1 /* Modified by Andrew.Vignaux@comp.vuw.ac.nz to get it to work :-) */
3 /* Copyright (C) 1985, 1986, 1987, 1988 Free Software Foundation, Inc.
5 This file is part of GNU Emacs.
7 GNU Emacs is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 GNU Emacs is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GNU Emacs; see the file COPYING. If not, write to
19 the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
20 Boston, MA 02111-1307, USA.
22 In other words, you are welcome to use, share and improve this program.
23 You are forbidden to forbid anyone else to use, share and improve
24 what you give them. Help stamp out software-hoarding! */
28 * unexec.c - Convert a running program into an a.out file.
30 * Author: Spencer W. Thomas
31 * Computer Science Dept.
33 * Date: Tue Mar 2 1982
34 * Modified heavily since then.
37 * unexec (new_name, a_name, data_start, bss_start, entry_address)
38 * char *new_name, *a_name;
39 * unsigned data_start, bss_start, entry_address;
41 * Takes a snapshot of the program and makes an a.out format file in the
42 * file named by the string argument new_name.
43 * If a_name is non-NULL, the symbol table will be taken from the given file.
44 * On some machines, an existing a_name file is required.
46 * The boundaries within the a.out file may be adjusted with the data_start
47 * and bss_start arguments. Either or both may be given as 0 for defaults.
49 * Data_start gives the boundary between the text segment and the data
50 * segment of the program. The text segment can contain shared, read-only
51 * program code and literal data, while the data segment is always unshared
52 * and unprotected. Data_start gives the lowest unprotected address.
53 * The value you specify may be rounded down to a suitable boundary
54 * as required by the machine you are using.
56 * Specifying zero for data_start means the boundary between text and data
57 * should not be the same as when the program was loaded.
58 * If NO_REMAP is defined, the argument data_start is ignored and the
59 * segment boundaries are never changed.
61 * Bss_start indicates how much of the data segment is to be saved in the
62 * a.out file and restored when the program is executed. It gives the lowest
63 * unsaved address, and is rounded up to a page boundary. The default when 0
64 * is given assumes that the entire data segment is to be stored, including
65 * the previous data and bss as well as any additional storage allocated with
68 * The new file is set up to start at entry_address.
70 * If you make improvements I'd like to get them too.
71 * harpo!utah-cs!thomas, thomas@Utah-20
75 /* There are several compilation parameters affecting unexec:
79 Define this if your system uses COFF for executables.
80 Otherwise we assume you use Berkeley format.
84 Define this if you do not want to try to save Emacs's pure data areas
85 as part of the text segment.
87 Saving them as text is good because it allows users to share more.
89 However, on machines that locate the text area far from the data area,
90 the boundary cannot feasibly be moved. Such machines require
93 Also, remapping can cause trouble with the built-in startup routine
94 /lib/crt0.o, which defines `environ' as an initialized variable.
95 Dumping `environ' as pure does not work! So, to use remapping,
96 you must write a startup routine for your machine in Emacs's crt0.c.
97 If NO_REMAP is defined, Emacs uses the system's crt0.o.
101 Some machines that use COFF executables require that each section
102 start on a certain boundary *in the COFF file*. Such machines should
103 define SECTION_ALIGNMENT to a mask of the low-order bits that must be
104 zero on such a boundary. This mask is used to control padding between
105 segments in the COFF file.
107 If SECTION_ALIGNMENT is not defined, the segments are written
108 consecutively with no attempt at alignment. This is right for
113 Some machines require that the beginnings and ends of segments
114 *in core* be on certain boundaries. For most machines, a page
115 boundary is sufficient. That is the default. When a larger
116 boundary is needed, define SEGMENT_MASK to a mask of
117 the bits that must be zero on such a boundary.
121 Some machines count the a.out header as part of the size of the text
122 segment (a_text); they may actually load the header into core as the
123 first data in the text segment. Some have additional padding between
124 the header and the real text of the program that is counted in a_text.
126 For these machines, define A_TEXT_OFFSET(HDR) to examine the header
127 structure HDR and return the number of bytes to add to `a_text'
128 before writing it (above and beyond the number of bytes of actual
129 program text). HDR's standard fields are already correct, except that
130 this adjustment to the `a_text' field has not yet been made;
131 thus, the amount of offset can depend on the data in the file.
135 If defined, this macro specifies the number of bytes to seek into the
136 a.out file before starting to write the text segment.a
140 For machines using COFF, this macro, if defined, is a value stored
141 into the magic number field of the output file.
145 This macro can be used to generate statements to adjust or
146 initialize nonstandard fields in the file header
150 Macro to correct an int which is the bit pattern of a pointer to a byte
151 into an int which is the number of a byte.
153 This macro has a default definition which is usually right.
154 This default definition is a no-op on most machines (where a
155 pointer looks like an int) but not on all machines.
164 #define PERROR(arg) perror (arg); return -1
167 #define PERROR(file) report_error (file, new)
171 /* Define getpagesize () if the system does not.
172 Note that this may depend on symbols defined in a.out.h
174 #include "getpagesize.h"
176 #ifndef makedev /* Try to detect types.h already loaded */
177 #include <sys/types.h>
180 #include <sys/stat.h>
183 extern char *start_of_text (); /* Start of text */
184 extern char *start_of_data (); /* Start of initialized data */
196 /* I have a suspicion that these are turned off on all systems
197 and can be deleted. Try it in version 19. */
202 #endif /* not sun386 */
203 #endif /* not UMAX */
204 #endif /* Not STRIDE */
206 static long block_copy_start
; /* Old executable start point */
207 static struct filehdr f_hdr
; /* File header */
208 static struct aouthdr f_ohdr
; /* Optional file header (a.out) */
209 long bias
; /* Bias to add for growth */
210 long lnnoptr
; /* Pointer to line-number info within file */
211 #define SYMS_START block_copy_start
213 static long text_scnptr
;
214 static long data_scnptr
;
216 static long load_scnptr
;
217 static long orig_load_scnptr
;
218 static long orig_data_scnptr
;
223 #define MAX_SECTIONS 10
230 /* Correct an int which is the bit pattern of a pointer to a byte
231 into an int which is the number of a byte.
232 This is a no-op on ordinary machines, but not on all. */
234 #ifndef ADDR_CORRECT /* Let m-*.h files override this definition */
235 #define ADDR_CORRECT(x) ((char *)(x) - (char*)0)
242 report_error (file
, fd
)
248 report_file_error ("Cannot unexec", Fcons (build_string (file
), Qnil
));
252 #define ERROR0(msg) report_error_1 (new, msg, 0, 0); return -1
253 #define ERROR1(msg,x) report_error_1 (new, msg, x, 0); return -1
254 #define ERROR2(msg,x,y) report_error_1 (new, msg, x, y); return -1
257 report_error_1 (fd
, msg
, a1
, a2
)
266 fprintf (stderr
, msg
, a1
, a2
);
267 fprintf (stderr
, "\n");
271 static int make_hdr ();
272 static void mark_x ();
273 static int copy_text_and_data ();
274 static int copy_sym ();
276 /* ****************************************************************
281 unexec (new_name
, a_name
, data_start
, bss_start
, entry_address
)
282 char *new_name
, *a_name
;
283 unsigned data_start
, bss_start
, entry_address
;
287 if (a_name
&& (a_out
= open (a_name
, 0)) < 0)
291 if ((new = creat (new_name
, 0666)) < 0)
295 if (make_hdr (new,a_out
,data_start
,bss_start
,entry_address
,a_name
,new_name
) < 0
296 || copy_text_and_data (new) < 0
297 || copy_sym (new, a_out
, a_name
, new_name
) < 0
299 || adjust_lnnoptrs (new, a_out
, new_name
) < 0
302 || unrelocate_symbols (new, a_out
, a_name
, new_name
) < 0
307 /* unlink (new_name); /* Failed, unlink new a.out */
318 /* ****************************************************************
321 * Make the header in the new a.out from the header in core.
322 * Modify the text and data sizes.
325 make_hdr (new, a_out
, data_start
, bss_start
, entry_address
, a_name
, new_name
)
327 unsigned data_start
, bss_start
, entry_address
;
332 unsigned int bss_end
;
334 struct scnhdr section
[MAX_SECTIONS
];
335 struct scnhdr
* f_thdr
; /* Text section header */
336 struct scnhdr
* f_dhdr
; /* Data section header */
337 struct scnhdr
* f_bhdr
; /* Bss section header */
338 struct scnhdr
* f_lhdr
; /* Loader section header */
339 struct scnhdr
* f_tchdr
; /* Typechk section header */
340 struct scnhdr
* f_dbhdr
; /* Debug section header */
341 struct scnhdr
* f_xhdr
; /* Except section header */
343 load_scnptr
= orig_load_scnptr
= lnnoptr
= 0;
344 pagemask
= getpagesize () - 1;
346 /* Adjust text/data boundary. */
348 data_start
= (long) start_of_data ();
349 #endif /* NO_REMAP */
350 data_start
= ADDR_CORRECT (data_start
);
353 data_start
= data_start
& ~SEGMENT_MASK
; /* (Down) to segment boundary. */
355 data_start
= data_start
& ~pagemask
; /* (Down) to page boundary. */
359 bss_end
= ADDR_CORRECT (sbrk (0)) + pagemask
;
360 bss_end
&= ~ pagemask
;
361 /* Adjust data/bss boundary. */
364 bss_start
= (ADDR_CORRECT (bss_start
) + pagemask
);
365 /* (Up) to page bdry. */
366 bss_start
&= ~ pagemask
;
367 if (bss_start
> bss_end
)
369 ERROR1 ("unexec: Specified bss_start (%u) is past end of program",
376 if (data_start
> bss_start
) /* Can't have negative data size. */
378 ERROR2 ("unexec: data_start (%u) can't be greater than bss_start (%u)",
379 data_start
, bss_start
);
383 /* Salvage as much info from the existing file as possible */
384 block_copy_start
= 0;
385 f_thdr
= NULL
; f_dhdr
= NULL
; f_bhdr
= NULL
;
386 f_lhdr
= NULL
; f_tchdr
= NULL
; f_dbhdr
= NULL
; f_xhdr
= NULL
;
389 if (read (a_out
, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
393 block_copy_start
+= sizeof (f_hdr
);
394 if (f_hdr
.f_opthdr
> 0)
396 if (read (a_out
, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
400 block_copy_start
+= sizeof (f_ohdr
);
402 if (f_hdr
.f_nscns
> MAX_SECTIONS
)
404 ERROR0 ("unexec: too many section headers -- increase MAX_SECTIONS");
406 /* Loop through section headers */
407 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
408 struct scnhdr
*s
= §ion
[scns
];
409 if (read (a_out
, s
, sizeof (*s
)) != sizeof (*s
))
413 if (s
->s_scnptr
> 0L)
415 if (block_copy_start
< s
->s_scnptr
+ s
->s_size
)
416 block_copy_start
= s
->s_scnptr
+ s
->s_size
;
419 #define CHECK_SCNHDR(ptr, name, flags) \
420 if (strcmp(s->s_name, name) == 0) { \
421 if (s->s_flags != flags) { \
422 fprintf(stderr, "unexec: %lx flags where %x expected in %s section.\n", \
423 (unsigned long)s->s_flags, flags, name); \
426 fprintf(stderr, "unexec: duplicate section header for section %s.\n", \
431 CHECK_SCNHDR(f_thdr
, _TEXT
, STYP_TEXT
);
432 CHECK_SCNHDR(f_dhdr
, _DATA
, STYP_DATA
);
433 CHECK_SCNHDR(f_bhdr
, _BSS
, STYP_BSS
);
434 CHECK_SCNHDR(f_lhdr
, _LOADER
, STYP_LOADER
);
435 CHECK_SCNHDR(f_dbhdr
, _DEBUG
, STYP_DEBUG
);
436 CHECK_SCNHDR(f_tchdr
, _TYPCHK
, STYP_TYPCHK
);
437 CHECK_SCNHDR(f_xhdr
, _EXCEPT
, STYP_EXCEPT
);
442 ERROR1 ("unexec: couldn't find \"%s\" section", _TEXT
);
446 ERROR1 ("unexec: couldn't find \"%s\" section", _DATA
);
450 ERROR1 ("unexec: couldn't find \"%s\" section", _BSS
);
455 ERROR0 ("can't build a COFF file from scratch yet");
457 orig_data_scnptr
= f_dhdr
->s_scnptr
;
458 orig_load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
460 /* Now we alter the contents of all the f_*hdr variables
461 to correspond to what we want to dump. */
462 f_hdr
.f_flags
|= (F_RELFLG
| F_EXEC
); /* Why? */
464 f_ohdr
.magic
= EXEC_MAGIC
;
467 f_ohdr
.tsize
= data_start
- f_ohdr
.text_start
;
468 f_ohdr
.text_start
= (long) start_of_text ();
470 f_ohdr
.dsize
= bss_start
- ((unsigned) &_data
);
471 f_ohdr
.bsize
= bss_end
- bss_start
;
473 f_dhdr
->s_size
= f_ohdr
.dsize
;
474 f_bhdr
->s_size
= f_ohdr
.bsize
;
475 f_bhdr
->s_paddr
= f_ohdr
.dsize
;
476 f_bhdr
->s_vaddr
= f_ohdr
.dsize
;
482 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
483 struct scnhdr
*s
= §ion
[scns
];
487 if (s
->s_scnptr
!= 0)
492 if ((s
->s_flags
& 0xffff) == STYP_PAD
)
495 * the text_start should probably be o_algntext but that doesn't
498 if (f_ohdr
.text_start
!= 0) /* && scns != 0 */
500 s
->s_size
= 512 - (s
->s_scnptr
% 512);
501 if (s
->s_size
== 512)
506 ptr
= ptr
+ s
->s_size
;
509 bias
= ptr
- block_copy_start
;
512 /* fix other pointers */
513 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
514 struct scnhdr
*s
= §ion
[scns
];
516 if (s
->s_relptr
!= 0)
520 if (s
->s_lnnoptr
!= 0)
522 if (lnnoptr
== 0) lnnoptr
= s
->s_lnnoptr
;
523 s
->s_lnnoptr
+= bias
;
527 if (f_hdr
.f_symptr
> 0L)
529 f_hdr
.f_symptr
+= bias
;
532 data_st
= data_start
;
533 text_scnptr
= f_thdr
->s_scnptr
;
534 data_scnptr
= f_dhdr
->s_scnptr
;
535 load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
536 block_copy_start
= orig_load_scnptr
;
538 #ifdef ADJUST_EXEC_HEADER
540 #endif /* ADJUST_EXEC_HEADER */
542 if (write (new, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
547 if (f_hdr
.f_opthdr
> 0)
549 if (write (new, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
555 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
556 struct scnhdr
*s
= §ion
[scns
];
557 if (write (new, s
, sizeof (*s
)) != sizeof (*s
))
568 /* ****************************************************************
571 * Copy the text and data segments from memory to the new a.out
574 copy_text_and_data (new)
580 lseek (new, (long) text_scnptr
, 0);
581 ptr
= start_of_text () + text_scnptr
;
582 end
= ptr
+ f_ohdr
.tsize
;
583 write_segment (new, ptr
, end
);
585 lseek (new, (long) data_scnptr
, 0);
586 ptr
= (char *) &_data
;
587 end
= ptr
+ f_ohdr
.dsize
;
588 write_segment (new, ptr
, end
);
593 write_segment (new, ptr
, end
)
595 register char *ptr
, *end
;
597 register int i
, nwrite
, ret
;
602 bzero (zeros
, sizeof zeros
);
604 for (i
= 0; ptr
< end
;)
606 /* distance to next multiple of 128. */
607 nwrite
= (((int) ptr
+ 128) & -128) - (int) ptr
;
608 /* But not beyond specified end. */
609 if (nwrite
> end
- ptr
) nwrite
= end
- ptr
;
610 ret
= write (new, ptr
, nwrite
);
611 /* If write gets a page fault, it means we reached
612 a gap between the old text segment and the old data segment.
613 This gap has probably been remapped into part of the text segment.
614 So write zeros for it. */
615 if (ret
== -1 && errno
== EFAULT
)
617 write (new, zeros
, nwrite
);
619 else if (nwrite
!= ret
)
622 "unexec write failure: addr 0x%lx, fileno %d, size 0x%x, wrote 0x%x, errno %d",
623 (unsigned long)ptr
, new, nwrite
, ret
, errno
);
631 /* ****************************************************************
634 * Copy the relocation information and symbol table from the a.out to the new
637 copy_sym (new, a_out
, a_name
, new_name
)
639 char *a_name
, *new_name
;
647 if (SYMS_START
== 0L)
650 if (lnnoptr
&& lnnoptr
< SYMS_START
) /* if there is line number info */
651 lseek (a_out
, lnnoptr
, 0); /* start copying from there */
653 lseek (a_out
, SYMS_START
, 0); /* Position a.out to symtab. */
655 while ((n
= read (a_out
, page
, sizeof page
)) > 0)
657 if (write (new, page
, n
) != n
)
669 /* ****************************************************************
672 * After successfully building the new a.out, mark it executable
680 int new = 0; /* for PERROR */
684 if (stat (name
, &sbuf
) == -1)
688 sbuf
.st_mode
|= 0111 & ~um
;
689 if (chmod (name
, sbuf
.st_mode
) == -1)
694 * If the COFF file contains a symbol table and a line number section,
695 * then any auxiliary entries that have values for x_lnnoptr must
696 * be adjusted by the amount that the line number section has moved
697 * in the file (bias computed in make_hdr). The #@$%&* designers of
698 * the auxiliary entry structures used the absolute file offsets for
699 * the line number entry rather than an offset from the start of the
700 * line number section!
702 * When I figure out how to scan through the symbol table and pick out
703 * the auxiliary entries that need adjustment, this routine will
704 * be fixed. As it is now, all such entries are wrong and sdb
705 * will complain. Fred Fish, UniSoft Systems Inc.
710 /* This function is probably very slow. Instead of reopening the new
711 file for input and output it should copy from the old to the new
712 using the two descriptors already open (WRITEDESC and READDESC).
713 Instead of reading one small structure at a time it should use
714 a reasonable size buffer. But I don't have time to work on such
715 things, so I am installing it as submitted to me. -- RMS. */
717 adjust_lnnoptrs (writedesc
, readdesc
, new_name
)
729 struct syment symentry
;
730 union auxent auxentry
;
733 if (!lnnoptr
|| !f_hdr
.f_symptr
)
736 if ((new = open (new_name
, 2)) < 0)
742 lseek (new, f_hdr
.f_symptr
, 0);
743 for (nsyms
= 0; nsyms
< f_hdr
.f_nsyms
; nsyms
++)
745 read (new, &symentry
, SYMESZ
);
746 for (naux
= 0; naux
< symentry
.n_numaux
; naux
++)
748 read (new, &auxentry
, AUXESZ
);
750 if (ISFCN (symentry
.n_type
)) {
751 auxentry
.x_sym
.x_fcnary
.x_fcn
.x_lnnoptr
+= bias
;
752 lseek (new, -AUXESZ
, 1);
753 write (new, &auxentry
, AUXESZ
);
764 /* It is probably a false economy to optimise this routine (it used to
765 read one LDREL and do do two lseeks per iteration) but the wrath of
766 RMS (see above :-) would be too much to bear */
768 unrelocate_symbols (new, a_out
, a_name
, new_name
)
770 char *a_name
, *new_name
;
774 register LDREL
*ldrel
;
776 LDREL ldrel_buf
[20];
777 ulong t_start
= (ulong
) &_text
;
778 ulong d_start
= (ulong
) &_data
;
782 if (load_scnptr
== 0)
785 lseek (a_out
, orig_load_scnptr
, 0);
786 if (read (a_out
, &ldhdr
, sizeof (ldhdr
)) != sizeof (ldhdr
))
791 #define SYMNDX_TEXT 0
792 #define SYMNDX_DATA 1
795 for (i
= 0; i
< ldhdr
.l_nreloc
; i
++, l
--, ldrel
++)
799 orig_load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
802 l
= ldhdr
.l_nreloc
- i
;
803 if (l
> sizeof (ldrel_buf
) / LDRELSZ
)
804 l
= sizeof (ldrel_buf
) / LDRELSZ
;
806 if (read (a_out
, ldrel_buf
, l
* LDRELSZ
) != l
* LDRELSZ
)
814 /* this code may not be necessary */
815 /* I originally had == in the "assignment" and it still unrelocated */
817 /* move the BSS loader symbols to the DATA segment */
818 if (ldrel
->l_rsecnm
== f_ohdr
.o_snbss
)
819 ldrel
->l_rsecnm
= f_ohdr
.o_sndata
, dirty
++;
821 if (ldrel
->l_symndx
== SYMNDX_BSS
)
822 ldrel
->l_symndx
= SYMNDX_DATA
, dirty
++;
827 load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
830 if (write (new, ldrel
, LDRELSZ
) != LDRELSZ
)
836 if (ldrel
->l_rsecnm
== f_ohdr
.o_sndata
)
841 lseek (a_out
, orig_data_scnptr
+ (ldrel
->l_vaddr
- d_start
), 0);
843 lseek (a_out
, orig_data_scnptr
+ ldrel
->l_vaddr
, 0);
846 if (read (a_out
, (void *) &orig_int
, sizeof (orig_int
)) != sizeof (orig_int
))
851 switch (ldrel
->l_symndx
) {
854 p
= (int *) (ldrel
->l_vaddr
);
857 p
= (int *) (d_start
+ ldrel
->l_vaddr
);
858 orig_int
= * p
- (t_start
- f_ohdr
.text_start
);
865 p
= (int *) (ldrel
->l_vaddr
);
868 p
= (int *) (d_start
+ ldrel
->l_vaddr
);
869 orig_int
= * p
- (d_start
- f_ohdr
.data_start
);
875 lseek (new, data_scnptr
+ (ldrel
->l_vaddr
- d_start
), 0);
877 lseek (new, data_scnptr
+ ldrel
->l_vaddr
, 0);
879 if (write (new, (void *) &orig_int
, sizeof (orig_int
)) != sizeof (orig_int
))