从一段代码看ELF .o文件（ET_REL）

网上有好多介绍ELF的文章，基本内容大多来自ELF的标准

Tool Interface Standard (TIS)
Executable and Linking Format (ELF)
Specification
Version 1.2

google elf pdf就能搜索到不少
通过一段代码来学习还是比较容易，下面的代码来自一个实用编译器tcc
主页是
因为tcc包含编译和链接功能，这段代码属于链接器的部分
有些项目（比如linux kernel）可能会有成千上万个.o，如果一个.c代码改变，从而链接成千上万个.o也是非常消耗时间的
所以一般也是会把不同的子目录下的.o合并成一个整体，这样，链接不再需要从头开始
ld -r参数就是由若干个.o 生成一个新的.o

/* load an object file and merge it with current files */
/* XXX: handle correctly stab (debug) info */
ST_FUNC int tcc_load_object_file(TCCState *s1,
                                int fd, unsigned long file_offset)
{//  -----
    ElfW(Ehdr) ehdr;
    ElfW(Shdr) *shdr, *sh;
    int size, i, j, offset, offseti, nb_syms, sym_index, ret;
    unsigned char *strsec, *strtab;
    int *old_to_new_syms;
    char *sh_name, *name;
    SectionMergeInfo *sm_table, *sm;
    ElfW(Sym) *sym, *symtab;
    ElfW_Rel *rel, *rel_end;
    Section *s;

    int stab_index;
    int stabstr_index;

    stab_index = stabstr_index = 0;

     //每个ELF文件都以一个elf头开始，对elf文件的分析起始于头部

    if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
        goto fail1;
    //
    if (ehdr.e_ident[0] != ELFMAG0 ||
        ehdr.e_ident[1] != ELFMAG1 ||
        ehdr.e_ident[2] != ELFMAG2 ||
        ehdr.e_ident[3] != ELFMAG3)
        goto fail1;

     该函数只分析可重定位目标文件

    /* test if object file */
    if (ehdr.e_type != ET_REL)
        goto fail1;


    /* test CPU specific stuff */
    if (ehdr.e_ident[5] != ELFDATA2LSB ||
        ehdr.e_machine != EM_TCC_TARGET) {
    tcc支持不少平台
    在i386-gen.c中 #define EM_TCC_TARGET EM_386
    fail1:
        error_noabort("invalid object file");
        return -1;
    }

    
     目标文件不需要 PHT, 目标文件包含若干节，比如数据，代码，符号表等等
    

每个节的大小是固定的大小， 比如32位上

typedef struct
{
  Elf32_Word    sh_name;                /* Section name (string tbl index) */
  Elf32_Word    sh_type;                /* Section type */
  Elf32_Word    sh_flags;               /* Section flags */
  Elf32_Addr    sh_addr;                /* Section virtual addr at execution */
  Elf32_Off     sh_offset;              /* Section file offset */
  Elf32_Word    sh_size;                /* Section size in bytes */
  Elf32_Word    sh_link;                /* Link to another section */
  Elf32_Word    sh_info;                /* Additional section information */
  Elf32_Word    sh_addralign;           /* Section alignment */
  Elf32_Word    sh_entsize;             /* Entry size if section holds table */
} Elf32_Shdr;


e_shnum指定有多少节
shdr指向一个Elf32_Shdr结构体数组, 即节头表，用来描述各个节的信息
Elf32_Shdr各个字段对于不同的节有不同的解释，这里只分析ELF_REL

    /* read sections */
    shdr = load_data(fd, file_offset + ehdr.e_shoff,
                     sizeof(ElfW(Shdr)) * ehdr.e_shnum);

    sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);

    每节都有名字， 所有的节名字也构成一个节，它在节头表中的位置由e_shstrndx给出

    /* load section names */
    sh = &shdr[ehdr.e_shstrndx];
    strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);

    /* load symtab and strtab */
    old_to_new_syms = NULL;
    symtab = NULL;
    strtab = NULL;
    nb_syms = 0;

    索引0是保留的，用来表示未定义

    for(i = 1; i < ehdr.e_shnum; i++) {
        sh = &shdr[i];
        if (sh->sh_type == SHT_SYMTAB) {
            if (symtab) {
                error_noabort("object must contain only one symtab");
            fail:
                ret = -1;
                goto the_end;
            }
            nb_syms = sh->sh_size / sizeof(ElfW(Sym));
            symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
            symtab_section由tcc建立，是符号表的内存表示
            sm_table[i].s = symtab_section;

            符号表中的每个符号都有名字，比如"printf", "fopen"等, 符号表的名字也有一个节
            节号由符号表的sh_link字段给出
            /* now load strtab */
            sh = &shdr[sh->sh_link];
            strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
        }
    }

    链接用于把所有obj, lib合并，并重定位，指出起始符号(默认是_start)，合并obj的功能在这里完成
    比如大多obj中都有的.text节，最后都会合并到text_section，这是text节在内存中的表示
    /* 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_RELX &&
#ifdef TCC_ARM_EABI
            sh->sh_type != SHT_ARM_EXIDX &&
#endif
            sh->sh_type != SHT_NOBITS &&
            sh->sh_type != SHT_PREINIT_ARRAY &&
            sh->sh_type != SHT_INIT_ARRAY &&
            sh->sh_type != SHT_FINI_ARRAY &&
            strcmp(sh_name, ".stabstr")
            )
            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. */
                    linkonce用于不同原文件需要，但是链接一次的情况
                    比如C++的函数模板需要根据参数的类型生成不同的代码，函数模板必须定义，编译器无法判断
                    别的原文件是否会生成同样的代码，链接器负责删除重复的，也就是linkonce
                    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;

        if (0 == strcmp(sh_name, ".stab")) {
            stab_index = i;
            goto no_align;
        }
        if (0 == strcmp(sh_name, ".stabstr")) {
            stabstr_index = i;
            goto no_align;
        }

        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;
    no_align:
        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: ;
    }

    /* //gr relocate stab strings */
    if (stab_index && stabstr_index) {
        Stab_Sym *a, *b;
        unsigned o;
        s = sm_table[stab_index].s;
        a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
        b = (Stab_Sym *)(s->data + s->data_offset);
        o = sm_table[stabstr_index].offset;
        while (a < b)
            a->n_strx += o, a++;
    }

    /* second short pass to update sh_link and sh_info fields of new
       sections */
    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_RELX) {
            s->sh_info = sm_table[sh->sh_info].s->sh_num;
            /* update backward link */
            该节需要被重定位，重定位节是s
            s1->sections[s->sh_info]->reloc = s;
        }
    }
    sm = sm_table;

    /* resolve symbols */
    old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
    合并之后符号的索引就变了

    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 (ELFW(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_other,
                                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_RELX:
            /* take relocation offset information */
            offseti = sm_table[sh->sh_info].offset;
            rel_end = (ElfW_Rel *)(s->data + s->data_offset);
            for(rel = (ElfW_Rel *)(s->data + offset);
                rel < rel_end;
                rel++) {
                int type;
                unsigned sym_index;
                /* convert symbol index */
                type = ELFW(R_TYPE)(rel->r_info);
                sym_index = ELFW(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];
                /* ignore link_once in rel section. */
                if (!sym_index && !sm->link_once
#ifdef TCC_TARGET_ARM
                    && type != R_ARM_V4BX
#endif
                   ) {
                invalid_reloc:
                    error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
                        i, strsec + sh->sh_name, rel->r_offset);
                    goto fail;
                }
                rel->r_info = ELFW(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;
}