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.
36 * Updated for AIX 4.1.3 by Bill_Mann @ PraxisInt.com, Feb 1996
37 * As of AIX 4.1, text, data, and bss are pre-relocated by the binder in
38 * such a way that the file can be mapped with code in one segment and
39 * data/bss in another segment, without reading or copying the file, by
40 * the AIX exec loader. Padding sections are omitted, nevertheless
41 * small amounts of 'padding' still occurs between sections in the file.
42 * As modified, this code handles both 3.2 and 4.1 conventions.
45 * unexec (new_name, a_name, data_start, bss_start, entry_address)
46 * char *new_name, *a_name;
47 * unsigned data_start, bss_start, entry_address;
49 * Takes a snapshot of the program and makes an a.out format file in the
50 * file named by the string argument new_name.
51 * If a_name is non-NULL, the symbol table will be taken from the given file.
52 * On some machines, an existing a_name file is required.
54 * The boundaries within the a.out file may be adjusted with the data_start
55 * and bss_start arguments. Either or both may be given as 0 for defaults.
57 * Data_start gives the boundary between the text segment and the data
58 * segment of the program. The text segment can contain shared, read-only
59 * program code and literal data, while the data segment is always unshared
60 * and unprotected. Data_start gives the lowest unprotected address.
61 * The value you specify may be rounded down to a suitable boundary
62 * as required by the machine you are using.
64 * Specifying zero for data_start means the boundary between text and data
65 * should not be the same as when the program was loaded.
66 * If NO_REMAP is defined, the argument data_start is ignored and the
67 * segment boundaries are never changed.
69 * Bss_start indicates how much of the data segment is to be saved in the
70 * a.out file and restored when the program is executed. It gives the lowest
71 * unsaved address, and is rounded up to a page boundary. The default when 0
72 * is given assumes that the entire data segment is to be stored, including
73 * the previous data and bss as well as any additional storage allocated with
76 * The new file is set up to start at entry_address.
78 * If you make improvements I'd like to get them too.
79 * harpo!utah-cs!thomas, thomas@Utah-20
83 /* There are several compilation parameters affecting unexec:
87 Define this if your system uses COFF for executables.
88 Otherwise we assume you use Berkeley format.
92 Define this if you do not want to try to save Emacs's pure data areas
93 as part of the text segment.
95 Saving them as text is good because it allows users to share more.
97 However, on machines that locate the text area far from the data area,
98 the boundary cannot feasibly be moved. Such machines require
101 Also, remapping can cause trouble with the built-in startup routine
102 /lib/crt0.o, which defines `environ' as an initialized variable.
103 Dumping `environ' as pure does not work! So, to use remapping,
104 you must write a startup routine for your machine in Emacs's crt0.c.
105 If NO_REMAP is defined, Emacs uses the system's crt0.o.
109 Some machines that use COFF executables require that each section
110 start on a certain boundary *in the COFF file*. Such machines should
111 define SECTION_ALIGNMENT to a mask of the low-order bits that must be
112 zero on such a boundary. This mask is used to control padding between
113 segments in the COFF file.
115 If SECTION_ALIGNMENT is not defined, the segments are written
116 consecutively with no attempt at alignment. This is right for
121 Some machines require that the beginnings and ends of segments
122 *in core* be on certain boundaries. For most machines, a page
123 boundary is sufficient. That is the default. When a larger
124 boundary is needed, define SEGMENT_MASK to a mask of
125 the bits that must be zero on such a boundary.
129 Some machines count the a.out header as part of the size of the text
130 segment (a_text); they may actually load the header into core as the
131 first data in the text segment. Some have additional padding between
132 the header and the real text of the program that is counted in a_text.
134 For these machines, define A_TEXT_OFFSET(HDR) to examine the header
135 structure HDR and return the number of bytes to add to `a_text'
136 before writing it (above and beyond the number of bytes of actual
137 program text). HDR's standard fields are already correct, except that
138 this adjustment to the `a_text' field has not yet been made;
139 thus, the amount of offset can depend on the data in the file.
143 If defined, this macro specifies the number of bytes to seek into the
144 a.out file before starting to write the text segment.a
148 For machines using COFF, this macro, if defined, is a value stored
149 into the magic number field of the output file.
153 This macro can be used to generate statements to adjust or
154 initialize nonstandard fields in the file header
158 Macro to correct an int which is the bit pattern of a pointer to a byte
159 into an int which is the number of a byte.
161 This macro has a default definition which is usually right.
162 This default definition is a no-op on most machines (where a
163 pointer looks like an int) but not on all machines.
172 #define PERROR(arg) perror (arg); return -1
175 #define PERROR(file) report_error (file, new)
179 /* Define getpagesize () if the system does not.
180 Note that this may depend on symbols defined in a.out.h
182 #include "getpagesize.h"
184 #ifndef makedev /* Try to detect types.h already loaded */
185 #include <sys/types.h>
188 #include <sys/stat.h>
191 extern char *start_of_text (); /* Start of text */
192 extern char *start_of_data (); /* Start of initialized data */
204 /* I have a suspicion that these are turned off on all systems
205 and can be deleted. Try it in version 19. */
210 #endif /* not sun386 */
211 #endif /* not UMAX */
212 #endif /* Not STRIDE */
214 static struct filehdr f_hdr
; /* File header */
215 static struct aouthdr f_ohdr
; /* Optional file header (a.out) */
216 long bias
; /* Bias to add for growth */
217 long lnnoptr
; /* Pointer to line-number info within file */
219 static long text_scnptr
;
220 static long data_scnptr
;
222 #define ALIGN(val, pwr) (((val) + ((1L<<(pwr))-1)) & ~((1L<<(pwr))-1))
223 static long load_scnptr
;
224 static long orig_load_scnptr
;
225 static long orig_data_scnptr
;
227 static ulong data_st
; /* start of data area written out */
230 #define MAX_SECTIONS 10
237 /* Correct an int which is the bit pattern of a pointer to a byte
238 into an int which is the number of a byte.
239 This is a no-op on ordinary machines, but not on all. */
241 #ifndef ADDR_CORRECT /* Let m-*.h files override this definition */
242 #define ADDR_CORRECT(x) ((char *)(x) - (char*)0)
249 report_error (file
, fd
)
255 report_file_error ("Cannot unexec", Fcons (build_string (file
), Qnil
));
259 #define ERROR0(msg) report_error_1 (new, msg, 0, 0); return -1
260 #define ERROR1(msg,x) report_error_1 (new, msg, x, 0); return -1
261 #define ERROR2(msg,x,y) report_error_1 (new, msg, x, y); return -1
264 report_error_1 (fd
, msg
, a1
, a2
)
273 fprintf (stderr
, msg
, a1
, a2
);
274 fprintf (stderr
, "\n");
278 static int make_hdr ();
279 static void mark_x ();
280 static int copy_text_and_data ();
281 static int copy_sym ();
283 /* ****************************************************************
288 unexec (new_name
, a_name
, data_start
, bss_start
, entry_address
)
289 char *new_name
, *a_name
;
290 unsigned data_start
, bss_start
, entry_address
;
294 if (a_name
&& (a_out
= open (a_name
, 0)) < 0)
298 if ((new = creat (new_name
, 0666)) < 0)
302 if (make_hdr (new,a_out
,data_start
,bss_start
,entry_address
,a_name
,new_name
) < 0
303 || copy_text_and_data (new) < 0
304 || copy_sym (new, a_out
, a_name
, new_name
) < 0
306 || adjust_lnnoptrs (new, a_out
, new_name
) < 0
309 || unrelocate_symbols (new, a_out
, a_name
, new_name
) < 0
314 /* unlink (new_name); /* Failed, unlink new a.out */
325 /* ****************************************************************
328 * Make the header in the new a.out from the header in core.
329 * Modify the text and data sizes.
332 make_hdr (new, a_out
, data_start
, bss_start
, entry_address
, a_name
, new_name
)
334 unsigned data_start
, bss_start
, entry_address
;
339 unsigned int bss_end
;
341 struct scnhdr section
[MAX_SECTIONS
];
342 struct scnhdr
* f_thdr
; /* Text section header */
343 struct scnhdr
* f_dhdr
; /* Data section header */
344 struct scnhdr
* f_bhdr
; /* Bss section header */
345 struct scnhdr
* f_lhdr
; /* Loader section header */
346 struct scnhdr
* f_tchdr
; /* Typechk section header */
347 struct scnhdr
* f_dbhdr
; /* Debug section header */
348 struct scnhdr
* f_xhdr
; /* Except section header */
350 load_scnptr
= orig_load_scnptr
= lnnoptr
= 0;
351 pagemask
= getpagesize () - 1;
353 /* Adjust text/data boundary. */
355 data_start
= (long) start_of_data ();
356 #endif /* NO_REMAP */
357 data_start
= ADDR_CORRECT (data_start
);
360 data_start
= data_start
& ~SEGMENT_MASK
; /* (Down) to segment boundary. */
362 data_start
= data_start
& ~pagemask
; /* (Down) to page boundary. */
366 bss_end
= ADDR_CORRECT (sbrk (0)) + pagemask
;
367 bss_end
&= ~ pagemask
;
368 /* Adjust data/bss boundary. */
371 bss_start
= (ADDR_CORRECT (bss_start
) + pagemask
);
372 /* (Up) to page bdry. */
373 bss_start
&= ~ pagemask
;
374 if (bss_start
> bss_end
)
376 ERROR1 ("unexec: Specified bss_start (%u) is past end of program",
383 if (data_start
> bss_start
) /* Can't have negative data size. */
385 ERROR2 ("unexec: data_start (%u) can't be greater than bss_start (%u)",
386 data_start
, bss_start
);
390 /* Salvage as much info from the existing file as possible */
391 f_thdr
= NULL
; f_dhdr
= NULL
; f_bhdr
= NULL
;
392 f_lhdr
= NULL
; f_tchdr
= NULL
; f_dbhdr
= NULL
; f_xhdr
= NULL
;
395 if (read (a_out
, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
399 if (f_hdr
.f_opthdr
> 0)
401 if (read (a_out
, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
406 if (f_hdr
.f_nscns
> MAX_SECTIONS
)
408 ERROR0 ("unexec: too many section headers -- increase MAX_SECTIONS");
410 /* Loop through section headers */
411 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
412 struct scnhdr
*s
= §ion
[scns
];
413 if (read (a_out
, s
, sizeof (*s
)) != sizeof (*s
))
418 #define CHECK_SCNHDR(ptr, name, flags) \
419 if (strcmp(s->s_name, name) == 0) { \
420 if (s->s_flags != flags) { \
421 fprintf(stderr, "unexec: %lx flags where %x expected in %s section.\n", \
422 (unsigned long)s->s_flags, flags, name); \
425 fprintf(stderr, "unexec: duplicate section header for section %s.\n", \
430 CHECK_SCNHDR(f_thdr
, _TEXT
, STYP_TEXT
);
431 CHECK_SCNHDR(f_dhdr
, _DATA
, STYP_DATA
);
432 CHECK_SCNHDR(f_bhdr
, _BSS
, STYP_BSS
);
433 CHECK_SCNHDR(f_lhdr
, _LOADER
, STYP_LOADER
);
434 CHECK_SCNHDR(f_dbhdr
, _DEBUG
, STYP_DEBUG
);
435 CHECK_SCNHDR(f_tchdr
, _TYPCHK
, STYP_TYPCHK
);
436 CHECK_SCNHDR(f_xhdr
, _EXCEPT
, STYP_EXCEPT
);
441 ERROR1 ("unexec: couldn't find \"%s\" section", _TEXT
);
445 ERROR1 ("unexec: couldn't find \"%s\" section", _DATA
);
449 ERROR1 ("unexec: couldn't find \"%s\" section", _BSS
);
454 ERROR0 ("can't build a COFF file from scratch yet");
456 orig_data_scnptr
= f_dhdr
->s_scnptr
;
457 orig_load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
459 /* Now we alter the contents of all the f_*hdr variables
460 to correspond to what we want to dump. */
462 /* Indicate that the reloc information is no longer valid for ld (bind);
463 we only update it enough to fake out the exec-time loader. */
464 f_hdr
.f_flags
|= (F_RELFLG
| F_EXEC
);
467 f_ohdr
.magic
= EXEC_MAGIC
;
470 f_ohdr
.tsize
= data_start
- f_ohdr
.text_start
;
471 f_ohdr
.text_start
= (long) start_of_text ();
473 data_st
= f_ohdr
.data_start
? f_ohdr
.data_start
: (ulong
) &_data
;
474 f_ohdr
.dsize
= bss_start
- data_st
;
475 f_ohdr
.bsize
= bss_end
- bss_start
;
477 f_dhdr
->s_size
= f_ohdr
.dsize
;
478 f_bhdr
->s_size
= f_ohdr
.bsize
;
479 f_bhdr
->s_paddr
= f_ohdr
.data_start
+ f_ohdr
.dsize
;
480 f_bhdr
->s_vaddr
= f_ohdr
.data_start
+ f_ohdr
.dsize
;
484 ulong ptr
= section
[0].s_scnptr
;
487 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++)
489 struct scnhdr
*s
= §ion
[scns
];
491 if (s
->s_flags
& STYP_PAD
) /* .pad sections omitted in AIX 4.1 */
494 * the text_start should probably be o_algntext but that doesn't
497 if (f_ohdr
.text_start
!= 0) /* && scns != 0 */
499 s
->s_size
= 512 - (ptr
% 512);
500 if (s
->s_size
== 512)
505 else if (s
->s_flags
& STYP_DATA
)
507 else if (!(s
->s_flags
& (STYP_TEXT
| STYP_BSS
)))
509 if (bias
== -1) /* if first section after bss */
510 bias
= ptr
- s
->s_scnptr
;
516 ptr
= ptr
+ s
->s_size
;
520 /* fix other pointers */
521 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++)
523 struct scnhdr
*s
= §ion
[scns
];
525 if (s
->s_relptr
!= 0)
529 if (s
->s_lnnoptr
!= 0)
531 if (lnnoptr
== 0) lnnoptr
= s
->s_lnnoptr
;
532 s
->s_lnnoptr
+= bias
;
536 if (f_hdr
.f_symptr
> 0L)
538 f_hdr
.f_symptr
+= bias
;
541 text_scnptr
= f_thdr
->s_scnptr
;
542 data_scnptr
= f_dhdr
->s_scnptr
;
543 load_scnptr
= f_lhdr
? f_lhdr
->s_scnptr
: 0;
545 #ifdef ADJUST_EXEC_HEADER
547 #endif /* ADJUST_EXEC_HEADER */
549 if (write (new, &f_hdr
, sizeof (f_hdr
)) != sizeof (f_hdr
))
554 if (f_hdr
.f_opthdr
> 0)
556 if (write (new, &f_ohdr
, sizeof (f_ohdr
)) != sizeof (f_ohdr
))
562 for (scns
= 0; scns
< f_hdr
.f_nscns
; scns
++) {
563 struct scnhdr
*s
= §ion
[scns
];
564 if (write (new, s
, sizeof (*s
)) != sizeof (*s
))
575 /* ****************************************************************
578 * Copy the text and data segments from memory to the new a.out
581 copy_text_and_data (new)
587 lseek (new, (long) text_scnptr
, 0);
588 ptr
= start_of_text () + text_scnptr
;
589 end
= ptr
+ f_ohdr
.tsize
;
590 write_segment (new, ptr
, end
);
592 lseek (new, (long) data_scnptr
, 0);
593 ptr
= (char *) data_st
;
594 end
= ptr
+ f_ohdr
.dsize
;
595 write_segment (new, ptr
, end
);
600 #define UnexBlockSz (1<<12) /* read/write block size */
601 write_segment (new, ptr
, end
)
603 register char *ptr
, *end
;
605 register int i
, nwrite
, ret
;
608 char zeros
[UnexBlockSz
];
610 for (i
= 0; ptr
< end
;)
612 /* distance to next block. */
613 nwrite
= (((int) ptr
+ UnexBlockSz
) & -UnexBlockSz
) - (int) ptr
;
614 /* But not beyond specified end. */
615 if (nwrite
> end
- ptr
) nwrite
= end
- ptr
;
616 ret
= write (new, ptr
, nwrite
);
617 /* If write gets a page fault, it means we reached
618 a gap between the old text segment and the old data segment.
619 This gap has probably been remapped into part of the text segment.
620 So write zeros for it. */
621 if (ret
== -1 && errno
== EFAULT
)
623 bzero (zeros
, nwrite
);
624 write (new, zeros
, nwrite
);
626 else if (nwrite
!= ret
)
629 "unexec write failure: addr 0x%lx, fileno %d, size 0x%x, wrote 0x%x, errno %d",
630 (unsigned long)ptr
, new, nwrite
, ret
, errno
);
638 /* ****************************************************************
641 * Copy the relocation information and symbol table from the a.out to the new
644 copy_sym (new, a_out
, a_name
, new_name
)
646 char *a_name
, *new_name
;
648 char page
[UnexBlockSz
];
654 if (orig_load_scnptr
== 0L)
657 if (lnnoptr
&& lnnoptr
< orig_load_scnptr
) /* if there is line number info */
658 lseek (a_out
, lnnoptr
, 0); /* start copying from there */
660 lseek (a_out
, orig_load_scnptr
, 0); /* Position a.out to symtab. */
662 while ((n
= read (a_out
, page
, sizeof page
)) > 0)
664 if (write (new, page
, n
) != n
)
676 /* ****************************************************************
679 * After successfully building the new a.out, mark it executable
687 int new = 0; /* for PERROR */
691 if (stat (name
, &sbuf
) == -1)
695 sbuf
.st_mode
|= 0111 & ~um
;
696 if (chmod (name
, sbuf
.st_mode
) == -1)
701 * If the COFF file contains a symbol table and a line number section,
702 * then any auxiliary entries that have values for x_lnnoptr must
703 * be adjusted by the amount that the line number section has moved
704 * in the file (bias computed in make_hdr). The #@$%&* designers of
705 * the auxiliary entry structures used the absolute file offsets for
706 * the line number entry rather than an offset from the start of the
707 * line number section!
709 * When I figure out how to scan through the symbol table and pick out
710 * the auxiliary entries that need adjustment, this routine will
711 * be fixed. As it is now, all such entries are wrong and sdb
712 * will complain. Fred Fish, UniSoft Systems Inc.
714 * I believe this is now fixed correctly. Bill Mann
719 /* This function is probably very slow. Instead of reopening the new
720 file for input and output it should copy from the old to the new
721 using the two descriptors already open (WRITEDESC and READDESC).
722 Instead of reading one small structure at a time it should use
723 a reasonable size buffer. But I don't have time to work on such
724 things, so I am installing it as submitted to me. -- RMS. */
726 adjust_lnnoptrs (writedesc
, readdesc
, new_name
)
738 struct syment symentry
;
739 union auxent auxentry
;
742 if (!lnnoptr
|| !f_hdr
.f_symptr
)
745 if ((new = open (new_name
, 2)) < 0)
751 lseek (new, f_hdr
.f_symptr
, 0);
752 for (nsyms
= 0; nsyms
< f_hdr
.f_nsyms
; nsyms
++)
754 read (new, &symentry
, SYMESZ
);
755 if (symentry
.n_sclass
== C_BINCL
|| symentry
.n_sclass
== C_EINCL
)
757 symentry
.n_value
+= bias
;
758 lseek (new, -SYMESZ
, 1);
759 write (new, &symentry
, SYMESZ
);
762 for (naux
= symentry
.n_numaux
; naux
-- != 0; )
764 read (new, &auxentry
, AUXESZ
);
766 if (naux
!= 0 /* skip csect auxentry (last entry) */
767 && (symentry
.n_sclass
== C_EXT
|| symentry
.n_sclass
== C_HIDEXT
))
769 auxentry
.x_sym
.x_fcnary
.x_fcn
.x_lnnoptr
+= bias
;
770 lseek (new, -AUXESZ
, 1);
771 write (new, &auxentry
, AUXESZ
);
782 /* It is probably a false economy to optimise this routine (it used to
783 read one LDREL and do do two lseeks per iteration) but the wrath of
784 RMS (see above :-) would be too much to bear */
786 unrelocate_symbols (new, a_out
, a_name
, new_name
)
788 char *a_name
, *new_name
;
792 register LDREL
*ldrel
;
794 LDREL ldrel_buf
[20];
795 ulong t_reloc
= (ulong
) &_text
- f_ohdr
.text_start
;
796 ulong d_reloc
= (ulong
) &_data
- ALIGN(f_ohdr
.data_start
, 2);
799 if (load_scnptr
== 0)
802 lseek (a_out
, orig_load_scnptr
, 0);
803 if (read (a_out
, &ldhdr
, sizeof (ldhdr
)) != sizeof (ldhdr
))
808 #define SYMNDX_TEXT 0
809 #define SYMNDX_DATA 1
812 for (i
= 0; i
< ldhdr
.l_nreloc
; i
++, l
--, ldrel
++)
816 orig_load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
819 l
= ldhdr
.l_nreloc
- i
;
820 if (l
> sizeof (ldrel_buf
) / LDRELSZ
)
821 l
= sizeof (ldrel_buf
) / LDRELSZ
;
823 if (read (a_out
, ldrel_buf
, l
* LDRELSZ
) != l
* LDRELSZ
)
830 /* move the BSS loader symbols to the DATA segment */
831 if (ldrel
->l_symndx
== SYMNDX_BSS
)
833 ldrel
->l_symndx
= SYMNDX_DATA
;
836 load_scnptr
+ LDHDRSZ
+ LDSYMSZ
*ldhdr
.l_nsyms
+ LDRELSZ
*i
,
839 if (write (new, ldrel
, LDRELSZ
) != LDRELSZ
)
845 if (ldrel
->l_rsecnm
== f_ohdr
.o_sndata
)
850 orig_data_scnptr
+ (ldrel
->l_vaddr
- f_ohdr
.data_start
), 0);
852 if (read (a_out
, (void *) &orig_int
, sizeof (orig_int
)) != sizeof (orig_int
))
857 p
= (int *) (ldrel
->l_vaddr
+ d_reloc
);
859 switch (ldrel
->l_symndx
) {
861 orig_int
= * p
- t_reloc
;
866 orig_int
= * p
- d_reloc
;
873 data_scnptr
+ (ldrel
->l_vaddr
- f_ohdr
.data_start
), 0);
874 if (write (new, (void *) &orig_int
, sizeof (orig_int
))
875 != sizeof (orig_int
))