dt_module.c revision 297953
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25/*
26 * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
27 * Copyright (c) 2016, Pedro Giffuni.  All rights reserved.
28 */
29
30#include <sys/types.h>
31#ifdef illumos
32#include <sys/modctl.h>
33#include <sys/kobj.h>
34#include <sys/kobj_impl.h>
35#include <sys/sysmacros.h>
36#include <sys/elf.h>
37#include <sys/task.h>
38#else
39#include <sys/param.h>
40#include <sys/linker.h>
41#include <sys/stat.h>
42#endif
43
44#include <unistd.h>
45#ifdef illumos
46#include <project.h>
47#endif
48#include <strings.h>
49#include <stdlib.h>
50#include <libelf.h>
51#include <limits.h>
52#include <assert.h>
53#include <errno.h>
54#include <dirent.h>
55#ifndef illumos
56#include <fcntl.h>
57#include <libproc_compat.h>
58#endif
59
60#include <dt_strtab.h>
61#include <dt_module.h>
62#include <dt_impl.h>
63
64static const char *dt_module_strtab; /* active strtab for qsort callbacks */
65
66static void
67dt_module_symhash_insert(dt_module_t *dmp, const char *name, uint_t id)
68{
69	dt_sym_t *dsp = &dmp->dm_symchains[dmp->dm_symfree];
70	uint_t h;
71
72	assert(dmp->dm_symfree < dmp->dm_nsymelems + 1);
73
74	dsp->ds_symid = id;
75	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;
76	dsp->ds_next = dmp->dm_symbuckets[h];
77	dmp->dm_symbuckets[h] = dmp->dm_symfree++;
78}
79
80static uint_t
81dt_module_syminit32(dt_module_t *dmp)
82{
83#if STT_NUM != (STT_TLS + 1)
84#error "STT_NUM has grown. update dt_module_syminit32()"
85#endif
86
87	Elf32_Sym *sym = dmp->dm_symtab.cts_data;
88	const char *base = dmp->dm_strtab.cts_data;
89	size_t ss_size = dmp->dm_strtab.cts_size;
90	uint_t i, n = dmp->dm_nsymelems;
91	uint_t asrsv = 0;
92
93#if defined(__FreeBSD__)
94	GElf_Ehdr ehdr;
95	int is_elf_obj;
96
97	gelf_getehdr(dmp->dm_elf, &ehdr);
98	is_elf_obj = (ehdr.e_type == ET_REL);
99#endif
100
101	for (i = 0; i < n; i++, sym++) {
102		const char *name = base + sym->st_name;
103		uchar_t type = ELF32_ST_TYPE(sym->st_info);
104
105		if (type >= STT_NUM || type == STT_SECTION)
106			continue; /* skip sections and unknown types */
107
108		if (sym->st_name == 0 || sym->st_name >= ss_size)
109			continue; /* skip null or invalid names */
110
111		if (sym->st_value != 0 &&
112		    (ELF32_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size)) {
113			asrsv++; /* reserve space in the address map */
114
115#if defined(__FreeBSD__)
116			sym->st_value += (Elf_Addr) dmp->dm_reloc_offset;
117			if (is_elf_obj && sym->st_shndx != SHN_UNDEF &&
118			    sym->st_shndx < ehdr.e_shnum)
119				sym->st_value +=
120				    dmp->dm_sec_offsets[sym->st_shndx];
121#endif
122		}
123
124		dt_module_symhash_insert(dmp, name, i);
125	}
126
127	return (asrsv);
128}
129
130static uint_t
131dt_module_syminit64(dt_module_t *dmp)
132{
133#if STT_NUM != (STT_TLS + 1)
134#error "STT_NUM has grown. update dt_module_syminit64()"
135#endif
136
137	Elf64_Sym *sym = dmp->dm_symtab.cts_data;
138	const char *base = dmp->dm_strtab.cts_data;
139	size_t ss_size = dmp->dm_strtab.cts_size;
140	uint_t i, n = dmp->dm_nsymelems;
141	uint_t asrsv = 0;
142
143#if defined(__FreeBSD__)
144	GElf_Ehdr ehdr;
145	int is_elf_obj;
146
147	gelf_getehdr(dmp->dm_elf, &ehdr);
148	is_elf_obj = (ehdr.e_type == ET_REL);
149#endif
150
151	for (i = 0; i < n; i++, sym++) {
152		const char *name = base + sym->st_name;
153		uchar_t type = ELF64_ST_TYPE(sym->st_info);
154
155		if (type >= STT_NUM || type == STT_SECTION)
156			continue; /* skip sections and unknown types */
157
158		if (sym->st_name == 0 || sym->st_name >= ss_size)
159			continue; /* skip null or invalid names */
160
161		if (sym->st_value != 0 &&
162		    (ELF64_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size)) {
163			asrsv++; /* reserve space in the address map */
164#if defined(__FreeBSD__)
165			sym->st_value += (Elf_Addr) dmp->dm_reloc_offset;
166			if (is_elf_obj && sym->st_shndx != SHN_UNDEF &&
167			    sym->st_shndx < ehdr.e_shnum)
168				sym->st_value +=
169				    dmp->dm_sec_offsets[sym->st_shndx];
170#endif
171		}
172
173		dt_module_symhash_insert(dmp, name, i);
174	}
175
176	return (asrsv);
177}
178
179/*
180 * Sort comparison function for 32-bit symbol address-to-name lookups.  We sort
181 * symbols by value.  If values are equal, we prefer the symbol that is
182 * non-zero sized, typed, not weak, or lexically first, in that order.
183 */
184static int
185dt_module_symcomp32(const void *lp, const void *rp)
186{
187	Elf32_Sym *lhs = *((Elf32_Sym **)lp);
188	Elf32_Sym *rhs = *((Elf32_Sym **)rp);
189
190	if (lhs->st_value != rhs->st_value)
191		return (lhs->st_value > rhs->st_value ? 1 : -1);
192
193	if ((lhs->st_size == 0) != (rhs->st_size == 0))
194		return (lhs->st_size == 0 ? 1 : -1);
195
196	if ((ELF32_ST_TYPE(lhs->st_info) == STT_NOTYPE) !=
197	    (ELF32_ST_TYPE(rhs->st_info) == STT_NOTYPE))
198		return (ELF32_ST_TYPE(lhs->st_info) == STT_NOTYPE ? 1 : -1);
199
200	if ((ELF32_ST_BIND(lhs->st_info) == STB_WEAK) !=
201	    (ELF32_ST_BIND(rhs->st_info) == STB_WEAK))
202		return (ELF32_ST_BIND(lhs->st_info) == STB_WEAK ? 1 : -1);
203
204	return (strcmp(dt_module_strtab + lhs->st_name,
205	    dt_module_strtab + rhs->st_name));
206}
207
208/*
209 * Sort comparison function for 64-bit symbol address-to-name lookups.  We sort
210 * symbols by value.  If values are equal, we prefer the symbol that is
211 * non-zero sized, typed, not weak, or lexically first, in that order.
212 */
213static int
214dt_module_symcomp64(const void *lp, const void *rp)
215{
216	Elf64_Sym *lhs = *((Elf64_Sym **)lp);
217	Elf64_Sym *rhs = *((Elf64_Sym **)rp);
218
219	if (lhs->st_value != rhs->st_value)
220		return (lhs->st_value > rhs->st_value ? 1 : -1);
221
222	if ((lhs->st_size == 0) != (rhs->st_size == 0))
223		return (lhs->st_size == 0 ? 1 : -1);
224
225	if ((ELF64_ST_TYPE(lhs->st_info) == STT_NOTYPE) !=
226	    (ELF64_ST_TYPE(rhs->st_info) == STT_NOTYPE))
227		return (ELF64_ST_TYPE(lhs->st_info) == STT_NOTYPE ? 1 : -1);
228
229	if ((ELF64_ST_BIND(lhs->st_info) == STB_WEAK) !=
230	    (ELF64_ST_BIND(rhs->st_info) == STB_WEAK))
231		return (ELF64_ST_BIND(lhs->st_info) == STB_WEAK ? 1 : -1);
232
233	return (strcmp(dt_module_strtab + lhs->st_name,
234	    dt_module_strtab + rhs->st_name));
235}
236
237static void
238dt_module_symsort32(dt_module_t *dmp)
239{
240	Elf32_Sym *symtab = (Elf32_Sym *)dmp->dm_symtab.cts_data;
241	Elf32_Sym **sympp = (Elf32_Sym **)dmp->dm_asmap;
242	const dt_sym_t *dsp = dmp->dm_symchains + 1;
243	uint_t i, n = dmp->dm_symfree;
244
245	for (i = 1; i < n; i++, dsp++) {
246		Elf32_Sym *sym = symtab + dsp->ds_symid;
247		if (sym->st_value != 0 &&
248		    (ELF32_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
249			*sympp++ = sym;
250	}
251
252	dmp->dm_aslen = (uint_t)(sympp - (Elf32_Sym **)dmp->dm_asmap);
253	assert(dmp->dm_aslen <= dmp->dm_asrsv);
254
255	dt_module_strtab = dmp->dm_strtab.cts_data;
256	qsort(dmp->dm_asmap, dmp->dm_aslen,
257	    sizeof (Elf32_Sym *), dt_module_symcomp32);
258	dt_module_strtab = NULL;
259}
260
261static void
262dt_module_symsort64(dt_module_t *dmp)
263{
264	Elf64_Sym *symtab = (Elf64_Sym *)dmp->dm_symtab.cts_data;
265	Elf64_Sym **sympp = (Elf64_Sym **)dmp->dm_asmap;
266	const dt_sym_t *dsp = dmp->dm_symchains + 1;
267	uint_t i, n = dmp->dm_symfree;
268
269	for (i = 1; i < n; i++, dsp++) {
270		Elf64_Sym *sym = symtab + dsp->ds_symid;
271		if (sym->st_value != 0 &&
272		    (ELF64_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
273			*sympp++ = sym;
274	}
275
276	dmp->dm_aslen = (uint_t)(sympp - (Elf64_Sym **)dmp->dm_asmap);
277	assert(dmp->dm_aslen <= dmp->dm_asrsv);
278
279	dt_module_strtab = dmp->dm_strtab.cts_data;
280	qsort(dmp->dm_asmap, dmp->dm_aslen,
281	    sizeof (Elf64_Sym *), dt_module_symcomp64);
282	dt_module_strtab = NULL;
283}
284
285static GElf_Sym *
286dt_module_symgelf32(const Elf32_Sym *src, GElf_Sym *dst)
287{
288	if (dst != NULL) {
289		dst->st_name = src->st_name;
290		dst->st_info = src->st_info;
291		dst->st_other = src->st_other;
292		dst->st_shndx = src->st_shndx;
293		dst->st_value = src->st_value;
294		dst->st_size = src->st_size;
295	}
296
297	return (dst);
298}
299
300static GElf_Sym *
301dt_module_symgelf64(const Elf64_Sym *src, GElf_Sym *dst)
302{
303	if (dst != NULL)
304		bcopy(src, dst, sizeof (GElf_Sym));
305
306	return (dst);
307}
308
309static GElf_Sym *
310dt_module_symname32(dt_module_t *dmp, const char *name,
311    GElf_Sym *symp, uint_t *idp)
312{
313	const Elf32_Sym *symtab = dmp->dm_symtab.cts_data;
314	const char *strtab = dmp->dm_strtab.cts_data;
315
316	const Elf32_Sym *sym;
317	const dt_sym_t *dsp;
318	uint_t i, h;
319
320	if (dmp->dm_nsymelems == 0)
321		return (NULL);
322
323	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;
324
325	for (i = dmp->dm_symbuckets[h]; i != 0; i = dsp->ds_next) {
326		dsp = &dmp->dm_symchains[i];
327		sym = symtab + dsp->ds_symid;
328
329		if (strcmp(name, strtab + sym->st_name) == 0) {
330			if (idp != NULL)
331				*idp = dsp->ds_symid;
332			return (dt_module_symgelf32(sym, symp));
333		}
334	}
335
336	return (NULL);
337}
338
339static GElf_Sym *
340dt_module_symname64(dt_module_t *dmp, const char *name,
341    GElf_Sym *symp, uint_t *idp)
342{
343	const Elf64_Sym *symtab = dmp->dm_symtab.cts_data;
344	const char *strtab = dmp->dm_strtab.cts_data;
345
346	const Elf64_Sym *sym;
347	const dt_sym_t *dsp;
348	uint_t i, h;
349
350	if (dmp->dm_nsymelems == 0)
351		return (NULL);
352
353	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;
354
355	for (i = dmp->dm_symbuckets[h]; i != 0; i = dsp->ds_next) {
356		dsp = &dmp->dm_symchains[i];
357		sym = symtab + dsp->ds_symid;
358
359		if (strcmp(name, strtab + sym->st_name) == 0) {
360			if (idp != NULL)
361				*idp = dsp->ds_symid;
362			return (dt_module_symgelf64(sym, symp));
363		}
364	}
365
366	return (NULL);
367}
368
369static GElf_Sym *
370dt_module_symaddr32(dt_module_t *dmp, GElf_Addr addr,
371    GElf_Sym *symp, uint_t *idp)
372{
373	const Elf32_Sym **asmap = (const Elf32_Sym **)dmp->dm_asmap;
374	const Elf32_Sym *symtab = dmp->dm_symtab.cts_data;
375	const Elf32_Sym *sym;
376
377	uint_t i, mid, lo = 0, hi = dmp->dm_aslen - 1;
378	Elf32_Addr v;
379
380	if (dmp->dm_aslen == 0)
381		return (NULL);
382
383	while (hi - lo > 1) {
384		mid = (lo + hi) / 2;
385		if (addr >= asmap[mid]->st_value)
386			lo = mid;
387		else
388			hi = mid;
389	}
390
391	i = addr < asmap[hi]->st_value ? lo : hi;
392	sym = asmap[i];
393	v = sym->st_value;
394
395	/*
396	 * If the previous entry has the same value, improve our choice.  The
397	 * order of equal-valued symbols is determined by the comparison func.
398	 */
399	while (i-- != 0 && asmap[i]->st_value == v)
400		sym = asmap[i];
401
402	if (addr - sym->st_value < MAX(sym->st_size, 1)) {
403		if (idp != NULL)
404			*idp = (uint_t)(sym - symtab);
405		return (dt_module_symgelf32(sym, symp));
406	}
407
408	return (NULL);
409}
410
411static GElf_Sym *
412dt_module_symaddr64(dt_module_t *dmp, GElf_Addr addr,
413    GElf_Sym *symp, uint_t *idp)
414{
415	const Elf64_Sym **asmap = (const Elf64_Sym **)dmp->dm_asmap;
416	const Elf64_Sym *symtab = dmp->dm_symtab.cts_data;
417	const Elf64_Sym *sym;
418
419	uint_t i, mid, lo = 0, hi = dmp->dm_aslen - 1;
420	Elf64_Addr v;
421
422	if (dmp->dm_aslen == 0)
423		return (NULL);
424
425	while (hi - lo > 1) {
426		mid = (lo + hi) / 2;
427		if (addr >= asmap[mid]->st_value)
428			lo = mid;
429		else
430			hi = mid;
431	}
432
433	i = addr < asmap[hi]->st_value ? lo : hi;
434	sym = asmap[i];
435	v = sym->st_value;
436
437	/*
438	 * If the previous entry has the same value, improve our choice.  The
439	 * order of equal-valued symbols is determined by the comparison func.
440	 */
441	while (i-- != 0 && asmap[i]->st_value == v)
442		sym = asmap[i];
443
444	if (addr - sym->st_value < MAX(sym->st_size, 1)) {
445		if (idp != NULL)
446			*idp = (uint_t)(sym - symtab);
447		return (dt_module_symgelf64(sym, symp));
448	}
449
450	return (NULL);
451}
452
453static const dt_modops_t dt_modops_32 = {
454	dt_module_syminit32,
455	dt_module_symsort32,
456	dt_module_symname32,
457	dt_module_symaddr32
458};
459
460static const dt_modops_t dt_modops_64 = {
461	dt_module_syminit64,
462	dt_module_symsort64,
463	dt_module_symname64,
464	dt_module_symaddr64
465};
466
467dt_module_t *
468dt_module_create(dtrace_hdl_t *dtp, const char *name)
469{
470	long pid;
471	char *eptr;
472	dt_ident_t *idp;
473	uint_t h = dt_strtab_hash(name, NULL) % dtp->dt_modbuckets;
474	dt_module_t *dmp;
475
476	for (dmp = dtp->dt_mods[h]; dmp != NULL; dmp = dmp->dm_next) {
477		if (strcmp(dmp->dm_name, name) == 0)
478			return (dmp);
479	}
480
481	if ((dmp = malloc(sizeof (dt_module_t))) == NULL)
482		return (NULL); /* caller must handle allocation failure */
483
484	bzero(dmp, sizeof (dt_module_t));
485	(void) strlcpy(dmp->dm_name, name, sizeof (dmp->dm_name));
486	dt_list_append(&dtp->dt_modlist, dmp);
487	dmp->dm_next = dtp->dt_mods[h];
488	dtp->dt_mods[h] = dmp;
489	dtp->dt_nmods++;
490
491	if (dtp->dt_conf.dtc_ctfmodel == CTF_MODEL_LP64)
492		dmp->dm_ops = &dt_modops_64;
493	else
494		dmp->dm_ops = &dt_modops_32;
495
496	/*
497	 * Modules for userland processes are special. They always refer to a
498	 * specific process and have a copy of their CTF data from a specific
499	 * instant in time. Any dt_module_t that begins with 'pid' is a module
500	 * for a specific process, much like how any probe description that
501	 * begins with 'pid' is special. pid123 refers to process 123. A module
502	 * that is just 'pid' refers specifically to pid$target. This is
503	 * generally done as D does not currently allow for macros to be
504	 * evaluated when working with types.
505	 */
506	if (strncmp(dmp->dm_name, "pid", 3) == 0) {
507		errno = 0;
508		if (dmp->dm_name[3] == '\0') {
509			idp = dt_idhash_lookup(dtp->dt_macros, "target");
510			if (idp != NULL && idp->di_id != 0)
511				dmp->dm_pid = idp->di_id;
512		} else {
513			pid = strtol(dmp->dm_name + 3, &eptr, 10);
514			if (errno == 0 && *eptr == '\0')
515				dmp->dm_pid = (pid_t)pid;
516			else
517				dt_dprintf("encountered malformed pid "
518				    "module: %s\n", dmp->dm_name);
519		}
520	}
521
522	return (dmp);
523}
524
525dt_module_t *
526dt_module_lookup_by_name(dtrace_hdl_t *dtp, const char *name)
527{
528	uint_t h = dt_strtab_hash(name, NULL) % dtp->dt_modbuckets;
529	dt_module_t *dmp;
530
531	for (dmp = dtp->dt_mods[h]; dmp != NULL; dmp = dmp->dm_next) {
532		if (strcmp(dmp->dm_name, name) == 0)
533			return (dmp);
534	}
535
536	return (NULL);
537}
538
539/*ARGSUSED*/
540dt_module_t *
541dt_module_lookup_by_ctf(dtrace_hdl_t *dtp, ctf_file_t *ctfp)
542{
543	return (ctfp ? ctf_getspecific(ctfp) : NULL);
544}
545
546static int
547dt_module_load_sect(dtrace_hdl_t *dtp, dt_module_t *dmp, ctf_sect_t *ctsp)
548{
549	const char *s;
550	size_t shstrs;
551	GElf_Shdr sh;
552	Elf_Data *dp;
553	Elf_Scn *sp;
554
555	if (elf_getshdrstrndx(dmp->dm_elf, &shstrs) == -1)
556		return (dt_set_errno(dtp, EDT_NOTLOADED));
557
558	for (sp = NULL; (sp = elf_nextscn(dmp->dm_elf, sp)) != NULL; ) {
559		if (gelf_getshdr(sp, &sh) == NULL || sh.sh_type == SHT_NULL ||
560		    (s = elf_strptr(dmp->dm_elf, shstrs, sh.sh_name)) == NULL)
561			continue; /* skip any malformed sections */
562
563		if (sh.sh_type == ctsp->cts_type &&
564		    sh.sh_entsize == ctsp->cts_entsize &&
565		    strcmp(s, ctsp->cts_name) == 0)
566			break; /* section matches specification */
567	}
568
569	/*
570	 * If the section isn't found, return success but leave cts_data set
571	 * to NULL and cts_size set to zero for our caller.
572	 */
573	if (sp == NULL || (dp = elf_getdata(sp, NULL)) == NULL)
574		return (0);
575
576#ifdef illumos
577	ctsp->cts_data = dp->d_buf;
578#else
579	if ((ctsp->cts_data = malloc(dp->d_size)) == NULL)
580		return (0);
581	memcpy(ctsp->cts_data, dp->d_buf, dp->d_size);
582#endif
583	ctsp->cts_size = dp->d_size;
584
585	dt_dprintf("loaded %s [%s] (%lu bytes)\n",
586	    dmp->dm_name, ctsp->cts_name, (ulong_t)ctsp->cts_size);
587
588	return (0);
589}
590
591typedef struct dt_module_cb_arg {
592	struct ps_prochandle *dpa_proc;
593	dtrace_hdl_t *dpa_dtp;
594	dt_module_t *dpa_dmp;
595	uint_t dpa_count;
596} dt_module_cb_arg_t;
597
598/* ARGSUSED */
599static int
600dt_module_load_proc_count(void *arg, const prmap_t *prmap, const char *obj)
601{
602	ctf_file_t *fp;
603	dt_module_cb_arg_t *dcp = arg;
604
605	/* Try to grab a ctf container if it exists */
606	fp = Pname_to_ctf(dcp->dpa_proc, obj);
607	if (fp != NULL)
608		dcp->dpa_count++;
609	return (0);
610}
611
612/* ARGSUSED */
613static int
614dt_module_load_proc_build(void *arg, const prmap_t *prmap, const char *obj)
615{
616	ctf_file_t *fp;
617	char buf[MAXPATHLEN], *p;
618	dt_module_cb_arg_t *dcp = arg;
619	int count = dcp->dpa_count;
620	Lmid_t lmid;
621
622	fp = Pname_to_ctf(dcp->dpa_proc, obj);
623	if (fp == NULL)
624		return (0);
625	fp = ctf_dup(fp);
626	if (fp == NULL)
627		return (0);
628	dcp->dpa_dmp->dm_libctfp[count] = fp;
629	/*
630	 * While it'd be nice to simply use objname here, because of our prior
631	 * actions we'll always get a resolved object name to its on disk file.
632	 * Like the pid provider, we need to tell a bit of a lie here. The type
633	 * that the user thinks of is in terms of the libraries they requested,
634	 * eg. libc.so.1, they don't care about the fact that it's
635	 * libc_hwcap.so.1.
636	 */
637	(void) Pobjname(dcp->dpa_proc, prmap->pr_vaddr, buf, sizeof (buf));
638	if ((p = strrchr(buf, '/')) == NULL)
639		p = buf;
640	else
641		p++;
642
643	/*
644	 * If for some reason we can't find a link map id for this module, which
645	 * would be really quite weird. We instead just say the link map id is
646	 * zero.
647	 */
648	if (Plmid(dcp->dpa_proc, prmap->pr_vaddr, &lmid) != 0)
649		lmid = 0;
650
651	if (lmid == 0)
652		dcp->dpa_dmp->dm_libctfn[count] = strdup(p);
653	else
654		(void) asprintf(&dcp->dpa_dmp->dm_libctfn[count],
655		    "LM%x`%s", lmid, p);
656	if (dcp->dpa_dmp->dm_libctfn[count] == NULL)
657		return (1);
658	ctf_setspecific(fp, dcp->dpa_dmp);
659	dcp->dpa_count++;
660	return (0);
661}
662
663/*
664 * We've been asked to load data that belongs to another process. As such we're
665 * going to pgrab it at this instant, load everything that we might ever care
666 * about, and then drive on. The reason for this is that the process that we're
667 * interested in might be changing. As long as we have grabbed it, then this
668 * can't be a problem for us.
669 *
670 * For now, we're actually going to punt on most things and just try to get CTF
671 * data, nothing else. Basically this is only useful as a source of type
672 * information, we can't go and do the stacktrace lookups, etc.
673 */
674static int
675dt_module_load_proc(dtrace_hdl_t *dtp, dt_module_t *dmp)
676{
677	struct ps_prochandle *p;
678	dt_module_cb_arg_t arg;
679
680	/*
681	 * Note that on success we do not release this hold. We must hold this
682	 * for our life time.
683	 */
684	p = dt_proc_grab(dtp, dmp->dm_pid, 0, PGRAB_RDONLY | PGRAB_FORCE);
685	if (p == NULL) {
686		dt_dprintf("failed to grab pid: %d\n", (int)dmp->dm_pid);
687		return (dt_set_errno(dtp, EDT_CANTLOAD));
688	}
689	dt_proc_lock(dtp, p);
690
691	arg.dpa_proc = p;
692	arg.dpa_dtp = dtp;
693	arg.dpa_dmp = dmp;
694	arg.dpa_count = 0;
695	if (Pobject_iter_resolved(p, dt_module_load_proc_count, &arg) != 0) {
696		dt_dprintf("failed to iterate objects\n");
697		dt_proc_unlock(dtp, p);
698		dt_proc_release(dtp, p);
699		return (dt_set_errno(dtp, EDT_CANTLOAD));
700	}
701
702	if (arg.dpa_count == 0) {
703		dt_dprintf("no ctf data present\n");
704		dt_proc_unlock(dtp, p);
705		dt_proc_release(dtp, p);
706		return (dt_set_errno(dtp, EDT_CANTLOAD));
707	}
708
709	dmp->dm_libctfp = calloc(arg.dpa_count, sizeof (ctf_file_t *));
710	if (dmp->dm_libctfp == NULL) {
711		dt_proc_unlock(dtp, p);
712		dt_proc_release(dtp, p);
713		return (dt_set_errno(dtp, EDT_NOMEM));
714	}
715
716	dmp->dm_libctfn = calloc(arg.dpa_count, sizeof (char *));
717	if (dmp->dm_libctfn == NULL) {
718		free(dmp->dm_libctfp);
719		dt_proc_unlock(dtp, p);
720		dt_proc_release(dtp, p);
721		return (dt_set_errno(dtp, EDT_NOMEM));
722	}
723
724	dmp->dm_nctflibs = arg.dpa_count;
725
726	arg.dpa_count = 0;
727	if (Pobject_iter_resolved(p, dt_module_load_proc_build, &arg) != 0) {
728		dt_proc_unlock(dtp, p);
729		dt_module_unload(dtp, dmp);
730		dt_proc_release(dtp, p);
731		return (dt_set_errno(dtp, EDT_CANTLOAD));
732	}
733	assert(arg.dpa_count == dmp->dm_nctflibs);
734	dt_dprintf("loaded %d ctf modules for pid %d\n", arg.dpa_count,
735	    (int)dmp->dm_pid);
736
737	dt_proc_unlock(dtp, p);
738	dt_proc_release(dtp, p);
739	dmp->dm_flags |= DT_DM_LOADED;
740
741	return (0);
742}
743
744int
745dt_module_load(dtrace_hdl_t *dtp, dt_module_t *dmp)
746{
747	if (dmp->dm_flags & DT_DM_LOADED)
748		return (0); /* module is already loaded */
749
750	if (dmp->dm_pid != 0)
751		return (dt_module_load_proc(dtp, dmp));
752
753	dmp->dm_ctdata.cts_name = ".SUNW_ctf";
754	dmp->dm_ctdata.cts_type = SHT_PROGBITS;
755	dmp->dm_ctdata.cts_flags = 0;
756	dmp->dm_ctdata.cts_data = NULL;
757	dmp->dm_ctdata.cts_size = 0;
758	dmp->dm_ctdata.cts_entsize = 0;
759	dmp->dm_ctdata.cts_offset = 0;
760
761	dmp->dm_symtab.cts_name = ".symtab";
762	dmp->dm_symtab.cts_type = SHT_SYMTAB;
763	dmp->dm_symtab.cts_flags = 0;
764	dmp->dm_symtab.cts_data = NULL;
765	dmp->dm_symtab.cts_size = 0;
766	dmp->dm_symtab.cts_entsize = dmp->dm_ops == &dt_modops_64 ?
767	    sizeof (Elf64_Sym) : sizeof (Elf32_Sym);
768	dmp->dm_symtab.cts_offset = 0;
769
770	dmp->dm_strtab.cts_name = ".strtab";
771	dmp->dm_strtab.cts_type = SHT_STRTAB;
772	dmp->dm_strtab.cts_flags = 0;
773	dmp->dm_strtab.cts_data = NULL;
774	dmp->dm_strtab.cts_size = 0;
775	dmp->dm_strtab.cts_entsize = 0;
776	dmp->dm_strtab.cts_offset = 0;
777
778	/*
779	 * Attempt to load the module's CTF section, symbol table section, and
780	 * string table section.  Note that modules may not contain CTF data:
781	 * this will result in a successful load_sect but data of size zero.
782	 * We will then fail if dt_module_getctf() is called, as shown below.
783	 */
784	if (dt_module_load_sect(dtp, dmp, &dmp->dm_ctdata) == -1 ||
785	    dt_module_load_sect(dtp, dmp, &dmp->dm_symtab) == -1 ||
786	    dt_module_load_sect(dtp, dmp, &dmp->dm_strtab) == -1) {
787		dt_module_unload(dtp, dmp);
788		return (-1); /* dt_errno is set for us */
789	}
790
791	/*
792	 * Allocate the hash chains and hash buckets for symbol name lookup.
793	 * This is relatively simple since the symbol table is of fixed size
794	 * and is known in advance.  We allocate one extra element since we
795	 * use element indices instead of pointers and zero is our sentinel.
796	 */
797	dmp->dm_nsymelems =
798	    dmp->dm_symtab.cts_size / dmp->dm_symtab.cts_entsize;
799
800	dmp->dm_nsymbuckets = _dtrace_strbuckets;
801	dmp->dm_symfree = 1;		/* first free element is index 1 */
802
803	dmp->dm_symbuckets = calloc(dmp->dm_nsymbuckets, sizeof (uint_t));
804	dmp->dm_symchains = calloc(dmp->dm_nsymelems + 1, sizeof (dt_sym_t));
805
806	if (dmp->dm_symbuckets == NULL || dmp->dm_symchains == NULL) {
807		dt_module_unload(dtp, dmp);
808		return (dt_set_errno(dtp, EDT_NOMEM));
809	}
810
811	/*
812	 * Iterate over the symbol table data buffer and insert each symbol
813	 * name into the name hash if the name and type are valid.  Then
814	 * allocate the address map, fill it in, and sort it.
815	 */
816	dmp->dm_asrsv = dmp->dm_ops->do_syminit(dmp);
817
818	dt_dprintf("hashed %s [%s] (%u symbols)\n",
819	    dmp->dm_name, dmp->dm_symtab.cts_name, dmp->dm_symfree - 1);
820
821	if ((dmp->dm_asmap = malloc(sizeof (void *) * dmp->dm_asrsv)) == NULL) {
822		dt_module_unload(dtp, dmp);
823		return (dt_set_errno(dtp, EDT_NOMEM));
824	}
825
826	dmp->dm_ops->do_symsort(dmp);
827
828	dt_dprintf("sorted %s [%s] (%u symbols)\n",
829	    dmp->dm_name, dmp->dm_symtab.cts_name, dmp->dm_aslen);
830
831	dmp->dm_flags |= DT_DM_LOADED;
832	return (0);
833}
834
835int
836dt_module_hasctf(dtrace_hdl_t *dtp, dt_module_t *dmp)
837{
838	if (dmp->dm_pid != 0 && dmp->dm_nctflibs > 0)
839		return (1);
840	return (dt_module_getctf(dtp, dmp) != NULL);
841}
842
843ctf_file_t *
844dt_module_getctf(dtrace_hdl_t *dtp, dt_module_t *dmp)
845{
846	const char *parent;
847	dt_module_t *pmp;
848	ctf_file_t *pfp;
849	int model;
850
851	if (dmp->dm_ctfp != NULL || dt_module_load(dtp, dmp) != 0)
852		return (dmp->dm_ctfp);
853
854	if (dmp->dm_ops == &dt_modops_64)
855		model = CTF_MODEL_LP64;
856	else
857		model = CTF_MODEL_ILP32;
858
859	/*
860	 * If the data model of the module does not match our program data
861	 * model, then do not permit CTF from this module to be opened and
862	 * returned to the compiler.  If we support mixed data models in the
863	 * future for combined kernel/user tracing, this can be removed.
864	 */
865	if (dtp->dt_conf.dtc_ctfmodel != model) {
866		(void) dt_set_errno(dtp, EDT_DATAMODEL);
867		return (NULL);
868	}
869
870	if (dmp->dm_ctdata.cts_size == 0) {
871		(void) dt_set_errno(dtp, EDT_NOCTF);
872		return (NULL);
873	}
874
875	dmp->dm_ctfp = ctf_bufopen(&dmp->dm_ctdata,
876	    &dmp->dm_symtab, &dmp->dm_strtab, &dtp->dt_ctferr);
877
878	if (dmp->dm_ctfp == NULL) {
879		(void) dt_set_errno(dtp, EDT_CTF);
880		return (NULL);
881	}
882
883	(void) ctf_setmodel(dmp->dm_ctfp, model);
884	ctf_setspecific(dmp->dm_ctfp, dmp);
885
886	if ((parent = ctf_parent_name(dmp->dm_ctfp)) != NULL) {
887		if ((pmp = dt_module_create(dtp, parent)) == NULL ||
888		    (pfp = dt_module_getctf(dtp, pmp)) == NULL) {
889			if (pmp == NULL)
890				(void) dt_set_errno(dtp, EDT_NOMEM);
891			goto err;
892		}
893
894		if (ctf_import(dmp->dm_ctfp, pfp) == CTF_ERR) {
895			dtp->dt_ctferr = ctf_errno(dmp->dm_ctfp);
896			(void) dt_set_errno(dtp, EDT_CTF);
897			goto err;
898		}
899	}
900
901	dt_dprintf("loaded CTF container for %s (%p)\n",
902	    dmp->dm_name, (void *)dmp->dm_ctfp);
903
904	return (dmp->dm_ctfp);
905
906err:
907	ctf_close(dmp->dm_ctfp);
908	dmp->dm_ctfp = NULL;
909	return (NULL);
910}
911
912/*ARGSUSED*/
913void
914dt_module_unload(dtrace_hdl_t *dtp, dt_module_t *dmp)
915{
916	int i;
917
918	ctf_close(dmp->dm_ctfp);
919	dmp->dm_ctfp = NULL;
920
921#ifndef illumos
922	if (dmp->dm_ctdata.cts_data != NULL) {
923		free(dmp->dm_ctdata.cts_data);
924	}
925	if (dmp->dm_symtab.cts_data != NULL) {
926		free(dmp->dm_symtab.cts_data);
927	}
928	if (dmp->dm_strtab.cts_data != NULL) {
929		free(dmp->dm_strtab.cts_data);
930	}
931#endif
932
933	if (dmp->dm_libctfp != NULL) {
934		for (i = 0; i < dmp->dm_nctflibs; i++) {
935			ctf_close(dmp->dm_libctfp[i]);
936			free(dmp->dm_libctfn[i]);
937		}
938		free(dmp->dm_libctfp);
939		free(dmp->dm_libctfn);
940		dmp->dm_libctfp = NULL;
941		dmp->dm_nctflibs = 0;
942	}
943
944	bzero(&dmp->dm_ctdata, sizeof (ctf_sect_t));
945	bzero(&dmp->dm_symtab, sizeof (ctf_sect_t));
946	bzero(&dmp->dm_strtab, sizeof (ctf_sect_t));
947
948	if (dmp->dm_symbuckets != NULL) {
949		free(dmp->dm_symbuckets);
950		dmp->dm_symbuckets = NULL;
951	}
952
953	if (dmp->dm_symchains != NULL) {
954		free(dmp->dm_symchains);
955		dmp->dm_symchains = NULL;
956	}
957
958	if (dmp->dm_asmap != NULL) {
959		free(dmp->dm_asmap);
960		dmp->dm_asmap = NULL;
961	}
962#if defined(__FreeBSD__)
963	if (dmp->dm_sec_offsets != NULL) {
964		free(dmp->dm_sec_offsets);
965		dmp->dm_sec_offsets = NULL;
966	}
967#endif
968	dmp->dm_symfree = 0;
969	dmp->dm_nsymbuckets = 0;
970	dmp->dm_nsymelems = 0;
971	dmp->dm_asrsv = 0;
972	dmp->dm_aslen = 0;
973
974	dmp->dm_text_va = 0;
975	dmp->dm_text_size = 0;
976	dmp->dm_data_va = 0;
977	dmp->dm_data_size = 0;
978	dmp->dm_bss_va = 0;
979	dmp->dm_bss_size = 0;
980
981	if (dmp->dm_extern != NULL) {
982		dt_idhash_destroy(dmp->dm_extern);
983		dmp->dm_extern = NULL;
984	}
985
986	(void) elf_end(dmp->dm_elf);
987	dmp->dm_elf = NULL;
988
989	dmp->dm_pid = 0;
990
991	dmp->dm_flags &= ~DT_DM_LOADED;
992}
993
994void
995dt_module_destroy(dtrace_hdl_t *dtp, dt_module_t *dmp)
996{
997	uint_t h = dt_strtab_hash(dmp->dm_name, NULL) % dtp->dt_modbuckets;
998	dt_module_t **dmpp = &dtp->dt_mods[h];
999
1000	dt_list_delete(&dtp->dt_modlist, dmp);
1001	assert(dtp->dt_nmods != 0);
1002	dtp->dt_nmods--;
1003
1004	/*
1005	 * Now remove this module from its hash chain.  We expect to always
1006	 * find the module on its hash chain, so in this loop we assert that
1007	 * we don't run off the end of the list.
1008	 */
1009	while (*dmpp != dmp) {
1010		dmpp = &((*dmpp)->dm_next);
1011		assert(*dmpp != NULL);
1012	}
1013
1014	*dmpp = dmp->dm_next;
1015
1016	dt_module_unload(dtp, dmp);
1017	free(dmp);
1018}
1019
1020/*
1021 * Insert a new external symbol reference into the specified module.  The new
1022 * symbol will be marked as undefined and is assigned a symbol index beyond
1023 * any existing cached symbols from this module.  We use the ident's di_data
1024 * field to store a pointer to a copy of the dtrace_syminfo_t for this symbol.
1025 */
1026dt_ident_t *
1027dt_module_extern(dtrace_hdl_t *dtp, dt_module_t *dmp,
1028    const char *name, const dtrace_typeinfo_t *tip)
1029{
1030	dtrace_syminfo_t *sip;
1031	dt_ident_t *idp;
1032	uint_t id;
1033
1034	if (dmp->dm_extern == NULL && (dmp->dm_extern = dt_idhash_create(
1035	    "extern", NULL, dmp->dm_nsymelems, UINT_MAX)) == NULL) {
1036		(void) dt_set_errno(dtp, EDT_NOMEM);
1037		return (NULL);
1038	}
1039
1040	if (dt_idhash_nextid(dmp->dm_extern, &id) == -1) {
1041		(void) dt_set_errno(dtp, EDT_SYMOFLOW);
1042		return (NULL);
1043	}
1044
1045	if ((sip = malloc(sizeof (dtrace_syminfo_t))) == NULL) {
1046		(void) dt_set_errno(dtp, EDT_NOMEM);
1047		return (NULL);
1048	}
1049
1050	idp = dt_idhash_insert(dmp->dm_extern, name, DT_IDENT_SYMBOL, 0, id,
1051	    _dtrace_symattr, 0, &dt_idops_thaw, NULL, dtp->dt_gen);
1052
1053	if (idp == NULL) {
1054		(void) dt_set_errno(dtp, EDT_NOMEM);
1055		free(sip);
1056		return (NULL);
1057	}
1058
1059	sip->dts_object = dmp->dm_name;
1060	sip->dts_name = idp->di_name;
1061	sip->dts_id = idp->di_id;
1062
1063	idp->di_data = sip;
1064	idp->di_ctfp = tip->dtt_ctfp;
1065	idp->di_type = tip->dtt_type;
1066
1067	return (idp);
1068}
1069
1070const char *
1071dt_module_modelname(dt_module_t *dmp)
1072{
1073	if (dmp->dm_ops == &dt_modops_64)
1074		return ("64-bit");
1075	else
1076		return ("32-bit");
1077}
1078
1079/* ARGSUSED */
1080int
1081dt_module_getlibid(dtrace_hdl_t *dtp, dt_module_t *dmp, const ctf_file_t *fp)
1082{
1083	int i;
1084
1085	for (i = 0; i < dmp->dm_nctflibs; i++) {
1086		if (dmp->dm_libctfp[i] == fp)
1087			return (i);
1088	}
1089
1090	return (-1);
1091}
1092
1093/* ARGSUSED */
1094ctf_file_t *
1095dt_module_getctflib(dtrace_hdl_t *dtp, dt_module_t *dmp, const char *name)
1096{
1097	int i;
1098
1099	for (i = 0; i < dmp->dm_nctflibs; i++) {
1100		if (strcmp(dmp->dm_libctfn[i], name) == 0)
1101			return (dmp->dm_libctfp[i]);
1102	}
1103
1104	return (NULL);
1105}
1106
1107/*
1108 * Update our module cache by adding an entry for the specified module 'name'.
1109 * We create the dt_module_t and populate it using /system/object/<name>/.
1110 *
1111 * On FreeBSD, the module name is passed as the full module file name,
1112 * including the path.
1113 */
1114static void
1115#ifdef illumos
1116dt_module_update(dtrace_hdl_t *dtp, const char *name)
1117#else
1118dt_module_update(dtrace_hdl_t *dtp, struct kld_file_stat *k_stat)
1119#endif
1120{
1121	char fname[MAXPATHLEN];
1122	struct stat64 st;
1123	int fd, err, bits;
1124
1125	dt_module_t *dmp;
1126	const char *s;
1127	size_t shstrs;
1128	GElf_Shdr sh;
1129	Elf_Data *dp;
1130	Elf_Scn *sp;
1131
1132#ifdef illumos
1133	(void) snprintf(fname, sizeof (fname),
1134	    "%s/%s/object", OBJFS_ROOT, name);
1135#else
1136	GElf_Ehdr ehdr;
1137	GElf_Phdr ph;
1138	char name[MAXPATHLEN];
1139	uintptr_t mapbase, alignmask;
1140	int i = 0;
1141	int is_elf_obj;
1142
1143	(void) strlcpy(name, k_stat->name, sizeof(name));
1144	(void) strlcpy(fname, k_stat->pathname, sizeof(fname));
1145#endif
1146
1147	if ((fd = open(fname, O_RDONLY)) == -1 || fstat64(fd, &st) == -1 ||
1148	    (dmp = dt_module_create(dtp, name)) == NULL) {
1149		dt_dprintf("failed to open %s: %s\n", fname, strerror(errno));
1150		(void) close(fd);
1151		return;
1152	}
1153
1154	/*
1155	 * Since the module can unload out from under us (and /system/object
1156	 * will return ENOENT), tell libelf to cook the entire file now and
1157	 * then close the underlying file descriptor immediately.  If this
1158	 * succeeds, we know that we can continue safely using dmp->dm_elf.
1159	 */
1160	dmp->dm_elf = elf_begin(fd, ELF_C_READ, NULL);
1161	err = elf_cntl(dmp->dm_elf, ELF_C_FDREAD);
1162	(void) close(fd);
1163
1164	if (dmp->dm_elf == NULL || err == -1 ||
1165	    elf_getshdrstrndx(dmp->dm_elf, &shstrs) == -1) {
1166		dt_dprintf("failed to load %s: %s\n",
1167		    fname, elf_errmsg(elf_errno()));
1168		dt_module_destroy(dtp, dmp);
1169		return;
1170	}
1171
1172	switch (gelf_getclass(dmp->dm_elf)) {
1173	case ELFCLASS32:
1174		dmp->dm_ops = &dt_modops_32;
1175		bits = 32;
1176		break;
1177	case ELFCLASS64:
1178		dmp->dm_ops = &dt_modops_64;
1179		bits = 64;
1180		break;
1181	default:
1182		dt_dprintf("failed to load %s: unknown ELF class\n", fname);
1183		dt_module_destroy(dtp, dmp);
1184		return;
1185	}
1186#if defined(__FreeBSD__)
1187	mapbase = (uintptr_t)k_stat->address;
1188	gelf_getehdr(dmp->dm_elf, &ehdr);
1189	is_elf_obj = (ehdr.e_type == ET_REL);
1190	if (is_elf_obj) {
1191		dmp->dm_sec_offsets =
1192		    malloc(ehdr.e_shnum * sizeof(*dmp->dm_sec_offsets));
1193		if (dmp->dm_sec_offsets == NULL) {
1194			dt_dprintf("failed to allocate memory\n");
1195			dt_module_destroy(dtp, dmp);
1196			return;
1197		}
1198	}
1199#endif
1200	/*
1201	 * Iterate over the section headers locating various sections of
1202	 * interest and use their attributes to flesh out the dt_module_t.
1203	 */
1204	for (sp = NULL; (sp = elf_nextscn(dmp->dm_elf, sp)) != NULL; ) {
1205		if (gelf_getshdr(sp, &sh) == NULL || sh.sh_type == SHT_NULL ||
1206		    (s = elf_strptr(dmp->dm_elf, shstrs, sh.sh_name)) == NULL)
1207			continue; /* skip any malformed sections */
1208#if defined(__FreeBSD__)
1209		if (sh.sh_size == 0)
1210			continue;
1211		if (sh.sh_type == SHT_PROGBITS || sh.sh_type == SHT_NOBITS) {
1212			alignmask = sh.sh_addralign - 1;
1213			mapbase += alignmask;
1214			mapbase &= ~alignmask;
1215			sh.sh_addr = mapbase;
1216			if (is_elf_obj)
1217				dmp->dm_sec_offsets[elf_ndxscn(sp)] = sh.sh_addr;
1218			mapbase += sh.sh_size;
1219		}
1220#endif
1221		if (strcmp(s, ".text") == 0) {
1222			dmp->dm_text_size = sh.sh_size;
1223			dmp->dm_text_va = sh.sh_addr;
1224		} else if (strcmp(s, ".data") == 0) {
1225			dmp->dm_data_size = sh.sh_size;
1226			dmp->dm_data_va = sh.sh_addr;
1227		} else if (strcmp(s, ".bss") == 0) {
1228			dmp->dm_bss_size = sh.sh_size;
1229			dmp->dm_bss_va = sh.sh_addr;
1230		} else if (strcmp(s, ".info") == 0 &&
1231		    (dp = elf_getdata(sp, NULL)) != NULL) {
1232			bcopy(dp->d_buf, &dmp->dm_info,
1233			    MIN(sh.sh_size, sizeof (dmp->dm_info)));
1234		} else if (strcmp(s, ".filename") == 0 &&
1235		    (dp = elf_getdata(sp, NULL)) != NULL) {
1236			(void) strlcpy(dmp->dm_file,
1237			    dp->d_buf, sizeof (dmp->dm_file));
1238		}
1239	}
1240
1241	dmp->dm_flags |= DT_DM_KERNEL;
1242#ifdef illumos
1243	dmp->dm_modid = (int)OBJFS_MODID(st.st_ino);
1244#else
1245	/*
1246	 * Include .rodata and special sections into .text.
1247	 * This depends on default section layout produced by GNU ld
1248	 * for ELF objects and libraries:
1249	 * [Text][R/O data][R/W data][Dynamic][BSS][Non loadable]
1250	 */
1251	dmp->dm_text_size = dmp->dm_data_va - dmp->dm_text_va;
1252#if defined(__i386__)
1253	/*
1254	 * Find the first load section and figure out the relocation
1255	 * offset for the symbols. The kernel module will not need
1256	 * relocation, but the kernel linker modules will.
1257	 */
1258	for (i = 0; gelf_getphdr(dmp->dm_elf, i, &ph) != NULL; i++) {
1259		if (ph.p_type == PT_LOAD) {
1260			dmp->dm_reloc_offset = k_stat->address - ph.p_vaddr;
1261			break;
1262		}
1263	}
1264#endif
1265#endif /* illumos */
1266
1267	if (dmp->dm_info.objfs_info_primary)
1268		dmp->dm_flags |= DT_DM_PRIMARY;
1269
1270	dt_dprintf("opened %d-bit module %s (%s) [%d]\n",
1271	    bits, dmp->dm_name, dmp->dm_file, dmp->dm_modid);
1272}
1273
1274/*
1275 * Unload all the loaded modules and then refresh the module cache with the
1276 * latest list of loaded modules and their address ranges.
1277 */
1278void
1279dtrace_update(dtrace_hdl_t *dtp)
1280{
1281	dt_module_t *dmp;
1282	DIR *dirp;
1283#if defined(__FreeBSD__)
1284	int fileid;
1285#endif
1286
1287	for (dmp = dt_list_next(&dtp->dt_modlist);
1288	    dmp != NULL; dmp = dt_list_next(dmp))
1289		dt_module_unload(dtp, dmp);
1290
1291#ifdef illumos
1292	/*
1293	 * Open /system/object and attempt to create a libdtrace module for
1294	 * each kernel module that is loaded on the current system.
1295	 */
1296	if (!(dtp->dt_oflags & DTRACE_O_NOSYS) &&
1297	    (dirp = opendir(OBJFS_ROOT)) != NULL) {
1298		struct dirent *dp;
1299
1300		while ((dp = readdir(dirp)) != NULL) {
1301			if (dp->d_name[0] != '.')
1302				dt_module_update(dtp, dp->d_name);
1303		}
1304
1305		(void) closedir(dirp);
1306	}
1307#elif defined(__FreeBSD__)
1308	/*
1309	 * Use FreeBSD's kernel loader interface to discover what kernel
1310	 * modules are loaded and create a libdtrace module for each one.
1311	 */
1312	for (fileid = kldnext(0); fileid > 0; fileid = kldnext(fileid)) {
1313		struct kld_file_stat k_stat;
1314		k_stat.version = sizeof(k_stat);
1315		if (kldstat(fileid, &k_stat) == 0)
1316			dt_module_update(dtp, &k_stat);
1317	}
1318#endif
1319
1320	/*
1321	 * Look up all the macro identifiers and set di_id to the latest value.
1322	 * This code collaborates with dt_lex.l on the use of di_id.  We will
1323	 * need to implement something fancier if we need to support non-ints.
1324	 */
1325	dt_idhash_lookup(dtp->dt_macros, "egid")->di_id = getegid();
1326	dt_idhash_lookup(dtp->dt_macros, "euid")->di_id = geteuid();
1327	dt_idhash_lookup(dtp->dt_macros, "gid")->di_id = getgid();
1328	dt_idhash_lookup(dtp->dt_macros, "pid")->di_id = getpid();
1329	dt_idhash_lookup(dtp->dt_macros, "pgid")->di_id = getpgid(0);
1330	dt_idhash_lookup(dtp->dt_macros, "ppid")->di_id = getppid();
1331#ifdef illumos
1332	dt_idhash_lookup(dtp->dt_macros, "projid")->di_id = getprojid();
1333#endif
1334	dt_idhash_lookup(dtp->dt_macros, "sid")->di_id = getsid(0);
1335#ifdef illumos
1336	dt_idhash_lookup(dtp->dt_macros, "taskid")->di_id = gettaskid();
1337#endif
1338	dt_idhash_lookup(dtp->dt_macros, "uid")->di_id = getuid();
1339
1340	/*
1341	 * Cache the pointers to the modules representing the base executable
1342	 * and the run-time linker in the dtrace client handle. Note that on
1343	 * x86 krtld is folded into unix, so if we don't find it, use unix
1344	 * instead.
1345	 */
1346	dtp->dt_exec = dt_module_lookup_by_name(dtp, "genunix");
1347	dtp->dt_rtld = dt_module_lookup_by_name(dtp, "krtld");
1348	if (dtp->dt_rtld == NULL)
1349		dtp->dt_rtld = dt_module_lookup_by_name(dtp, "unix");
1350
1351	/*
1352	 * If this is the first time we are initializing the module list,
1353	 * remove the module for genunix from the module list and then move it
1354	 * to the front of the module list.  We do this so that type and symbol
1355	 * queries encounter genunix and thereby optimize for the common case
1356	 * in dtrace_lookup_by_name() and dtrace_lookup_by_type(), below.
1357	 */
1358	if (dtp->dt_exec != NULL &&
1359	    dtp->dt_cdefs == NULL && dtp->dt_ddefs == NULL) {
1360		dt_list_delete(&dtp->dt_modlist, dtp->dt_exec);
1361		dt_list_prepend(&dtp->dt_modlist, dtp->dt_exec);
1362	}
1363}
1364
1365static dt_module_t *
1366dt_module_from_object(dtrace_hdl_t *dtp, const char *object)
1367{
1368	int err = EDT_NOMOD;
1369	dt_module_t *dmp;
1370
1371	switch ((uintptr_t)object) {
1372	case (uintptr_t)DTRACE_OBJ_EXEC:
1373		dmp = dtp->dt_exec;
1374		break;
1375	case (uintptr_t)DTRACE_OBJ_RTLD:
1376		dmp = dtp->dt_rtld;
1377		break;
1378	case (uintptr_t)DTRACE_OBJ_CDEFS:
1379		dmp = dtp->dt_cdefs;
1380		break;
1381	case (uintptr_t)DTRACE_OBJ_DDEFS:
1382		dmp = dtp->dt_ddefs;
1383		break;
1384	default:
1385		dmp = dt_module_create(dtp, object);
1386		err = EDT_NOMEM;
1387	}
1388
1389	if (dmp == NULL)
1390		(void) dt_set_errno(dtp, err);
1391
1392	return (dmp);
1393}
1394
1395/*
1396 * Exported interface to look up a symbol by name.  We return the GElf_Sym and
1397 * complete symbol information for the matching symbol.
1398 */
1399int
1400dtrace_lookup_by_name(dtrace_hdl_t *dtp, const char *object, const char *name,
1401    GElf_Sym *symp, dtrace_syminfo_t *sip)
1402{
1403	dt_module_t *dmp;
1404	dt_ident_t *idp;
1405	uint_t n, id;
1406	GElf_Sym sym;
1407
1408	uint_t mask = 0; /* mask of dt_module flags to match */
1409	uint_t bits = 0; /* flag bits that must be present */
1410
1411	if (object != DTRACE_OBJ_EVERY &&
1412	    object != DTRACE_OBJ_KMODS &&
1413	    object != DTRACE_OBJ_UMODS) {
1414		if ((dmp = dt_module_from_object(dtp, object)) == NULL)
1415			return (-1); /* dt_errno is set for us */
1416
1417		if (dt_module_load(dtp, dmp) == -1)
1418			return (-1); /* dt_errno is set for us */
1419		n = 1;
1420
1421	} else {
1422		if (object == DTRACE_OBJ_KMODS)
1423			mask = bits = DT_DM_KERNEL;
1424		else if (object == DTRACE_OBJ_UMODS)
1425			mask = DT_DM_KERNEL;
1426
1427		dmp = dt_list_next(&dtp->dt_modlist);
1428		n = dtp->dt_nmods;
1429	}
1430
1431	if (symp == NULL)
1432		symp = &sym;
1433
1434	for (; n > 0; n--, dmp = dt_list_next(dmp)) {
1435		if ((dmp->dm_flags & mask) != bits)
1436			continue; /* failed to match required attributes */
1437
1438		if (dt_module_load(dtp, dmp) == -1)
1439			continue; /* failed to load symbol table */
1440
1441		if (dmp->dm_ops->do_symname(dmp, name, symp, &id) != NULL) {
1442			if (sip != NULL) {
1443				sip->dts_object = dmp->dm_name;
1444				sip->dts_name = (const char *)
1445				    dmp->dm_strtab.cts_data + symp->st_name;
1446				sip->dts_id = id;
1447			}
1448			return (0);
1449		}
1450
1451		if (dmp->dm_extern != NULL &&
1452		    (idp = dt_idhash_lookup(dmp->dm_extern, name)) != NULL) {
1453			if (symp != &sym) {
1454				symp->st_name = (uintptr_t)idp->di_name;
1455				symp->st_info =
1456				    GELF_ST_INFO(STB_GLOBAL, STT_NOTYPE);
1457				symp->st_other = 0;
1458				symp->st_shndx = SHN_UNDEF;
1459				symp->st_value = 0;
1460				symp->st_size =
1461				    ctf_type_size(idp->di_ctfp, idp->di_type);
1462			}
1463
1464			if (sip != NULL) {
1465				sip->dts_object = dmp->dm_name;
1466				sip->dts_name = idp->di_name;
1467				sip->dts_id = idp->di_id;
1468			}
1469
1470			return (0);
1471		}
1472	}
1473
1474	return (dt_set_errno(dtp, EDT_NOSYM));
1475}
1476
1477/*
1478 * Exported interface to look up a symbol by address.  We return the GElf_Sym
1479 * and complete symbol information for the matching symbol.
1480 */
1481int
1482dtrace_lookup_by_addr(dtrace_hdl_t *dtp, GElf_Addr addr,
1483    GElf_Sym *symp, dtrace_syminfo_t *sip)
1484{
1485	dt_module_t *dmp;
1486	uint_t id;
1487	const dtrace_vector_t *v = dtp->dt_vector;
1488
1489	if (v != NULL)
1490		return (v->dtv_lookup_by_addr(dtp->dt_varg, addr, symp, sip));
1491
1492	for (dmp = dt_list_next(&dtp->dt_modlist); dmp != NULL;
1493	    dmp = dt_list_next(dmp)) {
1494		if (addr - dmp->dm_text_va < dmp->dm_text_size ||
1495		    addr - dmp->dm_data_va < dmp->dm_data_size ||
1496		    addr - dmp->dm_bss_va < dmp->dm_bss_size)
1497			break;
1498	}
1499
1500	if (dmp == NULL)
1501		return (dt_set_errno(dtp, EDT_NOSYMADDR));
1502
1503	if (dt_module_load(dtp, dmp) == -1)
1504		return (-1); /* dt_errno is set for us */
1505
1506	if (symp != NULL) {
1507		if (dmp->dm_ops->do_symaddr(dmp, addr, symp, &id) == NULL)
1508			return (dt_set_errno(dtp, EDT_NOSYMADDR));
1509	}
1510
1511	if (sip != NULL) {
1512		sip->dts_object = dmp->dm_name;
1513
1514		if (symp != NULL) {
1515			sip->dts_name = (const char *)
1516			    dmp->dm_strtab.cts_data + symp->st_name;
1517			sip->dts_id = id;
1518		} else {
1519			sip->dts_name = NULL;
1520			sip->dts_id = 0;
1521		}
1522	}
1523
1524	return (0);
1525}
1526
1527int
1528dtrace_lookup_by_type(dtrace_hdl_t *dtp, const char *object, const char *name,
1529    dtrace_typeinfo_t *tip)
1530{
1531	dtrace_typeinfo_t ti;
1532	dt_module_t *dmp;
1533	int found = 0;
1534	ctf_id_t id;
1535	uint_t n, i;
1536	int justone;
1537	ctf_file_t *fp;
1538	char *buf, *p, *q;
1539
1540	uint_t mask = 0; /* mask of dt_module flags to match */
1541	uint_t bits = 0; /* flag bits that must be present */
1542
1543	if (object != DTRACE_OBJ_EVERY &&
1544	    object != DTRACE_OBJ_KMODS &&
1545	    object != DTRACE_OBJ_UMODS) {
1546		if ((dmp = dt_module_from_object(dtp, object)) == NULL)
1547			return (-1); /* dt_errno is set for us */
1548
1549		if (dt_module_load(dtp, dmp) == -1)
1550			return (-1); /* dt_errno is set for us */
1551		n = 1;
1552		justone = 1;
1553	} else {
1554		if (object == DTRACE_OBJ_KMODS)
1555			mask = bits = DT_DM_KERNEL;
1556		else if (object == DTRACE_OBJ_UMODS)
1557			mask = DT_DM_KERNEL;
1558
1559		dmp = dt_list_next(&dtp->dt_modlist);
1560		n = dtp->dt_nmods;
1561		justone = 0;
1562	}
1563
1564	if (tip == NULL)
1565		tip = &ti;
1566
1567	for (; n > 0; n--, dmp = dt_list_next(dmp)) {
1568		if ((dmp->dm_flags & mask) != bits)
1569			continue; /* failed to match required attributes */
1570
1571		/*
1572		 * If we can't load the CTF container, continue on to the next
1573		 * module.  If our search was scoped to only one module then
1574		 * return immediately leaving dt_errno unmodified.
1575		 */
1576		if (dt_module_hasctf(dtp, dmp) == 0) {
1577			if (justone)
1578				return (-1);
1579			continue;
1580		}
1581
1582		/*
1583		 * Look up the type in the module's CTF container.  If our
1584		 * match is a forward declaration tag, save this choice in
1585		 * 'tip' and keep going in the hope that we will locate the
1586		 * underlying structure definition.  Otherwise just return.
1587		 */
1588		if (dmp->dm_pid == 0) {
1589			id = ctf_lookup_by_name(dmp->dm_ctfp, name);
1590			fp = dmp->dm_ctfp;
1591		} else {
1592			if ((p = strchr(name, '`')) != NULL) {
1593				buf = strdup(name);
1594				if (buf == NULL)
1595					return (dt_set_errno(dtp, EDT_NOMEM));
1596				p = strchr(buf, '`');
1597				if ((q = strchr(p + 1, '`')) != NULL)
1598					p = q;
1599				*p = '\0';
1600				fp = dt_module_getctflib(dtp, dmp, buf);
1601				if (fp == NULL || (id = ctf_lookup_by_name(fp,
1602				    p + 1)) == CTF_ERR)
1603					id = CTF_ERR;
1604				free(buf);
1605			} else {
1606				for (i = 0; i < dmp->dm_nctflibs; i++) {
1607					fp = dmp->dm_libctfp[i];
1608					id = ctf_lookup_by_name(fp, name);
1609					if (id != CTF_ERR)
1610						break;
1611				}
1612			}
1613		}
1614		if (id != CTF_ERR) {
1615			tip->dtt_object = dmp->dm_name;
1616			tip->dtt_ctfp = fp;
1617			tip->dtt_type = id;
1618			if (ctf_type_kind(fp, ctf_type_resolve(fp, id)) !=
1619			    CTF_K_FORWARD)
1620				return (0);
1621
1622			found++;
1623		}
1624	}
1625
1626	if (found == 0)
1627		return (dt_set_errno(dtp, EDT_NOTYPE));
1628
1629	return (0);
1630}
1631
1632int
1633dtrace_symbol_type(dtrace_hdl_t *dtp, const GElf_Sym *symp,
1634    const dtrace_syminfo_t *sip, dtrace_typeinfo_t *tip)
1635{
1636	dt_module_t *dmp;
1637
1638	tip->dtt_object = NULL;
1639	tip->dtt_ctfp = NULL;
1640	tip->dtt_type = CTF_ERR;
1641	tip->dtt_flags = 0;
1642
1643	if ((dmp = dt_module_lookup_by_name(dtp, sip->dts_object)) == NULL)
1644		return (dt_set_errno(dtp, EDT_NOMOD));
1645
1646	if (symp->st_shndx == SHN_UNDEF && dmp->dm_extern != NULL) {
1647		dt_ident_t *idp =
1648		    dt_idhash_lookup(dmp->dm_extern, sip->dts_name);
1649
1650		if (idp == NULL)
1651			return (dt_set_errno(dtp, EDT_NOSYM));
1652
1653		tip->dtt_ctfp = idp->di_ctfp;
1654		tip->dtt_type = idp->di_type;
1655
1656	} else if (GELF_ST_TYPE(symp->st_info) != STT_FUNC) {
1657		if (dt_module_getctf(dtp, dmp) == NULL)
1658			return (-1); /* errno is set for us */
1659
1660		tip->dtt_ctfp = dmp->dm_ctfp;
1661		tip->dtt_type = ctf_lookup_by_symbol(dmp->dm_ctfp, sip->dts_id);
1662
1663		if (tip->dtt_type == CTF_ERR) {
1664			dtp->dt_ctferr = ctf_errno(tip->dtt_ctfp);
1665			return (dt_set_errno(dtp, EDT_CTF));
1666		}
1667
1668	} else {
1669		tip->dtt_ctfp = DT_FPTR_CTFP(dtp);
1670		tip->dtt_type = DT_FPTR_TYPE(dtp);
1671	}
1672
1673	tip->dtt_object = dmp->dm_name;
1674	return (0);
1675}
1676
1677static dtrace_objinfo_t *
1678dt_module_info(const dt_module_t *dmp, dtrace_objinfo_t *dto)
1679{
1680	dto->dto_name = dmp->dm_name;
1681	dto->dto_file = dmp->dm_file;
1682	dto->dto_id = dmp->dm_modid;
1683	dto->dto_flags = 0;
1684
1685	if (dmp->dm_flags & DT_DM_KERNEL)
1686		dto->dto_flags |= DTRACE_OBJ_F_KERNEL;
1687	if (dmp->dm_flags & DT_DM_PRIMARY)
1688		dto->dto_flags |= DTRACE_OBJ_F_PRIMARY;
1689
1690	dto->dto_text_va = dmp->dm_text_va;
1691	dto->dto_text_size = dmp->dm_text_size;
1692	dto->dto_data_va = dmp->dm_data_va;
1693	dto->dto_data_size = dmp->dm_data_size;
1694	dto->dto_bss_va = dmp->dm_bss_va;
1695	dto->dto_bss_size = dmp->dm_bss_size;
1696
1697	return (dto);
1698}
1699
1700int
1701dtrace_object_iter(dtrace_hdl_t *dtp, dtrace_obj_f *func, void *data)
1702{
1703	const dt_module_t *dmp = dt_list_next(&dtp->dt_modlist);
1704	dtrace_objinfo_t dto;
1705	int rv;
1706
1707	for (; dmp != NULL; dmp = dt_list_next(dmp)) {
1708		if ((rv = (*func)(dtp, dt_module_info(dmp, &dto), data)) != 0)
1709			return (rv);
1710	}
1711
1712	return (0);
1713}
1714
1715int
1716dtrace_object_info(dtrace_hdl_t *dtp, const char *object, dtrace_objinfo_t *dto)
1717{
1718	dt_module_t *dmp;
1719
1720	if (object == DTRACE_OBJ_EVERY || object == DTRACE_OBJ_KMODS ||
1721	    object == DTRACE_OBJ_UMODS || dto == NULL)
1722		return (dt_set_errno(dtp, EINVAL));
1723
1724	if ((dmp = dt_module_from_object(dtp, object)) == NULL)
1725		return (-1); /* dt_errno is set for us */
1726
1727	if (dt_module_load(dtp, dmp) == -1)
1728		return (-1); /* dt_errno is set for us */
1729
1730	(void) dt_module_info(dmp, dto);
1731	return (0);
1732}
1733