_db, then always treat a request as one for all
2043	 * link maps.
2044	 */
2045	if (P->rap == NULL)
2046		lmid = PR_LMID_EVERY;
2047
2048	/*
2049	 * First pass: look for exact matches of the entire pathname or
2050	 * basename (cases 1 and 2 above):
2051	 */
2052	for (i = 0, mp = P->mappings; i < P->map_count; i++, mp++) {
2053
2054		if (mp->map_pmap.pr_mapname[0] == '\0' ||
2055		    (fp = mp->map_file) == NULL ||
2056		    ((fp->file_lname == NULL) && (fp->file_rname == NULL)))
2057			continue;
2058
2059		if (lmid != PR_LMID_EVERY &&
2060		    (fp->file_lo == NULL || lmid != fp->file_lo->rl_lmident))
2061			continue;
2062
2063		/*
2064		 * If we match, return the primary text mapping; otherwise
2065		 * just return the mapping we matched.
2066		 */
2067		if ((fp->file_lbase && strcmp(fp->file_lbase, objname) == 0) ||
2068		    (fp->file_rbase && strcmp(fp->file_rbase, objname) == 0) ||
2069		    (fp->file_lname && strcmp(fp->file_lname, objname) == 0) ||
2070		    (fp->file_rname && strcmp(fp->file_rname, objname) == 0))
2071			return (fp->file_map ? fp->file_map : mp);
2072	}
2073
2074	objlen = strlen(objname);
2075
2076	/*
2077	 * Second pass: look for partial matches (case 3 above):
2078	 */
2079	for (i = 0, mp = P->mappings; i < P->map_count; i++, mp++) {
2080
2081		if (mp->map_pmap.pr_mapname[0] == '\0' ||
2082		    (fp = mp->map_file) == NULL ||
2083		    ((fp->file_lname == NULL) && (fp->file_rname == NULL)))
2084			continue;
2085
2086		if (lmid != PR_LMID_EVERY &&
2087		    (fp->file_lo == NULL || lmid != fp->file_lo->rl_lmident))
2088			continue;
2089
2090		/*
2091		 * If we match, return the primary text mapping; otherwise
2092		 * just return the mapping we matched.
2093		 */
2094		if ((fp->file_lbase != NULL) &&
2095		    (strncmp(fp->file_lbase, objname, objlen) == 0) &&
2096		    (fp->file_lbase[objlen] == '.'))
2097			return (fp->file_map ? fp->file_map : mp);
2098		if ((fp->file_rbase != NULL) &&
2099		    (strncmp(fp->file_rbase, objname, objlen) == 0) &&
2100		    (fp->file_rbase[objlen] == '.'))
2101			return (fp->file_map ? fp->file_map : mp);
2102	}
2103
2104	/*
2105	 * One last check: we allow "a.out" to always alias the executable,
2106	 * assuming this name was not in use for something else.
2107	 */
2108	if ((lmid == PR_LMID_EVERY || lmid == LM_ID_BASE) &&
2109	    (strcmp(objname, "a.out") == 0))
2110		return (P->map_exec);
2111
2112	return (NULL);
2113}
2114
2115static map_info_t *
2116object_name_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *name)
2117{
2118	map_info_t *mptr;
2119
2120	if (!P->info_valid)
2121		Pupdate_maps(P);
2122
2123	if (P->map_exec == NULL && ((mptr = Paddr2mptr(P,
2124	    Pgetauxval(P, AT_ENTRY))) != NULL || (mptr = exec_map(P)) != NULL))
2125		P->map_exec = mptr;
2126
2127	if (P->map_ldso == NULL && (mptr = Paddr2mptr(P,
2128	    Pgetauxval(P, AT_BASE))) != NULL)
2129		P->map_ldso = mptr;
2130
2131	if (name == PR_OBJ_EXEC)
2132		mptr = P->map_exec;
2133	else if (name == PR_OBJ_LDSO)
2134		mptr = P->map_ldso;
2135	else if (Prd_agent(P) != NULL || P->state == PS_IDLE)
2136		mptr = object_to_map(P, lmid, name);
2137	else
2138		mptr = NULL;
2139
2140	return (mptr);
2141}
2142
2143/*
2144 * When two symbols are found by address, decide which one is to be preferred.
2145 */
2146static GElf_Sym *
2147sym_prefer(GElf_Sym *sym1, char *name1, GElf_Sym *sym2, char *name2)
2148{
2149	/*
2150	 * Prefer the non-NULL symbol.
2151	 */
2152	if (sym1 == NULL)
2153		return (sym2);
2154	if (sym2 == NULL)
2155		return (sym1);
2156
2157	/*
2158	 * Defer to the sort ordering...
2159	 */
2160	return (byaddr_cmp_common(sym1, name1, sym2, name2) <= 0 ? sym1 : sym2);
2161}
2162
2163/*
2164 * Use a binary search to do the work of sym_by_addr().
2165 */
2166static GElf_Sym *
2167sym_by_addr_binary(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symp,
2168    uint_t *idp)
2169{
2170	GElf_Sym sym, osym;
2171	uint_t i, oid, *byaddr = symtab->sym_byaddr;
2172	int min, max, mid, omid, found = 0;
2173
2174	if (symtab->sym_data_pri == NULL || symtab->sym_count == 0)
2175		return (NULL);
2176
2177	min = 0;
2178	max = symtab->sym_count - 1;
2179	osym.st_value = 0;
2180
2181	/*
2182	 * We can't return when we've found a match, we have to continue
2183	 * searching for the closest matching symbol.
2184	 */
2185	while (min <= max) {
2186		mid = (max + min) / 2;
2187
2188		i = byaddr[mid];
2189		(void) symtab_getsym(symtab, i, &sym);
2190
2191		if (addr >= sym.st_value &&
2192		    addr < sym.st_value + sym.st_size &&
2193		    (!found || sym.st_value > osym.st_value)) {
2194			osym = sym;
2195			omid = mid;
2196			oid = i;
2197			found = 1;
2198		}
2199
2200		if (addr < sym.st_value)
2201			max = mid - 1;
2202		else
2203			min = mid + 1;
2204	}
2205
2206	if (!found)
2207		return (NULL);
2208
2209	/*
2210	 * There may be many symbols with identical values so we walk
2211	 * backward in the byaddr table to find the best match.
2212	 */
2213	do {
2214		sym = osym;
2215		i = oid;
2216
2217		if (omid == 0)
2218			break;
2219
2220		oid = byaddr[--omid];
2221		(void) symtab_getsym(symtab, oid, &osym);
2222	} while (addr >= osym.st_value &&
2223	    addr < sym.st_value + osym.st_size &&
2224	    osym.st_value == sym.st_value);
2225
2226	*symp = sym;
2227	if (idp != NULL)
2228		*idp = i;
2229	return (symp);
2230}
2231
2232/*
2233 * Use a linear search to do the work of sym_by_addr().
2234 */
2235static GElf_Sym *
2236sym_by_addr_linear(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symbolp,
2237    uint_t *idp)
2238{
2239	size_t symn = symtab->sym_symn;
2240	char *strs = symtab->sym_strs;
2241	GElf_Sym sym, *symp = NULL;
2242	GElf_Sym osym, *osymp = NULL;
2243	int i, id;
2244
2245	if (symtab->sym_data_pri == NULL || symn == 0 || strs == NULL)
2246		return (NULL);
2247
2248	for (i = 0; i < symn; i++) {
2249		if ((symp = symtab_getsym(symtab, i, &sym)) != NULL) {
2250			if (addr >= sym.st_value &&
2251			    addr < sym.st_value + sym.st_size) {
2252				if (osymp)
2253					symp = sym_prefer(
2254					    symp, strs + symp->st_name,
2255					    osymp, strs + osymp->st_name);
2256				if (symp != osymp) {
2257					osym = sym;
2258					osymp = &osym;
2259					id = i;
2260				}
2261			}
2262		}
2263	}
2264	if (osymp) {
2265		*symbolp = osym;
2266		if (idp)
2267			*idp = id;
2268		return (symbolp);
2269	}
2270	return (NULL);
2271}
2272
2273/*
2274 * Look up a symbol by address in the specified symbol table.
2275 * Adjustment to 'addr' must already have been made for the
2276 * offset of the symbol if this is a dynamic library symbol table.
2277 *
2278 * Use a linear or a binary search depending on whether or not we
2279 * chose to sort the table in optimize_symtab().
2280 */
2281static GElf_Sym *
2282sym_by_addr(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symp, uint_t *idp)
2283{
2284	if (_libproc_no_qsort) {
2285		return (sym_by_addr_linear(symtab, addr, symp, idp));
2286	} else {
2287		return (sym_by_addr_binary(symtab, addr, symp, idp));
2288	}
2289}
2290
2291/*
2292 * Use a binary search to do the work of sym_by_name().
2293 */
2294static GElf_Sym *
2295sym_by_name_binary(sym_tbl_t *symtab, const char *name, GElf_Sym *symp,
2296    uint_t *idp)
2297{
2298	char *strs = symtab->sym_strs;
2299	uint_t i, *byname = symtab->sym_byname;
2300	int min, mid, max, cmp;
2301
2302	if (symtab->sym_data_pri == NULL || strs == NULL ||
2303	    symtab->sym_count == 0)
2304		return (NULL);
2305
2306	min = 0;
2307	max = symtab->sym_count - 1;
2308
2309	while (min <= max) {
2310		mid = (max + min) / 2;
2311
2312		i = byname[mid];
2313		(void) symtab_getsym(symtab, i, symp);
2314
2315		if ((cmp = strcmp(name, strs + symp->st_name)) == 0) {
2316			if (idp != NULL)
2317				*idp = i;
2318			return (symp);
2319		}
2320
2321		if (cmp < 0)
2322			max = mid - 1;
2323		else
2324			min = mid + 1;
2325	}
2326
2327	return (NULL);
2328}
2329
2330/*
2331 * Use a linear search to do the work of sym_by_name().
2332 */
2333static GElf_Sym *
2334sym_by_name_linear(sym_tbl_t *symtab, const char *name, GElf_Sym *symp,
2335    uint_t *idp)
2336{
2337	size_t symn = symtab->sym_symn;
2338	char *strs = symtab->sym_strs;
2339	int i;
2340
2341	if (symtab->sym_data_pri == NULL || symn == 0 || strs == NULL)
2342		return (NULL);
2343
2344	for (i = 0; i < symn; i++) {
2345		if (symtab_getsym(symtab, i, symp) &&
2346		    strcmp(name, strs + symp->st_name) == 0) {
2347			if (idp)
2348				*idp = i;
2349			return (symp);
2350		}
2351	}
2352
2353	return (NULL);
2354}
2355
2356/*
2357 * Look up a symbol by name in the specified symbol table.
2358 *
2359 * Use a linear or a binary search depending on whether or not we
2360 * chose to sort the table in optimize_symtab().
2361 */
2362static GElf_Sym *
2363sym_by_name(sym_tbl_t *symtab, const char *name, GElf_Sym *symp, uint_t *idp)
2364{
2365	if (_libproc_no_qsort) {
2366		return (sym_by_name_linear(symtab, name, symp, idp));
2367	} else {
2368		return (sym_by_name_binary(symtab, name, symp, idp));
2369	}
2370}
2371
2372/*
2373 * Search the process symbol tables looking for a symbol whose
2374 * value to value+size contain the address specified by addr.
2375 * Return values are:
2376 *	sym_name_buffer containing the symbol name
2377 *	GElf_Sym symbol table entry
2378 *	prsyminfo_t ancillary symbol information
2379 * Returns 0 on success, -1 on failure.
2380 */
2381static int
2382i_Pxlookup_by_addr(
2383	struct ps_prochandle *P,
2384	int lmresolve,			/* use resolve linker object names */
2385	uintptr_t addr,			/* process address being sought */
2386	char *sym_name_buffer,		/* buffer for the symbol name */
2387	size_t bufsize,			/* size of sym_name_buffer */
2388	GElf_Sym *symbolp,		/* returned symbol table entry */
2389	prsyminfo_t *sip)		/* returned symbol info */
2390{
2391	GElf_Sym	*symp;
2392	char		*name;
2393	GElf_Sym	sym1, *sym1p = NULL;
2394	GElf_Sym	sym2, *sym2p = NULL;
2395	char		*name1 = NULL;
2396	char		*name2 = NULL;
2397	uint_t		i1;
2398	uint_t		i2;
2399	map_info_t	*mptr;
2400	file_info_t	*fptr;
2401
2402	(void) Prd_agent(P);
2403
2404	if ((mptr = Paddr2mptr(P, addr)) == NULL ||	/* no such address */
2405	    (fptr = build_map_symtab(P, mptr)) == NULL || /* no mapped file */
2406	    fptr->file_elf == NULL)			/* not an ELF file */
2407		return (-1);
2408
2409	/*
2410	 * Adjust the address by the load object base address in
2411	 * case the address turns out to be in a shared library.
2412	 */
2413	addr -= fptr->file_dyn_base;
2414
2415	/*
2416	 * Search both symbol tables, symtab first, then dynsym.
2417	 */
2418	if ((sym1p = sym_by_addr(&fptr->file_symtab, addr, &sym1, &i1)) != NULL)
2419		name1 = fptr->file_symtab.sym_strs + sym1.st_name;
2420	if ((sym2p = sym_by_addr(&fptr->file_dynsym, addr, &sym2, &i2)) != NULL)
2421		name2 = fptr->file_dynsym.sym_strs + sym2.st_name;
2422
2423	if ((symp = sym_prefer(sym1p, name1, sym2p, name2)) == NULL)
2424		return (-1);
2425
2426	name = (symp == sym1p) ? name1 : name2;
2427	if (bufsize > 0) {
2428		(void) strncpy(sym_name_buffer, name, bufsize);
2429		sym_name_buffer[bufsize - 1] = '\0';
2430	}
2431
2432	*symbolp = *symp;
2433	if (sip != NULL) {
2434		sip->prs_name = bufsize == 0 ? NULL : sym_name_buffer;
2435		if (lmresolve && (fptr->file_rname != NULL))
2436			sip->prs_object = fptr->file_rbase;
2437		else
2438			sip->prs_object = fptr->file_lbase;
2439		sip->prs_id = (symp == sym1p) ? i1 : i2;
2440		sip->prs_table = (symp == sym1p) ? PR_SYMTAB : PR_DYNSYM;
2441		sip->prs_lmid = (fptr->file_lo == NULL) ? LM_ID_BASE :
2442		    fptr->file_lo->rl_lmident;
2443	}
2444
2445	if (GELF_ST_TYPE(symbolp->st_info) != STT_TLS)
2446		symbolp->st_value += fptr->file_dyn_base;
2447
2448	return (0);
2449}
2450
2451int
2452Pxlookup_by_addr(struct ps_prochandle *P, uintptr_t addr, char *buf,
2453    size_t bufsize, GElf_Sym *symp, prsyminfo_t *sip)
2454{
2455	return (i_Pxlookup_by_addr(P, B_FALSE, addr, buf, bufsize, symp, sip));
2456}
2457
2458int
2459Pxlookup_by_addr_resolved(struct ps_prochandle *P, uintptr_t addr, char *buf,
2460    size_t bufsize, GElf_Sym *symp, prsyminfo_t *sip)
2461{
2462	return (i_Pxlookup_by_addr(P, B_TRUE, addr, buf, bufsize, symp, sip));
2463}
2464
2465int
2466Plookup_by_addr(struct ps_prochandle *P, uintptr_t addr, char *buf,
2467    size_t size, GElf_Sym *symp)
2468{
2469	return (i_Pxlookup_by_addr(P, B_FALSE, addr, buf, size, symp, NULL));
2470}
2471
2472/*
2473 * Search the process symbol tables looking for a symbol whose name matches the
2474 * specified name and whose object and link map optionally match the specified
2475 * parameters.  On success, the function returns 0 and fills in the GElf_Sym
2476 * symbol table entry.  On failure, -1 is returned.
2477 */
2478int
2479Pxlookup_by_name(
2480	struct ps_prochandle *P,
2481	Lmid_t lmid,			/* link map to match, or -1 for any */
2482	const char *oname,		/* load object name */
2483	const char *sname,		/* symbol name */
2484	GElf_Sym *symp,			/* returned symbol table entry */
2485	prsyminfo_t *sip)		/* returned symbol info */
2486{
2487	map_info_t *mptr;
2488	file_info_t *fptr;
2489	int cnt;
2490
2491	GElf_Sym sym;
2492	prsyminfo_t si;
2493	int rv = -1;
2494	uint_t id;
2495
2496	if (oname == PR_OBJ_EVERY) {
2497		/* create all the file_info_t's for all the mappings */
2498		(void) Prd_agent(P);
2499		cnt = P->num_files;
2500		fptr = list_next(&P->file_head);
2501	} else {
2502		cnt = 1;
2503		if ((mptr = object_name_to_map(P, lmid, oname)) == NULL ||
2504		    (fptr = build_map_symtab(P, mptr)) == NULL)
2505			return (-1);
2506	}
2507
2508	/*
2509	 * Iterate through the loaded object files and look for the symbol
2510	 * name in the .symtab and .dynsym of each.  If we encounter a match
2511	 * with SHN_UNDEF, keep looking in hopes of finding a better match.
2512	 * This means that a name such as "puts" will match the puts function
2513	 * in libc instead of matching the puts PLT entry in the a.out file.
2514	 */
2515	for (; cnt > 0; cnt--, fptr = list_next(fptr)) {
2516		Pbuild_file_symtab(P, fptr);
2517
2518		if (fptr->file_elf == NULL)
2519			continue;
2520
2521		if (lmid != PR_LMID_EVERY && fptr->file_lo != NULL &&
2522		    lmid != fptr->file_lo->rl_lmident)
2523			continue;
2524
2525		if (fptr->file_symtab.sym_data_pri != NULL &&
2526		    sym_by_name(&fptr->file_symtab, sname, symp, &id)) {
2527			if (sip != NULL) {
2528				sip->prs_id = id;
2529				sip->prs_table = PR_SYMTAB;
2530				sip->prs_object = oname;
2531				sip->prs_name = sname;
2532				sip->prs_lmid = fptr->file_lo == NULL ?
2533				    LM_ID_BASE : fptr->file_lo->rl_lmident;
2534			}
2535		} else if (fptr->file_dynsym.sym_data_pri != NULL &&
2536		    sym_by_name(&fptr->file_dynsym, sname, symp, &id)) {
2537			if (sip != NULL) {
2538				sip->prs_id = id;
2539				sip->prs_table = PR_DYNSYM;
2540				sip->prs_object = oname;
2541				sip->prs_name = sname;
2542				sip->prs_lmid = fptr->file_lo == NULL ?
2543				    LM_ID_BASE : fptr->file_lo->rl_lmident;
2544			}
2545		} else {
2546			continue;
2547		}
2548
2549		if (GELF_ST_TYPE(symp->st_info) != STT_TLS)
2550			symp->st_value += fptr->file_dyn_base;
2551
2552		if (symp->st_shndx != SHN_UNDEF)
2553			return (0);
2554
2555		if (rv != 0) {
2556			if (sip != NULL)
2557				si = *sip;
2558			sym = *symp;
2559			rv = 0;
2560		}
2561	}
2562
2563	if (rv == 0) {
2564		if (sip != NULL)
2565			*sip = si;
2566		*symp = sym;
2567	}
2568
2569	return (rv);
2570}
2571
2572/*
2573 * Search the process symbol tables looking for a symbol whose name matches the
2574 * specified name, but without any restriction on the link map id.
2575 */
2576int
2577Plookup_by_name(struct ps_prochandle *P, const char *object,
2578	const char *symbol, GElf_Sym *symp)
2579{
2580	return (Pxlookup_by_name(P, PR_LMID_EVERY, object, symbol, symp, NULL));
2581}
2582
2583/*
2584 * Iterate over the process's address space mappings.
2585 */
2586static int
2587i_Pmapping_iter(struct ps_prochandle *P, boolean_t lmresolve,
2588    proc_map_f *func, void *cd)
2589{
2590	map_info_t *mptr;
2591	file_info_t *fptr;
2592	char *object_name;
2593	int rc = 0;
2594	int i;
2595
2596	/* create all the file_info_t's for all the mappings */
2597	(void) Prd_agent(P);
2598
2599	for (i = 0, mptr = P->mappings; i < P->map_count; i++, mptr++) {
2600		if ((fptr = mptr->map_file) == NULL)
2601			object_name = NULL;
2602		else if (lmresolve && (fptr->file_rname != NULL))
2603			object_name = fptr->file_rname;
2604		else
2605			object_name = fptr->file_lname;
2606		if ((rc = func(cd, &mptr->map_pmap, object_name)) != 0)
2607			return (rc);
2608	}
2609	return (0);
2610}
2611
2612int
2613Pmapping_iter(struct ps_prochandle *P, proc_map_f *func, void *cd)
2614{
2615	return (i_Pmapping_iter(P, B_FALSE, func, cd));
2616}
2617
2618int
2619Pmapping_iter_resolved(struct ps_prochandle *P, proc_map_f *func, void *cd)
2620{
2621	return (i_Pmapping_iter(P, B_TRUE, func, cd));
2622}
2623
2624/*
2625 * Iterate over the process's mapped objects.
2626 */
2627static int
2628i_Pobject_iter(struct ps_prochandle *P, boolean_t lmresolve,
2629    proc_map_f *func, void *cd)
2630{
2631	map_info_t *mptr;
2632	file_info_t *fptr;
2633	uint_t cnt;
2634	int rc = 0;
2635
2636	(void) Prd_agent(P); /* create file_info_t's for all the mappings */
2637	Pupdate_maps(P);
2638
2639	for (cnt = P->num_files, fptr = list_next(&P->file_head);
2640	    cnt; cnt--, fptr = list_next(fptr)) {
2641		const char *lname;
2642
2643		if (lmresolve && (fptr->file_rname != NULL))
2644			lname = fptr->file_rname;
2645		else if (fptr->file_lname != NULL)
2646			lname = fptr->file_lname;
2647		else
2648			lname = "";
2649
2650		if ((mptr = fptr->file_map) == NULL)
2651			continue;
2652
2653		if ((rc = func(cd, &mptr->map_pmap, lname)) != 0)
2654			return (rc);
2655
2656		if (!P->info_valid)
2657			Pupdate_maps(P);
2658	}
2659	return (0);
2660}
2661
2662int
2663Pobject_iter(struct ps_prochandle *P, proc_map_f *func, void *cd)
2664{
2665	return (i_Pobject_iter(P, B_FALSE, func, cd));
2666}
2667
2668int
2669Pobject_iter_resolved(struct ps_prochandle *P, proc_map_f *func, void *cd)
2670{
2671	return (i_Pobject_iter(P, B_TRUE, func, cd));
2672}
2673
2674static char *
2675i_Pobjname(struct ps_prochandle *P, boolean_t lmresolve, uintptr_t addr,
2676	char *buffer, size_t bufsize)
2677{
2678	map_info_t *mptr;
2679	file_info_t *fptr;
2680
2681	/* create all the file_info_t's for all the mappings */
2682	(void) Prd_agent(P);
2683
2684	if ((mptr = Paddr2mptr(P, addr)) == NULL)
2685		return (NULL);
2686
2687	if (!lmresolve) {
2688		if (((fptr = mptr->map_file) == NULL) ||
2689		    (fptr->file_lname == NULL))
2690			return (NULL);
2691		(void) strlcpy(buffer, fptr->file_lname, bufsize);
2692		return (buffer);
2693	}
2694
2695	/* Check for a cached copy of the resolved path */
2696	if (Pfindmap(P, mptr, buffer, bufsize) != NULL)
2697		return (buffer);
2698
2699	return (NULL);
2700}
2701
2702/*
2703 * Given a virtual address, return the name of the underlying
2704 * mapped object (file) as provided by the dynamic linker.
2705 * Return NULL if we can't find any name information for the object.
2706 */
2707char *
2708Pobjname(struct ps_prochandle *P, uintptr_t addr,
2709	char *buffer, size_t bufsize)
2710{
2711	return (i_Pobjname(P, B_FALSE, addr, buffer, bufsize));
2712}
2713
2714/*
2715 * Given a virtual address, try to return a filesystem path to the
2716 * underlying mapped object (file).  If we're in the global zone,
2717 * this path could resolve to an object in another zone.  If we're
2718 * unable return a valid filesystem path, we'll fall back to providing
2719 * the mapped object (file) name provided by the dynamic linker in
2720 * the target process (ie, the object reported by Pobjname()).
2721 */
2722char *
2723Pobjname_resolved(struct ps_prochandle *P, uintptr_t addr,
2724	char *buffer, size_t bufsize)
2725{
2726	return (i_Pobjname(P, B_TRUE, addr, buffer, bufsize));
2727}
2728
2729/*
2730 * Given a virtual address, return the link map id of the underlying mapped
2731 * object (file), as provided by the dynamic linker.  Return -1 on failure.
2732 */
2733int
2734Plmid(struct ps_prochandle *P, uintptr_t addr, Lmid_t *lmidp)
2735{
2736	map_info_t *mptr;
2737	file_info_t *fptr;
2738
2739	/* create all the file_info_t's for all the mappings */
2740	(void) Prd_agent(P);
2741
2742	if ((mptr = Paddr2mptr(P, addr)) != NULL &&
2743	    (fptr = mptr->map_file) != NULL && fptr->file_lo != NULL) {
2744		*lmidp = fptr->file_lo->rl_lmident;
2745		return (0);
2746	}
2747
2748	return (-1);
2749}
2750
2751/*
2752 * Given an object name and optional lmid, iterate over the object's symbols.
2753 * If which == PR_SYMTAB, search the normal symbol table.
2754 * If which == PR_DYNSYM, search the dynamic symbol table.
2755 */
2756static int
2757Psymbol_iter_com(struct ps_prochandle *P, Lmid_t lmid, const char *object_name,
2758    int which, int mask, pr_order_t order, proc_xsym_f *func, void *cd)
2759{
2760#if STT_NUM != (STT_TLS + 1)
2761#error "STT_NUM has grown. update Psymbol_iter_com()"
2762#endif
2763
2764	GElf_Sym sym;
2765	GElf_Shdr shdr;
2766	map_info_t *mptr;
2767	file_info_t *fptr;
2768	sym_tbl_t *symtab;
2769	size_t symn;
2770	const char *strs;
2771	size_t strsz;
2772	prsyminfo_t si;
2773	int rv;
2774	uint_t *map, i, count, ndx;
2775
2776	if ((mptr = object_name_to_map(P, lmid, object_name)) == NULL)
2777		return (-1);
2778
2779	if ((fptr = build_map_symtab(P, mptr)) == NULL || /* no mapped file */
2780	    fptr->file_elf == NULL)			/* not an ELF file */
2781		return (-1);
2782
2783	/*
2784	 * Search the specified symbol table.
2785	 */
2786	switch (which) {
2787	case PR_SYMTAB:
2788		symtab = &fptr->file_symtab;
2789		si.prs_table = PR_SYMTAB;
2790		break;
2791	case PR_DYNSYM:
2792		symtab = &fptr->file_dynsym;
2793		si.prs_table = PR_DYNSYM;
2794		break;
2795	default:
2796		return (-1);
2797	}
2798
2799	si.prs_object = object_name;
2800	si.prs_lmid = fptr->file_lo == NULL ?
2801	    LM_ID_BASE : fptr->file_lo->rl_lmident;
2802
2803	symn = symtab->sym_symn;
2804	strs = symtab->sym_strs;
2805	strsz = symtab->sym_strsz;
2806
2807	switch (order) {
2808	case PRO_NATURAL:
2809		map = NULL;
2810		count = symn;
2811		break;
2812	case PRO_BYNAME:
2813		map = symtab->sym_byname;
2814		count = symtab->sym_count;
2815		break;
2816	case PRO_BYADDR:
2817		map = symtab->sym_byaddr;
2818		count = symtab->sym_count;
2819		break;
2820	default:
2821		return (-1);
2822	}
2823
2824	if (symtab->sym_data_pri == NULL || strs == NULL || count == 0)
2825		return (-1);
2826
2827	rv = 0;
2828
2829	for (i = 0; i < count; i++) {
2830		ndx = map == NULL ? i : map[i];
2831		if (symtab_getsym(symtab, ndx, &sym) != NULL) {
2832			uint_t s_bind, s_type, type;
2833
2834			if (sym.st_name >= strsz)	/* invalid st_name */
2835				continue;
2836
2837			s_bind = GELF_ST_BIND(sym.st_info);
2838			s_type = GELF_ST_TYPE(sym.st_info);
2839
2840			/*
2841			 * In case you haven't already guessed, this relies on
2842			 * the bitmask used in <libproc.h> for encoding symbol
2843			 * type and binding matching the order of STB and STT
2844			 * constants in <sys/elf.h>.  Changes to ELF must
2845			 * maintain binary compatibility, so I think this is
2846			 * reasonably fair game.
2847			 */
2848			if (s_bind < STB_NUM && s_type < STT_NUM) {
2849				type = (1 << (s_type + 8)) | (1 << s_bind);
2850				if ((type & ~mask) != 0)
2851					continue;
2852			} else
2853				continue; /* Invalid type or binding */
2854
2855			if (GELF_ST_TYPE(sym.st_info) != STT_TLS)
2856				sym.st_value += fptr->file_dyn_base;
2857
2858			si.prs_name = strs + sym.st_name;
2859
2860			/*
2861			 * If symbol's type is STT_SECTION, then try to lookup
2862			 * the name of the corresponding section.
2863			 */
2864			if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
2865			    fptr->file_shstrs != NULL &&
2866			    gelf_getshdr(elf_getscn(fptr->file_elf,
2867			    sym.st_shndx), &shdr) != NULL &&
2868			    shdr.sh_name != 0 &&
2869			    shdr.sh_name < fptr->file_shstrsz)
2870				si.prs_name = fptr->file_shstrs + shdr.sh_name;
2871
2872			si.prs_id = ndx;
2873			if ((rv = func(cd, &sym, si.prs_name, &si)) != 0)
2874				break;
2875		}
2876	}
2877
2878	return (rv);
2879}
2880
2881int
2882Pxsymbol_iter(struct ps_prochandle *P, Lmid_t lmid, const char *object_name,
2883    int which, int mask, proc_xsym_f *func, void *cd)
2884{
2885	return (Psymbol_iter_com(P, lmid, object_name, which, mask,
2886	    PRO_NATURAL, func, cd));
2887}
2888
2889int
2890Psymbol_iter_by_lmid(struct ps_prochandle *P, Lmid_t lmid,
2891    const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
2892{
2893	return (Psymbol_iter_com(P, lmid, object_name, which, mask,
2894	    PRO_NATURAL, (proc_xsym_f *)func, cd));
2895}
2896
2897int
2898Psymbol_iter(struct ps_prochandle *P,
2899    const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
2900{
2901	return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
2902	    PRO_NATURAL, (proc_xsym_f *)func, cd));
2903}
2904
2905int
2906Psymbol_iter_by_addr(struct ps_prochandle *P,
2907    const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
2908{
2909	return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
2910	    PRO_BYADDR, (proc_xsym_f *)func, cd));
2911}
2912
2913int
2914Psymbol_iter_by_name(struct ps_prochandle *P,
2915    const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
2916{
2917	return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
2918	    PRO_BYNAME, (proc_xsym_f *)func, cd));
2919}
2920
2921/*
2922 * Get the platform string from the core file if we have it;
2923 * just perform the system call for the caller if this is a live process.
2924 */
2925char *
2926Pplatform(struct ps_prochandle *P, char *s, size_t n)
2927{
2928	if (P->state == PS_IDLE) {
2929		errno = ENODATA;
2930		return (NULL);
2931	}
2932
2933	if (P->state == PS_DEAD) {
2934		if (P->core->core_platform == NULL) {
2935			errno = ENODATA;
2936			return (NULL);
2937		}
2938		(void) strncpy(s, P->core->core_platform, n - 1);
2939		s[n - 1] = '\0';
2940
2941	} else if (sysinfo(SI_PLATFORM, s, n) == -1)
2942		return (NULL);
2943
2944	return (s);
2945}
2946
2947/*
2948 * Get the uname(2) information from the core file if we have it;
2949 * just perform the system call for the caller if this is a live process.
2950 */
2951int
2952Puname(struct ps_prochandle *P, struct utsname *u)
2953{
2954	if (P->state == PS_IDLE) {
2955		errno = ENODATA;
2956		return (-1);
2957	}
2958
2959	if (P->state == PS_DEAD) {
2960		if (P->core->core_uts == NULL) {
2961			errno = ENODATA;
2962			return (-1);
2963		}
2964		(void) memcpy(u, P->core->core_uts, sizeof (struct utsname));
2965		return (0);
2966	}
2967	return (uname(u));
2968}
2969
2970/*
2971 * Called from Pcreate(), Pgrab(), and Pfgrab_core() to initialize
2972 * the symbol table heads in the new ps_prochandle.
2973 */
2974void
2975Pinitsym(struct ps_prochandle *P)
2976{
2977	P->num_files = 0;
2978	list_link(&P->file_head, NULL);
2979}
2980
2981/*
2982 * Called from Prelease() to destroy the symbol tables.
2983 * Must be called by the client after an exec() in the victim process.
2984 */
2985void
2986Preset_maps(struct ps_prochandle *P)
2987{
2988	int i;
2989
2990	if (P->rap != NULL) {
2991		rd_delete(P->rap);
2992		P->rap = NULL;
2993	}
2994
2995	if (P->execname != NULL) {
2996		free(P->execname);
2997		P->execname = NULL;
2998	}
2999
3000	if (P->auxv != NULL) {
3001		free(P->auxv);
3002		P->auxv = NULL;
3003		P->nauxv = 0;
3004	}
3005
3006	for (i = 0; i < P->map_count; i++)
3007		map_info_free(P, &P->mappings[i]);
3008
3009	if (P->mappings != NULL) {
3010		free(P->mappings);
3011		P->mappings = NULL;
3012	}
3013	P->map_count = P->map_alloc = 0;
3014
3015	P->info_valid = 0;
3016}
3017
3018typedef struct getenv_data {
3019	char *buf;
3020	size_t bufsize;
3021	const char *search;
3022	size_t searchlen;
3023} getenv_data_t;
3024
3025/*ARGSUSED*/
3026static int
3027getenv_func(void *data, struct ps_prochandle *P, uintptr_t addr,
3028    const char *nameval)
3029{
3030	getenv_data_t *d = data;
3031	size_t len;
3032
3033	if (nameval == NULL)
3034		return (0);
3035
3036	if (d->searchlen < strlen(nameval) &&
3037	    strncmp(nameval, d->search, d->searchlen) == 0 &&
3038	    nameval[d->searchlen] == '=') {
3039		len = MIN(strlen(nameval), d->bufsize - 1);
3040		(void) strncpy(d->buf, nameval, len);
3041		d->buf[len] = '\0';
3042		return (1);
3043	}
3044
3045	return (0);
3046}
3047
3048char *
3049Pgetenv(struct ps_prochandle *P, const char *name, char *buf, size_t buflen)
3050{
3051	getenv_data_t d;
3052
3053	d.buf = buf;
3054	d.bufsize = buflen;
3055	d.search = name;
3056	d.searchlen = strlen(name);
3057
3058	if (Penv_iter(P, getenv_func, &d) == 1) {
3059		char *equals = strchr(d.buf, '=');
3060
3061		if (equals != NULL) {
3062			(void) memmove(d.buf, equals + 1,
3063			    d.buf + buflen - equals - 1);
3064			d.buf[d.buf + buflen - equals] = '\0';
3065
3066			return (buf);
3067		}
3068	}
3069
3070	return (NULL);
3071}
3072
3073/* number of argument or environment pointers to read all at once */
3074#define	NARG	100
3075
3076int
3077Penv_iter(struct ps_prochandle *P, proc_env_f *func, void *data)
3078{
3079	const psinfo_t *psp;
3080	uintptr_t envpoff;
3081	GElf_Sym sym;
3082	int ret;
3083	char *buf, *nameval;
3084	size_t buflen;
3085
3086	int nenv = NARG;
3087	long envp[NARG];
3088
3089	/*
3090	 * Attempt to find the "_environ" variable in the process.
3091	 * Failing that, use the original value provided by Ppsinfo().
3092	 */
3093	if ((psp = Ppsinfo(P)) == NULL)
3094		return (-1);
3095
3096	envpoff = psp->pr_envp; /* Default if no _environ found */
3097
3098	if (Plookup_by_name(P, PR_OBJ_EXEC, "_environ", &sym) == 0) {
3099		if (P->status.pr_dmodel == PR_MODEL_NATIVE) {
3100			if (Pread(P, &envpoff, sizeof (envpoff),
3101			    sym.st_value) != sizeof (envpoff))
3102				envpoff = psp->pr_envp;
3103		} else if (P->status.pr_dmodel == PR_MODEL_ILP32) {
3104			uint32_t envpoff32;
3105
3106			if (Pread(P, &envpoff32, sizeof (envpoff32),
3107			    sym.st_value) != sizeof (envpoff32))
3108				envpoff = psp->pr_envp;
3109			else
3110				envpoff = envpoff32;
3111		}
3112	}
3113
3114	buflen = 128;
3115	buf = malloc(buflen);
3116
3117	ret = 0;
3118	for (;;) {
3119		uintptr_t envoff;
3120
3121		if (nenv == NARG) {
3122			(void) memset(envp, 0, sizeof (envp));
3123			if (P->status.pr_dmodel == PR_MODEL_NATIVE) {
3124				if (Pread(P, envp,
3125				    sizeof (envp), envpoff) <= 0) {
3126					ret = -1;
3127					break;
3128				}
3129			} else if (P->status.pr_dmodel == PR_MODEL_ILP32) {
3130				uint32_t e32[NARG];
3131				int i;
3132
3133				(void) memset(e32, 0, sizeof (e32));
3134				if (Pread(P, e32, sizeof (e32), envpoff) <= 0) {
3135					ret = -1;
3136					break;
3137				}
3138				for (i = 0; i < NARG; i++)
3139					envp[i] = e32[i];
3140			}
3141			nenv = 0;
3142		}
3143
3144		if ((envoff = envp[nenv++]) == NULL)
3145			break;
3146
3147		/*
3148		 * Attempt to read the string from the process.
3149		 */
3150again:
3151		ret = Pread_string(P, buf, buflen, envoff);
3152
3153		if (ret <= 0) {
3154			nameval = NULL;
3155		} else if (ret == buflen - 1) {
3156			free(buf);
3157			/*
3158			 * Bail if we have a corrupted environment
3159			 */
3160			if (buflen >= ARG_MAX)
3161				return (-1);
3162			buflen *= 2;
3163			buf = malloc(buflen);
3164			goto again;
3165		} else {
3166			nameval = buf;
3167		}
3168
3169		if ((ret = func(data, P, envoff, nameval)) != 0)
3170			break;
3171
3172		envpoff += (P->status.pr_dmodel == PR_MODEL_LP64)? 8 : 4;
3173	}
3174
3175	free(buf);
3176
3177	return (ret);
3178}
3179