1 // SPDX-License-Identifier: GPL-2.0
2 #include <dirent.h>
3 #include <errno.h>
4 #include <stdlib.h>
5 #include <stdio.h>
6 #include <string.h>
7 #include <linux/capability.h>
8 #include <linux/kernel.h>
9 #include <linux/mman.h>
10 #include <linux/string.h>
11 #include <linux/time64.h>
12 #include <sys/types.h>
13 #include <sys/stat.h>
14 #include <sys/param.h>
15 #include <fcntl.h>
16 #include <unistd.h>
17 #include <inttypes.h>
18 #include "annotate.h"
19 #include "build-id.h"
20 #include "cap.h"
21 #include "dso.h"
22 #include "util.h" // lsdir()
23 #include "debug.h"
24 #include "event.h"
25 #include "machine.h"
26 #include "map.h"
27 #include "symbol.h"
28 #include "map_symbol.h"
29 #include "mem-events.h"
30 #include "mem-info.h"
31 #include "symsrc.h"
32 #include "strlist.h"
33 #include "intlist.h"
34 #include "namespaces.h"
35 #include "header.h"
36 #include "path.h"
37 #include <linux/ctype.h>
38 #include <linux/zalloc.h>
39
40 #include <elf.h>
41 #include <limits.h>
42 #include <symbol/kallsyms.h>
43 #include <sys/utsname.h>
44
45 static int dso__load_kernel_sym(struct dso *dso, struct map *map);
46 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map);
47 static bool symbol__is_idle(const char *name);
48
49 int vmlinux_path__nr_entries;
50 char **vmlinux_path;
51
52 struct symbol_conf symbol_conf = {
53 .nanosecs = false,
54 .use_modules = true,
55 .try_vmlinux_path = true,
56 .demangle = true,
57 .demangle_kernel = false,
58 .cumulate_callchain = true,
59 .time_quantum = 100 * NSEC_PER_MSEC, /* 100ms */
60 .show_hist_headers = true,
61 .symfs = "",
62 .event_group = true,
63 .inline_name = true,
64 .res_sample = 0,
65 };
66
67 struct map_list_node {
68 struct list_head node;
69 struct map *map;
70 };
71
map_list_node__new(void)72 static struct map_list_node *map_list_node__new(void)
73 {
74 return malloc(sizeof(struct map_list_node));
75 }
76
77 static enum dso_binary_type binary_type_symtab[] = {
78 DSO_BINARY_TYPE__KALLSYMS,
79 DSO_BINARY_TYPE__GUEST_KALLSYMS,
80 DSO_BINARY_TYPE__JAVA_JIT,
81 DSO_BINARY_TYPE__DEBUGLINK,
82 DSO_BINARY_TYPE__BUILD_ID_CACHE,
83 DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
84 DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
85 DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
86 DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
87 DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
88 DSO_BINARY_TYPE__GUEST_KMODULE,
89 DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
90 DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
91 DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
92 DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
93 DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
94 DSO_BINARY_TYPE__NOT_FOUND,
95 };
96
97 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab)
98
symbol_type__filter(char symbol_type)99 static bool symbol_type__filter(char symbol_type)
100 {
101 symbol_type = toupper(symbol_type);
102 return symbol_type == 'T' || symbol_type == 'W' || symbol_type == 'D' || symbol_type == 'B';
103 }
104
prefix_underscores_count(const char * str)105 static int prefix_underscores_count(const char *str)
106 {
107 const char *tail = str;
108
109 while (*tail == '_')
110 tail++;
111
112 return tail - str;
113 }
114
arch__normalize_symbol_name(const char * name)115 const char * __weak arch__normalize_symbol_name(const char *name)
116 {
117 return name;
118 }
119
arch__compare_symbol_names(const char * namea,const char * nameb)120 int __weak arch__compare_symbol_names(const char *namea, const char *nameb)
121 {
122 return strcmp(namea, nameb);
123 }
124
arch__compare_symbol_names_n(const char * namea,const char * nameb,unsigned int n)125 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb,
126 unsigned int n)
127 {
128 return strncmp(namea, nameb, n);
129 }
130
arch__choose_best_symbol(struct symbol * syma,struct symbol * symb __maybe_unused)131 int __weak arch__choose_best_symbol(struct symbol *syma,
132 struct symbol *symb __maybe_unused)
133 {
134 /* Avoid "SyS" kernel syscall aliases */
135 if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3))
136 return SYMBOL_B;
137 if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10))
138 return SYMBOL_B;
139
140 return SYMBOL_A;
141 }
142
choose_best_symbol(struct symbol * syma,struct symbol * symb)143 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
144 {
145 s64 a;
146 s64 b;
147 size_t na, nb;
148
149 /* Prefer a symbol with non zero length */
150 a = syma->end - syma->start;
151 b = symb->end - symb->start;
152 if ((b == 0) && (a > 0))
153 return SYMBOL_A;
154 else if ((a == 0) && (b > 0))
155 return SYMBOL_B;
156
157 /* Prefer a non weak symbol over a weak one */
158 a = syma->binding == STB_WEAK;
159 b = symb->binding == STB_WEAK;
160 if (b && !a)
161 return SYMBOL_A;
162 if (a && !b)
163 return SYMBOL_B;
164
165 /* Prefer a global symbol over a non global one */
166 a = syma->binding == STB_GLOBAL;
167 b = symb->binding == STB_GLOBAL;
168 if (a && !b)
169 return SYMBOL_A;
170 if (b && !a)
171 return SYMBOL_B;
172
173 /* Prefer a symbol with less underscores */
174 a = prefix_underscores_count(syma->name);
175 b = prefix_underscores_count(symb->name);
176 if (b > a)
177 return SYMBOL_A;
178 else if (a > b)
179 return SYMBOL_B;
180
181 /* Choose the symbol with the longest name */
182 na = strlen(syma->name);
183 nb = strlen(symb->name);
184 if (na > nb)
185 return SYMBOL_A;
186 else if (na < nb)
187 return SYMBOL_B;
188
189 return arch__choose_best_symbol(syma, symb);
190 }
191
symbols__fixup_duplicate(struct rb_root_cached * symbols)192 void symbols__fixup_duplicate(struct rb_root_cached *symbols)
193 {
194 struct rb_node *nd;
195 struct symbol *curr, *next;
196
197 if (symbol_conf.allow_aliases)
198 return;
199
200 nd = rb_first_cached(symbols);
201
202 while (nd) {
203 curr = rb_entry(nd, struct symbol, rb_node);
204 again:
205 nd = rb_next(&curr->rb_node);
206 if (!nd)
207 break;
208
209 next = rb_entry(nd, struct symbol, rb_node);
210 if (curr->start != next->start)
211 continue;
212
213 if (choose_best_symbol(curr, next) == SYMBOL_A) {
214 if (next->type == STT_GNU_IFUNC)
215 curr->ifunc_alias = true;
216 rb_erase_cached(&next->rb_node, symbols);
217 symbol__delete(next);
218 goto again;
219 } else {
220 if (curr->type == STT_GNU_IFUNC)
221 next->ifunc_alias = true;
222 nd = rb_next(&curr->rb_node);
223 rb_erase_cached(&curr->rb_node, symbols);
224 symbol__delete(curr);
225 }
226 }
227 }
228
229 /* Update zero-sized symbols using the address of the next symbol */
symbols__fixup_end(struct rb_root_cached * symbols,bool is_kallsyms)230 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms)
231 {
232 struct rb_node *nd, *prevnd = rb_first_cached(symbols);
233 struct symbol *curr, *prev;
234
235 if (prevnd == NULL)
236 return;
237
238 curr = rb_entry(prevnd, struct symbol, rb_node);
239
240 for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
241 prev = curr;
242 curr = rb_entry(nd, struct symbol, rb_node);
243
244 /*
245 * On some architecture kernel text segment start is located at
246 * some low memory address, while modules are located at high
247 * memory addresses (or vice versa). The gap between end of
248 * kernel text segment and beginning of first module's text
249 * segment is very big. Therefore do not fill this gap and do
250 * not assign it to the kernel dso map (kallsyms).
251 *
252 * Also BPF code can be allocated separately from text segments
253 * and modules. So the last entry in a module should not fill
254 * the gap too.
255 *
256 * In kallsyms, it determines module symbols using '[' character
257 * like in:
258 * ffffffffc1937000 T hdmi_driver_init [snd_hda_codec_hdmi]
259 */
260 if (prev->end == prev->start) {
261 const char *prev_mod;
262 const char *curr_mod;
263
264 if (!is_kallsyms) {
265 prev->end = curr->start;
266 continue;
267 }
268
269 prev_mod = strchr(prev->name, '[');
270 curr_mod = strchr(curr->name, '[');
271
272 /* Last kernel/module symbol mapped to end of page */
273 if (!prev_mod != !curr_mod)
274 prev->end = roundup(prev->end + 4096, 4096);
275 /* Last symbol in the previous module */
276 else if (prev_mod && strcmp(prev_mod, curr_mod))
277 prev->end = roundup(prev->end + 4096, 4096);
278 else
279 prev->end = curr->start;
280
281 pr_debug4("%s sym:%s end:%#" PRIx64 "\n",
282 __func__, prev->name, prev->end);
283 }
284 }
285
286 /* Last entry */
287 if (curr->end == curr->start)
288 curr->end = roundup(curr->start, 4096) + 4096;
289 }
290
symbol__new(u64 start,u64 len,u8 binding,u8 type,const char * name)291 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
292 {
293 size_t namelen = strlen(name) + 1;
294 struct symbol *sym = calloc(1, (symbol_conf.priv_size +
295 sizeof(*sym) + namelen));
296 if (sym == NULL)
297 return NULL;
298
299 if (symbol_conf.priv_size) {
300 if (symbol_conf.init_annotation) {
301 struct annotation *notes = (void *)sym;
302 annotation__init(notes);
303 }
304 sym = ((void *)sym) + symbol_conf.priv_size;
305 }
306
307 sym->start = start;
308 sym->end = len ? start + len : start;
309 sym->type = type;
310 sym->binding = binding;
311 sym->namelen = namelen - 1;
312
313 pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
314 __func__, name, start, sym->end);
315 memcpy(sym->name, name, namelen);
316
317 return sym;
318 }
319
symbol__delete(struct symbol * sym)320 void symbol__delete(struct symbol *sym)
321 {
322 if (symbol_conf.priv_size) {
323 if (symbol_conf.init_annotation) {
324 struct annotation *notes = symbol__annotation(sym);
325
326 annotation__exit(notes);
327 }
328 }
329 free(((void *)sym) - symbol_conf.priv_size);
330 }
331
symbols__delete(struct rb_root_cached * symbols)332 void symbols__delete(struct rb_root_cached *symbols)
333 {
334 struct symbol *pos;
335 struct rb_node *next = rb_first_cached(symbols);
336
337 while (next) {
338 pos = rb_entry(next, struct symbol, rb_node);
339 next = rb_next(&pos->rb_node);
340 rb_erase_cached(&pos->rb_node, symbols);
341 symbol__delete(pos);
342 }
343 }
344
__symbols__insert(struct rb_root_cached * symbols,struct symbol * sym,bool kernel)345 void __symbols__insert(struct rb_root_cached *symbols,
346 struct symbol *sym, bool kernel)
347 {
348 struct rb_node **p = &symbols->rb_root.rb_node;
349 struct rb_node *parent = NULL;
350 const u64 ip = sym->start;
351 struct symbol *s;
352 bool leftmost = true;
353
354 if (kernel) {
355 const char *name = sym->name;
356 /*
357 * ppc64 uses function descriptors and appends a '.' to the
358 * start of every instruction address. Remove it.
359 */
360 if (name[0] == '.')
361 name++;
362 sym->idle = symbol__is_idle(name);
363 }
364
365 while (*p != NULL) {
366 parent = *p;
367 s = rb_entry(parent, struct symbol, rb_node);
368 if (ip < s->start)
369 p = &(*p)->rb_left;
370 else {
371 p = &(*p)->rb_right;
372 leftmost = false;
373 }
374 }
375 rb_link_node(&sym->rb_node, parent, p);
376 rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
377 }
378
symbols__insert(struct rb_root_cached * symbols,struct symbol * sym)379 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
380 {
381 __symbols__insert(symbols, sym, false);
382 }
383
symbols__find(struct rb_root_cached * symbols,u64 ip)384 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
385 {
386 struct rb_node *n;
387
388 if (symbols == NULL)
389 return NULL;
390
391 n = symbols->rb_root.rb_node;
392
393 while (n) {
394 struct symbol *s = rb_entry(n, struct symbol, rb_node);
395
396 if (ip < s->start)
397 n = n->rb_left;
398 else if (ip > s->end || (ip == s->end && ip != s->start))
399 n = n->rb_right;
400 else
401 return s;
402 }
403
404 return NULL;
405 }
406
symbols__first(struct rb_root_cached * symbols)407 static struct symbol *symbols__first(struct rb_root_cached *symbols)
408 {
409 struct rb_node *n = rb_first_cached(symbols);
410
411 if (n)
412 return rb_entry(n, struct symbol, rb_node);
413
414 return NULL;
415 }
416
symbols__last(struct rb_root_cached * symbols)417 static struct symbol *symbols__last(struct rb_root_cached *symbols)
418 {
419 struct rb_node *n = rb_last(&symbols->rb_root);
420
421 if (n)
422 return rb_entry(n, struct symbol, rb_node);
423
424 return NULL;
425 }
426
symbols__next(struct symbol * sym)427 static struct symbol *symbols__next(struct symbol *sym)
428 {
429 struct rb_node *n = rb_next(&sym->rb_node);
430
431 if (n)
432 return rb_entry(n, struct symbol, rb_node);
433
434 return NULL;
435 }
436
symbols__sort_name_cmp(const void * vlhs,const void * vrhs)437 static int symbols__sort_name_cmp(const void *vlhs, const void *vrhs)
438 {
439 const struct symbol *lhs = *((const struct symbol **)vlhs);
440 const struct symbol *rhs = *((const struct symbol **)vrhs);
441
442 return strcmp(lhs->name, rhs->name);
443 }
444
symbols__sort_by_name(struct rb_root_cached * source,size_t * len)445 static struct symbol **symbols__sort_by_name(struct rb_root_cached *source, size_t *len)
446 {
447 struct rb_node *nd;
448 struct symbol **result;
449 size_t i = 0, size = 0;
450
451 for (nd = rb_first_cached(source); nd; nd = rb_next(nd))
452 size++;
453
454 result = malloc(sizeof(*result) * size);
455 if (!result)
456 return NULL;
457
458 for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
459 struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
460
461 result[i++] = pos;
462 }
463 qsort(result, size, sizeof(*result), symbols__sort_name_cmp);
464 *len = size;
465 return result;
466 }
467
symbol__match_symbol_name(const char * name,const char * str,enum symbol_tag_include includes)468 int symbol__match_symbol_name(const char *name, const char *str,
469 enum symbol_tag_include includes)
470 {
471 const char *versioning;
472
473 if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
474 (versioning = strstr(name, "@@"))) {
475 int len = strlen(str);
476
477 if (len < versioning - name)
478 len = versioning - name;
479
480 return arch__compare_symbol_names_n(name, str, len);
481 } else
482 return arch__compare_symbol_names(name, str);
483 }
484
symbols__find_by_name(struct symbol * symbols[],size_t symbols_len,const char * name,enum symbol_tag_include includes,size_t * found_idx)485 static struct symbol *symbols__find_by_name(struct symbol *symbols[],
486 size_t symbols_len,
487 const char *name,
488 enum symbol_tag_include includes,
489 size_t *found_idx)
490 {
491 size_t i, lower = 0, upper = symbols_len;
492 struct symbol *s = NULL;
493
494 if (found_idx)
495 *found_idx = SIZE_MAX;
496
497 if (!symbols_len)
498 return NULL;
499
500 while (lower < upper) {
501 int cmp;
502
503 i = (lower + upper) / 2;
504 cmp = symbol__match_symbol_name(symbols[i]->name, name, includes);
505
506 if (cmp > 0)
507 upper = i;
508 else if (cmp < 0)
509 lower = i + 1;
510 else {
511 if (found_idx)
512 *found_idx = i;
513 s = symbols[i];
514 break;
515 }
516 }
517 if (s && includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY) {
518 /* return first symbol that has same name (if any) */
519 for (; i > 0; i--) {
520 struct symbol *tmp = symbols[i - 1];
521
522 if (!arch__compare_symbol_names(tmp->name, s->name)) {
523 if (found_idx)
524 *found_idx = i - 1;
525 s = tmp;
526 } else
527 break;
528 }
529 }
530 assert(!found_idx || !s || s == symbols[*found_idx]);
531 return s;
532 }
533
dso__reset_find_symbol_cache(struct dso * dso)534 void dso__reset_find_symbol_cache(struct dso *dso)
535 {
536 dso__set_last_find_result_addr(dso, 0);
537 dso__set_last_find_result_symbol(dso, NULL);
538 }
539
dso__insert_symbol(struct dso * dso,struct symbol * sym)540 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
541 {
542 __symbols__insert(dso__symbols(dso), sym, dso__kernel(dso));
543
544 /* update the symbol cache if necessary */
545 if (dso__last_find_result_addr(dso) >= sym->start &&
546 (dso__last_find_result_addr(dso) < sym->end ||
547 sym->start == sym->end)) {
548 dso__set_last_find_result_symbol(dso, sym);
549 }
550 }
551
dso__delete_symbol(struct dso * dso,struct symbol * sym)552 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
553 {
554 rb_erase_cached(&sym->rb_node, dso__symbols(dso));
555 symbol__delete(sym);
556 dso__reset_find_symbol_cache(dso);
557 }
558
dso__find_symbol(struct dso * dso,u64 addr)559 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
560 {
561 if (dso__last_find_result_addr(dso) != addr || dso__last_find_result_symbol(dso) == NULL) {
562 dso__set_last_find_result_addr(dso, addr);
563 dso__set_last_find_result_symbol(dso, symbols__find(dso__symbols(dso), addr));
564 }
565
566 return dso__last_find_result_symbol(dso);
567 }
568
dso__find_symbol_nocache(struct dso * dso,u64 addr)569 struct symbol *dso__find_symbol_nocache(struct dso *dso, u64 addr)
570 {
571 return symbols__find(dso__symbols(dso), addr);
572 }
573
dso__first_symbol(struct dso * dso)574 struct symbol *dso__first_symbol(struct dso *dso)
575 {
576 return symbols__first(dso__symbols(dso));
577 }
578
dso__last_symbol(struct dso * dso)579 struct symbol *dso__last_symbol(struct dso *dso)
580 {
581 return symbols__last(dso__symbols(dso));
582 }
583
dso__next_symbol(struct symbol * sym)584 struct symbol *dso__next_symbol(struct symbol *sym)
585 {
586 return symbols__next(sym);
587 }
588
dso__next_symbol_by_name(struct dso * dso,size_t * idx)589 struct symbol *dso__next_symbol_by_name(struct dso *dso, size_t *idx)
590 {
591 if (*idx + 1 >= dso__symbol_names_len(dso))
592 return NULL;
593
594 ++*idx;
595 return dso__symbol_names(dso)[*idx];
596 }
597
598 /*
599 * Returns first symbol that matched with @name.
600 */
dso__find_symbol_by_name(struct dso * dso,const char * name,size_t * idx)601 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name, size_t *idx)
602 {
603 struct symbol *s = symbols__find_by_name(dso__symbol_names(dso),
604 dso__symbol_names_len(dso),
605 name, SYMBOL_TAG_INCLUDE__NONE, idx);
606 if (!s) {
607 s = symbols__find_by_name(dso__symbol_names(dso), dso__symbol_names_len(dso),
608 name, SYMBOL_TAG_INCLUDE__DEFAULT_ONLY, idx);
609 }
610 return s;
611 }
612
dso__sort_by_name(struct dso * dso)613 void dso__sort_by_name(struct dso *dso)
614 {
615 mutex_lock(dso__lock(dso));
616 if (!dso__sorted_by_name(dso)) {
617 size_t len;
618
619 dso__set_symbol_names(dso, symbols__sort_by_name(dso__symbols(dso), &len));
620 if (dso__symbol_names(dso)) {
621 dso__set_symbol_names_len(dso, len);
622 dso__set_sorted_by_name(dso);
623 }
624 }
625 mutex_unlock(dso__lock(dso));
626 }
627
628 /*
629 * While we find nice hex chars, build a long_val.
630 * Return number of chars processed.
631 */
hex2u64(const char * ptr,u64 * long_val)632 static int hex2u64(const char *ptr, u64 *long_val)
633 {
634 char *p;
635
636 *long_val = strtoull(ptr, &p, 16);
637
638 return p - ptr;
639 }
640
641
modules__parse(const char * filename,void * arg,int (* process_module)(void * arg,const char * name,u64 start,u64 size))642 int modules__parse(const char *filename, void *arg,
643 int (*process_module)(void *arg, const char *name,
644 u64 start, u64 size))
645 {
646 char *line = NULL;
647 size_t n;
648 FILE *file;
649 int err = 0;
650
651 file = fopen(filename, "r");
652 if (file == NULL)
653 return -1;
654
655 while (1) {
656 char name[PATH_MAX];
657 u64 start, size;
658 char *sep, *endptr;
659 ssize_t line_len;
660
661 line_len = getline(&line, &n, file);
662 if (line_len < 0) {
663 if (feof(file))
664 break;
665 err = -1;
666 goto out;
667 }
668
669 if (!line) {
670 err = -1;
671 goto out;
672 }
673
674 line[--line_len] = '\0'; /* \n */
675
676 sep = strrchr(line, 'x');
677 if (sep == NULL)
678 continue;
679
680 hex2u64(sep + 1, &start);
681
682 sep = strchr(line, ' ');
683 if (sep == NULL)
684 continue;
685
686 *sep = '\0';
687
688 scnprintf(name, sizeof(name), "[%s]", line);
689
690 size = strtoul(sep + 1, &endptr, 0);
691 if (*endptr != ' ' && *endptr != '\t')
692 continue;
693
694 err = process_module(arg, name, start, size);
695 if (err)
696 break;
697 }
698 out:
699 free(line);
700 fclose(file);
701 return err;
702 }
703
704 /*
705 * These are symbols in the kernel image, so make sure that
706 * sym is from a kernel DSO.
707 */
symbol__is_idle(const char * name)708 static bool symbol__is_idle(const char *name)
709 {
710 const char * const idle_symbols[] = {
711 "acpi_idle_do_entry",
712 "acpi_processor_ffh_cstate_enter",
713 "arch_cpu_idle",
714 "cpu_idle",
715 "cpu_startup_entry",
716 "idle_cpu",
717 "intel_idle",
718 "intel_idle_ibrs",
719 "default_idle",
720 "native_safe_halt",
721 "enter_idle",
722 "exit_idle",
723 "mwait_idle",
724 "mwait_idle_with_hints",
725 "mwait_idle_with_hints.constprop.0",
726 "poll_idle",
727 "ppc64_runlatch_off",
728 "pseries_dedicated_idle_sleep",
729 "psw_idle",
730 "psw_idle_exit",
731 NULL
732 };
733 int i;
734 static struct strlist *idle_symbols_list;
735
736 if (idle_symbols_list)
737 return strlist__has_entry(idle_symbols_list, name);
738
739 idle_symbols_list = strlist__new(NULL, NULL);
740
741 for (i = 0; idle_symbols[i]; i++)
742 strlist__add(idle_symbols_list, idle_symbols[i]);
743
744 return strlist__has_entry(idle_symbols_list, name);
745 }
746
map__process_kallsym_symbol(void * arg,const char * name,char type,u64 start)747 static int map__process_kallsym_symbol(void *arg, const char *name,
748 char type, u64 start)
749 {
750 struct symbol *sym;
751 struct dso *dso = arg;
752 struct rb_root_cached *root = dso__symbols(dso);
753
754 if (!symbol_type__filter(type))
755 return 0;
756
757 /* Ignore local symbols for ARM modules */
758 if (name[0] == '$')
759 return 0;
760
761 /*
762 * module symbols are not sorted so we add all
763 * symbols, setting length to 0, and rely on
764 * symbols__fixup_end() to fix it up.
765 */
766 sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
767 if (sym == NULL)
768 return -ENOMEM;
769 /*
770 * We will pass the symbols to the filter later, in
771 * map__split_kallsyms, when we have split the maps per module
772 */
773 __symbols__insert(root, sym, !strchr(name, '['));
774
775 return 0;
776 }
777
778 /*
779 * Loads the function entries in /proc/kallsyms into kernel_map->dso,
780 * so that we can in the next step set the symbol ->end address and then
781 * call kernel_maps__split_kallsyms.
782 */
dso__load_all_kallsyms(struct dso * dso,const char * filename)783 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
784 {
785 return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
786 }
787
maps__split_kallsyms_for_kcore(struct maps * kmaps,struct dso * dso)788 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
789 {
790 struct symbol *pos;
791 int count = 0;
792 struct rb_root_cached *root = dso__symbols(dso);
793 struct rb_root_cached old_root = *root;
794 struct rb_node *next = rb_first_cached(root);
795
796 if (!kmaps)
797 return -1;
798
799 *root = RB_ROOT_CACHED;
800
801 while (next) {
802 struct map *curr_map;
803 struct dso *curr_map_dso;
804 char *module;
805
806 pos = rb_entry(next, struct symbol, rb_node);
807 next = rb_next(&pos->rb_node);
808
809 rb_erase_cached(&pos->rb_node, &old_root);
810 RB_CLEAR_NODE(&pos->rb_node);
811 module = strchr(pos->name, '\t');
812 if (module)
813 *module = '\0';
814
815 curr_map = maps__find(kmaps, pos->start);
816
817 if (!curr_map) {
818 symbol__delete(pos);
819 continue;
820 }
821 curr_map_dso = map__dso(curr_map);
822 pos->start -= map__start(curr_map) - map__pgoff(curr_map);
823 if (pos->end > map__end(curr_map))
824 pos->end = map__end(curr_map);
825 if (pos->end)
826 pos->end -= map__start(curr_map) - map__pgoff(curr_map);
827 symbols__insert(dso__symbols(curr_map_dso), pos);
828 ++count;
829 map__put(curr_map);
830 }
831
832 /* Symbols have been adjusted */
833 dso__set_adjust_symbols(dso, true);
834
835 return count;
836 }
837
838 /*
839 * Split the symbols into maps, making sure there are no overlaps, i.e. the
840 * kernel range is broken in several maps, named [kernel].N, as we don't have
841 * the original ELF section names vmlinux have.
842 */
maps__split_kallsyms(struct maps * kmaps,struct dso * dso,u64 delta,struct map * initial_map)843 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
844 struct map *initial_map)
845 {
846 struct machine *machine;
847 struct map *curr_map = map__get(initial_map);
848 struct symbol *pos;
849 int count = 0, moved = 0;
850 struct rb_root_cached *root = dso__symbols(dso);
851 struct rb_node *next = rb_first_cached(root);
852 int kernel_range = 0;
853 bool x86_64;
854
855 if (!kmaps)
856 return -1;
857
858 machine = maps__machine(kmaps);
859
860 x86_64 = machine__is(machine, "x86_64");
861
862 while (next) {
863 char *module;
864
865 pos = rb_entry(next, struct symbol, rb_node);
866 next = rb_next(&pos->rb_node);
867
868 module = strchr(pos->name, '\t');
869 if (module) {
870 struct dso *curr_map_dso;
871
872 if (!symbol_conf.use_modules)
873 goto discard_symbol;
874
875 *module++ = '\0';
876 curr_map_dso = map__dso(curr_map);
877 if (strcmp(dso__short_name(curr_map_dso), module)) {
878 if (!RC_CHK_EQUAL(curr_map, initial_map) &&
879 dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST &&
880 machine__is_default_guest(machine)) {
881 /*
882 * We assume all symbols of a module are
883 * continuous in * kallsyms, so curr_map
884 * points to a module and all its
885 * symbols are in its kmap. Mark it as
886 * loaded.
887 */
888 dso__set_loaded(curr_map_dso);
889 }
890
891 map__zput(curr_map);
892 curr_map = maps__find_by_name(kmaps, module);
893 if (curr_map == NULL) {
894 pr_debug("%s/proc/{kallsyms,modules} "
895 "inconsistency while looking "
896 "for \"%s\" module!\n",
897 machine->root_dir, module);
898 curr_map = map__get(initial_map);
899 goto discard_symbol;
900 }
901 curr_map_dso = map__dso(curr_map);
902 if (dso__loaded(curr_map_dso) &&
903 !machine__is_default_guest(machine))
904 goto discard_symbol;
905 }
906 /*
907 * So that we look just like we get from .ko files,
908 * i.e. not prelinked, relative to initial_map->start.
909 */
910 pos->start = map__map_ip(curr_map, pos->start);
911 pos->end = map__map_ip(curr_map, pos->end);
912 } else if (x86_64 && is_entry_trampoline(pos->name)) {
913 /*
914 * These symbols are not needed anymore since the
915 * trampoline maps refer to the text section and it's
916 * symbols instead. Avoid having to deal with
917 * relocations, and the assumption that the first symbol
918 * is the start of kernel text, by simply removing the
919 * symbols at this point.
920 */
921 goto discard_symbol;
922 } else if (!RC_CHK_EQUAL(curr_map, initial_map)) {
923 char dso_name[PATH_MAX];
924 struct dso *ndso;
925
926 if (delta) {
927 /* Kernel was relocated at boot time */
928 pos->start -= delta;
929 pos->end -= delta;
930 }
931
932 if (count == 0) {
933 map__zput(curr_map);
934 curr_map = map__get(initial_map);
935 goto add_symbol;
936 }
937
938 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
939 snprintf(dso_name, sizeof(dso_name),
940 "[guest.kernel].%d",
941 kernel_range++);
942 else
943 snprintf(dso_name, sizeof(dso_name),
944 "[kernel].%d",
945 kernel_range++);
946
947 ndso = dso__new(dso_name);
948 map__zput(curr_map);
949 if (ndso == NULL)
950 return -1;
951
952 dso__set_kernel(ndso, dso__kernel(dso));
953
954 curr_map = map__new2(pos->start, ndso);
955 if (curr_map == NULL) {
956 dso__put(ndso);
957 return -1;
958 }
959
960 map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY);
961 if (maps__insert(kmaps, curr_map)) {
962 map__zput(curr_map);
963 dso__put(ndso);
964 return -1;
965 }
966 ++kernel_range;
967 } else if (delta) {
968 /* Kernel was relocated at boot time */
969 pos->start -= delta;
970 pos->end -= delta;
971 }
972 add_symbol:
973 if (!RC_CHK_EQUAL(curr_map, initial_map)) {
974 struct dso *curr_map_dso = map__dso(curr_map);
975
976 rb_erase_cached(&pos->rb_node, root);
977 symbols__insert(dso__symbols(curr_map_dso), pos);
978 ++moved;
979 } else
980 ++count;
981
982 continue;
983 discard_symbol:
984 rb_erase_cached(&pos->rb_node, root);
985 symbol__delete(pos);
986 }
987
988 if (!RC_CHK_EQUAL(curr_map, initial_map) &&
989 dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST &&
990 machine__is_default_guest(maps__machine(kmaps))) {
991 dso__set_loaded(map__dso(curr_map));
992 }
993 map__put(curr_map);
994 return count + moved;
995 }
996
symbol__restricted_filename(const char * filename,const char * restricted_filename)997 bool symbol__restricted_filename(const char *filename,
998 const char *restricted_filename)
999 {
1000 bool restricted = false;
1001
1002 if (symbol_conf.kptr_restrict) {
1003 char *r = realpath(filename, NULL);
1004
1005 if (r != NULL) {
1006 restricted = strcmp(r, restricted_filename) == 0;
1007 free(r);
1008 return restricted;
1009 }
1010 }
1011
1012 return restricted;
1013 }
1014
1015 struct module_info {
1016 struct rb_node rb_node;
1017 char *name;
1018 u64 start;
1019 };
1020
add_module(struct module_info * mi,struct rb_root * modules)1021 static void add_module(struct module_info *mi, struct rb_root *modules)
1022 {
1023 struct rb_node **p = &modules->rb_node;
1024 struct rb_node *parent = NULL;
1025 struct module_info *m;
1026
1027 while (*p != NULL) {
1028 parent = *p;
1029 m = rb_entry(parent, struct module_info, rb_node);
1030 if (strcmp(mi->name, m->name) < 0)
1031 p = &(*p)->rb_left;
1032 else
1033 p = &(*p)->rb_right;
1034 }
1035 rb_link_node(&mi->rb_node, parent, p);
1036 rb_insert_color(&mi->rb_node, modules);
1037 }
1038
delete_modules(struct rb_root * modules)1039 static void delete_modules(struct rb_root *modules)
1040 {
1041 struct module_info *mi;
1042 struct rb_node *next = rb_first(modules);
1043
1044 while (next) {
1045 mi = rb_entry(next, struct module_info, rb_node);
1046 next = rb_next(&mi->rb_node);
1047 rb_erase(&mi->rb_node, modules);
1048 zfree(&mi->name);
1049 free(mi);
1050 }
1051 }
1052
find_module(const char * name,struct rb_root * modules)1053 static struct module_info *find_module(const char *name,
1054 struct rb_root *modules)
1055 {
1056 struct rb_node *n = modules->rb_node;
1057
1058 while (n) {
1059 struct module_info *m;
1060 int cmp;
1061
1062 m = rb_entry(n, struct module_info, rb_node);
1063 cmp = strcmp(name, m->name);
1064 if (cmp < 0)
1065 n = n->rb_left;
1066 else if (cmp > 0)
1067 n = n->rb_right;
1068 else
1069 return m;
1070 }
1071
1072 return NULL;
1073 }
1074
__read_proc_modules(void * arg,const char * name,u64 start,u64 size __maybe_unused)1075 static int __read_proc_modules(void *arg, const char *name, u64 start,
1076 u64 size __maybe_unused)
1077 {
1078 struct rb_root *modules = arg;
1079 struct module_info *mi;
1080
1081 mi = zalloc(sizeof(struct module_info));
1082 if (!mi)
1083 return -ENOMEM;
1084
1085 mi->name = strdup(name);
1086 mi->start = start;
1087
1088 if (!mi->name) {
1089 free(mi);
1090 return -ENOMEM;
1091 }
1092
1093 add_module(mi, modules);
1094
1095 return 0;
1096 }
1097
read_proc_modules(const char * filename,struct rb_root * modules)1098 static int read_proc_modules(const char *filename, struct rb_root *modules)
1099 {
1100 if (symbol__restricted_filename(filename, "/proc/modules"))
1101 return -1;
1102
1103 if (modules__parse(filename, modules, __read_proc_modules)) {
1104 delete_modules(modules);
1105 return -1;
1106 }
1107
1108 return 0;
1109 }
1110
compare_proc_modules(const char * from,const char * to)1111 int compare_proc_modules(const char *from, const char *to)
1112 {
1113 struct rb_root from_modules = RB_ROOT;
1114 struct rb_root to_modules = RB_ROOT;
1115 struct rb_node *from_node, *to_node;
1116 struct module_info *from_m, *to_m;
1117 int ret = -1;
1118
1119 if (read_proc_modules(from, &from_modules))
1120 return -1;
1121
1122 if (read_proc_modules(to, &to_modules))
1123 goto out_delete_from;
1124
1125 from_node = rb_first(&from_modules);
1126 to_node = rb_first(&to_modules);
1127 while (from_node) {
1128 if (!to_node)
1129 break;
1130
1131 from_m = rb_entry(from_node, struct module_info, rb_node);
1132 to_m = rb_entry(to_node, struct module_info, rb_node);
1133
1134 if (from_m->start != to_m->start ||
1135 strcmp(from_m->name, to_m->name))
1136 break;
1137
1138 from_node = rb_next(from_node);
1139 to_node = rb_next(to_node);
1140 }
1141
1142 if (!from_node && !to_node)
1143 ret = 0;
1144
1145 delete_modules(&to_modules);
1146 out_delete_from:
1147 delete_modules(&from_modules);
1148
1149 return ret;
1150 }
1151
do_validate_kcore_modules_cb(struct map * old_map,void * data)1152 static int do_validate_kcore_modules_cb(struct map *old_map, void *data)
1153 {
1154 struct rb_root *modules = data;
1155 struct module_info *mi;
1156 struct dso *dso;
1157
1158 if (!__map__is_kmodule(old_map))
1159 return 0;
1160
1161 dso = map__dso(old_map);
1162 /* Module must be in memory at the same address */
1163 mi = find_module(dso__short_name(dso), modules);
1164 if (!mi || mi->start != map__start(old_map))
1165 return -EINVAL;
1166
1167 return 0;
1168 }
1169
do_validate_kcore_modules(const char * filename,struct maps * kmaps)1170 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1171 {
1172 struct rb_root modules = RB_ROOT;
1173 int err;
1174
1175 err = read_proc_modules(filename, &modules);
1176 if (err)
1177 return err;
1178
1179 err = maps__for_each_map(kmaps, do_validate_kcore_modules_cb, &modules);
1180
1181 delete_modules(&modules);
1182 return err;
1183 }
1184
1185 /*
1186 * If kallsyms is referenced by name then we look for filename in the same
1187 * directory.
1188 */
filename_from_kallsyms_filename(char * filename,const char * base_name,const char * kallsyms_filename)1189 static bool filename_from_kallsyms_filename(char *filename,
1190 const char *base_name,
1191 const char *kallsyms_filename)
1192 {
1193 char *name;
1194
1195 strcpy(filename, kallsyms_filename);
1196 name = strrchr(filename, '/');
1197 if (!name)
1198 return false;
1199
1200 name += 1;
1201
1202 if (!strcmp(name, "kallsyms")) {
1203 strcpy(name, base_name);
1204 return true;
1205 }
1206
1207 return false;
1208 }
1209
validate_kcore_modules(const char * kallsyms_filename,struct map * map)1210 static int validate_kcore_modules(const char *kallsyms_filename,
1211 struct map *map)
1212 {
1213 struct maps *kmaps = map__kmaps(map);
1214 char modules_filename[PATH_MAX];
1215
1216 if (!kmaps)
1217 return -EINVAL;
1218
1219 if (!filename_from_kallsyms_filename(modules_filename, "modules",
1220 kallsyms_filename))
1221 return -EINVAL;
1222
1223 if (do_validate_kcore_modules(modules_filename, kmaps))
1224 return -EINVAL;
1225
1226 return 0;
1227 }
1228
validate_kcore_addresses(const char * kallsyms_filename,struct map * map)1229 static int validate_kcore_addresses(const char *kallsyms_filename,
1230 struct map *map)
1231 {
1232 struct kmap *kmap = map__kmap(map);
1233
1234 if (!kmap)
1235 return -EINVAL;
1236
1237 if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) {
1238 u64 start;
1239
1240 if (kallsyms__get_function_start(kallsyms_filename,
1241 kmap->ref_reloc_sym->name, &start))
1242 return -ENOENT;
1243 if (start != kmap->ref_reloc_sym->addr)
1244 return -EINVAL;
1245 }
1246
1247 return validate_kcore_modules(kallsyms_filename, map);
1248 }
1249
1250 struct kcore_mapfn_data {
1251 struct dso *dso;
1252 struct list_head maps;
1253 };
1254
kcore_mapfn(u64 start,u64 len,u64 pgoff,void * data)1255 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
1256 {
1257 struct kcore_mapfn_data *md = data;
1258 struct map_list_node *list_node = map_list_node__new();
1259
1260 if (!list_node)
1261 return -ENOMEM;
1262
1263 list_node->map = map__new2(start, md->dso);
1264 if (!list_node->map) {
1265 free(list_node);
1266 return -ENOMEM;
1267 }
1268
1269 map__set_end(list_node->map, map__start(list_node->map) + len);
1270 map__set_pgoff(list_node->map, pgoff);
1271
1272 list_add(&list_node->node, &md->maps);
1273
1274 return 0;
1275 }
1276
remove_old_maps(struct map * map,void * data)1277 static bool remove_old_maps(struct map *map, void *data)
1278 {
1279 const struct map *map_to_save = data;
1280
1281 /*
1282 * We need to preserve eBPF maps even if they are covered by kcore,
1283 * because we need to access eBPF dso for source data.
1284 */
1285 return !RC_CHK_EQUAL(map, map_to_save) && !__map__is_bpf_prog(map);
1286 }
1287
dso__load_kcore(struct dso * dso,struct map * map,const char * kallsyms_filename)1288 static int dso__load_kcore(struct dso *dso, struct map *map,
1289 const char *kallsyms_filename)
1290 {
1291 struct maps *kmaps = map__kmaps(map);
1292 struct kcore_mapfn_data md;
1293 struct map *map_ref, *replacement_map = NULL;
1294 struct machine *machine;
1295 bool is_64_bit;
1296 int err, fd;
1297 char kcore_filename[PATH_MAX];
1298 u64 stext;
1299
1300 if (!kmaps)
1301 return -EINVAL;
1302
1303 machine = maps__machine(kmaps);
1304
1305 /* This function requires that the map is the kernel map */
1306 if (!__map__is_kernel(map))
1307 return -EINVAL;
1308
1309 if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1310 kallsyms_filename))
1311 return -EINVAL;
1312
1313 /* Modules and kernel must be present at their original addresses */
1314 if (validate_kcore_addresses(kallsyms_filename, map))
1315 return -EINVAL;
1316
1317 md.dso = dso;
1318 INIT_LIST_HEAD(&md.maps);
1319
1320 fd = open(kcore_filename, O_RDONLY);
1321 if (fd < 0) {
1322 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1323 kcore_filename);
1324 return -EINVAL;
1325 }
1326
1327 /* Read new maps into temporary lists */
1328 err = file__read_maps(fd, map__prot(map) & PROT_EXEC, kcore_mapfn, &md,
1329 &is_64_bit);
1330 if (err)
1331 goto out_err;
1332 dso__set_is_64_bit(dso, is_64_bit);
1333
1334 if (list_empty(&md.maps)) {
1335 err = -EINVAL;
1336 goto out_err;
1337 }
1338
1339 /* Remove old maps */
1340 maps__remove_maps(kmaps, remove_old_maps, map);
1341 machine->trampolines_mapped = false;
1342
1343 /* Find the kernel map using the '_stext' symbol */
1344 if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1345 u64 replacement_size = 0;
1346 struct map_list_node *new_node;
1347
1348 list_for_each_entry(new_node, &md.maps, node) {
1349 struct map *new_map = new_node->map;
1350 u64 new_size = map__size(new_map);
1351
1352 if (!(stext >= map__start(new_map) && stext < map__end(new_map)))
1353 continue;
1354
1355 /*
1356 * On some architectures, ARM64 for example, the kernel
1357 * text can get allocated inside of the vmalloc segment.
1358 * Select the smallest matching segment, in case stext
1359 * falls within more than one in the list.
1360 */
1361 if (!replacement_map || new_size < replacement_size) {
1362 replacement_map = new_map;
1363 replacement_size = new_size;
1364 }
1365 }
1366 }
1367
1368 if (!replacement_map)
1369 replacement_map = list_entry(md.maps.next, struct map_list_node, node)->map;
1370
1371 /*
1372 * Update addresses of vmlinux map. Re-insert it to ensure maps are
1373 * correctly ordered. Do this before using maps__merge_in() for the
1374 * remaining maps so vmlinux gets split if necessary.
1375 */
1376 map_ref = map__get(map);
1377 maps__remove(kmaps, map_ref);
1378
1379 map__set_start(map_ref, map__start(replacement_map));
1380 map__set_end(map_ref, map__end(replacement_map));
1381 map__set_pgoff(map_ref, map__pgoff(replacement_map));
1382 map__set_mapping_type(map_ref, map__mapping_type(replacement_map));
1383
1384 err = maps__insert(kmaps, map_ref);
1385 map__put(map_ref);
1386 if (err)
1387 goto out_err;
1388
1389 /* Add new maps */
1390 while (!list_empty(&md.maps)) {
1391 struct map_list_node *new_node = list_entry(md.maps.next, struct map_list_node, node);
1392 struct map *new_map = new_node->map;
1393
1394 list_del_init(&new_node->node);
1395
1396 /* skip if replacement_map, already inserted above */
1397 if (!RC_CHK_EQUAL(new_map, replacement_map)) {
1398 /*
1399 * Merge kcore map into existing maps,
1400 * and ensure that current maps (eBPF)
1401 * stay intact.
1402 */
1403 if (maps__merge_in(kmaps, new_map)) {
1404 err = -EINVAL;
1405 goto out_err;
1406 }
1407 }
1408 map__zput(new_node->map);
1409 free(new_node);
1410 }
1411
1412 if (machine__is(machine, "x86_64")) {
1413 u64 addr;
1414
1415 /*
1416 * If one of the corresponding symbols is there, assume the
1417 * entry trampoline maps are too.
1418 */
1419 if (!kallsyms__get_function_start(kallsyms_filename,
1420 ENTRY_TRAMPOLINE_NAME,
1421 &addr))
1422 machine->trampolines_mapped = true;
1423 }
1424
1425 /*
1426 * Set the data type and long name so that kcore can be read via
1427 * dso__data_read_addr().
1428 */
1429 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1430 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KCORE);
1431 else
1432 dso__set_binary_type(dso, DSO_BINARY_TYPE__KCORE);
1433 dso__set_long_name(dso, strdup(kcore_filename), true);
1434
1435 close(fd);
1436
1437 if (map__prot(map) & PROT_EXEC)
1438 pr_debug("Using %s for kernel object code\n", kcore_filename);
1439 else
1440 pr_debug("Using %s for kernel data\n", kcore_filename);
1441
1442 return 0;
1443
1444 out_err:
1445 while (!list_empty(&md.maps)) {
1446 struct map_list_node *list_node;
1447
1448 list_node = list_entry(md.maps.next, struct map_list_node, node);
1449 list_del_init(&list_node->node);
1450 map__zput(list_node->map);
1451 free(list_node);
1452 }
1453 close(fd);
1454 return err;
1455 }
1456
1457 /*
1458 * If the kernel is relocated at boot time, kallsyms won't match. Compute the
1459 * delta based on the relocation reference symbol.
1460 */
kallsyms__delta(struct kmap * kmap,const char * filename,u64 * delta)1461 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1462 {
1463 u64 addr;
1464
1465 if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1466 return 0;
1467
1468 if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1469 return -1;
1470
1471 *delta = addr - kmap->ref_reloc_sym->addr;
1472 return 0;
1473 }
1474
__dso__load_kallsyms(struct dso * dso,const char * filename,struct map * map,bool no_kcore)1475 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1476 struct map *map, bool no_kcore)
1477 {
1478 struct kmap *kmap = map__kmap(map);
1479 u64 delta = 0;
1480
1481 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1482 return -1;
1483
1484 if (!kmap || !kmap->kmaps)
1485 return -1;
1486
1487 if (dso__load_all_kallsyms(dso, filename) < 0)
1488 return -1;
1489
1490 if (kallsyms__delta(kmap, filename, &delta))
1491 return -1;
1492
1493 symbols__fixup_end(dso__symbols(dso), true);
1494 symbols__fixup_duplicate(dso__symbols(dso));
1495
1496 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1497 dso__set_symtab_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS);
1498 else
1499 dso__set_symtab_type(dso, DSO_BINARY_TYPE__KALLSYMS);
1500
1501 if (!no_kcore && !dso__load_kcore(dso, map, filename))
1502 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1503 else
1504 return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1505 }
1506
dso__load_kallsyms(struct dso * dso,const char * filename,struct map * map)1507 int dso__load_kallsyms(struct dso *dso, const char *filename,
1508 struct map *map)
1509 {
1510 return __dso__load_kallsyms(dso, filename, map, false);
1511 }
1512
dso__load_perf_map(const char * map_path,struct dso * dso)1513 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1514 {
1515 char *line = NULL;
1516 size_t n;
1517 FILE *file;
1518 int nr_syms = 0;
1519
1520 file = fopen(map_path, "r");
1521 if (file == NULL)
1522 goto out_failure;
1523
1524 while (!feof(file)) {
1525 u64 start, size;
1526 struct symbol *sym;
1527 int line_len, len;
1528
1529 line_len = getline(&line, &n, file);
1530 if (line_len < 0)
1531 break;
1532
1533 if (!line)
1534 goto out_failure;
1535
1536 line[--line_len] = '\0'; /* \n */
1537
1538 len = hex2u64(line, &start);
1539
1540 len++;
1541 if (len + 2 >= line_len)
1542 continue;
1543
1544 len += hex2u64(line + len, &size);
1545
1546 len++;
1547 if (len + 2 >= line_len)
1548 continue;
1549
1550 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1551
1552 if (sym == NULL)
1553 goto out_delete_line;
1554
1555 symbols__insert(dso__symbols(dso), sym);
1556 nr_syms++;
1557 }
1558
1559 free(line);
1560 fclose(file);
1561
1562 return nr_syms;
1563
1564 out_delete_line:
1565 free(line);
1566 out_failure:
1567 return -1;
1568 }
1569
1570 #ifdef HAVE_LIBBFD_SUPPORT
1571 #define PACKAGE 'perf'
1572 #include <bfd.h>
1573
bfd_symbols__cmpvalue(const void * a,const void * b)1574 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1575 {
1576 const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1577
1578 if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1579 return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1580
1581 return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1582 }
1583
bfd2elf_binding(asymbol * symbol)1584 static int bfd2elf_binding(asymbol *symbol)
1585 {
1586 if (symbol->flags & BSF_WEAK)
1587 return STB_WEAK;
1588 if (symbol->flags & BSF_GLOBAL)
1589 return STB_GLOBAL;
1590 if (symbol->flags & BSF_LOCAL)
1591 return STB_LOCAL;
1592 return -1;
1593 }
1594
dso__load_bfd_symbols(struct dso * dso,const char * debugfile)1595 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1596 {
1597 int err = -1;
1598 long symbols_size, symbols_count, i;
1599 asection *section;
1600 asymbol **symbols, *sym;
1601 struct symbol *symbol;
1602 bfd *abfd;
1603 u64 start, len;
1604
1605 abfd = bfd_openr(debugfile, NULL);
1606 if (!abfd)
1607 return -1;
1608
1609 if (!bfd_check_format(abfd, bfd_object)) {
1610 pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1611 dso__long_name(dso));
1612 goto out_close;
1613 }
1614
1615 if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1616 goto out_close;
1617
1618 symbols_size = bfd_get_symtab_upper_bound(abfd);
1619 if (symbols_size == 0) {
1620 bfd_close(abfd);
1621 return 0;
1622 }
1623
1624 if (symbols_size < 0)
1625 goto out_close;
1626
1627 symbols = malloc(symbols_size);
1628 if (!symbols)
1629 goto out_close;
1630
1631 symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1632 if (symbols_count < 0)
1633 goto out_free;
1634
1635 section = bfd_get_section_by_name(abfd, ".text");
1636 if (section) {
1637 for (i = 0; i < symbols_count; ++i) {
1638 if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1639 !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1640 break;
1641 }
1642 if (i < symbols_count) {
1643 /* PE symbols can only have 4 bytes, so use .text high bits */
1644 u64 text_offset = (section->vma - (u32)section->vma)
1645 + (u32)bfd_asymbol_value(symbols[i]);
1646 dso__set_text_offset(dso, text_offset);
1647 dso__set_text_end(dso, (section->vma - text_offset) + section->size);
1648 } else {
1649 dso__set_text_offset(dso, section->vma - section->filepos);
1650 dso__set_text_end(dso, section->filepos + section->size);
1651 }
1652 }
1653
1654 qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1655
1656 #ifdef bfd_get_section
1657 #define bfd_asymbol_section bfd_get_section
1658 #endif
1659 for (i = 0; i < symbols_count; ++i) {
1660 sym = symbols[i];
1661 section = bfd_asymbol_section(sym);
1662 if (bfd2elf_binding(sym) < 0)
1663 continue;
1664
1665 while (i + 1 < symbols_count &&
1666 bfd_asymbol_section(symbols[i + 1]) == section &&
1667 bfd2elf_binding(symbols[i + 1]) < 0)
1668 i++;
1669
1670 if (i + 1 < symbols_count &&
1671 bfd_asymbol_section(symbols[i + 1]) == section)
1672 len = symbols[i + 1]->value - sym->value;
1673 else
1674 len = section->size - sym->value;
1675
1676 start = bfd_asymbol_value(sym) - dso__text_offset(dso);
1677 symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1678 bfd_asymbol_name(sym));
1679 if (!symbol)
1680 goto out_free;
1681
1682 symbols__insert(dso__symbols(dso), symbol);
1683 }
1684 #ifdef bfd_get_section
1685 #undef bfd_asymbol_section
1686 #endif
1687
1688 symbols__fixup_end(dso__symbols(dso), false);
1689 symbols__fixup_duplicate(dso__symbols(dso));
1690 dso__set_adjust_symbols(dso, true);
1691
1692 err = 0;
1693 out_free:
1694 free(symbols);
1695 out_close:
1696 bfd_close(abfd);
1697 return err;
1698 }
1699 #endif
1700
dso__is_compatible_symtab_type(struct dso * dso,bool kmod,enum dso_binary_type type)1701 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1702 enum dso_binary_type type)
1703 {
1704 switch (type) {
1705 case DSO_BINARY_TYPE__JAVA_JIT:
1706 case DSO_BINARY_TYPE__DEBUGLINK:
1707 case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1708 case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1709 case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1710 case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1711 case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1712 case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1713 return !kmod && dso__kernel(dso) == DSO_SPACE__USER;
1714
1715 case DSO_BINARY_TYPE__KALLSYMS:
1716 case DSO_BINARY_TYPE__VMLINUX:
1717 case DSO_BINARY_TYPE__KCORE:
1718 return dso__kernel(dso) == DSO_SPACE__KERNEL;
1719
1720 case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1721 case DSO_BINARY_TYPE__GUEST_VMLINUX:
1722 case DSO_BINARY_TYPE__GUEST_KCORE:
1723 return dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST;
1724
1725 case DSO_BINARY_TYPE__GUEST_KMODULE:
1726 case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1727 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1728 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1729 /*
1730 * kernel modules know their symtab type - it's set when
1731 * creating a module dso in machine__addnew_module_map().
1732 */
1733 return kmod && dso__symtab_type(dso) == type;
1734
1735 case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1736 case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1737 return true;
1738
1739 case DSO_BINARY_TYPE__BPF_PROG_INFO:
1740 case DSO_BINARY_TYPE__BPF_IMAGE:
1741 case DSO_BINARY_TYPE__OOL:
1742 case DSO_BINARY_TYPE__NOT_FOUND:
1743 default:
1744 return false;
1745 }
1746 }
1747
1748 /* Checks for the existence of the perf-<pid>.map file in two different
1749 * locations. First, if the process is a separate mount namespace, check in
1750 * that namespace using the pid of the innermost pid namespace. If's not in a
1751 * namespace, or the file can't be found there, try in the mount namespace of
1752 * the tracing process using our view of its pid.
1753 */
dso__find_perf_map(char * filebuf,size_t bufsz,struct nsinfo ** nsip)1754 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1755 struct nsinfo **nsip)
1756 {
1757 struct nscookie nsc;
1758 struct nsinfo *nsi;
1759 struct nsinfo *nnsi;
1760 int rc = -1;
1761
1762 nsi = *nsip;
1763
1764 if (nsinfo__need_setns(nsi)) {
1765 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1766 nsinfo__mountns_enter(nsi, &nsc);
1767 rc = access(filebuf, R_OK);
1768 nsinfo__mountns_exit(&nsc);
1769 if (rc == 0)
1770 return rc;
1771 }
1772
1773 nnsi = nsinfo__copy(nsi);
1774 if (nnsi) {
1775 nsinfo__put(nsi);
1776
1777 nsinfo__clear_need_setns(nnsi);
1778 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1779 *nsip = nnsi;
1780 rc = 0;
1781 }
1782
1783 return rc;
1784 }
1785
dso__load(struct dso * dso,struct map * map)1786 int dso__load(struct dso *dso, struct map *map)
1787 {
1788 char *name;
1789 int ret = -1;
1790 u_int i;
1791 struct machine *machine = NULL;
1792 char *root_dir = (char *) "";
1793 int ss_pos = 0;
1794 struct symsrc ss_[2];
1795 struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1796 bool kmod;
1797 bool perfmap;
1798 struct build_id bid;
1799 struct nscookie nsc;
1800 char newmapname[PATH_MAX];
1801 const char *map_path = dso__long_name(dso);
1802
1803 mutex_lock(dso__lock(dso));
1804 perfmap = is_perf_pid_map_name(map_path);
1805
1806 if (perfmap) {
1807 if (dso__nsinfo(dso) &&
1808 (dso__find_perf_map(newmapname, sizeof(newmapname),
1809 dso__nsinfo_ptr(dso)) == 0)) {
1810 map_path = newmapname;
1811 }
1812 }
1813
1814 nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
1815
1816 /* check again under the dso->lock */
1817 if (dso__loaded(dso)) {
1818 ret = 1;
1819 goto out;
1820 }
1821
1822 kmod = dso__is_kmod(dso);
1823
1824 if (dso__kernel(dso) && !kmod) {
1825 if (dso__kernel(dso) == DSO_SPACE__KERNEL)
1826 ret = dso__load_kernel_sym(dso, map);
1827 else if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1828 ret = dso__load_guest_kernel_sym(dso, map);
1829
1830 machine = maps__machine(map__kmaps(map));
1831 if (machine__is(machine, "x86_64"))
1832 machine__map_x86_64_entry_trampolines(machine, dso);
1833 goto out;
1834 }
1835
1836 dso__set_adjust_symbols(dso, false);
1837
1838 if (perfmap) {
1839 ret = dso__load_perf_map(map_path, dso);
1840 dso__set_symtab_type(dso, ret > 0
1841 ? DSO_BINARY_TYPE__JAVA_JIT
1842 : DSO_BINARY_TYPE__NOT_FOUND);
1843 goto out;
1844 }
1845
1846 if (machine)
1847 root_dir = machine->root_dir;
1848
1849 name = malloc(PATH_MAX);
1850 if (!name)
1851 goto out;
1852
1853 /*
1854 * Read the build id if possible. This is required for
1855 * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1856 */
1857 if (!dso__has_build_id(dso) &&
1858 is_regular_file(dso__long_name(dso))) {
1859 __symbol__join_symfs(name, PATH_MAX, dso__long_name(dso));
1860 if (filename__read_build_id(name, &bid) > 0)
1861 dso__set_build_id(dso, &bid);
1862 }
1863
1864 /*
1865 * Iterate over candidate debug images.
1866 * Keep track of "interesting" ones (those which have a symtab, dynsym,
1867 * and/or opd section) for processing.
1868 */
1869 for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1870 struct symsrc *ss = &ss_[ss_pos];
1871 bool next_slot = false;
1872 bool is_reg;
1873 bool nsexit;
1874 int bfdrc = -1;
1875 int sirc = -1;
1876
1877 enum dso_binary_type symtab_type = binary_type_symtab[i];
1878
1879 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1880 symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1881
1882 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1883 continue;
1884
1885 if (dso__read_binary_type_filename(dso, symtab_type,
1886 root_dir, name, PATH_MAX))
1887 continue;
1888
1889 if (nsexit)
1890 nsinfo__mountns_exit(&nsc);
1891
1892 is_reg = is_regular_file(name);
1893 if (!is_reg && errno == ENOENT && dso__nsinfo(dso)) {
1894 char *new_name = dso__filename_with_chroot(dso, name);
1895 if (new_name) {
1896 is_reg = is_regular_file(new_name);
1897 strlcpy(name, new_name, PATH_MAX);
1898 free(new_name);
1899 }
1900 }
1901
1902 #ifdef HAVE_LIBBFD_SUPPORT
1903 if (is_reg)
1904 bfdrc = dso__load_bfd_symbols(dso, name);
1905 #endif
1906 if (is_reg && bfdrc < 0)
1907 sirc = symsrc__init(ss, dso, name, symtab_type);
1908
1909 if (nsexit)
1910 nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
1911
1912 if (bfdrc == 0) {
1913 ret = 0;
1914 break;
1915 }
1916
1917 if (!is_reg || sirc < 0)
1918 continue;
1919
1920 if (!syms_ss && symsrc__has_symtab(ss)) {
1921 syms_ss = ss;
1922 next_slot = true;
1923 if (!dso__symsrc_filename(dso))
1924 dso__set_symsrc_filename(dso, strdup(name));
1925 }
1926
1927 if (!runtime_ss && symsrc__possibly_runtime(ss)) {
1928 runtime_ss = ss;
1929 next_slot = true;
1930 }
1931
1932 if (next_slot) {
1933 ss_pos++;
1934
1935 if (dso__binary_type(dso) == DSO_BINARY_TYPE__NOT_FOUND)
1936 dso__set_binary_type(dso, symtab_type);
1937
1938 if (syms_ss && runtime_ss)
1939 break;
1940 } else {
1941 symsrc__destroy(ss);
1942 }
1943
1944 }
1945
1946 if (!runtime_ss && !syms_ss)
1947 goto out_free;
1948
1949 if (runtime_ss && !syms_ss) {
1950 syms_ss = runtime_ss;
1951 }
1952
1953 /* We'll have to hope for the best */
1954 if (!runtime_ss && syms_ss)
1955 runtime_ss = syms_ss;
1956
1957 if (syms_ss)
1958 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
1959 else
1960 ret = -1;
1961
1962 if (ret > 0) {
1963 int nr_plt;
1964
1965 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
1966 if (nr_plt > 0)
1967 ret += nr_plt;
1968 }
1969
1970 for (; ss_pos > 0; ss_pos--)
1971 symsrc__destroy(&ss_[ss_pos - 1]);
1972 out_free:
1973 free(name);
1974 if (ret < 0 && strstr(dso__name(dso), " (deleted)") != NULL)
1975 ret = 0;
1976 out:
1977 dso__set_loaded(dso);
1978 mutex_unlock(dso__lock(dso));
1979 nsinfo__mountns_exit(&nsc);
1980
1981 return ret;
1982 }
1983
1984 /*
1985 * Always takes ownership of vmlinux when vmlinux_allocated == true, even if
1986 * it returns an error.
1987 */
dso__load_vmlinux(struct dso * dso,struct map * map,const char * vmlinux,bool vmlinux_allocated)1988 int dso__load_vmlinux(struct dso *dso, struct map *map,
1989 const char *vmlinux, bool vmlinux_allocated)
1990 {
1991 int err = -1;
1992 struct symsrc ss;
1993 char symfs_vmlinux[PATH_MAX];
1994 enum dso_binary_type symtab_type;
1995
1996 if (vmlinux[0] == '/')
1997 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
1998 else
1999 symbol__join_symfs(symfs_vmlinux, vmlinux);
2000
2001 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
2002 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2003 else
2004 symtab_type = DSO_BINARY_TYPE__VMLINUX;
2005
2006 if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type)) {
2007 if (vmlinux_allocated)
2008 free((char *) vmlinux);
2009 return -1;
2010 }
2011
2012 /*
2013 * dso__load_sym() may copy 'dso' which will result in the copies having
2014 * an incorrect long name unless we set it here first.
2015 */
2016 dso__set_long_name(dso, vmlinux, vmlinux_allocated);
2017 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
2018 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_VMLINUX);
2019 else
2020 dso__set_binary_type(dso, DSO_BINARY_TYPE__VMLINUX);
2021
2022 err = dso__load_sym(dso, map, &ss, &ss, 0);
2023 symsrc__destroy(&ss);
2024
2025 if (err > 0) {
2026 dso__set_loaded(dso);
2027 pr_debug("Using %s for symbols\n", symfs_vmlinux);
2028 }
2029
2030 return err;
2031 }
2032
dso__load_vmlinux_path(struct dso * dso,struct map * map)2033 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2034 {
2035 int i, err = 0;
2036 char *filename = NULL;
2037
2038 pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2039 vmlinux_path__nr_entries + 1);
2040
2041 for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2042 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2043 if (err > 0)
2044 goto out;
2045 }
2046
2047 if (!symbol_conf.ignore_vmlinux_buildid)
2048 filename = dso__build_id_filename(dso, NULL, 0, false);
2049 if (filename != NULL) {
2050 err = dso__load_vmlinux(dso, map, filename, true);
2051 if (err > 0)
2052 goto out;
2053 }
2054 out:
2055 return err;
2056 }
2057
visible_dir_filter(const char * name,struct dirent * d)2058 static bool visible_dir_filter(const char *name, struct dirent *d)
2059 {
2060 if (d->d_type != DT_DIR)
2061 return false;
2062 return lsdir_no_dot_filter(name, d);
2063 }
2064
find_matching_kcore(struct map * map,char * dir,size_t dir_sz)2065 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2066 {
2067 char kallsyms_filename[PATH_MAX];
2068 int ret = -1;
2069 struct strlist *dirs;
2070 struct str_node *nd;
2071
2072 dirs = lsdir(dir, visible_dir_filter);
2073 if (!dirs)
2074 return -1;
2075
2076 strlist__for_each_entry(nd, dirs) {
2077 scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2078 "%s/%s/kallsyms", dir, nd->s);
2079 if (!validate_kcore_addresses(kallsyms_filename, map)) {
2080 strlcpy(dir, kallsyms_filename, dir_sz);
2081 ret = 0;
2082 break;
2083 }
2084 }
2085
2086 strlist__delete(dirs);
2087
2088 return ret;
2089 }
2090
2091 /*
2092 * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2093 * since access(R_OK) only checks with real UID/GID but open() use effective
2094 * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2095 */
filename__readable(const char * file)2096 static bool filename__readable(const char *file)
2097 {
2098 int fd = open(file, O_RDONLY);
2099 if (fd < 0)
2100 return false;
2101 close(fd);
2102 return true;
2103 }
2104
dso__find_kallsyms(struct dso * dso,struct map * map)2105 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2106 {
2107 struct build_id bid;
2108 char sbuild_id[SBUILD_ID_SIZE];
2109 bool is_host = false;
2110 char path[PATH_MAX];
2111
2112 if (!dso__has_build_id(dso)) {
2113 /*
2114 * Last resort, if we don't have a build-id and couldn't find
2115 * any vmlinux file, try the running kernel kallsyms table.
2116 */
2117 goto proc_kallsyms;
2118 }
2119
2120 if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2121 is_host = dso__build_id_equal(dso, &bid);
2122
2123 /* Try a fast path for /proc/kallsyms if possible */
2124 if (is_host) {
2125 /*
2126 * Do not check the build-id cache, unless we know we cannot use
2127 * /proc/kcore or module maps don't match to /proc/kallsyms.
2128 * To check readability of /proc/kcore, do not use access(R_OK)
2129 * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2130 * can't check it.
2131 */
2132 if (filename__readable("/proc/kcore") &&
2133 !validate_kcore_addresses("/proc/kallsyms", map))
2134 goto proc_kallsyms;
2135 }
2136
2137 build_id__sprintf(dso__bid(dso), sbuild_id);
2138
2139 /* Find kallsyms in build-id cache with kcore */
2140 scnprintf(path, sizeof(path), "%s/%s/%s",
2141 buildid_dir, DSO__NAME_KCORE, sbuild_id);
2142
2143 if (!find_matching_kcore(map, path, sizeof(path)))
2144 return strdup(path);
2145
2146 /* Use current /proc/kallsyms if possible */
2147 if (is_host) {
2148 proc_kallsyms:
2149 return strdup("/proc/kallsyms");
2150 }
2151
2152 /* Finally, find a cache of kallsyms */
2153 if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2154 pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2155 sbuild_id);
2156 return NULL;
2157 }
2158
2159 return strdup(path);
2160 }
2161
dso__load_kernel_sym(struct dso * dso,struct map * map)2162 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2163 {
2164 int err;
2165 const char *kallsyms_filename = NULL;
2166 char *kallsyms_allocated_filename = NULL;
2167 char *filename = NULL;
2168
2169 /*
2170 * Step 1: if the user specified a kallsyms or vmlinux filename, use
2171 * it and only it, reporting errors to the user if it cannot be used.
2172 *
2173 * For instance, try to analyse an ARM perf.data file _without_ a
2174 * build-id, or if the user specifies the wrong path to the right
2175 * vmlinux file, obviously we can't fallback to another vmlinux (a
2176 * x86_86 one, on the machine where analysis is being performed, say),
2177 * or worse, /proc/kallsyms.
2178 *
2179 * If the specified file _has_ a build-id and there is a build-id
2180 * section in the perf.data file, we will still do the expected
2181 * validation in dso__load_vmlinux and will bail out if they don't
2182 * match.
2183 */
2184 if (symbol_conf.kallsyms_name != NULL) {
2185 kallsyms_filename = symbol_conf.kallsyms_name;
2186 goto do_kallsyms;
2187 }
2188
2189 if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2190 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2191 }
2192
2193 /*
2194 * Before checking on common vmlinux locations, check if it's
2195 * stored as standard build id binary (not kallsyms) under
2196 * .debug cache.
2197 */
2198 if (!symbol_conf.ignore_vmlinux_buildid)
2199 filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2200 if (filename != NULL) {
2201 err = dso__load_vmlinux(dso, map, filename, true);
2202 if (err > 0)
2203 return err;
2204 }
2205
2206 if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2207 err = dso__load_vmlinux_path(dso, map);
2208 if (err > 0)
2209 return err;
2210 }
2211
2212 /* do not try local files if a symfs was given */
2213 if (symbol_conf.symfs[0] != 0)
2214 return -1;
2215
2216 kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2217 if (!kallsyms_allocated_filename)
2218 return -1;
2219
2220 kallsyms_filename = kallsyms_allocated_filename;
2221
2222 do_kallsyms:
2223 err = dso__load_kallsyms(dso, kallsyms_filename, map);
2224 if (err > 0)
2225 pr_debug("Using %s for symbols\n", kallsyms_filename);
2226 free(kallsyms_allocated_filename);
2227
2228 if (err > 0 && !dso__is_kcore(dso)) {
2229 dso__set_binary_type(dso, DSO_BINARY_TYPE__KALLSYMS);
2230 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2231 map__fixup_start(map);
2232 map__fixup_end(map);
2233 }
2234
2235 return err;
2236 }
2237
dso__load_guest_kernel_sym(struct dso * dso,struct map * map)2238 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2239 {
2240 int err;
2241 const char *kallsyms_filename;
2242 struct machine *machine = maps__machine(map__kmaps(map));
2243 char path[PATH_MAX];
2244
2245 if (machine->kallsyms_filename) {
2246 kallsyms_filename = machine->kallsyms_filename;
2247 } else if (machine__is_default_guest(machine)) {
2248 /*
2249 * if the user specified a vmlinux filename, use it and only
2250 * it, reporting errors to the user if it cannot be used.
2251 * Or use file guest_kallsyms inputted by user on commandline
2252 */
2253 if (symbol_conf.default_guest_vmlinux_name != NULL) {
2254 err = dso__load_vmlinux(dso, map,
2255 symbol_conf.default_guest_vmlinux_name,
2256 false);
2257 return err;
2258 }
2259
2260 kallsyms_filename = symbol_conf.default_guest_kallsyms;
2261 if (!kallsyms_filename)
2262 return -1;
2263 } else {
2264 sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2265 kallsyms_filename = path;
2266 }
2267
2268 err = dso__load_kallsyms(dso, kallsyms_filename, map);
2269 if (err > 0)
2270 pr_debug("Using %s for symbols\n", kallsyms_filename);
2271 if (err > 0 && !dso__is_kcore(dso)) {
2272 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS);
2273 dso__set_long_name(dso, machine->mmap_name, false);
2274 map__fixup_start(map);
2275 map__fixup_end(map);
2276 }
2277
2278 return err;
2279 }
2280
vmlinux_path__exit(void)2281 static void vmlinux_path__exit(void)
2282 {
2283 while (--vmlinux_path__nr_entries >= 0)
2284 zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2285 vmlinux_path__nr_entries = 0;
2286
2287 zfree(&vmlinux_path);
2288 }
2289
2290 static const char * const vmlinux_paths[] = {
2291 "vmlinux",
2292 "/boot/vmlinux"
2293 };
2294
2295 static const char * const vmlinux_paths_upd[] = {
2296 "/boot/vmlinux-%s",
2297 "/usr/lib/debug/boot/vmlinux-%s",
2298 "/lib/modules/%s/build/vmlinux",
2299 "/usr/lib/debug/lib/modules/%s/vmlinux",
2300 "/usr/lib/debug/boot/vmlinux-%s.debug"
2301 };
2302
vmlinux_path__add(const char * new_entry)2303 static int vmlinux_path__add(const char *new_entry)
2304 {
2305 vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2306 if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2307 return -1;
2308 ++vmlinux_path__nr_entries;
2309
2310 return 0;
2311 }
2312
vmlinux_path__init(struct perf_env * env)2313 static int vmlinux_path__init(struct perf_env *env)
2314 {
2315 struct utsname uts;
2316 char bf[PATH_MAX];
2317 char *kernel_version;
2318 unsigned int i;
2319
2320 vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2321 ARRAY_SIZE(vmlinux_paths_upd)));
2322 if (vmlinux_path == NULL)
2323 return -1;
2324
2325 for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2326 if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2327 goto out_fail;
2328
2329 /* only try kernel version if no symfs was given */
2330 if (symbol_conf.symfs[0] != 0)
2331 return 0;
2332
2333 if (env) {
2334 kernel_version = env->os_release;
2335 } else {
2336 if (uname(&uts) < 0)
2337 goto out_fail;
2338
2339 kernel_version = uts.release;
2340 }
2341
2342 for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2343 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2344 if (vmlinux_path__add(bf) < 0)
2345 goto out_fail;
2346 }
2347
2348 return 0;
2349
2350 out_fail:
2351 vmlinux_path__exit();
2352 return -1;
2353 }
2354
setup_list(struct strlist ** list,const char * list_str,const char * list_name)2355 int setup_list(struct strlist **list, const char *list_str,
2356 const char *list_name)
2357 {
2358 if (list_str == NULL)
2359 return 0;
2360
2361 *list = strlist__new(list_str, NULL);
2362 if (!*list) {
2363 pr_err("problems parsing %s list\n", list_name);
2364 return -1;
2365 }
2366
2367 symbol_conf.has_filter = true;
2368 return 0;
2369 }
2370
setup_intlist(struct intlist ** list,const char * list_str,const char * list_name)2371 int setup_intlist(struct intlist **list, const char *list_str,
2372 const char *list_name)
2373 {
2374 if (list_str == NULL)
2375 return 0;
2376
2377 *list = intlist__new(list_str);
2378 if (!*list) {
2379 pr_err("problems parsing %s list\n", list_name);
2380 return -1;
2381 }
2382 return 0;
2383 }
2384
setup_addrlist(struct intlist ** addr_list,struct strlist * sym_list)2385 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2386 {
2387 struct str_node *pos, *tmp;
2388 unsigned long val;
2389 char *sep;
2390 const char *end;
2391 int i = 0, err;
2392
2393 *addr_list = intlist__new(NULL);
2394 if (!*addr_list)
2395 return -1;
2396
2397 strlist__for_each_entry_safe(pos, tmp, sym_list) {
2398 errno = 0;
2399 val = strtoul(pos->s, &sep, 16);
2400 if (errno || (sep == pos->s))
2401 continue;
2402
2403 if (*sep != '\0') {
2404 end = pos->s + strlen(pos->s) - 1;
2405 while (end >= sep && isspace(*end))
2406 end--;
2407
2408 if (end >= sep)
2409 continue;
2410 }
2411
2412 err = intlist__add(*addr_list, val);
2413 if (err)
2414 break;
2415
2416 strlist__remove(sym_list, pos);
2417 i++;
2418 }
2419
2420 if (i == 0) {
2421 intlist__delete(*addr_list);
2422 *addr_list = NULL;
2423 }
2424
2425 return 0;
2426 }
2427
symbol__read_kptr_restrict(void)2428 static bool symbol__read_kptr_restrict(void)
2429 {
2430 bool value = false;
2431 FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2432 bool used_root;
2433 bool cap_syslog = perf_cap__capable(CAP_SYSLOG, &used_root);
2434
2435 if (fp != NULL) {
2436 char line[8];
2437
2438 if (fgets(line, sizeof(line), fp) != NULL)
2439 value = cap_syslog ? (atoi(line) >= 2) : (atoi(line) != 0);
2440
2441 fclose(fp);
2442 }
2443
2444 /* Per kernel/kallsyms.c:
2445 * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2446 */
2447 if (perf_event_paranoid() > 1 && !cap_syslog)
2448 value = true;
2449
2450 return value;
2451 }
2452
symbol__annotation_init(void)2453 int symbol__annotation_init(void)
2454 {
2455 if (symbol_conf.init_annotation)
2456 return 0;
2457
2458 if (symbol_conf.initialized) {
2459 pr_err("Annotation needs to be init before symbol__init()\n");
2460 return -1;
2461 }
2462
2463 symbol_conf.priv_size += sizeof(struct annotation);
2464 symbol_conf.init_annotation = true;
2465 return 0;
2466 }
2467
symbol__init(struct perf_env * env)2468 int symbol__init(struct perf_env *env)
2469 {
2470 const char *symfs;
2471
2472 if (symbol_conf.initialized)
2473 return 0;
2474
2475 symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2476
2477 symbol__elf_init();
2478
2479 if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2480 return -1;
2481
2482 if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2483 pr_err("'.' is the only non valid --field-separator argument\n");
2484 return -1;
2485 }
2486
2487 if (setup_list(&symbol_conf.dso_list,
2488 symbol_conf.dso_list_str, "dso") < 0)
2489 return -1;
2490
2491 if (setup_list(&symbol_conf.comm_list,
2492 symbol_conf.comm_list_str, "comm") < 0)
2493 goto out_free_dso_list;
2494
2495 if (setup_intlist(&symbol_conf.pid_list,
2496 symbol_conf.pid_list_str, "pid") < 0)
2497 goto out_free_comm_list;
2498
2499 if (setup_intlist(&symbol_conf.tid_list,
2500 symbol_conf.tid_list_str, "tid") < 0)
2501 goto out_free_pid_list;
2502
2503 if (setup_list(&symbol_conf.sym_list,
2504 symbol_conf.sym_list_str, "symbol") < 0)
2505 goto out_free_tid_list;
2506
2507 if (symbol_conf.sym_list &&
2508 setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2509 goto out_free_sym_list;
2510
2511 if (setup_list(&symbol_conf.bt_stop_list,
2512 symbol_conf.bt_stop_list_str, "symbol") < 0)
2513 goto out_free_sym_list;
2514
2515 /*
2516 * A path to symbols of "/" is identical to ""
2517 * reset here for simplicity.
2518 */
2519 symfs = realpath(symbol_conf.symfs, NULL);
2520 if (symfs == NULL)
2521 symfs = symbol_conf.symfs;
2522 if (strcmp(symfs, "/") == 0)
2523 symbol_conf.symfs = "";
2524 if (symfs != symbol_conf.symfs)
2525 free((void *)symfs);
2526
2527 symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2528
2529 symbol_conf.initialized = true;
2530 return 0;
2531
2532 out_free_sym_list:
2533 strlist__delete(symbol_conf.sym_list);
2534 intlist__delete(symbol_conf.addr_list);
2535 out_free_tid_list:
2536 intlist__delete(symbol_conf.tid_list);
2537 out_free_pid_list:
2538 intlist__delete(symbol_conf.pid_list);
2539 out_free_comm_list:
2540 strlist__delete(symbol_conf.comm_list);
2541 out_free_dso_list:
2542 strlist__delete(symbol_conf.dso_list);
2543 return -1;
2544 }
2545
symbol__exit(void)2546 void symbol__exit(void)
2547 {
2548 if (!symbol_conf.initialized)
2549 return;
2550 strlist__delete(symbol_conf.bt_stop_list);
2551 strlist__delete(symbol_conf.sym_list);
2552 strlist__delete(symbol_conf.dso_list);
2553 strlist__delete(symbol_conf.comm_list);
2554 intlist__delete(symbol_conf.tid_list);
2555 intlist__delete(symbol_conf.pid_list);
2556 intlist__delete(symbol_conf.addr_list);
2557 vmlinux_path__exit();
2558 symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2559 symbol_conf.bt_stop_list = NULL;
2560 symbol_conf.initialized = false;
2561 }
2562
symbol__config_symfs(const struct option * opt __maybe_unused,const char * dir,int unset __maybe_unused)2563 int symbol__config_symfs(const struct option *opt __maybe_unused,
2564 const char *dir, int unset __maybe_unused)
2565 {
2566 char *bf = NULL;
2567 int ret;
2568
2569 symbol_conf.symfs = strdup(dir);
2570 if (symbol_conf.symfs == NULL)
2571 return -ENOMEM;
2572
2573 /* skip the locally configured cache if a symfs is given, and
2574 * config buildid dir to symfs/.debug
2575 */
2576 ret = asprintf(&bf, "%s/%s", dir, ".debug");
2577 if (ret < 0)
2578 return -ENOMEM;
2579
2580 set_buildid_dir(bf);
2581
2582 free(bf);
2583 return 0;
2584 }
2585
2586 /*
2587 * Checks that user supplied symbol kernel files are accessible because
2588 * the default mechanism for accessing elf files fails silently. i.e. if
2589 * debug syms for a build ID aren't found perf carries on normally. When
2590 * they are user supplied we should assume that the user doesn't want to
2591 * silently fail.
2592 */
symbol__validate_sym_arguments(void)2593 int symbol__validate_sym_arguments(void)
2594 {
2595 if (symbol_conf.vmlinux_name &&
2596 access(symbol_conf.vmlinux_name, R_OK)) {
2597 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2598 return -EINVAL;
2599 }
2600 if (symbol_conf.kallsyms_name &&
2601 access(symbol_conf.kallsyms_name, R_OK)) {
2602 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2603 return -EINVAL;
2604 }
2605 return 0;
2606 }
2607