Merge remote-tracking branch 'torvalds/master' into perf/core
[linux-2.6-microblaze.git] / tools / perf / util / symbol-elf.c
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-ocaml.h"
16 #include "demangle-java.h"
17 #include "demangle-rust.h"
18 #include "machine.h"
19 #include "vdso.h"
20 #include "debug.h"
21 #include "util/copyfile.h"
22 #include <linux/ctype.h>
23 #include <linux/kernel.h>
24 #include <linux/zalloc.h>
25 #include <symbol/kallsyms.h>
26 #include <internal/lib.h>
27
28 #ifndef EM_AARCH64
29 #define EM_AARCH64      183  /* ARM 64 bit */
30 #endif
31
32 #ifndef ELF32_ST_VISIBILITY
33 #define ELF32_ST_VISIBILITY(o)  ((o) & 0x03)
34 #endif
35
36 /* For ELF64 the definitions are the same.  */
37 #ifndef ELF64_ST_VISIBILITY
38 #define ELF64_ST_VISIBILITY(o)  ELF32_ST_VISIBILITY (o)
39 #endif
40
41 /* How to extract information held in the st_other field.  */
42 #ifndef GELF_ST_VISIBILITY
43 #define GELF_ST_VISIBILITY(val) ELF64_ST_VISIBILITY (val)
44 #endif
45
46 typedef Elf64_Nhdr GElf_Nhdr;
47
48 #ifndef DMGL_PARAMS
49 #define DMGL_NO_OPTS     0              /* For readability... */
50 #define DMGL_PARAMS      (1 << 0)       /* Include function args */
51 #define DMGL_ANSI        (1 << 1)       /* Include const, volatile, etc */
52 #endif
53
54 #ifdef HAVE_LIBBFD_SUPPORT
55 #define PACKAGE 'perf'
56 #include <bfd.h>
57 #else
58 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
59 extern char *cplus_demangle(const char *, int);
60
61 static inline char *bfd_demangle(void __maybe_unused *v, const char *c, int i)
62 {
63         return cplus_demangle(c, i);
64 }
65 #else
66 #ifdef NO_DEMANGLE
67 static inline char *bfd_demangle(void __maybe_unused *v,
68                                  const char __maybe_unused *c,
69                                  int __maybe_unused i)
70 {
71         return NULL;
72 }
73 #endif
74 #endif
75 #endif
76
77 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
78 static int elf_getphdrnum(Elf *elf, size_t *dst)
79 {
80         GElf_Ehdr gehdr;
81         GElf_Ehdr *ehdr;
82
83         ehdr = gelf_getehdr(elf, &gehdr);
84         if (!ehdr)
85                 return -1;
86
87         *dst = ehdr->e_phnum;
88
89         return 0;
90 }
91 #endif
92
93 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
94 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
95 {
96         pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
97         return -1;
98 }
99 #endif
100
101 #ifndef NT_GNU_BUILD_ID
102 #define NT_GNU_BUILD_ID 3
103 #endif
104
105 /**
106  * elf_symtab__for_each_symbol - iterate thru all the symbols
107  *
108  * @syms: struct elf_symtab instance to iterate
109  * @idx: uint32_t idx
110  * @sym: GElf_Sym iterator
111  */
112 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
113         for (idx = 0, gelf_getsym(syms, idx, &sym);\
114              idx < nr_syms; \
115              idx++, gelf_getsym(syms, idx, &sym))
116
117 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
118 {
119         return GELF_ST_TYPE(sym->st_info);
120 }
121
122 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
123 {
124         return GELF_ST_VISIBILITY(sym->st_other);
125 }
126
127 #ifndef STT_GNU_IFUNC
128 #define STT_GNU_IFUNC 10
129 #endif
130
131 static inline int elf_sym__is_function(const GElf_Sym *sym)
132 {
133         return (elf_sym__type(sym) == STT_FUNC ||
134                 elf_sym__type(sym) == STT_GNU_IFUNC) &&
135                sym->st_name != 0 &&
136                sym->st_shndx != SHN_UNDEF;
137 }
138
139 static inline bool elf_sym__is_object(const GElf_Sym *sym)
140 {
141         return elf_sym__type(sym) == STT_OBJECT &&
142                 sym->st_name != 0 &&
143                 sym->st_shndx != SHN_UNDEF;
144 }
145
146 static inline int elf_sym__is_label(const GElf_Sym *sym)
147 {
148         return elf_sym__type(sym) == STT_NOTYPE &&
149                 sym->st_name != 0 &&
150                 sym->st_shndx != SHN_UNDEF &&
151                 sym->st_shndx != SHN_ABS &&
152                 elf_sym__visibility(sym) != STV_HIDDEN &&
153                 elf_sym__visibility(sym) != STV_INTERNAL;
154 }
155
156 static bool elf_sym__filter(GElf_Sym *sym)
157 {
158         return elf_sym__is_function(sym) || elf_sym__is_object(sym);
159 }
160
161 static inline const char *elf_sym__name(const GElf_Sym *sym,
162                                         const Elf_Data *symstrs)
163 {
164         return symstrs->d_buf + sym->st_name;
165 }
166
167 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
168                                         const Elf_Data *secstrs)
169 {
170         return secstrs->d_buf + shdr->sh_name;
171 }
172
173 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
174                                         const Elf_Data *secstrs)
175 {
176         return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
177 }
178
179 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
180                                     const Elf_Data *secstrs)
181 {
182         return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
183 }
184
185 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
186 {
187         return elf_sec__is_text(shdr, secstrs) || 
188                elf_sec__is_data(shdr, secstrs);
189 }
190
191 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
192 {
193         Elf_Scn *sec = NULL;
194         GElf_Shdr shdr;
195         size_t cnt = 1;
196
197         while ((sec = elf_nextscn(elf, sec)) != NULL) {
198                 gelf_getshdr(sec, &shdr);
199
200                 if ((addr >= shdr.sh_addr) &&
201                     (addr < (shdr.sh_addr + shdr.sh_size)))
202                         return cnt;
203
204                 ++cnt;
205         }
206
207         return -1;
208 }
209
210 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
211                              GElf_Shdr *shp, const char *name, size_t *idx)
212 {
213         Elf_Scn *sec = NULL;
214         size_t cnt = 1;
215
216         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
217         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
218                 return NULL;
219
220         while ((sec = elf_nextscn(elf, sec)) != NULL) {
221                 char *str;
222
223                 gelf_getshdr(sec, shp);
224                 str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
225                 if (str && !strcmp(name, str)) {
226                         if (idx)
227                                 *idx = cnt;
228                         return sec;
229                 }
230                 ++cnt;
231         }
232
233         return NULL;
234 }
235
236 static bool want_demangle(bool is_kernel_sym)
237 {
238         return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
239 }
240
241 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
242 {
243         int demangle_flags = verbose > 0 ? (DMGL_PARAMS | DMGL_ANSI) : DMGL_NO_OPTS;
244         char *demangled = NULL;
245
246         /*
247          * We need to figure out if the object was created from C++ sources
248          * DWARF DW_compile_unit has this, but we don't always have access
249          * to it...
250          */
251         if (!want_demangle(dso->kernel || kmodule))
252             return demangled;
253
254         demangled = bfd_demangle(NULL, elf_name, demangle_flags);
255         if (demangled == NULL) {
256                 demangled = ocaml_demangle_sym(elf_name);
257                 if (demangled == NULL) {
258                         demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
259                 }
260         }
261         else if (rust_is_mangled(demangled))
262                 /*
263                     * Input to Rust demangling is the BFD-demangled
264                     * name which it Rust-demangles in place.
265                     */
266                 rust_demangle_sym(demangled);
267
268         return demangled;
269 }
270
271 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
272         for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
273              idx < nr_entries; \
274              ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
275
276 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
277         for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
278              idx < nr_entries; \
279              ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
280
281 /*
282  * We need to check if we have a .dynsym, so that we can handle the
283  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
284  * .dynsym or .symtab).
285  * And always look at the original dso, not at debuginfo packages, that
286  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
287  */
288 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
289 {
290         uint32_t nr_rel_entries, idx;
291         GElf_Sym sym;
292         u64 plt_offset, plt_header_size, plt_entry_size;
293         GElf_Shdr shdr_plt;
294         struct symbol *f;
295         GElf_Shdr shdr_rel_plt, shdr_dynsym;
296         Elf_Data *reldata, *syms, *symstrs;
297         Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
298         size_t dynsym_idx;
299         GElf_Ehdr ehdr;
300         char sympltname[1024];
301         Elf *elf;
302         int nr = 0, symidx, err = 0;
303
304         if (!ss->dynsym)
305                 return 0;
306
307         elf = ss->elf;
308         ehdr = ss->ehdr;
309
310         scn_dynsym = ss->dynsym;
311         shdr_dynsym = ss->dynshdr;
312         dynsym_idx = ss->dynsym_idx;
313
314         if (scn_dynsym == NULL)
315                 goto out_elf_end;
316
317         scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
318                                           ".rela.plt", NULL);
319         if (scn_plt_rel == NULL) {
320                 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
321                                                   ".rel.plt", NULL);
322                 if (scn_plt_rel == NULL)
323                         goto out_elf_end;
324         }
325
326         err = -1;
327
328         if (shdr_rel_plt.sh_link != dynsym_idx)
329                 goto out_elf_end;
330
331         if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
332                 goto out_elf_end;
333
334         /*
335          * Fetch the relocation section to find the idxes to the GOT
336          * and the symbols in the .dynsym they refer to.
337          */
338         reldata = elf_getdata(scn_plt_rel, NULL);
339         if (reldata == NULL)
340                 goto out_elf_end;
341
342         syms = elf_getdata(scn_dynsym, NULL);
343         if (syms == NULL)
344                 goto out_elf_end;
345
346         scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
347         if (scn_symstrs == NULL)
348                 goto out_elf_end;
349
350         symstrs = elf_getdata(scn_symstrs, NULL);
351         if (symstrs == NULL)
352                 goto out_elf_end;
353
354         if (symstrs->d_size == 0)
355                 goto out_elf_end;
356
357         nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
358         plt_offset = shdr_plt.sh_offset;
359         switch (ehdr.e_machine) {
360                 case EM_ARM:
361                         plt_header_size = 20;
362                         plt_entry_size = 12;
363                         break;
364
365                 case EM_AARCH64:
366                         plt_header_size = 32;
367                         plt_entry_size = 16;
368                         break;
369
370                 case EM_SPARC:
371                         plt_header_size = 48;
372                         plt_entry_size = 12;
373                         break;
374
375                 case EM_SPARCV9:
376                         plt_header_size = 128;
377                         plt_entry_size = 32;
378                         break;
379
380                 default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
381                         plt_header_size = shdr_plt.sh_entsize;
382                         plt_entry_size = shdr_plt.sh_entsize;
383                         break;
384         }
385         plt_offset += plt_header_size;
386
387         if (shdr_rel_plt.sh_type == SHT_RELA) {
388                 GElf_Rela pos_mem, *pos;
389
390                 elf_section__for_each_rela(reldata, pos, pos_mem, idx,
391                                            nr_rel_entries) {
392                         const char *elf_name = NULL;
393                         char *demangled = NULL;
394                         symidx = GELF_R_SYM(pos->r_info);
395                         gelf_getsym(syms, symidx, &sym);
396
397                         elf_name = elf_sym__name(&sym, symstrs);
398                         demangled = demangle_sym(dso, 0, elf_name);
399                         if (demangled != NULL)
400                                 elf_name = demangled;
401                         snprintf(sympltname, sizeof(sympltname),
402                                  "%s@plt", elf_name);
403                         free(demangled);
404
405                         f = symbol__new(plt_offset, plt_entry_size,
406                                         STB_GLOBAL, STT_FUNC, sympltname);
407                         if (!f)
408                                 goto out_elf_end;
409
410                         plt_offset += plt_entry_size;
411                         symbols__insert(&dso->symbols, f);
412                         ++nr;
413                 }
414         } else if (shdr_rel_plt.sh_type == SHT_REL) {
415                 GElf_Rel pos_mem, *pos;
416                 elf_section__for_each_rel(reldata, pos, pos_mem, idx,
417                                           nr_rel_entries) {
418                         const char *elf_name = NULL;
419                         char *demangled = NULL;
420                         symidx = GELF_R_SYM(pos->r_info);
421                         gelf_getsym(syms, symidx, &sym);
422
423                         elf_name = elf_sym__name(&sym, symstrs);
424                         demangled = demangle_sym(dso, 0, elf_name);
425                         if (demangled != NULL)
426                                 elf_name = demangled;
427                         snprintf(sympltname, sizeof(sympltname),
428                                  "%s@plt", elf_name);
429                         free(demangled);
430
431                         f = symbol__new(plt_offset, plt_entry_size,
432                                         STB_GLOBAL, STT_FUNC, sympltname);
433                         if (!f)
434                                 goto out_elf_end;
435
436                         plt_offset += plt_entry_size;
437                         symbols__insert(&dso->symbols, f);
438                         ++nr;
439                 }
440         }
441
442         err = 0;
443 out_elf_end:
444         if (err == 0)
445                 return nr;
446         pr_debug("%s: problems reading %s PLT info.\n",
447                  __func__, dso->long_name);
448         return 0;
449 }
450
451 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
452 {
453         return demangle_sym(dso, kmodule, elf_name);
454 }
455
456 /*
457  * Align offset to 4 bytes as needed for note name and descriptor data.
458  */
459 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
460
461 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
462 {
463         int err = -1;
464         GElf_Ehdr ehdr;
465         GElf_Shdr shdr;
466         Elf_Data *data;
467         Elf_Scn *sec;
468         Elf_Kind ek;
469         void *ptr;
470
471         if (size < BUILD_ID_SIZE)
472                 goto out;
473
474         ek = elf_kind(elf);
475         if (ek != ELF_K_ELF)
476                 goto out;
477
478         if (gelf_getehdr(elf, &ehdr) == NULL) {
479                 pr_err("%s: cannot get elf header.\n", __func__);
480                 goto out;
481         }
482
483         /*
484          * Check following sections for notes:
485          *   '.note.gnu.build-id'
486          *   '.notes'
487          *   '.note' (VDSO specific)
488          */
489         do {
490                 sec = elf_section_by_name(elf, &ehdr, &shdr,
491                                           ".note.gnu.build-id", NULL);
492                 if (sec)
493                         break;
494
495                 sec = elf_section_by_name(elf, &ehdr, &shdr,
496                                           ".notes", NULL);
497                 if (sec)
498                         break;
499
500                 sec = elf_section_by_name(elf, &ehdr, &shdr,
501                                           ".note", NULL);
502                 if (sec)
503                         break;
504
505                 return err;
506
507         } while (0);
508
509         data = elf_getdata(sec, NULL);
510         if (data == NULL)
511                 goto out;
512
513         ptr = data->d_buf;
514         while (ptr < (data->d_buf + data->d_size)) {
515                 GElf_Nhdr *nhdr = ptr;
516                 size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
517                        descsz = NOTE_ALIGN(nhdr->n_descsz);
518                 const char *name;
519
520                 ptr += sizeof(*nhdr);
521                 name = ptr;
522                 ptr += namesz;
523                 if (nhdr->n_type == NT_GNU_BUILD_ID &&
524                     nhdr->n_namesz == sizeof("GNU")) {
525                         if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
526                                 size_t sz = min(size, descsz);
527                                 memcpy(bf, ptr, sz);
528                                 memset(bf + sz, 0, size - sz);
529                                 err = descsz;
530                                 break;
531                         }
532                 }
533                 ptr += descsz;
534         }
535
536 out:
537         return err;
538 }
539
540 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
541
542 static int read_build_id(const char *filename, struct build_id *bid)
543 {
544         size_t size = sizeof(bid->data);
545         int err = -1;
546         bfd *abfd;
547
548         abfd = bfd_openr(filename, NULL);
549         if (!abfd)
550                 return -1;
551
552         if (!bfd_check_format(abfd, bfd_object)) {
553                 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
554                 goto out_close;
555         }
556
557         if (!abfd->build_id || abfd->build_id->size > size)
558                 goto out_close;
559
560         memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
561         memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
562         err = bid->size = abfd->build_id->size;
563
564 out_close:
565         bfd_close(abfd);
566         return err;
567 }
568
569 #else // HAVE_LIBBFD_BUILDID_SUPPORT
570
571 static int read_build_id(const char *filename, struct build_id *bid)
572 {
573         size_t size = sizeof(bid->data);
574         int fd, err = -1;
575         Elf *elf;
576
577         if (size < BUILD_ID_SIZE)
578                 goto out;
579
580         fd = open(filename, O_RDONLY);
581         if (fd < 0)
582                 goto out;
583
584         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
585         if (elf == NULL) {
586                 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
587                 goto out_close;
588         }
589
590         err = elf_read_build_id(elf, bid->data, size);
591         if (err > 0)
592                 bid->size = err;
593
594         elf_end(elf);
595 out_close:
596         close(fd);
597 out:
598         return err;
599 }
600
601 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
602
603 int filename__read_build_id(const char *filename, struct build_id *bid)
604 {
605         struct kmod_path m = { .name = NULL, };
606         char path[PATH_MAX];
607         int err;
608
609         if (!filename)
610                 return -EFAULT;
611
612         err = kmod_path__parse(&m, filename);
613         if (err)
614                 return -1;
615
616         if (m.comp) {
617                 int error = 0, fd;
618
619                 fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
620                 if (fd < 0) {
621                         pr_debug("Failed to decompress (error %d) %s\n",
622                                  error, filename);
623                         return -1;
624                 }
625                 close(fd);
626                 filename = path;
627         }
628
629         err = read_build_id(filename, bid);
630
631         if (m.comp)
632                 unlink(filename);
633         return err;
634 }
635
636 int sysfs__read_build_id(const char *filename, struct build_id *bid)
637 {
638         size_t size = sizeof(bid->data);
639         int fd, err = -1;
640
641         fd = open(filename, O_RDONLY);
642         if (fd < 0)
643                 goto out;
644
645         while (1) {
646                 char bf[BUFSIZ];
647                 GElf_Nhdr nhdr;
648                 size_t namesz, descsz;
649
650                 if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
651                         break;
652
653                 namesz = NOTE_ALIGN(nhdr.n_namesz);
654                 descsz = NOTE_ALIGN(nhdr.n_descsz);
655                 if (nhdr.n_type == NT_GNU_BUILD_ID &&
656                     nhdr.n_namesz == sizeof("GNU")) {
657                         if (read(fd, bf, namesz) != (ssize_t)namesz)
658                                 break;
659                         if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
660                                 size_t sz = min(descsz, size);
661                                 if (read(fd, bid->data, sz) == (ssize_t)sz) {
662                                         memset(bid->data + sz, 0, size - sz);
663                                         bid->size = sz;
664                                         err = 0;
665                                         break;
666                                 }
667                         } else if (read(fd, bf, descsz) != (ssize_t)descsz)
668                                 break;
669                 } else {
670                         int n = namesz + descsz;
671
672                         if (n > (int)sizeof(bf)) {
673                                 n = sizeof(bf);
674                                 pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
675                                          __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
676                         }
677                         if (read(fd, bf, n) != n)
678                                 break;
679                 }
680         }
681         close(fd);
682 out:
683         return err;
684 }
685
686 #ifdef HAVE_LIBBFD_SUPPORT
687
688 int filename__read_debuglink(const char *filename, char *debuglink,
689                              size_t size)
690 {
691         int err = -1;
692         asection *section;
693         bfd *abfd;
694
695         abfd = bfd_openr(filename, NULL);
696         if (!abfd)
697                 return -1;
698
699         if (!bfd_check_format(abfd, bfd_object)) {
700                 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
701                 goto out_close;
702         }
703
704         section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
705         if (!section)
706                 goto out_close;
707
708         if (section->size > size)
709                 goto out_close;
710
711         if (!bfd_get_section_contents(abfd, section, debuglink, 0,
712                                       section->size))
713                 goto out_close;
714
715         err = 0;
716
717 out_close:
718         bfd_close(abfd);
719         return err;
720 }
721
722 #else
723
724 int filename__read_debuglink(const char *filename, char *debuglink,
725                              size_t size)
726 {
727         int fd, err = -1;
728         Elf *elf;
729         GElf_Ehdr ehdr;
730         GElf_Shdr shdr;
731         Elf_Data *data;
732         Elf_Scn *sec;
733         Elf_Kind ek;
734
735         fd = open(filename, O_RDONLY);
736         if (fd < 0)
737                 goto out;
738
739         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
740         if (elf == NULL) {
741                 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
742                 goto out_close;
743         }
744
745         ek = elf_kind(elf);
746         if (ek != ELF_K_ELF)
747                 goto out_elf_end;
748
749         if (gelf_getehdr(elf, &ehdr) == NULL) {
750                 pr_err("%s: cannot get elf header.\n", __func__);
751                 goto out_elf_end;
752         }
753
754         sec = elf_section_by_name(elf, &ehdr, &shdr,
755                                   ".gnu_debuglink", NULL);
756         if (sec == NULL)
757                 goto out_elf_end;
758
759         data = elf_getdata(sec, NULL);
760         if (data == NULL)
761                 goto out_elf_end;
762
763         /* the start of this section is a zero-terminated string */
764         strncpy(debuglink, data->d_buf, size);
765
766         err = 0;
767
768 out_elf_end:
769         elf_end(elf);
770 out_close:
771         close(fd);
772 out:
773         return err;
774 }
775
776 #endif
777
778 static int dso__swap_init(struct dso *dso, unsigned char eidata)
779 {
780         static unsigned int const endian = 1;
781
782         dso->needs_swap = DSO_SWAP__NO;
783
784         switch (eidata) {
785         case ELFDATA2LSB:
786                 /* We are big endian, DSO is little endian. */
787                 if (*(unsigned char const *)&endian != 1)
788                         dso->needs_swap = DSO_SWAP__YES;
789                 break;
790
791         case ELFDATA2MSB:
792                 /* We are little endian, DSO is big endian. */
793                 if (*(unsigned char const *)&endian != 0)
794                         dso->needs_swap = DSO_SWAP__YES;
795                 break;
796
797         default:
798                 pr_err("unrecognized DSO data encoding %d\n", eidata);
799                 return -EINVAL;
800         }
801
802         return 0;
803 }
804
805 bool symsrc__possibly_runtime(struct symsrc *ss)
806 {
807         return ss->dynsym || ss->opdsec;
808 }
809
810 bool symsrc__has_symtab(struct symsrc *ss)
811 {
812         return ss->symtab != NULL;
813 }
814
815 void symsrc__destroy(struct symsrc *ss)
816 {
817         zfree(&ss->name);
818         elf_end(ss->elf);
819         close(ss->fd);
820 }
821
822 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
823 {
824         /*
825          * Usually vmlinux is an ELF file with type ET_EXEC for most
826          * architectures; except Arm64 kernel is linked with option
827          * '-share', so need to check type ET_DYN.
828          */
829         return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
830                ehdr.e_type == ET_DYN;
831 }
832
833 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
834                  enum dso_binary_type type)
835 {
836         GElf_Ehdr ehdr;
837         Elf *elf;
838         int fd;
839
840         if (dso__needs_decompress(dso)) {
841                 fd = dso__decompress_kmodule_fd(dso, name);
842                 if (fd < 0)
843                         return -1;
844
845                 type = dso->symtab_type;
846         } else {
847                 fd = open(name, O_RDONLY);
848                 if (fd < 0) {
849                         dso->load_errno = errno;
850                         return -1;
851                 }
852         }
853
854         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
855         if (elf == NULL) {
856                 pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
857                 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
858                 goto out_close;
859         }
860
861         if (gelf_getehdr(elf, &ehdr) == NULL) {
862                 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
863                 pr_debug("%s: cannot get elf header.\n", __func__);
864                 goto out_elf_end;
865         }
866
867         if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
868                 dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
869                 goto out_elf_end;
870         }
871
872         /* Always reject images with a mismatched build-id: */
873         if (dso->has_build_id && !symbol_conf.ignore_vmlinux_buildid) {
874                 u8 build_id[BUILD_ID_SIZE];
875                 struct build_id bid;
876                 int size;
877
878                 size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
879                 if (size <= 0) {
880                         dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
881                         goto out_elf_end;
882                 }
883
884                 build_id__init(&bid, build_id, size);
885                 if (!dso__build_id_equal(dso, &bid)) {
886                         pr_debug("%s: build id mismatch for %s.\n", __func__, name);
887                         dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
888                         goto out_elf_end;
889                 }
890         }
891
892         ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
893
894         ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
895                         NULL);
896         if (ss->symshdr.sh_type != SHT_SYMTAB)
897                 ss->symtab = NULL;
898
899         ss->dynsym_idx = 0;
900         ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
901                         &ss->dynsym_idx);
902         if (ss->dynshdr.sh_type != SHT_DYNSYM)
903                 ss->dynsym = NULL;
904
905         ss->opdidx = 0;
906         ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
907                         &ss->opdidx);
908         if (ss->opdshdr.sh_type != SHT_PROGBITS)
909                 ss->opdsec = NULL;
910
911         if (dso->kernel == DSO_SPACE__USER)
912                 ss->adjust_symbols = true;
913         else
914                 ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
915
916         ss->name   = strdup(name);
917         if (!ss->name) {
918                 dso->load_errno = errno;
919                 goto out_elf_end;
920         }
921
922         ss->elf    = elf;
923         ss->fd     = fd;
924         ss->ehdr   = ehdr;
925         ss->type   = type;
926
927         return 0;
928
929 out_elf_end:
930         elf_end(elf);
931 out_close:
932         close(fd);
933         return -1;
934 }
935
936 /**
937  * ref_reloc_sym_not_found - has kernel relocation symbol been found.
938  * @kmap: kernel maps and relocation reference symbol
939  *
940  * This function returns %true if we are dealing with the kernel maps and the
941  * relocation reference symbol has not yet been found.  Otherwise %false is
942  * returned.
943  */
944 static bool ref_reloc_sym_not_found(struct kmap *kmap)
945 {
946         return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
947                !kmap->ref_reloc_sym->unrelocated_addr;
948 }
949
950 /**
951  * ref_reloc - kernel relocation offset.
952  * @kmap: kernel maps and relocation reference symbol
953  *
954  * This function returns the offset of kernel addresses as determined by using
955  * the relocation reference symbol i.e. if the kernel has not been relocated
956  * then the return value is zero.
957  */
958 static u64 ref_reloc(struct kmap *kmap)
959 {
960         if (kmap && kmap->ref_reloc_sym &&
961             kmap->ref_reloc_sym->unrelocated_addr)
962                 return kmap->ref_reloc_sym->addr -
963                        kmap->ref_reloc_sym->unrelocated_addr;
964         return 0;
965 }
966
967 void __weak arch__sym_update(struct symbol *s __maybe_unused,
968                 GElf_Sym *sym __maybe_unused) { }
969
970 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
971                                       GElf_Sym *sym, GElf_Shdr *shdr,
972                                       struct maps *kmaps, struct kmap *kmap,
973                                       struct dso **curr_dsop, struct map **curr_mapp,
974                                       const char *section_name,
975                                       bool adjust_kernel_syms, bool kmodule, bool *remap_kernel)
976 {
977         struct dso *curr_dso = *curr_dsop;
978         struct map *curr_map;
979         char dso_name[PATH_MAX];
980
981         /* Adjust symbol to map to file offset */
982         if (adjust_kernel_syms)
983                 sym->st_value -= shdr->sh_addr - shdr->sh_offset;
984
985         if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0)
986                 return 0;
987
988         if (strcmp(section_name, ".text") == 0) {
989                 /*
990                  * The initial kernel mapping is based on
991                  * kallsyms and identity maps.  Overwrite it to
992                  * map to the kernel dso.
993                  */
994                 if (*remap_kernel && dso->kernel && !kmodule) {
995                         *remap_kernel = false;
996                         map->start = shdr->sh_addr + ref_reloc(kmap);
997                         map->end = map->start + shdr->sh_size;
998                         map->pgoff = shdr->sh_offset;
999                         map->map_ip = map__map_ip;
1000                         map->unmap_ip = map__unmap_ip;
1001                         /* Ensure maps are correctly ordered */
1002                         if (kmaps) {
1003                                 map__get(map);
1004                                 maps__remove(kmaps, map);
1005                                 maps__insert(kmaps, map);
1006                                 map__put(map);
1007                         }
1008                 }
1009
1010                 /*
1011                  * The initial module mapping is based on
1012                  * /proc/modules mapped to offset zero.
1013                  * Overwrite it to map to the module dso.
1014                  */
1015                 if (*remap_kernel && kmodule) {
1016                         *remap_kernel = false;
1017                         map->pgoff = shdr->sh_offset;
1018                 }
1019
1020                 *curr_mapp = map;
1021                 *curr_dsop = dso;
1022                 return 0;
1023         }
1024
1025         if (!kmap)
1026                 return 0;
1027
1028         snprintf(dso_name, sizeof(dso_name), "%s%s", dso->short_name, section_name);
1029
1030         curr_map = maps__find_by_name(kmaps, dso_name);
1031         if (curr_map == NULL) {
1032                 u64 start = sym->st_value;
1033
1034                 if (kmodule)
1035                         start += map->start + shdr->sh_offset;
1036
1037                 curr_dso = dso__new(dso_name);
1038                 if (curr_dso == NULL)
1039                         return -1;
1040                 curr_dso->kernel = dso->kernel;
1041                 curr_dso->long_name = dso->long_name;
1042                 curr_dso->long_name_len = dso->long_name_len;
1043                 curr_map = map__new2(start, curr_dso);
1044                 dso__put(curr_dso);
1045                 if (curr_map == NULL)
1046                         return -1;
1047
1048                 if (curr_dso->kernel)
1049                         map__kmap(curr_map)->kmaps = kmaps;
1050
1051                 if (adjust_kernel_syms) {
1052                         curr_map->start  = shdr->sh_addr + ref_reloc(kmap);
1053                         curr_map->end    = curr_map->start + shdr->sh_size;
1054                         curr_map->pgoff  = shdr->sh_offset;
1055                 } else {
1056                         curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
1057                 }
1058                 curr_dso->symtab_type = dso->symtab_type;
1059                 maps__insert(kmaps, curr_map);
1060                 /*
1061                  * Add it before we drop the reference to curr_map, i.e. while
1062                  * we still are sure to have a reference to this DSO via
1063                  * *curr_map->dso.
1064                  */
1065                 dsos__add(&kmaps->machine->dsos, curr_dso);
1066                 /* kmaps already got it */
1067                 map__put(curr_map);
1068                 dso__set_loaded(curr_dso);
1069                 *curr_mapp = curr_map;
1070                 *curr_dsop = curr_dso;
1071         } else
1072                 *curr_dsop = curr_map->dso;
1073
1074         return 0;
1075 }
1076
1077 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1078                   struct symsrc *runtime_ss, int kmodule)
1079 {
1080         struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
1081         struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1082         struct map *curr_map = map;
1083         struct dso *curr_dso = dso;
1084         Elf_Data *symstrs, *secstrs;
1085         uint32_t nr_syms;
1086         int err = -1;
1087         uint32_t idx;
1088         GElf_Ehdr ehdr;
1089         GElf_Shdr shdr;
1090         GElf_Shdr tshdr;
1091         Elf_Data *syms, *opddata = NULL;
1092         GElf_Sym sym;
1093         Elf_Scn *sec, *sec_strndx;
1094         Elf *elf;
1095         int nr = 0;
1096         bool remap_kernel = false, adjust_kernel_syms = false;
1097
1098         if (kmap && !kmaps)
1099                 return -1;
1100
1101         dso->symtab_type = syms_ss->type;
1102         dso->is_64_bit = syms_ss->is_64_bit;
1103         dso->rel = syms_ss->ehdr.e_type == ET_REL;
1104
1105         /*
1106          * Modules may already have symbols from kallsyms, but those symbols
1107          * have the wrong values for the dso maps, so remove them.
1108          */
1109         if (kmodule && syms_ss->symtab)
1110                 symbols__delete(&dso->symbols);
1111
1112         if (!syms_ss->symtab) {
1113                 /*
1114                  * If the vmlinux is stripped, fail so we will fall back
1115                  * to using kallsyms. The vmlinux runtime symbols aren't
1116                  * of much use.
1117                  */
1118                 if (dso->kernel)
1119                         goto out_elf_end;
1120
1121                 syms_ss->symtab  = syms_ss->dynsym;
1122                 syms_ss->symshdr = syms_ss->dynshdr;
1123         }
1124
1125         elf = syms_ss->elf;
1126         ehdr = syms_ss->ehdr;
1127         sec = syms_ss->symtab;
1128         shdr = syms_ss->symshdr;
1129
1130         if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1131                                 ".text", NULL))
1132                 dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
1133
1134         if (runtime_ss->opdsec)
1135                 opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1136
1137         syms = elf_getdata(sec, NULL);
1138         if (syms == NULL)
1139                 goto out_elf_end;
1140
1141         sec = elf_getscn(elf, shdr.sh_link);
1142         if (sec == NULL)
1143                 goto out_elf_end;
1144
1145         symstrs = elf_getdata(sec, NULL);
1146         if (symstrs == NULL)
1147                 goto out_elf_end;
1148
1149         sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1150         if (sec_strndx == NULL)
1151                 goto out_elf_end;
1152
1153         secstrs = elf_getdata(sec_strndx, NULL);
1154         if (secstrs == NULL)
1155                 goto out_elf_end;
1156
1157         nr_syms = shdr.sh_size / shdr.sh_entsize;
1158
1159         memset(&sym, 0, sizeof(sym));
1160
1161         /*
1162          * The kernel relocation symbol is needed in advance in order to adjust
1163          * kernel maps correctly.
1164          */
1165         if (ref_reloc_sym_not_found(kmap)) {
1166                 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1167                         const char *elf_name = elf_sym__name(&sym, symstrs);
1168
1169                         if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1170                                 continue;
1171                         kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1172                         map->reloc = kmap->ref_reloc_sym->addr -
1173                                      kmap->ref_reloc_sym->unrelocated_addr;
1174                         break;
1175                 }
1176         }
1177
1178         /*
1179          * Handle any relocation of vdso necessary because older kernels
1180          * attempted to prelink vdso to its virtual address.
1181          */
1182         if (dso__is_vdso(dso))
1183                 map->reloc = map->start - dso->text_offset;
1184
1185         dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
1186         /*
1187          * Initial kernel and module mappings do not map to the dso.
1188          * Flag the fixups.
1189          */
1190         if (dso->kernel) {
1191                 remap_kernel = true;
1192                 adjust_kernel_syms = dso->adjust_symbols;
1193         }
1194         elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1195                 struct symbol *f;
1196                 const char *elf_name = elf_sym__name(&sym, symstrs);
1197                 char *demangled = NULL;
1198                 int is_label = elf_sym__is_label(&sym);
1199                 const char *section_name;
1200                 bool used_opd = false;
1201
1202                 if (!is_label && !elf_sym__filter(&sym))
1203                         continue;
1204
1205                 /* Reject ARM ELF "mapping symbols": these aren't unique and
1206                  * don't identify functions, so will confuse the profile
1207                  * output: */
1208                 if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1209                         if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1210                             && (elf_name[2] == '\0' || elf_name[2] == '.'))
1211                                 continue;
1212                 }
1213
1214                 if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1215                         u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1216                         u64 *opd = opddata->d_buf + offset;
1217                         sym.st_value = DSO__SWAP(dso, u64, *opd);
1218                         sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1219                                         sym.st_value);
1220                         used_opd = true;
1221                 }
1222                 /*
1223                  * When loading symbols in a data mapping, ABS symbols (which
1224                  * has a value of SHN_ABS in its st_shndx) failed at
1225                  * elf_getscn().  And it marks the loading as a failure so
1226                  * already loaded symbols cannot be fixed up.
1227                  *
1228                  * I'm not sure what should be done. Just ignore them for now.
1229                  * - Namhyung Kim
1230                  */
1231                 if (sym.st_shndx == SHN_ABS)
1232                         continue;
1233
1234                 sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1235                 if (!sec)
1236                         goto out_elf_end;
1237
1238                 gelf_getshdr(sec, &shdr);
1239
1240                 /*
1241                  * We have to fallback to runtime when syms' section header has
1242                  * NOBITS set. NOBITS results in file offset (sh_offset) not
1243                  * being incremented. So sh_offset used below has different
1244                  * values for syms (invalid) and runtime (valid).
1245                  */
1246                 if (shdr.sh_type == SHT_NOBITS) {
1247                         sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1248                         if (!sec)
1249                                 goto out_elf_end;
1250
1251                         gelf_getshdr(sec, &shdr);
1252                 }
1253
1254                 if (is_label && !elf_sec__filter(&shdr, secstrs))
1255                         continue;
1256
1257                 section_name = elf_sec__name(&shdr, secstrs);
1258
1259                 /* On ARM, symbols for thumb functions have 1 added to
1260                  * the symbol address as a flag - remove it */
1261                 if ((ehdr.e_machine == EM_ARM) &&
1262                     (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1263                     (sym.st_value & 1))
1264                         --sym.st_value;
1265
1266                 if (dso->kernel) {
1267                         if (dso__process_kernel_symbol(dso, map, &sym, &shdr, kmaps, kmap, &curr_dso, &curr_map,
1268                                                        section_name, adjust_kernel_syms, kmodule, &remap_kernel))
1269                                 goto out_elf_end;
1270                 } else if ((used_opd && runtime_ss->adjust_symbols) ||
1271                            (!used_opd && syms_ss->adjust_symbols)) {
1272                         pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1273                                   "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", __func__,
1274                                   (u64)sym.st_value, (u64)shdr.sh_addr,
1275                                   (u64)shdr.sh_offset);
1276                         sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1277                 }
1278
1279                 demangled = demangle_sym(dso, kmodule, elf_name);
1280                 if (demangled != NULL)
1281                         elf_name = demangled;
1282
1283                 f = symbol__new(sym.st_value, sym.st_size,
1284                                 GELF_ST_BIND(sym.st_info),
1285                                 GELF_ST_TYPE(sym.st_info), elf_name);
1286                 free(demangled);
1287                 if (!f)
1288                         goto out_elf_end;
1289
1290                 arch__sym_update(f, &sym);
1291
1292                 __symbols__insert(&curr_dso->symbols, f, dso->kernel);
1293                 nr++;
1294         }
1295
1296         /*
1297          * For misannotated, zeroed, ASM function sizes.
1298          */
1299         if (nr > 0) {
1300                 symbols__fixup_end(&dso->symbols);
1301                 symbols__fixup_duplicate(&dso->symbols);
1302                 if (kmap) {
1303                         /*
1304                          * We need to fixup this here too because we create new
1305                          * maps here, for things like vsyscall sections.
1306                          */
1307                         maps__fixup_end(kmaps);
1308                 }
1309         }
1310         err = nr;
1311 out_elf_end:
1312         return err;
1313 }
1314
1315 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1316 {
1317         GElf_Phdr phdr;
1318         size_t i, phdrnum;
1319         int err;
1320         u64 sz;
1321
1322         if (elf_getphdrnum(elf, &phdrnum))
1323                 return -1;
1324
1325         for (i = 0; i < phdrnum; i++) {
1326                 if (gelf_getphdr(elf, i, &phdr) == NULL)
1327                         return -1;
1328                 if (phdr.p_type != PT_LOAD)
1329                         continue;
1330                 if (exe) {
1331                         if (!(phdr.p_flags & PF_X))
1332                                 continue;
1333                 } else {
1334                         if (!(phdr.p_flags & PF_R))
1335                                 continue;
1336                 }
1337                 sz = min(phdr.p_memsz, phdr.p_filesz);
1338                 if (!sz)
1339                         continue;
1340                 err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1341                 if (err)
1342                         return err;
1343         }
1344         return 0;
1345 }
1346
1347 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1348                     bool *is_64_bit)
1349 {
1350         int err;
1351         Elf *elf;
1352
1353         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1354         if (elf == NULL)
1355                 return -1;
1356
1357         if (is_64_bit)
1358                 *is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1359
1360         err = elf_read_maps(elf, exe, mapfn, data);
1361
1362         elf_end(elf);
1363         return err;
1364 }
1365
1366 enum dso_type dso__type_fd(int fd)
1367 {
1368         enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1369         GElf_Ehdr ehdr;
1370         Elf_Kind ek;
1371         Elf *elf;
1372
1373         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1374         if (elf == NULL)
1375                 goto out;
1376
1377         ek = elf_kind(elf);
1378         if (ek != ELF_K_ELF)
1379                 goto out_end;
1380
1381         if (gelf_getclass(elf) == ELFCLASS64) {
1382                 dso_type = DSO__TYPE_64BIT;
1383                 goto out_end;
1384         }
1385
1386         if (gelf_getehdr(elf, &ehdr) == NULL)
1387                 goto out_end;
1388
1389         if (ehdr.e_machine == EM_X86_64)
1390                 dso_type = DSO__TYPE_X32BIT;
1391         else
1392                 dso_type = DSO__TYPE_32BIT;
1393 out_end:
1394         elf_end(elf);
1395 out:
1396         return dso_type;
1397 }
1398
1399 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1400 {
1401         ssize_t r;
1402         size_t n;
1403         int err = -1;
1404         char *buf = malloc(page_size);
1405
1406         if (buf == NULL)
1407                 return -1;
1408
1409         if (lseek(to, to_offs, SEEK_SET) != to_offs)
1410                 goto out;
1411
1412         if (lseek(from, from_offs, SEEK_SET) != from_offs)
1413                 goto out;
1414
1415         while (len) {
1416                 n = page_size;
1417                 if (len < n)
1418                         n = len;
1419                 /* Use read because mmap won't work on proc files */
1420                 r = read(from, buf, n);
1421                 if (r < 0)
1422                         goto out;
1423                 if (!r)
1424                         break;
1425                 n = r;
1426                 r = write(to, buf, n);
1427                 if (r < 0)
1428                         goto out;
1429                 if ((size_t)r != n)
1430                         goto out;
1431                 len -= n;
1432         }
1433
1434         err = 0;
1435 out:
1436         free(buf);
1437         return err;
1438 }
1439
1440 struct kcore {
1441         int fd;
1442         int elfclass;
1443         Elf *elf;
1444         GElf_Ehdr ehdr;
1445 };
1446
1447 static int kcore__open(struct kcore *kcore, const char *filename)
1448 {
1449         GElf_Ehdr *ehdr;
1450
1451         kcore->fd = open(filename, O_RDONLY);
1452         if (kcore->fd == -1)
1453                 return -1;
1454
1455         kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1456         if (!kcore->elf)
1457                 goto out_close;
1458
1459         kcore->elfclass = gelf_getclass(kcore->elf);
1460         if (kcore->elfclass == ELFCLASSNONE)
1461                 goto out_end;
1462
1463         ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1464         if (!ehdr)
1465                 goto out_end;
1466
1467         return 0;
1468
1469 out_end:
1470         elf_end(kcore->elf);
1471 out_close:
1472         close(kcore->fd);
1473         return -1;
1474 }
1475
1476 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1477                        bool temp)
1478 {
1479         kcore->elfclass = elfclass;
1480
1481         if (temp)
1482                 kcore->fd = mkstemp(filename);
1483         else
1484                 kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1485         if (kcore->fd == -1)
1486                 return -1;
1487
1488         kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1489         if (!kcore->elf)
1490                 goto out_close;
1491
1492         if (!gelf_newehdr(kcore->elf, elfclass))
1493                 goto out_end;
1494
1495         memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1496
1497         return 0;
1498
1499 out_end:
1500         elf_end(kcore->elf);
1501 out_close:
1502         close(kcore->fd);
1503         unlink(filename);
1504         return -1;
1505 }
1506
1507 static void kcore__close(struct kcore *kcore)
1508 {
1509         elf_end(kcore->elf);
1510         close(kcore->fd);
1511 }
1512
1513 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1514 {
1515         GElf_Ehdr *ehdr = &to->ehdr;
1516         GElf_Ehdr *kehdr = &from->ehdr;
1517
1518         memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1519         ehdr->e_type      = kehdr->e_type;
1520         ehdr->e_machine   = kehdr->e_machine;
1521         ehdr->e_version   = kehdr->e_version;
1522         ehdr->e_entry     = 0;
1523         ehdr->e_shoff     = 0;
1524         ehdr->e_flags     = kehdr->e_flags;
1525         ehdr->e_phnum     = count;
1526         ehdr->e_shentsize = 0;
1527         ehdr->e_shnum     = 0;
1528         ehdr->e_shstrndx  = 0;
1529
1530         if (from->elfclass == ELFCLASS32) {
1531                 ehdr->e_phoff     = sizeof(Elf32_Ehdr);
1532                 ehdr->e_ehsize    = sizeof(Elf32_Ehdr);
1533                 ehdr->e_phentsize = sizeof(Elf32_Phdr);
1534         } else {
1535                 ehdr->e_phoff     = sizeof(Elf64_Ehdr);
1536                 ehdr->e_ehsize    = sizeof(Elf64_Ehdr);
1537                 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1538         }
1539
1540         if (!gelf_update_ehdr(to->elf, ehdr))
1541                 return -1;
1542
1543         if (!gelf_newphdr(to->elf, count))
1544                 return -1;
1545
1546         return 0;
1547 }
1548
1549 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
1550                            u64 addr, u64 len)
1551 {
1552         GElf_Phdr phdr = {
1553                 .p_type         = PT_LOAD,
1554                 .p_flags        = PF_R | PF_W | PF_X,
1555                 .p_offset       = offset,
1556                 .p_vaddr        = addr,
1557                 .p_paddr        = 0,
1558                 .p_filesz       = len,
1559                 .p_memsz        = len,
1560                 .p_align        = page_size,
1561         };
1562
1563         if (!gelf_update_phdr(kcore->elf, idx, &phdr))
1564                 return -1;
1565
1566         return 0;
1567 }
1568
1569 static off_t kcore__write(struct kcore *kcore)
1570 {
1571         return elf_update(kcore->elf, ELF_C_WRITE);
1572 }
1573
1574 struct phdr_data {
1575         off_t offset;
1576         off_t rel;
1577         u64 addr;
1578         u64 len;
1579         struct list_head node;
1580         struct phdr_data *remaps;
1581 };
1582
1583 struct sym_data {
1584         u64 addr;
1585         struct list_head node;
1586 };
1587
1588 struct kcore_copy_info {
1589         u64 stext;
1590         u64 etext;
1591         u64 first_symbol;
1592         u64 last_symbol;
1593         u64 first_module;
1594         u64 first_module_symbol;
1595         u64 last_module_symbol;
1596         size_t phnum;
1597         struct list_head phdrs;
1598         struct list_head syms;
1599 };
1600
1601 #define kcore_copy__for_each_phdr(k, p) \
1602         list_for_each_entry((p), &(k)->phdrs, node)
1603
1604 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
1605 {
1606         struct phdr_data *p = zalloc(sizeof(*p));
1607
1608         if (p) {
1609                 p->addr   = addr;
1610                 p->len    = len;
1611                 p->offset = offset;
1612         }
1613
1614         return p;
1615 }
1616
1617 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
1618                                                  u64 addr, u64 len,
1619                                                  off_t offset)
1620 {
1621         struct phdr_data *p = phdr_data__new(addr, len, offset);
1622
1623         if (p)
1624                 list_add_tail(&p->node, &kci->phdrs);
1625
1626         return p;
1627 }
1628
1629 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
1630 {
1631         struct phdr_data *p, *tmp;
1632
1633         list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
1634                 list_del_init(&p->node);
1635                 free(p);
1636         }
1637 }
1638
1639 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
1640                                             u64 addr)
1641 {
1642         struct sym_data *s = zalloc(sizeof(*s));
1643
1644         if (s) {
1645                 s->addr = addr;
1646                 list_add_tail(&s->node, &kci->syms);
1647         }
1648
1649         return s;
1650 }
1651
1652 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
1653 {
1654         struct sym_data *s, *tmp;
1655
1656         list_for_each_entry_safe(s, tmp, &kci->syms, node) {
1657                 list_del_init(&s->node);
1658                 free(s);
1659         }
1660 }
1661
1662 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
1663                                         u64 start)
1664 {
1665         struct kcore_copy_info *kci = arg;
1666
1667         if (!kallsyms__is_function(type))
1668                 return 0;
1669
1670         if (strchr(name, '[')) {
1671                 if (!kci->first_module_symbol || start < kci->first_module_symbol)
1672                         kci->first_module_symbol = start;
1673                 if (start > kci->last_module_symbol)
1674                         kci->last_module_symbol = start;
1675                 return 0;
1676         }
1677
1678         if (!kci->first_symbol || start < kci->first_symbol)
1679                 kci->first_symbol = start;
1680
1681         if (!kci->last_symbol || start > kci->last_symbol)
1682                 kci->last_symbol = start;
1683
1684         if (!strcmp(name, "_stext")) {
1685                 kci->stext = start;
1686                 return 0;
1687         }
1688
1689         if (!strcmp(name, "_etext")) {
1690                 kci->etext = start;
1691                 return 0;
1692         }
1693
1694         if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
1695                 return -1;
1696
1697         return 0;
1698 }
1699
1700 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
1701                                       const char *dir)
1702 {
1703         char kallsyms_filename[PATH_MAX];
1704
1705         scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
1706
1707         if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
1708                 return -1;
1709
1710         if (kallsyms__parse(kallsyms_filename, kci,
1711                             kcore_copy__process_kallsyms) < 0)
1712                 return -1;
1713
1714         return 0;
1715 }
1716
1717 static int kcore_copy__process_modules(void *arg,
1718                                        const char *name __maybe_unused,
1719                                        u64 start, u64 size __maybe_unused)
1720 {
1721         struct kcore_copy_info *kci = arg;
1722
1723         if (!kci->first_module || start < kci->first_module)
1724                 kci->first_module = start;
1725
1726         return 0;
1727 }
1728
1729 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
1730                                      const char *dir)
1731 {
1732         char modules_filename[PATH_MAX];
1733
1734         scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
1735
1736         if (symbol__restricted_filename(modules_filename, "/proc/modules"))
1737                 return -1;
1738
1739         if (modules__parse(modules_filename, kci,
1740                            kcore_copy__process_modules) < 0)
1741                 return -1;
1742
1743         return 0;
1744 }
1745
1746 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
1747                            u64 pgoff, u64 s, u64 e)
1748 {
1749         u64 len, offset;
1750
1751         if (s < start || s >= end)
1752                 return 0;
1753
1754         offset = (s - start) + pgoff;
1755         len = e < end ? e - s : end - s;
1756
1757         return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
1758 }
1759
1760 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
1761 {
1762         struct kcore_copy_info *kci = data;
1763         u64 end = start + len;
1764         struct sym_data *sdat;
1765
1766         if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
1767                 return -1;
1768
1769         if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
1770                             kci->last_module_symbol))
1771                 return -1;
1772
1773         list_for_each_entry(sdat, &kci->syms, node) {
1774                 u64 s = round_down(sdat->addr, page_size);
1775
1776                 if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
1777                         return -1;
1778         }
1779
1780         return 0;
1781 }
1782
1783 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
1784 {
1785         if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
1786                 return -1;
1787
1788         return 0;
1789 }
1790
1791 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
1792 {
1793         struct phdr_data *p, *k = NULL;
1794         u64 kend;
1795
1796         if (!kci->stext)
1797                 return;
1798
1799         /* Find phdr that corresponds to the kernel map (contains stext) */
1800         kcore_copy__for_each_phdr(kci, p) {
1801                 u64 pend = p->addr + p->len - 1;
1802
1803                 if (p->addr <= kci->stext && pend >= kci->stext) {
1804                         k = p;
1805                         break;
1806                 }
1807         }
1808
1809         if (!k)
1810                 return;
1811
1812         kend = k->offset + k->len;
1813
1814         /* Find phdrs that remap the kernel */
1815         kcore_copy__for_each_phdr(kci, p) {
1816                 u64 pend = p->offset + p->len;
1817
1818                 if (p == k)
1819                         continue;
1820
1821                 if (p->offset >= k->offset && pend <= kend)
1822                         p->remaps = k;
1823         }
1824 }
1825
1826 static void kcore_copy__layout(struct kcore_copy_info *kci)
1827 {
1828         struct phdr_data *p;
1829         off_t rel = 0;
1830
1831         kcore_copy__find_remaps(kci);
1832
1833         kcore_copy__for_each_phdr(kci, p) {
1834                 if (!p->remaps) {
1835                         p->rel = rel;
1836                         rel += p->len;
1837                 }
1838                 kci->phnum += 1;
1839         }
1840
1841         kcore_copy__for_each_phdr(kci, p) {
1842                 struct phdr_data *k = p->remaps;
1843
1844                 if (k)
1845                         p->rel = p->offset - k->offset + k->rel;
1846         }
1847 }
1848
1849 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
1850                                  Elf *elf)
1851 {
1852         if (kcore_copy__parse_kallsyms(kci, dir))
1853                 return -1;
1854
1855         if (kcore_copy__parse_modules(kci, dir))
1856                 return -1;
1857
1858         if (kci->stext)
1859                 kci->stext = round_down(kci->stext, page_size);
1860         else
1861                 kci->stext = round_down(kci->first_symbol, page_size);
1862
1863         if (kci->etext) {
1864                 kci->etext = round_up(kci->etext, page_size);
1865         } else if (kci->last_symbol) {
1866                 kci->etext = round_up(kci->last_symbol, page_size);
1867                 kci->etext += page_size;
1868         }
1869
1870         if (kci->first_module_symbol &&
1871             (!kci->first_module || kci->first_module_symbol < kci->first_module))
1872                 kci->first_module = kci->first_module_symbol;
1873
1874         kci->first_module = round_down(kci->first_module, page_size);
1875
1876         if (kci->last_module_symbol) {
1877                 kci->last_module_symbol = round_up(kci->last_module_symbol,
1878                                                    page_size);
1879                 kci->last_module_symbol += page_size;
1880         }
1881
1882         if (!kci->stext || !kci->etext)
1883                 return -1;
1884
1885         if (kci->first_module && !kci->last_module_symbol)
1886                 return -1;
1887
1888         if (kcore_copy__read_maps(kci, elf))
1889                 return -1;
1890
1891         kcore_copy__layout(kci);
1892
1893         return 0;
1894 }
1895
1896 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
1897                                  const char *name)
1898 {
1899         char from_filename[PATH_MAX];
1900         char to_filename[PATH_MAX];
1901
1902         scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1903         scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1904
1905         return copyfile_mode(from_filename, to_filename, 0400);
1906 }
1907
1908 static int kcore_copy__unlink(const char *dir, const char *name)
1909 {
1910         char filename[PATH_MAX];
1911
1912         scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
1913
1914         return unlink(filename);
1915 }
1916
1917 static int kcore_copy__compare_fds(int from, int to)
1918 {
1919         char *buf_from;
1920         char *buf_to;
1921         ssize_t ret;
1922         size_t len;
1923         int err = -1;
1924
1925         buf_from = malloc(page_size);
1926         buf_to = malloc(page_size);
1927         if (!buf_from || !buf_to)
1928                 goto out;
1929
1930         while (1) {
1931                 /* Use read because mmap won't work on proc files */
1932                 ret = read(from, buf_from, page_size);
1933                 if (ret < 0)
1934                         goto out;
1935
1936                 if (!ret)
1937                         break;
1938
1939                 len = ret;
1940
1941                 if (readn(to, buf_to, len) != (int)len)
1942                         goto out;
1943
1944                 if (memcmp(buf_from, buf_to, len))
1945                         goto out;
1946         }
1947
1948         err = 0;
1949 out:
1950         free(buf_to);
1951         free(buf_from);
1952         return err;
1953 }
1954
1955 static int kcore_copy__compare_files(const char *from_filename,
1956                                      const char *to_filename)
1957 {
1958         int from, to, err = -1;
1959
1960         from = open(from_filename, O_RDONLY);
1961         if (from < 0)
1962                 return -1;
1963
1964         to = open(to_filename, O_RDONLY);
1965         if (to < 0)
1966                 goto out_close_from;
1967
1968         err = kcore_copy__compare_fds(from, to);
1969
1970         close(to);
1971 out_close_from:
1972         close(from);
1973         return err;
1974 }
1975
1976 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
1977                                     const char *name)
1978 {
1979         char from_filename[PATH_MAX];
1980         char to_filename[PATH_MAX];
1981
1982         scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
1983         scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
1984
1985         return kcore_copy__compare_files(from_filename, to_filename);
1986 }
1987
1988 /**
1989  * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
1990  * @from_dir: from directory
1991  * @to_dir: to directory
1992  *
1993  * This function copies kallsyms, modules and kcore files from one directory to
1994  * another.  kallsyms and modules are copied entirely.  Only code segments are
1995  * copied from kcore.  It is assumed that two segments suffice: one for the
1996  * kernel proper and one for all the modules.  The code segments are determined
1997  * from kallsyms and modules files.  The kernel map starts at _stext or the
1998  * lowest function symbol, and ends at _etext or the highest function symbol.
1999  * The module map starts at the lowest module address and ends at the highest
2000  * module symbol.  Start addresses are rounded down to the nearest page.  End
2001  * addresses are rounded up to the nearest page.  An extra page is added to the
2002  * highest kernel symbol and highest module symbol to, hopefully, encompass that
2003  * symbol too.  Because it contains only code sections, the resulting kcore is
2004  * unusual.  One significant peculiarity is that the mapping (start -> pgoff)
2005  * is not the same for the kernel map and the modules map.  That happens because
2006  * the data is copied adjacently whereas the original kcore has gaps.  Finally,
2007  * kallsyms and modules files are compared with their copies to check that
2008  * modules have not been loaded or unloaded while the copies were taking place.
2009  *
2010  * Return: %0 on success, %-1 on failure.
2011  */
2012 int kcore_copy(const char *from_dir, const char *to_dir)
2013 {
2014         struct kcore kcore;
2015         struct kcore extract;
2016         int idx = 0, err = -1;
2017         off_t offset, sz;
2018         struct kcore_copy_info kci = { .stext = 0, };
2019         char kcore_filename[PATH_MAX];
2020         char extract_filename[PATH_MAX];
2021         struct phdr_data *p;
2022
2023         INIT_LIST_HEAD(&kci.phdrs);
2024         INIT_LIST_HEAD(&kci.syms);
2025
2026         if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2027                 return -1;
2028
2029         if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2030                 goto out_unlink_kallsyms;
2031
2032         scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2033         scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2034
2035         if (kcore__open(&kcore, kcore_filename))
2036                 goto out_unlink_modules;
2037
2038         if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2039                 goto out_kcore_close;
2040
2041         if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2042                 goto out_kcore_close;
2043
2044         if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2045                 goto out_extract_close;
2046
2047         offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2048                  gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2049         offset = round_up(offset, page_size);
2050
2051         kcore_copy__for_each_phdr(&kci, p) {
2052                 off_t offs = p->rel + offset;
2053
2054                 if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2055                         goto out_extract_close;
2056         }
2057
2058         sz = kcore__write(&extract);
2059         if (sz < 0 || sz > offset)
2060                 goto out_extract_close;
2061
2062         kcore_copy__for_each_phdr(&kci, p) {
2063                 off_t offs = p->rel + offset;
2064
2065                 if (p->remaps)
2066                         continue;
2067                 if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2068                         goto out_extract_close;
2069         }
2070
2071         if (kcore_copy__compare_file(from_dir, to_dir, "modules"))
2072                 goto out_extract_close;
2073
2074         if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2075                 goto out_extract_close;
2076
2077         err = 0;
2078
2079 out_extract_close:
2080         kcore__close(&extract);
2081         if (err)
2082                 unlink(extract_filename);
2083 out_kcore_close:
2084         kcore__close(&kcore);
2085 out_unlink_modules:
2086         if (err)
2087                 kcore_copy__unlink(to_dir, "modules");
2088 out_unlink_kallsyms:
2089         if (err)
2090                 kcore_copy__unlink(to_dir, "kallsyms");
2091
2092         kcore_copy__free_phdrs(&kci);
2093         kcore_copy__free_syms(&kci);
2094
2095         return err;
2096 }
2097
2098 int kcore_extract__create(struct kcore_extract *kce)
2099 {
2100         struct kcore kcore;
2101         struct kcore extract;
2102         size_t count = 1;
2103         int idx = 0, err = -1;
2104         off_t offset = page_size, sz;
2105
2106         if (kcore__open(&kcore, kce->kcore_filename))
2107                 return -1;
2108
2109         strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2110         if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2111                 goto out_kcore_close;
2112
2113         if (kcore__copy_hdr(&kcore, &extract, count))
2114                 goto out_extract_close;
2115
2116         if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2117                 goto out_extract_close;
2118
2119         sz = kcore__write(&extract);
2120         if (sz < 0 || sz > offset)
2121                 goto out_extract_close;
2122
2123         if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2124                 goto out_extract_close;
2125
2126         err = 0;
2127
2128 out_extract_close:
2129         kcore__close(&extract);
2130         if (err)
2131                 unlink(kce->extract_filename);
2132 out_kcore_close:
2133         kcore__close(&kcore);
2134
2135         return err;
2136 }
2137
2138 void kcore_extract__delete(struct kcore_extract *kce)
2139 {
2140         unlink(kce->extract_filename);
2141 }
2142
2143 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2144
2145 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2146 {
2147         if (!base_off)
2148                 return;
2149
2150         if (tmp->bit32)
2151                 tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2152                         tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2153                         tmp->addr.a32[SDT_NOTE_IDX_BASE];
2154         else
2155                 tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2156                         tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2157                         tmp->addr.a64[SDT_NOTE_IDX_BASE];
2158 }
2159
2160 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2161                               GElf_Addr base_off)
2162 {
2163         if (!base_off)
2164                 return;
2165
2166         if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2167                 tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2168         else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2169                 tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2170 }
2171
2172 /**
2173  * populate_sdt_note : Parse raw data and identify SDT note
2174  * @elf: elf of the opened file
2175  * @data: raw data of a section with description offset applied
2176  * @len: note description size
2177  * @type: type of the note
2178  * @sdt_notes: List to add the SDT note
2179  *
2180  * Responsible for parsing the @data in section .note.stapsdt in @elf and
2181  * if its an SDT note, it appends to @sdt_notes list.
2182  */
2183 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2184                              struct list_head *sdt_notes)
2185 {
2186         const char *provider, *name, *args;
2187         struct sdt_note *tmp = NULL;
2188         GElf_Ehdr ehdr;
2189         GElf_Shdr shdr;
2190         int ret = -EINVAL;
2191
2192         union {
2193                 Elf64_Addr a64[NR_ADDR];
2194                 Elf32_Addr a32[NR_ADDR];
2195         } buf;
2196
2197         Elf_Data dst = {
2198                 .d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2199                 .d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2200                 .d_off = 0, .d_align = 0
2201         };
2202         Elf_Data src = {
2203                 .d_buf = (void *) data, .d_type = ELF_T_ADDR,
2204                 .d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2205                 .d_align = 0
2206         };
2207
2208         tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2209         if (!tmp) {
2210                 ret = -ENOMEM;
2211                 goto out_err;
2212         }
2213
2214         INIT_LIST_HEAD(&tmp->note_list);
2215
2216         if (len < dst.d_size + 3)
2217                 goto out_free_note;
2218
2219         /* Translation from file representation to memory representation */
2220         if (gelf_xlatetom(*elf, &dst, &src,
2221                           elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2222                 pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2223                 goto out_free_note;
2224         }
2225
2226         /* Populate the fields of sdt_note */
2227         provider = data + dst.d_size;
2228
2229         name = (const char *)memchr(provider, '\0', data + len - provider);
2230         if (name++ == NULL)
2231                 goto out_free_note;
2232
2233         tmp->provider = strdup(provider);
2234         if (!tmp->provider) {
2235                 ret = -ENOMEM;
2236                 goto out_free_note;
2237         }
2238         tmp->name = strdup(name);
2239         if (!tmp->name) {
2240                 ret = -ENOMEM;
2241                 goto out_free_prov;
2242         }
2243
2244         args = memchr(name, '\0', data + len - name);
2245
2246         /*
2247          * There is no argument if:
2248          * - We reached the end of the note;
2249          * - There is not enough room to hold a potential string;
2250          * - The argument string is empty or just contains ':'.
2251          */
2252         if (args == NULL || data + len - args < 2 ||
2253                 args[1] == ':' || args[1] == '\0')
2254                 tmp->args = NULL;
2255         else {
2256                 tmp->args = strdup(++args);
2257                 if (!tmp->args) {
2258                         ret = -ENOMEM;
2259                         goto out_free_name;
2260                 }
2261         }
2262
2263         if (gelf_getclass(*elf) == ELFCLASS32) {
2264                 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2265                 tmp->bit32 = true;
2266         } else {
2267                 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2268                 tmp->bit32 = false;
2269         }
2270
2271         if (!gelf_getehdr(*elf, &ehdr)) {
2272                 pr_debug("%s : cannot get elf header.\n", __func__);
2273                 ret = -EBADF;
2274                 goto out_free_args;
2275         }
2276
2277         /* Adjust the prelink effect :
2278          * Find out the .stapsdt.base section.
2279          * This scn will help us to handle prelinking (if present).
2280          * Compare the retrieved file offset of the base section with the
2281          * base address in the description of the SDT note. If its different,
2282          * then accordingly, adjust the note location.
2283          */
2284         if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2285                 sdt_adjust_loc(tmp, shdr.sh_offset);
2286
2287         /* Adjust reference counter offset */
2288         if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2289                 sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2290
2291         list_add_tail(&tmp->note_list, sdt_notes);
2292         return 0;
2293
2294 out_free_args:
2295         zfree(&tmp->args);
2296 out_free_name:
2297         zfree(&tmp->name);
2298 out_free_prov:
2299         zfree(&tmp->provider);
2300 out_free_note:
2301         free(tmp);
2302 out_err:
2303         return ret;
2304 }
2305
2306 /**
2307  * construct_sdt_notes_list : constructs a list of SDT notes
2308  * @elf : elf to look into
2309  * @sdt_notes : empty list_head
2310  *
2311  * Scans the sections in 'elf' for the section
2312  * .note.stapsdt. It, then calls populate_sdt_note to find
2313  * out the SDT events and populates the 'sdt_notes'.
2314  */
2315 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2316 {
2317         GElf_Ehdr ehdr;
2318         Elf_Scn *scn = NULL;
2319         Elf_Data *data;
2320         GElf_Shdr shdr;
2321         size_t shstrndx, next;
2322         GElf_Nhdr nhdr;
2323         size_t name_off, desc_off, offset;
2324         int ret = 0;
2325
2326         if (gelf_getehdr(elf, &ehdr) == NULL) {
2327                 ret = -EBADF;
2328                 goto out_ret;
2329         }
2330         if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2331                 ret = -EBADF;
2332                 goto out_ret;
2333         }
2334
2335         /* Look for the required section */
2336         scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2337         if (!scn) {
2338                 ret = -ENOENT;
2339                 goto out_ret;
2340         }
2341
2342         if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2343                 ret = -ENOENT;
2344                 goto out_ret;
2345         }
2346
2347         data = elf_getdata(scn, NULL);
2348
2349         /* Get the SDT notes */
2350         for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2351                                               &desc_off)) > 0; offset = next) {
2352                 if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2353                     !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2354                             sizeof(SDT_NOTE_NAME))) {
2355                         /* Check the type of the note */
2356                         if (nhdr.n_type != SDT_NOTE_TYPE)
2357                                 goto out_ret;
2358
2359                         ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2360                                                 nhdr.n_descsz, sdt_notes);
2361                         if (ret < 0)
2362                                 goto out_ret;
2363                 }
2364         }
2365         if (list_empty(sdt_notes))
2366                 ret = -ENOENT;
2367
2368 out_ret:
2369         return ret;
2370 }
2371
2372 /**
2373  * get_sdt_note_list : Wrapper to construct a list of sdt notes
2374  * @head : empty list_head
2375  * @target : file to find SDT notes from
2376  *
2377  * This opens the file, initializes
2378  * the ELF and then calls construct_sdt_notes_list.
2379  */
2380 int get_sdt_note_list(struct list_head *head, const char *target)
2381 {
2382         Elf *elf;
2383         int fd, ret;
2384
2385         fd = open(target, O_RDONLY);
2386         if (fd < 0)
2387                 return -EBADF;
2388
2389         elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2390         if (!elf) {
2391                 ret = -EBADF;
2392                 goto out_close;
2393         }
2394         ret = construct_sdt_notes_list(elf, head);
2395         elf_end(elf);
2396 out_close:
2397         close(fd);
2398         return ret;
2399 }
2400
2401 /**
2402  * cleanup_sdt_note_list : free the sdt notes' list
2403  * @sdt_notes: sdt notes' list
2404  *
2405  * Free up the SDT notes in @sdt_notes.
2406  * Returns the number of SDT notes free'd.
2407  */
2408 int cleanup_sdt_note_list(struct list_head *sdt_notes)
2409 {
2410         struct sdt_note *tmp, *pos;
2411         int nr_free = 0;
2412
2413         list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
2414                 list_del_init(&pos->note_list);
2415                 zfree(&pos->args);
2416                 zfree(&pos->name);
2417                 zfree(&pos->provider);
2418                 free(pos);
2419                 nr_free++;
2420         }
2421         return nr_free;
2422 }
2423
2424 /**
2425  * sdt_notes__get_count: Counts the number of sdt events
2426  * @start: list_head to sdt_notes list
2427  *
2428  * Returns the number of SDT notes in a list
2429  */
2430 int sdt_notes__get_count(struct list_head *start)
2431 {
2432         struct sdt_note *sdt_ptr;
2433         int count = 0;
2434
2435         list_for_each_entry(sdt_ptr, start, note_list)
2436                 count++;
2437         return count;
2438 }
2439 #endif
2440
2441 void symbol__elf_init(void)
2442 {
2443         elf_version(EV_CURRENT);
2444 }