blob: 2ec0a32da57937da5f992dc6ff6bc05981ba0b6d [file] [log] [blame]
Googler9398cc32022-12-02 17:21:52 +08001// SPDX-License-Identifier: GPL-2.0
Googleraf606d22022-10-26 21:40:12 -07002#include <fcntl.h>
3#include <stdio.h>
4#include <errno.h>
Googler9398cc32022-12-02 17:21:52 +08005#include <stdlib.h>
Googleraf606d22022-10-26 21:40:12 -07006#include <string.h>
7#include <unistd.h>
8#include <inttypes.h>
9
Googler9398cc32022-12-02 17:21:52 +080010#include "dso.h"
11#include "map.h"
12#include "map_groups.h"
Googleraf606d22022-10-26 21:40:12 -070013#include "symbol.h"
Googler9398cc32022-12-02 17:21:52 +080014#include "symsrc.h"
Googleraf606d22022-10-26 21:40:12 -070015#include "demangle-java.h"
16#include "demangle-rust.h"
17#include "machine.h"
18#include "vdso.h"
Googler9726be62022-12-14 05:53:31 +000019#include "debug.h"
Googler9398cc32022-12-02 17:21:52 +080020#include "util/copyfile.h"
21#include <linux/ctype.h>
22#include <linux/kernel.h>
23#include <linux/zalloc.h>
24#include <symbol/kallsyms.h>
25#include <internal/lib.h>
Googleraf606d22022-10-26 21:40:12 -070026
27#ifndef EM_AARCH64
28#define EM_AARCH64 183 /* ARM 64 bit */
29#endif
30
Googler9398cc32022-12-02 17:21:52 +080031#ifndef ELF32_ST_VISIBILITY
32#define ELF32_ST_VISIBILITY(o) ((o) & 0x03)
33#endif
34
35/* For ELF64 the definitions are the same. */
36#ifndef ELF64_ST_VISIBILITY
37#define ELF64_ST_VISIBILITY(o) ELF32_ST_VISIBILITY (o)
38#endif
39
40/* How to extract information held in the st_other field. */
41#ifndef GELF_ST_VISIBILITY
42#define GELF_ST_VISIBILITY(val) ELF64_ST_VISIBILITY (val)
43#endif
44
Googleraf606d22022-10-26 21:40:12 -070045typedef Elf64_Nhdr GElf_Nhdr;
46
Googler9398cc32022-12-02 17:21:52 +080047#ifndef DMGL_PARAMS
48#define DMGL_NO_OPTS 0 /* For readability... */
49#define DMGL_PARAMS (1 << 0) /* Include function args */
50#define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */
51#endif
52
Googleraf606d22022-10-26 21:40:12 -070053#ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
54extern char *cplus_demangle(const char *, int);
55
56static inline char *bfd_demangle(void __maybe_unused *v, const char *c, int i)
57{
58 return cplus_demangle(c, i);
59}
60#else
61#ifdef NO_DEMANGLE
62static inline char *bfd_demangle(void __maybe_unused *v,
63 const char __maybe_unused *c,
64 int __maybe_unused i)
65{
66 return NULL;
67}
68#else
69#define PACKAGE 'perf'
70#include <bfd.h>
71#endif
72#endif
73
74#ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
75static int elf_getphdrnum(Elf *elf, size_t *dst)
76{
77 GElf_Ehdr gehdr;
78 GElf_Ehdr *ehdr;
79
80 ehdr = gelf_getehdr(elf, &gehdr);
81 if (!ehdr)
82 return -1;
83
84 *dst = ehdr->e_phnum;
85
86 return 0;
87}
88#endif
89
90#ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
91static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
92{
93 pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
94 return -1;
95}
96#endif
97
98#ifndef NT_GNU_BUILD_ID
99#define NT_GNU_BUILD_ID 3
100#endif
101
102/**
103 * elf_symtab__for_each_symbol - iterate thru all the symbols
104 *
105 * @syms: struct elf_symtab instance to iterate
106 * @idx: uint32_t idx
107 * @sym: GElf_Sym iterator
108 */
109#define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
110 for (idx = 0, gelf_getsym(syms, idx, &sym);\
111 idx < nr_syms; \
112 idx++, gelf_getsym(syms, idx, &sym))
113
114static inline uint8_t elf_sym__type(const GElf_Sym *sym)
115{
116 return GELF_ST_TYPE(sym->st_info);
117}
118
119static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
120{
121 return GELF_ST_VISIBILITY(sym->st_other);
122}
123
124#ifndef STT_GNU_IFUNC
125#define STT_GNU_IFUNC 10
126#endif
127
128static inline int elf_sym__is_function(const GElf_Sym *sym)
129{
130 return (elf_sym__type(sym) == STT_FUNC ||
131 elf_sym__type(sym) == STT_GNU_IFUNC) &&
132 sym->st_name != 0 &&
133 sym->st_shndx != SHN_UNDEF;
134}
135
136static inline bool elf_sym__is_object(const GElf_Sym *sym)
137{
138 return elf_sym__type(sym) == STT_OBJECT &&
139 sym->st_name != 0 &&
140 sym->st_shndx != SHN_UNDEF;
141}
142
143static inline int elf_sym__is_label(const GElf_Sym *sym)
144{
145 return elf_sym__type(sym) == STT_NOTYPE &&
146 sym->st_name != 0 &&
147 sym->st_shndx != SHN_UNDEF &&
148 sym->st_shndx != SHN_ABS &&
149 elf_sym__visibility(sym) != STV_HIDDEN &&
150 elf_sym__visibility(sym) != STV_INTERNAL;
151}
152
Googler9398cc32022-12-02 17:21:52 +0800153static bool elf_sym__filter(GElf_Sym *sym)
Googleraf606d22022-10-26 21:40:12 -0700154{
Googler9398cc32022-12-02 17:21:52 +0800155 return elf_sym__is_function(sym) || elf_sym__is_object(sym);
Googleraf606d22022-10-26 21:40:12 -0700156}
157
158static inline const char *elf_sym__name(const GElf_Sym *sym,
159 const Elf_Data *symstrs)
160{
161 return symstrs->d_buf + sym->st_name;
162}
163
164static inline const char *elf_sec__name(const GElf_Shdr *shdr,
165 const Elf_Data *secstrs)
166{
167 return secstrs->d_buf + shdr->sh_name;
168}
169
170static inline int elf_sec__is_text(const GElf_Shdr *shdr,
171 const Elf_Data *secstrs)
172{
173 return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
174}
175
176static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
177 const Elf_Data *secstrs)
178{
179 return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
180}
181
Googler9398cc32022-12-02 17:21:52 +0800182static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
Googleraf606d22022-10-26 21:40:12 -0700183{
Googler9398cc32022-12-02 17:21:52 +0800184 return elf_sec__is_text(shdr, secstrs) ||
185 elf_sec__is_data(shdr, secstrs);
Googleraf606d22022-10-26 21:40:12 -0700186}
187
188static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
189{
190 Elf_Scn *sec = NULL;
191 GElf_Shdr shdr;
192 size_t cnt = 1;
193
194 while ((sec = elf_nextscn(elf, sec)) != NULL) {
195 gelf_getshdr(sec, &shdr);
196
197 if ((addr >= shdr.sh_addr) &&
198 (addr < (shdr.sh_addr + shdr.sh_size)))
199 return cnt;
200
201 ++cnt;
202 }
203
204 return -1;
205}
206
207Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
208 GElf_Shdr *shp, const char *name, size_t *idx)
209{
210 Elf_Scn *sec = NULL;
211 size_t cnt = 1;
212
213 /* Elf is corrupted/truncated, avoid calling elf_strptr. */
214 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
215 return NULL;
216
217 while ((sec = elf_nextscn(elf, sec)) != NULL) {
218 char *str;
219
220 gelf_getshdr(sec, shp);
221 str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
222 if (str && !strcmp(name, str)) {
223 if (idx)
224 *idx = cnt;
225 return sec;
226 }
227 ++cnt;
228 }
229
230 return NULL;
231}
232
233static bool want_demangle(bool is_kernel_sym)
234{
235 return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
236}
237
238static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
239{
Googler9398cc32022-12-02 17:21:52 +0800240 int demangle_flags = verbose > 0 ? (DMGL_PARAMS | DMGL_ANSI) : DMGL_NO_OPTS;
Googleraf606d22022-10-26 21:40:12 -0700241 char *demangled = NULL;
242
243 /*
244 * We need to figure out if the object was created from C++ sources
245 * DWARF DW_compile_unit has this, but we don't always have access
246 * to it...
247 */
248 if (!want_demangle(dso->kernel || kmodule))
249 return demangled;
250
251 demangled = bfd_demangle(NULL, elf_name, demangle_flags);
252 if (demangled == NULL)
253 demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
254 else if (rust_is_mangled(demangled))
255 /*
256 * Input to Rust demangling is the BFD-demangled
257 * name which it Rust-demangles in place.
258 */
259 rust_demangle_sym(demangled);
260
261 return demangled;
262}
263
264#define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
265 for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
266 idx < nr_entries; \
267 ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
268
269#define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
270 for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
271 idx < nr_entries; \
272 ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
273
274/*
275 * We need to check if we have a .dynsym, so that we can handle the
276 * .plt, synthesizing its symbols, that aren't on the symtabs (be it
277 * .dynsym or .symtab).
278 * And always look at the original dso, not at debuginfo packages, that
279 * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
280 */
Googler9398cc32022-12-02 17:21:52 +0800281int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
Googleraf606d22022-10-26 21:40:12 -0700282{
283 uint32_t nr_rel_entries, idx;
284 GElf_Sym sym;
Googler9398cc32022-12-02 17:21:52 +0800285 u64 plt_offset, plt_header_size, plt_entry_size;
Googleraf606d22022-10-26 21:40:12 -0700286 GElf_Shdr shdr_plt;
287 struct symbol *f;
288 GElf_Shdr shdr_rel_plt, shdr_dynsym;
289 Elf_Data *reldata, *syms, *symstrs;
290 Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
291 size_t dynsym_idx;
292 GElf_Ehdr ehdr;
293 char sympltname[1024];
294 Elf *elf;
295 int nr = 0, symidx, err = 0;
296
297 if (!ss->dynsym)
298 return 0;
299
300 elf = ss->elf;
301 ehdr = ss->ehdr;
302
303 scn_dynsym = ss->dynsym;
304 shdr_dynsym = ss->dynshdr;
305 dynsym_idx = ss->dynsym_idx;
306
307 if (scn_dynsym == NULL)
308 goto out_elf_end;
309
310 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
311 ".rela.plt", NULL);
312 if (scn_plt_rel == NULL) {
313 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
314 ".rel.plt", NULL);
315 if (scn_plt_rel == NULL)
316 goto out_elf_end;
317 }
318
319 err = -1;
320
321 if (shdr_rel_plt.sh_link != dynsym_idx)
322 goto out_elf_end;
323
324 if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
325 goto out_elf_end;
326
327 /*
328 * Fetch the relocation section to find the idxes to the GOT
329 * and the symbols in the .dynsym they refer to.
330 */
331 reldata = elf_getdata(scn_plt_rel, NULL);
332 if (reldata == NULL)
333 goto out_elf_end;
334
335 syms = elf_getdata(scn_dynsym, NULL);
336 if (syms == NULL)
337 goto out_elf_end;
338
339 scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
340 if (scn_symstrs == NULL)
341 goto out_elf_end;
342
343 symstrs = elf_getdata(scn_symstrs, NULL);
344 if (symstrs == NULL)
345 goto out_elf_end;
346
347 if (symstrs->d_size == 0)
348 goto out_elf_end;
349
350 nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
351 plt_offset = shdr_plt.sh_offset;
Googler9398cc32022-12-02 17:21:52 +0800352 switch (ehdr.e_machine) {
353 case EM_ARM:
354 plt_header_size = 20;
355 plt_entry_size = 12;
356 break;
357
358 case EM_AARCH64:
359 plt_header_size = 32;
360 plt_entry_size = 16;
361 break;
362
363 case EM_SPARC:
364 plt_header_size = 48;
365 plt_entry_size = 12;
366 break;
367
368 case EM_SPARCV9:
369 plt_header_size = 128;
370 plt_entry_size = 32;
371 break;
372
373 default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
374 plt_header_size = shdr_plt.sh_entsize;
375 plt_entry_size = shdr_plt.sh_entsize;
376 break;
377 }
378 plt_offset += plt_header_size;
Googleraf606d22022-10-26 21:40:12 -0700379
380 if (shdr_rel_plt.sh_type == SHT_RELA) {
381 GElf_Rela pos_mem, *pos;
382
383 elf_section__for_each_rela(reldata, pos, pos_mem, idx,
384 nr_rel_entries) {
385 const char *elf_name = NULL;
386 char *demangled = NULL;
387 symidx = GELF_R_SYM(pos->r_info);
388 gelf_getsym(syms, symidx, &sym);
389
390 elf_name = elf_sym__name(&sym, symstrs);
391 demangled = demangle_sym(dso, 0, elf_name);
392 if (demangled != NULL)
393 elf_name = demangled;
394 snprintf(sympltname, sizeof(sympltname),
395 "%s@plt", elf_name);
396 free(demangled);
397
Googler9398cc32022-12-02 17:21:52 +0800398 f = symbol__new(plt_offset, plt_entry_size,
399 STB_GLOBAL, STT_FUNC, sympltname);
Googleraf606d22022-10-26 21:40:12 -0700400 if (!f)
401 goto out_elf_end;
402
Googler9398cc32022-12-02 17:21:52 +0800403 plt_offset += plt_entry_size;
404 symbols__insert(&dso->symbols, f);
Googleraf606d22022-10-26 21:40:12 -0700405 ++nr;
406 }
407 } else if (shdr_rel_plt.sh_type == SHT_REL) {
408 GElf_Rel pos_mem, *pos;
409 elf_section__for_each_rel(reldata, pos, pos_mem, idx,
410 nr_rel_entries) {
411 const char *elf_name = NULL;
412 char *demangled = NULL;
413 symidx = GELF_R_SYM(pos->r_info);
414 gelf_getsym(syms, symidx, &sym);
415
416 elf_name = elf_sym__name(&sym, symstrs);
417 demangled = demangle_sym(dso, 0, elf_name);
418 if (demangled != NULL)
419 elf_name = demangled;
420 snprintf(sympltname, sizeof(sympltname),
421 "%s@plt", elf_name);
422 free(demangled);
423
Googler9398cc32022-12-02 17:21:52 +0800424 f = symbol__new(plt_offset, plt_entry_size,
425 STB_GLOBAL, STT_FUNC, sympltname);
Googleraf606d22022-10-26 21:40:12 -0700426 if (!f)
427 goto out_elf_end;
428
Googler9398cc32022-12-02 17:21:52 +0800429 plt_offset += plt_entry_size;
430 symbols__insert(&dso->symbols, f);
Googleraf606d22022-10-26 21:40:12 -0700431 ++nr;
432 }
433 }
434
435 err = 0;
436out_elf_end:
437 if (err == 0)
438 return nr;
439 pr_debug("%s: problems reading %s PLT info.\n",
440 __func__, dso->long_name);
441 return 0;
442}
443
Googler9398cc32022-12-02 17:21:52 +0800444char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
445{
446 return demangle_sym(dso, kmodule, elf_name);
447}
448
Googleraf606d22022-10-26 21:40:12 -0700449/*
450 * Align offset to 4 bytes as needed for note name and descriptor data.
451 */
452#define NOTE_ALIGN(n) (((n) + 3) & -4U)
453
454static int elf_read_build_id(Elf *elf, void *bf, size_t size)
455{
456 int err = -1;
457 GElf_Ehdr ehdr;
458 GElf_Shdr shdr;
459 Elf_Data *data;
460 Elf_Scn *sec;
461 Elf_Kind ek;
462 void *ptr;
463
464 if (size < BUILD_ID_SIZE)
465 goto out;
466
467 ek = elf_kind(elf);
468 if (ek != ELF_K_ELF)
469 goto out;
470
471 if (gelf_getehdr(elf, &ehdr) == NULL) {
472 pr_err("%s: cannot get elf header.\n", __func__);
473 goto out;
474 }
475
476 /*
477 * Check following sections for notes:
478 * '.note.gnu.build-id'
479 * '.notes'
480 * '.note' (VDSO specific)
481 */
482 do {
483 sec = elf_section_by_name(elf, &ehdr, &shdr,
484 ".note.gnu.build-id", NULL);
485 if (sec)
486 break;
487
488 sec = elf_section_by_name(elf, &ehdr, &shdr,
489 ".notes", NULL);
490 if (sec)
491 break;
492
493 sec = elf_section_by_name(elf, &ehdr, &shdr,
494 ".note", NULL);
495 if (sec)
496 break;
497
498 return err;
499
500 } while (0);
501
502 data = elf_getdata(sec, NULL);
503 if (data == NULL)
504 goto out;
505
506 ptr = data->d_buf;
507 while (ptr < (data->d_buf + data->d_size)) {
508 GElf_Nhdr *nhdr = ptr;
509 size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
510 descsz = NOTE_ALIGN(nhdr->n_descsz);
511 const char *name;
512
513 ptr += sizeof(*nhdr);
514 name = ptr;
515 ptr += namesz;
516 if (nhdr->n_type == NT_GNU_BUILD_ID &&
517 nhdr->n_namesz == sizeof("GNU")) {
518 if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
519 size_t sz = min(size, descsz);
520 memcpy(bf, ptr, sz);
521 memset(bf + sz, 0, size - sz);
522 err = descsz;
523 break;
524 }
525 }
526 ptr += descsz;
527 }
528
529out:
530 return err;
531}
532
533int filename__read_build_id(const char *filename, void *bf, size_t size)
534{
535 int fd, err = -1;
536 Elf *elf;
537
538 if (size < BUILD_ID_SIZE)
539 goto out;
540
541 fd = open(filename, O_RDONLY);
542 if (fd < 0)
543 goto out;
544
545 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
546 if (elf == NULL) {
547 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
548 goto out_close;
549 }
550
551 err = elf_read_build_id(elf, bf, size);
552
553 elf_end(elf);
554out_close:
555 close(fd);
556out:
557 return err;
558}
559
560int sysfs__read_build_id(const char *filename, void *build_id, size_t size)
561{
562 int fd, err = -1;
563
564 if (size < BUILD_ID_SIZE)
565 goto out;
566
567 fd = open(filename, O_RDONLY);
568 if (fd < 0)
569 goto out;
570
571 while (1) {
572 char bf[BUFSIZ];
573 GElf_Nhdr nhdr;
574 size_t namesz, descsz;
575
576 if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
577 break;
578
579 namesz = NOTE_ALIGN(nhdr.n_namesz);
580 descsz = NOTE_ALIGN(nhdr.n_descsz);
581 if (nhdr.n_type == NT_GNU_BUILD_ID &&
582 nhdr.n_namesz == sizeof("GNU")) {
583 if (read(fd, bf, namesz) != (ssize_t)namesz)
584 break;
585 if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
586 size_t sz = min(descsz, size);
587 if (read(fd, build_id, sz) == (ssize_t)sz) {
588 memset(build_id + sz, 0, size - sz);
589 err = 0;
590 break;
591 }
592 } else if (read(fd, bf, descsz) != (ssize_t)descsz)
593 break;
594 } else {
595 int n = namesz + descsz;
596
597 if (n > (int)sizeof(bf)) {
598 n = sizeof(bf);
599 pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
600 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
601 }
602 if (read(fd, bf, n) != n)
603 break;
604 }
605 }
606 close(fd);
607out:
608 return err;
609}
610
611int filename__read_debuglink(const char *filename, char *debuglink,
612 size_t size)
613{
614 int fd, err = -1;
615 Elf *elf;
616 GElf_Ehdr ehdr;
617 GElf_Shdr shdr;
618 Elf_Data *data;
619 Elf_Scn *sec;
620 Elf_Kind ek;
621
622 fd = open(filename, O_RDONLY);
623 if (fd < 0)
624 goto out;
625
626 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
627 if (elf == NULL) {
628 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
629 goto out_close;
630 }
631
632 ek = elf_kind(elf);
633 if (ek != ELF_K_ELF)
634 goto out_elf_end;
635
636 if (gelf_getehdr(elf, &ehdr) == NULL) {
637 pr_err("%s: cannot get elf header.\n", __func__);
638 goto out_elf_end;
639 }
640
641 sec = elf_section_by_name(elf, &ehdr, &shdr,
642 ".gnu_debuglink", NULL);
643 if (sec == NULL)
644 goto out_elf_end;
645
646 data = elf_getdata(sec, NULL);
647 if (data == NULL)
648 goto out_elf_end;
649
650 /* the start of this section is a zero-terminated string */
651 strncpy(debuglink, data->d_buf, size);
652
653 err = 0;
654
655out_elf_end:
656 elf_end(elf);
657out_close:
658 close(fd);
659out:
660 return err;
661}
662
663static int dso__swap_init(struct dso *dso, unsigned char eidata)
664{
665 static unsigned int const endian = 1;
666
667 dso->needs_swap = DSO_SWAP__NO;
668
669 switch (eidata) {
670 case ELFDATA2LSB:
671 /* We are big endian, DSO is little endian. */
672 if (*(unsigned char const *)&endian != 1)
673 dso->needs_swap = DSO_SWAP__YES;
674 break;
675
676 case ELFDATA2MSB:
677 /* We are little endian, DSO is big endian. */
678 if (*(unsigned char const *)&endian != 0)
679 dso->needs_swap = DSO_SWAP__YES;
680 break;
681
682 default:
683 pr_err("unrecognized DSO data encoding %d\n", eidata);
684 return -EINVAL;
685 }
686
687 return 0;
688}
689
690bool symsrc__possibly_runtime(struct symsrc *ss)
691{
692 return ss->dynsym || ss->opdsec;
693}
694
695bool symsrc__has_symtab(struct symsrc *ss)
696{
697 return ss->symtab != NULL;
698}
699
700void symsrc__destroy(struct symsrc *ss)
701{
702 zfree(&ss->name);
703 elf_end(ss->elf);
704 close(ss->fd);
705}
706
707bool __weak elf__needs_adjust_symbols(GElf_Ehdr ehdr)
708{
709 return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL;
710}
711
712int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
713 enum dso_binary_type type)
714{
715 GElf_Ehdr ehdr;
716 Elf *elf;
717 int fd;
718
719 if (dso__needs_decompress(dso)) {
Googler9398cc32022-12-02 17:21:52 +0800720 fd = dso__decompress_kmodule_fd(dso, name);
Googleraf606d22022-10-26 21:40:12 -0700721 if (fd < 0)
722 return -1;
Googler9398cc32022-12-02 17:21:52 +0800723
724 type = dso->symtab_type;
Googleraf606d22022-10-26 21:40:12 -0700725 } else {
726 fd = open(name, O_RDONLY);
727 if (fd < 0) {
728 dso->load_errno = errno;
729 return -1;
730 }
731 }
732
733 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
734 if (elf == NULL) {
735 pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
736 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
737 goto out_close;
738 }
739
740 if (gelf_getehdr(elf, &ehdr) == NULL) {
741 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
742 pr_debug("%s: cannot get elf header.\n", __func__);
743 goto out_elf_end;
744 }
745
746 if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
747 dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
748 goto out_elf_end;
749 }
750
751 /* Always reject images with a mismatched build-id: */
752 if (dso->has_build_id && !symbol_conf.ignore_vmlinux_buildid) {
753 u8 build_id[BUILD_ID_SIZE];
754
755 if (elf_read_build_id(elf, build_id, BUILD_ID_SIZE) < 0) {
756 dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
757 goto out_elf_end;
758 }
759
760 if (!dso__build_id_equal(dso, build_id)) {
761 pr_debug("%s: build id mismatch for %s.\n", __func__, name);
762 dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
763 goto out_elf_end;
764 }
765 }
766
767 ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
768
769 ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
770 NULL);
771 if (ss->symshdr.sh_type != SHT_SYMTAB)
772 ss->symtab = NULL;
773
774 ss->dynsym_idx = 0;
775 ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
776 &ss->dynsym_idx);
777 if (ss->dynshdr.sh_type != SHT_DYNSYM)
778 ss->dynsym = NULL;
779
780 ss->opdidx = 0;
781 ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
782 &ss->opdidx);
783 if (ss->opdshdr.sh_type != SHT_PROGBITS)
784 ss->opdsec = NULL;
785
786 if (dso->kernel == DSO_TYPE_USER)
787 ss->adjust_symbols = true;
788 else
789 ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
790
791 ss->name = strdup(name);
792 if (!ss->name) {
793 dso->load_errno = errno;
794 goto out_elf_end;
795 }
796
797 ss->elf = elf;
798 ss->fd = fd;
799 ss->ehdr = ehdr;
800 ss->type = type;
801
802 return 0;
803
804out_elf_end:
805 elf_end(elf);
806out_close:
807 close(fd);
Googler9398cc32022-12-02 17:21:52 +0800808 return -1;
Googleraf606d22022-10-26 21:40:12 -0700809}
810
811/**
812 * ref_reloc_sym_not_found - has kernel relocation symbol been found.
813 * @kmap: kernel maps and relocation reference symbol
814 *
815 * This function returns %true if we are dealing with the kernel maps and the
816 * relocation reference symbol has not yet been found. Otherwise %false is
817 * returned.
818 */
819static bool ref_reloc_sym_not_found(struct kmap *kmap)
820{
821 return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
822 !kmap->ref_reloc_sym->unrelocated_addr;
823}
824
825/**
826 * ref_reloc - kernel relocation offset.
827 * @kmap: kernel maps and relocation reference symbol
828 *
829 * This function returns the offset of kernel addresses as determined by using
830 * the relocation reference symbol i.e. if the kernel has not been relocated
831 * then the return value is zero.
832 */
833static u64 ref_reloc(struct kmap *kmap)
834{
835 if (kmap && kmap->ref_reloc_sym &&
836 kmap->ref_reloc_sym->unrelocated_addr)
837 return kmap->ref_reloc_sym->addr -
838 kmap->ref_reloc_sym->unrelocated_addr;
839 return 0;
840}
841
842void __weak arch__sym_update(struct symbol *s __maybe_unused,
843 GElf_Sym *sym __maybe_unused) { }
844
Googler9398cc32022-12-02 17:21:52 +0800845static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
846 GElf_Sym *sym, GElf_Shdr *shdr,
847 struct map_groups *kmaps, struct kmap *kmap,
848 struct dso **curr_dsop, struct map **curr_mapp,
849 const char *section_name,
850 bool adjust_kernel_syms, bool kmodule, bool *remap_kernel)
851{
852 struct dso *curr_dso = *curr_dsop;
853 struct map *curr_map;
854 char dso_name[PATH_MAX];
855
856 /* Adjust symbol to map to file offset */
857 if (adjust_kernel_syms)
858 sym->st_value -= shdr->sh_addr - shdr->sh_offset;
859
860 if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0)
861 return 0;
862
863 if (strcmp(section_name, ".text") == 0) {
864 /*
865 * The initial kernel mapping is based on
866 * kallsyms and identity maps. Overwrite it to
867 * map to the kernel dso.
868 */
869 if (*remap_kernel && dso->kernel) {
870 *remap_kernel = false;
871 map->start = shdr->sh_addr + ref_reloc(kmap);
872 map->end = map->start + shdr->sh_size;
873 map->pgoff = shdr->sh_offset;
874 map->map_ip = map__map_ip;
875 map->unmap_ip = map__unmap_ip;
876 /* Ensure maps are correctly ordered */
877 if (kmaps) {
878 map__get(map);
879 map_groups__remove(kmaps, map);
880 map_groups__insert(kmaps, map);
881 map__put(map);
882 }
883 }
884
885 /*
886 * The initial module mapping is based on
887 * /proc/modules mapped to offset zero.
888 * Overwrite it to map to the module dso.
889 */
890 if (*remap_kernel && kmodule) {
891 *remap_kernel = false;
892 map->pgoff = shdr->sh_offset;
893 }
894
895 *curr_mapp = map;
896 *curr_dsop = dso;
897 return 0;
898 }
899
900 if (!kmap)
901 return 0;
902
903 snprintf(dso_name, sizeof(dso_name), "%s%s", dso->short_name, section_name);
904
905 curr_map = map_groups__find_by_name(kmaps, dso_name);
906 if (curr_map == NULL) {
907 u64 start = sym->st_value;
908
909 if (kmodule)
910 start += map->start + shdr->sh_offset;
911
912 curr_dso = dso__new(dso_name);
913 if (curr_dso == NULL)
914 return -1;
915 curr_dso->kernel = dso->kernel;
916 curr_dso->long_name = dso->long_name;
917 curr_dso->long_name_len = dso->long_name_len;
918 curr_map = map__new2(start, curr_dso);
919 dso__put(curr_dso);
920 if (curr_map == NULL)
921 return -1;
922
923 if (adjust_kernel_syms) {
924 curr_map->start = shdr->sh_addr + ref_reloc(kmap);
925 curr_map->end = curr_map->start + shdr->sh_size;
926 curr_map->pgoff = shdr->sh_offset;
927 } else {
928 curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
929 }
930 curr_dso->symtab_type = dso->symtab_type;
931 map_groups__insert(kmaps, curr_map);
932 /*
933 * Add it before we drop the referece to curr_map, i.e. while
934 * we still are sure to have a reference to this DSO via
935 * *curr_map->dso.
936 */
937 dsos__add(&map->groups->machine->dsos, curr_dso);
938 /* kmaps already got it */
939 map__put(curr_map);
940 dso__set_loaded(curr_dso);
941 *curr_mapp = curr_map;
942 *curr_dsop = curr_dso;
943 } else
944 *curr_dsop = curr_map->dso;
945
946 return 0;
947}
948
Googleraf606d22022-10-26 21:40:12 -0700949int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
950 struct symsrc *runtime_ss, int kmodule)
951{
952 struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
953 struct map_groups *kmaps = kmap ? map__kmaps(map) : NULL;
954 struct map *curr_map = map;
955 struct dso *curr_dso = dso;
956 Elf_Data *symstrs, *secstrs;
957 uint32_t nr_syms;
958 int err = -1;
959 uint32_t idx;
960 GElf_Ehdr ehdr;
961 GElf_Shdr shdr;
962 GElf_Shdr tshdr;
963 Elf_Data *syms, *opddata = NULL;
964 GElf_Sym sym;
965 Elf_Scn *sec, *sec_strndx;
966 Elf *elf;
967 int nr = 0;
968 bool remap_kernel = false, adjust_kernel_syms = false;
969
970 if (kmap && !kmaps)
971 return -1;
972
973 dso->symtab_type = syms_ss->type;
974 dso->is_64_bit = syms_ss->is_64_bit;
975 dso->rel = syms_ss->ehdr.e_type == ET_REL;
976
977 /*
978 * Modules may already have symbols from kallsyms, but those symbols
979 * have the wrong values for the dso maps, so remove them.
980 */
981 if (kmodule && syms_ss->symtab)
Googler9398cc32022-12-02 17:21:52 +0800982 symbols__delete(&dso->symbols);
Googleraf606d22022-10-26 21:40:12 -0700983
984 if (!syms_ss->symtab) {
985 /*
986 * If the vmlinux is stripped, fail so we will fall back
987 * to using kallsyms. The vmlinux runtime symbols aren't
988 * of much use.
989 */
990 if (dso->kernel)
991 goto out_elf_end;
992
993 syms_ss->symtab = syms_ss->dynsym;
994 syms_ss->symshdr = syms_ss->dynshdr;
995 }
996
997 elf = syms_ss->elf;
998 ehdr = syms_ss->ehdr;
999 sec = syms_ss->symtab;
1000 shdr = syms_ss->symshdr;
1001
1002 if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1003 ".text", NULL))
1004 dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
1005
1006 if (runtime_ss->opdsec)
1007 opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1008
1009 syms = elf_getdata(sec, NULL);
1010 if (syms == NULL)
1011 goto out_elf_end;
1012
1013 sec = elf_getscn(elf, shdr.sh_link);
1014 if (sec == NULL)
1015 goto out_elf_end;
1016
1017 symstrs = elf_getdata(sec, NULL);
1018 if (symstrs == NULL)
1019 goto out_elf_end;
1020
1021 sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1022 if (sec_strndx == NULL)
1023 goto out_elf_end;
1024
1025 secstrs = elf_getdata(sec_strndx, NULL);
1026 if (secstrs == NULL)
1027 goto out_elf_end;
1028
1029 nr_syms = shdr.sh_size / shdr.sh_entsize;
1030
1031 memset(&sym, 0, sizeof(sym));
1032
1033 /*
1034 * The kernel relocation symbol is needed in advance in order to adjust
1035 * kernel maps correctly.
1036 */
1037 if (ref_reloc_sym_not_found(kmap)) {
1038 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1039 const char *elf_name = elf_sym__name(&sym, symstrs);
1040
1041 if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1042 continue;
1043 kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1044 map->reloc = kmap->ref_reloc_sym->addr -
1045 kmap->ref_reloc_sym->unrelocated_addr;
1046 break;
1047 }
1048 }
1049
1050 /*
1051 * Handle any relocation of vdso necessary because older kernels
1052 * attempted to prelink vdso to its virtual address.
1053 */
1054 if (dso__is_vdso(dso))
1055 map->reloc = map->start - dso->text_offset;
1056
1057 dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
1058 /*
Googler9398cc32022-12-02 17:21:52 +08001059 * Initial kernel and module mappings do not map to the dso.
1060 * Flag the fixups.
Googleraf606d22022-10-26 21:40:12 -07001061 */
Googler9398cc32022-12-02 17:21:52 +08001062 if (dso->kernel || kmodule) {
Googleraf606d22022-10-26 21:40:12 -07001063 remap_kernel = true;
1064 adjust_kernel_syms = dso->adjust_symbols;
1065 }
1066 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1067 struct symbol *f;
1068 const char *elf_name = elf_sym__name(&sym, symstrs);
1069 char *demangled = NULL;
1070 int is_label = elf_sym__is_label(&sym);
1071 const char *section_name;
1072 bool used_opd = false;
1073
Googler9398cc32022-12-02 17:21:52 +08001074 if (!is_label && !elf_sym__filter(&sym))
Googleraf606d22022-10-26 21:40:12 -07001075 continue;
1076
1077 /* Reject ARM ELF "mapping symbols": these aren't unique and
1078 * don't identify functions, so will confuse the profile
1079 * output: */
1080 if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1081 if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1082 && (elf_name[2] == '\0' || elf_name[2] == '.'))
1083 continue;
1084 }
1085
1086 if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1087 u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1088 u64 *opd = opddata->d_buf + offset;
1089 sym.st_value = DSO__SWAP(dso, u64, *opd);
1090 sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1091 sym.st_value);
1092 used_opd = true;
1093 }
1094 /*
1095 * When loading symbols in a data mapping, ABS symbols (which
1096 * has a value of SHN_ABS in its st_shndx) failed at
1097 * elf_getscn(). And it marks the loading as a failure so
1098 * already loaded symbols cannot be fixed up.
1099 *
1100 * I'm not sure what should be done. Just ignore them for now.
1101 * - Namhyung Kim
1102 */
1103 if (sym.st_shndx == SHN_ABS)
1104 continue;
1105
1106 sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1107 if (!sec)
1108 goto out_elf_end;
1109
1110 gelf_getshdr(sec, &shdr);
1111
Googler9398cc32022-12-02 17:21:52 +08001112 if (is_label && !elf_sec__filter(&shdr, secstrs))
Googleraf606d22022-10-26 21:40:12 -07001113 continue;
1114
1115 section_name = elf_sec__name(&shdr, secstrs);
1116
1117 /* On ARM, symbols for thumb functions have 1 added to
1118 * the symbol address as a flag - remove it */
1119 if ((ehdr.e_machine == EM_ARM) &&
Googler9398cc32022-12-02 17:21:52 +08001120 (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
Googleraf606d22022-10-26 21:40:12 -07001121 (sym.st_value & 1))
1122 --sym.st_value;
1123
1124 if (dso->kernel || kmodule) {
Googler9398cc32022-12-02 17:21:52 +08001125 if (dso__process_kernel_symbol(dso, map, &sym, &shdr, kmaps, kmap, &curr_dso, &curr_map,
1126 section_name, adjust_kernel_syms, kmodule, &remap_kernel))
1127 goto out_elf_end;
1128 } else if ((used_opd && runtime_ss->adjust_symbols) ||
1129 (!used_opd && syms_ss->adjust_symbols)) {
Googleraf606d22022-10-26 21:40:12 -07001130 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
Googlerb48fa912023-03-17 12:40:29 +05301131 "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", __func__,
1132 (u64)sym.st_value, (u64)shdr.sh_addr,
1133 (u64)shdr.sh_offset);
1134 sym.st_value -= shdr.sh_addr - shdr.sh_offset;
Googleraf606d22022-10-26 21:40:12 -07001135 }
Googler9398cc32022-12-02 17:21:52 +08001136
Googleraf606d22022-10-26 21:40:12 -07001137 demangled = demangle_sym(dso, kmodule, elf_name);
1138 if (demangled != NULL)
1139 elf_name = demangled;
1140
1141 f = symbol__new(sym.st_value, sym.st_size,
Googler9398cc32022-12-02 17:21:52 +08001142 GELF_ST_BIND(sym.st_info),
1143 GELF_ST_TYPE(sym.st_info), elf_name);
Googleraf606d22022-10-26 21:40:12 -07001144 free(demangled);
1145 if (!f)
1146 goto out_elf_end;
1147
1148 arch__sym_update(f, &sym);
1149
Googler9398cc32022-12-02 17:21:52 +08001150 __symbols__insert(&curr_dso->symbols, f, dso->kernel);
Googleraf606d22022-10-26 21:40:12 -07001151 nr++;
1152 }
1153
1154 /*
1155 * For misannotated, zeroed, ASM function sizes.
1156 */
1157 if (nr > 0) {
Googler9398cc32022-12-02 17:21:52 +08001158 symbols__fixup_end(&dso->symbols);
1159 symbols__fixup_duplicate(&dso->symbols);
Googleraf606d22022-10-26 21:40:12 -07001160 if (kmap) {
1161 /*
1162 * We need to fixup this here too because we create new
1163 * maps here, for things like vsyscall sections.
1164 */
Googler9398cc32022-12-02 17:21:52 +08001165 map_groups__fixup_end(kmaps);
Googleraf606d22022-10-26 21:40:12 -07001166 }
1167 }
1168 err = nr;
1169out_elf_end:
1170 return err;
1171}
1172
1173static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1174{
1175 GElf_Phdr phdr;
1176 size_t i, phdrnum;
1177 int err;
1178 u64 sz;
1179
1180 if (elf_getphdrnum(elf, &phdrnum))
1181 return -1;
1182
1183 for (i = 0; i < phdrnum; i++) {
1184 if (gelf_getphdr(elf, i, &phdr) == NULL)
1185 return -1;
1186 if (phdr.p_type != PT_LOAD)
1187 continue;
1188 if (exe) {
1189 if (!(phdr.p_flags & PF_X))
1190 continue;
1191 } else {
1192 if (!(phdr.p_flags & PF_R))
1193 continue;
1194 }
1195 sz = min(phdr.p_memsz, phdr.p_filesz);
1196 if (!sz)
1197 continue;
1198 err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1199 if (err)
1200 return err;
1201 }
1202 return 0;
1203}
1204
1205int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1206 bool *is_64_bit)
1207{
1208 int err;
1209 Elf *elf;
1210
1211 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1212 if (elf == NULL)
1213 return -1;
1214
1215 if (is_64_bit)
1216 *is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1217
1218 err = elf_read_maps(elf, exe, mapfn, data);
1219
1220 elf_end(elf);
1221 return err;
1222}
1223
1224enum dso_type dso__type_fd(int fd)
1225{
1226 enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1227 GElf_Ehdr ehdr;
1228 Elf_Kind ek;
1229 Elf *elf;
1230
1231 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1232 if (elf == NULL)
1233 goto out;
1234
1235 ek = elf_kind(elf);
1236 if (ek != ELF_K_ELF)
1237 goto out_end;
1238
1239 if (gelf_getclass(elf) == ELFCLASS64) {
1240 dso_type = DSO__TYPE_64BIT;
1241 goto out_end;
1242 }
1243
1244 if (gelf_getehdr(elf, &ehdr) == NULL)
1245 goto out_end;
1246
1247 if (ehdr.e_machine == EM_X86_64)
1248 dso_type = DSO__TYPE_X32BIT;
1249 else
1250 dso_type = DSO__TYPE_32BIT;
1251out_end:
1252 elf_end(elf);
1253out:
1254 return dso_type;
1255}
1256
1257static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1258{
1259 ssize_t r;
1260 size_t n;
1261 int err = -1;
1262 char *buf = malloc(page_size);
1263
1264 if (buf == NULL)
1265 return -1;
1266
1267 if (lseek(to, to_offs, SEEK_SET) != to_offs)
1268 goto out;
1269
1270 if (lseek(from, from_offs, SEEK_SET) != from_offs)
1271 goto out;
1272
1273 while (len) {
1274 n = page_size;
1275 if (len < n)
1276 n = len;
1277 /* Use read because mmap won't work on proc files */
1278 r = read(from, buf, n);
1279 if (r < 0)
1280 goto out;
1281 if (!r)
1282 break;
1283 n = r;
1284 r = write(to, buf, n);
1285 if (r < 0)
1286 goto out;
1287 if ((size_t)r != n)
1288 goto out;
1289 len -= n;
1290 }
1291
1292 err = 0;
1293out:
1294 free(buf);
1295 return err;
1296}
1297
1298struct kcore {
1299 int fd;
1300 int elfclass;
1301 Elf *elf;
1302 GElf_Ehdr ehdr;
1303};
1304
1305static int kcore__open(struct kcore *kcore, const char *filename)
1306{
1307 GElf_Ehdr *ehdr;
1308
1309 kcore->fd = open(filename, O_RDONLY);
1310 if (kcore->fd == -1)
1311 return -1;
1312
1313 kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1314 if (!kcore->elf)
1315 goto out_close;
1316
1317 kcore->elfclass = gelf_getclass(kcore->elf);
1318 if (kcore->elfclass == ELFCLASSNONE)
1319 goto out_end;
1320
1321 ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1322 if (!ehdr)
1323 goto out_end;
1324
1325 return 0;
1326
1327out_end:
1328 elf_end(kcore->elf);
1329out_close:
1330 close(kcore->fd);
1331 return -1;
1332}
1333
1334static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1335 bool temp)
1336{
1337 kcore->elfclass = elfclass;
1338
1339 if (temp)
1340 kcore->fd = mkstemp(filename);
1341 else
1342 kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1343 if (kcore->fd == -1)
1344 return -1;
1345
1346 kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1347 if (!kcore->elf)
1348 goto out_close;
1349
1350 if (!gelf_newehdr(kcore->elf, elfclass))
1351 goto out_end;
1352
1353 memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1354
1355 return 0;
1356
1357out_end:
1358 elf_end(kcore->elf);
1359out_close:
1360 close(kcore->fd);
1361 unlink(filename);
1362 return -1;
1363}
1364
1365static void kcore__close(struct kcore *kcore)
1366{
1367 elf_end(kcore->elf);
1368 close(kcore->fd);
1369}
1370
1371static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1372{
1373 GElf_Ehdr *ehdr = &to->ehdr;
1374 GElf_Ehdr *kehdr = &from->ehdr;
1375
1376 memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1377 ehdr->e_type = kehdr->e_type;
1378 ehdr->e_machine = kehdr->e_machine;
1379 ehdr->e_version = kehdr->e_version;
1380 ehdr->e_entry = 0;
1381 ehdr->e_shoff = 0;
1382 ehdr->e_flags = kehdr->e_flags;
1383 ehdr->e_phnum = count;
1384 ehdr->e_shentsize = 0;
1385 ehdr->e_shnum = 0;
1386 ehdr->e_shstrndx = 0;
1387
1388 if (from->elfclass == ELFCLASS32) {
1389 ehdr->e_phoff = sizeof(Elf32_Ehdr);
1390 ehdr->e_ehsize = sizeof(Elf32_Ehdr);
1391 ehdr->e_phentsize = sizeof(Elf32_Phdr);
1392 } else {
1393 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1394 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1395 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1396 }
1397
1398 if (!gelf_update_ehdr(to->elf, ehdr))
1399 return -1;
1400
1401 if (!gelf_newphdr(to->elf, count))
1402 return -1;
1403
1404 return 0;
1405}
1406
1407static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
1408 u64 addr, u64 len)
1409{
1410 GElf_Phdr phdr = {
1411 .p_type = PT_LOAD,
1412 .p_flags = PF_R | PF_W | PF_X,
1413 .p_offset = offset,
1414 .p_vaddr = addr,
1415 .p_paddr = 0,
1416 .p_filesz = len,
1417 .p_memsz = len,
1418 .p_align = page_size,
1419 };
1420
1421 if (!gelf_update_phdr(kcore->elf, idx, &phdr))
1422 return -1;
1423
1424 return 0;
1425}
1426
1427static off_t kcore__write(struct kcore *kcore)
1428{
1429 return elf_update(kcore->elf, ELF_C_WRITE);
1430}
1431
1432struct phdr_data {
1433 off_t offset;
Googler9398cc32022-12-02 17:21:52 +08001434 off_t rel;
Googleraf606d22022-10-26 21:40:12 -07001435 u64 addr;
1436 u64 len;
Googler9398cc32022-12-02 17:21:52 +08001437 struct list_head node;
1438 struct phdr_data *remaps;
1439};
1440
1441struct sym_data {
1442 u64 addr;
1443 struct list_head node;
Googleraf606d22022-10-26 21:40:12 -07001444};
1445
1446struct kcore_copy_info {
1447 u64 stext;
1448 u64 etext;
1449 u64 first_symbol;
1450 u64 last_symbol;
1451 u64 first_module;
1452 u64 first_module_symbol;
1453 u64 last_module_symbol;
Googler9398cc32022-12-02 17:21:52 +08001454 size_t phnum;
1455 struct list_head phdrs;
1456 struct list_head syms;
Googleraf606d22022-10-26 21:40:12 -07001457};
1458
Googler9398cc32022-12-02 17:21:52 +08001459#define kcore_copy__for_each_phdr(k, p) \
1460 list_for_each_entry((p), &(k)->phdrs, node)
1461
1462static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
1463{
1464 struct phdr_data *p = zalloc(sizeof(*p));
1465
1466 if (p) {
1467 p->addr = addr;
1468 p->len = len;
1469 p->offset = offset;
1470 }
1471
1472 return p;
1473}
1474
1475static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
1476 u64 addr, u64 len,
1477 off_t offset)
1478{
1479 struct phdr_data *p = phdr_data__new(addr, len, offset);
1480
1481 if (p)
1482 list_add_tail(&p->node, &kci->phdrs);
1483
1484 return p;
1485}
1486
1487static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
1488{
1489 struct phdr_data *p, *tmp;
1490
1491 list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
1492 list_del_init(&p->node);
1493 free(p);
1494 }
1495}
1496
1497static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
1498 u64 addr)
1499{
1500 struct sym_data *s = zalloc(sizeof(*s));
1501
1502 if (s) {
1503 s->addr = addr;
1504 list_add_tail(&s->node, &kci->syms);
1505 }
1506
1507 return s;
1508}
1509
1510static void kcore_copy__free_syms(struct kcore_copy_info *kci)
1511{
1512 struct sym_data *s, *tmp;
1513
1514 list_for_each_entry_safe(s, tmp, &kci->syms, node) {
1515 list_del_init(&s->node);
1516 free(s);
1517 }
1518}
1519
Googleraf606d22022-10-26 21:40:12 -07001520static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
1521 u64 start)
1522{
1523 struct kcore_copy_info *kci = arg;
1524
Googler9398cc32022-12-02 17:21:52 +08001525 if (!kallsyms__is_function(type))
Googleraf606d22022-10-26 21:40:12 -07001526 return 0;
1527
1528 if (strchr(name, '[')) {
1529 if (!kci->first_module_symbol || start < kci->first_module_symbol)
1530 kci->first_module_symbol = start;
1531 if (start > kci->last_module_symbol)
1532 kci->last_module_symbol = start;
1533 return 0;
1534 }
1535
1536 if (!kci->first_symbol || start < kci->first_symbol)
1537 kci->first_symbol = start;
1538
1539 if (!kci->last_symbol || start > kci->last_symbol)
1540 kci->last_symbol = start;
1541
1542 if (!strcmp(name, "_stext")) {
1543 kci->stext = start;
1544 return 0;
1545 }
1546
1547 if (!strcmp(name, "_etext")) {
1548 kci->etext = start;
1549 return 0;
1550 }
1551
Googler9398cc32022-12-02 17:21:52 +08001552 if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
1553 return -1;
1554
Googleraf606d22022-10-26 21:40:12 -07001555 return 0;
1556}
1557
1558static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
1559 const char *dir)
1560{
1561 char kallsyms_filename[PATH_MAX];
1562
1563 scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
1564
1565 if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
1566 return -1;
1567
1568 if (kallsyms__parse(kallsyms_filename, kci,
1569 kcore_copy__process_kallsyms) < 0)
1570 return -1;
1571
1572 return 0;
1573}
1574
1575static int kcore_copy__process_modules(void *arg,
1576 const char *name __maybe_unused,
1577 u64 start, u64 size __maybe_unused)
1578{
1579 struct kcore_copy_info *kci = arg;
1580
1581 if (!kci->first_module || start < kci->first_module)
1582 kci->first_module = start;
1583
1584 return 0;
1585}
1586
1587static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
1588 const char *dir)
1589{
1590 char modules_filename[PATH_MAX];
1591
1592 scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
1593
1594 if (symbol__restricted_filename(modules_filename, "/proc/modules"))
1595 return -1;
1596
1597 if (modules__parse(modules_filename, kci,
1598 kcore_copy__process_modules) < 0)
1599 return -1;
1600
1601 return 0;
1602}
1603
Googler9398cc32022-12-02 17:21:52 +08001604static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
1605 u64 pgoff, u64 s, u64 e)
Googleraf606d22022-10-26 21:40:12 -07001606{
Googler9398cc32022-12-02 17:21:52 +08001607 u64 len, offset;
Googleraf606d22022-10-26 21:40:12 -07001608
Googler9398cc32022-12-02 17:21:52 +08001609 if (s < start || s >= end)
1610 return 0;
1611
1612 offset = (s - start) + pgoff;
1613 len = e < end ? e - s : end - s;
1614
1615 return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
Googleraf606d22022-10-26 21:40:12 -07001616}
1617
1618static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
1619{
1620 struct kcore_copy_info *kci = data;
1621 u64 end = start + len;
Googler9398cc32022-12-02 17:21:52 +08001622 struct sym_data *sdat;
Googleraf606d22022-10-26 21:40:12 -07001623
Googler9398cc32022-12-02 17:21:52 +08001624 if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
1625 return -1;
Googleraf606d22022-10-26 21:40:12 -07001626
Googler9398cc32022-12-02 17:21:52 +08001627 if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
1628 kci->last_module_symbol))
1629 return -1;
1630
1631 list_for_each_entry(sdat, &kci->syms, node) {
1632 u64 s = round_down(sdat->addr, page_size);
1633
1634 if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
1635 return -1;
1636 }
Googleraf606d22022-10-26 21:40:12 -07001637
1638 return 0;
1639}
1640
1641static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
1642{
1643 if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
1644 return -1;
1645
1646 return 0;
1647}
1648
Googler9398cc32022-12-02 17:21:52 +08001649static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
1650{
1651 struct phdr_data *p, *k = NULL;
1652 u64 kend;
1653
1654 if (!kci->stext)
1655 return;
1656
1657 /* Find phdr that corresponds to the kernel map (contains stext) */
1658 kcore_copy__for_each_phdr(kci, p) {
1659 u64 pend = p->addr + p->len - 1;
1660
1661 if (p->addr <= kci->stext && pend >= kci->stext) {
1662 k = p;
1663 break;
1664 }
1665 }
1666
1667 if (!k)
1668 return;
1669
1670 kend = k->offset + k->len;
1671
1672 /* Find phdrs that remap the kernel */
1673 kcore_copy__for_each_phdr(kci, p) {
1674 u64 pend = p->offset + p->len;
1675
1676 if (p == k)
1677 continue;
1678
1679 if (p->offset >= k->offset && pend <= kend)
1680 p->remaps = k;
1681 }
1682}
1683
1684static void kcore_copy__layout(struct kcore_copy_info *kci)
1685{
1686 struct phdr_data *p;
1687 off_t rel = 0;
1688
1689 kcore_copy__find_remaps(kci);
1690
1691 kcore_copy__for_each_phdr(kci, p) {
1692 if (!p->remaps) {
1693 p->rel = rel;
1694 rel += p->len;
1695 }
1696 kci->phnum += 1;
1697 }
1698
1699 kcore_copy__for_each_phdr(kci, p) {
1700 struct phdr_data *k = p->remaps;
1701
1702 if (k)
1703 p->rel = p->offset - k->offset + k->rel;
1704 }
1705}
1706
Googleraf606d22022-10-26 21:40:12 -07001707static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
1708 Elf *elf)
1709{
1710 if (kcore_copy__parse_kallsyms(kci, dir))
1711 return -1;
1712
1713 if (kcore_copy__parse_modules(kci, dir))
1714 return -1;
1715
1716 if (kci->stext)
1717 kci->stext = round_down(kci->stext, page_size);
1718 else
1719 kci->stext = round_down(kci->first_symbol, page_size);
1720
1721 if (kci->etext) {
1722 kci->etext = round_up(kci->etext, page_size);
1723 } else if (kci->last_symbol) {
1724 kci->etext = round_up(kci->last_symbol, page_size);
1725 kci->etext += page_size;
1726 }
1727
1728 if (kci->first_module_symbol &&
1729 (!kci->first_module || kci->first_module_symbol < kci->first_module))
1730 kci->first_module = kci->first_module_symbol;
1731
1732 kci->first_module = round_down(kci->first_module, page_size);
1733
1734 if (kci->last_module_symbol) {
1735 kci->last_module_symbol = round_up(kci->last_module_symbol,
1736 page_size);
1737 kci->last_module_symbol += page_size;
1738 }
1739
1740 if (!kci->stext || !kci->etext)
1741 return -1;
1742
1743 if (kci->first_module && !kci->last_module_symbol)
1744 return -1;
1745
Googler9398cc32022-12-02 17:21:52 +08001746 if (kcore_copy__read_maps(kci, elf))
1747 return -1;
1748
1749 kcore_copy__layout(kci);
1750
1751 return 0;
Googleraf606d22022-10-26 21:40:12 -07001752}
1753
1754static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
1755 const char *name)
1756{
1757 char from_filename[PATH_MAX];
1758 char to_filename[PATH_MAX];
1759
1760 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1761 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1762
1763 return copyfile_mode(from_filename, to_filename, 0400);
1764}
1765
1766static int kcore_copy__unlink(const char *dir, const char *name)
1767{
1768 char filename[PATH_MAX];
1769
1770 scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
1771
1772 return unlink(filename);
1773}
1774
1775static int kcore_copy__compare_fds(int from, int to)
1776{
1777 char *buf_from;
1778 char *buf_to;
1779 ssize_t ret;
1780 size_t len;
1781 int err = -1;
1782
1783 buf_from = malloc(page_size);
1784 buf_to = malloc(page_size);
1785 if (!buf_from || !buf_to)
1786 goto out;
1787
1788 while (1) {
1789 /* Use read because mmap won't work on proc files */
1790 ret = read(from, buf_from, page_size);
1791 if (ret < 0)
1792 goto out;
1793
1794 if (!ret)
1795 break;
1796
1797 len = ret;
1798
1799 if (readn(to, buf_to, len) != (int)len)
1800 goto out;
1801
1802 if (memcmp(buf_from, buf_to, len))
1803 goto out;
1804 }
1805
1806 err = 0;
1807out:
1808 free(buf_to);
1809 free(buf_from);
1810 return err;
1811}
1812
1813static int kcore_copy__compare_files(const char *from_filename,
1814 const char *to_filename)
1815{
1816 int from, to, err = -1;
1817
1818 from = open(from_filename, O_RDONLY);
1819 if (from < 0)
1820 return -1;
1821
1822 to = open(to_filename, O_RDONLY);
1823 if (to < 0)
1824 goto out_close_from;
1825
1826 err = kcore_copy__compare_fds(from, to);
1827
1828 close(to);
1829out_close_from:
1830 close(from);
1831 return err;
1832}
1833
1834static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
1835 const char *name)
1836{
1837 char from_filename[PATH_MAX];
1838 char to_filename[PATH_MAX];
1839
1840 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1841 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1842
1843 return kcore_copy__compare_files(from_filename, to_filename);
1844}
1845
1846/**
1847 * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
1848 * @from_dir: from directory
1849 * @to_dir: to directory
1850 *
1851 * This function copies kallsyms, modules and kcore files from one directory to
1852 * another. kallsyms and modules are copied entirely. Only code segments are
1853 * copied from kcore. It is assumed that two segments suffice: one for the
1854 * kernel proper and one for all the modules. The code segments are determined
1855 * from kallsyms and modules files. The kernel map starts at _stext or the
1856 * lowest function symbol, and ends at _etext or the highest function symbol.
1857 * The module map starts at the lowest module address and ends at the highest
1858 * module symbol. Start addresses are rounded down to the nearest page. End
1859 * addresses are rounded up to the nearest page. An extra page is added to the
1860 * highest kernel symbol and highest module symbol to, hopefully, encompass that
1861 * symbol too. Because it contains only code sections, the resulting kcore is
1862 * unusual. One significant peculiarity is that the mapping (start -> pgoff)
1863 * is not the same for the kernel map and the modules map. That happens because
1864 * the data is copied adjacently whereas the original kcore has gaps. Finally,
1865 * kallsyms and modules files are compared with their copies to check that
1866 * modules have not been loaded or unloaded while the copies were taking place.
1867 *
1868 * Return: %0 on success, %-1 on failure.
1869 */
1870int kcore_copy(const char *from_dir, const char *to_dir)
1871{
1872 struct kcore kcore;
1873 struct kcore extract;
1874 int idx = 0, err = -1;
Googler9398cc32022-12-02 17:21:52 +08001875 off_t offset, sz;
Googleraf606d22022-10-26 21:40:12 -07001876 struct kcore_copy_info kci = { .stext = 0, };
1877 char kcore_filename[PATH_MAX];
1878 char extract_filename[PATH_MAX];
Googler9398cc32022-12-02 17:21:52 +08001879 struct phdr_data *p;
1880
1881 INIT_LIST_HEAD(&kci.phdrs);
1882 INIT_LIST_HEAD(&kci.syms);
Googleraf606d22022-10-26 21:40:12 -07001883
1884 if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
1885 return -1;
1886
1887 if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
1888 goto out_unlink_kallsyms;
1889
1890 scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
1891 scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
1892
1893 if (kcore__open(&kcore, kcore_filename))
1894 goto out_unlink_modules;
1895
1896 if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
1897 goto out_kcore_close;
1898
1899 if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
1900 goto out_kcore_close;
1901
Googler9398cc32022-12-02 17:21:52 +08001902 if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
Googleraf606d22022-10-26 21:40:12 -07001903 goto out_extract_close;
1904
Googler9398cc32022-12-02 17:21:52 +08001905 offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
1906 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
1907 offset = round_up(offset, page_size);
Googleraf606d22022-10-26 21:40:12 -07001908
Googler9398cc32022-12-02 17:21:52 +08001909 kcore_copy__for_each_phdr(&kci, p) {
1910 off_t offs = p->rel + offset;
1911
1912 if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
Googleraf606d22022-10-26 21:40:12 -07001913 goto out_extract_close;
1914 }
1915
1916 sz = kcore__write(&extract);
1917 if (sz < 0 || sz > offset)
1918 goto out_extract_close;
1919
Googler9398cc32022-12-02 17:21:52 +08001920 kcore_copy__for_each_phdr(&kci, p) {
1921 off_t offs = p->rel + offset;
Googleraf606d22022-10-26 21:40:12 -07001922
Googler9398cc32022-12-02 17:21:52 +08001923 if (p->remaps)
1924 continue;
1925 if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
1926 goto out_extract_close;
1927 }
Googleraf606d22022-10-26 21:40:12 -07001928
1929 if (kcore_copy__compare_file(from_dir, to_dir, "modules"))
1930 goto out_extract_close;
1931
1932 if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
1933 goto out_extract_close;
1934
1935 err = 0;
1936
1937out_extract_close:
1938 kcore__close(&extract);
1939 if (err)
1940 unlink(extract_filename);
1941out_kcore_close:
1942 kcore__close(&kcore);
1943out_unlink_modules:
1944 if (err)
1945 kcore_copy__unlink(to_dir, "modules");
1946out_unlink_kallsyms:
1947 if (err)
1948 kcore_copy__unlink(to_dir, "kallsyms");
1949
Googler9398cc32022-12-02 17:21:52 +08001950 kcore_copy__free_phdrs(&kci);
1951 kcore_copy__free_syms(&kci);
1952
Googleraf606d22022-10-26 21:40:12 -07001953 return err;
1954}
1955
1956int kcore_extract__create(struct kcore_extract *kce)
1957{
1958 struct kcore kcore;
1959 struct kcore extract;
1960 size_t count = 1;
1961 int idx = 0, err = -1;
1962 off_t offset = page_size, sz;
1963
1964 if (kcore__open(&kcore, kce->kcore_filename))
1965 return -1;
1966
1967 strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
1968 if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
1969 goto out_kcore_close;
1970
1971 if (kcore__copy_hdr(&kcore, &extract, count))
1972 goto out_extract_close;
1973
1974 if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
1975 goto out_extract_close;
1976
1977 sz = kcore__write(&extract);
1978 if (sz < 0 || sz > offset)
1979 goto out_extract_close;
1980
1981 if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
1982 goto out_extract_close;
1983
1984 err = 0;
1985
1986out_extract_close:
1987 kcore__close(&extract);
1988 if (err)
1989 unlink(kce->extract_filename);
1990out_kcore_close:
1991 kcore__close(&kcore);
1992
1993 return err;
1994}
1995
1996void kcore_extract__delete(struct kcore_extract *kce)
1997{
1998 unlink(kce->extract_filename);
1999}
2000
2001#ifdef HAVE_GELF_GETNOTE_SUPPORT
Googler9398cc32022-12-02 17:21:52 +08002002
2003static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2004{
2005 if (!base_off)
2006 return;
2007
2008 if (tmp->bit32)
2009 tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2010 tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2011 tmp->addr.a32[SDT_NOTE_IDX_BASE];
2012 else
2013 tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2014 tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2015 tmp->addr.a64[SDT_NOTE_IDX_BASE];
2016}
2017
2018static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2019 GElf_Addr base_off)
2020{
2021 if (!base_off)
2022 return;
2023
2024 if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2025 tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2026 else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2027 tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2028}
2029
Googleraf606d22022-10-26 21:40:12 -07002030/**
2031 * populate_sdt_note : Parse raw data and identify SDT note
2032 * @elf: elf of the opened file
2033 * @data: raw data of a section with description offset applied
2034 * @len: note description size
2035 * @type: type of the note
2036 * @sdt_notes: List to add the SDT note
2037 *
2038 * Responsible for parsing the @data in section .note.stapsdt in @elf and
2039 * if its an SDT note, it appends to @sdt_notes list.
2040 */
2041static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2042 struct list_head *sdt_notes)
2043{
Googler9398cc32022-12-02 17:21:52 +08002044 const char *provider, *name, *args;
Googleraf606d22022-10-26 21:40:12 -07002045 struct sdt_note *tmp = NULL;
2046 GElf_Ehdr ehdr;
2047 GElf_Shdr shdr;
2048 int ret = -EINVAL;
2049
2050 union {
2051 Elf64_Addr a64[NR_ADDR];
2052 Elf32_Addr a32[NR_ADDR];
2053 } buf;
2054
2055 Elf_Data dst = {
2056 .d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2057 .d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2058 .d_off = 0, .d_align = 0
2059 };
2060 Elf_Data src = {
2061 .d_buf = (void *) data, .d_type = ELF_T_ADDR,
2062 .d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2063 .d_align = 0
2064 };
2065
2066 tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2067 if (!tmp) {
2068 ret = -ENOMEM;
2069 goto out_err;
2070 }
2071
2072 INIT_LIST_HEAD(&tmp->note_list);
2073
2074 if (len < dst.d_size + 3)
2075 goto out_free_note;
2076
2077 /* Translation from file representation to memory representation */
2078 if (gelf_xlatetom(*elf, &dst, &src,
2079 elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2080 pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2081 goto out_free_note;
2082 }
2083
2084 /* Populate the fields of sdt_note */
2085 provider = data + dst.d_size;
2086
2087 name = (const char *)memchr(provider, '\0', data + len - provider);
2088 if (name++ == NULL)
2089 goto out_free_note;
2090
2091 tmp->provider = strdup(provider);
2092 if (!tmp->provider) {
2093 ret = -ENOMEM;
2094 goto out_free_note;
2095 }
2096 tmp->name = strdup(name);
2097 if (!tmp->name) {
2098 ret = -ENOMEM;
2099 goto out_free_prov;
2100 }
2101
Googler9398cc32022-12-02 17:21:52 +08002102 args = memchr(name, '\0', data + len - name);
2103
2104 /*
2105 * There is no argument if:
2106 * - We reached the end of the note;
2107 * - There is not enough room to hold a potential string;
2108 * - The argument string is empty or just contains ':'.
2109 */
2110 if (args == NULL || data + len - args < 2 ||
2111 args[1] == ':' || args[1] == '\0')
2112 tmp->args = NULL;
2113 else {
2114 tmp->args = strdup(++args);
2115 if (!tmp->args) {
2116 ret = -ENOMEM;
2117 goto out_free_name;
2118 }
2119 }
2120
Googleraf606d22022-10-26 21:40:12 -07002121 if (gelf_getclass(*elf) == ELFCLASS32) {
2122 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2123 tmp->bit32 = true;
2124 } else {
2125 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2126 tmp->bit32 = false;
2127 }
2128
2129 if (!gelf_getehdr(*elf, &ehdr)) {
2130 pr_debug("%s : cannot get elf header.\n", __func__);
2131 ret = -EBADF;
Googler9398cc32022-12-02 17:21:52 +08002132 goto out_free_args;
Googleraf606d22022-10-26 21:40:12 -07002133 }
2134
2135 /* Adjust the prelink effect :
2136 * Find out the .stapsdt.base section.
2137 * This scn will help us to handle prelinking (if present).
2138 * Compare the retrieved file offset of the base section with the
2139 * base address in the description of the SDT note. If its different,
2140 * then accordingly, adjust the note location.
2141 */
Googler9398cc32022-12-02 17:21:52 +08002142 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2143 sdt_adjust_loc(tmp, shdr.sh_offset);
2144
2145 /* Adjust reference counter offset */
2146 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2147 sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
Googleraf606d22022-10-26 21:40:12 -07002148
2149 list_add_tail(&tmp->note_list, sdt_notes);
2150 return 0;
2151
Googler9398cc32022-12-02 17:21:52 +08002152out_free_args:
2153 zfree(&tmp->args);
Googleraf606d22022-10-26 21:40:12 -07002154out_free_name:
Googler9398cc32022-12-02 17:21:52 +08002155 zfree(&tmp->name);
Googleraf606d22022-10-26 21:40:12 -07002156out_free_prov:
Googler9398cc32022-12-02 17:21:52 +08002157 zfree(&tmp->provider);
Googleraf606d22022-10-26 21:40:12 -07002158out_free_note:
2159 free(tmp);
2160out_err:
2161 return ret;
2162}
2163
2164/**
2165 * construct_sdt_notes_list : constructs a list of SDT notes
2166 * @elf : elf to look into
2167 * @sdt_notes : empty list_head
2168 *
2169 * Scans the sections in 'elf' for the section
2170 * .note.stapsdt. It, then calls populate_sdt_note to find
2171 * out the SDT events and populates the 'sdt_notes'.
2172 */
2173static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2174{
2175 GElf_Ehdr ehdr;
2176 Elf_Scn *scn = NULL;
2177 Elf_Data *data;
2178 GElf_Shdr shdr;
2179 size_t shstrndx, next;
2180 GElf_Nhdr nhdr;
2181 size_t name_off, desc_off, offset;
2182 int ret = 0;
2183
2184 if (gelf_getehdr(elf, &ehdr) == NULL) {
2185 ret = -EBADF;
2186 goto out_ret;
2187 }
2188 if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2189 ret = -EBADF;
2190 goto out_ret;
2191 }
2192
2193 /* Look for the required section */
2194 scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2195 if (!scn) {
2196 ret = -ENOENT;
2197 goto out_ret;
2198 }
2199
2200 if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2201 ret = -ENOENT;
2202 goto out_ret;
2203 }
2204
2205 data = elf_getdata(scn, NULL);
2206
2207 /* Get the SDT notes */
2208 for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2209 &desc_off)) > 0; offset = next) {
2210 if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2211 !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2212 sizeof(SDT_NOTE_NAME))) {
2213 /* Check the type of the note */
2214 if (nhdr.n_type != SDT_NOTE_TYPE)
2215 goto out_ret;
2216
2217 ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2218 nhdr.n_descsz, sdt_notes);
2219 if (ret < 0)
2220 goto out_ret;
2221 }
2222 }
2223 if (list_empty(sdt_notes))
2224 ret = -ENOENT;
2225
2226out_ret:
2227 return ret;
2228}
2229
2230/**
2231 * get_sdt_note_list : Wrapper to construct a list of sdt notes
2232 * @head : empty list_head
2233 * @target : file to find SDT notes from
2234 *
2235 * This opens the file, initializes
2236 * the ELF and then calls construct_sdt_notes_list.
2237 */
2238int get_sdt_note_list(struct list_head *head, const char *target)
2239{
2240 Elf *elf;
2241 int fd, ret;
2242
2243 fd = open(target, O_RDONLY);
2244 if (fd < 0)
2245 return -EBADF;
2246
2247 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2248 if (!elf) {
2249 ret = -EBADF;
2250 goto out_close;
2251 }
2252 ret = construct_sdt_notes_list(elf, head);
2253 elf_end(elf);
2254out_close:
2255 close(fd);
2256 return ret;
2257}
2258
2259/**
2260 * cleanup_sdt_note_list : free the sdt notes' list
2261 * @sdt_notes: sdt notes' list
2262 *
2263 * Free up the SDT notes in @sdt_notes.
2264 * Returns the number of SDT notes free'd.
2265 */
2266int cleanup_sdt_note_list(struct list_head *sdt_notes)
2267{
2268 struct sdt_note *tmp, *pos;
2269 int nr_free = 0;
2270
2271 list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
Googler9398cc32022-12-02 17:21:52 +08002272 list_del_init(&pos->note_list);
2273 zfree(&pos->name);
2274 zfree(&pos->provider);
Googleraf606d22022-10-26 21:40:12 -07002275 free(pos);
2276 nr_free++;
2277 }
2278 return nr_free;
2279}
2280
2281/**
2282 * sdt_notes__get_count: Counts the number of sdt events
2283 * @start: list_head to sdt_notes list
2284 *
2285 * Returns the number of SDT notes in a list
2286 */
2287int sdt_notes__get_count(struct list_head *start)
2288{
2289 struct sdt_note *sdt_ptr;
2290 int count = 0;
2291
2292 list_for_each_entry(sdt_ptr, start, note_list)
2293 count++;
2294 return count;
2295}
2296#endif
2297
2298void symbol__elf_init(void)
2299{
2300 elf_version(EV_CURRENT);
2301}