outelf.c

开源的nasm编译器源码,研究编译器原理很有帮且

/* outelf.c	output routines for the Netwide Assembler to produce
 *		ELF32 (i386 of course) object file format
 *
 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
 * Julian Hall. All rights reserved. The software is
 * redistributable under the licence given in the file "Licence"
 * distributed in the NASM archive.
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ctype.h>

#include "nasm.h"
#include "nasmlib.h"
#include "outform.h"

#ifdef OF_ELF

/*
 * Relocation types.
 */
enum reloc_type {
  R_386_32        =  1,		/* ordinary absolute relocation */
  R_386_PC32      =  2,		/* PC-relative relocation */
  R_386_GOT32     =  3,		/* an offset into GOT */
  R_386_PLT32     =  4,		/* a PC-relative offset into PLT */
  R_386_COPY      =  5,		/* ??? */
  R_386_GLOB_DAT  =  6,		/* ??? */
  R_386_JUMP_SLOT =  7,		/* ??? */
  R_386_RELATIVE  =  8,		/* ??? */
  R_386_GOTOFF    =  9,		/* an offset from GOT base */
  R_386_GOTPC     = 10,		/* a PC-relative offset _to_ GOT */
  /* These are GNU extensions, but useful */
  R_386_16        = 20,		/* A 16-bit absolute relocation */
  R_386_PC16      = 21,		/* A 16-bit PC-relative relocation */
  R_386_8         = 22,		/* An 8-bit absolute relocation */
  R_386_PC8       = 23		/* An 8-bit PC-relative relocation */
};

struct Reloc {
    struct Reloc *next;
    long address;		       /* relative to _start_ of section */
    long symbol;		       /* ELF symbol info thingy */
    int type;			       /* type of relocation */
};

struct Symbol {
    long strpos;		       /* string table position of name */
    long section;		       /* section ID of the symbol */
    int type;			       /* symbol type */
    long value;			       /* address, or COMMON variable align */
    long size;			       /* size of symbol */
    long globnum;		       /* symbol table offset if global */
    struct Symbol *next;	       /* list of globals in each section */
    struct Symbol *nextfwd;	       /* list of unresolved-size symbols */
    char *name;			       /* used temporarily if in above list */
};

#define SHT_PROGBITS 1
#define SHT_NOBITS 8

#define SHF_WRITE 1
#define SHF_ALLOC 2
#define SHF_EXECINSTR 4

struct Section {
    struct SAA *data;
    unsigned long len, size, nrelocs;
    long index;
    int type;			       /* SHT_PROGBITS or SHT_NOBITS */
    int align;			       /* alignment: power of two */
    unsigned long flags;	       /* section flags */
    char *name;
    struct SAA *rel;
    long rellen;
    struct Reloc *head, **tail;
    struct Symbol *gsyms;	       /* global symbols in section */
};

#define SECT_DELTA 32
static struct Section **sects;
static int nsects, sectlen;

#define SHSTR_DELTA 256
static char *shstrtab;
static int shstrtablen, shstrtabsize;

static struct SAA *syms;
static unsigned long nlocals, nglobs;

static long def_seg;

static struct RAA *bsym;

static struct SAA *strs;
static unsigned long strslen;

static FILE *elffp;
static efunc error;
static evalfunc evaluate;

static struct Symbol *fwds;

static char elf_module[FILENAME_MAX];

extern struct ofmt of_elf;

#define SHN_ABS 0xFFF1
#define SHN_COMMON 0xFFF2
#define SHN_UNDEF 0

#define SYM_SECTION 0x04
#define SYM_GLOBAL 0x10
#define SYM_DATA 0x01
#define SYM_FUNCTION 0x02

#define GLOBAL_TEMP_BASE 16	       /* bigger than any constant sym id */

#define SEG_ALIGN 16		       /* alignment of sections in file */
#define SEG_ALIGN_1 (SEG_ALIGN-1)

static const char align_str[SEG_ALIGN] = ""; /* ANSI will pad this with 0s */

#define ELF_MAX_SECTIONS 16	       /* really 10, but let's play safe */
static struct ELF_SECTDATA {
    void *data;
    long len;
    int is_saa;
} *elf_sects;
static int elf_nsect;
static long elf_foffs;

static void elf_write(void);
static void elf_sect_write(struct Section *, const unsigned char *, unsigned long);
static void elf_section_header (int, int, int, void *, int, long,
				int, int, int, int);
static void elf_write_sections (void);
static struct SAA *elf_build_symtab (long *, long *);
static struct SAA *elf_build_reltab (long *, struct Reloc *);
static void add_sectname (char *, char *);


/* this stuff is needed for the stabs debugging format */
#define N_SO 0x64     /* ID for main source file */
#define N_SOL 0x84    /* ID for sub-source file */
#define N_BINCL 0x82
#define N_EINCL 0xA2
#define N_SLINE 0x44
#define TY_STABSSYMLIN 0x40 /* ouch */

struct stabentry {
  unsigned long n_strx;
  unsigned char n_type;
  unsigned char n_other;
  unsigned short n_desc;
  unsigned long n_value;
};

struct erel {
  int offset,info;
};

struct symlininfo {
  int offset;
  int section; /* section index */
  char *name; /* shallow-copied pointer of section name */
};

struct linelist {
  struct symlininfo info;
  int line;
  char *filename;
  struct linelist * next;
  struct linelist * last;
};

static struct linelist *stabslines=0;
static int stabs_immcall=0;
static int currentline=0;
static int numlinestabs=0;
static char * stabs_filename=0;
static int symtabsection;
static unsigned char *stabbuf=0,*stabstrbuf=0,*stabrelbuf=0;
static int stablen,stabstrlen,stabrellen;

static struct dfmt df_stabs;

void stabs_init(struct ofmt *,void *,FILE *,efunc );
void stabs_linenum(const char *filename,long linenumber,long);
void stabs_deflabel(char *,long ,long ,int ,char *);
void stabs_directive(const char *,const char *);
void stabs_typevalue(long );
void stabs_output(int ,void *);
void stabs_generate();
void stabs_cleanup();

/* end of stabs debugging stuff */


/*
 * Special section numbers which are used to define ELF special
 * symbols, which can be used with WRT to provide PIC relocation
 * types.
 */
static long elf_gotpc_sect, elf_gotoff_sect;
static long elf_got_sect, elf_plt_sect;
static long elf_sym_sect;

static void elf_init(FILE *fp, efunc errfunc, ldfunc ldef, evalfunc eval)
{
    elffp = fp;
    error = errfunc;
    evaluate = eval;
    (void) ldef;		       /* placate optimisers */
    sects = NULL;
    nsects = sectlen = 0;
    syms = saa_init((long)sizeof(struct Symbol));
    nlocals = nglobs = 0;
    bsym = raa_init();
    strs = saa_init(1L);
    saa_wbytes (strs, "\0", 1L);
    saa_wbytes (strs, elf_module, (long)(strlen(elf_module)+1));
    strslen = 2+strlen(elf_module);
    shstrtab = NULL;
    shstrtablen = shstrtabsize = 0;;
    add_sectname ("", "");

    fwds = NULL;

    elf_gotpc_sect = seg_alloc();
    ldef("..gotpc", elf_gotpc_sect+1, 0L, NULL, FALSE, FALSE, &of_elf, error);
    elf_gotoff_sect = seg_alloc();
    ldef("..gotoff", elf_gotoff_sect+1, 0L, NULL, FALSE, FALSE,&of_elf,error);
    elf_got_sect = seg_alloc();
    ldef("..got", elf_got_sect+1, 0L, NULL, FALSE, FALSE, &of_elf, error);
    elf_plt_sect = seg_alloc();
    ldef("..plt", elf_plt_sect+1, 0L, NULL, FALSE, FALSE, &of_elf, error);
    elf_sym_sect = seg_alloc();
    ldef("..sym", elf_sym_sect+1, 0L, NULL, FALSE, FALSE, &of_elf, error);

    def_seg = seg_alloc();
}

static void elf_cleanup(int debuginfo)
{
    struct Reloc *r;
    int i;

    (void) debuginfo;

    elf_write();
    fclose (elffp);
    for (i=0; i<nsects; i++) {
	if (sects[i]->type != SHT_NOBITS)
	    saa_free (sects[i]->data);
	if (sects[i]->head)
	    saa_free (sects[i]->rel);
	while (sects[i]->head) {
	    r = sects[i]->head;
	    sects[i]->head = sects[i]->head->next;
	    nasm_free (r);
	}
    }
    nasm_free (sects);
    saa_free (syms);
    raa_free (bsym);
    saa_free (strs);
    if (of_elf.current_dfmt) {
      of_elf.current_dfmt->cleanup();
    }
}

static void add_sectname (char *firsthalf, char *secondhalf)
{
    int len = strlen(firsthalf)+strlen(secondhalf);
    while (shstrtablen + len + 1 > shstrtabsize)
	shstrtab = nasm_realloc (shstrtab, (shstrtabsize += SHSTR_DELTA));
    strcpy (shstrtab+shstrtablen, firsthalf);
    strcat (shstrtab+shstrtablen, secondhalf);
    shstrtablen += len+1;
}

static int elf_make_section (char *name, int type, int flags, int align)
{
    struct Section *s;

    s = nasm_malloc (sizeof(*s));

    if (type != SHT_NOBITS)
	s->data = saa_init (1L);
    s->head = NULL;
    s->tail = &s->head;
    s->len = s->size = 0;
    s->nrelocs = 0;
    if (!strcmp(name, ".text"))
	s->index = def_seg;
    else
	s->index = seg_alloc();
    add_sectname ("", name);
    s->name = nasm_malloc (1+strlen(name));
    strcpy (s->name, name);
    s->type = type;
    s->flags = flags;
    s->align = align;
    s->gsyms = NULL;

    if (nsects >= sectlen)
	sects = nasm_realloc (sects, (sectlen += SECT_DELTA)*sizeof(*sects));
    sects[nsects++] = s;

    return nsects-1;
}

static long elf_section_names (char *name, int pass, int *bits)
{
    char *p;
    int flags_and, flags_or, type, align, i;

    /*
     * Default is 32 bits.
     */
    if (!name) {
	*bits = 32;
	return def_seg;
    }

    p = name;
    while (*p && !isspace(*p)) p++;
    if (*p) *p++ = '\0';
    flags_and = flags_or = type = align = 0;

    while (*p && isspace(*p)) p++;
    while (*p) {
	char *q = p;
	while (*p && !isspace(*p)) p++;
	if (*p) *p++ = '\0';
	while (*p && isspace(*p)) p++;
	
	if (!nasm_strnicmp(q, "align=", 6)) {
	    align = atoi(q+6);
	    if (align == 0)
		align = 1;
	    if ( (align-1) & align ) {   /* means it's not a power of two */
		error (ERR_NONFATAL, "section alignment %d is not"
		       " a power of two", align);
		align = 1;
	    }
	} else if (!nasm_stricmp(q, "alloc")) {
	    flags_and |= SHF_ALLOC;
	    flags_or |= SHF_ALLOC;
	} else if (!nasm_stricmp(q, "noalloc")) {
	    flags_and |= SHF_ALLOC;
	    flags_or &= ~SHF_ALLOC;
	} else if (!nasm_stricmp(q, "exec")) {
	    flags_and |= SHF_EXECINSTR;
	    flags_or |= SHF_EXECINSTR;
	} else if (!nasm_stricmp(q, "noexec")) {
	    flags_and |= SHF_EXECINSTR;
	    flags_or &= ~SHF_EXECINSTR;
	} else if (!nasm_stricmp(q, "write")) {
	    flags_and |= SHF_WRITE;
	    flags_or |= SHF_WRITE;
	} else if (!nasm_stricmp(q, "nowrite")) {
	    flags_and |= SHF_WRITE;
	    flags_or &= ~SHF_WRITE;
	} else if (!nasm_stricmp(q, "progbits")) {
	    type = SHT_PROGBITS;
	} else if (!nasm_stricmp(q, "nobits")) {
	    type = SHT_NOBITS;
	}
    }

    if (!strcmp(name, ".comment") ||
	!strcmp(name, ".shstrtab") ||
	!strcmp(name, ".symtab") ||
	!strcmp(name, ".strtab")) {
	error (ERR_NONFATAL, "attempt to redefine reserved section"
	       "name `%s'", name);
	return NO_SEG;
    }

    for (i=0; i<nsects; i++)
	if (!strcmp(name, sects[i]->name))
	    break;
    if (i == nsects) {
	if (!strcmp(name, ".text"))
	    i = elf_make_section (name, SHT_PROGBITS,
				  SHF_ALLOC | SHF_EXECINSTR, 16);
	else if (!strcmp(name, ".rodata"))
	    i = elf_make_section (name, SHT_PROGBITS,
				  SHF_ALLOC, 4);
	else if (!strcmp(name, ".data"))
	    i = elf_make_section (name, SHT_PROGBITS,
				  SHF_ALLOC | SHF_WRITE, 4);
	else if (!strcmp(name, ".bss"))
	    i = elf_make_section (name, SHT_NOBITS,
				  SHF_ALLOC | SHF_WRITE, 4);
	else
	    i = elf_make_section (name, SHT_PROGBITS, SHF_ALLOC, 1);
	if (type)
	    sects[i]->type = type;
	if (align)
	    sects[i]->align = align;
	sects[i]->flags &= ~flags_and;
	sects[i]->flags |= flags_or;
    } else if (pass == 1) {
	if (type || align || flags_and)
	    error (ERR_WARNING, "section attributes ignored on"
		   " redeclaration of section `%s'", name);
    }

    return sects[i]->index;
}

static void elf_deflabel (char *name, long segment, long offset,
			   int is_global, char *special)
{
    int pos = strslen;
    struct Symbol *sym;
    int special_used = FALSE;

#if defined(DEBUG) && DEBUG>2
fprintf(stderr, " elf_deflabel: %s, seg=%ld, off=%ld, is_global=%d, %s\n",
      name, segment, offset, is_global, special);
#endif
    if (name[0] == '.' && name[1] == '.' && name[2] != '@') {
	/*
	 * This is a NASM special symbol. We never allow it into
	 * the ELF symbol table, even if it's a valid one. If it
	 * _isn't_ a valid one, we should barf immediately.
	 */
	if (strcmp(name, "..gotpc") && strcmp(name, "..gotoff") &&
	    strcmp(name, "..got") && strcmp(name, "..plt") &&
	    strcmp(name, "..sym"))
	    error (ERR_NONFATAL, "unrecognised special symbol `%s'", name);
	return;
    }

    if (is_global == 3) {
	struct Symbol **s;
	/*
	 * Fix up a forward-reference symbol size from the first
	 * pass.
	 */
	for (s = &fwds; *s; s = &(*s)->nextfwd)
	    if (!strcmp((*s)->name, name)) {
		struct tokenval tokval;
		expr *e;
		char *p = special;

		while (*p && !isspace(*p)) p++;
		while (*p && isspace(*p)) p++;
		stdscan_reset();
		stdscan_bufptr = p;
		tokval.t_type = TOKEN_INVALID;
		e = evaluate(stdscan, NULL, &tokval, NULL, 1, error, NULL);
		if (e) {
		    if (!is_simple(e))
			error (ERR_NONFATAL, "cannot use relocatable"
			       " expression as symbol size");
		    else
			(*s)->size = reloc_value(e);
		}

		/*
		 * Remove it from the list of unresolved sizes.
		 */
		nasm_free ((*s)->name);
		*s = (*s)->nextfwd;
		return;
	    }
	return;			       /* it wasn't an important one */
    }

    saa_wbytes (strs, name, (long)(1+strlen(name)));
    strslen += 1+strlen(name);

    sym = saa_wstruct (syms);

    sym->strpos = pos;
    sym->type = is_global ? SYM_GLOBAL : 0;
    sym->size = 0;
    if (segment == NO_SEG)
	sym->section = SHN_ABS;
    else {
	int i;
	sym->section = SHN_UNDEF;
	if (nsects == 0 && segment == def_seg) {
	    int tempint;
	    if (segment != elf_section_names (".text", 2, &tempint))
		error (ERR_PANIC, "strange segment conditions in ELF driver");
	    sym->section = nsects;
	} else {
	    for (i=0; i<nsects; i++)