1 // SPDX-License-Identifier: GPL-2.0
15 #include "demangle-cxx.h"
16 #include "demangle-ocaml.h"
17 #include "demangle-java.h"
18 #include "demangle-rust.h"
22 #include "util/copyfile.h"
23 #include <linux/ctype.h>
24 #include <linux/kernel.h>
25 #include <linux/zalloc.h>
26 #include <symbol/kallsyms.h>
27 #include <internal/lib.h>
29 #ifdef HAVE_LIBBFD_SUPPORT
30 #define PACKAGE 'perf'
34 #if defined(HAVE_LIBBFD_SUPPORT) || defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
36 #define DMGL_PARAMS (1 << 0) /* Include function args */
37 #define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */
42 #define EM_AARCH64 183 /* ARM 64 bit */
46 #define EM_LOONGARCH 258
49 #ifndef ELF32_ST_VISIBILITY
50 #define ELF32_ST_VISIBILITY(o) ((o) & 0x03)
53 /* For ELF64 the definitions are the same. */
54 #ifndef ELF64_ST_VISIBILITY
55 #define ELF64_ST_VISIBILITY(o) ELF32_ST_VISIBILITY (o)
58 /* How to extract information held in the st_other field. */
59 #ifndef GELF_ST_VISIBILITY
60 #define GELF_ST_VISIBILITY(val) ELF64_ST_VISIBILITY (val)
63 typedef Elf64_Nhdr GElf_Nhdr;
66 #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT
67 static int elf_getphdrnum(Elf *elf, size_t *dst)
72 ehdr = gelf_getehdr(elf, &gehdr);
82 #ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
83 static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
85 pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
90 #ifndef NT_GNU_BUILD_ID
91 #define NT_GNU_BUILD_ID 3
95 * elf_symtab__for_each_symbol - iterate thru all the symbols
97 * @syms: struct elf_symtab instance to iterate
99 * @sym: GElf_Sym iterator
101 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
102 for (idx = 0, gelf_getsym(syms, idx, &sym);\
104 idx++, gelf_getsym(syms, idx, &sym))
106 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
108 return GELF_ST_TYPE(sym->st_info);
111 static inline uint8_t elf_sym__visibility(const GElf_Sym *sym)
113 return GELF_ST_VISIBILITY(sym->st_other);
116 #ifndef STT_GNU_IFUNC
117 #define STT_GNU_IFUNC 10
120 static inline int elf_sym__is_function(const GElf_Sym *sym)
122 return (elf_sym__type(sym) == STT_FUNC ||
123 elf_sym__type(sym) == STT_GNU_IFUNC) &&
125 sym->st_shndx != SHN_UNDEF;
128 static inline bool elf_sym__is_object(const GElf_Sym *sym)
130 return elf_sym__type(sym) == STT_OBJECT &&
132 sym->st_shndx != SHN_UNDEF;
135 static inline int elf_sym__is_label(const GElf_Sym *sym)
137 return elf_sym__type(sym) == STT_NOTYPE &&
139 sym->st_shndx != SHN_UNDEF &&
140 sym->st_shndx != SHN_ABS &&
141 elf_sym__visibility(sym) != STV_HIDDEN &&
142 elf_sym__visibility(sym) != STV_INTERNAL;
145 static bool elf_sym__filter(GElf_Sym *sym)
147 return elf_sym__is_function(sym) || elf_sym__is_object(sym);
150 static inline const char *elf_sym__name(const GElf_Sym *sym,
151 const Elf_Data *symstrs)
153 return symstrs->d_buf + sym->st_name;
156 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
157 const Elf_Data *secstrs)
159 return secstrs->d_buf + shdr->sh_name;
162 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
163 const Elf_Data *secstrs)
165 return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
168 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
169 const Elf_Data *secstrs)
171 return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
174 static bool elf_sec__filter(GElf_Shdr *shdr, Elf_Data *secstrs)
176 return elf_sec__is_text(shdr, secstrs) ||
177 elf_sec__is_data(shdr, secstrs);
180 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
186 while ((sec = elf_nextscn(elf, sec)) != NULL) {
187 gelf_getshdr(sec, &shdr);
189 if ((addr >= shdr.sh_addr) &&
190 (addr < (shdr.sh_addr + shdr.sh_size)))
199 Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
200 GElf_Shdr *shp, const char *name, size_t *idx)
205 /* ELF is corrupted/truncated, avoid calling elf_strptr. */
206 if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL))
209 while ((sec = elf_nextscn(elf, sec)) != NULL) {
212 gelf_getshdr(sec, shp);
213 str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
214 if (str && !strcmp(name, str)) {
225 bool filename__has_section(const char *filename, const char *sec)
233 fd = open(filename, O_RDONLY);
237 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
241 if (gelf_getehdr(elf, &ehdr) == NULL)
244 found = !!elf_section_by_name(elf, &ehdr, &shdr, sec, NULL);
253 static int elf_read_program_header(Elf *elf, u64 vaddr, GElf_Phdr *phdr)
258 if (elf_getphdrnum(elf, &phdrnum))
261 for (i = 0; i < phdrnum; i++) {
262 if (gelf_getphdr(elf, i, phdr) == NULL)
265 if (phdr->p_type != PT_LOAD)
268 sz = max(phdr->p_memsz, phdr->p_filesz);
272 if (vaddr >= phdr->p_vaddr && (vaddr < phdr->p_vaddr + sz))
276 /* Not found any valid program header */
280 static bool want_demangle(bool is_kernel_sym)
282 return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle;
286 * Demangle C++ function signature, typically replaced by demangle-cxx.cpp
289 __weak char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused,
290 bool modifiers __maybe_unused)
292 #ifdef HAVE_LIBBFD_SUPPORT
293 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
295 return bfd_demangle(NULL, str, flags);
296 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT)
297 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0);
299 return cplus_demangle(str, flags);
305 static char *demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
307 char *demangled = NULL;
310 * We need to figure out if the object was created from C++ sources
311 * DWARF DW_compile_unit has this, but we don't always have access
314 if (!want_demangle(dso->kernel || kmodule))
317 demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0);
318 if (demangled == NULL) {
319 demangled = ocaml_demangle_sym(elf_name);
320 if (demangled == NULL) {
321 demangled = java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET);
324 else if (rust_is_mangled(demangled))
326 * Input to Rust demangling is the BFD-demangled
327 * name which it Rust-demangles in place.
329 rust_demangle_sym(demangled);
343 static u32 get_rel_symidx(struct rel_info *ri, u32 idx)
345 idx = ri->sorted ? ri->sorted[idx] : idx;
347 gelf_getrela(ri->reldata, idx, &ri->rela);
348 return GELF_R_SYM(ri->rela.r_info);
350 gelf_getrel(ri->reldata, idx, &ri->rel);
351 return GELF_R_SYM(ri->rel.r_info);
354 static u64 get_rel_offset(struct rel_info *ri, u32 x)
359 gelf_getrela(ri->reldata, x, &rela);
360 return rela.r_offset;
364 gelf_getrel(ri->reldata, x, &rel);
369 static int rel_cmp(const void *a, const void *b, void *r)
371 struct rel_info *ri = r;
372 u64 a_offset = get_rel_offset(ri, *(const u32 *)a);
373 u64 b_offset = get_rel_offset(ri, *(const u32 *)b);
375 return a_offset < b_offset ? -1 : (a_offset > b_offset ? 1 : 0);
378 static int sort_rel(struct rel_info *ri)
380 size_t sz = sizeof(ri->sorted[0]);
383 ri->sorted = calloc(ri->nr_entries, sz);
386 for (i = 0; i < ri->nr_entries; i++)
388 qsort_r(ri->sorted, ri->nr_entries, sz, rel_cmp, ri);
393 * For x86_64, the GNU linker is putting IFUNC information in the relocation
396 static bool addend_may_be_ifunc(GElf_Ehdr *ehdr, struct rel_info *ri)
398 return ehdr->e_machine == EM_X86_64 && ri->is_rela &&
399 GELF_R_TYPE(ri->rela.r_info) == R_X86_64_IRELATIVE;
402 static bool get_ifunc_name(Elf *elf, struct dso *dso, GElf_Ehdr *ehdr,
403 struct rel_info *ri, char *buf, size_t buf_sz)
405 u64 addr = ri->rela.r_addend;
409 if (!addend_may_be_ifunc(ehdr, ri))
412 if (elf_read_program_header(elf, addr, &phdr))
415 addr -= phdr.p_vaddr - phdr.p_offset;
417 sym = dso__find_symbol_nocache(dso, addr);
419 /* Expecting the address to be an IFUNC or IFUNC alias */
420 if (!sym || sym->start != addr || (sym->type != STT_GNU_IFUNC && !sym->ifunc_alias))
423 snprintf(buf, buf_sz, "%s@plt", sym->name);
428 static void exit_rel(struct rel_info *ri)
433 static bool get_plt_sizes(struct dso *dso, GElf_Ehdr *ehdr, GElf_Shdr *shdr_plt,
434 u64 *plt_header_size, u64 *plt_entry_size)
436 switch (ehdr->e_machine) {
438 *plt_header_size = 20;
439 *plt_entry_size = 12;
442 *plt_header_size = 32;
443 *plt_entry_size = 16;
446 *plt_header_size = 32;
447 *plt_entry_size = 16;
450 *plt_header_size = 48;
451 *plt_entry_size = 12;
454 *plt_header_size = 128;
455 *plt_entry_size = 32;
459 *plt_entry_size = shdr_plt->sh_entsize;
460 /* Size is 8 or 16, if not, assume alignment indicates size */
461 if (*plt_entry_size != 8 && *plt_entry_size != 16)
462 *plt_entry_size = shdr_plt->sh_addralign == 8 ? 8 : 16;
463 *plt_header_size = *plt_entry_size;
465 default: /* FIXME: s390/alpha/mips/parisc/poperpc/sh/xtensa need to be checked */
466 *plt_header_size = shdr_plt->sh_entsize;
467 *plt_entry_size = shdr_plt->sh_entsize;
472 pr_debug("Missing PLT entry size for %s\n", dso->long_name);
476 static bool machine_is_x86(GElf_Half e_machine)
478 return e_machine == EM_386 || e_machine == EM_X86_64;
486 struct rela_dyn_info {
488 Elf_Data *plt_got_data;
490 struct rela_dyn *sorted;
491 Elf_Data *dynsym_data;
492 Elf_Data *dynstr_data;
493 Elf_Data *rela_dyn_data;
496 static void exit_rela_dyn(struct rela_dyn_info *di)
501 static int cmp_offset(const void *a, const void *b)
503 const struct rela_dyn *va = a;
504 const struct rela_dyn *vb = b;
506 return va->offset < vb->offset ? -1 : (va->offset > vb->offset ? 1 : 0);
509 static int sort_rela_dyn(struct rela_dyn_info *di)
513 di->sorted = calloc(di->nr_entries, sizeof(di->sorted[0]));
517 /* Get data for sorting: the offset and symbol index */
518 for (i = 0, n = 0; i < di->nr_entries; i++) {
522 gelf_getrela(di->rela_dyn_data, i, &rela);
523 sym_idx = GELF_R_SYM(rela.r_info);
525 di->sorted[n].sym_idx = sym_idx;
526 di->sorted[n].offset = rela.r_offset;
533 qsort(di->sorted, n, sizeof(di->sorted[0]), cmp_offset);
538 static void get_rela_dyn_info(Elf *elf, GElf_Ehdr *ehdr, struct rela_dyn_info *di, Elf_Scn *scn)
540 GElf_Shdr rela_dyn_shdr;
543 di->plt_got_data = elf_getdata(scn, NULL);
545 scn = elf_section_by_name(elf, ehdr, &rela_dyn_shdr, ".rela.dyn", NULL);
546 if (!scn || !rela_dyn_shdr.sh_link || !rela_dyn_shdr.sh_entsize)
549 di->nr_entries = rela_dyn_shdr.sh_size / rela_dyn_shdr.sh_entsize;
550 di->rela_dyn_data = elf_getdata(scn, NULL);
552 scn = elf_getscn(elf, rela_dyn_shdr.sh_link);
553 if (!scn || !gelf_getshdr(scn, &shdr) || !shdr.sh_link)
556 di->dynsym_data = elf_getdata(scn, NULL);
557 di->dynstr_data = elf_getdata(elf_getscn(elf, shdr.sh_link), NULL);
559 if (!di->plt_got_data || !di->dynstr_data || !di->dynsym_data || !di->rela_dyn_data)
562 /* Sort into offset order */
566 /* Get instruction displacement from a plt entry for x86_64 */
567 static u32 get_x86_64_plt_disp(const u8 *p)
569 u8 endbr64[] = {0xf3, 0x0f, 0x1e, 0xfa};
573 if (!memcmp(p, endbr64, sizeof(endbr64)))
574 n += sizeof(endbr64);
575 /* Skip bnd prefix */
578 /* jmp with 4-byte displacement */
579 if (p[n] == 0xff && p[n + 1] == 0x25) {
583 /* Also add offset from start of entry to end of instruction */
584 memcpy(&disp, p + n, sizeof(disp));
585 return n + 4 + le32toh(disp);
590 static bool get_plt_got_name(GElf_Shdr *shdr, size_t i,
591 struct rela_dyn_info *di,
592 char *buf, size_t buf_sz)
594 struct rela_dyn vi, *vr;
595 const char *sym_name;
604 disp = get_x86_64_plt_disp(di->plt_got_data->d_buf + i);
608 /* Compute target offset of the .plt.got entry */
609 vi.offset = shdr->sh_offset + di->plt_got_data->d_off + i + disp;
611 /* Find that offset in .rela.dyn (sorted by offset) */
612 vr = bsearch(&vi, di->sorted, di->nr_entries, sizeof(di->sorted[0]), cmp_offset);
616 /* Get the associated symbol */
617 gelf_getsym(di->dynsym_data, vr->sym_idx, &sym);
618 sym_name = elf_sym__name(&sym, di->dynstr_data);
619 demangled = demangle_sym(di->dso, 0, sym_name);
620 if (demangled != NULL)
621 sym_name = demangled;
623 snprintf(buf, buf_sz, "%s@plt", sym_name);
632 static int dso__synthesize_plt_got_symbols(struct dso *dso, Elf *elf,
634 char *buf, size_t buf_sz)
636 struct rela_dyn_info di = { .dso = dso };
643 scn = elf_section_by_name(elf, ehdr, &shdr, ".plt.got", NULL);
644 if (!scn || !shdr.sh_entsize)
647 if (ehdr->e_machine == EM_X86_64)
648 get_rela_dyn_info(elf, ehdr, &di, scn);
650 for (i = 0; i < shdr.sh_size; i += shdr.sh_entsize) {
651 if (!get_plt_got_name(&shdr, i, &di, buf, buf_sz))
652 snprintf(buf, buf_sz, "offset_%#" PRIx64 "@plt", (u64)shdr.sh_offset + i);
653 sym = symbol__new(shdr.sh_offset + i, shdr.sh_entsize, STB_GLOBAL, STT_FUNC, buf);
656 symbols__insert(&dso->symbols, sym);
665 * We need to check if we have a .dynsym, so that we can handle the
666 * .plt, synthesizing its symbols, that aren't on the symtabs (be it
667 * .dynsym or .symtab).
668 * And always look at the original dso, not at debuginfo packages, that
669 * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
671 int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss)
675 u64 plt_offset, plt_header_size, plt_entry_size;
676 GElf_Shdr shdr_plt, plt_sec_shdr;
677 struct symbol *f, *plt_sym;
678 GElf_Shdr shdr_rel_plt, shdr_dynsym;
679 Elf_Data *syms, *symstrs;
680 Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
682 char sympltname[1024];
684 int nr = 0, err = -1;
685 struct rel_info ri = { .is_rela = false };
691 if (!elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL))
695 * A symbol from a previous section (e.g. .init) can have been expanded
696 * by symbols__fixup_end() to overlap .plt. Truncate it before adding
697 * a symbol for .plt header.
699 f = dso__find_symbol_nocache(dso, shdr_plt.sh_offset);
700 if (f && f->start < shdr_plt.sh_offset && f->end > shdr_plt.sh_offset)
701 f->end = shdr_plt.sh_offset;
703 if (!get_plt_sizes(dso, &ehdr, &shdr_plt, &plt_header_size, &plt_entry_size))
706 /* Add a symbol for .plt header */
707 plt_sym = symbol__new(shdr_plt.sh_offset, plt_header_size, STB_GLOBAL, STT_FUNC, ".plt");
710 symbols__insert(&dso->symbols, plt_sym);
712 /* Only x86 has .plt.got */
713 if (machine_is_x86(ehdr.e_machine) &&
714 dso__synthesize_plt_got_symbols(dso, elf, &ehdr, sympltname, sizeof(sympltname)))
717 /* Only x86 has .plt.sec */
718 if (machine_is_x86(ehdr.e_machine) &&
719 elf_section_by_name(elf, &ehdr, &plt_sec_shdr, ".plt.sec", NULL)) {
720 if (!get_plt_sizes(dso, &ehdr, &plt_sec_shdr, &plt_header_size, &plt_entry_size))
722 /* Extend .plt symbol to entire .plt */
723 plt_sym->end = plt_sym->start + shdr_plt.sh_size;
724 /* Use .plt.sec offset */
725 plt_offset = plt_sec_shdr.sh_offset;
728 plt_offset = shdr_plt.sh_offset;
732 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
734 if (scn_plt_rel == NULL) {
735 scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
737 if (scn_plt_rel == NULL)
741 if (shdr_rel_plt.sh_type != SHT_RELA &&
742 shdr_rel_plt.sh_type != SHT_REL)
745 if (!shdr_rel_plt.sh_link)
748 if (shdr_rel_plt.sh_link == ss->dynsym_idx) {
749 scn_dynsym = ss->dynsym;
750 shdr_dynsym = ss->dynshdr;
751 } else if (shdr_rel_plt.sh_link == ss->symtab_idx) {
753 * A static executable can have a .plt due to IFUNCs, in which
754 * case .symtab is used not .dynsym.
756 scn_dynsym = ss->symtab;
757 shdr_dynsym = ss->symshdr;
766 * Fetch the relocation section to find the idxes to the GOT
767 * and the symbols in the .dynsym they refer to.
769 ri.reldata = elf_getdata(scn_plt_rel, NULL);
773 syms = elf_getdata(scn_dynsym, NULL);
777 scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
778 if (scn_symstrs == NULL)
781 symstrs = elf_getdata(scn_symstrs, NULL);
785 if (symstrs->d_size == 0)
788 ri.nr_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
790 ri.is_rela = shdr_rel_plt.sh_type == SHT_RELA;
794 * Assume a .plt with the same number of entries as the number
795 * of relocation entries is not lazy and does not have a header.
797 if (ri.nr_entries * plt_entry_size == shdr_plt.sh_size)
798 dso__delete_symbol(dso, plt_sym);
800 plt_offset += plt_header_size;
804 * x86 doesn't insert IFUNC relocations in .plt order, so sort to get
807 if (machine_is_x86(ehdr.e_machine) && sort_rel(&ri))
810 for (idx = 0; idx < ri.nr_entries; idx++) {
811 const char *elf_name = NULL;
812 char *demangled = NULL;
814 gelf_getsym(syms, get_rel_symidx(&ri, idx), &sym);
816 elf_name = elf_sym__name(&sym, symstrs);
817 demangled = demangle_sym(dso, 0, elf_name);
819 elf_name = demangled;
821 snprintf(sympltname, sizeof(sympltname), "%s@plt", elf_name);
822 else if (!get_ifunc_name(elf, dso, &ehdr, &ri, sympltname, sizeof(sympltname)))
823 snprintf(sympltname, sizeof(sympltname),
824 "offset_%#" PRIx64 "@plt", plt_offset);
827 f = symbol__new(plt_offset, plt_entry_size, STB_GLOBAL, STT_FUNC, sympltname);
831 plt_offset += plt_entry_size;
832 symbols__insert(&dso->symbols, f);
841 pr_debug("%s: problems reading %s PLT info.\n",
842 __func__, dso->long_name);
846 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name)
848 return demangle_sym(dso, kmodule, elf_name);
852 * Align offset to 4 bytes as needed for note name and descriptor data.
854 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
856 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
866 if (size < BUILD_ID_SIZE)
873 if (gelf_getehdr(elf, &ehdr) == NULL) {
874 pr_err("%s: cannot get elf header.\n", __func__);
879 * Check following sections for notes:
880 * '.note.gnu.build-id'
882 * '.note' (VDSO specific)
885 sec = elf_section_by_name(elf, &ehdr, &shdr,
886 ".note.gnu.build-id", NULL);
890 sec = elf_section_by_name(elf, &ehdr, &shdr,
895 sec = elf_section_by_name(elf, &ehdr, &shdr,
904 data = elf_getdata(sec, NULL);
909 while (ptr < (data->d_buf + data->d_size)) {
910 GElf_Nhdr *nhdr = ptr;
911 size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
912 descsz = NOTE_ALIGN(nhdr->n_descsz);
915 ptr += sizeof(*nhdr);
918 if (nhdr->n_type == NT_GNU_BUILD_ID &&
919 nhdr->n_namesz == sizeof("GNU")) {
920 if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
921 size_t sz = min(size, descsz);
923 memset(bf + sz, 0, size - sz);
935 #ifdef HAVE_LIBBFD_BUILDID_SUPPORT
937 static int read_build_id(const char *filename, struct build_id *bid)
939 size_t size = sizeof(bid->data);
943 abfd = bfd_openr(filename, NULL);
947 if (!bfd_check_format(abfd, bfd_object)) {
948 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
952 if (!abfd->build_id || abfd->build_id->size > size)
955 memcpy(bid->data, abfd->build_id->data, abfd->build_id->size);
956 memset(bid->data + abfd->build_id->size, 0, size - abfd->build_id->size);
957 err = bid->size = abfd->build_id->size;
964 #else // HAVE_LIBBFD_BUILDID_SUPPORT
966 static int read_build_id(const char *filename, struct build_id *bid)
968 size_t size = sizeof(bid->data);
972 if (size < BUILD_ID_SIZE)
975 fd = open(filename, O_RDONLY);
979 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
981 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
985 err = elf_read_build_id(elf, bid->data, size);
996 #endif // HAVE_LIBBFD_BUILDID_SUPPORT
998 int filename__read_build_id(const char *filename, struct build_id *bid)
1000 struct kmod_path m = { .name = NULL, };
1001 char path[PATH_MAX];
1007 err = kmod_path__parse(&m, filename);
1014 fd = filename__decompress(filename, path, sizeof(path), m.comp, &error);
1016 pr_debug("Failed to decompress (error %d) %s\n",
1024 err = read_build_id(filename, bid);
1031 int sysfs__read_build_id(const char *filename, struct build_id *bid)
1033 size_t size = sizeof(bid->data);
1036 fd = open(filename, O_RDONLY);
1043 size_t namesz, descsz;
1045 if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
1048 namesz = NOTE_ALIGN(nhdr.n_namesz);
1049 descsz = NOTE_ALIGN(nhdr.n_descsz);
1050 if (nhdr.n_type == NT_GNU_BUILD_ID &&
1051 nhdr.n_namesz == sizeof("GNU")) {
1052 if (read(fd, bf, namesz) != (ssize_t)namesz)
1054 if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
1055 size_t sz = min(descsz, size);
1056 if (read(fd, bid->data, sz) == (ssize_t)sz) {
1057 memset(bid->data + sz, 0, size - sz);
1062 } else if (read(fd, bf, descsz) != (ssize_t)descsz)
1065 int n = namesz + descsz;
1067 if (n > (int)sizeof(bf)) {
1069 pr_debug("%s: truncating reading of build id in sysfs file %s: n_namesz=%u, n_descsz=%u.\n",
1070 __func__, filename, nhdr.n_namesz, nhdr.n_descsz);
1072 if (read(fd, bf, n) != n)
1081 #ifdef HAVE_LIBBFD_SUPPORT
1083 int filename__read_debuglink(const char *filename, char *debuglink,
1090 abfd = bfd_openr(filename, NULL);
1094 if (!bfd_check_format(abfd, bfd_object)) {
1095 pr_debug2("%s: cannot read %s bfd file.\n", __func__, filename);
1099 section = bfd_get_section_by_name(abfd, ".gnu_debuglink");
1103 if (section->size > size)
1106 if (!bfd_get_section_contents(abfd, section, debuglink, 0,
1119 int filename__read_debuglink(const char *filename, char *debuglink,
1130 fd = open(filename, O_RDONLY);
1134 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1136 pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1141 if (ek != ELF_K_ELF)
1144 if (gelf_getehdr(elf, &ehdr) == NULL) {
1145 pr_err("%s: cannot get elf header.\n", __func__);
1149 sec = elf_section_by_name(elf, &ehdr, &shdr,
1150 ".gnu_debuglink", NULL);
1154 data = elf_getdata(sec, NULL);
1158 /* the start of this section is a zero-terminated string */
1159 strncpy(debuglink, data->d_buf, size);
1173 static int dso__swap_init(struct dso *dso, unsigned char eidata)
1175 static unsigned int const endian = 1;
1177 dso->needs_swap = DSO_SWAP__NO;
1181 /* We are big endian, DSO is little endian. */
1182 if (*(unsigned char const *)&endian != 1)
1183 dso->needs_swap = DSO_SWAP__YES;
1187 /* We are little endian, DSO is big endian. */
1188 if (*(unsigned char const *)&endian != 0)
1189 dso->needs_swap = DSO_SWAP__YES;
1193 pr_err("unrecognized DSO data encoding %d\n", eidata);
1200 bool symsrc__possibly_runtime(struct symsrc *ss)
1202 return ss->dynsym || ss->opdsec;
1205 bool symsrc__has_symtab(struct symsrc *ss)
1207 return ss->symtab != NULL;
1210 void symsrc__destroy(struct symsrc *ss)
1217 bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
1220 * Usually vmlinux is an ELF file with type ET_EXEC for most
1221 * architectures; except Arm64 kernel is linked with option
1222 * '-share', so need to check type ET_DYN.
1224 return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
1225 ehdr.e_type == ET_DYN;
1228 int symsrc__init(struct symsrc *ss, struct dso *dso, const char *name,
1229 enum dso_binary_type type)
1235 if (dso__needs_decompress(dso)) {
1236 fd = dso__decompress_kmodule_fd(dso, name);
1240 type = dso->symtab_type;
1242 fd = open(name, O_RDONLY);
1244 dso->load_errno = errno;
1249 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1251 pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1252 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
1256 if (gelf_getehdr(elf, &ehdr) == NULL) {
1257 dso->load_errno = DSO_LOAD_ERRNO__INVALID_ELF;
1258 pr_debug("%s: cannot get elf header.\n", __func__);
1262 if (dso__swap_init(dso, ehdr.e_ident[EI_DATA])) {
1263 dso->load_errno = DSO_LOAD_ERRNO__INTERNAL_ERROR;
1267 /* Always reject images with a mismatched build-id: */
1268 if (dso->has_build_id && !symbol_conf.ignore_vmlinux_buildid) {
1269 u8 build_id[BUILD_ID_SIZE];
1270 struct build_id bid;
1273 size = elf_read_build_id(elf, build_id, BUILD_ID_SIZE);
1275 dso->load_errno = DSO_LOAD_ERRNO__CANNOT_READ_BUILDID;
1279 build_id__init(&bid, build_id, size);
1280 if (!dso__build_id_equal(dso, &bid)) {
1281 pr_debug("%s: build id mismatch for %s.\n", __func__, name);
1282 dso->load_errno = DSO_LOAD_ERRNO__MISMATCHING_BUILDID;
1287 ss->is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1290 ss->symtab = elf_section_by_name(elf, &ehdr, &ss->symshdr, ".symtab",
1292 if (ss->symshdr.sh_type != SHT_SYMTAB)
1296 ss->dynsym = elf_section_by_name(elf, &ehdr, &ss->dynshdr, ".dynsym",
1298 if (ss->dynshdr.sh_type != SHT_DYNSYM)
1302 ss->opdsec = elf_section_by_name(elf, &ehdr, &ss->opdshdr, ".opd",
1304 if (ss->opdshdr.sh_type != SHT_PROGBITS)
1307 if (dso->kernel == DSO_SPACE__USER)
1308 ss->adjust_symbols = true;
1310 ss->adjust_symbols = elf__needs_adjust_symbols(ehdr);
1312 ss->name = strdup(name);
1314 dso->load_errno = errno;
1333 * ref_reloc_sym_not_found - has kernel relocation symbol been found.
1334 * @kmap: kernel maps and relocation reference symbol
1336 * This function returns %true if we are dealing with the kernel maps and the
1337 * relocation reference symbol has not yet been found. Otherwise %false is
1340 static bool ref_reloc_sym_not_found(struct kmap *kmap)
1342 return kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
1343 !kmap->ref_reloc_sym->unrelocated_addr;
1347 * ref_reloc - kernel relocation offset.
1348 * @kmap: kernel maps and relocation reference symbol
1350 * This function returns the offset of kernel addresses as determined by using
1351 * the relocation reference symbol i.e. if the kernel has not been relocated
1352 * then the return value is zero.
1354 static u64 ref_reloc(struct kmap *kmap)
1356 if (kmap && kmap->ref_reloc_sym &&
1357 kmap->ref_reloc_sym->unrelocated_addr)
1358 return kmap->ref_reloc_sym->addr -
1359 kmap->ref_reloc_sym->unrelocated_addr;
1363 void __weak arch__sym_update(struct symbol *s __maybe_unused,
1364 GElf_Sym *sym __maybe_unused) { }
1366 static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
1367 GElf_Sym *sym, GElf_Shdr *shdr,
1368 struct maps *kmaps, struct kmap *kmap,
1369 struct dso **curr_dsop, struct map **curr_mapp,
1370 const char *section_name,
1371 bool adjust_kernel_syms, bool kmodule, bool *remap_kernel)
1373 struct dso *curr_dso = *curr_dsop;
1374 struct map *curr_map;
1375 char dso_name[PATH_MAX];
1377 /* Adjust symbol to map to file offset */
1378 if (adjust_kernel_syms)
1379 sym->st_value -= shdr->sh_addr - shdr->sh_offset;
1381 if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0)
1384 if (strcmp(section_name, ".text") == 0) {
1386 * The initial kernel mapping is based on
1387 * kallsyms and identity maps. Overwrite it to
1388 * map to the kernel dso.
1390 if (*remap_kernel && dso->kernel && !kmodule) {
1391 *remap_kernel = false;
1392 map__set_start(map, shdr->sh_addr + ref_reloc(kmap));
1393 map__set_end(map, map__start(map) + shdr->sh_size);
1394 map__set_pgoff(map, shdr->sh_offset);
1395 map__set_map_ip(map, map__dso_map_ip);
1396 map__set_unmap_ip(map, map__dso_unmap_ip);
1397 /* Ensure maps are correctly ordered */
1400 struct map *tmp = map__get(map);
1402 maps__remove(kmaps, map);
1403 err = maps__insert(kmaps, map);
1411 * The initial module mapping is based on
1412 * /proc/modules mapped to offset zero.
1413 * Overwrite it to map to the module dso.
1415 if (*remap_kernel && kmodule) {
1416 *remap_kernel = false;
1417 map__set_pgoff(map, shdr->sh_offset);
1428 snprintf(dso_name, sizeof(dso_name), "%s%s", dso->short_name, section_name);
1430 curr_map = maps__find_by_name(kmaps, dso_name);
1431 if (curr_map == NULL) {
1432 u64 start = sym->st_value;
1435 start += map__start(map) + shdr->sh_offset;
1437 curr_dso = dso__new(dso_name);
1438 if (curr_dso == NULL)
1440 curr_dso->kernel = dso->kernel;
1441 curr_dso->long_name = dso->long_name;
1442 curr_dso->long_name_len = dso->long_name_len;
1443 curr_map = map__new2(start, curr_dso);
1445 if (curr_map == NULL)
1448 if (curr_dso->kernel)
1449 map__kmap(curr_map)->kmaps = kmaps;
1451 if (adjust_kernel_syms) {
1452 map__set_start(curr_map, shdr->sh_addr + ref_reloc(kmap));
1453 map__set_end(curr_map, map__start(curr_map) + shdr->sh_size);
1454 map__set_pgoff(curr_map, shdr->sh_offset);
1456 map__set_map_ip(curr_map, identity__map_ip);
1457 map__set_unmap_ip(curr_map, identity__map_ip);
1459 curr_dso->symtab_type = dso->symtab_type;
1460 if (maps__insert(kmaps, curr_map))
1463 * Add it before we drop the reference to curr_map, i.e. while
1464 * we still are sure to have a reference to this DSO via
1467 dsos__add(&maps__machine(kmaps)->dsos, curr_dso);
1468 /* kmaps already got it */
1470 dso__set_loaded(curr_dso);
1471 *curr_mapp = curr_map;
1472 *curr_dsop = curr_dso;
1474 *curr_dsop = map__dso(curr_map);
1480 dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1481 struct symsrc *runtime_ss, int kmodule, int dynsym)
1483 struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
1484 struct maps *kmaps = kmap ? map__kmaps(map) : NULL;
1485 struct map *curr_map = map;
1486 struct dso *curr_dso = dso;
1487 Elf_Data *symstrs, *secstrs, *secstrs_run, *secstrs_sym;
1494 Elf_Data *syms, *opddata = NULL;
1496 Elf_Scn *sec, *sec_strndx;
1499 bool remap_kernel = false, adjust_kernel_syms = false;
1505 ehdr = syms_ss->ehdr;
1507 sec = syms_ss->dynsym;
1508 shdr = syms_ss->dynshdr;
1510 sec = syms_ss->symtab;
1511 shdr = syms_ss->symshdr;
1514 if (elf_section_by_name(runtime_ss->elf, &runtime_ss->ehdr, &tshdr,
1516 dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;
1518 if (runtime_ss->opdsec)
1519 opddata = elf_rawdata(runtime_ss->opdsec, NULL);
1521 syms = elf_getdata(sec, NULL);
1525 sec = elf_getscn(elf, shdr.sh_link);
1529 symstrs = elf_getdata(sec, NULL);
1530 if (symstrs == NULL)
1533 sec_strndx = elf_getscn(runtime_ss->elf, runtime_ss->ehdr.e_shstrndx);
1534 if (sec_strndx == NULL)
1537 secstrs_run = elf_getdata(sec_strndx, NULL);
1538 if (secstrs_run == NULL)
1541 sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1542 if (sec_strndx == NULL)
1545 secstrs_sym = elf_getdata(sec_strndx, NULL);
1546 if (secstrs_sym == NULL)
1549 nr_syms = shdr.sh_size / shdr.sh_entsize;
1551 memset(&sym, 0, sizeof(sym));
1554 * The kernel relocation symbol is needed in advance in order to adjust
1555 * kernel maps correctly.
1557 if (ref_reloc_sym_not_found(kmap)) {
1558 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1559 const char *elf_name = elf_sym__name(&sym, symstrs);
1561 if (strcmp(elf_name, kmap->ref_reloc_sym->name))
1563 kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1564 map__set_reloc(map, kmap->ref_reloc_sym->addr - kmap->ref_reloc_sym->unrelocated_addr);
1570 * Handle any relocation of vdso necessary because older kernels
1571 * attempted to prelink vdso to its virtual address.
1573 if (dso__is_vdso(dso))
1574 map__set_reloc(map, map__start(map) - dso->text_offset);
1576 dso->adjust_symbols = runtime_ss->adjust_symbols || ref_reloc(kmap);
1578 * Initial kernel and module mappings do not map to the dso.
1582 remap_kernel = true;
1583 adjust_kernel_syms = dso->adjust_symbols;
1585 elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1587 const char *elf_name = elf_sym__name(&sym, symstrs);
1588 char *demangled = NULL;
1589 int is_label = elf_sym__is_label(&sym);
1590 const char *section_name;
1591 bool used_opd = false;
1593 if (!is_label && !elf_sym__filter(&sym))
1596 /* Reject ARM ELF "mapping symbols": these aren't unique and
1597 * don't identify functions, so will confuse the profile
1599 if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
1600 if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
1601 && (elf_name[2] == '\0' || elf_name[2] == '.'))
1605 if (runtime_ss->opdsec && sym.st_shndx == runtime_ss->opdidx) {
1606 u32 offset = sym.st_value - syms_ss->opdshdr.sh_addr;
1607 u64 *opd = opddata->d_buf + offset;
1608 sym.st_value = DSO__SWAP(dso, u64, *opd);
1609 sym.st_shndx = elf_addr_to_index(runtime_ss->elf,
1615 * When loading symbols in a data mapping, ABS symbols (which
1616 * has a value of SHN_ABS in its st_shndx) failed at
1617 * elf_getscn(). And it marks the loading as a failure so
1618 * already loaded symbols cannot be fixed up.
1620 * I'm not sure what should be done. Just ignore them for now.
1623 if (sym.st_shndx == SHN_ABS)
1626 sec = elf_getscn(syms_ss->elf, sym.st_shndx);
1630 gelf_getshdr(sec, &shdr);
1633 * If the attribute bit SHF_ALLOC is not set, the section
1634 * doesn't occupy memory during process execution.
1635 * E.g. ".gnu.warning.*" section is used by linker to generate
1636 * warnings when calling deprecated functions, the symbols in
1637 * the section aren't loaded to memory during process execution,
1640 if (!(shdr.sh_flags & SHF_ALLOC))
1643 secstrs = secstrs_sym;
1646 * We have to fallback to runtime when syms' section header has
1647 * NOBITS set. NOBITS results in file offset (sh_offset) not
1648 * being incremented. So sh_offset used below has different
1649 * values for syms (invalid) and runtime (valid).
1651 if (shdr.sh_type == SHT_NOBITS) {
1652 sec = elf_getscn(runtime_ss->elf, sym.st_shndx);
1656 gelf_getshdr(sec, &shdr);
1657 secstrs = secstrs_run;
1660 if (is_label && !elf_sec__filter(&shdr, secstrs))
1663 section_name = elf_sec__name(&shdr, secstrs);
1665 /* On ARM, symbols for thumb functions have 1 added to
1666 * the symbol address as a flag - remove it */
1667 if ((ehdr.e_machine == EM_ARM) &&
1668 (GELF_ST_TYPE(sym.st_info) == STT_FUNC) &&
1673 if (dso__process_kernel_symbol(dso, map, &sym, &shdr, kmaps, kmap, &curr_dso, &curr_map,
1674 section_name, adjust_kernel_syms, kmodule, &remap_kernel))
1676 } else if ((used_opd && runtime_ss->adjust_symbols) ||
1677 (!used_opd && syms_ss->adjust_symbols)) {
1680 if (elf_read_program_header(runtime_ss->elf,
1681 (u64)sym.st_value, &phdr)) {
1682 pr_debug4("%s: failed to find program header for "
1683 "symbol: %s st_value: %#" PRIx64 "\n",
1684 __func__, elf_name, (u64)sym.st_value);
1685 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1686 "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n",
1687 __func__, (u64)sym.st_value, (u64)shdr.sh_addr,
1688 (u64)shdr.sh_offset);
1690 * Fail to find program header, let's rollback
1691 * to use shdr.sh_addr and shdr.sh_offset to
1692 * calibrate symbol's file address, though this
1693 * is not necessary for normal C ELF file, we
1694 * still need to handle java JIT symbols in this
1697 sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1699 pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1700 "p_vaddr: %#" PRIx64 " p_offset: %#" PRIx64 "\n",
1701 __func__, (u64)sym.st_value, (u64)phdr.p_vaddr,
1702 (u64)phdr.p_offset);
1703 sym.st_value -= phdr.p_vaddr - phdr.p_offset;
1707 demangled = demangle_sym(dso, kmodule, elf_name);
1708 if (demangled != NULL)
1709 elf_name = demangled;
1711 f = symbol__new(sym.st_value, sym.st_size,
1712 GELF_ST_BIND(sym.st_info),
1713 GELF_ST_TYPE(sym.st_info), elf_name);
1718 arch__sym_update(f, &sym);
1720 __symbols__insert(&curr_dso->symbols, f, dso->kernel);
1725 * For misannotated, zeroed, ASM function sizes.
1728 symbols__fixup_end(&dso->symbols, false);
1729 symbols__fixup_duplicate(&dso->symbols);
1732 * We need to fixup this here too because we create new
1733 * maps here, for things like vsyscall sections.
1735 maps__fixup_end(kmaps);
1743 int dso__load_sym(struct dso *dso, struct map *map, struct symsrc *syms_ss,
1744 struct symsrc *runtime_ss, int kmodule)
1749 dso->symtab_type = syms_ss->type;
1750 dso->is_64_bit = syms_ss->is_64_bit;
1751 dso->rel = syms_ss->ehdr.e_type == ET_REL;
1754 * Modules may already have symbols from kallsyms, but those symbols
1755 * have the wrong values for the dso maps, so remove them.
1757 if (kmodule && syms_ss->symtab)
1758 symbols__delete(&dso->symbols);
1760 if (!syms_ss->symtab) {
1762 * If the vmlinux is stripped, fail so we will fall back
1763 * to using kallsyms. The vmlinux runtime symbols aren't
1769 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1776 if (syms_ss->dynsym) {
1777 err = dso__load_sym_internal(dso, map, syms_ss, runtime_ss,
1787 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
1794 if (elf_getphdrnum(elf, &phdrnum))
1797 for (i = 0; i < phdrnum; i++) {
1798 if (gelf_getphdr(elf, i, &phdr) == NULL)
1800 if (phdr.p_type != PT_LOAD)
1803 if (!(phdr.p_flags & PF_X))
1806 if (!(phdr.p_flags & PF_R))
1809 sz = min(phdr.p_memsz, phdr.p_filesz);
1812 err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
1819 int file__read_maps(int fd, bool exe, mapfn_t mapfn, void *data,
1825 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1830 *is_64_bit = (gelf_getclass(elf) == ELFCLASS64);
1832 err = elf_read_maps(elf, exe, mapfn, data);
1838 enum dso_type dso__type_fd(int fd)
1840 enum dso_type dso_type = DSO__TYPE_UNKNOWN;
1845 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1850 if (ek != ELF_K_ELF)
1853 if (gelf_getclass(elf) == ELFCLASS64) {
1854 dso_type = DSO__TYPE_64BIT;
1858 if (gelf_getehdr(elf, &ehdr) == NULL)
1861 if (ehdr.e_machine == EM_X86_64)
1862 dso_type = DSO__TYPE_X32BIT;
1864 dso_type = DSO__TYPE_32BIT;
1871 static int copy_bytes(int from, off_t from_offs, int to, off_t to_offs, u64 len)
1876 char *buf = malloc(page_size);
1881 if (lseek(to, to_offs, SEEK_SET) != to_offs)
1884 if (lseek(from, from_offs, SEEK_SET) != from_offs)
1891 /* Use read because mmap won't work on proc files */
1892 r = read(from, buf, n);
1898 r = write(to, buf, n);
1919 static int kcore__open(struct kcore *kcore, const char *filename)
1923 kcore->fd = open(filename, O_RDONLY);
1924 if (kcore->fd == -1)
1927 kcore->elf = elf_begin(kcore->fd, ELF_C_READ, NULL);
1931 kcore->elfclass = gelf_getclass(kcore->elf);
1932 if (kcore->elfclass == ELFCLASSNONE)
1935 ehdr = gelf_getehdr(kcore->elf, &kcore->ehdr);
1942 elf_end(kcore->elf);
1948 static int kcore__init(struct kcore *kcore, char *filename, int elfclass,
1951 kcore->elfclass = elfclass;
1954 kcore->fd = mkstemp(filename);
1956 kcore->fd = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0400);
1957 if (kcore->fd == -1)
1960 kcore->elf = elf_begin(kcore->fd, ELF_C_WRITE, NULL);
1964 if (!gelf_newehdr(kcore->elf, elfclass))
1967 memset(&kcore->ehdr, 0, sizeof(GElf_Ehdr));
1972 elf_end(kcore->elf);
1979 static void kcore__close(struct kcore *kcore)
1981 elf_end(kcore->elf);
1985 static int kcore__copy_hdr(struct kcore *from, struct kcore *to, size_t count)
1987 GElf_Ehdr *ehdr = &to->ehdr;
1988 GElf_Ehdr *kehdr = &from->ehdr;
1990 memcpy(ehdr->e_ident, kehdr->e_ident, EI_NIDENT);
1991 ehdr->e_type = kehdr->e_type;
1992 ehdr->e_machine = kehdr->e_machine;
1993 ehdr->e_version = kehdr->e_version;
1996 ehdr->e_flags = kehdr->e_flags;
1997 ehdr->e_phnum = count;
1998 ehdr->e_shentsize = 0;
2000 ehdr->e_shstrndx = 0;
2002 if (from->elfclass == ELFCLASS32) {
2003 ehdr->e_phoff = sizeof(Elf32_Ehdr);
2004 ehdr->e_ehsize = sizeof(Elf32_Ehdr);
2005 ehdr->e_phentsize = sizeof(Elf32_Phdr);
2007 ehdr->e_phoff = sizeof(Elf64_Ehdr);
2008 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
2009 ehdr->e_phentsize = sizeof(Elf64_Phdr);
2012 if (!gelf_update_ehdr(to->elf, ehdr))
2015 if (!gelf_newphdr(to->elf, count))
2021 static int kcore__add_phdr(struct kcore *kcore, int idx, off_t offset,
2026 .p_flags = PF_R | PF_W | PF_X,
2032 .p_align = page_size,
2035 if (!gelf_update_phdr(kcore->elf, idx, &phdr))
2041 static off_t kcore__write(struct kcore *kcore)
2043 return elf_update(kcore->elf, ELF_C_WRITE);
2051 struct list_head node;
2052 struct phdr_data *remaps;
2057 struct list_head node;
2060 struct kcore_copy_info {
2066 u64 first_module_symbol;
2067 u64 last_module_symbol;
2069 struct list_head phdrs;
2070 struct list_head syms;
2073 #define kcore_copy__for_each_phdr(k, p) \
2074 list_for_each_entry((p), &(k)->phdrs, node)
2076 static struct phdr_data *phdr_data__new(u64 addr, u64 len, off_t offset)
2078 struct phdr_data *p = zalloc(sizeof(*p));
2089 static struct phdr_data *kcore_copy_info__addnew(struct kcore_copy_info *kci,
2093 struct phdr_data *p = phdr_data__new(addr, len, offset);
2096 list_add_tail(&p->node, &kci->phdrs);
2101 static void kcore_copy__free_phdrs(struct kcore_copy_info *kci)
2103 struct phdr_data *p, *tmp;
2105 list_for_each_entry_safe(p, tmp, &kci->phdrs, node) {
2106 list_del_init(&p->node);
2111 static struct sym_data *kcore_copy__new_sym(struct kcore_copy_info *kci,
2114 struct sym_data *s = zalloc(sizeof(*s));
2118 list_add_tail(&s->node, &kci->syms);
2124 static void kcore_copy__free_syms(struct kcore_copy_info *kci)
2126 struct sym_data *s, *tmp;
2128 list_for_each_entry_safe(s, tmp, &kci->syms, node) {
2129 list_del_init(&s->node);
2134 static int kcore_copy__process_kallsyms(void *arg, const char *name, char type,
2137 struct kcore_copy_info *kci = arg;
2139 if (!kallsyms__is_function(type))
2142 if (strchr(name, '[')) {
2143 if (!kci->first_module_symbol || start < kci->first_module_symbol)
2144 kci->first_module_symbol = start;
2145 if (start > kci->last_module_symbol)
2146 kci->last_module_symbol = start;
2150 if (!kci->first_symbol || start < kci->first_symbol)
2151 kci->first_symbol = start;
2153 if (!kci->last_symbol || start > kci->last_symbol)
2154 kci->last_symbol = start;
2156 if (!strcmp(name, "_stext")) {
2161 if (!strcmp(name, "_etext")) {
2166 if (is_entry_trampoline(name) && !kcore_copy__new_sym(kci, start))
2172 static int kcore_copy__parse_kallsyms(struct kcore_copy_info *kci,
2175 char kallsyms_filename[PATH_MAX];
2177 scnprintf(kallsyms_filename, PATH_MAX, "%s/kallsyms", dir);
2179 if (symbol__restricted_filename(kallsyms_filename, "/proc/kallsyms"))
2182 if (kallsyms__parse(kallsyms_filename, kci,
2183 kcore_copy__process_kallsyms) < 0)
2189 static int kcore_copy__process_modules(void *arg,
2190 const char *name __maybe_unused,
2191 u64 start, u64 size __maybe_unused)
2193 struct kcore_copy_info *kci = arg;
2195 if (!kci->first_module || start < kci->first_module)
2196 kci->first_module = start;
2201 static int kcore_copy__parse_modules(struct kcore_copy_info *kci,
2204 char modules_filename[PATH_MAX];
2206 scnprintf(modules_filename, PATH_MAX, "%s/modules", dir);
2208 if (symbol__restricted_filename(modules_filename, "/proc/modules"))
2211 if (modules__parse(modules_filename, kci,
2212 kcore_copy__process_modules) < 0)
2218 static int kcore_copy__map(struct kcore_copy_info *kci, u64 start, u64 end,
2219 u64 pgoff, u64 s, u64 e)
2223 if (s < start || s >= end)
2226 offset = (s - start) + pgoff;
2227 len = e < end ? e - s : end - s;
2229 return kcore_copy_info__addnew(kci, s, len, offset) ? 0 : -1;
2232 static int kcore_copy__read_map(u64 start, u64 len, u64 pgoff, void *data)
2234 struct kcore_copy_info *kci = data;
2235 u64 end = start + len;
2236 struct sym_data *sdat;
2238 if (kcore_copy__map(kci, start, end, pgoff, kci->stext, kci->etext))
2241 if (kcore_copy__map(kci, start, end, pgoff, kci->first_module,
2242 kci->last_module_symbol))
2245 list_for_each_entry(sdat, &kci->syms, node) {
2246 u64 s = round_down(sdat->addr, page_size);
2248 if (kcore_copy__map(kci, start, end, pgoff, s, s + len))
2255 static int kcore_copy__read_maps(struct kcore_copy_info *kci, Elf *elf)
2257 if (elf_read_maps(elf, true, kcore_copy__read_map, kci) < 0)
2263 static void kcore_copy__find_remaps(struct kcore_copy_info *kci)
2265 struct phdr_data *p, *k = NULL;
2271 /* Find phdr that corresponds to the kernel map (contains stext) */
2272 kcore_copy__for_each_phdr(kci, p) {
2273 u64 pend = p->addr + p->len - 1;
2275 if (p->addr <= kci->stext && pend >= kci->stext) {
2284 kend = k->offset + k->len;
2286 /* Find phdrs that remap the kernel */
2287 kcore_copy__for_each_phdr(kci, p) {
2288 u64 pend = p->offset + p->len;
2293 if (p->offset >= k->offset && pend <= kend)
2298 static void kcore_copy__layout(struct kcore_copy_info *kci)
2300 struct phdr_data *p;
2303 kcore_copy__find_remaps(kci);
2305 kcore_copy__for_each_phdr(kci, p) {
2313 kcore_copy__for_each_phdr(kci, p) {
2314 struct phdr_data *k = p->remaps;
2317 p->rel = p->offset - k->offset + k->rel;
2321 static int kcore_copy__calc_maps(struct kcore_copy_info *kci, const char *dir,
2324 if (kcore_copy__parse_kallsyms(kci, dir))
2327 if (kcore_copy__parse_modules(kci, dir))
2331 kci->stext = round_down(kci->stext, page_size);
2333 kci->stext = round_down(kci->first_symbol, page_size);
2336 kci->etext = round_up(kci->etext, page_size);
2337 } else if (kci->last_symbol) {
2338 kci->etext = round_up(kci->last_symbol, page_size);
2339 kci->etext += page_size;
2342 if (kci->first_module_symbol &&
2343 (!kci->first_module || kci->first_module_symbol < kci->first_module))
2344 kci->first_module = kci->first_module_symbol;
2346 kci->first_module = round_down(kci->first_module, page_size);
2348 if (kci->last_module_symbol) {
2349 kci->last_module_symbol = round_up(kci->last_module_symbol,
2351 kci->last_module_symbol += page_size;
2354 if (!kci->stext || !kci->etext)
2357 if (kci->first_module && !kci->last_module_symbol)
2360 if (kcore_copy__read_maps(kci, elf))
2363 kcore_copy__layout(kci);
2368 static int kcore_copy__copy_file(const char *from_dir, const char *to_dir,
2371 char from_filename[PATH_MAX];
2372 char to_filename[PATH_MAX];
2374 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2375 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2377 return copyfile_mode(from_filename, to_filename, 0400);
2380 static int kcore_copy__unlink(const char *dir, const char *name)
2382 char filename[PATH_MAX];
2384 scnprintf(filename, PATH_MAX, "%s/%s", dir, name);
2386 return unlink(filename);
2389 static int kcore_copy__compare_fds(int from, int to)
2397 buf_from = malloc(page_size);
2398 buf_to = malloc(page_size);
2399 if (!buf_from || !buf_to)
2403 /* Use read because mmap won't work on proc files */
2404 ret = read(from, buf_from, page_size);
2413 if (readn(to, buf_to, len) != (int)len)
2416 if (memcmp(buf_from, buf_to, len))
2427 static int kcore_copy__compare_files(const char *from_filename,
2428 const char *to_filename)
2430 int from, to, err = -1;
2432 from = open(from_filename, O_RDONLY);
2436 to = open(to_filename, O_RDONLY);
2438 goto out_close_from;
2440 err = kcore_copy__compare_fds(from, to);
2448 static int kcore_copy__compare_file(const char *from_dir, const char *to_dir,
2451 char from_filename[PATH_MAX];
2452 char to_filename[PATH_MAX];
2454 scnprintf(from_filename, PATH_MAX, "%s/%s", from_dir, name);
2455 scnprintf(to_filename, PATH_MAX, "%s/%s", to_dir, name);
2457 return kcore_copy__compare_files(from_filename, to_filename);
2461 * kcore_copy - copy kallsyms, modules and kcore from one directory to another.
2462 * @from_dir: from directory
2463 * @to_dir: to directory
2465 * This function copies kallsyms, modules and kcore files from one directory to
2466 * another. kallsyms and modules are copied entirely. Only code segments are
2467 * copied from kcore. It is assumed that two segments suffice: one for the
2468 * kernel proper and one for all the modules. The code segments are determined
2469 * from kallsyms and modules files. The kernel map starts at _stext or the
2470 * lowest function symbol, and ends at _etext or the highest function symbol.
2471 * The module map starts at the lowest module address and ends at the highest
2472 * module symbol. Start addresses are rounded down to the nearest page. End
2473 * addresses are rounded up to the nearest page. An extra page is added to the
2474 * highest kernel symbol and highest module symbol to, hopefully, encompass that
2475 * symbol too. Because it contains only code sections, the resulting kcore is
2476 * unusual. One significant peculiarity is that the mapping (start -> pgoff)
2477 * is not the same for the kernel map and the modules map. That happens because
2478 * the data is copied adjacently whereas the original kcore has gaps. Finally,
2479 * kallsyms file is compared with its copy to check that modules have not been
2480 * loaded or unloaded while the copies were taking place.
2482 * Return: %0 on success, %-1 on failure.
2484 int kcore_copy(const char *from_dir, const char *to_dir)
2487 struct kcore extract;
2488 int idx = 0, err = -1;
2490 struct kcore_copy_info kci = { .stext = 0, };
2491 char kcore_filename[PATH_MAX];
2492 char extract_filename[PATH_MAX];
2493 struct phdr_data *p;
2495 INIT_LIST_HEAD(&kci.phdrs);
2496 INIT_LIST_HEAD(&kci.syms);
2498 if (kcore_copy__copy_file(from_dir, to_dir, "kallsyms"))
2501 if (kcore_copy__copy_file(from_dir, to_dir, "modules"))
2502 goto out_unlink_kallsyms;
2504 scnprintf(kcore_filename, PATH_MAX, "%s/kcore", from_dir);
2505 scnprintf(extract_filename, PATH_MAX, "%s/kcore", to_dir);
2507 if (kcore__open(&kcore, kcore_filename))
2508 goto out_unlink_modules;
2510 if (kcore_copy__calc_maps(&kci, from_dir, kcore.elf))
2511 goto out_kcore_close;
2513 if (kcore__init(&extract, extract_filename, kcore.elfclass, false))
2514 goto out_kcore_close;
2516 if (kcore__copy_hdr(&kcore, &extract, kci.phnum))
2517 goto out_extract_close;
2519 offset = gelf_fsize(extract.elf, ELF_T_EHDR, 1, EV_CURRENT) +
2520 gelf_fsize(extract.elf, ELF_T_PHDR, kci.phnum, EV_CURRENT);
2521 offset = round_up(offset, page_size);
2523 kcore_copy__for_each_phdr(&kci, p) {
2524 off_t offs = p->rel + offset;
2526 if (kcore__add_phdr(&extract, idx++, offs, p->addr, p->len))
2527 goto out_extract_close;
2530 sz = kcore__write(&extract);
2531 if (sz < 0 || sz > offset)
2532 goto out_extract_close;
2534 kcore_copy__for_each_phdr(&kci, p) {
2535 off_t offs = p->rel + offset;
2539 if (copy_bytes(kcore.fd, p->offset, extract.fd, offs, p->len))
2540 goto out_extract_close;
2543 if (kcore_copy__compare_file(from_dir, to_dir, "kallsyms"))
2544 goto out_extract_close;
2549 kcore__close(&extract);
2551 unlink(extract_filename);
2553 kcore__close(&kcore);
2556 kcore_copy__unlink(to_dir, "modules");
2557 out_unlink_kallsyms:
2559 kcore_copy__unlink(to_dir, "kallsyms");
2561 kcore_copy__free_phdrs(&kci);
2562 kcore_copy__free_syms(&kci);
2567 int kcore_extract__create(struct kcore_extract *kce)
2570 struct kcore extract;
2572 int idx = 0, err = -1;
2573 off_t offset = page_size, sz;
2575 if (kcore__open(&kcore, kce->kcore_filename))
2578 strcpy(kce->extract_filename, PERF_KCORE_EXTRACT);
2579 if (kcore__init(&extract, kce->extract_filename, kcore.elfclass, true))
2580 goto out_kcore_close;
2582 if (kcore__copy_hdr(&kcore, &extract, count))
2583 goto out_extract_close;
2585 if (kcore__add_phdr(&extract, idx, offset, kce->addr, kce->len))
2586 goto out_extract_close;
2588 sz = kcore__write(&extract);
2589 if (sz < 0 || sz > offset)
2590 goto out_extract_close;
2592 if (copy_bytes(kcore.fd, kce->offs, extract.fd, offset, kce->len))
2593 goto out_extract_close;
2598 kcore__close(&extract);
2600 unlink(kce->extract_filename);
2602 kcore__close(&kcore);
2607 void kcore_extract__delete(struct kcore_extract *kce)
2609 unlink(kce->extract_filename);
2612 #ifdef HAVE_GELF_GETNOTE_SUPPORT
2614 static void sdt_adjust_loc(struct sdt_note *tmp, GElf_Addr base_off)
2620 tmp->addr.a32[SDT_NOTE_IDX_LOC] =
2621 tmp->addr.a32[SDT_NOTE_IDX_LOC] + base_off -
2622 tmp->addr.a32[SDT_NOTE_IDX_BASE];
2624 tmp->addr.a64[SDT_NOTE_IDX_LOC] =
2625 tmp->addr.a64[SDT_NOTE_IDX_LOC] + base_off -
2626 tmp->addr.a64[SDT_NOTE_IDX_BASE];
2629 static void sdt_adjust_refctr(struct sdt_note *tmp, GElf_Addr base_addr,
2635 if (tmp->bit32 && tmp->addr.a32[SDT_NOTE_IDX_REFCTR])
2636 tmp->addr.a32[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2637 else if (tmp->addr.a64[SDT_NOTE_IDX_REFCTR])
2638 tmp->addr.a64[SDT_NOTE_IDX_REFCTR] -= (base_addr - base_off);
2642 * populate_sdt_note : Parse raw data and identify SDT note
2643 * @elf: elf of the opened file
2644 * @data: raw data of a section with description offset applied
2645 * @len: note description size
2646 * @type: type of the note
2647 * @sdt_notes: List to add the SDT note
2649 * Responsible for parsing the @data in section .note.stapsdt in @elf and
2650 * if its an SDT note, it appends to @sdt_notes list.
2652 static int populate_sdt_note(Elf **elf, const char *data, size_t len,
2653 struct list_head *sdt_notes)
2655 const char *provider, *name, *args;
2656 struct sdt_note *tmp = NULL;
2662 Elf64_Addr a64[NR_ADDR];
2663 Elf32_Addr a32[NR_ADDR];
2667 .d_buf = &buf, .d_type = ELF_T_ADDR, .d_version = EV_CURRENT,
2668 .d_size = gelf_fsize((*elf), ELF_T_ADDR, NR_ADDR, EV_CURRENT),
2669 .d_off = 0, .d_align = 0
2672 .d_buf = (void *) data, .d_type = ELF_T_ADDR,
2673 .d_version = EV_CURRENT, .d_size = dst.d_size, .d_off = 0,
2677 tmp = (struct sdt_note *)calloc(1, sizeof(struct sdt_note));
2683 INIT_LIST_HEAD(&tmp->note_list);
2685 if (len < dst.d_size + 3)
2688 /* Translation from file representation to memory representation */
2689 if (gelf_xlatetom(*elf, &dst, &src,
2690 elf_getident(*elf, NULL)[EI_DATA]) == NULL) {
2691 pr_err("gelf_xlatetom : %s\n", elf_errmsg(-1));
2695 /* Populate the fields of sdt_note */
2696 provider = data + dst.d_size;
2698 name = (const char *)memchr(provider, '\0', data + len - provider);
2702 tmp->provider = strdup(provider);
2703 if (!tmp->provider) {
2707 tmp->name = strdup(name);
2713 args = memchr(name, '\0', data + len - name);
2716 * There is no argument if:
2717 * - We reached the end of the note;
2718 * - There is not enough room to hold a potential string;
2719 * - The argument string is empty or just contains ':'.
2721 if (args == NULL || data + len - args < 2 ||
2722 args[1] == ':' || args[1] == '\0')
2725 tmp->args = strdup(++args);
2732 if (gelf_getclass(*elf) == ELFCLASS32) {
2733 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf32_Addr));
2736 memcpy(&tmp->addr, &buf, 3 * sizeof(Elf64_Addr));
2740 if (!gelf_getehdr(*elf, &ehdr)) {
2741 pr_debug("%s : cannot get elf header.\n", __func__);
2746 /* Adjust the prelink effect :
2747 * Find out the .stapsdt.base section.
2748 * This scn will help us to handle prelinking (if present).
2749 * Compare the retrieved file offset of the base section with the
2750 * base address in the description of the SDT note. If its different,
2751 * then accordingly, adjust the note location.
2753 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_BASE_SCN, NULL))
2754 sdt_adjust_loc(tmp, shdr.sh_offset);
2756 /* Adjust reference counter offset */
2757 if (elf_section_by_name(*elf, &ehdr, &shdr, SDT_PROBES_SCN, NULL))
2758 sdt_adjust_refctr(tmp, shdr.sh_addr, shdr.sh_offset);
2760 list_add_tail(&tmp->note_list, sdt_notes);
2768 zfree(&tmp->provider);
2776 * construct_sdt_notes_list : constructs a list of SDT notes
2777 * @elf : elf to look into
2778 * @sdt_notes : empty list_head
2780 * Scans the sections in 'elf' for the section
2781 * .note.stapsdt. It, then calls populate_sdt_note to find
2782 * out the SDT events and populates the 'sdt_notes'.
2784 static int construct_sdt_notes_list(Elf *elf, struct list_head *sdt_notes)
2787 Elf_Scn *scn = NULL;
2790 size_t shstrndx, next;
2792 size_t name_off, desc_off, offset;
2795 if (gelf_getehdr(elf, &ehdr) == NULL) {
2799 if (elf_getshdrstrndx(elf, &shstrndx) != 0) {
2804 /* Look for the required section */
2805 scn = elf_section_by_name(elf, &ehdr, &shdr, SDT_NOTE_SCN, NULL);
2811 if ((shdr.sh_type != SHT_NOTE) || (shdr.sh_flags & SHF_ALLOC)) {
2816 data = elf_getdata(scn, NULL);
2818 /* Get the SDT notes */
2819 for (offset = 0; (next = gelf_getnote(data, offset, &nhdr, &name_off,
2820 &desc_off)) > 0; offset = next) {
2821 if (nhdr.n_namesz == sizeof(SDT_NOTE_NAME) &&
2822 !memcmp(data->d_buf + name_off, SDT_NOTE_NAME,
2823 sizeof(SDT_NOTE_NAME))) {
2824 /* Check the type of the note */
2825 if (nhdr.n_type != SDT_NOTE_TYPE)
2828 ret = populate_sdt_note(&elf, ((data->d_buf) + desc_off),
2829 nhdr.n_descsz, sdt_notes);
2834 if (list_empty(sdt_notes))
2842 * get_sdt_note_list : Wrapper to construct a list of sdt notes
2843 * @head : empty list_head
2844 * @target : file to find SDT notes from
2846 * This opens the file, initializes
2847 * the ELF and then calls construct_sdt_notes_list.
2849 int get_sdt_note_list(struct list_head *head, const char *target)
2854 fd = open(target, O_RDONLY);
2858 elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
2863 ret = construct_sdt_notes_list(elf, head);
2871 * cleanup_sdt_note_list : free the sdt notes' list
2872 * @sdt_notes: sdt notes' list
2874 * Free up the SDT notes in @sdt_notes.
2875 * Returns the number of SDT notes free'd.
2877 int cleanup_sdt_note_list(struct list_head *sdt_notes)
2879 struct sdt_note *tmp, *pos;
2882 list_for_each_entry_safe(pos, tmp, sdt_notes, note_list) {
2883 list_del_init(&pos->note_list);
2886 zfree(&pos->provider);
2894 * sdt_notes__get_count: Counts the number of sdt events
2895 * @start: list_head to sdt_notes list
2897 * Returns the number of SDT notes in a list
2899 int sdt_notes__get_count(struct list_head *start)
2901 struct sdt_note *sdt_ptr;
2904 list_for_each_entry(sdt_ptr, start, note_list)
2910 void symbol__elf_init(void)
2912 elf_version(EV_CURRENT);