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 program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 In other words, you are welcome to use, share and improve this program.
20 You are forbidden to forbid anyone else to use, share and improve
21 what you give them. Help stamp out software-hoarding! */
25 * unexec.c - Convert a running program into an a.out file.
27 * Author: Spencer W. Thomas
28 * Computer Science Dept.
30 * Date: Tue Mar 2 1982
31 * Modified heavily since then.
34 * unexec (new_name, a_name, data_start, bss_start, entry_address)
35 * char *new_name, *a_name;
36 * unsigned data_start, bss_start, entry_address;
38 * Takes a snapshot of the program and makes an a.out format file in the
39 * file named by the string argument new_name.
40 * If a_name is non-NULL, the symbol table will be taken from the given file.
41 * On some machines, an existing a_name file is required.
43 * The boundaries within the a.out file may be adjusted with the data_start
44 * and bss_start arguments. Either or both may be given as 0 for defaults.
46 * Data_start gives the boundary between the text segment and the data
47 * segment of the program. The text segment can contain shared, read-only
48 * program code and literal data, while the data segment is always unshared
49 * and unprotected. Data_start gives the lowest unprotected address.
50 * The value you specify may be rounded down to a suitable boundary
51 * as required by the machine you are using.
53 * Specifying zero for data_start means the boundary between text and data
54 * should not be the same as when the program was loaded.
55 * If NO_REMAP is defined, the argument data_start is ignored and the
56 * segment boundaries are never changed.
58 * Bss_start indicates how much of the data segment is to be saved in the
59 * a.out file and restored when the program is executed. It gives the lowest
60 * unsaved address, and is rounded up to a page boundary. The default when 0
61 * is given assumes that the entire data segment is to be stored, including
62 * the previous data and bss as well as any additional storage allocated with
65 * The new file is set up to start at entry_address.
67 * If you make improvements I'd like to get them too.
68 * harpo!utah-cs!thomas, thomas@Utah-20
72 /* There are several compilation parameters affecting unexec:
76 Define this if your system uses COFF for executables.
77 Otherwise we assume you use Berkeley format.
81 Define this if you do not want to try to save Emacs's pure data areas
82 as part of the text segment.
84 Saving them as text is good because it allows users to share more.
86 However, on machines that locate the text area far from the data area,
87 the boundary cannot feasibly be moved. Such machines require
90 Also, remapping can cause trouble with the built-in startup routine
91 /lib/crt0.o, which defines `environ' as an initialized variable.
92 Dumping `environ' as pure does not work! So, to use remapping,
93 you must write a startup routine for your machine in Emacs's crt0.c.
94 If NO_REMAP is defined, Emacs uses the system's crt0.o.
98 Some machines that use COFF executables require that each section
99 start on a certain boundary *in the COFF file*. Such machines should
100 define SECTION_ALIGNMENT to a mask of the low-order bits that must be
101 zero on such a boundary. This mask is used to control padding between
102 segments in the COFF file.
104 If SECTION_ALIGNMENT is not defined, the segments are written
105 consecutively with no attempt at alignment. This is right for
110 Some machines require that the beginnings and ends of segments
111 *in core* be on certain boundaries. For most machines, a page
112 boundary is sufficient. That is the default. When a larger
113 boundary is needed, define SEGMENT_MASK to a mask of
114 the bits that must be zero on such a boundary.
118 Some machines count the a.out header as part of the size of the text
119 segment (a_text); they may actually load the header into core as the
120 first data in the text segment. Some have additional padding between
121 the header and the real text of the program that is counted in a_text.
123 For these machines, define A_TEXT_OFFSET(HDR) to examine the header
124 structure HDR and return the number of bytes to add to `a_text'
125 before writing it (above and beyond the number of bytes of actual
126 program text). HDR's standard fields are already correct, except that
127 this adjustment to the `a_text' field has not yet been made;
128 thus, the amount of offset can depend on the data in the file.
132 If defined, this macro specifies the number of bytes to seek into the
133 a.out file before starting to write the text segment.a
137 For machines using COFF, this macro, if defined, is a value stored
138 into the magic number field of the output file.
142 This macro can be used to generate statements to adjust or
143 initialize nonstandard fields in the file header
147 Macro to correct an int which is the bit pattern of a pointer to a byte
148 into an int which is the number of a byte.
150 This macro has a default definition which is usually right.
151 This default definition is a no-op on most machines (where a
152 pointer looks like an int) but not on all machines.
161 #define PERROR(arg) perror (arg); return -1
164 #define PERROR(file) report_error (file, new)
168 /* Define getpagesize () if the system does not.
169 Note that this may depend on symbols defined in a.out.h
171 #include "getpagesize.h"
173 #ifndef makedev /* Try to detect types.h already loaded */
174 #include <sys/types.h>
177 #include <sys/stat.h>
180 extern char *start_of_text (); /* Start of text */
181 extern char *start_of_data (); /* Start of initialized data */
193 /* I have a suspicion that these are turned off on all systems
194 and can be deleted. Try it in version 19. */
199 #endif /* not sun386 */
200 #endif /* not UMAX */
201 #endif /* Not STRIDE */
203 static long block_copy_start
; /* Old executable start point */
204 static struct filehdr f_hdr
; /* File header */
205 static struct aouthdr f_ohdr
; /* Optional file header (a.out) */
206 long bias
; /* Bias to add for growth */
207 long lnnoptr
; /* Pointer to line-number info within file */
208 #define SYMS_START block_copy_start
210 static long text_scnptr
;
211 static long data_scnptr
;
213 static long load_scnptr
;
214 static long orig_load_scnptr
;
215 static long orig_data_scnptr
;
220 #define MAX_SECTIONS 10
227 /* Correct an int which is the bit pattern of a pointer to a byte
228 into an int which is the number of a byte.
229 This is a no-op on ordinary machines, but not on all. */
231 #ifndef ADDR_CORRECT /* Let m-*.h files override this definition */
232 #define ADDR_CORRECT(x) ((char *)(x) - (char*)0)
239 report_error (file
, fd
)
245 report_file_error ("Cannot unexec", Fcons (build_string (file
), Qnil
));
249 #define ERROR0(msg) report_error_1 (new, msg, 0, 0); return -1
250 #define ERROR1(msg,x) report_error_1 (new, msg, x, 0); return -1
251 #define ERROR2(msg,x,y) report_error_1 (new, msg, x, y); return -1
254 report_error_1 (fd
, msg
, a1
, a2
)
263 fprintf (stderr
, msg
, a1
, a2
);
264 fprintf (stderr
, "\n");
268 static int make_hdr ();
269 static void mark_x ();
270 static int copy_text_and_data ();
271 static int copy_sym ();
273 /* ****************************************************************
278 unexec (new_name
, a_name
, data_start
, bss_start
, entry_address
)
279 char *new_name
, *a_name
;
280 unsigned data_start
, bss_start
, entry_address
;
284 if (a_name
&& (a_out
= open (a_name
, 0)) < 0)
288 if ((new = creat (new_name
, 0666)) < 0)
292 if (make_hdr (new,a_out
,data_start
,bss_start
,entry_address
,a_name
,new_name
) < 0
293 || copy_text_and_data (new) < 0
294 || copy_sym (new, a_out
, a_name
, new_name
) < 0
296 || adjust_lnnoptrs (new, a_out
, new_name
) < 0
299 || unrelocate_symbols (new, a_out
, a_name
, new_name
) < 0
304 /* unlink (new_name); /* Failed, unlink new a.out */
315 /* ****************************************************************
318 * Make the header in the new a.out from the header in core.
319 * Modify the text and data sizes.
322 make_hdr (new, a_out
, data_start
, bss_start
, entry_address
, a_name
, new_name
)
324 unsigned data_start
, bss_start
, entry_address
;
329 unsigned int bss_end
;
331 struct scnhdr section
[MAX_SECTIONS
];
332 struct scnhdr
* f_thdr
; /* Text section header */
333 struct scnhdr
* f_dhdr
; /* Data section header */
334 struct scnhdr
* f_bhdr
; /* Bss section header */
335 struct scnhdr
* f_lhdr
; /* Loader section header */
336 struct scnhdr
* f_tchdr
; /* Typechk section header */
337 struct scnhdr
* f_dbhdr
; /* Debug section header */
338 struct scnhdr
* f_xhdr
; /* Except section header */
340 load_scnptr
= orig_load_scnptr
= lnnoptr
= 0;
341 pagemask
= getpagesize () - 1;
343 /* Adjust text/data boundary. */
345 data_start
= (long) start_of_data ();
346 #endif /* NO_REMAP */
347 data_start
= ADDR_CORRECT (data_start
);
350 data_start
= data_start
& ~SEGMENT_MASK
; /* (Down) to segment boundary. */
352 data_start
= data_start
& ~pagemask
; /* (Down) to page boundary. */
356 bss_end
= ADDR_CORRECT (sbrk (0)) + pagemask
;
357 bss_end
&= ~ pagemask
;
358 /* Adjust data/bss boundary. */
361 bss_start
= (ADDR_CORRECT (bss_start
) + pagemask
);
362 /* (Up) to page bdry. */
363 bss_start
&= ~ pagemask
;
364 if (bss_start
> bss_end
)
366 ERROR1 ("unexec: Specified bss_start (%u) is past end of program",
373 if (data_start
> bss_start
) /* Can't have negative data size. */
375 ERROR2 ("unexec: data_start (%u) can't be greater than bss_start (%u)",
376 data_start
, bss_start
);
380 /* Salvage as much info from the existing file as possible */
381 block_copy_start
= 0;
382 f_thdr
= NULL
; f_dhdr
= NULL
; f_bhdr
= NULL
;
383 f_lhdr
= NULL
; f_tchdr
= NULL
; f_dbhdr
= NULL
; f_xhdr
= NULL
;
386 if (read (a_out
, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
390 block_copy_start
+= sizeof (f_hdr
);
391 if (f_hdr
.f_opthdr
> 0)
393 if (read (a_out
, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
397 block_copy_start
+= sizeof (f_ohdr
);
399 if (f_hdr
.f_nscns
> MAX_SECTIONS
)
401 ERROR0 ("unexec: too many section headers -- increase MAX_SECTIONS");
403 /* Loop through section headers */
404 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
405 struct scnhdr
*s
= §ion
[scns
];
406 if (read (a_out
, s
, sizeof (*s
)) != sizeof (*s
))
410 if (s
->s_scnptr
> 0L)
412 if (block_copy_start
< s
->s_scnptr
+ s
->s_size
)
413 block_copy_start
= s
->s_scnptr
+ s
->s_size
;
416 #define CHECK_SCNHDR(ptr, name, flags) \
417 if (strcmp(s->s_name, name) == 0) { \
418 if (s->s_flags != flags) { \
419 fprintf(stderr, "unexec: %lx flags where %x expected in %s section.\n", \
420 (unsigned long)s->s_flags, flags, name); \
423 fprintf(stderr, "unexec: duplicate section header for section %s.\n", \
428 CHECK_SCNHDR(f_thdr
, _TEXT
, STYP_TEXT
);
429 CHECK_SCNHDR(f_dhdr
, _DATA
, STYP_DATA
);
430 CHECK_SCNHDR(f_bhdr
, _BSS
, STYP_BSS
);
431 CHECK_SCNHDR(f_lhdr
, _LOADER
, STYP_LOADER
);
432 CHECK_SCNHDR(f_dbhdr
, _DEBUG
, STYP_DEBUG
);
433 CHECK_SCNHDR(f_tchdr
, _TYPCHK
, STYP_TYPCHK
);
434 CHECK_SCNHDR(f_xhdr
, _EXCEPT
, STYP_EXCEPT
);
439 ERROR1 ("unexec: couldn't find \"%s\" section", _TEXT
);
443 ERROR1 ("unexec: couldn't find \"%s\" section", _DATA
);
447 ERROR1 ("unexec: couldn't find \"%s\" section", _BSS
);
452 ERROR0 ("can't build a COFF file from scratch yet");
454 orig_data_scnptr
= f_dhdr
->s_scnptr
;
455 orig_load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
457 /* Now we alter the contents of all the f_*hdr variables
458 to correspond to what we want to dump. */
459 f_hdr
.f_flags
|= (F_RELFLG
| F_EXEC
); /* Why? */
461 f_ohdr
.magic
= EXEC_MAGIC
;
464 f_ohdr
.tsize
= data_start
- f_ohdr
.text_start
;
465 f_ohdr
.text_start
= (long) start_of_text ();
467 f_ohdr
.dsize
= bss_start
- ((unsigned) &_data
);
468 f_ohdr
.bsize
= bss_end
- bss_start
;
470 f_dhdr
->s_size
= f_ohdr
.dsize
;
471 f_bhdr
->s_size
= f_ohdr
.bsize
;
472 f_bhdr
->s_paddr
= f_ohdr
.dsize
;
473 f_bhdr
->s_vaddr
= f_ohdr
.dsize
;
479 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
480 struct scnhdr
*s
= §ion
[scns
];
484 if (s
->s_scnptr
!= 0)
489 if ((s
->s_flags
& 0xffff) == STYP_PAD
)
492 * the text_start should probably be o_algntext but that doesn't
495 if (f_ohdr
.text_start
!= 0) /* && scns != 0 */
497 s
->s_size
= 512 - (s
->s_scnptr
% 512);
498 if (s
->s_size
== 512)
503 ptr
= ptr
+ s
->s_size
;
506 bias
= ptr
- block_copy_start
;
509 /* fix other pointers */
510 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
511 struct scnhdr
*s
= §ion
[scns
];
513 if (s
->s_relptr
!= 0)
517 if (s
->s_lnnoptr
!= 0)
519 if (lnnoptr
== 0) lnnoptr
= s
->s_lnnoptr
;
520 s
->s_lnnoptr
+= bias
;
524 if (f_hdr
.f_symptr
> 0L)
526 f_hdr
.f_symptr
+= bias
;
529 data_st
= data_start
;
530 text_scnptr
= f_thdr
->s_scnptr
;
531 data_scnptr
= f_dhdr
->s_scnptr
;
532 load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
533 block_copy_start
= orig_load_scnptr
;
535 #ifdef ADJUST_EXEC_HEADER
537 #endif /* ADJUST_EXEC_HEADER */
539 if (write (new, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
544 if (f_hdr
.f_opthdr
> 0)
546 if (write (new, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
552 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
553 struct scnhdr
*s
= §ion
[scns
];
554 if (write (new, s
, sizeof (*s
)) != sizeof (*s
))
565 /* ****************************************************************
568 * Copy the text and data segments from memory to the new a.out
571 copy_text_and_data (new)
577 lseek (new, (long) text_scnptr
, 0);
578 ptr
= start_of_text () + text_scnptr
;
579 end
= ptr
+ f_ohdr
.tsize
;
580 write_segment (new, ptr
, end
);
582 lseek (new, (long) data_scnptr
, 0);
583 ptr
= (char *) &_data
;
584 end
= ptr
+ f_ohdr
.dsize
;
585 write_segment (new, ptr
, end
);
590 write_segment (new, ptr
, end
)
592 register char *ptr
, *end
;
594 register int i
, nwrite
, ret
;
599 bzero (zeros
, sizeof zeros
);
601 for (i
= 0; ptr
< end
;)
603 /* distance to next multiple of 128. */
604 nwrite
= (((int) ptr
+ 128) & -128) - (int) ptr
;
605 /* But not beyond specified end. */
606 if (nwrite
> end
- ptr
) nwrite
= end
- ptr
;
607 ret
= write (new, ptr
, nwrite
);
608 /* If write gets a page fault, it means we reached
609 a gap between the old text segment and the old data segment.
610 This gap has probably been remapped into part of the text segment.
611 So write zeros for it. */
612 if (ret
== -1 && errno
== EFAULT
)
614 write (new, zeros
, nwrite
);
616 else if (nwrite
!= ret
)
619 "unexec write failure: addr 0x%lx, fileno %d, size 0x%x, wrote 0x%x, errno %d",
620 (unsigned long)ptr
, new, nwrite
, ret
, errno
);
628 /* ****************************************************************
631 * Copy the relocation information and symbol table from the a.out to the new
634 copy_sym (new, a_out
, a_name
, new_name
)
636 char *a_name
, *new_name
;
644 if (SYMS_START
== 0L)
647 if (lnnoptr
&& lnnoptr
< SYMS_START
) /* if there is line number info */
648 lseek (a_out
, lnnoptr
, 0); /* start copying from there */
650 lseek (a_out
, SYMS_START
, 0); /* Position a.out to symtab. */
652 while ((n
= read (a_out
, page
, sizeof page
)) > 0)
654 if (write (new, page
, n
) != n
)
666 /* ****************************************************************
669 * After successfully building the new a.out, mark it executable
677 int new = 0; /* for PERROR */
681 if (stat (name
, &sbuf
) == -1)
685 sbuf
.st_mode
|= 0111 & ~um
;
686 if (chmod (name
, sbuf
.st_mode
) == -1)
691 * If the COFF file contains a symbol table and a line number section,
692 * then any auxiliary entries that have values for x_lnnoptr must
693 * be adjusted by the amount that the line number section has moved
694 * in the file (bias computed in make_hdr). The #@$%&* designers of
695 * the auxiliary entry structures used the absolute file offsets for
696 * the line number entry rather than an offset from the start of the
697 * line number section!
699 * When I figure out how to scan through the symbol table and pick out
700 * the auxiliary entries that need adjustment, this routine will
701 * be fixed. As it is now, all such entries are wrong and sdb
702 * will complain. Fred Fish, UniSoft Systems Inc.
707 /* This function is probably very slow. Instead of reopening the new
708 file for input and output it should copy from the old to the new
709 using the two descriptors already open (WRITEDESC and READDESC).
710 Instead of reading one small structure at a time it should use
711 a reasonable size buffer. But I don't have time to work on such
712 things, so I am installing it as submitted to me. -- RMS. */
714 adjust_lnnoptrs (writedesc
, readdesc
, new_name
)
726 struct syment symentry
;
727 union auxent auxentry
;
730 if (!lnnoptr
|| !f_hdr
.f_symptr
)
733 if ((new = open (new_name
, 2)) < 0)
739 lseek (new, f_hdr
.f_symptr
, 0);
740 for (nsyms
= 0; nsyms
< f_hdr
.f_nsyms
; nsyms
++)
742 read (new, &symentry
, SYMESZ
);
743 for (naux
= 0; naux
< symentry
.n_numaux
; naux
++)
745 read (new, &auxentry
, AUXESZ
);
747 if (ISFCN (symentry
.n_type
)) {
748 auxentry
.x_sym
.x_fcnary
.x_fcn
.x_lnnoptr
+= bias
;
749 lseek (new, -AUXESZ
, 1);
750 write (new, &auxentry
, AUXESZ
);
761 /* It is probably a false economy to optimise this routine (it used to
762 read one LDREL and do do two lseeks per iteration) but the wrath of
763 RMS (see above :-) would be too much to bear */
765 unrelocate_symbols (new, a_out
, a_name
, new_name
)
767 char *a_name
, *new_name
;
771 register LDREL
*ldrel
;
773 LDREL ldrel_buf
[20];
774 ulong t_start
= (ulong
) &_text
;
775 ulong d_start
= (ulong
) &_data
;
779 if (load_scnptr
== 0)
782 lseek (a_out
, orig_load_scnptr
, 0);
783 if (read (a_out
, &ldhdr
, sizeof (ldhdr
)) != sizeof (ldhdr
))
788 #define SYMNDX_TEXT 0
789 #define SYMNDX_DATA 1
792 for (i
= 0; i
< ldhdr
.l_nreloc
; i
++, l
--, ldrel
++)
796 orig_load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
799 l
= ldhdr
.l_nreloc
- i
;
800 if (l
> sizeof (ldrel_buf
) / LDRELSZ
)
801 l
= sizeof (ldrel_buf
) / LDRELSZ
;
803 if (read (a_out
, ldrel_buf
, l
* LDRELSZ
) != l
* LDRELSZ
)
811 /* this code may not be necessary */
812 /* I originally had == in the "assignment" and it still unrelocated */
814 /* move the BSS loader symbols to the DATA segment */
815 if (ldrel
->l_rsecnm
== f_ohdr
.o_snbss
)
816 ldrel
->l_rsecnm
= f_ohdr
.o_sndata
, dirty
++;
818 if (ldrel
->l_symndx
== SYMNDX_BSS
)
819 ldrel
->l_symndx
= SYMNDX_DATA
, dirty
++;
824 load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
827 if (write (new, ldrel
, LDRELSZ
) != LDRELSZ
)
833 if (ldrel
->l_rsecnm
== f_ohdr
.o_sndata
)
838 lseek (a_out
, orig_data_scnptr
+ (ldrel
->l_vaddr
- d_start
), 0);
840 lseek (a_out
, orig_data_scnptr
+ ldrel
->l_vaddr
, 0);
843 if (read (a_out
, (void *) &orig_int
, sizeof (orig_int
)) != sizeof (orig_int
))
848 switch (ldrel
->l_symndx
) {
851 p
= (int *) (ldrel
->l_vaddr
);
854 p
= (int *) (d_start
+ ldrel
->l_vaddr
);
855 orig_int
= * p
- (t_start
- f_ohdr
.text_start
);
862 p
= (int *) (ldrel
->l_vaddr
);
865 p
= (int *) (d_start
+ ldrel
->l_vaddr
);
866 orig_int
= * p
- (d_start
- f_ohdr
.data_start
);
872 lseek (new, data_scnptr
+ (ldrel
->l_vaddr
- d_start
), 0);
874 lseek (new, data_scnptr
+ ldrel
->l_vaddr
, 0);
876 if (write (new, (void *) &orig_int
, sizeof (orig_int
)) != sizeof (orig_int
))