1/* Machine-dependent ELF dynamic relocation inline functions. x86-64 version.
2 Copyright (C) 2001-2018 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Andreas Jaeger <aj@suse.de>.
5
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
10
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <http://www.gnu.org/licenses/>. */
19
20#ifndef dl_machine_h
21#define dl_machine_h
22
23#define ELF_MACHINE_NAME "x86_64"
24
25#include <sys/param.h>
26#include <sysdep.h>
27#include <tls.h>
28#include <dl-tlsdesc.h>
29#include <cpu-features.c>
30
31/* Return nonzero iff ELF header is compatible with the running host. */
32static inline int __attribute__ ((unused))
33elf_machine_matches_host (const ElfW(Ehdr) *ehdr)
34{
35 return ehdr->e_machine == EM_X86_64;
36}
37
38
39/* Return the link-time address of _DYNAMIC. Conveniently, this is the
40 first element of the GOT. This must be inlined in a function which
41 uses global data. */
42static inline ElfW(Addr) __attribute__ ((unused))
43elf_machine_dynamic (void)
44{
45 /* This produces an IP-relative reloc which is resolved at link time. */
46 extern const ElfW(Addr) _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
47 return _GLOBAL_OFFSET_TABLE_[0];
48}
49
50
51/* Return the run-time load address of the shared object. */
52static inline ElfW(Addr) __attribute__ ((unused))
53elf_machine_load_address (void)
54{
55 /* Compute the difference between the runtime address of _DYNAMIC as seen
56 by an IP-relative reference, and the link-time address found in the
57 special unrelocated first GOT entry. */
58 extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
59 return (ElfW(Addr)) &_DYNAMIC - elf_machine_dynamic ();
60}
61
62/* Set up the loaded object described by L so its unrelocated PLT
63 entries will jump to the on-demand fixup code in dl-runtime.c. */
64
65static inline int __attribute__ ((unused, always_inline))
66elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
67{
68 Elf64_Addr *got;
69 extern void _dl_runtime_resolve_fxsave (ElfW(Word)) attribute_hidden;
70 extern void _dl_runtime_resolve_xsave (ElfW(Word)) attribute_hidden;
71 extern void _dl_runtime_resolve_xsavec (ElfW(Word)) attribute_hidden;
72 extern void _dl_runtime_profile_sse (ElfW(Word)) attribute_hidden;
73 extern void _dl_runtime_profile_avx (ElfW(Word)) attribute_hidden;
74 extern void _dl_runtime_profile_avx512 (ElfW(Word)) attribute_hidden;
75
76 if (l->l_info[DT_JMPREL] && lazy)
77 {
78 /* The GOT entries for functions in the PLT have not yet been filled
79 in. Their initial contents will arrange when called to push an
80 offset into the .rel.plt section, push _GLOBAL_OFFSET_TABLE_[1],
81 and then jump to _GLOBAL_OFFSET_TABLE_[2]. */
82 got = (Elf64_Addr *) D_PTR (l, l_info[DT_PLTGOT]);
83 /* If a library is prelinked but we have to relocate anyway,
84 we have to be able to undo the prelinking of .got.plt.
85 The prelinker saved us here address of .plt + 0x16. */
86 if (got[1])
87 {
88 l->l_mach.plt = got[1] + l->l_addr;
89 l->l_mach.gotplt = (ElfW(Addr)) &got[3];
90 }
91 /* Identify this shared object. */
92 *(ElfW(Addr) *) (got + 1) = (ElfW(Addr)) l;
93
94 /* The got[2] entry contains the address of a function which gets
95 called to get the address of a so far unresolved function and
96 jump to it. The profiling extension of the dynamic linker allows
97 to intercept the calls to collect information. In this case we
98 don't store the address in the GOT so that all future calls also
99 end in this function. */
100 if (__glibc_unlikely (profile))
101 {
102 if (HAS_ARCH_FEATURE (AVX512F_Usable))
103 *(ElfW(Addr) *) (got + 2) = (ElfW(Addr)) &_dl_runtime_profile_avx512;
104 else if (HAS_ARCH_FEATURE (AVX_Usable))
105 *(ElfW(Addr) *) (got + 2) = (ElfW(Addr)) &_dl_runtime_profile_avx;
106 else
107 *(ElfW(Addr) *) (got + 2) = (ElfW(Addr)) &_dl_runtime_profile_sse;
108
109 if (GLRO(dl_profile) != NULL
110 && _dl_name_match_p (GLRO(dl_profile), l))
111 /* This is the object we are looking for. Say that we really
112 want profiling and the timers are started. */
113 GL(dl_profile_map) = l;
114 }
115 else
116 {
117 /* This function will get called to fix up the GOT entry
118 indicated by the offset on the stack, and then jump to
119 the resolved address. */
120 if (GLRO(dl_x86_cpu_features).xsave_state_size != 0)
121 *(ElfW(Addr) *) (got + 2)
122 = (HAS_ARCH_FEATURE (XSAVEC_Usable)
123 ? (ElfW(Addr)) &_dl_runtime_resolve_xsavec
124 : (ElfW(Addr)) &_dl_runtime_resolve_xsave);
125 else
126 *(ElfW(Addr) *) (got + 2)
127 = (ElfW(Addr)) &_dl_runtime_resolve_fxsave;
128 }
129 }
130
131 if (l->l_info[ADDRIDX (DT_TLSDESC_GOT)] && lazy)
132 *(ElfW(Addr)*)(D_PTR (l, l_info[ADDRIDX (DT_TLSDESC_GOT)]) + l->l_addr)
133 = (ElfW(Addr)) &_dl_tlsdesc_resolve_rela;
134
135 return lazy;
136}
137
138/* Initial entry point code for the dynamic linker.
139 The C function `_dl_start' is the real entry point;
140 its return value is the user program's entry point. */
141#define RTLD_START asm ("\n\
142.text\n\
143 .align 16\n\
144.globl _start\n\
145.globl _dl_start_user\n\
146_start:\n\
147 movq %rsp, %rdi\n\
148 call _dl_start\n\
149_dl_start_user:\n\
150 # Save the user entry point address in %r12.\n\
151 movq %rax, %r12\n\
152 # See if we were run as a command with the executable file\n\
153 # name as an extra leading argument.\n\
154 movl _dl_skip_args(%rip), %eax\n\
155 # Pop the original argument count.\n\
156 popq %rdx\n\
157 # Adjust the stack pointer to skip _dl_skip_args words.\n\
158 leaq (%rsp,%rax,8), %rsp\n\
159 # Subtract _dl_skip_args from argc.\n\
160 subl %eax, %edx\n\
161 # Push argc back on the stack.\n\
162 pushq %rdx\n\
163 # Call _dl_init (struct link_map *main_map, int argc, char **argv, char **env)\n\
164 # argc -> rsi\n\
165 movq %rdx, %rsi\n\
166 # Save %rsp value in %r13.\n\
167 movq %rsp, %r13\n\
168 # And align stack for the _dl_init call. \n\
169 andq $-16, %rsp\n\
170 # _dl_loaded -> rdi\n\
171 movq _rtld_local(%rip), %rdi\n\
172 # env -> rcx\n\
173 leaq 16(%r13,%rdx,8), %rcx\n\
174 # argv -> rdx\n\
175 leaq 8(%r13), %rdx\n\
176 # Clear %rbp to mark outermost frame obviously even for constructors.\n\
177 xorl %ebp, %ebp\n\
178 # Call the function to run the initializers.\n\
179 call _dl_init\n\
180 # Pass our finalizer function to the user in %rdx, as per ELF ABI.\n\
181 leaq _dl_fini(%rip), %rdx\n\
182 # And make sure %rsp points to argc stored on the stack.\n\
183 movq %r13, %rsp\n\
184 # Jump to the user's entry point.\n\
185 jmp *%r12\n\
186.previous\n\
187");
188
189/* ELF_RTYPE_CLASS_PLT iff TYPE describes relocation of a PLT entry or
190 TLS variable, so undefined references should not be allowed to
191 define the value.
192 ELF_RTYPE_CLASS_COPY iff TYPE should not be allowed to resolve to one
193 of the main executable's symbols, as for a COPY reloc.
194 ELF_RTYPE_CLASS_EXTERN_PROTECTED_DATA iff TYPE describes relocation may
195 against protected data whose address be external due to copy relocation.
196 */
197#define elf_machine_type_class(type) \
198 ((((type) == R_X86_64_JUMP_SLOT \
199 || (type) == R_X86_64_DTPMOD64 \
200 || (type) == R_X86_64_DTPOFF64 \
201 || (type) == R_X86_64_TPOFF64 \
202 || (type) == R_X86_64_TLSDESC) \
203 * ELF_RTYPE_CLASS_PLT) \
204 | (((type) == R_X86_64_COPY) * ELF_RTYPE_CLASS_COPY) \
205 | (((type) == R_X86_64_GLOB_DAT) * ELF_RTYPE_CLASS_EXTERN_PROTECTED_DATA))
206
207/* A reloc type used for ld.so cmdline arg lookups to reject PLT entries. */
208#define ELF_MACHINE_JMP_SLOT R_X86_64_JUMP_SLOT
209
210/* The relative ifunc relocation. */
211// XXX This is a work-around for a broken linker. Remove!
212#define ELF_MACHINE_IRELATIVE R_X86_64_IRELATIVE
213
214/* The x86-64 never uses Elf64_Rel/Elf32_Rel relocations. */
215#define ELF_MACHINE_NO_REL 1
216#define ELF_MACHINE_NO_RELA 0
217
218/* We define an initialization function. This is called very early in
219 _dl_sysdep_start. */
220#define DL_PLATFORM_INIT dl_platform_init ()
221
222static inline void __attribute__ ((unused))
223dl_platform_init (void)
224{
225#if IS_IN (rtld)
226 /* init_cpu_features has been called early from __libc_start_main in
227 static executable. */
228 init_cpu_features (&GLRO(dl_x86_cpu_features));
229#else
230 if (GLRO(dl_platform) != NULL && *GLRO(dl_platform) == '\0')
231 /* Avoid an empty string which would disturb us. */
232 GLRO(dl_platform) = NULL;
233#endif
234}
235
236static inline ElfW(Addr)
237elf_machine_fixup_plt (struct link_map *map, lookup_t t,
238 const ElfW(Sym) *refsym, const ElfW(Sym) *sym,
239 const ElfW(Rela) *reloc,
240 ElfW(Addr) *reloc_addr, ElfW(Addr) value)
241{
242 return *reloc_addr = value;
243}
244
245/* Return the final value of a PLT relocation. On x86-64 the
246 JUMP_SLOT relocation ignores the addend. */
247static inline ElfW(Addr)
248elf_machine_plt_value (struct link_map *map, const ElfW(Rela) *reloc,
249 ElfW(Addr) value)
250{
251 return value;
252}
253
254
255/* Names of the architecture-specific auditing callback functions. */
256#define ARCH_LA_PLTENTER x86_64_gnu_pltenter
257#define ARCH_LA_PLTEXIT x86_64_gnu_pltexit
258
259#endif /* !dl_machine_h */
260
261#ifdef RESOLVE_MAP
262
263/* Perform the relocation specified by RELOC and SYM (which is fully resolved).
264 MAP is the object containing the reloc. */
265
266auto inline void
267__attribute__ ((always_inline))
268elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc,
269 const ElfW(Sym) *sym, const struct r_found_version *version,
270 void *const reloc_addr_arg, int skip_ifunc)
271{
272 ElfW(Addr) *const reloc_addr = reloc_addr_arg;
273 const unsigned long int r_type = ELFW(R_TYPE) (reloc->r_info);
274
275# if !defined RTLD_BOOTSTRAP || !defined HAVE_Z_COMBRELOC
276 if (__glibc_unlikely (r_type == R_X86_64_RELATIVE))
277 {
278# if !defined RTLD_BOOTSTRAP && !defined HAVE_Z_COMBRELOC
279 /* This is defined in rtld.c, but nowhere in the static libc.a;
280 make the reference weak so static programs can still link.
281 This declaration cannot be done when compiling rtld.c
282 (i.e. #ifdef RTLD_BOOTSTRAP) because rtld.c contains the
283 common defn for _dl_rtld_map, which is incompatible with a
284 weak decl in the same file. */
285# ifndef SHARED
286 weak_extern (GL(dl_rtld_map));
287# endif
288 if (map != &GL(dl_rtld_map)) /* Already done in rtld itself. */
289# endif
290 *reloc_addr = map->l_addr + reloc->r_addend;
291 }
292 else
293# endif
294# if !defined RTLD_BOOTSTRAP
295 /* l_addr + r_addend may be > 0xffffffff and R_X86_64_RELATIVE64
296 relocation updates the whole 64-bit entry. */
297 if (__glibc_unlikely (r_type == R_X86_64_RELATIVE64))
298 *(Elf64_Addr *) reloc_addr = (Elf64_Addr) map->l_addr + reloc->r_addend;
299 else
300# endif
301 if (__glibc_unlikely (r_type == R_X86_64_NONE))
302 return;
303 else
304 {
305# ifndef RTLD_BOOTSTRAP
306 const ElfW(Sym) *const refsym = sym;
307# endif
308 struct link_map *sym_map = RESOLVE_MAP (&sym, version, r_type);
309 ElfW(Addr) value = (sym == NULL ? 0
310 : (ElfW(Addr)) sym_map->l_addr + sym->st_value);
311
312 if (sym != NULL
313 && __builtin_expect (ELFW(ST_TYPE) (sym->st_info) == STT_GNU_IFUNC,
314 0)
315 && __builtin_expect (sym->st_shndx != SHN_UNDEF, 1)
316 && __builtin_expect (!skip_ifunc, 1))
317 {
318# ifndef RTLD_BOOTSTRAP
319 if (sym_map != map
320 && sym_map->l_type != lt_executable
321 && !sym_map->l_relocated)
322 {
323 const char *strtab
324 = (const char *) D_PTR (map, l_info[DT_STRTAB]);
325 _dl_error_printf ("\
326%s: Relink `%s' with `%s' for IFUNC symbol `%s'\n",
327 RTLD_PROGNAME, map->l_name,
328 sym_map->l_name,
329 strtab + refsym->st_name);
330 }
331# endif
332 value = ((ElfW(Addr) (*) (void)) value) ();
333 }
334
335 switch (r_type)
336 {
337# ifndef RTLD_BOOTSTRAP
338# ifdef __ILP32__
339 case R_X86_64_SIZE64:
340 /* Set to symbol size plus addend. */
341 *(Elf64_Addr *) (uintptr_t) reloc_addr
342 = (Elf64_Addr) sym->st_size + reloc->r_addend;
343 break;
344
345 case R_X86_64_SIZE32:
346# else
347 case R_X86_64_SIZE64:
348# endif
349 /* Set to symbol size plus addend. */
350 value = sym->st_size;
351# endif
352 case R_X86_64_GLOB_DAT:
353 case R_X86_64_JUMP_SLOT:
354 *reloc_addr = value + reloc->r_addend;
355 break;
356
357# ifndef RESOLVE_CONFLICT_FIND_MAP
358 case R_X86_64_DTPMOD64:
359# ifdef RTLD_BOOTSTRAP
360 /* During startup the dynamic linker is always the module
361 with index 1.
362 XXX If this relocation is necessary move before RESOLVE
363 call. */
364 *reloc_addr = 1;
365# else
366 /* Get the information from the link map returned by the
367 resolve function. */
368 if (sym_map != NULL)
369 *reloc_addr = sym_map->l_tls_modid;
370# endif
371 break;
372 case R_X86_64_DTPOFF64:
373# ifndef RTLD_BOOTSTRAP
374 /* During relocation all TLS symbols are defined and used.
375 Therefore the offset is already correct. */
376 if (sym != NULL)
377 {
378 value = sym->st_value + reloc->r_addend;
379# ifdef __ILP32__
380 /* This relocation type computes a signed offset that is
381 usually negative. The symbol and addend values are 32
382 bits but the GOT entry is 64 bits wide and the whole
383 64-bit entry is used as a signed quantity, so we need
384 to sign-extend the computed value to 64 bits. */
385 *(Elf64_Sxword *) reloc_addr = (Elf64_Sxword) (Elf32_Sword) value;
386# else
387 *reloc_addr = value;
388# endif
389 }
390# endif
391 break;
392 case R_X86_64_TLSDESC:
393 {
394 struct tlsdesc volatile *td =
395 (struct tlsdesc volatile *)reloc_addr;
396
397# ifndef RTLD_BOOTSTRAP
398 if (! sym)
399 {
400 td->arg = (void*)reloc->r_addend;
401 td->entry = _dl_tlsdesc_undefweak;
402 }
403 else
404# endif
405 {
406# ifndef RTLD_BOOTSTRAP
407# ifndef SHARED
408 CHECK_STATIC_TLS (map, sym_map);
409# else
410 if (!TRY_STATIC_TLS (map, sym_map))
411 {
412 td->arg = _dl_make_tlsdesc_dynamic
413 (sym_map, sym->st_value + reloc->r_addend);
414 td->entry = _dl_tlsdesc_dynamic;
415 }
416 else
417# endif
418# endif
419 {
420 td->arg = (void*)(sym->st_value - sym_map->l_tls_offset
421 + reloc->r_addend);
422 td->entry = _dl_tlsdesc_return;
423 }
424 }
425 break;
426 }
427 case R_X86_64_TPOFF64:
428 /* The offset is negative, forward from the thread pointer. */
429# ifndef RTLD_BOOTSTRAP
430 if (sym != NULL)
431# endif
432 {
433# ifndef RTLD_BOOTSTRAP
434 CHECK_STATIC_TLS (map, sym_map);
435# endif
436 /* We know the offset of the object the symbol is contained in.
437 It is a negative value which will be added to the
438 thread pointer. */
439 value = (sym->st_value + reloc->r_addend
440 - sym_map->l_tls_offset);
441# ifdef __ILP32__
442 /* The symbol and addend values are 32 bits but the GOT
443 entry is 64 bits wide and the whole 64-bit entry is used
444 as a signed quantity, so we need to sign-extend the
445 computed value to 64 bits. */
446 *(Elf64_Sxword *) reloc_addr = (Elf64_Sxword) (Elf32_Sword) value;
447# else
448 *reloc_addr = value;
449# endif
450 }
451 break;
452# endif
453
454# ifndef RTLD_BOOTSTRAP
455 case R_X86_64_64:
456 /* value + r_addend may be > 0xffffffff and R_X86_64_64
457 relocation updates the whole 64-bit entry. */
458 *(Elf64_Addr *) reloc_addr = (Elf64_Addr) value + reloc->r_addend;
459 break;
460# ifndef __ILP32__
461 case R_X86_64_SIZE32:
462 /* Set to symbol size plus addend. */
463 value = sym->st_size;
464# endif
465 case R_X86_64_32:
466 value += reloc->r_addend;
467 *(unsigned int *) reloc_addr = value;
468
469 const char *fmt;
470 if (__glibc_unlikely (value > UINT_MAX))
471 {
472 const char *strtab;
473
474 fmt = "\
475%s: Symbol `%s' causes overflow in R_X86_64_32 relocation\n";
476# ifndef RESOLVE_CONFLICT_FIND_MAP
477 print_err:
478# endif
479 strtab = (const char *) D_PTR (map, l_info[DT_STRTAB]);
480
481 _dl_error_printf (fmt, RTLD_PROGNAME, strtab + refsym->st_name);
482 }
483 break;
484# ifndef RESOLVE_CONFLICT_FIND_MAP
485 /* Not needed for dl-conflict.c. */
486 case R_X86_64_PC32:
487 value += reloc->r_addend - (ElfW(Addr)) reloc_addr;
488 *(unsigned int *) reloc_addr = value;
489 if (__glibc_unlikely (value != (int) value))
490 {
491 fmt = "\
492%s: Symbol `%s' causes overflow in R_X86_64_PC32 relocation\n";
493 goto print_err;
494 }
495 break;
496 case R_X86_64_COPY:
497 if (sym == NULL)
498 /* This can happen in trace mode if an object could not be
499 found. */
500 break;
501 memcpy (reloc_addr_arg, (void *) value,
502 MIN (sym->st_size, refsym->st_size));
503 if (__builtin_expect (sym->st_size > refsym->st_size, 0)
504 || (__builtin_expect (sym->st_size < refsym->st_size, 0)
505 && GLRO(dl_verbose)))
506 {
507 fmt = "\
508%s: Symbol `%s' has different size in shared object, consider re-linking\n";
509 goto print_err;
510 }
511 break;
512# endif
513 case R_X86_64_IRELATIVE:
514 value = map->l_addr + reloc->r_addend;
515 value = ((ElfW(Addr) (*) (void)) value) ();
516 *reloc_addr = value;
517 break;
518 default:
519 _dl_reloc_bad_type (map, r_type, 0);
520 break;
521# endif
522 }
523 }
524}
525
526auto inline void
527__attribute ((always_inline))
528elf_machine_rela_relative (ElfW(Addr) l_addr, const ElfW(Rela) *reloc,
529 void *const reloc_addr_arg)
530{
531 ElfW(Addr) *const reloc_addr = reloc_addr_arg;
532#if !defined RTLD_BOOTSTRAP
533 /* l_addr + r_addend may be > 0xffffffff and R_X86_64_RELATIVE64
534 relocation updates the whole 64-bit entry. */
535 if (__glibc_unlikely (ELFW(R_TYPE) (reloc->r_info) == R_X86_64_RELATIVE64))
536 *(Elf64_Addr *) reloc_addr = (Elf64_Addr) l_addr + reloc->r_addend;
537 else
538#endif
539 {
540 assert (ELFW(R_TYPE) (reloc->r_info) == R_X86_64_RELATIVE);
541 *reloc_addr = l_addr + reloc->r_addend;
542 }
543}
544
545auto inline void
546__attribute ((always_inline))
547elf_machine_lazy_rel (struct link_map *map,
548 ElfW(Addr) l_addr, const ElfW(Rela) *reloc,
549 int skip_ifunc)
550{
551 ElfW(Addr) *const reloc_addr = (void *) (l_addr + reloc->r_offset);
552 const unsigned long int r_type = ELFW(R_TYPE) (reloc->r_info);
553
554 /* Check for unexpected PLT reloc type. */
555 if (__glibc_likely (r_type == R_X86_64_JUMP_SLOT))
556 {
557 if (__builtin_expect (map->l_mach.plt, 0) == 0)
558 *reloc_addr += l_addr;
559 else
560 *reloc_addr =
561 map->l_mach.plt
562 + (((ElfW(Addr)) reloc_addr) - map->l_mach.gotplt) * 2;
563 }
564 else if (__glibc_likely (r_type == R_X86_64_TLSDESC))
565 {
566 struct tlsdesc volatile * __attribute__((__unused__)) td =
567 (struct tlsdesc volatile *)reloc_addr;
568
569 td->arg = (void*)reloc;
570 td->entry = (void*)(D_PTR (map, l_info[ADDRIDX (DT_TLSDESC_PLT)])
571 + map->l_addr);
572 }
573 else if (__glibc_unlikely (r_type == R_X86_64_IRELATIVE))
574 {
575 ElfW(Addr) value = map->l_addr + reloc->r_addend;
576 if (__glibc_likely (!skip_ifunc))
577 value = ((ElfW(Addr) (*) (void)) value) ();
578 *reloc_addr = value;
579 }
580 else
581 _dl_reloc_bad_type (map, r_type, 1);
582}
583
584#endif /* RESOLVE_MAP */
585