1 // SPDX-License-Identifier: GPL-2.0
2 #include <fcntl.h>
3 #include <stdio.h>
4 #include <errno.h>
5 #include <stdlib.h>
6 #include <string.h>
7 #include <unistd.h>
8 #include <inttypes.h>
9
10 #include "dso.h"
11 #include "map.h"
12 #include "maps.h"
13 #include "symbol.h"
14 #include "symsrc.h"
15 #include "demangle-cxx.h"
16 #include "demangle-ocaml.h"
17 #include "demangle-java.h"
18 #include "demangle-rust.h"
19 #include "machine.h"
20 #include "vdso.h"
21 #include "debug.h"
22 #include "util/copyfile.h"
23 #include <linux/ctype.h>
24 #include <linux/kernel.h>
25 #include <linux/zalloc.h>
26 #include <linux/string.h>
27 #include <symbol/kallsyms.h>
28 #include <internal/lib.h>
29
30 #ifdef HAVE_LIBBFD_SUPPORT
31 #define PACKAGE 'perf'
32 #include <bfd.h>
33 #endif
34
35 #if defined(HAVE_LIBBFD_SUPPORT) || defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
36 #ifndef DMGL_PARAMS
37 #define DMGL_PARAMS (1 << 0) /* Include function args */
38 #define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */
39 #endif
40 #endif
41
42 #ifndef EM_AARCH64
43 #define EM_AARCH64 183 /* ARM 64 bit */
44 #endif
45
46 #ifndef EM_LOONGARCH
47 #define EM_LOONGARCH 258
48 #endif
49
50 #ifndef ELF32_ST_VISIBILITY
51 #define ELF32_ST_VISIBILITY(o) ((o) & 0x03)
52 #endif
53
54 /* For ELF64 the definitions are the same. */
55 #ifndef ELF64_ST_VISIBILITY
56 #define ELF64_ST_VISIBILITY(o) ELF32_ST_VISIBILITY (o)
57 #endif
58
59 /* How to extract information held in the st_other field. */
60 #ifndef GELF_ST_VISIBILITY
61 #define GELF_ST_VISIBILITY(val) ELF64_ST_VISIBILITY (val)
62 #endif
63
64 typedef Elf64_Nhdr GElf_Nhdr;
65
66
67 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
elf_getphdrnum(Elf * elf,size_t * dst)68 static int elf_getphdrnum(Elf *elf, size_t *dst)
69 {
70 GElf_Ehdr gehdr;
71 GElf_Ehdr *ehdr;
72
73 ehdr = gelf_getehdr(elf, &gehdr);
74 if (!ehdr)
75 return -1;
76
77 *dst = ehdr->e_phnum;
78
79 return 0;
80 }
81 #endif
82
83 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
elf_getshdrstrndx(Elf * elf __maybe_unused,size_t * dst __maybe_unused)84 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
85 {
86 pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
87 return -1;
88 }
89 #endif
90
91 #ifndef NT_GNU_BUILD_ID
92 #define NT_GNU_BUILD_ID 3
93 #endif
94
95 /**
96 * elf_symtab__for_each_symbol - iterate thru all the symbols
97 *
98 * @syms: struct elf_symtab instance to iterate
99 * @idx: uint32_t idx
100 * @sym: GElf_Sym iterator
101 */
102 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
103 for (idx = 0, gelf_getsym(syms, idx, &sym);\
104 idx < nr_syms; \
105 idx++, gelf_getsym(syms, idx, &sym))
106
elf_sym__type(const GElf_Sym * sym)107 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
108 {
109 return GELF_ST_TYPE(sym->st_info);
110 }
111
elf_sym__visibility(const GElf_Sym * sym)112 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
113 {
114 return GELF_ST_VISIBILITY(sym->st_other);
115 }
116
117 #ifndef STT_GNU_IFUNC
118 #define STT_GNU_IFUNC 10
119 #endif
120
elf_sym__is_function(const GElf_Sym * sym)121 static inline int elf_sym__is_function(const GElf_Sym *sym)
122 {
123 return (elf_sym__type(sym) == STT_FUNC ||
124 elf_sym__type(sym) == STT_GNU_IFUNC) &&
125 sym->st_name != 0 &&
126 sym->st_shndx != SHN_UNDEF;
127 }
128
elf_sym__is_object(const GElf_Sym * sym)129 static inline bool elf_sym__is_object(const GElf_Sym *sym)
130 {
131 return elf_sym__type(sym) == STT_OBJECT &&
132 sym->st_name != 0 &&
133 sym->st_shndx != SHN_UNDEF;
134 }
135
elf_sym__is_label(const GElf_Sym * sym)136 static inline int elf_sym__is_label(const GElf_Sym *sym)
137 {
138 return elf_sym__type(sym) == STT_NOTYPE &&
139 sym->st_name != 0 &&
140 sym->st_shndx != SHN_UNDEF &&
141 sym->st_shndx != SHN_ABS &&
142 elf_sym__visibility(sym) != STV_HIDDEN &&
143 elf_sym__visibility(sym) != STV_INTERNAL;
144 }
145
elf_sym__filter(GElf_Sym * sym)146 static bool elf_sym__filter(GElf_Sym *sym)
147 {
148 return elf_sym__is_function(sym) || elf_sym__is_object(sym);
149 }
150
elf_sym__name(const GElf_Sym * sym,const Elf_Data * symstrs)151 static inline const char *elf_sym__name(const GElf_Sym *sym,
152 const Elf_Data *symstrs)
153 {
154 return symstrs->d_buf + sym->st_name;
155 }
156
elf_sec__name(const GElf_Shdr * shdr,const Elf_Data * secstrs)157 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
158 const Elf_Data *secstrs)
159 {
160 return secstrs->d_buf + shdr->sh_name;
161 }
162
elf_sec__is_text(const GElf_Shdr * shdr,const Elf_Data * secstrs)163 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
164 const Elf_Data *secstrs)
165 {
166 return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
167 }
168
elf_sec__is_data(const GElf_Shdr * shdr,const Elf_Data * secstrs)169 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
170 const Elf_Data *secstrs)
171 {
172 return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
173 }
174
elf_sec__filter(GElf_Shdr * shdr,Elf_Data * secstrs)175 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
176 {
177 return elf_sec__is_text(shdr, secstrs) ||
178 elf_sec__is_data(shdr, secstrs);
179 }
180
elf_addr_to_index(Elf * elf,GElf_Addr addr)181 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
182 {
183 Elf_Scn *sec = NULL;
184 GElf_Shdr shdr;
185 size_t cnt = 1;
186
187 while ((sec = elf_nextscn(elf, sec)) != NULL) {
188 gelf_getshdr(sec, &shdr);
189
190 if ((addr >= shdr.sh_addr) &&
191 (addr < (shdr.sh_addr + shdr.sh_size)))
192 return cnt;
193
194 ++cnt;
195 }
196
197 return -1;
198 }
199
elf_section_by_name(Elf * elf,GElf_Ehdr * ep,GElf_Shdr * shp,const char * name,size_t * idx)200 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
201 GElf_Shdr *shp, const char *name, size_t *idx)
202 {
203 Elf_Scn *sec = NULL;
204 size_t cnt = 1;
205
206 /* ELF is corrupted/truncated, avoid calling elf_strptr. */
207 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
208 return NULL;
209
210 while ((sec = elf_nextscn(elf, sec)) != NULL) {
211 char *str;
212
213 gelf_getshdr(sec, shp);
214 str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
215 if (str && !strcmp(name, str)) {
216 if (idx)
217 *idx = cnt;
218 return sec;
219 }
220 ++cnt;
221 }
222
223 return NULL;
224 }
225
filename__has_section(const char * filename,const char * sec)226 bool filename__has_section(const char *filename, const char *sec)
227 {
228 int fd;
229 Elf *elf;
230 GElf_Ehdr ehdr;
231 GElf_Shdr shdr;
232 bool found = false;
233
234 fd = open(filename, O_RDONLY);
235 if (fd < 0)
236 return false;
237
238 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
239 if (elf == NULL)
240 goto out;
241
242 if (gelf_getehdr(elf, &ehdr) == NULL)
243 goto elf_out;
244
245 found = !!elf_section_by_name(elf, &ehdr, &shdr, sec, NULL);
246
247 elf_out:
248 elf_end(elf);
249 out:
250 close(fd);
251 return found;
252 }
253
elf_read_program_header(Elf * elf,u64 vaddr,GElf_Phdr * phdr)254 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr)
255 {
256 size_t i, phdrnum;
257 u64 sz;
258
259 if (elf_getphdrnum(elf, &phdrnum))
260 return -1;
261
262 for (i = 0; i < phdrnum; i++) {
263 if (gelf_getphdr(elf, i, phdr) == NULL)
264 return -1;
265
266 if (phdr->p_type != PT_LOAD)
267 continue;
268
269 sz = max(phdr->p_memsz, phdr->p_filesz);
270 if (!sz)
271 continue;
272
273 if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz))
274 return 0;
275 }
276
277 /* Not found any valid program header */
278 return -1;
279 }
280
want_demangle(bool is_kernel_sym)281 static bool want_demangle(bool is_kernel_sym)
282 {
283 return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
284 }
285
286 /*
287 * Demangle C++ function signature, typically replaced by demangle-cxx.cpp
288 * version.
289 */
290 #ifndef HAVE_CXA_DEMANGLE_SUPPORT
cxx_demangle_sym(const char * str __maybe_unused,bool params __maybe_unused,bool modifiers __maybe_unused)291 char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused,
292 bool modifiers __maybe_unused)
293 {
294 #ifdef HAVE_LIBBFD_SUPPORT
295 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
296
297 return bfd_demangle(NULL, str, flags);
298 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
299 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
300
301 return cplus_demangle(str, flags);
302 #else
303 return NULL;
304 #endif
305 }
306 #endif /* !HAVE_CXA_DEMANGLE_SUPPORT */
307
demangle_sym(struct dso * dso,int kmodule,const char * elf_name)308 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
309 {
310 char *demangled = NULL;
311
312 /*
313 * We need to figure out if the object was created from C++ sources
314 * DWARF DW_compile_unit has this, but we don't always have access
315 * to it...
316 */
317 if (!want_demangle(dso__kernel(dso) || kmodule))
318 return demangled;
319
320 demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0);
321 if (demangled == NULL) {
322 demangled = ocaml_demangle_sym(elf_name);
323 if (demangled == NULL) {
324 demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
325 }
326 }
327 else if (rust_is_mangled(demangled))
328 /*
329 * Input to Rust demangling is the BFD-demangled
330 * name which it Rust-demangles in place.
331 */
332 rust_demangle_sym(demangled);
333
334 return demangled;
335 }
336
337 struct rel_info {
338 u32 nr_entries;
339 u32 *sorted;
340 bool is_rela;
341 Elf_Data *reldata;
342 GElf_Rela rela;
343 GElf_Rel rel;
344 };
345
get_rel_symidx(struct rel_info * ri,u32 idx)346 static u32 get_rel_symidx(struct rel_info *ri, u32 idx)
347 {
348 idx = ri->sorted ? ri->sorted[idx] : idx;
349 if (ri->is_rela) {
350 gelf_getrela(ri->reldata, idx, &ri->rela);
351 return GELF_R_SYM(ri->rela.r_info);
352 }
353 gelf_getrel(ri->reldata, idx, &ri->rel);
354 return GELF_R_SYM(ri->rel.r_info);
355 }
356
get_rel_offset(struct rel_info * ri,u32 x)357 static u64 get_rel_offset(struct rel_info *ri, u32 x)
358 {
359 if (ri->is_rela) {
360 GElf_Rela rela;
361
362 gelf_getrela(ri->reldata, x, &rela);
363 return rela.r_offset;
364 } else {
365 GElf_Rel rel;
366
367 gelf_getrel(ri->reldata, x, &rel);
368 return rel.r_offset;
369 }
370 }
371
rel_cmp(const void * a,const void * b,void * r)372 static int rel_cmp(const void *a, const void *b, void *r)
373 {
374 struct rel_info *ri = r;
375 u64 a_offset = get_rel_offset(ri, *(const u32 *)a);
376 u64 b_offset = get_rel_offset(ri, *(const u32 *)b);
377
378 return a_offset < b_offset ? -1 : (a_offset > b_offset ? 1 : 0);
379 }
380
sort_rel(struct rel_info * ri)381 static int sort_rel(struct rel_info *ri)
382 {
383 size_t sz = sizeof(ri->sorted[0]);
384 u32 i;
385
386 ri->sorted = calloc(ri->nr_entries, sz);
387 if (!ri->sorted)
388 return -1;
389 for (i = 0; i < ri->nr_entries; i++)
390 ri->sorted[i] = i;
391 qsort_r(ri->sorted, ri->nr_entries, sz, rel_cmp, ri);
392 return 0;
393 }
394
395 /*
396 * For x86_64, the GNU linker is putting IFUNC information in the relocation
397 * addend.
398 */
addend_may_be_ifunc(GElf_Ehdr * ehdr,struct rel_info * ri)399 static bool addend_may_be_ifunc(GElf_Ehdr *ehdr, struct rel_info *ri)
400 {
401 return ehdr->e_machine == EM_X86_64 && ri->is_rela &&
402 GELF_R_TYPE(ri->rela.r_info) == R_X86_64_IRELATIVE;
403 }
404
get_ifunc_name(Elf * elf,struct dso * dso,GElf_Ehdr * ehdr,struct rel_info * ri,char * buf,size_t buf_sz)405 static bool get_ifunc_name(Elf *elf, struct dso *dso, GElf_Ehdr *ehdr,
406 struct rel_info *ri, char *buf, size_t buf_sz)
407 {
408 u64 addr = ri->rela.r_addend;
409 struct symbol *sym;
410 GElf_Phdr phdr;
411
412 if (!addend_may_be_ifunc(ehdr, ri))
413 return false;
414
415 if (elf_read_program_header(elf, addr, &phdr))
416 return false;
417
418 addr -= phdr.p_vaddr - phdr.p_offset;
419
420 sym = dso__find_symbol_nocache(dso, addr);
421
422 /* Expecting the address to be an IFUNC or IFUNC alias */
423 if (!sym || sym->start != addr || (sym->type != STT_GNU_IFUNC && !sym->ifunc_alias))
424 return false;
425
426 snprintf(buf, buf_sz, "%s@plt", sym->name);
427
428 return true;
429 }
430
exit_rel(struct rel_info * ri)431 static void exit_rel(struct rel_info *ri)
432 {
433 zfree(&ri->sorted);
434 }
435
get_plt_sizes(struct dso * dso,GElf_Ehdr * ehdr,GElf_Shdr * shdr_plt,u64 * plt_header_size,u64 * plt_entry_size)436 static bool get_plt_sizes(struct dso *dso, GElf_Ehdr *ehdr, GElf_Shdr *shdr_plt,
437 u64 *plt_header_size, u64 *plt_entry_size)
438 {
439 switch (ehdr->e_machine) {
440 case EM_ARM:
441 *plt_header_size = 20;
442 *plt_entry_size = 12;
443 return true;
444 case EM_AARCH64:
445 *plt_header_size = 32;
446 *plt_entry_size = 16;
447 return true;
448 case EM_LOONGARCH:
449 *plt_header_size = 32;
450 *plt_entry_size = 16;
451 return true;
452 case EM_SPARC:
453 *plt_header_size = 48;
454 *plt_entry_size = 12;
455 return true;
456 case EM_SPARCV9:
457 *plt_header_size = 128;
458 *plt_entry_size = 32;
459 return true;
460 case EM_386:
461 case EM_X86_64:
462 *plt_entry_size = shdr_plt->sh_entsize;
463 /* Size is 8 or 16, if not, assume alignment indicates size */
464 if (*plt_entry_size != 8 && *plt_entry_size != 16)
465 *plt_entry_size = shdr_plt->sh_addralign == 8 ? 8 : 16;
466 *plt_header_size = *plt_entry_size;
467 break;
468 default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
469 *plt_header_size = shdr_plt->sh_entsize;
470 *plt_entry_size = shdr_plt->sh_entsize;
471 break;
472 }
473 if (*plt_entry_size)
474 return true;
475 pr_debug("Missing PLT entry size for %s\n", dso__long_name(dso));
476 return false;
477 }
478
machine_is_x86(GElf_Half e_machine)479 static bool machine_is_x86(GElf_Half e_machine)
480 {
481 return e_machine == EM_386 || e_machine == EM_X86_64;
482 }
483
484 struct rela_dyn {
485 GElf_Addr offset;
486 u32 sym_idx;
487 };
488
489 struct rela_dyn_info {
490 struct dso *dso;
491 Elf_Data *plt_got_data;
492 u32 nr_entries;
493 struct rela_dyn *sorted;
494 Elf_Data *dynsym_data;
495 Elf_Data *dynstr_data;
496 Elf_Data *rela_dyn_data;
497 };
498
exit_rela_dyn(struct rela_dyn_info * di)499 static void exit_rela_dyn(struct rela_dyn_info *di)
500 {
501 zfree(&di->sorted);
502 }
503
cmp_offset(const void * a,const void * b)504 static int cmp_offset(const void *a, const void *b)
505 {
506 const struct rela_dyn *va = a;
507 const struct rela_dyn *vb = b;
508
509 return va->offset < vb->offset ? -1 : (va->offset > vb->offset ? 1 : 0);
510 }
511
sort_rela_dyn(struct rela_dyn_info * di)512 static int sort_rela_dyn(struct rela_dyn_info *di)
513 {
514 u32 i, n;
515
516 di->sorted = calloc(di->nr_entries, sizeof(di->sorted[0]));
517 if (!di->sorted)
518 return -1;
519
520 /* Get data for sorting: the offset and symbol index */
521 for (i = 0, n = 0; i < di->nr_entries; i++) {
522 GElf_Rela rela;
523 u32 sym_idx;
524
525 gelf_getrela(di->rela_dyn_data, i, &rela);
526 sym_idx = GELF_R_SYM(rela.r_info);
527 if (sym_idx) {
528 di->sorted[n].sym_idx = sym_idx;
529 di->sorted[n].offset = rela.r_offset;
530 n += 1;
531 }
532 }
533
534 /* Sort by offset */
535 di->nr_entries = n;
536 qsort(di->sorted, n, sizeof(di->sorted[0]), cmp_offset);
537
538 return 0;
539 }
540
get_rela_dyn_info(Elf * elf,GElf_Ehdr * ehdr,struct rela_dyn_info * di,Elf_Scn * scn)541 static void get_rela_dyn_info(Elf *elf, GElf_Ehdr *ehdr, struct rela_dyn_info *di, Elf_Scn *scn)
542 {
543 GElf_Shdr rela_dyn_shdr;
544 GElf_Shdr shdr;
545
546 di->plt_got_data = elf_getdata(scn, NULL);
547
548 scn = elf_section_by_name(elf, ehdr, &rela_dyn_shdr, ".rela.dyn", NULL);
549 if (!scn || !rela_dyn_shdr.sh_link || !rela_dyn_shdr.sh_entsize)
550 return;
551
552 di->nr_entries = rela_dyn_shdr.sh_size / rela_dyn_shdr.sh_entsize;
553 di->rela_dyn_data = elf_getdata(scn, NULL);
554
555 scn = elf_getscn(elf, rela_dyn_shdr.sh_link);
556 if (!scn || !gelf_getshdr(scn, &shdr) || !shdr.sh_link)
557 return;
558
559 di->dynsym_data = elf_getdata(scn, NULL);
560 di->dynstr_data = elf_getdata(elf_getscn(elf, shdr.sh_link), NULL);
561
562 if (!di->plt_got_data || !di->dynstr_data || !di->dynsym_data || !di->rela_dyn_data)
563 return;
564
565 /* Sort into offset order */
566 sort_rela_dyn(di);
567 }
568
569 /* Get instruction displacement from a plt entry for x86_64 */
get_x86_64_plt_disp(const u8 * p)570 static u32 get_x86_64_plt_disp(const u8 *p)
571 {
572 u8 endbr64[] = {0xf3, 0x0f, 0x1e, 0xfa};
573 int n = 0;
574
575 /* Skip endbr64 */
576 if (!memcmp(p, endbr64, sizeof(endbr64)))
577 n += sizeof(endbr64);
578 /* Skip bnd prefix */
579 if (p[n] == 0xf2)
580 n += 1;
581 /* jmp with 4-byte displacement */
582 if (p[n] == 0xff && p[n + 1] == 0x25) {
583 u32 disp;
584
585 n += 2;
586 /* Also add offset from start of entry to end of instruction */
587 memcpy(&disp, p + n, sizeof(disp));
588 return n + 4 + le32toh(disp);
589 }
590 return 0;
591 }
592
get_plt_got_name(GElf_Shdr * shdr,size_t i,struct rela_dyn_info * di,char * buf,size_t buf_sz)593 static bool get_plt_got_name(GElf_Shdr *shdr, size_t i,
594 struct rela_dyn_info *di,
595 char *buf, size_t buf_sz)
596 {
597 struct rela_dyn vi, *vr;
598 const char *sym_name;
599 char *demangled;
600 GElf_Sym sym;
601 bool result;
602 u32 disp;
603
604 if (!di->sorted)
605 return false;
606
607 disp = get_x86_64_plt_disp(di->plt_got_data->d_buf + i);
608 if (!disp)
609 return false;
610
611 /* Compute target offset of the .plt.got entry */
612 vi.offset = shdr->sh_offset + di->plt_got_data->d_off + i + disp;
613
614 /* Find that offset in .rela.dyn (sorted by offset) */
615 vr = bsearch(&vi, di->sorted, di->nr_entries, sizeof(di->sorted[0]), cmp_offset);
616 if (!vr)
617 return false;
618
619 /* Get the associated symbol */
620 gelf_getsym(di->dynsym_data, vr->sym_idx, &sym);
621 sym_name = elf_sym__name(&sym, di->dynstr_data);
622 demangled = demangle_sym(di->dso, 0, sym_name);
623 if (demangled != NULL)
624 sym_name = demangled;
625
626 snprintf(buf, buf_sz, "%s@plt", sym_name);
627
628 result = *sym_name;
629
630 free(demangled);
631
632 return result;
633 }
634
dso__synthesize_plt_got_symbols(struct dso * dso,Elf * elf,GElf_Ehdr * ehdr,char * buf,size_t buf_sz)635 static int dso__synthesize_plt_got_symbols(struct dso *dso, Elf *elf,
636 GElf_Ehdr *ehdr,
637 char *buf, size_t buf_sz)
638 {
639 struct rela_dyn_info di = { .dso = dso };
640 struct symbol *sym;
641 GElf_Shdr shdr;
642 Elf_Scn *scn;
643 int err = -1;
644 size_t i;
645
646 scn = elf_section_by_name(elf, ehdr, &shdr, ".plt.got", NULL);
647 if (!scn || !shdr.sh_entsize)
648 return 0;
649
650 if (ehdr->e_machine == EM_X86_64)
651 get_rela_dyn_info(elf, ehdr, &di, scn);
652
653 for (i = 0; i < shdr.sh_size; i += shdr.sh_entsize) {
654 if (!get_plt_got_name(&shdr, i, &di, buf, buf_sz))
655 snprintf(buf, buf_sz, "offset_%#" PRIx64 "@plt", (u64)shdr.sh_offset + i);
656 sym = symbol__new(shdr.sh_offset + i, shdr.sh_entsize, STB_GLOBAL, STT_FUNC, buf);
657 if (!sym)
658 goto out;
659 symbols__insert(dso__symbols(dso), sym);
660 }
661 err = 0;
662 out:
663 exit_rela_dyn(&di);
664 return err;
665 }
666
667 /*
668 * We need to check if we have a .dynsym, so that we can handle the
669 * .plt, synthesizing its symbols, that aren't on the symtabs (be it
670 * .dynsym or .symtab).
671 * And always look at the original dso, not at debuginfo packages, that
672 * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
673 */
dso__synthesize_plt_symbols(struct dso * dso,struct symsrc * ss)674 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
675 {
676 uint32_t idx;
677 GElf_Sym sym;
678 u64 plt_offset, plt_header_size, plt_entry_size;
679 GElf_Shdr shdr_plt, plt_sec_shdr;
680 struct symbol *f, *plt_sym;
681 GElf_Shdr shdr_rel_plt, shdr_dynsym;
682 Elf_Data *syms, *symstrs;
683 Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
684 GElf_Ehdr ehdr;
685 char sympltname[1024];
686 Elf *elf;
687 int nr = 0, err = -1;
688 struct rel_info ri = { .is_rela = false };
689 bool lazy_plt;
690
691 elf = ss->elf;
692 ehdr = ss->ehdr;
693
694 if (!elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL))
695 return 0;
696
697 /*
698 * A symbol from a previous section (e.g. .init) can have been expanded
699 * by symbols__fixup_end() to overlap .plt. Truncate it before adding
700 * a symbol for .plt header.
701 */
702 f = dso__find_symbol_nocache(dso, shdr_plt.sh_offset);
703 if (f && f->start < shdr_plt.sh_offset && f->end > shdr_plt.sh_offset)
704 f->end = shdr_plt.sh_offset;
705
706 if (!get_plt_sizes(dso, &ehdr, &shdr_plt, &plt_header_size, &plt_entry_size))
707 return 0;
708
709 /* Add a symbol for .plt header */
710 plt_sym = symbol__new(shdr_plt.sh_offset, plt_header_size, STB_GLOBAL, STT_FUNC, ".plt");
711 if (!plt_sym)
712 goto out_elf_end;
713 symbols__insert(dso__symbols(dso), plt_sym);
714
715 /* Only x86 has .plt.got */
716 if (machine_is_x86(ehdr.e_machine) &&
717 dso__synthesize_plt_got_symbols(dso, elf, &ehdr, sympltname, sizeof(sympltname)))
718 goto out_elf_end;
719
720 /* Only x86 has .plt.sec */
721 if (machine_is_x86(ehdr.e_machine) &&
722 elf_section_by_name(elf, &ehdr, &plt_sec_shdr, ".plt.sec", NULL)) {
723 if (!get_plt_sizes(dso, &ehdr, &plt_sec_shdr, &plt_header_size, &plt_entry_size))
724 return 0;
725 /* Extend .plt symbol to entire .plt */
726 plt_sym->end = plt_sym->start + shdr_plt.sh_size;
727 /* Use .plt.sec offset */
728 plt_offset = plt_sec_shdr.sh_offset;
729 lazy_plt = false;
730 } else {
731 plt_offset = shdr_plt.sh_offset;
732 lazy_plt = true;
733 }
734
735 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
736 ".rela.plt", NULL);
737 if (scn_plt_rel == NULL) {
738 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
739 ".rel.plt", NULL);
740 if (scn_plt_rel == NULL)
741 return 0;
742 }
743
744 if (shdr_rel_plt.sh_type != SHT_RELA &&
745 shdr_rel_plt.sh_type != SHT_REL)
746 return 0;
747
748 if (!shdr_rel_plt.sh_link)
749 return 0;
750
751 if (shdr_rel_plt.sh_link == ss->dynsym_idx) {
752 scn_dynsym = ss->dynsym;
753 shdr_dynsym = ss->dynshdr;
754 } else if (shdr_rel_plt.sh_link == ss->symtab_idx) {
755 /*
756 * A static executable can have a .plt due to IFUNCs, in which
757 * case .symtab is used not .dynsym.
758 */
759 scn_dynsym = ss->symtab;
760 shdr_dynsym = ss->symshdr;
761 } else {
762 goto out_elf_end;
763 }
764
765 if (!scn_dynsym)
766 return 0;
767
768 /*
769 * Fetch the relocation section to find the idxes to the GOT
770 * and the symbols in the .dynsym they refer to.
771 */
772 ri.reldata = elf_getdata(scn_plt_rel, NULL);
773 if (!ri.reldata)
774 goto out_elf_end;
775
776 syms = elf_getdata(scn_dynsym, NULL);
777 if (syms == NULL)
778 goto out_elf_end;
779
780 scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
781 if (scn_symstrs == NULL)
782 goto out_elf_end;
783
784 symstrs = elf_getdata(scn_symstrs, NULL);
785 if (symstrs == NULL)
786 goto out_elf_end;
787
788 if (symstrs->d_size == 0)
789 goto out_elf_end;
790
791 ri.nr_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
792
793 ri.is_rela = shdr_rel_plt.sh_type == SHT_RELA;
794
795 if (lazy_plt) {
796 /*
797 * Assume a .plt with the same number of entries as the number
798 * of relocation entries is not lazy and does not have a header.
799 */
800 if (ri.nr_entries * plt_entry_size == shdr_plt.sh_size)
801 dso__delete_symbol(dso, plt_sym);
802 else
803 plt_offset += plt_header_size;
804 }
805
806 /*
807 * x86 doesn't insert IFUNC relocations in .plt order, so sort to get
808 * back in order.
809 */
810 if (machine_is_x86(ehdr.e_machine) && sort_rel(&ri))
811 goto out_elf_end;
812
813 for (idx = 0; idx < ri.nr_entries; idx++) {
814 const char *elf_name = NULL;
815 char *demangled = NULL;
816
817 gelf_getsym(syms, get_rel_symidx(&ri, idx), &sym);
818
819 elf_name = elf_sym__name(&sym, symstrs);
820 demangled = demangle_sym(dso, 0, elf_name);
821 if (demangled)
822 elf_name = demangled;
823 if (*elf_name)
824 snprintf(sympltname, sizeof(sympltname), "%s@plt", elf_name);
825 else if (!get_ifunc_name(elf, dso, &ehdr, &ri, sympltname, sizeof(sympltname)))
826 snprintf(sympltname, sizeof(sympltname),
827 "offset_%#" PRIx64 "@plt", plt_offset);
828 free(demangled);
829
830 f = symbol__new(plt_offset, plt_entry_size, STB_GLOBAL, STT_FUNC, sympltname);
831 if (!f)
832 goto out_elf_end;
833
834 plt_offset += plt_entry_size;
835 symbols__insert(dso__symbols(dso), f);
836 ++nr;
837 }
838
839 err = 0;
840 out_elf_end:
841 exit_rel(&ri);
842 if (err == 0)
843 return nr;
844 pr_debug("%s: problems reading %s PLT info.\n",
845 __func__, dso__long_name(dso));
846 return 0;
847 }
848
dso__demangle_sym(struct dso * dso,int kmodule,const char * elf_name)849 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
850 {
851 return demangle_sym(dso, kmodule, elf_name);
852 }
853
854 /*
855 * Align offset to 4 bytes as needed for note name and descriptor data.
856 */
857 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
858
elf_read_build_id(Elf * elf,void * bf,size_t size)859 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
860 {
861 int err = -1;
862 GElf_Ehdr ehdr;
863 GElf_Shdr shdr;
864 Elf_Data *data;
865 Elf_Scn *sec;
866 Elf_Kind ek;
867 void *ptr;
868
869 if (size < BUILD_ID_SIZE)
870 goto out;
871
872 ek = elf_kind(elf);
873 if (ek != ELF_K_ELF)
874 goto out;
875
876 if (gelf_getehdr(elf, &ehdr) == NULL) {
877 pr_err("%s: cannot get elf header.\n", __func__);
878 goto out;
879 }
880
881 /*
882 * Check following sections for notes:
883 * '.note.gnu.build-id'
884 * '.notes'
885 * '.note' (VDSO specific)
886 */
887 do {
888 sec = elf_section_by_name(elf, &ehdr, &shdr,
889 ".note.gnu.build-id", NULL);
890 if (sec)
891 break;
892
893 sec = elf_section_by_name(elf, &ehdr, &shdr,
894 ".notes", NULL);
895 if (sec)
896 break;
897
898 sec = elf_section_by_name(elf, &ehdr, &shdr,
899 ".note", NULL);
900 if (sec)
901 break;
902
903 return err;
904
905 } while (0);
906
907 data = elf_getdata(sec, NULL);
908 if (data == NULL)
909 goto out;
910
911 ptr = data->d_buf;
912 while (ptr < (data->d_buf + data->d_size)) {
913 GElf_Nhdr *nhdr = ptr;
914 size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
915 descsz = NOTE_ALIGN(nhdr->n_descsz);
916 const char *name;
917
918 ptr += sizeof(*nhdr);
919 name = ptr;
920 ptr += namesz;
921 if (nhdr->n_type == NT_GNU_BUILD_ID &&
922 nhdr->n_namesz == sizeof("GNU")) {
923 if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
924 size_t sz = min(size, descsz);
925 memcpy(bf, ptr, sz);
926 memset(bf + sz, 0, size - sz);
927 err = sz;
928 break;
929 }
930 }
931 ptr += descsz;
932 }
933
934 out:
935 return err;
936 }
937
938 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
939
read_build_id(const char * filename,struct build_id * bid)940 static int read_build_id(const char *filename, struct build_id *bid)
941 {
942 size_t size = sizeof(bid->data);
943 int err = -1;
944 bfd *abfd;
945
946 abfd = bfd_openr(filename, NULL);
947 if (!abfd)
948 return -1;
949
950 if (!bfd_check_format(abfd, bfd_object)) {
951 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
952 goto out_close;
953 }
954
955 if (!abfd->build_id || abfd->build_id->size > size)
956 goto out_close;
957
958 memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
959 memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
960 err = bid->size = abfd->build_id->size;
961
962 out_close:
963 bfd_close(abfd);
964 return err;
965 }
966
967 #else // HAVE_LIBBFD_BUILDID_SUPPORT
968
read_build_id(const char * filename,struct build_id * bid)969 static int read_build_id(const char *filename, struct build_id *bid)
970 {
971 size_t size = sizeof(bid->data);
972 int fd, err = -1;
973 Elf *elf;
974
975 if (size < BUILD_ID_SIZE)
976 goto out;
977
978 fd = open(filename, O_RDONLY);
979 if (fd < 0)
980 goto out;
981
982 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
983 if (elf == NULL) {
984 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
985 goto out_close;
986 }
987
988 err = elf_read_build_id(elf, bid->data, size);
989 if (err > 0)
990 bid->size = err;
991
992 elf_end(elf);
993 out_close:
994 close(fd);
995 out:
996 return err;
997 }
998
999 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
1000
filename__read_build_id(const char * filename,struct build_id * bid)1001 int filename__read_build_id(const char *filename, struct build_id *bid)
1002 {
1003 struct kmod_path m = { .name = NULL, };
1004 char path[PATH_MAX];
1005 int err;
1006
1007 if (!filename)
1008 return -EFAULT;
1009
1010 err = kmod_path__parse(&m, filename);
1011 if (err)
1012 return -1;
1013
1014 if (m.comp) {
1015 int error = 0, fd;
1016
1017 fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
1018 if (fd < 0) {
1019 pr_debug("Failed to decompress (error %d) %s\n",
1020 error, filename);
1021 return -1;
1022 }
1023 close(fd);
1024 filename = path;
1025 }
1026
1027 err = read_build_id(filename, bid);
1028
1029 if (m.comp)
1030 unlink(filename);
1031 return err;
1032 }
1033
sysfs__read_build_id(const char * filename,struct build_id * bid)1034 int sysfs__read_build_id(const char *filename, struct build_id *bid)
1035 {
1036 size_t size = sizeof(bid->data);
1037 int fd, err = -1;
1038
1039 fd = open(filename, O_RDONLY);
1040 if (fd < 0)
1041 goto out;
1042
1043 while (1) {
1044 char bf[BUFSIZ];
1045 GElf_Nhdr nhdr;
1046 size_t namesz, descsz;
1047
1048 if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
1049 break;
1050
1051 namesz = NOTE_ALIGN(nhdr.n_namesz);
1052 descsz = NOTE_ALIGN(nhdr.n_descsz);
1053 if (nhdr.n_type == NT_GNU_BUILD_ID &&
1054 nhdr.n_namesz == sizeof("GNU")) {
1055 if (read(fd, bf, namesz) != (ssize_t)namesz)
1056 break;
1057 if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
1058 size_t sz = min(descsz, size);
1059 if (read(fd, bid->data, sz) == (ssize_t)sz) {
1060 memset(bid->data + sz, 0, size - sz);
1061 bid->size = sz;
1062 err = 0;
1063 break;
1064 }
1065 } else if (read(fd, bf, descsz) != (ssize_t)descsz)
1066 break;
1067 } else {
1068 int n = namesz + descsz;
1069
1070 if (n > (int)sizeof(bf)) {
1071 n = sizeof(bf);
1072 pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
1073 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
1074 }
1075 if (read(fd, bf, n) != n)
1076 break;
1077 }
1078 }
1079 close(fd);
1080 out:
1081 return err;
1082 }
1083
1084 #ifdef HAVE_LIBBFD_SUPPORT
1085
filename__read_debuglink(const char * filename,char * debuglink,size_t size)1086 int filename__read_debuglink(const char *filename, char *debuglink,
1087 size_t size)
1088 {
1089 int err = -1;
1090 asection *section;
1091 bfd *abfd;
1092
1093 abfd = bfd_openr(filename, NULL);
1094 if (!abfd)
1095 return -1;
1096
1097 if (!bfd_check_format(abfd, bfd_object)) {
1098 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
1099 goto out_close;
1100 }
1101
1102 section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
1103 if (!section)
1104 goto out_close;
1105
1106 if (section->size > size)
1107 goto out_close;
1108
1109 if (!bfd_get_section_contents(abfd, section, debuglink, 0,
1110 section->size))
1111 goto out_close;
1112
1113 err = 0;
1114
1115 out_close:
1116 bfd_close(abfd);
1117 return err;
1118 }
1119
1120 #else
1121
filename__read_debuglink(const char * filename,char * debuglink,size_t size)1122 int filename__read_debuglink(const char *filename, char *debuglink,
1123 size_t size)
1124 {
1125 int fd, err = -1;
1126 Elf *elf;
1127 GElf_Ehdr ehdr;
1128 GElf_Shdr shdr;
1129 Elf_Data *data;
1130 Elf_Scn *sec;
1131 Elf_Kind ek;
1132
1133 fd = open(filename, O_RDONLY);
1134 if (fd < 0)
1135 goto out;
1136
1137 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1138 if (elf == NULL) {
1139 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1140 goto out_close;
1141 }
1142
1143 ek = elf_kind(elf);
1144 if (ek != ELF_K_ELF)
1145 goto out_elf_end;
1146
1147 if (gelf_getehdr(elf, &ehdr) == NULL) {
1148 pr_err("%s: cannot get elf header.\n", __func__);
1149 goto out_elf_end;
1150 }
1151
1152 sec = elf_section_by_name(elf, &ehdr, &shdr,
1153 ".gnu_debuglink", NULL);
1154 if (sec == NULL)
1155 goto out_elf_end;
1156
1157 data = elf_getdata(sec, NULL);
1158 if (data == NULL)
1159 goto out_elf_end;
1160
1161 /* the start of this section is a zero-terminated string */
1162 strncpy(debuglink, data->d_buf, size);
1163
1164 err = 0;
1165
1166 out_elf_end:
1167 elf_end(elf);
1168 out_close:
1169 close(fd);
1170 out:
1171 return err;
1172 }
1173
1174 #endif
1175
dso__swap_init(struct dso * dso,unsigned char eidata)1176 static int dso__swap_init(struct dso *dso, unsigned char eidata)
1177 {
1178 static unsigned int const endian = 1;
1179
1180 dso__set_needs_swap(dso, DSO_SWAP__NO);
1181
1182 switch (eidata) {
1183 case ELFDATA2LSB:
1184 /* We are big endian, DSO is little endian. */
1185 if (*(unsigned char const *)&endian != 1)
1186 dso__set_needs_swap(dso, DSO_SWAP__YES);
1187 break;
1188
1189 case ELFDATA2MSB:
1190 /* We are little endian, DSO is big endian. */
1191 if (*(unsigned char const *)&endian != 0)
1192 dso__set_needs_swap(dso, DSO_SWAP__YES);
1193 break;
1194
1195 default:
1196 pr_err("unrecognized DSO data encoding %d\n", eidata);
1197 return -EINVAL;
1198 }
1199
1200 return 0;
1201 }
1202
symsrc__possibly_runtime(struct symsrc * ss)1203 bool symsrc__possibly_runtime(struct symsrc *ss)
1204 {
1205 return ss->dynsym || ss->opdsec;
1206 }
1207
symsrc__has_symtab(struct symsrc * ss)1208 bool symsrc__has_symtab(struct symsrc *ss)
1209 {
1210 return ss->symtab != NULL;
1211 }
1212
symsrc__destroy(struct symsrc * ss)1213 void symsrc__destroy(struct symsrc *ss)
1214 {
1215 zfree(&ss->name);
1216 elf_end(ss->elf);
1217 close(ss->fd);
1218 }
1219
elf__needs_adjust_symbols(GElf_Ehdr ehdr)1220 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
1221 {
1222 /*
1223 * Usually vmlinux is an ELF file with type ET_EXEC for most
1224 * architectures; except Arm64 kernel is linked with option
1225 * '-share', so need to check type ET_DYN.
1226 */
1227 return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
1228 ehdr.e_type == ET_DYN;
1229 }
1230
symsrc__init(struct symsrc * ss,struct dso * dso,const char * name,enum dso_binary_type type)1231 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
1232 enum dso_binary_type type)
1233 {
1234 GElf_Ehdr ehdr;
1235 Elf *elf;
1236 int fd;
1237
1238 if (dso__needs_decompress(dso)) {
1239 fd = dso__decompress_kmodule_fd(dso, name);
1240 if (fd < 0)
1241 return -1;
1242
1243 type = dso__symtab_type(dso);
1244 } else {
1245 fd = open(name, O_RDONLY);
1246 if (fd < 0) {
1247 *dso__load_errno(dso) = errno;
1248 return -1;
1249 }
1250 }
1251
1252 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1253 if (elf == NULL) {
1254 pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1255 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1256 goto out_close;
1257 }
1258
1259 if (gelf_getehdr(elf, &ehdr) == NULL) {
1260 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INVALID_ELF;
1261 pr_debug("%s: cannot get elf header.\n", __func__);
1262 goto out_elf_end;
1263 }
1264
1265 if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
1266 *dso__load_errno(dso) = DSO_LOAD_ERRNO__INTERNAL_ERROR;
1267 goto out_elf_end;
1268 }
1269
1270 /* Always reject images with a mismatched build-id: */
1271 if (dso__has_build_id(dso) && !symbol_conf.ignore_vmlinux_buildid) {
1272 u8 build_id[BUILD_ID_SIZE];
1273 struct build_id bid;
1274 int size;
1275
1276 size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
1277 if (size <= 0) {
1278 *dso__load_errno(dso) = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
1279 goto out_elf_end;
1280 }
1281
1282 build_id__init(&bid, build_id, size);
1283 if (!dso__build_id_equal(dso, &bid)) {
1284 pr_debug("%s: build id mismatch for %s.\n", __func__, name);
1285 *dso__load_errno(dso) = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
1286 goto out_elf_end;
1287 }
1288 }
1289
1290 ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1291
1292 ss->symtab_idx = 0;
1293 ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
1294 &ss->symtab_idx);
1295 if (ss->symshdr.sh_type != SHT_SYMTAB)
1296 ss->symtab = NULL;
1297
1298 ss->dynsym_idx = 0;
1299 ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
1300 &ss->dynsym_idx);
1301 if (ss->dynshdr.sh_type != SHT_DYNSYM)
1302 ss->dynsym = NULL;
1303
1304 ss->opdidx = 0;
1305 ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
1306 &ss->opdidx);
1307 if (ss->opdshdr.sh_type != SHT_PROGBITS)
1308 ss->opdsec = NULL;
1309
1310 if (dso__kernel(dso) == DSO_SPACE__USER)
1311 ss->adjust_symbols = true;
1312 else
1313 ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
1314
1315 ss->name = strdup(name);
1316 if (!ss->name) {
1317 *dso__load_errno(dso) = errno;
1318 goto out_elf_end;
1319 }
1320
1321 ss->elf = elf;
1322 ss->fd = fd;
1323 ss->ehdr = ehdr;
1324 ss->type = type;
1325
1326 return 0;
1327
1328 out_elf_end:
1329 elf_end(elf);
1330 out_close:
1331 close(fd);
1332 return -1;
1333 }
1334
is_exe_text(int flags)1335 static bool is_exe_text(int flags)
1336 {
1337 return (flags & (SHF_ALLOC | SHF_EXECINSTR)) == (SHF_ALLOC | SHF_EXECINSTR);
1338 }
1339
1340 /*
1341 * Some executable module sections like .noinstr.text might be laid out with
1342 * .text so they can use the same mapping (memory address to file offset).
1343 * Check if that is the case. Refer to kernel layout_sections(). Return the
1344 * maximum offset.
1345 */
max_text_section(Elf * elf,GElf_Ehdr * ehdr)1346 static u64 max_text_section(Elf *elf, GElf_Ehdr *ehdr)
1347 {
1348 Elf_Scn *sec = NULL;
1349 GElf_Shdr shdr;
1350 u64 offs = 0;
1351
1352 /* Doesn't work for some arch */
1353 if (ehdr->e_machine == EM_PARISC ||
1354 ehdr->e_machine == EM_ALPHA)
1355 return 0;
1356
1357 /* ELF is corrupted/truncated, avoid calling elf_strptr. */
1358 if (!elf_rawdata(elf_getscn(elf, ehdr->e_shstrndx), NULL))
1359 return 0;
1360
1361 while ((sec = elf_nextscn(elf, sec)) != NULL) {
1362 char *sec_name;
1363
1364 if (!gelf_getshdr(sec, &shdr))
1365 break;
1366
1367 if (!is_exe_text(shdr.sh_flags))
1368 continue;
1369
1370 /* .init and .exit sections are not placed with .text */
1371 sec_name = elf_strptr(elf, ehdr->e_shstrndx, shdr.sh_name);
1372 if (!sec_name ||
1373 strstarts(sec_name, ".init") ||
1374 strstarts(sec_name, ".exit"))
1375 break;
1376
1377 /* Must be next to previous, assumes .text is first */
1378 if (offs && PERF_ALIGN(offs, shdr.sh_addralign ?: 1) != shdr.sh_offset)
1379 break;
1380
1381 offs = shdr.sh_offset + shdr.sh_size;
1382 }
1383
1384 return offs;
1385 }
1386
1387 /**
1388 * ref_reloc_sym_not_found - has kernel relocation symbol been found.
1389 * @kmap: kernel maps and relocation reference symbol
1390 *
1391 * This function returns %true if we are dealing with the kernel maps and the
1392 * relocation reference symbol has not yet been found. Otherwise %false is
1393 * returned.
1394 */
ref_reloc_sym_not_found(struct kmap * kmap)1395 static bool ref_reloc_sym_not_found(struct kmap *kmap)
1396 {
1397 return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
1398 !kmap->ref_reloc_sym->unrelocated_addr;
1399 }
1400
1401 /**
1402 * ref_reloc - kernel relocation offset.
1403 * @kmap: kernel maps and relocation reference symbol
1404 *
1405 * This function returns the offset of kernel addresses as determined by using
1406 * the relocation reference symbol i.e. if the kernel has not been relocated
1407 * then the return value is zero.
1408 */
ref_reloc(struct kmap * kmap)1409 static u64 ref_reloc(struct kmap *kmap)
1410 {
1411 if (kmap && kmap->ref_reloc_sym &&
1412 kmap->ref_reloc_sym->unrelocated_addr)
1413 return kmap->ref_reloc_sym->addr -
1414 kmap->ref_reloc_sym->unrelocated_addr;
1415 return 0;
1416 }
1417
arch__sym_update(struct symbol * s __maybe_unused,GElf_Sym * sym __maybe_unused)1418 void __weak arch__sym_update(struct symbol *s __maybe_unused,
1419 GElf_Sym *sym __maybe_unused) { }
1420
dso__process_kernel_symbol(struct dso * dso,struct map * map,GElf_Sym * sym,GElf_Shdr * shdr,struct maps * kmaps,struct kmap * kmap,struct dso ** curr_dsop,const char * section_name,bool adjust_kernel_syms,bool kmodule,bool * remap_kernel,u64 max_text_sh_offset)1421 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
1422 GElf_Sym *sym, GElf_Shdr *shdr,
1423 struct maps *kmaps, struct kmap *kmap,
1424 struct dso **curr_dsop,
1425 const char *section_name,
1426 bool adjust_kernel_syms, bool kmodule, bool *remap_kernel,
1427 u64 max_text_sh_offset)
1428 {
1429 struct dso *curr_dso = *curr_dsop;
1430 struct map *curr_map;
1431 char dso_name[PATH_MAX];
1432
1433 /* Adjust symbol to map to file offset */
1434 if (adjust_kernel_syms)
1435 sym->st_value -= shdr->sh_addr - shdr->sh_offset;
1436
1437 if (strcmp(section_name, (dso__short_name(curr_dso) + dso__short_name_len(dso))) == 0)
1438 return 0;
1439
1440 if (strcmp(section_name, ".text") == 0) {
1441 /*
1442 * The initial kernel mapping is based on
1443 * kallsyms and identity maps. Overwrite it to
1444 * map to the kernel dso.
1445 */
1446 if (*remap_kernel && dso__kernel(dso) && !kmodule) {
1447 *remap_kernel = false;
1448 map__set_start(map, shdr->sh_addr + ref_reloc(kmap));
1449 map__set_end(map, map__start(map) + shdr->sh_size);
1450 map__set_pgoff(map, shdr->sh_offset);
1451 map__set_mapping_type(map, MAPPING_TYPE__DSO);
1452 /* Ensure maps are correctly ordered */
1453 if (kmaps) {
1454 int err;
1455 struct map *tmp = map__get(map);
1456
1457 maps__remove(kmaps, map);
1458 err = maps__insert(kmaps, map);
1459 map__put(tmp);
1460 if (err)
1461 return err;
1462 }
1463 }
1464
1465 /*
1466 * The initial module mapping is based on
1467 * /proc/modules mapped to offset zero.
1468 * Overwrite it to map to the module dso.
1469 */
1470 if (*remap_kernel && kmodule) {
1471 *remap_kernel = false;
1472 map__set_pgoff(map, shdr->sh_offset);
1473 }
1474
1475 dso__put(*curr_dsop);
1476 *curr_dsop = dso__get(dso);
1477 return 0;
1478 }
1479
1480 if (!kmap)
1481 return 0;
1482
1483 /*
1484 * perf does not record module section addresses except for .text, but
1485 * some sections can use the same mapping as .text.
1486 */
1487 if (kmodule && adjust_kernel_syms && is_exe_text(shdr->sh_flags) &&
1488 shdr->sh_offset <= max_text_sh_offset) {
1489 dso__put(*curr_dsop);
1490 *curr_dsop = dso__get(dso);
1491 return 0;
1492 }
1493
1494 snprintf(dso_name, sizeof(dso_name), "%s%s", dso__short_name(dso), section_name);
1495
1496 curr_map = maps__find_by_name(kmaps, dso_name);
1497 if (curr_map == NULL) {
1498 u64 start = sym->st_value;
1499
1500 if (kmodule)
1501 start += map__start(map) + shdr->sh_offset;
1502
1503 curr_dso = dso__new(dso_name);
1504 if (curr_dso == NULL)
1505 return -1;
1506 dso__set_kernel(curr_dso, dso__kernel(dso));
1507 RC_CHK_ACCESS(curr_dso)->long_name = dso__long_name(dso);
1508 RC_CHK_ACCESS(curr_dso)->long_name_len = dso__long_name_len(dso);
1509 dso__set_binary_type(curr_dso, dso__binary_type(dso));
1510 dso__set_adjust_symbols(curr_dso, dso__adjust_symbols(dso));
1511 curr_map = map__new2(start, curr_dso);
1512 if (curr_map == NULL) {
1513 dso__put(curr_dso);
1514 return -1;
1515 }
1516 if (dso__kernel(curr_dso))
1517 map__kmap(curr_map)->kmaps = kmaps;
1518
1519 if (adjust_kernel_syms) {
1520 map__set_start(curr_map, shdr->sh_addr + ref_reloc(kmap));
1521 map__set_end(curr_map, map__start(curr_map) + shdr->sh_size);
1522 map__set_pgoff(curr_map, shdr->sh_offset);
1523 } else {
1524 map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY);
1525 }
1526 dso__set_symtab_type(curr_dso, dso__symtab_type(dso));
1527 if (maps__insert(kmaps, curr_map))
1528 return -1;
1529 dsos__add(&maps__machine(kmaps)->dsos, curr_dso);
1530 dso__set_loaded(curr_dso);
1531 dso__put(*curr_dsop);
1532 *curr_dsop = curr_dso;
1533 } else {
1534 dso__put(*curr_dsop);
1535 *curr_dsop = dso__get(map__dso(curr_map));
1536 }
1537 map__put(curr_map);
1538
1539 return 0;
1540 }
1541
1542 static int
dso__load_sym_internal(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule,int dynsym)1543 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1544 struct symsrc *runtime_ss, int kmodule, int dynsym)
1545 {
1546 struct kmap *kmap = dso__kernel(dso) ? map__kmap(map) : NULL;
1547 struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1548 struct dso *curr_dso = NULL;
1549 Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym;
1550 uint32_t nr_syms;
1551 uint32_t idx;
1552 GElf_Ehdr ehdr;
1553 GElf_Shdr shdr;
1554 GElf_Shdr tshdr;
1555 Elf_Data *syms, *opddata = NULL;
1556 GElf_Sym sym;
1557 Elf_Scn *sec, *sec_strndx;
1558 Elf *elf;
1559 int nr = 0;
1560 bool remap_kernel = false, adjust_kernel_syms = false;
1561 u64 max_text_sh_offset = 0;
1562
1563 if (kmap && !kmaps)
1564 return -1;
1565
1566 elf = syms_ss->elf;
1567 ehdr = syms_ss->ehdr;
1568 if (dynsym) {
1569 sec = syms_ss->dynsym;
1570 shdr = syms_ss->dynshdr;
1571 } else {
1572 sec = syms_ss->symtab;
1573 shdr = syms_ss->symshdr;
1574 }
1575
1576 if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1577 ".text", NULL)) {
1578 dso__set_text_offset(dso, tshdr.sh_addr - tshdr.sh_offset);
1579 dso__set_text_end(dso, tshdr.sh_offset + tshdr.sh_size);
1580 }
1581
1582 if (runtime_ss->opdsec)
1583 opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1584
1585 syms = elf_getdata(sec, NULL);
1586 if (syms == NULL)
1587 goto out_elf_end;
1588
1589 sec = elf_getscn(elf, shdr.sh_link);
1590 if (sec == NULL)
1591 goto out_elf_end;
1592
1593 symstrs = elf_getdata(sec, NULL);
1594 if (symstrs == NULL)
1595 goto out_elf_end;
1596
1597 sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1598 if (sec_strndx == NULL)
1599 goto out_elf_end;
1600
1601 secstrs_run = elf_getdata(sec_strndx, NULL);
1602 if (secstrs_run == NULL)
1603 goto out_elf_end;
1604
1605 sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1606 if (sec_strndx == NULL)
1607 goto out_elf_end;
1608
1609 secstrs_sym = elf_getdata(sec_strndx, NULL);
1610 if (secstrs_sym == NULL)
1611 goto out_elf_end;
1612
1613 nr_syms = shdr.sh_size / shdr.sh_entsize;
1614
1615 memset(&sym, 0, sizeof(sym));
1616
1617 /*
1618 * The kernel relocation symbol is needed in advance in order to adjust
1619 * kernel maps correctly.
1620 */
1621 if (ref_reloc_sym_not_found(kmap)) {
1622 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1623 const char *elf_name = elf_sym__name(&sym, symstrs);
1624
1625 if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1626 continue;
1627 kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1628 map__set_reloc(map, kmap->ref_reloc_sym->addr - kmap->ref_reloc_sym->unrelocated_addr);
1629 break;
1630 }
1631 }
1632
1633 /*
1634 * Handle any relocation of vdso necessary because older kernels
1635 * attempted to prelink vdso to its virtual address.
1636 */
1637 if (dso__is_vdso(dso))
1638 map__set_reloc(map, map__start(map) - dso__text_offset(dso));
1639
1640 dso__set_adjust_symbols(dso, runtime_ss->adjust_symbols || ref_reloc(kmap));
1641 /*
1642 * Initial kernel and module mappings do not map to the dso.
1643 * Flag the fixups.
1644 */
1645 if (dso__kernel(dso)) {
1646 remap_kernel = true;
1647 adjust_kernel_syms = dso__adjust_symbols(dso);
1648 }
1649
1650 if (kmodule && adjust_kernel_syms)
1651 max_text_sh_offset = max_text_section(runtime_ss->elf, &runtime_ss->ehdr);
1652
1653 curr_dso = dso__get(dso);
1654 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1655 struct symbol *f;
1656 const char *elf_name = elf_sym__name(&sym, symstrs);
1657 char *demangled = NULL;
1658 int is_label = elf_sym__is_label(&sym);
1659 const char *section_name;
1660 bool used_opd = false;
1661
1662 if (!is_label && !elf_sym__filter(&sym))
1663 continue;
1664
1665 /* Reject ARM ELF "mapping symbols": these aren't unique and
1666 * don't identify functions, so will confuse the profile
1667 * output: */
1668 if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1669 if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1670 && (elf_name[2] == '\0' || elf_name[2] == '.'))
1671 continue;
1672 }
1673
1674 if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1675 u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1676 u64 *opd = opddata->d_buf + offset;
1677 sym.st_value = DSO__SWAP(dso, u64, *opd);
1678 sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1679 sym.st_value);
1680 used_opd = true;
1681 }
1682
1683 /*
1684 * When loading symbols in a data mapping, ABS symbols (which
1685 * has a value of SHN_ABS in its st_shndx) failed at
1686 * elf_getscn(). And it marks the loading as a failure so
1687 * already loaded symbols cannot be fixed up.
1688 *
1689 * I'm not sure what should be done. Just ignore them for now.
1690 * - Namhyung Kim
1691 */
1692 if (sym.st_shndx == SHN_ABS)
1693 continue;
1694
1695 sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1696 if (!sec)
1697 goto out_elf_end;
1698
1699 gelf_getshdr(sec, &shdr);
1700
1701 /*
1702 * If the attribute bit SHF_ALLOC is not set, the section
1703 * doesn't occupy memory during process execution.
1704 * E.g. ".gnu.warning.*" section is used by linker to generate
1705 * warnings when calling deprecated functions, the symbols in
1706 * the section aren't loaded to memory during process execution,
1707 * so skip them.
1708 */
1709 if (!(shdr.sh_flags & SHF_ALLOC))
1710 continue;
1711
1712 secstrs = secstrs_sym;
1713
1714 /*
1715 * We have to fallback to runtime when syms' section header has
1716 * NOBITS set. NOBITS results in file offset (sh_offset) not
1717 * being incremented. So sh_offset used below has different
1718 * values for syms (invalid) and runtime (valid).
1719 */
1720 if (shdr.sh_type == SHT_NOBITS) {
1721 sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1722 if (!sec)
1723 goto out_elf_end;
1724
1725 gelf_getshdr(sec, &shdr);
1726 secstrs = secstrs_run;
1727 }
1728
1729 if (is_label && !elf_sec__filter(&shdr, secstrs))
1730 continue;
1731
1732 section_name = elf_sec__name(&shdr, secstrs);
1733
1734 /* On ARM, symbols for thumb functions have 1 added to
1735 * the symbol address as a flag - remove it */
1736 if ((ehdr.e_machine == EM_ARM) &&
1737 (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1738 (sym.st_value & 1))
1739 --sym.st_value;
1740
1741 if (dso__kernel(dso)) {
1742 if (dso__process_kernel_symbol(dso, map, &sym, &shdr,
1743 kmaps, kmap, &curr_dso,
1744 section_name,
1745 adjust_kernel_syms,
1746 kmodule,
1747 &remap_kernel,
1748 max_text_sh_offset))
1749 goto out_elf_end;
1750 } else if ((used_opd && runtime_ss->adjust_symbols) ||
1751 (!used_opd && syms_ss->adjust_symbols)) {
1752 GElf_Phdr phdr;
1753
1754 if (elf_read_program_header(runtime_ss->elf,
1755 (u64)sym.st_value, &phdr)) {
1756 pr_debug4("%s: failed to find program header for "
1757 "symbol: %s st_value: %#" PRIx64 "\n",
1758 __func__, elf_name, (u64)sym.st_value);
1759 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1760 "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n",
1761 __func__, (u64)sym.st_value, (u64)shdr.sh_addr,
1762 (u64)shdr.sh_offset);
1763 /*
1764 * Fail to find program header, let's rollback
1765 * to use shdr.sh_addr and shdr.sh_offset to
1766 * calibrate symbol's file address, though this
1767 * is not necessary for normal C ELF file, we
1768 * still need to handle java JIT symbols in this
1769 * case.
1770 */
1771 sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1772 } else {
1773 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1774 "p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n",
1775 __func__, (u64)sym.st_value, (u64)phdr.p_vaddr,
1776 (u64)phdr.p_offset);
1777 sym.st_value -= phdr.p_vaddr - phdr.p_offset;
1778 }
1779 }
1780
1781 demangled = demangle_sym(dso, kmodule, elf_name);
1782 if (demangled != NULL)
1783 elf_name = demangled;
1784
1785 f = symbol__new(sym.st_value, sym.st_size,
1786 GELF_ST_BIND(sym.st_info),
1787 GELF_ST_TYPE(sym.st_info), elf_name);
1788 free(demangled);
1789 if (!f)
1790 goto out_elf_end;
1791
1792 arch__sym_update(f, &sym);
1793
1794 __symbols__insert(dso__symbols(curr_dso), f, dso__kernel(dso));
1795 nr++;
1796 }
1797 dso__put(curr_dso);
1798
1799 /*
1800 * For misannotated, zeroed, ASM function sizes.
1801 */
1802 if (nr > 0) {
1803 symbols__fixup_end(dso__symbols(dso), false);
1804 symbols__fixup_duplicate(dso__symbols(dso));
1805 if (kmap) {
1806 /*
1807 * We need to fixup this here too because we create new
1808 * maps here, for things like vsyscall sections.
1809 */
1810 maps__fixup_end(kmaps);
1811 }
1812 }
1813 return nr;
1814 out_elf_end:
1815 dso__put(curr_dso);
1816 return -1;
1817 }
1818
dso__load_sym(struct dso * dso,struct map * map,struct symsrc * syms_ss,struct symsrc * runtime_ss,int kmodule)1819 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1820 struct symsrc *runtime_ss, int kmodule)
1821 {
1822 int nr = 0;
1823 int err = -1;
1824
1825 dso__set_symtab_type(dso, syms_ss->type);
1826 dso__set_is_64_bit(dso, syms_ss->is_64_bit);
1827 dso__set_rel(dso, syms_ss->ehdr.e_type == ET_REL);
1828
1829 /*
1830 * Modules may already have symbols from kallsyms, but those symbols
1831 * have the wrong values for the dso maps, so remove them.
1832 */
1833 if (kmodule && syms_ss->symtab)
1834 symbols__delete(dso__symbols(dso));
1835
1836 if (!syms_ss->symtab) {
1837 /*
1838 * If the vmlinux is stripped, fail so we will fall back
1839 * to using kallsyms. The vmlinux runtime symbols aren't
1840 * of much use.
1841 */
1842 if (dso__kernel(dso))
1843 return err;
1844 } else {
1845 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1846 kmodule, 0);
1847 if (err < 0)
1848 return err;
1849 nr = err;
1850 }
1851
1852 if (syms_ss->dynsym) {
1853 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1854 kmodule, 1);
1855 if (err < 0)
1856 return err;
1857 err += nr;
1858 }
1859
1860 return err;
1861 }
1862
elf_read_maps(Elf * elf,bool exe,mapfn_t mapfn,void * data)1863 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1864 {
1865 GElf_Phdr phdr;
1866 size_t i, phdrnum;
1867 int err;
1868 u64 sz;
1869
1870 if (elf_getphdrnum(elf, &phdrnum))
1871 return -1;
1872
1873 for (i = 0; i < phdrnum; i++) {
1874 if (gelf_getphdr(elf, i, &phdr) == NULL)
1875 return -1;
1876 if (phdr.p_type != PT_LOAD)
1877 continue;
1878 if (exe) {
1879 if (!(phdr.p_flags & PF_X))
1880 continue;
1881 } else {
1882 if (!(phdr.p_flags & PF_R))
1883 continue;
1884 }
1885 sz = min(phdr.p_memsz, phdr.p_filesz);
1886 if (!sz)
1887 continue;
1888 err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1889 if (err)
1890 return err;
1891 }
1892 return 0;
1893 }
1894
file__read_maps(int fd,bool exe,mapfn_t mapfn,void * data,bool * is_64_bit)1895 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1896 bool *is_64_bit)
1897 {
1898 int err;
1899 Elf *elf;
1900
1901 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1902 if (elf == NULL)
1903 return -1;
1904
1905 if (is_64_bit)
1906 *is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1907
1908 err = elf_read_maps(elf, exe, mapfn, data);
1909
1910 elf_end(elf);
1911 return err;
1912 }
1913
dso__type_fd(int fd)1914 enum dso_type dso__type_fd(int fd)
1915 {
1916 enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1917 GElf_Ehdr ehdr;
1918 Elf_Kind ek;
1919 Elf *elf;
1920
1921 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1922 if (elf == NULL)
1923 goto out;
1924
1925 ek = elf_kind(elf);
1926 if (ek != ELF_K_ELF)
1927 goto out_end;
1928
1929 if (gelf_getclass(elf) == ELFCLASS64) {
1930 dso_type = DSO__TYPE_64BIT;
1931 goto out_end;
1932 }
1933
1934 if (gelf_getehdr(elf, &ehdr) == NULL)
1935 goto out_end;
1936
1937 if (ehdr.e_machine == EM_X86_64)
1938 dso_type = DSO__TYPE_X32BIT;
1939 else
1940 dso_type = DSO__TYPE_32BIT;
1941 out_end:
1942 elf_end(elf);
1943 out:
1944 return dso_type;
1945 }
1946
copy_bytes(int from,off_t from_offs,int to,off_t to_offs,u64 len)1947 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1948 {
1949 ssize_t r;
1950 size_t n;
1951 int err = -1;
1952 char *buf = malloc(page_size);
1953
1954 if (buf == NULL)
1955 return -1;
1956
1957 if (lseek(to, to_offs, SEEK_SET) != to_offs)
1958 goto out;
1959
1960 if (lseek(from, from_offs, SEEK_SET) != from_offs)
1961 goto out;
1962
1963 while (len) {
1964 n = page_size;
1965 if (len < n)
1966 n = len;
1967 /* Use read because mmap won't work on proc files */
1968 r = read(from, buf, n);
1969 if (r < 0)
1970 goto out;
1971 if (!r)
1972 break;
1973 n = r;
1974 r = write(to, buf, n);
1975 if (r < 0)
1976 goto out;
1977 if ((size_t)r != n)
1978 goto out;
1979 len -= n;
1980 }
1981
1982 err = 0;
1983 out:
1984 free(buf);
1985 return err;
1986 }
1987
1988 struct kcore {
1989 int fd;
1990 int elfclass;
1991 Elf *elf;
1992 GElf_Ehdr ehdr;
1993 };
1994
kcore__open(struct kcore * kcore,const char * filename)1995 static int kcore__open(struct kcore *kcore, const char *filename)
1996 {
1997 GElf_Ehdr *ehdr;
1998
1999 kcore->fd = open(filename, O_RDONLY);
2000 if (kcore->fd == -1)
2001 return -1;
2002
2003 kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
2004 if (!kcore->elf)
2005 goto out_close;
2006
2007 kcore->elfclass = gelf_getclass(kcore->elf);
2008 if (kcore->elfclass == ELFCLASSNONE)
2009 goto out_end;
2010
2011 ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
2012 if (!ehdr)
2013 goto out_end;
2014
2015 return 0;
2016
2017 out_end:
2018 elf_end(kcore->elf);
2019 out_close:
2020 close(kcore->fd);
2021 return -1;
2022 }
2023
kcore__init(struct kcore * kcore,char * filename,int elfclass,bool temp)2024 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
2025 bool temp)
2026 {
2027 kcore->elfclass = elfclass;
2028
2029 if (temp)
2030 kcore->fd = mkstemp(filename);
2031 else
2032 kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
2033 if (kcore->fd == -1)
2034 return -1;
2035
2036 kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
2037 if (!kcore->elf)
2038 goto out_close;
2039
2040 if (!gelf_newehdr(kcore->elf, elfclass))
2041 goto out_end;
2042
2043 memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
2044
2045 return 0;
2046
2047 out_end:
2048 elf_end(kcore->elf);
2049 out_close:
2050 close(kcore->fd);
2051 unlink(filename);
2052 return -1;
2053 }
2054
kcore__close(struct kcore * kcore)2055 static void kcore__close(struct kcore *kcore)
2056 {
2057 elf_end(kcore->elf);
2058 close(kcore->fd);
2059 }
2060
kcore__copy_hdr(struct kcore * from,struct kcore * to,size_t count)2061 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
2062 {
2063 GElf_Ehdr *ehdr = &to->ehdr;
2064 GElf_Ehdr *kehdr = &from->ehdr;
2065
2066 memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
2067 ehdr->e_type = kehdr->e_type;
2068 ehdr->e_machine = kehdr->e_machine;
2069 ehdr->e_version = kehdr->e_version;
2070 ehdr->e_entry = 0;
2071 ehdr->e_shoff = 0;
2072 ehdr->e_flags = kehdr->e_flags;
2073 ehdr->e_phnum = count;
2074 ehdr->e_shentsize = 0;
2075 ehdr->e_shnum = 0;
2076 ehdr->e_shstrndx = 0;
2077
2078 if (from->elfclass == ELFCLASS32) {
2079 ehdr->e_phoff = sizeof(Elf32_Ehdr);
2080 ehdr->e_ehsize = sizeof(Elf32_Ehdr);
2081 ehdr->e_phentsize = sizeof(Elf32_Phdr);
2082 } else {
2083 ehdr->e_phoff = sizeof(Elf64_Ehdr);
2084 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
2085 ehdr->e_phentsize = sizeof(Elf64_Phdr);
2086 }
2087
2088 if (!gelf_update_ehdr(to->elf, ehdr))
2089 return -1;
2090
2091 if (!gelf_newphdr(to->elf, count))
2092 return -1;
2093
2094 return 0;
2095 }
2096
kcore__add_phdr(struct kcore * kcore,int idx,off_t offset,u64 addr,u64 len)2097 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
2098 u64 addr, u64 len)
2099 {
2100 GElf_Phdr phdr = {
2101 .p_type = PT_LOAD,
2102 .p_flags = PF_R | PF_W | PF_X,
2103 .p_offset = offset,
2104 .p_vaddr = addr,
2105 .p_paddr = 0,
2106 .p_filesz = len,
2107 .p_memsz = len,
2108 .p_align = page_size,
2109 };
2110
2111 if (!gelf_update_phdr(kcore->elf, idx, &phdr))
2112 return -1;
2113
2114 return 0;
2115 }
2116
kcore__write(struct kcore * kcore)2117 static off_t kcore__write(struct kcore *kcore)
2118 {
2119 return elf_update(kcore->elf, ELF_C_WRITE);
2120 }
2121
2122 struct phdr_data {
2123 off_t offset;
2124 off_t rel;
2125 u64 addr;
2126 u64 len;
2127 struct list_head node;
2128 struct phdr_data *remaps;
2129 };
2130
2131 struct sym_data {
2132 u64 addr;
2133 struct list_head node;
2134 };
2135
2136 struct kcore_copy_info {
2137 u64 stext;
2138 u64 etext;
2139 u64 first_symbol;
2140 u64 last_symbol;
2141 u64 first_module;
2142 u64 first_module_symbol;
2143 u64 last_module_symbol;
2144 size_t phnum;
2145 struct list_head phdrs;
2146 struct list_head syms;
2147 };
2148
2149 #define kcore_copy__for_each_phdr(k, p) \
2150 list_for_each_entry((p), &(k)->phdrs, node)
2151
phdr_data__new(u64 addr,u64 len,off_t offset)2152 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
2153 {
2154 struct phdr_data *p = zalloc(sizeof(*p));
2155
2156 if (p) {
2157 p->addr = addr;
2158 p->len = len;
2159 p->offset = offset;
2160 }
2161
2162 return p;
2163 }
2164
kcore_copy_info__addnew(struct kcore_copy_info * kci,u64 addr,u64 len,off_t offset)2165 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
2166 u64 addr, u64 len,
2167 off_t offset)
2168 {
2169 struct phdr_data *p = phdr_data__new(addr, len, offset);
2170
2171 if (p)
2172 list_add_tail(&p->node, &kci->phdrs);
2173
2174 return p;
2175 }
2176
kcore_copy__free_phdrs(struct kcore_copy_info * kci)2177 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
2178 {
2179 struct phdr_data *p, *tmp;
2180
2181 list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
2182 list_del_init(&p->node);
2183 free(p);
2184 }
2185 }
2186
kcore_copy__new_sym(struct kcore_copy_info * kci,u64 addr)2187 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
2188 u64 addr)
2189 {
2190 struct sym_data *s = zalloc(sizeof(*s));
2191
2192 if (s) {
2193 s->addr = addr;
2194 list_add_tail(&s->node, &kci->syms);
2195 }
2196
2197 return s;
2198 }
2199
kcore_copy__free_syms(struct kcore_copy_info * kci)2200 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
2201 {
2202 struct sym_data *s, *tmp;
2203
2204 list_for_each_entry_safe(s, tmp, &kci->syms, node) {
2205 list_del_init(&s->node);
2206 free(s);
2207 }
2208 }
2209
kcore_copy__process_kallsyms(void * arg,const char * name,char type,u64 start)2210 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
2211 u64 start)
2212 {
2213 struct kcore_copy_info *kci = arg;
2214
2215 if (!kallsyms__is_function(type))
2216 return 0;
2217
2218 if (strchr(name, '[')) {
2219 if (!kci->first_module_symbol || start < kci->first_module_symbol)
2220 kci->first_module_symbol = start;
2221 if (start > kci->last_module_symbol)
2222 kci->last_module_symbol = start;
2223 return 0;
2224 }
2225
2226 if (!kci->first_symbol || start < kci->first_symbol)
2227 kci->first_symbol = start;
2228
2229 if (!kci->last_symbol || start > kci->last_symbol)
2230 kci->last_symbol = start;
2231
2232 if (!strcmp(name, "_stext")) {
2233 kci->stext = start;
2234 return 0;
2235 }
2236
2237 if (!strcmp(name, "_etext")) {
2238 kci->etext = start;
2239 return 0;
2240 }
2241
2242 if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
2243 return -1;
2244
2245 return 0;
2246 }
2247
kcore_copy__parse_kallsyms(struct kcore_copy_info * kci,const char * dir)2248 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
2249 const char *dir)
2250 {
2251 char kallsyms_filename[PATH_MAX];
2252
2253 scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
2254
2255 if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
2256 return -1;
2257
2258 if (kallsyms__parse(kallsyms_filename, kci,
2259 kcore_copy__process_kallsyms) < 0)
2260 return -1;
2261
2262 return 0;
2263 }
2264
kcore_copy__process_modules(void * arg,const char * name __maybe_unused,u64 start,u64 size __maybe_unused)2265 static int kcore_copy__process_modules(void *arg,
2266 const char *name __maybe_unused,
2267 u64 start, u64 size __maybe_unused)
2268 {
2269 struct kcore_copy_info *kci = arg;
2270
2271 if (!kci->first_module || start < kci->first_module)
2272 kci->first_module = start;
2273
2274 return 0;
2275 }
2276
kcore_copy__parse_modules(struct kcore_copy_info * kci,const char * dir)2277 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
2278 const char *dir)
2279 {
2280 char modules_filename[PATH_MAX];
2281
2282 scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
2283
2284 if (symbol__restricted_filename(modules_filename, "/proc/modules"))
2285 return -1;
2286
2287 if (modules__parse(modules_filename, kci,
2288 kcore_copy__process_modules) < 0)
2289 return -1;
2290
2291 return 0;
2292 }
2293
kcore_copy__map(struct kcore_copy_info * kci,u64 start,u64 end,u64 pgoff,u64 s,u64 e)2294 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
2295 u64 pgoff, u64 s, u64 e)
2296 {
2297 u64 len, offset;
2298
2299 if (s < start || s >= end)
2300 return 0;
2301
2302 offset = (s - start) + pgoff;
2303 len = e < end ? e - s : end - s;
2304
2305 return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
2306 }
2307
kcore_copy__read_map(u64 start,u64 len,u64 pgoff,void * data)2308 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
2309 {
2310 struct kcore_copy_info *kci = data;
2311 u64 end = start + len;
2312 struct sym_data *sdat;
2313
2314 if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
2315 return -1;
2316
2317 if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
2318 kci->last_module_symbol))
2319 return -1;
2320
2321 list_for_each_entry(sdat, &kci->syms, node) {
2322 u64 s = round_down(sdat->addr, page_size);
2323
2324 if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
2325 return -1;
2326 }
2327
2328 return 0;
2329 }
2330
kcore_copy__read_maps(struct kcore_copy_info * kci,Elf * elf)2331 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
2332 {
2333 if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
2334 return -1;
2335
2336 return 0;
2337 }
2338
kcore_copy__find_remaps(struct kcore_copy_info * kci)2339 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
2340 {
2341 struct phdr_data *p, *k = NULL;
2342 u64 kend;
2343
2344 if (!kci->stext)
2345 return;
2346
2347 /* Find phdr that corresponds to the kernel map (contains stext) */
2348 kcore_copy__for_each_phdr(kci, p) {
2349 u64 pend = p->addr + p->len - 1;
2350
2351 if (p->addr <= kci->stext && pend >= kci->stext) {
2352 k = p;
2353 break;
2354 }
2355 }
2356
2357 if (!k)
2358 return;
2359
2360 kend = k->offset + k->len;
2361
2362 /* Find phdrs that remap the kernel */
2363 kcore_copy__for_each_phdr(kci, p) {
2364 u64 pend = p->offset + p->len;
2365
2366 if (p == k)
2367 continue;
2368
2369 if (p->offset >= k->offset && pend <= kend)
2370 p->remaps = k;
2371 }
2372 }
2373
kcore_copy__layout(struct kcore_copy_info * kci)2374 static void kcore_copy__layout(struct kcore_copy_info *kci)
2375 {
2376 struct phdr_data *p;
2377 off_t rel = 0;
2378
2379 kcore_copy__find_remaps(kci);
2380
2381 kcore_copy__for_each_phdr(kci, p) {
2382 if (!p->remaps) {
2383 p->rel = rel;
2384 rel += p->len;
2385 }
2386 kci->phnum += 1;
2387 }
2388
2389 kcore_copy__for_each_phdr(kci, p) {
2390 struct phdr_data *k = p->remaps;
2391
2392 if (k)
2393 p->rel = p->offset - k->offset + k->rel;
2394 }
2395 }
2396
kcore_copy__calc_maps(struct kcore_copy_info * kci,const char * dir,Elf * elf)2397 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
2398 Elf *elf)
2399 {
2400 if (kcore_copy__parse_kallsyms(kci, dir))
2401 return -1;
2402
2403 if (kcore_copy__parse_modules(kci, dir))
2404 return -1;
2405
2406 if (kci->stext)
2407 kci->stext = round_down(kci->stext, page_size);
2408 else
2409 kci->stext = round_down(kci->first_symbol, page_size);
2410
2411 if (kci->etext) {
2412 kci->etext = round_up(kci->etext, page_size);
2413 } else if (kci->last_symbol) {
2414 kci->etext = round_up(kci->last_symbol, page_size);
2415 kci->etext += page_size;
2416 }
2417
2418 if (kci->first_module_symbol &&
2419 (!kci->first_module || kci->first_module_symbol < kci->first_module))
2420 kci->first_module = kci->first_module_symbol;
2421
2422 kci->first_module = round_down(kci->first_module, page_size);
2423
2424 if (kci->last_module_symbol) {
2425 kci->last_module_symbol = round_up(kci->last_module_symbol,
2426 page_size);
2427 kci->last_module_symbol += page_size;
2428 }
2429
2430 if (!kci->stext || !kci->etext)
2431 return -1;
2432
2433 if (kci->first_module && !kci->last_module_symbol)
2434 return -1;
2435
2436 if (kcore_copy__read_maps(kci, elf))
2437 return -1;
2438
2439 kcore_copy__layout(kci);
2440
2441 return 0;
2442 }
2443
kcore_copy__copy_file(const char * from_dir,const char * to_dir,const char * name)2444 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
2445 const char *name)
2446 {
2447 char from_filename[PATH_MAX];
2448 char to_filename[PATH_MAX];
2449
2450 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2451 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2452
2453 return copyfile_mode(from_filename, to_filename, 0400);
2454 }
2455
kcore_copy__unlink(const char * dir,const char * name)2456 static int kcore_copy__unlink(const char *dir, const char *name)
2457 {
2458 char filename[PATH_MAX];
2459
2460 scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
2461
2462 return unlink(filename);
2463 }
2464
kcore_copy__compare_fds(int from,int to)2465 static int kcore_copy__compare_fds(int from, int to)
2466 {
2467 char *buf_from;
2468 char *buf_to;
2469 ssize_t ret;
2470 size_t len;
2471 int err = -1;
2472
2473 buf_from = malloc(page_size);
2474 buf_to = malloc(page_size);
2475 if (!buf_from || !buf_to)
2476 goto out;
2477
2478 while (1) {
2479 /* Use read because mmap won't work on proc files */
2480 ret = read(from, buf_from, page_size);
2481 if (ret < 0)
2482 goto out;
2483
2484 if (!ret)
2485 break;
2486
2487 len = ret;
2488
2489 if (readn(to, buf_to, len) != (int)len)
2490 goto out;
2491
2492 if (memcmp(buf_from, buf_to, len))
2493 goto out;
2494 }
2495
2496 err = 0;
2497 out:
2498 free(buf_to);
2499 free(buf_from);
2500 return err;
2501 }
2502
kcore_copy__compare_files(const char * from_filename,const char * to_filename)2503 static int kcore_copy__compare_files(const char *from_filename,
2504 const char *to_filename)
2505 {
2506 int from, to, err = -1;
2507
2508 from = open(from_filename, O_RDONLY);
2509 if (from < 0)
2510 return -1;
2511
2512 to = open(to_filename, O_RDONLY);
2513 if (to < 0)
2514 goto out_close_from;
2515
2516 err = kcore_copy__compare_fds(from, to);
2517
2518 close(to);
2519 out_close_from:
2520 close(from);
2521 return err;
2522 }
2523
kcore_copy__compare_file(const char * from_dir,const char * to_dir,const char * name)2524 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
2525 const char *name)
2526 {
2527 char from_filename[PATH_MAX];
2528 char to_filename[PATH_MAX];
2529
2530 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2531 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2532
2533 return kcore_copy__compare_files(from_filename, to_filename);
2534 }
2535
2536 /**
2537 * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
2538 * @from_dir: from directory
2539 * @to_dir: to directory
2540 *
2541 * This function copies kallsyms, modules and kcore files from one directory to
2542 * another. kallsyms and modules are copied entirely. Only code segments are
2543 * copied from kcore. It is assumed that two segments suffice: one for the
2544 * kernel proper and one for all the modules. The code segments are determined
2545 * from kallsyms and modules files. The kernel map starts at _stext or the
2546 * lowest function symbol, and ends at _etext or the highest function symbol.
2547 * The module map starts at the lowest module address and ends at the highest
2548 * module symbol. Start addresses are rounded down to the nearest page. End
2549 * addresses are rounded up to the nearest page. An extra page is added to the
2550 * highest kernel symbol and highest module symbol to, hopefully, encompass that
2551 * symbol too. Because it contains only code sections, the resulting kcore is
2552 * unusual. One significant peculiarity is that the mapping (start -> pgoff)
2553 * is not the same for the kernel map and the modules map. That happens because
2554 * the data is copied adjacently whereas the original kcore has gaps. Finally,
2555 * kallsyms file is compared with its copy to check that modules have not been
2556 * loaded or unloaded while the copies were taking place.
2557 *
2558 * Return: %0 on success, %-1 on failure.
2559 */
kcore_copy(const char * from_dir,const char * to_dir)2560 int kcore_copy(const char *from_dir, const char *to_dir)
2561 {
2562 struct kcore kcore;
2563 struct kcore extract;
2564 int idx = 0, err = -1;
2565 off_t offset, sz;
2566 struct kcore_copy_info kci = { .stext = 0, };
2567 char kcore_filename[PATH_MAX];
2568 char extract_filename[PATH_MAX];
2569 struct phdr_data *p;
2570
2571 INIT_LIST_HEAD(&kci.phdrs);
2572 INIT_LIST_HEAD(&kci.syms);
2573
2574 if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2575 return -1;
2576
2577 if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2578 goto out_unlink_kallsyms;
2579
2580 scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2581 scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2582
2583 if (kcore__open(&kcore, kcore_filename))
2584 goto out_unlink_modules;
2585
2586 if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2587 goto out_kcore_close;
2588
2589 if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2590 goto out_kcore_close;
2591
2592 if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2593 goto out_extract_close;
2594
2595 offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2596 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2597 offset = round_up(offset, page_size);
2598
2599 kcore_copy__for_each_phdr(&kci, p) {
2600 off_t offs = p->rel + offset;
2601
2602 if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2603 goto out_extract_close;
2604 }
2605
2606 sz = kcore__write(&extract);
2607 if (sz < 0 || sz > offset)
2608 goto out_extract_close;
2609
2610 kcore_copy__for_each_phdr(&kci, p) {
2611 off_t offs = p->rel + offset;
2612
2613 if (p->remaps)
2614 continue;
2615 if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2616 goto out_extract_close;
2617 }
2618
2619 if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2620 goto out_extract_close;
2621
2622 err = 0;
2623
2624 out_extract_close:
2625 kcore__close(&extract);
2626 if (err)
2627 unlink(extract_filename);
2628 out_kcore_close:
2629 kcore__close(&kcore);
2630 out_unlink_modules:
2631 if (err)
2632 kcore_copy__unlink(to_dir, "modules");
2633 out_unlink_kallsyms:
2634 if (err)
2635 kcore_copy__unlink(to_dir, "kallsyms");
2636
2637 kcore_copy__free_phdrs(&kci);
2638 kcore_copy__free_syms(&kci);
2639
2640 return err;
2641 }
2642
kcore_extract__create(struct kcore_extract * kce)2643 int kcore_extract__create(struct kcore_extract *kce)
2644 {
2645 struct kcore kcore;
2646 struct kcore extract;
2647 size_t count = 1;
2648 int idx = 0, err = -1;
2649 off_t offset = page_size, sz;
2650
2651 if (kcore__open(&kcore, kce->kcore_filename))
2652 return -1;
2653
2654 strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2655 if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2656 goto out_kcore_close;
2657
2658 if (kcore__copy_hdr(&kcore, &extract, count))
2659 goto out_extract_close;
2660
2661 if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2662 goto out_extract_close;
2663
2664 sz = kcore__write(&extract);
2665 if (sz < 0 || sz > offset)
2666 goto out_extract_close;
2667
2668 if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2669 goto out_extract_close;
2670
2671 err = 0;
2672
2673 out_extract_close:
2674 kcore__close(&extract);
2675 if (err)
2676 unlink(kce->extract_filename);
2677 out_kcore_close:
2678 kcore__close(&kcore);
2679
2680 return err;
2681 }
2682
kcore_extract__delete(struct kcore_extract * kce)2683 void kcore_extract__delete(struct kcore_extract *kce)
2684 {
2685 unlink(kce->extract_filename);
2686 }
2687
2688 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2689
sdt_adjust_loc(struct sdt_note * tmp,GElf_Addr base_off)2690 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2691 {
2692 if (!base_off)
2693 return;
2694
2695 if (tmp->bit32)
2696 tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2697 tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2698 tmp->addr.a32[SDT_NOTE_IDX_BASE];
2699 else
2700 tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2701 tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2702 tmp->addr.a64[SDT_NOTE_IDX_BASE];
2703 }
2704
sdt_adjust_refctr(struct sdt_note * tmp,GElf_Addr base_addr,GElf_Addr base_off)2705 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2706 GElf_Addr base_off)
2707 {
2708 if (!base_off)
2709 return;
2710
2711 if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2712 tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2713 else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2714 tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2715 }
2716
2717 /**
2718 * populate_sdt_note : Parse raw data and identify SDT note
2719 * @elf: elf of the opened file
2720 * @data: raw data of a section with description offset applied
2721 * @len: note description size
2722 * @type: type of the note
2723 * @sdt_notes: List to add the SDT note
2724 *
2725 * Responsible for parsing the @data in section .note.stapsdt in @elf and
2726 * if its an SDT note, it appends to @sdt_notes list.
2727 */
populate_sdt_note(Elf ** elf,const char * data,size_t len,struct list_head * sdt_notes)2728 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2729 struct list_head *sdt_notes)
2730 {
2731 const char *provider, *name, *args;
2732 struct sdt_note *tmp = NULL;
2733 GElf_Ehdr ehdr;
2734 GElf_Shdr shdr;
2735 int ret = -EINVAL;
2736
2737 union {
2738 Elf64_Addr a64[NR_ADDR];
2739 Elf32_Addr a32[NR_ADDR];
2740 } buf;
2741
2742 Elf_Data dst = {
2743 .d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2744 .d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2745 .d_off = 0, .d_align = 0
2746 };
2747 Elf_Data src = {
2748 .d_buf = (void *) data, .d_type = ELF_T_ADDR,
2749 .d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2750 .d_align = 0
2751 };
2752
2753 tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2754 if (!tmp) {
2755 ret = -ENOMEM;
2756 goto out_err;
2757 }
2758
2759 INIT_LIST_HEAD(&tmp->note_list);
2760
2761 if (len < dst.d_size + 3)
2762 goto out_free_note;
2763
2764 /* Translation from file representation to memory representation */
2765 if (gelf_xlatetom(*elf, &dst, &src,
2766 elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2767 pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2768 goto out_free_note;
2769 }
2770
2771 /* Populate the fields of sdt_note */
2772 provider = data + dst.d_size;
2773
2774 name = (const char *)memchr(provider, '\0', data + len - provider);
2775 if (name++ == NULL)
2776 goto out_free_note;
2777
2778 tmp->provider = strdup(provider);
2779 if (!tmp->provider) {
2780 ret = -ENOMEM;
2781 goto out_free_note;
2782 }
2783 tmp->name = strdup(name);
2784 if (!tmp->name) {
2785 ret = -ENOMEM;
2786 goto out_free_prov;
2787 }
2788
2789 args = memchr(name, '\0', data + len - name);
2790
2791 /*
2792 * There is no argument if:
2793 * - We reached the end of the note;
2794 * - There is not enough room to hold a potential string;
2795 * - The argument string is empty or just contains ':'.
2796 */
2797 if (args == NULL || data + len - args < 2 ||
2798 args[1] == ':' || args[1] == '\0')
2799 tmp->args = NULL;
2800 else {
2801 tmp->args = strdup(++args);
2802 if (!tmp->args) {
2803 ret = -ENOMEM;
2804 goto out_free_name;
2805 }
2806 }
2807
2808 if (gelf_getclass(*elf) == ELFCLASS32) {
2809 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2810 tmp->bit32 = true;
2811 } else {
2812 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2813 tmp->bit32 = false;
2814 }
2815
2816 if (!gelf_getehdr(*elf, &ehdr)) {
2817 pr_debug("%s : cannot get elf header.\n", __func__);
2818 ret = -EBADF;
2819 goto out_free_args;
2820 }
2821
2822 /* Adjust the prelink effect :
2823 * Find out the .stapsdt.base section.
2824 * This scn will help us to handle prelinking (if present).
2825 * Compare the retrieved file offset of the base section with the
2826 * base address in the description of the SDT note. If its different,
2827 * then accordingly, adjust the note location.
2828 */
2829 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2830 sdt_adjust_loc(tmp, shdr.sh_offset);
2831
2832 /* Adjust reference counter offset */
2833 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2834 sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2835
2836 list_add_tail(&tmp->note_list, sdt_notes);
2837 return 0;
2838
2839 out_free_args:
2840 zfree(&tmp->args);
2841 out_free_name:
2842 zfree(&tmp->name);
2843 out_free_prov:
2844 zfree(&tmp->provider);
2845 out_free_note:
2846 free(tmp);
2847 out_err:
2848 return ret;
2849 }
2850
2851 /**
2852 * construct_sdt_notes_list : constructs a list of SDT notes
2853 * @elf : elf to look into
2854 * @sdt_notes : empty list_head
2855 *
2856 * Scans the sections in 'elf' for the section
2857 * .note.stapsdt. It, then calls populate_sdt_note to find
2858 * out the SDT events and populates the 'sdt_notes'.
2859 */
construct_sdt_notes_list(Elf * elf,struct list_head * sdt_notes)2860 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2861 {
2862 GElf_Ehdr ehdr;
2863 Elf_Scn *scn = NULL;
2864 Elf_Data *data;
2865 GElf_Shdr shdr;
2866 size_t shstrndx, next;
2867 GElf_Nhdr nhdr;
2868 size_t name_off, desc_off, offset;
2869 int ret = 0;
2870
2871 if (gelf_getehdr(elf, &ehdr) == NULL) {
2872 ret = -EBADF;
2873 goto out_ret;
2874 }
2875 if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2876 ret = -EBADF;
2877 goto out_ret;
2878 }
2879
2880 /* Look for the required section */
2881 scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2882 if (!scn) {
2883 ret = -ENOENT;
2884 goto out_ret;
2885 }
2886
2887 if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2888 ret = -ENOENT;
2889 goto out_ret;
2890 }
2891
2892 data = elf_getdata(scn, NULL);
2893
2894 /* Get the SDT notes */
2895 for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2896 &desc_off)) > 0; offset = next) {
2897 if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2898 !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2899 sizeof(SDT_NOTE_NAME))) {
2900 /* Check the type of the note */
2901 if (nhdr.n_type != SDT_NOTE_TYPE)
2902 goto out_ret;
2903
2904 ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2905 nhdr.n_descsz, sdt_notes);
2906 if (ret < 0)
2907 goto out_ret;
2908 }
2909 }
2910 if (list_empty(sdt_notes))
2911 ret = -ENOENT;
2912
2913 out_ret:
2914 return ret;
2915 }
2916
2917 /**
2918 * get_sdt_note_list : Wrapper to construct a list of sdt notes
2919 * @head : empty list_head
2920 * @target : file to find SDT notes from
2921 *
2922 * This opens the file, initializes
2923 * the ELF and then calls construct_sdt_notes_list.
2924 */
get_sdt_note_list(struct list_head * head,const char * target)2925 int get_sdt_note_list(struct list_head *head, const char *target)
2926 {
2927 Elf *elf;
2928 int fd, ret;
2929
2930 fd = open(target, O_RDONLY);
2931 if (fd < 0)
2932 return -EBADF;
2933
2934 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2935 if (!elf) {
2936 ret = -EBADF;
2937 goto out_close;
2938 }
2939 ret = construct_sdt_notes_list(elf, head);
2940 elf_end(elf);
2941 out_close:
2942 close(fd);
2943 return ret;
2944 }
2945
2946 /**
2947 * cleanup_sdt_note_list : free the sdt notes' list
2948 * @sdt_notes: sdt notes' list
2949 *
2950 * Free up the SDT notes in @sdt_notes.
2951 * Returns the number of SDT notes free'd.
2952 */
cleanup_sdt_note_list(struct list_head * sdt_notes)2953 int cleanup_sdt_note_list(struct list_head *sdt_notes)
2954 {
2955 struct sdt_note *tmp, *pos;
2956 int nr_free = 0;
2957
2958 list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
2959 list_del_init(&pos->note_list);
2960 zfree(&pos->args);
2961 zfree(&pos->name);
2962 zfree(&pos->provider);
2963 free(pos);
2964 nr_free++;
2965 }
2966 return nr_free;
2967 }
2968
2969 /**
2970 * sdt_notes__get_count: Counts the number of sdt events
2971 * @start: list_head to sdt_notes list
2972 *
2973 * Returns the number of SDT notes in a list
2974 */
sdt_notes__get_count(struct list_head * start)2975 int sdt_notes__get_count(struct list_head *start)
2976 {
2977 struct sdt_note *sdt_ptr;
2978 int count = 0;
2979
2980 list_for_each_entry(sdt_ptr, start, note_list)
2981 count++;
2982 return count;
2983 }
2984 #endif
2985
symbol__elf_init(void)2986 void symbol__elf_init(void)
2987 {
2988 elf_version(EV_CURRENT);
2989 }
2990