1 /*
2 * This file and its contents are supplied under the terms of the
3 * Common Development and Distribution License ("CDDL"), version 1.0.
4 * You may only use this file in accordance with the terms of version
5 * 1.0 of the CDDL.
6 *
7 * A full copy of the text of the CDDL should have accompanied this
8 * source. A copy of the CDDL is also available via the Internet at
9 * http://www.illumos.org/license/CDDL.
10 */
11
12 /*
13 * Copyright (c) 2019 Joyent, Inc.
14 */
15
16 /*
17 * To perform a merge of two CTF containers, we first diff the two containers
18 * types. For every type that's in the src container, but not in the dst
19 * container, we note it and add it to dst container. If there are any objects
20 * or functions associated with src, we go through and update the types that
21 * they refer to such that they all refer to types in the dst container.
22 *
23 * The bulk of the logic for the merge, after we've run the diff, occurs in
24 * ctf_merge_common().
25 *
26 * In terms of exported APIs, we don't really export a simple merge two
27 * containers, as the general way this is used, in something like ctfmerge(1),
28 * is to add all the containers and then let us figure out the best way to merge
29 * it.
30 */
31
32 #include <libctf_impl.h>
33 #include <sys/debug.h>
34 #include <sys/list.h>
35 #include <stddef.h>
36 #include <fcntl.h>
37 #include <sys/types.h>
38 #include <sys/stat.h>
39 #include <mergeq.h>
40 #include <errno.h>
41
42 typedef struct ctf_merge_tinfo {
43 uint16_t cmt_map; /* Map to the type in out */
44 boolean_t cmt_fixup;
45 boolean_t cmt_forward;
46 boolean_t cmt_missing;
47 } ctf_merge_tinfo_t;
48
49 /*
50 * State required for doing an individual merge of two containers.
51 */
52 typedef struct ctf_merge_types {
53 ctf_file_t *cm_out; /* Output CTF file */
54 ctf_file_t *cm_src; /* Input CTF file */
55 ctf_merge_tinfo_t *cm_tmap; /* Type state information */
56 boolean_t cm_dedup; /* Are we doing a dedup? */
57 boolean_t cm_unique; /* are we doing a uniquify? */
58 } ctf_merge_types_t;
59
60 typedef struct ctf_merge_objmap {
61 list_node_t cmo_node;
62 const char *cmo_name; /* Symbol name */
63 const char *cmo_file; /* Symbol file */
64 ulong_t cmo_idx; /* Symbol ID */
65 Elf64_Sym cmo_sym; /* Symbol Entry */
66 ctf_id_t cmo_tid; /* Type ID */
67 } ctf_merge_objmap_t;
68
69 typedef struct ctf_merge_funcmap {
70 list_node_t cmf_node;
71 const char *cmf_name; /* Symbol name */
72 const char *cmf_file; /* Symbol file */
73 ulong_t cmf_idx; /* Symbol ID */
74 Elf64_Sym cmf_sym; /* Symbol Entry */
75 ctf_id_t cmf_rtid; /* Type ID */
76 uint_t cmf_flags; /* ctf_funcinfo_t ctc_flags */
77 uint_t cmf_argc; /* Number of arguments */
78 ctf_id_t cmf_args[]; /* Types of arguments */
79 } ctf_merge_funcmap_t;
80
81 typedef struct ctf_merge_input {
82 list_node_t cmi_node;
83 ctf_file_t *cmi_input;
84 list_t cmi_omap;
85 list_t cmi_fmap;
86 boolean_t cmi_created;
87 } ctf_merge_input_t;
88
89 struct ctf_merge_handle {
90 list_t cmh_inputs; /* Input list */
91 uint_t cmh_ninputs; /* Number of inputs */
92 uint_t cmh_nthreads; /* Number of threads to use */
93 ctf_file_t *cmh_unique; /* ctf to uniquify against */
94 boolean_t cmh_msyms; /* Should we merge symbols/funcs? */
95 int cmh_ofd; /* FD for output file */
96 int cmh_flags; /* Flags that control merge behavior */
97 char *cmh_label; /* Optional label */
98 char *cmh_pname; /* Parent name */
99 };
100
101 typedef struct ctf_merge_symbol_arg {
102 list_t *cmsa_objmap;
103 list_t *cmsa_funcmap;
104 ctf_file_t *cmsa_out;
105 boolean_t cmsa_dedup;
106 } ctf_merge_symbol_arg_t;
107
108 static int ctf_merge_add_type(ctf_merge_types_t *, ctf_id_t);
109
110 static ctf_id_t
111 ctf_merge_gettype(ctf_merge_types_t *cmp, ctf_id_t id)
112 {
113 if (cmp->cm_dedup == B_FALSE) {
114 VERIFY(cmp->cm_tmap[id].cmt_map != 0);
115 return (cmp->cm_tmap[id].cmt_map);
116 }
117
118 while (cmp->cm_tmap[id].cmt_missing == B_FALSE) {
119 VERIFY(cmp->cm_tmap[id].cmt_map != 0);
120 id = cmp->cm_tmap[id].cmt_map;
121 }
122 VERIFY(cmp->cm_tmap[id].cmt_map != 0);
123 return (cmp->cm_tmap[id].cmt_map);
124 }
125
126 static void
127 ctf_merge_diffcb(ctf_file_t *ifp, ctf_id_t iid, boolean_t same, ctf_file_t *ofp,
128 ctf_id_t oid, void *arg)
129 {
130 ctf_merge_types_t *cmp = arg;
131 ctf_merge_tinfo_t *cmt = cmp->cm_tmap;
132
133 if (same == B_TRUE) {
134 if (ctf_type_kind(ifp, iid) == CTF_K_FORWARD &&
135 ctf_type_kind(ofp, oid) != CTF_K_FORWARD) {
136 VERIFY(cmt[oid].cmt_map == 0);
137
138 /*
139 * If we're uniquifying types, it's possible for the
140 * container that we're uniquifying against to have a
141 * forward which exists in the container being reduced.
142 * For example, genunix has the machcpu structure as a
143 * forward which is actually in unix and we uniquify
144 * unix against genunix. In such cases, we explicitly do
145 * not do any mapping of the forward information, lest
146 * we risk losing the real definition. Instead, mark
147 * that it's missing.
148 */
149 if (cmp->cm_unique == B_TRUE) {
150 cmt[oid].cmt_missing = B_TRUE;
151 return;
152 }
153
154 cmt[oid].cmt_map = iid;
155 cmt[oid].cmt_forward = B_TRUE;
156 ctf_dprintf("merge diff forward mapped %d->%d\n", oid,
157 iid);
158 return;
159 }
160
161 /*
162 * We could have multiple things that a given type ends up
163 * matching in the world of forwards and pointers to forwards.
164 * For now just take the first one...
165 */
166 if (cmt[oid].cmt_map != 0)
167 return;
168 cmt[oid].cmt_map = iid;
169 ctf_dprintf("merge diff mapped %d->%d\n", oid, iid);
170 } else if (ifp == cmp->cm_src) {
171 VERIFY(cmt[iid].cmt_map == 0);
172 cmt[iid].cmt_missing = B_TRUE;
173 ctf_dprintf("merge diff said %d is missing\n", iid);
174 }
175 }
176
177 static int
178 ctf_merge_add_number(ctf_merge_types_t *cmp, ctf_id_t id)
179 {
180 int ret, flags;
181 const ctf_type_t *tp;
182 const char *name;
183 ctf_encoding_t en;
184
185 if (ctf_type_encoding(cmp->cm_src, id, &en) != 0)
186 return (CTF_ERR);
187
188 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
189 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
190 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
191 flags = CTF_ADD_ROOT;
192 else
193 flags = CTF_ADD_NONROOT;
194
195 ret = ctf_add_encoded(cmp->cm_out, flags, name, &en,
196 ctf_type_kind(cmp->cm_src, id));
197
198 if (ret == CTF_ERR)
199 return (ret);
200
201 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
202 cmp->cm_tmap[id].cmt_map = ret;
203 return (0);
204 }
205
206 static int
207 ctf_merge_add_array(ctf_merge_types_t *cmp, ctf_id_t id)
208 {
209 int ret, flags;
210 const ctf_type_t *tp;
211 ctf_arinfo_t ar;
212
213 if (ctf_array_info(cmp->cm_src, id, &ar) == CTF_ERR)
214 return (CTF_ERR);
215
216 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
217 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
218 flags = CTF_ADD_ROOT;
219 else
220 flags = CTF_ADD_NONROOT;
221
222 if (cmp->cm_tmap[ar.ctr_contents].cmt_map == 0) {
223 ret = ctf_merge_add_type(cmp, ar.ctr_contents);
224 if (ret != 0)
225 return (ret);
226 ASSERT(cmp->cm_tmap[ar.ctr_contents].cmt_map != 0);
227 }
228 ar.ctr_contents = ctf_merge_gettype(cmp, ar.ctr_contents);
229
230 if (cmp->cm_tmap[ar.ctr_index].cmt_map == 0) {
231 ret = ctf_merge_add_type(cmp, ar.ctr_index);
232 if (ret != 0)
233 return (ret);
234 ASSERT(cmp->cm_tmap[ar.ctr_index].cmt_map != 0);
235 }
236 ar.ctr_index = ctf_merge_gettype(cmp, ar.ctr_index);
237
238 ret = ctf_add_array(cmp->cm_out, flags, &ar);
239 if (ret == CTF_ERR)
240 return (ret);
241
242 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
243 cmp->cm_tmap[id].cmt_map = ret;
244
245 return (0);
246 }
247
248 static int
249 ctf_merge_add_reftype(ctf_merge_types_t *cmp, ctf_id_t id)
250 {
251 int ret, flags;
252 const ctf_type_t *tp;
253 ctf_id_t reftype;
254 const char *name;
255
256 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
257 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
258 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
259 flags = CTF_ADD_ROOT;
260 else
261 flags = CTF_ADD_NONROOT;
262
263 reftype = ctf_type_reference(cmp->cm_src, id);
264 if (reftype == CTF_ERR)
265 return (ctf_set_errno(cmp->cm_out, ctf_errno(cmp->cm_src)));
266
267 if (cmp->cm_tmap[reftype].cmt_map == 0) {
268 ret = ctf_merge_add_type(cmp, reftype);
269 if (ret != 0)
270 return (ret);
271 ASSERT(cmp->cm_tmap[reftype].cmt_map != 0);
272 }
273 reftype = ctf_merge_gettype(cmp, reftype);
274
275 ret = ctf_add_reftype(cmp->cm_out, flags, name, reftype,
276 ctf_type_kind(cmp->cm_src, id));
277 if (ret == CTF_ERR)
278 return (ret);
279
280 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
281 cmp->cm_tmap[id].cmt_map = ret;
282 return (0);
283 }
284
285 static int
286 ctf_merge_add_typedef(ctf_merge_types_t *cmp, ctf_id_t id)
287 {
288 int ret, flags;
289 const ctf_type_t *tp;
290 const char *name;
291 ctf_id_t reftype;
292
293 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
294 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
295 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
296 flags = CTF_ADD_ROOT;
297 else
298 flags = CTF_ADD_NONROOT;
299
300 reftype = ctf_type_reference(cmp->cm_src, id);
301 if (reftype == CTF_ERR)
302 return (ctf_set_errno(cmp->cm_out, ctf_errno(cmp->cm_src)));
303
304 if (cmp->cm_tmap[reftype].cmt_map == 0) {
305 ret = ctf_merge_add_type(cmp, reftype);
306 if (ret != 0)
307 return (ret);
308 ASSERT(cmp->cm_tmap[reftype].cmt_map != 0);
309 }
310 reftype = ctf_merge_gettype(cmp, reftype);
311
312 ret = ctf_add_typedef(cmp->cm_out, flags, name, reftype);
313 if (ret == CTF_ERR)
314 return (ret);
315
316 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
317 cmp->cm_tmap[id].cmt_map = ret;
318 return (0);
319 }
320
321 typedef struct ctf_merge_enum {
322 ctf_file_t *cme_fp;
323 ctf_id_t cme_id;
324 } ctf_merge_enum_t;
325
326 static int
327 ctf_merge_add_enumerator(const char *name, int value, void *arg)
328 {
329 ctf_merge_enum_t *cmep = arg;
330
331 return (ctf_add_enumerator(cmep->cme_fp, cmep->cme_id, name, value) ==
332 CTF_ERR);
333 }
334
335 static int
336 ctf_merge_add_enum(ctf_merge_types_t *cmp, ctf_id_t id)
337 {
338 int flags;
339 const ctf_type_t *tp;
340 const char *name;
341 ctf_id_t enumid;
342 ctf_merge_enum_t cme;
343
344 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
345 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
346 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
347 flags = CTF_ADD_ROOT;
348 else
349 flags = CTF_ADD_NONROOT;
350
351 enumid = ctf_add_enum(cmp->cm_out, flags, name);
352 if (enumid == CTF_ERR)
353 return (enumid);
354
355 cme.cme_fp = cmp->cm_out;
356 cme.cme_id = enumid;
357 if (ctf_enum_iter(cmp->cm_src, id, ctf_merge_add_enumerator,
358 &cme) != 0)
359 return (CTF_ERR);
360
361 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
362 cmp->cm_tmap[id].cmt_map = enumid;
363 return (0);
364 }
365
366 static int
367 ctf_merge_add_func(ctf_merge_types_t *cmp, ctf_id_t id)
368 {
369 int ret, flags, i;
370 const ctf_type_t *tp;
371 ctf_funcinfo_t ctc;
372 ctf_id_t *argv;
373
374 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
375 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
376 flags = CTF_ADD_ROOT;
377 else
378 flags = CTF_ADD_NONROOT;
379
380 if (ctf_func_info_by_id(cmp->cm_src, id, &ctc) == CTF_ERR)
381 return (ctf_set_errno(cmp->cm_out, ctf_errno(cmp->cm_src)));
382
383 argv = ctf_alloc(sizeof (ctf_id_t) * ctc.ctc_argc);
384 if (argv == NULL)
385 return (ctf_set_errno(cmp->cm_out, ENOMEM));
386 if (ctf_func_args_by_id(cmp->cm_src, id, ctc.ctc_argc, argv) ==
387 CTF_ERR) {
388 ctf_free(argv, sizeof (ctf_id_t) * ctc.ctc_argc);
389 return (ctf_set_errno(cmp->cm_out, ctf_errno(cmp->cm_src)));
390 }
391
392 if (cmp->cm_tmap[ctc.ctc_return].cmt_map == 0) {
393 ret = ctf_merge_add_type(cmp, ctc.ctc_return);
394 if (ret != 0)
395 return (ret);
396 ASSERT(cmp->cm_tmap[ctc.ctc_return].cmt_map != 0);
397 }
398 ctc.ctc_return = ctf_merge_gettype(cmp, ctc.ctc_return);
399
400 for (i = 0; i < ctc.ctc_argc; i++) {
401 if (cmp->cm_tmap[argv[i]].cmt_map == 0) {
402 ret = ctf_merge_add_type(cmp, argv[i]);
403 if (ret != 0)
404 return (ret);
405 ASSERT(cmp->cm_tmap[argv[i]].cmt_map != 0);
406 }
407 argv[i] = ctf_merge_gettype(cmp, argv[i]);
408 }
409
410 ret = ctf_add_funcptr(cmp->cm_out, flags, &ctc, argv);
411 ctf_free(argv, sizeof (ctf_id_t) * ctc.ctc_argc);
412 if (ret == CTF_ERR)
413 return (ret);
414
415 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
416 cmp->cm_tmap[id].cmt_map = ret;
417 return (0);
418 }
419
420 static int
421 ctf_merge_add_forward(ctf_merge_types_t *cmp, ctf_id_t id)
422 {
423 int ret, flags;
424 const ctf_type_t *tp;
425 const char *name;
426
427 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
428 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
429 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
430 flags = CTF_ADD_ROOT;
431 else
432 flags = CTF_ADD_NONROOT;
433
434 /*
435 * ctf_add_forward tries to check to see if a given forward already
436 * exists in one of its hash tables. If we're here then we know that we
437 * have a forward in a container that isn't present in another.
438 * Therefore, we choose a token hash table to satisfy the API choice
439 * here.
440 */
441 ret = ctf_add_forward(cmp->cm_out, flags, name, CTF_K_STRUCT);
442 if (ret == CTF_ERR)
443 return (CTF_ERR);
444
445 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
446 cmp->cm_tmap[id].cmt_map = ret;
447 return (0);
448 }
449
450 typedef struct ctf_merge_su {
451 ctf_merge_types_t *cms_cm;
452 ctf_id_t cms_id;
453 } ctf_merge_su_t;
454
455 static int
456 ctf_merge_add_member(const char *name, ctf_id_t type, ulong_t offset, void *arg)
457 {
458 ctf_merge_su_t *cms = arg;
459
460 VERIFY(cms->cms_cm->cm_tmap[type].cmt_map != 0);
461 type = cms->cms_cm->cm_tmap[type].cmt_map;
462
463 ctf_dprintf("Trying to add member %s to %d\n", name, cms->cms_id);
464 return (ctf_add_member(cms->cms_cm->cm_out, cms->cms_id, name,
465 type, offset) == CTF_ERR);
466 }
467
468 /*
469 * During the first pass, we always add the generic structure and union but none
470 * of its members as they might not all have been mapped yet. Instead we just
471 * mark all structures and unions as needing to be fixed up.
472 */
473 static int
474 ctf_merge_add_sou(ctf_merge_types_t *cmp, ctf_id_t id, boolean_t forward)
475 {
476 int flags, kind;
477 const ctf_type_t *tp;
478 const char *name;
479 ctf_id_t suid;
480
481 tp = LCTF_INDEX_TO_TYPEPTR(cmp->cm_src, id);
482 name = ctf_strraw(cmp->cm_src, tp->ctt_name);
483 if (CTF_INFO_ISROOT(tp->ctt_info) != 0)
484 flags = CTF_ADD_ROOT;
485 else
486 flags = CTF_ADD_NONROOT;
487 kind = ctf_type_kind(cmp->cm_src, id);
488
489 if (kind == CTF_K_STRUCT)
490 suid = ctf_add_struct(cmp->cm_out, flags, name);
491 else
492 suid = ctf_add_union(cmp->cm_out, flags, name);
493
494 if (suid == CTF_ERR)
495 return (suid);
496
497 /*
498 * If this is a forward reference then its mapping should already
499 * exist.
500 */
501 if (forward == B_FALSE) {
502 VERIFY(cmp->cm_tmap[id].cmt_map == 0);
503 cmp->cm_tmap[id].cmt_map = suid;
504 ctf_dprintf("added sou \"%s\" as (%d) %d->%d\n", name, kind, id,
505 suid);
506 } else {
507 VERIFY(cmp->cm_tmap[id].cmt_map == suid);
508 }
509 cmp->cm_tmap[id].cmt_fixup = B_TRUE;
510
511 return (0);
512 }
513
514 static int
515 ctf_merge_add_type(ctf_merge_types_t *cmp, ctf_id_t id)
516 {
517 int kind, ret;
518
519 /*
520 * We may end up evaluating a type more than once as we may deal with it
521 * as we recursively evaluate some kind of reference and then we may see
522 * it normally.
523 */
524 if (cmp->cm_tmap[id].cmt_map != 0)
525 return (0);
526
527 kind = ctf_type_kind(cmp->cm_src, id);
528 switch (kind) {
529 case CTF_K_INTEGER:
530 case CTF_K_FLOAT:
531 ret = ctf_merge_add_number(cmp, id);
532 break;
533 case CTF_K_ARRAY:
534 ret = ctf_merge_add_array(cmp, id);
535 break;
536 case CTF_K_POINTER:
537 case CTF_K_VOLATILE:
538 case CTF_K_CONST:
539 case CTF_K_RESTRICT:
540 ret = ctf_merge_add_reftype(cmp, id);
541 break;
542 case CTF_K_TYPEDEF:
543 ret = ctf_merge_add_typedef(cmp, id);
544 break;
545 case CTF_K_ENUM:
546 ret = ctf_merge_add_enum(cmp, id);
547 break;
548 case CTF_K_FUNCTION:
549 ret = ctf_merge_add_func(cmp, id);
550 break;
551 case CTF_K_FORWARD:
552 ret = ctf_merge_add_forward(cmp, id);
553 break;
554 case CTF_K_STRUCT:
555 case CTF_K_UNION:
556 ret = ctf_merge_add_sou(cmp, id, B_FALSE);
557 break;
558 case CTF_K_UNKNOWN:
559 /*
560 * We don't add unknown types, and we later assert that nothing
561 * should reference them.
562 */
563 return (0);
564 default:
565 abort();
566 }
567
568 return (ret);
569 }
570
571 static int
572 ctf_merge_fixup_sou(ctf_merge_types_t *cmp, ctf_id_t id)
573 {
574 ctf_dtdef_t *dtd;
575 ctf_merge_su_t cms;
576 ctf_id_t mapid;
577 ssize_t size;
578
579 mapid = cmp->cm_tmap[id].cmt_map;
580 VERIFY(mapid != 0);
581 dtd = ctf_dtd_lookup(cmp->cm_out, mapid);
582 VERIFY(dtd != NULL);
583
584 ctf_dprintf("Trying to fix up sou %d\n", id);
585 cms.cms_cm = cmp;
586 cms.cms_id = mapid;
587 if (ctf_member_iter(cmp->cm_src, id, ctf_merge_add_member, &cms) != 0)
588 return (CTF_ERR);
589
590 if ((size = ctf_type_size(cmp->cm_src, id)) == CTF_ERR)
591 return (CTF_ERR);
592 if (ctf_set_size(cmp->cm_out, mapid, size) == CTF_ERR)
593 return (CTF_ERR);
594
595 return (0);
596 }
597
598 static int
599 ctf_merge_fixup_type(ctf_merge_types_t *cmp, ctf_id_t id)
600 {
601 int kind, ret;
602
603 kind = ctf_type_kind(cmp->cm_src, id);
604 switch (kind) {
605 case CTF_K_STRUCT:
606 case CTF_K_UNION:
607 ret = ctf_merge_fixup_sou(cmp, id);
608 break;
609 default:
610 VERIFY(0);
611 ret = CTF_ERR;
612 }
613
614 return (ret);
615 }
616
617 /*
618 * Now that we've successfully merged everything, we're going to clean
619 * up the merge type table. Traditionally if we had just two different
620 * files that we were working between, the types would be fully
621 * resolved. However, because we were comparing with ourself every step
622 * of the way and not our reduced self, we need to go through and update
623 * every mapped entry to what it now points to in the deduped file.
624 */
625 static void
626 ctf_merge_fixup_dedup_map(ctf_merge_types_t *cmp)
627 {
628 int i;
629
630 for (i = 1; i < cmp->cm_src->ctf_typemax + 1; i++) {
631 ctf_id_t tid;
632
633 /*
634 * Missing types always have their id updated to exactly what it
635 * should be.
636 */
637 if (cmp->cm_tmap[i].cmt_missing == B_TRUE) {
638 VERIFY(cmp->cm_tmap[i].cmt_map != 0);
639 continue;
640 }
641
642 tid = i;
643 while (cmp->cm_tmap[tid].cmt_missing == B_FALSE) {
644 VERIFY(cmp->cm_tmap[tid].cmt_map != 0);
645 tid = cmp->cm_tmap[tid].cmt_map;
646 }
647 VERIFY(cmp->cm_tmap[tid].cmt_map != 0);
648 cmp->cm_tmap[i].cmt_map = cmp->cm_tmap[tid].cmt_map;
649 }
650 }
651
652
653 /*
654 * We're going to do three passes over the containers.
655 *
656 * Pass 1 checks for forward references in the output container that we know
657 * exist in the source container.
658 *
659 * Pass 2 adds all the missing types from the source container. As part of this
660 * we may be adding a type as a forward reference that doesn't exist yet.
661 * Any types that we encounter in this form, we need to add to a third pass.
662 *
663 * Pass 3 is the fixup pass. Here we go through and find all the types that were
664 * missing in the first.
665 *
666 * Importantly, we *must* call ctf_update between the second and third pass,
667 * otherwise several of the libctf functions will not properly find the data in
668 * the container. If we're doing a dedup we also fix up the type mapping.
669 */
670 static int
671 ctf_merge_common(ctf_merge_types_t *cmp)
672 {
673 int ret, i;
674
675 ctf_phase_dump(cmp->cm_src, "merge-common-src", NULL);
676 ctf_phase_dump(cmp->cm_out, "merge-common-dest", NULL);
677
678 /* Pass 1 */
679 for (i = 1; i <= cmp->cm_src->ctf_typemax; i++) {
680 if (cmp->cm_tmap[i].cmt_forward == B_TRUE) {
681 ret = ctf_merge_add_sou(cmp, i, B_TRUE);
682 if (ret != 0) {
683 return (ret);
684 }
685 }
686 }
687
688 /* Pass 2 */
689 for (i = 1; i <= cmp->cm_src->ctf_typemax; i++) {
690 if (cmp->cm_tmap[i].cmt_missing == B_TRUE) {
691 ret = ctf_merge_add_type(cmp, i);
692 if (ret != 0) {
693 ctf_dprintf("Failed to merge type %d\n", i);
694 return (ret);
695 }
696 }
697 }
698
699 ret = ctf_update(cmp->cm_out);
700 if (ret != 0)
701 return (ret);
702
703 if (cmp->cm_dedup == B_TRUE) {
704 ctf_merge_fixup_dedup_map(cmp);
705 }
706
707 ctf_dprintf("Beginning merge pass 3\n");
708 /* Pass 3 */
709 for (i = 1; i <= cmp->cm_src->ctf_typemax; i++) {
710 if (cmp->cm_tmap[i].cmt_fixup == B_TRUE) {
711 ret = ctf_merge_fixup_type(cmp, i);
712 if (ret != 0)
713 return (ret);
714 }
715 }
716
717 if (cmp->cm_dedup == B_TRUE) {
718 ctf_merge_fixup_dedup_map(cmp);
719 }
720
721 return (0);
722 }
723
724 /*
725 * Uniquification is slightly different from a stock merge. For starters, we
726 * don't need to replace any forward references in the output. In this case
727 * though, the types that already exist are in a parent container to the empty
728 * output container.
729 */
730 static int
731 ctf_merge_uniquify_types(ctf_merge_types_t *cmp)
732 {
733 int i, ret;
734
735 for (i = 1; i <= cmp->cm_src->ctf_typemax; i++) {
736 if (cmp->cm_tmap[i].cmt_missing == B_FALSE)
737 continue;
738 ret = ctf_merge_add_type(cmp, i);
739 if (ret != 0)
740 return (ret);
741 }
742
743 ret = ctf_update(cmp->cm_out);
744 if (ret != 0)
745 return (ret);
746
747 for (i = 1; i <= cmp->cm_src->ctf_typemax; i++) {
748 if (cmp->cm_tmap[i].cmt_fixup == B_FALSE)
749 continue;
750 ret = ctf_merge_fixup_type(cmp, i);
751 if (ret != 0)
752 return (ret);
753 }
754
755 return (0);
756 }
757
758 static int
759 ctf_merge_types_init(ctf_merge_types_t *cmp)
760 {
761 cmp->cm_tmap = ctf_alloc(sizeof (ctf_merge_tinfo_t) *
762 (cmp->cm_src->ctf_typemax + 1));
763 if (cmp->cm_tmap == NULL)
764 return (ctf_set_errno(cmp->cm_out, ENOMEM));
765 bzero(cmp->cm_tmap, sizeof (ctf_merge_tinfo_t) *
766 (cmp->cm_src->ctf_typemax + 1));
767 return (0);
768 }
769
770 static void
771 ctf_merge_types_fini(ctf_merge_types_t *cmp)
772 {
773 ctf_free(cmp->cm_tmap, sizeof (ctf_merge_tinfo_t) *
774 (cmp->cm_src->ctf_typemax + 1));
775 }
776
777 /*
778 * After performing a pass, we need to go through the object and function type
779 * maps and potentially fix them up based on the new maps that we have.
780 */
781 static void
782 ctf_merge_fixup_symmaps(ctf_merge_types_t *cmp, ctf_merge_input_t *cmi)
783 {
784 ctf_merge_objmap_t *cmo;
785 ctf_merge_funcmap_t *cmf;
786
787 for (cmo = list_head(&cmi->cmi_omap); cmo != NULL;
788 cmo = list_next(&cmi->cmi_omap, cmo)) {
789 VERIFY3S(cmo->cmo_tid, !=, 0);
790 VERIFY(cmp->cm_tmap[cmo->cmo_tid].cmt_map != 0);
791 cmo->cmo_tid = cmp->cm_tmap[cmo->cmo_tid].cmt_map;
792 }
793
794 for (cmf = list_head(&cmi->cmi_fmap); cmf != NULL;
795 cmf = list_next(&cmi->cmi_fmap, cmf)) {
796 int i;
797
798 VERIFY(cmp->cm_tmap[cmf->cmf_rtid].cmt_map != 0);
799 cmf->cmf_rtid = cmp->cm_tmap[cmf->cmf_rtid].cmt_map;
800 for (i = 0; i < cmf->cmf_argc; i++) {
801 VERIFY(cmp->cm_tmap[cmf->cmf_args[i]].cmt_map != 0);
802 cmf->cmf_args[i] =
803 cmp->cm_tmap[cmf->cmf_args[i]].cmt_map;
804 }
805 }
806 }
807
808 /*
809 * Merge the types contained inside of two input files. The second input file is
810 * always going to be the destination. We're guaranteed that it's always
811 * writeable.
812 */
813 static int
814 ctf_merge_types(void *arg, void *arg2, void **outp, void *unsued)
815 {
816 int ret;
817 ctf_merge_types_t cm;
818 ctf_diff_t *cdp;
819 ctf_merge_input_t *scmi = arg;
820 ctf_merge_input_t *dcmi = arg2;
821 ctf_file_t *out = dcmi->cmi_input;
822 ctf_file_t *source = scmi->cmi_input;
823
824 ctf_dprintf("merging %p->%p\n", source, out);
825
826 if (!(out->ctf_flags & LCTF_RDWR))
827 return (ctf_set_errno(out, ECTF_RDONLY));
828
829 if (ctf_getmodel(out) != ctf_getmodel(source))
830 return (ctf_set_errno(out, ECTF_DMODEL));
831
832 if ((ret = ctf_diff_init(out, source, &cdp)) != 0)
833 return (ret);
834
835 cm.cm_out = out;
836 cm.cm_src = source;
837 cm.cm_dedup = B_FALSE;
838 cm.cm_unique = B_FALSE;
839 ret = ctf_merge_types_init(&cm);
840 if (ret != 0) {
841 ctf_diff_fini(cdp);
842 return (ctf_set_errno(out, ret));
843 }
844
845 ret = ctf_diff_types(cdp, ctf_merge_diffcb, &cm);
846 if (ret != 0)
847 goto cleanup;
848 ret = ctf_merge_common(&cm);
849 ctf_dprintf("merge common returned with %d\n", ret);
850 if (ret == 0) {
851 ret = ctf_update(out);
852 ctf_dprintf("update returned with %d\n", ret);
853 } else {
854 goto cleanup;
855 }
856
857 /*
858 * Now we need to fix up the object and function maps.
859 */
860 ctf_merge_fixup_symmaps(&cm, scmi);
861
862 /*
863 * Now that we've fixed things up, we need to give our function and
864 * object maps to the destination, such that it can continue to update
865 * them going forward.
866 */
867 list_move_tail(&dcmi->cmi_fmap, &scmi->cmi_fmap);
868 list_move_tail(&dcmi->cmi_omap, &scmi->cmi_omap);
869
870 cleanup:
871 if (ret == 0)
872 *outp = dcmi;
873 ctf_merge_types_fini(&cm);
874 ctf_diff_fini(cdp);
875 if (ret != 0)
876 return (ctf_errno(out));
877 ctf_phase_bump();
878 return (0);
879 }
880
881 static int
882 ctf_uniquify_types(ctf_merge_t *cmh, ctf_file_t *src, ctf_file_t **outp)
883 {
884 int err, ret;
885 ctf_file_t *out;
886 ctf_merge_types_t cm;
887 ctf_diff_t *cdp;
888 ctf_merge_input_t *cmi;
889 ctf_file_t *parent = cmh->cmh_unique;
890
891 *outp = NULL;
892 out = ctf_fdcreate(cmh->cmh_ofd, &err);
893 if (out == NULL)
894 return (ctf_set_errno(src, err));
895
896 out->ctf_parname = cmh->cmh_pname;
897 if (ctf_setmodel(out, ctf_getmodel(parent)) != 0) {
898 (void) ctf_set_errno(src, ctf_errno(out));
899 ctf_close(out);
900 return (CTF_ERR);
901 }
902
903 if (ctf_import(out, parent) != 0) {
904 (void) ctf_set_errno(src, ctf_errno(out));
905 ctf_close(out);
906 return (CTF_ERR);
907 }
908
909 if ((ret = ctf_diff_init(parent, src, &cdp)) != 0) {
910 ctf_close(out);
911 return (ctf_set_errno(src, ctf_errno(parent)));
912 }
913
914 cm.cm_out = parent;
915 cm.cm_src = src;
916 cm.cm_dedup = B_FALSE;
917 cm.cm_unique = B_TRUE;
918 ret = ctf_merge_types_init(&cm);
919 if (ret != 0) {
920 ctf_close(out);
921 ctf_diff_fini(cdp);
922 return (ctf_set_errno(src, ret));
923 }
924
925 ret = ctf_diff_types(cdp, ctf_merge_diffcb, &cm);
926 if (ret == 0) {
927 cm.cm_out = out;
928 ret = ctf_merge_uniquify_types(&cm);
929 if (ret == 0)
930 ret = ctf_update(out);
931 }
932
933 if (ret != 0) {
934 ctf_merge_types_fini(&cm);
935 ctf_diff_fini(cdp);
936 return (ctf_set_errno(src, ctf_errno(cm.cm_out)));
937 }
938
939 for (cmi = list_head(&cmh->cmh_inputs); cmi != NULL;
940 cmi = list_next(&cmh->cmh_inputs, cmi)) {
941 ctf_merge_fixup_symmaps(&cm, cmi);
942 }
943
944 ctf_merge_types_fini(&cm);
945 ctf_diff_fini(cdp);
946 *outp = out;
947 return (0);
948 }
949
950 static void
951 ctf_merge_fini_input(ctf_merge_input_t *cmi)
952 {
953 ctf_merge_objmap_t *cmo;
954 ctf_merge_funcmap_t *cmf;
955
956 while ((cmo = list_remove_head(&cmi->cmi_omap)) != NULL)
957 ctf_free(cmo, sizeof (ctf_merge_objmap_t));
958
959 while ((cmf = list_remove_head(&cmi->cmi_fmap)) != NULL)
960 ctf_free(cmf, sizeof (ctf_merge_funcmap_t) +
961 sizeof (ctf_id_t) * cmf->cmf_argc);
962
963 if (cmi->cmi_created == B_TRUE && cmi->cmi_input != NULL)
964 ctf_close(cmi->cmi_input);
965
966 ctf_free(cmi, sizeof (ctf_merge_input_t));
967 }
968
969 void
970 ctf_merge_fini(ctf_merge_t *cmh)
971 {
972 size_t len;
973 ctf_merge_input_t *cmi;
974
975 if (cmh->cmh_label != NULL) {
976 len = strlen(cmh->cmh_label) + 1;
977 ctf_free(cmh->cmh_label, len);
978 }
979
980 if (cmh->cmh_pname != NULL) {
981 len = strlen(cmh->cmh_pname) + 1;
982 ctf_free(cmh->cmh_pname, len);
983 }
984
985 while ((cmi = list_remove_head(&cmh->cmh_inputs)) != NULL)
986 ctf_merge_fini_input(cmi);
987
988 ctf_free(cmh, sizeof (ctf_merge_t));
989 }
990
991 ctf_merge_t *
992 ctf_merge_init(int fd, int *errp)
993 {
994 int err;
995 ctf_merge_t *out;
996 struct stat st;
997
998 if (errp == NULL)
999 errp = &err;
1000
1001 if (fd != -1 && fstat(fd, &st) != 0) {
1002 *errp = EINVAL;
1003 return (NULL);
1004 }
1005
1006 out = ctf_alloc(sizeof (ctf_merge_t));
1007 if (out == NULL) {
1008 *errp = ENOMEM;
1009 return (NULL);
1010 }
1011
1012 if (fd == -1) {
1013 out->cmh_msyms = B_FALSE;
1014 } else {
1015 out->cmh_msyms = B_TRUE;
1016 }
1017
1018 list_create(&out->cmh_inputs, sizeof (ctf_merge_input_t),
1019 offsetof(ctf_merge_input_t, cmi_node));
1020 out->cmh_ninputs = 0;
1021 out->cmh_nthreads = 1;
1022 out->cmh_unique = NULL;
1023 out->cmh_ofd = fd;
1024 out->cmh_flags = 0;
1025 out->cmh_label = NULL;
1026 out->cmh_pname = NULL;
1027
1028 return (out);
1029 }
1030
1031 int
1032 ctf_merge_label(ctf_merge_t *cmh, const char *label)
1033 {
1034 char *dup;
1035
1036 if (label == NULL)
1037 return (EINVAL);
1038
1039 dup = ctf_strdup(label);
1040 if (dup == NULL)
1041 return (EAGAIN);
1042
1043 if (cmh->cmh_label != NULL) {
1044 size_t len = strlen(cmh->cmh_label) + 1;
1045 ctf_free(cmh->cmh_label, len);
1046 }
1047
1048 cmh->cmh_label = dup;
1049 return (0);
1050 }
1051
1052 static int
1053 ctf_merge_add_function(ctf_merge_input_t *cmi, ctf_funcinfo_t *fip, ulong_t idx,
1054 const char *file, const char *name, const Elf64_Sym *symp)
1055 {
1056 ctf_merge_funcmap_t *fmap;
1057
1058 fmap = ctf_alloc(sizeof (ctf_merge_funcmap_t) +
1059 sizeof (ctf_id_t) * fip->ctc_argc);
1060 if (fmap == NULL)
1061 return (ENOMEM);
1062
1063 fmap->cmf_idx = idx;
1064 fmap->cmf_sym = *symp;
1065 fmap->cmf_rtid = fip->ctc_return;
1066 fmap->cmf_flags = fip->ctc_flags;
1067 fmap->cmf_argc = fip->ctc_argc;
1068 fmap->cmf_name = name;
1069 if (ELF64_ST_BIND(symp->st_info) == STB_LOCAL) {
1070 fmap->cmf_file = file;
1071 } else {
1072 fmap->cmf_file = NULL;
1073 }
1074
1075 if (ctf_func_args(cmi->cmi_input, idx, fmap->cmf_argc,
1076 fmap->cmf_args) != 0) {
1077 ctf_free(fmap, sizeof (ctf_merge_funcmap_t) +
1078 sizeof (ctf_id_t) * fip->ctc_argc);
1079 return (ctf_errno(cmi->cmi_input));
1080 }
1081
1082 ctf_dprintf("added initial function %s, %lu, %s %u\n", name, idx,
1083 fmap->cmf_file != NULL ? fmap->cmf_file : "global",
1084 ELF64_ST_BIND(symp->st_info));
1085 list_insert_tail(&cmi->cmi_fmap, fmap);
1086 return (0);
1087 }
1088
1089 static int
1090 ctf_merge_add_object(ctf_merge_input_t *cmi, ctf_id_t id, ulong_t idx,
1091 const char *file, const char *name, const Elf64_Sym *symp)
1092 {
1093 ctf_merge_objmap_t *cmo;
1094
1095 cmo = ctf_alloc(sizeof (ctf_merge_objmap_t));
1096 if (cmo == NULL)
1097 return (ENOMEM);
1098
1099 cmo->cmo_name = name;
1100 if (ELF64_ST_BIND(symp->st_info) == STB_LOCAL) {
1101 cmo->cmo_file = file;
1102 } else {
1103 cmo->cmo_file = NULL;
1104 }
1105 cmo->cmo_idx = idx;
1106 cmo->cmo_tid = id;
1107 cmo->cmo_sym = *symp;
1108 list_insert_tail(&cmi->cmi_omap, cmo);
1109
1110 ctf_dprintf("added initial object %s, %lu, %ld, %s\n", name, idx, id,
1111 cmo->cmo_file != NULL ? cmo->cmo_file : "global");
1112
1113 return (0);
1114 }
1115
1116 static int
1117 ctf_merge_add_symbol(const Elf64_Sym *symp, ulong_t idx, const char *file,
1118 const char *name, boolean_t primary, void *arg)
1119 {
1120 ctf_merge_input_t *cmi = arg;
1121 ctf_file_t *fp = cmi->cmi_input;
1122 ushort_t *data, funcbase;
1123 uint_t type;
1124 ctf_funcinfo_t fi;
1125
1126 /*
1127 * See if there is type information for this. If there is no
1128 * type information for this entry or no translation, then we
1129 * will find the value zero. This indicates no type ID for
1130 * objects and encodes unknown information for functions.
1131 */
1132 if (fp->ctf_sxlate[idx] == -1u)
1133 return (0);
1134 data = (ushort_t *)((uintptr_t)fp->ctf_buf + fp->ctf_sxlate[idx]);
1135 if (*data == 0)
1136 return (0);
1137
1138 type = ELF64_ST_TYPE(symp->st_info);
1139
1140 switch (type) {
1141 case STT_FUNC:
1142 funcbase = *data;
1143 if (LCTF_INFO_KIND(fp, funcbase) != CTF_K_FUNCTION)
1144 return (0);
1145 data++;
1146 fi.ctc_return = *data;
1147 data++;
1148 fi.ctc_argc = LCTF_INFO_VLEN(fp, funcbase);
1149 fi.ctc_flags = 0;
1150
1151 if (fi.ctc_argc != 0 && data[fi.ctc_argc - 1] == 0) {
1152 fi.ctc_flags |= CTF_FUNC_VARARG;
1153 fi.ctc_argc--;
1154 }
1155 return (ctf_merge_add_function(cmi, &fi, idx, file, name,
1156 symp));
1157 case STT_OBJECT:
1158 return (ctf_merge_add_object(cmi, *data, idx, file, name,
1159 symp));
1160 default:
1161 return (0);
1162 }
1163 }
1164
1165 /*
1166 * Whenever we create an entry to merge, we then go and add a second empty
1167 * ctf_file_t which we use for the purposes of our merging. It's not the best,
1168 * but it's the best that we've got at the moment.
1169 */
1170 int
1171 ctf_merge_add(ctf_merge_t *cmh, ctf_file_t *input)
1172 {
1173 int ret;
1174 ctf_merge_input_t *cmi;
1175 ctf_file_t *empty;
1176
1177 ctf_dprintf("adding input %p\n", input);
1178
1179 if (input->ctf_flags & LCTF_CHILD)
1180 return (ECTF_MCHILD);
1181
1182 cmi = ctf_alloc(sizeof (ctf_merge_input_t));
1183 if (cmi == NULL)
1184 return (ENOMEM);
1185
1186 cmi->cmi_created = B_FALSE;
1187 cmi->cmi_input = input;
1188 list_create(&cmi->cmi_fmap, sizeof (ctf_merge_funcmap_t),
1189 offsetof(ctf_merge_funcmap_t, cmf_node));
1190 list_create(&cmi->cmi_omap, sizeof (ctf_merge_funcmap_t),
1191 offsetof(ctf_merge_objmap_t, cmo_node));
1192
1193 if (cmh->cmh_msyms == B_TRUE) {
1194 if ((ret = ctf_symtab_iter(input, ctf_merge_add_symbol,
1195 cmi)) != 0) {
1196 ctf_merge_fini_input(cmi);
1197 return (ret);
1198 }
1199 }
1200
1201 list_insert_tail(&cmh->cmh_inputs, cmi);
1202 cmh->cmh_ninputs++;
1203
1204 /* And now the empty one to merge into this */
1205 cmi = ctf_alloc(sizeof (ctf_merge_input_t));
1206 if (cmi == NULL)
1207 return (ENOMEM);
1208 list_create(&cmi->cmi_fmap, sizeof (ctf_merge_funcmap_t),
1209 offsetof(ctf_merge_funcmap_t, cmf_node));
1210 list_create(&cmi->cmi_omap, sizeof (ctf_merge_funcmap_t),
1211 offsetof(ctf_merge_objmap_t, cmo_node));
1212
1213 empty = ctf_fdcreate(cmh->cmh_ofd, &ret);
1214 if (empty == NULL)
1215 return (ret);
1216 cmi->cmi_input = empty;
1217 cmi->cmi_created = B_TRUE;
1218
1219 if (ctf_setmodel(empty, ctf_getmodel(input)) == CTF_ERR) {
1220 return (ctf_errno(empty));
1221 }
1222
1223 list_insert_tail(&cmh->cmh_inputs, cmi);
1224 cmh->cmh_ninputs++;
1225 ctf_dprintf("added containers %p and %p\n", input, empty);
1226 return (0);
1227 }
1228
1229 int
1230 ctf_merge_uniquify(ctf_merge_t *cmh, ctf_file_t *u, const char *pname)
1231 {
1232 char *dup;
1233
1234 if (u->ctf_flags & LCTF_CHILD)
1235 return (ECTF_MCHILD);
1236 if (pname == NULL)
1237 return (EINVAL);
1238 dup = ctf_strdup(pname);
1239 if (dup == NULL)
1240 return (EINVAL);
1241 if (cmh->cmh_pname != NULL) {
1242 size_t len = strlen(cmh->cmh_pname) + 1;
1243 ctf_free(cmh->cmh_pname, len);
1244 }
1245 cmh->cmh_pname = dup;
1246 cmh->cmh_unique = u;
1247 return (0);
1248 }
1249
1250 /*
1251 * Symbol matching rules: the purpose of this is to verify that the type
1252 * information that we have for a given symbol actually matches the output
1253 * symbol. This is unfortunately complicated by several different factors:
1254 *
1255 * 1. When merging multiple .o's into a single item, the symbol table index will
1256 * not match.
1257 *
1258 * 2. Visibility of a symbol may not be identical to the object file or the
1259 * DWARF information due to symbol reduction via a mapfile.
1260 *
1261 * As such, we have to employ the following rules:
1262 *
1263 * 1. A global symbol table entry always matches a global CTF symbol with the
1264 * same name.
1265 *
1266 * 2. A local symbol table entry always matches a local CTF symbol if they have
1267 * the same name and they belong to the same file.
1268 *
1269 * 3. A weak symbol matches a non-weak symbol. This happens if we find that the
1270 * types match, the values match, the sizes match, and the section indexes
1271 * match. This happens when we do a conversion in one pass, it almost never
1272 * happens when we're merging multiple object files. If we match a CTF global
1273 * symbol, that's a fixed match, otherwise it's a fuzzy match.
1274 *
1275 * 4. A local symbol table entry matches a global CTF entry if the
1276 * other pieces fail, but they have the same name. This is considered a fuzzy
1277 * match and is not used unless we have no other options.
1278 *
1279 * 5. A weak symbol table entry matches a weak CTF entry if the other pieces
1280 * fail, but they have the same name. This is considered a fuzzy match and is
1281 * not used unless we have no other options. When merging independent .o files,
1282 * this is often the only recourse we have to matching weak symbols.
1283 *
1284 * In the end, this would all be much simpler if we were able to do this as part
1285 * of libld which would be able to do all the symbol transformations.
1286 */
1287 static boolean_t
1288 ctf_merge_symbol_match(const char *ctf_file, const char *ctf_name,
1289 const Elf64_Sym *ctf_symp, const char *symtab_file, const char *symtab_name,
1290 const Elf64_Sym *symtab_symp, boolean_t *is_fuzzy)
1291 {
1292 *is_fuzzy = B_FALSE;
1293 uint_t symtab_bind, ctf_bind;
1294
1295 symtab_bind = ELF64_ST_BIND(symtab_symp->st_info);
1296 ctf_bind = ELF64_ST_BIND(ctf_symp->st_info);
1297
1298 ctf_dprintf("comparing merge match for %s/%s/%u->%s/%s/%u\n",
1299 symtab_file, symtab_name, symtab_bind,
1300 ctf_file, ctf_name, ctf_bind);
1301 if (strcmp(ctf_name, symtab_name) != 0) {
1302 return (B_FALSE);
1303 }
1304
1305 if (symtab_bind == STB_GLOBAL && ctf_bind == STB_GLOBAL) {
1306 return (B_TRUE);
1307 } else if (symtab_bind == STB_GLOBAL) {
1308 return (B_FALSE);
1309 }
1310
1311 if (ctf_bind == STB_LOCAL && ctf_bind == symtab_bind &&
1312 ctf_file != NULL && symtab_file != NULL &&
1313 strcmp(ctf_file, symtab_file) == 0) {
1314 return (B_TRUE);
1315 }
1316
1317 if (symtab_bind == STB_WEAK && ctf_bind != STB_WEAK &&
1318 ELF64_ST_TYPE(symtab_symp->st_info) ==
1319 ELF64_ST_TYPE(ctf_symp->st_info) &&
1320 symtab_symp->st_value == ctf_symp->st_value &&
1321 symtab_symp->st_size == ctf_symp->st_size &&
1322 symtab_symp->st_shndx == ctf_symp->st_shndx) {
1323 if (ctf_bind == STB_GLOBAL) {
1324 return (B_TRUE);
1325 }
1326
1327 if (ctf_bind == STB_LOCAL && ctf_file != NULL &&
1328 symtab_file != NULL && strcmp(ctf_file, symtab_file) == 0) {
1329 *is_fuzzy = B_TRUE;
1330 return (B_TRUE);
1331 }
1332 }
1333
1334 if (ctf_bind == STB_GLOBAL ||
1335 (ctf_bind == STB_WEAK && symtab_bind == STB_WEAK)) {
1336 *is_fuzzy = B_TRUE;
1337 return (B_TRUE);
1338 }
1339
1340 return (B_FALSE);
1341 }
1342
1343 /*
1344 * For each symbol, try and find a match. We will attempt to find an exact
1345 * match; however, we will settle for a fuzzy match in general. There is one
1346 * case where we will not opt to use a fuzzy match, which is when performing the
1347 * deduplication of a container. In such a case we are trying to reduce common
1348 * types and a fuzzy match would be inappropriate as if we're in the context of
1349 * a single container, the conversion process should have identified any exact
1350 * or fuzzy matches that were required.
1351 */
1352 static int
1353 ctf_merge_symbols(const Elf64_Sym *symp, ulong_t idx, const char *file,
1354 const char *name, boolean_t primary, void *arg)
1355 {
1356 int err;
1357 uint_t type, bind;
1358 ctf_merge_symbol_arg_t *csa = arg;
1359 ctf_file_t *fp = csa->cmsa_out;
1360
1361 type = ELF64_ST_TYPE(symp->st_info);
1362 bind = ELF64_ST_BIND(symp->st_info);
1363
1364 ctf_dprintf("Trying to find match for %s/%s/%u\n", file, name,
1365 ELF64_ST_BIND(symp->st_info));
1366
1367 if (type == STT_OBJECT) {
1368 ctf_merge_objmap_t *cmo, *match = NULL;
1369
1370 for (cmo = list_head(csa->cmsa_objmap); cmo != NULL;
1371 cmo = list_next(csa->cmsa_objmap, cmo)) {
1372 boolean_t is_fuzzy = B_FALSE;
1373 if (ctf_merge_symbol_match(cmo->cmo_file, cmo->cmo_name,
1374 &cmo->cmo_sym, file, name, symp, &is_fuzzy)) {
1375 if (is_fuzzy && csa->cmsa_dedup &&
1376 bind != STB_WEAK) {
1377 continue;
1378 }
1379 match = cmo;
1380 if (is_fuzzy) {
1381 continue;
1382 }
1383 break;
1384 }
1385 }
1386
1387 if (match == NULL) {
1388 return (0);
1389 }
1390
1391 if ((err = ctf_add_object(fp, idx, match->cmo_tid)) != 0) {
1392 ctf_dprintf("Failed to add symbol %s->%d: %s\n", name,
1393 match->cmo_tid, ctf_errmsg(ctf_errno(fp)));
1394 return (ctf_errno(fp));
1395 }
1396 ctf_dprintf("mapped object into output %s/%s->%ld\n", file,
1397 name, match->cmo_tid);
1398 } else {
1399 ctf_merge_funcmap_t *cmf, *match = NULL;
1400 ctf_funcinfo_t fi;
1401
1402 for (cmf = list_head(csa->cmsa_funcmap); cmf != NULL;
1403 cmf = list_next(csa->cmsa_funcmap, cmf)) {
1404 boolean_t is_fuzzy = B_FALSE;
1405 if (ctf_merge_symbol_match(cmf->cmf_file, cmf->cmf_name,
1406 &cmf->cmf_sym, file, name, symp, &is_fuzzy)) {
1407 if (is_fuzzy && csa->cmsa_dedup &&
1408 bind != STB_WEAK) {
1409 continue;
1410 }
1411 match = cmf;
1412 if (is_fuzzy) {
1413 continue;
1414 }
1415 break;
1416 }
1417 }
1418
1419 if (match == NULL) {
1420 return (0);
1421 }
1422
1423 fi.ctc_return = match->cmf_rtid;
1424 fi.ctc_argc = match->cmf_argc;
1425 fi.ctc_flags = match->cmf_flags;
1426 if ((err = ctf_add_function(fp, idx, &fi, match->cmf_args)) !=
1427 0) {
1428 ctf_dprintf("Failed to add function %s: %s\n", name,
1429 ctf_errmsg(ctf_errno(fp)));
1430 return (ctf_errno(fp));
1431 }
1432 ctf_dprintf("mapped function into output %s/%s\n", file,
1433 name);
1434 }
1435
1436 return (0);
1437 }
1438
1439 int
1440 ctf_merge_merge(ctf_merge_t *cmh, ctf_file_t **outp)
1441 {
1442 int err, merr;
1443 ctf_merge_input_t *cmi;
1444 ctf_id_t ltype;
1445 mergeq_t *mqp;
1446 ctf_merge_input_t *final;
1447 ctf_file_t *out;
1448
1449 ctf_dprintf("Beginning ctf_merge_merge()\n");
1450 if (cmh->cmh_label != NULL && cmh->cmh_unique != NULL) {
1451 const char *label = ctf_label_topmost(cmh->cmh_unique);
1452 if (label == NULL)
1453 return (ECTF_NOLABEL);
1454 if (strcmp(label, cmh->cmh_label) != 0)
1455 return (ECTF_LCONFLICT);
1456 }
1457
1458 if (mergeq_init(&mqp, cmh->cmh_nthreads) == -1) {
1459 return (errno);
1460 }
1461
1462 VERIFY(cmh->cmh_ninputs % 2 == 0);
1463 for (cmi = list_head(&cmh->cmh_inputs); cmi != NULL;
1464 cmi = list_next(&cmh->cmh_inputs, cmi)) {
1465 if (mergeq_add(mqp, cmi) == -1) {
1466 err = errno;
1467 mergeq_fini(mqp);
1468 }
1469 }
1470
1471 err = mergeq_merge(mqp, ctf_merge_types, NULL, (void **)&final, &merr);
1472 mergeq_fini(mqp);
1473
1474 if (err == MERGEQ_ERROR) {
1475 return (errno);
1476 } else if (err == MERGEQ_UERROR) {
1477 return (merr);
1478 }
1479
1480 /*
1481 * Disassociate the generated ctf_file_t from the original input. That
1482 * way when the input gets cleaned up, we don't accidentally kill the
1483 * final reference to the ctf_file_t. If it gets uniquified then we'll
1484 * kill it.
1485 */
1486 VERIFY(final->cmi_input != NULL);
1487 out = final->cmi_input;
1488 final->cmi_input = NULL;
1489
1490 ctf_dprintf("preparing to uniquify against: %p\n", cmh->cmh_unique);
1491 if (cmh->cmh_unique != NULL) {
1492 ctf_file_t *u;
1493 err = ctf_uniquify_types(cmh, out, &u);
1494 if (err != 0) {
1495 err = ctf_errno(out);
1496 ctf_close(out);
1497 return (err);
1498 }
1499 ctf_close(out);
1500 out = u;
1501 }
1502
1503 ltype = out->ctf_typemax;
1504 if ((out->ctf_flags & LCTF_CHILD) && ltype != 0)
1505 ltype += CTF_CHILD_START;
1506 ctf_dprintf("trying to add the label\n");
1507 if (cmh->cmh_label != NULL &&
1508 ctf_add_label(out, cmh->cmh_label, ltype, 0) != 0) {
1509 ctf_close(out);
1510 return (ctf_errno(out));
1511 }
1512
1513 ctf_dprintf("merging symbols and the like\n");
1514 if (cmh->cmh_msyms == B_TRUE) {
1515 ctf_merge_symbol_arg_t arg;
1516 arg.cmsa_objmap = &final->cmi_omap;
1517 arg.cmsa_funcmap = &final->cmi_fmap;
1518 arg.cmsa_out = out;
1519 arg.cmsa_dedup = B_FALSE;
1520 err = ctf_symtab_iter(out, ctf_merge_symbols, &arg);
1521 if (err != 0) {
1522 ctf_close(out);
1523 return (err);
1524 }
1525 }
1526
1527 err = ctf_update(out);
1528 if (err != 0) {
1529 err = ctf_errno(out);
1530 ctf_close(out);
1531 return (err);
1532 }
1533
1534 *outp = out;
1535 return (0);
1536 }
1537
1538 /*
1539 * When we get told that something is unique, eg. same is B_FALSE, then that
1540 * tells us that we need to add it to the output. If same is B_TRUE, then we'll
1541 * want to record it in the mapping table so that we know how to redirect types
1542 * to the extant ones.
1543 */
1544 static void
1545 ctf_dedup_cb(ctf_file_t *ifp, ctf_id_t iid, boolean_t same, ctf_file_t *ofp,
1546 ctf_id_t oid, void *arg)
1547 {
1548 ctf_merge_types_t *cmp = arg;
1549 ctf_merge_tinfo_t *cmt = cmp->cm_tmap;
1550
1551 if (same == B_TRUE) {
1552 /*
1553 * The output id here may itself map to something else.
1554 * Therefore, we need to basically walk a chain and see what it
1555 * points to until it itself points to a base type, eg. -1.
1556 * Otherwise we'll dedup to something which no longer exists.
1557 */
1558 while (cmt[oid].cmt_missing == B_FALSE)
1559 oid = cmt[oid].cmt_map;
1560 cmt[iid].cmt_map = oid;
1561 ctf_dprintf("%d->%d \n", iid, oid);
1562 } else {
1563 VERIFY(cmt[iid].cmt_map == 0);
1564 cmt[iid].cmt_missing = B_TRUE;
1565 ctf_dprintf("%d is missing\n", iid);
1566 }
1567 }
1568
1569 /*
1570 * Dedup a CTF container.
1571 *
1572 * DWARF and other encoding formats that we use to create CTF data may create
1573 * multiple copies of a given type. However, after doing a conversion, and
1574 * before doing a merge, we'd prefer, if possible, to have every input container
1575 * to be unique.
1576 *
1577 * Doing a deduplication is like a normal merge. However, when we diff the types
1578 * in the container, rather than doing a normal diff, we instead want to diff
1579 * against any already processed types. eg, for a given type i in a container,
1580 * we want to diff it from 0 to i - 1.
1581 */
1582 int
1583 ctf_merge_dedup(ctf_merge_t *cmp, ctf_file_t **outp)
1584 {
1585 int ret;
1586 ctf_diff_t *cdp = NULL;
1587 ctf_merge_input_t *cmi, *cmc;
1588 ctf_file_t *ifp, *ofp;
1589 ctf_merge_types_t cm;
1590
1591 if (cmp == NULL || outp == NULL)
1592 return (EINVAL);
1593
1594 ctf_dprintf("encountered %d inputs\n", cmp->cmh_ninputs);
1595 if (cmp->cmh_ninputs != 2)
1596 return (EINVAL);
1597
1598 ctf_dprintf("passed argument sanity check\n");
1599
1600 cmi = list_head(&cmp->cmh_inputs);
1601 VERIFY(cmi != NULL);
1602 cmc = list_next(&cmp->cmh_inputs, cmi);
1603 VERIFY(cmc != NULL);
1604 ifp = cmi->cmi_input;
1605 ofp = cmc->cmi_input;
1606 VERIFY(ifp != NULL);
1607 VERIFY(ofp != NULL);
1608 cm.cm_src = ifp;
1609 cm.cm_out = ofp;
1610 cm.cm_dedup = B_TRUE;
1611 cm.cm_unique = B_FALSE;
1612
1613 if ((ret = ctf_merge_types_init(&cm)) != 0) {
1614 return (ret);
1615 }
1616
1617 if ((ret = ctf_diff_init(ifp, ifp, &cdp)) != 0)
1618 goto err;
1619
1620 ctf_dprintf("Successfully initialized dedup\n");
1621 if ((ret = ctf_diff_self(cdp, ctf_dedup_cb, &cm)) != 0)
1622 goto err;
1623
1624 ctf_dprintf("Successfully diffed types\n");
1625 ret = ctf_merge_common(&cm);
1626 ctf_dprintf("deduping types result: %d\n", ret);
1627 if (ret == 0)
1628 ret = ctf_update(cm.cm_out);
1629 if (ret != 0)
1630 goto err;
1631
1632 ctf_dprintf("Successfully deduped types\n");
1633 ctf_phase_dump(cm.cm_out, "dedup-pre-syms", NULL);
1634
1635 /*
1636 * Now we need to fix up the object and function maps.
1637 */
1638 ctf_merge_fixup_symmaps(&cm, cmi);
1639
1640 if (cmp->cmh_msyms == B_TRUE) {
1641 ctf_merge_symbol_arg_t arg;
1642 arg.cmsa_objmap = &cmi->cmi_omap;
1643 arg.cmsa_funcmap = &cmi->cmi_fmap;
1644 arg.cmsa_out = cm.cm_out;
1645 arg.cmsa_dedup = B_TRUE;
1646 ret = ctf_symtab_iter(cm.cm_out, ctf_merge_symbols, &arg);
1647 if (ret != 0) {
1648 ctf_dprintf("failed to dedup symbols: %s\n",
1649 ctf_errmsg(ret));
1650 goto err;
1651 }
1652 }
1653
1654 ret = ctf_update(cm.cm_out);
1655 if (ret == 0) {
1656 cmc->cmi_input = NULL;
1657 *outp = cm.cm_out;
1658 }
1659 ctf_phase_dump(cm.cm_out, "dedup-post-syms", NULL);
1660 err:
1661 ctf_merge_types_fini(&cm);
1662 ctf_diff_fini(cdp);
1663 return (ret);
1664 }
1665
1666 int
1667 ctf_merge_set_nthreads(ctf_merge_t *cmp, const uint_t nthrs)
1668 {
1669 if (nthrs == 0)
1670 return (EINVAL);
1671 cmp->cmh_nthreads = nthrs;
1672 return (0);
1673 }