Add tcov support in Makefile
[tinycc.git] / tccelf.c
blobb46ad4ca31b68b0214650da007dd6141d91c8871
1 /*
2 * ELF file handling for TCC
4 * Copyright (c) 2001-2004 Fabrice Bellard
6 * This 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 of the License, or (at your option) any later version.
11 * This 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.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include "tcc.h"
23 /* Define this to get some debug output during relocation processing. */
24 #undef DEBUG_RELOC
26 /********************************************************/
27 /* global variables */
29 /* elf version information */
30 struct sym_version {
31 char *lib;
32 char *version;
33 int out_index;
34 int prev_same_lib;
37 #define nb_sym_versions s1->nb_sym_versions
38 #define sym_versions s1->sym_versions
39 #define nb_sym_to_version s1->nb_sym_to_version
40 #define sym_to_version s1->sym_to_version
41 #define dt_verneednum s1->dt_verneednum
42 #define versym_section s1->versym_section
43 #define verneed_section s1->verneed_section
45 /* special flag to indicate that the section should not be linked to the other ones */
46 #define SHF_PRIVATE 0x80000000
47 /* section is dynsymtab_section */
48 #define SHF_DYNSYM 0x40000000
50 #ifdef TCC_TARGET_PE
51 static const int shf_RELRO = SHF_ALLOC;
52 static const char rdata[] = ".rdata";
53 #else
54 static const int shf_RELRO = SHF_ALLOC | SHF_WRITE;
55 static const char rdata[] = ".data.ro";
56 #endif
58 /* ------------------------------------------------------------------------- */
60 ST_FUNC void tccelf_new(TCCState *s)
62 TCCState *s1 = s;
63 /* no section zero */
64 dynarray_add(&s->sections, &s->nb_sections, NULL);
66 /* create standard sections */
67 text_section = new_section(s, ".text", SHT_PROGBITS, SHF_ALLOC | SHF_EXECINSTR);
68 data_section = new_section(s, ".data", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
69 /* create ro data section (make ro after relocation done with GNU_RELRO) */
70 rodata_section = new_section(s, rdata, SHT_PROGBITS, shf_RELRO);
71 bss_section = new_section(s, ".bss", SHT_NOBITS, SHF_ALLOC | SHF_WRITE);
72 common_section = new_section(s, ".common", SHT_NOBITS, SHF_PRIVATE);
73 common_section->sh_num = SHN_COMMON;
75 /* symbols are always generated for linking stage */
76 symtab_section = new_symtab(s, ".symtab", SHT_SYMTAB, 0,
77 ".strtab",
78 ".hashtab", SHF_PRIVATE);
79 s->symtab = symtab_section;
81 /* private symbol table for dynamic symbols */
82 s->dynsymtab_section = new_symtab(s, ".dynsymtab", SHT_SYMTAB, SHF_PRIVATE|SHF_DYNSYM,
83 ".dynstrtab",
84 ".dynhashtab", SHF_PRIVATE);
85 get_sym_attr(s, 0, 1);
88 #ifdef CONFIG_TCC_BCHECK
89 ST_FUNC void tccelf_bounds_new(TCCState *s)
91 TCCState *s1 = s;
92 /* create bounds sections (make ro after relocation done with GNU_RELRO) */
93 bounds_section = new_section(s, ".bounds", SHT_PROGBITS, shf_RELRO);
94 lbounds_section = new_section(s, ".lbounds", SHT_PROGBITS, shf_RELRO);
96 #endif
98 static void free_section(Section *s)
100 tcc_free(s->data);
103 ST_FUNC void tccelf_delete(TCCState *s1)
105 int i;
107 #ifndef ELF_OBJ_ONLY
108 /* free symbol versions */
109 for (i = 0; i < nb_sym_versions; i++) {
110 tcc_free(sym_versions[i].version);
111 tcc_free(sym_versions[i].lib);
113 tcc_free(sym_versions);
114 tcc_free(sym_to_version);
115 #endif
117 /* free all sections */
118 for(i = 1; i < s1->nb_sections; i++)
119 free_section(s1->sections[i]);
120 dynarray_reset(&s1->sections, &s1->nb_sections);
122 for(i = 0; i < s1->nb_priv_sections; i++)
123 free_section(s1->priv_sections[i]);
124 dynarray_reset(&s1->priv_sections, &s1->nb_priv_sections);
126 /* free any loaded DLLs */
127 #ifdef TCC_IS_NATIVE
128 for ( i = 0; i < s1->nb_loaded_dlls; i++) {
129 DLLReference *ref = s1->loaded_dlls[i];
130 if ( ref->handle )
131 # ifdef _WIN32
132 FreeLibrary((HMODULE)ref->handle);
133 # else
134 dlclose(ref->handle);
135 # endif
137 #endif
138 /* free loaded dlls array */
139 dynarray_reset(&s1->loaded_dlls, &s1->nb_loaded_dlls);
140 tcc_free(s1->sym_attrs);
142 symtab_section = NULL; /* for tccrun.c:rt_printline() */
145 /* save section data state */
146 ST_FUNC void tccelf_begin_file(TCCState *s1)
148 Section *s; int i;
149 for (i = 1; i < s1->nb_sections; i++) {
150 s = s1->sections[i];
151 s->sh_offset = s->data_offset;
153 /* disable symbol hashing during compilation */
154 s = s1->symtab, s->reloc = s->hash, s->hash = NULL;
155 #if defined TCC_TARGET_X86_64 && defined TCC_TARGET_PE
156 s1->uw_sym = 0;
157 #endif
160 /* At the end of compilation, convert any UNDEF syms to global, and merge
161 with previously existing symbols */
162 ST_FUNC void tccelf_end_file(TCCState *s1)
164 Section *s = s1->symtab;
165 int first_sym, nb_syms, *tr, i;
167 first_sym = s->sh_offset / sizeof (ElfSym);
168 nb_syms = s->data_offset / sizeof (ElfSym) - first_sym;
169 s->data_offset = s->sh_offset;
170 s->link->data_offset = s->link->sh_offset;
171 s->hash = s->reloc, s->reloc = NULL;
172 tr = tcc_mallocz(nb_syms * sizeof *tr);
174 for (i = 0; i < nb_syms; ++i) {
175 ElfSym *sym = (ElfSym*)s->data + first_sym + i;
176 if (sym->st_shndx == SHN_UNDEF
177 && ELFW(ST_BIND)(sym->st_info) == STB_LOCAL)
178 sym->st_info = ELFW(ST_INFO)(STB_GLOBAL, ELFW(ST_TYPE)(sym->st_info));
179 tr[i] = set_elf_sym(s, sym->st_value, sym->st_size, sym->st_info,
180 sym->st_other, sym->st_shndx, (char*)s->link->data + sym->st_name);
182 /* now update relocations */
183 for (i = 1; i < s1->nb_sections; i++) {
184 Section *sr = s1->sections[i];
185 if (sr->sh_type == SHT_RELX && sr->link == s) {
186 ElfW_Rel *rel = (ElfW_Rel*)(sr->data + sr->sh_offset);
187 ElfW_Rel *rel_end = (ElfW_Rel*)(sr->data + sr->data_offset);
188 for (; rel < rel_end; ++rel) {
189 int n = ELFW(R_SYM)(rel->r_info) - first_sym;
190 if (n < 0) /* zero sym_index in reloc (can happen with asm) */
191 continue;
192 rel->r_info = ELFW(R_INFO)(tr[n], ELFW(R_TYPE)(rel->r_info));
196 tcc_free(tr);
198 /* record text/data/bss output for -bench info */
199 for (i = 0; i < 4; ++i) {
200 s = s1->sections[i + 1];
201 s1->total_output[i] += s->data_offset - s->sh_offset;
205 ST_FUNC Section *new_section(TCCState *s1, const char *name, int sh_type, int sh_flags)
207 Section *sec;
209 sec = tcc_mallocz(sizeof(Section) + strlen(name));
210 sec->s1 = s1;
211 strcpy(sec->name, name);
212 sec->sh_type = sh_type;
213 sec->sh_flags = sh_flags;
214 switch(sh_type) {
215 case SHT_GNU_versym:
216 sec->sh_addralign = 2;
217 break;
218 case SHT_HASH:
219 case SHT_GNU_HASH:
220 case SHT_REL:
221 case SHT_RELA:
222 case SHT_DYNSYM:
223 case SHT_SYMTAB:
224 case SHT_DYNAMIC:
225 case SHT_GNU_verneed:
226 case SHT_GNU_verdef:
227 sec->sh_addralign = PTR_SIZE;
228 break;
229 case SHT_STRTAB:
230 sec->sh_addralign = 1;
231 break;
232 default:
233 sec->sh_addralign = PTR_SIZE; /* gcc/pcc default alignment */
234 break;
237 if (sh_flags & SHF_PRIVATE) {
238 dynarray_add(&s1->priv_sections, &s1->nb_priv_sections, sec);
239 } else {
240 sec->sh_num = s1->nb_sections;
241 dynarray_add(&s1->sections, &s1->nb_sections, sec);
244 return sec;
247 ST_FUNC Section *new_symtab(TCCState *s1,
248 const char *symtab_name, int sh_type, int sh_flags,
249 const char *strtab_name,
250 const char *hash_name, int hash_sh_flags)
252 Section *symtab, *strtab, *hash;
253 int *ptr, nb_buckets;
255 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
256 symtab->sh_entsize = sizeof(ElfW(Sym));
257 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
258 put_elf_str(strtab, "");
259 symtab->link = strtab;
260 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
262 nb_buckets = 1;
264 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
265 hash->sh_entsize = sizeof(int);
266 symtab->hash = hash;
267 hash->link = symtab;
269 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
270 ptr[0] = nb_buckets;
271 ptr[1] = 1;
272 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
273 return symtab;
276 /* realloc section and set its content to zero */
277 ST_FUNC void section_realloc(Section *sec, unsigned long new_size)
279 unsigned long size;
280 unsigned char *data;
282 size = sec->data_allocated;
283 if (size == 0)
284 size = 1;
285 while (size < new_size)
286 size = size * 2;
287 data = tcc_realloc(sec->data, size);
288 memset(data + sec->data_allocated, 0, size - sec->data_allocated);
289 sec->data = data;
290 sec->data_allocated = size;
293 /* reserve at least 'size' bytes aligned per 'align' in section
294 'sec' from current offset, and return the aligned offset */
295 ST_FUNC size_t section_add(Section *sec, addr_t size, int align)
297 size_t offset, offset1;
299 offset = (sec->data_offset + align - 1) & -align;
300 offset1 = offset + size;
301 if (sec->sh_type != SHT_NOBITS && offset1 > sec->data_allocated)
302 section_realloc(sec, offset1);
303 sec->data_offset = offset1;
304 if (align > sec->sh_addralign)
305 sec->sh_addralign = align;
306 return offset;
309 /* reserve at least 'size' bytes in section 'sec' from
310 sec->data_offset. */
311 ST_FUNC void *section_ptr_add(Section *sec, addr_t size)
313 size_t offset = section_add(sec, size, 1);
314 return sec->data + offset;
317 #ifndef ELF_OBJ_ONLY
318 /* reserve at least 'size' bytes from section start */
319 static void section_reserve(Section *sec, unsigned long size)
321 if (size > sec->data_allocated)
322 section_realloc(sec, size);
323 if (size > sec->data_offset)
324 sec->data_offset = size;
326 #endif
328 static Section *have_section(TCCState *s1, const char *name)
330 Section *sec;
331 int i;
332 for(i = 1; i < s1->nb_sections; i++) {
333 sec = s1->sections[i];
334 if (!strcmp(name, sec->name))
335 return sec;
337 return NULL;
340 /* return a reference to a section, and create it if it does not
341 exists */
342 ST_FUNC Section *find_section(TCCState *s1, const char *name)
344 Section *sec = have_section(s1, name);
345 if (sec)
346 return sec;
347 /* sections are created as PROGBITS */
348 return new_section(s1, name, SHT_PROGBITS, SHF_ALLOC);
351 /* ------------------------------------------------------------------------- */
353 ST_FUNC int put_elf_str(Section *s, const char *sym)
355 int offset, len;
356 char *ptr;
358 len = strlen(sym) + 1;
359 offset = s->data_offset;
360 ptr = section_ptr_add(s, len);
361 memmove(ptr, sym, len);
362 return offset;
365 /* elf symbol hashing function */
366 static ElfW(Word) elf_hash(const unsigned char *name)
368 ElfW(Word) h = 0, g;
370 while (*name) {
371 h = (h << 4) + *name++;
372 g = h & 0xf0000000;
373 if (g)
374 h ^= g >> 24;
375 h &= ~g;
377 return h;
380 /* rebuild hash table of section s */
381 /* NOTE: we do factorize the hash table code to go faster */
382 static void rebuild_hash(Section *s, unsigned int nb_buckets)
384 ElfW(Sym) *sym;
385 int *ptr, *hash, nb_syms, sym_index, h;
386 unsigned char *strtab;
388 strtab = s->link->data;
389 nb_syms = s->data_offset / sizeof(ElfW(Sym));
391 if (!nb_buckets)
392 nb_buckets = ((int*)s->hash->data)[0];
394 s->hash->data_offset = 0;
395 ptr = section_ptr_add(s->hash, (2 + nb_buckets + nb_syms) * sizeof(int));
396 ptr[0] = nb_buckets;
397 ptr[1] = nb_syms;
398 ptr += 2;
399 hash = ptr;
400 memset(hash, 0, (nb_buckets + 1) * sizeof(int));
401 ptr += nb_buckets + 1;
403 sym = (ElfW(Sym) *)s->data + 1;
404 for(sym_index = 1; sym_index < nb_syms; sym_index++) {
405 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
406 h = elf_hash(strtab + sym->st_name) % nb_buckets;
407 *ptr = hash[h];
408 hash[h] = sym_index;
409 } else {
410 *ptr = 0;
412 ptr++;
413 sym++;
417 /* return the symbol number */
418 ST_FUNC int put_elf_sym(Section *s, addr_t value, unsigned long size,
419 int info, int other, int shndx, const char *name)
421 int name_offset, sym_index;
422 int nbuckets, h;
423 ElfW(Sym) *sym;
424 Section *hs;
426 sym = section_ptr_add(s, sizeof(ElfW(Sym)));
427 if (name && name[0])
428 name_offset = put_elf_str(s->link, name);
429 else
430 name_offset = 0;
431 /* XXX: endianness */
432 sym->st_name = name_offset;
433 sym->st_value = value;
434 sym->st_size = size;
435 sym->st_info = info;
436 sym->st_other = other;
437 sym->st_shndx = shndx;
438 sym_index = sym - (ElfW(Sym) *)s->data;
439 hs = s->hash;
440 if (hs) {
441 int *ptr, *base;
442 ptr = section_ptr_add(hs, sizeof(int));
443 base = (int *)hs->data;
444 /* only add global or weak symbols. */
445 if (ELFW(ST_BIND)(info) != STB_LOCAL) {
446 /* add another hashing entry */
447 nbuckets = base[0];
448 h = elf_hash((unsigned char *)s->link->data + name_offset) % nbuckets;
449 *ptr = base[2 + h];
450 base[2 + h] = sym_index;
451 base[1]++;
452 /* we resize the hash table */
453 hs->nb_hashed_syms++;
454 if (hs->nb_hashed_syms > 2 * nbuckets) {
455 rebuild_hash(s, 2 * nbuckets);
457 } else {
458 *ptr = 0;
459 base[1]++;
462 return sym_index;
465 ST_FUNC int find_elf_sym(Section *s, const char *name)
467 ElfW(Sym) *sym;
468 Section *hs;
469 int nbuckets, sym_index, h;
470 const char *name1;
472 hs = s->hash;
473 if (!hs)
474 return 0;
475 nbuckets = ((int *)hs->data)[0];
476 h = elf_hash((unsigned char *) name) % nbuckets;
477 sym_index = ((int *)hs->data)[2 + h];
478 while (sym_index != 0) {
479 sym = &((ElfW(Sym) *)s->data)[sym_index];
480 name1 = (char *) s->link->data + sym->st_name;
481 if (!strcmp(name, name1))
482 return sym_index;
483 sym_index = ((int *)hs->data)[2 + nbuckets + sym_index];
485 return 0;
488 /* return elf symbol value, signal error if 'err' is nonzero, decorate
489 name if FORC */
490 ST_FUNC addr_t get_sym_addr(TCCState *s1, const char *name, int err, int forc)
492 int sym_index;
493 ElfW(Sym) *sym;
494 char buf[256];
495 if (forc && s1->leading_underscore
496 #ifdef TCC_TARGET_PE
497 /* win32-32bit stdcall symbols always have _ already */
498 && !strchr(name, '@')
499 #endif
501 buf[0] = '_';
502 pstrcpy(buf + 1, sizeof(buf) - 1, name);
503 name = buf;
505 sym_index = find_elf_sym(s1->symtab, name);
506 sym = &((ElfW(Sym) *)s1->symtab->data)[sym_index];
507 if (!sym_index || sym->st_shndx == SHN_UNDEF) {
508 if (err)
509 tcc_error_noabort("%s not defined", name);
510 return (addr_t)-1;
512 return sym->st_value;
515 /* return elf symbol value */
516 LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
518 addr_t addr = get_sym_addr(s, name, 0, 1);
519 return addr == -1 ? NULL : (void*)(uintptr_t)addr;
522 /* list elf symbol names and values */
523 ST_FUNC void list_elf_symbols(TCCState *s, void *ctx,
524 void (*symbol_cb)(void *ctx, const char *name, const void *val))
526 ElfW(Sym) *sym;
527 Section *symtab;
528 int sym_index, end_sym;
529 const char *name;
530 unsigned char sym_vis, sym_bind;
532 symtab = s->symtab;
533 end_sym = symtab->data_offset / sizeof (ElfSym);
534 for (sym_index = 0; sym_index < end_sym; ++sym_index) {
535 sym = &((ElfW(Sym) *)symtab->data)[sym_index];
536 if (sym->st_value) {
537 name = (char *) symtab->link->data + sym->st_name;
538 sym_bind = ELFW(ST_BIND)(sym->st_info);
539 sym_vis = ELFW(ST_VISIBILITY)(sym->st_other);
540 if (sym_bind == STB_GLOBAL && sym_vis == STV_DEFAULT)
541 symbol_cb(ctx, name, (void*)(uintptr_t)sym->st_value);
546 /* list elf symbol names and values */
547 LIBTCCAPI void tcc_list_symbols(TCCState *s, void *ctx,
548 void (*symbol_cb)(void *ctx, const char *name, const void *val))
550 list_elf_symbols(s, ctx, symbol_cb);
553 #ifndef ELF_OBJ_ONLY
554 static void
555 version_add (TCCState *s1)
557 int i;
558 ElfW(Sym) *sym;
559 ElfW(Verneed) *vn = NULL;
560 Section *symtab;
561 int sym_index, end_sym, nb_versions = 2, nb_entries = 0;
562 ElfW(Half) *versym;
563 const char *name;
565 if (0 == nb_sym_versions)
566 return;
567 versym_section = new_section(s1, ".gnu.version", SHT_GNU_versym, SHF_ALLOC);
568 versym_section->sh_entsize = sizeof(ElfW(Half));
569 versym_section->link = s1->dynsym;
571 /* add needed symbols */
572 symtab = s1->dynsym;
573 end_sym = symtab->data_offset / sizeof (ElfSym);
574 versym = section_ptr_add(versym_section, end_sym * sizeof(ElfW(Half)));
575 for (sym_index = 1; sym_index < end_sym; ++sym_index) {
576 int dllindex, verndx;
577 sym = &((ElfW(Sym) *)symtab->data)[sym_index];
578 if (sym->st_shndx != SHN_UNDEF)
579 continue; /* defined symbol doesn't need library version */
580 name = (char *) symtab->link->data + sym->st_name;
581 dllindex = find_elf_sym(s1->dynsymtab_section, name);
582 verndx = (dllindex && dllindex < nb_sym_to_version)
583 ? sym_to_version[dllindex] : -1;
584 if (verndx >= 0) {
585 if (!sym_versions[verndx].out_index)
586 sym_versions[verndx].out_index = nb_versions++;
587 versym[sym_index] = sym_versions[verndx].out_index;
590 /* generate verneed section, but not when it will be empty. Some
591 dynamic linkers look at their contents even when DTVERNEEDNUM and
592 section size is zero. */
593 if (nb_versions > 2) {
594 verneed_section = new_section(s1, ".gnu.version_r",
595 SHT_GNU_verneed, SHF_ALLOC);
596 verneed_section->link = s1->dynsym->link;
597 for (i = nb_sym_versions; i-- > 0;) {
598 struct sym_version *sv = &sym_versions[i];
599 int n_same_libs = 0, prev;
600 size_t vnofs;
601 ElfW(Vernaux) *vna = 0;
602 if (sv->out_index < 1)
603 continue;
605 /* make sure that a DT_NEEDED tag is put */
606 /* abitest-tcc fails on older i386-linux with "ld-linux.so.2" DT_NEEDED
607 ret_int_test... Inconsistency detected by ld.so: dl-minimal.c: 148:
608 realloc: Assertion `ptr == alloc_last_block' failed! */
609 if (strcmp(sv->lib, "ld-linux.so.2"))
610 tcc_add_dllref(s1, sv->lib, 0);
612 vnofs = section_add(verneed_section, sizeof(*vn), 1);
613 vn = (ElfW(Verneed)*)(verneed_section->data + vnofs);
614 vn->vn_version = 1;
615 vn->vn_file = put_elf_str(verneed_section->link, sv->lib);
616 vn->vn_aux = sizeof (*vn);
617 do {
618 prev = sv->prev_same_lib;
619 if (sv->out_index > 0) {
620 vna = section_ptr_add(verneed_section, sizeof(*vna));
621 vna->vna_hash = elf_hash ((const unsigned char *)sv->version);
622 vna->vna_flags = 0;
623 vna->vna_other = sv->out_index;
624 sv->out_index = -2;
625 vna->vna_name = put_elf_str(verneed_section->link, sv->version);
626 vna->vna_next = sizeof (*vna);
627 n_same_libs++;
629 if (prev >= 0)
630 sv = &sym_versions[prev];
631 } while(prev >= 0);
632 vna->vna_next = 0;
633 vn = (ElfW(Verneed)*)(verneed_section->data + vnofs);
634 vn->vn_cnt = n_same_libs;
635 vn->vn_next = sizeof(*vn) + n_same_libs * sizeof(*vna);
636 nb_entries++;
638 if (vn)
639 vn->vn_next = 0;
640 verneed_section->sh_info = nb_entries;
642 dt_verneednum = nb_entries;
644 #endif /* ndef ELF_OBJ_ONLY */
646 /* add an elf symbol : check if it is already defined and patch
647 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
648 ST_FUNC int set_elf_sym(Section *s, addr_t value, unsigned long size,
649 int info, int other, int shndx, const char *name)
651 TCCState *s1 = s->s1;
652 ElfW(Sym) *esym;
653 int sym_bind, sym_index, sym_type, esym_bind;
654 unsigned char sym_vis, esym_vis, new_vis;
656 sym_bind = ELFW(ST_BIND)(info);
657 sym_type = ELFW(ST_TYPE)(info);
658 sym_vis = ELFW(ST_VISIBILITY)(other);
660 if (sym_bind != STB_LOCAL) {
661 /* we search global or weak symbols */
662 sym_index = find_elf_sym(s, name);
663 if (!sym_index)
664 goto do_def;
665 esym = &((ElfW(Sym) *)s->data)[sym_index];
666 if (esym->st_value == value && esym->st_size == size && esym->st_info == info
667 && esym->st_other == other && esym->st_shndx == shndx)
668 return sym_index;
669 if (esym->st_shndx != SHN_UNDEF) {
670 esym_bind = ELFW(ST_BIND)(esym->st_info);
671 /* propagate the most constraining visibility */
672 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
673 esym_vis = ELFW(ST_VISIBILITY)(esym->st_other);
674 if (esym_vis == STV_DEFAULT) {
675 new_vis = sym_vis;
676 } else if (sym_vis == STV_DEFAULT) {
677 new_vis = esym_vis;
678 } else {
679 new_vis = (esym_vis < sym_vis) ? esym_vis : sym_vis;
681 esym->st_other = (esym->st_other & ~ELFW(ST_VISIBILITY)(-1))
682 | new_vis;
683 if (shndx == SHN_UNDEF) {
684 /* ignore adding of undefined symbol if the
685 corresponding symbol is already defined */
686 } else if (sym_bind == STB_GLOBAL && esym_bind == STB_WEAK) {
687 /* global overrides weak, so patch */
688 goto do_patch;
689 } else if (sym_bind == STB_WEAK && esym_bind == STB_GLOBAL) {
690 /* weak is ignored if already global */
691 } else if (sym_bind == STB_WEAK && esym_bind == STB_WEAK) {
692 /* keep first-found weak definition, ignore subsequents */
693 } else if (sym_vis == STV_HIDDEN || sym_vis == STV_INTERNAL) {
694 /* ignore hidden symbols after */
695 } else if ((esym->st_shndx == SHN_COMMON
696 || esym->st_shndx == bss_section->sh_num)
697 && (shndx < SHN_LORESERVE
698 && shndx != bss_section->sh_num)) {
699 /* data symbol gets precedence over common/bss */
700 goto do_patch;
701 } else if (shndx == SHN_COMMON || shndx == bss_section->sh_num) {
702 /* data symbol keeps precedence over common/bss */
703 } else if (s->sh_flags & SHF_DYNSYM) {
704 /* we accept that two DLL define the same symbol */
705 } else if (esym->st_other & ST_ASM_SET) {
706 /* If the existing symbol came from an asm .set
707 we can override. */
708 goto do_patch;
709 } else {
710 #if 0
711 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
712 sym_bind, shndx, new_vis, esym_bind, esym->st_shndx, esym_vis);
713 #endif
714 tcc_error_noabort("'%s' defined twice", name);
716 } else {
717 esym->st_other = other;
718 do_patch:
719 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
720 esym->st_shndx = shndx;
721 s1->new_undef_sym = 1;
722 esym->st_value = value;
723 esym->st_size = size;
725 } else {
726 do_def:
727 sym_index = put_elf_sym(s, value, size,
728 ELFW(ST_INFO)(sym_bind, sym_type), other,
729 shndx, name);
731 return sym_index;
734 /* put relocation */
735 ST_FUNC void put_elf_reloca(Section *symtab, Section *s, unsigned long offset,
736 int type, int symbol, addr_t addend)
738 TCCState *s1 = s->s1;
739 char buf[256];
740 Section *sr;
741 ElfW_Rel *rel;
743 sr = s->reloc;
744 if (!sr) {
745 /* if no relocation section, create it */
746 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
747 /* if the symtab is allocated, then we consider the relocation
748 are also */
749 sr = new_section(s->s1, buf, SHT_RELX, symtab->sh_flags);
750 sr->sh_entsize = sizeof(ElfW_Rel);
751 sr->link = symtab;
752 sr->sh_info = s->sh_num;
753 s->reloc = sr;
755 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
756 rel->r_offset = offset;
757 rel->r_info = ELFW(R_INFO)(symbol, type);
758 #if SHT_RELX == SHT_RELA
759 rel->r_addend = addend;
760 #endif
761 if (SHT_RELX != SHT_RELA && addend)
762 tcc_error_noabort("non-zero addend on REL architecture");
765 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
766 int type, int symbol)
768 put_elf_reloca(symtab, s, offset, type, symbol, 0);
771 ST_FUNC struct sym_attr *get_sym_attr(TCCState *s1, int index, int alloc)
773 int n;
774 struct sym_attr *tab;
776 if (index >= s1->nb_sym_attrs) {
777 if (!alloc)
778 return s1->sym_attrs;
779 /* find immediately bigger power of 2 and reallocate array */
780 n = 1;
781 while (index >= n)
782 n *= 2;
783 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
784 s1->sym_attrs = tab;
785 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
786 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
787 s1->nb_sym_attrs = n;
789 return &s1->sym_attrs[index];
792 static void modify_reloctions_old_to_new(TCCState *s1, Section *s, int *old_to_new_syms)
794 int i, type, sym_index;
795 Section *sr;
796 ElfW_Rel *rel;
798 for(i = 1; i < s1->nb_sections; i++) {
799 sr = s1->sections[i];
800 if (sr->sh_type == SHT_RELX && sr->link == s) {
801 for_each_elem(sr, 0, rel, ElfW_Rel) {
802 sym_index = ELFW(R_SYM)(rel->r_info);
803 type = ELFW(R_TYPE)(rel->r_info);
804 sym_index = old_to_new_syms[sym_index];
805 rel->r_info = ELFW(R_INFO)(sym_index, type);
811 /* In an ELF file symbol table, the local symbols must appear below
812 the global and weak ones. Since TCC cannot sort it while generating
813 the code, we must do it after. All the relocation tables are also
814 modified to take into account the symbol table sorting */
815 static void sort_syms(TCCState *s1, Section *s)
817 int *old_to_new_syms;
818 ElfW(Sym) *new_syms;
819 int nb_syms, i;
820 ElfW(Sym) *p, *q;
822 nb_syms = s->data_offset / sizeof(ElfW(Sym));
823 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
824 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
826 /* first pass for local symbols */
827 p = (ElfW(Sym) *)s->data;
828 q = new_syms;
829 for(i = 0; i < nb_syms; i++) {
830 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
831 old_to_new_syms[i] = q - new_syms;
832 *q++ = *p;
834 p++;
836 /* save the number of local symbols in section header */
837 if( s->sh_size ) /* this 'if' makes IDA happy */
838 s->sh_info = q - new_syms;
840 /* then second pass for non local symbols */
841 p = (ElfW(Sym) *)s->data;
842 for(i = 0; i < nb_syms; i++) {
843 if (ELFW(ST_BIND)(p->st_info) != STB_LOCAL) {
844 old_to_new_syms[i] = q - new_syms;
845 *q++ = *p;
847 p++;
850 /* we copy the new symbols to the old */
851 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
852 tcc_free(new_syms);
854 modify_reloctions_old_to_new(s1, s, old_to_new_syms);
856 tcc_free(old_to_new_syms);
859 #ifndef ELF_OBJ_ONLY
860 /* See: https://flapenguin.me/elf-dt-gnu-hash */
861 #define ELFCLASS_BITS (PTR_SIZE * 8)
863 static Section *create_gnu_hash(TCCState *s1)
865 int nb_syms, i, ndef, nbuckets, symoffset, bloom_size, bloom_shift;
866 ElfW(Sym) *p;
867 Section *gnu_hash;
868 Section *dynsym = s1->dynsym;
869 Elf32_Word *ptr;
871 gnu_hash = new_section(s1, ".gnu.hash", SHT_GNU_HASH, SHF_ALLOC);
872 gnu_hash->link = dynsym->hash->link;
874 nb_syms = dynsym->data_offset / sizeof(ElfW(Sym));
876 /* count def symbols */
877 ndef = 0;
878 p = (ElfW(Sym) *)dynsym->data;
879 for(i = 0; i < nb_syms; i++, p++)
880 ndef += p->st_shndx != SHN_UNDEF;
882 /* calculate gnu hash sizes and fill header */
883 nbuckets = ndef / 4 + 1;
884 symoffset = nb_syms - ndef;
885 bloom_shift = PTR_SIZE == 8 ? 6 : 5;
886 bloom_size = 1; /* must be power of two */
887 while (ndef >= bloom_size * (1 << (bloom_shift - 3)))
888 bloom_size *= 2;
889 ptr = section_ptr_add(gnu_hash, 4 * 4 +
890 PTR_SIZE * bloom_size +
891 nbuckets * 4 +
892 ndef * 4);
893 ptr[0] = nbuckets;
894 ptr[1] = symoffset;
895 ptr[2] = bloom_size;
896 ptr[3] = bloom_shift;
897 return gnu_hash;
900 static Elf32_Word elf_gnu_hash (const unsigned char *name)
902 Elf32_Word h = 5381;
903 unsigned char c;
905 while ((c = *name++))
906 h = h * 33 + c;
907 return h;
910 static void update_gnu_hash(TCCState *s1, Section *gnu_hash)
912 int *old_to_new_syms;
913 ElfW(Sym) *new_syms;
914 int nb_syms, i, nbuckets, bloom_size, bloom_shift;
915 ElfW(Sym) *p, *q;
916 Section *vs;
917 Section *dynsym = s1->dynsym;
918 Elf32_Word *ptr, *buckets, *chain, *hash;
919 unsigned int *nextbuck;
920 addr_t *bloom;
921 unsigned char *strtab;
922 struct { int first, last; } *buck;
924 strtab = dynsym->link->data;
925 nb_syms = dynsym->data_offset / sizeof(ElfW(Sym));
926 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
927 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
928 hash = tcc_malloc(nb_syms * sizeof(Elf32_Word));
929 nextbuck = tcc_malloc(nb_syms * sizeof(int));
931 /* calculate hashes and copy undefs */
932 p = (ElfW(Sym) *)dynsym->data;
933 q = new_syms;
934 for(i = 0; i < nb_syms; i++, p++) {
935 if (p->st_shndx == SHN_UNDEF) {
936 old_to_new_syms[i] = q - new_syms;
937 *q++ = *p;
939 else
940 hash[i] = elf_gnu_hash(strtab + p->st_name);
943 ptr = (Elf32_Word *) gnu_hash->data;
944 nbuckets = ptr[0];
945 bloom_size = ptr[2];
946 bloom_shift = ptr[3];
947 bloom = (addr_t *) (void *) &ptr[4];
948 buckets = (Elf32_Word*) (void *) &bloom[bloom_size];
949 chain = &buckets[nbuckets];
950 buck = tcc_malloc(nbuckets * sizeof(*buck));
952 if (gnu_hash->data_offset != 4 * 4 +
953 PTR_SIZE * bloom_size +
954 nbuckets * 4 +
955 (nb_syms - (q - new_syms)) * 4)
956 tcc_error_noabort ("gnu_hash size incorrect");
958 /* find buckets */
959 for(i = 0; i < nbuckets; i++)
960 buck[i].first = -1;
962 p = (ElfW(Sym) *)dynsym->data;
963 for(i = 0; i < nb_syms; i++, p++)
964 if (p->st_shndx != SHN_UNDEF) {
965 int bucket = hash[i] % nbuckets;
967 if (buck[bucket].first == -1)
968 buck[bucket].first = buck[bucket].last = i;
969 else {
970 nextbuck[buck[bucket].last] = i;
971 buck[bucket].last = i;
975 /* fill buckets/chains/bloom and sort symbols */
976 p = (ElfW(Sym) *)dynsym->data;
977 for(i = 0; i < nbuckets; i++) {
978 int cur = buck[i].first;
980 if (cur != -1) {
981 buckets[i] = q - new_syms;
982 for (;;) {
983 old_to_new_syms[cur] = q - new_syms;
984 *q++ = p[cur];
985 *chain++ = hash[cur] & ~1;
986 bloom[(hash[cur] / ELFCLASS_BITS) % bloom_size] |=
987 (addr_t)1 << (hash[cur] % ELFCLASS_BITS) |
988 (addr_t)1 << ((hash[cur] >> bloom_shift) % ELFCLASS_BITS);
989 if (cur == buck[i].last)
990 break;
991 cur = nextbuck[cur];
993 chain[-1] |= 1;
997 memcpy(dynsym->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
998 tcc_free(new_syms);
999 tcc_free(hash);
1000 tcc_free(buck);
1001 tcc_free(nextbuck);
1003 modify_reloctions_old_to_new(s1, dynsym, old_to_new_syms);
1005 /* modify the versions */
1006 vs = versym_section;
1007 if (vs) {
1008 ElfW(Half) *newver, *versym = (ElfW(Half) *)vs->data;
1010 if (1/*versym*/) {
1011 newver = tcc_malloc(nb_syms * sizeof(*newver));
1012 for (i = 0; i < nb_syms; i++)
1013 newver[old_to_new_syms[i]] = versym[i];
1014 memcpy(vs->data, newver, nb_syms * sizeof(*newver));
1015 tcc_free(newver);
1019 tcc_free(old_to_new_syms);
1021 /* rebuild hash */
1022 ptr = (Elf32_Word *) dynsym->hash->data;
1023 rebuild_hash(dynsym, ptr[0]);
1025 #endif /* ELF_OBJ_ONLY */
1027 /* relocate symbol table, resolve undefined symbols if do_resolve is
1028 true and output error if undefined symbol. */
1029 ST_FUNC void relocate_syms(TCCState *s1, Section *symtab, int do_resolve)
1031 ElfW(Sym) *sym;
1032 int sym_bind, sh_num;
1033 const char *name;
1035 for_each_elem(symtab, 1, sym, ElfW(Sym)) {
1036 sh_num = sym->st_shndx;
1037 if (sh_num == SHN_UNDEF) {
1038 if (do_resolve == 2) /* relocating dynsym */
1039 continue;
1040 name = (char *) s1->symtab->link->data + sym->st_name;
1041 /* Use ld.so to resolve symbol for us (for tcc -run) */
1042 if (do_resolve) {
1043 #if defined TCC_IS_NATIVE && !defined TCC_TARGET_PE
1044 /* dlsym() needs the undecorated name. */
1045 void *addr = dlsym(RTLD_DEFAULT, &name[s1->leading_underscore]);
1046 #if TARGETOS_OpenBSD || TARGETOS_FreeBSD || TARGETOS_NetBSD || TARGETOS_ANDROID
1047 if (addr == NULL) {
1048 int i;
1049 for (i = 0; i < s1->nb_loaded_dlls; i++)
1050 if ((addr = dlsym(s1->loaded_dlls[i]->handle, name)))
1051 break;
1053 #endif
1054 if (addr) {
1055 sym->st_value = (addr_t) addr;
1056 #ifdef DEBUG_RELOC
1057 printf ("relocate_sym: %s -> 0x%lx\n", name, sym->st_value);
1058 #endif
1059 goto found;
1061 #endif
1062 /* if dynamic symbol exist, it will be used in relocate_section */
1063 } else if (s1->dynsym && find_elf_sym(s1->dynsym, name))
1064 goto found;
1065 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
1066 it */
1067 if (!strcmp(name, "_fp_hw"))
1068 goto found;
1069 /* only weak symbols are accepted to be undefined. Their
1070 value is zero */
1071 sym_bind = ELFW(ST_BIND)(sym->st_info);
1072 if (sym_bind == STB_WEAK)
1073 sym->st_value = 0;
1074 else
1075 tcc_error_noabort("undefined symbol '%s'", name);
1077 } else if (sh_num < SHN_LORESERVE) {
1078 /* add section base */
1079 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
1081 found: ;
1085 /* relocate a given section (CPU dependent) by applying the relocations
1086 in the associated relocation section */
1087 static void relocate_section(TCCState *s1, Section *s, Section *sr)
1089 ElfW_Rel *rel;
1090 ElfW(Sym) *sym;
1091 int type, sym_index;
1092 unsigned char *ptr;
1093 addr_t tgt, addr;
1094 int is_dwarf = s->sh_num >= s1->dwlo && s->sh_num < s1->dwhi;
1096 qrel = (ElfW_Rel *)sr->data;
1097 for_each_elem(sr, 0, rel, ElfW_Rel) {
1098 ptr = s->data + rel->r_offset;
1099 sym_index = ELFW(R_SYM)(rel->r_info);
1100 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1101 type = ELFW(R_TYPE)(rel->r_info);
1102 tgt = sym->st_value;
1103 #if SHT_RELX == SHT_RELA
1104 tgt += rel->r_addend;
1105 #endif
1106 if (is_dwarf && type == R_DATA_32DW
1107 && sym->st_shndx >= s1->dwlo && sym->st_shndx < s1->dwhi) {
1108 /* dwarf section relocation to each other */
1109 add32le(ptr, tgt - s1->sections[sym->st_shndx]->sh_addr);
1110 continue;
1112 addr = s->sh_addr + rel->r_offset;
1113 relocate(s1, rel, type, ptr, addr, tgt);
1115 #ifndef ELF_OBJ_ONLY
1116 /* if the relocation is allocated, we change its symbol table */
1117 if (sr->sh_flags & SHF_ALLOC) {
1118 sr->link = s1->dynsym;
1119 if (s1->output_type & TCC_OUTPUT_DYN) {
1120 size_t r = (uint8_t*)qrel - sr->data;
1121 if (sizeof ((Stab_Sym*)0)->n_value < PTR_SIZE
1122 && 0 == strcmp(s->name, ".stab"))
1123 r = 0; /* cannot apply 64bit relocation to 32bit value */
1124 sr->data_offset = sr->sh_size = r;
1125 #ifdef CONFIG_TCC_PIE
1126 if (r && 0 == (s->sh_flags & SHF_WRITE))
1127 tcc_warning("%d relocations to ro-section %s", (unsigned)(r / sizeof *qrel), s->name);
1128 #endif
1131 #endif
1134 /* relocate all sections */
1135 ST_FUNC void relocate_sections(TCCState *s1)
1137 int i;
1138 Section *s, *sr;
1140 for (i = 1; i < s1->nb_sections; ++i) {
1141 sr = s1->sections[i];
1142 if (sr->sh_type != SHT_RELX)
1143 continue;
1144 s = s1->sections[sr->sh_info];
1145 #ifndef TCC_TARGET_MACHO
1146 if (s != s1->got
1147 || s1->static_link
1148 || s1->output_type == TCC_OUTPUT_MEMORY)
1149 #endif
1151 relocate_section(s1, s, sr);
1153 #ifndef ELF_OBJ_ONLY
1154 if (sr->sh_flags & SHF_ALLOC) {
1155 ElfW_Rel *rel;
1156 /* relocate relocation table in 'sr' */
1157 for_each_elem(sr, 0, rel, ElfW_Rel)
1158 rel->r_offset += s->sh_addr;
1160 #endif
1164 #ifndef ELF_OBJ_ONLY
1165 /* count the number of dynamic relocations so that we can reserve
1166 their space */
1167 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
1169 int count = 0;
1170 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64) || \
1171 defined(TCC_TARGET_ARM) || defined(TCC_TARGET_ARM64) || \
1172 defined(TCC_TARGET_RISCV64)
1173 ElfW_Rel *rel;
1174 for_each_elem(sr, 0, rel, ElfW_Rel) {
1175 int sym_index = ELFW(R_SYM)(rel->r_info);
1176 int type = ELFW(R_TYPE)(rel->r_info);
1177 switch(type) {
1178 #if defined(TCC_TARGET_I386)
1179 case R_386_32:
1180 if (!get_sym_attr(s1, sym_index, 0)->dyn_index
1181 && ((ElfW(Sym)*)symtab_section->data + sym_index)->st_shndx == SHN_UNDEF) {
1182 /* don't fixup unresolved (weak) symbols */
1183 rel->r_info = ELFW(R_INFO)(sym_index, R_386_RELATIVE);
1184 break;
1186 #elif defined(TCC_TARGET_X86_64)
1187 case R_X86_64_32:
1188 case R_X86_64_32S:
1189 case R_X86_64_64:
1190 #elif defined(TCC_TARGET_ARM)
1191 case R_ARM_ABS32:
1192 case R_ARM_TARGET1:
1193 #elif defined(TCC_TARGET_ARM64)
1194 case R_AARCH64_ABS32:
1195 case R_AARCH64_ABS64:
1196 #elif defined(TCC_TARGET_RISCV64)
1197 case R_RISCV_32:
1198 case R_RISCV_64:
1199 #endif
1200 count++;
1201 break;
1202 #if defined(TCC_TARGET_I386)
1203 case R_386_PC32:
1204 #elif defined(TCC_TARGET_X86_64)
1205 case R_X86_64_PC32:
1207 ElfW(Sym) *sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1208 /* Hidden defined symbols can and must be resolved locally.
1209 We're misusing a PLT32 reloc for this, as that's always
1210 resolved to its address even in shared libs. */
1211 if (sym->st_shndx != SHN_UNDEF &&
1212 ELFW(ST_VISIBILITY)(sym->st_other) == STV_HIDDEN) {
1213 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PLT32);
1214 break;
1217 #elif defined(TCC_TARGET_ARM64)
1218 case R_AARCH64_PREL32:
1219 #endif
1220 if (s1->output_type != TCC_OUTPUT_DLL)
1221 break;
1222 if (get_sym_attr(s1, sym_index, 0)->dyn_index)
1223 count++;
1224 break;
1225 default:
1226 break;
1229 #endif
1230 return count;
1232 #endif
1234 #ifdef NEED_BUILD_GOT
1235 static int build_got(TCCState *s1)
1237 /* if no got, then create it */
1238 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1239 s1->got->sh_entsize = 4;
1240 /* keep space for _DYNAMIC pointer and two dummy got entries */
1241 section_ptr_add(s1->got, 3 * PTR_SIZE);
1242 return set_elf_sym(symtab_section, 0, 0, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1243 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1246 /* Create a GOT and (for function call) a PLT entry corresponding to a symbol
1247 in s1->symtab. When creating the dynamic symbol table entry for the GOT
1248 relocation, use 'size' and 'info' for the corresponding symbol metadata.
1249 Returns the offset of the GOT or (if any) PLT entry. */
1250 static struct sym_attr * put_got_entry(TCCState *s1, int dyn_reloc_type,
1251 int sym_index)
1253 int need_plt_entry;
1254 const char *name;
1255 ElfW(Sym) *sym;
1256 struct sym_attr *attr;
1257 unsigned got_offset;
1258 char plt_name[200];
1259 int len;
1260 Section *s_rel;
1262 need_plt_entry = (dyn_reloc_type == R_JMP_SLOT);
1263 attr = get_sym_attr(s1, sym_index, 1);
1265 /* In case a function is both called and its address taken 2 GOT entries
1266 are created, one for taking the address (GOT) and the other for the PLT
1267 entry (PLTGOT). */
1268 if (need_plt_entry ? attr->plt_offset : attr->got_offset)
1269 return attr;
1271 s_rel = s1->got;
1272 if (need_plt_entry) {
1273 if (!s1->plt) {
1274 s1->plt = new_section(s1, ".plt", SHT_PROGBITS, SHF_ALLOC | SHF_EXECINSTR);
1275 s1->plt->sh_entsize = 4;
1277 s_rel = s1->plt;
1280 /* create the GOT entry */
1281 got_offset = s1->got->data_offset;
1282 section_ptr_add(s1->got, PTR_SIZE);
1284 /* Create the GOT relocation that will insert the address of the object or
1285 function of interest in the GOT entry. This is a static relocation for
1286 memory output (dlsym will give us the address of symbols) and dynamic
1287 relocation otherwise (executable and DLLs). The relocation should be
1288 done lazily for GOT entry with *_JUMP_SLOT relocation type (the one
1289 associated to a PLT entry) but is currently done at load time for an
1290 unknown reason. */
1292 sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1293 name = (char *) symtab_section->link->data + sym->st_name;
1294 //printf("sym %d %s\n", need_plt_entry, name);
1296 if (s1->dynsym) {
1297 if (ELFW(ST_BIND)(sym->st_info) == STB_LOCAL) {
1298 /* Hack alarm. We don't want to emit dynamic symbols
1299 and symbol based relocs for STB_LOCAL symbols, but rather
1300 want to resolve them directly. At this point the symbol
1301 values aren't final yet, so we must defer this. We will later
1302 have to create a RELATIVE reloc anyway, so we misuse the
1303 relocation slot to smuggle the symbol reference until
1304 fill_local_got_entries. Not that the sym_index is
1305 relative to symtab_section, not s1->dynsym! Nevertheless
1306 we use s1->dyn_sym so that if this is the first call
1307 that got->reloc is correctly created. Also note that
1308 RELATIVE relocs are not normally created for the .got,
1309 so the types serves as a marker for later (and is retained
1310 also for the final output, which is okay because then the
1311 got is just normal data). */
1312 put_elf_reloc(s1->dynsym, s1->got, got_offset, R_RELATIVE,
1313 sym_index);
1314 } else {
1315 if (0 == attr->dyn_index)
1316 attr->dyn_index = set_elf_sym(s1->dynsym, sym->st_value,
1317 sym->st_size, sym->st_info, 0,
1318 sym->st_shndx, name);
1319 put_elf_reloc(s1->dynsym, s_rel, got_offset, dyn_reloc_type,
1320 attr->dyn_index);
1322 } else {
1323 put_elf_reloc(symtab_section, s1->got, got_offset, dyn_reloc_type,
1324 sym_index);
1327 if (need_plt_entry) {
1328 attr->plt_offset = create_plt_entry(s1, got_offset, attr);
1330 /* create a symbol 'sym@plt' for the PLT jump vector */
1331 len = strlen(name);
1332 if (len > sizeof plt_name - 5)
1333 len = sizeof plt_name - 5;
1334 memcpy(plt_name, name, len);
1335 strcpy(plt_name + len, "@plt");
1336 attr->plt_sym = put_elf_sym(s1->symtab, attr->plt_offset, 0,
1337 ELFW(ST_INFO)(STB_GLOBAL, STT_FUNC), 0, s1->plt->sh_num, plt_name);
1338 } else {
1339 attr->got_offset = got_offset;
1342 return attr;
1345 /* build GOT and PLT entries */
1346 /* Two passes because R_JMP_SLOT should become first. Some targets
1347 (arm, arm64) do not allow mixing R_JMP_SLOT and R_GLOB_DAT. */
1348 ST_FUNC void build_got_entries(TCCState *s1, int got_sym)
1350 Section *s;
1351 ElfW_Rel *rel;
1352 ElfW(Sym) *sym;
1353 int i, type, gotplt_entry, reloc_type, sym_index;
1354 struct sym_attr *attr;
1355 int pass = 0;
1356 redo:
1357 for(i = 1; i < s1->nb_sections; i++) {
1358 s = s1->sections[i];
1359 if (s->sh_type != SHT_RELX)
1360 continue;
1361 /* no need to handle got relocations */
1362 if (s->link != symtab_section)
1363 continue;
1364 for_each_elem(s, 0, rel, ElfW_Rel) {
1365 type = ELFW(R_TYPE)(rel->r_info);
1366 gotplt_entry = gotplt_entry_type(type);
1367 if (gotplt_entry == -1) {
1368 tcc_error_noabort ("Unknown relocation type for got: %d", type);
1369 continue;
1371 sym_index = ELFW(R_SYM)(rel->r_info);
1372 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1374 if (gotplt_entry == NO_GOTPLT_ENTRY) {
1375 continue;
1378 /* Automatically create PLT/GOT [entry] if it is an undefined
1379 reference (resolved at runtime), or the symbol is absolute,
1380 probably created by tcc_add_symbol, and thus on 64-bit
1381 targets might be too far from application code. */
1382 if (gotplt_entry == AUTO_GOTPLT_ENTRY) {
1383 if (sym->st_shndx == SHN_UNDEF) {
1384 ElfW(Sym) *esym;
1385 int dynindex;
1386 if (!PCRELATIVE_DLLPLT
1387 && (s1->output_type & TCC_OUTPUT_DYN))
1388 continue;
1389 /* Relocations for UNDEF symbols would normally need
1390 to be transferred into the executable or shared object.
1391 If that were done AUTO_GOTPLT_ENTRY wouldn't exist.
1392 But TCC doesn't do that (at least for exes), so we
1393 need to resolve all such relocs locally. And that
1394 means PLT slots for functions in DLLs and COPY relocs for
1395 data symbols. COPY relocs were generated in
1396 bind_exe_dynsyms (and the symbol adjusted to be defined),
1397 and for functions we were generated a dynamic symbol
1398 of function type. */
1399 if (s1->dynsym) {
1400 /* dynsym isn't set for -run :-/ */
1401 dynindex = get_sym_attr(s1, sym_index, 0)->dyn_index;
1402 esym = (ElfW(Sym) *)s1->dynsym->data + dynindex;
1403 if (dynindex
1404 && (ELFW(ST_TYPE)(esym->st_info) == STT_FUNC
1405 || (ELFW(ST_TYPE)(esym->st_info) == STT_NOTYPE
1406 && ELFW(ST_TYPE)(sym->st_info) == STT_FUNC)))
1407 goto jmp_slot;
1409 } else if (sym->st_shndx == SHN_ABS) {
1410 if (sym->st_value == 0) /* from tcc_add_btstub() */
1411 continue;
1412 #ifndef TCC_TARGET_ARM
1413 if (PTR_SIZE != 8)
1414 continue;
1415 #endif
1416 /* from tcc_add_symbol(): on 64 bit platforms these
1417 need to go through .got */
1418 } else
1419 continue;
1422 #ifdef TCC_TARGET_X86_64
1423 if ((type == R_X86_64_PLT32 || type == R_X86_64_PC32) &&
1424 sym->st_shndx != SHN_UNDEF &&
1425 (ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT ||
1426 ELFW(ST_BIND)(sym->st_info) == STB_LOCAL ||
1427 s1->output_type & TCC_OUTPUT_EXE)) {
1428 if (pass != 0)
1429 continue;
1430 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1431 continue;
1433 #endif
1434 reloc_type = code_reloc(type);
1435 if (reloc_type == -1) {
1436 tcc_error_noabort ("Unknown relocation type: %d", type);
1437 continue;
1440 if (reloc_type != 0) {
1441 jmp_slot:
1442 if (pass != 0)
1443 continue;
1444 reloc_type = R_JMP_SLOT;
1445 } else {
1446 if (pass != 1)
1447 continue;
1448 reloc_type = R_GLOB_DAT;
1451 if (!s1->got)
1452 got_sym = build_got(s1);
1454 if (gotplt_entry == BUILD_GOT_ONLY)
1455 continue;
1457 attr = put_got_entry(s1, reloc_type, sym_index);
1459 if (reloc_type == R_JMP_SLOT)
1460 rel->r_info = ELFW(R_INFO)(attr->plt_sym, type);
1463 if (++pass < 2)
1464 goto redo;
1465 /* .rel.plt refers to .got actually */
1466 if (s1->plt && s1->plt->reloc)
1467 s1->plt->reloc->sh_info = s1->got->sh_num;
1468 if (got_sym) /* set size */
1469 ((ElfW(Sym)*)symtab_section->data)[got_sym].st_size = s1->got->data_offset;
1471 #endif /* def NEED_BUILD_GOT */
1473 ST_FUNC int set_global_sym(TCCState *s1, const char *name, Section *sec, addr_t offs)
1475 int shn = sec ? sec->sh_num : offs || !name ? SHN_ABS : SHN_UNDEF;
1476 if (sec && offs == -1)
1477 offs = sec->data_offset;
1478 return set_elf_sym(symtab_section, offs, 0,
1479 ELFW(ST_INFO)(name ? STB_GLOBAL : STB_LOCAL, STT_NOTYPE), 0, shn, name);
1482 static void add_init_array_defines(TCCState *s1, const char *section_name)
1484 Section *s;
1485 addr_t end_offset;
1486 char buf[1024];
1487 s = have_section(s1, section_name);
1488 if (!s || !(s->sh_flags & SHF_ALLOC)) {
1489 end_offset = 0;
1490 s = data_section;
1491 } else {
1492 end_offset = s->data_offset;
1494 snprintf(buf, sizeof(buf), "__%s_start", section_name + 1);
1495 set_global_sym(s1, buf, s, 0);
1496 snprintf(buf, sizeof(buf), "__%s_end", section_name + 1);
1497 set_global_sym(s1, buf, s, end_offset);
1500 ST_FUNC void add_array (TCCState *s1, const char *sec, int c)
1502 Section *s;
1503 s = find_section(s1, sec);
1504 s->sh_flags = shf_RELRO;
1505 s->sh_type = sec[1] == 'i' ? SHT_INIT_ARRAY : SHT_FINI_ARRAY;
1506 put_elf_reloc (s1->symtab, s, s->data_offset, R_DATA_PTR, c);
1507 section_ptr_add(s, PTR_SIZE);
1510 #ifdef CONFIG_TCC_BCHECK
1511 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1513 if (0 == s1->do_bounds_check)
1514 return;
1515 section_ptr_add(bounds_section, sizeof(addr_t));
1517 #endif
1519 /* set symbol to STB_LOCAL and resolve. The point is to not export it as
1520 a dynamic symbol to allow so's to have one each with a different value. */
1521 static void set_local_sym(TCCState *s1, const char *name, Section *s, int offset)
1523 int c = find_elf_sym(s1->symtab, name);
1524 if (c) {
1525 ElfW(Sym) *esym = (ElfW(Sym)*)s1->symtab->data + c;
1526 esym->st_info = ELFW(ST_INFO)(STB_LOCAL, STT_NOTYPE);
1527 esym->st_value = offset;
1528 esym->st_shndx = s->sh_num;
1532 /* avoid generating debug/test_coverage code for stub functions */
1533 static void tcc_compile_string_no_debug(TCCState *s, const char *str)
1535 int save_do_debug = s->do_debug;
1536 int save_test_coverage = s->test_coverage;
1538 s->do_debug = 0;
1539 s->test_coverage = 0;
1540 tcc_compile_string(s, str);
1541 s->do_debug = save_do_debug;
1542 s->test_coverage = save_test_coverage;
1545 #ifdef CONFIG_TCC_BACKTRACE
1546 static void put_ptr(TCCState *s1, Section *s, int offs)
1548 int c;
1549 c = set_global_sym(s1, NULL, s, offs);
1550 s = data_section;
1551 put_elf_reloc (s1->symtab, s, s->data_offset, R_DATA_PTR, c);
1552 section_ptr_add(s, PTR_SIZE);
1555 ST_FUNC void tcc_add_btstub(TCCState *s1)
1557 Section *s;
1558 int n, o;
1559 CString cstr;
1561 s = data_section;
1562 /* Align to PTR_SIZE */
1563 section_ptr_add(s, -s->data_offset & (PTR_SIZE - 1));
1564 o = s->data_offset;
1565 /* create (part of) a struct rt_context (see tccrun.c) */
1566 if (s1->dwarf) {
1567 put_ptr(s1, dwarf_line_section, 0);
1568 put_ptr(s1, dwarf_line_section, -1);
1569 if (s1->dwarf >= 5)
1570 put_ptr(s1, dwarf_line_str_section, 0);
1571 else
1572 put_ptr(s1, dwarf_str_section, 0);
1574 else
1576 put_ptr(s1, stab_section, 0);
1577 put_ptr(s1, stab_section, -1);
1578 put_ptr(s1, stab_section->link, 0);
1580 *(addr_t *)section_ptr_add(s, PTR_SIZE) = s1->dwarf;
1581 /* skip esym_start/esym_end/elf_str (not loaded) */
1582 section_ptr_add(s, 3 * PTR_SIZE);
1583 /* prog_base : local nameless symbol with offset 0 at SHN_ABS */
1584 put_ptr(s1, NULL, 0);
1585 #if defined TCC_TARGET_MACHO
1586 /* adjust for __PAGEZERO */
1587 if (s1->dwarf == 0 && s1->output_type == TCC_OUTPUT_EXE)
1588 write64le(data_section->data + data_section->data_offset - PTR_SIZE,
1589 (uint64_t)1 << 32);
1590 #endif
1591 n = 2 * PTR_SIZE;
1592 #ifdef CONFIG_TCC_BCHECK
1593 if (s1->do_bounds_check) {
1594 put_ptr(s1, bounds_section, 0);
1595 n -= PTR_SIZE;
1597 #endif
1598 section_ptr_add(s, n);
1599 cstr_new(&cstr);
1600 cstr_printf(&cstr,
1601 "extern void __bt_init(),__bt_exit(),__bt_init_dll();"
1602 "static void *__rt_info[];"
1603 "__attribute__((constructor)) static void __bt_init_rt(){");
1604 #ifdef TCC_TARGET_PE
1605 if (s1->output_type == TCC_OUTPUT_DLL)
1606 #ifdef CONFIG_TCC_BCHECK
1607 cstr_printf(&cstr, "__bt_init_dll(%d);", s1->do_bounds_check);
1608 #else
1609 cstr_printf(&cstr, "__bt_init_dll(0);");
1610 #endif
1611 #endif
1612 cstr_printf(&cstr, "__bt_init(__rt_info,%d);}",
1613 s1->output_type == TCC_OUTPUT_DLL ? 0 : s1->rt_num_callers + 1);
1614 /* In case dlcose is called by application */
1615 cstr_printf(&cstr,
1616 "__attribute__((destructor)) static void __bt_exit_rt(){"
1617 "__bt_exit(__rt_info);}");
1618 tcc_compile_string_no_debug(s1, cstr.data);
1619 cstr_free(&cstr);
1620 set_local_sym(s1, &"___rt_info"[!s1->leading_underscore], s, o);
1622 #endif /* def CONFIG_TCC_BACKTRACE */
1624 static void tcc_tcov_add_file(TCCState *s1, const char *filename)
1626 CString cstr;
1627 void *ptr;
1628 char wd[1024];
1630 if (tcov_section == NULL)
1631 return;
1632 section_ptr_add(tcov_section, 1);
1633 write32le (tcov_section->data, tcov_section->data_offset);
1635 cstr_new (&cstr);
1636 if (filename[0] == '/')
1637 cstr_printf (&cstr, "%s.tcov", filename);
1638 else {
1639 getcwd (wd, sizeof(wd));
1640 cstr_printf (&cstr, "%s/%s.tcov", wd, filename);
1642 ptr = section_ptr_add(tcov_section, cstr.size + 1);
1643 strcpy((char *)ptr, cstr.data);
1644 unlink((char *)ptr);
1645 #ifdef _WIN32
1646 normalize_slashes((char *)ptr);
1647 #endif
1648 cstr_free (&cstr);
1650 cstr_new(&cstr);
1651 cstr_printf(&cstr,
1652 "extern char *__tcov_data[];"
1653 "extern void __store_test_coverage ();"
1654 "__attribute__((destructor)) static void __tcov_exit() {"
1655 "__store_test_coverage(__tcov_data);"
1656 "}");
1657 tcc_compile_string_no_debug(s1, cstr.data);
1658 cstr_free(&cstr);
1659 set_local_sym(s1, &"___tcov_data"[!s1->leading_underscore], tcov_section, 0);
1662 #ifndef TCC_TARGET_PE
1663 /* add tcc runtime libraries */
1664 ST_FUNC void tcc_add_runtime(TCCState *s1)
1666 s1->filetype = 0;
1668 #ifdef CONFIG_TCC_BCHECK
1669 tcc_add_bcheck(s1);
1670 #endif
1671 tcc_add_pragma_libs(s1);
1673 /* add libc */
1674 if (!s1->nostdlib) {
1675 int lpthread = s1->option_pthread;
1677 #ifdef CONFIG_TCC_BCHECK
1678 if (s1->do_bounds_check && s1->output_type != TCC_OUTPUT_DLL) {
1679 tcc_add_support(s1, "bcheck.o");
1680 # if !(TARGETOS_OpenBSD || TARGETOS_NetBSD)
1681 tcc_add_library_err(s1, "dl");
1682 # endif
1683 lpthread = 1;
1685 #endif
1686 #ifdef CONFIG_TCC_BACKTRACE
1687 if (s1->do_backtrace) {
1688 if (s1->output_type & TCC_OUTPUT_EXE)
1689 tcc_add_support(s1, "bt-exe.o");
1690 if (s1->output_type != TCC_OUTPUT_DLL)
1691 tcc_add_support(s1, "bt-log.o");
1692 if (s1->output_type != TCC_OUTPUT_MEMORY)
1693 tcc_add_btstub(s1);
1695 #endif
1696 if (lpthread)
1697 tcc_add_library_err(s1, "pthread");
1698 tcc_add_library_err(s1, "c");
1699 #ifdef TCC_LIBGCC
1700 if (!s1->static_link) {
1701 if (TCC_LIBGCC[0] == '/')
1702 tcc_add_file(s1, TCC_LIBGCC);
1703 else
1704 tcc_add_dll(s1, TCC_LIBGCC, 0);
1706 #endif
1707 #if defined TCC_TARGET_ARM && TARGETOS_FreeBSD
1708 tcc_add_library_err(s1, "gcc_s"); // unwind code
1709 #endif
1710 if (TCC_LIBTCC1[0])
1711 tcc_add_support(s1, TCC_LIBTCC1);
1713 /* add crt end if not memory output */
1714 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1715 #if defined TCC_TARGET_MACHO
1716 /* nothing to do */
1717 #elif TARGETOS_FreeBSD || TARGETOS_NetBSD
1718 if (s1->output_type & TCC_OUTPUT_DYN)
1719 tcc_add_crt(s1, "crtendS.o");
1720 else
1721 tcc_add_crt(s1, "crtend.o");
1722 tcc_add_crt(s1, "crtn.o");
1723 #elif TARGETOS_OpenBSD
1724 if (s1->output_type == TCC_OUTPUT_DLL)
1725 tcc_add_crt(s1, "crtendS.o");
1726 else
1727 tcc_add_crt(s1, "crtend.o");
1728 #elif TARGETOS_ANDROID
1729 if (s1->output_type == TCC_OUTPUT_DLL)
1730 tcc_add_crt(s1, "crtend_so.o");
1731 else
1732 tcc_add_crt(s1, "crtend_android.o");
1733 #else
1734 tcc_add_crt(s1, "crtn.o");
1735 #endif
1739 #endif /* ndef TCC_TARGET_PE */
1741 /* add various standard linker symbols (must be done after the
1742 sections are filled (for example after allocating common
1743 symbols)) */
1744 static void tcc_add_linker_symbols(TCCState *s1)
1746 char buf[1024];
1747 int i;
1748 Section *s;
1750 set_global_sym(s1, "_etext", text_section, -1);
1751 set_global_sym(s1, "_edata", data_section, -1);
1752 set_global_sym(s1, "_end", bss_section, -1);
1753 #if TARGETOS_OpenBSD
1754 set_global_sym(s1, "__executable_start", NULL, ELF_START_ADDR);
1755 #endif
1756 #ifdef TCC_TARGET_RISCV64
1757 /* XXX should be .sdata+0x800, not .data+0x800 */
1758 set_global_sym(s1, "__global_pointer$", data_section, 0x800);
1759 #endif
1760 /* horrible new standard ldscript defines */
1761 add_init_array_defines(s1, ".preinit_array");
1762 add_init_array_defines(s1, ".init_array");
1763 add_init_array_defines(s1, ".fini_array");
1764 /* add start and stop symbols for sections whose name can be
1765 expressed in C */
1766 for(i = 1; i < s1->nb_sections; i++) {
1767 s = s1->sections[i];
1768 if ((s->sh_flags & SHF_ALLOC)
1769 && (s->sh_type == SHT_PROGBITS
1770 || s->sh_type == SHT_STRTAB)) {
1771 const char *p;
1772 /* check if section name can be expressed in C */
1773 p = s->name;
1774 for(;;) {
1775 int c = *p;
1776 if (!c)
1777 break;
1778 if (!isid(c) && !isnum(c))
1779 goto next_sec;
1780 p++;
1782 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1783 set_global_sym(s1, buf, s, 0);
1784 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1785 set_global_sym(s1, buf, s, -1);
1787 next_sec: ;
1791 ST_FUNC void resolve_common_syms(TCCState *s1)
1793 ElfW(Sym) *sym;
1795 /* Allocate common symbols in BSS. */
1796 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1797 if (sym->st_shndx == SHN_COMMON) {
1798 /* symbol alignment is in st_value for SHN_COMMONs */
1799 sym->st_value = section_add(bss_section, sym->st_size,
1800 sym->st_value);
1801 sym->st_shndx = bss_section->sh_num;
1805 /* Now assign linker provided symbols their value. */
1806 tcc_add_linker_symbols(s1);
1809 #ifndef ELF_OBJ_ONLY
1810 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1812 int sym_index = ELFW(R_SYM) (rel->r_info);
1813 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1814 struct sym_attr *attr = get_sym_attr(s1, sym_index, 0);
1815 unsigned offset = attr->got_offset;
1817 if (0 == offset)
1818 return;
1819 section_reserve(s1->got, offset + PTR_SIZE);
1820 #if PTR_SIZE == 8
1821 write64le(s1->got->data + offset, sym->st_value);
1822 #else
1823 write32le(s1->got->data + offset, sym->st_value);
1824 #endif
1827 /* Perform relocation to GOT or PLT entries */
1828 ST_FUNC void fill_got(TCCState *s1)
1830 Section *s;
1831 ElfW_Rel *rel;
1832 int i;
1834 for(i = 1; i < s1->nb_sections; i++) {
1835 s = s1->sections[i];
1836 if (s->sh_type != SHT_RELX)
1837 continue;
1838 /* no need to handle got relocations */
1839 if (s->link != symtab_section)
1840 continue;
1841 for_each_elem(s, 0, rel, ElfW_Rel) {
1842 switch (ELFW(R_TYPE) (rel->r_info)) {
1843 case R_X86_64_GOT32:
1844 case R_X86_64_GOTPCREL:
1845 case R_X86_64_GOTPCRELX:
1846 case R_X86_64_REX_GOTPCRELX:
1847 case R_X86_64_PLT32:
1848 fill_got_entry(s1, rel);
1849 break;
1855 /* See put_got_entry for a description. This is the second stage
1856 where GOT references to local defined symbols are rewritten. */
1857 static void fill_local_got_entries(TCCState *s1)
1859 ElfW_Rel *rel;
1860 if (!s1->got->reloc)
1861 return;
1862 for_each_elem(s1->got->reloc, 0, rel, ElfW_Rel) {
1863 if (ELFW(R_TYPE)(rel->r_info) == R_RELATIVE) {
1864 int sym_index = ELFW(R_SYM) (rel->r_info);
1865 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1866 struct sym_attr *attr = get_sym_attr(s1, sym_index, 0);
1867 unsigned offset = attr->got_offset;
1868 if (offset != rel->r_offset - s1->got->sh_addr)
1869 tcc_error_noabort("fill_local_got_entries: huh?");
1870 rel->r_info = ELFW(R_INFO)(0, R_RELATIVE);
1871 #if SHT_RELX == SHT_RELA
1872 rel->r_addend = sym->st_value;
1873 #else
1874 /* All our REL architectures also happen to be 32bit LE. */
1875 write32le(s1->got->data + offset, sym->st_value);
1876 #endif
1881 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1882 in shared libraries */
1883 static void bind_exe_dynsyms(TCCState *s1)
1885 const char *name;
1886 int sym_index, index;
1887 ElfW(Sym) *sym, *esym;
1888 int type;
1890 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1891 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1892 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1893 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1894 if (sym->st_shndx == SHN_UNDEF) {
1895 name = (char *) symtab_section->link->data + sym->st_name;
1896 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1897 if (sym_index) {
1898 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1899 type = ELFW(ST_TYPE)(esym->st_info);
1900 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1901 /* Indirect functions shall have STT_FUNC type in executable
1902 * dynsym section. Indeed, a dlsym call following a lazy
1903 * resolution would pick the symbol value from the
1904 * executable dynsym entry which would contain the address
1905 * of the function wanted by the caller of dlsym instead of
1906 * the address of the function that would return that
1907 * address */
1908 int dynindex
1909 = put_elf_sym(s1->dynsym, 0, esym->st_size,
1910 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC), 0, 0,
1911 name);
1912 int index = sym - (ElfW(Sym) *) symtab_section->data;
1913 get_sym_attr(s1, index, 1)->dyn_index = dynindex;
1914 } else if (type == STT_OBJECT) {
1915 unsigned long offset;
1916 ElfW(Sym) *dynsym;
1917 offset = bss_section->data_offset;
1918 /* XXX: which alignment ? */
1919 offset = (offset + 16 - 1) & -16;
1920 set_elf_sym (s1->symtab, offset, esym->st_size,
1921 esym->st_info, 0, bss_section->sh_num, name);
1922 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1923 esym->st_info, 0, bss_section->sh_num,
1924 name);
1926 /* Ensure R_COPY works for weak symbol aliases */
1927 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1928 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1929 if ((dynsym->st_value == esym->st_value)
1930 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1931 char *dynname = (char *) s1->dynsymtab_section->link->data
1932 + dynsym->st_name;
1933 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1934 dynsym->st_info, 0,
1935 bss_section->sh_num, dynname);
1936 break;
1941 put_elf_reloc(s1->dynsym, bss_section,
1942 offset, R_COPY, index);
1943 offset += esym->st_size;
1944 bss_section->data_offset = offset;
1946 } else {
1947 /* STB_WEAK undefined symbols are accepted */
1948 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1949 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1950 !strcmp(name, "_fp_hw")) {
1951 } else {
1952 tcc_error_noabort("undefined symbol '%s'", name);
1959 /* Bind symbols of libraries: export all non local symbols of executable that
1960 are referenced by shared libraries. The reason is that the dynamic loader
1961 search symbol first in executable and then in libraries. Therefore a
1962 reference to a symbol already defined by a library can still be resolved by
1963 a symbol in the executable. With -rdynamic, export all defined symbols */
1964 static void bind_libs_dynsyms(TCCState *s1)
1966 const char *name;
1967 int dynsym_index;
1968 ElfW(Sym) *sym, *esym;
1970 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1971 name = (char *)symtab_section->link->data + sym->st_name;
1972 dynsym_index = find_elf_sym(s1->dynsymtab_section, name);
1973 if (sym->st_shndx != SHN_UNDEF
1974 && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1975 if (dynsym_index || s1->rdynamic)
1976 set_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1977 sym->st_info, 0, sym->st_shndx, name);
1978 } else if (dynsym_index) {
1979 esym = (ElfW(Sym) *)s1->dynsymtab_section->data + dynsym_index;
1980 if (esym->st_shndx == SHN_UNDEF) {
1981 /* weak symbols can stay undefined */
1982 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1983 tcc_warning("undefined dynamic symbol '%s'", name);
1989 /* Export all non local symbols. This is used by shared libraries so that the
1990 non local symbols they define can resolve a reference in another shared
1991 library or in the executable. Correspondingly, it allows undefined local
1992 symbols to be resolved by other shared libraries or by the executable. */
1993 static void export_global_syms(TCCState *s1)
1995 int dynindex, index;
1996 const char *name;
1997 ElfW(Sym) *sym;
1998 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1999 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2000 name = (char *) symtab_section->link->data + sym->st_name;
2001 dynindex = set_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
2002 sym->st_info, 0, sym->st_shndx, name);
2003 index = sym - (ElfW(Sym) *) symtab_section->data;
2004 get_sym_attr(s1, index, 1)->dyn_index = dynindex;
2009 /* decide if an unallocated section should be output. */
2010 static int set_sec_sizes(TCCState *s1)
2012 int i;
2013 Section *s;
2014 int textrel = 0;
2015 int file_type = s1->output_type;
2017 /* Allocate strings for section names */
2018 for(i = 1; i < s1->nb_sections; i++) {
2019 s = s1->sections[i];
2020 if (s->sh_type == SHT_RELX && !(s->sh_flags & SHF_ALLOC)) {
2021 /* when generating a DLL, we include relocations but
2022 we may patch them */
2023 if ((file_type & TCC_OUTPUT_DYN)
2024 && (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)) {
2025 int count = prepare_dynamic_rel(s1, s);
2026 if (count) {
2027 /* allocate the section */
2028 s->sh_flags |= SHF_ALLOC;
2029 s->sh_size = count * sizeof(ElfW_Rel);
2030 if (!(s1->sections[s->sh_info]->sh_flags & SHF_WRITE))
2031 textrel += count;
2034 } else if ((s->sh_flags & SHF_ALLOC)
2035 #ifdef TCC_TARGET_ARM
2036 || s->sh_type == SHT_ARM_ATTRIBUTES
2037 #endif
2038 || s1->do_debug) {
2039 s->sh_size = s->data_offset;
2042 #ifdef TCC_TARGET_ARM
2043 /* XXX: Suppress stack unwinding section. */
2044 if (s->sh_type == SHT_ARM_EXIDX) {
2045 s->sh_flags = 0;
2046 s->sh_size = 0;
2048 #endif
2051 return textrel;
2054 /* various data used under elf_output_file() */
2055 struct dyn_inf {
2056 Section *dynamic;
2057 Section *dynstr;
2058 struct {
2059 /* Info to be copied in dynamic section */
2060 unsigned long data_offset;
2061 addr_t rel_addr;
2062 addr_t rel_size;
2065 ElfW(Phdr) *phdr;
2066 int phnum;
2067 Section *interp;
2068 Section *note;
2069 Section *gnu_hash;
2071 /* read only segment mapping for GNU_RELRO */
2072 Section _roinf, *roinf;
2075 /* Decide the layout of sections loaded in memory. This must be done before
2076 program headers are filled since they contain info about the layout.
2077 We do the following ordering: interp, symbol tables, relocations, progbits,
2078 nobits */
2079 static int sort_sections(TCCState *s1, int *sec_order, Section *interp)
2081 Section *s;
2082 int i, j, k, f, f0, n;
2083 int nb_sections = s1->nb_sections;
2084 int *sec_cls = sec_order + nb_sections;
2086 for (i = 1; i < nb_sections; i++) {
2087 s = s1->sections[i];
2088 if (s->sh_flags & SHF_ALLOC) {
2089 j = 0x100;
2090 if (s->sh_flags & SHF_WRITE)
2091 j = 0x200;
2092 if (s->sh_flags & SHF_TLS)
2093 j += 0x200;
2094 } else if (s->sh_name) {
2095 j = 0x700;
2096 } else {
2097 j = 0x900; /* no sh_name: won't go to file */
2099 if (s->sh_type == SHT_SYMTAB || s->sh_type == SHT_DYNSYM) {
2100 k = 0x10;
2101 } else if (s->sh_type == SHT_STRTAB && strcmp(s->name, ".stabstr")) {
2102 k = 0x11;
2103 if (i == nb_sections - 1) /* ".shstrtab" assumed to remain last */
2104 k = 0xff;
2105 } else if (s->sh_type == SHT_HASH || s->sh_type == SHT_GNU_HASH) {
2106 k = 0x12;
2107 } else if (s->sh_type == SHT_RELX) {
2108 k = 0x20;
2109 if (s1->plt && s == s1->plt->reloc)
2110 k = 0x21;
2111 } else if (s->sh_type == SHT_PREINIT_ARRAY) {
2112 k = 0x41;
2113 } else if (s->sh_type == SHT_INIT_ARRAY) {
2114 k = 0x42;
2115 } else if (s->sh_type == SHT_FINI_ARRAY) {
2116 k = 0x43;
2117 #ifdef CONFIG_TCC_BCHECK
2118 } else if (s == bounds_section || s == lbounds_section) {
2119 k = 0x44;
2120 #endif
2121 } else if (s == rodata_section || 0 == strcmp(s->name, ".data.rel.ro")) {
2122 k = 0x45;
2123 } else if (s->sh_type == SHT_DYNAMIC) {
2124 k = 0x46;
2125 } else if (s == s1->got) {
2126 k = 0x47; /* .got as RELRO needs BIND_NOW in DT_FLAGS */
2127 } else {
2128 k = 0x50;
2129 if (s->sh_type == SHT_NOTE)
2130 k = 0x60;
2131 if (s->sh_flags & SHF_EXECINSTR)
2132 k = 0x70;
2133 if (s->sh_type == SHT_NOBITS)
2134 k = 0x80;
2135 if (s == interp)
2136 k = 0x00;
2138 k += j;
2140 for (n = i; n > 1 && k < (f = sec_cls[n - 1]); --n)
2141 sec_cls[n] = f, sec_order[n] = sec_order[n - 1];
2142 sec_cls[n] = k, sec_order[n] = i;
2144 sec_order[0] = 0;
2146 /* count PT_LOAD headers needed */
2147 n = f0 = 0;
2148 for (i = 1; i < nb_sections; i++) {
2149 s = s1->sections[sec_order[i]];
2150 k = sec_cls[i];
2151 f = 0;
2152 if (k < 0x700) {
2153 f = s->sh_flags & (SHF_ALLOC|SHF_WRITE|SHF_EXECINSTR|SHF_TLS);
2154 #if TARGETOS_NetBSD
2155 /* NetBSD only supports 2 PT_LOAD sections.
2156 See: https://blog.netbsd.org/tnf/entry/the_first_report_on_lld */
2157 if ((f & SHF_WRITE) == 0) f |= SHF_EXECINSTR;
2158 #else
2159 if ((k & 0xfff0) == 0x240) /* RELRO sections */
2160 f |= 1<<4;
2161 #endif
2162 if (f != f0) /* start new header when flags changed or relro */
2163 f0 = f, ++n, f |= 1<<8;
2165 sec_cls[i] = f;
2166 //printf("ph %d sec %02d : %3X %3X %8.2X %04X %s\n", !!f * n, i, f, k, s->sh_type, s->sh_size, s->name);
2168 return n;
2171 static ElfW(Phdr) *fill_phdr(ElfW(Phdr) *ph, int type, Section *s)
2173 if (s) {
2174 ph->p_offset = s->sh_offset;
2175 ph->p_vaddr = s->sh_addr;
2176 ph->p_filesz = s->sh_size;
2177 ph->p_align = s->sh_addralign;
2179 ph->p_type = type;
2180 ph->p_flags = PF_R;
2181 ph->p_paddr = ph->p_vaddr;
2182 ph->p_memsz = ph->p_filesz;
2183 return ph;
2186 /* Assign sections to segments and decide how are sections laid out when loaded
2187 in memory. This function also fills corresponding program headers. */
2188 static int layout_sections(TCCState *s1, int *sec_order, struct dyn_inf *d)
2190 Section *s;
2191 addr_t addr, tmp, align, s_align, base;
2192 ElfW(Phdr) *ph = NULL;
2193 int i, f, n, phnum, phfill;
2194 int file_offset;
2196 /* compute number of program headers */
2197 phnum = sort_sections(s1, sec_order, d->interp);
2198 phfill = 0; /* set to 1 to have dll's with a PT_PHDR */
2199 if (d->interp)
2200 phfill = 2;
2201 phnum += phfill;
2202 if (d->note)
2203 ++phnum;
2204 if (d->dynamic)
2205 ++phnum;
2206 if (d->roinf)
2207 ++phnum;
2208 d->phnum = phnum;
2209 d->phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2211 file_offset = 0;
2212 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2213 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2215 s_align = ELF_PAGE_SIZE;
2216 if (s1->section_align)
2217 s_align = s1->section_align;
2219 addr = ELF_START_ADDR;
2220 if (s1->output_type & TCC_OUTPUT_DYN)
2221 addr = 0;
2223 if (s1->has_text_addr) {
2224 addr = s1->text_addr;
2225 if (0) {
2226 int a_offset, p_offset;
2227 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2228 ELF_PAGE_SIZE */
2229 a_offset = (int) (addr & (s_align - 1));
2230 p_offset = file_offset & (s_align - 1);
2231 if (a_offset < p_offset)
2232 a_offset += s_align;
2233 file_offset += (a_offset - p_offset);
2236 base = addr;
2237 /* compute address after headers */
2238 addr = addr + (file_offset & (s_align - 1));
2240 n = 0;
2241 for(i = 1; i < s1->nb_sections; i++) {
2242 s = s1->sections[sec_order[i]];
2243 f = sec_order[i + s1->nb_sections];
2244 align = s->sh_addralign - 1;
2246 if (f == 0) { /* no alloc */
2247 file_offset = (file_offset + align) & ~align;
2248 s->sh_offset = file_offset;
2249 if (s->sh_type != SHT_NOBITS)
2250 file_offset += s->sh_size;
2251 continue;
2254 if ((f & 1<<8) && n) {
2255 /* different rwx section flags */
2256 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2257 /* if in the middle of a page, w e duplicate the page in
2258 memory so that one copy is RX and the other is RW */
2259 if ((addr & (s_align - 1)) != 0)
2260 addr += s_align;
2261 } else {
2262 align = s_align - 1;
2266 tmp = addr;
2267 addr = (addr + align) & ~align;
2268 file_offset += (int)(addr - tmp);
2269 s->sh_offset = file_offset;
2270 s->sh_addr = addr;
2272 if (f & 1<<8) {
2273 /* set new program header */
2274 ph = &d->phdr[phfill + n];
2275 ph->p_type = PT_LOAD;
2276 ph->p_align = s_align;
2277 ph->p_flags = PF_R;
2278 if (f & SHF_WRITE)
2279 ph->p_flags |= PF_W;
2280 if (f & SHF_EXECINSTR)
2281 ph->p_flags |= PF_X;
2282 if (f & SHF_TLS) {
2283 ph->p_type = PT_TLS;
2284 ph->p_align = align + 1;
2287 ph->p_offset = file_offset;
2288 ph->p_vaddr = addr;
2289 if (n == 0) {
2290 /* Make the first PT_LOAD segment include the program
2291 headers itself (and the ELF header as well), it'll
2292 come out with same memory use but will make various
2293 tools like binutils strip work better. */
2294 ph->p_offset = 0;
2295 ph->p_vaddr = base;
2297 ph->p_paddr = ph->p_vaddr;
2298 ++n;
2301 if (f & 1<<4) {
2302 Section *roinf = &d->_roinf;
2303 if (roinf->sh_size == 0) {
2304 roinf->sh_offset = s->sh_offset;
2305 roinf->sh_addr = s->sh_addr;
2306 roinf->sh_addralign = 1;
2308 roinf->sh_size = (addr - roinf->sh_addr) + s->sh_size;
2311 addr += s->sh_size;
2312 if (s->sh_type != SHT_NOBITS)
2313 file_offset += s->sh_size;
2315 ph->p_filesz = file_offset - ph->p_offset;
2316 ph->p_memsz = addr - ph->p_vaddr;
2319 /* Fill other headers */
2320 if (d->note)
2321 fill_phdr(++ph, PT_NOTE, d->note);
2322 if (d->dynamic)
2323 fill_phdr(++ph, PT_DYNAMIC, d->dynamic)->p_flags |= PF_W;
2324 if (d->roinf)
2325 fill_phdr(++ph, PT_GNU_RELRO, d->roinf)->p_flags |= PF_W;
2326 if (d->interp)
2327 fill_phdr(&d->phdr[1], PT_INTERP, d->interp);
2328 if (phfill) {
2329 ph = &d->phdr[0];
2330 ph->p_offset = sizeof(ElfW(Ehdr));
2331 ph->p_vaddr = base + ph->p_offset;
2332 ph->p_filesz = phnum * sizeof(ElfW(Phdr));
2333 ph->p_align = 4;
2334 fill_phdr(ph, PT_PHDR, NULL);
2336 return file_offset;
2339 /* put dynamic tag */
2340 static void put_dt(Section *dynamic, int dt, addr_t val)
2342 ElfW(Dyn) *dyn;
2343 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
2344 dyn->d_tag = dt;
2345 dyn->d_un.d_val = val;
2348 /* Fill the dynamic section with tags describing the address and size of
2349 sections */
2350 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2352 Section *dynamic = dyninf->dynamic;
2353 Section *s;
2355 /* put dynamic section entries */
2356 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2357 put_dt(dynamic, DT_GNU_HASH, dyninf->gnu_hash->sh_addr);
2358 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2359 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2360 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2361 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2362 #if PTR_SIZE == 8
2363 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2364 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2365 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2366 if (s1->plt && s1->plt->reloc) {
2367 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2368 put_dt(dynamic, DT_PLTRELSZ, s1->plt->reloc->data_offset);
2369 put_dt(dynamic, DT_JMPREL, s1->plt->reloc->sh_addr);
2370 put_dt(dynamic, DT_PLTREL, DT_RELA);
2372 put_dt(dynamic, DT_RELACOUNT, 0);
2373 #else
2374 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2375 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2376 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2377 if (s1->plt && s1->plt->reloc) {
2378 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2379 put_dt(dynamic, DT_PLTRELSZ, s1->plt->reloc->data_offset);
2380 put_dt(dynamic, DT_JMPREL, s1->plt->reloc->sh_addr);
2381 put_dt(dynamic, DT_PLTREL, DT_REL);
2383 put_dt(dynamic, DT_RELCOUNT, 0);
2384 #endif
2385 if (versym_section && verneed_section) {
2386 /* The dynamic linker can not handle VERSYM without VERNEED */
2387 put_dt(dynamic, DT_VERSYM, versym_section->sh_addr);
2388 put_dt(dynamic, DT_VERNEED, verneed_section->sh_addr);
2389 put_dt(dynamic, DT_VERNEEDNUM, dt_verneednum);
2391 s = have_section(s1, ".preinit_array");
2392 if (s && s->data_offset) {
2393 put_dt(dynamic, DT_PREINIT_ARRAY, s->sh_addr);
2394 put_dt(dynamic, DT_PREINIT_ARRAYSZ, s->data_offset);
2396 s = have_section(s1, ".init_array");
2397 if (s && s->data_offset) {
2398 put_dt(dynamic, DT_INIT_ARRAY, s->sh_addr);
2399 put_dt(dynamic, DT_INIT_ARRAYSZ, s->data_offset);
2401 s = have_section(s1, ".fini_array");
2402 if (s && s->data_offset) {
2403 put_dt(dynamic, DT_FINI_ARRAY, s->sh_addr);
2404 put_dt(dynamic, DT_FINI_ARRAYSZ, s->data_offset);
2406 s = have_section(s1, ".init");
2407 if (s && s->data_offset) {
2408 put_dt(dynamic, DT_INIT, s->sh_addr);
2410 s = have_section(s1, ".fini");
2411 if (s && s->data_offset) {
2412 put_dt(dynamic, DT_FINI, s->sh_addr);
2414 if (s1->do_debug)
2415 put_dt(dynamic, DT_DEBUG, 0);
2416 put_dt(dynamic, DT_NULL, 0);
2419 /* Remove gaps between RELX sections.
2420 These gaps are a result of final_sections_reloc. Here some relocs are removed.
2421 The gaps are then filled with 0 in tcc_output_elf. The 0 is intepreted as
2422 R_...NONE reloc. This does work on most targets but on OpenBSD/arm64 this
2423 is illegal. OpenBSD/arm64 does not support R_...NONE reloc. */
2424 static void update_reloc_sections(TCCState *s1, struct dyn_inf *dyninf)
2426 int i;
2427 unsigned long file_offset = 0;
2428 Section *s;
2429 Section *relocplt = s1->plt ? s1->plt->reloc : NULL;
2431 /* dynamic relocation table information, for .dynamic section */
2432 dyninf->rel_addr = dyninf->rel_size = 0;
2434 for(i = 1; i < s1->nb_sections; i++) {
2435 s = s1->sections[i];
2436 if (s->sh_type == SHT_RELX && s != relocplt) {
2437 if (dyninf->rel_size == 0) {
2438 dyninf->rel_addr = s->sh_addr;
2439 file_offset = s->sh_offset;
2441 else {
2442 s->sh_addr = dyninf->rel_addr + dyninf->rel_size;
2443 s->sh_offset = file_offset + dyninf->rel_size;
2445 dyninf->rel_size += s->sh_size;
2450 static int tidy_section_headers(TCCState *s1, int *sec_order);
2451 #endif /* ndef ELF_OBJ_ONLY */
2453 /* Create an ELF file on disk.
2454 This function handle ELF specific layout requirements */
2455 static int tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2456 int file_offset, int *sec_order)
2458 int i, shnum, offset, size, file_type;
2459 Section *s;
2460 ElfW(Ehdr) ehdr;
2461 ElfW(Shdr) shdr, *sh;
2463 file_type = s1->output_type;
2464 shnum = s1->nb_sections;
2466 memset(&ehdr, 0, sizeof(ehdr));
2468 if (phnum > 0) {
2469 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2470 ehdr.e_phnum = phnum;
2471 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2472 #ifndef ELF_OBJ_ONLY
2473 shnum = tidy_section_headers(s1, sec_order);
2474 #endif
2477 /* align to 4 */
2478 file_offset = (file_offset + 3) & -4;
2480 /* fill header */
2481 ehdr.e_ident[0] = ELFMAG0;
2482 ehdr.e_ident[1] = ELFMAG1;
2483 ehdr.e_ident[2] = ELFMAG2;
2484 ehdr.e_ident[3] = ELFMAG3;
2485 ehdr.e_ident[4] = ELFCLASSW;
2486 ehdr.e_ident[5] = ELFDATA2LSB;
2487 ehdr.e_ident[6] = EV_CURRENT;
2489 #if TARGETOS_FreeBSD || TARGETOS_FreeBSD_kernel
2490 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2491 #elif defined TCC_TARGET_ARM && defined TCC_ARM_EABI
2492 ehdr.e_flags = EF_ARM_EABI_VER5;
2493 ehdr.e_flags |= s1->float_abi == ARM_HARD_FLOAT
2494 ? EF_ARM_VFP_FLOAT : EF_ARM_SOFT_FLOAT;
2495 #elif defined TCC_TARGET_ARM
2496 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2497 #elif defined TCC_TARGET_RISCV64
2498 ehdr.e_flags = EF_RISCV_FLOAT_ABI_DOUBLE;
2499 #endif
2501 if (file_type == TCC_OUTPUT_OBJ) {
2502 ehdr.e_type = ET_REL;
2503 } else {
2504 if (file_type & TCC_OUTPUT_DYN)
2505 ehdr.e_type = ET_DYN;
2506 else
2507 ehdr.e_type = ET_EXEC;
2508 if (s1->elf_entryname)
2509 ehdr.e_entry = get_sym_addr(s1, s1->elf_entryname, 1, 0);
2510 else
2511 ehdr.e_entry = get_sym_addr(s1, "_start", !!(file_type & TCC_OUTPUT_EXE), 0);
2512 if (ehdr.e_entry == (addr_t)-1)
2513 ehdr.e_entry = text_section->sh_addr;
2514 if (s1->nb_errors)
2515 return -1;
2518 ehdr.e_machine = EM_TCC_TARGET;
2519 ehdr.e_version = EV_CURRENT;
2520 ehdr.e_shoff = file_offset;
2521 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2522 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2523 ehdr.e_shnum = shnum;
2524 ehdr.e_shstrndx = shnum - 1;
2526 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2527 if (phdr)
2528 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2529 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2531 sort_syms(s1, symtab_section);
2533 for(i = 1; i < shnum; i++) {
2534 s = s1->sections[sec_order ? sec_order[i] : i];
2535 if (s->sh_type != SHT_NOBITS) {
2536 while (offset < s->sh_offset) {
2537 fputc(0, f);
2538 offset++;
2540 size = s->sh_size;
2541 if (size)
2542 fwrite(s->data, 1, size, f);
2543 offset += size;
2547 /* output section headers */
2548 while (offset < ehdr.e_shoff) {
2549 fputc(0, f);
2550 offset++;
2553 for(i = 0; i < shnum; i++) {
2554 sh = &shdr;
2555 memset(sh, 0, sizeof(ElfW(Shdr)));
2556 s = s1->sections[i];
2557 if (s) {
2558 sh->sh_name = s->sh_name;
2559 sh->sh_type = s->sh_type;
2560 sh->sh_flags = s->sh_flags;
2561 sh->sh_entsize = s->sh_entsize;
2562 sh->sh_info = s->sh_info;
2563 if (s->link)
2564 sh->sh_link = s->link->sh_num;
2565 sh->sh_addralign = s->sh_addralign;
2566 sh->sh_addr = s->sh_addr;
2567 sh->sh_offset = s->sh_offset;
2568 sh->sh_size = s->sh_size;
2570 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2572 return 0;
2575 static int tcc_output_binary(TCCState *s1, FILE *f,
2576 const int *sec_order)
2578 Section *s;
2579 int i, offset, size;
2581 offset = 0;
2582 for(i=1;i<s1->nb_sections;i++) {
2583 s = s1->sections[sec_order[i]];
2584 if (s->sh_type != SHT_NOBITS &&
2585 (s->sh_flags & SHF_ALLOC)) {
2586 while (offset < s->sh_offset) {
2587 fputc(0, f);
2588 offset++;
2590 size = s->sh_size;
2591 fwrite(s->data, 1, size, f);
2592 offset += size;
2595 return 0;
2598 /* Write an elf, coff or "binary" file */
2599 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2600 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2602 int fd, mode, file_type, ret;
2603 FILE *f;
2605 file_type = s1->output_type;
2606 if (file_type == TCC_OUTPUT_OBJ)
2607 mode = 0666;
2608 else
2609 mode = 0777;
2610 unlink(filename);
2611 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2612 if (fd < 0 || (f = fdopen(fd, "wb")) == NULL)
2613 return tcc_error_noabort("could not write '%s: %s'", filename, strerror(errno));
2614 if (s1->verbose)
2615 printf("<- %s\n", filename);
2616 #ifdef TCC_TARGET_COFF
2617 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2618 tcc_output_coff(s1, f);
2619 else
2620 #endif
2621 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2622 ret = tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2623 else
2624 ret = tcc_output_binary(s1, f, sec_order);
2625 fclose(f);
2627 return ret;
2630 #ifndef ELF_OBJ_ONLY
2631 /* Sort section headers by assigned sh_addr, remove sections
2632 that we aren't going to output. */
2633 static int tidy_section_headers(TCCState *s1, int *sec_order)
2635 int i, nnew, l, *backmap;
2636 Section **snew, *s;
2637 ElfW(Sym) *sym;
2639 snew = tcc_malloc(s1->nb_sections * sizeof(snew[0]));
2640 backmap = tcc_malloc(s1->nb_sections * sizeof(backmap[0]));
2641 for (i = 0, nnew = 0, l = s1->nb_sections; i < s1->nb_sections; i++) {
2642 s = s1->sections[sec_order[i]];
2643 if (!i || s->sh_name) {
2644 backmap[sec_order[i]] = nnew;
2645 snew[nnew] = s;
2646 ++nnew;
2647 } else {
2648 backmap[sec_order[i]] = 0;
2649 snew[--l] = s;
2652 for (i = 0; i < nnew; i++) {
2653 s = snew[i];
2654 if (s) {
2655 s->sh_num = i;
2656 if (s->sh_type == SHT_RELX)
2657 s->sh_info = backmap[s->sh_info];
2661 for_each_elem(symtab_section, 1, sym, ElfW(Sym))
2662 if (sym->st_shndx != SHN_UNDEF && sym->st_shndx < SHN_LORESERVE)
2663 sym->st_shndx = backmap[sym->st_shndx];
2664 if ( !s1->static_link ) {
2665 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym))
2666 if (sym->st_shndx != SHN_UNDEF && sym->st_shndx < SHN_LORESERVE)
2667 sym->st_shndx = backmap[sym->st_shndx];
2669 for (i = 0; i < s1->nb_sections; i++)
2670 sec_order[i] = i;
2671 tcc_free(s1->sections);
2672 s1->sections = snew;
2673 tcc_free(backmap);
2674 return nnew;
2677 #ifdef TCC_TARGET_ARM
2678 static void create_arm_attribute_section(TCCState *s1)
2680 // Needed for DLL support.
2681 static const unsigned char arm_attr[] = {
2682 0x41, // 'A'
2683 0x2c, 0x00, 0x00, 0x00, // size 0x2c
2684 'a', 'e', 'a', 'b', 'i', 0x00, // "aeabi"
2685 0x01, 0x22, 0x00, 0x00, 0x00, // 'File Attributes', size 0x22
2686 0x05, 0x36, 0x00, // 'CPU_name', "6"
2687 0x06, 0x06, // 'CPU_arch', 'v6'
2688 0x08, 0x01, // 'ARM_ISA_use', 'Yes'
2689 0x09, 0x01, // 'THUMB_ISA_use', 'Thumb-1'
2690 0x0a, 0x02, // 'FP_arch', 'VFPv2'
2691 0x12, 0x04, // 'ABI_PCS_wchar_t', 4
2692 0x14, 0x01, // 'ABI_FP_denormal', 'Needed'
2693 0x15, 0x01, // 'ABI_FP_exceptions', 'Needed'
2694 0x17, 0x03, // 'ABI_FP_number_model', 'IEEE 754'
2695 0x18, 0x01, // 'ABI_align_needed', '8-byte'
2696 0x19, 0x01, // 'ABI_align_preserved', '8-byte, except leaf SP'
2697 0x1a, 0x02, // 'ABI_enum_size', 'int'
2698 0x1c, 0x01, // 'ABI_VFP_args', 'VFP registers'
2699 0x22, 0x01 // 'CPU_unaligned_access', 'v6'
2701 Section *attr = new_section(s1, ".ARM.attributes", SHT_ARM_ATTRIBUTES, 0);
2702 unsigned char *ptr = section_ptr_add(attr, sizeof(arm_attr));
2703 attr->sh_addralign = 1;
2704 memcpy(ptr, arm_attr, sizeof(arm_attr));
2705 if (s1->float_abi != ARM_HARD_FLOAT) {
2706 ptr[26] = 0x00; // 'FP_arch', 'No'
2707 ptr[41] = 0x1e; // 'ABI_optimization_goals'
2708 ptr[42] = 0x06; // 'Aggressive Debug'
2711 #endif
2713 #if TARGETOS_OpenBSD || TARGETOS_NetBSD
2714 static Section *create_bsd_note_section(TCCState *s1,
2715 const char *name,
2716 const char *value)
2718 Section *s = find_section (s1, name);
2720 if (s->data_offset == 0) {
2721 char *ptr = section_ptr_add(s, sizeof(ElfW(Nhdr)) + 8 + 4);
2722 ElfW(Nhdr) *note = (ElfW(Nhdr) *) ptr;
2724 s->sh_type = SHT_NOTE;
2725 note->n_namesz = 8;
2726 note->n_descsz = 4;
2727 note->n_type = ELF_NOTE_OS_GNU;
2728 strcpy (ptr + sizeof(ElfW(Nhdr)), value);
2730 return s;
2732 #endif
2734 static void alloc_sec_names(TCCState *s1, int is_obj);
2736 /* Output an elf, coff or binary file */
2737 /* XXX: suppress unneeded sections */
2738 static int elf_output_file(TCCState *s1, const char *filename)
2740 int i, ret, file_type, file_offset, *sec_order;
2741 struct dyn_inf dyninf = {0};
2742 Section *interp, *dynstr, *dynamic;
2743 int textrel, got_sym, dt_flags_1;
2745 file_type = s1->output_type;
2746 s1->nb_errors = 0;
2747 ret = -1;
2748 interp = dynstr = dynamic = NULL;
2749 sec_order = NULL;
2750 dyninf.roinf = &dyninf._roinf;
2752 #ifdef TCC_TARGET_ARM
2753 create_arm_attribute_section (s1);
2754 #endif
2756 #if TARGETOS_OpenBSD
2757 dyninf.note = create_bsd_note_section (s1, ".note.openbsd.ident", "OpenBSD");
2758 #endif
2760 #if TARGETOS_NetBSD
2761 dyninf.note = create_bsd_note_section (s1, ".note.netbsd.ident", "NetBSD");
2762 #endif
2764 #if TARGETOS_FreeBSD || TARGETOS_NetBSD
2765 dyninf.roinf = NULL;
2766 #endif
2767 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2768 tcc_add_runtime(s1);
2769 resolve_common_syms(s1);
2771 if (!s1->static_link) {
2772 if (file_type & TCC_OUTPUT_EXE) {
2773 char *ptr;
2774 /* allow override the dynamic loader */
2775 const char *elfint = getenv("LD_SO");
2776 if (elfint == NULL)
2777 elfint = DEFAULT_ELFINTERP(s1);
2778 /* add interpreter section only if executable */
2779 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2780 interp->sh_addralign = 1;
2781 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2782 strcpy(ptr, elfint);
2783 dyninf.interp = interp;
2786 /* add dynamic symbol table */
2787 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2788 ".dynstr",
2789 ".hash", SHF_ALLOC);
2790 /* Number of local symbols (readelf complains if not set) */
2791 s1->dynsym->sh_info = 1;
2792 dynstr = s1->dynsym->link;
2793 /* add dynamic section */
2794 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2795 SHF_ALLOC | SHF_WRITE);
2796 dynamic->link = dynstr;
2797 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2799 got_sym = build_got(s1);
2800 if (file_type == TCC_OUTPUT_EXE) {
2801 bind_exe_dynsyms(s1);
2802 if (s1->nb_errors)
2803 goto the_end;
2805 build_got_entries(s1, got_sym);
2806 if (file_type & TCC_OUTPUT_EXE) {
2807 bind_libs_dynsyms(s1);
2808 } else {
2809 /* shared library case: simply export all global symbols */
2810 export_global_syms(s1);
2812 dyninf.gnu_hash = create_gnu_hash(s1);
2813 } else {
2814 build_got_entries(s1, 0);
2816 version_add (s1);
2818 textrel = set_sec_sizes(s1);
2819 alloc_sec_names(s1, 0);
2821 if (!s1->static_link) {
2822 /* add a list of needed dlls */
2823 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2824 DLLReference *dllref = s1->loaded_dlls[i];
2825 if (dllref->level == 0)
2826 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2829 if (s1->rpath)
2830 put_dt(dynamic, s1->enable_new_dtags ? DT_RUNPATH : DT_RPATH,
2831 put_elf_str(dynstr, s1->rpath));
2833 dt_flags_1 = DF_1_NOW;
2834 if (file_type & TCC_OUTPUT_DYN) {
2835 if (s1->soname)
2836 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2837 /* XXX: currently, since we do not handle PIC code, we
2838 must relocate the readonly segments */
2839 if (textrel)
2840 put_dt(dynamic, DT_TEXTREL, 0);
2841 if (file_type & TCC_OUTPUT_EXE)
2842 dt_flags_1 = DF_1_NOW | DF_1_PIE;
2844 put_dt(dynamic, DT_FLAGS, DF_BIND_NOW);
2845 put_dt(dynamic, DT_FLAGS_1, dt_flags_1);
2846 if (s1->symbolic)
2847 put_dt(dynamic, DT_SYMBOLIC, 0);
2849 dyninf.dynamic = dynamic;
2850 dyninf.dynstr = dynstr;
2851 /* remember offset and reserve space for 2nd call below */
2852 dyninf.data_offset = dynamic->data_offset;
2853 fill_dynamic(s1, &dyninf);
2854 dynamic->sh_size = dynamic->data_offset;
2855 dynstr->sh_size = dynstr->data_offset;
2858 /* this array is used to reorder sections in the output file */
2859 sec_order = tcc_malloc(sizeof(int) * 2 * s1->nb_sections);
2860 /* compute section to program header mapping */
2861 file_offset = layout_sections(s1, sec_order, &dyninf);
2863 if (dynamic) {
2864 /* put in GOT the dynamic section address and relocate PLT */
2865 write32le(s1->got->data, dynamic->sh_addr);
2866 if (file_type == TCC_OUTPUT_EXE
2867 || (RELOCATE_DLLPLT && (file_type & TCC_OUTPUT_DYN)))
2868 relocate_plt(s1);
2869 /* relocate symbols in .dynsym now that final addresses are known */
2870 relocate_syms(s1, s1->dynsym, 2);
2873 /* if building executable or DLL, then relocate each section
2874 except the GOT which is already relocated */
2875 relocate_syms(s1, s1->symtab, 0);
2876 if (s1->nb_errors != 0)
2877 goto the_end;
2878 relocate_sections(s1);
2879 if (dynamic) {
2880 update_reloc_sections (s1, &dyninf);
2881 dynamic->data_offset = dyninf.data_offset;
2882 fill_dynamic(s1, &dyninf);
2884 /* Perform relocation to GOT or PLT entries */
2885 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2886 fill_got(s1);
2887 else if (s1->got)
2888 fill_local_got_entries(s1);
2890 if (dyninf.gnu_hash)
2891 update_gnu_hash(s1, dyninf.gnu_hash);
2893 /* Create the ELF file with name 'filename' */
2894 ret = tcc_write_elf_file(s1, filename, dyninf.phnum, dyninf.phdr, file_offset, sec_order);
2895 the_end:
2896 tcc_free(sec_order);
2897 tcc_free(dyninf.phdr);
2898 return ret;
2900 #endif /* ndef ELF_OBJ_ONLY */
2902 /* Allocate strings for section names */
2903 static void alloc_sec_names(TCCState *s1, int is_obj)
2905 int i;
2906 Section *s, *strsec;
2908 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2909 put_elf_str(strsec, "");
2910 for(i = 1; i < s1->nb_sections; i++) {
2911 s = s1->sections[i];
2912 if (is_obj)
2913 s->sh_size = s->data_offset;
2914 if (s == strsec || s->sh_size || (s->sh_flags & SHF_ALLOC))
2915 s->sh_name = put_elf_str(strsec, s->name);
2917 strsec->sh_size = strsec->data_offset;
2920 /* Output an elf .o file */
2921 static int elf_output_obj(TCCState *s1, const char *filename)
2923 Section *s;
2924 int i, ret, file_offset;
2925 s1->nb_errors = 0;
2926 /* Allocate strings for section names */
2927 alloc_sec_names(s1, 1);
2928 file_offset = sizeof (ElfW(Ehdr));
2929 for(i = 1; i < s1->nb_sections; i++) {
2930 s = s1->sections[i];
2931 file_offset = (file_offset + 15) & -16;
2932 s->sh_offset = file_offset;
2933 if (s->sh_type != SHT_NOBITS)
2934 file_offset += s->sh_size;
2936 /* Create the ELF file with name 'filename' */
2937 ret = tcc_write_elf_file(s1, filename, 0, NULL, file_offset, NULL);
2938 return ret;
2941 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2943 if (s->test_coverage)
2944 tcc_tcov_add_file(s, filename);
2945 if (s->output_type == TCC_OUTPUT_OBJ)
2946 return elf_output_obj(s, filename);
2947 #ifdef TCC_TARGET_PE
2948 return pe_output_file(s, filename);
2949 #elif TCC_TARGET_MACHO
2950 return macho_output_file(s, filename);
2951 #else
2952 return elf_output_file(s, filename);
2953 #endif
2956 ST_FUNC ssize_t full_read(int fd, void *buf, size_t count) {
2957 char *cbuf = buf;
2958 size_t rnum = 0;
2959 while (1) {
2960 ssize_t num = read(fd, cbuf, count-rnum);
2961 if (num < 0) return num;
2962 if (num == 0) return rnum;
2963 rnum += num;
2964 cbuf += num;
2968 ST_FUNC void *load_data(int fd, unsigned long file_offset, unsigned long size)
2970 void *data;
2972 data = tcc_malloc(size);
2973 lseek(fd, file_offset, SEEK_SET);
2974 full_read(fd, data, size);
2975 return data;
2978 typedef struct SectionMergeInfo {
2979 Section *s; /* corresponding existing section */
2980 unsigned long offset; /* offset of the new section in the existing section */
2981 uint8_t new_section; /* true if section 's' was added */
2982 uint8_t link_once; /* true if link once section */
2983 } SectionMergeInfo;
2985 ST_FUNC int tcc_object_type(int fd, ElfW(Ehdr) *h)
2987 int size = full_read(fd, h, sizeof *h);
2988 if (size == sizeof *h && 0 == memcmp(h, ELFMAG, 4)) {
2989 if (h->e_type == ET_REL)
2990 return AFF_BINTYPE_REL;
2991 if (h->e_type == ET_DYN)
2992 return AFF_BINTYPE_DYN;
2993 } else if (size >= 8) {
2994 if (0 == memcmp(h, ARMAG, 8))
2995 return AFF_BINTYPE_AR;
2996 #ifdef TCC_TARGET_COFF
2997 if (((struct filehdr*)h)->f_magic == COFF_C67_MAGIC)
2998 return AFF_BINTYPE_C67;
2999 #endif
3001 return 0;
3004 /* load an object file and merge it with current files */
3005 /* XXX: handle correctly stab (debug) info */
3006 ST_FUNC int tcc_load_object_file(TCCState *s1,
3007 int fd, unsigned long file_offset)
3009 ElfW(Ehdr) ehdr;
3010 ElfW(Shdr) *shdr, *sh;
3011 unsigned long size, offset, offseti;
3012 int i, j, nb_syms, sym_index, ret, seencompressed;
3013 char *strsec, *strtab;
3014 int stab_index, stabstr_index;
3015 int *old_to_new_syms;
3016 char *sh_name, *name;
3017 SectionMergeInfo *sm_table, *sm;
3018 ElfW(Sym) *sym, *symtab;
3019 ElfW_Rel *rel;
3020 Section *s;
3022 lseek(fd, file_offset, SEEK_SET);
3023 if (tcc_object_type(fd, &ehdr) != AFF_BINTYPE_REL)
3024 goto invalid;
3025 /* test CPU specific stuff */
3026 if (ehdr.e_ident[5] != ELFDATA2LSB ||
3027 ehdr.e_machine != EM_TCC_TARGET) {
3028 invalid:
3029 return tcc_error_noabort("invalid object file");
3031 /* read sections */
3032 shdr = load_data(fd, file_offset + ehdr.e_shoff,
3033 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
3034 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
3036 /* load section names */
3037 sh = &shdr[ehdr.e_shstrndx];
3038 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
3040 /* load symtab and strtab */
3041 old_to_new_syms = NULL;
3042 symtab = NULL;
3043 strtab = NULL;
3044 nb_syms = 0;
3045 seencompressed = 0;
3046 stab_index = stabstr_index = 0;
3047 ret = -1;
3049 for(i = 1; i < ehdr.e_shnum; i++) {
3050 sh = &shdr[i];
3051 if (sh->sh_type == SHT_SYMTAB) {
3052 if (symtab) {
3053 tcc_error_noabort("object must contain only one symtab");
3054 goto the_end;
3056 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
3057 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
3058 sm_table[i].s = symtab_section;
3060 /* now load strtab */
3061 sh = &shdr[sh->sh_link];
3062 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
3064 if (sh->sh_flags & SHF_COMPRESSED)
3065 seencompressed = 1;
3068 /* now examine each section and try to merge its content with the
3069 ones in memory */
3070 for(i = 1; i < ehdr.e_shnum; i++) {
3071 /* no need to examine section name strtab */
3072 if (i == ehdr.e_shstrndx)
3073 continue;
3074 sh = &shdr[i];
3075 if (sh->sh_type == SHT_RELX)
3076 sh = &shdr[sh->sh_info];
3077 /* ignore sections types we do not handle (plus relocs to those) */
3078 if (sh->sh_type != SHT_PROGBITS &&
3079 #ifdef TCC_ARM_EABI
3080 sh->sh_type != SHT_ARM_EXIDX &&
3081 #endif
3082 #if TARGETOS_OpenBSD || TARGETOS_FreeBSD || TARGETOS_NetBSD
3083 sh->sh_type != SHT_X86_64_UNWIND &&
3084 #endif
3085 sh->sh_type != SHT_NOTE &&
3086 sh->sh_type != SHT_NOBITS &&
3087 sh->sh_type != SHT_PREINIT_ARRAY &&
3088 sh->sh_type != SHT_INIT_ARRAY &&
3089 sh->sh_type != SHT_FINI_ARRAY &&
3090 strcmp(strsec + sh->sh_name, ".stabstr")
3092 continue;
3093 if (seencompressed && 0 == strncmp(strsec + sh->sh_name, ".debug_", 7))
3094 continue;
3096 sh = &shdr[i];
3097 sh_name = strsec + sh->sh_name;
3098 if (sh->sh_addralign < 1)
3099 sh->sh_addralign = 1;
3100 /* find corresponding section, if any */
3101 for(j = 1; j < s1->nb_sections;j++) {
3102 s = s1->sections[j];
3103 if (!strcmp(s->name, sh_name)) {
3104 if (!strncmp(sh_name, ".gnu.linkonce",
3105 sizeof(".gnu.linkonce") - 1)) {
3106 /* if a 'linkonce' section is already present, we
3107 do not add it again. It is a little tricky as
3108 symbols can still be defined in
3109 it. */
3110 sm_table[i].link_once = 1;
3111 goto next;
3113 if (stab_section) {
3114 if (s == stab_section)
3115 stab_index = i;
3116 if (s == stab_section->link)
3117 stabstr_index = i;
3119 goto found;
3122 /* not found: create new section */
3123 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags & ~SHF_GROUP);
3124 /* take as much info as possible from the section. sh_link and
3125 sh_info will be updated later */
3126 s->sh_addralign = sh->sh_addralign;
3127 s->sh_entsize = sh->sh_entsize;
3128 sm_table[i].new_section = 1;
3129 found:
3130 if (sh->sh_type != s->sh_type
3131 #if TARGETOS_OpenBSD || TARGETOS_FreeBSD || TARGETOS_NetBSD
3132 && strcmp (s->name, ".eh_frame")
3133 #endif
3135 tcc_error_noabort("invalid section type");
3136 goto the_end;
3138 /* align start of section */
3139 s->data_offset += -s->data_offset & (sh->sh_addralign - 1);
3140 if (sh->sh_addralign > s->sh_addralign)
3141 s->sh_addralign = sh->sh_addralign;
3142 sm_table[i].offset = s->data_offset;
3143 sm_table[i].s = s;
3144 /* concatenate sections */
3145 size = sh->sh_size;
3146 if (sh->sh_type != SHT_NOBITS) {
3147 unsigned char *ptr;
3148 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
3149 ptr = section_ptr_add(s, size);
3150 full_read(fd, ptr, size);
3151 } else {
3152 s->data_offset += size;
3154 next: ;
3157 /* gr relocate stab strings */
3158 if (stab_index && stabstr_index) {
3159 Stab_Sym *a, *b;
3160 unsigned o;
3161 s = sm_table[stab_index].s;
3162 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
3163 b = (Stab_Sym *)(s->data + s->data_offset);
3164 o = sm_table[stabstr_index].offset;
3165 while (a < b) {
3166 if (a->n_strx)
3167 a->n_strx += o;
3168 a++;
3172 /* second short pass to update sh_link and sh_info fields of new
3173 sections */
3174 for(i = 1; i < ehdr.e_shnum; i++) {
3175 s = sm_table[i].s;
3176 if (!s || !sm_table[i].new_section)
3177 continue;
3178 sh = &shdr[i];
3179 if (sh->sh_link > 0)
3180 s->link = sm_table[sh->sh_link].s;
3181 if (sh->sh_type == SHT_RELX) {
3182 s->sh_info = sm_table[sh->sh_info].s->sh_num;
3183 /* update backward link */
3184 s1->sections[s->sh_info]->reloc = s;
3188 /* resolve symbols */
3189 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
3191 sym = symtab + 1;
3192 for(i = 1; i < nb_syms; i++, sym++) {
3193 if (sym->st_shndx != SHN_UNDEF &&
3194 sym->st_shndx < SHN_LORESERVE) {
3195 sm = &sm_table[sym->st_shndx];
3196 if (sm->link_once) {
3197 /* if a symbol is in a link once section, we use the
3198 already defined symbol. It is very important to get
3199 correct relocations */
3200 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
3201 name = strtab + sym->st_name;
3202 sym_index = find_elf_sym(symtab_section, name);
3203 if (sym_index)
3204 old_to_new_syms[i] = sym_index;
3206 continue;
3208 /* if no corresponding section added, no need to add symbol */
3209 if (!sm->s)
3210 continue;
3211 /* convert section number */
3212 sym->st_shndx = sm->s->sh_num;
3213 /* offset value */
3214 sym->st_value += sm->offset;
3216 /* add symbol */
3217 name = strtab + sym->st_name;
3218 sym_index = set_elf_sym(symtab_section, sym->st_value, sym->st_size,
3219 sym->st_info, sym->st_other,
3220 sym->st_shndx, name);
3221 old_to_new_syms[i] = sym_index;
3224 /* third pass to patch relocation entries */
3225 for(i = 1; i < ehdr.e_shnum; i++) {
3226 s = sm_table[i].s;
3227 if (!s)
3228 continue;
3229 sh = &shdr[i];
3230 offset = sm_table[i].offset;
3231 size = sh->sh_size;
3232 switch(s->sh_type) {
3233 case SHT_RELX:
3234 /* take relocation offset information */
3235 offseti = sm_table[sh->sh_info].offset;
3236 for (rel = (ElfW_Rel *) s->data + (offset / sizeof(*rel));
3237 rel < (ElfW_Rel *) s->data + ((offset + size) / sizeof(*rel));
3238 rel++) {
3239 int type;
3240 unsigned sym_index;
3241 /* convert symbol index */
3242 type = ELFW(R_TYPE)(rel->r_info);
3243 sym_index = ELFW(R_SYM)(rel->r_info);
3244 /* NOTE: only one symtab assumed */
3245 if (sym_index >= nb_syms)
3246 goto invalid_reloc;
3247 sym_index = old_to_new_syms[sym_index];
3248 /* ignore link_once in rel section. */
3249 if (!sym_index && !sm_table[sh->sh_info].link_once
3250 #ifdef TCC_TARGET_ARM
3251 && type != R_ARM_V4BX
3252 #elif defined TCC_TARGET_RISCV64
3253 && type != R_RISCV_ALIGN
3254 && type != R_RISCV_RELAX
3255 #endif
3257 invalid_reloc:
3258 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
3259 i, strsec + sh->sh_name, (int)rel->r_offset);
3260 goto the_end;
3262 rel->r_info = ELFW(R_INFO)(sym_index, type);
3263 /* offset the relocation offset */
3264 rel->r_offset += offseti;
3265 #ifdef TCC_TARGET_ARM
3266 /* Jumps and branches from a Thumb code to a PLT entry need
3267 special handling since PLT entries are ARM code.
3268 Unconditional bl instructions referencing PLT entries are
3269 handled by converting these instructions into blx
3270 instructions. Other case of instructions referencing a PLT
3271 entry require to add a Thumb stub before the PLT entry to
3272 switch to ARM mode. We set bit plt_thumb_stub of the
3273 attribute of a symbol to indicate such a case. */
3274 if (type == R_ARM_THM_JUMP24)
3275 get_sym_attr(s1, sym_index, 1)->plt_thumb_stub = 1;
3276 #endif
3278 break;
3279 default:
3280 break;
3284 ret = 0;
3285 the_end:
3286 tcc_free(symtab);
3287 tcc_free(strtab);
3288 tcc_free(old_to_new_syms);
3289 tcc_free(sm_table);
3290 tcc_free(strsec);
3291 tcc_free(shdr);
3292 return ret;
3295 typedef struct ArchiveHeader {
3296 char ar_name[16]; /* name of this member */
3297 char ar_date[12]; /* file mtime */
3298 char ar_uid[6]; /* owner uid; printed as decimal */
3299 char ar_gid[6]; /* owner gid; printed as decimal */
3300 char ar_mode[8]; /* file mode, printed as octal */
3301 char ar_size[10]; /* file size, printed as decimal */
3302 char ar_fmag[2]; /* should contain ARFMAG */
3303 } ArchiveHeader;
3305 #define ARFMAG "`\n"
3307 static unsigned long long get_be(const uint8_t *b, int n)
3309 unsigned long long ret = 0;
3310 while (n)
3311 ret = (ret << 8) | *b++, --n;
3312 return ret;
3315 static int read_ar_header(int fd, int offset, ArchiveHeader *hdr)
3317 char *p, *e;
3318 int len;
3319 lseek(fd, offset, SEEK_SET);
3320 len = full_read(fd, hdr, sizeof(ArchiveHeader));
3321 if (len != sizeof(ArchiveHeader))
3322 return len ? -1 : 0;
3323 p = hdr->ar_name;
3324 for (e = p + sizeof hdr->ar_name; e > p && e[-1] == ' ';)
3325 --e;
3326 *e = '\0';
3327 hdr->ar_size[sizeof hdr->ar_size-1] = 0;
3328 return len;
3331 /* load only the objects which resolve undefined symbols */
3332 static int tcc_load_alacarte(TCCState *s1, int fd, int size, int entrysize)
3334 int i, bound, nsyms, sym_index, len, ret = -1;
3335 unsigned long long off;
3336 uint8_t *data;
3337 const char *ar_names, *p;
3338 const uint8_t *ar_index;
3339 ElfW(Sym) *sym;
3340 ArchiveHeader hdr;
3342 data = tcc_malloc(size);
3343 if (full_read(fd, data, size) != size)
3344 goto the_end;
3345 nsyms = get_be(data, entrysize);
3346 ar_index = data + entrysize;
3347 ar_names = (char *) ar_index + nsyms * entrysize;
3349 do {
3350 bound = 0;
3351 for (p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
3352 Section *s = symtab_section;
3353 sym_index = find_elf_sym(s, p);
3354 if (!sym_index)
3355 continue;
3356 sym = &((ElfW(Sym) *)s->data)[sym_index];
3357 if(sym->st_shndx != SHN_UNDEF)
3358 continue;
3359 off = get_be(ar_index + i * entrysize, entrysize);
3360 len = read_ar_header(fd, off, &hdr);
3361 if (len <= 0 || memcmp(hdr.ar_fmag, ARFMAG, 2)) {
3362 tcc_error_noabort("invalid archive");
3363 goto the_end;
3365 off += len;
3366 if (s1->verbose == 2)
3367 printf(" -> %s\n", hdr.ar_name);
3368 if (tcc_load_object_file(s1, fd, off) < 0)
3369 goto the_end;
3370 ++bound;
3372 } while(bound);
3373 ret = 0;
3374 the_end:
3375 tcc_free(data);
3376 return ret;
3379 /* load a '.a' file */
3380 ST_FUNC int tcc_load_archive(TCCState *s1, int fd, int alacarte)
3382 ArchiveHeader hdr;
3383 /* char magic[8]; */
3384 int size, len;
3385 unsigned long file_offset;
3386 ElfW(Ehdr) ehdr;
3388 /* skip magic which was already checked */
3389 /* full_read(fd, magic, sizeof(magic)); */
3390 file_offset = sizeof ARMAG - 1;
3392 for(;;) {
3393 len = read_ar_header(fd, file_offset, &hdr);
3394 if (len == 0)
3395 return 0;
3396 if (len < 0)
3397 return tcc_error_noabort("invalid archive");
3398 file_offset += len;
3399 size = strtol(hdr.ar_size, NULL, 0);
3400 /* align to even */
3401 size = (size + 1) & ~1;
3402 if (alacarte) {
3403 /* coff symbol table : we handle it */
3404 if (!strcmp(hdr.ar_name, "/"))
3405 return tcc_load_alacarte(s1, fd, size, 4);
3406 if (!strcmp(hdr.ar_name, "/SYM64/"))
3407 return tcc_load_alacarte(s1, fd, size, 8);
3408 } else if (tcc_object_type(fd, &ehdr) == AFF_BINTYPE_REL) {
3409 if (s1->verbose == 2)
3410 printf(" -> %s\n", hdr.ar_name);
3411 if (tcc_load_object_file(s1, fd, file_offset) < 0)
3412 return -1;
3414 file_offset += size;
3418 #ifndef ELF_OBJ_ONLY
3419 /* Set LV[I] to the global index of sym-version (LIB,VERSION). Maybe resizes
3420 LV, maybe create a new entry for (LIB,VERSION). */
3421 static void set_ver_to_ver(TCCState *s1, int *n, int **lv, int i, char *lib, char *version)
3423 while (i >= *n) {
3424 *lv = tcc_realloc(*lv, (*n + 1) * sizeof(**lv));
3425 (*lv)[(*n)++] = -1;
3427 if ((*lv)[i] == -1) {
3428 int v, prev_same_lib = -1;
3429 for (v = 0; v < nb_sym_versions; v++) {
3430 if (strcmp(sym_versions[v].lib, lib))
3431 continue;
3432 prev_same_lib = v;
3433 if (!strcmp(sym_versions[v].version, version))
3434 break;
3436 if (v == nb_sym_versions) {
3437 sym_versions = tcc_realloc (sym_versions,
3438 (v + 1) * sizeof(*sym_versions));
3439 sym_versions[v].lib = tcc_strdup(lib);
3440 sym_versions[v].version = tcc_strdup(version);
3441 sym_versions[v].out_index = 0;
3442 sym_versions[v].prev_same_lib = prev_same_lib;
3443 nb_sym_versions++;
3445 (*lv)[i] = v;
3449 /* Associates symbol SYM_INDEX (in dynsymtab) with sym-version index
3450 VERNDX. */
3451 static void
3452 set_sym_version(TCCState *s1, int sym_index, int verndx)
3454 if (sym_index >= nb_sym_to_version) {
3455 int newelems = sym_index ? sym_index * 2 : 1;
3456 sym_to_version = tcc_realloc(sym_to_version,
3457 newelems * sizeof(*sym_to_version));
3458 memset(sym_to_version + nb_sym_to_version, -1,
3459 (newelems - nb_sym_to_version) * sizeof(*sym_to_version));
3460 nb_sym_to_version = newelems;
3462 if (sym_to_version[sym_index] < 0)
3463 sym_to_version[sym_index] = verndx;
3466 struct versym_info {
3467 int nb_versyms;
3468 ElfW(Verdef) *verdef;
3469 ElfW(Verneed) *verneed;
3470 ElfW(Half) *versym;
3471 int nb_local_ver, *local_ver;
3475 static void store_version(TCCState *s1, struct versym_info *v, char *dynstr)
3477 char *lib, *version;
3478 uint32_t next;
3479 int i;
3481 #define DEBUG_VERSION 0
3483 if (v->versym && v->verdef) {
3484 ElfW(Verdef) *vdef = v->verdef;
3485 lib = NULL;
3486 do {
3487 ElfW(Verdaux) *verdaux =
3488 (ElfW(Verdaux) *) (((char *) vdef) + vdef->vd_aux);
3490 #if DEBUG_VERSION
3491 printf ("verdef: version:%u flags:%u index:%u, hash:%u\n",
3492 vdef->vd_version, vdef->vd_flags, vdef->vd_ndx,
3493 vdef->vd_hash);
3494 #endif
3495 if (vdef->vd_cnt) {
3496 version = dynstr + verdaux->vda_name;
3498 if (lib == NULL)
3499 lib = version;
3500 else
3501 set_ver_to_ver(s1, &v->nb_local_ver, &v->local_ver, vdef->vd_ndx,
3502 lib, version);
3503 #if DEBUG_VERSION
3504 printf (" verdaux(%u): %s\n", vdef->vd_ndx, version);
3505 #endif
3507 next = vdef->vd_next;
3508 vdef = (ElfW(Verdef) *) (((char *) vdef) + next);
3509 } while (next);
3511 if (v->versym && v->verneed) {
3512 ElfW(Verneed) *vneed = v->verneed;
3513 do {
3514 ElfW(Vernaux) *vernaux =
3515 (ElfW(Vernaux) *) (((char *) vneed) + vneed->vn_aux);
3517 lib = dynstr + vneed->vn_file;
3518 #if DEBUG_VERSION
3519 printf ("verneed: %u %s\n", vneed->vn_version, lib);
3520 #endif
3521 for (i = 0; i < vneed->vn_cnt; i++) {
3522 if ((vernaux->vna_other & 0x8000) == 0) { /* hidden */
3523 version = dynstr + vernaux->vna_name;
3524 set_ver_to_ver(s1, &v->nb_local_ver, &v->local_ver, vernaux->vna_other,
3525 lib, version);
3526 #if DEBUG_VERSION
3527 printf (" vernaux(%u): %u %u %s\n",
3528 vernaux->vna_other, vernaux->vna_hash,
3529 vernaux->vna_flags, version);
3530 #endif
3532 vernaux = (ElfW(Vernaux) *) (((char *) vernaux) + vernaux->vna_next);
3534 next = vneed->vn_next;
3535 vneed = (ElfW(Verneed) *) (((char *) vneed) + next);
3536 } while (next);
3539 #if DEBUG_VERSION
3540 for (i = 0; i < v->nb_local_ver; i++) {
3541 if (v->local_ver[i] > 0) {
3542 printf ("%d: lib: %s, version %s\n",
3543 i, sym_versions[v->local_ver[i]].lib,
3544 sym_versions[v->local_ver[i]].version);
3547 #endif
3550 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3551 is referenced by the user (so it should be added as DT_NEEDED in
3552 the generated ELF file) */
3553 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
3555 ElfW(Ehdr) ehdr;
3556 ElfW(Shdr) *shdr, *sh, *sh1;
3557 int i, nb_syms, nb_dts, sym_bind, ret = -1;
3558 ElfW(Sym) *sym, *dynsym;
3559 ElfW(Dyn) *dt, *dynamic;
3561 char *dynstr;
3562 int sym_index;
3563 const char *name, *soname;
3564 struct versym_info v;
3566 full_read(fd, &ehdr, sizeof(ehdr));
3568 /* test CPU specific stuff */
3569 if (ehdr.e_ident[5] != ELFDATA2LSB ||
3570 ehdr.e_machine != EM_TCC_TARGET) {
3571 return tcc_error_noabort("bad architecture");
3574 /* read sections */
3575 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
3577 /* load dynamic section and dynamic symbols */
3578 nb_syms = 0;
3579 nb_dts = 0;
3580 dynamic = NULL;
3581 dynsym = NULL; /* avoid warning */
3582 dynstr = NULL; /* avoid warning */
3583 memset(&v, 0, sizeof v);
3585 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
3586 switch(sh->sh_type) {
3587 case SHT_DYNAMIC:
3588 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
3589 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
3590 break;
3591 case SHT_DYNSYM:
3592 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
3593 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
3594 sh1 = &shdr[sh->sh_link];
3595 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
3596 break;
3597 case SHT_GNU_verdef:
3598 v.verdef = load_data(fd, sh->sh_offset, sh->sh_size);
3599 break;
3600 case SHT_GNU_verneed:
3601 v.verneed = load_data(fd, sh->sh_offset, sh->sh_size);
3602 break;
3603 case SHT_GNU_versym:
3604 v.nb_versyms = sh->sh_size / sizeof(ElfW(Half));
3605 v.versym = load_data(fd, sh->sh_offset, sh->sh_size);
3606 break;
3607 default:
3608 break;
3612 if (!dynamic)
3613 goto the_end;
3615 /* compute the real library name */
3616 soname = tcc_basename(filename);
3617 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++)
3618 if (dt->d_tag == DT_SONAME)
3619 soname = dynstr + dt->d_un.d_val;
3621 /* if the dll is already loaded, do not load it */
3622 if (tcc_add_dllref(s1, soname, level)->found)
3623 goto ret_success;
3625 if (v.nb_versyms != nb_syms)
3626 tcc_free (v.versym), v.versym = NULL;
3627 else
3628 store_version(s1, &v, dynstr);
3630 /* add dynamic symbols in dynsym_section */
3631 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3632 sym_bind = ELFW(ST_BIND)(sym->st_info);
3633 if (sym_bind == STB_LOCAL)
3634 continue;
3635 name = dynstr + sym->st_name;
3636 sym_index = set_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3637 sym->st_info, sym->st_other, sym->st_shndx, name);
3638 if (v.versym) {
3639 ElfW(Half) vsym = v.versym[i];
3640 if ((vsym & 0x8000) == 0 && vsym > 0 && vsym < v.nb_local_ver)
3641 set_sym_version(s1, sym_index, v.local_ver[vsym]);
3645 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++)
3646 if (dt->d_tag == DT_RPATH)
3647 tcc_add_library_path(s1, dynstr + dt->d_un.d_val);
3649 /* load all referenced DLLs */
3650 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3651 switch(dt->d_tag) {
3652 case DT_NEEDED:
3653 name = dynstr + dt->d_un.d_val;
3654 if (tcc_add_dllref(s1, name, -1))
3655 continue;
3656 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3657 ret = tcc_error_noabort("referenced dll '%s' not found", name);
3658 goto the_end;
3663 ret_success:
3664 ret = 0;
3665 the_end:
3666 tcc_free(dynstr);
3667 tcc_free(dynsym);
3668 tcc_free(dynamic);
3669 tcc_free(shdr);
3670 tcc_free(v.local_ver);
3671 tcc_free(v.verdef);
3672 tcc_free(v.verneed);
3673 tcc_free(v.versym);
3674 return ret;
3677 #define LD_TOK_NAME 256
3678 #define LD_TOK_EOF (-1)
3680 static int ld_inp(TCCState *s1)
3682 char b;
3683 if (s1->cc != -1) {
3684 int c = s1->cc;
3685 s1->cc = -1;
3686 return c;
3688 if (1 == read(s1->fd, &b, 1))
3689 return b;
3690 return CH_EOF;
3693 /* return next ld script token */
3694 static int ld_next(TCCState *s1, char *name, int name_size)
3696 int c, d, ch;
3697 char *q;
3699 redo:
3700 ch = ld_inp(s1);
3701 switch(ch) {
3702 case ' ':
3703 case '\t':
3704 case '\f':
3705 case '\v':
3706 case '\r':
3707 case '\n':
3708 goto redo;
3709 case '/':
3710 ch = ld_inp(s1);
3711 if (ch == '*') { /* comment */
3712 for (d = 0;; d = ch) {
3713 ch = ld_inp(s1);
3714 if (ch == CH_EOF || (ch == '/' && d == '*'))
3715 break;
3717 goto redo;
3718 } else {
3719 q = name;
3720 *q++ = '/';
3721 goto parse_name;
3723 break;
3724 case '\\':
3725 /* case 'a' ... 'z': */
3726 case 'a':
3727 case 'b':
3728 case 'c':
3729 case 'd':
3730 case 'e':
3731 case 'f':
3732 case 'g':
3733 case 'h':
3734 case 'i':
3735 case 'j':
3736 case 'k':
3737 case 'l':
3738 case 'm':
3739 case 'n':
3740 case 'o':
3741 case 'p':
3742 case 'q':
3743 case 'r':
3744 case 's':
3745 case 't':
3746 case 'u':
3747 case 'v':
3748 case 'w':
3749 case 'x':
3750 case 'y':
3751 case 'z':
3752 /* case 'A' ... 'z': */
3753 case 'A':
3754 case 'B':
3755 case 'C':
3756 case 'D':
3757 case 'E':
3758 case 'F':
3759 case 'G':
3760 case 'H':
3761 case 'I':
3762 case 'J':
3763 case 'K':
3764 case 'L':
3765 case 'M':
3766 case 'N':
3767 case 'O':
3768 case 'P':
3769 case 'Q':
3770 case 'R':
3771 case 'S':
3772 case 'T':
3773 case 'U':
3774 case 'V':
3775 case 'W':
3776 case 'X':
3777 case 'Y':
3778 case 'Z':
3779 case '_':
3780 case '.':
3781 case '$':
3782 case '~':
3783 q = name;
3784 parse_name:
3785 for(;;) {
3786 if (!((ch >= 'a' && ch <= 'z') ||
3787 (ch >= 'A' && ch <= 'Z') ||
3788 (ch >= '0' && ch <= '9') ||
3789 strchr("/.-_+=$:\\,~", ch)))
3790 break;
3791 if ((q - name) < name_size - 1) {
3792 *q++ = ch;
3794 ch = ld_inp(s1);
3796 s1->cc = ch;
3797 *q = '\0';
3798 c = LD_TOK_NAME;
3799 break;
3800 case CH_EOF:
3801 c = LD_TOK_EOF;
3802 break;
3803 default:
3804 c = ch;
3805 break;
3807 return c;
3810 static int ld_add_file(TCCState *s1, const char filename[])
3812 if (filename[0] == '/') {
3813 if (CONFIG_SYSROOT[0] == '\0'
3814 && tcc_add_file_internal(s1, filename, AFF_TYPE_BIN) == 0)
3815 return 0;
3816 filename = tcc_basename(filename);
3818 return tcc_add_dll(s1, filename, 0);
3821 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3823 char filename[1024], libname[1024];
3824 int t, group, nblibs = 0, ret = 0;
3825 char **libs = NULL;
3827 group = !strcmp(cmd, "GROUP");
3828 if (!as_needed)
3829 s1->new_undef_sym = 0;
3830 t = ld_next(s1, filename, sizeof(filename));
3831 if (t != '(') {
3832 ret = tcc_error_noabort("( expected");
3833 goto lib_parse_error;
3835 t = ld_next(s1, filename, sizeof(filename));
3836 for(;;) {
3837 libname[0] = '\0';
3838 if (t == LD_TOK_EOF) {
3839 ret = tcc_error_noabort("unexpected end of file");
3840 goto lib_parse_error;
3841 } else if (t == ')') {
3842 break;
3843 } else if (t == '-') {
3844 t = ld_next(s1, filename, sizeof(filename));
3845 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3846 ret = tcc_error_noabort("library name expected");
3847 goto lib_parse_error;
3849 pstrcpy(libname, sizeof libname, &filename[1]);
3850 if (s1->static_link) {
3851 snprintf(filename, sizeof filename, "lib%s.a", libname);
3852 } else {
3853 snprintf(filename, sizeof filename, "lib%s.so", libname);
3855 } else if (t != LD_TOK_NAME) {
3856 ret = tcc_error_noabort("filename expected");
3857 goto lib_parse_error;
3859 if (!strcmp(filename, "AS_NEEDED")) {
3860 ret = ld_add_file_list(s1, cmd, 1);
3861 if (ret)
3862 goto lib_parse_error;
3863 } else {
3864 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3865 if (!as_needed) {
3866 ret = ld_add_file(s1, filename);
3867 if (ret)
3868 goto lib_parse_error;
3869 if (group) {
3870 /* Add the filename *and* the libname to avoid future conversions */
3871 dynarray_add(&libs, &nblibs, tcc_strdup(filename));
3872 if (libname[0] != '\0')
3873 dynarray_add(&libs, &nblibs, tcc_strdup(libname));
3877 t = ld_next(s1, filename, sizeof(filename));
3878 if (t == ',') {
3879 t = ld_next(s1, filename, sizeof(filename));
3882 if (group && !as_needed) {
3883 while (s1->new_undef_sym) {
3884 int i;
3885 s1->new_undef_sym = 0;
3886 for (i = 0; i < nblibs; i ++)
3887 ld_add_file(s1, libs[i]);
3890 lib_parse_error:
3891 dynarray_reset(&libs, &nblibs);
3892 return ret;
3895 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3896 files */
3897 ST_FUNC int tcc_load_ldscript(TCCState *s1, int fd)
3899 char cmd[64];
3900 char filename[1024];
3901 int t, ret;
3903 s1->fd = fd;
3904 s1->cc = -1;
3905 for(;;) {
3906 t = ld_next(s1, cmd, sizeof(cmd));
3907 if (t == LD_TOK_EOF)
3908 return 0;
3909 else if (t != LD_TOK_NAME)
3910 return -1;
3911 if (!strcmp(cmd, "INPUT") ||
3912 !strcmp(cmd, "GROUP")) {
3913 ret = ld_add_file_list(s1, cmd, 0);
3914 if (ret)
3915 return ret;
3916 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3917 !strcmp(cmd, "TARGET")) {
3918 /* ignore some commands */
3919 t = ld_next(s1, cmd, sizeof(cmd));
3920 if (t != '(')
3921 return tcc_error_noabort("( expected");
3922 for(;;) {
3923 t = ld_next(s1, filename, sizeof(filename));
3924 if (t == LD_TOK_EOF) {
3925 return tcc_error_noabort("unexpected end of file");
3926 } else if (t == ')') {
3927 break;
3930 } else {
3931 return -1;
3934 return 0;
3936 #endif /* !ELF_OBJ_ONLY */