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) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2016 Joyent, Inc.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 * Copyright 2023 Oxide Computer Company
27 */
28
29 #include <assert.h>
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <stddef.h>
33 #include <unistd.h>
34 #include <ctype.h>
35 #include <fcntl.h>
36 #include <string.h>
37 #include <strings.h>
38 #include <memory.h>
39 #include <errno.h>
40 #include <dirent.h>
41 #include <signal.h>
42 #include <limits.h>
43 #include <libgen.h>
44 #include <sys/types.h>
45 #include <sys/stat.h>
46 #include <sys/sysmacros.h>
47 #include <sys/crc32.h>
48
49 #include "libproc.h"
50 #include "Pcontrol.h"
51 #include "Putil.h"
52 #include "Psymtab_machelf.h"
53
54 static file_info_t *build_map_symtab(struct ps_prochandle *, map_info_t *);
55 static map_info_t *exec_map(struct ps_prochandle *);
56 static map_info_t *object_to_map(struct ps_prochandle *, Lmid_t, const char *);
57 static map_info_t *object_name_to_map(struct ps_prochandle *,
58 Lmid_t, const char *);
59 static GElf_Sym *sym_by_name(sym_tbl_t *, const char *, GElf_Sym *, uint_t *);
60 static int read_ehdr32(struct ps_prochandle *, Elf32_Ehdr *, uint_t *,
61 uintptr_t);
62 #ifdef _LP64
63 static int read_ehdr64(struct ps_prochandle *, Elf64_Ehdr *, uint_t *,
64 uintptr_t);
65 #endif
66 static uint32_t psym_crc32[] = { CRC32_TABLE };
67
68 #define DATA_TYPES \
69 ((1 << STT_OBJECT) | (1 << STT_FUNC) | \
70 (1 << STT_COMMON) | (1 << STT_TLS))
71 #define IS_DATA_TYPE(tp) (((1 << (tp)) & DATA_TYPES) != 0)
72
73 #define MA_RWX (MA_READ | MA_WRITE | MA_EXEC)
74
75 /*
76 * Minimum and maximum length of a build-id that we'll accept. Generally it's a
77 * 20 byte SHA1 and it's expected that the first byte (which is two ascii
78 * characters) indicates a directory and the remaining bytes become the file
79 * name. Therefore, our minimum length is at least 2 bytes (one for the
80 * directory and one for the name) and the max is a bit over the minimum -- 64,
81 * just in case folks do something odd. The string length is three times the max
82 * length. This accounts for the fact that each byte is two characters, a null
83 * terminator, and the directory '/' character.
84 */
85 #define MINBUILDID 2
86 #define MAXBUILDID 64
87 #define BUILDID_STRLEN (3*MAXBUILDID)
88 #define BUILDID_NAME ".note.gnu.build-id"
89 #define DBGLINK_NAME ".gnu_debuglink"
90
91 typedef enum {
92 PRO_NATURAL,
93 PRO_BYADDR,
94 PRO_BYNAME
95 } pr_order_t;
96
97 static int
addr_cmp(const void * aa,const void * bb)98 addr_cmp(const void *aa, const void *bb)
99 {
100 uintptr_t a = *((uintptr_t *)aa);
101 uintptr_t b = *((uintptr_t *)bb);
102
103 if (a > b)
104 return (1);
105 if (a < b)
106 return (-1);
107 return (0);
108 }
109
110 /*
111 * This function creates a list of addresses for a load object's sections.
112 * The list is in ascending address order and alternates start address
113 * then end address for each section we're interested in. The function
114 * returns a pointer to the list, which must be freed by the caller.
115 */
116 static uintptr_t *
get_saddrs(struct ps_prochandle * P,uintptr_t ehdr_start,uint_t * n)117 get_saddrs(struct ps_prochandle *P, uintptr_t ehdr_start, uint_t *n)
118 {
119 uintptr_t a, addr, *addrs, last = 0;
120 uint_t i, naddrs = 0, unordered = 0;
121
122 if (P->status.pr_dmodel == PR_MODEL_ILP32) {
123 Elf32_Ehdr ehdr;
124 Elf32_Phdr phdr;
125 uint_t phnum;
126
127 if (read_ehdr32(P, &ehdr, &phnum, ehdr_start) != 0)
128 return (NULL);
129
130 addrs = malloc(sizeof (uintptr_t) * phnum * 2);
131 a = ehdr_start + ehdr.e_phoff;
132 for (i = 0; i < phnum; i++, a += ehdr.e_phentsize) {
133 if (Pread(P, &phdr, sizeof (phdr), a) !=
134 sizeof (phdr)) {
135 free(addrs);
136 return (NULL);
137 }
138 if (phdr.p_type != PT_LOAD || phdr.p_memsz == 0)
139 continue;
140
141 addr = phdr.p_vaddr;
142 if (ehdr.e_type == ET_DYN)
143 addr += ehdr_start;
144 if (last > addr)
145 unordered = 1;
146 addrs[naddrs++] = addr;
147 addrs[naddrs++] = last = addr + phdr.p_memsz - 1;
148 }
149 #ifdef _LP64
150 } else {
151 Elf64_Ehdr ehdr;
152 Elf64_Phdr phdr;
153 uint_t phnum;
154
155 if (read_ehdr64(P, &ehdr, &phnum, ehdr_start) != 0)
156 return (NULL);
157
158 addrs = malloc(sizeof (uintptr_t) * phnum * 2);
159 a = ehdr_start + ehdr.e_phoff;
160 for (i = 0; i < phnum; i++, a += ehdr.e_phentsize) {
161 if (Pread(P, &phdr, sizeof (phdr), a) !=
162 sizeof (phdr)) {
163 free(addrs);
164 return (NULL);
165 }
166 if (phdr.p_type != PT_LOAD || phdr.p_memsz == 0)
167 continue;
168
169 addr = phdr.p_vaddr;
170 if (ehdr.e_type == ET_DYN)
171 addr += ehdr_start;
172 if (last > addr)
173 unordered = 1;
174 addrs[naddrs++] = addr;
175 addrs[naddrs++] = last = addr + phdr.p_memsz - 1;
176 }
177 #endif
178 }
179
180 if (unordered)
181 qsort(addrs, naddrs, sizeof (uintptr_t), addr_cmp);
182
183 *n = naddrs;
184 return (addrs);
185 }
186
187 /*
188 * Allocation function for a new file_info_t
189 */
190 file_info_t *
file_info_new(struct ps_prochandle * P,map_info_t * mptr)191 file_info_new(struct ps_prochandle *P, map_info_t *mptr)
192 {
193 file_info_t *fptr;
194 map_info_t *mp;
195 uintptr_t mstart, mend, sstart, send;
196 uint_t i;
197
198 if ((fptr = calloc(1, sizeof (file_info_t))) == NULL)
199 return (NULL);
200
201 list_insert_tail(&P->file_head, fptr);
202 (void) strcpy(fptr->file_pname, mptr->map_pmap.pr_mapname);
203 mptr->map_file = fptr;
204 fptr->file_ref = 1;
205 fptr->file_fd = -1;
206 fptr->file_dbgfile = -1;
207 P->num_files++;
208
209 /*
210 * To figure out which map_info_t instances correspond to the mappings
211 * for this load object we try to obtain the start and end address
212 * for each section of our in-memory ELF image. If successful, we
213 * walk down the list of addresses and the list of map_info_t
214 * instances in lock step to correctly find the mappings that
215 * correspond to this load object.
216 */
217 if ((fptr->file_saddrs = get_saddrs(P, mptr->map_pmap.pr_vaddr,
218 &fptr->file_nsaddrs)) == NULL)
219 return (fptr);
220
221 mp = P->mappings;
222 i = 0;
223 while (mp < P->mappings + P->map_count && i < fptr->file_nsaddrs) {
224
225 /* Calculate the start and end of the mapping and section */
226 mstart = mp->map_pmap.pr_vaddr;
227 mend = mp->map_pmap.pr_vaddr + mp->map_pmap.pr_size;
228 sstart = fptr->file_saddrs[i];
229 send = fptr->file_saddrs[i + 1];
230
231 if (mend <= sstart) {
232 /* This mapping is below the current section */
233 mp++;
234 } else if (mstart >= send) {
235 /* This mapping is above the current section */
236 i += 2;
237 } else {
238 /* This mapping overlaps the current section */
239 if (mp->map_file == NULL) {
240 dprintf("file_info_new: associating "
241 "segment at %p\n",
242 (void *)mp->map_pmap.pr_vaddr);
243 mp->map_file = fptr;
244 fptr->file_ref++;
245 } else {
246 dprintf("file_info_new: segment at %p "
247 "already associated with %s\n",
248 (void *)mp->map_pmap.pr_vaddr,
249 (mp == mptr ? "this file" :
250 mp->map_file->file_pname));
251 }
252 mp++;
253 }
254 }
255
256 return (fptr);
257 }
258
259 /*
260 * Deallocation function for a file_info_t
261 */
262 static void
file_info_free(struct ps_prochandle * P,file_info_t * fptr)263 file_info_free(struct ps_prochandle *P, file_info_t *fptr)
264 {
265 if (--fptr->file_ref == 0) {
266 list_remove(&P->file_head, fptr);
267 if (fptr->file_symtab.sym_elf) {
268 (void) elf_end(fptr->file_symtab.sym_elf);
269 free(fptr->file_symtab.sym_elfmem);
270 }
271 if (fptr->file_symtab.sym_byname)
272 free(fptr->file_symtab.sym_byname);
273 if (fptr->file_symtab.sym_byaddr)
274 free(fptr->file_symtab.sym_byaddr);
275
276 if (fptr->file_dynsym.sym_elf) {
277 (void) elf_end(fptr->file_dynsym.sym_elf);
278 free(fptr->file_dynsym.sym_elfmem);
279 }
280 if (fptr->file_dynsym.sym_byname)
281 free(fptr->file_dynsym.sym_byname);
282 if (fptr->file_dynsym.sym_byaddr)
283 free(fptr->file_dynsym.sym_byaddr);
284
285 if (fptr->file_lo)
286 free(fptr->file_lo);
287 if (fptr->file_lname)
288 free(fptr->file_lname);
289 if (fptr->file_rname)
290 free(fptr->file_rname);
291 if (fptr->file_elf)
292 (void) elf_end(fptr->file_elf);
293 if (fptr->file_elfmem != NULL)
294 free(fptr->file_elfmem);
295 if (fptr->file_fd >= 0)
296 (void) close(fptr->file_fd);
297 if (fptr->file_dbgelf)
298 (void) elf_end(fptr->file_dbgelf);
299 if (fptr->file_dbgfile >= 0)
300 (void) close(fptr->file_dbgfile);
301 ctf_close(fptr->file_ctfp);
302 free(fptr->file_ctf_buf);
303 if (fptr->file_saddrs)
304 free(fptr->file_saddrs);
305 free(fptr);
306 P->num_files--;
307 }
308 }
309
310 /*
311 * Deallocation function for a map_info_t
312 */
313 static void
map_info_free(struct ps_prochandle * P,map_info_t * mptr)314 map_info_free(struct ps_prochandle *P, map_info_t *mptr)
315 {
316 file_info_t *fptr;
317
318 if ((fptr = mptr->map_file) != NULL) {
319 if (fptr->file_map == mptr)
320 fptr->file_map = NULL;
321 file_info_free(P, fptr);
322 }
323 if (P->execname && mptr == P->map_exec) {
324 free(P->execname);
325 P->execname = NULL;
326 }
327 if (P->auxv && (mptr == P->map_exec || mptr == P->map_ldso)) {
328 free(P->auxv);
329 P->auxv = NULL;
330 P->nauxv = 0;
331 }
332 if (mptr == P->map_exec)
333 P->map_exec = NULL;
334 if (mptr == P->map_ldso)
335 P->map_ldso = NULL;
336 }
337
338 /*
339 * Call-back function for librtld_db to iterate through all of its shared
340 * libraries. We use this to get the load object names for the mappings.
341 */
342 static int
map_iter(const rd_loadobj_t * lop,void * cd)343 map_iter(const rd_loadobj_t *lop, void *cd)
344 {
345 char buf[PATH_MAX];
346 struct ps_prochandle *P = cd;
347 map_info_t *mptr;
348 file_info_t *fptr;
349
350 dprintf("encountered rd object at %p\n", (void *)lop->rl_base);
351
352 if ((mptr = Paddr2mptr(P, lop->rl_base)) == NULL) {
353 dprintf("map_iter: base address doesn't match any mapping\n");
354 return (1); /* Base address does not match any mapping */
355 }
356
357 if ((fptr = mptr->map_file) == NULL &&
358 (fptr = file_info_new(P, mptr)) == NULL) {
359 dprintf("map_iter: failed to allocate a new file_info_t\n");
360 return (1); /* Failed to allocate a new file_info_t */
361 }
362
363 if ((fptr->file_lo == NULL) &&
364 (fptr->file_lo = malloc(sizeof (rd_loadobj_t))) == NULL) {
365 dprintf("map_iter: failed to allocate rd_loadobj_t\n");
366 file_info_free(P, fptr);
367 return (1); /* Failed to allocate rd_loadobj_t */
368 }
369
370 fptr->file_map = mptr;
371 *fptr->file_lo = *lop;
372
373 fptr->file_lo->rl_plt_base = fptr->file_plt_base;
374 fptr->file_lo->rl_plt_size = fptr->file_plt_size;
375
376 if (fptr->file_lname) {
377 free(fptr->file_lname);
378 fptr->file_lname = NULL;
379 fptr->file_lbase = NULL;
380 }
381 if (fptr->file_rname) {
382 free(fptr->file_rname);
383 fptr->file_rname = NULL;
384 fptr->file_rbase = NULL;
385 }
386
387 if (Pread_string(P, buf, sizeof (buf), lop->rl_nameaddr) > 0) {
388 if ((fptr->file_lname = strdup(buf)) != NULL)
389 fptr->file_lbase = basename(fptr->file_lname);
390 } else {
391 dprintf("map_iter: failed to read string at %p\n",
392 (void *)lop->rl_nameaddr);
393 }
394
395 if ((Pfindmap(P, mptr, buf, sizeof (buf)) != NULL) &&
396 ((fptr->file_rname = strdup(buf)) != NULL))
397 fptr->file_rbase = basename(fptr->file_rname);
398
399 dprintf("loaded rd object %s lmid %lx\n",
400 fptr->file_lname ? buf : "<NULL>", lop->rl_lmident);
401 return (1);
402 }
403
404 static void
map_set(struct ps_prochandle * P,map_info_t * mptr,const char * lname)405 map_set(struct ps_prochandle *P, map_info_t *mptr, const char *lname)
406 {
407 file_info_t *fptr;
408 char buf[PATH_MAX];
409
410 if ((fptr = mptr->map_file) == NULL &&
411 (fptr = file_info_new(P, mptr)) == NULL)
412 return; /* Failed to allocate a new file_info_t */
413
414 fptr->file_map = mptr;
415
416 if ((fptr->file_lo == NULL) &&
417 (fptr->file_lo = malloc(sizeof (rd_loadobj_t))) == NULL) {
418 file_info_free(P, fptr);
419 return; /* Failed to allocate rd_loadobj_t */
420 }
421
422 (void) memset(fptr->file_lo, 0, sizeof (rd_loadobj_t));
423 fptr->file_lo->rl_base = mptr->map_pmap.pr_vaddr;
424 fptr->file_lo->rl_bend =
425 mptr->map_pmap.pr_vaddr + mptr->map_pmap.pr_size;
426
427 fptr->file_lo->rl_plt_base = fptr->file_plt_base;
428 fptr->file_lo->rl_plt_size = fptr->file_plt_size;
429
430 if ((fptr->file_lname == NULL) &&
431 (fptr->file_lname = strdup(lname)) != NULL)
432 fptr->file_lbase = basename(fptr->file_lname);
433
434 if ((Pfindmap(P, mptr, buf, sizeof (buf)) != NULL) &&
435 ((fptr->file_rname = strdup(buf)) != NULL))
436 fptr->file_rbase = basename(fptr->file_rname);
437 }
438
439 static void
load_static_maps(struct ps_prochandle * P)440 load_static_maps(struct ps_prochandle *P)
441 {
442 map_info_t *mptr;
443
444 /*
445 * Construct the map for the a.out.
446 */
447 if ((mptr = object_name_to_map(P, PR_LMID_EVERY, PR_OBJ_EXEC)) != NULL)
448 map_set(P, mptr, "a.out");
449
450 /*
451 * If the dynamic linker exists for this process,
452 * construct the map for it.
453 */
454 if (Pgetauxval(P, AT_BASE) != -1L &&
455 (mptr = object_name_to_map(P, PR_LMID_EVERY, PR_OBJ_LDSO)) != NULL)
456 map_set(P, mptr, "ld.so.1");
457 }
458
459 int
Preadmaps(struct ps_prochandle * P,prmap_t ** Pmapp,ssize_t * nmapp)460 Preadmaps(struct ps_prochandle *P, prmap_t **Pmapp, ssize_t *nmapp)
461 {
462 return (P->ops.pop_read_maps(P, Pmapp, nmapp, P->data));
463 }
464
465 /*
466 * Go through all the address space mappings, validating or updating
467 * the information already gathered, or gathering new information.
468 *
469 * This function is only called when we suspect that the mappings have changed
470 * because this is the first time we're calling it or because of rtld activity.
471 */
472 void
Pupdate_maps(struct ps_prochandle * P)473 Pupdate_maps(struct ps_prochandle *P)
474 {
475 prmap_t *Pmap = NULL;
476 prmap_t *pmap;
477 ssize_t nmap;
478 int i;
479 uint_t oldmapcount;
480 map_info_t *newmap, *newp;
481 map_info_t *mptr;
482
483 if (P->info_valid || P->state == PS_UNDEAD)
484 return;
485
486 Preadauxvec(P);
487
488 if (Preadmaps(P, &Pmap, &nmap) != 0)
489 return;
490
491 if ((newmap = calloc(1, nmap * sizeof (map_info_t))) == NULL)
492 return;
493
494 /*
495 * We try to merge any file information we may have for existing
496 * mappings, to avoid having to rebuild the file info.
497 */
498 mptr = P->mappings;
499 pmap = Pmap;
500 newp = newmap;
501 oldmapcount = P->map_count;
502 for (i = 0; i < nmap; i++, pmap++, newp++) {
503
504 if (oldmapcount == 0) {
505 /*
506 * We've exhausted all the old mappings. Every new
507 * mapping should be added.
508 */
509 newp->map_pmap = *pmap;
510
511 } else if (pmap->pr_vaddr == mptr->map_pmap.pr_vaddr &&
512 pmap->pr_size == mptr->map_pmap.pr_size &&
513 pmap->pr_offset == mptr->map_pmap.pr_offset &&
514 (pmap->pr_mflags & ~(MA_BREAK | MA_STACK)) ==
515 (mptr->map_pmap.pr_mflags & ~(MA_BREAK | MA_STACK)) &&
516 pmap->pr_pagesize == mptr->map_pmap.pr_pagesize &&
517 pmap->pr_shmid == mptr->map_pmap.pr_shmid &&
518 strcmp(pmap->pr_mapname, mptr->map_pmap.pr_mapname) == 0) {
519
520 /*
521 * This mapping matches exactly. Copy over the old
522 * mapping, taking care to get the latest flags.
523 * Make sure the associated file_info_t is updated
524 * appropriately.
525 */
526 *newp = *mptr;
527 if (P->map_exec == mptr)
528 P->map_exec = newp;
529 if (P->map_ldso == mptr)
530 P->map_ldso = newp;
531 newp->map_pmap.pr_mflags = pmap->pr_mflags;
532 if (mptr->map_file != NULL &&
533 mptr->map_file->file_map == mptr)
534 mptr->map_file->file_map = newp;
535 oldmapcount--;
536 mptr++;
537
538 } else if (pmap->pr_vaddr + pmap->pr_size >
539 mptr->map_pmap.pr_vaddr) {
540
541 /*
542 * The old mapping doesn't exist any more, remove it
543 * from the list.
544 */
545 map_info_free(P, mptr);
546 oldmapcount--;
547 i--;
548 newp--;
549 pmap--;
550 mptr++;
551
552 } else {
553
554 /*
555 * This is a new mapping, add it directly.
556 */
557 newp->map_pmap = *pmap;
558 }
559 }
560
561 /*
562 * Free any old maps
563 */
564 while (oldmapcount) {
565 map_info_free(P, mptr);
566 oldmapcount--;
567 mptr++;
568 }
569
570 free(Pmap);
571 if (P->mappings != NULL)
572 free(P->mappings);
573 P->mappings = newmap;
574 P->map_count = P->map_alloc = nmap;
575 P->info_valid = 1;
576
577 /*
578 * Consult librtld_db to get the load object
579 * names for all of the shared libraries.
580 */
581 if (P->rap != NULL)
582 (void) rd_loadobj_iter(P->rap, map_iter, P);
583 }
584
585 /*
586 * Update all of the mappings and rtld_db as if by Pupdate_maps(), and then
587 * forcibly cache all of the symbol tables associated with all object files.
588 */
589 void
Pupdate_syms(struct ps_prochandle * P)590 Pupdate_syms(struct ps_prochandle *P)
591 {
592 file_info_t *fptr;
593
594 Pupdate_maps(P);
595
596 for (fptr = list_head(&P->file_head); fptr != NULL;
597 fptr = list_next(&P->file_head, fptr)) {
598 Pbuild_file_symtab(P, fptr);
599 (void) Pbuild_file_ctf(P, fptr);
600 }
601 }
602
603 /*
604 * Return the librtld_db agent handle for the victim process.
605 * The handle will become invalid at the next successful exec() and the
606 * client (caller of proc_rd_agent()) must not use it beyond that point.
607 * If the process is already dead, we've already tried our best to
608 * create the agent during core file initialization.
609 */
610 rd_agent_t *
Prd_agent(struct ps_prochandle * P)611 Prd_agent(struct ps_prochandle *P)
612 {
613 if (P->rap == NULL && P->state != PS_DEAD && P->state != PS_IDLE) {
614 Pupdate_maps(P);
615 if (P->num_files == 0)
616 load_static_maps(P);
617 rd_log(_libproc_debug);
618 if ((P->rap = rd_new(P)) != NULL)
619 (void) rd_loadobj_iter(P->rap, map_iter, P);
620 }
621 return (P->rap);
622 }
623
624 /*
625 * Return the prmap_t structure containing 'addr', but only if it
626 * is in the dynamic linker's link map and is the text section.
627 */
628 const prmap_t *
Paddr_to_text_map(struct ps_prochandle * P,uintptr_t addr)629 Paddr_to_text_map(struct ps_prochandle *P, uintptr_t addr)
630 {
631 map_info_t *mptr;
632
633 if (!P->info_valid)
634 Pupdate_maps(P);
635
636 if ((mptr = Paddr2mptr(P, addr)) != NULL) {
637 file_info_t *fptr = build_map_symtab(P, mptr);
638 const prmap_t *pmp = &mptr->map_pmap;
639
640 /*
641 * Assume that if rl_data_base is NULL, it means that no
642 * data section was found for this load object, and that
643 * a section must be text. Otherwise, a section will be
644 * text unless it ends above the start of the data
645 * section.
646 */
647 if (fptr != NULL && fptr->file_lo != NULL &&
648 (fptr->file_lo->rl_data_base == (uintptr_t)NULL ||
649 pmp->pr_vaddr + pmp->pr_size <=
650 fptr->file_lo->rl_data_base))
651 return (pmp);
652 }
653
654 return (NULL);
655 }
656
657 /*
658 * Return the prmap_t structure containing 'addr' (no restrictions on
659 * the type of mapping).
660 */
661 const prmap_t *
Paddr_to_map(struct ps_prochandle * P,uintptr_t addr)662 Paddr_to_map(struct ps_prochandle *P, uintptr_t addr)
663 {
664 map_info_t *mptr;
665
666 if (!P->info_valid)
667 Pupdate_maps(P);
668
669 if ((mptr = Paddr2mptr(P, addr)) != NULL)
670 return (&mptr->map_pmap);
671
672 return (NULL);
673 }
674
675 /*
676 * Convert a full or partial load object name to the prmap_t for its
677 * corresponding primary text mapping.
678 */
679 const prmap_t *
Plmid_to_map(struct ps_prochandle * P,Lmid_t lmid,const char * name)680 Plmid_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *name)
681 {
682 map_info_t *mptr;
683
684 if (name == PR_OBJ_EVERY)
685 return (NULL); /* A reasonable mistake */
686
687 if ((mptr = object_name_to_map(P, lmid, name)) != NULL)
688 return (&mptr->map_pmap);
689
690 return (NULL);
691 }
692
693 const prmap_t *
Pname_to_map(struct ps_prochandle * P,const char * name)694 Pname_to_map(struct ps_prochandle *P, const char *name)
695 {
696 return (Plmid_to_map(P, PR_LMID_EVERY, name));
697 }
698
699 const rd_loadobj_t *
Paddr_to_loadobj(struct ps_prochandle * P,uintptr_t addr)700 Paddr_to_loadobj(struct ps_prochandle *P, uintptr_t addr)
701 {
702 map_info_t *mptr;
703
704 if (!P->info_valid)
705 Pupdate_maps(P);
706
707 if ((mptr = Paddr2mptr(P, addr)) == NULL)
708 return (NULL);
709
710 /*
711 * By building the symbol table, we implicitly bring the PLT
712 * information up to date in the load object.
713 */
714 (void) build_map_symtab(P, mptr);
715
716 return (mptr->map_file->file_lo);
717 }
718
719 const rd_loadobj_t *
Plmid_to_loadobj(struct ps_prochandle * P,Lmid_t lmid,const char * name)720 Plmid_to_loadobj(struct ps_prochandle *P, Lmid_t lmid, const char *name)
721 {
722 map_info_t *mptr;
723
724 if (name == PR_OBJ_EVERY)
725 return (NULL);
726
727 if ((mptr = object_name_to_map(P, lmid, name)) == NULL)
728 return (NULL);
729
730 /*
731 * By building the symbol table, we implicitly bring the PLT
732 * information up to date in the load object.
733 */
734 (void) build_map_symtab(P, mptr);
735
736 return (mptr->map_file->file_lo);
737 }
738
739 const rd_loadobj_t *
Pname_to_loadobj(struct ps_prochandle * P,const char * name)740 Pname_to_loadobj(struct ps_prochandle *P, const char *name)
741 {
742 return (Plmid_to_loadobj(P, PR_LMID_EVERY, name));
743 }
744
745 ctf_file_t *
Pbuild_file_ctf(struct ps_prochandle * P,file_info_t * fptr)746 Pbuild_file_ctf(struct ps_prochandle *P, file_info_t *fptr)
747 {
748 ctf_sect_t ctdata, symtab, strtab;
749 sym_tbl_t *symp;
750 int err;
751
752 if (fptr->file_ctfp != NULL)
753 return (fptr->file_ctfp);
754
755 Pbuild_file_symtab(P, fptr);
756
757 if (fptr->file_ctf_size == 0)
758 return (NULL);
759
760 symp = fptr->file_ctf_dyn ? &fptr->file_dynsym : &fptr->file_symtab;
761 if (symp->sym_data_pri == NULL)
762 return (NULL);
763
764 /*
765 * The buffer may alread be allocated if this is a core file that
766 * contained CTF data for this file.
767 */
768 if (fptr->file_ctf_buf == NULL) {
769 fptr->file_ctf_buf = malloc(fptr->file_ctf_size);
770 if (fptr->file_ctf_buf == NULL) {
771 dprintf("failed to allocate ctf buffer\n");
772 return (NULL);
773 }
774
775 if (pread(fptr->file_fd, fptr->file_ctf_buf,
776 fptr->file_ctf_size, fptr->file_ctf_off) !=
777 fptr->file_ctf_size) {
778 free(fptr->file_ctf_buf);
779 fptr->file_ctf_buf = NULL;
780 dprintf("failed to read ctf data\n");
781 return (NULL);
782 }
783 }
784
785 ctdata.cts_name = ".SUNW_ctf";
786 ctdata.cts_type = SHT_PROGBITS;
787 ctdata.cts_flags = 0;
788 ctdata.cts_data = fptr->file_ctf_buf;
789 ctdata.cts_size = fptr->file_ctf_size;
790 ctdata.cts_entsize = 1;
791 ctdata.cts_offset = 0;
792
793 symtab.cts_name = fptr->file_ctf_dyn ? ".dynsym" : ".symtab";
794 symtab.cts_type = symp->sym_hdr_pri.sh_type;
795 symtab.cts_flags = symp->sym_hdr_pri.sh_flags;
796 symtab.cts_data = symp->sym_data_pri->d_buf;
797 symtab.cts_size = symp->sym_hdr_pri.sh_size;
798 symtab.cts_entsize = symp->sym_hdr_pri.sh_entsize;
799 symtab.cts_offset = symp->sym_hdr_pri.sh_offset;
800
801 strtab.cts_name = fptr->file_ctf_dyn ? ".dynstr" : ".strtab";
802 strtab.cts_type = symp->sym_strhdr.sh_type;
803 strtab.cts_flags = symp->sym_strhdr.sh_flags;
804 strtab.cts_data = symp->sym_strs;
805 strtab.cts_size = symp->sym_strhdr.sh_size;
806 strtab.cts_entsize = symp->sym_strhdr.sh_entsize;
807 strtab.cts_offset = symp->sym_strhdr.sh_offset;
808
809 fptr->file_ctfp = ctf_bufopen(&ctdata, &symtab, &strtab, &err);
810 if (fptr->file_ctfp == NULL) {
811 dprintf("ctf_bufopen() failed, error code %d\n", err);
812 free(fptr->file_ctf_buf);
813 fptr->file_ctf_buf = NULL;
814 return (NULL);
815 }
816
817 dprintf("loaded %lu bytes of CTF data for %s\n",
818 (ulong_t)fptr->file_ctf_size, fptr->file_pname);
819
820 return (fptr->file_ctfp);
821 }
822
823 ctf_file_t *
Paddr_to_ctf(struct ps_prochandle * P,uintptr_t addr)824 Paddr_to_ctf(struct ps_prochandle *P, uintptr_t addr)
825 {
826 map_info_t *mptr;
827 file_info_t *fptr;
828
829 if (!P->info_valid)
830 Pupdate_maps(P);
831
832 if ((mptr = Paddr2mptr(P, addr)) == NULL ||
833 (fptr = mptr->map_file) == NULL)
834 return (NULL);
835
836 return (Pbuild_file_ctf(P, fptr));
837 }
838
839 ctf_file_t *
Plmid_to_ctf(struct ps_prochandle * P,Lmid_t lmid,const char * name)840 Plmid_to_ctf(struct ps_prochandle *P, Lmid_t lmid, const char *name)
841 {
842 map_info_t *mptr;
843 file_info_t *fptr = NULL;
844
845 if (name == PR_OBJ_EVERY)
846 return (NULL);
847
848 /*
849 * While most idle files are all ELF objects, not all of them have
850 * mapping information available. There's nothing which would make
851 * sense to fake up for ET_REL. Instead, if we're being asked for their
852 * executable object and we know that the information is valid and they
853 * only have a single file, we jump straight to that file pointer.
854 */
855 if (P->state == PS_IDLE && name == PR_OBJ_EXEC && P->info_valid == 1 &&
856 P->num_files == 1 && P->mappings == NULL) {
857 fptr = list_head(&P->file_head);
858 }
859
860 if (fptr == NULL) {
861 if ((mptr = object_name_to_map(P, lmid, name)) == NULL ||
862 (fptr = mptr->map_file) == NULL)
863 return (NULL);
864 }
865
866 return (Pbuild_file_ctf(P, fptr));
867 }
868
869 ctf_file_t *
Pname_to_ctf(struct ps_prochandle * P,const char * name)870 Pname_to_ctf(struct ps_prochandle *P, const char *name)
871 {
872 return (Plmid_to_ctf(P, PR_LMID_EVERY, name));
873 }
874
875 void
Preadauxvec(struct ps_prochandle * P)876 Preadauxvec(struct ps_prochandle *P)
877 {
878 if (P->auxv != NULL) {
879 free(P->auxv);
880 P->auxv = NULL;
881 P->nauxv = 0;
882 }
883
884 P->ops.pop_read_aux(P, &P->auxv, &P->nauxv, P->data);
885 }
886
887 /*
888 * Return a requested element from the process's aux vector.
889 * Return -1 on failure (this is adequate for our purposes).
890 */
891 long
Pgetauxval(struct ps_prochandle * P,int type)892 Pgetauxval(struct ps_prochandle *P, int type)
893 {
894 auxv_t *auxv;
895
896 if (P->auxv == NULL)
897 Preadauxvec(P);
898
899 if (P->auxv == NULL)
900 return (-1);
901
902 for (auxv = P->auxv; auxv->a_type != AT_NULL; auxv++) {
903 if (auxv->a_type == type)
904 return (auxv->a_un.a_val);
905 }
906
907 return (-1);
908 }
909
910 /*
911 * Return a pointer to our internal copy of the process's aux vector.
912 * The caller should not hold on to this pointer across any libproc calls.
913 */
914 const auxv_t *
Pgetauxvec(struct ps_prochandle * P)915 Pgetauxvec(struct ps_prochandle *P)
916 {
917 static const auxv_t empty = { AT_NULL, 0L };
918
919 if (P->auxv == NULL)
920 Preadauxvec(P);
921
922 if (P->auxv == NULL)
923 return (&empty);
924
925 return (P->auxv);
926 }
927
928 /*
929 * Return 1 if the given mapping corresponds to the given file_info_t's
930 * load object; return 0 otherwise.
931 */
932 static int
is_mapping_in_file(struct ps_prochandle * P,map_info_t * mptr,file_info_t * fptr)933 is_mapping_in_file(struct ps_prochandle *P, map_info_t *mptr, file_info_t *fptr)
934 {
935 prmap_t *pmap = &mptr->map_pmap;
936 rd_loadobj_t *lop = fptr->file_lo;
937 uint_t i;
938 uintptr_t mstart, mend, sstart, send;
939
940 /*
941 * We can get for free the start address of the text and data
942 * sections of the load object. Start by seeing if the mapping
943 * encloses either of these.
944 */
945 if ((pmap->pr_vaddr <= lop->rl_base &&
946 lop->rl_base < pmap->pr_vaddr + pmap->pr_size) ||
947 (pmap->pr_vaddr <= lop->rl_data_base &&
948 lop->rl_data_base < pmap->pr_vaddr + pmap->pr_size))
949 return (1);
950
951 /*
952 * It's still possible that this mapping correponds to the load
953 * object. Consider the example of a mapping whose start and end
954 * addresses correspond to those of the load object's text section.
955 * If the mapping splits, e.g. as a result of a segment demotion,
956 * then although both mappings are still backed by the same section,
957 * only one will be seen to enclose that section's start address.
958 * Thus, to be rigorous, we ask not whether this mapping encloses
959 * the start of a section, but whether there exists a section that
960 * overlaps this mapping.
961 *
962 * If we don't already have the section addresses, and we successfully
963 * get them, then we cache them in case we come here again.
964 */
965 if (fptr->file_saddrs == NULL &&
966 (fptr->file_saddrs = get_saddrs(P,
967 fptr->file_map->map_pmap.pr_vaddr, &fptr->file_nsaddrs)) == NULL)
968 return (0);
969
970 mstart = mptr->map_pmap.pr_vaddr;
971 mend = mptr->map_pmap.pr_vaddr + mptr->map_pmap.pr_size;
972 for (i = 0; i < fptr->file_nsaddrs; i += 2) {
973 /* Does this section overlap the mapping? */
974 sstart = fptr->file_saddrs[i];
975 send = fptr->file_saddrs[i + 1];
976 if (!(mend <= sstart || mstart >= send))
977 return (1);
978 }
979
980 return (0);
981 }
982
983 /*
984 * Find or build the symbol table for the given mapping.
985 */
986 static file_info_t *
build_map_symtab(struct ps_prochandle * P,map_info_t * mptr)987 build_map_symtab(struct ps_prochandle *P, map_info_t *mptr)
988 {
989 prmap_t *pmap = &mptr->map_pmap;
990 file_info_t *fptr;
991
992 if ((fptr = mptr->map_file) != NULL) {
993 Pbuild_file_symtab(P, fptr);
994 return (fptr);
995 }
996
997 if (pmap->pr_mapname[0] == '\0')
998 return (NULL);
999
1000 /*
1001 * Attempt to find a matching file.
1002 * (A file can be mapped at several different addresses.)
1003 */
1004 for (fptr = list_head(&P->file_head); fptr != NULL;
1005 fptr = list_next(&P->file_head, fptr)) {
1006 if (strcmp(fptr->file_pname, pmap->pr_mapname) == 0 &&
1007 fptr->file_lo && is_mapping_in_file(P, mptr, fptr)) {
1008 mptr->map_file = fptr;
1009 fptr->file_ref++;
1010 Pbuild_file_symtab(P, fptr);
1011 return (fptr);
1012 }
1013 }
1014
1015 /*
1016 * If we need to create a new file_info structure, iterate
1017 * through the load objects in order to attempt to connect
1018 * this new file with its primary text mapping. We again
1019 * need to handle ld.so as a special case because we need
1020 * to be able to bootstrap librtld_db.
1021 */
1022 if ((fptr = file_info_new(P, mptr)) == NULL)
1023 return (NULL);
1024
1025 if (P->map_ldso != mptr) {
1026 if (P->rap != NULL)
1027 (void) rd_loadobj_iter(P->rap, map_iter, P);
1028 else
1029 (void) Prd_agent(P);
1030 } else {
1031 fptr->file_map = mptr;
1032 }
1033
1034 /*
1035 * If librtld_db wasn't able to help us connect the file to a primary
1036 * text mapping, set file_map to the current mapping because we require
1037 * fptr->file_map to be set in Pbuild_file_symtab. librtld_db may be
1038 * unaware of what's going on in the rare case that a legitimate ELF
1039 * file has been mmap(2)ed into the process address space *without*
1040 * the use of dlopen(3x).
1041 */
1042 if (fptr->file_map == NULL)
1043 fptr->file_map = mptr;
1044
1045 Pbuild_file_symtab(P, fptr);
1046
1047 return (fptr);
1048 }
1049
1050 static int
read_ehdr32(struct ps_prochandle * P,Elf32_Ehdr * ehdr,uint_t * phnum,uintptr_t addr)1051 read_ehdr32(struct ps_prochandle *P, Elf32_Ehdr *ehdr, uint_t *phnum,
1052 uintptr_t addr)
1053 {
1054 if (Pread(P, ehdr, sizeof (*ehdr), addr) != sizeof (*ehdr))
1055 return (-1);
1056
1057 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1058 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1059 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1060 ehdr->e_ident[EI_MAG3] != ELFMAG3 ||
1061 ehdr->e_ident[EI_CLASS] != ELFCLASS32 ||
1062 #ifdef _BIG_ENDIAN
1063 ehdr->e_ident[EI_DATA] != ELFDATA2MSB ||
1064 #else
1065 ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
1066 #endif
1067 ehdr->e_ident[EI_VERSION] != EV_CURRENT)
1068 return (-1);
1069
1070 if ((*phnum = ehdr->e_phnum) == PN_XNUM) {
1071 Elf32_Shdr shdr0;
1072
1073 if (ehdr->e_shoff == 0 || ehdr->e_shentsize < sizeof (shdr0) ||
1074 Pread(P, &shdr0, sizeof (shdr0), addr + ehdr->e_shoff) !=
1075 sizeof (shdr0))
1076 return (-1);
1077
1078 if (shdr0.sh_info != 0)
1079 *phnum = shdr0.sh_info;
1080 }
1081
1082 return (0);
1083 }
1084
1085 static int
read_dynamic_phdr32(struct ps_prochandle * P,const Elf32_Ehdr * ehdr,uint_t phnum,Elf32_Phdr * phdr,uintptr_t addr)1086 read_dynamic_phdr32(struct ps_prochandle *P, const Elf32_Ehdr *ehdr,
1087 uint_t phnum, Elf32_Phdr *phdr, uintptr_t addr)
1088 {
1089 uint_t i;
1090
1091 for (i = 0; i < phnum; i++) {
1092 uintptr_t a = addr + ehdr->e_phoff + i * ehdr->e_phentsize;
1093 if (Pread(P, phdr, sizeof (*phdr), a) != sizeof (*phdr))
1094 return (-1);
1095
1096 if (phdr->p_type == PT_DYNAMIC)
1097 return (0);
1098 }
1099
1100 return (-1);
1101 }
1102
1103 #ifdef _LP64
1104 static int
read_ehdr64(struct ps_prochandle * P,Elf64_Ehdr * ehdr,uint_t * phnum,uintptr_t addr)1105 read_ehdr64(struct ps_prochandle *P, Elf64_Ehdr *ehdr, uint_t *phnum,
1106 uintptr_t addr)
1107 {
1108 if (Pread(P, ehdr, sizeof (Elf64_Ehdr), addr) != sizeof (Elf64_Ehdr))
1109 return (-1);
1110
1111 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
1112 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
1113 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
1114 ehdr->e_ident[EI_MAG3] != ELFMAG3 ||
1115 ehdr->e_ident[EI_CLASS] != ELFCLASS64 ||
1116 #ifdef _BIG_ENDIAN
1117 ehdr->e_ident[EI_DATA] != ELFDATA2MSB ||
1118 #else
1119 ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
1120 #endif
1121 ehdr->e_ident[EI_VERSION] != EV_CURRENT)
1122 return (-1);
1123
1124 if ((*phnum = ehdr->e_phnum) == PN_XNUM) {
1125 Elf64_Shdr shdr0;
1126
1127 if (ehdr->e_shoff == 0 || ehdr->e_shentsize < sizeof (shdr0) ||
1128 Pread(P, &shdr0, sizeof (shdr0), addr + ehdr->e_shoff) !=
1129 sizeof (shdr0))
1130 return (-1);
1131
1132 if (shdr0.sh_info != 0)
1133 *phnum = shdr0.sh_info;
1134 }
1135
1136 return (0);
1137 }
1138
1139 static int
read_dynamic_phdr64(struct ps_prochandle * P,const Elf64_Ehdr * ehdr,uint_t phnum,Elf64_Phdr * phdr,uintptr_t addr)1140 read_dynamic_phdr64(struct ps_prochandle *P, const Elf64_Ehdr *ehdr,
1141 uint_t phnum, Elf64_Phdr *phdr, uintptr_t addr)
1142 {
1143 uint_t i;
1144
1145 for (i = 0; i < phnum; i++) {
1146 uintptr_t a = addr + ehdr->e_phoff + i * ehdr->e_phentsize;
1147 if (Pread(P, phdr, sizeof (*phdr), a) != sizeof (*phdr))
1148 return (-1);
1149
1150 if (phdr->p_type == PT_DYNAMIC)
1151 return (0);
1152 }
1153
1154 return (-1);
1155 }
1156 #endif /* _LP64 */
1157
1158 /*
1159 * The text segment for each load object contains the elf header and
1160 * program headers. We can use this information to determine if the
1161 * file that corresponds to the load object is the same file that
1162 * was loaded into the process's address space. There can be a discrepency
1163 * if a file is recompiled after the process is started or if the target
1164 * represents a core file from a differently configured system -- two
1165 * common examples. The DT_CHECKSUM entry in the dynamic section
1166 * provides an easy method of comparison. It is important to note that
1167 * the dynamic section usually lives in the data segment, but the meta
1168 * data we use to find the dynamic section lives in the text segment so
1169 * if either of those segments is absent we can't proceed.
1170 *
1171 * We're looking through the elf file for several items: the symbol tables
1172 * (both dynsym and symtab), the procedure linkage table (PLT) base,
1173 * size, and relocation base, and the CTF information. Most of this can
1174 * be recovered from the loaded image of the file itself, the exceptions
1175 * being the symtab and CTF data.
1176 *
1177 * First we try to open the file that we think corresponds to the load
1178 * object, if the DT_CHECKSUM values match, we're all set, and can simply
1179 * recover all the information we need from the file. If the values of
1180 * DT_CHECKSUM don't match, or if we can't access the file for whatever
1181 * reasaon, we fake up a elf file to use in its stead. If we can't read
1182 * the elf data in the process's address space, we fall back to using
1183 * the file even though it may give inaccurate information.
1184 *
1185 * The elf file that we fake up has to consist of sections for the
1186 * dynsym, the PLT and the dynamic section. Note that in the case of a
1187 * core file, we'll get the CTF data in the file_info_t later on from
1188 * a section embedded the core file (if it's present).
1189 *
1190 * file_differs() conservatively looks for mismatched files, identifying
1191 * a match when there is any ambiguity (since that's the legacy behavior).
1192 */
1193 static int
file_differs(struct ps_prochandle * P,Elf * elf,file_info_t * fptr)1194 file_differs(struct ps_prochandle *P, Elf *elf, file_info_t *fptr)
1195 {
1196 Elf_Scn *scn;
1197 GElf_Shdr shdr;
1198 GElf_Dyn dyn;
1199 Elf_Data *data;
1200 uint_t i, ndyn;
1201 GElf_Xword cksum;
1202 uintptr_t addr;
1203
1204 if (fptr->file_map == NULL)
1205 return (0);
1206
1207 if ((Pcontent(P) & (CC_CONTENT_TEXT | CC_CONTENT_DATA)) !=
1208 (CC_CONTENT_TEXT | CC_CONTENT_DATA))
1209 return (0);
1210
1211 /*
1212 * First, we find the checksum value in the elf file.
1213 */
1214 scn = NULL;
1215 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1216 if (gelf_getshdr(scn, &shdr) != NULL &&
1217 shdr.sh_type == SHT_DYNAMIC)
1218 goto found_shdr;
1219 }
1220 return (0);
1221
1222 found_shdr:
1223 if ((data = elf_getdata(scn, NULL)) == NULL)
1224 return (0);
1225
1226 if (P->status.pr_dmodel == PR_MODEL_ILP32)
1227 ndyn = shdr.sh_size / sizeof (Elf32_Dyn);
1228 #ifdef _LP64
1229 else if (P->status.pr_dmodel == PR_MODEL_LP64)
1230 ndyn = shdr.sh_size / sizeof (Elf64_Dyn);
1231 #endif
1232 else
1233 return (0);
1234
1235 for (i = 0; i < ndyn; i++) {
1236 if (gelf_getdyn(data, i, &dyn) != NULL &&
1237 dyn.d_tag == DT_CHECKSUM)
1238 goto found_cksum;
1239 }
1240
1241 /*
1242 * The in-memory ELF has no DT_CHECKSUM section, but we will report it
1243 * as matching the file anyhow.
1244 */
1245 return (0);
1246
1247 found_cksum:
1248 cksum = dyn.d_un.d_val;
1249 dprintf("elf cksum value is %llx\n", (u_longlong_t)cksum);
1250
1251 /*
1252 * Get the base of the text mapping that corresponds to this file.
1253 */
1254 addr = fptr->file_map->map_pmap.pr_vaddr;
1255
1256 if (P->status.pr_dmodel == PR_MODEL_ILP32) {
1257 Elf32_Ehdr ehdr;
1258 Elf32_Phdr phdr;
1259 Elf32_Dyn dync, *dynp;
1260 uint_t phnum, i;
1261
1262 if (read_ehdr32(P, &ehdr, &phnum, addr) != 0 ||
1263 read_dynamic_phdr32(P, &ehdr, phnum, &phdr, addr) != 0)
1264 return (0);
1265
1266 if (ehdr.e_type == ET_DYN)
1267 phdr.p_vaddr += addr;
1268 if ((dynp = malloc(phdr.p_filesz)) == NULL)
1269 return (0);
1270 dync.d_tag = DT_NULL;
1271 if (Pread(P, dynp, phdr.p_filesz, phdr.p_vaddr) !=
1272 phdr.p_filesz) {
1273 free(dynp);
1274 return (0);
1275 }
1276
1277 for (i = 0; i < phdr.p_filesz / sizeof (Elf32_Dyn); i++) {
1278 if (dynp[i].d_tag == DT_CHECKSUM)
1279 dync = dynp[i];
1280 }
1281
1282 free(dynp);
1283
1284 if (dync.d_tag != DT_CHECKSUM)
1285 return (0);
1286
1287 dprintf("image cksum value is %llx\n",
1288 (u_longlong_t)dync.d_un.d_val);
1289 return (dync.d_un.d_val != cksum);
1290 #ifdef _LP64
1291 } else if (P->status.pr_dmodel == PR_MODEL_LP64) {
1292 Elf64_Ehdr ehdr;
1293 Elf64_Phdr phdr;
1294 Elf64_Dyn dync, *dynp;
1295 uint_t phnum, i;
1296
1297 if (read_ehdr64(P, &ehdr, &phnum, addr) != 0 ||
1298 read_dynamic_phdr64(P, &ehdr, phnum, &phdr, addr) != 0)
1299 return (0);
1300
1301 if (ehdr.e_type == ET_DYN)
1302 phdr.p_vaddr += addr;
1303 if ((dynp = malloc(phdr.p_filesz)) == NULL)
1304 return (0);
1305 dync.d_tag = DT_NULL;
1306 if (Pread(P, dynp, phdr.p_filesz, phdr.p_vaddr) !=
1307 phdr.p_filesz) {
1308 free(dynp);
1309 return (0);
1310 }
1311
1312 for (i = 0; i < phdr.p_filesz / sizeof (Elf64_Dyn); i++) {
1313 if (dynp[i].d_tag == DT_CHECKSUM)
1314 dync = dynp[i];
1315 }
1316
1317 free(dynp);
1318
1319 if (dync.d_tag != DT_CHECKSUM)
1320 return (0);
1321
1322 dprintf("image cksum value is %llx\n",
1323 (u_longlong_t)dync.d_un.d_val);
1324 return (dync.d_un.d_val != cksum);
1325 #endif /* _LP64 */
1326 }
1327
1328 return (0);
1329 }
1330
1331 /*
1332 * Read data from the specified process and construct an in memory
1333 * image of an ELF file that represents it well enough to let
1334 * us probe it for information.
1335 */
1336 static Elf *
fake_elf(struct ps_prochandle * P,file_info_t * fptr)1337 fake_elf(struct ps_prochandle *P, file_info_t *fptr)
1338 {
1339 Elf *elf;
1340 uintptr_t addr;
1341 uint_t phnum;
1342
1343 if (fptr->file_map == NULL)
1344 return (NULL);
1345
1346 if ((Pcontent(P) & (CC_CONTENT_TEXT | CC_CONTENT_DATA)) !=
1347 (CC_CONTENT_TEXT | CC_CONTENT_DATA))
1348 return (NULL);
1349
1350 addr = fptr->file_map->map_pmap.pr_vaddr;
1351
1352 if (P->status.pr_dmodel == PR_MODEL_ILP32) {
1353 Elf32_Ehdr ehdr;
1354 Elf32_Phdr phdr;
1355
1356 if ((read_ehdr32(P, &ehdr, &phnum, addr) != 0) ||
1357 read_dynamic_phdr32(P, &ehdr, phnum, &phdr, addr) != 0)
1358 return (NULL);
1359
1360 elf = fake_elf32(P, fptr, addr, &ehdr, phnum, &phdr);
1361 #ifdef _LP64
1362 } else {
1363 Elf64_Ehdr ehdr;
1364 Elf64_Phdr phdr;
1365
1366 if (read_ehdr64(P, &ehdr, &phnum, addr) != 0 ||
1367 read_dynamic_phdr64(P, &ehdr, phnum, &phdr, addr) != 0)
1368 return (NULL);
1369
1370 elf = fake_elf64(P, fptr, addr, &ehdr, phnum, &phdr);
1371 #endif
1372 }
1373
1374 return (elf);
1375 }
1376
1377 /*
1378 * We wouldn't need these if qsort(3C) took an argument for the callback...
1379 */
1380 static mutex_t sort_mtx = DEFAULTMUTEX;
1381 static char *sort_strs;
1382 static GElf_Sym *sort_syms;
1383
1384 int
byaddr_cmp_common(GElf_Sym * a,char * aname,GElf_Sym * b,char * bname)1385 byaddr_cmp_common(GElf_Sym *a, char *aname, GElf_Sym *b, char *bname)
1386 {
1387 if (a->st_value < b->st_value)
1388 return (-1);
1389 if (a->st_value > b->st_value)
1390 return (1);
1391
1392 /*
1393 * Prefer the function to the non-function.
1394 */
1395 if (GELF_ST_TYPE(a->st_info) != GELF_ST_TYPE(b->st_info)) {
1396 if (GELF_ST_TYPE(a->st_info) == STT_FUNC)
1397 return (-1);
1398 if (GELF_ST_TYPE(b->st_info) == STT_FUNC)
1399 return (1);
1400 }
1401
1402 /*
1403 * Prefer the weak or strong global symbol to the local symbol.
1404 */
1405 if (GELF_ST_BIND(a->st_info) != GELF_ST_BIND(b->st_info)) {
1406 if (GELF_ST_BIND(b->st_info) == STB_LOCAL)
1407 return (-1);
1408 if (GELF_ST_BIND(a->st_info) == STB_LOCAL)
1409 return (1);
1410 }
1411
1412 /*
1413 * Prefer the symbol that doesn't begin with a '$' since compilers and
1414 * other symbol generators often use it as a prefix.
1415 */
1416 if (*bname == '$')
1417 return (-1);
1418 if (*aname == '$')
1419 return (1);
1420
1421 /*
1422 * Prefer the name with fewer leading underscores in the name.
1423 */
1424 while (*aname == '_' && *bname == '_') {
1425 aname++;
1426 bname++;
1427 }
1428
1429 if (*bname == '_')
1430 return (-1);
1431 if (*aname == '_')
1432 return (1);
1433
1434 /*
1435 * Prefer the symbol with the smaller size.
1436 */
1437 if (a->st_size < b->st_size)
1438 return (-1);
1439 if (a->st_size > b->st_size)
1440 return (1);
1441
1442 /*
1443 * All other factors being equal, fall back to lexicographic order.
1444 */
1445 return (strcmp(aname, bname));
1446 }
1447
1448 static int
byaddr_cmp(const void * aa,const void * bb)1449 byaddr_cmp(const void *aa, const void *bb)
1450 {
1451 GElf_Sym *a = &sort_syms[*(uint_t *)aa];
1452 GElf_Sym *b = &sort_syms[*(uint_t *)bb];
1453 char *aname = sort_strs + a->st_name;
1454 char *bname = sort_strs + b->st_name;
1455
1456 return (byaddr_cmp_common(a, aname, b, bname));
1457 }
1458
1459 static int
byname_cmp(const void * aa,const void * bb)1460 byname_cmp(const void *aa, const void *bb)
1461 {
1462 GElf_Sym *a = &sort_syms[*(uint_t *)aa];
1463 GElf_Sym *b = &sort_syms[*(uint_t *)bb];
1464 char *aname = sort_strs + a->st_name;
1465 char *bname = sort_strs + b->st_name;
1466
1467 return (strcmp(aname, bname));
1468 }
1469
1470 /*
1471 * Given a symbol index, look up the corresponding symbol from the
1472 * given symbol table.
1473 *
1474 * This function allows the caller to treat the symbol table as a single
1475 * logical entity even though there may be 2 actual ELF symbol tables
1476 * involved. See the comments in Pcontrol.h for details.
1477 */
1478 static GElf_Sym *
symtab_getsym(sym_tbl_t * symtab,int ndx,GElf_Sym * dst)1479 symtab_getsym(sym_tbl_t *symtab, int ndx, GElf_Sym *dst)
1480 {
1481 /* If index is in range of primary symtab, look it up there */
1482 if (ndx >= symtab->sym_symn_aux) {
1483 return (gelf_getsym(symtab->sym_data_pri,
1484 ndx - symtab->sym_symn_aux, dst));
1485 }
1486
1487 /* Not in primary: Look it up in the auxiliary symtab */
1488 return (gelf_getsym(symtab->sym_data_aux, ndx, dst));
1489 }
1490
1491 void
optimize_symtab(sym_tbl_t * symtab)1492 optimize_symtab(sym_tbl_t *symtab)
1493 {
1494 GElf_Sym *symp, *syms;
1495 uint_t i, *indexa, *indexb;
1496 size_t symn, strsz, count;
1497
1498 if (symtab == NULL || symtab->sym_data_pri == NULL ||
1499 symtab->sym_byaddr != NULL)
1500 return;
1501
1502 symn = symtab->sym_symn;
1503 strsz = symtab->sym_strsz;
1504
1505 symp = syms = malloc(sizeof (GElf_Sym) * symn);
1506 if (symp == NULL) {
1507 dprintf("optimize_symtab: failed to malloc symbol array");
1508 return;
1509 }
1510
1511 /*
1512 * First record all the symbols into a table and count up the ones
1513 * that we're interested in. We mark symbols as invalid by setting
1514 * the st_name to an illegal value.
1515 */
1516 for (i = 0, count = 0; i < symn; i++, symp++) {
1517 if (symtab_getsym(symtab, i, symp) != NULL &&
1518 symp->st_name < strsz &&
1519 IS_DATA_TYPE(GELF_ST_TYPE(symp->st_info)))
1520 count++;
1521 else
1522 symp->st_name = strsz;
1523 }
1524
1525 /*
1526 * Allocate sufficient space for both tables and populate them
1527 * with the same symbols we just counted.
1528 */
1529 symtab->sym_count = count;
1530 indexa = symtab->sym_byaddr = calloc(sizeof (uint_t), count);
1531 indexb = symtab->sym_byname = calloc(sizeof (uint_t), count);
1532 if (indexa == NULL || indexb == NULL) {
1533 dprintf(
1534 "optimize_symtab: failed to malloc symbol index arrays");
1535 symtab->sym_count = 0;
1536 if (indexa != NULL) { /* First alloc succeeded. Free it */
1537 free(indexa);
1538 symtab->sym_byaddr = NULL;
1539 }
1540 free(syms);
1541 return;
1542 }
1543 for (i = 0, symp = syms; i < symn; i++, symp++) {
1544 if (symp->st_name < strsz)
1545 *indexa++ = *indexb++ = i;
1546 }
1547
1548 /*
1549 * Sort the two tables according to the appropriate criteria,
1550 * unless the user has overridden this behaviour.
1551 *
1552 * An example where we might not sort the tables is the relatively
1553 * unusual case of a process with very large symbol tables in which
1554 * we perform few lookups. In such a case the total time would be
1555 * dominated by the sort. It is difficult to determine a priori
1556 * how many lookups an arbitrary client will perform, and
1557 * hence whether the symbol tables should be sorted. We therefore
1558 * sort the tables by default, but provide the user with a
1559 * "chicken switch" in the form of the LIBPROC_NO_QSORT
1560 * environment variable.
1561 */
1562 if (!_libproc_no_qsort) {
1563 (void) mutex_lock(&sort_mtx);
1564 sort_strs = symtab->sym_strs;
1565 sort_syms = syms;
1566
1567 qsort(symtab->sym_byaddr, count, sizeof (uint_t), byaddr_cmp);
1568 qsort(symtab->sym_byname, count, sizeof (uint_t), byname_cmp);
1569
1570 sort_strs = NULL;
1571 sort_syms = NULL;
1572 (void) mutex_unlock(&sort_mtx);
1573 }
1574
1575 free(syms);
1576 }
1577
1578
1579 static Elf *
build_fake_elf(struct ps_prochandle * P,file_info_t * fptr,GElf_Ehdr * ehdr,size_t * nshdrs,Elf_Data ** shdata)1580 build_fake_elf(struct ps_prochandle *P, file_info_t *fptr, GElf_Ehdr *ehdr,
1581 size_t *nshdrs, Elf_Data **shdata)
1582 {
1583 size_t shstrndx;
1584 Elf_Scn *scn;
1585 Elf *elf;
1586
1587 if ((elf = fake_elf(P, fptr)) == NULL ||
1588 elf_kind(elf) != ELF_K_ELF ||
1589 gelf_getehdr(elf, ehdr) == NULL ||
1590 elf_getshdrnum(elf, nshdrs) == -1 ||
1591 elf_getshdrstrndx(elf, &shstrndx) == -1 ||
1592 (scn = elf_getscn(elf, shstrndx)) == NULL ||
1593 (*shdata = elf_getdata(scn, NULL)) == NULL) {
1594 if (elf != NULL)
1595 (void) elf_end(elf);
1596 dprintf("failed to fake up ELF file\n");
1597 return (NULL);
1598 }
1599
1600 return (elf);
1601 }
1602
1603 /*
1604 * Try and find the file described by path in the file system and validate that
1605 * it matches our CRC before we try and process it for symbol information. If we
1606 * instead have an ELF data section, then that means we're checking a build-id
1607 * section instead. In that case we just need to find and bcmp the corresponding
1608 * section.
1609 *
1610 * Before we validate if it's a valid CRC or data section, we check to ensure
1611 * that it's a normal file and not anything else.
1612 */
1613 static boolean_t
build_alt_debug(file_info_t * fptr,const char * path,uint32_t crc,Elf_Data * data)1614 build_alt_debug(file_info_t *fptr, const char *path, uint32_t crc,
1615 Elf_Data *data)
1616 {
1617 int fd;
1618 struct stat st;
1619 Elf *elf;
1620 Elf_Scn *scn;
1621 GElf_Shdr symshdr, strshdr;
1622 Elf_Data *symdata, *strdata;
1623 boolean_t valid;
1624 uint32_t c = -1U;
1625
1626 if ((fd = open(path, O_RDONLY)) < 0)
1627 return (B_FALSE);
1628
1629 if (fstat(fd, &st) != 0) {
1630 (void) close(fd);
1631 return (B_FALSE);
1632 }
1633
1634 if (S_ISREG(st.st_mode) == 0) {
1635 (void) close(fd);
1636 return (B_FALSE);
1637 }
1638
1639 /*
1640 * Only check the CRC if we've come here through a GNU debug link
1641 * section as opposed to the build id. This is indicated by having the
1642 * value of data be NULL.
1643 */
1644 if (data == NULL) {
1645 for (;;) {
1646 char buf[4096];
1647 ssize_t ret = read(fd, buf, sizeof (buf));
1648 if (ret == -1) {
1649 if (ret == EINTR)
1650 continue;
1651 (void) close(fd);
1652 return (B_FALSE);
1653 }
1654 if (ret == 0) {
1655 c = ~c;
1656 if (c != crc) {
1657 dprintf("crc mismatch, found: 0x%x "
1658 "expected 0x%x\n", c, crc);
1659 (void) close(fd);
1660 return (B_FALSE);
1661 }
1662 break;
1663 }
1664 CRC32(c, buf, ret, c, psym_crc32);
1665 }
1666 }
1667
1668 elf = elf_begin(fd, ELF_C_READ, NULL);
1669 if (elf == NULL) {
1670 (void) close(fd);
1671 return (B_FALSE);
1672 }
1673
1674 if (elf_kind(elf) != ELF_K_ELF) {
1675 goto fail;
1676 }
1677
1678 /*
1679 * If we have a data section, that indicates we have a build-id which
1680 * means we need to find the corresponding build-id section and compare
1681 * it.
1682 */
1683 scn = NULL;
1684 valid = B_FALSE;
1685 for (scn = elf_nextscn(elf, scn); data != NULL && scn != NULL;
1686 scn = elf_nextscn(elf, scn)) {
1687 GElf_Shdr hdr;
1688 Elf_Data *ntdata;
1689
1690 if (gelf_getshdr(scn, &hdr) == NULL)
1691 goto fail;
1692
1693 if (hdr.sh_type != SHT_NOTE)
1694 continue;
1695
1696 if ((ntdata = elf_getdata(scn, NULL)) == NULL)
1697 goto fail;
1698
1699 /*
1700 * First verify the data section sizes are equal, then the
1701 * section name. If that's all true, then we can just do a bcmp.
1702 */
1703 if (data->d_size != ntdata->d_size)
1704 continue;
1705
1706 dprintf("found corresponding section in alternate file\n");
1707 if (bcmp(ntdata->d_buf, data->d_buf, data->d_size) != 0)
1708 goto fail;
1709
1710 valid = B_TRUE;
1711 break;
1712 }
1713 if (data != NULL && valid == B_FALSE) {
1714 dprintf("failed to find a matching %s section in %s\n",
1715 BUILDID_NAME, path);
1716 goto fail;
1717 }
1718
1719
1720 /*
1721 * Do two passes, first see if we have a symbol header, then see if we
1722 * can find the corresponding linked string table.
1723 */
1724 scn = NULL;
1725 for (scn = elf_nextscn(elf, scn); scn != NULL;
1726 scn = elf_nextscn(elf, scn)) {
1727
1728 if (gelf_getshdr(scn, &symshdr) == NULL)
1729 goto fail;
1730
1731 if (symshdr.sh_type != SHT_SYMTAB)
1732 continue;
1733
1734 if ((symdata = elf_getdata(scn, NULL)) == NULL)
1735 goto fail;
1736
1737 break;
1738 }
1739 if (scn == NULL)
1740 goto fail;
1741
1742 if ((scn = elf_getscn(elf, symshdr.sh_link)) == NULL)
1743 goto fail;
1744
1745 if (gelf_getshdr(scn, &strshdr) == NULL)
1746 goto fail;
1747
1748 if ((strdata = elf_getdata(scn, NULL)) == NULL)
1749 goto fail;
1750
1751 fptr->file_symtab.sym_data_pri = symdata;
1752 fptr->file_symtab.sym_symn += symshdr.sh_size / symshdr.sh_entsize;
1753 fptr->file_symtab.sym_strs = strdata->d_buf;
1754 fptr->file_symtab.sym_strsz = strdata->d_size;
1755 fptr->file_symtab.sym_hdr_pri = symshdr;
1756 fptr->file_symtab.sym_strhdr = strshdr;
1757
1758 dprintf("successfully loaded additional debug symbols for %s from %s\n",
1759 fptr->file_rname, path);
1760
1761 fptr->file_dbgfile = fd;
1762 fptr->file_dbgelf = elf;
1763 return (B_TRUE);
1764 fail:
1765 (void) elf_end(elf);
1766 (void) close(fd);
1767 return (B_FALSE);
1768 }
1769
1770 /*
1771 * We're here because the object in question has no symbol information, that's a
1772 * bit unfortunate. However, we've found that there's a .gnu_debuglink sitting
1773 * around. By convention that means that given the current location of the
1774 * object on disk, and the debug name that we found in the binary we need to
1775 * search the following locations for a matching file.
1776 *
1777 * <dirname>/.debug/<debug-name>
1778 * /usr/lib/debug/<dirname>/<debug-name>
1779 *
1780 * In the future, we should consider supporting looking in the prefix's
1781 * lib/debug directory for a matching object or supporting an arbitrary user
1782 * defined set of places to look.
1783 */
1784 static void
find_alt_debuglink(file_info_t * fptr,const char * name,uint32_t crc)1785 find_alt_debuglink(file_info_t *fptr, const char *name, uint32_t crc)
1786 {
1787 boolean_t r;
1788 char *dup = NULL, *path = NULL, *dname;
1789
1790 dprintf("find_alt_debug: looking for %s, crc 0x%x\n", name, crc);
1791 if (fptr->file_rname == NULL) {
1792 dprintf("find_alt_debug: encountered null file_rname\n");
1793 return;
1794 }
1795
1796 dup = strdup(fptr->file_rname);
1797 if (dup == NULL)
1798 return;
1799
1800 dname = dirname(dup);
1801 if (asprintf(&path, "%s/.debug/%s", dname, name) != -1) {
1802 dprintf("attempting to load alternate debug information "
1803 "from %s\n", path);
1804 r = build_alt_debug(fptr, path, crc, NULL);
1805 free(path);
1806 if (r == B_TRUE)
1807 goto out;
1808 }
1809
1810 if (asprintf(&path, "/usr/lib/debug/%s/%s", dname, name) != -1) {
1811 dprintf("attempting to load alternate debug information "
1812 "from %s\n", path);
1813 r = build_alt_debug(fptr, path, crc, NULL);
1814 free(path);
1815 if (r == B_TRUE)
1816 goto out;
1817 }
1818 out:
1819 free(dup);
1820 }
1821
1822 /*
1823 * Build the symbol table for the given mapped file.
1824 */
1825 void
Pbuild_file_symtab(struct ps_prochandle * P,file_info_t * fptr)1826 Pbuild_file_symtab(struct ps_prochandle *P, file_info_t *fptr)
1827 {
1828 char objectfile[PATH_MAX];
1829 uint_t i;
1830
1831 GElf_Ehdr ehdr;
1832 GElf_Sym s;
1833
1834 Elf_Data *shdata;
1835 Elf_Scn *scn;
1836 Elf *elf;
1837 size_t nshdrs, shstrndx;
1838
1839 struct {
1840 GElf_Shdr c_shdr;
1841 Elf_Data *c_data;
1842 const char *c_name;
1843 } *cp, *cache = NULL, *dyn = NULL, *plt = NULL, *ctf = NULL,
1844 *dbglink = NULL, *buildid = NULL;
1845
1846 if (fptr->file_init)
1847 return; /* We've already processed this file */
1848
1849 /*
1850 * Mark the file_info struct as having the symbol table initialized
1851 * even if we fail below. We tried once; we don't try again.
1852 */
1853 fptr->file_init = 1;
1854
1855 if (elf_version(EV_CURRENT) == EV_NONE) {
1856 dprintf("libproc ELF version is more recent than libelf\n");
1857 return;
1858 }
1859
1860 if (P->state == PS_DEAD || P->state == PS_IDLE) {
1861 char *name;
1862 /*
1863 * If we're a not live, we can't open files from the /proc
1864 * object directory; we have only the mapping and file names
1865 * to guide us. We prefer the file_lname, but need to handle
1866 * the case of it being NULL in order to bootstrap: we first
1867 * come here during rd_new() when the only information we have
1868 * is interpreter name associated with the AT_BASE mapping.
1869 *
1870 * Also, if the zone associated with the core file seems
1871 * to exists on this machine we'll try to open the object
1872 * file within the zone.
1873 */
1874 if (fptr->file_rname != NULL)
1875 name = fptr->file_rname;
1876 else if (fptr->file_lname != NULL)
1877 name = fptr->file_lname;
1878 else
1879 name = fptr->file_pname;
1880 (void) strlcpy(objectfile, name, sizeof (objectfile));
1881 } else {
1882 (void) snprintf(objectfile, sizeof (objectfile),
1883 "%s/%d/object/%s",
1884 procfs_path, (int)P->pid, fptr->file_pname);
1885 }
1886
1887 /*
1888 * Open the object file, create the elf file, and then get the elf
1889 * header and .shstrtab data buffer so we can process sections by
1890 * name. If anything goes wrong try to fake up an elf file from
1891 * the in-core elf image.
1892 */
1893
1894 if (_libproc_incore_elf || (P->flags & INCORE)) {
1895 dprintf("Pbuild_file_symtab: using in-core data for: %s\n",
1896 fptr->file_pname);
1897
1898 if ((elf = build_fake_elf(P, fptr, &ehdr, &nshdrs, &shdata)) ==
1899 NULL)
1900 return;
1901
1902 } else if ((fptr->file_fd = open(objectfile, O_RDONLY)) < 0) {
1903 dprintf("Pbuild_file_symtab: failed to open %s: %s\n",
1904 objectfile, strerror(errno));
1905
1906 if ((elf = build_fake_elf(P, fptr, &ehdr, &nshdrs, &shdata)) ==
1907 NULL)
1908 return;
1909
1910 } else if ((elf = elf_begin(fptr->file_fd, ELF_C_READ, NULL)) == NULL ||
1911 elf_kind(elf) != ELF_K_ELF ||
1912 gelf_getehdr(elf, &ehdr) == NULL ||
1913 elf_getshdrnum(elf, &nshdrs) == -1 ||
1914 elf_getshdrstrndx(elf, &shstrndx) == -1 ||
1915 (scn = elf_getscn(elf, shstrndx)) == NULL ||
1916 (shdata = elf_getdata(scn, NULL)) == NULL) {
1917 int err = elf_errno();
1918
1919 dprintf("failed to process ELF file %s: %s\n",
1920 objectfile, (err == 0) ? "<null>" : elf_errmsg(err));
1921 (void) elf_end(elf);
1922
1923 if ((elf = build_fake_elf(P, fptr, &ehdr, &nshdrs, &shdata)) ==
1924 NULL)
1925 return;
1926
1927 } else if (file_differs(P, elf, fptr)) {
1928 Elf *newelf;
1929
1930 /*
1931 * Before we get too excited about this elf file, we'll check
1932 * its checksum value against the value we have in memory. If
1933 * they don't agree, we try to fake up a new elf file and
1934 * proceed with that instead.
1935 */
1936 dprintf("ELF file %s (%lx) doesn't match in-core image\n",
1937 fptr->file_pname,
1938 (ulong_t)fptr->file_map->map_pmap.pr_vaddr);
1939
1940 if ((newelf = build_fake_elf(P, fptr, &ehdr, &nshdrs, &shdata))
1941 != NULL) {
1942 (void) elf_end(elf);
1943 elf = newelf;
1944 dprintf("switched to faked up ELF file\n");
1945
1946 /*
1947 * Check to see if the file that we just discovered
1948 * to be an imposter matches the execname that was
1949 * determined by Pfindexec(). If it does, we (clearly)
1950 * don't have the right binary, and we zero out
1951 * execname before anyone gets hurt.
1952 */
1953 if (fptr->file_rname != NULL && P->execname != NULL &&
1954 strcmp(fptr->file_rname, P->execname) == 0) {
1955 dprintf("file/in-core image mismatch was "
1956 "on P->execname; discarding\n");
1957 free(P->execname);
1958 P->execname = NULL;
1959 }
1960 }
1961 }
1962
1963 if ((cache = malloc(nshdrs * sizeof (*cache))) == NULL) {
1964 dprintf("failed to malloc section cache for %s\n", objectfile);
1965 goto bad;
1966 }
1967
1968 dprintf("processing ELF file %s\n", objectfile);
1969 fptr->file_class = ehdr.e_ident[EI_CLASS];
1970 fptr->file_etype = ehdr.e_type;
1971 fptr->file_elf = elf;
1972 fptr->file_shstrs = shdata->d_buf;
1973 fptr->file_shstrsz = shdata->d_size;
1974
1975 /*
1976 * Iterate through each section, caching its section header, data
1977 * pointer, and name. We use this for handling sh_link values below.
1978 */
1979 for (cp = cache + 1, scn = NULL; scn = elf_nextscn(elf, scn); cp++) {
1980 if (gelf_getshdr(scn, &cp->c_shdr) == NULL) {
1981 dprintf("Pbuild_file_symtab: Failed to get section "
1982 "header\n");
1983 goto bad; /* Failed to get section header */
1984 }
1985
1986 if ((cp->c_data = elf_getdata(scn, NULL)) == NULL) {
1987 dprintf("Pbuild_file_symtab: Failed to get section "
1988 "data\n");
1989 goto bad; /* Failed to get section data */
1990 }
1991
1992 if (cp->c_shdr.sh_name >= shdata->d_size) {
1993 dprintf("Pbuild_file_symtab: corrupt section name");
1994 goto bad; /* Corrupt section name */
1995 }
1996
1997 cp->c_name = (const char *)shdata->d_buf + cp->c_shdr.sh_name;
1998 }
1999
2000 /*
2001 * Now iterate through the section cache in order to locate info
2002 * for the .symtab, .dynsym, .SUNW_ldynsym, .dynamic, .plt,
2003 * and .SUNW_ctf sections:
2004 */
2005 for (i = 1, cp = cache + 1; i < nshdrs; i++, cp++) {
2006 GElf_Shdr *shp = &cp->c_shdr;
2007
2008 if (shp->sh_type == SHT_SYMTAB || shp->sh_type == SHT_DYNSYM) {
2009 sym_tbl_t *symp = shp->sh_type == SHT_SYMTAB ?
2010 &fptr->file_symtab : &fptr->file_dynsym;
2011 /*
2012 * It's possible that the we already got the symbol
2013 * table from the core file itself. Either the file
2014 * differs in which case our faked up elf file will
2015 * only contain the dynsym (not the symtab) or the
2016 * file matches in which case we'll just be replacing
2017 * the symbol table we pulled out of the core file
2018 * with an equivalent one. In either case, this
2019 * check isn't essential, but it's a good idea.
2020 */
2021 if (symp->sym_data_pri == NULL) {
2022 dprintf("Symbol table found for %s\n",
2023 objectfile);
2024 symp->sym_data_pri = cp->c_data;
2025 symp->sym_symn +=
2026 shp->sh_size / shp->sh_entsize;
2027 symp->sym_strs =
2028 cache[shp->sh_link].c_data->d_buf;
2029 symp->sym_strsz =
2030 cache[shp->sh_link].c_data->d_size;
2031 symp->sym_hdr_pri = cp->c_shdr;
2032 symp->sym_strhdr = cache[shp->sh_link].c_shdr;
2033 } else {
2034 dprintf("Symbol table already there for %s\n",
2035 objectfile);
2036 }
2037 } else if (shp->sh_type == SHT_SUNW_LDYNSYM) {
2038 /* .SUNW_ldynsym section is auxiliary to .dynsym */
2039 if (fptr->file_dynsym.sym_data_aux == NULL) {
2040 dprintf(".SUNW_ldynsym symbol table"
2041 " found for %s\n", objectfile);
2042 fptr->file_dynsym.sym_data_aux = cp->c_data;
2043 fptr->file_dynsym.sym_symn_aux =
2044 shp->sh_size / shp->sh_entsize;
2045 fptr->file_dynsym.sym_symn +=
2046 fptr->file_dynsym.sym_symn_aux;
2047 fptr->file_dynsym.sym_hdr_aux = cp->c_shdr;
2048 } else {
2049 dprintf(".SUNW_ldynsym symbol table already"
2050 " there for %s\n", objectfile);
2051 }
2052 } else if (shp->sh_type == SHT_DYNAMIC) {
2053 dyn = cp;
2054 } else if (strcmp(cp->c_name, ".plt") == 0) {
2055 plt = cp;
2056 } else if (strcmp(cp->c_name, ".SUNW_ctf") == 0) {
2057 /*
2058 * Skip over bogus CTF sections so they don't come back
2059 * to haunt us later.
2060 */
2061 if (shp->sh_link == 0 ||
2062 shp->sh_link >= nshdrs ||
2063 (cache[shp->sh_link].c_shdr.sh_type != SHT_DYNSYM &&
2064 cache[shp->sh_link].c_shdr.sh_type != SHT_SYMTAB)) {
2065 dprintf("Bad sh_link %d for "
2066 "CTF\n", shp->sh_link);
2067 continue;
2068 }
2069 ctf = cp;
2070 } else if (strcmp(cp->c_name, BUILDID_NAME) == 0) {
2071 dprintf("Found a %s section for %s\n", BUILDID_NAME,
2072 fptr->file_rname);
2073 /* The ElfXX_Nhdr is 32/64-bit neutral */
2074 if (cp->c_shdr.sh_type == SHT_NOTE &&
2075 cp->c_data->d_buf != NULL &&
2076 cp->c_data->d_size >= sizeof (Elf32_Nhdr)) {
2077 Elf32_Nhdr *hdr = cp->c_data->d_buf;
2078 if (hdr->n_type != 3)
2079 continue;
2080 if (hdr->n_namesz != 4)
2081 continue;
2082 if (hdr->n_descsz < MINBUILDID)
2083 continue;
2084 /* Set a reasonable upper bound */
2085 if (hdr->n_descsz > MAXBUILDID) {
2086 dprintf("Skipped %s as too large "
2087 "(%ld)\n", BUILDID_NAME,
2088 (unsigned long)hdr->n_descsz);
2089 continue;
2090 }
2091
2092 if (cp->c_data->d_size < sizeof (hdr) +
2093 hdr->n_namesz + hdr->n_descsz)
2094 continue;
2095 buildid = cp;
2096 }
2097 } else if (strcmp(cp->c_name, DBGLINK_NAME) == 0) {
2098 dprintf("found %s section for %s\n", DBGLINK_NAME,
2099 fptr->file_rname);
2100 /*
2101 * Let's make sure of a few things before we do this.
2102 */
2103 if (cp->c_shdr.sh_type == SHT_PROGBITS &&
2104 cp->c_data->d_buf != NULL &&
2105 cp->c_data->d_size) {
2106 dbglink = cp;
2107 }
2108 }
2109 }
2110
2111 /*
2112 * If we haven't found any symbol table information and we have found
2113 * either a .note.gnu.build-id or a .gnu_debuglink, it's time to try and
2114 * figure out where we might find this. Originally, GNU used the
2115 * .gnu_debuglink solely, but then they added a .note.gnu.build-id. The
2116 * build-id is some size, usually 16 or 20 bytes, often a SHA1 sum of
2117 * parts of the original file. This is maintained across all versions of
2118 * the subsequent file.
2119 *
2120 * For the .note.gnu.build-id, we're going to check a few things before
2121 * using it, first that the name is 4 bytes, and is GNU and that the
2122 * type is 3, which they say is the build-id identifier.
2123 *
2124 * To verify that the elf data for the .gnu_debuglink seems somewhat
2125 * sane, eg. the elf data should be a string, so we want to verify we
2126 * have a null-terminator.
2127 */
2128 if (fptr->file_symtab.sym_data_pri == NULL && buildid != NULL) {
2129 int i, bo;
2130 uint8_t *dp;
2131 char buf[BUILDID_STRLEN], *path;
2132 Elf32_Nhdr *hdr = buildid->c_data->d_buf;
2133
2134 /*
2135 * This was checked for validity when assigning the buildid
2136 * variable.
2137 */
2138 bzero(buf, sizeof (buf));
2139 dp = (uint8_t *)((uintptr_t)hdr + sizeof (*hdr) +
2140 hdr->n_namesz);
2141 for (i = 0, bo = 0; i < hdr->n_descsz; i++, bo += 2, dp++) {
2142 assert(sizeof (buf) - bo > 0);
2143
2144 /*
2145 * Recall that the build-id is structured as a series of
2146 * bytes. However, the first two characters are supposed
2147 * to represent a directory. Hence, once we reach offset
2148 * two, we insert a '/' character.
2149 */
2150 if (bo == 2) {
2151 buf[bo] = '/';
2152 bo++;
2153 }
2154 (void) snprintf(buf + bo, sizeof (buf) - bo, "%2x",
2155 *dp);
2156 }
2157
2158 if (asprintf(&path, "/usr/lib/debug/.build-id/%s.debug",
2159 buf) != -1) {
2160 boolean_t r;
2161 dprintf("attempting to find build id alternate debug "
2162 "file at %s\n", path);
2163 r = build_alt_debug(fptr, path, 0, buildid->c_data);
2164 dprintf("attempt %s\n", r == B_TRUE ?
2165 "succeeded" : "failed");
2166 free(path);
2167 } else {
2168 dprintf("failed to construct build id path: %s\n",
2169 strerror(errno));
2170 }
2171 }
2172
2173 if (fptr->file_symtab.sym_data_pri == NULL && dbglink != NULL) {
2174 char *c = dbglink->c_data->d_buf;
2175 size_t i;
2176 boolean_t found = B_FALSE;
2177 Elf_Data *ed = dbglink->c_data;
2178 uint32_t crc;
2179
2180 for (i = 0; i < ed->d_size; i++) {
2181 if (c[i] == '\0') {
2182 uintptr_t off;
2183 dprintf("got .gnu_debuglink terminator at "
2184 "offset %lu\n", (unsigned long)i);
2185 /*
2186 * After the null terminator, there should be
2187 * padding, followed by a 4 byte CRC of the
2188 * file. If we don't see this, we're going to
2189 * assume this is bogus.
2190 */
2191 if ((i % sizeof (uint32_t)) == 0) {
2192 i += 4;
2193 } else {
2194 i += sizeof (uint32_t) -
2195 (i % sizeof (uint32_t));
2196 }
2197 if (i + sizeof (uint32_t) ==
2198 dbglink->c_data->d_size) {
2199 found = B_TRUE;
2200 off = (uintptr_t)ed->d_buf + i;
2201 crc = *(uint32_t *)off;
2202 } else {
2203 dprintf(".gnu_debuglink size mismatch, "
2204 "expected: %lu, found: %lu\n",
2205 (unsigned long)i,
2206 (unsigned long)ed->d_size);
2207 }
2208 break;
2209 }
2210 }
2211
2212 if (found == B_TRUE)
2213 find_alt_debuglink(fptr, dbglink->c_data->d_buf, crc);
2214 }
2215
2216 /*
2217 * At this point, we've found all the symbol tables we're ever going
2218 * to find: the ones in the loop above and possibly the symtab that
2219 * was included in the core file. Before we perform any lookups, we
2220 * create sorted versions to optimize for lookups.
2221 */
2222 optimize_symtab(&fptr->file_symtab);
2223 optimize_symtab(&fptr->file_dynsym);
2224
2225 /*
2226 * Fill in the base address of the text mapping for shared libraries.
2227 * This allows us to translate symbols before librtld_db is ready.
2228 */
2229 if (fptr->file_etype == ET_DYN) {
2230 fptr->file_dyn_base = fptr->file_map->map_pmap.pr_vaddr -
2231 fptr->file_map->map_pmap.pr_offset;
2232 dprintf("setting file_dyn_base for %s to %lx\n",
2233 objectfile, (long)fptr->file_dyn_base);
2234 }
2235
2236 /*
2237 * Record the CTF section information in the file info structure.
2238 */
2239 if (ctf != NULL) {
2240 fptr->file_ctf_off = ctf->c_shdr.sh_offset;
2241 fptr->file_ctf_size = ctf->c_shdr.sh_size;
2242 if (ctf->c_shdr.sh_link != 0 &&
2243 cache[ctf->c_shdr.sh_link].c_shdr.sh_type == SHT_DYNSYM)
2244 fptr->file_ctf_dyn = 1;
2245 }
2246
2247 if (fptr->file_lo == NULL)
2248 goto done; /* Nothing else to do if no load object info */
2249
2250 /*
2251 * If the object is a shared library and we have a different rl_base
2252 * value, reset file_dyn_base according to librtld_db's information.
2253 */
2254 if (fptr->file_etype == ET_DYN &&
2255 fptr->file_lo->rl_base != fptr->file_dyn_base) {
2256 dprintf("resetting file_dyn_base for %s to %lx\n",
2257 objectfile, (long)fptr->file_lo->rl_base);
2258 fptr->file_dyn_base = fptr->file_lo->rl_base;
2259 }
2260
2261 /*
2262 * Fill in the PLT information for this file if a PLT symbol is found.
2263 */
2264 if (sym_by_name(&fptr->file_dynsym, "_PROCEDURE_LINKAGE_TABLE_", &s,
2265 NULL) != NULL) {
2266 fptr->file_plt_base = s.st_value + fptr->file_dyn_base;
2267 fptr->file_plt_size = (plt != NULL) ? plt->c_shdr.sh_size : 0;
2268
2269 /*
2270 * Bring the load object up to date; it is the only way the
2271 * user has to access the PLT data. The PLT information in the
2272 * rd_loadobj_t is not set in the call to map_iter() (the
2273 * callback for rd_loadobj_iter) where we set file_lo.
2274 */
2275 fptr->file_lo->rl_plt_base = fptr->file_plt_base;
2276 fptr->file_lo->rl_plt_size = fptr->file_plt_size;
2277
2278 dprintf("PLT found at %p, size = %lu\n",
2279 (void *)fptr->file_plt_base, (ulong_t)fptr->file_plt_size);
2280 }
2281
2282 /*
2283 * Fill in the PLT information.
2284 */
2285 if (dyn != NULL) {
2286 uintptr_t dynaddr = dyn->c_shdr.sh_addr + fptr->file_dyn_base;
2287 size_t ndyn = dyn->c_shdr.sh_size / dyn->c_shdr.sh_entsize;
2288 GElf_Dyn d;
2289
2290 for (i = 0; i < ndyn; i++) {
2291 if (gelf_getdyn(dyn->c_data, i, &d) == NULL)
2292 continue;
2293
2294 switch (d.d_tag) {
2295 case DT_JMPREL:
2296 dprintf("DT_JMPREL is %p\n",
2297 (void *)(uintptr_t)d.d_un.d_ptr);
2298 fptr->file_jmp_rel =
2299 d.d_un.d_ptr + fptr->file_dyn_base;
2300 break;
2301 case DT_STRTAB:
2302 dprintf("DT_STRTAB is %p\n",
2303 (void *)(uintptr_t)d.d_un.d_ptr);
2304 break;
2305 case DT_PLTGOT:
2306 dprintf("DT_PLTGOT is %p\n",
2307 (void *)(uintptr_t)d.d_un.d_ptr);
2308 break;
2309 case DT_SUNW_SYMTAB:
2310 dprintf("DT_SUNW_SYMTAB is %p\n",
2311 (void *)(uintptr_t)d.d_un.d_ptr);
2312 break;
2313 case DT_SYMTAB:
2314 dprintf("DT_SYMTAB is %p\n",
2315 (void *)(uintptr_t)d.d_un.d_ptr);
2316 break;
2317 case DT_HASH:
2318 dprintf("DT_HASH is %p\n",
2319 (void *)(uintptr_t)d.d_un.d_ptr);
2320 break;
2321 }
2322 }
2323
2324 dprintf("_DYNAMIC found at %p, %lu entries, DT_JMPREL = %p\n",
2325 (void *)dynaddr, (ulong_t)ndyn, (void *)fptr->file_jmp_rel);
2326 }
2327
2328 done:
2329 free(cache);
2330 return;
2331
2332 bad:
2333 if (cache != NULL)
2334 free(cache);
2335
2336 (void) elf_end(elf);
2337 fptr->file_elf = NULL;
2338 if (fptr->file_elfmem != NULL) {
2339 free(fptr->file_elfmem);
2340 fptr->file_elfmem = NULL;
2341 }
2342 (void) close(fptr->file_fd);
2343 if (fptr->file_dbgelf != NULL)
2344 (void) elf_end(fptr->file_dbgelf);
2345 fptr->file_dbgelf = NULL;
2346 if (fptr->file_dbgfile >= 0)
2347 (void) close(fptr->file_dbgfile);
2348 fptr->file_fd = -1;
2349 fptr->file_dbgfile = -1;
2350 }
2351
2352 /*
2353 * Given a process virtual address, return the map_info_t containing it.
2354 * If none found, return NULL.
2355 */
2356 map_info_t *
Paddr2mptr(struct ps_prochandle * P,uintptr_t addr)2357 Paddr2mptr(struct ps_prochandle *P, uintptr_t addr)
2358 {
2359 int lo = 0;
2360 int hi = P->map_count - 1;
2361 int mid;
2362 map_info_t *mp;
2363
2364 while (lo <= hi) {
2365
2366 mid = (lo + hi) / 2;
2367 mp = &P->mappings[mid];
2368
2369 /* check that addr is in [vaddr, vaddr + size) */
2370 if ((addr - mp->map_pmap.pr_vaddr) < mp->map_pmap.pr_size)
2371 return (mp);
2372
2373 if (addr < mp->map_pmap.pr_vaddr)
2374 hi = mid - 1;
2375 else
2376 lo = mid + 1;
2377 }
2378
2379 return (NULL);
2380 }
2381
2382 /*
2383 * Return the map_info_t for the executable file.
2384 * If not found, return NULL.
2385 */
2386 static map_info_t *
exec_map(struct ps_prochandle * P)2387 exec_map(struct ps_prochandle *P)
2388 {
2389 uint_t i;
2390 map_info_t *mptr;
2391 map_info_t *mold = NULL;
2392 file_info_t *fptr;
2393 uintptr_t base;
2394
2395 for (i = 0, mptr = P->mappings; i < P->map_count; i++, mptr++) {
2396 if (mptr->map_pmap.pr_mapname[0] == '\0')
2397 continue;
2398 if (strcmp(mptr->map_pmap.pr_mapname, "a.out") == 0) {
2399 if ((fptr = mptr->map_file) != NULL &&
2400 fptr->file_lo != NULL) {
2401 base = fptr->file_lo->rl_base;
2402 if (base >= mptr->map_pmap.pr_vaddr &&
2403 base < mptr->map_pmap.pr_vaddr +
2404 mptr->map_pmap.pr_size) /* text space */
2405 return (mptr);
2406 mold = mptr; /* must be the data */
2407 continue;
2408 }
2409 /* This is a poor way to test for text space */
2410 if (!(mptr->map_pmap.pr_mflags & MA_EXEC) ||
2411 (mptr->map_pmap.pr_mflags & MA_WRITE)) {
2412 mold = mptr;
2413 continue;
2414 }
2415 return (mptr);
2416 }
2417 }
2418
2419 return (mold);
2420 }
2421
2422 /*
2423 * Given a shared object name, return the map_info_t for it. If no matching
2424 * object is found, return NULL. Normally, the link maps contain the full
2425 * object pathname, e.g. /usr/lib/libc.so.1. We allow the object name to
2426 * take one of the following forms:
2427 *
2428 * 1. An exact match (i.e. a full pathname): "/usr/lib/libc.so.1"
2429 * 2. An exact basename match: "libc.so.1"
2430 * 3. An initial basename match up to a '.' suffix: "libc.so" or "libc"
2431 * 4. The literal string "a.out" is an alias for the executable mapping
2432 *
2433 * The third case is a convenience for callers and may not be necessary.
2434 *
2435 * As the exact same object name may be loaded on different link maps (see
2436 * dlmopen(3DL)), we also allow the caller to resolve the object name by
2437 * specifying a particular link map id. If lmid is PR_LMID_EVERY, the
2438 * first matching name will be returned, regardless of the link map id.
2439 */
2440 static map_info_t *
object_to_map(struct ps_prochandle * P,Lmid_t lmid,const char * objname)2441 object_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *objname)
2442 {
2443 map_info_t *mp;
2444 file_info_t *fp;
2445 size_t objlen;
2446 uint_t i;
2447
2448 /*
2449 * If we have no rtld_db, then always treat a request as one for all
2450 * link maps.
2451 */
2452 if (P->rap == NULL)
2453 lmid = PR_LMID_EVERY;
2454
2455 /*
2456 * First pass: look for exact matches of the entire pathname or
2457 * basename (cases 1 and 2 above):
2458 */
2459 for (i = 0, mp = P->mappings; i < P->map_count; i++, mp++) {
2460
2461 if (mp->map_pmap.pr_mapname[0] == '\0' ||
2462 (fp = mp->map_file) == NULL ||
2463 ((fp->file_lname == NULL) && (fp->file_rname == NULL)))
2464 continue;
2465
2466 if (lmid != PR_LMID_EVERY &&
2467 (fp->file_lo == NULL || lmid != fp->file_lo->rl_lmident))
2468 continue;
2469
2470 /*
2471 * If we match, return the primary text mapping; otherwise
2472 * just return the mapping we matched.
2473 */
2474 if ((fp->file_lbase && strcmp(fp->file_lbase, objname) == 0) ||
2475 (fp->file_rbase && strcmp(fp->file_rbase, objname) == 0) ||
2476 (fp->file_lname && strcmp(fp->file_lname, objname) == 0) ||
2477 (fp->file_rname && strcmp(fp->file_rname, objname) == 0))
2478 return (fp->file_map ? fp->file_map : mp);
2479 }
2480
2481 objlen = strlen(objname);
2482
2483 /*
2484 * Second pass: look for partial matches (case 3 above):
2485 */
2486 for (i = 0, mp = P->mappings; i < P->map_count; i++, mp++) {
2487
2488 if (mp->map_pmap.pr_mapname[0] == '\0' ||
2489 (fp = mp->map_file) == NULL ||
2490 ((fp->file_lname == NULL) && (fp->file_rname == NULL)))
2491 continue;
2492
2493 if (lmid != PR_LMID_EVERY &&
2494 (fp->file_lo == NULL || lmid != fp->file_lo->rl_lmident))
2495 continue;
2496
2497 /*
2498 * If we match, return the primary text mapping; otherwise
2499 * just return the mapping we matched.
2500 */
2501 if ((fp->file_lbase != NULL) &&
2502 (strncmp(fp->file_lbase, objname, objlen) == 0) &&
2503 (fp->file_lbase[objlen] == '.'))
2504 return (fp->file_map ? fp->file_map : mp);
2505 if ((fp->file_rbase != NULL) &&
2506 (strncmp(fp->file_rbase, objname, objlen) == 0) &&
2507 (fp->file_rbase[objlen] == '.'))
2508 return (fp->file_map ? fp->file_map : mp);
2509 }
2510
2511 /*
2512 * One last check: we allow "a.out" to always alias the executable,
2513 * assuming this name was not in use for something else.
2514 */
2515 if ((lmid == PR_LMID_EVERY || lmid == LM_ID_BASE) &&
2516 (strcmp(objname, "a.out") == 0))
2517 return (P->map_exec);
2518
2519 return (NULL);
2520 }
2521
2522 static map_info_t *
object_name_to_map(struct ps_prochandle * P,Lmid_t lmid,const char * name)2523 object_name_to_map(struct ps_prochandle *P, Lmid_t lmid, const char *name)
2524 {
2525 map_info_t *mptr;
2526
2527 if (!P->info_valid)
2528 Pupdate_maps(P);
2529
2530 if (P->map_exec == NULL && ((mptr = Paddr2mptr(P,
2531 Pgetauxval(P, AT_ENTRY))) != NULL || (mptr = exec_map(P)) != NULL))
2532 P->map_exec = mptr;
2533
2534 if (P->map_ldso == NULL && (mptr = Paddr2mptr(P,
2535 Pgetauxval(P, AT_BASE))) != NULL)
2536 P->map_ldso = mptr;
2537
2538 if (name == PR_OBJ_EXEC)
2539 mptr = P->map_exec;
2540 else if (name == PR_OBJ_LDSO)
2541 mptr = P->map_ldso;
2542 else if (Prd_agent(P) != NULL || P->state == PS_IDLE)
2543 mptr = object_to_map(P, lmid, name);
2544 else
2545 mptr = NULL;
2546
2547 return (mptr);
2548 }
2549
2550 /*
2551 * When two symbols are found by address, decide which one is to be preferred.
2552 */
2553 static GElf_Sym *
sym_prefer(GElf_Sym * sym1,char * name1,GElf_Sym * sym2,char * name2)2554 sym_prefer(GElf_Sym *sym1, char *name1, GElf_Sym *sym2, char *name2)
2555 {
2556 /*
2557 * Prefer the non-NULL symbol.
2558 */
2559 if (sym1 == NULL)
2560 return (sym2);
2561 if (sym2 == NULL)
2562 return (sym1);
2563
2564 /*
2565 * Defer to the sort ordering...
2566 */
2567 return (byaddr_cmp_common(sym1, name1, sym2, name2) <= 0 ? sym1 : sym2);
2568 }
2569
2570 /*
2571 * Use a binary search to do the work of sym_by_addr().
2572 */
2573 static GElf_Sym *
sym_by_addr_binary(sym_tbl_t * symtab,GElf_Addr addr,GElf_Sym * symp,uint_t * idp)2574 sym_by_addr_binary(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symp,
2575 uint_t *idp)
2576 {
2577 GElf_Sym sym, osym;
2578 uint_t i, oid, *byaddr = symtab->sym_byaddr;
2579 int min, max, mid, omid, found = 0;
2580
2581 if (symtab->sym_data_pri == NULL || symtab->sym_count == 0)
2582 return (NULL);
2583
2584 min = 0;
2585 max = symtab->sym_count - 1;
2586 osym.st_value = 0;
2587
2588 /*
2589 * We can't return when we've found a match, we have to continue
2590 * searching for the closest matching symbol.
2591 */
2592 while (min <= max) {
2593 mid = (max + min) / 2;
2594
2595 i = byaddr[mid];
2596 (void) symtab_getsym(symtab, i, &sym);
2597
2598 if (addr >= sym.st_value &&
2599 addr < sym.st_value + sym.st_size &&
2600 (!found || sym.st_value > osym.st_value)) {
2601 osym = sym;
2602 omid = mid;
2603 oid = i;
2604 found = 1;
2605 }
2606
2607 if (addr < sym.st_value)
2608 max = mid - 1;
2609 else
2610 min = mid + 1;
2611 }
2612
2613 if (!found)
2614 return (NULL);
2615
2616 /*
2617 * There may be many symbols with identical values so we walk
2618 * backward in the byaddr table to find the best match.
2619 */
2620 do {
2621 sym = osym;
2622 i = oid;
2623
2624 if (omid == 0)
2625 break;
2626
2627 oid = byaddr[--omid];
2628 (void) symtab_getsym(symtab, oid, &osym);
2629 } while (addr >= osym.st_value &&
2630 addr < sym.st_value + osym.st_size &&
2631 osym.st_value == sym.st_value);
2632
2633 *symp = sym;
2634 if (idp != NULL)
2635 *idp = i;
2636 return (symp);
2637 }
2638
2639 /*
2640 * Use a linear search to do the work of sym_by_addr().
2641 */
2642 static GElf_Sym *
sym_by_addr_linear(sym_tbl_t * symtab,GElf_Addr addr,GElf_Sym * symbolp,uint_t * idp)2643 sym_by_addr_linear(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symbolp,
2644 uint_t *idp)
2645 {
2646 size_t symn = symtab->sym_symn;
2647 char *strs = symtab->sym_strs;
2648 GElf_Sym sym, *symp = NULL;
2649 GElf_Sym osym, *osymp = NULL;
2650 int i, id;
2651
2652 if (symtab->sym_data_pri == NULL || symn == 0 || strs == NULL)
2653 return (NULL);
2654
2655 for (i = 0; i < symn; i++) {
2656 if ((symp = symtab_getsym(symtab, i, &sym)) != NULL) {
2657 if (addr >= sym.st_value &&
2658 addr < sym.st_value + sym.st_size) {
2659 if (osymp)
2660 symp = sym_prefer(
2661 symp, strs + symp->st_name,
2662 osymp, strs + osymp->st_name);
2663 if (symp != osymp) {
2664 osym = sym;
2665 osymp = &osym;
2666 id = i;
2667 }
2668 }
2669 }
2670 }
2671 if (osymp) {
2672 *symbolp = osym;
2673 if (idp)
2674 *idp = id;
2675 return (symbolp);
2676 }
2677 return (NULL);
2678 }
2679
2680 /*
2681 * Look up a symbol by address in the specified symbol table.
2682 * Adjustment to 'addr' must already have been made for the
2683 * offset of the symbol if this is a dynamic library symbol table.
2684 *
2685 * Use a linear or a binary search depending on whether or not we
2686 * chose to sort the table in optimize_symtab().
2687 */
2688 static GElf_Sym *
sym_by_addr(sym_tbl_t * symtab,GElf_Addr addr,GElf_Sym * symp,uint_t * idp)2689 sym_by_addr(sym_tbl_t *symtab, GElf_Addr addr, GElf_Sym *symp, uint_t *idp)
2690 {
2691 if (_libproc_no_qsort) {
2692 return (sym_by_addr_linear(symtab, addr, symp, idp));
2693 } else {
2694 return (sym_by_addr_binary(symtab, addr, symp, idp));
2695 }
2696 }
2697
2698 /*
2699 * Use a binary search to do the work of sym_by_name().
2700 */
2701 static GElf_Sym *
sym_by_name_binary(sym_tbl_t * symtab,const char * name,GElf_Sym * symp,uint_t * idp)2702 sym_by_name_binary(sym_tbl_t *symtab, const char *name, GElf_Sym *symp,
2703 uint_t *idp)
2704 {
2705 char *strs = symtab->sym_strs;
2706 uint_t i, *byname = symtab->sym_byname;
2707 int min, mid, max, cmp;
2708
2709 if (symtab->sym_data_pri == NULL || strs == NULL ||
2710 symtab->sym_count == 0)
2711 return (NULL);
2712
2713 min = 0;
2714 max = symtab->sym_count - 1;
2715
2716 while (min <= max) {
2717 mid = (max + min) / 2;
2718
2719 i = byname[mid];
2720 (void) symtab_getsym(symtab, i, symp);
2721
2722 if ((cmp = strcmp(name, strs + symp->st_name)) == 0) {
2723 if (idp != NULL)
2724 *idp = i;
2725 return (symp);
2726 }
2727
2728 if (cmp < 0)
2729 max = mid - 1;
2730 else
2731 min = mid + 1;
2732 }
2733
2734 return (NULL);
2735 }
2736
2737 /*
2738 * Use a linear search to do the work of sym_by_name().
2739 */
2740 static GElf_Sym *
sym_by_name_linear(sym_tbl_t * symtab,const char * name,GElf_Sym * symp,uint_t * idp)2741 sym_by_name_linear(sym_tbl_t *symtab, const char *name, GElf_Sym *symp,
2742 uint_t *idp)
2743 {
2744 size_t symn = symtab->sym_symn;
2745 char *strs = symtab->sym_strs;
2746 int i;
2747
2748 if (symtab->sym_data_pri == NULL || symn == 0 || strs == NULL)
2749 return (NULL);
2750
2751 for (i = 0; i < symn; i++) {
2752 if (symtab_getsym(symtab, i, symp) &&
2753 strcmp(name, strs + symp->st_name) == 0) {
2754 if (idp)
2755 *idp = i;
2756 return (symp);
2757 }
2758 }
2759
2760 return (NULL);
2761 }
2762
2763 /*
2764 * Look up a symbol by name in the specified symbol table.
2765 *
2766 * Use a linear or a binary search depending on whether or not we
2767 * chose to sort the table in optimize_symtab().
2768 */
2769 static GElf_Sym *
sym_by_name(sym_tbl_t * symtab,const char * name,GElf_Sym * symp,uint_t * idp)2770 sym_by_name(sym_tbl_t *symtab, const char *name, GElf_Sym *symp, uint_t *idp)
2771 {
2772 if (_libproc_no_qsort) {
2773 return (sym_by_name_linear(symtab, name, symp, idp));
2774 } else {
2775 return (sym_by_name_binary(symtab, name, symp, idp));
2776 }
2777 }
2778
2779 /*
2780 * Search the process symbol tables looking for a symbol whose
2781 * value to value+size contain the address specified by addr.
2782 * Return values are:
2783 * sym_name_buffer containing the symbol name
2784 * GElf_Sym symbol table entry
2785 * prsyminfo_t ancillary symbol information
2786 * Returns 0 on success, -1 on failure.
2787 */
2788 static int
i_Pxlookup_by_addr(struct ps_prochandle * P,int lmresolve,uintptr_t addr,char * sym_name_buffer,size_t bufsize,GElf_Sym * symbolp,prsyminfo_t * sip)2789 i_Pxlookup_by_addr(
2790 struct ps_prochandle *P,
2791 int lmresolve, /* use resolve linker object names */
2792 uintptr_t addr, /* process address being sought */
2793 char *sym_name_buffer, /* buffer for the symbol name */
2794 size_t bufsize, /* size of sym_name_buffer */
2795 GElf_Sym *symbolp, /* returned symbol table entry */
2796 prsyminfo_t *sip) /* returned symbol info */
2797 {
2798 GElf_Sym *symp;
2799 char *name;
2800 GElf_Sym sym1, *sym1p = NULL;
2801 GElf_Sym sym2, *sym2p = NULL;
2802 char *name1 = NULL;
2803 char *name2 = NULL;
2804 uint_t i1;
2805 uint_t i2;
2806 map_info_t *mptr;
2807 file_info_t *fptr;
2808
2809 (void) Prd_agent(P);
2810
2811 if ((mptr = Paddr2mptr(P, addr)) == NULL || /* no such address */
2812 (fptr = build_map_symtab(P, mptr)) == NULL || /* no mapped file */
2813 fptr->file_elf == NULL) /* not an ELF file */
2814 return (-1);
2815
2816 /*
2817 * Adjust the address by the load object base address in
2818 * case the address turns out to be in a shared library.
2819 */
2820 addr -= fptr->file_dyn_base;
2821
2822 /*
2823 * Search both symbol tables, symtab first, then dynsym.
2824 */
2825 if ((sym1p = sym_by_addr(&fptr->file_symtab, addr, &sym1, &i1)) != NULL)
2826 name1 = fptr->file_symtab.sym_strs + sym1.st_name;
2827 if ((sym2p = sym_by_addr(&fptr->file_dynsym, addr, &sym2, &i2)) != NULL)
2828 name2 = fptr->file_dynsym.sym_strs + sym2.st_name;
2829
2830 if ((symp = sym_prefer(sym1p, name1, sym2p, name2)) == NULL)
2831 return (-1);
2832
2833 name = (symp == sym1p) ? name1 : name2;
2834 if (bufsize > 0) {
2835 (void) strncpy(sym_name_buffer, name, bufsize);
2836 sym_name_buffer[bufsize - 1] = '\0';
2837 }
2838
2839 *symbolp = *symp;
2840 if (sip != NULL) {
2841 sip->prs_name = bufsize == 0 ? NULL : sym_name_buffer;
2842 if (lmresolve && (fptr->file_rname != NULL))
2843 sip->prs_object = fptr->file_rbase;
2844 else
2845 sip->prs_object = fptr->file_lbase;
2846 sip->prs_id = (symp == sym1p) ? i1 : i2;
2847 sip->prs_table = (symp == sym1p) ? PR_SYMTAB : PR_DYNSYM;
2848 sip->prs_lmid = (fptr->file_lo == NULL) ? LM_ID_BASE :
2849 fptr->file_lo->rl_lmident;
2850 }
2851
2852 if (GELF_ST_TYPE(symbolp->st_info) != STT_TLS)
2853 symbolp->st_value += fptr->file_dyn_base;
2854
2855 return (0);
2856 }
2857
2858 int
Pxlookup_by_addr(struct ps_prochandle * P,uintptr_t addr,char * buf,size_t bufsize,GElf_Sym * symp,prsyminfo_t * sip)2859 Pxlookup_by_addr(struct ps_prochandle *P, uintptr_t addr, char *buf,
2860 size_t bufsize, GElf_Sym *symp, prsyminfo_t *sip)
2861 {
2862 return (i_Pxlookup_by_addr(P, B_FALSE, addr, buf, bufsize, symp, sip));
2863 }
2864
2865 int
Pxlookup_by_addr_resolved(struct ps_prochandle * P,uintptr_t addr,char * buf,size_t bufsize,GElf_Sym * symp,prsyminfo_t * sip)2866 Pxlookup_by_addr_resolved(struct ps_prochandle *P, uintptr_t addr, char *buf,
2867 size_t bufsize, GElf_Sym *symp, prsyminfo_t *sip)
2868 {
2869 return (i_Pxlookup_by_addr(P, B_TRUE, addr, buf, bufsize, symp, sip));
2870 }
2871
2872 int
Plookup_by_addr(struct ps_prochandle * P,uintptr_t addr,char * buf,size_t size,GElf_Sym * symp)2873 Plookup_by_addr(struct ps_prochandle *P, uintptr_t addr, char *buf,
2874 size_t size, GElf_Sym *symp)
2875 {
2876 return (i_Pxlookup_by_addr(P, B_FALSE, addr, buf, size, symp, NULL));
2877 }
2878
2879 /*
2880 * Search the process symbol tables looking for a symbol whose name matches the
2881 * specified name and whose object and link map optionally match the specified
2882 * parameters. On success, the function returns 0 and fills in the GElf_Sym
2883 * symbol table entry. On failure, -1 is returned.
2884 */
2885 int
Pxlookup_by_name(struct ps_prochandle * P,Lmid_t lmid,const char * oname,const char * sname,GElf_Sym * symp,prsyminfo_t * sip)2886 Pxlookup_by_name(
2887 struct ps_prochandle *P,
2888 Lmid_t lmid, /* link map to match, or -1 for any */
2889 const char *oname, /* load object name */
2890 const char *sname, /* symbol name */
2891 GElf_Sym *symp, /* returned symbol table entry */
2892 prsyminfo_t *sip) /* returned symbol info */
2893 {
2894 map_info_t *mptr;
2895 file_info_t *fptr;
2896 int cnt;
2897
2898 GElf_Sym sym;
2899 prsyminfo_t si;
2900 int rv = -1;
2901 uint_t id;
2902
2903 if (oname == PR_OBJ_EVERY) {
2904 /* create all the file_info_t's for all the mappings */
2905 (void) Prd_agent(P);
2906 cnt = P->num_files;
2907 fptr = list_head(&P->file_head);
2908 } else {
2909 cnt = 1;
2910 if ((mptr = object_name_to_map(P, lmid, oname)) == NULL ||
2911 (fptr = build_map_symtab(P, mptr)) == NULL)
2912 return (-1);
2913 }
2914
2915 /*
2916 * Iterate through the loaded object files and look for the symbol
2917 * name in the .symtab and .dynsym of each. If we encounter a match
2918 * with SHN_UNDEF, keep looking in hopes of finding a better match.
2919 * This means that a name such as "puts" will match the puts function
2920 * in libc instead of matching the puts PLT entry in the a.out file.
2921 */
2922 for (; cnt > 0; cnt--, fptr = list_next(&P->file_head, fptr)) {
2923 Pbuild_file_symtab(P, fptr);
2924
2925 if (fptr->file_elf == NULL)
2926 continue;
2927
2928 if (lmid != PR_LMID_EVERY && fptr->file_lo != NULL &&
2929 lmid != fptr->file_lo->rl_lmident)
2930 continue;
2931
2932 if (fptr->file_symtab.sym_data_pri != NULL &&
2933 sym_by_name(&fptr->file_symtab, sname, symp, &id)) {
2934 if (sip != NULL) {
2935 sip->prs_id = id;
2936 sip->prs_table = PR_SYMTAB;
2937 sip->prs_object = oname;
2938 sip->prs_name = sname;
2939 sip->prs_lmid = fptr->file_lo == NULL ?
2940 LM_ID_BASE : fptr->file_lo->rl_lmident;
2941 }
2942 } else if (fptr->file_dynsym.sym_data_pri != NULL &&
2943 sym_by_name(&fptr->file_dynsym, sname, symp, &id)) {
2944 if (sip != NULL) {
2945 sip->prs_id = id;
2946 sip->prs_table = PR_DYNSYM;
2947 sip->prs_object = oname;
2948 sip->prs_name = sname;
2949 sip->prs_lmid = fptr->file_lo == NULL ?
2950 LM_ID_BASE : fptr->file_lo->rl_lmident;
2951 }
2952 } else {
2953 continue;
2954 }
2955
2956 if (GELF_ST_TYPE(symp->st_info) != STT_TLS)
2957 symp->st_value += fptr->file_dyn_base;
2958
2959 if (symp->st_shndx != SHN_UNDEF)
2960 return (0);
2961
2962 if (rv != 0) {
2963 if (sip != NULL)
2964 si = *sip;
2965 sym = *symp;
2966 rv = 0;
2967 }
2968 }
2969
2970 if (rv == 0) {
2971 if (sip != NULL)
2972 *sip = si;
2973 *symp = sym;
2974 }
2975
2976 return (rv);
2977 }
2978
2979 /*
2980 * Search the process symbol tables looking for a symbol whose name matches the
2981 * specified name, but without any restriction on the link map id.
2982 */
2983 int
Plookup_by_name(struct ps_prochandle * P,const char * object,const char * symbol,GElf_Sym * symp)2984 Plookup_by_name(struct ps_prochandle *P, const char *object,
2985 const char *symbol, GElf_Sym *symp)
2986 {
2987 return (Pxlookup_by_name(P, PR_LMID_EVERY, object, symbol, symp, NULL));
2988 }
2989
2990 /*
2991 * Iterate over the process's address space mappings.
2992 */
2993 static int
i_Pmapping_iter(struct ps_prochandle * P,boolean_t lmresolve,proc_map_f * func,void * cd)2994 i_Pmapping_iter(struct ps_prochandle *P, boolean_t lmresolve,
2995 proc_map_f *func, void *cd)
2996 {
2997 map_info_t *mptr;
2998 file_info_t *fptr;
2999 char *object_name;
3000 int rc = 0;
3001 int i;
3002
3003 /* create all the file_info_t's for all the mappings */
3004 (void) Prd_agent(P);
3005
3006 for (i = 0, mptr = P->mappings; i < P->map_count; i++, mptr++) {
3007 if ((fptr = mptr->map_file) == NULL)
3008 object_name = NULL;
3009 else if (lmresolve && (fptr->file_rname != NULL))
3010 object_name = fptr->file_rname;
3011 else
3012 object_name = fptr->file_lname;
3013 if ((rc = func(cd, &mptr->map_pmap, object_name)) != 0)
3014 return (rc);
3015 }
3016 return (0);
3017 }
3018
3019 int
Pmapping_iter(struct ps_prochandle * P,proc_map_f * func,void * cd)3020 Pmapping_iter(struct ps_prochandle *P, proc_map_f *func, void *cd)
3021 {
3022 return (i_Pmapping_iter(P, B_FALSE, func, cd));
3023 }
3024
3025 int
Pmapping_iter_resolved(struct ps_prochandle * P,proc_map_f * func,void * cd)3026 Pmapping_iter_resolved(struct ps_prochandle *P, proc_map_f *func, void *cd)
3027 {
3028 return (i_Pmapping_iter(P, B_TRUE, func, cd));
3029 }
3030
3031 /*
3032 * Iterate over the process's mapped objects.
3033 */
3034 static int
i_Pobject_iter(struct ps_prochandle * P,boolean_t lmresolve,proc_map_f * func,void * cd)3035 i_Pobject_iter(struct ps_prochandle *P, boolean_t lmresolve,
3036 proc_map_f *func, void *cd)
3037 {
3038 map_info_t *mptr;
3039 file_info_t *fptr;
3040 int rc = 0;
3041
3042 (void) Prd_agent(P); /* create file_info_t's for all the mappings */
3043 Pupdate_maps(P);
3044
3045 for (fptr = list_head(&P->file_head); fptr != NULL;
3046 fptr = list_next(&P->file_head, fptr)) {
3047 const char *lname;
3048
3049 if (lmresolve && (fptr->file_rname != NULL))
3050 lname = fptr->file_rname;
3051 else if (fptr->file_lname != NULL)
3052 lname = fptr->file_lname;
3053 else
3054 lname = "";
3055
3056 if ((mptr = fptr->file_map) == NULL)
3057 continue;
3058
3059 if ((rc = func(cd, &mptr->map_pmap, lname)) != 0)
3060 return (rc);
3061
3062 if (!P->info_valid)
3063 Pupdate_maps(P);
3064 }
3065 return (0);
3066 }
3067
3068 int
Pobject_iter(struct ps_prochandle * P,proc_map_f * func,void * cd)3069 Pobject_iter(struct ps_prochandle *P, proc_map_f *func, void *cd)
3070 {
3071 return (i_Pobject_iter(P, B_FALSE, func, cd));
3072 }
3073
3074 int
Pobject_iter_resolved(struct ps_prochandle * P,proc_map_f * func,void * cd)3075 Pobject_iter_resolved(struct ps_prochandle *P, proc_map_f *func, void *cd)
3076 {
3077 return (i_Pobject_iter(P, B_TRUE, func, cd));
3078 }
3079
3080 static char *
i_Pobjname(struct ps_prochandle * P,boolean_t lmresolve,uintptr_t addr,char * buffer,size_t bufsize)3081 i_Pobjname(struct ps_prochandle *P, boolean_t lmresolve, uintptr_t addr,
3082 char *buffer, size_t bufsize)
3083 {
3084 map_info_t *mptr;
3085 file_info_t *fptr;
3086
3087 /* create all the file_info_t's for all the mappings */
3088 (void) Prd_agent(P);
3089
3090 if ((mptr = Paddr2mptr(P, addr)) == NULL)
3091 return (NULL);
3092
3093 if (!lmresolve) {
3094 if (((fptr = mptr->map_file) == NULL) ||
3095 (fptr->file_lname == NULL))
3096 return (NULL);
3097 (void) strlcpy(buffer, fptr->file_lname, bufsize);
3098 return (buffer);
3099 }
3100
3101 /* Check for a cached copy of the resolved path */
3102 if (Pfindmap(P, mptr, buffer, bufsize) != NULL)
3103 return (buffer);
3104
3105 return (NULL);
3106 }
3107
3108 /*
3109 * Given a virtual address, return the name of the underlying
3110 * mapped object (file) as provided by the dynamic linker.
3111 * Return NULL if we can't find any name information for the object.
3112 */
3113 char *
Pobjname(struct ps_prochandle * P,uintptr_t addr,char * buffer,size_t bufsize)3114 Pobjname(struct ps_prochandle *P, uintptr_t addr,
3115 char *buffer, size_t bufsize)
3116 {
3117 return (i_Pobjname(P, B_FALSE, addr, buffer, bufsize));
3118 }
3119
3120 /*
3121 * Given a virtual address, try to return a filesystem path to the
3122 * underlying mapped object (file). If we're in the global zone,
3123 * this path could resolve to an object in another zone. If we're
3124 * unable return a valid filesystem path, we'll fall back to providing
3125 * the mapped object (file) name provided by the dynamic linker in
3126 * the target process (ie, the object reported by Pobjname()).
3127 */
3128 char *
Pobjname_resolved(struct ps_prochandle * P,uintptr_t addr,char * buffer,size_t bufsize)3129 Pobjname_resolved(struct ps_prochandle *P, uintptr_t addr,
3130 char *buffer, size_t bufsize)
3131 {
3132 return (i_Pobjname(P, B_TRUE, addr, buffer, bufsize));
3133 }
3134
3135 /*
3136 * Given a virtual address, return the link map id of the underlying mapped
3137 * object (file), as provided by the dynamic linker. Return -1 on failure.
3138 */
3139 int
Plmid(struct ps_prochandle * P,uintptr_t addr,Lmid_t * lmidp)3140 Plmid(struct ps_prochandle *P, uintptr_t addr, Lmid_t *lmidp)
3141 {
3142 map_info_t *mptr;
3143 file_info_t *fptr;
3144
3145 /* create all the file_info_t's for all the mappings */
3146 (void) Prd_agent(P);
3147
3148 if ((mptr = Paddr2mptr(P, addr)) != NULL &&
3149 (fptr = mptr->map_file) != NULL && fptr->file_lo != NULL) {
3150 *lmidp = fptr->file_lo->rl_lmident;
3151 return (0);
3152 }
3153
3154 return (-1);
3155 }
3156
3157 /*
3158 * Given an object name and optional lmid, iterate over the object's symbols.
3159 * If which == PR_SYMTAB, search the normal symbol table.
3160 * If which == PR_DYNSYM, search the dynamic symbol table.
3161 */
3162 static int
Psymbol_iter_com(struct ps_prochandle * P,Lmid_t lmid,const char * object_name,int which,int mask,pr_order_t order,proc_xsym_f * func,void * cd)3163 Psymbol_iter_com(struct ps_prochandle *P, Lmid_t lmid, const char *object_name,
3164 int which, int mask, pr_order_t order, proc_xsym_f *func, void *cd)
3165 {
3166 #if STT_NUM != (STT_TLS + 1)
3167 #error "STT_NUM has grown. update Psymbol_iter_com()"
3168 #endif
3169
3170 GElf_Sym sym;
3171 GElf_Shdr shdr;
3172 map_info_t *mptr;
3173 file_info_t *fptr;
3174 sym_tbl_t *symtab;
3175 size_t symn;
3176 const char *strs;
3177 size_t strsz;
3178 prsyminfo_t si;
3179 int rv;
3180 uint_t *map, i, count, ndx;
3181
3182 if ((mptr = object_name_to_map(P, lmid, object_name)) == NULL)
3183 return (-1);
3184
3185 if ((fptr = build_map_symtab(P, mptr)) == NULL || /* no mapped file */
3186 fptr->file_elf == NULL) /* not an ELF file */
3187 return (-1);
3188
3189 /*
3190 * Search the specified symbol table.
3191 */
3192 switch (which) {
3193 case PR_SYMTAB:
3194 symtab = &fptr->file_symtab;
3195 si.prs_table = PR_SYMTAB;
3196 break;
3197 case PR_DYNSYM:
3198 symtab = &fptr->file_dynsym;
3199 si.prs_table = PR_DYNSYM;
3200 break;
3201 default:
3202 return (-1);
3203 }
3204
3205 si.prs_object = object_name;
3206 si.prs_lmid = fptr->file_lo == NULL ?
3207 LM_ID_BASE : fptr->file_lo->rl_lmident;
3208
3209 symn = symtab->sym_symn;
3210 strs = symtab->sym_strs;
3211 strsz = symtab->sym_strsz;
3212
3213 switch (order) {
3214 case PRO_NATURAL:
3215 map = NULL;
3216 count = symn;
3217 break;
3218 case PRO_BYNAME:
3219 map = symtab->sym_byname;
3220 count = symtab->sym_count;
3221 break;
3222 case PRO_BYADDR:
3223 map = symtab->sym_byaddr;
3224 count = symtab->sym_count;
3225 break;
3226 default:
3227 return (-1);
3228 }
3229
3230 if (symtab->sym_data_pri == NULL || strs == NULL || count == 0)
3231 return (-1);
3232
3233 rv = 0;
3234
3235 for (i = 0; i < count; i++) {
3236 ndx = map == NULL ? i : map[i];
3237 if (symtab_getsym(symtab, ndx, &sym) != NULL) {
3238 uint_t s_bind, s_type, type;
3239
3240 if (sym.st_name >= strsz) /* invalid st_name */
3241 continue;
3242
3243 s_bind = GELF_ST_BIND(sym.st_info);
3244 s_type = GELF_ST_TYPE(sym.st_info);
3245
3246 /*
3247 * In case you haven't already guessed, this relies on
3248 * the bitmask used in <libproc.h> for encoding symbol
3249 * type and binding matching the order of STB and STT
3250 * constants in <sys/elf.h>. Changes to ELF must
3251 * maintain binary compatibility, so I think this is
3252 * reasonably fair game.
3253 */
3254 if (s_bind < STB_NUM && s_type < STT_NUM) {
3255 type = (1 << (s_type + 8)) | (1 << s_bind);
3256 if ((type & ~mask) != 0)
3257 continue;
3258 } else
3259 continue; /* Invalid type or binding */
3260
3261 if (GELF_ST_TYPE(sym.st_info) != STT_TLS)
3262 sym.st_value += fptr->file_dyn_base;
3263
3264 si.prs_name = strs + sym.st_name;
3265
3266 /*
3267 * If symbol's type is STT_SECTION, then try to lookup
3268 * the name of the corresponding section.
3269 */
3270 if (GELF_ST_TYPE(sym.st_info) == STT_SECTION &&
3271 fptr->file_shstrs != NULL &&
3272 gelf_getshdr(elf_getscn(fptr->file_elf,
3273 sym.st_shndx), &shdr) != NULL &&
3274 shdr.sh_name != 0 &&
3275 shdr.sh_name < fptr->file_shstrsz)
3276 si.prs_name = fptr->file_shstrs + shdr.sh_name;
3277
3278 si.prs_id = ndx;
3279 if ((rv = func(cd, &sym, si.prs_name, &si)) != 0)
3280 break;
3281 }
3282 }
3283
3284 return (rv);
3285 }
3286
3287 int
Pxsymbol_iter(struct ps_prochandle * P,Lmid_t lmid,const char * object_name,int which,int mask,proc_xsym_f * func,void * cd)3288 Pxsymbol_iter(struct ps_prochandle *P, Lmid_t lmid, const char *object_name,
3289 int which, int mask, proc_xsym_f *func, void *cd)
3290 {
3291 return (Psymbol_iter_com(P, lmid, object_name, which, mask,
3292 PRO_NATURAL, func, cd));
3293 }
3294
3295 int
Psymbol_iter_by_lmid(struct ps_prochandle * P,Lmid_t lmid,const char * object_name,int which,int mask,proc_sym_f * func,void * cd)3296 Psymbol_iter_by_lmid(struct ps_prochandle *P, Lmid_t lmid,
3297 const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
3298 {
3299 return (Psymbol_iter_com(P, lmid, object_name, which, mask,
3300 PRO_NATURAL, (proc_xsym_f *)(uintptr_t)func, cd));
3301 }
3302
3303 int
Psymbol_iter(struct ps_prochandle * P,const char * object_name,int which,int mask,proc_sym_f * func,void * cd)3304 Psymbol_iter(struct ps_prochandle *P,
3305 const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
3306 {
3307 return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
3308 PRO_NATURAL, (proc_xsym_f *)(uintptr_t)func, cd));
3309 }
3310
3311 int
Psymbol_iter_by_addr(struct ps_prochandle * P,const char * object_name,int which,int mask,proc_sym_f * func,void * cd)3312 Psymbol_iter_by_addr(struct ps_prochandle *P,
3313 const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
3314 {
3315 return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
3316 PRO_BYADDR, (proc_xsym_f *)(uintptr_t)func, cd));
3317 }
3318
3319 int
Psymbol_iter_by_name(struct ps_prochandle * P,const char * object_name,int which,int mask,proc_sym_f * func,void * cd)3320 Psymbol_iter_by_name(struct ps_prochandle *P,
3321 const char *object_name, int which, int mask, proc_sym_f *func, void *cd)
3322 {
3323 return (Psymbol_iter_com(P, PR_LMID_EVERY, object_name, which, mask,
3324 PRO_BYNAME, (proc_xsym_f *)(uintptr_t)func, cd));
3325 }
3326
3327 /*
3328 * Get the platform string.
3329 */
3330 char *
Pplatform(struct ps_prochandle * P,char * s,size_t n)3331 Pplatform(struct ps_prochandle *P, char *s, size_t n)
3332 {
3333 return (P->ops.pop_platform(P, s, n, P->data));
3334 }
3335
3336 /*
3337 * Get the uname(2) information.
3338 */
3339 int
Puname(struct ps_prochandle * P,struct utsname * u)3340 Puname(struct ps_prochandle *P, struct utsname *u)
3341 {
3342 return (P->ops.pop_uname(P, u, P->data));
3343 }
3344
3345 /*
3346 * Called from Pcreate(), Pgrab(), and Pfgrab_core() to initialize
3347 * the symbol table heads in the new ps_prochandle.
3348 */
3349 void
Pinitsym(struct ps_prochandle * P)3350 Pinitsym(struct ps_prochandle *P)
3351 {
3352 P->num_files = 0;
3353 list_create(&P->file_head, sizeof (file_info_t),
3354 offsetof(file_info_t, file_list));
3355 }
3356
3357 /*
3358 * Called from Prelease() to destroy the symbol tables.
3359 * Must be called by the client after an exec() in the victim process.
3360 */
3361 void
Preset_maps(struct ps_prochandle * P)3362 Preset_maps(struct ps_prochandle *P)
3363 {
3364 int i;
3365
3366 if (P->rap != NULL) {
3367 rd_delete(P->rap);
3368 P->rap = NULL;
3369 }
3370
3371 if (P->execname != NULL) {
3372 free(P->execname);
3373 P->execname = NULL;
3374 }
3375
3376 if (P->auxv != NULL) {
3377 free(P->auxv);
3378 P->auxv = NULL;
3379 P->nauxv = 0;
3380 }
3381
3382 for (i = 0; i < P->map_count; i++)
3383 map_info_free(P, &P->mappings[i]);
3384
3385 if (P->mappings != NULL) {
3386 free(P->mappings);
3387 P->mappings = NULL;
3388 }
3389 P->map_count = P->map_alloc = 0;
3390
3391 P->info_valid = 0;
3392 }
3393
3394 typedef struct getenv_data {
3395 char *buf;
3396 size_t bufsize;
3397 const char *search;
3398 size_t searchlen;
3399 } getenv_data_t;
3400
3401 /*ARGSUSED*/
3402 static int
getenv_func(void * data,struct ps_prochandle * P,uintptr_t addr,const char * nameval)3403 getenv_func(void *data, struct ps_prochandle *P, uintptr_t addr,
3404 const char *nameval)
3405 {
3406 getenv_data_t *d = data;
3407 size_t len;
3408
3409 if (nameval == NULL)
3410 return (0);
3411
3412 if (d->searchlen < strlen(nameval) &&
3413 strncmp(nameval, d->search, d->searchlen) == 0 &&
3414 nameval[d->searchlen] == '=') {
3415 len = MIN(strlen(nameval), d->bufsize - 1);
3416 (void) strncpy(d->buf, nameval, len);
3417 d->buf[len] = '\0';
3418 return (1);
3419 }
3420
3421 return (0);
3422 }
3423
3424 char *
Pgetenv(struct ps_prochandle * P,const char * name,char * buf,size_t buflen)3425 Pgetenv(struct ps_prochandle *P, const char *name, char *buf, size_t buflen)
3426 {
3427 getenv_data_t d;
3428
3429 d.buf = buf;
3430 d.bufsize = buflen;
3431 d.search = name;
3432 d.searchlen = strlen(name);
3433
3434 if (Penv_iter(P, getenv_func, &d) == 1) {
3435 char *equals = strchr(d.buf, '=');
3436
3437 if (equals != NULL) {
3438 (void) memmove(d.buf, equals + 1,
3439 d.buf + buflen - equals - 1);
3440 d.buf[d.buf + buflen - equals] = '\0';
3441
3442 return (buf);
3443 }
3444 }
3445
3446 return (NULL);
3447 }
3448
3449 /* number of argument or environment pointers to read all at once */
3450 #define NARG 100
3451
3452 int
Penv_iter(struct ps_prochandle * P,proc_env_f * func,void * data)3453 Penv_iter(struct ps_prochandle *P, proc_env_f *func, void *data)
3454 {
3455 const psinfo_t *psp;
3456 uintptr_t envpoff;
3457 GElf_Sym sym;
3458 int ret;
3459 char *buf, *nameval;
3460 size_t buflen;
3461
3462 int nenv = NARG;
3463 long envp[NARG];
3464
3465 /*
3466 * Attempt to find the "_environ" variable in the process.
3467 * Failing that, use the original value provided by Ppsinfo().
3468 *
3469 * The "_environ" variable is initialized by the CRT. We use a rough
3470 * heuristic to try and figure out if we have started running before the
3471 * CRT has executed by checking if the _environ pointer points to NULL
3472 * or not. Once initialized, it will never point to NULL absent an
3473 * application manipulating it directly, libc does not do so, even if
3474 * one calls clearenv(). There is a rare chance that an application is
3475 * messing with the _environ pointer directly; however, in practice that
3476 * is much rarer than this case and if someone is, libc is unlikely to
3477 * have a good day.
3478 *
3479 * While it's tempting to look towards libc variables such as
3480 * initenv_done and related, we have to remember that we're here because
3481 * we haven't actually called libc_init() or even loaded it!
3482 */
3483 if ((psp = Ppsinfo(P)) == NULL)
3484 return (-1);
3485
3486 envpoff = psp->pr_envp; /* Default if no _environ found */
3487
3488 if (Plookup_by_name(P, PR_OBJ_EXEC, "_environ", &sym) == 0) {
3489 if (P->status.pr_dmodel == PR_MODEL_NATIVE) {
3490 if (Pread(P, &envpoff, sizeof (envpoff),
3491 sym.st_value) != sizeof (envpoff))
3492 envpoff = psp->pr_envp;
3493 } else if (P->status.pr_dmodel == PR_MODEL_ILP32) {
3494 uint32_t envpoff32;
3495
3496 if (Pread(P, &envpoff32, sizeof (envpoff32),
3497 sym.st_value) != sizeof (envpoff32))
3498 envpoff = psp->pr_envp;
3499 else
3500 envpoff = envpoff32;
3501 }
3502
3503 if (envpoff == 0) {
3504 envpoff = psp->pr_envp;
3505 }
3506 }
3507
3508 buflen = 128;
3509 buf = malloc(buflen);
3510
3511 ret = 0;
3512 for (;;) {
3513 uintptr_t envoff;
3514
3515 if (nenv == NARG) {
3516 (void) memset(envp, 0, sizeof (envp));
3517 if (P->status.pr_dmodel == PR_MODEL_NATIVE) {
3518 if (Pread(P, envp,
3519 sizeof (envp), envpoff) <= 0) {
3520 ret = -1;
3521 break;
3522 }
3523 } else if (P->status.pr_dmodel == PR_MODEL_ILP32) {
3524 uint32_t e32[NARG];
3525 int i;
3526
3527 (void) memset(e32, 0, sizeof (e32));
3528 if (Pread(P, e32, sizeof (e32), envpoff) <= 0) {
3529 ret = -1;
3530 break;
3531 }
3532 for (i = 0; i < NARG; i++)
3533 envp[i] = e32[i];
3534 }
3535 nenv = 0;
3536 }
3537
3538 if ((envoff = envp[nenv++]) == (uintptr_t)NULL)
3539 break;
3540
3541 /*
3542 * Attempt to read the string from the process.
3543 */
3544 again:
3545 ret = Pread_string(P, buf, buflen, envoff);
3546
3547 if (ret <= 0) {
3548 nameval = NULL;
3549 } else if (ret == buflen - 1) {
3550 free(buf);
3551 /*
3552 * Bail if we have a corrupted environment
3553 */
3554 if (buflen >= ARG_MAX)
3555 return (-1);
3556 buflen *= 2;
3557 buf = malloc(buflen);
3558 goto again;
3559 } else {
3560 nameval = buf;
3561 }
3562
3563 if ((ret = func(data, P, envoff, nameval)) != 0)
3564 break;
3565
3566 envpoff += (P->status.pr_dmodel == PR_MODEL_LP64)? 8 : 4;
3567 }
3568
3569 free(buf);
3570
3571 return (ret);
3572 }
3573