tcc_add_file(): preserve s->filetype
[tinycc.git] / tccelf.c
blobf3dc1f3b389e10aed0e6b780d8cac2b40bc5b031
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 ST_DATA Section *text_section, *data_section, *bss_section; /* predefined sections */
30 ST_DATA Section *common_section;
31 ST_DATA Section *cur_text_section; /* current section where function code is generated */
32 #ifdef CONFIG_TCC_ASM
33 ST_DATA Section *last_text_section; /* to handle .previous asm directive */
34 #endif
35 #ifdef CONFIG_TCC_BCHECK
36 /* bound check related sections */
37 ST_DATA Section *bounds_section; /* contains global data bound description */
38 ST_DATA Section *lbounds_section; /* contains local data bound description */
39 #endif
40 /* symbol sections */
41 ST_DATA Section *symtab_section;
42 /* debug sections */
43 ST_DATA Section *stab_section, *stabstr_section;
45 /* XXX: avoid static variable */
46 static int new_undef_sym = 0; /* Is there a new undefined sym since last new_undef_sym() */
48 /* special flag to indicate that the section should not be linked to the other ones */
49 #define SHF_PRIVATE 0x80000000
50 /* section is dynsymtab_section */
51 #define SHF_DYNSYM 0x40000000
53 /* ------------------------------------------------------------------------- */
55 ST_FUNC void tccelf_new(TCCState *s)
57 /* no section zero */
58 dynarray_add(&s->sections, &s->nb_sections, NULL);
60 /* create standard sections */
61 text_section = new_section(s, ".text", SHT_PROGBITS, SHF_ALLOC | SHF_EXECINSTR);
62 data_section = new_section(s, ".data", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
63 bss_section = new_section(s, ".bss", SHT_NOBITS, SHF_ALLOC | SHF_WRITE);
64 common_section = new_section(s, ".common", SHT_NOBITS, SHF_PRIVATE);
65 common_section->sh_num = SHN_COMMON;
67 /* symbols are always generated for linking stage */
68 symtab_section = new_symtab(s, ".symtab", SHT_SYMTAB, 0,
69 ".strtab",
70 ".hashtab", SHF_PRIVATE);
71 s->symtab = symtab_section;
73 /* private symbol table for dynamic symbols */
74 s->dynsymtab_section = new_symtab(s, ".dynsymtab", SHT_SYMTAB, SHF_PRIVATE|SHF_DYNSYM,
75 ".dynstrtab",
76 ".dynhashtab", SHF_PRIVATE);
77 get_sym_attr(s, 0, 1);
80 #ifdef CONFIG_TCC_BCHECK
81 ST_FUNC void tccelf_bounds_new(TCCState *s)
83 /* create bounds sections */
84 bounds_section = new_section(s, ".bounds",
85 SHT_PROGBITS, SHF_ALLOC);
86 lbounds_section = new_section(s, ".lbounds",
87 SHT_PROGBITS, SHF_ALLOC);
89 #endif
91 ST_FUNC void tccelf_stab_new(TCCState *s)
93 stab_section = new_section(s, ".stab", SHT_PROGBITS, 0);
94 stab_section->sh_entsize = sizeof(Stab_Sym);
95 stabstr_section = new_section(s, ".stabstr", SHT_STRTAB, 0);
96 put_elf_str(stabstr_section, "");
97 stab_section->link = stabstr_section;
98 /* put first entry */
99 put_stabs("", 0, 0, 0, 0);
102 static void free_section(Section *s)
104 tcc_free(s->data);
107 ST_FUNC void tccelf_delete(TCCState *s1)
109 int i;
111 /* free all sections */
112 for(i = 1; i < s1->nb_sections; i++)
113 free_section(s1->sections[i]);
114 dynarray_reset(&s1->sections, &s1->nb_sections);
116 for(i = 0; i < s1->nb_priv_sections; i++)
117 free_section(s1->priv_sections[i]);
118 dynarray_reset(&s1->priv_sections, &s1->nb_priv_sections);
120 /* free any loaded DLLs */
121 #ifdef TCC_IS_NATIVE
122 for ( i = 0; i < s1->nb_loaded_dlls; i++) {
123 DLLReference *ref = s1->loaded_dlls[i];
124 if ( ref->handle )
125 # ifdef _WIN32
126 FreeLibrary((HMODULE)ref->handle);
127 # else
128 dlclose(ref->handle);
129 # endif
131 #endif
132 /* free loaded dlls array */
133 dynarray_reset(&s1->loaded_dlls, &s1->nb_loaded_dlls);
134 tcc_free(s1->sym_attrs);
136 symtab_section = NULL; /* for tccrun.c:rt_printline() */
139 /* save section data state */
140 ST_FUNC void tccelf_begin_file(TCCState *s1)
142 Section *s; int i;
143 for (i = 1; i < s1->nb_sections; i++) {
144 s = s1->sections[i];
145 s->sh_offset = s->data_offset;
147 /* disable symbol hashing during compilation */
148 s = s1->symtab, s->reloc = s->hash, s->hash = NULL;
149 #if defined TCC_TARGET_X86_64 && defined TCC_TARGET_PE
150 s1->uw_sym = 0;
151 #endif
154 /* At the end of compilation, convert any UNDEF syms to global, and merge
155 with previously existing symbols */
156 ST_FUNC void tccelf_end_file(TCCState *s1)
158 Section *s = s1->symtab;
159 int first_sym, nb_syms, *tr, i;
161 first_sym = s->sh_offset / sizeof (ElfSym);
162 nb_syms = s->data_offset / sizeof (ElfSym) - first_sym;
163 s->data_offset = s->sh_offset;
164 s->link->data_offset = s->link->sh_offset;
165 s->hash = s->reloc, s->reloc = NULL;
166 tr = tcc_mallocz(nb_syms * sizeof *tr);
168 for (i = 0; i < nb_syms; ++i) {
169 ElfSym *sym = (ElfSym*)s->data + first_sym + i;
170 if (sym->st_shndx == SHN_UNDEF
171 && ELFW(ST_BIND)(sym->st_info) == STB_LOCAL)
172 sym->st_info = ELFW(ST_INFO)(STB_GLOBAL, ELFW(ST_TYPE)(sym->st_info));
173 tr[i] = set_elf_sym(s, sym->st_value, sym->st_size, sym->st_info,
174 sym->st_other, sym->st_shndx, s->link->data + sym->st_name);
176 /* now update relocations */
177 for (i = 1; i < s1->nb_sections; i++) {
178 Section *sr = s1->sections[i];
179 if (sr->sh_type == SHT_RELX && sr->link == s) {
180 ElfW_Rel *rel = (ElfW_Rel*)(sr->data + sr->sh_offset);
181 ElfW_Rel *rel_end = (ElfW_Rel*)(sr->data + sr->data_offset);
182 for (; rel < rel_end; ++rel) {
183 int n = ELFW(R_SYM)(rel->r_info) - first_sym;
184 //if (n < 0) tcc_error("internal: invalid symbol index in relocation");
185 rel->r_info = ELFW(R_INFO)(tr[n], ELFW(R_TYPE)(rel->r_info));
189 tcc_free(tr);
192 ST_FUNC Section *new_section(TCCState *s1, const char *name, int sh_type, int sh_flags)
194 Section *sec;
196 sec = tcc_mallocz(sizeof(Section) + strlen(name));
197 strcpy(sec->name, name);
198 sec->sh_type = sh_type;
199 sec->sh_flags = sh_flags;
200 switch(sh_type) {
201 case SHT_HASH:
202 case SHT_REL:
203 case SHT_RELA:
204 case SHT_DYNSYM:
205 case SHT_SYMTAB:
206 case SHT_DYNAMIC:
207 sec->sh_addralign = 4;
208 break;
209 case SHT_STRTAB:
210 sec->sh_addralign = 1;
211 break;
212 default:
213 sec->sh_addralign = PTR_SIZE; /* gcc/pcc default alignment */
214 break;
217 if (sh_flags & SHF_PRIVATE) {
218 dynarray_add(&s1->priv_sections, &s1->nb_priv_sections, sec);
219 } else {
220 sec->sh_num = s1->nb_sections;
221 dynarray_add(&s1->sections, &s1->nb_sections, sec);
224 return sec;
227 ST_FUNC Section *new_symtab(TCCState *s1,
228 const char *symtab_name, int sh_type, int sh_flags,
229 const char *strtab_name,
230 const char *hash_name, int hash_sh_flags)
232 Section *symtab, *strtab, *hash;
233 int *ptr, nb_buckets;
235 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
236 symtab->sh_entsize = sizeof(ElfW(Sym));
237 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
238 put_elf_str(strtab, "");
239 symtab->link = strtab;
240 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
242 nb_buckets = 1;
244 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
245 hash->sh_entsize = sizeof(int);
246 symtab->hash = hash;
247 hash->link = symtab;
249 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
250 ptr[0] = nb_buckets;
251 ptr[1] = 1;
252 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
253 return symtab;
256 /* realloc section and set its content to zero */
257 ST_FUNC void section_realloc(Section *sec, unsigned long new_size)
259 unsigned long size;
260 unsigned char *data;
262 size = sec->data_allocated;
263 if (size == 0)
264 size = 1;
265 while (size < new_size)
266 size = size * 2;
267 data = tcc_realloc(sec->data, size);
268 memset(data + sec->data_allocated, 0, size - sec->data_allocated);
269 sec->data = data;
270 sec->data_allocated = size;
273 /* reserve at least 'size' bytes aligned per 'align' in section
274 'sec' from current offset, and return the aligned offset */
275 ST_FUNC size_t section_add(Section *sec, addr_t size, int align)
277 size_t offset, offset1;
279 offset = (sec->data_offset + align - 1) & -align;
280 offset1 = offset + size;
281 if (sec->sh_type != SHT_NOBITS && offset1 > sec->data_allocated)
282 section_realloc(sec, offset1);
283 sec->data_offset = offset1;
284 if (align > sec->sh_addralign)
285 sec->sh_addralign = align;
286 return offset;
289 /* reserve at least 'size' bytes in section 'sec' from
290 sec->data_offset. */
291 ST_FUNC void *section_ptr_add(Section *sec, addr_t size)
293 size_t offset = section_add(sec, size, 1);
294 return sec->data + offset;
297 /* reserve at least 'size' bytes from section start */
298 ST_FUNC void section_reserve(Section *sec, unsigned long size)
300 if (size > sec->data_allocated)
301 section_realloc(sec, size);
302 if (size > sec->data_offset)
303 sec->data_offset = size;
306 /* return a reference to a section, and create it if it does not
307 exists */
308 ST_FUNC Section *find_section(TCCState *s1, const char *name)
310 Section *sec;
311 int i;
312 for(i = 1; i < s1->nb_sections; i++) {
313 sec = s1->sections[i];
314 if (!strcmp(name, sec->name))
315 return sec;
317 /* sections are created as PROGBITS */
318 return new_section(s1, name, SHT_PROGBITS, SHF_ALLOC);
321 /* ------------------------------------------------------------------------- */
323 ST_FUNC int put_elf_str(Section *s, const char *sym)
325 int offset, len;
326 char *ptr;
328 len = strlen(sym) + 1;
329 offset = s->data_offset;
330 ptr = section_ptr_add(s, len);
331 memmove(ptr, sym, len);
332 return offset;
335 /* elf symbol hashing function */
336 static unsigned long elf_hash(const unsigned char *name)
338 unsigned long h = 0, g;
340 while (*name) {
341 h = (h << 4) + *name++;
342 g = h & 0xf0000000;
343 if (g)
344 h ^= g >> 24;
345 h &= ~g;
347 return h;
350 /* rebuild hash table of section s */
351 /* NOTE: we do factorize the hash table code to go faster */
352 static void rebuild_hash(Section *s, unsigned int nb_buckets)
354 ElfW(Sym) *sym;
355 int *ptr, *hash, nb_syms, sym_index, h;
356 unsigned char *strtab;
358 strtab = s->link->data;
359 nb_syms = s->data_offset / sizeof(ElfW(Sym));
361 if (!nb_buckets)
362 nb_buckets = ((int*)s->hash->data)[0];
364 s->hash->data_offset = 0;
365 ptr = section_ptr_add(s->hash, (2 + nb_buckets + nb_syms) * sizeof(int));
366 ptr[0] = nb_buckets;
367 ptr[1] = nb_syms;
368 ptr += 2;
369 hash = ptr;
370 memset(hash, 0, (nb_buckets + 1) * sizeof(int));
371 ptr += nb_buckets + 1;
373 sym = (ElfW(Sym) *)s->data + 1;
374 for(sym_index = 1; sym_index < nb_syms; sym_index++) {
375 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
376 h = elf_hash(strtab + sym->st_name) % nb_buckets;
377 *ptr = hash[h];
378 hash[h] = sym_index;
379 } else {
380 *ptr = 0;
382 ptr++;
383 sym++;
387 /* return the symbol number */
388 ST_FUNC int put_elf_sym(Section *s, addr_t value, unsigned long size,
389 int info, int other, int shndx, const char *name)
391 int name_offset, sym_index;
392 int nbuckets, h;
393 ElfW(Sym) *sym;
394 Section *hs;
396 sym = section_ptr_add(s, sizeof(ElfW(Sym)));
397 if (name && name[0])
398 name_offset = put_elf_str(s->link, name);
399 else
400 name_offset = 0;
401 /* XXX: endianness */
402 sym->st_name = name_offset;
403 sym->st_value = value;
404 sym->st_size = size;
405 sym->st_info = info;
406 sym->st_other = other;
407 sym->st_shndx = shndx;
408 sym_index = sym - (ElfW(Sym) *)s->data;
409 hs = s->hash;
410 if (hs) {
411 int *ptr, *base;
412 ptr = section_ptr_add(hs, sizeof(int));
413 base = (int *)hs->data;
414 /* only add global or weak symbols. */
415 if (ELFW(ST_BIND)(info) != STB_LOCAL) {
416 /* add another hashing entry */
417 nbuckets = base[0];
418 h = elf_hash((unsigned char *)s->link->data + name_offset) % nbuckets;
419 *ptr = base[2 + h];
420 base[2 + h] = sym_index;
421 base[1]++;
422 /* we resize the hash table */
423 hs->nb_hashed_syms++;
424 if (hs->nb_hashed_syms > 2 * nbuckets) {
425 rebuild_hash(s, 2 * nbuckets);
427 } else {
428 *ptr = 0;
429 base[1]++;
432 return sym_index;
435 ST_FUNC int find_elf_sym(Section *s, const char *name)
437 ElfW(Sym) *sym;
438 Section *hs;
439 int nbuckets, sym_index, h;
440 const char *name1;
442 hs = s->hash;
443 if (!hs)
444 return 0;
445 nbuckets = ((int *)hs->data)[0];
446 h = elf_hash((unsigned char *) name) % nbuckets;
447 sym_index = ((int *)hs->data)[2 + h];
448 while (sym_index != 0) {
449 sym = &((ElfW(Sym) *)s->data)[sym_index];
450 name1 = (char *) s->link->data + sym->st_name;
451 if (!strcmp(name, name1))
452 return sym_index;
453 sym_index = ((int *)hs->data)[2 + nbuckets + sym_index];
455 return 0;
458 /* return elf symbol value, signal error if 'err' is nonzero */
459 ST_FUNC addr_t get_elf_sym_addr(TCCState *s, const char *name, int err)
461 int sym_index;
462 ElfW(Sym) *sym;
464 sym_index = find_elf_sym(s->symtab, name);
465 sym = &((ElfW(Sym) *)s->symtab->data)[sym_index];
466 if (!sym_index || sym->st_shndx == SHN_UNDEF) {
467 if (err)
468 tcc_error("%s not defined", name);
469 return 0;
471 return sym->st_value;
474 /* return elf symbol value */
475 LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
477 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 0);
480 #if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
481 /* return elf symbol value or error */
482 ST_FUNC void* tcc_get_symbol_err(TCCState *s, const char *name)
484 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 1);
486 #endif
488 /* add an elf symbol : check if it is already defined and patch
489 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
490 ST_FUNC int set_elf_sym(Section *s, addr_t value, unsigned long size,
491 int info, int other, int shndx, const char *name)
493 ElfW(Sym) *esym;
494 int sym_bind, sym_index, sym_type, esym_bind;
495 unsigned char sym_vis, esym_vis, new_vis;
497 sym_bind = ELFW(ST_BIND)(info);
498 sym_type = ELFW(ST_TYPE)(info);
499 sym_vis = ELFW(ST_VISIBILITY)(other);
501 if (sym_bind != STB_LOCAL) {
502 /* we search global or weak symbols */
503 sym_index = find_elf_sym(s, name);
504 if (!sym_index)
505 goto do_def;
506 esym = &((ElfW(Sym) *)s->data)[sym_index];
507 if (esym->st_value == value && esym->st_size == size && esym->st_info == info
508 && esym->st_other == other && esym->st_shndx == shndx)
509 return sym_index;
510 if (esym->st_shndx != SHN_UNDEF) {
511 esym_bind = ELFW(ST_BIND)(esym->st_info);
512 /* propagate the most constraining visibility */
513 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
514 esym_vis = ELFW(ST_VISIBILITY)(esym->st_other);
515 if (esym_vis == STV_DEFAULT) {
516 new_vis = sym_vis;
517 } else if (sym_vis == STV_DEFAULT) {
518 new_vis = esym_vis;
519 } else {
520 new_vis = (esym_vis < sym_vis) ? esym_vis : sym_vis;
522 esym->st_other = (esym->st_other & ~ELFW(ST_VISIBILITY)(-1))
523 | new_vis;
524 other = esym->st_other; /* in case we have to patch esym */
525 if (shndx == SHN_UNDEF) {
526 /* ignore adding of undefined symbol if the
527 corresponding symbol is already defined */
528 } else if (sym_bind == STB_GLOBAL && esym_bind == STB_WEAK) {
529 /* global overrides weak, so patch */
530 goto do_patch;
531 } else if (sym_bind == STB_WEAK && esym_bind == STB_GLOBAL) {
532 /* weak is ignored if already global */
533 } else if (sym_bind == STB_WEAK && esym_bind == STB_WEAK) {
534 /* keep first-found weak definition, ignore subsequents */
535 } else if (sym_vis == STV_HIDDEN || sym_vis == STV_INTERNAL) {
536 /* ignore hidden symbols after */
537 } else if ((esym->st_shndx == SHN_COMMON
538 || esym->st_shndx == bss_section->sh_num)
539 && (shndx < SHN_LORESERVE
540 && shndx != bss_section->sh_num)) {
541 /* data symbol gets precedence over common/bss */
542 goto do_patch;
543 } else if (shndx == SHN_COMMON || shndx == bss_section->sh_num) {
544 /* data symbol keeps precedence over common/bss */
545 } else if (s->sh_flags & SHF_DYNSYM) {
546 /* we accept that two DLL define the same symbol */
547 } else if (esym->st_other & ST_ASM_SET) {
548 /* If the existing symbol came from an asm .set
549 we can override. */
550 goto do_patch;
551 } else {
552 #if 0
553 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
554 sym_bind, shndx, new_vis, esym_bind, esym->st_shndx, esym_vis);
555 #endif
556 tcc_error_noabort("'%s' defined twice", name);
558 } else {
559 do_patch:
560 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
561 esym->st_shndx = shndx;
562 new_undef_sym = 1;
563 esym->st_value = value;
564 esym->st_size = size;
565 esym->st_other = other;
567 } else {
568 do_def:
569 sym_index = put_elf_sym(s, value, size,
570 ELFW(ST_INFO)(sym_bind, sym_type), other,
571 shndx, name);
573 return sym_index;
576 /* put relocation */
577 ST_FUNC void put_elf_reloca(Section *symtab, Section *s, unsigned long offset,
578 int type, int symbol, addr_t addend)
580 char buf[256];
581 Section *sr;
582 ElfW_Rel *rel;
584 sr = s->reloc;
585 if (!sr) {
586 /* if no relocation section, create it */
587 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
588 /* if the symtab is allocated, then we consider the relocation
589 are also */
590 sr = new_section(tcc_state, buf, SHT_RELX, symtab->sh_flags);
591 sr->sh_entsize = sizeof(ElfW_Rel);
592 sr->link = symtab;
593 sr->sh_info = s->sh_num;
594 s->reloc = sr;
596 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
597 rel->r_offset = offset;
598 rel->r_info = ELFW(R_INFO)(symbol, type);
599 #if SHT_RELX == SHT_RELA
600 rel->r_addend = addend;
601 #else
602 if (addend)
603 tcc_error("non-zero addend on REL architecture");
604 #endif
607 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
608 int type, int symbol)
610 put_elf_reloca(symtab, s, offset, type, symbol, 0);
613 /* Remove relocations for section S->reloc starting at oldrelocoffset
614 that are to the same place, retaining the last of them. As side effect
615 the relocations are sorted. Possibly reduces the number of relocs. */
616 ST_FUNC void squeeze_multi_relocs(Section *s, size_t oldrelocoffset)
618 Section *sr = s->reloc;
619 ElfW_Rel *r, *dest;
620 ssize_t a;
621 ElfW(Addr) addr;
623 if (oldrelocoffset + sizeof(*r) >= sr->data_offset)
624 return;
625 /* The relocs we're dealing with are the result of initializer parsing.
626 So they will be mostly in order and there aren't many of them.
627 Secondly we need a stable sort (which qsort isn't). We use
628 a simple insertion sort. */
629 for (a = oldrelocoffset + sizeof(*r); a < sr->data_offset; a += sizeof(*r)) {
630 ssize_t i = a - sizeof(*r);
631 addr = ((ElfW_Rel*)(sr->data + a))->r_offset;
632 for (; i >= (ssize_t)oldrelocoffset &&
633 ((ElfW_Rel*)(sr->data + i))->r_offset > addr; i -= sizeof(*r)) {
634 ElfW_Rel tmp = *(ElfW_Rel*)(sr->data + a);
635 *(ElfW_Rel*)(sr->data + a) = *(ElfW_Rel*)(sr->data + i);
636 *(ElfW_Rel*)(sr->data + i) = tmp;
640 r = (ElfW_Rel*)(sr->data + oldrelocoffset);
641 dest = r;
642 for (; r < (ElfW_Rel*)(sr->data + sr->data_offset); r++) {
643 if (dest->r_offset != r->r_offset)
644 dest++;
645 *dest = *r;
647 sr->data_offset = (unsigned char*)dest - sr->data + sizeof(*r);
650 /* put stab debug information */
652 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
653 unsigned long value)
655 Stab_Sym *sym;
657 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
658 if (str) {
659 sym->n_strx = put_elf_str(stabstr_section, str);
660 } else {
661 sym->n_strx = 0;
663 sym->n_type = type;
664 sym->n_other = other;
665 sym->n_desc = desc;
666 sym->n_value = value;
669 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
670 unsigned long value, Section *sec, int sym_index)
672 put_stabs(str, type, other, desc, value);
673 put_elf_reloc(symtab_section, stab_section,
674 stab_section->data_offset - sizeof(unsigned int),
675 R_DATA_32, sym_index);
678 ST_FUNC void put_stabn(int type, int other, int desc, int value)
680 put_stabs(NULL, type, other, desc, value);
683 ST_FUNC void put_stabd(int type, int other, int desc)
685 put_stabs(NULL, type, other, desc, 0);
688 ST_FUNC struct sym_attr *get_sym_attr(TCCState *s1, int index, int alloc)
690 int n;
691 struct sym_attr *tab;
693 if (index >= s1->nb_sym_attrs) {
694 if (!alloc)
695 return s1->sym_attrs;
696 /* find immediately bigger power of 2 and reallocate array */
697 n = 1;
698 while (index >= n)
699 n *= 2;
700 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
701 s1->sym_attrs = tab;
702 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
703 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
704 s1->nb_sym_attrs = n;
706 return &s1->sym_attrs[index];
709 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
710 using variable <elem> */
711 #define for_each_elem(sec, startoff, elem, type) \
712 for (elem = (type *) sec->data + startoff; \
713 elem < (type *) (sec->data + sec->data_offset); elem++)
715 /* In an ELF file symbol table, the local symbols must appear below
716 the global and weak ones. Since TCC cannot sort it while generating
717 the code, we must do it after. All the relocation tables are also
718 modified to take into account the symbol table sorting */
719 static void sort_syms(TCCState *s1, Section *s)
721 int *old_to_new_syms;
722 ElfW(Sym) *new_syms;
723 int nb_syms, i;
724 ElfW(Sym) *p, *q;
725 ElfW_Rel *rel;
726 Section *sr;
727 int type, sym_index;
729 nb_syms = s->data_offset / sizeof(ElfW(Sym));
730 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
731 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
733 /* first pass for local symbols */
734 p = (ElfW(Sym) *)s->data;
735 q = new_syms;
736 for(i = 0; i < nb_syms; i++) {
737 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
738 old_to_new_syms[i] = q - new_syms;
739 *q++ = *p;
741 p++;
743 /* save the number of local symbols in section header */
744 if( s->sh_size ) /* this 'if' makes IDA happy */
745 s->sh_info = q - new_syms;
747 /* then second pass for non local symbols */
748 p = (ElfW(Sym) *)s->data;
749 for(i = 0; i < nb_syms; i++) {
750 if (ELFW(ST_BIND)(p->st_info) != STB_LOCAL) {
751 old_to_new_syms[i] = q - new_syms;
752 *q++ = *p;
754 p++;
757 /* we copy the new symbols to the old */
758 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
759 tcc_free(new_syms);
761 /* now we modify all the relocations */
762 for(i = 1; i < s1->nb_sections; i++) {
763 sr = s1->sections[i];
764 if (sr->sh_type == SHT_RELX && sr->link == s) {
765 for_each_elem(sr, 0, rel, ElfW_Rel) {
766 sym_index = ELFW(R_SYM)(rel->r_info);
767 type = ELFW(R_TYPE)(rel->r_info);
768 sym_index = old_to_new_syms[sym_index];
769 rel->r_info = ELFW(R_INFO)(sym_index, type);
774 tcc_free(old_to_new_syms);
777 /* relocate symbol table, resolve undefined symbols if do_resolve is
778 true and output error if undefined symbol. */
779 ST_FUNC void relocate_syms(TCCState *s1, Section *symtab, int do_resolve)
781 ElfW(Sym) *sym;
782 int sym_bind, sh_num;
783 const char *name;
785 for_each_elem(symtab, 1, sym, ElfW(Sym)) {
786 sh_num = sym->st_shndx;
787 if (sh_num == SHN_UNDEF) {
788 name = (char *) s1->symtab->link->data + sym->st_name;
789 /* Use ld.so to resolve symbol for us (for tcc -run) */
790 if (do_resolve) {
791 #if defined TCC_IS_NATIVE && !defined TCC_TARGET_PE
792 void *addr = dlsym(RTLD_DEFAULT, name);
793 if (addr) {
794 sym->st_value = (addr_t) addr;
795 #ifdef DEBUG_RELOC
796 printf ("relocate_sym: %s -> 0x%lx\n", name, sym->st_value);
797 #endif
798 goto found;
800 #endif
801 /* if dynamic symbol exist, it will be used in relocate_section */
802 } else if (s1->dynsym && find_elf_sym(s1->dynsym, name))
803 goto found;
804 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
805 it */
806 if (!strcmp(name, "_fp_hw"))
807 goto found;
808 /* only weak symbols are accepted to be undefined. Their
809 value is zero */
810 sym_bind = ELFW(ST_BIND)(sym->st_info);
811 if (sym_bind == STB_WEAK)
812 sym->st_value = 0;
813 else
814 tcc_error_noabort("undefined symbol '%s'", name);
815 } else if (sh_num < SHN_LORESERVE) {
816 /* add section base */
817 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
819 found: ;
823 /* relocate a given section (CPU dependent) by applying the relocations
824 in the associated relocation section */
825 ST_FUNC void relocate_section(TCCState *s1, Section *s)
827 Section *sr = s->reloc;
828 ElfW_Rel *rel;
829 ElfW(Sym) *sym;
830 int type, sym_index;
831 unsigned char *ptr;
832 addr_t tgt, addr;
834 relocate_init(sr);
836 for_each_elem(sr, 0, rel, ElfW_Rel) {
837 ptr = s->data + rel->r_offset;
838 sym_index = ELFW(R_SYM)(rel->r_info);
839 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
840 type = ELFW(R_TYPE)(rel->r_info);
841 tgt = sym->st_value;
842 #if SHT_RELX == SHT_RELA
843 tgt += rel->r_addend;
844 #endif
845 addr = s->sh_addr + rel->r_offset;
846 relocate(s1, rel, type, ptr, addr, tgt);
848 /* if the relocation is allocated, we change its symbol table */
849 if (sr->sh_flags & SHF_ALLOC)
850 sr->link = s1->dynsym;
853 /* relocate relocation table in 'sr' */
854 static void relocate_rel(TCCState *s1, Section *sr)
856 Section *s;
857 ElfW_Rel *rel;
859 s = s1->sections[sr->sh_info];
860 for_each_elem(sr, 0, rel, ElfW_Rel)
861 rel->r_offset += s->sh_addr;
864 /* count the number of dynamic relocations so that we can reserve
865 their space */
866 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
868 ElfW_Rel *rel;
869 int sym_index, type, count;
871 count = 0;
872 for_each_elem(sr, 0, rel, ElfW_Rel) {
873 sym_index = ELFW(R_SYM)(rel->r_info);
874 type = ELFW(R_TYPE)(rel->r_info);
875 switch(type) {
876 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
877 #if defined(TCC_TARGET_I386)
878 case R_386_32:
879 if (!get_sym_attr(s1, sym_index, 0)->dyn_index
880 && ((ElfW(Sym)*)symtab_section->data + sym_index)->st_shndx == SHN_UNDEF) {
881 /* don't fixup unresolved (weak) symbols */
882 rel->r_info = ELFW(R_INFO)(sym_index, R_386_RELATIVE);
883 break;
885 #elif defined(TCC_TARGET_X86_64)
886 case R_X86_64_32:
887 case R_X86_64_32S:
888 case R_X86_64_64:
889 #endif
890 count++;
891 break;
892 #if defined(TCC_TARGET_I386)
893 case R_386_PC32:
894 #elif defined(TCC_TARGET_X86_64)
895 case R_X86_64_PC32:
896 #endif
897 if (get_sym_attr(s1, sym_index, 0)->dyn_index)
898 count++;
899 break;
900 #endif
901 default:
902 break;
905 if (count) {
906 /* allocate the section */
907 sr->sh_flags |= SHF_ALLOC;
908 sr->sh_size = count * sizeof(ElfW_Rel);
910 return count;
913 static void build_got(TCCState *s1)
915 /* if no got, then create it */
916 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
917 s1->got->sh_entsize = 4;
918 set_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
919 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
920 /* keep space for _DYNAMIC pointer and two dummy got entries */
921 section_ptr_add(s1->got, 3 * PTR_SIZE);
924 /* Create a GOT and (for function call) a PLT entry corresponding to a symbol
925 in s1->symtab. When creating the dynamic symbol table entry for the GOT
926 relocation, use 'size' and 'info' for the corresponding symbol metadata.
927 Returns the offset of the GOT or (if any) PLT entry. */
928 static struct sym_attr * put_got_entry(TCCState *s1, int dyn_reloc_type,
929 unsigned long size,
930 int info, int sym_index)
932 int need_plt_entry;
933 const char *name;
934 ElfW(Sym) *sym;
935 struct sym_attr *attr;
936 unsigned got_offset;
937 char plt_name[100];
938 int len;
940 need_plt_entry = (dyn_reloc_type == R_JMP_SLOT);
941 attr = get_sym_attr(s1, sym_index, 1);
943 /* In case a function is both called and its address taken 2 GOT entries
944 are created, one for taking the address (GOT) and the other for the PLT
945 entry (PLTGOT). */
946 if (need_plt_entry ? attr->plt_offset : attr->got_offset)
947 return attr;
949 /* create the GOT entry */
950 got_offset = s1->got->data_offset;
951 section_ptr_add(s1->got, PTR_SIZE);
953 /* Create the GOT relocation that will insert the address of the object or
954 function of interest in the GOT entry. This is a static relocation for
955 memory output (dlsym will give us the address of symbols) and dynamic
956 relocation otherwise (executable and DLLs). The relocation should be
957 done lazily for GOT entry with *_JUMP_SLOT relocation type (the one
958 associated to a PLT entry) but is currently done at load time for an
959 unknown reason. */
961 sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
962 name = (char *) symtab_section->link->data + sym->st_name;
964 if (s1->dynsym) {
965 if (ELFW(ST_BIND)(sym->st_info) == STB_LOCAL) {
966 /* Hack alarm. We don't want to emit dynamic symbols
967 and symbol based relocs for STB_LOCAL symbols, but rather
968 want to resolve them directly. At this point the symbol
969 values aren't final yet, so we must defer this. We will later
970 have to create a RELATIVE reloc anyway, so we misuse the
971 relocation slot to smuggle the symbol reference until
972 fill_local_got_entries. Not that the sym_index is
973 relative to symtab_section, not s1->dynsym! Nevertheless
974 we use s1->dyn_sym so that if this is the first call
975 that got->reloc is correctly created. Also note that
976 RELATIVE relocs are not normally created for the .got,
977 so the types serves as a marker for later (and is retained
978 also for the final output, which is okay because then the
979 got is just normal data). */
980 put_elf_reloc(s1->dynsym, s1->got, got_offset, R_RELATIVE,
981 sym_index);
982 } else {
983 if (0 == attr->dyn_index)
984 attr->dyn_index = set_elf_sym(s1->dynsym, sym->st_value, size,
985 info, 0, sym->st_shndx, name);
986 put_elf_reloc(s1->dynsym, s1->got, got_offset, dyn_reloc_type,
987 attr->dyn_index);
989 } else {
990 put_elf_reloc(symtab_section, s1->got, got_offset, dyn_reloc_type,
991 sym_index);
994 if (need_plt_entry) {
995 if (!s1->plt) {
996 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
997 SHF_ALLOC | SHF_EXECINSTR);
998 s1->plt->sh_entsize = 4;
1001 attr->plt_offset = create_plt_entry(s1, got_offset, attr);
1003 /* create a symbol 'sym@plt' for the PLT jump vector */
1004 len = strlen(name);
1005 if (len > sizeof plt_name - 5)
1006 len = sizeof plt_name - 5;
1007 memcpy(plt_name, name, len);
1008 strcpy(plt_name + len, "@plt");
1009 attr->plt_sym = put_elf_sym(s1->symtab, attr->plt_offset, sym->st_size,
1010 ELFW(ST_INFO)(STB_GLOBAL, STT_FUNC), 0, s1->plt->sh_num, plt_name);
1012 } else {
1013 attr->got_offset = got_offset;
1016 return attr;
1019 /* build GOT and PLT entries */
1020 ST_FUNC void build_got_entries(TCCState *s1)
1022 Section *s;
1023 ElfW_Rel *rel;
1024 ElfW(Sym) *sym;
1025 int i, type, gotplt_entry, reloc_type, sym_index;
1026 struct sym_attr *attr;
1028 for(i = 1; i < s1->nb_sections; i++) {
1029 s = s1->sections[i];
1030 if (s->sh_type != SHT_RELX)
1031 continue;
1032 /* no need to handle got relocations */
1033 if (s->link != symtab_section)
1034 continue;
1035 for_each_elem(s, 0, rel, ElfW_Rel) {
1036 type = ELFW(R_TYPE)(rel->r_info);
1037 gotplt_entry = gotplt_entry_type(type);
1038 sym_index = ELFW(R_SYM)(rel->r_info);
1039 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1041 if (gotplt_entry == NO_GOTPLT_ENTRY) {
1042 continue;
1045 /* Automatically create PLT/GOT [entry] if it is an undefined
1046 reference (resolved at runtime), or the symbol is absolute,
1047 probably created by tcc_add_symbol, and thus on 64-bit
1048 targets might be too far from application code. */
1049 if (gotplt_entry == AUTO_GOTPLT_ENTRY) {
1050 if (sym->st_shndx == SHN_UNDEF) {
1051 ElfW(Sym) *esym;
1052 int dynindex;
1053 if (s1->output_type == TCC_OUTPUT_DLL && ! PCRELATIVE_DLLPLT)
1054 continue;
1055 /* Relocations for UNDEF symbols would normally need
1056 to be transferred into the executable or shared object.
1057 If that were done AUTO_GOTPLT_ENTRY wouldn't exist.
1058 But TCC doesn't do that (at least for exes), so we
1059 need to resolve all such relocs locally. And that
1060 means PLT slots for functions in DLLs and COPY relocs for
1061 data symbols. COPY relocs were generated in
1062 bind_exe_dynsyms (and the symbol adjusted to be defined),
1063 and for functions we were generated a dynamic symbol
1064 of function type. */
1065 if (s1->dynsym) {
1066 /* dynsym isn't set for -run :-/ */
1067 dynindex = get_sym_attr(s1, sym_index, 0)->dyn_index;
1068 esym = (ElfW(Sym) *)s1->dynsym->data + dynindex;
1069 if (dynindex
1070 && (ELFW(ST_TYPE)(esym->st_info) == STT_FUNC
1071 || (ELFW(ST_TYPE)(esym->st_info) == STT_NOTYPE
1072 && ELFW(ST_TYPE)(sym->st_info) == STT_FUNC)))
1073 goto jmp_slot;
1075 } else if (!(sym->st_shndx == SHN_ABS
1076 #ifndef TCC_TARGET_ARM
1077 && PTR_SIZE == 8
1078 #endif
1080 continue;
1083 #ifdef TCC_TARGET_X86_64
1084 if ((type == R_X86_64_PLT32 || type == R_X86_64_PC32) &&
1085 (ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT ||
1086 ELFW(ST_BIND)(sym->st_info) == STB_LOCAL)) {
1087 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1088 continue;
1090 #endif
1091 if (code_reloc(type)) {
1092 jmp_slot:
1093 reloc_type = R_JMP_SLOT;
1094 } else
1095 reloc_type = R_GLOB_DAT;
1097 if (!s1->got)
1098 build_got(s1);
1100 if (gotplt_entry == BUILD_GOT_ONLY)
1101 continue;
1103 attr = put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1104 sym_index);
1106 if (reloc_type == R_JMP_SLOT)
1107 rel->r_info = ELFW(R_INFO)(attr->plt_sym, type);
1112 /* put dynamic tag */
1113 static void put_dt(Section *dynamic, int dt, addr_t val)
1115 ElfW(Dyn) *dyn;
1116 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1117 dyn->d_tag = dt;
1118 dyn->d_un.d_val = val;
1121 #ifndef TCC_TARGET_PE
1122 static void add_init_array_defines(TCCState *s1, const char *section_name)
1124 Section *s;
1125 long end_offset;
1126 char sym_start[1024];
1127 char sym_end[1024];
1129 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1130 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1132 s = find_section(s1, section_name);
1133 if (!s) {
1134 end_offset = 0;
1135 s = data_section;
1136 } else {
1137 end_offset = s->data_offset;
1140 set_elf_sym(symtab_section,
1141 0, 0,
1142 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1143 s->sh_num, sym_start);
1144 set_elf_sym(symtab_section,
1145 end_offset, 0,
1146 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1147 s->sh_num, sym_end);
1150 static int tcc_add_support(TCCState *s1, const char *filename)
1152 char buf[1024];
1153 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1154 return tcc_add_file(s1, buf);
1156 #endif
1158 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1160 #ifdef CONFIG_TCC_BCHECK
1161 addr_t *ptr;
1162 int sym_index;
1164 if (0 == s1->do_bounds_check)
1165 return;
1166 /* XXX: add an object file to do that */
1167 ptr = section_ptr_add(bounds_section, sizeof(*ptr));
1168 *ptr = 0;
1169 set_elf_sym(symtab_section, 0, 0,
1170 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1171 bounds_section->sh_num, "__bounds_start");
1172 /* pull bcheck.o from libtcc1.a */
1173 sym_index = set_elf_sym(symtab_section, 0, 0,
1174 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1175 SHN_UNDEF, "__bound_init");
1176 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1177 /* add 'call __bound_init()' in .init section */
1178 Section *init_section = find_section(s1, ".init");
1179 unsigned char *pinit = section_ptr_add(init_section, 5);
1180 pinit[0] = 0xe8;
1181 write32le(pinit + 1, -4);
1182 put_elf_reloc(symtab_section, init_section,
1183 init_section->data_offset - 4, R_386_PC32, sym_index);
1184 /* R_386_PC32 = R_X86_64_PC32 = 2 */
1186 #endif
1189 /* add tcc runtime libraries */
1190 ST_FUNC void tcc_add_runtime(TCCState *s1)
1192 s1->filetype = 0;
1193 tcc_add_bcheck(s1);
1194 tcc_add_pragma_libs(s1);
1195 #ifndef TCC_TARGET_PE
1196 /* add libc */
1197 if (!s1->nostdlib) {
1198 tcc_add_library_err(s1, "c");
1199 #ifdef TCC_LIBGCC
1200 if (!s1->static_link) {
1201 if (TCC_LIBGCC[0] == '/')
1202 tcc_add_file(s1, TCC_LIBGCC);
1203 else
1204 tcc_add_dll(s1, TCC_LIBGCC, 0);
1206 #endif
1207 tcc_add_support(s1, TCC_LIBTCC1);
1208 /* add crt end if not memory output */
1209 if (s1->output_type != TCC_OUTPUT_MEMORY)
1210 tcc_add_crt(s1, "crtn.o");
1212 #endif
1215 /* add various standard linker symbols (must be done after the
1216 sections are filled (for example after allocating common
1217 symbols)) */
1218 static void tcc_add_linker_symbols(TCCState *s1)
1220 char buf[1024];
1221 int i;
1222 Section *s;
1224 set_elf_sym(symtab_section,
1225 text_section->data_offset, 0,
1226 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1227 text_section->sh_num, "_etext");
1228 set_elf_sym(symtab_section,
1229 data_section->data_offset, 0,
1230 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1231 data_section->sh_num, "_edata");
1232 set_elf_sym(symtab_section,
1233 bss_section->data_offset, 0,
1234 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1235 bss_section->sh_num, "_end");
1236 #ifndef TCC_TARGET_PE
1237 /* horrible new standard ldscript defines */
1238 add_init_array_defines(s1, ".preinit_array");
1239 add_init_array_defines(s1, ".init_array");
1240 add_init_array_defines(s1, ".fini_array");
1241 #endif
1243 /* add start and stop symbols for sections whose name can be
1244 expressed in C */
1245 for(i = 1; i < s1->nb_sections; i++) {
1246 s = s1->sections[i];
1247 if (s->sh_type == SHT_PROGBITS &&
1248 (s->sh_flags & SHF_ALLOC)) {
1249 const char *p;
1250 int ch;
1252 /* check if section name can be expressed in C */
1253 p = s->name;
1254 for(;;) {
1255 ch = *p;
1256 if (!ch)
1257 break;
1258 if (!isid(ch) && !isnum(ch))
1259 goto next_sec;
1260 p++;
1262 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1263 set_elf_sym(symtab_section,
1264 0, 0,
1265 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1266 s->sh_num, buf);
1267 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1268 set_elf_sym(symtab_section,
1269 s->data_offset, 0,
1270 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1271 s->sh_num, buf);
1273 next_sec: ;
1277 ST_FUNC void resolve_common_syms(TCCState *s1)
1279 ElfW(Sym) *sym;
1281 /* Allocate common symbols in BSS. */
1282 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1283 if (sym->st_shndx == SHN_COMMON) {
1284 /* symbol alignment is in st_value for SHN_COMMONs */
1285 sym->st_value = section_add(bss_section, sym->st_size,
1286 sym->st_value);
1287 sym->st_shndx = bss_section->sh_num;
1291 /* Now assign linker provided symbols their value. */
1292 tcc_add_linker_symbols(s1);
1295 static void tcc_output_binary(TCCState *s1, FILE *f,
1296 const int *sec_order)
1298 Section *s;
1299 int i, offset, size;
1301 offset = 0;
1302 for(i=1;i<s1->nb_sections;i++) {
1303 s = s1->sections[sec_order[i]];
1304 if (s->sh_type != SHT_NOBITS &&
1305 (s->sh_flags & SHF_ALLOC)) {
1306 while (offset < s->sh_offset) {
1307 fputc(0, f);
1308 offset++;
1310 size = s->sh_size;
1311 fwrite(s->data, 1, size, f);
1312 offset += size;
1317 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1319 int sym_index = ELFW(R_SYM) (rel->r_info);
1320 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1321 struct sym_attr *attr = get_sym_attr(s1, sym_index, 0);
1322 unsigned offset = attr->got_offset;
1324 if (0 == offset)
1325 return;
1326 section_reserve(s1->got, offset + PTR_SIZE);
1327 #ifdef TCC_TARGET_X86_64
1328 write64le(s1->got->data + offset, sym->st_value);
1329 #else
1330 write32le(s1->got->data + offset, sym->st_value);
1331 #endif
1334 /* Perform relocation to GOT or PLT entries */
1335 ST_FUNC void fill_got(TCCState *s1)
1337 Section *s;
1338 ElfW_Rel *rel;
1339 int i;
1341 for(i = 1; i < s1->nb_sections; i++) {
1342 s = s1->sections[i];
1343 if (s->sh_type != SHT_RELX)
1344 continue;
1345 /* no need to handle got relocations */
1346 if (s->link != symtab_section)
1347 continue;
1348 for_each_elem(s, 0, rel, ElfW_Rel) {
1349 switch (ELFW(R_TYPE) (rel->r_info)) {
1350 case R_X86_64_GOT32:
1351 case R_X86_64_GOTPCREL:
1352 case R_X86_64_GOTPCRELX:
1353 case R_X86_64_REX_GOTPCRELX:
1354 case R_X86_64_PLT32:
1355 fill_got_entry(s1, rel);
1356 break;
1362 /* See put_got_entry for a description. This is the second stage
1363 where GOT references to local defined symbols are rewritten. */
1364 static void fill_local_got_entries(TCCState *s1)
1366 ElfW_Rel *rel;
1367 for_each_elem(s1->got->reloc, 0, rel, ElfW_Rel) {
1368 if (ELFW(R_TYPE)(rel->r_info) == R_RELATIVE) {
1369 int sym_index = ELFW(R_SYM) (rel->r_info);
1370 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1371 struct sym_attr *attr = get_sym_attr(s1, sym_index, 0);
1372 unsigned offset = attr->got_offset;
1373 if (offset != rel->r_offset - s1->got->sh_addr)
1374 tcc_error_noabort("huh");
1375 rel->r_info = ELFW(R_INFO)(0, R_RELATIVE);
1376 #if SHT_RELX == SHT_RELA
1377 rel->r_addend = sym->st_value;
1378 #else
1379 /* All our REL architectures also happen to be 32bit LE. */
1380 write32le(s1->got->data + offset, sym->st_value);
1381 #endif
1386 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1387 in shared libraries and export non local defined symbols to shared libraries
1388 if -rdynamic switch was given on command line */
1389 static void bind_exe_dynsyms(TCCState *s1)
1391 const char *name;
1392 int sym_index, index;
1393 ElfW(Sym) *sym, *esym;
1394 int type;
1396 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1397 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1398 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1399 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1400 if (sym->st_shndx == SHN_UNDEF) {
1401 name = (char *) symtab_section->link->data + sym->st_name;
1402 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1403 if (sym_index) {
1404 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1405 type = ELFW(ST_TYPE)(esym->st_info);
1406 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1407 /* Indirect functions shall have STT_FUNC type in executable
1408 * dynsym section. Indeed, a dlsym call following a lazy
1409 * resolution would pick the symbol value from the
1410 * executable dynsym entry which would contain the address
1411 * of the function wanted by the caller of dlsym instead of
1412 * the address of the function that would return that
1413 * address */
1414 int dynindex
1415 = put_elf_sym(s1->dynsym, 0, esym->st_size,
1416 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC), 0, 0,
1417 name);
1418 int index = sym - (ElfW(Sym) *) symtab_section->data;
1419 get_sym_attr(s1, index, 1)->dyn_index = dynindex;
1420 } else if (type == STT_OBJECT) {
1421 unsigned long offset;
1422 ElfW(Sym) *dynsym;
1423 offset = bss_section->data_offset;
1424 /* XXX: which alignment ? */
1425 offset = (offset + 16 - 1) & -16;
1426 set_elf_sym (s1->symtab, offset, esym->st_size,
1427 esym->st_info, 0, bss_section->sh_num, name);
1428 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1429 esym->st_info, 0, bss_section->sh_num,
1430 name);
1432 /* Ensure R_COPY works for weak symbol aliases */
1433 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1434 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1435 if ((dynsym->st_value == esym->st_value)
1436 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1437 char *dynname = (char *) s1->dynsymtab_section->link->data
1438 + dynsym->st_name;
1439 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1440 dynsym->st_info, 0,
1441 bss_section->sh_num, dynname);
1442 break;
1447 put_elf_reloc(s1->dynsym, bss_section,
1448 offset, R_COPY, index);
1449 offset += esym->st_size;
1450 bss_section->data_offset = offset;
1452 } else {
1453 /* STB_WEAK undefined symbols are accepted */
1454 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1455 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1456 !strcmp(name, "_fp_hw")) {
1457 } else {
1458 tcc_error_noabort("undefined symbol '%s'", name);
1461 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1462 /* if -rdynamic option, then export all non local symbols */
1463 name = (char *) symtab_section->link->data + sym->st_name;
1464 set_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1465 0, sym->st_shndx, name);
1470 /* Bind symbols of libraries: export all non local symbols of executable that
1471 are referenced by shared libraries. The reason is that the dynamic loader
1472 search symbol first in executable and then in libraries. Therefore a
1473 reference to a symbol already defined by a library can still be resolved by
1474 a symbol in the executable. */
1475 static void bind_libs_dynsyms(TCCState *s1)
1477 const char *name;
1478 int sym_index;
1479 ElfW(Sym) *sym, *esym;
1481 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1482 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1483 sym_index = find_elf_sym(symtab_section, name);
1484 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1485 if (sym_index && sym->st_shndx != SHN_UNDEF
1486 && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1487 set_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1488 sym->st_info, 0, sym->st_shndx, name);
1489 } else if (esym->st_shndx == SHN_UNDEF) {
1490 /* weak symbols can stay undefined */
1491 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1492 tcc_warning("undefined dynamic symbol '%s'", name);
1497 /* Export all non local symbols. This is used by shared libraries so that the
1498 non local symbols they define can resolve a reference in another shared
1499 library or in the executable. Correspondingly, it allows undefined local
1500 symbols to be resolved by other shared libraries or by the executable. */
1501 static void export_global_syms(TCCState *s1)
1503 int dynindex, index;
1504 const char *name;
1505 ElfW(Sym) *sym;
1507 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1508 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1509 name = (char *) symtab_section->link->data + sym->st_name;
1510 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1511 sym->st_info, 0, sym->st_shndx, name);
1512 index = sym - (ElfW(Sym) *) symtab_section->data;
1513 get_sym_attr(s1, index, 1)->dyn_index = dynindex;
1518 /* Allocate strings for section names and decide if an unallocated section
1519 should be output.
1520 NOTE: the strsec section comes last, so its size is also correct ! */
1521 static int alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1523 int i;
1524 Section *s;
1525 int textrel = 0;
1527 /* Allocate strings for section names */
1528 for(i = 1; i < s1->nb_sections; i++) {
1529 s = s1->sections[i];
1530 /* when generating a DLL, we include relocations but we may
1531 patch them */
1532 if (file_type == TCC_OUTPUT_DLL &&
1533 s->sh_type == SHT_RELX &&
1534 !(s->sh_flags & SHF_ALLOC) &&
1535 (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC) &&
1536 prepare_dynamic_rel(s1, s)) {
1537 if (s1->sections[s->sh_info]->sh_flags & SHF_EXECINSTR)
1538 textrel = 1;
1539 } else if ((s1->do_debug && s->sh_type != SHT_RELX) ||
1540 file_type == TCC_OUTPUT_OBJ ||
1541 (s->sh_flags & SHF_ALLOC) ||
1542 i == (s1->nb_sections - 1)) {
1543 /* we output all sections if debug or object file */
1544 s->sh_size = s->data_offset;
1546 if (s->sh_size || (s->sh_flags & SHF_ALLOC))
1547 s->sh_name = put_elf_str(strsec, s->name);
1549 strsec->sh_size = strsec->data_offset;
1550 return textrel;
1553 /* Info to be copied in dynamic section */
1554 struct dyn_inf {
1555 Section *dynamic;
1556 Section *dynstr;
1557 unsigned long data_offset;
1558 addr_t rel_addr;
1559 addr_t rel_size;
1560 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1561 addr_t bss_addr;
1562 addr_t bss_size;
1563 #endif
1566 /* Assign sections to segments and decide how are sections laid out when loaded
1567 in memory. This function also fills corresponding program headers. */
1568 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1569 Section *interp, Section* strsec,
1570 struct dyn_inf *dyninf, int *sec_order)
1572 int i, j, k, file_type, sh_order_index, file_offset;
1573 unsigned long s_align;
1574 long long tmp;
1575 addr_t addr;
1576 ElfW(Phdr) *ph;
1577 Section *s;
1579 file_type = s1->output_type;
1580 sh_order_index = 1;
1581 file_offset = 0;
1582 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1583 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1584 s_align = ELF_PAGE_SIZE;
1585 if (s1->section_align)
1586 s_align = s1->section_align;
1588 if (phnum > 0) {
1589 if (s1->has_text_addr) {
1590 int a_offset, p_offset;
1591 addr = s1->text_addr;
1592 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1593 ELF_PAGE_SIZE */
1594 a_offset = (int) (addr & (s_align - 1));
1595 p_offset = file_offset & (s_align - 1);
1596 if (a_offset < p_offset)
1597 a_offset += s_align;
1598 file_offset += (a_offset - p_offset);
1599 } else {
1600 if (file_type == TCC_OUTPUT_DLL)
1601 addr = 0;
1602 else
1603 addr = ELF_START_ADDR;
1604 /* compute address after headers */
1605 addr += (file_offset & (s_align - 1));
1608 ph = &phdr[0];
1609 /* Leave one program headers for the program interpreter and one for
1610 the program header table itself if needed. These are done later as
1611 they require section layout to be done first. */
1612 if (interp)
1613 ph += 2;
1615 /* dynamic relocation table information, for .dynamic section */
1616 dyninf->rel_addr = dyninf->rel_size = 0;
1617 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1618 dyninf->bss_addr = dyninf->bss_size = 0;
1619 #endif
1621 for(j = 0; j < 2; j++) {
1622 ph->p_type = PT_LOAD;
1623 if (j == 0)
1624 ph->p_flags = PF_R | PF_X;
1625 else
1626 ph->p_flags = PF_R | PF_W;
1627 ph->p_align = s_align;
1629 /* Decide the layout of sections loaded in memory. This must
1630 be done before program headers are filled since they contain
1631 info about the layout. We do the following ordering: interp,
1632 symbol tables, relocations, progbits, nobits */
1633 /* XXX: do faster and simpler sorting */
1634 for(k = 0; k < 5; k++) {
1635 for(i = 1; i < s1->nb_sections; i++) {
1636 s = s1->sections[i];
1637 /* compute if section should be included */
1638 if (j == 0) {
1639 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1640 SHF_ALLOC)
1641 continue;
1642 } else {
1643 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1644 (SHF_ALLOC | SHF_WRITE))
1645 continue;
1647 if (s == interp) {
1648 if (k != 0)
1649 continue;
1650 } else if (s->sh_type == SHT_DYNSYM ||
1651 s->sh_type == SHT_STRTAB ||
1652 s->sh_type == SHT_HASH) {
1653 if (k != 1)
1654 continue;
1655 } else if (s->sh_type == SHT_RELX) {
1656 if (k != 2)
1657 continue;
1658 } else if (s->sh_type == SHT_NOBITS) {
1659 if (k != 4)
1660 continue;
1661 } else {
1662 if (k != 3)
1663 continue;
1665 sec_order[sh_order_index++] = i;
1667 /* section matches: we align it and add its size */
1668 tmp = addr;
1669 addr = (addr + s->sh_addralign - 1) &
1670 ~(s->sh_addralign - 1);
1671 file_offset += (int) ( addr - tmp );
1672 s->sh_offset = file_offset;
1673 s->sh_addr = addr;
1675 /* update program header infos */
1676 if (ph->p_offset == 0) {
1677 ph->p_offset = file_offset;
1678 ph->p_vaddr = addr;
1679 ph->p_paddr = ph->p_vaddr;
1681 /* update dynamic relocation infos */
1682 if (s->sh_type == SHT_RELX) {
1683 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1684 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1685 dyninf->rel_addr = addr;
1686 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1688 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1689 dyninf->bss_addr = addr;
1690 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1692 #else
1693 if (dyninf->rel_size == 0)
1694 dyninf->rel_addr = addr;
1695 dyninf->rel_size += s->sh_size;
1696 #endif
1698 addr += s->sh_size;
1699 if (s->sh_type != SHT_NOBITS)
1700 file_offset += s->sh_size;
1703 if (j == 0) {
1704 /* Make the first PT_LOAD segment include the program
1705 headers itself (and the ELF header as well), it'll
1706 come out with same memory use but will make various
1707 tools like binutils strip work better. */
1708 ph->p_offset &= ~(ph->p_align - 1);
1709 ph->p_vaddr &= ~(ph->p_align - 1);
1710 ph->p_paddr &= ~(ph->p_align - 1);
1712 ph->p_filesz = file_offset - ph->p_offset;
1713 ph->p_memsz = addr - ph->p_vaddr;
1714 ph++;
1715 if (j == 0) {
1716 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1717 /* if in the middle of a page, we duplicate the page in
1718 memory so that one copy is RX and the other is RW */
1719 if ((addr & (s_align - 1)) != 0)
1720 addr += s_align;
1721 } else {
1722 addr = (addr + s_align - 1) & ~(s_align - 1);
1723 file_offset = (file_offset + s_align - 1) & ~(s_align - 1);
1729 /* all other sections come after */
1730 for(i = 1; i < s1->nb_sections; i++) {
1731 s = s1->sections[i];
1732 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
1733 continue;
1734 sec_order[sh_order_index++] = i;
1736 file_offset = (file_offset + s->sh_addralign - 1) &
1737 ~(s->sh_addralign - 1);
1738 s->sh_offset = file_offset;
1739 if (s->sh_type != SHT_NOBITS)
1740 file_offset += s->sh_size;
1743 return file_offset;
1746 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
1747 Section *dynamic)
1749 ElfW(Phdr) *ph;
1751 /* if interpreter, then add corresponding program header */
1752 if (interp) {
1753 ph = &phdr[0];
1755 ph->p_type = PT_PHDR;
1756 ph->p_offset = sizeof(ElfW(Ehdr));
1757 ph->p_filesz = ph->p_memsz = phnum * sizeof(ElfW(Phdr));
1758 ph->p_vaddr = interp->sh_addr - ph->p_filesz;
1759 ph->p_paddr = ph->p_vaddr;
1760 ph->p_flags = PF_R | PF_X;
1761 ph->p_align = 4; /* interp->sh_addralign; */
1762 ph++;
1764 ph->p_type = PT_INTERP;
1765 ph->p_offset = interp->sh_offset;
1766 ph->p_vaddr = interp->sh_addr;
1767 ph->p_paddr = ph->p_vaddr;
1768 ph->p_filesz = interp->sh_size;
1769 ph->p_memsz = interp->sh_size;
1770 ph->p_flags = PF_R;
1771 ph->p_align = interp->sh_addralign;
1774 /* if dynamic section, then add corresponding program header */
1775 if (dynamic) {
1776 ph = &phdr[phnum - 1];
1778 ph->p_type = PT_DYNAMIC;
1779 ph->p_offset = dynamic->sh_offset;
1780 ph->p_vaddr = dynamic->sh_addr;
1781 ph->p_paddr = ph->p_vaddr;
1782 ph->p_filesz = dynamic->sh_size;
1783 ph->p_memsz = dynamic->sh_size;
1784 ph->p_flags = PF_R | PF_W;
1785 ph->p_align = dynamic->sh_addralign;
1789 /* Fill the dynamic section with tags describing the address and size of
1790 sections */
1791 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
1793 Section *dynamic = dyninf->dynamic;
1795 /* put dynamic section entries */
1796 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
1797 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
1798 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
1799 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
1800 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
1801 #if PTR_SIZE == 8
1802 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
1803 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
1804 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
1805 #else
1806 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1807 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
1808 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
1809 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
1810 put_dt(dynamic, DT_PLTREL, DT_REL);
1811 put_dt(dynamic, DT_REL, dyninf->bss_addr);
1812 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
1813 #else
1814 put_dt(dynamic, DT_REL, dyninf->rel_addr);
1815 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
1816 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
1817 #endif
1818 #endif
1819 if (s1->do_debug)
1820 put_dt(dynamic, DT_DEBUG, 0);
1821 put_dt(dynamic, DT_NULL, 0);
1824 /* Relocate remaining sections and symbols (that is those not related to
1825 dynamic linking) */
1826 static int final_sections_reloc(TCCState *s1)
1828 int i;
1829 Section *s;
1831 relocate_syms(s1, s1->symtab, 0);
1833 if (s1->nb_errors != 0)
1834 return -1;
1836 /* relocate sections */
1837 /* XXX: ignore sections with allocated relocations ? */
1838 for(i = 1; i < s1->nb_sections; i++) {
1839 s = s1->sections[i];
1840 if (s->reloc && s != s1->got)
1841 relocate_section(s1, s);
1844 /* relocate relocation entries if the relocation tables are
1845 allocated in the executable */
1846 for(i = 1; i < s1->nb_sections; i++) {
1847 s = s1->sections[i];
1848 if ((s->sh_flags & SHF_ALLOC) &&
1849 s->sh_type == SHT_RELX) {
1850 relocate_rel(s1, s);
1853 return 0;
1856 /* Create an ELF file on disk.
1857 This function handle ELF specific layout requirements */
1858 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
1859 int file_offset, int *sec_order)
1861 int i, shnum, offset, size, file_type;
1862 Section *s;
1863 ElfW(Ehdr) ehdr;
1864 ElfW(Shdr) shdr, *sh;
1866 file_type = s1->output_type;
1867 shnum = s1->nb_sections;
1869 memset(&ehdr, 0, sizeof(ehdr));
1871 if (phnum > 0) {
1872 ehdr.e_phentsize = sizeof(ElfW(Phdr));
1873 ehdr.e_phnum = phnum;
1874 ehdr.e_phoff = sizeof(ElfW(Ehdr));
1877 /* align to 4 */
1878 file_offset = (file_offset + 3) & -4;
1880 /* fill header */
1881 ehdr.e_ident[0] = ELFMAG0;
1882 ehdr.e_ident[1] = ELFMAG1;
1883 ehdr.e_ident[2] = ELFMAG2;
1884 ehdr.e_ident[3] = ELFMAG3;
1885 ehdr.e_ident[4] = ELFCLASSW;
1886 ehdr.e_ident[5] = ELFDATA2LSB;
1887 ehdr.e_ident[6] = EV_CURRENT;
1888 #if !defined(TCC_TARGET_PE) && (defined(__FreeBSD__) || defined(__FreeBSD_kernel__))
1889 /* FIXME: should set only for freebsd _target_, but we exclude only PE target */
1890 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1891 #endif
1892 #ifdef TCC_TARGET_ARM
1893 #ifdef TCC_ARM_EABI
1894 ehdr.e_ident[EI_OSABI] = 0;
1895 ehdr.e_flags = EF_ARM_EABI_VER4;
1896 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
1897 ehdr.e_flags |= EF_ARM_HASENTRY;
1898 if (s1->float_abi == ARM_HARD_FLOAT)
1899 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
1900 else
1901 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
1902 #else
1903 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
1904 #endif
1905 #endif
1906 switch(file_type) {
1907 default:
1908 case TCC_OUTPUT_EXE:
1909 ehdr.e_type = ET_EXEC;
1910 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
1911 break;
1912 case TCC_OUTPUT_DLL:
1913 ehdr.e_type = ET_DYN;
1914 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
1915 break;
1916 case TCC_OUTPUT_OBJ:
1917 ehdr.e_type = ET_REL;
1918 break;
1920 ehdr.e_machine = EM_TCC_TARGET;
1921 ehdr.e_version = EV_CURRENT;
1922 ehdr.e_shoff = file_offset;
1923 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
1924 ehdr.e_shentsize = sizeof(ElfW(Shdr));
1925 ehdr.e_shnum = shnum;
1926 ehdr.e_shstrndx = shnum - 1;
1928 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
1929 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
1930 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1932 sort_syms(s1, symtab_section);
1933 for(i = 1; i < s1->nb_sections; i++) {
1934 s = s1->sections[sec_order[i]];
1935 if (s->sh_type != SHT_NOBITS) {
1936 while (offset < s->sh_offset) {
1937 fputc(0, f);
1938 offset++;
1940 size = s->sh_size;
1941 if (size)
1942 fwrite(s->data, 1, size, f);
1943 offset += size;
1947 /* output section headers */
1948 while (offset < ehdr.e_shoff) {
1949 fputc(0, f);
1950 offset++;
1953 for(i = 0; i < s1->nb_sections; i++) {
1954 sh = &shdr;
1955 memset(sh, 0, sizeof(ElfW(Shdr)));
1956 s = s1->sections[i];
1957 if (s) {
1958 sh->sh_name = s->sh_name;
1959 sh->sh_type = s->sh_type;
1960 sh->sh_flags = s->sh_flags;
1961 sh->sh_entsize = s->sh_entsize;
1962 sh->sh_info = s->sh_info;
1963 if (s->link)
1964 sh->sh_link = s->link->sh_num;
1965 sh->sh_addralign = s->sh_addralign;
1966 sh->sh_addr = s->sh_addr;
1967 sh->sh_offset = s->sh_offset;
1968 sh->sh_size = s->sh_size;
1970 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
1974 /* Write an elf, coff or "binary" file */
1975 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
1976 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
1978 int fd, mode, file_type;
1979 FILE *f;
1981 file_type = s1->output_type;
1982 if (file_type == TCC_OUTPUT_OBJ)
1983 mode = 0666;
1984 else
1985 mode = 0777;
1986 unlink(filename);
1987 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
1988 if (fd < 0) {
1989 tcc_error_noabort("could not write '%s'", filename);
1990 return -1;
1992 f = fdopen(fd, "wb");
1993 if (s1->verbose)
1994 printf("<- %s\n", filename);
1996 #ifdef TCC_TARGET_COFF
1997 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
1998 tcc_output_coff(s1, f);
1999 else
2000 #endif
2001 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2002 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2003 else
2004 tcc_output_binary(s1, f, sec_order);
2005 fclose(f);
2007 return 0;
2010 /* Sort section headers by assigned sh_addr, remove sections
2011 that we aren't going to output. */
2012 static void tidy_section_headers(TCCState *s1, int *sec_order)
2014 int i, nnew, l, *backmap;
2015 Section **snew, *s;
2016 ElfW(Sym) *sym;
2018 snew = tcc_malloc(s1->nb_sections * sizeof(snew[0]));
2019 backmap = tcc_malloc(s1->nb_sections * sizeof(backmap[0]));
2020 for (i = 0, nnew = 0, l = s1->nb_sections; i < s1->nb_sections; i++) {
2021 s = s1->sections[sec_order[i]];
2022 if (!i || s->sh_name) {
2023 backmap[sec_order[i]] = nnew;
2024 snew[nnew] = s;
2025 ++nnew;
2026 } else {
2027 backmap[sec_order[i]] = 0;
2028 snew[--l] = s;
2031 for (i = 0; i < nnew; i++) {
2032 s = snew[i];
2033 if (s) {
2034 s->sh_num = i;
2035 if (s->sh_type == SHT_RELX)
2036 s->sh_info = backmap[s->sh_info];
2040 for_each_elem(symtab_section, 1, sym, ElfW(Sym))
2041 if (sym->st_shndx != SHN_UNDEF && sym->st_shndx < SHN_LORESERVE)
2042 sym->st_shndx = backmap[sym->st_shndx];
2043 if( !s1->static_link ) {
2044 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym))
2045 if (sym->st_shndx != SHN_UNDEF && sym->st_shndx < SHN_LORESERVE)
2046 sym->st_shndx = backmap[sym->st_shndx];
2048 for (i = 0; i < s1->nb_sections; i++)
2049 sec_order[i] = i;
2050 tcc_free(s1->sections);
2051 s1->sections = snew;
2052 s1->nb_sections = nnew;
2053 tcc_free(backmap);
2056 /* Output an elf, coff or binary file */
2057 /* XXX: suppress unneeded sections */
2058 static int elf_output_file(TCCState *s1, const char *filename)
2060 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2061 struct dyn_inf dyninf = {0};
2062 ElfW(Phdr) *phdr;
2063 ElfW(Sym) *sym;
2064 Section *strsec, *interp, *dynamic, *dynstr;
2065 int textrel;
2067 file_type = s1->output_type;
2068 s1->nb_errors = 0;
2069 ret = -1;
2070 phdr = NULL;
2071 sec_order = NULL;
2072 interp = dynamic = dynstr = NULL; /* avoid warning */
2073 textrel = 0;
2075 if (file_type != TCC_OUTPUT_OBJ) {
2076 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2077 tcc_add_runtime(s1);
2078 resolve_common_syms(s1);
2080 if (!s1->static_link) {
2081 if (file_type == TCC_OUTPUT_EXE) {
2082 char *ptr;
2083 /* allow override the dynamic loader */
2084 const char *elfint = getenv("LD_SO");
2085 if (elfint == NULL)
2086 elfint = DEFAULT_ELFINTERP(s1);
2087 /* add interpreter section only if executable */
2088 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2089 interp->sh_addralign = 1;
2090 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2091 strcpy(ptr, elfint);
2094 /* add dynamic symbol table */
2095 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2096 ".dynstr",
2097 ".hash", SHF_ALLOC);
2098 dynstr = s1->dynsym->link;
2100 /* add dynamic section */
2101 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2102 SHF_ALLOC | SHF_WRITE);
2103 dynamic->link = dynstr;
2104 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2106 build_got(s1);
2108 if (file_type == TCC_OUTPUT_EXE) {
2109 bind_exe_dynsyms(s1);
2110 if (s1->nb_errors)
2111 goto the_end;
2112 bind_libs_dynsyms(s1);
2113 } else {
2114 /* shared library case: simply export all global symbols */
2115 export_global_syms(s1);
2118 build_got_entries(s1);
2121 /* we add a section for symbols */
2122 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2123 put_elf_str(strsec, "");
2125 /* Allocate strings for section names */
2126 textrel = alloc_sec_names(s1, file_type, strsec);
2128 if (dynamic) {
2129 /* add a list of needed dlls */
2130 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2131 DLLReference *dllref = s1->loaded_dlls[i];
2132 if (dllref->level == 0)
2133 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2136 if (s1->rpath)
2137 put_dt(dynamic, s1->enable_new_dtags ? DT_RUNPATH : DT_RPATH,
2138 put_elf_str(dynstr, s1->rpath));
2140 if (file_type == TCC_OUTPUT_DLL) {
2141 if (s1->soname)
2142 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2143 /* XXX: currently, since we do not handle PIC code, we
2144 must relocate the readonly segments */
2145 if (textrel)
2146 put_dt(dynamic, DT_TEXTREL, 0);
2149 if (s1->symbolic)
2150 put_dt(dynamic, DT_SYMBOLIC, 0);
2152 dyninf.dynamic = dynamic;
2153 dyninf.dynstr = dynstr;
2154 /* remember offset and reserve space for 2nd call below */
2155 dyninf.data_offset = dynamic->data_offset;
2156 fill_dynamic(s1, &dyninf);
2157 dynamic->sh_size = dynamic->data_offset;
2158 dynstr->sh_size = dynstr->data_offset;
2161 /* compute number of program headers */
2162 if (file_type == TCC_OUTPUT_OBJ)
2163 phnum = 0;
2164 else if (file_type == TCC_OUTPUT_DLL)
2165 phnum = 3;
2166 else if (s1->static_link)
2167 phnum = 2;
2168 else
2169 phnum = 5;
2171 /* allocate program segment headers */
2172 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2174 /* compute number of sections */
2175 shnum = s1->nb_sections;
2177 /* this array is used to reorder sections in the output file */
2178 sec_order = tcc_malloc(sizeof(int) * shnum);
2179 sec_order[0] = 0;
2181 /* compute section to program header mapping */
2182 file_offset = layout_sections(s1, phdr, phnum, interp, strsec, &dyninf,
2183 sec_order);
2185 /* Fill remaining program header and finalize relocation related to dynamic
2186 linking. */
2187 if (file_type != TCC_OUTPUT_OBJ) {
2188 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2189 if (dynamic) {
2190 dynamic->data_offset = dyninf.data_offset;
2191 fill_dynamic(s1, &dyninf);
2193 /* put in GOT the dynamic section address and relocate PLT */
2194 write32le(s1->got->data, dynamic->sh_addr);
2195 if (file_type == TCC_OUTPUT_EXE
2196 || (RELOCATE_DLLPLT && file_type == TCC_OUTPUT_DLL))
2197 relocate_plt(s1);
2199 /* relocate symbols in .dynsym now that final addresses are known */
2200 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2201 if (sym->st_shndx != SHN_UNDEF && sym->st_shndx < SHN_LORESERVE) {
2202 /* do symbol relocation */
2203 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2208 /* if building executable or DLL, then relocate each section
2209 except the GOT which is already relocated */
2210 ret = final_sections_reloc(s1);
2211 if (ret)
2212 goto the_end;
2213 tidy_section_headers(s1, sec_order);
2215 /* Perform relocation to GOT or PLT entries */
2216 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2217 fill_got(s1);
2218 else if (s1->got)
2219 fill_local_got_entries(s1);
2222 /* Create the ELF file with name 'filename' */
2223 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2224 s1->nb_sections = shnum;
2225 the_end:
2226 tcc_free(sec_order);
2227 tcc_free(phdr);
2228 return ret;
2231 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2233 int ret;
2234 #ifdef TCC_TARGET_PE
2235 if (s->output_type != TCC_OUTPUT_OBJ) {
2236 ret = pe_output_file(s, filename);
2237 } else
2238 #endif
2239 ret = elf_output_file(s, filename);
2240 return ret;
2243 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2245 void *data;
2247 data = tcc_malloc(size);
2248 lseek(fd, file_offset, SEEK_SET);
2249 read(fd, data, size);
2250 return data;
2253 typedef struct SectionMergeInfo {
2254 Section *s; /* corresponding existing section */
2255 unsigned long offset; /* offset of the new section in the existing section */
2256 uint8_t new_section; /* true if section 's' was added */
2257 uint8_t link_once; /* true if link once section */
2258 } SectionMergeInfo;
2260 ST_FUNC int tcc_object_type(int fd, ElfW(Ehdr) *h)
2262 int size = read(fd, h, sizeof *h);
2263 if (size == sizeof *h && 0 == memcmp(h, ELFMAG, 4)) {
2264 if (h->e_type == ET_REL)
2265 return AFF_BINTYPE_REL;
2266 if (h->e_type == ET_DYN)
2267 return AFF_BINTYPE_DYN;
2268 } else if (size >= 8) {
2269 if (0 == memcmp(h, ARMAG, 8))
2270 return AFF_BINTYPE_AR;
2271 #ifdef TCC_TARGET_COFF
2272 if (((struct filehdr*)h)->f_magic == COFF_C67_MAGIC)
2273 return AFF_BINTYPE_C67;
2274 #endif
2276 return 0;
2279 /* load an object file and merge it with current files */
2280 /* XXX: handle correctly stab (debug) info */
2281 ST_FUNC int tcc_load_object_file(TCCState *s1,
2282 int fd, unsigned long file_offset)
2284 ElfW(Ehdr) ehdr;
2285 ElfW(Shdr) *shdr, *sh;
2286 int size, i, j, offset, offseti, nb_syms, sym_index, ret, seencompressed;
2287 unsigned char *strsec, *strtab;
2288 int *old_to_new_syms;
2289 char *sh_name, *name;
2290 SectionMergeInfo *sm_table, *sm;
2291 ElfW(Sym) *sym, *symtab;
2292 ElfW_Rel *rel;
2293 Section *s;
2295 int stab_index;
2296 int stabstr_index;
2298 stab_index = stabstr_index = 0;
2300 lseek(fd, file_offset, SEEK_SET);
2301 if (tcc_object_type(fd, &ehdr) != AFF_BINTYPE_REL)
2302 goto fail1;
2303 /* test CPU specific stuff */
2304 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2305 ehdr.e_machine != EM_TCC_TARGET) {
2306 fail1:
2307 tcc_error_noabort("invalid object file");
2308 return -1;
2310 /* read sections */
2311 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2312 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2313 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2315 /* load section names */
2316 sh = &shdr[ehdr.e_shstrndx];
2317 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2319 /* load symtab and strtab */
2320 old_to_new_syms = NULL;
2321 symtab = NULL;
2322 strtab = NULL;
2323 nb_syms = 0;
2324 seencompressed = 0;
2325 for(i = 1; i < ehdr.e_shnum; i++) {
2326 sh = &shdr[i];
2327 if (sh->sh_type == SHT_SYMTAB) {
2328 if (symtab) {
2329 tcc_error_noabort("object must contain only one symtab");
2330 fail:
2331 ret = -1;
2332 goto the_end;
2334 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2335 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2336 sm_table[i].s = symtab_section;
2338 /* now load strtab */
2339 sh = &shdr[sh->sh_link];
2340 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2342 if (sh->sh_flags & SHF_COMPRESSED)
2343 seencompressed = 1;
2346 /* now examine each section and try to merge its content with the
2347 ones in memory */
2348 for(i = 1; i < ehdr.e_shnum; i++) {
2349 /* no need to examine section name strtab */
2350 if (i == ehdr.e_shstrndx)
2351 continue;
2352 sh = &shdr[i];
2353 sh_name = (char *) strsec + sh->sh_name;
2354 /* ignore sections types we do not handle */
2355 if (sh->sh_type != SHT_PROGBITS &&
2356 sh->sh_type != SHT_RELX &&
2357 #ifdef TCC_ARM_EABI
2358 sh->sh_type != SHT_ARM_EXIDX &&
2359 #endif
2360 sh->sh_type != SHT_NOBITS &&
2361 sh->sh_type != SHT_PREINIT_ARRAY &&
2362 sh->sh_type != SHT_INIT_ARRAY &&
2363 sh->sh_type != SHT_FINI_ARRAY &&
2364 strcmp(sh_name, ".stabstr")
2366 continue;
2367 if (seencompressed
2368 && (!strncmp(sh_name, ".debug_", sizeof(".debug_")-1)
2369 || (sh->sh_type == SHT_RELX
2370 && !strncmp((char*)strsec + shdr[sh->sh_info].sh_name,
2371 ".debug_", sizeof(".debug_")-1))))
2372 continue;
2373 if (sh->sh_addralign < 1)
2374 sh->sh_addralign = 1;
2375 /* find corresponding section, if any */
2376 for(j = 1; j < s1->nb_sections;j++) {
2377 s = s1->sections[j];
2378 if (!strcmp(s->name, sh_name)) {
2379 if (!strncmp(sh_name, ".gnu.linkonce",
2380 sizeof(".gnu.linkonce") - 1)) {
2381 /* if a 'linkonce' section is already present, we
2382 do not add it again. It is a little tricky as
2383 symbols can still be defined in
2384 it. */
2385 sm_table[i].link_once = 1;
2386 goto next;
2387 } else {
2388 goto found;
2392 /* not found: create new section */
2393 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags & ~SHF_GROUP);
2394 /* take as much info as possible from the section. sh_link and
2395 sh_info will be updated later */
2396 s->sh_addralign = sh->sh_addralign;
2397 s->sh_entsize = sh->sh_entsize;
2398 sm_table[i].new_section = 1;
2399 found:
2400 if (sh->sh_type != s->sh_type) {
2401 tcc_error_noabort("invalid section type");
2402 goto fail;
2405 /* align start of section */
2406 offset = s->data_offset;
2408 if (0 == strcmp(sh_name, ".stab")) {
2409 stab_index = i;
2410 goto no_align;
2412 if (0 == strcmp(sh_name, ".stabstr")) {
2413 stabstr_index = i;
2414 goto no_align;
2417 size = sh->sh_addralign - 1;
2418 offset = (offset + size) & ~size;
2419 if (sh->sh_addralign > s->sh_addralign)
2420 s->sh_addralign = sh->sh_addralign;
2421 s->data_offset = offset;
2422 no_align:
2423 sm_table[i].offset = offset;
2424 sm_table[i].s = s;
2425 /* concatenate sections */
2426 size = sh->sh_size;
2427 if (sh->sh_type != SHT_NOBITS) {
2428 unsigned char *ptr;
2429 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2430 ptr = section_ptr_add(s, size);
2431 read(fd, ptr, size);
2432 } else {
2433 s->data_offset += size;
2435 next: ;
2438 /* gr relocate stab strings */
2439 if (stab_index && stabstr_index) {
2440 Stab_Sym *a, *b;
2441 unsigned o;
2442 s = sm_table[stab_index].s;
2443 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2444 b = (Stab_Sym *)(s->data + s->data_offset);
2445 o = sm_table[stabstr_index].offset;
2446 while (a < b)
2447 a->n_strx += o, a++;
2450 /* second short pass to update sh_link and sh_info fields of new
2451 sections */
2452 for(i = 1; i < ehdr.e_shnum; i++) {
2453 s = sm_table[i].s;
2454 if (!s || !sm_table[i].new_section)
2455 continue;
2456 sh = &shdr[i];
2457 if (sh->sh_link > 0)
2458 s->link = sm_table[sh->sh_link].s;
2459 if (sh->sh_type == SHT_RELX) {
2460 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2461 /* update backward link */
2462 s1->sections[s->sh_info]->reloc = s;
2465 sm = sm_table;
2467 /* resolve symbols */
2468 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2470 sym = symtab + 1;
2471 for(i = 1; i < nb_syms; i++, sym++) {
2472 if (sym->st_shndx != SHN_UNDEF &&
2473 sym->st_shndx < SHN_LORESERVE) {
2474 sm = &sm_table[sym->st_shndx];
2475 if (sm->link_once) {
2476 /* if a symbol is in a link once section, we use the
2477 already defined symbol. It is very important to get
2478 correct relocations */
2479 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2480 name = (char *) strtab + sym->st_name;
2481 sym_index = find_elf_sym(symtab_section, name);
2482 if (sym_index)
2483 old_to_new_syms[i] = sym_index;
2485 continue;
2487 /* if no corresponding section added, no need to add symbol */
2488 if (!sm->s)
2489 continue;
2490 /* convert section number */
2491 sym->st_shndx = sm->s->sh_num;
2492 /* offset value */
2493 sym->st_value += sm->offset;
2495 /* add symbol */
2496 name = (char *) strtab + sym->st_name;
2497 sym_index = set_elf_sym(symtab_section, sym->st_value, sym->st_size,
2498 sym->st_info, sym->st_other,
2499 sym->st_shndx, name);
2500 old_to_new_syms[i] = sym_index;
2503 /* third pass to patch relocation entries */
2504 for(i = 1; i < ehdr.e_shnum; i++) {
2505 s = sm_table[i].s;
2506 if (!s)
2507 continue;
2508 sh = &shdr[i];
2509 offset = sm_table[i].offset;
2510 switch(s->sh_type) {
2511 case SHT_RELX:
2512 /* take relocation offset information */
2513 offseti = sm_table[sh->sh_info].offset;
2514 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2515 int type;
2516 unsigned sym_index;
2517 /* convert symbol index */
2518 type = ELFW(R_TYPE)(rel->r_info);
2519 sym_index = ELFW(R_SYM)(rel->r_info);
2520 /* NOTE: only one symtab assumed */
2521 if (sym_index >= nb_syms)
2522 goto invalid_reloc;
2523 sym_index = old_to_new_syms[sym_index];
2524 /* ignore link_once in rel section. */
2525 if (!sym_index && !sm->link_once
2526 #ifdef TCC_TARGET_ARM
2527 && type != R_ARM_V4BX
2528 #endif
2530 invalid_reloc:
2531 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2532 i, strsec + sh->sh_name, rel->r_offset);
2533 goto fail;
2535 rel->r_info = ELFW(R_INFO)(sym_index, type);
2536 /* offset the relocation offset */
2537 rel->r_offset += offseti;
2538 #ifdef TCC_TARGET_ARM
2539 /* Jumps and branches from a Thumb code to a PLT entry need
2540 special handling since PLT entries are ARM code.
2541 Unconditional bl instructions referencing PLT entries are
2542 handled by converting these instructions into blx
2543 instructions. Other case of instructions referencing a PLT
2544 entry require to add a Thumb stub before the PLT entry to
2545 switch to ARM mode. We set bit plt_thumb_stub of the
2546 attribute of a symbol to indicate such a case. */
2547 if (type == R_ARM_THM_JUMP24)
2548 get_sym_attr(s1, sym_index, 1)->plt_thumb_stub = 1;
2549 #endif
2551 break;
2552 default:
2553 break;
2557 ret = 0;
2558 the_end:
2559 tcc_free(symtab);
2560 tcc_free(strtab);
2561 tcc_free(old_to_new_syms);
2562 tcc_free(sm_table);
2563 tcc_free(strsec);
2564 tcc_free(shdr);
2565 return ret;
2568 typedef struct ArchiveHeader {
2569 char ar_name[16]; /* name of this member */
2570 char ar_date[12]; /* file mtime */
2571 char ar_uid[6]; /* owner uid; printed as decimal */
2572 char ar_gid[6]; /* owner gid; printed as decimal */
2573 char ar_mode[8]; /* file mode, printed as octal */
2574 char ar_size[10]; /* file size, printed as decimal */
2575 char ar_fmag[2]; /* should contain ARFMAG */
2576 } ArchiveHeader;
2578 static int get_be32(const uint8_t *b)
2580 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2583 static long get_be64(const uint8_t *b)
2585 long long ret = get_be32(b);
2586 ret = (ret << 32) | (unsigned)get_be32(b+4);
2587 return (long)ret;
2590 /* load only the objects which resolve undefined symbols */
2591 static int tcc_load_alacarte(TCCState *s1, int fd, int size, int entrysize)
2593 long i, bound, nsyms, sym_index, off, ret;
2594 uint8_t *data;
2595 const char *ar_names, *p;
2596 const uint8_t *ar_index;
2597 ElfW(Sym) *sym;
2599 data = tcc_malloc(size);
2600 if (read(fd, data, size) != size)
2601 goto fail;
2602 nsyms = entrysize == 4 ? get_be32(data) : get_be64(data);
2603 ar_index = data + entrysize;
2604 ar_names = (char *) ar_index + nsyms * entrysize;
2606 do {
2607 bound = 0;
2608 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2609 sym_index = find_elf_sym(symtab_section, p);
2610 if(sym_index) {
2611 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2612 if(sym->st_shndx == SHN_UNDEF) {
2613 off = (entrysize == 4
2614 ? get_be32(ar_index + i * 4)
2615 : get_be64(ar_index + i * 8))
2616 + sizeof(ArchiveHeader);
2617 ++bound;
2618 if(tcc_load_object_file(s1, fd, off) < 0) {
2619 fail:
2620 ret = -1;
2621 goto the_end;
2626 } while(bound);
2627 ret = 0;
2628 the_end:
2629 tcc_free(data);
2630 return ret;
2633 /* load a '.a' file */
2634 ST_FUNC int tcc_load_archive(TCCState *s1, int fd, int alacarte)
2636 ArchiveHeader hdr;
2637 char ar_size[11];
2638 char ar_name[17];
2639 char magic[8];
2640 int size, len, i;
2641 unsigned long file_offset;
2643 /* skip magic which was already checked */
2644 read(fd, magic, sizeof(magic));
2646 for(;;) {
2647 len = read(fd, &hdr, sizeof(hdr));
2648 if (len == 0)
2649 break;
2650 if (len != sizeof(hdr)) {
2651 tcc_error_noabort("invalid archive");
2652 return -1;
2654 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2655 ar_size[sizeof(hdr.ar_size)] = '\0';
2656 size = strtol(ar_size, NULL, 0);
2657 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2658 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2659 if (ar_name[i] != ' ')
2660 break;
2662 ar_name[i + 1] = '\0';
2663 file_offset = lseek(fd, 0, SEEK_CUR);
2664 /* align to even */
2665 size = (size + 1) & ~1;
2666 if (!strcmp(ar_name, "/")) {
2667 /* coff symbol table : we handle it */
2668 if (alacarte)
2669 return tcc_load_alacarte(s1, fd, size, 4);
2670 } else if (!strcmp(ar_name, "/SYM64/")) {
2671 if (alacarte)
2672 return tcc_load_alacarte(s1, fd, size, 8);
2673 } else {
2674 ElfW(Ehdr) ehdr;
2675 if (tcc_object_type(fd, &ehdr) == AFF_BINTYPE_REL) {
2676 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2677 return -1;
2680 lseek(fd, file_offset + size, SEEK_SET);
2682 return 0;
2685 #ifndef TCC_TARGET_PE
2686 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2687 is referenced by the user (so it should be added as DT_NEEDED in
2688 the generated ELF file) */
2689 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2691 ElfW(Ehdr) ehdr;
2692 ElfW(Shdr) *shdr, *sh, *sh1;
2693 int i, j, nb_syms, nb_dts, sym_bind, ret;
2694 ElfW(Sym) *sym, *dynsym;
2695 ElfW(Dyn) *dt, *dynamic;
2696 unsigned char *dynstr;
2697 const char *name, *soname;
2698 DLLReference *dllref;
2700 read(fd, &ehdr, sizeof(ehdr));
2702 /* test CPU specific stuff */
2703 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2704 ehdr.e_machine != EM_TCC_TARGET) {
2705 tcc_error_noabort("bad architecture");
2706 return -1;
2709 /* read sections */
2710 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2712 /* load dynamic section and dynamic symbols */
2713 nb_syms = 0;
2714 nb_dts = 0;
2715 dynamic = NULL;
2716 dynsym = NULL; /* avoid warning */
2717 dynstr = NULL; /* avoid warning */
2718 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2719 switch(sh->sh_type) {
2720 case SHT_DYNAMIC:
2721 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2722 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2723 break;
2724 case SHT_DYNSYM:
2725 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2726 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2727 sh1 = &shdr[sh->sh_link];
2728 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2729 break;
2730 default:
2731 break;
2735 /* compute the real library name */
2736 soname = tcc_basename(filename);
2738 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2739 if (dt->d_tag == DT_SONAME) {
2740 soname = (char *) dynstr + dt->d_un.d_val;
2744 /* if the dll is already loaded, do not load it */
2745 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2746 dllref = s1->loaded_dlls[i];
2747 if (!strcmp(soname, dllref->name)) {
2748 /* but update level if needed */
2749 if (level < dllref->level)
2750 dllref->level = level;
2751 ret = 0;
2752 goto the_end;
2756 /* add the dll and its level */
2757 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2758 dllref->level = level;
2759 strcpy(dllref->name, soname);
2760 dynarray_add(&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2762 /* add dynamic symbols in dynsym_section */
2763 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2764 sym_bind = ELFW(ST_BIND)(sym->st_info);
2765 if (sym_bind == STB_LOCAL)
2766 continue;
2767 name = (char *) dynstr + sym->st_name;
2768 set_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
2769 sym->st_info, sym->st_other, sym->st_shndx, name);
2772 /* load all referenced DLLs */
2773 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2774 switch(dt->d_tag) {
2775 case DT_NEEDED:
2776 name = (char *) dynstr + dt->d_un.d_val;
2777 for(j = 0; j < s1->nb_loaded_dlls; j++) {
2778 dllref = s1->loaded_dlls[j];
2779 if (!strcmp(name, dllref->name))
2780 goto already_loaded;
2782 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
2783 tcc_error_noabort("referenced dll '%s' not found", name);
2784 ret = -1;
2785 goto the_end;
2787 already_loaded:
2788 break;
2791 ret = 0;
2792 the_end:
2793 tcc_free(dynstr);
2794 tcc_free(dynsym);
2795 tcc_free(dynamic);
2796 tcc_free(shdr);
2797 return ret;
2800 #define LD_TOK_NAME 256
2801 #define LD_TOK_EOF (-1)
2803 /* return next ld script token */
2804 static int ld_next(TCCState *s1, char *name, int name_size)
2806 int c;
2807 char *q;
2809 redo:
2810 switch(ch) {
2811 case ' ':
2812 case '\t':
2813 case '\f':
2814 case '\v':
2815 case '\r':
2816 case '\n':
2817 inp();
2818 goto redo;
2819 case '/':
2820 minp();
2821 if (ch == '*') {
2822 file->buf_ptr = parse_comment(file->buf_ptr);
2823 ch = file->buf_ptr[0];
2824 goto redo;
2825 } else {
2826 q = name;
2827 *q++ = '/';
2828 goto parse_name;
2830 break;
2831 case '\\':
2832 ch = handle_eob();
2833 if (ch != '\\')
2834 goto redo;
2835 /* fall through */
2836 /* case 'a' ... 'z': */
2837 case 'a':
2838 case 'b':
2839 case 'c':
2840 case 'd':
2841 case 'e':
2842 case 'f':
2843 case 'g':
2844 case 'h':
2845 case 'i':
2846 case 'j':
2847 case 'k':
2848 case 'l':
2849 case 'm':
2850 case 'n':
2851 case 'o':
2852 case 'p':
2853 case 'q':
2854 case 'r':
2855 case 's':
2856 case 't':
2857 case 'u':
2858 case 'v':
2859 case 'w':
2860 case 'x':
2861 case 'y':
2862 case 'z':
2863 /* case 'A' ... 'z': */
2864 case 'A':
2865 case 'B':
2866 case 'C':
2867 case 'D':
2868 case 'E':
2869 case 'F':
2870 case 'G':
2871 case 'H':
2872 case 'I':
2873 case 'J':
2874 case 'K':
2875 case 'L':
2876 case 'M':
2877 case 'N':
2878 case 'O':
2879 case 'P':
2880 case 'Q':
2881 case 'R':
2882 case 'S':
2883 case 'T':
2884 case 'U':
2885 case 'V':
2886 case 'W':
2887 case 'X':
2888 case 'Y':
2889 case 'Z':
2890 case '_':
2891 case '.':
2892 case '$':
2893 case '~':
2894 q = name;
2895 parse_name:
2896 for(;;) {
2897 if (!((ch >= 'a' && ch <= 'z') ||
2898 (ch >= 'A' && ch <= 'Z') ||
2899 (ch >= '0' && ch <= '9') ||
2900 strchr("/.-_+=$:\\,~", ch)))
2901 break;
2902 if ((q - name) < name_size - 1) {
2903 *q++ = ch;
2905 minp();
2907 *q = '\0';
2908 c = LD_TOK_NAME;
2909 break;
2910 case CH_EOF:
2911 c = LD_TOK_EOF;
2912 break;
2913 default:
2914 c = ch;
2915 inp();
2916 break;
2918 return c;
2921 static int ld_add_file(TCCState *s1, const char filename[])
2923 if (filename[0] == '/') {
2924 if (CONFIG_SYSROOT[0] == '\0'
2925 && tcc_add_file_internal(s1, filename, AFF_TYPE_BIN) == 0)
2926 return 0;
2927 filename = tcc_basename(filename);
2929 return tcc_add_dll(s1, filename, 0);
2932 static inline int new_undef_syms(void)
2934 int ret = 0;
2935 ret = new_undef_sym;
2936 new_undef_sym = 0;
2937 return ret;
2940 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
2942 char filename[1024], libname[1024];
2943 int t, group, nblibs = 0, ret = 0;
2944 char **libs = NULL;
2946 group = !strcmp(cmd, "GROUP");
2947 if (!as_needed)
2948 new_undef_syms();
2949 t = ld_next(s1, filename, sizeof(filename));
2950 if (t != '(')
2951 expect("(");
2952 t = ld_next(s1, filename, sizeof(filename));
2953 for(;;) {
2954 libname[0] = '\0';
2955 if (t == LD_TOK_EOF) {
2956 tcc_error_noabort("unexpected end of file");
2957 ret = -1;
2958 goto lib_parse_error;
2959 } else if (t == ')') {
2960 break;
2961 } else if (t == '-') {
2962 t = ld_next(s1, filename, sizeof(filename));
2963 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
2964 tcc_error_noabort("library name expected");
2965 ret = -1;
2966 goto lib_parse_error;
2968 pstrcpy(libname, sizeof libname, &filename[1]);
2969 if (s1->static_link) {
2970 snprintf(filename, sizeof filename, "lib%s.a", libname);
2971 } else {
2972 snprintf(filename, sizeof filename, "lib%s.so", libname);
2974 } else if (t != LD_TOK_NAME) {
2975 tcc_error_noabort("filename expected");
2976 ret = -1;
2977 goto lib_parse_error;
2979 if (!strcmp(filename, "AS_NEEDED")) {
2980 ret = ld_add_file_list(s1, cmd, 1);
2981 if (ret)
2982 goto lib_parse_error;
2983 } else {
2984 /* TODO: Implement AS_NEEDED support. Ignore it for now */
2985 if (!as_needed) {
2986 ret = ld_add_file(s1, filename);
2987 if (ret)
2988 goto lib_parse_error;
2989 if (group) {
2990 /* Add the filename *and* the libname to avoid future conversions */
2991 dynarray_add(&libs, &nblibs, tcc_strdup(filename));
2992 if (libname[0] != '\0')
2993 dynarray_add(&libs, &nblibs, tcc_strdup(libname));
2997 t = ld_next(s1, filename, sizeof(filename));
2998 if (t == ',') {
2999 t = ld_next(s1, filename, sizeof(filename));
3002 if (group && !as_needed) {
3003 while (new_undef_syms()) {
3004 int i;
3006 for (i = 0; i < nblibs; i ++)
3007 ld_add_file(s1, libs[i]);
3010 lib_parse_error:
3011 dynarray_reset(&libs, &nblibs);
3012 return ret;
3015 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3016 files */
3017 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3019 char cmd[64];
3020 char filename[1024];
3021 int t, ret;
3023 ch = handle_eob();
3024 for(;;) {
3025 t = ld_next(s1, cmd, sizeof(cmd));
3026 if (t == LD_TOK_EOF)
3027 return 0;
3028 else if (t != LD_TOK_NAME)
3029 return -1;
3030 if (!strcmp(cmd, "INPUT") ||
3031 !strcmp(cmd, "GROUP")) {
3032 ret = ld_add_file_list(s1, cmd, 0);
3033 if (ret)
3034 return ret;
3035 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3036 !strcmp(cmd, "TARGET")) {
3037 /* ignore some commands */
3038 t = ld_next(s1, cmd, sizeof(cmd));
3039 if (t != '(')
3040 expect("(");
3041 for(;;) {
3042 t = ld_next(s1, filename, sizeof(filename));
3043 if (t == LD_TOK_EOF) {
3044 tcc_error_noabort("unexpected end of file");
3045 return -1;
3046 } else if (t == ')') {
3047 break;
3050 } else {
3051 return -1;
3054 return 0;
3056 #endif /* !TCC_TARGET_PE */