tccelf.c

一个GNU的C语言编译器源代码,源代码过万行.

                error_noabort("object must contain only one symtab");
            fail:
                ret = -1;
                goto the_end;
            }
            nb_syms = sh->sh_size / sizeof(Elf32_Sym);
            symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
            sm_table[i].s = symtab_section;

            /* now load strtab */
            sh = &shdr[sh->sh_link];
            strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
        }
    }
        
    /* now examine each section and try to merge its content with the
       ones in memory */
    for(i = 1; i < ehdr.e_shnum; i++) {
        /* no need to examine section name strtab */
        if (i == ehdr.e_shstrndx)
            continue;
        sh = &shdr[i];
        sh_name = strsec + sh->sh_name;
        /* ignore sections types we do not handle */
        if (sh->sh_type != SHT_PROGBITS &&
            sh->sh_type != SHT_REL && 
            sh->sh_type != SHT_NOBITS)
            continue;
        if (sh->sh_addralign < 1)
            sh->sh_addralign = 1;
        /* find corresponding section, if any */
        for(j = 1; j < s1->nb_sections;j++) {
            s = s1->sections[j];
            if (!strcmp(s->name, sh_name)) {
                if (!strncmp(sh_name, ".gnu.linkonce", 
                             sizeof(".gnu.linkonce") - 1)) {
                    /* if a 'linkonce' section is already present, we
                       do not add it again. It is a little tricky as
                       symbols can still be defined in
                       it. */
                    sm_table[i].link_once = 1;
                    goto next;
                } else {
                    goto found;
                }
            }
        }
        /* not found: create new section */
        s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
        /* take as much info as possible from the section. sh_link and
           sh_info will be updated later */
        s->sh_addralign = sh->sh_addralign;
        s->sh_entsize = sh->sh_entsize;
        sm_table[i].new_section = 1;
    found:
        if (sh->sh_type != s->sh_type) {
            error_noabort("invalid section type");
            goto fail;
        }

        /* align start of section */
        offset = s->data_offset;
        size = sh->sh_addralign - 1;
        offset = (offset + size) & ~size;
        if (sh->sh_addralign > s->sh_addralign)
            s->sh_addralign = sh->sh_addralign;
        s->data_offset = offset;
        sm_table[i].offset = offset;
        sm_table[i].s = s;
        /* concatenate sections */
        size = sh->sh_size;
        if (sh->sh_type != SHT_NOBITS) {
            unsigned char *ptr;
            lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
            ptr = section_ptr_add(s, size);
            read(fd, ptr, size);
        } else {
            s->data_offset += size;
        }
    next: ;
    }

    /* second short pass to update sh_link and sh_info fields of new
       sections */
    sm = sm_table;
    for(i = 1; i < ehdr.e_shnum; i++) {
        s = sm_table[i].s;
        if (!s || !sm_table[i].new_section)
            continue;
        sh = &shdr[i];
        if (sh->sh_link > 0)
            s->link = sm_table[sh->sh_link].s;
        if (sh->sh_type == SHT_REL) {
            s->sh_info = sm_table[sh->sh_info].s->sh_num;
            /* update backward link */
            s1->sections[s->sh_info]->reloc = s;
        }
    }

    /* resolve symbols */
    old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));

    sym = symtab + 1;
    for(i = 1; i < nb_syms; i++, sym++) {
        if (sym->st_shndx != SHN_UNDEF &&
            sym->st_shndx < SHN_LORESERVE) {
            sm = &sm_table[sym->st_shndx];
            if (sm->link_once) {
                /* if a symbol is in a link once section, we use the
                   already defined symbol. It is very important to get
                   correct relocations */
                if (ELF32_ST_BIND(sym->st_info) != STB_LOCAL) {
                    name = strtab + sym->st_name;
                    sym_index = find_elf_sym(symtab_section, name);
                    if (sym_index)
                        old_to_new_syms[i] = sym_index;
                }
                continue;
            }
            /* if no corresponding section added, no need to add symbol */
            if (!sm->s)
                continue;
            /* convert section number */
            sym->st_shndx = sm->s->sh_num;
            /* offset value */
            sym->st_value += sm->offset;
        }
        /* add symbol */
        name = strtab + sym->st_name;
        sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size, 
                                sym->st_info, sym->st_shndx, name);
        old_to_new_syms[i] = sym_index;
    }

    /* third pass to patch relocation entries */
    for(i = 1; i < ehdr.e_shnum; i++) {
        s = sm_table[i].s;
        if (!s)
            continue;
        sh = &shdr[i];
        offset = sm_table[i].offset;
        switch(s->sh_type) {
        case SHT_REL:
            /* take relocation offset information */
            offseti = sm_table[sh->sh_info].offset;
            rel_end = (Elf32_Rel *)(s->data + s->data_offset);
            for(rel = (Elf32_Rel *)(s->data + offset);
                rel < rel_end;
                rel++) {
                int type;
                unsigned sym_index;
                /* convert symbol index */
                type = ELF32_R_TYPE(rel->r_info);
                sym_index = ELF32_R_SYM(rel->r_info);
                /* NOTE: only one symtab assumed */
                if (sym_index >= nb_syms)
                    goto invalid_reloc;
                sym_index = old_to_new_syms[sym_index];
                if (!sym_index) {
                invalid_reloc:
                    error_noabort("Invalid relocation entry");
                    goto fail;
                }
                rel->r_info = ELF32_R_INFO(sym_index, type);
                /* offset the relocation offset */
                rel->r_offset += offseti;
            }
            break;
        default:
            break;
        }
    }
    
    ret = 0;
 the_end:
    tcc_free(symtab);
    tcc_free(strtab);
    tcc_free(old_to_new_syms);
    tcc_free(sm_table);
    tcc_free(strsec);
    tcc_free(shdr);
    return ret;
}

#define ARMAG  "!<arch>\012"	/* For COFF and a.out archives */

typedef struct ArchiveHeader {
    char ar_name[16];		/* name of this member */
    char ar_date[12];		/* file mtime */
    char ar_uid[6];		/* owner uid; printed as decimal */
    char ar_gid[6];		/* owner gid; printed as decimal */
    char ar_mode[8];		/* file mode, printed as octal   */
    char ar_size[10];		/* file size, printed as decimal */
    char ar_fmag[2];		/* should contain ARFMAG */
} ArchiveHeader;

static int get_be32(const uint8_t *b)
{
    return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
}

/* load only the objects which resolve undefined symbols */
static int tcc_load_alacarte(TCCState *s1, int fd, int size)
{
    int i, bound, nsyms, sym_index, off, ret;
    uint8_t *data;
    const char *ar_names, *p;
    const uint8_t *ar_index;
    Elf32_Sym *sym;

    data = tcc_malloc(size);
    if (read(fd, data, size) != size)
        goto fail;
    nsyms = get_be32(data);
    ar_index = data + 4;
    ar_names = ar_index + nsyms * 4;

    do {
	bound = 0;
	for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
	    sym_index = find_elf_sym(symtab_section, p);
	    if(sym_index) {
		sym = &((Elf32_Sym *)symtab_section->data)[sym_index];
		if(sym->st_shndx == SHN_UNDEF) {
		    off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
#if 0
		    printf("%5d\t%s\t%08x\n", i, p, sym->st_shndx);
#endif
		    ++bound;
		    lseek(fd, off, SEEK_SET);
		    if(tcc_load_object_file(s1, fd, off) < 0) {
                    fail:
                        ret = -1;
                        goto the_end;
                    }
		}
	    }
	}
    } while(bound);
    ret = 0;
 the_end:
    tcc_free(data);
    return ret;
}

/* load a '.a' file */
static int tcc_load_archive(TCCState *s1, int fd)
{
    ArchiveHeader hdr;
    char ar_size[11];
    char ar_name[17];
    char magic[8];
    int size, len, i;
    unsigned long file_offset;

    /* skip magic which was already checked */
    read(fd, magic, sizeof(magic));
    
    for(;;) {
        len = read(fd, &hdr, sizeof(hdr));
        if (len == 0)
            break;
        if (len != sizeof(hdr)) {
            error_noabort("invalid archive");
            return -1;
        }
        memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
        ar_size[sizeof(hdr.ar_size)] = '\0';
        size = strtol(ar_size, NULL, 0);
        memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
        for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
            if (ar_name[i] != ' ')
                break;
        }
        ar_name[i + 1] = '\0';
        //        printf("name='%s' size=%d %s\n", ar_name, size, ar_size);
        file_offset = lseek(fd, 0, SEEK_CUR);
        /* align to even */
        size = (size + 1) & ~1;
        if (!strcmp(ar_name, "/")) {
            /* coff symbol table : we handle it */
	    if(s1->alacarte_link)
		return tcc_load_alacarte(s1, fd, size);
        } else if (!strcmp(ar_name, "//") ||
                   !strcmp(ar_name, "__.SYMDEF") ||
                   !strcmp(ar_name, "__.SYMDEF/") ||
                   !strcmp(ar_name, "ARFILENAMES/")) {
            /* skip symbol table or archive names */
        } else {
            if (tcc_load_object_file(s1, fd, file_offset) < 0)
                return -1;
        }
        lseek(fd, file_offset + size, SEEK_SET);
    }
    return 0;
}

/* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
   is referenced by the user (so it should be added as DT_NEEDED in
   the generated ELF file) */
static int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
{ 
    Elf32_Ehdr ehdr;
    Elf32_Shdr *shdr, *sh, *sh1;
    int i, nb_syms, nb_dts, sym_bind, ret;
    Elf32_Sym *sym, *dynsym;
    Elf32_Dyn *dt, *dynamic;
    unsigned char *dynstr;
    const char *name, *soname, *p;
    DLLReference *dllref;
    
    read(fd, &ehdr, sizeof(ehdr));

    /* test CPU specific stuff */
    if (ehdr.e_ident[5] != ELFDATA2LSB ||
        ehdr.e_machine != EM_386) {
        error_noabort("bad architecture");
        return -1;
    }

    /* read sections */
    shdr = load_data(fd, ehdr.e_shoff, sizeof(Elf32_Shdr) * ehdr.e_shnum);

    /* load dynamic section and dynamic symbols */
    nb_syms = 0;
    nb_dts = 0;
    dynamic = NULL;
    dynsym = NULL; /* avoid warning */
    dynstr = NULL; /* avoid warning */
    for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
        switch(sh->sh_type) {
        case SHT_DYNAMIC:
            nb_dts = sh->sh_size / sizeof(Elf32_Dyn);
            dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
            break;
        case SHT_DYNSYM:
            nb_syms = sh->sh_size / sizeof(Elf32_Sym);
            dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
            sh1 = &shdr[sh->sh_link];
            dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
            break;
        default:
            break;
        }
    }
    
    /* compute the real library name */
    soname = filename;
    p = strrchr(soname, '/');
    if (p)
        soname = p + 1;
        
    for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
        if (dt->d_tag == DT_SONAME) {
            soname = dynstr + dt->d_un.d_val;
        }
    }

    /* if the dll is already loaded, do not load it */
    for(i = 0; i < s1->nb_loaded_dlls; i++) {
        dllref = s1->loaded_dlls[i];
        if (!strcmp(soname, dllref->name)) {
            /* but update level if needed */
            if (level < dllref->level)
                dllref->level = level;
            ret = 0;
            goto the_end;
        }
    }
    
    //    printf("loading dll '%s'\n", soname);

    /* add the dll and its level */
    dllref = tcc_malloc(sizeof(DLLReference) + strlen(soname));
    dllref->level = level;
    strcpy(dllref->name, soname);
    dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);

    /* add dynamic symbols in dynsym_section */
    for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
        sym_bind = ELF32_ST_BIND(sym->st_info);
        if (sym_bind == STB_LOCAL)
            continue;
        name = dynstr + sym->st_name;
        add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
                    sym->st_info, sym->st_shndx, name);
    }