1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * resolve_btfids scans Elf object for .BTF_ids section and resolves
5 * its symbols with BTF ID values.
6 *
7 * Each symbol points to 4 bytes data and is expected to have
8 * following name syntax:
9 *
10 * __BTF_ID__<type>__<symbol>[__<id>]
11 *
12 * type is:
13 *
14 * func - lookup BTF_KIND_FUNC symbol with <symbol> name
15 * and store its ID into the data:
16 *
17 * __BTF_ID__func__vfs_close__1:
18 * .zero 4
19 *
20 * struct - lookup BTF_KIND_STRUCT symbol with <symbol> name
21 * and store its ID into the data:
22 *
23 * __BTF_ID__struct__sk_buff__1:
24 * .zero 4
25 *
26 * union - lookup BTF_KIND_UNION symbol with <symbol> name
27 * and store its ID into the data:
28 *
29 * __BTF_ID__union__thread_union__1:
30 * .zero 4
31 *
32 * typedef - lookup BTF_KIND_TYPEDEF symbol with <symbol> name
33 * and store its ID into the data:
34 *
35 * __BTF_ID__typedef__pid_t__1:
36 * .zero 4
37 *
38 * set - store symbol size into first 4 bytes and sort following
39 * ID list
40 *
41 * __BTF_ID__set__list:
42 * .zero 4
43 * list:
44 * __BTF_ID__func__vfs_getattr__3:
45 * .zero 4
46 * __BTF_ID__func__vfs_fallocate__4:
47 * .zero 4
48 */
49
50 #define _GNU_SOURCE
51 #include <stdio.h>
52 #include <string.h>
53 #include <unistd.h>
54 #include <stdlib.h>
55 #include <libelf.h>
56 #include <gelf.h>
57 #include <sys/stat.h>
58 #include <fcntl.h>
59 #include <errno.h>
60 #include <linux/rbtree.h>
61 #include <linux/zalloc.h>
62 #include <linux/err.h>
63 #include <bpf/btf.h>
64 #include <bpf/libbpf.h>
65 #include <parse-options.h>
66
67 #define BTF_IDS_SECTION ".BTF_ids"
68 #define BTF_ID "__BTF_ID__"
69
70 #define BTF_STRUCT "struct"
71 #define BTF_UNION "union"
72 #define BTF_TYPEDEF "typedef"
73 #define BTF_FUNC "func"
74 #define BTF_SET "set"
75
76 #define ADDR_CNT 100
77
78 struct btf_id {
79 struct rb_node rb_node;
80 char *name;
81 union {
82 int id;
83 int cnt;
84 };
85 int addr_cnt;
86 bool is_set;
87 Elf64_Addr addr[ADDR_CNT];
88 };
89
90 struct object {
91 const char *path;
92 const char *btf;
93 const char *base_btf_path;
94
95 struct {
96 int fd;
97 Elf *elf;
98 Elf_Data *symbols;
99 Elf_Data *idlist;
100 int symbols_shndx;
101 int idlist_shndx;
102 size_t strtabidx;
103 unsigned long idlist_addr;
104 } efile;
105
106 struct rb_root sets;
107 struct rb_root structs;
108 struct rb_root unions;
109 struct rb_root typedefs;
110 struct rb_root funcs;
111
112 int nr_funcs;
113 int nr_structs;
114 int nr_unions;
115 int nr_typedefs;
116 };
117
118 static int verbose;
119
eprintf(int level,int var,const char * fmt,...)120 static int eprintf(int level, int var, const char *fmt, ...)
121 {
122 va_list args;
123 int ret = 0;
124
125 if (var >= level) {
126 va_start(args, fmt);
127 ret = vfprintf(stderr, fmt, args);
128 va_end(args);
129 }
130 return ret;
131 }
132
133 #ifndef pr_fmt
134 #define pr_fmt(fmt) fmt
135 #endif
136
137 #define pr_debug(fmt, ...) \
138 eprintf(1, verbose, pr_fmt(fmt), ##__VA_ARGS__)
139 #define pr_debugN(n, fmt, ...) \
140 eprintf(n, verbose, pr_fmt(fmt), ##__VA_ARGS__)
141 #define pr_debug2(fmt, ...) pr_debugN(2, pr_fmt(fmt), ##__VA_ARGS__)
142 #define pr_err(fmt, ...) \
143 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__)
144 #define pr_info(fmt, ...) \
145 eprintf(0, verbose, pr_fmt(fmt), ##__VA_ARGS__)
146
is_btf_id(const char * name)147 static bool is_btf_id(const char *name)
148 {
149 return name && !strncmp(name, BTF_ID, sizeof(BTF_ID) - 1);
150 }
151
btf_id__find(struct rb_root * root,const char * name)152 static struct btf_id *btf_id__find(struct rb_root *root, const char *name)
153 {
154 struct rb_node *p = root->rb_node;
155 struct btf_id *id;
156 int cmp;
157
158 while (p) {
159 id = rb_entry(p, struct btf_id, rb_node);
160 cmp = strcmp(id->name, name);
161 if (cmp < 0)
162 p = p->rb_left;
163 else if (cmp > 0)
164 p = p->rb_right;
165 else
166 return id;
167 }
168 return NULL;
169 }
170
171 static struct btf_id *
btf_id__add(struct rb_root * root,char * name,bool unique)172 btf_id__add(struct rb_root *root, char *name, bool unique)
173 {
174 struct rb_node **p = &root->rb_node;
175 struct rb_node *parent = NULL;
176 struct btf_id *id;
177 int cmp;
178
179 while (*p != NULL) {
180 parent = *p;
181 id = rb_entry(parent, struct btf_id, rb_node);
182 cmp = strcmp(id->name, name);
183 if (cmp < 0)
184 p = &(*p)->rb_left;
185 else if (cmp > 0)
186 p = &(*p)->rb_right;
187 else
188 return unique ? NULL : id;
189 }
190
191 id = zalloc(sizeof(*id));
192 if (id) {
193 pr_debug("adding symbol %s\n", name);
194 id->name = name;
195 rb_link_node(&id->rb_node, parent, p);
196 rb_insert_color(&id->rb_node, root);
197 }
198 return id;
199 }
200
get_id(const char * prefix_end)201 static char *get_id(const char *prefix_end)
202 {
203 /*
204 * __BTF_ID__func__vfs_truncate__0
205 * prefix_end = ^
206 * pos = ^
207 */
208 int len = strlen(prefix_end);
209 int pos = sizeof("__") - 1;
210 char *p, *id;
211
212 if (pos >= len)
213 return NULL;
214
215 id = strdup(prefix_end + pos);
216 if (id) {
217 /*
218 * __BTF_ID__func__vfs_truncate__0
219 * id = ^
220 *
221 * cut the unique id part
222 */
223 p = strrchr(id, '_');
224 p--;
225 if (*p != '_') {
226 free(id);
227 return NULL;
228 }
229 *p = '\0';
230 }
231 return id;
232 }
233
add_set(struct object * obj,char * name)234 static struct btf_id *add_set(struct object *obj, char *name)
235 {
236 /*
237 * __BTF_ID__set__name
238 * name = ^
239 * id = ^
240 */
241 char *id = name + sizeof(BTF_SET "__") - 1;
242 int len = strlen(name);
243
244 if (id >= name + len) {
245 pr_err("FAILED to parse set name: %s\n", name);
246 return NULL;
247 }
248
249 return btf_id__add(&obj->sets, id, true);
250 }
251
add_symbol(struct rb_root * root,char * name,size_t size)252 static struct btf_id *add_symbol(struct rb_root *root, char *name, size_t size)
253 {
254 char *id;
255
256 id = get_id(name + size);
257 if (!id) {
258 pr_err("FAILED to parse symbol name: %s\n", name);
259 return NULL;
260 }
261
262 return btf_id__add(root, id, false);
263 }
264
265 /* Older libelf.h and glibc elf.h might not yet define the ELF compression types. */
266 #ifndef SHF_COMPRESSED
267 #define SHF_COMPRESSED (1 << 11) /* Section with compressed data. */
268 #endif
269
270 /*
271 * The data of compressed section should be aligned to 4
272 * (for 32bit) or 8 (for 64 bit) bytes. The binutils ld
273 * sets sh_addralign to 1, which makes libelf fail with
274 * misaligned section error during the update:
275 * FAILED elf_update(WRITE): invalid section alignment
276 *
277 * While waiting for ld fix, we fix the compressed sections
278 * sh_addralign value manualy.
279 */
compressed_section_fix(Elf * elf,Elf_Scn * scn,GElf_Shdr * sh)280 static int compressed_section_fix(Elf *elf, Elf_Scn *scn, GElf_Shdr *sh)
281 {
282 int expected = gelf_getclass(elf) == ELFCLASS32 ? 4 : 8;
283
284 if (!(sh->sh_flags & SHF_COMPRESSED))
285 return 0;
286
287 if (sh->sh_addralign == expected)
288 return 0;
289
290 pr_debug2(" - fixing wrong alignment sh_addralign %u, expected %u\n",
291 sh->sh_addralign, expected);
292
293 sh->sh_addralign = expected;
294
295 if (gelf_update_shdr(scn, sh) == 0) {
296 pr_err("FAILED cannot update section header: %s\n",
297 elf_errmsg(-1));
298 return -1;
299 }
300 return 0;
301 }
302
elf_collect(struct object * obj)303 static int elf_collect(struct object *obj)
304 {
305 Elf_Scn *scn = NULL;
306 size_t shdrstrndx;
307 int idx = 0;
308 Elf *elf;
309 int fd;
310
311 fd = open(obj->path, O_RDWR, 0666);
312 if (fd == -1) {
313 pr_err("FAILED cannot open %s: %s\n",
314 obj->path, strerror(errno));
315 return -1;
316 }
317
318 elf_version(EV_CURRENT);
319
320 elf = elf_begin(fd, ELF_C_RDWR_MMAP, NULL);
321 if (!elf) {
322 close(fd);
323 pr_err("FAILED cannot create ELF descriptor: %s\n",
324 elf_errmsg(-1));
325 return -1;
326 }
327
328 obj->efile.fd = fd;
329 obj->efile.elf = elf;
330
331 elf_flagelf(elf, ELF_C_SET, ELF_F_LAYOUT);
332
333 if (elf_getshdrstrndx(elf, &shdrstrndx) != 0) {
334 pr_err("FAILED cannot get shdr str ndx\n");
335 return -1;
336 }
337
338 /*
339 * Scan all the elf sections and look for save data
340 * from .BTF_ids section and symbols.
341 */
342 while ((scn = elf_nextscn(elf, scn)) != NULL) {
343 Elf_Data *data;
344 GElf_Shdr sh;
345 char *name;
346
347 idx++;
348 if (gelf_getshdr(scn, &sh) != &sh) {
349 pr_err("FAILED get section(%d) header\n", idx);
350 return -1;
351 }
352
353 name = elf_strptr(elf, shdrstrndx, sh.sh_name);
354 if (!name) {
355 pr_err("FAILED get section(%d) name\n", idx);
356 return -1;
357 }
358
359 data = elf_getdata(scn, 0);
360 if (!data) {
361 pr_err("FAILED to get section(%d) data from %s\n",
362 idx, name);
363 return -1;
364 }
365
366 pr_debug2("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
367 idx, name, (unsigned long) data->d_size,
368 (int) sh.sh_link, (unsigned long) sh.sh_flags,
369 (int) sh.sh_type);
370
371 if (sh.sh_type == SHT_SYMTAB) {
372 obj->efile.symbols = data;
373 obj->efile.symbols_shndx = idx;
374 obj->efile.strtabidx = sh.sh_link;
375 } else if (!strcmp(name, BTF_IDS_SECTION)) {
376 obj->efile.idlist = data;
377 obj->efile.idlist_shndx = idx;
378 obj->efile.idlist_addr = sh.sh_addr;
379 }
380
381 if (compressed_section_fix(elf, scn, &sh))
382 return -1;
383 }
384
385 return 0;
386 }
387
symbols_collect(struct object * obj)388 static int symbols_collect(struct object *obj)
389 {
390 Elf_Scn *scn = NULL;
391 int n, i;
392 GElf_Shdr sh;
393 char *name;
394
395 scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
396 if (!scn)
397 return -1;
398
399 if (gelf_getshdr(scn, &sh) != &sh)
400 return -1;
401
402 n = sh.sh_size / sh.sh_entsize;
403
404 /*
405 * Scan symbols and look for the ones starting with
406 * __BTF_ID__* over .BTF_ids section.
407 */
408 for (i = 0; i < n; i++) {
409 char *prefix;
410 struct btf_id *id;
411 GElf_Sym sym;
412
413 if (!gelf_getsym(obj->efile.symbols, i, &sym))
414 return -1;
415
416 if (sym.st_shndx != obj->efile.idlist_shndx)
417 continue;
418
419 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
420 sym.st_name);
421
422 if (!is_btf_id(name))
423 continue;
424
425 /*
426 * __BTF_ID__TYPE__vfs_truncate__0
427 * prefix = ^
428 */
429 prefix = name + sizeof(BTF_ID) - 1;
430
431 /* struct */
432 if (!strncmp(prefix, BTF_STRUCT, sizeof(BTF_STRUCT) - 1)) {
433 obj->nr_structs++;
434 id = add_symbol(&obj->structs, prefix, sizeof(BTF_STRUCT) - 1);
435 /* union */
436 } else if (!strncmp(prefix, BTF_UNION, sizeof(BTF_UNION) - 1)) {
437 obj->nr_unions++;
438 id = add_symbol(&obj->unions, prefix, sizeof(BTF_UNION) - 1);
439 /* typedef */
440 } else if (!strncmp(prefix, BTF_TYPEDEF, sizeof(BTF_TYPEDEF) - 1)) {
441 obj->nr_typedefs++;
442 id = add_symbol(&obj->typedefs, prefix, sizeof(BTF_TYPEDEF) - 1);
443 /* func */
444 } else if (!strncmp(prefix, BTF_FUNC, sizeof(BTF_FUNC) - 1)) {
445 obj->nr_funcs++;
446 id = add_symbol(&obj->funcs, prefix, sizeof(BTF_FUNC) - 1);
447 /* set */
448 } else if (!strncmp(prefix, BTF_SET, sizeof(BTF_SET) - 1)) {
449 id = add_set(obj, prefix);
450 /*
451 * SET objects store list's count, which is encoded
452 * in symbol's size, together with 'cnt' field hence
453 * that - 1.
454 */
455 if (id) {
456 id->cnt = sym.st_size / sizeof(int) - 1;
457 id->is_set = true;
458 }
459 } else {
460 pr_err("FAILED unsupported prefix %s\n", prefix);
461 return -1;
462 }
463
464 if (!id)
465 return -ENOMEM;
466
467 if (id->addr_cnt >= ADDR_CNT) {
468 pr_err("FAILED symbol %s crossed the number of allowed lists\n",
469 id->name);
470 return -1;
471 }
472 id->addr[id->addr_cnt++] = sym.st_value;
473 }
474
475 return 0;
476 }
477
symbols_resolve(struct object * obj)478 static int symbols_resolve(struct object *obj)
479 {
480 int nr_typedefs = obj->nr_typedefs;
481 int nr_structs = obj->nr_structs;
482 int nr_unions = obj->nr_unions;
483 int nr_funcs = obj->nr_funcs;
484 struct btf *base_btf = NULL;
485 int err, type_id;
486 struct btf *btf;
487 __u32 nr_types;
488
489 if (obj->base_btf_path) {
490 base_btf = btf__parse(obj->base_btf_path, NULL);
491 err = libbpf_get_error(base_btf);
492 if (err) {
493 pr_err("FAILED: load base BTF from %s: %s\n",
494 obj->base_btf_path, strerror(-err));
495 return -1;
496 }
497 }
498
499 btf = btf__parse_split(obj->btf ?: obj->path, base_btf);
500 err = libbpf_get_error(btf);
501 if (err) {
502 pr_err("FAILED: load BTF from %s: %s\n",
503 obj->btf ?: obj->path, strerror(-err));
504 goto out;
505 }
506
507 err = -1;
508 nr_types = btf__type_cnt(btf);
509
510 /*
511 * Iterate all the BTF types and search for collected symbol IDs.
512 */
513 for (type_id = 1; type_id < nr_types; type_id++) {
514 const struct btf_type *type;
515 struct rb_root *root;
516 struct btf_id *id;
517 const char *str;
518 int *nr;
519
520 type = btf__type_by_id(btf, type_id);
521 if (!type) {
522 pr_err("FAILED: malformed BTF, can't resolve type for ID %d\n",
523 type_id);
524 goto out;
525 }
526
527 if (btf_is_func(type) && nr_funcs) {
528 nr = &nr_funcs;
529 root = &obj->funcs;
530 } else if (btf_is_struct(type) && nr_structs) {
531 nr = &nr_structs;
532 root = &obj->structs;
533 } else if (btf_is_union(type) && nr_unions) {
534 nr = &nr_unions;
535 root = &obj->unions;
536 } else if (btf_is_typedef(type) && nr_typedefs) {
537 nr = &nr_typedefs;
538 root = &obj->typedefs;
539 } else
540 continue;
541
542 str = btf__name_by_offset(btf, type->name_off);
543 if (!str) {
544 pr_err("FAILED: malformed BTF, can't resolve name for ID %d\n",
545 type_id);
546 goto out;
547 }
548
549 id = btf_id__find(root, str);
550 if (id) {
551 if (id->id) {
552 pr_info("WARN: multiple IDs found for '%s': %d, %d - using %d\n",
553 str, id->id, type_id, id->id);
554 } else {
555 id->id = type_id;
556 (*nr)--;
557 }
558 }
559 }
560
561 err = 0;
562 out:
563 btf__free(base_btf);
564 btf__free(btf);
565 return err;
566 }
567
id_patch(struct object * obj,struct btf_id * id)568 static int id_patch(struct object *obj, struct btf_id *id)
569 {
570 Elf_Data *data = obj->efile.idlist;
571 int *ptr = data->d_buf;
572 int i;
573
574 if (!id->id && !id->is_set)
575 pr_err("WARN: resolve_btfids: unresolved symbol %s\n", id->name);
576
577 for (i = 0; i < id->addr_cnt; i++) {
578 unsigned long addr = id->addr[i];
579 unsigned long idx = addr - obj->efile.idlist_addr;
580
581 pr_debug("patching addr %5lu: ID %7d [%s]\n",
582 idx, id->id, id->name);
583
584 if (idx >= data->d_size) {
585 pr_err("FAILED patching index %lu out of bounds %lu\n",
586 idx, data->d_size);
587 return -1;
588 }
589
590 idx = idx / sizeof(int);
591 ptr[idx] = id->id;
592 }
593
594 return 0;
595 }
596
__symbols_patch(struct object * obj,struct rb_root * root)597 static int __symbols_patch(struct object *obj, struct rb_root *root)
598 {
599 struct rb_node *next;
600 struct btf_id *id;
601
602 next = rb_first(root);
603 while (next) {
604 id = rb_entry(next, struct btf_id, rb_node);
605
606 if (id_patch(obj, id))
607 return -1;
608
609 next = rb_next(next);
610 }
611 return 0;
612 }
613
cmp_id(const void * pa,const void * pb)614 static int cmp_id(const void *pa, const void *pb)
615 {
616 const int *a = pa, *b = pb;
617
618 return *a - *b;
619 }
620
sets_patch(struct object * obj)621 static int sets_patch(struct object *obj)
622 {
623 Elf_Data *data = obj->efile.idlist;
624 int *ptr = data->d_buf;
625 struct rb_node *next;
626
627 next = rb_first(&obj->sets);
628 while (next) {
629 unsigned long addr, idx;
630 struct btf_id *id;
631 int *base;
632 int cnt;
633
634 id = rb_entry(next, struct btf_id, rb_node);
635 addr = id->addr[0];
636 idx = addr - obj->efile.idlist_addr;
637
638 /* sets are unique */
639 if (id->addr_cnt != 1) {
640 pr_err("FAILED malformed data for set '%s'\n",
641 id->name);
642 return -1;
643 }
644
645 idx = idx / sizeof(int);
646 base = &ptr[idx] + 1;
647 cnt = ptr[idx];
648
649 pr_debug("sorting addr %5lu: cnt %6d [%s]\n",
650 (idx + 1) * sizeof(int), cnt, id->name);
651
652 qsort(base, cnt, sizeof(int), cmp_id);
653
654 next = rb_next(next);
655 }
656 return 0;
657 }
658
symbols_patch(struct object * obj)659 static int symbols_patch(struct object *obj)
660 {
661 int err;
662
663 if (__symbols_patch(obj, &obj->structs) ||
664 __symbols_patch(obj, &obj->unions) ||
665 __symbols_patch(obj, &obj->typedefs) ||
666 __symbols_patch(obj, &obj->funcs) ||
667 __symbols_patch(obj, &obj->sets))
668 return -1;
669
670 if (sets_patch(obj))
671 return -1;
672
673 /* Set type to ensure endian translation occurs. */
674 obj->efile.idlist->d_type = ELF_T_WORD;
675
676 elf_flagdata(obj->efile.idlist, ELF_C_SET, ELF_F_DIRTY);
677
678 err = elf_update(obj->efile.elf, ELF_C_WRITE);
679 if (err < 0) {
680 pr_err("FAILED elf_update(WRITE): %s\n",
681 elf_errmsg(-1));
682 }
683
684 pr_debug("update %s for %s\n",
685 err >= 0 ? "ok" : "failed", obj->path);
686 return err < 0 ? -1 : 0;
687 }
688
689 static const char * const resolve_btfids_usage[] = {
690 "resolve_btfids [<options>] <ELF object>",
691 NULL
692 };
693
main(int argc,const char ** argv)694 int main(int argc, const char **argv)
695 {
696 struct object obj = {
697 .efile = {
698 .idlist_shndx = -1,
699 .symbols_shndx = -1,
700 },
701 .structs = RB_ROOT,
702 .unions = RB_ROOT,
703 .typedefs = RB_ROOT,
704 .funcs = RB_ROOT,
705 .sets = RB_ROOT,
706 };
707 struct option btfid_options[] = {
708 OPT_INCR('v', "verbose", &verbose,
709 "be more verbose (show errors, etc)"),
710 OPT_STRING(0, "btf", &obj.btf, "BTF data",
711 "BTF data"),
712 OPT_STRING('b', "btf_base", &obj.base_btf_path, "file",
713 "path of file providing base BTF"),
714 OPT_END()
715 };
716 int err = -1;
717
718 argc = parse_options(argc, argv, btfid_options, resolve_btfids_usage,
719 PARSE_OPT_STOP_AT_NON_OPTION);
720 if (argc != 1)
721 usage_with_options(resolve_btfids_usage, btfid_options);
722
723 obj.path = argv[0];
724
725 if (elf_collect(&obj))
726 goto out;
727
728 /*
729 * We did not find .BTF_ids section or symbols section,
730 * nothing to do..
731 */
732 if (obj.efile.idlist_shndx == -1 ||
733 obj.efile.symbols_shndx == -1) {
734 pr_debug("Cannot find .BTF_ids or symbols sections, nothing to do\n");
735 err = 0;
736 goto out;
737 }
738
739 if (symbols_collect(&obj))
740 goto out;
741
742 if (symbols_resolve(&obj))
743 goto out;
744
745 if (symbols_patch(&obj))
746 goto out;
747
748 err = 0;
749 out:
750 if (obj.efile.elf) {
751 elf_end(obj.efile.elf);
752 close(obj.efile.fd);
753 }
754 return err;
755 }
756